spa.c revision 307122
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, 2014 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 */ 31 32/* 33 * SPA: Storage Pool Allocator 34 * 35 * This file contains all the routines used when modifying on-disk SPA state. 36 * This includes opening, importing, destroying, exporting a pool, and syncing a 37 * pool. 38 */ 39 40#include <sys/zfs_context.h> 41#include <sys/fm/fs/zfs.h> 42#include <sys/spa_impl.h> 43#include <sys/zio.h> 44#include <sys/zio_checksum.h> 45#include <sys/dmu.h> 46#include <sys/dmu_tx.h> 47#include <sys/zap.h> 48#include <sys/zil.h> 49#include <sys/ddt.h> 50#include <sys/vdev_impl.h> 51#include <sys/metaslab.h> 52#include <sys/metaslab_impl.h> 53#include <sys/uberblock_impl.h> 54#include <sys/txg.h> 55#include <sys/avl.h> 56#include <sys/dmu_traverse.h> 57#include <sys/dmu_objset.h> 58#include <sys/unique.h> 59#include <sys/dsl_pool.h> 60#include <sys/dsl_dataset.h> 61#include <sys/dsl_dir.h> 62#include <sys/dsl_prop.h> 63#include <sys/dsl_synctask.h> 64#include <sys/fs/zfs.h> 65#include <sys/arc.h> 66#include <sys/callb.h> 67#include <sys/spa_boot.h> 68#include <sys/zfs_ioctl.h> 69#include <sys/dsl_scan.h> 70#include <sys/dmu_send.h> 71#include <sys/dsl_destroy.h> 72#include <sys/dsl_userhold.h> 73#include <sys/zfeature.h> 74#include <sys/zvol.h> 75#include <sys/trim_map.h> 76 77#ifdef _KERNEL 78#include <sys/callb.h> 79#include <sys/cpupart.h> 80#include <sys/zone.h> 81#endif /* _KERNEL */ 82 83#include "zfs_prop.h" 84#include "zfs_comutil.h" 85 86/* Check hostid on import? */ 87static int check_hostid = 1; 88 89/* 90 * The interval, in seconds, at which failed configuration cache file writes 91 * should be retried. 92 */ 93static int zfs_ccw_retry_interval = 300; 94 95SYSCTL_DECL(_vfs_zfs); 96TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 97SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 98 "Check hostid on import?"); 99TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 100SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 101 &zfs_ccw_retry_interval, 0, 102 "Configuration cache file write, retry after failure, interval (seconds)"); 103 104typedef enum zti_modes { 105 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 106 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 107 ZTI_MODE_NULL, /* don't create a taskq */ 108 ZTI_NMODES 109} zti_modes_t; 110 111#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 112#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 113#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 114 115#define ZTI_N(n) ZTI_P(n, 1) 116#define ZTI_ONE ZTI_N(1) 117 118typedef struct zio_taskq_info { 119 zti_modes_t zti_mode; 120 uint_t zti_value; 121 uint_t zti_count; 122} zio_taskq_info_t; 123 124static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 125 "issue", "issue_high", "intr", "intr_high" 126}; 127 128/* 129 * This table defines the taskq settings for each ZFS I/O type. When 130 * initializing a pool, we use this table to create an appropriately sized 131 * taskq. Some operations are low volume and therefore have a small, static 132 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 133 * macros. Other operations process a large amount of data; the ZTI_BATCH 134 * macro causes us to create a taskq oriented for throughput. Some operations 135 * are so high frequency and short-lived that the taskq itself can become a a 136 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 137 * additional degree of parallelism specified by the number of threads per- 138 * taskq and the number of taskqs; when dispatching an event in this case, the 139 * particular taskq is chosen at random. 140 * 141 * The different taskq priorities are to handle the different contexts (issue 142 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 143 * need to be handled with minimum delay. 144 */ 145const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 146 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 148 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 149 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 150 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 153}; 154 155static void spa_sync_version(void *arg, dmu_tx_t *tx); 156static void spa_sync_props(void *arg, dmu_tx_t *tx); 157static boolean_t spa_has_active_shared_spare(spa_t *spa); 158static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 159 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 160 char **ereport); 161static void spa_vdev_resilver_done(spa_t *spa); 162 163uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 164#ifdef PSRSET_BIND 165id_t zio_taskq_psrset_bind = PS_NONE; 166#endif 167#ifdef SYSDC 168boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 169#endif 170uint_t zio_taskq_basedc = 80; /* base duty cycle */ 171 172boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 173extern int zfs_sync_pass_deferred_free; 174 175#ifndef illumos 176extern void spa_deadman(void *arg); 177#endif 178 179/* 180 * This (illegal) pool name is used when temporarily importing a spa_t in order 181 * to get the vdev stats associated with the imported devices. 182 */ 183#define TRYIMPORT_NAME "$import" 184 185/* 186 * ========================================================================== 187 * SPA properties routines 188 * ========================================================================== 189 */ 190 191/* 192 * Add a (source=src, propname=propval) list to an nvlist. 193 */ 194static void 195spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 196 uint64_t intval, zprop_source_t src) 197{ 198 const char *propname = zpool_prop_to_name(prop); 199 nvlist_t *propval; 200 201 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 202 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 203 204 if (strval != NULL) 205 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 206 else 207 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 208 209 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 210 nvlist_free(propval); 211} 212 213/* 214 * Get property values from the spa configuration. 215 */ 216static void 217spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 218{ 219 vdev_t *rvd = spa->spa_root_vdev; 220 dsl_pool_t *pool = spa->spa_dsl_pool; 221 uint64_t size, alloc, cap, version; 222 zprop_source_t src = ZPROP_SRC_NONE; 223 spa_config_dirent_t *dp; 224 metaslab_class_t *mc = spa_normal_class(spa); 225 226 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 227 228 if (rvd != NULL) { 229 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 230 size = metaslab_class_get_space(spa_normal_class(spa)); 231 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 232 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 233 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 234 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 235 size - alloc, src); 236 237 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 238 metaslab_class_fragmentation(mc), src); 239 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 240 metaslab_class_expandable_space(mc), src); 241 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 242 (spa_mode(spa) == FREAD), src); 243 244 cap = (size == 0) ? 0 : (alloc * 100 / size); 245 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 246 247 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 248 ddt_get_pool_dedup_ratio(spa), src); 249 250 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 251 rvd->vdev_state, src); 252 253 version = spa_version(spa); 254 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 255 src = ZPROP_SRC_DEFAULT; 256 else 257 src = ZPROP_SRC_LOCAL; 258 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 259 } 260 261 if (pool != NULL) { 262 /* 263 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 264 * when opening pools before this version freedir will be NULL. 265 */ 266 if (pool->dp_free_dir != NULL) { 267 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 268 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 269 src); 270 } else { 271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 272 NULL, 0, src); 273 } 274 275 if (pool->dp_leak_dir != NULL) { 276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 277 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 278 src); 279 } else { 280 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 281 NULL, 0, src); 282 } 283 } 284 285 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 286 287 if (spa->spa_comment != NULL) { 288 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 289 0, ZPROP_SRC_LOCAL); 290 } 291 292 if (spa->spa_root != NULL) 293 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 294 0, ZPROP_SRC_LOCAL); 295 296 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 297 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 298 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 299 } else { 300 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 301 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 302 } 303 304 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 305 if (dp->scd_path == NULL) { 306 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 307 "none", 0, ZPROP_SRC_LOCAL); 308 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 309 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 310 dp->scd_path, 0, ZPROP_SRC_LOCAL); 311 } 312 } 313} 314 315/* 316 * Get zpool property values. 317 */ 318int 319spa_prop_get(spa_t *spa, nvlist_t **nvp) 320{ 321 objset_t *mos = spa->spa_meta_objset; 322 zap_cursor_t zc; 323 zap_attribute_t za; 324 int err; 325 326 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 327 328 mutex_enter(&spa->spa_props_lock); 329 330 /* 331 * Get properties from the spa config. 332 */ 333 spa_prop_get_config(spa, nvp); 334 335 /* If no pool property object, no more prop to get. */ 336 if (mos == NULL || spa->spa_pool_props_object == 0) { 337 mutex_exit(&spa->spa_props_lock); 338 return (0); 339 } 340 341 /* 342 * Get properties from the MOS pool property object. 343 */ 344 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 345 (err = zap_cursor_retrieve(&zc, &za)) == 0; 346 zap_cursor_advance(&zc)) { 347 uint64_t intval = 0; 348 char *strval = NULL; 349 zprop_source_t src = ZPROP_SRC_DEFAULT; 350 zpool_prop_t prop; 351 352 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 353 continue; 354 355 switch (za.za_integer_length) { 356 case 8: 357 /* integer property */ 358 if (za.za_first_integer != 359 zpool_prop_default_numeric(prop)) 360 src = ZPROP_SRC_LOCAL; 361 362 if (prop == ZPOOL_PROP_BOOTFS) { 363 dsl_pool_t *dp; 364 dsl_dataset_t *ds = NULL; 365 366 dp = spa_get_dsl(spa); 367 dsl_pool_config_enter(dp, FTAG); 368 if (err = dsl_dataset_hold_obj(dp, 369 za.za_first_integer, FTAG, &ds)) { 370 dsl_pool_config_exit(dp, FTAG); 371 break; 372 } 373 374 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 375 KM_SLEEP); 376 dsl_dataset_name(ds, strval); 377 dsl_dataset_rele(ds, FTAG); 378 dsl_pool_config_exit(dp, FTAG); 379 } else { 380 strval = NULL; 381 intval = za.za_first_integer; 382 } 383 384 spa_prop_add_list(*nvp, prop, strval, intval, src); 385 386 if (strval != NULL) 387 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 388 389 break; 390 391 case 1: 392 /* string property */ 393 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 394 err = zap_lookup(mos, spa->spa_pool_props_object, 395 za.za_name, 1, za.za_num_integers, strval); 396 if (err) { 397 kmem_free(strval, za.za_num_integers); 398 break; 399 } 400 spa_prop_add_list(*nvp, prop, strval, 0, src); 401 kmem_free(strval, za.za_num_integers); 402 break; 403 404 default: 405 break; 406 } 407 } 408 zap_cursor_fini(&zc); 409 mutex_exit(&spa->spa_props_lock); 410out: 411 if (err && err != ENOENT) { 412 nvlist_free(*nvp); 413 *nvp = NULL; 414 return (err); 415 } 416 417 return (0); 418} 419 420/* 421 * Validate the given pool properties nvlist and modify the list 422 * for the property values to be set. 423 */ 424static int 425spa_prop_validate(spa_t *spa, nvlist_t *props) 426{ 427 nvpair_t *elem; 428 int error = 0, reset_bootfs = 0; 429 uint64_t objnum = 0; 430 boolean_t has_feature = B_FALSE; 431 432 elem = NULL; 433 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 434 uint64_t intval; 435 char *strval, *slash, *check, *fname; 436 const char *propname = nvpair_name(elem); 437 zpool_prop_t prop = zpool_name_to_prop(propname); 438 439 switch (prop) { 440 case ZPROP_INVAL: 441 if (!zpool_prop_feature(propname)) { 442 error = SET_ERROR(EINVAL); 443 break; 444 } 445 446 /* 447 * Sanitize the input. 448 */ 449 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 450 error = SET_ERROR(EINVAL); 451 break; 452 } 453 454 if (nvpair_value_uint64(elem, &intval) != 0) { 455 error = SET_ERROR(EINVAL); 456 break; 457 } 458 459 if (intval != 0) { 460 error = SET_ERROR(EINVAL); 461 break; 462 } 463 464 fname = strchr(propname, '@') + 1; 465 if (zfeature_lookup_name(fname, NULL) != 0) { 466 error = SET_ERROR(EINVAL); 467 break; 468 } 469 470 has_feature = B_TRUE; 471 break; 472 473 case ZPOOL_PROP_VERSION: 474 error = nvpair_value_uint64(elem, &intval); 475 if (!error && 476 (intval < spa_version(spa) || 477 intval > SPA_VERSION_BEFORE_FEATURES || 478 has_feature)) 479 error = SET_ERROR(EINVAL); 480 break; 481 482 case ZPOOL_PROP_DELEGATION: 483 case ZPOOL_PROP_AUTOREPLACE: 484 case ZPOOL_PROP_LISTSNAPS: 485 case ZPOOL_PROP_AUTOEXPAND: 486 error = nvpair_value_uint64(elem, &intval); 487 if (!error && intval > 1) 488 error = SET_ERROR(EINVAL); 489 break; 490 491 case ZPOOL_PROP_BOOTFS: 492 /* 493 * If the pool version is less than SPA_VERSION_BOOTFS, 494 * or the pool is still being created (version == 0), 495 * the bootfs property cannot be set. 496 */ 497 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 498 error = SET_ERROR(ENOTSUP); 499 break; 500 } 501 502 /* 503 * Make sure the vdev config is bootable 504 */ 505 if (!vdev_is_bootable(spa->spa_root_vdev)) { 506 error = SET_ERROR(ENOTSUP); 507 break; 508 } 509 510 reset_bootfs = 1; 511 512 error = nvpair_value_string(elem, &strval); 513 514 if (!error) { 515 objset_t *os; 516 uint64_t propval; 517 518 if (strval == NULL || strval[0] == '\0') { 519 objnum = zpool_prop_default_numeric( 520 ZPOOL_PROP_BOOTFS); 521 break; 522 } 523 524 if (error = dmu_objset_hold(strval, FTAG, &os)) 525 break; 526 527 /* 528 * Must be ZPL, and its property settings 529 * must be supported by GRUB (compression 530 * is not gzip, and large blocks are not used). 531 */ 532 533 if (dmu_objset_type(os) != DMU_OST_ZFS) { 534 error = SET_ERROR(ENOTSUP); 535 } else if ((error = 536 dsl_prop_get_int_ds(dmu_objset_ds(os), 537 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 538 &propval)) == 0 && 539 !BOOTFS_COMPRESS_VALID(propval)) { 540 error = SET_ERROR(ENOTSUP); 541 } else if ((error = 542 dsl_prop_get_int_ds(dmu_objset_ds(os), 543 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 544 &propval)) == 0 && 545 propval > SPA_OLD_MAXBLOCKSIZE) { 546 error = SET_ERROR(ENOTSUP); 547 } else { 548 objnum = dmu_objset_id(os); 549 } 550 dmu_objset_rele(os, FTAG); 551 } 552 break; 553 554 case ZPOOL_PROP_FAILUREMODE: 555 error = nvpair_value_uint64(elem, &intval); 556 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 557 intval > ZIO_FAILURE_MODE_PANIC)) 558 error = SET_ERROR(EINVAL); 559 560 /* 561 * This is a special case which only occurs when 562 * the pool has completely failed. This allows 563 * the user to change the in-core failmode property 564 * without syncing it out to disk (I/Os might 565 * currently be blocked). We do this by returning 566 * EIO to the caller (spa_prop_set) to trick it 567 * into thinking we encountered a property validation 568 * error. 569 */ 570 if (!error && spa_suspended(spa)) { 571 spa->spa_failmode = intval; 572 error = SET_ERROR(EIO); 573 } 574 break; 575 576 case ZPOOL_PROP_CACHEFILE: 577 if ((error = nvpair_value_string(elem, &strval)) != 0) 578 break; 579 580 if (strval[0] == '\0') 581 break; 582 583 if (strcmp(strval, "none") == 0) 584 break; 585 586 if (strval[0] != '/') { 587 error = SET_ERROR(EINVAL); 588 break; 589 } 590 591 slash = strrchr(strval, '/'); 592 ASSERT(slash != NULL); 593 594 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 595 strcmp(slash, "/..") == 0) 596 error = SET_ERROR(EINVAL); 597 break; 598 599 case ZPOOL_PROP_COMMENT: 600 if ((error = nvpair_value_string(elem, &strval)) != 0) 601 break; 602 for (check = strval; *check != '\0'; check++) { 603 /* 604 * The kernel doesn't have an easy isprint() 605 * check. For this kernel check, we merely 606 * check ASCII apart from DEL. Fix this if 607 * there is an easy-to-use kernel isprint(). 608 */ 609 if (*check >= 0x7f) { 610 error = SET_ERROR(EINVAL); 611 break; 612 } 613 } 614 if (strlen(strval) > ZPROP_MAX_COMMENT) 615 error = E2BIG; 616 break; 617 618 case ZPOOL_PROP_DEDUPDITTO: 619 if (spa_version(spa) < SPA_VERSION_DEDUP) 620 error = SET_ERROR(ENOTSUP); 621 else 622 error = nvpair_value_uint64(elem, &intval); 623 if (error == 0 && 624 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 625 error = SET_ERROR(EINVAL); 626 break; 627 } 628 629 if (error) 630 break; 631 } 632 633 if (!error && reset_bootfs) { 634 error = nvlist_remove(props, 635 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 636 637 if (!error) { 638 error = nvlist_add_uint64(props, 639 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 640 } 641 } 642 643 return (error); 644} 645 646void 647spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 648{ 649 char *cachefile; 650 spa_config_dirent_t *dp; 651 652 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 653 &cachefile) != 0) 654 return; 655 656 dp = kmem_alloc(sizeof (spa_config_dirent_t), 657 KM_SLEEP); 658 659 if (cachefile[0] == '\0') 660 dp->scd_path = spa_strdup(spa_config_path); 661 else if (strcmp(cachefile, "none") == 0) 662 dp->scd_path = NULL; 663 else 664 dp->scd_path = spa_strdup(cachefile); 665 666 list_insert_head(&spa->spa_config_list, dp); 667 if (need_sync) 668 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 669} 670 671int 672spa_prop_set(spa_t *spa, nvlist_t *nvp) 673{ 674 int error; 675 nvpair_t *elem = NULL; 676 boolean_t need_sync = B_FALSE; 677 678 if ((error = spa_prop_validate(spa, nvp)) != 0) 679 return (error); 680 681 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 682 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 683 684 if (prop == ZPOOL_PROP_CACHEFILE || 685 prop == ZPOOL_PROP_ALTROOT || 686 prop == ZPOOL_PROP_READONLY) 687 continue; 688 689 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 690 uint64_t ver; 691 692 if (prop == ZPOOL_PROP_VERSION) { 693 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 694 } else { 695 ASSERT(zpool_prop_feature(nvpair_name(elem))); 696 ver = SPA_VERSION_FEATURES; 697 need_sync = B_TRUE; 698 } 699 700 /* Save time if the version is already set. */ 701 if (ver == spa_version(spa)) 702 continue; 703 704 /* 705 * In addition to the pool directory object, we might 706 * create the pool properties object, the features for 707 * read object, the features for write object, or the 708 * feature descriptions object. 709 */ 710 error = dsl_sync_task(spa->spa_name, NULL, 711 spa_sync_version, &ver, 712 6, ZFS_SPACE_CHECK_RESERVED); 713 if (error) 714 return (error); 715 continue; 716 } 717 718 need_sync = B_TRUE; 719 break; 720 } 721 722 if (need_sync) { 723 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 724 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 725 } 726 727 return (0); 728} 729 730/* 731 * If the bootfs property value is dsobj, clear it. 732 */ 733void 734spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 735{ 736 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 737 VERIFY(zap_remove(spa->spa_meta_objset, 738 spa->spa_pool_props_object, 739 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 740 spa->spa_bootfs = 0; 741 } 742} 743 744/*ARGSUSED*/ 745static int 746spa_change_guid_check(void *arg, dmu_tx_t *tx) 747{ 748 uint64_t *newguid = arg; 749 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 750 vdev_t *rvd = spa->spa_root_vdev; 751 uint64_t vdev_state; 752 753 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 754 vdev_state = rvd->vdev_state; 755 spa_config_exit(spa, SCL_STATE, FTAG); 756 757 if (vdev_state != VDEV_STATE_HEALTHY) 758 return (SET_ERROR(ENXIO)); 759 760 ASSERT3U(spa_guid(spa), !=, *newguid); 761 762 return (0); 763} 764 765static void 766spa_change_guid_sync(void *arg, dmu_tx_t *tx) 767{ 768 uint64_t *newguid = arg; 769 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 770 uint64_t oldguid; 771 vdev_t *rvd = spa->spa_root_vdev; 772 773 oldguid = spa_guid(spa); 774 775 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 776 rvd->vdev_guid = *newguid; 777 rvd->vdev_guid_sum += (*newguid - oldguid); 778 vdev_config_dirty(rvd); 779 spa_config_exit(spa, SCL_STATE, FTAG); 780 781 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 782 oldguid, *newguid); 783} 784 785/* 786 * Change the GUID for the pool. This is done so that we can later 787 * re-import a pool built from a clone of our own vdevs. We will modify 788 * the root vdev's guid, our own pool guid, and then mark all of our 789 * vdevs dirty. Note that we must make sure that all our vdevs are 790 * online when we do this, or else any vdevs that weren't present 791 * would be orphaned from our pool. We are also going to issue a 792 * sysevent to update any watchers. 793 */ 794int 795spa_change_guid(spa_t *spa) 796{ 797 int error; 798 uint64_t guid; 799 800 mutex_enter(&spa->spa_vdev_top_lock); 801 mutex_enter(&spa_namespace_lock); 802 guid = spa_generate_guid(NULL); 803 804 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 805 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 806 807 if (error == 0) { 808 spa_config_sync(spa, B_FALSE, B_TRUE); 809 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 810 } 811 812 mutex_exit(&spa_namespace_lock); 813 mutex_exit(&spa->spa_vdev_top_lock); 814 815 return (error); 816} 817 818/* 819 * ========================================================================== 820 * SPA state manipulation (open/create/destroy/import/export) 821 * ========================================================================== 822 */ 823 824static int 825spa_error_entry_compare(const void *a, const void *b) 826{ 827 spa_error_entry_t *sa = (spa_error_entry_t *)a; 828 spa_error_entry_t *sb = (spa_error_entry_t *)b; 829 int ret; 830 831 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 832 sizeof (zbookmark_phys_t)); 833 834 if (ret < 0) 835 return (-1); 836 else if (ret > 0) 837 return (1); 838 else 839 return (0); 840} 841 842/* 843 * Utility function which retrieves copies of the current logs and 844 * re-initializes them in the process. 845 */ 846void 847spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 848{ 849 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 850 851 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 852 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 853 854 avl_create(&spa->spa_errlist_scrub, 855 spa_error_entry_compare, sizeof (spa_error_entry_t), 856 offsetof(spa_error_entry_t, se_avl)); 857 avl_create(&spa->spa_errlist_last, 858 spa_error_entry_compare, sizeof (spa_error_entry_t), 859 offsetof(spa_error_entry_t, se_avl)); 860} 861 862static void 863spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 864{ 865 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 866 enum zti_modes mode = ztip->zti_mode; 867 uint_t value = ztip->zti_value; 868 uint_t count = ztip->zti_count; 869 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 870 char name[32]; 871 uint_t flags = 0; 872 boolean_t batch = B_FALSE; 873 874 if (mode == ZTI_MODE_NULL) { 875 tqs->stqs_count = 0; 876 tqs->stqs_taskq = NULL; 877 return; 878 } 879 880 ASSERT3U(count, >, 0); 881 882 tqs->stqs_count = count; 883 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 884 885 switch (mode) { 886 case ZTI_MODE_FIXED: 887 ASSERT3U(value, >=, 1); 888 value = MAX(value, 1); 889 break; 890 891 case ZTI_MODE_BATCH: 892 batch = B_TRUE; 893 flags |= TASKQ_THREADS_CPU_PCT; 894 value = zio_taskq_batch_pct; 895 break; 896 897 default: 898 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 899 "spa_activate()", 900 zio_type_name[t], zio_taskq_types[q], mode, value); 901 break; 902 } 903 904 for (uint_t i = 0; i < count; i++) { 905 taskq_t *tq; 906 907 if (count > 1) { 908 (void) snprintf(name, sizeof (name), "%s_%s_%u", 909 zio_type_name[t], zio_taskq_types[q], i); 910 } else { 911 (void) snprintf(name, sizeof (name), "%s_%s", 912 zio_type_name[t], zio_taskq_types[q]); 913 } 914 915#ifdef SYSDC 916 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 917 if (batch) 918 flags |= TASKQ_DC_BATCH; 919 920 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 921 spa->spa_proc, zio_taskq_basedc, flags); 922 } else { 923#endif 924 pri_t pri = maxclsyspri; 925 /* 926 * The write issue taskq can be extremely CPU 927 * intensive. Run it at slightly lower priority 928 * than the other taskqs. 929 */ 930 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 931 pri++; 932 933 tq = taskq_create_proc(name, value, pri, 50, 934 INT_MAX, spa->spa_proc, flags); 935#ifdef SYSDC 936 } 937#endif 938 939 tqs->stqs_taskq[i] = tq; 940 } 941} 942 943static void 944spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 945{ 946 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 947 948 if (tqs->stqs_taskq == NULL) { 949 ASSERT0(tqs->stqs_count); 950 return; 951 } 952 953 for (uint_t i = 0; i < tqs->stqs_count; i++) { 954 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 955 taskq_destroy(tqs->stqs_taskq[i]); 956 } 957 958 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 959 tqs->stqs_taskq = NULL; 960} 961 962/* 963 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 964 * Note that a type may have multiple discrete taskqs to avoid lock contention 965 * on the taskq itself. In that case we choose which taskq at random by using 966 * the low bits of gethrtime(). 