spa.c revision 346676
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 2018 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/vdev_initialize.h> 59#include <sys/metaslab.h> 60#include <sys/metaslab_impl.h> 61#include <sys/uberblock_impl.h> 62#include <sys/txg.h> 63#include <sys/avl.h> 64#include <sys/bpobj.h> 65#include <sys/dmu_traverse.h> 66#include <sys/dmu_objset.h> 67#include <sys/unique.h> 68#include <sys/dsl_pool.h> 69#include <sys/dsl_dataset.h> 70#include <sys/dsl_dir.h> 71#include <sys/dsl_prop.h> 72#include <sys/dsl_synctask.h> 73#include <sys/fs/zfs.h> 74#include <sys/arc.h> 75#include <sys/callb.h> 76#include <sys/spa_boot.h> 77#include <sys/zfs_ioctl.h> 78#include <sys/dsl_scan.h> 79#include <sys/dmu_send.h> 80#include <sys/dsl_destroy.h> 81#include <sys/dsl_userhold.h> 82#include <sys/zfeature.h> 83#include <sys/zvol.h> 84#include <sys/trim_map.h> 85#include <sys/abd.h> 86 87#ifdef _KERNEL 88#include <sys/callb.h> 89#include <sys/cpupart.h> 90#include <sys/zone.h> 91#endif /* _KERNEL */ 92 93#include "zfs_prop.h" 94#include "zfs_comutil.h" 95 96/* Check hostid on import? */ 97static int check_hostid = 1; 98 99/* 100 * The interval, in seconds, at which failed configuration cache file writes 101 * should be retried. 102 */ 103int zfs_ccw_retry_interval = 300; 104 105SYSCTL_DECL(_vfs_zfs); 106SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 107 "Check hostid on import?"); 108TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 109SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 110 &zfs_ccw_retry_interval, 0, 111 "Configuration cache file write, retry after failure, interval (seconds)"); 112 113typedef enum zti_modes { 114 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 115 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 116 ZTI_MODE_NULL, /* don't create a taskq */ 117 ZTI_NMODES 118} zti_modes_t; 119 120#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 121#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 122#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 123 124#define ZTI_N(n) ZTI_P(n, 1) 125#define ZTI_ONE ZTI_N(1) 126 127typedef struct zio_taskq_info { 128 zti_modes_t zti_mode; 129 uint_t zti_value; 130 uint_t zti_count; 131} zio_taskq_info_t; 132 133static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 134 "issue", "issue_high", "intr", "intr_high" 135}; 136 137/* 138 * This table defines the taskq settings for each ZFS I/O type. When 139 * initializing a pool, we use this table to create an appropriately sized 140 * taskq. Some operations are low volume and therefore have a small, static 141 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 142 * macros. Other operations process a large amount of data; the ZTI_BATCH 143 * macro causes us to create a taskq oriented for throughput. Some operations 144 * are so high frequency and short-lived that the taskq itself can become a a 145 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 146 * additional degree of parallelism specified by the number of threads per- 147 * taskq and the number of taskqs; when dispatching an event in this case, the 148 * particular taskq is chosen at random. 149 * 150 * The different taskq priorities are to handle the different contexts (issue 151 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 152 * need to be handled with minimum delay. 153 */ 154const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 155 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 156 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 157 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 158 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 159 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 161 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 162}; 163 164static void spa_sync_version(void *arg, dmu_tx_t *tx); 165static void spa_sync_props(void *arg, dmu_tx_t *tx); 166static boolean_t spa_has_active_shared_spare(spa_t *spa); 167static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport); 168static void spa_vdev_resilver_done(spa_t *spa); 169 170uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 171#ifdef PSRSET_BIND 172id_t zio_taskq_psrset_bind = PS_NONE; 173#endif 174#ifdef SYSDC 175boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 176uint_t zio_taskq_basedc = 80; /* base duty cycle */ 177#endif 178 179boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 180extern int zfs_sync_pass_deferred_free; 181 182/* 183 * Report any spa_load_verify errors found, but do not fail spa_load. 184 * This is used by zdb to analyze non-idle pools. 185 */ 186boolean_t spa_load_verify_dryrun = B_FALSE; 187 188/* 189 * This (illegal) pool name is used when temporarily importing a spa_t in order 190 * to get the vdev stats associated with the imported devices. 191 */ 192#define TRYIMPORT_NAME "$import" 193 194/* 195 * For debugging purposes: print out vdev tree during pool import. 196 */ 197int spa_load_print_vdev_tree = B_FALSE; 198 199/* 200 * A non-zero value for zfs_max_missing_tvds means that we allow importing 201 * pools with missing top-level vdevs. This is strictly intended for advanced 202 * pool recovery cases since missing data is almost inevitable. Pools with 203 * missing devices can only be imported read-only for safety reasons, and their 204 * fail-mode will be automatically set to "continue". 205 * 206 * With 1 missing vdev we should be able to import the pool and mount all 207 * datasets. User data that was not modified after the missing device has been 208 * added should be recoverable. This means that snapshots created prior to the 209 * addition of that device should be completely intact. 210 * 211 * With 2 missing vdevs, some datasets may fail to mount since there are 212 * dataset statistics that are stored as regular metadata. Some data might be 213 * recoverable if those vdevs were added recently. 214 * 215 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries 216 * may be missing entirely. Chances of data recovery are very low. Note that 217 * there are also risks of performing an inadvertent rewind as we might be 218 * missing all the vdevs with the latest uberblocks. 219 */ 220uint64_t zfs_max_missing_tvds = 0; 221 222/* 223 * The parameters below are similar to zfs_max_missing_tvds but are only 224 * intended for a preliminary open of the pool with an untrusted config which 225 * might be incomplete or out-dated. 226 * 227 * We are more tolerant for pools opened from a cachefile since we could have 228 * an out-dated cachefile where a device removal was not registered. 229 * We could have set the limit arbitrarily high but in the case where devices 230 * are really missing we would want to return the proper error codes; we chose 231 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available 232 * and we get a chance to retrieve the trusted config. 233 */ 234uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1; 235 236/* 237 * In the case where config was assembled by scanning device paths (/dev/dsks 238 * by default) we are less tolerant since all the existing devices should have 239 * been detected and we want spa_load to return the right error codes. 240 */ 241uint64_t zfs_max_missing_tvds_scan = 0; 242 243 244SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN, 245 &spa_load_print_vdev_tree, 0, 246 "print out vdev tree during pool import"); 247SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN, 248 &zfs_max_missing_tvds, 0, 249 "allow importing pools with missing top-level vdevs"); 250SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN, 251 &zfs_max_missing_tvds_cachefile, 0, 252 "allow importing pools with missing top-level vdevs in cache file"); 253SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN, 254 &zfs_max_missing_tvds_scan, 0, 255 "allow importing pools with missing top-level vdevs during scan"); 256 257/* 258 * Debugging aid that pauses spa_sync() towards the end. 259 */ 260boolean_t zfs_pause_spa_sync = B_FALSE; 261 262/* 263 * ========================================================================== 264 * SPA properties routines 265 * ========================================================================== 266 */ 267 268/* 269 * Add a (source=src, propname=propval) list to an nvlist. 270 */ 271static void 272spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 273 uint64_t intval, zprop_source_t src) 274{ 275 const char *propname = zpool_prop_to_name(prop); 276 nvlist_t *propval; 277 278 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 279 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 280 281 if (strval != NULL) 282 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 283 else 284 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 285 286 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 287 nvlist_free(propval); 288} 289 290/* 291 * Get property values from the spa configuration. 292 */ 293static void 294spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 295{ 296 vdev_t *rvd = spa->spa_root_vdev; 297 dsl_pool_t *pool = spa->spa_dsl_pool; 298 uint64_t size, alloc, cap, version; 299 zprop_source_t src = ZPROP_SRC_NONE; 300 spa_config_dirent_t *dp; 301 metaslab_class_t *mc = spa_normal_class(spa); 302 303 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 304 305 if (rvd != NULL) { 306 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 307 size = metaslab_class_get_space(spa_normal_class(spa)); 308 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 309 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 310 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 311 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 312 size - alloc, src); 313 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL, 314 spa->spa_checkpoint_info.sci_dspace, src); 315 316 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 317 metaslab_class_fragmentation(mc), src); 318 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 319 metaslab_class_expandable_space(mc), src); 320 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 321 (spa_mode(spa) == FREAD), src); 322 323 cap = (size == 0) ? 0 : (alloc * 100 / size); 324 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 325 326 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 327 ddt_get_pool_dedup_ratio(spa), src); 328 329 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 330 rvd->vdev_state, src); 331 332 version = spa_version(spa); 333 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 334 src = ZPROP_SRC_DEFAULT; 335 else 336 src = ZPROP_SRC_LOCAL; 337 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 338 } 339 340 if (pool != NULL) { 341 /* 342 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 343 * when opening pools before this version freedir will be NULL. 344 */ 345 if (pool->dp_free_dir != NULL) { 346 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 347 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 348 src); 349 } else { 350 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 351 NULL, 0, src); 352 } 353 354 if (pool->dp_leak_dir != NULL) { 355 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 356 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 357 src); 358 } else { 359 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 360 NULL, 0, src); 361 } 362 } 363 364 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 365 366 if (spa->spa_comment != NULL) { 367 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 368 0, ZPROP_SRC_LOCAL); 369 } 370 371 if (spa->spa_root != NULL) 372 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 373 0, ZPROP_SRC_LOCAL); 374 375 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 376 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 377 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 378 } else { 379 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 380 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 381 } 382 383 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 384 if (dp->scd_path == NULL) { 385 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 386 "none", 0, ZPROP_SRC_LOCAL); 387 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 388 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 389 dp->scd_path, 0, ZPROP_SRC_LOCAL); 390 } 391 } 392} 393 394/* 395 * Get zpool property values. 396 */ 397int 398spa_prop_get(spa_t *spa, nvlist_t **nvp) 399{ 400 objset_t *mos = spa->spa_meta_objset; 401 zap_cursor_t zc; 402 zap_attribute_t za; 403 int err; 404 405 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 406 407 mutex_enter(&spa->spa_props_lock); 408 409 /* 410 * Get properties from the spa config. 411 */ 412 spa_prop_get_config(spa, nvp); 413 414 /* If no pool property object, no more prop to get. */ 415 if (mos == NULL || spa->spa_pool_props_object == 0) { 416 mutex_exit(&spa->spa_props_lock); 417 return (0); 418 } 419 420 /* 421 * Get properties from the MOS pool property object. 422 */ 423 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 424 (err = zap_cursor_retrieve(&zc, &za)) == 0; 425 zap_cursor_advance(&zc)) { 426 uint64_t intval = 0; 427 char *strval = NULL; 428 zprop_source_t src = ZPROP_SRC_DEFAULT; 429 zpool_prop_t prop; 430 431 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) 432 continue; 433 434 switch (za.za_integer_length) { 435 case 8: 436 /* integer property */ 437 if (za.za_first_integer != 438 zpool_prop_default_numeric(prop)) 439 src = ZPROP_SRC_LOCAL; 440 441 if (prop == ZPOOL_PROP_BOOTFS) { 442 dsl_pool_t *dp; 443 dsl_dataset_t *ds = NULL; 444 445 dp = spa_get_dsl(spa); 446 dsl_pool_config_enter(dp, FTAG); 447 err = dsl_dataset_hold_obj(dp, 448 za.za_first_integer, FTAG, &ds); 449 if (err != 0) { 450 dsl_pool_config_exit(dp, FTAG); 451 break; 452 } 453 454 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 455 KM_SLEEP); 456 dsl_dataset_name(ds, strval); 457 dsl_dataset_rele(ds, FTAG); 458 dsl_pool_config_exit(dp, FTAG); 459 } else { 460 strval = NULL; 461 intval = za.za_first_integer; 462 } 463 464 spa_prop_add_list(*nvp, prop, strval, intval, src); 465 466 if (strval != NULL) 467 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 468 469 break; 470 471 case 1: 472 /* string property */ 473 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 474 err = zap_lookup(mos, spa->spa_pool_props_object, 475 za.za_name, 1, za.za_num_integers, strval); 476 if (err) { 477 kmem_free(strval, za.za_num_integers); 478 break; 479 } 480 spa_prop_add_list(*nvp, prop, strval, 0, src); 481 kmem_free(strval, za.za_num_integers); 482 break; 483 484 default: 485 break; 486 } 487 } 488 zap_cursor_fini(&zc); 489 mutex_exit(&spa->spa_props_lock); 490out: 491 if (err && err != ENOENT) { 492 nvlist_free(*nvp); 493 *nvp = NULL; 494 return (err); 495 } 496 497 return (0); 498} 499 500/* 501 * Validate the given pool properties nvlist and modify the list 502 * for the property values to be set. 503 */ 504static int 505spa_prop_validate(spa_t *spa, nvlist_t *props) 506{ 507 nvpair_t *elem; 508 int error = 0, reset_bootfs = 0; 509 uint64_t objnum = 0; 510 boolean_t has_feature = B_FALSE; 511 512 elem = NULL; 513 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 514 uint64_t intval; 515 char *strval, *slash, *check, *fname; 516 const char *propname = nvpair_name(elem); 517 zpool_prop_t prop = zpool_name_to_prop(propname); 518 519 switch (prop) { 520 case ZPOOL_PROP_INVAL: 521 if (!zpool_prop_feature(propname)) { 522 error = SET_ERROR(EINVAL); 523 break; 524 } 525 526 /* 527 * Sanitize the input. 528 */ 529 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 530 error = SET_ERROR(EINVAL); 531 break; 532 } 533 534 if (nvpair_value_uint64(elem, &intval) != 0) { 535 error = SET_ERROR(EINVAL); 536 break; 537 } 538 539 if (intval != 0) { 540 error = SET_ERROR(EINVAL); 541 break; 542 } 543 544 fname = strchr(propname, '@') + 1; 545 if (zfeature_lookup_name(fname, NULL) != 0) { 546 error = SET_ERROR(EINVAL); 547 break; 548 } 549 550 has_feature = B_TRUE; 551 break; 552 553 case ZPOOL_PROP_VERSION: 554 error = nvpair_value_uint64(elem, &intval); 555 if (!error && 556 (intval < spa_version(spa) || 557 intval > SPA_VERSION_BEFORE_FEATURES || 558 has_feature)) 559 error = SET_ERROR(EINVAL); 560 break; 561 562 case ZPOOL_PROP_DELEGATION: 563 case ZPOOL_PROP_AUTOREPLACE: 564 case ZPOOL_PROP_LISTSNAPS: 565 case ZPOOL_PROP_AUTOEXPAND: 566 error = nvpair_value_uint64(elem, &intval); 567 if (!error && intval > 1) 568 error = SET_ERROR(EINVAL); 569 break; 570 571 case ZPOOL_PROP_BOOTFS: 572 /* 573 * If the pool version is less than SPA_VERSION_BOOTFS, 574 * or the pool is still being created (version == 0), 575 * the bootfs property cannot be set. 576 */ 577 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 578 error = SET_ERROR(ENOTSUP); 579 break; 580 } 581 582 /* 583 * Make sure the vdev config is bootable 584 */ 585 if (!vdev_is_bootable(spa->spa_root_vdev)) { 586 error = SET_ERROR(ENOTSUP); 587 break; 588 } 589 590 reset_bootfs = 1; 591 592 error = nvpair_value_string(elem, &strval); 593 594 if (!error) { 595 objset_t *os; 596 uint64_t propval; 597 598 if (strval == NULL || strval[0] == '\0') { 599 objnum = zpool_prop_default_numeric( 600 ZPOOL_PROP_BOOTFS); 601 break; 602 } 603 604 error = dmu_objset_hold(strval, FTAG, &os); 605 if (error != 0) 606 break; 607 608 /* 609 * Must be ZPL, and its property settings 610 * must be supported by GRUB (compression 611 * is not gzip, and large blocks are not used). 612 */ 613 614 if (dmu_objset_type(os) != DMU_OST_ZFS) { 615 error = SET_ERROR(ENOTSUP); 616 } else if ((error = 617 dsl_prop_get_int_ds(dmu_objset_ds(os), 618 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 619 &propval)) == 0 && 620 !BOOTFS_COMPRESS_VALID(propval)) { 621 error = SET_ERROR(ENOTSUP); 622 } else { 623 objnum = dmu_objset_id(os); 624 } 625 dmu_objset_rele(os, FTAG); 626 } 627 break; 628 629 case ZPOOL_PROP_FAILUREMODE: 630 error = nvpair_value_uint64(elem, &intval); 631 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 632 intval > ZIO_FAILURE_MODE_PANIC)) 633 error = SET_ERROR(EINVAL); 634 635 /* 636 * This is a special case which only occurs when 637 * the pool has completely failed. This allows 638 * the user to change the in-core failmode property 639 * without syncing it out to disk (I/Os might 640 * currently be blocked). We do this by returning 641 * EIO to the caller (spa_prop_set) to trick it 642 * into thinking we encountered a property validation 643 * error. 644 */ 645 if (!error && spa_suspended(spa)) { 646 spa->spa_failmode = intval; 647 error = SET_ERROR(EIO); 648 } 649 break; 650 651 case ZPOOL_PROP_CACHEFILE: 652 if ((error = nvpair_value_string(elem, &strval)) != 0) 653 break; 654 655 if (strval[0] == '\0') 656 break; 657 658 if (strcmp(strval, "none") == 0) 659 break; 660 661 if (strval[0] != '/') { 662 error = SET_ERROR(EINVAL); 663 break; 664 } 665 666 slash = strrchr(strval, '/'); 667 ASSERT(slash != NULL); 668 669 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 670 strcmp(slash, "/..") == 0) 671 error = SET_ERROR(EINVAL); 672 break; 673 674 case ZPOOL_PROP_COMMENT: 675 if ((error = nvpair_value_string(elem, &strval)) != 0) 676 break; 677 for (check = strval; *check != '\0'; check++) { 678 /* 679 * The kernel doesn't have an easy isprint() 680 * check. For this kernel check, we merely 681 * check ASCII apart from DEL. Fix this if 682 * there is an easy-to-use kernel isprint(). 683 */ 684 if (*check >= 0x7f) { 685 error = SET_ERROR(EINVAL); 686 break; 687 } 688 } 689 if (strlen(strval) > ZPROP_MAX_COMMENT) 690 error = E2BIG; 691 break; 692 693 case ZPOOL_PROP_DEDUPDITTO: 694 if (spa_version(spa) < SPA_VERSION_DEDUP) 695 error = SET_ERROR(ENOTSUP); 696 else 697 error = nvpair_value_uint64(elem, &intval); 698 if (error == 0 && 699 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 700 error = SET_ERROR(EINVAL); 701 break; 702 } 703 704 if (error) 705 break; 706 } 707 708 if (!error && reset_bootfs) { 709 error = nvlist_remove(props, 710 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 711 712 if (!error) { 713 error = nvlist_add_uint64(props, 714 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 715 } 716 } 717 718 return (error); 719} 720 721void 722spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 723{ 724 char *cachefile; 725 spa_config_dirent_t *dp; 726 727 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 728 &cachefile) != 0) 729 return; 730 731 dp = kmem_alloc(sizeof (spa_config_dirent_t), 732 KM_SLEEP); 733 734 if (cachefile[0] == '\0') 735 dp->scd_path = spa_strdup(spa_config_path); 736 else if (strcmp(cachefile, "none") == 0) 737 dp->scd_path = NULL; 738 else 739 dp->scd_path = spa_strdup(cachefile); 740 741 list_insert_head(&spa->spa_config_list, dp); 742 if (need_sync) 743 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 744} 745 746int 747spa_prop_set(spa_t *spa, nvlist_t *nvp) 748{ 749 int error; 750 nvpair_t *elem = NULL; 751 boolean_t need_sync = B_FALSE; 752 753 if ((error = spa_prop_validate(spa, nvp)) != 0) 754 return (error); 755 756 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 757 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 758 759 if (prop == ZPOOL_PROP_CACHEFILE || 760 prop == ZPOOL_PROP_ALTROOT || 761 prop == ZPOOL_PROP_READONLY) 762 continue; 763 764 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { 765 uint64_t ver; 766 767 if (prop == ZPOOL_PROP_VERSION) { 768 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 769 } else { 770 ASSERT(zpool_prop_feature(nvpair_name(elem))); 771 ver = SPA_VERSION_FEATURES; 772 need_sync = B_TRUE; 773 } 774 775 /* Save time if the version is already set. */ 776 if (ver == spa_version(spa)) 777 continue; 778 779 /* 780 * In addition to the pool directory object, we might 781 * create the pool properties object, the features for 782 * read object, the features for write object, or the 783 * feature descriptions object. 784 */ 785 error = dsl_sync_task(spa->spa_name, NULL, 786 spa_sync_version, &ver, 787 6, ZFS_SPACE_CHECK_RESERVED); 788 if (error) 789 return (error); 790 continue; 791 } 792 793 need_sync = B_TRUE; 794 break; 795 } 796 797 if (need_sync) { 798 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 799 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 800 } 801 802 return (0); 803} 804 805/* 806 * If the bootfs property value is dsobj, clear it. 807 */ 808void 809spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 810{ 811 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 812 VERIFY(zap_remove(spa->spa_meta_objset, 813 spa->spa_pool_props_object, 814 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 815 spa->spa_bootfs = 0; 816 } 817} 818 819/*ARGSUSED*/ 820static int 821spa_change_guid_check(void *arg, dmu_tx_t *tx) 822{ 823 uint64_t *newguid = arg; 824 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 825 vdev_t *rvd = spa->spa_root_vdev; 826 uint64_t vdev_state; 827 828 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 829 int error = (spa_has_checkpoint(spa)) ? 830 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 831 return (SET_ERROR(error)); 832 } 833 834 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 835 vdev_state = rvd->vdev_state; 836 spa_config_exit(spa, SCL_STATE, FTAG); 837 838 if (vdev_state != VDEV_STATE_HEALTHY) 839 return (SET_ERROR(ENXIO)); 840 841 ASSERT3U(spa_guid(spa), !=, *newguid); 842 843 return (0); 844} 845 846static void 847spa_change_guid_sync(void *arg, dmu_tx_t *tx) 848{ 849 uint64_t *newguid = arg; 850 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 851 uint64_t oldguid; 852 vdev_t *rvd = spa->spa_root_vdev; 853 854 oldguid = spa_guid(spa); 855 856 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 857 rvd->vdev_guid = *newguid; 858 rvd->vdev_guid_sum += (*newguid - oldguid); 859 vdev_config_dirty(rvd); 860 spa_config_exit(spa, SCL_STATE, FTAG); 861 862 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 863 oldguid, *newguid); 864} 865 866/* 867 * Change the GUID for the pool. This is done so that we can later 868 * re-import a pool built from a clone of our own vdevs. We will modify 869 * the root vdev's guid, our own pool guid, and then mark all of our 870 * vdevs dirty. Note that we must make sure that all our vdevs are 871 * online when we do this, or else any vdevs that weren't present 872 * would be orphaned from our pool. We are also going to issue a 873 * sysevent to update any watchers. 874 */ 875int 876spa_change_guid(spa_t *spa) 877{ 878 int error; 879 uint64_t guid; 880 881 mutex_enter(&spa->spa_vdev_top_lock); 882 mutex_enter(&spa_namespace_lock); 883 guid = spa_generate_guid(NULL); 884 885 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 886 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 887 888 if (error == 0) { 889 spa_write_cachefile(spa, B_FALSE, B_TRUE); 890 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); 891 } 892 893 mutex_exit(&spa_namespace_lock); 894 mutex_exit(&spa->spa_vdev_top_lock); 895 896 return (error); 897} 898 899/* 900 * ========================================================================== 901 * SPA state manipulation (open/create/destroy/import/export) 902 * ========================================================================== 903 */ 904 905static int 906spa_error_entry_compare(const void *a, const void *b) 907{ 908 const spa_error_entry_t *sa = (const spa_error_entry_t *)a; 909 const spa_error_entry_t *sb = (const spa_error_entry_t *)b; 910 int ret; 911 912 ret = memcmp(&sa->se_bookmark, &sb->se_bookmark, 913 sizeof (zbookmark_phys_t)); 914 915 return (AVL_ISIGN(ret)); 916} 917 918/* 919 * Utility function which retrieves copies of the current logs and 920 * re-initializes them in the process. 921 */ 922void 923spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 924{ 925 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 926 927 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 928 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 929 930 avl_create(&spa->spa_errlist_scrub, 931 spa_error_entry_compare, sizeof (spa_error_entry_t), 932 offsetof(spa_error_entry_t, se_avl)); 933 avl_create(&spa->spa_errlist_last, 934 spa_error_entry_compare, sizeof (spa_error_entry_t), 935 offsetof(spa_error_entry_t, se_avl)); 936} 937 938static void 939spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 940{ 941 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 942 enum zti_modes mode = ztip->zti_mode; 943 uint_t value = ztip->zti_value; 944 uint_t count = ztip->zti_count; 945 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 946 char name[32]; 947 uint_t flags = 0; 948 boolean_t batch = B_FALSE; 949 950 if (mode == ZTI_MODE_NULL) { 951 tqs->stqs_count = 0; 952 tqs->stqs_taskq = NULL; 953 return; 954 } 955 956 ASSERT3U(count, >, 0); 957 958 tqs->stqs_count = count; 959 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 960 961 switch (mode) { 962 case ZTI_MODE_FIXED: 963 ASSERT3U(value, >=, 1); 964 value = MAX(value, 1); 965 break; 966 967 case ZTI_MODE_BATCH: 968 batch = B_TRUE; 969 flags |= TASKQ_THREADS_CPU_PCT; 970 value = zio_taskq_batch_pct; 971 break; 972 973 default: 974 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 975 "spa_activate()", 976 zio_type_name[t], zio_taskq_types[q], mode, value); 977 break; 978 } 979 980 for (uint_t i = 0; i < count; i++) { 981 taskq_t *tq; 982 983 if (count > 1) { 984 (void) snprintf(name, sizeof (name), "%s_%s_%u", 985 zio_type_name[t], zio_taskq_types[q], i); 986 } else { 987 (void) snprintf(name, sizeof (name), "%s_%s", 988 zio_type_name[t], zio_taskq_types[q]); 989 } 990 991#ifdef SYSDC 992 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 993 if (batch) 994 flags |= TASKQ_DC_BATCH; 995 996 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 997 spa->spa_proc, zio_taskq_basedc, flags); 998 } else { 999#endif 1000 pri_t pri = maxclsyspri; 1001 /* 1002 * The write issue taskq can be extremely CPU 1003 * intensive. Run it at slightly lower priority 1004 * than the other taskqs. 1005 * FreeBSD notes: 1006 * - numerically higher priorities are lower priorities; 1007 * - if priorities divided by four (RQ_PPQ) are equal 1008 * then a difference between them is insignificant. 1009 */ 1010 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 1011#ifdef illumos 1012 pri--; 1013#else 1014 pri += 4; 1015#endif 1016 1017 tq = taskq_create_proc(name, value, pri, 50, 1018 INT_MAX, spa->spa_proc, flags); 1019#ifdef SYSDC 1020 } 1021#endif 1022 1023 tqs->stqs_taskq[i] = tq; 1024 } 1025} 1026 1027static void 1028spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 1029{ 1030 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1031 1032 if (tqs->stqs_taskq == NULL) { 1033 ASSERT0(tqs->stqs_count); 1034 return; 1035 } 1036 1037 for (uint_t i = 0; i < tqs->stqs_count; i++) { 1038 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 1039 taskq_destroy(tqs->stqs_taskq[i]); 1040 } 1041 1042 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 1043 tqs->stqs_taskq = NULL; 1044} 1045 1046/* 1047 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 1048 * Note that a type may have multiple discrete taskqs to avoid lock contention 1049 * on the taskq itself. In that case we choose which taskq at random by using 1050 * the low bits of gethrtime(). 1051 */ 1052void 1053spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 1054 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 1055{ 1056 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1057 taskq_t *tq; 1058 1059 ASSERT3P(tqs->stqs_taskq, !=, NULL); 1060 ASSERT3U(tqs->stqs_count, !=, 0); 1061 1062 if (tqs->stqs_count == 1) { 1063 tq = tqs->stqs_taskq[0]; 1064 } else { 1065#ifdef _KERNEL 1066 tq = tqs->stqs_taskq[(u_int)(sbinuptime() + curcpu) % 1067 tqs->stqs_count]; 1068#else 1069 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 1070#endif 1071 } 1072 1073 taskq_dispatch_ent(tq, func, arg, flags, ent); 1074} 1075 1076static void 1077spa_create_zio_taskqs(spa_t *spa) 1078{ 1079 for (int t = 0; t < ZIO_TYPES; t++) { 1080 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1081 spa_taskqs_init(spa, t, q); 1082 } 1083 } 1084} 1085 1086#ifdef _KERNEL 1087#ifdef SPA_PROCESS 1088static void 1089spa_thread(void *arg) 1090{ 1091 callb_cpr_t cprinfo; 1092 1093 spa_t *spa = arg; 1094 user_t *pu = PTOU(curproc); 1095 1096 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1097 spa->spa_name); 1098 1099 ASSERT(curproc != &p0); 1100 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1101 "zpool-%s", spa->spa_name); 1102 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1103 1104#ifdef PSRSET_BIND 1105 /* bind this thread to the requested psrset */ 1106 if (zio_taskq_psrset_bind != PS_NONE) { 1107 pool_lock(); 1108 mutex_enter(&cpu_lock); 1109 mutex_enter(&pidlock); 1110 mutex_enter(&curproc->p_lock); 1111 1112 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1113 0, NULL, NULL) == 0) { 1114 curthread->t_bind_pset = zio_taskq_psrset_bind; 1115 } else { 1116 cmn_err(CE_WARN, 1117 "Couldn't bind process for zfs pool \"%s\" to " 1118 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1119 } 1120 1121 mutex_exit(&curproc->p_lock); 1122 mutex_exit(&pidlock); 1123 mutex_exit(&cpu_lock); 1124 pool_unlock(); 1125 } 1126#endif 1127 1128#ifdef SYSDC 1129 if (zio_taskq_sysdc) { 1130 sysdc_thread_enter(curthread, 100, 0); 1131 } 1132#endif 1133 1134 spa->spa_proc = curproc; 1135 spa->spa_did = curthread->t_did; 1136 1137 spa_create_zio_taskqs(spa); 1138 1139 mutex_enter(&spa->spa_proc_lock); 1140 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1141 1142 spa->spa_proc_state = SPA_PROC_ACTIVE; 1143 cv_broadcast(&spa->spa_proc_cv); 1144 1145 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1146 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1147 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1148 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1149 1150 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1151 spa->spa_proc_state = SPA_PROC_GONE; 1152 spa->spa_proc = &p0; 1153 cv_broadcast(&spa->spa_proc_cv); 1154 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1155 1156 mutex_enter(&curproc->p_lock); 1157 lwp_exit(); 1158} 1159#endif /* SPA_PROCESS */ 1160#endif 1161 1162/* 1163 * Activate an uninitialized pool. 1164 */ 1165static void 1166spa_activate(spa_t *spa, int mode) 1167{ 1168 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1169 1170 spa->spa_state = POOL_STATE_ACTIVE; 1171 spa->spa_mode = mode; 1172 1173 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1174 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1175 1176 /* Try to create a covering process */ 1177 mutex_enter(&spa->spa_proc_lock); 1178 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1179 ASSERT(spa->spa_proc == &p0); 1180 spa->spa_did = 0; 1181 1182#ifdef SPA_PROCESS 1183 /* Only create a process if we're going to be around a while. */ 1184 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1185 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1186 NULL, 0) == 0) { 1187 spa->spa_proc_state = SPA_PROC_CREATED; 1188 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1189 cv_wait(&spa->spa_proc_cv, 1190 &spa->spa_proc_lock); 1191 } 1192 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1193 ASSERT(spa->spa_proc != &p0); 1194 ASSERT(spa->spa_did != 0); 1195 } else { 1196#ifdef _KERNEL 1197 cmn_err(CE_WARN, 1198 "Couldn't create process for zfs pool \"%s\"\n", 1199 spa->spa_name); 1200#endif 1201 } 1202 } 1203#endif /* SPA_PROCESS */ 1204 mutex_exit(&spa->spa_proc_lock); 1205 1206 /* If we didn't create a process, we need to create our taskqs. */ 1207 ASSERT(spa->spa_proc == &p0); 1208 if (spa->spa_proc == &p0) { 1209 spa_create_zio_taskqs(spa); 1210 } 1211 1212 /* 1213 * Start TRIM thread. 