spa.c revision 287745
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 */ 29 30/* 31 * SPA: Storage Pool Allocator 32 * 33 * This file contains all the routines used when modifying on-disk SPA state. 34 * This includes opening, importing, destroying, exporting a pool, and syncing a 35 * pool. 36 */ 37 38#include <sys/zfs_context.h> 39#include <sys/fm/fs/zfs.h> 40#include <sys/spa_impl.h> 41#include <sys/zio.h> 42#include <sys/zio_checksum.h> 43#include <sys/dmu.h> 44#include <sys/dmu_tx.h> 45#include <sys/zap.h> 46#include <sys/zil.h> 47#include <sys/ddt.h> 48#include <sys/vdev_impl.h> 49#include <sys/metaslab.h> 50#include <sys/metaslab_impl.h> 51#include <sys/uberblock_impl.h> 52#include <sys/txg.h> 53#include <sys/avl.h> 54#include <sys/dmu_traverse.h> 55#include <sys/dmu_objset.h> 56#include <sys/unique.h> 57#include <sys/dsl_pool.h> 58#include <sys/dsl_dataset.h> 59#include <sys/dsl_dir.h> 60#include <sys/dsl_prop.h> 61#include <sys/dsl_synctask.h> 62#include <sys/fs/zfs.h> 63#include <sys/arc.h> 64#include <sys/callb.h> 65#include <sys/spa_boot.h> 66#include <sys/zfs_ioctl.h> 67#include <sys/dsl_scan.h> 68#include <sys/dmu_send.h> 69#include <sys/dsl_destroy.h> 70#include <sys/dsl_userhold.h> 71#include <sys/zfeature.h> 72#include <sys/zvol.h> 73#include <sys/trim_map.h> 74 75#ifdef _KERNEL 76#include <sys/callb.h> 77#include <sys/cpupart.h> 78#include <sys/zone.h> 79#endif /* _KERNEL */ 80 81#include "zfs_prop.h" 82#include "zfs_comutil.h" 83 84/* Check hostid on import? */ 85static int check_hostid = 1; 86 87/* 88 * The interval, in seconds, at which failed configuration cache file writes 89 * should be retried. 90 */ 91static int zfs_ccw_retry_interval = 300; 92 93SYSCTL_DECL(_vfs_zfs); 94SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 95 "Check hostid on import?"); 96TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 97SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 98 &zfs_ccw_retry_interval, 0, 99 "Configuration cache file write, retry after failure, interval (seconds)"); 100 101typedef enum zti_modes { 102 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 103 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 104 ZTI_MODE_NULL, /* don't create a taskq */ 105 ZTI_NMODES 106} zti_modes_t; 107 108#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 109#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 110#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 111 112#define ZTI_N(n) ZTI_P(n, 1) 113#define ZTI_ONE ZTI_N(1) 114 115typedef struct zio_taskq_info { 116 zti_modes_t zti_mode; 117 uint_t zti_value; 118 uint_t zti_count; 119} zio_taskq_info_t; 120 121static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 122 "issue", "issue_high", "intr", "intr_high" 123}; 124 125/* 126 * This table defines the taskq settings for each ZFS I/O type. When 127 * initializing a pool, we use this table to create an appropriately sized 128 * taskq. Some operations are low volume and therefore have a small, static 129 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 130 * macros. Other operations process a large amount of data; the ZTI_BATCH 131 * macro causes us to create a taskq oriented for throughput. Some operations 132 * are so high frequency and short-lived that the taskq itself can become a a 133 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 134 * additional degree of parallelism specified by the number of threads per- 135 * taskq and the number of taskqs; when dispatching an event in this case, the 136 * particular taskq is chosen at random. 137 * 138 * The different taskq priorities are to handle the different contexts (issue 139 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 140 * need to be handled with minimum delay. 141 */ 142const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 143 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 145 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 146 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 147 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 149 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 150}; 151 152static void spa_sync_version(void *arg, dmu_tx_t *tx); 153static void spa_sync_props(void *arg, dmu_tx_t *tx); 154static boolean_t spa_has_active_shared_spare(spa_t *spa); 155static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 156 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 157 char **ereport); 158static void spa_vdev_resilver_done(spa_t *spa); 159 160uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 161#ifdef PSRSET_BIND 162id_t zio_taskq_psrset_bind = PS_NONE; 163#endif 164#ifdef SYSDC 165boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 166#endif 167uint_t zio_taskq_basedc = 80; /* base duty cycle */ 168 169boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 170extern int zfs_sync_pass_deferred_free; 171 172#ifndef illumos 173extern void spa_deadman(void *arg); 174#endif 175 176/* 177 * This (illegal) pool name is used when temporarily importing a spa_t in order 178 * to get the vdev stats associated with the imported devices. 179 */ 180#define TRYIMPORT_NAME "$import" 181 182/* 183 * ========================================================================== 184 * SPA properties routines 185 * ========================================================================== 186 */ 187 188/* 189 * Add a (source=src, propname=propval) list to an nvlist. 190 */ 191static void 192spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 193 uint64_t intval, zprop_source_t src) 194{ 195 const char *propname = zpool_prop_to_name(prop); 196 nvlist_t *propval; 197 198 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 199 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 200 201 if (strval != NULL) 202 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 203 else 204 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 205 206 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 207 nvlist_free(propval); 208} 209 210/* 211 * Get property values from the spa configuration. 212 */ 213static void 214spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 215{ 216 vdev_t *rvd = spa->spa_root_vdev; 217 dsl_pool_t *pool = spa->spa_dsl_pool; 218 uint64_t size, alloc, cap, version; 219 zprop_source_t src = ZPROP_SRC_NONE; 220 spa_config_dirent_t *dp; 221 metaslab_class_t *mc = spa_normal_class(spa); 222 223 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 224 225 if (rvd != NULL) { 226 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 227 size = metaslab_class_get_space(spa_normal_class(spa)); 228 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 229 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 230 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 231 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 232 size - alloc, src); 233 234 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 235 metaslab_class_fragmentation(mc), src); 236 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 237 metaslab_class_expandable_space(mc), src); 238 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 239 (spa_mode(spa) == FREAD), src); 240 241 cap = (size == 0) ? 0 : (alloc * 100 / size); 242 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 243 244 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 245 ddt_get_pool_dedup_ratio(spa), src); 246 247 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 248 rvd->vdev_state, src); 249 250 version = spa_version(spa); 251 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 252 src = ZPROP_SRC_DEFAULT; 253 else 254 src = ZPROP_SRC_LOCAL; 255 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 256 } 257 258 if (pool != NULL) { 259 /* 260 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 261 * when opening pools before this version freedir will be NULL. 262 */ 263 if (pool->dp_free_dir != NULL) { 264 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 265 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 266 src); 267 } else { 268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 269 NULL, 0, src); 270 } 271 272 if (pool->dp_leak_dir != NULL) { 273 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 274 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 275 src); 276 } else { 277 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 278 NULL, 0, src); 279 } 280 } 281 282 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 283 284 if (spa->spa_comment != NULL) { 285 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 286 0, ZPROP_SRC_LOCAL); 287 } 288 289 if (spa->spa_root != NULL) 290 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 291 0, ZPROP_SRC_LOCAL); 292 293 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 294 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 295 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 296 } else { 297 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 298 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 299 } 300 301 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 302 if (dp->scd_path == NULL) { 303 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 304 "none", 0, ZPROP_SRC_LOCAL); 305 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 306 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 307 dp->scd_path, 0, ZPROP_SRC_LOCAL); 308 } 309 } 310} 311 312/* 313 * Get zpool property values. 314 */ 315int 316spa_prop_get(spa_t *spa, nvlist_t **nvp) 317{ 318 objset_t *mos = spa->spa_meta_objset; 319 zap_cursor_t zc; 320 zap_attribute_t za; 321 int err; 322 323 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 324 325 mutex_enter(&spa->spa_props_lock); 326 327 /* 328 * Get properties from the spa config. 329 */ 330 spa_prop_get_config(spa, nvp); 331 332 /* If no pool property object, no more prop to get. */ 333 if (mos == NULL || spa->spa_pool_props_object == 0) { 334 mutex_exit(&spa->spa_props_lock); 335 return (0); 336 } 337 338 /* 339 * Get properties from the MOS pool property object. 340 */ 341 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 342 (err = zap_cursor_retrieve(&zc, &za)) == 0; 343 zap_cursor_advance(&zc)) { 344 uint64_t intval = 0; 345 char *strval = NULL; 346 zprop_source_t src = ZPROP_SRC_DEFAULT; 347 zpool_prop_t prop; 348 349 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 350 continue; 351 352 switch (za.za_integer_length) { 353 case 8: 354 /* integer property */ 355 if (za.za_first_integer != 356 zpool_prop_default_numeric(prop)) 357 src = ZPROP_SRC_LOCAL; 358 359 if (prop == ZPOOL_PROP_BOOTFS) { 360 dsl_pool_t *dp; 361 dsl_dataset_t *ds = NULL; 362 363 dp = spa_get_dsl(spa); 364 dsl_pool_config_enter(dp, FTAG); 365 if (err = dsl_dataset_hold_obj(dp, 366 za.za_first_integer, FTAG, &ds)) { 367 dsl_pool_config_exit(dp, FTAG); 368 break; 369 } 370 371 strval = kmem_alloc( 372 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 373 KM_SLEEP); 374 dsl_dataset_name(ds, strval); 375 dsl_dataset_rele(ds, FTAG); 376 dsl_pool_config_exit(dp, FTAG); 377 } else { 378 strval = NULL; 379 intval = za.za_first_integer; 380 } 381 382 spa_prop_add_list(*nvp, prop, strval, intval, src); 383 384 if (strval != NULL) 385 kmem_free(strval, 386 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 387 388 break; 389 390 case 1: 391 /* string property */ 392 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 393 err = zap_lookup(mos, spa->spa_pool_props_object, 394 za.za_name, 1, za.za_num_integers, strval); 395 if (err) { 396 kmem_free(strval, za.za_num_integers); 397 break; 398 } 399 spa_prop_add_list(*nvp, prop, strval, 0, src); 400 kmem_free(strval, za.za_num_integers); 401 break; 402 403 default: 404 break; 405 } 406 } 407 zap_cursor_fini(&zc); 408 mutex_exit(&spa->spa_props_lock); 409out: 410 if (err && err != ENOENT) { 411 nvlist_free(*nvp); 412 *nvp = NULL; 413 return (err); 414 } 415 416 return (0); 417} 418 419/* 420 * Validate the given pool properties nvlist and modify the list 421 * for the property values to be set. 422 */ 423static int 424spa_prop_validate(spa_t *spa, nvlist_t *props) 425{ 426 nvpair_t *elem; 427 int error = 0, reset_bootfs = 0; 428 uint64_t objnum = 0; 429 boolean_t has_feature = B_FALSE; 430 431 elem = NULL; 432 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 433 uint64_t intval; 434 char *strval, *slash, *check, *fname; 435 const char *propname = nvpair_name(elem); 436 zpool_prop_t prop = zpool_name_to_prop(propname); 437 438 switch (prop) { 439 case ZPROP_INVAL: 440 if (!zpool_prop_feature(propname)) { 441 error = SET_ERROR(EINVAL); 442 break; 443 } 444 445 /* 446 * Sanitize the input. 447 */ 448 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 449 error = SET_ERROR(EINVAL); 450 break; 451 } 452 453 if (nvpair_value_uint64(elem, &intval) != 0) { 454 error = SET_ERROR(EINVAL); 455 break; 456 } 457 458 if (intval != 0) { 459 error = SET_ERROR(EINVAL); 460 break; 461 } 462 463 fname = strchr(propname, '@') + 1; 464 if (zfeature_lookup_name(fname, NULL) != 0) { 465 error = SET_ERROR(EINVAL); 466 break; 467 } 468 469 has_feature = B_TRUE; 470 break; 471 472 case ZPOOL_PROP_VERSION: 473 error = nvpair_value_uint64(elem, &intval); 474 if (!error && 475 (intval < spa_version(spa) || 476 intval > SPA_VERSION_BEFORE_FEATURES || 477 has_feature)) 478 error = SET_ERROR(EINVAL); 479 break; 480 481 case ZPOOL_PROP_DELEGATION: 482 case ZPOOL_PROP_AUTOREPLACE: 483 case ZPOOL_PROP_LISTSNAPS: 484 case ZPOOL_PROP_AUTOEXPAND: 485 error = nvpair_value_uint64(elem, &intval); 486 if (!error && intval > 1) 487 error = SET_ERROR(EINVAL); 488 break; 489 490 case ZPOOL_PROP_BOOTFS: 491 /* 492 * If the pool version is less than SPA_VERSION_BOOTFS, 493 * or the pool is still being created (version == 0), 494 * the bootfs property cannot be set. 495 */ 496 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 497 error = SET_ERROR(ENOTSUP); 498 break; 499 } 500 501 /* 502 * Make sure the vdev config is bootable 503 */ 504 if (!vdev_is_bootable(spa->spa_root_vdev)) { 505 error = SET_ERROR(ENOTSUP); 506 break; 507 } 508 509 reset_bootfs = 1; 510 511 error = nvpair_value_string(elem, &strval); 512 513 if (!error) { 514 objset_t *os; 515 uint64_t propval; 516 517 if (strval == NULL || strval[0] == '\0') { 518 objnum = zpool_prop_default_numeric( 519 ZPOOL_PROP_BOOTFS); 520 break; 521 } 522 523 if (error = dmu_objset_hold(strval, FTAG, &os)) 524 break; 525 526 /* 527 * Must be ZPL, and its property settings 528 * must be supported by GRUB (compression 529 * is not gzip, and large blocks are not used). 530 */ 531 532 if (dmu_objset_type(os) != DMU_OST_ZFS) { 533 error = SET_ERROR(ENOTSUP); 534 } else if ((error = 535 dsl_prop_get_int_ds(dmu_objset_ds(os), 536 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 537 &propval)) == 0 && 538 !BOOTFS_COMPRESS_VALID(propval)) { 539 error = SET_ERROR(ENOTSUP); 540 } else if ((error = 541 dsl_prop_get_int_ds(dmu_objset_ds(os), 542 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 543 &propval)) == 0 && 544 propval > SPA_OLD_MAXBLOCKSIZE) { 545 error = SET_ERROR(ENOTSUP); 546 } else { 547 objnum = dmu_objset_id(os); 548 } 549 dmu_objset_rele(os, FTAG); 550 } 551 break; 552 553 case ZPOOL_PROP_FAILUREMODE: 554 error = nvpair_value_uint64(elem, &intval); 555 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 556 intval > ZIO_FAILURE_MODE_PANIC)) 557 error = SET_ERROR(EINVAL); 558 559 /* 560 * This is a special case which only occurs when 561 * the pool has completely failed. This allows 562 * the user to change the in-core failmode property 563 * without syncing it out to disk (I/Os might 564 * currently be blocked). We do this by returning 565 * EIO to the caller (spa_prop_set) to trick it 566 * into thinking we encountered a property validation 567 * error. 568 */ 569 if (!error && spa_suspended(spa)) { 570 spa->spa_failmode = intval; 571 error = SET_ERROR(EIO); 572 } 573 break; 574 575 case ZPOOL_PROP_CACHEFILE: 576 if ((error = nvpair_value_string(elem, &strval)) != 0) 577 break; 578 579 if (strval[0] == '\0') 580 break; 581 582 if (strcmp(strval, "none") == 0) 583 break; 584 585 if (strval[0] != '/') { 586 error = SET_ERROR(EINVAL); 587 break; 588 } 589 590 slash = strrchr(strval, '/'); 591 ASSERT(slash != NULL); 592 593 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 594 strcmp(slash, "/..") == 0) 595 error = SET_ERROR(EINVAL); 596 break; 597 598 case ZPOOL_PROP_COMMENT: 599 if ((error = nvpair_value_string(elem, &strval)) != 0) 600 break; 601 for (check = strval; *check != '\0'; check++) { 602 /* 603 * The kernel doesn't have an easy isprint() 604 * check. For this kernel check, we merely 605 * check ASCII apart from DEL. Fix this if 606 * there is an easy-to-use kernel isprint(). 607 */ 608 if (*check >= 0x7f) { 609 error = SET_ERROR(EINVAL); 610 break; 611 } 612 check++; 613 } 614 if (strlen(strval) > ZPROP_MAX_COMMENT) 615 error = E2BIG; 616 break; 617 618 case ZPOOL_PROP_DEDUPDITTO: 619 if (spa_version(spa) < SPA_VERSION_DEDUP) 620 error = SET_ERROR(ENOTSUP); 621 else 622 error = nvpair_value_uint64(elem, &intval); 623 if (error == 0 && 624 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 625 error = SET_ERROR(EINVAL); 626 break; 627 } 628 629 if (error) 630 break; 631 } 632 633 if (!error && reset_bootfs) { 634 error = nvlist_remove(props, 635 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 636 637 if (!error) { 638 error = nvlist_add_uint64(props, 639 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 640 } 641 } 642 643 return (error); 644} 645 646void 647spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 648{ 649 char *cachefile; 650 spa_config_dirent_t *dp; 651 652 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 653 &cachefile) != 0) 654 return; 655 656 dp = kmem_alloc(sizeof (spa_config_dirent_t), 657 KM_SLEEP); 658 659 if (cachefile[0] == '\0') 660 dp->scd_path = spa_strdup(spa_config_path); 661 else if (strcmp(cachefile, "none") == 0) 662 dp->scd_path = NULL; 663 else 664 dp->scd_path = spa_strdup(cachefile); 665 666 list_insert_head(&spa->spa_config_list, dp); 667 if (need_sync) 668 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 669} 670 671int 672spa_prop_set(spa_t *spa, nvlist_t *nvp) 673{ 674 int error; 675 nvpair_t *elem = NULL; 676 boolean_t need_sync = B_FALSE; 677 678 if ((error = spa_prop_validate(spa, nvp)) != 0) 679 return (error); 680 681 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 682 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 683 684 if (prop == ZPOOL_PROP_CACHEFILE || 685 prop == ZPOOL_PROP_ALTROOT || 686 prop == ZPOOL_PROP_READONLY) 687 continue; 688 689 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 690 uint64_t ver; 691 692 if (prop == ZPOOL_PROP_VERSION) { 693 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 694 } else { 695 ASSERT(zpool_prop_feature(nvpair_name(elem))); 696 ver = SPA_VERSION_FEATURES; 697 need_sync = B_TRUE; 698 } 699 700 /* Save time if the version is already set. */ 701 if (ver == spa_version(spa)) 702 continue; 703 704 /* 705 * In addition to the pool directory object, we might 706 * create the pool properties object, the features for 707 * read object, the features for write object, or the 708 * feature descriptions object. 