spa.c revision 324010
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, 2017 by Delphix. All rights reserved. 25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2013 Saso Kiselkov. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright (c) 2017 Datto Inc. 31 */ 32 33/* 34 * SPA: Storage Pool Allocator 35 * 36 * This file contains all the routines used when modifying on-disk SPA state. 37 * This includes opening, importing, destroying, exporting a pool, and syncing a 38 * pool. 39 */ 40 41#include <sys/zfs_context.h> 42#include <sys/fm/fs/zfs.h> 43#include <sys/spa_impl.h> 44#include <sys/zio.h> 45#include <sys/zio_checksum.h> 46#include <sys/dmu.h> 47#include <sys/dmu_tx.h> 48#include <sys/zap.h> 49#include <sys/zil.h> 50#include <sys/ddt.h> 51#include <sys/vdev_impl.h> 52#include <sys/metaslab.h> 53#include <sys/metaslab_impl.h> 54#include <sys/uberblock_impl.h> 55#include <sys/txg.h> 56#include <sys/avl.h> 57#include <sys/dmu_traverse.h> 58#include <sys/dmu_objset.h> 59#include <sys/unique.h> 60#include <sys/dsl_pool.h> 61#include <sys/dsl_dataset.h> 62#include <sys/dsl_dir.h> 63#include <sys/dsl_prop.h> 64#include <sys/dsl_synctask.h> 65#include <sys/fs/zfs.h> 66#include <sys/arc.h> 67#include <sys/callb.h> 68#include <sys/spa_boot.h> 69#include <sys/zfs_ioctl.h> 70#include <sys/dsl_scan.h> 71#include <sys/dmu_send.h> 72#include <sys/dsl_destroy.h> 73#include <sys/dsl_userhold.h> 74#include <sys/zfeature.h> 75#include <sys/zvol.h> 76#include <sys/trim_map.h> 77#include <sys/abd.h> 78 79#ifdef _KERNEL 80#include <sys/callb.h> 81#include <sys/cpupart.h> 82#include <sys/zone.h> 83#endif /* _KERNEL */ 84 85#include "zfs_prop.h" 86#include "zfs_comutil.h" 87 88/* Check hostid on import? */ 89static int check_hostid = 1; 90 91/* 92 * The interval, in seconds, at which failed configuration cache file writes 93 * should be retried. 94 */ 95static int zfs_ccw_retry_interval = 300; 96 97SYSCTL_DECL(_vfs_zfs); 98SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 99 "Check hostid on import?"); 100TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 101SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 102 &zfs_ccw_retry_interval, 0, 103 "Configuration cache file write, retry after failure, interval (seconds)"); 104 105typedef enum zti_modes { 106 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 107 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 108 ZTI_MODE_NULL, /* don't create a taskq */ 109 ZTI_NMODES 110} zti_modes_t; 111 112#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 113#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 114#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 115 116#define ZTI_N(n) ZTI_P(n, 1) 117#define ZTI_ONE ZTI_N(1) 118 119typedef struct zio_taskq_info { 120 zti_modes_t zti_mode; 121 uint_t zti_value; 122 uint_t zti_count; 123} zio_taskq_info_t; 124 125static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 126 "issue", "issue_high", "intr", "intr_high" 127}; 128 129/* 130 * This table defines the taskq settings for each ZFS I/O type. When 131 * initializing a pool, we use this table to create an appropriately sized 132 * taskq. Some operations are low volume and therefore have a small, static 133 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 134 * macros. Other operations process a large amount of data; the ZTI_BATCH 135 * macro causes us to create a taskq oriented for throughput. Some operations 136 * are so high frequency and short-lived that the taskq itself can become a a 137 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 138 * additional degree of parallelism specified by the number of threads per- 139 * taskq and the number of taskqs; when dispatching an event in this case, the 140 * particular taskq is chosen at random. 141 * 142 * The different taskq priorities are to handle the different contexts (issue 143 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 144 * need to be handled with minimum delay. 145 */ 146const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 147 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 149 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 150 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 151 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 153 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 154}; 155 156static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name); 157static void spa_event_post(sysevent_t *ev); 158static void spa_sync_version(void *arg, dmu_tx_t *tx); 159static void spa_sync_props(void *arg, dmu_tx_t *tx); 160static boolean_t spa_has_active_shared_spare(spa_t *spa); 161static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 162 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 163 char **ereport); 164static void spa_vdev_resilver_done(spa_t *spa); 165 166uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 167#ifdef PSRSET_BIND 168id_t zio_taskq_psrset_bind = PS_NONE; 169#endif 170#ifdef SYSDC 171boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 172uint_t zio_taskq_basedc = 80; /* base duty cycle */ 173#endif 174 175boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 176extern int zfs_sync_pass_deferred_free; 177 178/* 179 * This (illegal) pool name is used when temporarily importing a spa_t in order 180 * to get the vdev stats associated with the imported devices. 181 */ 182#define TRYIMPORT_NAME "$import" 183 184/* 185 * ========================================================================== 186 * SPA properties routines 187 * ========================================================================== 188 */ 189 190/* 191 * Add a (source=src, propname=propval) list to an nvlist. 192 */ 193static void 194spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 195 uint64_t intval, zprop_source_t src) 196{ 197 const char *propname = zpool_prop_to_name(prop); 198 nvlist_t *propval; 199 200 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 201 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 202 203 if (strval != NULL) 204 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 205 else 206 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 207 208 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 209 nvlist_free(propval); 210} 211 212/* 213 * Get property values from the spa configuration. 214 */ 215static void 216spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 217{ 218 vdev_t *rvd = spa->spa_root_vdev; 219 dsl_pool_t *pool = spa->spa_dsl_pool; 220 uint64_t size, alloc, cap, version; 221 zprop_source_t src = ZPROP_SRC_NONE; 222 spa_config_dirent_t *dp; 223 metaslab_class_t *mc = spa_normal_class(spa); 224 225 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 226 227 if (rvd != NULL) { 228 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 229 size = metaslab_class_get_space(spa_normal_class(spa)); 230 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 231 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 232 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 233 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 234 size - alloc, src); 235 236 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 237 metaslab_class_fragmentation(mc), src); 238 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 239 metaslab_class_expandable_space(mc), src); 240 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 241 (spa_mode(spa) == FREAD), src); 242 243 cap = (size == 0) ? 0 : (alloc * 100 / size); 244 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 245 246 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 247 ddt_get_pool_dedup_ratio(spa), src); 248 249 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 250 rvd->vdev_state, src); 251 252 version = spa_version(spa); 253 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 254 src = ZPROP_SRC_DEFAULT; 255 else 256 src = ZPROP_SRC_LOCAL; 257 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 258 } 259 260 if (pool != NULL) { 261 /* 262 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 263 * when opening pools before this version freedir will be NULL. 264 */ 265 if (pool->dp_free_dir != NULL) { 266 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 267 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 268 src); 269 } else { 270 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 271 NULL, 0, src); 272 } 273 274 if (pool->dp_leak_dir != NULL) { 275 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 276 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 277 src); 278 } else { 279 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 280 NULL, 0, src); 281 } 282 } 283 284 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 285 286 if (spa->spa_comment != NULL) { 287 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 288 0, ZPROP_SRC_LOCAL); 289 } 290 291 if (spa->spa_root != NULL) 292 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 293 0, ZPROP_SRC_LOCAL); 294 295 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 296 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 297 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 298 } else { 299 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 300 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 301 } 302 303 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 304 if (dp->scd_path == NULL) { 305 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 306 "none", 0, ZPROP_SRC_LOCAL); 307 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 308 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 309 dp->scd_path, 0, ZPROP_SRC_LOCAL); 310 } 311 } 312} 313 314/* 315 * Get zpool property values. 316 */ 317int 318spa_prop_get(spa_t *spa, nvlist_t **nvp) 319{ 320 objset_t *mos = spa->spa_meta_objset; 321 zap_cursor_t zc; 322 zap_attribute_t za; 323 int err; 324 325 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 326 327 mutex_enter(&spa->spa_props_lock); 328 329 /* 330 * Get properties from the spa config. 331 */ 332 spa_prop_get_config(spa, nvp); 333 334 /* If no pool property object, no more prop to get. */ 335 if (mos == NULL || spa->spa_pool_props_object == 0) { 336 mutex_exit(&spa->spa_props_lock); 337 return (0); 338 } 339 340 /* 341 * Get properties from the MOS pool property object. 342 */ 343 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 344 (err = zap_cursor_retrieve(&zc, &za)) == 0; 345 zap_cursor_advance(&zc)) { 346 uint64_t intval = 0; 347 char *strval = NULL; 348 zprop_source_t src = ZPROP_SRC_DEFAULT; 349 zpool_prop_t prop; 350 351 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 352 continue; 353 354 switch (za.za_integer_length) { 355 case 8: 356 /* integer property */ 357 if (za.za_first_integer != 358 zpool_prop_default_numeric(prop)) 359 src = ZPROP_SRC_LOCAL; 360 361 if (prop == ZPOOL_PROP_BOOTFS) { 362 dsl_pool_t *dp; 363 dsl_dataset_t *ds = NULL; 364 365 dp = spa_get_dsl(spa); 366 dsl_pool_config_enter(dp, FTAG); 367 if (err = dsl_dataset_hold_obj(dp, 368 za.za_first_integer, FTAG, &ds)) { 369 dsl_pool_config_exit(dp, FTAG); 370 break; 371 } 372 373 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 374 KM_SLEEP); 375 dsl_dataset_name(ds, strval); 376 dsl_dataset_rele(ds, FTAG); 377 dsl_pool_config_exit(dp, FTAG); 378 } else { 379 strval = NULL; 380 intval = za.za_first_integer; 381 } 382 383 spa_prop_add_list(*nvp, prop, strval, intval, src); 384 385 if (strval != NULL) 386 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 387 388 break; 389 390 case 1: 391 /* string property */ 392 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 393 err = zap_lookup(mos, spa->spa_pool_props_object, 394 za.za_name, 1, za.za_num_integers, strval); 395 if (err) { 396 kmem_free(strval, za.za_num_integers); 397 break; 398 } 399 spa_prop_add_list(*nvp, prop, strval, 0, src); 400 kmem_free(strval, za.za_num_integers); 401 break; 402 403 default: 404 break; 405 } 406 } 407 zap_cursor_fini(&zc); 408 mutex_exit(&spa->spa_props_lock); 409out: 410 if (err && err != ENOENT) { 411 nvlist_free(*nvp); 412 *nvp = NULL; 413 return (err); 414 } 415 416 return (0); 417} 418 419/* 420 * Validate the given pool properties nvlist and modify the list 421 * for the property values to be set. 422 */ 423static int 424spa_prop_validate(spa_t *spa, nvlist_t *props) 425{ 426 nvpair_t *elem; 427 int error = 0, reset_bootfs = 0; 428 uint64_t objnum = 0; 429 boolean_t has_feature = B_FALSE; 430 431 elem = NULL; 432 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 433 uint64_t intval; 434 char *strval, *slash, *check, *fname; 435 const char *propname = nvpair_name(elem); 436 zpool_prop_t prop = zpool_name_to_prop(propname); 437 438 switch (prop) { 439 case ZPROP_INVAL: 440 if (!zpool_prop_feature(propname)) { 441 error = SET_ERROR(EINVAL); 442 break; 443 } 444 445 /* 446 * Sanitize the input. 447 */ 448 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 449 error = SET_ERROR(EINVAL); 450 break; 451 } 452 453 if (nvpair_value_uint64(elem, &intval) != 0) { 454 error = SET_ERROR(EINVAL); 455 break; 456 } 457 458 if (intval != 0) { 459 error = SET_ERROR(EINVAL); 460 break; 461 } 462 463 fname = strchr(propname, '@') + 1; 464 if (zfeature_lookup_name(fname, NULL) != 0) { 465 error = SET_ERROR(EINVAL); 466 break; 467 } 468 469 has_feature = B_TRUE; 470 break; 471 472 case ZPOOL_PROP_VERSION: 473 error = nvpair_value_uint64(elem, &intval); 474 if (!error && 475 (intval < spa_version(spa) || 476 intval > SPA_VERSION_BEFORE_FEATURES || 477 has_feature)) 478 error = SET_ERROR(EINVAL); 479 break; 480 481 case ZPOOL_PROP_DELEGATION: 482 case ZPOOL_PROP_AUTOREPLACE: 483 case ZPOOL_PROP_LISTSNAPS: 484 case ZPOOL_PROP_AUTOEXPAND: 485 error = nvpair_value_uint64(elem, &intval); 486 if (!error && intval > 1) 487 error = SET_ERROR(EINVAL); 488 break; 489 490 case ZPOOL_PROP_BOOTFS: 491 /* 492 * If the pool version is less than SPA_VERSION_BOOTFS, 493 * or the pool is still being created (version == 0), 494 * the bootfs property cannot be set. 495 */ 496 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 497 error = SET_ERROR(ENOTSUP); 498 break; 499 } 500 501 /* 502 * Make sure the vdev config is bootable 503 */ 504 if (!vdev_is_bootable(spa->spa_root_vdev)) { 505 error = SET_ERROR(ENOTSUP); 506 break; 507 } 508 509 reset_bootfs = 1; 510 511 error = nvpair_value_string(elem, &strval); 512 513 if (!error) { 514 objset_t *os; 515 uint64_t propval; 516 517 if (strval == NULL || strval[0] == '\0') { 518 objnum = zpool_prop_default_numeric( 519 ZPOOL_PROP_BOOTFS); 520 break; 521 } 522 523 if (error = dmu_objset_hold(strval, FTAG, &os)) 524 break; 525 526 /* 527 * Must be ZPL, and its property settings 528 * must be supported by GRUB (compression 529 * is not gzip, and large blocks are not used). 530 */ 531 532 if (dmu_objset_type(os) != DMU_OST_ZFS) { 533 error = SET_ERROR(ENOTSUP); 534 } else if ((error = 535 dsl_prop_get_int_ds(dmu_objset_ds(os), 536 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 537 &propval)) == 0 && 538 !BOOTFS_COMPRESS_VALID(propval)) { 539 error = SET_ERROR(ENOTSUP); 540 } else if ((error = 541 dsl_prop_get_int_ds(dmu_objset_ds(os), 542 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 543 &propval)) == 0 && 544 propval > SPA_OLD_MAXBLOCKSIZE) { 545 error = SET_ERROR(ENOTSUP); 546 } else { 547 objnum = dmu_objset_id(os); 548 } 549 dmu_objset_rele(os, FTAG); 550 } 551 break; 552 553 case ZPOOL_PROP_FAILUREMODE: 554 error = nvpair_value_uint64(elem, &intval); 555 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 556 intval > ZIO_FAILURE_MODE_PANIC)) 557 error = SET_ERROR(EINVAL); 558 559 /* 560 * This is a special case which only occurs when 561 * the pool has completely failed. This allows 562 * the user to change the in-core failmode property 563 * without syncing it out to disk (I/Os might 564 * currently be blocked). We do this by returning 565 * EIO to the caller (spa_prop_set) to trick it 566 * into thinking we encountered a property validation 567 * error. 568 */ 569 if (!error && spa_suspended(spa)) { 570 spa->spa_failmode = intval; 571 error = SET_ERROR(EIO); 572 } 573 break; 574 575 case ZPOOL_PROP_CACHEFILE: 576 if ((error = nvpair_value_string(elem, &strval)) != 0) 577 break; 578 579 if (strval[0] == '\0') 580 break; 581 582 if (strcmp(strval, "none") == 0) 583 break; 584 585 if (strval[0] != '/') { 586 error = SET_ERROR(EINVAL); 587 break; 588 } 589 590 slash = strrchr(strval, '/'); 591 ASSERT(slash != NULL); 592 593 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 594 strcmp(slash, "/..") == 0) 595 error = SET_ERROR(EINVAL); 596 break; 597 598 case ZPOOL_PROP_COMMENT: 599 if ((error = nvpair_value_string(elem, &strval)) != 0) 600 break; 601 for (check = strval; *check != '\0'; check++) { 602 /* 603 * The kernel doesn't have an easy isprint() 604 * check. For this kernel check, we merely 605 * check ASCII apart from DEL. Fix this if 606 * there is an easy-to-use kernel isprint(). 607 */ 608 if (*check >= 0x7f) { 609 error = SET_ERROR(EINVAL); 610 break; 611 } 612 } 613 if (strlen(strval) > ZPROP_MAX_COMMENT) 614 error = E2BIG; 615 break; 616 617 case ZPOOL_PROP_DEDUPDITTO: 618 if (spa_version(spa) < SPA_VERSION_DEDUP) 619 error = SET_ERROR(ENOTSUP); 620 else 621 error = nvpair_value_uint64(elem, &intval); 622 if (error == 0 && 623 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 624 error = SET_ERROR(EINVAL); 625 break; 626 } 627 628 if (error) 629 break; 630 } 631 632 if (!error && reset_bootfs) { 633 error = nvlist_remove(props, 634 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 635 636 if (!error) { 637 error = nvlist_add_uint64(props, 638 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 639 } 640 } 641 642 return (error); 643} 644 645void 646spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 647{ 648 char *cachefile; 649 spa_config_dirent_t *dp; 650 651 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 652 &cachefile) != 0) 653 return; 654 655 dp = kmem_alloc(sizeof (spa_config_dirent_t), 656 KM_SLEEP); 657 658 if (cachefile[0] == '\0') 659 dp->scd_path = spa_strdup(spa_config_path); 660 else if (strcmp(cachefile, "none") == 0) 661 dp->scd_path = NULL; 662 else 663 dp->scd_path = spa_strdup(cachefile); 664 665 list_insert_head(&spa->spa_config_list, dp); 666 if (need_sync) 667 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 668} 669 670int 671spa_prop_set(spa_t *spa, nvlist_t *nvp) 672{ 673 int error; 674 nvpair_t *elem = NULL; 675 boolean_t need_sync = B_FALSE; 676 677 if ((error = spa_prop_validate(spa, nvp)) != 0) 678 return (error); 679 680 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 681 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 682 683 if (prop == ZPOOL_PROP_CACHEFILE || 684 prop == ZPOOL_PROP_ALTROOT || 685 prop == ZPOOL_PROP_READONLY) 686 continue; 687 688 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 689 uint64_t ver; 690 691 if (prop == ZPOOL_PROP_VERSION) { 692 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 693 } else { 694 ASSERT(zpool_prop_feature(nvpair_name(elem))); 695 ver = SPA_VERSION_FEATURES; 696 need_sync = B_TRUE; 697 } 698 699 /* Save time if the version is already set. */ 700 if (ver == spa_version(spa)) 701 continue; 702 703 /* 704 * In addition to the pool directory object, we might 705 * create the pool properties object, the features for 706 * read object, the features for write object, or the 707 * feature descriptions object. 708 */ 709 error = dsl_sync_task(spa->spa_name, NULL, 710 spa_sync_version, &ver, 711 6, ZFS_SPACE_CHECK_RESERVED); 712 if (error) 713 return (error); 714 continue; 715 } 716 717 need_sync = B_TRUE; 718 break; 719 } 720 721 if (need_sync) { 722 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 723 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 724 } 725 726 return (0); 727} 728 729/* 730 * If the bootfs property value is dsobj, clear it. 731 */ 732void 733spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 734{ 735 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 736 VERIFY(zap_remove(spa->spa_meta_objset, 737 spa->spa_pool_props_object, 738 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 739 spa->spa_bootfs = 0; 740 } 741} 742 743/*ARGSUSED*/ 744static int 745spa_change_guid_check(void *arg, dmu_tx_t *tx) 746{ 747 uint64_t *newguid = arg; 748 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 749 vdev_t *rvd = spa->spa_root_vdev; 750 uint64_t vdev_state; 751 752 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 753 vdev_state = rvd->vdev_state; 754 spa_config_exit(spa, SCL_STATE, FTAG); 755 756 if (vdev_state != VDEV_STATE_HEALTHY) 757 return (SET_ERROR(ENXIO)); 758 759 ASSERT3U(spa_guid(spa), !=, *newguid); 760 761 return (0); 762} 763 764static void 765spa_change_guid_sync(void *arg, dmu_tx_t *tx) 766{ 767 uint64_t *newguid = arg; 768 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 769 uint64_t oldguid; 770 vdev_t *rvd = spa->spa_root_vdev; 771 772 oldguid = spa_guid(spa); 773 774 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 775 rvd->vdev_guid = *newguid; 776 rvd->vdev_guid_sum += (*newguid - oldguid); 777 vdev_config_dirty(rvd); 778 spa_config_exit(spa, SCL_STATE, FTAG); 779 780 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 781 oldguid, *newguid); 782} 783 784/* 785 * Change the GUID for the pool. This is done so that we can later 786 * re-import a pool built from a clone of our own vdevs. We will modify 787 * the root vdev's guid, our own pool guid, and then mark all of our 788 * vdevs dirty. Note that we must make sure that all our vdevs are 789 * online when we do this, or else any vdevs that weren't present 790 * would be orphaned from our pool. We are also going to issue a 791 * sysevent to update any watchers. 792 */ 793int 794spa_change_guid(spa_t *spa) 795{ 796 int error; 797 uint64_t guid; 798 799 mutex_enter(&spa->spa_vdev_top_lock); 800 mutex_enter(&spa_namespace_lock); 801 guid = spa_generate_guid(NULL); 802 803 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 804 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 805 806 if (error == 0) { 807 spa_config_sync(spa, B_FALSE, B_TRUE); 808 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 809 } 810 811 mutex_exit(&spa_namespace_lock); 812 mutex_exit(&spa->spa_vdev_top_lock); 813 814 return (error); 815} 816 817/* 818 * ========================================================================== 819 * SPA state manipulation (open/create/destroy/import/export) 820 * ========================================================================== 821 */ 822 823static int 824spa_error_entry_compare(const void *a, const void *b) 825{ 826 spa_error_entry_t *sa = (spa_error_entry_t *)a; 827 spa_error_entry_t *sb = (spa_error_entry_t *)b; 828 int ret; 829 830 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 831 sizeof (zbookmark_phys_t)); 832 833 if (ret < 0) 834 return (-1); 835 else if (ret > 0) 836 return (1); 837 else 838 return (0); 839} 840 841/* 842 * Utility function which retrieves copies of the current logs and 843 * re-initializes them in the process. 844 */ 845void 846spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 847{ 848 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 849 850 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 851 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 852 853 avl_create(&spa->spa_errlist_scrub, 854 spa_error_entry_compare, sizeof (spa_error_entry_t), 855 offsetof(spa_error_entry_t, se_avl)); 856 avl_create(&spa->spa_errlist_last, 857 spa_error_entry_compare, sizeof (spa_error_entry_t), 858 offsetof(spa_error_entry_t, se_avl)); 859} 860 861static void 862spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 863{ 864 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 865 enum zti_modes mode = ztip->zti_mode; 866 uint_t value = ztip->zti_value; 867 uint_t count = ztip->zti_count; 868 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 869 char name[32]; 870 uint_t flags = 0; 871 boolean_t batch = B_FALSE; 872 873 if (mode == ZTI_MODE_NULL) { 874 tqs->stqs_count = 0; 875 tqs->stqs_taskq = NULL; 876 return; 877 } 878 879 ASSERT3U(count, >, 0); 880 881 tqs->stqs_count = count; 882 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 883 884 switch (mode) { 885 case ZTI_MODE_FIXED: 886 ASSERT3U(value, >=, 1); 887 value = MAX(value, 1); 888 break; 889 890 case ZTI_MODE_BATCH: 891 batch = B_TRUE; 892 flags |= TASKQ_THREADS_CPU_PCT; 893 value = zio_taskq_batch_pct; 894 break; 895 896 default: 897 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 898 "spa_activate()", 899 zio_type_name[t], zio_taskq_types[q], mode, value); 900 break; 901 } 902 903 for (uint_t i = 0; i < count; i++) { 904 taskq_t *tq; 905 906 if (count > 1) { 907 (void) snprintf(name, sizeof (name), "%s_%s_%u", 908 zio_type_name[t], zio_taskq_types[q], i); 909 } else { 910 (void) snprintf(name, sizeof (name), "%s_%s", 911 zio_type_name[t], zio_taskq_types[q]); 912 } 913 914#ifdef SYSDC 915 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 916 if (batch) 917 flags |= TASKQ_DC_BATCH; 918 919 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 920 spa->spa_proc, zio_taskq_basedc, flags); 921 } else { 922#endif 923 pri_t pri = maxclsyspri; 924 /* 925 * The write issue taskq can be extremely CPU 926 * intensive. Run it at slightly lower priority 927 * than the other taskqs. 928 * FreeBSD notes: 929 * - numerically higher priorities are lower priorities; 930 * - if priorities divided by four (RQ_PPQ) are equal 931 * then a difference between them is insignificant. 932 */ 933 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 934#ifdef illumos 935 pri--; 936#else 937 pri += 4; 938#endif 939 940 tq = taskq_create_proc(name, value, pri, 50, 941 INT_MAX, spa->spa_proc, flags); 942#ifdef SYSDC 943 } 944#endif 945 946 tqs->stqs_taskq[i] = tq; 947 } 948} 949 950static void 951spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 952{ 953 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 954 955 if (tqs->stqs_taskq == NULL) { 956 ASSERT0(tqs->stqs_count); 957 return; 958 } 959 960 for (uint_t i = 0; i < tqs->stqs_count; i++) { 961 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 962 taskq_destroy(tqs->stqs_taskq[i]); 963 } 964 965 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 966 tqs->stqs_taskq = NULL; 967} 968 969/* 970 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 971 * Note that a type may have multiple discrete taskqs to avoid lock contention 972 * on the taskq itself. In that case we choose which taskq at random by using 973 * the low bits of gethrtime(). 974 */ 975void 976spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 977 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 978{ 979 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 980 taskq_t *tq; 981 982 ASSERT3P(tqs->stqs_taskq, !=, NULL); 983 ASSERT3U(tqs->stqs_count, !