1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * 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 */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 23*b01c3b58Seschrock * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24fa9e4066Sahrens * Use is subject to license terms. 25fa9e4066Sahrens */ 26fa9e4066Sahrens 27fa9e4066Sahrens #pragma ident "%Z%%M% %I% %E% SMI" 28fa9e4066Sahrens 29fa9e4066Sahrens /* 30fa9e4066Sahrens * This file contains all the routines used when modifying on-disk SPA state. 31fa9e4066Sahrens * This includes opening, importing, destroying, exporting a pool, and syncing a 32fa9e4066Sahrens * pool. 33fa9e4066Sahrens */ 34fa9e4066Sahrens 35fa9e4066Sahrens #include <sys/zfs_context.h> 36ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 37fa9e4066Sahrens #include <sys/spa_impl.h> 38fa9e4066Sahrens #include <sys/zio.h> 39fa9e4066Sahrens #include <sys/zio_checksum.h> 40fa9e4066Sahrens #include <sys/zio_compress.h> 41fa9e4066Sahrens #include <sys/dmu.h> 42fa9e4066Sahrens #include <sys/dmu_tx.h> 43fa9e4066Sahrens #include <sys/zap.h> 44fa9e4066Sahrens #include <sys/zil.h> 45fa9e4066Sahrens #include <sys/vdev_impl.h> 46fa9e4066Sahrens #include <sys/metaslab.h> 47fa9e4066Sahrens #include <sys/uberblock_impl.h> 48fa9e4066Sahrens #include <sys/txg.h> 49fa9e4066Sahrens #include <sys/avl.h> 50fa9e4066Sahrens #include <sys/dmu_traverse.h> 51b1b8ab34Slling #include <sys/dmu_objset.h> 52fa9e4066Sahrens #include <sys/unique.h> 53fa9e4066Sahrens #include <sys/dsl_pool.h> 54b1b8ab34Slling #include <sys/dsl_dataset.h> 55fa9e4066Sahrens #include <sys/dsl_dir.h> 56fa9e4066Sahrens #include <sys/dsl_prop.h> 57b1b8ab34Slling #include <sys/dsl_synctask.h> 58fa9e4066Sahrens #include <sys/fs/zfs.h> 59fa94a07fSbrendan #include <sys/arc.h> 60fa9e4066Sahrens #include <sys/callb.h> 6195173954Sek #include <sys/systeminfo.h> 6295173954Sek #include <sys/sunddi.h> 63fa9e4066Sahrens 64990b4856Slling #include "zfs_prop.h" 65990b4856Slling 66416e0cd8Sek int zio_taskq_threads = 8; 67416e0cd8Sek 68990b4856Slling static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx); 69990b4856Slling 70990b4856Slling /* 71990b4856Slling * ========================================================================== 72990b4856Slling * SPA properties routines 73990b4856Slling * ========================================================================== 74990b4856Slling */ 75990b4856Slling 76990b4856Slling /* 77990b4856Slling * Add a (source=src, propname=propval) list to an nvlist. 78990b4856Slling */ 79990b4856Slling static int 80990b4856Slling spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 81990b4856Slling uint64_t intval, zprop_source_t src) 82990b4856Slling { 83990b4856Slling const char *propname = zpool_prop_to_name(prop); 84990b4856Slling nvlist_t *propval; 85990b4856Slling int err = 0; 86990b4856Slling 87990b4856Slling if (err = nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP)) 88990b4856Slling return (err); 89990b4856Slling 90990b4856Slling if (err = nvlist_add_uint64(propval, ZPROP_SOURCE, src)) 91990b4856Slling goto out; 92990b4856Slling 93990b4856Slling if (strval != NULL) { 94990b4856Slling if (err = nvlist_add_string(propval, ZPROP_VALUE, strval)) 95990b4856Slling goto out; 96990b4856Slling } else { 97990b4856Slling if (err = nvlist_add_uint64(propval, ZPROP_VALUE, intval)) 98990b4856Slling goto out; 99990b4856Slling } 100990b4856Slling 101990b4856Slling err = nvlist_add_nvlist(nvl, propname, propval); 102990b4856Slling out: 103990b4856Slling nvlist_free(propval); 104990b4856Slling return (err); 105990b4856Slling } 106990b4856Slling 107990b4856Slling /* 108990b4856Slling * Get property values from the spa configuration. 109990b4856Slling */ 110990b4856Slling static int 111990b4856Slling spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 112990b4856Slling { 113990b4856Slling uint64_t size = spa_get_space(spa); 114990b4856Slling uint64_t used = spa_get_alloc(spa); 115990b4856Slling uint64_t cap, version; 116990b4856Slling zprop_source_t src = ZPROP_SRC_NONE; 117990b4856Slling int err; 1182f8aaab3Seschrock char *cachefile; 1192f8aaab3Seschrock size_t len; 120990b4856Slling 121990b4856Slling /* 122990b4856Slling * readonly properties 123990b4856Slling */ 124990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa->spa_name, 125990b4856Slling 0, src)) 126990b4856Slling return (err); 127990b4856Slling 128990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src)) 129990b4856Slling return (err); 130990b4856Slling 131990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src)) 132990b4856Slling return (err); 133990b4856Slling 134990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL, 135990b4856Slling size - used, src)) 136990b4856Slling return (err); 137990b4856Slling 138990b4856Slling cap = (size == 0) ? 0 : (used * 100 / size); 139990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src)) 140990b4856Slling return (err); 141990b4856Slling 142990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, 143990b4856Slling spa_guid(spa), src)) 144990b4856Slling return (err); 145990b4856Slling 146990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 147990b4856Slling spa->spa_root_vdev->vdev_state, src)) 148990b4856Slling return (err); 149990b4856Slling 150990b4856Slling /* 151990b4856Slling * settable properties that are not stored in the pool property object. 152990b4856Slling */ 153990b4856Slling version = spa_version(spa); 154990b4856Slling if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 155990b4856Slling src = ZPROP_SRC_DEFAULT; 156990b4856Slling else 157990b4856Slling src = ZPROP_SRC_LOCAL; 158990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, 159990b4856Slling version, src)) 160990b4856Slling return (err); 161990b4856Slling 162990b4856Slling if (spa->spa_root != NULL) { 163990b4856Slling src = ZPROP_SRC_LOCAL; 164990b4856Slling if (err = spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, 165990b4856Slling spa->spa_root, 0, src)) 166990b4856Slling return (err); 167990b4856Slling } 168990b4856Slling 1692f8aaab3Seschrock if (spa->spa_config_dir != NULL) { 1702f8aaab3Seschrock if (strcmp(spa->spa_config_dir, "none") == 0) { 1712f8aaab3Seschrock err = spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 1722f8aaab3Seschrock spa->spa_config_dir, 0, ZPROP_SRC_LOCAL); 1732f8aaab3Seschrock } else { 1742f8aaab3Seschrock len = strlen(spa->spa_config_dir) + 1752f8aaab3Seschrock strlen(spa->spa_config_file) + 2; 1762f8aaab3Seschrock cachefile = kmem_alloc(len, KM_SLEEP); 1772f8aaab3Seschrock (void) snprintf(cachefile, len, "%s/%s", 1782f8aaab3Seschrock spa->spa_config_dir, spa->spa_config_file); 1792f8aaab3Seschrock err = spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 1802f8aaab3Seschrock cachefile, 0, ZPROP_SRC_LOCAL); 1812f8aaab3Seschrock kmem_free(cachefile, len); 1822f8aaab3Seschrock } 1832f8aaab3Seschrock 1842f8aaab3Seschrock if (err) 1852f8aaab3Seschrock return (err); 1862f8aaab3Seschrock } 187990b4856Slling 188990b4856Slling return (0); 189990b4856Slling } 190990b4856Slling 191990b4856Slling /* 192990b4856Slling * Get zpool property values. 193990b4856Slling */ 194990b4856Slling int 195990b4856Slling spa_prop_get(spa_t *spa, nvlist_t **nvp) 196990b4856Slling { 197990b4856Slling zap_cursor_t zc; 198990b4856Slling zap_attribute_t za; 199990b4856Slling objset_t *mos = spa->spa_meta_objset; 200990b4856Slling int err; 201990b4856Slling 202990b4856Slling if (err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)) 203990b4856Slling return (err); 204990b4856Slling 205990b4856Slling /* 206990b4856Slling * Get properties from the spa config. 207990b4856Slling */ 208990b4856Slling if (err = spa_prop_get_config(spa, nvp)) 209990b4856Slling goto out; 210990b4856Slling 211990b4856Slling mutex_enter(&spa->spa_props_lock); 212990b4856Slling /* If no pool property object, no more prop to get. */ 213990b4856Slling if (spa->spa_pool_props_object == 0) { 214990b4856Slling mutex_exit(&spa->spa_props_lock); 215990b4856Slling return (0); 216990b4856Slling } 217990b4856Slling 218990b4856Slling /* 219990b4856Slling * Get properties from the MOS pool property object. 220990b4856Slling */ 221990b4856Slling for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 222990b4856Slling (err = zap_cursor_retrieve(&zc, &za)) == 0; 223990b4856Slling zap_cursor_advance(&zc)) { 224990b4856Slling uint64_t intval = 0; 225990b4856Slling char *strval = NULL; 226990b4856Slling zprop_source_t src = ZPROP_SRC_DEFAULT; 227990b4856Slling zpool_prop_t prop; 228990b4856Slling 229990b4856Slling if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 230990b4856Slling continue; 231990b4856Slling 232990b4856Slling switch (za.za_integer_length) { 233990b4856Slling case 8: 234990b4856Slling /* integer property */ 235990b4856Slling if (za.za_first_integer != 236990b4856Slling zpool_prop_default_numeric(prop)) 237990b4856Slling src = ZPROP_SRC_LOCAL; 238990b4856Slling 239990b4856Slling if (prop == ZPOOL_PROP_BOOTFS) { 240990b4856Slling dsl_pool_t *dp; 241990b4856Slling dsl_dataset_t *ds = NULL; 242990b4856Slling 243990b4856Slling dp = spa_get_dsl(spa); 244990b4856Slling rw_enter(&dp->dp_config_rwlock, RW_READER); 245990b4856Slling if (err = dsl_dataset_open_obj(dp, 246990b4856Slling za.za_first_integer, NULL, DS_MODE_NONE, 247990b4856Slling FTAG, &ds)) { 248990b4856Slling rw_exit(&dp->dp_config_rwlock); 249990b4856Slling break; 250990b4856Slling } 251990b4856Slling 252990b4856Slling strval = kmem_alloc( 253990b4856Slling MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 254990b4856Slling KM_SLEEP); 255990b4856Slling dsl_dataset_name(ds, strval); 256990b4856Slling dsl_dataset_close(ds, DS_MODE_NONE, FTAG); 257990b4856Slling rw_exit(&dp->dp_config_rwlock); 258990b4856Slling } else { 259990b4856Slling strval = NULL; 260990b4856Slling intval = za.za_first_integer; 261990b4856Slling } 262990b4856Slling 263990b4856Slling err = spa_prop_add_list(*nvp, prop, strval, 264990b4856Slling intval, src); 265990b4856Slling 266990b4856Slling if (strval != NULL) 267990b4856Slling kmem_free(strval, 268990b4856Slling MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 269990b4856Slling 270990b4856Slling break; 271990b4856Slling 272990b4856Slling case 1: 273990b4856Slling /* string property */ 274990b4856Slling strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 275990b4856Slling err = zap_lookup(mos, spa->spa_pool_props_object, 276990b4856Slling za.za_name, 1, za.za_num_integers, strval); 277990b4856Slling if (err) { 278990b4856Slling kmem_free(strval, za.za_num_integers); 279990b4856Slling break; 280990b4856Slling } 281990b4856Slling err = spa_prop_add_list(*nvp, prop, strval, 0, src); 282990b4856Slling kmem_free(strval, za.za_num_integers); 283990b4856Slling break; 284990b4856Slling 285990b4856Slling default: 286990b4856Slling break; 287990b4856Slling } 288990b4856Slling } 289990b4856Slling zap_cursor_fini(&zc); 290990b4856Slling mutex_exit(&spa->spa_props_lock); 291990b4856Slling out: 292990b4856Slling if (err && err != ENOENT) { 293990b4856Slling nvlist_free(*nvp); 294990b4856Slling return (err); 295990b4856Slling } 296990b4856Slling 297990b4856Slling return (0); 298990b4856Slling } 299990b4856Slling 300990b4856Slling /* 301990b4856Slling * Validate the given pool properties nvlist and modify the list 302990b4856Slling * for the property values to be set. 303990b4856Slling */ 304990b4856Slling static int 305990b4856Slling spa_prop_validate(spa_t *spa, nvlist_t *props) 306990b4856Slling { 307990b4856Slling nvpair_t *elem; 308990b4856Slling int error = 0, reset_bootfs = 0; 309990b4856Slling uint64_t objnum; 310990b4856Slling 311990b4856Slling elem = NULL; 312990b4856Slling while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 313990b4856Slling zpool_prop_t prop; 314990b4856Slling char *propname, *strval; 315990b4856Slling uint64_t intval; 316990b4856Slling vdev_t *rvdev; 317990b4856Slling char *vdev_type; 318990b4856Slling objset_t *os; 3192f8aaab3Seschrock char *slash; 320990b4856Slling 321990b4856Slling propname = nvpair_name(elem); 322990b4856Slling 323990b4856Slling if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) 324990b4856Slling return (EINVAL); 325990b4856Slling 326990b4856Slling switch (prop) { 327990b4856Slling case ZPOOL_PROP_VERSION: 328990b4856Slling error = nvpair_value_uint64(elem, &intval); 329990b4856Slling if (!error && 330990b4856Slling (intval < spa_version(spa) || intval > SPA_VERSION)) 331990b4856Slling error = EINVAL; 332990b4856Slling break; 333990b4856Slling 334990b4856Slling case ZPOOL_PROP_DELEGATION: 335990b4856Slling case ZPOOL_PROP_AUTOREPLACE: 336990b4856Slling error = nvpair_value_uint64(elem, &intval); 337990b4856Slling if (!error && intval > 1) 338990b4856Slling error = EINVAL; 339990b4856Slling break; 340990b4856Slling 341990b4856Slling case ZPOOL_PROP_BOOTFS: 342990b4856Slling if (spa_version(spa) < SPA_VERSION_BOOTFS) { 343990b4856Slling error = ENOTSUP; 344990b4856Slling break; 345990b4856Slling } 346990b4856Slling 347990b4856Slling /* 348990b4856Slling * A bootable filesystem can not be on a RAIDZ pool 349990b4856Slling * nor a striped pool with more than 1 device. 350990b4856Slling */ 351990b4856Slling rvdev = spa->spa_root_vdev; 352990b4856Slling vdev_type = 353990b4856Slling rvdev->vdev_child[0]->vdev_ops->vdev_op_type; 354990b4856Slling if (rvdev->vdev_children > 1 || 355990b4856Slling strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 356990b4856Slling strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 357990b4856Slling error = ENOTSUP; 358990b4856Slling break; 359990b4856Slling } 360990b4856Slling 361990b4856Slling reset_bootfs = 1; 362990b4856Slling 363990b4856Slling error = nvpair_value_string(elem, &strval); 364990b4856Slling 365990b4856Slling if (!error) { 366990b4856Slling if (strval == NULL || strval[0] == '\0') { 367990b4856Slling objnum = zpool_prop_default_numeric( 368990b4856Slling ZPOOL_PROP_BOOTFS); 369990b4856Slling break; 370990b4856Slling } 371990b4856Slling 372990b4856Slling if (error = dmu_objset_open(strval, DMU_OST_ZFS, 373990b4856Slling DS_MODE_STANDARD | DS_MODE_READONLY, &os)) 374990b4856Slling break; 375990b4856Slling objnum = dmu_objset_id(os); 376990b4856Slling dmu_objset_close(os); 377990b4856Slling } 378990b4856Slling break; 3790a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 3800a4e9518Sgw error = nvpair_value_uint64(elem, &intval); 3810a4e9518Sgw if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 3820a4e9518Sgw intval > ZIO_FAILURE_MODE_PANIC)) 3830a4e9518Sgw error = EINVAL; 3840a4e9518Sgw 3850a4e9518Sgw /* 3860a4e9518Sgw * This is a special case which only occurs when 3870a4e9518Sgw * the pool has completely failed. This allows 3880a4e9518Sgw * the user to change the in-core failmode property 3890a4e9518Sgw * without syncing it out to disk (I/Os might 3900a4e9518Sgw * currently be blocked). We do this by returning 3910a4e9518Sgw * EIO to the caller (spa_prop_set) to trick it 3920a4e9518Sgw * into thinking we encountered a property validation 3930a4e9518Sgw * error. 3940a4e9518Sgw */ 3950a4e9518Sgw if (!error && spa_state(spa) == POOL_STATE_IO_FAILURE) { 3960a4e9518Sgw spa->spa_failmode = intval; 3970a4e9518Sgw error = EIO; 3980a4e9518Sgw } 3990a4e9518Sgw break; 4002f8aaab3Seschrock 4012f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 4022f8aaab3Seschrock if ((error = nvpair_value_string(elem, &strval)) != 0) 4032f8aaab3Seschrock break; 4042f8aaab3Seschrock 4052f8aaab3Seschrock if (strval[0] == '\0') 4062f8aaab3Seschrock break; 4072f8aaab3Seschrock 4082f8aaab3Seschrock if (strcmp(strval, "none") == 0) 4092f8aaab3Seschrock break; 4102f8aaab3Seschrock 4112f8aaab3Seschrock if (strval[0] != '/') { 4122f8aaab3Seschrock error = EINVAL; 4132f8aaab3Seschrock break; 4142f8aaab3Seschrock } 4152f8aaab3Seschrock 4162f8aaab3Seschrock slash = strrchr(strval, '/'); 4172f8aaab3Seschrock ASSERT(slash != NULL); 4182f8aaab3Seschrock 4192f8aaab3Seschrock if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 4202f8aaab3Seschrock strcmp(slash, "/..") == 0) 4212f8aaab3Seschrock error = EINVAL; 4222f8aaab3Seschrock break; 423990b4856Slling } 424990b4856Slling 425990b4856Slling if (error) 426990b4856Slling break; 427990b4856Slling } 428990b4856Slling 429990b4856Slling if (!error && reset_bootfs) { 430990b4856Slling error = nvlist_remove(props, 431990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 432990b4856Slling 433990b4856Slling if (!error) { 434990b4856Slling error = nvlist_add_uint64(props, 435990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 436990b4856Slling } 437990b4856Slling } 438990b4856Slling 439990b4856Slling return (error); 440990b4856Slling } 441990b4856Slling 442990b4856Slling int 443990b4856Slling spa_prop_set(spa_t *spa, nvlist_t *nvp) 444990b4856Slling { 445990b4856Slling int error; 446990b4856Slling 447990b4856Slling if ((error = spa_prop_validate(spa, nvp)) != 0) 448990b4856Slling return (error); 449990b4856Slling 450990b4856Slling return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, 451990b4856Slling spa, nvp, 3)); 452990b4856Slling } 453990b4856Slling 454990b4856Slling /* 455990b4856Slling * If the bootfs property value is dsobj, clear it. 456990b4856Slling */ 457990b4856Slling void 458990b4856Slling spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 459990b4856Slling { 460990b4856Slling if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 461990b4856Slling VERIFY(zap_remove(spa->spa_meta_objset, 462990b4856Slling spa->spa_pool_props_object, 463990b4856Slling zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 464990b4856Slling spa->spa_bootfs = 0; 465990b4856Slling } 466990b4856Slling } 467990b4856Slling 468fa9e4066Sahrens /* 469fa9e4066Sahrens * ========================================================================== 470fa9e4066Sahrens * SPA state manipulation (open/create/destroy/import/export) 471fa9e4066Sahrens * ========================================================================== 472fa9e4066Sahrens */ 473fa9e4066Sahrens 474ea8dc4b6Seschrock static int 475ea8dc4b6Seschrock spa_error_entry_compare(const void *a, const void *b) 476ea8dc4b6Seschrock { 477ea8dc4b6Seschrock spa_error_entry_t *sa = (spa_error_entry_t *)a; 478ea8dc4b6Seschrock spa_error_entry_t *sb = (spa_error_entry_t *)b; 479ea8dc4b6Seschrock int ret; 480ea8dc4b6Seschrock 481ea8dc4b6Seschrock ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 482ea8dc4b6Seschrock sizeof (zbookmark_t)); 483ea8dc4b6Seschrock 484ea8dc4b6Seschrock if (ret < 0) 485ea8dc4b6Seschrock return (-1); 486ea8dc4b6Seschrock else if (ret > 0) 487ea8dc4b6Seschrock return (1); 488ea8dc4b6Seschrock else 489ea8dc4b6Seschrock return (0); 490ea8dc4b6Seschrock } 491ea8dc4b6Seschrock 492ea8dc4b6Seschrock /* 493ea8dc4b6Seschrock * Utility function which retrieves copies of the current logs and 494ea8dc4b6Seschrock * re-initializes them in the process. 495ea8dc4b6Seschrock */ 496ea8dc4b6Seschrock void 497ea8dc4b6Seschrock spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 498ea8dc4b6Seschrock { 499ea8dc4b6Seschrock ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 500ea8dc4b6Seschrock 501ea8dc4b6Seschrock bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 502ea8dc4b6Seschrock bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 503ea8dc4b6Seschrock 504ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 505ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 506ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 507ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 508ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 509ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 510ea8dc4b6Seschrock } 511ea8dc4b6Seschrock 512fa9e4066Sahrens /* 513fa9e4066Sahrens * Activate an uninitialized pool. 514fa9e4066Sahrens */ 515fa9e4066Sahrens static void 516fa9e4066Sahrens spa_activate(spa_t *spa) 517fa9e4066Sahrens { 518fa9e4066Sahrens int t; 519fa9e4066Sahrens 520fa9e4066Sahrens ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 521fa9e4066Sahrens 522fa9e4066Sahrens spa->spa_state = POOL_STATE_ACTIVE; 523fa9e4066Sahrens 524fa9e4066Sahrens spa->spa_normal_class = metaslab_class_create(); 5258654d025Sperrin spa->spa_log_class = metaslab_class_create(); 526fa9e4066Sahrens 527fa9e4066Sahrens for (t = 0; t < ZIO_TYPES; t++) { 528fa9e4066Sahrens spa->spa_zio_issue_taskq[t] = taskq_create("spa_zio_issue", 529416e0cd8Sek zio_taskq_threads, maxclsyspri, 50, INT_MAX, 530fa9e4066Sahrens TASKQ_PREPOPULATE); 531fa9e4066Sahrens spa->spa_zio_intr_taskq[t] = taskq_create("spa_zio_intr", 532416e0cd8Sek zio_taskq_threads, maxclsyspri, 50, INT_MAX, 533fa9e4066Sahrens TASKQ_PREPOPULATE); 534fa9e4066Sahrens } 535fa9e4066Sahrens 536fa9e4066Sahrens list_create(&spa->spa_dirty_list, sizeof (vdev_t), 537fa9e4066Sahrens offsetof(vdev_t, vdev_dirty_node)); 5380a4e9518Sgw list_create(&spa->spa_zio_list, sizeof (zio_t), 5390a4e9518Sgw offsetof(zio_t, zio_link_node)); 540fa9e4066Sahrens 541fa9e4066Sahrens txg_list_create(&spa->spa_vdev_txg_list, 542fa9e4066Sahrens offsetof(struct vdev, vdev_txg_node)); 543ea8dc4b6Seschrock 544ea8dc4b6Seschrock avl_create(&spa->spa_errlist_scrub, 545ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 546ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 547ea8dc4b6Seschrock avl_create(&spa->spa_errlist_last, 548ea8dc4b6Seschrock spa_error_entry_compare, sizeof (spa_error_entry_t), 549ea8dc4b6Seschrock offsetof(spa_error_entry_t, se_avl)); 550fa9e4066Sahrens } 551fa9e4066Sahrens 552fa9e4066Sahrens /* 553fa9e4066Sahrens * Opposite of spa_activate(). 554fa9e4066Sahrens */ 555fa9e4066Sahrens static void 556fa9e4066Sahrens spa_deactivate(spa_t *spa) 557fa9e4066Sahrens { 558fa9e4066Sahrens int t; 559fa9e4066Sahrens 560fa9e4066Sahrens ASSERT(spa->spa_sync_on == B_FALSE); 561fa9e4066Sahrens ASSERT(spa->spa_dsl_pool == NULL); 562fa9e4066Sahrens ASSERT(spa->spa_root_vdev == NULL); 563fa9e4066Sahrens 564fa9e4066Sahrens ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 565fa9e4066Sahrens 566fa9e4066Sahrens txg_list_destroy(&spa->spa_vdev_txg_list); 567fa9e4066Sahrens 568fa9e4066Sahrens list_destroy(&spa->spa_dirty_list); 5690a4e9518Sgw list_destroy(&spa->spa_zio_list); 570fa9e4066Sahrens 571fa9e4066Sahrens for (t = 0; t < ZIO_TYPES; t++) { 572fa9e4066Sahrens taskq_destroy(spa->spa_zio_issue_taskq[t]); 573fa9e4066Sahrens taskq_destroy(spa->spa_zio_intr_taskq[t]); 574fa9e4066Sahrens spa->spa_zio_issue_taskq[t] = NULL; 575fa9e4066Sahrens spa->spa_zio_intr_taskq[t] = NULL; 576fa9e4066Sahrens } 577fa9e4066Sahrens 578fa9e4066Sahrens metaslab_class_destroy(spa->spa_normal_class); 579fa9e4066Sahrens spa->spa_normal_class = NULL; 580fa9e4066Sahrens 5818654d025Sperrin metaslab_class_destroy(spa->spa_log_class); 5828654d025Sperrin spa->spa_log_class = NULL; 5838654d025Sperrin 584ea8dc4b6Seschrock /* 585ea8dc4b6Seschrock * If this was part of an import or the open otherwise failed, we may 586ea8dc4b6Seschrock * still have errors left in the queues. Empty them just in case. 587ea8dc4b6Seschrock */ 588ea8dc4b6Seschrock spa_errlog_drain(spa); 589ea8dc4b6Seschrock 590ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_scrub); 591ea8dc4b6Seschrock avl_destroy(&spa->spa_errlist_last); 592ea8dc4b6Seschrock 593fa9e4066Sahrens spa->spa_state = POOL_STATE_UNINITIALIZED; 594fa9e4066Sahrens } 595fa9e4066Sahrens 596fa9e4066Sahrens /* 597fa9e4066Sahrens * Verify a pool configuration, and construct the vdev tree appropriately. This 598fa9e4066Sahrens * will create all the necessary vdevs in the appropriate layout, with each vdev 599fa9e4066Sahrens * in the CLOSED state. This will prep the pool before open/creation/import. 600fa9e4066Sahrens * All vdev validation is done by the vdev_alloc() routine. 