spa.c revision 339105
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2013 Saso Kiselkov. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright 2016 Toomas Soome <tsoome@me.com> 31 * Copyright 2017 Joyent, Inc. 32 * Copyright (c) 2017 Datto Inc. 33 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association. 34 */ 35 36/* 37 * SPA: Storage Pool Allocator 38 * 39 * This file contains all the routines used when modifying on-disk SPA state. 40 * This includes opening, importing, destroying, exporting a pool, and syncing a 41 * pool. 42 */ 43 44#include <sys/zfs_context.h> 45#include <sys/fm/fs/zfs.h> 46#include <sys/spa_impl.h> 47#include <sys/zio.h> 48#include <sys/zio_checksum.h> 49#include <sys/dmu.h> 50#include <sys/dmu_tx.h> 51#include <sys/zap.h> 52#include <sys/zil.h> 53#include <sys/ddt.h> 54#include <sys/vdev_impl.h> 55#include <sys/vdev_removal.h> 56#include <sys/vdev_indirect_mapping.h> 57#include <sys/vdev_indirect_births.h> 58#include <sys/metaslab.h> 59#include <sys/metaslab_impl.h> 60#include <sys/uberblock_impl.h> 61#include <sys/txg.h> 62#include <sys/avl.h> 63#include <sys/bpobj.h> 64#include <sys/dmu_traverse.h> 65#include <sys/dmu_objset.h> 66#include <sys/unique.h> 67#include <sys/dsl_pool.h> 68#include <sys/dsl_dataset.h> 69#include <sys/dsl_dir.h> 70#include <sys/dsl_prop.h> 71#include <sys/dsl_synctask.h> 72#include <sys/fs/zfs.h> 73#include <sys/arc.h> 74#include <sys/callb.h> 75#include <sys/spa_boot.h> 76#include <sys/zfs_ioctl.h> 77#include <sys/dsl_scan.h> 78#include <sys/dmu_send.h> 79#include <sys/dsl_destroy.h> 80#include <sys/dsl_userhold.h> 81#include <sys/zfeature.h> 82#include <sys/zvol.h> 83#include <sys/trim_map.h> 84#include <sys/abd.h> 85 86#ifdef _KERNEL 87#include <sys/callb.h> 88#include <sys/cpupart.h> 89#include <sys/zone.h> 90#endif /* _KERNEL */ 91 92#include "zfs_prop.h" 93#include "zfs_comutil.h" 94 95/* Check hostid on import? */ 96static int check_hostid = 1; 97 98/* 99 * The interval, in seconds, at which failed configuration cache file writes 100 * should be retried. 101 */ 102int zfs_ccw_retry_interval = 300; 103 104SYSCTL_DECL(_vfs_zfs); 105SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 106 "Check hostid on import?"); 107TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 108SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 109 &zfs_ccw_retry_interval, 0, 110 "Configuration cache file write, retry after failure, interval (seconds)"); 111 112typedef enum zti_modes { 113 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 114 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 115 ZTI_MODE_NULL, /* don't create a taskq */ 116 ZTI_NMODES 117} zti_modes_t; 118 119#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 120#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 121#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 122 123#define ZTI_N(n) ZTI_P(n, 1) 124#define ZTI_ONE ZTI_N(1) 125 126typedef struct zio_taskq_info { 127 zti_modes_t zti_mode; 128 uint_t zti_value; 129 uint_t zti_count; 130} zio_taskq_info_t; 131 132static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 133 "issue", "issue_high", "intr", "intr_high" 134}; 135 136/* 137 * This table defines the taskq settings for each ZFS I/O type. When 138 * initializing a pool, we use this table to create an appropriately sized 139 * taskq. Some operations are low volume and therefore have a small, static 140 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 141 * macros. Other operations process a large amount of data; the ZTI_BATCH 142 * macro causes us to create a taskq oriented for throughput. Some operations 143 * are so high frequency and short-lived that the taskq itself can become a a 144 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 145 * additional degree of parallelism specified by the number of threads per- 146 * taskq and the number of taskqs; when dispatching an event in this case, the 147 * particular taskq is chosen at random. 148 * 149 * The different taskq priorities are to handle the different contexts (issue 150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 151 * need to be handled with minimum delay. 152 */ 153const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 156 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 157 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 158 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 159 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 161}; 162 163static void spa_sync_version(void *arg, dmu_tx_t *tx); 164static void spa_sync_props(void *arg, dmu_tx_t *tx); 165static boolean_t spa_has_active_shared_spare(spa_t *spa); 166static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport); 167static void spa_vdev_resilver_done(spa_t *spa); 168 169uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 170#ifdef PSRSET_BIND 171id_t zio_taskq_psrset_bind = PS_NONE; 172#endif 173#ifdef SYSDC 174boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 175uint_t zio_taskq_basedc = 80; /* base duty cycle */ 176#endif 177 178boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 179extern int zfs_sync_pass_deferred_free; 180 181/* 182 * Report any spa_load_verify errors found, but do not fail spa_load. 183 * This is used by zdb to analyze non-idle pools. 184 */ 185boolean_t spa_load_verify_dryrun = B_FALSE; 186 187/* 188 * This (illegal) pool name is used when temporarily importing a spa_t in order 189 * to get the vdev stats associated with the imported devices. 190 */ 191#define TRYIMPORT_NAME "$import" 192 193/* 194 * For debugging purposes: print out vdev tree during pool import. 195 */ 196int spa_load_print_vdev_tree = B_FALSE; 197 198/* 199 * A non-zero value for zfs_max_missing_tvds means that we allow importing 200 * pools with missing top-level vdevs. This is strictly intended for advanced 201 * pool recovery cases since missing data is almost inevitable. Pools with 202 * missing devices can only be imported read-only for safety reasons, and their 203 * fail-mode will be automatically set to "continue". 204 * 205 * With 1 missing vdev we should be able to import the pool and mount all 206 * datasets. User data that was not modified after the missing device has been 207 * added should be recoverable. This means that snapshots created prior to the 208 * addition of that device should be completely intact. 209 * 210 * With 2 missing vdevs, some datasets may fail to mount since there are 211 * dataset statistics that are stored as regular metadata. Some data might be 212 * recoverable if those vdevs were added recently. 213 * 214 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries 215 * may be missing entirely. Chances of data recovery are very low. Note that 216 * there are also risks of performing an inadvertent rewind as we might be 217 * missing all the vdevs with the latest uberblocks. 218 */ 219uint64_t zfs_max_missing_tvds = 0; 220 221/* 222 * The parameters below are similar to zfs_max_missing_tvds but are only 223 * intended for a preliminary open of the pool with an untrusted config which 224 * might be incomplete or out-dated. 225 * 226 * We are more tolerant for pools opened from a cachefile since we could have 227 * an out-dated cachefile where a device removal was not registered. 228 * We could have set the limit arbitrarily high but in the case where devices 229 * are really missing we would want to return the proper error codes; we chose 230 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available 231 * and we get a chance to retrieve the trusted config. 232 */ 233uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; 234 235/* 236 * In the case where config was assembled by scanning device paths (/dev/dsks 237 * by default) we are less tolerant since all the existing devices should have 238 * been detected and we want spa_load to return the right error codes. 239 */ 240uint64_t zfs_max_missing_tvds_scan = 0; 241 242 243SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN, 244 &spa_load_print_vdev_tree, 0, 245 "print out vdev tree during pool import"); 246SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN, 247 &zfs_max_missing_tvds, 0, 248 "allow importing pools with missing top-level vdevs"); 249SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN, 250 &zfs_max_missing_tvds_cachefile, 0, 251 "allow importing pools with missing top-level vdevs in cache file"); 252SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN, 253 &zfs_max_missing_tvds_scan, 0, 254 "allow importing pools with missing top-level vdevs during scan"); 255 256/* 257 * Debugging aid that pauses spa_sync() towards the end. 258 */ 259boolean_t zfs_pause_spa_sync = B_FALSE; 260 261/* 262 * ========================================================================== 263 * SPA properties routines 264 * ========================================================================== 265 */ 266 267/* 268 * Add a (source=src, propname=propval) list to an nvlist. 269 */ 270static void 271spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 272 uint64_t intval, zprop_source_t src) 273{ 274 const char *propname = zpool_prop_to_name(prop); 275 nvlist_t *propval; 276 277 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 278 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 279 280 if (strval != NULL) 281 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 282 else 283 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 284 285 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 286 nvlist_free(propval); 287} 288 289/* 290 * Get property values from the spa configuration. 291 */ 292static void 293spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 294{ 295 vdev_t *rvd = spa->spa_root_vdev; 296 dsl_pool_t *pool = spa->spa_dsl_pool; 297 uint64_t size, alloc, cap, version; 298 zprop_source_t src = ZPROP_SRC_NONE; 299 spa_config_dirent_t *dp; 300 metaslab_class_t *mc = spa_normal_class(spa); 301 302 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 303 304 if (rvd != NULL) { 305 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 306 size = metaslab_class_get_space(spa_normal_class(spa)); 307 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 308 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 309 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 310 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 311 size - alloc, src); 312 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, 313 spa->spa_checkpoint_info.sci_dspace, src); 314 315 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 316 metaslab_class_fragmentation(mc), src); 317 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 318 metaslab_class_expandable_space(mc), src); 319 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 320 (spa_mode(spa) == FREAD), src); 321 322 cap = (size == 0) ? 0 : (alloc * 100 / size); 323 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 324 325 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 326 ddt_get_pool_dedup_ratio(spa), src); 327 328 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 329 rvd->vdev_state, src); 330 331 version = spa_version(spa); 332 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 333 src = ZPROP_SRC_DEFAULT; 334 else 335 src = ZPROP_SRC_LOCAL; 336 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 337 } 338 339 if (pool != NULL) { 340 /* 341 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 342 * when opening pools before this version freedir will be NULL. 343 */ 344 if (pool->dp_free_dir != NULL) { 345 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 346 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 347 src); 348 } else { 349 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 350 NULL, 0, src); 351 } 352 353 if (pool->dp_leak_dir != NULL) { 354 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 355 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 356 src); 357 } else { 358 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 359 NULL, 0, src); 360 } 361 } 362 363 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 364 365 if (spa->spa_comment != NULL) { 366 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 367 0, ZPROP_SRC_LOCAL); 368 } 369 370 if (spa->spa_root != NULL) 371 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 372 0, ZPROP_SRC_LOCAL); 373 374 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 375 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 376 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 377 } else { 378 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 379 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 380 } 381 382 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 383 if (dp->scd_path == NULL) { 384 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 385 "none", 0, ZPROP_SRC_LOCAL); 386 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 387 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 388 dp->scd_path, 0, ZPROP_SRC_LOCAL); 389 } 390 } 391} 392 393/* 394 * Get zpool property values. 395 */ 396int 397spa_prop_get(spa_t *spa, nvlist_t **nvp) 398{ 399 objset_t *mos = spa->spa_meta_objset; 400 zap_cursor_t zc; 401 zap_attribute_t za; 402 int err; 403 404 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 405 406 mutex_enter(&spa->spa_props_lock); 407 408 /* 409 * Get properties from the spa config. 410 */ 411 spa_prop_get_config(spa, nvp); 412 413 /* If no pool property object, no more prop to get. */ 414 if (mos == NULL || spa->spa_pool_props_object == 0) { 415 mutex_exit(&spa->spa_props_lock); 416 return (0); 417 } 418 419 /* 420 * Get properties from the MOS pool property object. 421 */ 422 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 423 (err = zap_cursor_retrieve(&zc, &za)) == 0; 424 zap_cursor_advance(&zc)) { 425 uint64_t intval = 0; 426 char *strval = NULL; 427 zprop_source_t src = ZPROP_SRC_DEFAULT; 428 zpool_prop_t prop; 429 430 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) 431 continue; 432 433 switch (za.za_integer_length) { 434 case 8: 435 /* integer property */ 436 if (za.za_first_integer != 437 zpool_prop_default_numeric(prop)) 438 src = ZPROP_SRC_LOCAL; 439 440 if (prop == ZPOOL_PROP_BOOTFS) { 441 dsl_pool_t *dp; 442 dsl_dataset_t *ds = NULL; 443 444 dp = spa_get_dsl(spa); 445 dsl_pool_config_enter(dp, FTAG); 446 if (err = dsl_dataset_hold_obj(dp, 447 za.za_first_integer, FTAG, &ds)) { 448 dsl_pool_config_exit(dp, FTAG); 449 break; 450 } 451 452 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 453 KM_SLEEP); 454 dsl_dataset_name(ds, strval); 455 dsl_dataset_rele(ds, FTAG); 456 dsl_pool_config_exit(dp, FTAG); 457 } else { 458 strval = NULL; 459 intval = za.za_first_integer; 460 } 461 462 spa_prop_add_list(*nvp, prop, strval, intval, src); 463 464 if (strval != NULL) 465 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 466 467 break; 468 469 case 1: 470 /* string property */ 471 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 472 err = zap_lookup(mos, spa->spa_pool_props_object, 473 za.za_name, 1, za.za_num_integers, strval); 474 if (err) { 475 kmem_free(strval, za.za_num_integers); 476 break; 477 } 478 spa_prop_add_list(*nvp, prop, strval, 0, src); 479 kmem_free(strval, za.za_num_integers); 480 break; 481 482 default: 483 break; 484 } 485 } 486 zap_cursor_fini(&zc); 487 mutex_exit(&spa->spa_props_lock); 488out: 489 if (err && err != ENOENT) { 490 nvlist_free(*nvp); 491 *nvp = NULL; 492 return (err); 493 } 494 495 return (0); 496} 497 498/* 499 * Validate the given pool properties nvlist and modify the list 500 * for the property values to be set. 501 */ 502static int 503spa_prop_validate(spa_t *spa, nvlist_t *props) 504{ 505 nvpair_t *elem; 506 int error = 0, reset_bootfs = 0; 507 uint64_t objnum = 0; 508 boolean_t has_feature = B_FALSE; 509 510 elem = NULL; 511 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 512 uint64_t intval; 513 char *strval, *slash, *check, *fname; 514 const char *propname = nvpair_name(elem); 515 zpool_prop_t prop = zpool_name_to_prop(propname); 516 517 switch (prop) { 518 case ZPOOL_PROP_INVAL: 519 if (!zpool_prop_feature(propname)) { 520 error = SET_ERROR(EINVAL); 521 break; 522 } 523 524 /* 525 * Sanitize the input. 526 */ 527 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 528 error = SET_ERROR(EINVAL); 529 break; 530 } 531 532 if (nvpair_value_uint64(elem, &intval) != 0) { 533 error = SET_ERROR(EINVAL); 534 break; 535 } 536 537 if (intval != 0) { 538 error = SET_ERROR(EINVAL); 539 break; 540 } 541 542 fname = strchr(propname, '@') + 1; 543 if (zfeature_lookup_name(fname, NULL) != 0) { 544 error = SET_ERROR(EINVAL); 545 break; 546 } 547 548 has_feature = B_TRUE; 549 break; 550 551 case ZPOOL_PROP_VERSION: 552 error = nvpair_value_uint64(elem, &intval); 553 if (!error && 554 (intval < spa_version(spa) || 555 intval > SPA_VERSION_BEFORE_FEATURES || 556 has_feature)) 557 error = SET_ERROR(EINVAL); 558 break; 559 560 case ZPOOL_PROP_DELEGATION: 561 case ZPOOL_PROP_AUTOREPLACE: 562 case ZPOOL_PROP_LISTSNAPS: 563 case ZPOOL_PROP_AUTOEXPAND: 564 error = nvpair_value_uint64(elem, &intval); 565 if (!error && intval > 1) 566 error = SET_ERROR(EINVAL); 567 break; 568 569 case ZPOOL_PROP_BOOTFS: 570 /* 571 * If the pool version is less than SPA_VERSION_BOOTFS, 572 * or the pool is still being created (version == 0), 573 * the bootfs property cannot be set. 574 */ 575 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 576 error = SET_ERROR(ENOTSUP); 577 break; 578 } 579 580 /* 581 * Make sure the vdev config is bootable 582 */ 583 if (!vdev_is_bootable(spa->spa_root_vdev)) { 584 error = SET_ERROR(ENOTSUP); 585 break; 586 } 587 588 reset_bootfs = 1; 589 590 error = nvpair_value_string(elem, &strval); 591 592 if (!error) { 593 objset_t *os; 594 uint64_t propval; 595 596 if (strval == NULL || strval[0] == '\0') { 597 objnum = zpool_prop_default_numeric( 598 ZPOOL_PROP_BOOTFS); 599 break; 600 } 601 602 if (error = dmu_objset_hold(strval, FTAG, &os)) 603 break; 604 605 /* 606 * Must be ZPL, and its property settings 607 * must be supported by GRUB (compression 608 * is not gzip, and large blocks are not used). 609 */ 610 611 if (dmu_objset_type(os) != DMU_OST_ZFS) { 612 error = SET_ERROR(ENOTSUP); 613 } else if ((error = 614 dsl_prop_get_int_ds(dmu_objset_ds(os), 615 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 616 &propval)) == 0 && 617 !BOOTFS_COMPRESS_VALID(propval)) { 618 error = SET_ERROR(ENOTSUP); 619 } else { 620 objnum = dmu_objset_id(os); 621 } 622 dmu_objset_rele(os, FTAG); 623 } 624 break; 625 626 case ZPOOL_PROP_FAILUREMODE: 627 error = nvpair_value_uint64(elem, &intval); 628 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 629 intval > ZIO_FAILURE_MODE_PANIC)) 630 error = SET_ERROR(EINVAL); 631 632 /* 633 * This is a special case which only occurs when 634 * the pool has completely failed. This allows 635 * the user to change the in-core failmode property 636 * without syncing it out to disk (I/Os might 637 * currently be blocked). We do this by returning 638 * EIO to the caller (spa_prop_set) to trick it 639 * into thinking we encountered a property validation 640 * error. 641 */ 642 if (!error && spa_suspended(spa)) { 643 spa->spa_failmode = intval; 644 error = SET_ERROR(EIO); 645 } 646 break; 647 648 case ZPOOL_PROP_CACHEFILE: 649 if ((error = nvpair_value_string(elem, &strval)) != 0) 650 break; 651 652 if (strval[0] == '\0') 653 break; 654 655 if (strcmp(strval, "none") == 0) 656 break; 657 658 if (strval[0] != '/') { 659 error = SET_ERROR(EINVAL); 660 break; 661 } 662 663 slash = strrchr(strval, '/'); 664 ASSERT(slash != NULL); 665 666 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 667 strcmp(slash, "/..") == 0) 668 error = SET_ERROR(EINVAL); 669 break; 670 671 case ZPOOL_PROP_COMMENT: 672 if ((error = nvpair_value_string(elem, &strval)) != 0) 673 break; 674 for (check = strval; *check != '\0'; check++) { 675 /* 676 * The kernel doesn't have an easy isprint() 677 * check. For this kernel check, we merely 678 * check ASCII apart from DEL. Fix this if 679 * there is an easy-to-use kernel isprint(). 680 */ 681 if (*check >= 0x7f) { 682 error = SET_ERROR(EINVAL); 683 break; 684 } 685 } 686 if (strlen(strval) > ZPROP_MAX_COMMENT) 687 error = E2BIG; 688 break; 689 690 case ZPOOL_PROP_DEDUPDITTO: 691 if (spa_version(spa) < SPA_VERSION_DEDUP) 692 error = SET_ERROR(ENOTSUP); 693 else 694 error = nvpair_value_uint64(elem, &intval); 695 if (error == 0 && 696 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 697 error = SET_ERROR(EINVAL); 698 break; 699 } 700 701 if (error) 702 break; 703 } 704 705 if (!error && reset_bootfs) { 706 error = nvlist_remove(props, 707 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 708 709 if (!error) { 710 error = nvlist_add_uint64(props, 711 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 712 } 713 } 714 715 return (error); 716} 717 718void 719spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 720{ 721 char *cachefile; 722 spa_config_dirent_t *dp; 723 724 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 725 &cachefile) != 0) 726 return; 727 728 dp = kmem_alloc(sizeof (spa_config_dirent_t), 729 KM_SLEEP); 730 731 if (cachefile[0] == '\0') 732 dp->scd_path = spa_strdup(spa_config_path); 733 else if (strcmp(cachefile, "none") == 0) 734 dp->scd_path = NULL; 735 else 736 dp->scd_path = spa_strdup(cachefile); 737 738 list_insert_head(&spa->spa_config_list, dp); 739 if (need_sync) 740 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 741} 742 743int 744spa_prop_set(spa_t *spa, nvlist_t *nvp) 745{ 746 int error; 747 nvpair_t *elem = NULL; 748 boolean_t need_sync = B_FALSE; 749 750 if ((error = spa_prop_validate(spa, nvp)) != 0) 751 return (error); 752 753 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 754 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 755 756 if (prop == ZPOOL_PROP_CACHEFILE || 757 prop == ZPOOL_PROP_ALTROOT || 758 prop == ZPOOL_PROP_READONLY) 759 continue; 760 761 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { 762 uint64_t ver; 763 764 if (prop == ZPOOL_PROP_VERSION) { 765 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 766 } else { 767 ASSERT(zpool_prop_feature(nvpair_name(elem))); 768 ver = SPA_VERSION_FEATURES; 769 need_sync = B_TRUE; 770 } 771 772 /* Save time if the version is already set. */ 773 if (ver == spa_version(spa)) 774 continue; 775 776 /* 777 * In addition to the pool directory object, we might 778 * create the pool properties object, the features for 779 * read object, the features for write object, or the 780 * feature descriptions object. 781 */ 782 error = dsl_sync_task(spa->spa_name, NULL, 783 spa_sync_version, &ver, 784 6, ZFS_SPACE_CHECK_RESERVED); 785 if (error) 786 return (error); 787 continue; 788 } 789 790 need_sync = B_TRUE; 791 break; 792 } 793 794 if (need_sync) { 795 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 796 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 797 } 798 799 return (0); 800} 801 802/* 803 * If the bootfs property value is dsobj, clear it. 804 */ 805void 806spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 807{ 808 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 809 VERIFY(zap_remove(spa->spa_meta_objset, 810 spa->spa_pool_props_object, 811 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 812 spa->spa_bootfs = 0; 813 } 814} 815 816/*ARGSUSED*/ 817static int 818spa_change_guid_check(void *arg, dmu_tx_t *tx) 819{ 820 uint64_t *newguid = arg; 821 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 822 vdev_t *rvd = spa->spa_root_vdev; 823 uint64_t vdev_state; 824 825 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 826 int error = (spa_has_checkpoint(spa)) ? 827 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 828 return (SET_ERROR(error)); 829 } 830 831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 832 vdev_state = rvd->vdev_state; 833 spa_config_exit(spa, SCL_STATE, FTAG); 834 835 if (vdev_state != VDEV_STATE_HEALTHY) 836 return (SET_ERROR(ENXIO)); 837 838 ASSERT3U(spa_guid(spa), !=, *newguid); 839 840 return (0); 841} 842 843static void 844spa_change_guid_sync(void *arg, dmu_tx_t *tx) 845{ 846 uint64_t *newguid = arg; 847 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 848 uint64_t oldguid; 849 vdev_t *rvd = spa->spa_root_vdev; 850 851 oldguid = spa_guid(spa); 852 853 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 854 rvd->vdev_guid = *newguid; 855 rvd->vdev_guid_sum += (*newguid - oldguid); 856 vdev_config_dirty(rvd); 857 spa_config_exit(spa, SCL_STATE, FTAG); 858 859 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 860 oldguid, *newguid); 861} 862 863/* 864 * Change the GUID for the pool. This is done so that we can later 865 * re-import a pool built from a clone of our own vdevs. We will modify 866 * the root vdev's guid, our own pool guid, and then mark all of our 867 * vdevs dirty. Note that we must make sure that all our vdevs are 868 * online when we do this, or else any vdevs that weren't present 869 * would be orphaned from our pool. We are also going to issue a 870 * sysevent to update any watchers. 871 */ 872int 873spa_change_guid(spa_t *spa) 874{ 875 int error; 876 uint64_t guid; 877 878 mutex_enter(&spa->spa_vdev_top_lock); 879 mutex_enter(&spa_namespace_lock); 880 guid = spa_generate_guid(NULL); 881 882 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 883 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 884 885 if (error == 0) { 886 spa_write_cachefile(spa, B_FALSE, B_TRUE); 887 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); 888 } 889 890 mutex_exit(&spa_namespace_lock); 891 mutex_exit(&spa->spa_vdev_top_lock); 892 893 return (error); 894} 895 896/* 897 * ========================================================================== 898 * SPA state manipulation (open/create/destroy/import/export) 899 * ========================================================================== 900 */ 901 902static int 903spa_error_entry_compare(const void *a, const void *b) 904{ 905 spa_error_entry_t *sa = (spa_error_entry_t *)a; 906 spa_error_entry_t *sb = (spa_error_entry_t *)b; 907 int ret; 908 909 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 910 sizeof (zbookmark_phys_t)); 911 912 if (ret < 0) 913 return (-1); 914 else if (ret > 0) 915 return (1); 916 else 917 return (0); 918} 919 920/* 921 * Utility function which retrieves copies of the current logs and 922 * re-initializes them in the process. 923 */ 924void 925spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 926{ 927 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 928 929 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 930 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 931 932 avl_create(&spa->spa_errlist_scrub, 933 spa_error_entry_compare, sizeof (spa_error_entry_t), 934 offsetof(spa_error_entry_t, se_avl)); 935 avl_create(&spa->spa_errlist_last, 936 spa_error_entry_compare, sizeof (spa_error_entry_t), 937 offsetof(spa_error_entry_t, se_avl)); 938} 939 940static void 941spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 942{ 943 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 944 enum zti_modes mode = ztip->zti_mode; 945 uint_t value = ztip->zti_value; 946 uint_t count = ztip->zti_count; 947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 948 char name[32]; 949 uint_t flags = 0; 950 boolean_t batch = B_FALSE; 951 952 if (mode == ZTI_MODE_NULL) { 953 tqs->stqs_count = 0; 954 tqs->stqs_taskq = NULL; 955 return; 956 } 957 958 ASSERT3U(count, >, 0); 959 960 tqs->stqs_count = count; 961 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 962 963 switch (mode) { 964 case ZTI_MODE_FIXED: 965 ASSERT3U(value, >=, 1); 966 value = MAX(value, 1); 967 break; 968 969 case ZTI_MODE_BATCH: 970 batch = B_TRUE; 971 flags |= TASKQ_THREADS_CPU_PCT; 972 value = zio_taskq_batch_pct; 973 break; 974 975 default: 976 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 977 "spa_activate()", 978 zio_type_name[t], zio_taskq_types[q], mode, value); 979 break; 980 } 981 982 for (uint_t i = 0; i < count; i++) { 983 taskq_t *tq; 984 985 if (count > 1) { 986 (void) snprintf(name, sizeof (name), "%s_%s_%u", 987 zio_type_name[t], zio_taskq_types[q], i); 988 } else { 989 (void) snprintf(name, sizeof (name), "%s_%s", 990 zio_type_name[t], zio_taskq_types[q]); 991 } 992 993#ifdef SYSDC 994 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 995 if (batch) 996 flags |= TASKQ_DC_BATCH; 997 998 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 999 spa->spa_proc, zio_taskq_basedc, flags); 1000 } else { 1001#endif 1002 pri_t pri = maxclsyspri; 1003 /* 1004 * The write issue taskq can be extremely CPU 1005 * intensive. Run it at slightly lower priority 1006 * than the other taskqs. 1007 * FreeBSD notes: 1008 * - numerically higher priorities are lower priorities; 1009 * - if priorities divided by four (RQ_PPQ) are equal 1010 * then a difference between them is insignificant. 1011 */ 1012 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 1013#ifdef illumos 1014 pri--; 1015#else 1016 pri += 4; 1017#endif 1018 1019 tq = taskq_create_proc(name, value, pri, 50, 1020 INT_MAX, spa->spa_proc, flags); 1021#ifdef SYSDC 1022 } 1023#endif 1024 1025 tqs->stqs_taskq[i] = tq; 1026 } 1027} 1028 1029static void 1030spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 1031{ 1032 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1033 1034 if (tqs->stqs_taskq == NULL) { 1035 ASSERT0(tqs->stqs_count); 1036 return; 1037 } 1038 1039 for (uint_t i = 0; i < tqs->stqs_count; i++) { 1040 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 1041 taskq_destroy(tqs->stqs_taskq[i]); 1042 } 1043 1044 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 1045 tqs->stqs_taskq = NULL; 1046} 1047 1048/* 1049 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 1050 * Note that a type may have multiple discrete taskqs to avoid lock contention 1051 * on the taskq itself. In that case we choose which taskq at random by using 1052 * the low bits of gethrtime(). 1053 */ 1054void 1055spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 1056 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 1057{ 1058 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1059 taskq_t *tq; 1060 1061 ASSERT3P(tqs->stqs_taskq, !=, NULL); 1062 ASSERT3U(tqs->stqs_count, !