spa.c revision 332536
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2013 Saso Kiselkov. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright 2016 Toomas Soome <tsoome@me.com> 31 * Copyright 2017 Joyent, Inc. 32 * Copyright (c) 2017 Datto Inc. 33 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association. 34 */ 35 36/* 37 * SPA: Storage Pool Allocator 38 * 39 * This file contains all the routines used when modifying on-disk SPA state. 40 * This includes opening, importing, destroying, exporting a pool, and syncing a 41 * pool. 42 */ 43 44#include <sys/zfs_context.h> 45#include <sys/fm/fs/zfs.h> 46#include <sys/spa_impl.h> 47#include <sys/zio.h> 48#include <sys/zio_checksum.h> 49#include <sys/dmu.h> 50#include <sys/dmu_tx.h> 51#include <sys/zap.h> 52#include <sys/zil.h> 53#include <sys/ddt.h> 54#include <sys/vdev_impl.h> 55#include <sys/vdev_removal.h> 56#include <sys/vdev_indirect_mapping.h> 57#include <sys/vdev_indirect_births.h> 58#include <sys/metaslab.h> 59#include <sys/metaslab_impl.h> 60#include <sys/uberblock_impl.h> 61#include <sys/txg.h> 62#include <sys/avl.h> 63#include <sys/bpobj.h> 64#include <sys/dmu_traverse.h> 65#include <sys/dmu_objset.h> 66#include <sys/unique.h> 67#include <sys/dsl_pool.h> 68#include <sys/dsl_dataset.h> 69#include <sys/dsl_dir.h> 70#include <sys/dsl_prop.h> 71#include <sys/dsl_synctask.h> 72#include <sys/fs/zfs.h> 73#include <sys/arc.h> 74#include <sys/callb.h> 75#include <sys/spa_boot.h> 76#include <sys/zfs_ioctl.h> 77#include <sys/dsl_scan.h> 78#include <sys/dmu_send.h> 79#include <sys/dsl_destroy.h> 80#include <sys/dsl_userhold.h> 81#include <sys/zfeature.h> 82#include <sys/zvol.h> 83#include <sys/trim_map.h> 84#include <sys/abd.h> 85 86#ifdef _KERNEL 87#include <sys/callb.h> 88#include <sys/cpupart.h> 89#include <sys/zone.h> 90#endif /* _KERNEL */ 91 92#include "zfs_prop.h" 93#include "zfs_comutil.h" 94 95/* Check hostid on import? */ 96static int check_hostid = 1; 97 98/* 99 * The interval, in seconds, at which failed configuration cache file writes 100 * should be retried. 101 */ 102int zfs_ccw_retry_interval = 300; 103 104SYSCTL_DECL(_vfs_zfs); 105SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 106 "Check hostid on import?"); 107TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 108SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 109 &zfs_ccw_retry_interval, 0, 110 "Configuration cache file write, retry after failure, interval (seconds)"); 111 112typedef enum zti_modes { 113 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 114 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 115 ZTI_MODE_NULL, /* don't create a taskq */ 116 ZTI_NMODES 117} zti_modes_t; 118 119#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 120#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 121#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 122 123#define ZTI_N(n) ZTI_P(n, 1) 124#define ZTI_ONE ZTI_N(1) 125 126typedef struct zio_taskq_info { 127 zti_modes_t zti_mode; 128 uint_t zti_value; 129 uint_t zti_count; 130} zio_taskq_info_t; 131 132static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 133 "issue", "issue_high", "intr", "intr_high" 134}; 135 136/* 137 * This table defines the taskq settings for each ZFS I/O type. When 138 * initializing a pool, we use this table to create an appropriately sized 139 * taskq. Some operations are low volume and therefore have a small, static 140 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 141 * macros. Other operations process a large amount of data; the ZTI_BATCH 142 * macro causes us to create a taskq oriented for throughput. Some operations 143 * are so high frequency and short-lived that the taskq itself can become a a 144 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 145 * additional degree of parallelism specified by the number of threads per- 146 * taskq and the number of taskqs; when dispatching an event in this case, the 147 * particular taskq is chosen at random. 148 * 149 * The different taskq priorities are to handle the different contexts (issue 150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 151 * need to be handled with minimum delay. 152 */ 153const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 156 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 157 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 158 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 159 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 161}; 162 163static void spa_sync_version(void *arg, dmu_tx_t *tx); 164static void spa_sync_props(void *arg, dmu_tx_t *tx); 165static boolean_t spa_has_active_shared_spare(spa_t *spa); 166static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport, 167 boolean_t reloading); 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 * In the case where config was assembled by scanning device paths (/dev/dsks 237 * by default) we are less tolerant since all the existing devices should have 238 * been detected and we want spa_load to return the right error codes. 239 */ 240uint64_t zfs_max_missing_tvds_scan = 0; 241 242 243SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN, 244 &spa_load_print_vdev_tree, 0, 245 "print out vdev tree during pool import"); 246SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN, 247 &zfs_max_missing_tvds, 0, 248 "allow importing pools with missing top-level vdevs"); 249SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN, 250 &zfs_max_missing_tvds_cachefile, 0, 251 "allow importing pools with missing top-level vdevs in cache file"); 252SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN, 253 &zfs_max_missing_tvds_scan, 0, 254 "allow importing pools with missing top-level vdevs during scan"); 255 256/* 257 * ========================================================================== 258 * SPA properties routines 259 * ========================================================================== 260 */ 261 262/* 263 * Add a (source=src, propname=propval) list to an nvlist. 264 */ 265static void 266spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 267 uint64_t intval, zprop_source_t src) 268{ 269 const char *propname = zpool_prop_to_name(prop); 270 nvlist_t *propval; 271 272 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 273 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 274 275 if (strval != NULL) 276 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 277 else 278 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 279 280 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 281 nvlist_free(propval); 282} 283 284/* 285 * Get property values from the spa configuration. 286 */ 287static void 288spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 289{ 290 vdev_t *rvd = spa->spa_root_vdev; 291 dsl_pool_t *pool = spa->spa_dsl_pool; 292 uint64_t size, alloc, cap, version; 293 zprop_source_t src = ZPROP_SRC_NONE; 294 spa_config_dirent_t *dp; 295 metaslab_class_t *mc = spa_normal_class(spa); 296 297 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 298 299 if (rvd != NULL) { 300 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 301 size = metaslab_class_get_space(spa_normal_class(spa)); 302 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 303 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 304 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 305 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 306 size - alloc, src); 307 308 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 309 metaslab_class_fragmentation(mc), src); 310 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 311 metaslab_class_expandable_space(mc), src); 312 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 313 (spa_mode(spa) == FREAD), src); 314 315 cap = (size == 0) ? 0 : (alloc * 100 / size); 316 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 317 318 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 319 ddt_get_pool_dedup_ratio(spa), src); 320 321 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 322 rvd->vdev_state, src); 323 324 version = spa_version(spa); 325 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 326 src = ZPROP_SRC_DEFAULT; 327 else 328 src = ZPROP_SRC_LOCAL; 329 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 330 } 331 332 if (pool != NULL) { 333 /* 334 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 335 * when opening pools before this version freedir will be NULL. 336 */ 337 if (pool->dp_free_dir != NULL) { 338 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 339 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 340 src); 341 } else { 342 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 343 NULL, 0, src); 344 } 345 346 if (pool->dp_leak_dir != NULL) { 347 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 348 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 349 src); 350 } else { 351 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 352 NULL, 0, src); 353 } 354 } 355 356 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 357 358 if (spa->spa_comment != NULL) { 359 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 360 0, ZPROP_SRC_LOCAL); 361 } 362 363 if (spa->spa_root != NULL) 364 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 365 0, ZPROP_SRC_LOCAL); 366 367 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 368 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 369 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 370 } else { 371 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 372 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 373 } 374 375 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 376 if (dp->scd_path == NULL) { 377 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 378 "none", 0, ZPROP_SRC_LOCAL); 379 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 380 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 381 dp->scd_path, 0, ZPROP_SRC_LOCAL); 382 } 383 } 384} 385 386/* 387 * Get zpool property values. 388 */ 389int 390spa_prop_get(spa_t *spa, nvlist_t **nvp) 391{ 392 objset_t *mos = spa->spa_meta_objset; 393 zap_cursor_t zc; 394 zap_attribute_t za; 395 int err; 396 397 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 398 399 mutex_enter(&spa->spa_props_lock); 400 401 /* 402 * Get properties from the spa config. 403 */ 404 spa_prop_get_config(spa, nvp); 405 406 /* If no pool property object, no more prop to get. */ 407 if (mos == NULL || spa->spa_pool_props_object == 0) { 408 mutex_exit(&spa->spa_props_lock); 409 return (0); 410 } 411 412 /* 413 * Get properties from the MOS pool property object. 414 */ 415 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 416 (err = zap_cursor_retrieve(&zc, &za)) == 0; 417 zap_cursor_advance(&zc)) { 418 uint64_t intval = 0; 419 char *strval = NULL; 420 zprop_source_t src = ZPROP_SRC_DEFAULT; 421 zpool_prop_t prop; 422 423 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) 424 continue; 425 426 switch (za.za_integer_length) { 427 case 8: 428 /* integer property */ 429 if (za.za_first_integer != 430 zpool_prop_default_numeric(prop)) 431 src = ZPROP_SRC_LOCAL; 432 433 if (prop == ZPOOL_PROP_BOOTFS) { 434 dsl_pool_t *dp; 435 dsl_dataset_t *ds = NULL; 436 437 dp = spa_get_dsl(spa); 438 dsl_pool_config_enter(dp, FTAG); 439 if (err = dsl_dataset_hold_obj(dp, 440 za.za_first_integer, FTAG, &ds)) { 441 dsl_pool_config_exit(dp, FTAG); 442 break; 443 } 444 445 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 446 KM_SLEEP); 447 dsl_dataset_name(ds, strval); 448 dsl_dataset_rele(ds, FTAG); 449 dsl_pool_config_exit(dp, FTAG); 450 } else { 451 strval = NULL; 452 intval = za.za_first_integer; 453 } 454 455 spa_prop_add_list(*nvp, prop, strval, intval, src); 456 457 if (strval != NULL) 458 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 459 460 break; 461 462 case 1: 463 /* string property */ 464 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 465 err = zap_lookup(mos, spa->spa_pool_props_object, 466 za.za_name, 1, za.za_num_integers, strval); 467 if (err) { 468 kmem_free(strval, za.za_num_integers); 469 break; 470 } 471 spa_prop_add_list(*nvp, prop, strval, 0, src); 472 kmem_free(strval, za.za_num_integers); 473 break; 474 475 default: 476 break; 477 } 478 } 479 zap_cursor_fini(&zc); 480 mutex_exit(&spa->spa_props_lock); 481out: 482 if (err && err != ENOENT) { 483 nvlist_free(*nvp); 484 *nvp = NULL; 485 return (err); 486 } 487 488 return (0); 489} 490 491/* 492 * Validate the given pool properties nvlist and modify the list 493 * for the property values to be set. 494 */ 495static int 496spa_prop_validate(spa_t *spa, nvlist_t *props) 497{ 498 nvpair_t *elem; 499 int error = 0, reset_bootfs = 0; 500 uint64_t objnum = 0; 501 boolean_t has_feature = B_FALSE; 502 503 elem = NULL; 504 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 505 uint64_t intval; 506 char *strval, *slash, *check, *fname; 507 const char *propname = nvpair_name(elem); 508 zpool_prop_t prop = zpool_name_to_prop(propname); 509 510 switch (prop) { 511 case ZPOOL_PROP_INVAL: 512 if (!zpool_prop_feature(propname)) { 513 error = SET_ERROR(EINVAL); 514 break; 515 } 516 517 /* 518 * Sanitize the input. 519 */ 520 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 521 error = SET_ERROR(EINVAL); 522 break; 523 } 524 525 if (nvpair_value_uint64(elem, &intval) != 0) { 526 error = SET_ERROR(EINVAL); 527 break; 528 } 529 530 if (intval != 0) { 531 error = SET_ERROR(EINVAL); 532 break; 533 } 534 535 fname = strchr(propname, '@') + 1; 536 if (zfeature_lookup_name(fname, NULL) != 0) { 537 error = SET_ERROR(EINVAL); 538 break; 539 } 540 541 has_feature = B_TRUE; 542 break; 543 544 case ZPOOL_PROP_VERSION: 545 error = nvpair_value_uint64(elem, &intval); 546 if (!error && 547 (intval < spa_version(spa) || 548 intval > SPA_VERSION_BEFORE_FEATURES || 549 has_feature)) 550 error = SET_ERROR(EINVAL); 551 break; 552 553 case ZPOOL_PROP_DELEGATION: 554 case ZPOOL_PROP_AUTOREPLACE: 555 case ZPOOL_PROP_LISTSNAPS: 556 case ZPOOL_PROP_AUTOEXPAND: 557 error = nvpair_value_uint64(elem, &intval); 558 if (!error && intval > 1) 559 error = SET_ERROR(EINVAL); 560 break; 561 562 case ZPOOL_PROP_BOOTFS: 563 /* 564 * If the pool version is less than SPA_VERSION_BOOTFS, 565 * or the pool is still being created (version == 0), 566 * the bootfs property cannot be set. 567 */ 568 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 569 error = SET_ERROR(ENOTSUP); 570 break; 571 } 572 573 /* 574 * Make sure the vdev config is bootable 575 */ 576 if (!vdev_is_bootable(spa->spa_root_vdev)) { 577 error = SET_ERROR(ENOTSUP); 578 break; 579 } 580 581 reset_bootfs = 1; 582 583 error = nvpair_value_string(elem, &strval); 584 585 if (!error) { 586 objset_t *os; 587 uint64_t propval; 588 589 if (strval == NULL || strval[0] == '\0') { 590 objnum = zpool_prop_default_numeric( 591 ZPOOL_PROP_BOOTFS); 592 break; 593 } 594 595 if (error = dmu_objset_hold(strval, FTAG, &os)) 596 break; 597 598 /* 599 * Must be ZPL, and its property settings 600 * must be supported by GRUB (compression 601 * is not gzip, and large blocks are not used). 602 */ 603 604 if (dmu_objset_type(os) != DMU_OST_ZFS) { 605 error = SET_ERROR(ENOTSUP); 606 } else if ((error = 607 dsl_prop_get_int_ds(dmu_objset_ds(os), 608 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 609 &propval)) == 0 && 610 !BOOTFS_COMPRESS_VALID(propval)) { 611 error = SET_ERROR(ENOTSUP); 612 } else { 613 objnum = dmu_objset_id(os); 614 } 615 dmu_objset_rele(os, FTAG); 616 } 617 break; 618 619 case ZPOOL_PROP_FAILUREMODE: 620 error = nvpair_value_uint64(elem, &intval); 621 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 622 intval > ZIO_FAILURE_MODE_PANIC)) 623 error = SET_ERROR(EINVAL); 624 625 /* 626 * This is a special case which only occurs when 627 * the pool has completely failed. This allows 628 * the user to change the in-core failmode property 629 * without syncing it out to disk (I/Os might 630 * currently be blocked). We do this by returning 631 * EIO to the caller (spa_prop_set) to trick it 632 * into thinking we encountered a property validation 633 * error. 634 */ 635 if (!error && spa_suspended(spa)) { 636 spa->spa_failmode = intval; 637 error = SET_ERROR(EIO); 638 } 639 break; 640 641 case ZPOOL_PROP_CACHEFILE: 642 if ((error = nvpair_value_string(elem, &strval)) != 0) 643 break; 644 645 if (strval[0] == '\0') 646 break; 647 648 if (strcmp(strval, "none") == 0) 649 break; 650 651 if (strval[0] != '/') { 652 error = SET_ERROR(EINVAL); 653 break; 654 } 655 656 slash = strrchr(strval, '/'); 657 ASSERT(slash != NULL); 658 659 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 660 strcmp(slash, "/..") == 0) 661 error = SET_ERROR(EINVAL); 662 break; 663 664 case ZPOOL_PROP_COMMENT: 665 if ((error = nvpair_value_string(elem, &strval)) != 0) 666 break; 667 for (check = strval; *check != '\0'; check++) { 668 /* 669 * The kernel doesn't have an easy isprint() 670 * check. For this kernel check, we merely 671 * check ASCII apart from DEL. Fix this if 672 * there is an easy-to-use kernel isprint(). 673 */ 674 if (*check >= 0x7f) { 675 error = SET_ERROR(EINVAL); 676 break; 677 } 678 } 679 if (strlen(strval) > ZPROP_MAX_COMMENT) 680 error = E2BIG; 681 break; 682 683 case ZPOOL_PROP_DEDUPDITTO: 684 if (spa_version(spa) < SPA_VERSION_DEDUP) 685 error = SET_ERROR(ENOTSUP); 686 else 687 error = nvpair_value_uint64(elem, &intval); 688 if (error == 0 && 689 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 690 error = SET_ERROR(EINVAL); 691 break; 692 } 693 694 if (error) 695 break; 696 } 697 698 if (!error && reset_bootfs) { 699 error = nvlist_remove(props, 700 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 701 702 if (!error) { 703 error = nvlist_add_uint64(props, 704 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 705 } 706 } 707 708 return (error); 709} 710 711void 712spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 713{ 714 char *cachefile; 715 spa_config_dirent_t *dp; 716 717 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 718 &cachefile) != 0) 719 return; 720 721 dp = kmem_alloc(sizeof (spa_config_dirent_t), 722 KM_SLEEP); 723 724 if (cachefile[0] == '\0') 725 dp->scd_path = spa_strdup(spa_config_path); 726 else if (strcmp(cachefile, "none") == 0) 727 dp->scd_path = NULL; 728 else 729 dp->scd_path = spa_strdup(cachefile); 730 731 list_insert_head(&spa->spa_config_list, dp); 732 if (need_sync) 733 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 734} 735 736int 737spa_prop_set(spa_t *spa, nvlist_t *nvp) 738{ 739 int error; 740 nvpair_t *elem = NULL; 741 boolean_t need_sync = B_FALSE; 742 743 if ((error = spa_prop_validate(spa, nvp)) != 0) 744 return (error); 745 746 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 747 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 748 749 if (prop == ZPOOL_PROP_CACHEFILE || 750 prop == ZPOOL_PROP_ALTROOT || 751 prop == ZPOOL_PROP_READONLY) 752 continue; 753 754 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { 755 uint64_t ver; 756 757 if (prop == ZPOOL_PROP_VERSION) { 758 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 759 } else { 760 ASSERT(zpool_prop_feature(nvpair_name(elem))); 761 ver = SPA_VERSION_FEATURES; 762 need_sync = B_TRUE; 763 } 764 765 /* Save time if the version is already set. */ 766 if (ver == spa_version(spa)) 767 continue; 768 769 /* 770 * In addition to the pool directory object, we might 771 * create the pool properties object, the features for 772 * read object, the features for write object, or the 773 * feature descriptions object. 774 */ 775 error = dsl_sync_task(spa->spa_name, NULL, 776 spa_sync_version, &ver, 777 6, ZFS_SPACE_CHECK_RESERVED); 778 if (error) 779 return (error); 780 continue; 781 } 782 783 need_sync = B_TRUE; 784 break; 785 } 786 787 if (need_sync) { 788 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 789 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 790 } 791 792 return (0); 793} 794 795/* 796 * If the bootfs property value is dsobj, clear it. 797 */ 798void 799spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 800{ 801 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 802 VERIFY(zap_remove(spa->spa_meta_objset, 803 spa->spa_pool_props_object, 804 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 805 spa->spa_bootfs = 0; 806 } 807} 808 809/*ARGSUSED*/ 810static int 811spa_change_guid_check(void *arg, dmu_tx_t *tx) 812{ 813 uint64_t *newguid = arg; 814 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 815 vdev_t *rvd = spa->spa_root_vdev; 816 uint64_t vdev_state; 817 818 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 819 vdev_state = rvd->vdev_state; 820 spa_config_exit(spa, SCL_STATE, FTAG); 821 822 if (vdev_state != VDEV_STATE_HEALTHY) 823 return (SET_ERROR(ENXIO)); 824 825 ASSERT3U(spa_guid(spa), !=, *newguid); 826 827 return (0); 828} 829 830static void 831spa_change_guid_sync(void *arg, dmu_tx_t *tx) 832{ 833 uint64_t *newguid = arg; 834 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 835 uint64_t oldguid; 836 vdev_t *rvd = spa->spa_root_vdev; 837 838 oldguid = spa_guid(spa); 839 840 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 841 rvd->vdev_guid = *newguid; 842 rvd->vdev_guid_sum += (*newguid - oldguid); 843 vdev_config_dirty(rvd); 844 spa_config_exit(spa, SCL_STATE, FTAG); 845 846 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 847 oldguid, *newguid); 848} 849 850/* 851 * Change the GUID for the pool. This is done so that we can later 852 * re-import a pool built from a clone of our own vdevs. We will modify 853 * the root vdev's guid, our own pool guid, and then mark all of our 854 * vdevs dirty. Note that we must make sure that all our vdevs are 855 * online when we do this, or else any vdevs that weren't present 856 * would be orphaned from our pool. We are also going to issue a 857 * sysevent to update any watchers. 858 */ 859int 860spa_change_guid(spa_t *spa) 861{ 862 int error; 863 uint64_t guid; 864 865 mutex_enter(&spa->spa_vdev_top_lock); 866 mutex_enter(&spa_namespace_lock); 867 guid = spa_generate_guid(NULL); 868 869 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 870 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 871 872 if (error == 0) { 873 spa_write_cachefile(spa, B_FALSE, B_TRUE); 874 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); 875 } 876 877 mutex_exit(&spa_namespace_lock); 878 mutex_exit(&spa->spa_vdev_top_lock); 879 880 return (error); 881} 882 883/* 884 * ========================================================================== 885 * SPA state manipulation (open/create/destroy/import/export) 886 * ========================================================================== 887 */ 888 889static int 890spa_error_entry_compare(const void *a, const void *b) 891{ 892 spa_error_entry_t *sa = (spa_error_entry_t *)a; 893 spa_error_entry_t *sb = (spa_error_entry_t *)b; 894 int ret; 895 896 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 897 sizeof (zbookmark_phys_t)); 898 899 if (ret < 0) 900 return (-1); 901 else if (ret > 0) 902 return (1); 903 else 904 return (0); 905} 906 907/* 908 * Utility function which retrieves copies of the current logs and 909 * re-initializes them in the process. 910 */ 911void 912spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 913{ 914 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 915 916 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 917 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 918 919 avl_create(&spa->spa_errlist_scrub, 920 spa_error_entry_compare, sizeof (spa_error_entry_t), 921 offsetof(spa_error_entry_t, se_avl)); 922 avl_create(&spa->spa_errlist_last, 923 spa_error_entry_compare, sizeof (spa_error_entry_t), 924 offsetof(spa_error_entry_t, se_avl)); 925} 926 927static void 928spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 929{ 930 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 931 enum zti_modes mode = ztip->zti_mode; 932 uint_t value = ztip->zti_value; 933 uint_t count = ztip->zti_count; 934 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 935 char name[32]; 936 uint_t flags = 0; 937 boolean_t batch = B_FALSE; 938 939 if (mode == ZTI_MODE_NULL) { 940 tqs->stqs_count = 0; 941 tqs->stqs_taskq = NULL; 942 return; 943 } 944 945 ASSERT3U(count, >, 0); 946 947 tqs->stqs_count = count; 948 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 949 950 switch (mode) { 951 case ZTI_MODE_FIXED: 952 ASSERT3U(value, >=, 1); 953 value = MAX(value, 1); 954 break; 955 956 case ZTI_MODE_BATCH: 957 batch = B_TRUE; 958 flags |= TASKQ_THREADS_CPU_PCT; 959 value = zio_taskq_batch_pct; 960 break; 961 962 default: 963 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 964 "spa_activate()", 965 zio_type_name[t], zio_taskq_types[q], mode, value); 966 break; 967 } 968 969 for (uint_t i = 0; i < count; i++) { 970 taskq_t *tq; 971 972 if (count > 1) { 973 (void) snprintf(name, sizeof (name), "%s_%s_%u", 974 zio_type_name[t], zio_taskq_types[q], i); 975 } else { 976 (void) snprintf(name, sizeof (name), "%s_%s", 977 zio_type_name[t], zio_taskq_types[q]); 978 } 979 980#ifdef SYSDC 981 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 982 if (batch) 983 flags |= TASKQ_DC_BATCH; 984 985 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 986 spa->spa_proc, zio_taskq_basedc, flags); 987 } else { 988#endif 989 pri_t pri = maxclsyspri; 990 /* 991 * The write issue taskq can be extremely CPU 992 * intensive. Run it at slightly lower priority 993 * than the other taskqs. 994 * FreeBSD notes: 995 * - numerically higher priorities are lower priorities; 996 * - if priorities divided by four (RQ_PPQ) are equal 997 * then a difference between them is insignificant. 998 */ 999 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 1000#ifdef illumos 1001 pri--; 1002#else 1003 pri += 4; 1004#endif 1005 1006 tq = taskq_create_proc(name, value, pri, 50, 1007 INT_MAX, spa->spa_proc, flags); 1008#ifdef SYSDC 1009 } 1010#endif 1011 1012 tqs->stqs_taskq[i] = tq; 1013 } 1014} 1015 1016static void 1017spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 1018{ 1019 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1020 1021 if (tqs->stqs_taskq == NULL) { 1022 ASSERT0(tqs->stqs_count); 1023 return; 1024 } 1025 1026 for (uint_t i = 0; i < tqs->stqs_count; i++) { 1027 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 1028 taskq_destroy(tqs->stqs_taskq[i]); 1029 } 1030 1031 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 1032 tqs->stqs_taskq = NULL; 1033} 1034 1035/* 1036 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 1037 * Note that a type may have multiple discrete taskqs to avoid lock contention 1038 * on the taskq itself. In that case we choose which taskq at random by using 1039 * the low bits of gethrtime(). 1040 */ 1041void 1042spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 1043 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 1044{ 1045 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 1046 taskq_t *tq; 1047 1048 ASSERT3P(tqs->stqs_taskq, !=, NULL); 1049 ASSERT3U(tqs->stqs_count, !=, 0); 1050 1051 if (tqs->stqs_count == 1) { 1052 tq = tqs->stqs_taskq[0]; 1053 } else { 1054#ifdef _KERNEL 1055 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 1056#else 1057 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 1058#endif 1059 } 1060 1061 taskq_dispatch_ent(tq, func, arg, flags, ent); 1062} 1063 1064static void 1065spa_create_zio_taskqs(spa_t *spa) 1066{ 1067 for (int t = 0; t < ZIO_TYPES; t++) { 1068 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1069 spa_taskqs_init(spa, t, q); 1070 } 1071 } 1072} 1073 1074#ifdef _KERNEL 1075#ifdef SPA_PROCESS 1076static void 1077spa_thread(void *arg) 1078{ 1079 callb_cpr_t cprinfo; 1080 1081 spa_t *spa = arg; 1082 user_t *pu = PTOU(curproc); 1083 1084 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1085 spa->spa_name); 1086 1087 ASSERT(curproc != &p0); 1088 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1089 "zpool-%s", spa->spa_name); 1090 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1091 1092#ifdef PSRSET_BIND 1093 /* bind this thread to the requested psrset */ 1094 if (zio_taskq_psrset_bind != PS_NONE) { 1095 pool_lock(); 1096 mutex_enter(&cpu_lock); 1097 mutex_enter(&pidlock); 1098 mutex_enter(&curproc->p_lock); 1099 1100 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1101 0, NULL, NULL) == 0) { 1102 curthread->t_bind_pset = zio_taskq_psrset_bind; 1103 } else { 1104 cmn_err(CE_WARN, 1105 "Couldn't bind process for zfs pool \"%s\" to " 1106 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1107 } 1108 1109 mutex_exit(&curproc->p_lock); 1110 mutex_exit(&pidlock); 1111 mutex_exit(&cpu_lock); 1112 pool_unlock(); 1113 } 1114#endif 1115 1116#ifdef SYSDC 1117 if (zio_taskq_sysdc) { 1118 sysdc_thread_enter(curthread, 100, 0); 1119 } 1120#endif 1121 1122 spa->spa_proc = curproc; 1123 spa->spa_did = curthread->t_did; 1124 1125 spa_create_zio_taskqs(spa); 1126 1127 mutex_enter(&spa->spa_proc_lock); 1128 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1129 1130 spa->spa_proc_state = SPA_PROC_ACTIVE; 1131 cv_broadcast(&spa->spa_proc_cv); 1132 1133 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1134 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1135 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1136 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1137 1138 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1139 spa->spa_proc_state = SPA_PROC_GONE; 1140 spa->spa_proc = &p0; 1141 cv_broadcast(&spa->spa_proc_cv); 1142 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1143 1144 mutex_enter(&curproc->p_lock); 1145 lwp_exit(); 1146} 1147#endif /* SPA_PROCESS */ 1148#endif 1149 1150/* 1151 * Activate an uninitialized pool. 1152 */ 1153static void 1154spa_activate(spa_t *spa, int mode) 1155{ 1156 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1157 1158 spa->spa_state = POOL_STATE_ACTIVE; 1159 spa->spa_mode = mode; 1160 1161 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1162 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1163 1164 /* Try to create a covering process */ 1165 mutex_enter(&spa->spa_proc_lock); 1166 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1167 ASSERT(spa->spa_proc == &p0); 1168 spa->spa_did = 0; 1169 1170#ifdef SPA_PROCESS 1171 /* Only create a process if we're going to be around a while. */ 1172 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1173 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1174 NULL, 0) == 0) { 1175 spa->spa_proc_state = SPA_PROC_CREATED; 1176 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1177 cv_wait(&spa->spa_proc_cv, 1178 &spa->spa_proc_lock); 1179 } 1180 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1181 ASSERT(spa->spa_proc != &p0); 1182 ASSERT(spa->spa_did != 0); 1183 } else { 1184#ifdef _KERNEL 1185 cmn_err(CE_WARN, 1186 "Couldn't create process for zfs pool \"%s\"\n", 1187 spa->spa_name); 1188#endif 1189 } 1190 } 1191#endif /* SPA_PROCESS */ 1192 mutex_exit(&spa->spa_proc_lock); 1193 1194 /* If we didn't create a process, we need to create our taskqs. */ 1195 ASSERT(spa->spa_proc == &p0); 1196 if (spa->spa_proc == &p0) { 1197 spa_create_zio_taskqs(spa); 1198 } 1199 1200 /* 1201 * Start TRIM thread. 