601fa9e4066Sahrens */ 60299653d4eSeschrock static int 60399653d4eSeschrock spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 60499653d4eSeschrock uint_t id, int atype) 605fa9e4066Sahrens { 606fa9e4066Sahrens nvlist_t **child; 607fa9e4066Sahrens uint_t c, children; 60899653d4eSeschrock int error; 609fa9e4066Sahrens 61099653d4eSeschrock if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 61199653d4eSeschrock return (error); 612fa9e4066Sahrens 61399653d4eSeschrock if ((*vdp)->vdev_ops->vdev_op_leaf) 61499653d4eSeschrock return (0); 615fa9e4066Sahrens 616fa9e4066Sahrens if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 617fa9e4066Sahrens &child, &children) != 0) { 61899653d4eSeschrock vdev_free(*vdp); 61999653d4eSeschrock *vdp = NULL; 62099653d4eSeschrock return (EINVAL); 621fa9e4066Sahrens } 622fa9e4066Sahrens 623fa9e4066Sahrens for (c = 0; c < children; c++) { 62499653d4eSeschrock vdev_t *vd; 62599653d4eSeschrock if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 62699653d4eSeschrock atype)) != 0) { 62799653d4eSeschrock vdev_free(*vdp); 62899653d4eSeschrock *vdp = NULL; 62999653d4eSeschrock return (error); 630fa9e4066Sahrens } 631fa9e4066Sahrens } 632fa9e4066Sahrens 63399653d4eSeschrock ASSERT(*vdp != NULL); 63499653d4eSeschrock 63599653d4eSeschrock return (0); 636fa9e4066Sahrens } 637fa9e4066Sahrens 638fa9e4066Sahrens /* 639fa9e4066Sahrens * Opposite of spa_load(). 640fa9e4066Sahrens */ 641fa9e4066Sahrens static void 642fa9e4066Sahrens spa_unload(spa_t *spa) 643fa9e4066Sahrens { 64499653d4eSeschrock int i; 64599653d4eSeschrock 646ea8dc4b6Seschrock /* 647ea8dc4b6Seschrock * Stop async tasks. 648ea8dc4b6Seschrock */ 649ea8dc4b6Seschrock spa_async_suspend(spa); 650ea8dc4b6Seschrock 651fa9e4066Sahrens /* 652fa9e4066Sahrens * Stop syncing. 653fa9e4066Sahrens */ 654fa9e4066Sahrens if (spa->spa_sync_on) { 655fa9e4066Sahrens txg_sync_stop(spa->spa_dsl_pool); 656fa9e4066Sahrens spa->spa_sync_on = B_FALSE; 657fa9e4066Sahrens } 658fa9e4066Sahrens 659fa9e4066Sahrens /* 660fa9e4066Sahrens * Wait for any outstanding prefetch I/O to complete. 661fa9e4066Sahrens */ 662ea8dc4b6Seschrock spa_config_enter(spa, RW_WRITER, FTAG); 663ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 664fa9e4066Sahrens 665fa94a07fSbrendan /* 666fa94a07fSbrendan * Drop and purge level 2 cache 667fa94a07fSbrendan */ 668fa94a07fSbrendan spa_l2cache_drop(spa); 669fa94a07fSbrendan 670fa9e4066Sahrens /* 671fa9e4066Sahrens * Close the dsl pool. 672fa9e4066Sahrens */ 673fa9e4066Sahrens if (spa->spa_dsl_pool) { 674fa9e4066Sahrens dsl_pool_close(spa->spa_dsl_pool); 675fa9e4066Sahrens spa->spa_dsl_pool = NULL; 676fa9e4066Sahrens } 677fa9e4066Sahrens 678fa9e4066Sahrens /* 679fa9e4066Sahrens * Close all vdevs. 680fa9e4066Sahrens */ 6810e34b6a7Sbonwick if (spa->spa_root_vdev) 682fa9e4066Sahrens vdev_free(spa->spa_root_vdev); 6830e34b6a7Sbonwick ASSERT(spa->spa_root_vdev == NULL); 684ea8dc4b6Seschrock 685fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 686fa94a07fSbrendan vdev_free(spa->spa_spares.sav_vdevs[i]); 687fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) { 688fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 689fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 690fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 69199653d4eSeschrock } 692fa94a07fSbrendan if (spa->spa_spares.sav_config) { 693fa94a07fSbrendan nvlist_free(spa->spa_spares.sav_config); 694fa94a07fSbrendan spa->spa_spares.sav_config = NULL; 695fa94a07fSbrendan } 696fa94a07fSbrendan 697fa94a07fSbrendan for (i = 0; i < spa->spa_l2cache.sav_count; i++) 698fa94a07fSbrendan vdev_free(spa->spa_l2cache.sav_vdevs[i]); 699fa94a07fSbrendan if (spa->spa_l2cache.sav_vdevs) { 700fa94a07fSbrendan kmem_free(spa->spa_l2cache.sav_vdevs, 701fa94a07fSbrendan spa->spa_l2cache.sav_count * sizeof (void *)); 702fa94a07fSbrendan spa->spa_l2cache.sav_vdevs = NULL; 703fa94a07fSbrendan } 704fa94a07fSbrendan if (spa->spa_l2cache.sav_config) { 705fa94a07fSbrendan nvlist_free(spa->spa_l2cache.sav_config); 706fa94a07fSbrendan spa->spa_l2cache.sav_config = NULL; 70799653d4eSeschrock } 70899653d4eSeschrock 709ea8dc4b6Seschrock spa->spa_async_suspended = 0; 710fa9e4066Sahrens } 711fa9e4066Sahrens 71299653d4eSeschrock /* 71399653d4eSeschrock * Load (or re-load) the current list of vdevs describing the active spares for 71499653d4eSeschrock * this pool. When this is called, we have some form of basic information in 715fa94a07fSbrendan * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 716fa94a07fSbrendan * then re-generate a more complete list including status information. 71799653d4eSeschrock */ 71899653d4eSeschrock static void 71999653d4eSeschrock spa_load_spares(spa_t *spa) 72099653d4eSeschrock { 72199653d4eSeschrock nvlist_t **spares; 72299653d4eSeschrock uint_t nspares; 72399653d4eSeschrock int i; 72439c23413Seschrock vdev_t *vd, *tvd; 72599653d4eSeschrock 72699653d4eSeschrock /* 72799653d4eSeschrock * First, close and free any existing spare vdevs. 72899653d4eSeschrock */ 729fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 730fa94a07fSbrendan vd = spa->spa_spares.sav_vdevs[i]; 73139c23413Seschrock 73239c23413Seschrock /* Undo the call to spa_activate() below */ 73339c23413Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL && 73439c23413Seschrock tvd->vdev_isspare) 73539c23413Seschrock spa_spare_remove(tvd); 73639c23413Seschrock vdev_close(vd); 73739c23413Seschrock vdev_free(vd); 73899653d4eSeschrock } 73939c23413Seschrock 740fa94a07fSbrendan if (spa->spa_spares.sav_vdevs) 741fa94a07fSbrendan kmem_free(spa->spa_spares.sav_vdevs, 742fa94a07fSbrendan spa->spa_spares.sav_count * sizeof (void *)); 74399653d4eSeschrock 744fa94a07fSbrendan if (spa->spa_spares.sav_config == NULL) 74599653d4eSeschrock nspares = 0; 74699653d4eSeschrock else 747fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 74899653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 74999653d4eSeschrock 750fa94a07fSbrendan spa->spa_spares.sav_count = (int)nspares; 751fa94a07fSbrendan spa->spa_spares.sav_vdevs = NULL; 75299653d4eSeschrock 75399653d4eSeschrock if (nspares == 0) 75499653d4eSeschrock return; 75599653d4eSeschrock 75699653d4eSeschrock /* 75799653d4eSeschrock * Construct the array of vdevs, opening them to get status in the 75839c23413Seschrock * process. For each spare, there is potentially two different vdev_t 75939c23413Seschrock * structures associated with it: one in the list of spares (used only 76039c23413Seschrock * for basic validation purposes) and one in the active vdev 76139c23413Seschrock * configuration (if it's spared in). During this phase we open and 76239c23413Seschrock * validate each vdev on the spare list. If the vdev also exists in the 76339c23413Seschrock * active configuration, then we also mark this vdev as an active spare. 76499653d4eSeschrock */ 765fa94a07fSbrendan spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 766fa94a07fSbrendan KM_SLEEP); 767fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) { 76899653d4eSeschrock VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 76999653d4eSeschrock VDEV_ALLOC_SPARE) == 0); 77099653d4eSeschrock ASSERT(vd != NULL); 77199653d4eSeschrock 772fa94a07fSbrendan spa->spa_spares.sav_vdevs[i] = vd; 77399653d4eSeschrock 77439c23413Seschrock if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid)) != NULL) { 77539c23413Seschrock if (!tvd->vdev_isspare) 77639c23413Seschrock spa_spare_add(tvd); 77739c23413Seschrock 77839c23413Seschrock /* 77939c23413Seschrock * We only mark the spare active if we were successfully 78039c23413Seschrock * able to load the vdev. Otherwise, importing a pool 78139c23413Seschrock * with a bad active spare would result in strange 78239c23413Seschrock * behavior, because multiple pool would think the spare 78339c23413Seschrock * is actively in use. 78439c23413Seschrock * 78539c23413Seschrock * There is a vulnerability here to an equally bizarre 78639c23413Seschrock * circumstance, where a dead active spare is later 78739c23413Seschrock * brought back to life (onlined or otherwise). Given 78839c23413Seschrock * the rarity of this scenario, and the extra complexity 78939c23413Seschrock * it adds, we ignore the possibility. 79039c23413Seschrock */ 79139c23413Seschrock if (!vdev_is_dead(tvd)) 79239c23413Seschrock spa_spare_activate(tvd); 79339c23413Seschrock } 79439c23413Seschrock 79599653d4eSeschrock if (vdev_open(vd) != 0) 79699653d4eSeschrock continue; 79799653d4eSeschrock 79899653d4eSeschrock vd->vdev_top = vd; 799fa94a07fSbrendan if (vdev_validate_aux(vd) == 0) 800fa94a07fSbrendan spa_spare_add(vd); 80199653d4eSeschrock } 80299653d4eSeschrock 80399653d4eSeschrock /* 80499653d4eSeschrock * Recompute the stashed list of spares, with status information 80599653d4eSeschrock * this time. 80699653d4eSeschrock */ 807fa94a07fSbrendan VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 80899653d4eSeschrock DATA_TYPE_NVLIST_ARRAY) == 0); 80999653d4eSeschrock 810fa94a07fSbrendan spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 811fa94a07fSbrendan KM_SLEEP); 812fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 813fa94a07fSbrendan spares[i] = vdev_config_generate(spa, 814fa94a07fSbrendan spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE); 815fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 816fa94a07fSbrendan ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 817fa94a07fSbrendan for (i = 0; i < spa->spa_spares.sav_count; i++) 81899653d4eSeschrock nvlist_free(spares[i]); 819fa94a07fSbrendan kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 820fa94a07fSbrendan } 821fa94a07fSbrendan 822fa94a07fSbrendan /* 823fa94a07fSbrendan * Load (or re-load) the current list of vdevs describing the active l2cache for 824fa94a07fSbrendan * this pool. When this is called, we have some form of basic information in 825fa94a07fSbrendan * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 826fa94a07fSbrendan * then re-generate a more complete list including status information. 827fa94a07fSbrendan * Devices which are already active have their details maintained, and are 828fa94a07fSbrendan * not re-opened. 829fa94a07fSbrendan */ 830fa94a07fSbrendan static void 831fa94a07fSbrendan spa_load_l2cache(spa_t *spa) 832fa94a07fSbrendan { 833fa94a07fSbrendan nvlist_t **l2cache; 834fa94a07fSbrendan uint_t nl2cache; 835fa94a07fSbrendan int i, j, oldnvdevs; 836fa94a07fSbrendan uint64_t guid; 837fa94a07fSbrendan vdev_t *vd, **oldvdevs, **newvdevs; 838fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 839fa94a07fSbrendan 840fa94a07fSbrendan if (sav->sav_config != NULL) { 841fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 842fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 843fa94a07fSbrendan newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 844fa94a07fSbrendan } else { 845fa94a07fSbrendan nl2cache = 0; 846fa94a07fSbrendan } 847fa94a07fSbrendan 848fa94a07fSbrendan oldvdevs = sav->sav_vdevs; 849fa94a07fSbrendan oldnvdevs = sav->sav_count; 850fa94a07fSbrendan sav->sav_vdevs = NULL; 851fa94a07fSbrendan sav->sav_count = 0; 852fa94a07fSbrendan 853fa94a07fSbrendan /* 854fa94a07fSbrendan * Process new nvlist of vdevs. 855fa94a07fSbrendan */ 856fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 857fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 858fa94a07fSbrendan &guid) == 0); 859fa94a07fSbrendan 860fa94a07fSbrendan newvdevs[i] = NULL; 861fa94a07fSbrendan for (j = 0; j < oldnvdevs; j++) { 862fa94a07fSbrendan vd = oldvdevs[j]; 863fa94a07fSbrendan if (vd != NULL && guid == vd->vdev_guid) { 864fa94a07fSbrendan /* 865fa94a07fSbrendan * Retain previous vdev for add/remove ops. 866fa94a07fSbrendan */ 867fa94a07fSbrendan newvdevs[i] = vd; 868fa94a07fSbrendan oldvdevs[j] = NULL; 869fa94a07fSbrendan break; 870fa94a07fSbrendan } 871fa94a07fSbrendan } 872fa94a07fSbrendan 873fa94a07fSbrendan if (newvdevs[i] == NULL) { 874fa94a07fSbrendan /* 875fa94a07fSbrendan * Create new vdev 876fa94a07fSbrendan */ 877fa94a07fSbrendan VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 878fa94a07fSbrendan VDEV_ALLOC_L2CACHE) == 0); 879fa94a07fSbrendan ASSERT(vd != NULL); 880fa94a07fSbrendan newvdevs[i] = vd; 881fa94a07fSbrendan 882fa94a07fSbrendan /* 883fa94a07fSbrendan * Commit this vdev as an l2cache device, 884fa94a07fSbrendan * even if it fails to open. 885fa94a07fSbrendan */ 886fa94a07fSbrendan spa_l2cache_add(vd); 887fa94a07fSbrendan 888fa94a07fSbrendan if (vdev_open(vd) != 0) 889fa94a07fSbrendan continue; 890fa94a07fSbrendan 891fa94a07fSbrendan vd->vdev_top = vd; 892fa94a07fSbrendan (void) vdev_validate_aux(vd); 893fa94a07fSbrendan 894fa94a07fSbrendan if (!vdev_is_dead(vd)) { 895fa94a07fSbrendan uint64_t size; 896fa94a07fSbrendan size = vdev_get_rsize(vd); 897fa94a07fSbrendan ASSERT3U(size, >, 0); 898fa94a07fSbrendan if (spa_mode & FWRITE) { 899fa94a07fSbrendan l2arc_add_vdev(spa, vd, 900fa94a07fSbrendan VDEV_LABEL_START_SIZE, 901fa94a07fSbrendan size - VDEV_LABEL_START_SIZE); 902fa94a07fSbrendan } 903fa94a07fSbrendan spa_l2cache_activate(vd); 904fa94a07fSbrendan } 905fa94a07fSbrendan } 906fa94a07fSbrendan } 907fa94a07fSbrendan 908fa94a07fSbrendan /* 909fa94a07fSbrendan * Purge vdevs that were dropped 910fa94a07fSbrendan */ 911fa94a07fSbrendan for (i = 0; i < oldnvdevs; i++) { 912fa94a07fSbrendan uint64_t pool; 913fa94a07fSbrendan 914fa94a07fSbrendan vd = oldvdevs[i]; 915fa94a07fSbrendan if (vd != NULL) { 916fa94a07fSbrendan if (spa_mode & FWRITE && 917fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, &pool) && 918fa94a07fSbrendan pool != 0ULL) { 919fa94a07fSbrendan l2arc_remove_vdev(vd); 920fa94a07fSbrendan } 921fa94a07fSbrendan (void) vdev_close(vd); 922fa94a07fSbrendan spa_l2cache_remove(vd); 923fa94a07fSbrendan } 924fa94a07fSbrendan } 925fa94a07fSbrendan 926fa94a07fSbrendan if (oldvdevs) 927fa94a07fSbrendan kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 928fa94a07fSbrendan 929fa94a07fSbrendan if (sav->sav_config == NULL) 930fa94a07fSbrendan goto out; 931fa94a07fSbrendan 932fa94a07fSbrendan sav->sav_vdevs = newvdevs; 933fa94a07fSbrendan sav->sav_count = (int)nl2cache; 934fa94a07fSbrendan 935fa94a07fSbrendan /* 936fa94a07fSbrendan * Recompute the stashed list of l2cache devices, with status 937fa94a07fSbrendan * information this time. 938fa94a07fSbrendan */ 939fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 940fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 941fa94a07fSbrendan 942fa94a07fSbrendan l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 943fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 944fa94a07fSbrendan l2cache[i] = vdev_config_generate(spa, 945fa94a07fSbrendan sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE); 946fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 947fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 948fa94a07fSbrendan out: 949fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 950fa94a07fSbrendan nvlist_free(l2cache[i]); 951fa94a07fSbrendan if (sav->sav_count) 952fa94a07fSbrendan kmem_free(l2cache, sav->sav_count * sizeof (void *)); 95399653d4eSeschrock } 95499653d4eSeschrock 95599653d4eSeschrock static int 95699653d4eSeschrock load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 95799653d4eSeschrock { 95899653d4eSeschrock dmu_buf_t *db; 95999653d4eSeschrock char *packed = NULL; 96099653d4eSeschrock size_t nvsize = 0; 96199653d4eSeschrock int error; 96299653d4eSeschrock *value = NULL; 96399653d4eSeschrock 96499653d4eSeschrock VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 96599653d4eSeschrock nvsize = *(uint64_t *)db->db_data; 96699653d4eSeschrock dmu_buf_rele(db, FTAG); 96799653d4eSeschrock 96899653d4eSeschrock packed = kmem_alloc(nvsize, KM_SLEEP); 96999653d4eSeschrock error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed); 97099653d4eSeschrock if (error == 0) 97199653d4eSeschrock error = nvlist_unpack(packed, nvsize, value, 0); 97299653d4eSeschrock kmem_free(packed, nvsize); 97399653d4eSeschrock 97499653d4eSeschrock return (error); 97599653d4eSeschrock } 97699653d4eSeschrock 9773d7072f8Seschrock /* 9783d7072f8Seschrock * Checks to see if the given vdev could not be opened, in which case we post a 9793d7072f8Seschrock * sysevent to notify the autoreplace code that the device has been removed. 9803d7072f8Seschrock */ 9813d7072f8Seschrock static void 9823d7072f8Seschrock spa_check_removed(vdev_t *vd) 9833d7072f8Seschrock { 9843d7072f8Seschrock int c; 9853d7072f8Seschrock 9863d7072f8Seschrock for (c = 0; c < vd->vdev_children; c++) 9873d7072f8Seschrock spa_check_removed(vd->vdev_child[c]); 9883d7072f8Seschrock 9893d7072f8Seschrock if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) { 9903d7072f8Seschrock zfs_post_autoreplace(vd->vdev_spa, vd); 9913d7072f8Seschrock spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 9923d7072f8Seschrock } 9933d7072f8Seschrock } 9943d7072f8Seschrock 995fa9e4066Sahrens /* 996fa9e4066Sahrens * Load an existing storage pool, using the pool's builtin spa_config as a 997ea8dc4b6Seschrock * source of configuration information. 998fa9e4066Sahrens */ 999fa9e4066Sahrens static int 1000ea8dc4b6Seschrock spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig) 1001fa9e4066Sahrens { 1002fa9e4066Sahrens int error = 0; 1003fa9e4066Sahrens nvlist_t *nvroot = NULL; 1004fa9e4066Sahrens vdev_t *rvd; 1005fa9e4066Sahrens uberblock_t *ub = &spa->spa_uberblock; 10060373e76bSbonwick uint64_t config_cache_txg = spa->spa_config_txg; 1007fa9e4066Sahrens uint64_t pool_guid; 100899653d4eSeschrock uint64_t version; 1009fa9e4066Sahrens zio_t *zio; 10103d7072f8Seschrock uint64_t autoreplace = 0; 1011fa9e4066Sahrens 1012ea8dc4b6Seschrock spa->spa_load_state = state; 10130373e76bSbonwick 1014fa9e4066Sahrens if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || 1015a9926bf0Sbonwick nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { 1016ea8dc4b6Seschrock error = EINVAL; 1017ea8dc4b6Seschrock goto out; 1018ea8dc4b6Seschrock } 1019fa9e4066Sahrens 102099653d4eSeschrock /* 102199653d4eSeschrock * Versioning wasn't explicitly added to the label until later, so if 102299653d4eSeschrock * it's not present treat it as the initial version. 102399653d4eSeschrock */ 102499653d4eSeschrock if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0) 1025e7437265Sahrens version = SPA_VERSION_INITIAL; 102699653d4eSeschrock 1027a9926bf0Sbonwick (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 1028a9926bf0Sbonwick &spa->spa_config_txg); 1029a9926bf0Sbonwick 10300373e76bSbonwick if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 1031ea8dc4b6Seschrock spa_guid_exists(pool_guid, 0)) { 1032ea8dc4b6Seschrock error = EEXIST; 1033ea8dc4b6Seschrock goto out; 1034ea8dc4b6Seschrock } 1035fa9e4066Sahrens 1036b5989ec7Seschrock spa->spa_load_guid = pool_guid; 1037b5989ec7Seschrock 1038fa9e4066Sahrens /* 103999653d4eSeschrock * Parse the configuration into a vdev tree. We explicitly set the 104099653d4eSeschrock * value that will be returned by spa_version() since parsing the 104199653d4eSeschrock * configuration requires knowing the version number. 1042fa9e4066Sahrens */ 1043ea8dc4b6Seschrock spa_config_enter(spa, RW_WRITER, FTAG); 104499653d4eSeschrock spa->spa_ubsync.ub_version = version; 104599653d4eSeschrock error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD); 1046ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 1047fa9e4066Sahrens 104899653d4eSeschrock if (error != 0) 1049ea8dc4b6Seschrock goto out; 1050fa9e4066Sahrens 10510e34b6a7Sbonwick ASSERT(spa->spa_root_vdev == rvd); 1052fa9e4066Sahrens ASSERT(spa_guid(spa) == pool_guid); 1053fa9e4066Sahrens 1054fa9e4066Sahrens /* 1055fa9e4066Sahrens * Try to open all vdevs, loading each label in the process. 1056fa9e4066Sahrens */ 10570bf246f5Smc error = vdev_open(rvd); 10580bf246f5Smc if (error != 0) 1059ea8dc4b6Seschrock goto out; 1060fa9e4066Sahrens 1061560e6e96Seschrock /* 1062560e6e96Seschrock * Validate the labels for all leaf vdevs. We need to grab the config 1063560e6e96Seschrock * lock because all label I/O is done with the ZIO_FLAG_CONFIG_HELD 1064560e6e96Seschrock * flag. 1065560e6e96Seschrock */ 1066560e6e96Seschrock spa_config_enter(spa, RW_READER, FTAG); 1067560e6e96Seschrock error = vdev_validate(rvd); 1068560e6e96Seschrock spa_config_exit(spa, FTAG); 1069560e6e96Seschrock 10700bf246f5Smc if (error != 0) 1071560e6e96Seschrock goto out; 1072560e6e96Seschrock 1073560e6e96Seschrock if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 1074560e6e96Seschrock error = ENXIO; 1075560e6e96Seschrock goto out; 1076560e6e96Seschrock } 1077560e6e96Seschrock 1078fa9e4066Sahrens /* 1079fa9e4066Sahrens * Find the best uberblock. 1080fa9e4066Sahrens */ 1081fa9e4066Sahrens bzero(ub, sizeof (uberblock_t)); 1082fa9e4066Sahrens 1083fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, 1084fa9e4066Sahrens ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1085fa9e4066Sahrens vdev_uberblock_load(zio, rvd, ub); 1086fa9e4066Sahrens error = zio_wait(zio); 1087fa9e4066Sahrens 1088fa9e4066Sahrens /* 1089fa9e4066Sahrens * If we weren't able to find a single valid uberblock, return failure. 1090fa9e4066Sahrens */ 1091fa9e4066Sahrens if (ub->ub_txg == 0) { 1092eaca9bbdSeschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1093eaca9bbdSeschrock VDEV_AUX_CORRUPT_DATA); 1094ea8dc4b6Seschrock error = ENXIO; 1095ea8dc4b6Seschrock goto out; 1096ea8dc4b6Seschrock } 1097ea8dc4b6Seschrock 1098ea8dc4b6Seschrock /* 1099ea8dc4b6Seschrock * If the pool is newer than the code, we can't open it. 1100ea8dc4b6Seschrock */ 1101e7437265Sahrens if (ub->ub_version > SPA_VERSION) { 1102eaca9bbdSeschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1103eaca9bbdSeschrock VDEV_AUX_VERSION_NEWER); 1104ea8dc4b6Seschrock error = ENOTSUP; 1105ea8dc4b6Seschrock goto out; 1106fa9e4066Sahrens } 1107fa9e4066Sahrens 1108fa9e4066Sahrens /* 1109fa9e4066Sahrens * If the vdev guid sum doesn't match the uberblock, we have an 1110fa9e4066Sahrens * incomplete configuration. 1111fa9e4066Sahrens */ 1112ecc2d604Sbonwick if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) { 1113ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1114ea8dc4b6Seschrock VDEV_AUX_BAD_GUID_SUM); 1115ea8dc4b6Seschrock error = ENXIO; 1116ea8dc4b6Seschrock goto out; 1117fa9e4066Sahrens } 1118fa9e4066Sahrens 1119fa9e4066Sahrens /* 1120fa9e4066Sahrens * Initialize internal SPA structures. 1121fa9e4066Sahrens */ 1122fa9e4066Sahrens spa->spa_state = POOL_STATE_ACTIVE; 1123fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 1124fa9e4066Sahrens spa->spa_first_txg = spa_last_synced_txg(spa) + 1; 1125ea8dc4b6Seschrock error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 1126ea8dc4b6Seschrock if (error) { 1127ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1128ea8dc4b6Seschrock VDEV_AUX_CORRUPT_DATA); 1129ea8dc4b6Seschrock goto out; 1130ea8dc4b6Seschrock } 1131fa9e4066Sahrens spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 1132fa9e4066Sahrens 1133ea8dc4b6Seschrock if (zap_lookup(spa->spa_meta_objset, 1134fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 1135ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_config_object) != 0) { 1136ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1137ea8dc4b6Seschrock VDEV_AUX_CORRUPT_DATA); 1138ea8dc4b6Seschrock error = EIO; 1139ea8dc4b6Seschrock goto out; 1140ea8dc4b6Seschrock } 1141fa9e4066Sahrens 1142fa9e4066Sahrens if (!mosconfig) { 114399653d4eSeschrock nvlist_t *newconfig; 114495173954Sek uint64_t hostid; 1145fa9e4066Sahrens 114699653d4eSeschrock if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) { 1147ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1148ea8dc4b6Seschrock VDEV_AUX_CORRUPT_DATA); 1149ea8dc4b6Seschrock error = EIO; 1150ea8dc4b6Seschrock goto out; 1151ea8dc4b6Seschrock } 1152fa9e4066Sahrens 115395173954Sek if (nvlist_lookup_uint64(newconfig, ZPOOL_CONFIG_HOSTID, 115495173954Sek &hostid) == 0) { 115595173954Sek char *hostname; 115695173954Sek unsigned long myhostid = 0; 115795173954Sek 115895173954Sek VERIFY(nvlist_lookup_string(newconfig, 115995173954Sek ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 116095173954Sek 116195173954Sek (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 116217194a52Slling if (hostid != 0 && myhostid != 0 && 116317194a52Slling (unsigned long)hostid != myhostid) { 116495173954Sek cmn_err(CE_WARN, "pool '%s' could not be " 116595173954Sek "loaded as it was last accessed by " 116695173954Sek "another system (host: %s hostid: 0x%lx). " 116795173954Sek "See: http://www.sun.com/msg/ZFS-8000-EY", 116895173954Sek spa->spa_name, hostname, 116995173954Sek (unsigned long)hostid); 117095173954Sek error = EBADF; 117195173954Sek goto out; 117295173954Sek } 117395173954Sek } 117495173954Sek 1175fa9e4066Sahrens spa_config_set(spa, newconfig); 1176fa9e4066Sahrens spa_unload(spa); 1177fa9e4066Sahrens spa_deactivate(spa); 1178fa9e4066Sahrens spa_activate(spa); 1179fa9e4066Sahrens 1180ea8dc4b6Seschrock return (spa_load(spa, newconfig, state, B_TRUE)); 1181fa9e4066Sahrens } 1182fa9e4066Sahrens 1183ea8dc4b6Seschrock if (zap_lookup(spa->spa_meta_objset, 1184fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, 1185ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) { 1186ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1187ea8dc4b6Seschrock VDEV_AUX_CORRUPT_DATA); 1188ea8dc4b6Seschrock error = EIO; 1189ea8dc4b6Seschrock goto out; 1190ea8dc4b6Seschrock } 1191fa9e4066Sahrens 119299653d4eSeschrock /* 119399653d4eSeschrock * Load the bit that tells us to use the new accounting function 119499653d4eSeschrock * (raid-z deflation). If we have an older pool, this will not 119599653d4eSeschrock * be present. 