709 */ 710 error = dsl_sync_task(spa->spa_name, NULL, 711 spa_sync_version, &ver, 712 6, ZFS_SPACE_CHECK_RESERVED); 713 if (error) 714 return (error); 715 continue; 716 } 717 718 need_sync = B_TRUE; 719 break; 720 } 721 722 if (need_sync) { 723 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 724 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 725 } 726 727 return (0); 728} 729 730/* 731 * If the bootfs property value is dsobj, clear it. 732 */ 733void 734spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 735{ 736 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 737 VERIFY(zap_remove(spa->spa_meta_objset, 738 spa->spa_pool_props_object, 739 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 740 spa->spa_bootfs = 0; 741 } 742} 743 744/*ARGSUSED*/ 745static int 746spa_change_guid_check(void *arg, dmu_tx_t *tx) 747{ 748 uint64_t *newguid = arg; 749 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 750 vdev_t *rvd = spa->spa_root_vdev; 751 uint64_t vdev_state; 752 753 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 754 vdev_state = rvd->vdev_state; 755 spa_config_exit(spa, SCL_STATE, FTAG); 756 757 if (vdev_state != VDEV_STATE_HEALTHY) 758 return (SET_ERROR(ENXIO)); 759 760 ASSERT3U(spa_guid(spa), !=, *newguid); 761 762 return (0); 763} 764 765static void 766spa_change_guid_sync(void *arg, dmu_tx_t *tx) 767{ 768 uint64_t *newguid = arg; 769 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 770 uint64_t oldguid; 771 vdev_t *rvd = spa->spa_root_vdev; 772 773 oldguid = spa_guid(spa); 774 775 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 776 rvd->vdev_guid = *newguid; 777 rvd->vdev_guid_sum += (*newguid - oldguid); 778 vdev_config_dirty(rvd); 779 spa_config_exit(spa, SCL_STATE, FTAG); 780 781 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 782 oldguid, *newguid); 783} 784 785/* 786 * Change the GUID for the pool. This is done so that we can later 787 * re-import a pool built from a clone of our own vdevs. We will modify 788 * the root vdev's guid, our own pool guid, and then mark all of our 789 * vdevs dirty. Note that we must make sure that all our vdevs are 790 * online when we do this, or else any vdevs that weren't present 791 * would be orphaned from our pool. We are also going to issue a 792 * sysevent to update any watchers. 793 */ 794int 795spa_change_guid(spa_t *spa) 796{ 797 int error; 798 uint64_t guid; 799 800 mutex_enter(&spa->spa_vdev_top_lock); 801 mutex_enter(&spa_namespace_lock); 802 guid = spa_generate_guid(NULL); 803 804 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 805 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 806 807 if (error == 0) { 808 spa_config_sync(spa, B_FALSE, B_TRUE); 809 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 810 } 811 812 mutex_exit(&spa_namespace_lock); 813 mutex_exit(&spa->spa_vdev_top_lock); 814 815 return (error); 816} 817 818/* 819 * ========================================================================== 820 * SPA state manipulation (open/create/destroy/import/export) 821 * ========================================================================== 822 */ 823 824static int 825spa_error_entry_compare(const void *a, const void *b) 826{ 827 spa_error_entry_t *sa = (spa_error_entry_t *)a; 828 spa_error_entry_t *sb = (spa_error_entry_t *)b; 829 int ret; 830 831 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 832 sizeof (zbookmark_phys_t)); 833 834 if (ret < 0) 835 return (-1); 836 else if (ret > 0) 837 return (1); 838 else 839 return (0); 840} 841 842/* 843 * Utility function which retrieves copies of the current logs and 844 * re-initializes them in the process. 845 */ 846void 847spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 848{ 849 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 850 851 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 852 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 853 854 avl_create(&spa->spa_errlist_scrub, 855 spa_error_entry_compare, sizeof (spa_error_entry_t), 856 offsetof(spa_error_entry_t, se_avl)); 857 avl_create(&spa->spa_errlist_last, 858 spa_error_entry_compare, sizeof (spa_error_entry_t), 859 offsetof(spa_error_entry_t, se_avl)); 860} 861 862static void 863spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 864{ 865 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 866 enum zti_modes mode = ztip->zti_mode; 867 uint_t value = ztip->zti_value; 868 uint_t count = ztip->zti_count; 869 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 870 char name[32]; 871 uint_t flags = 0; 872 boolean_t batch = B_FALSE; 873 874 if (mode == ZTI_MODE_NULL) { 875 tqs->stqs_count = 0; 876 tqs->stqs_taskq = NULL; 877 return; 878 } 879 880 ASSERT3U(count, >, 0); 881 882 tqs->stqs_count = count; 883 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 884 885 switch (mode) { 886 case ZTI_MODE_FIXED: 887 ASSERT3U(value, >=, 1); 888 value = MAX(value, 1); 889 break; 890 891 case ZTI_MODE_BATCH: 892 batch = B_TRUE; 893 flags |= TASKQ_THREADS_CPU_PCT; 894 value = zio_taskq_batch_pct; 895 break; 896 897 default: 898 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 899 "spa_activate()", 900 zio_type_name[t], zio_taskq_types[q], mode, value); 901 break; 902 } 903 904 for (uint_t i = 0; i < count; i++) { 905 taskq_t *tq; 906 907 if (count > 1) { 908 (void) snprintf(name, sizeof (name), "%s_%s_%u", 909 zio_type_name[t], zio_taskq_types[q], i); 910 } else { 911 (void) snprintf(name, sizeof (name), "%s_%s", 912 zio_type_name[t], zio_taskq_types[q]); 913 } 914 915#ifdef SYSDC 916 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 917 if (batch) 918 flags |= TASKQ_DC_BATCH; 919 920 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 921 spa->spa_proc, zio_taskq_basedc, flags); 922 } else { 923#endif 924 pri_t pri = maxclsyspri; 925 /* 926 * The write issue taskq can be extremely CPU 927 * intensive. Run it at slightly lower priority 928 * than the other taskqs. 929 */ 930 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 931 pri--; 932 933 tq = taskq_create_proc(name, value, pri, 50, 934 INT_MAX, spa->spa_proc, flags); 935#ifdef SYSDC 936 } 937#endif 938 939 tqs->stqs_taskq[i] = tq; 940 } 941} 942 943static void 944spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 945{ 946 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 947 948 if (tqs->stqs_taskq == NULL) { 949 ASSERT0(tqs->stqs_count); 950 return; 951 } 952 953 for (uint_t i = 0; i < tqs->stqs_count; i++) { 954 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 955 taskq_destroy(tqs->stqs_taskq[i]); 956 } 957 958 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 959 tqs->stqs_taskq = NULL; 960} 961 962/* 963 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 964 * Note that a type may have multiple discrete taskqs to avoid lock contention 965 * on the taskq itself. In that case we choose which taskq at random by using 966 * the low bits of gethrtime(). 967 */ 968void 969spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 970 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 971{ 972 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 973 taskq_t *tq; 974 975 ASSERT3P(tqs->stqs_taskq, !=, NULL); 976 ASSERT3U(tqs->stqs_count, !=, 0); 977 978 if (tqs->stqs_count == 1) { 979 tq = tqs->stqs_taskq[0]; 980 } else { 981#ifdef _KERNEL 982 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 983#else 984 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 985#endif 986 } 987 988 taskq_dispatch_ent(tq, func, arg, flags, ent); 989} 990 991static void 992spa_create_zio_taskqs(spa_t *spa) 993{ 994 for (int t = 0; t < ZIO_TYPES; t++) { 995 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 996 spa_taskqs_init(spa, t, q); 997 } 998 } 999} 1000 1001#ifdef _KERNEL 1002#ifdef SPA_PROCESS 1003static void 1004spa_thread(void *arg) 1005{ 1006 callb_cpr_t cprinfo; 1007 1008 spa_t *spa = arg; 1009 user_t *pu = PTOU(curproc); 1010 1011 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1012 spa->spa_name); 1013 1014 ASSERT(curproc != &p0); 1015 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1016 "zpool-%s", spa->spa_name); 1017 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1018 1019#ifdef PSRSET_BIND 1020 /* bind this thread to the requested psrset */ 1021 if (zio_taskq_psrset_bind != PS_NONE) { 1022 pool_lock(); 1023 mutex_enter(&cpu_lock); 1024 mutex_enter(&pidlock); 1025 mutex_enter(&curproc->p_lock); 1026 1027 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1028 0, NULL, NULL) == 0) { 1029 curthread->t_bind_pset = zio_taskq_psrset_bind; 1030 } else { 1031 cmn_err(CE_WARN, 1032 "Couldn't bind process for zfs pool \"%s\" to " 1033 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1034 } 1035 1036 mutex_exit(&curproc->p_lock); 1037 mutex_exit(&pidlock); 1038 mutex_exit(&cpu_lock); 1039 pool_unlock(); 1040 } 1041#endif 1042 1043#ifdef SYSDC 1044 if (zio_taskq_sysdc) { 1045 sysdc_thread_enter(curthread, 100, 0); 1046 } 1047#endif 1048 1049 spa->spa_proc = curproc; 1050 spa->spa_did = curthread->t_did; 1051 1052 spa_create_zio_taskqs(spa); 1053 1054 mutex_enter(&spa->spa_proc_lock); 1055 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1056 1057 spa->spa_proc_state = SPA_PROC_ACTIVE; 1058 cv_broadcast(&spa->spa_proc_cv); 1059 1060 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1061 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1062 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1063 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1064 1065 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1066 spa->spa_proc_state = SPA_PROC_GONE; 1067 spa->spa_proc = &p0; 1068 cv_broadcast(&spa->spa_proc_cv); 1069 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1070 1071 mutex_enter(&curproc->p_lock); 1072 lwp_exit(); 1073} 1074#endif /* SPA_PROCESS */ 1075#endif 1076 1077/* 1078 * Activate an uninitialized pool. 1079 */ 1080static void 1081spa_activate(spa_t *spa, int mode) 1082{ 1083 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1084 1085 spa->spa_state = POOL_STATE_ACTIVE; 1086 spa->spa_mode = mode; 1087 1088 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1089 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1090 1091 /* Try to create a covering process */ 1092 mutex_enter(&spa->spa_proc_lock); 1093 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1094 ASSERT(spa->spa_proc == &p0); 1095 spa->spa_did = 0; 1096 1097#ifdef SPA_PROCESS 1098 /* Only create a process if we're going to be around a while. */ 1099 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1100 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1101 NULL, 0) == 0) { 1102 spa->spa_proc_state = SPA_PROC_CREATED; 1103 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1104 cv_wait(&spa->spa_proc_cv, 1105 &spa->spa_proc_lock); 1106 } 1107 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1108 ASSERT(spa->spa_proc != &p0); 1109 ASSERT(spa->spa_did != 0); 1110 } else { 1111#ifdef _KERNEL 1112 cmn_err(CE_WARN, 1113 "Couldn't create process for zfs pool \"%s\"\n", 1114 spa->spa_name); 1115#endif 1116 } 1117 } 1118#endif /* SPA_PROCESS */ 1119 mutex_exit(&spa->spa_proc_lock); 1120 1121 /* If we didn't create a process, we need to create our taskqs. */ 1122 ASSERT(spa->spa_proc == &p0); 1123 if (spa->spa_proc == &p0) { 1124 spa_create_zio_taskqs(spa); 1125 } 1126 1127 /* 1128 * Start TRIM thread. 1129 */ 1130 trim_thread_create(spa); 1131 1132 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1133 offsetof(vdev_t, vdev_config_dirty_node)); 1134 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1135 offsetof(objset_t, os_evicting_node)); 1136 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1137 offsetof(vdev_t, vdev_state_dirty_node)); 1138 1139 txg_list_create(&spa->spa_vdev_txg_list, 1140 offsetof(struct vdev, vdev_txg_node)); 1141 1142 avl_create(&spa->spa_errlist_scrub, 1143 spa_error_entry_compare, sizeof (spa_error_entry_t), 1144 offsetof(spa_error_entry_t, se_avl)); 1145 avl_create(&spa->spa_errlist_last, 1146 spa_error_entry_compare, sizeof (spa_error_entry_t), 1147 offsetof(spa_error_entry_t, se_avl)); 1148} 1149 1150/* 1151 * Opposite of spa_activate(). 1152 */ 1153static void 1154spa_deactivate(spa_t *spa) 1155{ 1156 ASSERT(spa->spa_sync_on == B_FALSE); 1157 ASSERT(spa->spa_dsl_pool == NULL); 1158 ASSERT(spa->spa_root_vdev == NULL); 1159 ASSERT(spa->spa_async_zio_root == NULL); 1160 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1161 1162 /* 1163 * Stop TRIM thread in case spa_unload() wasn't called directly 1164 * before spa_deactivate(). 1165 */ 1166 trim_thread_destroy(spa); 1167 1168 spa_evicting_os_wait(spa); 1169 1170 txg_list_destroy(&spa->spa_vdev_txg_list); 1171 1172 list_destroy(&spa->spa_config_dirty_list); 1173 list_destroy(&spa->spa_evicting_os_list); 1174 list_destroy(&spa->spa_state_dirty_list); 1175 1176 for (int t = 0; t < ZIO_TYPES; t++) { 1177 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1178 spa_taskqs_fini(spa, t, q); 1179 } 1180 } 1181 1182 metaslab_class_destroy(spa->spa_normal_class); 1183 spa->spa_normal_class = NULL; 1184 1185 metaslab_class_destroy(spa->spa_log_class); 1186 spa->spa_log_class = NULL; 1187 1188 /* 1189 * If this was part of an import or the open otherwise failed, we may 1190 * still have errors left in the queues. Empty them just in case. 1191 */ 1192 spa_errlog_drain(spa); 1193 1194 avl_destroy(&spa->spa_errlist_scrub); 1195 avl_destroy(&spa->spa_errlist_last); 1196 1197 spa->spa_state = POOL_STATE_UNINITIALIZED; 1198 1199 mutex_enter(&spa->spa_proc_lock); 1200 if (spa->spa_proc_state != SPA_PROC_NONE) { 1201 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1202 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1203 cv_broadcast(&spa->spa_proc_cv); 1204 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1205 ASSERT(spa->spa_proc != &p0); 1206 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1207 } 1208 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1209 spa->spa_proc_state = SPA_PROC_NONE; 1210 } 1211 ASSERT(spa->spa_proc == &p0); 1212 mutex_exit(&spa->spa_proc_lock); 1213 1214#ifdef SPA_PROCESS 1215 /* 1216 * We want to make sure spa_thread() has actually exited the ZFS 1217 * module, so that the module can't be unloaded out from underneath 1218 * it. 1219 */ 1220 if (spa->spa_did != 0) { 1221 thread_join(spa->spa_did); 1222 spa->spa_did = 0; 1223 } 1224#endif /* SPA_PROCESS */ 1225} 1226 1227/* 1228 * Verify a pool configuration, and construct the vdev tree appropriately. This 1229 * will create all the necessary vdevs in the appropriate layout, with each vdev 1230 * in the CLOSED state. This will prep the pool before open/creation/import. 1231 * All vdev validation is done by the vdev_alloc() routine. 1232 */ 1233static int 1234spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1235 uint_t id, int atype) 1236{ 1237 nvlist_t **child; 1238 uint_t children; 1239 int error; 1240 1241 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1242 return (error); 1243 1244 if ((*vdp)->vdev_ops->vdev_op_leaf) 1245 return (0); 1246 1247 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1248 &child, &children); 1249 1250 if (error == ENOENT) 1251 return (0); 1252 1253 if (error) { 1254 vdev_free(*vdp); 1255 *vdp = NULL; 1256 return (SET_ERROR(EINVAL)); 1257 } 1258 1259 for (int c = 0; c < children; c++) { 1260 vdev_t *vd; 1261 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1262 atype)) != 0) { 1263 vdev_free(*vdp); 1264 *vdp = NULL; 1265 return (error); 1266 } 1267 } 1268 1269 ASSERT(*vdp != NULL); 1270 1271 return (0); 1272} 1273 1274/* 1275 * Opposite of spa_load(). 1276 */ 1277static void 1278spa_unload(spa_t *spa) 1279{ 1280 int i; 1281 1282 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1283 1284 /* 1285 * Stop TRIM thread. 1286 */ 1287 trim_thread_destroy(spa); 1288 1289 /* 1290 * Stop async tasks. 1291 */ 1292 spa_async_suspend(spa); 1293 1294 /* 1295 * Stop syncing. 1296 */ 1297 if (spa->spa_sync_on) { 1298 txg_sync_stop(spa->spa_dsl_pool); 1299 spa->spa_sync_on = B_FALSE; 1300 } 1301 1302 /* 1303 * Wait for any outstanding async I/O to complete. 1304 */ 1305 if (spa->spa_async_zio_root != NULL) { 1306 for (int i = 0; i < max_ncpus; i++) 1307 (void) zio_wait(spa->spa_async_zio_root[i]); 1308 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1309 spa->spa_async_zio_root = NULL; 1310 } 1311 1312 bpobj_close(&spa->spa_deferred_bpobj); 1313 1314 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1315 1316 /* 1317 * Close all vdevs. 1318 */ 1319 if (spa->spa_root_vdev) 1320 vdev_free(spa->spa_root_vdev); 1321 ASSERT(spa->spa_root_vdev == NULL); 1322 1323 /* 1324 * Close the dsl pool. 1325 */ 1326 if (spa->spa_dsl_pool) { 1327 dsl_pool_close(spa->spa_dsl_pool); 1328 spa->spa_dsl_pool = NULL; 1329 spa->spa_meta_objset = NULL; 1330 } 1331 1332 ddt_unload(spa); 1333 1334 1335 /* 1336 * Drop and purge level 2 cache 1337 */ 1338 spa_l2cache_drop(spa); 1339 1340 for (i = 0; i < spa->spa_spares.sav_count; i++) 1341 vdev_free(spa->spa_spares.sav_vdevs[i]); 1342 if (spa->spa_spares.sav_vdevs) { 1343 kmem_free(spa->spa_spares.sav_vdevs, 1344 spa->spa_spares.sav_count * sizeof (void *)); 1345 spa->spa_spares.sav_vdevs = NULL; 1346 } 1347 if (spa->spa_spares.sav_config) { 1348 nvlist_free(spa->spa_spares.sav_config); 1349 spa->spa_spares.sav_config = NULL; 1350 } 1351 spa->spa_spares.sav_count = 0; 1352 1353 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1354 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1355 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1356 } 1357 if (spa->spa_l2cache.sav_vdevs) { 1358 kmem_free(spa->spa_l2cache.sav_vdevs, 1359 spa->spa_l2cache.sav_count * sizeof (void *)); 1360 spa->spa_l2cache.sav_vdevs = NULL; 1361 } 1362 if (spa->spa_l2cache.sav_config) { 1363 nvlist_free(spa->spa_l2cache.sav_config); 1364 spa->spa_l2cache.sav_config = NULL; 1365 } 1366 spa->spa_l2cache.sav_count = 0; 1367 1368 spa->spa_async_suspended = 0; 1369 1370 if (spa->spa_comment != NULL) { 1371 spa_strfree(spa->spa_comment); 1372 spa->spa_comment = NULL; 1373 } 1374 1375 spa_config_exit(spa, SCL_ALL, FTAG); 1376} 1377 1378/* 1379 * Load (or re-load) the current list of vdevs describing the active spares for 1380 * this pool. When this is called, we have some form of basic information in 1381 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1382 * then re-generate a more complete list including status information. 1383 */ 1384static void 1385spa_load_spares(spa_t *spa) 1386{ 1387 nvlist_t **spares; 1388 uint_t nspares; 1389 int i; 1390 vdev_t *vd, *tvd; 1391 1392 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1393 1394 /* 1395 * First, close and free any existing spare vdevs. 1396 */ 1397 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1398 vd = spa->spa_spares.sav_vdevs[i]; 1399 1400 /* Undo the call to spa_activate() below */ 1401 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1402 B_FALSE)) != NULL && tvd->vdev_isspare) 1403 spa_spare_remove(tvd); 1404 vdev_close(vd); 1405 vdev_free(vd); 1406 } 1407 1408 if (spa->spa_spares.sav_vdevs) 1409 kmem_free(spa->spa_spares.sav_vdevs, 1410 spa->spa_spares.sav_count * sizeof (void *)); 1411 1412 if (spa->spa_spares.sav_config == NULL) 1413 nspares = 0; 1414 else 1415 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1416 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1417 1418 spa->spa_spares.sav_count = (int)nspares; 1419 spa->spa_spares.sav_vdevs = NULL; 1420 1421 if (nspares == 0) 1422 return; 1423 1424 /* 1425 * Construct the array of vdevs, opening them to get status in the 1426 * process. For each spare, there is potentially two different vdev_t 1427 * structures associated with it: one in the list of spares (used only 1428 * for basic validation purposes) and one in the active vdev 1429 * configuration (if it's spared in). During this phase we open and 1430 * validate each vdev on the spare list. If the vdev also exists in the 1431 * active configuration, then we also mark this vdev as an active spare. 