=, 0); 1063 1064 if (tqs->stqs_count == 1) { 1065 tq = tqs->stqs_taskq[0]; 1066 } else { 1067#ifdef _KERNEL 1068 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 1069#else 1070 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 1071#endif 1072 } 1073 1074 taskq_dispatch_ent(tq, func, arg, flags, ent); 1075} 1076 1077static void 1078spa_create_zio_taskqs(spa_t *spa) 1079{ 1080 for (int t = 0; t < ZIO_TYPES; t++) { 1081 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1082 spa_taskqs_init(spa, t, q); 1083 } 1084 } 1085} 1086 1087#ifdef _KERNEL 1088#ifdef SPA_PROCESS 1089static void 1090spa_thread(void *arg) 1091{ 1092 callb_cpr_t cprinfo; 1093 1094 spa_t *spa = arg; 1095 user_t *pu = PTOU(curproc); 1096 1097 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1098 spa->spa_name); 1099 1100 ASSERT(curproc != &p0); 1101 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1102 "zpool-%s", spa->spa_name); 1103 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1104 1105#ifdef PSRSET_BIND 1106 /* bind this thread to the requested psrset */ 1107 if (zio_taskq_psrset_bind != PS_NONE) { 1108 pool_lock(); 1109 mutex_enter(&cpu_lock); 1110 mutex_enter(&pidlock); 1111 mutex_enter(&curproc->p_lock); 1112 1113 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1114 0, NULL, NULL) == 0) { 1115 curthread->t_bind_pset = zio_taskq_psrset_bind; 1116 } else { 1117 cmn_err(CE_WARN, 1118 "Couldn't bind process for zfs pool \"%s\" to " 1119 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1120 } 1121 1122 mutex_exit(&curproc->p_lock); 1123 mutex_exit(&pidlock); 1124 mutex_exit(&cpu_lock); 1125 pool_unlock(); 1126 } 1127#endif 1128 1129#ifdef SYSDC 1130 if (zio_taskq_sysdc) { 1131 sysdc_thread_enter(curthread, 100, 0); 1132 } 1133#endif 1134 1135 spa->spa_proc = curproc; 1136 spa->spa_did = curthread->t_did; 1137 1138 spa_create_zio_taskqs(spa); 1139 1140 mutex_enter(&spa->spa_proc_lock); 1141 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1142 1143 spa->spa_proc_state = SPA_PROC_ACTIVE; 1144 cv_broadcast(&spa->spa_proc_cv); 1145 1146 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1147 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1148 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1149 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1150 1151 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1152 spa->spa_proc_state = SPA_PROC_GONE; 1153 spa->spa_proc = &p0; 1154 cv_broadcast(&spa->spa_proc_cv); 1155 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1156 1157 mutex_enter(&curproc->p_lock); 1158 lwp_exit(); 1159} 1160#endif /* SPA_PROCESS */ 1161#endif 1162 1163/* 1164 * Activate an uninitialized pool. 1165 */ 1166static void 1167spa_activate(spa_t *spa, int mode) 1168{ 1169 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1170 1171 spa->spa_state = POOL_STATE_ACTIVE; 1172 spa->spa_mode = mode; 1173 1174 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1175 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1176 1177 /* Try to create a covering process */ 1178 mutex_enter(&spa->spa_proc_lock); 1179 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1180 ASSERT(spa->spa_proc == &p0); 1181 spa->spa_did = 0; 1182 1183#ifdef SPA_PROCESS 1184 /* Only create a process if we're going to be around a while. */ 1185 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1186 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1187 NULL, 0) == 0) { 1188 spa->spa_proc_state = SPA_PROC_CREATED; 1189 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1190 cv_wait(&spa->spa_proc_cv, 1191 &spa->spa_proc_lock); 1192 } 1193 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1194 ASSERT(spa->spa_proc != &p0); 1195 ASSERT(spa->spa_did != 0); 1196 } else { 1197#ifdef _KERNEL 1198 cmn_err(CE_WARN, 1199 "Couldn't create process for zfs pool \"%s\"\n", 1200 spa->spa_name); 1201#endif 1202 } 1203 } 1204#endif /* SPA_PROCESS */ 1205 mutex_exit(&spa->spa_proc_lock); 1206 1207 /* If we didn't create a process, we need to create our taskqs. */ 1208 ASSERT(spa->spa_proc == &p0); 1209 if (spa->spa_proc == &p0) { 1210 spa_create_zio_taskqs(spa); 1211 } 1212 1213 /* 1214 * Start TRIM thread. 1215 */ 1216 trim_thread_create(spa); 1217 1218 for (size_t i = 0; i < TXG_SIZE; i++) 1219 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0); 1220 1221 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1222 offsetof(vdev_t, vdev_config_dirty_node)); 1223 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1224 offsetof(objset_t, os_evicting_node)); 1225 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1226 offsetof(vdev_t, vdev_state_dirty_node)); 1227 1228 txg_list_create(&spa->spa_vdev_txg_list, spa, 1229 offsetof(struct vdev, vdev_txg_node)); 1230 1231 avl_create(&spa->spa_errlist_scrub, 1232 spa_error_entry_compare, sizeof (spa_error_entry_t), 1233 offsetof(spa_error_entry_t, se_avl)); 1234 avl_create(&spa->spa_errlist_last, 1235 spa_error_entry_compare, sizeof (spa_error_entry_t), 1236 offsetof(spa_error_entry_t, se_avl)); 1237} 1238 1239/* 1240 * Opposite of spa_activate(). 1241 */ 1242static void 1243spa_deactivate(spa_t *spa) 1244{ 1245 ASSERT(spa->spa_sync_on == B_FALSE); 1246 ASSERT(spa->spa_dsl_pool == NULL); 1247 ASSERT(spa->spa_root_vdev == NULL); 1248 ASSERT(spa->spa_async_zio_root == NULL); 1249 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1250 1251 /* 1252 * Stop TRIM thread in case spa_unload() wasn't called directly 1253 * before spa_deactivate(). 1254 */ 1255 trim_thread_destroy(spa); 1256 1257 spa_evicting_os_wait(spa); 1258 1259 txg_list_destroy(&spa->spa_vdev_txg_list); 1260 1261 list_destroy(&spa->spa_config_dirty_list); 1262 list_destroy(&spa->spa_evicting_os_list); 1263 list_destroy(&spa->spa_state_dirty_list); 1264 1265 for (int t = 0; t < ZIO_TYPES; t++) { 1266 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1267 spa_taskqs_fini(spa, t, q); 1268 } 1269 } 1270 1271 for (size_t i = 0; i < TXG_SIZE; i++) { 1272 ASSERT3P(spa->spa_txg_zio[i], !=, NULL); 1273 VERIFY0(zio_wait(spa->spa_txg_zio[i])); 1274 spa->spa_txg_zio[i] = NULL; 1275 } 1276 1277 metaslab_class_destroy(spa->spa_normal_class); 1278 spa->spa_normal_class = NULL; 1279 1280 metaslab_class_destroy(spa->spa_log_class); 1281 spa->spa_log_class = NULL; 1282 1283 /* 1284 * If this was part of an import or the open otherwise failed, we may 1285 * still have errors left in the queues. Empty them just in case. 1286 */ 1287 spa_errlog_drain(spa); 1288 1289 avl_destroy(&spa->spa_errlist_scrub); 1290 avl_destroy(&spa->spa_errlist_last); 1291 1292 spa->spa_state = POOL_STATE_UNINITIALIZED; 1293 1294 mutex_enter(&spa->spa_proc_lock); 1295 if (spa->spa_proc_state != SPA_PROC_NONE) { 1296 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1297 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1298 cv_broadcast(&spa->spa_proc_cv); 1299 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1300 ASSERT(spa->spa_proc != &p0); 1301 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1302 } 1303 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1304 spa->spa_proc_state = SPA_PROC_NONE; 1305 } 1306 ASSERT(spa->spa_proc == &p0); 1307 mutex_exit(&spa->spa_proc_lock); 1308 1309#ifdef SPA_PROCESS 1310 /* 1311 * We want to make sure spa_thread() has actually exited the ZFS 1312 * module, so that the module can't be unloaded out from underneath 1313 * it. 1314 */ 1315 if (spa->spa_did != 0) { 1316 thread_join(spa->spa_did); 1317 spa->spa_did = 0; 1318 } 1319#endif /* SPA_PROCESS */ 1320} 1321 1322/* 1323 * Verify a pool configuration, and construct the vdev tree appropriately. This 1324 * will create all the necessary vdevs in the appropriate layout, with each vdev 1325 * in the CLOSED state. This will prep the pool before open/creation/import. 1326 * All vdev validation is done by the vdev_alloc() routine. 1327 */ 1328static int 1329spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1330 uint_t id, int atype) 1331{ 1332 nvlist_t **child; 1333 uint_t children; 1334 int error; 1335 1336 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1337 return (error); 1338 1339 if ((*vdp)->vdev_ops->vdev_op_leaf) 1340 return (0); 1341 1342 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1343 &child, &children); 1344 1345 if (error == ENOENT) 1346 return (0); 1347 1348 if (error) { 1349 vdev_free(*vdp); 1350 *vdp = NULL; 1351 return (SET_ERROR(EINVAL)); 1352 } 1353 1354 for (int c = 0; c < children; c++) { 1355 vdev_t *vd; 1356 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1357 atype)) != 0) { 1358 vdev_free(*vdp); 1359 *vdp = NULL; 1360 return (error); 1361 } 1362 } 1363 1364 ASSERT(*vdp != NULL); 1365 1366 return (0); 1367} 1368 1369/* 1370 * Opposite of spa_load(). 1371 */ 1372static void 1373spa_unload(spa_t *spa) 1374{ 1375 int i; 1376 1377 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1378 1379 spa_load_note(spa, "UNLOADING"); 1380 1381 /* 1382 * Stop TRIM thread. 1383 */ 1384 trim_thread_destroy(spa); 1385 1386 /* 1387 * Stop async tasks. 1388 */ 1389 spa_async_suspend(spa); 1390 1391 /* 1392 * Stop syncing. 1393 */ 1394 if (spa->spa_sync_on) { 1395 txg_sync_stop(spa->spa_dsl_pool); 1396 spa->spa_sync_on = B_FALSE; 1397 } 1398 1399 /* 1400 * Even though vdev_free() also calls vdev_metaslab_fini, we need 1401 * to call it earlier, before we wait for async i/o to complete. 1402 * This ensures that there is no async metaslab prefetching, by 1403 * calling taskq_wait(mg_taskq). 1404 */ 1405 if (spa->spa_root_vdev != NULL) { 1406 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1407 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) 1408 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); 1409 spa_config_exit(spa, SCL_ALL, FTAG); 1410 } 1411 1412 /* 1413 * Wait for any outstanding async I/O to complete. 1414 */ 1415 if (spa->spa_async_zio_root != NULL) { 1416 for (int i = 0; i < max_ncpus; i++) 1417 (void) zio_wait(spa->spa_async_zio_root[i]); 1418 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1419 spa->spa_async_zio_root = NULL; 1420 } 1421 1422 if (spa->spa_vdev_removal != NULL) { 1423 spa_vdev_removal_destroy(spa->spa_vdev_removal); 1424 spa->spa_vdev_removal = NULL; 1425 } 1426 1427 if (spa->spa_condense_zthr != NULL) { 1428 ASSERT(!zthr_isrunning(spa->spa_condense_zthr)); 1429 zthr_destroy(spa->spa_condense_zthr); 1430 spa->spa_condense_zthr = NULL; 1431 } 1432 1433 if (spa->spa_checkpoint_discard_zthr != NULL) { 1434 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr)); 1435 zthr_destroy(spa->spa_checkpoint_discard_zthr); 1436 spa->spa_checkpoint_discard_zthr = NULL; 1437 } 1438 1439 spa_condense_fini(spa); 1440 1441 bpobj_close(&spa->spa_deferred_bpobj); 1442 1443 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1444 1445 /* 1446 * Close all vdevs. 1447 */ 1448 if (spa->spa_root_vdev) 1449 vdev_free(spa->spa_root_vdev); 1450 ASSERT(spa->spa_root_vdev == NULL); 1451 1452 /* 1453 * Close the dsl pool. 1454 */ 1455 if (spa->spa_dsl_pool) { 1456 dsl_pool_close(spa->spa_dsl_pool); 1457 spa->spa_dsl_pool = NULL; 1458 spa->spa_meta_objset = NULL; 1459 } 1460 1461 ddt_unload(spa); 1462 1463 /* 1464 * Drop and purge level 2 cache 1465 */ 1466 spa_l2cache_drop(spa); 1467 1468 for (i = 0; i < spa->spa_spares.sav_count; i++) 1469 vdev_free(spa->spa_spares.sav_vdevs[i]); 1470 if (spa->spa_spares.sav_vdevs) { 1471 kmem_free(spa->spa_spares.sav_vdevs, 1472 spa->spa_spares.sav_count * sizeof (void *)); 1473 spa->spa_spares.sav_vdevs = NULL; 1474 } 1475 if (spa->spa_spares.sav_config) { 1476 nvlist_free(spa->spa_spares.sav_config); 1477 spa->spa_spares.sav_config = NULL; 1478 } 1479 spa->spa_spares.sav_count = 0; 1480 1481 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1482 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1483 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1484 } 1485 if (spa->spa_l2cache.sav_vdevs) { 1486 kmem_free(spa->spa_l2cache.sav_vdevs, 1487 spa->spa_l2cache.sav_count * sizeof (void *)); 1488 spa->spa_l2cache.sav_vdevs = NULL; 1489 } 1490 if (spa->spa_l2cache.sav_config) { 1491 nvlist_free(spa->spa_l2cache.sav_config); 1492 spa->spa_l2cache.sav_config = NULL; 1493 } 1494 spa->spa_l2cache.sav_count = 0; 1495 1496 spa->spa_async_suspended = 0; 1497 1498 spa->spa_indirect_vdevs_loaded = B_FALSE; 1499 1500 if (spa->spa_comment != NULL) { 1501 spa_strfree(spa->spa_comment); 1502 spa->spa_comment = NULL; 1503 } 1504 1505 spa_config_exit(spa, SCL_ALL, FTAG); 1506} 1507 1508/* 1509 * Load (or re-load) the current list of vdevs describing the active spares for 1510 * this pool. When this is called, we have some form of basic information in 1511 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1512 * then re-generate a more complete list including status information. 1513 */ 1514void 1515spa_load_spares(spa_t *spa) 1516{ 1517 nvlist_t **spares; 1518 uint_t nspares; 1519 int i; 1520 vdev_t *vd, *tvd; 1521 1522#ifndef _KERNEL 1523 /* 1524 * zdb opens both the current state of the pool and the 1525 * checkpointed state (if present), with a different spa_t. 1526 * 1527 * As spare vdevs are shared among open pools, we skip loading 1528 * them when we load the checkpointed state of the pool. 1529 */ 1530 if (!spa_writeable(spa)) 1531 return; 1532#endif 1533 1534 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1535 1536 /* 1537 * First, close and free any existing spare vdevs. 1538 */ 1539 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1540 vd = spa->spa_spares.sav_vdevs[i]; 1541 1542 /* Undo the call to spa_activate() below */ 1543 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1544 B_FALSE)) != NULL && tvd->vdev_isspare) 1545 spa_spare_remove(tvd); 1546 vdev_close(vd); 1547 vdev_free(vd); 1548 } 1549 1550 if (spa->spa_spares.sav_vdevs) 1551 kmem_free(spa->spa_spares.sav_vdevs, 1552 spa->spa_spares.sav_count * sizeof (void *)); 1553 1554 if (spa->spa_spares.sav_config == NULL) 1555 nspares = 0; 1556 else 1557 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1558 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1559 1560 spa->spa_spares.sav_count = (int)nspares; 1561 spa->spa_spares.sav_vdevs = NULL; 1562 1563 if (nspares == 0) 1564 return; 1565 1566 /* 1567 * Construct the array of vdevs, opening them to get status in the 1568 * process. For each spare, there is potentially two different vdev_t 1569 * structures associated with it: one in the list of spares (used only 1570 * for basic validation purposes) and one in the active vdev 1571 * configuration (if it's spared in). During this phase we open and 1572 * validate each vdev on the spare list. If the vdev also exists in the 1573 * active configuration, then we also mark this vdev as an active spare. 1574 */ 1575 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1576 KM_SLEEP); 1577 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1578 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1579 VDEV_ALLOC_SPARE) == 0); 1580 ASSERT(vd != NULL); 1581 1582 spa->spa_spares.sav_vdevs[i] = vd; 1583 1584 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1585 B_FALSE)) != NULL) { 1586 if (!tvd->vdev_isspare) 1587 spa_spare_add(tvd); 1588 1589 /* 1590 * We only mark the spare active if we were successfully 1591 * able to load the vdev. Otherwise, importing a pool 1592 * with a bad active spare would result in strange 1593 * behavior, because multiple pool would think the spare 1594 * is actively in use. 1595 * 1596 * There is a vulnerability here to an equally bizarre 1597 * circumstance, where a dead active spare is later 1598 * brought back to life (onlined or otherwise). Given 1599 * the rarity of this scenario, and the extra complexity 1600 * it adds, we ignore the possibility. 1601 */ 1602 if (!vdev_is_dead(tvd)) 1603 spa_spare_activate(tvd); 1604 } 1605 1606 vd->vdev_top = vd; 1607 vd->vdev_aux = &spa->spa_spares; 1608 1609 if (vdev_open(vd) != 0) 1610 continue; 1611 1612 if (vdev_validate_aux(vd) == 0) 1613 spa_spare_add(vd); 1614 } 1615 1616 /* 1617 * Recompute the stashed list of spares, with status information 1618 * this time. 1619 */ 1620 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1621 DATA_TYPE_NVLIST_ARRAY) == 0); 1622 1623 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1624 KM_SLEEP); 1625 for (i = 0; i < spa->spa_spares.sav_count; i++) 1626 spares[i] = vdev_config_generate(spa, 1627 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1628 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1629 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1630 for (i = 0; i < spa->spa_spares.sav_count; i++) 1631 nvlist_free(spares[i]); 1632 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1633} 1634 1635/* 1636 * Load (or re-load) the current list of vdevs describing the active l2cache for 1637 * this pool. When this is called, we have some form of basic information in 1638 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1639 * then re-generate a more complete list including status information. 1640 * Devices which are already active have their details maintained, and are 1641 * not re-opened. 1642 */ 1643void 1644spa_load_l2cache(spa_t *spa) 1645{ 1646 nvlist_t **l2cache; 1647 uint_t nl2cache; 1648 int i, j, oldnvdevs; 1649 uint64_t guid; 1650 vdev_t *vd, **oldvdevs, **newvdevs; 1651 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1652 1653#ifndef _KERNEL 1654 /* 1655 * zdb opens both the current state of the pool and the 1656 * checkpointed state (if present), with a different spa_t. 1657 * 1658 * As L2 caches are part of the ARC which is shared among open 1659 * pools, we skip loading them when we load the checkpointed 1660 * state of the pool. 1661 */ 1662 if (!spa_writeable(spa)) 1663 return; 1664#endif 1665 1666 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1667 1668 if (sav->sav_config != NULL) { 1669 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1670 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1671 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1672 } else { 1673 nl2cache = 0; 1674 newvdevs = NULL; 1675 } 1676 1677 oldvdevs = sav->sav_vdevs; 1678 oldnvdevs = sav->sav_count; 1679 sav->sav_vdevs = NULL; 1680 sav->sav_count = 0; 1681 1682 /* 1683 * Process new nvlist of vdevs. 1684 */ 1685 for (i = 0; i < nl2cache; i++) { 1686 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1687 &guid) == 0); 1688 1689 newvdevs[i] = NULL; 1690 for (j = 0; j < oldnvdevs; j++) { 1691 vd = oldvdevs[j]; 1692 if (vd != NULL && guid == vd->vdev_guid) { 1693 /* 1694 * Retain previous vdev for add/remove ops. 1695 */ 1696 newvdevs[i] = vd; 1697 oldvdevs[j] = NULL; 1698 break; 1699 } 1700 } 1701 1702 if (newvdevs[i] == NULL) { 1703 /* 1704 * Create new vdev 1705 */ 1706 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1707 VDEV_ALLOC_L2CACHE) == 0); 1708 ASSERT(vd != NULL); 1709 newvdevs[i] = vd; 1710 1711 /* 1712 * Commit this vdev as an l2cache device, 1713 * even if it fails to open. 1714 */ 1715 spa_l2cache_add(vd); 1716 1717 vd->vdev_top = vd; 1718 vd->vdev_aux = sav; 1719 1720 spa_l2cache_activate(vd); 1721 1722 if (vdev_open(vd) != 0) 1723 continue; 1724 1725 (void) vdev_validate_aux(vd); 1726 1727 if (!vdev_is_dead(vd)) 1728 l2arc_add_vdev(spa, vd); 1729 } 1730 } 1731 1732 /* 1733 * Purge vdevs that were dropped 1734 */ 1735 for (i = 0; i < oldnvdevs; i++) { 1736 uint64_t pool; 1737 1738 vd = oldvdevs[i]; 1739 if (vd != NULL) { 1740 ASSERT(vd->vdev_isl2cache); 1741 1742 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1743 pool != 0ULL && l2arc_vdev_present(vd)) 1744 l2arc_remove_vdev(vd); 1745 vdev_clear_stats(vd); 1746 vdev_free(vd); 1747 } 1748 } 1749 1750 if (oldvdevs) 1751 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1752 1753 if (sav->sav_config == NULL) 1754 goto out; 1755 1756 sav->sav_vdevs = newvdevs; 1757 sav->sav_count = (int)nl2cache; 1758 1759 /* 1760 * Recompute the stashed list of l2cache devices, with status 1761 * information this time. 1762 */ 1763 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1764 DATA_TYPE_NVLIST_ARRAY) == 0); 1765 1766 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1767 for (i = 0; i < sav->sav_count; i++) 1768 l2cache[i] = vdev_config_generate(spa, 1769 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1770 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1771 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1772out: 1773 for (i = 0; i < sav->sav_count; i++) 1774 nvlist_free(l2cache[i]); 1775 if (sav->sav_count) 1776 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1777} 1778 1779static int 1780load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1781{ 1782 dmu_buf_t *db; 1783 char *packed = NULL; 1784 size_t nvsize = 0; 1785 int error; 1786 *value = NULL; 1787 1788 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1789 if (error != 0) 1790 return (error); 1791 1792 nvsize = *(uint64_t *)db->db_data; 1793 dmu_buf_rele(db, FTAG); 1794 1795 packed = kmem_alloc(nvsize, KM_SLEEP); 1796 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1797 DMU_READ_PREFETCH); 1798 if (error == 0) 1799 error = nvlist_unpack(packed, nvsize, value, 0); 1800 kmem_free(packed, nvsize); 1801 1802 return (error); 1803} 1804 1805/* 1806 * Concrete top-level vdevs that are not missing and are not logs. At every 1807 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. 1808 */ 1809static uint64_t 1810spa_healthy_core_tvds(spa_t *spa) 1811{ 1812 vdev_t *rvd = spa->spa_root_vdev; 1813 uint64_t tvds = 0; 1814 1815 for (uint64_t i = 0; i < rvd->vdev_children; i++) { 1816 vdev_t *vd = rvd->vdev_child[i]; 1817 if (vd->vdev_islog) 1818 continue; 1819 if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) 1820 tvds++; 1821 } 1822 1823 return (tvds); 1824} 1825 1826/* 1827 * Checks to see if the given vdev could not be opened, in which case we post a 1828 * sysevent to notify the autoreplace code that the device has been removed. 1829 */ 1830static void 1831spa_check_removed(vdev_t *vd) 1832{ 1833 for (uint64_t c = 0; c < vd->vdev_children; c++) 1834 spa_check_removed(vd->vdev_child[c]); 1835 1836 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1837 vdev_is_concrete(vd)) { 1838 zfs_post_autoreplace(vd->vdev_spa, vd); 1839 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); 1840 } 1841} 1842 1843static int 1844spa_check_for_missing_logs(spa_t *spa) 1845{ 1846 vdev_t *rvd = spa->spa_root_vdev; 1847 1848 /* 1849 * If we're doing a normal import, then build up any additional 1850 * diagnostic information about missing log devices. 1851 * We'll pass this up to the user for further processing. 1852 */ 1853 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1854 nvlist_t **child, *nv; 1855 uint64_t idx = 0; 1856 1857 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1858 KM_SLEEP); 1859 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1860 1861 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1862 vdev_t *tvd = rvd->vdev_child[c]; 1863 1864 /* 1865 * We consider a device as missing only if it failed 1866 * to open (i.e. offline or faulted is not considered 1867 * as missing). 1868 */ 1869 if (tvd->vdev_islog && 1870 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1871 child[idx++] = vdev_config_generate(spa, tvd, 1872 B_FALSE, VDEV_CONFIG_MISSING); 1873 } 1874 } 1875 1876 if (idx > 0) { 1877 fnvlist_add_nvlist_array(nv, 1878 ZPOOL_CONFIG_CHILDREN, child, idx); 1879 fnvlist_add_nvlist(spa->spa_load_info, 1880 ZPOOL_CONFIG_MISSING_DEVICES, nv); 1881 1882 for (uint64_t i = 0; i < idx; i++) 1883 nvlist_free(child[i]); 1884 } 1885 nvlist_free(nv); 1886 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1887 1888 if (idx > 0) { 1889 spa_load_failed(spa, "some log devices are missing"); 1890 vdev_dbgmsg_print_tree(rvd, 2); 1891 return (SET_ERROR(ENXIO)); 1892 } 1893 } else { 1894 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1895 vdev_t *tvd = rvd->vdev_child[c]; 1896 1897 if (tvd->vdev_islog && 1898 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1899 spa_set_log_state(spa, SPA_LOG_CLEAR); 1900 spa_load_note(spa, "some log devices are " 1901 "missing, ZIL is dropped."); 1902 vdev_dbgmsg_print_tree(rvd, 2); 1903 break; 1904 } 1905 } 1906 } 1907 1908 return (0); 1909} 1910 1911/* 1912 * Check for missing log devices 1913 */ 1914static boolean_t 1915spa_check_logs(spa_t *spa) 1916{ 1917 boolean_t rv = B_FALSE; 1918 dsl_pool_t *dp = spa_get_dsl(spa); 1919 1920 switch (spa->spa_log_state) { 1921 case SPA_LOG_MISSING: 1922 /* need to recheck in case slog has been restored */ 1923 case SPA_LOG_UNKNOWN: 1924 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1925 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1926 if (rv) 1927 spa_set_log_state(spa, SPA_LOG_MISSING); 1928 break; 1929 } 1930 return (rv); 1931} 1932 1933static boolean_t 1934spa_passivate_log(spa_t *spa) 1935{ 1936 vdev_t *rvd = spa->spa_root_vdev; 1937 boolean_t slog_found = B_FALSE; 1938 1939 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1940 1941 if (!spa_has_slogs(spa)) 1942 return (B_FALSE); 1943 1944 for (int c = 0; c < rvd->vdev_children; c++) { 1945 vdev_t *tvd = rvd->vdev_child[c]; 1946 metaslab_group_t *mg = tvd->vdev_mg; 1947 1948 if (tvd->vdev_islog) { 1949 metaslab_group_passivate(mg); 1950 slog_found = B_TRUE; 1951 } 1952 } 1953 1954 return (slog_found); 1955} 1956 1957static void 1958spa_activate_log(spa_t *spa) 1959{ 1960 vdev_t *rvd = spa->spa_root_vdev; 1961 1962 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1963 1964 for (int c = 0; c < rvd->vdev_children; c++) { 1965 vdev_t *tvd = rvd->vdev_child[c]; 1966 metaslab_group_t *mg = tvd->vdev_mg; 1967 1968 if (tvd->vdev_islog) 1969 metaslab_group_activate(mg); 1970 } 1971} 1972 1973int 1974spa_reset_logs(spa_t *spa) 1975{ 1976 int error; 1977 1978 error = dmu_objset_find(spa_name(spa), zil_reset, 1979 NULL, DS_FIND_CHILDREN); 1980 if (error == 0) { 1981 /* 1982 * We successfully offlined the log device, sync out the 1983 * current txg so that the "stubby" block can be removed 1984 * by zil_sync(). 1985 */ 1986 txg_wait_synced(spa->spa_dsl_pool, 0); 1987 } 1988 return (error); 1989} 1990 1991static void 1992spa_aux_check_removed(spa_aux_vdev_t *sav) 1993{ 1994 int i; 1995 1996 for (i = 0; i < sav->sav_count; i++) 1997 spa_check_removed(sav->sav_vdevs[i]); 1998} 1999 2000void 2001spa_claim_notify(zio_t *zio) 2002{ 2003 spa_t *spa = zio->io_spa; 2004 2005 if (zio->io_error) 2006 return; 2007 2008 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 2009 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 2010 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 2011 mutex_exit(&spa->spa_props_lock); 2012} 2013 2014typedef struct spa_load_error { 2015 uint64_t sle_meta_count; 2016 uint64_t sle_data_count; 2017} spa_load_error_t; 2018 2019static void 2020spa_load_verify_done(zio_t *zio) 2021{ 2022 blkptr_t *bp = zio->io_bp; 2023 spa_load_error_t *sle = zio->io_private; 2024 dmu_object_type_t type = BP_GET_TYPE(bp); 2025 int error = zio->io_error; 2026 spa_t *spa = zio->io_spa; 2027 2028 abd_free(zio->io_abd); 2029 if (error) { 2030 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 2031 type != DMU_OT_INTENT_LOG) 2032 atomic_inc_64(&sle->sle_meta_count); 2033 else 2034 atomic_inc_64(&sle->sle_data_count); 2035 } 2036 2037 mutex_enter(&spa->spa_scrub_lock); 2038 spa->spa_load_verify_ios--; 2039 cv_broadcast(&spa->spa_scrub_io_cv); 2040 mutex_exit(&spa->spa_scrub_lock); 2041} 2042 2043/* 2044 * Maximum number of concurrent scrub i/os to create while verifying 2045 * a pool while importing it. 2046 */ 2047int spa_load_verify_maxinflight = 10000; 2048boolean_t spa_load_verify_metadata = B_TRUE; 2049boolean_t spa_load_verify_data = B_TRUE; 2050 2051SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 2052 &spa_load_verify_maxinflight, 0, 2053 "Maximum number of concurrent scrub I/Os to create while verifying a " 2054 "pool while importing it"); 2055 2056SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 2057 &spa_load_verify_metadata, 0, 2058 "Check metadata on import?"); 2059 2060SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 2061 &spa_load_verify_data, 0, 2062 "Check user data on import?"); 2063 2064/*ARGSUSED*/ 2065static int 2066spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 2067 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 2068{ 2069 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 2070 return (0); 2071 /* 2072 * Note: normally this routine will not be called if 2073 * spa_load_verify_metadata is not set. However, it may be useful 2074 * to manually set the flag after the traversal has begun. 2075 */ 2076 if (!spa_load_verify_metadata) 2077 return (0); 2078 if (!BP_IS_METADATA(bp) && !spa_load_verify_data) 2079 return (0); 2080 2081 zio_t *rio = arg; 2082 size_t size = BP_GET_PSIZE(bp); 2083 2084 mutex_enter(&spa->spa_scrub_lock); 2085 while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight) 2086 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 2087 spa->spa_load_verify_ios++; 2088 mutex_exit(&spa->spa_scrub_lock); 2089 2090 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, 2091 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 2092 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 2093 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 2094 return (0); 2095} 2096 2097/* ARGSUSED */ 2098int 2099verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 2100{ 2101 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 2102 return (SET_ERROR(ENAMETOOLONG)); 2103 2104 return (0); 2105} 2106 2107static int 2108spa_load_verify(spa_t *spa) 2109{ 2110 zio_t *rio; 2111 spa_load_error_t sle = { 0 }; 2112 zpool_load_policy_t policy; 2113 boolean_t verify_ok = B_FALSE; 2114 int error = 0; 2115 2116 zpool_get_load_policy(spa->spa_config, &policy); 2117 2118 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND) 2119 return (0); 2120 2121 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2122 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2123 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2124 DS_FIND_CHILDREN); 2125 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2126 if (error != 0) 2127 return (error); 2128 2129 rio = zio_root(spa, NULL, &sle, 2130 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2131 2132 if (spa_load_verify_metadata) { 2133 if (spa->spa_extreme_rewind) { 2134 spa_load_note(spa, "performing a complete scan of the " 2135 "pool since extreme rewind is on. This may take " 2136 "a very long time.\n (spa_load_verify_data=%u, " 2137 "spa_load_verify_metadata=%u)", 2138 spa_load_verify_data, spa_load_verify_metadata); 2139 } 2140 error = traverse_pool(spa, spa->spa_verify_min_txg, 2141 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2142 spa_load_verify_cb, rio); 2143 } 2144 2145 (void) zio_wait(rio); 2146 2147 spa->spa_load_meta_errors = sle.sle_meta_count; 2148 spa->spa_load_data_errors = sle.sle_data_count; 2149 2150 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { 2151 spa_load_note(spa, "spa_load_verify found %llu metadata errors " 2152 "and %llu data errors", (u_longlong_t)sle.sle_meta_count, 2153 (u_longlong_t)sle.sle_data_count); 2154 } 2155 2156 if (spa_load_verify_dryrun || 2157 (!error && sle.sle_meta_count <= policy.zlp_maxmeta && 2158 sle.sle_data_count <= policy.zlp_maxdata)) { 2159 int64_t loss = 0; 2160 2161 verify_ok = B_TRUE; 2162 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2163 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2164 2165 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2166 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2167 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2168 VERIFY(nvlist_add_int64(spa->spa_load_info, 2169 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2170 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2171 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2172 } else { 2173 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2174 } 2175 2176 if (spa_load_verify_dryrun) 2177 return (0); 2178 2179 if (error) { 2180 if (error != ENXIO && error != EIO) 2181 error = SET_ERROR(EIO); 2182 return (error); 2183 } 2184 2185 return (verify_ok ? 0 : EIO); 2186} 2187 2188/* 2189 * Find a value in the pool props object. 2190 */ 2191static void 2192spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2193{ 2194 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2195 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2196} 2197 2198/* 2199 * Find a value in the pool directory object. 