1202 */ 1203 trim_thread_create(spa); 1204 1205 for (size_t i = 0; i < TXG_SIZE; i++) 1206 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0); 1207 1208 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1209 offsetof(vdev_t, vdev_config_dirty_node)); 1210 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1211 offsetof(objset_t, os_evicting_node)); 1212 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1213 offsetof(vdev_t, vdev_state_dirty_node)); 1214 1215 txg_list_create(&spa->spa_vdev_txg_list, spa, 1216 offsetof(struct vdev, vdev_txg_node)); 1217 1218 avl_create(&spa->spa_errlist_scrub, 1219 spa_error_entry_compare, sizeof (spa_error_entry_t), 1220 offsetof(spa_error_entry_t, se_avl)); 1221 avl_create(&spa->spa_errlist_last, 1222 spa_error_entry_compare, sizeof (spa_error_entry_t), 1223 offsetof(spa_error_entry_t, se_avl)); 1224} 1225 1226/* 1227 * Opposite of spa_activate(). 1228 */ 1229static void 1230spa_deactivate(spa_t *spa) 1231{ 1232 ASSERT(spa->spa_sync_on == B_FALSE); 1233 ASSERT(spa->spa_dsl_pool == NULL); 1234 ASSERT(spa->spa_root_vdev == NULL); 1235 ASSERT(spa->spa_async_zio_root == NULL); 1236 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1237 1238 /* 1239 * Stop TRIM thread in case spa_unload() wasn't called directly 1240 * before spa_deactivate(). 1241 */ 1242 trim_thread_destroy(spa); 1243 1244 spa_evicting_os_wait(spa); 1245 1246 txg_list_destroy(&spa->spa_vdev_txg_list); 1247 1248 list_destroy(&spa->spa_config_dirty_list); 1249 list_destroy(&spa->spa_evicting_os_list); 1250 list_destroy(&spa->spa_state_dirty_list); 1251 1252 for (int t = 0; t < ZIO_TYPES; t++) { 1253 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1254 spa_taskqs_fini(spa, t, q); 1255 } 1256 } 1257 1258 for (size_t i = 0; i < TXG_SIZE; i++) { 1259 ASSERT3P(spa->spa_txg_zio[i], !=, NULL); 1260 VERIFY0(zio_wait(spa->spa_txg_zio[i])); 1261 spa->spa_txg_zio[i] = NULL; 1262 } 1263 1264 metaslab_class_destroy(spa->spa_normal_class); 1265 spa->spa_normal_class = NULL; 1266 1267 metaslab_class_destroy(spa->spa_log_class); 1268 spa->spa_log_class = NULL; 1269 1270 /* 1271 * If this was part of an import or the open otherwise failed, we may 1272 * still have errors left in the queues. Empty them just in case. 1273 */ 1274 spa_errlog_drain(spa); 1275 1276 avl_destroy(&spa->spa_errlist_scrub); 1277 avl_destroy(&spa->spa_errlist_last); 1278 1279 spa->spa_state = POOL_STATE_UNINITIALIZED; 1280 1281 mutex_enter(&spa->spa_proc_lock); 1282 if (spa->spa_proc_state != SPA_PROC_NONE) { 1283 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1284 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1285 cv_broadcast(&spa->spa_proc_cv); 1286 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1287 ASSERT(spa->spa_proc != &p0); 1288 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1289 } 1290 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1291 spa->spa_proc_state = SPA_PROC_NONE; 1292 } 1293 ASSERT(spa->spa_proc == &p0); 1294 mutex_exit(&spa->spa_proc_lock); 1295 1296#ifdef SPA_PROCESS 1297 /* 1298 * We want to make sure spa_thread() has actually exited the ZFS 1299 * module, so that the module can't be unloaded out from underneath 1300 * it. 1301 */ 1302 if (spa->spa_did != 0) { 1303 thread_join(spa->spa_did); 1304 spa->spa_did = 0; 1305 } 1306#endif /* SPA_PROCESS */ 1307} 1308 1309/* 1310 * Verify a pool configuration, and construct the vdev tree appropriately. This 1311 * will create all the necessary vdevs in the appropriate layout, with each vdev 1312 * in the CLOSED state. This will prep the pool before open/creation/import. 1313 * All vdev validation is done by the vdev_alloc() routine. 1314 */ 1315static int 1316spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1317 uint_t id, int atype) 1318{ 1319 nvlist_t **child; 1320 uint_t children; 1321 int error; 1322 1323 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1324 return (error); 1325 1326 if ((*vdp)->vdev_ops->vdev_op_leaf) 1327 return (0); 1328 1329 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1330 &child, &children); 1331 1332 if (error == ENOENT) 1333 return (0); 1334 1335 if (error) { 1336 vdev_free(*vdp); 1337 *vdp = NULL; 1338 return (SET_ERROR(EINVAL)); 1339 } 1340 1341 for (int c = 0; c < children; c++) { 1342 vdev_t *vd; 1343 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1344 atype)) != 0) { 1345 vdev_free(*vdp); 1346 *vdp = NULL; 1347 return (error); 1348 } 1349 } 1350 1351 ASSERT(*vdp != NULL); 1352 1353 return (0); 1354} 1355 1356/* 1357 * Opposite of spa_load(). 1358 */ 1359static void 1360spa_unload(spa_t *spa) 1361{ 1362 int i; 1363 1364 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1365 1366 spa_load_note(spa, "UNLOADING"); 1367 1368 /* 1369 * Stop TRIM thread. 1370 */ 1371 trim_thread_destroy(spa); 1372 1373 /* 1374 * Stop async tasks. 1375 */ 1376 spa_async_suspend(spa); 1377 1378 /* 1379 * Stop syncing. 1380 */ 1381 if (spa->spa_sync_on) { 1382 txg_sync_stop(spa->spa_dsl_pool); 1383 spa->spa_sync_on = B_FALSE; 1384 } 1385 1386 /* 1387 * Even though vdev_free() also calls vdev_metaslab_fini, we need 1388 * to call it earlier, before we wait for async i/o to complete. 1389 * This ensures that there is no async metaslab prefetching, by 1390 * calling taskq_wait(mg_taskq). 1391 */ 1392 if (spa->spa_root_vdev != NULL) { 1393 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1394 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) 1395 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); 1396 spa_config_exit(spa, SCL_ALL, FTAG); 1397 } 1398 1399 /* 1400 * Wait for any outstanding async I/O to complete. 1401 */ 1402 if (spa->spa_async_zio_root != NULL) { 1403 for (int i = 0; i < max_ncpus; i++) 1404 (void) zio_wait(spa->spa_async_zio_root[i]); 1405 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1406 spa->spa_async_zio_root = NULL; 1407 } 1408 1409 if (spa->spa_vdev_removal != NULL) { 1410 spa_vdev_removal_destroy(spa->spa_vdev_removal); 1411 spa->spa_vdev_removal = NULL; 1412 } 1413 1414 spa_condense_fini(spa); 1415 1416 bpobj_close(&spa->spa_deferred_bpobj); 1417 1418 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1419 1420 /* 1421 * Close all vdevs. 1422 */ 1423 if (spa->spa_root_vdev) 1424 vdev_free(spa->spa_root_vdev); 1425 ASSERT(spa->spa_root_vdev == NULL); 1426 1427 /* 1428 * Close the dsl pool. 1429 */ 1430 if (spa->spa_dsl_pool) { 1431 dsl_pool_close(spa->spa_dsl_pool); 1432 spa->spa_dsl_pool = NULL; 1433 spa->spa_meta_objset = NULL; 1434 } 1435 1436 ddt_unload(spa); 1437 1438 /* 1439 * Drop and purge level 2 cache 1440 */ 1441 spa_l2cache_drop(spa); 1442 1443 for (i = 0; i < spa->spa_spares.sav_count; i++) 1444 vdev_free(spa->spa_spares.sav_vdevs[i]); 1445 if (spa->spa_spares.sav_vdevs) { 1446 kmem_free(spa->spa_spares.sav_vdevs, 1447 spa->spa_spares.sav_count * sizeof (void *)); 1448 spa->spa_spares.sav_vdevs = NULL; 1449 } 1450 if (spa->spa_spares.sav_config) { 1451 nvlist_free(spa->spa_spares.sav_config); 1452 spa->spa_spares.sav_config = NULL; 1453 } 1454 spa->spa_spares.sav_count = 0; 1455 1456 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1457 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1458 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1459 } 1460 if (spa->spa_l2cache.sav_vdevs) { 1461 kmem_free(spa->spa_l2cache.sav_vdevs, 1462 spa->spa_l2cache.sav_count * sizeof (void *)); 1463 spa->spa_l2cache.sav_vdevs = NULL; 1464 } 1465 if (spa->spa_l2cache.sav_config) { 1466 nvlist_free(spa->spa_l2cache.sav_config); 1467 spa->spa_l2cache.sav_config = NULL; 1468 } 1469 spa->spa_l2cache.sav_count = 0; 1470 1471 spa->spa_async_suspended = 0; 1472 1473 spa->spa_indirect_vdevs_loaded = B_FALSE; 1474 1475 if (spa->spa_comment != NULL) { 1476 spa_strfree(spa->spa_comment); 1477 spa->spa_comment = NULL; 1478 } 1479 1480 spa_config_exit(spa, SCL_ALL, FTAG); 1481} 1482 1483/* 1484 * Load (or re-load) the current list of vdevs describing the active spares for 1485 * this pool. When this is called, we have some form of basic information in 1486 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1487 * then re-generate a more complete list including status information. 1488 */ 1489void 1490spa_load_spares(spa_t *spa) 1491{ 1492 nvlist_t **spares; 1493 uint_t nspares; 1494 int i; 1495 vdev_t *vd, *tvd; 1496 1497 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1498 1499 /* 1500 * First, close and free any existing spare vdevs. 1501 */ 1502 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1503 vd = spa->spa_spares.sav_vdevs[i]; 1504 1505 /* Undo the call to spa_activate() below */ 1506 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1507 B_FALSE)) != NULL && tvd->vdev_isspare) 1508 spa_spare_remove(tvd); 1509 vdev_close(vd); 1510 vdev_free(vd); 1511 } 1512 1513 if (spa->spa_spares.sav_vdevs) 1514 kmem_free(spa->spa_spares.sav_vdevs, 1515 spa->spa_spares.sav_count * sizeof (void *)); 1516 1517 if (spa->spa_spares.sav_config == NULL) 1518 nspares = 0; 1519 else 1520 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1521 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1522 1523 spa->spa_spares.sav_count = (int)nspares; 1524 spa->spa_spares.sav_vdevs = NULL; 1525 1526 if (nspares == 0) 1527 return; 1528 1529 /* 1530 * Construct the array of vdevs, opening them to get status in the 1531 * process. For each spare, there is potentially two different vdev_t 1532 * structures associated with it: one in the list of spares (used only 1533 * for basic validation purposes) and one in the active vdev 1534 * configuration (if it's spared in). During this phase we open and 1535 * validate each vdev on the spare list. If the vdev also exists in the 1536 * active configuration, then we also mark this vdev as an active spare. 1537 */ 1538 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1539 KM_SLEEP); 1540 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1541 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1542 VDEV_ALLOC_SPARE) == 0); 1543 ASSERT(vd != NULL); 1544 1545 spa->spa_spares.sav_vdevs[i] = vd; 1546 1547 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1548 B_FALSE)) != NULL) { 1549 if (!tvd->vdev_isspare) 1550 spa_spare_add(tvd); 1551 1552 /* 1553 * We only mark the spare active if we were successfully 1554 * able to load the vdev. Otherwise, importing a pool 1555 * with a bad active spare would result in strange 1556 * behavior, because multiple pool would think the spare 1557 * is actively in use. 1558 * 1559 * There is a vulnerability here to an equally bizarre 1560 * circumstance, where a dead active spare is later 1561 * brought back to life (onlined or otherwise). Given 1562 * the rarity of this scenario, and the extra complexity 1563 * it adds, we ignore the possibility. 1564 */ 1565 if (!vdev_is_dead(tvd)) 1566 spa_spare_activate(tvd); 1567 } 1568 1569 vd->vdev_top = vd; 1570 vd->vdev_aux = &spa->spa_spares; 1571 1572 if (vdev_open(vd) != 0) 1573 continue; 1574 1575 if (vdev_validate_aux(vd) == 0) 1576 spa_spare_add(vd); 1577 } 1578 1579 /* 1580 * Recompute the stashed list of spares, with status information 1581 * this time. 1582 */ 1583 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1584 DATA_TYPE_NVLIST_ARRAY) == 0); 1585 1586 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1587 KM_SLEEP); 1588 for (i = 0; i < spa->spa_spares.sav_count; i++) 1589 spares[i] = vdev_config_generate(spa, 1590 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1591 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1592 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1593 for (i = 0; i < spa->spa_spares.sav_count; i++) 1594 nvlist_free(spares[i]); 1595 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1596} 1597 1598/* 1599 * Load (or re-load) the current list of vdevs describing the active l2cache for 1600 * this pool. When this is called, we have some form of basic information in 1601 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1602 * then re-generate a more complete list including status information. 1603 * Devices which are already active have their details maintained, and are 1604 * not re-opened. 1605 */ 1606void 1607spa_load_l2cache(spa_t *spa) 1608{ 1609 nvlist_t **l2cache; 1610 uint_t nl2cache; 1611 int i, j, oldnvdevs; 1612 uint64_t guid; 1613 vdev_t *vd, **oldvdevs, **newvdevs; 1614 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1615 1616 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1617 1618 if (sav->sav_config != NULL) { 1619 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1620 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1621 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1622 } else { 1623 nl2cache = 0; 1624 newvdevs = NULL; 1625 } 1626 1627 oldvdevs = sav->sav_vdevs; 1628 oldnvdevs = sav->sav_count; 1629 sav->sav_vdevs = NULL; 1630 sav->sav_count = 0; 1631 1632 /* 1633 * Process new nvlist of vdevs. 1634 */ 1635 for (i = 0; i < nl2cache; i++) { 1636 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1637 &guid) == 0); 1638 1639 newvdevs[i] = NULL; 1640 for (j = 0; j < oldnvdevs; j++) { 1641 vd = oldvdevs[j]; 1642 if (vd != NULL && guid == vd->vdev_guid) { 1643 /* 1644 * Retain previous vdev for add/remove ops. 1645 */ 1646 newvdevs[i] = vd; 1647 oldvdevs[j] = NULL; 1648 break; 1649 } 1650 } 1651 1652 if (newvdevs[i] == NULL) { 1653 /* 1654 * Create new vdev 1655 */ 1656 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1657 VDEV_ALLOC_L2CACHE) == 0); 1658 ASSERT(vd != NULL); 1659 newvdevs[i] = vd; 1660 1661 /* 1662 * Commit this vdev as an l2cache device, 1663 * even if it fails to open. 1664 */ 1665 spa_l2cache_add(vd); 1666 1667 vd->vdev_top = vd; 1668 vd->vdev_aux = sav; 1669 1670 spa_l2cache_activate(vd); 1671 1672 if (vdev_open(vd) != 0) 1673 continue; 1674 1675 (void) vdev_validate_aux(vd); 1676 1677 if (!vdev_is_dead(vd)) 1678 l2arc_add_vdev(spa, vd); 1679 } 1680 } 1681 1682 /* 1683 * Purge vdevs that were dropped 1684 */ 1685 for (i = 0; i < oldnvdevs; i++) { 1686 uint64_t pool; 1687 1688 vd = oldvdevs[i]; 1689 if (vd != NULL) { 1690 ASSERT(vd->vdev_isl2cache); 1691 1692 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1693 pool != 0ULL && l2arc_vdev_present(vd)) 1694 l2arc_remove_vdev(vd); 1695 vdev_clear_stats(vd); 1696 vdev_free(vd); 1697 } 1698 } 1699 1700 if (oldvdevs) 1701 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1702 1703 if (sav->sav_config == NULL) 1704 goto out; 1705 1706 sav->sav_vdevs = newvdevs; 1707 sav->sav_count = (int)nl2cache; 1708 1709 /* 1710 * Recompute the stashed list of l2cache devices, with status 1711 * information this time. 1712 */ 1713 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1714 DATA_TYPE_NVLIST_ARRAY) == 0); 1715 1716 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1717 for (i = 0; i < sav->sav_count; i++) 1718 l2cache[i] = vdev_config_generate(spa, 1719 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1720 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1721 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1722out: 1723 for (i = 0; i < sav->sav_count; i++) 1724 nvlist_free(l2cache[i]); 1725 if (sav->sav_count) 1726 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1727} 1728 1729static int 1730load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1731{ 1732 dmu_buf_t *db; 1733 char *packed = NULL; 1734 size_t nvsize = 0; 1735 int error; 1736 *value = NULL; 1737 1738 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1739 if (error != 0) 1740 return (error); 1741 1742 nvsize = *(uint64_t *)db->db_data; 1743 dmu_buf_rele(db, FTAG); 1744 1745 packed = kmem_alloc(nvsize, KM_SLEEP); 1746 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1747 DMU_READ_PREFETCH); 1748 if (error == 0) 1749 error = nvlist_unpack(packed, nvsize, value, 0); 1750 kmem_free(packed, nvsize); 1751 1752 return (error); 1753} 1754 1755/* 1756 * Concrete top-level vdevs that are not missing and are not logs. At every 1757 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds. 1758 */ 1759static uint64_t 1760spa_healthy_core_tvds(spa_t *spa) 1761{ 1762 vdev_t *rvd = spa->spa_root_vdev; 1763 uint64_t tvds = 0; 1764 1765 for (uint64_t i = 0; i < rvd->vdev_children; i++) { 1766 vdev_t *vd = rvd->vdev_child[i]; 1767 if (vd->vdev_islog) 1768 continue; 1769 if (vdev_is_concrete(vd) && !vdev_is_dead(vd)) 1770 tvds++; 1771 } 1772 1773 return (tvds); 1774} 1775 1776/* 1777 * Checks to see if the given vdev could not be opened, in which case we post a 1778 * sysevent to notify the autoreplace code that the device has been removed. 1779 */ 1780static void 1781spa_check_removed(vdev_t *vd) 1782{ 1783 for (uint64_t c = 0; c < vd->vdev_children; c++) 1784 spa_check_removed(vd->vdev_child[c]); 1785 1786 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1787 vdev_is_concrete(vd)) { 1788 zfs_post_autoreplace(vd->vdev_spa, vd); 1789 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); 1790 } 1791} 1792 1793static int 1794spa_check_for_missing_logs(spa_t *spa) 1795{ 1796 vdev_t *rvd = spa->spa_root_vdev; 1797 1798 /* 1799 * If we're doing a normal import, then build up any additional 1800 * diagnostic information about missing log devices. 1801 * We'll pass this up to the user for further processing. 1802 */ 1803 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1804 nvlist_t **child, *nv; 1805 uint64_t idx = 0; 1806 1807 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1808 KM_SLEEP); 1809 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1810 1811 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1812 vdev_t *tvd = rvd->vdev_child[c]; 1813 1814 /* 1815 * We consider a device as missing only if it failed 1816 * to open (i.e. offline or faulted is not considered 1817 * as missing). 1818 */ 1819 if (tvd->vdev_islog && 1820 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1821 child[idx++] = vdev_config_generate(spa, tvd, 1822 B_FALSE, VDEV_CONFIG_MISSING); 1823 } 1824 } 1825 1826 if (idx > 0) { 1827 fnvlist_add_nvlist_array(nv, 1828 ZPOOL_CONFIG_CHILDREN, child, idx); 1829 fnvlist_add_nvlist(spa->spa_load_info, 1830 ZPOOL_CONFIG_MISSING_DEVICES, nv); 1831 1832 for (uint64_t i = 0; i < idx; i++) 1833 nvlist_free(child[i]); 1834 } 1835 nvlist_free(nv); 1836 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1837 1838 if (idx > 0) { 1839 spa_load_failed(spa, "some log devices are missing"); 1840 return (SET_ERROR(ENXIO)); 1841 } 1842 } else { 1843 for (uint64_t c = 0; c < rvd->vdev_children; c++) { 1844 vdev_t *tvd = rvd->vdev_child[c]; 1845 1846 if (tvd->vdev_islog && 1847 tvd->vdev_state == VDEV_STATE_CANT_OPEN) { 1848 spa_set_log_state(spa, SPA_LOG_CLEAR); 1849 spa_load_note(spa, "some log devices are " 1850 "missing, ZIL is dropped."); 1851 break; 1852 } 1853 } 1854 } 1855 1856 return (0); 1857} 1858 1859/* 1860 * Check for missing log devices 1861 */ 1862static boolean_t 1863spa_check_logs(spa_t *spa) 1864{ 1865 boolean_t rv = B_FALSE; 1866 dsl_pool_t *dp = spa_get_dsl(spa); 1867 1868 switch (spa->spa_log_state) { 1869 case SPA_LOG_MISSING: 1870 /* need to recheck in case slog has been restored */ 1871 case SPA_LOG_UNKNOWN: 1872 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1873 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1874 if (rv) 1875 spa_set_log_state(spa, SPA_LOG_MISSING); 1876 break; 1877 } 1878 return (rv); 1879} 1880 1881static boolean_t 1882spa_passivate_log(spa_t *spa) 1883{ 1884 vdev_t *rvd = spa->spa_root_vdev; 1885 boolean_t slog_found = B_FALSE; 1886 1887 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1888 1889 if (!spa_has_slogs(spa)) 1890 return (B_FALSE); 1891 1892 for (int c = 0; c < rvd->vdev_children; c++) { 1893 vdev_t *tvd = rvd->vdev_child[c]; 1894 metaslab_group_t *mg = tvd->vdev_mg; 1895 1896 if (tvd->vdev_islog) { 1897 metaslab_group_passivate(mg); 1898 slog_found = B_TRUE; 1899 } 1900 } 1901 1902 return (slog_found); 1903} 1904 1905static void 1906spa_activate_log(spa_t *spa) 1907{ 1908 vdev_t *rvd = spa->spa_root_vdev; 1909 1910 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1911 1912 for (int c = 0; c < rvd->vdev_children; c++) { 1913 vdev_t *tvd = rvd->vdev_child[c]; 1914 metaslab_group_t *mg = tvd->vdev_mg; 1915 1916 if (tvd->vdev_islog) 1917 metaslab_group_activate(mg); 1918 } 1919} 1920 1921int 1922spa_reset_logs(spa_t *spa) 1923{ 1924 int error; 1925 1926 error = dmu_objset_find(spa_name(spa), zil_reset, 1927 NULL, DS_FIND_CHILDREN); 1928 if (error == 0) { 1929 /* 1930 * We successfully offlined the log device, sync out the 1931 * current txg so that the "stubby" block can be removed 1932 * by zil_sync(). 1933 */ 1934 txg_wait_synced(spa->spa_dsl_pool, 0); 1935 } 1936 return (error); 1937} 1938 1939static void 1940spa_aux_check_removed(spa_aux_vdev_t *sav) 1941{ 1942 int i; 1943 1944 for (i = 0; i < sav->sav_count; i++) 1945 spa_check_removed(sav->sav_vdevs[i]); 1946} 1947 1948void 1949spa_claim_notify(zio_t *zio) 1950{ 1951 spa_t *spa = zio->io_spa; 1952 1953 if (zio->io_error) 1954 return; 1955 1956 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1957 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1958 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1959 mutex_exit(&spa->spa_props_lock); 1960} 1961 1962typedef struct spa_load_error { 1963 uint64_t sle_meta_count; 1964 uint64_t sle_data_count; 1965} spa_load_error_t; 1966 1967static void 1968spa_load_verify_done(zio_t *zio) 1969{ 1970 blkptr_t *bp = zio->io_bp; 1971 spa_load_error_t *sle = zio->io_private; 1972 dmu_object_type_t type = BP_GET_TYPE(bp); 1973 int error = zio->io_error; 1974 spa_t *spa = zio->io_spa; 1975 1976 abd_free(zio->io_abd); 1977 if (error) { 1978 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1979 type != DMU_OT_INTENT_LOG) 1980 atomic_inc_64(&sle->sle_meta_count); 1981 else 1982 atomic_inc_64(&sle->sle_data_count); 1983 } 1984 1985 mutex_enter(&spa->spa_scrub_lock); 1986 spa->spa_scrub_inflight--; 1987 cv_broadcast(&spa->spa_scrub_io_cv); 1988 mutex_exit(&spa->spa_scrub_lock); 1989} 1990 1991/* 1992 * Maximum number of concurrent scrub i/os to create while verifying 1993 * a pool while importing it. 1994 */ 1995int spa_load_verify_maxinflight = 10000; 1996boolean_t spa_load_verify_metadata = B_TRUE; 1997boolean_t spa_load_verify_data = B_TRUE; 1998 1999SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 2000 &spa_load_verify_maxinflight, 0, 2001 "Maximum number of concurrent scrub I/Os to create while verifying a " 2002 "pool while importing it"); 2003 2004SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 2005 &spa_load_verify_metadata, 0, 2006 "Check metadata on import?"); 2007 2008SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 2009 &spa_load_verify_data, 0, 2010 "Check user data on import?"); 2011 2012/*ARGSUSED*/ 2013static int 2014spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 2015 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 2016{ 2017 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 2018 return (0); 2019 /* 2020 * Note: normally this routine will not be called if 2021 * spa_load_verify_metadata is not set. However, it may be useful 2022 * to manually set the flag after the traversal has begun. 2023 */ 2024 if (!spa_load_verify_metadata) 2025 return (0); 2026 if (!BP_IS_METADATA(bp) && !spa_load_verify_data) 2027 return (0); 2028 2029 zio_t *rio = arg; 2030 size_t size = BP_GET_PSIZE(bp); 2031 2032 mutex_enter(&spa->spa_scrub_lock); 2033 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 2034 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 2035 spa->spa_scrub_inflight++; 2036 mutex_exit(&spa->spa_scrub_lock); 2037 2038 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, 2039 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 2040 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 2041 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 2042 return (0); 2043} 2044 2045/* ARGSUSED */ 2046int 2047verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 2048{ 2049 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 2050 return (SET_ERROR(ENAMETOOLONG)); 2051 2052 return (0); 2053} 2054 2055static int 2056spa_load_verify(spa_t *spa) 2057{ 2058 zio_t *rio; 2059 spa_load_error_t sle = { 0 }; 2060 zpool_rewind_policy_t policy; 2061 boolean_t verify_ok = B_FALSE; 2062 int error = 0; 2063 2064 zpool_get_rewind_policy(spa->spa_config, &policy); 2065 2066 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 2067 return (0); 2068 2069 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2070 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2071 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2072 DS_FIND_CHILDREN); 2073 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2074 if (error != 0) 2075 return (error); 2076 2077 rio = zio_root(spa, NULL, &sle, 2078 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2079 2080 if (spa_load_verify_metadata) { 2081 if (spa->spa_extreme_rewind) { 2082 spa_load_note(spa, "performing a complete scan of the " 2083 "pool since extreme rewind is on. This may take " 2084 "a very long time.\n (spa_load_verify_data=%u, " 2085 "spa_load_verify_metadata=%u)", 2086 spa_load_verify_data, spa_load_verify_metadata); 2087 } 2088 error = traverse_pool(spa, spa->spa_verify_min_txg, 2089 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2090 spa_load_verify_cb, rio); 2091 } 2092 2093 (void) zio_wait(rio); 2094 2095 spa->spa_load_meta_errors = sle.sle_meta_count; 2096 spa->spa_load_data_errors = sle.sle_data_count; 2097 2098 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) { 2099 spa_load_note(spa, "spa_load_verify found %llu metadata errors " 2100 "and %llu data errors", (u_longlong_t)sle.sle_meta_count, 2101 (u_longlong_t)sle.sle_data_count); 2102 } 2103 2104 if (spa_load_verify_dryrun || 2105 (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2106 sle.sle_data_count <= policy.zrp_maxdata)) { 2107 int64_t loss = 0; 2108 2109 verify_ok = B_TRUE; 2110 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2111 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2112 2113 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2114 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2115 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2116 VERIFY(nvlist_add_int64(spa->spa_load_info, 2117 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2118 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2119 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2120 } else { 2121 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2122 } 2123 2124 if (spa_load_verify_dryrun) 2125 return (0); 2126 2127 if (error) { 2128 if (error != ENXIO && error != EIO) 2129 error = SET_ERROR(EIO); 2130 return (error); 2131 } 2132 2133 return (verify_ok ? 0 : EIO); 2134} 2135 2136/* 2137 * Find a value in the pool props object. 2138 */ 2139static void 2140spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2141{ 2142 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2143 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2144} 2145 2146/* 2147 * Find a value in the pool directory object. 2148 */ 2149static int 2150spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent) 2151{ 2152 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2153 name, sizeof (uint64_t), 1, val); 2154 2155 if (error != 0 && (error != ENOENT || log_enoent)) { 2156 spa_load_failed(spa, "couldn't get '%s' value in MOS directory " 2157 "[error=%d]", name, error); 2158 } 2159 2160 return (error); 2161} 2162 2163static int 2164spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2165{ 2166 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2167 return (SET_ERROR(err)); 2168} 2169 2170/* 2171 * Fix up config after a partly-completed split. This is done with the 2172 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2173 * pool have that entry in their config, but only the splitting one contains 2174 * a list of all the guids of the vdevs that are being split off. 2175 * 2176 * This function determines what to do with that list: either rejoin 2177 * all the disks to the pool, or complete the splitting process. To attempt 2178 * the rejoin, each disk that is offlined is marked online again, and 2179 * we do a reopen() call. If the vdev label for every disk that was 2180 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2181 * then we call vdev_split() on each disk, and complete the split. 