spa.c revision 297709
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2013 Saso Kiselkov. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 */ 31 32/* 33 * SPA: Storage Pool Allocator 34 * 35 * This file contains all the routines used when modifying on-disk SPA state. 36 * This includes opening, importing, destroying, exporting a pool, and syncing a 37 * pool. 38 */ 39 40#include <sys/zfs_context.h> 41#include <sys/fm/fs/zfs.h> 42#include <sys/spa_impl.h> 43#include <sys/zio.h> 44#include <sys/zio_checksum.h> 45#include <sys/dmu.h> 46#include <sys/dmu_tx.h> 47#include <sys/zap.h> 48#include <sys/zil.h> 49#include <sys/ddt.h> 50#include <sys/vdev_impl.h> 51#include <sys/metaslab.h> 52#include <sys/metaslab_impl.h> 53#include <sys/uberblock_impl.h> 54#include <sys/txg.h> 55#include <sys/avl.h> 56#include <sys/dmu_traverse.h> 57#include <sys/dmu_objset.h> 58#include <sys/unique.h> 59#include <sys/dsl_pool.h> 60#include <sys/dsl_dataset.h> 61#include <sys/dsl_dir.h> 62#include <sys/dsl_prop.h> 63#include <sys/dsl_synctask.h> 64#include <sys/fs/zfs.h> 65#include <sys/arc.h> 66#include <sys/callb.h> 67#include <sys/spa_boot.h> 68#include <sys/zfs_ioctl.h> 69#include <sys/dsl_scan.h> 70#include <sys/dmu_send.h> 71#include <sys/dsl_destroy.h> 72#include <sys/dsl_userhold.h> 73#include <sys/zfeature.h> 74#include <sys/zvol.h> 75#include <sys/trim_map.h> 76 77#ifdef _KERNEL 78#include <sys/callb.h> 79#include <sys/cpupart.h> 80#include <sys/zone.h> 81#endif /* _KERNEL */ 82 83#include "zfs_prop.h" 84#include "zfs_comutil.h" 85 86/* Check hostid on import? */ 87static int check_hostid = 1; 88 89/* 90 * The interval, in seconds, at which failed configuration cache file writes 91 * should be retried. 92 */ 93static int zfs_ccw_retry_interval = 300; 94 95SYSCTL_DECL(_vfs_zfs); 96SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 97 "Check hostid on import?"); 98TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 99SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 100 &zfs_ccw_retry_interval, 0, 101 "Configuration cache file write, retry after failure, interval (seconds)"); 102 103typedef enum zti_modes { 104 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 105 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 106 ZTI_MODE_NULL, /* don't create a taskq */ 107 ZTI_NMODES 108} zti_modes_t; 109 110#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 111#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 112#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 113 114#define ZTI_N(n) ZTI_P(n, 1) 115#define ZTI_ONE ZTI_N(1) 116 117typedef struct zio_taskq_info { 118 zti_modes_t zti_mode; 119 uint_t zti_value; 120 uint_t zti_count; 121} zio_taskq_info_t; 122 123static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 124 "issue", "issue_high", "intr", "intr_high" 125}; 126 127/* 128 * This table defines the taskq settings for each ZFS I/O type. When 129 * initializing a pool, we use this table to create an appropriately sized 130 * taskq. Some operations are low volume and therefore have a small, static 131 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 132 * macros. Other operations process a large amount of data; the ZTI_BATCH 133 * macro causes us to create a taskq oriented for throughput. Some operations 134 * are so high frequency and short-lived that the taskq itself can become a a 135 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 136 * additional degree of parallelism specified by the number of threads per- 137 * taskq and the number of taskqs; when dispatching an event in this case, the 138 * particular taskq is chosen at random. 139 * 140 * The different taskq priorities are to handle the different contexts (issue 141 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 142 * need to be handled with minimum delay. 143 */ 144const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 145 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 146 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 147 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 148 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 149 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 150 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 152}; 153 154static void spa_sync_version(void *arg, dmu_tx_t *tx); 155static void spa_sync_props(void *arg, dmu_tx_t *tx); 156static boolean_t spa_has_active_shared_spare(spa_t *spa); 157static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 158 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 159 char **ereport); 160static void spa_vdev_resilver_done(spa_t *spa); 161 162uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 163#ifdef PSRSET_BIND 164id_t zio_taskq_psrset_bind = PS_NONE; 165#endif 166#ifdef SYSDC 167boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 168#endif 169uint_t zio_taskq_basedc = 80; /* base duty cycle */ 170 171boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 172extern int zfs_sync_pass_deferred_free; 173 174#ifndef illumos 175extern void spa_deadman(void *arg); 176#endif 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( 374 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 375 KM_SLEEP); 376 dsl_dataset_name(ds, strval); 377 dsl_dataset_rele(ds, FTAG); 378 dsl_pool_config_exit(dp, FTAG); 379 } else { 380 strval = NULL; 381 intval = za.za_first_integer; 382 } 383 384 spa_prop_add_list(*nvp, prop, strval, intval, src); 385 386 if (strval != NULL) 387 kmem_free(strval, 388 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 389 390 break; 391 392 case 1: 393 /* string property */ 394 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 395 err = zap_lookup(mos, spa->spa_pool_props_object, 396 za.za_name, 1, za.za_num_integers, strval); 397 if (err) { 398 kmem_free(strval, za.za_num_integers); 399 break; 400 } 401 spa_prop_add_list(*nvp, prop, strval, 0, src); 402 kmem_free(strval, za.za_num_integers); 403 break; 404 405 default: 406 break; 407 } 408 } 409 zap_cursor_fini(&zc); 410 mutex_exit(&spa->spa_props_lock); 411out: 412 if (err && err != ENOENT) { 413 nvlist_free(*nvp); 414 *nvp = NULL; 415 return (err); 416 } 417 418 return (0); 419} 420 421/* 422 * Validate the given pool properties nvlist and modify the list 423 * for the property values to be set. 424 */ 425static int 426spa_prop_validate(spa_t *spa, nvlist_t *props) 427{ 428 nvpair_t *elem; 429 int error = 0, reset_bootfs = 0; 430 uint64_t objnum = 0; 431 boolean_t has_feature = B_FALSE; 432 433 elem = NULL; 434 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 435 uint64_t intval; 436 char *strval, *slash, *check, *fname; 437 const char *propname = nvpair_name(elem); 438 zpool_prop_t prop = zpool_name_to_prop(propname); 439 440 switch (prop) { 441 case ZPROP_INVAL: 442 if (!zpool_prop_feature(propname)) { 443 error = SET_ERROR(EINVAL); 444 break; 445 } 446 447 /* 448 * Sanitize the input. 449 */ 450 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 451 error = SET_ERROR(EINVAL); 452 break; 453 } 454 455 if (nvpair_value_uint64(elem, &intval) != 0) { 456 error = SET_ERROR(EINVAL); 457 break; 458 } 459 460 if (intval != 0) { 461 error = SET_ERROR(EINVAL); 462 break; 463 } 464 465 fname = strchr(propname, '@') + 1; 466 if (zfeature_lookup_name(fname, NULL) != 0) { 467 error = SET_ERROR(EINVAL); 468 break; 469 } 470 471 has_feature = B_TRUE; 472 break; 473 474 case ZPOOL_PROP_VERSION: 475 error = nvpair_value_uint64(elem, &intval); 476 if (!error && 477 (intval < spa_version(spa) || 478 intval > SPA_VERSION_BEFORE_FEATURES || 479 has_feature)) 480 error = SET_ERROR(EINVAL); 481 break; 482 483 case ZPOOL_PROP_DELEGATION: 484 case ZPOOL_PROP_AUTOREPLACE: 485 case ZPOOL_PROP_LISTSNAPS: 486 case ZPOOL_PROP_AUTOEXPAND: 487 error = nvpair_value_uint64(elem, &intval); 488 if (!error && intval > 1) 489 error = SET_ERROR(EINVAL); 490 break; 491 492 case ZPOOL_PROP_BOOTFS: 493 /* 494 * If the pool version is less than SPA_VERSION_BOOTFS, 495 * or the pool is still being created (version == 0), 496 * the bootfs property cannot be set. 497 */ 498 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 499 error = SET_ERROR(ENOTSUP); 500 break; 501 } 502 503 /* 504 * Make sure the vdev config is bootable 505 */ 506 if (!vdev_is_bootable(spa->spa_root_vdev)) { 507 error = SET_ERROR(ENOTSUP); 508 break; 509 } 510 511 reset_bootfs = 1; 512 513 error = nvpair_value_string(elem, &strval); 514 515 if (!error) { 516 objset_t *os; 517 uint64_t propval; 518 519 if (strval == NULL || strval[0] == '\0') { 520 objnum = zpool_prop_default_numeric( 521 ZPOOL_PROP_BOOTFS); 522 break; 523 } 524 525 if (error = dmu_objset_hold(strval, FTAG, &os)) 526 break; 527 528 /* 529 * Must be ZPL, and its property settings 530 * must be supported by GRUB (compression 531 * is not gzip, and large blocks are not used). 532 */ 533 534 if (dmu_objset_type(os) != DMU_OST_ZFS) { 535 error = SET_ERROR(ENOTSUP); 536 } else if ((error = 537 dsl_prop_get_int_ds(dmu_objset_ds(os), 538 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 539 &propval)) == 0 && 540 !BOOTFS_COMPRESS_VALID(propval)) { 541 error = SET_ERROR(ENOTSUP); 542 } else if ((error = 543 dsl_prop_get_int_ds(dmu_objset_ds(os), 544 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 545 &propval)) == 0 && 546 propval > SPA_OLD_MAXBLOCKSIZE) { 547 error = SET_ERROR(ENOTSUP); 548 } else { 549 objnum = dmu_objset_id(os); 550 } 551 dmu_objset_rele(os, FTAG); 552 } 553 break; 554 555 case ZPOOL_PROP_FAILUREMODE: 556 error = nvpair_value_uint64(elem, &intval); 557 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 558 intval > ZIO_FAILURE_MODE_PANIC)) 559 error = SET_ERROR(EINVAL); 560 561 /* 562 * This is a special case which only occurs when 563 * the pool has completely failed. This allows 564 * the user to change the in-core failmode property 565 * without syncing it out to disk (I/Os might 566 * currently be blocked). We do this by returning 567 * EIO to the caller (spa_prop_set) to trick it 568 * into thinking we encountered a property validation 569 * error. 570 */ 571 if (!error && spa_suspended(spa)) { 572 spa->spa_failmode = intval; 573 error = SET_ERROR(EIO); 574 } 575 break; 576 577 case ZPOOL_PROP_CACHEFILE: 578 if ((error = nvpair_value_string(elem, &strval)) != 0) 579 break; 580 581 if (strval[0] == '\0') 582 break; 583 584 if (strcmp(strval, "none") == 0) 585 break; 586 587 if (strval[0] != '/') { 588 error = SET_ERROR(EINVAL); 589 break; 590 } 591 592 slash = strrchr(strval, '/'); 593 ASSERT(slash != NULL); 594 595 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 596 strcmp(slash, "/..") == 0) 597 error = SET_ERROR(EINVAL); 598 break; 599 600 case ZPOOL_PROP_COMMENT: 601 if ((error = nvpair_value_string(elem, &strval)) != 0) 602 break; 603 for (check = strval; *check != '\0'; check++) { 604 /* 605 * The kernel doesn't have an easy isprint() 606 * check. For this kernel check, we merely 607 * check ASCII apart from DEL. Fix this if 608 * there is an easy-to-use kernel isprint(). 609 */ 610 if (*check >= 0x7f) { 611 error = SET_ERROR(EINVAL); 612 break; 613 } 614 } 615 if (strlen(strval) > ZPROP_MAX_COMMENT) 616 error = E2BIG; 617 break; 618 619 case ZPOOL_PROP_DEDUPDITTO: 620 if (spa_version(spa) < SPA_VERSION_DEDUP) 621 error = SET_ERROR(ENOTSUP); 622 else 623 error = nvpair_value_uint64(elem, &intval); 624 if (error == 0 && 625 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 626 error = SET_ERROR(EINVAL); 627 break; 628 } 629 630 if (error) 631 break; 632 } 633 634 if (!error && reset_bootfs) { 635 error = nvlist_remove(props, 636 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 637 638 if (!error) { 639 error = nvlist_add_uint64(props, 640 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 641 } 642 } 643 644 return (error); 645} 646 647void 648spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 649{ 650 char *cachefile; 651 spa_config_dirent_t *dp; 652 653 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 654 &cachefile) != 0) 655 return; 656 657 dp = kmem_alloc(sizeof (spa_config_dirent_t), 658 KM_SLEEP); 659 660 if (cachefile[0] == '\0') 661 dp->scd_path = spa_strdup(spa_config_path); 662 else if (strcmp(cachefile, "none") == 0) 663 dp->scd_path = NULL; 664 else 665 dp->scd_path = spa_strdup(cachefile); 666 667 list_insert_head(&spa->spa_config_list, dp); 668 if (need_sync) 669 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 670} 671 672int 673spa_prop_set(spa_t *spa, nvlist_t *nvp) 674{ 675 int error; 676 nvpair_t *elem = NULL; 677 boolean_t need_sync = B_FALSE; 678 679 if ((error = spa_prop_validate(spa, nvp)) != 0) 680 return (error); 681 682 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 683 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 684 685 if (prop == ZPOOL_PROP_CACHEFILE || 686 prop == ZPOOL_PROP_ALTROOT || 687 prop == ZPOOL_PROP_READONLY) 688 continue; 689 690 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 691 uint64_t ver; 692 693 if (prop == ZPOOL_PROP_VERSION) { 694 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 695 } else { 696 ASSERT(zpool_prop_feature(nvpair_name(elem))); 697 ver = SPA_VERSION_FEATURES; 698 need_sync = B_TRUE; 699 } 700 701 /* Save time if the version is already set. */ 702 if (ver == spa_version(spa)) 703 continue; 704 705 /* 706 * In addition to the pool directory object, we might 707 * create the pool properties object, the features for 708 * read object, the features for write object, or the 709 * feature descriptions object. 710 */ 711 error = dsl_sync_task(spa->spa_name, NULL, 712 spa_sync_version, &ver, 713 6, ZFS_SPACE_CHECK_RESERVED); 714 if (error) 715 return (error); 716 continue; 717 } 718 719 need_sync = B_TRUE; 720 break; 721 } 722 723 if (need_sync) { 724 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 725 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 726 } 727 728 return (0); 729} 730 731/* 732 * If the bootfs property value is dsobj, clear it. 733 */ 734void 735spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 736{ 737 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 738 VERIFY(zap_remove(spa->spa_meta_objset, 739 spa->spa_pool_props_object, 740 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 741 spa->spa_bootfs = 0; 742 } 743} 744 745/*ARGSUSED*/ 746static int 747spa_change_guid_check(void *arg, dmu_tx_t *tx) 748{ 749 uint64_t *newguid = arg; 750 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 751 vdev_t *rvd = spa->spa_root_vdev; 752 uint64_t vdev_state; 753 754 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 755 vdev_state = rvd->vdev_state; 756 spa_config_exit(spa, SCL_STATE, FTAG); 757 758 if (vdev_state != VDEV_STATE_HEALTHY) 759 return (SET_ERROR(ENXIO)); 760 761 ASSERT3U(spa_guid(spa), !=, *newguid); 762 763 return (0); 764} 765 766static void 767spa_change_guid_sync(void *arg, dmu_tx_t *tx) 768{ 769 uint64_t *newguid = arg; 770 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 771 uint64_t oldguid; 772 vdev_t *rvd = spa->spa_root_vdev; 773 774 oldguid = spa_guid(spa); 775 776 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 777 rvd->vdev_guid = *newguid; 778 rvd->vdev_guid_sum += (*newguid - oldguid); 779 vdev_config_dirty(rvd); 780 spa_config_exit(spa, SCL_STATE, FTAG); 781 782 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 783 oldguid, *newguid); 784} 785 786/* 787 * Change the GUID for the pool. This is done so that we can later 788 * re-import a pool built from a clone of our own vdevs. We will modify 789 * the root vdev's guid, our own pool guid, and then mark all of our 790 * vdevs dirty. Note that we must make sure that all our vdevs are 791 * online when we do this, or else any vdevs that weren't present 792 * would be orphaned from our pool. We are also going to issue a 793 * sysevent to update any watchers. 794 */ 795int 796spa_change_guid(spa_t *spa) 797{ 798 int error; 799 uint64_t guid; 800 801 mutex_enter(&spa->spa_vdev_top_lock); 802 mutex_enter(&spa_namespace_lock); 803 guid = spa_generate_guid(NULL); 804 805 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 806 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 807 808 if (error == 0) { 809 spa_config_sync(spa, B_FALSE, B_TRUE); 810 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 811 } 812 813 mutex_exit(&spa_namespace_lock); 814 mutex_exit(&spa->spa_vdev_top_lock); 815 816 return (error); 817} 818 819/* 820 * ========================================================================== 821 * SPA state manipulation (open/create/destroy/import/export) 822 * ========================================================================== 823 */ 824 825static int 826spa_error_entry_compare(const void *a, const void *b) 827{ 828 spa_error_entry_t *sa = (spa_error_entry_t *)a; 829 spa_error_entry_t *sb = (spa_error_entry_t *)b; 830 int ret; 831 832 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 833 sizeof (zbookmark_phys_t)); 834 835 if (ret < 0) 836 return (-1); 837 else if (ret > 0) 838 return (1); 839 else 840 return (0); 841} 842 843/* 844 * Utility function which retrieves copies of the current logs and 845 * re-initializes them in the process. 846 */ 847void 848spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 849{ 850 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 851 852 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 853 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 854 855 avl_create(&spa->spa_errlist_scrub, 856 spa_error_entry_compare, sizeof (spa_error_entry_t), 857 offsetof(spa_error_entry_t, se_avl)); 858 avl_create(&spa->spa_errlist_last, 859 spa_error_entry_compare, sizeof (spa_error_entry_t), 860 offsetof(spa_error_entry_t, se_avl)); 861} 862 863static void 864spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 865{ 866 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 867 enum zti_modes mode = ztip->zti_mode; 868 uint_t value = ztip->zti_value; 869 uint_t count = ztip->zti_count; 870 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 871 char name[32]; 872 uint_t flags = 0; 873 boolean_t batch = B_FALSE; 874 875 if (mode == ZTI_MODE_NULL) { 876 tqs->stqs_count = 0; 877 tqs->stqs_taskq = NULL; 878 return; 879 } 880 881 ASSERT3U(count, >, 0); 882 883 tqs->stqs_count = count; 884 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 885 886 switch (mode) { 887 case ZTI_MODE_FIXED: 888 ASSERT3U(value, >=, 1); 889 value = MAX(value, 1); 890 break; 891 892 case ZTI_MODE_BATCH: 893 batch = B_TRUE; 894 flags |= TASKQ_THREADS_CPU_PCT; 895 value = zio_taskq_batch_pct; 896 break; 897 898 default: 899 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 900 "spa_activate()", 901 zio_type_name[t], zio_taskq_types[q], mode, value); 902 break; 903 } 904 905 for (uint_t i = 0; i < count; i++) { 906 taskq_t *tq; 907 908 if (count > 1) { 909 (void) snprintf(name, sizeof (name), "%s_%s_%u", 910 zio_type_name[t], zio_taskq_types[q], i); 911 } else { 912 (void) snprintf(name, sizeof (name), "%s_%s", 913 zio_type_name[t], zio_taskq_types[q]); 914 } 915 916#ifdef SYSDC 917 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 918 if (batch) 919 flags |= TASKQ_DC_BATCH; 920 921 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 922 spa->spa_proc, zio_taskq_basedc, flags); 923 } else { 924#endif 925 pri_t pri = maxclsyspri; 926 /* 927 * The write issue taskq can be extremely CPU 928 * intensive. Run it at slightly lower priority 929 * than the other taskqs. 930 */ 931 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 932 pri++; 933 934 tq = taskq_create_proc(name, value, pri, 50, 935 INT_MAX, spa->spa_proc, flags); 936#ifdef SYSDC 937 } 938#endif 939 940 tqs->stqs_taskq[i] = tq; 941 } 942} 943 944static void 945spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 946{ 947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 948 949 if (tqs->stqs_taskq == NULL) { 950 ASSERT0(tqs->stqs_count); 951 return; 952 } 953 954 for (uint_t i = 0; i < tqs->stqs_count; i++) { 955 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 956 taskq_destroy(tqs->stqs_taskq[i]); 957 } 958 959 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 960 tqs->stqs_taskq = NULL; 961} 962 963/* 964 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 965 * Note that a type may have multiple discrete taskqs to avoid lock contention 966 * on the taskq itself. In that case we choose which taskq at random by using 967 * the low bits of gethrtime(). 968 */ 969void 970spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 971 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 972{ 973 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 974 taskq_t *tq; 975 976 ASSERT3P(tqs->stqs_taskq, !=, NULL); 977 ASSERT3U(tqs->stqs_count, !=, 0); 978 979 if (tqs->stqs_count == 1) { 980 tq = tqs->stqs_taskq[0]; 981 } else { 982#ifdef _KERNEL 983 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 984#else 985 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 986#endif 987 } 988 989 taskq_dispatch_ent(tq, func, arg, flags, ent); 990} 991 992static void 993spa_create_zio_taskqs(spa_t *spa) 994{ 995 for (int t = 0; t < ZIO_TYPES; t++) { 996 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 997 spa_taskqs_init(spa, t, q); 998 } 999 } 1000} 1001 1002#ifdef _KERNEL 1003#ifdef SPA_PROCESS 1004static void 1005spa_thread(void *arg) 1006{ 1007 callb_cpr_t cprinfo; 1008 1009 spa_t *spa = arg; 1010 user_t *pu = PTOU(curproc); 1011 1012 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1013 spa->spa_name); 1014 1015 ASSERT(curproc != &p0); 1016 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1017 "zpool-%s", spa->spa_name); 1018 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1019 1020#ifdef PSRSET_BIND 1021 /* bind this thread to the requested psrset */ 1022 if (zio_taskq_psrset_bind != PS_NONE) { 1023 pool_lock(); 1024 mutex_enter(&cpu_lock); 1025 mutex_enter(&pidlock); 1026 mutex_enter(&curproc->p_lock); 1027 1028 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1029 0, NULL, NULL) == 0) { 1030 curthread->t_bind_pset = zio_taskq_psrset_bind; 1031 } else { 1032 cmn_err(CE_WARN, 1033 "Couldn't bind process for zfs pool \"%s\" to " 1034 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1035 } 1036 1037 mutex_exit(&curproc->p_lock); 1038 mutex_exit(&pidlock); 1039 mutex_exit(&cpu_lock); 1040 pool_unlock(); 1041 } 1042#endif 1043 1044#ifdef SYSDC 1045 if (zio_taskq_sysdc) { 1046 sysdc_thread_enter(curthread, 100, 0); 1047 } 1048#endif 1049 1050 spa->spa_proc = curproc; 1051 spa->spa_did = curthread->t_did; 1052 1053 spa_create_zio_taskqs(spa); 1054 1055 mutex_enter(&spa->spa_proc_lock); 1056 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1057 1058 spa->spa_proc_state = SPA_PROC_ACTIVE; 1059 cv_broadcast(&spa->spa_proc_cv); 1060 1061 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1062 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1063 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1064 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1065 1066 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1067 spa->spa_proc_state = SPA_PROC_GONE; 1068 spa->spa_proc = &p0; 1069 cv_broadcast(&spa->spa_proc_cv); 1070 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1071 1072 mutex_enter(&curproc->p_lock); 1073 lwp_exit(); 1074} 1075#endif /* SPA_PROCESS */ 1076#endif 1077 1078/* 1079 * Activate an uninitialized pool. 1080 */ 1081static void 1082spa_activate(spa_t *spa, int mode) 1083{ 1084 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1085 1086 spa->spa_state = POOL_STATE_ACTIVE; 1087 spa->spa_mode = mode; 1088 1089 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1090 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1091 1092 /* Try to create a covering process */ 1093 mutex_enter(&spa->spa_proc_lock); 1094 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1095 ASSERT(spa->spa_proc == &p0); 1096 spa->spa_did = 0; 1097 1098#ifdef SPA_PROCESS 1099 /* Only create a process if we're going to be around a while. */ 1100 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1101 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1102 NULL, 0) == 0) { 1103 spa->spa_proc_state = SPA_PROC_CREATED; 1104 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1105 cv_wait(&spa->spa_proc_cv, 1106 &spa->spa_proc_lock); 1107 } 1108 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1109 ASSERT(spa->spa_proc != &p0); 1110 ASSERT(spa->spa_did != 0); 1111 } else { 1112#ifdef _KERNEL 1113 cmn_err(CE_WARN, 1114 "Couldn't create process for zfs pool \"%s\"\n", 1115 spa->spa_name); 1116#endif 1117 } 1118 } 1119#endif /* SPA_PROCESS */ 1120 mutex_exit(&spa->spa_proc_lock); 1121 1122 /* If we didn't create a process, we need to create our taskqs. */ 1123 ASSERT(spa->spa_proc == &p0); 1124 if (spa->spa_proc == &p0) { 1125 spa_create_zio_taskqs(spa); 1126 } 1127 1128 /* 1129 * Start TRIM thread. 1130 */ 1131 trim_thread_create(spa); 1132 1133 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1134 offsetof(vdev_t, vdev_config_dirty_node)); 1135 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1136 offsetof(objset_t, os_evicting_node)); 1137 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1138 offsetof(vdev_t, vdev_state_dirty_node)); 1139 1140 txg_list_create(&spa->spa_vdev_txg_list, 1141 offsetof(struct vdev, vdev_txg_node)); 1142 1143 avl_create(&spa->spa_errlist_scrub, 1144 spa_error_entry_compare, sizeof (spa_error_entry_t), 1145 offsetof(spa_error_entry_t, se_avl)); 1146 avl_create(&spa->spa_errlist_last, 1147 spa_error_entry_compare, sizeof (spa_error_entry_t), 1148 offsetof(spa_error_entry_t, se_avl)); 1149} 1150 1151/* 1152 * Opposite of spa_activate(). 