2200 */ 2201static int 2202spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) 2203{ 2204 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2205 name, sizeof (uint64_t), 1, val); 2206 2207 if (error != 0 && (error != ENOENT || log_enoent)) { 2208 spa_load_failed(spa, "couldn't get '%s' value in MOS directory " 2209 "[error=%d]", name, error); 2210 } 2211 2212 return (error); 2213} 2214 2215static int 2216spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2217{ 2218 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2219 return (SET_ERROR(err)); 2220} 2221 2222static void 2223spa_spawn_aux_threads(spa_t *spa) 2224{ 2225 ASSERT(spa_writeable(spa)); 2226 2227 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2228 2229 spa_start_indirect_condensing_thread(spa); 2230 2231 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); 2232 spa->spa_checkpoint_discard_zthr = 2233 zthr_create(spa_checkpoint_discard_thread_check, 2234 spa_checkpoint_discard_thread, spa); 2235} 2236 2237/* 2238 * Fix up config after a partly-completed split. This is done with the 2239 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2240 * pool have that entry in their config, but only the splitting one contains 2241 * a list of all the guids of the vdevs that are being split off. 2242 * 2243 * This function determines what to do with that list: either rejoin 2244 * all the disks to the pool, or complete the splitting process. To attempt 2245 * the rejoin, each disk that is offlined is marked online again, and 2246 * we do a reopen() call. If the vdev label for every disk that was 2247 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2248 * then we call vdev_split() on each disk, and complete the split. 2249 * 2250 * Otherwise we leave the config alone, with all the vdevs in place in 2251 * the original pool. 2252 */ 2253static void 2254spa_try_repair(spa_t *spa, nvlist_t *config) 2255{ 2256 uint_t extracted; 2257 uint64_t *glist; 2258 uint_t i, gcount; 2259 nvlist_t *nvl; 2260 vdev_t **vd; 2261 boolean_t attempt_reopen; 2262 2263 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2264 return; 2265 2266 /* check that the config is complete */ 2267 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2268 &glist, &gcount) != 0) 2269 return; 2270 2271 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2272 2273 /* attempt to online all the vdevs & validate */ 2274 attempt_reopen = B_TRUE; 2275 for (i = 0; i < gcount; i++) { 2276 if (glist[i] == 0) /* vdev is hole */ 2277 continue; 2278 2279 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2280 if (vd[i] == NULL) { 2281 /* 2282 * Don't bother attempting to reopen the disks; 2283 * just do the split. 2284 */ 2285 attempt_reopen = B_FALSE; 2286 } else { 2287 /* attempt to re-online it */ 2288 vd[i]->vdev_offline = B_FALSE; 2289 } 2290 } 2291 2292 if (attempt_reopen) { 2293 vdev_reopen(spa->spa_root_vdev); 2294 2295 /* check each device to see what state it's in */ 2296 for (extracted = 0, i = 0; i < gcount; i++) { 2297 if (vd[i] != NULL && 2298 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2299 break; 2300 ++extracted; 2301 } 2302 } 2303 2304 /* 2305 * If every disk has been moved to the new pool, or if we never 2306 * even attempted to look at them, then we split them off for 2307 * good. 2308 */ 2309 if (!attempt_reopen || gcount == extracted) { 2310 for (i = 0; i < gcount; i++) 2311 if (vd[i] != NULL) 2312 vdev_split(vd[i]); 2313 vdev_reopen(spa->spa_root_vdev); 2314 } 2315 2316 kmem_free(vd, gcount * sizeof (vdev_t *)); 2317} 2318 2319static int 2320spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) 2321{ 2322 char *ereport = FM_EREPORT_ZFS_POOL; 2323 int error; 2324 2325 spa->spa_load_state = state; 2326 2327 gethrestime(&spa->spa_loaded_ts); 2328 error = spa_load_impl(spa, type, &ereport); 2329 2330 /* 2331 * Don't count references from objsets that are already closed 2332 * and are making their way through the eviction process. 2333 */ 2334 spa_evicting_os_wait(spa); 2335 spa->spa_minref = refcount_count(&spa->spa_refcount); 2336 if (error) { 2337 if (error != EEXIST) { 2338 spa->spa_loaded_ts.tv_sec = 0; 2339 spa->spa_loaded_ts.tv_nsec = 0; 2340 } 2341 if (error != EBADF) { 2342 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2343 } 2344 } 2345 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2346 spa->spa_ena = 0; 2347 2348 return (error); 2349} 2350 2351/* 2352 * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2353 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2354 * spa's per-vdev ZAP list. 2355 */ 2356static uint64_t 2357vdev_count_verify_zaps(vdev_t *vd) 2358{ 2359 spa_t *spa = vd->vdev_spa; 2360 uint64_t total = 0; 2361 if (vd->vdev_top_zap != 0) { 2362 total++; 2363 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2364 spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2365 } 2366 if (vd->vdev_leaf_zap != 0) { 2367 total++; 2368 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2369 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2370 } 2371 2372 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2373 total += vdev_count_verify_zaps(vd->vdev_child[i]); 2374 } 2375 2376 return (total); 2377} 2378 2379static int 2380spa_verify_host(spa_t *spa, nvlist_t *mos_config) 2381{ 2382 uint64_t hostid; 2383 char *hostname; 2384 uint64_t myhostid = 0; 2385 2386 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, 2387 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2388 hostname = fnvlist_lookup_string(mos_config, 2389 ZPOOL_CONFIG_HOSTNAME); 2390 2391 myhostid = zone_get_hostid(NULL); 2392 2393 if (hostid != 0 && myhostid != 0 && hostid != myhostid) { 2394 cmn_err(CE_WARN, "pool '%s' could not be " 2395 "loaded as it was last accessed by " 2396 "another system (host: %s hostid: 0x%llx). " 2397 "See: http://illumos.org/msg/ZFS-8000-EY", 2398 spa_name(spa), hostname, (u_longlong_t)hostid); 2399 spa_load_failed(spa, "hostid verification failed: pool " 2400 "last accessed by host: %s (hostid: 0x%llx)", 2401 hostname, (u_longlong_t)hostid); 2402 return (SET_ERROR(EBADF)); 2403 } 2404 } 2405 2406 return (0); 2407} 2408 2409static int 2410spa_ld_parse_config(spa_t *spa, spa_import_type_t type) 2411{ 2412 int error = 0; 2413 nvlist_t *nvtree, *nvl, *config = spa->spa_config; 2414 int parse; 2415 vdev_t *rvd; 2416 uint64_t pool_guid; 2417 char *comment; 2418 2419 /* 2420 * Versioning wasn't explicitly added to the label until later, so if 2421 * it's not present treat it as the initial version. 2422 */ 2423 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2424 &spa->spa_ubsync.ub_version) != 0) 2425 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2426 2427 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { 2428 spa_load_failed(spa, "invalid config provided: '%s' missing", 2429 ZPOOL_CONFIG_POOL_GUID); 2430 return (SET_ERROR(EINVAL)); 2431 } 2432 2433 /* 2434 * If we are doing an import, ensure that the pool is not already 2435 * imported by checking if its pool guid already exists in the 2436 * spa namespace. 2437 * 2438 * The only case that we allow an already imported pool to be 2439 * imported again, is when the pool is checkpointed and we want to 2440 * look at its checkpointed state from userland tools like zdb. 2441 */ 2442#ifdef _KERNEL 2443 if ((spa->spa_load_state == SPA_LOAD_IMPORT || 2444 spa->spa_load_state == SPA_LOAD_TRYIMPORT) && 2445 spa_guid_exists(pool_guid, 0)) { 2446#else 2447 if ((spa->spa_load_state == SPA_LOAD_IMPORT || 2448 spa->spa_load_state == SPA_LOAD_TRYIMPORT) && 2449 spa_guid_exists(pool_guid, 0) && 2450 !spa_importing_readonly_checkpoint(spa)) { 2451#endif 2452 spa_load_failed(spa, "a pool with guid %llu is already open", 2453 (u_longlong_t)pool_guid); 2454 return (SET_ERROR(EEXIST)); 2455 } 2456 2457 spa->spa_config_guid = pool_guid; 2458 2459 nvlist_free(spa->spa_load_info); 2460 spa->spa_load_info = fnvlist_alloc(); 2461 2462 ASSERT(spa->spa_comment == NULL); 2463 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2464 spa->spa_comment = spa_strdup(comment); 2465 2466 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2467 &spa->spa_config_txg); 2468 2469 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) 2470 spa->spa_config_splitting = fnvlist_dup(nvl); 2471 2472 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { 2473 spa_load_failed(spa, "invalid config provided: '%s' missing", 2474 ZPOOL_CONFIG_VDEV_TREE); 2475 return (SET_ERROR(EINVAL)); 2476 } 2477 2478 /* 2479 * Create "The Godfather" zio to hold all async IOs 2480 */ 2481 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2482 KM_SLEEP); 2483 for (int i = 0; i < max_ncpus; i++) { 2484 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2485 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2486 ZIO_FLAG_GODFATHER); 2487 } 2488 2489 /* 2490 * Parse the configuration into a vdev tree. We explicitly set the 2491 * value that will be returned by spa_version() since parsing the 2492 * configuration requires knowing the version number. 2493 */ 2494 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2495 parse = (type == SPA_IMPORT_EXISTING ? 2496 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2497 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); 2498 spa_config_exit(spa, SCL_ALL, FTAG); 2499 2500 if (error != 0) { 2501 spa_load_failed(spa, "unable to parse config [error=%d]", 2502 error); 2503 return (error); 2504 } 2505 2506 ASSERT(spa->spa_root_vdev == rvd); 2507 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2508 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2509 2510 if (type != SPA_IMPORT_ASSEMBLE) { 2511 ASSERT(spa_guid(spa) == pool_guid); 2512 } 2513 2514 return (0); 2515} 2516 2517/* 2518 * Recursively open all vdevs in the vdev tree. This function is called twice: 2519 * first with the untrusted config, then with the trusted config. 2520 */ 2521static int 2522spa_ld_open_vdevs(spa_t *spa) 2523{ 2524 int error = 0; 2525 2526 /* 2527 * spa_missing_tvds_allowed defines how many top-level vdevs can be 2528 * missing/unopenable for the root vdev to be still considered openable. 2529 */ 2530 if (spa->spa_trust_config) { 2531 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; 2532 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { 2533 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; 2534 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { 2535 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; 2536 } else { 2537 spa->spa_missing_tvds_allowed = 0; 2538 } 2539 2540 spa->spa_missing_tvds_allowed = 2541 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); 2542 2543 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2544 error = vdev_open(spa->spa_root_vdev); 2545 spa_config_exit(spa, SCL_ALL, FTAG); 2546 2547 if (spa->spa_missing_tvds != 0) { 2548 spa_load_note(spa, "vdev tree has %lld missing top-level " 2549 "vdevs.", (u_longlong_t)spa->spa_missing_tvds); 2550 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) { 2551 /* 2552 * Although theoretically we could allow users to open 2553 * incomplete pools in RW mode, we'd need to add a lot 2554 * of extra logic (e.g. adjust pool space to account 2555 * for missing vdevs). 2556 * This limitation also prevents users from accidentally 2557 * opening the pool in RW mode during data recovery and 2558 * damaging it further. 2559 */ 2560 spa_load_note(spa, "pools with missing top-level " 2561 "vdevs can only be opened in read-only mode."); 2562 error = SET_ERROR(ENXIO); 2563 } else { 2564 spa_load_note(spa, "current settings allow for maximum " 2565 "%lld missing top-level vdevs at this stage.", 2566 (u_longlong_t)spa->spa_missing_tvds_allowed); 2567 } 2568 } 2569 if (error != 0) { 2570 spa_load_failed(spa, "unable to open vdev tree [error=%d]", 2571 error); 2572 } 2573 if (spa->spa_missing_tvds != 0 || error != 0) 2574 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); 2575 2576 return (error); 2577} 2578 2579/* 2580 * We need to validate the vdev labels against the configuration that 2581 * we have in hand. This function is called twice: first with an untrusted 2582 * config, then with a trusted config. The validation is more strict when the 2583 * config is trusted. 2584 */ 2585static int 2586spa_ld_validate_vdevs(spa_t *spa) 2587{ 2588 int error = 0; 2589 vdev_t *rvd = spa->spa_root_vdev; 2590 2591 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2592 error = vdev_validate(rvd); 2593 spa_config_exit(spa, SCL_ALL, FTAG); 2594 2595 if (error != 0) { 2596 spa_load_failed(spa, "vdev_validate failed [error=%d]", error); 2597 return (error); 2598 } 2599 2600 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 2601 spa_load_failed(spa, "cannot open vdev tree after invalidating " 2602 "some vdevs"); 2603 vdev_dbgmsg_print_tree(rvd, 2); 2604 return (SET_ERROR(ENXIO)); 2605 } 2606 2607 return (0); 2608} 2609 2610static void 2611spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) 2612{ 2613 spa->spa_state = POOL_STATE_ACTIVE; 2614 spa->spa_ubsync = spa->spa_uberblock; 2615 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2616 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2617 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2618 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2619 spa->spa_claim_max_txg = spa->spa_first_txg; 2620 spa->spa_prev_software_version = ub->ub_software_version; 2621} 2622 2623static int 2624spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) 2625{ 2626 vdev_t *rvd = spa->spa_root_vdev; 2627 nvlist_t *label; 2628 uberblock_t *ub = &spa->spa_uberblock; 2629 2630 /* 2631 * If we are opening the checkpointed state of the pool by 2632 * rewinding to it, at this point we will have written the 2633 * checkpointed uberblock to the vdev labels, so searching 2634 * the labels will find the right uberblock. However, if 2635 * we are opening the checkpointed state read-only, we have 2636 * not modified the labels. Therefore, we must ignore the 2637 * labels and continue using the spa_uberblock that was set 2638 * by spa_ld_checkpoint_rewind. 2639 * 2640 * Note that it would be fine to ignore the labels when 2641 * rewinding (opening writeable) as well. However, if we 2642 * crash just after writing the labels, we will end up 2643 * searching the labels. Doing so in the common case means 2644 * that this code path gets exercised normally, rather than 2645 * just in the edge case. 2646 */ 2647 if (ub->ub_checkpoint_txg != 0 && 2648 spa_importing_readonly_checkpoint(spa)) { 2649 spa_ld_select_uberblock_done(spa, ub); 2650 return (0); 2651 } 2652 2653 /* 2654 * Find the best uberblock. 2655 */ 2656 vdev_uberblock_load(rvd, ub, &label); 2657 2658 /* 2659 * If we weren't able to find a single valid uberblock, return failure. 2660 */ 2661 if (ub->ub_txg == 0) { 2662 nvlist_free(label); 2663 spa_load_failed(spa, "no valid uberblock found"); 2664 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2665 } 2666 2667 spa_load_note(spa, "using uberblock with txg=%llu", 2668 (u_longlong_t)ub->ub_txg); 2669 2670 /* 2671 * If the pool has an unsupported version we can't open it. 2672 */ 2673 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2674 nvlist_free(label); 2675 spa_load_failed(spa, "version %llu is not supported", 2676 (u_longlong_t)ub->ub_version); 2677 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2678 } 2679 2680 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2681 nvlist_t *features; 2682 2683 /* 2684 * If we weren't able to find what's necessary for reading the 2685 * MOS in the label, return failure. 2686 */ 2687 if (label == NULL) { 2688 spa_load_failed(spa, "label config unavailable"); 2689 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2690 ENXIO)); 2691 } 2692 2693 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, 2694 &features) != 0) { 2695 nvlist_free(label); 2696 spa_load_failed(spa, "invalid label: '%s' missing", 2697 ZPOOL_CONFIG_FEATURES_FOR_READ); 2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2699 ENXIO)); 2700 } 2701 2702 /* 2703 * Update our in-core representation with the definitive values 2704 * from the label. 2705 */ 2706 nvlist_free(spa->spa_label_features); 2707 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2708 } 2709 2710 nvlist_free(label); 2711 2712 /* 2713 * Look through entries in the label nvlist's features_for_read. If 2714 * there is a feature listed there which we don't understand then we 2715 * cannot open a pool. 2716 */ 2717 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2718 nvlist_t *unsup_feat; 2719 2720 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2721 0); 2722 2723 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2724 NULL); nvp != NULL; 2725 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2726 if (!zfeature_is_supported(nvpair_name(nvp))) { 2727 VERIFY(nvlist_add_string(unsup_feat, 2728 nvpair_name(nvp), "") == 0); 2729 } 2730 } 2731 2732 if (!nvlist_empty(unsup_feat)) { 2733 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2734 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2735 nvlist_free(unsup_feat); 2736 spa_load_failed(spa, "some features are unsupported"); 2737 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2738 ENOTSUP)); 2739 } 2740 2741 nvlist_free(unsup_feat); 2742 } 2743 2744 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2745 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2746 spa_try_repair(spa, spa->spa_config); 2747 spa_config_exit(spa, SCL_ALL, FTAG); 2748 nvlist_free(spa->spa_config_splitting); 2749 spa->spa_config_splitting = NULL; 2750 } 2751 2752 /* 2753 * Initialize internal SPA structures. 2754 */ 2755 spa_ld_select_uberblock_done(spa, ub); 2756 2757 return (0); 2758} 2759 2760static int 2761spa_ld_open_rootbp(spa_t *spa) 2762{ 2763 int error = 0; 2764 vdev_t *rvd = spa->spa_root_vdev; 2765 2766 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2767 if (error != 0) { 2768 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " 2769 "[error=%d]", error); 2770 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2771 } 2772 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2773 2774 return (0); 2775} 2776 2777static int 2778spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, 2779 boolean_t reloading) 2780{ 2781 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 2782 nvlist_t *nv, *mos_config, *policy; 2783 int error = 0, copy_error; 2784 uint64_t healthy_tvds, healthy_tvds_mos; 2785 uint64_t mos_config_txg; 2786 2787 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) 2788 != 0) 2789 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2790 2791 /* 2792 * If we're assembling a pool from a split, the config provided is 2793 * already trusted so there is nothing to do. 2794 */ 2795 if (type == SPA_IMPORT_ASSEMBLE) 2796 return (0); 2797 2798 healthy_tvds = spa_healthy_core_tvds(spa); 2799 2800 if (load_nvlist(spa, spa->spa_config_object, &mos_config) 2801 != 0) { 2802 spa_load_failed(spa, "unable to retrieve MOS config"); 2803 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2804 } 2805 2806 /* 2807 * If we are doing an open, pool owner wasn't verified yet, thus do 2808 * the verification here. 2809 */ 2810 if (spa->spa_load_state == SPA_LOAD_OPEN) { 2811 error = spa_verify_host(spa, mos_config); 2812 if (error != 0) { 2813 nvlist_free(mos_config); 2814 return (error); 2815 } 2816 } 2817 2818 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); 2819 2820 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2821 2822 /* 2823 * Build a new vdev tree from the trusted config 2824 */ 2825 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 2826 2827 /* 2828 * Vdev paths in the MOS may be obsolete. If the untrusted config was 2829 * obtained by scanning /dev/dsk, then it will have the right vdev 2830 * paths. We update the trusted MOS config with this information. 2831 * We first try to copy the paths with vdev_copy_path_strict, which 2832 * succeeds only when both configs have exactly the same vdev tree. 2833 * If that fails, we fall back to a more flexible method that has a 2834 * best effort policy. 2835 */ 2836 copy_error = vdev_copy_path_strict(rvd, mrvd); 2837 if (copy_error != 0 || spa_load_print_vdev_tree) { 2838 spa_load_note(spa, "provided vdev tree:"); 2839 vdev_dbgmsg_print_tree(rvd, 2); 2840 spa_load_note(spa, "MOS vdev tree:"); 2841 vdev_dbgmsg_print_tree(mrvd, 2); 2842 } 2843 if (copy_error != 0) { 2844 spa_load_note(spa, "vdev_copy_path_strict failed, falling " 2845 "back to vdev_copy_path_relaxed"); 2846 vdev_copy_path_relaxed(rvd, mrvd); 2847 } 2848 2849 vdev_close(rvd); 2850 vdev_free(rvd); 2851 spa->spa_root_vdev = mrvd; 2852 rvd = mrvd; 2853 spa_config_exit(spa, SCL_ALL, FTAG); 2854 2855 /* 2856 * We will use spa_config if we decide to reload the spa or if spa_load 2857 * fails and we rewind. We must thus regenerate the config using the 2858 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to 2859 * pass settings on how to load the pool and is not stored in the MOS. 2860 * We copy it over to our new, trusted config. 2861 */ 2862 mos_config_txg = fnvlist_lookup_uint64(mos_config, 2863 ZPOOL_CONFIG_POOL_TXG); 2864 nvlist_free(mos_config); 2865 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); 2866 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, 2867 &policy) == 0) 2868 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); 2869 spa_config_set(spa, mos_config); 2870 spa->spa_config_source = SPA_CONFIG_SRC_MOS; 2871 2872 /* 2873 * Now that we got the config from the MOS, we should be more strict 2874 * in checking blkptrs and can make assumptions about the consistency 2875 * of the vdev tree. spa_trust_config must be set to true before opening 2876 * vdevs in order for them to be writeable. 2877 */ 2878 spa->spa_trust_config = B_TRUE; 2879 2880 /* 2881 * Open and validate the new vdev tree 2882 */ 2883 error = spa_ld_open_vdevs(spa); 2884 if (error != 0) 2885 return (error); 2886 2887 error = spa_ld_validate_vdevs(spa); 2888 if (error != 0) 2889 return (error); 2890 2891 if (copy_error != 0 || spa_load_print_vdev_tree) { 2892 spa_load_note(spa, "final vdev tree:"); 2893 vdev_dbgmsg_print_tree(rvd, 2); 2894 } 2895 2896 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && 2897 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { 2898 /* 2899 * Sanity check to make sure that we are indeed loading the 2900 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds 2901 * in the config provided and they happened to be the only ones 2902 * to have the latest uberblock, we could involuntarily perform 2903 * an extreme rewind. 2904 */ 2905 healthy_tvds_mos = spa_healthy_core_tvds(spa); 2906 if (healthy_tvds_mos - healthy_tvds >= 2907 SPA_SYNC_MIN_VDEVS) { 2908 spa_load_note(spa, "config provided misses too many " 2909 "top-level vdevs compared to MOS (%lld vs %lld). ", 2910 (u_longlong_t)healthy_tvds, 2911 (u_longlong_t)healthy_tvds_mos); 2912 spa_load_note(spa, "vdev tree:"); 2913 vdev_dbgmsg_print_tree(rvd, 2); 2914 if (reloading) { 2915 spa_load_failed(spa, "config was already " 2916 "provided from MOS. Aborting."); 2917 return (spa_vdev_err(rvd, 2918 VDEV_AUX_CORRUPT_DATA, EIO)); 2919 } 2920 spa_load_note(spa, "spa must be reloaded using MOS " 2921 "config"); 2922 return (SET_ERROR(EAGAIN)); 2923 } 2924 } 2925 2926 error = spa_check_for_missing_logs(spa); 2927 if (error != 0) 2928 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2929 2930 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { 2931 spa_load_failed(spa, "uberblock guid sum doesn't match MOS " 2932 "guid sum (%llu != %llu)", 2933 (u_longlong_t)spa->spa_uberblock.ub_guid_sum, 2934 (u_longlong_t)rvd->vdev_guid_sum); 2935 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2936 ENXIO)); 2937 } 2938 2939 return (0); 2940} 2941 2942static int 2943spa_ld_open_indirect_vdev_metadata(spa_t *spa) 2944{ 2945 int error = 0; 2946 vdev_t *rvd = spa->spa_root_vdev; 2947 2948 /* 2949 * Everything that we read before spa_remove_init() must be stored 2950 * on concreted vdevs. Therefore we do this as early as possible. 2951 */ 2952 error = spa_remove_init(spa); 2953 if (error != 0) { 2954 spa_load_failed(spa, "spa_remove_init failed [error=%d]", 2955 error); 2956 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2957 } 2958 2959 /* 2960 * Retrieve information needed to condense indirect vdev mappings. 2961 */ 2962 error = spa_condense_init(spa); 2963 if (error != 0) { 2964 spa_load_failed(spa, "spa_condense_init failed [error=%d]", 2965 error); 2966 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 2967 } 2968 2969 return (0); 2970} 2971 2972static int 2973spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) 2974{ 2975 int error = 0; 2976 vdev_t *rvd = spa->spa_root_vdev; 2977 2978 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2979 boolean_t missing_feat_read = B_FALSE; 2980 nvlist_t *unsup_feat, *enabled_feat; 2981 2982 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2983 &spa->spa_feat_for_read_obj, B_TRUE) != 0) { 2984 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2985 } 2986 2987 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2988 &spa->spa_feat_for_write_obj, B_TRUE) != 0) { 2989 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2990 } 2991 2992 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2993 &spa->spa_feat_desc_obj, B_TRUE) != 0) { 2994 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2995 } 2996 2997 enabled_feat = fnvlist_alloc(); 2998 unsup_feat = fnvlist_alloc(); 2999 3000 if (!spa_features_check(spa, B_FALSE, 3001 unsup_feat, enabled_feat)) 3002 missing_feat_read = B_TRUE; 3003 3004 if (spa_writeable(spa) || 3005 spa->spa_load_state == SPA_LOAD_TRYIMPORT) { 3006 if (!spa_features_check(spa, B_TRUE, 3007 unsup_feat, enabled_feat)) { 3008 *missing_feat_writep = B_TRUE; 3009 } 3010 } 3011 3012 fnvlist_add_nvlist(spa->spa_load_info, 3013 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 3014 3015 if (!nvlist_empty(unsup_feat)) { 3016 fnvlist_add_nvlist(spa->spa_load_info, 3017 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 3018 } 3019 3020 fnvlist_free(enabled_feat); 3021 fnvlist_free(unsup_feat); 3022 3023 if (!missing_feat_read) { 3024 fnvlist_add_boolean(spa->spa_load_info, 3025 ZPOOL_CONFIG_CAN_RDONLY); 3026 } 3027 3028 /* 3029 * If the state is SPA_LOAD_TRYIMPORT, our objective is 3030 * twofold: to determine whether the pool is available for 3031 * import in read-write mode and (if it is not) whether the 3032 * pool is available for import in read-only mode. If the pool 3033 * is available for import in read-write mode, it is displayed 3034 * as available in userland; if it is not available for import 3035 * in read-only mode, it is displayed as unavailable in 3036 * userland. If the pool is available for import in read-only 3037 * mode but not read-write mode, it is displayed as unavailable 3038 * in userland with a special note that the pool is actually 3039 * available for open in read-only mode. 3040 * 3041 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 3042 * missing a feature for write, we must first determine whether 3043 * the pool can be opened read-only before returning to 3044 * userland in order to know whether to display the 3045 * abovementioned note. 3046 */ 3047 if (missing_feat_read || (*missing_feat_writep && 3048 spa_writeable(spa))) { 3049 spa_load_failed(spa, "pool uses unsupported features"); 3050 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 3051 ENOTSUP)); 3052 } 3053 3054 /* 3055 * Load refcounts for ZFS features from disk into an in-memory 3056 * cache during SPA initialization. 3057 */ 3058 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 3059 uint64_t refcount; 3060 3061 error = feature_get_refcount_from_disk(spa, 3062 &spa_feature_table[i], &refcount); 3063 if (error == 0) { 3064 spa->spa_feat_refcount_cache[i] = refcount; 3065 } else if (error == ENOTSUP) { 3066 spa->spa_feat_refcount_cache[i] = 3067 SPA_FEATURE_DISABLED; 3068 } else { 3069 spa_load_failed(spa, "error getting refcount " 3070 "for feature %s [error=%d]", 3071 spa_feature_table[i].fi_guid, error); 3072 return (spa_vdev_err(rvd, 3073 VDEV_AUX_CORRUPT_DATA, EIO)); 3074 } 3075 } 3076 } 3077 3078 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 3079 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 3080 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) 3081 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3082 } 3083 3084 return (0); 3085} 3086 3087static int 3088spa_ld_load_special_directories(spa_t *spa) 3089{ 3090 int error = 0; 3091 vdev_t *rvd = spa->spa_root_vdev; 3092 3093 spa->spa_is_initializing = B_TRUE; 3094 error = dsl_pool_open(spa->spa_dsl_pool); 3095 spa->spa_is_initializing = B_FALSE; 3096 if (error != 0) { 3097 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); 3098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3099 } 3100 3101 return (0); 3102} 3103 3104static int 3105spa_ld_get_props(spa_t *spa) 3106{ 3107 int error = 0; 3108 uint64_t obj; 3109 vdev_t *rvd = spa->spa_root_vdev; 3110 3111 /* Grab the secret checksum salt from the MOS. */ 3112 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3113 DMU_POOL_CHECKSUM_SALT, 1, 3114 sizeof (spa->spa_cksum_salt.zcs_bytes), 3115 spa->spa_cksum_salt.zcs_bytes); 3116 if (error == ENOENT) { 3117 /* Generate a new salt for subsequent use */ 3118 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3119 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3120 } else if (error != 0) { 3121 spa_load_failed(spa, "unable to retrieve checksum salt from " 3122 "MOS [error=%d]", error); 3123 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3124 } 3125 3126 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) 3127 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3128 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 3129 if (error != 0) { 3130 spa_load_failed(spa, "error opening deferred-frees bpobj " 3131 "[error=%d]", error); 3132 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3133 } 3134 3135 /* 3136 * Load the bit that tells us to use the new accounting function 3137 * (raid-z deflation). If we have an older pool, this will not 3138 * be present. 3139 */ 3140 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); 3141 if (error != 0 && error != ENOENT) 3142 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3143 3144 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 3145 &spa->spa_creation_version, B_FALSE); 3146 if (error != 0 && error != ENOENT) 3147 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3148 3149 /* 3150 * Load the persistent error log. If we have an older pool, this will 3151 * not be present. 3152 */ 3153 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, 3154 B_FALSE); 3155 if (error != 0 && error != ENOENT) 3156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3157 3158 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 3159 &spa->spa_errlog_scrub, B_FALSE); 3160 if (error != 0 && error != ENOENT) 3161 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3162 3163 /* 3164 * Load the history object. If we have an older pool, this 3165 * will not be present. 3166 */ 3167 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); 3168 if (error != 0 && error != ENOENT) 3169 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3170 3171 /* 3172 * Load the per-vdev ZAP map. If we have an older pool, this will not 3173 * be present; in this case, defer its creation to a later time to 3174 * avoid dirtying the MOS this early / out of sync context. See 3175 * spa_sync_config_object. 