2182 * 2183 * Otherwise we leave the config alone, with all the vdevs in place in 2184 * the original pool. 2185 */ 2186static void 2187spa_try_repair(spa_t *spa, nvlist_t *config) 2188{ 2189 uint_t extracted; 2190 uint64_t *glist; 2191 uint_t i, gcount; 2192 nvlist_t *nvl; 2193 vdev_t **vd; 2194 boolean_t attempt_reopen; 2195 2196 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2197 return; 2198 2199 /* check that the config is complete */ 2200 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2201 &glist, &gcount) != 0) 2202 return; 2203 2204 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2205 2206 /* attempt to online all the vdevs & validate */ 2207 attempt_reopen = B_TRUE; 2208 for (i = 0; i < gcount; i++) { 2209 if (glist[i] == 0) /* vdev is hole */ 2210 continue; 2211 2212 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2213 if (vd[i] == NULL) { 2214 /* 2215 * Don't bother attempting to reopen the disks; 2216 * just do the split. 2217 */ 2218 attempt_reopen = B_FALSE; 2219 } else { 2220 /* attempt to re-online it */ 2221 vd[i]->vdev_offline = B_FALSE; 2222 } 2223 } 2224 2225 if (attempt_reopen) { 2226 vdev_reopen(spa->spa_root_vdev); 2227 2228 /* check each device to see what state it's in */ 2229 for (extracted = 0, i = 0; i < gcount; i++) { 2230 if (vd[i] != NULL && 2231 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2232 break; 2233 ++extracted; 2234 } 2235 } 2236 2237 /* 2238 * If every disk has been moved to the new pool, or if we never 2239 * even attempted to look at them, then we split them off for 2240 * good. 2241 */ 2242 if (!attempt_reopen || gcount == extracted) { 2243 for (i = 0; i < gcount; i++) 2244 if (vd[i] != NULL) 2245 vdev_split(vd[i]); 2246 vdev_reopen(spa->spa_root_vdev); 2247 } 2248 2249 kmem_free(vd, gcount * sizeof (vdev_t *)); 2250} 2251 2252static int 2253spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type) 2254{ 2255 char *ereport = FM_EREPORT_ZFS_POOL; 2256 int error; 2257 2258 spa->spa_load_state = state; 2259 2260 gethrestime(&spa->spa_loaded_ts); 2261 error = spa_load_impl(spa, type, &ereport, B_FALSE); 2262 2263 /* 2264 * Don't count references from objsets that are already closed 2265 * and are making their way through the eviction process. 2266 */ 2267 spa_evicting_os_wait(spa); 2268 spa->spa_minref = refcount_count(&spa->spa_refcount); 2269 if (error) { 2270 if (error != EEXIST) { 2271 spa->spa_loaded_ts.tv_sec = 0; 2272 spa->spa_loaded_ts.tv_nsec = 0; 2273 } 2274 if (error != EBADF) { 2275 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2276 } 2277 } 2278 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2279 spa->spa_ena = 0; 2280 2281 return (error); 2282} 2283 2284/* 2285 * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2286 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2287 * spa's per-vdev ZAP list. 2288 */ 2289static uint64_t 2290vdev_count_verify_zaps(vdev_t *vd) 2291{ 2292 spa_t *spa = vd->vdev_spa; 2293 uint64_t total = 0; 2294 if (vd->vdev_top_zap != 0) { 2295 total++; 2296 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2297 spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2298 } 2299 if (vd->vdev_leaf_zap != 0) { 2300 total++; 2301 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2302 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2303 } 2304 2305 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2306 total += vdev_count_verify_zaps(vd->vdev_child[i]); 2307 } 2308 2309 return (total); 2310} 2311 2312static int 2313spa_verify_host(spa_t *spa, nvlist_t *mos_config) 2314{ 2315 uint64_t hostid; 2316 char *hostname; 2317 uint64_t myhostid = 0; 2318 2319 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config, 2320 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2321 hostname = fnvlist_lookup_string(mos_config, 2322 ZPOOL_CONFIG_HOSTNAME); 2323 2324 myhostid = zone_get_hostid(NULL); 2325 2326 if (hostid != 0 && myhostid != 0 && hostid != myhostid) { 2327 cmn_err(CE_WARN, "pool '%s' could not be " 2328 "loaded as it was last accessed by " 2329 "another system (host: %s hostid: 0x%llx). " 2330 "See: http://illumos.org/msg/ZFS-8000-EY", 2331 spa_name(spa), hostname, (u_longlong_t)hostid); 2332 spa_load_failed(spa, "hostid verification failed: pool " 2333 "last accessed by host: %s (hostid: 0x%llx)", 2334 hostname, (u_longlong_t)hostid); 2335 return (SET_ERROR(EBADF)); 2336 } 2337 } 2338 2339 return (0); 2340} 2341 2342static int 2343spa_ld_parse_config(spa_t *spa, spa_import_type_t type) 2344{ 2345 int error = 0; 2346 nvlist_t *nvtree, *nvl, *config = spa->spa_config; 2347 int parse; 2348 vdev_t *rvd; 2349 uint64_t pool_guid; 2350 char *comment; 2351 2352 /* 2353 * Versioning wasn't explicitly added to the label until later, so if 2354 * it's not present treat it as the initial version. 2355 */ 2356 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2357 &spa->spa_ubsync.ub_version) != 0) 2358 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2359 2360 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) { 2361 spa_load_failed(spa, "invalid config provided: '%s' missing", 2362 ZPOOL_CONFIG_POOL_GUID); 2363 return (SET_ERROR(EINVAL)); 2364 } 2365 2366 if ((spa->spa_load_state == SPA_LOAD_IMPORT || spa->spa_load_state == 2367 SPA_LOAD_TRYIMPORT) && spa_guid_exists(pool_guid, 0)) { 2368 spa_load_failed(spa, "a pool with guid %llu is already open", 2369 (u_longlong_t)pool_guid); 2370 return (SET_ERROR(EEXIST)); 2371 } 2372 2373 spa->spa_config_guid = pool_guid; 2374 2375 nvlist_free(spa->spa_load_info); 2376 spa->spa_load_info = fnvlist_alloc(); 2377 2378 ASSERT(spa->spa_comment == NULL); 2379 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2380 spa->spa_comment = spa_strdup(comment); 2381 2382 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2383 &spa->spa_config_txg); 2384 2385 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0) 2386 spa->spa_config_splitting = fnvlist_dup(nvl); 2387 2388 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) { 2389 spa_load_failed(spa, "invalid config provided: '%s' missing", 2390 ZPOOL_CONFIG_VDEV_TREE); 2391 return (SET_ERROR(EINVAL)); 2392 } 2393 2394 /* 2395 * Create "The Godfather" zio to hold all async IOs 2396 */ 2397 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2398 KM_SLEEP); 2399 for (int i = 0; i < max_ncpus; i++) { 2400 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2401 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2402 ZIO_FLAG_GODFATHER); 2403 } 2404 2405 /* 2406 * Parse the configuration into a vdev tree. We explicitly set the 2407 * value that will be returned by spa_version() since parsing the 2408 * configuration requires knowing the version number. 2409 */ 2410 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2411 parse = (type == SPA_IMPORT_EXISTING ? 2412 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2413 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse); 2414 spa_config_exit(spa, SCL_ALL, FTAG); 2415 2416 if (error != 0) { 2417 spa_load_failed(spa, "unable to parse config [error=%d]", 2418 error); 2419 return (error); 2420 } 2421 2422 ASSERT(spa->spa_root_vdev == rvd); 2423 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2424 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2425 2426 if (type != SPA_IMPORT_ASSEMBLE) { 2427 ASSERT(spa_guid(spa) == pool_guid); 2428 } 2429 2430 return (0); 2431} 2432 2433/* 2434 * Recursively open all vdevs in the vdev tree. This function is called twice: 2435 * first with the untrusted config, then with the trusted config. 2436 */ 2437static int 2438spa_ld_open_vdevs(spa_t *spa) 2439{ 2440 int error = 0; 2441 2442 /* 2443 * spa_missing_tvds_allowed defines how many top-level vdevs can be 2444 * missing/unopenable for the root vdev to be still considered openable. 2445 */ 2446 if (spa->spa_trust_config) { 2447 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds; 2448 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) { 2449 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile; 2450 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) { 2451 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan; 2452 } else { 2453 spa->spa_missing_tvds_allowed = 0; 2454 } 2455 2456 spa->spa_missing_tvds_allowed = 2457 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed); 2458 2459 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2460 error = vdev_open(spa->spa_root_vdev); 2461 spa_config_exit(spa, SCL_ALL, FTAG); 2462 2463 if (spa->spa_missing_tvds != 0) { 2464 spa_load_note(spa, "vdev tree has %lld missing top-level " 2465 "vdevs.", (u_longlong_t)spa->spa_missing_tvds); 2466 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) { 2467 /* 2468 * Although theoretically we could allow users to open 2469 * incomplete pools in RW mode, we'd need to add a lot 2470 * of extra logic (e.g. adjust pool space to account 2471 * for missing vdevs). 2472 * This limitation also prevents users from accidentally 2473 * opening the pool in RW mode during data recovery and 2474 * damaging it further. 2475 */ 2476 spa_load_note(spa, "pools with missing top-level " 2477 "vdevs can only be opened in read-only mode."); 2478 error = SET_ERROR(ENXIO); 2479 } else { 2480 spa_load_note(spa, "current settings allow for maximum " 2481 "%lld missing top-level vdevs at this stage.", 2482 (u_longlong_t)spa->spa_missing_tvds_allowed); 2483 } 2484 } 2485 if (error != 0) { 2486 spa_load_failed(spa, "unable to open vdev tree [error=%d]", 2487 error); 2488 } 2489 if (spa->spa_missing_tvds != 0 || error != 0) 2490 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2); 2491 2492 return (error); 2493} 2494 2495/* 2496 * We need to validate the vdev labels against the configuration that 2497 * we have in hand. This function is called twice: first with an untrusted 2498 * config, then with a trusted config. The validation is more strict when the 2499 * config is trusted. 2500 */ 2501static int 2502spa_ld_validate_vdevs(spa_t *spa) 2503{ 2504 int error = 0; 2505 vdev_t *rvd = spa->spa_root_vdev; 2506 2507 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2508 error = vdev_validate(rvd); 2509 spa_config_exit(spa, SCL_ALL, FTAG); 2510 2511 if (error != 0) { 2512 spa_load_failed(spa, "vdev_validate failed [error=%d]", error); 2513 return (error); 2514 } 2515 2516 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) { 2517 spa_load_failed(spa, "cannot open vdev tree after invalidating " 2518 "some vdevs"); 2519 vdev_dbgmsg_print_tree(rvd, 2); 2520 return (SET_ERROR(ENXIO)); 2521 } 2522 2523 return (0); 2524} 2525 2526static int 2527spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type) 2528{ 2529 vdev_t *rvd = spa->spa_root_vdev; 2530 nvlist_t *label; 2531 uberblock_t *ub = &spa->spa_uberblock; 2532 2533 /* 2534 * Find the best uberblock. 2535 */ 2536 vdev_uberblock_load(rvd, ub, &label); 2537 2538 /* 2539 * If we weren't able to find a single valid uberblock, return failure. 2540 */ 2541 if (ub->ub_txg == 0) { 2542 nvlist_free(label); 2543 spa_load_failed(spa, "no valid uberblock found"); 2544 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2545 } 2546 2547 spa_load_note(spa, "using uberblock with txg=%llu", 2548 (u_longlong_t)ub->ub_txg); 2549 2550 /* 2551 * If the pool has an unsupported version we can't open it. 2552 */ 2553 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2554 nvlist_free(label); 2555 spa_load_failed(spa, "version %llu is not supported", 2556 (u_longlong_t)ub->ub_version); 2557 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2558 } 2559 2560 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2561 nvlist_t *features; 2562 2563 /* 2564 * If we weren't able to find what's necessary for reading the 2565 * MOS in the label, return failure. 2566 */ 2567 if (label == NULL) { 2568 spa_load_failed(spa, "label config unavailable"); 2569 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2570 ENXIO)); 2571 } 2572 2573 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ, 2574 &features) != 0) { 2575 nvlist_free(label); 2576 spa_load_failed(spa, "invalid label: '%s' missing", 2577 ZPOOL_CONFIG_FEATURES_FOR_READ); 2578 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2579 ENXIO)); 2580 } 2581 2582 /* 2583 * Update our in-core representation with the definitive values 2584 * from the label. 2585 */ 2586 nvlist_free(spa->spa_label_features); 2587 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2588 } 2589 2590 nvlist_free(label); 2591 2592 /* 2593 * Look through entries in the label nvlist's features_for_read. If 2594 * there is a feature listed there which we don't understand then we 2595 * cannot open a pool. 2596 */ 2597 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2598 nvlist_t *unsup_feat; 2599 2600 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2601 0); 2602 2603 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2604 NULL); nvp != NULL; 2605 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2606 if (!zfeature_is_supported(nvpair_name(nvp))) { 2607 VERIFY(nvlist_add_string(unsup_feat, 2608 nvpair_name(nvp), "") == 0); 2609 } 2610 } 2611 2612 if (!nvlist_empty(unsup_feat)) { 2613 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2614 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2615 nvlist_free(unsup_feat); 2616 spa_load_failed(spa, "some features are unsupported"); 2617 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2618 ENOTSUP)); 2619 } 2620 2621 nvlist_free(unsup_feat); 2622 } 2623 2624 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2625 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2626 spa_try_repair(spa, spa->spa_config); 2627 spa_config_exit(spa, SCL_ALL, FTAG); 2628 nvlist_free(spa->spa_config_splitting); 2629 spa->spa_config_splitting = NULL; 2630 } 2631 2632 /* 2633 * Initialize internal SPA structures. 2634 */ 2635 spa->spa_state = POOL_STATE_ACTIVE; 2636 spa->spa_ubsync = spa->spa_uberblock; 2637 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2638 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2639 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2640 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2641 spa->spa_claim_max_txg = spa->spa_first_txg; 2642 spa->spa_prev_software_version = ub->ub_software_version; 2643 2644 return (0); 2645} 2646 2647static int 2648spa_ld_open_rootbp(spa_t *spa) 2649{ 2650 int error = 0; 2651 vdev_t *rvd = spa->spa_root_vdev; 2652 2653 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2654 if (error != 0) { 2655 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init " 2656 "[error=%d]", error); 2657 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2658 } 2659 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2660 2661 return (0); 2662} 2663 2664static int 2665spa_ld_load_trusted_config(spa_t *spa, spa_import_type_t type, 2666 boolean_t reloading) 2667{ 2668 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 2669 nvlist_t *nv, *mos_config, *policy; 2670 int error = 0, copy_error; 2671 uint64_t healthy_tvds, healthy_tvds_mos; 2672 uint64_t mos_config_txg; 2673 2674 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE) 2675 != 0) 2676 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2677 2678 /* 2679 * If we're assembling a pool from a split, the config provided is 2680 * already trusted so there is nothing to do. 2681 */ 2682 if (type == SPA_IMPORT_ASSEMBLE) 2683 return (0); 2684 2685 healthy_tvds = spa_healthy_core_tvds(spa); 2686 2687 if (load_nvlist(spa, spa->spa_config_object, &mos_config) 2688 != 0) { 2689 spa_load_failed(spa, "unable to retrieve MOS config"); 2690 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2691 } 2692 2693 /* 2694 * If we are doing an open, pool owner wasn't verified yet, thus do 2695 * the verification here. 2696 */ 2697 if (spa->spa_load_state == SPA_LOAD_OPEN) { 2698 error = spa_verify_host(spa, mos_config); 2699 if (error != 0) { 2700 nvlist_free(mos_config); 2701 return (error); 2702 } 2703 } 2704 2705 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE); 2706 2707 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2708 2709 /* 2710 * Build a new vdev tree from the trusted config 2711 */ 2712 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 2713 2714 /* 2715 * Vdev paths in the MOS may be obsolete. If the untrusted config was 2716 * obtained by scanning /dev/dsk, then it will have the right vdev 2717 * paths. We update the trusted MOS config with this information. 2718 * We first try to copy the paths with vdev_copy_path_strict, which 2719 * succeeds only when both configs have exactly the same vdev tree. 2720 * If that fails, we fall back to a more flexible method that has a 2721 * best effort policy. 2722 */ 2723 copy_error = vdev_copy_path_strict(rvd, mrvd); 2724 if (copy_error != 0 || spa_load_print_vdev_tree) { 2725 spa_load_note(spa, "provided vdev tree:"); 2726 vdev_dbgmsg_print_tree(rvd, 2); 2727 spa_load_note(spa, "MOS vdev tree:"); 2728 vdev_dbgmsg_print_tree(mrvd, 2); 2729 } 2730 if (copy_error != 0) { 2731 spa_load_note(spa, "vdev_copy_path_strict failed, falling " 2732 "back to vdev_copy_path_relaxed"); 2733 vdev_copy_path_relaxed(rvd, mrvd); 2734 } 2735 2736 vdev_close(rvd); 2737 vdev_free(rvd); 2738 spa->spa_root_vdev = mrvd; 2739 rvd = mrvd; 2740 spa_config_exit(spa, SCL_ALL, FTAG); 2741 2742 /* 2743 * We will use spa_config if we decide to reload the spa or if spa_load 2744 * fails and we rewind. We must thus regenerate the config using the 2745 * MOS information with the updated paths. Rewind policy is an import 2746 * setting and is not in the MOS. We copy it over to our new, trusted 2747 * config. 2748 */ 2749 mos_config_txg = fnvlist_lookup_uint64(mos_config, 2750 ZPOOL_CONFIG_POOL_TXG); 2751 nvlist_free(mos_config); 2752 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE); 2753 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_REWIND_POLICY, 2754 &policy) == 0) 2755 fnvlist_add_nvlist(mos_config, ZPOOL_REWIND_POLICY, policy); 2756 spa_config_set(spa, mos_config); 2757 spa->spa_config_source = SPA_CONFIG_SRC_MOS; 2758 2759 /* 2760 * Now that we got the config from the MOS, we should be more strict 2761 * in checking blkptrs and can make assumptions about the consistency 2762 * of the vdev tree. spa_trust_config must be set to true before opening 2763 * vdevs in order for them to be writeable. 2764 */ 2765 spa->spa_trust_config = B_TRUE; 2766 2767 /* 2768 * Open and validate the new vdev tree 2769 */ 2770 error = spa_ld_open_vdevs(spa); 2771 if (error != 0) 2772 return (error); 2773 2774 error = spa_ld_validate_vdevs(spa); 2775 if (error != 0) 2776 return (error); 2777 2778 if (copy_error != 0 || spa_load_print_vdev_tree) { 2779 spa_load_note(spa, "final vdev tree:"); 2780 vdev_dbgmsg_print_tree(rvd, 2); 2781 } 2782 2783 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT && 2784 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) { 2785 /* 2786 * Sanity check to make sure that we are indeed loading the 2787 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds 2788 * in the config provided and they happened to be the only ones 2789 * to have the latest uberblock, we could involuntarily perform 2790 * an extreme rewind. 2791 */ 2792 healthy_tvds_mos = spa_healthy_core_tvds(spa); 2793 if (healthy_tvds_mos - healthy_tvds >= 2794 SPA_SYNC_MIN_VDEVS) { 2795 spa_load_note(spa, "config provided misses too many " 2796 "top-level vdevs compared to MOS (%lld vs %lld). ", 2797 (u_longlong_t)healthy_tvds, 2798 (u_longlong_t)healthy_tvds_mos); 2799 spa_load_note(spa, "vdev tree:"); 2800 vdev_dbgmsg_print_tree(rvd, 2); 2801 if (reloading) { 2802 spa_load_failed(spa, "config was already " 2803 "provided from MOS. Aborting."); 2804 return (spa_vdev_err(rvd, 2805 VDEV_AUX_CORRUPT_DATA, EIO)); 2806 } 2807 spa_load_note(spa, "spa must be reloaded using MOS " 2808 "config"); 2809 return (SET_ERROR(EAGAIN)); 2810 } 2811 } 2812 2813 error = spa_check_for_missing_logs(spa); 2814 if (error != 0) 2815 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2816 2817 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) { 2818 spa_load_failed(spa, "uberblock guid sum doesn't match MOS " 2819 "guid sum (%llu != %llu)", 2820 (u_longlong_t)spa->spa_uberblock.ub_guid_sum, 2821 (u_longlong_t)rvd->vdev_guid_sum); 2822 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2823 ENXIO)); 2824 } 2825 2826 return (0); 2827} 2828 2829static int 2830spa_ld_open_indirect_vdev_metadata(spa_t *spa) 2831{ 2832 int error = 0; 2833 vdev_t *rvd = spa->spa_root_vdev; 2834 2835 /* 2836 * Everything that we read before spa_remove_init() must be stored 2837 * on concreted vdevs. Therefore we do this as early as possible. 2838 */ 2839 error = spa_remove_init(spa); 2840 if (error != 0) { 2841 spa_load_failed(spa, "spa_remove_init failed [error=%d]", 2842 error); 2843 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2844 } 2845 2846 /* 2847 * Retrieve information needed to condense indirect vdev mappings. 2848 */ 2849 error = spa_condense_init(spa); 2850 if (error != 0) { 2851 spa_load_failed(spa, "spa_condense_init failed [error=%d]", 2852 error); 2853 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 2854 } 2855 2856 return (0); 2857} 2858 2859static int 2860spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep) 2861{ 2862 int error = 0; 2863 vdev_t *rvd = spa->spa_root_vdev; 2864 2865 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2866 boolean_t missing_feat_read = B_FALSE; 2867 nvlist_t *unsup_feat, *enabled_feat; 2868 2869 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2870 &spa->spa_feat_for_read_obj, B_TRUE) != 0) { 2871 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2872 } 2873 2874 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2875 &spa->spa_feat_for_write_obj, B_TRUE) != 0) { 2876 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2877 } 2878 2879 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2880 &spa->spa_feat_desc_obj, B_TRUE) != 0) { 2881 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2882 } 2883 2884 enabled_feat = fnvlist_alloc(); 2885 unsup_feat = fnvlist_alloc(); 2886 2887 if (!spa_features_check(spa, B_FALSE, 2888 unsup_feat, enabled_feat)) 2889 missing_feat_read = B_TRUE; 2890 2891 if (spa_writeable(spa) || 2892 spa->spa_load_state == SPA_LOAD_TRYIMPORT) { 2893 if (!spa_features_check(spa, B_TRUE, 2894 unsup_feat, enabled_feat)) { 2895 *missing_feat_writep = B_TRUE; 2896 } 2897 } 2898 2899 fnvlist_add_nvlist(spa->spa_load_info, 2900 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2901 2902 if (!nvlist_empty(unsup_feat)) { 2903 fnvlist_add_nvlist(spa->spa_load_info, 2904 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2905 } 2906 2907 fnvlist_free(enabled_feat); 2908 fnvlist_free(unsup_feat); 2909 2910 if (!missing_feat_read) { 2911 fnvlist_add_boolean(spa->spa_load_info, 2912 ZPOOL_CONFIG_CAN_RDONLY); 2913 } 2914 2915 /* 2916 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2917 * twofold: to determine whether the pool is available for 2918 * import in read-write mode and (if it is not) whether the 2919 * pool is available for import in read-only mode. If the pool 2920 * is available for import in read-write mode, it is displayed 2921 * as available in userland; if it is not available for import 2922 * in read-only mode, it is displayed as unavailable in 2923 * userland. If the pool is available for import in read-only 2924 * mode but not read-write mode, it is displayed as unavailable 2925 * in userland with a special note that the pool is actually 2926 * available for open in read-only mode. 2927 * 2928 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2929 * missing a feature for write, we must first determine whether 2930 * the pool can be opened read-only before returning to 2931 * userland in order to know whether to display the 2932 * abovementioned note. 2933 */ 2934 if (missing_feat_read || (*missing_feat_writep && 2935 spa_writeable(spa))) { 2936 spa_load_failed(spa, "pool uses unsupported features"); 2937 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2938 ENOTSUP)); 2939 } 2940 2941 /* 2942 * Load refcounts for ZFS features from disk into an in-memory 2943 * cache during SPA initialization. 2944 */ 2945 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2946 uint64_t refcount; 2947 2948 error = feature_get_refcount_from_disk(spa, 2949 &spa_feature_table[i], &refcount); 2950 if (error == 0) { 2951 spa->spa_feat_refcount_cache[i] = refcount; 2952 } else if (error == ENOTSUP) { 2953 spa->spa_feat_refcount_cache[i] = 2954 SPA_FEATURE_DISABLED; 2955 } else { 2956 spa_load_failed(spa, "error getting refcount " 2957 "for feature %s [error=%d]", 2958 spa_feature_table[i].fi_guid, error); 2959 return (spa_vdev_err(rvd, 2960 VDEV_AUX_CORRUPT_DATA, EIO)); 2961 } 2962 } 2963 } 2964 2965 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2966 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2967 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0) 2968 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2969 } 2970 2971 return (0); 2972} 2973 2974static int 2975spa_ld_load_special_directories(spa_t *spa) 2976{ 2977 int error = 0; 2978 vdev_t *rvd = spa->spa_root_vdev; 2979 2980 spa->spa_is_initializing = B_TRUE; 2981 error = dsl_pool_open(spa->spa_dsl_pool); 2982 spa->spa_is_initializing = B_FALSE; 2983 if (error != 0) { 2984 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error); 2985 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2986 } 2987 2988 return (0); 2989} 2990 2991static int 2992spa_ld_get_props(spa_t *spa) 2993{ 2994 int error = 0; 2995 uint64_t obj; 2996 vdev_t *rvd = spa->spa_root_vdev; 2997 2998 /* Grab the secret checksum salt from the MOS. */ 2999 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 3000 DMU_POOL_CHECKSUM_SALT, 1, 3001 sizeof (spa->spa_cksum_salt.zcs_bytes), 3002 spa->spa_cksum_salt.zcs_bytes); 3003 if (error == ENOENT) { 3004 /* Generate a new salt for subsequent use */ 3005 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3006 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3007 } else if (error != 0) { 3008 spa_load_failed(spa, "unable to retrieve checksum salt from " 3009 "MOS [error=%d]", error); 3010 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3011 } 3012 3013 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0) 3014 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3015 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 3016 if (error != 0) { 3017 spa_load_failed(spa, "error opening deferred-frees bpobj " 3018 "[error=%d]", error); 3019 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3020 } 3021 3022 /* 3023 * Load the bit that tells us to use the new accounting function 3024 * (raid-z deflation). If we have an older pool, this will not 3025 * be present. 3026 */ 3027 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE); 3028 if (error != 0 && error != ENOENT) 3029 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3030 3031 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 3032 &spa->spa_creation_version, B_FALSE); 3033 if (error != 0 && error != ENOENT) 3034 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3035 3036 /* 3037 * Load the persistent error log. If we have an older pool, this will 3038 * not be present. 3039 */ 3040 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last, 3041 B_FALSE); 3042 if (error != 0 && error != ENOENT) 3043 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3044 3045 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 3046 &spa->spa_errlog_scrub, B_FALSE); 3047 if (error != 0 && error != ENOENT) 3048 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3049 3050 /* 3051 * Load the history object. If we have an older pool, this 3052 * will not be present. 3053 */ 3054 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE); 3055 if (error != 0 && error != ENOENT) 3056 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3057 3058 /* 3059 * Load the per-vdev ZAP map. If we have an older pool, this will not 3060 * be present; in this case, defer its creation to a later time to 3061 * avoid dirtying the MOS this early / out of sync context. See 3062 * spa_sync_config_object. 3063 */ 3064 3065 /* The sentinel is only available in the MOS config. */ 3066 nvlist_t *mos_config; 3067 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) { 3068 spa_load_failed(spa, "unable to retrieve MOS config"); 3069 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3070 } 3071 3072 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 3073 &spa->spa_all_vdev_zaps, B_FALSE); 3074 3075 if (error == ENOENT) { 3076 VERIFY(!nvlist_exists(mos_config, 3077 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 3078 spa->spa_avz_action = AVZ_ACTION_INITIALIZE; 3079 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3080 } else if (error != 0) { 3081 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3082 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 3083 /* 3084 * An older version of ZFS overwrote the sentinel value, so 3085 * we have orphaned per-vdev ZAPs in the MOS. Defer their 3086 * destruction to later; see spa_sync_config_object. 