1153 */ 1154static void 1155spa_deactivate(spa_t *spa) 1156{ 1157 ASSERT(spa->spa_sync_on == B_FALSE); 1158 ASSERT(spa->spa_dsl_pool == NULL); 1159 ASSERT(spa->spa_root_vdev == NULL); 1160 ASSERT(spa->spa_async_zio_root == NULL); 1161 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1162 1163 /* 1164 * Stop TRIM thread in case spa_unload() wasn't called directly 1165 * before spa_deactivate(). 1166 */ 1167 trim_thread_destroy(spa); 1168 1169 spa_evicting_os_wait(spa); 1170 1171 txg_list_destroy(&spa->spa_vdev_txg_list); 1172 1173 list_destroy(&spa->spa_config_dirty_list); 1174 list_destroy(&spa->spa_evicting_os_list); 1175 list_destroy(&spa->spa_state_dirty_list); 1176 1177 for (int t = 0; t < ZIO_TYPES; t++) { 1178 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1179 spa_taskqs_fini(spa, t, q); 1180 } 1181 } 1182 1183 metaslab_class_destroy(spa->spa_normal_class); 1184 spa->spa_normal_class = NULL; 1185 1186 metaslab_class_destroy(spa->spa_log_class); 1187 spa->spa_log_class = NULL; 1188 1189 /* 1190 * If this was part of an import or the open otherwise failed, we may 1191 * still have errors left in the queues. Empty them just in case. 1192 */ 1193 spa_errlog_drain(spa); 1194 1195 avl_destroy(&spa->spa_errlist_scrub); 1196 avl_destroy(&spa->spa_errlist_last); 1197 1198 spa->spa_state = POOL_STATE_UNINITIALIZED; 1199 1200 mutex_enter(&spa->spa_proc_lock); 1201 if (spa->spa_proc_state != SPA_PROC_NONE) { 1202 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1203 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1204 cv_broadcast(&spa->spa_proc_cv); 1205 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1206 ASSERT(spa->spa_proc != &p0); 1207 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1208 } 1209 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1210 spa->spa_proc_state = SPA_PROC_NONE; 1211 } 1212 ASSERT(spa->spa_proc == &p0); 1213 mutex_exit(&spa->spa_proc_lock); 1214 1215#ifdef SPA_PROCESS 1216 /* 1217 * We want to make sure spa_thread() has actually exited the ZFS 1218 * module, so that the module can't be unloaded out from underneath 1219 * it. 1220 */ 1221 if (spa->spa_did != 0) { 1222 thread_join(spa->spa_did); 1223 spa->spa_did = 0; 1224 } 1225#endif /* SPA_PROCESS */ 1226} 1227 1228/* 1229 * Verify a pool configuration, and construct the vdev tree appropriately. This 1230 * will create all the necessary vdevs in the appropriate layout, with each vdev 1231 * in the CLOSED state. This will prep the pool before open/creation/import. 1232 * All vdev validation is done by the vdev_alloc() routine. 1233 */ 1234static int 1235spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1236 uint_t id, int atype) 1237{ 1238 nvlist_t **child; 1239 uint_t children; 1240 int error; 1241 1242 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1243 return (error); 1244 1245 if ((*vdp)->vdev_ops->vdev_op_leaf) 1246 return (0); 1247 1248 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1249 &child, &children); 1250 1251 if (error == ENOENT) 1252 return (0); 1253 1254 if (error) { 1255 vdev_free(*vdp); 1256 *vdp = NULL; 1257 return (SET_ERROR(EINVAL)); 1258 } 1259 1260 for (int c = 0; c < children; c++) { 1261 vdev_t *vd; 1262 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1263 atype)) != 0) { 1264 vdev_free(*vdp); 1265 *vdp = NULL; 1266 return (error); 1267 } 1268 } 1269 1270 ASSERT(*vdp != NULL); 1271 1272 return (0); 1273} 1274 1275/* 1276 * Opposite of spa_load(). 1277 */ 1278static void 1279spa_unload(spa_t *spa) 1280{ 1281 int i; 1282 1283 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1284 1285 /* 1286 * Stop TRIM thread. 1287 */ 1288 trim_thread_destroy(spa); 1289 1290 /* 1291 * Stop async tasks. 1292 */ 1293 spa_async_suspend(spa); 1294 1295 /* 1296 * Stop syncing. 1297 */ 1298 if (spa->spa_sync_on) { 1299 txg_sync_stop(spa->spa_dsl_pool); 1300 spa->spa_sync_on = B_FALSE; 1301 } 1302 1303 /* 1304 * Wait for any outstanding async I/O to complete. 1305 */ 1306 if (spa->spa_async_zio_root != NULL) { 1307 for (int i = 0; i < max_ncpus; i++) 1308 (void) zio_wait(spa->spa_async_zio_root[i]); 1309 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1310 spa->spa_async_zio_root = NULL; 1311 } 1312 1313 bpobj_close(&spa->spa_deferred_bpobj); 1314 1315 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1316 1317 /* 1318 * Close all vdevs. 1319 */ 1320 if (spa->spa_root_vdev) 1321 vdev_free(spa->spa_root_vdev); 1322 ASSERT(spa->spa_root_vdev == NULL); 1323 1324 /* 1325 * Close the dsl pool. 1326 */ 1327 if (spa->spa_dsl_pool) { 1328 dsl_pool_close(spa->spa_dsl_pool); 1329 spa->spa_dsl_pool = NULL; 1330 spa->spa_meta_objset = NULL; 1331 } 1332 1333 ddt_unload(spa); 1334 1335 1336 /* 1337 * Drop and purge level 2 cache 1338 */ 1339 spa_l2cache_drop(spa); 1340 1341 for (i = 0; i < spa->spa_spares.sav_count; i++) 1342 vdev_free(spa->spa_spares.sav_vdevs[i]); 1343 if (spa->spa_spares.sav_vdevs) { 1344 kmem_free(spa->spa_spares.sav_vdevs, 1345 spa->spa_spares.sav_count * sizeof (void *)); 1346 spa->spa_spares.sav_vdevs = NULL; 1347 } 1348 if (spa->spa_spares.sav_config) { 1349 nvlist_free(spa->spa_spares.sav_config); 1350 spa->spa_spares.sav_config = NULL; 1351 } 1352 spa->spa_spares.sav_count = 0; 1353 1354 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1355 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1356 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1357 } 1358 if (spa->spa_l2cache.sav_vdevs) { 1359 kmem_free(spa->spa_l2cache.sav_vdevs, 1360 spa->spa_l2cache.sav_count * sizeof (void *)); 1361 spa->spa_l2cache.sav_vdevs = NULL; 1362 } 1363 if (spa->spa_l2cache.sav_config) { 1364 nvlist_free(spa->spa_l2cache.sav_config); 1365 spa->spa_l2cache.sav_config = NULL; 1366 } 1367 spa->spa_l2cache.sav_count = 0; 1368 1369 spa->spa_async_suspended = 0; 1370 1371 if (spa->spa_comment != NULL) { 1372 spa_strfree(spa->spa_comment); 1373 spa->spa_comment = NULL; 1374 } 1375 1376 spa_config_exit(spa, SCL_ALL, FTAG); 1377} 1378 1379/* 1380 * Load (or re-load) the current list of vdevs describing the active spares for 1381 * this pool. When this is called, we have some form of basic information in 1382 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1383 * then re-generate a more complete list including status information. 1384 */ 1385static void 1386spa_load_spares(spa_t *spa) 1387{ 1388 nvlist_t **spares; 1389 uint_t nspares; 1390 int i; 1391 vdev_t *vd, *tvd; 1392 1393 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1394 1395 /* 1396 * First, close and free any existing spare vdevs. 1397 */ 1398 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1399 vd = spa->spa_spares.sav_vdevs[i]; 1400 1401 /* Undo the call to spa_activate() below */ 1402 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1403 B_FALSE)) != NULL && tvd->vdev_isspare) 1404 spa_spare_remove(tvd); 1405 vdev_close(vd); 1406 vdev_free(vd); 1407 } 1408 1409 if (spa->spa_spares.sav_vdevs) 1410 kmem_free(spa->spa_spares.sav_vdevs, 1411 spa->spa_spares.sav_count * sizeof (void *)); 1412 1413 if (spa->spa_spares.sav_config == NULL) 1414 nspares = 0; 1415 else 1416 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1417 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1418 1419 spa->spa_spares.sav_count = (int)nspares; 1420 spa->spa_spares.sav_vdevs = NULL; 1421 1422 if (nspares == 0) 1423 return; 1424 1425 /* 1426 * Construct the array of vdevs, opening them to get status in the 1427 * process. For each spare, there is potentially two different vdev_t 1428 * structures associated with it: one in the list of spares (used only 1429 * for basic validation purposes) and one in the active vdev 1430 * configuration (if it's spared in). During this phase we open and 1431 * validate each vdev on the spare list. If the vdev also exists in the 1432 * active configuration, then we also mark this vdev as an active spare. 1433 */ 1434 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1435 KM_SLEEP); 1436 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1437 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1438 VDEV_ALLOC_SPARE) == 0); 1439 ASSERT(vd != NULL); 1440 1441 spa->spa_spares.sav_vdevs[i] = vd; 1442 1443 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1444 B_FALSE)) != NULL) { 1445 if (!tvd->vdev_isspare) 1446 spa_spare_add(tvd); 1447 1448 /* 1449 * We only mark the spare active if we were successfully 1450 * able to load the vdev. Otherwise, importing a pool 1451 * with a bad active spare would result in strange 1452 * behavior, because multiple pool would think the spare 1453 * is actively in use. 1454 * 1455 * There is a vulnerability here to an equally bizarre 1456 * circumstance, where a dead active spare is later 1457 * brought back to life (onlined or otherwise). Given 1458 * the rarity of this scenario, and the extra complexity 1459 * it adds, we ignore the possibility. 1460 */ 1461 if (!vdev_is_dead(tvd)) 1462 spa_spare_activate(tvd); 1463 } 1464 1465 vd->vdev_top = vd; 1466 vd->vdev_aux = &spa->spa_spares; 1467 1468 if (vdev_open(vd) != 0) 1469 continue; 1470 1471 if (vdev_validate_aux(vd) == 0) 1472 spa_spare_add(vd); 1473 } 1474 1475 /* 1476 * Recompute the stashed list of spares, with status information 1477 * this time. 1478 */ 1479 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1480 DATA_TYPE_NVLIST_ARRAY) == 0); 1481 1482 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1483 KM_SLEEP); 1484 for (i = 0; i < spa->spa_spares.sav_count; i++) 1485 spares[i] = vdev_config_generate(spa, 1486 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1487 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1488 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1489 for (i = 0; i < spa->spa_spares.sav_count; i++) 1490 nvlist_free(spares[i]); 1491 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1492} 1493 1494/* 1495 * Load (or re-load) the current list of vdevs describing the active l2cache for 1496 * this pool. When this is called, we have some form of basic information in 1497 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1498 * then re-generate a more complete list including status information. 1499 * Devices which are already active have their details maintained, and are 1500 * not re-opened. 1501 */ 1502static void 1503spa_load_l2cache(spa_t *spa) 1504{ 1505 nvlist_t **l2cache; 1506 uint_t nl2cache; 1507 int i, j, oldnvdevs; 1508 uint64_t guid; 1509 vdev_t *vd, **oldvdevs, **newvdevs; 1510 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1511 1512 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1513 1514 if (sav->sav_config != NULL) { 1515 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1516 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1517 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1518 } else { 1519 nl2cache = 0; 1520 newvdevs = NULL; 1521 } 1522 1523 oldvdevs = sav->sav_vdevs; 1524 oldnvdevs = sav->sav_count; 1525 sav->sav_vdevs = NULL; 1526 sav->sav_count = 0; 1527 1528 /* 1529 * Process new nvlist of vdevs. 1530 */ 1531 for (i = 0; i < nl2cache; i++) { 1532 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1533 &guid) == 0); 1534 1535 newvdevs[i] = NULL; 1536 for (j = 0; j < oldnvdevs; j++) { 1537 vd = oldvdevs[j]; 1538 if (vd != NULL && guid == vd->vdev_guid) { 1539 /* 1540 * Retain previous vdev for add/remove ops. 1541 */ 1542 newvdevs[i] = vd; 1543 oldvdevs[j] = NULL; 1544 break; 1545 } 1546 } 1547 1548 if (newvdevs[i] == NULL) { 1549 /* 1550 * Create new vdev 1551 */ 1552 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1553 VDEV_ALLOC_L2CACHE) == 0); 1554 ASSERT(vd != NULL); 1555 newvdevs[i] = vd; 1556 1557 /* 1558 * Commit this vdev as an l2cache device, 1559 * even if it fails to open. 1560 */ 1561 spa_l2cache_add(vd); 1562 1563 vd->vdev_top = vd; 1564 vd->vdev_aux = sav; 1565 1566 spa_l2cache_activate(vd); 1567 1568 if (vdev_open(vd) != 0) 1569 continue; 1570 1571 (void) vdev_validate_aux(vd); 1572 1573 if (!vdev_is_dead(vd)) 1574 l2arc_add_vdev(spa, vd); 1575 } 1576 } 1577 1578 /* 1579 * Purge vdevs that were dropped 1580 */ 1581 for (i = 0; i < oldnvdevs; i++) { 1582 uint64_t pool; 1583 1584 vd = oldvdevs[i]; 1585 if (vd != NULL) { 1586 ASSERT(vd->vdev_isl2cache); 1587 1588 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1589 pool != 0ULL && l2arc_vdev_present(vd)) 1590 l2arc_remove_vdev(vd); 1591 vdev_clear_stats(vd); 1592 vdev_free(vd); 1593 } 1594 } 1595 1596 if (oldvdevs) 1597 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1598 1599 if (sav->sav_config == NULL) 1600 goto out; 1601 1602 sav->sav_vdevs = newvdevs; 1603 sav->sav_count = (int)nl2cache; 1604 1605 /* 1606 * Recompute the stashed list of l2cache devices, with status 1607 * information this time. 1608 */ 1609 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1610 DATA_TYPE_NVLIST_ARRAY) == 0); 1611 1612 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1613 for (i = 0; i < sav->sav_count; i++) 1614 l2cache[i] = vdev_config_generate(spa, 1615 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1616 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1617 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1618out: 1619 for (i = 0; i < sav->sav_count; i++) 1620 nvlist_free(l2cache[i]); 1621 if (sav->sav_count) 1622 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1623} 1624 1625static int 1626load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1627{ 1628 dmu_buf_t *db; 1629 char *packed = NULL; 1630 size_t nvsize = 0; 1631 int error; 1632 *value = NULL; 1633 1634 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1635 if (error != 0) 1636 return (error); 1637 1638 nvsize = *(uint64_t *)db->db_data; 1639 dmu_buf_rele(db, FTAG); 1640 1641 packed = kmem_alloc(nvsize, KM_SLEEP); 1642 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1643 DMU_READ_PREFETCH); 1644 if (error == 0) 1645 error = nvlist_unpack(packed, nvsize, value, 0); 1646 kmem_free(packed, nvsize); 1647 1648 return (error); 1649} 1650 1651/* 1652 * Checks to see if the given vdev could not be opened, in which case we post a 1653 * sysevent to notify the autoreplace code that the device has been removed. 1654 */ 1655static void 1656spa_check_removed(vdev_t *vd) 1657{ 1658 for (int c = 0; c < vd->vdev_children; c++) 1659 spa_check_removed(vd->vdev_child[c]); 1660 1661 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1662 !vd->vdev_ishole) { 1663 zfs_post_autoreplace(vd->vdev_spa, vd); 1664 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1665 } 1666} 1667 1668/* 1669 * Validate the current config against the MOS config 1670 */ 1671static boolean_t 1672spa_config_valid(spa_t *spa, nvlist_t *config) 1673{ 1674 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1675 nvlist_t *nv; 1676 1677 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1678 1679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1680 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1681 1682 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1683 1684 /* 1685 * If we're doing a normal import, then build up any additional 1686 * diagnostic information about missing devices in this config. 1687 * We'll pass this up to the user for further processing. 1688 */ 1689 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1690 nvlist_t **child, *nv; 1691 uint64_t idx = 0; 1692 1693 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1694 KM_SLEEP); 1695 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1696 1697 for (int c = 0; c < rvd->vdev_children; c++) { 1698 vdev_t *tvd = rvd->vdev_child[c]; 1699 vdev_t *mtvd = mrvd->vdev_child[c]; 1700 1701 if (tvd->vdev_ops == &vdev_missing_ops && 1702 mtvd->vdev_ops != &vdev_missing_ops && 1703 mtvd->vdev_islog) 1704 child[idx++] = vdev_config_generate(spa, mtvd, 1705 B_FALSE, 0); 1706 } 1707 1708 if (idx) { 1709 VERIFY(nvlist_add_nvlist_array(nv, 1710 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1711 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1712 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1713 1714 for (int i = 0; i < idx; i++) 1715 nvlist_free(child[i]); 1716 } 1717 nvlist_free(nv); 1718 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1719 } 1720 1721 /* 1722 * Compare the root vdev tree with the information we have 1723 * from the MOS config (mrvd). Check each top-level vdev 1724 * with the corresponding MOS config top-level (mtvd). 1725 */ 1726 for (int c = 0; c < rvd->vdev_children; c++) { 1727 vdev_t *tvd = rvd->vdev_child[c]; 1728 vdev_t *mtvd = mrvd->vdev_child[c]; 1729 1730 /* 1731 * Resolve any "missing" vdevs in the current configuration. 1732 * If we find that the MOS config has more accurate information 1733 * about the top-level vdev then use that vdev instead. 1734 */ 1735 if (tvd->vdev_ops == &vdev_missing_ops && 1736 mtvd->vdev_ops != &vdev_missing_ops) { 1737 1738 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1739 continue; 1740 1741 /* 1742 * Device specific actions. 1743 */ 1744 if (mtvd->vdev_islog) { 1745 spa_set_log_state(spa, SPA_LOG_CLEAR); 1746 } else { 1747 /* 1748 * XXX - once we have 'readonly' pool 1749 * support we should be able to handle 1750 * missing data devices by transitioning 1751 * the pool to readonly. 1752 */ 1753 continue; 1754 } 1755 1756 /* 1757 * Swap the missing vdev with the data we were 1758 * able to obtain from the MOS config. 1759 */ 1760 vdev_remove_child(rvd, tvd); 1761 vdev_remove_child(mrvd, mtvd); 1762 1763 vdev_add_child(rvd, mtvd); 1764 vdev_add_child(mrvd, tvd); 1765 1766 spa_config_exit(spa, SCL_ALL, FTAG); 1767 vdev_load(mtvd); 1768 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1769 1770 vdev_reopen(rvd); 1771 } else if (mtvd->vdev_islog) { 1772 /* 1773 * Load the slog device's state from the MOS config 1774 * since it's possible that the label does not 1775 * contain the most up-to-date information. 1776 */ 1777 vdev_load_log_state(tvd, mtvd); 1778 vdev_reopen(tvd); 1779 } 1780 } 1781 vdev_free(mrvd); 1782 spa_config_exit(spa, SCL_ALL, FTAG); 1783 1784 /* 1785 * Ensure we were able to validate the config. 1786 */ 1787 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1788} 1789 1790/* 1791 * Check for missing log devices 1792 */ 1793static boolean_t 1794spa_check_logs(spa_t *spa) 1795{ 1796 boolean_t rv = B_FALSE; 1797 dsl_pool_t *dp = spa_get_dsl(spa); 1798 1799 switch (spa->spa_log_state) { 1800 case SPA_LOG_MISSING: 1801 /* need to recheck in case slog has been restored */ 1802 case SPA_LOG_UNKNOWN: 1803 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1804 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1805 if (rv) 1806 spa_set_log_state(spa, SPA_LOG_MISSING); 1807 break; 1808 } 1809 return (rv); 1810} 1811 1812static boolean_t 1813spa_passivate_log(spa_t *spa) 1814{ 1815 vdev_t *rvd = spa->spa_root_vdev; 1816 boolean_t slog_found = B_FALSE; 1817 1818 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1819 1820 if (!spa_has_slogs(spa)) 1821 return (B_FALSE); 1822 1823 for (int c = 0; c < rvd->vdev_children; c++) { 1824 vdev_t *tvd = rvd->vdev_child[c]; 1825 metaslab_group_t *mg = tvd->vdev_mg; 1826 1827 if (tvd->vdev_islog) { 1828 metaslab_group_passivate(mg); 1829 slog_found = B_TRUE; 1830 } 1831 } 1832 1833 return (slog_found); 1834} 1835 1836static void 1837spa_activate_log(spa_t *spa) 1838{ 1839 vdev_t *rvd = spa->spa_root_vdev; 1840 1841 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1842 1843 for (int c = 0; c < rvd->vdev_children; c++) { 1844 vdev_t *tvd = rvd->vdev_child[c]; 1845 metaslab_group_t *mg = tvd->vdev_mg; 1846 1847 if (tvd->vdev_islog) 1848 metaslab_group_activate(mg); 1849 } 1850} 1851 1852int 1853spa_offline_log(spa_t *spa) 1854{ 1855 int error; 1856 1857 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1858 NULL, DS_FIND_CHILDREN); 1859 if (error == 0) { 1860 /* 1861 * We successfully offlined the log device, sync out the 1862 * current txg so that the "stubby" block can be removed 1863 * by zil_sync(). 1864 */ 1865 txg_wait_synced(spa->spa_dsl_pool, 0); 1866 } 1867 return (error); 1868} 1869 1870static void 1871spa_aux_check_removed(spa_aux_vdev_t *sav) 1872{ 1873 int i; 1874 1875 for (i = 0; i < sav->sav_count; i++) 1876 spa_check_removed(sav->sav_vdevs[i]); 1877} 1878 1879void 1880spa_claim_notify(zio_t *zio) 1881{ 1882 spa_t *spa = zio->io_spa; 1883 1884 if (zio->io_error) 1885 return; 1886 1887 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1888 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1889 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1890 mutex_exit(&spa->spa_props_lock); 1891} 1892 1893typedef struct spa_load_error { 1894 uint64_t sle_meta_count; 1895 uint64_t sle_data_count; 1896} spa_load_error_t; 1897 1898static void 1899spa_load_verify_done(zio_t *zio) 1900{ 1901 blkptr_t *bp = zio->io_bp; 1902 spa_load_error_t *sle = zio->io_private; 1903 dmu_object_type_t type = BP_GET_TYPE(bp); 1904 int error = zio->io_error; 1905 spa_t *spa = zio->io_spa; 1906 1907 if (error) { 1908 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1909 type != DMU_OT_INTENT_LOG) 1910 atomic_inc_64(&sle->sle_meta_count); 1911 else 1912 atomic_inc_64(&sle->sle_data_count); 1913 } 1914 zio_data_buf_free(zio->io_data, zio->io_size); 1915 1916 mutex_enter(&spa->spa_scrub_lock); 1917 spa->spa_scrub_inflight--; 1918 cv_broadcast(&spa->spa_scrub_io_cv); 1919 mutex_exit(&spa->spa_scrub_lock); 1920} 1921 1922/* 1923 * Maximum number of concurrent scrub i/os to create while verifying 1924 * a pool while importing it. 1925 */ 1926int spa_load_verify_maxinflight = 10000; 1927boolean_t spa_load_verify_metadata = B_TRUE; 1928boolean_t spa_load_verify_data = B_TRUE; 1929 1930SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1931 &spa_load_verify_maxinflight, 0, 1932 "Maximum number of concurrent scrub I/Os to create while verifying a " 1933 "pool while importing it"); 1934 1935SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1936 &spa_load_verify_metadata, 0, 1937 "Check metadata on import?"); 1938 1939SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1940 &spa_load_verify_data, 0, 1941 "Check user data on import?"); 1942 1943/*ARGSUSED*/ 1944static int 1945spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1946 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1947{ 1948 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1949 return (0); 1950 /* 1951 * Note: normally this routine will not be called if 1952 * spa_load_verify_metadata is not set. However, it may be useful 1953 * to manually set the flag after the traversal has begun. 1954 */ 1955 if (!spa_load_verify_metadata) 1956 return (0); 1957 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1958 return (0); 1959 1960 zio_t *rio = arg; 1961 size_t size = BP_GET_PSIZE(bp); 1962 void *data = zio_data_buf_alloc(size); 1963 1964 mutex_enter(&spa->spa_scrub_lock); 1965 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1966 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1967 spa->spa_scrub_inflight++; 1968 mutex_exit(&spa->spa_scrub_lock); 1969 1970 zio_nowait(zio_read(rio, spa, bp, data, size, 1971 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1972 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1973 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1974 return (0); 1975} 1976 1977static int 1978spa_load_verify(spa_t *spa) 1979{ 1980 zio_t *rio; 1981 spa_load_error_t sle = { 0 }; 1982 zpool_rewind_policy_t policy; 1983 boolean_t verify_ok = B_FALSE; 1984 int error = 0; 1985 1986 zpool_get_rewind_policy(spa->spa_config, &policy); 1987 1988 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1989 return (0); 1990 1991 rio = zio_root(spa, NULL, &sle, 1992 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1993 1994 if (spa_load_verify_metadata) { 1995 error = traverse_pool(spa, spa->spa_verify_min_txg, 1996 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1997 spa_load_verify_cb, rio); 1998 } 1999 2000 (void) zio_wait(rio); 2001 2002 spa->spa_load_meta_errors = sle.sle_meta_count; 2003 spa->spa_load_data_errors = sle.sle_data_count; 2004 2005 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2006 sle.sle_data_count <= policy.zrp_maxdata) { 2007 int64_t loss = 0; 2008 2009 verify_ok = B_TRUE; 2010 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2011 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2012 2013 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2014 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2015 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2016 VERIFY(nvlist_add_int64(spa->spa_load_info, 2017 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2018 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2019 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2020 } else { 2021 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2022 } 2023 2024 if (error) { 2025 if (error != ENXIO && error != EIO) 2026 error = SET_ERROR(EIO); 2027 return (error); 2028 } 2029 2030 return (verify_ok ? 