967 */ 968void 969spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 970 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 971{ 972 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 973 taskq_t *tq; 974 975 ASSERT3P(tqs->stqs_taskq, !=, NULL); 976 ASSERT3U(tqs->stqs_count, !=, 0); 977 978 if (tqs->stqs_count == 1) { 979 tq = tqs->stqs_taskq[0]; 980 } else { 981#ifdef _KERNEL 982 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 983#else 984 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 985#endif 986 } 987 988 taskq_dispatch_ent(tq, func, arg, flags, ent); 989} 990 991static void 992spa_create_zio_taskqs(spa_t *spa) 993{ 994 for (int t = 0; t < ZIO_TYPES; t++) { 995 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 996 spa_taskqs_init(spa, t, q); 997 } 998 } 999} 1000 1001#ifdef _KERNEL 1002#ifdef SPA_PROCESS 1003static void 1004spa_thread(void *arg) 1005{ 1006 callb_cpr_t cprinfo; 1007 1008 spa_t *spa = arg; 1009 user_t *pu = PTOU(curproc); 1010 1011 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1012 spa->spa_name); 1013 1014 ASSERT(curproc != &p0); 1015 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1016 "zpool-%s", spa->spa_name); 1017 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1018 1019#ifdef PSRSET_BIND 1020 /* bind this thread to the requested psrset */ 1021 if (zio_taskq_psrset_bind != PS_NONE) { 1022 pool_lock(); 1023 mutex_enter(&cpu_lock); 1024 mutex_enter(&pidlock); 1025 mutex_enter(&curproc->p_lock); 1026 1027 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1028 0, NULL, NULL) == 0) { 1029 curthread->t_bind_pset = zio_taskq_psrset_bind; 1030 } else { 1031 cmn_err(CE_WARN, 1032 "Couldn't bind process for zfs pool \"%s\" to " 1033 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1034 } 1035 1036 mutex_exit(&curproc->p_lock); 1037 mutex_exit(&pidlock); 1038 mutex_exit(&cpu_lock); 1039 pool_unlock(); 1040 } 1041#endif 1042 1043#ifdef SYSDC 1044 if (zio_taskq_sysdc) { 1045 sysdc_thread_enter(curthread, 100, 0); 1046 } 1047#endif 1048 1049 spa->spa_proc = curproc; 1050 spa->spa_did = curthread->t_did; 1051 1052 spa_create_zio_taskqs(spa); 1053 1054 mutex_enter(&spa->spa_proc_lock); 1055 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1056 1057 spa->spa_proc_state = SPA_PROC_ACTIVE; 1058 cv_broadcast(&spa->spa_proc_cv); 1059 1060 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1061 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1062 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1063 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1064 1065 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1066 spa->spa_proc_state = SPA_PROC_GONE; 1067 spa->spa_proc = &p0; 1068 cv_broadcast(&spa->spa_proc_cv); 1069 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1070 1071 mutex_enter(&curproc->p_lock); 1072 lwp_exit(); 1073} 1074#endif /* SPA_PROCESS */ 1075#endif 1076 1077/* 1078 * Activate an uninitialized pool. 1079 */ 1080static void 1081spa_activate(spa_t *spa, int mode) 1082{ 1083 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1084 1085 spa->spa_state = POOL_STATE_ACTIVE; 1086 spa->spa_mode = mode; 1087 1088 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1089 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1090 1091 /* Try to create a covering process */ 1092 mutex_enter(&spa->spa_proc_lock); 1093 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1094 ASSERT(spa->spa_proc == &p0); 1095 spa->spa_did = 0; 1096 1097#ifdef SPA_PROCESS 1098 /* Only create a process if we're going to be around a while. */ 1099 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1100 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1101 NULL, 0) == 0) { 1102 spa->spa_proc_state = SPA_PROC_CREATED; 1103 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1104 cv_wait(&spa->spa_proc_cv, 1105 &spa->spa_proc_lock); 1106 } 1107 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1108 ASSERT(spa->spa_proc != &p0); 1109 ASSERT(spa->spa_did != 0); 1110 } else { 1111#ifdef _KERNEL 1112 cmn_err(CE_WARN, 1113 "Couldn't create process for zfs pool \"%s\"\n", 1114 spa->spa_name); 1115#endif 1116 } 1117 } 1118#endif /* SPA_PROCESS */ 1119 mutex_exit(&spa->spa_proc_lock); 1120 1121 /* If we didn't create a process, we need to create our taskqs. */ 1122 ASSERT(spa->spa_proc == &p0); 1123 if (spa->spa_proc == &p0) { 1124 spa_create_zio_taskqs(spa); 1125 } 1126 1127 /* 1128 * Start TRIM thread. 1129 */ 1130 trim_thread_create(spa); 1131 1132 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1133 offsetof(vdev_t, vdev_config_dirty_node)); 1134 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1135 offsetof(objset_t, os_evicting_node)); 1136 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1137 offsetof(vdev_t, vdev_state_dirty_node)); 1138 1139 txg_list_create(&spa->spa_vdev_txg_list, 1140 offsetof(struct vdev, vdev_txg_node)); 1141 1142 avl_create(&spa->spa_errlist_scrub, 1143 spa_error_entry_compare, sizeof (spa_error_entry_t), 1144 offsetof(spa_error_entry_t, se_avl)); 1145 avl_create(&spa->spa_errlist_last, 1146 spa_error_entry_compare, sizeof (spa_error_entry_t), 1147 offsetof(spa_error_entry_t, se_avl)); 1148} 1149 1150/* 1151 * Opposite of spa_activate(). 1152 */ 1153static void 1154spa_deactivate(spa_t *spa) 1155{ 1156 ASSERT(spa->spa_sync_on == B_FALSE); 1157 ASSERT(spa->spa_dsl_pool == NULL); 1158 ASSERT(spa->spa_root_vdev == NULL); 1159 ASSERT(spa->spa_async_zio_root == NULL); 1160 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1161 1162 /* 1163 * Stop TRIM thread in case spa_unload() wasn't called directly 1164 * before spa_deactivate(). 1165 */ 1166 trim_thread_destroy(spa); 1167 1168 spa_evicting_os_wait(spa); 1169 1170 txg_list_destroy(&spa->spa_vdev_txg_list); 1171 1172 list_destroy(&spa->spa_config_dirty_list); 1173 list_destroy(&spa->spa_evicting_os_list); 1174 list_destroy(&spa->spa_state_dirty_list); 1175 1176 for (int t = 0; t < ZIO_TYPES; t++) { 1177 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1178 spa_taskqs_fini(spa, t, q); 1179 } 1180 } 1181 1182 metaslab_class_destroy(spa->spa_normal_class); 1183 spa->spa_normal_class = NULL; 1184 1185 metaslab_class_destroy(spa->spa_log_class); 1186 spa->spa_log_class = NULL; 1187 1188 /* 1189 * If this was part of an import or the open otherwise failed, we may 1190 * still have errors left in the queues. Empty them just in case. 1191 */ 1192 spa_errlog_drain(spa); 1193 1194 avl_destroy(&spa->spa_errlist_scrub); 1195 avl_destroy(&spa->spa_errlist_last); 1196 1197 spa->spa_state = POOL_STATE_UNINITIALIZED; 1198 1199 mutex_enter(&spa->spa_proc_lock); 1200 if (spa->spa_proc_state != SPA_PROC_NONE) { 1201 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1202 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1203 cv_broadcast(&spa->spa_proc_cv); 1204 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1205 ASSERT(spa->spa_proc != &p0); 1206 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1207 } 1208 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1209 spa->spa_proc_state = SPA_PROC_NONE; 1210 } 1211 ASSERT(spa->spa_proc == &p0); 1212 mutex_exit(&spa->spa_proc_lock); 1213 1214#ifdef SPA_PROCESS 1215 /* 1216 * We want to make sure spa_thread() has actually exited the ZFS 1217 * module, so that the module can't be unloaded out from underneath 1218 * it. 1219 */ 1220 if (spa->spa_did != 0) { 1221 thread_join(spa->spa_did); 1222 spa->spa_did = 0; 1223 } 1224#endif /* SPA_PROCESS */ 1225} 1226 1227/* 1228 * Verify a pool configuration, and construct the vdev tree appropriately. This 1229 * will create all the necessary vdevs in the appropriate layout, with each vdev 1230 * in the CLOSED state. This will prep the pool before open/creation/import. 1231 * All vdev validation is done by the vdev_alloc() routine. 1232 */ 1233static int 1234spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1235 uint_t id, int atype) 1236{ 1237 nvlist_t **child; 1238 uint_t children; 1239 int error; 1240 1241 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1242 return (error); 1243 1244 if ((*vdp)->vdev_ops->vdev_op_leaf) 1245 return (0); 1246 1247 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1248 &child, &children); 1249 1250 if (error == ENOENT) 1251 return (0); 1252 1253 if (error) { 1254 vdev_free(*vdp); 1255 *vdp = NULL; 1256 return (SET_ERROR(EINVAL)); 1257 } 1258 1259 for (int c = 0; c < children; c++) { 1260 vdev_t *vd; 1261 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1262 atype)) != 0) { 1263 vdev_free(*vdp); 1264 *vdp = NULL; 1265 return (error); 1266 } 1267 } 1268 1269 ASSERT(*vdp != NULL); 1270 1271 return (0); 1272} 1273 1274/* 1275 * Opposite of spa_load(). 1276 */ 1277static void 1278spa_unload(spa_t *spa) 1279{ 1280 int i; 1281 1282 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1283 1284 /* 1285 * Stop TRIM thread. 1286 */ 1287 trim_thread_destroy(spa); 1288 1289 /* 1290 * Stop async tasks. 1291 */ 1292 spa_async_suspend(spa); 1293 1294 /* 1295 * Stop syncing. 1296 */ 1297 if (spa->spa_sync_on) { 1298 txg_sync_stop(spa->spa_dsl_pool); 1299 spa->spa_sync_on = B_FALSE; 1300 } 1301 1302 /* 1303 * Wait for any outstanding async I/O to complete. 1304 */ 1305 if (spa->spa_async_zio_root != NULL) { 1306 for (int i = 0; i < max_ncpus; i++) 1307 (void) zio_wait(spa->spa_async_zio_root[i]); 1308 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1309 spa->spa_async_zio_root = NULL; 1310 } 1311 1312 bpobj_close(&spa->spa_deferred_bpobj); 1313 1314 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1315 1316 /* 1317 * Close all vdevs. 1318 */ 1319 if (spa->spa_root_vdev) 1320 vdev_free(spa->spa_root_vdev); 1321 ASSERT(spa->spa_root_vdev == NULL); 1322 1323 /* 1324 * Close the dsl pool. 1325 */ 1326 if (spa->spa_dsl_pool) { 1327 dsl_pool_close(spa->spa_dsl_pool); 1328 spa->spa_dsl_pool = NULL; 1329 spa->spa_meta_objset = NULL; 1330 } 1331 1332 ddt_unload(spa); 1333 1334 1335 /* 1336 * Drop and purge level 2 cache 1337 */ 1338 spa_l2cache_drop(spa); 1339 1340 for (i = 0; i < spa->spa_spares.sav_count; i++) 1341 vdev_free(spa->spa_spares.sav_vdevs[i]); 1342 if (spa->spa_spares.sav_vdevs) { 1343 kmem_free(spa->spa_spares.sav_vdevs, 1344 spa->spa_spares.sav_count * sizeof (void *)); 1345 spa->spa_spares.sav_vdevs = NULL; 1346 } 1347 if (spa->spa_spares.sav_config) { 1348 nvlist_free(spa->spa_spares.sav_config); 1349 spa->spa_spares.sav_config = NULL; 1350 } 1351 spa->spa_spares.sav_count = 0; 1352 1353 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1354 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1355 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1356 } 1357 if (spa->spa_l2cache.sav_vdevs) { 1358 kmem_free(spa->spa_l2cache.sav_vdevs, 1359 spa->spa_l2cache.sav_count * sizeof (void *)); 1360 spa->spa_l2cache.sav_vdevs = NULL; 1361 } 1362 if (spa->spa_l2cache.sav_config) { 1363 nvlist_free(spa->spa_l2cache.sav_config); 1364 spa->spa_l2cache.sav_config = NULL; 1365 } 1366 spa->spa_l2cache.sav_count = 0; 1367 1368 spa->spa_async_suspended = 0; 1369 1370 if (spa->spa_comment != NULL) { 1371 spa_strfree(spa->spa_comment); 1372 spa->spa_comment = NULL; 1373 } 1374 1375 spa_config_exit(spa, SCL_ALL, FTAG); 1376} 1377 1378/* 1379 * Load (or re-load) the current list of vdevs describing the active spares for 1380 * this pool. When this is called, we have some form of basic information in 1381 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1382 * then re-generate a more complete list including status information. 1383 */ 1384static void 1385spa_load_spares(spa_t *spa) 1386{ 1387 nvlist_t **spares; 1388 uint_t nspares; 1389 int i; 1390 vdev_t *vd, *tvd; 1391 1392 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1393 1394 /* 1395 * First, close and free any existing spare vdevs. 1396 */ 1397 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1398 vd = spa->spa_spares.sav_vdevs[i]; 1399 1400 /* Undo the call to spa_activate() below */ 1401 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1402 B_FALSE)) != NULL && tvd->vdev_isspare) 1403 spa_spare_remove(tvd); 1404 vdev_close(vd); 1405 vdev_free(vd); 1406 } 1407 1408 if (spa->spa_spares.sav_vdevs) 1409 kmem_free(spa->spa_spares.sav_vdevs, 1410 spa->spa_spares.sav_count * sizeof (void *)); 1411 1412 if (spa->spa_spares.sav_config == NULL) 1413 nspares = 0; 1414 else 1415 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1416 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1417 1418 spa->spa_spares.sav_count = (int)nspares; 1419 spa->spa_spares.sav_vdevs = NULL; 1420 1421 if (nspares == 0) 1422 return; 1423 1424 /* 1425 * Construct the array of vdevs, opening them to get status in the 1426 * process. For each spare, there is potentially two different vdev_t 1427 * structures associated with it: one in the list of spares (used only 1428 * for basic validation purposes) and one in the active vdev 1429 * configuration (if it's spared in). During this phase we open and 1430 * validate each vdev on the spare list. If the vdev also exists in the 1431 * active configuration, then we also mark this vdev as an active spare. 1432 */ 1433 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1434 KM_SLEEP); 1435 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1436 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1437 VDEV_ALLOC_SPARE) == 0); 1438 ASSERT(vd != NULL); 1439 1440 spa->spa_spares.sav_vdevs[i] = vd; 1441 1442 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1443 B_FALSE)) != NULL) { 1444 if (!tvd->vdev_isspare) 1445 spa_spare_add(tvd); 1446 1447 /* 1448 * We only mark the spare active if we were successfully 1449 * able to load the vdev. Otherwise, importing a pool 1450 * with a bad active spare would result in strange 1451 * behavior, because multiple pool would think the spare 1452 * is actively in use. 1453 * 1454 * There is a vulnerability here to an equally bizarre 1455 * circumstance, where a dead active spare is later 1456 * brought back to life (onlined or otherwise). Given 1457 * the rarity of this scenario, and the extra complexity 1458 * it adds, we ignore the possibility. 1459 */ 1460 if (!vdev_is_dead(tvd)) 1461 spa_spare_activate(tvd); 1462 } 1463 1464 vd->vdev_top = vd; 1465 vd->vdev_aux = &spa->spa_spares; 1466 1467 if (vdev_open(vd) != 0) 1468 continue; 1469 1470 if (vdev_validate_aux(vd) == 0) 1471 spa_spare_add(vd); 1472 } 1473 1474 /* 1475 * Recompute the stashed list of spares, with status information 1476 * this time. 1477 */ 1478 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1479 DATA_TYPE_NVLIST_ARRAY) == 0); 1480 1481 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1482 KM_SLEEP); 1483 for (i = 0; i < spa->spa_spares.sav_count; i++) 1484 spares[i] = vdev_config_generate(spa, 1485 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1486 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1487 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1488 for (i = 0; i < spa->spa_spares.sav_count; i++) 1489 nvlist_free(spares[i]); 1490 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1491} 1492 1493/* 1494 * Load (or re-load) the current list of vdevs describing the active l2cache for 1495 * this pool. When this is called, we have some form of basic information in 1496 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1497 * then re-generate a more complete list including status information. 1498 * Devices which are already active have their details maintained, and are 1499 * not re-opened. 1500 */ 1501static void 1502spa_load_l2cache(spa_t *spa) 1503{ 1504 nvlist_t **l2cache; 1505 uint_t nl2cache; 1506 int i, j, oldnvdevs; 1507 uint64_t guid; 1508 vdev_t *vd, **oldvdevs, **newvdevs; 1509 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1510 1511 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1512 1513 if (sav->sav_config != NULL) { 1514 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1515 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1516 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1517 } else { 1518 nl2cache = 0; 1519 newvdevs = NULL; 1520 } 1521 1522 oldvdevs = sav->sav_vdevs; 1523 oldnvdevs = sav->sav_count; 1524 sav->sav_vdevs = NULL; 1525 sav->sav_count = 0; 1526 1527 /* 1528 * Process new nvlist of vdevs. 1529 */ 1530 for (i = 0; i < nl2cache; i++) { 1531 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1532 &guid) == 0); 1533 1534 newvdevs[i] = NULL; 1535 for (j = 0; j < oldnvdevs; j++) { 1536 vd = oldvdevs[j]; 1537 if (vd != NULL && guid == vd->vdev_guid) { 1538 /* 1539 * Retain previous vdev for add/remove ops. 1540 */ 1541 newvdevs[i] = vd; 1542 oldvdevs[j] = NULL; 1543 break; 1544 } 1545 } 1546 1547 if (newvdevs[i] == NULL) { 1548 /* 1549 * Create new vdev 1550 */ 1551 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1552 VDEV_ALLOC_L2CACHE) == 0); 1553 ASSERT(vd != NULL); 1554 newvdevs[i] = vd; 1555 1556 /* 1557 * Commit this vdev as an l2cache device, 1558 * even if it fails to open. 1559 */ 1560 spa_l2cache_add(vd); 1561 1562 vd->vdev_top = vd; 1563 vd->vdev_aux = sav; 1564 1565 spa_l2cache_activate(vd); 1566 1567 if (vdev_open(vd) != 0) 1568 continue; 1569 1570 (void) vdev_validate_aux(vd); 1571 1572 if (!vdev_is_dead(vd)) 1573 l2arc_add_vdev(spa, vd); 1574 } 1575 } 1576 1577 /* 1578 * Purge vdevs that were dropped 1579 */ 1580 for (i = 0; i < oldnvdevs; i++) { 1581 uint64_t pool; 1582 1583 vd = oldvdevs[i]; 1584 if (vd != NULL) { 1585 ASSERT(vd->vdev_isl2cache); 1586 1587 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1588 pool != 0ULL && l2arc_vdev_present(vd)) 1589 l2arc_remove_vdev(vd); 1590 vdev_clear_stats(vd); 1591 vdev_free(vd); 1592 } 1593 } 1594 1595 if (oldvdevs) 1596 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1597 1598 if (sav->sav_config == NULL) 1599 goto out; 1600 1601 sav->sav_vdevs = newvdevs; 1602 sav->sav_count = (int)nl2cache; 1603 1604 /* 1605 * Recompute the stashed list of l2cache devices, with status 1606 * information this time. 1607 */ 1608 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1609 DATA_TYPE_NVLIST_ARRAY) == 0); 1610 1611 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1612 for (i = 0; i < sav->sav_count; i++) 1613 l2cache[i] = vdev_config_generate(spa, 1614 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1615 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1616 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1617out: 1618 for (i = 0; i < sav->sav_count; i++) 1619 nvlist_free(l2cache[i]); 1620 if (sav->sav_count) 1621 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1622} 1623 1624static int 1625load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1626{ 1627 dmu_buf_t *db; 1628 char *packed = NULL; 1629 size_t nvsize = 0; 1630 int error; 1631 *value = NULL; 1632 1633 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1634 if (error != 0) 1635 return (error); 1636 1637 nvsize = *(uint64_t *)db->db_data; 1638 dmu_buf_rele(db, FTAG); 1639 1640 packed = kmem_alloc(nvsize, KM_SLEEP); 1641 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1642 DMU_READ_PREFETCH); 1643 if (error == 0) 1644 error = nvlist_unpack(packed, nvsize, value, 0); 1645 kmem_free(packed, nvsize); 1646 1647 return (error); 1648} 1649 1650/* 1651 * Checks to see if the given vdev could not be opened, in which case we post a 1652 * sysevent to notify the autoreplace code that the device has been removed. 1653 */ 1654static void 1655spa_check_removed(vdev_t *vd) 1656{ 1657 for (int c = 0; c < vd->vdev_children; c++) 1658 spa_check_removed(vd->vdev_child[c]); 1659 1660 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1661 !vd->vdev_ishole) { 1662 zfs_post_autoreplace(vd->vdev_spa, vd); 1663 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1664 } 1665} 1666 1667/* 1668 * Validate the current config against the MOS config 1669 */ 1670static boolean_t 1671spa_config_valid(spa_t *spa, nvlist_t *config) 1672{ 1673 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1674 nvlist_t *nv; 1675 1676 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1677 1678 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1679 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1680 1681 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1682 1683 /* 1684 * If we're doing a normal import, then build up any additional 1685 * diagnostic information about missing devices in this config. 1686 * We'll pass this up to the user for further processing. 1687 */ 1688 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1689 nvlist_t **child, *nv; 1690 uint64_t idx = 0; 1691 1692 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1693 KM_SLEEP); 1694 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1695 1696 for (int c = 0; c < rvd->vdev_children; c++) { 1697 vdev_t *tvd = rvd->vdev_child[c]; 1698 vdev_t *mtvd = mrvd->vdev_child[c]; 1699 1700 if (tvd->vdev_ops == &vdev_missing_ops && 1701 mtvd->vdev_ops != &vdev_missing_ops && 1702 mtvd->vdev_islog) 1703 child[idx++] = vdev_config_generate(spa, mtvd, 1704 B_FALSE, 0); 1705 } 1706 1707 if (idx) { 1708 VERIFY(nvlist_add_nvlist_array(nv, 1709 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1710 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1711 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1712 1713 for (int i = 0; i < idx; i++) 1714 nvlist_free(child[i]); 1715 } 1716 nvlist_free(nv); 1717 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1718 } 1719 1720 /* 1721 * Compare the root vdev tree with the information we have 1722 * from the MOS config (mrvd). Check each top-level vdev 1723 * with the corresponding MOS config top-level (mtvd). 1724 */ 1725 for (int c = 0; c < rvd->vdev_children; c++) { 1726 vdev_t *tvd = rvd->vdev_child[c]; 1727 vdev_t *mtvd = mrvd->vdev_child[c]; 1728 1729 /* 1730 * Resolve any "missing" vdevs in the current configuration. 1731 * If we find that the MOS config has more accurate information 1732 * about the top-level vdev then use that vdev instead. 1733 */ 1734 if (tvd->vdev_ops == &vdev_missing_ops && 1735 mtvd->vdev_ops != &vdev_missing_ops) { 1736 1737 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1738 continue; 1739 1740 /* 1741 * Device specific actions. 1742 */ 1743 if (mtvd->vdev_islog) { 1744 spa_set_log_state(spa, SPA_LOG_CLEAR); 1745 } else { 1746 /* 1747 * XXX - once we have 'readonly' pool 1748 * support we should be able to handle 1749 * missing data devices by transitioning 1750 * the pool to readonly. 1751 */ 1752 continue; 1753 } 1754 1755 /* 1756 * Swap the missing vdev with the data we were 1757 * able to obtain from the MOS config. 1758 */ 1759 vdev_remove_child(rvd, tvd); 1760 vdev_remove_child(mrvd, mtvd); 1761 1762 vdev_add_child(rvd, mtvd); 1763 vdev_add_child(mrvd, tvd); 1764 1765 spa_config_exit(spa, SCL_ALL, FTAG); 1766 vdev_load(mtvd); 1767 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1768 1769 vdev_reopen(rvd); 1770 } else if (mtvd->vdev_islog) { 1771 /* 1772 * Load the slog device's state from the MOS config 1773 * since it's possible that the label does not 1774 * contain the most up-to-date information. 1775 */ 1776 vdev_load_log_state(tvd, mtvd); 1777 vdev_reopen(tvd); 1778 } 1779 } 1780 vdev_free(mrvd); 1781 spa_config_exit(spa, SCL_ALL, FTAG); 1782 1783 /* 1784 * Ensure we were able to validate the config. 1785 */ 1786 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1787} 1788 1789/* 1790 * Check for missing log devices 1791 */ 1792static boolean_t 1793spa_check_logs(spa_t *spa) 1794{ 1795 boolean_t rv = B_FALSE; 1796 dsl_pool_t *dp = spa_get_dsl(spa); 1797 1798 switch (spa->spa_log_state) { 1799 case SPA_LOG_MISSING: 1800 /* need to recheck in case slog has been restored */ 1801 case SPA_LOG_UNKNOWN: 1802 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1803 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1804 if (rv) 1805 spa_set_log_state(spa, SPA_LOG_MISSING); 1806 break; 1807 } 1808 return (rv); 1809} 1810 1811static boolean_t 1812spa_passivate_log(spa_t *spa) 1813{ 1814 vdev_t *rvd = spa->spa_root_vdev; 1815 boolean_t slog_found = B_FALSE; 1816 1817 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1818 1819 if (!spa_has_slogs(spa)) 1820 return (B_FALSE); 1821 1822 for (int c = 0; c < rvd->vdev_children; c++) { 1823 vdev_t *tvd = rvd->vdev_child[c]; 1824 metaslab_group_t *mg = tvd->vdev_mg; 1825 1826 if (tvd->vdev_islog) { 1827 metaslab_group_passivate(mg); 1828 slog_found = B_TRUE; 1829 } 1830 } 1831 1832 return (slog_found); 1833} 1834 1835static void 1836spa_activate_log(spa_t *spa) 1837{ 1838 vdev_t *rvd = spa->spa_root_vdev; 1839 1840 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1841 1842 for (int c = 0; c < rvd->vdev_children; c++) { 1843 vdev_t *tvd = rvd->vdev_child[c]; 1844 metaslab_group_t *mg = tvd->vdev_mg; 1845 1846 if (tvd->vdev_islog) 1847 metaslab_group_activate(mg); 1848 } 1849} 1850 1851int 1852spa_offline_log(spa_t *spa) 1853{ 1854 int error; 1855 1856 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1857 NULL, DS_FIND_CHILDREN); 1858 if (error == 0) { 1859 /* 1860 * We successfully offlined the log device, sync out the 1861 * current txg so that the "stubby" block can be removed 1862 * by zil_sync(). 