1214 */ 1215 trim_thread_create(spa); 1216 1217 for (size_t i = 0; i < TXG_SIZE; i++) { 1218 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 1219 ZIO_FLAG_CANFAIL); 1220 } 1221 1222 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1223 offsetof(vdev_t, vdev_config_dirty_node)); 1224 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1225 offsetof(objset_t, os_evicting_node)); 1226 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1227 offsetof(vdev_t, vdev_state_dirty_node)); 1228 1229 txg_list_create(&spa->spa_vdev_txg_list, spa, 1230 offsetof(struct vdev, vdev_txg_node)); 1231 1232 avl_create(&spa->spa_errlist_scrub, 1233 spa_error_entry_compare, sizeof (spa_error_entry_t), 1234 offsetof(spa_error_entry_t, se_avl)); 1235 avl_create(&spa->spa_errlist_last, 1236 spa_error_entry_compare, sizeof (spa_error_entry_t), 1237 offsetof(spa_error_entry_t, se_avl)); 1238} 1239 1240/* 1241 * Opposite of spa_activate(). 1242 */ 1243static void 1244spa_deactivate(spa_t *spa) 1245{ 1246 ASSERT(spa->spa_sync_on == B_FALSE); 1247 ASSERT(spa->spa_dsl_pool == NULL); 1248 ASSERT(spa->spa_root_vdev == NULL); 1249 ASSERT(spa->spa_async_zio_root == NULL); 1250 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1251 1252 /* 1253 * Stop TRIM thread in case spa_unload() wasn't called directly 1254 * before spa_deactivate(). 1255 */ 1256 trim_thread_destroy(spa); 1257 1258 spa_evicting_os_wait(spa); 1259 1260 txg_list_destroy(&spa->spa_vdev_txg_list); 1261 1262 list_destroy(&spa->spa_config_dirty_list); 1263 list_destroy(&spa->spa_evicting_os_list); 1264 list_destroy(&spa->spa_state_dirty_list); 1265 1266 for (int t = 0; t < ZIO_TYPES; t++) { 1267 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1268 spa_taskqs_fini(spa, t, q); 1269 } 1270 } 1271 1272 for (size_t i = 0; i < TXG_SIZE; i++) { 1273 ASSERT3P(spa->spa_txg_zio[i], !=, NULL); 1274 VERIFY0(zio_wait(spa->spa_txg_zio[i])); 1275 spa->spa_txg_zio[i] = NULL; 1276 } 1277 1278 metaslab_class_destroy(spa->spa_normal_class); 1279 spa->spa_normal_class = NULL; 1280 1281 metaslab_class_destroy(spa->spa_log_class); 1282 spa->spa_log_class = NULL; 1283 1284 /* 1285 * If this was part of an import or the open otherwise failed, we may 1286 * still have errors left in the queues. Empty them just in case. 1287 */ 1288 spa_errlog_drain(spa); 1289 1290 avl_destroy(&spa->spa_errlist_scrub); 1291 avl_destroy(&spa->spa_errlist_last); 1292 1293 spa->spa_state = POOL_STATE_UNINITIALIZED; 1294 1295 mutex_enter(&spa->spa_proc_lock); 1296 if (spa->spa_proc_state != SPA_PROC_NONE) { 1297 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1298 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1299 cv_broadcast(&spa->spa_proc_cv); 1300 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1301 ASSERT(spa->spa_proc != &p0); 1302 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1303 } 1304 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1305 spa->spa_proc_state = SPA_PROC_NONE; 1306 } 1307 ASSERT(spa->spa_proc == &p0); 1308 mutex_exit(&spa->spa_proc_lock); 1309 1310#ifdef SPA_PROCESS 1311 /* 1312 * We want to make sure spa_thread() has actually exited the ZFS 1313 * module, so that the module can't be unloaded out from underneath 1314 * it. 1315 */ 1316 if (spa->spa_did != 0) { 1317 thread_join(spa->spa_did); 1318 spa->spa_did = 0; 1319 } 1320#endif /* SPA_PROCESS */ 1321} 1322 1323/* 1324 * Verify a pool configuration, and construct the vdev tree appropriately. This 1325 * will create all the necessary vdevs in the appropriate layout, with each vdev 1326 * in the CLOSED state. This will prep the pool before open/creation/import. 1327 * All vdev validation is done by the vdev_alloc() routine. 1328 */ 1329static int 1330spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1331 uint_t id, int atype) 1332{ 1333 nvlist_t **child; 1334 uint_t children; 1335 int error; 1336 1337 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1338 return (error); 1339 1340 if ((*vdp)->vdev_ops->vdev_op_leaf) 1341 return (0); 1342 1343 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1344 &child, &children); 1345 1346 if (error == ENOENT) 1347 return (0); 1348 1349 if (error) { 1350 vdev_free(*vdp); 1351 *vdp = NULL; 1352 return (SET_ERROR(EINVAL)); 1353 } 1354 1355 for (int c = 0; c < children; c++) { 1356 vdev_t *vd; 1357 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1358 atype)) != 0) { 1359 vdev_free(*vdp); 1360 *vdp = NULL; 1361 return (error); 1362 } 1363 } 1364 1365 ASSERT(*vdp != NULL); 1366 1367 return (0); 1368} 1369 1370/* 1371 * Opposite of spa_load(). 1372 */ 1373static void 1374spa_unload(spa_t *spa) 1375{ 1376 int i; 1377 1378 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1379 1380 spa_load_note(spa, "UNLOADING"); 1381 1382 /* 1383 * Stop TRIM thread. 1384 */ 1385 trim_thread_destroy(spa); 1386 1387 /* 1388 * Stop async tasks. 1389 */ 1390 spa_async_suspend(spa); 1391 1392 if (spa->spa_root_vdev) { 1393 vdev_initialize_stop_all(spa->spa_root_vdev, 1394 VDEV_INITIALIZE_ACTIVE); 1395 } 1396 1397 /* 1398 * Stop syncing. 1399 */ 1400 if (spa->spa_sync_on) { 1401 txg_sync_stop(spa->spa_dsl_pool); 1402 spa->spa_sync_on = B_FALSE; 1403 } 1404 1405 /* 1406 * Even though vdev_free() also calls vdev_metaslab_fini, we need 1407 * to call it earlier, before we wait for async i/o to complete. 1408 * This ensures that there is no async metaslab prefetching, by 1409 * calling taskq_wait(mg_taskq). 1410 */ 1411 if (spa->spa_root_vdev != NULL) { 1412 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1413 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) 1414 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); 1415 spa_config_exit(spa, SCL_ALL, spa); 1416 } 1417 1418 /* 1419 * Wait for any outstanding async I/O to complete. 1420 */ 1421 if (spa->spa_async_zio_root != NULL) { 1422 for (int i = 0; i < max_ncpus; i++) 1423 (void) zio_wait(spa->spa_async_zio_root[i]); 1424 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1425 spa->spa_async_zio_root = NULL; 1426 } 1427 1428 if (spa->spa_vdev_removal != NULL) { 1429 spa_vdev_removal_destroy(spa->spa_vdev_removal); 1430 spa->spa_vdev_removal = NULL; 1431 } 1432 1433 if (spa->spa_condense_zthr != NULL) { 1434 ASSERT(!zthr_isrunning(spa->spa_condense_zthr)); 1435 zthr_destroy(spa->spa_condense_zthr); 1436 spa->spa_condense_zthr = NULL; 1437 } 1438 1439 if (spa->spa_checkpoint_discard_zthr != NULL) { 1440 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr)); 1441 zthr_destroy(spa->spa_checkpoint_discard_zthr); 1442 spa->spa_checkpoint_discard_zthr = NULL; 1443 } 1444 1445 spa_condense_fini(spa); 1446 1447 bpobj_close(&spa->spa_deferred_bpobj); 1448 1449 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER); 1450 1451 /* 1452 * Close all vdevs. 1453 */ 1454 if (spa->spa_root_vdev) 1455 vdev_free(spa->spa_root_vdev); 1456 ASSERT(spa->spa_root_vdev == NULL); 1457 1458 /* 1459 * Close the dsl pool. 1460 */ 1461 if (spa->spa_dsl_pool) { 1462 dsl_pool_close(spa->spa_dsl_pool); 1463 spa->spa_dsl_pool = NULL; 1464 spa->spa_meta_objset = NULL; 1465 } 1466 1467 ddt_unload(spa); 1468 1469 /* 1470 * Drop and purge level 2 cache 1471 */ 1472 spa_l2cache_drop(spa); 1473 1474 for (i = 0; i < spa->spa_spares.sav_count; i++) 1475 vdev_free(spa->spa_spares.sav_vdevs[i]); 1476 if (spa->spa_spares.sav_vdevs) { 1477 kmem_free(spa->spa_spares.sav_vdevs, 1478 spa->spa_spares.sav_count * sizeof (void *)); 1479 spa->spa_spares.sav_vdevs = NULL; 1480 } 1481 if (spa->spa_spares.sav_config) { 1482 nvlist_free(spa->spa_spares.sav_config); 1483 spa->spa_spares.sav_config = NULL; 1484 } 1485 spa->spa_spares.sav_count = 0; 1486 1487 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1488 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1489 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1490 } 1491 if (spa->spa_l2cache.sav_vdevs) { 1492 kmem_free(spa->spa_l2cache.sav_vdevs, 1493 spa->spa_l2cache.sav_count * sizeof (void *)); 1494 spa->spa_l2cache.sav_vdevs = NULL; 1495 } 1496 if (spa->spa_l2cache.sav_config) { 1497 nvlist_free(spa->spa_l2cache.sav_config); 1498 spa->spa_l2cache.sav_config = NULL; 1499 } 1500 spa->spa_l2cache.sav_count = 0; 1501 1502 spa->spa_async_suspended = 0; 1503 1504 spa->spa_indirect_vdevs_loaded = B_FALSE; 1505 1506 if (spa->spa_comment != NULL) { 1507 spa_strfree(spa->spa_comment); 1508 spa->spa_comment = NULL; 1509 } 1510 1511 spa_config_exit(spa, SCL_ALL, spa); 1512} 1513 1514/* 1515 * Load (or re-load) the current list of vdevs describing the active spares for 1516 * this pool. When this is called, we have some form of basic information in 1517 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1518 * then re-generate a more complete list including status information. 1519 */ 1520void 1521spa_load_spares(spa_t *spa) 1522{ 1523 nvlist_t **spares; 1524 uint_t nspares; 1525 int i; 1526 vdev_t *vd, *tvd; 1527 1528#ifndef _KERNEL 1529 /* 1530 * zdb opens both the current state of the pool and the 1531 * checkpointed state (if present), with a different spa_t. 1532 * 1533 * As spare vdevs are shared among open pools, we skip loading 1534 * them when we load the checkpointed state of the pool. 1535 */ 1536 if (!spa_writeable(spa)) 1537 return; 1538#endif 1539 1540 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1541 1542 /* 1543 * First, close and free any existing spare vdevs. 1544 */ 1545 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1546 vd = spa->spa_spares.sav_vdevs[i]; 1547 1548 /* Undo the call to spa_activate() below */ 1549 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1550 B_FALSE)) != NULL && tvd->vdev_isspare) 1551 spa_spare_remove(tvd); 1552 vdev_close(vd); 1553 vdev_free(vd); 1554 } 1555 1556 if (spa->spa_spares.sav_vdevs) 1557 kmem_free(spa->spa_spares.sav_vdevs, 1558 spa->spa_spares.sav_count * sizeof (void *)); 1559 1560 if (spa->spa_spares.sav_config == NULL) 1561 nspares = 0; 1562 else 1563 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1564 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1565 1566 spa->spa_spares.sav_count = (int)nspares; 1567 spa->spa_spares.sav_vdevs = NULL; 1568 1569 if (nspares == 0) 1570 return; 1571 1572 /* 1573 * Construct the array of vdevs, opening them to get status in the 1574 * process. For each spare, there is potentially two different vdev_t 1575 * structures associated with it: one in the list of spares (used only 1576 * for basic validation purposes) and one in the active vdev 1577 * configuration (if it's spared in). During this phase we open and 1578 * validate each vdev on the spare list. If the vdev also exists in the 1579 * active configuration, then we also mark this vdev as an active spare. 1580 */ 1581 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1582 KM_SLEEP); 1583 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1584 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1585 VDEV_ALLOC_SPARE) == 0); 1586 ASSERT(vd != NULL); 1587 1588 spa->spa_spares.sav_vdevs[i] = vd; 1589 1590 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1591 B_FALSE)) != NULL) { 1592 if (!tvd->vdev_isspare) 1593 spa_spare_add(tvd); 1594 1595 /* 1596 * We only mark the spare active if we were successfully 1597 * able to load the vdev. Otherwise, importing a pool 1598 * with a bad active spare would result in strange 1599 * behavior, because multiple pool would think the spare 1600 * is actively in use. 1601 * 1602 * There is a vulnerability here to an equally bizarre 1603 * circumstance, where a dead active spare is later 1604 * brought back to life (onlined or otherwise). Given 1605 * the rarity of this scenario, and the extra complexity 1606 * it adds, we ignore the possibility. 1607 */ 1608 if (!vdev_is_dead(tvd)) 1609 spa_spare_activate(tvd); 1610 } 1611 1612 vd->vdev_top = vd; 1613 vd->vdev_aux = &spa->spa_spares; 1614 1615 if (vdev_open(vd) != 0) 1616 continue; 1617 1618 if (vdev_validate_aux(vd) == 0) 1619 spa_spare_add(vd); 1620 } 1621 1622 /* 1623 * Recompute the stashed list of spares, with status information 1624 * this time. 1625 */ 1626 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1627 DATA_TYPE_NVLIST_ARRAY) == 0); 1628 1629 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1630 KM_SLEEP); 1631 for (i = 0; i < spa->spa_spares.sav_count; i++) 1632 spares[i] = vdev_config_generate(spa, 1633 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1634 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1635 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1636 for (i = 0; i < spa->spa_spares.sav_count; i++) 1637 nvlist_free(spares[i]); 1638 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1639} 1640 1641/* 1642 * Load (or re-load) the current list of vdevs describing the active l2cache for 1643 * this pool. When this is called, we have some form of basic information in 1644 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1645 * then re-generate a more complete list including status information. 1646 * Devices which are already active have their details maintained, and are 1647 * not re-opened. 1648 */ 1649void 1650spa_load_l2cache(spa_t *spa) 1651{ 1652 nvlist_t **l2cache; 1653 uint_t nl2cache; 1654 int i, j, oldnvdevs; 1655 uint64_t guid; 1656 vdev_t *vd, **oldvdevs, **newvdevs; 1657 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1658 1659#ifndef _KERNEL 1660 /* 1661 * zdb opens both the current state of the pool and the 1662 * checkpointed state (if present), with a different spa_t. 1663 * 1664 * As L2 caches are part of the ARC which is shared among open 1665 * pools, we skip loading them when we load the checkpointed 1666 * state of the pool. 1667 */ 1668 if (!spa_writeable(spa)) 1669 return; 1670#endif 1671 1672 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1673 1674 if (sav->sav_config != NULL) { 1675 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1676 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1677 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1678 } else { 1679 nl2cache = 0; 1680 newvdevs = NULL; 1681 } 1682 1683 oldvdevs = sav->sav_vdevs; 1684 oldnvdevs = sav->sav_count; 1685 sav->sav_vdevs = NULL; 1686 sav->sav_count = 0; 1687 1688 /* 1689 * Process new nvlist of vdevs. 1690 */ 1691 for (i = 0; i < nl2cache; i++) { 1692 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1693 &guid) == 0); 1694 1695 newvdevs[i] = NULL; 1696 for (j = 0; j < oldnvdevs; j++) { 1697 vd = oldvdevs[j]; 1698 if (vd != NULL && guid == vd->vdev_guid) { 1699 /* 1700 * Retain previous vdev for add/remove ops. 1701 */ 1702 newvdevs[i] = vd; 1703 oldvdevs[j] = NULL; 1704 break; 1705 } 1706 } 1707 1708 if (newvdevs[i] == NULL) { 1709 /* 1710 * Create new vdev 1711 */ 1712 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1713 VDEV_ALLOC_L2CACHE) == 0); 1714 ASSERT(vd != NULL); 1715 newvdevs[i] = vd; 1716 1717 /* 1718 * Commit this vdev as an l2cache device, 1719 * even if it fails to open. 1720 */ 1721 spa_l2cache_add(vd); 1722 1723 vd->vdev_top = vd; 1724 vd->vdev_aux = sav; 1725 1726 spa_l2cache_activate(vd); 1727 1728 if (vdev_open(vd) != 0) 1729 continue; 1730 1731 (void) vdev_validate_aux(vd); 1732 1733 if (!vdev_is_dead(vd)) 1734 l2arc_add_vdev(spa, vd); 1735 } 1736 } 1737 1738 /* 1739 * Purge vdevs that were dropped 1740 */ 1741 for (i = 0; i < oldnvdevs; i++) { 1742 uint64_t pool; 1743 1744 vd = oldvdevs[i]; 1745 if (vd != NULL) { 1746 ASSERT(vd->vdev_isl2cache); 1747 1748 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1749 pool != 0ULL && l2arc_vdev_present(vd)) 1750 l2arc_remove_vdev(vd); 1751 vdev_clear_stats(vd); 1752 vdev_free(vd); 1753 } 1754 } 1755 1756 if (oldvdevs) 1757 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1758 1759 if (sav->sav_config == NULL) 1760 goto out; 1761 1762 sav->sav_vdevs = newvdevs; 1763 sav->sav_count = (int)nl2cache; 1764 1765 /* 1766 * Recompute the stashed list of l2cache devices, with status 1767 * information this time. 1768 */ 1769 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1770 DATA_TYPE_NVLIST_ARRAY) == 0); 1771 1772 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1773 for (i = 0; i < sav->sav_count; i++) 1774 l2cache[i] = vdev_config_generate(spa, 1775 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1776 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1777 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1778out: 1779 for (i = 0; i < sav->sav_count; i++) 1780 nvlist_free(l2cache[i]); 1781 if (sav->sav_count) 1782 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1783} 1784 1785static int 1786load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1787{ 1788 dmu_buf_t *db; 1789 char *packed = NULL; 1790 size_t nvsize = 0; 1791 int error; 1792 *value = NULL; 1793 1794 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1795 if (error != 0) 1796 return (error); 1797 1798 nvsize = *(uint64_t *)db->db_data; 1799 dmu_buf_rele(db, FTAG); 1800 1801 packed = kmem_alloc(nvsize, KM_SLEEP); 1802 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1803 DMU_READ_PREFETCH); 1804 if (error == 0) 1805 error = nvlist_unpack(packed, nvsize, value, 0); 1806 kmem_free(packed, nvsize); 1807 1808 return (error); 1809} 1810 1811/* 1812 * Concrete top-level vdevs that are not missing and are not logs. At every 1813 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. 1814 */ 1815static uint64_t 1816spa_healthy_core_tvds(spa_t *spa) 1817{ 1818 vdev_t *rvd = spa->spa_root_vdev; 1819 uint64_t tvds = 0; 1820 1821 for (uint64_t i = 0; i < rvd->vdev_children; i++) { 1822 vdev_t *vd = rvd->vdev_child[i]; 1823 if (vd->vdev_islog) 1824 continue; 1825 if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) 1826 tvds++; 1827 } 1828 1829 return (tvds); 1830} 1831 1832/* 1833 * Checks to see if the given vdev could not be opened, in which case we post a 1834 * sysevent to notify the autoreplace code that the device has been removed. 1835 */ 1836static void 1837spa_check_removed(vdev_t *vd) 1838{ 1839 for (uint64_t c = 0; c < vd->vdev_children; c++) 1840 spa_check_removed(vd->vdev_child[c]); 1841 1842 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1843 vdev_is_concrete(vd)) { 1844 zfs_post_autoreplace(vd->vdev_spa, vd); 1845 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); 1846 } 1847} 1848 1849static int 1850spa_check_for_missing_logs(spa_t *spa) 1851{ 1852 vdev_t *rvd = spa->spa_root_vdev; 1853 1854 /* 1855 * If we're doing a normal import, then build up any additional 1856 * diagnostic information about missing log devices. 1857 * We'll pass this up to the user for further processing. 1858 */ 1859 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1860 nvlist_t **child, *nv; 1861 uint64_t idx = 0; 1862 1863 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1864 KM_SLEEP); 1865 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1866 1867 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1868 vdev_t *tvd = rvd->vdev_child[c]; 1869 1870 /* 1871 * We consider a device as missing only if it failed 1872 * to open (i.e. offline or faulted is not considered 1873 * as missing). 1874 */ 1875 if (tvd->vdev_islog && 1876 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1877 child[idx++] = vdev_config_generate(spa, tvd, 1878 B_FALSE, VDEV_CONFIG_MISSING); 1879 } 1880 } 1881 1882 if (idx > 0) { 1883 fnvlist_add_nvlist_array(nv, 1884 ZPOOL_CONFIG_CHILDREN, child, idx); 1885 fnvlist_add_nvlist(spa->spa_load_info, 1886 ZPOOL_CONFIG_MISSING_DEVICES, nv); 1887 1888 for (uint64_t i = 0; i < idx; i++) 1889 nvlist_free(child[i]); 1890 } 1891 nvlist_free(nv); 1892 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1893 1894 if (idx > 0) { 1895 spa_load_failed(spa, "some log devices are missing"); 1896 vdev_dbgmsg_print_tree(rvd, 2); 1897 return (SET_ERROR(ENXIO)); 1898 } 1899 } else { 1900 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1901 vdev_t *tvd = rvd->vdev_child[c]; 1902 1903 if (tvd->vdev_islog && 1904 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1905 spa_set_log_state(spa, SPA_LOG_CLEAR); 1906 spa_load_note(spa, "some log devices are " 1907 "missing, ZIL is dropped."); 1908 vdev_dbgmsg_print_tree(rvd, 2); 1909 break; 1910 } 1911 } 1912 } 1913 1914 return (0); 1915} 1916 1917/* 1918 * Check for missing log devices 1919 */ 1920static boolean_t 1921spa_check_logs(spa_t *spa) 1922{ 1923 boolean_t rv = B_FALSE; 1924 dsl_pool_t *dp = spa_get_dsl(spa); 1925 1926 switch (spa->spa_log_state) { 1927 case SPA_LOG_MISSING: 1928 /* need to recheck in case slog has been restored */ 1929 case SPA_LOG_UNKNOWN: 1930 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1931 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1932 if (rv) 1933 spa_set_log_state(spa, SPA_LOG_MISSING); 1934 break; 1935 } 1936 return (rv); 1937} 1938 1939static boolean_t 1940spa_passivate_log(spa_t *spa) 1941{ 1942 vdev_t *rvd = spa->spa_root_vdev; 1943 boolean_t slog_found = B_FALSE; 1944 1945 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1946 1947 if (!spa_has_slogs(spa)) 1948 return (B_FALSE); 1949 1950 for (int c = 0; c < rvd->vdev_children; c++) { 1951 vdev_t *tvd = rvd->vdev_child[c]; 1952 metaslab_group_t *mg = tvd->vdev_mg; 1953 1954 if (tvd->vdev_islog) { 1955 metaslab_group_passivate(mg); 1956 slog_found = B_TRUE; 1957 } 1958 } 1959 1960 return (slog_found); 1961} 1962 1963static void 1964spa_activate_log(spa_t *spa) 1965{ 1966 vdev_t *rvd = spa->spa_root_vdev; 1967 1968 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1969 1970 for (int c = 0; c < rvd->vdev_children; c++) { 1971 vdev_t *tvd = rvd->vdev_child[c]; 1972 metaslab_group_t *mg = tvd->vdev_mg; 1973 1974 if (tvd->vdev_islog) 1975 metaslab_group_activate(mg); 1976 } 1977} 1978 1979int 1980spa_reset_logs(spa_t *spa) 1981{ 1982 int error; 1983 1984 error = dmu_objset_find(spa_name(spa), zil_reset, 1985 NULL, DS_FIND_CHILDREN); 1986 if (error == 0) { 1987 /* 1988 * We successfully offlined the log device, sync out the 1989 * current txg so that the "stubby" block can be removed 1990 * by zil_sync(). 1991 */ 1992 txg_wait_synced(spa->spa_dsl_pool, 0); 1993 } 1994 return (error); 1995} 1996 1997static void 1998spa_aux_check_removed(spa_aux_vdev_t *sav) 1999{ 2000 int i; 2001 2002 for (i = 0; i < sav->sav_count; i++) 2003 spa_check_removed(sav->sav_vdevs[i]); 2004} 2005 2006void 2007spa_claim_notify(zio_t *zio) 2008{ 2009 spa_t *spa = zio->io_spa; 2010 2011 if (zio->io_error) 2012 return; 2013 2014 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 2015 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 2016 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 2017 mutex_exit(&spa->spa_props_lock); 2018} 2019 2020typedef struct spa_load_error { 2021 uint64_t sle_meta_count; 2022 uint64_t sle_data_count; 2023} spa_load_error_t; 2024 2025static void 2026spa_load_verify_done(zio_t *zio) 2027{ 2028 blkptr_t *bp = zio->io_bp; 2029 spa_load_error_t *sle = zio->io_private; 2030 dmu_object_type_t type = BP_GET_TYPE(bp); 2031 int error = zio->io_error; 2032 spa_t *spa = zio->io_spa; 2033 2034 abd_free(zio->io_abd); 2035 if (error) { 2036 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 2037 type != DMU_OT_INTENT_LOG) 2038 atomic_inc_64(&sle->sle_meta_count); 2039 else 2040 atomic_inc_64(&sle->sle_data_count); 2041 } 2042 2043 mutex_enter(&spa->spa_scrub_lock); 2044 spa->spa_load_verify_ios--; 2045 cv_broadcast(&spa->spa_scrub_io_cv); 2046 mutex_exit(&spa->spa_scrub_lock); 2047} 2048 2049/* 2050 * Maximum number of concurrent scrub i/os to create while verifying 2051 * a pool while importing it. 2052 */ 2053int spa_load_verify_maxinflight = 10000; 2054boolean_t spa_load_verify_metadata = B_TRUE; 2055boolean_t spa_load_verify_data = B_TRUE; 2056 2057SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 2058 &spa_load_verify_maxinflight, 0, 2059 "Maximum number of concurrent scrub I/Os to create while verifying a " 2060 "pool while importing it"); 2061 2062SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 2063 &spa_load_verify_metadata, 0, 2064 "Check metadata on import?"); 2065 2066SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 2067 &spa_load_verify_data, 0, 2068 "Check user data on import?"); 2069 2070/*ARGSUSED*/ 2071static int 2072spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 2073 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 2074{ 2075 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 2076 return (0); 2077 /* 2078 * Note: normally this routine will not be called if 2079 * spa_load_verify_metadata is not set. However, it may be useful 2080 * to manually set the flag after the traversal has begun. 2081 */ 2082 if (!spa_load_verify_metadata) 2083 return (0); 2084 if (!BP_IS_METADATA(bp) && !spa_load_verify_data) 2085 return (0); 2086 2087 zio_t *rio = arg; 2088 size_t size = BP_GET_PSIZE(bp); 2089 2090 mutex_enter(&spa->spa_scrub_lock); 2091 while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight) 2092 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 2093 spa->spa_load_verify_ios++; 2094 mutex_exit(&spa->spa_scrub_lock); 2095 2096 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, 2097 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 2098 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 2099 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 2100 return (0); 2101} 2102 2103/* ARGSUSED */ 2104int 2105verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 2106{ 2107 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 2108 return (SET_ERROR(ENAMETOOLONG)); 2109 2110 return (0); 2111} 2112 2113static int 2114spa_load_verify(spa_t *spa) 2115{ 2116 zio_t *rio; 2117 spa_load_error_t sle = { 0 }; 2118 zpool_load_policy_t policy; 2119 boolean_t verify_ok = B_FALSE; 2120 int error = 0; 2121 2122 zpool_get_load_policy(spa->spa_config, &policy); 2123 2124 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND) 2125 return (0); 2126 2127 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2128 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2129 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2130 DS_FIND_CHILDREN); 2131 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2132 if (error != 0) 2133 return (error); 2134 2135 rio = zio_root(spa, NULL, &sle, 2136 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2137 2138 if (spa_load_verify_metadata) { 2139 if (spa->spa_extreme_rewind) { 2140 spa_load_note(spa, "performing a complete scan of the " 2141 "pool since extreme rewind is on. This may take " 2142 "a very long time.\n (spa_load_verify_data=%u, " 2143 "spa_load_verify_metadata=%u)", 2144 spa_load_verify_data, spa_load_verify_metadata); 2145 } 2146 error = traverse_pool(spa, spa->spa_verify_min_txg, 2147 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2148 spa_load_verify_cb, rio); 2149 } 2150 2151 (void) zio_wait(rio); 2152 2153 spa->spa_load_meta_errors = sle.sle_meta_count; 2154 spa->spa_load_data_errors = sle.sle_data_count; 2155 2156 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { 2157 spa_load_note(spa, "spa_load_verify found %llu metadata errors " 2158 "and %llu data errors", (u_longlong_t)sle.sle_meta_count, 2159 (u_longlong_t)sle.sle_data_count); 2160 } 2161 2162 if (spa_load_verify_dryrun || 2163 (!error && sle.sle_meta_count <= policy.zlp_maxmeta && 2164 sle.sle_data_count <= policy.zlp_maxdata)) { 2165 int64_t loss = 0; 2166 2167 verify_ok = B_TRUE; 2168 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2169 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2170 2171 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2172 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2173 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2174 VERIFY(nvlist_add_int64(spa->spa_load_info, 2175 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2176 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2177 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2178 } else { 2179 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2180 } 2181 2182 if (spa_load_verify_dryrun) 2183 return (0); 2184 2185 if (error) { 2186 if (error != ENXIO && error != EIO) 2187 error = SET_ERROR(EIO); 2188 return (error); 2189 } 2190 2191 return (verify_ok ? 0 : EIO); 2192} 2193 2194/* 2195 * Find a value in the pool props object. 2196 */ 2197static void 2198spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2199{ 2200 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2201 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2202} 2203 2204/* 2205 * Find a value in the pool directory object. 2206 */ 2207static int 2208spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) 2209{ 2210 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2211 name, sizeof (uint64_t), 1, val); 2212 2213 if (error != 0 && (error != ENOENT || log_enoent)) { 2214 spa_load_failed(spa, "couldn't get '%s' value in MOS directory " 2215 "[error=%d]", name, error); 2216 } 2217 2218 return (error); 2219} 2220 2221static int 2222spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2223{ 2224 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2225 return (SET_ERROR(err)); 2226} 2227 2228static void 2229spa_spawn_aux_threads(spa_t *spa) 2230{ 2231 ASSERT(spa_writeable(spa)); 2232 2233 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2234 2235 spa_start_indirect_condensing_thread(spa); 2236 2237 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL); 2238 spa->spa_checkpoint_discard_zthr = 2239 zthr_create(spa_checkpoint_discard_thread_check, 2240 spa_checkpoint_discard_thread, spa); 2241} 2242 2243/* 2244 * Fix up config after a partly-completed split. This is done with the 2245 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2246 * pool have that entry in their config, but only the splitting one contains 2247 * a list of all the guids of the vdevs that are being split off. 2248 * 2249 * This function determines what to do with that list: either rejoin 2250 * all the disks to the pool, or complete the splitting process. To attempt 2251 * the rejoin, each disk that is offlined is marked online again, and 2252 * we do a reopen() call. If the vdev label for every disk that was 2253 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2254 * then we call vdev_split() on each disk, and complete the split. 