3087 */ 3088 spa->spa_avz_action = AVZ_ACTION_DESTROY; 3089 /* 3090 * We're assuming that no vdevs have had their ZAPs created 3091 * before this. Better be sure of it. 3092 */ 3093 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 3094 } 3095 nvlist_free(mos_config); 3096 3097 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3098 3099 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object, 3100 B_FALSE); 3101 if (error && error != ENOENT) 3102 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3103 3104 if (error == 0) { 3105 uint64_t autoreplace; 3106 3107 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 3108 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 3109 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 3110 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 3111 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 3112 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 3113 &spa->spa_dedup_ditto); 3114 3115 spa->spa_autoreplace = (autoreplace != 0); 3116 } 3117 3118 /* 3119 * If we are importing a pool with missing top-level vdevs, 3120 * we enforce that the pool doesn't panic or get suspended on 3121 * error since the likelihood of missing data is extremely high. 3122 */ 3123 if (spa->spa_missing_tvds > 0 && 3124 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE && 3125 spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3126 spa_load_note(spa, "forcing failmode to 'continue' " 3127 "as some top level vdevs are missing"); 3128 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE; 3129 } 3130 3131 return (0); 3132} 3133 3134static int 3135spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type) 3136{ 3137 int error = 0; 3138 vdev_t *rvd = spa->spa_root_vdev; 3139 3140 /* 3141 * If we're assembling the pool from the split-off vdevs of 3142 * an existing pool, we don't want to attach the spares & cache 3143 * devices. 3144 */ 3145 3146 /* 3147 * Load any hot spares for this pool. 3148 */ 3149 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object, 3150 B_FALSE); 3151 if (error != 0 && error != ENOENT) 3152 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3153 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3154 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 3155 if (load_nvlist(spa, spa->spa_spares.sav_object, 3156 &spa->spa_spares.sav_config) != 0) { 3157 spa_load_failed(spa, "error loading spares nvlist"); 3158 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3159 } 3160 3161 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3162 spa_load_spares(spa); 3163 spa_config_exit(spa, SCL_ALL, FTAG); 3164 } else if (error == 0) { 3165 spa->spa_spares.sav_sync = B_TRUE; 3166 } 3167 3168 /* 3169 * Load any level 2 ARC devices for this pool. 3170 */ 3171 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 3172 &spa->spa_l2cache.sav_object, B_FALSE); 3173 if (error != 0 && error != ENOENT) 3174 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3175 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 3176 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 3177 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 3178 &spa->spa_l2cache.sav_config) != 0) { 3179 spa_load_failed(spa, "error loading l2cache nvlist"); 3180 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3181 } 3182 3183 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3184 spa_load_l2cache(spa); 3185 spa_config_exit(spa, SCL_ALL, FTAG); 3186 } else if (error == 0) { 3187 spa->spa_l2cache.sav_sync = B_TRUE; 3188 } 3189 3190 return (0); 3191} 3192 3193static int 3194spa_ld_load_vdev_metadata(spa_t *spa) 3195{ 3196 int error = 0; 3197 vdev_t *rvd = spa->spa_root_vdev; 3198 3199 /* 3200 * If the 'autoreplace' property is set, then post a resource notifying 3201 * the ZFS DE that it should not issue any faults for unopenable 3202 * devices. We also iterate over the vdevs, and post a sysevent for any 3203 * unopenable vdevs so that the normal autoreplace handler can take 3204 * over. 3205 */ 3206 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3207 spa_check_removed(spa->spa_root_vdev); 3208 /* 3209 * For the import case, this is done in spa_import(), because 3210 * at this point we're using the spare definitions from 3211 * the MOS config, not necessarily from the userland config. 3212 */ 3213 if (spa->spa_load_state != SPA_LOAD_IMPORT) { 3214 spa_aux_check_removed(&spa->spa_spares); 3215 spa_aux_check_removed(&spa->spa_l2cache); 3216 } 3217 } 3218 3219 /* 3220 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc. 3221 */ 3222 error = vdev_load(rvd); 3223 if (error != 0) { 3224 spa_load_failed(spa, "vdev_load failed [error=%d]", error); 3225 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error)); 3226 } 3227 3228 /* 3229 * Propagate the leaf DTLs we just loaded all the way up the vdev tree. 3230 */ 3231 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3232 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 3233 spa_config_exit(spa, SCL_ALL, FTAG); 3234 3235 return (0); 3236} 3237 3238static int 3239spa_ld_load_dedup_tables(spa_t *spa) 3240{ 3241 int error = 0; 3242 vdev_t *rvd = spa->spa_root_vdev; 3243 3244 error = ddt_load(spa); 3245 if (error != 0) { 3246 spa_load_failed(spa, "ddt_load failed [error=%d]", error); 3247 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 3248 } 3249 3250 return (0); 3251} 3252 3253static int 3254spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport) 3255{ 3256 vdev_t *rvd = spa->spa_root_vdev; 3257 3258 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) { 3259 boolean_t missing = spa_check_logs(spa); 3260 if (missing) { 3261 if (spa->spa_missing_tvds != 0) { 3262 spa_load_note(spa, "spa_check_logs failed " 3263 "so dropping the logs"); 3264 } else { 3265 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 3266 spa_load_failed(spa, "spa_check_logs failed"); 3267 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, 3268 ENXIO)); 3269 } 3270 } 3271 } 3272 3273 return (0); 3274} 3275 3276static int 3277spa_ld_verify_pool_data(spa_t *spa) 3278{ 3279 int error = 0; 3280 vdev_t *rvd = spa->spa_root_vdev; 3281 3282 /* 3283 * We've successfully opened the pool, verify that we're ready 3284 * to start pushing transactions. 3285 */ 3286 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) { 3287 error = spa_load_verify(spa); 3288 if (error != 0) { 3289 spa_load_failed(spa, "spa_load_verify failed " 3290 "[error=%d]", error); 3291 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 3292 error)); 3293 } 3294 } 3295 3296 return (0); 3297} 3298 3299static void 3300spa_ld_claim_log_blocks(spa_t *spa) 3301{ 3302 dmu_tx_t *tx; 3303 dsl_pool_t *dp = spa_get_dsl(spa); 3304 3305 /* 3306 * Claim log blocks that haven't been committed yet. 3307 * This must all happen in a single txg. 3308 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 3309 * invoked from zil_claim_log_block()'s i/o done callback. 3310 * Price of rollback is that we abandon the log. 3311 */ 3312 spa->spa_claiming = B_TRUE; 3313 3314 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 3315 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 3316 zil_claim, tx, DS_FIND_CHILDREN); 3317 dmu_tx_commit(tx); 3318 3319 spa->spa_claiming = B_FALSE; 3320 3321 spa_set_log_state(spa, SPA_LOG_GOOD); 3322} 3323 3324static void 3325spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg, 3326 boolean_t reloading) 3327{ 3328 vdev_t *rvd = spa->spa_root_vdev; 3329 int need_update = B_FALSE; 3330 3331 /* 3332 * If the config cache is stale, or we have uninitialized 3333 * metaslabs (see spa_vdev_add()), then update the config. 3334 * 3335 * If this is a verbatim import, trust the current 3336 * in-core spa_config and update the disk labels. 3337 */ 3338 if (reloading || config_cache_txg != spa->spa_config_txg || 3339 spa->spa_load_state == SPA_LOAD_IMPORT || 3340 spa->spa_load_state == SPA_LOAD_RECOVER || 3341 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 3342 need_update = B_TRUE; 3343 3344 for (int c = 0; c < rvd->vdev_children; c++) 3345 if (rvd->vdev_child[c]->vdev_ms_array == 0) 3346 need_update = B_TRUE; 3347 3348 /* 3349 * Update the config cache asychronously in case we're the 3350 * root pool, in which case the config cache isn't writable yet. 3351 */ 3352 if (need_update) 3353 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3354} 3355 3356static void 3357spa_ld_prepare_for_reload(spa_t *spa) 3358{ 3359 int mode = spa->spa_mode; 3360 int async_suspended = spa->spa_async_suspended; 3361 3362 spa_unload(spa); 3363 spa_deactivate(spa); 3364 spa_activate(spa, mode); 3365 3366 /* 3367 * We save the value of spa_async_suspended as it gets reset to 0 by 3368 * spa_unload(). We want to restore it back to the original value before 3369 * returning as we might be calling spa_async_resume() later. 3370 */ 3371 spa->spa_async_suspended = async_suspended; 3372} 3373 3374/* 3375 * Load an existing storage pool, using the config provided. This config 3376 * describes which vdevs are part of the pool and is later validated against 3377 * partial configs present in each vdev's label and an entire copy of the 3378 * config stored in the MOS. 3379 */ 3380static int 3381spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport, 3382 boolean_t reloading) 3383{ 3384 int error = 0; 3385 boolean_t missing_feat_write = B_FALSE; 3386 3387 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 3388 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE); 3389 3390 /* 3391 * Never trust the config that is provided unless we are assembling 3392 * a pool following a split. 3393 * This means don't trust blkptrs and the vdev tree in general. This 3394 * also effectively puts the spa in read-only mode since 3395 * spa_writeable() checks for spa_trust_config to be true. 3396 * We will later load a trusted config from the MOS. 3397 */ 3398 if (type != SPA_IMPORT_ASSEMBLE) 3399 spa->spa_trust_config = B_FALSE; 3400 3401 if (reloading) 3402 spa_load_note(spa, "RELOADING"); 3403 else 3404 spa_load_note(spa, "LOADING"); 3405 3406 /* 3407 * Parse the config provided to create a vdev tree. 3408 */ 3409 error = spa_ld_parse_config(spa, type); 3410 if (error != 0) 3411 return (error); 3412 3413 /* 3414 * Now that we have the vdev tree, try to open each vdev. This involves 3415 * opening the underlying physical device, retrieving its geometry and 3416 * probing the vdev with a dummy I/O. The state of each vdev will be set 3417 * based on the success of those operations. After this we'll be ready 3418 * to read from the vdevs. 3419 */ 3420 error = spa_ld_open_vdevs(spa); 3421 if (error != 0) 3422 return (error); 3423 3424 /* 3425 * Read the label of each vdev and make sure that the GUIDs stored 3426 * there match the GUIDs in the config provided. 3427 * If we're assembling a new pool that's been split off from an 3428 * existing pool, the labels haven't yet been updated so we skip 3429 * validation for now. 3430 */ 3431 if (type != SPA_IMPORT_ASSEMBLE) { 3432 error = spa_ld_validate_vdevs(spa); 3433 if (error != 0) 3434 return (error); 3435 } 3436 3437 /* 3438 * Read vdev labels to find the best uberblock (i.e. latest, unless 3439 * spa_load_max_txg is set) and store it in spa_uberblock. We get the 3440 * list of features required to read blkptrs in the MOS from the vdev 3441 * label with the best uberblock and verify that our version of zfs 3442 * supports them all. 3443 */ 3444 error = spa_ld_select_uberblock(spa, type); 3445 if (error != 0) 3446 return (error); 3447 3448 /* 3449 * Pass that uberblock to the dsl_pool layer which will open the root 3450 * blkptr. This blkptr points to the latest version of the MOS and will 3451 * allow us to read its contents. 3452 */ 3453 error = spa_ld_open_rootbp(spa); 3454 if (error != 0) 3455 return (error); 3456 3457 /* 3458 * Retrieve the trusted config stored in the MOS and use it to create 3459 * a new, exact version of the vdev tree, then reopen all vdevs. 3460 */ 3461 error = spa_ld_load_trusted_config(spa, type, reloading); 3462 if (error == EAGAIN) { 3463 VERIFY(!reloading); 3464 /* 3465 * Redo the loading process with the trusted config if it is 3466 * too different from the untrusted config. 3467 */ 3468 spa_ld_prepare_for_reload(spa); 3469 return (spa_load_impl(spa, type, ereport, B_TRUE)); 3470 } else if (error != 0) { 3471 return (error); 3472 } 3473 3474 /* 3475 * Retrieve the mapping of indirect vdevs. Those vdevs were removed 3476 * from the pool and their contents were re-mapped to other vdevs. Note 3477 * that everything that we read before this step must have been 3478 * rewritten on concrete vdevs after the last device removal was 3479 * initiated. Otherwise we could be reading from indirect vdevs before 3480 * we have loaded their mappings. 3481 */ 3482 error = spa_ld_open_indirect_vdev_metadata(spa); 3483 if (error != 0) 3484 return (error); 3485 3486 /* 3487 * Retrieve the full list of active features from the MOS and check if 3488 * they are all supported. 3489 */ 3490 error = spa_ld_check_features(spa, &missing_feat_write); 3491 if (error != 0) 3492 return (error); 3493 3494 /* 3495 * Load several special directories from the MOS needed by the dsl_pool 3496 * layer. 3497 */ 3498 error = spa_ld_load_special_directories(spa); 3499 if (error != 0) 3500 return (error); 3501 3502 /* 3503 * Retrieve pool properties from the MOS. 3504 */ 3505 error = spa_ld_get_props(spa); 3506 if (error != 0) 3507 return (error); 3508 3509 /* 3510 * Retrieve the list of auxiliary devices - cache devices and spares - 3511 * and open them. 3512 */ 3513 error = spa_ld_open_aux_vdevs(spa, type); 3514 if (error != 0) 3515 return (error); 3516 3517 /* 3518 * Load the metadata for all vdevs. Also check if unopenable devices 3519 * should be autoreplaced. 3520 */ 3521 error = spa_ld_load_vdev_metadata(spa); 3522 if (error != 0) 3523 return (error); 3524 3525 error = spa_ld_load_dedup_tables(spa); 3526 if (error != 0) 3527 return (error); 3528 3529 /* 3530 * Verify the logs now to make sure we don't have any unexpected errors 3531 * when we claim log blocks later. 3532 */ 3533 error = spa_ld_verify_logs(spa, type, ereport); 3534 if (error != 0) 3535 return (error); 3536 3537 if (missing_feat_write) { 3538 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT); 3539 3540 /* 3541 * At this point, we know that we can open the pool in 3542 * read-only mode but not read-write mode. We now have enough 3543 * information and can return to userland. 3544 */ 3545 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT, 3546 ENOTSUP)); 3547 } 3548 3549 /* 3550 * Traverse the last txgs to make sure the pool was left off in a safe 3551 * state. When performing an extreme rewind, we verify the whole pool, 3552 * which can take a very long time. 3553 */ 3554 error = spa_ld_verify_pool_data(spa); 3555 if (error != 0) 3556 return (error); 3557 3558 /* 3559 * Calculate the deflated space for the pool. This must be done before 3560 * we write anything to the pool because we'd need to update the space 3561 * accounting using the deflated sizes. 3562 */ 3563 spa_update_dspace(spa); 3564 3565 /* 3566 * We have now retrieved all the information we needed to open the 3567 * pool. If we are importing the pool in read-write mode, a few 3568 * additional steps must be performed to finish the import. 3569 */ 3570 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER || 3571 spa->spa_load_max_txg == UINT64_MAX)) { 3572 uint64_t config_cache_txg = spa->spa_config_txg; 3573 3574 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT); 3575 3576 /* 3577 * We must check this before we start the sync thread, because 3578 * we only want to start a condense thread for condense 3579 * operations that were in progress when the pool was 3580 * imported. Once we start syncing, spa_sync() could 3581 * initiate a condense (and start a thread for it). In 3582 * that case it would be wrong to start a second 3583 * condense thread. 3584 */ 3585 boolean_t condense_in_progress = 3586 (spa->spa_condensing_indirect != NULL); 3587 3588 /* 3589 * Traverse the ZIL and claim all blocks. 3590 */ 3591 spa_ld_claim_log_blocks(spa); 3592 3593 /* 3594 * Kick-off the syncing thread. 3595 */ 3596 spa->spa_sync_on = B_TRUE; 3597 txg_sync_start(spa->spa_dsl_pool); 3598 3599 /* 3600 * Wait for all claims to sync. We sync up to the highest 3601 * claimed log block birth time so that claimed log blocks 3602 * don't appear to be from the future. spa_claim_max_txg 3603 * will have been set for us by ZIL traversal operations 3604 * performed above. 3605 */ 3606 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 3607 3608 /* 3609 * Check if we need to request an update of the config. On the 3610 * next sync, we would update the config stored in vdev labels 3611 * and the cachefile (by default /etc/zfs/zpool.cache). 3612 */ 3613 spa_ld_check_for_config_update(spa, config_cache_txg, 3614 reloading); 3615 3616 /* 3617 * Check all DTLs to see if anything needs resilvering. 3618 */ 3619 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 3620 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) 3621 spa_async_request(spa, SPA_ASYNC_RESILVER); 3622 3623 /* 3624 * Log the fact that we booted up (so that we can detect if 3625 * we rebooted in the middle of an operation). 3626 */ 3627 spa_history_log_version(spa, "open"); 3628 3629 /* 3630 * Delete any inconsistent datasets. 3631 */ 3632 (void) dmu_objset_find(spa_name(spa), 3633 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 3634 3635 /* 3636 * Clean up any stale temporary dataset userrefs. 3637 */ 3638 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 3639 3640 /* 3641 * Note: unlike condensing, we don't need an analogous 3642 * "removal_in_progress" dance because no other thread 3643 * can start a removal while we hold the spa_namespace_lock. 3644 */ 3645 spa_restart_removal(spa); 3646 3647 if (condense_in_progress) 3648 spa_condense_indirect_restart(spa); 3649 } 3650 3651 spa_load_note(spa, "LOADED"); 3652 3653 return (0); 3654} 3655 3656static int 3657spa_load_retry(spa_t *spa, spa_load_state_t state) 3658{ 3659 int mode = spa->spa_mode; 3660 3661 spa_unload(spa); 3662 spa_deactivate(spa); 3663 3664 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 3665 3666 spa_activate(spa, mode); 3667 spa_async_suspend(spa); 3668 3669 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu", 3670 (u_longlong_t)spa->spa_load_max_txg); 3671 3672 return (spa_load(spa, state, SPA_IMPORT_EXISTING)); 3673} 3674 3675/* 3676 * If spa_load() fails this function will try loading prior txg's. If 3677 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 3678 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 3679 * function will not rewind the pool and will return the same error as 3680 * spa_load(). 3681 */ 3682static int 3683spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request, 3684 int rewind_flags) 3685{ 3686 nvlist_t *loadinfo = NULL; 3687 nvlist_t *config = NULL; 3688 int load_error, rewind_error; 3689 uint64_t safe_rewind_txg; 3690 uint64_t min_txg; 3691 3692 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 3693 spa->spa_load_max_txg = spa->spa_load_txg; 3694 spa_set_log_state(spa, SPA_LOG_CLEAR); 3695 } else { 3696 spa->spa_load_max_txg = max_request; 3697 if (max_request != UINT64_MAX) 3698 spa->spa_extreme_rewind = B_TRUE; 3699 } 3700 3701 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING); 3702 if (load_error == 0) 3703 return (0); 3704 3705 if (spa->spa_root_vdev != NULL) 3706 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3707 3708 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 3709 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 3710 3711 if (rewind_flags & ZPOOL_NEVER_REWIND) { 3712 nvlist_free(config); 3713 return (load_error); 3714 } 3715 3716 if (state == SPA_LOAD_RECOVER) { 3717 /* Price of rolling back is discarding txgs, including log */ 3718 spa_set_log_state(spa, SPA_LOG_CLEAR); 3719 } else { 3720 /* 3721 * If we aren't rolling back save the load info from our first 3722 * import attempt so that we can restore it after attempting 3723 * to rewind. 3724 */ 3725 loadinfo = spa->spa_load_info; 3726 spa->spa_load_info = fnvlist_alloc(); 3727 } 3728 3729 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 3730 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 3731 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 3732 TXG_INITIAL : safe_rewind_txg; 3733 3734 /* 3735 * Continue as long as we're finding errors, we're still within 3736 * the acceptable rewind range, and we're still finding uberblocks 3737 */ 3738 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 3739 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 3740 if (spa->spa_load_max_txg < safe_rewind_txg) 3741 spa->spa_extreme_rewind = B_TRUE; 3742 rewind_error = spa_load_retry(spa, state); 3743 } 3744 3745 spa->spa_extreme_rewind = B_FALSE; 3746 spa->spa_load_max_txg = UINT64_MAX; 3747 3748 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 3749 spa_config_set(spa, config); 3750 else 3751 nvlist_free(config); 3752 3753 if (state == SPA_LOAD_RECOVER) { 3754 ASSERT3P(loadinfo, ==, NULL); 3755 return (rewind_error); 3756 } else { 3757 /* Store the rewind info as part of the initial load info */ 3758 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3759 spa->spa_load_info); 3760 3761 /* Restore the initial load info */ 3762 fnvlist_free(spa->spa_load_info); 3763 spa->spa_load_info = loadinfo; 3764 3765 return (load_error); 3766 } 3767} 3768 3769/* 3770 * Pool Open/Import 3771 * 3772 * The import case is identical to an open except that the configuration is sent 3773 * down from userland, instead of grabbed from the configuration cache. For the 3774 * case of an open, the pool configuration will exist in the 3775 * POOL_STATE_UNINITIALIZED state. 3776 * 3777 * The stats information (gen/count/ustats) is used to gather vdev statistics at 3778 * the same time open the pool, without having to keep around the spa_t in some 3779 * ambiguous state. 3780 */ 3781static int 3782spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3783 nvlist_t **config) 3784{ 3785 spa_t *spa; 3786 spa_load_state_t state = SPA_LOAD_OPEN; 3787 int error; 3788 int locked = B_FALSE; 3789 int firstopen = B_FALSE; 3790 3791 *spapp = NULL; 3792 3793 /* 3794 * As disgusting as this is, we need to support recursive calls to this 3795 * function because dsl_dir_open() is called during spa_load(), and ends 3796 * up calling spa_open() again. The real fix is to figure out how to 3797 * avoid dsl_dir_open() calling this in the first place. 3798 */ 3799 if (mutex_owner(&spa_namespace_lock) != curthread) { 3800 mutex_enter(&spa_namespace_lock); 3801 locked = B_TRUE; 3802 } 3803 3804 if ((spa = spa_lookup(pool)) == NULL) { 3805 if (locked) 3806 mutex_exit(&spa_namespace_lock); 3807 return (SET_ERROR(ENOENT)); 3808 } 3809 3810 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3811 zpool_rewind_policy_t policy; 3812 3813 firstopen = B_TRUE; 3814 3815 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3816 &policy); 3817 if (policy.zrp_request & ZPOOL_DO_REWIND) 3818 state = SPA_LOAD_RECOVER; 3819 3820 spa_activate(spa, spa_mode_global); 3821 3822 if (state != SPA_LOAD_RECOVER) 3823 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3824 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 3825 3826 zfs_dbgmsg("spa_open_common: opening %s", pool); 3827 error = spa_load_best(spa, state, policy.zrp_txg, 3828 policy.zrp_request); 3829 3830 if (error == EBADF) { 3831 /* 3832 * If vdev_validate() returns failure (indicated by 3833 * EBADF), it indicates that one of the vdevs indicates 3834 * that the pool has been exported or destroyed. If 3835 * this is the case, the config cache is out of sync and 3836 * we should remove the pool from the namespace. 3837 */ 3838 spa_unload(spa); 3839 spa_deactivate(spa); 3840 spa_write_cachefile(spa, B_TRUE, B_TRUE); 3841 spa_remove(spa); 3842 if (locked) 3843 mutex_exit(&spa_namespace_lock); 3844 return (SET_ERROR(ENOENT)); 3845 } 3846 3847 if (error) { 3848 /* 3849 * We can't open the pool, but we still have useful 3850 * information: the state of each vdev after the 3851 * attempted vdev_open(). Return this to the user. 3852 */ 3853 if (config != NULL && spa->spa_config) { 3854 VERIFY(nvlist_dup(spa->spa_config, config, 3855 KM_SLEEP) == 0); 3856 VERIFY(nvlist_add_nvlist(*config, 3857 ZPOOL_CONFIG_LOAD_INFO, 3858 spa->spa_load_info) == 0); 3859 } 3860 spa_unload(spa); 3861 spa_deactivate(spa); 3862 spa->spa_last_open_failed = error; 3863 if (locked) 3864 mutex_exit(&spa_namespace_lock); 3865 *spapp = NULL; 3866 return (error); 3867 } 3868 } 3869 3870 spa_open_ref(spa, tag); 3871 3872 if (config != NULL) 3873 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3874 3875 /* 3876 * If we've recovered the pool, pass back any information we 3877 * gathered while doing the load. 3878 */ 3879 if (state == SPA_LOAD_RECOVER) { 3880 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3881 spa->spa_load_info) == 0); 3882 } 3883 3884 if (locked) { 3885 spa->spa_last_open_failed = 0; 3886 spa->spa_last_ubsync_txg = 0; 3887 spa->spa_load_txg = 0; 3888 mutex_exit(&spa_namespace_lock); 3889#ifdef __FreeBSD__ 3890#ifdef _KERNEL 3891 if (firstopen) 3892 zvol_create_minors(spa->spa_name); 3893#endif 3894#endif 3895 } 3896 3897 *spapp = spa; 3898 3899 return (0); 3900} 3901 3902int 3903spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3904 nvlist_t **config) 3905{ 3906 return (spa_open_common(name, spapp, tag, policy, config)); 3907} 3908 3909int 3910spa_open(const char *name, spa_t **spapp, void *tag) 3911{ 3912 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3913} 3914 3915/* 3916 * Lookup the given spa_t, incrementing the inject count in the process, 3917 * preventing it from being exported or destroyed. 3918 */ 3919spa_t * 3920spa_inject_addref(char *name) 3921{ 3922 spa_t *spa; 3923 3924 mutex_enter(&spa_namespace_lock); 3925 if ((spa = spa_lookup(name)) == NULL) { 3926 mutex_exit(&spa_namespace_lock); 3927 return (NULL); 3928 } 3929 spa->spa_inject_ref++; 3930 mutex_exit(&spa_namespace_lock); 3931 3932 return (spa); 3933} 3934 3935void 3936spa_inject_delref(spa_t *spa) 3937{ 3938 mutex_enter(&spa_namespace_lock); 3939 spa->spa_inject_ref--; 3940 mutex_exit(&spa_namespace_lock); 3941} 3942 3943/* 3944 * Add spares device information to the nvlist. 3945 */ 3946static void 3947spa_add_spares(spa_t *spa, nvlist_t *config) 3948{ 3949 nvlist_t **spares; 3950 uint_t i, nspares; 3951 nvlist_t *nvroot; 3952 uint64_t guid; 3953 vdev_stat_t *vs; 3954 uint_t vsc; 3955 uint64_t pool; 3956 3957 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3958 3959 if (spa->spa_spares.sav_count == 0) 3960 return; 3961 3962 VERIFY(nvlist_lookup_nvlist(config, 3963 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3964 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3965 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3966 if (nspares != 0) { 3967 VERIFY(nvlist_add_nvlist_array(nvroot, 3968 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3969 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3970 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3971 3972 /* 3973 * Go through and find any spares which have since been 3974 * repurposed as an active spare. If this is the case, update 3975 * their status appropriately. 3976 */ 3977 for (i = 0; i < nspares; i++) { 3978 VERIFY(nvlist_lookup_uint64(spares[i], 3979 ZPOOL_CONFIG_GUID, &guid) == 0); 3980 if (spa_spare_exists(guid, &pool, NULL) && 3981 pool != 0ULL) { 3982 VERIFY(nvlist_lookup_uint64_array( 3983 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3984 (uint64_t **)&vs, &vsc) == 0); 3985 vs->vs_state = VDEV_STATE_CANT_OPEN; 3986 vs->vs_aux = VDEV_AUX_SPARED; 3987 } 3988 } 3989 } 3990} 3991 3992/* 3993 * Add l2cache device information to the nvlist, including vdev stats. 3994 */ 3995static void 3996spa_add_l2cache(spa_t *spa, nvlist_t *config) 3997{ 3998 nvlist_t **l2cache; 3999 uint_t i, j, nl2cache; 4000 nvlist_t *nvroot; 4001 uint64_t guid; 4002 vdev_t *vd; 4003 vdev_stat_t *vs; 4004 uint_t vsc; 4005 4006 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4007 4008 if (spa->spa_l2cache.sav_count == 0) 4009 return; 4010 4011 VERIFY(nvlist_lookup_nvlist(config, 4012 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 4013 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 4014 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4015 if (nl2cache != 0) { 4016 VERIFY(nvlist_add_nvlist_array(nvroot, 4017 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4018 VERIFY(nvlist_lookup_nvlist_array(nvroot, 4019 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 4020 4021 /* 4022 * Update level 2 cache device stats. 