spa.c revision 204073
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27/* 28 * This file contains all the routines used when modifying on-disk SPA state. 29 * This includes opening, importing, destroying, exporting a pool, and syncing a 30 * pool. 31 */ 32 33#include <sys/zfs_context.h> 34#include <sys/fm/fs/zfs.h> 35#include <sys/spa_impl.h> 36#include <sys/zio.h> 37#include <sys/zio_checksum.h> 38#include <sys/zio_compress.h> 39#include <sys/dmu.h> 40#include <sys/dmu_tx.h> 41#include <sys/zap.h> 42#include <sys/zil.h> 43#include <sys/vdev_impl.h> 44#include <sys/metaslab.h> 45#include <sys/uberblock_impl.h> 46#include <sys/txg.h> 47#include <sys/avl.h> 48#include <sys/dmu_traverse.h> 49#include <sys/dmu_objset.h> 50#include <sys/unique.h> 51#include <sys/dsl_pool.h> 52#include <sys/dsl_dataset.h> 53#include <sys/dsl_dir.h> 54#include <sys/dsl_prop.h> 55#include <sys/dsl_synctask.h> 56#include <sys/fs/zfs.h> 57#include <sys/arc.h> 58#include <sys/callb.h> 59#include <sys/sunddi.h> 60#include <sys/spa_boot.h> 61 62#include "zfs_prop.h" 63#include "zfs_comutil.h" 64 65/* Check hostid on import? */ 66static int check_hostid = 1; 67 68SYSCTL_DECL(_vfs_zfs); 69TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 70SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 71 "Check hostid on import?"); 72 73int zio_taskq_threads[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 74 /* ISSUE INTR */ 75 { 1, 1 }, /* ZIO_TYPE_NULL */ 76 { 1, 8 }, /* ZIO_TYPE_READ */ 77 { 8, 1 }, /* ZIO_TYPE_WRITE */ 78 { 1, 1 }, /* ZIO_TYPE_FREE */ 79 { 1, 1 }, /* ZIO_TYPE_CLAIM */ 80 { 1, 1 }, /* ZIO_TYPE_IOCTL */ 81}; 82 83static void spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx); 84static boolean_t spa_has_active_shared_spare(spa_t *spa); 85 86/* 87 * ========================================================================== 88 * SPA properties routines 89 * ========================================================================== 90 */ 91 92/* 93 * Add a (source=src, propname=propval) list to an nvlist. 94 */ 95static void 96spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 97 uint64_t intval, zprop_source_t src) 98{ 99 const char *propname = zpool_prop_to_name(prop); 100 nvlist_t *propval; 101 102 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 103 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 104 105 if (strval != NULL) 106 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 107 else 108 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 109 110 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 111 nvlist_free(propval); 112} 113 114/* 115 * Get property values from the spa configuration. 116 */ 117static void 118spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 119{ 120 uint64_t size = spa_get_space(spa); 121 uint64_t used = spa_get_alloc(spa); 122 uint64_t cap, version; 123 zprop_source_t src = ZPROP_SRC_NONE; 124 spa_config_dirent_t *dp; 125 126 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 127 128 /* 129 * readonly properties 130 */ 131 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 132 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 133 spa_prop_add_list(*nvp, ZPOOL_PROP_USED, NULL, used, src); 134 spa_prop_add_list(*nvp, ZPOOL_PROP_AVAILABLE, NULL, size - used, src); 135 136 cap = (size == 0) ? 0 : (used * 100 / size); 137 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 138 139 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 140 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 141 spa->spa_root_vdev->vdev_state, src); 142 143 /* 144 * settable properties that are not stored in the pool property object. 145 */ 146 version = spa_version(spa); 147 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 148 src = ZPROP_SRC_DEFAULT; 149 else 150 src = ZPROP_SRC_LOCAL; 151 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 152 153 if (spa->spa_root != NULL) 154 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 155 0, ZPROP_SRC_LOCAL); 156 157 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 158 if (dp->scd_path == NULL) { 159 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 160 "none", 0, ZPROP_SRC_LOCAL); 161 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 162 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 163 dp->scd_path, 0, ZPROP_SRC_LOCAL); 164 } 165 } 166} 167 168/* 169 * Get zpool property values. 170 */ 171int 172spa_prop_get(spa_t *spa, nvlist_t **nvp) 173{ 174 zap_cursor_t zc; 175 zap_attribute_t za; 176 objset_t *mos = spa->spa_meta_objset; 177 int err; 178 179 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 180 181 mutex_enter(&spa->spa_props_lock); 182 183 /* 184 * Get properties from the spa config. 185 */ 186 spa_prop_get_config(spa, nvp); 187 188 /* If no pool property object, no more prop to get. */ 189 if (spa->spa_pool_props_object == 0) { 190 mutex_exit(&spa->spa_props_lock); 191 return (0); 192 } 193 194 /* 195 * Get properties from the MOS pool property object. 196 */ 197 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 198 (err = zap_cursor_retrieve(&zc, &za)) == 0; 199 zap_cursor_advance(&zc)) { 200 uint64_t intval = 0; 201 char *strval = NULL; 202 zprop_source_t src = ZPROP_SRC_DEFAULT; 203 zpool_prop_t prop; 204 205 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 206 continue; 207 208 switch (za.za_integer_length) { 209 case 8: 210 /* integer property */ 211 if (za.za_first_integer != 212 zpool_prop_default_numeric(prop)) 213 src = ZPROP_SRC_LOCAL; 214 215 if (prop == ZPOOL_PROP_BOOTFS) { 216 dsl_pool_t *dp; 217 dsl_dataset_t *ds = NULL; 218 219 dp = spa_get_dsl(spa); 220 rw_enter(&dp->dp_config_rwlock, RW_READER); 221 if (err = dsl_dataset_hold_obj(dp, 222 za.za_first_integer, FTAG, &ds)) { 223 rw_exit(&dp->dp_config_rwlock); 224 break; 225 } 226 227 strval = kmem_alloc( 228 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 229 KM_SLEEP); 230 dsl_dataset_name(ds, strval); 231 dsl_dataset_rele(ds, FTAG); 232 rw_exit(&dp->dp_config_rwlock); 233 } else { 234 strval = NULL; 235 intval = za.za_first_integer; 236 } 237 238 spa_prop_add_list(*nvp, prop, strval, intval, src); 239 240 if (strval != NULL) 241 kmem_free(strval, 242 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 243 244 break; 245 246 case 1: 247 /* string property */ 248 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 249 err = zap_lookup(mos, spa->spa_pool_props_object, 250 za.za_name, 1, za.za_num_integers, strval); 251 if (err) { 252 kmem_free(strval, za.za_num_integers); 253 break; 254 } 255 spa_prop_add_list(*nvp, prop, strval, 0, src); 256 kmem_free(strval, za.za_num_integers); 257 break; 258 259 default: 260 break; 261 } 262 } 263 zap_cursor_fini(&zc); 264 mutex_exit(&spa->spa_props_lock); 265out: 266 if (err && err != ENOENT) { 267 nvlist_free(*nvp); 268 *nvp = NULL; 269 return (err); 270 } 271 272 return (0); 273} 274 275/* 276 * Validate the given pool properties nvlist and modify the list 277 * for the property values to be set. 278 */ 279static int 280spa_prop_validate(spa_t *spa, nvlist_t *props) 281{ 282 nvpair_t *elem; 283 int error = 0, reset_bootfs = 0; 284 uint64_t objnum; 285 286 elem = NULL; 287 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 288 zpool_prop_t prop; 289 char *propname, *strval; 290 uint64_t intval; 291 objset_t *os; 292 char *slash; 293 294 propname = nvpair_name(elem); 295 296 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) 297 return (EINVAL); 298 299 switch (prop) { 300 case ZPOOL_PROP_VERSION: 301 error = nvpair_value_uint64(elem, &intval); 302 if (!error && 303 (intval < spa_version(spa) || intval > SPA_VERSION)) 304 error = EINVAL; 305 break; 306 307 case ZPOOL_PROP_DELEGATION: 308 case ZPOOL_PROP_AUTOREPLACE: 309 case ZPOOL_PROP_LISTSNAPS: 310 error = nvpair_value_uint64(elem, &intval); 311 if (!error && intval > 1) 312 error = EINVAL; 313 break; 314 315 case ZPOOL_PROP_BOOTFS: 316 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 317 error = ENOTSUP; 318 break; 319 } 320 321 /* 322 * Make sure the vdev config is bootable 323 */ 324 if (!vdev_is_bootable(spa->spa_root_vdev)) { 325 error = ENOTSUP; 326 break; 327 } 328 329 reset_bootfs = 1; 330 331 error = nvpair_value_string(elem, &strval); 332 333 if (!error) { 334 uint64_t compress; 335 336 if (strval == NULL || strval[0] == '\0') { 337 objnum = zpool_prop_default_numeric( 338 ZPOOL_PROP_BOOTFS); 339 break; 340 } 341 342 if (error = dmu_objset_open(strval, DMU_OST_ZFS, 343 DS_MODE_USER | DS_MODE_READONLY, &os)) 344 break; 345 346 /* We don't support gzip bootable datasets */ 347 if ((error = dsl_prop_get_integer(strval, 348 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 349 &compress, NULL)) == 0 && 350 !BOOTFS_COMPRESS_VALID(compress)) { 351 error = ENOTSUP; 352 } else { 353 objnum = dmu_objset_id(os); 354 } 355 dmu_objset_close(os); 356 } 357 break; 358 359 case ZPOOL_PROP_FAILUREMODE: 360 error = nvpair_value_uint64(elem, &intval); 361 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 362 intval > ZIO_FAILURE_MODE_PANIC)) 363 error = EINVAL; 364 365 /* 366 * This is a special case which only occurs when 367 * the pool has completely failed. This allows 368 * the user to change the in-core failmode property 369 * without syncing it out to disk (I/Os might 370 * currently be blocked). We do this by returning 371 * EIO to the caller (spa_prop_set) to trick it 372 * into thinking we encountered a property validation 373 * error. 374 */ 375 if (!error && spa_suspended(spa)) { 376 spa->spa_failmode = intval; 377 error = EIO; 378 } 379 break; 380 381 case ZPOOL_PROP_CACHEFILE: 382 if ((error = nvpair_value_string(elem, &strval)) != 0) 383 break; 384 385 if (strval[0] == '\0') 386 break; 387 388 if (strcmp(strval, "none") == 0) 389 break; 390 391 if (strval[0] != '/') { 392 error = EINVAL; 393 break; 394 } 395 396 slash = strrchr(strval, '/'); 397 ASSERT(slash != NULL); 398 399 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 400 strcmp(slash, "/..") == 0) 401 error = EINVAL; 402 break; 403 } 404 405 if (error) 406 break; 407 } 408 409 if (!error && reset_bootfs) { 410 error = nvlist_remove(props, 411 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 412 413 if (!error) { 414 error = nvlist_add_uint64(props, 415 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 416 } 417 } 418 419 return (error); 420} 421 422int 423spa_prop_set(spa_t *spa, nvlist_t *nvp) 424{ 425 int error; 426 427 if ((error = spa_prop_validate(spa, nvp)) != 0) 428 return (error); 429 430 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, 431 spa, nvp, 3)); 432} 433 434/* 435 * If the bootfs property value is dsobj, clear it. 436 */ 437void 438spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 439{ 440 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 441 VERIFY(zap_remove(spa->spa_meta_objset, 442 spa->spa_pool_props_object, 443 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 444 spa->spa_bootfs = 0; 445 } 446} 447 448/* 449 * ========================================================================== 450 * SPA state manipulation (open/create/destroy/import/export) 451 * ========================================================================== 452 */ 453 454static int 455spa_error_entry_compare(const void *a, const void *b) 456{ 457 spa_error_entry_t *sa = (spa_error_entry_t *)a; 458 spa_error_entry_t *sb = (spa_error_entry_t *)b; 459 int ret; 460 461 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 462 sizeof (zbookmark_t)); 463 464 if (ret < 0) 465 return (-1); 466 else if (ret > 0) 467 return (1); 468 else 469 return (0); 470} 471 472/* 473 * Utility function which retrieves copies of the current logs and 474 * re-initializes them in the process. 475 */ 476void 477spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 478{ 479 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 480 481 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 482 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 483 484 avl_create(&spa->spa_errlist_scrub, 485 spa_error_entry_compare, sizeof (spa_error_entry_t), 486 offsetof(spa_error_entry_t, se_avl)); 487 avl_create(&spa->spa_errlist_last, 488 spa_error_entry_compare, sizeof (spa_error_entry_t), 489 offsetof(spa_error_entry_t, se_avl)); 490} 491 492/* 493 * Activate an uninitialized pool. 494 */ 495static void 496spa_activate(spa_t *spa) 497{ 498 499 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 500 501 spa->spa_state = POOL_STATE_ACTIVE; 502 503 spa->spa_normal_class = metaslab_class_create(); 504 spa->spa_log_class = metaslab_class_create(); 505 506 for (int t = 0; t < ZIO_TYPES; t++) { 507 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 508 spa->spa_zio_taskq[t][q] = taskq_create("spa_zio", 509 zio_taskq_threads[t][q], maxclsyspri, 50, 510 INT_MAX, TASKQ_PREPOPULATE); 511 } 512 } 513 514 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 515 offsetof(vdev_t, vdev_config_dirty_node)); 516 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 517 offsetof(vdev_t, vdev_state_dirty_node)); 518 519 txg_list_create(&spa->spa_vdev_txg_list, 520 offsetof(struct vdev, vdev_txg_node)); 521 522 avl_create(&spa->spa_errlist_scrub, 523 spa_error_entry_compare, sizeof (spa_error_entry_t), 524 offsetof(spa_error_entry_t, se_avl)); 525 avl_create(&spa->spa_errlist_last, 526 spa_error_entry_compare, sizeof (spa_error_entry_t), 527 offsetof(spa_error_entry_t, se_avl)); 528} 529 530/* 531 * Opposite of spa_activate(). 532 */ 533static void 534spa_deactivate(spa_t *spa) 535{ 536 ASSERT(spa->spa_sync_on == B_FALSE); 537 ASSERT(spa->spa_dsl_pool == NULL); 538 ASSERT(spa->spa_root_vdev == NULL); 539 540 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 541 542 txg_list_destroy(&spa->spa_vdev_txg_list); 543 544 list_destroy(&spa->spa_config_dirty_list); 545 list_destroy(&spa->spa_state_dirty_list); 546 547 for (int t = 0; t < ZIO_TYPES; t++) { 548 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 549 taskq_destroy(spa->spa_zio_taskq[t][q]); 550 spa->spa_zio_taskq[t][q] = NULL; 551 } 552 } 553 554 metaslab_class_destroy(spa->spa_normal_class); 555 spa->spa_normal_class = NULL; 556 557 metaslab_class_destroy(spa->spa_log_class); 558 spa->spa_log_class = NULL; 559 560 /* 561 * If this was part of an import or the open otherwise failed, we may 562 * still have errors left in the queues. Empty them just in case. 563 */ 564 spa_errlog_drain(spa); 565 566 avl_destroy(&spa->spa_errlist_scrub); 567 avl_destroy(&spa->spa_errlist_last); 568 569 spa->spa_state = POOL_STATE_UNINITIALIZED; 570} 571 572/* 573 * Verify a pool configuration, and construct the vdev tree appropriately. This 574 * will create all the necessary vdevs in the appropriate layout, with each vdev 575 * in the CLOSED state. This will prep the pool before open/creation/import. 576 * All vdev validation is done by the vdev_alloc() routine. 577 */ 578static int 579spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 580 uint_t id, int atype) 581{ 582 nvlist_t **child; 583 uint_t c, children; 584 int error; 585 586 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 587 return (error); 588 589 if ((*vdp)->vdev_ops->vdev_op_leaf) 590 return (0); 591 592 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 593 &child, &children); 594 595 if (error == ENOENT) 596 return (0); 597 598 if (error) { 599 vdev_free(*vdp); 600 *vdp = NULL; 601 return (EINVAL); 602 } 603 604 for (c = 0; c < children; c++) { 605 vdev_t *vd; 606 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 607 atype)) != 0) { 608 vdev_free(*vdp); 609 *vdp = NULL; 610 return (error); 611 } 612 } 613 614 ASSERT(*vdp != NULL); 615 616 return (0); 617} 618 619/* 620 * Opposite of spa_load(). 