1863 */ 1864 txg_wait_synced(spa->spa_dsl_pool, 0); 1865 } 1866 return (error); 1867} 1868 1869static void 1870spa_aux_check_removed(spa_aux_vdev_t *sav) 1871{ 1872 int i; 1873 1874 for (i = 0; i < sav->sav_count; i++) 1875 spa_check_removed(sav->sav_vdevs[i]); 1876} 1877 1878void 1879spa_claim_notify(zio_t *zio) 1880{ 1881 spa_t *spa = zio->io_spa; 1882 1883 if (zio->io_error) 1884 return; 1885 1886 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1887 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1888 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1889 mutex_exit(&spa->spa_props_lock); 1890} 1891 1892typedef struct spa_load_error { 1893 uint64_t sle_meta_count; 1894 uint64_t sle_data_count; 1895} spa_load_error_t; 1896 1897static void 1898spa_load_verify_done(zio_t *zio) 1899{ 1900 blkptr_t *bp = zio->io_bp; 1901 spa_load_error_t *sle = zio->io_private; 1902 dmu_object_type_t type = BP_GET_TYPE(bp); 1903 int error = zio->io_error; 1904 spa_t *spa = zio->io_spa; 1905 1906 if (error) { 1907 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1908 type != DMU_OT_INTENT_LOG) 1909 atomic_inc_64(&sle->sle_meta_count); 1910 else 1911 atomic_inc_64(&sle->sle_data_count); 1912 } 1913 zio_data_buf_free(zio->io_data, zio->io_size); 1914 1915 mutex_enter(&spa->spa_scrub_lock); 1916 spa->spa_scrub_inflight--; 1917 cv_broadcast(&spa->spa_scrub_io_cv); 1918 mutex_exit(&spa->spa_scrub_lock); 1919} 1920 1921/* 1922 * Maximum number of concurrent scrub i/os to create while verifying 1923 * a pool while importing it. 1924 */ 1925int spa_load_verify_maxinflight = 10000; 1926boolean_t spa_load_verify_metadata = B_TRUE; 1927boolean_t spa_load_verify_data = B_TRUE; 1928 1929SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1930 &spa_load_verify_maxinflight, 0, 1931 "Maximum number of concurrent scrub I/Os to create while verifying a " 1932 "pool while importing it"); 1933 1934SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1935 &spa_load_verify_metadata, 0, 1936 "Check metadata on import?"); 1937 1938SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1939 &spa_load_verify_data, 0, 1940 "Check user data on import?"); 1941 1942/*ARGSUSED*/ 1943static int 1944spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1945 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1946{ 1947 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1948 return (0); 1949 /* 1950 * Note: normally this routine will not be called if 1951 * spa_load_verify_metadata is not set. However, it may be useful 1952 * to manually set the flag after the traversal has begun. 1953 */ 1954 if (!spa_load_verify_metadata) 1955 return (0); 1956 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1957 return (0); 1958 1959 zio_t *rio = arg; 1960 size_t size = BP_GET_PSIZE(bp); 1961 void *data = zio_data_buf_alloc(size); 1962 1963 mutex_enter(&spa->spa_scrub_lock); 1964 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1965 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1966 spa->spa_scrub_inflight++; 1967 mutex_exit(&spa->spa_scrub_lock); 1968 1969 zio_nowait(zio_read(rio, spa, bp, data, size, 1970 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1971 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1972 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1973 return (0); 1974} 1975 1976/* ARGSUSED */ 1977int 1978verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1979{ 1980 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 1981 return (SET_ERROR(ENAMETOOLONG)); 1982 1983 return (0); 1984} 1985 1986static int 1987spa_load_verify(spa_t *spa) 1988{ 1989 zio_t *rio; 1990 spa_load_error_t sle = { 0 }; 1991 zpool_rewind_policy_t policy; 1992 boolean_t verify_ok = B_FALSE; 1993 int error = 0; 1994 1995 zpool_get_rewind_policy(spa->spa_config, &policy); 1996 1997 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1998 return (0); 1999 2000 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2001 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2002 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2003 DS_FIND_CHILDREN); 2004 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2005 if (error != 0) 2006 return (error); 2007 2008 rio = zio_root(spa, NULL, &sle, 2009 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2010 2011 if (spa_load_verify_metadata) { 2012 error = traverse_pool(spa, spa->spa_verify_min_txg, 2013 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2014 spa_load_verify_cb, rio); 2015 } 2016 2017 (void) zio_wait(rio); 2018 2019 spa->spa_load_meta_errors = sle.sle_meta_count; 2020 spa->spa_load_data_errors = sle.sle_data_count; 2021 2022 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2023 sle.sle_data_count <= policy.zrp_maxdata) { 2024 int64_t loss = 0; 2025 2026 verify_ok = B_TRUE; 2027 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2028 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2029 2030 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2031 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2032 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2033 VERIFY(nvlist_add_int64(spa->spa_load_info, 2034 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2035 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2036 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2037 } else { 2038 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2039 } 2040 2041 if (error) { 2042 if (error != ENXIO && error != EIO) 2043 error = SET_ERROR(EIO); 2044 return (error); 2045 } 2046 2047 return (verify_ok ? 0 : EIO); 2048} 2049 2050/* 2051 * Find a value in the pool props object. 2052 */ 2053static void 2054spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2055{ 2056 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2057 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2058} 2059 2060/* 2061 * Find a value in the pool directory object. 2062 */ 2063static int 2064spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2065{ 2066 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2067 name, sizeof (uint64_t), 1, val)); 2068} 2069 2070static int 2071spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2072{ 2073 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2074 return (err); 2075} 2076 2077/* 2078 * Fix up config after a partly-completed split. This is done with the 2079 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2080 * pool have that entry in their config, but only the splitting one contains 2081 * a list of all the guids of the vdevs that are being split off. 2082 * 2083 * This function determines what to do with that list: either rejoin 2084 * all the disks to the pool, or complete the splitting process. To attempt 2085 * the rejoin, each disk that is offlined is marked online again, and 2086 * we do a reopen() call. If the vdev label for every disk that was 2087 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2088 * then we call vdev_split() on each disk, and complete the split. 2089 * 2090 * Otherwise we leave the config alone, with all the vdevs in place in 2091 * the original pool. 2092 */ 2093static void 2094spa_try_repair(spa_t *spa, nvlist_t *config) 2095{ 2096 uint_t extracted; 2097 uint64_t *glist; 2098 uint_t i, gcount; 2099 nvlist_t *nvl; 2100 vdev_t **vd; 2101 boolean_t attempt_reopen; 2102 2103 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2104 return; 2105 2106 /* check that the config is complete */ 2107 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2108 &glist, &gcount) != 0) 2109 return; 2110 2111 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2112 2113 /* attempt to online all the vdevs & validate */ 2114 attempt_reopen = B_TRUE; 2115 for (i = 0; i < gcount; i++) { 2116 if (glist[i] == 0) /* vdev is hole */ 2117 continue; 2118 2119 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2120 if (vd[i] == NULL) { 2121 /* 2122 * Don't bother attempting to reopen the disks; 2123 * just do the split. 2124 */ 2125 attempt_reopen = B_FALSE; 2126 } else { 2127 /* attempt to re-online it */ 2128 vd[i]->vdev_offline = B_FALSE; 2129 } 2130 } 2131 2132 if (attempt_reopen) { 2133 vdev_reopen(spa->spa_root_vdev); 2134 2135 /* check each device to see what state it's in */ 2136 for (extracted = 0, i = 0; i < gcount; i++) { 2137 if (vd[i] != NULL && 2138 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2139 break; 2140 ++extracted; 2141 } 2142 } 2143 2144 /* 2145 * If every disk has been moved to the new pool, or if we never 2146 * even attempted to look at them, then we split them off for 2147 * good. 2148 */ 2149 if (!attempt_reopen || gcount == extracted) { 2150 for (i = 0; i < gcount; i++) 2151 if (vd[i] != NULL) 2152 vdev_split(vd[i]); 2153 vdev_reopen(spa->spa_root_vdev); 2154 } 2155 2156 kmem_free(vd, gcount * sizeof (vdev_t *)); 2157} 2158 2159static int 2160spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2161 boolean_t mosconfig) 2162{ 2163 nvlist_t *config = spa->spa_config; 2164 char *ereport = FM_EREPORT_ZFS_POOL; 2165 char *comment; 2166 int error; 2167 uint64_t pool_guid; 2168 nvlist_t *nvl; 2169 2170 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2171 return (SET_ERROR(EINVAL)); 2172 2173 ASSERT(spa->spa_comment == NULL); 2174 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2175 spa->spa_comment = spa_strdup(comment); 2176 2177 /* 2178 * Versioning wasn't explicitly added to the label until later, so if 2179 * it's not present treat it as the initial version. 2180 */ 2181 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2182 &spa->spa_ubsync.ub_version) != 0) 2183 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2184 2185 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2186 &spa->spa_config_txg); 2187 2188 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2189 spa_guid_exists(pool_guid, 0)) { 2190 error = SET_ERROR(EEXIST); 2191 } else { 2192 spa->spa_config_guid = pool_guid; 2193 2194 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2195 &nvl) == 0) { 2196 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2197 KM_SLEEP) == 0); 2198 } 2199 2200 nvlist_free(spa->spa_load_info); 2201 spa->spa_load_info = fnvlist_alloc(); 2202 2203 gethrestime(&spa->spa_loaded_ts); 2204 error = spa_load_impl(spa, pool_guid, config, state, type, 2205 mosconfig, &ereport); 2206 } 2207 2208 /* 2209 * Don't count references from objsets that are already closed 2210 * and are making their way through the eviction process. 2211 */ 2212 spa_evicting_os_wait(spa); 2213 spa->spa_minref = refcount_count(&spa->spa_refcount); 2214 if (error) { 2215 if (error != EEXIST) { 2216 spa->spa_loaded_ts.tv_sec = 0; 2217 spa->spa_loaded_ts.tv_nsec = 0; 2218 } 2219 if (error != EBADF) { 2220 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2221 } 2222 } 2223 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2224 spa->spa_ena = 0; 2225 2226 return (error); 2227} 2228 2229/* 2230 * Load an existing storage pool, using the pool's builtin spa_config as a 2231 * source of configuration information. 2232 */ 2233static int 2234spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2235 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2236 char **ereport) 2237{ 2238 int error = 0; 2239 nvlist_t *nvroot = NULL; 2240 nvlist_t *label; 2241 vdev_t *rvd; 2242 uberblock_t *ub = &spa->spa_uberblock; 2243 uint64_t children, config_cache_txg = spa->spa_config_txg; 2244 int orig_mode = spa->spa_mode; 2245 int parse; 2246 uint64_t obj; 2247 boolean_t missing_feat_write = B_FALSE; 2248 2249 /* 2250 * If this is an untrusted config, access the pool in read-only mode. 2251 * This prevents things like resilvering recently removed devices. 2252 */ 2253 if (!mosconfig) 2254 spa->spa_mode = FREAD; 2255 2256 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2257 2258 spa->spa_load_state = state; 2259 2260 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2261 return (SET_ERROR(EINVAL)); 2262 2263 parse = (type == SPA_IMPORT_EXISTING ? 2264 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2265 2266 /* 2267 * Create "The Godfather" zio to hold all async IOs 2268 */ 2269 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2270 KM_SLEEP); 2271 for (int i = 0; i < max_ncpus; i++) { 2272 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2273 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2274 ZIO_FLAG_GODFATHER); 2275 } 2276 2277 /* 2278 * Parse the configuration into a vdev tree. We explicitly set the 2279 * value that will be returned by spa_version() since parsing the 2280 * configuration requires knowing the version number. 2281 */ 2282 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2283 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2284 spa_config_exit(spa, SCL_ALL, FTAG); 2285 2286 if (error != 0) 2287 return (error); 2288 2289 ASSERT(spa->spa_root_vdev == rvd); 2290 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2291 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2292 2293 if (type != SPA_IMPORT_ASSEMBLE) { 2294 ASSERT(spa_guid(spa) == pool_guid); 2295 } 2296 2297 /* 2298 * Try to open all vdevs, loading each label in the process. 2299 */ 2300 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2301 error = vdev_open(rvd); 2302 spa_config_exit(spa, SCL_ALL, FTAG); 2303 if (error != 0) 2304 return (error); 2305 2306 /* 2307 * We need to validate the vdev labels against the configuration that 2308 * we have in hand, which is dependent on the setting of mosconfig. If 2309 * mosconfig is true then we're validating the vdev labels based on 2310 * that config. Otherwise, we're validating against the cached config 2311 * (zpool.cache) that was read when we loaded the zfs module, and then 2312 * later we will recursively call spa_load() and validate against 2313 * the vdev config. 2314 * 2315 * If we're assembling a new pool that's been split off from an 2316 * existing pool, the labels haven't yet been updated so we skip 2317 * validation for now. 2318 */ 2319 if (type != SPA_IMPORT_ASSEMBLE) { 2320 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2321 error = vdev_validate(rvd, mosconfig); 2322 spa_config_exit(spa, SCL_ALL, FTAG); 2323 2324 if (error != 0) 2325 return (error); 2326 2327 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2328 return (SET_ERROR(ENXIO)); 2329 } 2330 2331 /* 2332 * Find the best uberblock. 2333 */ 2334 vdev_uberblock_load(rvd, ub, &label); 2335 2336 /* 2337 * If we weren't able to find a single valid uberblock, return failure. 2338 */ 2339 if (ub->ub_txg == 0) { 2340 nvlist_free(label); 2341 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2342 } 2343 2344 /* 2345 * If the pool has an unsupported version we can't open it. 2346 */ 2347 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2348 nvlist_free(label); 2349 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2350 } 2351 2352 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2353 nvlist_t *features; 2354 2355 /* 2356 * If we weren't able to find what's necessary for reading the 2357 * MOS in the label, return failure. 2358 */ 2359 if (label == NULL || nvlist_lookup_nvlist(label, 2360 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2361 nvlist_free(label); 2362 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2363 ENXIO)); 2364 } 2365 2366 /* 2367 * Update our in-core representation with the definitive values 2368 * from the label. 2369 */ 2370 nvlist_free(spa->spa_label_features); 2371 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2372 } 2373 2374 nvlist_free(label); 2375 2376 /* 2377 * Look through entries in the label nvlist's features_for_read. If 2378 * there is a feature listed there which we don't understand then we 2379 * cannot open a pool. 2380 */ 2381 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2382 nvlist_t *unsup_feat; 2383 2384 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2385 0); 2386 2387 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2388 NULL); nvp != NULL; 2389 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2390 if (!zfeature_is_supported(nvpair_name(nvp))) { 2391 VERIFY(nvlist_add_string(unsup_feat, 2392 nvpair_name(nvp), "") == 0); 2393 } 2394 } 2395 2396 if (!nvlist_empty(unsup_feat)) { 2397 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2398 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2399 nvlist_free(unsup_feat); 2400 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2401 ENOTSUP)); 2402 } 2403 2404 nvlist_free(unsup_feat); 2405 } 2406 2407 /* 2408 * If the vdev guid sum doesn't match the uberblock, we have an 2409 * incomplete configuration. We first check to see if the pool 2410 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2411 * If it is, defer the vdev_guid_sum check till later so we 2412 * can handle missing vdevs. 2413 */ 2414 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2415 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2416 rvd->vdev_guid_sum != ub->ub_guid_sum) 2417 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2418 2419 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2420 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2421 spa_try_repair(spa, config); 2422 spa_config_exit(spa, SCL_ALL, FTAG); 2423 nvlist_free(spa->spa_config_splitting); 2424 spa->spa_config_splitting = NULL; 2425 } 2426 2427 /* 2428 * Initialize internal SPA structures. 2429 */ 2430 spa->spa_state = POOL_STATE_ACTIVE; 2431 spa->spa_ubsync = spa->spa_uberblock; 2432 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2433 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2434 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2435 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2436 spa->spa_claim_max_txg = spa->spa_first_txg; 2437 spa->spa_prev_software_version = ub->ub_software_version; 2438 2439 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2440 if (error) 2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2442 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2443 2444 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2445 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2446 2447 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2448 boolean_t missing_feat_read = B_FALSE; 2449 nvlist_t *unsup_feat, *enabled_feat; 2450 2451 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2452 &spa->spa_feat_for_read_obj) != 0) { 2453 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2454 } 2455 2456 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2457 &spa->spa_feat_for_write_obj) != 0) { 2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2459 } 2460 2461 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2462 &spa->spa_feat_desc_obj) != 0) { 2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2464 } 2465 2466 enabled_feat = fnvlist_alloc(); 2467 unsup_feat = fnvlist_alloc(); 2468 2469 if (!spa_features_check(spa, B_FALSE, 2470 unsup_feat, enabled_feat)) 2471 missing_feat_read = B_TRUE; 2472 2473 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2474 if (!spa_features_check(spa, B_TRUE, 2475 unsup_feat, enabled_feat)) { 2476 missing_feat_write = B_TRUE; 2477 } 2478 } 2479 2480 fnvlist_add_nvlist(spa->spa_load_info, 2481 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2482 2483 if (!nvlist_empty(unsup_feat)) { 2484 fnvlist_add_nvlist(spa->spa_load_info, 2485 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2486 } 2487 2488 fnvlist_free(enabled_feat); 2489 fnvlist_free(unsup_feat); 2490 2491 if (!missing_feat_read) { 2492 fnvlist_add_boolean(spa->spa_load_info, 2493 ZPOOL_CONFIG_CAN_RDONLY); 2494 } 2495 2496 /* 2497 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2498 * twofold: to determine whether the pool is available for 2499 * import in read-write mode and (if it is not) whether the 2500 * pool is available for import in read-only mode. If the pool 2501 * is available for import in read-write mode, it is displayed 2502 * as available in userland; if it is not available for import 2503 * in read-only mode, it is displayed as unavailable in 2504 * userland. If the pool is available for import in read-only 2505 * mode but not read-write mode, it is displayed as unavailable 2506 * in userland with a special note that the pool is actually 2507 * available for open in read-only mode. 2508 * 2509 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2510 * missing a feature for write, we must first determine whether 2511 * the pool can be opened read-only before returning to 2512 * userland in order to know whether to display the 2513 * abovementioned note. 2514 */ 2515 if (missing_feat_read || (missing_feat_write && 2516 spa_writeable(spa))) { 2517 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2518 ENOTSUP)); 2519 } 2520 2521 /* 2522 * Load refcounts for ZFS features from disk into an in-memory 2523 * cache during SPA initialization. 2524 */ 2525 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2526 uint64_t refcount; 2527 2528 error = feature_get_refcount_from_disk(spa, 2529 &spa_feature_table[i], &refcount); 2530 if (error == 0) { 2531 spa->spa_feat_refcount_cache[i] = refcount; 2532 } else if (error == ENOTSUP) { 2533 spa->spa_feat_refcount_cache[i] = 2534 SPA_FEATURE_DISABLED; 2535 } else { 2536 return (spa_vdev_err(rvd, 2537 VDEV_AUX_CORRUPT_DATA, EIO)); 2538 } 2539 } 2540 } 2541 2542 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2543 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2544 &spa->spa_feat_enabled_txg_obj) != 0) 2545 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2546 } 2547 2548 spa->spa_is_initializing = B_TRUE; 2549 error = dsl_pool_open(spa->spa_dsl_pool); 2550 spa->spa_is_initializing = B_FALSE; 2551 if (error != 0) 2552 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2553 2554 if (!mosconfig) { 2555 uint64_t hostid; 2556 nvlist_t *policy = NULL, *nvconfig; 2557 2558 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2559 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2560 2561 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2562 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2563 char *hostname; 2564 unsigned long myhostid = 0; 2565 2566 VERIFY(nvlist_lookup_string(nvconfig, 2567 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2568 2569#ifdef _KERNEL 2570 myhostid = zone_get_hostid(NULL); 2571#else /* _KERNEL */ 2572 /* 2573 * We're emulating the system's hostid in userland, so 2574 * we can't use zone_get_hostid(). 2575 */ 2576 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2577#endif /* _KERNEL */ 2578 if (check_hostid && hostid != 0 && myhostid != 0 && 2579 hostid != myhostid) { 2580 nvlist_free(nvconfig); 2581 cmn_err(CE_WARN, "pool '%s' could not be " 2582 "loaded as it was last accessed by " 2583 "another system (host: %s hostid: 0x%lx). " 2584 "See: http://illumos.org/msg/ZFS-8000-EY", 2585 spa_name(spa), hostname, 2586 (unsigned long)hostid); 2587 return (SET_ERROR(EBADF)); 2588 } 2589 } 2590 if (nvlist_lookup_nvlist(spa->spa_config, 2591 ZPOOL_REWIND_POLICY, &policy) == 0) 2592 VERIFY(nvlist_add_nvlist(nvconfig, 2593 ZPOOL_REWIND_POLICY, policy) == 0); 2594 2595 spa_config_set(spa, nvconfig); 2596 spa_unload(spa); 2597 spa_deactivate(spa); 2598 spa_activate(spa, orig_mode); 2599 2600 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2601 } 2602 2603 /* Grab the secret checksum salt from the MOS. */ 2604 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2605 DMU_POOL_CHECKSUM_SALT, 1, 2606 sizeof (spa->spa_cksum_salt.zcs_bytes), 2607 spa->spa_cksum_salt.zcs_bytes); 2608 if (error == ENOENT) { 2609 /* Generate a new salt for subsequent use */ 2610 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 2611 sizeof (spa->spa_cksum_salt.zcs_bytes)); 2612 } else if (error != 0) { 2613 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2614 } 2615 2616 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2617 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2618 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2619 if (error != 0) 2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2621 2622 /* 2623 * Load the bit that tells us to use the new accounting function 2624 * (raid-z deflation). If we have an older pool, this will not 2625 * be present. 2626 */ 2627 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2628 if (error != 0 && error != ENOENT) 2629 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2630 2631 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2632 &spa->spa_creation_version); 2633 if (error != 0 && error != ENOENT) 2634 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2635 2636 /* 2637 * Load the persistent error log. If we have an older pool, this will 2638 * not be present. 2639 */ 2640 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2641 if (error != 0 && error != ENOENT) 2642 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2643 2644 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2645 &spa->spa_errlog_scrub); 2646 if (error != 0 && error != ENOENT) 2647 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2648 2649 /* 2650 * Load the history object. If we have an older pool, this 2651 * will not be present. 2652 */ 2653 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2654 if (error != 0 && error != ENOENT) 2655 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2656 2657 /* 2658 * If we're assembling the pool from the split-off vdevs of 2659 * an existing pool, we don't want to attach the spares & cache 2660 * devices. 2661 */ 2662 2663 /* 2664 * Load any hot spares for this pool. 2665 */ 2666 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2667 if (error != 0 && error != ENOENT) 2668 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2669 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2670 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2671 if (load_nvlist(spa, spa->spa_spares.sav_object, 2672 &spa->spa_spares.sav_config) != 0) 2673 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2674 2675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2676 spa_load_spares(spa); 2677 spa_config_exit(spa, SCL_ALL, FTAG); 2678 } else if (error == 0) { 2679 spa->spa_spares.sav_sync = B_TRUE; 2680 } 2681 2682 /* 2683 * Load any level 2 ARC devices for this pool. 2684 */ 2685 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2686 &spa->spa_l2cache.sav_object); 2687 if (error != 0 && error != ENOENT) 2688 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2689 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2690 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2691 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2692 &spa->spa_l2cache.sav_config) != 0) 2693 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2694 2695 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2696 spa_load_l2cache(spa); 2697 spa_config_exit(spa, SCL_ALL, FTAG); 2698 } else if (error == 0) { 2699 spa->spa_l2cache.sav_sync = B_TRUE; 2700 } 2701 2702 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2703 2704 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2705 if (error && error != ENOENT) 2706 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2707 2708 if (error == 0) { 2709 uint64_t autoreplace; 2710 2711 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2712 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2713 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2714 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2715 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2716 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2717 &spa->spa_dedup_ditto); 2718 2719 spa->spa_autoreplace = (autoreplace != 0); 2720 } 2721 2722 /* 2723 * If the 'autoreplace' property is set, then post a resource notifying 2724 * the ZFS DE that it should not issue any faults for unopenable 2725 * devices. We also iterate over the vdevs, and post a sysevent for any 2726 * unopenable vdevs so that the normal autoreplace handler can take 2727 * over. 2728 */ 2729 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2730 spa_check_removed(spa->spa_root_vdev); 2731 /* 2732 * For the import case, this is done in spa_import(), because 2733 * at this point we're using the spare definitions from 2734 * the MOS config, not necessarily from the userland config. 