2255 * 2256 * Otherwise we leave the config alone, with all the vdevs in place in 2257 * the original pool. 2258 */ 2259static void 2260spa_try_repair(spa_t *spa, nvlist_t *config) 2261{ 2262 uint_t extracted; 2263 uint64_t *glist; 2264 uint_t i, gcount; 2265 nvlist_t *nvl; 2266 vdev_t **vd; 2267 boolean_t attempt_reopen; 2268 2269 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2270 return; 2271 2272 /* check that the config is complete */ 2273 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2274 &glist, &gcount) != 0) 2275 return; 2276 2277 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2278 2279 /* attempt to online all the vdevs & validate */ 2280 attempt_reopen = B_TRUE; 2281 for (i = 0; i < gcount; i++) { 2282 if (glist[i] == 0) /* vdev is hole */ 2283 continue; 2284 2285 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2286 if (vd[i] == NULL) { 2287 /* 2288 * Don't bother attempting to reopen the disks; 2289 * just do the split. 2290 */ 2291 attempt_reopen = B_FALSE; 2292 } else { 2293 /* attempt to re-online it */ 2294 vd[i]->vdev_offline = B_FALSE; 2295 } 2296 } 2297 2298 if (attempt_reopen) { 2299 vdev_reopen(spa->spa_root_vdev); 2300 2301 /* check each device to see what state it's in */ 2302 for (extracted = 0, i = 0; i < gcount; i++) { 2303 if (vd[i] != NULL && 2304 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2305 break; 2306 ++extracted; 2307 } 2308 } 2309 2310 /* 2311 * If every disk has been moved to the new pool, or if we never 2312 * even attempted to look at them, then we split them off for 2313 * good. 2314 */ 2315 if (!attempt_reopen || gcount == extracted) { 2316 for (i = 0; i < gcount; i++) 2317 if (vd[i] != NULL) 2318 vdev_split(vd[i]); 2319 vdev_reopen(spa->spa_root_vdev); 2320 } 2321 2322 kmem_free(vd, gcount * sizeof (vdev_t *)); 2323} 2324 2325static int 2326spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) 2327{ 2328 char *ereport = FM_EREPORT_ZFS_POOL; 2329 int error; 2330 2331 spa->spa_load_state = state; 2332 2333 gethrestime(&spa->spa_loaded_ts); 2334 error = spa_load_impl(spa, type, &ereport); 2335 2336 /* 2337 * Don't count references from objsets that are already closed 2338 * and are making their way through the eviction process. 2339 */ 2340 spa_evicting_os_wait(spa); 2341 spa->spa_minref = refcount_count(&spa->spa_refcount); 2342 if (error) { 2343 if (error != EEXIST) { 2344 spa->spa_loaded_ts.tv_sec = 0; 2345 spa->spa_loaded_ts.tv_nsec = 0; 2346 } 2347 if (error != EBADF) { 2348 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2349 } 2350 } 2351 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2352 spa->spa_ena = 0; 2353 2354 return (error); 2355} 2356 2357/* 2358 * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2359 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2360 * spa's per-vdev ZAP list. 2361 */ 2362static uint64_t 2363vdev_count_verify_zaps(vdev_t *vd) 2364{ 2365 spa_t *spa = vd->vdev_spa; 2366 uint64_t total = 0; 2367 if (vd->vdev_top_zap != 0) { 2368 total++; 2369 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2370 spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2371 } 2372 if (vd->vdev_leaf_zap != 0) { 2373 total++; 2374 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2375 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2376 } 2377 2378 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2379 total += vdev_count_verify_zaps(vd->vdev_child[i]); 2380 } 2381 2382 return (total); 2383} 2384 2385static int 2386spa_verify_host(spa_t *spa, nvlist_t *mos_config) 2387{ 2388 uint64_t hostid; 2389 char *hostname; 2390 uint64_t myhostid = 0; 2391 2392 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, 2393 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2394 hostname = fnvlist_lookup_string(mos_config, 2395 ZPOOL_CONFIG_HOSTNAME); 2396 2397 myhostid = zone_get_hostid(NULL); 2398 2399 if (hostid != 0 && myhostid != 0 && hostid != myhostid) { 2400 cmn_err(CE_WARN, "pool '%s' could not be " 2401 "loaded as it was last accessed by " 2402 "another system (host: %s hostid: 0x%llx). " 2403 "See: http://illumos.org/msg/ZFS-8000-EY", 2404 spa_name(spa), hostname, (u_longlong_t)hostid); 2405 spa_load_failed(spa, "hostid verification failed: pool " 2406 "last accessed by host: %s (hostid: 0x%llx)", 2407 hostname, (u_longlong_t)hostid); 2408 return (SET_ERROR(EBADF)); 2409 } 2410 } 2411 2412 return (0); 2413} 2414 2415static int 2416spa_ld_parse_config(spa_t *spa, spa_import_type_t type) 2417{ 2418 int error = 0; 2419 nvlist_t *nvtree, *nvl, *config = spa->spa_config; 2420 int parse; 2421 vdev_t *rvd; 2422 uint64_t pool_guid; 2423 char *comment; 2424 2425 /* 2426 * Versioning wasn't explicitly added to the label until later, so if 2427 * it's not present treat it as the initial version. 2428 */ 2429 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2430 &spa->spa_ubsync.ub_version) != 0) 2431 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2432 2433 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { 2434 spa_load_failed(spa, "invalid config provided: '%s' missing", 2435 ZPOOL_CONFIG_POOL_GUID); 2436 return (SET_ERROR(EINVAL)); 2437 } 2438 2439 /* 2440 * If we are doing an import, ensure that the pool is not already 2441 * imported by checking if its pool guid already exists in the 2442 * spa namespace. 2443 * 2444 * The only case that we allow an already imported pool to be 2445 * imported again, is when the pool is checkpointed and we want to 2446 * look at its checkpointed state from userland tools like zdb. 2447 */ 2448#ifdef _KERNEL 2449 if ((spa->spa_load_state == SPA_LOAD_IMPORT || 2450 spa->spa_load_state == SPA_LOAD_TRYIMPORT) && 2451 spa_guid_exists(pool_guid, 0)) { 2452#else 2453 if ((spa->spa_load_state == SPA_LOAD_IMPORT || 2454 spa->spa_load_state == SPA_LOAD_TRYIMPORT) && 2455 spa_guid_exists(pool_guid, 0) && 2456 !spa_importing_readonly_checkpoint(spa)) { 2457#endif 2458 spa_load_failed(spa, "a pool with guid %llu is already open", 2459 (u_longlong_t)pool_guid); 2460 return (SET_ERROR(EEXIST)); 2461 } 2462 2463 spa->spa_config_guid = pool_guid; 2464 2465 nvlist_free(spa->spa_load_info); 2466 spa->spa_load_info = fnvlist_alloc(); 2467 2468 ASSERT(spa->spa_comment == NULL); 2469 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2470 spa->spa_comment = spa_strdup(comment); 2471 2472 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2473 &spa->spa_config_txg); 2474 2475 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) 2476 spa->spa_config_splitting = fnvlist_dup(nvl); 2477 2478 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { 2479 spa_load_failed(spa, "invalid config provided: '%s' missing", 2480 ZPOOL_CONFIG_VDEV_TREE); 2481 return (SET_ERROR(EINVAL)); 2482 } 2483 2484 /* 2485 * Create "The Godfather" zio to hold all async IOs 2486 */ 2487 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2488 KM_SLEEP); 2489 for (int i = 0; i < max_ncpus; i++) { 2490 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2491 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2492 ZIO_FLAG_GODFATHER); 2493 } 2494 2495 /* 2496 * Parse the configuration into a vdev tree. We explicitly set the 2497 * value that will be returned by spa_version() since parsing the 2498 * configuration requires knowing the version number. 2499 */ 2500 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2501 parse = (type == SPA_IMPORT_EXISTING ? 2502 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2503 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); 2504 spa_config_exit(spa, SCL_ALL, FTAG); 2505 2506 if (error != 0) { 2507 spa_load_failed(spa, "unable to parse config [error=%d]", 2508 error); 2509 return (error); 2510 } 2511 2512 ASSERT(spa->spa_root_vdev == rvd); 2513 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2514 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2515 2516 if (type != SPA_IMPORT_ASSEMBLE) { 2517 ASSERT(spa_guid(spa) == pool_guid); 2518 } 2519 2520 return (0); 2521} 2522 2523/* 2524 * Recursively open all vdevs in the vdev tree. This function is called twice: 2525 * first with the untrusted config, then with the trusted config. 2526 */ 2527static int 2528spa_ld_open_vdevs(spa_t *spa) 2529{ 2530 int error = 0; 2531 2532 /* 2533 * spa_missing_tvds_allowed defines how many top-level vdevs can be 2534 * missing/unopenable for the root vdev to be still considered openable. 2535 */ 2536 if (spa->spa_trust_config) { 2537 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; 2538 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { 2539 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; 2540 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { 2541 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; 2542 } else { 2543 spa->spa_missing_tvds_allowed = 0; 2544 } 2545 2546 spa->spa_missing_tvds_allowed = 2547 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); 2548 2549 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2550 error = vdev_open(spa->spa_root_vdev); 2551 spa_config_exit(spa, SCL_ALL, FTAG); 2552 2553 if (spa->spa_missing_tvds != 0) { 2554 spa_load_note(spa, "vdev tree has %lld missing top-level " 2555 "vdevs.", (u_longlong_t)spa->spa_missing_tvds); 2556 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) { 2557 /* 2558 * Although theoretically we could allow users to open 2559 * incomplete pools in RW mode, we'd need to add a lot 2560 * of extra logic (e.g. adjust pool space to account 2561 * for missing vdevs). 2562 * This limitation also prevents users from accidentally 2563 * opening the pool in RW mode during data recovery and 2564 * damaging it further. 2565 */ 2566 spa_load_note(spa, "pools with missing top-level " 2567 "vdevs can only be opened in read-only mode."); 2568 error = SET_ERROR(ENXIO); 2569 } else { 2570 spa_load_note(spa, "current settings allow for maximum " 2571 "%lld missing top-level vdevs at this stage.", 2572 (u_longlong_t)spa->spa_missing_tvds_allowed); 2573 } 2574 } 2575 if (error != 0) { 2576 spa_load_failed(spa, "unable to open vdev tree [error=%d]", 2577 error); 2578 } 2579 if (spa->spa_missing_tvds != 0 || error != 0) 2580 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); 2581 2582 return (error); 2583} 2584 2585/* 2586 * We need to validate the vdev labels against the configuration that 2587 * we have in hand. This function is called twice: first with an untrusted 2588 * config, then with a trusted config. The validation is more strict when the 2589 * config is trusted. 2590 */ 2591static int 2592spa_ld_validate_vdevs(spa_t *spa) 2593{ 2594 int error = 0; 2595 vdev_t *rvd = spa->spa_root_vdev; 2596 2597 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2598 error = vdev_validate(rvd); 2599 spa_config_exit(spa, SCL_ALL, FTAG); 2600 2601 if (error != 0) { 2602 spa_load_failed(spa, "vdev_validate failed [error=%d]", error); 2603 return (error); 2604 } 2605 2606 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 2607 spa_load_failed(spa, "cannot open vdev tree after invalidating " 2608 "some vdevs"); 2609 vdev_dbgmsg_print_tree(rvd, 2); 2610 return (SET_ERROR(ENXIO)); 2611 } 2612 2613 return (0); 2614} 2615 2616static void 2617spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub) 2618{ 2619 spa->spa_state = POOL_STATE_ACTIVE; 2620 spa->spa_ubsync = spa->spa_uberblock; 2621 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2622 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2623 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2624 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2625 spa->spa_claim_max_txg = spa->spa_first_txg; 2626 spa->spa_prev_software_version = ub->ub_software_version; 2627} 2628 2629static int 2630spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) 2631{ 2632 vdev_t *rvd = spa->spa_root_vdev; 2633 nvlist_t *label; 2634 uberblock_t *ub = &spa->spa_uberblock; 2635 2636 /* 2637 * If we are opening the checkpointed state of the pool by 2638 * rewinding to it, at this point we will have written the 2639 * checkpointed uberblock to the vdev labels, so searching 2640 * the labels will find the right uberblock. However, if 2641 * we are opening the checkpointed state read-only, we have 2642 * not modified the labels. Therefore, we must ignore the 2643 * labels and continue using the spa_uberblock that was set 2644 * by spa_ld_checkpoint_rewind. 2645 * 2646 * Note that it would be fine to ignore the labels when 2647 * rewinding (opening writeable) as well. However, if we 2648 * crash just after writing the labels, we will end up 2649 * searching the labels. Doing so in the common case means 2650 * that this code path gets exercised normally, rather than 2651 * just in the edge case. 2652 */ 2653 if (ub->ub_checkpoint_txg != 0 && 2654 spa_importing_readonly_checkpoint(spa)) { 2655 spa_ld_select_uberblock_done(spa, ub); 2656 return (0); 2657 } 2658 2659 /* 2660 * Find the best uberblock. 2661 */ 2662 vdev_uberblock_load(rvd, ub, &label); 2663 2664 /* 2665 * If we weren't able to find a single valid uberblock, return failure. 2666 */ 2667 if (ub->ub_txg == 0) { 2668 nvlist_free(label); 2669 spa_load_failed(spa, "no valid uberblock found"); 2670 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2671 } 2672 2673 spa_load_note(spa, "using uberblock with txg=%llu", 2674 (u_longlong_t)ub->ub_txg); 2675 2676 /* 2677 * If the pool has an unsupported version we can't open it. 2678 */ 2679 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2680 nvlist_free(label); 2681 spa_load_failed(spa, "version %llu is not supported", 2682 (u_longlong_t)ub->ub_version); 2683 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2684 } 2685 2686 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2687 nvlist_t *features; 2688 2689 /* 2690 * If we weren't able to find what's necessary for reading the 2691 * MOS in the label, return failure. 2692 */ 2693 if (label == NULL) { 2694 spa_load_failed(spa, "label config unavailable"); 2695 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2696 ENXIO)); 2697 } 2698 2699 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, 2700 &features) != 0) { 2701 nvlist_free(label); 2702 spa_load_failed(spa, "invalid label: '%s' missing", 2703 ZPOOL_CONFIG_FEATURES_FOR_READ); 2704 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2705 ENXIO)); 2706 } 2707 2708 /* 2709 * Update our in-core representation with the definitive values 2710 * from the label. 2711 */ 2712 nvlist_free(spa->spa_label_features); 2713 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2714 } 2715 2716 nvlist_free(label); 2717 2718 /* 2719 * Look through entries in the label nvlist's features_for_read. If 2720 * there is a feature listed there which we don't understand then we 2721 * cannot open a pool. 2722 */ 2723 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2724 nvlist_t *unsup_feat; 2725 2726 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2727 0); 2728 2729 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2730 NULL); nvp != NULL; 2731 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2732 if (!zfeature_is_supported(nvpair_name(nvp))) { 2733 VERIFY(nvlist_add_string(unsup_feat, 2734 nvpair_name(nvp), "") == 0); 2735 } 2736 } 2737 2738 if (!nvlist_empty(unsup_feat)) { 2739 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2740 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2741 nvlist_free(unsup_feat); 2742 spa_load_failed(spa, "some features are unsupported"); 2743 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2744 ENOTSUP)); 2745 } 2746 2747 nvlist_free(unsup_feat); 2748 } 2749 2750 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2751 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2752 spa_try_repair(spa, spa->spa_config); 2753 spa_config_exit(spa, SCL_ALL, FTAG); 2754 nvlist_free(spa->spa_config_splitting); 2755 spa->spa_config_splitting = NULL; 2756 } 2757 2758 /* 2759 * Initialize internal SPA structures. 2760 */ 2761 spa_ld_select_uberblock_done(spa, ub); 2762 2763 return (0); 2764} 2765 2766static int 2767spa_ld_open_rootbp(spa_t *spa) 2768{ 2769 int error = 0; 2770 vdev_t *rvd = spa->spa_root_vdev; 2771 2772 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2773 if (error != 0) { 2774 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " 2775 "[error=%d]", error); 2776 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2777 } 2778 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2779 2780 return (0); 2781} 2782 2783static int 2784spa_ld_trusted_config(spa_t *spa, spa_import_type_t type, 2785 boolean_t reloading) 2786{ 2787 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 2788 nvlist_t *nv, *mos_config, *policy; 2789 int error = 0, copy_error; 2790 uint64_t healthy_tvds, healthy_tvds_mos; 2791 uint64_t mos_config_txg; 2792 2793 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) 2794 != 0) 2795 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2796 2797 /* 2798 * If we're assembling a pool from a split, the config provided is 2799 * already trusted so there is nothing to do. 2800 */ 2801 if (type == SPA_IMPORT_ASSEMBLE) 2802 return (0); 2803 2804 healthy_tvds = spa_healthy_core_tvds(spa); 2805 2806 if (load_nvlist(spa, spa->spa_config_object, &mos_config) 2807 != 0) { 2808 spa_load_failed(spa, "unable to retrieve MOS config"); 2809 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2810 } 2811 2812 /* 2813 * If we are doing an open, pool owner wasn't verified yet, thus do 2814 * the verification here. 2815 */ 2816 if (spa->spa_load_state == SPA_LOAD_OPEN) { 2817 error = spa_verify_host(spa, mos_config); 2818 if (error != 0) { 2819 nvlist_free(mos_config); 2820 return (error); 2821 } 2822 } 2823 2824 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); 2825 2826 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2827 2828 /* 2829 * Build a new vdev tree from the trusted config 2830 */ 2831 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 2832 2833 /* 2834 * Vdev paths in the MOS may be obsolete. If the untrusted config was 2835 * obtained by scanning /dev/dsk, then it will have the right vdev 2836 * paths. We update the trusted MOS config with this information. 2837 * We first try to copy the paths with vdev_copy_path_strict, which 2838 * succeeds only when both configs have exactly the same vdev tree. 2839 * If that fails, we fall back to a more flexible method that has a 2840 * best effort policy. 2841 */ 2842 copy_error = vdev_copy_path_strict(rvd, mrvd); 2843 if (copy_error != 0 || spa_load_print_vdev_tree) { 2844 spa_load_note(spa, "provided vdev tree:"); 2845 vdev_dbgmsg_print_tree(rvd, 2); 2846 spa_load_note(spa, "MOS vdev tree:"); 2847 vdev_dbgmsg_print_tree(mrvd, 2); 2848 } 2849 if (copy_error != 0) { 2850 spa_load_note(spa, "vdev_copy_path_strict failed, falling " 2851 "back to vdev_copy_path_relaxed"); 2852 vdev_copy_path_relaxed(rvd, mrvd); 2853 } 2854 2855 vdev_close(rvd); 2856 vdev_free(rvd); 2857 spa->spa_root_vdev = mrvd; 2858 rvd = mrvd; 2859 spa_config_exit(spa, SCL_ALL, FTAG); 2860 2861 /* 2862 * We will use spa_config if we decide to reload the spa or if spa_load 2863 * fails and we rewind. We must thus regenerate the config using the 2864 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to 2865 * pass settings on how to load the pool and is not stored in the MOS. 2866 * We copy it over to our new, trusted config. 2867 */ 2868 mos_config_txg = fnvlist_lookup_uint64(mos_config, 2869 ZPOOL_CONFIG_POOL_TXG); 2870 nvlist_free(mos_config); 2871 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); 2872 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY, 2873 &policy) == 0) 2874 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy); 2875 spa_config_set(spa, mos_config); 2876 spa->spa_config_source = SPA_CONFIG_SRC_MOS; 2877 2878 /* 2879 * Now that we got the config from the MOS, we should be more strict 2880 * in checking blkptrs and can make assumptions about the consistency 2881 * of the vdev tree. spa_trust_config must be set to true before opening 2882 * vdevs in order for them to be writeable. 2883 */ 2884 spa->spa_trust_config = B_TRUE; 2885 2886 /* 2887 * Open and validate the new vdev tree 2888 */ 2889 error = spa_ld_open_vdevs(spa); 2890 if (error != 0) 2891 return (error); 2892 2893 error = spa_ld_validate_vdevs(spa); 2894 if (error != 0) 2895 return (error); 2896 2897 if (copy_error != 0 || spa_load_print_vdev_tree) { 2898 spa_load_note(spa, "final vdev tree:"); 2899 vdev_dbgmsg_print_tree(rvd, 2); 2900 } 2901 2902 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && 2903 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { 2904 /* 2905 * Sanity check to make sure that we are indeed loading the 2906 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds 2907 * in the config provided and they happened to be the only ones 2908 * to have the latest uberblock, we could involuntarily perform 2909 * an extreme rewind. 2910 */ 2911 healthy_tvds_mos = spa_healthy_core_tvds(spa); 2912 if (healthy_tvds_mos - healthy_tvds >= 2913 SPA_SYNC_MIN_VDEVS) { 2914 spa_load_note(spa, "config provided misses too many " 2915 "top-level vdevs compared to MOS (%lld vs %lld). ", 2916 (u_longlong_t)healthy_tvds, 2917 (u_longlong_t)healthy_tvds_mos); 2918 spa_load_note(spa, "vdev tree:"); 2919 vdev_dbgmsg_print_tree(rvd, 2); 2920 if (reloading) { 2921 spa_load_failed(spa, "config was already " 2922 "provided from MOS. Aborting."); 2923 return (spa_vdev_err(rvd, 2924 VDEV_AUX_CORRUPT_DATA, EIO)); 2925 } 2926 spa_load_note(spa, "spa must be reloaded using MOS " 2927 "config"); 2928 return (SET_ERROR(EAGAIN)); 2929 } 2930 } 2931 2932 error = spa_check_for_missing_logs(spa); 2933 if (error != 0) 2934 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2935 2936 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { 2937 spa_load_failed(spa, "uberblock guid sum doesn't match MOS " 2938 "guid sum (%llu != %llu)", 2939 (u_longlong_t)spa->spa_uberblock.ub_guid_sum, 2940 (u_longlong_t)rvd->vdev_guid_sum); 2941 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2942 ENXIO)); 2943 } 2944 2945 return (0); 2946} 2947 2948static int 2949spa_ld_open_indirect_vdev_metadata(spa_t *spa) 2950{ 2951 int error = 0; 2952 vdev_t *rvd = spa->spa_root_vdev; 2953 2954 /* 2955 * Everything that we read before spa_remove_init() must be stored 2956 * on concreted vdevs. Therefore we do this as early as possible. 2957 */ 2958 error = spa_remove_init(spa); 2959 if (error != 0) { 2960 spa_load_failed(spa, "spa_remove_init failed [error=%d]", 2961 error); 2962 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2963 } 2964 2965 /* 2966 * Retrieve information needed to condense indirect vdev mappings. 2967 */ 2968 error = spa_condense_init(spa); 2969 if (error != 0) { 2970 spa_load_failed(spa, "spa_condense_init failed [error=%d]", 2971 error); 2972 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 2973 } 2974 2975 return (0); 2976} 2977 2978static int 2979spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) 2980{ 2981 int error = 0; 2982 vdev_t *rvd = spa->spa_root_vdev; 2983 2984 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2985 boolean_t missing_feat_read = B_FALSE; 2986 nvlist_t *unsup_feat, *enabled_feat; 2987 2988 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2989 &spa->spa_feat_for_read_obj, B_TRUE) != 0) { 2990 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2991 } 2992 2993 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2994 &spa->spa_feat_for_write_obj, B_TRUE) != 0) { 2995 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2996 } 2997 2998 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2999 &spa->spa_feat_desc_obj, B_TRUE) != 0) { 3000 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3001 } 3002 3003 enabled_feat = fnvlist_alloc(); 3004 unsup_feat = fnvlist_alloc(); 3005 3006 if (!spa_features_check(spa, B_FALSE, 3007 unsup_feat, enabled_feat)) 3008 missing_feat_read = B_TRUE; 3009 3010 if (spa_writeable(spa) || 3011 spa->spa_load_state == SPA_LOAD_TRYIMPORT) { 3012 if (!spa_features_check(spa, B_TRUE, 3013 unsup_feat, enabled_feat)) { 3014 *missing_feat_writep = B_TRUE; 3015 } 3016 } 3017 3018 fnvlist_add_nvlist(spa->spa_load_info, 3019 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 3020 3021 if (!nvlist_empty(unsup_feat)) { 3022 fnvlist_add_nvlist(spa->spa_load_info, 3023 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 3024 } 3025 3026 fnvlist_free(enabled_feat); 3027 fnvlist_free(unsup_feat); 3028 3029 if (!missing_feat_read) { 3030 fnvlist_add_boolean(spa->spa_load_info, 3031 ZPOOL_CONFIG_CAN_RDONLY); 3032 } 3033 3034 /* 3035 * If the state is SPA_LOAD_TRYIMPORT, our objective is 3036 * twofold: to determine whether the pool is available for 3037 * import in read-write mode and (if it is not) whether the 3038 * pool is available for import in read-only mode. If the pool 3039 * is available for import in read-write mode, it is displayed 3040 * as available in userland; if it is not available for import 3041 * in read-only mode, it is displayed as unavailable in 3042 * userland. If the pool is available for import in read-only 3043 * mode but not read-write mode, it is displayed as unavailable 3044 * in userland with a special note that the pool is actually 3045 * available for open in read-only mode. 3046 * 3047 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 3048 * missing a feature for write, we must first determine whether 3049 * the pool can be opened read-only before returning to 3050 * userland in order to know whether to display the 3051 * abovementioned note. 3052 */ 3053 if (missing_feat_read || (*missing_feat_writep && 3054 spa_writeable(spa))) { 3055 spa_load_failed(spa, "pool uses unsupported features"); 3056 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 3057 ENOTSUP)); 3058 } 3059 3060 /* 3061 * Load refcounts for ZFS features from disk into an in-memory 3062 * cache during SPA initialization. 3063 */ 3064 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 3065 uint64_t refcount; 3066 3067 error = feature_get_refcount_from_disk(spa, 3068 &spa_feature_table[i], &refcount); 3069 if (error == 0) { 3070 spa->spa_feat_refcount_cache[i] = refcount; 3071 } else if (error == ENOTSUP) { 3072 spa->spa_feat_refcount_cache[i] = 3073 SPA_FEATURE_DISABLED; 3074 } else { 3075 spa_load_failed(spa, "error getting refcount " 3076 "for feature %s [error=%d]", 3077 spa_feature_table[i].fi_guid, error); 3078 return (spa_vdev_err(rvd, 3079 VDEV_AUX_CORRUPT_DATA, EIO)); 3080 } 3081 } 3082 } 3083 3084 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 3085 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 3086 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) 3087 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3088 } 3089 3090 return (0); 3091} 3092 3093static int 3094spa_ld_load_special_directories(spa_t *spa) 3095{ 3096 int error = 0; 3097 vdev_t *rvd = spa->spa_root_vdev; 3098 3099 spa->spa_is_initializing = B_TRUE; 3100 error = dsl_pool_open(spa->spa_dsl_pool); 3101 spa->spa_is_initializing = B_FALSE; 3102 if (error != 0) { 3103 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); 3104 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3105 } 3106 3107 return (0); 3108} 3109 3110static int 3111spa_ld_get_props(spa_t *spa) 3112{ 3113 int error = 0; 3114 uint64_t obj; 3115 vdev_t *rvd = spa->spa_root_vdev; 3116 3117 /* Grab the secret checksum salt from the MOS. */ 3118 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3119 DMU_POOL_CHECKSUM_SALT, 1, 3120 sizeof (spa->spa_cksum_salt.zcs_bytes), 3121 spa->spa_cksum_salt.zcs_bytes); 3122 if (error == ENOENT) { 3123 /* Generate a new salt for subsequent use */ 3124 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3125 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3126 } else if (error != 0) { 3127 spa_load_failed(spa, "unable to retrieve checksum salt from " 3128 "MOS [error=%d]", error); 3129 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3130 } 3131 3132 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) 3133 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3134 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 3135 if (error != 0) { 3136 spa_load_failed(spa, "error opening deferred-frees bpobj " 3137 "[error=%d]", error); 3138 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3139 } 3140 3141 /* 3142 * Load the bit that tells us to use the new accounting function 3143 * (raid-z deflation). If we have an older pool, this will not 3144 * be present. 3145 */ 3146 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); 3147 if (error != 0 && error != ENOENT) 3148 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3149 3150 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 3151 &spa->spa_creation_version, B_FALSE); 3152 if (error != 0 && error != ENOENT) 3153 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3154 3155 /* 3156 * Load the persistent error log. If we have an older pool, this will 3157 * not be present. 3158 */ 3159 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, 3160 B_FALSE); 3161 if (error != 0 && error != ENOENT) 3162 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3163 3164 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 3165 &spa->spa_errlog_scrub, B_FALSE); 3166 if (error != 0 && error != ENOENT) 3167 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3168 3169 /* 3170 * Load the history object. If we have an older pool, this 3171 * will not be present. 3172 */ 3173 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); 3174 if (error != 0 && error != ENOENT) 3175 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3176 3177 /* 3178 * Load the per-vdev ZAP map. If we have an older pool, this will not 3179 * be present; in this case, defer its creation to a later time to 3180 * avoid dirtying the MOS this early / out of sync context. See 3181 * spa_sync_config_object. 3182 */ 3183 3184 /* The sentinel is only available in the MOS config. */ 3185 nvlist_t *mos_config; 3186 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { 3187 spa_load_failed(spa, "unable to retrieve MOS config"); 3188 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3189 } 3190 3191 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 3192 &spa->spa_all_vdev_zaps, B_FALSE); 3193 3194 if (error == ENOENT) { 3195 VERIFY(!nvlist_exists(mos_config, 3196 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 3197 spa->spa_avz_action = AVZ_ACTION_INITIALIZE; 3198 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3199 } else if (error != 0) { 3200 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3201 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 3202 /* 3203 * An older version of ZFS overwrote the sentinel value, so 3204 * we have orphaned per-vdev ZAPs in the MOS. Defer their 3205 * destruction to later; see spa_sync_config_object. 3206 */ 3207 spa->spa_avz_action = AVZ_ACTION_DESTROY; 3208 /* 3209 * We're assuming that no vdevs have had their ZAPs created 3210 * before this. Better be sure of it. 3211 */ 3212 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3213 } 3214 nvlist_free(mos_config); 3215 3216 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3217 3218 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, 3219 B_FALSE); 3220 if (error && error != ENOENT) 3221 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3222 3223 if (error == 0) { 3224 uint64_t autoreplace; 3225 3226 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 3227 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 3228 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 3229 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 3230 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 3231 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 3232 &spa->spa_dedup_ditto); 3233 3234 spa->spa_autoreplace = (autoreplace != 0); 3235 } 3236 3237 /* 3238 * If we are importing a pool with missing top-level vdevs, 3239 * we enforce that the pool doesn't panic or get suspended on 3240 * error since the likelihood of missing data is extremely high. 3241 */ 3242 if (spa->spa_missing_tvds > 0 && 3243 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && 3244 spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3245 spa_load_note(spa, "forcing failmode to 'continue' " 3246 "as some top level vdevs are missing"); 3247 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; 3248 } 3249 3250 return (0); 3251} 3252 3253static int 3254spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) 3255{ 3256 int error = 0; 3257 vdev_t *rvd = spa->spa_root_vdev; 3258 3259 /* 3260 * If we're assembling the pool from the split-off vdevs of 3261 * an existing pool, we don't want to attach the spares & cache 3262 * devices. 