4023 */ 4024 4025 for (i = 0; i < nl2cache; i++) { 4026 VERIFY(nvlist_lookup_uint64(l2cache[i], 4027 ZPOOL_CONFIG_GUID, &guid) == 0); 4028 4029 vd = NULL; 4030 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 4031 if (guid == 4032 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 4033 vd = spa->spa_l2cache.sav_vdevs[j]; 4034 break; 4035 } 4036 } 4037 ASSERT(vd != NULL); 4038 4039 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 4040 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 4041 == 0); 4042 vdev_get_stats(vd, vs); 4043 } 4044 } 4045} 4046 4047static void 4048spa_add_feature_stats(spa_t *spa, nvlist_t *config) 4049{ 4050 nvlist_t *features; 4051 zap_cursor_t zc; 4052 zap_attribute_t za; 4053 4054 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 4055 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4056 4057 /* We may be unable to read features if pool is suspended. */ 4058 if (spa_suspended(spa)) 4059 goto out; 4060 4061 if (spa->spa_feat_for_read_obj != 0) { 4062 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4063 spa->spa_feat_for_read_obj); 4064 zap_cursor_retrieve(&zc, &za) == 0; 4065 zap_cursor_advance(&zc)) { 4066 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4067 za.za_num_integers == 1); 4068 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 4069 za.za_first_integer)); 4070 } 4071 zap_cursor_fini(&zc); 4072 } 4073 4074 if (spa->spa_feat_for_write_obj != 0) { 4075 for (zap_cursor_init(&zc, spa->spa_meta_objset, 4076 spa->spa_feat_for_write_obj); 4077 zap_cursor_retrieve(&zc, &za) == 0; 4078 zap_cursor_advance(&zc)) { 4079 ASSERT(za.za_integer_length == sizeof (uint64_t) && 4080 za.za_num_integers == 1); 4081 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 4082 za.za_first_integer)); 4083 } 4084 zap_cursor_fini(&zc); 4085 } 4086 4087out: 4088 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 4089 features) == 0); 4090 nvlist_free(features); 4091} 4092 4093int 4094spa_get_stats(const char *name, nvlist_t **config, 4095 char *altroot, size_t buflen) 4096{ 4097 int error; 4098 spa_t *spa; 4099 4100 *config = NULL; 4101 error = spa_open_common(name, &spa, FTAG, NULL, config); 4102 4103 if (spa != NULL) { 4104 /* 4105 * This still leaves a window of inconsistency where the spares 4106 * or l2cache devices could change and the config would be 4107 * self-inconsistent. 4108 */ 4109 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4110 4111 if (*config != NULL) { 4112 uint64_t loadtimes[2]; 4113 4114 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 4115 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 4116 VERIFY(nvlist_add_uint64_array(*config, 4117 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 4118 4119 VERIFY(nvlist_add_uint64(*config, 4120 ZPOOL_CONFIG_ERRCOUNT, 4121 spa_get_errlog_size(spa)) == 0); 4122 4123 if (spa_suspended(spa)) 4124 VERIFY(nvlist_add_uint64(*config, 4125 ZPOOL_CONFIG_SUSPENDED, 4126 spa->spa_failmode) == 0); 4127 4128 spa_add_spares(spa, *config); 4129 spa_add_l2cache(spa, *config); 4130 spa_add_feature_stats(spa, *config); 4131 } 4132 } 4133 4134 /* 4135 * We want to get the alternate root even for faulted pools, so we cheat 4136 * and call spa_lookup() directly. 4137 */ 4138 if (altroot) { 4139 if (spa == NULL) { 4140 mutex_enter(&spa_namespace_lock); 4141 spa = spa_lookup(name); 4142 if (spa) 4143 spa_altroot(spa, altroot, buflen); 4144 else 4145 altroot[0] = '\0'; 4146 spa = NULL; 4147 mutex_exit(&spa_namespace_lock); 4148 } else { 4149 spa_altroot(spa, altroot, buflen); 4150 } 4151 } 4152 4153 if (spa != NULL) { 4154 spa_config_exit(spa, SCL_CONFIG, FTAG); 4155 spa_close(spa, FTAG); 4156 } 4157 4158 return (error); 4159} 4160 4161/* 4162 * Validate that the auxiliary device array is well formed. We must have an 4163 * array of nvlists, each which describes a valid leaf vdev. If this is an 4164 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 4165 * specified, as long as they are well-formed. 4166 */ 4167static int 4168spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 4169 spa_aux_vdev_t *sav, const char *config, uint64_t version, 4170 vdev_labeltype_t label) 4171{ 4172 nvlist_t **dev; 4173 uint_t i, ndev; 4174 vdev_t *vd; 4175 int error; 4176 4177 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4178 4179 /* 4180 * It's acceptable to have no devs specified. 4181 */ 4182 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 4183 return (0); 4184 4185 if (ndev == 0) 4186 return (SET_ERROR(EINVAL)); 4187 4188 /* 4189 * Make sure the pool is formatted with a version that supports this 4190 * device type. 4191 */ 4192 if (spa_version(spa) < version) 4193 return (SET_ERROR(ENOTSUP)); 4194 4195 /* 4196 * Set the pending device list so we correctly handle device in-use 4197 * checking. 4198 */ 4199 sav->sav_pending = dev; 4200 sav->sav_npending = ndev; 4201 4202 for (i = 0; i < ndev; i++) { 4203 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 4204 mode)) != 0) 4205 goto out; 4206 4207 if (!vd->vdev_ops->vdev_op_leaf) { 4208 vdev_free(vd); 4209 error = SET_ERROR(EINVAL); 4210 goto out; 4211 } 4212 4213 /* 4214 * The L2ARC currently only supports disk devices in 4215 * kernel context. For user-level testing, we allow it. 4216 */ 4217#ifdef _KERNEL 4218 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 4219 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 4220 error = SET_ERROR(ENOTBLK); 4221 vdev_free(vd); 4222 goto out; 4223 } 4224#endif 4225 vd->vdev_top = vd; 4226 4227 if ((error = vdev_open(vd)) == 0 && 4228 (error = vdev_label_init(vd, crtxg, label)) == 0) { 4229 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 4230 vd->vdev_guid) == 0); 4231 } 4232 4233 vdev_free(vd); 4234 4235 if (error && 4236 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 4237 goto out; 4238 else 4239 error = 0; 4240 } 4241 4242out: 4243 sav->sav_pending = NULL; 4244 sav->sav_npending = 0; 4245 return (error); 4246} 4247 4248static int 4249spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 4250{ 4251 int error; 4252 4253 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 4254 4255 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4256 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 4257 VDEV_LABEL_SPARE)) != 0) { 4258 return (error); 4259 } 4260 4261 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 4262 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 4263 VDEV_LABEL_L2CACHE)); 4264} 4265 4266static void 4267spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 4268 const char *config) 4269{ 4270 int i; 4271 4272 if (sav->sav_config != NULL) { 4273 nvlist_t **olddevs; 4274 uint_t oldndevs; 4275 nvlist_t **newdevs; 4276 4277 /* 4278 * Generate new dev list by concatentating with the 4279 * current dev list. 4280 */ 4281 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 4282 &olddevs, &oldndevs) == 0); 4283 4284 newdevs = kmem_alloc(sizeof (void *) * 4285 (ndevs + oldndevs), KM_SLEEP); 4286 for (i = 0; i < oldndevs; i++) 4287 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 4288 KM_SLEEP) == 0); 4289 for (i = 0; i < ndevs; i++) 4290 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 4291 KM_SLEEP) == 0); 4292 4293 VERIFY(nvlist_remove(sav->sav_config, config, 4294 DATA_TYPE_NVLIST_ARRAY) == 0); 4295 4296 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 4297 config, newdevs, ndevs + oldndevs) == 0); 4298 for (i = 0; i < oldndevs + ndevs; i++) 4299 nvlist_free(newdevs[i]); 4300 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 4301 } else { 4302 /* 4303 * Generate a new dev list. 4304 */ 4305 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 4306 KM_SLEEP) == 0); 4307 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 4308 devs, ndevs) == 0); 4309 } 4310} 4311 4312/* 4313 * Stop and drop level 2 ARC devices 4314 */ 4315void 4316spa_l2cache_drop(spa_t *spa) 4317{ 4318 vdev_t *vd; 4319 int i; 4320 spa_aux_vdev_t *sav = &spa->spa_l2cache; 4321 4322 for (i = 0; i < sav->sav_count; i++) { 4323 uint64_t pool; 4324 4325 vd = sav->sav_vdevs[i]; 4326 ASSERT(vd != NULL); 4327 4328 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 4329 pool != 0ULL && l2arc_vdev_present(vd)) 4330 l2arc_remove_vdev(vd); 4331 } 4332} 4333 4334/* 4335 * Pool Creation 4336 */ 4337int 4338spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 4339 nvlist_t *zplprops) 4340{ 4341 spa_t *spa; 4342 char *altroot = NULL; 4343 vdev_t *rvd; 4344 dsl_pool_t *dp; 4345 dmu_tx_t *tx; 4346 int error = 0; 4347 uint64_t txg = TXG_INITIAL; 4348 nvlist_t **spares, **l2cache; 4349 uint_t nspares, nl2cache; 4350 uint64_t version, obj; 4351 boolean_t has_features; 4352 4353 /* 4354 * If this pool already exists, return failure. 4355 */ 4356 mutex_enter(&spa_namespace_lock); 4357 if (spa_lookup(pool) != NULL) { 4358 mutex_exit(&spa_namespace_lock); 4359 return (SET_ERROR(EEXIST)); 4360 } 4361 4362 /* 4363 * Allocate a new spa_t structure. 4364 */ 4365 (void) nvlist_lookup_string(props, 4366 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4367 spa = spa_add(pool, NULL, altroot); 4368 spa_activate(spa, spa_mode_global); 4369 4370 if (props && (error = spa_prop_validate(spa, props))) { 4371 spa_deactivate(spa); 4372 spa_remove(spa); 4373 mutex_exit(&spa_namespace_lock); 4374 return (error); 4375 } 4376 4377 has_features = B_FALSE; 4378 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 4379 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 4380 if (zpool_prop_feature(nvpair_name(elem))) 4381 has_features = B_TRUE; 4382 } 4383 4384 if (has_features || nvlist_lookup_uint64(props, 4385 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 4386 version = SPA_VERSION; 4387 } 4388 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 4389 4390 spa->spa_first_txg = txg; 4391 spa->spa_uberblock.ub_txg = txg - 1; 4392 spa->spa_uberblock.ub_version = version; 4393 spa->spa_ubsync = spa->spa_uberblock; 4394 spa->spa_load_state = SPA_LOAD_CREATE; 4395 spa->spa_removing_phys.sr_state = DSS_NONE; 4396 spa->spa_removing_phys.sr_removing_vdev = -1; 4397 spa->spa_removing_phys.sr_prev_indirect_vdev = -1; 4398 4399 /* 4400 * Create "The Godfather" zio to hold all async IOs 4401 */ 4402 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 4403 KM_SLEEP); 4404 for (int i = 0; i < max_ncpus; i++) { 4405 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 4406 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 4407 ZIO_FLAG_GODFATHER); 4408 } 4409 4410 /* 4411 * Create the root vdev. 4412 */ 4413 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4414 4415 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 4416 4417 ASSERT(error != 0 || rvd != NULL); 4418 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 4419 4420 if (error == 0 && !zfs_allocatable_devs(nvroot)) 4421 error = SET_ERROR(EINVAL); 4422 4423 if (error == 0 && 4424 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 4425 (error = spa_validate_aux(spa, nvroot, txg, 4426 VDEV_ALLOC_ADD)) == 0) { 4427 for (int c = 0; c < rvd->vdev_children; c++) { 4428 vdev_ashift_optimize(rvd->vdev_child[c]); 4429 vdev_metaslab_set_size(rvd->vdev_child[c]); 4430 vdev_expand(rvd->vdev_child[c], txg); 4431 } 4432 } 4433 4434 spa_config_exit(spa, SCL_ALL, FTAG); 4435 4436 if (error != 0) { 4437 spa_unload(spa); 4438 spa_deactivate(spa); 4439 spa_remove(spa); 4440 mutex_exit(&spa_namespace_lock); 4441 return (error); 4442 } 4443 4444 /* 4445 * Get the list of spares, if specified. 4446 */ 4447 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4448 &spares, &nspares) == 0) { 4449 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 4450 KM_SLEEP) == 0); 4451 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4452 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4453 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4454 spa_load_spares(spa); 4455 spa_config_exit(spa, SCL_ALL, FTAG); 4456 spa->spa_spares.sav_sync = B_TRUE; 4457 } 4458 4459 /* 4460 * Get the list of level 2 cache devices, if specified. 4461 */ 4462 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4463 &l2cache, &nl2cache) == 0) { 4464 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4465 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4466 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4467 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4468 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4469 spa_load_l2cache(spa); 4470 spa_config_exit(spa, SCL_ALL, FTAG); 4471 spa->spa_l2cache.sav_sync = B_TRUE; 4472 } 4473 4474 spa->spa_is_initializing = B_TRUE; 4475 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 4476 spa->spa_meta_objset = dp->dp_meta_objset; 4477 spa->spa_is_initializing = B_FALSE; 4478 4479 /* 4480 * Create DDTs (dedup tables). 4481 */ 4482 ddt_create(spa); 4483 4484 spa_update_dspace(spa); 4485 4486 tx = dmu_tx_create_assigned(dp, txg); 4487 4488 /* 4489 * Create the pool config object. 4490 */ 4491 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 4492 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 4493 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 4494 4495 if (zap_add(spa->spa_meta_objset, 4496 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 4497 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 4498 cmn_err(CE_PANIC, "failed to add pool config"); 4499 } 4500 4501 if (spa_version(spa) >= SPA_VERSION_FEATURES) 4502 spa_feature_create_zap_objects(spa, tx); 4503 4504 if (zap_add(spa->spa_meta_objset, 4505 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 4506 sizeof (uint64_t), 1, &version, tx) != 0) { 4507 cmn_err(CE_PANIC, "failed to add pool version"); 4508 } 4509 4510 /* Newly created pools with the right version are always deflated. */ 4511 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 4512 spa->spa_deflate = TRUE; 4513 if (zap_add(spa->spa_meta_objset, 4514 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 4515 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 4516 cmn_err(CE_PANIC, "failed to add deflate"); 4517 } 4518 } 4519 4520 /* 4521 * Create the deferred-free bpobj. Turn off compression 4522 * because sync-to-convergence takes longer if the blocksize 4523 * keeps changing. 4524 */ 4525 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 4526 dmu_object_set_compress(spa->spa_meta_objset, obj, 4527 ZIO_COMPRESS_OFF, tx); 4528 if (zap_add(spa->spa_meta_objset, 4529 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 4530 sizeof (uint64_t), 1, &obj, tx) != 0) { 4531 cmn_err(CE_PANIC, "failed to add bpobj"); 4532 } 4533 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 4534 spa->spa_meta_objset, obj)); 4535 4536 /* 4537 * Create the pool's history object. 4538 */ 4539 if (version >= SPA_VERSION_ZPOOL_HISTORY) 4540 spa_history_create_obj(spa, tx); 4541 4542 /* 4543 * Generate some random noise for salted checksums to operate on. 4544 */ 4545 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 4546 sizeof (spa->spa_cksum_salt.zcs_bytes)); 4547 4548 /* 4549 * Set pool properties. 4550 */ 4551 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 4552 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 4553 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 4554 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 4555 4556 if (props != NULL) { 4557 spa_configfile_set(spa, props, B_FALSE); 4558 spa_sync_props(props, tx); 4559 } 4560 4561 dmu_tx_commit(tx); 4562 4563 spa->spa_sync_on = B_TRUE; 4564 txg_sync_start(spa->spa_dsl_pool); 4565 4566 /* 4567 * We explicitly wait for the first transaction to complete so that our 4568 * bean counters are appropriately updated. 4569 */ 4570 txg_wait_synced(spa->spa_dsl_pool, txg); 4571 4572 spa_write_cachefile(spa, B_FALSE, B_TRUE); 4573 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); 4574 4575 spa_history_log_version(spa, "create"); 4576 4577 /* 4578 * Don't count references from objsets that are already closed 4579 * and are making their way through the eviction process. 4580 */ 4581 spa_evicting_os_wait(spa); 4582 spa->spa_minref = refcount_count(&spa->spa_refcount); 4583 spa->spa_load_state = SPA_LOAD_NONE; 4584 4585 mutex_exit(&spa_namespace_lock); 4586 4587 return (0); 4588} 4589 4590#ifdef _KERNEL 4591#ifdef illumos 4592/* 4593 * Get the root pool information from the root disk, then import the root pool 4594 * during the system boot up time. 4595 */ 4596extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 4597 4598static nvlist_t * 4599spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 4600{ 4601 nvlist_t *config; 4602 nvlist_t *nvtop, *nvroot; 4603 uint64_t pgid; 4604 4605 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 4606 return (NULL); 4607 4608 /* 4609 * Add this top-level vdev to the child array. 4610 */ 4611 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4612 &nvtop) == 0); 4613 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4614 &pgid) == 0); 4615 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 4616 4617 /* 4618 * Put this pool's top-level vdevs into a root vdev. 4619 */ 4620 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4621 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4622 VDEV_TYPE_ROOT) == 0); 4623 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4624 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4625 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4626 &nvtop, 1) == 0); 4627 4628 /* 4629 * Replace the existing vdev_tree with the new root vdev in 4630 * this pool's configuration (remove the old, add the new). 4631 */ 4632 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4633 nvlist_free(nvroot); 4634 return (config); 4635} 4636 4637/* 4638 * Walk the vdev tree and see if we can find a device with "better" 4639 * configuration. A configuration is "better" if the label on that 4640 * device has a more recent txg. 4641 */ 4642static void 4643spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 4644{ 4645 for (int c = 0; c < vd->vdev_children; c++) 4646 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 4647 4648 if (vd->vdev_ops->vdev_op_leaf) { 4649 nvlist_t *label; 4650 uint64_t label_txg; 4651 4652 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 4653 &label) != 0) 4654 return; 4655 4656 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 4657 &label_txg) == 0); 4658 4659 /* 4660 * Do we have a better boot device? 4661 */ 4662 if (label_txg > *txg) { 4663 *txg = label_txg; 4664 *avd = vd; 4665 } 4666 nvlist_free(label); 4667 } 4668} 4669 4670/* 4671 * Import a root pool. 4672 * 4673 * For x86. devpath_list will consist of devid and/or physpath name of 4674 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 4675 * The GRUB "findroot" command will return the vdev we should boot. 4676 * 4677 * For Sparc, devpath_list consists the physpath name of the booting device 4678 * no matter the rootpool is a single device pool or a mirrored pool. 4679 * e.g. 4680 * "/pci@1f,0/ide@d/disk@0,0:a" 4681 */ 4682int 4683spa_import_rootpool(char *devpath, char *devid) 4684{ 4685 spa_t *spa; 4686 vdev_t *rvd, *bvd, *avd = NULL; 4687 nvlist_t *config, *nvtop; 4688 uint64_t guid, txg; 4689 char *pname; 4690 int error; 4691 4692 /* 4693 * Read the label from the boot device and generate a configuration. 4694 */ 4695 config = spa_generate_rootconf(devpath, devid, &guid); 4696#if defined(_OBP) && defined(_KERNEL) 4697 if (config == NULL) { 4698 if (strstr(devpath, "/iscsi/ssd") != NULL) { 4699 /* iscsi boot */ 4700 get_iscsi_bootpath_phy(devpath); 4701 config = spa_generate_rootconf(devpath, devid, &guid); 4702 } 4703 } 4704#endif 4705 if (config == NULL) { 4706 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 4707 devpath); 4708 return (SET_ERROR(EIO)); 4709 } 4710 4711 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4712 &pname) == 0); 4713 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 4714 4715 mutex_enter(&spa_namespace_lock); 4716 if ((spa = spa_lookup(pname)) != NULL) { 4717 /* 4718 * Remove the existing root pool from the namespace so that we 4719 * can replace it with the correct config we just read in. 4720 */ 4721 spa_remove(spa); 4722 } 4723 4724 spa = spa_add(pname, config, NULL); 4725 spa->spa_is_root = B_TRUE; 4726 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4727 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4728 &spa->spa_ubsync.ub_version) != 0) 4729 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4730 4731 /* 4732 * Build up a vdev tree based on the boot device's label config. 4733 */ 4734 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4735 &nvtop) == 0); 4736 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4737 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4738 VDEV_ALLOC_ROOTPOOL); 4739 spa_config_exit(spa, SCL_ALL, FTAG); 4740 if (error) { 4741 mutex_exit(&spa_namespace_lock); 4742 nvlist_free(config); 4743 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4744 pname); 4745 return (error); 4746 } 4747 4748 /* 4749 * Get the boot vdev. 4750 */ 4751 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 4752 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 4753 (u_longlong_t)guid); 4754 error = SET_ERROR(ENOENT); 4755 goto out; 4756 } 4757 4758 /* 4759 * Determine if there is a better boot device. 4760 */ 4761 avd = bvd; 4762 spa_alt_rootvdev(rvd, &avd, &txg); 4763 if (avd != bvd) { 4764 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 4765 "try booting from '%s'", avd->vdev_path); 4766 error = SET_ERROR(EINVAL); 4767 goto out; 4768 } 4769 4770 /* 4771 * If the boot device is part of a spare vdev then ensure that 4772 * we're booting off the active spare. 4773 */ 4774 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 4775 !bvd->vdev_isspare) { 4776 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 4777 "try booting from '%s'", 4778 bvd->vdev_parent-> 4779 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4780 error = SET_ERROR(EINVAL); 4781 goto out; 4782 } 4783 4784 error = 0; 4785out: 4786 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4787 vdev_free(rvd); 4788 spa_config_exit(spa, SCL_ALL, FTAG); 4789 mutex_exit(&spa_namespace_lock); 4790 4791 nvlist_free(config); 4792 return (error); 4793} 4794 4795#else /* !illumos */ 4796 4797extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 4798 uint64_t *count); 4799 4800static nvlist_t * 4801spa_generate_rootconf(const char *name) 4802{ 4803 nvlist_t **configs, **tops; 4804 nvlist_t *config; 4805 nvlist_t *best_cfg, *nvtop, *nvroot; 4806 uint64_t *holes; 4807 uint64_t best_txg; 4808 uint64_t nchildren; 4809 uint64_t pgid; 4810 uint64_t count; 4811 uint64_t i; 4812 uint_t nholes; 4813 4814 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4815 return (NULL); 4816 4817 ASSERT3U(count, !=, 0); 4818 best_txg = 0; 4819 for (i = 0; i < count; i++) { 4820 uint64_t txg; 4821 4822 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4823 &txg) == 0); 4824 if (txg > best_txg) { 4825 best_txg = txg; 4826 best_cfg = configs[i]; 4827 } 4828 } 4829 4830 nchildren = 1; 4831 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4832 holes = NULL; 4833 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4834 &holes, &nholes); 4835 4836 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4837 for (i = 0; i < nchildren; i++) { 4838 if (i >= count) 4839 break; 4840 if (configs[i] == NULL) 4841 continue; 4842 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4843 &nvtop) == 0); 4844 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4845 } 4846 for (i = 0; holes != NULL && i < nholes; i++) { 4847 if (i >= nchildren) 4848 continue; 4849 if (tops[holes[i]] != NULL) 4850 continue; 4851 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4852 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4853 VDEV_TYPE_HOLE) == 0); 4854 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4855 holes[i]) == 0); 4856 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4857 0) == 0); 4858 } 4859 for (i = 0; i < nchildren; i++) { 4860 if (tops[i] != NULL) 4861 continue; 4862 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4863 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4864 VDEV_TYPE_MISSING) == 0); 4865 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4866 i) == 0); 4867 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4868 0) == 0); 4869 } 4870 4871 /* 4872 * Create pool config based on the best vdev config. 4873 */ 4874 nvlist_dup(best_cfg, &config, KM_SLEEP); 4875 4876 /* 4877 * Put this pool's top-level vdevs into a root vdev. 4878 */ 4879 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4880 &pgid) == 0); 4881 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4882 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4883 VDEV_TYPE_ROOT) == 0); 4884 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4885 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4886 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4887 tops, nchildren) == 0); 4888 4889 /* 4890 * Replace the existing vdev_tree with the new root vdev in 4891 * this pool's configuration (remove the old, add the new). 4892 */ 4893 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4894 4895 /* 4896 * Drop vdev config elements that should not be present at pool level. 4897 */ 4898 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4899 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4900 4901 for (i = 0; i < count; i++) 4902 nvlist_free(configs[i]); 4903 kmem_free(configs, count * sizeof(void *)); 4904 for (i = 0; i < nchildren; i++) 4905 nvlist_free(tops[i]); 4906 kmem_free(tops, nchildren * sizeof(void *)); 4907 nvlist_free(nvroot); 4908 return (config); 4909} 4910 4911int 4912spa_import_rootpool(const char *name) 4913{ 4914 spa_t *spa; 4915 vdev_t *rvd, *bvd, *avd = NULL; 4916 nvlist_t *config, *nvtop; 4917 uint64_t txg; 4918 char *pname; 4919 int error; 4920 4921 /* 4922 * Read the label from the boot device and generate a configuration. 4923 */ 4924 config = spa_generate_rootconf(name); 4925 4926 mutex_enter(&spa_namespace_lock); 4927 if (config != NULL) { 4928 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4929 &pname) == 0 && strcmp(name, pname) == 0); 4930 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4931 == 0); 4932 4933 if ((spa = spa_lookup(pname)) != NULL) { 4934 /* 4935 * The pool could already be imported, 4936 * e.g., after reboot -r. 4937 */ 4938 if (spa->spa_state == POOL_STATE_ACTIVE) { 4939 mutex_exit(&spa_namespace_lock); 4940 nvlist_free(config); 4941 return (0); 4942 } 4943 4944 /* 4945 * Remove the existing root pool from the namespace so 4946 * that we can replace it with the correct config 4947 * we just read in. 4948 */ 4949 spa_remove(spa); 4950 } 4951 spa = spa_add(pname, config, NULL); 4952 4953 /* 4954 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4955 * via spa_version(). 4956 */ 4957 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4958 &spa->spa_ubsync.ub_version) != 0) 4959 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4960 } else if ((spa = spa_lookup(name)) == NULL) { 4961 mutex_exit(&spa_namespace_lock); 4962 nvlist_free(config); 4963 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4964 name); 4965 return (EIO); 4966 } else { 4967 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4968 } 4969 spa->spa_is_root = B_TRUE; 4970 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4971 4972 /* 4973 * Build up a vdev tree based on the boot device's label config. 4974 */ 4975 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4976 &nvtop) == 0); 4977 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4978 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4979 VDEV_ALLOC_ROOTPOOL); 4980 spa_config_exit(spa, SCL_ALL, FTAG); 4981 if (error) { 4982 mutex_exit(&spa_namespace_lock); 4983 nvlist_free(config); 4984 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4985 pname); 4986 return (error); 4987 } 4988 4989 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4990 vdev_free(rvd); 4991 spa_config_exit(spa, SCL_ALL, FTAG); 4992 mutex_exit(&spa_namespace_lock); 4993 4994 nvlist_free(config); 4995 return (0); 4996} 4997 4998#endif /* illumos */ 4999#endif /* _KERNEL */ 5000 5001/* 5002 * Import a non-root pool into the system. 5003 */ 5004int 5005spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 5006{ 5007 spa_t *spa; 5008 char *altroot = NULL; 5009 spa_load_state_t state = SPA_LOAD_IMPORT; 5010 zpool_rewind_policy_t policy; 5011 uint64_t mode = spa_mode_global; 5012 uint64_t readonly = B_FALSE; 5013 int error; 5014 nvlist_t *nvroot; 5015 nvlist_t **spares, **l2cache; 5016 uint_t nspares, nl2cache; 5017 5018 /* 5019 * If a pool with this name exists, return failure. 5020 */ 5021 mutex_enter(&spa_namespace_lock); 5022 if (spa_lookup(pool) != NULL) { 5023 mutex_exit(&spa_namespace_lock); 5024 return (SET_ERROR(EEXIST)); 5025 } 5026 5027 /* 5028 * Create and initialize the spa structure. 5029 */ 5030 (void) nvlist_lookup_string(props, 5031 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5032 (void) nvlist_lookup_uint64(props, 5033 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 5034 if (readonly) 5035 mode = FREAD; 5036 spa = spa_add(pool, config, altroot); 5037 spa->spa_import_flags = flags; 5038 5039 /* 5040 * Verbatim import - Take a pool and insert it into the namespace 5041 * as if it had been loaded at boot. 5042 */ 5043 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 5044 if (props != NULL) 5045 spa_configfile_set(spa, props, B_FALSE); 5046 5047 spa_write_cachefile(spa, B_FALSE, B_TRUE); 5048 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5049 zfs_dbgmsg("spa_import: verbatim import of %s", pool); 5050 mutex_exit(&spa_namespace_lock); 5051 return (0); 5052 } 5053 5054 spa_activate(spa, mode); 5055 5056 /* 5057 * Don't start async tasks until we know everything is healthy. 