2735 */ 2736 if (state != SPA_LOAD_IMPORT) { 2737 spa_aux_check_removed(&spa->spa_spares); 2738 spa_aux_check_removed(&spa->spa_l2cache); 2739 } 2740 } 2741 2742 /* 2743 * Load the vdev state for all toplevel vdevs. 2744 */ 2745 vdev_load(rvd); 2746 2747 /* 2748 * Propagate the leaf DTLs we just loaded all the way up the tree. 2749 */ 2750 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2751 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2752 spa_config_exit(spa, SCL_ALL, FTAG); 2753 2754 /* 2755 * Load the DDTs (dedup tables). 2756 */ 2757 error = ddt_load(spa); 2758 if (error != 0) 2759 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2760 2761 spa_update_dspace(spa); 2762 2763 /* 2764 * Validate the config, using the MOS config to fill in any 2765 * information which might be missing. If we fail to validate 2766 * the config then declare the pool unfit for use. If we're 2767 * assembling a pool from a split, the log is not transferred 2768 * over. 2769 */ 2770 if (type != SPA_IMPORT_ASSEMBLE) { 2771 nvlist_t *nvconfig; 2772 2773 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2774 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2775 2776 if (!spa_config_valid(spa, nvconfig)) { 2777 nvlist_free(nvconfig); 2778 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2779 ENXIO)); 2780 } 2781 nvlist_free(nvconfig); 2782 2783 /* 2784 * Now that we've validated the config, check the state of the 2785 * root vdev. If it can't be opened, it indicates one or 2786 * more toplevel vdevs are faulted. 2787 */ 2788 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2789 return (SET_ERROR(ENXIO)); 2790 2791 if (spa_writeable(spa) && spa_check_logs(spa)) { 2792 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2793 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2794 } 2795 } 2796 2797 if (missing_feat_write) { 2798 ASSERT(state == SPA_LOAD_TRYIMPORT); 2799 2800 /* 2801 * At this point, we know that we can open the pool in 2802 * read-only mode but not read-write mode. We now have enough 2803 * information and can return to userland. 2804 */ 2805 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2806 } 2807 2808 /* 2809 * We've successfully opened the pool, verify that we're ready 2810 * to start pushing transactions. 2811 */ 2812 if (state != SPA_LOAD_TRYIMPORT) { 2813 if (error = spa_load_verify(spa)) 2814 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2815 error)); 2816 } 2817 2818 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2819 spa->spa_load_max_txg == UINT64_MAX)) { 2820 dmu_tx_t *tx; 2821 int need_update = B_FALSE; 2822 dsl_pool_t *dp = spa_get_dsl(spa); 2823 2824 ASSERT(state != SPA_LOAD_TRYIMPORT); 2825 2826 /* 2827 * Claim log blocks that haven't been committed yet. 2828 * This must all happen in a single txg. 2829 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2830 * invoked from zil_claim_log_block()'s i/o done callback. 2831 * Price of rollback is that we abandon the log. 2832 */ 2833 spa->spa_claiming = B_TRUE; 2834 2835 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2836 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2837 zil_claim, tx, DS_FIND_CHILDREN); 2838 dmu_tx_commit(tx); 2839 2840 spa->spa_claiming = B_FALSE; 2841 2842 spa_set_log_state(spa, SPA_LOG_GOOD); 2843 spa->spa_sync_on = B_TRUE; 2844 txg_sync_start(spa->spa_dsl_pool); 2845 2846 /* 2847 * Wait for all claims to sync. We sync up to the highest 2848 * claimed log block birth time so that claimed log blocks 2849 * don't appear to be from the future. spa_claim_max_txg 2850 * will have been set for us by either zil_check_log_chain() 2851 * (invoked from spa_check_logs()) or zil_claim() above. 2852 */ 2853 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2854 2855 /* 2856 * If the config cache is stale, or we have uninitialized 2857 * metaslabs (see spa_vdev_add()), then update the config. 2858 * 2859 * If this is a verbatim import, trust the current 2860 * in-core spa_config and update the disk labels. 2861 */ 2862 if (config_cache_txg != spa->spa_config_txg || 2863 state == SPA_LOAD_IMPORT || 2864 state == SPA_LOAD_RECOVER || 2865 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2866 need_update = B_TRUE; 2867 2868 for (int c = 0; c < rvd->vdev_children; c++) 2869 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2870 need_update = B_TRUE; 2871 2872 /* 2873 * Update the config cache asychronously in case we're the 2874 * root pool, in which case the config cache isn't writable yet. 2875 */ 2876 if (need_update) 2877 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2878 2879 /* 2880 * Check all DTLs to see if anything needs resilvering. 2881 */ 2882 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2883 vdev_resilver_needed(rvd, NULL, NULL)) 2884 spa_async_request(spa, SPA_ASYNC_RESILVER); 2885 2886 /* 2887 * Log the fact that we booted up (so that we can detect if 2888 * we rebooted in the middle of an operation). 2889 */ 2890 spa_history_log_version(spa, "open"); 2891 2892 /* 2893 * Delete any inconsistent datasets. 2894 */ 2895 (void) dmu_objset_find(spa_name(spa), 2896 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2897 2898 /* 2899 * Clean up any stale temporary dataset userrefs. 2900 */ 2901 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2902 } 2903 2904 return (0); 2905} 2906 2907static int 2908spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2909{ 2910 int mode = spa->spa_mode; 2911 2912 spa_unload(spa); 2913 spa_deactivate(spa); 2914 2915 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2916 2917 spa_activate(spa, mode); 2918 spa_async_suspend(spa); 2919 2920 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2921} 2922 2923/* 2924 * If spa_load() fails this function will try loading prior txg's. If 2925 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2926 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2927 * function will not rewind the pool and will return the same error as 2928 * spa_load(). 2929 */ 2930static int 2931spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2932 uint64_t max_request, int rewind_flags) 2933{ 2934 nvlist_t *loadinfo = NULL; 2935 nvlist_t *config = NULL; 2936 int load_error, rewind_error; 2937 uint64_t safe_rewind_txg; 2938 uint64_t min_txg; 2939 2940 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2941 spa->spa_load_max_txg = spa->spa_load_txg; 2942 spa_set_log_state(spa, SPA_LOG_CLEAR); 2943 } else { 2944 spa->spa_load_max_txg = max_request; 2945 if (max_request != UINT64_MAX) 2946 spa->spa_extreme_rewind = B_TRUE; 2947 } 2948 2949 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2950 mosconfig); 2951 if (load_error == 0) 2952 return (0); 2953 2954 if (spa->spa_root_vdev != NULL) 2955 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2956 2957 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2958 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2959 2960 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2961 nvlist_free(config); 2962 return (load_error); 2963 } 2964 2965 if (state == SPA_LOAD_RECOVER) { 2966 /* Price of rolling back is discarding txgs, including log */ 2967 spa_set_log_state(spa, SPA_LOG_CLEAR); 2968 } else { 2969 /* 2970 * If we aren't rolling back save the load info from our first 2971 * import attempt so that we can restore it after attempting 2972 * to rewind. 2973 */ 2974 loadinfo = spa->spa_load_info; 2975 spa->spa_load_info = fnvlist_alloc(); 2976 } 2977 2978 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2979 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2980 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2981 TXG_INITIAL : safe_rewind_txg; 2982 2983 /* 2984 * Continue as long as we're finding errors, we're still within 2985 * the acceptable rewind range, and we're still finding uberblocks 2986 */ 2987 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2988 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2989 if (spa->spa_load_max_txg < safe_rewind_txg) 2990 spa->spa_extreme_rewind = B_TRUE; 2991 rewind_error = spa_load_retry(spa, state, mosconfig); 2992 } 2993 2994 spa->spa_extreme_rewind = B_FALSE; 2995 spa->spa_load_max_txg = UINT64_MAX; 2996 2997 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2998 spa_config_set(spa, config); 2999 3000 if (state == SPA_LOAD_RECOVER) { 3001 ASSERT3P(loadinfo, ==, NULL); 3002 return (rewind_error); 3003 } else { 3004 /* Store the rewind info as part of the initial load info */ 3005 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3006 spa->spa_load_info); 3007 3008 /* Restore the initial load info */ 3009 fnvlist_free(spa->spa_load_info); 3010 spa->spa_load_info = loadinfo; 3011 3012 return (load_error); 3013 } 3014} 3015 3016/* 3017 * Pool Open/Import 3018 * 3019 * The import case is identical to an open except that the configuration is sent 3020 * down from userland, instead of grabbed from the configuration cache. For the 3021 * case of an open, the pool configuration will exist in the 3022 * POOL_STATE_UNINITIALIZED state. 3023 * 3024 * The stats information (gen/count/ustats) is used to gather vdev statistics at 3025 * the same time open the pool, without having to keep around the spa_t in some 3026 * ambiguous state. 3027 */ 3028static int 3029spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3030 nvlist_t **config) 3031{ 3032 spa_t *spa; 3033 spa_load_state_t state = SPA_LOAD_OPEN; 3034 int error; 3035 int locked = B_FALSE; 3036 int firstopen = B_FALSE; 3037 3038 *spapp = NULL; 3039 3040 /* 3041 * As disgusting as this is, we need to support recursive calls to this 3042 * function because dsl_dir_open() is called during spa_load(), and ends 3043 * up calling spa_open() again. The real fix is to figure out how to 3044 * avoid dsl_dir_open() calling this in the first place. 3045 */ 3046 if (mutex_owner(&spa_namespace_lock) != curthread) { 3047 mutex_enter(&spa_namespace_lock); 3048 locked = B_TRUE; 3049 } 3050 3051 if ((spa = spa_lookup(pool)) == NULL) { 3052 if (locked) 3053 mutex_exit(&spa_namespace_lock); 3054 return (SET_ERROR(ENOENT)); 3055 } 3056 3057 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3058 zpool_rewind_policy_t policy; 3059 3060 firstopen = B_TRUE; 3061 3062 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3063 &policy); 3064 if (policy.zrp_request & ZPOOL_DO_REWIND) 3065 state = SPA_LOAD_RECOVER; 3066 3067 spa_activate(spa, spa_mode_global); 3068 3069 if (state != SPA_LOAD_RECOVER) 3070 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3071 3072 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3073 policy.zrp_request); 3074 3075 if (error == EBADF) { 3076 /* 3077 * If vdev_validate() returns failure (indicated by 3078 * EBADF), it indicates that one of the vdevs indicates 3079 * that the pool has been exported or destroyed. If 3080 * this is the case, the config cache is out of sync and 3081 * we should remove the pool from the namespace. 3082 */ 3083 spa_unload(spa); 3084 spa_deactivate(spa); 3085 spa_config_sync(spa, B_TRUE, B_TRUE); 3086 spa_remove(spa); 3087 if (locked) 3088 mutex_exit(&spa_namespace_lock); 3089 return (SET_ERROR(ENOENT)); 3090 } 3091 3092 if (error) { 3093 /* 3094 * We can't open the pool, but we still have useful 3095 * information: the state of each vdev after the 3096 * attempted vdev_open(). Return this to the user. 3097 */ 3098 if (config != NULL && spa->spa_config) { 3099 VERIFY(nvlist_dup(spa->spa_config, config, 3100 KM_SLEEP) == 0); 3101 VERIFY(nvlist_add_nvlist(*config, 3102 ZPOOL_CONFIG_LOAD_INFO, 3103 spa->spa_load_info) == 0); 3104 } 3105 spa_unload(spa); 3106 spa_deactivate(spa); 3107 spa->spa_last_open_failed = error; 3108 if (locked) 3109 mutex_exit(&spa_namespace_lock); 3110 *spapp = NULL; 3111 return (error); 3112 } 3113 } 3114 3115 spa_open_ref(spa, tag); 3116 3117 if (config != NULL) 3118 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3119 3120 /* 3121 * If we've recovered the pool, pass back any information we 3122 * gathered while doing the load. 3123 */ 3124 if (state == SPA_LOAD_RECOVER) { 3125 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3126 spa->spa_load_info) == 0); 3127 } 3128 3129 if (locked) { 3130 spa->spa_last_open_failed = 0; 3131 spa->spa_last_ubsync_txg = 0; 3132 spa->spa_load_txg = 0; 3133 mutex_exit(&spa_namespace_lock); 3134#ifdef __FreeBSD__ 3135#ifdef _KERNEL 3136 if (firstopen) 3137 zvol_create_minors(spa->spa_name); 3138#endif 3139#endif 3140 } 3141 3142 *spapp = spa; 3143 3144 return (0); 3145} 3146 3147int 3148spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3149 nvlist_t **config) 3150{ 3151 return (spa_open_common(name, spapp, tag, policy, config)); 3152} 3153 3154int 3155spa_open(const char *name, spa_t **spapp, void *tag) 3156{ 3157 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3158} 3159 3160/* 3161 * Lookup the given spa_t, incrementing the inject count in the process, 3162 * preventing it from being exported or destroyed. 3163 */ 3164spa_t * 3165spa_inject_addref(char *name) 3166{ 3167 spa_t *spa; 3168 3169 mutex_enter(&spa_namespace_lock); 3170 if ((spa = spa_lookup(name)) == NULL) { 3171 mutex_exit(&spa_namespace_lock); 3172 return (NULL); 3173 } 3174 spa->spa_inject_ref++; 3175 mutex_exit(&spa_namespace_lock); 3176 3177 return (spa); 3178} 3179 3180void 3181spa_inject_delref(spa_t *spa) 3182{ 3183 mutex_enter(&spa_namespace_lock); 3184 spa->spa_inject_ref--; 3185 mutex_exit(&spa_namespace_lock); 3186} 3187 3188/* 3189 * Add spares device information to the nvlist. 3190 */ 3191static void 3192spa_add_spares(spa_t *spa, nvlist_t *config) 3193{ 3194 nvlist_t **spares; 3195 uint_t i, nspares; 3196 nvlist_t *nvroot; 3197 uint64_t guid; 3198 vdev_stat_t *vs; 3199 uint_t vsc; 3200 uint64_t pool; 3201 3202 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3203 3204 if (spa->spa_spares.sav_count == 0) 3205 return; 3206 3207 VERIFY(nvlist_lookup_nvlist(config, 3208 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3209 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3210 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3211 if (nspares != 0) { 3212 VERIFY(nvlist_add_nvlist_array(nvroot, 3213 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3214 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3215 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3216 3217 /* 3218 * Go through and find any spares which have since been 3219 * repurposed as an active spare. If this is the case, update 3220 * their status appropriately. 3221 */ 3222 for (i = 0; i < nspares; i++) { 3223 VERIFY(nvlist_lookup_uint64(spares[i], 3224 ZPOOL_CONFIG_GUID, &guid) == 0); 3225 if (spa_spare_exists(guid, &pool, NULL) && 3226 pool != 0ULL) { 3227 VERIFY(nvlist_lookup_uint64_array( 3228 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3229 (uint64_t **)&vs, &vsc) == 0); 3230 vs->vs_state = VDEV_STATE_CANT_OPEN; 3231 vs->vs_aux = VDEV_AUX_SPARED; 3232 } 3233 } 3234 } 3235} 3236 3237/* 3238 * Add l2cache device information to the nvlist, including vdev stats. 3239 */ 3240static void 3241spa_add_l2cache(spa_t *spa, nvlist_t *config) 3242{ 3243 nvlist_t **l2cache; 3244 uint_t i, j, nl2cache; 3245 nvlist_t *nvroot; 3246 uint64_t guid; 3247 vdev_t *vd; 3248 vdev_stat_t *vs; 3249 uint_t vsc; 3250 3251 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3252 3253 if (spa->spa_l2cache.sav_count == 0) 3254 return; 3255 3256 VERIFY(nvlist_lookup_nvlist(config, 3257 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3258 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3259 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3260 if (nl2cache != 0) { 3261 VERIFY(nvlist_add_nvlist_array(nvroot, 3262 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3263 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3264 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3265 3266 /* 3267 * Update level 2 cache device stats. 3268 */ 3269 3270 for (i = 0; i < nl2cache; i++) { 3271 VERIFY(nvlist_lookup_uint64(l2cache[i], 3272 ZPOOL_CONFIG_GUID, &guid) == 0); 3273 3274 vd = NULL; 3275 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3276 if (guid == 3277 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3278 vd = spa->spa_l2cache.sav_vdevs[j]; 3279 break; 3280 } 3281 } 3282 ASSERT(vd != NULL); 3283 3284 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3285 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3286 == 0); 3287 vdev_get_stats(vd, vs); 3288 } 3289 } 3290} 3291 3292static void 3293spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3294{ 3295 nvlist_t *features; 3296 zap_cursor_t zc; 3297 zap_attribute_t za; 3298 3299 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3300 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3301 3302 /* We may be unable to read features if pool is suspended. */ 3303 if (spa_suspended(spa)) 3304 goto out; 3305 3306 if (spa->spa_feat_for_read_obj != 0) { 3307 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3308 spa->spa_feat_for_read_obj); 3309 zap_cursor_retrieve(&zc, &za) == 0; 3310 zap_cursor_advance(&zc)) { 3311 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3312 za.za_num_integers == 1); 3313 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3314 za.za_first_integer)); 3315 } 3316 zap_cursor_fini(&zc); 3317 } 3318 3319 if (spa->spa_feat_for_write_obj != 0) { 3320 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3321 spa->spa_feat_for_write_obj); 3322 zap_cursor_retrieve(&zc, &za) == 0; 3323 zap_cursor_advance(&zc)) { 3324 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3325 za.za_num_integers == 1); 3326 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3327 za.za_first_integer)); 3328 } 3329 zap_cursor_fini(&zc); 3330 } 3331 3332out: 3333 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3334 features) == 0); 3335 nvlist_free(features); 3336} 3337 3338int 3339spa_get_stats(const char *name, nvlist_t **config, 3340 char *altroot, size_t buflen) 3341{ 3342 int error; 3343 spa_t *spa; 3344 3345 *config = NULL; 3346 error = spa_open_common(name, &spa, FTAG, NULL, config); 3347 3348 if (spa != NULL) { 3349 /* 3350 * This still leaves a window of inconsistency where the spares 3351 * or l2cache devices could change and the config would be 3352 * self-inconsistent. 3353 */ 3354 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3355 3356 if (*config != NULL) { 3357 uint64_t loadtimes[2]; 3358 3359 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3360 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3361 VERIFY(nvlist_add_uint64_array(*config, 3362 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3363 3364 VERIFY(nvlist_add_uint64(*config, 3365 ZPOOL_CONFIG_ERRCOUNT, 3366 spa_get_errlog_size(spa)) == 0); 3367 3368 if (spa_suspended(spa)) 3369 VERIFY(nvlist_add_uint64(*config, 3370 ZPOOL_CONFIG_SUSPENDED, 3371 spa->spa_failmode) == 0); 3372 3373 spa_add_spares(spa, *config); 3374 spa_add_l2cache(spa, *config); 3375 spa_add_feature_stats(spa, *config); 3376 } 3377 } 3378 3379 /* 3380 * We want to get the alternate root even for faulted pools, so we cheat 3381 * and call spa_lookup() directly. 3382 */ 3383 if (altroot) { 3384 if (spa == NULL) { 3385 mutex_enter(&spa_namespace_lock); 3386 spa = spa_lookup(name); 3387 if (spa) 3388 spa_altroot(spa, altroot, buflen); 3389 else 3390 altroot[0] = '\0'; 3391 spa = NULL; 3392 mutex_exit(&spa_namespace_lock); 3393 } else { 3394 spa_altroot(spa, altroot, buflen); 3395 } 3396 } 3397 3398 if (spa != NULL) { 3399 spa_config_exit(spa, SCL_CONFIG, FTAG); 3400 spa_close(spa, FTAG); 3401 } 3402 3403 return (error); 3404} 3405 3406/* 3407 * Validate that the auxiliary device array is well formed. We must have an 3408 * array of nvlists, each which describes a valid leaf vdev. If this is an 3409 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3410 * specified, as long as they are well-formed. 3411 */ 3412static int 3413spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3414 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3415 vdev_labeltype_t label) 3416{ 3417 nvlist_t **dev; 3418 uint_t i, ndev; 3419 vdev_t *vd; 3420 int error; 3421 3422 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3423 3424 /* 3425 * It's acceptable to have no devs specified. 3426 */ 3427 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3428 return (0); 3429 3430 if (ndev == 0) 3431 return (SET_ERROR(EINVAL)); 3432 3433 /* 3434 * Make sure the pool is formatted with a version that supports this 3435 * device type. 3436 */ 3437 if (spa_version(spa) < version) 3438 return (SET_ERROR(ENOTSUP)); 3439 3440 /* 3441 * Set the pending device list so we correctly handle device in-use 3442 * checking. 3443 */ 3444 sav->sav_pending = dev; 3445 sav->sav_npending = ndev; 3446 3447 for (i = 0; i < ndev; i++) { 3448 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3449 mode)) != 0) 3450 goto out; 3451 3452 if (!vd->vdev_ops->vdev_op_leaf) { 3453 vdev_free(vd); 3454 error = SET_ERROR(EINVAL); 3455 goto out; 3456 } 3457 3458 /* 3459 * The L2ARC currently only supports disk devices in 3460 * kernel context. For user-level testing, we allow it. 3461 */ 3462#ifdef _KERNEL 3463 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3464 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3465 error = SET_ERROR(ENOTBLK); 3466 vdev_free(vd); 3467 goto out; 3468 } 3469#endif 3470 vd->vdev_top = vd; 3471 3472 if ((error = vdev_open(vd)) == 0 && 3473 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3474 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3475 vd->vdev_guid) == 0); 3476 } 3477 3478 vdev_free(vd); 3479 3480 if (error && 3481 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3482 goto out; 3483 else 3484 error = 0; 3485 } 3486 3487out: 3488 sav->sav_pending = NULL; 3489 sav->sav_npending = 0; 3490 return (error); 3491} 3492 3493static int 3494spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3495{ 3496 int error; 3497 3498 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3499 3500 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3501 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3502 VDEV_LABEL_SPARE)) != 0) { 3503 return (error); 3504 } 3505 3506 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3507 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3508 VDEV_LABEL_L2CACHE)); 3509} 3510 3511static void 3512spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3513 const char *config) 3514{ 3515 int i; 3516 3517 if (sav->sav_config != NULL) { 3518 nvlist_t **olddevs; 3519 uint_t oldndevs; 3520 nvlist_t **newdevs; 3521 3522 /* 3523 * Generate new dev list by concatentating with the 3524 * current dev list. 3525 */ 3526 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3527 &olddevs, &oldndevs) == 0); 3528 3529 newdevs = kmem_alloc(sizeof (void *) * 3530 (ndevs + oldndevs), KM_SLEEP); 3531 for (i = 0; i < oldndevs; i++) 3532 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3533 KM_SLEEP) == 0); 3534 for (i = 0; i < ndevs; i++) 3535 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3536 KM_SLEEP) == 0); 3537 3538 VERIFY(nvlist_remove(sav->sav_config, config, 3539 DATA_TYPE_NVLIST_ARRAY) == 0); 3540 3541 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3542 config, newdevs, ndevs + oldndevs) == 0); 3543 for (i = 0; i < oldndevs + ndevs; i++) 3544 nvlist_free(newdevs[i]); 3545 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3546 } else { 3547 /* 3548 * Generate a new dev list. 3549 */ 3550 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3551 KM_SLEEP) == 0); 3552 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3553 devs, ndevs) == 0); 3554 } 3555} 3556 3557/* 3558 * Stop and drop level 2 ARC devices 3559 */ 3560void 3561spa_l2cache_drop(spa_t *spa) 3562{ 3563 vdev_t *vd; 3564 int i; 3565 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3566 3567 for (i = 0; i < sav->sav_count; i++) { 3568 uint64_t pool; 3569 3570 vd = sav->sav_vdevs[i]; 3571 ASSERT(vd != NULL); 3572 3573 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3574 pool != 0ULL && l2arc_vdev_present(vd)) 3575 l2arc_remove_vdev(vd); 3576 } 3577} 3578 3579/* 3580 * Pool Creation 3581 */ 3582int 3583spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3584 nvlist_t *zplprops) 3585{ 3586 spa_t *spa; 3587 char *altroot = NULL; 3588 vdev_t *rvd; 3589 dsl_pool_t *dp; 3590 dmu_tx_t *tx; 3591 int error = 0; 3592 uint64_t txg = TXG_INITIAL; 3593 nvlist_t **spares, **l2cache; 3594 uint_t nspares, nl2cache; 3595 uint64_t version, obj; 3596 boolean_t has_features; 3597 3598 /* 3599 * If this pool already exists, return failure. 3600 */ 3601 mutex_enter(&spa_namespace_lock); 3602 if (spa_lookup(pool) != NULL) { 3603 mutex_exit(&spa_namespace_lock); 3604 return (SET_ERROR(EEXIST)); 3605 } 3606 3607 /* 3608 * Allocate a new spa_t structure. 