1432 */ 1433 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1434 KM_SLEEP); 1435 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1436 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1437 VDEV_ALLOC_SPARE) == 0); 1438 ASSERT(vd != NULL); 1439 1440 spa->spa_spares.sav_vdevs[i] = vd; 1441 1442 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1443 B_FALSE)) != NULL) { 1444 if (!tvd->vdev_isspare) 1445 spa_spare_add(tvd); 1446 1447 /* 1448 * We only mark the spare active if we were successfully 1449 * able to load the vdev. Otherwise, importing a pool 1450 * with a bad active spare would result in strange 1451 * behavior, because multiple pool would think the spare 1452 * is actively in use. 1453 * 1454 * There is a vulnerability here to an equally bizarre 1455 * circumstance, where a dead active spare is later 1456 * brought back to life (onlined or otherwise). Given 1457 * the rarity of this scenario, and the extra complexity 1458 * it adds, we ignore the possibility. 1459 */ 1460 if (!vdev_is_dead(tvd)) 1461 spa_spare_activate(tvd); 1462 } 1463 1464 vd->vdev_top = vd; 1465 vd->vdev_aux = &spa->spa_spares; 1466 1467 if (vdev_open(vd) != 0) 1468 continue; 1469 1470 if (vdev_validate_aux(vd) == 0) 1471 spa_spare_add(vd); 1472 } 1473 1474 /* 1475 * Recompute the stashed list of spares, with status information 1476 * this time. 1477 */ 1478 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1479 DATA_TYPE_NVLIST_ARRAY) == 0); 1480 1481 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1482 KM_SLEEP); 1483 for (i = 0; i < spa->spa_spares.sav_count; i++) 1484 spares[i] = vdev_config_generate(spa, 1485 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1486 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1487 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1488 for (i = 0; i < spa->spa_spares.sav_count; i++) 1489 nvlist_free(spares[i]); 1490 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1491} 1492 1493/* 1494 * Load (or re-load) the current list of vdevs describing the active l2cache for 1495 * this pool. When this is called, we have some form of basic information in 1496 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1497 * then re-generate a more complete list including status information. 1498 * Devices which are already active have their details maintained, and are 1499 * not re-opened. 1500 */ 1501static void 1502spa_load_l2cache(spa_t *spa) 1503{ 1504 nvlist_t **l2cache; 1505 uint_t nl2cache; 1506 int i, j, oldnvdevs; 1507 uint64_t guid; 1508 vdev_t *vd, **oldvdevs, **newvdevs; 1509 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1510 1511 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1512 1513 if (sav->sav_config != NULL) { 1514 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1515 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1516 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1517 } else { 1518 nl2cache = 0; 1519 newvdevs = NULL; 1520 } 1521 1522 oldvdevs = sav->sav_vdevs; 1523 oldnvdevs = sav->sav_count; 1524 sav->sav_vdevs = NULL; 1525 sav->sav_count = 0; 1526 1527 /* 1528 * Process new nvlist of vdevs. 1529 */ 1530 for (i = 0; i < nl2cache; i++) { 1531 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1532 &guid) == 0); 1533 1534 newvdevs[i] = NULL; 1535 for (j = 0; j < oldnvdevs; j++) { 1536 vd = oldvdevs[j]; 1537 if (vd != NULL && guid == vd->vdev_guid) { 1538 /* 1539 * Retain previous vdev for add/remove ops. 1540 */ 1541 newvdevs[i] = vd; 1542 oldvdevs[j] = NULL; 1543 break; 1544 } 1545 } 1546 1547 if (newvdevs[i] == NULL) { 1548 /* 1549 * Create new vdev 1550 */ 1551 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1552 VDEV_ALLOC_L2CACHE) == 0); 1553 ASSERT(vd != NULL); 1554 newvdevs[i] = vd; 1555 1556 /* 1557 * Commit this vdev as an l2cache device, 1558 * even if it fails to open. 1559 */ 1560 spa_l2cache_add(vd); 1561 1562 vd->vdev_top = vd; 1563 vd->vdev_aux = sav; 1564 1565 spa_l2cache_activate(vd); 1566 1567 if (vdev_open(vd) != 0) 1568 continue; 1569 1570 (void) vdev_validate_aux(vd); 1571 1572 if (!vdev_is_dead(vd)) 1573 l2arc_add_vdev(spa, vd); 1574 } 1575 } 1576 1577 /* 1578 * Purge vdevs that were dropped 1579 */ 1580 for (i = 0; i < oldnvdevs; i++) { 1581 uint64_t pool; 1582 1583 vd = oldvdevs[i]; 1584 if (vd != NULL) { 1585 ASSERT(vd->vdev_isl2cache); 1586 1587 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1588 pool != 0ULL && l2arc_vdev_present(vd)) 1589 l2arc_remove_vdev(vd); 1590 vdev_clear_stats(vd); 1591 vdev_free(vd); 1592 } 1593 } 1594 1595 if (oldvdevs) 1596 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1597 1598 if (sav->sav_config == NULL) 1599 goto out; 1600 1601 sav->sav_vdevs = newvdevs; 1602 sav->sav_count = (int)nl2cache; 1603 1604 /* 1605 * Recompute the stashed list of l2cache devices, with status 1606 * information this time. 1607 */ 1608 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1609 DATA_TYPE_NVLIST_ARRAY) == 0); 1610 1611 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1612 for (i = 0; i < sav->sav_count; i++) 1613 l2cache[i] = vdev_config_generate(spa, 1614 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1615 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1616 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1617out: 1618 for (i = 0; i < sav->sav_count; i++) 1619 nvlist_free(l2cache[i]); 1620 if (sav->sav_count) 1621 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1622} 1623 1624static int 1625load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1626{ 1627 dmu_buf_t *db; 1628 char *packed = NULL; 1629 size_t nvsize = 0; 1630 int error; 1631 *value = NULL; 1632 1633 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1634 if (error != 0) 1635 return (error); 1636 1637 nvsize = *(uint64_t *)db->db_data; 1638 dmu_buf_rele(db, FTAG); 1639 1640 packed = kmem_alloc(nvsize, KM_SLEEP); 1641 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1642 DMU_READ_PREFETCH); 1643 if (error == 0) 1644 error = nvlist_unpack(packed, nvsize, value, 0); 1645 kmem_free(packed, nvsize); 1646 1647 return (error); 1648} 1649 1650/* 1651 * Checks to see if the given vdev could not be opened, in which case we post a 1652 * sysevent to notify the autoreplace code that the device has been removed. 1653 */ 1654static void 1655spa_check_removed(vdev_t *vd) 1656{ 1657 for (int c = 0; c < vd->vdev_children; c++) 1658 spa_check_removed(vd->vdev_child[c]); 1659 1660 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1661 !vd->vdev_ishole) { 1662 zfs_post_autoreplace(vd->vdev_spa, vd); 1663 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1664 } 1665} 1666 1667/* 1668 * Validate the current config against the MOS config 1669 */ 1670static boolean_t 1671spa_config_valid(spa_t *spa, nvlist_t *config) 1672{ 1673 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1674 nvlist_t *nv; 1675 1676 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1677 1678 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1679 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1680 1681 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1682 1683 /* 1684 * If we're doing a normal import, then build up any additional 1685 * diagnostic information about missing devices in this config. 1686 * We'll pass this up to the user for further processing. 1687 */ 1688 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1689 nvlist_t **child, *nv; 1690 uint64_t idx = 0; 1691 1692 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1693 KM_SLEEP); 1694 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1695 1696 for (int c = 0; c < rvd->vdev_children; c++) { 1697 vdev_t *tvd = rvd->vdev_child[c]; 1698 vdev_t *mtvd = mrvd->vdev_child[c]; 1699 1700 if (tvd->vdev_ops == &vdev_missing_ops && 1701 mtvd->vdev_ops != &vdev_missing_ops && 1702 mtvd->vdev_islog) 1703 child[idx++] = vdev_config_generate(spa, mtvd, 1704 B_FALSE, 0); 1705 } 1706 1707 if (idx) { 1708 VERIFY(nvlist_add_nvlist_array(nv, 1709 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1710 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1711 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1712 1713 for (int i = 0; i < idx; i++) 1714 nvlist_free(child[i]); 1715 } 1716 nvlist_free(nv); 1717 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1718 } 1719 1720 /* 1721 * Compare the root vdev tree with the information we have 1722 * from the MOS config (mrvd). Check each top-level vdev 1723 * with the corresponding MOS config top-level (mtvd). 1724 */ 1725 for (int c = 0; c < rvd->vdev_children; c++) { 1726 vdev_t *tvd = rvd->vdev_child[c]; 1727 vdev_t *mtvd = mrvd->vdev_child[c]; 1728 1729 /* 1730 * Resolve any "missing" vdevs in the current configuration. 1731 * If we find that the MOS config has more accurate information 1732 * about the top-level vdev then use that vdev instead. 1733 */ 1734 if (tvd->vdev_ops == &vdev_missing_ops && 1735 mtvd->vdev_ops != &vdev_missing_ops) { 1736 1737 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1738 continue; 1739 1740 /* 1741 * Device specific actions. 1742 */ 1743 if (mtvd->vdev_islog) { 1744 spa_set_log_state(spa, SPA_LOG_CLEAR); 1745 } else { 1746 /* 1747 * XXX - once we have 'readonly' pool 1748 * support we should be able to handle 1749 * missing data devices by transitioning 1750 * the pool to readonly. 1751 */ 1752 continue; 1753 } 1754 1755 /* 1756 * Swap the missing vdev with the data we were 1757 * able to obtain from the MOS config. 1758 */ 1759 vdev_remove_child(rvd, tvd); 1760 vdev_remove_child(mrvd, mtvd); 1761 1762 vdev_add_child(rvd, mtvd); 1763 vdev_add_child(mrvd, tvd); 1764 1765 spa_config_exit(spa, SCL_ALL, FTAG); 1766 vdev_load(mtvd); 1767 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1768 1769 vdev_reopen(rvd); 1770 } else if (mtvd->vdev_islog) { 1771 /* 1772 * Load the slog device's state from the MOS config 1773 * since it's possible that the label does not 1774 * contain the most up-to-date information. 1775 */ 1776 vdev_load_log_state(tvd, mtvd); 1777 vdev_reopen(tvd); 1778 } 1779 } 1780 vdev_free(mrvd); 1781 spa_config_exit(spa, SCL_ALL, FTAG); 1782 1783 /* 1784 * Ensure we were able to validate the config. 1785 */ 1786 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1787} 1788 1789/* 1790 * Check for missing log devices 1791 */ 1792static boolean_t 1793spa_check_logs(spa_t *spa) 1794{ 1795 boolean_t rv = B_FALSE; 1796 dsl_pool_t *dp = spa_get_dsl(spa); 1797 1798 switch (spa->spa_log_state) { 1799 case SPA_LOG_MISSING: 1800 /* need to recheck in case slog has been restored */ 1801 case SPA_LOG_UNKNOWN: 1802 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1803 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1804 if (rv) 1805 spa_set_log_state(spa, SPA_LOG_MISSING); 1806 break; 1807 } 1808 return (rv); 1809} 1810 1811static boolean_t 1812spa_passivate_log(spa_t *spa) 1813{ 1814 vdev_t *rvd = spa->spa_root_vdev; 1815 boolean_t slog_found = B_FALSE; 1816 1817 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1818 1819 if (!spa_has_slogs(spa)) 1820 return (B_FALSE); 1821 1822 for (int c = 0; c < rvd->vdev_children; c++) { 1823 vdev_t *tvd = rvd->vdev_child[c]; 1824 metaslab_group_t *mg = tvd->vdev_mg; 1825 1826 if (tvd->vdev_islog) { 1827 metaslab_group_passivate(mg); 1828 slog_found = B_TRUE; 1829 } 1830 } 1831 1832 return (slog_found); 1833} 1834 1835static void 1836spa_activate_log(spa_t *spa) 1837{ 1838 vdev_t *rvd = spa->spa_root_vdev; 1839 1840 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1841 1842 for (int c = 0; c < rvd->vdev_children; c++) { 1843 vdev_t *tvd = rvd->vdev_child[c]; 1844 metaslab_group_t *mg = tvd->vdev_mg; 1845 1846 if (tvd->vdev_islog) 1847 metaslab_group_activate(mg); 1848 } 1849} 1850 1851int 1852spa_offline_log(spa_t *spa) 1853{ 1854 int error; 1855 1856 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1857 NULL, DS_FIND_CHILDREN); 1858 if (error == 0) { 1859 /* 1860 * We successfully offlined the log device, sync out the 1861 * current txg so that the "stubby" block can be removed 1862 * by zil_sync(). 1863 */ 1864 txg_wait_synced(spa->spa_dsl_pool, 0); 1865 } 1866 return (error); 1867} 1868 1869static void 1870spa_aux_check_removed(spa_aux_vdev_t *sav) 1871{ 1872 int i; 1873 1874 for (i = 0; i < sav->sav_count; i++) 1875 spa_check_removed(sav->sav_vdevs[i]); 1876} 1877 1878void 1879spa_claim_notify(zio_t *zio) 1880{ 1881 spa_t *spa = zio->io_spa; 1882 1883 if (zio->io_error) 1884 return; 1885 1886 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1887 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1888 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1889 mutex_exit(&spa->spa_props_lock); 1890} 1891 1892typedef struct spa_load_error { 1893 uint64_t sle_meta_count; 1894 uint64_t sle_data_count; 1895} spa_load_error_t; 1896 1897static void 1898spa_load_verify_done(zio_t *zio) 1899{ 1900 blkptr_t *bp = zio->io_bp; 1901 spa_load_error_t *sle = zio->io_private; 1902 dmu_object_type_t type = BP_GET_TYPE(bp); 1903 int error = zio->io_error; 1904 spa_t *spa = zio->io_spa; 1905 1906 if (error) { 1907 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1908 type != DMU_OT_INTENT_LOG) 1909 atomic_inc_64(&sle->sle_meta_count); 1910 else 1911 atomic_inc_64(&sle->sle_data_count); 1912 } 1913 zio_data_buf_free(zio->io_data, zio->io_size); 1914 1915 mutex_enter(&spa->spa_scrub_lock); 1916 spa->spa_scrub_inflight--; 1917 cv_broadcast(&spa->spa_scrub_io_cv); 1918 mutex_exit(&spa->spa_scrub_lock); 1919} 1920 1921/* 1922 * Maximum number of concurrent scrub i/os to create while verifying 1923 * a pool while importing it. 1924 */ 1925int spa_load_verify_maxinflight = 10000; 1926boolean_t spa_load_verify_metadata = B_TRUE; 1927boolean_t spa_load_verify_data = B_TRUE; 1928 1929SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1930 &spa_load_verify_maxinflight, 0, 1931 "Maximum number of concurrent scrub I/Os to create while verifying a " 1932 "pool while importing it"); 1933 1934SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1935 &spa_load_verify_metadata, 0, 1936 "Check metadata on import?"); 1937 1938SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1939 &spa_load_verify_data, 0, 1940 "Check user data on import?"); 1941 1942/*ARGSUSED*/ 1943static int 1944spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1945 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1946{ 1947 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1948 return (0); 1949 /* 1950 * Note: normally this routine will not be called if 1951 * spa_load_verify_metadata is not set. However, it may be useful 1952 * to manually set the flag after the traversal has begun. 1953 */ 1954 if (!spa_load_verify_metadata) 1955 return (0); 1956 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1957 return (0); 1958 1959 zio_t *rio = arg; 1960 size_t size = BP_GET_PSIZE(bp); 1961 void *data = zio_data_buf_alloc(size); 1962 1963 mutex_enter(&spa->spa_scrub_lock); 1964 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1965 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1966 spa->spa_scrub_inflight++; 1967 mutex_exit(&spa->spa_scrub_lock); 1968 1969 zio_nowait(zio_read(rio, spa, bp, data, size, 1970 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1971 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1972 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1973 return (0); 1974} 1975 1976static int 1977spa_load_verify(spa_t *spa) 1978{ 1979 zio_t *rio; 1980 spa_load_error_t sle = { 0 }; 1981 zpool_rewind_policy_t policy; 1982 boolean_t verify_ok = B_FALSE; 1983 int error = 0; 1984 1985 zpool_get_rewind_policy(spa->spa_config, &policy); 1986 1987 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1988 return (0); 1989 1990 rio = zio_root(spa, NULL, &sle, 1991 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1992 1993 if (spa_load_verify_metadata) { 1994 error = traverse_pool(spa, spa->spa_verify_min_txg, 1995 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1996 spa_load_verify_cb, rio); 1997 } 1998 1999 (void) zio_wait(rio); 2000 2001 spa->spa_load_meta_errors = sle.sle_meta_count; 2002 spa->spa_load_data_errors = sle.sle_data_count; 2003 2004 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2005 sle.sle_data_count <= policy.zrp_maxdata) { 2006 int64_t loss = 0; 2007 2008 verify_ok = B_TRUE; 2009 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2010 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2011 2012 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2013 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2014 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2015 VERIFY(nvlist_add_int64(spa->spa_load_info, 2016 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2017 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2018 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2019 } else { 2020 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2021 } 2022 2023 if (error) { 2024 if (error != ENXIO && error != EIO) 2025 error = SET_ERROR(EIO); 2026 return (error); 2027 } 2028 2029 return (verify_ok ? 0 : EIO); 2030} 2031 2032/* 2033 * Find a value in the pool props object. 2034 */ 2035static void 2036spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2037{ 2038 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2039 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2040} 2041 2042/* 2043 * Find a value in the pool directory object. 2044 */ 2045static int 2046spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2047{ 2048 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2049 name, sizeof (uint64_t), 1, val)); 2050} 2051 2052static int 2053spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2054{ 2055 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2056 return (err); 2057} 2058 2059/* 2060 * Fix up config after a partly-completed split. This is done with the 2061 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2062 * pool have that entry in their config, but only the splitting one contains 2063 * a list of all the guids of the vdevs that are being split off. 2064 * 2065 * This function determines what to do with that list: either rejoin 2066 * all the disks to the pool, or complete the splitting process. To attempt 2067 * the rejoin, each disk that is offlined is marked online again, and 2068 * we do a reopen() call. If the vdev label for every disk that was 2069 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2070 * then we call vdev_split() on each disk, and complete the split. 2071 * 2072 * Otherwise we leave the config alone, with all the vdevs in place in 2073 * the original pool. 2074 */ 2075static void 2076spa_try_repair(spa_t *spa, nvlist_t *config) 2077{ 2078 uint_t extracted; 2079 uint64_t *glist; 2080 uint_t i, gcount; 2081 nvlist_t *nvl; 2082 vdev_t **vd; 2083 boolean_t attempt_reopen; 2084 2085 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2086 return; 2087 2088 /* check that the config is complete */ 2089 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2090 &glist, &gcount) != 0) 2091 return; 2092 2093 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2094 2095 /* attempt to online all the vdevs & validate */ 2096 attempt_reopen = B_TRUE; 2097 for (i = 0; i < gcount; i++) { 2098 if (glist[i] == 0) /* vdev is hole */ 2099 continue; 2100 2101 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2102 if (vd[i] == NULL) { 2103 /* 2104 * Don't bother attempting to reopen the disks; 2105 * just do the split. 