621 */ 622static void 623spa_unload(spa_t *spa) 624{ 625 int i; 626 627 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 628 629 /* 630 * Stop async tasks. 631 */ 632 spa_async_suspend(spa); 633 634 /* 635 * Stop syncing. 636 */ 637 if (spa->spa_sync_on) { 638 txg_sync_stop(spa->spa_dsl_pool); 639 spa->spa_sync_on = B_FALSE; 640 } 641 642 /* 643 * Wait for any outstanding async I/O to complete. 644 */ 645 mutex_enter(&spa->spa_async_root_lock); 646 while (spa->spa_async_root_count != 0) 647 cv_wait(&spa->spa_async_root_cv, &spa->spa_async_root_lock); 648 mutex_exit(&spa->spa_async_root_lock); 649 650 /* 651 * Drop and purge level 2 cache 652 */ 653 spa_l2cache_drop(spa); 654 655 /* 656 * Close the dsl pool. 657 */ 658 if (spa->spa_dsl_pool) { 659 dsl_pool_close(spa->spa_dsl_pool); 660 spa->spa_dsl_pool = NULL; 661 } 662 663 /* 664 * Close all vdevs. 665 */ 666 if (spa->spa_root_vdev) 667 vdev_free(spa->spa_root_vdev); 668 ASSERT(spa->spa_root_vdev == NULL); 669 670 for (i = 0; i < spa->spa_spares.sav_count; i++) 671 vdev_free(spa->spa_spares.sav_vdevs[i]); 672 if (spa->spa_spares.sav_vdevs) { 673 kmem_free(spa->spa_spares.sav_vdevs, 674 spa->spa_spares.sav_count * sizeof (void *)); 675 spa->spa_spares.sav_vdevs = NULL; 676 } 677 if (spa->spa_spares.sav_config) { 678 nvlist_free(spa->spa_spares.sav_config); 679 spa->spa_spares.sav_config = NULL; 680 } 681 spa->spa_spares.sav_count = 0; 682 683 for (i = 0; i < spa->spa_l2cache.sav_count; i++) 684 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 685 if (spa->spa_l2cache.sav_vdevs) { 686 kmem_free(spa->spa_l2cache.sav_vdevs, 687 spa->spa_l2cache.sav_count * sizeof (void *)); 688 spa->spa_l2cache.sav_vdevs = NULL; 689 } 690 if (spa->spa_l2cache.sav_config) { 691 nvlist_free(spa->spa_l2cache.sav_config); 692 spa->spa_l2cache.sav_config = NULL; 693 } 694 spa->spa_l2cache.sav_count = 0; 695 696 spa->spa_async_suspended = 0; 697} 698 699/* 700 * Load (or re-load) the current list of vdevs describing the active spares for 701 * this pool. When this is called, we have some form of basic information in 702 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 703 * then re-generate a more complete list including status information. 704 */ 705static void 706spa_load_spares(spa_t *spa) 707{ 708 nvlist_t **spares; 709 uint_t nspares; 710 int i; 711 vdev_t *vd, *tvd; 712 713 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 714 715 /* 716 * First, close and free any existing spare vdevs. 717 */ 718 for (i = 0; i < spa->spa_spares.sav_count; i++) { 719 vd = spa->spa_spares.sav_vdevs[i]; 720 721 /* Undo the call to spa_activate() below */ 722 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 723 B_FALSE)) != NULL && tvd->vdev_isspare) 724 spa_spare_remove(tvd); 725 vdev_close(vd); 726 vdev_free(vd); 727 } 728 729 if (spa->spa_spares.sav_vdevs) 730 kmem_free(spa->spa_spares.sav_vdevs, 731 spa->spa_spares.sav_count * sizeof (void *)); 732 733 if (spa->spa_spares.sav_config == NULL) 734 nspares = 0; 735 else 736 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 737 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 738 739 spa->spa_spares.sav_count = (int)nspares; 740 spa->spa_spares.sav_vdevs = NULL; 741 742 if (nspares == 0) 743 return; 744 745 /* 746 * Construct the array of vdevs, opening them to get status in the 747 * process. For each spare, there is potentially two different vdev_t 748 * structures associated with it: one in the list of spares (used only 749 * for basic validation purposes) and one in the active vdev 750 * configuration (if it's spared in). During this phase we open and 751 * validate each vdev on the spare list. If the vdev also exists in the 752 * active configuration, then we also mark this vdev as an active spare. 753 */ 754 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 755 KM_SLEEP); 756 for (i = 0; i < spa->spa_spares.sav_count; i++) { 757 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 758 VDEV_ALLOC_SPARE) == 0); 759 ASSERT(vd != NULL); 760 761 spa->spa_spares.sav_vdevs[i] = vd; 762 763 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 764 B_FALSE)) != NULL) { 765 if (!tvd->vdev_isspare) 766 spa_spare_add(tvd); 767 768 /* 769 * We only mark the spare active if we were successfully 770 * able to load the vdev. Otherwise, importing a pool 771 * with a bad active spare would result in strange 772 * behavior, because multiple pool would think the spare 773 * is actively in use. 774 * 775 * There is a vulnerability here to an equally bizarre 776 * circumstance, where a dead active spare is later 777 * brought back to life (onlined or otherwise). Given 778 * the rarity of this scenario, and the extra complexity 779 * it adds, we ignore the possibility. 780 */ 781 if (!vdev_is_dead(tvd)) 782 spa_spare_activate(tvd); 783 } 784 785 vd->vdev_top = vd; 786 787 if (vdev_open(vd) != 0) 788 continue; 789 790 if (vdev_validate_aux(vd) == 0) 791 spa_spare_add(vd); 792 } 793 794 /* 795 * Recompute the stashed list of spares, with status information 796 * this time. 797 */ 798 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 799 DATA_TYPE_NVLIST_ARRAY) == 0); 800 801 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 802 KM_SLEEP); 803 for (i = 0; i < spa->spa_spares.sav_count; i++) 804 spares[i] = vdev_config_generate(spa, 805 spa->spa_spares.sav_vdevs[i], B_TRUE, B_TRUE, B_FALSE); 806 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 807 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 808 for (i = 0; i < spa->spa_spares.sav_count; i++) 809 nvlist_free(spares[i]); 810 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 811} 812 813/* 814 * Load (or re-load) the current list of vdevs describing the active l2cache for 815 * this pool. When this is called, we have some form of basic information in 816 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 817 * then re-generate a more complete list including status information. 818 * Devices which are already active have their details maintained, and are 819 * not re-opened. 820 */ 821static void 822spa_load_l2cache(spa_t *spa) 823{ 824 nvlist_t **l2cache; 825 uint_t nl2cache; 826 int i, j, oldnvdevs; 827 uint64_t guid, size; 828 vdev_t *vd, **oldvdevs, **newvdevs; 829 spa_aux_vdev_t *sav = &spa->spa_l2cache; 830 831 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 832 833 if (sav->sav_config != NULL) { 834 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 835 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 836 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 837 } else { 838 nl2cache = 0; 839 } 840 841 oldvdevs = sav->sav_vdevs; 842 oldnvdevs = sav->sav_count; 843 sav->sav_vdevs = NULL; 844 sav->sav_count = 0; 845 846 /* 847 * Process new nvlist of vdevs. 848 */ 849 for (i = 0; i < nl2cache; i++) { 850 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 851 &guid) == 0); 852 853 newvdevs[i] = NULL; 854 for (j = 0; j < oldnvdevs; j++) { 855 vd = oldvdevs[j]; 856 if (vd != NULL && guid == vd->vdev_guid) { 857 /* 858 * Retain previous vdev for add/remove ops. 859 */ 860 newvdevs[i] = vd; 861 oldvdevs[j] = NULL; 862 break; 863 } 864 } 865 866 if (newvdevs[i] == NULL) { 867 /* 868 * Create new vdev 869 */ 870 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 871 VDEV_ALLOC_L2CACHE) == 0); 872 ASSERT(vd != NULL); 873 newvdevs[i] = vd; 874 875 /* 876 * Commit this vdev as an l2cache device, 877 * even if it fails to open. 878 */ 879 spa_l2cache_add(vd); 880 881 vd->vdev_top = vd; 882 vd->vdev_aux = sav; 883 884 spa_l2cache_activate(vd); 885 886 if (vdev_open(vd) != 0) 887 continue; 888 889 (void) vdev_validate_aux(vd); 890 891 if (!vdev_is_dead(vd)) { 892 size = vdev_get_rsize(vd); 893 l2arc_add_vdev(spa, vd, 894 VDEV_LABEL_START_SIZE, 895 size - VDEV_LABEL_START_SIZE); 896 } 897 } 898 } 899 900 /* 901 * Purge vdevs that were dropped 902 */ 903 for (i = 0; i < oldnvdevs; i++) { 904 uint64_t pool; 905 906 vd = oldvdevs[i]; 907 if (vd != NULL) { 908 if ((spa_mode & FWRITE) && 909 spa_l2cache_exists(vd->vdev_guid, &pool) && 910 pool != 0ULL && 911 l2arc_vdev_present(vd)) { 912 l2arc_remove_vdev(vd); 913 } 914 (void) vdev_close(vd); 915 spa_l2cache_remove(vd); 916 } 917 } 918 919 if (oldvdevs) 920 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 921 922 if (sav->sav_config == NULL) 923 goto out; 924 925 sav->sav_vdevs = newvdevs; 926 sav->sav_count = (int)nl2cache; 927 928 /* 929 * Recompute the stashed list of l2cache devices, with status 930 * information this time. 931 */ 932 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 933 DATA_TYPE_NVLIST_ARRAY) == 0); 934 935 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 936 for (i = 0; i < sav->sav_count; i++) 937 l2cache[i] = vdev_config_generate(spa, 938 sav->sav_vdevs[i], B_TRUE, B_FALSE, B_TRUE); 939 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 940 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 941out: 942 for (i = 0; i < sav->sav_count; i++) 943 nvlist_free(l2cache[i]); 944 if (sav->sav_count) 945 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 946} 947 948static int 949load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 950{ 951 dmu_buf_t *db; 952 char *packed = NULL; 953 size_t nvsize = 0; 954 int error; 955 *value = NULL; 956 957 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 958 nvsize = *(uint64_t *)db->db_data; 959 dmu_buf_rele(db, FTAG); 960 961 packed = kmem_alloc(nvsize, KM_SLEEP); 962 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed); 963 if (error == 0) 964 error = nvlist_unpack(packed, nvsize, value, 0); 965 kmem_free(packed, nvsize); 966 967 return (error); 968} 969 970/* 971 * Checks to see if the given vdev could not be opened, in which case we post a 972 * sysevent to notify the autoreplace code that the device has been removed. 973 */ 974static void 975spa_check_removed(vdev_t *vd) 976{ 977 int c; 978 979 for (c = 0; c < vd->vdev_children; c++) 980 spa_check_removed(vd->vdev_child[c]); 981 982 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) { 983 zfs_post_autoreplace(vd->vdev_spa, vd); 984 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 985 } 986} 987 988/* 989 * Check for missing log devices 990 */ 991int 992spa_check_logs(spa_t *spa) 993{ 994 switch (spa->spa_log_state) { 995 case SPA_LOG_MISSING: 996 /* need to recheck in case slog has been restored */ 997 case SPA_LOG_UNKNOWN: 998 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL, 999 DS_FIND_CHILDREN)) { 1000 spa->spa_log_state = SPA_LOG_MISSING; 1001 return (1); 1002 } 1003 break; 1004 1005 case SPA_LOG_CLEAR: 1006 (void) dmu_objset_find(spa->spa_name, zil_clear_log_chain, NULL, 1007 DS_FIND_CHILDREN); 1008 break; 1009 } 1010 spa->spa_log_state = SPA_LOG_GOOD; 1011 return (0); 1012} 1013 1014/* 1015 * Load an existing storage pool, using the pool's builtin spa_config as a 1016 * source of configuration information. 1017 */ 1018static int 1019spa_load(spa_t *spa, nvlist_t *config, spa_load_state_t state, int mosconfig) 1020{ 1021 int error = 0; 1022 nvlist_t *nvroot = NULL; 1023 vdev_t *rvd; 1024 uberblock_t *ub = &spa->spa_uberblock; 1025 uint64_t config_cache_txg = spa->spa_config_txg; 1026 uint64_t pool_guid; 1027 uint64_t version; 1028 uint64_t autoreplace = 0; 1029 char *ereport = FM_EREPORT_ZFS_POOL; 1030 1031 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1032 1033 spa->spa_load_state = state; 1034 1035 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot) || 1036 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { 1037 error = EINVAL; 1038 goto out; 1039 } 1040 1041 /* 1042 * Versioning wasn't explicitly added to the label until later, so if 1043 * it's not present treat it as the initial version. 1044 */ 1045 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &version) != 0) 1046 version = SPA_VERSION_INITIAL; 1047 1048 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 1049 &spa->spa_config_txg); 1050 1051 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 1052 spa_guid_exists(pool_guid, 0)) { 1053 error = EEXIST; 1054 goto out; 1055 } 1056 1057 spa->spa_load_guid = pool_guid; 1058 1059 /* 1060 * Parse the configuration into a vdev tree. We explicitly set the 1061 * value that will be returned by spa_version() since parsing the 1062 * configuration requires knowing the version number. 1063 */ 1064 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1065 spa->spa_ubsync.ub_version = version; 1066 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_LOAD); 1067 spa_config_exit(spa, SCL_ALL, FTAG); 1068 1069 if (error != 0) 1070 goto out; 1071 1072 ASSERT(spa->spa_root_vdev == rvd); 1073 ASSERT(spa_guid(spa) == pool_guid); 1074 1075 /* 1076 * Try to open all vdevs, loading each label in the process. 1077 */ 1078 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1079 error = vdev_open(rvd); 1080 spa_config_exit(spa, SCL_ALL, FTAG); 1081 if (error != 0) 1082 goto out; 1083 1084 /* 1085 * Validate the labels for all leaf vdevs. We need to grab the config 1086 * lock because all label I/O is done with ZIO_FLAG_CONFIG_WRITER. 1087 */ 1088 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1089 error = vdev_validate(rvd); 1090 spa_config_exit(spa, SCL_ALL, FTAG); 1091 1092 if (error != 0) 1093 goto out; 1094 1095 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 1096 error = ENXIO; 1097 goto out; 1098 } 1099 1100 /* 1101 * Find the best uberblock. 1102 */ 1103 vdev_uberblock_load(NULL, rvd, ub); 1104 1105 /* 1106 * If we weren't able to find a single valid uberblock, return failure. 1107 */ 1108 if (ub->ub_txg == 0) { 1109 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1110 VDEV_AUX_CORRUPT_DATA); 1111 error = ENXIO; 1112 goto out; 1113 } 1114 1115 /* 1116 * If the pool is newer than the code, we can't open it. 1117 */ 1118 if (ub->ub_version > SPA_VERSION) { 1119 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1120 VDEV_AUX_VERSION_NEWER); 1121 error = ENOTSUP; 1122 goto out; 1123 } 1124 1125 /* 1126 * If the vdev guid sum doesn't match the uberblock, we have an 1127 * incomplete configuration. 1128 */ 1129 if (rvd->vdev_guid_sum != ub->ub_guid_sum && mosconfig) { 1130 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1131 VDEV_AUX_BAD_GUID_SUM); 1132 error = ENXIO; 1133 goto out; 1134 } 1135 1136 /* 1137 * Initialize internal SPA structures. 