5058 */ 5059 spa_async_suspend(spa); 5060 5061 zpool_get_rewind_policy(config, &policy); 5062 if (policy.zrp_request & ZPOOL_DO_REWIND) 5063 state = SPA_LOAD_RECOVER; 5064 5065 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT; 5066 5067 if (state != SPA_LOAD_RECOVER) { 5068 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 5069 zfs_dbgmsg("spa_import: importing %s", pool); 5070 } else { 5071 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld " 5072 "(RECOVERY MODE)", pool, (longlong_t)policy.zrp_txg); 5073 } 5074 error = spa_load_best(spa, state, policy.zrp_txg, policy.zrp_request); 5075 5076 /* 5077 * Propagate anything learned while loading the pool and pass it 5078 * back to caller (i.e. rewind info, missing devices, etc). 5079 */ 5080 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5081 spa->spa_load_info) == 0); 5082 5083 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5084 /* 5085 * Toss any existing sparelist, as it doesn't have any validity 5086 * anymore, and conflicts with spa_has_spare(). 5087 */ 5088 if (spa->spa_spares.sav_config) { 5089 nvlist_free(spa->spa_spares.sav_config); 5090 spa->spa_spares.sav_config = NULL; 5091 spa_load_spares(spa); 5092 } 5093 if (spa->spa_l2cache.sav_config) { 5094 nvlist_free(spa->spa_l2cache.sav_config); 5095 spa->spa_l2cache.sav_config = NULL; 5096 spa_load_l2cache(spa); 5097 } 5098 5099 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 5100 &nvroot) == 0); 5101 if (error == 0) 5102 error = spa_validate_aux(spa, nvroot, -1ULL, 5103 VDEV_ALLOC_SPARE); 5104 if (error == 0) 5105 error = spa_validate_aux(spa, nvroot, -1ULL, 5106 VDEV_ALLOC_L2CACHE); 5107 spa_config_exit(spa, SCL_ALL, FTAG); 5108 5109 if (props != NULL) 5110 spa_configfile_set(spa, props, B_FALSE); 5111 5112 if (error != 0 || (props && spa_writeable(spa) && 5113 (error = spa_prop_set(spa, props)))) { 5114 spa_unload(spa); 5115 spa_deactivate(spa); 5116 spa_remove(spa); 5117 mutex_exit(&spa_namespace_lock); 5118 return (error); 5119 } 5120 5121 spa_async_resume(spa); 5122 5123 /* 5124 * Override any spares and level 2 cache devices as specified by 5125 * the user, as these may have correct device names/devids, etc. 5126 */ 5127 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 5128 &spares, &nspares) == 0) { 5129 if (spa->spa_spares.sav_config) 5130 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 5131 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 5132 else 5133 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 5134 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5135 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 5136 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 5137 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5138 spa_load_spares(spa); 5139 spa_config_exit(spa, SCL_ALL, FTAG); 5140 spa->spa_spares.sav_sync = B_TRUE; 5141 } 5142 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 5143 &l2cache, &nl2cache) == 0) { 5144 if (spa->spa_l2cache.sav_config) 5145 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 5146 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 5147 else 5148 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 5149 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5150 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 5151 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 5152 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5153 spa_load_l2cache(spa); 5154 spa_config_exit(spa, SCL_ALL, FTAG); 5155 spa->spa_l2cache.sav_sync = B_TRUE; 5156 } 5157 5158 /* 5159 * Check for any removed devices. 5160 */ 5161 if (spa->spa_autoreplace) { 5162 spa_aux_check_removed(&spa->spa_spares); 5163 spa_aux_check_removed(&spa->spa_l2cache); 5164 } 5165 5166 if (spa_writeable(spa)) { 5167 /* 5168 * Update the config cache to include the newly-imported pool. 5169 */ 5170 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5171 } 5172 5173 /* 5174 * It's possible that the pool was expanded while it was exported. 5175 * We kick off an async task to handle this for us. 5176 */ 5177 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 5178 5179 spa_history_log_version(spa, "import"); 5180 5181 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 5182 5183 mutex_exit(&spa_namespace_lock); 5184 5185#ifdef __FreeBSD__ 5186#ifdef _KERNEL 5187 zvol_create_minors(pool); 5188#endif 5189#endif 5190 return (0); 5191} 5192 5193nvlist_t * 5194spa_tryimport(nvlist_t *tryconfig) 5195{ 5196 nvlist_t *config = NULL; 5197 char *poolname, *cachefile; 5198 spa_t *spa; 5199 uint64_t state; 5200 int error; 5201 zpool_rewind_policy_t policy; 5202 5203 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 5204 return (NULL); 5205 5206 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 5207 return (NULL); 5208 5209 /* 5210 * Create and initialize the spa structure. 5211 */ 5212 mutex_enter(&spa_namespace_lock); 5213 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 5214 spa_activate(spa, FREAD); 5215 5216 /* 5217 * Rewind pool if a max txg was provided. Note that even though we 5218 * retrieve the complete rewind policy, only the rewind txg is relevant 5219 * for tryimport. 5220 */ 5221 zpool_get_rewind_policy(spa->spa_config, &policy); 5222 if (policy.zrp_txg != UINT64_MAX) { 5223 spa->spa_load_max_txg = policy.zrp_txg; 5224 spa->spa_extreme_rewind = B_TRUE; 5225 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld", 5226 poolname, (longlong_t)policy.zrp_txg); 5227 } else { 5228 zfs_dbgmsg("spa_tryimport: importing %s", poolname); 5229 } 5230 5231 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile) 5232 == 0) { 5233 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile); 5234 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE; 5235 } else { 5236 spa->spa_config_source = SPA_CONFIG_SRC_SCAN; 5237 } 5238 5239 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING); 5240 5241 /* 5242 * If 'tryconfig' was at least parsable, return the current config. 5243 */ 5244 if (spa->spa_root_vdev != NULL) { 5245 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 5246 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 5247 poolname) == 0); 5248 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5249 state) == 0); 5250 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 5251 spa->spa_uberblock.ub_timestamp) == 0); 5252 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 5253 spa->spa_load_info) == 0); 5254 5255 /* 5256 * If the bootfs property exists on this pool then we 5257 * copy it out so that external consumers can tell which 5258 * pools are bootable. 5259 */ 5260 if ((!error || error == EEXIST) && spa->spa_bootfs) { 5261 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5262 5263 /* 5264 * We have to play games with the name since the 5265 * pool was opened as TRYIMPORT_NAME. 5266 */ 5267 if (dsl_dsobj_to_dsname(spa_name(spa), 5268 spa->spa_bootfs, tmpname) == 0) { 5269 char *cp; 5270 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 5271 5272 cp = strchr(tmpname, '/'); 5273 if (cp == NULL) { 5274 (void) strlcpy(dsname, tmpname, 5275 MAXPATHLEN); 5276 } else { 5277 (void) snprintf(dsname, MAXPATHLEN, 5278 "%s/%s", poolname, ++cp); 5279 } 5280 VERIFY(nvlist_add_string(config, 5281 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 5282 kmem_free(dsname, MAXPATHLEN); 5283 } 5284 kmem_free(tmpname, MAXPATHLEN); 5285 } 5286 5287 /* 5288 * Add the list of hot spares and level 2 cache devices. 5289 */ 5290 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5291 spa_add_spares(spa, config); 5292 spa_add_l2cache(spa, config); 5293 spa_config_exit(spa, SCL_CONFIG, FTAG); 5294 } 5295 5296 spa_unload(spa); 5297 spa_deactivate(spa); 5298 spa_remove(spa); 5299 mutex_exit(&spa_namespace_lock); 5300 5301 return (config); 5302} 5303 5304/* 5305 * Pool export/destroy 5306 * 5307 * The act of destroying or exporting a pool is very simple. We make sure there 5308 * is no more pending I/O and any references to the pool are gone. Then, we 5309 * update the pool state and sync all the labels to disk, removing the 5310 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 5311 * we don't sync the labels or remove the configuration cache. 5312 */ 5313static int 5314spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 5315 boolean_t force, boolean_t hardforce) 5316{ 5317 spa_t *spa; 5318 5319 if (oldconfig) 5320 *oldconfig = NULL; 5321 5322 if (!(spa_mode_global & FWRITE)) 5323 return (SET_ERROR(EROFS)); 5324 5325 mutex_enter(&spa_namespace_lock); 5326 if ((spa = spa_lookup(pool)) == NULL) { 5327 mutex_exit(&spa_namespace_lock); 5328 return (SET_ERROR(ENOENT)); 5329 } 5330 5331 /* 5332 * Put a hold on the pool, drop the namespace lock, stop async tasks, 5333 * reacquire the namespace lock, and see if we can export. 5334 */ 5335 spa_open_ref(spa, FTAG); 5336 mutex_exit(&spa_namespace_lock); 5337 spa_async_suspend(spa); 5338 mutex_enter(&spa_namespace_lock); 5339 spa_close(spa, FTAG); 5340 5341 /* 5342 * The pool will be in core if it's openable, 5343 * in which case we can modify its state. 5344 */ 5345 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 5346 /* 5347 * Objsets may be open only because they're dirty, so we 5348 * have to force it to sync before checking spa_refcnt. 5349 */ 5350 txg_wait_synced(spa->spa_dsl_pool, 0); 5351 spa_evicting_os_wait(spa); 5352 5353 /* 5354 * A pool cannot be exported or destroyed if there are active 5355 * references. If we are resetting a pool, allow references by 5356 * fault injection handlers. 5357 */ 5358 if (!spa_refcount_zero(spa) || 5359 (spa->spa_inject_ref != 0 && 5360 new_state != POOL_STATE_UNINITIALIZED)) { 5361 spa_async_resume(spa); 5362 mutex_exit(&spa_namespace_lock); 5363 return (SET_ERROR(EBUSY)); 5364 } 5365 5366 /* 5367 * A pool cannot be exported if it has an active shared spare. 5368 * This is to prevent other pools stealing the active spare 5369 * from an exported pool. At user's own will, such pool can 5370 * be forcedly exported. 5371 */ 5372 if (!force && new_state == POOL_STATE_EXPORTED && 5373 spa_has_active_shared_spare(spa)) { 5374 spa_async_resume(spa); 5375 mutex_exit(&spa_namespace_lock); 5376 return (SET_ERROR(EXDEV)); 5377 } 5378 5379 /* 5380 * We want this to be reflected on every label, 5381 * so mark them all dirty. spa_unload() will do the 5382 * final sync that pushes these changes out. 5383 */ 5384 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 5385 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5386 spa->spa_state = new_state; 5387 spa->spa_final_txg = spa_last_synced_txg(spa) + 5388 TXG_DEFER_SIZE + 1; 5389 vdev_config_dirty(spa->spa_root_vdev); 5390 spa_config_exit(spa, SCL_ALL, FTAG); 5391 } 5392 } 5393 5394 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); 5395 5396 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 5397 spa_unload(spa); 5398 spa_deactivate(spa); 5399 } 5400 5401 if (oldconfig && spa->spa_config) 5402 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 5403 5404 if (new_state != POOL_STATE_UNINITIALIZED) { 5405 if (!hardforce) 5406 spa_write_cachefile(spa, B_TRUE, B_TRUE); 5407 spa_remove(spa); 5408 } 5409 mutex_exit(&spa_namespace_lock); 5410 5411 return (0); 5412} 5413 5414/* 5415 * Destroy a storage pool. 5416 */ 5417int 5418spa_destroy(char *pool) 5419{ 5420 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 5421 B_FALSE, B_FALSE)); 5422} 5423 5424/* 5425 * Export a storage pool. 5426 */ 5427int 5428spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 5429 boolean_t hardforce) 5430{ 5431 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 5432 force, hardforce)); 5433} 5434 5435/* 5436 * Similar to spa_export(), this unloads the spa_t without actually removing it 5437 * from the namespace in any way. 5438 */ 5439int 5440spa_reset(char *pool) 5441{ 5442 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 5443 B_FALSE, B_FALSE)); 5444} 5445 5446/* 5447 * ========================================================================== 5448 * Device manipulation 5449 * ========================================================================== 5450 */ 5451 5452/* 5453 * Add a device to a storage pool. 5454 */ 5455int 5456spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 5457{ 5458 uint64_t txg, id; 5459 int error; 5460 vdev_t *rvd = spa->spa_root_vdev; 5461 vdev_t *vd, *tvd; 5462 nvlist_t **spares, **l2cache; 5463 uint_t nspares, nl2cache; 5464 5465 ASSERT(spa_writeable(spa)); 5466 5467 txg = spa_vdev_enter(spa); 5468 5469 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 5470 VDEV_ALLOC_ADD)) != 0) 5471 return (spa_vdev_exit(spa, NULL, txg, error)); 5472 5473 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 5474 5475 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 5476 &nspares) != 0) 5477 nspares = 0; 5478 5479 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 5480 &nl2cache) != 0) 5481 nl2cache = 0; 5482 5483 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 5484 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5485 5486 if (vd->vdev_children != 0 && 5487 (error = vdev_create(vd, txg, B_FALSE)) != 0) 5488 return (spa_vdev_exit(spa, vd, txg, error)); 5489 5490 /* 5491 * We must validate the spares and l2cache devices after checking the 5492 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 5493 */ 5494 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 5495 return (spa_vdev_exit(spa, vd, txg, error)); 5496 5497 /* 5498 * If we are in the middle of a device removal, we can only add 5499 * devices which match the existing devices in the pool. 5500 * If we are in the middle of a removal, or have some indirect 5501 * vdevs, we can not add raidz toplevels. 5502 */ 5503 if (spa->spa_vdev_removal != NULL || 5504 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5505 for (int c = 0; c < vd->vdev_children; c++) { 5506 tvd = vd->vdev_child[c]; 5507 if (spa->spa_vdev_removal != NULL && 5508 tvd->vdev_ashift != 5509 spa->spa_vdev_removal->svr_vdev->vdev_ashift) { 5510 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5511 } 5512 /* Fail if top level vdev is raidz */ 5513 if (tvd->vdev_ops == &vdev_raidz_ops) { 5514 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 5515 } 5516 /* 5517 * Need the top level mirror to be 5518 * a mirror of leaf vdevs only 5519 */ 5520 if (tvd->vdev_ops == &vdev_mirror_ops) { 5521 for (uint64_t cid = 0; 5522 cid < tvd->vdev_children; cid++) { 5523 vdev_t *cvd = tvd->vdev_child[cid]; 5524 if (!cvd->vdev_ops->vdev_op_leaf) { 5525 return (spa_vdev_exit(spa, vd, 5526 txg, EINVAL)); 5527 } 5528 } 5529 } 5530 } 5531 } 5532 5533 for (int c = 0; c < vd->vdev_children; c++) { 5534 5535 /* 5536 * Set the vdev id to the first hole, if one exists. 5537 */ 5538 for (id = 0; id < rvd->vdev_children; id++) { 5539 if (rvd->vdev_child[id]->vdev_ishole) { 5540 vdev_free(rvd->vdev_child[id]); 5541 break; 5542 } 5543 } 5544 tvd = vd->vdev_child[c]; 5545 vdev_remove_child(vd, tvd); 5546 tvd->vdev_id = id; 5547 vdev_add_child(rvd, tvd); 5548 vdev_config_dirty(tvd); 5549 } 5550 5551 if (nspares != 0) { 5552 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 5553 ZPOOL_CONFIG_SPARES); 5554 spa_load_spares(spa); 5555 spa->spa_spares.sav_sync = B_TRUE; 5556 } 5557 5558 if (nl2cache != 0) { 5559 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 5560 ZPOOL_CONFIG_L2CACHE); 5561 spa_load_l2cache(spa); 5562 spa->spa_l2cache.sav_sync = B_TRUE; 5563 } 5564 5565 /* 5566 * We have to be careful when adding new vdevs to an existing pool. 5567 * If other threads start allocating from these vdevs before we 5568 * sync the config cache, and we lose power, then upon reboot we may 5569 * fail to open the pool because there are DVAs that the config cache 5570 * can't translate. Therefore, we first add the vdevs without 5571 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 5572 * and then let spa_config_update() initialize the new metaslabs. 5573 * 5574 * spa_load() checks for added-but-not-initialized vdevs, so that 5575 * if we lose power at any point in this sequence, the remaining 5576 * steps will be completed the next time we load the pool. 5577 */ 5578 (void) spa_vdev_exit(spa, vd, txg, 0); 5579 5580 mutex_enter(&spa_namespace_lock); 5581 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5582 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); 5583 mutex_exit(&spa_namespace_lock); 5584 5585 return (0); 5586} 5587 5588/* 5589 * Attach a device to a mirror. The arguments are the path to any device 5590 * in the mirror, and the nvroot for the new device. If the path specifies 5591 * a device that is not mirrored, we automatically insert the mirror vdev. 5592 * 5593 * If 'replacing' is specified, the new device is intended to replace the 5594 * existing device; in this case the two devices are made into their own 5595 * mirror using the 'replacing' vdev, which is functionally identical to 5596 * the mirror vdev (it actually reuses all the same ops) but has a few 5597 * extra rules: you can't attach to it after it's been created, and upon 5598 * completion of resilvering, the first disk (the one being replaced) 5599 * is automatically detached. 5600 */ 5601int 5602spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 5603{ 5604 uint64_t txg, dtl_max_txg; 5605 vdev_t *rvd = spa->spa_root_vdev; 5606 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 5607 vdev_ops_t *pvops; 5608 char *oldvdpath, *newvdpath; 5609 int newvd_isspare; 5610 int error; 5611 5612 ASSERT(spa_writeable(spa)); 5613 5614 txg = spa_vdev_enter(spa); 5615 5616 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 5617 5618 if (spa->spa_vdev_removal != NULL || 5619 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) { 5620 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5621 } 5622 5623 if (oldvd == NULL) 5624 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5625 5626 if (!oldvd->vdev_ops->vdev_op_leaf) 5627 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5628 5629 pvd = oldvd->vdev_parent; 5630 5631 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 5632 VDEV_ALLOC_ATTACH)) != 0) 5633 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5634 5635 if (newrootvd->vdev_children != 1) 5636 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 5637 5638 newvd = newrootvd->vdev_child[0]; 5639 5640 if (!newvd->vdev_ops->vdev_op_leaf) 5641 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 5642 5643 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 5644 return (spa_vdev_exit(spa, newrootvd, txg, error)); 5645 5646 /* 5647 * Spares can't replace logs 5648 */ 5649 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 5650 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5651 5652 if (!replacing) { 5653 /* 5654 * For attach, the only allowable parent is a mirror or the root 5655 * vdev. 5656 */ 5657 if (pvd->vdev_ops != &vdev_mirror_ops && 5658 pvd->vdev_ops != &vdev_root_ops) 5659 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5660 5661 pvops = &vdev_mirror_ops; 5662 } else { 5663 /* 5664 * Active hot spares can only be replaced by inactive hot 5665 * spares. 5666 */ 5667 if (pvd->vdev_ops == &vdev_spare_ops && 5668 oldvd->vdev_isspare && 5669 !spa_has_spare(spa, newvd->vdev_guid)) 5670 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5671 5672 /* 5673 * If the source is a hot spare, and the parent isn't already a 5674 * spare, then we want to create a new hot spare. Otherwise, we 5675 * want to create a replacing vdev. The user is not allowed to 5676 * attach to a spared vdev child unless the 'isspare' state is 5677 * the same (spare replaces spare, non-spare replaces 5678 * non-spare). 5679 */ 5680 if (pvd->vdev_ops == &vdev_replacing_ops && 5681 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 5682 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5683 } else if (pvd->vdev_ops == &vdev_spare_ops && 5684 newvd->vdev_isspare != oldvd->vdev_isspare) { 5685 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 5686 } 5687 5688 if (newvd->vdev_isspare) 5689 pvops = &vdev_spare_ops; 5690 else 5691 pvops = &vdev_replacing_ops; 5692 } 5693 5694 /* 5695 * Make sure the new device is big enough. 5696 */ 5697 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 5698 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 5699 5700 /* 5701 * The new device cannot have a higher alignment requirement 5702 * than the top-level vdev. 5703 */ 5704 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 5705 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 5706 5707 /* 5708 * If this is an in-place replacement, update oldvd's path and devid 5709 * to make it distinguishable from newvd, and unopenable from now on. 5710 */ 5711 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 5712 spa_strfree(oldvd->vdev_path); 5713 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 5714 KM_SLEEP); 5715 (void) sprintf(oldvd->vdev_path, "%s/%s", 5716 newvd->vdev_path, "old"); 5717 if (oldvd->vdev_devid != NULL) { 5718 spa_strfree(oldvd->vdev_devid); 5719 oldvd->vdev_devid = NULL; 5720 } 5721 } 5722 5723 /* mark the device being resilvered */ 5724 newvd->vdev_resilver_txg = txg; 5725 5726 /* 5727 * If the parent is not a mirror, or if we're replacing, insert the new 5728 * mirror/replacing/spare vdev above oldvd. 5729 */ 5730 if (pvd->vdev_ops != pvops) 5731 pvd = vdev_add_parent(oldvd, pvops); 5732 5733 ASSERT(pvd->vdev_top->vdev_parent == rvd); 5734 ASSERT(pvd->vdev_ops == pvops); 5735 ASSERT(oldvd->vdev_parent == pvd); 5736 5737 /* 5738 * Extract the new device from its root and add it to pvd. 5739 */ 5740 vdev_remove_child(newrootvd, newvd); 5741 newvd->vdev_id = pvd->vdev_children; 5742 newvd->vdev_crtxg = oldvd->vdev_crtxg; 5743 vdev_add_child(pvd, newvd); 5744 5745 tvd = newvd->vdev_top; 5746 ASSERT(pvd->vdev_top == tvd); 5747 ASSERT(tvd->vdev_parent == rvd); 5748 5749 vdev_config_dirty(tvd); 5750 5751 /* 5752 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 5753 * for any dmu_sync-ed blocks. It will propagate upward when 5754 * spa_vdev_exit() calls vdev_dtl_reassess(). 5755 */ 5756 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 5757 5758 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 5759 dtl_max_txg - TXG_INITIAL); 5760 5761 if (newvd->vdev_isspare) { 5762 spa_spare_activate(newvd); 5763 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); 5764 } 5765 5766 oldvdpath = spa_strdup(oldvd->vdev_path); 5767 newvdpath = spa_strdup(newvd->vdev_path); 5768 newvd_isspare = newvd->vdev_isspare; 5769 5770 /* 5771 * Mark newvd's DTL dirty in this txg. 5772 */ 5773 vdev_dirty(tvd, VDD_DTL, newvd, txg); 5774 5775 /* 5776 * Schedule the resilver to restart in the future. We do this to 5777 * ensure that dmu_sync-ed blocks have been stitched into the 5778 * respective datasets. 5779 */ 5780 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 5781 5782 if (spa->spa_bootfs) 5783 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); 5784 5785 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); 5786 5787 /* 5788 * Commit the config 5789 */ 5790 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 5791 5792 spa_history_log_internal(spa, "vdev attach", NULL, 5793 "%s vdev=%s %s vdev=%s", 5794 replacing && newvd_isspare ? "spare in" : 5795 replacing ? "replace" : "attach", newvdpath, 5796 replacing ? "for" : "to", oldvdpath); 5797 5798 spa_strfree(oldvdpath); 5799 spa_strfree(newvdpath); 5800 5801 return (0); 5802} 5803 5804/* 5805 * Detach a device from a mirror or replacing vdev. 5806 * 5807 * If 'replace_done' is specified, only detach if the parent 5808 * is a replacing vdev. 5809 */ 5810int 5811spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 5812{ 5813 uint64_t txg; 5814 int error; 5815 vdev_t *rvd = spa->spa_root_vdev; 5816 vdev_t *vd, *pvd, *cvd, *tvd; 5817 boolean_t unspare = B_FALSE; 5818 uint64_t unspare_guid = 0; 5819 char *vdpath; 5820 5821 ASSERT(spa_writeable(spa)); 5822 5823 txg = spa_vdev_enter(spa); 5824 5825 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5826 5827 if (vd == NULL) 5828 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5829 5830 if (!vd->vdev_ops->vdev_op_leaf) 5831 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5832 5833 pvd = vd->vdev_parent; 5834 5835 /* 5836 * If the parent/child relationship is not as expected, don't do it. 5837 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 5838 * vdev that's replacing B with C. The user's intent in replacing 5839 * is to go from M(A,B) to M(A,C). If the user decides to cancel 5840 * the replace by detaching C, the expected behavior is to end up 5841 * M(A,B). But suppose that right after deciding to detach C, 5842 * the replacement of B completes. We would have M(A,C), and then 5843 * ask to detach C, which would leave us with just A -- not what 5844 * the user wanted. To prevent this, we make sure that the 5845 * parent/child relationship hasn't changed -- in this example, 5846 * that C's parent is still the replacing vdev R. 5847 */ 5848 if (pvd->vdev_guid != pguid && pguid != 0) 5849 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5850 5851 /* 5852 * Only 'replacing' or 'spare' vdevs can be replaced. 5853 */ 5854 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5855 pvd->vdev_ops != &vdev_spare_ops) 5856 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5857 5858 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5859 spa_version(spa) >= SPA_VERSION_SPARES); 5860 5861 /* 5862 * Only mirror, replacing, and spare vdevs support detach. 5863 */ 5864 if (pvd->vdev_ops != &vdev_replacing_ops && 5865 pvd->vdev_ops != &vdev_mirror_ops && 5866 pvd->vdev_ops != &vdev_spare_ops) 5867 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5868 5869 /* 5870 * If this device has the only valid copy of some data, 5871 * we cannot safely detach it. 5872 */ 5873 if (vdev_dtl_required(vd)) 5874 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5875 5876 ASSERT(pvd->vdev_children >= 2); 5877 5878 /* 5879 * If we are detaching the second disk from a replacing vdev, then 5880 * check to see if we changed the original vdev's path to have "/old" 5881 * at the end in spa_vdev_attach(). If so, undo that change now. 5882 */ 5883 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5884 vd->vdev_path != NULL) { 5885 size_t len = strlen(vd->vdev_path); 5886 5887 for (int c = 0; c < pvd->vdev_children; c++) { 5888 cvd = pvd->vdev_child[c]; 5889 5890 if (cvd == vd || cvd->vdev_path == NULL) 5891 continue; 5892 5893 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5894 strcmp(cvd->vdev_path + len, "/old") == 0) { 5895 spa_strfree(cvd->vdev_path); 5896 cvd->vdev_path = spa_strdup(vd->vdev_path); 5897 break; 5898 } 5899 } 5900 } 5901 5902 /* 5903 * If we are detaching the original disk from a spare, then it implies 5904 * that the spare should become a real disk, and be removed from the 5905 * active spare list for the pool. 5906 */ 5907 if (pvd->vdev_ops == &vdev_spare_ops && 5908 vd->vdev_id == 0 && 5909 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5910 unspare = B_TRUE; 5911 5912 /* 5913 * Erase the disk labels so the disk can be used for other things. 5914 * This must be done after all other error cases are handled, 5915 * but before we disembowel vd (so we can still do I/O to it). 5916 * But if we can't do it, don't treat the error as fatal -- 5917 * it may be that the unwritability of the disk is the reason 5918 * it's being detached! 5919 */ 5920 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5921 5922 /* 5923 * Remove vd from its parent and compact the parent's children. 5924 */ 5925 vdev_remove_child(pvd, vd); 5926 vdev_compact_children(pvd); 5927 5928 /* 5929 * Remember one of the remaining children so we can get tvd below. 5930 */ 5931 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5932 5933 /* 5934 * If we need to remove the remaining child from the list of hot spares, 5935 * do it now, marking the vdev as no longer a spare in the process. 5936 * We must do this before vdev_remove_parent(), because that can 5937 * change the GUID if it creates a new toplevel GUID. For a similar 5938 * reason, we must remove the spare now, in the same txg as the detach; 5939 * otherwise someone could attach a new sibling, change the GUID, and 5940 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5941 */ 5942 if (unspare) { 5943 ASSERT(cvd->vdev_isspare); 5944 spa_spare_remove(cvd); 5945 unspare_guid = cvd->vdev_guid; 5946 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5947 cvd->vdev_unspare = B_TRUE; 5948 } 5949 5950 /* 5951 * If the parent mirror/replacing vdev only has one child, 5952 * the parent is no longer needed. Remove it from the tree. 5953 */ 5954 if (pvd->vdev_children == 1) { 5955 if (pvd->vdev_ops == &vdev_spare_ops) 5956 cvd->vdev_unspare = B_FALSE; 5957 vdev_remove_parent(cvd); 5958 } 5959 5960 5961 /* 5962 * We don't set tvd until now because the parent we just removed 5963 * may have been the previous top-level vdev. 5964 */ 5965 tvd = cvd->vdev_top; 5966 ASSERT(tvd->vdev_parent == rvd); 5967 5968 /* 5969 * Reevaluate the parent vdev state. 5970 */ 5971 vdev_propagate_state(cvd); 5972 5973 /* 5974 * If the 'autoexpand' property is set on the pool then automatically 5975 * try to expand the size of the pool. For example if the device we 5976 * just detached was smaller than the others, it may be possible to 5977 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5978 * first so that we can obtain the updated sizes of the leaf vdevs. 5979 */ 5980 if (spa->spa_autoexpand) { 5981 vdev_reopen(tvd); 5982 vdev_expand(tvd, txg); 5983 } 5984 5985 vdev_config_dirty(tvd); 5986 5987 /* 5988 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5989 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5990 * But first make sure we're not on any *other* txg's DTL list, to 5991 * prevent vd from being accessed after it's freed. 