3176 */ 3177 3178 /* The sentinel is only available in the MOS config. */ 3179 nvlist_t *mos_config; 3180 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { 3181 spa_load_failed(spa, "unable to retrieve MOS config"); 3182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3183 } 3184 3185 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 3186 &spa->spa_all_vdev_zaps, B_FALSE); 3187 3188 if (error == ENOENT) { 3189 VERIFY(!nvlist_exists(mos_config, 3190 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 3191 spa->spa_avz_action = AVZ_ACTION_INITIALIZE; 3192 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3193 } else if (error != 0) { 3194 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3195 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 3196 /* 3197 * An older version of ZFS overwrote the sentinel value, so 3198 * we have orphaned per-vdev ZAPs in the MOS. Defer their 3199 * destruction to later; see spa_sync_config_object. 3200 */ 3201 spa->spa_avz_action = AVZ_ACTION_DESTROY; 3202 /* 3203 * We're assuming that no vdevs have had their ZAPs created 3204 * before this. Better be sure of it. 3205 */ 3206 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3207 } 3208 nvlist_free(mos_config); 3209 3210 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3211 3212 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, 3213 B_FALSE); 3214 if (error && error != ENOENT) 3215 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3216 3217 if (error == 0) { 3218 uint64_t autoreplace; 3219 3220 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 3221 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 3222 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 3223 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 3224 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 3225 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 3226 &spa->spa_dedup_ditto); 3227 3228 spa->spa_autoreplace = (autoreplace != 0); 3229 } 3230 3231 /* 3232 * If we are importing a pool with missing top-level vdevs, 3233 * we enforce that the pool doesn't panic or get suspended on 3234 * error since the likelihood of missing data is extremely high. 3235 */ 3236 if (spa->spa_missing_tvds > 0 && 3237 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && 3238 spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3239 spa_load_note(spa, "forcing failmode to 'continue' " 3240 "as some top level vdevs are missing"); 3241 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; 3242 } 3243 3244 return (0); 3245} 3246 3247static int 3248spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) 3249{ 3250 int error = 0; 3251 vdev_t *rvd = spa->spa_root_vdev; 3252 3253 /* 3254 * If we're assembling the pool from the split-off vdevs of 3255 * an existing pool, we don't want to attach the spares & cache 3256 * devices. 3257 */ 3258 3259 /* 3260 * Load any hot spares for this pool. 3261 */ 3262 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, 3263 B_FALSE); 3264 if (error != 0 && error != ENOENT) 3265 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3266 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3267 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 3268 if (load_nvlist(spa, spa->spa_spares.sav_object, 3269 &spa->spa_spares.sav_config) != 0) { 3270 spa_load_failed(spa, "error loading spares nvlist"); 3271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3272 } 3273 3274 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3275 spa_load_spares(spa); 3276 spa_config_exit(spa, SCL_ALL, FTAG); 3277 } else if (error == 0) { 3278 spa->spa_spares.sav_sync = B_TRUE; 3279 } 3280 3281 /* 3282 * Load any level 2 ARC devices for this pool. 3283 */ 3284 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 3285 &spa->spa_l2cache.sav_object, B_FALSE); 3286 if (error != 0 && error != ENOENT) 3287 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3288 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3289 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 3290 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 3291 &spa->spa_l2cache.sav_config) != 0) { 3292 spa_load_failed(spa, "error loading l2cache nvlist"); 3293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3294 } 3295 3296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3297 spa_load_l2cache(spa); 3298 spa_config_exit(spa, SCL_ALL, FTAG); 3299 } else if (error == 0) { 3300 spa->spa_l2cache.sav_sync = B_TRUE; 3301 } 3302 3303 return (0); 3304} 3305 3306static int 3307spa_ld_load_vdev_metadata(spa_t *spa) 3308{ 3309 int error = 0; 3310 vdev_t *rvd = spa->spa_root_vdev; 3311 3312 /* 3313 * If the 'autoreplace' property is set, then post a resource notifying 3314 * the ZFS DE that it should not issue any faults for unopenable 3315 * devices. We also iterate over the vdevs, and post a sysevent for any 3316 * unopenable vdevs so that the normal autoreplace handler can take 3317 * over. 3318 */ 3319 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3320 spa_check_removed(spa->spa_root_vdev); 3321 /* 3322 * For the import case, this is done in spa_import(), because 3323 * at this point we're using the spare definitions from 3324 * the MOS config, not necessarily from the userland config. 3325 */ 3326 if (spa->spa_load_state != SPA_LOAD_IMPORT) { 3327 spa_aux_check_removed(&spa->spa_spares); 3328 spa_aux_check_removed(&spa->spa_l2cache); 3329 } 3330 } 3331 3332 /* 3333 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. 3334 */ 3335 error = vdev_load(rvd); 3336 if (error != 0) { 3337 spa_load_failed(spa, "vdev_load failed [error=%d]", error); 3338 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 3339 } 3340 3341 /* 3342 * Propagate the leaf DTLs we just loaded all the way up the vdev tree. 3343 */ 3344 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3345 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 3346 spa_config_exit(spa, SCL_ALL, FTAG); 3347 3348 return (0); 3349} 3350 3351static int 3352spa_ld_load_dedup_tables(spa_t *spa) 3353{ 3354 int error = 0; 3355 vdev_t *rvd = spa->spa_root_vdev; 3356 3357 error = ddt_load(spa); 3358 if (error != 0) { 3359 spa_load_failed(spa, "ddt_load failed [error=%d]", error); 3360 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3361 } 3362 3363 return (0); 3364} 3365 3366static int 3367spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport) 3368{ 3369 vdev_t *rvd = spa->spa_root_vdev; 3370 3371 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { 3372 boolean_t missing = spa_check_logs(spa); 3373 if (missing) { 3374 if (spa->spa_missing_tvds != 0) { 3375 spa_load_note(spa, "spa_check_logs failed " 3376 "so dropping the logs"); 3377 } else { 3378 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 3379 spa_load_failed(spa, "spa_check_logs failed"); 3380 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, 3381 ENXIO)); 3382 } 3383 } 3384 } 3385 3386 return (0); 3387} 3388 3389static int 3390spa_ld_verify_pool_data(spa_t *spa) 3391{ 3392 int error = 0; 3393 vdev_t *rvd = spa->spa_root_vdev; 3394 3395 /* 3396 * We've successfully opened the pool, verify that we're ready 3397 * to start pushing transactions. 3398 */ 3399 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3400 error = spa_load_verify(spa); 3401 if (error != 0) { 3402 spa_load_failed(spa, "spa_load_verify failed " 3403 "[error=%d]", error); 3404 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 3405 error)); 3406 } 3407 } 3408 3409 return (0); 3410} 3411 3412static void 3413spa_ld_claim_log_blocks(spa_t *spa) 3414{ 3415 dmu_tx_t *tx; 3416 dsl_pool_t *dp = spa_get_dsl(spa); 3417 3418 /* 3419 * Claim log blocks that haven't been committed yet. 3420 * This must all happen in a single txg. 3421 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 3422 * invoked from zil_claim_log_block()'s i/o done callback. 3423 * Price of rollback is that we abandon the log. 3424 */ 3425 spa->spa_claiming = B_TRUE; 3426 3427 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 3428 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 3429 zil_claim, tx, DS_FIND_CHILDREN); 3430 dmu_tx_commit(tx); 3431 3432 spa->spa_claiming = B_FALSE; 3433 3434 spa_set_log_state(spa, SPA_LOG_GOOD); 3435} 3436 3437static void 3438spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, 3439 boolean_t update_config_cache) 3440{ 3441 vdev_t *rvd = spa->spa_root_vdev; 3442 int need_update = B_FALSE; 3443 3444 /* 3445 * If the config cache is stale, or we have uninitialized 3446 * metaslabs (see spa_vdev_add()), then update the config. 3447 * 3448 * If this is a verbatim import, trust the current 3449 * in-core spa_config and update the disk labels. 3450 */ 3451 if (update_config_cache || config_cache_txg != spa->spa_config_txg || 3452 spa->spa_load_state == SPA_LOAD_IMPORT || 3453 spa->spa_load_state == SPA_LOAD_RECOVER || 3454 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 3455 need_update = B_TRUE; 3456 3457 for (int c = 0; c < rvd->vdev_children; c++) 3458 if (rvd->vdev_child[c]->vdev_ms_array == 0) 3459 need_update = B_TRUE; 3460 3461 /* 3462 * Update the config cache asychronously in case we're the 3463 * root pool, in which case the config cache isn't writable yet. 3464 */ 3465 if (need_update) 3466 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3467} 3468 3469static void 3470spa_ld_prepare_for_reload(spa_t *spa) 3471{ 3472 int mode = spa->spa_mode; 3473 int async_suspended = spa->spa_async_suspended; 3474 3475 spa_unload(spa); 3476 spa_deactivate(spa); 3477 spa_activate(spa, mode); 3478 3479 /* 3480 * We save the value of spa_async_suspended as it gets reset to 0 by 3481 * spa_unload(). We want to restore it back to the original value before 3482 * returning as we might be calling spa_async_resume() later. 3483 */ 3484 spa->spa_async_suspended = async_suspended; 3485} 3486 3487static int 3488spa_ld_read_checkpoint_txg(spa_t *spa) 3489{ 3490 uberblock_t checkpoint; 3491 int error = 0; 3492 3493 ASSERT0(spa->spa_checkpoint_txg); 3494 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3495 3496 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3497 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 3498 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 3499 3500 if (error == ENOENT) 3501 return (0); 3502 3503 if (error != 0) 3504 return (error); 3505 3506 ASSERT3U(checkpoint.ub_txg, !=, 0); 3507 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); 3508 ASSERT3U(checkpoint.ub_timestamp, !=, 0); 3509 spa->spa_checkpoint_txg = checkpoint.ub_txg; 3510 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; 3511 3512 return (0); 3513} 3514 3515static int 3516spa_ld_mos_init(spa_t *spa, spa_import_type_t type) 3517{ 3518 int error = 0; 3519 3520 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3521 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3522 3523 /* 3524 * Never trust the config that is provided unless we are assembling 3525 * a pool following a split. 3526 * This means don't trust blkptrs and the vdev tree in general. This 3527 * also effectively puts the spa in read-only mode since 3528 * spa_writeable() checks for spa_trust_config to be true. 3529 * We will later load a trusted config from the MOS. 3530 */ 3531 if (type != SPA_IMPORT_ASSEMBLE) 3532 spa->spa_trust_config = B_FALSE; 3533 3534 /* 3535 * Parse the config provided to create a vdev tree. 3536 */ 3537 error = spa_ld_parse_config(spa, type); 3538 if (error != 0) 3539 return (error); 3540 3541 /* 3542 * Now that we have the vdev tree, try to open each vdev. This involves 3543 * opening the underlying physical device, retrieving its geometry and 3544 * probing the vdev with a dummy I/O. The state of each vdev will be set 3545 * based on the success of those operations. After this we'll be ready 3546 * to read from the vdevs. 3547 */ 3548 error = spa_ld_open_vdevs(spa); 3549 if (error != 0) 3550 return (error); 3551 3552 /* 3553 * Read the label of each vdev and make sure that the GUIDs stored 3554 * there match the GUIDs in the config provided. 3555 * If we're assembling a new pool that's been split off from an 3556 * existing pool, the labels haven't yet been updated so we skip 3557 * validation for now. 3558 */ 3559 if (type != SPA_IMPORT_ASSEMBLE) { 3560 error = spa_ld_validate_vdevs(spa); 3561 if (error != 0) 3562 return (error); 3563 } 3564 3565 /* 3566 * Read all vdev labels to find the best uberblock (i.e. latest, 3567 * unless spa_load_max_txg is set) and store it in spa_uberblock. We 3568 * get the list of features required to read blkptrs in the MOS from 3569 * the vdev label with the best uberblock and verify that our version 3570 * of zfs supports them all. 3571 */ 3572 error = spa_ld_select_uberblock(spa, type); 3573 if (error != 0) 3574 return (error); 3575 3576 /* 3577 * Pass that uberblock to the dsl_pool layer which will open the root 3578 * blkptr. This blkptr points to the latest version of the MOS and will 3579 * allow us to read its contents. 3580 */ 3581 error = spa_ld_open_rootbp(spa); 3582 if (error != 0) 3583 return (error); 3584 3585 return (0); 3586} 3587 3588static int 3589spa_ld_checkpoint_rewind(spa_t *spa) 3590{ 3591 uberblock_t checkpoint; 3592 int error = 0; 3593 3594 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3595 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3596 3597 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3598 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 3599 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 3600 3601 if (error != 0) { 3602 spa_load_failed(spa, "unable to retrieve checkpointed " 3603 "uberblock from the MOS config [error=%d]", error); 3604 3605 if (error == ENOENT) 3606 error = ZFS_ERR_NO_CHECKPOINT; 3607 3608 return (error); 3609 } 3610 3611 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); 3612 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); 3613 3614 /* 3615 * We need to update the txg and timestamp of the checkpointed 3616 * uberblock to be higher than the latest one. This ensures that 3617 * the checkpointed uberblock is selected if we were to close and 3618 * reopen the pool right after we've written it in the vdev labels. 3619 * (also see block comment in vdev_uberblock_compare) 3620 */ 3621 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; 3622 checkpoint.ub_timestamp = gethrestime_sec(); 3623 3624 /* 3625 * Set current uberblock to be the checkpointed uberblock. 3626 */ 3627 spa->spa_uberblock = checkpoint; 3628 3629 /* 3630 * If we are doing a normal rewind, then the pool is open for 3631 * writing and we sync the "updated" checkpointed uberblock to 3632 * disk. Once this is done, we've basically rewound the whole 3633 * pool and there is no way back. 3634 * 3635 * There are cases when we don't want to attempt and sync the 3636 * checkpointed uberblock to disk because we are opening a 3637 * pool as read-only. Specifically, verifying the checkpointed 3638 * state with zdb, and importing the checkpointed state to get 3639 * a "preview" of its content. 3640 */ 3641 if (spa_writeable(spa)) { 3642 vdev_t *rvd = spa->spa_root_vdev; 3643 3644 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3645 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; 3646 int svdcount = 0; 3647 int children = rvd->vdev_children; 3648 int c0 = spa_get_random(children); 3649 3650 for (int c = 0; c < children; c++) { 3651 vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; 3652 3653 /* Stop when revisiting the first vdev */ 3654 if (c > 0 && svd[0] == vd) 3655 break; 3656 3657 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 3658 !vdev_is_concrete(vd)) 3659 continue; 3660 3661 svd[svdcount++] = vd; 3662 if (svdcount == SPA_SYNC_MIN_VDEVS) 3663 break; 3664 } 3665 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); 3666 if (error == 0) 3667 spa->spa_last_synced_guid = rvd->vdev_guid; 3668 spa_config_exit(spa, SCL_ALL, FTAG); 3669 3670 if (error != 0) { 3671 spa_load_failed(spa, "failed to write checkpointed " 3672 "uberblock to the vdev labels [error=%d]", error); 3673 return (error); 3674 } 3675 } 3676 3677 return (0); 3678} 3679 3680static int 3681spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, 3682 boolean_t *update_config_cache) 3683{ 3684 int error; 3685 3686 /* 3687 * Parse the config for pool, open and validate vdevs, 3688 * select an uberblock, and use that uberblock to open 3689 * the MOS. 3690 */ 3691 error = spa_ld_mos_init(spa, type); 3692 if (error != 0) 3693 return (error); 3694 3695 /* 3696 * Retrieve the trusted config stored in the MOS and use it to create 3697 * a new, exact version of the vdev tree, then reopen all vdevs. 3698 */ 3699 error = spa_ld_trusted_config(spa, type, B_FALSE); 3700 if (error == EAGAIN) { 3701 if (update_config_cache != NULL) 3702 *update_config_cache = B_TRUE; 3703 3704 /* 3705 * Redo the loading process with the trusted config if it is 3706 * too different from the untrusted config. 3707 */ 3708 spa_ld_prepare_for_reload(spa); 3709 spa_load_note(spa, "RELOADING"); 3710 error = spa_ld_mos_init(spa, type); 3711 if (error != 0) 3712 return (error); 3713 3714 error = spa_ld_trusted_config(spa, type, B_TRUE); 3715 if (error != 0) 3716 return (error); 3717 3718 } else if (error != 0) { 3719 return (error); 3720 } 3721 3722 return (0); 3723} 3724 3725/* 3726 * Load an existing storage pool, using the config provided. This config 3727 * describes which vdevs are part of the pool and is later validated against 3728 * partial configs present in each vdev's label and an entire copy of the 3729 * config stored in the MOS. 3730 */ 3731static int 3732spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport) 3733{ 3734 int error = 0; 3735 boolean_t missing_feat_write = B_FALSE; 3736 boolean_t checkpoint_rewind = 3737 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3738 boolean_t update_config_cache = B_FALSE; 3739 3740 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3741 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3742 3743 spa_load_note(spa, "LOADING"); 3744 3745 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); 3746 if (error != 0) 3747 return (error); 3748 3749 /* 3750 * If we are rewinding to the checkpoint then we need to repeat 3751 * everything we've done so far in this function but this time 3752 * selecting the checkpointed uberblock and using that to open 3753 * the MOS. 3754 */ 3755 if (checkpoint_rewind) { 3756 /* 3757 * If we are rewinding to the checkpoint update config cache 3758 * anyway. 3759 */ 3760 update_config_cache = B_TRUE; 3761 3762 /* 3763 * Extract the checkpointed uberblock from the current MOS 3764 * and use this as the pool's uberblock from now on. If the 3765 * pool is imported as writeable we also write the checkpoint 3766 * uberblock to the labels, making the rewind permanent. 3767 */ 3768 error = spa_ld_checkpoint_rewind(spa); 3769 if (error != 0) 3770 return (error); 3771 3772 /* 3773 * Redo the loading process process again with the 3774 * checkpointed uberblock. 3775 */ 3776 spa_ld_prepare_for_reload(spa); 3777 spa_load_note(spa, "LOADING checkpointed uberblock"); 3778 error = spa_ld_mos_with_trusted_config(spa, type, NULL); 3779 if (error != 0) 3780 return (error); 3781 } 3782 3783 /* 3784 * Retrieve the checkpoint txg if the pool has a checkpoint. 3785 */ 3786 error = spa_ld_read_checkpoint_txg(spa); 3787 if (error != 0) 3788 return (error); 3789 3790 /* 3791 * Retrieve the mapping of indirect vdevs. Those vdevs were removed 3792 * from the pool and their contents were re-mapped to other vdevs. Note 3793 * that everything that we read before this step must have been 3794 * rewritten on concrete vdevs after the last device removal was 3795 * initiated. Otherwise we could be reading from indirect vdevs before 3796 * we have loaded their mappings. 3797 */ 3798 error = spa_ld_open_indirect_vdev_metadata(spa); 3799 if (error != 0) 3800 return (error); 3801 3802 /* 3803 * Retrieve the full list of active features from the MOS and check if 3804 * they are all supported. 3805 */ 3806 error = spa_ld_check_features(spa, &missing_feat_write); 3807 if (error != 0) 3808 return (error); 3809 3810 /* 3811 * Load several special directories from the MOS needed by the dsl_pool 3812 * layer. 3813 */ 3814 error = spa_ld_load_special_directories(spa); 3815 if (error != 0) 3816 return (error); 3817 3818 /* 3819 * Retrieve pool properties from the MOS. 3820 */ 3821 error = spa_ld_get_props(spa); 3822 if (error != 0) 3823 return (error); 3824 3825 /* 3826 * Retrieve the list of auxiliary devices - cache devices and spares - 3827 * and open them. 3828 */ 3829 error = spa_ld_open_aux_vdevs(spa, type); 3830 if (error != 0) 3831 return (error); 3832 3833 /* 3834 * Load the metadata for all vdevs. Also check if unopenable devices 3835 * should be autoreplaced. 3836 */ 3837 error = spa_ld_load_vdev_metadata(spa); 3838 if (error != 0) 3839 return (error); 3840 3841 error = spa_ld_load_dedup_tables(spa); 3842 if (error != 0) 3843 return (error); 3844 3845 /* 3846 * Verify the logs now to make sure we don't have any unexpected errors 3847 * when we claim log blocks later. 3848 */ 3849 error = spa_ld_verify_logs(spa, type, ereport); 3850 if (error != 0) 3851 return (error); 3852 3853 if (missing_feat_write) { 3854 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); 3855 3856 /* 3857 * At this point, we know that we can open the pool in 3858 * read-only mode but not read-write mode. We now have enough 3859 * information and can return to userland. 3860 */ 3861 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, 3862 ENOTSUP)); 3863 } 3864 3865 /* 3866 * Traverse the last txgs to make sure the pool was left off in a safe 3867 * state. When performing an extreme rewind, we verify the whole pool, 3868 * which can take a very long time. 3869 */ 3870 error = spa_ld_verify_pool_data(spa); 3871 if (error != 0) 3872 return (error); 3873 3874 /* 3875 * Calculate the deflated space for the pool. This must be done before 3876 * we write anything to the pool because we'd need to update the space 3877 * accounting using the deflated sizes. 3878 */ 3879 spa_update_dspace(spa); 3880 3881 /* 3882 * We have now retrieved all the information we needed to open the 3883 * pool. If we are importing the pool in read-write mode, a few 3884 * additional steps must be performed to finish the import. 3885 */ 3886 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || 3887 spa->spa_load_max_txg == UINT64_MAX)) { 3888 uint64_t config_cache_txg = spa->spa_config_txg; 3889 3890 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); 3891 3892 /* 3893 * In case of a checkpoint rewind, log the original txg 3894 * of the checkpointed uberblock. 3895 */ 3896 if (checkpoint_rewind) { 3897 spa_history_log_internal(spa, "checkpoint rewind", 3898 NULL, "rewound state to txg=%llu", 3899 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); 3900 } 3901 3902 /* 3903 * Traverse the ZIL and claim all blocks. 3904 */ 3905 spa_ld_claim_log_blocks(spa); 3906 3907 /* 3908 * Kick-off the syncing thread. 3909 */ 3910 spa->spa_sync_on = B_TRUE; 3911 txg_sync_start(spa->spa_dsl_pool); 3912 3913 /* 3914 * Wait for all claims to sync. We sync up to the highest 3915 * claimed log block birth time so that claimed log blocks 3916 * don't appear to be from the future. spa_claim_max_txg 3917 * will have been set for us by ZIL traversal operations 3918 * performed above. 3919 */ 3920 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 3921 3922 /* 3923 * Check if we need to request an update of the config. On the 3924 * next sync, we would update the config stored in vdev labels 3925 * and the cachefile (by default /etc/zfs/zpool.cache). 3926 */ 3927 spa_ld_check_for_config_update(spa, config_cache_txg, 3928 update_config_cache); 3929 3930 /* 3931 * Check all DTLs to see if anything needs resilvering. 3932 */ 3933 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 3934 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) 3935 spa_async_request(spa, SPA_ASYNC_RESILVER); 3936 3937 /* 3938 * Log the fact that we booted up (so that we can detect if 3939 * we rebooted in the middle of an operation). 3940 */ 3941 spa_history_log_version(spa, "open"); 3942 3943 /* 3944 * Delete any inconsistent datasets. 3945 */ 3946 (void) dmu_objset_find(spa_name(spa), 3947 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 3948 3949 /* 3950 * Clean up any stale temporary dataset userrefs. 3951 */ 3952 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 3953 3954 spa_restart_removal(spa); 3955 3956 spa_spawn_aux_threads(spa); 3957 } 3958 3959 spa_load_note(spa, "LOADED"); 3960 3961 return (0); 3962} 3963 3964static int 3965spa_load_retry(spa_t *spa, spa_load_state_t state) 3966{ 3967 int mode = spa->spa_mode; 3968 3969 spa_unload(spa); 3970 spa_deactivate(spa); 3971 3972 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 3973 3974 spa_activate(spa, mode); 3975 spa_async_suspend(spa); 3976 3977 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", 3978 (u_longlong_t)spa->spa_load_max_txg); 3979 3980 return (spa_load(spa, state, SPA_IMPORT_EXISTING)); 3981} 3982 3983/* 3984 * If spa_load() fails this function will try loading prior txg's. If 3985 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 3986 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 3987 * function will not rewind the pool and will return the same error as 3988 * spa_load(). 3989 */ 3990static int 3991spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, 3992 int rewind_flags) 3993{ 3994 nvlist_t *loadinfo = NULL; 3995 nvlist_t *config = NULL; 3996 int load_error, rewind_error; 3997 uint64_t safe_rewind_txg; 3998 uint64_t min_txg; 3999 4000 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 4001 spa->spa_load_max_txg = spa->spa_load_txg; 4002 spa_set_log_state(spa, SPA_LOG_CLEAR); 4003 } else { 4004 spa->spa_load_max_txg = max_request; 4005 if (max_request != UINT64_MAX) 4006 spa->spa_extreme_rewind = B_TRUE; 4007 } 4008 4009 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); 4010 if (load_error == 0) 4011 return (0); 4012 if (load_error == ZFS_ERR_NO_CHECKPOINT) { 4013 /* 4014 * When attempting checkpoint-rewind on a pool with no 4015 * checkpoint, we should not attempt to load uberblocks 4016 * from previous txgs when spa_load fails. 4017 */ 4018 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 4019 return (load_error); 4020 } 4021 4022 if (spa->spa_root_vdev != NULL) 4023 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4024 4025 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 4026 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 4027 4028 if (rewind_flags & ZPOOL_NEVER_REWIND) { 4029 nvlist_free(config); 4030 return (load_error); 4031 } 4032 4033 if (state == SPA_LOAD_RECOVER) { 4034 /* Price of rolling back is discarding txgs, including log */ 4035 spa_set_log_state(spa, SPA_LOG_CLEAR); 4036 } else { 4037 /* 4038 * If we aren't rolling back save the load info from our first 4039 * import attempt so that we can restore it after attempting 4040 * to rewind. 4041 */ 4042 loadinfo = spa->spa_load_info; 4043 spa->spa_load_info = fnvlist_alloc(); 4044 } 4045 4046 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 4047 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 4048 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 4049 TXG_INITIAL : safe_rewind_txg; 4050 4051 /* 4052 * Continue as long as we're finding errors, we're still within 4053 * the acceptable rewind range, and we're still finding uberblocks 4054 */ 4055 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 4056 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 4057 if (spa->spa_load_max_txg < safe_rewind_txg) 4058 spa->spa_extreme_rewind = B_TRUE; 4059 rewind_error = spa_load_retry(spa, state); 4060 } 4061 4062 spa->spa_extreme_rewind = B_FALSE; 4063 spa->spa_load_max_txg = UINT64_MAX; 4064 4065 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 4066 spa_config_set(spa, config); 4067 else 4068 nvlist_free(config); 4069 4070 if (state == SPA_LOAD_RECOVER) { 4071 ASSERT3P(loadinfo, ==, NULL); 4072 return (rewind_error); 4073 } else { 4074 /* Store the rewind info as part of the initial load info */ 4075 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 4076 spa->spa_load_info); 4077 4078 /* Restore the initial load info */ 4079 fnvlist_free(spa->spa_load_info); 4080 spa->spa_load_info = loadinfo; 4081 4082 return (load_error); 4083 } 4084} 4085 4086/* 4087 * Pool Open/Import 4088 * 4089 * The import case is identical to an open except that the configuration is sent 4090 * down from userland, instead of grabbed from the configuration cache. For the 4091 * case of an open, the pool configuration will exist in the 4092 * POOL_STATE_UNINITIALIZED state. 4093 * 4094 * The stats information (gen/count/ustats) is used to gather vdev statistics at 4095 * the same time open the pool, without having to keep around the spa_t in some 4096 * ambiguous state. 4097 */ 4098static int 4099spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 4100 nvlist_t **config) 4101{ 4102 spa_t *spa; 4103 spa_load_state_t state = SPA_LOAD_OPEN; 4104 int error; 4105 int locked = B_FALSE; 4106 int firstopen = B_FALSE; 4107 4108 *spapp = NULL; 4109 4110 /* 4111 * As disgusting as this is, we need to support recursive calls to this 4112 * function because dsl_dir_open() is called during spa_load(), and ends 4113 * up calling spa_open() again. The real fix is to figure out how to 4114 * avoid dsl_dir_open() calling this in the first place. 4115 */ 4116 if (mutex_owner(&spa_namespace_lock) != curthread) { 4117 mutex_enter(&spa_namespace_lock); 4118 locked = B_TRUE; 4119 } 4120 4121 if ((spa = spa_lookup(pool)) == NULL) { 4122 if (locked) 4123 mutex_exit(&spa_namespace_lock); 4124 return (SET_ERROR(ENOENT)); 4125 } 4126 4127 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 4128 zpool_load_policy_t policy; 4129 4130 firstopen = B_TRUE; 4131 4132 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, 4133 &policy); 4134 if (policy.zlp_rewind & ZPOOL_DO_REWIND) 4135 state = SPA_LOAD_RECOVER; 4136 4137 spa_activate(spa, spa_mode_global); 4138 4139 if (state != SPA_LOAD_RECOVER) 4140 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4141 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 4142 4143 zfs_dbgmsg("spa_open_common: opening %s", pool); 4144 error = spa_load_best(spa, state, policy.zlp_txg, 4145 policy.zlp_rewind); 4146 4147 if (error == EBADF) { 4148 /* 4149 * If vdev_validate() returns failure (indicated by 4150 * EBADF), it indicates that one of the vdevs indicates 4151 * that the pool has been exported or destroyed. If 4152 * this is the case, the config cache is out of sync and 4153 * we should remove the pool from the namespace. 