1138 */ 1139 spa->spa_state = POOL_STATE_ACTIVE; 1140 spa->spa_ubsync = spa->spa_uberblock; 1141 spa->spa_first_txg = spa_last_synced_txg(spa) + 1; 1142 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 1143 if (error) { 1144 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1145 VDEV_AUX_CORRUPT_DATA); 1146 goto out; 1147 } 1148 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 1149 1150 if (zap_lookup(spa->spa_meta_objset, 1151 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 1152 sizeof (uint64_t), 1, &spa->spa_config_object) != 0) { 1153 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1154 VDEV_AUX_CORRUPT_DATA); 1155 error = EIO; 1156 goto out; 1157 } 1158 1159 if (!mosconfig) { 1160 nvlist_t *newconfig; 1161 uint64_t hostid; 1162 1163 if (load_nvlist(spa, spa->spa_config_object, &newconfig) != 0) { 1164 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1165 VDEV_AUX_CORRUPT_DATA); 1166 error = EIO; 1167 goto out; 1168 } 1169 1170 if (!spa_is_root(spa) && nvlist_lookup_uint64(newconfig, 1171 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 1172 char *hostname; 1173 unsigned long myhostid = 0; 1174 1175 VERIFY(nvlist_lookup_string(newconfig, 1176 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 1177 1178 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 1179 if (check_hostid && hostid != 0 && myhostid != 0 && 1180 (unsigned long)hostid != myhostid) { 1181 cmn_err(CE_WARN, "pool '%s' could not be " 1182 "loaded as it was last accessed by " 1183 "another system (host: %s hostid: 0x%lx). " 1184 "See: http://www.sun.com/msg/ZFS-8000-EY", 1185 spa_name(spa), hostname, 1186 (unsigned long)hostid); 1187 error = EBADF; 1188 goto out; 1189 } 1190 } 1191 1192 spa_config_set(spa, newconfig); 1193 spa_unload(spa); 1194 spa_deactivate(spa); 1195 spa_activate(spa); 1196 1197 return (spa_load(spa, newconfig, state, B_TRUE)); 1198 } 1199 1200 if (zap_lookup(spa->spa_meta_objset, 1201 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, 1202 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj) != 0) { 1203 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1204 VDEV_AUX_CORRUPT_DATA); 1205 error = EIO; 1206 goto out; 1207 } 1208 1209 /* 1210 * Load the bit that tells us to use the new accounting function 1211 * (raid-z deflation). If we have an older pool, this will not 1212 * be present. 1213 */ 1214 error = zap_lookup(spa->spa_meta_objset, 1215 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 1216 sizeof (uint64_t), 1, &spa->spa_deflate); 1217 if (error != 0 && error != ENOENT) { 1218 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1219 VDEV_AUX_CORRUPT_DATA); 1220 error = EIO; 1221 goto out; 1222 } 1223 1224 /* 1225 * Load the persistent error log. If we have an older pool, this will 1226 * not be present. 1227 */ 1228 error = zap_lookup(spa->spa_meta_objset, 1229 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_LAST, 1230 sizeof (uint64_t), 1, &spa->spa_errlog_last); 1231 if (error != 0 && error != ENOENT) { 1232 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1233 VDEV_AUX_CORRUPT_DATA); 1234 error = EIO; 1235 goto out; 1236 } 1237 1238 error = zap_lookup(spa->spa_meta_objset, 1239 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ERRLOG_SCRUB, 1240 sizeof (uint64_t), 1, &spa->spa_errlog_scrub); 1241 if (error != 0 && error != ENOENT) { 1242 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1243 VDEV_AUX_CORRUPT_DATA); 1244 error = EIO; 1245 goto out; 1246 } 1247 1248 /* 1249 * Load the history object. If we have an older pool, this 1250 * will not be present. 1251 */ 1252 error = zap_lookup(spa->spa_meta_objset, 1253 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_HISTORY, 1254 sizeof (uint64_t), 1, &spa->spa_history); 1255 if (error != 0 && error != ENOENT) { 1256 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1257 VDEV_AUX_CORRUPT_DATA); 1258 error = EIO; 1259 goto out; 1260 } 1261 1262 /* 1263 * Load any hot spares for this pool. 1264 */ 1265 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1266 DMU_POOL_SPARES, sizeof (uint64_t), 1, &spa->spa_spares.sav_object); 1267 if (error != 0 && error != ENOENT) { 1268 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1269 VDEV_AUX_CORRUPT_DATA); 1270 error = EIO; 1271 goto out; 1272 } 1273 if (error == 0) { 1274 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 1275 if (load_nvlist(spa, spa->spa_spares.sav_object, 1276 &spa->spa_spares.sav_config) != 0) { 1277 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1278 VDEV_AUX_CORRUPT_DATA); 1279 error = EIO; 1280 goto out; 1281 } 1282 1283 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1284 spa_load_spares(spa); 1285 spa_config_exit(spa, SCL_ALL, FTAG); 1286 } 1287 1288 /* 1289 * Load any level 2 ARC devices for this pool. 1290 */ 1291 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1292 DMU_POOL_L2CACHE, sizeof (uint64_t), 1, 1293 &spa->spa_l2cache.sav_object); 1294 if (error != 0 && error != ENOENT) { 1295 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1296 VDEV_AUX_CORRUPT_DATA); 1297 error = EIO; 1298 goto out; 1299 } 1300 if (error == 0) { 1301 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 1302 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 1303 &spa->spa_l2cache.sav_config) != 0) { 1304 vdev_set_state(rvd, B_TRUE, 1305 VDEV_STATE_CANT_OPEN, 1306 VDEV_AUX_CORRUPT_DATA); 1307 error = EIO; 1308 goto out; 1309 } 1310 1311 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1312 spa_load_l2cache(spa); 1313 spa_config_exit(spa, SCL_ALL, FTAG); 1314 } 1315 1316 if (spa_check_logs(spa)) { 1317 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1318 VDEV_AUX_BAD_LOG); 1319 error = ENXIO; 1320 ereport = FM_EREPORT_ZFS_LOG_REPLAY; 1321 goto out; 1322 } 1323 1324 1325 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 1326 1327 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1328 DMU_POOL_PROPS, sizeof (uint64_t), 1, &spa->spa_pool_props_object); 1329 1330 if (error && error != ENOENT) { 1331 vdev_set_state(rvd, B_TRUE, VDEV_STATE_CANT_OPEN, 1332 VDEV_AUX_CORRUPT_DATA); 1333 error = EIO; 1334 goto out; 1335 } 1336 1337 if (error == 0) { 1338 (void) zap_lookup(spa->spa_meta_objset, 1339 spa->spa_pool_props_object, 1340 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), 1341 sizeof (uint64_t), 1, &spa->spa_bootfs); 1342 (void) zap_lookup(spa->spa_meta_objset, 1343 spa->spa_pool_props_object, 1344 zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1345 sizeof (uint64_t), 1, &autoreplace); 1346 (void) zap_lookup(spa->spa_meta_objset, 1347 spa->spa_pool_props_object, 1348 zpool_prop_to_name(ZPOOL_PROP_DELEGATION), 1349 sizeof (uint64_t), 1, &spa->spa_delegation); 1350 (void) zap_lookup(spa->spa_meta_objset, 1351 spa->spa_pool_props_object, 1352 zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE), 1353 sizeof (uint64_t), 1, &spa->spa_failmode); 1354 } 1355 1356 /* 1357 * If the 'autoreplace' property is set, then post a resource notifying 1358 * the ZFS DE that it should not issue any faults for unopenable 1359 * devices. We also iterate over the vdevs, and post a sysevent for any 1360 * unopenable vdevs so that the normal autoreplace handler can take 1361 * over. 1362 */ 1363 if (autoreplace && state != SPA_LOAD_TRYIMPORT) 1364 spa_check_removed(spa->spa_root_vdev); 1365 1366 /* 1367 * Load the vdev state for all toplevel vdevs. 1368 */ 1369 vdev_load(rvd); 1370 1371 /* 1372 * Propagate the leaf DTLs we just loaded all the way up the tree. 1373 */ 1374 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1375 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 1376 spa_config_exit(spa, SCL_ALL, FTAG); 1377 1378 /* 1379 * Check the state of the root vdev. If it can't be opened, it 1380 * indicates one or more toplevel vdevs are faulted. 1381 */ 1382 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 1383 error = ENXIO; 1384 goto out; 1385 } 1386 1387 if ((spa_mode & FWRITE) && state != SPA_LOAD_TRYIMPORT) { 1388 dmu_tx_t *tx; 1389 int need_update = B_FALSE; 1390 int c; 1391 1392 /* 1393 * Claim log blocks that haven't been committed yet. 1394 * This must all happen in a single txg. 1395 */ 1396 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 1397 spa_first_txg(spa)); 1398 (void) dmu_objset_find(spa_name(spa), 1399 zil_claim, tx, DS_FIND_CHILDREN); 1400 dmu_tx_commit(tx); 1401 1402 spa->spa_sync_on = B_TRUE; 1403 txg_sync_start(spa->spa_dsl_pool); 1404 1405 /* 1406 * Wait for all claims to sync. 1407 */ 1408 txg_wait_synced(spa->spa_dsl_pool, 0); 1409 1410 /* 1411 * If the config cache is stale, or we have uninitialized 1412 * metaslabs (see spa_vdev_add()), then update the config. 1413 */ 1414 if (config_cache_txg != spa->spa_config_txg || 1415 state == SPA_LOAD_IMPORT) 1416 need_update = B_TRUE; 1417 1418 for (c = 0; c < rvd->vdev_children; c++) 1419 if (rvd->vdev_child[c]->vdev_ms_array == 0) 1420 need_update = B_TRUE; 1421 1422 /* 1423 * Update the config cache asychronously in case we're the 1424 * root pool, in which case the config cache isn't writable yet. 1425 */ 1426 if (need_update) 1427 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 1428 } 1429 1430 error = 0; 1431out: 1432 spa->spa_minref = refcount_count(&spa->spa_refcount); 1433 if (error && error != EBADF) 1434 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 1435 spa->spa_load_state = SPA_LOAD_NONE; 1436 spa->spa_ena = 0; 1437 1438 return (error); 1439} 1440 1441/* 1442 * Pool Open/Import 1443 * 1444 * The import case is identical to an open except that the configuration is sent 1445 * down from userland, instead of grabbed from the configuration cache. For the 1446 * case of an open, the pool configuration will exist in the 1447 * POOL_STATE_UNINITIALIZED state. 1448 * 1449 * The stats information (gen/count/ustats) is used to gather vdev statistics at 1450 * the same time open the pool, without having to keep around the spa_t in some 1451 * ambiguous state. 1452 */ 1453static int 1454spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t **config) 1455{ 1456 spa_t *spa; 1457 int error; 1458 int locked = B_FALSE; 1459 1460 *spapp = NULL; 1461 1462 /* 1463 * As disgusting as this is, we need to support recursive calls to this 1464 * function because dsl_dir_open() is called during spa_load(), and ends 1465 * up calling spa_open() again. The real fix is to figure out how to 1466 * avoid dsl_dir_open() calling this in the first place. 1467 */ 1468 if (mutex_owner(&spa_namespace_lock) != curthread) { 1469 mutex_enter(&spa_namespace_lock); 1470 locked = B_TRUE; 1471 } 1472 1473 if ((spa = spa_lookup(pool)) == NULL) { 1474 if (locked) 1475 mutex_exit(&spa_namespace_lock); 1476 return (ENOENT); 1477 } 1478 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 1479 1480 spa_activate(spa); 1481 1482 error = spa_load(spa, spa->spa_config, SPA_LOAD_OPEN, B_FALSE); 1483 1484 if (error == EBADF) { 1485 /* 1486 * If vdev_validate() returns failure (indicated by 1487 * EBADF), it indicates that one of the vdevs indicates 1488 * that the pool has been exported or destroyed. If 1489 * this is the case, the config cache is out of sync and 1490 * we should remove the pool from the namespace. 1491 */ 1492 spa_unload(spa); 1493 spa_deactivate(spa); 1494 spa_config_sync(spa, B_TRUE, B_TRUE); 1495 spa_remove(spa); 1496 if (locked) 1497 mutex_exit(&spa_namespace_lock); 1498 return (ENOENT); 1499 } 1500 1501 if (error) { 1502 /* 1503 * We can't open the pool, but we still have useful 1504 * information: the state of each vdev after the 1505 * attempted vdev_open(). Return this to the user. 1506 */ 1507 if (config != NULL && spa->spa_root_vdev != NULL) 1508 *config = spa_config_generate(spa, NULL, -1ULL, 1509 B_TRUE); 1510 spa_unload(spa); 1511 spa_deactivate(spa); 1512 spa->spa_last_open_failed = B_TRUE; 1513 if (locked) 1514 mutex_exit(&spa_namespace_lock); 1515 *spapp = NULL; 1516 return (error); 1517 } else { 1518 spa->spa_last_open_failed = B_FALSE; 1519 } 1520 } 1521 1522 spa_open_ref(spa, tag); 1523 1524 if (locked) 1525 mutex_exit(&spa_namespace_lock); 1526 1527 *spapp = spa; 1528 1529 if (config != NULL) 1530 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 1531 1532 return (0); 1533} 1534 1535int 1536spa_open(const char *name, spa_t **spapp, void *tag) 1537{ 1538 return (spa_open_common(name, spapp, tag, NULL)); 1539} 1540 1541/* 1542 * Lookup the given spa_t, incrementing the inject count in the process, 1543 * preventing it from being exported or destroyed. 1544 */ 1545spa_t * 1546spa_inject_addref(char *name) 1547{ 1548 spa_t *spa; 1549 1550 mutex_enter(&spa_namespace_lock); 1551 if ((spa = spa_lookup(name)) == NULL) { 1552 mutex_exit(&spa_namespace_lock); 1553 return (NULL); 1554 } 1555 spa->spa_inject_ref++; 1556 mutex_exit(&spa_namespace_lock); 1557 1558 return (spa); 1559} 1560 1561void 1562spa_inject_delref(spa_t *spa) 1563{ 1564 mutex_enter(&spa_namespace_lock); 1565 spa->spa_inject_ref--; 1566 mutex_exit(&spa_namespace_lock); 1567} 1568 1569/* 1570 * Add spares device information to the nvlist. 1571 */ 1572static void 1573spa_add_spares(spa_t *spa, nvlist_t *config) 1574{ 1575 nvlist_t **spares; 1576 uint_t i, nspares; 1577 nvlist_t *nvroot; 1578 uint64_t guid; 1579 vdev_stat_t *vs; 1580 uint_t vsc; 1581 uint64_t pool; 1582 1583 if (spa->spa_spares.sav_count == 0) 1584 return; 1585 1586 VERIFY(nvlist_lookup_nvlist(config, 1587 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 1588 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1589 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1590 if (nspares != 0) { 1591 VERIFY(nvlist_add_nvlist_array(nvroot, 1592 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 1593 VERIFY(nvlist_lookup_nvlist_array(nvroot, 1594 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1595 1596 /* 1597 * Go through and find any spares which have since been 1598 * repurposed as an active spare. If this is the case, update 1599 * their status appropriately. 1600 */ 1601 for (i = 0; i < nspares; i++) { 1602 VERIFY(nvlist_lookup_uint64(spares[i], 1603 ZPOOL_CONFIG_GUID, &guid) == 0); 1604 if (spa_spare_exists(guid, &pool, NULL) && 1605 pool != 0ULL) { 1606 VERIFY(nvlist_lookup_uint64_array( 1607 spares[i], ZPOOL_CONFIG_STATS, 1608 (uint64_t **)&vs, &vsc) == 0); 1609 vs->vs_state = VDEV_STATE_CANT_OPEN; 1610 vs->vs_aux = VDEV_AUX_SPARED; 1611 } 1612 } 1613 } 1614} 1615 1616/* 1617 * Add l2cache device information to the nvlist, including vdev stats. 1618 */ 1619static void 1620spa_add_l2cache(spa_t *spa, nvlist_t *config) 1621{ 1622 nvlist_t **l2cache; 1623 uint_t i, j, nl2cache; 1624 nvlist_t *nvroot; 1625 uint64_t guid; 1626 vdev_t *vd; 1627 vdev_stat_t *vs; 1628 uint_t vsc; 1629 1630 if (spa->spa_l2cache.sav_count == 0) 1631 return; 1632 1633 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 1634 1635 VERIFY(nvlist_lookup_nvlist(config, 1636 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 1637 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 1638 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1639 if (nl2cache != 0) { 1640 VERIFY(nvlist_add_nvlist_array(nvroot, 1641 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 1642 VERIFY(nvlist_lookup_nvlist_array(nvroot, 1643 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1644 1645 /* 1646 * Update level 2 cache device stats. 1647 */ 1648 1649 for (i = 0; i < nl2cache; i++) { 1650 VERIFY(nvlist_lookup_uint64(l2cache[i], 1651 ZPOOL_CONFIG_GUID, &guid) == 0); 1652 1653 vd = NULL; 1654 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 1655 if (guid == 1656 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 1657 vd = spa->spa_l2cache.sav_vdevs[j]; 1658 break; 1659 } 1660 } 1661 ASSERT(vd != NULL); 1662 1663 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 1664 ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0); 1665 vdev_get_stats(vd, vs); 1666 } 1667 } 1668 1669 spa_config_exit(spa, SCL_CONFIG, FTAG); 1670} 1671 1672int 1673spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen) 1674{ 1675 int error; 1676 spa_t *spa; 1677 1678 *config = NULL; 1679 error = spa_open_common(name, &spa, FTAG, config); 1680 1681 if (spa && *config != NULL) { 1682 VERIFY(nvlist_add_uint64(*config, ZPOOL_CONFIG_ERRCOUNT, 1683 spa_get_errlog_size(spa)) == 0); 1684 1685 if (spa_suspended(spa)) 1686 VERIFY(nvlist_add_uint64(*config, 1687 ZPOOL_CONFIG_SUSPENDED, spa->spa_failmode) == 0); 1688 1689 spa_add_spares(spa, *config); 1690 spa_add_l2cache(spa, *config); 1691 } 1692 1693 /* 1694 * We want to get the alternate root even for faulted pools, so we cheat 1695 * and call spa_lookup() directly. 