3263 */ 3264 3265 /* 3266 * Load any hot spares for this pool. 3267 */ 3268 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, 3269 B_FALSE); 3270 if (error != 0 && error != ENOENT) 3271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3272 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3273 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 3274 if (load_nvlist(spa, spa->spa_spares.sav_object, 3275 &spa->spa_spares.sav_config) != 0) { 3276 spa_load_failed(spa, "error loading spares nvlist"); 3277 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3278 } 3279 3280 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3281 spa_load_spares(spa); 3282 spa_config_exit(spa, SCL_ALL, FTAG); 3283 } else if (error == 0) { 3284 spa->spa_spares.sav_sync = B_TRUE; 3285 } 3286 3287 /* 3288 * Load any level 2 ARC devices for this pool. 3289 */ 3290 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 3291 &spa->spa_l2cache.sav_object, B_FALSE); 3292 if (error != 0 && error != ENOENT) 3293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3294 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3295 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 3296 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 3297 &spa->spa_l2cache.sav_config) != 0) { 3298 spa_load_failed(spa, "error loading l2cache nvlist"); 3299 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3300 } 3301 3302 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3303 spa_load_l2cache(spa); 3304 spa_config_exit(spa, SCL_ALL, FTAG); 3305 } else if (error == 0) { 3306 spa->spa_l2cache.sav_sync = B_TRUE; 3307 } 3308 3309 return (0); 3310} 3311 3312static int 3313spa_ld_load_vdev_metadata(spa_t *spa) 3314{ 3315 int error = 0; 3316 vdev_t *rvd = spa->spa_root_vdev; 3317 3318 /* 3319 * If the 'autoreplace' property is set, then post a resource notifying 3320 * the ZFS DE that it should not issue any faults for unopenable 3321 * devices. We also iterate over the vdevs, and post a sysevent for any 3322 * unopenable vdevs so that the normal autoreplace handler can take 3323 * over. 3324 */ 3325 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3326 spa_check_removed(spa->spa_root_vdev); 3327 /* 3328 * For the import case, this is done in spa_import(), because 3329 * at this point we're using the spare definitions from 3330 * the MOS config, not necessarily from the userland config. 3331 */ 3332 if (spa->spa_load_state != SPA_LOAD_IMPORT) { 3333 spa_aux_check_removed(&spa->spa_spares); 3334 spa_aux_check_removed(&spa->spa_l2cache); 3335 } 3336 } 3337 3338 /* 3339 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. 3340 */ 3341 error = vdev_load(rvd); 3342 if (error != 0) { 3343 spa_load_failed(spa, "vdev_load failed [error=%d]", error); 3344 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 3345 } 3346 3347 /* 3348 * Propagate the leaf DTLs we just loaded all the way up the vdev tree. 3349 */ 3350 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3351 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 3352 spa_config_exit(spa, SCL_ALL, FTAG); 3353 3354 return (0); 3355} 3356 3357static int 3358spa_ld_load_dedup_tables(spa_t *spa) 3359{ 3360 int error = 0; 3361 vdev_t *rvd = spa->spa_root_vdev; 3362 3363 error = ddt_load(spa); 3364 if (error != 0) { 3365 spa_load_failed(spa, "ddt_load failed [error=%d]", error); 3366 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3367 } 3368 3369 return (0); 3370} 3371 3372static int 3373spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport) 3374{ 3375 vdev_t *rvd = spa->spa_root_vdev; 3376 3377 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { 3378 boolean_t missing = spa_check_logs(spa); 3379 if (missing) { 3380 if (spa->spa_missing_tvds != 0) { 3381 spa_load_note(spa, "spa_check_logs failed " 3382 "so dropping the logs"); 3383 } else { 3384 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 3385 spa_load_failed(spa, "spa_check_logs failed"); 3386 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, 3387 ENXIO)); 3388 } 3389 } 3390 } 3391 3392 return (0); 3393} 3394 3395static int 3396spa_ld_verify_pool_data(spa_t *spa) 3397{ 3398 int error = 0; 3399 vdev_t *rvd = spa->spa_root_vdev; 3400 3401 /* 3402 * We've successfully opened the pool, verify that we're ready 3403 * to start pushing transactions. 3404 */ 3405 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3406 error = spa_load_verify(spa); 3407 if (error != 0) { 3408 spa_load_failed(spa, "spa_load_verify failed " 3409 "[error=%d]", error); 3410 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 3411 error)); 3412 } 3413 } 3414 3415 return (0); 3416} 3417 3418static void 3419spa_ld_claim_log_blocks(spa_t *spa) 3420{ 3421 dmu_tx_t *tx; 3422 dsl_pool_t *dp = spa_get_dsl(spa); 3423 3424 /* 3425 * Claim log blocks that haven't been committed yet. 3426 * This must all happen in a single txg. 3427 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 3428 * invoked from zil_claim_log_block()'s i/o done callback. 3429 * Price of rollback is that we abandon the log. 3430 */ 3431 spa->spa_claiming = B_TRUE; 3432 3433 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 3434 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 3435 zil_claim, tx, DS_FIND_CHILDREN); 3436 dmu_tx_commit(tx); 3437 3438 spa->spa_claiming = B_FALSE; 3439 3440 spa_set_log_state(spa, SPA_LOG_GOOD); 3441} 3442 3443static void 3444spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, 3445 boolean_t update_config_cache) 3446{ 3447 vdev_t *rvd = spa->spa_root_vdev; 3448 int need_update = B_FALSE; 3449 3450 /* 3451 * If the config cache is stale, or we have uninitialized 3452 * metaslabs (see spa_vdev_add()), then update the config. 3453 * 3454 * If this is a verbatim import, trust the current 3455 * in-core spa_config and update the disk labels. 3456 */ 3457 if (update_config_cache || config_cache_txg != spa->spa_config_txg || 3458 spa->spa_load_state == SPA_LOAD_IMPORT || 3459 spa->spa_load_state == SPA_LOAD_RECOVER || 3460 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 3461 need_update = B_TRUE; 3462 3463 for (int c = 0; c < rvd->vdev_children; c++) 3464 if (rvd->vdev_child[c]->vdev_ms_array == 0) 3465 need_update = B_TRUE; 3466 3467 /* 3468 * Update the config cache asychronously in case we're the 3469 * root pool, in which case the config cache isn't writable yet. 3470 */ 3471 if (need_update) 3472 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3473} 3474 3475static void 3476spa_ld_prepare_for_reload(spa_t *spa) 3477{ 3478 int mode = spa->spa_mode; 3479 int async_suspended = spa->spa_async_suspended; 3480 3481 spa_unload(spa); 3482 spa_deactivate(spa); 3483 spa_activate(spa, mode); 3484 3485 /* 3486 * We save the value of spa_async_suspended as it gets reset to 0 by 3487 * spa_unload(). We want to restore it back to the original value before 3488 * returning as we might be calling spa_async_resume() later. 3489 */ 3490 spa->spa_async_suspended = async_suspended; 3491} 3492 3493static int 3494spa_ld_read_checkpoint_txg(spa_t *spa) 3495{ 3496 uberblock_t checkpoint; 3497 int error = 0; 3498 3499 ASSERT0(spa->spa_checkpoint_txg); 3500 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3501 3502 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3503 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 3504 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 3505 3506 if (error == ENOENT) 3507 return (0); 3508 3509 if (error != 0) 3510 return (error); 3511 3512 ASSERT3U(checkpoint.ub_txg, !=, 0); 3513 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0); 3514 ASSERT3U(checkpoint.ub_timestamp, !=, 0); 3515 spa->spa_checkpoint_txg = checkpoint.ub_txg; 3516 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp; 3517 3518 return (0); 3519} 3520 3521static int 3522spa_ld_mos_init(spa_t *spa, spa_import_type_t type) 3523{ 3524 int error = 0; 3525 3526 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3527 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3528 3529 /* 3530 * Never trust the config that is provided unless we are assembling 3531 * a pool following a split. 3532 * This means don't trust blkptrs and the vdev tree in general. This 3533 * also effectively puts the spa in read-only mode since 3534 * spa_writeable() checks for spa_trust_config to be true. 3535 * We will later load a trusted config from the MOS. 3536 */ 3537 if (type != SPA_IMPORT_ASSEMBLE) 3538 spa->spa_trust_config = B_FALSE; 3539 3540 /* 3541 * Parse the config provided to create a vdev tree. 3542 */ 3543 error = spa_ld_parse_config(spa, type); 3544 if (error != 0) 3545 return (error); 3546 3547 /* 3548 * Now that we have the vdev tree, try to open each vdev. This involves 3549 * opening the underlying physical device, retrieving its geometry and 3550 * probing the vdev with a dummy I/O. The state of each vdev will be set 3551 * based on the success of those operations. After this we'll be ready 3552 * to read from the vdevs. 3553 */ 3554 error = spa_ld_open_vdevs(spa); 3555 if (error != 0) 3556 return (error); 3557 3558 /* 3559 * Read the label of each vdev and make sure that the GUIDs stored 3560 * there match the GUIDs in the config provided. 3561 * If we're assembling a new pool that's been split off from an 3562 * existing pool, the labels haven't yet been updated so we skip 3563 * validation for now. 3564 */ 3565 if (type != SPA_IMPORT_ASSEMBLE) { 3566 error = spa_ld_validate_vdevs(spa); 3567 if (error != 0) 3568 return (error); 3569 } 3570 3571 /* 3572 * Read all vdev labels to find the best uberblock (i.e. latest, 3573 * unless spa_load_max_txg is set) and store it in spa_uberblock. We 3574 * get the list of features required to read blkptrs in the MOS from 3575 * the vdev label with the best uberblock and verify that our version 3576 * of zfs supports them all. 3577 */ 3578 error = spa_ld_select_uberblock(spa, type); 3579 if (error != 0) 3580 return (error); 3581 3582 /* 3583 * Pass that uberblock to the dsl_pool layer which will open the root 3584 * blkptr. This blkptr points to the latest version of the MOS and will 3585 * allow us to read its contents. 3586 */ 3587 error = spa_ld_open_rootbp(spa); 3588 if (error != 0) 3589 return (error); 3590 3591 return (0); 3592} 3593 3594static int 3595spa_ld_checkpoint_rewind(spa_t *spa) 3596{ 3597 uberblock_t checkpoint; 3598 int error = 0; 3599 3600 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3601 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3602 3603 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3604 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t), 3605 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint); 3606 3607 if (error != 0) { 3608 spa_load_failed(spa, "unable to retrieve checkpointed " 3609 "uberblock from the MOS config [error=%d]", error); 3610 3611 if (error == ENOENT) 3612 error = ZFS_ERR_NO_CHECKPOINT; 3613 3614 return (error); 3615 } 3616 3617 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg); 3618 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg); 3619 3620 /* 3621 * We need to update the txg and timestamp of the checkpointed 3622 * uberblock to be higher than the latest one. This ensures that 3623 * the checkpointed uberblock is selected if we were to close and 3624 * reopen the pool right after we've written it in the vdev labels. 3625 * (also see block comment in vdev_uberblock_compare) 3626 */ 3627 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1; 3628 checkpoint.ub_timestamp = gethrestime_sec(); 3629 3630 /* 3631 * Set current uberblock to be the checkpointed uberblock. 3632 */ 3633 spa->spa_uberblock = checkpoint; 3634 3635 /* 3636 * If we are doing a normal rewind, then the pool is open for 3637 * writing and we sync the "updated" checkpointed uberblock to 3638 * disk. Once this is done, we've basically rewound the whole 3639 * pool and there is no way back. 3640 * 3641 * There are cases when we don't want to attempt and sync the 3642 * checkpointed uberblock to disk because we are opening a 3643 * pool as read-only. Specifically, verifying the checkpointed 3644 * state with zdb, and importing the checkpointed state to get 3645 * a "preview" of its content. 3646 */ 3647 if (spa_writeable(spa)) { 3648 vdev_t *rvd = spa->spa_root_vdev; 3649 3650 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3651 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; 3652 int svdcount = 0; 3653 int children = rvd->vdev_children; 3654 int c0 = spa_get_random(children); 3655 3656 for (int c = 0; c < children; c++) { 3657 vdev_t *vd = rvd->vdev_child[(c0 + c) % children]; 3658 3659 /* Stop when revisiting the first vdev */ 3660 if (c > 0 && svd[0] == vd) 3661 break; 3662 3663 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 3664 !vdev_is_concrete(vd)) 3665 continue; 3666 3667 svd[svdcount++] = vd; 3668 if (svdcount == SPA_SYNC_MIN_VDEVS) 3669 break; 3670 } 3671 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg); 3672 if (error == 0) 3673 spa->spa_last_synced_guid = rvd->vdev_guid; 3674 spa_config_exit(spa, SCL_ALL, FTAG); 3675 3676 if (error != 0) { 3677 spa_load_failed(spa, "failed to write checkpointed " 3678 "uberblock to the vdev labels [error=%d]", error); 3679 return (error); 3680 } 3681 } 3682 3683 return (0); 3684} 3685 3686static int 3687spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type, 3688 boolean_t *update_config_cache) 3689{ 3690 int error; 3691 3692 /* 3693 * Parse the config for pool, open and validate vdevs, 3694 * select an uberblock, and use that uberblock to open 3695 * the MOS. 3696 */ 3697 error = spa_ld_mos_init(spa, type); 3698 if (error != 0) 3699 return (error); 3700 3701 /* 3702 * Retrieve the trusted config stored in the MOS and use it to create 3703 * a new, exact version of the vdev tree, then reopen all vdevs. 3704 */ 3705 error = spa_ld_trusted_config(spa, type, B_FALSE); 3706 if (error == EAGAIN) { 3707 if (update_config_cache != NULL) 3708 *update_config_cache = B_TRUE; 3709 3710 /* 3711 * Redo the loading process with the trusted config if it is 3712 * too different from the untrusted config. 3713 */ 3714 spa_ld_prepare_for_reload(spa); 3715 spa_load_note(spa, "RELOADING"); 3716 error = spa_ld_mos_init(spa, type); 3717 if (error != 0) 3718 return (error); 3719 3720 error = spa_ld_trusted_config(spa, type, B_TRUE); 3721 if (error != 0) 3722 return (error); 3723 3724 } else if (error != 0) { 3725 return (error); 3726 } 3727 3728 return (0); 3729} 3730 3731/* 3732 * Load an existing storage pool, using the config provided. This config 3733 * describes which vdevs are part of the pool and is later validated against 3734 * partial configs present in each vdev's label and an entire copy of the 3735 * config stored in the MOS. 3736 */ 3737static int 3738spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport) 3739{ 3740 int error = 0; 3741 boolean_t missing_feat_write = B_FALSE; 3742 boolean_t checkpoint_rewind = 3743 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 3744 boolean_t update_config_cache = B_FALSE; 3745 3746 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3747 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3748 3749 spa_load_note(spa, "LOADING"); 3750 3751 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache); 3752 if (error != 0) 3753 return (error); 3754 3755 /* 3756 * If we are rewinding to the checkpoint then we need to repeat 3757 * everything we've done so far in this function but this time 3758 * selecting the checkpointed uberblock and using that to open 3759 * the MOS. 3760 */ 3761 if (checkpoint_rewind) { 3762 /* 3763 * If we are rewinding to the checkpoint update config cache 3764 * anyway. 3765 */ 3766 update_config_cache = B_TRUE; 3767 3768 /* 3769 * Extract the checkpointed uberblock from the current MOS 3770 * and use this as the pool's uberblock from now on. If the 3771 * pool is imported as writeable we also write the checkpoint 3772 * uberblock to the labels, making the rewind permanent. 3773 */ 3774 error = spa_ld_checkpoint_rewind(spa); 3775 if (error != 0) 3776 return (error); 3777 3778 /* 3779 * Redo the loading process process again with the 3780 * checkpointed uberblock. 3781 */ 3782 spa_ld_prepare_for_reload(spa); 3783 spa_load_note(spa, "LOADING checkpointed uberblock"); 3784 error = spa_ld_mos_with_trusted_config(spa, type, NULL); 3785 if (error != 0) 3786 return (error); 3787 } 3788 3789 /* 3790 * Retrieve the checkpoint txg if the pool has a checkpoint. 3791 */ 3792 error = spa_ld_read_checkpoint_txg(spa); 3793 if (error != 0) 3794 return (error); 3795 3796 /* 3797 * Retrieve the mapping of indirect vdevs. Those vdevs were removed 3798 * from the pool and their contents were re-mapped to other vdevs. Note 3799 * that everything that we read before this step must have been 3800 * rewritten on concrete vdevs after the last device removal was 3801 * initiated. Otherwise we could be reading from indirect vdevs before 3802 * we have loaded their mappings. 3803 */ 3804 error = spa_ld_open_indirect_vdev_metadata(spa); 3805 if (error != 0) 3806 return (error); 3807 3808 /* 3809 * Retrieve the full list of active features from the MOS and check if 3810 * they are all supported. 3811 */ 3812 error = spa_ld_check_features(spa, &missing_feat_write); 3813 if (error != 0) 3814 return (error); 3815 3816 /* 3817 * Load several special directories from the MOS needed by the dsl_pool 3818 * layer. 3819 */ 3820 error = spa_ld_load_special_directories(spa); 3821 if (error != 0) 3822 return (error); 3823 3824 /* 3825 * Retrieve pool properties from the MOS. 3826 */ 3827 error = spa_ld_get_props(spa); 3828 if (error != 0) 3829 return (error); 3830 3831 /* 3832 * Retrieve the list of auxiliary devices - cache devices and spares - 3833 * and open them. 3834 */ 3835 error = spa_ld_open_aux_vdevs(spa, type); 3836 if (error != 0) 3837 return (error); 3838 3839 /* 3840 * Load the metadata for all vdevs. Also check if unopenable devices 3841 * should be autoreplaced. 3842 */ 3843 error = spa_ld_load_vdev_metadata(spa); 3844 if (error != 0) 3845 return (error); 3846 3847 error = spa_ld_load_dedup_tables(spa); 3848 if (error != 0) 3849 return (error); 3850 3851 /* 3852 * Verify the logs now to make sure we don't have any unexpected errors 3853 * when we claim log blocks later. 3854 */ 3855 error = spa_ld_verify_logs(spa, type, ereport); 3856 if (error != 0) 3857 return (error); 3858 3859 if (missing_feat_write) { 3860 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); 3861 3862 /* 3863 * At this point, we know that we can open the pool in 3864 * read-only mode but not read-write mode. We now have enough 3865 * information and can return to userland. 3866 */ 3867 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, 3868 ENOTSUP)); 3869 } 3870 3871 /* 3872 * Traverse the last txgs to make sure the pool was left off in a safe 3873 * state. When performing an extreme rewind, we verify the whole pool, 3874 * which can take a very long time. 3875 */ 3876 error = spa_ld_verify_pool_data(spa); 3877 if (error != 0) 3878 return (error); 3879 3880 /* 3881 * Calculate the deflated space for the pool. This must be done before 3882 * we write anything to the pool because we'd need to update the space 3883 * accounting using the deflated sizes. 3884 */ 3885 spa_update_dspace(spa); 3886 3887 /* 3888 * We have now retrieved all the information we needed to open the 3889 * pool. If we are importing the pool in read-write mode, a few 3890 * additional steps must be performed to finish the import. 3891 */ 3892 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || 3893 spa->spa_load_max_txg == UINT64_MAX)) { 3894 uint64_t config_cache_txg = spa->spa_config_txg; 3895 3896 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); 3897 3898 /* 3899 * In case of a checkpoint rewind, log the original txg 3900 * of the checkpointed uberblock. 3901 */ 3902 if (checkpoint_rewind) { 3903 spa_history_log_internal(spa, "checkpoint rewind", 3904 NULL, "rewound state to txg=%llu", 3905 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg); 3906 } 3907 3908 /* 3909 * Traverse the ZIL and claim all blocks. 3910 */ 3911 spa_ld_claim_log_blocks(spa); 3912 3913 /* 3914 * Kick-off the syncing thread. 3915 */ 3916 spa->spa_sync_on = B_TRUE; 3917 txg_sync_start(spa->spa_dsl_pool); 3918 3919 /* 3920 * Wait for all claims to sync. We sync up to the highest 3921 * claimed log block birth time so that claimed log blocks 3922 * don't appear to be from the future. spa_claim_max_txg 3923 * will have been set for us by ZIL traversal operations 3924 * performed above. 3925 */ 3926 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 3927 3928 /* 3929 * Check if we need to request an update of the config. On the 3930 * next sync, we would update the config stored in vdev labels 3931 * and the cachefile (by default /etc/zfs/zpool.cache). 3932 */ 3933 spa_ld_check_for_config_update(spa, config_cache_txg, 3934 update_config_cache); 3935 3936 /* 3937 * Check all DTLs to see if anything needs resilvering. 3938 */ 3939 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 3940 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) 3941 spa_async_request(spa, SPA_ASYNC_RESILVER); 3942 3943 /* 3944 * Log the fact that we booted up (so that we can detect if 3945 * we rebooted in the middle of an operation). 3946 */ 3947 spa_history_log_version(spa, "open"); 3948 3949 /* 3950 * Delete any inconsistent datasets. 3951 */ 3952 (void) dmu_objset_find(spa_name(spa), 3953 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 3954 3955 /* 3956 * Clean up any stale temporary dataset userrefs. 3957 */ 3958 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 3959 3960 spa_restart_removal(spa); 3961 3962 spa_spawn_aux_threads(spa); 3963 3964 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3965 vdev_initialize_restart(spa->spa_root_vdev); 3966 spa_config_exit(spa, SCL_CONFIG, FTAG); 3967 } 3968 3969 spa_load_note(spa, "LOADED"); 3970 3971 return (0); 3972} 3973 3974static int 3975spa_load_retry(spa_t *spa, spa_load_state_t state) 3976{ 3977 int mode = spa->spa_mode; 3978 3979 spa_unload(spa); 3980 spa_deactivate(spa); 3981 3982 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 3983 3984 spa_activate(spa, mode); 3985 spa_async_suspend(spa); 3986 3987 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", 3988 (u_longlong_t)spa->spa_load_max_txg); 3989 3990 return (spa_load(spa, state, SPA_IMPORT_EXISTING)); 3991} 3992 3993/* 3994 * If spa_load() fails this function will try loading prior txg's. If 3995 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 3996 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 3997 * function will not rewind the pool and will return the same error as 3998 * spa_load(). 3999 */ 4000static int 4001spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, 4002 int rewind_flags) 4003{ 4004 nvlist_t *loadinfo = NULL; 4005 nvlist_t *config = NULL; 4006 int load_error, rewind_error; 4007 uint64_t safe_rewind_txg; 4008 uint64_t min_txg; 4009 4010 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 4011 spa->spa_load_max_txg = spa->spa_load_txg; 4012 spa_set_log_state(spa, SPA_LOG_CLEAR); 4013 } else { 4014 spa->spa_load_max_txg = max_request; 4015 if (max_request != UINT64_MAX) 4016 spa->spa_extreme_rewind = B_TRUE; 4017 } 4018 4019 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); 4020 if (load_error == 0) 4021 return (0); 4022 if (load_error == ZFS_ERR_NO_CHECKPOINT) { 4023 /* 4024 * When attempting checkpoint-rewind on a pool with no 4025 * checkpoint, we should not attempt to load uberblocks 4026 * from previous txgs when spa_load fails. 4027 */ 4028 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT); 4029 return (load_error); 4030 } 4031 4032 if (spa->spa_root_vdev != NULL) 4033 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4034 4035 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 4036 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 4037 4038 if (rewind_flags & ZPOOL_NEVER_REWIND) { 4039 nvlist_free(config); 4040 return (load_error); 4041 } 4042 4043 if (state == SPA_LOAD_RECOVER) { 4044 /* Price of rolling back is discarding txgs, including log */ 4045 spa_set_log_state(spa, SPA_LOG_CLEAR); 4046 } else { 4047 /* 4048 * If we aren't rolling back save the load info from our first 4049 * import attempt so that we can restore it after attempting 4050 * to rewind. 4051 */ 4052 loadinfo = spa->spa_load_info; 4053 spa->spa_load_info = fnvlist_alloc(); 4054 } 4055 4056 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 4057 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 4058 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 4059 TXG_INITIAL : safe_rewind_txg; 4060 4061 /* 4062 * Continue as long as we're finding errors, we're still within 4063 * the acceptable rewind range, and we're still finding uberblocks 4064 */ 4065 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 4066 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 4067 if (spa->spa_load_max_txg < safe_rewind_txg) 4068 spa->spa_extreme_rewind = B_TRUE; 4069 rewind_error = spa_load_retry(spa, state); 4070 } 4071 4072 spa->spa_extreme_rewind = B_FALSE; 4073 spa->spa_load_max_txg = UINT64_MAX; 4074 4075 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 4076 spa_config_set(spa, config); 4077 else 4078 nvlist_free(config); 4079 4080 if (state == SPA_LOAD_RECOVER) { 4081 ASSERT3P(loadinfo, ==, NULL); 4082 return (rewind_error); 4083 } else { 4084 /* Store the rewind info as part of the initial load info */ 4085 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 4086 spa->spa_load_info); 4087 4088 /* Restore the initial load info */ 4089 fnvlist_free(spa->spa_load_info); 4090 spa->spa_load_info = loadinfo; 4091 4092 return (load_error); 4093 } 4094} 4095 4096/* 4097 * Pool Open/Import 4098 * 4099 * The import case is identical to an open except that the configuration is sent 4100 * down from userland, instead of grabbed from the configuration cache. For the 4101 * case of an open, the pool configuration will exist in the 4102 * POOL_STATE_UNINITIALIZED state. 4103 * 4104 * The stats information (gen/count/ustats) is used to gather vdev statistics at 4105 * the same time open the pool, without having to keep around the spa_t in some 4106 * ambiguous state. 4107 */ 4108static int 4109spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 4110 nvlist_t **config) 4111{ 4112 spa_t *spa; 4113 spa_load_state_t state = SPA_LOAD_OPEN; 4114 int error; 4115 int locked = B_FALSE; 4116 int firstopen = B_FALSE; 4117 4118 *spapp = NULL; 4119 4120 /* 4121 * As disgusting as this is, we need to support recursive calls to this 4122 * function because dsl_dir_open() is called during spa_load(), and ends 4123 * up calling spa_open() again. The real fix is to figure out how to 4124 * avoid dsl_dir_open() calling this in the first place. 4125 */ 4126 if (mutex_owner(&spa_namespace_lock) != curthread) { 4127 mutex_enter(&spa_namespace_lock); 4128 locked = B_TRUE; 4129 } 4130 4131 if ((spa = spa_lookup(pool)) == NULL) { 4132 if (locked) 4133 mutex_exit(&spa_namespace_lock); 4134 return (SET_ERROR(ENOENT)); 4135 } 4136 4137 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 4138 zpool_load_policy_t policy; 4139 4140 firstopen = B_TRUE; 4141 4142 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config, 4143 &policy); 4144 if (policy.zlp_rewind & ZPOOL_DO_REWIND) 4145 state = SPA_LOAD_RECOVER; 4146 4147 spa_activate(spa, spa_mode_global); 4148 4149 if (state != SPA_LOAD_RECOVER) 4150 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4151 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 4152 4153 zfs_dbgmsg("spa_open_common: opening %s", pool); 4154 error = spa_load_best(spa, state, policy.zlp_txg, 4155 policy.zlp_rewind); 4156 4157 if (error == EBADF) { 4158 /* 4159 * If vdev_validate() returns failure (indicated by 4160 * EBADF), it indicates that one of the vdevs indicates 4161 * that the pool has been exported or destroyed. If 4162 * this is the case, the config cache is out of sync and 4163 * we should remove the pool from the namespace. 4164 */ 4165 spa_unload(spa); 4166 spa_deactivate(spa); 4167 spa_write_cachefile(spa, B_TRUE, B_TRUE); 4168 spa_remove(spa); 4169 if (locked) 4170 mutex_exit(&spa_namespace_lock); 4171 return (SET_ERROR(ENOENT)); 4172 } 4173 4174 if (error) { 4175 /* 4176 * We can't open the pool, but we still have useful 4177 * information: the state of each vdev after the 4178 * attempted vdev_open(). Return this to the user. 4179 */ 4180 if (config != NULL && spa->spa_config) { 4181 VERIFY(nvlist_dup(spa->spa_config, config, 4182 KM_SLEEP) == 0); 4183 VERIFY(nvlist_add_nvlist(*config, 4184 ZPOOL_CONFIG_LOAD_INFO, 4185 spa->spa_load_info) == 0); 4186 } 4187 spa_unload(spa); 4188 spa_deactivate(spa); 4189 spa->spa_last_open_failed = error; 4190 if (locked) 4191 mutex_exit(&spa_namespace_lock); 4192 *spapp = NULL; 4193 return (error); 4194 } 4195 } 4196 4197 spa_open_ref(spa, tag); 4198 4199 if (config != NULL) 4200 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4201 4202 /* 4203 * If we've recovered the pool, pass back any information we 4204 * gathered while doing the load. 4205 */ 4206 if (state == SPA_LOAD_RECOVER) { 4207 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 4208 spa->spa_load_info) == 0); 4209 } 4210 4211 if (locked) { 4212 spa->spa_last_open_failed = 0; 4213 spa->spa_last_ubsync_txg = 0; 4214 spa->spa_load_txg = 0; 4215 mutex_exit(&spa_namespace_lock); 4216#ifdef __FreeBSD__ 4217#ifdef _KERNEL 4218 if (firstopen) 4219 zvol_create_minors(spa->spa_name); 4220#endif 4221#endif 4222 } 4223 4224 *spapp = spa; 4225 4226 return (0); 4227} 4228 4229int 4230spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 4231 nvlist_t **config) 4232{ 4233 return (spa_open_common(name, spapp, tag, policy, config)); 4234} 4235 4236int 4237spa_open(const char *name, spa_t **spapp, void *tag) 4238{ 4239 return (spa_open_common(name, spapp, tag, NULL, NULL)); 4240} 4241 4242/* 4243 * Lookup the given spa_t, incrementing the inject count in the process, 4244 * preventing it from being exported or destroyed. 4245 */ 4246spa_t * 4247spa_inject_addref(char *name) 4248{ 4249 spa_t *spa; 4250 4251 mutex_enter(&spa_namespace_lock); 4252 if ((spa = spa_lookup(name)) == NULL) { 4253 mutex_exit(&spa_namespace_lock); 4254 return (NULL); 4255 } 4256 spa->spa_inject_ref++; 4257 mutex_exit(&spa_namespace_lock); 4258 4259 return (spa); 4260} 4261 4262void 4263spa_inject_delref(spa_t *spa) 4264{ 4265 mutex_enter(&spa_namespace_lock); 4266 spa->spa_inject_ref--; 4267 mutex_exit(&spa_namespace_lock); 4268} 4269 4270/* 4271 * Add spares device information to the nvlist. 4272 */ 4273static void 4274spa_add_spares(spa_t *spa, nvlist_t *config) 4275{ 4276 nvlist_t **spares; 4277 uint_t i, nspares; 4278 nvlist_t *nvroot; 4279 uint64_t guid; 4280 vdev_stat_t *vs; 4281 uint_t vsc; 4282 uint64_t pool; 4283 4284 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4285 4286 if (spa->spa_spares.sav_count == 0) 4287 return; 4288 4289 VERIFY(nvlist_lookup_nvlist(config, 4290 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4291 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 4292 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 4293 if (nspares != 0) { 4294 VERIFY(nvlist_add_nvlist_array(nvroot, 4295 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4296 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4297 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 4298 4299 /* 4300 * Go through and find any spares which have since been 4301 * repurposed as an active spare. If this is the case, update 4302 * their status appropriately. 