119699653d4eSeschrock */ 119799653d4eSeschrock error = zap_lookup(spa->spa_meta_objset, 119899653d4eSeschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 119999653d4eSeschrock sizeof (uint64_t), 1, &spa->spa_deflate); 120099653d4eSeschrock if (error != 0 && error != ENOENT) { 120199653d4eSeschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 120299653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 120399653d4eSeschrock error = EIO; 120499653d4eSeschrock goto out; 120599653d4eSeschrock } 120699653d4eSeschrock 1207fa9e4066Sahrens /* 1208ea8dc4b6Seschrock * Load the persistent error log. If we have an older pool, this will 1209ea8dc4b6Seschrock * not be present. 1210fa9e4066Sahrens */ 1211ea8dc4b6Seschrock error = zap_lookup(spa->spa_meta_objset, 1212ea8dc4b6Seschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST, 1213ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_errlog_last); 1214d80c45e0Sbonwick if (error != 0 && error != ENOENT) { 1215ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1216ea8dc4b6Seschrock VDEV_AUX_CORRUPT_DATA); 1217ea8dc4b6Seschrock error = EIO; 1218ea8dc4b6Seschrock goto out; 1219ea8dc4b6Seschrock } 1220ea8dc4b6Seschrock 1221ea8dc4b6Seschrock error = zap_lookup(spa->spa_meta_objset, 1222ea8dc4b6Seschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB, 1223ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_errlog_scrub); 1224ea8dc4b6Seschrock if (error != 0 && error != ENOENT) { 1225ea8dc4b6Seschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1226ea8dc4b6Seschrock VDEV_AUX_CORRUPT_DATA); 1227ea8dc4b6Seschrock error = EIO; 1228ea8dc4b6Seschrock goto out; 1229ea8dc4b6Seschrock } 1230ea8dc4b6Seschrock 123106eeb2adSek /* 123206eeb2adSek * Load the history object. If we have an older pool, this 123306eeb2adSek * will not be present. 123406eeb2adSek */ 123506eeb2adSek error = zap_lookup(spa->spa_meta_objset, 123606eeb2adSek DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY, 123706eeb2adSek sizeof (uint64_t), 1, &spa->spa_history); 123806eeb2adSek if (error != 0 && error != ENOENT) { 123906eeb2adSek vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 124006eeb2adSek VDEV_AUX_CORRUPT_DATA); 124106eeb2adSek error = EIO; 124206eeb2adSek goto out; 124306eeb2adSek } 124406eeb2adSek 124599653d4eSeschrock /* 124699653d4eSeschrock * Load any hot spares for this pool. 124799653d4eSeschrock */ 124899653d4eSeschrock error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1249fa94a07fSbrendan DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object); 125099653d4eSeschrock if (error != 0 && error != ENOENT) { 125199653d4eSeschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 125299653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 125399653d4eSeschrock error = EIO; 125499653d4eSeschrock goto out; 125599653d4eSeschrock } 125699653d4eSeschrock if (error == 0) { 1257e7437265Sahrens ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 1258fa94a07fSbrendan if (load_nvlist(spa, spa->spa_spares.sav_object, 1259fa94a07fSbrendan &spa->spa_spares.sav_config) != 0) { 126099653d4eSeschrock vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 126199653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 126299653d4eSeschrock error = EIO; 126399653d4eSeschrock goto out; 126499653d4eSeschrock } 126599653d4eSeschrock 126699653d4eSeschrock spa_config_enter(spa, RW_WRITER, FTAG); 126799653d4eSeschrock spa_load_spares(spa); 126899653d4eSeschrock spa_config_exit(spa, FTAG); 126999653d4eSeschrock } 127099653d4eSeschrock 1271fa94a07fSbrendan /* 1272fa94a07fSbrendan * Load any level 2 ARC devices for this pool. 1273fa94a07fSbrendan */ 1274fa94a07fSbrendan error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1275fa94a07fSbrendan DMU_POOL_L2CACHE, sizeof (uint64_t), 1, 1276fa94a07fSbrendan &spa->spa_l2cache.sav_object); 1277fa94a07fSbrendan if (error != 0 && error != ENOENT) { 1278fa94a07fSbrendan vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1279fa94a07fSbrendan VDEV_AUX_CORRUPT_DATA); 1280fa94a07fSbrendan error = EIO; 1281fa94a07fSbrendan goto out; 1282fa94a07fSbrendan } 1283fa94a07fSbrendan if (error == 0) { 1284fa94a07fSbrendan ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 1285fa94a07fSbrendan if (load_nvlist(spa, spa->spa_l2cache.sav_object, 1286fa94a07fSbrendan &spa->spa_l2cache.sav_config) != 0) { 1287fa94a07fSbrendan vdev_set_state(rvd, B_TRUE, 1288fa94a07fSbrendan VDEV_STATE_CANT_OPEN, 1289fa94a07fSbrendan VDEV_AUX_CORRUPT_DATA); 1290fa94a07fSbrendan error = EIO; 1291fa94a07fSbrendan goto out; 1292fa94a07fSbrendan } 1293fa94a07fSbrendan 1294fa94a07fSbrendan spa_config_enter(spa, RW_WRITER, FTAG); 1295fa94a07fSbrendan spa_load_l2cache(spa); 1296fa94a07fSbrendan spa_config_exit(spa, FTAG); 1297fa94a07fSbrendan } 1298fa94a07fSbrendan 1299990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 1300ecd6cf80Smarks 1301b1b8ab34Slling error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1302b1b8ab34Slling DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object); 1303b1b8ab34Slling 1304b1b8ab34Slling if (error && error != ENOENT) { 1305b1b8ab34Slling vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1306b1b8ab34Slling VDEV_AUX_CORRUPT_DATA); 1307b1b8ab34Slling error = EIO; 1308b1b8ab34Slling goto out; 1309b1b8ab34Slling } 1310b1b8ab34Slling 1311b1b8ab34Slling if (error == 0) { 1312b1b8ab34Slling (void) zap_lookup(spa->spa_meta_objset, 1313b1b8ab34Slling spa->spa_pool_props_object, 13143d7072f8Seschrock zpool_prop_to_name(ZPOOL_PROP_BOOTFS), 1315b1b8ab34Slling sizeof (uint64_t), 1, &spa->spa_bootfs); 13163d7072f8Seschrock (void) zap_lookup(spa->spa_meta_objset, 13173d7072f8Seschrock spa->spa_pool_props_object, 13183d7072f8Seschrock zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 13193d7072f8Seschrock sizeof (uint64_t), 1, &autoreplace); 1320ecd6cf80Smarks (void) zap_lookup(spa->spa_meta_objset, 1321ecd6cf80Smarks spa->spa_pool_props_object, 1322ecd6cf80Smarks zpool_prop_to_name(ZPOOL_PROP_DELEGATION), 1323ecd6cf80Smarks sizeof (uint64_t), 1, &spa->spa_delegation); 13240a4e9518Sgw (void) zap_lookup(spa->spa_meta_objset, 13250a4e9518Sgw spa->spa_pool_props_object, 13260a4e9518Sgw zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE), 13270a4e9518Sgw sizeof (uint64_t), 1, &spa->spa_failmode); 1328b1b8ab34Slling } 1329b1b8ab34Slling 13303d7072f8Seschrock /* 13313d7072f8Seschrock * If the 'autoreplace' property is set, then post a resource notifying 13323d7072f8Seschrock * the ZFS DE that it should not issue any faults for unopenable 13333d7072f8Seschrock * devices. We also iterate over the vdevs, and post a sysevent for any 13343d7072f8Seschrock * unopenable vdevs so that the normal autoreplace handler can take 13353d7072f8Seschrock * over. 13363d7072f8Seschrock */ 1337*b01c3b58Seschrock if (autoreplace && state != SPA_LOAD_TRYIMPORT) 13383d7072f8Seschrock spa_check_removed(spa->spa_root_vdev); 13393d7072f8Seschrock 1340ea8dc4b6Seschrock /* 1341560e6e96Seschrock * Load the vdev state for all toplevel vdevs. 1342ea8dc4b6Seschrock */ 1343560e6e96Seschrock vdev_load(rvd); 13440373e76bSbonwick 1345fa9e4066Sahrens /* 1346fa9e4066Sahrens * Propagate the leaf DTLs we just loaded all the way up the tree. 1347fa9e4066Sahrens */ 1348ea8dc4b6Seschrock spa_config_enter(spa, RW_WRITER, FTAG); 1349fa9e4066Sahrens vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 1350ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 1351fa9e4066Sahrens 1352fa9e4066Sahrens /* 1353fa9e4066Sahrens * Check the state of the root vdev. If it can't be opened, it 1354fa9e4066Sahrens * indicates one or more toplevel vdevs are faulted. 1355fa9e4066Sahrens */ 1356ea8dc4b6Seschrock if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 1357ea8dc4b6Seschrock error = ENXIO; 1358ea8dc4b6Seschrock goto out; 1359ea8dc4b6Seschrock } 1360fa9e4066Sahrens 1361ea8dc4b6Seschrock if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) { 13625dabedeeSbonwick dmu_tx_t *tx; 13630373e76bSbonwick int need_update = B_FALSE; 13640373e76bSbonwick int c; 13655dabedeeSbonwick 13660373e76bSbonwick /* 13670373e76bSbonwick * Claim log blocks that haven't been committed yet. 13680373e76bSbonwick * This must all happen in a single txg. 13690373e76bSbonwick */ 13705dabedeeSbonwick tx = dmu_tx_create_assigned(spa_get_dsl(spa), 1371fa9e4066Sahrens spa_first_txg(spa)); 13720b69c2f0Sahrens (void) dmu_objset_find(spa->spa_name, 13730b69c2f0Sahrens zil_claim, tx, DS_FIND_CHILDREN); 1374fa9e4066Sahrens dmu_tx_commit(tx); 1375fa9e4066Sahrens 1376fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 1377fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 1378fa9e4066Sahrens 1379fa9e4066Sahrens /* 1380fa9e4066Sahrens * Wait for all claims to sync. 1381fa9e4066Sahrens */ 1382fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, 0); 13830e34b6a7Sbonwick 13840e34b6a7Sbonwick /* 13850373e76bSbonwick * If the config cache is stale, or we have uninitialized 13860373e76bSbonwick * metaslabs (see spa_vdev_add()), then update the config. 13870e34b6a7Sbonwick */ 13880373e76bSbonwick if (config_cache_txg != spa->spa_config_txg || 13890373e76bSbonwick state == SPA_LOAD_IMPORT) 13900373e76bSbonwick need_update = B_TRUE; 13910373e76bSbonwick 13920373e76bSbonwick for (c = 0; c < rvd->vdev_children; c++) 13930373e76bSbonwick if (rvd->vdev_child[c]->vdev_ms_array == 0) 13940373e76bSbonwick need_update = B_TRUE; 13950e34b6a7Sbonwick 13960e34b6a7Sbonwick /* 13970373e76bSbonwick * Update the config cache asychronously in case we're the 13980373e76bSbonwick * root pool, in which case the config cache isn't writable yet. 13990e34b6a7Sbonwick */ 14000373e76bSbonwick if (need_update) 14010373e76bSbonwick spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 1402fa9e4066Sahrens } 1403fa9e4066Sahrens 1404ea8dc4b6Seschrock error = 0; 1405ea8dc4b6Seschrock out: 140699653d4eSeschrock if (error && error != EBADF) 1407ea8dc4b6Seschrock zfs_ereport_post(FM_EREPORT_ZFS_POOL, spa, NULL, NULL, 0, 0); 1408ea8dc4b6Seschrock spa->spa_load_state = SPA_LOAD_NONE; 1409ea8dc4b6Seschrock spa->spa_ena = 0; 1410ea8dc4b6Seschrock 1411ea8dc4b6Seschrock return (error); 1412fa9e4066Sahrens } 1413fa9e4066Sahrens 1414fa9e4066Sahrens /* 1415fa9e4066Sahrens * Pool Open/Import 1416fa9e4066Sahrens * 1417fa9e4066Sahrens * The import case is identical to an open except that the configuration is sent 1418fa9e4066Sahrens * down from userland, instead of grabbed from the configuration cache. For the 1419fa9e4066Sahrens * case of an open, the pool configuration will exist in the 14203d7072f8Seschrock * POOL_STATE_UNINITIALIZED state. 1421fa9e4066Sahrens * 1422fa9e4066Sahrens * The stats information (gen/count/ustats) is used to gather vdev statistics at 1423fa9e4066Sahrens * the same time open the pool, without having to keep around the spa_t in some 1424fa9e4066Sahrens * ambiguous state. 1425fa9e4066Sahrens */ 1426fa9e4066Sahrens static int 1427fa9e4066Sahrens spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config) 1428fa9e4066Sahrens { 1429fa9e4066Sahrens spa_t *spa; 1430fa9e4066Sahrens int error; 1431fa9e4066Sahrens int loaded = B_FALSE; 1432fa9e4066Sahrens int locked = B_FALSE; 1433fa9e4066Sahrens 1434fa9e4066Sahrens *spapp = NULL; 1435fa9e4066Sahrens 1436fa9e4066Sahrens /* 1437fa9e4066Sahrens * As disgusting as this is, we need to support recursive calls to this 1438fa9e4066Sahrens * function because dsl_dir_open() is called during spa_load(), and ends 1439fa9e4066Sahrens * up calling spa_open() again. The real fix is to figure out how to 1440fa9e4066Sahrens * avoid dsl_dir_open() calling this in the first place. 1441fa9e4066Sahrens */ 1442fa9e4066Sahrens if (mutex_owner(&spa_namespace_lock) != curthread) { 1443fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 1444fa9e4066Sahrens locked = B_TRUE; 1445fa9e4066Sahrens } 1446fa9e4066Sahrens 1447fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 1448fa9e4066Sahrens if (locked) 1449fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1450fa9e4066Sahrens return (ENOENT); 1451fa9e4066Sahrens } 1452fa9e4066Sahrens if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 1453fa9e4066Sahrens 1454fa9e4066Sahrens spa_activate(spa); 1455fa9e4066Sahrens 14560373e76bSbonwick error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE); 1457fa9e4066Sahrens 1458fa9e4066Sahrens if (error == EBADF) { 1459fa9e4066Sahrens /* 1460560e6e96Seschrock * If vdev_validate() returns failure (indicated by 1461560e6e96Seschrock * EBADF), it indicates that one of the vdevs indicates 1462560e6e96Seschrock * that the pool has been exported or destroyed. If 1463560e6e96Seschrock * this is the case, the config cache is out of sync and 1464560e6e96Seschrock * we should remove the pool from the namespace. 1465fa9e4066Sahrens */ 146699653d4eSeschrock zfs_post_ok(spa, NULL); 1467fa9e4066Sahrens spa_unload(spa); 1468fa9e4066Sahrens spa_deactivate(spa); 1469fa9e4066Sahrens spa_remove(spa); 1470fa9e4066Sahrens spa_config_sync(); 1471fa9e4066Sahrens if (locked) 1472fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1473fa9e4066Sahrens return (ENOENT); 1474ea8dc4b6Seschrock } 1475ea8dc4b6Seschrock 1476ea8dc4b6Seschrock if (error) { 1477fa9e4066Sahrens /* 1478fa9e4066Sahrens * We can't open the pool, but we still have useful 1479fa9e4066Sahrens * information: the state of each vdev after the 1480fa9e4066Sahrens * attempted vdev_open(). Return this to the user. 1481fa9e4066Sahrens */ 14820373e76bSbonwick if (config != NULL && spa->spa_root_vdev != NULL) { 14830373e76bSbonwick spa_config_enter(spa, RW_READER, FTAG); 1484fa9e4066Sahrens *config = spa_config_generate(spa, NULL, -1ULL, 1485fa9e4066Sahrens B_TRUE); 14860373e76bSbonwick spa_config_exit(spa, FTAG); 14870373e76bSbonwick } 1488fa9e4066Sahrens spa_unload(spa); 1489fa9e4066Sahrens spa_deactivate(spa); 1490ea8dc4b6Seschrock spa->spa_last_open_failed = B_TRUE; 1491fa9e4066Sahrens if (locked) 1492fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1493fa9e4066Sahrens *spapp = NULL; 1494fa9e4066Sahrens return (error); 1495ea8dc4b6Seschrock } else { 1496ea8dc4b6Seschrock zfs_post_ok(spa, NULL); 1497ea8dc4b6Seschrock spa->spa_last_open_failed = B_FALSE; 1498fa9e4066Sahrens } 1499fa9e4066Sahrens 1500fa9e4066Sahrens loaded = B_TRUE; 1501fa9e4066Sahrens } 1502fa9e4066Sahrens 1503fa9e4066Sahrens spa_open_ref(spa, tag); 15043d7072f8Seschrock 15053d7072f8Seschrock /* 15063d7072f8Seschrock * If we just loaded the pool, resilver anything that's out of date. 15073d7072f8Seschrock */ 15083d7072f8Seschrock if (loaded && (spa_mode & FWRITE)) 15093d7072f8Seschrock VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); 15103d7072f8Seschrock 1511fa9e4066Sahrens if (locked) 1512fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1513fa9e4066Sahrens 1514fa9e4066Sahrens *spapp = spa; 1515fa9e4066Sahrens 1516fa9e4066Sahrens if (config != NULL) { 1517ea8dc4b6Seschrock spa_config_enter(spa, RW_READER, FTAG); 1518fa9e4066Sahrens *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 1519ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 1520fa9e4066Sahrens } 1521fa9e4066Sahrens 1522fa9e4066Sahrens return (0); 1523fa9e4066Sahrens } 1524fa9e4066Sahrens 1525fa9e4066Sahrens int 1526fa9e4066Sahrens spa_open(const char *name, spa_t **spapp, void *tag) 1527fa9e4066Sahrens { 1528fa9e4066Sahrens return (spa_open_common(name, spapp, tag, NULL)); 1529fa9e4066Sahrens } 1530fa9e4066Sahrens 1531ea8dc4b6Seschrock /* 1532ea8dc4b6Seschrock * Lookup the given spa_t, incrementing the inject count in the process, 1533ea8dc4b6Seschrock * preventing it from being exported or destroyed. 1534ea8dc4b6Seschrock */ 1535ea8dc4b6Seschrock spa_t * 1536ea8dc4b6Seschrock spa_inject_addref(char *name) 1537ea8dc4b6Seschrock { 1538ea8dc4b6Seschrock spa_t *spa; 1539ea8dc4b6Seschrock 1540ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 1541ea8dc4b6Seschrock if ((spa = spa_lookup(name)) == NULL) { 1542ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1543ea8dc4b6Seschrock return (NULL); 1544ea8dc4b6Seschrock } 1545ea8dc4b6Seschrock spa->spa_inject_ref++; 1546ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1547ea8dc4b6Seschrock 1548ea8dc4b6Seschrock return (spa); 1549ea8dc4b6Seschrock } 1550ea8dc4b6Seschrock 1551ea8dc4b6Seschrock void 1552ea8dc4b6Seschrock spa_inject_delref(spa_t *spa) 1553ea8dc4b6Seschrock { 1554ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 1555ea8dc4b6Seschrock spa->spa_inject_ref--; 1556ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1557ea8dc4b6Seschrock } 1558ea8dc4b6Seschrock 1559fa94a07fSbrendan /* 1560fa94a07fSbrendan * Add spares device information to the nvlist. 1561fa94a07fSbrendan */ 156299653d4eSeschrock static void 156399653d4eSeschrock spa_add_spares(spa_t *spa, nvlist_t *config) 156499653d4eSeschrock { 156599653d4eSeschrock nvlist_t **spares; 156699653d4eSeschrock uint_t i, nspares; 156799653d4eSeschrock nvlist_t *nvroot; 156899653d4eSeschrock uint64_t guid; 156999653d4eSeschrock vdev_stat_t *vs; 157099653d4eSeschrock uint_t vsc; 157139c23413Seschrock uint64_t pool; 157299653d4eSeschrock 1573fa94a07fSbrendan if (spa->spa_spares.sav_count == 0) 157499653d4eSeschrock return; 157599653d4eSeschrock 157699653d4eSeschrock VERIFY(nvlist_lookup_nvlist(config, 157799653d4eSeschrock ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 1578fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 157999653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 158099653d4eSeschrock if (nspares != 0) { 158199653d4eSeschrock VERIFY(nvlist_add_nvlist_array(nvroot, 158299653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 158399653d4eSeschrock VERIFY(nvlist_lookup_nvlist_array(nvroot, 158499653d4eSeschrock ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 158599653d4eSeschrock 158699653d4eSeschrock /* 158799653d4eSeschrock * Go through and find any spares which have since been 158899653d4eSeschrock * repurposed as an active spare. If this is the case, update 158999653d4eSeschrock * their status appropriately. 159099653d4eSeschrock */ 159199653d4eSeschrock for (i = 0; i < nspares; i++) { 159299653d4eSeschrock VERIFY(nvlist_lookup_uint64(spares[i], 159399653d4eSeschrock ZPOOL_CONFIG_GUID, &guid) == 0); 159439c23413Seschrock if (spa_spare_exists(guid, &pool) && pool != 0ULL) { 159599653d4eSeschrock VERIFY(nvlist_lookup_uint64_array( 159699653d4eSeschrock spares[i], ZPOOL_CONFIG_STATS, 159799653d4eSeschrock (uint64_t **)&vs, &vsc) == 0); 159899653d4eSeschrock vs->vs_state = VDEV_STATE_CANT_OPEN; 159999653d4eSeschrock vs->vs_aux = VDEV_AUX_SPARED; 160099653d4eSeschrock } 160199653d4eSeschrock } 160299653d4eSeschrock } 160399653d4eSeschrock } 160499653d4eSeschrock 1605fa94a07fSbrendan /* 1606fa94a07fSbrendan * Add l2cache device information to the nvlist, including vdev stats. 1607fa94a07fSbrendan */ 1608fa94a07fSbrendan static void 1609fa94a07fSbrendan spa_add_l2cache(spa_t *spa, nvlist_t *config) 1610fa94a07fSbrendan { 1611fa94a07fSbrendan nvlist_t **l2cache; 1612fa94a07fSbrendan uint_t i, j, nl2cache; 1613fa94a07fSbrendan nvlist_t *nvroot; 1614fa94a07fSbrendan uint64_t guid; 1615fa94a07fSbrendan vdev_t *vd; 1616fa94a07fSbrendan vdev_stat_t *vs; 1617fa94a07fSbrendan uint_t vsc; 1618fa94a07fSbrendan 1619fa94a07fSbrendan if (spa->spa_l2cache.sav_count == 0) 1620fa94a07fSbrendan return; 1621fa94a07fSbrendan 1622fa94a07fSbrendan spa_config_enter(spa, RW_READER, FTAG); 1623fa94a07fSbrendan 1624fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist(config, 1625fa94a07fSbrendan ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 1626fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 1627fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1628fa94a07fSbrendan if (nl2cache != 0) { 1629fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, 1630fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 1631fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(nvroot, 1632fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1633fa94a07fSbrendan 1634fa94a07fSbrendan /* 1635fa94a07fSbrendan * Update level 2 cache device stats. 1636fa94a07fSbrendan */ 1637fa94a07fSbrendan 1638fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 1639fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], 1640fa94a07fSbrendan ZPOOL_CONFIG_GUID, &guid) == 0); 1641fa94a07fSbrendan 1642fa94a07fSbrendan vd = NULL; 1643fa94a07fSbrendan for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 1644fa94a07fSbrendan if (guid == 1645fa94a07fSbrendan spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 1646fa94a07fSbrendan vd = spa->spa_l2cache.sav_vdevs[j]; 1647fa94a07fSbrendan break; 1648fa94a07fSbrendan } 1649fa94a07fSbrendan } 1650fa94a07fSbrendan ASSERT(vd != NULL); 1651fa94a07fSbrendan 1652fa94a07fSbrendan VERIFY(nvlist_lookup_uint64_array(l2cache[i], 1653fa94a07fSbrendan ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0); 1654fa94a07fSbrendan vdev_get_stats(vd, vs); 1655fa94a07fSbrendan } 1656fa94a07fSbrendan } 1657fa94a07fSbrendan 1658fa94a07fSbrendan spa_config_exit(spa, FTAG); 1659fa94a07fSbrendan } 1660fa94a07fSbrendan 1661fa9e4066Sahrens int 1662ea8dc4b6Seschrock spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) 1663fa9e4066Sahrens { 1664fa9e4066Sahrens int error; 1665fa9e4066Sahrens spa_t *spa; 1666fa9e4066Sahrens 1667fa9e4066Sahrens *config = NULL; 1668fa9e4066Sahrens error = spa_open_common(name, &spa, FTAG, config); 1669fa9e4066Sahrens 167099653d4eSeschrock if (spa && *config != NULL) { 1671ea8dc4b6Seschrock VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, 1672ea8dc4b6Seschrock spa_get_errlog_size(spa)) == 0); 1673ea8dc4b6Seschrock 167499653d4eSeschrock spa_add_spares(spa, *config); 1675fa94a07fSbrendan spa_add_l2cache(spa, *config); 167699653d4eSeschrock } 167799653d4eSeschrock 1678ea8dc4b6Seschrock /* 1679ea8dc4b6Seschrock * We want to get the alternate root even for faulted pools, so we cheat 1680ea8dc4b6Seschrock * and call spa_lookup() directly. 1681ea8dc4b6Seschrock */ 1682ea8dc4b6Seschrock if (altroot) { 1683ea8dc4b6Seschrock if (spa == NULL) { 1684ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 1685ea8dc4b6Seschrock spa = spa_lookup(name); 1686ea8dc4b6Seschrock if (spa) 1687ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 1688ea8dc4b6Seschrock else 1689ea8dc4b6Seschrock altroot[0] = '\0'; 1690ea8dc4b6Seschrock spa = NULL; 1691ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 1692ea8dc4b6Seschrock } else { 1693ea8dc4b6Seschrock spa_altroot(spa, altroot, buflen); 1694ea8dc4b6Seschrock } 1695ea8dc4b6Seschrock } 1696ea8dc4b6Seschrock 1697fa9e4066Sahrens if (spa != NULL) 1698fa9e4066Sahrens spa_close(spa, FTAG); 1699fa9e4066Sahrens 1700fa9e4066Sahrens return (error); 1701fa9e4066Sahrens } 1702fa9e4066Sahrens 170399653d4eSeschrock /* 1704fa94a07fSbrendan * Validate that the auxiliary device array is well formed. We must have an 1705fa94a07fSbrendan * array of nvlists, each which describes a valid leaf vdev. If this is an 1706fa94a07fSbrendan * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 1707fa94a07fSbrendan * specified, as long as they are well-formed. 