5992 */ 5993 vdpath = spa_strdup(vd->vdev_path); 5994 for (int t = 0; t < TXG_SIZE; t++) 5995 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5996 vd->vdev_detached = B_TRUE; 5997 vdev_dirty(tvd, VDD_DTL, vd, txg); 5998 5999 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); 6000 6001 /* hang on to the spa before we release the lock */ 6002 spa_open_ref(spa, FTAG); 6003 6004 error = spa_vdev_exit(spa, vd, txg, 0); 6005 6006 spa_history_log_internal(spa, "detach", NULL, 6007 "vdev=%s", vdpath); 6008 spa_strfree(vdpath); 6009 6010 /* 6011 * If this was the removal of the original device in a hot spare vdev, 6012 * then we want to go through and remove the device from the hot spare 6013 * list of every other pool. 6014 */ 6015 if (unspare) { 6016 spa_t *altspa = NULL; 6017 6018 mutex_enter(&spa_namespace_lock); 6019 while ((altspa = spa_next(altspa)) != NULL) { 6020 if (altspa->spa_state != POOL_STATE_ACTIVE || 6021 altspa == spa) 6022 continue; 6023 6024 spa_open_ref(altspa, FTAG); 6025 mutex_exit(&spa_namespace_lock); 6026 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 6027 mutex_enter(&spa_namespace_lock); 6028 spa_close(altspa, FTAG); 6029 } 6030 mutex_exit(&spa_namespace_lock); 6031 6032 /* search the rest of the vdevs for spares to remove */ 6033 spa_vdev_resilver_done(spa); 6034 } 6035 6036 /* all done with the spa; OK to release */ 6037 mutex_enter(&spa_namespace_lock); 6038 spa_close(spa, FTAG); 6039 mutex_exit(&spa_namespace_lock); 6040 6041 return (error); 6042} 6043 6044/* 6045 * Split a set of devices from their mirrors, and create a new pool from them. 6046 */ 6047int 6048spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 6049 nvlist_t *props, boolean_t exp) 6050{ 6051 int error = 0; 6052 uint64_t txg, *glist; 6053 spa_t *newspa; 6054 uint_t c, children, lastlog; 6055 nvlist_t **child, *nvl, *tmp; 6056 dmu_tx_t *tx; 6057 char *altroot = NULL; 6058 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 6059 boolean_t activate_slog; 6060 6061 ASSERT(spa_writeable(spa)); 6062 6063 txg = spa_vdev_enter(spa); 6064 6065 /* clear the log and flush everything up to now */ 6066 activate_slog = spa_passivate_log(spa); 6067 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6068 error = spa_reset_logs(spa); 6069 txg = spa_vdev_config_enter(spa); 6070 6071 if (activate_slog) 6072 spa_activate_log(spa); 6073 6074 if (error != 0) 6075 return (spa_vdev_exit(spa, NULL, txg, error)); 6076 6077 /* check new spa name before going any further */ 6078 if (spa_lookup(newname) != NULL) 6079 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 6080 6081 /* 6082 * scan through all the children to ensure they're all mirrors 6083 */ 6084 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 6085 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 6086 &children) != 0) 6087 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6088 6089 /* first, check to ensure we've got the right child count */ 6090 rvd = spa->spa_root_vdev; 6091 lastlog = 0; 6092 for (c = 0; c < rvd->vdev_children; c++) { 6093 vdev_t *vd = rvd->vdev_child[c]; 6094 6095 /* don't count the holes & logs as children */ 6096 if (vd->vdev_islog || !vdev_is_concrete(vd)) { 6097 if (lastlog == 0) 6098 lastlog = c; 6099 continue; 6100 } 6101 6102 lastlog = 0; 6103 } 6104 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 6105 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6106 6107 /* next, ensure no spare or cache devices are part of the split */ 6108 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 6109 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 6110 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 6111 6112 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 6113 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 6114 6115 /* then, loop over each vdev and validate it */ 6116 for (c = 0; c < children; c++) { 6117 uint64_t is_hole = 0; 6118 6119 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 6120 &is_hole); 6121 6122 if (is_hole != 0) { 6123 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 6124 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 6125 continue; 6126 } else { 6127 error = SET_ERROR(EINVAL); 6128 break; 6129 } 6130 } 6131 6132 /* which disk is going to be split? */ 6133 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 6134 &glist[c]) != 0) { 6135 error = SET_ERROR(EINVAL); 6136 break; 6137 } 6138 6139 /* look it up in the spa */ 6140 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 6141 if (vml[c] == NULL) { 6142 error = SET_ERROR(ENODEV); 6143 break; 6144 } 6145 6146 /* make sure there's nothing stopping the split */ 6147 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 6148 vml[c]->vdev_islog || 6149 !vdev_is_concrete(vml[c]) || 6150 vml[c]->vdev_isspare || 6151 vml[c]->vdev_isl2cache || 6152 !vdev_writeable(vml[c]) || 6153 vml[c]->vdev_children != 0 || 6154 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 6155 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 6156 error = SET_ERROR(EINVAL); 6157 break; 6158 } 6159 6160 if (vdev_dtl_required(vml[c])) { 6161 error = SET_ERROR(EBUSY); 6162 break; 6163 } 6164 6165 /* we need certain info from the top level */ 6166 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 6167 vml[c]->vdev_top->vdev_ms_array) == 0); 6168 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 6169 vml[c]->vdev_top->vdev_ms_shift) == 0); 6170 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 6171 vml[c]->vdev_top->vdev_asize) == 0); 6172 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 6173 vml[c]->vdev_top->vdev_ashift) == 0); 6174 6175 /* transfer per-vdev ZAPs */ 6176 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 6177 VERIFY0(nvlist_add_uint64(child[c], 6178 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 6179 6180 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 6181 VERIFY0(nvlist_add_uint64(child[c], 6182 ZPOOL_CONFIG_VDEV_TOP_ZAP, 6183 vml[c]->vdev_parent->vdev_top_zap)); 6184 } 6185 6186 if (error != 0) { 6187 kmem_free(vml, children * sizeof (vdev_t *)); 6188 kmem_free(glist, children * sizeof (uint64_t)); 6189 return (spa_vdev_exit(spa, NULL, txg, error)); 6190 } 6191 6192 /* stop writers from using the disks */ 6193 for (c = 0; c < children; c++) { 6194 if (vml[c] != NULL) 6195 vml[c]->vdev_offline = B_TRUE; 6196 } 6197 vdev_reopen(spa->spa_root_vdev); 6198 6199 /* 6200 * Temporarily record the splitting vdevs in the spa config. This 6201 * will disappear once the config is regenerated. 6202 */ 6203 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6204 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 6205 glist, children) == 0); 6206 kmem_free(glist, children * sizeof (uint64_t)); 6207 6208 mutex_enter(&spa->spa_props_lock); 6209 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 6210 nvl) == 0); 6211 mutex_exit(&spa->spa_props_lock); 6212 spa->spa_config_splitting = nvl; 6213 vdev_config_dirty(spa->spa_root_vdev); 6214 6215 /* configure and create the new pool */ 6216 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 6217 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 6218 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 6219 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6220 spa_version(spa)) == 0); 6221 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 6222 spa->spa_config_txg) == 0); 6223 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 6224 spa_generate_guid(NULL)) == 0); 6225 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 6226 (void) nvlist_lookup_string(props, 6227 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 6228 6229 /* add the new pool to the namespace */ 6230 newspa = spa_add(newname, config, altroot); 6231 newspa->spa_avz_action = AVZ_ACTION_REBUILD; 6232 newspa->spa_config_txg = spa->spa_config_txg; 6233 spa_set_log_state(newspa, SPA_LOG_CLEAR); 6234 6235 /* release the spa config lock, retaining the namespace lock */ 6236 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 6237 6238 if (zio_injection_enabled) 6239 zio_handle_panic_injection(spa, FTAG, 1); 6240 6241 spa_activate(newspa, spa_mode_global); 6242 spa_async_suspend(newspa); 6243 6244#ifndef illumos 6245 /* mark that we are creating new spa by splitting */ 6246 newspa->spa_splitting_newspa = B_TRUE; 6247#endif 6248 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT; 6249 6250 /* create the new pool from the disks of the original pool */ 6251 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE); 6252#ifndef illumos 6253 newspa->spa_splitting_newspa = B_FALSE; 6254#endif 6255 if (error) 6256 goto out; 6257 6258 /* if that worked, generate a real config for the new pool */ 6259 if (newspa->spa_root_vdev != NULL) { 6260 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 6261 NV_UNIQUE_NAME, KM_SLEEP) == 0); 6262 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 6263 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 6264 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 6265 B_TRUE)); 6266 } 6267 6268 /* set the props */ 6269 if (props != NULL) { 6270 spa_configfile_set(newspa, props, B_FALSE); 6271 error = spa_prop_set(newspa, props); 6272 if (error) 6273 goto out; 6274 } 6275 6276 /* flush everything */ 6277 txg = spa_vdev_config_enter(newspa); 6278 vdev_config_dirty(newspa->spa_root_vdev); 6279 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 6280 6281 if (zio_injection_enabled) 6282 zio_handle_panic_injection(spa, FTAG, 2); 6283 6284 spa_async_resume(newspa); 6285 6286 /* finally, update the original pool's config */ 6287 txg = spa_vdev_config_enter(spa); 6288 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 6289 error = dmu_tx_assign(tx, TXG_WAIT); 6290 if (error != 0) 6291 dmu_tx_abort(tx); 6292 for (c = 0; c < children; c++) { 6293 if (vml[c] != NULL) { 6294 vdev_split(vml[c]); 6295 if (error == 0) 6296 spa_history_log_internal(spa, "detach", tx, 6297 "vdev=%s", vml[c]->vdev_path); 6298 6299 vdev_free(vml[c]); 6300 } 6301 } 6302 spa->spa_avz_action = AVZ_ACTION_REBUILD; 6303 vdev_config_dirty(spa->spa_root_vdev); 6304 spa->spa_config_splitting = NULL; 6305 nvlist_free(nvl); 6306 if (error == 0) 6307 dmu_tx_commit(tx); 6308 (void) spa_vdev_exit(spa, NULL, txg, 0); 6309 6310 if (zio_injection_enabled) 6311 zio_handle_panic_injection(spa, FTAG, 3); 6312 6313 /* split is complete; log a history record */ 6314 spa_history_log_internal(newspa, "split", NULL, 6315 "from pool %s", spa_name(spa)); 6316 6317 kmem_free(vml, children * sizeof (vdev_t *)); 6318 6319 /* if we're not going to mount the filesystems in userland, export */ 6320 if (exp) 6321 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 6322 B_FALSE, B_FALSE); 6323 6324 return (error); 6325 6326out: 6327 spa_unload(newspa); 6328 spa_deactivate(newspa); 6329 spa_remove(newspa); 6330 6331 txg = spa_vdev_config_enter(spa); 6332 6333 /* re-online all offlined disks */ 6334 for (c = 0; c < children; c++) { 6335 if (vml[c] != NULL) 6336 vml[c]->vdev_offline = B_FALSE; 6337 } 6338 vdev_reopen(spa->spa_root_vdev); 6339 6340 nvlist_free(spa->spa_config_splitting); 6341 spa->spa_config_splitting = NULL; 6342 (void) spa_vdev_exit(spa, NULL, txg, error); 6343 6344 kmem_free(vml, children * sizeof (vdev_t *)); 6345 return (error); 6346} 6347 6348/* 6349 * Find any device that's done replacing, or a vdev marked 'unspare' that's 6350 * currently spared, so we can detach it. 6351 */ 6352static vdev_t * 6353spa_vdev_resilver_done_hunt(vdev_t *vd) 6354{ 6355 vdev_t *newvd, *oldvd; 6356 6357 for (int c = 0; c < vd->vdev_children; c++) { 6358 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 6359 if (oldvd != NULL) 6360 return (oldvd); 6361 } 6362 6363 /* 6364 * Check for a completed replacement. We always consider the first 6365 * vdev in the list to be the oldest vdev, and the last one to be 6366 * the newest (see spa_vdev_attach() for how that works). In 6367 * the case where the newest vdev is faulted, we will not automatically 6368 * remove it after a resilver completes. This is OK as it will require 6369 * user intervention to determine which disk the admin wishes to keep. 6370 */ 6371 if (vd->vdev_ops == &vdev_replacing_ops) { 6372 ASSERT(vd->vdev_children > 1); 6373 6374 newvd = vd->vdev_child[vd->vdev_children - 1]; 6375 oldvd = vd->vdev_child[0]; 6376 6377 if (vdev_dtl_empty(newvd, DTL_MISSING) && 6378 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6379 !vdev_dtl_required(oldvd)) 6380 return (oldvd); 6381 } 6382 6383 /* 6384 * Check for a completed resilver with the 'unspare' flag set. 6385 */ 6386 if (vd->vdev_ops == &vdev_spare_ops) { 6387 vdev_t *first = vd->vdev_child[0]; 6388 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 6389 6390 if (last->vdev_unspare) { 6391 oldvd = first; 6392 newvd = last; 6393 } else if (first->vdev_unspare) { 6394 oldvd = last; 6395 newvd = first; 6396 } else { 6397 oldvd = NULL; 6398 } 6399 6400 if (oldvd != NULL && 6401 vdev_dtl_empty(newvd, DTL_MISSING) && 6402 vdev_dtl_empty(newvd, DTL_OUTAGE) && 6403 !vdev_dtl_required(oldvd)) 6404 return (oldvd); 6405 6406 /* 6407 * If there are more than two spares attached to a disk, 6408 * and those spares are not required, then we want to 6409 * attempt to free them up now so that they can be used 6410 * by other pools. Once we're back down to a single 6411 * disk+spare, we stop removing them. 6412 */ 6413 if (vd->vdev_children > 2) { 6414 newvd = vd->vdev_child[1]; 6415 6416 if (newvd->vdev_isspare && last->vdev_isspare && 6417 vdev_dtl_empty(last, DTL_MISSING) && 6418 vdev_dtl_empty(last, DTL_OUTAGE) && 6419 !vdev_dtl_required(newvd)) 6420 return (newvd); 6421 } 6422 } 6423 6424 return (NULL); 6425} 6426 6427static void 6428spa_vdev_resilver_done(spa_t *spa) 6429{ 6430 vdev_t *vd, *pvd, *ppvd; 6431 uint64_t guid, sguid, pguid, ppguid; 6432 6433 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6434 6435 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 6436 pvd = vd->vdev_parent; 6437 ppvd = pvd->vdev_parent; 6438 guid = vd->vdev_guid; 6439 pguid = pvd->vdev_guid; 6440 ppguid = ppvd->vdev_guid; 6441 sguid = 0; 6442 /* 6443 * If we have just finished replacing a hot spared device, then 6444 * we need to detach the parent's first child (the original hot 6445 * spare) as well. 6446 */ 6447 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 6448 ppvd->vdev_children == 2) { 6449 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 6450 sguid = ppvd->vdev_child[1]->vdev_guid; 6451 } 6452 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 6453 6454 spa_config_exit(spa, SCL_ALL, FTAG); 6455 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 6456 return; 6457 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 6458 return; 6459 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6460 } 6461 6462 spa_config_exit(spa, SCL_ALL, FTAG); 6463} 6464 6465/* 6466 * Update the stored path or FRU for this vdev. 6467 */ 6468int 6469spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 6470 boolean_t ispath) 6471{ 6472 vdev_t *vd; 6473 boolean_t sync = B_FALSE; 6474 6475 ASSERT(spa_writeable(spa)); 6476 6477 spa_vdev_state_enter(spa, SCL_ALL); 6478 6479 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 6480 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 6481 6482 if (!vd->vdev_ops->vdev_op_leaf) 6483 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 6484 6485 if (ispath) { 6486 if (strcmp(value, vd->vdev_path) != 0) { 6487 spa_strfree(vd->vdev_path); 6488 vd->vdev_path = spa_strdup(value); 6489 sync = B_TRUE; 6490 } 6491 } else { 6492 if (vd->vdev_fru == NULL) { 6493 vd->vdev_fru = spa_strdup(value); 6494 sync = B_TRUE; 6495 } else if (strcmp(value, vd->vdev_fru) != 0) { 6496 spa_strfree(vd->vdev_fru); 6497 vd->vdev_fru = spa_strdup(value); 6498 sync = B_TRUE; 6499 } 6500 } 6501 6502 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 6503} 6504 6505int 6506spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 6507{ 6508 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 6509} 6510 6511int 6512spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 6513{ 6514 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 6515} 6516 6517/* 6518 * ========================================================================== 6519 * SPA Scanning 6520 * ========================================================================== 6521 */ 6522int 6523spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) 6524{ 6525 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6526 6527 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6528 return (SET_ERROR(EBUSY)); 6529 6530 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); 6531} 6532 6533int 6534spa_scan_stop(spa_t *spa) 6535{ 6536 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6537 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6538 return (SET_ERROR(EBUSY)); 6539 return (dsl_scan_cancel(spa->spa_dsl_pool)); 6540} 6541 6542int 6543spa_scan(spa_t *spa, pool_scan_func_t func) 6544{ 6545 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6546 6547 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 6548 return (SET_ERROR(ENOTSUP)); 6549 6550 /* 6551 * If a resilver was requested, but there is no DTL on a 6552 * writeable leaf device, we have nothing to do. 6553 */ 6554 if (func == POOL_SCAN_RESILVER && 6555 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 6556 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 6557 return (0); 6558 } 6559 6560 return (dsl_scan(spa->spa_dsl_pool, func)); 6561} 6562 6563/* 6564 * ========================================================================== 6565 * SPA async task processing 6566 * ========================================================================== 6567 */ 6568 6569static void 6570spa_async_remove(spa_t *spa, vdev_t *vd) 6571{ 6572 if (vd->vdev_remove_wanted) { 6573 vd->vdev_remove_wanted = B_FALSE; 6574 vd->vdev_delayed_close = B_FALSE; 6575 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 6576 6577 /* 6578 * We want to clear the stats, but we don't want to do a full 6579 * vdev_clear() as that will cause us to throw away 6580 * degraded/faulted state as well as attempt to reopen the 6581 * device, all of which is a waste. 6582 */ 6583 vd->vdev_stat.vs_read_errors = 0; 6584 vd->vdev_stat.vs_write_errors = 0; 6585 vd->vdev_stat.vs_checksum_errors = 0; 6586 6587 vdev_state_dirty(vd->vdev_top); 6588 /* Tell userspace that the vdev is gone. */ 6589 zfs_post_remove(spa, vd); 6590 } 6591 6592 for (int c = 0; c < vd->vdev_children; c++) 6593 spa_async_remove(spa, vd->vdev_child[c]); 6594} 6595 6596static void 6597spa_async_probe(spa_t *spa, vdev_t *vd) 6598{ 6599 if (vd->vdev_probe_wanted) { 6600 vd->vdev_probe_wanted = B_FALSE; 6601 vdev_reopen(vd); /* vdev_open() does the actual probe */ 6602 } 6603 6604 for (int c = 0; c < vd->vdev_children; c++) 6605 spa_async_probe(spa, vd->vdev_child[c]); 6606} 6607 6608static void 6609spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6610{ 6611 sysevent_id_t eid; 6612 nvlist_t *attr; 6613 char *physpath; 6614 6615 if (!spa->spa_autoexpand) 6616 return; 6617 6618 for (int c = 0; c < vd->vdev_children; c++) { 6619 vdev_t *cvd = vd->vdev_child[c]; 6620 spa_async_autoexpand(spa, cvd); 6621 } 6622 6623 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6624 return; 6625 6626 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6627 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6628 6629 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6630 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6631 6632 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6633 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 6634 6635 nvlist_free(attr); 6636 kmem_free(physpath, MAXPATHLEN); 6637} 6638 6639static void 6640spa_async_thread(void *arg) 6641{ 6642 spa_t *spa = (spa_t *)arg; 6643 int tasks; 6644 6645 ASSERT(spa->spa_sync_on); 6646 6647 mutex_enter(&spa->spa_async_lock); 6648 tasks = spa->spa_async_tasks; 6649 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 6650 mutex_exit(&spa->spa_async_lock); 6651 6652 /* 6653 * See if the config needs to be updated. 6654 */ 6655 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6656 uint64_t old_space, new_space; 6657 6658 mutex_enter(&spa_namespace_lock); 6659 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6660 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6661 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6662 mutex_exit(&spa_namespace_lock); 6663 6664 /* 6665 * If the pool grew as a result of the config update, 6666 * then log an internal history event. 6667 */ 6668 if (new_space != old_space) { 6669 spa_history_log_internal(spa, "vdev online", NULL, 6670 "pool '%s' size: %llu(+%llu)", 6671 spa_name(spa), new_space, new_space - old_space); 6672 } 6673 } 6674 6675 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6676 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6677 spa_async_autoexpand(spa, spa->spa_root_vdev); 6678 spa_config_exit(spa, SCL_CONFIG, FTAG); 6679 } 6680 6681 /* 6682 * See if any devices need to be probed. 6683 */ 6684 if (tasks & SPA_ASYNC_PROBE) { 6685 spa_vdev_state_enter(spa, SCL_NONE); 6686 spa_async_probe(spa, spa->spa_root_vdev); 6687 (void) spa_vdev_state_exit(spa, NULL, 0); 6688 } 6689 6690 /* 6691 * If any devices are done replacing, detach them. 6692 */ 6693 if (tasks & SPA_ASYNC_RESILVER_DONE) 6694 spa_vdev_resilver_done(spa); 6695 6696 /* 6697 * Kick off a resilver. 6698 */ 6699 if (tasks & SPA_ASYNC_RESILVER) 6700 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6701 6702 /* 6703 * Let the world know that we're done. 6704 */ 6705 mutex_enter(&spa->spa_async_lock); 6706 spa->spa_async_thread = NULL; 6707 cv_broadcast(&spa->spa_async_cv); 6708 mutex_exit(&spa->spa_async_lock); 6709 thread_exit(); 6710} 6711 6712static void 6713spa_async_thread_vd(void *arg) 6714{ 6715 spa_t *spa = arg; 6716 int tasks; 6717 6718 mutex_enter(&spa->spa_async_lock); 6719 tasks = spa->spa_async_tasks; 6720retry: 6721 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6722 mutex_exit(&spa->spa_async_lock); 6723 6724 /* 6725 * See if any devices need to be marked REMOVED. 6726 */ 6727 if (tasks & SPA_ASYNC_REMOVE) { 6728 spa_vdev_state_enter(spa, SCL_NONE); 6729 spa_async_remove(spa, spa->spa_root_vdev); 6730 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6731 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6732 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6733 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6734 (void) spa_vdev_state_exit(spa, NULL, 0); 6735 } 6736 6737 /* 6738 * Let the world know that we're done. 6739 */ 6740 mutex_enter(&spa->spa_async_lock); 6741 tasks = spa->spa_async_tasks; 6742 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6743 goto retry; 6744 spa->spa_async_thread_vd = NULL; 6745 cv_broadcast(&spa->spa_async_cv); 6746 mutex_exit(&spa->spa_async_lock); 6747 thread_exit(); 6748} 6749 6750void 6751spa_async_suspend(spa_t *spa) 6752{ 6753 mutex_enter(&spa->spa_async_lock); 6754 spa->spa_async_suspended++; 6755 while (spa->spa_async_thread != NULL || 6756 spa->spa_async_thread_vd != NULL || 6757 spa->spa_condense_thread != NULL) 6758 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6759 mutex_exit(&spa->spa_async_lock); 6760 6761 spa_vdev_remove_suspend(spa); 6762} 6763 6764void 6765spa_async_resume(spa_t *spa) 6766{ 6767 mutex_enter(&spa->spa_async_lock); 6768 ASSERT(spa->spa_async_suspended != 0); 6769 spa->spa_async_suspended--; 6770 mutex_exit(&spa->spa_async_lock); 6771 spa_restart_removal(spa); 6772} 6773 6774static boolean_t 6775spa_async_tasks_pending(spa_t *spa) 6776{ 6777 uint_t non_config_tasks; 6778 uint_t config_task; 6779 boolean_t config_task_suspended; 6780 6781 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6782 SPA_ASYNC_REMOVE); 6783 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6784 if (spa->spa_ccw_fail_time == 0) { 6785 config_task_suspended = B_FALSE; 6786 } else { 6787 config_task_suspended = 6788 (gethrtime() - spa->spa_ccw_fail_time) < 6789 (zfs_ccw_retry_interval * NANOSEC); 6790 } 6791 6792 return (non_config_tasks || (config_task && !config_task_suspended)); 6793} 6794 6795static void 6796spa_async_dispatch(spa_t *spa) 6797{ 6798 mutex_enter(&spa->spa_async_lock); 6799 if (spa_async_tasks_pending(spa) && 6800 !spa->spa_async_suspended && 6801 spa->spa_async_thread == NULL && 6802 rootdir != NULL) 6803 spa->spa_async_thread = thread_create(NULL, 0, 6804 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6805 mutex_exit(&spa->spa_async_lock); 6806} 6807 6808static void 6809spa_async_dispatch_vd(spa_t *spa) 6810{ 6811 mutex_enter(&spa->spa_async_lock); 6812 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6813 !spa->spa_async_suspended && 6814 spa->spa_async_thread_vd == NULL && 6815 rootdir != NULL) 6816 spa->spa_async_thread_vd = thread_create(NULL, 0, 6817 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6818 mutex_exit(&spa->spa_async_lock); 6819} 6820 6821void 6822spa_async_request(spa_t *spa, int task) 6823{ 6824 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6825 mutex_enter(&spa->spa_async_lock); 6826 spa->spa_async_tasks |= task; 6827 mutex_exit(&spa->spa_async_lock); 6828 spa_async_dispatch_vd(spa); 6829} 6830 6831/* 6832 * ========================================================================== 6833 * SPA syncing routines 6834 * ========================================================================== 6835 */ 6836 6837static int 6838bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6839{ 6840 bpobj_t *bpo = arg; 6841 bpobj_enqueue(bpo, bp, tx); 6842 return (0); 6843} 6844 6845static int 6846spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6847{ 6848 zio_t *zio = arg; 6849 6850 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6851 BP_GET_PSIZE(bp), zio->io_flags)); 6852 return (0); 6853} 6854 6855/* 6856 * Note: this simple function is not inlined to make it easier to dtrace the 6857 * amount of time spent syncing frees. 6858 */ 6859static void 6860spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6861{ 6862 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6863 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6864 VERIFY(zio_wait(zio) == 0); 6865} 6866 6867/* 6868 * Note: this simple function is not inlined to make it easier to dtrace the 6869 * amount of time spent syncing deferred frees. 6870 */ 6871static void 6872spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6873{ 6874 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6875 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6876 spa_free_sync_cb, zio, tx), ==, 0); 6877 VERIFY0(zio_wait(zio)); 6878} 6879 6880 6881static void 6882spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6883{ 6884 char *packed = NULL; 6885 size_t bufsize; 6886 size_t nvsize = 0; 6887 dmu_buf_t *db; 6888 6889 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6890 6891 /* 6892 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6893 * information. This avoids the dmu_buf_will_dirty() path and 6894 * saves us a pre-read to get data we don't actually care about. 6895 */ 6896 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6897 packed = kmem_alloc(bufsize, KM_SLEEP); 6898 6899 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6900 KM_SLEEP) == 0); 6901 bzero(packed + nvsize, bufsize - nvsize); 6902 6903 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6904 6905 kmem_free(packed, bufsize); 6906 6907 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6908 dmu_buf_will_dirty(db, tx); 6909 *(uint64_t *)db->db_data = nvsize; 6910 dmu_buf_rele(db, FTAG); 6911} 6912 6913static void 6914spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6915 const char *config, const char *entry) 6916{ 6917 nvlist_t *nvroot; 6918 nvlist_t **list; 6919 int i; 6920 6921 if (!sav->sav_sync) 6922 return; 6923 6924 /* 6925 * Update the MOS nvlist describing the list of available devices. 6926 * spa_validate_aux() will have already made sure this nvlist is 6927 * valid and the vdevs are labeled appropriately. 6928 */ 6929 if (sav->sav_object == 0) { 6930 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6931 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6932 sizeof (uint64_t), tx); 6933 VERIFY(zap_update(spa->spa_meta_objset, 6934 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6935 &sav->sav_object, tx) == 0); 6936 } 6937 6938 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6939 if (sav->sav_count == 0) { 6940 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6941 } else { 6942 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6943 for (i = 0; i < sav->sav_count; i++) 6944 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6945 B_FALSE, VDEV_CONFIG_L2CACHE); 6946 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6947 sav->sav_count) == 0); 6948 for (i = 0; i < sav->sav_count; i++) 6949 nvlist_free(list[i]); 6950 kmem_free(list, sav->sav_count * sizeof (void *)); 6951 } 6952 6953 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6954 nvlist_free(nvroot); 6955 6956 sav->sav_sync = B_FALSE; 6957} 6958 6959/* 6960 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 6961 * The all-vdev ZAP must be empty. 