4303 */ 4304 for (i = 0; i < nspares; i++) { 4305 VERIFY(nvlist_lookup_uint64(spares[i], 4306 ZPOOL_CONFIG_GUID, &guid) == 0); 4307 if (spa_spare_exists(guid, &pool, NULL) && 4308 pool != 0ULL) { 4309 VERIFY(nvlist_lookup_uint64_array( 4310 spares[i], ZPOOL_CONFIG_VDEV_STATS, 4311 (uint64_t **)&vs, &vsc) == 0); 4312 vs->vs_state = VDEV_STATE_CANT_OPEN; 4313 vs->vs_aux = VDEV_AUX_SPARED; 4314 } 4315 } 4316 } 4317} 4318 4319/* 4320 * Add l2cache device information to the nvlist, including vdev stats. 4321 */ 4322static void 4323spa_add_l2cache(spa_t *spa, nvlist_t *config) 4324{ 4325 nvlist_t **l2cache; 4326 uint_t i, j, nl2cache; 4327 nvlist_t *nvroot; 4328 uint64_t guid; 4329 vdev_t *vd; 4330 vdev_stat_t *vs; 4331 uint_t vsc; 4332 4333 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4334 4335 if (spa->spa_l2cache.sav_count == 0) 4336 return; 4337 4338 VERIFY(nvlist_lookup_nvlist(config, 4339 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4340 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 4341 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4342 if (nl2cache != 0) { 4343 VERIFY(nvlist_add_nvlist_array(nvroot, 4344 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4345 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4346 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4347 4348 /* 4349 * Update level 2 cache device stats. 4350 */ 4351 4352 for (i = 0; i < nl2cache; i++) { 4353 VERIFY(nvlist_lookup_uint64(l2cache[i], 4354 ZPOOL_CONFIG_GUID, &guid) == 0); 4355 4356 vd = NULL; 4357 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 4358 if (guid == 4359 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 4360 vd = spa->spa_l2cache.sav_vdevs[j]; 4361 break; 4362 } 4363 } 4364 ASSERT(vd != NULL); 4365 4366 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 4367 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 4368 == 0); 4369 vdev_get_stats(vd, vs); 4370 } 4371 } 4372} 4373 4374static void 4375spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features) 4376{ 4377 zap_cursor_t zc; 4378 zap_attribute_t za; 4379 4380 /* We may be unable to read features if pool is suspended. */ 4381 if (spa_suspended(spa)) 4382 return; 4383 4384 if (spa->spa_feat_for_read_obj != 0) { 4385 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4386 spa->spa_feat_for_read_obj); 4387 zap_cursor_retrieve(&zc, &za) == 0; 4388 zap_cursor_advance(&zc)) { 4389 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4390 za.za_num_integers == 1); 4391 VERIFY0(nvlist_add_uint64(features, za.za_name, 4392 za.za_first_integer)); 4393 } 4394 zap_cursor_fini(&zc); 4395 } 4396 4397 if (spa->spa_feat_for_write_obj != 0) { 4398 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4399 spa->spa_feat_for_write_obj); 4400 zap_cursor_retrieve(&zc, &za) == 0; 4401 zap_cursor_advance(&zc)) { 4402 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4403 za.za_num_integers == 1); 4404 VERIFY0(nvlist_add_uint64(features, za.za_name, 4405 za.za_first_integer)); 4406 } 4407 zap_cursor_fini(&zc); 4408 } 4409} 4410 4411static void 4412spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features) 4413{ 4414 int i; 4415 4416 for (i = 0; i < SPA_FEATURES; i++) { 4417 zfeature_info_t feature = spa_feature_table[i]; 4418 uint64_t refcount; 4419 4420 if (feature_get_refcount(spa, &feature, &refcount) != 0) 4421 continue; 4422 4423 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount)); 4424 } 4425} 4426 4427/* 4428 * Store a list of pool features and their reference counts in the 4429 * config. 4430 * 4431 * The first time this is called on a spa, allocate a new nvlist, fetch 4432 * the pool features and reference counts from disk, then save the list 4433 * in the spa. In subsequent calls on the same spa use the saved nvlist 4434 * and refresh its values from the cached reference counts. This 4435 * ensures we don't block here on I/O on a suspended pool so 'zpool 4436 * clear' can resume the pool. 4437 */ 4438static void 4439spa_add_feature_stats(spa_t *spa, nvlist_t *config) 4440{ 4441 nvlist_t *features; 4442 4443 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4444 4445 mutex_enter(&spa->spa_feat_stats_lock); 4446 features = spa->spa_feat_stats; 4447 4448 if (features != NULL) { 4449 spa_feature_stats_from_cache(spa, features); 4450 } else { 4451 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP)); 4452 spa->spa_feat_stats = features; 4453 spa_feature_stats_from_disk(spa, features); 4454 } 4455 4456 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 4457 features)); 4458 4459 mutex_exit(&spa->spa_feat_stats_lock); 4460} 4461 4462int 4463spa_get_stats(const char *name, nvlist_t **config, 4464 char *altroot, size_t buflen) 4465{ 4466 int error; 4467 spa_t *spa; 4468 4469 *config = NULL; 4470 error = spa_open_common(name, &spa, FTAG, NULL, config); 4471 4472 if (spa != NULL) { 4473 /* 4474 * This still leaves a window of inconsistency where the spares 4475 * or l2cache devices could change and the config would be 4476 * self-inconsistent. 4477 */ 4478 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4479 4480 if (*config != NULL) { 4481 uint64_t loadtimes[2]; 4482 4483 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 4484 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 4485 VERIFY(nvlist_add_uint64_array(*config, 4486 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 4487 4488 VERIFY(nvlist_add_uint64(*config, 4489 ZPOOL_CONFIG_ERRCOUNT, 4490 spa_get_errlog_size(spa)) == 0); 4491 4492 if (spa_suspended(spa)) 4493 VERIFY(nvlist_add_uint64(*config, 4494 ZPOOL_CONFIG_SUSPENDED, 4495 spa->spa_failmode) == 0); 4496 4497 spa_add_spares(spa, *config); 4498 spa_add_l2cache(spa, *config); 4499 spa_add_feature_stats(spa, *config); 4500 } 4501 } 4502 4503 /* 4504 * We want to get the alternate root even for faulted pools, so we cheat 4505 * and call spa_lookup() directly. 4506 */ 4507 if (altroot) { 4508 if (spa == NULL) { 4509 mutex_enter(&spa_namespace_lock); 4510 spa = spa_lookup(name); 4511 if (spa) 4512 spa_altroot(spa, altroot, buflen); 4513 else 4514 altroot[0] = '\0'; 4515 spa = NULL; 4516 mutex_exit(&spa_namespace_lock); 4517 } else { 4518 spa_altroot(spa, altroot, buflen); 4519 } 4520 } 4521 4522 if (spa != NULL) { 4523 spa_config_exit(spa, SCL_CONFIG, FTAG); 4524 spa_close(spa, FTAG); 4525 } 4526 4527 return (error); 4528} 4529 4530/* 4531 * Validate that the auxiliary device array is well formed. We must have an 4532 * array of nvlists, each which describes a valid leaf vdev. If this is an 4533 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 4534 * specified, as long as they are well-formed. 4535 */ 4536static int 4537spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 4538 spa_aux_vdev_t *sav, const char *config, uint64_t version, 4539 vdev_labeltype_t label) 4540{ 4541 nvlist_t **dev; 4542 uint_t i, ndev; 4543 vdev_t *vd; 4544 int error; 4545 4546 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4547 4548 /* 4549 * It's acceptable to have no devs specified. 4550 */ 4551 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 4552 return (0); 4553 4554 if (ndev == 0) 4555 return (SET_ERROR(EINVAL)); 4556 4557 /* 4558 * Make sure the pool is formatted with a version that supports this 4559 * device type. 4560 */ 4561 if (spa_version(spa) < version) 4562 return (SET_ERROR(ENOTSUP)); 4563 4564 /* 4565 * Set the pending device list so we correctly handle device in-use 4566 * checking. 4567 */ 4568 sav->sav_pending = dev; 4569 sav->sav_npending = ndev; 4570 4571 for (i = 0; i < ndev; i++) { 4572 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 4573 mode)) != 0) 4574 goto out; 4575 4576 if (!vd->vdev_ops->vdev_op_leaf) { 4577 vdev_free(vd); 4578 error = SET_ERROR(EINVAL); 4579 goto out; 4580 } 4581 4582 /* 4583 * The L2ARC currently only supports disk devices in 4584 * kernel context. For user-level testing, we allow it. 4585 */ 4586#ifdef _KERNEL 4587 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 4588 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 4589 error = SET_ERROR(ENOTBLK); 4590 vdev_free(vd); 4591 goto out; 4592 } 4593#endif 4594 vd->vdev_top = vd; 4595 4596 if ((error = vdev_open(vd)) == 0 && 4597 (error = vdev_label_init(vd, crtxg, label)) == 0) { 4598 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 4599 vd->vdev_guid) == 0); 4600 } 4601 4602 vdev_free(vd); 4603 4604 if (error && 4605 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 4606 goto out; 4607 else 4608 error = 0; 4609 } 4610 4611out: 4612 sav->sav_pending = NULL; 4613 sav->sav_npending = 0; 4614 return (error); 4615} 4616 4617static int 4618spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 4619{ 4620 int error; 4621 4622 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4623 4624 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4625 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 4626 VDEV_LABEL_SPARE)) != 0) { 4627 return (error); 4628 } 4629 4630 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4631 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 4632 VDEV_LABEL_L2CACHE)); 4633} 4634 4635static void 4636spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 4637 const char *config) 4638{ 4639 int i; 4640 4641 if (sav->sav_config != NULL) { 4642 nvlist_t **olddevs; 4643 uint_t oldndevs; 4644 nvlist_t **newdevs; 4645 4646 /* 4647 * Generate new dev list by concatentating with the 4648 * current dev list. 4649 */ 4650 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 4651 &olddevs, &oldndevs) == 0); 4652 4653 newdevs = kmem_alloc(sizeof (void *) * 4654 (ndevs + oldndevs), KM_SLEEP); 4655 for (i = 0; i < oldndevs; i++) 4656 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 4657 KM_SLEEP) == 0); 4658 for (i = 0; i < ndevs; i++) 4659 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 4660 KM_SLEEP) == 0); 4661 4662 VERIFY(nvlist_remove(sav->sav_config, config, 4663 DATA_TYPE_NVLIST_ARRAY) == 0); 4664 4665 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 4666 config, newdevs, ndevs + oldndevs) == 0); 4667 for (i = 0; i < oldndevs + ndevs; i++) 4668 nvlist_free(newdevs[i]); 4669 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 4670 } else { 4671 /* 4672 * Generate a new dev list. 4673 */ 4674 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 4675 KM_SLEEP) == 0); 4676 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 4677 devs, ndevs) == 0); 4678 } 4679} 4680 4681/* 4682 * Stop and drop level 2 ARC devices 4683 */ 4684void 4685spa_l2cache_drop(spa_t *spa) 4686{ 4687 vdev_t *vd; 4688 int i; 4689 spa_aux_vdev_t *sav = &spa->spa_l2cache; 4690 4691 for (i = 0; i < sav->sav_count; i++) { 4692 uint64_t pool; 4693 4694 vd = sav->sav_vdevs[i]; 4695 ASSERT(vd != NULL); 4696 4697 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 4698 pool != 0ULL && l2arc_vdev_present(vd)) 4699 l2arc_remove_vdev(vd); 4700 } 4701} 4702 4703/* 4704 * Pool Creation 4705 */ 4706int 4707spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 4708 nvlist_t *zplprops) 4709{ 4710 spa_t *spa; 4711 char *altroot = NULL; 4712 vdev_t *rvd; 4713 dsl_pool_t *dp; 4714 dmu_tx_t *tx; 4715 int error = 0; 4716 uint64_t txg = TXG_INITIAL; 4717 nvlist_t **spares, **l2cache; 4718 uint_t nspares, nl2cache; 4719 uint64_t version, obj; 4720 boolean_t has_features; 4721 char *poolname; 4722 nvlist_t *nvl; 4723 4724 if (nvlist_lookup_string(props, 4725 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0) 4726 poolname = (char *)pool; 4727 4728 /* 4729 * If this pool already exists, return failure. 4730 */ 4731 mutex_enter(&spa_namespace_lock); 4732 if (spa_lookup(poolname) != NULL) { 4733 mutex_exit(&spa_namespace_lock); 4734 return (SET_ERROR(EEXIST)); 4735 } 4736 4737 /* 4738 * Allocate a new spa_t structure. 4739 */ 4740 nvl = fnvlist_alloc(); 4741 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); 4742 (void) nvlist_lookup_string(props, 4743 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4744 spa = spa_add(poolname, nvl, altroot); 4745 fnvlist_free(nvl); 4746 spa_activate(spa, spa_mode_global); 4747 4748 if (props && (error = spa_prop_validate(spa, props))) { 4749 spa_deactivate(spa); 4750 spa_remove(spa); 4751 mutex_exit(&spa_namespace_lock); 4752 return (error); 4753 } 4754 4755 /* 4756 * Temporary pool names should never be written to disk. 4757 */ 4758 if (poolname != pool) 4759 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; 4760 4761 has_features = B_FALSE; 4762 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 4763 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 4764 if (zpool_prop_feature(nvpair_name(elem))) 4765 has_features = B_TRUE; 4766 } 4767 4768 if (has_features || nvlist_lookup_uint64(props, 4769 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 4770 version = SPA_VERSION; 4771 } 4772 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 4773 4774 spa->spa_first_txg = txg; 4775 spa->spa_uberblock.ub_txg = txg - 1; 4776 spa->spa_uberblock.ub_version = version; 4777 spa->spa_ubsync = spa->spa_uberblock; 4778 spa->spa_load_state = SPA_LOAD_CREATE; 4779 spa->spa_removing_phys.sr_state = DSS_NONE; 4780 spa->spa_removing_phys.sr_removing_vdev = -1; 4781 spa->spa_removing_phys.sr_prev_indirect_vdev = -1; 4782 spa->spa_indirect_vdevs_loaded = B_TRUE; 4783 4784 /* 4785 * Create "The Godfather" zio to hold all async IOs 4786 */ 4787 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 4788 KM_SLEEP); 4789 for (int i = 0; i < max_ncpus; i++) { 4790 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 4791 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 4792 ZIO_FLAG_GODFATHER); 4793 } 4794 4795 /* 4796 * Create the root vdev. 4797 */ 4798 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4799 4800 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 4801 4802 ASSERT(error != 0 || rvd != NULL); 4803 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 4804 4805 if (error == 0 && !zfs_allocatable_devs(nvroot)) 4806 error = SET_ERROR(EINVAL); 4807 4808 if (error == 0 && 4809 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 4810 (error = spa_validate_aux(spa, nvroot, txg, 4811 VDEV_ALLOC_ADD)) == 0) { 4812 for (int c = 0; c < rvd->vdev_children; c++) { 4813 vdev_ashift_optimize(rvd->vdev_child[c]); 4814 vdev_metaslab_set_size(rvd->vdev_child[c]); 4815 vdev_expand(rvd->vdev_child[c], txg); 4816 } 4817 } 4818 4819 spa_config_exit(spa, SCL_ALL, FTAG); 4820 4821 if (error != 0) { 4822 spa_unload(spa); 4823 spa_deactivate(spa); 4824 spa_remove(spa); 4825 mutex_exit(&spa_namespace_lock); 4826 return (error); 4827 } 4828 4829 /* 4830 * Get the list of spares, if specified. 4831 */ 4832 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4833 &spares, &nspares) == 0) { 4834 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 4835 KM_SLEEP) == 0); 4836 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4837 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4838 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4839 spa_load_spares(spa); 4840 spa_config_exit(spa, SCL_ALL, FTAG); 4841 spa->spa_spares.sav_sync = B_TRUE; 4842 } 4843 4844 /* 4845 * Get the list of level 2 cache devices, if specified. 4846 */ 4847 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4848 &l2cache, &nl2cache) == 0) { 4849 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4850 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4851 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4852 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4853 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4854 spa_load_l2cache(spa); 4855 spa_config_exit(spa, SCL_ALL, FTAG); 4856 spa->spa_l2cache.sav_sync = B_TRUE; 4857 } 4858 4859 spa->spa_is_initializing = B_TRUE; 4860 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 4861 spa->spa_meta_objset = dp->dp_meta_objset; 4862 spa->spa_is_initializing = B_FALSE; 4863 4864 /* 4865 * Create DDTs (dedup tables). 4866 */ 4867 ddt_create(spa); 4868 4869 spa_update_dspace(spa); 4870 4871 tx = dmu_tx_create_assigned(dp, txg); 4872 4873 /* 4874 * Create the pool config object. 4875 */ 4876 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 4877 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 4878 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 4879 4880 if (zap_add(spa->spa_meta_objset, 4881 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 4882 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 4883 cmn_err(CE_PANIC, "failed to add pool config"); 4884 } 4885 4886 if (spa_version(spa) >= SPA_VERSION_FEATURES) 4887 spa_feature_create_zap_objects(spa, tx); 4888 4889 if (zap_add(spa->spa_meta_objset, 4890 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 4891 sizeof (uint64_t), 1, &version, tx) != 0) { 4892 cmn_err(CE_PANIC, "failed to add pool version"); 4893 } 4894 4895 /* Newly created pools with the right version are always deflated. */ 4896 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 4897 spa->spa_deflate = TRUE; 4898 if (zap_add(spa->spa_meta_objset, 4899 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 4900 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 4901 cmn_err(CE_PANIC, "failed to add deflate"); 4902 } 4903 } 4904 4905 /* 4906 * Create the deferred-free bpobj. Turn off compression 4907 * because sync-to-convergence takes longer if the blocksize 4908 * keeps changing. 4909 */ 4910 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 4911 dmu_object_set_compress(spa->spa_meta_objset, obj, 4912 ZIO_COMPRESS_OFF, tx); 4913 if (zap_add(spa->spa_meta_objset, 4914 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 4915 sizeof (uint64_t), 1, &obj, tx) != 0) { 4916 cmn_err(CE_PANIC, "failed to add bpobj"); 4917 } 4918 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 4919 spa->spa_meta_objset, obj)); 4920 4921 /* 4922 * Create the pool's history object. 4923 */ 4924 if (version >= SPA_VERSION_ZPOOL_HISTORY) 4925 spa_history_create_obj(spa, tx); 4926 4927 /* 4928 * Generate some random noise for salted checksums to operate on. 4929 */ 4930 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 4931 sizeof (spa->spa_cksum_salt.zcs_bytes)); 4932 4933 /* 4934 * Set pool properties. 4935 */ 4936 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 4937 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 4938 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 4939 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 4940 4941 if (props != NULL) { 4942 spa_configfile_set(spa, props, B_FALSE); 4943 spa_sync_props(props, tx); 4944 } 4945 4946 dmu_tx_commit(tx); 4947 4948 spa->spa_sync_on = B_TRUE; 4949 txg_sync_start(spa->spa_dsl_pool); 4950 4951 /* 4952 * We explicitly wait for the first transaction to complete so that our 4953 * bean counters are appropriately updated. 4954 */ 4955 txg_wait_synced(spa->spa_dsl_pool, txg); 4956 4957 spa_spawn_aux_threads(spa); 4958 4959 spa_write_cachefile(spa, B_FALSE, B_TRUE); 4960 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); 4961 4962 spa_history_log_version(spa, "create"); 4963 4964 /* 4965 * Don't count references from objsets that are already closed 4966 * and are making their way through the eviction process. 4967 */ 4968 spa_evicting_os_wait(spa); 4969 spa->spa_minref = refcount_count(&spa->spa_refcount); 4970 spa->spa_load_state = SPA_LOAD_NONE; 4971 4972 mutex_exit(&spa_namespace_lock); 4973 4974 return (0); 4975} 4976 4977#ifdef _KERNEL 4978#ifdef illumos 4979/* 4980 * Get the root pool information from the root disk, then import the root pool 4981 * during the system boot up time. 4982 */ 4983extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 4984 4985static nvlist_t * 4986spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 4987{ 4988 nvlist_t *config; 4989 nvlist_t *nvtop, *nvroot; 4990 uint64_t pgid; 4991 4992 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 4993 return (NULL); 4994 4995 /* 4996 * Add this top-level vdev to the child array. 4997 */ 4998 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4999 &nvtop) == 0); 5000 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5001 &pgid) == 0); 5002 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 5003 5004 /* 5005 * Put this pool's top-level vdevs into a root vdev. 5006 */ 5007 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5008 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 5009 VDEV_TYPE_ROOT) == 0); 5010 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 5011 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 5012 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 5013 &nvtop, 1) == 0); 5014 5015 /* 5016 * Replace the existing vdev_tree with the new root vdev in 5017 * this pool's configuration (remove the old, add the new). 5018 */ 5019 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 5020 nvlist_free(nvroot); 5021 return (config); 5022} 5023 5024/* 5025 * Walk the vdev tree and see if we can find a device with "better" 5026 * configuration. A configuration is "better" if the label on that 5027 * device has a more recent txg. 5028 */ 5029static void 5030spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 5031{ 5032 for (int c = 0; c < vd->vdev_children; c++) 5033 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 5034 5035 if (vd->vdev_ops->vdev_op_leaf) { 5036 nvlist_t *label; 5037 uint64_t label_txg; 5038 5039 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 5040 &label) != 0) 5041 return; 5042 5043 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 5044 &label_txg) == 0); 5045 5046 /* 5047 * Do we have a better boot device? 5048 */ 5049 if (label_txg > *txg) { 5050 *txg = label_txg; 5051 *avd = vd; 5052 } 5053 nvlist_free(label); 5054 } 5055} 5056 5057/* 5058 * Import a root pool. 5059 * 5060 * For x86. devpath_list will consist of devid and/or physpath name of 5061 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 5062 * The GRUB "findroot" command will return the vdev we should boot. 5063 * 5064 * For Sparc, devpath_list consists the physpath name of the booting device 5065 * no matter the rootpool is a single device pool or a mirrored pool. 5066 * e.g. 5067 * "/pci@1f,0/ide@d/disk@0,0:a" 5068 */ 5069int 5070spa_import_rootpool(char *devpath, char *devid) 5071{ 5072 spa_t *spa; 5073 vdev_t *rvd, *bvd, *avd = NULL; 5074 nvlist_t *config, *nvtop; 5075 uint64_t guid, txg; 5076 char *pname; 5077 int error; 5078 5079 /* 5080 * Read the label from the boot device and generate a configuration. 5081 */ 5082 config = spa_generate_rootconf(devpath, devid, &guid); 5083#if defined(_OBP) && defined(_KERNEL) 5084 if (config == NULL) { 5085 if (strstr(devpath, "/iscsi/ssd") != NULL) { 5086 /* iscsi boot */ 5087 get_iscsi_bootpath_phy(devpath); 5088 config = spa_generate_rootconf(devpath, devid, &guid); 5089 } 5090 } 5091#endif 5092 if (config == NULL) { 5093 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 5094 devpath); 5095 return (SET_ERROR(EIO)); 5096 } 5097 5098 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 5099 &pname) == 0); 5100 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 5101 5102 mutex_enter(&spa_namespace_lock); 5103 if ((spa = spa_lookup(pname)) != NULL) { 5104 /* 5105 * Remove the existing root pool from the namespace so that we 5106 * can replace it with the correct config we just read in. 5107 */ 5108 spa_remove(spa); 5109 } 5110 5111 spa = spa_add(pname, config, NULL); 5112 spa->spa_is_root = B_TRUE; 5113 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 5114 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 5115 &spa->spa_ubsync.ub_version) != 0) 5116 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 5117 5118 /* 5119 * Build up a vdev tree based on the boot device's label config. 5120 */ 5121 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5122 &nvtop) == 0); 5123 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5124 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 5125 VDEV_ALLOC_ROOTPOOL); 5126 spa_config_exit(spa, SCL_ALL, FTAG); 5127 if (error) { 5128 mutex_exit(&spa_namespace_lock); 5129 nvlist_free(config); 5130 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 5131 pname); 5132 return (error); 5133 } 5134 5135 /* 5136 * Get the boot vdev. 5137 */ 5138 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 5139 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 5140 (u_longlong_t)guid); 5141 error = SET_ERROR(ENOENT); 5142 goto out; 5143 } 5144 5145 /* 5146 * Determine if there is a better boot device. 5147 */ 5148 avd = bvd; 5149 spa_alt_rootvdev(rvd, &avd, &txg); 5150 if (avd != bvd) { 5151 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 5152 "try booting from '%s'", avd->vdev_path); 5153 error = SET_ERROR(EINVAL); 5154 goto out; 5155 } 5156 5157 /* 5158 * If the boot device is part of a spare vdev then ensure that 5159 * we're booting off the active spare. 5160 */ 5161 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 5162 !bvd->vdev_isspare) { 5163 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 5164 "try booting from '%s'", 5165 bvd->vdev_parent-> 5166 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 5167 error = SET_ERROR(EINVAL); 5168 goto out; 5169 } 5170 5171 error = 0; 5172out: 5173 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5174 vdev_free(rvd); 5175 spa_config_exit(spa, SCL_ALL, FTAG); 5176 mutex_exit(&spa_namespace_lock); 5177 5178 nvlist_free(config); 5179 return (error); 5180} 5181 5182#else /* !illumos */ 5183 5184extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 5185 uint64_t *count); 5186 5187static nvlist_t * 5188spa_generate_rootconf(const char *name) 5189{ 5190 nvlist_t **configs, **tops; 5191 nvlist_t *config; 5192 nvlist_t *best_cfg, *nvtop, *nvroot; 5193 uint64_t *holes; 5194 uint64_t best_txg; 5195 uint64_t nchildren; 5196 uint64_t pgid; 5197 uint64_t count; 5198 uint64_t i; 5199 uint_t nholes; 5200 5201 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 5202 return (NULL); 5203 5204 ASSERT3U(count, !=, 0); 5205 best_txg = 0; 5206 for (i = 0; i < count; i++) { 5207 uint64_t txg; 5208 5209 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 5210 &txg) == 0); 5211 if (txg > best_txg) { 5212 best_txg = txg; 5213 best_cfg = configs[i]; 5214 } 5215 } 5216 5217 nchildren = 1; 5218 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 5219 holes = NULL; 5220 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 5221 &holes, &nholes); 5222 5223 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 5224 for (i = 0; i < nchildren; i++) { 5225 if (i >= count) 5226 break; 5227 if (configs[i] == NULL) 5228 continue; 5229 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 5230 &nvtop) == 0); 5231 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 5232 } 5233 for (i = 0; holes != NULL && i < nholes; i++) { 5234 if (i >= nchildren) 5235 continue; 5236 if (tops[holes[i]] != NULL) 5237 continue; 5238 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 5239 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 5240 VDEV_TYPE_HOLE) == 0); 5241 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 5242 holes[i]) == 0); 5243 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 5244 0) == 0); 5245 } 5246 for (i = 0; i < nchildren; i++) { 5247 if (tops[i] != NULL) 5248 continue; 5249 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 5250 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 5251 VDEV_TYPE_MISSING) == 0); 5252 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 5253 i) == 0); 5254 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 5255 0) == 0); 5256 } 5257 5258 /* 5259 * Create pool config based on the best vdev config. 5260 */ 5261 nvlist_dup(best_cfg, &config, KM_SLEEP); 5262 5263 /* 5264 * Put this pool's top-level vdevs into a root vdev. 5265 */ 5266 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5267 &pgid) == 0); 5268 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5269 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 5270 VDEV_TYPE_ROOT) == 0); 5271 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 5272 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 5273 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 5274 tops, nchildren) == 0); 5275 5276 /* 5277 * Replace the existing vdev_tree with the new root vdev in 5278 * this pool's configuration (remove the old, add the new). 5279 */ 5280 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 5281 5282 /* 5283 * Drop vdev config elements that should not be present at pool level. 5284 */ 5285 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 5286 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 5287 5288 for (i = 0; i < count; i++) 5289 nvlist_free(configs[i]); 5290 kmem_free(configs, count * sizeof(void *)); 5291 for (i = 0; i < nchildren; i++) 5292 nvlist_free(tops[i]); 5293 kmem_free(tops, nchildren * sizeof(void *)); 5294 nvlist_free(nvroot); 5295 return (config); 5296} 5297 5298int 5299spa_import_rootpool(const char *name) 5300{ 5301 spa_t *spa; 5302 vdev_t *rvd, *bvd, *avd = NULL; 5303 nvlist_t *config, *nvtop; 5304 uint64_t txg; 5305 char *pname; 5306 int error; 5307 5308 /* 5309 * Read the label from the boot device and generate a configuration. 5310 */ 5311 config = spa_generate_rootconf(name); 5312 5313 mutex_enter(&spa_namespace_lock); 5314 if (config != NULL) { 5315 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 5316 &pname) == 0 && strcmp(name, pname) == 0); 5317 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 5318 == 0); 5319 5320 if ((spa = spa_lookup(pname)) != NULL) { 5321 /* 5322 * The pool could already be imported, 5323 * e.g., after reboot -r. 5324 */ 5325 if (spa->spa_state == POOL_STATE_ACTIVE) { 5326 mutex_exit(&spa_namespace_lock); 5327 nvlist_free(config); 5328 return (0); 5329 } 5330 5331 /* 5332 * Remove the existing root pool from the namespace so 5333 * that we can replace it with the correct config 5334 * we just read in. 5335 */ 5336 spa_remove(spa); 5337 } 5338 spa = spa_add(pname, config, NULL); 5339 5340 /* 5341 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 5342 * via spa_version(). 5343 */ 5344 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 5345 &spa->spa_ubsync.ub_version) != 0) 5346 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 5347 } else if ((spa = spa_lookup(name)) == NULL) { 5348 mutex_exit(&spa_namespace_lock); 5349 nvlist_free(config); 5350 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 5351 name); 5352 return (EIO); 5353 } else { 5354 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 5355 } 5356 spa->spa_is_root = B_TRUE; 5357 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 5358 5359 /* 5360 * Build up a vdev tree based on the boot device's label config. 5361 */ 5362 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5363 &nvtop) == 0); 5364 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5365 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 5366 VDEV_ALLOC_ROOTPOOL); 5367 spa_config_exit(spa, SCL_ALL, FTAG); 5368 if (error) { 5369 mutex_exit(&spa_namespace_lock); 5370 nvlist_free(config); 5371 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 5372 pname); 5373 return (error); 5374 } 5375 5376 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5377 vdev_free(rvd); 5378 spa_config_exit(spa, SCL_ALL, FTAG); 5379 mutex_exit(&spa_namespace_lock); 5380 5381 nvlist_free(config); 5382 return (0); 5383} 5384 5385#endif /* illumos */ 5386#endif /* _KERNEL */ 5387 5388/* 5389 * Import a non-root pool into the system. 5390 */ 5391int 5392spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 5393{ 5394 spa_t *spa; 5395 char *altroot = NULL; 5396 spa_load_state_t state = SPA_LOAD_IMPORT; 5397 zpool_load_policy_t policy; 5398 uint64_t mode = spa_mode_global; 5399 uint64_t readonly = B_FALSE; 5400 int error; 5401 nvlist_t *nvroot; 5402 nvlist_t **spares, **l2cache; 5403 uint_t nspares, nl2cache; 5404 5405 /* 5406 * If a pool with this name exists, return failure. 5407 */ 5408 mutex_enter(&spa_namespace_lock); 5409 if (spa_lookup(pool) != NULL) { 5410 mutex_exit(&spa_namespace_lock); 5411 return (SET_ERROR(EEXIST)); 5412 } 5413 5414 /* 5415 * Create and initialize the spa structure. 