170899653d4eSeschrock */ 170999653d4eSeschrock static int 1710fa94a07fSbrendan spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 1711fa94a07fSbrendan spa_aux_vdev_t *sav, const char *config, uint64_t version, 1712fa94a07fSbrendan vdev_labeltype_t label) 171399653d4eSeschrock { 1714fa94a07fSbrendan nvlist_t **dev; 1715fa94a07fSbrendan uint_t i, ndev; 171699653d4eSeschrock vdev_t *vd; 171799653d4eSeschrock int error; 171899653d4eSeschrock 171999653d4eSeschrock /* 1720fa94a07fSbrendan * It's acceptable to have no devs specified. 172199653d4eSeschrock */ 1722fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 172399653d4eSeschrock return (0); 172499653d4eSeschrock 1725fa94a07fSbrendan if (ndev == 0) 172699653d4eSeschrock return (EINVAL); 172799653d4eSeschrock 172899653d4eSeschrock /* 1729fa94a07fSbrendan * Make sure the pool is formatted with a version that supports this 1730fa94a07fSbrendan * device type. 173199653d4eSeschrock */ 1732fa94a07fSbrendan if (spa_version(spa) < version) 173399653d4eSeschrock return (ENOTSUP); 173499653d4eSeschrock 173539c23413Seschrock /* 1736fa94a07fSbrendan * Set the pending device list so we correctly handle device in-use 173739c23413Seschrock * checking. 173839c23413Seschrock */ 1739fa94a07fSbrendan sav->sav_pending = dev; 1740fa94a07fSbrendan sav->sav_npending = ndev; 174139c23413Seschrock 1742fa94a07fSbrendan for (i = 0; i < ndev; i++) { 1743fa94a07fSbrendan if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 174499653d4eSeschrock mode)) != 0) 174539c23413Seschrock goto out; 174699653d4eSeschrock 174799653d4eSeschrock if (!vd->vdev_ops->vdev_op_leaf) { 174899653d4eSeschrock vdev_free(vd); 174939c23413Seschrock error = EINVAL; 175039c23413Seschrock goto out; 175199653d4eSeschrock } 175299653d4eSeschrock 1753fa94a07fSbrendan /* 1754fa94a07fSbrendan * The L2ARC currently only supports disk devices. 1755fa94a07fSbrendan */ 1756fa94a07fSbrendan if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 1757fa94a07fSbrendan strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 1758fa94a07fSbrendan error = ENOTBLK; 1759fa94a07fSbrendan goto out; 1760fa94a07fSbrendan } 1761fa94a07fSbrendan 176299653d4eSeschrock vd->vdev_top = vd; 176399653d4eSeschrock 176439c23413Seschrock if ((error = vdev_open(vd)) == 0 && 1765fa94a07fSbrendan (error = vdev_label_init(vd, crtxg, label)) == 0) { 1766fa94a07fSbrendan VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 176739c23413Seschrock vd->vdev_guid) == 0); 176839c23413Seschrock } 176999653d4eSeschrock 177099653d4eSeschrock vdev_free(vd); 177139c23413Seschrock 1772fa94a07fSbrendan if (error && 1773fa94a07fSbrendan (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 177439c23413Seschrock goto out; 177539c23413Seschrock else 177639c23413Seschrock error = 0; 177799653d4eSeschrock } 177899653d4eSeschrock 177939c23413Seschrock out: 1780fa94a07fSbrendan sav->sav_pending = NULL; 1781fa94a07fSbrendan sav->sav_npending = 0; 178239c23413Seschrock return (error); 178399653d4eSeschrock } 178499653d4eSeschrock 1785fa94a07fSbrendan static int 1786fa94a07fSbrendan spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 1787fa94a07fSbrendan { 1788fa94a07fSbrendan int error; 1789fa94a07fSbrendan 1790fa94a07fSbrendan if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 1791fa94a07fSbrendan &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 1792fa94a07fSbrendan VDEV_LABEL_SPARE)) != 0) { 1793fa94a07fSbrendan return (error); 1794fa94a07fSbrendan } 1795fa94a07fSbrendan 1796fa94a07fSbrendan return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 1797fa94a07fSbrendan &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 1798fa94a07fSbrendan VDEV_LABEL_L2CACHE)); 1799fa94a07fSbrendan } 1800fa94a07fSbrendan 1801fa94a07fSbrendan static void 1802fa94a07fSbrendan spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 1803fa94a07fSbrendan const char *config) 1804fa94a07fSbrendan { 1805fa94a07fSbrendan int i; 1806fa94a07fSbrendan 1807fa94a07fSbrendan if (sav->sav_config != NULL) { 1808fa94a07fSbrendan nvlist_t **olddevs; 1809fa94a07fSbrendan uint_t oldndevs; 1810fa94a07fSbrendan nvlist_t **newdevs; 1811fa94a07fSbrendan 1812fa94a07fSbrendan /* 1813fa94a07fSbrendan * Generate new dev list by concatentating with the 1814fa94a07fSbrendan * current dev list. 1815fa94a07fSbrendan */ 1816fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 1817fa94a07fSbrendan &olddevs, &oldndevs) == 0); 1818fa94a07fSbrendan 1819fa94a07fSbrendan newdevs = kmem_alloc(sizeof (void *) * 1820fa94a07fSbrendan (ndevs + oldndevs), KM_SLEEP); 1821fa94a07fSbrendan for (i = 0; i < oldndevs; i++) 1822fa94a07fSbrendan VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 1823fa94a07fSbrendan KM_SLEEP) == 0); 1824fa94a07fSbrendan for (i = 0; i < ndevs; i++) 1825fa94a07fSbrendan VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 1826fa94a07fSbrendan KM_SLEEP) == 0); 1827fa94a07fSbrendan 1828fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, config, 1829fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 1830fa94a07fSbrendan 1831fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1832fa94a07fSbrendan config, newdevs, ndevs + oldndevs) == 0); 1833fa94a07fSbrendan for (i = 0; i < oldndevs + ndevs; i++) 1834fa94a07fSbrendan nvlist_free(newdevs[i]); 1835fa94a07fSbrendan kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 1836fa94a07fSbrendan } else { 1837fa94a07fSbrendan /* 1838fa94a07fSbrendan * Generate a new dev list. 1839fa94a07fSbrendan */ 1840fa94a07fSbrendan VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 1841fa94a07fSbrendan KM_SLEEP) == 0); 1842fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 1843fa94a07fSbrendan devs, ndevs) == 0); 1844fa94a07fSbrendan } 1845fa94a07fSbrendan } 1846fa94a07fSbrendan 1847fa94a07fSbrendan /* 1848fa94a07fSbrendan * Stop and drop level 2 ARC devices 1849fa94a07fSbrendan */ 1850fa94a07fSbrendan void 1851fa94a07fSbrendan spa_l2cache_drop(spa_t *spa) 1852fa94a07fSbrendan { 1853fa94a07fSbrendan vdev_t *vd; 1854fa94a07fSbrendan int i; 1855fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_l2cache; 1856fa94a07fSbrendan 1857fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) { 1858fa94a07fSbrendan uint64_t pool; 1859fa94a07fSbrendan 1860fa94a07fSbrendan vd = sav->sav_vdevs[i]; 1861fa94a07fSbrendan ASSERT(vd != NULL); 1862fa94a07fSbrendan 1863fa94a07fSbrendan if (spa_mode & FWRITE && 1864fa94a07fSbrendan spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL) { 1865fa94a07fSbrendan l2arc_remove_vdev(vd); 1866fa94a07fSbrendan } 1867fa94a07fSbrendan if (vd->vdev_isl2cache) 1868fa94a07fSbrendan spa_l2cache_remove(vd); 1869fa94a07fSbrendan vdev_clear_stats(vd); 1870fa94a07fSbrendan (void) vdev_close(vd); 1871fa94a07fSbrendan } 1872fa94a07fSbrendan } 1873fa94a07fSbrendan 1874fa9e4066Sahrens /* 1875fa9e4066Sahrens * Pool Creation 1876fa9e4066Sahrens */ 1877fa9e4066Sahrens int 1878990b4856Slling spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 1879228975ccSek const char *history_str) 1880fa9e4066Sahrens { 1881fa9e4066Sahrens spa_t *spa; 1882990b4856Slling char *altroot = NULL; 18830373e76bSbonwick vdev_t *rvd; 1884fa9e4066Sahrens dsl_pool_t *dp; 1885fa9e4066Sahrens dmu_tx_t *tx; 188699653d4eSeschrock int c, error = 0; 1887fa9e4066Sahrens uint64_t txg = TXG_INITIAL; 1888fa94a07fSbrendan nvlist_t **spares, **l2cache; 1889fa94a07fSbrendan uint_t nspares, nl2cache; 1890990b4856Slling uint64_t version; 1891fa9e4066Sahrens 1892fa9e4066Sahrens /* 1893fa9e4066Sahrens * If this pool already exists, return failure. 1894fa9e4066Sahrens */ 1895fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 1896fa9e4066Sahrens if (spa_lookup(pool) != NULL) { 1897fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1898fa9e4066Sahrens return (EEXIST); 1899fa9e4066Sahrens } 1900fa9e4066Sahrens 1901fa9e4066Sahrens /* 1902fa9e4066Sahrens * Allocate a new spa_t structure. 1903fa9e4066Sahrens */ 1904990b4856Slling (void) nvlist_lookup_string(props, 1905990b4856Slling zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 19060373e76bSbonwick spa = spa_add(pool, altroot); 1907fa9e4066Sahrens spa_activate(spa); 1908fa9e4066Sahrens 1909fa9e4066Sahrens spa->spa_uberblock.ub_txg = txg - 1; 1910990b4856Slling 1911990b4856Slling if (props && (error = spa_prop_validate(spa, props))) { 1912990b4856Slling spa_unload(spa); 1913990b4856Slling spa_deactivate(spa); 1914990b4856Slling spa_remove(spa); 1915990b4856Slling return (error); 1916990b4856Slling } 1917990b4856Slling 1918990b4856Slling if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), 1919990b4856Slling &version) != 0) 1920990b4856Slling version = SPA_VERSION; 1921990b4856Slling ASSERT(version <= SPA_VERSION); 1922990b4856Slling spa->spa_uberblock.ub_version = version; 1923fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 1924fa9e4066Sahrens 19250373e76bSbonwick /* 19260373e76bSbonwick * Create the root vdev. 19270373e76bSbonwick */ 19280373e76bSbonwick spa_config_enter(spa, RW_WRITER, FTAG); 19290373e76bSbonwick 193099653d4eSeschrock error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 19310373e76bSbonwick 193299653d4eSeschrock ASSERT(error != 0 || rvd != NULL); 193399653d4eSeschrock ASSERT(error != 0 || spa->spa_root_vdev == rvd); 19340373e76bSbonwick 193599653d4eSeschrock if (error == 0 && rvd->vdev_children == 0) 19360373e76bSbonwick error = EINVAL; 193799653d4eSeschrock 193899653d4eSeschrock if (error == 0 && 193999653d4eSeschrock (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 1940fa94a07fSbrendan (error = spa_validate_aux(spa, nvroot, txg, 194199653d4eSeschrock VDEV_ALLOC_ADD)) == 0) { 194299653d4eSeschrock for (c = 0; c < rvd->vdev_children; c++) 194399653d4eSeschrock vdev_init(rvd->vdev_child[c], txg); 194499653d4eSeschrock vdev_config_dirty(rvd); 19450373e76bSbonwick } 19460373e76bSbonwick 19470373e76bSbonwick spa_config_exit(spa, FTAG); 1948fa9e4066Sahrens 194999653d4eSeschrock if (error != 0) { 1950fa9e4066Sahrens spa_unload(spa); 1951fa9e4066Sahrens spa_deactivate(spa); 1952fa9e4066Sahrens spa_remove(spa); 1953fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 1954fa9e4066Sahrens return (error); 1955fa9e4066Sahrens } 1956fa9e4066Sahrens 195799653d4eSeschrock /* 195899653d4eSeschrock * Get the list of spares, if specified. 195999653d4eSeschrock */ 196099653d4eSeschrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 196199653d4eSeschrock &spares, &nspares) == 0) { 1962fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 196399653d4eSeschrock KM_SLEEP) == 0); 1964fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 196599653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 196699653d4eSeschrock spa_config_enter(spa, RW_WRITER, FTAG); 196799653d4eSeschrock spa_load_spares(spa); 196899653d4eSeschrock spa_config_exit(spa, FTAG); 1969fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 1970fa94a07fSbrendan } 1971fa94a07fSbrendan 1972fa94a07fSbrendan /* 1973fa94a07fSbrendan * Get the list of level 2 cache devices, if specified. 1974fa94a07fSbrendan */ 1975fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 1976fa94a07fSbrendan &l2cache, &nl2cache) == 0) { 1977fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 1978fa94a07fSbrendan NV_UNIQUE_NAME, KM_SLEEP) == 0); 1979fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 1980fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 1981fa94a07fSbrendan spa_config_enter(spa, RW_WRITER, FTAG); 1982fa94a07fSbrendan spa_load_l2cache(spa); 1983fa94a07fSbrendan spa_config_exit(spa, FTAG); 1984fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 198599653d4eSeschrock } 198699653d4eSeschrock 1987fa9e4066Sahrens spa->spa_dsl_pool = dp = dsl_pool_create(spa, txg); 1988fa9e4066Sahrens spa->spa_meta_objset = dp->dp_meta_objset; 1989fa9e4066Sahrens 1990fa9e4066Sahrens tx = dmu_tx_create_assigned(dp, txg); 1991fa9e4066Sahrens 1992fa9e4066Sahrens /* 1993fa9e4066Sahrens * Create the pool config object. 1994fa9e4066Sahrens */ 1995fa9e4066Sahrens spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 1996fa9e4066Sahrens DMU_OT_PACKED_NVLIST, 1 << 14, 1997fa9e4066Sahrens DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 1998fa9e4066Sahrens 1999ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 2000fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 2001ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 2002ea8dc4b6Seschrock cmn_err(CE_PANIC, "failed to add pool config"); 2003ea8dc4b6Seschrock } 2004fa9e4066Sahrens 2005990b4856Slling /* Newly created pools with the right version are always deflated. */ 2006990b4856Slling if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 2007990b4856Slling spa->spa_deflate = TRUE; 2008990b4856Slling if (zap_add(spa->spa_meta_objset, 2009990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 2010990b4856Slling sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 2011990b4856Slling cmn_err(CE_PANIC, "failed to add deflate"); 2012990b4856Slling } 201399653d4eSeschrock } 201499653d4eSeschrock 2015fa9e4066Sahrens /* 2016fa9e4066Sahrens * Create the deferred-free bplist object. Turn off compression 2017fa9e4066Sahrens * because sync-to-convergence takes longer if the blocksize 2018fa9e4066Sahrens * keeps changing. 2019fa9e4066Sahrens */ 2020fa9e4066Sahrens spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset, 2021fa9e4066Sahrens 1 << 14, tx); 2022fa9e4066Sahrens dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 2023fa9e4066Sahrens ZIO_COMPRESS_OFF, tx); 2024fa9e4066Sahrens 2025ea8dc4b6Seschrock if (zap_add(spa->spa_meta_objset, 2026fa9e4066Sahrens DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, 2027ea8dc4b6Seschrock sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) { 2028ea8dc4b6Seschrock cmn_err(CE_PANIC, "failed to add bplist"); 2029ea8dc4b6Seschrock } 2030fa9e4066Sahrens 203106eeb2adSek /* 203206eeb2adSek * Create the pool's history object. 203306eeb2adSek */ 2034990b4856Slling if (version >= SPA_VERSION_ZPOOL_HISTORY) 2035990b4856Slling spa_history_create_obj(spa, tx); 2036990b4856Slling 2037990b4856Slling /* 2038990b4856Slling * Set pool properties. 2039990b4856Slling */ 2040990b4856Slling spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 2041990b4856Slling spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 20420a4e9518Sgw spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 2043990b4856Slling if (props) 2044990b4856Slling spa_sync_props(spa, props, CRED(), tx); 204506eeb2adSek 2046fa9e4066Sahrens dmu_tx_commit(tx); 2047fa9e4066Sahrens 2048fa9e4066Sahrens spa->spa_sync_on = B_TRUE; 2049fa9e4066Sahrens txg_sync_start(spa->spa_dsl_pool); 2050fa9e4066Sahrens 2051fa9e4066Sahrens /* 2052fa9e4066Sahrens * We explicitly wait for the first transaction to complete so that our 2053fa9e4066Sahrens * bean counters are appropriately updated. 2054fa9e4066Sahrens */ 2055fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, txg); 2056fa9e4066Sahrens 2057fa9e4066Sahrens spa_config_sync(); 2058fa9e4066Sahrens 2059990b4856Slling if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL) 2060228975ccSek (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE); 2061228975ccSek 2062fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2063fa9e4066Sahrens 2064fa9e4066Sahrens return (0); 2065fa9e4066Sahrens } 2066fa9e4066Sahrens 2067fa9e4066Sahrens /* 2068fa9e4066Sahrens * Import the given pool into the system. We set up the necessary spa_t and 2069fa9e4066Sahrens * then call spa_load() to do the dirty work. 2070fa9e4066Sahrens */ 2071fa9e4066Sahrens int 2072990b4856Slling spa_import(const char *pool, nvlist_t *config, nvlist_t *props) 2073fa9e4066Sahrens { 2074fa9e4066Sahrens spa_t *spa; 2075990b4856Slling char *altroot = NULL; 2076fa9e4066Sahrens int error; 207799653d4eSeschrock nvlist_t *nvroot; 2078fa94a07fSbrendan nvlist_t **spares, **l2cache; 2079fa94a07fSbrendan uint_t nspares, nl2cache; 2080fa9e4066Sahrens 2081fa9e4066Sahrens /* 2082fa9e4066Sahrens * If a pool with this name exists, return failure. 2083fa9e4066Sahrens */ 2084fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 2085fa9e4066Sahrens if (spa_lookup(pool) != NULL) { 2086fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2087fa9e4066Sahrens return (EEXIST); 2088fa9e4066Sahrens } 2089fa9e4066Sahrens 2090fa9e4066Sahrens /* 20910373e76bSbonwick * Create and initialize the spa structure. 2092fa9e4066Sahrens */ 2093990b4856Slling (void) nvlist_lookup_string(props, 2094990b4856Slling zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 20950373e76bSbonwick spa = spa_add(pool, altroot); 2096fa9e4066Sahrens spa_activate(spa); 2097fa9e4066Sahrens 20985dabedeeSbonwick /* 20990373e76bSbonwick * Pass off the heavy lifting to spa_load(). 2100ecc2d604Sbonwick * Pass TRUE for mosconfig because the user-supplied config 2101ecc2d604Sbonwick * is actually the one to trust when doing an import. 21025dabedeeSbonwick */ 2103ecc2d604Sbonwick error = spa_load(spa, config, SPA_LOAD_IMPORT, B_TRUE); 2104fa9e4066Sahrens 210599653d4eSeschrock spa_config_enter(spa, RW_WRITER, FTAG); 210699653d4eSeschrock /* 210799653d4eSeschrock * Toss any existing sparelist, as it doesn't have any validity anymore, 210899653d4eSeschrock * and conflicts with spa_has_spare(). 210999653d4eSeschrock */ 2110fa94a07fSbrendan if (spa->spa_spares.sav_config) { 2111fa94a07fSbrendan nvlist_free(spa->spa_spares.sav_config); 2112fa94a07fSbrendan spa->spa_spares.sav_config = NULL; 211399653d4eSeschrock spa_load_spares(spa); 211499653d4eSeschrock } 2115fa94a07fSbrendan if (spa->spa_l2cache.sav_config) { 2116fa94a07fSbrendan nvlist_free(spa->spa_l2cache.sav_config); 2117fa94a07fSbrendan spa->spa_l2cache.sav_config = NULL; 2118fa94a07fSbrendan spa_load_l2cache(spa); 2119fa94a07fSbrendan } 212099653d4eSeschrock 212199653d4eSeschrock VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 212299653d4eSeschrock &nvroot) == 0); 2123fa94a07fSbrendan if (error == 0) 2124fa94a07fSbrendan error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE); 2125fa94a07fSbrendan if (error == 0) 2126fa94a07fSbrendan error = spa_validate_aux(spa, nvroot, -1ULL, 2127fa94a07fSbrendan VDEV_ALLOC_L2CACHE); 212899653d4eSeschrock spa_config_exit(spa, FTAG); 212999653d4eSeschrock 2130990b4856Slling if (error != 0 || (props && (error = spa_prop_set(spa, props)))) { 2131fa9e4066Sahrens spa_unload(spa); 2132fa9e4066Sahrens spa_deactivate(spa); 2133fa9e4066Sahrens spa_remove(spa); 2134fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2135fa9e4066Sahrens return (error); 2136fa9e4066Sahrens } 2137fa9e4066Sahrens 213899653d4eSeschrock /* 2139fa94a07fSbrendan * Override any spares and level 2 cache devices as specified by 2140fa94a07fSbrendan * the user, as these may have correct device names/devids, etc. 214199653d4eSeschrock */ 214299653d4eSeschrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 214399653d4eSeschrock &spares, &nspares) == 0) { 2144fa94a07fSbrendan if (spa->spa_spares.sav_config) 2145fa94a07fSbrendan VERIFY(nvlist_remove(spa->spa_spares.sav_config, 214699653d4eSeschrock ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 214799653d4eSeschrock else 2148fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 214999653d4eSeschrock NV_UNIQUE_NAME, KM_SLEEP) == 0); 2150fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 215199653d4eSeschrock ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 215299653d4eSeschrock spa_config_enter(spa, RW_WRITER, FTAG); 215399653d4eSeschrock spa_load_spares(spa); 215499653d4eSeschrock spa_config_exit(spa, FTAG); 2155fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 2156fa94a07fSbrendan } 2157fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 2158fa94a07fSbrendan &l2cache, &nl2cache) == 0) { 2159fa94a07fSbrendan if (spa->spa_l2cache.sav_config) 2160fa94a07fSbrendan VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 2161fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 2162fa94a07fSbrendan else 2163fa94a07fSbrendan VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 2164fa94a07fSbrendan NV_UNIQUE_NAME, KM_SLEEP) == 0); 2165fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 2166fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 2167fa94a07fSbrendan spa_config_enter(spa, RW_WRITER, FTAG); 2168fa94a07fSbrendan spa_load_l2cache(spa); 2169fa94a07fSbrendan spa_config_exit(spa, FTAG); 2170fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 217199653d4eSeschrock } 217299653d4eSeschrock 21730373e76bSbonwick /* 21740373e76bSbonwick * Update the config cache to include the newly-imported pool. 21750373e76bSbonwick */ 2176de6628f0Sck if (spa_mode & FWRITE) 2177de6628f0Sck spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 21780373e76bSbonwick 2179fa9e4066Sahrens /* 2180fa9e4066Sahrens * Resilver anything that's out of date. 2181fa9e4066Sahrens */ 2182fa9e4066Sahrens if (spa_mode & FWRITE) 2183fa9e4066Sahrens VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); 2184fa9e4066Sahrens 21853d7072f8Seschrock mutex_exit(&spa_namespace_lock); 21863d7072f8Seschrock 2187fa9e4066Sahrens return (0); 2188fa9e4066Sahrens } 2189fa9e4066Sahrens 2190fa9e4066Sahrens /* 2191fa9e4066Sahrens * This (illegal) pool name is used when temporarily importing a spa_t in order 2192fa9e4066Sahrens * to get the vdev stats associated with the imported devices. 2193fa9e4066Sahrens */ 2194fa9e4066Sahrens #define TRYIMPORT_NAME "$import" 2195fa9e4066Sahrens 2196fa9e4066Sahrens nvlist_t * 2197fa9e4066Sahrens spa_tryimport(nvlist_t *tryconfig) 2198fa9e4066Sahrens { 2199fa9e4066Sahrens nvlist_t *config = NULL; 2200fa9e4066Sahrens char *poolname; 2201fa9e4066Sahrens spa_t *spa; 2202fa9e4066Sahrens uint64_t state; 2203fa9e4066Sahrens 2204fa9e4066Sahrens if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 2205fa9e4066Sahrens return (NULL); 2206fa9e4066Sahrens 2207fa9e4066Sahrens if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 2208fa9e4066Sahrens return (NULL); 2209fa9e4066Sahrens 2210fa9e4066Sahrens /* 22110373e76bSbonwick * Create and initialize the spa structure. 2212fa9e4066Sahrens */ 22130373e76bSbonwick mutex_enter(&spa_namespace_lock); 22140373e76bSbonwick spa = spa_add(TRYIMPORT_NAME, NULL); 2215fa9e4066Sahrens spa_activate(spa); 2216fa9e4066Sahrens 2217fa9e4066Sahrens /* 22180373e76bSbonwick * Pass off the heavy lifting to spa_load(). 2219ecc2d604Sbonwick * Pass TRUE for mosconfig because the user-supplied config 2220ecc2d604Sbonwick * is actually the one to trust when doing an import. 2221fa9e4066Sahrens */ 2222ecc2d604Sbonwick (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE); 2223fa9e4066Sahrens 2224fa9e4066Sahrens /* 2225fa9e4066Sahrens * If 'tryconfig' was at least parsable, return the current config. 2226fa9e4066Sahrens */ 2227fa9e4066Sahrens if (spa->spa_root_vdev != NULL) { 22280373e76bSbonwick spa_config_enter(spa, RW_READER, FTAG); 2229fa9e4066Sahrens config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 22300373e76bSbonwick spa_config_exit(spa, FTAG); 2231fa9e4066Sahrens VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 2232fa9e4066Sahrens poolname) == 0); 2233fa9e4066Sahrens VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 2234fa9e4066Sahrens state) == 0); 223595173954Sek VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 223695173954Sek spa->spa_uberblock.ub_timestamp) == 0); 223799653d4eSeschrock 223899653d4eSeschrock /* 2239fa94a07fSbrendan * Add the list of hot spares and level 2 cache devices. 