1696 */ 1697 if (altroot) { 1698 if (spa == NULL) { 1699 mutex_enter(&spa_namespace_lock); 1700 spa = spa_lookup(name); 1701 if (spa) 1702 spa_altroot(spa, altroot, buflen); 1703 else 1704 altroot[0] = '\0'; 1705 spa = NULL; 1706 mutex_exit(&spa_namespace_lock); 1707 } else { 1708 spa_altroot(spa, altroot, buflen); 1709 } 1710 } 1711 1712 if (spa != NULL) 1713 spa_close(spa, FTAG); 1714 1715 return (error); 1716} 1717 1718/* 1719 * Validate that the auxiliary device array is well formed. We must have an 1720 * array of nvlists, each which describes a valid leaf vdev. If this is an 1721 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 1722 * specified, as long as they are well-formed. 1723 */ 1724static int 1725spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 1726 spa_aux_vdev_t *sav, const char *config, uint64_t version, 1727 vdev_labeltype_t label) 1728{ 1729 nvlist_t **dev; 1730 uint_t i, ndev; 1731 vdev_t *vd; 1732 int error; 1733 1734 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1735 1736 /* 1737 * It's acceptable to have no devs specified. 1738 */ 1739 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 1740 return (0); 1741 1742 if (ndev == 0) 1743 return (EINVAL); 1744 1745 /* 1746 * Make sure the pool is formatted with a version that supports this 1747 * device type. 1748 */ 1749 if (spa_version(spa) < version) 1750 return (ENOTSUP); 1751 1752 /* 1753 * Set the pending device list so we correctly handle device in-use 1754 * checking. 1755 */ 1756 sav->sav_pending = dev; 1757 sav->sav_npending = ndev; 1758 1759 for (i = 0; i < ndev; i++) { 1760 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 1761 mode)) != 0) 1762 goto out; 1763 1764 if (!vd->vdev_ops->vdev_op_leaf) { 1765 vdev_free(vd); 1766 error = EINVAL; 1767 goto out; 1768 } 1769 1770 /* 1771 * The L2ARC currently only supports disk devices in 1772 * kernel context. For user-level testing, we allow it. 1773 */ 1774#ifdef _KERNEL 1775 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 1776 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 1777 error = ENOTBLK; 1778 goto out; 1779 } 1780#endif 1781 vd->vdev_top = vd; 1782 1783 if ((error = vdev_open(vd)) == 0 && 1784 (error = vdev_label_init(vd, crtxg, label)) == 0) { 1785 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 1786 vd->vdev_guid) == 0); 1787 } 1788 1789 vdev_free(vd); 1790 1791 if (error && 1792 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 1793 goto out; 1794 else 1795 error = 0; 1796 } 1797 1798out: 1799 sav->sav_pending = NULL; 1800 sav->sav_npending = 0; 1801 return (error); 1802} 1803 1804static int 1805spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 1806{ 1807 int error; 1808 1809 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1810 1811 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 1812 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 1813 VDEV_LABEL_SPARE)) != 0) { 1814 return (error); 1815 } 1816 1817 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 1818 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 1819 VDEV_LABEL_L2CACHE)); 1820} 1821 1822static void 1823spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 1824 const char *config) 1825{ 1826 int i; 1827 1828 if (sav->sav_config != NULL) { 1829 nvlist_t **olddevs; 1830 uint_t oldndevs; 1831 nvlist_t **newdevs; 1832 1833 /* 1834 * Generate new dev list by concatentating with the 1835 * current dev list. 1836 */ 1837 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 1838 &olddevs, &oldndevs) == 0); 1839 1840 newdevs = kmem_alloc(sizeof (void *) * 1841 (ndevs + oldndevs), KM_SLEEP); 1842 for (i = 0; i < oldndevs; i++) 1843 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 1844 KM_SLEEP) == 0); 1845 for (i = 0; i < ndevs; i++) 1846 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 1847 KM_SLEEP) == 0); 1848 1849 VERIFY(nvlist_remove(sav->sav_config, config, 1850 DATA_TYPE_NVLIST_ARRAY) == 0); 1851 1852 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1853 config, newdevs, ndevs + oldndevs) == 0); 1854 for (i = 0; i < oldndevs + ndevs; i++) 1855 nvlist_free(newdevs[i]); 1856 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 1857 } else { 1858 /* 1859 * Generate a new dev list. 1860 */ 1861 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 1862 KM_SLEEP) == 0); 1863 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 1864 devs, ndevs) == 0); 1865 } 1866} 1867 1868/* 1869 * Stop and drop level 2 ARC devices 1870 */ 1871void 1872spa_l2cache_drop(spa_t *spa) 1873{ 1874 vdev_t *vd; 1875 int i; 1876 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1877 1878 for (i = 0; i < sav->sav_count; i++) { 1879 uint64_t pool; 1880 1881 vd = sav->sav_vdevs[i]; 1882 ASSERT(vd != NULL); 1883 1884 if ((spa_mode & FWRITE) && 1885 spa_l2cache_exists(vd->vdev_guid, &pool) && pool != 0ULL && 1886 l2arc_vdev_present(vd)) { 1887 l2arc_remove_vdev(vd); 1888 } 1889 if (vd->vdev_isl2cache) 1890 spa_l2cache_remove(vd); 1891 vdev_clear_stats(vd); 1892 (void) vdev_close(vd); 1893 } 1894} 1895 1896/* 1897 * Pool Creation 1898 */ 1899int 1900spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 1901 const char *history_str, nvlist_t *zplprops) 1902{ 1903 spa_t *spa; 1904 char *altroot = NULL; 1905 vdev_t *rvd; 1906 dsl_pool_t *dp; 1907 dmu_tx_t *tx; 1908 int c, error = 0; 1909 uint64_t txg = TXG_INITIAL; 1910 nvlist_t **spares, **l2cache; 1911 uint_t nspares, nl2cache; 1912 uint64_t version; 1913 1914 /* 1915 * If this pool already exists, return failure. 1916 */ 1917 mutex_enter(&spa_namespace_lock); 1918 if (spa_lookup(pool) != NULL) { 1919 mutex_exit(&spa_namespace_lock); 1920 return (EEXIST); 1921 } 1922 1923 /* 1924 * Allocate a new spa_t structure. 1925 */ 1926 (void) nvlist_lookup_string(props, 1927 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 1928 spa = spa_add(pool, altroot); 1929 spa_activate(spa); 1930 1931 spa->spa_uberblock.ub_txg = txg - 1; 1932 1933 if (props && (error = spa_prop_validate(spa, props))) { 1934 spa_unload(spa); 1935 spa_deactivate(spa); 1936 spa_remove(spa); 1937 mutex_exit(&spa_namespace_lock); 1938 return (error); 1939 } 1940 1941 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION), 1942 &version) != 0) 1943 version = SPA_VERSION; 1944 ASSERT(version <= SPA_VERSION); 1945 spa->spa_uberblock.ub_version = version; 1946 spa->spa_ubsync = spa->spa_uberblock; 1947 1948 /* 1949 * Create the root vdev. 1950 */ 1951 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1952 1953 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 1954 1955 ASSERT(error != 0 || rvd != NULL); 1956 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 1957 1958 if (error == 0 && !zfs_allocatable_devs(nvroot)) 1959 error = EINVAL; 1960 1961 if (error == 0 && 1962 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 1963 (error = spa_validate_aux(spa, nvroot, txg, 1964 VDEV_ALLOC_ADD)) == 0) { 1965 for (c = 0; c < rvd->vdev_children; c++) 1966 vdev_init(rvd->vdev_child[c], txg); 1967 vdev_config_dirty(rvd); 1968 } 1969 1970 spa_config_exit(spa, SCL_ALL, FTAG); 1971 1972 if (error != 0) { 1973 spa_unload(spa); 1974 spa_deactivate(spa); 1975 spa_remove(spa); 1976 mutex_exit(&spa_namespace_lock); 1977 return (error); 1978 } 1979 1980 /* 1981 * Get the list of spares, if specified. 1982 */ 1983 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 1984 &spares, &nspares) == 0) { 1985 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 1986 KM_SLEEP) == 0); 1987 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1988 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 1989 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1990 spa_load_spares(spa); 1991 spa_config_exit(spa, SCL_ALL, FTAG); 1992 spa->spa_spares.sav_sync = B_TRUE; 1993 } 1994 1995 /* 1996 * Get the list of level 2 cache devices, if specified. 1997 */ 1998 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 1999 &l2cache, &nl2cache) == 0) { 2000 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 2001 NV_UNIQUE_NAME, KM_SLEEP) == 0); 2002 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 2003 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 2004 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2005 spa_load_l2cache(spa); 2006 spa_config_exit(spa, SCL_ALL, FTAG); 2007 spa->spa_l2cache.sav_sync = B_TRUE; 2008 } 2009 2010 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 2011 spa->spa_meta_objset = dp->dp_meta_objset; 2012 2013 tx = dmu_tx_create_assigned(dp, txg); 2014 2015 /* 2016 * Create the pool config object. 2017 */ 2018 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 2019 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 2020 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 2021 2022 if (zap_add(spa->spa_meta_objset, 2023 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 2024 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 2025 cmn_err(CE_PANIC, "failed to add pool config"); 2026 } 2027 2028 /* Newly created pools with the right version are always deflated. */ 2029 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 2030 spa->spa_deflate = TRUE; 2031 if (zap_add(spa->spa_meta_objset, 2032 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 2033 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 2034 cmn_err(CE_PANIC, "failed to add deflate"); 2035 } 2036 } 2037 2038 /* 2039 * Create the deferred-free bplist object. Turn off compression 2040 * because sync-to-convergence takes longer if the blocksize 2041 * keeps changing. 2042 */ 2043 spa->spa_sync_bplist_obj = bplist_create(spa->spa_meta_objset, 2044 1 << 14, tx); 2045 dmu_object_set_compress(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 2046 ZIO_COMPRESS_OFF, tx); 2047 2048 if (zap_add(spa->spa_meta_objset, 2049 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPLIST, 2050 sizeof (uint64_t), 1, &spa->spa_sync_bplist_obj, tx) != 0) { 2051 cmn_err(CE_PANIC, "failed to add bplist"); 2052 } 2053 2054 /* 2055 * Create the pool's history object. 2056 */ 2057 if (version >= SPA_VERSION_ZPOOL_HISTORY) 2058 spa_history_create_obj(spa, tx); 2059 2060 /* 2061 * Set pool properties. 2062 */ 2063 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 2064 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2065 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 2066 if (props) 2067 spa_sync_props(spa, props, CRED(), tx); 2068 2069 dmu_tx_commit(tx); 2070 2071 spa->spa_sync_on = B_TRUE; 2072 txg_sync_start(spa->spa_dsl_pool); 2073 2074 /* 2075 * We explicitly wait for the first transaction to complete so that our 2076 * bean counters are appropriately updated. 2077 */ 2078 txg_wait_synced(spa->spa_dsl_pool, txg); 2079 2080 spa_config_sync(spa, B_FALSE, B_TRUE); 2081 2082 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL) 2083 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE); 2084 2085 mutex_exit(&spa_namespace_lock); 2086 2087 spa->spa_minref = refcount_count(&spa->spa_refcount); 2088 2089 return (0); 2090} 2091 2092/* 2093 * Import the given pool into the system. We set up the necessary spa_t and 2094 * then call spa_load() to do the dirty work. 2095 */ 2096static int 2097spa_import_common(const char *pool, nvlist_t *config, nvlist_t *props, 2098 boolean_t isroot, boolean_t allowfaulted) 2099{ 2100 spa_t *spa; 2101 char *altroot = NULL; 2102 int error, loaderr; 2103 nvlist_t *nvroot; 2104 nvlist_t **spares, **l2cache; 2105 uint_t nspares, nl2cache; 2106 2107 /* 2108 * If a pool with this name exists, return failure. 2109 */ 2110 mutex_enter(&spa_namespace_lock); 2111 if (spa_lookup(pool) != NULL) { 2112 mutex_exit(&spa_namespace_lock); 2113 return (EEXIST); 2114 } 2115 2116 /* 2117 * Create and initialize the spa structure. 2118 */ 2119 (void) nvlist_lookup_string(props, 2120 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 2121 spa = spa_add(pool, altroot); 2122 spa_activate(spa); 2123 2124 if (allowfaulted) 2125 spa->spa_import_faulted = B_TRUE; 2126 spa->spa_is_root = isroot; 2127 2128 /* 2129 * Pass off the heavy lifting to spa_load(). 2130 * Pass TRUE for mosconfig (unless this is a root pool) because 2131 * the user-supplied config is actually the one to trust when 2132 * doing an import. 2133 */ 2134 loaderr = error = spa_load(spa, config, SPA_LOAD_IMPORT, !isroot); 2135 2136 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2137 /* 2138 * Toss any existing sparelist, as it doesn't have any validity anymore, 2139 * and conflicts with spa_has_spare(). 2140 */ 2141 if (!isroot && spa->spa_spares.sav_config) { 2142 nvlist_free(spa->spa_spares.sav_config); 2143 spa->spa_spares.sav_config = NULL; 2144 spa_load_spares(spa); 2145 } 2146 if (!isroot && spa->spa_l2cache.sav_config) { 2147 nvlist_free(spa->spa_l2cache.sav_config); 2148 spa->spa_l2cache.sav_config = NULL; 2149 spa_load_l2cache(spa); 2150 } 2151 2152 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 2153 &nvroot) == 0); 2154 if (error == 0) 2155 error = spa_validate_aux(spa, nvroot, -1ULL, VDEV_ALLOC_SPARE); 2156 if (error == 0) 2157 error = spa_validate_aux(spa, nvroot, -1ULL, 2158 VDEV_ALLOC_L2CACHE); 2159 spa_config_exit(spa, SCL_ALL, FTAG); 2160 2161 if (error != 0 || (props && (error = spa_prop_set(spa, props)))) { 2162 if (loaderr != 0 && loaderr != EINVAL && allowfaulted) { 2163 /* 2164 * If we failed to load the pool, but 'allowfaulted' is 2165 * set, then manually set the config as if the config 2166 * passed in was specified in the cache file. 2167 */ 2168 error = 0; 2169 spa->spa_import_faulted = B_FALSE; 2170 if (spa->spa_config == NULL) 2171 spa->spa_config = spa_config_generate(spa, 2172 NULL, -1ULL, B_TRUE); 2173 spa_unload(spa); 2174 spa_deactivate(spa); 2175 spa_config_sync(spa, B_FALSE, B_TRUE); 2176 } else { 2177 spa_unload(spa); 2178 spa_deactivate(spa); 2179 spa_remove(spa); 2180 } 2181 mutex_exit(&spa_namespace_lock); 2182 return (error); 2183 } 2184 2185 /* 2186 * Override any spares and level 2 cache devices as specified by 2187 * the user, as these may have correct device names/devids, etc. 2188 */ 2189 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 2190 &spares, &nspares) == 0) { 2191 if (spa->spa_spares.sav_config) 2192 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 2193 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 2194 else 2195 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 2196 NV_UNIQUE_NAME, KM_SLEEP) == 0); 2197 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 2198 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 2199 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2200 spa_load_spares(spa); 2201 spa_config_exit(spa, SCL_ALL, FTAG); 2202 spa->spa_spares.sav_sync = B_TRUE; 2203 } 2204 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 2205 &l2cache, &nl2cache) == 0) { 2206 if (spa->spa_l2cache.sav_config) 2207 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 2208 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 2209 else 2210 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 2211 NV_UNIQUE_NAME, KM_SLEEP) == 0); 2212 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 2213 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 2214 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2215 spa_load_l2cache(spa); 2216 spa_config_exit(spa, SCL_ALL, FTAG); 2217 spa->spa_l2cache.sav_sync = B_TRUE; 2218 } 2219 2220 if (spa_mode & FWRITE) { 2221 /* 2222 * Update the config cache to include the newly-imported pool. 2223 */ 2224 spa_config_update_common(spa, SPA_CONFIG_UPDATE_POOL, isroot); 2225 } 2226 2227 spa->spa_import_faulted = B_FALSE; 2228 mutex_exit(&spa_namespace_lock); 2229 2230 return (0); 2231} 2232 2233#if defined(sun) 2234#ifdef _KERNEL 2235/* 2236 * Build a "root" vdev for a top level vdev read in from a rootpool 2237 * device label. 