5416 */ 5417 (void) nvlist_lookup_string(props, 5418 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5419 (void) nvlist_lookup_uint64(props, 5420 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 5421 if (readonly) 5422 mode = FREAD; 5423 spa = spa_add(pool, config, altroot); 5424 spa->spa_import_flags = flags; 5425 5426 /* 5427 * Verbatim import - Take a pool and insert it into the namespace 5428 * as if it had been loaded at boot. 5429 */ 5430 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 5431 if (props != NULL) 5432 spa_configfile_set(spa, props, B_FALSE); 5433 5434 spa_write_cachefile(spa, B_FALSE, B_TRUE); 5435 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5436 zfs_dbgmsg("spa_import: verbatim import of %s", pool); 5437 mutex_exit(&spa_namespace_lock); 5438 return (0); 5439 } 5440 5441 spa_activate(spa, mode); 5442 5443 /* 5444 * Don't start async tasks until we know everything is healthy. 5445 */ 5446 spa_async_suspend(spa); 5447 5448 zpool_get_load_policy(config, &policy); 5449 if (policy.zlp_rewind & ZPOOL_DO_REWIND) 5450 state = SPA_LOAD_RECOVER; 5451 5452 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; 5453 5454 if (state != SPA_LOAD_RECOVER) { 5455 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 5456 zfs_dbgmsg("spa_import: importing %s", pool); 5457 } else { 5458 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " 5459 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg); 5460 } 5461 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind); 5462 5463 /* 5464 * Propagate anything learned while loading the pool and pass it 5465 * back to caller (i.e. rewind info, missing devices, etc). 5466 */ 5467 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5468 spa->spa_load_info) == 0); 5469 5470 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5471 /* 5472 * Toss any existing sparelist, as it doesn't have any validity 5473 * anymore, and conflicts with spa_has_spare(). 5474 */ 5475 if (spa->spa_spares.sav_config) { 5476 nvlist_free(spa->spa_spares.sav_config); 5477 spa->spa_spares.sav_config = NULL; 5478 spa_load_spares(spa); 5479 } 5480 if (spa->spa_l2cache.sav_config) { 5481 nvlist_free(spa->spa_l2cache.sav_config); 5482 spa->spa_l2cache.sav_config = NULL; 5483 spa_load_l2cache(spa); 5484 } 5485 5486 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5487 &nvroot) == 0); 5488 if (error == 0) 5489 error = spa_validate_aux(spa, nvroot, -1ULL, 5490 VDEV_ALLOC_SPARE); 5491 if (error == 0) 5492 error = spa_validate_aux(spa, nvroot, -1ULL, 5493 VDEV_ALLOC_L2CACHE); 5494 spa_config_exit(spa, SCL_ALL, FTAG); 5495 5496 if (props != NULL) 5497 spa_configfile_set(spa, props, B_FALSE); 5498 5499 if (error != 0 || (props && spa_writeable(spa) && 5500 (error = spa_prop_set(spa, props)))) { 5501 spa_unload(spa); 5502 spa_deactivate(spa); 5503 spa_remove(spa); 5504 mutex_exit(&spa_namespace_lock); 5505 return (error); 5506 } 5507 5508 spa_async_resume(spa); 5509 5510 /* 5511 * Override any spares and level 2 cache devices as specified by 5512 * the user, as these may have correct device names/devids, etc. 5513 */ 5514 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 5515 &spares, &nspares) == 0) { 5516 if (spa->spa_spares.sav_config) 5517 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 5518 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 5519 else 5520 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 5521 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5522 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 5523 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 5524 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5525 spa_load_spares(spa); 5526 spa_config_exit(spa, SCL_ALL, FTAG); 5527 spa->spa_spares.sav_sync = B_TRUE; 5528 } 5529 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 5530 &l2cache, &nl2cache) == 0) { 5531 if (spa->spa_l2cache.sav_config) 5532 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 5533 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 5534 else 5535 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 5536 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5537 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 5538 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 5539 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5540 spa_load_l2cache(spa); 5541 spa_config_exit(spa, SCL_ALL, FTAG); 5542 spa->spa_l2cache.sav_sync = B_TRUE; 5543 } 5544 5545 /* 5546 * Check for any removed devices. 5547 */ 5548 if (spa->spa_autoreplace) { 5549 spa_aux_check_removed(&spa->spa_spares); 5550 spa_aux_check_removed(&spa->spa_l2cache); 5551 } 5552 5553 if (spa_writeable(spa)) { 5554 /* 5555 * Update the config cache to include the newly-imported pool. 5556 */ 5557 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5558 } 5559 5560 /* 5561 * It's possible that the pool was expanded while it was exported. 5562 * We kick off an async task to handle this for us. 5563 */ 5564 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 5565 5566 spa_history_log_version(spa, "import"); 5567 5568 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5569 5570 mutex_exit(&spa_namespace_lock); 5571 5572#ifdef __FreeBSD__ 5573#ifdef _KERNEL 5574 zvol_create_minors(pool); 5575#endif 5576#endif 5577 return (0); 5578} 5579 5580nvlist_t * 5581spa_tryimport(nvlist_t *tryconfig) 5582{ 5583 nvlist_t *config = NULL; 5584 char *poolname, *cachefile; 5585 spa_t *spa; 5586 uint64_t state; 5587 int error; 5588 zpool_load_policy_t policy; 5589 5590 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 5591 return (NULL); 5592 5593 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 5594 return (NULL); 5595 5596 /* 5597 * Create and initialize the spa structure. 5598 */ 5599 mutex_enter(&spa_namespace_lock); 5600 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 5601 spa_activate(spa, FREAD); 5602 5603 /* 5604 * Rewind pool if a max txg was provided. 5605 */ 5606 zpool_get_load_policy(spa->spa_config, &policy); 5607 if (policy.zlp_txg != UINT64_MAX) { 5608 spa->spa_load_max_txg = policy.zlp_txg; 5609 spa->spa_extreme_rewind = B_TRUE; 5610 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", 5611 poolname, (longlong_t)policy.zlp_txg); 5612 } else { 5613 zfs_dbgmsg("spa_tryimport: importing %s", poolname); 5614 } 5615 5616 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) 5617 == 0) { 5618 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); 5619 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 5620 } else { 5621 spa->spa_config_source = SPA_CONFIG_SRC_SCAN; 5622 } 5623 5624 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); 5625 5626 /* 5627 * If 'tryconfig' was at least parsable, return the current config. 5628 */ 5629 if (spa->spa_root_vdev != NULL) { 5630 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 5631 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 5632 poolname) == 0); 5633 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5634 state) == 0); 5635 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 5636 spa->spa_uberblock.ub_timestamp) == 0); 5637 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5638 spa->spa_load_info) == 0); 5639 5640 /* 5641 * If the bootfs property exists on this pool then we 5642 * copy it out so that external consumers can tell which 5643 * pools are bootable. 5644 */ 5645 if ((!error || error == EEXIST) && spa->spa_bootfs) { 5646 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5647 5648 /* 5649 * We have to play games with the name since the 5650 * pool was opened as TRYIMPORT_NAME. 5651 */ 5652 if (dsl_dsobj_to_dsname(spa_name(spa), 5653 spa->spa_bootfs, tmpname) == 0) { 5654 char *cp; 5655 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5656 5657 cp = strchr(tmpname, '/'); 5658 if (cp == NULL) { 5659 (void) strlcpy(dsname, tmpname, 5660 MAXPATHLEN); 5661 } else { 5662 (void) snprintf(dsname, MAXPATHLEN, 5663 "%s/%s", poolname, ++cp); 5664 } 5665 VERIFY(nvlist_add_string(config, 5666 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 5667 kmem_free(dsname, MAXPATHLEN); 5668 } 5669 kmem_free(tmpname, MAXPATHLEN); 5670 } 5671 5672 /* 5673 * Add the list of hot spares and level 2 cache devices. 5674 */ 5675 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5676 spa_add_spares(spa, config); 5677 spa_add_l2cache(spa, config); 5678 spa_config_exit(spa, SCL_CONFIG, FTAG); 5679 } 5680 5681 spa_unload(spa); 5682 spa_deactivate(spa); 5683 spa_remove(spa); 5684 mutex_exit(&spa_namespace_lock); 5685 5686 return (config); 5687} 5688 5689/* 5690 * Pool export/destroy 5691 * 5692 * The act of destroying or exporting a pool is very simple. We make sure there 5693 * is no more pending I/O and any references to the pool are gone. Then, we 5694 * update the pool state and sync all the labels to disk, removing the 5695 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 5696 * we don't sync the labels or remove the configuration cache. 5697 */ 5698static int 5699spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 5700 boolean_t force, boolean_t hardforce) 5701{ 5702 spa_t *spa; 5703 5704 if (oldconfig) 5705 *oldconfig = NULL; 5706 5707 if (!(spa_mode_global & FWRITE)) 5708 return (SET_ERROR(EROFS)); 5709 5710 mutex_enter(&spa_namespace_lock); 5711 if ((spa = spa_lookup(pool)) == NULL) { 5712 mutex_exit(&spa_namespace_lock); 5713 return (SET_ERROR(ENOENT)); 5714 } 5715 5716 /* 5717 * Put a hold on the pool, drop the namespace lock, stop async tasks, 5718 * reacquire the namespace lock, and see if we can export. 5719 */ 5720 spa_open_ref(spa, FTAG); 5721 mutex_exit(&spa_namespace_lock); 5722 spa_async_suspend(spa); 5723 mutex_enter(&spa_namespace_lock); 5724 spa_close(spa, FTAG); 5725 5726 /* 5727 * The pool will be in core if it's openable, 5728 * in which case we can modify its state. 5729 */ 5730 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 5731 5732 /* 5733 * Objsets may be open only because they're dirty, so we 5734 * have to force it to sync before checking spa_refcnt. 5735 */ 5736 txg_wait_synced(spa->spa_dsl_pool, 0); 5737 spa_evicting_os_wait(spa); 5738 5739 /* 5740 * A pool cannot be exported or destroyed if there are active 5741 * references. If we are resetting a pool, allow references by 5742 * fault injection handlers. 5743 */ 5744 if (!spa_refcount_zero(spa) || 5745 (spa->spa_inject_ref != 0 && 5746 new_state != POOL_STATE_UNINITIALIZED)) { 5747 spa_async_resume(spa); 5748 mutex_exit(&spa_namespace_lock); 5749 return (SET_ERROR(EBUSY)); 5750 } 5751 5752 /* 5753 * A pool cannot be exported if it has an active shared spare. 5754 * This is to prevent other pools stealing the active spare 5755 * from an exported pool. At user's own will, such pool can 5756 * be forcedly exported. 5757 */ 5758 if (!force && new_state == POOL_STATE_EXPORTED && 5759 spa_has_active_shared_spare(spa)) { 5760 spa_async_resume(spa); 5761 mutex_exit(&spa_namespace_lock); 5762 return (SET_ERROR(EXDEV)); 5763 } 5764 5765 /* 5766 * We're about to export or destroy this pool. Make sure 5767 * we stop all initializtion activity here before we 5768 * set the spa_final_txg. This will ensure that all 5769 * dirty data resulting from the initialization is 5770 * committed to disk before we unload the pool. 5771 */ 5772 if (spa->spa_root_vdev != NULL) { 5773 vdev_initialize_stop_all(spa->spa_root_vdev, 5774 VDEV_INITIALIZE_ACTIVE); 5775 } 5776 5777 /* 5778 * We want this to be reflected on every label, 5779 * so mark them all dirty. spa_unload() will do the 5780 * final sync that pushes these changes out. 5781 */ 5782 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 5783 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5784 spa->spa_state = new_state; 5785 spa->spa_final_txg = spa_last_synced_txg(spa) + 5786 TXG_DEFER_SIZE + 1; 5787 vdev_config_dirty(spa->spa_root_vdev); 5788 spa_config_exit(spa, SCL_ALL, FTAG); 5789 } 5790 } 5791 5792 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); 5793 5794 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 5795 spa_unload(spa); 5796 spa_deactivate(spa); 5797 } 5798 5799 if (oldconfig && spa->spa_config) 5800 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 5801 5802 if (new_state != POOL_STATE_UNINITIALIZED) { 5803 if (!hardforce) 5804 spa_write_cachefile(spa, B_TRUE, B_TRUE); 5805 spa_remove(spa); 5806 } 5807 mutex_exit(&spa_namespace_lock); 5808 5809 return (0); 5810} 5811 5812/* 5813 * Destroy a storage pool. 5814 */ 5815int 5816spa_destroy(char *pool) 5817{ 5818 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 5819 B_FALSE, B_FALSE)); 5820} 5821 5822/* 5823 * Export a storage pool. 5824 */ 5825int 5826spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 5827 boolean_t hardforce) 5828{ 5829 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 5830 force, hardforce)); 5831} 5832 5833/* 5834 * Similar to spa_export(), this unloads the spa_t without actually removing it 5835 * from the namespace in any way. 5836 */ 5837int 5838spa_reset(char *pool) 5839{ 5840 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 5841 B_FALSE, B_FALSE)); 5842} 5843 5844/* 5845 * ========================================================================== 5846 * Device manipulation 5847 * ========================================================================== 5848 */ 5849 5850/* 5851 * Add a device to a storage pool. 5852 */ 5853int 5854spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 5855{ 5856 uint64_t txg, id; 5857 int error; 5858 vdev_t *rvd = spa->spa_root_vdev; 5859 vdev_t *vd, *tvd; 5860 nvlist_t **spares, **l2cache; 5861 uint_t nspares, nl2cache; 5862 5863 ASSERT(spa_writeable(spa)); 5864 5865 txg = spa_vdev_enter(spa); 5866 5867 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 5868 VDEV_ALLOC_ADD)) != 0) 5869 return (spa_vdev_exit(spa, NULL, txg, error)); 5870 5871 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 5872 5873 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 5874 &nspares) != 0) 5875 nspares = 0; 5876 5877 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 5878 &nl2cache) != 0) 5879 nl2cache = 0; 5880 5881 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 5882 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5883 5884 if (vd->vdev_children != 0 && 5885 (error = vdev_create(vd, txg, B_FALSE)) != 0) 5886 return (spa_vdev_exit(spa, vd, txg, error)); 5887 5888 /* 5889 * We must validate the spares and l2cache devices after checking the 5890 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 5891 */ 5892 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 5893 return (spa_vdev_exit(spa, vd, txg, error)); 5894 5895 /* 5896 * If we are in the middle of a device removal, we can only add 5897 * devices which match the existing devices in the pool. 5898 * If we are in the middle of a removal, or have some indirect 5899 * vdevs, we can not add raidz toplevels. 5900 */ 5901 if (spa->spa_vdev_removal != NULL || 5902 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5903 for (int c = 0; c < vd->vdev_children; c++) { 5904 tvd = vd->vdev_child[c]; 5905 if (spa->spa_vdev_removal != NULL && 5906 tvd->vdev_ashift != spa->spa_max_ashift) { 5907 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5908 } 5909 /* Fail if top level vdev is raidz */ 5910 if (tvd->vdev_ops == &vdev_raidz_ops) { 5911 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5912 } 5913 /* 5914 * Need the top level mirror to be 5915 * a mirror of leaf vdevs only 5916 */ 5917 if (tvd->vdev_ops == &vdev_mirror_ops) { 5918 for (uint64_t cid = 0; 5919 cid < tvd->vdev_children; cid++) { 5920 vdev_t *cvd = tvd->vdev_child[cid]; 5921 if (!cvd->vdev_ops->vdev_op_leaf) { 5922 return (spa_vdev_exit(spa, vd, 5923 txg, EINVAL)); 5924 } 5925 } 5926 } 5927 } 5928 } 5929 5930 for (int c = 0; c < vd->vdev_children; c++) { 5931 5932 /* 5933 * Set the vdev id to the first hole, if one exists. 5934 */ 5935 for (id = 0; id < rvd->vdev_children; id++) { 5936 if (rvd->vdev_child[id]->vdev_ishole) { 5937 vdev_free(rvd->vdev_child[id]); 5938 break; 5939 } 5940 } 5941 tvd = vd->vdev_child[c]; 5942 vdev_remove_child(vd, tvd); 5943 tvd->vdev_id = id; 5944 vdev_add_child(rvd, tvd); 5945 vdev_config_dirty(tvd); 5946 } 5947 5948 if (nspares != 0) { 5949 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 5950 ZPOOL_CONFIG_SPARES); 5951 spa_load_spares(spa); 5952 spa->spa_spares.sav_sync = B_TRUE; 5953 } 5954 5955 if (nl2cache != 0) { 5956 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 5957 ZPOOL_CONFIG_L2CACHE); 5958 spa_load_l2cache(spa); 5959 spa->spa_l2cache.sav_sync = B_TRUE; 5960 } 5961 5962 /* 5963 * We have to be careful when adding new vdevs to an existing pool. 5964 * If other threads start allocating from these vdevs before we 5965 * sync the config cache, and we lose power, then upon reboot we may 5966 * fail to open the pool because there are DVAs that the config cache 5967 * can't translate. Therefore, we first add the vdevs without 5968 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 5969 * and then let spa_config_update() initialize the new metaslabs. 5970 * 5971 * spa_load() checks for added-but-not-initialized vdevs, so that 5972 * if we lose power at any point in this sequence, the remaining 5973 * steps will be completed the next time we load the pool. 5974 */ 5975 (void) spa_vdev_exit(spa, vd, txg, 0); 5976 5977 mutex_enter(&spa_namespace_lock); 5978 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5979 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); 5980 mutex_exit(&spa_namespace_lock); 5981 5982 return (0); 5983} 5984 5985/* 5986 * Attach a device to a mirror. The arguments are the path to any device 5987 * in the mirror, and the nvroot for the new device. If the path specifies 5988 * a device that is not mirrored, we automatically insert the mirror vdev. 5989 * 5990 * If 'replacing' is specified, the new device is intended to replace the 5991 * existing device; in this case the two devices are made into their own 5992 * mirror using the 'replacing' vdev, which is functionally identical to 5993 * the mirror vdev (it actually reuses all the same ops) but has a few 5994 * extra rules: you can't attach to it after it's been created, and upon 5995 * completion of resilvering, the first disk (the one being replaced) 5996 * is automatically detached. 5997 */ 5998int 5999spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 6000{ 6001 uint64_t txg, dtl_max_txg; 6002 vdev_t *rvd = spa->spa_root_vdev; 6003 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 6004 vdev_ops_t *pvops; 6005 char *oldvdpath, *newvdpath; 6006 int newvd_isspare; 6007 int error; 6008 6009 ASSERT(spa_writeable(spa)); 6010 6011 txg = spa_vdev_enter(spa); 6012 6013 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 6014 6015 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 6016 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 6017 error = (spa_has_checkpoint(spa)) ? 6018 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 6019 return (spa_vdev_exit(spa, NULL, txg, error)); 6020 } 6021 6022 if (spa->spa_vdev_removal != NULL) 6023 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 6024 6025 if (oldvd == NULL) 6026 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 6027 6028 if (!oldvd->vdev_ops->vdev_op_leaf) 6029 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6030 6031 pvd = oldvd->vdev_parent; 6032 6033 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 6034 VDEV_ALLOC_ATTACH)) != 0) 6035 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6036 6037 if (newrootvd->vdev_children != 1) 6038 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 6039 6040 newvd = newrootvd->vdev_child[0]; 6041 6042 if (!newvd->vdev_ops->vdev_op_leaf) 6043 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 6044 6045 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 6046 return (spa_vdev_exit(spa, newrootvd, txg, error)); 6047 6048 /* 6049 * Spares can't replace logs 6050 */ 6051 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 6052 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6053 6054 if (!replacing) { 6055 /* 6056 * For attach, the only allowable parent is a mirror or the root 6057 * vdev. 6058 */ 6059 if (pvd->vdev_ops != &vdev_mirror_ops && 6060 pvd->vdev_ops != &vdev_root_ops) 6061 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6062 6063 pvops = &vdev_mirror_ops; 6064 } else { 6065 /* 6066 * Active hot spares can only be replaced by inactive hot 6067 * spares. 6068 */ 6069 if (pvd->vdev_ops == &vdev_spare_ops && 6070 oldvd->vdev_isspare && 6071 !spa_has_spare(spa, newvd->vdev_guid)) 6072 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6073 6074 /* 6075 * If the source is a hot spare, and the parent isn't already a 6076 * spare, then we want to create a new hot spare. Otherwise, we 6077 * want to create a replacing vdev. The user is not allowed to 6078 * attach to a spared vdev child unless the 'isspare' state is 6079 * the same (spare replaces spare, non-spare replaces 6080 * non-spare). 6081 */ 6082 if (pvd->vdev_ops == &vdev_replacing_ops && 6083 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 6084 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6085 } else if (pvd->vdev_ops == &vdev_spare_ops && 6086 newvd->vdev_isspare != oldvd->vdev_isspare) { 6087 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 6088 } 6089 6090 if (newvd->vdev_isspare) 6091 pvops = &vdev_spare_ops; 6092 else 6093 pvops = &vdev_replacing_ops; 6094 } 6095 6096 /* 6097 * Make sure the new device is big enough. 6098 */ 6099 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 6100 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 6101 6102 /* 6103 * The new device cannot have a higher alignment requirement 6104 * than the top-level vdev. 6105 */ 6106 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 6107 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 6108 6109 /* 6110 * If this is an in-place replacement, update oldvd's path and devid 6111 * to make it distinguishable from newvd, and unopenable from now on. 6112 */ 6113 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 6114 spa_strfree(oldvd->vdev_path); 6115 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 6116 KM_SLEEP); 6117 (void) sprintf(oldvd->vdev_path, "%s/%s", 6118 newvd->vdev_path, "old"); 6119 if (oldvd->vdev_devid != NULL) { 6120 spa_strfree(oldvd->vdev_devid); 6121 oldvd->vdev_devid = NULL; 6122 } 6123 } 6124 6125 /* mark the device being resilvered */ 6126 newvd->vdev_resilver_txg = txg; 6127 6128 /* 6129 * If the parent is not a mirror, or if we're replacing, insert the new 6130 * mirror/replacing/spare vdev above oldvd. 6131 */ 6132 if (pvd->vdev_ops != pvops) 6133 pvd = vdev_add_parent(oldvd, pvops); 6134 6135 ASSERT(pvd->vdev_top->vdev_parent == rvd); 6136 ASSERT(pvd->vdev_ops == pvops); 6137 ASSERT(oldvd->vdev_parent == pvd); 6138 6139 /* 6140 * Extract the new device from its root and add it to pvd. 6141 */ 6142 vdev_remove_child(newrootvd, newvd); 6143 newvd->vdev_id = pvd->vdev_children; 6144 newvd->vdev_crtxg = oldvd->vdev_crtxg; 6145 vdev_add_child(pvd, newvd); 6146 6147 tvd = newvd->vdev_top; 6148 ASSERT(pvd->vdev_top == tvd); 6149 ASSERT(tvd->vdev_parent == rvd); 6150 6151 vdev_config_dirty(tvd); 6152 6153 /* 6154 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 6155 * for any dmu_sync-ed blocks. It will propagate upward when 6156 * spa_vdev_exit() calls vdev_dtl_reassess(). 6157 */ 6158 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 6159 6160 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 6161 dtl_max_txg - TXG_INITIAL); 6162 6163 if (newvd->vdev_isspare) { 6164 spa_spare_activate(newvd); 6165 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); 6166 } 6167 6168 oldvdpath = spa_strdup(oldvd->vdev_path); 6169 newvdpath = spa_strdup(newvd->vdev_path); 6170 newvd_isspare = newvd->vdev_isspare; 6171 6172 /* 6173 * Mark newvd's DTL dirty in this txg. 6174 */ 6175 vdev_dirty(tvd, VDD_DTL, newvd, txg); 6176 6177 /* 6178 * Schedule the resilver to restart in the future. We do this to 6179 * ensure that dmu_sync-ed blocks have been stitched into the 6180 * respective datasets. 6181 */ 6182 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 6183 6184 if (spa->spa_bootfs) 6185 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); 6186 6187 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); 6188 6189 /* 6190 * Commit the config 6191 */ 6192 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 6193 6194 spa_history_log_internal(spa, "vdev attach", NULL, 6195 "%s vdev=%s %s vdev=%s", 6196 replacing && newvd_isspare ? "spare in" : 6197 replacing ? "replace" : "attach", newvdpath, 6198 replacing ? "for" : "to", oldvdpath); 6199 6200 spa_strfree(oldvdpath); 6201 spa_strfree(newvdpath); 6202 6203 return (0); 6204} 6205 6206/* 6207 * Detach a device from a mirror or replacing vdev. 6208 * 6209 * If 'replace_done' is specified, only detach if the parent 6210 * is a replacing vdev. 6211 */ 6212int 6213spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 6214{ 6215 uint64_t txg; 6216 int error; 6217 vdev_t *rvd = spa->spa_root_vdev; 6218 vdev_t *vd, *pvd, *cvd, *tvd; 6219 boolean_t unspare = B_FALSE; 6220 uint64_t unspare_guid = 0; 6221 char *vdpath; 6222 6223 ASSERT(spa_writeable(spa)); 6224 6225 txg = spa_vdev_enter(spa); 6226 6227 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 6228 6229 /* 6230 * Besides being called directly from the userland through the 6231 * ioctl interface, spa_vdev_detach() can be potentially called 6232 * at the end of spa_vdev_resilver_done(). 6233 * 6234 * In the regular case, when we have a checkpoint this shouldn't 6235 * happen as we never empty the DTLs of a vdev during the scrub 6236 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done() 6237 * should never get here when we have a checkpoint. 6238 * 6239 * That said, even in a case when we checkpoint the pool exactly 6240 * as spa_vdev_resilver_done() calls this function everything 6241 * should be fine as the resilver will return right away. 6242 */ 6243 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 6244 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 6245 error = (spa_has_checkpoint(spa)) ? 6246 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 6247 return (spa_vdev_exit(spa, NULL, txg, error)); 6248 } 6249 6250 if (vd == NULL) 6251 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 6252 6253 if (!vd->vdev_ops->vdev_op_leaf) 6254 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6255 6256 pvd = vd->vdev_parent; 6257 6258 /* 6259 * If the parent/child relationship is not as expected, don't do it. 6260 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 6261 * vdev that's replacing B with C. The user's intent in replacing 6262 * is to go from M(A,B) to M(A,C). If the user decides to cancel 6263 * the replace by detaching C, the expected behavior is to end up 6264 * M(A,B). But suppose that right after deciding to detach C, 6265 * the replacement of B completes. We would have M(A,C), and then 6266 * ask to detach C, which would leave us with just A -- not what 6267 * the user wanted. To prevent this, we make sure that the 6268 * parent/child relationship hasn't changed -- in this example, 6269 * that C's parent is still the replacing vdev R. 6270 */ 6271 if (pvd->vdev_guid != pguid && pguid != 0) 6272 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 6273 6274 /* 6275 * Only 'replacing' or 'spare' vdevs can be replaced. 6276 */ 6277 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 6278 pvd->vdev_ops != &vdev_spare_ops) 6279 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6280 6281 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 6282 spa_version(spa) >= SPA_VERSION_SPARES); 6283 6284 /* 6285 * Only mirror, replacing, and spare vdevs support detach. 6286 */ 6287 if (pvd->vdev_ops != &vdev_replacing_ops && 6288 pvd->vdev_ops != &vdev_mirror_ops && 6289 pvd->vdev_ops != &vdev_spare_ops) 6290 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 6291 6292 /* 6293 * If this device has the only valid copy of some data, 6294 * we cannot safely detach it. 6295 */ 6296 if (vdev_dtl_required(vd)) 6297 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 6298 6299 ASSERT(pvd->vdev_children >= 2); 6300 6301 /* 6302 * If we are detaching the second disk from a replacing vdev, then 6303 * check to see if we changed the original vdev's path to have "/old" 6304 * at the end in spa_vdev_attach(). If so, undo that change now. 6305 */ 6306 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 6307 vd->vdev_path != NULL) { 6308 size_t len = strlen(vd->vdev_path); 6309 6310 for (int c = 0; c < pvd->vdev_children; c++) { 6311 cvd = pvd->vdev_child[c]; 6312 6313 if (cvd == vd || cvd->vdev_path == NULL) 6314 continue; 6315 6316 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 6317 strcmp(cvd->vdev_path + len, "/old") == 0) { 6318 spa_strfree(cvd->vdev_path); 6319 cvd->vdev_path = spa_strdup(vd->vdev_path); 6320 break; 6321 } 6322 } 6323 } 6324 6325 /* 6326 * If we are detaching the original disk from a spare, then it implies 6327 * that the spare should become a real disk, and be removed from the 6328 * active spare list for the pool. 6329 */ 6330 if (pvd->vdev_ops == &vdev_spare_ops && 6331 vd->vdev_id == 0 && 6332 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 6333 unspare = B_TRUE; 6334 6335 /* 6336 * Erase the disk labels so the disk can be used for other things. 6337 * This must be done after all other error cases are handled, 6338 * but before we disembowel vd (so we can still do I/O to it). 6339 * But if we can't do it, don't treat the error as fatal -- 6340 * it may be that the unwritability of the disk is the reason 6341 * it's being detached! 6342 */ 6343 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 6344 6345 /* 6346 * Remove vd from its parent and compact the parent's children. 6347 */ 6348 vdev_remove_child(pvd, vd); 6349 vdev_compact_children(pvd); 6350 6351 /* 6352 * Remember one of the remaining children so we can get tvd below. 6353 */ 6354 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 6355 6356 /* 6357 * If we need to remove the remaining child from the list of hot spares, 6358 * do it now, marking the vdev as no longer a spare in the process. 6359 * We must do this before vdev_remove_parent(), because that can 6360 * change the GUID if it creates a new toplevel GUID. For a similar 6361 * reason, we must remove the spare now, in the same txg as the detach; 6362 * otherwise someone could attach a new sibling, change the GUID, and 6363 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 6364 */ 6365 if (unspare) { 6366 ASSERT(cvd->vdev_isspare); 6367 spa_spare_remove(cvd); 6368 unspare_guid = cvd->vdev_guid; 6369 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 6370 cvd->vdev_unspare = B_TRUE; 6371 } 6372 6373 /* 6374 * If the parent mirror/replacing vdev only has one child, 6375 * the parent is no longer needed. Remove it from the tree. 6376 */ 6377 if (pvd->vdev_children == 1) { 6378 if (pvd->vdev_ops == &vdev_spare_ops) 6379 cvd->vdev_unspare = B_FALSE; 6380 vdev_remove_parent(cvd); 6381 } 6382 6383 6384 /* 6385 * We don't set tvd until now because the parent we just removed 6386 * may have been the previous top-level vdev. 6387 */ 6388 tvd = cvd->vdev_top; 6389 ASSERT(tvd->vdev_parent == rvd); 6390 6391 /* 6392 * Reevaluate the parent vdev state. 6393 */ 6394 vdev_propagate_state(cvd); 6395 6396 /* 6397 * If the 'autoexpand' property is set on the pool then automatically 6398 * try to expand the size of the pool. For example if the device we 6399 * just detached was smaller than the others, it may be possible to 6400 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 6401 * first so that we can obtain the updated sizes of the leaf vdevs. 6402 */ 6403 if (spa->spa_autoexpand) { 6404 vdev_reopen(tvd); 6405 vdev_expand(tvd, txg); 6406 } 6407 6408 vdev_config_dirty(tvd); 6409 6410 /* 6411 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 6412 * vd->vdev_detached is set and free vd's DTL object in syncing context. 6413 * But first make sure we're not on any *other* txg's DTL list, to 6414 * prevent vd from being accessed after it's freed. 