224099653d4eSeschrock */ 224199653d4eSeschrock spa_add_spares(spa, config); 2242fa94a07fSbrendan spa_add_l2cache(spa, config); 2243fa9e4066Sahrens } 2244fa9e4066Sahrens 2245fa9e4066Sahrens spa_unload(spa); 2246fa9e4066Sahrens spa_deactivate(spa); 2247fa9e4066Sahrens spa_remove(spa); 2248fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2249fa9e4066Sahrens 2250fa9e4066Sahrens return (config); 2251fa9e4066Sahrens } 2252fa9e4066Sahrens 2253fa9e4066Sahrens /* 2254fa9e4066Sahrens * Pool export/destroy 2255fa9e4066Sahrens * 2256fa9e4066Sahrens * The act of destroying or exporting a pool is very simple. We make sure there 2257fa9e4066Sahrens * is no more pending I/O and any references to the pool are gone. Then, we 2258fa9e4066Sahrens * update the pool state and sync all the labels to disk, removing the 2259fa9e4066Sahrens * configuration from the cache afterwards. 2260fa9e4066Sahrens */ 2261fa9e4066Sahrens static int 226244cd46caSbillm spa_export_common(char *pool, int new_state, nvlist_t **oldconfig) 2263fa9e4066Sahrens { 2264fa9e4066Sahrens spa_t *spa; 2265fa9e4066Sahrens 226644cd46caSbillm if (oldconfig) 226744cd46caSbillm *oldconfig = NULL; 226844cd46caSbillm 2269fa9e4066Sahrens if (!(spa_mode & FWRITE)) 2270fa9e4066Sahrens return (EROFS); 2271fa9e4066Sahrens 2272fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 2273fa9e4066Sahrens if ((spa = spa_lookup(pool)) == NULL) { 2274fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2275fa9e4066Sahrens return (ENOENT); 2276fa9e4066Sahrens } 2277fa9e4066Sahrens 2278ea8dc4b6Seschrock /* 2279ea8dc4b6Seschrock * Put a hold on the pool, drop the namespace lock, stop async tasks, 2280ea8dc4b6Seschrock * reacquire the namespace lock, and see if we can export. 2281ea8dc4b6Seschrock */ 2282ea8dc4b6Seschrock spa_open_ref(spa, FTAG); 2283ea8dc4b6Seschrock mutex_exit(&spa_namespace_lock); 2284ea8dc4b6Seschrock spa_async_suspend(spa); 2285ea8dc4b6Seschrock mutex_enter(&spa_namespace_lock); 2286ea8dc4b6Seschrock spa_close(spa, FTAG); 2287ea8dc4b6Seschrock 2288fa9e4066Sahrens /* 2289fa9e4066Sahrens * The pool will be in core if it's openable, 2290fa9e4066Sahrens * in which case we can modify its state. 2291fa9e4066Sahrens */ 2292fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 2293fa9e4066Sahrens /* 2294fa9e4066Sahrens * Objsets may be open only because they're dirty, so we 2295fa9e4066Sahrens * have to force it to sync before checking spa_refcnt. 2296fa9e4066Sahrens */ 2297fa9e4066Sahrens spa_scrub_suspend(spa); 2298fa9e4066Sahrens txg_wait_synced(spa->spa_dsl_pool, 0); 2299fa9e4066Sahrens 2300ea8dc4b6Seschrock /* 2301ea8dc4b6Seschrock * A pool cannot be exported or destroyed if there are active 2302ea8dc4b6Seschrock * references. If we are resetting a pool, allow references by 2303ea8dc4b6Seschrock * fault injection handlers. 2304ea8dc4b6Seschrock */ 2305ea8dc4b6Seschrock if (!spa_refcount_zero(spa) || 2306ea8dc4b6Seschrock (spa->spa_inject_ref != 0 && 2307ea8dc4b6Seschrock new_state != POOL_STATE_UNINITIALIZED)) { 2308fa9e4066Sahrens spa_scrub_resume(spa); 2309ea8dc4b6Seschrock spa_async_resume(spa); 2310fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2311fa9e4066Sahrens return (EBUSY); 2312fa9e4066Sahrens } 2313fa9e4066Sahrens 2314fa9e4066Sahrens spa_scrub_resume(spa); 2315fa9e4066Sahrens VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); 2316fa9e4066Sahrens 2317fa9e4066Sahrens /* 2318fa9e4066Sahrens * We want this to be reflected on every label, 2319fa9e4066Sahrens * so mark them all dirty. spa_unload() will do the 2320fa9e4066Sahrens * final sync that pushes these changes out. 2321fa9e4066Sahrens */ 2322ea8dc4b6Seschrock if (new_state != POOL_STATE_UNINITIALIZED) { 23235dabedeeSbonwick spa_config_enter(spa, RW_WRITER, FTAG); 2324ea8dc4b6Seschrock spa->spa_state = new_state; 23250373e76bSbonwick spa->spa_final_txg = spa_last_synced_txg(spa) + 1; 2326ea8dc4b6Seschrock vdev_config_dirty(spa->spa_root_vdev); 23275dabedeeSbonwick spa_config_exit(spa, FTAG); 2328ea8dc4b6Seschrock } 2329fa9e4066Sahrens } 2330fa9e4066Sahrens 23313d7072f8Seschrock spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 23323d7072f8Seschrock 2333fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 2334fa9e4066Sahrens spa_unload(spa); 2335fa9e4066Sahrens spa_deactivate(spa); 2336fa9e4066Sahrens } 2337fa9e4066Sahrens 233844cd46caSbillm if (oldconfig && spa->spa_config) 233944cd46caSbillm VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 234044cd46caSbillm 2341ea8dc4b6Seschrock if (new_state != POOL_STATE_UNINITIALIZED) { 23422f8aaab3Seschrock spa_config_check(spa->spa_config_dir, 23432f8aaab3Seschrock spa->spa_config_file); 2344ea8dc4b6Seschrock spa_remove(spa); 2345ea8dc4b6Seschrock spa_config_sync(); 2346ea8dc4b6Seschrock } 2347fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 2348fa9e4066Sahrens 2349fa9e4066Sahrens return (0); 2350fa9e4066Sahrens } 2351fa9e4066Sahrens 2352fa9e4066Sahrens /* 2353fa9e4066Sahrens * Destroy a storage pool. 2354fa9e4066Sahrens */ 2355fa9e4066Sahrens int 2356fa9e4066Sahrens spa_destroy(char *pool) 2357fa9e4066Sahrens { 235844cd46caSbillm return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL)); 2359fa9e4066Sahrens } 2360fa9e4066Sahrens 2361fa9e4066Sahrens /* 2362fa9e4066Sahrens * Export a storage pool. 2363fa9e4066Sahrens */ 2364fa9e4066Sahrens int 236544cd46caSbillm spa_export(char *pool, nvlist_t **oldconfig) 2366fa9e4066Sahrens { 236744cd46caSbillm return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig)); 2368fa9e4066Sahrens } 2369fa9e4066Sahrens 2370ea8dc4b6Seschrock /* 2371ea8dc4b6Seschrock * Similar to spa_export(), this unloads the spa_t without actually removing it 2372ea8dc4b6Seschrock * from the namespace in any way. 2373ea8dc4b6Seschrock */ 2374ea8dc4b6Seschrock int 2375ea8dc4b6Seschrock spa_reset(char *pool) 2376ea8dc4b6Seschrock { 237744cd46caSbillm return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL)); 2378ea8dc4b6Seschrock } 2379ea8dc4b6Seschrock 2380ea8dc4b6Seschrock 2381fa9e4066Sahrens /* 2382fa9e4066Sahrens * ========================================================================== 2383fa9e4066Sahrens * Device manipulation 2384fa9e4066Sahrens * ========================================================================== 2385fa9e4066Sahrens */ 2386fa9e4066Sahrens 2387fa9e4066Sahrens /* 23888654d025Sperrin * Add a device to a storage pool. 2389fa9e4066Sahrens */ 2390fa9e4066Sahrens int 2391fa9e4066Sahrens spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 2392fa9e4066Sahrens { 2393fa9e4066Sahrens uint64_t txg; 23940373e76bSbonwick int c, error; 2395fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 23960e34b6a7Sbonwick vdev_t *vd, *tvd; 2397fa94a07fSbrendan nvlist_t **spares, **l2cache; 2398fa94a07fSbrendan uint_t nspares, nl2cache; 2399fa9e4066Sahrens 2400fa9e4066Sahrens txg = spa_vdev_enter(spa); 2401fa9e4066Sahrens 240299653d4eSeschrock if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 240399653d4eSeschrock VDEV_ALLOC_ADD)) != 0) 240499653d4eSeschrock return (spa_vdev_exit(spa, NULL, txg, error)); 2405fa9e4066Sahrens 240639c23413Seschrock spa->spa_pending_vdev = vd; 240799653d4eSeschrock 2408fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 2409fa94a07fSbrendan &nspares) != 0) 241099653d4eSeschrock nspares = 0; 241199653d4eSeschrock 2412fa94a07fSbrendan if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 2413fa94a07fSbrendan &nl2cache) != 0) 2414fa94a07fSbrendan nl2cache = 0; 2415fa94a07fSbrendan 2416fa94a07fSbrendan if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) { 241739c23413Seschrock spa->spa_pending_vdev = NULL; 2418fa9e4066Sahrens return (spa_vdev_exit(spa, vd, txg, EINVAL)); 241939c23413Seschrock } 2420fa9e4066Sahrens 242199653d4eSeschrock if (vd->vdev_children != 0) { 242239c23413Seschrock if ((error = vdev_create(vd, txg, B_FALSE)) != 0) { 242339c23413Seschrock spa->spa_pending_vdev = NULL; 242499653d4eSeschrock return (spa_vdev_exit(spa, vd, txg, error)); 242599653d4eSeschrock } 242699653d4eSeschrock } 242799653d4eSeschrock 242839c23413Seschrock /* 2429fa94a07fSbrendan * We must validate the spares and l2cache devices after checking the 2430fa94a07fSbrendan * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 243139c23413Seschrock */ 2432fa94a07fSbrendan if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) { 243339c23413Seschrock spa->spa_pending_vdev = NULL; 243439c23413Seschrock return (spa_vdev_exit(spa, vd, txg, error)); 243539c23413Seschrock } 243639c23413Seschrock 243739c23413Seschrock spa->spa_pending_vdev = NULL; 243839c23413Seschrock 243939c23413Seschrock /* 244039c23413Seschrock * Transfer each new top-level vdev from vd to rvd. 244139c23413Seschrock */ 244239c23413Seschrock for (c = 0; c < vd->vdev_children; c++) { 244339c23413Seschrock tvd = vd->vdev_child[c]; 244439c23413Seschrock vdev_remove_child(vd, tvd); 244539c23413Seschrock tvd->vdev_id = rvd->vdev_children; 244639c23413Seschrock vdev_add_child(rvd, tvd); 244739c23413Seschrock vdev_config_dirty(tvd); 244839c23413Seschrock } 244939c23413Seschrock 245099653d4eSeschrock if (nspares != 0) { 2451fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 2452fa94a07fSbrendan ZPOOL_CONFIG_SPARES); 245399653d4eSeschrock spa_load_spares(spa); 2454fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 2455fa94a07fSbrendan } 2456fa94a07fSbrendan 2457fa94a07fSbrendan if (nl2cache != 0) { 2458fa94a07fSbrendan spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 2459fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE); 2460fa94a07fSbrendan spa_load_l2cache(spa); 2461fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 2462fa9e4066Sahrens } 2463fa9e4066Sahrens 2464fa9e4066Sahrens /* 24650e34b6a7Sbonwick * We have to be careful when adding new vdevs to an existing pool. 24660e34b6a7Sbonwick * If other threads start allocating from these vdevs before we 24670e34b6a7Sbonwick * sync the config cache, and we lose power, then upon reboot we may 24680e34b6a7Sbonwick * fail to open the pool because there are DVAs that the config cache 24690e34b6a7Sbonwick * can't translate. Therefore, we first add the vdevs without 24700e34b6a7Sbonwick * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 24710373e76bSbonwick * and then let spa_config_update() initialize the new metaslabs. 24720e34b6a7Sbonwick * 24730e34b6a7Sbonwick * spa_load() checks for added-but-not-initialized vdevs, so that 24740e34b6a7Sbonwick * if we lose power at any point in this sequence, the remaining 24750e34b6a7Sbonwick * steps will be completed the next time we load the pool. 24760e34b6a7Sbonwick */ 24770373e76bSbonwick (void) spa_vdev_exit(spa, vd, txg, 0); 24780e34b6a7Sbonwick 24790373e76bSbonwick mutex_enter(&spa_namespace_lock); 24800373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 24810373e76bSbonwick mutex_exit(&spa_namespace_lock); 2482fa9e4066Sahrens 24830373e76bSbonwick return (0); 2484fa9e4066Sahrens } 2485fa9e4066Sahrens 2486fa9e4066Sahrens /* 2487fa9e4066Sahrens * Attach a device to a mirror. The arguments are the path to any device 2488fa9e4066Sahrens * in the mirror, and the nvroot for the new device. If the path specifies 2489fa9e4066Sahrens * a device that is not mirrored, we automatically insert the mirror vdev. 2490fa9e4066Sahrens * 2491fa9e4066Sahrens * If 'replacing' is specified, the new device is intended to replace the 2492fa9e4066Sahrens * existing device; in this case the two devices are made into their own 24933d7072f8Seschrock * mirror using the 'replacing' vdev, which is functionally identical to 2494fa9e4066Sahrens * the mirror vdev (it actually reuses all the same ops) but has a few 2495fa9e4066Sahrens * extra rules: you can't attach to it after it's been created, and upon 2496fa9e4066Sahrens * completion of resilvering, the first disk (the one being replaced) 2497fa9e4066Sahrens * is automatically detached. 2498fa9e4066Sahrens */ 2499fa9e4066Sahrens int 2500ea8dc4b6Seschrock spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 2501fa9e4066Sahrens { 2502fa9e4066Sahrens uint64_t txg, open_txg; 2503fa9e4066Sahrens int error; 2504fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2505fa9e4066Sahrens vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 250699653d4eSeschrock vdev_ops_t *pvops; 25078654d025Sperrin int is_log; 2508fa9e4066Sahrens 2509fa9e4066Sahrens txg = spa_vdev_enter(spa); 2510fa9e4066Sahrens 2511ea8dc4b6Seschrock oldvd = vdev_lookup_by_guid(rvd, guid); 2512fa9e4066Sahrens 2513fa9e4066Sahrens if (oldvd == NULL) 2514fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 2515fa9e4066Sahrens 25160e34b6a7Sbonwick if (!oldvd->vdev_ops->vdev_op_leaf) 25170e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 25180e34b6a7Sbonwick 2519fa9e4066Sahrens pvd = oldvd->vdev_parent; 2520fa9e4066Sahrens 252199653d4eSeschrock if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 25223d7072f8Seschrock VDEV_ALLOC_ADD)) != 0) 25233d7072f8Seschrock return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 25243d7072f8Seschrock 25253d7072f8Seschrock if (newrootvd->vdev_children != 1) 2526fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 2527fa9e4066Sahrens 2528fa9e4066Sahrens newvd = newrootvd->vdev_child[0]; 2529fa9e4066Sahrens 2530fa9e4066Sahrens if (!newvd->vdev_ops->vdev_op_leaf) 2531fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 2532fa9e4066Sahrens 253399653d4eSeschrock if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 2534fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, error)); 2535fa9e4066Sahrens 25368654d025Sperrin /* 25378654d025Sperrin * Spares can't replace logs 25388654d025Sperrin */ 25398654d025Sperrin is_log = oldvd->vdev_islog; 25408654d025Sperrin if (is_log && newvd->vdev_isspare) 25418654d025Sperrin return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 25428654d025Sperrin 254399653d4eSeschrock if (!replacing) { 254499653d4eSeschrock /* 254599653d4eSeschrock * For attach, the only allowable parent is a mirror or the root 254699653d4eSeschrock * vdev. 254799653d4eSeschrock */ 254899653d4eSeschrock if (pvd->vdev_ops != &vdev_mirror_ops && 254999653d4eSeschrock pvd->vdev_ops != &vdev_root_ops) 255099653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 255199653d4eSeschrock 255299653d4eSeschrock pvops = &vdev_mirror_ops; 255399653d4eSeschrock } else { 255499653d4eSeschrock /* 255599653d4eSeschrock * Active hot spares can only be replaced by inactive hot 255699653d4eSeschrock * spares. 255799653d4eSeschrock */ 255899653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 255999653d4eSeschrock pvd->vdev_child[1] == oldvd && 256099653d4eSeschrock !spa_has_spare(spa, newvd->vdev_guid)) 256199653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 256299653d4eSeschrock 256399653d4eSeschrock /* 256499653d4eSeschrock * If the source is a hot spare, and the parent isn't already a 256599653d4eSeschrock * spare, then we want to create a new hot spare. Otherwise, we 256639c23413Seschrock * want to create a replacing vdev. The user is not allowed to 256739c23413Seschrock * attach to a spared vdev child unless the 'isspare' state is 256839c23413Seschrock * the same (spare replaces spare, non-spare replaces 256939c23413Seschrock * non-spare). 257099653d4eSeschrock */ 257199653d4eSeschrock if (pvd->vdev_ops == &vdev_replacing_ops) 257299653d4eSeschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 257339c23413Seschrock else if (pvd->vdev_ops == &vdev_spare_ops && 257439c23413Seschrock newvd->vdev_isspare != oldvd->vdev_isspare) 257539c23413Seschrock return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 257699653d4eSeschrock else if (pvd->vdev_ops != &vdev_spare_ops && 257799653d4eSeschrock newvd->vdev_isspare) 257899653d4eSeschrock pvops = &vdev_spare_ops; 257999653d4eSeschrock else 258099653d4eSeschrock pvops = &vdev_replacing_ops; 258199653d4eSeschrock } 258299653d4eSeschrock 25832a79c5feSlling /* 25842a79c5feSlling * Compare the new device size with the replaceable/attachable 25852a79c5feSlling * device size. 25862a79c5feSlling */ 25872a79c5feSlling if (newvd->vdev_psize < vdev_get_rsize(oldvd)) 2588fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 2589fa9e4066Sahrens 2590ecc2d604Sbonwick /* 2591ecc2d604Sbonwick * The new device cannot have a higher alignment requirement 2592ecc2d604Sbonwick * than the top-level vdev. 2593ecc2d604Sbonwick */ 2594ecc2d604Sbonwick if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 2595fa9e4066Sahrens return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 2596fa9e4066Sahrens 2597fa9e4066Sahrens /* 2598fa9e4066Sahrens * If this is an in-place replacement, update oldvd's path and devid 2599fa9e4066Sahrens * to make it distinguishable from newvd, and unopenable from now on. 2600fa9e4066Sahrens */ 2601fa9e4066Sahrens if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 2602fa9e4066Sahrens spa_strfree(oldvd->vdev_path); 2603fa9e4066Sahrens oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 2604fa9e4066Sahrens KM_SLEEP); 2605fa9e4066Sahrens (void) sprintf(oldvd->vdev_path, "%s/%s", 2606fa9e4066Sahrens newvd->vdev_path, "old"); 2607fa9e4066Sahrens if (oldvd->vdev_devid != NULL) { 2608fa9e4066Sahrens spa_strfree(oldvd->vdev_devid); 2609fa9e4066Sahrens oldvd->vdev_devid = NULL; 2610fa9e4066Sahrens } 2611fa9e4066Sahrens } 2612fa9e4066Sahrens 2613fa9e4066Sahrens /* 261499653d4eSeschrock * If the parent is not a mirror, or if we're replacing, insert the new 261599653d4eSeschrock * mirror/replacing/spare vdev above oldvd. 2616fa9e4066Sahrens */ 2617fa9e4066Sahrens if (pvd->vdev_ops != pvops) 2618fa9e4066Sahrens pvd = vdev_add_parent(oldvd, pvops); 2619fa9e4066Sahrens 2620fa9e4066Sahrens ASSERT(pvd->vdev_top->vdev_parent == rvd); 2621fa9e4066Sahrens ASSERT(pvd->vdev_ops == pvops); 2622fa9e4066Sahrens ASSERT(oldvd->vdev_parent == pvd); 2623fa9e4066Sahrens 2624fa9e4066Sahrens /* 2625fa9e4066Sahrens * Extract the new device from its root and add it to pvd. 2626fa9e4066Sahrens */ 2627fa9e4066Sahrens vdev_remove_child(newrootvd, newvd); 2628fa9e4066Sahrens newvd->vdev_id = pvd->vdev_children; 2629fa9e4066Sahrens vdev_add_child(pvd, newvd); 2630fa9e4066Sahrens 2631ea8dc4b6Seschrock /* 2632ea8dc4b6Seschrock * If newvd is smaller than oldvd, but larger than its rsize, 2633ea8dc4b6Seschrock * the addition of newvd may have decreased our parent's asize. 2634ea8dc4b6Seschrock */ 2635ea8dc4b6Seschrock pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize); 2636ea8dc4b6Seschrock 2637fa9e4066Sahrens tvd = newvd->vdev_top; 2638fa9e4066Sahrens ASSERT(pvd->vdev_top == tvd); 2639fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 2640fa9e4066Sahrens 2641fa9e4066Sahrens vdev_config_dirty(tvd); 2642fa9e4066Sahrens 2643fa9e4066Sahrens /* 2644fa9e4066Sahrens * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate 2645fa9e4066Sahrens * upward when spa_vdev_exit() calls vdev_dtl_reassess(). 2646fa9e4066Sahrens */ 2647fa9e4066Sahrens open_txg = txg + TXG_CONCURRENT_STATES - 1; 2648fa9e4066Sahrens 2649fa9e4066Sahrens mutex_enter(&newvd->vdev_dtl_lock); 2650fa9e4066Sahrens space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL, 2651fa9e4066Sahrens open_txg - TXG_INITIAL + 1); 2652fa9e4066Sahrens mutex_exit(&newvd->vdev_dtl_lock); 2653fa9e4066Sahrens 265439c23413Seschrock if (newvd->vdev_isspare) 265539c23413Seschrock spa_spare_activate(newvd); 2656ea8dc4b6Seschrock 2657fa9e4066Sahrens /* 2658fa9e4066Sahrens * Mark newvd's DTL dirty in this txg. 2659fa9e4066Sahrens */ 2660ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, newvd, txg); 2661fa9e4066Sahrens 2662fa9e4066Sahrens (void) spa_vdev_exit(spa, newrootvd, open_txg, 0); 2663fa9e4066Sahrens 2664fa9e4066Sahrens /* 26653d7072f8Seschrock * Kick off a resilver to update newvd. We need to grab the namespace 26663d7072f8Seschrock * lock because spa_scrub() needs to post a sysevent with the pool name. 2667fa9e4066Sahrens */ 26683d7072f8Seschrock mutex_enter(&spa_namespace_lock); 2669fa9e4066Sahrens VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); 26703d7072f8Seschrock mutex_exit(&spa_namespace_lock); 2671fa9e4066Sahrens 2672fa9e4066Sahrens return (0); 2673fa9e4066Sahrens } 2674fa9e4066Sahrens 2675fa9e4066Sahrens /* 2676fa9e4066Sahrens * Detach a device from a mirror or replacing vdev. 2677fa9e4066Sahrens * If 'replace_done' is specified, only detach if the parent 2678fa9e4066Sahrens * is a replacing vdev. 2679fa9e4066Sahrens */ 2680fa9e4066Sahrens int 2681ea8dc4b6Seschrock spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done) 2682fa9e4066Sahrens { 2683fa9e4066Sahrens uint64_t txg; 2684fa9e4066Sahrens int c, t, error; 2685fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 2686fa9e4066Sahrens vdev_t *vd, *pvd, *cvd, *tvd; 268799653d4eSeschrock boolean_t unspare = B_FALSE; 268899653d4eSeschrock uint64_t unspare_guid; 2689fa9e4066Sahrens 2690fa9e4066Sahrens txg = spa_vdev_enter(spa); 2691fa9e4066Sahrens 2692ea8dc4b6Seschrock vd = vdev_lookup_by_guid(rvd, guid); 2693fa9e4066Sahrens 2694fa9e4066Sahrens if (vd == NULL) 2695fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 2696fa9e4066Sahrens 26970e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 26980e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 26990e34b6a7Sbonwick 2700fa9e4066Sahrens pvd = vd->vdev_parent; 2701fa9e4066Sahrens 2702fa9e4066Sahrens /* 2703fa9e4066Sahrens * If replace_done is specified, only remove this device if it's 270499653d4eSeschrock * the first child of a replacing vdev. For the 'spare' vdev, either 270599653d4eSeschrock * disk can be removed. 270699653d4eSeschrock */ 270799653d4eSeschrock if (replace_done) { 270899653d4eSeschrock if (pvd->vdev_ops == &vdev_replacing_ops) { 270999653d4eSeschrock if (vd->vdev_id != 0) 271099653d4eSeschrock return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 271199653d4eSeschrock } else if (pvd->vdev_ops != &vdev_spare_ops) { 271299653d4eSeschrock return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 271399653d4eSeschrock } 271499653d4eSeschrock } 271599653d4eSeschrock 271699653d4eSeschrock ASSERT(pvd->vdev_ops != &vdev_spare_ops || 2717e7437265Sahrens spa_version(spa) >= SPA_VERSION_SPARES); 2718fa9e4066Sahrens 2719fa9e4066Sahrens /* 272099653d4eSeschrock * Only mirror, replacing, and spare vdevs support detach. 2721fa9e4066Sahrens */ 2722fa9e4066Sahrens if (pvd->vdev_ops != &vdev_replacing_ops && 272399653d4eSeschrock pvd->vdev_ops != &vdev_mirror_ops && 272499653d4eSeschrock pvd->vdev_ops != &vdev_spare_ops) 2725fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 2726fa9e4066Sahrens 2727fa9e4066Sahrens /* 2728fa9e4066Sahrens * If there's only one replica, you can't detach it. 2729fa9e4066Sahrens */ 2730fa9e4066Sahrens if (pvd->vdev_children <= 1) 2731fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 2732fa9e4066Sahrens 2733fa9e4066Sahrens /* 2734fa9e4066Sahrens * If all siblings have non-empty DTLs, this device may have the only 2735fa9e4066Sahrens * valid copy of the data, which means we cannot safely detach it. 2736fa9e4066Sahrens * 2737fa9e4066Sahrens * XXX -- as in the vdev_offline() case, we really want a more 2738fa9e4066Sahrens * precise DTL check. 2739fa9e4066Sahrens */ 2740fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) { 2741fa9e4066Sahrens uint64_t dirty; 2742fa9e4066Sahrens 2743fa9e4066Sahrens cvd = pvd->vdev_child[c]; 2744fa9e4066Sahrens if (cvd == vd) 2745fa9e4066Sahrens continue; 2746fa9e4066Sahrens if (vdev_is_dead(cvd)) 2747fa9e4066Sahrens continue; 2748fa9e4066Sahrens mutex_enter(&cvd->vdev_dtl_lock); 2749fa9e4066Sahrens dirty = cvd->vdev_dtl_map.sm_space | 2750fa9e4066Sahrens cvd->vdev_dtl_scrub.sm_space; 2751fa9e4066Sahrens mutex_exit(&cvd->vdev_dtl_lock); 2752fa9e4066Sahrens if (!dirty) 2753fa9e4066Sahrens break; 2754fa9e4066Sahrens } 275599653d4eSeschrock 275699653d4eSeschrock /* 275799653d4eSeschrock * If we are a replacing or spare vdev, then we can always detach the 275899653d4eSeschrock * latter child, as that is how one cancels the operation. 275999653d4eSeschrock */ 276099653d4eSeschrock if ((pvd->vdev_ops == &vdev_mirror_ops || vd->vdev_id != 1) && 276199653d4eSeschrock c == pvd->vdev_children) 2762fa9e4066Sahrens return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 2763fa9e4066Sahrens 276499653d4eSeschrock /* 276599653d4eSeschrock * If we are detaching the original disk from a spare, then it implies 276699653d4eSeschrock * that the spare should become a real disk, and be removed from the 276799653d4eSeschrock * active spare list for the pool. 276899653d4eSeschrock */ 276999653d4eSeschrock if (pvd->vdev_ops == &vdev_spare_ops && 277099653d4eSeschrock vd->vdev_id == 0) 277199653d4eSeschrock unspare = B_TRUE; 277299653d4eSeschrock 2773fa9e4066Sahrens /* 2774fa9e4066Sahrens * Erase the disk labels so the disk can be used for other things. 