0 : EIO); 2031} 2032 2033/* 2034 * Find a value in the pool props object. 2035 */ 2036static void 2037spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2038{ 2039 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2040 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2041} 2042 2043/* 2044 * Find a value in the pool directory object. 2045 */ 2046static int 2047spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2048{ 2049 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2050 name, sizeof (uint64_t), 1, val)); 2051} 2052 2053static int 2054spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2055{ 2056 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2057 return (err); 2058} 2059 2060/* 2061 * Fix up config after a partly-completed split. This is done with the 2062 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2063 * pool have that entry in their config, but only the splitting one contains 2064 * a list of all the guids of the vdevs that are being split off. 2065 * 2066 * This function determines what to do with that list: either rejoin 2067 * all the disks to the pool, or complete the splitting process. To attempt 2068 * the rejoin, each disk that is offlined is marked online again, and 2069 * we do a reopen() call. If the vdev label for every disk that was 2070 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2071 * then we call vdev_split() on each disk, and complete the split. 2072 * 2073 * Otherwise we leave the config alone, with all the vdevs in place in 2074 * the original pool. 2075 */ 2076static void 2077spa_try_repair(spa_t *spa, nvlist_t *config) 2078{ 2079 uint_t extracted; 2080 uint64_t *glist; 2081 uint_t i, gcount; 2082 nvlist_t *nvl; 2083 vdev_t **vd; 2084 boolean_t attempt_reopen; 2085 2086 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2087 return; 2088 2089 /* check that the config is complete */ 2090 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2091 &glist, &gcount) != 0) 2092 return; 2093 2094 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2095 2096 /* attempt to online all the vdevs & validate */ 2097 attempt_reopen = B_TRUE; 2098 for (i = 0; i < gcount; i++) { 2099 if (glist[i] == 0) /* vdev is hole */ 2100 continue; 2101 2102 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2103 if (vd[i] == NULL) { 2104 /* 2105 * Don't bother attempting to reopen the disks; 2106 * just do the split. 2107 */ 2108 attempt_reopen = B_FALSE; 2109 } else { 2110 /* attempt to re-online it */ 2111 vd[i]->vdev_offline = B_FALSE; 2112 } 2113 } 2114 2115 if (attempt_reopen) { 2116 vdev_reopen(spa->spa_root_vdev); 2117 2118 /* check each device to see what state it's in */ 2119 for (extracted = 0, i = 0; i < gcount; i++) { 2120 if (vd[i] != NULL && 2121 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2122 break; 2123 ++extracted; 2124 } 2125 } 2126 2127 /* 2128 * If every disk has been moved to the new pool, or if we never 2129 * even attempted to look at them, then we split them off for 2130 * good. 2131 */ 2132 if (!attempt_reopen || gcount == extracted) { 2133 for (i = 0; i < gcount; i++) 2134 if (vd[i] != NULL) 2135 vdev_split(vd[i]); 2136 vdev_reopen(spa->spa_root_vdev); 2137 } 2138 2139 kmem_free(vd, gcount * sizeof (vdev_t *)); 2140} 2141 2142static int 2143spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2144 boolean_t mosconfig) 2145{ 2146 nvlist_t *config = spa->spa_config; 2147 char *ereport = FM_EREPORT_ZFS_POOL; 2148 char *comment; 2149 int error; 2150 uint64_t pool_guid; 2151 nvlist_t *nvl; 2152 2153 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2154 return (SET_ERROR(EINVAL)); 2155 2156 ASSERT(spa->spa_comment == NULL); 2157 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2158 spa->spa_comment = spa_strdup(comment); 2159 2160 /* 2161 * Versioning wasn't explicitly added to the label until later, so if 2162 * it's not present treat it as the initial version. 2163 */ 2164 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2165 &spa->spa_ubsync.ub_version) != 0) 2166 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2167 2168 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2169 &spa->spa_config_txg); 2170 2171 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2172 spa_guid_exists(pool_guid, 0)) { 2173 error = SET_ERROR(EEXIST); 2174 } else { 2175 spa->spa_config_guid = pool_guid; 2176 2177 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2178 &nvl) == 0) { 2179 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2180 KM_SLEEP) == 0); 2181 } 2182 2183 nvlist_free(spa->spa_load_info); 2184 spa->spa_load_info = fnvlist_alloc(); 2185 2186 gethrestime(&spa->spa_loaded_ts); 2187 error = spa_load_impl(spa, pool_guid, config, state, type, 2188 mosconfig, &ereport); 2189 } 2190 2191 /* 2192 * Don't count references from objsets that are already closed 2193 * and are making their way through the eviction process. 2194 */ 2195 spa_evicting_os_wait(spa); 2196 spa->spa_minref = refcount_count(&spa->spa_refcount); 2197 if (error) { 2198 if (error != EEXIST) { 2199 spa->spa_loaded_ts.tv_sec = 0; 2200 spa->spa_loaded_ts.tv_nsec = 0; 2201 } 2202 if (error != EBADF) { 2203 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2204 } 2205 } 2206 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2207 spa->spa_ena = 0; 2208 2209 return (error); 2210} 2211 2212/* 2213 * Load an existing storage pool, using the pool's builtin spa_config as a 2214 * source of configuration information. 2215 */ 2216static int 2217spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2218 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2219 char **ereport) 2220{ 2221 int error = 0; 2222 nvlist_t *nvroot = NULL; 2223 nvlist_t *label; 2224 vdev_t *rvd; 2225 uberblock_t *ub = &spa->spa_uberblock; 2226 uint64_t children, config_cache_txg = spa->spa_config_txg; 2227 int orig_mode = spa->spa_mode; 2228 int parse; 2229 uint64_t obj; 2230 boolean_t missing_feat_write = B_FALSE; 2231 2232 /* 2233 * If this is an untrusted config, access the pool in read-only mode. 2234 * This prevents things like resilvering recently removed devices. 2235 */ 2236 if (!mosconfig) 2237 spa->spa_mode = FREAD; 2238 2239 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2240 2241 spa->spa_load_state = state; 2242 2243 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2244 return (SET_ERROR(EINVAL)); 2245 2246 parse = (type == SPA_IMPORT_EXISTING ? 2247 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2248 2249 /* 2250 * Create "The Godfather" zio to hold all async IOs 2251 */ 2252 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2253 KM_SLEEP); 2254 for (int i = 0; i < max_ncpus; i++) { 2255 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2256 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2257 ZIO_FLAG_GODFATHER); 2258 } 2259 2260 /* 2261 * Parse the configuration into a vdev tree. We explicitly set the 2262 * value that will be returned by spa_version() since parsing the 2263 * configuration requires knowing the version number. 2264 */ 2265 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2266 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2267 spa_config_exit(spa, SCL_ALL, FTAG); 2268 2269 if (error != 0) 2270 return (error); 2271 2272 ASSERT(spa->spa_root_vdev == rvd); 2273 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2274 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2275 2276 if (type != SPA_IMPORT_ASSEMBLE) { 2277 ASSERT(spa_guid(spa) == pool_guid); 2278 } 2279 2280 /* 2281 * Try to open all vdevs, loading each label in the process. 2282 */ 2283 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2284 error = vdev_open(rvd); 2285 spa_config_exit(spa, SCL_ALL, FTAG); 2286 if (error != 0) 2287 return (error); 2288 2289 /* 2290 * We need to validate the vdev labels against the configuration that 2291 * we have in hand, which is dependent on the setting of mosconfig. If 2292 * mosconfig is true then we're validating the vdev labels based on 2293 * that config. Otherwise, we're validating against the cached config 2294 * (zpool.cache) that was read when we loaded the zfs module, and then 2295 * later we will recursively call spa_load() and validate against 2296 * the vdev config. 2297 * 2298 * If we're assembling a new pool that's been split off from an 2299 * existing pool, the labels haven't yet been updated so we skip 2300 * validation for now. 2301 */ 2302 if (type != SPA_IMPORT_ASSEMBLE) { 2303 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2304 error = vdev_validate(rvd, mosconfig); 2305 spa_config_exit(spa, SCL_ALL, FTAG); 2306 2307 if (error != 0) 2308 return (error); 2309 2310 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2311 return (SET_ERROR(ENXIO)); 2312 } 2313 2314 /* 2315 * Find the best uberblock. 2316 */ 2317 vdev_uberblock_load(rvd, ub, &label); 2318 2319 /* 2320 * If we weren't able to find a single valid uberblock, return failure. 2321 */ 2322 if (ub->ub_txg == 0) { 2323 nvlist_free(label); 2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2325 } 2326 2327 /* 2328 * If the pool has an unsupported version we can't open it. 2329 */ 2330 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2331 nvlist_free(label); 2332 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2333 } 2334 2335 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2336 nvlist_t *features; 2337 2338 /* 2339 * If we weren't able to find what's necessary for reading the 2340 * MOS in the label, return failure. 2341 */ 2342 if (label == NULL || nvlist_lookup_nvlist(label, 2343 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2344 nvlist_free(label); 2345 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2346 ENXIO)); 2347 } 2348 2349 /* 2350 * Update our in-core representation with the definitive values 2351 * from the label. 2352 */ 2353 nvlist_free(spa->spa_label_features); 2354 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2355 } 2356 2357 nvlist_free(label); 2358 2359 /* 2360 * Look through entries in the label nvlist's features_for_read. If 2361 * there is a feature listed there which we don't understand then we 2362 * cannot open a pool. 2363 */ 2364 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2365 nvlist_t *unsup_feat; 2366 2367 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2368 0); 2369 2370 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2371 NULL); nvp != NULL; 2372 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2373 if (!zfeature_is_supported(nvpair_name(nvp))) { 2374 VERIFY(nvlist_add_string(unsup_feat, 2375 nvpair_name(nvp), "") == 0); 2376 } 2377 } 2378 2379 if (!nvlist_empty(unsup_feat)) { 2380 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2381 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2382 nvlist_free(unsup_feat); 2383 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2384 ENOTSUP)); 2385 } 2386 2387 nvlist_free(unsup_feat); 2388 } 2389 2390 /* 2391 * If the vdev guid sum doesn't match the uberblock, we have an 2392 * incomplete configuration. We first check to see if the pool 2393 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2394 * If it is, defer the vdev_guid_sum check till later so we 2395 * can handle missing vdevs. 2396 */ 2397 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2398 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2399 rvd->vdev_guid_sum != ub->ub_guid_sum) 2400 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2401 2402 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2403 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2404 spa_try_repair(spa, config); 2405 spa_config_exit(spa, SCL_ALL, FTAG); 2406 nvlist_free(spa->spa_config_splitting); 2407 spa->spa_config_splitting = NULL; 2408 } 2409 2410 /* 2411 * Initialize internal SPA structures. 2412 */ 2413 spa->spa_state = POOL_STATE_ACTIVE; 2414 spa->spa_ubsync = spa->spa_uberblock; 2415 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2416 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2417 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2418 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2419 spa->spa_claim_max_txg = spa->spa_first_txg; 2420 spa->spa_prev_software_version = ub->ub_software_version; 2421 2422 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2423 if (error) 2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2425 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2426 2427 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2428 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2429 2430 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2431 boolean_t missing_feat_read = B_FALSE; 2432 nvlist_t *unsup_feat, *enabled_feat; 2433 2434 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2435 &spa->spa_feat_for_read_obj) != 0) { 2436 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2437 } 2438 2439 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2440 &spa->spa_feat_for_write_obj) != 0) { 2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2442 } 2443 2444 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2445 &spa->spa_feat_desc_obj) != 0) { 2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2447 } 2448 2449 enabled_feat = fnvlist_alloc(); 2450 unsup_feat = fnvlist_alloc(); 2451 2452 if (!spa_features_check(spa, B_FALSE, 2453 unsup_feat, enabled_feat)) 2454 missing_feat_read = B_TRUE; 2455 2456 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2457 if (!spa_features_check(spa, B_TRUE, 2458 unsup_feat, enabled_feat)) { 2459 missing_feat_write = B_TRUE; 2460 } 2461 } 2462 2463 fnvlist_add_nvlist(spa->spa_load_info, 2464 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2465 2466 if (!nvlist_empty(unsup_feat)) { 2467 fnvlist_add_nvlist(spa->spa_load_info, 2468 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2469 } 2470 2471 fnvlist_free(enabled_feat); 2472 fnvlist_free(unsup_feat); 2473 2474 if (!missing_feat_read) { 2475 fnvlist_add_boolean(spa->spa_load_info, 2476 ZPOOL_CONFIG_CAN_RDONLY); 2477 } 2478 2479 /* 2480 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2481 * twofold: to determine whether the pool is available for 2482 * import in read-write mode and (if it is not) whether the 2483 * pool is available for import in read-only mode. If the pool 2484 * is available for import in read-write mode, it is displayed 2485 * as available in userland; if it is not available for import 2486 * in read-only mode, it is displayed as unavailable in 2487 * userland. If the pool is available for import in read-only 2488 * mode but not read-write mode, it is displayed as unavailable 2489 * in userland with a special note that the pool is actually 2490 * available for open in read-only mode. 2491 * 2492 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2493 * missing a feature for write, we must first determine whether 2494 * the pool can be opened read-only before returning to 2495 * userland in order to know whether to display the 2496 * abovementioned note. 2497 */ 2498 if (missing_feat_read || (missing_feat_write && 2499 spa_writeable(spa))) { 2500 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2501 ENOTSUP)); 2502 } 2503 2504 /* 2505 * Load refcounts for ZFS features from disk into an in-memory 2506 * cache during SPA initialization. 2507 */ 2508 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2509 uint64_t refcount; 2510 2511 error = feature_get_refcount_from_disk(spa, 2512 &spa_feature_table[i], &refcount); 2513 if (error == 0) { 2514 spa->spa_feat_refcount_cache[i] = refcount; 2515 } else if (error == ENOTSUP) { 2516 spa->spa_feat_refcount_cache[i] = 2517 SPA_FEATURE_DISABLED; 2518 } else { 2519 return (spa_vdev_err(rvd, 2520 VDEV_AUX_CORRUPT_DATA, EIO)); 2521 } 2522 } 2523 } 2524 2525 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2526 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2527 &spa->spa_feat_enabled_txg_obj) != 0) 2528 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2529 } 2530 2531 spa->spa_is_initializing = B_TRUE; 2532 error = dsl_pool_open(spa->spa_dsl_pool); 2533 spa->spa_is_initializing = B_FALSE; 2534 if (error != 0) 2535 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2536 2537 if (!mosconfig) { 2538 uint64_t hostid; 2539 nvlist_t *policy = NULL, *nvconfig; 2540 2541 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2542 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2543 2544 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2545 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2546 char *hostname; 2547 unsigned long myhostid = 0; 2548 2549 VERIFY(nvlist_lookup_string(nvconfig, 2550 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2551 2552#ifdef _KERNEL 2553 myhostid = zone_get_hostid(NULL); 2554#else /* _KERNEL */ 2555 /* 2556 * We're emulating the system's hostid in userland, so 2557 * we can't use zone_get_hostid(). 2558 */ 2559 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2560#endif /* _KERNEL */ 2561 if (check_hostid && hostid != 0 && myhostid != 0 && 2562 hostid != myhostid) { 2563 nvlist_free(nvconfig); 2564 cmn_err(CE_WARN, "pool '%s' could not be " 2565 "loaded as it was last accessed by " 2566 "another system (host: %s hostid: 0x%lx). " 2567 "See: http://illumos.org/msg/ZFS-8000-EY", 2568 spa_name(spa), hostname, 2569 (unsigned long)hostid); 2570 return (SET_ERROR(EBADF)); 2571 } 2572 } 2573 if (nvlist_lookup_nvlist(spa->spa_config, 2574 ZPOOL_REWIND_POLICY, &policy) == 0) 2575 VERIFY(nvlist_add_nvlist(nvconfig, 2576 ZPOOL_REWIND_POLICY, policy) == 0); 2577 2578 spa_config_set(spa, nvconfig); 2579 spa_unload(spa); 2580 spa_deactivate(spa); 2581 spa_activate(spa, orig_mode); 2582 2583 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2584 } 2585 2586 /* Grab the secret checksum salt from the MOS. */ 2587 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2588 DMU_POOL_CHECKSUM_SALT, 1, 2589 sizeof (spa->spa_cksum_salt.zcs_bytes), 2590 spa->spa_cksum_salt.zcs_bytes); 2591 if (error == ENOENT) { 2592 /* Generate a new salt for subsequent use */ 2593 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 2594 sizeof (spa->spa_cksum_salt.zcs_bytes)); 2595 } else if (error != 0) { 2596 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2597 } 2598 2599 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2600 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2601 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2602 if (error != 0) 2603 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2604 2605 /* 2606 * Load the bit that tells us to use the new accounting function 2607 * (raid-z deflation). If we have an older pool, this will not 2608 * be present. 2609 */ 2610 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2611 if (error != 0 && error != ENOENT) 2612 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2613 2614 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2615 &spa->spa_creation_version); 2616 if (error != 0 && error != ENOENT) 2617 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2618 2619 /* 2620 * Load the persistent error log. If we have an older pool, this will 2621 * not be present. 2622 */ 2623 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2624 if (error != 0 && error != ENOENT) 2625 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2626 2627 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2628 &spa->spa_errlog_scrub); 2629 if (error != 0 && error != ENOENT) 2630 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2631 2632 /* 2633 * Load the history object. If we have an older pool, this 2634 * will not be present. 2635 */ 2636 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2637 if (error != 0 && error != ENOENT) 2638 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2639 2640 /* 2641 * If we're assembling the pool from the split-off vdevs of 2642 * an existing pool, we don't want to attach the spares & cache 2643 * devices. 2644 */ 2645 2646 /* 2647 * Load any hot spares for this pool. 2648 */ 2649 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2650 if (error != 0 && error != ENOENT) 2651 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2652 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2653 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2654 if (load_nvlist(spa, spa->spa_spares.sav_object, 2655 &spa->spa_spares.sav_config) != 0) 2656 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2657 2658 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2659 spa_load_spares(spa); 2660 spa_config_exit(spa, SCL_ALL, FTAG); 2661 } else if (error == 0) { 2662 spa->spa_spares.sav_sync = B_TRUE; 2663 } 2664 2665 /* 2666 * Load any level 2 ARC devices for this pool. 2667 */ 2668 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2669 &spa->spa_l2cache.sav_object); 2670 if (error != 0 && error != ENOENT) 2671 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2672 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2673 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2674 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2675 &spa->spa_l2cache.sav_config) != 0) 2676 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2677 2678 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2679 spa_load_l2cache(spa); 2680 spa_config_exit(spa, SCL_ALL, FTAG); 2681 } else if (error == 0) { 2682 spa->spa_l2cache.sav_sync = B_TRUE; 2683 } 2684 2685 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2686 2687 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2688 if (error && error != ENOENT) 2689 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2690 2691 if (error == 0) { 2692 uint64_t autoreplace; 2693 2694 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2695 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2696 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2697 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2698 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2699 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2700 &spa->spa_dedup_ditto); 2701 2702 spa->spa_autoreplace = (autoreplace != 0); 2703 } 2704 2705 /* 2706 * If the 'autoreplace' property is set, then post a resource notifying 2707 * the ZFS DE that it should not issue any faults for unopenable 2708 * devices. We also iterate over the vdevs, and post a sysevent for any 2709 * unopenable vdevs so that the normal autoreplace handler can take 2710 * over. 2711 */ 2712 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2713 spa_check_removed(spa->spa_root_vdev); 2714 /* 2715 * For the import case, this is done in spa_import(), because 2716 * at this point we're using the spare definitions from 2717 * the MOS config, not necessarily from the userland config. 2718 */ 2719 if (state != SPA_LOAD_IMPORT) { 2720 spa_aux_check_removed(&spa->spa_spares); 2721 spa_aux_check_removed(&spa->spa_l2cache); 2722 } 2723 } 2724 2725 /* 2726 * Load the vdev state for all toplevel vdevs. 2727 */ 2728 vdev_load(rvd); 2729 2730 /* 2731 * Propagate the leaf DTLs we just loaded all the way up the tree. 2732 */ 2733 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2734 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2735 spa_config_exit(spa, SCL_ALL, FTAG); 2736 2737 /* 2738 * Load the DDTs (dedup tables). 2739 */ 2740 error = ddt_load(spa); 2741 if (error != 0) 2742 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2743 2744 spa_update_dspace(spa); 2745 2746 /* 2747 * Validate the config, using the MOS config to fill in any 2748 * information which might be missing. If we fail to validate 2749 * the config then declare the pool unfit for use. If we're 2750 * assembling a pool from a split, the log is not transferred 2751 * over. 2752 */ 2753 if (type != SPA_IMPORT_ASSEMBLE) { 2754 nvlist_t *nvconfig; 2755 2756 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2757 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2758 2759 if (!spa_config_valid(spa, nvconfig)) { 2760 nvlist_free(nvconfig); 2761 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2762 ENXIO)); 2763 } 2764 nvlist_free(nvconfig); 2765 2766 /* 2767 * Now that we've validated the config, check the state of the 2768 * root vdev. If it can't be opened, it indicates one or 2769 * more toplevel vdevs are faulted. 2770 */ 2771 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2772 return (SET_ERROR(ENXIO)); 2773 2774 if (spa_writeable(spa) && spa_check_logs(spa)) { 2775 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2776 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2777 } 2778 } 2779 2780 if (missing_feat_write) { 2781 ASSERT(state == SPA_LOAD_TRYIMPORT); 2782 2783 /* 2784 * At this point, we know that we can open the pool in 2785 * read-only mode but not read-write mode. We now have enough 2786 * information and can return to userland. 2787 */ 2788 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2789 } 2790 2791 /* 2792 * We've successfully opened the pool, verify that we're ready 2793 * to start pushing transactions. 