3609 */ 3610 (void) nvlist_lookup_string(props, 3611 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3612 spa = spa_add(pool, NULL, altroot); 3613 spa_activate(spa, spa_mode_global); 3614 3615 if (props && (error = spa_prop_validate(spa, props))) { 3616 spa_deactivate(spa); 3617 spa_remove(spa); 3618 mutex_exit(&spa_namespace_lock); 3619 return (error); 3620 } 3621 3622 has_features = B_FALSE; 3623 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3624 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3625 if (zpool_prop_feature(nvpair_name(elem))) 3626 has_features = B_TRUE; 3627 } 3628 3629 if (has_features || nvlist_lookup_uint64(props, 3630 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3631 version = SPA_VERSION; 3632 } 3633 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3634 3635 spa->spa_first_txg = txg; 3636 spa->spa_uberblock.ub_txg = txg - 1; 3637 spa->spa_uberblock.ub_version = version; 3638 spa->spa_ubsync = spa->spa_uberblock; 3639 3640 /* 3641 * Create "The Godfather" zio to hold all async IOs 3642 */ 3643 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3644 KM_SLEEP); 3645 for (int i = 0; i < max_ncpus; i++) { 3646 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3647 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3648 ZIO_FLAG_GODFATHER); 3649 } 3650 3651 /* 3652 * Create the root vdev. 3653 */ 3654 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3655 3656 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3657 3658 ASSERT(error != 0 || rvd != NULL); 3659 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3660 3661 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3662 error = SET_ERROR(EINVAL); 3663 3664 if (error == 0 && 3665 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3666 (error = spa_validate_aux(spa, nvroot, txg, 3667 VDEV_ALLOC_ADD)) == 0) { 3668 for (int c = 0; c < rvd->vdev_children; c++) { 3669 vdev_ashift_optimize(rvd->vdev_child[c]); 3670 vdev_metaslab_set_size(rvd->vdev_child[c]); 3671 vdev_expand(rvd->vdev_child[c], txg); 3672 } 3673 } 3674 3675 spa_config_exit(spa, SCL_ALL, FTAG); 3676 3677 if (error != 0) { 3678 spa_unload(spa); 3679 spa_deactivate(spa); 3680 spa_remove(spa); 3681 mutex_exit(&spa_namespace_lock); 3682 return (error); 3683 } 3684 3685 /* 3686 * Get the list of spares, if specified. 3687 */ 3688 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3689 &spares, &nspares) == 0) { 3690 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3691 KM_SLEEP) == 0); 3692 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3693 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3694 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3695 spa_load_spares(spa); 3696 spa_config_exit(spa, SCL_ALL, FTAG); 3697 spa->spa_spares.sav_sync = B_TRUE; 3698 } 3699 3700 /* 3701 * Get the list of level 2 cache devices, if specified. 3702 */ 3703 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3704 &l2cache, &nl2cache) == 0) { 3705 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3706 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3707 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3708 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3709 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3710 spa_load_l2cache(spa); 3711 spa_config_exit(spa, SCL_ALL, FTAG); 3712 spa->spa_l2cache.sav_sync = B_TRUE; 3713 } 3714 3715 spa->spa_is_initializing = B_TRUE; 3716 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3717 spa->spa_meta_objset = dp->dp_meta_objset; 3718 spa->spa_is_initializing = B_FALSE; 3719 3720 /* 3721 * Create DDTs (dedup tables). 3722 */ 3723 ddt_create(spa); 3724 3725 spa_update_dspace(spa); 3726 3727 tx = dmu_tx_create_assigned(dp, txg); 3728 3729 /* 3730 * Create the pool config object. 3731 */ 3732 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3733 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3734 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3735 3736 if (zap_add(spa->spa_meta_objset, 3737 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3738 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3739 cmn_err(CE_PANIC, "failed to add pool config"); 3740 } 3741 3742 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3743 spa_feature_create_zap_objects(spa, tx); 3744 3745 if (zap_add(spa->spa_meta_objset, 3746 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3747 sizeof (uint64_t), 1, &version, tx) != 0) { 3748 cmn_err(CE_PANIC, "failed to add pool version"); 3749 } 3750 3751 /* Newly created pools with the right version are always deflated. */ 3752 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3753 spa->spa_deflate = TRUE; 3754 if (zap_add(spa->spa_meta_objset, 3755 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3756 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3757 cmn_err(CE_PANIC, "failed to add deflate"); 3758 } 3759 } 3760 3761 /* 3762 * Create the deferred-free bpobj. Turn off compression 3763 * because sync-to-convergence takes longer if the blocksize 3764 * keeps changing. 3765 */ 3766 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3767 dmu_object_set_compress(spa->spa_meta_objset, obj, 3768 ZIO_COMPRESS_OFF, tx); 3769 if (zap_add(spa->spa_meta_objset, 3770 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3771 sizeof (uint64_t), 1, &obj, tx) != 0) { 3772 cmn_err(CE_PANIC, "failed to add bpobj"); 3773 } 3774 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3775 spa->spa_meta_objset, obj)); 3776 3777 /* 3778 * Create the pool's history object. 3779 */ 3780 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3781 spa_history_create_obj(spa, tx); 3782 3783 /* 3784 * Generate some random noise for salted checksums to operate on. 3785 */ 3786 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3787 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3788 3789 /* 3790 * Set pool properties. 3791 */ 3792 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3793 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3794 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3795 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3796 3797 if (props != NULL) { 3798 spa_configfile_set(spa, props, B_FALSE); 3799 spa_sync_props(props, tx); 3800 } 3801 3802 dmu_tx_commit(tx); 3803 3804 spa->spa_sync_on = B_TRUE; 3805 txg_sync_start(spa->spa_dsl_pool); 3806 3807 /* 3808 * We explicitly wait for the first transaction to complete so that our 3809 * bean counters are appropriately updated. 3810 */ 3811 txg_wait_synced(spa->spa_dsl_pool, txg); 3812 3813 spa_config_sync(spa, B_FALSE, B_TRUE); 3814 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3815 3816 spa_history_log_version(spa, "create"); 3817 3818 /* 3819 * Don't count references from objsets that are already closed 3820 * and are making their way through the eviction process. 3821 */ 3822 spa_evicting_os_wait(spa); 3823 spa->spa_minref = refcount_count(&spa->spa_refcount); 3824 3825 mutex_exit(&spa_namespace_lock); 3826 3827 return (0); 3828} 3829 3830#ifdef _KERNEL 3831#ifdef illumos 3832/* 3833 * Get the root pool information from the root disk, then import the root pool 3834 * during the system boot up time. 3835 */ 3836extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3837 3838static nvlist_t * 3839spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3840{ 3841 nvlist_t *config; 3842 nvlist_t *nvtop, *nvroot; 3843 uint64_t pgid; 3844 3845 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3846 return (NULL); 3847 3848 /* 3849 * Add this top-level vdev to the child array. 3850 */ 3851 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3852 &nvtop) == 0); 3853 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3854 &pgid) == 0); 3855 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3856 3857 /* 3858 * Put this pool's top-level vdevs into a root vdev. 3859 */ 3860 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3861 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3862 VDEV_TYPE_ROOT) == 0); 3863 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3864 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3865 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3866 &nvtop, 1) == 0); 3867 3868 /* 3869 * Replace the existing vdev_tree with the new root vdev in 3870 * this pool's configuration (remove the old, add the new). 3871 */ 3872 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3873 nvlist_free(nvroot); 3874 return (config); 3875} 3876 3877/* 3878 * Walk the vdev tree and see if we can find a device with "better" 3879 * configuration. A configuration is "better" if the label on that 3880 * device has a more recent txg. 3881 */ 3882static void 3883spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3884{ 3885 for (int c = 0; c < vd->vdev_children; c++) 3886 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3887 3888 if (vd->vdev_ops->vdev_op_leaf) { 3889 nvlist_t *label; 3890 uint64_t label_txg; 3891 3892 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3893 &label) != 0) 3894 return; 3895 3896 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3897 &label_txg) == 0); 3898 3899 /* 3900 * Do we have a better boot device? 3901 */ 3902 if (label_txg > *txg) { 3903 *txg = label_txg; 3904 *avd = vd; 3905 } 3906 nvlist_free(label); 3907 } 3908} 3909 3910/* 3911 * Import a root pool. 3912 * 3913 * For x86. devpath_list will consist of devid and/or physpath name of 3914 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3915 * The GRUB "findroot" command will return the vdev we should boot. 3916 * 3917 * For Sparc, devpath_list consists the physpath name of the booting device 3918 * no matter the rootpool is a single device pool or a mirrored pool. 3919 * e.g. 3920 * "/pci@1f,0/ide@d/disk@0,0:a" 3921 */ 3922int 3923spa_import_rootpool(char *devpath, char *devid) 3924{ 3925 spa_t *spa; 3926 vdev_t *rvd, *bvd, *avd = NULL; 3927 nvlist_t *config, *nvtop; 3928 uint64_t guid, txg; 3929 char *pname; 3930 int error; 3931 3932 /* 3933 * Read the label from the boot device and generate a configuration. 3934 */ 3935 config = spa_generate_rootconf(devpath, devid, &guid); 3936#if defined(_OBP) && defined(_KERNEL) 3937 if (config == NULL) { 3938 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3939 /* iscsi boot */ 3940 get_iscsi_bootpath_phy(devpath); 3941 config = spa_generate_rootconf(devpath, devid, &guid); 3942 } 3943 } 3944#endif 3945 if (config == NULL) { 3946 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3947 devpath); 3948 return (SET_ERROR(EIO)); 3949 } 3950 3951 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3952 &pname) == 0); 3953 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3954 3955 mutex_enter(&spa_namespace_lock); 3956 if ((spa = spa_lookup(pname)) != NULL) { 3957 /* 3958 * Remove the existing root pool from the namespace so that we 3959 * can replace it with the correct config we just read in. 3960 */ 3961 spa_remove(spa); 3962 } 3963 3964 spa = spa_add(pname, config, NULL); 3965 spa->spa_is_root = B_TRUE; 3966 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3967 3968 /* 3969 * Build up a vdev tree based on the boot device's label config. 3970 */ 3971 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3972 &nvtop) == 0); 3973 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3974 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3975 VDEV_ALLOC_ROOTPOOL); 3976 spa_config_exit(spa, SCL_ALL, FTAG); 3977 if (error) { 3978 mutex_exit(&spa_namespace_lock); 3979 nvlist_free(config); 3980 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3981 pname); 3982 return (error); 3983 } 3984 3985 /* 3986 * Get the boot vdev. 3987 */ 3988 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3989 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3990 (u_longlong_t)guid); 3991 error = SET_ERROR(ENOENT); 3992 goto out; 3993 } 3994 3995 /* 3996 * Determine if there is a better boot device. 3997 */ 3998 avd = bvd; 3999 spa_alt_rootvdev(rvd, &avd, &txg); 4000 if (avd != bvd) { 4001 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 4002 "try booting from '%s'", avd->vdev_path); 4003 error = SET_ERROR(EINVAL); 4004 goto out; 4005 } 4006 4007 /* 4008 * If the boot device is part of a spare vdev then ensure that 4009 * we're booting off the active spare. 4010 */ 4011 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 4012 !bvd->vdev_isspare) { 4013 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 4014 "try booting from '%s'", 4015 bvd->vdev_parent-> 4016 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4017 error = SET_ERROR(EINVAL); 4018 goto out; 4019 } 4020 4021 error = 0; 4022out: 4023 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4024 vdev_free(rvd); 4025 spa_config_exit(spa, SCL_ALL, FTAG); 4026 mutex_exit(&spa_namespace_lock); 4027 4028 nvlist_free(config); 4029 return (error); 4030} 4031 4032#else /* !illumos */ 4033 4034extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 4035 uint64_t *count); 4036 4037static nvlist_t * 4038spa_generate_rootconf(const char *name) 4039{ 4040 nvlist_t **configs, **tops; 4041 nvlist_t *config; 4042 nvlist_t *best_cfg, *nvtop, *nvroot; 4043 uint64_t *holes; 4044 uint64_t best_txg; 4045 uint64_t nchildren; 4046 uint64_t pgid; 4047 uint64_t count; 4048 uint64_t i; 4049 uint_t nholes; 4050 4051 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4052 return (NULL); 4053 4054 ASSERT3U(count, !=, 0); 4055 best_txg = 0; 4056 for (i = 0; i < count; i++) { 4057 uint64_t txg; 4058 4059 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4060 &txg) == 0); 4061 if (txg > best_txg) { 4062 best_txg = txg; 4063 best_cfg = configs[i]; 4064 } 4065 } 4066 4067 /* 4068 * Multi-vdev root pool configuration discovery is not supported yet. 4069 */ 4070 nchildren = 1; 4071 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4072 holes = NULL; 4073 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4074 &holes, &nholes); 4075 4076 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4077 for (i = 0; i < nchildren; i++) { 4078 if (i >= count) 4079 break; 4080 if (configs[i] == NULL) 4081 continue; 4082 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4083 &nvtop) == 0); 4084 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4085 } 4086 for (i = 0; holes != NULL && i < nholes; i++) { 4087 if (i >= nchildren) 4088 continue; 4089 if (tops[holes[i]] != NULL) 4090 continue; 4091 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4092 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4093 VDEV_TYPE_HOLE) == 0); 4094 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4095 holes[i]) == 0); 4096 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4097 0) == 0); 4098 } 4099 for (i = 0; i < nchildren; i++) { 4100 if (tops[i] != NULL) 4101 continue; 4102 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4103 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4104 VDEV_TYPE_MISSING) == 0); 4105 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4106 i) == 0); 4107 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4108 0) == 0); 4109 } 4110 4111 /* 4112 * Create pool config based on the best vdev config. 4113 */ 4114 nvlist_dup(best_cfg, &config, KM_SLEEP); 4115 4116 /* 4117 * Put this pool's top-level vdevs into a root vdev. 4118 */ 4119 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4120 &pgid) == 0); 4121 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4122 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4123 VDEV_TYPE_ROOT) == 0); 4124 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4125 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4126 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4127 tops, nchildren) == 0); 4128 4129 /* 4130 * Replace the existing vdev_tree with the new root vdev in 4131 * this pool's configuration (remove the old, add the new). 4132 */ 4133 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4134 4135 /* 4136 * Drop vdev config elements that should not be present at pool level. 4137 */ 4138 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4139 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4140 4141 for (i = 0; i < count; i++) 4142 nvlist_free(configs[i]); 4143 kmem_free(configs, count * sizeof(void *)); 4144 for (i = 0; i < nchildren; i++) 4145 nvlist_free(tops[i]); 4146 kmem_free(tops, nchildren * sizeof(void *)); 4147 nvlist_free(nvroot); 4148 return (config); 4149} 4150 4151int 4152spa_import_rootpool(const char *name) 4153{ 4154 spa_t *spa; 4155 vdev_t *rvd, *bvd, *avd = NULL; 4156 nvlist_t *config, *nvtop; 4157 uint64_t txg; 4158 char *pname; 4159 int error; 4160 4161 /* 4162 * Read the label from the boot device and generate a configuration. 4163 */ 4164 config = spa_generate_rootconf(name); 4165 4166 mutex_enter(&spa_namespace_lock); 4167 if (config != NULL) { 4168 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4169 &pname) == 0 && strcmp(name, pname) == 0); 4170 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4171 == 0); 4172 4173 if ((spa = spa_lookup(pname)) != NULL) { 4174 /* 4175 * Remove the existing root pool from the namespace so 4176 * that we can replace it with the correct config 4177 * we just read in. 4178 */ 4179 spa_remove(spa); 4180 } 4181 spa = spa_add(pname, config, NULL); 4182 4183 /* 4184 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4185 * via spa_version(). 4186 */ 4187 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4188 &spa->spa_ubsync.ub_version) != 0) 4189 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4190 } else if ((spa = spa_lookup(name)) == NULL) { 4191 mutex_exit(&spa_namespace_lock); 4192 nvlist_free(config); 4193 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4194 name); 4195 return (EIO); 4196 } else { 4197 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4198 } 4199 spa->spa_is_root = B_TRUE; 4200 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4201 4202 /* 4203 * Build up a vdev tree based on the boot device's label config. 4204 */ 4205 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4206 &nvtop) == 0); 4207 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4208 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4209 VDEV_ALLOC_ROOTPOOL); 4210 spa_config_exit(spa, SCL_ALL, FTAG); 4211 if (error) { 4212 mutex_exit(&spa_namespace_lock); 4213 nvlist_free(config); 4214 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4215 pname); 4216 return (error); 4217 } 4218 4219 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4220 vdev_free(rvd); 4221 spa_config_exit(spa, SCL_ALL, FTAG); 4222 mutex_exit(&spa_namespace_lock); 4223 4224 nvlist_free(config); 4225 return (0); 4226} 4227 4228#endif /* illumos */ 4229#endif /* _KERNEL */ 4230 4231/* 4232 * Import a non-root pool into the system. 4233 */ 4234int 4235spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4236{ 4237 spa_t *spa; 4238 char *altroot = NULL; 4239 spa_load_state_t state = SPA_LOAD_IMPORT; 4240 zpool_rewind_policy_t policy; 4241 uint64_t mode = spa_mode_global; 4242 uint64_t readonly = B_FALSE; 4243 int error; 4244 nvlist_t *nvroot; 4245 nvlist_t **spares, **l2cache; 4246 uint_t nspares, nl2cache; 4247 4248 /* 4249 * If a pool with this name exists, return failure. 4250 */ 4251 mutex_enter(&spa_namespace_lock); 4252 if (spa_lookup(pool) != NULL) { 4253 mutex_exit(&spa_namespace_lock); 4254 return (SET_ERROR(EEXIST)); 4255 } 4256 4257 /* 4258 * Create and initialize the spa structure. 4259 */ 4260 (void) nvlist_lookup_string(props, 4261 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4262 (void) nvlist_lookup_uint64(props, 4263 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4264 if (readonly) 4265 mode = FREAD; 4266 spa = spa_add(pool, config, altroot); 4267 spa->spa_import_flags = flags; 4268 4269 /* 4270 * Verbatim import - Take a pool and insert it into the namespace 4271 * as if it had been loaded at boot. 4272 */ 4273 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4274 if (props != NULL) 4275 spa_configfile_set(spa, props, B_FALSE); 4276 4277 spa_config_sync(spa, B_FALSE, B_TRUE); 4278 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4279 4280 mutex_exit(&spa_namespace_lock); 4281 return (0); 4282 } 4283 4284 spa_activate(spa, mode); 4285 4286 /* 4287 * Don't start async tasks until we know everything is healthy. 4288 */ 4289 spa_async_suspend(spa); 4290 4291 zpool_get_rewind_policy(config, &policy); 4292 if (policy.zrp_request & ZPOOL_DO_REWIND) 4293 state = SPA_LOAD_RECOVER; 4294 4295 /* 4296 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4297 * because the user-supplied config is actually the one to trust when 4298 * doing an import. 4299 */ 4300 if (state != SPA_LOAD_RECOVER) 4301 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4302 4303 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4304 policy.zrp_request); 4305 4306 /* 4307 * Propagate anything learned while loading the pool and pass it 4308 * back to caller (i.e. rewind info, missing devices, etc). 4309 */ 4310 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4311 spa->spa_load_info) == 0); 4312 4313 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4314 /* 4315 * Toss any existing sparelist, as it doesn't have any validity 4316 * anymore, and conflicts with spa_has_spare(). 4317 */ 4318 if (spa->spa_spares.sav_config) { 4319 nvlist_free(spa->spa_spares.sav_config); 4320 spa->spa_spares.sav_config = NULL; 4321 spa_load_spares(spa); 4322 } 4323 if (spa->spa_l2cache.sav_config) { 4324 nvlist_free(spa->spa_l2cache.sav_config); 4325 spa->spa_l2cache.sav_config = NULL; 4326 spa_load_l2cache(spa); 4327 } 4328 4329 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4330 &nvroot) == 0); 4331 if (error == 0) 4332 error = spa_validate_aux(spa, nvroot, -1ULL, 4333 VDEV_ALLOC_SPARE); 4334 if (error == 0) 4335 error = spa_validate_aux(spa, nvroot, -1ULL, 4336 VDEV_ALLOC_L2CACHE); 4337 spa_config_exit(spa, SCL_ALL, FTAG); 4338 4339 if (props != NULL) 4340 spa_configfile_set(spa, props, B_FALSE); 4341 4342 if (error != 0 || (props && spa_writeable(spa) && 4343 (error = spa_prop_set(spa, props)))) { 4344 spa_unload(spa); 4345 spa_deactivate(spa); 4346 spa_remove(spa); 4347 mutex_exit(&spa_namespace_lock); 4348 return (error); 4349 } 4350 4351 spa_async_resume(spa); 4352 4353 /* 4354 * Override any spares and level 2 cache devices as specified by 4355 * the user, as these may have correct device names/devids, etc. 4356 */ 4357 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4358 &spares, &nspares) == 0) { 4359 if (spa->spa_spares.sav_config) 4360 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4361 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4362 else 4363 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4364 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4365 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4366 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4367 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4368 spa_load_spares(spa); 4369 spa_config_exit(spa, SCL_ALL, FTAG); 4370 spa->spa_spares.sav_sync = B_TRUE; 4371 } 4372 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4373 &l2cache, &nl2cache) == 0) { 4374 if (spa->spa_l2cache.sav_config) 4375 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4376 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4377 else 4378 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4379 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4380 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4381 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4382 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4383 spa_load_l2cache(spa); 4384 spa_config_exit(spa, SCL_ALL, FTAG); 4385 spa->spa_l2cache.sav_sync = B_TRUE; 4386 } 4387 4388 /* 4389 * Check for any removed devices. 4390 */ 4391 if (spa->spa_autoreplace) { 4392 spa_aux_check_removed(&spa->spa_spares); 4393 spa_aux_check_removed(&spa->spa_l2cache); 4394 } 4395 4396 if (spa_writeable(spa)) { 4397 /* 4398 * Update the config cache to include the newly-imported pool. 4399 */ 4400 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4401 } 4402 4403 /* 4404 * It's possible that the pool was expanded while it was exported. 4405 * We kick off an async task to handle this for us. 4406 */ 4407 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4408 4409 spa_history_log_version(spa, "import"); 4410 4411 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4412 4413 mutex_exit(&spa_namespace_lock); 4414 4415#ifdef __FreeBSD__ 4416#ifdef _KERNEL 4417 zvol_create_minors(pool); 4418#endif 4419#endif 4420 return (0); 4421} 4422 4423nvlist_t * 4424spa_tryimport(nvlist_t *tryconfig) 4425{ 4426 nvlist_t *config = NULL; 4427 char *poolname; 4428 spa_t *spa; 4429 uint64_t state; 4430 int error; 4431 4432 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4433 return (NULL); 4434 4435 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4436 return (NULL); 4437 4438 /* 4439 * Create and initialize the spa structure. 4440 */ 4441 mutex_enter(&spa_namespace_lock); 4442 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4443 spa_activate(spa, FREAD); 4444 4445 /* 4446 * Pass off the heavy lifting to spa_load(). 4447 * Pass TRUE for mosconfig because the user-supplied config 4448 * is actually the one to trust when doing an import. 4449 */ 4450 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4451 4452 /* 4453 * If 'tryconfig' was at least parsable, return the current config. 4454 */ 4455 if (spa->spa_root_vdev != NULL) { 4456 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4457 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4458 poolname) == 0); 4459 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4460 state) == 0); 4461 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4462 spa->spa_uberblock.ub_timestamp) == 0); 4463 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4464 spa->spa_load_info) == 0); 4465 4466 /* 4467 * If the bootfs property exists on this pool then we 4468 * copy it out so that external consumers can tell which 4469 * pools are bootable. 