=, 0); 984 985 if (tqs->stqs_count == 1) { 986 tq = tqs->stqs_taskq[0]; 987 } else { 988#ifdef _KERNEL 989 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 990#else 991 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 992#endif 993 } 994 995 taskq_dispatch_ent(tq, func, arg, flags, ent); 996} 997 998static void 999spa_create_zio_taskqs(spa_t *spa) 1000{ 1001 for (int t = 0; t < ZIO_TYPES; t++) { 1002 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1003 spa_taskqs_init(spa, t, q); 1004 } 1005 } 1006} 1007 1008#ifdef _KERNEL 1009#ifdef SPA_PROCESS 1010static void 1011spa_thread(void *arg) 1012{ 1013 callb_cpr_t cprinfo; 1014 1015 spa_t *spa = arg; 1016 user_t *pu = PTOU(curproc); 1017 1018 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1019 spa->spa_name); 1020 1021 ASSERT(curproc != &p0); 1022 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1023 "zpool-%s", spa->spa_name); 1024 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1025 1026#ifdef PSRSET_BIND 1027 /* bind this thread to the requested psrset */ 1028 if (zio_taskq_psrset_bind != PS_NONE) { 1029 pool_lock(); 1030 mutex_enter(&cpu_lock); 1031 mutex_enter(&pidlock); 1032 mutex_enter(&curproc->p_lock); 1033 1034 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1035 0, NULL, NULL) == 0) { 1036 curthread->t_bind_pset = zio_taskq_psrset_bind; 1037 } else { 1038 cmn_err(CE_WARN, 1039 "Couldn't bind process for zfs pool \"%s\" to " 1040 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1041 } 1042 1043 mutex_exit(&curproc->p_lock); 1044 mutex_exit(&pidlock); 1045 mutex_exit(&cpu_lock); 1046 pool_unlock(); 1047 } 1048#endif 1049 1050#ifdef SYSDC 1051 if (zio_taskq_sysdc) { 1052 sysdc_thread_enter(curthread, 100, 0); 1053 } 1054#endif 1055 1056 spa->spa_proc = curproc; 1057 spa->spa_did = curthread->t_did; 1058 1059 spa_create_zio_taskqs(spa); 1060 1061 mutex_enter(&spa->spa_proc_lock); 1062 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1063 1064 spa->spa_proc_state = SPA_PROC_ACTIVE; 1065 cv_broadcast(&spa->spa_proc_cv); 1066 1067 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1068 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1069 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1070 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1071 1072 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1073 spa->spa_proc_state = SPA_PROC_GONE; 1074 spa->spa_proc = &p0; 1075 cv_broadcast(&spa->spa_proc_cv); 1076 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1077 1078 mutex_enter(&curproc->p_lock); 1079 lwp_exit(); 1080} 1081#endif /* SPA_PROCESS */ 1082#endif 1083 1084/* 1085 * Activate an uninitialized pool. 1086 */ 1087static void 1088spa_activate(spa_t *spa, int mode) 1089{ 1090 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1091 1092 spa->spa_state = POOL_STATE_ACTIVE; 1093 spa->spa_mode = mode; 1094 1095 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1096 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1097 1098 /* Try to create a covering process */ 1099 mutex_enter(&spa->spa_proc_lock); 1100 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1101 ASSERT(spa->spa_proc == &p0); 1102 spa->spa_did = 0; 1103 1104#ifdef SPA_PROCESS 1105 /* Only create a process if we're going to be around a while. */ 1106 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1107 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1108 NULL, 0) == 0) { 1109 spa->spa_proc_state = SPA_PROC_CREATED; 1110 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1111 cv_wait(&spa->spa_proc_cv, 1112 &spa->spa_proc_lock); 1113 } 1114 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1115 ASSERT(spa->spa_proc != &p0); 1116 ASSERT(spa->spa_did != 0); 1117 } else { 1118#ifdef _KERNEL 1119 cmn_err(CE_WARN, 1120 "Couldn't create process for zfs pool \"%s\"\n", 1121 spa->spa_name); 1122#endif 1123 } 1124 } 1125#endif /* SPA_PROCESS */ 1126 mutex_exit(&spa->spa_proc_lock); 1127 1128 /* If we didn't create a process, we need to create our taskqs. */ 1129 ASSERT(spa->spa_proc == &p0); 1130 if (spa->spa_proc == &p0) { 1131 spa_create_zio_taskqs(spa); 1132 } 1133 1134 /* 1135 * Start TRIM thread. 1136 */ 1137 trim_thread_create(spa); 1138 1139 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1140 offsetof(vdev_t, vdev_config_dirty_node)); 1141 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1142 offsetof(objset_t, os_evicting_node)); 1143 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1144 offsetof(vdev_t, vdev_state_dirty_node)); 1145 1146 txg_list_create(&spa->spa_vdev_txg_list, spa, 1147 offsetof(struct vdev, vdev_txg_node)); 1148 1149 avl_create(&spa->spa_errlist_scrub, 1150 spa_error_entry_compare, sizeof (spa_error_entry_t), 1151 offsetof(spa_error_entry_t, se_avl)); 1152 avl_create(&spa->spa_errlist_last, 1153 spa_error_entry_compare, sizeof (spa_error_entry_t), 1154 offsetof(spa_error_entry_t, se_avl)); 1155} 1156 1157/* 1158 * Opposite of spa_activate(). 1159 */ 1160static void 1161spa_deactivate(spa_t *spa) 1162{ 1163 ASSERT(spa->spa_sync_on == B_FALSE); 1164 ASSERT(spa->spa_dsl_pool == NULL); 1165 ASSERT(spa->spa_root_vdev == NULL); 1166 ASSERT(spa->spa_async_zio_root == NULL); 1167 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1168 1169 /* 1170 * Stop TRIM thread in case spa_unload() wasn't called directly 1171 * before spa_deactivate(). 1172 */ 1173 trim_thread_destroy(spa); 1174 1175 spa_evicting_os_wait(spa); 1176 1177 txg_list_destroy(&spa->spa_vdev_txg_list); 1178 1179 list_destroy(&spa->spa_config_dirty_list); 1180 list_destroy(&spa->spa_evicting_os_list); 1181 list_destroy(&spa->spa_state_dirty_list); 1182 1183 for (int t = 0; t < ZIO_TYPES; t++) { 1184 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1185 spa_taskqs_fini(spa, t, q); 1186 } 1187 } 1188 1189 metaslab_class_destroy(spa->spa_normal_class); 1190 spa->spa_normal_class = NULL; 1191 1192 metaslab_class_destroy(spa->spa_log_class); 1193 spa->spa_log_class = NULL; 1194 1195 /* 1196 * If this was part of an import or the open otherwise failed, we may 1197 * still have errors left in the queues. Empty them just in case. 1198 */ 1199 spa_errlog_drain(spa); 1200 1201 avl_destroy(&spa->spa_errlist_scrub); 1202 avl_destroy(&spa->spa_errlist_last); 1203 1204 spa->spa_state = POOL_STATE_UNINITIALIZED; 1205 1206 mutex_enter(&spa->spa_proc_lock); 1207 if (spa->spa_proc_state != SPA_PROC_NONE) { 1208 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1209 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1210 cv_broadcast(&spa->spa_proc_cv); 1211 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1212 ASSERT(spa->spa_proc != &p0); 1213 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1214 } 1215 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1216 spa->spa_proc_state = SPA_PROC_NONE; 1217 } 1218 ASSERT(spa->spa_proc == &p0); 1219 mutex_exit(&spa->spa_proc_lock); 1220 1221#ifdef SPA_PROCESS 1222 /* 1223 * We want to make sure spa_thread() has actually exited the ZFS 1224 * module, so that the module can't be unloaded out from underneath 1225 * it. 1226 */ 1227 if (spa->spa_did != 0) { 1228 thread_join(spa->spa_did); 1229 spa->spa_did = 0; 1230 } 1231#endif /* SPA_PROCESS */ 1232} 1233 1234/* 1235 * Verify a pool configuration, and construct the vdev tree appropriately. This 1236 * will create all the necessary vdevs in the appropriate layout, with each vdev 1237 * in the CLOSED state. This will prep the pool before open/creation/import. 1238 * All vdev validation is done by the vdev_alloc() routine. 1239 */ 1240static int 1241spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1242 uint_t id, int atype) 1243{ 1244 nvlist_t **child; 1245 uint_t children; 1246 int error; 1247 1248 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1249 return (error); 1250 1251 if ((*vdp)->vdev_ops->vdev_op_leaf) 1252 return (0); 1253 1254 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1255 &child, &children); 1256 1257 if (error == ENOENT) 1258 return (0); 1259 1260 if (error) { 1261 vdev_free(*vdp); 1262 *vdp = NULL; 1263 return (SET_ERROR(EINVAL)); 1264 } 1265 1266 for (int c = 0; c < children; c++) { 1267 vdev_t *vd; 1268 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1269 atype)) != 0) { 1270 vdev_free(*vdp); 1271 *vdp = NULL; 1272 return (error); 1273 } 1274 } 1275 1276 ASSERT(*vdp != NULL); 1277 1278 return (0); 1279} 1280 1281/* 1282 * Opposite of spa_load(). 1283 */ 1284static void 1285spa_unload(spa_t *spa) 1286{ 1287 int i; 1288 1289 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1290 1291 /* 1292 * Stop TRIM thread. 1293 */ 1294 trim_thread_destroy(spa); 1295 1296 /* 1297 * Stop async tasks. 1298 */ 1299 spa_async_suspend(spa); 1300 1301 /* 1302 * Stop syncing. 1303 */ 1304 if (spa->spa_sync_on) { 1305 txg_sync_stop(spa->spa_dsl_pool); 1306 spa->spa_sync_on = B_FALSE; 1307 } 1308 1309 /* 1310 * Even though vdev_free() also calls vdev_metaslab_fini, we need 1311 * to call it earlier, before we wait for async i/o to complete. 1312 * This ensures that there is no async metaslab prefetching, by 1313 * calling taskq_wait(mg_taskq). 1314 */ 1315 if (spa->spa_root_vdev != NULL) { 1316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1317 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) 1318 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); 1319 spa_config_exit(spa, SCL_ALL, FTAG); 1320 } 1321 1322 /* 1323 * Wait for any outstanding async I/O to complete. 1324 */ 1325 if (spa->spa_async_zio_root != NULL) { 1326 for (int i = 0; i < max_ncpus; i++) 1327 (void) zio_wait(spa->spa_async_zio_root[i]); 1328 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1329 spa->spa_async_zio_root = NULL; 1330 } 1331 1332 bpobj_close(&spa->spa_deferred_bpobj); 1333 1334 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1335 1336 /* 1337 * Close all vdevs. 1338 */ 1339 if (spa->spa_root_vdev) 1340 vdev_free(spa->spa_root_vdev); 1341 ASSERT(spa->spa_root_vdev == NULL); 1342 1343 /* 1344 * Close the dsl pool. 1345 */ 1346 if (spa->spa_dsl_pool) { 1347 dsl_pool_close(spa->spa_dsl_pool); 1348 spa->spa_dsl_pool = NULL; 1349 spa->spa_meta_objset = NULL; 1350 } 1351 1352 ddt_unload(spa); 1353 1354 /* 1355 * Drop and purge level 2 cache 1356 */ 1357 spa_l2cache_drop(spa); 1358 1359 for (i = 0; i < spa->spa_spares.sav_count; i++) 1360 vdev_free(spa->spa_spares.sav_vdevs[i]); 1361 if (spa->spa_spares.sav_vdevs) { 1362 kmem_free(spa->spa_spares.sav_vdevs, 1363 spa->spa_spares.sav_count * sizeof (void *)); 1364 spa->spa_spares.sav_vdevs = NULL; 1365 } 1366 if (spa->spa_spares.sav_config) { 1367 nvlist_free(spa->spa_spares.sav_config); 1368 spa->spa_spares.sav_config = NULL; 1369 } 1370 spa->spa_spares.sav_count = 0; 1371 1372 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1373 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1374 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1375 } 1376 if (spa->spa_l2cache.sav_vdevs) { 1377 kmem_free(spa->spa_l2cache.sav_vdevs, 1378 spa->spa_l2cache.sav_count * sizeof (void *)); 1379 spa->spa_l2cache.sav_vdevs = NULL; 1380 } 1381 if (spa->spa_l2cache.sav_config) { 1382 nvlist_free(spa->spa_l2cache.sav_config); 1383 spa->spa_l2cache.sav_config = NULL; 1384 } 1385 spa->spa_l2cache.sav_count = 0; 1386 1387 spa->spa_async_suspended = 0; 1388 1389 if (spa->spa_comment != NULL) { 1390 spa_strfree(spa->spa_comment); 1391 spa->spa_comment = NULL; 1392 } 1393 1394 spa_config_exit(spa, SCL_ALL, FTAG); 1395} 1396 1397/* 1398 * Load (or re-load) the current list of vdevs describing the active spares for 1399 * this pool. When this is called, we have some form of basic information in 1400 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1401 * then re-generate a more complete list including status information. 1402 */ 1403static void 1404spa_load_spares(spa_t *spa) 1405{ 1406 nvlist_t **spares; 1407 uint_t nspares; 1408 int i; 1409 vdev_t *vd, *tvd; 1410 1411 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1412 1413 /* 1414 * First, close and free any existing spare vdevs. 1415 */ 1416 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1417 vd = spa->spa_spares.sav_vdevs[i]; 1418 1419 /* Undo the call to spa_activate() below */ 1420 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1421 B_FALSE)) != NULL && tvd->vdev_isspare) 1422 spa_spare_remove(tvd); 1423 vdev_close(vd); 1424 vdev_free(vd); 1425 } 1426 1427 if (spa->spa_spares.sav_vdevs) 1428 kmem_free(spa->spa_spares.sav_vdevs, 1429 spa->spa_spares.sav_count * sizeof (void *)); 1430 1431 if (spa->spa_spares.sav_config == NULL) 1432 nspares = 0; 1433 else 1434 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1435 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1436 1437 spa->spa_spares.sav_count = (int)nspares; 1438 spa->spa_spares.sav_vdevs = NULL; 1439 1440 if (nspares == 0) 1441 return; 1442 1443 /* 1444 * Construct the array of vdevs, opening them to get status in the 1445 * process. For each spare, there is potentially two different vdev_t 1446 * structures associated with it: one in the list of spares (used only 1447 * for basic validation purposes) and one in the active vdev 1448 * configuration (if it's spared in). During this phase we open and 1449 * validate each vdev on the spare list. If the vdev also exists in the 1450 * active configuration, then we also mark this vdev as an active spare. 1451 */ 1452 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1453 KM_SLEEP); 1454 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1455 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1456 VDEV_ALLOC_SPARE) == 0); 1457 ASSERT(vd != NULL); 1458 1459 spa->spa_spares.sav_vdevs[i] = vd; 1460 1461 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1462 B_FALSE)) != NULL) { 1463 if (!tvd->vdev_isspare) 1464 spa_spare_add(tvd); 1465 1466 /* 1467 * We only mark the spare active if we were successfully 1468 * able to load the vdev. Otherwise, importing a pool 1469 * with a bad active spare would result in strange 1470 * behavior, because multiple pool would think the spare 1471 * is actively in use. 1472 * 1473 * There is a vulnerability here to an equally bizarre 1474 * circumstance, where a dead active spare is later 1475 * brought back to life (onlined or otherwise). Given 1476 * the rarity of this scenario, and the extra complexity 1477 * it adds, we ignore the possibility. 1478 */ 1479 if (!vdev_is_dead(tvd)) 1480 spa_spare_activate(tvd); 1481 } 1482 1483 vd->vdev_top = vd; 1484 vd->vdev_aux = &spa->spa_spares; 1485 1486 if (vdev_open(vd) != 0) 1487 continue; 1488 1489 if (vdev_validate_aux(vd) == 0) 1490 spa_spare_add(vd); 1491 } 1492 1493 /* 1494 * Recompute the stashed list of spares, with status information 1495 * this time. 1496 */ 1497 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1498 DATA_TYPE_NVLIST_ARRAY) == 0); 1499 1500 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1501 KM_SLEEP); 1502 for (i = 0; i < spa->spa_spares.sav_count; i++) 1503 spares[i] = vdev_config_generate(spa, 1504 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1505 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1506 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1507 for (i = 0; i < spa->spa_spares.sav_count; i++) 1508 nvlist_free(spares[i]); 1509 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1510} 1511 1512/* 1513 * Load (or re-load) the current list of vdevs describing the active l2cache for 1514 * this pool. When this is called, we have some form of basic information in 1515 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1516 * then re-generate a more complete list including status information. 1517 * Devices which are already active have their details maintained, and are 1518 * not re-opened. 1519 */ 1520static void 1521spa_load_l2cache(spa_t *spa) 1522{ 1523 nvlist_t **l2cache; 1524 uint_t nl2cache; 1525 int i, j, oldnvdevs; 1526 uint64_t guid; 1527 vdev_t *vd, **oldvdevs, **newvdevs; 1528 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1529 1530 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1531 1532 if (sav->sav_config != NULL) { 1533 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1534 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1535 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1536 } else { 1537 nl2cache = 0; 1538 newvdevs = NULL; 1539 } 1540 1541 oldvdevs = sav->sav_vdevs; 1542 oldnvdevs = sav->sav_count; 1543 sav->sav_vdevs = NULL; 1544 sav->sav_count = 0; 1545 1546 /* 1547 * Process new nvlist of vdevs. 1548 */ 1549 for (i = 0; i < nl2cache; i++) { 1550 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1551 &guid) == 0); 1552 1553 newvdevs[i] = NULL; 1554 for (j = 0; j < oldnvdevs; j++) { 1555 vd = oldvdevs[j]; 1556 if (vd != NULL && guid == vd->vdev_guid) { 1557 /* 1558 * Retain previous vdev for add/remove ops. 1559 */ 1560 newvdevs[i] = vd; 1561 oldvdevs[j] = NULL; 1562 break; 1563 } 1564 } 1565 1566 if (newvdevs[i] == NULL) { 1567 /* 1568 * Create new vdev 1569 */ 1570 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1571 VDEV_ALLOC_L2CACHE) == 0); 1572 ASSERT(vd != NULL); 1573 newvdevs[i] = vd; 1574 1575 /* 1576 * Commit this vdev as an l2cache device, 1577 * even if it fails to open. 1578 */ 1579 spa_l2cache_add(vd); 1580 1581 vd->vdev_top = vd; 1582 vd->vdev_aux = sav; 1583 1584 spa_l2cache_activate(vd); 1585 1586 if (vdev_open(vd) != 0) 1587 continue; 1588 1589 (void) vdev_validate_aux(vd); 1590 1591 if (!vdev_is_dead(vd)) 1592 l2arc_add_vdev(spa, vd); 1593 } 1594 } 1595 1596 /* 1597 * Purge vdevs that were dropped 1598 */ 1599 for (i = 0; i < oldnvdevs; i++) { 1600 uint64_t pool; 1601 1602 vd = oldvdevs[i]; 1603 if (vd != NULL) { 1604 ASSERT(vd->vdev_isl2cache); 1605 1606 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1607 pool != 0ULL && l2arc_vdev_present(vd)) 1608 l2arc_remove_vdev(vd); 1609 vdev_clear_stats(vd); 1610 vdev_free(vd); 1611 } 1612 } 1613 1614 if (oldvdevs) 1615 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1616 1617 if (sav->sav_config == NULL) 1618 goto out; 1619 1620 sav->sav_vdevs = newvdevs; 1621 sav->sav_count = (int)nl2cache; 1622 1623 /* 1624 * Recompute the stashed list of l2cache devices, with status 1625 * information this time. 1626 */ 1627 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1628 DATA_TYPE_NVLIST_ARRAY) == 0); 1629 1630 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1631 for (i = 0; i < sav->sav_count; i++) 1632 l2cache[i] = vdev_config_generate(spa, 1633 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1634 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1635 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1636out: 1637 for (i = 0; i < sav->sav_count; i++) 1638 nvlist_free(l2cache[i]); 1639 if (sav->sav_count) 1640 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1641} 1642 1643static int 1644load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1645{ 1646 dmu_buf_t *db; 1647 char *packed = NULL; 1648 size_t nvsize = 0; 1649 int error; 1650 *value = NULL; 1651 1652 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1653 if (error != 0) 1654 return (error); 1655 1656 nvsize = *(uint64_t *)db->db_data; 1657 dmu_buf_rele(db, FTAG); 1658 1659 packed = kmem_alloc(nvsize, KM_SLEEP); 1660 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1661 DMU_READ_PREFETCH); 1662 if (error == 0) 1663 error = nvlist_unpack(packed, nvsize, value, 0); 1664 kmem_free(packed, nvsize); 1665 1666 return (error); 1667} 1668 1669/* 1670 * Checks to see if the given vdev could not be opened, in which case we post a 1671 * sysevent to notify the autoreplace code that the device has been removed. 1672 */ 1673static void 1674spa_check_removed(vdev_t *vd) 1675{ 1676 for (int c = 0; c < vd->vdev_children; c++) 1677 spa_check_removed(vd->vdev_child[c]); 1678 1679 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1680 !vd->vdev_ishole) { 1681 zfs_post_autoreplace(vd->vdev_spa, vd); 1682 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1683 } 1684} 1685 1686static void 1687spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd) 1688{ 1689 ASSERT3U(vd->vdev_children, ==, mvd->vdev_children); 1690 1691 vd->vdev_top_zap = mvd->vdev_top_zap; 1692 vd->vdev_leaf_zap = mvd->vdev_leaf_zap; 1693 1694 for (uint64_t i = 0; i < vd->vdev_children; i++) { 1695 spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]); 1696 } 1697} 1698 1699/* 1700 * Validate the current config against the MOS config 1701 */ 1702static boolean_t 1703spa_config_valid(spa_t *spa, nvlist_t *config) 1704{ 1705 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1706 nvlist_t *nv; 1707 1708 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1709 1710 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1711 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1712 1713 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1714 1715 /* 1716 * If we're doing a normal import, then build up any additional 1717 * diagnostic information about missing devices in this config. 1718 * We'll pass this up to the user for further processing. 1719 */ 1720 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1721 nvlist_t **child, *nv; 1722 uint64_t idx = 0; 1723 1724 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1725 KM_SLEEP); 1726 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1727 1728 for (int c = 0; c < rvd->vdev_children; c++) { 1729 vdev_t *tvd = rvd->vdev_child[c]; 1730 vdev_t *mtvd = mrvd->vdev_child[c]; 1731 1732 if (tvd->vdev_ops == &vdev_missing_ops && 1733 mtvd->vdev_ops != &vdev_missing_ops && 1734 mtvd->vdev_islog) 1735 child[idx++] = vdev_config_generate(spa, mtvd, 1736 B_FALSE, 0); 1737 } 1738 1739 if (idx) { 1740 VERIFY(nvlist_add_nvlist_array(nv, 1741 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1742 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1743 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1744 1745 for (int i = 0; i < idx; i++) 1746 nvlist_free(child[i]); 1747 } 1748 nvlist_free(nv); 1749 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1750 } 1751 1752 /* 1753 * Compare the root vdev tree with the information we have 1754 * from the MOS config (mrvd). Check each top-level vdev 1755 * with the corresponding MOS config top-level (mtvd). 1756 */ 1757 for (int c = 0; c < rvd->vdev_children; c++) { 1758 vdev_t *tvd = rvd->vdev_child[c]; 1759 vdev_t *mtvd = mrvd->vdev_child[c]; 1760 1761 /* 1762 * Resolve any "missing" vdevs in the current configuration. 1763 * If we find that the MOS config has more accurate information 1764 * about the top-level vdev then use that vdev instead. 1765 */ 1766 if (tvd->vdev_ops == &vdev_missing_ops && 1767 mtvd->vdev_ops != &vdev_missing_ops) { 1768 1769 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1770 continue; 1771 1772 /* 1773 * Device specific actions. 1774 */ 1775 if (mtvd->vdev_islog) { 1776 spa_set_log_state(spa, SPA_LOG_CLEAR); 1777 } else { 1778 /* 1779 * XXX - once we have 'readonly' pool 1780 * support we should be able to handle 1781 * missing data devices by transitioning 1782 * the pool to readonly. 1783 */ 1784 continue; 1785 } 1786 1787 /* 1788 * Swap the missing vdev with the data we were 1789 * able to obtain from the MOS config. 1790 */ 1791 vdev_remove_child(rvd, tvd); 1792 vdev_remove_child(mrvd, mtvd); 1793 1794 vdev_add_child(rvd, mtvd); 1795 vdev_add_child(mrvd, tvd); 1796 1797 spa_config_exit(spa, SCL_ALL, FTAG); 1798 vdev_load(mtvd); 1799 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1800 1801 vdev_reopen(rvd); 1802 } else { 1803 if (mtvd->vdev_islog) { 1804 /* 1805 * Load the slog device's state from the MOS 1806 * config since it's possible that the label 1807 * does not contain the most up-to-date 1808 * information. 1809 */ 1810 vdev_load_log_state(tvd, mtvd); 1811 vdev_reopen(tvd); 1812 } 1813 1814 /* 1815 * Per-vdev ZAP info is stored exclusively in the MOS. 1816 */ 1817 spa_config_valid_zaps(tvd, mtvd); 1818 } 1819 } 1820 1821 vdev_free(mrvd); 1822 spa_config_exit(spa, SCL_ALL, FTAG); 1823 1824 /* 1825 * Ensure we were able to validate the config. 1826 */ 1827 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1828} 1829 1830/* 1831 * Check for missing log devices 1832 */ 1833static boolean_t 1834spa_check_logs(spa_t *spa) 1835{ 1836 boolean_t rv = B_FALSE; 1837 dsl_pool_t *dp = spa_get_dsl(spa); 1838 1839 switch (spa->spa_log_state) { 1840 case SPA_LOG_MISSING: 1841 /* need to recheck in case slog has been restored */ 1842 case SPA_LOG_UNKNOWN: 1843 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1844 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1845 if (rv) 1846 spa_set_log_state(spa, SPA_LOG_MISSING); 1847 break; 1848 } 1849 return (rv); 1850} 1851 1852static boolean_t 1853spa_passivate_log(spa_t *spa) 1854{ 1855 vdev_t *rvd = spa->spa_root_vdev; 1856 boolean_t slog_found = B_FALSE; 1857 1858 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1859 1860 if (!spa_has_slogs(spa)) 1861 return (B_FALSE); 1862 1863 for (int c = 0; c < rvd->vdev_children; c++) { 1864 vdev_t *tvd = rvd->vdev_child[c]; 1865 metaslab_group_t *mg = tvd->vdev_mg; 1866 1867 if (tvd->vdev_islog) { 1868 metaslab_group_passivate(mg); 1869 slog_found = B_TRUE; 1870 } 1871 } 1872 1873 return (slog_found); 1874} 1875 1876static void 1877spa_activate_log(spa_t *spa) 1878{ 1879 vdev_t *rvd = spa->spa_root_vdev; 1880 1881 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1882 1883 for (int c = 0; c < rvd->vdev_children; c++) { 1884 vdev_t *tvd = rvd->vdev_child[c]; 1885 metaslab_group_t *mg = tvd->vdev_mg; 1886 1887 if (tvd->vdev_islog) 1888 metaslab_group_activate(mg); 1889 } 1890} 1891 1892int 1893spa_offline_log(spa_t *spa) 1894{ 1895 int error; 1896 1897 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1898 NULL, DS_FIND_CHILDREN); 1899 if (error == 0) { 1900 /* 1901 * We successfully offlined the log device, sync out the 1902 * current txg so that the "stubby" block can be removed 1903 * by zil_sync(). 1904 */ 1905 txg_wait_synced(spa->spa_dsl_pool, 0); 1906 } 1907 return (error); 1908} 1909 1910static void 1911spa_aux_check_removed(spa_aux_vdev_t *sav) 1912{ 1913 int i; 1914 1915 for (i = 0; i < sav->sav_count; i++) 1916 spa_check_removed(sav->sav_vdevs[i]); 1917} 1918 1919void 1920spa_claim_notify(zio_t *zio) 1921{ 1922 spa_t *spa = zio->io_spa; 1923 1924 if (zio->io_error) 1925 return; 1926 1927 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1928 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1929 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1930 mutex_exit(&spa->spa_props_lock); 1931} 1932 1933typedef struct spa_load_error { 1934 uint64_t sle_meta_count; 1935 uint64_t sle_data_count; 1936} spa_load_error_t; 1937 1938static void 1939spa_load_verify_done(zio_t *zio) 1940{ 1941 blkptr_t *bp = zio->io_bp; 1942 spa_load_error_t *sle = zio->io_private; 1943 dmu_object_type_t type = BP_GET_TYPE(bp); 1944 int error = zio->io_error; 1945 spa_t *spa = zio->io_spa; 1946 1947 abd_free(zio->io_abd); 1948 if (error) { 1949 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1950 type != DMU_OT_INTENT_LOG) 1951 atomic_inc_64(&sle->sle_meta_count); 1952 else 1953 atomic_inc_64(&sle->sle_data_count); 1954 } 1955 1956 mutex_enter(&spa->spa_scrub_lock); 1957 spa->spa_scrub_inflight--; 1958 cv_broadcast(&spa->spa_scrub_io_cv); 1959 mutex_exit(&spa->spa_scrub_lock); 1960} 1961 1962/* 1963 * Maximum number of concurrent scrub i/os to create while verifying 1964 * a pool while importing it. 1965 */ 1966int spa_load_verify_maxinflight = 10000; 1967boolean_t spa_load_verify_metadata = B_TRUE; 1968boolean_t spa_load_verify_data = B_TRUE; 1969 1970SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1971 &spa_load_verify_maxinflight, 0, 1972 "Maximum number of concurrent scrub I/Os to create while verifying a " 1973 "pool while importing it"); 1974 1975SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1976 &spa_load_verify_metadata, 0, 1977 "Check metadata on import?"); 1978 1979SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1980 &spa_load_verify_data, 0, 1981 "Check user data on import?"); 1982 1983/*ARGSUSED*/ 1984static int 1985spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1986 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1987{ 1988 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1989 return (0); 1990 /* 1991 * Note: normally this routine will not be called if 1992 * spa_load_verify_metadata is not set. However, it may be useful 1993 * to manually set the flag after the traversal has begun. 