2794 */ 2795 if (state != SPA_LOAD_TRYIMPORT) { 2796 if (error = spa_load_verify(spa)) 2797 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2798 error)); 2799 } 2800 2801 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2802 spa->spa_load_max_txg == UINT64_MAX)) { 2803 dmu_tx_t *tx; 2804 int need_update = B_FALSE; 2805 dsl_pool_t *dp = spa_get_dsl(spa); 2806 2807 ASSERT(state != SPA_LOAD_TRYIMPORT); 2808 2809 /* 2810 * Claim log blocks that haven't been committed yet. 2811 * This must all happen in a single txg. 2812 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2813 * invoked from zil_claim_log_block()'s i/o done callback. 2814 * Price of rollback is that we abandon the log. 2815 */ 2816 spa->spa_claiming = B_TRUE; 2817 2818 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2819 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2820 zil_claim, tx, DS_FIND_CHILDREN); 2821 dmu_tx_commit(tx); 2822 2823 spa->spa_claiming = B_FALSE; 2824 2825 spa_set_log_state(spa, SPA_LOG_GOOD); 2826 spa->spa_sync_on = B_TRUE; 2827 txg_sync_start(spa->spa_dsl_pool); 2828 2829 /* 2830 * Wait for all claims to sync. We sync up to the highest 2831 * claimed log block birth time so that claimed log blocks 2832 * don't appear to be from the future. spa_claim_max_txg 2833 * will have been set for us by either zil_check_log_chain() 2834 * (invoked from spa_check_logs()) or zil_claim() above. 2835 */ 2836 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2837 2838 /* 2839 * If the config cache is stale, or we have uninitialized 2840 * metaslabs (see spa_vdev_add()), then update the config. 2841 * 2842 * If this is a verbatim import, trust the current 2843 * in-core spa_config and update the disk labels. 2844 */ 2845 if (config_cache_txg != spa->spa_config_txg || 2846 state == SPA_LOAD_IMPORT || 2847 state == SPA_LOAD_RECOVER || 2848 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2849 need_update = B_TRUE; 2850 2851 for (int c = 0; c < rvd->vdev_children; c++) 2852 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2853 need_update = B_TRUE; 2854 2855 /* 2856 * Update the config cache asychronously in case we're the 2857 * root pool, in which case the config cache isn't writable yet. 2858 */ 2859 if (need_update) 2860 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2861 2862 /* 2863 * Check all DTLs to see if anything needs resilvering. 2864 */ 2865 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2866 vdev_resilver_needed(rvd, NULL, NULL)) 2867 spa_async_request(spa, SPA_ASYNC_RESILVER); 2868 2869 /* 2870 * Log the fact that we booted up (so that we can detect if 2871 * we rebooted in the middle of an operation). 2872 */ 2873 spa_history_log_version(spa, "open"); 2874 2875 /* 2876 * Delete any inconsistent datasets. 2877 */ 2878 (void) dmu_objset_find(spa_name(spa), 2879 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2880 2881 /* 2882 * Clean up any stale temporary dataset userrefs. 2883 */ 2884 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2885 } 2886 2887 return (0); 2888} 2889 2890static int 2891spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2892{ 2893 int mode = spa->spa_mode; 2894 2895 spa_unload(spa); 2896 spa_deactivate(spa); 2897 2898 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2899 2900 spa_activate(spa, mode); 2901 spa_async_suspend(spa); 2902 2903 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2904} 2905 2906/* 2907 * If spa_load() fails this function will try loading prior txg's. If 2908 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2909 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2910 * function will not rewind the pool and will return the same error as 2911 * spa_load(). 2912 */ 2913static int 2914spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2915 uint64_t max_request, int rewind_flags) 2916{ 2917 nvlist_t *loadinfo = NULL; 2918 nvlist_t *config = NULL; 2919 int load_error, rewind_error; 2920 uint64_t safe_rewind_txg; 2921 uint64_t min_txg; 2922 2923 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2924 spa->spa_load_max_txg = spa->spa_load_txg; 2925 spa_set_log_state(spa, SPA_LOG_CLEAR); 2926 } else { 2927 spa->spa_load_max_txg = max_request; 2928 if (max_request != UINT64_MAX) 2929 spa->spa_extreme_rewind = B_TRUE; 2930 } 2931 2932 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2933 mosconfig); 2934 if (load_error == 0) 2935 return (0); 2936 2937 if (spa->spa_root_vdev != NULL) 2938 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2939 2940 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2941 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2942 2943 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2944 nvlist_free(config); 2945 return (load_error); 2946 } 2947 2948 if (state == SPA_LOAD_RECOVER) { 2949 /* Price of rolling back is discarding txgs, including log */ 2950 spa_set_log_state(spa, SPA_LOG_CLEAR); 2951 } else { 2952 /* 2953 * If we aren't rolling back save the load info from our first 2954 * import attempt so that we can restore it after attempting 2955 * to rewind. 2956 */ 2957 loadinfo = spa->spa_load_info; 2958 spa->spa_load_info = fnvlist_alloc(); 2959 } 2960 2961 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2962 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2963 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2964 TXG_INITIAL : safe_rewind_txg; 2965 2966 /* 2967 * Continue as long as we're finding errors, we're still within 2968 * the acceptable rewind range, and we're still finding uberblocks 2969 */ 2970 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2971 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2972 if (spa->spa_load_max_txg < safe_rewind_txg) 2973 spa->spa_extreme_rewind = B_TRUE; 2974 rewind_error = spa_load_retry(spa, state, mosconfig); 2975 } 2976 2977 spa->spa_extreme_rewind = B_FALSE; 2978 spa->spa_load_max_txg = UINT64_MAX; 2979 2980 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2981 spa_config_set(spa, config); 2982 2983 if (state == SPA_LOAD_RECOVER) { 2984 ASSERT3P(loadinfo, ==, NULL); 2985 return (rewind_error); 2986 } else { 2987 /* Store the rewind info as part of the initial load info */ 2988 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2989 spa->spa_load_info); 2990 2991 /* Restore the initial load info */ 2992 fnvlist_free(spa->spa_load_info); 2993 spa->spa_load_info = loadinfo; 2994 2995 return (load_error); 2996 } 2997} 2998 2999/* 3000 * Pool Open/Import 3001 * 3002 * The import case is identical to an open except that the configuration is sent 3003 * down from userland, instead of grabbed from the configuration cache. For the 3004 * case of an open, the pool configuration will exist in the 3005 * POOL_STATE_UNINITIALIZED state. 3006 * 3007 * The stats information (gen/count/ustats) is used to gather vdev statistics at 3008 * the same time open the pool, without having to keep around the spa_t in some 3009 * ambiguous state. 3010 */ 3011static int 3012spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3013 nvlist_t **config) 3014{ 3015 spa_t *spa; 3016 spa_load_state_t state = SPA_LOAD_OPEN; 3017 int error; 3018 int locked = B_FALSE; 3019 int firstopen = B_FALSE; 3020 3021 *spapp = NULL; 3022 3023 /* 3024 * As disgusting as this is, we need to support recursive calls to this 3025 * function because dsl_dir_open() is called during spa_load(), and ends 3026 * up calling spa_open() again. The real fix is to figure out how to 3027 * avoid dsl_dir_open() calling this in the first place. 3028 */ 3029 if (mutex_owner(&spa_namespace_lock) != curthread) { 3030 mutex_enter(&spa_namespace_lock); 3031 locked = B_TRUE; 3032 } 3033 3034 if ((spa = spa_lookup(pool)) == NULL) { 3035 if (locked) 3036 mutex_exit(&spa_namespace_lock); 3037 return (SET_ERROR(ENOENT)); 3038 } 3039 3040 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3041 zpool_rewind_policy_t policy; 3042 3043 firstopen = B_TRUE; 3044 3045 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3046 &policy); 3047 if (policy.zrp_request & ZPOOL_DO_REWIND) 3048 state = SPA_LOAD_RECOVER; 3049 3050 spa_activate(spa, spa_mode_global); 3051 3052 if (state != SPA_LOAD_RECOVER) 3053 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3054 3055 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3056 policy.zrp_request); 3057 3058 if (error == EBADF) { 3059 /* 3060 * If vdev_validate() returns failure (indicated by 3061 * EBADF), it indicates that one of the vdevs indicates 3062 * that the pool has been exported or destroyed. If 3063 * this is the case, the config cache is out of sync and 3064 * we should remove the pool from the namespace. 3065 */ 3066 spa_unload(spa); 3067 spa_deactivate(spa); 3068 spa_config_sync(spa, B_TRUE, B_TRUE); 3069 spa_remove(spa); 3070 if (locked) 3071 mutex_exit(&spa_namespace_lock); 3072 return (SET_ERROR(ENOENT)); 3073 } 3074 3075 if (error) { 3076 /* 3077 * We can't open the pool, but we still have useful 3078 * information: the state of each vdev after the 3079 * attempted vdev_open(). Return this to the user. 3080 */ 3081 if (config != NULL && spa->spa_config) { 3082 VERIFY(nvlist_dup(spa->spa_config, config, 3083 KM_SLEEP) == 0); 3084 VERIFY(nvlist_add_nvlist(*config, 3085 ZPOOL_CONFIG_LOAD_INFO, 3086 spa->spa_load_info) == 0); 3087 } 3088 spa_unload(spa); 3089 spa_deactivate(spa); 3090 spa->spa_last_open_failed = error; 3091 if (locked) 3092 mutex_exit(&spa_namespace_lock); 3093 *spapp = NULL; 3094 return (error); 3095 } 3096 } 3097 3098 spa_open_ref(spa, tag); 3099 3100 if (config != NULL) 3101 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3102 3103 /* 3104 * If we've recovered the pool, pass back any information we 3105 * gathered while doing the load. 3106 */ 3107 if (state == SPA_LOAD_RECOVER) { 3108 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3109 spa->spa_load_info) == 0); 3110 } 3111 3112 if (locked) { 3113 spa->spa_last_open_failed = 0; 3114 spa->spa_last_ubsync_txg = 0; 3115 spa->spa_load_txg = 0; 3116 mutex_exit(&spa_namespace_lock); 3117#ifdef __FreeBSD__ 3118#ifdef _KERNEL 3119 if (firstopen) 3120 zvol_create_minors(spa->spa_name); 3121#endif 3122#endif 3123 } 3124 3125 *spapp = spa; 3126 3127 return (0); 3128} 3129 3130int 3131spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3132 nvlist_t **config) 3133{ 3134 return (spa_open_common(name, spapp, tag, policy, config)); 3135} 3136 3137int 3138spa_open(const char *name, spa_t **spapp, void *tag) 3139{ 3140 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3141} 3142 3143/* 3144 * Lookup the given spa_t, incrementing the inject count in the process, 3145 * preventing it from being exported or destroyed. 3146 */ 3147spa_t * 3148spa_inject_addref(char *name) 3149{ 3150 spa_t *spa; 3151 3152 mutex_enter(&spa_namespace_lock); 3153 if ((spa = spa_lookup(name)) == NULL) { 3154 mutex_exit(&spa_namespace_lock); 3155 return (NULL); 3156 } 3157 spa->spa_inject_ref++; 3158 mutex_exit(&spa_namespace_lock); 3159 3160 return (spa); 3161} 3162 3163void 3164spa_inject_delref(spa_t *spa) 3165{ 3166 mutex_enter(&spa_namespace_lock); 3167 spa->spa_inject_ref--; 3168 mutex_exit(&spa_namespace_lock); 3169} 3170 3171/* 3172 * Add spares device information to the nvlist. 3173 */ 3174static void 3175spa_add_spares(spa_t *spa, nvlist_t *config) 3176{ 3177 nvlist_t **spares; 3178 uint_t i, nspares; 3179 nvlist_t *nvroot; 3180 uint64_t guid; 3181 vdev_stat_t *vs; 3182 uint_t vsc; 3183 uint64_t pool; 3184 3185 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3186 3187 if (spa->spa_spares.sav_count == 0) 3188 return; 3189 3190 VERIFY(nvlist_lookup_nvlist(config, 3191 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3192 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3193 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3194 if (nspares != 0) { 3195 VERIFY(nvlist_add_nvlist_array(nvroot, 3196 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3197 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3198 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3199 3200 /* 3201 * Go through and find any spares which have since been 3202 * repurposed as an active spare. If this is the case, update 3203 * their status appropriately. 3204 */ 3205 for (i = 0; i < nspares; i++) { 3206 VERIFY(nvlist_lookup_uint64(spares[i], 3207 ZPOOL_CONFIG_GUID, &guid) == 0); 3208 if (spa_spare_exists(guid, &pool, NULL) && 3209 pool != 0ULL) { 3210 VERIFY(nvlist_lookup_uint64_array( 3211 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3212 (uint64_t **)&vs, &vsc) == 0); 3213 vs->vs_state = VDEV_STATE_CANT_OPEN; 3214 vs->vs_aux = VDEV_AUX_SPARED; 3215 } 3216 } 3217 } 3218} 3219 3220/* 3221 * Add l2cache device information to the nvlist, including vdev stats. 3222 */ 3223static void 3224spa_add_l2cache(spa_t *spa, nvlist_t *config) 3225{ 3226 nvlist_t **l2cache; 3227 uint_t i, j, nl2cache; 3228 nvlist_t *nvroot; 3229 uint64_t guid; 3230 vdev_t *vd; 3231 vdev_stat_t *vs; 3232 uint_t vsc; 3233 3234 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3235 3236 if (spa->spa_l2cache.sav_count == 0) 3237 return; 3238 3239 VERIFY(nvlist_lookup_nvlist(config, 3240 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3241 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3242 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3243 if (nl2cache != 0) { 3244 VERIFY(nvlist_add_nvlist_array(nvroot, 3245 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3246 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3247 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3248 3249 /* 3250 * Update level 2 cache device stats. 3251 */ 3252 3253 for (i = 0; i < nl2cache; i++) { 3254 VERIFY(nvlist_lookup_uint64(l2cache[i], 3255 ZPOOL_CONFIG_GUID, &guid) == 0); 3256 3257 vd = NULL; 3258 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3259 if (guid == 3260 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3261 vd = spa->spa_l2cache.sav_vdevs[j]; 3262 break; 3263 } 3264 } 3265 ASSERT(vd != NULL); 3266 3267 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3268 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3269 == 0); 3270 vdev_get_stats(vd, vs); 3271 } 3272 } 3273} 3274 3275static void 3276spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3277{ 3278 nvlist_t *features; 3279 zap_cursor_t zc; 3280 zap_attribute_t za; 3281 3282 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3283 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3284 3285 /* We may be unable to read features if pool is suspended. */ 3286 if (spa_suspended(spa)) 3287 goto out; 3288 3289 if (spa->spa_feat_for_read_obj != 0) { 3290 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3291 spa->spa_feat_for_read_obj); 3292 zap_cursor_retrieve(&zc, &za) == 0; 3293 zap_cursor_advance(&zc)) { 3294 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3295 za.za_num_integers == 1); 3296 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3297 za.za_first_integer)); 3298 } 3299 zap_cursor_fini(&zc); 3300 } 3301 3302 if (spa->spa_feat_for_write_obj != 0) { 3303 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3304 spa->spa_feat_for_write_obj); 3305 zap_cursor_retrieve(&zc, &za) == 0; 3306 zap_cursor_advance(&zc)) { 3307 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3308 za.za_num_integers == 1); 3309 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3310 za.za_first_integer)); 3311 } 3312 zap_cursor_fini(&zc); 3313 } 3314 3315out: 3316 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3317 features) == 0); 3318 nvlist_free(features); 3319} 3320 3321int 3322spa_get_stats(const char *name, nvlist_t **config, 3323 char *altroot, size_t buflen) 3324{ 3325 int error; 3326 spa_t *spa; 3327 3328 *config = NULL; 3329 error = spa_open_common(name, &spa, FTAG, NULL, config); 3330 3331 if (spa != NULL) { 3332 /* 3333 * This still leaves a window of inconsistency where the spares 3334 * or l2cache devices could change and the config would be 3335 * self-inconsistent. 3336 */ 3337 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3338 3339 if (*config != NULL) { 3340 uint64_t loadtimes[2]; 3341 3342 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3343 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3344 VERIFY(nvlist_add_uint64_array(*config, 3345 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3346 3347 VERIFY(nvlist_add_uint64(*config, 3348 ZPOOL_CONFIG_ERRCOUNT, 3349 spa_get_errlog_size(spa)) == 0); 3350 3351 if (spa_suspended(spa)) 3352 VERIFY(nvlist_add_uint64(*config, 3353 ZPOOL_CONFIG_SUSPENDED, 3354 spa->spa_failmode) == 0); 3355 3356 spa_add_spares(spa, *config); 3357 spa_add_l2cache(spa, *config); 3358 spa_add_feature_stats(spa, *config); 3359 } 3360 } 3361 3362 /* 3363 * We want to get the alternate root even for faulted pools, so we cheat 3364 * and call spa_lookup() directly. 3365 */ 3366 if (altroot) { 3367 if (spa == NULL) { 3368 mutex_enter(&spa_namespace_lock); 3369 spa = spa_lookup(name); 3370 if (spa) 3371 spa_altroot(spa, altroot, buflen); 3372 else 3373 altroot[0] = '\0'; 3374 spa = NULL; 3375 mutex_exit(&spa_namespace_lock); 3376 } else { 3377 spa_altroot(spa, altroot, buflen); 3378 } 3379 } 3380 3381 if (spa != NULL) { 3382 spa_config_exit(spa, SCL_CONFIG, FTAG); 3383 spa_close(spa, FTAG); 3384 } 3385 3386 return (error); 3387} 3388 3389/* 3390 * Validate that the auxiliary device array is well formed. We must have an 3391 * array of nvlists, each which describes a valid leaf vdev. If this is an 3392 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3393 * specified, as long as they are well-formed. 3394 */ 3395static int 3396spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3397 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3398 vdev_labeltype_t label) 3399{ 3400 nvlist_t **dev; 3401 uint_t i, ndev; 3402 vdev_t *vd; 3403 int error; 3404 3405 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3406 3407 /* 3408 * It's acceptable to have no devs specified. 3409 */ 3410 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3411 return (0); 3412 3413 if (ndev == 0) 3414 return (SET_ERROR(EINVAL)); 3415 3416 /* 3417 * Make sure the pool is formatted with a version that supports this 3418 * device type. 3419 */ 3420 if (spa_version(spa) < version) 3421 return (SET_ERROR(ENOTSUP)); 3422 3423 /* 3424 * Set the pending device list so we correctly handle device in-use 3425 * checking. 3426 */ 3427 sav->sav_pending = dev; 3428 sav->sav_npending = ndev; 3429 3430 for (i = 0; i < ndev; i++) { 3431 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3432 mode)) != 0) 3433 goto out; 3434 3435 if (!vd->vdev_ops->vdev_op_leaf) { 3436 vdev_free(vd); 3437 error = SET_ERROR(EINVAL); 3438 goto out; 3439 } 3440 3441 /* 3442 * The L2ARC currently only supports disk devices in 3443 * kernel context. For user-level testing, we allow it. 3444 */ 3445#ifdef _KERNEL 3446 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3447 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3448 error = SET_ERROR(ENOTBLK); 3449 vdev_free(vd); 3450 goto out; 3451 } 3452#endif 3453 vd->vdev_top = vd; 3454 3455 if ((error = vdev_open(vd)) == 0 && 3456 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3457 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3458 vd->vdev_guid) == 0); 3459 } 3460 3461 vdev_free(vd); 3462 3463 if (error && 3464 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3465 goto out; 3466 else 3467 error = 0; 3468 } 3469 3470out: 3471 sav->sav_pending = NULL; 3472 sav->sav_npending = 0; 3473 return (error); 3474} 3475 3476static int 3477spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3478{ 3479 int error; 3480 3481 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3482 3483 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3484 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3485 VDEV_LABEL_SPARE)) != 0) { 3486 return (error); 3487 } 3488 3489 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3490 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3491 VDEV_LABEL_L2CACHE)); 3492} 3493 3494static void 3495spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3496 const char *config) 3497{ 3498 int i; 3499 3500 if (sav->sav_config != NULL) { 3501 nvlist_t **olddevs; 3502 uint_t oldndevs; 3503 nvlist_t **newdevs; 3504 3505 /* 3506 * Generate new dev list by concatentating with the 3507 * current dev list. 3508 */ 3509 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3510 &olddevs, &oldndevs) == 0); 3511 3512 newdevs = kmem_alloc(sizeof (void *) * 3513 (ndevs + oldndevs), KM_SLEEP); 3514 for (i = 0; i < oldndevs; i++) 3515 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3516 KM_SLEEP) == 0); 3517 for (i = 0; i < ndevs; i++) 3518 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3519 KM_SLEEP) == 0); 3520 3521 VERIFY(nvlist_remove(sav->sav_config, config, 3522 DATA_TYPE_NVLIST_ARRAY) == 0); 3523 3524 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3525 config, newdevs, ndevs + oldndevs) == 0); 3526 for (i = 0; i < oldndevs + ndevs; i++) 3527 nvlist_free(newdevs[i]); 3528 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3529 } else { 3530 /* 3531 * Generate a new dev list. 3532 */ 3533 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3534 KM_SLEEP) == 0); 3535 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3536 devs, ndevs) == 0); 3537 } 3538} 3539 3540/* 3541 * Stop and drop level 2 ARC devices 3542 */ 3543void 3544spa_l2cache_drop(spa_t *spa) 3545{ 3546 vdev_t *vd; 3547 int i; 3548 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3549 3550 for (i = 0; i < sav->sav_count; i++) { 3551 uint64_t pool; 3552 3553 vd = sav->sav_vdevs[i]; 3554 ASSERT(vd != NULL); 3555 3556 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3557 pool != 0ULL && l2arc_vdev_present(vd)) 3558 l2arc_remove_vdev(vd); 3559 } 3560} 3561 3562/* 3563 * Pool Creation 3564 */ 3565int 3566spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3567 nvlist_t *zplprops) 3568{ 3569 spa_t *spa; 3570 char *altroot = NULL; 3571 vdev_t *rvd; 3572 dsl_pool_t *dp; 3573 dmu_tx_t *tx; 3574 int error = 0; 3575 uint64_t txg = TXG_INITIAL; 3576 nvlist_t **spares, **l2cache; 3577 uint_t nspares, nl2cache; 3578 uint64_t version, obj; 3579 boolean_t has_features; 3580 3581 /* 3582 * If this pool already exists, return failure. 3583 */ 3584 mutex_enter(&spa_namespace_lock); 3585 if (spa_lookup(pool) != NULL) { 3586 mutex_exit(&spa_namespace_lock); 3587 return (SET_ERROR(EEXIST)); 3588 } 3589 3590 /* 3591 * Allocate a new spa_t structure. 3592 */ 3593 (void) nvlist_lookup_string(props, 3594 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3595 spa = spa_add(pool, NULL, altroot); 3596 spa_activate(spa, spa_mode_global); 3597 3598 if (props && (error = spa_prop_validate(spa, props))) { 3599 spa_deactivate(spa); 3600 spa_remove(spa); 3601 mutex_exit(&spa_namespace_lock); 3602 return (error); 3603 } 3604 3605 has_features = B_FALSE; 3606 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3607 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3608 if (zpool_prop_feature(nvpair_name(elem))) 3609 has_features = B_TRUE; 3610 } 3611 3612 if (has_features || nvlist_lookup_uint64(props, 3613 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3614 version = SPA_VERSION; 3615 } 3616 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3617 3618 spa->spa_first_txg = txg; 3619 spa->spa_uberblock.ub_txg = txg - 1; 3620 spa->spa_uberblock.ub_version = version; 3621 spa->spa_ubsync = spa->spa_uberblock; 3622 3623 /* 3624 * Create "The Godfather" zio to hold all async IOs 3625 */ 3626 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3627 KM_SLEEP); 3628 for (int i = 0; i < max_ncpus; i++) { 3629 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3630 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3631 ZIO_FLAG_GODFATHER); 3632 } 3633 3634 /* 3635 * Create the root vdev. 3636 */ 3637 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3638 3639 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3640 3641 ASSERT(error != 0 || rvd != NULL); 3642 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3643 3644 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3645 error = SET_ERROR(EINVAL); 3646 3647 if (error == 0 && 3648 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3649 (error = spa_validate_aux(spa, nvroot, txg, 3650 VDEV_ALLOC_ADD)) == 0) { 3651 for (int c = 0; c < rvd->vdev_children; c++) { 3652 vdev_ashift_optimize(rvd->vdev_child[c]); 3653 vdev_metaslab_set_size(rvd->vdev_child[c]); 3654 vdev_expand(rvd->vdev_child[c], txg); 3655 } 3656 } 3657 3658 spa_config_exit(spa, SCL_ALL, FTAG); 3659 3660 if (error != 0) { 3661 spa_unload(spa); 3662 spa_deactivate(spa); 3663 spa_remove(spa); 3664 mutex_exit(&spa_namespace_lock); 3665 return (error); 3666 } 3667 3668 /* 3669 * Get the list of spares, if specified. 3670 */ 3671 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3672 &spares, &nspares) == 0) { 3673 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3674 KM_SLEEP) == 0); 3675 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3676 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3677 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3678 spa_load_spares(spa); 3679 spa_config_exit(spa, SCL_ALL, FTAG); 3680 spa->spa_spares.sav_sync = B_TRUE; 3681 } 3682 3683 /* 3684 * Get the list of level 2 cache devices, if specified. 