4154 */ 4155 spa_unload(spa); 4156 spa_deactivate(spa); 4157 spa_write_cachefile(spa, B_TRUE, B_TRUE); 4158 spa_remove(spa); 4159 if (locked) 4160 mutex_exit(&spa_namespace_lock); 4161 return (SET_ERROR(ENOENT)); 4162 } 4163 4164 if (error) { 4165 /* 4166 * We can't open the pool, but we still have useful 4167 * information: the state of each vdev after the 4168 * attempted vdev_open(). Return this to the user. 4169 */ 4170 if (config != NULL && spa->spa_config) { 4171 VERIFY(nvlist_dup(spa->spa_config, config, 4172 KM_SLEEP) == 0); 4173 VERIFY(nvlist_add_nvlist(*config, 4174 ZPOOL_CONFIG_LOAD_INFO, 4175 spa->spa_load_info) == 0); 4176 } 4177 spa_unload(spa); 4178 spa_deactivate(spa); 4179 spa->spa_last_open_failed = error; 4180 if (locked) 4181 mutex_exit(&spa_namespace_lock); 4182 *spapp = NULL; 4183 return (error); 4184 } 4185 } 4186 4187 spa_open_ref(spa, tag); 4188 4189 if (config != NULL) 4190 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4191 4192 /* 4193 * If we've recovered the pool, pass back any information we 4194 * gathered while doing the load. 4195 */ 4196 if (state == SPA_LOAD_RECOVER) { 4197 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 4198 spa->spa_load_info) == 0); 4199 } 4200 4201 if (locked) { 4202 spa->spa_last_open_failed = 0; 4203 spa->spa_last_ubsync_txg = 0; 4204 spa->spa_load_txg = 0; 4205 mutex_exit(&spa_namespace_lock); 4206#ifdef __FreeBSD__ 4207#ifdef _KERNEL 4208 if (firstopen) 4209 zvol_create_minors(spa->spa_name); 4210#endif 4211#endif 4212 } 4213 4214 *spapp = spa; 4215 4216 return (0); 4217} 4218 4219int 4220spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 4221 nvlist_t **config) 4222{ 4223 return (spa_open_common(name, spapp, tag, policy, config)); 4224} 4225 4226int 4227spa_open(const char *name, spa_t **spapp, void *tag) 4228{ 4229 return (spa_open_common(name, spapp, tag, NULL, NULL)); 4230} 4231 4232/* 4233 * Lookup the given spa_t, incrementing the inject count in the process, 4234 * preventing it from being exported or destroyed. 4235 */ 4236spa_t * 4237spa_inject_addref(char *name) 4238{ 4239 spa_t *spa; 4240 4241 mutex_enter(&spa_namespace_lock); 4242 if ((spa = spa_lookup(name)) == NULL) { 4243 mutex_exit(&spa_namespace_lock); 4244 return (NULL); 4245 } 4246 spa->spa_inject_ref++; 4247 mutex_exit(&spa_namespace_lock); 4248 4249 return (spa); 4250} 4251 4252void 4253spa_inject_delref(spa_t *spa) 4254{ 4255 mutex_enter(&spa_namespace_lock); 4256 spa->spa_inject_ref--; 4257 mutex_exit(&spa_namespace_lock); 4258} 4259 4260/* 4261 * Add spares device information to the nvlist. 4262 */ 4263static void 4264spa_add_spares(spa_t *spa, nvlist_t *config) 4265{ 4266 nvlist_t **spares; 4267 uint_t i, nspares; 4268 nvlist_t *nvroot; 4269 uint64_t guid; 4270 vdev_stat_t *vs; 4271 uint_t vsc; 4272 uint64_t pool; 4273 4274 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4275 4276 if (spa->spa_spares.sav_count == 0) 4277 return; 4278 4279 VERIFY(nvlist_lookup_nvlist(config, 4280 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4281 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 4282 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 4283 if (nspares != 0) { 4284 VERIFY(nvlist_add_nvlist_array(nvroot, 4285 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4286 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4287 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 4288 4289 /* 4290 * Go through and find any spares which have since been 4291 * repurposed as an active spare. If this is the case, update 4292 * their status appropriately. 4293 */ 4294 for (i = 0; i < nspares; i++) { 4295 VERIFY(nvlist_lookup_uint64(spares[i], 4296 ZPOOL_CONFIG_GUID, &guid) == 0); 4297 if (spa_spare_exists(guid, &pool, NULL) && 4298 pool != 0ULL) { 4299 VERIFY(nvlist_lookup_uint64_array( 4300 spares[i], ZPOOL_CONFIG_VDEV_STATS, 4301 (uint64_t **)&vs, &vsc) == 0); 4302 vs->vs_state = VDEV_STATE_CANT_OPEN; 4303 vs->vs_aux = VDEV_AUX_SPARED; 4304 } 4305 } 4306 } 4307} 4308 4309/* 4310 * Add l2cache device information to the nvlist, including vdev stats. 4311 */ 4312static void 4313spa_add_l2cache(spa_t *spa, nvlist_t *config) 4314{ 4315 nvlist_t **l2cache; 4316 uint_t i, j, nl2cache; 4317 nvlist_t *nvroot; 4318 uint64_t guid; 4319 vdev_t *vd; 4320 vdev_stat_t *vs; 4321 uint_t vsc; 4322 4323 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4324 4325 if (spa->spa_l2cache.sav_count == 0) 4326 return; 4327 4328 VERIFY(nvlist_lookup_nvlist(config, 4329 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4330 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 4331 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4332 if (nl2cache != 0) { 4333 VERIFY(nvlist_add_nvlist_array(nvroot, 4334 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4335 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4336 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4337 4338 /* 4339 * Update level 2 cache device stats. 4340 */ 4341 4342 for (i = 0; i < nl2cache; i++) { 4343 VERIFY(nvlist_lookup_uint64(l2cache[i], 4344 ZPOOL_CONFIG_GUID, &guid) == 0); 4345 4346 vd = NULL; 4347 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 4348 if (guid == 4349 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 4350 vd = spa->spa_l2cache.sav_vdevs[j]; 4351 break; 4352 } 4353 } 4354 ASSERT(vd != NULL); 4355 4356 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 4357 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 4358 == 0); 4359 vdev_get_stats(vd, vs); 4360 } 4361 } 4362} 4363 4364static void 4365spa_add_feature_stats(spa_t *spa, nvlist_t *config) 4366{ 4367 nvlist_t *features; 4368 zap_cursor_t zc; 4369 zap_attribute_t za; 4370 4371 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4372 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4373 4374 /* We may be unable to read features if pool is suspended. */ 4375 if (spa_suspended(spa)) 4376 goto out; 4377 4378 if (spa->spa_feat_for_read_obj != 0) { 4379 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4380 spa->spa_feat_for_read_obj); 4381 zap_cursor_retrieve(&zc, &za) == 0; 4382 zap_cursor_advance(&zc)) { 4383 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4384 za.za_num_integers == 1); 4385 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 4386 za.za_first_integer)); 4387 } 4388 zap_cursor_fini(&zc); 4389 } 4390 4391 if (spa->spa_feat_for_write_obj != 0) { 4392 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4393 spa->spa_feat_for_write_obj); 4394 zap_cursor_retrieve(&zc, &za) == 0; 4395 zap_cursor_advance(&zc)) { 4396 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4397 za.za_num_integers == 1); 4398 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 4399 za.za_first_integer)); 4400 } 4401 zap_cursor_fini(&zc); 4402 } 4403 4404out: 4405 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 4406 features) == 0); 4407 nvlist_free(features); 4408} 4409 4410int 4411spa_get_stats(const char *name, nvlist_t **config, 4412 char *altroot, size_t buflen) 4413{ 4414 int error; 4415 spa_t *spa; 4416 4417 *config = NULL; 4418 error = spa_open_common(name, &spa, FTAG, NULL, config); 4419 4420 if (spa != NULL) { 4421 /* 4422 * This still leaves a window of inconsistency where the spares 4423 * or l2cache devices could change and the config would be 4424 * self-inconsistent. 4425 */ 4426 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4427 4428 if (*config != NULL) { 4429 uint64_t loadtimes[2]; 4430 4431 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 4432 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 4433 VERIFY(nvlist_add_uint64_array(*config, 4434 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 4435 4436 VERIFY(nvlist_add_uint64(*config, 4437 ZPOOL_CONFIG_ERRCOUNT, 4438 spa_get_errlog_size(spa)) == 0); 4439 4440 if (spa_suspended(spa)) 4441 VERIFY(nvlist_add_uint64(*config, 4442 ZPOOL_CONFIG_SUSPENDED, 4443 spa->spa_failmode) == 0); 4444 4445 spa_add_spares(spa, *config); 4446 spa_add_l2cache(spa, *config); 4447 spa_add_feature_stats(spa, *config); 4448 } 4449 } 4450 4451 /* 4452 * We want to get the alternate root even for faulted pools, so we cheat 4453 * and call spa_lookup() directly. 4454 */ 4455 if (altroot) { 4456 if (spa == NULL) { 4457 mutex_enter(&spa_namespace_lock); 4458 spa = spa_lookup(name); 4459 if (spa) 4460 spa_altroot(spa, altroot, buflen); 4461 else 4462 altroot[0] = '\0'; 4463 spa = NULL; 4464 mutex_exit(&spa_namespace_lock); 4465 } else { 4466 spa_altroot(spa, altroot, buflen); 4467 } 4468 } 4469 4470 if (spa != NULL) { 4471 spa_config_exit(spa, SCL_CONFIG, FTAG); 4472 spa_close(spa, FTAG); 4473 } 4474 4475 return (error); 4476} 4477 4478/* 4479 * Validate that the auxiliary device array is well formed. We must have an 4480 * array of nvlists, each which describes a valid leaf vdev. If this is an 4481 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 4482 * specified, as long as they are well-formed. 4483 */ 4484static int 4485spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 4486 spa_aux_vdev_t *sav, const char *config, uint64_t version, 4487 vdev_labeltype_t label) 4488{ 4489 nvlist_t **dev; 4490 uint_t i, ndev; 4491 vdev_t *vd; 4492 int error; 4493 4494 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4495 4496 /* 4497 * It's acceptable to have no devs specified. 4498 */ 4499 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 4500 return (0); 4501 4502 if (ndev == 0) 4503 return (SET_ERROR(EINVAL)); 4504 4505 /* 4506 * Make sure the pool is formatted with a version that supports this 4507 * device type. 4508 */ 4509 if (spa_version(spa) < version) 4510 return (SET_ERROR(ENOTSUP)); 4511 4512 /* 4513 * Set the pending device list so we correctly handle device in-use 4514 * checking. 4515 */ 4516 sav->sav_pending = dev; 4517 sav->sav_npending = ndev; 4518 4519 for (i = 0; i < ndev; i++) { 4520 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 4521 mode)) != 0) 4522 goto out; 4523 4524 if (!vd->vdev_ops->vdev_op_leaf) { 4525 vdev_free(vd); 4526 error = SET_ERROR(EINVAL); 4527 goto out; 4528 } 4529 4530 /* 4531 * The L2ARC currently only supports disk devices in 4532 * kernel context. For user-level testing, we allow it. 4533 */ 4534#ifdef _KERNEL 4535 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 4536 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 4537 error = SET_ERROR(ENOTBLK); 4538 vdev_free(vd); 4539 goto out; 4540 } 4541#endif 4542 vd->vdev_top = vd; 4543 4544 if ((error = vdev_open(vd)) == 0 && 4545 (error = vdev_label_init(vd, crtxg, label)) == 0) { 4546 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 4547 vd->vdev_guid) == 0); 4548 } 4549 4550 vdev_free(vd); 4551 4552 if (error && 4553 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 4554 goto out; 4555 else 4556 error = 0; 4557 } 4558 4559out: 4560 sav->sav_pending = NULL; 4561 sav->sav_npending = 0; 4562 return (error); 4563} 4564 4565static int 4566spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 4567{ 4568 int error; 4569 4570 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4571 4572 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4573 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 4574 VDEV_LABEL_SPARE)) != 0) { 4575 return (error); 4576 } 4577 4578 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4579 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 4580 VDEV_LABEL_L2CACHE)); 4581} 4582 4583static void 4584spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 4585 const char *config) 4586{ 4587 int i; 4588 4589 if (sav->sav_config != NULL) { 4590 nvlist_t **olddevs; 4591 uint_t oldndevs; 4592 nvlist_t **newdevs; 4593 4594 /* 4595 * Generate new dev list by concatentating with the 4596 * current dev list. 4597 */ 4598 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 4599 &olddevs, &oldndevs) == 0); 4600 4601 newdevs = kmem_alloc(sizeof (void *) * 4602 (ndevs + oldndevs), KM_SLEEP); 4603 for (i = 0; i < oldndevs; i++) 4604 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 4605 KM_SLEEP) == 0); 4606 for (i = 0; i < ndevs; i++) 4607 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 4608 KM_SLEEP) == 0); 4609 4610 VERIFY(nvlist_remove(sav->sav_config, config, 4611 DATA_TYPE_NVLIST_ARRAY) == 0); 4612 4613 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 4614 config, newdevs, ndevs + oldndevs) == 0); 4615 for (i = 0; i < oldndevs + ndevs; i++) 4616 nvlist_free(newdevs[i]); 4617 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 4618 } else { 4619 /* 4620 * Generate a new dev list. 4621 */ 4622 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 4623 KM_SLEEP) == 0); 4624 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 4625 devs, ndevs) == 0); 4626 } 4627} 4628 4629/* 4630 * Stop and drop level 2 ARC devices 4631 */ 4632void 4633spa_l2cache_drop(spa_t *spa) 4634{ 4635 vdev_t *vd; 4636 int i; 4637 spa_aux_vdev_t *sav = &spa->spa_l2cache; 4638 4639 for (i = 0; i < sav->sav_count; i++) { 4640 uint64_t pool; 4641 4642 vd = sav->sav_vdevs[i]; 4643 ASSERT(vd != NULL); 4644 4645 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 4646 pool != 0ULL && l2arc_vdev_present(vd)) 4647 l2arc_remove_vdev(vd); 4648 } 4649} 4650 4651/* 4652 * Pool Creation 4653 */ 4654int 4655spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 4656 nvlist_t *zplprops) 4657{ 4658 spa_t *spa; 4659 char *altroot = NULL; 4660 vdev_t *rvd; 4661 dsl_pool_t *dp; 4662 dmu_tx_t *tx; 4663 int error = 0; 4664 uint64_t txg = TXG_INITIAL; 4665 nvlist_t **spares, **l2cache; 4666 uint_t nspares, nl2cache; 4667 uint64_t version, obj; 4668 boolean_t has_features; 4669 char *poolname; 4670 nvlist_t *nvl; 4671 4672 if (nvlist_lookup_string(props, 4673 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0) 4674 poolname = (char *)pool; 4675 4676 /* 4677 * If this pool already exists, return failure. 4678 */ 4679 mutex_enter(&spa_namespace_lock); 4680 if (spa_lookup(poolname) != NULL) { 4681 mutex_exit(&spa_namespace_lock); 4682 return (SET_ERROR(EEXIST)); 4683 } 4684 4685 /* 4686 * Allocate a new spa_t structure. 4687 */ 4688 nvl = fnvlist_alloc(); 4689 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); 4690 (void) nvlist_lookup_string(props, 4691 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4692 spa = spa_add(poolname, nvl, altroot); 4693 fnvlist_free(nvl); 4694 spa_activate(spa, spa_mode_global); 4695 4696 if (props && (error = spa_prop_validate(spa, props))) { 4697 spa_deactivate(spa); 4698 spa_remove(spa); 4699 mutex_exit(&spa_namespace_lock); 4700 return (error); 4701 } 4702 4703 /* 4704 * Temporary pool names should never be written to disk. 4705 */ 4706 if (poolname != pool) 4707 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; 4708 4709 has_features = B_FALSE; 4710 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 4711 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 4712 if (zpool_prop_feature(nvpair_name(elem))) 4713 has_features = B_TRUE; 4714 } 4715 4716 if (has_features || nvlist_lookup_uint64(props, 4717 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 4718 version = SPA_VERSION; 4719 } 4720 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 4721 4722 spa->spa_first_txg = txg; 4723 spa->spa_uberblock.ub_txg = txg - 1; 4724 spa->spa_uberblock.ub_version = version; 4725 spa->spa_ubsync = spa->spa_uberblock; 4726 spa->spa_load_state = SPA_LOAD_CREATE; 4727 spa->spa_removing_phys.sr_state = DSS_NONE; 4728 spa->spa_removing_phys.sr_removing_vdev = -1; 4729 spa->spa_removing_phys.sr_prev_indirect_vdev = -1; 4730 spa->spa_indirect_vdevs_loaded = B_TRUE; 4731 4732 /* 4733 * Create "The Godfather" zio to hold all async IOs 4734 */ 4735 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 4736 KM_SLEEP); 4737 for (int i = 0; i < max_ncpus; i++) { 4738 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 4739 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 4740 ZIO_FLAG_GODFATHER); 4741 } 4742 4743 /* 4744 * Create the root vdev. 4745 */ 4746 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4747 4748 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 4749 4750 ASSERT(error != 0 || rvd != NULL); 4751 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 4752 4753 if (error == 0 && !zfs_allocatable_devs(nvroot)) 4754 error = SET_ERROR(EINVAL); 4755 4756 if (error == 0 && 4757 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 4758 (error = spa_validate_aux(spa, nvroot, txg, 4759 VDEV_ALLOC_ADD)) == 0) { 4760 for (int c = 0; c < rvd->vdev_children; c++) { 4761 vdev_ashift_optimize(rvd->vdev_child[c]); 4762 vdev_metaslab_set_size(rvd->vdev_child[c]); 4763 vdev_expand(rvd->vdev_child[c], txg); 4764 } 4765 } 4766 4767 spa_config_exit(spa, SCL_ALL, FTAG); 4768 4769 if (error != 0) { 4770 spa_unload(spa); 4771 spa_deactivate(spa); 4772 spa_remove(spa); 4773 mutex_exit(&spa_namespace_lock); 4774 return (error); 4775 } 4776 4777 /* 4778 * Get the list of spares, if specified. 4779 */ 4780 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4781 &spares, &nspares) == 0) { 4782 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 4783 KM_SLEEP) == 0); 4784 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4785 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4786 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4787 spa_load_spares(spa); 4788 spa_config_exit(spa, SCL_ALL, FTAG); 4789 spa->spa_spares.sav_sync = B_TRUE; 4790 } 4791 4792 /* 4793 * Get the list of level 2 cache devices, if specified. 4794 */ 4795 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4796 &l2cache, &nl2cache) == 0) { 4797 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4798 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4799 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4800 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4801 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4802 spa_load_l2cache(spa); 4803 spa_config_exit(spa, SCL_ALL, FTAG); 4804 spa->spa_l2cache.sav_sync = B_TRUE; 4805 } 4806 4807 spa->spa_is_initializing = B_TRUE; 4808 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 4809 spa->spa_meta_objset = dp->dp_meta_objset; 4810 spa->spa_is_initializing = B_FALSE; 4811 4812 /* 4813 * Create DDTs (dedup tables). 4814 */ 4815 ddt_create(spa); 4816 4817 spa_update_dspace(spa); 4818 4819 tx = dmu_tx_create_assigned(dp, txg); 4820 4821 /* 4822 * Create the pool config object. 4823 */ 4824 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 4825 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 4826 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 4827 4828 if (zap_add(spa->spa_meta_objset, 4829 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 4830 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 4831 cmn_err(CE_PANIC, "failed to add pool config"); 4832 } 4833 4834 if (spa_version(spa) >= SPA_VERSION_FEATURES) 4835 spa_feature_create_zap_objects(spa, tx); 4836 4837 if (zap_add(spa->spa_meta_objset, 4838 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 4839 sizeof (uint64_t), 1, &version, tx) != 0) { 4840 cmn_err(CE_PANIC, "failed to add pool version"); 4841 } 4842 4843 /* Newly created pools with the right version are always deflated. */ 4844 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 4845 spa->spa_deflate = TRUE; 4846 if (zap_add(spa->spa_meta_objset, 4847 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 4848 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 4849 cmn_err(CE_PANIC, "failed to add deflate"); 4850 } 4851 } 4852 4853 /* 4854 * Create the deferred-free bpobj. Turn off compression 4855 * because sync-to-convergence takes longer if the blocksize 4856 * keeps changing. 4857 */ 4858 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 4859 dmu_object_set_compress(spa->spa_meta_objset, obj, 4860 ZIO_COMPRESS_OFF, tx); 4861 if (zap_add(spa->spa_meta_objset, 4862 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 4863 sizeof (uint64_t), 1, &obj, tx) != 0) { 4864 cmn_err(CE_PANIC, "failed to add bpobj"); 4865 } 4866 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 4867 spa->spa_meta_objset, obj)); 4868 4869 /* 4870 * Create the pool's history object. 4871 */ 4872 if (version >= SPA_VERSION_ZPOOL_HISTORY) 4873 spa_history_create_obj(spa, tx); 4874 4875 /* 4876 * Generate some random noise for salted checksums to operate on. 4877 */ 4878 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 4879 sizeof (spa->spa_cksum_salt.zcs_bytes)); 4880 4881 /* 4882 * Set pool properties. 4883 */ 4884 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 4885 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 4886 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 4887 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 4888 4889 if (props != NULL) { 4890 spa_configfile_set(spa, props, B_FALSE); 4891 spa_sync_props(props, tx); 4892 } 4893 4894 dmu_tx_commit(tx); 4895 4896 spa->spa_sync_on = B_TRUE; 4897 txg_sync_start(spa->spa_dsl_pool); 4898 4899 /* 4900 * We explicitly wait for the first transaction to complete so that our 4901 * bean counters are appropriately updated. 4902 */ 4903 txg_wait_synced(spa->spa_dsl_pool, txg); 4904 4905 spa_spawn_aux_threads(spa); 4906 4907 spa_write_cachefile(spa, B_FALSE, B_TRUE); 4908 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); 4909 4910 spa_history_log_version(spa, "create"); 4911 4912 /* 4913 * Don't count references from objsets that are already closed 4914 * and are making their way through the eviction process. 4915 */ 4916 spa_evicting_os_wait(spa); 4917 spa->spa_minref = refcount_count(&spa->spa_refcount); 4918 spa->spa_load_state = SPA_LOAD_NONE; 4919 4920 mutex_exit(&spa_namespace_lock); 4921 4922 return (0); 4923} 4924 4925#ifdef _KERNEL 4926#ifdef illumos 4927/* 4928 * Get the root pool information from the root disk, then import the root pool 4929 * during the system boot up time. 4930 */ 4931extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 4932 4933static nvlist_t * 4934spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 4935{ 4936 nvlist_t *config; 4937 nvlist_t *nvtop, *nvroot; 4938 uint64_t pgid; 4939 4940 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 4941 return (NULL); 4942 4943 /* 4944 * Add this top-level vdev to the child array. 4945 */ 4946 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4947 &nvtop) == 0); 4948 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4949 &pgid) == 0); 4950 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 4951 4952 /* 4953 * Put this pool's top-level vdevs into a root vdev. 4954 */ 4955 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4956 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4957 VDEV_TYPE_ROOT) == 0); 4958 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4959 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4960 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4961 &nvtop, 1) == 0); 4962 4963 /* 4964 * Replace the existing vdev_tree with the new root vdev in 4965 * this pool's configuration (remove the old, add the new). 4966 */ 4967 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4968 nvlist_free(nvroot); 4969 return (config); 4970} 4971 4972/* 4973 * Walk the vdev tree and see if we can find a device with "better" 4974 * configuration. A configuration is "better" if the label on that 4975 * device has a more recent txg. 4976 */ 4977static void 4978spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 4979{ 4980 for (int c = 0; c < vd->vdev_children; c++) 4981 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 4982 4983 if (vd->vdev_ops->vdev_op_leaf) { 4984 nvlist_t *label; 4985 uint64_t label_txg; 4986 4987 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 4988 &label) != 0) 4989 return; 4990 4991 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 4992 &label_txg) == 0); 4993 4994 /* 4995 * Do we have a better boot device? 4996 */ 4997 if (label_txg > *txg) { 4998 *txg = label_txg; 4999 *avd = vd; 5000 } 5001 nvlist_free(label); 5002 } 5003} 5004 5005/* 5006 * Import a root pool. 5007 * 5008 * For x86. devpath_list will consist of devid and/or physpath name of 5009 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 5010 * The GRUB "findroot" command will return the vdev we should boot. 5011 * 5012 * For Sparc, devpath_list consists the physpath name of the booting device 5013 * no matter the rootpool is a single device pool or a mirrored pool. 5014 * e.g. 5015 * "/pci@1f,0/ide@d/disk@0,0:a" 5016 */ 5017int 5018spa_import_rootpool(char *devpath, char *devid) 5019{ 5020 spa_t *spa; 5021 vdev_t *rvd, *bvd, *avd = NULL; 5022 nvlist_t *config, *nvtop; 5023 uint64_t guid, txg; 5024 char *pname; 5025 int error; 5026 5027 /* 5028 * Read the label from the boot device and generate a configuration. 5029 */ 5030 config = spa_generate_rootconf(devpath, devid, &guid); 5031#if defined(_OBP) && defined(_KERNEL) 5032 if (config == NULL) { 5033 if (strstr(devpath, "/iscsi/ssd") != NULL) { 5034 /* iscsi boot */ 5035 get_iscsi_bootpath_phy(devpath); 5036 config = spa_generate_rootconf(devpath, devid, &guid); 5037 } 5038 } 5039#endif 5040 if (config == NULL) { 5041 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 5042 devpath); 5043 return (SET_ERROR(EIO)); 5044 } 5045 5046 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 5047 &pname) == 0); 5048 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 5049 5050 mutex_enter(&spa_namespace_lock); 5051 if ((spa = spa_lookup(pname)) != NULL) { 5052 /* 5053 * Remove the existing root pool from the namespace so that we 5054 * can replace it with the correct config we just read in. 5055 */ 5056 spa_remove(spa); 5057 } 5058 5059 spa = spa_add(pname, config, NULL); 5060 spa->spa_is_root = B_TRUE; 5061 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 5062 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 5063 &spa->spa_ubsync.ub_version) != 0) 5064 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 5065 5066 /* 5067 * Build up a vdev tree based on the boot device's label config. 5068 */ 5069 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5070 &nvtop) == 0); 5071 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5072 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 5073 VDEV_ALLOC_ROOTPOOL); 5074 spa_config_exit(spa, SCL_ALL, FTAG); 5075 if (error) { 5076 mutex_exit(&spa_namespace_lock); 5077 nvlist_free(config); 5078 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 5079 pname); 5080 return (error); 5081 } 5082 5083 /* 5084 * Get the boot vdev. 5085 */ 5086 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 5087 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 5088 (u_longlong_t)guid); 5089 error = SET_ERROR(ENOENT); 5090 goto out; 5091 } 5092 5093 /* 5094 * Determine if there is a better boot device. 5095 */ 5096 avd = bvd; 5097 spa_alt_rootvdev(rvd, &avd, &txg); 5098 if (avd != bvd) { 5099 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 5100 "try booting from '%s'", avd->vdev_path); 5101 error = SET_ERROR(EINVAL); 5102 goto out; 5103 } 5104 5105 /* 5106 * If the boot device is part of a spare vdev then ensure that 5107 * we're booting off the active spare. 5108 */ 5109 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 5110 !bvd->vdev_isspare) { 5111 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 5112 "try booting from '%s'", 5113 bvd->vdev_parent-> 5114 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 5115 error = SET_ERROR(EINVAL); 5116 goto out; 5117 } 5118 5119 error = 0; 5120out: 5121 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5122 vdev_free(rvd); 5123 spa_config_exit(spa, SCL_ALL, FTAG); 5124 mutex_exit(&spa_namespace_lock); 5125 5126 nvlist_free(config); 5127 return (error); 5128} 5129 5130#else /* !illumos */ 5131 5132extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 5133 uint64_t *count); 5134 5135static nvlist_t * 5136spa_generate_rootconf(const char *name) 5137{ 5138 nvlist_t **configs, **tops; 5139 nvlist_t *config; 5140 nvlist_t *best_cfg, *nvtop, *nvroot; 5141 uint64_t *holes; 5142 uint64_t best_txg; 5143 uint64_t nchildren; 5144 uint64_t pgid; 5145 uint64_t count; 5146 uint64_t i; 5147 uint_t nholes; 5148 5149 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 5150 return (NULL); 5151 5152 ASSERT3U(count, !=, 0); 5153 best_txg = 0; 5154 for (i = 0; i < count; i++) { 5155 uint64_t txg; 5156 5157 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 5158 &txg) == 0); 5159 if (txg > best_txg) { 5160 best_txg = txg; 5161 best_cfg = configs[i]; 5162 } 5163 } 5164 5165 nchildren = 1; 5166 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 5167 holes = NULL; 5168 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 5169 &holes, &nholes); 5170 5171 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 5172 for (i = 0; i < nchildren; i++) { 5173 if (i >= count) 5174 break; 5175 if (configs[i] == NULL) 5176 continue; 5177 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 5178 &nvtop) == 0); 5179 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 5180 } 5181 for (i = 0; holes != NULL && i < nholes; i++) { 5182 if (i >= nchildren) 5183 continue; 5184 if (tops[holes[i]] != NULL) 5185 continue; 5186 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 5187 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 5188 VDEV_TYPE_HOLE) == 0); 5189 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 5190 holes[i]) == 0); 5191 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 5192 0) == 0); 5193 } 5194 for (i = 0; i < nchildren; i++) { 5195 if (tops[i] != NULL) 5196 continue; 5197 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 5198 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 5199 VDEV_TYPE_MISSING) == 0); 5200 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 5201 i) == 0); 5202 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 5203 0) == 0); 5204 } 5205 5206 /* 5207 * Create pool config based on the best vdev config. 5208 */ 5209 nvlist_dup(best_cfg, &config, KM_SLEEP); 5210 5211 /* 5212 * Put this pool's top-level vdevs into a root vdev. 5213 */ 5214 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5215 &pgid) == 0); 5216 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5217 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 5218 VDEV_TYPE_ROOT) == 0); 5219 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 5220 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 5221 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 5222 tops, nchildren) == 0); 5223 5224 /* 5225 * Replace the existing vdev_tree with the new root vdev in 5226 * this pool's configuration (remove the old, add the new). 5227 */ 5228 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 5229 5230 /* 5231 * Drop vdev config elements that should not be present at pool level. 5232 */ 5233 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 5234 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 5235 5236 for (i = 0; i < count; i++) 5237 nvlist_free(configs[i]); 5238 kmem_free(configs, count * sizeof(void *)); 5239 for (i = 0; i < nchildren; i++) 5240 nvlist_free(tops[i]); 5241 kmem_free(tops, nchildren * sizeof(void *)); 5242 nvlist_free(nvroot); 5243 return (config); 5244} 5245 5246int 5247spa_import_rootpool(const char *name) 5248{ 5249 spa_t *spa; 5250 vdev_t *rvd, *bvd, *avd = NULL; 5251 nvlist_t *config, *nvtop; 5252 uint64_t txg; 5253 char *pname; 5254 int error; 5255 5256 /* 5257 * Read the label from the boot device and generate a configuration. 5258 */ 5259 config = spa_generate_rootconf(name); 5260 5261 mutex_enter(&spa_namespace_lock); 5262 if (config != NULL) { 5263 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 5264 &pname) == 0 && strcmp(name, pname) == 0); 5265 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 5266 == 0); 5267 5268 if ((spa = spa_lookup(pname)) != NULL) { 5269 /* 5270 * The pool could already be imported, 5271 * e.g., after reboot -r. 5272 */ 5273 if (spa->spa_state == POOL_STATE_ACTIVE) { 5274 mutex_exit(&spa_namespace_lock); 5275 nvlist_free(config); 5276 return (0); 5277 } 5278 5279 /* 5280 * Remove the existing root pool from the namespace so 5281 * that we can replace it with the correct config 5282 * we just read in. 5283 */ 5284 spa_remove(spa); 5285 } 5286 spa = spa_add(pname, config, NULL); 5287 5288 /* 5289 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 5290 * via spa_version(). 