1994 */ 1995 if (!spa_load_verify_metadata) 1996 return (0); 1997 if (!BP_IS_METADATA(bp) && !spa_load_verify_data) 1998 return (0); 1999 2000 zio_t *rio = arg; 2001 size_t size = BP_GET_PSIZE(bp); 2002 2003 mutex_enter(&spa->spa_scrub_lock); 2004 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 2005 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 2006 spa->spa_scrub_inflight++; 2007 mutex_exit(&spa->spa_scrub_lock); 2008 2009 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, 2010 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 2011 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 2012 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 2013 return (0); 2014} 2015 2016/* ARGSUSED */ 2017int 2018verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 2019{ 2020 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 2021 return (SET_ERROR(ENAMETOOLONG)); 2022 2023 return (0); 2024} 2025 2026static int 2027spa_load_verify(spa_t *spa) 2028{ 2029 zio_t *rio; 2030 spa_load_error_t sle = { 0 }; 2031 zpool_rewind_policy_t policy; 2032 boolean_t verify_ok = B_FALSE; 2033 int error = 0; 2034 2035 zpool_get_rewind_policy(spa->spa_config, &policy); 2036 2037 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 2038 return (0); 2039 2040 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2041 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2042 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2043 DS_FIND_CHILDREN); 2044 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2045 if (error != 0) 2046 return (error); 2047 2048 rio = zio_root(spa, NULL, &sle, 2049 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2050 2051 if (spa_load_verify_metadata) { 2052 error = traverse_pool(spa, spa->spa_verify_min_txg, 2053 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2054 spa_load_verify_cb, rio); 2055 } 2056 2057 (void) zio_wait(rio); 2058 2059 spa->spa_load_meta_errors = sle.sle_meta_count; 2060 spa->spa_load_data_errors = sle.sle_data_count; 2061 2062 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2063 sle.sle_data_count <= policy.zrp_maxdata) { 2064 int64_t loss = 0; 2065 2066 verify_ok = B_TRUE; 2067 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2068 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2069 2070 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2071 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2072 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2073 VERIFY(nvlist_add_int64(spa->spa_load_info, 2074 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2075 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2076 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2077 } else { 2078 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2079 } 2080 2081 if (error) { 2082 if (error != ENXIO && error != EIO) 2083 error = SET_ERROR(EIO); 2084 return (error); 2085 } 2086 2087 return (verify_ok ? 0 : EIO); 2088} 2089 2090/* 2091 * Find a value in the pool props object. 2092 */ 2093static void 2094spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2095{ 2096 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2097 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2098} 2099 2100/* 2101 * Find a value in the pool directory object. 2102 */ 2103static int 2104spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2105{ 2106 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2107 name, sizeof (uint64_t), 1, val)); 2108} 2109 2110static int 2111spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2112{ 2113 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2114 return (err); 2115} 2116 2117/* 2118 * Fix up config after a partly-completed split. This is done with the 2119 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2120 * pool have that entry in their config, but only the splitting one contains 2121 * a list of all the guids of the vdevs that are being split off. 2122 * 2123 * This function determines what to do with that list: either rejoin 2124 * all the disks to the pool, or complete the splitting process. To attempt 2125 * the rejoin, each disk that is offlined is marked online again, and 2126 * we do a reopen() call. If the vdev label for every disk that was 2127 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2128 * then we call vdev_split() on each disk, and complete the split. 2129 * 2130 * Otherwise we leave the config alone, with all the vdevs in place in 2131 * the original pool. 2132 */ 2133static void 2134spa_try_repair(spa_t *spa, nvlist_t *config) 2135{ 2136 uint_t extracted; 2137 uint64_t *glist; 2138 uint_t i, gcount; 2139 nvlist_t *nvl; 2140 vdev_t **vd; 2141 boolean_t attempt_reopen; 2142 2143 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2144 return; 2145 2146 /* check that the config is complete */ 2147 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2148 &glist, &gcount) != 0) 2149 return; 2150 2151 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2152 2153 /* attempt to online all the vdevs & validate */ 2154 attempt_reopen = B_TRUE; 2155 for (i = 0; i < gcount; i++) { 2156 if (glist[i] == 0) /* vdev is hole */ 2157 continue; 2158 2159 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2160 if (vd[i] == NULL) { 2161 /* 2162 * Don't bother attempting to reopen the disks; 2163 * just do the split. 2164 */ 2165 attempt_reopen = B_FALSE; 2166 } else { 2167 /* attempt to re-online it */ 2168 vd[i]->vdev_offline = B_FALSE; 2169 } 2170 } 2171 2172 if (attempt_reopen) { 2173 vdev_reopen(spa->spa_root_vdev); 2174 2175 /* check each device to see what state it's in */ 2176 for (extracted = 0, i = 0; i < gcount; i++) { 2177 if (vd[i] != NULL && 2178 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2179 break; 2180 ++extracted; 2181 } 2182 } 2183 2184 /* 2185 * If every disk has been moved to the new pool, or if we never 2186 * even attempted to look at them, then we split them off for 2187 * good. 2188 */ 2189 if (!attempt_reopen || gcount == extracted) { 2190 for (i = 0; i < gcount; i++) 2191 if (vd[i] != NULL) 2192 vdev_split(vd[i]); 2193 vdev_reopen(spa->spa_root_vdev); 2194 } 2195 2196 kmem_free(vd, gcount * sizeof (vdev_t *)); 2197} 2198 2199static int 2200spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2201 boolean_t mosconfig) 2202{ 2203 nvlist_t *config = spa->spa_config; 2204 char *ereport = FM_EREPORT_ZFS_POOL; 2205 char *comment; 2206 int error; 2207 uint64_t pool_guid; 2208 nvlist_t *nvl; 2209 2210 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2211 return (SET_ERROR(EINVAL)); 2212 2213 ASSERT(spa->spa_comment == NULL); 2214 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2215 spa->spa_comment = spa_strdup(comment); 2216 2217 /* 2218 * Versioning wasn't explicitly added to the label until later, so if 2219 * it's not present treat it as the initial version. 2220 */ 2221 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2222 &spa->spa_ubsync.ub_version) != 0) 2223 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2224 2225 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2226 &spa->spa_config_txg); 2227 2228 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2229 spa_guid_exists(pool_guid, 0)) { 2230 error = SET_ERROR(EEXIST); 2231 } else { 2232 spa->spa_config_guid = pool_guid; 2233 2234 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2235 &nvl) == 0) { 2236 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2237 KM_SLEEP) == 0); 2238 } 2239 2240 nvlist_free(spa->spa_load_info); 2241 spa->spa_load_info = fnvlist_alloc(); 2242 2243 gethrestime(&spa->spa_loaded_ts); 2244 error = spa_load_impl(spa, pool_guid, config, state, type, 2245 mosconfig, &ereport); 2246 } 2247 2248 /* 2249 * Don't count references from objsets that are already closed 2250 * and are making their way through the eviction process. 2251 */ 2252 spa_evicting_os_wait(spa); 2253 spa->spa_minref = refcount_count(&spa->spa_refcount); 2254 if (error) { 2255 if (error != EEXIST) { 2256 spa->spa_loaded_ts.tv_sec = 0; 2257 spa->spa_loaded_ts.tv_nsec = 0; 2258 } 2259 if (error != EBADF) { 2260 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2261 } 2262 } 2263 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2264 spa->spa_ena = 0; 2265 2266 return (error); 2267} 2268 2269/* 2270 * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2271 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2272 * spa's per-vdev ZAP list. 2273 */ 2274static uint64_t 2275vdev_count_verify_zaps(vdev_t *vd) 2276{ 2277 spa_t *spa = vd->vdev_spa; 2278 uint64_t total = 0; 2279 if (vd->vdev_top_zap != 0) { 2280 total++; 2281 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2282 spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2283 } 2284 if (vd->vdev_leaf_zap != 0) { 2285 total++; 2286 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2287 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2288 } 2289 2290 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2291 total += vdev_count_verify_zaps(vd->vdev_child[i]); 2292 } 2293 2294 return (total); 2295} 2296 2297/* 2298 * Load an existing storage pool, using the pool's builtin spa_config as a 2299 * source of configuration information. 2300 */ 2301static int 2302spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2303 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2304 char **ereport) 2305{ 2306 int error = 0; 2307 nvlist_t *nvroot = NULL; 2308 nvlist_t *label; 2309 vdev_t *rvd; 2310 uberblock_t *ub = &spa->spa_uberblock; 2311 uint64_t children, config_cache_txg = spa->spa_config_txg; 2312 int orig_mode = spa->spa_mode; 2313 int parse; 2314 uint64_t obj; 2315 boolean_t missing_feat_write = B_FALSE; 2316 2317 /* 2318 * If this is an untrusted config, access the pool in read-only mode. 2319 * This prevents things like resilvering recently removed devices. 2320 */ 2321 if (!mosconfig) 2322 spa->spa_mode = FREAD; 2323 2324 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2325 2326 spa->spa_load_state = state; 2327 2328 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2329 return (SET_ERROR(EINVAL)); 2330 2331 parse = (type == SPA_IMPORT_EXISTING ? 2332 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2333 2334 /* 2335 * Create "The Godfather" zio to hold all async IOs 2336 */ 2337 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2338 KM_SLEEP); 2339 for (int i = 0; i < max_ncpus; i++) { 2340 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2341 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2342 ZIO_FLAG_GODFATHER); 2343 } 2344 2345 /* 2346 * Parse the configuration into a vdev tree. We explicitly set the 2347 * value that will be returned by spa_version() since parsing the 2348 * configuration requires knowing the version number. 2349 */ 2350 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2351 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2352 spa_config_exit(spa, SCL_ALL, FTAG); 2353 2354 if (error != 0) 2355 return (error); 2356 2357 ASSERT(spa->spa_root_vdev == rvd); 2358 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2359 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2360 2361 if (type != SPA_IMPORT_ASSEMBLE) { 2362 ASSERT(spa_guid(spa) == pool_guid); 2363 } 2364 2365 /* 2366 * Try to open all vdevs, loading each label in the process. 2367 */ 2368 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2369 error = vdev_open(rvd); 2370 spa_config_exit(spa, SCL_ALL, FTAG); 2371 if (error != 0) 2372 return (error); 2373 2374 /* 2375 * We need to validate the vdev labels against the configuration that 2376 * we have in hand, which is dependent on the setting of mosconfig. If 2377 * mosconfig is true then we're validating the vdev labels based on 2378 * that config. Otherwise, we're validating against the cached config 2379 * (zpool.cache) that was read when we loaded the zfs module, and then 2380 * later we will recursively call spa_load() and validate against 2381 * the vdev config. 2382 * 2383 * If we're assembling a new pool that's been split off from an 2384 * existing pool, the labels haven't yet been updated so we skip 2385 * validation for now. 2386 */ 2387 if (type != SPA_IMPORT_ASSEMBLE) { 2388 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2389 error = vdev_validate(rvd, mosconfig); 2390 spa_config_exit(spa, SCL_ALL, FTAG); 2391 2392 if (error != 0) 2393 return (error); 2394 2395 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2396 return (SET_ERROR(ENXIO)); 2397 } 2398 2399 /* 2400 * Find the best uberblock. 2401 */ 2402 vdev_uberblock_load(rvd, ub, &label); 2403 2404 /* 2405 * If we weren't able to find a single valid uberblock, return failure. 2406 */ 2407 if (ub->ub_txg == 0) { 2408 nvlist_free(label); 2409 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2410 } 2411 2412 /* 2413 * If the pool has an unsupported version we can't open it. 2414 */ 2415 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2416 nvlist_free(label); 2417 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2418 } 2419 2420 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2421 nvlist_t *features; 2422 2423 /* 2424 * If we weren't able to find what's necessary for reading the 2425 * MOS in the label, return failure. 2426 */ 2427 if (label == NULL || nvlist_lookup_nvlist(label, 2428 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2429 nvlist_free(label); 2430 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2431 ENXIO)); 2432 } 2433 2434 /* 2435 * Update our in-core representation with the definitive values 2436 * from the label. 2437 */ 2438 nvlist_free(spa->spa_label_features); 2439 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2440 } 2441 2442 nvlist_free(label); 2443 2444 /* 2445 * Look through entries in the label nvlist's features_for_read. If 2446 * there is a feature listed there which we don't understand then we 2447 * cannot open a pool. 2448 */ 2449 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2450 nvlist_t *unsup_feat; 2451 2452 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2453 0); 2454 2455 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2456 NULL); nvp != NULL; 2457 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2458 if (!zfeature_is_supported(nvpair_name(nvp))) { 2459 VERIFY(nvlist_add_string(unsup_feat, 2460 nvpair_name(nvp), "") == 0); 2461 } 2462 } 2463 2464 if (!nvlist_empty(unsup_feat)) { 2465 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2466 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2467 nvlist_free(unsup_feat); 2468 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2469 ENOTSUP)); 2470 } 2471 2472 nvlist_free(unsup_feat); 2473 } 2474 2475 /* 2476 * If the vdev guid sum doesn't match the uberblock, we have an 2477 * incomplete configuration. We first check to see if the pool 2478 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2479 * If it is, defer the vdev_guid_sum check till later so we 2480 * can handle missing vdevs. 2481 */ 2482 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2483 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2484 rvd->vdev_guid_sum != ub->ub_guid_sum) 2485 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2486 2487 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2488 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2489 spa_try_repair(spa, config); 2490 spa_config_exit(spa, SCL_ALL, FTAG); 2491 nvlist_free(spa->spa_config_splitting); 2492 spa->spa_config_splitting = NULL; 2493 } 2494 2495 /* 2496 * Initialize internal SPA structures. 2497 */ 2498 spa->spa_state = POOL_STATE_ACTIVE; 2499 spa->spa_ubsync = spa->spa_uberblock; 2500 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2501 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2502 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2503 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2504 spa->spa_claim_max_txg = spa->spa_first_txg; 2505 spa->spa_prev_software_version = ub->ub_software_version; 2506 2507 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2508 if (error) 2509 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2510 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2511 2512 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2513 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2514 2515 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2516 boolean_t missing_feat_read = B_FALSE; 2517 nvlist_t *unsup_feat, *enabled_feat; 2518 2519 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2520 &spa->spa_feat_for_read_obj) != 0) { 2521 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2522 } 2523 2524 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2525 &spa->spa_feat_for_write_obj) != 0) { 2526 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2527 } 2528 2529 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2530 &spa->spa_feat_desc_obj) != 0) { 2531 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2532 } 2533 2534 enabled_feat = fnvlist_alloc(); 2535 unsup_feat = fnvlist_alloc(); 2536 2537 if (!spa_features_check(spa, B_FALSE, 2538 unsup_feat, enabled_feat)) 2539 missing_feat_read = B_TRUE; 2540 2541 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2542 if (!spa_features_check(spa, B_TRUE, 2543 unsup_feat, enabled_feat)) { 2544 missing_feat_write = B_TRUE; 2545 } 2546 } 2547 2548 fnvlist_add_nvlist(spa->spa_load_info, 2549 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2550 2551 if (!nvlist_empty(unsup_feat)) { 2552 fnvlist_add_nvlist(spa->spa_load_info, 2553 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2554 } 2555 2556 fnvlist_free(enabled_feat); 2557 fnvlist_free(unsup_feat); 2558 2559 if (!missing_feat_read) { 2560 fnvlist_add_boolean(spa->spa_load_info, 2561 ZPOOL_CONFIG_CAN_RDONLY); 2562 } 2563 2564 /* 2565 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2566 * twofold: to determine whether the pool is available for 2567 * import in read-write mode and (if it is not) whether the 2568 * pool is available for import in read-only mode. If the pool 2569 * is available for import in read-write mode, it is displayed 2570 * as available in userland; if it is not available for import 2571 * in read-only mode, it is displayed as unavailable in 2572 * userland. If the pool is available for import in read-only 2573 * mode but not read-write mode, it is displayed as unavailable 2574 * in userland with a special note that the pool is actually 2575 * available for open in read-only mode. 2576 * 2577 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2578 * missing a feature for write, we must first determine whether 2579 * the pool can be opened read-only before returning to 2580 * userland in order to know whether to display the 2581 * abovementioned note. 2582 */ 2583 if (missing_feat_read || (missing_feat_write && 2584 spa_writeable(spa))) { 2585 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2586 ENOTSUP)); 2587 } 2588 2589 /* 2590 * Load refcounts for ZFS features from disk into an in-memory 2591 * cache during SPA initialization. 2592 */ 2593 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2594 uint64_t refcount; 2595 2596 error = feature_get_refcount_from_disk(spa, 2597 &spa_feature_table[i], &refcount); 2598 if (error == 0) { 2599 spa->spa_feat_refcount_cache[i] = refcount; 2600 } else if (error == ENOTSUP) { 2601 spa->spa_feat_refcount_cache[i] = 2602 SPA_FEATURE_DISABLED; 2603 } else { 2604 return (spa_vdev_err(rvd, 2605 VDEV_AUX_CORRUPT_DATA, EIO)); 2606 } 2607 } 2608 } 2609 2610 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2611 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2612 &spa->spa_feat_enabled_txg_obj) != 0) 2613 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2614 } 2615 2616 spa->spa_is_initializing = B_TRUE; 2617 error = dsl_pool_open(spa->spa_dsl_pool); 2618 spa->spa_is_initializing = B_FALSE; 2619 if (error != 0) 2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2621 2622 if (!mosconfig) { 2623 uint64_t hostid; 2624 nvlist_t *policy = NULL, *nvconfig; 2625 2626 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2627 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2628 2629 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2630 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2631 char *hostname; 2632 unsigned long myhostid = 0; 2633 2634 VERIFY(nvlist_lookup_string(nvconfig, 2635 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2636 2637#ifdef _KERNEL 2638 myhostid = zone_get_hostid(NULL); 2639#else /* _KERNEL */ 2640 /* 2641 * We're emulating the system's hostid in userland, so 2642 * we can't use zone_get_hostid(). 2643 */ 2644 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2645#endif /* _KERNEL */ 2646 if (check_hostid && hostid != 0 && myhostid != 0 && 2647 hostid != myhostid) { 2648 nvlist_free(nvconfig); 2649 cmn_err(CE_WARN, "pool '%s' could not be " 2650 "loaded as it was last accessed by " 2651 "another system (host: %s hostid: 0x%lx). " 2652 "See: http://illumos.org/msg/ZFS-8000-EY", 2653 spa_name(spa), hostname, 2654 (unsigned long)hostid); 2655 return (SET_ERROR(EBADF)); 2656 } 2657 } 2658 if (nvlist_lookup_nvlist(spa->spa_config, 2659 ZPOOL_REWIND_POLICY, &policy) == 0) 2660 VERIFY(nvlist_add_nvlist(nvconfig, 2661 ZPOOL_REWIND_POLICY, policy) == 0); 2662 2663 spa_config_set(spa, nvconfig); 2664 spa_unload(spa); 2665 spa_deactivate(spa); 2666 spa_activate(spa, orig_mode); 2667 2668 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2669 } 2670 2671 /* Grab the secret checksum salt from the MOS. */ 2672 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2673 DMU_POOL_CHECKSUM_SALT, 1, 2674 sizeof (spa->spa_cksum_salt.zcs_bytes), 2675 spa->spa_cksum_salt.zcs_bytes); 2676 if (error == ENOENT) { 2677 /* Generate a new salt for subsequent use */ 2678 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 2679 sizeof (spa->spa_cksum_salt.zcs_bytes)); 2680 } else if (error != 0) { 2681 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2682 } 2683 2684 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2685 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2686 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2687 if (error != 0) 2688 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2689 2690 /* 2691 * Load the bit that tells us to use the new accounting function 2692 * (raid-z deflation). If we have an older pool, this will not 2693 * be present. 2694 */ 2695 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2696 if (error != 0 && error != ENOENT) 2697 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2698 2699 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2700 &spa->spa_creation_version); 2701 if (error != 0 && error != ENOENT) 2702 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2703 2704 /* 2705 * Load the persistent error log. If we have an older pool, this will 2706 * not be present. 2707 */ 2708 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2709 if (error != 0 && error != ENOENT) 2710 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2711 2712 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2713 &spa->spa_errlog_scrub); 2714 if (error != 0 && error != ENOENT) 2715 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2716 2717 /* 2718 * Load the history object. If we have an older pool, this 2719 * will not be present. 2720 */ 2721 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2722 if (error != 0 && error != ENOENT) 2723 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2724 2725 /* 2726 * Load the per-vdev ZAP map. If we have an older pool, this will not 2727 * be present; in this case, defer its creation to a later time to 2728 * avoid dirtying the MOS this early / out of sync context. See 2729 * spa_sync_config_object. 2730 */ 2731 2732 /* The sentinel is only available in the MOS config. */ 2733 nvlist_t *mos_config; 2734 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) 2735 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2736 2737 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 2738 &spa->spa_all_vdev_zaps); 2739 2740 if (error == ENOENT) { 2741 VERIFY(!nvlist_exists(mos_config, 2742 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 2743 spa->spa_avz_action = AVZ_ACTION_INITIALIZE; 2744 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 2745 } else if (error != 0) { 2746 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2747 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 2748 /* 2749 * An older version of ZFS overwrote the sentinel value, so 2750 * we have orphaned per-vdev ZAPs in the MOS. Defer their 2751 * destruction to later; see spa_sync_config_object. 2752 */ 2753 spa->spa_avz_action = AVZ_ACTION_DESTROY; 2754 /* 2755 * We're assuming that no vdevs have had their ZAPs created 2756 * before this. Better be sure of it. 2757 */ 2758 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 2759 } 2760 nvlist_free(mos_config); 2761 2762 /* 2763 * If we're assembling the pool from the split-off vdevs of 2764 * an existing pool, we don't want to attach the spares & cache 2765 * devices. 2766 */ 2767 2768 /* 2769 * Load any hot spares for this pool. 2770 */ 2771 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2772 if (error != 0 && error != ENOENT) 2773 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2774 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2775 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2776 if (load_nvlist(spa, spa->spa_spares.sav_object, 2777 &spa->spa_spares.sav_config) != 0) 2778 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2779 2780 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2781 spa_load_spares(spa); 2782 spa_config_exit(spa, SCL_ALL, FTAG); 2783 } else if (error == 0) { 2784 spa->spa_spares.sav_sync = B_TRUE; 2785 } 2786 2787 /* 2788 * Load any level 2 ARC devices for this pool. 2789 */ 2790 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2791 &spa->spa_l2cache.sav_object); 2792 if (error != 0 && error != ENOENT) 2793 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2794 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2795 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2796 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2797 &spa->spa_l2cache.sav_config) != 0) 2798 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2799 2800 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2801 spa_load_l2cache(spa); 2802 spa_config_exit(spa, SCL_ALL, FTAG); 2803 } else if (error == 0) { 2804 spa->spa_l2cache.sav_sync = B_TRUE; 2805 } 2806 2807 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2808 2809 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2810 if (error && error != ENOENT) 2811 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2812 2813 if (error == 0) { 2814 uint64_t autoreplace; 2815 2816 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2817 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2818 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2819 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2820 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2821 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2822 &spa->spa_dedup_ditto); 2823 2824 spa->spa_autoreplace = (autoreplace != 0); 2825 } 2826 2827 /* 2828 * If the 'autoreplace' property is set, then post a resource notifying 2829 * the ZFS DE that it should not issue any faults for unopenable 2830 * devices. We also iterate over the vdevs, and post a sysevent for any 2831 * unopenable vdevs so that the normal autoreplace handler can take 2832 * over. 2833 */ 2834 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2835 spa_check_removed(spa->spa_root_vdev); 2836 /* 2837 * For the import case, this is done in spa_import(), because 2838 * at this point we're using the spare definitions from 2839 * the MOS config, not necessarily from the userland config. 