3685 */ 3686 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3687 &l2cache, &nl2cache) == 0) { 3688 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3689 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3690 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3691 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3692 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3693 spa_load_l2cache(spa); 3694 spa_config_exit(spa, SCL_ALL, FTAG); 3695 spa->spa_l2cache.sav_sync = B_TRUE; 3696 } 3697 3698 spa->spa_is_initializing = B_TRUE; 3699 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3700 spa->spa_meta_objset = dp->dp_meta_objset; 3701 spa->spa_is_initializing = B_FALSE; 3702 3703 /* 3704 * Create DDTs (dedup tables). 3705 */ 3706 ddt_create(spa); 3707 3708 spa_update_dspace(spa); 3709 3710 tx = dmu_tx_create_assigned(dp, txg); 3711 3712 /* 3713 * Create the pool config object. 3714 */ 3715 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3716 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3717 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3718 3719 if (zap_add(spa->spa_meta_objset, 3720 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3721 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3722 cmn_err(CE_PANIC, "failed to add pool config"); 3723 } 3724 3725 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3726 spa_feature_create_zap_objects(spa, tx); 3727 3728 if (zap_add(spa->spa_meta_objset, 3729 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3730 sizeof (uint64_t), 1, &version, tx) != 0) { 3731 cmn_err(CE_PANIC, "failed to add pool version"); 3732 } 3733 3734 /* Newly created pools with the right version are always deflated. */ 3735 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3736 spa->spa_deflate = TRUE; 3737 if (zap_add(spa->spa_meta_objset, 3738 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3739 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3740 cmn_err(CE_PANIC, "failed to add deflate"); 3741 } 3742 } 3743 3744 /* 3745 * Create the deferred-free bpobj. Turn off compression 3746 * because sync-to-convergence takes longer if the blocksize 3747 * keeps changing. 3748 */ 3749 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3750 dmu_object_set_compress(spa->spa_meta_objset, obj, 3751 ZIO_COMPRESS_OFF, tx); 3752 if (zap_add(spa->spa_meta_objset, 3753 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3754 sizeof (uint64_t), 1, &obj, tx) != 0) { 3755 cmn_err(CE_PANIC, "failed to add bpobj"); 3756 } 3757 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3758 spa->spa_meta_objset, obj)); 3759 3760 /* 3761 * Create the pool's history object. 3762 */ 3763 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3764 spa_history_create_obj(spa, tx); 3765 3766 /* 3767 * Generate some random noise for salted checksums to operate on. 3768 */ 3769 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3770 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3771 3772 /* 3773 * Set pool properties. 3774 */ 3775 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3776 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3777 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3778 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3779 3780 if (props != NULL) { 3781 spa_configfile_set(spa, props, B_FALSE); 3782 spa_sync_props(props, tx); 3783 } 3784 3785 dmu_tx_commit(tx); 3786 3787 spa->spa_sync_on = B_TRUE; 3788 txg_sync_start(spa->spa_dsl_pool); 3789 3790 /* 3791 * We explicitly wait for the first transaction to complete so that our 3792 * bean counters are appropriately updated. 3793 */ 3794 txg_wait_synced(spa->spa_dsl_pool, txg); 3795 3796 spa_config_sync(spa, B_FALSE, B_TRUE); 3797 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3798 3799 spa_history_log_version(spa, "create"); 3800 3801 /* 3802 * Don't count references from objsets that are already closed 3803 * and are making their way through the eviction process. 3804 */ 3805 spa_evicting_os_wait(spa); 3806 spa->spa_minref = refcount_count(&spa->spa_refcount); 3807 3808 mutex_exit(&spa_namespace_lock); 3809 3810 return (0); 3811} 3812 3813#ifdef _KERNEL 3814#ifdef illumos 3815/* 3816 * Get the root pool information from the root disk, then import the root pool 3817 * during the system boot up time. 3818 */ 3819extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3820 3821static nvlist_t * 3822spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3823{ 3824 nvlist_t *config; 3825 nvlist_t *nvtop, *nvroot; 3826 uint64_t pgid; 3827 3828 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3829 return (NULL); 3830 3831 /* 3832 * Add this top-level vdev to the child array. 3833 */ 3834 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3835 &nvtop) == 0); 3836 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3837 &pgid) == 0); 3838 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3839 3840 /* 3841 * Put this pool's top-level vdevs into a root vdev. 3842 */ 3843 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3844 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3845 VDEV_TYPE_ROOT) == 0); 3846 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3847 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3848 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3849 &nvtop, 1) == 0); 3850 3851 /* 3852 * Replace the existing vdev_tree with the new root vdev in 3853 * this pool's configuration (remove the old, add the new). 3854 */ 3855 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3856 nvlist_free(nvroot); 3857 return (config); 3858} 3859 3860/* 3861 * Walk the vdev tree and see if we can find a device with "better" 3862 * configuration. A configuration is "better" if the label on that 3863 * device has a more recent txg. 3864 */ 3865static void 3866spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3867{ 3868 for (int c = 0; c < vd->vdev_children; c++) 3869 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3870 3871 if (vd->vdev_ops->vdev_op_leaf) { 3872 nvlist_t *label; 3873 uint64_t label_txg; 3874 3875 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3876 &label) != 0) 3877 return; 3878 3879 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3880 &label_txg) == 0); 3881 3882 /* 3883 * Do we have a better boot device? 3884 */ 3885 if (label_txg > *txg) { 3886 *txg = label_txg; 3887 *avd = vd; 3888 } 3889 nvlist_free(label); 3890 } 3891} 3892 3893/* 3894 * Import a root pool. 3895 * 3896 * For x86. devpath_list will consist of devid and/or physpath name of 3897 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3898 * The GRUB "findroot" command will return the vdev we should boot. 3899 * 3900 * For Sparc, devpath_list consists the physpath name of the booting device 3901 * no matter the rootpool is a single device pool or a mirrored pool. 3902 * e.g. 3903 * "/pci@1f,0/ide@d/disk@0,0:a" 3904 */ 3905int 3906spa_import_rootpool(char *devpath, char *devid) 3907{ 3908 spa_t *spa; 3909 vdev_t *rvd, *bvd, *avd = NULL; 3910 nvlist_t *config, *nvtop; 3911 uint64_t guid, txg; 3912 char *pname; 3913 int error; 3914 3915 /* 3916 * Read the label from the boot device and generate a configuration. 3917 */ 3918 config = spa_generate_rootconf(devpath, devid, &guid); 3919#if defined(_OBP) && defined(_KERNEL) 3920 if (config == NULL) { 3921 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3922 /* iscsi boot */ 3923 get_iscsi_bootpath_phy(devpath); 3924 config = spa_generate_rootconf(devpath, devid, &guid); 3925 } 3926 } 3927#endif 3928 if (config == NULL) { 3929 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3930 devpath); 3931 return (SET_ERROR(EIO)); 3932 } 3933 3934 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3935 &pname) == 0); 3936 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3937 3938 mutex_enter(&spa_namespace_lock); 3939 if ((spa = spa_lookup(pname)) != NULL) { 3940 /* 3941 * Remove the existing root pool from the namespace so that we 3942 * can replace it with the correct config we just read in. 3943 */ 3944 spa_remove(spa); 3945 } 3946 3947 spa = spa_add(pname, config, NULL); 3948 spa->spa_is_root = B_TRUE; 3949 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3950 3951 /* 3952 * Build up a vdev tree based on the boot device's label config. 3953 */ 3954 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3955 &nvtop) == 0); 3956 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3957 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3958 VDEV_ALLOC_ROOTPOOL); 3959 spa_config_exit(spa, SCL_ALL, FTAG); 3960 if (error) { 3961 mutex_exit(&spa_namespace_lock); 3962 nvlist_free(config); 3963 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3964 pname); 3965 return (error); 3966 } 3967 3968 /* 3969 * Get the boot vdev. 3970 */ 3971 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3972 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3973 (u_longlong_t)guid); 3974 error = SET_ERROR(ENOENT); 3975 goto out; 3976 } 3977 3978 /* 3979 * Determine if there is a better boot device. 3980 */ 3981 avd = bvd; 3982 spa_alt_rootvdev(rvd, &avd, &txg); 3983 if (avd != bvd) { 3984 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3985 "try booting from '%s'", avd->vdev_path); 3986 error = SET_ERROR(EINVAL); 3987 goto out; 3988 } 3989 3990 /* 3991 * If the boot device is part of a spare vdev then ensure that 3992 * we're booting off the active spare. 3993 */ 3994 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3995 !bvd->vdev_isspare) { 3996 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3997 "try booting from '%s'", 3998 bvd->vdev_parent-> 3999 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4000 error = SET_ERROR(EINVAL); 4001 goto out; 4002 } 4003 4004 error = 0; 4005out: 4006 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4007 vdev_free(rvd); 4008 spa_config_exit(spa, SCL_ALL, FTAG); 4009 mutex_exit(&spa_namespace_lock); 4010 4011 nvlist_free(config); 4012 return (error); 4013} 4014 4015#else /* !illumos */ 4016 4017extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 4018 uint64_t *count); 4019 4020static nvlist_t * 4021spa_generate_rootconf(const char *name) 4022{ 4023 nvlist_t **configs, **tops; 4024 nvlist_t *config; 4025 nvlist_t *best_cfg, *nvtop, *nvroot; 4026 uint64_t *holes; 4027 uint64_t best_txg; 4028 uint64_t nchildren; 4029 uint64_t pgid; 4030 uint64_t count; 4031 uint64_t i; 4032 uint_t nholes; 4033 4034 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4035 return (NULL); 4036 4037 ASSERT3U(count, !=, 0); 4038 best_txg = 0; 4039 for (i = 0; i < count; i++) { 4040 uint64_t txg; 4041 4042 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4043 &txg) == 0); 4044 if (txg > best_txg) { 4045 best_txg = txg; 4046 best_cfg = configs[i]; 4047 } 4048 } 4049 4050 nchildren = 1; 4051 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4052 holes = NULL; 4053 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4054 &holes, &nholes); 4055 4056 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4057 for (i = 0; i < nchildren; i++) { 4058 if (i >= count) 4059 break; 4060 if (configs[i] == NULL) 4061 continue; 4062 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4063 &nvtop) == 0); 4064 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4065 } 4066 for (i = 0; holes != NULL && i < nholes; i++) { 4067 if (i >= nchildren) 4068 continue; 4069 if (tops[holes[i]] != NULL) 4070 continue; 4071 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4072 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4073 VDEV_TYPE_HOLE) == 0); 4074 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4075 holes[i]) == 0); 4076 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4077 0) == 0); 4078 } 4079 for (i = 0; i < nchildren; i++) { 4080 if (tops[i] != NULL) 4081 continue; 4082 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4083 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4084 VDEV_TYPE_MISSING) == 0); 4085 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4086 i) == 0); 4087 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4088 0) == 0); 4089 } 4090 4091 /* 4092 * Create pool config based on the best vdev config. 4093 */ 4094 nvlist_dup(best_cfg, &config, KM_SLEEP); 4095 4096 /* 4097 * Put this pool's top-level vdevs into a root vdev. 4098 */ 4099 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4100 &pgid) == 0); 4101 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4102 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4103 VDEV_TYPE_ROOT) == 0); 4104 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4105 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4106 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4107 tops, nchildren) == 0); 4108 4109 /* 4110 * Replace the existing vdev_tree with the new root vdev in 4111 * this pool's configuration (remove the old, add the new). 4112 */ 4113 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4114 4115 /* 4116 * Drop vdev config elements that should not be present at pool level. 4117 */ 4118 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4119 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4120 4121 for (i = 0; i < count; i++) 4122 nvlist_free(configs[i]); 4123 kmem_free(configs, count * sizeof(void *)); 4124 for (i = 0; i < nchildren; i++) 4125 nvlist_free(tops[i]); 4126 kmem_free(tops, nchildren * sizeof(void *)); 4127 nvlist_free(nvroot); 4128 return (config); 4129} 4130 4131int 4132spa_import_rootpool(const char *name) 4133{ 4134 spa_t *spa; 4135 vdev_t *rvd, *bvd, *avd = NULL; 4136 nvlist_t *config, *nvtop; 4137 uint64_t txg; 4138 char *pname; 4139 int error; 4140 4141 /* 4142 * Read the label from the boot device and generate a configuration. 4143 */ 4144 config = spa_generate_rootconf(name); 4145 4146 mutex_enter(&spa_namespace_lock); 4147 if (config != NULL) { 4148 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4149 &pname) == 0 && strcmp(name, pname) == 0); 4150 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4151 == 0); 4152 4153 if ((spa = spa_lookup(pname)) != NULL) { 4154 /* 4155 * Remove the existing root pool from the namespace so 4156 * that we can replace it with the correct config 4157 * we just read in. 4158 */ 4159 spa_remove(spa); 4160 } 4161 spa = spa_add(pname, config, NULL); 4162 4163 /* 4164 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4165 * via spa_version(). 4166 */ 4167 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4168 &spa->spa_ubsync.ub_version) != 0) 4169 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4170 } else if ((spa = spa_lookup(name)) == NULL) { 4171 mutex_exit(&spa_namespace_lock); 4172 nvlist_free(config); 4173 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4174 name); 4175 return (EIO); 4176 } else { 4177 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4178 } 4179 spa->spa_is_root = B_TRUE; 4180 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4181 4182 /* 4183 * Build up a vdev tree based on the boot device's label config. 4184 */ 4185 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4186 &nvtop) == 0); 4187 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4188 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4189 VDEV_ALLOC_ROOTPOOL); 4190 spa_config_exit(spa, SCL_ALL, FTAG); 4191 if (error) { 4192 mutex_exit(&spa_namespace_lock); 4193 nvlist_free(config); 4194 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4195 pname); 4196 return (error); 4197 } 4198 4199 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4200 vdev_free(rvd); 4201 spa_config_exit(spa, SCL_ALL, FTAG); 4202 mutex_exit(&spa_namespace_lock); 4203 4204 nvlist_free(config); 4205 return (0); 4206} 4207 4208#endif /* illumos */ 4209#endif /* _KERNEL */ 4210 4211/* 4212 * Import a non-root pool into the system. 4213 */ 4214int 4215spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4216{ 4217 spa_t *spa; 4218 char *altroot = NULL; 4219 spa_load_state_t state = SPA_LOAD_IMPORT; 4220 zpool_rewind_policy_t policy; 4221 uint64_t mode = spa_mode_global; 4222 uint64_t readonly = B_FALSE; 4223 int error; 4224 nvlist_t *nvroot; 4225 nvlist_t **spares, **l2cache; 4226 uint_t nspares, nl2cache; 4227 4228 /* 4229 * If a pool with this name exists, return failure. 4230 */ 4231 mutex_enter(&spa_namespace_lock); 4232 if (spa_lookup(pool) != NULL) { 4233 mutex_exit(&spa_namespace_lock); 4234 return (SET_ERROR(EEXIST)); 4235 } 4236 4237 /* 4238 * Create and initialize the spa structure. 4239 */ 4240 (void) nvlist_lookup_string(props, 4241 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4242 (void) nvlist_lookup_uint64(props, 4243 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4244 if (readonly) 4245 mode = FREAD; 4246 spa = spa_add(pool, config, altroot); 4247 spa->spa_import_flags = flags; 4248 4249 /* 4250 * Verbatim import - Take a pool and insert it into the namespace 4251 * as if it had been loaded at boot. 4252 */ 4253 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4254 if (props != NULL) 4255 spa_configfile_set(spa, props, B_FALSE); 4256 4257 spa_config_sync(spa, B_FALSE, B_TRUE); 4258 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4259 4260 mutex_exit(&spa_namespace_lock); 4261 return (0); 4262 } 4263 4264 spa_activate(spa, mode); 4265 4266 /* 4267 * Don't start async tasks until we know everything is healthy. 4268 */ 4269 spa_async_suspend(spa); 4270 4271 zpool_get_rewind_policy(config, &policy); 4272 if (policy.zrp_request & ZPOOL_DO_REWIND) 4273 state = SPA_LOAD_RECOVER; 4274 4275 /* 4276 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4277 * because the user-supplied config is actually the one to trust when 4278 * doing an import. 4279 */ 4280 if (state != SPA_LOAD_RECOVER) 4281 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4282 4283 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4284 policy.zrp_request); 4285 4286 /* 4287 * Propagate anything learned while loading the pool and pass it 4288 * back to caller (i.e. rewind info, missing devices, etc). 4289 */ 4290 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4291 spa->spa_load_info) == 0); 4292 4293 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4294 /* 4295 * Toss any existing sparelist, as it doesn't have any validity 4296 * anymore, and conflicts with spa_has_spare(). 4297 */ 4298 if (spa->spa_spares.sav_config) { 4299 nvlist_free(spa->spa_spares.sav_config); 4300 spa->spa_spares.sav_config = NULL; 4301 spa_load_spares(spa); 4302 } 4303 if (spa->spa_l2cache.sav_config) { 4304 nvlist_free(spa->spa_l2cache.sav_config); 4305 spa->spa_l2cache.sav_config = NULL; 4306 spa_load_l2cache(spa); 4307 } 4308 4309 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4310 &nvroot) == 0); 4311 if (error == 0) 4312 error = spa_validate_aux(spa, nvroot, -1ULL, 4313 VDEV_ALLOC_SPARE); 4314 if (error == 0) 4315 error = spa_validate_aux(spa, nvroot, -1ULL, 4316 VDEV_ALLOC_L2CACHE); 4317 spa_config_exit(spa, SCL_ALL, FTAG); 4318 4319 if (props != NULL) 4320 spa_configfile_set(spa, props, B_FALSE); 4321 4322 if (error != 0 || (props && spa_writeable(spa) && 4323 (error = spa_prop_set(spa, props)))) { 4324 spa_unload(spa); 4325 spa_deactivate(spa); 4326 spa_remove(spa); 4327 mutex_exit(&spa_namespace_lock); 4328 return (error); 4329 } 4330 4331 spa_async_resume(spa); 4332 4333 /* 4334 * Override any spares and level 2 cache devices as specified by 4335 * the user, as these may have correct device names/devids, etc. 4336 */ 4337 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4338 &spares, &nspares) == 0) { 4339 if (spa->spa_spares.sav_config) 4340 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4341 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4342 else 4343 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4344 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4345 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4346 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4347 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4348 spa_load_spares(spa); 4349 spa_config_exit(spa, SCL_ALL, FTAG); 4350 spa->spa_spares.sav_sync = B_TRUE; 4351 } 4352 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4353 &l2cache, &nl2cache) == 0) { 4354 if (spa->spa_l2cache.sav_config) 4355 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4356 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4357 else 4358 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4359 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4360 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4361 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4362 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4363 spa_load_l2cache(spa); 4364 spa_config_exit(spa, SCL_ALL, FTAG); 4365 spa->spa_l2cache.sav_sync = B_TRUE; 4366 } 4367 4368 /* 4369 * Check for any removed devices. 4370 */ 4371 if (spa->spa_autoreplace) { 4372 spa_aux_check_removed(&spa->spa_spares); 4373 spa_aux_check_removed(&spa->spa_l2cache); 4374 } 4375 4376 if (spa_writeable(spa)) { 4377 /* 4378 * Update the config cache to include the newly-imported pool. 4379 */ 4380 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4381 } 4382 4383 /* 4384 * It's possible that the pool was expanded while it was exported. 4385 * We kick off an async task to handle this for us. 4386 */ 4387 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4388 4389 spa_history_log_version(spa, "import"); 4390 4391 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4392 4393 mutex_exit(&spa_namespace_lock); 4394 4395#ifdef __FreeBSD__ 4396#ifdef _KERNEL 4397 zvol_create_minors(pool); 4398#endif 4399#endif 4400 return (0); 4401} 4402 4403nvlist_t * 4404spa_tryimport(nvlist_t *tryconfig) 4405{ 4406 nvlist_t *config = NULL; 4407 char *poolname; 4408 spa_t *spa; 4409 uint64_t state; 4410 int error; 4411 4412 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4413 return (NULL); 4414 4415 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4416 return (NULL); 4417 4418 /* 4419 * Create and initialize the spa structure. 4420 */ 4421 mutex_enter(&spa_namespace_lock); 4422 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4423 spa_activate(spa, FREAD); 4424 4425 /* 4426 * Pass off the heavy lifting to spa_load(). 4427 * Pass TRUE for mosconfig because the user-supplied config 4428 * is actually the one to trust when doing an import. 4429 */ 4430 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4431 4432 /* 4433 * If 'tryconfig' was at least parsable, return the current config. 4434 */ 4435 if (spa->spa_root_vdev != NULL) { 4436 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4437 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4438 poolname) == 0); 4439 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4440 state) == 0); 4441 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4442 spa->spa_uberblock.ub_timestamp) == 0); 4443 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4444 spa->spa_load_info) == 0); 4445 4446 /* 4447 * If the bootfs property exists on this pool then we 4448 * copy it out so that external consumers can tell which 4449 * pools are bootable. 4450 */ 4451 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4452 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4453 4454 /* 4455 * We have to play games with the name since the 4456 * pool was opened as TRYIMPORT_NAME. 4457 */ 4458 if (dsl_dsobj_to_dsname(spa_name(spa), 4459 spa->spa_bootfs, tmpname) == 0) { 4460 char *cp; 4461 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4462 4463 cp = strchr(tmpname, '/'); 4464 if (cp == NULL) { 4465 (void) strlcpy(dsname, tmpname, 4466 MAXPATHLEN); 4467 } else { 4468 (void) snprintf(dsname, MAXPATHLEN, 4469 "%s/%s", poolname, ++cp); 4470 } 4471 VERIFY(nvlist_add_string(config, 4472 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4473 kmem_free(dsname, MAXPATHLEN); 4474 } 4475 kmem_free(tmpname, MAXPATHLEN); 4476 } 4477 4478 /* 4479 * Add the list of hot spares and level 2 cache devices. 4480 */ 4481 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4482 spa_add_spares(spa, config); 4483 spa_add_l2cache(spa, config); 4484 spa_config_exit(spa, SCL_CONFIG, FTAG); 4485 } 4486 4487 spa_unload(spa); 4488 spa_deactivate(spa); 4489 spa_remove(spa); 4490 mutex_exit(&spa_namespace_lock); 4491 4492 return (config); 4493} 4494 4495/* 4496 * Pool export/destroy 4497 * 4498 * The act of destroying or exporting a pool is very simple. We make sure there 4499 * is no more pending I/O and any references to the pool are gone. Then, we 4500 * update the pool state and sync all the labels to disk, removing the 4501 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4502 * we don't sync the labels or remove the configuration cache. 4503 */ 4504static int 4505spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4506 boolean_t force, boolean_t hardforce) 4507{ 4508 spa_t *spa; 4509 4510 if (oldconfig) 4511 *oldconfig = NULL; 4512 4513 if (!(spa_mode_global & FWRITE)) 4514 return (SET_ERROR(EROFS)); 4515 4516 mutex_enter(&spa_namespace_lock); 4517 if ((spa = spa_lookup(pool)) == NULL) { 4518 mutex_exit(&spa_namespace_lock); 4519 return (SET_ERROR(ENOENT)); 4520 } 4521 4522 /* 4523 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4524 * reacquire the namespace lock, and see if we can export. 4525 */ 4526 spa_open_ref(spa, FTAG); 4527 mutex_exit(&spa_namespace_lock); 4528 spa_async_suspend(spa); 4529 mutex_enter(&spa_namespace_lock); 4530 spa_close(spa, FTAG); 4531 4532 /* 4533 * The pool will be in core if it's openable, 4534 * in which case we can modify its state. 