2106 */ 2107 attempt_reopen = B_FALSE; 2108 } else { 2109 /* attempt to re-online it */ 2110 vd[i]->vdev_offline = B_FALSE; 2111 } 2112 } 2113 2114 if (attempt_reopen) { 2115 vdev_reopen(spa->spa_root_vdev); 2116 2117 /* check each device to see what state it's in */ 2118 for (extracted = 0, i = 0; i < gcount; i++) { 2119 if (vd[i] != NULL && 2120 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2121 break; 2122 ++extracted; 2123 } 2124 } 2125 2126 /* 2127 * If every disk has been moved to the new pool, or if we never 2128 * even attempted to look at them, then we split them off for 2129 * good. 2130 */ 2131 if (!attempt_reopen || gcount == extracted) { 2132 for (i = 0; i < gcount; i++) 2133 if (vd[i] != NULL) 2134 vdev_split(vd[i]); 2135 vdev_reopen(spa->spa_root_vdev); 2136 } 2137 2138 kmem_free(vd, gcount * sizeof (vdev_t *)); 2139} 2140 2141static int 2142spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2143 boolean_t mosconfig) 2144{ 2145 nvlist_t *config = spa->spa_config; 2146 char *ereport = FM_EREPORT_ZFS_POOL; 2147 char *comment; 2148 int error; 2149 uint64_t pool_guid; 2150 nvlist_t *nvl; 2151 2152 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2153 return (SET_ERROR(EINVAL)); 2154 2155 ASSERT(spa->spa_comment == NULL); 2156 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2157 spa->spa_comment = spa_strdup(comment); 2158 2159 /* 2160 * Versioning wasn't explicitly added to the label until later, so if 2161 * it's not present treat it as the initial version. 2162 */ 2163 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2164 &spa->spa_ubsync.ub_version) != 0) 2165 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2166 2167 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2168 &spa->spa_config_txg); 2169 2170 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2171 spa_guid_exists(pool_guid, 0)) { 2172 error = SET_ERROR(EEXIST); 2173 } else { 2174 spa->spa_config_guid = pool_guid; 2175 2176 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2177 &nvl) == 0) { 2178 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2179 KM_SLEEP) == 0); 2180 } 2181 2182 nvlist_free(spa->spa_load_info); 2183 spa->spa_load_info = fnvlist_alloc(); 2184 2185 gethrestime(&spa->spa_loaded_ts); 2186 error = spa_load_impl(spa, pool_guid, config, state, type, 2187 mosconfig, &ereport); 2188 } 2189 2190 /* 2191 * Don't count references from objsets that are already closed 2192 * and are making their way through the eviction process. 2193 */ 2194 spa_evicting_os_wait(spa); 2195 spa->spa_minref = refcount_count(&spa->spa_refcount); 2196 if (error) { 2197 if (error != EEXIST) { 2198 spa->spa_loaded_ts.tv_sec = 0; 2199 spa->spa_loaded_ts.tv_nsec = 0; 2200 } 2201 if (error != EBADF) { 2202 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2203 } 2204 } 2205 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2206 spa->spa_ena = 0; 2207 2208 return (error); 2209} 2210 2211/* 2212 * Load an existing storage pool, using the pool's builtin spa_config as a 2213 * source of configuration information. 2214 */ 2215static int 2216spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2217 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2218 char **ereport) 2219{ 2220 int error = 0; 2221 nvlist_t *nvroot = NULL; 2222 nvlist_t *label; 2223 vdev_t *rvd; 2224 uberblock_t *ub = &spa->spa_uberblock; 2225 uint64_t children, config_cache_txg = spa->spa_config_txg; 2226 int orig_mode = spa->spa_mode; 2227 int parse; 2228 uint64_t obj; 2229 boolean_t missing_feat_write = B_FALSE; 2230 2231 /* 2232 * If this is an untrusted config, access the pool in read-only mode. 2233 * This prevents things like resilvering recently removed devices. 2234 */ 2235 if (!mosconfig) 2236 spa->spa_mode = FREAD; 2237 2238 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2239 2240 spa->spa_load_state = state; 2241 2242 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2243 return (SET_ERROR(EINVAL)); 2244 2245 parse = (type == SPA_IMPORT_EXISTING ? 2246 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2247 2248 /* 2249 * Create "The Godfather" zio to hold all async IOs 2250 */ 2251 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2252 KM_SLEEP); 2253 for (int i = 0; i < max_ncpus; i++) { 2254 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2255 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2256 ZIO_FLAG_GODFATHER); 2257 } 2258 2259 /* 2260 * Parse the configuration into a vdev tree. We explicitly set the 2261 * value that will be returned by spa_version() since parsing the 2262 * configuration requires knowing the version number. 2263 */ 2264 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2265 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2266 spa_config_exit(spa, SCL_ALL, FTAG); 2267 2268 if (error != 0) 2269 return (error); 2270 2271 ASSERT(spa->spa_root_vdev == rvd); 2272 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2273 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2274 2275 if (type != SPA_IMPORT_ASSEMBLE) { 2276 ASSERT(spa_guid(spa) == pool_guid); 2277 } 2278 2279 /* 2280 * Try to open all vdevs, loading each label in the process. 2281 */ 2282 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2283 error = vdev_open(rvd); 2284 spa_config_exit(spa, SCL_ALL, FTAG); 2285 if (error != 0) 2286 return (error); 2287 2288 /* 2289 * We need to validate the vdev labels against the configuration that 2290 * we have in hand, which is dependent on the setting of mosconfig. If 2291 * mosconfig is true then we're validating the vdev labels based on 2292 * that config. Otherwise, we're validating against the cached config 2293 * (zpool.cache) that was read when we loaded the zfs module, and then 2294 * later we will recursively call spa_load() and validate against 2295 * the vdev config. 2296 * 2297 * If we're assembling a new pool that's been split off from an 2298 * existing pool, the labels haven't yet been updated so we skip 2299 * validation for now. 2300 */ 2301 if (type != SPA_IMPORT_ASSEMBLE) { 2302 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2303 error = vdev_validate(rvd, mosconfig); 2304 spa_config_exit(spa, SCL_ALL, FTAG); 2305 2306 if (error != 0) 2307 return (error); 2308 2309 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2310 return (SET_ERROR(ENXIO)); 2311 } 2312 2313 /* 2314 * Find the best uberblock. 2315 */ 2316 vdev_uberblock_load(rvd, ub, &label); 2317 2318 /* 2319 * If we weren't able to find a single valid uberblock, return failure. 2320 */ 2321 if (ub->ub_txg == 0) { 2322 nvlist_free(label); 2323 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2324 } 2325 2326 /* 2327 * If the pool has an unsupported version we can't open it. 2328 */ 2329 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2330 nvlist_free(label); 2331 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2332 } 2333 2334 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2335 nvlist_t *features; 2336 2337 /* 2338 * If we weren't able to find what's necessary for reading the 2339 * MOS in the label, return failure. 2340 */ 2341 if (label == NULL || nvlist_lookup_nvlist(label, 2342 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2343 nvlist_free(label); 2344 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2345 ENXIO)); 2346 } 2347 2348 /* 2349 * Update our in-core representation with the definitive values 2350 * from the label. 2351 */ 2352 nvlist_free(spa->spa_label_features); 2353 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2354 } 2355 2356 nvlist_free(label); 2357 2358 /* 2359 * Look through entries in the label nvlist's features_for_read. If 2360 * there is a feature listed there which we don't understand then we 2361 * cannot open a pool. 2362 */ 2363 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2364 nvlist_t *unsup_feat; 2365 2366 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2367 0); 2368 2369 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2370 NULL); nvp != NULL; 2371 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2372 if (!zfeature_is_supported(nvpair_name(nvp))) { 2373 VERIFY(nvlist_add_string(unsup_feat, 2374 nvpair_name(nvp), "") == 0); 2375 } 2376 } 2377 2378 if (!nvlist_empty(unsup_feat)) { 2379 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2380 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2381 nvlist_free(unsup_feat); 2382 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2383 ENOTSUP)); 2384 } 2385 2386 nvlist_free(unsup_feat); 2387 } 2388 2389 /* 2390 * If the vdev guid sum doesn't match the uberblock, we have an 2391 * incomplete configuration. We first check to see if the pool 2392 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2393 * If it is, defer the vdev_guid_sum check till later so we 2394 * can handle missing vdevs. 2395 */ 2396 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2397 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2398 rvd->vdev_guid_sum != ub->ub_guid_sum) 2399 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2400 2401 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2402 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2403 spa_try_repair(spa, config); 2404 spa_config_exit(spa, SCL_ALL, FTAG); 2405 nvlist_free(spa->spa_config_splitting); 2406 spa->spa_config_splitting = NULL; 2407 } 2408 2409 /* 2410 * Initialize internal SPA structures. 2411 */ 2412 spa->spa_state = POOL_STATE_ACTIVE; 2413 spa->spa_ubsync = spa->spa_uberblock; 2414 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2415 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2416 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2417 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2418 spa->spa_claim_max_txg = spa->spa_first_txg; 2419 spa->spa_prev_software_version = ub->ub_software_version; 2420 2421 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2422 if (error) 2423 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2424 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2425 2426 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2427 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2428 2429 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2430 boolean_t missing_feat_read = B_FALSE; 2431 nvlist_t *unsup_feat, *enabled_feat; 2432 2433 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2434 &spa->spa_feat_for_read_obj) != 0) { 2435 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2436 } 2437 2438 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2439 &spa->spa_feat_for_write_obj) != 0) { 2440 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2441 } 2442 2443 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2444 &spa->spa_feat_desc_obj) != 0) { 2445 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2446 } 2447 2448 enabled_feat = fnvlist_alloc(); 2449 unsup_feat = fnvlist_alloc(); 2450 2451 if (!spa_features_check(spa, B_FALSE, 2452 unsup_feat, enabled_feat)) 2453 missing_feat_read = B_TRUE; 2454 2455 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2456 if (!spa_features_check(spa, B_TRUE, 2457 unsup_feat, enabled_feat)) { 2458 missing_feat_write = B_TRUE; 2459 } 2460 } 2461 2462 fnvlist_add_nvlist(spa->spa_load_info, 2463 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2464 2465 if (!nvlist_empty(unsup_feat)) { 2466 fnvlist_add_nvlist(spa->spa_load_info, 2467 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2468 } 2469 2470 fnvlist_free(enabled_feat); 2471 fnvlist_free(unsup_feat); 2472 2473 if (!missing_feat_read) { 2474 fnvlist_add_boolean(spa->spa_load_info, 2475 ZPOOL_CONFIG_CAN_RDONLY); 2476 } 2477 2478 /* 2479 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2480 * twofold: to determine whether the pool is available for 2481 * import in read-write mode and (if it is not) whether the 2482 * pool is available for import in read-only mode. If the pool 2483 * is available for import in read-write mode, it is displayed 2484 * as available in userland; if it is not available for import 2485 * in read-only mode, it is displayed as unavailable in 2486 * userland. If the pool is available for import in read-only 2487 * mode but not read-write mode, it is displayed as unavailable 2488 * in userland with a special note that the pool is actually 2489 * available for open in read-only mode. 2490 * 2491 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2492 * missing a feature for write, we must first determine whether 2493 * the pool can be opened read-only before returning to 2494 * userland in order to know whether to display the 2495 * abovementioned note. 2496 */ 2497 if (missing_feat_read || (missing_feat_write && 2498 spa_writeable(spa))) { 2499 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2500 ENOTSUP)); 2501 } 2502 2503 /* 2504 * Load refcounts for ZFS features from disk into an in-memory 2505 * cache during SPA initialization. 2506 */ 2507 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2508 uint64_t refcount; 2509 2510 error = feature_get_refcount_from_disk(spa, 2511 &spa_feature_table[i], &refcount); 2512 if (error == 0) { 2513 spa->spa_feat_refcount_cache[i] = refcount; 2514 } else if (error == ENOTSUP) { 2515 spa->spa_feat_refcount_cache[i] = 2516 SPA_FEATURE_DISABLED; 2517 } else { 2518 return (spa_vdev_err(rvd, 2519 VDEV_AUX_CORRUPT_DATA, EIO)); 2520 } 2521 } 2522 } 2523 2524 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2525 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2526 &spa->spa_feat_enabled_txg_obj) != 0) 2527 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2528 } 2529 2530 spa->spa_is_initializing = B_TRUE; 2531 error = dsl_pool_open(spa->spa_dsl_pool); 2532 spa->spa_is_initializing = B_FALSE; 2533 if (error != 0) 2534 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2535 2536 if (!mosconfig) { 2537 uint64_t hostid; 2538 nvlist_t *policy = NULL, *nvconfig; 2539 2540 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2541 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2542 2543 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2544 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2545 char *hostname; 2546 unsigned long myhostid = 0; 2547 2548 VERIFY(nvlist_lookup_string(nvconfig, 2549 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2550 2551#ifdef _KERNEL 2552 myhostid = zone_get_hostid(NULL); 2553#else /* _KERNEL */ 2554 /* 2555 * We're emulating the system's hostid in userland, so 2556 * we can't use zone_get_hostid(). 2557 */ 2558 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2559#endif /* _KERNEL */ 2560 if (check_hostid && hostid != 0 && myhostid != 0 && 2561 hostid != myhostid) { 2562 nvlist_free(nvconfig); 2563 cmn_err(CE_WARN, "pool '%s' could not be " 2564 "loaded as it was last accessed by " 2565 "another system (host: %s hostid: 0x%lx). " 2566 "See: http://illumos.org/msg/ZFS-8000-EY", 2567 spa_name(spa), hostname, 2568 (unsigned long)hostid); 2569 return (SET_ERROR(EBADF)); 2570 } 2571 } 2572 if (nvlist_lookup_nvlist(spa->spa_config, 2573 ZPOOL_REWIND_POLICY, &policy) == 0) 2574 VERIFY(nvlist_add_nvlist(nvconfig, 2575 ZPOOL_REWIND_POLICY, policy) == 0); 2576 2577 spa_config_set(spa, nvconfig); 2578 spa_unload(spa); 2579 spa_deactivate(spa); 2580 spa_activate(spa, orig_mode); 2581 2582 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2583 } 2584 2585 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2586 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2587 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2588 if (error != 0) 2589 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2590 2591 /* 2592 * Load the bit that tells us to use the new accounting function 2593 * (raid-z deflation). If we have an older pool, this will not 2594 * be present. 2595 */ 2596 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2597 if (error != 0 && error != ENOENT) 2598 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2599 2600 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2601 &spa->spa_creation_version); 2602 if (error != 0 && error != ENOENT) 2603 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2604 2605 /* 2606 * Load the persistent error log. If we have an older pool, this will 2607 * not be present. 2608 */ 2609 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2610 if (error != 0 && error != ENOENT) 2611 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2612 2613 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2614 &spa->spa_errlog_scrub); 2615 if (error != 0 && error != ENOENT) 2616 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2617 2618 /* 2619 * Load the history object. If we have an older pool, this 2620 * will not be present. 2621 */ 2622 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2623 if (error != 0 && error != ENOENT) 2624 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2625 2626 /* 2627 * If we're assembling the pool from the split-off vdevs of 2628 * an existing pool, we don't want to attach the spares & cache 2629 * devices. 2630 */ 2631 2632 /* 2633 * Load any hot spares for this pool. 2634 */ 2635 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2636 if (error != 0 && error != ENOENT) 2637 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2638 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2639 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2640 if (load_nvlist(spa, spa->spa_spares.sav_object, 2641 &spa->spa_spares.sav_config) != 0) 2642 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2643 2644 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2645 spa_load_spares(spa); 2646 spa_config_exit(spa, SCL_ALL, FTAG); 2647 } else if (error == 0) { 2648 spa->spa_spares.sav_sync = B_TRUE; 2649 } 2650 2651 /* 2652 * Load any level 2 ARC devices for this pool. 2653 */ 2654 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2655 &spa->spa_l2cache.sav_object); 2656 if (error != 0 && error != ENOENT) 2657 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2658 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2659 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2660 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2661 &spa->spa_l2cache.sav_config) != 0) 2662 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2663 2664 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2665 spa_load_l2cache(spa); 2666 spa_config_exit(spa, SCL_ALL, FTAG); 2667 } else if (error == 0) { 2668 spa->spa_l2cache.sav_sync = B_TRUE; 2669 } 2670 2671 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2672 2673 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2674 if (error && error != ENOENT) 2675 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2676 2677 if (error == 0) { 2678 uint64_t autoreplace; 2679 2680 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2681 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2682 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2683 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2684 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2685 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2686 &spa->spa_dedup_ditto); 2687 2688 spa->spa_autoreplace = (autoreplace != 0); 2689 } 2690 2691 /* 2692 * If the 'autoreplace' property is set, then post a resource notifying 2693 * the ZFS DE that it should not issue any faults for unopenable 2694 * devices. We also iterate over the vdevs, and post a sysevent for any 2695 * unopenable vdevs so that the normal autoreplace handler can take 2696 * over. 2697 */ 2698 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2699 spa_check_removed(spa->spa_root_vdev); 2700 /* 2701 * For the import case, this is done in spa_import(), because 2702 * at this point we're using the spare definitions from 2703 * the MOS config, not necessarily from the userland config. 