2840 */ 2841 if (state != SPA_LOAD_IMPORT) { 2842 spa_aux_check_removed(&spa->spa_spares); 2843 spa_aux_check_removed(&spa->spa_l2cache); 2844 } 2845 } 2846 2847 /* 2848 * Load the vdev state for all toplevel vdevs. 2849 */ 2850 vdev_load(rvd); 2851 2852 /* 2853 * Propagate the leaf DTLs we just loaded all the way up the tree. 2854 */ 2855 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2856 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2857 spa_config_exit(spa, SCL_ALL, FTAG); 2858 2859 /* 2860 * Load the DDTs (dedup tables). 2861 */ 2862 error = ddt_load(spa); 2863 if (error != 0) 2864 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2865 2866 spa_update_dspace(spa); 2867 2868 /* 2869 * Validate the config, using the MOS config to fill in any 2870 * information which might be missing. If we fail to validate 2871 * the config then declare the pool unfit for use. If we're 2872 * assembling a pool from a split, the log is not transferred 2873 * over. 2874 */ 2875 if (type != SPA_IMPORT_ASSEMBLE) { 2876 nvlist_t *nvconfig; 2877 2878 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2879 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2880 2881 if (!spa_config_valid(spa, nvconfig)) { 2882 nvlist_free(nvconfig); 2883 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2884 ENXIO)); 2885 } 2886 nvlist_free(nvconfig); 2887 2888 /* 2889 * Now that we've validated the config, check the state of the 2890 * root vdev. If it can't be opened, it indicates one or 2891 * more toplevel vdevs are faulted. 2892 */ 2893 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2894 return (SET_ERROR(ENXIO)); 2895 2896 if (spa_writeable(spa) && spa_check_logs(spa)) { 2897 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2898 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2899 } 2900 } 2901 2902 if (missing_feat_write) { 2903 ASSERT(state == SPA_LOAD_TRYIMPORT); 2904 2905 /* 2906 * At this point, we know that we can open the pool in 2907 * read-only mode but not read-write mode. We now have enough 2908 * information and can return to userland. 2909 */ 2910 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2911 } 2912 2913 /* 2914 * We've successfully opened the pool, verify that we're ready 2915 * to start pushing transactions. 2916 */ 2917 if (state != SPA_LOAD_TRYIMPORT) { 2918 if (error = spa_load_verify(spa)) 2919 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2920 error)); 2921 } 2922 2923 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2924 spa->spa_load_max_txg == UINT64_MAX)) { 2925 dmu_tx_t *tx; 2926 int need_update = B_FALSE; 2927 dsl_pool_t *dp = spa_get_dsl(spa); 2928 2929 ASSERT(state != SPA_LOAD_TRYIMPORT); 2930 2931 /* 2932 * Claim log blocks that haven't been committed yet. 2933 * This must all happen in a single txg. 2934 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2935 * invoked from zil_claim_log_block()'s i/o done callback. 2936 * Price of rollback is that we abandon the log. 2937 */ 2938 spa->spa_claiming = B_TRUE; 2939 2940 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2941 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2942 zil_claim, tx, DS_FIND_CHILDREN); 2943 dmu_tx_commit(tx); 2944 2945 spa->spa_claiming = B_FALSE; 2946 2947 spa_set_log_state(spa, SPA_LOG_GOOD); 2948 spa->spa_sync_on = B_TRUE; 2949 txg_sync_start(spa->spa_dsl_pool); 2950 2951 /* 2952 * Wait for all claims to sync. We sync up to the highest 2953 * claimed log block birth time so that claimed log blocks 2954 * don't appear to be from the future. spa_claim_max_txg 2955 * will have been set for us by either zil_check_log_chain() 2956 * (invoked from spa_check_logs()) or zil_claim() above. 2957 */ 2958 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2959 2960 /* 2961 * If the config cache is stale, or we have uninitialized 2962 * metaslabs (see spa_vdev_add()), then update the config. 2963 * 2964 * If this is a verbatim import, trust the current 2965 * in-core spa_config and update the disk labels. 2966 */ 2967 if (config_cache_txg != spa->spa_config_txg || 2968 state == SPA_LOAD_IMPORT || 2969 state == SPA_LOAD_RECOVER || 2970 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2971 need_update = B_TRUE; 2972 2973 for (int c = 0; c < rvd->vdev_children; c++) 2974 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2975 need_update = B_TRUE; 2976 2977 /* 2978 * Update the config cache asychronously in case we're the 2979 * root pool, in which case the config cache isn't writable yet. 2980 */ 2981 if (need_update) 2982 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2983 2984 /* 2985 * Check all DTLs to see if anything needs resilvering. 2986 */ 2987 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2988 vdev_resilver_needed(rvd, NULL, NULL)) 2989 spa_async_request(spa, SPA_ASYNC_RESILVER); 2990 2991 /* 2992 * Log the fact that we booted up (so that we can detect if 2993 * we rebooted in the middle of an operation). 2994 */ 2995 spa_history_log_version(spa, "open"); 2996 2997 /* 2998 * Delete any inconsistent datasets. 2999 */ 3000 (void) dmu_objset_find(spa_name(spa), 3001 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 3002 3003 /* 3004 * Clean up any stale temporary dataset userrefs. 3005 */ 3006 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 3007 } 3008 3009 return (0); 3010} 3011 3012static int 3013spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 3014{ 3015 int mode = spa->spa_mode; 3016 3017 spa_unload(spa); 3018 spa_deactivate(spa); 3019 3020 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 3021 3022 spa_activate(spa, mode); 3023 spa_async_suspend(spa); 3024 3025 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 3026} 3027 3028/* 3029 * If spa_load() fails this function will try loading prior txg's. If 3030 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 3031 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 3032 * function will not rewind the pool and will return the same error as 3033 * spa_load(). 3034 */ 3035static int 3036spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 3037 uint64_t max_request, int rewind_flags) 3038{ 3039 nvlist_t *loadinfo = NULL; 3040 nvlist_t *config = NULL; 3041 int load_error, rewind_error; 3042 uint64_t safe_rewind_txg; 3043 uint64_t min_txg; 3044 3045 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 3046 spa->spa_load_max_txg = spa->spa_load_txg; 3047 spa_set_log_state(spa, SPA_LOG_CLEAR); 3048 } else { 3049 spa->spa_load_max_txg = max_request; 3050 if (max_request != UINT64_MAX) 3051 spa->spa_extreme_rewind = B_TRUE; 3052 } 3053 3054 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 3055 mosconfig); 3056 if (load_error == 0) 3057 return (0); 3058 3059 if (spa->spa_root_vdev != NULL) 3060 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3061 3062 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 3063 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 3064 3065 if (rewind_flags & ZPOOL_NEVER_REWIND) { 3066 nvlist_free(config); 3067 return (load_error); 3068 } 3069 3070 if (state == SPA_LOAD_RECOVER) { 3071 /* Price of rolling back is discarding txgs, including log */ 3072 spa_set_log_state(spa, SPA_LOG_CLEAR); 3073 } else { 3074 /* 3075 * If we aren't rolling back save the load info from our first 3076 * import attempt so that we can restore it after attempting 3077 * to rewind. 3078 */ 3079 loadinfo = spa->spa_load_info; 3080 spa->spa_load_info = fnvlist_alloc(); 3081 } 3082 3083 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 3084 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 3085 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 3086 TXG_INITIAL : safe_rewind_txg; 3087 3088 /* 3089 * Continue as long as we're finding errors, we're still within 3090 * the acceptable rewind range, and we're still finding uberblocks 3091 */ 3092 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 3093 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 3094 if (spa->spa_load_max_txg < safe_rewind_txg) 3095 spa->spa_extreme_rewind = B_TRUE; 3096 rewind_error = spa_load_retry(spa, state, mosconfig); 3097 } 3098 3099 spa->spa_extreme_rewind = B_FALSE; 3100 spa->spa_load_max_txg = UINT64_MAX; 3101 3102 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 3103 spa_config_set(spa, config); 3104 3105 if (state == SPA_LOAD_RECOVER) { 3106 ASSERT3P(loadinfo, ==, NULL); 3107 return (rewind_error); 3108 } else { 3109 /* Store the rewind info as part of the initial load info */ 3110 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3111 spa->spa_load_info); 3112 3113 /* Restore the initial load info */ 3114 fnvlist_free(spa->spa_load_info); 3115 spa->spa_load_info = loadinfo; 3116 3117 return (load_error); 3118 } 3119} 3120 3121/* 3122 * Pool Open/Import 3123 * 3124 * The import case is identical to an open except that the configuration is sent 3125 * down from userland, instead of grabbed from the configuration cache. For the 3126 * case of an open, the pool configuration will exist in the 3127 * POOL_STATE_UNINITIALIZED state. 3128 * 3129 * The stats information (gen/count/ustats) is used to gather vdev statistics at 3130 * the same time open the pool, without having to keep around the spa_t in some 3131 * ambiguous state. 3132 */ 3133static int 3134spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3135 nvlist_t **config) 3136{ 3137 spa_t *spa; 3138 spa_load_state_t state = SPA_LOAD_OPEN; 3139 int error; 3140 int locked = B_FALSE; 3141 int firstopen = B_FALSE; 3142 3143 *spapp = NULL; 3144 3145 /* 3146 * As disgusting as this is, we need to support recursive calls to this 3147 * function because dsl_dir_open() is called during spa_load(), and ends 3148 * up calling spa_open() again. The real fix is to figure out how to 3149 * avoid dsl_dir_open() calling this in the first place. 3150 */ 3151 if (mutex_owner(&spa_namespace_lock) != curthread) { 3152 mutex_enter(&spa_namespace_lock); 3153 locked = B_TRUE; 3154 } 3155 3156 if ((spa = spa_lookup(pool)) == NULL) { 3157 if (locked) 3158 mutex_exit(&spa_namespace_lock); 3159 return (SET_ERROR(ENOENT)); 3160 } 3161 3162 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3163 zpool_rewind_policy_t policy; 3164 3165 firstopen = B_TRUE; 3166 3167 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3168 &policy); 3169 if (policy.zrp_request & ZPOOL_DO_REWIND) 3170 state = SPA_LOAD_RECOVER; 3171 3172 spa_activate(spa, spa_mode_global); 3173 3174 if (state != SPA_LOAD_RECOVER) 3175 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3176 3177 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3178 policy.zrp_request); 3179 3180 if (error == EBADF) { 3181 /* 3182 * If vdev_validate() returns failure (indicated by 3183 * EBADF), it indicates that one of the vdevs indicates 3184 * that the pool has been exported or destroyed. If 3185 * this is the case, the config cache is out of sync and 3186 * we should remove the pool from the namespace. 3187 */ 3188 spa_unload(spa); 3189 spa_deactivate(spa); 3190 spa_config_sync(spa, B_TRUE, B_TRUE); 3191 spa_remove(spa); 3192 if (locked) 3193 mutex_exit(&spa_namespace_lock); 3194 return (SET_ERROR(ENOENT)); 3195 } 3196 3197 if (error) { 3198 /* 3199 * We can't open the pool, but we still have useful 3200 * information: the state of each vdev after the 3201 * attempted vdev_open(). Return this to the user. 3202 */ 3203 if (config != NULL && spa->spa_config) { 3204 VERIFY(nvlist_dup(spa->spa_config, config, 3205 KM_SLEEP) == 0); 3206 VERIFY(nvlist_add_nvlist(*config, 3207 ZPOOL_CONFIG_LOAD_INFO, 3208 spa->spa_load_info) == 0); 3209 } 3210 spa_unload(spa); 3211 spa_deactivate(spa); 3212 spa->spa_last_open_failed = error; 3213 if (locked) 3214 mutex_exit(&spa_namespace_lock); 3215 *spapp = NULL; 3216 return (error); 3217 } 3218 } 3219 3220 spa_open_ref(spa, tag); 3221 3222 if (config != NULL) 3223 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3224 3225 /* 3226 * If we've recovered the pool, pass back any information we 3227 * gathered while doing the load. 3228 */ 3229 if (state == SPA_LOAD_RECOVER) { 3230 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3231 spa->spa_load_info) == 0); 3232 } 3233 3234 if (locked) { 3235 spa->spa_last_open_failed = 0; 3236 spa->spa_last_ubsync_txg = 0; 3237 spa->spa_load_txg = 0; 3238 mutex_exit(&spa_namespace_lock); 3239#ifdef __FreeBSD__ 3240#ifdef _KERNEL 3241 if (firstopen) 3242 zvol_create_minors(spa->spa_name); 3243#endif 3244#endif 3245 } 3246 3247 *spapp = spa; 3248 3249 return (0); 3250} 3251 3252int 3253spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3254 nvlist_t **config) 3255{ 3256 return (spa_open_common(name, spapp, tag, policy, config)); 3257} 3258 3259int 3260spa_open(const char *name, spa_t **spapp, void *tag) 3261{ 3262 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3263} 3264 3265/* 3266 * Lookup the given spa_t, incrementing the inject count in the process, 3267 * preventing it from being exported or destroyed. 3268 */ 3269spa_t * 3270spa_inject_addref(char *name) 3271{ 3272 spa_t *spa; 3273 3274 mutex_enter(&spa_namespace_lock); 3275 if ((spa = spa_lookup(name)) == NULL) { 3276 mutex_exit(&spa_namespace_lock); 3277 return (NULL); 3278 } 3279 spa->spa_inject_ref++; 3280 mutex_exit(&spa_namespace_lock); 3281 3282 return (spa); 3283} 3284 3285void 3286spa_inject_delref(spa_t *spa) 3287{ 3288 mutex_enter(&spa_namespace_lock); 3289 spa->spa_inject_ref--; 3290 mutex_exit(&spa_namespace_lock); 3291} 3292 3293/* 3294 * Add spares device information to the nvlist. 3295 */ 3296static void 3297spa_add_spares(spa_t *spa, nvlist_t *config) 3298{ 3299 nvlist_t **spares; 3300 uint_t i, nspares; 3301 nvlist_t *nvroot; 3302 uint64_t guid; 3303 vdev_stat_t *vs; 3304 uint_t vsc; 3305 uint64_t pool; 3306 3307 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3308 3309 if (spa->spa_spares.sav_count == 0) 3310 return; 3311 3312 VERIFY(nvlist_lookup_nvlist(config, 3313 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3314 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3315 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3316 if (nspares != 0) { 3317 VERIFY(nvlist_add_nvlist_array(nvroot, 3318 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3319 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3320 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3321 3322 /* 3323 * Go through and find any spares which have since been 3324 * repurposed as an active spare. If this is the case, update 3325 * their status appropriately. 3326 */ 3327 for (i = 0; i < nspares; i++) { 3328 VERIFY(nvlist_lookup_uint64(spares[i], 3329 ZPOOL_CONFIG_GUID, &guid) == 0); 3330 if (spa_spare_exists(guid, &pool, NULL) && 3331 pool != 0ULL) { 3332 VERIFY(nvlist_lookup_uint64_array( 3333 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3334 (uint64_t **)&vs, &vsc) == 0); 3335 vs->vs_state = VDEV_STATE_CANT_OPEN; 3336 vs->vs_aux = VDEV_AUX_SPARED; 3337 } 3338 } 3339 } 3340} 3341 3342/* 3343 * Add l2cache device information to the nvlist, including vdev stats. 3344 */ 3345static void 3346spa_add_l2cache(spa_t *spa, nvlist_t *config) 3347{ 3348 nvlist_t **l2cache; 3349 uint_t i, j, nl2cache; 3350 nvlist_t *nvroot; 3351 uint64_t guid; 3352 vdev_t *vd; 3353 vdev_stat_t *vs; 3354 uint_t vsc; 3355 3356 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3357 3358 if (spa->spa_l2cache.sav_count == 0) 3359 return; 3360 3361 VERIFY(nvlist_lookup_nvlist(config, 3362 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3363 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3364 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3365 if (nl2cache != 0) { 3366 VERIFY(nvlist_add_nvlist_array(nvroot, 3367 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3368 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3369 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3370 3371 /* 3372 * Update level 2 cache device stats. 3373 */ 3374 3375 for (i = 0; i < nl2cache; i++) { 3376 VERIFY(nvlist_lookup_uint64(l2cache[i], 3377 ZPOOL_CONFIG_GUID, &guid) == 0); 3378 3379 vd = NULL; 3380 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3381 if (guid == 3382 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3383 vd = spa->spa_l2cache.sav_vdevs[j]; 3384 break; 3385 } 3386 } 3387 ASSERT(vd != NULL); 3388 3389 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3390 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3391 == 0); 3392 vdev_get_stats(vd, vs); 3393 } 3394 } 3395} 3396 3397static void 3398spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3399{ 3400 nvlist_t *features; 3401 zap_cursor_t zc; 3402 zap_attribute_t za; 3403 3404 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3405 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3406 3407 /* We may be unable to read features if pool is suspended. */ 3408 if (spa_suspended(spa)) 3409 goto out; 3410 3411 if (spa->spa_feat_for_read_obj != 0) { 3412 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3413 spa->spa_feat_for_read_obj); 3414 zap_cursor_retrieve(&zc, &za) == 0; 3415 zap_cursor_advance(&zc)) { 3416 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3417 za.za_num_integers == 1); 3418 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3419 za.za_first_integer)); 3420 } 3421 zap_cursor_fini(&zc); 3422 } 3423 3424 if (spa->spa_feat_for_write_obj != 0) { 3425 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3426 spa->spa_feat_for_write_obj); 3427 zap_cursor_retrieve(&zc, &za) == 0; 3428 zap_cursor_advance(&zc)) { 3429 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3430 za.za_num_integers == 1); 3431 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3432 za.za_first_integer)); 3433 } 3434 zap_cursor_fini(&zc); 3435 } 3436 3437out: 3438 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3439 features) == 0); 3440 nvlist_free(features); 3441} 3442 3443int 3444spa_get_stats(const char *name, nvlist_t **config, 3445 char *altroot, size_t buflen) 3446{ 3447 int error; 3448 spa_t *spa; 3449 3450 *config = NULL; 3451 error = spa_open_common(name, &spa, FTAG, NULL, config); 3452 3453 if (spa != NULL) { 3454 /* 3455 * This still leaves a window of inconsistency where the spares 3456 * or l2cache devices could change and the config would be 3457 * self-inconsistent. 3458 */ 3459 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3460 3461 if (*config != NULL) { 3462 uint64_t loadtimes[2]; 3463 3464 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3465 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3466 VERIFY(nvlist_add_uint64_array(*config, 3467 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3468 3469 VERIFY(nvlist_add_uint64(*config, 3470 ZPOOL_CONFIG_ERRCOUNT, 3471 spa_get_errlog_size(spa)) == 0); 3472 3473 if (spa_suspended(spa)) 3474 VERIFY(nvlist_add_uint64(*config, 3475 ZPOOL_CONFIG_SUSPENDED, 3476 spa->spa_failmode) == 0); 3477 3478 spa_add_spares(spa, *config); 3479 spa_add_l2cache(spa, *config); 3480 spa_add_feature_stats(spa, *config); 3481 } 3482 } 3483 3484 /* 3485 * We want to get the alternate root even for faulted pools, so we cheat 3486 * and call spa_lookup() directly. 3487 */ 3488 if (altroot) { 3489 if (spa == NULL) { 3490 mutex_enter(&spa_namespace_lock); 3491 spa = spa_lookup(name); 3492 if (spa) 3493 spa_altroot(spa, altroot, buflen); 3494 else 3495 altroot[0] = '\0'; 3496 spa = NULL; 3497 mutex_exit(&spa_namespace_lock); 3498 } else { 3499 spa_altroot(spa, altroot, buflen); 3500 } 3501 } 3502 3503 if (spa != NULL) { 3504 spa_config_exit(spa, SCL_CONFIG, FTAG); 3505 spa_close(spa, FTAG); 3506 } 3507 3508 return (error); 3509} 3510 3511/* 3512 * Validate that the auxiliary device array is well formed. We must have an 3513 * array of nvlists, each which describes a valid leaf vdev. If this is an 3514 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3515 * specified, as long as they are well-formed. 3516 */ 3517static int 3518spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3519 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3520 vdev_labeltype_t label) 3521{ 3522 nvlist_t **dev; 3523 uint_t i, ndev; 3524 vdev_t *vd; 3525 int error; 3526 3527 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3528 3529 /* 3530 * It's acceptable to have no devs specified. 3531 */ 3532 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3533 return (0); 3534 3535 if (ndev == 0) 3536 return (SET_ERROR(EINVAL)); 3537 3538 /* 3539 * Make sure the pool is formatted with a version that supports this 3540 * device type. 3541 */ 3542 if (spa_version(spa) < version) 3543 return (SET_ERROR(ENOTSUP)); 3544 3545 /* 3546 * Set the pending device list so we correctly handle device in-use 3547 * checking. 3548 */ 3549 sav->sav_pending = dev; 3550 sav->sav_npending = ndev; 3551 3552 for (i = 0; i < ndev; i++) { 3553 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3554 mode)) != 0) 3555 goto out; 3556 3557 if (!vd->vdev_ops->vdev_op_leaf) { 3558 vdev_free(vd); 3559 error = SET_ERROR(EINVAL); 3560 goto out; 3561 } 3562 3563 /* 3564 * The L2ARC currently only supports disk devices in 3565 * kernel context. For user-level testing, we allow it. 3566 */ 3567#ifdef _KERNEL 3568 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3569 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3570 error = SET_ERROR(ENOTBLK); 3571 vdev_free(vd); 3572 goto out; 3573 } 3574#endif 3575 vd->vdev_top = vd; 3576 3577 if ((error = vdev_open(vd)) == 0 && 3578 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3579 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3580 vd->vdev_guid) == 0); 3581 } 3582 3583 vdev_free(vd); 3584 3585 if (error && 3586 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3587 goto out; 3588 else 3589 error = 0; 3590 } 3591 3592out: 3593 sav->sav_pending = NULL; 3594 sav->sav_npending = 0; 3595 return (error); 3596} 3597 3598static int 3599spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3600{ 3601 int error; 3602 3603 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3604 3605 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3606 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3607 VDEV_LABEL_SPARE)) != 0) { 3608 return (error); 3609 } 3610 3611 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3612 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3613 VDEV_LABEL_L2CACHE)); 3614} 3615 3616static void 3617spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3618 const char *config) 3619{ 3620 int i; 3621 3622 if (sav->sav_config != NULL) { 3623 nvlist_t **olddevs; 3624 uint_t oldndevs; 3625 nvlist_t **newdevs; 3626 3627 /* 3628 * Generate new dev list by concatentating with the 3629 * current dev list. 3630 */ 3631 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3632 &olddevs, &oldndevs) == 0); 3633 3634 newdevs = kmem_alloc(sizeof (void *) * 3635 (ndevs + oldndevs), KM_SLEEP); 3636 for (i = 0; i < oldndevs; i++) 3637 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3638 KM_SLEEP) == 0); 3639 for (i = 0; i < ndevs; i++) 3640 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3641 KM_SLEEP) == 0); 3642 3643 VERIFY(nvlist_remove(sav->sav_config, config, 3644 DATA_TYPE_NVLIST_ARRAY) == 0); 3645 3646 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3647 config, newdevs, ndevs + oldndevs) == 0); 3648 for (i = 0; i < oldndevs + ndevs; i++) 3649 nvlist_free(newdevs[i]); 3650 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3651 } else { 3652 /* 3653 * Generate a new dev list. 3654 */ 3655 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3656 KM_SLEEP) == 0); 3657 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3658 devs, ndevs) == 0); 3659 } 3660} 3661 3662/* 3663 * Stop and drop level 2 ARC devices 3664 */ 3665void 3666spa_l2cache_drop(spa_t *spa) 3667{ 3668 vdev_t *vd; 3669 int i; 3670 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3671 3672 for (i = 0; i < sav->sav_count; i++) { 3673 uint64_t pool; 3674 3675 vd = sav->sav_vdevs[i]; 3676 ASSERT(vd != NULL); 3677 3678 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3679 pool != 0ULL && l2arc_vdev_present(vd)) 3680 l2arc_remove_vdev(vd); 3681 } 3682} 3683 3684/* 3685 * Pool Creation 3686 */ 3687int 3688spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3689 nvlist_t *zplprops) 3690{ 3691 spa_t *spa; 3692 char *altroot = NULL; 3693 vdev_t *rvd; 3694 dsl_pool_t *dp; 3695 dmu_tx_t *tx; 3696 int error = 0; 3697 uint64_t txg = TXG_INITIAL; 3698 nvlist_t **spares, **l2cache; 3699 uint_t nspares, nl2cache; 3700 uint64_t version, obj; 3701 boolean_t has_features; 3702 3703 /* 3704 * If this pool already exists, return failure. 3705 */ 3706 mutex_enter(&spa_namespace_lock); 3707 if (spa_lookup(pool) != NULL) { 3708 mutex_exit(&spa_namespace_lock); 3709 return (SET_ERROR(EEXIST)); 3710 } 3711 3712 /* 3713 * Allocate a new spa_t structure. 3714 */ 3715 (void) nvlist_lookup_string(props, 3716 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3717 spa = spa_add(pool, NULL, altroot); 3718 spa_activate(spa, spa_mode_global); 3719 3720 if (props && (error = spa_prop_validate(spa, props))) { 3721 spa_deactivate(spa); 3722 spa_remove(spa); 3723 mutex_exit(&spa_namespace_lock); 3724 return (error); 3725 } 3726 3727 has_features = B_FALSE; 3728 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3729 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3730 if (zpool_prop_feature(nvpair_name(elem))) 3731 has_features = B_TRUE; 3732 } 3733 3734 if (has_features || nvlist_lookup_uint64(props, 3735 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3736 version = SPA_VERSION; 3737 } 3738 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3739 3740 spa->spa_first_txg = txg; 3741 spa->spa_uberblock.ub_txg = txg - 1; 3742 spa->spa_uberblock.ub_version = version; 3743 spa->spa_ubsync = spa->spa_uberblock; 3744 spa->spa_load_state = SPA_LOAD_CREATE; 3745 3746 /* 3747 * Create "The Godfather" zio to hold all async IOs 3748 */ 3749 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3750 KM_SLEEP); 3751 for (int i = 0; i < max_ncpus; i++) { 3752 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3753 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3754 ZIO_FLAG_GODFATHER); 3755 } 3756 3757 /* 3758 * Create the root vdev. 3759 */ 3760 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3761 3762 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3763 3764 ASSERT(error != 0 || rvd != NULL); 3765 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3766 3767 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3768 error = SET_ERROR(EINVAL); 3769 3770 if (error == 0 && 3771 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3772 (error = spa_validate_aux(spa, nvroot, txg, 3773 VDEV_ALLOC_ADD)) == 0) { 3774 for (int c = 0; c < rvd->vdev_children; c++) { 3775 vdev_ashift_optimize(rvd->vdev_child[c]); 3776 vdev_metaslab_set_size(rvd->vdev_child[c]); 3777 vdev_expand(rvd->vdev_child[c], txg); 3778 } 3779 } 3780 3781 spa_config_exit(spa, SCL_ALL, FTAG); 3782 3783 if (error != 0) { 3784 spa_unload(spa); 3785 spa_deactivate(spa); 3786 spa_remove(spa); 3787 mutex_exit(&spa_namespace_lock); 3788 return (error); 3789 } 3790 3791 /* 3792 * Get the list of spares, if specified. 