4535 */ 4536 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4537 /* 4538 * Objsets may be open only because they're dirty, so we 4539 * have to force it to sync before checking spa_refcnt. 4540 */ 4541 txg_wait_synced(spa->spa_dsl_pool, 0); 4542 spa_evicting_os_wait(spa); 4543 4544 /* 4545 * A pool cannot be exported or destroyed if there are active 4546 * references. If we are resetting a pool, allow references by 4547 * fault injection handlers. 4548 */ 4549 if (!spa_refcount_zero(spa) || 4550 (spa->spa_inject_ref != 0 && 4551 new_state != POOL_STATE_UNINITIALIZED)) { 4552 spa_async_resume(spa); 4553 mutex_exit(&spa_namespace_lock); 4554 return (SET_ERROR(EBUSY)); 4555 } 4556 4557 /* 4558 * A pool cannot be exported if it has an active shared spare. 4559 * This is to prevent other pools stealing the active spare 4560 * from an exported pool. At user's own will, such pool can 4561 * be forcedly exported. 4562 */ 4563 if (!force && new_state == POOL_STATE_EXPORTED && 4564 spa_has_active_shared_spare(spa)) { 4565 spa_async_resume(spa); 4566 mutex_exit(&spa_namespace_lock); 4567 return (SET_ERROR(EXDEV)); 4568 } 4569 4570 /* 4571 * We want this to be reflected on every label, 4572 * so mark them all dirty. spa_unload() will do the 4573 * final sync that pushes these changes out. 4574 */ 4575 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4576 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4577 spa->spa_state = new_state; 4578 spa->spa_final_txg = spa_last_synced_txg(spa) + 4579 TXG_DEFER_SIZE + 1; 4580 vdev_config_dirty(spa->spa_root_vdev); 4581 spa_config_exit(spa, SCL_ALL, FTAG); 4582 } 4583 } 4584 4585 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4586 4587 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4588 spa_unload(spa); 4589 spa_deactivate(spa); 4590 } 4591 4592 if (oldconfig && spa->spa_config) 4593 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4594 4595 if (new_state != POOL_STATE_UNINITIALIZED) { 4596 if (!hardforce) 4597 spa_config_sync(spa, B_TRUE, B_TRUE); 4598 spa_remove(spa); 4599 } 4600 mutex_exit(&spa_namespace_lock); 4601 4602 return (0); 4603} 4604 4605/* 4606 * Destroy a storage pool. 4607 */ 4608int 4609spa_destroy(char *pool) 4610{ 4611 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4612 B_FALSE, B_FALSE)); 4613} 4614 4615/* 4616 * Export a storage pool. 4617 */ 4618int 4619spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4620 boolean_t hardforce) 4621{ 4622 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4623 force, hardforce)); 4624} 4625 4626/* 4627 * Similar to spa_export(), this unloads the spa_t without actually removing it 4628 * from the namespace in any way. 4629 */ 4630int 4631spa_reset(char *pool) 4632{ 4633 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4634 B_FALSE, B_FALSE)); 4635} 4636 4637/* 4638 * ========================================================================== 4639 * Device manipulation 4640 * ========================================================================== 4641 */ 4642 4643/* 4644 * Add a device to a storage pool. 4645 */ 4646int 4647spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4648{ 4649 uint64_t txg, id; 4650 int error; 4651 vdev_t *rvd = spa->spa_root_vdev; 4652 vdev_t *vd, *tvd; 4653 nvlist_t **spares, **l2cache; 4654 uint_t nspares, nl2cache; 4655 4656 ASSERT(spa_writeable(spa)); 4657 4658 txg = spa_vdev_enter(spa); 4659 4660 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4661 VDEV_ALLOC_ADD)) != 0) 4662 return (spa_vdev_exit(spa, NULL, txg, error)); 4663 4664 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4665 4666 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4667 &nspares) != 0) 4668 nspares = 0; 4669 4670 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4671 &nl2cache) != 0) 4672 nl2cache = 0; 4673 4674 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4675 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4676 4677 if (vd->vdev_children != 0 && 4678 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4679 return (spa_vdev_exit(spa, vd, txg, error)); 4680 4681 /* 4682 * We must validate the spares and l2cache devices after checking the 4683 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4684 */ 4685 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4686 return (spa_vdev_exit(spa, vd, txg, error)); 4687 4688 /* 4689 * Transfer each new top-level vdev from vd to rvd. 4690 */ 4691 for (int c = 0; c < vd->vdev_children; c++) { 4692 4693 /* 4694 * Set the vdev id to the first hole, if one exists. 4695 */ 4696 for (id = 0; id < rvd->vdev_children; id++) { 4697 if (rvd->vdev_child[id]->vdev_ishole) { 4698 vdev_free(rvd->vdev_child[id]); 4699 break; 4700 } 4701 } 4702 tvd = vd->vdev_child[c]; 4703 vdev_remove_child(vd, tvd); 4704 tvd->vdev_id = id; 4705 vdev_add_child(rvd, tvd); 4706 vdev_config_dirty(tvd); 4707 } 4708 4709 if (nspares != 0) { 4710 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4711 ZPOOL_CONFIG_SPARES); 4712 spa_load_spares(spa); 4713 spa->spa_spares.sav_sync = B_TRUE; 4714 } 4715 4716 if (nl2cache != 0) { 4717 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4718 ZPOOL_CONFIG_L2CACHE); 4719 spa_load_l2cache(spa); 4720 spa->spa_l2cache.sav_sync = B_TRUE; 4721 } 4722 4723 /* 4724 * We have to be careful when adding new vdevs to an existing pool. 4725 * If other threads start allocating from these vdevs before we 4726 * sync the config cache, and we lose power, then upon reboot we may 4727 * fail to open the pool because there are DVAs that the config cache 4728 * can't translate. Therefore, we first add the vdevs without 4729 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4730 * and then let spa_config_update() initialize the new metaslabs. 4731 * 4732 * spa_load() checks for added-but-not-initialized vdevs, so that 4733 * if we lose power at any point in this sequence, the remaining 4734 * steps will be completed the next time we load the pool. 4735 */ 4736 (void) spa_vdev_exit(spa, vd, txg, 0); 4737 4738 mutex_enter(&spa_namespace_lock); 4739 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4740 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 4741 mutex_exit(&spa_namespace_lock); 4742 4743 return (0); 4744} 4745 4746/* 4747 * Attach a device to a mirror. The arguments are the path to any device 4748 * in the mirror, and the nvroot for the new device. If the path specifies 4749 * a device that is not mirrored, we automatically insert the mirror vdev. 4750 * 4751 * If 'replacing' is specified, the new device is intended to replace the 4752 * existing device; in this case the two devices are made into their own 4753 * mirror using the 'replacing' vdev, which is functionally identical to 4754 * the mirror vdev (it actually reuses all the same ops) but has a few 4755 * extra rules: you can't attach to it after it's been created, and upon 4756 * completion of resilvering, the first disk (the one being replaced) 4757 * is automatically detached. 4758 */ 4759int 4760spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4761{ 4762 uint64_t txg, dtl_max_txg; 4763 vdev_t *rvd = spa->spa_root_vdev; 4764 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4765 vdev_ops_t *pvops; 4766 char *oldvdpath, *newvdpath; 4767 int newvd_isspare; 4768 int error; 4769 4770 ASSERT(spa_writeable(spa)); 4771 4772 txg = spa_vdev_enter(spa); 4773 4774 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4775 4776 if (oldvd == NULL) 4777 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4778 4779 if (!oldvd->vdev_ops->vdev_op_leaf) 4780 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4781 4782 pvd = oldvd->vdev_parent; 4783 4784 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4785 VDEV_ALLOC_ATTACH)) != 0) 4786 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4787 4788 if (newrootvd->vdev_children != 1) 4789 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4790 4791 newvd = newrootvd->vdev_child[0]; 4792 4793 if (!newvd->vdev_ops->vdev_op_leaf) 4794 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4795 4796 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4797 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4798 4799 /* 4800 * Spares can't replace logs 4801 */ 4802 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4803 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4804 4805 if (!replacing) { 4806 /* 4807 * For attach, the only allowable parent is a mirror or the root 4808 * vdev. 4809 */ 4810 if (pvd->vdev_ops != &vdev_mirror_ops && 4811 pvd->vdev_ops != &vdev_root_ops) 4812 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4813 4814 pvops = &vdev_mirror_ops; 4815 } else { 4816 /* 4817 * Active hot spares can only be replaced by inactive hot 4818 * spares. 4819 */ 4820 if (pvd->vdev_ops == &vdev_spare_ops && 4821 oldvd->vdev_isspare && 4822 !spa_has_spare(spa, newvd->vdev_guid)) 4823 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4824 4825 /* 4826 * If the source is a hot spare, and the parent isn't already a 4827 * spare, then we want to create a new hot spare. Otherwise, we 4828 * want to create a replacing vdev. The user is not allowed to 4829 * attach to a spared vdev child unless the 'isspare' state is 4830 * the same (spare replaces spare, non-spare replaces 4831 * non-spare). 4832 */ 4833 if (pvd->vdev_ops == &vdev_replacing_ops && 4834 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4835 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4836 } else if (pvd->vdev_ops == &vdev_spare_ops && 4837 newvd->vdev_isspare != oldvd->vdev_isspare) { 4838 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4839 } 4840 4841 if (newvd->vdev_isspare) 4842 pvops = &vdev_spare_ops; 4843 else 4844 pvops = &vdev_replacing_ops; 4845 } 4846 4847 /* 4848 * Make sure the new device is big enough. 4849 */ 4850 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4851 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4852 4853 /* 4854 * The new device cannot have a higher alignment requirement 4855 * than the top-level vdev. 4856 */ 4857 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4858 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4859 4860 /* 4861 * If this is an in-place replacement, update oldvd's path and devid 4862 * to make it distinguishable from newvd, and unopenable from now on. 4863 */ 4864 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4865 spa_strfree(oldvd->vdev_path); 4866 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4867 KM_SLEEP); 4868 (void) sprintf(oldvd->vdev_path, "%s/%s", 4869 newvd->vdev_path, "old"); 4870 if (oldvd->vdev_devid != NULL) { 4871 spa_strfree(oldvd->vdev_devid); 4872 oldvd->vdev_devid = NULL; 4873 } 4874 } 4875 4876 /* mark the device being resilvered */ 4877 newvd->vdev_resilver_txg = txg; 4878 4879 /* 4880 * If the parent is not a mirror, or if we're replacing, insert the new 4881 * mirror/replacing/spare vdev above oldvd. 4882 */ 4883 if (pvd->vdev_ops != pvops) 4884 pvd = vdev_add_parent(oldvd, pvops); 4885 4886 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4887 ASSERT(pvd->vdev_ops == pvops); 4888 ASSERT(oldvd->vdev_parent == pvd); 4889 4890 /* 4891 * Extract the new device from its root and add it to pvd. 4892 */ 4893 vdev_remove_child(newrootvd, newvd); 4894 newvd->vdev_id = pvd->vdev_children; 4895 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4896 vdev_add_child(pvd, newvd); 4897 4898 tvd = newvd->vdev_top; 4899 ASSERT(pvd->vdev_top == tvd); 4900 ASSERT(tvd->vdev_parent == rvd); 4901 4902 vdev_config_dirty(tvd); 4903 4904 /* 4905 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4906 * for any dmu_sync-ed blocks. It will propagate upward when 4907 * spa_vdev_exit() calls vdev_dtl_reassess(). 4908 */ 4909 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4910 4911 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4912 dtl_max_txg - TXG_INITIAL); 4913 4914 if (newvd->vdev_isspare) { 4915 spa_spare_activate(newvd); 4916 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4917 } 4918 4919 oldvdpath = spa_strdup(oldvd->vdev_path); 4920 newvdpath = spa_strdup(newvd->vdev_path); 4921 newvd_isspare = newvd->vdev_isspare; 4922 4923 /* 4924 * Mark newvd's DTL dirty in this txg. 4925 */ 4926 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4927 4928 /* 4929 * Schedule the resilver to restart in the future. We do this to 4930 * ensure that dmu_sync-ed blocks have been stitched into the 4931 * respective datasets. 4932 */ 4933 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4934 4935 if (spa->spa_bootfs) 4936 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4937 4938 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 4939 4940 /* 4941 * Commit the config 4942 */ 4943 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4944 4945 spa_history_log_internal(spa, "vdev attach", NULL, 4946 "%s vdev=%s %s vdev=%s", 4947 replacing && newvd_isspare ? "spare in" : 4948 replacing ? "replace" : "attach", newvdpath, 4949 replacing ? "for" : "to", oldvdpath); 4950 4951 spa_strfree(oldvdpath); 4952 spa_strfree(newvdpath); 4953 4954 return (0); 4955} 4956 4957/* 4958 * Detach a device from a mirror or replacing vdev. 4959 * 4960 * If 'replace_done' is specified, only detach if the parent 4961 * is a replacing vdev. 4962 */ 4963int 4964spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4965{ 4966 uint64_t txg; 4967 int error; 4968 vdev_t *rvd = spa->spa_root_vdev; 4969 vdev_t *vd, *pvd, *cvd, *tvd; 4970 boolean_t unspare = B_FALSE; 4971 uint64_t unspare_guid = 0; 4972 char *vdpath; 4973 4974 ASSERT(spa_writeable(spa)); 4975 4976 txg = spa_vdev_enter(spa); 4977 4978 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4979 4980 if (vd == NULL) 4981 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4982 4983 if (!vd->vdev_ops->vdev_op_leaf) 4984 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4985 4986 pvd = vd->vdev_parent; 4987 4988 /* 4989 * If the parent/child relationship is not as expected, don't do it. 4990 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4991 * vdev that's replacing B with C. The user's intent in replacing 4992 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4993 * the replace by detaching C, the expected behavior is to end up 4994 * M(A,B). But suppose that right after deciding to detach C, 4995 * the replacement of B completes. We would have M(A,C), and then 4996 * ask to detach C, which would leave us with just A -- not what 4997 * the user wanted. To prevent this, we make sure that the 4998 * parent/child relationship hasn't changed -- in this example, 4999 * that C's parent is still the replacing vdev R. 5000 */ 5001 if (pvd->vdev_guid != pguid && pguid != 0) 5002 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5003 5004 /* 5005 * Only 'replacing' or 'spare' vdevs can be replaced. 5006 */ 5007 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5008 pvd->vdev_ops != &vdev_spare_ops) 5009 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5010 5011 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5012 spa_version(spa) >= SPA_VERSION_SPARES); 5013 5014 /* 5015 * Only mirror, replacing, and spare vdevs support detach. 5016 */ 5017 if (pvd->vdev_ops != &vdev_replacing_ops && 5018 pvd->vdev_ops != &vdev_mirror_ops && 5019 pvd->vdev_ops != &vdev_spare_ops) 5020 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5021 5022 /* 5023 * If this device has the only valid copy of some data, 5024 * we cannot safely detach it. 5025 */ 5026 if (vdev_dtl_required(vd)) 5027 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5028 5029 ASSERT(pvd->vdev_children >= 2); 5030 5031 /* 5032 * If we are detaching the second disk from a replacing vdev, then 5033 * check to see if we changed the original vdev's path to have "/old" 5034 * at the end in spa_vdev_attach(). If so, undo that change now. 5035 */ 5036 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5037 vd->vdev_path != NULL) { 5038 size_t len = strlen(vd->vdev_path); 5039 5040 for (int c = 0; c < pvd->vdev_children; c++) { 5041 cvd = pvd->vdev_child[c]; 5042 5043 if (cvd == vd || cvd->vdev_path == NULL) 5044 continue; 5045 5046 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5047 strcmp(cvd->vdev_path + len, "/old") == 0) { 5048 spa_strfree(cvd->vdev_path); 5049 cvd->vdev_path = spa_strdup(vd->vdev_path); 5050 break; 5051 } 5052 } 5053 } 5054 5055 /* 5056 * If we are detaching the original disk from a spare, then it implies 5057 * that the spare should become a real disk, and be removed from the 5058 * active spare list for the pool. 5059 */ 5060 if (pvd->vdev_ops == &vdev_spare_ops && 5061 vd->vdev_id == 0 && 5062 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5063 unspare = B_TRUE; 5064 5065 /* 5066 * Erase the disk labels so the disk can be used for other things. 5067 * This must be done after all other error cases are handled, 5068 * but before we disembowel vd (so we can still do I/O to it). 5069 * But if we can't do it, don't treat the error as fatal -- 5070 * it may be that the unwritability of the disk is the reason 5071 * it's being detached! 5072 */ 5073 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5074 5075 /* 5076 * Remove vd from its parent and compact the parent's children. 5077 */ 5078 vdev_remove_child(pvd, vd); 5079 vdev_compact_children(pvd); 5080 5081 /* 5082 * Remember one of the remaining children so we can get tvd below. 5083 */ 5084 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5085 5086 /* 5087 * If we need to remove the remaining child from the list of hot spares, 5088 * do it now, marking the vdev as no longer a spare in the process. 5089 * We must do this before vdev_remove_parent(), because that can 5090 * change the GUID if it creates a new toplevel GUID. For a similar 5091 * reason, we must remove the spare now, in the same txg as the detach; 5092 * otherwise someone could attach a new sibling, change the GUID, and 5093 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5094 */ 5095 if (unspare) { 5096 ASSERT(cvd->vdev_isspare); 5097 spa_spare_remove(cvd); 5098 unspare_guid = cvd->vdev_guid; 5099 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5100 cvd->vdev_unspare = B_TRUE; 5101 } 5102 5103 /* 5104 * If the parent mirror/replacing vdev only has one child, 5105 * the parent is no longer needed. Remove it from the tree. 5106 */ 5107 if (pvd->vdev_children == 1) { 5108 if (pvd->vdev_ops == &vdev_spare_ops) 5109 cvd->vdev_unspare = B_FALSE; 5110 vdev_remove_parent(cvd); 5111 } 5112 5113 5114 /* 5115 * We don't set tvd until now because the parent we just removed 5116 * may have been the previous top-level vdev. 5117 */ 5118 tvd = cvd->vdev_top; 5119 ASSERT(tvd->vdev_parent == rvd); 5120 5121 /* 5122 * Reevaluate the parent vdev state. 5123 */ 5124 vdev_propagate_state(cvd); 5125 5126 /* 5127 * If the 'autoexpand' property is set on the pool then automatically 5128 * try to expand the size of the pool. For example if the device we 5129 * just detached was smaller than the others, it may be possible to 5130 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5131 * first so that we can obtain the updated sizes of the leaf vdevs. 5132 */ 5133 if (spa->spa_autoexpand) { 5134 vdev_reopen(tvd); 5135 vdev_expand(tvd, txg); 5136 } 5137 5138 vdev_config_dirty(tvd); 5139 5140 /* 5141 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5142 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5143 * But first make sure we're not on any *other* txg's DTL list, to 5144 * prevent vd from being accessed after it's freed. 5145 */ 5146 vdpath = spa_strdup(vd->vdev_path); 5147 for (int t = 0; t < TXG_SIZE; t++) 5148 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5149 vd->vdev_detached = B_TRUE; 5150 vdev_dirty(tvd, VDD_DTL, vd, txg); 5151 5152 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5153 5154 /* hang on to the spa before we release the lock */ 5155 spa_open_ref(spa, FTAG); 5156 5157 error = spa_vdev_exit(spa, vd, txg, 0); 5158 5159 spa_history_log_internal(spa, "detach", NULL, 5160 "vdev=%s", vdpath); 5161 spa_strfree(vdpath); 5162 5163 /* 5164 * If this was the removal of the original device in a hot spare vdev, 5165 * then we want to go through and remove the device from the hot spare 5166 * list of every other pool. 5167 */ 5168 if (unspare) { 5169 spa_t *altspa = NULL; 5170 5171 mutex_enter(&spa_namespace_lock); 5172 while ((altspa = spa_next(altspa)) != NULL) { 5173 if (altspa->spa_state != POOL_STATE_ACTIVE || 5174 altspa == spa) 5175 continue; 5176 5177 spa_open_ref(altspa, FTAG); 5178 mutex_exit(&spa_namespace_lock); 5179 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5180 mutex_enter(&spa_namespace_lock); 5181 spa_close(altspa, FTAG); 5182 } 5183 mutex_exit(&spa_namespace_lock); 5184 5185 /* search the rest of the vdevs for spares to remove */ 5186 spa_vdev_resilver_done(spa); 5187 } 5188 5189 /* all done with the spa; OK to release */ 5190 mutex_enter(&spa_namespace_lock); 5191 spa_close(spa, FTAG); 5192 mutex_exit(&spa_namespace_lock); 5193 5194 return (error); 5195} 5196 5197/* 5198 * Split a set of devices from their mirrors, and create a new pool from them. 5199 */ 5200int 5201spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5202 nvlist_t *props, boolean_t exp) 5203{ 5204 int error = 0; 5205 uint64_t txg, *glist; 5206 spa_t *newspa; 5207 uint_t c, children, lastlog; 5208 nvlist_t **child, *nvl, *tmp; 5209 dmu_tx_t *tx; 5210 char *altroot = NULL; 5211 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5212 boolean_t activate_slog; 5213 5214 ASSERT(spa_writeable(spa)); 5215 5216 txg = spa_vdev_enter(spa); 5217 5218 /* clear the log and flush everything up to now */ 5219 activate_slog = spa_passivate_log(spa); 5220 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5221 error = spa_offline_log(spa); 5222 txg = spa_vdev_config_enter(spa); 5223 5224 if (activate_slog) 5225 spa_activate_log(spa); 5226 5227 if (error != 0) 5228 return (spa_vdev_exit(spa, NULL, txg, error)); 5229 5230 /* check new spa name before going any further */ 5231 if (spa_lookup(newname) != NULL) 5232 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5233 5234 /* 5235 * scan through all the children to ensure they're all mirrors 5236 */ 5237 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5238 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5239 &children) != 0) 5240 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5241 5242 /* first, check to ensure we've got the right child count */ 5243 rvd = spa->spa_root_vdev; 5244 lastlog = 0; 5245 for (c = 0; c < rvd->vdev_children; c++) { 5246 vdev_t *vd = rvd->vdev_child[c]; 5247 5248 /* don't count the holes & logs as children */ 5249 if (vd->vdev_islog || vd->vdev_ishole) { 5250 if (lastlog == 0) 5251 lastlog = c; 5252 continue; 5253 } 5254 5255 lastlog = 0; 5256 } 5257 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5258 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5259 5260 /* next, ensure no spare or cache devices are part of the split */ 5261 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5262 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5263 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5264 5265 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5266 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5267 5268 /* then, loop over each vdev and validate it */ 5269 for (c = 0; c < children; c++) { 5270 uint64_t is_hole = 0; 5271 5272 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5273 &is_hole); 5274 5275 if (is_hole != 0) { 5276 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5277 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5278 continue; 5279 } else { 5280 error = SET_ERROR(EINVAL); 5281 break; 5282 } 5283 } 5284 5285 /* which disk is going to be split? */ 5286 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5287 &glist[c]) != 0) { 5288 error = SET_ERROR(EINVAL); 5289 break; 5290 } 5291 5292 /* look it up in the spa */ 5293 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5294 if (vml[c] == NULL) { 5295 error = SET_ERROR(ENODEV); 5296 break; 5297 } 5298 5299 /* make sure there's nothing stopping the split */ 5300 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5301 vml[c]->vdev_islog || 5302 vml[c]->vdev_ishole || 5303 vml[c]->vdev_isspare || 5304 vml[c]->vdev_isl2cache || 5305 !vdev_writeable(vml[c]) || 5306 vml[c]->vdev_children != 0 || 5307 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5308 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5309 error = SET_ERROR(EINVAL); 5310 break; 5311 } 5312 5313 if (vdev_dtl_required(vml[c])) { 5314 error = SET_ERROR(EBUSY); 5315 break; 5316 } 5317 5318 /* we need certain info from the top level */ 5319 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5320 vml[c]->vdev_top->vdev_ms_array) == 0); 5321 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5322 vml[c]->vdev_top->vdev_ms_shift) == 0); 5323 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5324 vml[c]->vdev_top->vdev_asize) == 0); 5325 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5326 vml[c]->vdev_top->vdev_ashift) == 0); 5327 } 5328 5329 if (error != 0) { 5330 kmem_free(vml, children * sizeof (vdev_t *)); 5331 kmem_free(glist, children * sizeof (uint64_t)); 5332 return (spa_vdev_exit(spa, NULL, txg, error)); 5333 } 5334 5335 /* stop writers from using the disks */ 5336 for (c = 0; c < children; c++) { 5337 if (vml[c] != NULL) 5338 vml[c]->vdev_offline = B_TRUE; 5339 } 5340 vdev_reopen(spa->spa_root_vdev); 5341 5342 /* 5343 * Temporarily record the splitting vdevs in the spa config. This 5344 * will disappear once the config is regenerated. 