2704 */ 2705 if (state != SPA_LOAD_IMPORT) { 2706 spa_aux_check_removed(&spa->spa_spares); 2707 spa_aux_check_removed(&spa->spa_l2cache); 2708 } 2709 } 2710 2711 /* 2712 * Load the vdev state for all toplevel vdevs. 2713 */ 2714 vdev_load(rvd); 2715 2716 /* 2717 * Propagate the leaf DTLs we just loaded all the way up the tree. 2718 */ 2719 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2720 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2721 spa_config_exit(spa, SCL_ALL, FTAG); 2722 2723 /* 2724 * Load the DDTs (dedup tables). 2725 */ 2726 error = ddt_load(spa); 2727 if (error != 0) 2728 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2729 2730 spa_update_dspace(spa); 2731 2732 /* 2733 * Validate the config, using the MOS config to fill in any 2734 * information which might be missing. If we fail to validate 2735 * the config then declare the pool unfit for use. If we're 2736 * assembling a pool from a split, the log is not transferred 2737 * over. 2738 */ 2739 if (type != SPA_IMPORT_ASSEMBLE) { 2740 nvlist_t *nvconfig; 2741 2742 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2743 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2744 2745 if (!spa_config_valid(spa, nvconfig)) { 2746 nvlist_free(nvconfig); 2747 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2748 ENXIO)); 2749 } 2750 nvlist_free(nvconfig); 2751 2752 /* 2753 * Now that we've validated the config, check the state of the 2754 * root vdev. If it can't be opened, it indicates one or 2755 * more toplevel vdevs are faulted. 2756 */ 2757 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2758 return (SET_ERROR(ENXIO)); 2759 2760 if (spa_writeable(spa) && spa_check_logs(spa)) { 2761 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2762 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2763 } 2764 } 2765 2766 if (missing_feat_write) { 2767 ASSERT(state == SPA_LOAD_TRYIMPORT); 2768 2769 /* 2770 * At this point, we know that we can open the pool in 2771 * read-only mode but not read-write mode. We now have enough 2772 * information and can return to userland. 2773 */ 2774 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2775 } 2776 2777 /* 2778 * We've successfully opened the pool, verify that we're ready 2779 * to start pushing transactions. 2780 */ 2781 if (state != SPA_LOAD_TRYIMPORT) { 2782 if (error = spa_load_verify(spa)) 2783 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2784 error)); 2785 } 2786 2787 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2788 spa->spa_load_max_txg == UINT64_MAX)) { 2789 dmu_tx_t *tx; 2790 int need_update = B_FALSE; 2791 dsl_pool_t *dp = spa_get_dsl(spa); 2792 2793 ASSERT(state != SPA_LOAD_TRYIMPORT); 2794 2795 /* 2796 * Claim log blocks that haven't been committed yet. 2797 * This must all happen in a single txg. 2798 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2799 * invoked from zil_claim_log_block()'s i/o done callback. 2800 * Price of rollback is that we abandon the log. 2801 */ 2802 spa->spa_claiming = B_TRUE; 2803 2804 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2805 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2806 zil_claim, tx, DS_FIND_CHILDREN); 2807 dmu_tx_commit(tx); 2808 2809 spa->spa_claiming = B_FALSE; 2810 2811 spa_set_log_state(spa, SPA_LOG_GOOD); 2812 spa->spa_sync_on = B_TRUE; 2813 txg_sync_start(spa->spa_dsl_pool); 2814 2815 /* 2816 * Wait for all claims to sync. We sync up to the highest 2817 * claimed log block birth time so that claimed log blocks 2818 * don't appear to be from the future. spa_claim_max_txg 2819 * will have been set for us by either zil_check_log_chain() 2820 * (invoked from spa_check_logs()) or zil_claim() above. 2821 */ 2822 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2823 2824 /* 2825 * If the config cache is stale, or we have uninitialized 2826 * metaslabs (see spa_vdev_add()), then update the config. 2827 * 2828 * If this is a verbatim import, trust the current 2829 * in-core spa_config and update the disk labels. 2830 */ 2831 if (config_cache_txg != spa->spa_config_txg || 2832 state == SPA_LOAD_IMPORT || 2833 state == SPA_LOAD_RECOVER || 2834 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2835 need_update = B_TRUE; 2836 2837 for (int c = 0; c < rvd->vdev_children; c++) 2838 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2839 need_update = B_TRUE; 2840 2841 /* 2842 * Update the config cache asychronously in case we're the 2843 * root pool, in which case the config cache isn't writable yet. 2844 */ 2845 if (need_update) 2846 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2847 2848 /* 2849 * Check all DTLs to see if anything needs resilvering. 2850 */ 2851 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2852 vdev_resilver_needed(rvd, NULL, NULL)) 2853 spa_async_request(spa, SPA_ASYNC_RESILVER); 2854 2855 /* 2856 * Log the fact that we booted up (so that we can detect if 2857 * we rebooted in the middle of an operation). 2858 */ 2859 spa_history_log_version(spa, "open"); 2860 2861 /* 2862 * Delete any inconsistent datasets. 2863 */ 2864 (void) dmu_objset_find(spa_name(spa), 2865 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2866 2867 /* 2868 * Clean up any stale temporary dataset userrefs. 2869 */ 2870 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2871 } 2872 2873 return (0); 2874} 2875 2876static int 2877spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2878{ 2879 int mode = spa->spa_mode; 2880 2881 spa_unload(spa); 2882 spa_deactivate(spa); 2883 2884 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2885 2886 spa_activate(spa, mode); 2887 spa_async_suspend(spa); 2888 2889 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2890} 2891 2892/* 2893 * If spa_load() fails this function will try loading prior txg's. If 2894 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2895 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2896 * function will not rewind the pool and will return the same error as 2897 * spa_load(). 2898 */ 2899static int 2900spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2901 uint64_t max_request, int rewind_flags) 2902{ 2903 nvlist_t *loadinfo = NULL; 2904 nvlist_t *config = NULL; 2905 int load_error, rewind_error; 2906 uint64_t safe_rewind_txg; 2907 uint64_t min_txg; 2908 2909 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2910 spa->spa_load_max_txg = spa->spa_load_txg; 2911 spa_set_log_state(spa, SPA_LOG_CLEAR); 2912 } else { 2913 spa->spa_load_max_txg = max_request; 2914 if (max_request != UINT64_MAX) 2915 spa->spa_extreme_rewind = B_TRUE; 2916 } 2917 2918 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2919 mosconfig); 2920 if (load_error == 0) 2921 return (0); 2922 2923 if (spa->spa_root_vdev != NULL) 2924 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2925 2926 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2927 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2928 2929 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2930 nvlist_free(config); 2931 return (load_error); 2932 } 2933 2934 if (state == SPA_LOAD_RECOVER) { 2935 /* Price of rolling back is discarding txgs, including log */ 2936 spa_set_log_state(spa, SPA_LOG_CLEAR); 2937 } else { 2938 /* 2939 * If we aren't rolling back save the load info from our first 2940 * import attempt so that we can restore it after attempting 2941 * to rewind. 2942 */ 2943 loadinfo = spa->spa_load_info; 2944 spa->spa_load_info = fnvlist_alloc(); 2945 } 2946 2947 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2948 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2949 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2950 TXG_INITIAL : safe_rewind_txg; 2951 2952 /* 2953 * Continue as long as we're finding errors, we're still within 2954 * the acceptable rewind range, and we're still finding uberblocks 2955 */ 2956 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2957 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2958 if (spa->spa_load_max_txg < safe_rewind_txg) 2959 spa->spa_extreme_rewind = B_TRUE; 2960 rewind_error = spa_load_retry(spa, state, mosconfig); 2961 } 2962 2963 spa->spa_extreme_rewind = B_FALSE; 2964 spa->spa_load_max_txg = UINT64_MAX; 2965 2966 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2967 spa_config_set(spa, config); 2968 2969 if (state == SPA_LOAD_RECOVER) { 2970 ASSERT3P(loadinfo, ==, NULL); 2971 return (rewind_error); 2972 } else { 2973 /* Store the rewind info as part of the initial load info */ 2974 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2975 spa->spa_load_info); 2976 2977 /* Restore the initial load info */ 2978 fnvlist_free(spa->spa_load_info); 2979 spa->spa_load_info = loadinfo; 2980 2981 return (load_error); 2982 } 2983} 2984 2985/* 2986 * Pool Open/Import 2987 * 2988 * The import case is identical to an open except that the configuration is sent 2989 * down from userland, instead of grabbed from the configuration cache. For the 2990 * case of an open, the pool configuration will exist in the 2991 * POOL_STATE_UNINITIALIZED state. 2992 * 2993 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2994 * the same time open the pool, without having to keep around the spa_t in some 2995 * ambiguous state. 2996 */ 2997static int 2998spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2999 nvlist_t **config) 3000{ 3001 spa_t *spa; 3002 spa_load_state_t state = SPA_LOAD_OPEN; 3003 int error; 3004 int locked = B_FALSE; 3005 int firstopen = B_FALSE; 3006 3007 *spapp = NULL; 3008 3009 /* 3010 * As disgusting as this is, we need to support recursive calls to this 3011 * function because dsl_dir_open() is called during spa_load(), and ends 3012 * up calling spa_open() again. The real fix is to figure out how to 3013 * avoid dsl_dir_open() calling this in the first place. 3014 */ 3015 if (mutex_owner(&spa_namespace_lock) != curthread) { 3016 mutex_enter(&spa_namespace_lock); 3017 locked = B_TRUE; 3018 } 3019 3020 if ((spa = spa_lookup(pool)) == NULL) { 3021 if (locked) 3022 mutex_exit(&spa_namespace_lock); 3023 return (SET_ERROR(ENOENT)); 3024 } 3025 3026 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3027 zpool_rewind_policy_t policy; 3028 3029 firstopen = B_TRUE; 3030 3031 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3032 &policy); 3033 if (policy.zrp_request & ZPOOL_DO_REWIND) 3034 state = SPA_LOAD_RECOVER; 3035 3036 spa_activate(spa, spa_mode_global); 3037 3038 if (state != SPA_LOAD_RECOVER) 3039 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3040 3041 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3042 policy.zrp_request); 3043 3044 if (error == EBADF) { 3045 /* 3046 * If vdev_validate() returns failure (indicated by 3047 * EBADF), it indicates that one of the vdevs indicates 3048 * that the pool has been exported or destroyed. If 3049 * this is the case, the config cache is out of sync and 3050 * we should remove the pool from the namespace. 3051 */ 3052 spa_unload(spa); 3053 spa_deactivate(spa); 3054 spa_config_sync(spa, B_TRUE, B_TRUE); 3055 spa_remove(spa); 3056 if (locked) 3057 mutex_exit(&spa_namespace_lock); 3058 return (SET_ERROR(ENOENT)); 3059 } 3060 3061 if (error) { 3062 /* 3063 * We can't open the pool, but we still have useful 3064 * information: the state of each vdev after the 3065 * attempted vdev_open(). Return this to the user. 3066 */ 3067 if (config != NULL && spa->spa_config) { 3068 VERIFY(nvlist_dup(spa->spa_config, config, 3069 KM_SLEEP) == 0); 3070 VERIFY(nvlist_add_nvlist(*config, 3071 ZPOOL_CONFIG_LOAD_INFO, 3072 spa->spa_load_info) == 0); 3073 } 3074 spa_unload(spa); 3075 spa_deactivate(spa); 3076 spa->spa_last_open_failed = error; 3077 if (locked) 3078 mutex_exit(&spa_namespace_lock); 3079 *spapp = NULL; 3080 return (error); 3081 } 3082 } 3083 3084 spa_open_ref(spa, tag); 3085 3086 if (config != NULL) 3087 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3088 3089 /* 3090 * If we've recovered the pool, pass back any information we 3091 * gathered while doing the load. 3092 */ 3093 if (state == SPA_LOAD_RECOVER) { 3094 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3095 spa->spa_load_info) == 0); 3096 } 3097 3098 if (locked) { 3099 spa->spa_last_open_failed = 0; 3100 spa->spa_last_ubsync_txg = 0; 3101 spa->spa_load_txg = 0; 3102 mutex_exit(&spa_namespace_lock); 3103#ifdef __FreeBSD__ 3104#ifdef _KERNEL 3105 if (firstopen) 3106 zvol_create_minors(spa->spa_name); 3107#endif 3108#endif 3109 } 3110 3111 *spapp = spa; 3112 3113 return (0); 3114} 3115 3116int 3117spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3118 nvlist_t **config) 3119{ 3120 return (spa_open_common(name, spapp, tag, policy, config)); 3121} 3122 3123int 3124spa_open(const char *name, spa_t **spapp, void *tag) 3125{ 3126 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3127} 3128 3129/* 3130 * Lookup the given spa_t, incrementing the inject count in the process, 3131 * preventing it from being exported or destroyed. 3132 */ 3133spa_t * 3134spa_inject_addref(char *name) 3135{ 3136 spa_t *spa; 3137 3138 mutex_enter(&spa_namespace_lock); 3139 if ((spa = spa_lookup(name)) == NULL) { 3140 mutex_exit(&spa_namespace_lock); 3141 return (NULL); 3142 } 3143 spa->spa_inject_ref++; 3144 mutex_exit(&spa_namespace_lock); 3145 3146 return (spa); 3147} 3148 3149void 3150spa_inject_delref(spa_t *spa) 3151{ 3152 mutex_enter(&spa_namespace_lock); 3153 spa->spa_inject_ref--; 3154 mutex_exit(&spa_namespace_lock); 3155} 3156 3157/* 3158 * Add spares device information to the nvlist. 3159 */ 3160static void 3161spa_add_spares(spa_t *spa, nvlist_t *config) 3162{ 3163 nvlist_t **spares; 3164 uint_t i, nspares; 3165 nvlist_t *nvroot; 3166 uint64_t guid; 3167 vdev_stat_t *vs; 3168 uint_t vsc; 3169 uint64_t pool; 3170 3171 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3172 3173 if (spa->spa_spares.sav_count == 0) 3174 return; 3175 3176 VERIFY(nvlist_lookup_nvlist(config, 3177 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3178 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3179 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3180 if (nspares != 0) { 3181 VERIFY(nvlist_add_nvlist_array(nvroot, 3182 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3183 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3184 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3185 3186 /* 3187 * Go through and find any spares which have since been 3188 * repurposed as an active spare. If this is the case, update 3189 * their status appropriately. 3190 */ 3191 for (i = 0; i < nspares; i++) { 3192 VERIFY(nvlist_lookup_uint64(spares[i], 3193 ZPOOL_CONFIG_GUID, &guid) == 0); 3194 if (spa_spare_exists(guid, &pool, NULL) && 3195 pool != 0ULL) { 3196 VERIFY(nvlist_lookup_uint64_array( 3197 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3198 (uint64_t **)&vs, &vsc) == 0); 3199 vs->vs_state = VDEV_STATE_CANT_OPEN; 3200 vs->vs_aux = VDEV_AUX_SPARED; 3201 } 3202 } 3203 } 3204} 3205 3206/* 3207 * Add l2cache device information to the nvlist, including vdev stats. 3208 */ 3209static void 3210spa_add_l2cache(spa_t *spa, nvlist_t *config) 3211{ 3212 nvlist_t **l2cache; 3213 uint_t i, j, nl2cache; 3214 nvlist_t *nvroot; 3215 uint64_t guid; 3216 vdev_t *vd; 3217 vdev_stat_t *vs; 3218 uint_t vsc; 3219 3220 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3221 3222 if (spa->spa_l2cache.sav_count == 0) 3223 return; 3224 3225 VERIFY(nvlist_lookup_nvlist(config, 3226 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3227 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3228 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3229 if (nl2cache != 0) { 3230 VERIFY(nvlist_add_nvlist_array(nvroot, 3231 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3232 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3233 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3234 3235 /* 3236 * Update level 2 cache device stats. 3237 */ 3238 3239 for (i = 0; i < nl2cache; i++) { 3240 VERIFY(nvlist_lookup_uint64(l2cache[i], 3241 ZPOOL_CONFIG_GUID, &guid) == 0); 3242 3243 vd = NULL; 3244 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3245 if (guid == 3246 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3247 vd = spa->spa_l2cache.sav_vdevs[j]; 3248 break; 3249 } 3250 } 3251 ASSERT(vd != NULL); 3252 3253 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3254 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3255 == 0); 3256 vdev_get_stats(vd, vs); 3257 } 3258 } 3259} 3260 3261static void 3262spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3263{ 3264 nvlist_t *features; 3265 zap_cursor_t zc; 3266 zap_attribute_t za; 3267 3268 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3269 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3270 3271 /* We may be unable to read features if pool is suspended. */ 3272 if (spa_suspended(spa)) 3273 goto out; 3274 3275 if (spa->spa_feat_for_read_obj != 0) { 3276 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3277 spa->spa_feat_for_read_obj); 3278 zap_cursor_retrieve(&zc, &za) == 0; 3279 zap_cursor_advance(&zc)) { 3280 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3281 za.za_num_integers == 1); 3282 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3283 za.za_first_integer)); 3284 } 3285 zap_cursor_fini(&zc); 3286 } 3287 3288 if (spa->spa_feat_for_write_obj != 0) { 3289 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3290 spa->spa_feat_for_write_obj); 3291 zap_cursor_retrieve(&zc, &za) == 0; 3292 zap_cursor_advance(&zc)) { 3293 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3294 za.za_num_integers == 1); 3295 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3296 za.za_first_integer)); 3297 } 3298 zap_cursor_fini(&zc); 3299 } 3300 3301out: 3302 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3303 features) == 0); 3304 nvlist_free(features); 3305} 3306 3307int 3308spa_get_stats(const char *name, nvlist_t **config, 3309 char *altroot, size_t buflen) 3310{ 3311 int error; 3312 spa_t *spa; 3313 3314 *config = NULL; 3315 error = spa_open_common(name, &spa, FTAG, NULL, config); 3316 3317 if (spa != NULL) { 3318 /* 3319 * This still leaves a window of inconsistency where the spares 3320 * or l2cache devices could change and the config would be 3321 * self-inconsistent. 3322 */ 3323 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3324 3325 if (*config != NULL) { 3326 uint64_t loadtimes[2]; 3327 3328 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3329 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3330 VERIFY(nvlist_add_uint64_array(*config, 3331 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3332 3333 VERIFY(nvlist_add_uint64(*config, 3334 ZPOOL_CONFIG_ERRCOUNT, 3335 spa_get_errlog_size(spa)) == 0); 3336 3337 if (spa_suspended(spa)) 3338 VERIFY(nvlist_add_uint64(*config, 3339 ZPOOL_CONFIG_SUSPENDED, 3340 spa->spa_failmode) == 0); 3341 3342 spa_add_spares(spa, *config); 3343 spa_add_l2cache(spa, *config); 3344 spa_add_feature_stats(spa, *config); 3345 } 3346 } 3347 3348 /* 3349 * We want to get the alternate root even for faulted pools, so we cheat 3350 * and call spa_lookup() directly. 