3793 */ 3794 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3795 &spares, &nspares) == 0) { 3796 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3797 KM_SLEEP) == 0); 3798 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3799 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3800 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3801 spa_load_spares(spa); 3802 spa_config_exit(spa, SCL_ALL, FTAG); 3803 spa->spa_spares.sav_sync = B_TRUE; 3804 } 3805 3806 /* 3807 * Get the list of level 2 cache devices, if specified. 3808 */ 3809 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3810 &l2cache, &nl2cache) == 0) { 3811 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3812 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3813 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3814 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3815 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3816 spa_load_l2cache(spa); 3817 spa_config_exit(spa, SCL_ALL, FTAG); 3818 spa->spa_l2cache.sav_sync = B_TRUE; 3819 } 3820 3821 spa->spa_is_initializing = B_TRUE; 3822 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3823 spa->spa_meta_objset = dp->dp_meta_objset; 3824 spa->spa_is_initializing = B_FALSE; 3825 3826 /* 3827 * Create DDTs (dedup tables). 3828 */ 3829 ddt_create(spa); 3830 3831 spa_update_dspace(spa); 3832 3833 tx = dmu_tx_create_assigned(dp, txg); 3834 3835 /* 3836 * Create the pool config object. 3837 */ 3838 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3839 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3840 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3841 3842 if (zap_add(spa->spa_meta_objset, 3843 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3844 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3845 cmn_err(CE_PANIC, "failed to add pool config"); 3846 } 3847 3848 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3849 spa_feature_create_zap_objects(spa, tx); 3850 3851 if (zap_add(spa->spa_meta_objset, 3852 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3853 sizeof (uint64_t), 1, &version, tx) != 0) { 3854 cmn_err(CE_PANIC, "failed to add pool version"); 3855 } 3856 3857 /* Newly created pools with the right version are always deflated. */ 3858 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3859 spa->spa_deflate = TRUE; 3860 if (zap_add(spa->spa_meta_objset, 3861 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3862 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3863 cmn_err(CE_PANIC, "failed to add deflate"); 3864 } 3865 } 3866 3867 /* 3868 * Create the deferred-free bpobj. Turn off compression 3869 * because sync-to-convergence takes longer if the blocksize 3870 * keeps changing. 3871 */ 3872 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3873 dmu_object_set_compress(spa->spa_meta_objset, obj, 3874 ZIO_COMPRESS_OFF, tx); 3875 if (zap_add(spa->spa_meta_objset, 3876 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3877 sizeof (uint64_t), 1, &obj, tx) != 0) { 3878 cmn_err(CE_PANIC, "failed to add bpobj"); 3879 } 3880 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3881 spa->spa_meta_objset, obj)); 3882 3883 /* 3884 * Create the pool's history object. 3885 */ 3886 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3887 spa_history_create_obj(spa, tx); 3888 3889 /* 3890 * Generate some random noise for salted checksums to operate on. 3891 */ 3892 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3893 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3894 3895 /* 3896 * Set pool properties. 3897 */ 3898 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3899 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3900 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3901 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3902 3903 if (props != NULL) { 3904 spa_configfile_set(spa, props, B_FALSE); 3905 spa_sync_props(props, tx); 3906 } 3907 3908 dmu_tx_commit(tx); 3909 3910 spa->spa_sync_on = B_TRUE; 3911 txg_sync_start(spa->spa_dsl_pool); 3912 3913 /* 3914 * We explicitly wait for the first transaction to complete so that our 3915 * bean counters are appropriately updated. 3916 */ 3917 txg_wait_synced(spa->spa_dsl_pool, txg); 3918 3919 spa_config_sync(spa, B_FALSE, B_TRUE); 3920 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3921 3922 spa_history_log_version(spa, "create"); 3923 3924 /* 3925 * Don't count references from objsets that are already closed 3926 * and are making their way through the eviction process. 3927 */ 3928 spa_evicting_os_wait(spa); 3929 spa->spa_minref = refcount_count(&spa->spa_refcount); 3930 spa->spa_load_state = SPA_LOAD_NONE; 3931 3932 mutex_exit(&spa_namespace_lock); 3933 3934 return (0); 3935} 3936 3937#ifdef _KERNEL 3938#ifdef illumos 3939/* 3940 * Get the root pool information from the root disk, then import the root pool 3941 * during the system boot up time. 3942 */ 3943extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3944 3945static nvlist_t * 3946spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3947{ 3948 nvlist_t *config; 3949 nvlist_t *nvtop, *nvroot; 3950 uint64_t pgid; 3951 3952 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3953 return (NULL); 3954 3955 /* 3956 * Add this top-level vdev to the child array. 3957 */ 3958 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3959 &nvtop) == 0); 3960 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3961 &pgid) == 0); 3962 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3963 3964 /* 3965 * Put this pool's top-level vdevs into a root vdev. 3966 */ 3967 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3968 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3969 VDEV_TYPE_ROOT) == 0); 3970 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3971 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3972 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3973 &nvtop, 1) == 0); 3974 3975 /* 3976 * Replace the existing vdev_tree with the new root vdev in 3977 * this pool's configuration (remove the old, add the new). 3978 */ 3979 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3980 nvlist_free(nvroot); 3981 return (config); 3982} 3983 3984/* 3985 * Walk the vdev tree and see if we can find a device with "better" 3986 * configuration. A configuration is "better" if the label on that 3987 * device has a more recent txg. 3988 */ 3989static void 3990spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3991{ 3992 for (int c = 0; c < vd->vdev_children; c++) 3993 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3994 3995 if (vd->vdev_ops->vdev_op_leaf) { 3996 nvlist_t *label; 3997 uint64_t label_txg; 3998 3999 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 4000 &label) != 0) 4001 return; 4002 4003 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 4004 &label_txg) == 0); 4005 4006 /* 4007 * Do we have a better boot device? 4008 */ 4009 if (label_txg > *txg) { 4010 *txg = label_txg; 4011 *avd = vd; 4012 } 4013 nvlist_free(label); 4014 } 4015} 4016 4017/* 4018 * Import a root pool. 4019 * 4020 * For x86. devpath_list will consist of devid and/or physpath name of 4021 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 4022 * The GRUB "findroot" command will return the vdev we should boot. 4023 * 4024 * For Sparc, devpath_list consists the physpath name of the booting device 4025 * no matter the rootpool is a single device pool or a mirrored pool. 4026 * e.g. 4027 * "/pci@1f,0/ide@d/disk@0,0:a" 4028 */ 4029int 4030spa_import_rootpool(char *devpath, char *devid) 4031{ 4032 spa_t *spa; 4033 vdev_t *rvd, *bvd, *avd = NULL; 4034 nvlist_t *config, *nvtop; 4035 uint64_t guid, txg; 4036 char *pname; 4037 int error; 4038 4039 /* 4040 * Read the label from the boot device and generate a configuration. 4041 */ 4042 config = spa_generate_rootconf(devpath, devid, &guid); 4043#if defined(_OBP) && defined(_KERNEL) 4044 if (config == NULL) { 4045 if (strstr(devpath, "/iscsi/ssd") != NULL) { 4046 /* iscsi boot */ 4047 get_iscsi_bootpath_phy(devpath); 4048 config = spa_generate_rootconf(devpath, devid, &guid); 4049 } 4050 } 4051#endif 4052 if (config == NULL) { 4053 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 4054 devpath); 4055 return (SET_ERROR(EIO)); 4056 } 4057 4058 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4059 &pname) == 0); 4060 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 4061 4062 mutex_enter(&spa_namespace_lock); 4063 if ((spa = spa_lookup(pname)) != NULL) { 4064 /* 4065 * Remove the existing root pool from the namespace so that we 4066 * can replace it with the correct config we just read in. 4067 */ 4068 spa_remove(spa); 4069 } 4070 4071 spa = spa_add(pname, config, NULL); 4072 spa->spa_is_root = B_TRUE; 4073 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4074 4075 /* 4076 * Build up a vdev tree based on the boot device's label config. 4077 */ 4078 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4079 &nvtop) == 0); 4080 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4081 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4082 VDEV_ALLOC_ROOTPOOL); 4083 spa_config_exit(spa, SCL_ALL, FTAG); 4084 if (error) { 4085 mutex_exit(&spa_namespace_lock); 4086 nvlist_free(config); 4087 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4088 pname); 4089 return (error); 4090 } 4091 4092 /* 4093 * Get the boot vdev. 4094 */ 4095 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 4096 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 4097 (u_longlong_t)guid); 4098 error = SET_ERROR(ENOENT); 4099 goto out; 4100 } 4101 4102 /* 4103 * Determine if there is a better boot device. 4104 */ 4105 avd = bvd; 4106 spa_alt_rootvdev(rvd, &avd, &txg); 4107 if (avd != bvd) { 4108 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 4109 "try booting from '%s'", avd->vdev_path); 4110 error = SET_ERROR(EINVAL); 4111 goto out; 4112 } 4113 4114 /* 4115 * If the boot device is part of a spare vdev then ensure that 4116 * we're booting off the active spare. 4117 */ 4118 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 4119 !bvd->vdev_isspare) { 4120 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 4121 "try booting from '%s'", 4122 bvd->vdev_parent-> 4123 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4124 error = SET_ERROR(EINVAL); 4125 goto out; 4126 } 4127 4128 error = 0; 4129out: 4130 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4131 vdev_free(rvd); 4132 spa_config_exit(spa, SCL_ALL, FTAG); 4133 mutex_exit(&spa_namespace_lock); 4134 4135 nvlist_free(config); 4136 return (error); 4137} 4138 4139#else /* !illumos */ 4140 4141extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 4142 uint64_t *count); 4143 4144static nvlist_t * 4145spa_generate_rootconf(const char *name) 4146{ 4147 nvlist_t **configs, **tops; 4148 nvlist_t *config; 4149 nvlist_t *best_cfg, *nvtop, *nvroot; 4150 uint64_t *holes; 4151 uint64_t best_txg; 4152 uint64_t nchildren; 4153 uint64_t pgid; 4154 uint64_t count; 4155 uint64_t i; 4156 uint_t nholes; 4157 4158 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4159 return (NULL); 4160 4161 ASSERT3U(count, !=, 0); 4162 best_txg = 0; 4163 for (i = 0; i < count; i++) { 4164 uint64_t txg; 4165 4166 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4167 &txg) == 0); 4168 if (txg > best_txg) { 4169 best_txg = txg; 4170 best_cfg = configs[i]; 4171 } 4172 } 4173 4174 nchildren = 1; 4175 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4176 holes = NULL; 4177 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4178 &holes, &nholes); 4179 4180 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4181 for (i = 0; i < nchildren; i++) { 4182 if (i >= count) 4183 break; 4184 if (configs[i] == NULL) 4185 continue; 4186 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4187 &nvtop) == 0); 4188 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4189 } 4190 for (i = 0; holes != NULL && i < nholes; i++) { 4191 if (i >= nchildren) 4192 continue; 4193 if (tops[holes[i]] != NULL) 4194 continue; 4195 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4196 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4197 VDEV_TYPE_HOLE) == 0); 4198 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4199 holes[i]) == 0); 4200 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4201 0) == 0); 4202 } 4203 for (i = 0; i < nchildren; i++) { 4204 if (tops[i] != NULL) 4205 continue; 4206 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4207 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4208 VDEV_TYPE_MISSING) == 0); 4209 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4210 i) == 0); 4211 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4212 0) == 0); 4213 } 4214 4215 /* 4216 * Create pool config based on the best vdev config. 4217 */ 4218 nvlist_dup(best_cfg, &config, KM_SLEEP); 4219 4220 /* 4221 * Put this pool's top-level vdevs into a root vdev. 4222 */ 4223 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4224 &pgid) == 0); 4225 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4226 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4227 VDEV_TYPE_ROOT) == 0); 4228 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4229 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4230 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4231 tops, nchildren) == 0); 4232 4233 /* 4234 * Replace the existing vdev_tree with the new root vdev in 4235 * this pool's configuration (remove the old, add the new). 4236 */ 4237 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4238 4239 /* 4240 * Drop vdev config elements that should not be present at pool level. 4241 */ 4242 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4243 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4244 4245 for (i = 0; i < count; i++) 4246 nvlist_free(configs[i]); 4247 kmem_free(configs, count * sizeof(void *)); 4248 for (i = 0; i < nchildren; i++) 4249 nvlist_free(tops[i]); 4250 kmem_free(tops, nchildren * sizeof(void *)); 4251 nvlist_free(nvroot); 4252 return (config); 4253} 4254 4255int 4256spa_import_rootpool(const char *name) 4257{ 4258 spa_t *spa; 4259 vdev_t *rvd, *bvd, *avd = NULL; 4260 nvlist_t *config, *nvtop; 4261 uint64_t txg; 4262 char *pname; 4263 int error; 4264 4265 /* 4266 * Read the label from the boot device and generate a configuration. 4267 */ 4268 config = spa_generate_rootconf(name); 4269 4270 mutex_enter(&spa_namespace_lock); 4271 if (config != NULL) { 4272 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4273 &pname) == 0 && strcmp(name, pname) == 0); 4274 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4275 == 0); 4276 4277 if ((spa = spa_lookup(pname)) != NULL) { 4278 /* 4279 * The pool could already be imported, 4280 * e.g., after reboot -r. 4281 */ 4282 if (spa->spa_state == POOL_STATE_ACTIVE) { 4283 mutex_exit(&spa_namespace_lock); 4284 nvlist_free(config); 4285 return (0); 4286 } 4287 4288 /* 4289 * Remove the existing root pool from the namespace so 4290 * that we can replace it with the correct config 4291 * we just read in. 4292 */ 4293 spa_remove(spa); 4294 } 4295 spa = spa_add(pname, config, NULL); 4296 4297 /* 4298 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4299 * via spa_version(). 4300 */ 4301 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4302 &spa->spa_ubsync.ub_version) != 0) 4303 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4304 } else if ((spa = spa_lookup(name)) == NULL) { 4305 mutex_exit(&spa_namespace_lock); 4306 nvlist_free(config); 4307 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4308 name); 4309 return (EIO); 4310 } else { 4311 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4312 } 4313 spa->spa_is_root = B_TRUE; 4314 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4315 4316 /* 4317 * Build up a vdev tree based on the boot device's label config. 4318 */ 4319 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4320 &nvtop) == 0); 4321 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4322 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4323 VDEV_ALLOC_ROOTPOOL); 4324 spa_config_exit(spa, SCL_ALL, FTAG); 4325 if (error) { 4326 mutex_exit(&spa_namespace_lock); 4327 nvlist_free(config); 4328 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4329 pname); 4330 return (error); 4331 } 4332 4333 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4334 vdev_free(rvd); 4335 spa_config_exit(spa, SCL_ALL, FTAG); 4336 mutex_exit(&spa_namespace_lock); 4337 4338 nvlist_free(config); 4339 return (0); 4340} 4341 4342#endif /* illumos */ 4343#endif /* _KERNEL */ 4344 4345/* 4346 * Import a non-root pool into the system. 4347 */ 4348int 4349spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4350{ 4351 spa_t *spa; 4352 char *altroot = NULL; 4353 spa_load_state_t state = SPA_LOAD_IMPORT; 4354 zpool_rewind_policy_t policy; 4355 uint64_t mode = spa_mode_global; 4356 uint64_t readonly = B_FALSE; 4357 int error; 4358 nvlist_t *nvroot; 4359 nvlist_t **spares, **l2cache; 4360 uint_t nspares, nl2cache; 4361 4362 /* 4363 * If a pool with this name exists, return failure. 4364 */ 4365 mutex_enter(&spa_namespace_lock); 4366 if (spa_lookup(pool) != NULL) { 4367 mutex_exit(&spa_namespace_lock); 4368 return (SET_ERROR(EEXIST)); 4369 } 4370 4371 /* 4372 * Create and initialize the spa structure. 4373 */ 4374 (void) nvlist_lookup_string(props, 4375 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4376 (void) nvlist_lookup_uint64(props, 4377 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4378 if (readonly) 4379 mode = FREAD; 4380 spa = spa_add(pool, config, altroot); 4381 spa->spa_import_flags = flags; 4382 4383 /* 4384 * Verbatim import - Take a pool and insert it into the namespace 4385 * as if it had been loaded at boot. 4386 */ 4387 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4388 if (props != NULL) 4389 spa_configfile_set(spa, props, B_FALSE); 4390 4391 spa_config_sync(spa, B_FALSE, B_TRUE); 4392 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4393 4394 mutex_exit(&spa_namespace_lock); 4395 return (0); 4396 } 4397 4398 spa_activate(spa, mode); 4399 4400 /* 4401 * Don't start async tasks until we know everything is healthy. 4402 */ 4403 spa_async_suspend(spa); 4404 4405 zpool_get_rewind_policy(config, &policy); 4406 if (policy.zrp_request & ZPOOL_DO_REWIND) 4407 state = SPA_LOAD_RECOVER; 4408 4409 /* 4410 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4411 * because the user-supplied config is actually the one to trust when 4412 * doing an import. 4413 */ 4414 if (state != SPA_LOAD_RECOVER) 4415 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4416 4417 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4418 policy.zrp_request); 4419 4420 /* 4421 * Propagate anything learned while loading the pool and pass it 4422 * back to caller (i.e. rewind info, missing devices, etc). 4423 */ 4424 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4425 spa->spa_load_info) == 0); 4426 4427 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4428 /* 4429 * Toss any existing sparelist, as it doesn't have any validity 4430 * anymore, and conflicts with spa_has_spare(). 4431 */ 4432 if (spa->spa_spares.sav_config) { 4433 nvlist_free(spa->spa_spares.sav_config); 4434 spa->spa_spares.sav_config = NULL; 4435 spa_load_spares(spa); 4436 } 4437 if (spa->spa_l2cache.sav_config) { 4438 nvlist_free(spa->spa_l2cache.sav_config); 4439 spa->spa_l2cache.sav_config = NULL; 4440 spa_load_l2cache(spa); 4441 } 4442 4443 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4444 &nvroot) == 0); 4445 if (error == 0) 4446 error = spa_validate_aux(spa, nvroot, -1ULL, 4447 VDEV_ALLOC_SPARE); 4448 if (error == 0) 4449 error = spa_validate_aux(spa, nvroot, -1ULL, 4450 VDEV_ALLOC_L2CACHE); 4451 spa_config_exit(spa, SCL_ALL, FTAG); 4452 4453 if (props != NULL) 4454 spa_configfile_set(spa, props, B_FALSE); 4455 4456 if (error != 0 || (props && spa_writeable(spa) && 4457 (error = spa_prop_set(spa, props)))) { 4458 spa_unload(spa); 4459 spa_deactivate(spa); 4460 spa_remove(spa); 4461 mutex_exit(&spa_namespace_lock); 4462 return (error); 4463 } 4464 4465 spa_async_resume(spa); 4466 4467 /* 4468 * Override any spares and level 2 cache devices as specified by 4469 * the user, as these may have correct device names/devids, etc. 4470 */ 4471 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4472 &spares, &nspares) == 0) { 4473 if (spa->spa_spares.sav_config) 4474 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4475 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4476 else 4477 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4478 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4479 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4480 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4481 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4482 spa_load_spares(spa); 4483 spa_config_exit(spa, SCL_ALL, FTAG); 4484 spa->spa_spares.sav_sync = B_TRUE; 4485 } 4486 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4487 &l2cache, &nl2cache) == 0) { 4488 if (spa->spa_l2cache.sav_config) 4489 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4490 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4491 else 4492 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4493 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4494 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4495 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4496 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4497 spa_load_l2cache(spa); 4498 spa_config_exit(spa, SCL_ALL, FTAG); 4499 spa->spa_l2cache.sav_sync = B_TRUE; 4500 } 4501 4502 /* 4503 * Check for any removed devices. 4504 */ 4505 if (spa->spa_autoreplace) { 4506 spa_aux_check_removed(&spa->spa_spares); 4507 spa_aux_check_removed(&spa->spa_l2cache); 4508 } 4509 4510 if (spa_writeable(spa)) { 4511 /* 4512 * Update the config cache to include the newly-imported pool. 4513 */ 4514 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4515 } 4516 4517 /* 4518 * It's possible that the pool was expanded while it was exported. 4519 * We kick off an async task to handle this for us. 4520 */ 4521 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4522 4523 spa_history_log_version(spa, "import"); 4524 4525 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4526 4527 mutex_exit(&spa_namespace_lock); 4528 4529#ifdef __FreeBSD__ 4530#ifdef _KERNEL 4531 zvol_create_minors(pool); 4532#endif 4533#endif 4534 return (0); 4535} 4536 4537nvlist_t * 4538spa_tryimport(nvlist_t *tryconfig) 4539{ 4540 nvlist_t *config = NULL; 4541 char *poolname; 4542 spa_t *spa; 4543 uint64_t state; 4544 int error; 4545 4546 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4547 return (NULL); 4548 4549 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4550 return (NULL); 4551 4552 /* 4553 * Create and initialize the spa structure. 4554 */ 4555 mutex_enter(&spa_namespace_lock); 4556 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4557 spa_activate(spa, FREAD); 4558 4559 /* 4560 * Pass off the heavy lifting to spa_load(). 4561 * Pass TRUE for mosconfig because the user-supplied config 4562 * is actually the one to trust when doing an import. 4563 */ 4564 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4565 4566 /* 4567 * If 'tryconfig' was at least parsable, return the current config. 4568 */ 4569 if (spa->spa_root_vdev != NULL) { 4570 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4571 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4572 poolname) == 0); 4573 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4574 state) == 0); 4575 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4576 spa->spa_uberblock.ub_timestamp) == 0); 4577 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4578 spa->spa_load_info) == 0); 4579 4580 /* 4581 * If the bootfs property exists on this pool then we 4582 * copy it out so that external consumers can tell which 4583 * pools are bootable. 4584 */ 4585 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4586 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4587 4588 /* 4589 * We have to play games with the name since the 4590 * pool was opened as TRYIMPORT_NAME. 4591 */ 4592 if (dsl_dsobj_to_dsname(spa_name(spa), 4593 spa->spa_bootfs, tmpname) == 0) { 4594 char *cp; 4595 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4596 4597 cp = strchr(tmpname, '/'); 4598 if (cp == NULL) { 4599 (void) strlcpy(dsname, tmpname, 4600 MAXPATHLEN); 4601 } else { 4602 (void) snprintf(dsname, MAXPATHLEN, 4603 "%s/%s", poolname, ++cp); 4604 } 4605 VERIFY(nvlist_add_string(config, 4606 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4607 kmem_free(dsname, MAXPATHLEN); 4608 } 4609 kmem_free(tmpname, MAXPATHLEN); 4610 } 4611 4612 /* 4613 * Add the list of hot spares and level 2 cache devices. 4614 */ 4615 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4616 spa_add_spares(spa, config); 4617 spa_add_l2cache(spa, config); 4618 spa_config_exit(spa, SCL_CONFIG, FTAG); 4619 } 4620 4621 spa_unload(spa); 4622 spa_deactivate(spa); 4623 spa_remove(spa); 4624 mutex_exit(&spa_namespace_lock); 4625 4626 return (config); 4627} 4628 4629/* 4630 * Pool export/destroy 4631 * 4632 * The act of destroying or exporting a pool is very simple. We make sure there 4633 * is no more pending I/O and any references to the pool are gone. Then, we 4634 * update the pool state and sync all the labels to disk, removing the 4635 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4636 * we don't sync the labels or remove the configuration cache. 4637 */ 4638static int 4639spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4640 boolean_t force, boolean_t hardforce) 4641{ 4642 spa_t *spa; 4643 4644 if (oldconfig) 4645 *oldconfig = NULL; 4646 4647 if (!(spa_mode_global & FWRITE)) 4648 return (SET_ERROR(EROFS)); 4649 4650 mutex_enter(&spa_namespace_lock); 4651 if ((spa = spa_lookup(pool)) == NULL) { 4652 mutex_exit(&spa_namespace_lock); 4653 return (SET_ERROR(ENOENT)); 4654 } 4655 4656 /* 4657 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4658 * reacquire the namespace lock, and see if we can export. 4659 */ 4660 spa_open_ref(spa, FTAG); 4661 mutex_exit(&spa_namespace_lock); 4662 spa_async_suspend(spa); 4663 mutex_enter(&spa_namespace_lock); 4664 spa_close(spa, FTAG); 4665 4666 /* 4667 * The pool will be in core if it's openable, 4668 * in which case we can modify its state. 4669 */ 4670 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4671 /* 4672 * Objsets may be open only because they're dirty, so we 4673 * have to force it to sync before checking spa_refcnt. 4674 */ 4675 txg_wait_synced(spa->spa_dsl_pool, 0); 4676 spa_evicting_os_wait(spa); 4677 4678 /* 4679 * A pool cannot be exported or destroyed if there are active 4680 * references. If we are resetting a pool, allow references by 4681 * fault injection handlers. 4682 */ 4683 if (!spa_refcount_zero(spa) || 4684 (spa->spa_inject_ref != 0 && 4685 new_state != POOL_STATE_UNINITIALIZED)) { 4686 spa_async_resume(spa); 4687 mutex_exit(&spa_namespace_lock); 4688 return (SET_ERROR(EBUSY)); 4689 } 4690 4691 /* 4692 * A pool cannot be exported if it has an active shared spare. 