5345 */ 5346 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5347 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5348 glist, children) == 0); 5349 kmem_free(glist, children * sizeof (uint64_t)); 5350 5351 mutex_enter(&spa->spa_props_lock); 5352 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5353 nvl) == 0); 5354 mutex_exit(&spa->spa_props_lock); 5355 spa->spa_config_splitting = nvl; 5356 vdev_config_dirty(spa->spa_root_vdev); 5357 5358 /* configure and create the new pool */ 5359 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5360 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5361 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5362 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5363 spa_version(spa)) == 0); 5364 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5365 spa->spa_config_txg) == 0); 5366 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5367 spa_generate_guid(NULL)) == 0); 5368 (void) nvlist_lookup_string(props, 5369 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5370 5371 /* add the new pool to the namespace */ 5372 newspa = spa_add(newname, config, altroot); 5373 newspa->spa_config_txg = spa->spa_config_txg; 5374 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5375 5376 /* release the spa config lock, retaining the namespace lock */ 5377 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5378 5379 if (zio_injection_enabled) 5380 zio_handle_panic_injection(spa, FTAG, 1); 5381 5382 spa_activate(newspa, spa_mode_global); 5383 spa_async_suspend(newspa); 5384 5385#ifndef illumos 5386 /* mark that we are creating new spa by splitting */ 5387 newspa->spa_splitting_newspa = B_TRUE; 5388#endif 5389 /* create the new pool from the disks of the original pool */ 5390 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5391#ifndef illumos 5392 newspa->spa_splitting_newspa = B_FALSE; 5393#endif 5394 if (error) 5395 goto out; 5396 5397 /* if that worked, generate a real config for the new pool */ 5398 if (newspa->spa_root_vdev != NULL) { 5399 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5400 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5401 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5402 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5403 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5404 B_TRUE)); 5405 } 5406 5407 /* set the props */ 5408 if (props != NULL) { 5409 spa_configfile_set(newspa, props, B_FALSE); 5410 error = spa_prop_set(newspa, props); 5411 if (error) 5412 goto out; 5413 } 5414 5415 /* flush everything */ 5416 txg = spa_vdev_config_enter(newspa); 5417 vdev_config_dirty(newspa->spa_root_vdev); 5418 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5419 5420 if (zio_injection_enabled) 5421 zio_handle_panic_injection(spa, FTAG, 2); 5422 5423 spa_async_resume(newspa); 5424 5425 /* finally, update the original pool's config */ 5426 txg = spa_vdev_config_enter(spa); 5427 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5428 error = dmu_tx_assign(tx, TXG_WAIT); 5429 if (error != 0) 5430 dmu_tx_abort(tx); 5431 for (c = 0; c < children; c++) { 5432 if (vml[c] != NULL) { 5433 vdev_split(vml[c]); 5434 if (error == 0) 5435 spa_history_log_internal(spa, "detach", tx, 5436 "vdev=%s", vml[c]->vdev_path); 5437 vdev_free(vml[c]); 5438 } 5439 } 5440 vdev_config_dirty(spa->spa_root_vdev); 5441 spa->spa_config_splitting = NULL; 5442 nvlist_free(nvl); 5443 if (error == 0) 5444 dmu_tx_commit(tx); 5445 (void) spa_vdev_exit(spa, NULL, txg, 0); 5446 5447 if (zio_injection_enabled) 5448 zio_handle_panic_injection(spa, FTAG, 3); 5449 5450 /* split is complete; log a history record */ 5451 spa_history_log_internal(newspa, "split", NULL, 5452 "from pool %s", spa_name(spa)); 5453 5454 kmem_free(vml, children * sizeof (vdev_t *)); 5455 5456 /* if we're not going to mount the filesystems in userland, export */ 5457 if (exp) 5458 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5459 B_FALSE, B_FALSE); 5460 5461 return (error); 5462 5463out: 5464 spa_unload(newspa); 5465 spa_deactivate(newspa); 5466 spa_remove(newspa); 5467 5468 txg = spa_vdev_config_enter(spa); 5469 5470 /* re-online all offlined disks */ 5471 for (c = 0; c < children; c++) { 5472 if (vml[c] != NULL) 5473 vml[c]->vdev_offline = B_FALSE; 5474 } 5475 vdev_reopen(spa->spa_root_vdev); 5476 5477 nvlist_free(spa->spa_config_splitting); 5478 spa->spa_config_splitting = NULL; 5479 (void) spa_vdev_exit(spa, NULL, txg, error); 5480 5481 kmem_free(vml, children * sizeof (vdev_t *)); 5482 return (error); 5483} 5484 5485static nvlist_t * 5486spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5487{ 5488 for (int i = 0; i < count; i++) { 5489 uint64_t guid; 5490 5491 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5492 &guid) == 0); 5493 5494 if (guid == target_guid) 5495 return (nvpp[i]); 5496 } 5497 5498 return (NULL); 5499} 5500 5501static void 5502spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5503 nvlist_t *dev_to_remove) 5504{ 5505 nvlist_t **newdev = NULL; 5506 5507 if (count > 1) 5508 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5509 5510 for (int i = 0, j = 0; i < count; i++) { 5511 if (dev[i] == dev_to_remove) 5512 continue; 5513 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5514 } 5515 5516 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5517 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5518 5519 for (int i = 0; i < count - 1; i++) 5520 nvlist_free(newdev[i]); 5521 5522 if (count > 1) 5523 kmem_free(newdev, (count - 1) * sizeof (void *)); 5524} 5525 5526/* 5527 * Evacuate the device. 5528 */ 5529static int 5530spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5531{ 5532 uint64_t txg; 5533 int error = 0; 5534 5535 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5536 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5537 ASSERT(vd == vd->vdev_top); 5538 5539 /* 5540 * Evacuate the device. We don't hold the config lock as writer 5541 * since we need to do I/O but we do keep the 5542 * spa_namespace_lock held. Once this completes the device 5543 * should no longer have any blocks allocated on it. 5544 */ 5545 if (vd->vdev_islog) { 5546 if (vd->vdev_stat.vs_alloc != 0) 5547 error = spa_offline_log(spa); 5548 } else { 5549 error = SET_ERROR(ENOTSUP); 5550 } 5551 5552 if (error) 5553 return (error); 5554 5555 /* 5556 * The evacuation succeeded. Remove any remaining MOS metadata 5557 * associated with this vdev, and wait for these changes to sync. 5558 */ 5559 ASSERT0(vd->vdev_stat.vs_alloc); 5560 txg = spa_vdev_config_enter(spa); 5561 vd->vdev_removing = B_TRUE; 5562 vdev_dirty_leaves(vd, VDD_DTL, txg); 5563 vdev_config_dirty(vd); 5564 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5565 5566 return (0); 5567} 5568 5569/* 5570 * Complete the removal by cleaning up the namespace. 5571 */ 5572static void 5573spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5574{ 5575 vdev_t *rvd = spa->spa_root_vdev; 5576 uint64_t id = vd->vdev_id; 5577 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5578 5579 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5580 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5581 ASSERT(vd == vd->vdev_top); 5582 5583 /* 5584 * Only remove any devices which are empty. 5585 */ 5586 if (vd->vdev_stat.vs_alloc != 0) 5587 return; 5588 5589 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5590 5591 if (list_link_active(&vd->vdev_state_dirty_node)) 5592 vdev_state_clean(vd); 5593 if (list_link_active(&vd->vdev_config_dirty_node)) 5594 vdev_config_clean(vd); 5595 5596 vdev_free(vd); 5597 5598 if (last_vdev) { 5599 vdev_compact_children(rvd); 5600 } else { 5601 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5602 vdev_add_child(rvd, vd); 5603 } 5604 vdev_config_dirty(rvd); 5605 5606 /* 5607 * Reassess the health of our root vdev. 5608 */ 5609 vdev_reopen(rvd); 5610} 5611 5612/* 5613 * Remove a device from the pool - 5614 * 5615 * Removing a device from the vdev namespace requires several steps 5616 * and can take a significant amount of time. As a result we use 5617 * the spa_vdev_config_[enter/exit] functions which allow us to 5618 * grab and release the spa_config_lock while still holding the namespace 5619 * lock. During each step the configuration is synced out. 5620 * 5621 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5622 * devices. 5623 */ 5624int 5625spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5626{ 5627 vdev_t *vd; 5628 metaslab_group_t *mg; 5629 nvlist_t **spares, **l2cache, *nv; 5630 uint64_t txg = 0; 5631 uint_t nspares, nl2cache; 5632 int error = 0; 5633 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5634 5635 ASSERT(spa_writeable(spa)); 5636 5637 if (!locked) 5638 txg = spa_vdev_enter(spa); 5639 5640 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5641 5642 if (spa->spa_spares.sav_vdevs != NULL && 5643 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5644 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5645 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5646 /* 5647 * Only remove the hot spare if it's not currently in use 5648 * in this pool. 5649 */ 5650 if (vd == NULL || unspare) { 5651 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5652 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5653 spa_load_spares(spa); 5654 spa->spa_spares.sav_sync = B_TRUE; 5655 } else { 5656 error = SET_ERROR(EBUSY); 5657 } 5658 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5659 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5660 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5661 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5662 /* 5663 * Cache devices can always be removed. 5664 */ 5665 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5666 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5667 spa_load_l2cache(spa); 5668 spa->spa_l2cache.sav_sync = B_TRUE; 5669 } else if (vd != NULL && vd->vdev_islog) { 5670 ASSERT(!locked); 5671 ASSERT(vd == vd->vdev_top); 5672 5673 mg = vd->vdev_mg; 5674 5675 /* 5676 * Stop allocating from this vdev. 5677 */ 5678 metaslab_group_passivate(mg); 5679 5680 /* 5681 * Wait for the youngest allocations and frees to sync, 5682 * and then wait for the deferral of those frees to finish. 5683 */ 5684 spa_vdev_config_exit(spa, NULL, 5685 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5686 5687 /* 5688 * Attempt to evacuate the vdev. 5689 */ 5690 error = spa_vdev_remove_evacuate(spa, vd); 5691 5692 txg = spa_vdev_config_enter(spa); 5693 5694 /* 5695 * If we couldn't evacuate the vdev, unwind. 5696 */ 5697 if (error) { 5698 metaslab_group_activate(mg); 5699 return (spa_vdev_exit(spa, NULL, txg, error)); 5700 } 5701 5702 /* 5703 * Clean up the vdev namespace. 5704 */ 5705 spa_vdev_remove_from_namespace(spa, vd); 5706 5707 } else if (vd != NULL) { 5708 /* 5709 * Normal vdevs cannot be removed (yet). 5710 */ 5711 error = SET_ERROR(ENOTSUP); 5712 } else { 5713 /* 5714 * There is no vdev of any kind with the specified guid. 5715 */ 5716 error = SET_ERROR(ENOENT); 5717 } 5718 5719 if (!locked) 5720 return (spa_vdev_exit(spa, NULL, txg, error)); 5721 5722 return (error); 5723} 5724 5725/* 5726 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5727 * currently spared, so we can detach it. 5728 */ 5729static vdev_t * 5730spa_vdev_resilver_done_hunt(vdev_t *vd) 5731{ 5732 vdev_t *newvd, *oldvd; 5733 5734 for (int c = 0; c < vd->vdev_children; c++) { 5735 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5736 if (oldvd != NULL) 5737 return (oldvd); 5738 } 5739 5740 /* 5741 * Check for a completed replacement. We always consider the first 5742 * vdev in the list to be the oldest vdev, and the last one to be 5743 * the newest (see spa_vdev_attach() for how that works). In 5744 * the case where the newest vdev is faulted, we will not automatically 5745 * remove it after a resilver completes. This is OK as it will require 5746 * user intervention to determine which disk the admin wishes to keep. 5747 */ 5748 if (vd->vdev_ops == &vdev_replacing_ops) { 5749 ASSERT(vd->vdev_children > 1); 5750 5751 newvd = vd->vdev_child[vd->vdev_children - 1]; 5752 oldvd = vd->vdev_child[0]; 5753 5754 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5755 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5756 !vdev_dtl_required(oldvd)) 5757 return (oldvd); 5758 } 5759 5760 /* 5761 * Check for a completed resilver with the 'unspare' flag set. 5762 */ 5763 if (vd->vdev_ops == &vdev_spare_ops) { 5764 vdev_t *first = vd->vdev_child[0]; 5765 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5766 5767 if (last->vdev_unspare) { 5768 oldvd = first; 5769 newvd = last; 5770 } else if (first->vdev_unspare) { 5771 oldvd = last; 5772 newvd = first; 5773 } else { 5774 oldvd = NULL; 5775 } 5776 5777 if (oldvd != NULL && 5778 vdev_dtl_empty(newvd, DTL_MISSING) && 5779 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5780 !vdev_dtl_required(oldvd)) 5781 return (oldvd); 5782 5783 /* 5784 * If there are more than two spares attached to a disk, 5785 * and those spares are not required, then we want to 5786 * attempt to free them up now so that they can be used 5787 * by other pools. Once we're back down to a single 5788 * disk+spare, we stop removing them. 5789 */ 5790 if (vd->vdev_children > 2) { 5791 newvd = vd->vdev_child[1]; 5792 5793 if (newvd->vdev_isspare && last->vdev_isspare && 5794 vdev_dtl_empty(last, DTL_MISSING) && 5795 vdev_dtl_empty(last, DTL_OUTAGE) && 5796 !vdev_dtl_required(newvd)) 5797 return (newvd); 5798 } 5799 } 5800 5801 return (NULL); 5802} 5803 5804static void 5805spa_vdev_resilver_done(spa_t *spa) 5806{ 5807 vdev_t *vd, *pvd, *ppvd; 5808 uint64_t guid, sguid, pguid, ppguid; 5809 5810 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5811 5812 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5813 pvd = vd->vdev_parent; 5814 ppvd = pvd->vdev_parent; 5815 guid = vd->vdev_guid; 5816 pguid = pvd->vdev_guid; 5817 ppguid = ppvd->vdev_guid; 5818 sguid = 0; 5819 /* 5820 * If we have just finished replacing a hot spared device, then 5821 * we need to detach the parent's first child (the original hot 5822 * spare) as well. 5823 */ 5824 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5825 ppvd->vdev_children == 2) { 5826 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5827 sguid = ppvd->vdev_child[1]->vdev_guid; 5828 } 5829 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5830 5831 spa_config_exit(spa, SCL_ALL, FTAG); 5832 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5833 return; 5834 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5835 return; 5836 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5837 } 5838 5839 spa_config_exit(spa, SCL_ALL, FTAG); 5840} 5841 5842/* 5843 * Update the stored path or FRU for this vdev. 5844 */ 5845int 5846spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5847 boolean_t ispath) 5848{ 5849 vdev_t *vd; 5850 boolean_t sync = B_FALSE; 5851 5852 ASSERT(spa_writeable(spa)); 5853 5854 spa_vdev_state_enter(spa, SCL_ALL); 5855 5856 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5857 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5858 5859 if (!vd->vdev_ops->vdev_op_leaf) 5860 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5861 5862 if (ispath) { 5863 if (strcmp(value, vd->vdev_path) != 0) { 5864 spa_strfree(vd->vdev_path); 5865 vd->vdev_path = spa_strdup(value); 5866 sync = B_TRUE; 5867 } 5868 } else { 5869 if (vd->vdev_fru == NULL) { 5870 vd->vdev_fru = spa_strdup(value); 5871 sync = B_TRUE; 5872 } else if (strcmp(value, vd->vdev_fru) != 0) { 5873 spa_strfree(vd->vdev_fru); 5874 vd->vdev_fru = spa_strdup(value); 5875 sync = B_TRUE; 5876 } 5877 } 5878 5879 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5880} 5881 5882int 5883spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5884{ 5885 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5886} 5887 5888int 5889spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5890{ 5891 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5892} 5893 5894/* 5895 * ========================================================================== 5896 * SPA Scanning 5897 * ========================================================================== 5898 */ 5899 5900int 5901spa_scan_stop(spa_t *spa) 5902{ 5903 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5904 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5905 return (SET_ERROR(EBUSY)); 5906 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5907} 5908 5909int 5910spa_scan(spa_t *spa, pool_scan_func_t func) 5911{ 5912 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5913 5914 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5915 return (SET_ERROR(ENOTSUP)); 5916 5917 /* 5918 * If a resilver was requested, but there is no DTL on a 5919 * writeable leaf device, we have nothing to do. 5920 */ 5921 if (func == POOL_SCAN_RESILVER && 5922 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5923 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5924 return (0); 5925 } 5926 5927 return (dsl_scan(spa->spa_dsl_pool, func)); 5928} 5929 5930/* 5931 * ========================================================================== 5932 * SPA async task processing 5933 * ========================================================================== 5934 */ 5935 5936static void 5937spa_async_remove(spa_t *spa, vdev_t *vd) 5938{ 5939 if (vd->vdev_remove_wanted) { 5940 vd->vdev_remove_wanted = B_FALSE; 5941 vd->vdev_delayed_close = B_FALSE; 5942 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5943 5944 /* 5945 * We want to clear the stats, but we don't want to do a full 5946 * vdev_clear() as that will cause us to throw away 5947 * degraded/faulted state as well as attempt to reopen the 5948 * device, all of which is a waste. 5949 */ 5950 vd->vdev_stat.vs_read_errors = 0; 5951 vd->vdev_stat.vs_write_errors = 0; 5952 vd->vdev_stat.vs_checksum_errors = 0; 5953 5954 vdev_state_dirty(vd->vdev_top); 5955 /* Tell userspace that the vdev is gone. */ 5956 zfs_post_remove(spa, vd); 5957 } 5958 5959 for (int c = 0; c < vd->vdev_children; c++) 5960 spa_async_remove(spa, vd->vdev_child[c]); 5961} 5962 5963static void 5964spa_async_probe(spa_t *spa, vdev_t *vd) 5965{ 5966 if (vd->vdev_probe_wanted) { 5967 vd->vdev_probe_wanted = B_FALSE; 5968 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5969 } 5970 5971 for (int c = 0; c < vd->vdev_children; c++) 5972 spa_async_probe(spa, vd->vdev_child[c]); 5973} 5974 5975static void 5976spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5977{ 5978 sysevent_id_t eid; 5979 nvlist_t *attr; 5980 char *physpath; 5981 5982 if (!spa->spa_autoexpand) 5983 return; 5984 5985 for (int c = 0; c < vd->vdev_children; c++) { 5986 vdev_t *cvd = vd->vdev_child[c]; 5987 spa_async_autoexpand(spa, cvd); 5988 } 5989 5990 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5991 return; 5992 5993 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5994 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5995 5996 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5997 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5998 5999 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6000 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 6001 6002 nvlist_free(attr); 6003 kmem_free(physpath, MAXPATHLEN); 6004} 6005 6006static void 6007spa_async_thread(void *arg) 6008{ 6009 spa_t *spa = arg; 6010 int tasks; 6011 6012 ASSERT(spa->spa_sync_on); 6013 6014 mutex_enter(&spa->spa_async_lock); 6015 tasks = spa->spa_async_tasks; 6016 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 6017 mutex_exit(&spa->spa_async_lock); 6018 6019 /* 6020 * See if the config needs to be updated. 6021 */ 6022 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6023 uint64_t old_space, new_space; 6024 6025 mutex_enter(&spa_namespace_lock); 6026 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6027 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6028 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6029 mutex_exit(&spa_namespace_lock); 6030 6031 /* 6032 * If the pool grew as a result of the config update, 6033 * then log an internal history event. 6034 */ 6035 if (new_space != old_space) { 6036 spa_history_log_internal(spa, "vdev online", NULL, 6037 "pool '%s' size: %llu(+%llu)", 6038 spa_name(spa), new_space, new_space - old_space); 6039 } 6040 } 6041 6042 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6043 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6044 spa_async_autoexpand(spa, spa->spa_root_vdev); 6045 spa_config_exit(spa, SCL_CONFIG, FTAG); 6046 } 6047 6048 /* 6049 * See if any devices need to be probed. 6050 */ 6051 if (tasks & SPA_ASYNC_PROBE) { 6052 spa_vdev_state_enter(spa, SCL_NONE); 6053 spa_async_probe(spa, spa->spa_root_vdev); 6054 (void) spa_vdev_state_exit(spa, NULL, 0); 6055 } 6056 6057 /* 6058 * If any devices are done replacing, detach them. 6059 */ 6060 if (tasks & SPA_ASYNC_RESILVER_DONE) 6061 spa_vdev_resilver_done(spa); 6062 6063 /* 6064 * Kick off a resilver. 6065 */ 6066 if (tasks & SPA_ASYNC_RESILVER) 6067 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6068 6069 /* 6070 * Let the world know that we're done. 6071 */ 6072 mutex_enter(&spa->spa_async_lock); 6073 spa->spa_async_thread = NULL; 6074 cv_broadcast(&spa->spa_async_cv); 6075 mutex_exit(&spa->spa_async_lock); 6076 thread_exit(); 6077} 6078 6079static void 6080spa_async_thread_vd(void *arg) 6081{ 6082 spa_t *spa = arg; 6083 int tasks; 6084 6085 ASSERT(spa->spa_sync_on); 6086 6087 mutex_enter(&spa->spa_async_lock); 6088 tasks = spa->spa_async_tasks; 6089retry: 6090 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6091 mutex_exit(&spa->spa_async_lock); 6092 6093 /* 6094 * See if any devices need to be marked REMOVED. 6095 */ 6096 if (tasks & SPA_ASYNC_REMOVE) { 6097 spa_vdev_state_enter(spa, SCL_NONE); 6098 spa_async_remove(spa, spa->spa_root_vdev); 6099 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6100 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6101 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6102 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6103 (void) spa_vdev_state_exit(spa, NULL, 0); 6104 } 6105 6106 /* 6107 * Let the world know that we're done. 6108 */ 6109 mutex_enter(&spa->spa_async_lock); 6110 tasks = spa->spa_async_tasks; 6111 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6112 goto retry; 6113 spa->spa_async_thread_vd = NULL; 6114 cv_broadcast(&spa->spa_async_cv); 6115 mutex_exit(&spa->spa_async_lock); 6116 thread_exit(); 6117} 6118 6119void 6120spa_async_suspend(spa_t *spa) 6121{ 6122 mutex_enter(&spa->spa_async_lock); 6123 spa->spa_async_suspended++; 6124 while (spa->spa_async_thread != NULL && 6125 spa->spa_async_thread_vd != NULL) 6126 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6127 mutex_exit(&spa->spa_async_lock); 6128} 6129 6130void 6131spa_async_resume(spa_t *spa) 6132{ 6133 mutex_enter(&spa->spa_async_lock); 6134 ASSERT(spa->spa_async_suspended != 0); 6135 spa->spa_async_suspended--; 6136 mutex_exit(&spa->spa_async_lock); 6137} 6138 6139static boolean_t 6140spa_async_tasks_pending(spa_t *spa) 6141{ 6142 uint_t non_config_tasks; 6143 uint_t config_task; 6144 boolean_t config_task_suspended; 6145 6146 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6147 SPA_ASYNC_REMOVE); 6148 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6149 if (spa->spa_ccw_fail_time == 0) { 6150 config_task_suspended = B_FALSE; 6151 } else { 6152 config_task_suspended = 6153 (gethrtime() - spa->spa_ccw_fail_time) < 6154 (zfs_ccw_retry_interval * NANOSEC); 6155 } 6156 6157 return (non_config_tasks || (config_task && !config_task_suspended)); 6158} 6159 6160static void 6161spa_async_dispatch(spa_t *spa) 6162{ 6163 mutex_enter(&spa->spa_async_lock); 6164 if (spa_async_tasks_pending(spa) && 6165 !spa->spa_async_suspended && 6166 spa->spa_async_thread == NULL && 6167 rootdir != NULL) 6168 spa->spa_async_thread = thread_create(NULL, 0, 6169 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6170 mutex_exit(&spa->spa_async_lock); 6171} 6172 6173static void 6174spa_async_dispatch_vd(spa_t *spa) 6175{ 6176 mutex_enter(&spa->spa_async_lock); 6177 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6178 !spa->spa_async_suspended && 6179 spa->spa_async_thread_vd == NULL && 6180 rootdir != NULL) 6181 spa->spa_async_thread_vd = thread_create(NULL, 0, 6182 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6183 mutex_exit(&spa->spa_async_lock); 6184} 6185 6186void 6187spa_async_request(spa_t *spa, int task) 6188{ 6189 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6190 mutex_enter(&spa->spa_async_lock); 6191 spa->spa_async_tasks |= task; 6192 mutex_exit(&spa->spa_async_lock); 6193 spa_async_dispatch_vd(spa); 6194} 6195 6196/* 6197 * ========================================================================== 6198 * SPA syncing routines 6199 * ========================================================================== 6200 */ 6201 6202static int 6203bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6204{ 6205 bpobj_t *bpo = arg; 6206 bpobj_enqueue(bpo, bp, tx); 6207 return (0); 6208} 6209 6210static int 6211spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6212{ 6213 zio_t *zio = arg; 6214 6215 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6216 BP_GET_PSIZE(bp), zio->io_flags)); 6217 return (0); 6218} 6219 6220/* 6221 * Note: this simple function is not inlined to make it easier to dtrace the 6222 * amount of time spent syncing frees. 