5291 */ 5292 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 5293 &spa->spa_ubsync.ub_version) != 0) 5294 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 5295 } else if ((spa = spa_lookup(name)) == NULL) { 5296 mutex_exit(&spa_namespace_lock); 5297 nvlist_free(config); 5298 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 5299 name); 5300 return (EIO); 5301 } else { 5302 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 5303 } 5304 spa->spa_is_root = B_TRUE; 5305 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 5306 5307 /* 5308 * Build up a vdev tree based on the boot device's label config. 5309 */ 5310 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5311 &nvtop) == 0); 5312 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5313 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 5314 VDEV_ALLOC_ROOTPOOL); 5315 spa_config_exit(spa, SCL_ALL, FTAG); 5316 if (error) { 5317 mutex_exit(&spa_namespace_lock); 5318 nvlist_free(config); 5319 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 5320 pname); 5321 return (error); 5322 } 5323 5324 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5325 vdev_free(rvd); 5326 spa_config_exit(spa, SCL_ALL, FTAG); 5327 mutex_exit(&spa_namespace_lock); 5328 5329 nvlist_free(config); 5330 return (0); 5331} 5332 5333#endif /* illumos */ 5334#endif /* _KERNEL */ 5335 5336/* 5337 * Import a non-root pool into the system. 5338 */ 5339int 5340spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 5341{ 5342 spa_t *spa; 5343 char *altroot = NULL; 5344 spa_load_state_t state = SPA_LOAD_IMPORT; 5345 zpool_load_policy_t policy; 5346 uint64_t mode = spa_mode_global; 5347 uint64_t readonly = B_FALSE; 5348 int error; 5349 nvlist_t *nvroot; 5350 nvlist_t **spares, **l2cache; 5351 uint_t nspares, nl2cache; 5352 5353 /* 5354 * If a pool with this name exists, return failure. 5355 */ 5356 mutex_enter(&spa_namespace_lock); 5357 if (spa_lookup(pool) != NULL) { 5358 mutex_exit(&spa_namespace_lock); 5359 return (SET_ERROR(EEXIST)); 5360 } 5361 5362 /* 5363 * Create and initialize the spa structure. 5364 */ 5365 (void) nvlist_lookup_string(props, 5366 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5367 (void) nvlist_lookup_uint64(props, 5368 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 5369 if (readonly) 5370 mode = FREAD; 5371 spa = spa_add(pool, config, altroot); 5372 spa->spa_import_flags = flags; 5373 5374 /* 5375 * Verbatim import - Take a pool and insert it into the namespace 5376 * as if it had been loaded at boot. 5377 */ 5378 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 5379 if (props != NULL) 5380 spa_configfile_set(spa, props, B_FALSE); 5381 5382 spa_write_cachefile(spa, B_FALSE, B_TRUE); 5383 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5384 zfs_dbgmsg("spa_import: verbatim import of %s", pool); 5385 mutex_exit(&spa_namespace_lock); 5386 return (0); 5387 } 5388 5389 spa_activate(spa, mode); 5390 5391 /* 5392 * Don't start async tasks until we know everything is healthy. 5393 */ 5394 spa_async_suspend(spa); 5395 5396 zpool_get_load_policy(config, &policy); 5397 if (policy.zlp_rewind & ZPOOL_DO_REWIND) 5398 state = SPA_LOAD_RECOVER; 5399 5400 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; 5401 5402 if (state != SPA_LOAD_RECOVER) { 5403 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 5404 zfs_dbgmsg("spa_import: importing %s", pool); 5405 } else { 5406 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " 5407 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); 5408 } 5409 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); 5410 5411 /* 5412 * Propagate anything learned while loading the pool and pass it 5413 * back to caller (i.e. rewind info, missing devices, etc). 5414 */ 5415 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5416 spa->spa_load_info) == 0); 5417 5418 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5419 /* 5420 * Toss any existing sparelist, as it doesn't have any validity 5421 * anymore, and conflicts with spa_has_spare(). 5422 */ 5423 if (spa->spa_spares.sav_config) { 5424 nvlist_free(spa->spa_spares.sav_config); 5425 spa->spa_spares.sav_config = NULL; 5426 spa_load_spares(spa); 5427 } 5428 if (spa->spa_l2cache.sav_config) { 5429 nvlist_free(spa->spa_l2cache.sav_config); 5430 spa->spa_l2cache.sav_config = NULL; 5431 spa_load_l2cache(spa); 5432 } 5433 5434 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5435 &nvroot) == 0); 5436 if (error == 0) 5437 error = spa_validate_aux(spa, nvroot, -1ULL, 5438 VDEV_ALLOC_SPARE); 5439 if (error == 0) 5440 error = spa_validate_aux(spa, nvroot, -1ULL, 5441 VDEV_ALLOC_L2CACHE); 5442 spa_config_exit(spa, SCL_ALL, FTAG); 5443 5444 if (props != NULL) 5445 spa_configfile_set(spa, props, B_FALSE); 5446 5447 if (error != 0 || (props && spa_writeable(spa) && 5448 (error = spa_prop_set(spa, props)))) { 5449 spa_unload(spa); 5450 spa_deactivate(spa); 5451 spa_remove(spa); 5452 mutex_exit(&spa_namespace_lock); 5453 return (error); 5454 } 5455 5456 spa_async_resume(spa); 5457 5458 /* 5459 * Override any spares and level 2 cache devices as specified by 5460 * the user, as these may have correct device names/devids, etc. 5461 */ 5462 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 5463 &spares, &nspares) == 0) { 5464 if (spa->spa_spares.sav_config) 5465 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 5466 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 5467 else 5468 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 5469 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5470 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 5471 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 5472 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5473 spa_load_spares(spa); 5474 spa_config_exit(spa, SCL_ALL, FTAG); 5475 spa->spa_spares.sav_sync = B_TRUE; 5476 } 5477 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 5478 &l2cache, &nl2cache) == 0) { 5479 if (spa->spa_l2cache.sav_config) 5480 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 5481 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 5482 else 5483 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 5484 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5485 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 5486 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 5487 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5488 spa_load_l2cache(spa); 5489 spa_config_exit(spa, SCL_ALL, FTAG); 5490 spa->spa_l2cache.sav_sync = B_TRUE; 5491 } 5492 5493 /* 5494 * Check for any removed devices. 5495 */ 5496 if (spa->spa_autoreplace) { 5497 spa_aux_check_removed(&spa->spa_spares); 5498 spa_aux_check_removed(&spa->spa_l2cache); 5499 } 5500 5501 if (spa_writeable(spa)) { 5502 /* 5503 * Update the config cache to include the newly-imported pool. 5504 */ 5505 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5506 } 5507 5508 /* 5509 * It's possible that the pool was expanded while it was exported. 5510 * We kick off an async task to handle this for us. 5511 */ 5512 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 5513 5514 spa_history_log_version(spa, "import"); 5515 5516 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5517 5518 mutex_exit(&spa_namespace_lock); 5519 5520#ifdef __FreeBSD__ 5521#ifdef _KERNEL 5522 zvol_create_minors(pool); 5523#endif 5524#endif 5525 return (0); 5526} 5527 5528nvlist_t * 5529spa_tryimport(nvlist_t *tryconfig) 5530{ 5531 nvlist_t *config = NULL; 5532 char *poolname, *cachefile; 5533 spa_t *spa; 5534 uint64_t state; 5535 int error; 5536 zpool_load_policy_t policy; 5537 5538 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 5539 return (NULL); 5540 5541 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 5542 return (NULL); 5543 5544 /* 5545 * Create and initialize the spa structure. 5546 */ 5547 mutex_enter(&spa_namespace_lock); 5548 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 5549 spa_activate(spa, FREAD); 5550 5551 /* 5552 * Rewind pool if a max txg was provided. 5553 */ 5554 zpool_get_load_policy(spa->spa_config, &policy); 5555 if (policy.zlp_txg != UINT64_MAX) { 5556 spa->spa_load_max_txg = policy.zlp_txg; 5557 spa->spa_extreme_rewind = B_TRUE; 5558 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", 5559 poolname, (longlong_t)policy.zlp_txg); 5560 } else { 5561 zfs_dbgmsg("spa_tryimport: importing %s", poolname); 5562 } 5563 5564 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) 5565 == 0) { 5566 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); 5567 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 5568 } else { 5569 spa->spa_config_source = SPA_CONFIG_SRC_SCAN; 5570 } 5571 5572 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); 5573 5574 /* 5575 * If 'tryconfig' was at least parsable, return the current config. 5576 */ 5577 if (spa->spa_root_vdev != NULL) { 5578 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 5579 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 5580 poolname) == 0); 5581 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5582 state) == 0); 5583 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 5584 spa->spa_uberblock.ub_timestamp) == 0); 5585 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5586 spa->spa_load_info) == 0); 5587 5588 /* 5589 * If the bootfs property exists on this pool then we 5590 * copy it out so that external consumers can tell which 5591 * pools are bootable. 5592 */ 5593 if ((!error || error == EEXIST) && spa->spa_bootfs) { 5594 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5595 5596 /* 5597 * We have to play games with the name since the 5598 * pool was opened as TRYIMPORT_NAME. 5599 */ 5600 if (dsl_dsobj_to_dsname(spa_name(spa), 5601 spa->spa_bootfs, tmpname) == 0) { 5602 char *cp; 5603 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5604 5605 cp = strchr(tmpname, '/'); 5606 if (cp == NULL) { 5607 (void) strlcpy(dsname, tmpname, 5608 MAXPATHLEN); 5609 } else { 5610 (void) snprintf(dsname, MAXPATHLEN, 5611 "%s/%s", poolname, ++cp); 5612 } 5613 VERIFY(nvlist_add_string(config, 5614 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 5615 kmem_free(dsname, MAXPATHLEN); 5616 } 5617 kmem_free(tmpname, MAXPATHLEN); 5618 } 5619 5620 /* 5621 * Add the list of hot spares and level 2 cache devices. 5622 */ 5623 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5624 spa_add_spares(spa, config); 5625 spa_add_l2cache(spa, config); 5626 spa_config_exit(spa, SCL_CONFIG, FTAG); 5627 } 5628 5629 spa_unload(spa); 5630 spa_deactivate(spa); 5631 spa_remove(spa); 5632 mutex_exit(&spa_namespace_lock); 5633 5634 return (config); 5635} 5636 5637/* 5638 * Pool export/destroy 5639 * 5640 * The act of destroying or exporting a pool is very simple. We make sure there 5641 * is no more pending I/O and any references to the pool are gone. Then, we 5642 * update the pool state and sync all the labels to disk, removing the 5643 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 5644 * we don't sync the labels or remove the configuration cache. 5645 */ 5646static int 5647spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 5648 boolean_t force, boolean_t hardforce) 5649{ 5650 spa_t *spa; 5651 5652 if (oldconfig) 5653 *oldconfig = NULL; 5654 5655 if (!(spa_mode_global & FWRITE)) 5656 return (SET_ERROR(EROFS)); 5657 5658 mutex_enter(&spa_namespace_lock); 5659 if ((spa = spa_lookup(pool)) == NULL) { 5660 mutex_exit(&spa_namespace_lock); 5661 return (SET_ERROR(ENOENT)); 5662 } 5663 5664 /* 5665 * Put a hold on the pool, drop the namespace lock, stop async tasks, 5666 * reacquire the namespace lock, and see if we can export. 5667 */ 5668 spa_open_ref(spa, FTAG); 5669 mutex_exit(&spa_namespace_lock); 5670 spa_async_suspend(spa); 5671 mutex_enter(&spa_namespace_lock); 5672 spa_close(spa, FTAG); 5673 5674 /* 5675 * The pool will be in core if it's openable, 5676 * in which case we can modify its state. 5677 */ 5678 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 5679 /* 5680 * Objsets may be open only because they're dirty, so we 5681 * have to force it to sync before checking spa_refcnt. 5682 */ 5683 txg_wait_synced(spa->spa_dsl_pool, 0); 5684 spa_evicting_os_wait(spa); 5685 5686 /* 5687 * A pool cannot be exported or destroyed if there are active 5688 * references. If we are resetting a pool, allow references by 5689 * fault injection handlers. 5690 */ 5691 if (!spa_refcount_zero(spa) || 5692 (spa->spa_inject_ref != 0 && 5693 new_state != POOL_STATE_UNINITIALIZED)) { 5694 spa_async_resume(spa); 5695 mutex_exit(&spa_namespace_lock); 5696 return (SET_ERROR(EBUSY)); 5697 } 5698 5699 /* 5700 * A pool cannot be exported if it has an active shared spare. 5701 * This is to prevent other pools stealing the active spare 5702 * from an exported pool. At user's own will, such pool can 5703 * be forcedly exported. 5704 */ 5705 if (!force && new_state == POOL_STATE_EXPORTED && 5706 spa_has_active_shared_spare(spa)) { 5707 spa_async_resume(spa); 5708 mutex_exit(&spa_namespace_lock); 5709 return (SET_ERROR(EXDEV)); 5710 } 5711 5712 /* 5713 * We want this to be reflected on every label, 5714 * so mark them all dirty. spa_unload() will do the 5715 * final sync that pushes these changes out. 5716 */ 5717 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 5718 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5719 spa->spa_state = new_state; 5720 spa->spa_final_txg = spa_last_synced_txg(spa) + 5721 TXG_DEFER_SIZE + 1; 5722 vdev_config_dirty(spa->spa_root_vdev); 5723 spa_config_exit(spa, SCL_ALL, FTAG); 5724 } 5725 } 5726 5727 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); 5728 5729 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 5730 spa_unload(spa); 5731 spa_deactivate(spa); 5732 } 5733 5734 if (oldconfig && spa->spa_config) 5735 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 5736 5737 if (new_state != POOL_STATE_UNINITIALIZED) { 5738 if (!hardforce) 5739 spa_write_cachefile(spa, B_TRUE, B_TRUE); 5740 spa_remove(spa); 5741 } 5742 mutex_exit(&spa_namespace_lock); 5743 5744 return (0); 5745} 5746 5747/* 5748 * Destroy a storage pool. 5749 */ 5750int 5751spa_destroy(char *pool) 5752{ 5753 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 5754 B_FALSE, B_FALSE)); 5755} 5756 5757/* 5758 * Export a storage pool. 5759 */ 5760int 5761spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 5762 boolean_t hardforce) 5763{ 5764 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 5765 force, hardforce)); 5766} 5767 5768/* 5769 * Similar to spa_export(), this unloads the spa_t without actually removing it 5770 * from the namespace in any way. 5771 */ 5772int 5773spa_reset(char *pool) 5774{ 5775 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 5776 B_FALSE, B_FALSE)); 5777} 5778 5779/* 5780 * ========================================================================== 5781 * Device manipulation 5782 * ========================================================================== 5783 */ 5784 5785/* 5786 * Add a device to a storage pool. 5787 */ 5788int 5789spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 5790{ 5791 uint64_t txg, id; 5792 int error; 5793 vdev_t *rvd = spa->spa_root_vdev; 5794 vdev_t *vd, *tvd; 5795 nvlist_t **spares, **l2cache; 5796 uint_t nspares, nl2cache; 5797 5798 ASSERT(spa_writeable(spa)); 5799 5800 txg = spa_vdev_enter(spa); 5801 5802 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 5803 VDEV_ALLOC_ADD)) != 0) 5804 return (spa_vdev_exit(spa, NULL, txg, error)); 5805 5806 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 5807 5808 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 5809 &nspares) != 0) 5810 nspares = 0; 5811 5812 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 5813 &nl2cache) != 0) 5814 nl2cache = 0; 5815 5816 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 5817 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5818 5819 if (vd->vdev_children != 0 && 5820 (error = vdev_create(vd, txg, B_FALSE)) != 0) 5821 return (spa_vdev_exit(spa, vd, txg, error)); 5822 5823 /* 5824 * We must validate the spares and l2cache devices after checking the 5825 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 5826 */ 5827 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 5828 return (spa_vdev_exit(spa, vd, txg, error)); 5829 5830 /* 5831 * If we are in the middle of a device removal, we can only add 5832 * devices which match the existing devices in the pool. 5833 * If we are in the middle of a removal, or have some indirect 5834 * vdevs, we can not add raidz toplevels. 5835 */ 5836 if (spa->spa_vdev_removal != NULL || 5837 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5838 for (int c = 0; c < vd->vdev_children; c++) { 5839 tvd = vd->vdev_child[c]; 5840 if (spa->spa_vdev_removal != NULL && 5841 tvd->vdev_ashift != 5842 spa->spa_vdev_removal->svr_vdev->vdev_ashift) { 5843 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5844 } 5845 /* Fail if top level vdev is raidz */ 5846 if (tvd->vdev_ops == &vdev_raidz_ops) { 5847 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5848 } 5849 /* 5850 * Need the top level mirror to be 5851 * a mirror of leaf vdevs only 5852 */ 5853 if (tvd->vdev_ops == &vdev_mirror_ops) { 5854 for (uint64_t cid = 0; 5855 cid < tvd->vdev_children; cid++) { 5856 vdev_t *cvd = tvd->vdev_child[cid]; 5857 if (!cvd->vdev_ops->vdev_op_leaf) { 5858 return (spa_vdev_exit(spa, vd, 5859 txg, EINVAL)); 5860 } 5861 } 5862 } 5863 } 5864 } 5865 5866 for (int c = 0; c < vd->vdev_children; c++) { 5867 5868 /* 5869 * Set the vdev id to the first hole, if one exists. 5870 */ 5871 for (id = 0; id < rvd->vdev_children; id++) { 5872 if (rvd->vdev_child[id]->vdev_ishole) { 5873 vdev_free(rvd->vdev_child[id]); 5874 break; 5875 } 5876 } 5877 tvd = vd->vdev_child[c]; 5878 vdev_remove_child(vd, tvd); 5879 tvd->vdev_id = id; 5880 vdev_add_child(rvd, tvd); 5881 vdev_config_dirty(tvd); 5882 } 5883 5884 if (nspares != 0) { 5885 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 5886 ZPOOL_CONFIG_SPARES); 5887 spa_load_spares(spa); 5888 spa->spa_spares.sav_sync = B_TRUE; 5889 } 5890 5891 if (nl2cache != 0) { 5892 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 5893 ZPOOL_CONFIG_L2CACHE); 5894 spa_load_l2cache(spa); 5895 spa->spa_l2cache.sav_sync = B_TRUE; 5896 } 5897 5898 /* 5899 * We have to be careful when adding new vdevs to an existing pool. 5900 * If other threads start allocating from these vdevs before we 5901 * sync the config cache, and we lose power, then upon reboot we may 5902 * fail to open the pool because there are DVAs that the config cache 5903 * can't translate. Therefore, we first add the vdevs without 5904 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 5905 * and then let spa_config_update() initialize the new metaslabs. 5906 * 5907 * spa_load() checks for added-but-not-initialized vdevs, so that 5908 * if we lose power at any point in this sequence, the remaining 5909 * steps will be completed the next time we load the pool. 5910 */ 5911 (void) spa_vdev_exit(spa, vd, txg, 0); 5912 5913 mutex_enter(&spa_namespace_lock); 5914 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5915 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); 5916 mutex_exit(&spa_namespace_lock); 5917 5918 return (0); 5919} 5920 5921/* 5922 * Attach a device to a mirror. The arguments are the path to any device 5923 * in the mirror, and the nvroot for the new device. If the path specifies 5924 * a device that is not mirrored, we automatically insert the mirror vdev. 5925 * 5926 * If 'replacing' is specified, the new device is intended to replace the 5927 * existing device; in this case the two devices are made into their own 5928 * mirror using the 'replacing' vdev, which is functionally identical to 5929 * the mirror vdev (it actually reuses all the same ops) but has a few 5930 * extra rules: you can't attach to it after it's been created, and upon 5931 * completion of resilvering, the first disk (the one being replaced) 5932 * is automatically detached. 5933 */ 5934int 5935spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 5936{ 5937 uint64_t txg, dtl_max_txg; 5938 vdev_t *rvd = spa->spa_root_vdev; 5939 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 5940 vdev_ops_t *pvops; 5941 char *oldvdpath, *newvdpath; 5942 int newvd_isspare; 5943 int error; 5944 5945 ASSERT(spa_writeable(spa)); 5946 5947 txg = spa_vdev_enter(spa); 5948 5949 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 5950 5951 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5952 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 5953 error = (spa_has_checkpoint(spa)) ? 5954 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 5955 return (spa_vdev_exit(spa, NULL, txg, error)); 5956 } 5957 5958 if (spa->spa_vdev_removal != NULL || 5959 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5960 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5961 } 5962 5963 if (oldvd == NULL) 5964 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5965 5966 if (!oldvd->vdev_ops->vdev_op_leaf) 5967 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5968 5969 pvd = oldvd->vdev_parent; 5970 5971 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 5972 VDEV_ALLOC_ATTACH)) != 0) 5973 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5974 5975 if (newrootvd->vdev_children != 1) 5976 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 5977 5978 newvd = newrootvd->vdev_child[0]; 5979 5980 if (!newvd->vdev_ops->vdev_op_leaf) 5981 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 5982 5983 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 5984 return (spa_vdev_exit(spa, newrootvd, txg, error)); 5985 5986 /* 5987 * Spares can't replace logs 5988 */ 5989 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 5990 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5991 5992 if (!replacing) { 5993 /* 5994 * For attach, the only allowable parent is a mirror or the root 5995 * vdev. 5996 */ 5997 if (pvd->vdev_ops != &vdev_mirror_ops && 5998 pvd->vdev_ops != &vdev_root_ops) 5999 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6000 6001 pvops = &vdev_mirror_ops; 6002 } else { 6003 /* 6004 * Active hot spares can only be replaced by inactive hot 6005 * spares. 6006 */ 6007 if (pvd->vdev_ops == &vdev_spare_ops && 6008 oldvd->vdev_isspare && 6009 !spa_has_spare(spa, newvd->vdev_guid)) 6010 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6011 6012 /* 6013 * If the source is a hot spare, and the parent isn't already a 6014 * spare, then we want to create a new hot spare. Otherwise, we 6015 * want to create a replacing vdev. The user is not allowed to 6016 * attach to a spared vdev child unless the 'isspare' state is 6017 * the same (spare replaces spare, non-spare replaces 6018 * non-spare). 6019 */ 6020 if (pvd->vdev_ops == &vdev_replacing_ops && 6021 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 6022 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6023 } else if (pvd->vdev_ops == &vdev_spare_ops && 6024 newvd->vdev_isspare != oldvd->vdev_isspare) { 6025 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6026 } 6027 6028 if (newvd->vdev_isspare) 6029 pvops = &vdev_spare_ops; 6030 else 6031 pvops = &vdev_replacing_ops; 6032 } 6033 6034 /* 6035 * Make sure the new device is big enough. 6036 */ 6037 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 6038 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 6039 6040 /* 6041 * The new device cannot have a higher alignment requirement 6042 * than the top-level vdev. 6043 */ 6044 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 6045 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 6046 6047 /* 6048 * If this is an in-place replacement, update oldvd's path and devid 6049 * to make it distinguishable from newvd, and unopenable from now on. 6050 */ 6051 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 6052 spa_strfree(oldvd->vdev_path); 6053 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 6054 KM_SLEEP); 6055 (void) sprintf(oldvd->vdev_path, "%s/%s", 6056 newvd->vdev_path, "old"); 6057 if (oldvd->vdev_devid != NULL) { 6058 spa_strfree(oldvd->vdev_devid); 6059 oldvd->vdev_devid = NULL; 6060 } 6061 } 6062 6063 /* mark the device being resilvered */ 6064 newvd->vdev_resilver_txg = txg; 6065 6066 /* 6067 * If the parent is not a mirror, or if we're replacing, insert the new 6068 * mirror/replacing/spare vdev above oldvd. 6069 */ 6070 if (pvd->vdev_ops != pvops) 6071 pvd = vdev_add_parent(oldvd, pvops); 6072 6073 ASSERT(pvd->vdev_top->vdev_parent == rvd); 6074 ASSERT(pvd->vdev_ops == pvops); 6075 ASSERT(oldvd->vdev_parent == pvd); 6076 6077 /* 6078 * Extract the new device from its root and add it to pvd. 6079 */ 6080 vdev_remove_child(newrootvd, newvd); 6081 newvd->vdev_id = pvd->vdev_children; 6082 newvd->vdev_crtxg = oldvd->vdev_crtxg; 6083 vdev_add_child(pvd, newvd); 6084 6085 tvd = newvd->vdev_top; 6086 ASSERT(pvd->vdev_top == tvd); 6087 ASSERT(tvd->vdev_parent == rvd); 6088 6089 vdev_config_dirty(tvd); 6090 6091 /* 6092 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 6093 * for any dmu_sync-ed blocks. It will propagate upward when 6094 * spa_vdev_exit() calls vdev_dtl_reassess(). 6095 */ 6096 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 6097 6098 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 6099 dtl_max_txg - TXG_INITIAL); 6100 6101 if (newvd->vdev_isspare) { 6102 spa_spare_activate(newvd); 6103 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); 6104 } 6105 6106 oldvdpath = spa_strdup(oldvd->vdev_path); 6107 newvdpath = spa_strdup(newvd->vdev_path); 6108 newvd_isspare = newvd->vdev_isspare; 6109 6110 /* 6111 * Mark newvd's DTL dirty in this txg. 6112 */ 6113 vdev_dirty(tvd, VDD_DTL, newvd, txg); 6114 6115 /* 6116 * Schedule the resilver to restart in the future. We do this to 6117 * ensure that dmu_sync-ed blocks have been stitched into the 6118 * respective datasets. 6119 */ 6120 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 6121 6122 if (spa->spa_bootfs) 6123 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); 6124 6125 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); 6126 6127 /* 6128 * Commit the config 6129 */ 6130 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 6131 6132 spa_history_log_internal(spa, "vdev attach", NULL, 6133 "%s vdev=%s %s vdev=%s", 6134 replacing && newvd_isspare ? "spare in" : 6135 replacing ? "replace" : "attach", newvdpath, 6136 replacing ? "for" : "to", oldvdpath); 6137 6138 spa_strfree(oldvdpath); 6139 spa_strfree(newvdpath); 6140 6141 return (0); 6142} 6143 6144/* 6145 * Detach a device from a mirror or replacing vdev. 6146 * 6147 * If 'replace_done' is specified, only detach if the parent 6148 * is a replacing vdev. 6149 */ 6150int 6151spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 6152{ 6153 uint64_t txg; 6154 int error; 6155 vdev_t *rvd = spa->spa_root_vdev; 6156 vdev_t *vd, *pvd, *cvd, *tvd; 6157 boolean_t unspare = B_FALSE; 6158 uint64_t unspare_guid = 0; 6159 char *vdpath; 6160 6161 ASSERT(spa_writeable(spa)); 6162 6163 txg = spa_vdev_enter(spa); 6164 6165 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 6166 6167 /* 6168 * Besides being called directly from the userland through the 6169 * ioctl interface, spa_vdev_detach() can be potentially called 6170 * at the end of spa_vdev_resilver_done(). 6171 * 6172 * In the regular case, when we have a checkpoint this shouldn't 6173 * happen as we never empty the DTLs of a vdev during the scrub 6174 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() 6175 * should never get here when we have a checkpoint. 6176 * 6177 * That said, even in a case when we checkpoint the pool exactly 6178 * as spa_vdev_resilver_done() calls this function everything 6179 * should be fine as the resilver will return right away. 6180 */ 6181 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 6182 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 6183 error = (spa_has_checkpoint(spa)) ? 6184 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 6185 return (spa_vdev_exit(spa, NULL, txg, error)); 6186 } 6187 6188 if (vd == NULL) 6189 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 6190 6191 if (!vd->vdev_ops->vdev_op_leaf) 6192 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6193 6194 pvd = vd->vdev_parent; 6195 6196 /* 6197 * If the parent/child relationship is not as expected, don't do it. 6198 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 6199 * vdev that's replacing B with C. The user's intent in replacing 6200 * is to go from M(A,B) to M(A,C). If the user decides to cancel 6201 * the replace by detaching C, the expected behavior is to end up 6202 * M(A,B). But suppose that right after deciding to detach C, 6203 * the replacement of B completes. We would have M(A,C), and then 6204 * ask to detach C, which would leave us with just A -- not what 6205 * the user wanted. To prevent this, we make sure that the 6206 * parent/child relationship hasn't changed -- in this example, 6207 * that C's parent is still the replacing vdev R. 6208 */ 6209 if (pvd->vdev_guid != pguid && pguid != 0) 6210 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 6211 6212 /* 6213 * Only 'replacing' or 'spare' vdevs can be replaced. 6214 */ 6215 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 6216 pvd->vdev_ops != &vdev_spare_ops) 6217 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6218 6219 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 6220 spa_version(spa) >= SPA_VERSION_SPARES); 6221 6222 /* 6223 * Only mirror, replacing, and spare vdevs support detach. 6224 */ 6225 if (pvd->vdev_ops != &vdev_replacing_ops && 6226 pvd->vdev_ops != &vdev_mirror_ops && 6227 pvd->vdev_ops != &vdev_spare_ops) 6228 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6229 6230 /* 6231 * If this device has the only valid copy of some data, 6232 * we cannot safely detach it. 6233 */ 6234 if (vdev_dtl_required(vd)) 6235 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 6236 6237 ASSERT(pvd->vdev_children >= 2); 6238 6239 /* 6240 * If we are detaching the second disk from a replacing vdev, then 6241 * check to see if we changed the original vdev's path to have "/old" 6242 * at the end in spa_vdev_attach(). If so, undo that change now. 6243 */ 6244 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 6245 vd->vdev_path != NULL) { 6246 size_t len = strlen(vd->vdev_path); 6247 6248 for (int c = 0; c < pvd->vdev_children; c++) { 6249 cvd = pvd->vdev_child[c]; 6250 6251 if (cvd == vd || cvd->vdev_path == NULL) 6252 continue; 6253 6254 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 6255 strcmp(cvd->vdev_path + len, "/old") == 0) { 6256 spa_strfree(cvd->vdev_path); 6257 cvd->vdev_path = spa_strdup(vd->vdev_path); 6258 break; 6259 } 6260 } 6261 } 6262 6263 /* 6264 * If we are detaching the original disk from a spare, then it implies 6265 * that the spare should become a real disk, and be removed from the 6266 * active spare list for the pool. 6267 */ 6268 if (pvd->vdev_ops == &vdev_spare_ops && 6269 vd->vdev_id == 0 && 6270 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 6271 unspare = B_TRUE; 6272 6273 /* 6274 * Erase the disk labels so the disk can be used for other things. 6275 * This must be done after all other error cases are handled, 6276 * but before we disembowel vd (so we can still do I/O to it). 6277 * But if we can't do it, don't treat the error as fatal -- 6278 * it may be that the unwritability of the disk is the reason 6279 * it's being detached! 