2775fa9e4066Sahrens * This must be done after all other error cases are handled, 2776fa9e4066Sahrens * but before we disembowel vd (so we can still do I/O to it). 2777fa9e4066Sahrens * But if we can't do it, don't treat the error as fatal -- 2778fa9e4066Sahrens * it may be that the unwritability of the disk is the reason 2779fa9e4066Sahrens * it's being detached! 2780fa9e4066Sahrens */ 278139c23413Seschrock error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 2782fa9e4066Sahrens 2783fa9e4066Sahrens /* 2784fa9e4066Sahrens * Remove vd from its parent and compact the parent's children. 2785fa9e4066Sahrens */ 2786fa9e4066Sahrens vdev_remove_child(pvd, vd); 2787fa9e4066Sahrens vdev_compact_children(pvd); 2788fa9e4066Sahrens 2789fa9e4066Sahrens /* 2790fa9e4066Sahrens * Remember one of the remaining children so we can get tvd below. 2791fa9e4066Sahrens */ 2792fa9e4066Sahrens cvd = pvd->vdev_child[0]; 2793fa9e4066Sahrens 279499653d4eSeschrock /* 279599653d4eSeschrock * If we need to remove the remaining child from the list of hot spares, 279699653d4eSeschrock * do it now, marking the vdev as no longer a spare in the process. We 279799653d4eSeschrock * must do this before vdev_remove_parent(), because that can change the 279899653d4eSeschrock * GUID if it creates a new toplevel GUID. 279999653d4eSeschrock */ 280099653d4eSeschrock if (unspare) { 280199653d4eSeschrock ASSERT(cvd->vdev_isspare); 280239c23413Seschrock spa_spare_remove(cvd); 280399653d4eSeschrock unspare_guid = cvd->vdev_guid; 280499653d4eSeschrock } 280599653d4eSeschrock 2806fa9e4066Sahrens /* 2807fa9e4066Sahrens * If the parent mirror/replacing vdev only has one child, 2808fa9e4066Sahrens * the parent is no longer needed. Remove it from the tree. 2809fa9e4066Sahrens */ 2810fa9e4066Sahrens if (pvd->vdev_children == 1) 2811fa9e4066Sahrens vdev_remove_parent(cvd); 2812fa9e4066Sahrens 2813fa9e4066Sahrens /* 2814fa9e4066Sahrens * We don't set tvd until now because the parent we just removed 2815fa9e4066Sahrens * may have been the previous top-level vdev. 2816fa9e4066Sahrens */ 2817fa9e4066Sahrens tvd = cvd->vdev_top; 2818fa9e4066Sahrens ASSERT(tvd->vdev_parent == rvd); 2819fa9e4066Sahrens 2820fa9e4066Sahrens /* 282139c23413Seschrock * Reevaluate the parent vdev state. 2822fa9e4066Sahrens */ 28233d7072f8Seschrock vdev_propagate_state(cvd); 2824fa9e4066Sahrens 2825fa9e4066Sahrens /* 282639c23413Seschrock * If the device we just detached was smaller than the others, it may be 282739c23413Seschrock * possible to add metaslabs (i.e. grow the pool). vdev_metaslab_init() 282839c23413Seschrock * can't fail because the existing metaslabs are already in core, so 282939c23413Seschrock * there's nothing to read from disk. 2830fa9e4066Sahrens */ 2831ecc2d604Sbonwick VERIFY(vdev_metaslab_init(tvd, txg) == 0); 2832fa9e4066Sahrens 2833fa9e4066Sahrens vdev_config_dirty(tvd); 2834fa9e4066Sahrens 2835fa9e4066Sahrens /* 283639c23413Seschrock * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 283739c23413Seschrock * vd->vdev_detached is set and free vd's DTL object in syncing context. 283839c23413Seschrock * But first make sure we're not on any *other* txg's DTL list, to 283939c23413Seschrock * prevent vd from being accessed after it's freed. 2840fa9e4066Sahrens */ 2841fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) 2842fa9e4066Sahrens (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 2843ecc2d604Sbonwick vd->vdev_detached = B_TRUE; 2844ecc2d604Sbonwick vdev_dirty(tvd, VDD_DTL, vd, txg); 2845fa9e4066Sahrens 28463d7072f8Seschrock spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 28473d7072f8Seschrock 284899653d4eSeschrock error = spa_vdev_exit(spa, vd, txg, 0); 284999653d4eSeschrock 285099653d4eSeschrock /* 285139c23413Seschrock * If this was the removal of the original device in a hot spare vdev, 285239c23413Seschrock * then we want to go through and remove the device from the hot spare 285339c23413Seschrock * list of every other pool. 285499653d4eSeschrock */ 285599653d4eSeschrock if (unspare) { 285699653d4eSeschrock spa = NULL; 285799653d4eSeschrock mutex_enter(&spa_namespace_lock); 285899653d4eSeschrock while ((spa = spa_next(spa)) != NULL) { 285999653d4eSeschrock if (spa->spa_state != POOL_STATE_ACTIVE) 286099653d4eSeschrock continue; 286199653d4eSeschrock 286299653d4eSeschrock (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 286399653d4eSeschrock } 286499653d4eSeschrock mutex_exit(&spa_namespace_lock); 286599653d4eSeschrock } 286699653d4eSeschrock 286799653d4eSeschrock return (error); 286899653d4eSeschrock } 286999653d4eSeschrock 287099653d4eSeschrock /* 2871fa94a07fSbrendan * Remove a spares vdev from the nvlist config. 287299653d4eSeschrock */ 2873fa94a07fSbrendan static int 2874fa94a07fSbrendan spa_remove_spares(spa_aux_vdev_t *sav, uint64_t guid, boolean_t unspare, 2875fa94a07fSbrendan nvlist_t **spares, int nspares, vdev_t *vd) 287699653d4eSeschrock { 2877fa94a07fSbrendan nvlist_t *nv, **newspares; 2878fa94a07fSbrendan int i, j; 287999653d4eSeschrock 288099653d4eSeschrock nv = NULL; 2881fa94a07fSbrendan for (i = 0; i < nspares; i++) { 2882fa94a07fSbrendan uint64_t theguid; 288399653d4eSeschrock 2884fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(spares[i], 2885fa94a07fSbrendan ZPOOL_CONFIG_GUID, &theguid) == 0); 2886fa94a07fSbrendan if (theguid == guid) { 2887fa94a07fSbrendan nv = spares[i]; 2888fa94a07fSbrendan break; 288999653d4eSeschrock } 289099653d4eSeschrock } 289199653d4eSeschrock 289299653d4eSeschrock /* 2893fa94a07fSbrendan * Only remove the hot spare if it's not currently in use in this pool. 289499653d4eSeschrock */ 2895fa94a07fSbrendan if (nv == NULL && vd == NULL) 2896fa94a07fSbrendan return (ENOENT); 289799653d4eSeschrock 2898fa94a07fSbrendan if (nv == NULL && vd != NULL) 2899fa94a07fSbrendan return (ENOTSUP); 290099653d4eSeschrock 2901fa94a07fSbrendan if (!unspare && nv != NULL && vd != NULL) 2902fa94a07fSbrendan return (EBUSY); 290399653d4eSeschrock 290499653d4eSeschrock if (nspares == 1) { 290599653d4eSeschrock newspares = NULL; 290699653d4eSeschrock } else { 290799653d4eSeschrock newspares = kmem_alloc((nspares - 1) * sizeof (void *), 290899653d4eSeschrock KM_SLEEP); 290999653d4eSeschrock for (i = 0, j = 0; i < nspares; i++) { 291099653d4eSeschrock if (spares[i] != nv) 291199653d4eSeschrock VERIFY(nvlist_dup(spares[i], 291299653d4eSeschrock &newspares[j++], KM_SLEEP) == 0); 291399653d4eSeschrock } 291499653d4eSeschrock } 291599653d4eSeschrock 2916fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_SPARES, 291799653d4eSeschrock DATA_TYPE_NVLIST_ARRAY) == 0); 2918fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 2919fa94a07fSbrendan ZPOOL_CONFIG_SPARES, newspares, nspares - 1) == 0); 292099653d4eSeschrock for (i = 0; i < nspares - 1; i++) 292199653d4eSeschrock nvlist_free(newspares[i]); 292299653d4eSeschrock kmem_free(newspares, (nspares - 1) * sizeof (void *)); 2923fa94a07fSbrendan 2924fa94a07fSbrendan return (0); 2925fa94a07fSbrendan } 2926fa94a07fSbrendan 2927fa94a07fSbrendan /* 2928fa94a07fSbrendan * Remove an l2cache vdev from the nvlist config. 2929fa94a07fSbrendan */ 2930fa94a07fSbrendan static int 2931fa94a07fSbrendan spa_remove_l2cache(spa_aux_vdev_t *sav, uint64_t guid, nvlist_t **l2cache, 2932fa94a07fSbrendan int nl2cache, vdev_t *vd) 2933fa94a07fSbrendan { 2934fa94a07fSbrendan nvlist_t *nv, **newl2cache; 2935fa94a07fSbrendan int i, j; 2936fa94a07fSbrendan 2937fa94a07fSbrendan nv = NULL; 2938fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 2939fa94a07fSbrendan uint64_t theguid; 2940fa94a07fSbrendan 2941fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], 2942fa94a07fSbrendan ZPOOL_CONFIG_GUID, &theguid) == 0); 2943fa94a07fSbrendan if (theguid == guid) { 2944fa94a07fSbrendan nv = l2cache[i]; 2945fa94a07fSbrendan break; 2946fa94a07fSbrendan } 2947fa94a07fSbrendan } 2948fa94a07fSbrendan 2949fa94a07fSbrendan if (vd == NULL) { 2950fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 2951fa94a07fSbrendan if (sav->sav_vdevs[i]->vdev_guid == guid) { 2952fa94a07fSbrendan vd = sav->sav_vdevs[i]; 2953fa94a07fSbrendan break; 2954fa94a07fSbrendan } 2955fa94a07fSbrendan } 2956fa94a07fSbrendan } 2957fa94a07fSbrendan 2958fa94a07fSbrendan if (nv == NULL && vd == NULL) 2959fa94a07fSbrendan return (ENOENT); 2960fa94a07fSbrendan 2961fa94a07fSbrendan if (nv == NULL && vd != NULL) 2962fa94a07fSbrendan return (ENOTSUP); 2963fa94a07fSbrendan 2964fa94a07fSbrendan if (nl2cache == 1) { 2965fa94a07fSbrendan newl2cache = NULL; 2966fa94a07fSbrendan } else { 2967fa94a07fSbrendan newl2cache = kmem_alloc((nl2cache - 1) * sizeof (void *), 2968fa94a07fSbrendan KM_SLEEP); 2969fa94a07fSbrendan for (i = 0, j = 0; i < nl2cache; i++) { 2970fa94a07fSbrendan if (l2cache[i] != nv) 2971fa94a07fSbrendan VERIFY(nvlist_dup(l2cache[i], 2972fa94a07fSbrendan &newl2cache[j++], KM_SLEEP) == 0); 2973fa94a07fSbrendan } 2974fa94a07fSbrendan } 2975fa94a07fSbrendan 2976fa94a07fSbrendan VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 2977fa94a07fSbrendan DATA_TYPE_NVLIST_ARRAY) == 0); 2978fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(sav->sav_config, 2979fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, newl2cache, nl2cache - 1) == 0); 2980fa94a07fSbrendan for (i = 0; i < nl2cache - 1; i++) 2981fa94a07fSbrendan nvlist_free(newl2cache[i]); 2982fa94a07fSbrendan kmem_free(newl2cache, (nl2cache - 1) * sizeof (void *)); 2983fa94a07fSbrendan 2984fa94a07fSbrendan return (0); 2985fa94a07fSbrendan } 2986fa94a07fSbrendan 2987fa94a07fSbrendan /* 2988fa94a07fSbrendan * Remove a device from the pool. Currently, this supports removing only hot 2989fa94a07fSbrendan * spares and level 2 ARC devices. 2990fa94a07fSbrendan */ 2991fa94a07fSbrendan int 2992fa94a07fSbrendan spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 2993fa94a07fSbrendan { 2994fa94a07fSbrendan vdev_t *vd; 2995fa94a07fSbrendan nvlist_t **spares, **l2cache; 2996fa94a07fSbrendan uint_t nspares, nl2cache; 2997fa94a07fSbrendan int error = 0; 2998fa94a07fSbrendan 2999fa94a07fSbrendan spa_config_enter(spa, RW_WRITER, FTAG); 3000fa94a07fSbrendan 3001fa94a07fSbrendan vd = spa_lookup_by_guid(spa, guid); 3002fa94a07fSbrendan 3003fa94a07fSbrendan if (spa->spa_spares.sav_vdevs != NULL && 3004fa94a07fSbrendan spa_spare_exists(guid, NULL) && 3005fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3006fa94a07fSbrendan ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) { 3007fa94a07fSbrendan if ((error = spa_remove_spares(&spa->spa_spares, guid, unspare, 3008fa94a07fSbrendan spares, nspares, vd)) != 0) 3009fa94a07fSbrendan goto out; 3010fa94a07fSbrendan spa_load_spares(spa); 3011fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 3012fa94a07fSbrendan goto out; 3013fa94a07fSbrendan } 3014fa94a07fSbrendan 3015fa94a07fSbrendan if (spa->spa_l2cache.sav_vdevs != NULL && 3016fa94a07fSbrendan spa_l2cache_exists(guid, NULL) && 3017fa94a07fSbrendan nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3018fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0) { 3019fa94a07fSbrendan if ((error = spa_remove_l2cache(&spa->spa_l2cache, guid, 3020fa94a07fSbrendan l2cache, nl2cache, vd)) != 0) 3021fa94a07fSbrendan goto out; 3022fa94a07fSbrendan spa_load_l2cache(spa); 3023fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 3024fa94a07fSbrendan } 302599653d4eSeschrock 302699653d4eSeschrock out: 302799653d4eSeschrock spa_config_exit(spa, FTAG); 3028fa94a07fSbrendan return (error); 3029fa9e4066Sahrens } 3030fa9e4066Sahrens 3031fa9e4066Sahrens /* 30323d7072f8Seschrock * Find any device that's done replacing, or a vdev marked 'unspare' that's 30333d7072f8Seschrock * current spared, so we can detach it. 3034fa9e4066Sahrens */ 3035ea8dc4b6Seschrock static vdev_t * 30363d7072f8Seschrock spa_vdev_resilver_done_hunt(vdev_t *vd) 3037fa9e4066Sahrens { 3038ea8dc4b6Seschrock vdev_t *newvd, *oldvd; 3039fa9e4066Sahrens int c; 3040fa9e4066Sahrens 3041ea8dc4b6Seschrock for (c = 0; c < vd->vdev_children; c++) { 30423d7072f8Seschrock oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 3043ea8dc4b6Seschrock if (oldvd != NULL) 3044ea8dc4b6Seschrock return (oldvd); 3045ea8dc4b6Seschrock } 3046fa9e4066Sahrens 30473d7072f8Seschrock /* 30483d7072f8Seschrock * Check for a completed replacement. 30493d7072f8Seschrock */ 3050fa9e4066Sahrens if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) { 3051ea8dc4b6Seschrock oldvd = vd->vdev_child[0]; 3052ea8dc4b6Seschrock newvd = vd->vdev_child[1]; 3053ea8dc4b6Seschrock 3054ea8dc4b6Seschrock mutex_enter(&newvd->vdev_dtl_lock); 3055ea8dc4b6Seschrock if (newvd->vdev_dtl_map.sm_space == 0 && 3056ea8dc4b6Seschrock newvd->vdev_dtl_scrub.sm_space == 0) { 3057ea8dc4b6Seschrock mutex_exit(&newvd->vdev_dtl_lock); 3058ea8dc4b6Seschrock return (oldvd); 3059fa9e4066Sahrens } 3060ea8dc4b6Seschrock mutex_exit(&newvd->vdev_dtl_lock); 3061fa9e4066Sahrens } 3062ea8dc4b6Seschrock 30633d7072f8Seschrock /* 30643d7072f8Seschrock * Check for a completed resilver with the 'unspare' flag set. 30653d7072f8Seschrock */ 30663d7072f8Seschrock if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) { 30673d7072f8Seschrock newvd = vd->vdev_child[0]; 30683d7072f8Seschrock oldvd = vd->vdev_child[1]; 30693d7072f8Seschrock 30703d7072f8Seschrock mutex_enter(&newvd->vdev_dtl_lock); 30713d7072f8Seschrock if (newvd->vdev_unspare && 30723d7072f8Seschrock newvd->vdev_dtl_map.sm_space == 0 && 30733d7072f8Seschrock newvd->vdev_dtl_scrub.sm_space == 0) { 30743d7072f8Seschrock newvd->vdev_unspare = 0; 30753d7072f8Seschrock mutex_exit(&newvd->vdev_dtl_lock); 30763d7072f8Seschrock return (oldvd); 30773d7072f8Seschrock } 30783d7072f8Seschrock mutex_exit(&newvd->vdev_dtl_lock); 30793d7072f8Seschrock } 30803d7072f8Seschrock 3081ea8dc4b6Seschrock return (NULL); 3082fa9e4066Sahrens } 3083fa9e4066Sahrens 3084ea8dc4b6Seschrock static void 30853d7072f8Seschrock spa_vdev_resilver_done(spa_t *spa) 3086fa9e4066Sahrens { 3087ea8dc4b6Seschrock vdev_t *vd; 308899653d4eSeschrock vdev_t *pvd; 3089ea8dc4b6Seschrock uint64_t guid; 309099653d4eSeschrock uint64_t pguid = 0; 3091ea8dc4b6Seschrock 3092ea8dc4b6Seschrock spa_config_enter(spa, RW_READER, FTAG); 3093ea8dc4b6Seschrock 30943d7072f8Seschrock while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 3095ea8dc4b6Seschrock guid = vd->vdev_guid; 309699653d4eSeschrock /* 309799653d4eSeschrock * If we have just finished replacing a hot spared device, then 309899653d4eSeschrock * we need to detach the parent's first child (the original hot 309999653d4eSeschrock * spare) as well. 310099653d4eSeschrock */ 310199653d4eSeschrock pvd = vd->vdev_parent; 310299653d4eSeschrock if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops && 310399653d4eSeschrock pvd->vdev_id == 0) { 310499653d4eSeschrock ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 310599653d4eSeschrock ASSERT(pvd->vdev_parent->vdev_children == 2); 310699653d4eSeschrock pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid; 310799653d4eSeschrock } 3108ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 3109ea8dc4b6Seschrock if (spa_vdev_detach(spa, guid, B_TRUE) != 0) 3110ea8dc4b6Seschrock return; 311199653d4eSeschrock if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0) 311299653d4eSeschrock return; 3113ea8dc4b6Seschrock spa_config_enter(spa, RW_READER, FTAG); 3114fa9e4066Sahrens } 3115fa9e4066Sahrens 3116ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 3117fa9e4066Sahrens } 3118fa9e4066Sahrens 3119c67d9675Seschrock /* 3120c67d9675Seschrock * Update the stored path for this vdev. Dirty the vdev configuration, relying 3121c67d9675Seschrock * on spa_vdev_enter/exit() to synchronize the labels and cache. 3122c67d9675Seschrock */ 3123c67d9675Seschrock int 3124c67d9675Seschrock spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 3125c67d9675Seschrock { 3126c67d9675Seschrock vdev_t *rvd, *vd; 3127c67d9675Seschrock uint64_t txg; 3128c67d9675Seschrock 3129c67d9675Seschrock rvd = spa->spa_root_vdev; 3130c67d9675Seschrock 3131c67d9675Seschrock txg = spa_vdev_enter(spa); 3132c67d9675Seschrock 313399653d4eSeschrock if ((vd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 313499653d4eSeschrock /* 3135fa94a07fSbrendan * Determine if this is a reference to a hot spare or l2cache 3136fa94a07fSbrendan * device. If it is, update the path as stored in their 3137fa94a07fSbrendan * device list. 313899653d4eSeschrock */ 3139fa94a07fSbrendan nvlist_t **spares, **l2cache; 3140fa94a07fSbrendan uint_t i, nspares, nl2cache; 3141fa94a07fSbrendan 3142fa94a07fSbrendan if (spa->spa_spares.sav_config != NULL) { 3143fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array( 3144fa94a07fSbrendan spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 3145fa94a07fSbrendan &spares, &nspares) == 0); 314699653d4eSeschrock for (i = 0; i < nspares; i++) { 314799653d4eSeschrock uint64_t theguid; 314899653d4eSeschrock VERIFY(nvlist_lookup_uint64(spares[i], 314999653d4eSeschrock ZPOOL_CONFIG_GUID, &theguid) == 0); 3150fa94a07fSbrendan if (theguid == guid) { 3151fa94a07fSbrendan VERIFY(nvlist_add_string(spares[i], 3152fa94a07fSbrendan ZPOOL_CONFIG_PATH, newpath) == 0); 3153fa94a07fSbrendan spa_load_spares(spa); 3154fa94a07fSbrendan spa->spa_spares.sav_sync = B_TRUE; 3155fa94a07fSbrendan return (spa_vdev_exit(spa, NULL, txg, 3156fa94a07fSbrendan 0)); 3157fa94a07fSbrendan } 315899653d4eSeschrock } 3159fa94a07fSbrendan } 316099653d4eSeschrock 3161fa94a07fSbrendan if (spa->spa_l2cache.sav_config != NULL) { 3162fa94a07fSbrendan VERIFY(nvlist_lookup_nvlist_array( 3163fa94a07fSbrendan spa->spa_l2cache.sav_config, ZPOOL_CONFIG_L2CACHE, 3164fa94a07fSbrendan &l2cache, &nl2cache) == 0); 3165fa94a07fSbrendan for (i = 0; i < nl2cache; i++) { 3166fa94a07fSbrendan uint64_t theguid; 3167fa94a07fSbrendan VERIFY(nvlist_lookup_uint64(l2cache[i], 3168fa94a07fSbrendan ZPOOL_CONFIG_GUID, &theguid) == 0); 3169fa94a07fSbrendan if (theguid == guid) { 3170fa94a07fSbrendan VERIFY(nvlist_add_string(l2cache[i], 3171fa94a07fSbrendan ZPOOL_CONFIG_PATH, newpath) == 0); 3172fa94a07fSbrendan spa_load_l2cache(spa); 3173fa94a07fSbrendan spa->spa_l2cache.sav_sync = B_TRUE; 3174fa94a07fSbrendan return (spa_vdev_exit(spa, NULL, txg, 3175fa94a07fSbrendan 0)); 3176fa94a07fSbrendan } 3177fa94a07fSbrendan } 317899653d4eSeschrock } 3179fa94a07fSbrendan 3180fa94a07fSbrendan return (spa_vdev_exit(spa, NULL, txg, ENOENT)); 318199653d4eSeschrock } 3182c67d9675Seschrock 31830e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 31840e34b6a7Sbonwick return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 31850e34b6a7Sbonwick 3186c67d9675Seschrock spa_strfree(vd->vdev_path); 3187c67d9675Seschrock vd->vdev_path = spa_strdup(newpath); 3188c67d9675Seschrock 3189c67d9675Seschrock vdev_config_dirty(vd->vdev_top); 3190c67d9675Seschrock 3191c67d9675Seschrock return (spa_vdev_exit(spa, NULL, txg, 0)); 3192c67d9675Seschrock } 3193c67d9675Seschrock 3194fa9e4066Sahrens /* 3195fa9e4066Sahrens * ========================================================================== 3196fa9e4066Sahrens * SPA Scrubbing 3197fa9e4066Sahrens * ========================================================================== 3198fa9e4066Sahrens */ 3199fa9e4066Sahrens 3200fa9e4066Sahrens static void 3201fa9e4066Sahrens spa_scrub_io_done(zio_t *zio) 3202fa9e4066Sahrens { 3203fa9e4066Sahrens spa_t *spa = zio->io_spa; 3204fa9e4066Sahrens 32050e8c6158Smaybee arc_data_buf_free(zio->io_data, zio->io_size); 3206fa9e4066Sahrens 3207fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3208ea8dc4b6Seschrock if (zio->io_error && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { 320944cd46caSbillm vdev_t *vd = zio->io_vd ? zio->io_vd : spa->spa_root_vdev; 3210ea8dc4b6Seschrock spa->spa_scrub_errors++; 3211fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 3212fa9e4066Sahrens vd->vdev_stat.vs_scrub_errors++; 3213fa9e4066Sahrens mutex_exit(&vd->vdev_stat_lock); 3214fa9e4066Sahrens } 321505b2b3b8Smishra 321605b2b3b8Smishra if (--spa->spa_scrub_inflight < spa->spa_scrub_maxinflight) 3217ea8dc4b6Seschrock cv_broadcast(&spa->spa_scrub_io_cv); 321805b2b3b8Smishra 321905b2b3b8Smishra ASSERT(spa->spa_scrub_inflight >= 0); 322005b2b3b8Smishra 3221ea8dc4b6Seschrock mutex_exit(&spa->spa_scrub_lock); 3222fa9e4066Sahrens } 3223fa9e4066Sahrens 3224fa9e4066Sahrens static void 3225ea8dc4b6Seschrock spa_scrub_io_start(spa_t *spa, blkptr_t *bp, int priority, int flags, 3226ea8dc4b6Seschrock zbookmark_t *zb) 3227fa9e4066Sahrens { 3228fa9e4066Sahrens size_t size = BP_GET_LSIZE(bp); 322905b2b3b8Smishra void *data; 3230fa9e4066Sahrens 3231fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 323205b2b3b8Smishra /* 323305b2b3b8Smishra * Do not give too much work to vdev(s). 323405b2b3b8Smishra */ 323505b2b3b8Smishra while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight) { 323605b2b3b8Smishra cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 323705b2b3b8Smishra } 3238fa9e4066Sahrens spa->spa_scrub_inflight++; 3239fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3240fa9e4066Sahrens 32410e8c6158Smaybee data = arc_data_buf_alloc(size); 324205b2b3b8Smishra 3243ea8dc4b6Seschrock if (zb->zb_level == -1 && BP_GET_TYPE(bp) != DMU_OT_OBJSET) 3244ea8dc4b6Seschrock flags |= ZIO_FLAG_SPECULATIVE; /* intent log block */ 3245ea8dc4b6Seschrock 3246d80c45e0Sbonwick flags |= ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_CANFAIL; 3247ea8dc4b6Seschrock 3248fa9e4066Sahrens zio_nowait(zio_read(NULL, spa, bp, data, size, 3249ea8dc4b6Seschrock spa_scrub_io_done, NULL, priority, flags, zb)); 3250fa9e4066Sahrens } 3251fa9e4066Sahrens 3252fa9e4066Sahrens /* ARGSUSED */ 3253fa9e4066Sahrens static int 3254fa9e4066Sahrens spa_scrub_cb(traverse_blk_cache_t *bc, spa_t *spa, void *a) 3255fa9e4066Sahrens { 3256fa9e4066Sahrens blkptr_t *bp = &bc->bc_blkptr; 325744cd46caSbillm vdev_t *vd = spa->spa_root_vdev; 325844cd46caSbillm dva_t *dva = bp->blk_dva; 325944cd46caSbillm int needs_resilver = B_FALSE; 326044cd46caSbillm int d; 3261fa9e4066Sahrens 326244cd46caSbillm if (bc->bc_errno) { 3263fa9e4066Sahrens /* 3264fa9e4066Sahrens * We can't scrub this block, but we can continue to scrub 3265fa9e4066Sahrens * the rest of the pool. Note the error and move along. 3266fa9e4066Sahrens */ 3267fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3268fa9e4066Sahrens spa->spa_scrub_errors++; 3269fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3270fa9e4066Sahrens 327144cd46caSbillm mutex_enter(&vd->vdev_stat_lock); 327244cd46caSbillm vd->vdev_stat.vs_scrub_errors++; 327344cd46caSbillm mutex_exit(&vd->vdev_stat_lock); 3274fa9e4066Sahrens 3275fa9e4066Sahrens return (ERESTART); 3276fa9e4066Sahrens } 3277fa9e4066Sahrens 3278fa9e4066Sahrens ASSERT(bp->blk_birth < spa->spa_scrub_maxtxg); 3279fa9e4066Sahrens 328044cd46caSbillm for (d = 0; d < BP_GET_NDVAS(bp); d++) { 328144cd46caSbillm vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d])); 3282fa9e4066Sahrens 328344cd46caSbillm ASSERT(vd != NULL); 328444cd46caSbillm 328544cd46caSbillm /* 328644cd46caSbillm * Keep track of how much data we've examined so that 328744cd46caSbillm * zpool(1M) status can make useful progress reports. 328844cd46caSbillm */ 328944cd46caSbillm mutex_enter(&vd->vdev_stat_lock); 329044cd46caSbillm vd->vdev_stat.vs_scrub_examined += DVA_GET_ASIZE(&dva[d]); 329144cd46caSbillm mutex_exit(&vd->vdev_stat_lock); 329244cd46caSbillm 329344cd46caSbillm if (spa->spa_scrub_type == POOL_SCRUB_RESILVER) { 329444cd46caSbillm if (DVA_GET_GANG(&dva[d])) { 329544cd46caSbillm /* 329644cd46caSbillm * Gang members may be spread across multiple 329744cd46caSbillm * vdevs, so the best we can do is look at the 329844cd46caSbillm * pool-wide DTL. 329944cd46caSbillm * XXX -- it would be better to change our 330044cd46caSbillm * allocation policy to ensure that this can't 330144cd46caSbillm * happen. 