2238 */ 2239static void 2240spa_build_rootpool_config(nvlist_t *config) 2241{ 2242 nvlist_t *nvtop, *nvroot; 2243 uint64_t pgid; 2244 2245 /* 2246 * Add this top-level vdev to the child array. 2247 */ 2248 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtop) 2249 == 0); 2250 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pgid) 2251 == 0); 2252 2253 /* 2254 * Put this pool's top-level vdevs into a root vdev. 2255 */ 2256 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 2257 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) 2258 == 0); 2259 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 2260 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 2261 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 2262 &nvtop, 1) == 0); 2263 2264 /* 2265 * Replace the existing vdev_tree with the new root vdev in 2266 * this pool's configuration (remove the old, add the new). 2267 */ 2268 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 2269 nvlist_free(nvroot); 2270} 2271 2272/* 2273 * Get the root pool information from the root disk, then import the root pool 2274 * during the system boot up time. 2275 */ 2276extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 2277 2278int 2279spa_check_rootconf(char *devpath, char *devid, nvlist_t **bestconf, 2280 uint64_t *besttxg) 2281{ 2282 nvlist_t *config; 2283 uint64_t txg; 2284 int error; 2285 2286 if (error = vdev_disk_read_rootlabel(devpath, devid, &config)) 2287 return (error); 2288 2289 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 2290 2291 if (bestconf != NULL) 2292 *bestconf = config; 2293 else 2294 nvlist_free(config); 2295 *besttxg = txg; 2296 return (0); 2297} 2298 2299boolean_t 2300spa_rootdev_validate(nvlist_t *nv) 2301{ 2302 uint64_t ival; 2303 2304 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 || 2305 nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 || 2306 nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0) 2307 return (B_FALSE); 2308 2309 return (B_TRUE); 2310} 2311 2312 2313/* 2314 * Given the boot device's physical path or devid, check if the device 2315 * is in a valid state. If so, return the configuration from the vdev 2316 * label. 2317 */ 2318int 2319spa_get_rootconf(char *devpath, char *devid, nvlist_t **bestconf) 2320{ 2321 nvlist_t *conf = NULL; 2322 uint64_t txg = 0; 2323 nvlist_t *nvtop, **child; 2324 char *type; 2325 char *bootpath = NULL; 2326 uint_t children, c; 2327 char *tmp; 2328 int error; 2329 2330 if (devpath && ((tmp = strchr(devpath, ' ')) != NULL)) 2331 *tmp = '\0'; 2332 if (error = spa_check_rootconf(devpath, devid, &conf, &txg)) { 2333 cmn_err(CE_NOTE, "error reading device label"); 2334 return (error); 2335 } 2336 if (txg == 0) { 2337 cmn_err(CE_NOTE, "this device is detached"); 2338 nvlist_free(conf); 2339 return (EINVAL); 2340 } 2341 2342 VERIFY(nvlist_lookup_nvlist(conf, ZPOOL_CONFIG_VDEV_TREE, 2343 &nvtop) == 0); 2344 VERIFY(nvlist_lookup_string(nvtop, ZPOOL_CONFIG_TYPE, &type) == 0); 2345 2346 if (strcmp(type, VDEV_TYPE_DISK) == 0) { 2347 if (spa_rootdev_validate(nvtop)) { 2348 goto out; 2349 } else { 2350 nvlist_free(conf); 2351 return (EINVAL); 2352 } 2353 } 2354 2355 ASSERT(strcmp(type, VDEV_TYPE_MIRROR) == 0); 2356 2357 VERIFY(nvlist_lookup_nvlist_array(nvtop, ZPOOL_CONFIG_CHILDREN, 2358 &child, &children) == 0); 2359 2360 /* 2361 * Go thru vdevs in the mirror to see if the given device 2362 * has the most recent txg. Only the device with the most 2363 * recent txg has valid information and should be booted. 2364 */ 2365 for (c = 0; c < children; c++) { 2366 char *cdevid, *cpath; 2367 uint64_t tmptxg; 2368 2369 if (nvlist_lookup_string(child[c], ZPOOL_CONFIG_PHYS_PATH, 2370 &cpath) != 0) 2371 return (EINVAL); 2372 if (nvlist_lookup_string(child[c], ZPOOL_CONFIG_DEVID, 2373 &cdevid) != 0) 2374 return (EINVAL); 2375 if ((spa_check_rootconf(cpath, cdevid, NULL, 2376 &tmptxg) == 0) && (tmptxg > txg)) { 2377 txg = tmptxg; 2378 VERIFY(nvlist_lookup_string(child[c], 2379 ZPOOL_CONFIG_PATH, &bootpath) == 0); 2380 } 2381 } 2382 2383 /* Does the best device match the one we've booted from? */ 2384 if (bootpath) { 2385 cmn_err(CE_NOTE, "try booting from '%s'", bootpath); 2386 return (EINVAL); 2387 } 2388out: 2389 *bestconf = conf; 2390 return (0); 2391} 2392 2393/* 2394 * Import a root pool. 2395 * 2396 * For x86. devpath_list will consist of devid and/or physpath name of 2397 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 2398 * The GRUB "findroot" command will return the vdev we should boot. 2399 * 2400 * For Sparc, devpath_list consists the physpath name of the booting device 2401 * no matter the rootpool is a single device pool or a mirrored pool. 2402 * e.g. 2403 * "/pci@1f,0/ide@d/disk@0,0:a" 2404 */ 2405int 2406spa_import_rootpool(char *devpath, char *devid) 2407{ 2408 nvlist_t *conf = NULL; 2409 char *pname; 2410 int error; 2411 2412 /* 2413 * Get the vdev pathname and configuation from the most 2414 * recently updated vdev (highest txg). 2415 */ 2416 if (error = spa_get_rootconf(devpath, devid, &conf)) 2417 goto msg_out; 2418 2419 /* 2420 * Add type "root" vdev to the config. 2421 */ 2422 spa_build_rootpool_config(conf); 2423 2424 VERIFY(nvlist_lookup_string(conf, ZPOOL_CONFIG_POOL_NAME, &pname) == 0); 2425 2426 /* 2427 * We specify 'allowfaulted' for this to be treated like spa_open() 2428 * instead of spa_import(). This prevents us from marking vdevs as 2429 * persistently unavailable, and generates FMA ereports as if it were a 2430 * pool open, not import. 2431 */ 2432 error = spa_import_common(pname, conf, NULL, B_TRUE, B_TRUE); 2433 if (error == EEXIST) 2434 error = 0; 2435 2436 nvlist_free(conf); 2437 return (error); 2438 2439msg_out: 2440 cmn_err(CE_NOTE, "\n" 2441 " *************************************************** \n" 2442 " * This device is not bootable! * \n" 2443 " * It is either offlined or detached or faulted. * \n" 2444 " * Please try to boot from a different device. * \n" 2445 " *************************************************** "); 2446 2447 return (error); 2448} 2449#endif 2450#endif 2451 2452/* 2453 * Import a non-root pool into the system. 2454 */ 2455int 2456spa_import(const char *pool, nvlist_t *config, nvlist_t *props) 2457{ 2458 return (spa_import_common(pool, config, props, B_FALSE, B_FALSE)); 2459} 2460 2461int 2462spa_import_faulted(const char *pool, nvlist_t *config, nvlist_t *props) 2463{ 2464 return (spa_import_common(pool, config, props, B_FALSE, B_TRUE)); 2465} 2466 2467 2468/* 2469 * This (illegal) pool name is used when temporarily importing a spa_t in order 2470 * to get the vdev stats associated with the imported devices. 2471 */ 2472#define TRYIMPORT_NAME "$import" 2473 2474nvlist_t * 2475spa_tryimport(nvlist_t *tryconfig) 2476{ 2477 nvlist_t *config = NULL; 2478 char *poolname; 2479 spa_t *spa; 2480 uint64_t state; 2481 2482 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 2483 return (NULL); 2484 2485 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 2486 return (NULL); 2487 2488 /* 2489 * Create and initialize the spa structure. 2490 */ 2491 mutex_enter(&spa_namespace_lock); 2492 spa = spa_add(TRYIMPORT_NAME, NULL); 2493 spa_activate(spa); 2494 2495 /* 2496 * Pass off the heavy lifting to spa_load(). 2497 * Pass TRUE for mosconfig because the user-supplied config 2498 * is actually the one to trust when doing an import. 2499 */ 2500 (void) spa_load(spa, tryconfig, SPA_LOAD_TRYIMPORT, B_TRUE); 2501 2502 /* 2503 * If 'tryconfig' was at least parsable, return the current config. 2504 */ 2505 if (spa->spa_root_vdev != NULL) { 2506 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2507 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 2508 poolname) == 0); 2509 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 2510 state) == 0); 2511 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 2512 spa->spa_uberblock.ub_timestamp) == 0); 2513 2514 /* 2515 * If the bootfs property exists on this pool then we 2516 * copy it out so that external consumers can tell which 2517 * pools are bootable. 2518 */ 2519 if (spa->spa_bootfs) { 2520 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2521 2522 /* 2523 * We have to play games with the name since the 2524 * pool was opened as TRYIMPORT_NAME. 2525 */ 2526 if (dsl_dsobj_to_dsname(spa_name(spa), 2527 spa->spa_bootfs, tmpname) == 0) { 2528 char *cp; 2529 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 2530 2531 cp = strchr(tmpname, '/'); 2532 if (cp == NULL) { 2533 (void) strlcpy(dsname, tmpname, 2534 MAXPATHLEN); 2535 } else { 2536 (void) snprintf(dsname, MAXPATHLEN, 2537 "%s/%s", poolname, ++cp); 2538 } 2539 VERIFY(nvlist_add_string(config, 2540 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 2541 kmem_free(dsname, MAXPATHLEN); 2542 } 2543 kmem_free(tmpname, MAXPATHLEN); 2544 } 2545 2546 /* 2547 * Add the list of hot spares and level 2 cache devices. 2548 */ 2549 spa_add_spares(spa, config); 2550 spa_add_l2cache(spa, config); 2551 } 2552 2553 spa_unload(spa); 2554 spa_deactivate(spa); 2555 spa_remove(spa); 2556 mutex_exit(&spa_namespace_lock); 2557 2558 return (config); 2559} 2560 2561/* 2562 * Pool export/destroy 2563 * 2564 * The act of destroying or exporting a pool is very simple. We make sure there 2565 * is no more pending I/O and any references to the pool are gone. Then, we 2566 * update the pool state and sync all the labels to disk, removing the 2567 * configuration from the cache afterwards. 2568 */ 2569static int 2570spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 2571 boolean_t force) 2572{ 2573 spa_t *spa; 2574 2575 if (oldconfig) 2576 *oldconfig = NULL; 2577 2578 if (!(spa_mode & FWRITE)) 2579 return (EROFS); 2580 2581 mutex_enter(&spa_namespace_lock); 2582 if ((spa = spa_lookup(pool)) == NULL) { 2583 mutex_exit(&spa_namespace_lock); 2584 return (ENOENT); 2585 } 2586 2587 /* 2588 * Put a hold on the pool, drop the namespace lock, stop async tasks, 2589 * reacquire the namespace lock, and see if we can export. 2590 */ 2591 spa_open_ref(spa, FTAG); 2592 mutex_exit(&spa_namespace_lock); 2593 spa_async_suspend(spa); 2594 mutex_enter(&spa_namespace_lock); 2595 spa_close(spa, FTAG); 2596 2597 /* 2598 * The pool will be in core if it's openable, 2599 * in which case we can modify its state. 2600 */ 2601 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 2602 /* 2603 * Objsets may be open only because they're dirty, so we 2604 * have to force it to sync before checking spa_refcnt. 2605 */ 2606 txg_wait_synced(spa->spa_dsl_pool, 0); 2607 2608 /* 2609 * A pool cannot be exported or destroyed if there are active 2610 * references. If we are resetting a pool, allow references by 2611 * fault injection handlers. 2612 */ 2613 if (!spa_refcount_zero(spa) || 2614 (spa->spa_inject_ref != 0 && 2615 new_state != POOL_STATE_UNINITIALIZED)) { 2616 spa_async_resume(spa); 2617 mutex_exit(&spa_namespace_lock); 2618 return (EBUSY); 2619 } 2620 2621 /* 2622 * A pool cannot be exported if it has an active shared spare. 2623 * This is to prevent other pools stealing the active spare 2624 * from an exported pool. At user's own will, such pool can 2625 * be forcedly exported. 2626 */ 2627 if (!force && new_state == POOL_STATE_EXPORTED && 2628 spa_has_active_shared_spare(spa)) { 2629 spa_async_resume(spa); 2630 mutex_exit(&spa_namespace_lock); 2631 return (EXDEV); 2632 } 2633 2634 /* 2635 * We want this to be reflected on every label, 2636 * so mark them all dirty. spa_unload() will do the 2637 * final sync that pushes these changes out. 2638 */ 2639 if (new_state != POOL_STATE_UNINITIALIZED) { 2640 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2641 spa->spa_state = new_state; 2642 spa->spa_final_txg = spa_last_synced_txg(spa) + 1; 2643 vdev_config_dirty(spa->spa_root_vdev); 2644 spa_config_exit(spa, SCL_ALL, FTAG); 2645 } 2646 } 2647 2648 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 2649 2650 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 2651 spa_unload(spa); 2652 spa_deactivate(spa); 2653 } 2654 2655 if (oldconfig && spa->spa_config) 2656 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 2657 2658 if (new_state != POOL_STATE_UNINITIALIZED) { 2659 spa_config_sync(spa, B_TRUE, B_TRUE); 2660 spa_remove(spa); 2661 } 2662 mutex_exit(&spa_namespace_lock); 2663 2664 return (0); 2665} 2666 2667/* 2668 * Destroy a storage pool. 2669 */ 2670int 2671spa_destroy(char *pool) 2672{ 2673 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, B_FALSE)); 2674} 2675 2676/* 2677 * Export a storage pool. 2678 */ 2679int 2680spa_export(char *pool, nvlist_t **oldconfig, boolean_t force) 2681{ 2682 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, force)); 2683} 2684 2685/* 2686 * Similar to spa_export(), this unloads the spa_t without actually removing it 2687 * from the namespace in any way. 2688 */ 2689int 2690spa_reset(char *pool) 2691{ 2692 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 2693 B_FALSE)); 2694} 2695 2696/* 2697 * ========================================================================== 2698 * Device manipulation 2699 * ========================================================================== 2700 */ 2701 2702/* 2703 * Add a device to a storage pool. 2704 */ 2705int 2706spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 2707{ 2708 uint64_t txg; 2709 int c, error; 2710 vdev_t *rvd = spa->spa_root_vdev; 2711 vdev_t *vd, *tvd; 2712 nvlist_t **spares, **l2cache; 2713 uint_t nspares, nl2cache; 2714 2715 txg = spa_vdev_enter(spa); 2716 2717 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 2718 VDEV_ALLOC_ADD)) != 0) 2719 return (spa_vdev_exit(spa, NULL, txg, error)); 2720 2721 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 2722 2723 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 2724 &nspares) != 0) 2725 nspares = 0; 2726 2727 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 2728 &nl2cache) != 0) 2729 nl2cache = 0; 2730 2731 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 2732 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 2733 2734 if (vd->vdev_children != 0 && 2735 (error = vdev_create(vd, txg, B_FALSE)) != 0) 2736 return (spa_vdev_exit(spa, vd, txg, error)); 2737 2738 /* 2739 * We must validate the spares and l2cache devices after checking the 2740 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 2741 */ 2742 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 2743 return (spa_vdev_exit(spa, vd, txg, error)); 2744 2745 /* 2746 * Transfer each new top-level vdev from vd to rvd. 2747 */ 2748 for (c = 0; c < vd->vdev_children; c++) { 2749 tvd = vd->vdev_child[c]; 2750 vdev_remove_child(vd, tvd); 2751 tvd->vdev_id = rvd->vdev_children; 2752 vdev_add_child(rvd, tvd); 2753 vdev_config_dirty(tvd); 2754 } 2755 2756 if (nspares != 0) { 2757 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 2758 ZPOOL_CONFIG_SPARES); 2759 spa_load_spares(spa); 2760 spa->spa_spares.sav_sync = B_TRUE; 2761 } 2762 2763 if (nl2cache != 0) { 2764 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 2765 ZPOOL_CONFIG_L2CACHE); 2766 spa_load_l2cache(spa); 2767 spa->spa_l2cache.sav_sync = B_TRUE; 2768 } 2769 2770 /* 2771 * We have to be careful when adding new vdevs to an existing pool. 2772 * If other threads start allocating from these vdevs before we 2773 * sync the config cache, and we lose power, then upon reboot we may 2774 * fail to open the pool because there are DVAs that the config cache 2775 * can't translate. Therefore, we first add the vdevs without 2776 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 2777 * and then let spa_config_update() initialize the new metaslabs. 