6280 */ 6281 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 6282 6283 /* 6284 * Remove vd from its parent and compact the parent's children. 6285 */ 6286 vdev_remove_child(pvd, vd); 6287 vdev_compact_children(pvd); 6288 6289 /* 6290 * Remember one of the remaining children so we can get tvd below. 6291 */ 6292 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 6293 6294 /* 6295 * If we need to remove the remaining child from the list of hot spares, 6296 * do it now, marking the vdev as no longer a spare in the process. 6297 * We must do this before vdev_remove_parent(), because that can 6298 * change the GUID if it creates a new toplevel GUID. For a similar 6299 * reason, we must remove the spare now, in the same txg as the detach; 6300 * otherwise someone could attach a new sibling, change the GUID, and 6301 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 6302 */ 6303 if (unspare) { 6304 ASSERT(cvd->vdev_isspare); 6305 spa_spare_remove(cvd); 6306 unspare_guid = cvd->vdev_guid; 6307 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 6308 cvd->vdev_unspare = B_TRUE; 6309 } 6310 6311 /* 6312 * If the parent mirror/replacing vdev only has one child, 6313 * the parent is no longer needed. Remove it from the tree. 6314 */ 6315 if (pvd->vdev_children == 1) { 6316 if (pvd->vdev_ops == &vdev_spare_ops) 6317 cvd->vdev_unspare = B_FALSE; 6318 vdev_remove_parent(cvd); 6319 } 6320 6321 6322 /* 6323 * We don't set tvd until now because the parent we just removed 6324 * may have been the previous top-level vdev. 6325 */ 6326 tvd = cvd->vdev_top; 6327 ASSERT(tvd->vdev_parent == rvd); 6328 6329 /* 6330 * Reevaluate the parent vdev state. 6331 */ 6332 vdev_propagate_state(cvd); 6333 6334 /* 6335 * If the 'autoexpand' property is set on the pool then automatically 6336 * try to expand the size of the pool. For example if the device we 6337 * just detached was smaller than the others, it may be possible to 6338 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 6339 * first so that we can obtain the updated sizes of the leaf vdevs. 6340 */ 6341 if (spa->spa_autoexpand) { 6342 vdev_reopen(tvd); 6343 vdev_expand(tvd, txg); 6344 } 6345 6346 vdev_config_dirty(tvd); 6347 6348 /* 6349 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 6350 * vd->vdev_detached is set and free vd's DTL object in syncing context. 6351 * But first make sure we're not on any *other* txg's DTL list, to 6352 * prevent vd from being accessed after it's freed. 6353 */ 6354 vdpath = spa_strdup(vd->vdev_path); 6355 for (int t = 0; t < TXG_SIZE; t++) 6356 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 6357 vd->vdev_detached = B_TRUE; 6358 vdev_dirty(tvd, VDD_DTL, vd, txg); 6359 6360 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); 6361 6362 /* hang on to the spa before we release the lock */ 6363 spa_open_ref(spa, FTAG); 6364 6365 error = spa_vdev_exit(spa, vd, txg, 0); 6366 6367 spa_history_log_internal(spa, "detach", NULL, 6368 "vdev=%s", vdpath); 6369 spa_strfree(vdpath); 6370 6371 /* 6372 * If this was the removal of the original device in a hot spare vdev, 6373 * then we want to go through and remove the device from the hot spare 6374 * list of every other pool. 6375 */ 6376 if (unspare) { 6377 spa_t *altspa = NULL; 6378 6379 mutex_enter(&spa_namespace_lock); 6380 while ((altspa = spa_next(altspa)) != NULL) { 6381 if (altspa->spa_state != POOL_STATE_ACTIVE || 6382 altspa == spa) 6383 continue; 6384 6385 spa_open_ref(altspa, FTAG); 6386 mutex_exit(&spa_namespace_lock); 6387 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 6388 mutex_enter(&spa_namespace_lock); 6389 spa_close(altspa, FTAG); 6390 } 6391 mutex_exit(&spa_namespace_lock); 6392 6393 /* search the rest of the vdevs for spares to remove */ 6394 spa_vdev_resilver_done(spa); 6395 } 6396 6397 /* all done with the spa; OK to release */ 6398 mutex_enter(&spa_namespace_lock); 6399 spa_close(spa, FTAG); 6400 mutex_exit(&spa_namespace_lock); 6401 6402 return (error); 6403} 6404 6405/* 6406 * Split a set of devices from their mirrors, and create a new pool from them. 6407 */ 6408int 6409spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 6410 nvlist_t *props, boolean_t exp) 6411{ 6412 int error = 0; 6413 uint64_t txg, *glist; 6414 spa_t *newspa; 6415 uint_t c, children, lastlog; 6416 nvlist_t **child, *nvl, *tmp; 6417 dmu_tx_t *tx; 6418 char *altroot = NULL; 6419 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 6420 boolean_t activate_slog; 6421 6422 ASSERT(spa_writeable(spa)); 6423 6424 txg = spa_vdev_enter(spa); 6425 6426 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 6427 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 6428 error = (spa_has_checkpoint(spa)) ? 6429 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 6430 return (spa_vdev_exit(spa, NULL, txg, error)); 6431 } 6432 6433 /* clear the log and flush everything up to now */ 6434 activate_slog = spa_passivate_log(spa); 6435 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6436 error = spa_reset_logs(spa); 6437 txg = spa_vdev_config_enter(spa); 6438 6439 if (activate_slog) 6440 spa_activate_log(spa); 6441 6442 if (error != 0) 6443 return (spa_vdev_exit(spa, NULL, txg, error)); 6444 6445 /* check new spa name before going any further */ 6446 if (spa_lookup(newname) != NULL) 6447 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 6448 6449 /* 6450 * scan through all the children to ensure they're all mirrors 6451 */ 6452 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 6453 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 6454 &children) != 0) 6455 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6456 6457 /* first, check to ensure we've got the right child count */ 6458 rvd = spa->spa_root_vdev; 6459 lastlog = 0; 6460 for (c = 0; c < rvd->vdev_children; c++) { 6461 vdev_t *vd = rvd->vdev_child[c]; 6462 6463 /* don't count the holes & logs as children */ 6464 if (vd->vdev_islog || !vdev_is_concrete(vd)) { 6465 if (lastlog == 0) 6466 lastlog = c; 6467 continue; 6468 } 6469 6470 lastlog = 0; 6471 } 6472 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 6473 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6474 6475 /* next, ensure no spare or cache devices are part of the split */ 6476 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 6477 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 6478 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6479 6480 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 6481 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 6482 6483 /* then, loop over each vdev and validate it */ 6484 for (c = 0; c < children; c++) { 6485 uint64_t is_hole = 0; 6486 6487 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 6488 &is_hole); 6489 6490 if (is_hole != 0) { 6491 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 6492 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 6493 continue; 6494 } else { 6495 error = SET_ERROR(EINVAL); 6496 break; 6497 } 6498 } 6499 6500 /* which disk is going to be split? */ 6501 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 6502 &glist[c]) != 0) { 6503 error = SET_ERROR(EINVAL); 6504 break; 6505 } 6506 6507 /* look it up in the spa */ 6508 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 6509 if (vml[c] == NULL) { 6510 error = SET_ERROR(ENODEV); 6511 break; 6512 } 6513 6514 /* make sure there's nothing stopping the split */ 6515 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 6516 vml[c]->vdev_islog || 6517 !vdev_is_concrete(vml[c]) || 6518 vml[c]->vdev_isspare || 6519 vml[c]->vdev_isl2cache || 6520 !vdev_writeable(vml[c]) || 6521 vml[c]->vdev_children != 0 || 6522 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 6523 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 6524 error = SET_ERROR(EINVAL); 6525 break; 6526 } 6527 6528 if (vdev_dtl_required(vml[c])) { 6529 error = SET_ERROR(EBUSY); 6530 break; 6531 } 6532 6533 /* we need certain info from the top level */ 6534 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 6535 vml[c]->vdev_top->vdev_ms_array) == 0); 6536 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 6537 vml[c]->vdev_top->vdev_ms_shift) == 0); 6538 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 6539 vml[c]->vdev_top->vdev_asize) == 0); 6540 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 6541 vml[c]->vdev_top->vdev_ashift) == 0); 6542 6543 /* transfer per-vdev ZAPs */ 6544 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 6545 VERIFY0(nvlist_add_uint64(child[c], 6546 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 6547 6548 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 6549 VERIFY0(nvlist_add_uint64(child[c], 6550 ZPOOL_CONFIG_VDEV_TOP_ZAP, 6551 vml[c]->vdev_parent->vdev_top_zap)); 6552 } 6553 6554 if (error != 0) { 6555 kmem_free(vml, children * sizeof (vdev_t *)); 6556 kmem_free(glist, children * sizeof (uint64_t)); 6557 return (spa_vdev_exit(spa, NULL, txg, error)); 6558 } 6559 6560 /* stop writers from using the disks */ 6561 for (c = 0; c < children; c++) { 6562 if (vml[c] != NULL) 6563 vml[c]->vdev_offline = B_TRUE; 6564 } 6565 vdev_reopen(spa->spa_root_vdev); 6566 6567 /* 6568 * Temporarily record the splitting vdevs in the spa config. This 6569 * will disappear once the config is regenerated. 6570 */ 6571 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6572 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 6573 glist, children) == 0); 6574 kmem_free(glist, children * sizeof (uint64_t)); 6575 6576 mutex_enter(&spa->spa_props_lock); 6577 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 6578 nvl) == 0); 6579 mutex_exit(&spa->spa_props_lock); 6580 spa->spa_config_splitting = nvl; 6581 vdev_config_dirty(spa->spa_root_vdev); 6582 6583 /* configure and create the new pool */ 6584 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 6585 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 6586 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 6587 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6588 spa_version(spa)) == 0); 6589 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 6590 spa->spa_config_txg) == 0); 6591 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 6592 spa_generate_guid(NULL)) == 0); 6593 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 6594 (void) nvlist_lookup_string(props, 6595 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 6596 6597 /* add the new pool to the namespace */ 6598 newspa = spa_add(newname, config, altroot); 6599 newspa->spa_avz_action = AVZ_ACTION_REBUILD; 6600 newspa->spa_config_txg = spa->spa_config_txg; 6601 spa_set_log_state(newspa, SPA_LOG_CLEAR); 6602 6603 /* release the spa config lock, retaining the namespace lock */ 6604 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6605 6606 if (zio_injection_enabled) 6607 zio_handle_panic_injection(spa, FTAG, 1); 6608 6609 spa_activate(newspa, spa_mode_global); 6610 spa_async_suspend(newspa); 6611 6612#ifndef illumos 6613 /* mark that we are creating new spa by splitting */ 6614 newspa->spa_splitting_newspa = B_TRUE; 6615#endif 6616 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; 6617 6618 /* create the new pool from the disks of the original pool */ 6619 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); 6620#ifndef illumos 6621 newspa->spa_splitting_newspa = B_FALSE; 6622#endif 6623 if (error) 6624 goto out; 6625 6626 /* if that worked, generate a real config for the new pool */ 6627 if (newspa->spa_root_vdev != NULL) { 6628 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 6629 NV_UNIQUE_NAME, KM_SLEEP) == 0); 6630 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 6631 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 6632 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 6633 B_TRUE)); 6634 } 6635 6636 /* set the props */ 6637 if (props != NULL) { 6638 spa_configfile_set(newspa, props, B_FALSE); 6639 error = spa_prop_set(newspa, props); 6640 if (error) 6641 goto out; 6642 } 6643 6644 /* flush everything */ 6645 txg = spa_vdev_config_enter(newspa); 6646 vdev_config_dirty(newspa->spa_root_vdev); 6647 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 6648 6649 if (zio_injection_enabled) 6650 zio_handle_panic_injection(spa, FTAG, 2); 6651 6652 spa_async_resume(newspa); 6653 6654 /* finally, update the original pool's config */ 6655 txg = spa_vdev_config_enter(spa); 6656 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 6657 error = dmu_tx_assign(tx, TXG_WAIT); 6658 if (error != 0) 6659 dmu_tx_abort(tx); 6660 for (c = 0; c < children; c++) { 6661 if (vml[c] != NULL) { 6662 vdev_split(vml[c]); 6663 if (error == 0) 6664 spa_history_log_internal(spa, "detach", tx, 6665 "vdev=%s", vml[c]->vdev_path); 6666 6667 vdev_free(vml[c]); 6668 } 6669 } 6670 spa->spa_avz_action = AVZ_ACTION_REBUILD; 6671 vdev_config_dirty(spa->spa_root_vdev); 6672 spa->spa_config_splitting = NULL; 6673 nvlist_free(nvl); 6674 if (error == 0) 6675 dmu_tx_commit(tx); 6676 (void) spa_vdev_exit(spa, NULL, txg, 0); 6677 6678 if (zio_injection_enabled) 6679 zio_handle_panic_injection(spa, FTAG, 3); 6680 6681 /* split is complete; log a history record */ 6682 spa_history_log_internal(newspa, "split", NULL, 6683 "from pool %s", spa_name(spa)); 6684 6685 kmem_free(vml, children * sizeof (vdev_t *)); 6686 6687 /* if we're not going to mount the filesystems in userland, export */ 6688 if (exp) 6689 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 6690 B_FALSE, B_FALSE); 6691 6692 return (error); 6693 6694out: 6695 spa_unload(newspa); 6696 spa_deactivate(newspa); 6697 spa_remove(newspa); 6698 6699 txg = spa_vdev_config_enter(spa); 6700 6701 /* re-online all offlined disks */ 6702 for (c = 0; c < children; c++) { 6703 if (vml[c] != NULL) 6704 vml[c]->vdev_offline = B_FALSE; 6705 } 6706 vdev_reopen(spa->spa_root_vdev); 6707 6708 nvlist_free(spa->spa_config_splitting); 6709 spa->spa_config_splitting = NULL; 6710 (void) spa_vdev_exit(spa, NULL, txg, error); 6711 6712 kmem_free(vml, children * sizeof (vdev_t *)); 6713 return (error); 6714} 6715 6716/* 6717 * Find any device that's done replacing, or a vdev marked 'unspare' that's 6718 * currently spared, so we can detach it. 6719 */ 6720static vdev_t * 6721spa_vdev_resilver_done_hunt(vdev_t *vd) 6722{ 6723 vdev_t *newvd, *oldvd; 6724 6725 for (int c = 0; c < vd->vdev_children; c++) { 6726 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 6727 if (oldvd != NULL) 6728 return (oldvd); 6729 } 6730 6731 /* 6732 * Check for a completed replacement. We always consider the first 6733 * vdev in the list to be the oldest vdev, and the last one to be 6734 * the newest (see spa_vdev_attach() for how that works). In 6735 * the case where the newest vdev is faulted, we will not automatically 6736 * remove it after a resilver completes. This is OK as it will require 6737 * user intervention to determine which disk the admin wishes to keep. 6738 */ 6739 if (vd->vdev_ops == &vdev_replacing_ops) { 6740 ASSERT(vd->vdev_children > 1); 6741 6742 newvd = vd->vdev_child[vd->vdev_children - 1]; 6743 oldvd = vd->vdev_child[0]; 6744 6745 if (vdev_dtl_empty(newvd, DTL_MISSING) && 6746 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6747 !vdev_dtl_required(oldvd)) 6748 return (oldvd); 6749 } 6750 6751 /* 6752 * Check for a completed resilver with the 'unspare' flag set. 6753 */ 6754 if (vd->vdev_ops == &vdev_spare_ops) { 6755 vdev_t *first = vd->vdev_child[0]; 6756 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 6757 6758 if (last->vdev_unspare) { 6759 oldvd = first; 6760 newvd = last; 6761 } else if (first->vdev_unspare) { 6762 oldvd = last; 6763 newvd = first; 6764 } else { 6765 oldvd = NULL; 6766 } 6767 6768 if (oldvd != NULL && 6769 vdev_dtl_empty(newvd, DTL_MISSING) && 6770 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6771 !vdev_dtl_required(oldvd)) 6772 return (oldvd); 6773 6774 /* 6775 * If there are more than two spares attached to a disk, 6776 * and those spares are not required, then we want to 6777 * attempt to free them up now so that they can be used 6778 * by other pools. Once we're back down to a single 6779 * disk+spare, we stop removing them. 6780 */ 6781 if (vd->vdev_children > 2) { 6782 newvd = vd->vdev_child[1]; 6783 6784 if (newvd->vdev_isspare && last->vdev_isspare && 6785 vdev_dtl_empty(last, DTL_MISSING) && 6786 vdev_dtl_empty(last, DTL_OUTAGE) && 6787 !vdev_dtl_required(newvd)) 6788 return (newvd); 6789 } 6790 } 6791 6792 return (NULL); 6793} 6794 6795static void 6796spa_vdev_resilver_done(spa_t *spa) 6797{ 6798 vdev_t *vd, *pvd, *ppvd; 6799 uint64_t guid, sguid, pguid, ppguid; 6800 6801 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6802 6803 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 6804 pvd = vd->vdev_parent; 6805 ppvd = pvd->vdev_parent; 6806 guid = vd->vdev_guid; 6807 pguid = pvd->vdev_guid; 6808 ppguid = ppvd->vdev_guid; 6809 sguid = 0; 6810 /* 6811 * If we have just finished replacing a hot spared device, then 6812 * we need to detach the parent's first child (the original hot 6813 * spare) as well. 6814 */ 6815 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 6816 ppvd->vdev_children == 2) { 6817 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 6818 sguid = ppvd->vdev_child[1]->vdev_guid; 6819 } 6820 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 6821 6822 spa_config_exit(spa, SCL_ALL, FTAG); 6823 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 6824 return; 6825 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 6826 return; 6827 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6828 } 6829 6830 spa_config_exit(spa, SCL_ALL, FTAG); 6831} 6832 6833/* 6834 * Update the stored path or FRU for this vdev. 6835 */ 6836int 6837spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 6838 boolean_t ispath) 6839{ 6840 vdev_t *vd; 6841 boolean_t sync = B_FALSE; 6842 6843 ASSERT(spa_writeable(spa)); 6844 6845 spa_vdev_state_enter(spa, SCL_ALL); 6846 6847 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 6848 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 6849 6850 if (!vd->vdev_ops->vdev_op_leaf) 6851 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 6852 6853 if (ispath) { 6854 if (strcmp(value, vd->vdev_path) != 0) { 6855 spa_strfree(vd->vdev_path); 6856 vd->vdev_path = spa_strdup(value); 6857 sync = B_TRUE; 6858 } 6859 } else { 6860 if (vd->vdev_fru == NULL) { 6861 vd->vdev_fru = spa_strdup(value); 6862 sync = B_TRUE; 6863 } else if (strcmp(value, vd->vdev_fru) != 0) { 6864 spa_strfree(vd->vdev_fru); 6865 vd->vdev_fru = spa_strdup(value); 6866 sync = B_TRUE; 6867 } 6868 } 6869 6870 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 6871} 6872 6873int 6874spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 6875{ 6876 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 6877} 6878 6879int 6880spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 6881{ 6882 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 6883} 6884 6885/* 6886 * ========================================================================== 6887 * SPA Scanning 6888 * ========================================================================== 6889 */ 6890int 6891spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) 6892{ 6893 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6894 6895 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6896 return (SET_ERROR(EBUSY)); 6897 6898 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); 6899} 6900 6901int 6902spa_scan_stop(spa_t *spa) 6903{ 6904 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6905 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6906 return (SET_ERROR(EBUSY)); 6907 return (dsl_scan_cancel(spa->spa_dsl_pool)); 6908} 6909 6910int 6911spa_scan(spa_t *spa, pool_scan_func_t func) 6912{ 6913 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6914 6915 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 6916 return (SET_ERROR(ENOTSUP)); 6917 6918 /* 6919 * If a resilver was requested, but there is no DTL on a 6920 * writeable leaf device, we have nothing to do. 6921 */ 6922 if (func == POOL_SCAN_RESILVER && 6923 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 6924 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 6925 return (0); 6926 } 6927 6928 return (dsl_scan(spa->spa_dsl_pool, func)); 6929} 6930 6931/* 6932 * ========================================================================== 6933 * SPA async task processing 6934 * ========================================================================== 6935 */ 6936 6937static void 6938spa_async_remove(spa_t *spa, vdev_t *vd) 6939{ 6940 if (vd->vdev_remove_wanted) { 6941 vd->vdev_remove_wanted = B_FALSE; 6942 vd->vdev_delayed_close = B_FALSE; 6943 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 6944 6945 /* 6946 * We want to clear the stats, but we don't want to do a full 6947 * vdev_clear() as that will cause us to throw away 6948 * degraded/faulted state as well as attempt to reopen the 6949 * device, all of which is a waste. 6950 */ 6951 vd->vdev_stat.vs_read_errors = 0; 6952 vd->vdev_stat.vs_write_errors = 0; 6953 vd->vdev_stat.vs_checksum_errors = 0; 6954 6955 vdev_state_dirty(vd->vdev_top); 6956 /* Tell userspace that the vdev is gone. */ 6957 zfs_post_remove(spa, vd); 6958 } 6959 6960 for (int c = 0; c < vd->vdev_children; c++) 6961 spa_async_remove(spa, vd->vdev_child[c]); 6962} 6963 6964static void 6965spa_async_probe(spa_t *spa, vdev_t *vd) 6966{ 6967 if (vd->vdev_probe_wanted) { 6968 vd->vdev_probe_wanted = B_FALSE; 6969 vdev_reopen(vd); /* vdev_open() does the actual probe */ 6970 } 6971 6972 for (int c = 0; c < vd->vdev_children; c++) 6973 spa_async_probe(spa, vd->vdev_child[c]); 6974} 6975 6976static void 6977spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6978{ 6979 sysevent_id_t eid; 6980 nvlist_t *attr; 6981 char *physpath; 6982 6983 if (!spa->spa_autoexpand) 6984 return; 6985 6986 for (int c = 0; c < vd->vdev_children; c++) { 6987 vdev_t *cvd = vd->vdev_child[c]; 6988 spa_async_autoexpand(spa, cvd); 6989 } 6990 6991 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6992 return; 6993 6994 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6995 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6996 6997 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6998 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6999 7000 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 7001 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 7002 7003 nvlist_free(attr); 7004 kmem_free(physpath, MAXPATHLEN); 7005} 7006 7007static void 7008spa_async_thread(void *arg) 7009{ 7010 spa_t *spa = (spa_t *)arg; 7011 int tasks; 7012 7013 ASSERT(spa->spa_sync_on); 7014 7015 mutex_enter(&spa->spa_async_lock); 7016 tasks = spa->spa_async_tasks; 7017 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 7018 mutex_exit(&spa->spa_async_lock); 7019 7020 /* 7021 * See if the config needs to be updated. 7022 */ 7023 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 7024 uint64_t old_space, new_space; 7025 7026 mutex_enter(&spa_namespace_lock); 7027 old_space = metaslab_class_get_space(spa_normal_class(spa)); 7028 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 7029 new_space = metaslab_class_get_space(spa_normal_class(spa)); 7030 mutex_exit(&spa_namespace_lock); 7031 7032 /* 7033 * If the pool grew as a result of the config update, 7034 * then log an internal history event. 7035 */ 7036 if (new_space != old_space) { 7037 spa_history_log_internal(spa, "vdev online", NULL, 7038 "pool '%s' size: %llu(+%llu)", 7039 spa_name(spa), new_space, new_space - old_space); 7040 } 7041 } 7042 7043 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 7044 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7045 spa_async_autoexpand(spa, spa->spa_root_vdev); 7046 spa_config_exit(spa, SCL_CONFIG, FTAG); 7047 } 7048 7049 /* 7050 * See if any devices need to be probed. 7051 */ 7052 if (tasks & SPA_ASYNC_PROBE) { 7053 spa_vdev_state_enter(spa, SCL_NONE); 7054 spa_async_probe(spa, spa->spa_root_vdev); 7055 (void) spa_vdev_state_exit(spa, NULL, 0); 7056 } 7057 7058 /* 7059 * If any devices are done replacing, detach them. 7060 */ 7061 if (tasks & SPA_ASYNC_RESILVER_DONE) 7062 spa_vdev_resilver_done(spa); 7063 7064 /* 7065 * Kick off a resilver. 7066 */ 7067 if (tasks & SPA_ASYNC_RESILVER) 7068 dsl_resilver_restart(spa->spa_dsl_pool, 0); 7069 7070 /* 7071 * Let the world know that we're done. 7072 */ 7073 mutex_enter(&spa->spa_async_lock); 7074 spa->spa_async_thread = NULL; 7075 cv_broadcast(&spa->spa_async_cv); 7076 mutex_exit(&spa->spa_async_lock); 7077 thread_exit(); 7078} 7079 7080static void 7081spa_async_thread_vd(void *arg) 7082{ 7083 spa_t *spa = arg; 7084 int tasks; 7085 7086 mutex_enter(&spa->spa_async_lock); 7087 tasks = spa->spa_async_tasks; 7088retry: 7089 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 7090 mutex_exit(&spa->spa_async_lock); 7091 7092 /* 7093 * See if any devices need to be marked REMOVED. 7094 */ 7095 if (tasks & SPA_ASYNC_REMOVE) { 7096 spa_vdev_state_enter(spa, SCL_NONE); 7097 spa_async_remove(spa, spa->spa_root_vdev); 7098 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 7099 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 7100 for (int i = 0; i < spa->spa_spares.sav_count; i++) 7101 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 7102 (void) spa_vdev_state_exit(spa, NULL, 0); 7103 } 7104 7105 /* 7106 * Let the world know that we're done. 7107 */ 7108 mutex_enter(&spa->spa_async_lock); 7109 tasks = spa->spa_async_tasks; 7110 if ((tasks & SPA_ASYNC_REMOVE) != 0) 7111 goto retry; 7112 spa->spa_async_thread_vd = NULL; 7113 cv_broadcast(&spa->spa_async_cv); 7114 mutex_exit(&spa->spa_async_lock); 7115 thread_exit(); 7116} 7117 7118void 7119spa_async_suspend(spa_t *spa) 7120{ 7121 mutex_enter(&spa->spa_async_lock); 7122 spa->spa_async_suspended++; 7123 while (spa->spa_async_thread != NULL || 7124 spa->spa_async_thread_vd != NULL) 7125 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 7126 mutex_exit(&spa->spa_async_lock); 7127 7128 spa_vdev_remove_suspend(spa); 7129 7130 zthr_t *condense_thread = spa->spa_condense_zthr; 7131 if (condense_thread != NULL && zthr_isrunning(condense_thread)) 7132 VERIFY0(zthr_cancel(condense_thread)); 7133 7134 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; 7135 if (discard_thread != NULL && zthr_isrunning(discard_thread)) 7136 VERIFY0(zthr_cancel(discard_thread)); 7137} 7138 7139void 7140spa_async_resume(spa_t *spa) 7141{ 7142 mutex_enter(&spa->spa_async_lock); 7143 ASSERT(spa->spa_async_suspended != 0); 7144 spa->spa_async_suspended--; 7145 mutex_exit(&spa->spa_async_lock); 7146 spa_restart_removal(spa); 7147 7148 zthr_t *condense_thread = spa->spa_condense_zthr; 7149 if (condense_thread != NULL && !zthr_isrunning(condense_thread)) 7150 zthr_resume(condense_thread); 7151 7152 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; 7153 if (discard_thread != NULL && !zthr_isrunning(discard_thread)) 7154 zthr_resume(discard_thread); 7155} 7156 7157static boolean_t 7158spa_async_tasks_pending(spa_t *spa) 7159{ 7160 uint_t non_config_tasks; 7161 uint_t config_task; 7162 boolean_t config_task_suspended; 7163 7164 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 7165 SPA_ASYNC_REMOVE); 7166 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 7167 if (spa->spa_ccw_fail_time == 0) { 7168 config_task_suspended = B_FALSE; 7169 } else { 7170 config_task_suspended = 7171 (gethrtime() - spa->spa_ccw_fail_time) < 7172 (zfs_ccw_retry_interval * NANOSEC); 7173 } 7174 7175 return (non_config_tasks || (config_task && !config_task_suspended)); 7176} 7177 7178static void 7179spa_async_dispatch(spa_t *spa) 7180{ 7181 mutex_enter(&spa->spa_async_lock); 7182 if (spa_async_tasks_pending(spa) && 7183 !spa->spa_async_suspended && 7184 spa->spa_async_thread == NULL && 7185 rootdir != NULL) 7186 spa->spa_async_thread = thread_create(NULL, 0, 7187 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 7188 mutex_exit(&spa->spa_async_lock); 7189} 7190 7191static void 7192spa_async_dispatch_vd(spa_t *spa) 7193{ 7194 mutex_enter(&spa->spa_async_lock); 7195 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 7196 !spa->spa_async_suspended && 7197 spa->spa_async_thread_vd == NULL && 7198 rootdir != NULL) 7199 spa->spa_async_thread_vd = thread_create(NULL, 0, 7200 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 7201 mutex_exit(&spa->spa_async_lock); 7202} 7203 7204void 7205spa_async_request(spa_t *spa, int task) 7206{ 7207 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 7208 mutex_enter(&spa->spa_async_lock); 7209 spa->spa_async_tasks |= task; 7210 mutex_exit(&spa->spa_async_lock); 7211 spa_async_dispatch_vd(spa); 7212} 7213 7214/* 7215 * ========================================================================== 7216 * SPA syncing routines 7217 * ========================================================================== 7218 */ 7219 7220static int 7221bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 7222{ 7223 bpobj_t *bpo = arg; 7224 bpobj_enqueue(bpo, bp, tx); 7225 return (0); 7226} 7227 7228static int 7229spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 7230{ 7231 zio_t *zio = arg; 7232 7233 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 7234 BP_GET_PSIZE(bp), zio->io_flags)); 7235 return (0); 7236} 7237 7238/* 7239 * Note: this simple function is not inlined to make it easier to dtrace the 7240 * amount of time spent syncing frees. 7241 */ 7242static void 7243spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 7244{ 7245 zio_t *zio = zio_root(spa, NULL, NULL, 0); 7246 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 7247 VERIFY(zio_wait(zio) == 0); 7248} 7249 7250/* 7251 * Note: this simple function is not inlined to make it easier to dtrace the 7252 * amount of time spent syncing deferred frees. 7253 */ 7254static void 7255spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 7256{ 7257 zio_t *zio = zio_root(spa, NULL, NULL, 0); 7258 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 7259 spa_free_sync_cb, zio, tx), ==, 0); 7260 VERIFY0(zio_wait(zio)); 7261} 7262 7263 7264static void 7265spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 7266{ 7267 char *packed = NULL; 7268 size_t bufsize; 7269 size_t nvsize = 0; 7270 dmu_buf_t *db; 7271 7272 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 7273 7274 /* 7275 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 7276 * information. This avoids the dmu_buf_will_dirty() path and 7277 * saves us a pre-read to get data we don't actually care about. 7278 */ 7279 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 7280 packed = kmem_alloc(bufsize, KM_SLEEP); 7281 7282 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 7283 KM_SLEEP) == 0); 7284 bzero(packed + nvsize, bufsize - nvsize); 7285 7286 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 7287 7288 kmem_free(packed, bufsize); 7289 7290 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 7291 dmu_buf_will_dirty(db, tx); 7292 *(uint64_t *)db->db_data = nvsize; 7293 dmu_buf_rele(db, FTAG); 7294} 7295 7296static void 7297spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 7298 const char *config, const char *entry) 7299{ 7300 nvlist_t *nvroot; 7301 nvlist_t **list; 7302 int i; 7303 7304 if (!sav->sav_sync) 7305 return; 7306 7307 /* 7308 * Update the MOS nvlist describing the list of available devices. 7309 * spa_validate_aux() will have already made sure this nvlist is 7310 * valid and the vdevs are labeled appropriately. 