4693 * This is to prevent other pools stealing the active spare 4694 * from an exported pool. At user's own will, such pool can 4695 * be forcedly exported. 4696 */ 4697 if (!force && new_state == POOL_STATE_EXPORTED && 4698 spa_has_active_shared_spare(spa)) { 4699 spa_async_resume(spa); 4700 mutex_exit(&spa_namespace_lock); 4701 return (SET_ERROR(EXDEV)); 4702 } 4703 4704 /* 4705 * We want this to be reflected on every label, 4706 * so mark them all dirty. spa_unload() will do the 4707 * final sync that pushes these changes out. 4708 */ 4709 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4710 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4711 spa->spa_state = new_state; 4712 spa->spa_final_txg = spa_last_synced_txg(spa) + 4713 TXG_DEFER_SIZE + 1; 4714 vdev_config_dirty(spa->spa_root_vdev); 4715 spa_config_exit(spa, SCL_ALL, FTAG); 4716 } 4717 } 4718 4719 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4720 4721 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4722 spa_unload(spa); 4723 spa_deactivate(spa); 4724 } 4725 4726 if (oldconfig && spa->spa_config) 4727 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4728 4729 if (new_state != POOL_STATE_UNINITIALIZED) { 4730 if (!hardforce) 4731 spa_config_sync(spa, B_TRUE, B_TRUE); 4732 spa_remove(spa); 4733 } 4734 mutex_exit(&spa_namespace_lock); 4735 4736 return (0); 4737} 4738 4739/* 4740 * Destroy a storage pool. 4741 */ 4742int 4743spa_destroy(char *pool) 4744{ 4745 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4746 B_FALSE, B_FALSE)); 4747} 4748 4749/* 4750 * Export a storage pool. 4751 */ 4752int 4753spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4754 boolean_t hardforce) 4755{ 4756 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4757 force, hardforce)); 4758} 4759 4760/* 4761 * Similar to spa_export(), this unloads the spa_t without actually removing it 4762 * from the namespace in any way. 4763 */ 4764int 4765spa_reset(char *pool) 4766{ 4767 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4768 B_FALSE, B_FALSE)); 4769} 4770 4771/* 4772 * ========================================================================== 4773 * Device manipulation 4774 * ========================================================================== 4775 */ 4776 4777/* 4778 * Add a device to a storage pool. 4779 */ 4780int 4781spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4782{ 4783 uint64_t txg, id; 4784 int error; 4785 vdev_t *rvd = spa->spa_root_vdev; 4786 vdev_t *vd, *tvd; 4787 nvlist_t **spares, **l2cache; 4788 uint_t nspares, nl2cache; 4789 4790 ASSERT(spa_writeable(spa)); 4791 4792 txg = spa_vdev_enter(spa); 4793 4794 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4795 VDEV_ALLOC_ADD)) != 0) 4796 return (spa_vdev_exit(spa, NULL, txg, error)); 4797 4798 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4799 4800 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4801 &nspares) != 0) 4802 nspares = 0; 4803 4804 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4805 &nl2cache) != 0) 4806 nl2cache = 0; 4807 4808 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4809 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4810 4811 if (vd->vdev_children != 0 && 4812 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4813 return (spa_vdev_exit(spa, vd, txg, error)); 4814 4815 /* 4816 * We must validate the spares and l2cache devices after checking the 4817 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4818 */ 4819 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4820 return (spa_vdev_exit(spa, vd, txg, error)); 4821 4822 /* 4823 * Transfer each new top-level vdev from vd to rvd. 4824 */ 4825 for (int c = 0; c < vd->vdev_children; c++) { 4826 4827 /* 4828 * Set the vdev id to the first hole, if one exists. 4829 */ 4830 for (id = 0; id < rvd->vdev_children; id++) { 4831 if (rvd->vdev_child[id]->vdev_ishole) { 4832 vdev_free(rvd->vdev_child[id]); 4833 break; 4834 } 4835 } 4836 tvd = vd->vdev_child[c]; 4837 vdev_remove_child(vd, tvd); 4838 tvd->vdev_id = id; 4839 vdev_add_child(rvd, tvd); 4840 vdev_config_dirty(tvd); 4841 } 4842 4843 if (nspares != 0) { 4844 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4845 ZPOOL_CONFIG_SPARES); 4846 spa_load_spares(spa); 4847 spa->spa_spares.sav_sync = B_TRUE; 4848 } 4849 4850 if (nl2cache != 0) { 4851 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4852 ZPOOL_CONFIG_L2CACHE); 4853 spa_load_l2cache(spa); 4854 spa->spa_l2cache.sav_sync = B_TRUE; 4855 } 4856 4857 /* 4858 * We have to be careful when adding new vdevs to an existing pool. 4859 * If other threads start allocating from these vdevs before we 4860 * sync the config cache, and we lose power, then upon reboot we may 4861 * fail to open the pool because there are DVAs that the config cache 4862 * can't translate. Therefore, we first add the vdevs without 4863 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4864 * and then let spa_config_update() initialize the new metaslabs. 4865 * 4866 * spa_load() checks for added-but-not-initialized vdevs, so that 4867 * if we lose power at any point in this sequence, the remaining 4868 * steps will be completed the next time we load the pool. 4869 */ 4870 (void) spa_vdev_exit(spa, vd, txg, 0); 4871 4872 mutex_enter(&spa_namespace_lock); 4873 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4874 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 4875 mutex_exit(&spa_namespace_lock); 4876 4877 return (0); 4878} 4879 4880/* 4881 * Attach a device to a mirror. The arguments are the path to any device 4882 * in the mirror, and the nvroot for the new device. If the path specifies 4883 * a device that is not mirrored, we automatically insert the mirror vdev. 4884 * 4885 * If 'replacing' is specified, the new device is intended to replace the 4886 * existing device; in this case the two devices are made into their own 4887 * mirror using the 'replacing' vdev, which is functionally identical to 4888 * the mirror vdev (it actually reuses all the same ops) but has a few 4889 * extra rules: you can't attach to it after it's been created, and upon 4890 * completion of resilvering, the first disk (the one being replaced) 4891 * is automatically detached. 4892 */ 4893int 4894spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4895{ 4896 uint64_t txg, dtl_max_txg; 4897 vdev_t *rvd = spa->spa_root_vdev; 4898 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4899 vdev_ops_t *pvops; 4900 char *oldvdpath, *newvdpath; 4901 int newvd_isspare; 4902 int error; 4903 4904 ASSERT(spa_writeable(spa)); 4905 4906 txg = spa_vdev_enter(spa); 4907 4908 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4909 4910 if (oldvd == NULL) 4911 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4912 4913 if (!oldvd->vdev_ops->vdev_op_leaf) 4914 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4915 4916 pvd = oldvd->vdev_parent; 4917 4918 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4919 VDEV_ALLOC_ATTACH)) != 0) 4920 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4921 4922 if (newrootvd->vdev_children != 1) 4923 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4924 4925 newvd = newrootvd->vdev_child[0]; 4926 4927 if (!newvd->vdev_ops->vdev_op_leaf) 4928 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4929 4930 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4931 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4932 4933 /* 4934 * Spares can't replace logs 4935 */ 4936 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4937 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4938 4939 if (!replacing) { 4940 /* 4941 * For attach, the only allowable parent is a mirror or the root 4942 * vdev. 4943 */ 4944 if (pvd->vdev_ops != &vdev_mirror_ops && 4945 pvd->vdev_ops != &vdev_root_ops) 4946 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4947 4948 pvops = &vdev_mirror_ops; 4949 } else { 4950 /* 4951 * Active hot spares can only be replaced by inactive hot 4952 * spares. 4953 */ 4954 if (pvd->vdev_ops == &vdev_spare_ops && 4955 oldvd->vdev_isspare && 4956 !spa_has_spare(spa, newvd->vdev_guid)) 4957 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4958 4959 /* 4960 * If the source is a hot spare, and the parent isn't already a 4961 * spare, then we want to create a new hot spare. Otherwise, we 4962 * want to create a replacing vdev. The user is not allowed to 4963 * attach to a spared vdev child unless the 'isspare' state is 4964 * the same (spare replaces spare, non-spare replaces 4965 * non-spare). 4966 */ 4967 if (pvd->vdev_ops == &vdev_replacing_ops && 4968 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4969 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4970 } else if (pvd->vdev_ops == &vdev_spare_ops && 4971 newvd->vdev_isspare != oldvd->vdev_isspare) { 4972 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4973 } 4974 4975 if (newvd->vdev_isspare) 4976 pvops = &vdev_spare_ops; 4977 else 4978 pvops = &vdev_replacing_ops; 4979 } 4980 4981 /* 4982 * Make sure the new device is big enough. 4983 */ 4984 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4985 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4986 4987 /* 4988 * The new device cannot have a higher alignment requirement 4989 * than the top-level vdev. 4990 */ 4991 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4992 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4993 4994 /* 4995 * If this is an in-place replacement, update oldvd's path and devid 4996 * to make it distinguishable from newvd, and unopenable from now on. 4997 */ 4998 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4999 spa_strfree(oldvd->vdev_path); 5000 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 5001 KM_SLEEP); 5002 (void) sprintf(oldvd->vdev_path, "%s/%s", 5003 newvd->vdev_path, "old"); 5004 if (oldvd->vdev_devid != NULL) { 5005 spa_strfree(oldvd->vdev_devid); 5006 oldvd->vdev_devid = NULL; 5007 } 5008 } 5009 5010 /* mark the device being resilvered */ 5011 newvd->vdev_resilver_txg = txg; 5012 5013 /* 5014 * If the parent is not a mirror, or if we're replacing, insert the new 5015 * mirror/replacing/spare vdev above oldvd. 5016 */ 5017 if (pvd->vdev_ops != pvops) 5018 pvd = vdev_add_parent(oldvd, pvops); 5019 5020 ASSERT(pvd->vdev_top->vdev_parent == rvd); 5021 ASSERT(pvd->vdev_ops == pvops); 5022 ASSERT(oldvd->vdev_parent == pvd); 5023 5024 /* 5025 * Extract the new device from its root and add it to pvd. 5026 */ 5027 vdev_remove_child(newrootvd, newvd); 5028 newvd->vdev_id = pvd->vdev_children; 5029 newvd->vdev_crtxg = oldvd->vdev_crtxg; 5030 vdev_add_child(pvd, newvd); 5031 5032 tvd = newvd->vdev_top; 5033 ASSERT(pvd->vdev_top == tvd); 5034 ASSERT(tvd->vdev_parent == rvd); 5035 5036 vdev_config_dirty(tvd); 5037 5038 /* 5039 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 5040 * for any dmu_sync-ed blocks. It will propagate upward when 5041 * spa_vdev_exit() calls vdev_dtl_reassess(). 5042 */ 5043 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 5044 5045 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 5046 dtl_max_txg - TXG_INITIAL); 5047 5048 if (newvd->vdev_isspare) { 5049 spa_spare_activate(newvd); 5050 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 5051 } 5052 5053 oldvdpath = spa_strdup(oldvd->vdev_path); 5054 newvdpath = spa_strdup(newvd->vdev_path); 5055 newvd_isspare = newvd->vdev_isspare; 5056 5057 /* 5058 * Mark newvd's DTL dirty in this txg. 5059 */ 5060 vdev_dirty(tvd, VDD_DTL, newvd, txg); 5061 5062 /* 5063 * Schedule the resilver to restart in the future. We do this to 5064 * ensure that dmu_sync-ed blocks have been stitched into the 5065 * respective datasets. 5066 */ 5067 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 5068 5069 if (spa->spa_bootfs) 5070 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 5071 5072 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 5073 5074 /* 5075 * Commit the config 5076 */ 5077 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 5078 5079 spa_history_log_internal(spa, "vdev attach", NULL, 5080 "%s vdev=%s %s vdev=%s", 5081 replacing && newvd_isspare ? "spare in" : 5082 replacing ? "replace" : "attach", newvdpath, 5083 replacing ? "for" : "to", oldvdpath); 5084 5085 spa_strfree(oldvdpath); 5086 spa_strfree(newvdpath); 5087 5088 return (0); 5089} 5090 5091/* 5092 * Detach a device from a mirror or replacing vdev. 5093 * 5094 * If 'replace_done' is specified, only detach if the parent 5095 * is a replacing vdev. 5096 */ 5097int 5098spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 5099{ 5100 uint64_t txg; 5101 int error; 5102 vdev_t *rvd = spa->spa_root_vdev; 5103 vdev_t *vd, *pvd, *cvd, *tvd; 5104 boolean_t unspare = B_FALSE; 5105 uint64_t unspare_guid = 0; 5106 char *vdpath; 5107 5108 ASSERT(spa_writeable(spa)); 5109 5110 txg = spa_vdev_enter(spa); 5111 5112 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5113 5114 if (vd == NULL) 5115 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5116 5117 if (!vd->vdev_ops->vdev_op_leaf) 5118 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5119 5120 pvd = vd->vdev_parent; 5121 5122 /* 5123 * If the parent/child relationship is not as expected, don't do it. 5124 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 5125 * vdev that's replacing B with C. The user's intent in replacing 5126 * is to go from M(A,B) to M(A,C). If the user decides to cancel 5127 * the replace by detaching C, the expected behavior is to end up 5128 * M(A,B). But suppose that right after deciding to detach C, 5129 * the replacement of B completes. We would have M(A,C), and then 5130 * ask to detach C, which would leave us with just A -- not what 5131 * the user wanted. To prevent this, we make sure that the 5132 * parent/child relationship hasn't changed -- in this example, 5133 * that C's parent is still the replacing vdev R. 5134 */ 5135 if (pvd->vdev_guid != pguid && pguid != 0) 5136 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5137 5138 /* 5139 * Only 'replacing' or 'spare' vdevs can be replaced. 5140 */ 5141 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5142 pvd->vdev_ops != &vdev_spare_ops) 5143 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5144 5145 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5146 spa_version(spa) >= SPA_VERSION_SPARES); 5147 5148 /* 5149 * Only mirror, replacing, and spare vdevs support detach. 5150 */ 5151 if (pvd->vdev_ops != &vdev_replacing_ops && 5152 pvd->vdev_ops != &vdev_mirror_ops && 5153 pvd->vdev_ops != &vdev_spare_ops) 5154 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5155 5156 /* 5157 * If this device has the only valid copy of some data, 5158 * we cannot safely detach it. 5159 */ 5160 if (vdev_dtl_required(vd)) 5161 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5162 5163 ASSERT(pvd->vdev_children >= 2); 5164 5165 /* 5166 * If we are detaching the second disk from a replacing vdev, then 5167 * check to see if we changed the original vdev's path to have "/old" 5168 * at the end in spa_vdev_attach(). If so, undo that change now. 5169 */ 5170 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5171 vd->vdev_path != NULL) { 5172 size_t len = strlen(vd->vdev_path); 5173 5174 for (int c = 0; c < pvd->vdev_children; c++) { 5175 cvd = pvd->vdev_child[c]; 5176 5177 if (cvd == vd || cvd->vdev_path == NULL) 5178 continue; 5179 5180 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5181 strcmp(cvd->vdev_path + len, "/old") == 0) { 5182 spa_strfree(cvd->vdev_path); 5183 cvd->vdev_path = spa_strdup(vd->vdev_path); 5184 break; 5185 } 5186 } 5187 } 5188 5189 /* 5190 * If we are detaching the original disk from a spare, then it implies 5191 * that the spare should become a real disk, and be removed from the 5192 * active spare list for the pool. 5193 */ 5194 if (pvd->vdev_ops == &vdev_spare_ops && 5195 vd->vdev_id == 0 && 5196 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5197 unspare = B_TRUE; 5198 5199 /* 5200 * Erase the disk labels so the disk can be used for other things. 5201 * This must be done after all other error cases are handled, 5202 * but before we disembowel vd (so we can still do I/O to it). 5203 * But if we can't do it, don't treat the error as fatal -- 5204 * it may be that the unwritability of the disk is the reason 5205 * it's being detached! 5206 */ 5207 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5208 5209 /* 5210 * Remove vd from its parent and compact the parent's children. 5211 */ 5212 vdev_remove_child(pvd, vd); 5213 vdev_compact_children(pvd); 5214 5215 /* 5216 * Remember one of the remaining children so we can get tvd below. 5217 */ 5218 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5219 5220 /* 5221 * If we need to remove the remaining child from the list of hot spares, 5222 * do it now, marking the vdev as no longer a spare in the process. 5223 * We must do this before vdev_remove_parent(), because that can 5224 * change the GUID if it creates a new toplevel GUID. For a similar 5225 * reason, we must remove the spare now, in the same txg as the detach; 5226 * otherwise someone could attach a new sibling, change the GUID, and 5227 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5228 */ 5229 if (unspare) { 5230 ASSERT(cvd->vdev_isspare); 5231 spa_spare_remove(cvd); 5232 unspare_guid = cvd->vdev_guid; 5233 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5234 cvd->vdev_unspare = B_TRUE; 5235 } 5236 5237 /* 5238 * If the parent mirror/replacing vdev only has one child, 5239 * the parent is no longer needed. Remove it from the tree. 5240 */ 5241 if (pvd->vdev_children == 1) { 5242 if (pvd->vdev_ops == &vdev_spare_ops) 5243 cvd->vdev_unspare = B_FALSE; 5244 vdev_remove_parent(cvd); 5245 } 5246 5247 5248 /* 5249 * We don't set tvd until now because the parent we just removed 5250 * may have been the previous top-level vdev. 5251 */ 5252 tvd = cvd->vdev_top; 5253 ASSERT(tvd->vdev_parent == rvd); 5254 5255 /* 5256 * Reevaluate the parent vdev state. 5257 */ 5258 vdev_propagate_state(cvd); 5259 5260 /* 5261 * If the 'autoexpand' property is set on the pool then automatically 5262 * try to expand the size of the pool. For example if the device we 5263 * just detached was smaller than the others, it may be possible to 5264 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5265 * first so that we can obtain the updated sizes of the leaf vdevs. 5266 */ 5267 if (spa->spa_autoexpand) { 5268 vdev_reopen(tvd); 5269 vdev_expand(tvd, txg); 5270 } 5271 5272 vdev_config_dirty(tvd); 5273 5274 /* 5275 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5276 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5277 * But first make sure we're not on any *other* txg's DTL list, to 5278 * prevent vd from being accessed after it's freed. 5279 */ 5280 vdpath = spa_strdup(vd->vdev_path); 5281 for (int t = 0; t < TXG_SIZE; t++) 5282 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5283 vd->vdev_detached = B_TRUE; 5284 vdev_dirty(tvd, VDD_DTL, vd, txg); 5285 5286 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5287 5288 /* hang on to the spa before we release the lock */ 5289 spa_open_ref(spa, FTAG); 5290 5291 error = spa_vdev_exit(spa, vd, txg, 0); 5292 5293 spa_history_log_internal(spa, "detach", NULL, 5294 "vdev=%s", vdpath); 5295 spa_strfree(vdpath); 5296 5297 /* 5298 * If this was the removal of the original device in a hot spare vdev, 5299 * then we want to go through and remove the device from the hot spare 5300 * list of every other pool. 5301 */ 5302 if (unspare) { 5303 spa_t *altspa = NULL; 5304 5305 mutex_enter(&spa_namespace_lock); 5306 while ((altspa = spa_next(altspa)) != NULL) { 5307 if (altspa->spa_state != POOL_STATE_ACTIVE || 5308 altspa == spa) 5309 continue; 5310 5311 spa_open_ref(altspa, FTAG); 5312 mutex_exit(&spa_namespace_lock); 5313 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5314 mutex_enter(&spa_namespace_lock); 5315 spa_close(altspa, FTAG); 5316 } 5317 mutex_exit(&spa_namespace_lock); 5318 5319 /* search the rest of the vdevs for spares to remove */ 5320 spa_vdev_resilver_done(spa); 5321 } 5322 5323 /* all done with the spa; OK to release */ 5324 mutex_enter(&spa_namespace_lock); 5325 spa_close(spa, FTAG); 5326 mutex_exit(&spa_namespace_lock); 5327 5328 return (error); 5329} 5330 5331/* 5332 * Split a set of devices from their mirrors, and create a new pool from them. 5333 */ 5334int 5335spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5336 nvlist_t *props, boolean_t exp) 5337{ 5338 int error = 0; 5339 uint64_t txg, *glist; 5340 spa_t *newspa; 5341 uint_t c, children, lastlog; 5342 nvlist_t **child, *nvl, *tmp; 5343 dmu_tx_t *tx; 5344 char *altroot = NULL; 5345 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5346 boolean_t activate_slog; 5347 5348 ASSERT(spa_writeable(spa)); 5349 5350 txg = spa_vdev_enter(spa); 5351 5352 /* clear the log and flush everything up to now */ 5353 activate_slog = spa_passivate_log(spa); 5354 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5355 error = spa_offline_log(spa); 5356 txg = spa_vdev_config_enter(spa); 5357 5358 if (activate_slog) 5359 spa_activate_log(spa); 5360 5361 if (error != 0) 5362 return (spa_vdev_exit(spa, NULL, txg, error)); 5363 5364 /* check new spa name before going any further */ 5365 if (spa_lookup(newname) != NULL) 5366 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5367 5368 /* 5369 * scan through all the children to ensure they're all mirrors 5370 */ 5371 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5372 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5373 &children) != 0) 5374 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5375 5376 /* first, check to ensure we've got the right child count */ 5377 rvd = spa->spa_root_vdev; 5378 lastlog = 0; 5379 for (c = 0; c < rvd->vdev_children; c++) { 5380 vdev_t *vd = rvd->vdev_child[c]; 5381 5382 /* don't count the holes & logs as children */ 5383 if (vd->vdev_islog || vd->vdev_ishole) { 5384 if (lastlog == 0) 5385 lastlog = c; 5386 continue; 5387 } 5388 5389 lastlog = 0; 5390 } 5391 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5392 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5393 5394 /* next, ensure no spare or cache devices are part of the split */ 5395 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5396 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5397 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5398 5399 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5400 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5401 5402 /* then, loop over each vdev and validate it */ 5403 for (c = 0; c < children; c++) { 5404 uint64_t is_hole = 0; 5405 5406 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5407 &is_hole); 5408 5409 if (is_hole != 0) { 5410 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5411 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5412 continue; 5413 } else { 5414 error = SET_ERROR(EINVAL); 5415 break; 5416 } 5417 } 5418 5419 /* which disk is going to be split? */ 5420 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5421 &glist[c]) != 0) { 5422 error = SET_ERROR(EINVAL); 5423 break; 5424 } 5425 5426 /* look it up in the spa */ 5427 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5428 if (vml[c] == NULL) { 5429 error = SET_ERROR(ENODEV); 5430 break; 5431 } 5432 5433 /* make sure there's nothing stopping the split */ 5434 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5435 vml[c]->vdev_islog || 5436 vml[c]->vdev_ishole || 5437 vml[c]->vdev_isspare || 5438 vml[c]->vdev_isl2cache || 5439 !vdev_writeable(vml[c]) || 5440 vml[c]->vdev_children != 0 || 5441 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5442 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5443 error = SET_ERROR(EINVAL); 5444 break; 5445 } 5446 5447 if (vdev_dtl_required(vml[c])) { 5448 error = SET_ERROR(EBUSY); 5449 break; 5450 } 5451 5452 /* we need certain info from the top level */ 5453 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5454 vml[c]->vdev_top->vdev_ms_array) == 0); 5455 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5456 vml[c]->vdev_top->vdev_ms_shift) == 0); 5457 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5458 vml[c]->vdev_top->vdev_asize) == 0); 5459 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5460 vml[c]->vdev_top->vdev_ashift) == 0); 5461 5462 /* transfer per-vdev ZAPs */ 5463 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 5464 VERIFY0(nvlist_add_uint64(child[c], 5465 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 5466 5467 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 5468 VERIFY0(nvlist_add_uint64(child[c], 5469 ZPOOL_CONFIG_VDEV_TOP_ZAP, 5470 vml[c]->vdev_parent->vdev_top_zap)); 5471 } 5472 5473 if (error != 0) { 5474 kmem_free(vml, children * sizeof (vdev_t *)); 5475 kmem_free(glist, children * sizeof (uint64_t)); 5476 return (spa_vdev_exit(spa, NULL, txg, error)); 5477 } 5478 5479 /* stop writers from using the disks */ 5480 for (c = 0; c < children; c++) { 5481 if (vml[c] != NULL) 5482 vml[c]->vdev_offline = B_TRUE; 5483 } 5484 vdev_reopen(spa->spa_root_vdev); 5485 5486 /* 5487 * Temporarily record the splitting vdevs in the spa config. This 5488 * will disappear once the config is regenerated. 