2778 * 2779 * spa_load() checks for added-but-not-initialized vdevs, so that 2780 * if we lose power at any point in this sequence, the remaining 2781 * steps will be completed the next time we load the pool. 2782 */ 2783 (void) spa_vdev_exit(spa, vd, txg, 0); 2784 2785 mutex_enter(&spa_namespace_lock); 2786 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 2787 mutex_exit(&spa_namespace_lock); 2788 2789 return (0); 2790} 2791 2792/* 2793 * Attach a device to a mirror. The arguments are the path to any device 2794 * in the mirror, and the nvroot for the new device. If the path specifies 2795 * a device that is not mirrored, we automatically insert the mirror vdev. 2796 * 2797 * If 'replacing' is specified, the new device is intended to replace the 2798 * existing device; in this case the two devices are made into their own 2799 * mirror using the 'replacing' vdev, which is functionally identical to 2800 * the mirror vdev (it actually reuses all the same ops) but has a few 2801 * extra rules: you can't attach to it after it's been created, and upon 2802 * completion of resilvering, the first disk (the one being replaced) 2803 * is automatically detached. 2804 */ 2805int 2806spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 2807{ 2808 uint64_t txg, open_txg; 2809 vdev_t *rvd = spa->spa_root_vdev; 2810 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 2811 vdev_ops_t *pvops; 2812 dmu_tx_t *tx; 2813 char *oldvdpath, *newvdpath; 2814 int newvd_isspare; 2815 int error; 2816 2817 txg = spa_vdev_enter(spa); 2818 2819 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 2820 2821 if (oldvd == NULL) 2822 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 2823 2824 if (!oldvd->vdev_ops->vdev_op_leaf) 2825 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 2826 2827 pvd = oldvd->vdev_parent; 2828 2829 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 2830 VDEV_ALLOC_ADD)) != 0) 2831 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 2832 2833 if (newrootvd->vdev_children != 1) 2834 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 2835 2836 newvd = newrootvd->vdev_child[0]; 2837 2838 if (!newvd->vdev_ops->vdev_op_leaf) 2839 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 2840 2841 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 2842 return (spa_vdev_exit(spa, newrootvd, txg, error)); 2843 2844 /* 2845 * Spares can't replace logs 2846 */ 2847 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 2848 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 2849 2850 if (!replacing) { 2851 /* 2852 * For attach, the only allowable parent is a mirror or the root 2853 * vdev. 2854 */ 2855 if (pvd->vdev_ops != &vdev_mirror_ops && 2856 pvd->vdev_ops != &vdev_root_ops) 2857 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 2858 2859 pvops = &vdev_mirror_ops; 2860 } else { 2861 /* 2862 * Active hot spares can only be replaced by inactive hot 2863 * spares. 2864 */ 2865 if (pvd->vdev_ops == &vdev_spare_ops && 2866 pvd->vdev_child[1] == oldvd && 2867 !spa_has_spare(spa, newvd->vdev_guid)) 2868 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 2869 2870 /* 2871 * If the source is a hot spare, and the parent isn't already a 2872 * spare, then we want to create a new hot spare. Otherwise, we 2873 * want to create a replacing vdev. The user is not allowed to 2874 * attach to a spared vdev child unless the 'isspare' state is 2875 * the same (spare replaces spare, non-spare replaces 2876 * non-spare). 2877 */ 2878 if (pvd->vdev_ops == &vdev_replacing_ops) 2879 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 2880 else if (pvd->vdev_ops == &vdev_spare_ops && 2881 newvd->vdev_isspare != oldvd->vdev_isspare) 2882 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 2883 else if (pvd->vdev_ops != &vdev_spare_ops && 2884 newvd->vdev_isspare) 2885 pvops = &vdev_spare_ops; 2886 else 2887 pvops = &vdev_replacing_ops; 2888 } 2889 2890 /* 2891 * Compare the new device size with the replaceable/attachable 2892 * device size. 2893 */ 2894 if (newvd->vdev_psize < vdev_get_rsize(oldvd)) 2895 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 2896 2897 /* 2898 * The new device cannot have a higher alignment requirement 2899 * than the top-level vdev. 2900 */ 2901 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 2902 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 2903 2904 /* 2905 * If this is an in-place replacement, update oldvd's path and devid 2906 * to make it distinguishable from newvd, and unopenable from now on. 2907 */ 2908 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 2909 spa_strfree(oldvd->vdev_path); 2910 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 2911 KM_SLEEP); 2912 (void) sprintf(oldvd->vdev_path, "%s/%s", 2913 newvd->vdev_path, "old"); 2914 if (oldvd->vdev_devid != NULL) { 2915 spa_strfree(oldvd->vdev_devid); 2916 oldvd->vdev_devid = NULL; 2917 } 2918 } 2919 2920 /* 2921 * If the parent is not a mirror, or if we're replacing, insert the new 2922 * mirror/replacing/spare vdev above oldvd. 2923 */ 2924 if (pvd->vdev_ops != pvops) 2925 pvd = vdev_add_parent(oldvd, pvops); 2926 2927 ASSERT(pvd->vdev_top->vdev_parent == rvd); 2928 ASSERT(pvd->vdev_ops == pvops); 2929 ASSERT(oldvd->vdev_parent == pvd); 2930 2931 /* 2932 * Extract the new device from its root and add it to pvd. 2933 */ 2934 vdev_remove_child(newrootvd, newvd); 2935 newvd->vdev_id = pvd->vdev_children; 2936 vdev_add_child(pvd, newvd); 2937 2938 /* 2939 * If newvd is smaller than oldvd, but larger than its rsize, 2940 * the addition of newvd may have decreased our parent's asize. 2941 */ 2942 pvd->vdev_asize = MIN(pvd->vdev_asize, newvd->vdev_asize); 2943 2944 tvd = newvd->vdev_top; 2945 ASSERT(pvd->vdev_top == tvd); 2946 ASSERT(tvd->vdev_parent == rvd); 2947 2948 vdev_config_dirty(tvd); 2949 2950 /* 2951 * Set newvd's DTL to [TXG_INITIAL, open_txg]. It will propagate 2952 * upward when spa_vdev_exit() calls vdev_dtl_reassess(). 2953 */ 2954 open_txg = txg + TXG_CONCURRENT_STATES - 1; 2955 2956 mutex_enter(&newvd->vdev_dtl_lock); 2957 space_map_add(&newvd->vdev_dtl_map, TXG_INITIAL, 2958 open_txg - TXG_INITIAL + 1); 2959 mutex_exit(&newvd->vdev_dtl_lock); 2960 2961 if (newvd->vdev_isspare) 2962 spa_spare_activate(newvd); 2963 oldvdpath = spa_strdup(oldvd->vdev_path); 2964 newvdpath = spa_strdup(newvd->vdev_path); 2965 newvd_isspare = newvd->vdev_isspare; 2966 2967 /* 2968 * Mark newvd's DTL dirty in this txg. 2969 */ 2970 vdev_dirty(tvd, VDD_DTL, newvd, txg); 2971 2972 (void) spa_vdev_exit(spa, newrootvd, open_txg, 0); 2973 2974 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 2975 if (dmu_tx_assign(tx, TXG_WAIT) == 0) { 2976 spa_history_internal_log(LOG_POOL_VDEV_ATTACH, spa, tx, 2977 CRED(), "%s vdev=%s %s vdev=%s", 2978 replacing && newvd_isspare ? "spare in" : 2979 replacing ? "replace" : "attach", newvdpath, 2980 replacing ? "for" : "to", oldvdpath); 2981 dmu_tx_commit(tx); 2982 } else { 2983 dmu_tx_abort(tx); 2984 } 2985 2986 spa_strfree(oldvdpath); 2987 spa_strfree(newvdpath); 2988 2989 /* 2990 * Kick off a resilver to update newvd. 2991 */ 2992 VERIFY3U(spa_scrub(spa, POOL_SCRUB_RESILVER), ==, 0); 2993 2994 return (0); 2995} 2996 2997/* 2998 * Detach a device from a mirror or replacing vdev. 2999 * If 'replace_done' is specified, only detach if the parent 3000 * is a replacing vdev. 3001 */ 3002int 3003spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done) 3004{ 3005 uint64_t txg; 3006 int c, t, error; 3007 vdev_t *rvd = spa->spa_root_vdev; 3008 vdev_t *vd, *pvd, *cvd, *tvd; 3009 boolean_t unspare = B_FALSE; 3010 uint64_t unspare_guid; 3011 size_t len; 3012 3013 txg = spa_vdev_enter(spa); 3014 3015 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 3016 3017 if (vd == NULL) 3018 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 3019 3020 if (!vd->vdev_ops->vdev_op_leaf) 3021 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 3022 3023 pvd = vd->vdev_parent; 3024 3025 /* 3026 * If replace_done is specified, only remove this device if it's 3027 * the first child of a replacing vdev. For the 'spare' vdev, either 3028 * disk can be removed. 3029 */ 3030 if (replace_done) { 3031 if (pvd->vdev_ops == &vdev_replacing_ops) { 3032 if (vd->vdev_id != 0) 3033 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 3034 } else if (pvd->vdev_ops != &vdev_spare_ops) { 3035 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 3036 } 3037 } 3038 3039 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 3040 spa_version(spa) >= SPA_VERSION_SPARES); 3041 3042 /* 3043 * Only mirror, replacing, and spare vdevs support detach. 3044 */ 3045 if (pvd->vdev_ops != &vdev_replacing_ops && 3046 pvd->vdev_ops != &vdev_mirror_ops && 3047 pvd->vdev_ops != &vdev_spare_ops) 3048 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 3049 3050 /* 3051 * If there's only one replica, you can't detach it. 3052 */ 3053 if (pvd->vdev_children <= 1) 3054 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 3055 3056 /* 3057 * If all siblings have non-empty DTLs, this device may have the only 3058 * valid copy of the data, which means we cannot safely detach it. 3059 * 3060 * XXX -- as in the vdev_offline() case, we really want a more 3061 * precise DTL check. 3062 */ 3063 for (c = 0; c < pvd->vdev_children; c++) { 3064 uint64_t dirty; 3065 3066 cvd = pvd->vdev_child[c]; 3067 if (cvd == vd) 3068 continue; 3069 if (vdev_is_dead(cvd)) 3070 continue; 3071 mutex_enter(&cvd->vdev_dtl_lock); 3072 dirty = cvd->vdev_dtl_map.sm_space | 3073 cvd->vdev_dtl_scrub.sm_space; 3074 mutex_exit(&cvd->vdev_dtl_lock); 3075 if (!dirty) 3076 break; 3077 } 3078 3079 if (c == pvd->vdev_children) 3080 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 3081 3082 /* 3083 * If we are detaching the second disk from a replacing vdev, then 3084 * check to see if we changed the original vdev's path to have "/old" 3085 * at the end in spa_vdev_attach(). If so, undo that change now. 3086 */ 3087 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id == 1 && 3088 pvd->vdev_child[0]->vdev_path != NULL && 3089 pvd->vdev_child[1]->vdev_path != NULL) { 3090 ASSERT(pvd->vdev_child[1] == vd); 3091 cvd = pvd->vdev_child[0]; 3092 len = strlen(vd->vdev_path); 3093 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 3094 strcmp(cvd->vdev_path + len, "/old") == 0) { 3095 spa_strfree(cvd->vdev_path); 3096 cvd->vdev_path = spa_strdup(vd->vdev_path); 3097 } 3098 } 3099 3100 /* 3101 * If we are detaching the original disk from a spare, then it implies 3102 * that the spare should become a real disk, and be removed from the 3103 * active spare list for the pool. 3104 */ 3105 if (pvd->vdev_ops == &vdev_spare_ops && 3106 vd->vdev_id == 0) 3107 unspare = B_TRUE; 3108 3109 /* 3110 * Erase the disk labels so the disk can be used for other things. 3111 * This must be done after all other error cases are handled, 3112 * but before we disembowel vd (so we can still do I/O to it). 3113 * But if we can't do it, don't treat the error as fatal -- 3114 * it may be that the unwritability of the disk is the reason 3115 * it's being detached! 3116 */ 3117 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 3118 3119 /* 3120 * Remove vd from its parent and compact the parent's children. 3121 */ 3122 vdev_remove_child(pvd, vd); 3123 vdev_compact_children(pvd); 3124 3125 /* 3126 * Remember one of the remaining children so we can get tvd below. 3127 */ 3128 cvd = pvd->vdev_child[0]; 3129 3130 /* 3131 * If we need to remove the remaining child from the list of hot spares, 3132 * do it now, marking the vdev as no longer a spare in the process. We 3133 * must do this before vdev_remove_parent(), because that can change the 3134 * GUID if it creates a new toplevel GUID. 3135 */ 3136 if (unspare) { 3137 ASSERT(cvd->vdev_isspare); 3138 spa_spare_remove(cvd); 3139 unspare_guid = cvd->vdev_guid; 3140 } 3141 3142 /* 3143 * If the parent mirror/replacing vdev only has one child, 3144 * the parent is no longer needed. Remove it from the tree. 3145 */ 3146 if (pvd->vdev_children == 1) 3147 vdev_remove_parent(cvd); 3148 3149 /* 3150 * We don't set tvd until now because the parent we just removed 3151 * may have been the previous top-level vdev. 3152 */ 3153 tvd = cvd->vdev_top; 3154 ASSERT(tvd->vdev_parent == rvd); 3155 3156 /* 3157 * Reevaluate the parent vdev state. 3158 */ 3159 vdev_propagate_state(cvd); 3160 3161 /* 3162 * If the device we just detached was smaller than the others, it may be 3163 * possible to add metaslabs (i.e. grow the pool). vdev_metaslab_init() 3164 * can't fail because the existing metaslabs are already in core, so 3165 * there's nothing to read from disk. 3166 */ 3167 VERIFY(vdev_metaslab_init(tvd, txg) == 0); 3168 3169 vdev_config_dirty(tvd); 3170 3171 /* 3172 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 3173 * vd->vdev_detached is set and free vd's DTL object in syncing context. 3174 * But first make sure we're not on any *other* txg's DTL list, to 3175 * prevent vd from being accessed after it's freed. 3176 */ 3177 for (t = 0; t < TXG_SIZE; t++) 3178 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 3179 vd->vdev_detached = B_TRUE; 3180 vdev_dirty(tvd, VDD_DTL, vd, txg); 3181 3182 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 3183 3184 error = spa_vdev_exit(spa, vd, txg, 0); 3185 3186 /* 3187 * If this was the removal of the original device in a hot spare vdev, 3188 * then we want to go through and remove the device from the hot spare 3189 * list of every other pool. 3190 */ 3191 if (unspare) { 3192 spa = NULL; 3193 mutex_enter(&spa_namespace_lock); 3194 while ((spa = spa_next(spa)) != NULL) { 3195 if (spa->spa_state != POOL_STATE_ACTIVE) 3196 continue; 3197 spa_open_ref(spa, FTAG); 3198 mutex_exit(&spa_namespace_lock); 3199 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 3200 mutex_enter(&spa_namespace_lock); 3201 spa_close(spa, FTAG); 3202 } 3203 mutex_exit(&spa_namespace_lock); 3204 } 3205 3206 return (error); 3207} 3208 3209static nvlist_t * 3210spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 3211{ 3212 for (int i = 0; i < count; i++) { 3213 uint64_t guid; 3214 3215 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 3216 &guid) == 0); 3217 3218 if (guid == target_guid) 3219 return (nvpp[i]); 3220 } 3221 3222 return (NULL); 3223} 3224 3225static void 3226spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 3227 nvlist_t *dev_to_remove) 3228{ 3229 nvlist_t **newdev = NULL; 3230 3231 if (count > 1) 3232 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 3233 3234 for (int i = 0, j = 0; i < count; i++) { 3235 if (dev[i] == dev_to_remove) 3236 continue; 3237 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 3238 } 3239 3240 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 3241 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 3242 3243 for (int i = 0; i < count - 1; i++) 3244 nvlist_free(newdev[i]); 3245 3246 if (count > 1) 3247 kmem_free(newdev, (count - 1) * sizeof (void *)); 3248} 3249 3250/* 3251 * Remove a device from the pool. Currently, this supports removing only hot 3252 * spares and level 2 ARC devices. 