3351 */ 3352 if (altroot) { 3353 if (spa == NULL) { 3354 mutex_enter(&spa_namespace_lock); 3355 spa = spa_lookup(name); 3356 if (spa) 3357 spa_altroot(spa, altroot, buflen); 3358 else 3359 altroot[0] = '\0'; 3360 spa = NULL; 3361 mutex_exit(&spa_namespace_lock); 3362 } else { 3363 spa_altroot(spa, altroot, buflen); 3364 } 3365 } 3366 3367 if (spa != NULL) { 3368 spa_config_exit(spa, SCL_CONFIG, FTAG); 3369 spa_close(spa, FTAG); 3370 } 3371 3372 return (error); 3373} 3374 3375/* 3376 * Validate that the auxiliary device array is well formed. We must have an 3377 * array of nvlists, each which describes a valid leaf vdev. If this is an 3378 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3379 * specified, as long as they are well-formed. 3380 */ 3381static int 3382spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3383 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3384 vdev_labeltype_t label) 3385{ 3386 nvlist_t **dev; 3387 uint_t i, ndev; 3388 vdev_t *vd; 3389 int error; 3390 3391 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3392 3393 /* 3394 * It's acceptable to have no devs specified. 3395 */ 3396 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3397 return (0); 3398 3399 if (ndev == 0) 3400 return (SET_ERROR(EINVAL)); 3401 3402 /* 3403 * Make sure the pool is formatted with a version that supports this 3404 * device type. 3405 */ 3406 if (spa_version(spa) < version) 3407 return (SET_ERROR(ENOTSUP)); 3408 3409 /* 3410 * Set the pending device list so we correctly handle device in-use 3411 * checking. 3412 */ 3413 sav->sav_pending = dev; 3414 sav->sav_npending = ndev; 3415 3416 for (i = 0; i < ndev; i++) { 3417 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3418 mode)) != 0) 3419 goto out; 3420 3421 if (!vd->vdev_ops->vdev_op_leaf) { 3422 vdev_free(vd); 3423 error = SET_ERROR(EINVAL); 3424 goto out; 3425 } 3426 3427 /* 3428 * The L2ARC currently only supports disk devices in 3429 * kernel context. For user-level testing, we allow it. 3430 */ 3431#ifdef _KERNEL 3432 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3433 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3434 error = SET_ERROR(ENOTBLK); 3435 vdev_free(vd); 3436 goto out; 3437 } 3438#endif 3439 vd->vdev_top = vd; 3440 3441 if ((error = vdev_open(vd)) == 0 && 3442 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3443 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3444 vd->vdev_guid) == 0); 3445 } 3446 3447 vdev_free(vd); 3448 3449 if (error && 3450 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3451 goto out; 3452 else 3453 error = 0; 3454 } 3455 3456out: 3457 sav->sav_pending = NULL; 3458 sav->sav_npending = 0; 3459 return (error); 3460} 3461 3462static int 3463spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3464{ 3465 int error; 3466 3467 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3468 3469 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3470 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3471 VDEV_LABEL_SPARE)) != 0) { 3472 return (error); 3473 } 3474 3475 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3476 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3477 VDEV_LABEL_L2CACHE)); 3478} 3479 3480static void 3481spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3482 const char *config) 3483{ 3484 int i; 3485 3486 if (sav->sav_config != NULL) { 3487 nvlist_t **olddevs; 3488 uint_t oldndevs; 3489 nvlist_t **newdevs; 3490 3491 /* 3492 * Generate new dev list by concatentating with the 3493 * current dev list. 3494 */ 3495 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3496 &olddevs, &oldndevs) == 0); 3497 3498 newdevs = kmem_alloc(sizeof (void *) * 3499 (ndevs + oldndevs), KM_SLEEP); 3500 for (i = 0; i < oldndevs; i++) 3501 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3502 KM_SLEEP) == 0); 3503 for (i = 0; i < ndevs; i++) 3504 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3505 KM_SLEEP) == 0); 3506 3507 VERIFY(nvlist_remove(sav->sav_config, config, 3508 DATA_TYPE_NVLIST_ARRAY) == 0); 3509 3510 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3511 config, newdevs, ndevs + oldndevs) == 0); 3512 for (i = 0; i < oldndevs + ndevs; i++) 3513 nvlist_free(newdevs[i]); 3514 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3515 } else { 3516 /* 3517 * Generate a new dev list. 3518 */ 3519 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3520 KM_SLEEP) == 0); 3521 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3522 devs, ndevs) == 0); 3523 } 3524} 3525 3526/* 3527 * Stop and drop level 2 ARC devices 3528 */ 3529void 3530spa_l2cache_drop(spa_t *spa) 3531{ 3532 vdev_t *vd; 3533 int i; 3534 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3535 3536 for (i = 0; i < sav->sav_count; i++) { 3537 uint64_t pool; 3538 3539 vd = sav->sav_vdevs[i]; 3540 ASSERT(vd != NULL); 3541 3542 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3543 pool != 0ULL && l2arc_vdev_present(vd)) 3544 l2arc_remove_vdev(vd); 3545 } 3546} 3547 3548/* 3549 * Pool Creation 3550 */ 3551int 3552spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3553 nvlist_t *zplprops) 3554{ 3555 spa_t *spa; 3556 char *altroot = NULL; 3557 vdev_t *rvd; 3558 dsl_pool_t *dp; 3559 dmu_tx_t *tx; 3560 int error = 0; 3561 uint64_t txg = TXG_INITIAL; 3562 nvlist_t **spares, **l2cache; 3563 uint_t nspares, nl2cache; 3564 uint64_t version, obj; 3565 boolean_t has_features; 3566 3567 /* 3568 * If this pool already exists, return failure. 3569 */ 3570 mutex_enter(&spa_namespace_lock); 3571 if (spa_lookup(pool) != NULL) { 3572 mutex_exit(&spa_namespace_lock); 3573 return (SET_ERROR(EEXIST)); 3574 } 3575 3576 /* 3577 * Allocate a new spa_t structure. 3578 */ 3579 (void) nvlist_lookup_string(props, 3580 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3581 spa = spa_add(pool, NULL, altroot); 3582 spa_activate(spa, spa_mode_global); 3583 3584 if (props && (error = spa_prop_validate(spa, props))) { 3585 spa_deactivate(spa); 3586 spa_remove(spa); 3587 mutex_exit(&spa_namespace_lock); 3588 return (error); 3589 } 3590 3591 has_features = B_FALSE; 3592 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3593 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3594 if (zpool_prop_feature(nvpair_name(elem))) 3595 has_features = B_TRUE; 3596 } 3597 3598 if (has_features || nvlist_lookup_uint64(props, 3599 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3600 version = SPA_VERSION; 3601 } 3602 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3603 3604 spa->spa_first_txg = txg; 3605 spa->spa_uberblock.ub_txg = txg - 1; 3606 spa->spa_uberblock.ub_version = version; 3607 spa->spa_ubsync = spa->spa_uberblock; 3608 3609 /* 3610 * Create "The Godfather" zio to hold all async IOs 3611 */ 3612 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3613 KM_SLEEP); 3614 for (int i = 0; i < max_ncpus; i++) { 3615 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3616 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3617 ZIO_FLAG_GODFATHER); 3618 } 3619 3620 /* 3621 * Create the root vdev. 3622 */ 3623 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3624 3625 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3626 3627 ASSERT(error != 0 || rvd != NULL); 3628 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3629 3630 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3631 error = SET_ERROR(EINVAL); 3632 3633 if (error == 0 && 3634 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3635 (error = spa_validate_aux(spa, nvroot, txg, 3636 VDEV_ALLOC_ADD)) == 0) { 3637 for (int c = 0; c < rvd->vdev_children; c++) { 3638 vdev_ashift_optimize(rvd->vdev_child[c]); 3639 vdev_metaslab_set_size(rvd->vdev_child[c]); 3640 vdev_expand(rvd->vdev_child[c], txg); 3641 } 3642 } 3643 3644 spa_config_exit(spa, SCL_ALL, FTAG); 3645 3646 if (error != 0) { 3647 spa_unload(spa); 3648 spa_deactivate(spa); 3649 spa_remove(spa); 3650 mutex_exit(&spa_namespace_lock); 3651 return (error); 3652 } 3653 3654 /* 3655 * Get the list of spares, if specified. 3656 */ 3657 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3658 &spares, &nspares) == 0) { 3659 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3660 KM_SLEEP) == 0); 3661 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3662 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3663 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3664 spa_load_spares(spa); 3665 spa_config_exit(spa, SCL_ALL, FTAG); 3666 spa->spa_spares.sav_sync = B_TRUE; 3667 } 3668 3669 /* 3670 * Get the list of level 2 cache devices, if specified. 3671 */ 3672 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3673 &l2cache, &nl2cache) == 0) { 3674 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3675 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3676 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3677 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3678 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3679 spa_load_l2cache(spa); 3680 spa_config_exit(spa, SCL_ALL, FTAG); 3681 spa->spa_l2cache.sav_sync = B_TRUE; 3682 } 3683 3684 spa->spa_is_initializing = B_TRUE; 3685 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3686 spa->spa_meta_objset = dp->dp_meta_objset; 3687 spa->spa_is_initializing = B_FALSE; 3688 3689 /* 3690 * Create DDTs (dedup tables). 3691 */ 3692 ddt_create(spa); 3693 3694 spa_update_dspace(spa); 3695 3696 tx = dmu_tx_create_assigned(dp, txg); 3697 3698 /* 3699 * Create the pool config object. 3700 */ 3701 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3702 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3703 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3704 3705 if (zap_add(spa->spa_meta_objset, 3706 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3707 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3708 cmn_err(CE_PANIC, "failed to add pool config"); 3709 } 3710 3711 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3712 spa_feature_create_zap_objects(spa, tx); 3713 3714 if (zap_add(spa->spa_meta_objset, 3715 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3716 sizeof (uint64_t), 1, &version, tx) != 0) { 3717 cmn_err(CE_PANIC, "failed to add pool version"); 3718 } 3719 3720 /* Newly created pools with the right version are always deflated. */ 3721 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3722 spa->spa_deflate = TRUE; 3723 if (zap_add(spa->spa_meta_objset, 3724 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3725 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3726 cmn_err(CE_PANIC, "failed to add deflate"); 3727 } 3728 } 3729 3730 /* 3731 * Create the deferred-free bpobj. Turn off compression 3732 * because sync-to-convergence takes longer if the blocksize 3733 * keeps changing. 3734 */ 3735 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3736 dmu_object_set_compress(spa->spa_meta_objset, obj, 3737 ZIO_COMPRESS_OFF, tx); 3738 if (zap_add(spa->spa_meta_objset, 3739 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3740 sizeof (uint64_t), 1, &obj, tx) != 0) { 3741 cmn_err(CE_PANIC, "failed to add bpobj"); 3742 } 3743 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3744 spa->spa_meta_objset, obj)); 3745 3746 /* 3747 * Create the pool's history object. 3748 */ 3749 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3750 spa_history_create_obj(spa, tx); 3751 3752 /* 3753 * Set pool properties. 3754 */ 3755 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3756 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3757 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3758 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3759 3760 if (props != NULL) { 3761 spa_configfile_set(spa, props, B_FALSE); 3762 spa_sync_props(props, tx); 3763 } 3764 3765 dmu_tx_commit(tx); 3766 3767 spa->spa_sync_on = B_TRUE; 3768 txg_sync_start(spa->spa_dsl_pool); 3769 3770 /* 3771 * We explicitly wait for the first transaction to complete so that our 3772 * bean counters are appropriately updated. 3773 */ 3774 txg_wait_synced(spa->spa_dsl_pool, txg); 3775 3776 spa_config_sync(spa, B_FALSE, B_TRUE); 3777 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3778 3779 spa_history_log_version(spa, "create"); 3780 3781 /* 3782 * Don't count references from objsets that are already closed 3783 * and are making their way through the eviction process. 3784 */ 3785 spa_evicting_os_wait(spa); 3786 spa->spa_minref = refcount_count(&spa->spa_refcount); 3787 3788 mutex_exit(&spa_namespace_lock); 3789 3790 return (0); 3791} 3792 3793#ifdef _KERNEL 3794#ifdef illumos 3795/* 3796 * Get the root pool information from the root disk, then import the root pool 3797 * during the system boot up time. 3798 */ 3799extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3800 3801static nvlist_t * 3802spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3803{ 3804 nvlist_t *config; 3805 nvlist_t *nvtop, *nvroot; 3806 uint64_t pgid; 3807 3808 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3809 return (NULL); 3810 3811 /* 3812 * Add this top-level vdev to the child array. 3813 */ 3814 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3815 &nvtop) == 0); 3816 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3817 &pgid) == 0); 3818 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3819 3820 /* 3821 * Put this pool's top-level vdevs into a root vdev. 3822 */ 3823 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3824 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3825 VDEV_TYPE_ROOT) == 0); 3826 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3827 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3828 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3829 &nvtop, 1) == 0); 3830 3831 /* 3832 * Replace the existing vdev_tree with the new root vdev in 3833 * this pool's configuration (remove the old, add the new). 3834 */ 3835 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3836 nvlist_free(nvroot); 3837 return (config); 3838} 3839 3840/* 3841 * Walk the vdev tree and see if we can find a device with "better" 3842 * configuration. A configuration is "better" if the label on that 3843 * device has a more recent txg. 3844 */ 3845static void 3846spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3847{ 3848 for (int c = 0; c < vd->vdev_children; c++) 3849 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3850 3851 if (vd->vdev_ops->vdev_op_leaf) { 3852 nvlist_t *label; 3853 uint64_t label_txg; 3854 3855 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3856 &label) != 0) 3857 return; 3858 3859 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3860 &label_txg) == 0); 3861 3862 /* 3863 * Do we have a better boot device? 3864 */ 3865 if (label_txg > *txg) { 3866 *txg = label_txg; 3867 *avd = vd; 3868 } 3869 nvlist_free(label); 3870 } 3871} 3872 3873/* 3874 * Import a root pool. 3875 * 3876 * For x86. devpath_list will consist of devid and/or physpath name of 3877 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3878 * The GRUB "findroot" command will return the vdev we should boot. 3879 * 3880 * For Sparc, devpath_list consists the physpath name of the booting device 3881 * no matter the rootpool is a single device pool or a mirrored pool. 3882 * e.g. 3883 * "/pci@1f,0/ide@d/disk@0,0:a" 3884 */ 3885int 3886spa_import_rootpool(char *devpath, char *devid) 3887{ 3888 spa_t *spa; 3889 vdev_t *rvd, *bvd, *avd = NULL; 3890 nvlist_t *config, *nvtop; 3891 uint64_t guid, txg; 3892 char *pname; 3893 int error; 3894 3895 /* 3896 * Read the label from the boot device and generate a configuration. 3897 */ 3898 config = spa_generate_rootconf(devpath, devid, &guid); 3899#if defined(_OBP) && defined(_KERNEL) 3900 if (config == NULL) { 3901 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3902 /* iscsi boot */ 3903 get_iscsi_bootpath_phy(devpath); 3904 config = spa_generate_rootconf(devpath, devid, &guid); 3905 } 3906 } 3907#endif 3908 if (config == NULL) { 3909 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3910 devpath); 3911 return (SET_ERROR(EIO)); 3912 } 3913 3914 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3915 &pname) == 0); 3916 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3917 3918 mutex_enter(&spa_namespace_lock); 3919 if ((spa = spa_lookup(pname)) != NULL) { 3920 /* 3921 * Remove the existing root pool from the namespace so that we 3922 * can replace it with the correct config we just read in. 3923 */ 3924 spa_remove(spa); 3925 } 3926 3927 spa = spa_add(pname, config, NULL); 3928 spa->spa_is_root = B_TRUE; 3929 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3930 3931 /* 3932 * Build up a vdev tree based on the boot device's label config. 3933 */ 3934 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3935 &nvtop) == 0); 3936 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3937 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3938 VDEV_ALLOC_ROOTPOOL); 3939 spa_config_exit(spa, SCL_ALL, FTAG); 3940 if (error) { 3941 mutex_exit(&spa_namespace_lock); 3942 nvlist_free(config); 3943 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3944 pname); 3945 return (error); 3946 } 3947 3948 /* 3949 * Get the boot vdev. 3950 */ 3951 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3952 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3953 (u_longlong_t)guid); 3954 error = SET_ERROR(ENOENT); 3955 goto out; 3956 } 3957 3958 /* 3959 * Determine if there is a better boot device. 3960 */ 3961 avd = bvd; 3962 spa_alt_rootvdev(rvd, &avd, &txg); 3963 if (avd != bvd) { 3964 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3965 "try booting from '%s'", avd->vdev_path); 3966 error = SET_ERROR(EINVAL); 3967 goto out; 3968 } 3969 3970 /* 3971 * If the boot device is part of a spare vdev then ensure that 3972 * we're booting off the active spare. 3973 */ 3974 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3975 !bvd->vdev_isspare) { 3976 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3977 "try booting from '%s'", 3978 bvd->vdev_parent-> 3979 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3980 error = SET_ERROR(EINVAL); 3981 goto out; 3982 } 3983 3984 error = 0; 3985out: 3986 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3987 vdev_free(rvd); 3988 spa_config_exit(spa, SCL_ALL, FTAG); 3989 mutex_exit(&spa_namespace_lock); 3990 3991 nvlist_free(config); 3992 return (error); 3993} 3994 3995#else /* !illumos */ 3996 3997extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3998 uint64_t *count); 3999 4000static nvlist_t * 4001spa_generate_rootconf(const char *name) 4002{ 4003 nvlist_t **configs, **tops; 4004 nvlist_t *config; 4005 nvlist_t *best_cfg, *nvtop, *nvroot; 4006 uint64_t *holes; 4007 uint64_t best_txg; 4008 uint64_t nchildren; 4009 uint64_t pgid; 4010 uint64_t count; 4011 uint64_t i; 4012 uint_t nholes; 4013 4014 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4015 return (NULL); 4016 4017 ASSERT3U(count, !