330244cd46caSbillm */ 330344cd46caSbillm vd = spa->spa_root_vdev; 330444cd46caSbillm } 330544cd46caSbillm if (vdev_dtl_contains(&vd->vdev_dtl_map, 330644cd46caSbillm bp->blk_birth, 1)) 330744cd46caSbillm needs_resilver = B_TRUE; 3308fa9e4066Sahrens } 330944cd46caSbillm } 331044cd46caSbillm 331144cd46caSbillm if (spa->spa_scrub_type == POOL_SCRUB_EVERYTHING) 3312fa9e4066Sahrens spa_scrub_io_start(spa, bp, ZIO_PRIORITY_SCRUB, 3313ea8dc4b6Seschrock ZIO_FLAG_SCRUB, &bc->bc_bookmark); 331444cd46caSbillm else if (needs_resilver) 331544cd46caSbillm spa_scrub_io_start(spa, bp, ZIO_PRIORITY_RESILVER, 331644cd46caSbillm ZIO_FLAG_RESILVER, &bc->bc_bookmark); 3317fa9e4066Sahrens 3318fa9e4066Sahrens return (0); 3319fa9e4066Sahrens } 3320fa9e4066Sahrens 3321fa9e4066Sahrens static void 3322fa9e4066Sahrens spa_scrub_thread(spa_t *spa) 3323fa9e4066Sahrens { 3324fa9e4066Sahrens callb_cpr_t cprinfo; 3325fa9e4066Sahrens traverse_handle_t *th = spa->spa_scrub_th; 3326fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3327fa9e4066Sahrens pool_scrub_type_t scrub_type = spa->spa_scrub_type; 3328fa9e4066Sahrens int error = 0; 3329fa9e4066Sahrens boolean_t complete; 3330fa9e4066Sahrens 3331fa9e4066Sahrens CALLB_CPR_INIT(&cprinfo, &spa->spa_scrub_lock, callb_generic_cpr, FTAG); 3332fa9e4066Sahrens 3333f0aa80d4Sbonwick /* 3334f0aa80d4Sbonwick * If we're restarting due to a snapshot create/delete, 3335f0aa80d4Sbonwick * wait for that to complete. 3336f0aa80d4Sbonwick */ 3337f0aa80d4Sbonwick txg_wait_synced(spa_get_dsl(spa), 0); 3338f0aa80d4Sbonwick 3339ea8dc4b6Seschrock dprintf("start %s mintxg=%llu maxtxg=%llu\n", 3340ea8dc4b6Seschrock scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", 3341ea8dc4b6Seschrock spa->spa_scrub_mintxg, spa->spa_scrub_maxtxg); 3342ea8dc4b6Seschrock 3343ea8dc4b6Seschrock spa_config_enter(spa, RW_WRITER, FTAG); 3344ea8dc4b6Seschrock vdev_reopen(rvd); /* purge all vdev caches */ 3345fa9e4066Sahrens vdev_config_dirty(rvd); /* rewrite all disk labels */ 3346fa9e4066Sahrens vdev_scrub_stat_update(rvd, scrub_type, B_FALSE); 3347ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 3348fa9e4066Sahrens 3349fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3350fa9e4066Sahrens spa->spa_scrub_errors = 0; 3351fa9e4066Sahrens spa->spa_scrub_active = 1; 3352ea8dc4b6Seschrock ASSERT(spa->spa_scrub_inflight == 0); 3353fa9e4066Sahrens 3354fa9e4066Sahrens while (!spa->spa_scrub_stop) { 3355fa9e4066Sahrens CALLB_CPR_SAFE_BEGIN(&cprinfo); 3356ea8dc4b6Seschrock while (spa->spa_scrub_suspended) { 3357fa9e4066Sahrens spa->spa_scrub_active = 0; 3358fa9e4066Sahrens cv_broadcast(&spa->spa_scrub_cv); 3359fa9e4066Sahrens cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); 3360fa9e4066Sahrens spa->spa_scrub_active = 1; 3361fa9e4066Sahrens } 3362fa9e4066Sahrens CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_scrub_lock); 3363fa9e4066Sahrens 3364fa9e4066Sahrens if (spa->spa_scrub_restart_txg != 0) 3365fa9e4066Sahrens break; 3366fa9e4066Sahrens 3367fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3368fa9e4066Sahrens error = traverse_more(th); 3369fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3370fa9e4066Sahrens if (error != EAGAIN) 3371fa9e4066Sahrens break; 3372fa9e4066Sahrens } 3373fa9e4066Sahrens 3374fa9e4066Sahrens while (spa->spa_scrub_inflight) 3375fa9e4066Sahrens cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 3376fa9e4066Sahrens 33775dabedeeSbonwick spa->spa_scrub_active = 0; 33785dabedeeSbonwick cv_broadcast(&spa->spa_scrub_cv); 33795dabedeeSbonwick 33805dabedeeSbonwick mutex_exit(&spa->spa_scrub_lock); 33815dabedeeSbonwick 33825dabedeeSbonwick spa_config_enter(spa, RW_WRITER, FTAG); 33835dabedeeSbonwick 33845dabedeeSbonwick mutex_enter(&spa->spa_scrub_lock); 33855dabedeeSbonwick 33865dabedeeSbonwick /* 33875dabedeeSbonwick * Note: we check spa_scrub_restart_txg under both spa_scrub_lock 33885dabedeeSbonwick * AND the spa config lock to synchronize with any config changes 33895dabedeeSbonwick * that revise the DTLs under spa_vdev_enter() / spa_vdev_exit(). 33905dabedeeSbonwick */ 3391fa9e4066Sahrens if (spa->spa_scrub_restart_txg != 0) 3392fa9e4066Sahrens error = ERESTART; 3393fa9e4066Sahrens 3394ea8dc4b6Seschrock if (spa->spa_scrub_stop) 3395ea8dc4b6Seschrock error = EINTR; 3396ea8dc4b6Seschrock 3397fa9e4066Sahrens /* 3398ea8dc4b6Seschrock * Even if there were uncorrectable errors, we consider the scrub 3399ea8dc4b6Seschrock * completed. The downside is that if there is a transient error during 3400ea8dc4b6Seschrock * a resilver, we won't resilver the data properly to the target. But 3401ea8dc4b6Seschrock * if the damage is permanent (more likely) we will resilver forever, 3402ea8dc4b6Seschrock * which isn't really acceptable. Since there is enough information for 3403ea8dc4b6Seschrock * the user to know what has failed and why, this seems like a more 3404ea8dc4b6Seschrock * tractable approach. 3405fa9e4066Sahrens */ 3406ea8dc4b6Seschrock complete = (error == 0); 3407fa9e4066Sahrens 3408ea8dc4b6Seschrock dprintf("end %s to maxtxg=%llu %s, traverse=%d, %llu errors, stop=%u\n", 3409ea8dc4b6Seschrock scrub_type == POOL_SCRUB_RESILVER ? "resilver" : "scrub", 3410fa9e4066Sahrens spa->spa_scrub_maxtxg, complete ? "done" : "FAILED", 3411fa9e4066Sahrens error, spa->spa_scrub_errors, spa->spa_scrub_stop); 3412fa9e4066Sahrens 3413fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3414fa9e4066Sahrens 3415fa9e4066Sahrens /* 3416fa9e4066Sahrens * If the scrub/resilver completed, update all DTLs to reflect this. 3417fa9e4066Sahrens * Whether it succeeded or not, vacate all temporary scrub DTLs. 3418fa9e4066Sahrens */ 3419fa9e4066Sahrens vdev_dtl_reassess(rvd, spa_last_synced_txg(spa) + 1, 3420fa9e4066Sahrens complete ? spa->spa_scrub_maxtxg : 0, B_TRUE); 3421fa9e4066Sahrens vdev_scrub_stat_update(rvd, POOL_SCRUB_NONE, complete); 3422ea8dc4b6Seschrock spa_errlog_rotate(spa); 34235dabedeeSbonwick 34243d7072f8Seschrock if (scrub_type == POOL_SCRUB_RESILVER && complete) 34253d7072f8Seschrock spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_FINISH); 34263d7072f8Seschrock 3427ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 3428fa9e4066Sahrens 3429fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3430fa9e4066Sahrens 3431ea8dc4b6Seschrock /* 3432ea8dc4b6Seschrock * We may have finished replacing a device. 3433ea8dc4b6Seschrock * Let the async thread assess this and handle the detach. 3434ea8dc4b6Seschrock */ 34353d7072f8Seschrock spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 3436fa9e4066Sahrens 3437fa9e4066Sahrens /* 3438fa9e4066Sahrens * If we were told to restart, our final act is to start a new scrub. 3439fa9e4066Sahrens */ 3440fa9e4066Sahrens if (error == ERESTART) 3441ea8dc4b6Seschrock spa_async_request(spa, scrub_type == POOL_SCRUB_RESILVER ? 3442ea8dc4b6Seschrock SPA_ASYNC_RESILVER : SPA_ASYNC_SCRUB); 3443fa9e4066Sahrens 3444ea8dc4b6Seschrock spa->spa_scrub_type = POOL_SCRUB_NONE; 3445ea8dc4b6Seschrock spa->spa_scrub_active = 0; 3446ea8dc4b6Seschrock spa->spa_scrub_thread = NULL; 3447ea8dc4b6Seschrock cv_broadcast(&spa->spa_scrub_cv); 3448fa9e4066Sahrens CALLB_CPR_EXIT(&cprinfo); /* drops &spa->spa_scrub_lock */ 3449fa9e4066Sahrens thread_exit(); 3450fa9e4066Sahrens } 3451fa9e4066Sahrens 3452fa9e4066Sahrens void 3453fa9e4066Sahrens spa_scrub_suspend(spa_t *spa) 3454fa9e4066Sahrens { 3455fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3456ea8dc4b6Seschrock spa->spa_scrub_suspended++; 3457fa9e4066Sahrens while (spa->spa_scrub_active) { 3458fa9e4066Sahrens cv_broadcast(&spa->spa_scrub_cv); 3459fa9e4066Sahrens cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); 3460fa9e4066Sahrens } 3461fa9e4066Sahrens while (spa->spa_scrub_inflight) 3462fa9e4066Sahrens cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 3463fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3464fa9e4066Sahrens } 3465fa9e4066Sahrens 3466fa9e4066Sahrens void 3467fa9e4066Sahrens spa_scrub_resume(spa_t *spa) 3468fa9e4066Sahrens { 3469fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3470ea8dc4b6Seschrock ASSERT(spa->spa_scrub_suspended != 0); 3471ea8dc4b6Seschrock if (--spa->spa_scrub_suspended == 0) 3472fa9e4066Sahrens cv_broadcast(&spa->spa_scrub_cv); 3473fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3474fa9e4066Sahrens } 3475fa9e4066Sahrens 3476fa9e4066Sahrens void 3477fa9e4066Sahrens spa_scrub_restart(spa_t *spa, uint64_t txg) 3478fa9e4066Sahrens { 3479fa9e4066Sahrens /* 3480fa9e4066Sahrens * Something happened (e.g. snapshot create/delete) that means 3481fa9e4066Sahrens * we must restart any in-progress scrubs. The itinerary will 3482fa9e4066Sahrens * fix this properly. 3483fa9e4066Sahrens */ 3484fa9e4066Sahrens mutex_enter(&spa->spa_scrub_lock); 3485fa9e4066Sahrens spa->spa_scrub_restart_txg = txg; 3486fa9e4066Sahrens mutex_exit(&spa->spa_scrub_lock); 3487fa9e4066Sahrens } 3488fa9e4066Sahrens 3489ea8dc4b6Seschrock int 3490ea8dc4b6Seschrock spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force) 3491fa9e4066Sahrens { 3492fa9e4066Sahrens space_seg_t *ss; 3493fa9e4066Sahrens uint64_t mintxg, maxtxg; 3494fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3495fa9e4066Sahrens 3496bb8b5132Sek ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3497bb8b5132Sek ASSERT(!spa_config_held(spa, RW_WRITER)); 3498bb8b5132Sek 3499fa9e4066Sahrens if ((uint_t)type >= POOL_SCRUB_TYPES) 3500fa9e4066Sahrens return (ENOTSUP); 3501fa9e4066Sahrens 3502ea8dc4b6Seschrock mutex_enter(&spa->spa_scrub_lock); 3503ea8dc4b6Seschrock 3504fa9e4066Sahrens /* 3505fa9e4066Sahrens * If there's a scrub or resilver already in progress, stop it. 3506fa9e4066Sahrens */ 3507fa9e4066Sahrens while (spa->spa_scrub_thread != NULL) { 3508fa9e4066Sahrens /* 3509fa9e4066Sahrens * Don't stop a resilver unless forced. 3510fa9e4066Sahrens */ 3511ea8dc4b6Seschrock if (spa->spa_scrub_type == POOL_SCRUB_RESILVER && !force) { 3512ea8dc4b6Seschrock mutex_exit(&spa->spa_scrub_lock); 3513fa9e4066Sahrens return (EBUSY); 3514ea8dc4b6Seschrock } 3515fa9e4066Sahrens spa->spa_scrub_stop = 1; 3516fa9e4066Sahrens cv_broadcast(&spa->spa_scrub_cv); 3517fa9e4066Sahrens cv_wait(&spa->spa_scrub_cv, &spa->spa_scrub_lock); 3518fa9e4066Sahrens } 3519fa9e4066Sahrens 3520fa9e4066Sahrens /* 3521fa9e4066Sahrens * Terminate the previous traverse. 3522fa9e4066Sahrens */ 3523fa9e4066Sahrens if (spa->spa_scrub_th != NULL) { 3524fa9e4066Sahrens traverse_fini(spa->spa_scrub_th); 3525fa9e4066Sahrens spa->spa_scrub_th = NULL; 3526fa9e4066Sahrens } 3527fa9e4066Sahrens 3528ea8dc4b6Seschrock if (rvd == NULL) { 3529ea8dc4b6Seschrock ASSERT(spa->spa_scrub_stop == 0); 3530ea8dc4b6Seschrock ASSERT(spa->spa_scrub_type == type); 3531ea8dc4b6Seschrock ASSERT(spa->spa_scrub_restart_txg == 0); 3532ea8dc4b6Seschrock mutex_exit(&spa->spa_scrub_lock); 3533ea8dc4b6Seschrock return (0); 3534ea8dc4b6Seschrock } 3535fa9e4066Sahrens 3536fa9e4066Sahrens mintxg = TXG_INITIAL - 1; 3537fa9e4066Sahrens maxtxg = spa_last_synced_txg(spa) + 1; 3538fa9e4066Sahrens 3539ea8dc4b6Seschrock mutex_enter(&rvd->vdev_dtl_lock); 3540fa9e4066Sahrens 3541ea8dc4b6Seschrock if (rvd->vdev_dtl_map.sm_space == 0) { 3542ea8dc4b6Seschrock /* 3543ea8dc4b6Seschrock * The pool-wide DTL is empty. 3544ecc2d604Sbonwick * If this is a resilver, there's nothing to do except 3545ecc2d604Sbonwick * check whether any in-progress replacements have completed. 3546ea8dc4b6Seschrock */ 3547ecc2d604Sbonwick if (type == POOL_SCRUB_RESILVER) { 3548ea8dc4b6Seschrock type = POOL_SCRUB_NONE; 35493d7072f8Seschrock spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 3550ecc2d604Sbonwick } 3551ea8dc4b6Seschrock } else { 3552ea8dc4b6Seschrock /* 3553ea8dc4b6Seschrock * The pool-wide DTL is non-empty. 3554ea8dc4b6Seschrock * If this is a normal scrub, upgrade to a resilver instead. 3555ea8dc4b6Seschrock */ 3556ea8dc4b6Seschrock if (type == POOL_SCRUB_EVERYTHING) 3557ea8dc4b6Seschrock type = POOL_SCRUB_RESILVER; 3558ea8dc4b6Seschrock } 3559fa9e4066Sahrens 3560ea8dc4b6Seschrock if (type == POOL_SCRUB_RESILVER) { 3561fa9e4066Sahrens /* 3562fa9e4066Sahrens * Determine the resilvering boundaries. 3563fa9e4066Sahrens * 3564fa9e4066Sahrens * Note: (mintxg, maxtxg) is an open interval, 3565fa9e4066Sahrens * i.e. mintxg and maxtxg themselves are not included. 3566fa9e4066Sahrens * 3567fa9e4066Sahrens * Note: for maxtxg, we MIN with spa_last_synced_txg(spa) + 1 3568fa9e4066Sahrens * so we don't claim to resilver a txg that's still changing. 3569fa9e4066Sahrens */ 3570fa9e4066Sahrens ss = avl_first(&rvd->vdev_dtl_map.sm_root); 3571ea8dc4b6Seschrock mintxg = ss->ss_start - 1; 3572fa9e4066Sahrens ss = avl_last(&rvd->vdev_dtl_map.sm_root); 3573ea8dc4b6Seschrock maxtxg = MIN(ss->ss_end, maxtxg); 35743d7072f8Seschrock 35753d7072f8Seschrock spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START); 3576fa9e4066Sahrens } 3577fa9e4066Sahrens 3578ea8dc4b6Seschrock mutex_exit(&rvd->vdev_dtl_lock); 3579ea8dc4b6Seschrock 3580ea8dc4b6Seschrock spa->spa_scrub_stop = 0; 3581ea8dc4b6Seschrock spa->spa_scrub_type = type; 3582ea8dc4b6Seschrock spa->spa_scrub_restart_txg = 0; 3583ea8dc4b6Seschrock 3584ea8dc4b6Seschrock if (type != POOL_SCRUB_NONE) { 3585ea8dc4b6Seschrock spa->spa_scrub_mintxg = mintxg; 3586fa9e4066Sahrens spa->spa_scrub_maxtxg = maxtxg; 3587fa9e4066Sahrens spa->spa_scrub_th = traverse_init(spa, spa_scrub_cb, NULL, 35880373e76bSbonwick ADVANCE_PRE | ADVANCE_PRUNE | ADVANCE_ZIL, 35890373e76bSbonwick ZIO_FLAG_CANFAIL); 3590fa9e4066Sahrens traverse_add_pool(spa->spa_scrub_th, mintxg, maxtxg); 3591fa9e4066Sahrens spa->spa_scrub_thread = thread_create(NULL, 0, 3592fa9e4066Sahrens spa_scrub_thread, spa, 0, &p0, TS_RUN, minclsyspri); 3593fa9e4066Sahrens } 3594fa9e4066Sahrens 3595ea8dc4b6Seschrock mutex_exit(&spa->spa_scrub_lock); 3596ea8dc4b6Seschrock 3597fa9e4066Sahrens return (0); 3598fa9e4066Sahrens } 3599fa9e4066Sahrens 3600ea8dc4b6Seschrock /* 3601ea8dc4b6Seschrock * ========================================================================== 3602ea8dc4b6Seschrock * SPA async task processing 3603ea8dc4b6Seschrock * ========================================================================== 3604ea8dc4b6Seschrock */ 3605ea8dc4b6Seschrock 3606ea8dc4b6Seschrock static void 36073d7072f8Seschrock spa_async_remove(spa_t *spa, vdev_t *vd) 3608fa9e4066Sahrens { 3609ea8dc4b6Seschrock vdev_t *tvd; 3610ea8dc4b6Seschrock int c; 3611fa9e4066Sahrens 36123d7072f8Seschrock for (c = 0; c < vd->vdev_children; c++) { 36133d7072f8Seschrock tvd = vd->vdev_child[c]; 36143d7072f8Seschrock if (tvd->vdev_remove_wanted) { 36153d7072f8Seschrock tvd->vdev_remove_wanted = 0; 36163d7072f8Seschrock vdev_set_state(tvd, B_FALSE, VDEV_STATE_REMOVED, 36173d7072f8Seschrock VDEV_AUX_NONE); 36180a4e9518Sgw vdev_clear(spa, tvd, B_TRUE); 36193d7072f8Seschrock vdev_config_dirty(tvd->vdev_top); 3620ea8dc4b6Seschrock } 36213d7072f8Seschrock spa_async_remove(spa, tvd); 3622ea8dc4b6Seschrock } 3623ea8dc4b6Seschrock } 3624fa9e4066Sahrens 3625ea8dc4b6Seschrock static void 3626ea8dc4b6Seschrock spa_async_thread(spa_t *spa) 3627ea8dc4b6Seschrock { 3628ea8dc4b6Seschrock int tasks; 36293d7072f8Seschrock uint64_t txg; 3630ea8dc4b6Seschrock 3631ea8dc4b6Seschrock ASSERT(spa->spa_sync_on); 3632ea8dc4b6Seschrock 3633ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 3634ea8dc4b6Seschrock tasks = spa->spa_async_tasks; 3635ea8dc4b6Seschrock spa->spa_async_tasks = 0; 3636ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 3637ea8dc4b6Seschrock 36380373e76bSbonwick /* 36390373e76bSbonwick * See if the config needs to be updated. 36400373e76bSbonwick */ 36410373e76bSbonwick if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 36420373e76bSbonwick mutex_enter(&spa_namespace_lock); 36430373e76bSbonwick spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 36440373e76bSbonwick mutex_exit(&spa_namespace_lock); 36450373e76bSbonwick } 36460373e76bSbonwick 3647ea8dc4b6Seschrock /* 36483d7072f8Seschrock * See if any devices need to be marked REMOVED. 36490a4e9518Sgw * 36500a4e9518Sgw * XXX - We avoid doing this when we are in 36510a4e9518Sgw * I/O failure state since spa_vdev_enter() grabs 36520a4e9518Sgw * the namespace lock and would not be able to obtain 36530a4e9518Sgw * the writer config lock. 3654ea8dc4b6Seschrock */ 36550a4e9518Sgw if (tasks & SPA_ASYNC_REMOVE && 36560a4e9518Sgw spa_state(spa) != POOL_STATE_IO_FAILURE) { 36573d7072f8Seschrock txg = spa_vdev_enter(spa); 36583d7072f8Seschrock spa_async_remove(spa, spa->spa_root_vdev); 36593d7072f8Seschrock (void) spa_vdev_exit(spa, NULL, txg, 0); 36603d7072f8Seschrock } 3661ea8dc4b6Seschrock 3662ea8dc4b6Seschrock /* 3663ea8dc4b6Seschrock * If any devices are done replacing, detach them. 3664ea8dc4b6Seschrock */ 36653d7072f8Seschrock if (tasks & SPA_ASYNC_RESILVER_DONE) 36663d7072f8Seschrock spa_vdev_resilver_done(spa); 3667fa9e4066Sahrens 3668ea8dc4b6Seschrock /* 36693d7072f8Seschrock * Kick off a scrub. When starting a RESILVER scrub (or an EVERYTHING 36703d7072f8Seschrock * scrub which can become a resilver), we need to hold 36713d7072f8Seschrock * spa_namespace_lock() because the sysevent we post via 36723d7072f8Seschrock * spa_event_notify() needs to get the name of the pool. 3673ea8dc4b6Seschrock */ 36743d7072f8Seschrock if (tasks & SPA_ASYNC_SCRUB) { 36753d7072f8Seschrock mutex_enter(&spa_namespace_lock); 3676ea8dc4b6Seschrock VERIFY(spa_scrub(spa, POOL_SCRUB_EVERYTHING, B_TRUE) == 0); 36773d7072f8Seschrock mutex_exit(&spa_namespace_lock); 36783d7072f8Seschrock } 3679ea8dc4b6Seschrock 3680ea8dc4b6Seschrock /* 3681ea8dc4b6Seschrock * Kick off a resilver. 3682ea8dc4b6Seschrock */ 36833d7072f8Seschrock if (tasks & SPA_ASYNC_RESILVER) { 36843d7072f8Seschrock mutex_enter(&spa_namespace_lock); 3685ea8dc4b6Seschrock VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER, B_TRUE) == 0); 36863d7072f8Seschrock mutex_exit(&spa_namespace_lock); 36873d7072f8Seschrock } 3688ea8dc4b6Seschrock 3689ea8dc4b6Seschrock /* 3690ea8dc4b6Seschrock * Let the world know that we're done. 3691ea8dc4b6Seschrock */ 3692ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 3693ea8dc4b6Seschrock spa->spa_async_thread = NULL; 3694ea8dc4b6Seschrock cv_broadcast(&spa->spa_async_cv); 3695ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 3696ea8dc4b6Seschrock thread_exit(); 3697ea8dc4b6Seschrock } 3698ea8dc4b6Seschrock 3699ea8dc4b6Seschrock void 3700ea8dc4b6Seschrock spa_async_suspend(spa_t *spa) 3701ea8dc4b6Seschrock { 3702ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 3703ea8dc4b6Seschrock spa->spa_async_suspended++; 3704ea8dc4b6Seschrock while (spa->spa_async_thread != NULL) 3705ea8dc4b6Seschrock cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 3706ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 3707ea8dc4b6Seschrock } 3708ea8dc4b6Seschrock 3709ea8dc4b6Seschrock void 3710ea8dc4b6Seschrock spa_async_resume(spa_t *spa) 3711ea8dc4b6Seschrock { 3712ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 3713ea8dc4b6Seschrock ASSERT(spa->spa_async_suspended != 0); 3714ea8dc4b6Seschrock spa->spa_async_suspended--; 3715ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 3716ea8dc4b6Seschrock } 3717ea8dc4b6Seschrock 3718ea8dc4b6Seschrock static void 3719ea8dc4b6Seschrock spa_async_dispatch(spa_t *spa) 3720ea8dc4b6Seschrock { 3721ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 3722ea8dc4b6Seschrock if (spa->spa_async_tasks && !spa->spa_async_suspended && 37230373e76bSbonwick spa->spa_async_thread == NULL && 37240373e76bSbonwick rootdir != NULL && !vn_is_readonly(rootdir)) 3725ea8dc4b6Seschrock spa->spa_async_thread = thread_create(NULL, 0, 3726ea8dc4b6Seschrock spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 3727ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 3728ea8dc4b6Seschrock } 3729ea8dc4b6Seschrock 3730ea8dc4b6Seschrock void 3731ea8dc4b6Seschrock spa_async_request(spa_t *spa, int task) 3732ea8dc4b6Seschrock { 3733ea8dc4b6Seschrock mutex_enter(&spa->spa_async_lock); 3734ea8dc4b6Seschrock spa->spa_async_tasks |= task; 3735ea8dc4b6Seschrock mutex_exit(&spa->spa_async_lock); 3736fa9e4066Sahrens } 3737fa9e4066Sahrens 3738fa9e4066Sahrens /* 3739fa9e4066Sahrens * ========================================================================== 3740fa9e4066Sahrens * SPA syncing routines 3741fa9e4066Sahrens * ========================================================================== 3742fa9e4066Sahrens */ 3743fa9e4066Sahrens 3744fa9e4066Sahrens static void 3745fa9e4066Sahrens spa_sync_deferred_frees(spa_t *spa, uint64_t txg) 3746fa9e4066Sahrens { 3747fa9e4066Sahrens bplist_t *bpl = &spa->spa_sync_bplist; 3748fa9e4066Sahrens dmu_tx_t *tx; 3749fa9e4066Sahrens blkptr_t blk; 3750fa9e4066Sahrens uint64_t itor = 0; 3751fa9e4066Sahrens zio_t *zio; 3752fa9e4066Sahrens int error; 3753fa9e4066Sahrens uint8_t c = 1; 3754fa9e4066Sahrens 3755fa9e4066Sahrens zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CONFIG_HELD); 3756fa9e4066Sahrens 3757fa9e4066Sahrens while (bplist_iterate(bpl, &itor, &blk) == 0) 3758fa9e4066Sahrens zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL)); 3759fa9e4066Sahrens 3760fa9e4066Sahrens error = zio_wait(zio); 3761fa9e4066Sahrens ASSERT3U(error, ==, 0); 3762fa9e4066Sahrens 3763fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3764fa9e4066Sahrens bplist_vacate(bpl, tx); 3765fa9e4066Sahrens 3766fa9e4066Sahrens /* 3767fa9e4066Sahrens * Pre-dirty the first block so we sync to convergence faster. 3768fa9e4066Sahrens * (Usually only the first block is needed.) 3769fa9e4066Sahrens */ 3770fa9e4066Sahrens dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx); 3771fa9e4066Sahrens dmu_tx_commit(tx); 3772fa9e4066Sahrens } 3773fa9e4066Sahrens 3774fa9e4066Sahrens static void 377599653d4eSeschrock spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 3776fa9e4066Sahrens { 3777fa9e4066Sahrens char *packed = NULL; 3778fa9e4066Sahrens size_t nvsize = 0; 3779fa9e4066Sahrens dmu_buf_t *db; 3780fa9e4066Sahrens 378199653d4eSeschrock VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 3782fa9e4066Sahrens 3783fa9e4066Sahrens packed = kmem_alloc(nvsize, KM_SLEEP); 3784fa9e4066Sahrens 378599653d4eSeschrock VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 3786ea8dc4b6Seschrock KM_SLEEP) == 0); 3787fa9e4066Sahrens 378899653d4eSeschrock dmu_write(spa->spa_meta_objset, obj, 0, nvsize, packed, tx); 3789fa9e4066Sahrens 3790fa9e4066Sahrens kmem_free(packed, nvsize); 3791fa9e4066Sahrens 379299653d4eSeschrock VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 3793fa9e4066Sahrens dmu_buf_will_dirty(db, tx); 3794fa9e4066Sahrens *(uint64_t *)db->db_data = nvsize; 3795ea8dc4b6Seschrock dmu_buf_rele(db, FTAG); 3796fa9e4066Sahrens } 3797fa9e4066Sahrens 379899653d4eSeschrock static void 3799fa94a07fSbrendan spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 3800fa94a07fSbrendan const char *config, const char *entry) 380199653d4eSeschrock { 380299653d4eSeschrock nvlist_t *nvroot; 3803fa94a07fSbrendan nvlist_t **list; 380499653d4eSeschrock int i; 380599653d4eSeschrock 3806fa94a07fSbrendan if (!sav->sav_sync) 380799653d4eSeschrock return; 380899653d4eSeschrock 380999653d4eSeschrock /* 3810fa94a07fSbrendan * Update the MOS nvlist describing the list of available devices. 3811fa94a07fSbrendan * spa_validate_aux() will have already made sure this nvlist is 38123d7072f8Seschrock * valid and the vdevs are labeled appropriately. 