3253 */ 3254int 3255spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 3256{ 3257 vdev_t *vd; 3258 nvlist_t **spares, **l2cache, *nv; 3259 uint_t nspares, nl2cache; 3260 uint64_t txg; 3261 int error = 0; 3262 3263 txg = spa_vdev_enter(spa); 3264 3265 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 3266 3267 if (spa->spa_spares.sav_vdevs != NULL && 3268 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3269 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 3270 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 3271 /* 3272 * Only remove the hot spare if it's not currently in use 3273 * in this pool. 3274 */ 3275 if (vd == NULL || unspare) { 3276 spa_vdev_remove_aux(spa->spa_spares.sav_config, 3277 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 3278 spa_load_spares(spa); 3279 spa->spa_spares.sav_sync = B_TRUE; 3280 } else { 3281 error = EBUSY; 3282 } 3283 } else if (spa->spa_l2cache.sav_vdevs != NULL && 3284 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3285 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 3286 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 3287 /* 3288 * Cache devices can always be removed. 3289 */ 3290 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 3291 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 3292 spa_load_l2cache(spa); 3293 spa->spa_l2cache.sav_sync = B_TRUE; 3294 } else if (vd != NULL) { 3295 /* 3296 * Normal vdevs cannot be removed (yet). 3297 */ 3298 error = ENOTSUP; 3299 } else { 3300 /* 3301 * There is no vdev of any kind with the specified guid. 3302 */ 3303 error = ENOENT; 3304 } 3305 3306 return (spa_vdev_exit(spa, NULL, txg, error)); 3307} 3308 3309/* 3310 * Find any device that's done replacing, or a vdev marked 'unspare' that's 3311 * current spared, so we can detach it. 3312 */ 3313static vdev_t * 3314spa_vdev_resilver_done_hunt(vdev_t *vd) 3315{ 3316 vdev_t *newvd, *oldvd; 3317 int c; 3318 3319 for (c = 0; c < vd->vdev_children; c++) { 3320 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 3321 if (oldvd != NULL) 3322 return (oldvd); 3323 } 3324 3325 /* 3326 * Check for a completed replacement. 3327 */ 3328 if (vd->vdev_ops == &vdev_replacing_ops && vd->vdev_children == 2) { 3329 oldvd = vd->vdev_child[0]; 3330 newvd = vd->vdev_child[1]; 3331 3332 mutex_enter(&newvd->vdev_dtl_lock); 3333 if (newvd->vdev_dtl_map.sm_space == 0 && 3334 newvd->vdev_dtl_scrub.sm_space == 0) { 3335 mutex_exit(&newvd->vdev_dtl_lock); 3336 return (oldvd); 3337 } 3338 mutex_exit(&newvd->vdev_dtl_lock); 3339 } 3340 3341 /* 3342 * Check for a completed resilver with the 'unspare' flag set. 3343 */ 3344 if (vd->vdev_ops == &vdev_spare_ops && vd->vdev_children == 2) { 3345 newvd = vd->vdev_child[0]; 3346 oldvd = vd->vdev_child[1]; 3347 3348 mutex_enter(&newvd->vdev_dtl_lock); 3349 if (newvd->vdev_unspare && 3350 newvd->vdev_dtl_map.sm_space == 0 && 3351 newvd->vdev_dtl_scrub.sm_space == 0) { 3352 newvd->vdev_unspare = 0; 3353 mutex_exit(&newvd->vdev_dtl_lock); 3354 return (oldvd); 3355 } 3356 mutex_exit(&newvd->vdev_dtl_lock); 3357 } 3358 3359 return (NULL); 3360} 3361 3362static void 3363spa_vdev_resilver_done(spa_t *spa) 3364{ 3365 vdev_t *vd; 3366 vdev_t *pvd; 3367 uint64_t guid; 3368 uint64_t pguid = 0; 3369 3370 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3371 3372 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 3373 guid = vd->vdev_guid; 3374 /* 3375 * If we have just finished replacing a hot spared device, then 3376 * we need to detach the parent's first child (the original hot 3377 * spare) as well. 3378 */ 3379 pvd = vd->vdev_parent; 3380 if (pvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3381 pvd->vdev_id == 0) { 3382 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 3383 ASSERT(pvd->vdev_parent->vdev_children == 2); 3384 pguid = pvd->vdev_parent->vdev_child[1]->vdev_guid; 3385 } 3386 spa_config_exit(spa, SCL_CONFIG, FTAG); 3387 if (spa_vdev_detach(spa, guid, B_TRUE) != 0) 3388 return; 3389 if (pguid != 0 && spa_vdev_detach(spa, pguid, B_TRUE) != 0) 3390 return; 3391 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3392 } 3393 3394 spa_config_exit(spa, SCL_CONFIG, FTAG); 3395} 3396 3397/* 3398 * Update the stored path for this vdev. Dirty the vdev configuration, relying 3399 * on spa_vdev_enter/exit() to synchronize the labels and cache. 3400 */ 3401int 3402spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 3403{ 3404 vdev_t *vd; 3405 uint64_t txg; 3406 3407 txg = spa_vdev_enter(spa); 3408 3409 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) { 3410 /* 3411 * Determine if this is a reference to a hot spare device. If 3412 * it is, update the path manually as there is no associated 3413 * vdev_t that can be synced to disk. 3414 */ 3415 nvlist_t **spares; 3416 uint_t i, nspares; 3417 3418 if (spa->spa_spares.sav_config != NULL) { 3419 VERIFY(nvlist_lookup_nvlist_array( 3420 spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 3421 &spares, &nspares) == 0); 3422 for (i = 0; i < nspares; i++) { 3423 uint64_t theguid; 3424 VERIFY(nvlist_lookup_uint64(spares[i], 3425 ZPOOL_CONFIG_GUID, &theguid) == 0); 3426 if (theguid == guid) { 3427 VERIFY(nvlist_add_string(spares[i], 3428 ZPOOL_CONFIG_PATH, newpath) == 0); 3429 spa_load_spares(spa); 3430 spa->spa_spares.sav_sync = B_TRUE; 3431 return (spa_vdev_exit(spa, NULL, txg, 3432 0)); 3433 } 3434 } 3435 } 3436 3437 return (spa_vdev_exit(spa, NULL, txg, ENOENT)); 3438 } 3439 3440 if (!vd->vdev_ops->vdev_op_leaf) 3441 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 3442 3443 spa_strfree(vd->vdev_path); 3444 vd->vdev_path = spa_strdup(newpath); 3445 3446 vdev_config_dirty(vd->vdev_top); 3447 3448 return (spa_vdev_exit(spa, NULL, txg, 0)); 3449} 3450 3451/* 3452 * ========================================================================== 3453 * SPA Scrubbing 3454 * ========================================================================== 3455 */ 3456 3457int 3458spa_scrub(spa_t *spa, pool_scrub_type_t type) 3459{ 3460 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 3461 3462 if ((uint_t)type >= POOL_SCRUB_TYPES) 3463 return (ENOTSUP); 3464 3465 /* 3466 * If a resilver was requested, but there is no DTL on a 3467 * writeable leaf device, we have nothing to do. 3468 */ 3469 if (type == POOL_SCRUB_RESILVER && 3470 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 3471 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 3472 return (0); 3473 } 3474 3475 if (type == POOL_SCRUB_EVERYTHING && 3476 spa->spa_dsl_pool->dp_scrub_func != SCRUB_FUNC_NONE && 3477 spa->spa_dsl_pool->dp_scrub_isresilver) 3478 return (EBUSY); 3479 3480 if (type == POOL_SCRUB_EVERYTHING || type == POOL_SCRUB_RESILVER) { 3481 return (dsl_pool_scrub_clean(spa->spa_dsl_pool)); 3482 } else if (type == POOL_SCRUB_NONE) { 3483 return (dsl_pool_scrub_cancel(spa->spa_dsl_pool)); 3484 } else { 3485 return (EINVAL); 3486 } 3487} 3488 3489/* 3490 * ========================================================================== 3491 * SPA async task processing 3492 * ========================================================================== 3493 */ 3494 3495static void 3496spa_async_remove(spa_t *spa, vdev_t *vd) 3497{ 3498 if (vd->vdev_remove_wanted) { 3499 vd->vdev_remove_wanted = 0; 3500 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 3501 vdev_clear(spa, vd); 3502 vdev_state_dirty(vd->vdev_top); 3503 } 3504 3505 for (int c = 0; c < vd->vdev_children; c++) 3506 spa_async_remove(spa, vd->vdev_child[c]); 3507} 3508 3509static void 3510spa_async_probe(spa_t *spa, vdev_t *vd) 3511{ 3512 if (vd->vdev_probe_wanted) { 3513 vd->vdev_probe_wanted = 0; 3514 vdev_reopen(vd); /* vdev_open() does the actual probe */ 3515 } 3516 3517 for (int c = 0; c < vd->vdev_children; c++) 3518 spa_async_probe(spa, vd->vdev_child[c]); 3519} 3520 3521static void 3522spa_async_thread(void *arg) 3523{ 3524 spa_t *spa = arg; 3525 int tasks; 3526 3527 ASSERT(spa->spa_sync_on); 3528 3529 mutex_enter(&spa->spa_async_lock); 3530 tasks = spa->spa_async_tasks; 3531 spa->spa_async_tasks = 0; 3532 mutex_exit(&spa->spa_async_lock); 3533 3534 /* 3535 * See if the config needs to be updated. 3536 */ 3537 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 3538 mutex_enter(&spa_namespace_lock); 3539 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 3540 mutex_exit(&spa_namespace_lock); 3541 } 3542 3543 /* 3544 * See if any devices need to be marked REMOVED. 3545 */ 3546 if (tasks & SPA_ASYNC_REMOVE) { 3547 spa_vdev_state_enter(spa); 3548 spa_async_remove(spa, spa->spa_root_vdev); 3549 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 3550 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 3551 for (int i = 0; i < spa->spa_spares.sav_count; i++) 3552 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 3553 (void) spa_vdev_state_exit(spa, NULL, 0); 3554 } 3555 3556 /* 3557 * See if any devices need to be probed. 3558 */ 3559 if (tasks & SPA_ASYNC_PROBE) { 3560 spa_vdev_state_enter(spa); 3561 spa_async_probe(spa, spa->spa_root_vdev); 3562 (void) spa_vdev_state_exit(spa, NULL, 0); 3563 } 3564 3565 /* 3566 * If any devices are done replacing, detach them. 3567 */ 3568 if (tasks & SPA_ASYNC_RESILVER_DONE) 3569 spa_vdev_resilver_done(spa); 3570 3571 /* 3572 * Kick off a resilver. 3573 */ 3574 if (tasks & SPA_ASYNC_RESILVER) 3575 VERIFY(spa_scrub(spa, POOL_SCRUB_RESILVER) == 0); 3576 3577 /* 3578 * Let the world know that we're done. 3579 */ 3580 mutex_enter(&spa->spa_async_lock); 3581 spa->spa_async_thread = NULL; 3582 cv_broadcast(&spa->spa_async_cv); 3583 mutex_exit(&spa->spa_async_lock); 3584 thread_exit(); 3585} 3586 3587void 3588spa_async_suspend(spa_t *spa) 3589{ 3590 mutex_enter(&spa->spa_async_lock); 3591 spa->spa_async_suspended++; 3592 while (spa->spa_async_thread != NULL) 3593 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 3594 mutex_exit(&spa->spa_async_lock); 3595} 3596 3597void 3598spa_async_resume(spa_t *spa) 3599{ 3600 mutex_enter(&spa->spa_async_lock); 3601 ASSERT(spa->spa_async_suspended != 0); 3602 spa->spa_async_suspended--; 3603 mutex_exit(&spa->spa_async_lock); 3604} 3605 3606static void 3607spa_async_dispatch(spa_t *spa) 3608{ 3609 mutex_enter(&spa->spa_async_lock); 3610 if (spa->spa_async_tasks && !spa->spa_async_suspended && 3611 spa->spa_async_thread == NULL && 3612 rootdir != NULL && !vn_is_readonly(rootdir)) 3613 spa->spa_async_thread = thread_create(NULL, 0, 3614 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 3615 mutex_exit(&spa->spa_async_lock); 3616} 3617 3618void 3619spa_async_request(spa_t *spa, int task) 3620{ 3621 mutex_enter(&spa->spa_async_lock); 3622 spa->spa_async_tasks |= task; 3623 mutex_exit(&spa->spa_async_lock); 3624} 3625 3626/* 3627 * ========================================================================== 3628 * SPA syncing routines 3629 * ========================================================================== 3630 */ 3631 3632static void 3633spa_sync_deferred_frees(spa_t *spa, uint64_t txg) 3634{ 3635 bplist_t *bpl = &spa->spa_sync_bplist; 3636 dmu_tx_t *tx; 3637 blkptr_t blk; 3638 uint64_t itor = 0; 3639 zio_t *zio; 3640 int error; 3641 uint8_t c = 1; 3642 3643 zio = zio_root(spa, NULL, NULL, ZIO_FLAG_CANFAIL); 3644 3645 while (bplist_iterate(bpl, &itor, &blk) == 0) { 3646 ASSERT(blk.blk_birth < txg); 3647 zio_nowait(zio_free(zio, spa, txg, &blk, NULL, NULL, 3648 ZIO_FLAG_MUSTSUCCEED)); 3649 } 3650 3651 error = zio_wait(zio); 3652 ASSERT3U(error, ==, 0); 3653 3654 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 3655 bplist_vacate(bpl, tx); 3656 3657 /* 3658 * Pre-dirty the first block so we sync to convergence faster. 3659 * (Usually only the first block is needed.) 3660 */ 3661 dmu_write(spa->spa_meta_objset, spa->spa_sync_bplist_obj, 0, 1, &c, tx); 3662 dmu_tx_commit(tx); 3663} 3664 3665static void 3666spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 3667{ 3668 char *packed = NULL; 3669 size_t bufsize; 3670 size_t nvsize = 0; 3671 dmu_buf_t *db; 3672 3673 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 3674 3675 /* 3676 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 3677 * information. This avoids the dbuf_will_dirty() path and 3678 * saves us a pre-read to get data we don't actually care about. 3679 */ 3680 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE); 3681 packed = kmem_alloc(bufsize, KM_SLEEP); 3682 3683 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 3684 KM_SLEEP) == 0); 3685 bzero(packed + nvsize, bufsize - nvsize); 3686 3687 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 3688 3689 kmem_free(packed, bufsize); 3690 3691 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 3692 dmu_buf_will_dirty(db, tx); 3693 *(uint64_t *)db->db_data = nvsize; 3694 dmu_buf_rele(db, FTAG); 3695} 3696 3697static void 3698spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 3699 const char *config, const char *entry) 3700{ 3701 nvlist_t *nvroot; 3702 nvlist_t **list; 3703 int i; 3704 3705 if (!sav->sav_sync) 3706 return; 3707 3708 /* 3709 * Update the MOS nvlist describing the list of available devices. 3710 * spa_validate_aux() will have already made sure this nvlist is 3711 * valid and the vdevs are labeled appropriately. 3712 */ 3713 if (sav->sav_object == 0) { 3714 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 3715 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 3716 sizeof (uint64_t), tx); 3717 VERIFY(zap_update(spa->spa_meta_objset, 3718 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 3719 &sav->sav_object, tx) == 0); 3720 } 3721 3722 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3723 if (sav->sav_count == 0) { 3724 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 3725 } else { 3726 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 3727 for (i = 0; i < sav->sav_count; i++) 3728 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 3729 B_FALSE, B_FALSE, B_TRUE); 3730 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 3731 sav->sav_count) == 0); 3732 for (i = 0; i < sav->sav_count; i++) 3733 nvlist_free(list[i]); 3734 kmem_free(list, sav->sav_count * sizeof (void *)); 3735 } 3736 3737 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 3738 nvlist_free(nvroot); 3739 3740 sav->sav_sync = B_FALSE; 3741} 3742 3743static void 3744spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 3745{ 3746 nvlist_t *config; 3747 3748 if (list_is_empty(&spa->spa_config_dirty_list)) 3749 return; 3750 3751 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 3752 3753 config = spa_config_generate(spa, spa->spa_root_vdev, 3754 dmu_tx_get_txg(tx), B_FALSE); 3755 3756 spa_config_exit(spa, SCL_STATE, FTAG); 3757 3758 if (spa->spa_config_syncing) 3759 nvlist_free(spa->spa_config_syncing); 3760 spa->spa_config_syncing = config; 3761 3762 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 3763} 3764 3765/* 3766 * Set zpool properties. 