6962 */ 6963static void 6964spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 6965{ 6966 spa_t *spa = vd->vdev_spa; 6967 if (vd->vdev_top_zap != 0) { 6968 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6969 vd->vdev_top_zap, tx)); 6970 } 6971 if (vd->vdev_leaf_zap != 0) { 6972 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6973 vd->vdev_leaf_zap, tx)); 6974 } 6975 for (uint64_t i = 0; i < vd->vdev_children; i++) { 6976 spa_avz_build(vd->vdev_child[i], avz, tx); 6977 } 6978} 6979 6980static void 6981spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6982{ 6983 nvlist_t *config; 6984 6985 /* 6986 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 6987 * its config may not be dirty but we still need to build per-vdev ZAPs. 6988 * Similarly, if the pool is being assembled (e.g. after a split), we 6989 * need to rebuild the AVZ although the config may not be dirty. 6990 */ 6991 if (list_is_empty(&spa->spa_config_dirty_list) && 6992 spa->spa_avz_action == AVZ_ACTION_NONE) 6993 return; 6994 6995 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6996 6997 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 6998 spa->spa_avz_action == AVZ_ACTION_INITIALIZE || 6999 spa->spa_all_vdev_zaps != 0); 7000 7001 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 7002 /* Make and build the new AVZ */ 7003 uint64_t new_avz = zap_create(spa->spa_meta_objset, 7004 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 7005 spa_avz_build(spa->spa_root_vdev, new_avz, tx); 7006 7007 /* Diff old AVZ with new one */ 7008 zap_cursor_t zc; 7009 zap_attribute_t za; 7010 7011 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7012 spa->spa_all_vdev_zaps); 7013 zap_cursor_retrieve(&zc, &za) == 0; 7014 zap_cursor_advance(&zc)) { 7015 uint64_t vdzap = za.za_first_integer; 7016 if (zap_lookup_int(spa->spa_meta_objset, new_avz, 7017 vdzap) == ENOENT) { 7018 /* 7019 * ZAP is listed in old AVZ but not in new one; 7020 * destroy it 7021 */ 7022 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 7023 tx)); 7024 } 7025 } 7026 7027 zap_cursor_fini(&zc); 7028 7029 /* Destroy the old AVZ */ 7030 VERIFY0(zap_destroy(spa->spa_meta_objset, 7031 spa->spa_all_vdev_zaps, tx)); 7032 7033 /* Replace the old AVZ in the dir obj with the new one */ 7034 VERIFY0(zap_update(spa->spa_meta_objset, 7035 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 7036 sizeof (new_avz), 1, &new_avz, tx)); 7037 7038 spa->spa_all_vdev_zaps = new_avz; 7039 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 7040 zap_cursor_t zc; 7041 zap_attribute_t za; 7042 7043 /* Walk through the AVZ and destroy all listed ZAPs */ 7044 for (zap_cursor_init(&zc, spa->spa_meta_objset, 7045 spa->spa_all_vdev_zaps); 7046 zap_cursor_retrieve(&zc, &za) == 0; 7047 zap_cursor_advance(&zc)) { 7048 uint64_t zap = za.za_first_integer; 7049 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 7050 } 7051 7052 zap_cursor_fini(&zc); 7053 7054 /* Destroy and unlink the AVZ itself */ 7055 VERIFY0(zap_destroy(spa->spa_meta_objset, 7056 spa->spa_all_vdev_zaps, tx)); 7057 VERIFY0(zap_remove(spa->spa_meta_objset, 7058 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 7059 spa->spa_all_vdev_zaps = 0; 7060 } 7061 7062 if (spa->spa_all_vdev_zaps == 0) { 7063 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 7064 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 7065 DMU_POOL_VDEV_ZAP_MAP, tx); 7066 } 7067 spa->spa_avz_action = AVZ_ACTION_NONE; 7068 7069 /* Create ZAPs for vdevs that don't have them. */ 7070 vdev_construct_zaps(spa->spa_root_vdev, tx); 7071 7072 config = spa_config_generate(spa, spa->spa_root_vdev, 7073 dmu_tx_get_txg(tx), B_FALSE); 7074 7075 /* 7076 * If we're upgrading the spa version then make sure that 7077 * the config object gets updated with the correct version. 7078 */ 7079 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 7080 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 7081 spa->spa_uberblock.ub_version); 7082 7083 spa_config_exit(spa, SCL_STATE, FTAG); 7084 7085 nvlist_free(spa->spa_config_syncing); 7086 spa->spa_config_syncing = config; 7087 7088 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 7089} 7090 7091static void 7092spa_sync_version(void *arg, dmu_tx_t *tx) 7093{ 7094 uint64_t *versionp = arg; 7095 uint64_t version = *versionp; 7096 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7097 7098 /* 7099 * Setting the version is special cased when first creating the pool. 7100 */ 7101 ASSERT(tx->tx_txg != TXG_INITIAL); 7102 7103 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 7104 ASSERT(version >= spa_version(spa)); 7105 7106 spa->spa_uberblock.ub_version = version; 7107 vdev_config_dirty(spa->spa_root_vdev); 7108 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 7109} 7110 7111/* 7112 * Set zpool properties. 7113 */ 7114static void 7115spa_sync_props(void *arg, dmu_tx_t *tx) 7116{ 7117 nvlist_t *nvp = arg; 7118 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 7119 objset_t *mos = spa->spa_meta_objset; 7120 nvpair_t *elem = NULL; 7121 7122 mutex_enter(&spa->spa_props_lock); 7123 7124 while ((elem = nvlist_next_nvpair(nvp, elem))) { 7125 uint64_t intval; 7126 char *strval, *fname; 7127 zpool_prop_t prop; 7128 const char *propname; 7129 zprop_type_t proptype; 7130 spa_feature_t fid; 7131 7132 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 7133 case ZPOOL_PROP_INVAL: 7134 /* 7135 * We checked this earlier in spa_prop_validate(). 7136 */ 7137 ASSERT(zpool_prop_feature(nvpair_name(elem))); 7138 7139 fname = strchr(nvpair_name(elem), '@') + 1; 7140 VERIFY0(zfeature_lookup_name(fname, &fid)); 7141 7142 spa_feature_enable(spa, fid, tx); 7143 spa_history_log_internal(spa, "set", tx, 7144 "%s=enabled", nvpair_name(elem)); 7145 break; 7146 7147 case ZPOOL_PROP_VERSION: 7148 intval = fnvpair_value_uint64(elem); 7149 /* 7150 * The version is synced seperatly before other 7151 * properties and should be correct by now. 7152 */ 7153 ASSERT3U(spa_version(spa), >=, intval); 7154 break; 7155 7156 case ZPOOL_PROP_ALTROOT: 7157 /* 7158 * 'altroot' is a non-persistent property. It should 7159 * have been set temporarily at creation or import time. 7160 */ 7161 ASSERT(spa->spa_root != NULL); 7162 break; 7163 7164 case ZPOOL_PROP_READONLY: 7165 case ZPOOL_PROP_CACHEFILE: 7166 /* 7167 * 'readonly' and 'cachefile' are also non-persisitent 7168 * properties. 7169 */ 7170 break; 7171 case ZPOOL_PROP_COMMENT: 7172 strval = fnvpair_value_string(elem); 7173 if (spa->spa_comment != NULL) 7174 spa_strfree(spa->spa_comment); 7175 spa->spa_comment = spa_strdup(strval); 7176 /* 7177 * We need to dirty the configuration on all the vdevs 7178 * so that their labels get updated. It's unnecessary 7179 * to do this for pool creation since the vdev's 7180 * configuratoin has already been dirtied. 7181 */ 7182 if (tx->tx_txg != TXG_INITIAL) 7183 vdev_config_dirty(spa->spa_root_vdev); 7184 spa_history_log_internal(spa, "set", tx, 7185 "%s=%s", nvpair_name(elem), strval); 7186 break; 7187 default: 7188 /* 7189 * Set pool property values in the poolprops mos object. 7190 */ 7191 if (spa->spa_pool_props_object == 0) { 7192 spa->spa_pool_props_object = 7193 zap_create_link(mos, DMU_OT_POOL_PROPS, 7194 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 7195 tx); 7196 } 7197 7198 /* normalize the property name */ 7199 propname = zpool_prop_to_name(prop); 7200 proptype = zpool_prop_get_type(prop); 7201 7202 if (nvpair_type(elem) == DATA_TYPE_STRING) { 7203 ASSERT(proptype == PROP_TYPE_STRING); 7204 strval = fnvpair_value_string(elem); 7205 VERIFY0(zap_update(mos, 7206 spa->spa_pool_props_object, propname, 7207 1, strlen(strval) + 1, strval, tx)); 7208 spa_history_log_internal(spa, "set", tx, 7209 "%s=%s", nvpair_name(elem), strval); 7210 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 7211 intval = fnvpair_value_uint64(elem); 7212 7213 if (proptype == PROP_TYPE_INDEX) { 7214 const char *unused; 7215 VERIFY0(zpool_prop_index_to_string( 7216 prop, intval, &unused)); 7217 } 7218 VERIFY0(zap_update(mos, 7219 spa->spa_pool_props_object, propname, 7220 8, 1, &intval, tx)); 7221 spa_history_log_internal(spa, "set", tx, 7222 "%s=%lld", nvpair_name(elem), intval); 7223 } else { 7224 ASSERT(0); /* not allowed */ 7225 } 7226 7227 switch (prop) { 7228 case ZPOOL_PROP_DELEGATION: 7229 spa->spa_delegation = intval; 7230 break; 7231 case ZPOOL_PROP_BOOTFS: 7232 spa->spa_bootfs = intval; 7233 break; 7234 case ZPOOL_PROP_FAILUREMODE: 7235 spa->spa_failmode = intval; 7236 break; 7237 case ZPOOL_PROP_AUTOEXPAND: 7238 spa->spa_autoexpand = intval; 7239 if (tx->tx_txg != TXG_INITIAL) 7240 spa_async_request(spa, 7241 SPA_ASYNC_AUTOEXPAND); 7242 break; 7243 case ZPOOL_PROP_DEDUPDITTO: 7244 spa->spa_dedup_ditto = intval; 7245 break; 7246 default: 7247 break; 7248 } 7249 } 7250 7251 } 7252 7253 mutex_exit(&spa->spa_props_lock); 7254} 7255 7256/* 7257 * Perform one-time upgrade on-disk changes. spa_version() does not 7258 * reflect the new version this txg, so there must be no changes this 7259 * txg to anything that the upgrade code depends on after it executes. 7260 * Therefore this must be called after dsl_pool_sync() does the sync 7261 * tasks. 7262 */ 7263static void 7264spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 7265{ 7266 dsl_pool_t *dp = spa->spa_dsl_pool; 7267 7268 ASSERT(spa->spa_sync_pass == 1); 7269 7270 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 7271 7272 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 7273 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 7274 dsl_pool_create_origin(dp, tx); 7275 7276 /* Keeping the origin open increases spa_minref */ 7277 spa->spa_minref += 3; 7278 } 7279 7280 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 7281 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 7282 dsl_pool_upgrade_clones(dp, tx); 7283 } 7284 7285 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 7286 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 7287 dsl_pool_upgrade_dir_clones(dp, tx); 7288 7289 /* Keeping the freedir open increases spa_minref */ 7290 spa->spa_minref += 3; 7291 } 7292 7293 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 7294 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7295 spa_feature_create_zap_objects(spa, tx); 7296 } 7297 7298 /* 7299 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 7300 * when possibility to use lz4 compression for metadata was added 7301 * Old pools that have this feature enabled must be upgraded to have 7302 * this feature active 7303 */ 7304 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 7305 boolean_t lz4_en = spa_feature_is_enabled(spa, 7306 SPA_FEATURE_LZ4_COMPRESS); 7307 boolean_t lz4_ac = spa_feature_is_active(spa, 7308 SPA_FEATURE_LZ4_COMPRESS); 7309 7310 if (lz4_en && !lz4_ac) 7311 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 7312 } 7313 7314 /* 7315 * If we haven't written the salt, do so now. Note that the 7316 * feature may not be activated yet, but that's fine since 7317 * the presence of this ZAP entry is backwards compatible. 7318 */ 7319 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 7320 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 7321 VERIFY0(zap_add(spa->spa_meta_objset, 7322 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 7323 sizeof (spa->spa_cksum_salt.zcs_bytes), 7324 spa->spa_cksum_salt.zcs_bytes, tx)); 7325 } 7326 7327 rrw_exit(&dp->dp_config_rwlock, FTAG); 7328} 7329 7330static void 7331vdev_indirect_state_sync_verify(vdev_t *vd) 7332{ 7333 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping; 7334 vdev_indirect_births_t *vib = vd->vdev_indirect_births; 7335 7336 if (vd->vdev_ops == &vdev_indirect_ops) { 7337 ASSERT(vim != NULL); 7338 ASSERT(vib != NULL); 7339 } 7340 7341 if (vdev_obsolete_sm_object(vd) != 0) { 7342 ASSERT(vd->vdev_obsolete_sm != NULL); 7343 ASSERT(vd->vdev_removing || 7344 vd->vdev_ops == &vdev_indirect_ops); 7345 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0); 7346 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0); 7347 7348 ASSERT3U(vdev_obsolete_sm_object(vd), ==, 7349 space_map_object(vd->vdev_obsolete_sm)); 7350 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=, 7351 space_map_allocated(vd->vdev_obsolete_sm)); 7352 } 7353 ASSERT(vd->vdev_obsolete_segments != NULL); 7354 7355 /* 7356 * Since frees / remaps to an indirect vdev can only 7357 * happen in syncing context, the obsolete segments 7358 * tree must be empty when we start syncing. 7359 */ 7360 ASSERT0(range_tree_space(vd->vdev_obsolete_segments)); 7361} 7362 7363/* 7364 * Sync the specified transaction group. New blocks may be dirtied as 7365 * part of the process, so we iterate until it converges. 7366 */ 7367void 7368spa_sync(spa_t *spa, uint64_t txg) 7369{ 7370 dsl_pool_t *dp = spa->spa_dsl_pool; 7371 objset_t *mos = spa->spa_meta_objset; 7372 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 7373 vdev_t *rvd = spa->spa_root_vdev; 7374 vdev_t *vd; 7375 dmu_tx_t *tx; 7376 int error; 7377 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 7378 zfs_vdev_queue_depth_pct / 100; 7379 7380 VERIFY(spa_writeable(spa)); 7381 7382 /* 7383 * Wait for i/os issued in open context that need to complete 7384 * before this txg syncs. 7385 */ 7386 VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK])); 7387 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0); 7388 7389 /* 7390 * Lock out configuration changes. 7391 */ 7392 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 7393 7394 spa->spa_syncing_txg = txg; 7395 spa->spa_sync_pass = 0; 7396 7397 mutex_enter(&spa->spa_alloc_lock); 7398 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 7399 mutex_exit(&spa->spa_alloc_lock); 7400 7401 /* 7402 * If there are any pending vdev state changes, convert them 7403 * into config changes that go out with this transaction group. 7404 */ 7405 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7406 while (list_head(&spa->spa_state_dirty_list) != NULL) { 7407 /* 7408 * We need the write lock here because, for aux vdevs, 7409 * calling vdev_config_dirty() modifies sav_config. 7410 * This is ugly and will become unnecessary when we 7411 * eliminate the aux vdev wart by integrating all vdevs 7412 * into the root vdev tree. 7413 */ 7414 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7415 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 7416 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 7417 vdev_state_clean(vd); 7418 vdev_config_dirty(vd); 7419 } 7420 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 7421 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 7422 } 7423 spa_config_exit(spa, SCL_STATE, FTAG); 7424 7425 tx = dmu_tx_create_assigned(dp, txg); 7426 7427 spa->spa_sync_starttime = gethrtime(); 7428#ifdef illumos 7429 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 7430 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 7431#else /* !illumos */ 7432#ifdef _KERNEL 7433 callout_schedule(&spa->spa_deadman_cycid, 7434 hz * spa->spa_deadman_synctime / NANOSEC); 7435#endif 7436#endif /* illumos */ 7437 7438 /* 7439 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 7440 * set spa_deflate if we have no raid-z vdevs. 7441 */ 7442 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 7443 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 7444 int i; 7445 7446 for (i = 0; i < rvd->vdev_children; i++) { 7447 vd = rvd->vdev_child[i]; 7448 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 7449 break; 7450 } 7451 if (i == rvd->vdev_children) { 7452 spa->spa_deflate = TRUE; 7453 VERIFY(0 == zap_add(spa->spa_meta_objset, 7454 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 7455 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 7456 } 7457 } 7458 7459 /* 7460 * Set the top-level vdev's max queue depth. Evaluate each 7461 * top-level's async write queue depth in case it changed. 7462 * The max queue depth will not change in the middle of syncing 7463 * out this txg. 7464 */ 7465 uint64_t queue_depth_total = 0; 7466 for (int c = 0; c < rvd->vdev_children; c++) { 7467 vdev_t *tvd = rvd->vdev_child[c]; 7468 metaslab_group_t *mg = tvd->vdev_mg; 7469 7470 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 7471 !metaslab_group_initialized(mg)) 7472 continue; 7473 7474 /* 7475 * It is safe to do a lock-free check here because only async 7476 * allocations look at mg_max_alloc_queue_depth, and async 7477 * allocations all happen from spa_sync(). 7478 */ 7479 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); 7480 mg->mg_max_alloc_queue_depth = max_queue_depth; 7481 queue_depth_total += mg->mg_max_alloc_queue_depth; 7482 } 7483 metaslab_class_t *mc = spa_normal_class(spa); 7484 ASSERT0(refcount_count(&mc->mc_alloc_slots)); 7485 mc->mc_alloc_max_slots = queue_depth_total; 7486 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 7487 7488 ASSERT3U(mc->mc_alloc_max_slots, <=, 7489 max_queue_depth * rvd->vdev_children); 7490 7491 for (int c = 0; c < rvd->vdev_children; c++) { 7492 vdev_t *vd = rvd->vdev_child[c]; 7493 vdev_indirect_state_sync_verify(vd); 7494 7495 if (vdev_indirect_should_condense(vd)) { 7496 spa_condense_indirect_start_sync(vd, tx); 7497 break; 7498 } 7499 } 7500 7501 /* 7502 * Iterate to convergence. 7503 */ 7504 do { 7505 int pass = ++spa->spa_sync_pass; 7506 7507 spa_sync_config_object(spa, tx); 7508 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 7509 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 7510 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 7511 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 7512 spa_errlog_sync(spa, txg); 7513 dsl_pool_sync(dp, txg); 7514 7515 if (pass < zfs_sync_pass_deferred_free) { 7516 spa_sync_frees(spa, free_bpl, tx); 7517 } else { 7518 /* 7519 * We can not defer frees in pass 1, because 7520 * we sync the deferred frees later in pass 1. 7521 */ 7522 ASSERT3U(pass, >, 1); 7523 bplist_iterate(free_bpl, bpobj_enqueue_cb, 7524 &spa->spa_deferred_bpobj, tx); 7525 } 7526 7527 ddt_sync(spa, txg); 7528 dsl_scan_sync(dp, tx); 7529 7530 if (spa->spa_vdev_removal != NULL) 7531 svr_sync(spa, tx); 7532 7533 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 7534 != NULL) 7535 vdev_sync(vd, txg); 7536 7537 if (pass == 1) { 7538 spa_sync_upgrades(spa, tx); 7539 ASSERT3U(txg, >=, 7540 spa->spa_uberblock.ub_rootbp.blk_birth); 7541 /* 7542 * Note: We need to check if the MOS is dirty 7543 * because we could have marked the MOS dirty 7544 * without updating the uberblock (e.g. if we 7545 * have sync tasks but no dirty user data). We 7546 * need to check the uberblock's rootbp because 7547 * it is updated if we have synced out dirty 7548 * data (though in this case the MOS will most 7549 * likely also be dirty due to second order 7550 * effects, we don't want to rely on that here). 7551 */ 7552 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 7553 !dmu_objset_is_dirty(mos, txg)) { 7554 /* 7555 * Nothing changed on the first pass, 7556 * therefore this TXG is a no-op. Avoid 7557 * syncing deferred frees, so that we 7558 * can keep this TXG as a no-op. 7559 */ 7560 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 7561 txg)); 7562 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7563 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 7564 break; 7565 } 7566 spa_sync_deferred_frees(spa, tx); 7567 } 7568 7569 } while (dmu_objset_is_dirty(mos, txg)); 7570 7571 if (!list_is_empty(&spa->spa_config_dirty_list)) { 7572 /* 7573 * Make sure that the number of ZAPs for all the vdevs matches 7574 * the number of ZAPs in the per-vdev ZAP list. This only gets 7575 * called if the config is dirty; otherwise there may be 7576 * outstanding AVZ operations that weren't completed in 7577 * spa_sync_config_object. 7578 */ 7579 uint64_t all_vdev_zap_entry_count; 7580 ASSERT0(zap_count(spa->spa_meta_objset, 7581 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 7582 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 7583 all_vdev_zap_entry_count); 7584 } 7585 7586 if (spa->spa_vdev_removal != NULL) { 7587 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]); 7588 } 7589 7590 /* 7591 * Rewrite the vdev configuration (which includes the uberblock) 7592 * to commit the transaction group. 7593 * 7594 * If there are no dirty vdevs, we sync the uberblock to a few 7595 * random top-level vdevs that are known to be visible in the 7596 * config cache (see spa_vdev_add() for a complete description). 7597 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 7598 */ 7599 for (;;) { 7600 /* 7601 * We hold SCL_STATE to prevent vdev open/close/etc. 7602 * while we're attempting to write the vdev labels. 7603 */ 7604 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7605 7606 if (list_is_empty(&spa->spa_config_dirty_list)) { 7607 vdev_t *svd[SPA_SYNC_MIN_VDEVS]; 7608 int svdcount = 0; 7609 int children = rvd->vdev_children; 7610 int c0 = spa_get_random(children); 7611 7612 for (int c = 0; c < children; c++) { 7613 vd = rvd->vdev_child[(c0 + c) % children]; 7614 if (vd->vdev_ms_array == 0 || vd->vdev_islog || 7615 !vdev_is_concrete(vd)) 7616 continue; 7617 svd[svdcount++] = vd; 7618 if (svdcount == SPA_SYNC_MIN_VDEVS) 7619 break; 7620 } 7621 error = vdev_config_sync(svd, svdcount, txg); 7622 } else { 7623 error = vdev_config_sync(rvd->vdev_child, 7624 rvd->vdev_children, txg); 7625 } 7626 7627 if (error == 0) 7628 spa->spa_last_synced_guid = rvd->vdev_guid; 7629 7630 spa_config_exit(spa, SCL_STATE, FTAG); 7631 7632 if (error == 0) 7633 break; 7634 zio_suspend(spa, NULL); 7635 zio_resume_wait(spa); 7636 } 7637 dmu_tx_commit(tx); 7638 7639#ifdef illumos 7640 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 7641#else /* !illumos */ 7642#ifdef _KERNEL 7643 callout_drain(&spa->spa_deadman_cycid); 7644#endif 7645#endif /* illumos */ 7646 7647 /* 7648 * Clear the dirty config list. 7649 */ 7650 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 7651 vdev_config_clean(vd); 7652 7653 /* 7654 * Now that the new config has synced transactionally, 7655 * let it become visible to the config cache. 7656 */ 7657 if (spa->spa_config_syncing != NULL) { 7658 spa_config_set(spa, spa->spa_config_syncing); 7659 spa->spa_config_txg = txg; 7660 spa->spa_config_syncing = NULL; 7661 } 7662 7663 dsl_pool_sync_done(dp, txg); 7664 7665 mutex_enter(&spa->spa_alloc_lock); 7666 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 7667 mutex_exit(&spa->spa_alloc_lock); 7668 7669 /* 7670 * Update usable space statistics. 7671 */ 7672 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 7673 vdev_sync_done(vd, txg); 7674 7675 spa_update_dspace(spa); 7676 7677 /* 7678 * It had better be the case that we didn't dirty anything 7679 * since vdev_config_sync(). 7680 */ 7681 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 7682 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7683 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 7684 7685 spa->spa_sync_pass = 0; 7686 7687 /* 7688 * Update the last synced uberblock here. We want to do this at 7689 * the end of spa_sync() so that consumers of spa_last_synced_txg() 7690 * will be guaranteed that all the processing associated with 7691 * that txg has been completed. 7692 */ 7693 spa->spa_ubsync = spa->spa_uberblock; 7694 spa_config_exit(spa, SCL_CONFIG, FTAG); 7695 7696 spa_handle_ignored_writes(spa); 7697 7698 /* 7699 * If any async tasks have been requested, kick them off. 7700 */ 7701 spa_async_dispatch(spa); 7702 spa_async_dispatch_vd(spa); 7703} 7704 7705/* 7706 * Sync all pools. We don't want to hold the namespace lock across these 7707 * operations, so we take a reference on the spa_t and drop the lock during the 7708 * sync. 7709 */ 7710void 7711spa_sync_allpools(void) 7712{ 7713 spa_t *spa = NULL; 7714 mutex_enter(&spa_namespace_lock); 7715 while ((spa = spa_next(spa)) != NULL) { 7716 if (spa_state(spa) != POOL_STATE_ACTIVE || 7717 !spa_writeable(spa) || spa_suspended(spa)) 7718 continue; 7719 spa_open_ref(spa, FTAG); 7720 mutex_exit(&spa_namespace_lock); 7721 txg_wait_synced(spa_get_dsl(spa), 0); 7722 mutex_enter(&spa_namespace_lock); 7723 spa_close(spa, FTAG); 7724 } 7725 mutex_exit(&spa_namespace_lock); 7726} 7727 7728/* 7729 * ========================================================================== 7730 * Miscellaneous routines 7731 * ========================================================================== 7732 */ 7733 7734/* 7735 * Remove all pools in the system. 7736 */ 7737void 7738spa_evict_all(void) 7739{ 7740 spa_t *spa; 7741 7742 /* 7743 * Remove all cached state. All pools should be closed now, 7744 * so every spa in the AVL tree should be unreferenced. 7745 */ 7746 mutex_enter(&spa_namespace_lock); 7747 while ((spa = spa_next(NULL)) != NULL) { 7748 /* 7749 * Stop async tasks. The async thread may need to detach 7750 * a device that's been replaced, which requires grabbing 7751 * spa_namespace_lock, so we must drop it here. 7752 */ 7753 spa_open_ref(spa, FTAG); 7754 mutex_exit(&spa_namespace_lock); 7755 spa_async_suspend(spa); 7756 mutex_enter(&spa_namespace_lock); 7757 spa_close(spa, FTAG); 7758 7759 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 7760 spa_unload(spa); 7761 spa_deactivate(spa); 7762 } 7763 spa_remove(spa); 7764 } 7765 mutex_exit(&spa_namespace_lock); 7766} 7767 7768vdev_t * 7769spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 7770{ 7771 vdev_t *vd; 7772 int i; 7773 7774 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 7775 return (vd); 7776 7777 if (aux) { 7778 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 7779 vd = spa->spa_l2cache.sav_vdevs[i]; 7780 if (vd->vdev_guid == guid) 7781 return (vd); 7782 } 7783 7784 for (i = 0; i < spa->spa_spares.sav_count; i++) { 7785 vd = spa->spa_spares.sav_vdevs[i]; 7786 if (vd->vdev_guid == guid) 7787 return (vd); 7788 } 7789 } 7790 7791 return (NULL); 7792} 7793 7794void 7795spa_upgrade(spa_t *spa, uint64_t version) 7796{ 7797 ASSERT(spa_writeable(spa)); 7798 7799 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 7800 7801 /* 7802 * This should only be called for a non-faulted pool, and since a 7803 * future version would result in an unopenable pool, this shouldn't be 7804 * possible. 7805 */ 7806 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 7807 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 7808 7809 spa->spa_uberblock.ub_version = version; 7810 vdev_config_dirty(spa->spa_root_vdev); 7811 7812 spa_config_exit(spa, SCL_ALL, FTAG); 7813 7814 txg_wait_synced(spa_get_dsl(spa), 0); 7815} 7816 7817boolean_t 7818spa_has_spare(spa_t *spa, uint64_t guid) 7819{ 7820 int i; 7821 uint64_t spareguid; 7822 spa_aux_vdev_t *sav = &spa->spa_spares; 7823 7824 for (i = 0; i < sav->sav_count; i++) 7825 if (sav->sav_vdevs[i]->vdev_guid == guid) 7826 return (B_TRUE); 7827 7828 for (i = 0; i < sav->sav_npending; i++) { 7829 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 7830 &spareguid) == 0 && spareguid == guid) 7831 return (B_TRUE); 7832 } 7833 7834 return (B_FALSE); 7835} 7836 7837/* 7838 * Check if a pool has an active shared spare device. 7839 * Note: reference count of an active spare is 2, as a spare and as a replace 7840 */ 7841static boolean_t 7842spa_has_active_shared_spare(spa_t *spa) 7843{ 7844 int i, refcnt; 7845 uint64_t pool; 7846 spa_aux_vdev_t *sav = &spa->spa_spares; 7847 7848 for (i = 0; i < sav->sav_count; i++) { 7849 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 7850 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 7851 refcnt > 2) 7852 return (B_TRUE); 7853 } 7854 7855 return (B_FALSE); 7856} 7857 7858sysevent_t * 7859spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 7860{ 7861 sysevent_t *ev = NULL; 7862#ifdef _KERNEL 7863 sysevent_attr_list_t *attr = NULL; 7864 sysevent_value_t value; 7865 7866 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7867 SE_SLEEP); 7868 ASSERT(ev != NULL); 7869 7870 value.value_type = SE_DATA_TYPE_STRING; 7871 value.value.sv_string = spa_name(spa); 7872 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7873 goto done; 7874 7875 value.value_type = SE_DATA_TYPE_UINT64; 7876 value.value.sv_uint64 = spa_guid(spa); 7877 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7878 goto done; 7879 7880 if (vd) { 7881 value.value_type = SE_DATA_TYPE_UINT64; 7882 value.value.sv_uint64 = vd->vdev_guid; 7883 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7884 SE_SLEEP) != 0) 7885 goto done; 7886 7887 if (vd->vdev_path) { 7888 value.value_type = SE_DATA_TYPE_STRING; 7889 value.value.sv_string = vd->vdev_path; 7890 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7891 &value, SE_SLEEP) != 0) 7892 goto done; 7893 } 7894 } 7895 7896 if (hist_nvl != NULL) { 7897 fnvlist_merge((nvlist_t *)attr, hist_nvl); 7898 } 7899 7900 if (sysevent_attach_attributes(ev, attr) != 0) 7901 goto done; 7902 attr = NULL; 7903 7904done: 7905 if (attr) 7906 sysevent_free_attr(attr); 7907 7908#endif 7909 return (ev); 7910} 7911 7912void 7913spa_event_post(sysevent_t *ev) 7914{ 7915#ifdef _KERNEL 7916 sysevent_id_t eid; 7917 7918 (void) log_sysevent(ev, SE_SLEEP, &eid); 7919 sysevent_free(ev); 7920#endif 7921} 7922 7923void 7924spa_event_discard(sysevent_t *ev) 7925{ 7926#ifdef _KERNEL 7927 sysevent_free(ev); 7928#endif 7929} 7930 7931/* 7932 * Post a sysevent corresponding to the given event. The 'name' must be one of 7933 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7934 * filled in from the spa and (optionally) the vdev and history nvl. This 7935 * doesn't do anything in the userland libzpool, as we don't want consumers to 7936 * misinterpret ztest or zdb as real changes. 7937 */ 7938void 7939spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 7940{ 7941 spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); 7942} 7943