5489 */ 5490 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5491 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5492 glist, children) == 0); 5493 kmem_free(glist, children * sizeof (uint64_t)); 5494 5495 mutex_enter(&spa->spa_props_lock); 5496 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5497 nvl) == 0); 5498 mutex_exit(&spa->spa_props_lock); 5499 spa->spa_config_splitting = nvl; 5500 vdev_config_dirty(spa->spa_root_vdev); 5501 5502 /* configure and create the new pool */ 5503 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5504 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5505 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5506 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5507 spa_version(spa)) == 0); 5508 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5509 spa->spa_config_txg) == 0); 5510 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5511 spa_generate_guid(NULL)) == 0); 5512 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 5513 (void) nvlist_lookup_string(props, 5514 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5515 5516 /* add the new pool to the namespace */ 5517 newspa = spa_add(newname, config, altroot); 5518 newspa->spa_avz_action = AVZ_ACTION_REBUILD; 5519 newspa->spa_config_txg = spa->spa_config_txg; 5520 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5521 5522 /* release the spa config lock, retaining the namespace lock */ 5523 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5524 5525 if (zio_injection_enabled) 5526 zio_handle_panic_injection(spa, FTAG, 1); 5527 5528 spa_activate(newspa, spa_mode_global); 5529 spa_async_suspend(newspa); 5530 5531#ifndef illumos 5532 /* mark that we are creating new spa by splitting */ 5533 newspa->spa_splitting_newspa = B_TRUE; 5534#endif 5535 /* create the new pool from the disks of the original pool */ 5536 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5537#ifndef illumos 5538 newspa->spa_splitting_newspa = B_FALSE; 5539#endif 5540 if (error) 5541 goto out; 5542 5543 /* if that worked, generate a real config for the new pool */ 5544 if (newspa->spa_root_vdev != NULL) { 5545 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5546 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5547 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5548 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5549 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5550 B_TRUE)); 5551 } 5552 5553 /* set the props */ 5554 if (props != NULL) { 5555 spa_configfile_set(newspa, props, B_FALSE); 5556 error = spa_prop_set(newspa, props); 5557 if (error) 5558 goto out; 5559 } 5560 5561 /* flush everything */ 5562 txg = spa_vdev_config_enter(newspa); 5563 vdev_config_dirty(newspa->spa_root_vdev); 5564 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5565 5566 if (zio_injection_enabled) 5567 zio_handle_panic_injection(spa, FTAG, 2); 5568 5569 spa_async_resume(newspa); 5570 5571 /* finally, update the original pool's config */ 5572 txg = spa_vdev_config_enter(spa); 5573 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5574 error = dmu_tx_assign(tx, TXG_WAIT); 5575 if (error != 0) 5576 dmu_tx_abort(tx); 5577 for (c = 0; c < children; c++) { 5578 if (vml[c] != NULL) { 5579 vdev_split(vml[c]); 5580 if (error == 0) 5581 spa_history_log_internal(spa, "detach", tx, 5582 "vdev=%s", vml[c]->vdev_path); 5583 5584 vdev_free(vml[c]); 5585 } 5586 } 5587 spa->spa_avz_action = AVZ_ACTION_REBUILD; 5588 vdev_config_dirty(spa->spa_root_vdev); 5589 spa->spa_config_splitting = NULL; 5590 nvlist_free(nvl); 5591 if (error == 0) 5592 dmu_tx_commit(tx); 5593 (void) spa_vdev_exit(spa, NULL, txg, 0); 5594 5595 if (zio_injection_enabled) 5596 zio_handle_panic_injection(spa, FTAG, 3); 5597 5598 /* split is complete; log a history record */ 5599 spa_history_log_internal(newspa, "split", NULL, 5600 "from pool %s", spa_name(spa)); 5601 5602 kmem_free(vml, children * sizeof (vdev_t *)); 5603 5604 /* if we're not going to mount the filesystems in userland, export */ 5605 if (exp) 5606 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5607 B_FALSE, B_FALSE); 5608 5609 return (error); 5610 5611out: 5612 spa_unload(newspa); 5613 spa_deactivate(newspa); 5614 spa_remove(newspa); 5615 5616 txg = spa_vdev_config_enter(spa); 5617 5618 /* re-online all offlined disks */ 5619 for (c = 0; c < children; c++) { 5620 if (vml[c] != NULL) 5621 vml[c]->vdev_offline = B_FALSE; 5622 } 5623 vdev_reopen(spa->spa_root_vdev); 5624 5625 nvlist_free(spa->spa_config_splitting); 5626 spa->spa_config_splitting = NULL; 5627 (void) spa_vdev_exit(spa, NULL, txg, error); 5628 5629 kmem_free(vml, children * sizeof (vdev_t *)); 5630 return (error); 5631} 5632 5633static nvlist_t * 5634spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5635{ 5636 for (int i = 0; i < count; i++) { 5637 uint64_t guid; 5638 5639 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5640 &guid) == 0); 5641 5642 if (guid == target_guid) 5643 return (nvpp[i]); 5644 } 5645 5646 return (NULL); 5647} 5648 5649static void 5650spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5651 nvlist_t *dev_to_remove) 5652{ 5653 nvlist_t **newdev = NULL; 5654 5655 if (count > 1) 5656 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5657 5658 for (int i = 0, j = 0; i < count; i++) { 5659 if (dev[i] == dev_to_remove) 5660 continue; 5661 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5662 } 5663 5664 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5665 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5666 5667 for (int i = 0; i < count - 1; i++) 5668 nvlist_free(newdev[i]); 5669 5670 if (count > 1) 5671 kmem_free(newdev, (count - 1) * sizeof (void *)); 5672} 5673 5674/* 5675 * Evacuate the device. 5676 */ 5677static int 5678spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5679{ 5680 uint64_t txg; 5681 int error = 0; 5682 5683 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5684 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5685 ASSERT(vd == vd->vdev_top); 5686 5687 /* 5688 * Evacuate the device. We don't hold the config lock as writer 5689 * since we need to do I/O but we do keep the 5690 * spa_namespace_lock held. Once this completes the device 5691 * should no longer have any blocks allocated on it. 5692 */ 5693 if (vd->vdev_islog) { 5694 if (vd->vdev_stat.vs_alloc != 0) 5695 error = spa_offline_log(spa); 5696 } else { 5697 error = SET_ERROR(ENOTSUP); 5698 } 5699 5700 if (error) 5701 return (error); 5702 5703 /* 5704 * The evacuation succeeded. Remove any remaining MOS metadata 5705 * associated with this vdev, and wait for these changes to sync. 5706 */ 5707 ASSERT0(vd->vdev_stat.vs_alloc); 5708 txg = spa_vdev_config_enter(spa); 5709 vd->vdev_removing = B_TRUE; 5710 vdev_dirty_leaves(vd, VDD_DTL, txg); 5711 vdev_config_dirty(vd); 5712 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5713 5714 return (0); 5715} 5716 5717/* 5718 * Complete the removal by cleaning up the namespace. 5719 */ 5720static void 5721spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5722{ 5723 vdev_t *rvd = spa->spa_root_vdev; 5724 uint64_t id = vd->vdev_id; 5725 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5726 5727 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5728 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5729 ASSERT(vd == vd->vdev_top); 5730 5731 /* 5732 * Only remove any devices which are empty. 5733 */ 5734 if (vd->vdev_stat.vs_alloc != 0) 5735 return; 5736 5737 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5738 5739 if (list_link_active(&vd->vdev_state_dirty_node)) 5740 vdev_state_clean(vd); 5741 if (list_link_active(&vd->vdev_config_dirty_node)) 5742 vdev_config_clean(vd); 5743 5744 vdev_free(vd); 5745 5746 if (last_vdev) { 5747 vdev_compact_children(rvd); 5748 } else { 5749 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5750 vdev_add_child(rvd, vd); 5751 } 5752 vdev_config_dirty(rvd); 5753 5754 /* 5755 * Reassess the health of our root vdev. 5756 */ 5757 vdev_reopen(rvd); 5758} 5759 5760/* 5761 * Remove a device from the pool - 5762 * 5763 * Removing a device from the vdev namespace requires several steps 5764 * and can take a significant amount of time. As a result we use 5765 * the spa_vdev_config_[enter/exit] functions which allow us to 5766 * grab and release the spa_config_lock while still holding the namespace 5767 * lock. During each step the configuration is synced out. 5768 * 5769 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5770 * devices. 5771 */ 5772int 5773spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5774{ 5775 vdev_t *vd; 5776 sysevent_t *ev = NULL; 5777 metaslab_group_t *mg; 5778 nvlist_t **spares, **l2cache, *nv; 5779 uint64_t txg = 0; 5780 uint_t nspares, nl2cache; 5781 int error = 0; 5782 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5783 5784 ASSERT(spa_writeable(spa)); 5785 5786 if (!locked) 5787 txg = spa_vdev_enter(spa); 5788 5789 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5790 5791 if (spa->spa_spares.sav_vdevs != NULL && 5792 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5793 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5794 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5795 /* 5796 * Only remove the hot spare if it's not currently in use 5797 * in this pool. 5798 */ 5799 if (vd == NULL || unspare) { 5800 if (vd == NULL) 5801 vd = spa_lookup_by_guid(spa, guid, B_TRUE); 5802 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5803 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5804 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5805 spa_load_spares(spa); 5806 spa->spa_spares.sav_sync = B_TRUE; 5807 } else { 5808 error = SET_ERROR(EBUSY); 5809 } 5810 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5811 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5812 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5813 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5814 /* 5815 * Cache devices can always be removed. 5816 */ 5817 vd = spa_lookup_by_guid(spa, guid, B_TRUE); 5818 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5819 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5820 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5821 spa_load_l2cache(spa); 5822 spa->spa_l2cache.sav_sync = B_TRUE; 5823 } else if (vd != NULL && vd->vdev_islog) { 5824 ASSERT(!locked); 5825 ASSERT(vd == vd->vdev_top); 5826 5827 mg = vd->vdev_mg; 5828 5829 /* 5830 * Stop allocating from this vdev. 5831 */ 5832 metaslab_group_passivate(mg); 5833 5834 /* 5835 * Wait for the youngest allocations and frees to sync, 5836 * and then wait for the deferral of those frees to finish. 5837 */ 5838 spa_vdev_config_exit(spa, NULL, 5839 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5840 5841 /* 5842 * Attempt to evacuate the vdev. 5843 */ 5844 error = spa_vdev_remove_evacuate(spa, vd); 5845 5846 txg = spa_vdev_config_enter(spa); 5847 5848 /* 5849 * If we couldn't evacuate the vdev, unwind. 5850 */ 5851 if (error) { 5852 metaslab_group_activate(mg); 5853 return (spa_vdev_exit(spa, NULL, txg, error)); 5854 } 5855 5856 /* 5857 * Clean up the vdev namespace. 5858 */ 5859 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV); 5860 spa_vdev_remove_from_namespace(spa, vd); 5861 5862 } else if (vd != NULL) { 5863 /* 5864 * Normal vdevs cannot be removed (yet). 5865 */ 5866 error = SET_ERROR(ENOTSUP); 5867 } else { 5868 /* 5869 * There is no vdev of any kind with the specified guid. 5870 */ 5871 error = SET_ERROR(ENOENT); 5872 } 5873 5874 if (!locked) 5875 error = spa_vdev_exit(spa, NULL, txg, error); 5876 5877 if (ev) 5878 spa_event_post(ev); 5879 5880 return (error); 5881} 5882 5883/* 5884 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5885 * currently spared, so we can detach it. 5886 */ 5887static vdev_t * 5888spa_vdev_resilver_done_hunt(vdev_t *vd) 5889{ 5890 vdev_t *newvd, *oldvd; 5891 5892 for (int c = 0; c < vd->vdev_children; c++) { 5893 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5894 if (oldvd != NULL) 5895 return (oldvd); 5896 } 5897 5898 /* 5899 * Check for a completed replacement. We always consider the first 5900 * vdev in the list to be the oldest vdev, and the last one to be 5901 * the newest (see spa_vdev_attach() for how that works). In 5902 * the case where the newest vdev is faulted, we will not automatically 5903 * remove it after a resilver completes. This is OK as it will require 5904 * user intervention to determine which disk the admin wishes to keep. 5905 */ 5906 if (vd->vdev_ops == &vdev_replacing_ops) { 5907 ASSERT(vd->vdev_children > 1); 5908 5909 newvd = vd->vdev_child[vd->vdev_children - 1]; 5910 oldvd = vd->vdev_child[0]; 5911 5912 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5913 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5914 !vdev_dtl_required(oldvd)) 5915 return (oldvd); 5916 } 5917 5918 /* 5919 * Check for a completed resilver with the 'unspare' flag set. 5920 */ 5921 if (vd->vdev_ops == &vdev_spare_ops) { 5922 vdev_t *first = vd->vdev_child[0]; 5923 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5924 5925 if (last->vdev_unspare) { 5926 oldvd = first; 5927 newvd = last; 5928 } else if (first->vdev_unspare) { 5929 oldvd = last; 5930 newvd = first; 5931 } else { 5932 oldvd = NULL; 5933 } 5934 5935 if (oldvd != NULL && 5936 vdev_dtl_empty(newvd, DTL_MISSING) && 5937 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5938 !vdev_dtl_required(oldvd)) 5939 return (oldvd); 5940 5941 /* 5942 * If there are more than two spares attached to a disk, 5943 * and those spares are not required, then we want to 5944 * attempt to free them up now so that they can be used 5945 * by other pools. Once we're back down to a single 5946 * disk+spare, we stop removing them. 5947 */ 5948 if (vd->vdev_children > 2) { 5949 newvd = vd->vdev_child[1]; 5950 5951 if (newvd->vdev_isspare && last->vdev_isspare && 5952 vdev_dtl_empty(last, DTL_MISSING) && 5953 vdev_dtl_empty(last, DTL_OUTAGE) && 5954 !vdev_dtl_required(newvd)) 5955 return (newvd); 5956 } 5957 } 5958 5959 return (NULL); 5960} 5961 5962static void 5963spa_vdev_resilver_done(spa_t *spa) 5964{ 5965 vdev_t *vd, *pvd, *ppvd; 5966 uint64_t guid, sguid, pguid, ppguid; 5967 5968 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5969 5970 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5971 pvd = vd->vdev_parent; 5972 ppvd = pvd->vdev_parent; 5973 guid = vd->vdev_guid; 5974 pguid = pvd->vdev_guid; 5975 ppguid = ppvd->vdev_guid; 5976 sguid = 0; 5977 /* 5978 * If we have just finished replacing a hot spared device, then 5979 * we need to detach the parent's first child (the original hot 5980 * spare) as well. 5981 */ 5982 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5983 ppvd->vdev_children == 2) { 5984 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5985 sguid = ppvd->vdev_child[1]->vdev_guid; 5986 } 5987 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5988 5989 spa_config_exit(spa, SCL_ALL, FTAG); 5990 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5991 return; 5992 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5993 return; 5994 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5995 } 5996 5997 spa_config_exit(spa, SCL_ALL, FTAG); 5998} 5999 6000/* 6001 * Update the stored path or FRU for this vdev. 6002 */ 6003int 6004spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 6005 boolean_t ispath) 6006{ 6007 vdev_t *vd; 6008 boolean_t sync = B_FALSE; 6009 6010 ASSERT(spa_writeable(spa)); 6011 6012 spa_vdev_state_enter(spa, SCL_ALL); 6013 6014 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 6015 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 6016 6017 if (!vd->vdev_ops->vdev_op_leaf) 6018 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 6019 6020 if (ispath) { 6021 if (strcmp(value, vd->vdev_path) != 0) { 6022 spa_strfree(vd->vdev_path); 6023 vd->vdev_path = spa_strdup(value); 6024 sync = B_TRUE; 6025 } 6026 } else { 6027 if (vd->vdev_fru == NULL) { 6028 vd->vdev_fru = spa_strdup(value); 6029 sync = B_TRUE; 6030 } else if (strcmp(value, vd->vdev_fru) != 0) { 6031 spa_strfree(vd->vdev_fru); 6032 vd->vdev_fru = spa_strdup(value); 6033 sync = B_TRUE; 6034 } 6035 } 6036 6037 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 6038} 6039 6040int 6041spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 6042{ 6043 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 6044} 6045 6046int 6047spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 6048{ 6049 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 6050} 6051 6052/* 6053 * ========================================================================== 6054 * SPA Scanning 6055 * ========================================================================== 6056 */ 6057int 6058spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) 6059{ 6060 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6061 6062 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6063 return (SET_ERROR(EBUSY)); 6064 6065 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); 6066} 6067 6068int 6069spa_scan_stop(spa_t *spa) 6070{ 6071 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6072 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6073 return (SET_ERROR(EBUSY)); 6074 return (dsl_scan_cancel(spa->spa_dsl_pool)); 6075} 6076 6077int 6078spa_scan(spa_t *spa, pool_scan_func_t func) 6079{ 6080 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6081 6082 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 6083 return (SET_ERROR(ENOTSUP)); 6084 6085 /* 6086 * If a resilver was requested, but there is no DTL on a 6087 * writeable leaf device, we have nothing to do. 6088 */ 6089 if (func == POOL_SCAN_RESILVER && 6090 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 6091 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 6092 return (0); 6093 } 6094 6095 return (dsl_scan(spa->spa_dsl_pool, func)); 6096} 6097 6098/* 6099 * ========================================================================== 6100 * SPA async task processing 6101 * ========================================================================== 6102 */ 6103 6104static void 6105spa_async_remove(spa_t *spa, vdev_t *vd) 6106{ 6107 if (vd->vdev_remove_wanted) { 6108 vd->vdev_remove_wanted = B_FALSE; 6109 vd->vdev_delayed_close = B_FALSE; 6110 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 6111 6112 /* 6113 * We want to clear the stats, but we don't want to do a full 6114 * vdev_clear() as that will cause us to throw away 6115 * degraded/faulted state as well as attempt to reopen the 6116 * device, all of which is a waste. 6117 */ 6118 vd->vdev_stat.vs_read_errors = 0; 6119 vd->vdev_stat.vs_write_errors = 0; 6120 vd->vdev_stat.vs_checksum_errors = 0; 6121 6122 vdev_state_dirty(vd->vdev_top); 6123 /* Tell userspace that the vdev is gone. */ 6124 zfs_post_remove(spa, vd); 6125 } 6126 6127 for (int c = 0; c < vd->vdev_children; c++) 6128 spa_async_remove(spa, vd->vdev_child[c]); 6129} 6130 6131static void 6132spa_async_probe(spa_t *spa, vdev_t *vd) 6133{ 6134 if (vd->vdev_probe_wanted) { 6135 vd->vdev_probe_wanted = B_FALSE; 6136 vdev_reopen(vd); /* vdev_open() does the actual probe */ 6137 } 6138 6139 for (int c = 0; c < vd->vdev_children; c++) 6140 spa_async_probe(spa, vd->vdev_child[c]); 6141} 6142 6143static void 6144spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6145{ 6146 sysevent_id_t eid; 6147 nvlist_t *attr; 6148 char *physpath; 6149 6150 if (!spa->spa_autoexpand) 6151 return; 6152 6153 for (int c = 0; c < vd->vdev_children; c++) { 6154 vdev_t *cvd = vd->vdev_child[c]; 6155 spa_async_autoexpand(spa, cvd); 6156 } 6157 6158 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6159 return; 6160 6161 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6162 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6163 6164 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6165 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6166 6167 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6168 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 6169 6170 nvlist_free(attr); 6171 kmem_free(physpath, MAXPATHLEN); 6172} 6173 6174static void 6175spa_async_thread(void *arg) 6176{ 6177 spa_t *spa = arg; 6178 int tasks; 6179 6180 ASSERT(spa->spa_sync_on); 6181 6182 mutex_enter(&spa->spa_async_lock); 6183 tasks = spa->spa_async_tasks; 6184 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 6185 mutex_exit(&spa->spa_async_lock); 6186 6187 /* 6188 * See if the config needs to be updated. 6189 */ 6190 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6191 uint64_t old_space, new_space; 6192 6193 mutex_enter(&spa_namespace_lock); 6194 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6195 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6196 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6197 mutex_exit(&spa_namespace_lock); 6198 6199 /* 6200 * If the pool grew as a result of the config update, 6201 * then log an internal history event. 6202 */ 6203 if (new_space != old_space) { 6204 spa_history_log_internal(spa, "vdev online", NULL, 6205 "pool '%s' size: %llu(+%llu)", 6206 spa_name(spa), new_space, new_space - old_space); 6207 } 6208 } 6209 6210 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6211 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6212 spa_async_autoexpand(spa, spa->spa_root_vdev); 6213 spa_config_exit(spa, SCL_CONFIG, FTAG); 6214 } 6215 6216 /* 6217 * See if any devices need to be probed. 6218 */ 6219 if (tasks & SPA_ASYNC_PROBE) { 6220 spa_vdev_state_enter(spa, SCL_NONE); 6221 spa_async_probe(spa, spa->spa_root_vdev); 6222 (void) spa_vdev_state_exit(spa, NULL, 0); 6223 } 6224 6225 /* 6226 * If any devices are done replacing, detach them. 6227 */ 6228 if (tasks & SPA_ASYNC_RESILVER_DONE) 6229 spa_vdev_resilver_done(spa); 6230 6231 /* 6232 * Kick off a resilver. 6233 */ 6234 if (tasks & SPA_ASYNC_RESILVER) 6235 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6236 6237 /* 6238 * Let the world know that we're done. 6239 */ 6240 mutex_enter(&spa->spa_async_lock); 6241 spa->spa_async_thread = NULL; 6242 cv_broadcast(&spa->spa_async_cv); 6243 mutex_exit(&spa->spa_async_lock); 6244 thread_exit(); 6245} 6246 6247static void 6248spa_async_thread_vd(void *arg) 6249{ 6250 spa_t *spa = arg; 6251 int tasks; 6252 6253 ASSERT(spa->spa_sync_on); 6254 6255 mutex_enter(&spa->spa_async_lock); 6256 tasks = spa->spa_async_tasks; 6257retry: 6258 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6259 mutex_exit(&spa->spa_async_lock); 6260 6261 /* 6262 * See if any devices need to be marked REMOVED. 6263 */ 6264 if (tasks & SPA_ASYNC_REMOVE) { 6265 spa_vdev_state_enter(spa, SCL_NONE); 6266 spa_async_remove(spa, spa->spa_root_vdev); 6267 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6268 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6269 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6270 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6271 (void) spa_vdev_state_exit(spa, NULL, 0); 6272 } 6273 6274 /* 6275 * Let the world know that we're done. 6276 */ 6277 mutex_enter(&spa->spa_async_lock); 6278 tasks = spa->spa_async_tasks; 6279 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6280 goto retry; 6281 spa->spa_async_thread_vd = NULL; 6282 cv_broadcast(&spa->spa_async_cv); 6283 mutex_exit(&spa->spa_async_lock); 6284 thread_exit(); 6285} 6286 6287void 6288spa_async_suspend(spa_t *spa) 6289{ 6290 mutex_enter(&spa->spa_async_lock); 6291 spa->spa_async_suspended++; 6292 while (spa->spa_async_thread != NULL && 6293 spa->spa_async_thread_vd != NULL) 6294 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6295 mutex_exit(&spa->spa_async_lock); 6296} 6297 6298void 6299spa_async_resume(spa_t *spa) 6300{ 6301 mutex_enter(&spa->spa_async_lock); 6302 ASSERT(spa->spa_async_suspended != 0); 6303 spa->spa_async_suspended--; 6304 mutex_exit(&spa->spa_async_lock); 6305} 6306 6307static boolean_t 6308spa_async_tasks_pending(spa_t *spa) 6309{ 6310 uint_t non_config_tasks; 6311 uint_t config_task; 6312 boolean_t config_task_suspended; 6313 6314 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6315 SPA_ASYNC_REMOVE); 6316 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6317 if (spa->spa_ccw_fail_time == 0) { 6318 config_task_suspended = B_FALSE; 6319 } else { 6320 config_task_suspended = 6321 (gethrtime() - spa->spa_ccw_fail_time) < 6322 (zfs_ccw_retry_interval * NANOSEC); 6323 } 6324 6325 return (non_config_tasks || (config_task && !config_task_suspended)); 6326} 6327 6328static void 6329spa_async_dispatch(spa_t *spa) 6330{ 6331 mutex_enter(&spa->spa_async_lock); 6332 if (spa_async_tasks_pending(spa) && 6333 !spa->spa_async_suspended && 6334 spa->spa_async_thread == NULL && 6335 rootdir != NULL) 6336 spa->spa_async_thread = thread_create(NULL, 0, 6337 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6338 mutex_exit(&spa->spa_async_lock); 6339} 6340 6341static void 6342spa_async_dispatch_vd(spa_t *spa) 6343{ 6344 mutex_enter(&spa->spa_async_lock); 6345 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6346 !spa->spa_async_suspended && 6347 spa->spa_async_thread_vd == NULL && 6348 rootdir != NULL) 6349 spa->spa_async_thread_vd = thread_create(NULL, 0, 6350 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6351 mutex_exit(&spa->spa_async_lock); 6352} 6353 6354void 6355spa_async_request(spa_t *spa, int task) 6356{ 6357 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6358 mutex_enter(&spa->spa_async_lock); 6359 spa->spa_async_tasks |= task; 6360 mutex_exit(&spa->spa_async_lock); 6361 spa_async_dispatch_vd(spa); 6362} 6363 6364/* 6365 * ========================================================================== 6366 * SPA syncing routines 6367 * ========================================================================== 6368 */ 6369 6370static int 6371bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6372{ 6373 bpobj_t *bpo = arg; 6374 bpobj_enqueue(bpo, bp, tx); 6375 return (0); 6376} 6377 6378static int 6379spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6380{ 6381 zio_t *zio = arg; 6382 6383 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6384 BP_GET_PSIZE(bp), zio->io_flags)); 6385 return (0); 6386} 6387 6388/* 6389 * Note: this simple function is not inlined to make it easier to dtrace the 6390 * amount of time spent syncing frees. 6391 */ 6392static void 6393spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6394{ 6395 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6396 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6397 VERIFY(zio_wait(zio) == 0); 6398} 6399 6400/* 6401 * Note: this simple function is not inlined to make it easier to dtrace the 6402 * amount of time spent syncing deferred frees. 6403 */ 6404static void 6405spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6406{ 6407 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6408 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6409 spa_free_sync_cb, zio, tx), ==, 0); 6410 VERIFY0(zio_wait(zio)); 6411} 6412 6413 6414static void 6415spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6416{ 6417 char *packed = NULL; 6418 size_t bufsize; 6419 size_t nvsize = 0; 6420 dmu_buf_t *db; 6421 6422 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6423 6424 /* 6425 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6426 * information. This avoids the dmu_buf_will_dirty() path and 6427 * saves us a pre-read to get data we don't actually care about. 6428 */ 6429 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6430 packed = kmem_alloc(bufsize, KM_SLEEP); 6431 6432 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6433 KM_SLEEP) == 0); 6434 bzero(packed + nvsize, bufsize - nvsize); 6435 6436 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6437 6438 kmem_free(packed, bufsize); 6439 6440 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6441 dmu_buf_will_dirty(db, tx); 6442 *(uint64_t *)db->db_data = nvsize; 6443 dmu_buf_rele(db, FTAG); 6444} 6445 6446static void 6447spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6448 const char *config, const char *entry) 6449{ 6450 nvlist_t *nvroot; 6451 nvlist_t **list; 6452 int i; 6453 6454 if (!sav->sav_sync) 6455 return; 6456 6457 /* 6458 * Update the MOS nvlist describing the list of available devices. 6459 * spa_validate_aux() will have already made sure this nvlist is 6460 * valid and the vdevs are labeled appropriately. 6461 */ 6462 if (sav->sav_object == 0) { 6463 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6464 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6465 sizeof (uint64_t), tx); 6466 VERIFY(zap_update(spa->spa_meta_objset, 6467 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6468 &sav->sav_object, tx) == 0); 6469 } 6470 6471 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6472 if (sav->sav_count == 0) { 6473 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6474 } else { 6475 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6476 for (i = 0; i < sav->sav_count; i++) 6477 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6478 B_FALSE, VDEV_CONFIG_L2CACHE); 6479 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6480 sav->sav_count) == 0); 6481 for (i = 0; i < sav->sav_count; i++) 6482 nvlist_free(list[i]); 6483 kmem_free(list, sav->sav_count * sizeof (void *)); 6484 } 6485 6486 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6487 nvlist_free(nvroot); 6488 6489 sav->sav_sync = B_FALSE; 6490} 6491 6492/* 6493 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 6494 * The all-vdev ZAP must be empty. 