6415 */ 6416 vdpath = spa_strdup(vd->vdev_path); 6417 for (int t = 0; t < TXG_SIZE; t++) 6418 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 6419 vd->vdev_detached = B_TRUE; 6420 vdev_dirty(tvd, VDD_DTL, vd, txg); 6421 6422 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); 6423 6424 /* hang on to the spa before we release the lock */ 6425 spa_open_ref(spa, FTAG); 6426 6427 error = spa_vdev_exit(spa, vd, txg, 0); 6428 6429 spa_history_log_internal(spa, "detach", NULL, 6430 "vdev=%s", vdpath); 6431 spa_strfree(vdpath); 6432 6433 /* 6434 * If this was the removal of the original device in a hot spare vdev, 6435 * then we want to go through and remove the device from the hot spare 6436 * list of every other pool. 6437 */ 6438 if (unspare) { 6439 spa_t *altspa = NULL; 6440 6441 mutex_enter(&spa_namespace_lock); 6442 while ((altspa = spa_next(altspa)) != NULL) { 6443 if (altspa->spa_state != POOL_STATE_ACTIVE || 6444 altspa == spa) 6445 continue; 6446 6447 spa_open_ref(altspa, FTAG); 6448 mutex_exit(&spa_namespace_lock); 6449 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 6450 mutex_enter(&spa_namespace_lock); 6451 spa_close(altspa, FTAG); 6452 } 6453 mutex_exit(&spa_namespace_lock); 6454 6455 /* search the rest of the vdevs for spares to remove */ 6456 spa_vdev_resilver_done(spa); 6457 } 6458 6459 /* all done with the spa; OK to release */ 6460 mutex_enter(&spa_namespace_lock); 6461 spa_close(spa, FTAG); 6462 mutex_exit(&spa_namespace_lock); 6463 6464 return (error); 6465} 6466 6467int 6468spa_vdev_initialize(spa_t *spa, uint64_t guid, uint64_t cmd_type) 6469{ 6470 /* 6471 * We hold the namespace lock through the whole function 6472 * to prevent any changes to the pool while we're starting or 6473 * stopping initialization. The config and state locks are held so that 6474 * we can properly assess the vdev state before we commit to 6475 * the initializing operation. 6476 */ 6477 mutex_enter(&spa_namespace_lock); 6478 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6479 6480 /* Look up vdev and ensure it's a leaf. */ 6481 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE); 6482 if (vd == NULL || vd->vdev_detached) { 6483 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6484 mutex_exit(&spa_namespace_lock); 6485 return (SET_ERROR(ENODEV)); 6486 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) { 6487 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6488 mutex_exit(&spa_namespace_lock); 6489 return (SET_ERROR(EINVAL)); 6490 } else if (!vdev_writeable(vd)) { 6491 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6492 mutex_exit(&spa_namespace_lock); 6493 return (SET_ERROR(EROFS)); 6494 } 6495 mutex_enter(&vd->vdev_initialize_lock); 6496 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6497 6498 /* 6499 * When we activate an initialize action we check to see 6500 * if the vdev_initialize_thread is NULL. We do this instead 6501 * of using the vdev_initialize_state since there might be 6502 * a previous initialization process which has completed but 6503 * the thread is not exited. 6504 */ 6505 if (cmd_type == POOL_INITIALIZE_DO && 6506 (vd->vdev_initialize_thread != NULL || 6507 vd->vdev_top->vdev_removing)) { 6508 mutex_exit(&vd->vdev_initialize_lock); 6509 mutex_exit(&spa_namespace_lock); 6510 return (SET_ERROR(EBUSY)); 6511 } else if (cmd_type == POOL_INITIALIZE_CANCEL && 6512 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE && 6513 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) { 6514 mutex_exit(&vd->vdev_initialize_lock); 6515 mutex_exit(&spa_namespace_lock); 6516 return (SET_ERROR(ESRCH)); 6517 } else if (cmd_type == POOL_INITIALIZE_SUSPEND && 6518 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) { 6519 mutex_exit(&vd->vdev_initialize_lock); 6520 mutex_exit(&spa_namespace_lock); 6521 return (SET_ERROR(ESRCH)); 6522 } 6523 6524 switch (cmd_type) { 6525 case POOL_INITIALIZE_DO: 6526 vdev_initialize(vd); 6527 break; 6528 case POOL_INITIALIZE_CANCEL: 6529 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED); 6530 break; 6531 case POOL_INITIALIZE_SUSPEND: 6532 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED); 6533 break; 6534 default: 6535 panic("invalid cmd_type %llu", (unsigned long long)cmd_type); 6536 } 6537 mutex_exit(&vd->vdev_initialize_lock); 6538 6539 /* Sync out the initializing state */ 6540 txg_wait_synced(spa->spa_dsl_pool, 0); 6541 mutex_exit(&spa_namespace_lock); 6542 6543 return (0); 6544} 6545 6546 6547/* 6548 * Split a set of devices from their mirrors, and create a new pool from them. 6549 */ 6550int 6551spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 6552 nvlist_t *props, boolean_t exp) 6553{ 6554 int error = 0; 6555 uint64_t txg, *glist; 6556 spa_t *newspa; 6557 uint_t c, children, lastlog; 6558 nvlist_t **child, *nvl, *tmp; 6559 dmu_tx_t *tx; 6560 char *altroot = NULL; 6561 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 6562 boolean_t activate_slog; 6563 6564 ASSERT(spa_writeable(spa)); 6565 6566 txg = spa_vdev_enter(spa); 6567 6568 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 6569 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) { 6570 error = (spa_has_checkpoint(spa)) ? 6571 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT; 6572 return (spa_vdev_exit(spa, NULL, txg, error)); 6573 } 6574 6575 /* clear the log and flush everything up to now */ 6576 activate_slog = spa_passivate_log(spa); 6577 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6578 error = spa_reset_logs(spa); 6579 txg = spa_vdev_config_enter(spa); 6580 6581 if (activate_slog) 6582 spa_activate_log(spa); 6583 6584 if (error != 0) 6585 return (spa_vdev_exit(spa, NULL, txg, error)); 6586 6587 /* check new spa name before going any further */ 6588 if (spa_lookup(newname) != NULL) 6589 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 6590 6591 /* 6592 * scan through all the children to ensure they're all mirrors 6593 */ 6594 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 6595 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 6596 &children) != 0) 6597 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6598 6599 /* first, check to ensure we've got the right child count */ 6600 rvd = spa->spa_root_vdev; 6601 lastlog = 0; 6602 for (c = 0; c < rvd->vdev_children; c++) { 6603 vdev_t *vd = rvd->vdev_child[c]; 6604 6605 /* don't count the holes & logs as children */ 6606 if (vd->vdev_islog || !vdev_is_concrete(vd)) { 6607 if (lastlog == 0) 6608 lastlog = c; 6609 continue; 6610 } 6611 6612 lastlog = 0; 6613 } 6614 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 6615 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6616 6617 /* next, ensure no spare or cache devices are part of the split */ 6618 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 6619 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 6620 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6621 6622 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 6623 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 6624 6625 /* then, loop over each vdev and validate it */ 6626 for (c = 0; c < children; c++) { 6627 uint64_t is_hole = 0; 6628 6629 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 6630 &is_hole); 6631 6632 if (is_hole != 0) { 6633 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 6634 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 6635 continue; 6636 } else { 6637 error = SET_ERROR(EINVAL); 6638 break; 6639 } 6640 } 6641 6642 /* which disk is going to be split? */ 6643 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 6644 &glist[c]) != 0) { 6645 error = SET_ERROR(EINVAL); 6646 break; 6647 } 6648 6649 /* look it up in the spa */ 6650 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 6651 if (vml[c] == NULL) { 6652 error = SET_ERROR(ENODEV); 6653 break; 6654 } 6655 6656 /* make sure there's nothing stopping the split */ 6657 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 6658 vml[c]->vdev_islog || 6659 !vdev_is_concrete(vml[c]) || 6660 vml[c]->vdev_isspare || 6661 vml[c]->vdev_isl2cache || 6662 !vdev_writeable(vml[c]) || 6663 vml[c]->vdev_children != 0 || 6664 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 6665 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 6666 error = SET_ERROR(EINVAL); 6667 break; 6668 } 6669 6670 if (vdev_dtl_required(vml[c])) { 6671 error = SET_ERROR(EBUSY); 6672 break; 6673 } 6674 6675 /* we need certain info from the top level */ 6676 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 6677 vml[c]->vdev_top->vdev_ms_array) == 0); 6678 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 6679 vml[c]->vdev_top->vdev_ms_shift) == 0); 6680 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 6681 vml[c]->vdev_top->vdev_asize) == 0); 6682 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 6683 vml[c]->vdev_top->vdev_ashift) == 0); 6684 6685 /* transfer per-vdev ZAPs */ 6686 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 6687 VERIFY0(nvlist_add_uint64(child[c], 6688 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 6689 6690 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 6691 VERIFY0(nvlist_add_uint64(child[c], 6692 ZPOOL_CONFIG_VDEV_TOP_ZAP, 6693 vml[c]->vdev_parent->vdev_top_zap)); 6694 } 6695 6696 if (error != 0) { 6697 kmem_free(vml, children * sizeof (vdev_t *)); 6698 kmem_free(glist, children * sizeof (uint64_t)); 6699 return (spa_vdev_exit(spa, NULL, txg, error)); 6700 } 6701 6702 /* stop writers from using the disks */ 6703 for (c = 0; c < children; c++) { 6704 if (vml[c] != NULL) 6705 vml[c]->vdev_offline = B_TRUE; 6706 } 6707 vdev_reopen(spa->spa_root_vdev); 6708 6709 /* 6710 * Temporarily record the splitting vdevs in the spa config. This 6711 * will disappear once the config is regenerated. 6712 */ 6713 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6714 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 6715 glist, children) == 0); 6716 kmem_free(glist, children * sizeof (uint64_t)); 6717 6718 mutex_enter(&spa->spa_props_lock); 6719 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 6720 nvl) == 0); 6721 mutex_exit(&spa->spa_props_lock); 6722 spa->spa_config_splitting = nvl; 6723 vdev_config_dirty(spa->spa_root_vdev); 6724 6725 /* configure and create the new pool */ 6726 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 6727 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 6728 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 6729 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6730 spa_version(spa)) == 0); 6731 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 6732 spa->spa_config_txg) == 0); 6733 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 6734 spa_generate_guid(NULL)) == 0); 6735 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 6736 (void) nvlist_lookup_string(props, 6737 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 6738 6739 /* add the new pool to the namespace */ 6740 newspa = spa_add(newname, config, altroot); 6741 newspa->spa_avz_action = AVZ_ACTION_REBUILD; 6742 newspa->spa_config_txg = spa->spa_config_txg; 6743 spa_set_log_state(newspa, SPA_LOG_CLEAR); 6744 6745 /* release the spa config lock, retaining the namespace lock */ 6746 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6747 6748 if (zio_injection_enabled) 6749 zio_handle_panic_injection(spa, FTAG, 1); 6750 6751 spa_activate(newspa, spa_mode_global); 6752 spa_async_suspend(newspa); 6753 6754 for (c = 0; c < children; c++) { 6755 if (vml[c] != NULL) { 6756 /* 6757 * Temporarily stop the initializing activity. We set 6758 * the state to ACTIVE so that we know to resume 6759 * the initializing once the split has completed. 6760 */ 6761 mutex_enter(&vml[c]->vdev_initialize_lock); 6762 vdev_initialize_stop(vml[c], VDEV_INITIALIZE_ACTIVE); 6763 mutex_exit(&vml[c]->vdev_initialize_lock); 6764 } 6765 } 6766 6767#ifndef illumos 6768 /* mark that we are creating new spa by splitting */ 6769 newspa->spa_splitting_newspa = B_TRUE; 6770#endif 6771 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; 6772 6773 /* create the new pool from the disks of the original pool */ 6774 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); 6775#ifndef illumos 6776 newspa->spa_splitting_newspa = B_FALSE; 6777#endif 6778 if (error) 6779 goto out; 6780 6781 /* if that worked, generate a real config for the new pool */ 6782 if (newspa->spa_root_vdev != NULL) { 6783 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 6784 NV_UNIQUE_NAME, KM_SLEEP) == 0); 6785 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 6786 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 6787 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 6788 B_TRUE)); 6789 } 6790 6791 /* set the props */ 6792 if (props != NULL) { 6793 spa_configfile_set(newspa, props, B_FALSE); 6794 error = spa_prop_set(newspa, props); 6795 if (error) 6796 goto out; 6797 } 6798 6799 /* flush everything */ 6800 txg = spa_vdev_config_enter(newspa); 6801 vdev_config_dirty(newspa->spa_root_vdev); 6802 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 6803 6804 if (zio_injection_enabled) 6805 zio_handle_panic_injection(spa, FTAG, 2); 6806 6807 spa_async_resume(newspa); 6808 6809 /* finally, update the original pool's config */ 6810 txg = spa_vdev_config_enter(spa); 6811 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 6812 error = dmu_tx_assign(tx, TXG_WAIT); 6813 if (error != 0) 6814 dmu_tx_abort(tx); 6815 for (c = 0; c < children; c++) { 6816 if (vml[c] != NULL) { 6817 vdev_split(vml[c]); 6818 if (error == 0) 6819 spa_history_log_internal(spa, "detach", tx, 6820 "vdev=%s", vml[c]->vdev_path); 6821 6822 vdev_free(vml[c]); 6823 } 6824 } 6825 spa->spa_avz_action = AVZ_ACTION_REBUILD; 6826 vdev_config_dirty(spa->spa_root_vdev); 6827 spa->spa_config_splitting = NULL; 6828 nvlist_free(nvl); 6829 if (error == 0) 6830 dmu_tx_commit(tx); 6831 (void) spa_vdev_exit(spa, NULL, txg, 0); 6832 6833 if (zio_injection_enabled) 6834 zio_handle_panic_injection(spa, FTAG, 3); 6835 6836 /* split is complete; log a history record */ 6837 spa_history_log_internal(newspa, "split", NULL, 6838 "from pool %s", spa_name(spa)); 6839 6840 kmem_free(vml, children * sizeof (vdev_t *)); 6841 6842 /* if we're not going to mount the filesystems in userland, export */ 6843 if (exp) 6844 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 6845 B_FALSE, B_FALSE); 6846 6847 return (error); 6848 6849out: 6850 spa_unload(newspa); 6851 spa_deactivate(newspa); 6852 spa_remove(newspa); 6853 6854 txg = spa_vdev_config_enter(spa); 6855 6856 /* re-online all offlined disks */ 6857 for (c = 0; c < children; c++) { 6858 if (vml[c] != NULL) 6859 vml[c]->vdev_offline = B_FALSE; 6860 } 6861 6862 /* restart initializing disks as necessary */ 6863 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART); 6864 6865 vdev_reopen(spa->spa_root_vdev); 6866 6867 nvlist_free(spa->spa_config_splitting); 6868 spa->spa_config_splitting = NULL; 6869 (void) spa_vdev_exit(spa, NULL, txg, error); 6870 6871 kmem_free(vml, children * sizeof (vdev_t *)); 6872 return (error); 6873} 6874 6875/* 6876 * Find any device that's done replacing, or a vdev marked 'unspare' that's 6877 * currently spared, so we can detach it. 6878 */ 6879static vdev_t * 6880spa_vdev_resilver_done_hunt(vdev_t *vd) 6881{ 6882 vdev_t *newvd, *oldvd; 6883 6884 for (int c = 0; c < vd->vdev_children; c++) { 6885 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 6886 if (oldvd != NULL) 6887 return (oldvd); 6888 } 6889 6890 /* 6891 * Check for a completed replacement. We always consider the first 6892 * vdev in the list to be the oldest vdev, and the last one to be 6893 * the newest (see spa_vdev_attach() for how that works). In 6894 * the case where the newest vdev is faulted, we will not automatically 6895 * remove it after a resilver completes. This is OK as it will require 6896 * user intervention to determine which disk the admin wishes to keep. 6897 */ 6898 if (vd->vdev_ops == &vdev_replacing_ops) { 6899 ASSERT(vd->vdev_children > 1); 6900 6901 newvd = vd->vdev_child[vd->vdev_children - 1]; 6902 oldvd = vd->vdev_child[0]; 6903 6904 if (vdev_dtl_empty(newvd, DTL_MISSING) && 6905 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6906 !vdev_dtl_required(oldvd)) 6907 return (oldvd); 6908 } 6909 6910 /* 6911 * Check for a completed resilver with the 'unspare' flag set. 6912 * Also potentially update faulted state. 6913 */ 6914 if (vd->vdev_ops == &vdev_spare_ops) { 6915 vdev_t *first = vd->vdev_child[0]; 6916 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 6917 6918 if (last->vdev_unspare) { 6919 oldvd = first; 6920 newvd = last; 6921 } else if (first->vdev_unspare) { 6922 oldvd = last; 6923 newvd = first; 6924 } else { 6925 oldvd = NULL; 6926 } 6927 6928 if (oldvd != NULL && 6929 vdev_dtl_empty(newvd, DTL_MISSING) && 6930 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6931 !vdev_dtl_required(oldvd)) 6932 return (oldvd); 6933 6934 vdev_propagate_state(vd); 6935 6936 /* 6937 * If there are more than two spares attached to a disk, 6938 * and those spares are not required, then we want to 6939 * attempt to free them up now so that they can be used 6940 * by other pools. Once we're back down to a single 6941 * disk+spare, we stop removing them. 6942 */ 6943 if (vd->vdev_children > 2) { 6944 newvd = vd->vdev_child[1]; 6945 6946 if (newvd->vdev_isspare && last->vdev_isspare && 6947 vdev_dtl_empty(last, DTL_MISSING) && 6948 vdev_dtl_empty(last, DTL_OUTAGE) && 6949 !vdev_dtl_required(newvd)) 6950 return (newvd); 6951 } 6952 } 6953 6954 return (NULL); 6955} 6956 6957static void 6958spa_vdev_resilver_done(spa_t *spa) 6959{ 6960 vdev_t *vd, *pvd, *ppvd; 6961 uint64_t guid, sguid, pguid, ppguid; 6962 6963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6964 6965 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 6966 pvd = vd->vdev_parent; 6967 ppvd = pvd->vdev_parent; 6968 guid = vd->vdev_guid; 6969 pguid = pvd->vdev_guid; 6970 ppguid = ppvd->vdev_guid; 6971 sguid = 0; 6972 /* 6973 * If we have just finished replacing a hot spared device, then 6974 * we need to detach the parent's first child (the original hot 6975 * spare) as well. 6976 */ 6977 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 6978 ppvd->vdev_children == 2) { 6979 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 6980 sguid = ppvd->vdev_child[1]->vdev_guid; 6981 } 6982 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 6983 6984 spa_config_exit(spa, SCL_ALL, FTAG); 6985 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 6986 return; 6987 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 6988 return; 6989 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6990 } 6991 6992 spa_config_exit(spa, SCL_ALL, FTAG); 6993} 6994 6995/* 6996 * Update the stored path or FRU for this vdev. 6997 */ 6998int 6999spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 7000 boolean_t ispath) 7001{ 7002 vdev_t *vd; 7003 boolean_t sync = B_FALSE; 7004 7005 ASSERT(spa_writeable(spa)); 7006 7007 spa_vdev_state_enter(spa, SCL_ALL); 7008 7009 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 7010 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 7011 7012 if (!vd->vdev_ops->vdev_op_leaf) 7013 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 7014 7015 if (ispath) { 7016 if (strcmp(value, vd->vdev_path) != 0) { 7017 spa_strfree(vd->vdev_path); 7018 vd->vdev_path = spa_strdup(value); 7019 sync = B_TRUE; 7020 } 7021 } else { 7022 if (vd->vdev_fru == NULL) { 7023 vd->vdev_fru = spa_strdup(value); 7024 sync = B_TRUE; 7025 } else if (strcmp(value, vd->vdev_fru) != 0) { 7026 spa_strfree(vd->vdev_fru); 7027 vd->vdev_fru = spa_strdup(value); 7028 sync = B_TRUE; 7029 } 7030 } 7031 7032 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 7033} 7034 7035int 7036spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 7037{ 7038 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 7039} 7040 7041int 7042spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 7043{ 7044 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 7045} 7046 7047/* 7048 * ========================================================================== 7049 * SPA Scanning 7050 * ========================================================================== 7051 */ 7052int 7053spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) 7054{ 7055 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 7056 7057 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 7058 return (SET_ERROR(EBUSY)); 7059 7060 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); 7061} 7062 7063int 7064spa_scan_stop(spa_t *spa) 7065{ 7066 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 7067 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 7068 return (SET_ERROR(EBUSY)); 7069 return (dsl_scan_cancel(spa->spa_dsl_pool)); 7070} 7071 7072int 7073spa_scan(spa_t *spa, pool_scan_func_t func) 7074{ 7075 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 7076 7077 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 7078 return (SET_ERROR(ENOTSUP)); 7079 7080 /* 7081 * If a resilver was requested, but there is no DTL on a 7082 * writeable leaf device, we have nothing to do. 7083 */ 7084 if (func == POOL_SCAN_RESILVER && 7085 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 7086 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 7087 return (0); 7088 } 7089 7090 return (dsl_scan(spa->spa_dsl_pool, func)); 7091} 7092 7093/* 7094 * ========================================================================== 7095 * SPA async task processing 7096 * ========================================================================== 7097 */ 7098 7099static void 7100spa_async_remove(spa_t *spa, vdev_t *vd) 7101{ 7102 if (vd->vdev_remove_wanted) { 7103 vd->vdev_remove_wanted = B_FALSE; 7104 vd->vdev_delayed_close = B_FALSE; 7105 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 7106 7107 /* 7108 * We want to clear the stats, but we don't want to do a full 7109 * vdev_clear() as that will cause us to throw away 7110 * degraded/faulted state as well as attempt to reopen the 7111 * device, all of which is a waste. 7112 */ 7113 vd->vdev_stat.vs_read_errors = 0; 7114 vd->vdev_stat.vs_write_errors = 0; 7115 vd->vdev_stat.vs_checksum_errors = 0; 7116 7117 vdev_state_dirty(vd->vdev_top); 7118 /* Tell userspace that the vdev is gone. */ 7119 zfs_post_remove(spa, vd); 7120 } 7121 7122 for (int c = 0; c < vd->vdev_children; c++) 7123 spa_async_remove(spa, vd->vdev_child[c]); 7124} 7125 7126static void 7127spa_async_probe(spa_t *spa, vdev_t *vd) 7128{ 7129 if (vd->vdev_probe_wanted) { 7130 vd->vdev_probe_wanted = B_FALSE; 7131 vdev_reopen(vd); /* vdev_open() does the actual probe */ 7132 } 7133 7134 for (int c = 0; c < vd->vdev_children; c++) 7135 spa_async_probe(spa, vd->vdev_child[c]); 7136} 7137 7138static void 7139spa_async_autoexpand(spa_t *spa, vdev_t *vd) 7140{ 7141 sysevent_id_t eid; 7142 nvlist_t *attr; 7143 char *physpath; 7144 7145 if (!spa->spa_autoexpand) 7146 return; 7147 7148 for (int c = 0; c < vd->vdev_children; c++) { 7149 vdev_t *cvd = vd->vdev_child[c]; 7150 spa_async_autoexpand(spa, cvd); 7151 } 7152 7153 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 7154 return; 7155 7156 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 7157 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 7158 7159 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 7160 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 7161 7162 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 7163 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 7164 7165 nvlist_free(attr); 7166 kmem_free(physpath, MAXPATHLEN); 7167} 7168 7169static void 7170spa_async_thread(void *arg) 7171{ 7172 spa_t *spa = (spa_t *)arg; 7173 int tasks; 7174 7175 ASSERT(spa->spa_sync_on); 7176 7177 mutex_enter(&spa->spa_async_lock); 7178 tasks = spa->spa_async_tasks; 7179 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 7180 mutex_exit(&spa->spa_async_lock); 7181 7182 /* 7183 * See if the config needs to be updated. 7184 */ 7185 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 7186 uint64_t old_space, new_space; 7187 7188 mutex_enter(&spa_namespace_lock); 7189 old_space = metaslab_class_get_space(spa_normal_class(spa)); 7190 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 7191 new_space = metaslab_class_get_space(spa_normal_class(spa)); 7192 mutex_exit(&spa_namespace_lock); 7193 7194 /* 7195 * If the pool grew as a result of the config update, 7196 * then log an internal history event. 7197 */ 7198 if (new_space != old_space) { 7199 spa_history_log_internal(spa, "vdev online", NULL, 7200 "pool '%s' size: %llu(+%llu)", 7201 spa_name(spa), new_space, new_space - old_space); 7202 } 7203 } 7204 7205 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 7206 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7207 spa_async_autoexpand(spa, spa->spa_root_vdev); 7208 spa_config_exit(spa, SCL_CONFIG, FTAG); 7209 } 7210 7211 /* 7212 * See if any devices need to be probed. 7213 */ 7214 if (tasks & SPA_ASYNC_PROBE) { 7215 spa_vdev_state_enter(spa, SCL_NONE); 7216 spa_async_probe(spa, spa->spa_root_vdev); 7217 (void) spa_vdev_state_exit(spa, NULL, 0); 7218 } 7219 7220 /* 7221 * If any devices are done replacing, detach them. 7222 */ 7223 if (tasks & SPA_ASYNC_RESILVER_DONE) 7224 spa_vdev_resilver_done(spa); 7225 7226 /* 7227 * Kick off a resilver. 7228 */ 7229 if (tasks & SPA_ASYNC_RESILVER) 7230 dsl_resilver_restart(spa->spa_dsl_pool, 0); 7231 7232 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) { 7233 mutex_enter(&spa_namespace_lock); 7234 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7235 vdev_initialize_restart(spa->spa_root_vdev); 7236 spa_config_exit(spa, SCL_CONFIG, FTAG); 7237 mutex_exit(&spa_namespace_lock); 7238 } 7239 7240 /* 7241 * Let the world know that we're done. 7242 */ 7243 mutex_enter(&spa->spa_async_lock); 7244 spa->spa_async_thread = NULL; 7245 cv_broadcast(&spa->spa_async_cv); 7246 mutex_exit(&spa->spa_async_lock); 7247 thread_exit(); 7248} 7249 7250static void 7251spa_async_thread_vd(void *arg) 7252{ 7253 spa_t *spa = arg; 7254 int tasks; 7255 7256 mutex_enter(&spa->spa_async_lock); 7257 tasks = spa->spa_async_tasks; 7258retry: 7259 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 7260 mutex_exit(&spa->spa_async_lock); 7261 7262 /* 7263 * See if any devices need to be marked REMOVED. 7264 */ 7265 if (tasks & SPA_ASYNC_REMOVE) { 7266 spa_vdev_state_enter(spa, SCL_NONE); 7267 spa_async_remove(spa, spa->spa_root_vdev); 7268 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 7269 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 7270 for (int i = 0; i < spa->spa_spares.sav_count; i++) 7271 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 7272 (void) spa_vdev_state_exit(spa, NULL, 0); 7273 } 7274 7275 /* 7276 * Let the world know that we're done. 7277 */ 7278 mutex_enter(&spa->spa_async_lock); 7279 tasks = spa->spa_async_tasks; 7280 if ((tasks & SPA_ASYNC_REMOVE) != 0) 7281 goto retry; 7282 spa->spa_async_thread_vd = NULL; 7283 cv_broadcast(&spa->spa_async_cv); 7284 mutex_exit(&spa->spa_async_lock); 7285 thread_exit(); 7286} 7287 7288void 7289spa_async_suspend(spa_t *spa) 7290{ 7291 mutex_enter(&spa->spa_async_lock); 7292 spa->spa_async_suspended++; 7293 while (spa->spa_async_thread != NULL || 7294 spa->spa_async_thread_vd != NULL) 7295 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 7296 mutex_exit(&spa->spa_async_lock); 7297 7298 spa_vdev_remove_suspend(spa); 7299 7300 zthr_t *condense_thread = spa->spa_condense_zthr; 7301 if (condense_thread != NULL && zthr_isrunning(condense_thread)) 7302 VERIFY0(zthr_cancel(condense_thread)); 7303 7304 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; 7305 if (discard_thread != NULL && zthr_isrunning(discard_thread)) 7306 VERIFY0(zthr_cancel(discard_thread)); 7307} 7308 7309void 7310spa_async_resume(spa_t *spa) 7311{ 7312 mutex_enter(&spa->spa_async_lock); 7313 ASSERT(spa->spa_async_suspended != 0); 7314 spa->spa_async_suspended--; 7315 mutex_exit(&spa->spa_async_lock); 7316 spa_restart_removal(spa); 7317 7318 zthr_t *condense_thread = spa->spa_condense_zthr; 7319 if (condense_thread != NULL && !zthr_isrunning(condense_thread)) 7320 zthr_resume(condense_thread); 7321 7322 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr; 7323 if (discard_thread != NULL && !zthr_isrunning(discard_thread)) 7324 zthr_resume(discard_thread); 7325} 7326 7327static boolean_t 7328spa_async_tasks_pending(spa_t *spa) 7329{ 7330 uint_t non_config_tasks; 7331 uint_t config_task; 7332 boolean_t config_task_suspended; 7333 7334 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 7335 SPA_ASYNC_REMOVE); 7336 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 7337 if (spa->spa_ccw_fail_time == 0) { 7338 config_task_suspended = B_FALSE; 7339 } else { 7340 config_task_suspended = 7341 (gethrtime() - spa->spa_ccw_fail_time) < 7342 (zfs_ccw_retry_interval * NANOSEC); 7343 } 7344 7345 return (non_config_tasks || (config_task && !config_task_suspended)); 7346} 7347 7348static void 7349spa_async_dispatch(spa_t *spa) 7350{ 7351 mutex_enter(&spa->spa_async_lock); 7352 if (spa_async_tasks_pending(spa) && 7353 !spa->spa_async_suspended && 7354 spa->spa_async_thread == NULL && 7355 rootdir != NULL) 7356 spa->spa_async_thread = thread_create(NULL, 0, 7357 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 7358 mutex_exit(&spa->spa_async_lock); 7359} 7360 7361static void 7362spa_async_dispatch_vd(spa_t *spa) 7363{ 7364 mutex_enter(&spa->spa_async_lock); 7365 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 7366 !spa->spa_async_suspended && 7367 spa->spa_async_thread_vd == NULL && 7368 rootdir != NULL) 7369 spa->spa_async_thread_vd = thread_create(NULL, 0, 7370 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 7371 mutex_exit(&spa->spa_async_lock); 7372} 7373 7374void 7375spa_async_request(spa_t *spa, int task) 7376{ 7377 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 7378 mutex_enter(&spa->spa_async_lock); 7379 spa->spa_async_tasks |= task; 7380 mutex_exit(&spa->spa_async_lock); 7381 spa_async_dispatch_vd(spa); 7382} 7383 7384/* 7385 * ========================================================================== 7386 * SPA syncing routines 7387 * ========================================================================== 7388 */ 7389 7390static int 7391bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 7392{ 7393 bpobj_t *bpo = arg; 7394 bpobj_enqueue(bpo, bp, tx); 7395 return (0); 7396} 7397 7398static int 7399spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 7400{ 7401 zio_t *zio = arg; 7402 7403 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 7404 BP_GET_PSIZE(bp), zio->io_flags)); 7405 return (0); 7406} 7407 7408/* 7409 * Note: this simple function is not inlined to make it easier to dtrace the 7410 * amount of time spent syncing frees. 7411 */ 7412static void 7413spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 7414{ 7415 zio_t *zio = zio_root(spa, NULL, NULL, 0); 7416 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 7417 VERIFY(zio_wait(zio) == 0); 7418} 7419 7420/* 7421 * Note: this simple function is not inlined to make it easier to dtrace the 7422 * amount of time spent syncing deferred frees. 7423 */ 7424static void 7425spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 7426{ 7427 zio_t *zio = zio_root(spa, NULL, NULL, 0); 7428 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 7429 spa_free_sync_cb, zio, tx), ==, 0); 7430 VERIFY0(zio_wait(zio)); 7431} 7432 7433 7434static void 7435spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 7436{ 7437 char *packed = NULL; 7438 size_t bufsize; 7439 size_t nvsize = 0; 7440 dmu_buf_t *db; 7441 7442 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 7443 7444 /* 7445 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 7446 * information. This avoids the dmu_buf_will_dirty() path and 7447 * saves us a pre-read to get data we don't actually care about. 