4470 */ 4471 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4472 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4473 4474 /* 4475 * We have to play games with the name since the 4476 * pool was opened as TRYIMPORT_NAME. 4477 */ 4478 if (dsl_dsobj_to_dsname(spa_name(spa), 4479 spa->spa_bootfs, tmpname) == 0) { 4480 char *cp; 4481 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4482 4483 cp = strchr(tmpname, '/'); 4484 if (cp == NULL) { 4485 (void) strlcpy(dsname, tmpname, 4486 MAXPATHLEN); 4487 } else { 4488 (void) snprintf(dsname, MAXPATHLEN, 4489 "%s/%s", poolname, ++cp); 4490 } 4491 VERIFY(nvlist_add_string(config, 4492 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4493 kmem_free(dsname, MAXPATHLEN); 4494 } 4495 kmem_free(tmpname, MAXPATHLEN); 4496 } 4497 4498 /* 4499 * Add the list of hot spares and level 2 cache devices. 4500 */ 4501 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4502 spa_add_spares(spa, config); 4503 spa_add_l2cache(spa, config); 4504 spa_config_exit(spa, SCL_CONFIG, FTAG); 4505 } 4506 4507 spa_unload(spa); 4508 spa_deactivate(spa); 4509 spa_remove(spa); 4510 mutex_exit(&spa_namespace_lock); 4511 4512 return (config); 4513} 4514 4515/* 4516 * Pool export/destroy 4517 * 4518 * The act of destroying or exporting a pool is very simple. We make sure there 4519 * is no more pending I/O and any references to the pool are gone. Then, we 4520 * update the pool state and sync all the labels to disk, removing the 4521 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4522 * we don't sync the labels or remove the configuration cache. 4523 */ 4524static int 4525spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4526 boolean_t force, boolean_t hardforce) 4527{ 4528 spa_t *spa; 4529 4530 if (oldconfig) 4531 *oldconfig = NULL; 4532 4533 if (!(spa_mode_global & FWRITE)) 4534 return (SET_ERROR(EROFS)); 4535 4536 mutex_enter(&spa_namespace_lock); 4537 if ((spa = spa_lookup(pool)) == NULL) { 4538 mutex_exit(&spa_namespace_lock); 4539 return (SET_ERROR(ENOENT)); 4540 } 4541 4542 /* 4543 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4544 * reacquire the namespace lock, and see if we can export. 4545 */ 4546 spa_open_ref(spa, FTAG); 4547 mutex_exit(&spa_namespace_lock); 4548 spa_async_suspend(spa); 4549 mutex_enter(&spa_namespace_lock); 4550 spa_close(spa, FTAG); 4551 4552 /* 4553 * The pool will be in core if it's openable, 4554 * in which case we can modify its state. 4555 */ 4556 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4557 /* 4558 * Objsets may be open only because they're dirty, so we 4559 * have to force it to sync before checking spa_refcnt. 4560 */ 4561 txg_wait_synced(spa->spa_dsl_pool, 0); 4562 spa_evicting_os_wait(spa); 4563 4564 /* 4565 * A pool cannot be exported or destroyed if there are active 4566 * references. If we are resetting a pool, allow references by 4567 * fault injection handlers. 4568 */ 4569 if (!spa_refcount_zero(spa) || 4570 (spa->spa_inject_ref != 0 && 4571 new_state != POOL_STATE_UNINITIALIZED)) { 4572 spa_async_resume(spa); 4573 mutex_exit(&spa_namespace_lock); 4574 return (SET_ERROR(EBUSY)); 4575 } 4576 4577 /* 4578 * A pool cannot be exported if it has an active shared spare. 4579 * This is to prevent other pools stealing the active spare 4580 * from an exported pool. At user's own will, such pool can 4581 * be forcedly exported. 4582 */ 4583 if (!force && new_state == POOL_STATE_EXPORTED && 4584 spa_has_active_shared_spare(spa)) { 4585 spa_async_resume(spa); 4586 mutex_exit(&spa_namespace_lock); 4587 return (SET_ERROR(EXDEV)); 4588 } 4589 4590 /* 4591 * We want this to be reflected on every label, 4592 * so mark them all dirty. spa_unload() will do the 4593 * final sync that pushes these changes out. 4594 */ 4595 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4596 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4597 spa->spa_state = new_state; 4598 spa->spa_final_txg = spa_last_synced_txg(spa) + 4599 TXG_DEFER_SIZE + 1; 4600 vdev_config_dirty(spa->spa_root_vdev); 4601 spa_config_exit(spa, SCL_ALL, FTAG); 4602 } 4603 } 4604 4605 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4606 4607 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4608 spa_unload(spa); 4609 spa_deactivate(spa); 4610 } 4611 4612 if (oldconfig && spa->spa_config) 4613 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4614 4615 if (new_state != POOL_STATE_UNINITIALIZED) { 4616 if (!hardforce) 4617 spa_config_sync(spa, B_TRUE, B_TRUE); 4618 spa_remove(spa); 4619 } 4620 mutex_exit(&spa_namespace_lock); 4621 4622 return (0); 4623} 4624 4625/* 4626 * Destroy a storage pool. 4627 */ 4628int 4629spa_destroy(char *pool) 4630{ 4631 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4632 B_FALSE, B_FALSE)); 4633} 4634 4635/* 4636 * Export a storage pool. 4637 */ 4638int 4639spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4640 boolean_t hardforce) 4641{ 4642 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4643 force, hardforce)); 4644} 4645 4646/* 4647 * Similar to spa_export(), this unloads the spa_t without actually removing it 4648 * from the namespace in any way. 4649 */ 4650int 4651spa_reset(char *pool) 4652{ 4653 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4654 B_FALSE, B_FALSE)); 4655} 4656 4657/* 4658 * ========================================================================== 4659 * Device manipulation 4660 * ========================================================================== 4661 */ 4662 4663/* 4664 * Add a device to a storage pool. 4665 */ 4666int 4667spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4668{ 4669 uint64_t txg, id; 4670 int error; 4671 vdev_t *rvd = spa->spa_root_vdev; 4672 vdev_t *vd, *tvd; 4673 nvlist_t **spares, **l2cache; 4674 uint_t nspares, nl2cache; 4675 4676 ASSERT(spa_writeable(spa)); 4677 4678 txg = spa_vdev_enter(spa); 4679 4680 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4681 VDEV_ALLOC_ADD)) != 0) 4682 return (spa_vdev_exit(spa, NULL, txg, error)); 4683 4684 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4685 4686 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4687 &nspares) != 0) 4688 nspares = 0; 4689 4690 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4691 &nl2cache) != 0) 4692 nl2cache = 0; 4693 4694 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4695 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4696 4697 if (vd->vdev_children != 0 && 4698 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4699 return (spa_vdev_exit(spa, vd, txg, error)); 4700 4701 /* 4702 * We must validate the spares and l2cache devices after checking the 4703 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4704 */ 4705 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4706 return (spa_vdev_exit(spa, vd, txg, error)); 4707 4708 /* 4709 * Transfer each new top-level vdev from vd to rvd. 4710 */ 4711 for (int c = 0; c < vd->vdev_children; c++) { 4712 4713 /* 4714 * Set the vdev id to the first hole, if one exists. 4715 */ 4716 for (id = 0; id < rvd->vdev_children; id++) { 4717 if (rvd->vdev_child[id]->vdev_ishole) { 4718 vdev_free(rvd->vdev_child[id]); 4719 break; 4720 } 4721 } 4722 tvd = vd->vdev_child[c]; 4723 vdev_remove_child(vd, tvd); 4724 tvd->vdev_id = id; 4725 vdev_add_child(rvd, tvd); 4726 vdev_config_dirty(tvd); 4727 } 4728 4729 if (nspares != 0) { 4730 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4731 ZPOOL_CONFIG_SPARES); 4732 spa_load_spares(spa); 4733 spa->spa_spares.sav_sync = B_TRUE; 4734 } 4735 4736 if (nl2cache != 0) { 4737 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4738 ZPOOL_CONFIG_L2CACHE); 4739 spa_load_l2cache(spa); 4740 spa->spa_l2cache.sav_sync = B_TRUE; 4741 } 4742 4743 /* 4744 * We have to be careful when adding new vdevs to an existing pool. 4745 * If other threads start allocating from these vdevs before we 4746 * sync the config cache, and we lose power, then upon reboot we may 4747 * fail to open the pool because there are DVAs that the config cache 4748 * can't translate. Therefore, we first add the vdevs without 4749 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4750 * and then let spa_config_update() initialize the new metaslabs. 4751 * 4752 * spa_load() checks for added-but-not-initialized vdevs, so that 4753 * if we lose power at any point in this sequence, the remaining 4754 * steps will be completed the next time we load the pool. 4755 */ 4756 (void) spa_vdev_exit(spa, vd, txg, 0); 4757 4758 mutex_enter(&spa_namespace_lock); 4759 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4760 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 4761 mutex_exit(&spa_namespace_lock); 4762 4763 return (0); 4764} 4765 4766/* 4767 * Attach a device to a mirror. The arguments are the path to any device 4768 * in the mirror, and the nvroot for the new device. If the path specifies 4769 * a device that is not mirrored, we automatically insert the mirror vdev. 4770 * 4771 * If 'replacing' is specified, the new device is intended to replace the 4772 * existing device; in this case the two devices are made into their own 4773 * mirror using the 'replacing' vdev, which is functionally identical to 4774 * the mirror vdev (it actually reuses all the same ops) but has a few 4775 * extra rules: you can't attach to it after it's been created, and upon 4776 * completion of resilvering, the first disk (the one being replaced) 4777 * is automatically detached. 4778 */ 4779int 4780spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4781{ 4782 uint64_t txg, dtl_max_txg; 4783 vdev_t *rvd = spa->spa_root_vdev; 4784 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4785 vdev_ops_t *pvops; 4786 char *oldvdpath, *newvdpath; 4787 int newvd_isspare; 4788 int error; 4789 4790 ASSERT(spa_writeable(spa)); 4791 4792 txg = spa_vdev_enter(spa); 4793 4794 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4795 4796 if (oldvd == NULL) 4797 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4798 4799 if (!oldvd->vdev_ops->vdev_op_leaf) 4800 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4801 4802 pvd = oldvd->vdev_parent; 4803 4804 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4805 VDEV_ALLOC_ATTACH)) != 0) 4806 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4807 4808 if (newrootvd->vdev_children != 1) 4809 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4810 4811 newvd = newrootvd->vdev_child[0]; 4812 4813 if (!newvd->vdev_ops->vdev_op_leaf) 4814 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4815 4816 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4817 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4818 4819 /* 4820 * Spares can't replace logs 4821 */ 4822 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4823 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4824 4825 if (!replacing) { 4826 /* 4827 * For attach, the only allowable parent is a mirror or the root 4828 * vdev. 4829 */ 4830 if (pvd->vdev_ops != &vdev_mirror_ops && 4831 pvd->vdev_ops != &vdev_root_ops) 4832 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4833 4834 pvops = &vdev_mirror_ops; 4835 } else { 4836 /* 4837 * Active hot spares can only be replaced by inactive hot 4838 * spares. 4839 */ 4840 if (pvd->vdev_ops == &vdev_spare_ops && 4841 oldvd->vdev_isspare && 4842 !spa_has_spare(spa, newvd->vdev_guid)) 4843 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4844 4845 /* 4846 * If the source is a hot spare, and the parent isn't already a 4847 * spare, then we want to create a new hot spare. Otherwise, we 4848 * want to create a replacing vdev. The user is not allowed to 4849 * attach to a spared vdev child unless the 'isspare' state is 4850 * the same (spare replaces spare, non-spare replaces 4851 * non-spare). 4852 */ 4853 if (pvd->vdev_ops == &vdev_replacing_ops && 4854 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4855 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4856 } else if (pvd->vdev_ops == &vdev_spare_ops && 4857 newvd->vdev_isspare != oldvd->vdev_isspare) { 4858 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4859 } 4860 4861 if (newvd->vdev_isspare) 4862 pvops = &vdev_spare_ops; 4863 else 4864 pvops = &vdev_replacing_ops; 4865 } 4866 4867 /* 4868 * Make sure the new device is big enough. 4869 */ 4870 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4871 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4872 4873 /* 4874 * The new device cannot have a higher alignment requirement 4875 * than the top-level vdev. 4876 */ 4877 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4878 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4879 4880 /* 4881 * If this is an in-place replacement, update oldvd's path and devid 4882 * to make it distinguishable from newvd, and unopenable from now on. 4883 */ 4884 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4885 spa_strfree(oldvd->vdev_path); 4886 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4887 KM_SLEEP); 4888 (void) sprintf(oldvd->vdev_path, "%s/%s", 4889 newvd->vdev_path, "old"); 4890 if (oldvd->vdev_devid != NULL) { 4891 spa_strfree(oldvd->vdev_devid); 4892 oldvd->vdev_devid = NULL; 4893 } 4894 } 4895 4896 /* mark the device being resilvered */ 4897 newvd->vdev_resilver_txg = txg; 4898 4899 /* 4900 * If the parent is not a mirror, or if we're replacing, insert the new 4901 * mirror/replacing/spare vdev above oldvd. 4902 */ 4903 if (pvd->vdev_ops != pvops) 4904 pvd = vdev_add_parent(oldvd, pvops); 4905 4906 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4907 ASSERT(pvd->vdev_ops == pvops); 4908 ASSERT(oldvd->vdev_parent == pvd); 4909 4910 /* 4911 * Extract the new device from its root and add it to pvd. 4912 */ 4913 vdev_remove_child(newrootvd, newvd); 4914 newvd->vdev_id = pvd->vdev_children; 4915 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4916 vdev_add_child(pvd, newvd); 4917 4918 tvd = newvd->vdev_top; 4919 ASSERT(pvd->vdev_top == tvd); 4920 ASSERT(tvd->vdev_parent == rvd); 4921 4922 vdev_config_dirty(tvd); 4923 4924 /* 4925 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4926 * for any dmu_sync-ed blocks. It will propagate upward when 4927 * spa_vdev_exit() calls vdev_dtl_reassess(). 4928 */ 4929 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4930 4931 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4932 dtl_max_txg - TXG_INITIAL); 4933 4934 if (newvd->vdev_isspare) { 4935 spa_spare_activate(newvd); 4936 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4937 } 4938 4939 oldvdpath = spa_strdup(oldvd->vdev_path); 4940 newvdpath = spa_strdup(newvd->vdev_path); 4941 newvd_isspare = newvd->vdev_isspare; 4942 4943 /* 4944 * Mark newvd's DTL dirty in this txg. 4945 */ 4946 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4947 4948 /* 4949 * Schedule the resilver to restart in the future. We do this to 4950 * ensure that dmu_sync-ed blocks have been stitched into the 4951 * respective datasets. 4952 */ 4953 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4954 4955 if (spa->spa_bootfs) 4956 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4957 4958 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 4959 4960 /* 4961 * Commit the config 4962 */ 4963 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4964 4965 spa_history_log_internal(spa, "vdev attach", NULL, 4966 "%s vdev=%s %s vdev=%s", 4967 replacing && newvd_isspare ? "spare in" : 4968 replacing ? "replace" : "attach", newvdpath, 4969 replacing ? "for" : "to", oldvdpath); 4970 4971 spa_strfree(oldvdpath); 4972 spa_strfree(newvdpath); 4973 4974 return (0); 4975} 4976 4977/* 4978 * Detach a device from a mirror or replacing vdev. 4979 * 4980 * If 'replace_done' is specified, only detach if the parent 4981 * is a replacing vdev. 4982 */ 4983int 4984spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4985{ 4986 uint64_t txg; 4987 int error; 4988 vdev_t *rvd = spa->spa_root_vdev; 4989 vdev_t *vd, *pvd, *cvd, *tvd; 4990 boolean_t unspare = B_FALSE; 4991 uint64_t unspare_guid = 0; 4992 char *vdpath; 4993 4994 ASSERT(spa_writeable(spa)); 4995 4996 txg = spa_vdev_enter(spa); 4997 4998 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4999 5000 if (vd == NULL) 5001 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5002 5003 if (!vd->vdev_ops->vdev_op_leaf) 5004 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5005 5006 pvd = vd->vdev_parent; 5007 5008 /* 5009 * If the parent/child relationship is not as expected, don't do it. 5010 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 5011 * vdev that's replacing B with C. The user's intent in replacing 5012 * is to go from M(A,B) to M(A,C). If the user decides to cancel 5013 * the replace by detaching C, the expected behavior is to end up 5014 * M(A,B). But suppose that right after deciding to detach C, 5015 * the replacement of B completes. We would have M(A,C), and then 5016 * ask to detach C, which would leave us with just A -- not what 5017 * the user wanted. To prevent this, we make sure that the 5018 * parent/child relationship hasn't changed -- in this example, 5019 * that C's parent is still the replacing vdev R. 5020 */ 5021 if (pvd->vdev_guid != pguid && pguid != 0) 5022 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5023 5024 /* 5025 * Only 'replacing' or 'spare' vdevs can be replaced. 5026 */ 5027 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5028 pvd->vdev_ops != &vdev_spare_ops) 5029 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5030 5031 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5032 spa_version(spa) >= SPA_VERSION_SPARES); 5033 5034 /* 5035 * Only mirror, replacing, and spare vdevs support detach. 5036 */ 5037 if (pvd->vdev_ops != &vdev_replacing_ops && 5038 pvd->vdev_ops != &vdev_mirror_ops && 5039 pvd->vdev_ops != &vdev_spare_ops) 5040 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5041 5042 /* 5043 * If this device has the only valid copy of some data, 5044 * we cannot safely detach it. 5045 */ 5046 if (vdev_dtl_required(vd)) 5047 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5048 5049 ASSERT(pvd->vdev_children >= 2); 5050 5051 /* 5052 * If we are detaching the second disk from a replacing vdev, then 5053 * check to see if we changed the original vdev's path to have "/old" 5054 * at the end in spa_vdev_attach(). If so, undo that change now. 5055 */ 5056 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5057 vd->vdev_path != NULL) { 5058 size_t len = strlen(vd->vdev_path); 5059 5060 for (int c = 0; c < pvd->vdev_children; c++) { 5061 cvd = pvd->vdev_child[c]; 5062 5063 if (cvd == vd || cvd->vdev_path == NULL) 5064 continue; 5065 5066 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5067 strcmp(cvd->vdev_path + len, "/old") == 0) { 5068 spa_strfree(cvd->vdev_path); 5069 cvd->vdev_path = spa_strdup(vd->vdev_path); 5070 break; 5071 } 5072 } 5073 } 5074 5075 /* 5076 * If we are detaching the original disk from a spare, then it implies 5077 * that the spare should become a real disk, and be removed from the 5078 * active spare list for the pool. 5079 */ 5080 if (pvd->vdev_ops == &vdev_spare_ops && 5081 vd->vdev_id == 0 && 5082 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5083 unspare = B_TRUE; 5084 5085 /* 5086 * Erase the disk labels so the disk can be used for other things. 5087 * This must be done after all other error cases are handled, 5088 * but before we disembowel vd (so we can still do I/O to it). 5089 * But if we can't do it, don't treat the error as fatal -- 5090 * it may be that the unwritability of the disk is the reason 5091 * it's being detached! 5092 */ 5093 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5094 5095 /* 5096 * Remove vd from its parent and compact the parent's children. 5097 */ 5098 vdev_remove_child(pvd, vd); 5099 vdev_compact_children(pvd); 5100 5101 /* 5102 * Remember one of the remaining children so we can get tvd below. 5103 */ 5104 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5105 5106 /* 5107 * If we need to remove the remaining child from the list of hot spares, 5108 * do it now, marking the vdev as no longer a spare in the process. 5109 * We must do this before vdev_remove_parent(), because that can 5110 * change the GUID if it creates a new toplevel GUID. For a similar 5111 * reason, we must remove the spare now, in the same txg as the detach; 5112 * otherwise someone could attach a new sibling, change the GUID, and 5113 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5114 */ 5115 if (unspare) { 5116 ASSERT(cvd->vdev_isspare); 5117 spa_spare_remove(cvd); 5118 unspare_guid = cvd->vdev_guid; 5119 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5120 cvd->vdev_unspare = B_TRUE; 5121 } 5122 5123 /* 5124 * If the parent mirror/replacing vdev only has one child, 5125 * the parent is no longer needed. Remove it from the tree. 5126 */ 5127 if (pvd->vdev_children == 1) { 5128 if (pvd->vdev_ops == &vdev_spare_ops) 5129 cvd->vdev_unspare = B_FALSE; 5130 vdev_remove_parent(cvd); 5131 } 5132 5133 5134 /* 5135 * We don't set tvd until now because the parent we just removed 5136 * may have been the previous top-level vdev. 5137 */ 5138 tvd = cvd->vdev_top; 5139 ASSERT(tvd->vdev_parent == rvd); 5140 5141 /* 5142 * Reevaluate the parent vdev state. 5143 */ 5144 vdev_propagate_state(cvd); 5145 5146 /* 5147 * If the 'autoexpand' property is set on the pool then automatically 5148 * try to expand the size of the pool. For example if the device we 5149 * just detached was smaller than the others, it may be possible to 5150 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5151 * first so that we can obtain the updated sizes of the leaf vdevs. 5152 */ 5153 if (spa->spa_autoexpand) { 5154 vdev_reopen(tvd); 5155 vdev_expand(tvd, txg); 5156 } 5157 5158 vdev_config_dirty(tvd); 5159 5160 /* 5161 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5162 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5163 * But first make sure we're not on any *other* txg's DTL list, to 5164 * prevent vd from being accessed after it's freed. 5165 */ 5166 vdpath = spa_strdup(vd->vdev_path); 5167 for (int t = 0; t < TXG_SIZE; t++) 5168 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5169 vd->vdev_detached = B_TRUE; 5170 vdev_dirty(tvd, VDD_DTL, vd, txg); 5171 5172 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5173 5174 /* hang on to the spa before we release the lock */ 5175 spa_open_ref(spa, FTAG); 5176 5177 error = spa_vdev_exit(spa, vd, txg, 0); 5178 5179 spa_history_log_internal(spa, "detach", NULL, 5180 "vdev=%s", vdpath); 5181 spa_strfree(vdpath); 5182 5183 /* 5184 * If this was the removal of the original device in a hot spare vdev, 5185 * then we want to go through and remove the device from the hot spare 5186 * list of every other pool. 5187 */ 5188 if (unspare) { 5189 spa_t *altspa = NULL; 5190 5191 mutex_enter(&spa_namespace_lock); 5192 while ((altspa = spa_next(altspa)) != NULL) { 5193 if (altspa->spa_state != POOL_STATE_ACTIVE || 5194 altspa == spa) 5195 continue; 5196 5197 spa_open_ref(altspa, FTAG); 5198 mutex_exit(&spa_namespace_lock); 5199 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5200 mutex_enter(&spa_namespace_lock); 5201 spa_close(altspa, FTAG); 5202 } 5203 mutex_exit(&spa_namespace_lock); 5204 5205 /* search the rest of the vdevs for spares to remove */ 5206 spa_vdev_resilver_done(spa); 5207 } 5208 5209 /* all done with the spa; OK to release */ 5210 mutex_enter(&spa_namespace_lock); 5211 spa_close(spa, FTAG); 5212 mutex_exit(&spa_namespace_lock); 5213 5214 return (error); 5215} 5216 5217/* 5218 * Split a set of devices from their mirrors, and create a new pool from them. 5219 */ 5220int 5221spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5222 nvlist_t *props, boolean_t exp) 5223{ 5224 int error = 0; 5225 uint64_t txg, *glist; 5226 spa_t *newspa; 5227 uint_t c, children, lastlog; 5228 nvlist_t **child, *nvl, *tmp; 5229 dmu_tx_t *tx; 5230 char *altroot = NULL; 5231 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5232 boolean_t activate_slog; 5233 5234 ASSERT(spa_writeable(spa)); 5235 5236 txg = spa_vdev_enter(spa); 5237 5238 /* clear the log and flush everything up to now */ 5239 activate_slog = spa_passivate_log(spa); 5240 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5241 error = spa_offline_log(spa); 5242 txg = spa_vdev_config_enter(spa); 5243 5244 if (activate_slog) 5245 spa_activate_log(spa); 5246 5247 if (error != 0) 5248 return (spa_vdev_exit(spa, NULL, txg, error)); 5249 5250 /* check new spa name before going any further */ 5251 if (spa_lookup(newname) != NULL) 5252 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5253 5254 /* 5255 * scan through all the children to ensure they're all mirrors 5256 */ 5257 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5258 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5259 &children) != 0) 5260 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5261 5262 /* first, check to ensure we've got the right child count */ 5263 rvd = spa->spa_root_vdev; 5264 lastlog = 0; 5265 for (c = 0; c < rvd->vdev_children; c++) { 5266 vdev_t *vd = rvd->vdev_child[c]; 5267 5268 /* don't count the holes & logs as children */ 5269 if (vd->vdev_islog || vd->vdev_ishole) { 5270 if (lastlog == 0) 5271 lastlog = c; 5272 continue; 5273 } 5274 5275 lastlog = 0; 5276 } 5277 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5278 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5279 5280 /* next, ensure no spare or cache devices are part of the split */ 5281 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5282 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5283 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5284 5285 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5286 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5287 5288 /* then, loop over each vdev and validate it */ 5289 for (c = 0; c < children; c++) { 5290 uint64_t is_hole = 0; 5291 5292 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5293 &is_hole); 5294 5295 if (is_hole != 0) { 5296 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5297 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5298 continue; 5299 } else { 5300 error = SET_ERROR(EINVAL); 5301 break; 5302 } 5303 } 5304 5305 /* which disk is going to be split? */ 5306 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5307 &glist[c]) != 0) { 5308 error = SET_ERROR(EINVAL); 5309 break; 5310 } 5311 5312 /* look it up in the spa */ 5313 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5314 if (vml[c] == NULL) { 5315 error = SET_ERROR(ENODEV); 5316 break; 5317 } 5318 5319 /* make sure there's nothing stopping the split */ 5320 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5321 vml[c]->vdev_islog || 5322 vml[c]->vdev_ishole || 5323 vml[c]->vdev_isspare || 5324 vml[c]->vdev_isl2cache || 5325 !vdev_writeable(vml[c]) || 5326 vml[c]->vdev_children != 0 || 5327 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5328 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5329 error = SET_ERROR(EINVAL); 5330 break; 5331 } 5332 5333 if (vdev_dtl_required(vml[c])) { 5334 error = SET_ERROR(EBUSY); 5335 break; 5336 } 5337 5338 /* we need certain info from the top level */ 5339 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5340 vml[c]->vdev_top->vdev_ms_array) == 0); 5341 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5342 vml[c]->vdev_top->vdev_ms_shift) == 0); 5343 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5344 vml[c]->vdev_top->vdev_asize) == 0); 5345 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5346 vml[c]->vdev_top->vdev_ashift) == 0); 5347 } 5348 5349 if (error != 0) { 5350 kmem_free(vml, children * sizeof (vdev_t *)); 5351 kmem_free(glist, children * sizeof (uint64_t)); 5352 return (spa_vdev_exit(spa, NULL, txg, error)); 5353 } 5354 5355 /* stop writers from using the disks */ 5356 for (c = 0; c < children; c++) { 5357 if (vml[c] != NULL) 5358 vml[c]->vdev_offline = B_TRUE; 5359 } 5360 vdev_reopen(spa->spa_root_vdev); 5361 5362 /* 5363 * Temporarily record the splitting vdevs in the spa config. This 5364 * will disappear once the config is regenerated. 