381399653d4eSeschrock */ 3814fa94a07fSbrendan if (sav->sav_object == 0) { 3815fa94a07fSbrendan sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 3816fa94a07fSbrendan DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 3817fa94a07fSbrendan sizeof (uint64_t), tx); 381899653d4eSeschrock VERIFY(zap_update(spa->spa_meta_objset, 3819fa94a07fSbrendan DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 3820fa94a07fSbrendan &sav->sav_object, tx) == 0); 382199653d4eSeschrock } 382299653d4eSeschrock 382399653d4eSeschrock VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3824fa94a07fSbrendan if (sav->sav_count == 0) { 3825fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 382699653d4eSeschrock } else { 3827fa94a07fSbrendan list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 3828fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 3829fa94a07fSbrendan list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 3830fa94a07fSbrendan B_FALSE, B_FALSE, B_TRUE); 3831fa94a07fSbrendan VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 3832fa94a07fSbrendan sav->sav_count) == 0); 3833fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 3834fa94a07fSbrendan nvlist_free(list[i]); 3835fa94a07fSbrendan kmem_free(list, sav->sav_count * sizeof (void *)); 383699653d4eSeschrock } 383799653d4eSeschrock 3838fa94a07fSbrendan spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 383906eeb2adSek nvlist_free(nvroot); 384099653d4eSeschrock 3841fa94a07fSbrendan sav->sav_sync = B_FALSE; 384299653d4eSeschrock } 384399653d4eSeschrock 384499653d4eSeschrock static void 384599653d4eSeschrock spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 384699653d4eSeschrock { 384799653d4eSeschrock nvlist_t *config; 384899653d4eSeschrock 384999653d4eSeschrock if (list_is_empty(&spa->spa_dirty_list)) 385099653d4eSeschrock return; 385199653d4eSeschrock 385299653d4eSeschrock config = spa_config_generate(spa, NULL, dmu_tx_get_txg(tx), B_FALSE); 385399653d4eSeschrock 385499653d4eSeschrock if (spa->spa_config_syncing) 385599653d4eSeschrock nvlist_free(spa->spa_config_syncing); 385699653d4eSeschrock spa->spa_config_syncing = config; 385799653d4eSeschrock 385899653d4eSeschrock spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 385999653d4eSeschrock } 386099653d4eSeschrock 3861990b4856Slling /* 3862990b4856Slling * Set zpool properties. 3863990b4856Slling */ 3864b1b8ab34Slling static void 3865ecd6cf80Smarks spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 3866b1b8ab34Slling { 3867b1b8ab34Slling spa_t *spa = arg1; 3868b1b8ab34Slling objset_t *mos = spa->spa_meta_objset; 3869990b4856Slling nvlist_t *nvp = arg2; 3870990b4856Slling nvpair_t *elem; 38713d7072f8Seschrock uint64_t intval; 38722f8aaab3Seschrock char *strval, *slash; 3873990b4856Slling zpool_prop_t prop; 3874990b4856Slling const char *propname; 3875990b4856Slling zprop_type_t proptype; 3876b1b8ab34Slling 3877990b4856Slling elem = NULL; 3878990b4856Slling while ((elem = nvlist_next_nvpair(nvp, elem))) { 3879990b4856Slling switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 3880990b4856Slling case ZPOOL_PROP_VERSION: 3881990b4856Slling /* 3882990b4856Slling * Only set version for non-zpool-creation cases 3883990b4856Slling * (set/import). spa_create() needs special care 3884990b4856Slling * for version setting. 3885990b4856Slling */ 3886990b4856Slling if (tx->tx_txg != TXG_INITIAL) { 3887990b4856Slling VERIFY(nvpair_value_uint64(elem, 3888990b4856Slling &intval) == 0); 3889990b4856Slling ASSERT(intval <= SPA_VERSION); 3890990b4856Slling ASSERT(intval >= spa_version(spa)); 3891990b4856Slling spa->spa_uberblock.ub_version = intval; 3892990b4856Slling vdev_config_dirty(spa->spa_root_vdev); 3893990b4856Slling } 3894ecd6cf80Smarks break; 3895990b4856Slling 3896990b4856Slling case ZPOOL_PROP_ALTROOT: 3897990b4856Slling /* 3898990b4856Slling * 'altroot' is a non-persistent property. It should 3899990b4856Slling * have been set temporarily at creation or import time. 3900990b4856Slling */ 3901990b4856Slling ASSERT(spa->spa_root != NULL); 3902b1b8ab34Slling break; 39033d7072f8Seschrock 39042f8aaab3Seschrock case ZPOOL_PROP_CACHEFILE: 3905990b4856Slling /* 39062f8aaab3Seschrock * 'cachefile' is a non-persistent property, but note 39072f8aaab3Seschrock * an async request that the config cache needs to be 39082f8aaab3Seschrock * udpated. 3909990b4856Slling */ 39102f8aaab3Seschrock VERIFY(nvpair_value_string(elem, &strval) == 0); 39112f8aaab3Seschrock if (spa->spa_config_dir) 39122f8aaab3Seschrock spa_strfree(spa->spa_config_dir); 39132f8aaab3Seschrock if (spa->spa_config_file) 39142f8aaab3Seschrock spa_strfree(spa->spa_config_file); 39152f8aaab3Seschrock 39162f8aaab3Seschrock if (strval[0] == '\0') { 39172f8aaab3Seschrock spa->spa_config_dir = NULL; 39182f8aaab3Seschrock spa->spa_config_file = NULL; 39192f8aaab3Seschrock } else if (strcmp(strval, "none") == 0) { 39202f8aaab3Seschrock spa->spa_config_dir = spa_strdup(strval); 39212f8aaab3Seschrock spa->spa_config_file = NULL; 39222f8aaab3Seschrock } else { 39232c32020fSeschrock /* 39242c32020fSeschrock * If the cachefile is in the root directory, 39252c32020fSeschrock * we will end up with an empty string for 39262c32020fSeschrock * spa_config_dir. This value is only ever 39272c32020fSeschrock * used when concatenated with '/', so an empty 39282c32020fSeschrock * string still behaves correctly and keeps the 39292c32020fSeschrock * rest of the code simple. 39302c32020fSeschrock */ 39312f8aaab3Seschrock slash = strrchr(strval, '/'); 39322f8aaab3Seschrock ASSERT(slash != NULL); 39332f8aaab3Seschrock *slash = '\0'; 39342c32020fSeschrock if (strcmp(strval, spa_config_dir) == 0 && 39352c32020fSeschrock strcmp(slash + 1, ZPOOL_CACHE_FILE) == 0) { 39362c32020fSeschrock spa->spa_config_dir = NULL; 39372c32020fSeschrock spa->spa_config_file = NULL; 39382c32020fSeschrock } else { 39392c32020fSeschrock spa->spa_config_dir = 39402c32020fSeschrock spa_strdup(strval); 39412c32020fSeschrock spa->spa_config_file = 39422c32020fSeschrock spa_strdup(slash + 1); 39432c32020fSeschrock } 39442f8aaab3Seschrock } 39452f8aaab3Seschrock spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 39463d7072f8Seschrock break; 3947990b4856Slling default: 3948990b4856Slling /* 3949990b4856Slling * Set pool property values in the poolprops mos object. 3950990b4856Slling */ 3951990b4856Slling mutex_enter(&spa->spa_props_lock); 3952990b4856Slling if (spa->spa_pool_props_object == 0) { 3953990b4856Slling objset_t *mos = spa->spa_meta_objset; 3954990b4856Slling 3955990b4856Slling VERIFY((spa->spa_pool_props_object = 3956990b4856Slling zap_create(mos, DMU_OT_POOL_PROPS, 3957990b4856Slling DMU_OT_NONE, 0, tx)) > 0); 3958990b4856Slling 3959990b4856Slling VERIFY(zap_update(mos, 3960990b4856Slling DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 3961990b4856Slling 8, 1, &spa->spa_pool_props_object, tx) 3962990b4856Slling == 0); 3963990b4856Slling } 3964990b4856Slling mutex_exit(&spa->spa_props_lock); 3965990b4856Slling 3966990b4856Slling /* normalize the property name */ 3967990b4856Slling propname = zpool_prop_to_name(prop); 3968990b4856Slling proptype = zpool_prop_get_type(prop); 3969990b4856Slling 3970990b4856Slling if (nvpair_type(elem) == DATA_TYPE_STRING) { 3971990b4856Slling ASSERT(proptype == PROP_TYPE_STRING); 3972990b4856Slling VERIFY(nvpair_value_string(elem, &strval) == 0); 3973990b4856Slling VERIFY(zap_update(mos, 3974990b4856Slling spa->spa_pool_props_object, propname, 3975990b4856Slling 1, strlen(strval) + 1, strval, tx) == 0); 3976990b4856Slling 3977990b4856Slling } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 3978990b4856Slling VERIFY(nvpair_value_uint64(elem, &intval) == 0); 3979990b4856Slling 3980990b4856Slling if (proptype == PROP_TYPE_INDEX) { 3981990b4856Slling const char *unused; 3982990b4856Slling VERIFY(zpool_prop_index_to_string( 3983990b4856Slling prop, intval, &unused) == 0); 3984990b4856Slling } 3985990b4856Slling VERIFY(zap_update(mos, 3986990b4856Slling spa->spa_pool_props_object, propname, 3987990b4856Slling 8, 1, &intval, tx) == 0); 3988990b4856Slling } else { 3989990b4856Slling ASSERT(0); /* not allowed */ 3990990b4856Slling } 3991990b4856Slling 39920a4e9518Sgw switch (prop) { 39930a4e9518Sgw case ZPOOL_PROP_DELEGATION: 3994990b4856Slling spa->spa_delegation = intval; 39950a4e9518Sgw break; 39960a4e9518Sgw case ZPOOL_PROP_BOOTFS: 3997990b4856Slling spa->spa_bootfs = intval; 39980a4e9518Sgw break; 39990a4e9518Sgw case ZPOOL_PROP_FAILUREMODE: 40000a4e9518Sgw spa->spa_failmode = intval; 40010a4e9518Sgw break; 40020a4e9518Sgw default: 40030a4e9518Sgw break; 40040a4e9518Sgw } 4005990b4856Slling } 4006990b4856Slling 4007990b4856Slling /* log internal history if this is not a zpool create */ 4008990b4856Slling if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY && 4009990b4856Slling tx->tx_txg != TXG_INITIAL) { 4010990b4856Slling spa_history_internal_log(LOG_POOL_PROPSET, 4011990b4856Slling spa, tx, cr, "%s %lld %s", 4012990b4856Slling nvpair_name(elem), intval, spa->spa_name); 4013b1b8ab34Slling } 4014b1b8ab34Slling } 4015b1b8ab34Slling } 4016b1b8ab34Slling 4017fa9e4066Sahrens /* 4018fa9e4066Sahrens * Sync the specified transaction group. New blocks may be dirtied as 4019fa9e4066Sahrens * part of the process, so we iterate until it converges. 4020fa9e4066Sahrens */ 4021fa9e4066Sahrens void 4022fa9e4066Sahrens spa_sync(spa_t *spa, uint64_t txg) 4023fa9e4066Sahrens { 4024fa9e4066Sahrens dsl_pool_t *dp = spa->spa_dsl_pool; 4025fa9e4066Sahrens objset_t *mos = spa->spa_meta_objset; 4026fa9e4066Sahrens bplist_t *bpl = &spa->spa_sync_bplist; 40270373e76bSbonwick vdev_t *rvd = spa->spa_root_vdev; 4028fa9e4066Sahrens vdev_t *vd; 402917f17c2dSbonwick vdev_t *svd[SPA_DVAS_PER_BP]; 403017f17c2dSbonwick int svdcount = 0; 4031fa9e4066Sahrens dmu_tx_t *tx; 4032fa9e4066Sahrens int dirty_vdevs; 4033fa9e4066Sahrens 4034fa9e4066Sahrens /* 4035fa9e4066Sahrens * Lock out configuration changes. 4036fa9e4066Sahrens */ 4037ea8dc4b6Seschrock spa_config_enter(spa, RW_READER, FTAG); 4038fa9e4066Sahrens 4039fa9e4066Sahrens spa->spa_syncing_txg = txg; 4040fa9e4066Sahrens spa->spa_sync_pass = 0; 4041fa9e4066Sahrens 4042ea8dc4b6Seschrock VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj)); 4043fa9e4066Sahrens 404499653d4eSeschrock tx = dmu_tx_create_assigned(dp, txg); 404599653d4eSeschrock 404699653d4eSeschrock /* 4047e7437265Sahrens * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 404899653d4eSeschrock * set spa_deflate if we have no raid-z vdevs. 404999653d4eSeschrock */ 4050e7437265Sahrens if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 4051e7437265Sahrens spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 405299653d4eSeschrock int i; 405399653d4eSeschrock 405499653d4eSeschrock for (i = 0; i < rvd->vdev_children; i++) { 405599653d4eSeschrock vd = rvd->vdev_child[i]; 405699653d4eSeschrock if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 405799653d4eSeschrock break; 405899653d4eSeschrock } 405999653d4eSeschrock if (i == rvd->vdev_children) { 406099653d4eSeschrock spa->spa_deflate = TRUE; 406199653d4eSeschrock VERIFY(0 == zap_add(spa->spa_meta_objset, 406299653d4eSeschrock DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 406399653d4eSeschrock sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 406499653d4eSeschrock } 406599653d4eSeschrock } 406699653d4eSeschrock 4067fa9e4066Sahrens /* 4068fa9e4066Sahrens * If anything has changed in this txg, push the deferred frees 4069fa9e4066Sahrens * from the previous txg. If not, leave them alone so that we 4070fa9e4066Sahrens * don't generate work on an otherwise idle system. 4071fa9e4066Sahrens */ 4072fa9e4066Sahrens if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 40731615a317Sek !txg_list_empty(&dp->dp_dirty_dirs, txg) || 40741615a317Sek !txg_list_empty(&dp->dp_sync_tasks, txg)) 4075fa9e4066Sahrens spa_sync_deferred_frees(spa, txg); 4076fa9e4066Sahrens 4077fa9e4066Sahrens /* 4078fa9e4066Sahrens * Iterate to convergence. 4079fa9e4066Sahrens */ 4080fa9e4066Sahrens do { 4081fa9e4066Sahrens spa->spa_sync_pass++; 4082fa9e4066Sahrens 4083fa9e4066Sahrens spa_sync_config_object(spa, tx); 4084fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_spares, tx, 4085fa94a07fSbrendan ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 4086fa94a07fSbrendan spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 4087fa94a07fSbrendan ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 4088ea8dc4b6Seschrock spa_errlog_sync(spa, txg); 4089fa9e4066Sahrens dsl_pool_sync(dp, txg); 4090fa9e4066Sahrens 4091fa9e4066Sahrens dirty_vdevs = 0; 4092fa9e4066Sahrens while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) { 4093fa9e4066Sahrens vdev_sync(vd, txg); 4094fa9e4066Sahrens dirty_vdevs++; 4095fa9e4066Sahrens } 4096fa9e4066Sahrens 4097fa9e4066Sahrens bplist_sync(bpl, tx); 4098fa9e4066Sahrens } while (dirty_vdevs); 4099fa9e4066Sahrens 4100fa9e4066Sahrens bplist_close(bpl); 4101fa9e4066Sahrens 4102fa9e4066Sahrens dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass); 4103fa9e4066Sahrens 4104fa9e4066Sahrens /* 4105fa9e4066Sahrens * Rewrite the vdev configuration (which includes the uberblock) 4106fa9e4066Sahrens * to commit the transaction group. 41070373e76bSbonwick * 410817f17c2dSbonwick * If there are no dirty vdevs, we sync the uberblock to a few 410917f17c2dSbonwick * random top-level vdevs that are known to be visible in the 411017f17c2dSbonwick * config cache (see spa_vdev_add() for details). If there *are* 411117f17c2dSbonwick * dirty vdevs -- or if the sync to our random subset fails -- 411217f17c2dSbonwick * then sync the uberblock to all vdevs. 41130373e76bSbonwick */ 411417f17c2dSbonwick if (list_is_empty(&spa->spa_dirty_list)) { 41150373e76bSbonwick int children = rvd->vdev_children; 41160373e76bSbonwick int c0 = spa_get_random(children); 41170373e76bSbonwick int c; 41180373e76bSbonwick 41190373e76bSbonwick for (c = 0; c < children; c++) { 41200373e76bSbonwick vd = rvd->vdev_child[(c0 + c) % children]; 412117f17c2dSbonwick if (vd->vdev_ms_array == 0 || vd->vdev_islog) 41220373e76bSbonwick continue; 412317f17c2dSbonwick svd[svdcount++] = vd; 412417f17c2dSbonwick if (svdcount == SPA_DVAS_PER_BP) 41250373e76bSbonwick break; 41260373e76bSbonwick } 41270373e76bSbonwick } 412817f17c2dSbonwick if (svdcount == 0 || vdev_config_sync(svd, svdcount, txg) != 0) 412917f17c2dSbonwick VERIFY3U(vdev_config_sync(rvd->vdev_child, 413017f17c2dSbonwick rvd->vdev_children, txg), ==, 0); 41310373e76bSbonwick 413299653d4eSeschrock dmu_tx_commit(tx); 413399653d4eSeschrock 41340373e76bSbonwick /* 41350373e76bSbonwick * Clear the dirty config list. 4136fa9e4066Sahrens */ 41370373e76bSbonwick while ((vd = list_head(&spa->spa_dirty_list)) != NULL) 41380373e76bSbonwick vdev_config_clean(vd); 41390373e76bSbonwick 41400373e76bSbonwick /* 41410373e76bSbonwick * Now that the new config has synced transactionally, 41420373e76bSbonwick * let it become visible to the config cache. 41430373e76bSbonwick */ 41440373e76bSbonwick if (spa->spa_config_syncing != NULL) { 41450373e76bSbonwick spa_config_set(spa, spa->spa_config_syncing); 41460373e76bSbonwick spa->spa_config_txg = txg; 41470373e76bSbonwick spa->spa_config_syncing = NULL; 41480373e76bSbonwick } 4149fa9e4066Sahrens 4150fa9e4066Sahrens /* 4151fa9e4066Sahrens * Make a stable copy of the fully synced uberblock. 4152fa9e4066Sahrens * We use this as the root for pool traversals. 4153fa9e4066Sahrens */ 4154fa9e4066Sahrens spa->spa_traverse_wanted = 1; /* tells traverse_more() to stop */ 4155fa9e4066Sahrens 4156fa9e4066Sahrens spa_scrub_suspend(spa); /* stop scrubbing and finish I/Os */ 4157fa9e4066Sahrens 4158fa9e4066Sahrens rw_enter(&spa->spa_traverse_lock, RW_WRITER); 4159fa9e4066Sahrens spa->spa_traverse_wanted = 0; 4160fa9e4066Sahrens spa->spa_ubsync = spa->spa_uberblock; 4161fa9e4066Sahrens rw_exit(&spa->spa_traverse_lock); 4162fa9e4066Sahrens 4163fa9e4066Sahrens spa_scrub_resume(spa); /* resume scrub with new ubsync */ 4164fa9e4066Sahrens 4165fa9e4066Sahrens /* 4166fa9e4066Sahrens * Clean up the ZIL records for the synced txg. 4167fa9e4066Sahrens */ 4168fa9e4066Sahrens dsl_pool_zil_clean(dp); 4169fa9e4066Sahrens 4170fa9e4066Sahrens /* 4171fa9e4066Sahrens * Update usable space statistics. 4172fa9e4066Sahrens */ 4173fa9e4066Sahrens while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 4174fa9e4066Sahrens vdev_sync_done(vd, txg); 4175fa9e4066Sahrens 4176fa9e4066Sahrens /* 4177fa9e4066Sahrens * It had better be the case that we didn't dirty anything 417899653d4eSeschrock * since vdev_config_sync(). 4179fa9e4066Sahrens */ 4180fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 4181fa9e4066Sahrens ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 4182fa9e4066Sahrens ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 4183fa9e4066Sahrens ASSERT(bpl->bpl_queue == NULL); 4184fa9e4066Sahrens 4185ea8dc4b6Seschrock spa_config_exit(spa, FTAG); 4186ea8dc4b6Seschrock 4187ea8dc4b6Seschrock /* 4188ea8dc4b6Seschrock * If any async tasks have been requested, kick them off. 4189ea8dc4b6Seschrock */ 4190ea8dc4b6Seschrock spa_async_dispatch(spa); 4191fa9e4066Sahrens } 4192fa9e4066Sahrens 4193fa9e4066Sahrens /* 4194fa9e4066Sahrens * Sync all pools. We don't want to hold the namespace lock across these 4195fa9e4066Sahrens * operations, so we take a reference on the spa_t and drop the lock during the 4196fa9e4066Sahrens * sync. 4197fa9e4066Sahrens */ 4198fa9e4066Sahrens void 4199fa9e4066Sahrens spa_sync_allpools(void) 4200fa9e4066Sahrens { 4201fa9e4066Sahrens spa_t *spa = NULL; 4202fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 4203fa9e4066Sahrens while ((spa = spa_next(spa)) != NULL) { 4204fa9e4066Sahrens if (spa_state(spa) != POOL_STATE_ACTIVE) 4205fa9e4066Sahrens continue; 4206fa9e4066Sahrens spa_open_ref(spa, FTAG); 4207fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4208fa9e4066Sahrens txg_wait_synced(spa_get_dsl(spa), 0); 4209fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 4210fa9e4066Sahrens spa_close(spa, FTAG); 4211fa9e4066Sahrens } 4212fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4213fa9e4066Sahrens } 4214fa9e4066Sahrens 4215fa9e4066Sahrens /* 4216fa9e4066Sahrens * ========================================================================== 4217fa9e4066Sahrens * Miscellaneous routines 4218fa9e4066Sahrens * ========================================================================== 4219fa9e4066Sahrens */ 4220fa9e4066Sahrens 4221fa9e4066Sahrens /* 4222fa9e4066Sahrens * Remove all pools in the system. 4223fa9e4066Sahrens */ 4224fa9e4066Sahrens void 4225fa9e4066Sahrens spa_evict_all(void) 4226fa9e4066Sahrens { 4227fa9e4066Sahrens spa_t *spa; 4228fa9e4066Sahrens 4229fa9e4066Sahrens /* 4230fa9e4066Sahrens * Remove all cached state. All pools should be closed now, 4231fa9e4066Sahrens * so every spa in the AVL tree should be unreferenced. 4232fa9e4066Sahrens */ 4233fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 4234fa9e4066Sahrens while ((spa = spa_next(NULL)) != NULL) { 4235fa9e4066Sahrens /* 4236ea8dc4b6Seschrock * Stop async tasks. The async thread may need to detach 4237ea8dc4b6Seschrock * a device that's been replaced, which requires grabbing 4238ea8dc4b6Seschrock * spa_namespace_lock, so we must drop it here. 4239fa9e4066Sahrens */ 4240fa9e4066Sahrens spa_open_ref(spa, FTAG); 4241fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4242ea8dc4b6Seschrock spa_async_suspend(spa); 4243fa9e4066Sahrens mutex_enter(&spa_namespace_lock); 4244bb8b5132Sek VERIFY(spa_scrub(spa, POOL_SCRUB_NONE, B_TRUE) == 0); 4245fa9e4066Sahrens spa_close(spa, FTAG); 4246fa9e4066Sahrens 4247fa9e4066Sahrens if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4248fa9e4066Sahrens spa_unload(spa); 4249fa9e4066Sahrens spa_deactivate(spa); 4250fa9e4066Sahrens } 4251fa9e4066Sahrens spa_remove(spa); 4252fa9e4066Sahrens } 4253fa9e4066Sahrens mutex_exit(&spa_namespace_lock); 4254fa9e4066Sahrens } 4255ea8dc4b6Seschrock 4256ea8dc4b6Seschrock vdev_t * 4257ea8dc4b6Seschrock spa_lookup_by_guid(spa_t *spa, uint64_t guid) 4258ea8dc4b6Seschrock { 4259ea8dc4b6Seschrock return (vdev_lookup_by_guid(spa->spa_root_vdev, guid)); 4260ea8dc4b6Seschrock } 4261eaca9bbdSeschrock 4262eaca9bbdSeschrock void 4263990b4856Slling spa_upgrade(spa_t *spa, uint64_t version) 4264eaca9bbdSeschrock { 4265eaca9bbdSeschrock spa_config_enter(spa, RW_WRITER, FTAG); 4266eaca9bbdSeschrock 4267eaca9bbdSeschrock /* 4268eaca9bbdSeschrock * This should only be called for a non-faulted pool, and since a 4269eaca9bbdSeschrock * future version would result in an unopenable pool, this shouldn't be 4270eaca9bbdSeschrock * possible. 4271eaca9bbdSeschrock */ 4272e7437265Sahrens ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION); 4273990b4856Slling ASSERT(version >= spa->spa_uberblock.ub_version); 4274eaca9bbdSeschrock 4275990b4856Slling spa->spa_uberblock.ub_version = version; 4276eaca9bbdSeschrock vdev_config_dirty(spa->spa_root_vdev); 4277eaca9bbdSeschrock 4278eaca9bbdSeschrock spa_config_exit(spa, FTAG); 427999653d4eSeschrock 428099653d4eSeschrock txg_wait_synced(spa_get_dsl(spa), 0); 428199653d4eSeschrock } 428299653d4eSeschrock 428399653d4eSeschrock boolean_t 428499653d4eSeschrock spa_has_spare(spa_t *spa, uint64_t guid) 428599653d4eSeschrock { 428699653d4eSeschrock int i; 428739c23413Seschrock uint64_t spareguid; 4288fa94a07fSbrendan spa_aux_vdev_t *sav = &spa->spa_spares; 428999653d4eSeschrock 4290fa94a07fSbrendan for (i = 0; i < sav->sav_count; i++) 4291fa94a07fSbrendan if (sav->sav_vdevs[i]->vdev_guid == guid) 429299653d4eSeschrock return (B_TRUE); 429399653d4eSeschrock 4294fa94a07fSbrendan for (i = 0; i < sav->sav_npending; i++) { 4295fa94a07fSbrendan if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 4296fa94a07fSbrendan &spareguid) == 0 && spareguid == guid) 429739c23413Seschrock return (B_TRUE); 429839c23413Seschrock } 429939c23413Seschrock 430099653d4eSeschrock return (B_FALSE); 4301eaca9bbdSeschrock } 4302b1b8ab34Slling 43033d7072f8Seschrock /* 43043d7072f8Seschrock * Post a sysevent corresponding to the given event. The 'name' must be one of 43053d7072f8Seschrock * the event definitions in sys/sysevent/eventdefs.h. The payload will be 43063d7072f8Seschrock * filled in from the spa and (optionally) the vdev. This doesn't do anything 43073d7072f8Seschrock * in the userland libzpool, as we don't want consumers to misinterpret ztest 43083d7072f8Seschrock * or zdb as real changes. 43093d7072f8Seschrock */ 43103d7072f8Seschrock void 43113d7072f8Seschrock spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 43123d7072f8Seschrock { 43133d7072f8Seschrock #ifdef _KERNEL 43143d7072f8Seschrock sysevent_t *ev; 43153d7072f8Seschrock sysevent_attr_list_t *attr = NULL; 43163d7072f8Seschrock sysevent_value_t value; 43173d7072f8Seschrock sysevent_id_t eid; 43183d7072f8Seschrock 43193d7072f8Seschrock ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 43203d7072f8Seschrock SE_SLEEP); 43213d7072f8Seschrock 43223d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 43233d7072f8Seschrock value.value.sv_string = spa_name(spa); 43243d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 43253d7072f8Seschrock goto done; 43263d7072f8Seschrock 43273d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 43283d7072f8Seschrock value.value.sv_uint64 = spa_guid(spa); 43293d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 43303d7072f8Seschrock goto done; 43313d7072f8Seschrock 43323d7072f8Seschrock if (vd) { 43333d7072f8Seschrock value.value_type = SE_DATA_TYPE_UINT64; 43343d7072f8Seschrock value.value.sv_uint64 = vd->vdev_guid; 43353d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 43363d7072f8Seschrock SE_SLEEP) != 0) 43373d7072f8Seschrock goto done; 43383d7072f8Seschrock 43393d7072f8Seschrock if (vd->vdev_path) { 43403d7072f8Seschrock value.value_type = SE_DATA_TYPE_STRING; 43413d7072f8Seschrock value.value.sv_string = vd->vdev_path; 43423d7072f8Seschrock if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 43433d7072f8Seschrock &value, SE_SLEEP) != 0) 43443d7072f8Seschrock goto done; 43453d7072f8Seschrock } 43463d7072f8Seschrock } 43473d7072f8Seschrock 4348*b01c3b58Seschrock if (sysevent_attach_attributes(ev, attr) != 0) 4349*b01c3b58Seschrock goto done; 4350*b01c3b58Seschrock attr = NULL; 4351*b01c3b58Seschrock 43523d7072f8Seschrock (void) log_sysevent(ev, SE_SLEEP, &eid); 43533d7072f8Seschrock 43543d7072f8Seschrock done: 43553d7072f8Seschrock if (attr) 43563d7072f8Seschrock sysevent_free_attr(attr); 43573d7072f8Seschrock sysevent_free(ev); 43583d7072f8Seschrock #endif 43593d7072f8Seschrock } 4360