=, 0); 4018 best_txg = 0; 4019 for (i = 0; i < count; i++) { 4020 uint64_t txg; 4021 4022 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4023 &txg) == 0); 4024 if (txg > best_txg) { 4025 best_txg = txg; 4026 best_cfg = configs[i]; 4027 } 4028 } 4029 4030 nchildren = 1; 4031 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4032 holes = NULL; 4033 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4034 &holes, &nholes); 4035 4036 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4037 for (i = 0; i < nchildren; i++) { 4038 if (i >= count) 4039 break; 4040 if (configs[i] == NULL) 4041 continue; 4042 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4043 &nvtop) == 0); 4044 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4045 } 4046 for (i = 0; holes != NULL && i < nholes; i++) { 4047 if (i >= nchildren) 4048 continue; 4049 if (tops[holes[i]] != NULL) 4050 continue; 4051 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4052 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4053 VDEV_TYPE_HOLE) == 0); 4054 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4055 holes[i]) == 0); 4056 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4057 0) == 0); 4058 } 4059 for (i = 0; i < nchildren; i++) { 4060 if (tops[i] != NULL) 4061 continue; 4062 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4063 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4064 VDEV_TYPE_MISSING) == 0); 4065 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4066 i) == 0); 4067 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4068 0) == 0); 4069 } 4070 4071 /* 4072 * Create pool config based on the best vdev config. 4073 */ 4074 nvlist_dup(best_cfg, &config, KM_SLEEP); 4075 4076 /* 4077 * Put this pool's top-level vdevs into a root vdev. 4078 */ 4079 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4080 &pgid) == 0); 4081 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4082 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4083 VDEV_TYPE_ROOT) == 0); 4084 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4085 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4086 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4087 tops, nchildren) == 0); 4088 4089 /* 4090 * Replace the existing vdev_tree with the new root vdev in 4091 * this pool's configuration (remove the old, add the new). 4092 */ 4093 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4094 4095 /* 4096 * Drop vdev config elements that should not be present at pool level. 4097 */ 4098 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4099 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4100 4101 for (i = 0; i < count; i++) 4102 nvlist_free(configs[i]); 4103 kmem_free(configs, count * sizeof(void *)); 4104 for (i = 0; i < nchildren; i++) 4105 nvlist_free(tops[i]); 4106 kmem_free(tops, nchildren * sizeof(void *)); 4107 nvlist_free(nvroot); 4108 return (config); 4109} 4110 4111int 4112spa_import_rootpool(const char *name) 4113{ 4114 spa_t *spa; 4115 vdev_t *rvd, *bvd, *avd = NULL; 4116 nvlist_t *config, *nvtop; 4117 uint64_t txg; 4118 char *pname; 4119 int error; 4120 4121 /* 4122 * Read the label from the boot device and generate a configuration. 4123 */ 4124 config = spa_generate_rootconf(name); 4125 4126 mutex_enter(&spa_namespace_lock); 4127 if (config != NULL) { 4128 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4129 &pname) == 0 && strcmp(name, pname) == 0); 4130 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4131 == 0); 4132 4133 if ((spa = spa_lookup(pname)) != NULL) { 4134 /* 4135 * Remove the existing root pool from the namespace so 4136 * that we can replace it with the correct config 4137 * we just read in. 4138 */ 4139 spa_remove(spa); 4140 } 4141 spa = spa_add(pname, config, NULL); 4142 4143 /* 4144 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4145 * via spa_version(). 4146 */ 4147 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4148 &spa->spa_ubsync.ub_version) != 0) 4149 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4150 } else if ((spa = spa_lookup(name)) == NULL) { 4151 mutex_exit(&spa_namespace_lock); 4152 nvlist_free(config); 4153 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4154 name); 4155 return (EIO); 4156 } else { 4157 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4158 } 4159 spa->spa_is_root = B_TRUE; 4160 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4161 4162 /* 4163 * Build up a vdev tree based on the boot device's label config. 4164 */ 4165 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4166 &nvtop) == 0); 4167 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4168 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4169 VDEV_ALLOC_ROOTPOOL); 4170 spa_config_exit(spa, SCL_ALL, FTAG); 4171 if (error) { 4172 mutex_exit(&spa_namespace_lock); 4173 nvlist_free(config); 4174 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4175 pname); 4176 return (error); 4177 } 4178 4179 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4180 vdev_free(rvd); 4181 spa_config_exit(spa, SCL_ALL, FTAG); 4182 mutex_exit(&spa_namespace_lock); 4183 4184 nvlist_free(config); 4185 return (0); 4186} 4187 4188#endif /* illumos */ 4189#endif /* _KERNEL */ 4190 4191/* 4192 * Import a non-root pool into the system. 4193 */ 4194int 4195spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4196{ 4197 spa_t *spa; 4198 char *altroot = NULL; 4199 spa_load_state_t state = SPA_LOAD_IMPORT; 4200 zpool_rewind_policy_t policy; 4201 uint64_t mode = spa_mode_global; 4202 uint64_t readonly = B_FALSE; 4203 int error; 4204 nvlist_t *nvroot; 4205 nvlist_t **spares, **l2cache; 4206 uint_t nspares, nl2cache; 4207 4208 /* 4209 * If a pool with this name exists, return failure. 4210 */ 4211 mutex_enter(&spa_namespace_lock); 4212 if (spa_lookup(pool) != NULL) { 4213 mutex_exit(&spa_namespace_lock); 4214 return (SET_ERROR(EEXIST)); 4215 } 4216 4217 /* 4218 * Create and initialize the spa structure. 4219 */ 4220 (void) nvlist_lookup_string(props, 4221 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4222 (void) nvlist_lookup_uint64(props, 4223 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4224 if (readonly) 4225 mode = FREAD; 4226 spa = spa_add(pool, config, altroot); 4227 spa->spa_import_flags = flags; 4228 4229 /* 4230 * Verbatim import - Take a pool and insert it into the namespace 4231 * as if it had been loaded at boot. 4232 */ 4233 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4234 if (props != NULL) 4235 spa_configfile_set(spa, props, B_FALSE); 4236 4237 spa_config_sync(spa, B_FALSE, B_TRUE); 4238 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4239 4240 mutex_exit(&spa_namespace_lock); 4241 return (0); 4242 } 4243 4244 spa_activate(spa, mode); 4245 4246 /* 4247 * Don't start async tasks until we know everything is healthy. 4248 */ 4249 spa_async_suspend(spa); 4250 4251 zpool_get_rewind_policy(config, &policy); 4252 if (policy.zrp_request & ZPOOL_DO_REWIND) 4253 state = SPA_LOAD_RECOVER; 4254 4255 /* 4256 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4257 * because the user-supplied config is actually the one to trust when 4258 * doing an import. 4259 */ 4260 if (state != SPA_LOAD_RECOVER) 4261 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4262 4263 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4264 policy.zrp_request); 4265 4266 /* 4267 * Propagate anything learned while loading the pool and pass it 4268 * back to caller (i.e. rewind info, missing devices, etc). 4269 */ 4270 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4271 spa->spa_load_info) == 0); 4272 4273 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4274 /* 4275 * Toss any existing sparelist, as it doesn't have any validity 4276 * anymore, and conflicts with spa_has_spare(). 4277 */ 4278 if (spa->spa_spares.sav_config) { 4279 nvlist_free(spa->spa_spares.sav_config); 4280 spa->spa_spares.sav_config = NULL; 4281 spa_load_spares(spa); 4282 } 4283 if (spa->spa_l2cache.sav_config) { 4284 nvlist_free(spa->spa_l2cache.sav_config); 4285 spa->spa_l2cache.sav_config = NULL; 4286 spa_load_l2cache(spa); 4287 } 4288 4289 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4290 &nvroot) == 0); 4291 if (error == 0) 4292 error = spa_validate_aux(spa, nvroot, -1ULL, 4293 VDEV_ALLOC_SPARE); 4294 if (error == 0) 4295 error = spa_validate_aux(spa, nvroot, -1ULL, 4296 VDEV_ALLOC_L2CACHE); 4297 spa_config_exit(spa, SCL_ALL, FTAG); 4298 4299 if (props != NULL) 4300 spa_configfile_set(spa, props, B_FALSE); 4301 4302 if (error != 0 || (props && spa_writeable(spa) && 4303 (error = spa_prop_set(spa, props)))) { 4304 spa_unload(spa); 4305 spa_deactivate(spa); 4306 spa_remove(spa); 4307 mutex_exit(&spa_namespace_lock); 4308 return (error); 4309 } 4310 4311 spa_async_resume(spa); 4312 4313 /* 4314 * Override any spares and level 2 cache devices as specified by 4315 * the user, as these may have correct device names/devids, etc. 4316 */ 4317 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4318 &spares, &nspares) == 0) { 4319 if (spa->spa_spares.sav_config) 4320 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4321 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4322 else 4323 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4324 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4325 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4326 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4327 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4328 spa_load_spares(spa); 4329 spa_config_exit(spa, SCL_ALL, FTAG); 4330 spa->spa_spares.sav_sync = B_TRUE; 4331 } 4332 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4333 &l2cache, &nl2cache) == 0) { 4334 if (spa->spa_l2cache.sav_config) 4335 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4336 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4337 else 4338 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4339 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4340 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4341 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4342 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4343 spa_load_l2cache(spa); 4344 spa_config_exit(spa, SCL_ALL, FTAG); 4345 spa->spa_l2cache.sav_sync = B_TRUE; 4346 } 4347 4348 /* 4349 * Check for any removed devices. 4350 */ 4351 if (spa->spa_autoreplace) { 4352 spa_aux_check_removed(&spa->spa_spares); 4353 spa_aux_check_removed(&spa->spa_l2cache); 4354 } 4355 4356 if (spa_writeable(spa)) { 4357 /* 4358 * Update the config cache to include the newly-imported pool. 4359 */ 4360 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4361 } 4362 4363 /* 4364 * It's possible that the pool was expanded while it was exported. 4365 * We kick off an async task to handle this for us. 4366 */ 4367 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4368 4369 spa_history_log_version(spa, "import"); 4370 4371 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4372 4373 mutex_exit(&spa_namespace_lock); 4374 4375#ifdef __FreeBSD__ 4376#ifdef _KERNEL 4377 zvol_create_minors(pool); 4378#endif 4379#endif 4380 return (0); 4381} 4382 4383nvlist_t * 4384spa_tryimport(nvlist_t *tryconfig) 4385{ 4386 nvlist_t *config = NULL; 4387 char *poolname; 4388 spa_t *spa; 4389 uint64_t state; 4390 int error; 4391 4392 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4393 return (NULL); 4394 4395 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4396 return (NULL); 4397 4398 /* 4399 * Create and initialize the spa structure. 4400 */ 4401 mutex_enter(&spa_namespace_lock); 4402 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4403 spa_activate(spa, FREAD); 4404 4405 /* 4406 * Pass off the heavy lifting to spa_load(). 4407 * Pass TRUE for mosconfig because the user-supplied config 4408 * is actually the one to trust when doing an import. 4409 */ 4410 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4411 4412 /* 4413 * If 'tryconfig' was at least parsable, return the current config. 4414 */ 4415 if (spa->spa_root_vdev != NULL) { 4416 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4417 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4418 poolname) == 0); 4419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4420 state) == 0); 4421 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4422 spa->spa_uberblock.ub_timestamp) == 0); 4423 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4424 spa->spa_load_info) == 0); 4425 4426 /* 4427 * If the bootfs property exists on this pool then we 4428 * copy it out so that external consumers can tell which 4429 * pools are bootable. 4430 */ 4431 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4432 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4433 4434 /* 4435 * We have to play games with the name since the 4436 * pool was opened as TRYIMPORT_NAME. 4437 */ 4438 if (dsl_dsobj_to_dsname(spa_name(spa), 4439 spa->spa_bootfs, tmpname) == 0) { 4440 char *cp; 4441 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4442 4443 cp = strchr(tmpname, '/'); 4444 if (cp == NULL) { 4445 (void) strlcpy(dsname, tmpname, 4446 MAXPATHLEN); 4447 } else { 4448 (void) snprintf(dsname, MAXPATHLEN, 4449 "%s/%s", poolname, ++cp); 4450 } 4451 VERIFY(nvlist_add_string(config, 4452 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4453 kmem_free(dsname, MAXPATHLEN); 4454 } 4455 kmem_free(tmpname, MAXPATHLEN); 4456 } 4457 4458 /* 4459 * Add the list of hot spares and level 2 cache devices. 4460 */ 4461 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4462 spa_add_spares(spa, config); 4463 spa_add_l2cache(spa, config); 4464 spa_config_exit(spa, SCL_CONFIG, FTAG); 4465 } 4466 4467 spa_unload(spa); 4468 spa_deactivate(spa); 4469 spa_remove(spa); 4470 mutex_exit(&spa_namespace_lock); 4471 4472 return (config); 4473} 4474 4475/* 4476 * Pool export/destroy 4477 * 4478 * The act of destroying or exporting a pool is very simple. We make sure there 4479 * is no more pending I/O and any references to the pool are gone. Then, we 4480 * update the pool state and sync all the labels to disk, removing the 4481 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4482 * we don't sync the labels or remove the configuration cache. 4483 */ 4484static int 4485spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4486 boolean_t force, boolean_t hardforce) 4487{ 4488 spa_t *spa; 4489 4490 if (oldconfig) 4491 *oldconfig = NULL; 4492 4493 if (!(spa_mode_global & FWRITE)) 4494 return (SET_ERROR(EROFS)); 4495 4496 mutex_enter(&spa_namespace_lock); 4497 if ((spa = spa_lookup(pool)) == NULL) { 4498 mutex_exit(&spa_namespace_lock); 4499 return (SET_ERROR(ENOENT)); 4500 } 4501 4502 /* 4503 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4504 * reacquire the namespace lock, and see if we can export. 4505 */ 4506 spa_open_ref(spa, FTAG); 4507 mutex_exit(&spa_namespace_lock); 4508 spa_async_suspend(spa); 4509 mutex_enter(&spa_namespace_lock); 4510 spa_close(spa, FTAG); 4511 4512 /* 4513 * The pool will be in core if it's openable, 4514 * in which case we can modify its state. 4515 */ 4516 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4517 /* 4518 * Objsets may be open only because they're dirty, so we 4519 * have to force it to sync before checking spa_refcnt. 4520 */ 4521 txg_wait_synced(spa->spa_dsl_pool, 0); 4522 spa_evicting_os_wait(spa); 4523 4524 /* 4525 * A pool cannot be exported or destroyed if there are active 4526 * references. If we are resetting a pool, allow references by 4527 * fault injection handlers. 4528 */ 4529 if (!spa_refcount_zero(spa) || 4530 (spa->spa_inject_ref != 0 && 4531 new_state != POOL_STATE_UNINITIALIZED)) { 4532 spa_async_resume(spa); 4533 mutex_exit(&spa_namespace_lock); 4534 return (SET_ERROR(EBUSY)); 4535 } 4536 4537 /* 4538 * A pool cannot be exported if it has an active shared spare. 4539 * This is to prevent other pools stealing the active spare 4540 * from an exported pool. At user's own will, such pool can 4541 * be forcedly exported. 4542 */ 4543 if (!force && new_state == POOL_STATE_EXPORTED && 4544 spa_has_active_shared_spare(spa)) { 4545 spa_async_resume(spa); 4546 mutex_exit(&spa_namespace_lock); 4547 return (SET_ERROR(EXDEV)); 4548 } 4549 4550 /* 4551 * We want this to be reflected on every label, 4552 * so mark them all dirty. spa_unload() will do the 4553 * final sync that pushes these changes out. 4554 */ 4555 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4556 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4557 spa->spa_state = new_state; 4558 spa->spa_final_txg = spa_last_synced_txg(spa) + 4559 TXG_DEFER_SIZE + 1; 4560 vdev_config_dirty(spa->spa_root_vdev); 4561 spa_config_exit(spa, SCL_ALL, FTAG); 4562 } 4563 } 4564 4565 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4566 4567 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4568 spa_unload(spa); 4569 spa_deactivate(spa); 4570 } 4571 4572 if (oldconfig && spa->spa_config) 4573 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4574 4575 if (new_state != POOL_STATE_UNINITIALIZED) { 4576 if (!hardforce) 4577 spa_config_sync(spa, B_TRUE, B_TRUE); 4578 spa_remove(spa); 4579 } 4580 mutex_exit(&spa_namespace_lock); 4581 4582 return (0); 4583} 4584 4585/* 4586 * Destroy a storage pool. 4587 */ 4588int 4589spa_destroy(char *pool) 4590{ 4591 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4592 B_FALSE, B_FALSE)); 4593} 4594 4595/* 4596 * Export a storage pool. 4597 */ 4598int 4599spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4600 boolean_t hardforce) 4601{ 4602 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4603 force, hardforce)); 4604} 4605 4606/* 4607 * Similar to spa_export(), this unloads the spa_t without actually removing it 4608 * from the namespace in any way. 4609 */ 4610int 4611spa_reset(char *pool) 4612{ 4613 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4614 B_FALSE, B_FALSE)); 4615} 4616 4617/* 4618 * ========================================================================== 4619 * Device manipulation 4620 * ========================================================================== 4621 */ 4622 4623/* 4624 * Add a device to a storage pool. 4625 */ 4626int 4627spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4628{ 4629 uint64_t txg, id; 4630 int error; 4631 vdev_t *rvd = spa->spa_root_vdev; 4632 vdev_t *vd, *tvd; 4633 nvlist_t **spares, **l2cache; 4634 uint_t nspares, nl2cache; 4635 4636 ASSERT(spa_writeable(spa)); 4637 4638 txg = spa_vdev_enter(spa); 4639 4640 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4641 VDEV_ALLOC_ADD)) != 0) 4642 return (spa_vdev_exit(spa, NULL, txg, error)); 4643 4644 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4645 4646 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4647 &nspares) != 0) 4648 nspares = 0; 4649 4650 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4651 &nl2cache) != 0) 4652 nl2cache = 0; 4653 4654 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4655 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4656 4657 if (vd->vdev_children != 0 && 4658 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4659 return (spa_vdev_exit(spa, vd, txg, error)); 4660 4661 /* 4662 * We must validate the spares and l2cache devices after checking the 4663 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4664 */ 4665 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4666 return (spa_vdev_exit(spa, vd, txg, error)); 4667 4668 /* 4669 * Transfer each new top-level vdev from vd to rvd. 4670 */ 4671 for (int c = 0; c < vd->vdev_children; c++) { 4672 4673 /* 4674 * Set the vdev id to the first hole, if one exists. 