5365 */ 5366 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5367 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5368 glist, children) == 0); 5369 kmem_free(glist, children * sizeof (uint64_t)); 5370 5371 mutex_enter(&spa->spa_props_lock); 5372 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5373 nvl) == 0); 5374 mutex_exit(&spa->spa_props_lock); 5375 spa->spa_config_splitting = nvl; 5376 vdev_config_dirty(spa->spa_root_vdev); 5377 5378 /* configure and create the new pool */ 5379 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5380 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5381 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5382 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5383 spa_version(spa)) == 0); 5384 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5385 spa->spa_config_txg) == 0); 5386 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5387 spa_generate_guid(NULL)) == 0); 5388 (void) nvlist_lookup_string(props, 5389 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5390 5391 /* add the new pool to the namespace */ 5392 newspa = spa_add(newname, config, altroot); 5393 newspa->spa_config_txg = spa->spa_config_txg; 5394 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5395 5396 /* release the spa config lock, retaining the namespace lock */ 5397 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5398 5399 if (zio_injection_enabled) 5400 zio_handle_panic_injection(spa, FTAG, 1); 5401 5402 spa_activate(newspa, spa_mode_global); 5403 spa_async_suspend(newspa); 5404 5405#ifndef illumos 5406 /* mark that we are creating new spa by splitting */ 5407 newspa->spa_splitting_newspa = B_TRUE; 5408#endif 5409 /* create the new pool from the disks of the original pool */ 5410 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5411#ifndef illumos 5412 newspa->spa_splitting_newspa = B_FALSE; 5413#endif 5414 if (error) 5415 goto out; 5416 5417 /* if that worked, generate a real config for the new pool */ 5418 if (newspa->spa_root_vdev != NULL) { 5419 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5420 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5421 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5422 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5423 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5424 B_TRUE)); 5425 } 5426 5427 /* set the props */ 5428 if (props != NULL) { 5429 spa_configfile_set(newspa, props, B_FALSE); 5430 error = spa_prop_set(newspa, props); 5431 if (error) 5432 goto out; 5433 } 5434 5435 /* flush everything */ 5436 txg = spa_vdev_config_enter(newspa); 5437 vdev_config_dirty(newspa->spa_root_vdev); 5438 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5439 5440 if (zio_injection_enabled) 5441 zio_handle_panic_injection(spa, FTAG, 2); 5442 5443 spa_async_resume(newspa); 5444 5445 /* finally, update the original pool's config */ 5446 txg = spa_vdev_config_enter(spa); 5447 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5448 error = dmu_tx_assign(tx, TXG_WAIT); 5449 if (error != 0) 5450 dmu_tx_abort(tx); 5451 for (c = 0; c < children; c++) { 5452 if (vml[c] != NULL) { 5453 vdev_split(vml[c]); 5454 if (error == 0) 5455 spa_history_log_internal(spa, "detach", tx, 5456 "vdev=%s", vml[c]->vdev_path); 5457 vdev_free(vml[c]); 5458 } 5459 } 5460 vdev_config_dirty(spa->spa_root_vdev); 5461 spa->spa_config_splitting = NULL; 5462 nvlist_free(nvl); 5463 if (error == 0) 5464 dmu_tx_commit(tx); 5465 (void) spa_vdev_exit(spa, NULL, txg, 0); 5466 5467 if (zio_injection_enabled) 5468 zio_handle_panic_injection(spa, FTAG, 3); 5469 5470 /* split is complete; log a history record */ 5471 spa_history_log_internal(newspa, "split", NULL, 5472 "from pool %s", spa_name(spa)); 5473 5474 kmem_free(vml, children * sizeof (vdev_t *)); 5475 5476 /* if we're not going to mount the filesystems in userland, export */ 5477 if (exp) 5478 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5479 B_FALSE, B_FALSE); 5480 5481 return (error); 5482 5483out: 5484 spa_unload(newspa); 5485 spa_deactivate(newspa); 5486 spa_remove(newspa); 5487 5488 txg = spa_vdev_config_enter(spa); 5489 5490 /* re-online all offlined disks */ 5491 for (c = 0; c < children; c++) { 5492 if (vml[c] != NULL) 5493 vml[c]->vdev_offline = B_FALSE; 5494 } 5495 vdev_reopen(spa->spa_root_vdev); 5496 5497 nvlist_free(spa->spa_config_splitting); 5498 spa->spa_config_splitting = NULL; 5499 (void) spa_vdev_exit(spa, NULL, txg, error); 5500 5501 kmem_free(vml, children * sizeof (vdev_t *)); 5502 return (error); 5503} 5504 5505static nvlist_t * 5506spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5507{ 5508 for (int i = 0; i < count; i++) { 5509 uint64_t guid; 5510 5511 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5512 &guid) == 0); 5513 5514 if (guid == target_guid) 5515 return (nvpp[i]); 5516 } 5517 5518 return (NULL); 5519} 5520 5521static void 5522spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5523 nvlist_t *dev_to_remove) 5524{ 5525 nvlist_t **newdev = NULL; 5526 5527 if (count > 1) 5528 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5529 5530 for (int i = 0, j = 0; i < count; i++) { 5531 if (dev[i] == dev_to_remove) 5532 continue; 5533 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5534 } 5535 5536 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5537 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5538 5539 for (int i = 0; i < count - 1; i++) 5540 nvlist_free(newdev[i]); 5541 5542 if (count > 1) 5543 kmem_free(newdev, (count - 1) * sizeof (void *)); 5544} 5545 5546/* 5547 * Evacuate the device. 5548 */ 5549static int 5550spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5551{ 5552 uint64_t txg; 5553 int error = 0; 5554 5555 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5556 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5557 ASSERT(vd == vd->vdev_top); 5558 5559 /* 5560 * Evacuate the device. We don't hold the config lock as writer 5561 * since we need to do I/O but we do keep the 5562 * spa_namespace_lock held. Once this completes the device 5563 * should no longer have any blocks allocated on it. 5564 */ 5565 if (vd->vdev_islog) { 5566 if (vd->vdev_stat.vs_alloc != 0) 5567 error = spa_offline_log(spa); 5568 } else { 5569 error = SET_ERROR(ENOTSUP); 5570 } 5571 5572 if (error) 5573 return (error); 5574 5575 /* 5576 * The evacuation succeeded. Remove any remaining MOS metadata 5577 * associated with this vdev, and wait for these changes to sync. 5578 */ 5579 ASSERT0(vd->vdev_stat.vs_alloc); 5580 txg = spa_vdev_config_enter(spa); 5581 vd->vdev_removing = B_TRUE; 5582 vdev_dirty_leaves(vd, VDD_DTL, txg); 5583 vdev_config_dirty(vd); 5584 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5585 5586 return (0); 5587} 5588 5589/* 5590 * Complete the removal by cleaning up the namespace. 5591 */ 5592static void 5593spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5594{ 5595 vdev_t *rvd = spa->spa_root_vdev; 5596 uint64_t id = vd->vdev_id; 5597 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5598 5599 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5600 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5601 ASSERT(vd == vd->vdev_top); 5602 5603 /* 5604 * Only remove any devices which are empty. 5605 */ 5606 if (vd->vdev_stat.vs_alloc != 0) 5607 return; 5608 5609 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5610 5611 if (list_link_active(&vd->vdev_state_dirty_node)) 5612 vdev_state_clean(vd); 5613 if (list_link_active(&vd->vdev_config_dirty_node)) 5614 vdev_config_clean(vd); 5615 5616 vdev_free(vd); 5617 5618 if (last_vdev) { 5619 vdev_compact_children(rvd); 5620 } else { 5621 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5622 vdev_add_child(rvd, vd); 5623 } 5624 vdev_config_dirty(rvd); 5625 5626 /* 5627 * Reassess the health of our root vdev. 5628 */ 5629 vdev_reopen(rvd); 5630} 5631 5632/* 5633 * Remove a device from the pool - 5634 * 5635 * Removing a device from the vdev namespace requires several steps 5636 * and can take a significant amount of time. As a result we use 5637 * the spa_vdev_config_[enter/exit] functions which allow us to 5638 * grab and release the spa_config_lock while still holding the namespace 5639 * lock. During each step the configuration is synced out. 5640 * 5641 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5642 * devices. 5643 */ 5644int 5645spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5646{ 5647 vdev_t *vd; 5648 metaslab_group_t *mg; 5649 nvlist_t **spares, **l2cache, *nv; 5650 uint64_t txg = 0; 5651 uint_t nspares, nl2cache; 5652 int error = 0; 5653 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5654 5655 ASSERT(spa_writeable(spa)); 5656 5657 if (!locked) 5658 txg = spa_vdev_enter(spa); 5659 5660 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5661 5662 if (spa->spa_spares.sav_vdevs != NULL && 5663 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5664 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5665 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5666 /* 5667 * Only remove the hot spare if it's not currently in use 5668 * in this pool. 5669 */ 5670 if (vd == NULL || unspare) { 5671 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5672 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5673 spa_load_spares(spa); 5674 spa->spa_spares.sav_sync = B_TRUE; 5675 } else { 5676 error = SET_ERROR(EBUSY); 5677 } 5678 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5679 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5680 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5681 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5682 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5683 /* 5684 * Cache devices can always be removed. 5685 */ 5686 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5687 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5688 spa_load_l2cache(spa); 5689 spa->spa_l2cache.sav_sync = B_TRUE; 5690 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5691 } else if (vd != NULL && vd->vdev_islog) { 5692 ASSERT(!locked); 5693 ASSERT(vd == vd->vdev_top); 5694 5695 mg = vd->vdev_mg; 5696 5697 /* 5698 * Stop allocating from this vdev. 5699 */ 5700 metaslab_group_passivate(mg); 5701 5702 /* 5703 * Wait for the youngest allocations and frees to sync, 5704 * and then wait for the deferral of those frees to finish. 5705 */ 5706 spa_vdev_config_exit(spa, NULL, 5707 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5708 5709 /* 5710 * Attempt to evacuate the vdev. 5711 */ 5712 error = spa_vdev_remove_evacuate(spa, vd); 5713 5714 txg = spa_vdev_config_enter(spa); 5715 5716 /* 5717 * If we couldn't evacuate the vdev, unwind. 5718 */ 5719 if (error) { 5720 metaslab_group_activate(mg); 5721 return (spa_vdev_exit(spa, NULL, txg, error)); 5722 } 5723 5724 /* 5725 * Clean up the vdev namespace. 5726 */ 5727 spa_vdev_remove_from_namespace(spa, vd); 5728 5729 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV); 5730 } else if (vd != NULL) { 5731 /* 5732 * Normal vdevs cannot be removed (yet). 5733 */ 5734 error = SET_ERROR(ENOTSUP); 5735 } else { 5736 /* 5737 * There is no vdev of any kind with the specified guid. 5738 */ 5739 error = SET_ERROR(ENOENT); 5740 } 5741 5742 if (!locked) 5743 error = spa_vdev_exit(spa, NULL, txg, error); 5744 5745 return (error); 5746} 5747 5748/* 5749 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5750 * currently spared, so we can detach it. 5751 */ 5752static vdev_t * 5753spa_vdev_resilver_done_hunt(vdev_t *vd) 5754{ 5755 vdev_t *newvd, *oldvd; 5756 5757 for (int c = 0; c < vd->vdev_children; c++) { 5758 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5759 if (oldvd != NULL) 5760 return (oldvd); 5761 } 5762 5763 /* 5764 * Check for a completed replacement. We always consider the first 5765 * vdev in the list to be the oldest vdev, and the last one to be 5766 * the newest (see spa_vdev_attach() for how that works). In 5767 * the case where the newest vdev is faulted, we will not automatically 5768 * remove it after a resilver completes. This is OK as it will require 5769 * user intervention to determine which disk the admin wishes to keep. 5770 */ 5771 if (vd->vdev_ops == &vdev_replacing_ops) { 5772 ASSERT(vd->vdev_children > 1); 5773 5774 newvd = vd->vdev_child[vd->vdev_children - 1]; 5775 oldvd = vd->vdev_child[0]; 5776 5777 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5778 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5779 !vdev_dtl_required(oldvd)) 5780 return (oldvd); 5781 } 5782 5783 /* 5784 * Check for a completed resilver with the 'unspare' flag set. 5785 */ 5786 if (vd->vdev_ops == &vdev_spare_ops) { 5787 vdev_t *first = vd->vdev_child[0]; 5788 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5789 5790 if (last->vdev_unspare) { 5791 oldvd = first; 5792 newvd = last; 5793 } else if (first->vdev_unspare) { 5794 oldvd = last; 5795 newvd = first; 5796 } else { 5797 oldvd = NULL; 5798 } 5799 5800 if (oldvd != NULL && 5801 vdev_dtl_empty(newvd, DTL_MISSING) && 5802 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5803 !vdev_dtl_required(oldvd)) 5804 return (oldvd); 5805 5806 /* 5807 * If there are more than two spares attached to a disk, 5808 * and those spares are not required, then we want to 5809 * attempt to free them up now so that they can be used 5810 * by other pools. Once we're back down to a single 5811 * disk+spare, we stop removing them. 5812 */ 5813 if (vd->vdev_children > 2) { 5814 newvd = vd->vdev_child[1]; 5815 5816 if (newvd->vdev_isspare && last->vdev_isspare && 5817 vdev_dtl_empty(last, DTL_MISSING) && 5818 vdev_dtl_empty(last, DTL_OUTAGE) && 5819 !vdev_dtl_required(newvd)) 5820 return (newvd); 5821 } 5822 } 5823 5824 return (NULL); 5825} 5826 5827static void 5828spa_vdev_resilver_done(spa_t *spa) 5829{ 5830 vdev_t *vd, *pvd, *ppvd; 5831 uint64_t guid, sguid, pguid, ppguid; 5832 5833 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5834 5835 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5836 pvd = vd->vdev_parent; 5837 ppvd = pvd->vdev_parent; 5838 guid = vd->vdev_guid; 5839 pguid = pvd->vdev_guid; 5840 ppguid = ppvd->vdev_guid; 5841 sguid = 0; 5842 /* 5843 * If we have just finished replacing a hot spared device, then 5844 * we need to detach the parent's first child (the original hot 5845 * spare) as well. 5846 */ 5847 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5848 ppvd->vdev_children == 2) { 5849 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5850 sguid = ppvd->vdev_child[1]->vdev_guid; 5851 } 5852 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5853 5854 spa_config_exit(spa, SCL_ALL, FTAG); 5855 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5856 return; 5857 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5858 return; 5859 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5860 } 5861 5862 spa_config_exit(spa, SCL_ALL, FTAG); 5863} 5864 5865/* 5866 * Update the stored path or FRU for this vdev. 5867 */ 5868int 5869spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5870 boolean_t ispath) 5871{ 5872 vdev_t *vd; 5873 boolean_t sync = B_FALSE; 5874 5875 ASSERT(spa_writeable(spa)); 5876 5877 spa_vdev_state_enter(spa, SCL_ALL); 5878 5879 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5880 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5881 5882 if (!vd->vdev_ops->vdev_op_leaf) 5883 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5884 5885 if (ispath) { 5886 if (strcmp(value, vd->vdev_path) != 0) { 5887 spa_strfree(vd->vdev_path); 5888 vd->vdev_path = spa_strdup(value); 5889 sync = B_TRUE; 5890 } 5891 } else { 5892 if (vd->vdev_fru == NULL) { 5893 vd->vdev_fru = spa_strdup(value); 5894 sync = B_TRUE; 5895 } else if (strcmp(value, vd->vdev_fru) != 0) { 5896 spa_strfree(vd->vdev_fru); 5897 vd->vdev_fru = spa_strdup(value); 5898 sync = B_TRUE; 5899 } 5900 } 5901 5902 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5903} 5904 5905int 5906spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5907{ 5908 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5909} 5910 5911int 5912spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5913{ 5914 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5915} 5916 5917/* 5918 * ========================================================================== 5919 * SPA Scanning 5920 * ========================================================================== 5921 */ 5922 5923int 5924spa_scan_stop(spa_t *spa) 5925{ 5926 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5927 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5928 return (SET_ERROR(EBUSY)); 5929 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5930} 5931 5932int 5933spa_scan(spa_t *spa, pool_scan_func_t func) 5934{ 5935 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5936 5937 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5938 return (SET_ERROR(ENOTSUP)); 5939 5940 /* 5941 * If a resilver was requested, but there is no DTL on a 5942 * writeable leaf device, we have nothing to do. 5943 */ 5944 if (func == POOL_SCAN_RESILVER && 5945 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5946 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5947 return (0); 5948 } 5949 5950 return (dsl_scan(spa->spa_dsl_pool, func)); 5951} 5952 5953/* 5954 * ========================================================================== 5955 * SPA async task processing 5956 * ========================================================================== 5957 */ 5958 5959static void 5960spa_async_remove(spa_t *spa, vdev_t *vd) 5961{ 5962 if (vd->vdev_remove_wanted) { 5963 vd->vdev_remove_wanted = B_FALSE; 5964 vd->vdev_delayed_close = B_FALSE; 5965 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5966 5967 /* 5968 * We want to clear the stats, but we don't want to do a full 5969 * vdev_clear() as that will cause us to throw away 5970 * degraded/faulted state as well as attempt to reopen the 5971 * device, all of which is a waste. 5972 */ 5973 vd->vdev_stat.vs_read_errors = 0; 5974 vd->vdev_stat.vs_write_errors = 0; 5975 vd->vdev_stat.vs_checksum_errors = 0; 5976 5977 vdev_state_dirty(vd->vdev_top); 5978 /* Tell userspace that the vdev is gone. */ 5979 zfs_post_remove(spa, vd); 5980 } 5981 5982 for (int c = 0; c < vd->vdev_children; c++) 5983 spa_async_remove(spa, vd->vdev_child[c]); 5984} 5985 5986static void 5987spa_async_probe(spa_t *spa, vdev_t *vd) 5988{ 5989 if (vd->vdev_probe_wanted) { 5990 vd->vdev_probe_wanted = B_FALSE; 5991 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5992 } 5993 5994 for (int c = 0; c < vd->vdev_children; c++) 5995 spa_async_probe(spa, vd->vdev_child[c]); 5996} 5997 5998static void 5999spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6000{ 6001 sysevent_id_t eid; 6002 nvlist_t *attr; 6003 char *physpath; 6004 6005 if (!spa->spa_autoexpand) 6006 return; 6007 6008 for (int c = 0; c < vd->vdev_children; c++) { 6009 vdev_t *cvd = vd->vdev_child[c]; 6010 spa_async_autoexpand(spa, cvd); 6011 } 6012 6013 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6014 return; 6015 6016 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6017 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6018 6019 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6020 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6021 6022 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6023 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 6024 6025 nvlist_free(attr); 6026 kmem_free(physpath, MAXPATHLEN); 6027} 6028 6029static void 6030spa_async_thread(void *arg) 6031{ 6032 spa_t *spa = arg; 6033 int tasks; 6034 6035 ASSERT(spa->spa_sync_on); 6036 6037 mutex_enter(&spa->spa_async_lock); 6038 tasks = spa->spa_async_tasks; 6039 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 6040 mutex_exit(&spa->spa_async_lock); 6041 6042 /* 6043 * See if the config needs to be updated. 6044 */ 6045 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6046 uint64_t old_space, new_space; 6047 6048 mutex_enter(&spa_namespace_lock); 6049 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6050 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6051 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6052 mutex_exit(&spa_namespace_lock); 6053 6054 /* 6055 * If the pool grew as a result of the config update, 6056 * then log an internal history event. 6057 */ 6058 if (new_space != old_space) { 6059 spa_history_log_internal(spa, "vdev online", NULL, 6060 "pool '%s' size: %llu(+%llu)", 6061 spa_name(spa), new_space, new_space - old_space); 6062 } 6063 } 6064 6065 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6066 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6067 spa_async_autoexpand(spa, spa->spa_root_vdev); 6068 spa_config_exit(spa, SCL_CONFIG, FTAG); 6069 } 6070 6071 /* 6072 * See if any devices need to be probed. 6073 */ 6074 if (tasks & SPA_ASYNC_PROBE) { 6075 spa_vdev_state_enter(spa, SCL_NONE); 6076 spa_async_probe(spa, spa->spa_root_vdev); 6077 (void) spa_vdev_state_exit(spa, NULL, 0); 6078 } 6079 6080 /* 6081 * If any devices are done replacing, detach them. 6082 */ 6083 if (tasks & SPA_ASYNC_RESILVER_DONE) 6084 spa_vdev_resilver_done(spa); 6085 6086 /* 6087 * Kick off a resilver. 6088 */ 6089 if (tasks & SPA_ASYNC_RESILVER) 6090 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6091 6092 /* 6093 * Let the world know that we're done. 6094 */ 6095 mutex_enter(&spa->spa_async_lock); 6096 spa->spa_async_thread = NULL; 6097 cv_broadcast(&spa->spa_async_cv); 6098 mutex_exit(&spa->spa_async_lock); 6099 thread_exit(); 6100} 6101 6102static void 6103spa_async_thread_vd(void *arg) 6104{ 6105 spa_t *spa = arg; 6106 int tasks; 6107 6108 ASSERT(spa->spa_sync_on); 6109 6110 mutex_enter(&spa->spa_async_lock); 6111 tasks = spa->spa_async_tasks; 6112retry: 6113 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6114 mutex_exit(&spa->spa_async_lock); 6115 6116 /* 6117 * See if any devices need to be marked REMOVED. 6118 */ 6119 if (tasks & SPA_ASYNC_REMOVE) { 6120 spa_vdev_state_enter(spa, SCL_NONE); 6121 spa_async_remove(spa, spa->spa_root_vdev); 6122 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6123 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6124 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6125 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6126 (void) spa_vdev_state_exit(spa, NULL, 0); 6127 } 6128 6129 /* 6130 * Let the world know that we're done. 6131 */ 6132 mutex_enter(&spa->spa_async_lock); 6133 tasks = spa->spa_async_tasks; 6134 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6135 goto retry; 6136 spa->spa_async_thread_vd = NULL; 6137 cv_broadcast(&spa->spa_async_cv); 6138 mutex_exit(&spa->spa_async_lock); 6139 thread_exit(); 6140} 6141 6142void 6143spa_async_suspend(spa_t *spa) 6144{ 6145 mutex_enter(&spa->spa_async_lock); 6146 spa->spa_async_suspended++; 6147 while (spa->spa_async_thread != NULL && 6148 spa->spa_async_thread_vd != NULL) 6149 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6150 mutex_exit(&spa->spa_async_lock); 6151} 6152 6153void 6154spa_async_resume(spa_t *spa) 6155{ 6156 mutex_enter(&spa->spa_async_lock); 6157 ASSERT(spa->spa_async_suspended != 0); 6158 spa->spa_async_suspended--; 6159 mutex_exit(&spa->spa_async_lock); 6160} 6161 6162static boolean_t 6163spa_async_tasks_pending(spa_t *spa) 6164{ 6165 uint_t non_config_tasks; 6166 uint_t config_task; 6167 boolean_t config_task_suspended; 6168 6169 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6170 SPA_ASYNC_REMOVE); 6171 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6172 if (spa->spa_ccw_fail_time == 0) { 6173 config_task_suspended = B_FALSE; 6174 } else { 6175 config_task_suspended = 6176 (gethrtime() - spa->spa_ccw_fail_time) < 6177 (zfs_ccw_retry_interval * NANOSEC); 6178 } 6179 6180 return (non_config_tasks || (config_task && !config_task_suspended)); 6181} 6182 6183static void 6184spa_async_dispatch(spa_t *spa) 6185{ 6186 mutex_enter(&spa->spa_async_lock); 6187 if (spa_async_tasks_pending(spa) && 6188 !spa->spa_async_suspended && 6189 spa->spa_async_thread == NULL && 6190 rootdir != NULL) 6191 spa->spa_async_thread = thread_create(NULL, 0, 6192 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6193 mutex_exit(&spa->spa_async_lock); 6194} 6195 6196static void 6197spa_async_dispatch_vd(spa_t *spa) 6198{ 6199 mutex_enter(&spa->spa_async_lock); 6200 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6201 !spa->spa_async_suspended && 6202 spa->spa_async_thread_vd == NULL && 6203 rootdir != NULL) 6204 spa->spa_async_thread_vd = thread_create(NULL, 0, 6205 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6206 mutex_exit(&spa->spa_async_lock); 6207} 6208 6209void 6210spa_async_request(spa_t *spa, int task) 6211{ 6212 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6213 mutex_enter(&spa->spa_async_lock); 6214 spa->spa_async_tasks |= task; 6215 mutex_exit(&spa->spa_async_lock); 6216 spa_async_dispatch_vd(spa); 6217} 6218 6219/* 6220 * ========================================================================== 6221 * SPA syncing routines 6222 * ========================================================================== 6223 */ 6224 6225static int 6226bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6227{ 6228 bpobj_t *bpo = arg; 6229 bpobj_enqueue(bpo, bp, tx); 6230 return (0); 6231} 6232 6233static int 6234spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6235{ 6236 zio_t *zio = arg; 6237 6238 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6239 BP_GET_PSIZE(bp), zio->io_flags)); 6240 return (0); 6241} 6242 6243/* 6244 * Note: this simple function is not inlined to make it easier to dtrace the 6245 * amount of time spent syncing frees. 