6495 */ 6496static void 6497spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 6498{ 6499 spa_t *spa = vd->vdev_spa; 6500 if (vd->vdev_top_zap != 0) { 6501 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6502 vd->vdev_top_zap, tx)); 6503 } 6504 if (vd->vdev_leaf_zap != 0) { 6505 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6506 vd->vdev_leaf_zap, tx)); 6507 } 6508 for (uint64_t i = 0; i < vd->vdev_children; i++) { 6509 spa_avz_build(vd->vdev_child[i], avz, tx); 6510 } 6511} 6512 6513static void 6514spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6515{ 6516 nvlist_t *config; 6517 6518 /* 6519 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 6520 * its config may not be dirty but we still need to build per-vdev ZAPs. 6521 * Similarly, if the pool is being assembled (e.g. after a split), we 6522 * need to rebuild the AVZ although the config may not be dirty. 6523 */ 6524 if (list_is_empty(&spa->spa_config_dirty_list) && 6525 spa->spa_avz_action == AVZ_ACTION_NONE) 6526 return; 6527 6528 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6529 6530 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 6531 spa->spa_avz_action == AVZ_ACTION_INITIALIZE || 6532 spa->spa_all_vdev_zaps != 0); 6533 6534 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 6535 /* Make and build the new AVZ */ 6536 uint64_t new_avz = zap_create(spa->spa_meta_objset, 6537 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 6538 spa_avz_build(spa->spa_root_vdev, new_avz, tx); 6539 6540 /* Diff old AVZ with new one */ 6541 zap_cursor_t zc; 6542 zap_attribute_t za; 6543 6544 for (zap_cursor_init(&zc, spa->spa_meta_objset, 6545 spa->spa_all_vdev_zaps); 6546 zap_cursor_retrieve(&zc, &za) == 0; 6547 zap_cursor_advance(&zc)) { 6548 uint64_t vdzap = za.za_first_integer; 6549 if (zap_lookup_int(spa->spa_meta_objset, new_avz, 6550 vdzap) == ENOENT) { 6551 /* 6552 * ZAP is listed in old AVZ but not in new one; 6553 * destroy it 6554 */ 6555 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 6556 tx)); 6557 } 6558 } 6559 6560 zap_cursor_fini(&zc); 6561 6562 /* Destroy the old AVZ */ 6563 VERIFY0(zap_destroy(spa->spa_meta_objset, 6564 spa->spa_all_vdev_zaps, tx)); 6565 6566 /* Replace the old AVZ in the dir obj with the new one */ 6567 VERIFY0(zap_update(spa->spa_meta_objset, 6568 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 6569 sizeof (new_avz), 1, &new_avz, tx)); 6570 6571 spa->spa_all_vdev_zaps = new_avz; 6572 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 6573 zap_cursor_t zc; 6574 zap_attribute_t za; 6575 6576 /* Walk through the AVZ and destroy all listed ZAPs */ 6577 for (zap_cursor_init(&zc, spa->spa_meta_objset, 6578 spa->spa_all_vdev_zaps); 6579 zap_cursor_retrieve(&zc, &za) == 0; 6580 zap_cursor_advance(&zc)) { 6581 uint64_t zap = za.za_first_integer; 6582 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 6583 } 6584 6585 zap_cursor_fini(&zc); 6586 6587 /* Destroy and unlink the AVZ itself */ 6588 VERIFY0(zap_destroy(spa->spa_meta_objset, 6589 spa->spa_all_vdev_zaps, tx)); 6590 VERIFY0(zap_remove(spa->spa_meta_objset, 6591 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 6592 spa->spa_all_vdev_zaps = 0; 6593 } 6594 6595 if (spa->spa_all_vdev_zaps == 0) { 6596 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 6597 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 6598 DMU_POOL_VDEV_ZAP_MAP, tx); 6599 } 6600 spa->spa_avz_action = AVZ_ACTION_NONE; 6601 6602 /* Create ZAPs for vdevs that don't have them. */ 6603 vdev_construct_zaps(spa->spa_root_vdev, tx); 6604 6605 config = spa_config_generate(spa, spa->spa_root_vdev, 6606 dmu_tx_get_txg(tx), B_FALSE); 6607 6608 /* 6609 * If we're upgrading the spa version then make sure that 6610 * the config object gets updated with the correct version. 6611 */ 6612 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6613 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6614 spa->spa_uberblock.ub_version); 6615 6616 spa_config_exit(spa, SCL_STATE, FTAG); 6617 6618 nvlist_free(spa->spa_config_syncing); 6619 spa->spa_config_syncing = config; 6620 6621 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6622} 6623 6624static void 6625spa_sync_version(void *arg, dmu_tx_t *tx) 6626{ 6627 uint64_t *versionp = arg; 6628 uint64_t version = *versionp; 6629 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6630 6631 /* 6632 * Setting the version is special cased when first creating the pool. 6633 */ 6634 ASSERT(tx->tx_txg != TXG_INITIAL); 6635 6636 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6637 ASSERT(version >= spa_version(spa)); 6638 6639 spa->spa_uberblock.ub_version = version; 6640 vdev_config_dirty(spa->spa_root_vdev); 6641 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6642} 6643 6644/* 6645 * Set zpool properties. 6646 */ 6647static void 6648spa_sync_props(void *arg, dmu_tx_t *tx) 6649{ 6650 nvlist_t *nvp = arg; 6651 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6652 objset_t *mos = spa->spa_meta_objset; 6653 nvpair_t *elem = NULL; 6654 6655 mutex_enter(&spa->spa_props_lock); 6656 6657 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6658 uint64_t intval; 6659 char *strval, *fname; 6660 zpool_prop_t prop; 6661 const char *propname; 6662 zprop_type_t proptype; 6663 spa_feature_t fid; 6664 6665 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6666 case ZPROP_INVAL: 6667 /* 6668 * We checked this earlier in spa_prop_validate(). 6669 */ 6670 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6671 6672 fname = strchr(nvpair_name(elem), '@') + 1; 6673 VERIFY0(zfeature_lookup_name(fname, &fid)); 6674 6675 spa_feature_enable(spa, fid, tx); 6676 spa_history_log_internal(spa, "set", tx, 6677 "%s=enabled", nvpair_name(elem)); 6678 break; 6679 6680 case ZPOOL_PROP_VERSION: 6681 intval = fnvpair_value_uint64(elem); 6682 /* 6683 * The version is synced seperatly before other 6684 * properties and should be correct by now. 6685 */ 6686 ASSERT3U(spa_version(spa), >=, intval); 6687 break; 6688 6689 case ZPOOL_PROP_ALTROOT: 6690 /* 6691 * 'altroot' is a non-persistent property. It should 6692 * have been set temporarily at creation or import time. 6693 */ 6694 ASSERT(spa->spa_root != NULL); 6695 break; 6696 6697 case ZPOOL_PROP_READONLY: 6698 case ZPOOL_PROP_CACHEFILE: 6699 /* 6700 * 'readonly' and 'cachefile' are also non-persisitent 6701 * properties. 6702 */ 6703 break; 6704 case ZPOOL_PROP_COMMENT: 6705 strval = fnvpair_value_string(elem); 6706 if (spa->spa_comment != NULL) 6707 spa_strfree(spa->spa_comment); 6708 spa->spa_comment = spa_strdup(strval); 6709 /* 6710 * We need to dirty the configuration on all the vdevs 6711 * so that their labels get updated. It's unnecessary 6712 * to do this for pool creation since the vdev's 6713 * configuratoin has already been dirtied. 6714 */ 6715 if (tx->tx_txg != TXG_INITIAL) 6716 vdev_config_dirty(spa->spa_root_vdev); 6717 spa_history_log_internal(spa, "set", tx, 6718 "%s=%s", nvpair_name(elem), strval); 6719 break; 6720 default: 6721 /* 6722 * Set pool property values in the poolprops mos object. 6723 */ 6724 if (spa->spa_pool_props_object == 0) { 6725 spa->spa_pool_props_object = 6726 zap_create_link(mos, DMU_OT_POOL_PROPS, 6727 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6728 tx); 6729 } 6730 6731 /* normalize the property name */ 6732 propname = zpool_prop_to_name(prop); 6733 proptype = zpool_prop_get_type(prop); 6734 6735 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6736 ASSERT(proptype == PROP_TYPE_STRING); 6737 strval = fnvpair_value_string(elem); 6738 VERIFY0(zap_update(mos, 6739 spa->spa_pool_props_object, propname, 6740 1, strlen(strval) + 1, strval, tx)); 6741 spa_history_log_internal(spa, "set", tx, 6742 "%s=%s", nvpair_name(elem), strval); 6743 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6744 intval = fnvpair_value_uint64(elem); 6745 6746 if (proptype == PROP_TYPE_INDEX) { 6747 const char *unused; 6748 VERIFY0(zpool_prop_index_to_string( 6749 prop, intval, &unused)); 6750 } 6751 VERIFY0(zap_update(mos, 6752 spa->spa_pool_props_object, propname, 6753 8, 1, &intval, tx)); 6754 spa_history_log_internal(spa, "set", tx, 6755 "%s=%lld", nvpair_name(elem), intval); 6756 } else { 6757 ASSERT(0); /* not allowed */ 6758 } 6759 6760 switch (prop) { 6761 case ZPOOL_PROP_DELEGATION: 6762 spa->spa_delegation = intval; 6763 break; 6764 case ZPOOL_PROP_BOOTFS: 6765 spa->spa_bootfs = intval; 6766 break; 6767 case ZPOOL_PROP_FAILUREMODE: 6768 spa->spa_failmode = intval; 6769 break; 6770 case ZPOOL_PROP_AUTOEXPAND: 6771 spa->spa_autoexpand = intval; 6772 if (tx->tx_txg != TXG_INITIAL) 6773 spa_async_request(spa, 6774 SPA_ASYNC_AUTOEXPAND); 6775 break; 6776 case ZPOOL_PROP_DEDUPDITTO: 6777 spa->spa_dedup_ditto = intval; 6778 break; 6779 default: 6780 break; 6781 } 6782 } 6783 6784 } 6785 6786 mutex_exit(&spa->spa_props_lock); 6787} 6788 6789/* 6790 * Perform one-time upgrade on-disk changes. spa_version() does not 6791 * reflect the new version this txg, so there must be no changes this 6792 * txg to anything that the upgrade code depends on after it executes. 6793 * Therefore this must be called after dsl_pool_sync() does the sync 6794 * tasks. 6795 */ 6796static void 6797spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6798{ 6799 dsl_pool_t *dp = spa->spa_dsl_pool; 6800 6801 ASSERT(spa->spa_sync_pass == 1); 6802 6803 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6804 6805 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6806 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6807 dsl_pool_create_origin(dp, tx); 6808 6809 /* Keeping the origin open increases spa_minref */ 6810 spa->spa_minref += 3; 6811 } 6812 6813 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6814 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6815 dsl_pool_upgrade_clones(dp, tx); 6816 } 6817 6818 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6819 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6820 dsl_pool_upgrade_dir_clones(dp, tx); 6821 6822 /* Keeping the freedir open increases spa_minref */ 6823 spa->spa_minref += 3; 6824 } 6825 6826 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6827 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6828 spa_feature_create_zap_objects(spa, tx); 6829 } 6830 6831 /* 6832 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6833 * when possibility to use lz4 compression for metadata was added 6834 * Old pools that have this feature enabled must be upgraded to have 6835 * this feature active 6836 */ 6837 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6838 boolean_t lz4_en = spa_feature_is_enabled(spa, 6839 SPA_FEATURE_LZ4_COMPRESS); 6840 boolean_t lz4_ac = spa_feature_is_active(spa, 6841 SPA_FEATURE_LZ4_COMPRESS); 6842 6843 if (lz4_en && !lz4_ac) 6844 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6845 } 6846 6847 /* 6848 * If we haven't written the salt, do so now. Note that the 6849 * feature may not be activated yet, but that's fine since 6850 * the presence of this ZAP entry is backwards compatible. 6851 */ 6852 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 6853 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 6854 VERIFY0(zap_add(spa->spa_meta_objset, 6855 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 6856 sizeof (spa->spa_cksum_salt.zcs_bytes), 6857 spa->spa_cksum_salt.zcs_bytes, tx)); 6858 } 6859 6860 rrw_exit(&dp->dp_config_rwlock, FTAG); 6861} 6862 6863/* 6864 * Sync the specified transaction group. New blocks may be dirtied as 6865 * part of the process, so we iterate until it converges. 6866 */ 6867void 6868spa_sync(spa_t *spa, uint64_t txg) 6869{ 6870 dsl_pool_t *dp = spa->spa_dsl_pool; 6871 objset_t *mos = spa->spa_meta_objset; 6872 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6873 vdev_t *rvd = spa->spa_root_vdev; 6874 vdev_t *vd; 6875 dmu_tx_t *tx; 6876 int error; 6877 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 6878 zfs_vdev_queue_depth_pct / 100; 6879 6880 VERIFY(spa_writeable(spa)); 6881 6882 /* 6883 * Lock out configuration changes. 6884 */ 6885 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6886 6887 spa->spa_syncing_txg = txg; 6888 spa->spa_sync_pass = 0; 6889 6890 mutex_enter(&spa->spa_alloc_lock); 6891 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 6892 mutex_exit(&spa->spa_alloc_lock); 6893 6894 /* 6895 * If there are any pending vdev state changes, convert them 6896 * into config changes that go out with this transaction group. 6897 */ 6898 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6899 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6900 /* 6901 * We need the write lock here because, for aux vdevs, 6902 * calling vdev_config_dirty() modifies sav_config. 6903 * This is ugly and will become unnecessary when we 6904 * eliminate the aux vdev wart by integrating all vdevs 6905 * into the root vdev tree. 6906 */ 6907 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6908 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6909 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6910 vdev_state_clean(vd); 6911 vdev_config_dirty(vd); 6912 } 6913 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6914 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6915 } 6916 spa_config_exit(spa, SCL_STATE, FTAG); 6917 6918 tx = dmu_tx_create_assigned(dp, txg); 6919 6920 spa->spa_sync_starttime = gethrtime(); 6921#ifdef illumos 6922 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6923 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6924#else /* !illumos */ 6925#ifdef _KERNEL 6926 callout_schedule(&spa->spa_deadman_cycid, 6927 hz * spa->spa_deadman_synctime / NANOSEC); 6928#endif 6929#endif /* illumos */ 6930 6931 /* 6932 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6933 * set spa_deflate if we have no raid-z vdevs. 6934 */ 6935 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6936 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6937 int i; 6938 6939 for (i = 0; i < rvd->vdev_children; i++) { 6940 vd = rvd->vdev_child[i]; 6941 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6942 break; 6943 } 6944 if (i == rvd->vdev_children) { 6945 spa->spa_deflate = TRUE; 6946 VERIFY(0 == zap_add(spa->spa_meta_objset, 6947 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6948 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6949 } 6950 } 6951 6952 /* 6953 * Set the top-level vdev's max queue depth. Evaluate each 6954 * top-level's async write queue depth in case it changed. 6955 * The max queue depth will not change in the middle of syncing 6956 * out this txg. 6957 */ 6958 uint64_t queue_depth_total = 0; 6959 for (int c = 0; c < rvd->vdev_children; c++) { 6960 vdev_t *tvd = rvd->vdev_child[c]; 6961 metaslab_group_t *mg = tvd->vdev_mg; 6962 6963 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 6964 !metaslab_group_initialized(mg)) 6965 continue; 6966 6967 /* 6968 * It is safe to do a lock-free check here because only async 6969 * allocations look at mg_max_alloc_queue_depth, and async 6970 * allocations all happen from spa_sync(). 6971 */ 6972 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); 6973 mg->mg_max_alloc_queue_depth = max_queue_depth; 6974 queue_depth_total += mg->mg_max_alloc_queue_depth; 6975 } 6976 metaslab_class_t *mc = spa_normal_class(spa); 6977 ASSERT0(refcount_count(&mc->mc_alloc_slots)); 6978 mc->mc_alloc_max_slots = queue_depth_total; 6979 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 6980 6981 ASSERT3U(mc->mc_alloc_max_slots, <=, 6982 max_queue_depth * rvd->vdev_children); 6983 6984 /* 6985 * Iterate to convergence. 6986 */ 6987 do { 6988 int pass = ++spa->spa_sync_pass; 6989 6990 spa_sync_config_object(spa, tx); 6991 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6992 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6993 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6994 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6995 spa_errlog_sync(spa, txg); 6996 dsl_pool_sync(dp, txg); 6997 6998 if (pass < zfs_sync_pass_deferred_free) { 6999 spa_sync_frees(spa, free_bpl, tx); 7000 } else { 7001 /* 7002 * We can not defer frees in pass 1, because 7003 * we sync the deferred frees later in pass 1. 7004 */ 7005 ASSERT3U(pass, >, 1); 7006 bplist_iterate(free_bpl, bpobj_enqueue_cb, 7007 &spa->spa_deferred_bpobj, tx); 7008 } 7009 7010 ddt_sync(spa, txg); 7011 dsl_scan_sync(dp, tx); 7012 7013 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 7014 vdev_sync(vd, txg); 7015 7016 if (pass == 1) { 7017 spa_sync_upgrades(spa, tx); 7018 ASSERT3U(txg, >=, 7019 spa->spa_uberblock.ub_rootbp.blk_birth); 7020 /* 7021 * Note: We need to check if the MOS is dirty 7022 * because we could have marked the MOS dirty 7023 * without updating the uberblock (e.g. if we 7024 * have sync tasks but no dirty user data). We 7025 * need to check the uberblock's rootbp because 7026 * it is updated if we have synced out dirty 7027 * data (though in this case the MOS will most 7028 * likely also be dirty due to second order 7029 * effects, we don't want to rely on that here). 7030 */ 7031 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 7032 !dmu_objset_is_dirty(mos, txg)) { 7033 /* 7034 * Nothing changed on the first pass, 7035 * therefore this TXG is a no-op. Avoid 7036 * syncing deferred frees, so that we 7037 * can keep this TXG as a no-op. 7038 */ 7039 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 7040 txg)); 7041 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7042 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 7043 break; 7044 } 7045 spa_sync_deferred_frees(spa, tx); 7046 } 7047 7048 } while (dmu_objset_is_dirty(mos, txg)); 7049 7050 if (!list_is_empty(&spa->spa_config_dirty_list)) { 7051 /* 7052 * Make sure that the number of ZAPs for all the vdevs matches 7053 * the number of ZAPs in the per-vdev ZAP list. This only gets 7054 * called if the config is dirty; otherwise there may be 7055 * outstanding AVZ operations that weren't completed in 7056 * spa_sync_config_object. 7057 */ 7058 uint64_t all_vdev_zap_entry_count; 7059 ASSERT0(zap_count(spa->spa_meta_objset, 7060 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 7061 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 7062 all_vdev_zap_entry_count); 7063 } 7064 7065 /* 7066 * Rewrite the vdev configuration (which includes the uberblock) 7067 * to commit the transaction group. 7068 * 7069 * If there are no dirty vdevs, we sync the uberblock to a few 7070 * random top-level vdevs that are known to be visible in the 7071 * config cache (see spa_vdev_add() for a complete description). 7072 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 7073 */ 7074 for (;;) { 7075 /* 7076 * We hold SCL_STATE to prevent vdev open/close/etc. 7077 * while we're attempting to write the vdev labels. 7078 */ 7079 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7080 7081 if (list_is_empty(&spa->spa_config_dirty_list)) { 7082 vdev_t *svd[SPA_DVAS_PER_BP]; 7083 int svdcount = 0; 7084 int children = rvd->vdev_children; 7085 int c0 = spa_get_random(children); 7086 7087 for (int c = 0; c < children; c++) { 7088 vd = rvd->vdev_child[(c0 + c) % children]; 7089 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 7090 continue; 7091 svd[svdcount++] = vd; 7092 if (svdcount == SPA_DVAS_PER_BP) 7093 break; 7094 } 7095 error = vdev_config_sync(svd, svdcount, txg); 7096 } else { 7097 error = vdev_config_sync(rvd->vdev_child, 7098 rvd->vdev_children, txg); 7099 } 7100 7101 if (error == 0) 7102 spa->spa_last_synced_guid = rvd->vdev_guid; 7103 7104 spa_config_exit(spa, SCL_STATE, FTAG); 7105 7106 if (error == 0) 7107 break; 7108 zio_suspend(spa, NULL); 7109 zio_resume_wait(spa); 7110 } 7111 dmu_tx_commit(tx); 7112 7113#ifdef illumos 7114 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 7115#else /* !illumos */ 7116#ifdef _KERNEL 7117 callout_drain(&spa->spa_deadman_cycid); 7118#endif 7119#endif /* illumos */ 7120 7121 /* 7122 * Clear the dirty config list. 7123 */ 7124 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 7125 vdev_config_clean(vd); 7126 7127 /* 7128 * Now that the new config has synced transactionally, 7129 * let it become visible to the config cache. 7130 */ 7131 if (spa->spa_config_syncing != NULL) { 7132 spa_config_set(spa, spa->spa_config_syncing); 7133 spa->spa_config_txg = txg; 7134 spa->spa_config_syncing = NULL; 7135 } 7136 7137 dsl_pool_sync_done(dp, txg); 7138 7139 mutex_enter(&spa->spa_alloc_lock); 7140 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 7141 mutex_exit(&spa->spa_alloc_lock); 7142 7143 /* 7144 * Update usable space statistics. 7145 */ 7146 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 7147 vdev_sync_done(vd, txg); 7148 7149 spa_update_dspace(spa); 7150 7151 /* 7152 * It had better be the case that we didn't dirty anything 7153 * since vdev_config_sync(). 7154 */ 7155 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 7156 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7157 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 7158 7159 spa->spa_sync_pass = 0; 7160 7161 /* 7162 * Update the last synced uberblock here. We want to do this at 7163 * the end of spa_sync() so that consumers of spa_last_synced_txg() 7164 * will be guaranteed that all the processing associated with 7165 * that txg has been completed. 7166 */ 7167 spa->spa_ubsync = spa->spa_uberblock; 7168 spa_config_exit(spa, SCL_CONFIG, FTAG); 7169 7170 spa_handle_ignored_writes(spa); 7171 7172 /* 7173 * If any async tasks have been requested, kick them off. 7174 */ 7175 spa_async_dispatch(spa); 7176 spa_async_dispatch_vd(spa); 7177} 7178 7179/* 7180 * Sync all pools. We don't want to hold the namespace lock across these 7181 * operations, so we take a reference on the spa_t and drop the lock during the 7182 * sync. 7183 */ 7184void 7185spa_sync_allpools(void) 7186{ 7187 spa_t *spa = NULL; 7188 mutex_enter(&spa_namespace_lock); 7189 while ((spa = spa_next(spa)) != NULL) { 7190 if (spa_state(spa) != POOL_STATE_ACTIVE || 7191 !spa_writeable(spa) || spa_suspended(spa)) 7192 continue; 7193 spa_open_ref(spa, FTAG); 7194 mutex_exit(&spa_namespace_lock); 7195 txg_wait_synced(spa_get_dsl(spa), 0); 7196 mutex_enter(&spa_namespace_lock); 7197 spa_close(spa, FTAG); 7198 } 7199 mutex_exit(&spa_namespace_lock); 7200} 7201 7202/* 7203 * ========================================================================== 7204 * Miscellaneous routines 7205 * ========================================================================== 7206 */ 7207 7208/* 7209 * Remove all pools in the system. 7210 */ 7211void 7212spa_evict_all(void) 7213{ 7214 spa_t *spa; 7215 7216 /* 7217 * Remove all cached state. All pools should be closed now, 7218 * so every spa in the AVL tree should be unreferenced. 7219 */ 7220 mutex_enter(&spa_namespace_lock); 7221 while ((spa = spa_next(NULL)) != NULL) { 7222 /* 7223 * Stop async tasks. The async thread may need to detach 7224 * a device that's been replaced, which requires grabbing 7225 * spa_namespace_lock, so we must drop it here. 7226 */ 7227 spa_open_ref(spa, FTAG); 7228 mutex_exit(&spa_namespace_lock); 7229 spa_async_suspend(spa); 7230 mutex_enter(&spa_namespace_lock); 7231 spa_close(spa, FTAG); 7232 7233 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 7234 spa_unload(spa); 7235 spa_deactivate(spa); 7236 } 7237 spa_remove(spa); 7238 } 7239 mutex_exit(&spa_namespace_lock); 7240} 7241 7242vdev_t * 7243spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 7244{ 7245 vdev_t *vd; 7246 int i; 7247 7248 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 7249 return (vd); 7250 7251 if (aux) { 7252 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 7253 vd = spa->spa_l2cache.sav_vdevs[i]; 7254 if (vd->vdev_guid == guid) 7255 return (vd); 7256 } 7257 7258 for (i = 0; i < spa->spa_spares.sav_count; i++) { 7259 vd = spa->spa_spares.sav_vdevs[i]; 7260 if (vd->vdev_guid == guid) 7261 return (vd); 7262 } 7263 } 7264 7265 return (NULL); 7266} 7267 7268void 7269spa_upgrade(spa_t *spa, uint64_t version) 7270{ 7271 ASSERT(spa_writeable(spa)); 7272 7273 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 7274 7275 /* 7276 * This should only be called for a non-faulted pool, and since a 7277 * future version would result in an unopenable pool, this shouldn't be 7278 * possible. 7279 */ 7280 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 7281 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 7282 7283 spa->spa_uberblock.ub_version = version; 7284 vdev_config_dirty(spa->spa_root_vdev); 7285 7286 spa_config_exit(spa, SCL_ALL, FTAG); 7287 7288 txg_wait_synced(spa_get_dsl(spa), 0); 7289} 7290 7291boolean_t 7292spa_has_spare(spa_t *spa, uint64_t guid) 7293{ 7294 int i; 7295 uint64_t spareguid; 7296 spa_aux_vdev_t *sav = &spa->spa_spares; 7297 7298 for (i = 0; i < sav->sav_count; i++) 7299 if (sav->sav_vdevs[i]->vdev_guid == guid) 7300 return (B_TRUE); 7301 7302 for (i = 0; i < sav->sav_npending; i++) { 7303 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 7304 &spareguid) == 0 && spareguid == guid) 7305 return (B_TRUE); 7306 } 7307 7308 return (B_FALSE); 7309} 7310 7311/* 7312 * Check if a pool has an active shared spare device. 7313 * Note: reference count of an active spare is 2, as a spare and as a replace 7314 */ 7315static boolean_t 7316spa_has_active_shared_spare(spa_t *spa) 7317{ 7318 int i, refcnt; 7319 uint64_t pool; 7320 spa_aux_vdev_t *sav = &spa->spa_spares; 7321 7322 for (i = 0; i < sav->sav_count; i++) { 7323 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 7324 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 7325 refcnt > 2) 7326 return (B_TRUE); 7327 } 7328 7329 return (B_FALSE); 7330} 7331 7332static sysevent_t * 7333spa_event_create(spa_t *spa, vdev_t *vd, const char *name) 7334{ 7335 sysevent_t *ev = NULL; 7336#ifdef _KERNEL 7337 sysevent_attr_list_t *attr = NULL; 7338 sysevent_value_t value; 7339 7340 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7341 SE_SLEEP); 7342 ASSERT(ev != NULL); 7343 7344 value.value_type = SE_DATA_TYPE_STRING; 7345 value.value.sv_string = spa_name(spa); 7346 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7347 goto done; 7348 7349 value.value_type = SE_DATA_TYPE_UINT64; 7350 value.value.sv_uint64 = spa_guid(spa); 7351 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7352 goto done; 7353 7354 if (vd) { 7355 value.value_type = SE_DATA_TYPE_UINT64; 7356 value.value.sv_uint64 = vd->vdev_guid; 7357 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7358 SE_SLEEP) != 0) 7359 goto done; 7360 7361 if (vd->vdev_path) { 7362 value.value_type = SE_DATA_TYPE_STRING; 7363 value.value.sv_string = vd->vdev_path; 7364 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7365 &value, SE_SLEEP) != 0) 7366 goto done; 7367 } 7368 } 7369 7370 if (sysevent_attach_attributes(ev, attr) != 0) 7371 goto done; 7372 attr = NULL; 7373 7374done: 7375 if (attr) 7376 sysevent_free_attr(attr); 7377 7378#endif 7379 return (ev); 7380} 7381 7382static void 7383spa_event_post(sysevent_t *ev) 7384{ 7385#ifdef _KERNEL 7386 sysevent_id_t eid; 7387 7388 (void) log_sysevent(ev, SE_SLEEP, &eid); 7389 sysevent_free(ev); 7390#endif 7391} 7392 7393/* 7394 * Post a sysevent corresponding to the given event. The 'name' must be one of 7395 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7396 * filled in from the spa and (optionally) the vdev. This doesn't do anything 7397 * in the userland libzpool, as we don't want consumers to misinterpret ztest 7398 * or zdb as real changes. 7399 */ 7400void 7401spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 7402{ 7403 spa_event_post(spa_event_create(spa, vd, name)); 7404} 7405