7448 */ 7449 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 7450 packed = kmem_alloc(bufsize, KM_SLEEP); 7451 7452 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 7453 KM_SLEEP) == 0); 7454 bzero(packed + nvsize, bufsize - nvsize); 7455 7456 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 7457 7458 kmem_free(packed, bufsize); 7459 7460 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 7461 dmu_buf_will_dirty(db, tx); 7462 *(uint64_t *)db->db_data = nvsize; 7463 dmu_buf_rele(db, FTAG); 7464} 7465 7466static void 7467spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 7468 const char *config, const char *entry) 7469{ 7470 nvlist_t *nvroot; 7471 nvlist_t **list; 7472 int i; 7473 7474 if (!sav->sav_sync) 7475 return; 7476 7477 /* 7478 * Update the MOS nvlist describing the list of available devices. 7479 * spa_validate_aux() will have already made sure this nvlist is 7480 * valid and the vdevs are labeled appropriately. 7481 */ 7482 if (sav->sav_object == 0) { 7483 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 7484 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 7485 sizeof (uint64_t), tx); 7486 VERIFY(zap_update(spa->spa_meta_objset, 7487 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 7488 &sav->sav_object, tx) == 0); 7489 } 7490 7491 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 7492 if (sav->sav_count == 0) { 7493 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 7494 } else { 7495 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 7496 for (i = 0; i < sav->sav_count; i++) 7497 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 7498 B_FALSE, VDEV_CONFIG_L2CACHE); 7499 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 7500 sav->sav_count) == 0); 7501 for (i = 0; i < sav->sav_count; i++) 7502 nvlist_free(list[i]); 7503 kmem_free(list, sav->sav_count * sizeof (void *)); 7504 } 7505 7506 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 7507 nvlist_free(nvroot); 7508 7509 sav->sav_sync = B_FALSE; 7510} 7511 7512/* 7513 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 7514 * The all-vdev ZAP must be empty. 7515 */ 7516static void 7517spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 7518{ 7519 spa_t *spa = vd->vdev_spa; 7520 if (vd->vdev_top_zap != 0) { 7521 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 7522 vd->vdev_top_zap, tx)); 7523 } 7524 if (vd->vdev_leaf_zap != 0) { 7525 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 7526 vd->vdev_leaf_zap, tx)); 7527 } 7528 for (uint64_t i = 0; i < vd->vdev_children; i++) { 7529 spa_avz_build(vd->vdev_child[i], avz, tx); 7530 } 7531} 7532 7533static void 7534spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 7535{ 7536 nvlist_t *config; 7537 7538 /* 7539 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 7540 * its config may not be dirty but we still need to build per-vdev ZAPs. 7541 * Similarly, if the pool is being assembled (e.g. after a split), we 7542 * need to rebuild the AVZ although the config may not be dirty. 7543 */ 7544 if (list_is_empty(&spa->spa_config_dirty_list) && 7545 spa->spa_avz_action == AVZ_ACTION_NONE) 7546 return; 7547 7548 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7549 7550 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 7551 spa->spa_avz_action == AVZ_ACTION_INITIALIZE || 7552 spa->spa_all_vdev_zaps != 0); 7553 7554 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 7555 /* Make and build the new AVZ */ 7556 uint64_t new_avz = zap_create(spa->spa_meta_objset, 7557 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 7558 spa_avz_build(spa->spa_root_vdev, new_avz, tx); 7559 7560 /* Diff old AVZ with new one */ 7561 zap_cursor_t zc; 7562 zap_attribute_t za; 7563 7564 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7565 spa->spa_all_vdev_zaps); 7566 zap_cursor_retrieve(&zc, &za) == 0; 7567 zap_cursor_advance(&zc)) { 7568 uint64_t vdzap = za.za_first_integer; 7569 if (zap_lookup_int(spa->spa_meta_objset, new_avz, 7570 vdzap) == ENOENT) { 7571 /* 7572 * ZAP is listed in old AVZ but not in new one; 7573 * destroy it 7574 */ 7575 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 7576 tx)); 7577 } 7578 } 7579 7580 zap_cursor_fini(&zc); 7581 7582 /* Destroy the old AVZ */ 7583 VERIFY0(zap_destroy(spa->spa_meta_objset, 7584 spa->spa_all_vdev_zaps, tx)); 7585 7586 /* Replace the old AVZ in the dir obj with the new one */ 7587 VERIFY0(zap_update(spa->spa_meta_objset, 7588 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 7589 sizeof (new_avz), 1, &new_avz, tx)); 7590 7591 spa->spa_all_vdev_zaps = new_avz; 7592 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 7593 zap_cursor_t zc; 7594 zap_attribute_t za; 7595 7596 /* Walk through the AVZ and destroy all listed ZAPs */ 7597 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7598 spa->spa_all_vdev_zaps); 7599 zap_cursor_retrieve(&zc, &za) == 0; 7600 zap_cursor_advance(&zc)) { 7601 uint64_t zap = za.za_first_integer; 7602 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 7603 } 7604 7605 zap_cursor_fini(&zc); 7606 7607 /* Destroy and unlink the AVZ itself */ 7608 VERIFY0(zap_destroy(spa->spa_meta_objset, 7609 spa->spa_all_vdev_zaps, tx)); 7610 VERIFY0(zap_remove(spa->spa_meta_objset, 7611 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 7612 spa->spa_all_vdev_zaps = 0; 7613 } 7614 7615 if (spa->spa_all_vdev_zaps == 0) { 7616 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 7617 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 7618 DMU_POOL_VDEV_ZAP_MAP, tx); 7619 } 7620 spa->spa_avz_action = AVZ_ACTION_NONE; 7621 7622 /* Create ZAPs for vdevs that don't have them. */ 7623 vdev_construct_zaps(spa->spa_root_vdev, tx); 7624 7625 config = spa_config_generate(spa, spa->spa_root_vdev, 7626 dmu_tx_get_txg(tx), B_FALSE); 7627 7628 /* 7629 * If we're upgrading the spa version then make sure that 7630 * the config object gets updated with the correct version. 7631 */ 7632 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 7633 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 7634 spa->spa_uberblock.ub_version); 7635 7636 spa_config_exit(spa, SCL_STATE, FTAG); 7637 7638 nvlist_free(spa->spa_config_syncing); 7639 spa->spa_config_syncing = config; 7640 7641 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 7642} 7643 7644static void 7645spa_sync_version(void *arg, dmu_tx_t *tx) 7646{ 7647 uint64_t *versionp = arg; 7648 uint64_t version = *versionp; 7649 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7650 7651 /* 7652 * Setting the version is special cased when first creating the pool. 7653 */ 7654 ASSERT(tx->tx_txg != TXG_INITIAL); 7655 7656 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 7657 ASSERT(version >= spa_version(spa)); 7658 7659 spa->spa_uberblock.ub_version = version; 7660 vdev_config_dirty(spa->spa_root_vdev); 7661 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 7662} 7663 7664/* 7665 * Set zpool properties. 7666 */ 7667static void 7668spa_sync_props(void *arg, dmu_tx_t *tx) 7669{ 7670 nvlist_t *nvp = arg; 7671 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7672 objset_t *mos = spa->spa_meta_objset; 7673 nvpair_t *elem = NULL; 7674 7675 mutex_enter(&spa->spa_props_lock); 7676 7677 while ((elem = nvlist_next_nvpair(nvp, elem))) { 7678 uint64_t intval; 7679 char *strval, *fname; 7680 zpool_prop_t prop; 7681 const char *propname; 7682 zprop_type_t proptype; 7683 spa_feature_t fid; 7684 7685 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 7686 case ZPOOL_PROP_INVAL: 7687 /* 7688 * We checked this earlier in spa_prop_validate(). 7689 */ 7690 ASSERT(zpool_prop_feature(nvpair_name(elem))); 7691 7692 fname = strchr(nvpair_name(elem), '@') + 1; 7693 VERIFY0(zfeature_lookup_name(fname, &fid)); 7694 7695 spa_feature_enable(spa, fid, tx); 7696 spa_history_log_internal(spa, "set", tx, 7697 "%s=enabled", nvpair_name(elem)); 7698 break; 7699 7700 case ZPOOL_PROP_VERSION: 7701 intval = fnvpair_value_uint64(elem); 7702 /* 7703 * The version is synced seperatly before other 7704 * properties and should be correct by now. 7705 */ 7706 ASSERT3U(spa_version(spa), >=, intval); 7707 break; 7708 7709 case ZPOOL_PROP_ALTROOT: 7710 /* 7711 * 'altroot' is a non-persistent property. It should 7712 * have been set temporarily at creation or import time. 7713 */ 7714 ASSERT(spa->spa_root != NULL); 7715 break; 7716 7717 case ZPOOL_PROP_READONLY: 7718 case ZPOOL_PROP_CACHEFILE: 7719 /* 7720 * 'readonly' and 'cachefile' are also non-persisitent 7721 * properties. 7722 */ 7723 break; 7724 case ZPOOL_PROP_COMMENT: 7725 strval = fnvpair_value_string(elem); 7726 if (spa->spa_comment != NULL) 7727 spa_strfree(spa->spa_comment); 7728 spa->spa_comment = spa_strdup(strval); 7729 /* 7730 * We need to dirty the configuration on all the vdevs 7731 * so that their labels get updated. It's unnecessary 7732 * to do this for pool creation since the vdev's 7733 * configuratoin has already been dirtied. 7734 */ 7735 if (tx->tx_txg != TXG_INITIAL) 7736 vdev_config_dirty(spa->spa_root_vdev); 7737 spa_history_log_internal(spa, "set", tx, 7738 "%s=%s", nvpair_name(elem), strval); 7739 break; 7740 default: 7741 /* 7742 * Set pool property values in the poolprops mos object. 7743 */ 7744 if (spa->spa_pool_props_object == 0) { 7745 spa->spa_pool_props_object = 7746 zap_create_link(mos, DMU_OT_POOL_PROPS, 7747 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 7748 tx); 7749 } 7750 7751 /* normalize the property name */ 7752 propname = zpool_prop_to_name(prop); 7753 proptype = zpool_prop_get_type(prop); 7754 7755 if (nvpair_type(elem) == DATA_TYPE_STRING) { 7756 ASSERT(proptype == PROP_TYPE_STRING); 7757 strval = fnvpair_value_string(elem); 7758 VERIFY0(zap_update(mos, 7759 spa->spa_pool_props_object, propname, 7760 1, strlen(strval) + 1, strval, tx)); 7761 spa_history_log_internal(spa, "set", tx, 7762 "%s=%s", nvpair_name(elem), strval); 7763 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 7764 intval = fnvpair_value_uint64(elem); 7765 7766 if (proptype == PROP_TYPE_INDEX) { 7767 const char *unused; 7768 VERIFY0(zpool_prop_index_to_string( 7769 prop, intval, &unused)); 7770 } 7771 VERIFY0(zap_update(mos, 7772 spa->spa_pool_props_object, propname, 7773 8, 1, &intval, tx)); 7774 spa_history_log_internal(spa, "set", tx, 7775 "%s=%lld", nvpair_name(elem), intval); 7776 } else { 7777 ASSERT(0); /* not allowed */ 7778 } 7779 7780 switch (prop) { 7781 case ZPOOL_PROP_DELEGATION: 7782 spa->spa_delegation = intval; 7783 break; 7784 case ZPOOL_PROP_BOOTFS: 7785 spa->spa_bootfs = intval; 7786 break; 7787 case ZPOOL_PROP_FAILUREMODE: 7788 spa->spa_failmode = intval; 7789 break; 7790 case ZPOOL_PROP_AUTOEXPAND: 7791 spa->spa_autoexpand = intval; 7792 if (tx->tx_txg != TXG_INITIAL) 7793 spa_async_request(spa, 7794 SPA_ASYNC_AUTOEXPAND); 7795 break; 7796 case ZPOOL_PROP_DEDUPDITTO: 7797 spa->spa_dedup_ditto = intval; 7798 break; 7799 default: 7800 break; 7801 } 7802 } 7803 7804 } 7805 7806 mutex_exit(&spa->spa_props_lock); 7807} 7808 7809/* 7810 * Perform one-time upgrade on-disk changes. spa_version() does not 7811 * reflect the new version this txg, so there must be no changes this 7812 * txg to anything that the upgrade code depends on after it executes. 7813 * Therefore this must be called after dsl_pool_sync() does the sync 7814 * tasks. 7815 */ 7816static void 7817spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 7818{ 7819 dsl_pool_t *dp = spa->spa_dsl_pool; 7820 7821 ASSERT(spa->spa_sync_pass == 1); 7822 7823 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 7824 7825 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 7826 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 7827 dsl_pool_create_origin(dp, tx); 7828 7829 /* Keeping the origin open increases spa_minref */ 7830 spa->spa_minref += 3; 7831 } 7832 7833 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 7834 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 7835 dsl_pool_upgrade_clones(dp, tx); 7836 } 7837 7838 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 7839 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 7840 dsl_pool_upgrade_dir_clones(dp, tx); 7841 7842 /* Keeping the freedir open increases spa_minref */ 7843 spa->spa_minref += 3; 7844 } 7845 7846 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 7847 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7848 spa_feature_create_zap_objects(spa, tx); 7849 } 7850 7851 /* 7852 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 7853 * when possibility to use lz4 compression for metadata was added 7854 * Old pools that have this feature enabled must be upgraded to have 7855 * this feature active 7856 */ 7857 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7858 boolean_t lz4_en = spa_feature_is_enabled(spa, 7859 SPA_FEATURE_LZ4_COMPRESS); 7860 boolean_t lz4_ac = spa_feature_is_active(spa, 7861 SPA_FEATURE_LZ4_COMPRESS); 7862 7863 if (lz4_en && !lz4_ac) 7864 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 7865 } 7866 7867 /* 7868 * If we haven't written the salt, do so now. Note that the 7869 * feature may not be activated yet, but that's fine since 7870 * the presence of this ZAP entry is backwards compatible. 7871 */ 7872 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 7873 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 7874 VERIFY0(zap_add(spa->spa_meta_objset, 7875 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 7876 sizeof (spa->spa_cksum_salt.zcs_bytes), 7877 spa->spa_cksum_salt.zcs_bytes, tx)); 7878 } 7879 7880 rrw_exit(&dp->dp_config_rwlock, FTAG); 7881} 7882 7883static void 7884vdev_indirect_state_sync_verify(vdev_t *vd) 7885{ 7886 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 7887 vdev_indirect_births_t *vib = vd->vdev_indirect_births; 7888 7889 if (vd->vdev_ops == &vdev_indirect_ops) { 7890 ASSERT(vim != NULL); 7891 ASSERT(vib != NULL); 7892 } 7893 7894 if (vdev_obsolete_sm_object(vd) != 0) { 7895 ASSERT(vd->vdev_obsolete_sm != NULL); 7896 ASSERT(vd->vdev_removing || 7897 vd->vdev_ops == &vdev_indirect_ops); 7898 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); 7899 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); 7900 7901 ASSERT3U(vdev_obsolete_sm_object(vd), ==, 7902 space_map_object(vd->vdev_obsolete_sm)); 7903 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, 7904 space_map_allocated(vd->vdev_obsolete_sm)); 7905 } 7906 ASSERT(vd->vdev_obsolete_segments != NULL); 7907 7908 /* 7909 * Since frees / remaps to an indirect vdev can only 7910 * happen in syncing context, the obsolete segments 7911 * tree must be empty when we start syncing. 7912 */ 7913 ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); 7914} 7915 7916/* 7917 * Sync the specified transaction group. New blocks may be dirtied as 7918 * part of the process, so we iterate until it converges. 7919 */ 7920void 7921spa_sync(spa_t *spa, uint64_t txg) 7922{ 7923 dsl_pool_t *dp = spa->spa_dsl_pool; 7924 objset_t *mos = spa->spa_meta_objset; 7925 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 7926 vdev_t *rvd = spa->spa_root_vdev; 7927 vdev_t *vd; 7928 dmu_tx_t *tx; 7929 int error; 7930 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 7931 zfs_vdev_queue_depth_pct / 100; 7932 7933 VERIFY(spa_writeable(spa)); 7934 7935 /* 7936 * Wait for i/os issued in open context that need to complete 7937 * before this txg syncs. 7938 */ 7939 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]); 7940 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 7941 ZIO_FLAG_CANFAIL); 7942 7943 /* 7944 * Lock out configuration changes. 7945 */ 7946 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7947 7948 spa->spa_syncing_txg = txg; 7949 spa->spa_sync_pass = 0; 7950 7951 for (int i = 0; i < spa->spa_alloc_count; i++) { 7952 mutex_enter(&spa->spa_alloc_locks[i]); 7953 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i])); 7954 mutex_exit(&spa->spa_alloc_locks[i]); 7955 } 7956 7957 /* 7958 * If there are any pending vdev state changes, convert them 7959 * into config changes that go out with this transaction group. 7960 */ 7961 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7962 while (list_head(&spa->spa_state_dirty_list) != NULL) { 7963 /* 7964 * We need the write lock here because, for aux vdevs, 7965 * calling vdev_config_dirty() modifies sav_config. 7966 * This is ugly and will become unnecessary when we 7967 * eliminate the aux vdev wart by integrating all vdevs 7968 * into the root vdev tree. 7969 */ 7970 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7971 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 7972 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 7973 vdev_state_clean(vd); 7974 vdev_config_dirty(vd); 7975 } 7976 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7977 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 7978 } 7979 spa_config_exit(spa, SCL_STATE, FTAG); 7980 7981 tx = dmu_tx_create_assigned(dp, txg); 7982 7983 spa->spa_sync_starttime = gethrtime(); 7984#ifdef illumos 7985 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 7986 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 7987#else /* !illumos */ 7988#ifdef _KERNEL 7989 callout_schedule(&spa->spa_deadman_cycid, 7990 hz * spa->spa_deadman_synctime / NANOSEC); 7991#endif 7992#endif /* illumos */ 7993 7994 /* 7995 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 7996 * set spa_deflate if we have no raid-z vdevs. 7997 */ 7998 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 7999 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 8000 int i; 8001 8002 for (i = 0; i < rvd->vdev_children; i++) { 8003 vd = rvd->vdev_child[i]; 8004 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 8005 break; 8006 } 8007 if (i == rvd->vdev_children) { 8008 spa->spa_deflate = TRUE; 8009 VERIFY(0 == zap_add(spa->spa_meta_objset, 8010 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 8011 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 8012 } 8013 } 8014 8015 /* 8016 * Set the top-level vdev's max queue depth. Evaluate each 8017 * top-level's async write queue depth in case it changed. 8018 * The max queue depth will not change in the middle of syncing 8019 * out this txg. 8020 */ 8021 uint64_t slots_per_allocator = 0; 8022 for (int c = 0; c < rvd->vdev_children; c++) { 8023 vdev_t *tvd = rvd->vdev_child[c]; 8024 metaslab_group_t *mg = tvd->vdev_mg; 8025 8026 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 8027 !metaslab_group_initialized(mg)) 8028 continue; 8029 8030 /* 8031 * It is safe to do a lock-free check here because only async 8032 * allocations look at mg_max_alloc_queue_depth, and async 8033 * allocations all happen from spa_sync(). 8034 */ 8035 for (int i = 0; i < spa->spa_alloc_count; i++) 8036 ASSERT0(refcount_count(&(mg->mg_alloc_queue_depth[i]))); 8037 mg->mg_max_alloc_queue_depth = max_queue_depth; 8038 8039 for (int i = 0; i < spa->spa_alloc_count; i++) { 8040 mg->mg_cur_max_alloc_queue_depth[i] = 8041 zfs_vdev_def_queue_depth; 8042 } 8043 slots_per_allocator += zfs_vdev_def_queue_depth; 8044 } 8045 metaslab_class_t *mc = spa_normal_class(spa); 8046 for (int i = 0; i < spa->spa_alloc_count; i++) { 8047 ASSERT0(refcount_count(&mc->mc_alloc_slots[i])); 8048 mc->mc_alloc_max_slots[i] = slots_per_allocator; 8049 } 8050 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 8051 8052 for (int c = 0; c < rvd->vdev_children; c++) { 8053 vdev_t *vd = rvd->vdev_child[c]; 8054 vdev_indirect_state_sync_verify(vd); 8055 8056 if (vdev_indirect_should_condense(vd)) { 8057 spa_condense_indirect_start_sync(vd, tx); 8058 break; 8059 } 8060 } 8061 8062 /* 8063 * Iterate to convergence. 8064 */ 8065 do { 8066 int pass = ++spa->spa_sync_pass; 8067 8068 spa_sync_config_object(spa, tx); 8069 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 8070 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 8071 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 8072 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 8073 spa_errlog_sync(spa, txg); 8074 dsl_pool_sync(dp, txg); 8075 8076 if (pass < zfs_sync_pass_deferred_free) { 8077 spa_sync_frees(spa, free_bpl, tx); 8078 } else { 8079 /* 8080 * We can not defer frees in pass 1, because 8081 * we sync the deferred frees later in pass 1. 8082 */ 8083 ASSERT3U(pass, >, 1); 8084 bplist_iterate(free_bpl, bpobj_enqueue_cb, 8085 &spa->spa_deferred_bpobj, tx); 8086 } 8087 8088 ddt_sync(spa, txg); 8089 dsl_scan_sync(dp, tx); 8090 8091 if (spa->spa_vdev_removal != NULL) 8092 svr_sync(spa, tx); 8093 8094 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 8095 != NULL) 8096 vdev_sync(vd, txg); 8097 8098 if (pass == 1) { 8099 spa_sync_upgrades(spa, tx); 8100 ASSERT3U(txg, >=, 8101 spa->spa_uberblock.ub_rootbp.blk_birth); 8102 /* 8103 * Note: We need to check if the MOS is dirty 8104 * because we could have marked the MOS dirty 8105 * without updating the uberblock (e.g. if we 8106 * have sync tasks but no dirty user data). We 8107 * need to check the uberblock's rootbp because 8108 * it is updated if we have synced out dirty 8109 * data (though in this case the MOS will most 8110 * likely also be dirty due to second order 8111 * effects, we don't want to rely on that here). 8112 */ 8113 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 8114 !dmu_objset_is_dirty(mos, txg)) { 8115 /* 8116 * Nothing changed on the first pass, 8117 * therefore this TXG is a no-op. Avoid 8118 * syncing deferred frees, so that we 8119 * can keep this TXG as a no-op. 8120 */ 8121 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 8122 txg)); 8123 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 8124 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 8125 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks, 8126 txg)); 8127 break; 8128 } 8129 spa_sync_deferred_frees(spa, tx); 8130 } 8131 8132 } while (dmu_objset_is_dirty(mos, txg)); 8133 8134 if (!list_is_empty(&spa->spa_config_dirty_list)) { 8135 /* 8136 * Make sure that the number of ZAPs for all the vdevs matches 8137 * the number of ZAPs in the per-vdev ZAP list. This only gets 8138 * called if the config is dirty; otherwise there may be 8139 * outstanding AVZ operations that weren't completed in 8140 * spa_sync_config_object. 8141 */ 8142 uint64_t all_vdev_zap_entry_count; 8143 ASSERT0(zap_count(spa->spa_meta_objset, 8144 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 8145 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 8146 all_vdev_zap_entry_count); 8147 } 8148 8149 if (spa->spa_vdev_removal != NULL) { 8150 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); 8151 } 8152 8153 /* 8154 * Rewrite the vdev configuration (which includes the uberblock) 8155 * to commit the transaction group. 8156 * 8157 * If there are no dirty vdevs, we sync the uberblock to a few 8158 * random top-level vdevs that are known to be visible in the 8159 * config cache (see spa_vdev_add() for a complete description). 8160 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 8161 */ 8162 for (;;) { 8163 /* 8164 * We hold SCL_STATE to prevent vdev open/close/etc. 8165 * while we're attempting to write the vdev labels. 8166 */ 8167 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 8168 8169 if (list_is_empty(&spa->spa_config_dirty_list)) { 8170 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL }; 8171 int svdcount = 0; 8172 int children = rvd->vdev_children; 8173 int c0 = spa_get_random(children); 8174 8175 for (int c = 0; c < children; c++) { 8176 vd = rvd->vdev_child[(c0 + c) % children]; 8177 8178 /* Stop when revisiting the first vdev */ 8179 if (c > 0 && svd[0] == vd) 8180 break; 8181 8182 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 8183 !vdev_is_concrete(vd)) 8184 continue; 8185 8186 svd[svdcount++] = vd; 8187 if (svdcount == SPA_SYNC_MIN_VDEVS) 8188 break; 8189 } 8190 error = vdev_config_sync(svd, svdcount, txg); 8191 } else { 8192 error = vdev_config_sync(rvd->vdev_child, 8193 rvd->vdev_children, txg); 8194 } 8195 8196 if (error == 0) 8197 spa->spa_last_synced_guid = rvd->vdev_guid; 8198 8199 spa_config_exit(spa, SCL_STATE, FTAG); 8200 8201 if (error == 0) 8202 break; 8203 zio_suspend(spa, NULL); 8204 zio_resume_wait(spa); 8205 } 8206 dmu_tx_commit(tx); 8207 8208#ifdef illumos 8209 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 8210#else /* !illumos */ 8211#ifdef _KERNEL 8212 callout_drain(&spa->spa_deadman_cycid); 8213#endif 8214#endif /* illumos */ 8215 8216 /* 8217 * Clear the dirty config list. 8218 */ 8219 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 8220 vdev_config_clean(vd); 8221 8222 /* 8223 * Now that the new config has synced transactionally, 8224 * let it become visible to the config cache. 8225 */ 8226 if (spa->spa_config_syncing != NULL) { 8227 spa_config_set(spa, spa->spa_config_syncing); 8228 spa->spa_config_txg = txg; 8229 spa->spa_config_syncing = NULL; 8230 } 8231 8232 dsl_pool_sync_done(dp, txg); 8233 8234 for (int i = 0; i < spa->spa_alloc_count; i++) { 8235 mutex_enter(&spa->spa_alloc_locks[i]); 8236 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i])); 8237 mutex_exit(&spa->spa_alloc_locks[i]); 8238 } 8239 8240 /* 8241 * Update usable space statistics. 8242 */ 8243 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 8244 != NULL) 8245 vdev_sync_done(vd, txg); 8246 8247 spa_update_dspace(spa); 8248 8249 /* 8250 * It had better be the case that we didn't dirty anything 8251 * since vdev_config_sync(). 8252 */ 8253 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 8254 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 8255 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 8256 8257 while (zfs_pause_spa_sync) 8258 delay(1); 8259 8260 spa->spa_sync_pass = 0; 8261 8262 /* 8263 * Update the last synced uberblock here. We want to do this at 8264 * the end of spa_sync() so that consumers of spa_last_synced_txg() 8265 * will be guaranteed that all the processing associated with 8266 * that txg has been completed. 8267 */ 8268 spa->spa_ubsync = spa->spa_uberblock; 8269 spa_config_exit(spa, SCL_CONFIG, FTAG); 8270 8271 spa_handle_ignored_writes(spa); 8272 8273 /* 8274 * If any async tasks have been requested, kick them off. 8275 */ 8276 spa_async_dispatch(spa); 8277 spa_async_dispatch_vd(spa); 8278} 8279 8280/* 8281 * Sync all pools. We don't want to hold the namespace lock across these 8282 * operations, so we take a reference on the spa_t and drop the lock during the 8283 * sync. 8284 */ 8285void 8286spa_sync_allpools(void) 8287{ 8288 spa_t *spa = NULL; 8289 mutex_enter(&spa_namespace_lock); 8290 while ((spa = spa_next(spa)) != NULL) { 8291 if (spa_state(spa) != POOL_STATE_ACTIVE || 8292 !spa_writeable(spa) || spa_suspended(spa)) 8293 continue; 8294 spa_open_ref(spa, FTAG); 8295 mutex_exit(&spa_namespace_lock); 8296 txg_wait_synced(spa_get_dsl(spa), 0); 8297 mutex_enter(&spa_namespace_lock); 8298 spa_close(spa, FTAG); 8299 } 8300 mutex_exit(&spa_namespace_lock); 8301} 8302 8303/* 8304 * ========================================================================== 8305 * Miscellaneous routines 8306 * ========================================================================== 8307 */ 8308 8309/* 8310 * Remove all pools in the system. 8311 */ 8312void 8313spa_evict_all(void) 8314{ 8315 spa_t *spa; 8316 8317 /* 8318 * Remove all cached state. All pools should be closed now, 8319 * so every spa in the AVL tree should be unreferenced. 8320 */ 8321 mutex_enter(&spa_namespace_lock); 8322 while ((spa = spa_next(NULL)) != NULL) { 8323 /* 8324 * Stop async tasks. The async thread may need to detach 8325 * a device that's been replaced, which requires grabbing 8326 * spa_namespace_lock, so we must drop it here. 8327 */ 8328 spa_open_ref(spa, FTAG); 8329 mutex_exit(&spa_namespace_lock); 8330 spa_async_suspend(spa); 8331 mutex_enter(&spa_namespace_lock); 8332 spa_close(spa, FTAG); 8333 8334 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 8335 spa_unload(spa); 8336 spa_deactivate(spa); 8337 } 8338 spa_remove(spa); 8339 } 8340 mutex_exit(&spa_namespace_lock); 8341} 8342 8343vdev_t * 8344spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 8345{ 8346 vdev_t *vd; 8347 int i; 8348 8349 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 8350 return (vd); 8351 8352 if (aux) { 8353 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 8354 vd = spa->spa_l2cache.sav_vdevs[i]; 8355 if (vd->vdev_guid == guid) 8356 return (vd); 8357 } 8358 8359 for (i = 0; i < spa->spa_spares.sav_count; i++) { 8360 vd = spa->spa_spares.sav_vdevs[i]; 8361 if (vd->vdev_guid == guid) 8362 return (vd); 8363 } 8364 } 8365 8366 return (NULL); 8367} 8368 8369void 8370spa_upgrade(spa_t *spa, uint64_t version) 8371{ 8372 ASSERT(spa_writeable(spa)); 8373 8374 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 8375 8376 /* 8377 * This should only be called for a non-faulted pool, and since a 8378 * future version would result in an unopenable pool, this shouldn't be 8379 * possible. 8380 */ 8381 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 8382 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 8383 8384 spa->spa_uberblock.ub_version = version; 8385 vdev_config_dirty(spa->spa_root_vdev); 8386 8387 spa_config_exit(spa, SCL_ALL, FTAG); 8388 8389 txg_wait_synced(spa_get_dsl(spa), 0); 8390} 8391 8392boolean_t 8393spa_has_spare(spa_t *spa, uint64_t guid) 8394{ 8395 int i; 8396 uint64_t spareguid; 8397 spa_aux_vdev_t *sav = &spa->spa_spares; 8398 8399 for (i = 0; i < sav->sav_count; i++) 8400 if (sav->sav_vdevs[i]->vdev_guid == guid) 8401 return (B_TRUE); 8402 8403 for (i = 0; i < sav->sav_npending; i++) { 8404 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 8405 &spareguid) == 0 && spareguid == guid) 8406 return (B_TRUE); 8407 } 8408 8409 return (B_FALSE); 8410} 8411 8412/* 8413 * Check if a pool has an active shared spare device. 8414 * Note: reference count of an active spare is 2, as a spare and as a replace 8415 */ 8416static boolean_t 8417spa_has_active_shared_spare(spa_t *spa) 8418{ 8419 int i, refcnt; 8420 uint64_t pool; 8421 spa_aux_vdev_t *sav = &spa->spa_spares; 8422 8423 for (i = 0; i < sav->sav_count; i++) { 8424 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 8425 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 8426 refcnt > 2) 8427 return (B_TRUE); 8428 } 8429 8430 return (B_FALSE); 8431} 8432 8433sysevent_t * 8434spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 8435{ 8436 sysevent_t *ev = NULL; 8437#ifdef _KERNEL 8438 sysevent_attr_list_t *attr = NULL; 8439 sysevent_value_t value; 8440 8441 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 8442 SE_SLEEP); 8443 ASSERT(ev != NULL); 8444 8445 value.value_type = SE_DATA_TYPE_STRING; 8446 value.value.sv_string = spa_name(spa); 8447 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 8448 goto done; 8449 8450 value.value_type = SE_DATA_TYPE_UINT64; 8451 value.value.sv_uint64 = spa_guid(spa); 8452 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 8453 goto done; 8454 8455 if (vd) { 8456 value.value_type = SE_DATA_TYPE_UINT64; 8457 value.value.sv_uint64 = vd->vdev_guid; 8458 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 8459 SE_SLEEP) != 0) 8460 goto done; 8461 8462 if (vd->vdev_path) { 8463 value.value_type = SE_DATA_TYPE_STRING; 8464 value.value.sv_string = vd->vdev_path; 8465 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 8466 &value, SE_SLEEP) != 0) 8467 goto done; 8468 } 8469 } 8470 8471 if (hist_nvl != NULL) { 8472 fnvlist_merge((nvlist_t *)attr, hist_nvl); 8473 } 8474 8475 if (sysevent_attach_attributes(ev, attr) != 0) 8476 goto done; 8477 attr = NULL; 8478 8479done: 8480 if (attr) 8481 sysevent_free_attr(attr); 8482 8483#endif 8484 return (ev); 8485} 8486 8487void 8488spa_event_post(sysevent_t *ev) 8489{ 8490#ifdef _KERNEL 8491 sysevent_id_t eid; 8492 8493 (void) log_sysevent(ev, SE_SLEEP, &eid); 8494 sysevent_free(ev); 8495#endif 8496} 8497 8498void 8499spa_event_discard(sysevent_t *ev) 8500{ 8501#ifdef _KERNEL 8502 sysevent_free(ev); 8503#endif 8504} 8505 8506/* 8507 * Post a sysevent corresponding to the given event. The 'name' must be one of 8508 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 8509 * filled in from the spa and (optionally) the vdev and history nvl. This 8510 * doesn't do anything in the userland libzpool, as we don't want consumers to 8511 * misinterpret ztest or zdb as real changes. 8512 */ 8513void 8514spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 8515{ 8516 spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); 8517} 8518