6246 */ 6247static void 6248spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6249{ 6250 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6251 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6252 VERIFY(zio_wait(zio) == 0); 6253} 6254 6255/* 6256 * Note: this simple function is not inlined to make it easier to dtrace the 6257 * amount of time spent syncing deferred frees. 6258 */ 6259static void 6260spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6261{ 6262 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6263 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6264 spa_free_sync_cb, zio, tx), ==, 0); 6265 VERIFY0(zio_wait(zio)); 6266} 6267 6268 6269static void 6270spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6271{ 6272 char *packed = NULL; 6273 size_t bufsize; 6274 size_t nvsize = 0; 6275 dmu_buf_t *db; 6276 6277 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6278 6279 /* 6280 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6281 * information. This avoids the dmu_buf_will_dirty() path and 6282 * saves us a pre-read to get data we don't actually care about. 6283 */ 6284 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6285 packed = kmem_alloc(bufsize, KM_SLEEP); 6286 6287 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6288 KM_SLEEP) == 0); 6289 bzero(packed + nvsize, bufsize - nvsize); 6290 6291 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6292 6293 kmem_free(packed, bufsize); 6294 6295 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6296 dmu_buf_will_dirty(db, tx); 6297 *(uint64_t *)db->db_data = nvsize; 6298 dmu_buf_rele(db, FTAG); 6299} 6300 6301static void 6302spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6303 const char *config, const char *entry) 6304{ 6305 nvlist_t *nvroot; 6306 nvlist_t **list; 6307 int i; 6308 6309 if (!sav->sav_sync) 6310 return; 6311 6312 /* 6313 * Update the MOS nvlist describing the list of available devices. 6314 * spa_validate_aux() will have already made sure this nvlist is 6315 * valid and the vdevs are labeled appropriately. 6316 */ 6317 if (sav->sav_object == 0) { 6318 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6319 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6320 sizeof (uint64_t), tx); 6321 VERIFY(zap_update(spa->spa_meta_objset, 6322 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6323 &sav->sav_object, tx) == 0); 6324 } 6325 6326 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6327 if (sav->sav_count == 0) { 6328 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6329 } else { 6330 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6331 for (i = 0; i < sav->sav_count; i++) 6332 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6333 B_FALSE, VDEV_CONFIG_L2CACHE); 6334 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6335 sav->sav_count) == 0); 6336 for (i = 0; i < sav->sav_count; i++) 6337 nvlist_free(list[i]); 6338 kmem_free(list, sav->sav_count * sizeof (void *)); 6339 } 6340 6341 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6342 nvlist_free(nvroot); 6343 6344 sav->sav_sync = B_FALSE; 6345} 6346 6347static void 6348spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6349{ 6350 nvlist_t *config; 6351 6352 if (list_is_empty(&spa->spa_config_dirty_list)) 6353 return; 6354 6355 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6356 6357 config = spa_config_generate(spa, spa->spa_root_vdev, 6358 dmu_tx_get_txg(tx), B_FALSE); 6359 6360 /* 6361 * If we're upgrading the spa version then make sure that 6362 * the config object gets updated with the correct version. 6363 */ 6364 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6365 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6366 spa->spa_uberblock.ub_version); 6367 6368 spa_config_exit(spa, SCL_STATE, FTAG); 6369 6370 nvlist_free(spa->spa_config_syncing); 6371 spa->spa_config_syncing = config; 6372 6373 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6374} 6375 6376static void 6377spa_sync_version(void *arg, dmu_tx_t *tx) 6378{ 6379 uint64_t *versionp = arg; 6380 uint64_t version = *versionp; 6381 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6382 6383 /* 6384 * Setting the version is special cased when first creating the pool. 6385 */ 6386 ASSERT(tx->tx_txg != TXG_INITIAL); 6387 6388 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6389 ASSERT(version >= spa_version(spa)); 6390 6391 spa->spa_uberblock.ub_version = version; 6392 vdev_config_dirty(spa->spa_root_vdev); 6393 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6394} 6395 6396/* 6397 * Set zpool properties. 6398 */ 6399static void 6400spa_sync_props(void *arg, dmu_tx_t *tx) 6401{ 6402 nvlist_t *nvp = arg; 6403 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6404 objset_t *mos = spa->spa_meta_objset; 6405 nvpair_t *elem = NULL; 6406 6407 mutex_enter(&spa->spa_props_lock); 6408 6409 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6410 uint64_t intval; 6411 char *strval, *fname; 6412 zpool_prop_t prop; 6413 const char *propname; 6414 zprop_type_t proptype; 6415 spa_feature_t fid; 6416 6417 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6418 case ZPROP_INVAL: 6419 /* 6420 * We checked this earlier in spa_prop_validate(). 6421 */ 6422 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6423 6424 fname = strchr(nvpair_name(elem), '@') + 1; 6425 VERIFY0(zfeature_lookup_name(fname, &fid)); 6426 6427 spa_feature_enable(spa, fid, tx); 6428 spa_history_log_internal(spa, "set", tx, 6429 "%s=enabled", nvpair_name(elem)); 6430 break; 6431 6432 case ZPOOL_PROP_VERSION: 6433 intval = fnvpair_value_uint64(elem); 6434 /* 6435 * The version is synced seperatly before other 6436 * properties and should be correct by now. 6437 */ 6438 ASSERT3U(spa_version(spa), >=, intval); 6439 break; 6440 6441 case ZPOOL_PROP_ALTROOT: 6442 /* 6443 * 'altroot' is a non-persistent property. It should 6444 * have been set temporarily at creation or import time. 6445 */ 6446 ASSERT(spa->spa_root != NULL); 6447 break; 6448 6449 case ZPOOL_PROP_READONLY: 6450 case ZPOOL_PROP_CACHEFILE: 6451 /* 6452 * 'readonly' and 'cachefile' are also non-persisitent 6453 * properties. 6454 */ 6455 break; 6456 case ZPOOL_PROP_COMMENT: 6457 strval = fnvpair_value_string(elem); 6458 if (spa->spa_comment != NULL) 6459 spa_strfree(spa->spa_comment); 6460 spa->spa_comment = spa_strdup(strval); 6461 /* 6462 * We need to dirty the configuration on all the vdevs 6463 * so that their labels get updated. It's unnecessary 6464 * to do this for pool creation since the vdev's 6465 * configuratoin has already been dirtied. 6466 */ 6467 if (tx->tx_txg != TXG_INITIAL) 6468 vdev_config_dirty(spa->spa_root_vdev); 6469 spa_history_log_internal(spa, "set", tx, 6470 "%s=%s", nvpair_name(elem), strval); 6471 break; 6472 default: 6473 /* 6474 * Set pool property values in the poolprops mos object. 6475 */ 6476 if (spa->spa_pool_props_object == 0) { 6477 spa->spa_pool_props_object = 6478 zap_create_link(mos, DMU_OT_POOL_PROPS, 6479 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6480 tx); 6481 } 6482 6483 /* normalize the property name */ 6484 propname = zpool_prop_to_name(prop); 6485 proptype = zpool_prop_get_type(prop); 6486 6487 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6488 ASSERT(proptype == PROP_TYPE_STRING); 6489 strval = fnvpair_value_string(elem); 6490 VERIFY0(zap_update(mos, 6491 spa->spa_pool_props_object, propname, 6492 1, strlen(strval) + 1, strval, tx)); 6493 spa_history_log_internal(spa, "set", tx, 6494 "%s=%s", nvpair_name(elem), strval); 6495 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6496 intval = fnvpair_value_uint64(elem); 6497 6498 if (proptype == PROP_TYPE_INDEX) { 6499 const char *unused; 6500 VERIFY0(zpool_prop_index_to_string( 6501 prop, intval, &unused)); 6502 } 6503 VERIFY0(zap_update(mos, 6504 spa->spa_pool_props_object, propname, 6505 8, 1, &intval, tx)); 6506 spa_history_log_internal(spa, "set", tx, 6507 "%s=%lld", nvpair_name(elem), intval); 6508 } else { 6509 ASSERT(0); /* not allowed */ 6510 } 6511 6512 switch (prop) { 6513 case ZPOOL_PROP_DELEGATION: 6514 spa->spa_delegation = intval; 6515 break; 6516 case ZPOOL_PROP_BOOTFS: 6517 spa->spa_bootfs = intval; 6518 break; 6519 case ZPOOL_PROP_FAILUREMODE: 6520 spa->spa_failmode = intval; 6521 break; 6522 case ZPOOL_PROP_AUTOEXPAND: 6523 spa->spa_autoexpand = intval; 6524 if (tx->tx_txg != TXG_INITIAL) 6525 spa_async_request(spa, 6526 SPA_ASYNC_AUTOEXPAND); 6527 break; 6528 case ZPOOL_PROP_DEDUPDITTO: 6529 spa->spa_dedup_ditto = intval; 6530 break; 6531 default: 6532 break; 6533 } 6534 } 6535 6536 } 6537 6538 mutex_exit(&spa->spa_props_lock); 6539} 6540 6541/* 6542 * Perform one-time upgrade on-disk changes. spa_version() does not 6543 * reflect the new version this txg, so there must be no changes this 6544 * txg to anything that the upgrade code depends on after it executes. 6545 * Therefore this must be called after dsl_pool_sync() does the sync 6546 * tasks. 6547 */ 6548static void 6549spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6550{ 6551 dsl_pool_t *dp = spa->spa_dsl_pool; 6552 6553 ASSERT(spa->spa_sync_pass == 1); 6554 6555 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6556 6557 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6558 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6559 dsl_pool_create_origin(dp, tx); 6560 6561 /* Keeping the origin open increases spa_minref */ 6562 spa->spa_minref += 3; 6563 } 6564 6565 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6566 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6567 dsl_pool_upgrade_clones(dp, tx); 6568 } 6569 6570 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6571 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6572 dsl_pool_upgrade_dir_clones(dp, tx); 6573 6574 /* Keeping the freedir open increases spa_minref */ 6575 spa->spa_minref += 3; 6576 } 6577 6578 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6579 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6580 spa_feature_create_zap_objects(spa, tx); 6581 } 6582 6583 /* 6584 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6585 * when possibility to use lz4 compression for metadata was added 6586 * Old pools that have this feature enabled must be upgraded to have 6587 * this feature active 6588 */ 6589 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6590 boolean_t lz4_en = spa_feature_is_enabled(spa, 6591 SPA_FEATURE_LZ4_COMPRESS); 6592 boolean_t lz4_ac = spa_feature_is_active(spa, 6593 SPA_FEATURE_LZ4_COMPRESS); 6594 6595 if (lz4_en && !lz4_ac) 6596 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6597 } 6598 6599 /* 6600 * If we haven't written the salt, do so now. Note that the 6601 * feature may not be activated yet, but that's fine since 6602 * the presence of this ZAP entry is backwards compatible. 6603 */ 6604 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 6605 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 6606 VERIFY0(zap_add(spa->spa_meta_objset, 6607 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 6608 sizeof (spa->spa_cksum_salt.zcs_bytes), 6609 spa->spa_cksum_salt.zcs_bytes, tx)); 6610 } 6611 6612 rrw_exit(&dp->dp_config_rwlock, FTAG); 6613} 6614 6615/* 6616 * Sync the specified transaction group. New blocks may be dirtied as 6617 * part of the process, so we iterate until it converges. 6618 */ 6619void 6620spa_sync(spa_t *spa, uint64_t txg) 6621{ 6622 dsl_pool_t *dp = spa->spa_dsl_pool; 6623 objset_t *mos = spa->spa_meta_objset; 6624 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6625 vdev_t *rvd = spa->spa_root_vdev; 6626 vdev_t *vd; 6627 dmu_tx_t *tx; 6628 int error; 6629 6630 VERIFY(spa_writeable(spa)); 6631 6632 /* 6633 * Lock out configuration changes. 6634 */ 6635 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6636 6637 spa->spa_syncing_txg = txg; 6638 spa->spa_sync_pass = 0; 6639 6640 /* 6641 * If there are any pending vdev state changes, convert them 6642 * into config changes that go out with this transaction group. 6643 */ 6644 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6645 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6646 /* 6647 * We need the write lock here because, for aux vdevs, 6648 * calling vdev_config_dirty() modifies sav_config. 6649 * This is ugly and will become unnecessary when we 6650 * eliminate the aux vdev wart by integrating all vdevs 6651 * into the root vdev tree. 6652 */ 6653 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6654 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6655 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6656 vdev_state_clean(vd); 6657 vdev_config_dirty(vd); 6658 } 6659 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6660 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6661 } 6662 spa_config_exit(spa, SCL_STATE, FTAG); 6663 6664 tx = dmu_tx_create_assigned(dp, txg); 6665 6666 spa->spa_sync_starttime = gethrtime(); 6667#ifdef illumos 6668 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6669 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6670#else /* !illumos */ 6671#ifdef _KERNEL 6672 callout_reset(&spa->spa_deadman_cycid, 6673 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6674#endif 6675#endif /* illumos */ 6676 6677 /* 6678 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6679 * set spa_deflate if we have no raid-z vdevs. 6680 */ 6681 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6682 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6683 int i; 6684 6685 for (i = 0; i < rvd->vdev_children; i++) { 6686 vd = rvd->vdev_child[i]; 6687 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6688 break; 6689 } 6690 if (i == rvd->vdev_children) { 6691 spa->spa_deflate = TRUE; 6692 VERIFY(0 == zap_add(spa->spa_meta_objset, 6693 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6694 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6695 } 6696 } 6697 6698 /* 6699 * Iterate to convergence. 6700 */ 6701 do { 6702 int pass = ++spa->spa_sync_pass; 6703 6704 spa_sync_config_object(spa, tx); 6705 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6706 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6707 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6708 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6709 spa_errlog_sync(spa, txg); 6710 dsl_pool_sync(dp, txg); 6711 6712 if (pass < zfs_sync_pass_deferred_free) { 6713 spa_sync_frees(spa, free_bpl, tx); 6714 } else { 6715 /* 6716 * We can not defer frees in pass 1, because 6717 * we sync the deferred frees later in pass 1. 6718 */ 6719 ASSERT3U(pass, >, 1); 6720 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6721 &spa->spa_deferred_bpobj, tx); 6722 } 6723 6724 ddt_sync(spa, txg); 6725 dsl_scan_sync(dp, tx); 6726 6727 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6728 vdev_sync(vd, txg); 6729 6730 if (pass == 1) { 6731 spa_sync_upgrades(spa, tx); 6732 ASSERT3U(txg, >=, 6733 spa->spa_uberblock.ub_rootbp.blk_birth); 6734 /* 6735 * Note: We need to check if the MOS is dirty 6736 * because we could have marked the MOS dirty 6737 * without updating the uberblock (e.g. if we 6738 * have sync tasks but no dirty user data). We 6739 * need to check the uberblock's rootbp because 6740 * it is updated if we have synced out dirty 6741 * data (though in this case the MOS will most 6742 * likely also be dirty due to second order 6743 * effects, we don't want to rely on that here). 6744 */ 6745 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6746 !dmu_objset_is_dirty(mos, txg)) { 6747 /* 6748 * Nothing changed on the first pass, 6749 * therefore this TXG is a no-op. Avoid 6750 * syncing deferred frees, so that we 6751 * can keep this TXG as a no-op. 6752 */ 6753 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6754 txg)); 6755 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6756 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6757 break; 6758 } 6759 spa_sync_deferred_frees(spa, tx); 6760 } 6761 6762 } while (dmu_objset_is_dirty(mos, txg)); 6763 6764 /* 6765 * Rewrite the vdev configuration (which includes the uberblock) 6766 * to commit the transaction group. 6767 * 6768 * If there are no dirty vdevs, we sync the uberblock to a few 6769 * random top-level vdevs that are known to be visible in the 6770 * config cache (see spa_vdev_add() for a complete description). 6771 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6772 */ 6773 for (;;) { 6774 /* 6775 * We hold SCL_STATE to prevent vdev open/close/etc. 6776 * while we're attempting to write the vdev labels. 6777 */ 6778 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6779 6780 if (list_is_empty(&spa->spa_config_dirty_list)) { 6781 vdev_t *svd[SPA_DVAS_PER_BP]; 6782 int svdcount = 0; 6783 int children = rvd->vdev_children; 6784 int c0 = spa_get_random(children); 6785 6786 for (int c = 0; c < children; c++) { 6787 vd = rvd->vdev_child[(c0 + c) % children]; 6788 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6789 continue; 6790 svd[svdcount++] = vd; 6791 if (svdcount == SPA_DVAS_PER_BP) 6792 break; 6793 } 6794 error = vdev_config_sync(svd, svdcount, txg); 6795 } else { 6796 error = vdev_config_sync(rvd->vdev_child, 6797 rvd->vdev_children, txg); 6798 } 6799 6800 if (error == 0) 6801 spa->spa_last_synced_guid = rvd->vdev_guid; 6802 6803 spa_config_exit(spa, SCL_STATE, FTAG); 6804 6805 if (error == 0) 6806 break; 6807 zio_suspend(spa, NULL); 6808 zio_resume_wait(spa); 6809 } 6810 dmu_tx_commit(tx); 6811 6812#ifdef illumos 6813 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6814#else /* !illumos */ 6815#ifdef _KERNEL 6816 callout_drain(&spa->spa_deadman_cycid); 6817#endif 6818#endif /* illumos */ 6819 6820 /* 6821 * Clear the dirty config list. 6822 */ 6823 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6824 vdev_config_clean(vd); 6825 6826 /* 6827 * Now that the new config has synced transactionally, 6828 * let it become visible to the config cache. 6829 */ 6830 if (spa->spa_config_syncing != NULL) { 6831 spa_config_set(spa, spa->spa_config_syncing); 6832 spa->spa_config_txg = txg; 6833 spa->spa_config_syncing = NULL; 6834 } 6835 6836 spa->spa_ubsync = spa->spa_uberblock; 6837 6838 dsl_pool_sync_done(dp, txg); 6839 6840 /* 6841 * Update usable space statistics. 6842 */ 6843 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6844 vdev_sync_done(vd, txg); 6845 6846 spa_update_dspace(spa); 6847 6848 /* 6849 * It had better be the case that we didn't dirty anything 6850 * since vdev_config_sync(). 6851 */ 6852 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6853 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6854 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6855 6856 spa->spa_sync_pass = 0; 6857 6858 spa_config_exit(spa, SCL_CONFIG, FTAG); 6859 6860 spa_handle_ignored_writes(spa); 6861 6862 /* 6863 * If any async tasks have been requested, kick them off. 6864 */ 6865 spa_async_dispatch(spa); 6866 spa_async_dispatch_vd(spa); 6867} 6868 6869/* 6870 * Sync all pools. We don't want to hold the namespace lock across these 6871 * operations, so we take a reference on the spa_t and drop the lock during the 6872 * sync. 6873 */ 6874void 6875spa_sync_allpools(void) 6876{ 6877 spa_t *spa = NULL; 6878 mutex_enter(&spa_namespace_lock); 6879 while ((spa = spa_next(spa)) != NULL) { 6880 if (spa_state(spa) != POOL_STATE_ACTIVE || 6881 !spa_writeable(spa) || spa_suspended(spa)) 6882 continue; 6883 spa_open_ref(spa, FTAG); 6884 mutex_exit(&spa_namespace_lock); 6885 txg_wait_synced(spa_get_dsl(spa), 0); 6886 mutex_enter(&spa_namespace_lock); 6887 spa_close(spa, FTAG); 6888 } 6889 mutex_exit(&spa_namespace_lock); 6890} 6891 6892/* 6893 * ========================================================================== 6894 * Miscellaneous routines 6895 * ========================================================================== 6896 */ 6897 6898/* 6899 * Remove all pools in the system. 6900 */ 6901void 6902spa_evict_all(void) 6903{ 6904 spa_t *spa; 6905 6906 /* 6907 * Remove all cached state. All pools should be closed now, 6908 * so every spa in the AVL tree should be unreferenced. 6909 */ 6910 mutex_enter(&spa_namespace_lock); 6911 while ((spa = spa_next(NULL)) != NULL) { 6912 /* 6913 * Stop async tasks. The async thread may need to detach 6914 * a device that's been replaced, which requires grabbing 6915 * spa_namespace_lock, so we must drop it here. 6916 */ 6917 spa_open_ref(spa, FTAG); 6918 mutex_exit(&spa_namespace_lock); 6919 spa_async_suspend(spa); 6920 mutex_enter(&spa_namespace_lock); 6921 spa_close(spa, FTAG); 6922 6923 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6924 spa_unload(spa); 6925 spa_deactivate(spa); 6926 } 6927 spa_remove(spa); 6928 } 6929 mutex_exit(&spa_namespace_lock); 6930} 6931 6932vdev_t * 6933spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6934{ 6935 vdev_t *vd; 6936 int i; 6937 6938 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6939 return (vd); 6940 6941 if (aux) { 6942 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6943 vd = spa->spa_l2cache.sav_vdevs[i]; 6944 if (vd->vdev_guid == guid) 6945 return (vd); 6946 } 6947 6948 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6949 vd = spa->spa_spares.sav_vdevs[i]; 6950 if (vd->vdev_guid == guid) 6951 return (vd); 6952 } 6953 } 6954 6955 return (NULL); 6956} 6957 6958void 6959spa_upgrade(spa_t *spa, uint64_t version) 6960{ 6961 ASSERT(spa_writeable(spa)); 6962 6963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6964 6965 /* 6966 * This should only be called for a non-faulted pool, and since a 6967 * future version would result in an unopenable pool, this shouldn't be 6968 * possible. 6969 */ 6970 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6971 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6972 6973 spa->spa_uberblock.ub_version = version; 6974 vdev_config_dirty(spa->spa_root_vdev); 6975 6976 spa_config_exit(spa, SCL_ALL, FTAG); 6977 6978 txg_wait_synced(spa_get_dsl(spa), 0); 6979} 6980 6981boolean_t 6982spa_has_spare(spa_t *spa, uint64_t guid) 6983{ 6984 int i; 6985 uint64_t spareguid; 6986 spa_aux_vdev_t *sav = &spa->spa_spares; 6987 6988 for (i = 0; i < sav->sav_count; i++) 6989 if (sav->sav_vdevs[i]->vdev_guid == guid) 6990 return (B_TRUE); 6991 6992 for (i = 0; i < sav->sav_npending; i++) { 6993 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6994 &spareguid) == 0 && spareguid == guid) 6995 return (B_TRUE); 6996 } 6997 6998 return (B_FALSE); 6999} 7000 7001/* 7002 * Check if a pool has an active shared spare device. 7003 * Note: reference count of an active spare is 2, as a spare and as a replace 7004 */ 7005static boolean_t 7006spa_has_active_shared_spare(spa_t *spa) 7007{ 7008 int i, refcnt; 7009 uint64_t pool; 7010 spa_aux_vdev_t *sav = &spa->spa_spares; 7011 7012 for (i = 0; i < sav->sav_count; i++) { 7013 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 7014 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 7015 refcnt > 2) 7016 return (B_TRUE); 7017 } 7018 7019 return (B_FALSE); 7020} 7021 7022/* 7023 * Post a sysevent corresponding to the given event. The 'name' must be one of 7024 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7025 * filled in from the spa and (optionally) the vdev. This doesn't do anything 7026 * in the userland libzpool, as we don't want consumers to misinterpret ztest 7027 * or zdb as real changes. 7028 */ 7029void 7030spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 7031{ 7032#ifdef _KERNEL 7033 sysevent_t *ev; 7034 sysevent_attr_list_t *attr = NULL; 7035 sysevent_value_t value; 7036 sysevent_id_t eid; 7037 7038 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7039 SE_SLEEP); 7040 7041 value.value_type = SE_DATA_TYPE_STRING; 7042 value.value.sv_string = spa_name(spa); 7043 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7044 goto done; 7045 7046 value.value_type = SE_DATA_TYPE_UINT64; 7047 value.value.sv_uint64 = spa_guid(spa); 7048 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7049 goto done; 7050 7051 if (vd) { 7052 value.value_type = SE_DATA_TYPE_UINT64; 7053 value.value.sv_uint64 = vd->vdev_guid; 7054 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7055 SE_SLEEP) != 0) 7056 goto done; 7057 7058 if (vd->vdev_path) { 7059 value.value_type = SE_DATA_TYPE_STRING; 7060 value.value.sv_string = vd->vdev_path; 7061 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7062 &value, SE_SLEEP) != 0) 7063 goto done; 7064 } 7065 } 7066 7067 if (sysevent_attach_attributes(ev, attr) != 0) 7068 goto done; 7069 attr = NULL; 7070 7071 (void) log_sysevent(ev, SE_SLEEP, &eid); 7072 7073done: 7074 if (attr) 7075 sysevent_free_attr(attr); 7076 sysevent_free(ev); 7077#endif 7078} 7079