7311 */ 7312 if (sav->sav_object == 0) { 7313 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 7314 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 7315 sizeof (uint64_t), tx); 7316 VERIFY(zap_update(spa->spa_meta_objset, 7317 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 7318 &sav->sav_object, tx) == 0); 7319 } 7320 7321 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 7322 if (sav->sav_count == 0) { 7323 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 7324 } else { 7325 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 7326 for (i = 0; i < sav->sav_count; i++) 7327 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 7328 B_FALSE, VDEV_CONFIG_L2CACHE); 7329 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 7330 sav->sav_count) == 0); 7331 for (i = 0; i < sav->sav_count; i++) 7332 nvlist_free(list[i]); 7333 kmem_free(list, sav->sav_count * sizeof (void *)); 7334 } 7335 7336 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 7337 nvlist_free(nvroot); 7338 7339 sav->sav_sync = B_FALSE; 7340} 7341 7342/* 7343 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 7344 * The all-vdev ZAP must be empty. 7345 */ 7346static void 7347spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 7348{ 7349 spa_t *spa = vd->vdev_spa; 7350 if (vd->vdev_top_zap != 0) { 7351 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 7352 vd->vdev_top_zap, tx)); 7353 } 7354 if (vd->vdev_leaf_zap != 0) { 7355 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 7356 vd->vdev_leaf_zap, tx)); 7357 } 7358 for (uint64_t i = 0; i < vd->vdev_children; i++) { 7359 spa_avz_build(vd->vdev_child[i], avz, tx); 7360 } 7361} 7362 7363static void 7364spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 7365{ 7366 nvlist_t *config; 7367 7368 /* 7369 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 7370 * its config may not be dirty but we still need to build per-vdev ZAPs. 7371 * Similarly, if the pool is being assembled (e.g. after a split), we 7372 * need to rebuild the AVZ although the config may not be dirty. 7373 */ 7374 if (list_is_empty(&spa->spa_config_dirty_list) && 7375 spa->spa_avz_action == AVZ_ACTION_NONE) 7376 return; 7377 7378 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7379 7380 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 7381 spa->spa_avz_action == AVZ_ACTION_INITIALIZE || 7382 spa->spa_all_vdev_zaps != 0); 7383 7384 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 7385 /* Make and build the new AVZ */ 7386 uint64_t new_avz = zap_create(spa->spa_meta_objset, 7387 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 7388 spa_avz_build(spa->spa_root_vdev, new_avz, tx); 7389 7390 /* Diff old AVZ with new one */ 7391 zap_cursor_t zc; 7392 zap_attribute_t za; 7393 7394 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7395 spa->spa_all_vdev_zaps); 7396 zap_cursor_retrieve(&zc, &za) == 0; 7397 zap_cursor_advance(&zc)) { 7398 uint64_t vdzap = za.za_first_integer; 7399 if (zap_lookup_int(spa->spa_meta_objset, new_avz, 7400 vdzap) == ENOENT) { 7401 /* 7402 * ZAP is listed in old AVZ but not in new one; 7403 * destroy it 7404 */ 7405 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 7406 tx)); 7407 } 7408 } 7409 7410 zap_cursor_fini(&zc); 7411 7412 /* Destroy the old AVZ */ 7413 VERIFY0(zap_destroy(spa->spa_meta_objset, 7414 spa->spa_all_vdev_zaps, tx)); 7415 7416 /* Replace the old AVZ in the dir obj with the new one */ 7417 VERIFY0(zap_update(spa->spa_meta_objset, 7418 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 7419 sizeof (new_avz), 1, &new_avz, tx)); 7420 7421 spa->spa_all_vdev_zaps = new_avz; 7422 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 7423 zap_cursor_t zc; 7424 zap_attribute_t za; 7425 7426 /* Walk through the AVZ and destroy all listed ZAPs */ 7427 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7428 spa->spa_all_vdev_zaps); 7429 zap_cursor_retrieve(&zc, &za) == 0; 7430 zap_cursor_advance(&zc)) { 7431 uint64_t zap = za.za_first_integer; 7432 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 7433 } 7434 7435 zap_cursor_fini(&zc); 7436 7437 /* Destroy and unlink the AVZ itself */ 7438 VERIFY0(zap_destroy(spa->spa_meta_objset, 7439 spa->spa_all_vdev_zaps, tx)); 7440 VERIFY0(zap_remove(spa->spa_meta_objset, 7441 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 7442 spa->spa_all_vdev_zaps = 0; 7443 } 7444 7445 if (spa->spa_all_vdev_zaps == 0) { 7446 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 7447 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 7448 DMU_POOL_VDEV_ZAP_MAP, tx); 7449 } 7450 spa->spa_avz_action = AVZ_ACTION_NONE; 7451 7452 /* Create ZAPs for vdevs that don't have them. */ 7453 vdev_construct_zaps(spa->spa_root_vdev, tx); 7454 7455 config = spa_config_generate(spa, spa->spa_root_vdev, 7456 dmu_tx_get_txg(tx), B_FALSE); 7457 7458 /* 7459 * If we're upgrading the spa version then make sure that 7460 * the config object gets updated with the correct version. 7461 */ 7462 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 7463 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 7464 spa->spa_uberblock.ub_version); 7465 7466 spa_config_exit(spa, SCL_STATE, FTAG); 7467 7468 nvlist_free(spa->spa_config_syncing); 7469 spa->spa_config_syncing = config; 7470 7471 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 7472} 7473 7474static void 7475spa_sync_version(void *arg, dmu_tx_t *tx) 7476{ 7477 uint64_t *versionp = arg; 7478 uint64_t version = *versionp; 7479 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7480 7481 /* 7482 * Setting the version is special cased when first creating the pool. 7483 */ 7484 ASSERT(tx->tx_txg != TXG_INITIAL); 7485 7486 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 7487 ASSERT(version >= spa_version(spa)); 7488 7489 spa->spa_uberblock.ub_version = version; 7490 vdev_config_dirty(spa->spa_root_vdev); 7491 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 7492} 7493 7494/* 7495 * Set zpool properties. 7496 */ 7497static void 7498spa_sync_props(void *arg, dmu_tx_t *tx) 7499{ 7500 nvlist_t *nvp = arg; 7501 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7502 objset_t *mos = spa->spa_meta_objset; 7503 nvpair_t *elem = NULL; 7504 7505 mutex_enter(&spa->spa_props_lock); 7506 7507 while ((elem = nvlist_next_nvpair(nvp, elem))) { 7508 uint64_t intval; 7509 char *strval, *fname; 7510 zpool_prop_t prop; 7511 const char *propname; 7512 zprop_type_t proptype; 7513 spa_feature_t fid; 7514 7515 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 7516 case ZPOOL_PROP_INVAL: 7517 /* 7518 * We checked this earlier in spa_prop_validate(). 7519 */ 7520 ASSERT(zpool_prop_feature(nvpair_name(elem))); 7521 7522 fname = strchr(nvpair_name(elem), '@') + 1; 7523 VERIFY0(zfeature_lookup_name(fname, &fid)); 7524 7525 spa_feature_enable(spa, fid, tx); 7526 spa_history_log_internal(spa, "set", tx, 7527 "%s=enabled", nvpair_name(elem)); 7528 break; 7529 7530 case ZPOOL_PROP_VERSION: 7531 intval = fnvpair_value_uint64(elem); 7532 /* 7533 * The version is synced seperatly before other 7534 * properties and should be correct by now. 7535 */ 7536 ASSERT3U(spa_version(spa), >=, intval); 7537 break; 7538 7539 case ZPOOL_PROP_ALTROOT: 7540 /* 7541 * 'altroot' is a non-persistent property. It should 7542 * have been set temporarily at creation or import time. 7543 */ 7544 ASSERT(spa->spa_root != NULL); 7545 break; 7546 7547 case ZPOOL_PROP_READONLY: 7548 case ZPOOL_PROP_CACHEFILE: 7549 /* 7550 * 'readonly' and 'cachefile' are also non-persisitent 7551 * properties. 7552 */ 7553 break; 7554 case ZPOOL_PROP_COMMENT: 7555 strval = fnvpair_value_string(elem); 7556 if (spa->spa_comment != NULL) 7557 spa_strfree(spa->spa_comment); 7558 spa->spa_comment = spa_strdup(strval); 7559 /* 7560 * We need to dirty the configuration on all the vdevs 7561 * so that their labels get updated. It's unnecessary 7562 * to do this for pool creation since the vdev's 7563 * configuratoin has already been dirtied. 7564 */ 7565 if (tx->tx_txg != TXG_INITIAL) 7566 vdev_config_dirty(spa->spa_root_vdev); 7567 spa_history_log_internal(spa, "set", tx, 7568 "%s=%s", nvpair_name(elem), strval); 7569 break; 7570 default: 7571 /* 7572 * Set pool property values in the poolprops mos object. 7573 */ 7574 if (spa->spa_pool_props_object == 0) { 7575 spa->spa_pool_props_object = 7576 zap_create_link(mos, DMU_OT_POOL_PROPS, 7577 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 7578 tx); 7579 } 7580 7581 /* normalize the property name */ 7582 propname = zpool_prop_to_name(prop); 7583 proptype = zpool_prop_get_type(prop); 7584 7585 if (nvpair_type(elem) == DATA_TYPE_STRING) { 7586 ASSERT(proptype == PROP_TYPE_STRING); 7587 strval = fnvpair_value_string(elem); 7588 VERIFY0(zap_update(mos, 7589 spa->spa_pool_props_object, propname, 7590 1, strlen(strval) + 1, strval, tx)); 7591 spa_history_log_internal(spa, "set", tx, 7592 "%s=%s", nvpair_name(elem), strval); 7593 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 7594 intval = fnvpair_value_uint64(elem); 7595 7596 if (proptype == PROP_TYPE_INDEX) { 7597 const char *unused; 7598 VERIFY0(zpool_prop_index_to_string( 7599 prop, intval, &unused)); 7600 } 7601 VERIFY0(zap_update(mos, 7602 spa->spa_pool_props_object, propname, 7603 8, 1, &intval, tx)); 7604 spa_history_log_internal(spa, "set", tx, 7605 "%s=%lld", nvpair_name(elem), intval); 7606 } else { 7607 ASSERT(0); /* not allowed */ 7608 } 7609 7610 switch (prop) { 7611 case ZPOOL_PROP_DELEGATION: 7612 spa->spa_delegation = intval; 7613 break; 7614 case ZPOOL_PROP_BOOTFS: 7615 spa->spa_bootfs = intval; 7616 break; 7617 case ZPOOL_PROP_FAILUREMODE: 7618 spa->spa_failmode = intval; 7619 break; 7620 case ZPOOL_PROP_AUTOEXPAND: 7621 spa->spa_autoexpand = intval; 7622 if (tx->tx_txg != TXG_INITIAL) 7623 spa_async_request(spa, 7624 SPA_ASYNC_AUTOEXPAND); 7625 break; 7626 case ZPOOL_PROP_DEDUPDITTO: 7627 spa->spa_dedup_ditto = intval; 7628 break; 7629 default: 7630 break; 7631 } 7632 } 7633 7634 } 7635 7636 mutex_exit(&spa->spa_props_lock); 7637} 7638 7639/* 7640 * Perform one-time upgrade on-disk changes. spa_version() does not 7641 * reflect the new version this txg, so there must be no changes this 7642 * txg to anything that the upgrade code depends on after it executes. 7643 * Therefore this must be called after dsl_pool_sync() does the sync 7644 * tasks. 7645 */ 7646static void 7647spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 7648{ 7649 dsl_pool_t *dp = spa->spa_dsl_pool; 7650 7651 ASSERT(spa->spa_sync_pass == 1); 7652 7653 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 7654 7655 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 7656 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 7657 dsl_pool_create_origin(dp, tx); 7658 7659 /* Keeping the origin open increases spa_minref */ 7660 spa->spa_minref += 3; 7661 } 7662 7663 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 7664 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 7665 dsl_pool_upgrade_clones(dp, tx); 7666 } 7667 7668 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 7669 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 7670 dsl_pool_upgrade_dir_clones(dp, tx); 7671 7672 /* Keeping the freedir open increases spa_minref */ 7673 spa->spa_minref += 3; 7674 } 7675 7676 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 7677 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7678 spa_feature_create_zap_objects(spa, tx); 7679 } 7680 7681 /* 7682 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 7683 * when possibility to use lz4 compression for metadata was added 7684 * Old pools that have this feature enabled must be upgraded to have 7685 * this feature active 7686 */ 7687 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7688 boolean_t lz4_en = spa_feature_is_enabled(spa, 7689 SPA_FEATURE_LZ4_COMPRESS); 7690 boolean_t lz4_ac = spa_feature_is_active(spa, 7691 SPA_FEATURE_LZ4_COMPRESS); 7692 7693 if (lz4_en && !lz4_ac) 7694 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 7695 } 7696 7697 /* 7698 * If we haven't written the salt, do so now. Note that the 7699 * feature may not be activated yet, but that's fine since 7700 * the presence of this ZAP entry is backwards compatible. 7701 */ 7702 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 7703 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 7704 VERIFY0(zap_add(spa->spa_meta_objset, 7705 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 7706 sizeof (spa->spa_cksum_salt.zcs_bytes), 7707 spa->spa_cksum_salt.zcs_bytes, tx)); 7708 } 7709 7710 rrw_exit(&dp->dp_config_rwlock, FTAG); 7711} 7712 7713static void 7714vdev_indirect_state_sync_verify(vdev_t *vd) 7715{ 7716 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 7717 vdev_indirect_births_t *vib = vd->vdev_indirect_births; 7718 7719 if (vd->vdev_ops == &vdev_indirect_ops) { 7720 ASSERT(vim != NULL); 7721 ASSERT(vib != NULL); 7722 } 7723 7724 if (vdev_obsolete_sm_object(vd) != 0) { 7725 ASSERT(vd->vdev_obsolete_sm != NULL); 7726 ASSERT(vd->vdev_removing || 7727 vd->vdev_ops == &vdev_indirect_ops); 7728 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); 7729 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); 7730 7731 ASSERT3U(vdev_obsolete_sm_object(vd), ==, 7732 space_map_object(vd->vdev_obsolete_sm)); 7733 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, 7734 space_map_allocated(vd->vdev_obsolete_sm)); 7735 } 7736 ASSERT(vd->vdev_obsolete_segments != NULL); 7737 7738 /* 7739 * Since frees / remaps to an indirect vdev can only 7740 * happen in syncing context, the obsolete segments 7741 * tree must be empty when we start syncing. 7742 */ 7743 ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); 7744} 7745 7746/* 7747 * Sync the specified transaction group. New blocks may be dirtied as 7748 * part of the process, so we iterate until it converges. 7749 */ 7750void 7751spa_sync(spa_t *spa, uint64_t txg) 7752{ 7753 dsl_pool_t *dp = spa->spa_dsl_pool; 7754 objset_t *mos = spa->spa_meta_objset; 7755 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 7756 vdev_t *rvd = spa->spa_root_vdev; 7757 vdev_t *vd; 7758 dmu_tx_t *tx; 7759 int error; 7760 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 7761 zfs_vdev_queue_depth_pct / 100; 7762 7763 VERIFY(spa_writeable(spa)); 7764 7765 /* 7766 * Wait for i/os issued in open context that need to complete 7767 * before this txg syncs. 7768 */ 7769 VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK])); 7770 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0); 7771 7772 /* 7773 * Lock out configuration changes. 7774 */ 7775 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7776 7777 spa->spa_syncing_txg = txg; 7778 spa->spa_sync_pass = 0; 7779 7780 for (int i = 0; i < spa->spa_alloc_count; i++) { 7781 mutex_enter(&spa->spa_alloc_locks[i]); 7782 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i])); 7783 mutex_exit(&spa->spa_alloc_locks[i]); 7784 } 7785 7786 /* 7787 * If there are any pending vdev state changes, convert them 7788 * into config changes that go out with this transaction group. 7789 */ 7790 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7791 while (list_head(&spa->spa_state_dirty_list) != NULL) { 7792 /* 7793 * We need the write lock here because, for aux vdevs, 7794 * calling vdev_config_dirty() modifies sav_config. 7795 * This is ugly and will become unnecessary when we 7796 * eliminate the aux vdev wart by integrating all vdevs 7797 * into the root vdev tree. 7798 */ 7799 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7800 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 7801 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 7802 vdev_state_clean(vd); 7803 vdev_config_dirty(vd); 7804 } 7805 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7806 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 7807 } 7808 spa_config_exit(spa, SCL_STATE, FTAG); 7809 7810 tx = dmu_tx_create_assigned(dp, txg); 7811 7812 spa->spa_sync_starttime = gethrtime(); 7813#ifdef illumos 7814 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 7815 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 7816#else /* !illumos */ 7817#ifdef _KERNEL 7818 callout_schedule(&spa->spa_deadman_cycid, 7819 hz * spa->spa_deadman_synctime / NANOSEC); 7820#endif 7821#endif /* illumos */ 7822 7823 /* 7824 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 7825 * set spa_deflate if we have no raid-z vdevs. 7826 */ 7827 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 7828 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 7829 int i; 7830 7831 for (i = 0; i < rvd->vdev_children; i++) { 7832 vd = rvd->vdev_child[i]; 7833 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 7834 break; 7835 } 7836 if (i == rvd->vdev_children) { 7837 spa->spa_deflate = TRUE; 7838 VERIFY(0 == zap_add(spa->spa_meta_objset, 7839 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 7840 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 7841 } 7842 } 7843 7844 /* 7845 * Set the top-level vdev's max queue depth. Evaluate each 7846 * top-level's async write queue depth in case it changed. 7847 * The max queue depth will not change in the middle of syncing 7848 * out this txg. 7849 */ 7850 uint64_t slots_per_allocator = 0; 7851 for (int c = 0; c < rvd->vdev_children; c++) { 7852 vdev_t *tvd = rvd->vdev_child[c]; 7853 metaslab_group_t *mg = tvd->vdev_mg; 7854 7855 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 7856 !metaslab_group_initialized(mg)) 7857 continue; 7858 7859 /* 7860 * It is safe to do a lock-free check here because only async 7861 * allocations look at mg_max_alloc_queue_depth, and async 7862 * allocations all happen from spa_sync(). 7863 */ 7864 for (int i = 0; i < spa->spa_alloc_count; i++) 7865 ASSERT0(refcount_count(&(mg->mg_alloc_queue_depth[i]))); 7866 mg->mg_max_alloc_queue_depth = max_queue_depth; 7867 7868 for (int i = 0; i < spa->spa_alloc_count; i++) { 7869 mg->mg_cur_max_alloc_queue_depth[i] = 7870 zfs_vdev_def_queue_depth; 7871 } 7872 slots_per_allocator += zfs_vdev_def_queue_depth; 7873 } 7874 metaslab_class_t *mc = spa_normal_class(spa); 7875 for (int i = 0; i < spa->spa_alloc_count; i++) { 7876 ASSERT0(refcount_count(&mc->mc_alloc_slots[i])); 7877 mc->mc_alloc_max_slots[i] = slots_per_allocator; 7878 } 7879 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 7880 7881 for (int c = 0; c < rvd->vdev_children; c++) { 7882 vdev_t *vd = rvd->vdev_child[c]; 7883 vdev_indirect_state_sync_verify(vd); 7884 7885 if (vdev_indirect_should_condense(vd)) { 7886 spa_condense_indirect_start_sync(vd, tx); 7887 break; 7888 } 7889 } 7890 7891 /* 7892 * Iterate to convergence. 7893 */ 7894 do { 7895 int pass = ++spa->spa_sync_pass; 7896 7897 spa_sync_config_object(spa, tx); 7898 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 7899 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 7900 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 7901 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 7902 spa_errlog_sync(spa, txg); 7903 dsl_pool_sync(dp, txg); 7904 7905 if (pass < zfs_sync_pass_deferred_free) { 7906 spa_sync_frees(spa, free_bpl, tx); 7907 } else { 7908 /* 7909 * We can not defer frees in pass 1, because 7910 * we sync the deferred frees later in pass 1. 7911 */ 7912 ASSERT3U(pass, >, 1); 7913 bplist_iterate(free_bpl, bpobj_enqueue_cb, 7914 &spa->spa_deferred_bpobj, tx); 7915 } 7916 7917 ddt_sync(spa, txg); 7918 dsl_scan_sync(dp, tx); 7919 7920 if (spa->spa_vdev_removal != NULL) 7921 svr_sync(spa, tx); 7922 7923 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 7924 != NULL) 7925 vdev_sync(vd, txg); 7926 7927 if (pass == 1) { 7928 spa_sync_upgrades(spa, tx); 7929 ASSERT3U(txg, >=, 7930 spa->spa_uberblock.ub_rootbp.blk_birth); 7931 /* 7932 * Note: We need to check if the MOS is dirty 7933 * because we could have marked the MOS dirty 7934 * without updating the uberblock (e.g. if we 7935 * have sync tasks but no dirty user data). We 7936 * need to check the uberblock's rootbp because 7937 * it is updated if we have synced out dirty 7938 * data (though in this case the MOS will most 7939 * likely also be dirty due to second order 7940 * effects, we don't want to rely on that here). 7941 */ 7942 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 7943 !dmu_objset_is_dirty(mos, txg)) { 7944 /* 7945 * Nothing changed on the first pass, 7946 * therefore this TXG is a no-op. Avoid 7947 * syncing deferred frees, so that we 7948 * can keep this TXG as a no-op. 7949 */ 7950 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 7951 txg)); 7952 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7953 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 7954 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, 7955 txg)); 7956 break; 7957 } 7958 spa_sync_deferred_frees(spa, tx); 7959 } 7960 7961 } while (dmu_objset_is_dirty(mos, txg)); 7962 7963 if (!list_is_empty(&spa->spa_config_dirty_list)) { 7964 /* 7965 * Make sure that the number of ZAPs for all the vdevs matches 7966 * the number of ZAPs in the per-vdev ZAP list. This only gets 7967 * called if the config is dirty; otherwise there may be 7968 * outstanding AVZ operations that weren't completed in 7969 * spa_sync_config_object. 7970 */ 7971 uint64_t all_vdev_zap_entry_count; 7972 ASSERT0(zap_count(spa->spa_meta_objset, 7973 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 7974 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 7975 all_vdev_zap_entry_count); 7976 } 7977 7978 if (spa->spa_vdev_removal != NULL) { 7979 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); 7980 } 7981 7982 /* 7983 * Rewrite the vdev configuration (which includes the uberblock) 7984 * to commit the transaction group. 7985 * 7986 * If there are no dirty vdevs, we sync the uberblock to a few 7987 * random top-level vdevs that are known to be visible in the 7988 * config cache (see spa_vdev_add() for a complete description). 7989 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 7990 */ 7991 for (;;) { 7992 /* 7993 * We hold SCL_STATE to prevent vdev open/close/etc. 7994 * while we're attempting to write the vdev labels. 7995 */ 7996 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7997 7998 if (list_is_empty(&spa->spa_config_dirty_list)) { 7999 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; 8000 int svdcount = 0; 8001 int children = rvd->vdev_children; 8002 int c0 = spa_get_random(children); 8003 8004 for (int c = 0; c < children; c++) { 8005 vd = rvd->vdev_child[(c0 + c) % children]; 8006 8007 /* Stop when revisiting the first vdev */ 8008 if (c > 0 && svd[0] == vd) 8009 break; 8010 8011 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 8012 !vdev_is_concrete(vd)) 8013 continue; 8014 8015 svd[svdcount++] = vd; 8016 if (svdcount == SPA_SYNC_MIN_VDEVS) 8017 break; 8018 } 8019 error = vdev_config_sync(svd, svdcount, txg); 8020 } else { 8021 error = vdev_config_sync(rvd->vdev_child, 8022 rvd->vdev_children, txg); 8023 } 8024 8025 if (error == 0) 8026 spa->spa_last_synced_guid = rvd->vdev_guid; 8027 8028 spa_config_exit(spa, SCL_STATE, FTAG); 8029 8030 if (error == 0) 8031 break; 8032 zio_suspend(spa, NULL); 8033 zio_resume_wait(spa); 8034 } 8035 dmu_tx_commit(tx); 8036 8037#ifdef illumos 8038 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 8039#else /* !illumos */ 8040#ifdef _KERNEL 8041 callout_drain(&spa->spa_deadman_cycid); 8042#endif 8043#endif /* illumos */ 8044 8045 /* 8046 * Clear the dirty config list. 8047 */ 8048 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 8049 vdev_config_clean(vd); 8050 8051 /* 8052 * Now that the new config has synced transactionally, 8053 * let it become visible to the config cache. 8054 */ 8055 if (spa->spa_config_syncing != NULL) { 8056 spa_config_set(spa, spa->spa_config_syncing); 8057 spa->spa_config_txg = txg; 8058 spa->spa_config_syncing = NULL; 8059 } 8060 8061 dsl_pool_sync_done(dp, txg); 8062 8063 for (int i = 0; i < spa->spa_alloc_count; i++) { 8064 mutex_enter(&spa->spa_alloc_locks[i]); 8065 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i])); 8066 mutex_exit(&spa->spa_alloc_locks[i]); 8067 } 8068 8069 /* 8070 * Update usable space statistics. 8071 */ 8072 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 8073 vdev_sync_done(vd, txg); 8074 8075 spa_update_dspace(spa); 8076 8077 /* 8078 * It had better be the case that we didn't dirty anything 8079 * since vdev_config_sync(). 8080 */ 8081 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 8082 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 8083 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 8084 8085 while (zfs_pause_spa_sync) 8086 delay(1); 8087 8088 spa->spa_sync_pass = 0; 8089 8090 /* 8091 * Update the last synced uberblock here. We want to do this at 8092 * the end of spa_sync() so that consumers of spa_last_synced_txg() 8093 * will be guaranteed that all the processing associated with 8094 * that txg has been completed. 8095 */ 8096 spa->spa_ubsync = spa->spa_uberblock; 8097 spa_config_exit(spa, SCL_CONFIG, FTAG); 8098 8099 spa_handle_ignored_writes(spa); 8100 8101 /* 8102 * If any async tasks have been requested, kick them off. 8103 */ 8104 spa_async_dispatch(spa); 8105 spa_async_dispatch_vd(spa); 8106} 8107 8108/* 8109 * Sync all pools. We don't want to hold the namespace lock across these 8110 * operations, so we take a reference on the spa_t and drop the lock during the 8111 * sync. 8112 */ 8113void 8114spa_sync_allpools(void) 8115{ 8116 spa_t *spa = NULL; 8117 mutex_enter(&spa_namespace_lock); 8118 while ((spa = spa_next(spa)) != NULL) { 8119 if (spa_state(spa) != POOL_STATE_ACTIVE || 8120 !spa_writeable(spa) || spa_suspended(spa)) 8121 continue; 8122 spa_open_ref(spa, FTAG); 8123 mutex_exit(&spa_namespace_lock); 8124 txg_wait_synced(spa_get_dsl(spa), 0); 8125 mutex_enter(&spa_namespace_lock); 8126 spa_close(spa, FTAG); 8127 } 8128 mutex_exit(&spa_namespace_lock); 8129} 8130 8131/* 8132 * ========================================================================== 8133 * Miscellaneous routines 8134 * ========================================================================== 8135 */ 8136 8137/* 8138 * Remove all pools in the system. 8139 */ 8140void 8141spa_evict_all(void) 8142{ 8143 spa_t *spa; 8144 8145 /* 8146 * Remove all cached state. All pools should be closed now, 8147 * so every spa in the AVL tree should be unreferenced. 8148 */ 8149 mutex_enter(&spa_namespace_lock); 8150 while ((spa = spa_next(NULL)) != NULL) { 8151 /* 8152 * Stop async tasks. The async thread may need to detach 8153 * a device that's been replaced, which requires grabbing 8154 * spa_namespace_lock, so we must drop it here. 8155 */ 8156 spa_open_ref(spa, FTAG); 8157 mutex_exit(&spa_namespace_lock); 8158 spa_async_suspend(spa); 8159 mutex_enter(&spa_namespace_lock); 8160 spa_close(spa, FTAG); 8161 8162 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 8163 spa_unload(spa); 8164 spa_deactivate(spa); 8165 } 8166 spa_remove(spa); 8167 } 8168 mutex_exit(&spa_namespace_lock); 8169} 8170 8171vdev_t * 8172spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 8173{ 8174 vdev_t *vd; 8175 int i; 8176 8177 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 8178 return (vd); 8179 8180 if (aux) { 8181 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 8182 vd = spa->spa_l2cache.sav_vdevs[i]; 8183 if (vd->vdev_guid == guid) 8184 return (vd); 8185 } 8186 8187 for (i = 0; i < spa->spa_spares.sav_count; i++) { 8188 vd = spa->spa_spares.sav_vdevs[i]; 8189 if (vd->vdev_guid == guid) 8190 return (vd); 8191 } 8192 } 8193 8194 return (NULL); 8195} 8196 8197void 8198spa_upgrade(spa_t *spa, uint64_t version) 8199{ 8200 ASSERT(spa_writeable(spa)); 8201 8202 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 8203 8204 /* 8205 * This should only be called for a non-faulted pool, and since a 8206 * future version would result in an unopenable pool, this shouldn't be 8207 * possible. 8208 */ 8209 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 8210 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 8211 8212 spa->spa_uberblock.ub_version = version; 8213 vdev_config_dirty(spa->spa_root_vdev); 8214 8215 spa_config_exit(spa, SCL_ALL, FTAG); 8216 8217 txg_wait_synced(spa_get_dsl(spa), 0); 8218} 8219 8220boolean_t 8221spa_has_spare(spa_t *spa, uint64_t guid) 8222{ 8223 int i; 8224 uint64_t spareguid; 8225 spa_aux_vdev_t *sav = &spa->spa_spares; 8226 8227 for (i = 0; i < sav->sav_count; i++) 8228 if (sav->sav_vdevs[i]->vdev_guid == guid) 8229 return (B_TRUE); 8230 8231 for (i = 0; i < sav->sav_npending; i++) { 8232 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 8233 &spareguid) == 0 && spareguid == guid) 8234 return (B_TRUE); 8235 } 8236 8237 return (B_FALSE); 8238} 8239 8240/* 8241 * Check if a pool has an active shared spare device. 8242 * Note: reference count of an active spare is 2, as a spare and as a replace 8243 */ 8244static boolean_t 8245spa_has_active_shared_spare(spa_t *spa) 8246{ 8247 int i, refcnt; 8248 uint64_t pool; 8249 spa_aux_vdev_t *sav = &spa->spa_spares; 8250 8251 for (i = 0; i < sav->sav_count; i++) { 8252 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 8253 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 8254 refcnt > 2) 8255 return (B_TRUE); 8256 } 8257 8258 return (B_FALSE); 8259} 8260 8261sysevent_t * 8262spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 8263{ 8264 sysevent_t *ev = NULL; 8265#ifdef _KERNEL 8266 sysevent_attr_list_t *attr = NULL; 8267 sysevent_value_t value; 8268 8269 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 8270 SE_SLEEP); 8271 ASSERT(ev != NULL); 8272 8273 value.value_type = SE_DATA_TYPE_STRING; 8274 value.value.sv_string = spa_name(spa); 8275 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 8276 goto done; 8277 8278 value.value_type = SE_DATA_TYPE_UINT64; 8279 value.value.sv_uint64 = spa_guid(spa); 8280 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 8281 goto done; 8282 8283 if (vd) { 8284 value.value_type = SE_DATA_TYPE_UINT64; 8285 value.value.sv_uint64 = vd->vdev_guid; 8286 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 8287 SE_SLEEP) != 0) 8288 goto done; 8289 8290 if (vd->vdev_path) { 8291 value.value_type = SE_DATA_TYPE_STRING; 8292 value.value.sv_string = vd->vdev_path; 8293 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 8294 &value, SE_SLEEP) != 0) 8295 goto done; 8296 } 8297 } 8298 8299 if (hist_nvl != NULL) { 8300 fnvlist_merge((nvlist_t *)attr, hist_nvl); 8301 } 8302 8303 if (sysevent_attach_attributes(ev, attr) != 0) 8304 goto done; 8305 attr = NULL; 8306 8307done: 8308 if (attr) 8309 sysevent_free_attr(attr); 8310 8311#endif 8312 return (ev); 8313} 8314 8315void 8316spa_event_post(sysevent_t *ev) 8317{ 8318#ifdef _KERNEL 8319 sysevent_id_t eid; 8320 8321 (void) log_sysevent(ev, SE_SLEEP, &eid); 8322 sysevent_free(ev); 8323#endif 8324} 8325 8326void 8327spa_event_discard(sysevent_t *ev) 8328{ 8329#ifdef _KERNEL 8330 sysevent_free(ev); 8331#endif 8332} 8333 8334/* 8335 * Post a sysevent corresponding to the given event. The 'name' must be one of 8336 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 8337 * filled in from the spa and (optionally) the vdev and history nvl. This 8338 * doesn't do anything in the userland libzpool, as we don't want consumers to 8339 * misinterpret ztest or zdb as real changes. 8340 */ 8341void 8342spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 8343{ 8344 spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); 8345} 8346