6223 */ 6224static void 6225spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6226{ 6227 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6228 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6229 VERIFY(zio_wait(zio) == 0); 6230} 6231 6232/* 6233 * Note: this simple function is not inlined to make it easier to dtrace the 6234 * amount of time spent syncing deferred frees. 6235 */ 6236static void 6237spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6238{ 6239 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6240 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6241 spa_free_sync_cb, zio, tx), ==, 0); 6242 VERIFY0(zio_wait(zio)); 6243} 6244 6245 6246static void 6247spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6248{ 6249 char *packed = NULL; 6250 size_t bufsize; 6251 size_t nvsize = 0; 6252 dmu_buf_t *db; 6253 6254 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6255 6256 /* 6257 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6258 * information. This avoids the dmu_buf_will_dirty() path and 6259 * saves us a pre-read to get data we don't actually care about. 6260 */ 6261 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6262 packed = kmem_alloc(bufsize, KM_SLEEP); 6263 6264 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6265 KM_SLEEP) == 0); 6266 bzero(packed + nvsize, bufsize - nvsize); 6267 6268 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6269 6270 kmem_free(packed, bufsize); 6271 6272 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6273 dmu_buf_will_dirty(db, tx); 6274 *(uint64_t *)db->db_data = nvsize; 6275 dmu_buf_rele(db, FTAG); 6276} 6277 6278static void 6279spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6280 const char *config, const char *entry) 6281{ 6282 nvlist_t *nvroot; 6283 nvlist_t **list; 6284 int i; 6285 6286 if (!sav->sav_sync) 6287 return; 6288 6289 /* 6290 * Update the MOS nvlist describing the list of available devices. 6291 * spa_validate_aux() will have already made sure this nvlist is 6292 * valid and the vdevs are labeled appropriately. 6293 */ 6294 if (sav->sav_object == 0) { 6295 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6296 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6297 sizeof (uint64_t), tx); 6298 VERIFY(zap_update(spa->spa_meta_objset, 6299 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6300 &sav->sav_object, tx) == 0); 6301 } 6302 6303 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6304 if (sav->sav_count == 0) { 6305 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6306 } else { 6307 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6308 for (i = 0; i < sav->sav_count; i++) 6309 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6310 B_FALSE, VDEV_CONFIG_L2CACHE); 6311 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6312 sav->sav_count) == 0); 6313 for (i = 0; i < sav->sav_count; i++) 6314 nvlist_free(list[i]); 6315 kmem_free(list, sav->sav_count * sizeof (void *)); 6316 } 6317 6318 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6319 nvlist_free(nvroot); 6320 6321 sav->sav_sync = B_FALSE; 6322} 6323 6324static void 6325spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6326{ 6327 nvlist_t *config; 6328 6329 if (list_is_empty(&spa->spa_config_dirty_list)) 6330 return; 6331 6332 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6333 6334 config = spa_config_generate(spa, spa->spa_root_vdev, 6335 dmu_tx_get_txg(tx), B_FALSE); 6336 6337 /* 6338 * If we're upgrading the spa version then make sure that 6339 * the config object gets updated with the correct version. 6340 */ 6341 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6342 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6343 spa->spa_uberblock.ub_version); 6344 6345 spa_config_exit(spa, SCL_STATE, FTAG); 6346 6347 nvlist_free(spa->spa_config_syncing); 6348 spa->spa_config_syncing = config; 6349 6350 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6351} 6352 6353static void 6354spa_sync_version(void *arg, dmu_tx_t *tx) 6355{ 6356 uint64_t *versionp = arg; 6357 uint64_t version = *versionp; 6358 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6359 6360 /* 6361 * Setting the version is special cased when first creating the pool. 6362 */ 6363 ASSERT(tx->tx_txg != TXG_INITIAL); 6364 6365 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6366 ASSERT(version >= spa_version(spa)); 6367 6368 spa->spa_uberblock.ub_version = version; 6369 vdev_config_dirty(spa->spa_root_vdev); 6370 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6371} 6372 6373/* 6374 * Set zpool properties. 6375 */ 6376static void 6377spa_sync_props(void *arg, dmu_tx_t *tx) 6378{ 6379 nvlist_t *nvp = arg; 6380 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6381 objset_t *mos = spa->spa_meta_objset; 6382 nvpair_t *elem = NULL; 6383 6384 mutex_enter(&spa->spa_props_lock); 6385 6386 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6387 uint64_t intval; 6388 char *strval, *fname; 6389 zpool_prop_t prop; 6390 const char *propname; 6391 zprop_type_t proptype; 6392 spa_feature_t fid; 6393 6394 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6395 case ZPROP_INVAL: 6396 /* 6397 * We checked this earlier in spa_prop_validate(). 6398 */ 6399 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6400 6401 fname = strchr(nvpair_name(elem), '@') + 1; 6402 VERIFY0(zfeature_lookup_name(fname, &fid)); 6403 6404 spa_feature_enable(spa, fid, tx); 6405 spa_history_log_internal(spa, "set", tx, 6406 "%s=enabled", nvpair_name(elem)); 6407 break; 6408 6409 case ZPOOL_PROP_VERSION: 6410 intval = fnvpair_value_uint64(elem); 6411 /* 6412 * The version is synced seperatly before other 6413 * properties and should be correct by now. 6414 */ 6415 ASSERT3U(spa_version(spa), >=, intval); 6416 break; 6417 6418 case ZPOOL_PROP_ALTROOT: 6419 /* 6420 * 'altroot' is a non-persistent property. It should 6421 * have been set temporarily at creation or import time. 6422 */ 6423 ASSERT(spa->spa_root != NULL); 6424 break; 6425 6426 case ZPOOL_PROP_READONLY: 6427 case ZPOOL_PROP_CACHEFILE: 6428 /* 6429 * 'readonly' and 'cachefile' are also non-persisitent 6430 * properties. 6431 */ 6432 break; 6433 case ZPOOL_PROP_COMMENT: 6434 strval = fnvpair_value_string(elem); 6435 if (spa->spa_comment != NULL) 6436 spa_strfree(spa->spa_comment); 6437 spa->spa_comment = spa_strdup(strval); 6438 /* 6439 * We need to dirty the configuration on all the vdevs 6440 * so that their labels get updated. It's unnecessary 6441 * to do this for pool creation since the vdev's 6442 * configuratoin has already been dirtied. 6443 */ 6444 if (tx->tx_txg != TXG_INITIAL) 6445 vdev_config_dirty(spa->spa_root_vdev); 6446 spa_history_log_internal(spa, "set", tx, 6447 "%s=%s", nvpair_name(elem), strval); 6448 break; 6449 default: 6450 /* 6451 * Set pool property values in the poolprops mos object. 6452 */ 6453 if (spa->spa_pool_props_object == 0) { 6454 spa->spa_pool_props_object = 6455 zap_create_link(mos, DMU_OT_POOL_PROPS, 6456 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6457 tx); 6458 } 6459 6460 /* normalize the property name */ 6461 propname = zpool_prop_to_name(prop); 6462 proptype = zpool_prop_get_type(prop); 6463 6464 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6465 ASSERT(proptype == PROP_TYPE_STRING); 6466 strval = fnvpair_value_string(elem); 6467 VERIFY0(zap_update(mos, 6468 spa->spa_pool_props_object, propname, 6469 1, strlen(strval) + 1, strval, tx)); 6470 spa_history_log_internal(spa, "set", tx, 6471 "%s=%s", nvpair_name(elem), strval); 6472 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6473 intval = fnvpair_value_uint64(elem); 6474 6475 if (proptype == PROP_TYPE_INDEX) { 6476 const char *unused; 6477 VERIFY0(zpool_prop_index_to_string( 6478 prop, intval, &unused)); 6479 } 6480 VERIFY0(zap_update(mos, 6481 spa->spa_pool_props_object, propname, 6482 8, 1, &intval, tx)); 6483 spa_history_log_internal(spa, "set", tx, 6484 "%s=%lld", nvpair_name(elem), intval); 6485 } else { 6486 ASSERT(0); /* not allowed */ 6487 } 6488 6489 switch (prop) { 6490 case ZPOOL_PROP_DELEGATION: 6491 spa->spa_delegation = intval; 6492 break; 6493 case ZPOOL_PROP_BOOTFS: 6494 spa->spa_bootfs = intval; 6495 break; 6496 case ZPOOL_PROP_FAILUREMODE: 6497 spa->spa_failmode = intval; 6498 break; 6499 case ZPOOL_PROP_AUTOEXPAND: 6500 spa->spa_autoexpand = intval; 6501 if (tx->tx_txg != TXG_INITIAL) 6502 spa_async_request(spa, 6503 SPA_ASYNC_AUTOEXPAND); 6504 break; 6505 case ZPOOL_PROP_DEDUPDITTO: 6506 spa->spa_dedup_ditto = intval; 6507 break; 6508 default: 6509 break; 6510 } 6511 } 6512 6513 } 6514 6515 mutex_exit(&spa->spa_props_lock); 6516} 6517 6518/* 6519 * Perform one-time upgrade on-disk changes. spa_version() does not 6520 * reflect the new version this txg, so there must be no changes this 6521 * txg to anything that the upgrade code depends on after it executes. 6522 * Therefore this must be called after dsl_pool_sync() does the sync 6523 * tasks. 6524 */ 6525static void 6526spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6527{ 6528 dsl_pool_t *dp = spa->spa_dsl_pool; 6529 6530 ASSERT(spa->spa_sync_pass == 1); 6531 6532 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6533 6534 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6535 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6536 dsl_pool_create_origin(dp, tx); 6537 6538 /* Keeping the origin open increases spa_minref */ 6539 spa->spa_minref += 3; 6540 } 6541 6542 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6543 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6544 dsl_pool_upgrade_clones(dp, tx); 6545 } 6546 6547 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6548 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6549 dsl_pool_upgrade_dir_clones(dp, tx); 6550 6551 /* Keeping the freedir open increases spa_minref */ 6552 spa->spa_minref += 3; 6553 } 6554 6555 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6556 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6557 spa_feature_create_zap_objects(spa, tx); 6558 } 6559 6560 /* 6561 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6562 * when possibility to use lz4 compression for metadata was added 6563 * Old pools that have this feature enabled must be upgraded to have 6564 * this feature active 6565 */ 6566 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6567 boolean_t lz4_en = spa_feature_is_enabled(spa, 6568 SPA_FEATURE_LZ4_COMPRESS); 6569 boolean_t lz4_ac = spa_feature_is_active(spa, 6570 SPA_FEATURE_LZ4_COMPRESS); 6571 6572 if (lz4_en && !lz4_ac) 6573 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6574 } 6575 6576 /* 6577 * If we haven't written the salt, do so now. Note that the 6578 * feature may not be activated yet, but that's fine since 6579 * the presence of this ZAP entry is backwards compatible. 6580 */ 6581 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 6582 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 6583 VERIFY0(zap_add(spa->spa_meta_objset, 6584 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 6585 sizeof (spa->spa_cksum_salt.zcs_bytes), 6586 spa->spa_cksum_salt.zcs_bytes, tx)); 6587 } 6588 6589 rrw_exit(&dp->dp_config_rwlock, FTAG); 6590} 6591 6592/* 6593 * Sync the specified transaction group. New blocks may be dirtied as 6594 * part of the process, so we iterate until it converges. 6595 */ 6596void 6597spa_sync(spa_t *spa, uint64_t txg) 6598{ 6599 dsl_pool_t *dp = spa->spa_dsl_pool; 6600 objset_t *mos = spa->spa_meta_objset; 6601 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6602 vdev_t *rvd = spa->spa_root_vdev; 6603 vdev_t *vd; 6604 dmu_tx_t *tx; 6605 int error; 6606 6607 VERIFY(spa_writeable(spa)); 6608 6609 /* 6610 * Lock out configuration changes. 6611 */ 6612 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6613 6614 spa->spa_syncing_txg = txg; 6615 spa->spa_sync_pass = 0; 6616 6617 /* 6618 * If there are any pending vdev state changes, convert them 6619 * into config changes that go out with this transaction group. 6620 */ 6621 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6622 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6623 /* 6624 * We need the write lock here because, for aux vdevs, 6625 * calling vdev_config_dirty() modifies sav_config. 6626 * This is ugly and will become unnecessary when we 6627 * eliminate the aux vdev wart by integrating all vdevs 6628 * into the root vdev tree. 6629 */ 6630 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6631 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6632 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6633 vdev_state_clean(vd); 6634 vdev_config_dirty(vd); 6635 } 6636 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6637 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6638 } 6639 spa_config_exit(spa, SCL_STATE, FTAG); 6640 6641 tx = dmu_tx_create_assigned(dp, txg); 6642 6643 spa->spa_sync_starttime = gethrtime(); 6644#ifdef illumos 6645 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6646 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6647#else /* !illumos */ 6648#ifdef _KERNEL 6649 callout_reset(&spa->spa_deadman_cycid, 6650 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6651#endif 6652#endif /* illumos */ 6653 6654 /* 6655 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6656 * set spa_deflate if we have no raid-z vdevs. 6657 */ 6658 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6659 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6660 int i; 6661 6662 for (i = 0; i < rvd->vdev_children; i++) { 6663 vd = rvd->vdev_child[i]; 6664 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6665 break; 6666 } 6667 if (i == rvd->vdev_children) { 6668 spa->spa_deflate = TRUE; 6669 VERIFY(0 == zap_add(spa->spa_meta_objset, 6670 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6671 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6672 } 6673 } 6674 6675 /* 6676 * Iterate to convergence. 6677 */ 6678 do { 6679 int pass = ++spa->spa_sync_pass; 6680 6681 spa_sync_config_object(spa, tx); 6682 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6683 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6684 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6685 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6686 spa_errlog_sync(spa, txg); 6687 dsl_pool_sync(dp, txg); 6688 6689 if (pass < zfs_sync_pass_deferred_free) { 6690 spa_sync_frees(spa, free_bpl, tx); 6691 } else { 6692 /* 6693 * We can not defer frees in pass 1, because 6694 * we sync the deferred frees later in pass 1. 6695 */ 6696 ASSERT3U(pass, >, 1); 6697 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6698 &spa->spa_deferred_bpobj, tx); 6699 } 6700 6701 ddt_sync(spa, txg); 6702 dsl_scan_sync(dp, tx); 6703 6704 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6705 vdev_sync(vd, txg); 6706 6707 if (pass == 1) { 6708 spa_sync_upgrades(spa, tx); 6709 ASSERT3U(txg, >=, 6710 spa->spa_uberblock.ub_rootbp.blk_birth); 6711 /* 6712 * Note: We need to check if the MOS is dirty 6713 * because we could have marked the MOS dirty 6714 * without updating the uberblock (e.g. if we 6715 * have sync tasks but no dirty user data). We 6716 * need to check the uberblock's rootbp because 6717 * it is updated if we have synced out dirty 6718 * data (though in this case the MOS will most 6719 * likely also be dirty due to second order 6720 * effects, we don't want to rely on that here). 6721 */ 6722 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6723 !dmu_objset_is_dirty(mos, txg)) { 6724 /* 6725 * Nothing changed on the first pass, 6726 * therefore this TXG is a no-op. Avoid 6727 * syncing deferred frees, so that we 6728 * can keep this TXG as a no-op. 6729 */ 6730 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6731 txg)); 6732 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6733 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6734 break; 6735 } 6736 spa_sync_deferred_frees(spa, tx); 6737 } 6738 6739 } while (dmu_objset_is_dirty(mos, txg)); 6740 6741 /* 6742 * Rewrite the vdev configuration (which includes the uberblock) 6743 * to commit the transaction group. 6744 * 6745 * If there are no dirty vdevs, we sync the uberblock to a few 6746 * random top-level vdevs that are known to be visible in the 6747 * config cache (see spa_vdev_add() for a complete description). 6748 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6749 */ 6750 for (;;) { 6751 /* 6752 * We hold SCL_STATE to prevent vdev open/close/etc. 6753 * while we're attempting to write the vdev labels. 6754 */ 6755 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6756 6757 if (list_is_empty(&spa->spa_config_dirty_list)) { 6758 vdev_t *svd[SPA_DVAS_PER_BP]; 6759 int svdcount = 0; 6760 int children = rvd->vdev_children; 6761 int c0 = spa_get_random(children); 6762 6763 for (int c = 0; c < children; c++) { 6764 vd = rvd->vdev_child[(c0 + c) % children]; 6765 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6766 continue; 6767 svd[svdcount++] = vd; 6768 if (svdcount == SPA_DVAS_PER_BP) 6769 break; 6770 } 6771 error = vdev_config_sync(svd, svdcount, txg); 6772 } else { 6773 error = vdev_config_sync(rvd->vdev_child, 6774 rvd->vdev_children, txg); 6775 } 6776 6777 if (error == 0) 6778 spa->spa_last_synced_guid = rvd->vdev_guid; 6779 6780 spa_config_exit(spa, SCL_STATE, FTAG); 6781 6782 if (error == 0) 6783 break; 6784 zio_suspend(spa, NULL); 6785 zio_resume_wait(spa); 6786 } 6787 dmu_tx_commit(tx); 6788 6789#ifdef illumos 6790 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6791#else /* !illumos */ 6792#ifdef _KERNEL 6793 callout_drain(&spa->spa_deadman_cycid); 6794#endif 6795#endif /* illumos */ 6796 6797 /* 6798 * Clear the dirty config list. 6799 */ 6800 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6801 vdev_config_clean(vd); 6802 6803 /* 6804 * Now that the new config has synced transactionally, 6805 * let it become visible to the config cache. 6806 */ 6807 if (spa->spa_config_syncing != NULL) { 6808 spa_config_set(spa, spa->spa_config_syncing); 6809 spa->spa_config_txg = txg; 6810 spa->spa_config_syncing = NULL; 6811 } 6812 6813 spa->spa_ubsync = spa->spa_uberblock; 6814 6815 dsl_pool_sync_done(dp, txg); 6816 6817 /* 6818 * Update usable space statistics. 6819 */ 6820 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6821 vdev_sync_done(vd, txg); 6822 6823 spa_update_dspace(spa); 6824 6825 /* 6826 * It had better be the case that we didn't dirty anything 6827 * since vdev_config_sync(). 6828 */ 6829 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6830 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6831 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6832 6833 spa->spa_sync_pass = 0; 6834 6835 spa_config_exit(spa, SCL_CONFIG, FTAG); 6836 6837 spa_handle_ignored_writes(spa); 6838 6839 /* 6840 * If any async tasks have been requested, kick them off. 6841 */ 6842 spa_async_dispatch(spa); 6843 spa_async_dispatch_vd(spa); 6844} 6845 6846/* 6847 * Sync all pools. We don't want to hold the namespace lock across these 6848 * operations, so we take a reference on the spa_t and drop the lock during the 6849 * sync. 6850 */ 6851void 6852spa_sync_allpools(void) 6853{ 6854 spa_t *spa = NULL; 6855 mutex_enter(&spa_namespace_lock); 6856 while ((spa = spa_next(spa)) != NULL) { 6857 if (spa_state(spa) != POOL_STATE_ACTIVE || 6858 !spa_writeable(spa) || spa_suspended(spa)) 6859 continue; 6860 spa_open_ref(spa, FTAG); 6861 mutex_exit(&spa_namespace_lock); 6862 txg_wait_synced(spa_get_dsl(spa), 0); 6863 mutex_enter(&spa_namespace_lock); 6864 spa_close(spa, FTAG); 6865 } 6866 mutex_exit(&spa_namespace_lock); 6867} 6868 6869/* 6870 * ========================================================================== 6871 * Miscellaneous routines 6872 * ========================================================================== 6873 */ 6874 6875/* 6876 * Remove all pools in the system. 6877 */ 6878void 6879spa_evict_all(void) 6880{ 6881 spa_t *spa; 6882 6883 /* 6884 * Remove all cached state. All pools should be closed now, 6885 * so every spa in the AVL tree should be unreferenced. 6886 */ 6887 mutex_enter(&spa_namespace_lock); 6888 while ((spa = spa_next(NULL)) != NULL) { 6889 /* 6890 * Stop async tasks. The async thread may need to detach 6891 * a device that's been replaced, which requires grabbing 6892 * spa_namespace_lock, so we must drop it here. 6893 */ 6894 spa_open_ref(spa, FTAG); 6895 mutex_exit(&spa_namespace_lock); 6896 spa_async_suspend(spa); 6897 mutex_enter(&spa_namespace_lock); 6898 spa_close(spa, FTAG); 6899 6900 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6901 spa_unload(spa); 6902 spa_deactivate(spa); 6903 } 6904 spa_remove(spa); 6905 } 6906 mutex_exit(&spa_namespace_lock); 6907} 6908 6909vdev_t * 6910spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6911{ 6912 vdev_t *vd; 6913 int i; 6914 6915 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6916 return (vd); 6917 6918 if (aux) { 6919 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6920 vd = spa->spa_l2cache.sav_vdevs[i]; 6921 if (vd->vdev_guid == guid) 6922 return (vd); 6923 } 6924 6925 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6926 vd = spa->spa_spares.sav_vdevs[i]; 6927 if (vd->vdev_guid == guid) 6928 return (vd); 6929 } 6930 } 6931 6932 return (NULL); 6933} 6934 6935void 6936spa_upgrade(spa_t *spa, uint64_t version) 6937{ 6938 ASSERT(spa_writeable(spa)); 6939 6940 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6941 6942 /* 6943 * This should only be called for a non-faulted pool, and since a 6944 * future version would result in an unopenable pool, this shouldn't be 6945 * possible. 6946 */ 6947 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6948 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6949 6950 spa->spa_uberblock.ub_version = version; 6951 vdev_config_dirty(spa->spa_root_vdev); 6952 6953 spa_config_exit(spa, SCL_ALL, FTAG); 6954 6955 txg_wait_synced(spa_get_dsl(spa), 0); 6956} 6957 6958boolean_t 6959spa_has_spare(spa_t *spa, uint64_t guid) 6960{ 6961 int i; 6962 uint64_t spareguid; 6963 spa_aux_vdev_t *sav = &spa->spa_spares; 6964 6965 for (i = 0; i < sav->sav_count; i++) 6966 if (sav->sav_vdevs[i]->vdev_guid == guid) 6967 return (B_TRUE); 6968 6969 for (i = 0; i < sav->sav_npending; i++) { 6970 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6971 &spareguid) == 0 && spareguid == guid) 6972 return (B_TRUE); 6973 } 6974 6975 return (B_FALSE); 6976} 6977 6978/* 6979 * Check if a pool has an active shared spare device. 6980 * Note: reference count of an active spare is 2, as a spare and as a replace 6981 */ 6982static boolean_t 6983spa_has_active_shared_spare(spa_t *spa) 6984{ 6985 int i, refcnt; 6986 uint64_t pool; 6987 spa_aux_vdev_t *sav = &spa->spa_spares; 6988 6989 for (i = 0; i < sav->sav_count; i++) { 6990 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6991 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6992 refcnt > 2) 6993 return (B_TRUE); 6994 } 6995 6996 return (B_FALSE); 6997} 6998 6999/* 7000 * Post a sysevent corresponding to the given event. The 'name' must be one of 7001 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7002 * filled in from the spa and (optionally) the vdev. This doesn't do anything 7003 * in the userland libzpool, as we don't want consumers to misinterpret ztest 7004 * or zdb as real changes. 7005 */ 7006void 7007spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 7008{ 7009#ifdef _KERNEL 7010 sysevent_t *ev; 7011 sysevent_attr_list_t *attr = NULL; 7012 sysevent_value_t value; 7013 sysevent_id_t eid; 7014 7015 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7016 SE_SLEEP); 7017 7018 value.value_type = SE_DATA_TYPE_STRING; 7019 value.value.sv_string = spa_name(spa); 7020 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7021 goto done; 7022 7023 value.value_type = SE_DATA_TYPE_UINT64; 7024 value.value.sv_uint64 = spa_guid(spa); 7025 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7026 goto done; 7027 7028 if (vd) { 7029 value.value_type = SE_DATA_TYPE_UINT64; 7030 value.value.sv_uint64 = vd->vdev_guid; 7031 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7032 SE_SLEEP) != 0) 7033 goto done; 7034 7035 if (vd->vdev_path) { 7036 value.value_type = SE_DATA_TYPE_STRING; 7037 value.value.sv_string = vd->vdev_path; 7038 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7039 &value, SE_SLEEP) != 0) 7040 goto done; 7041 } 7042 } 7043 7044 if (sysevent_attach_attributes(ev, attr) != 0) 7045 goto done; 7046 attr = NULL; 7047 7048 (void) log_sysevent(ev, SE_SLEEP, &eid); 7049 7050done: 7051 if (attr) 7052 sysevent_free_attr(attr); 7053 sysevent_free(ev); 7054#endif 7055} 7056