3767 */ 3768static void 3769spa_sync_props(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) 3770{ 3771 spa_t *spa = arg1; 3772 objset_t *mos = spa->spa_meta_objset; 3773 nvlist_t *nvp = arg2; 3774 nvpair_t *elem; 3775 uint64_t intval; 3776 char *strval; 3777 zpool_prop_t prop; 3778 const char *propname; 3779 zprop_type_t proptype; 3780 spa_config_dirent_t *dp; 3781 3782 mutex_enter(&spa->spa_props_lock); 3783 3784 elem = NULL; 3785 while ((elem = nvlist_next_nvpair(nvp, elem))) { 3786 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 3787 case ZPOOL_PROP_VERSION: 3788 /* 3789 * Only set version for non-zpool-creation cases 3790 * (set/import). spa_create() needs special care 3791 * for version setting. 3792 */ 3793 if (tx->tx_txg != TXG_INITIAL) { 3794 VERIFY(nvpair_value_uint64(elem, 3795 &intval) == 0); 3796 ASSERT(intval <= SPA_VERSION); 3797 ASSERT(intval >= spa_version(spa)); 3798 spa->spa_uberblock.ub_version = intval; 3799 vdev_config_dirty(spa->spa_root_vdev); 3800 } 3801 break; 3802 3803 case ZPOOL_PROP_ALTROOT: 3804 /* 3805 * 'altroot' is a non-persistent property. It should 3806 * have been set temporarily at creation or import time. 3807 */ 3808 ASSERT(spa->spa_root != NULL); 3809 break; 3810 3811 case ZPOOL_PROP_CACHEFILE: 3812 /* 3813 * 'cachefile' is a non-persistent property, but note 3814 * an async request that the config cache needs to be 3815 * udpated. 3816 */ 3817 VERIFY(nvpair_value_string(elem, &strval) == 0); 3818 3819 dp = kmem_alloc(sizeof (spa_config_dirent_t), KM_SLEEP); 3820 3821 if (strval[0] == '\0') 3822 dp->scd_path = spa_strdup(spa_config_path); 3823 else if (strcmp(strval, "none") == 0) 3824 dp->scd_path = NULL; 3825 else 3826 dp->scd_path = spa_strdup(strval); 3827 3828 list_insert_head(&spa->spa_config_list, dp); 3829 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3830 break; 3831 default: 3832 /* 3833 * Set pool property values in the poolprops mos object. 3834 */ 3835 if (spa->spa_pool_props_object == 0) { 3836 objset_t *mos = spa->spa_meta_objset; 3837 3838 VERIFY((spa->spa_pool_props_object = 3839 zap_create(mos, DMU_OT_POOL_PROPS, 3840 DMU_OT_NONE, 0, tx)) > 0); 3841 3842 VERIFY(zap_update(mos, 3843 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 3844 8, 1, &spa->spa_pool_props_object, tx) 3845 == 0); 3846 } 3847 3848 /* normalize the property name */ 3849 propname = zpool_prop_to_name(prop); 3850 proptype = zpool_prop_get_type(prop); 3851 3852 if (nvpair_type(elem) == DATA_TYPE_STRING) { 3853 ASSERT(proptype == PROP_TYPE_STRING); 3854 VERIFY(nvpair_value_string(elem, &strval) == 0); 3855 VERIFY(zap_update(mos, 3856 spa->spa_pool_props_object, propname, 3857 1, strlen(strval) + 1, strval, tx) == 0); 3858 3859 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 3860 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 3861 3862 if (proptype == PROP_TYPE_INDEX) { 3863 const char *unused; 3864 VERIFY(zpool_prop_index_to_string( 3865 prop, intval, &unused) == 0); 3866 } 3867 VERIFY(zap_update(mos, 3868 spa->spa_pool_props_object, propname, 3869 8, 1, &intval, tx) == 0); 3870 } else { 3871 ASSERT(0); /* not allowed */ 3872 } 3873 3874 switch (prop) { 3875 case ZPOOL_PROP_DELEGATION: 3876 spa->spa_delegation = intval; 3877 break; 3878 case ZPOOL_PROP_BOOTFS: 3879 spa->spa_bootfs = intval; 3880 break; 3881 case ZPOOL_PROP_FAILUREMODE: 3882 spa->spa_failmode = intval; 3883 break; 3884 default: 3885 break; 3886 } 3887 } 3888 3889 /* log internal history if this is not a zpool create */ 3890 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY && 3891 tx->tx_txg != TXG_INITIAL) { 3892 spa_history_internal_log(LOG_POOL_PROPSET, 3893 spa, tx, cr, "%s %lld %s", 3894 nvpair_name(elem), intval, spa_name(spa)); 3895 } 3896 } 3897 3898 mutex_exit(&spa->spa_props_lock); 3899} 3900 3901/* 3902 * Sync the specified transaction group. New blocks may be dirtied as 3903 * part of the process, so we iterate until it converges. 3904 */ 3905void 3906spa_sync(spa_t *spa, uint64_t txg) 3907{ 3908 dsl_pool_t *dp = spa->spa_dsl_pool; 3909 objset_t *mos = spa->spa_meta_objset; 3910 bplist_t *bpl = &spa->spa_sync_bplist; 3911 vdev_t *rvd = spa->spa_root_vdev; 3912 vdev_t *vd; 3913 dmu_tx_t *tx; 3914 int dirty_vdevs; 3915 int error; 3916 3917 /* 3918 * Lock out configuration changes. 3919 */ 3920 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3921 3922 spa->spa_syncing_txg = txg; 3923 spa->spa_sync_pass = 0; 3924 3925 /* 3926 * If there are any pending vdev state changes, convert them 3927 * into config changes that go out with this transaction group. 3928 */ 3929 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 3930 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 3931 vdev_state_clean(vd); 3932 vdev_config_dirty(vd); 3933 } 3934 spa_config_exit(spa, SCL_STATE, FTAG); 3935 3936 VERIFY(0 == bplist_open(bpl, mos, spa->spa_sync_bplist_obj)); 3937 3938 tx = dmu_tx_create_assigned(dp, txg); 3939 3940 /* 3941 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 3942 * set spa_deflate if we have no raid-z vdevs. 3943 */ 3944 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 3945 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 3946 int i; 3947 3948 for (i = 0; i < rvd->vdev_children; i++) { 3949 vd = rvd->vdev_child[i]; 3950 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 3951 break; 3952 } 3953 if (i == rvd->vdev_children) { 3954 spa->spa_deflate = TRUE; 3955 VERIFY(0 == zap_add(spa->spa_meta_objset, 3956 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3957 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 3958 } 3959 } 3960 3961 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 3962 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 3963 dsl_pool_create_origin(dp, tx); 3964 3965 /* Keeping the origin open increases spa_minref */ 3966 spa->spa_minref += 3; 3967 } 3968 3969 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 3970 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 3971 dsl_pool_upgrade_clones(dp, tx); 3972 } 3973 3974 /* 3975 * If anything has changed in this txg, push the deferred frees 3976 * from the previous txg. If not, leave them alone so that we 3977 * don't generate work on an otherwise idle system. 3978 */ 3979 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 3980 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 3981 !txg_list_empty(&dp->dp_sync_tasks, txg)) 3982 spa_sync_deferred_frees(spa, txg); 3983 3984 /* 3985 * Iterate to convergence. 3986 */ 3987 do { 3988 spa->spa_sync_pass++; 3989 3990 spa_sync_config_object(spa, tx); 3991 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 3992 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 3993 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 3994 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 3995 spa_errlog_sync(spa, txg); 3996 dsl_pool_sync(dp, txg); 3997 3998 dirty_vdevs = 0; 3999 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) { 4000 vdev_sync(vd, txg); 4001 dirty_vdevs++; 4002 } 4003 4004 bplist_sync(bpl, tx); 4005 } while (dirty_vdevs); 4006 4007 bplist_close(bpl); 4008 4009 dprintf("txg %llu passes %d\n", txg, spa->spa_sync_pass); 4010 4011 /* 4012 * Rewrite the vdev configuration (which includes the uberblock) 4013 * to commit the transaction group. 4014 * 4015 * If there are no dirty vdevs, we sync the uberblock to a few 4016 * random top-level vdevs that are known to be visible in the 4017 * config cache (see spa_vdev_add() for a complete description). 4018 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 4019 */ 4020 for (;;) { 4021 /* 4022 * We hold SCL_STATE to prevent vdev open/close/etc. 4023 * while we're attempting to write the vdev labels. 4024 */ 4025 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 4026 4027 if (list_is_empty(&spa->spa_config_dirty_list)) { 4028 vdev_t *svd[SPA_DVAS_PER_BP]; 4029 int svdcount = 0; 4030 int children = rvd->vdev_children; 4031 int c0 = spa_get_random(children); 4032 int c; 4033 4034 for (c = 0; c < children; c++) { 4035 vd = rvd->vdev_child[(c0 + c) % children]; 4036 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 4037 continue; 4038 svd[svdcount++] = vd; 4039 if (svdcount == SPA_DVAS_PER_BP) 4040 break; 4041 } 4042 error = vdev_config_sync(svd, svdcount, txg); 4043 } else { 4044 error = vdev_config_sync(rvd->vdev_child, 4045 rvd->vdev_children, txg); 4046 } 4047 4048 spa_config_exit(spa, SCL_STATE, FTAG); 4049 4050 if (error == 0) 4051 break; 4052 zio_suspend(spa, NULL); 4053 zio_resume_wait(spa); 4054 } 4055 dmu_tx_commit(tx); 4056 4057 /* 4058 * Clear the dirty config list. 4059 */ 4060 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 4061 vdev_config_clean(vd); 4062 4063 /* 4064 * Now that the new config has synced transactionally, 4065 * let it become visible to the config cache. 4066 */ 4067 if (spa->spa_config_syncing != NULL) { 4068 spa_config_set(spa, spa->spa_config_syncing); 4069 spa->spa_config_txg = txg; 4070 spa->spa_config_syncing = NULL; 4071 } 4072 4073 spa->spa_traverse_wanted = B_TRUE; 4074 rw_enter(&spa->spa_traverse_lock, RW_WRITER); 4075 spa->spa_traverse_wanted = B_FALSE; 4076 spa->spa_ubsync = spa->spa_uberblock; 4077 rw_exit(&spa->spa_traverse_lock); 4078 4079 /* 4080 * Clean up the ZIL records for the synced txg. 4081 */ 4082 dsl_pool_zil_clean(dp); 4083 4084 /* 4085 * Update usable space statistics. 4086 */ 4087 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 4088 vdev_sync_done(vd, txg); 4089 4090 /* 4091 * It had better be the case that we didn't dirty anything 4092 * since vdev_config_sync(). 4093 */ 4094 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 4095 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 4096 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 4097 ASSERT(bpl->bpl_queue == NULL); 4098 4099 spa_config_exit(spa, SCL_CONFIG, FTAG); 4100 4101 /* 4102 * If any async tasks have been requested, kick them off. 4103 */ 4104 spa_async_dispatch(spa); 4105} 4106 4107/* 4108 * Sync all pools. We don't want to hold the namespace lock across these 4109 * operations, so we take a reference on the spa_t and drop the lock during the 4110 * sync. 4111 */ 4112void 4113spa_sync_allpools(void) 4114{ 4115 spa_t *spa = NULL; 4116 mutex_enter(&spa_namespace_lock); 4117 while ((spa = spa_next(spa)) != NULL) { 4118 if (spa_state(spa) != POOL_STATE_ACTIVE || spa_suspended(spa)) 4119 continue; 4120 spa_open_ref(spa, FTAG); 4121 mutex_exit(&spa_namespace_lock); 4122 txg_wait_synced(spa_get_dsl(spa), 0); 4123 mutex_enter(&spa_namespace_lock); 4124 spa_close(spa, FTAG); 4125 } 4126 mutex_exit(&spa_namespace_lock); 4127} 4128 4129/* 4130 * ========================================================================== 4131 * Miscellaneous routines 4132 * ========================================================================== 4133 */ 4134 4135/* 4136 * Remove all pools in the system. 4137 */ 4138void 4139spa_evict_all(void) 4140{ 4141 spa_t *spa; 4142 4143 /* 4144 * Remove all cached state. All pools should be closed now, 4145 * so every spa in the AVL tree should be unreferenced. 4146 */ 4147 mutex_enter(&spa_namespace_lock); 4148 while ((spa = spa_next(NULL)) != NULL) { 4149 /* 4150 * Stop async tasks. The async thread may need to detach 4151 * a device that's been replaced, which requires grabbing 4152 * spa_namespace_lock, so we must drop it here. 4153 */ 4154 spa_open_ref(spa, FTAG); 4155 mutex_exit(&spa_namespace_lock); 4156 spa_async_suspend(spa); 4157 mutex_enter(&spa_namespace_lock); 4158 spa_close(spa, FTAG); 4159 4160 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4161 spa_unload(spa); 4162 spa_deactivate(spa); 4163 } 4164 spa_remove(spa); 4165 } 4166 mutex_exit(&spa_namespace_lock); 4167} 4168 4169vdev_t * 4170spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t l2cache) 4171{ 4172 vdev_t *vd; 4173 int i; 4174 4175 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 4176 return (vd); 4177 4178 if (l2cache) { 4179 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 4180 vd = spa->spa_l2cache.sav_vdevs[i]; 4181 if (vd->vdev_guid == guid) 4182 return (vd); 4183 } 4184 } 4185 4186 return (NULL); 4187} 4188 4189void 4190spa_upgrade(spa_t *spa, uint64_t version) 4191{ 4192 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4193 4194 /* 4195 * This should only be called for a non-faulted pool, and since a 4196 * future version would result in an unopenable pool, this shouldn't be 4197 * possible. 4198 */ 4199 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION); 4200 ASSERT(version >= spa->spa_uberblock.ub_version); 4201 4202 spa->spa_uberblock.ub_version = version; 4203 vdev_config_dirty(spa->spa_root_vdev); 4204 4205 spa_config_exit(spa, SCL_ALL, FTAG); 4206 4207 txg_wait_synced(spa_get_dsl(spa), 0); 4208} 4209 4210boolean_t 4211spa_has_spare(spa_t *spa, uint64_t guid) 4212{ 4213 int i; 4214 uint64_t spareguid; 4215 spa_aux_vdev_t *sav = &spa->spa_spares; 4216 4217 for (i = 0; i < sav->sav_count; i++) 4218 if (sav->sav_vdevs[i]->vdev_guid == guid) 4219 return (B_TRUE); 4220 4221 for (i = 0; i < sav->sav_npending; i++) { 4222 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 4223 &spareguid) == 0 && spareguid == guid) 4224 return (B_TRUE); 4225 } 4226 4227 return (B_FALSE); 4228} 4229 4230/* 4231 * Check if a pool has an active shared spare device. 4232 * Note: reference count of an active spare is 2, as a spare and as a replace 4233 */ 4234static boolean_t 4235spa_has_active_shared_spare(spa_t *spa) 4236{ 4237 int i, refcnt; 4238 uint64_t pool; 4239 spa_aux_vdev_t *sav = &spa->spa_spares; 4240 4241 for (i = 0; i < sav->sav_count; i++) { 4242 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 4243 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 4244 refcnt > 2) 4245 return (B_TRUE); 4246 } 4247 4248 return (B_FALSE); 4249} 4250 4251/* 4252 * Post a sysevent corresponding to the given event. The 'name' must be one of 4253 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 4254 * filled in from the spa and (optionally) the vdev. This doesn't do anything 4255 * in the userland libzpool, as we don't want consumers to misinterpret ztest 4256 * or zdb as real changes. 4257 */ 4258void 4259spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 4260{ 4261#if 0 4262#ifdef _KERNEL 4263 sysevent_t *ev; 4264 sysevent_attr_list_t *attr = NULL; 4265 sysevent_value_t value; 4266 sysevent_id_t eid; 4267 4268 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 4269 SE_SLEEP); 4270 4271 value.value_type = SE_DATA_TYPE_STRING; 4272 value.value.sv_string = spa_name(spa); 4273 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 4274 goto done; 4275 4276 value.value_type = SE_DATA_TYPE_UINT64; 4277 value.value.sv_uint64 = spa_guid(spa); 4278 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 4279 goto done; 4280 4281 if (vd) { 4282 value.value_type = SE_DATA_TYPE_UINT64; 4283 value.value.sv_uint64 = vd->vdev_guid; 4284 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 4285 SE_SLEEP) != 0) 4286 goto done; 4287 4288 if (vd->vdev_path) { 4289 value.value_type = SE_DATA_TYPE_STRING; 4290 value.value.sv_string = vd->vdev_path; 4291 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 4292 &value, SE_SLEEP) != 0) 4293 goto done; 4294 } 4295 } 4296 4297 if (sysevent_attach_attributes(ev, attr) != 0) 4298 goto done; 4299 attr = NULL; 4300 4301 (void) log_sysevent(ev, SE_SLEEP, &eid); 4302 4303done: 4304 if (attr) 4305 sysevent_free_attr(attr); 4306 sysevent_free(ev); 4307#endif 4308#endif 4309} 4310