4675 */ 4676 for (id = 0; id < rvd->vdev_children; id++) { 4677 if (rvd->vdev_child[id]->vdev_ishole) { 4678 vdev_free(rvd->vdev_child[id]); 4679 break; 4680 } 4681 } 4682 tvd = vd->vdev_child[c]; 4683 vdev_remove_child(vd, tvd); 4684 tvd->vdev_id = id; 4685 vdev_add_child(rvd, tvd); 4686 vdev_config_dirty(tvd); 4687 } 4688 4689 if (nspares != 0) { 4690 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4691 ZPOOL_CONFIG_SPARES); 4692 spa_load_spares(spa); 4693 spa->spa_spares.sav_sync = B_TRUE; 4694 } 4695 4696 if (nl2cache != 0) { 4697 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4698 ZPOOL_CONFIG_L2CACHE); 4699 spa_load_l2cache(spa); 4700 spa->spa_l2cache.sav_sync = B_TRUE; 4701 } 4702 4703 /* 4704 * We have to be careful when adding new vdevs to an existing pool. 4705 * If other threads start allocating from these vdevs before we 4706 * sync the config cache, and we lose power, then upon reboot we may 4707 * fail to open the pool because there are DVAs that the config cache 4708 * can't translate. Therefore, we first add the vdevs without 4709 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4710 * and then let spa_config_update() initialize the new metaslabs. 4711 * 4712 * spa_load() checks for added-but-not-initialized vdevs, so that 4713 * if we lose power at any point in this sequence, the remaining 4714 * steps will be completed the next time we load the pool. 4715 */ 4716 (void) spa_vdev_exit(spa, vd, txg, 0); 4717 4718 mutex_enter(&spa_namespace_lock); 4719 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4720 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 4721 mutex_exit(&spa_namespace_lock); 4722 4723 return (0); 4724} 4725 4726/* 4727 * Attach a device to a mirror. The arguments are the path to any device 4728 * in the mirror, and the nvroot for the new device. If the path specifies 4729 * a device that is not mirrored, we automatically insert the mirror vdev. 4730 * 4731 * If 'replacing' is specified, the new device is intended to replace the 4732 * existing device; in this case the two devices are made into their own 4733 * mirror using the 'replacing' vdev, which is functionally identical to 4734 * the mirror vdev (it actually reuses all the same ops) but has a few 4735 * extra rules: you can't attach to it after it's been created, and upon 4736 * completion of resilvering, the first disk (the one being replaced) 4737 * is automatically detached. 4738 */ 4739int 4740spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4741{ 4742 uint64_t txg, dtl_max_txg; 4743 vdev_t *rvd = spa->spa_root_vdev; 4744 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4745 vdev_ops_t *pvops; 4746 char *oldvdpath, *newvdpath; 4747 int newvd_isspare; 4748 int error; 4749 4750 ASSERT(spa_writeable(spa)); 4751 4752 txg = spa_vdev_enter(spa); 4753 4754 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4755 4756 if (oldvd == NULL) 4757 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4758 4759 if (!oldvd->vdev_ops->vdev_op_leaf) 4760 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4761 4762 pvd = oldvd->vdev_parent; 4763 4764 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4765 VDEV_ALLOC_ATTACH)) != 0) 4766 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4767 4768 if (newrootvd->vdev_children != 1) 4769 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4770 4771 newvd = newrootvd->vdev_child[0]; 4772 4773 if (!newvd->vdev_ops->vdev_op_leaf) 4774 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4775 4776 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4777 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4778 4779 /* 4780 * Spares can't replace logs 4781 */ 4782 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4783 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4784 4785 if (!replacing) { 4786 /* 4787 * For attach, the only allowable parent is a mirror or the root 4788 * vdev. 4789 */ 4790 if (pvd->vdev_ops != &vdev_mirror_ops && 4791 pvd->vdev_ops != &vdev_root_ops) 4792 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4793 4794 pvops = &vdev_mirror_ops; 4795 } else { 4796 /* 4797 * Active hot spares can only be replaced by inactive hot 4798 * spares. 4799 */ 4800 if (pvd->vdev_ops == &vdev_spare_ops && 4801 oldvd->vdev_isspare && 4802 !spa_has_spare(spa, newvd->vdev_guid)) 4803 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4804 4805 /* 4806 * If the source is a hot spare, and the parent isn't already a 4807 * spare, then we want to create a new hot spare. Otherwise, we 4808 * want to create a replacing vdev. The user is not allowed to 4809 * attach to a spared vdev child unless the 'isspare' state is 4810 * the same (spare replaces spare, non-spare replaces 4811 * non-spare). 4812 */ 4813 if (pvd->vdev_ops == &vdev_replacing_ops && 4814 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4815 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4816 } else if (pvd->vdev_ops == &vdev_spare_ops && 4817 newvd->vdev_isspare != oldvd->vdev_isspare) { 4818 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4819 } 4820 4821 if (newvd->vdev_isspare) 4822 pvops = &vdev_spare_ops; 4823 else 4824 pvops = &vdev_replacing_ops; 4825 } 4826 4827 /* 4828 * Make sure the new device is big enough. 4829 */ 4830 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4831 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4832 4833 /* 4834 * The new device cannot have a higher alignment requirement 4835 * than the top-level vdev. 4836 */ 4837 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4838 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4839 4840 /* 4841 * If this is an in-place replacement, update oldvd's path and devid 4842 * to make it distinguishable from newvd, and unopenable from now on. 4843 */ 4844 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4845 spa_strfree(oldvd->vdev_path); 4846 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4847 KM_SLEEP); 4848 (void) sprintf(oldvd->vdev_path, "%s/%s", 4849 newvd->vdev_path, "old"); 4850 if (oldvd->vdev_devid != NULL) { 4851 spa_strfree(oldvd->vdev_devid); 4852 oldvd->vdev_devid = NULL; 4853 } 4854 } 4855 4856 /* mark the device being resilvered */ 4857 newvd->vdev_resilver_txg = txg; 4858 4859 /* 4860 * If the parent is not a mirror, or if we're replacing, insert the new 4861 * mirror/replacing/spare vdev above oldvd. 4862 */ 4863 if (pvd->vdev_ops != pvops) 4864 pvd = vdev_add_parent(oldvd, pvops); 4865 4866 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4867 ASSERT(pvd->vdev_ops == pvops); 4868 ASSERT(oldvd->vdev_parent == pvd); 4869 4870 /* 4871 * Extract the new device from its root and add it to pvd. 4872 */ 4873 vdev_remove_child(newrootvd, newvd); 4874 newvd->vdev_id = pvd->vdev_children; 4875 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4876 vdev_add_child(pvd, newvd); 4877 4878 tvd = newvd->vdev_top; 4879 ASSERT(pvd->vdev_top == tvd); 4880 ASSERT(tvd->vdev_parent == rvd); 4881 4882 vdev_config_dirty(tvd); 4883 4884 /* 4885 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4886 * for any dmu_sync-ed blocks. It will propagate upward when 4887 * spa_vdev_exit() calls vdev_dtl_reassess(). 4888 */ 4889 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4890 4891 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4892 dtl_max_txg - TXG_INITIAL); 4893 4894 if (newvd->vdev_isspare) { 4895 spa_spare_activate(newvd); 4896 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4897 } 4898 4899 oldvdpath = spa_strdup(oldvd->vdev_path); 4900 newvdpath = spa_strdup(newvd->vdev_path); 4901 newvd_isspare = newvd->vdev_isspare; 4902 4903 /* 4904 * Mark newvd's DTL dirty in this txg. 4905 */ 4906 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4907 4908 /* 4909 * Schedule the resilver to restart in the future. We do this to 4910 * ensure that dmu_sync-ed blocks have been stitched into the 4911 * respective datasets. 4912 */ 4913 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4914 4915 if (spa->spa_bootfs) 4916 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4917 4918 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 4919 4920 /* 4921 * Commit the config 4922 */ 4923 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4924 4925 spa_history_log_internal(spa, "vdev attach", NULL, 4926 "%s vdev=%s %s vdev=%s", 4927 replacing && newvd_isspare ? "spare in" : 4928 replacing ? "replace" : "attach", newvdpath, 4929 replacing ? "for" : "to", oldvdpath); 4930 4931 spa_strfree(oldvdpath); 4932 spa_strfree(newvdpath); 4933 4934 return (0); 4935} 4936 4937/* 4938 * Detach a device from a mirror or replacing vdev. 4939 * 4940 * If 'replace_done' is specified, only detach if the parent 4941 * is a replacing vdev. 4942 */ 4943int 4944spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4945{ 4946 uint64_t txg; 4947 int error; 4948 vdev_t *rvd = spa->spa_root_vdev; 4949 vdev_t *vd, *pvd, *cvd, *tvd; 4950 boolean_t unspare = B_FALSE; 4951 uint64_t unspare_guid = 0; 4952 char *vdpath; 4953 4954 ASSERT(spa_writeable(spa)); 4955 4956 txg = spa_vdev_enter(spa); 4957 4958 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4959 4960 if (vd == NULL) 4961 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4962 4963 if (!vd->vdev_ops->vdev_op_leaf) 4964 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4965 4966 pvd = vd->vdev_parent; 4967 4968 /* 4969 * If the parent/child relationship is not as expected, don't do it. 4970 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4971 * vdev that's replacing B with C. The user's intent in replacing 4972 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4973 * the replace by detaching C, the expected behavior is to end up 4974 * M(A,B). But suppose that right after deciding to detach C, 4975 * the replacement of B completes. We would have M(A,C), and then 4976 * ask to detach C, which would leave us with just A -- not what 4977 * the user wanted. To prevent this, we make sure that the 4978 * parent/child relationship hasn't changed -- in this example, 4979 * that C's parent is still the replacing vdev R. 4980 */ 4981 if (pvd->vdev_guid != pguid && pguid != 0) 4982 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4983 4984 /* 4985 * Only 'replacing' or 'spare' vdevs can be replaced. 4986 */ 4987 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4988 pvd->vdev_ops != &vdev_spare_ops) 4989 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4990 4991 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4992 spa_version(spa) >= SPA_VERSION_SPARES); 4993 4994 /* 4995 * Only mirror, replacing, and spare vdevs support detach. 4996 */ 4997 if (pvd->vdev_ops != &vdev_replacing_ops && 4998 pvd->vdev_ops != &vdev_mirror_ops && 4999 pvd->vdev_ops != &vdev_spare_ops) 5000 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5001 5002 /* 5003 * If this device has the only valid copy of some data, 5004 * we cannot safely detach it. 5005 */ 5006 if (vdev_dtl_required(vd)) 5007 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5008 5009 ASSERT(pvd->vdev_children >= 2); 5010 5011 /* 5012 * If we are detaching the second disk from a replacing vdev, then 5013 * check to see if we changed the original vdev's path to have "/old" 5014 * at the end in spa_vdev_attach(). If so, undo that change now. 5015 */ 5016 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5017 vd->vdev_path != NULL) { 5018 size_t len = strlen(vd->vdev_path); 5019 5020 for (int c = 0; c < pvd->vdev_children; c++) { 5021 cvd = pvd->vdev_child[c]; 5022 5023 if (cvd == vd || cvd->vdev_path == NULL) 5024 continue; 5025 5026 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5027 strcmp(cvd->vdev_path + len, "/old") == 0) { 5028 spa_strfree(cvd->vdev_path); 5029 cvd->vdev_path = spa_strdup(vd->vdev_path); 5030 break; 5031 } 5032 } 5033 } 5034 5035 /* 5036 * If we are detaching the original disk from a spare, then it implies 5037 * that the spare should become a real disk, and be removed from the 5038 * active spare list for the pool. 5039 */ 5040 if (pvd->vdev_ops == &vdev_spare_ops && 5041 vd->vdev_id == 0 && 5042 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5043 unspare = B_TRUE; 5044 5045 /* 5046 * Erase the disk labels so the disk can be used for other things. 5047 * This must be done after all other error cases are handled, 5048 * but before we disembowel vd (so we can still do I/O to it). 5049 * But if we can't do it, don't treat the error as fatal -- 5050 * it may be that the unwritability of the disk is the reason 5051 * it's being detached! 5052 */ 5053 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5054 5055 /* 5056 * Remove vd from its parent and compact the parent's children. 5057 */ 5058 vdev_remove_child(pvd, vd); 5059 vdev_compact_children(pvd); 5060 5061 /* 5062 * Remember one of the remaining children so we can get tvd below. 5063 */ 5064 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5065 5066 /* 5067 * If we need to remove the remaining child from the list of hot spares, 5068 * do it now, marking the vdev as no longer a spare in the process. 5069 * We must do this before vdev_remove_parent(), because that can 5070 * change the GUID if it creates a new toplevel GUID. For a similar 5071 * reason, we must remove the spare now, in the same txg as the detach; 5072 * otherwise someone could attach a new sibling, change the GUID, and 5073 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5074 */ 5075 if (unspare) { 5076 ASSERT(cvd->vdev_isspare); 5077 spa_spare_remove(cvd); 5078 unspare_guid = cvd->vdev_guid; 5079 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5080 cvd->vdev_unspare = B_TRUE; 5081 } 5082 5083 /* 5084 * If the parent mirror/replacing vdev only has one child, 5085 * the parent is no longer needed. Remove it from the tree. 5086 */ 5087 if (pvd->vdev_children == 1) { 5088 if (pvd->vdev_ops == &vdev_spare_ops) 5089 cvd->vdev_unspare = B_FALSE; 5090 vdev_remove_parent(cvd); 5091 } 5092 5093 5094 /* 5095 * We don't set tvd until now because the parent we just removed 5096 * may have been the previous top-level vdev. 5097 */ 5098 tvd = cvd->vdev_top; 5099 ASSERT(tvd->vdev_parent == rvd); 5100 5101 /* 5102 * Reevaluate the parent vdev state. 5103 */ 5104 vdev_propagate_state(cvd); 5105 5106 /* 5107 * If the 'autoexpand' property is set on the pool then automatically 5108 * try to expand the size of the pool. For example if the device we 5109 * just detached was smaller than the others, it may be possible to 5110 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5111 * first so that we can obtain the updated sizes of the leaf vdevs. 5112 */ 5113 if (spa->spa_autoexpand) { 5114 vdev_reopen(tvd); 5115 vdev_expand(tvd, txg); 5116 } 5117 5118 vdev_config_dirty(tvd); 5119 5120 /* 5121 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5122 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5123 * But first make sure we're not on any *other* txg's DTL list, to 5124 * prevent vd from being accessed after it's freed. 5125 */ 5126 vdpath = spa_strdup(vd->vdev_path); 5127 for (int t = 0; t < TXG_SIZE; t++) 5128 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5129 vd->vdev_detached = B_TRUE; 5130 vdev_dirty(tvd, VDD_DTL, vd, txg); 5131 5132 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5133 5134 /* hang on to the spa before we release the lock */ 5135 spa_open_ref(spa, FTAG); 5136 5137 error = spa_vdev_exit(spa, vd, txg, 0); 5138 5139 spa_history_log_internal(spa, "detach", NULL, 5140 "vdev=%s", vdpath); 5141 spa_strfree(vdpath); 5142 5143 /* 5144 * If this was the removal of the original device in a hot spare vdev, 5145 * then we want to go through and remove the device from the hot spare 5146 * list of every other pool. 5147 */ 5148 if (unspare) { 5149 spa_t *altspa = NULL; 5150 5151 mutex_enter(&spa_namespace_lock); 5152 while ((altspa = spa_next(altspa)) != NULL) { 5153 if (altspa->spa_state != POOL_STATE_ACTIVE || 5154 altspa == spa) 5155 continue; 5156 5157 spa_open_ref(altspa, FTAG); 5158 mutex_exit(&spa_namespace_lock); 5159 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5160 mutex_enter(&spa_namespace_lock); 5161 spa_close(altspa, FTAG); 5162 } 5163 mutex_exit(&spa_namespace_lock); 5164 5165 /* search the rest of the vdevs for spares to remove */ 5166 spa_vdev_resilver_done(spa); 5167 } 5168 5169 /* all done with the spa; OK to release */ 5170 mutex_enter(&spa_namespace_lock); 5171 spa_close(spa, FTAG); 5172 mutex_exit(&spa_namespace_lock); 5173 5174 return (error); 5175} 5176 5177/* 5178 * Split a set of devices from their mirrors, and create a new pool from them. 5179 */ 5180int 5181spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5182 nvlist_t *props, boolean_t exp) 5183{ 5184 int error = 0; 5185 uint64_t txg, *glist; 5186 spa_t *newspa; 5187 uint_t c, children, lastlog; 5188 nvlist_t **child, *nvl, *tmp; 5189 dmu_tx_t *tx; 5190 char *altroot = NULL; 5191 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5192 boolean_t activate_slog; 5193 5194 ASSERT(spa_writeable(spa)); 5195 5196 txg = spa_vdev_enter(spa); 5197 5198 /* clear the log and flush everything up to now */ 5199 activate_slog = spa_passivate_log(spa); 5200 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5201 error = spa_offline_log(spa); 5202 txg = spa_vdev_config_enter(spa); 5203 5204 if (activate_slog) 5205 spa_activate_log(spa); 5206 5207 if (error != 0) 5208 return (spa_vdev_exit(spa, NULL, txg, error)); 5209 5210 /* check new spa name before going any further */ 5211 if (spa_lookup(newname) != NULL) 5212 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5213 5214 /* 5215 * scan through all the children to ensure they're all mirrors 5216 */ 5217 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5218 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5219 &children) != 0) 5220 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5221 5222 /* first, check to ensure we've got the right child count */ 5223 rvd = spa->spa_root_vdev; 5224 lastlog = 0; 5225 for (c = 0; c < rvd->vdev_children; c++) { 5226 vdev_t *vd = rvd->vdev_child[c]; 5227 5228 /* don't count the holes & logs as children */ 5229 if (vd->vdev_islog || vd->vdev_ishole) { 5230 if (lastlog == 0) 5231 lastlog = c; 5232 continue; 5233 } 5234 5235 lastlog = 0; 5236 } 5237 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5238 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5239 5240 /* next, ensure no spare or cache devices are part of the split */ 5241 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5242 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5243 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5244 5245 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5246 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5247 5248 /* then, loop over each vdev and validate it */ 5249 for (c = 0; c < children; c++) { 5250 uint64_t is_hole = 0; 5251 5252 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5253 &is_hole); 5254 5255 if (is_hole != 0) { 5256 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5257 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5258 continue; 5259 } else { 5260 error = SET_ERROR(EINVAL); 5261 break; 5262 } 5263 } 5264 5265 /* which disk is going to be split? */ 5266 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5267 &glist[c]) != 0) { 5268 error = SET_ERROR(EINVAL); 5269 break; 5270 } 5271 5272 /* look it up in the spa */ 5273 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5274 if (vml[c] == NULL) { 5275 error = SET_ERROR(ENODEV); 5276 break; 5277 } 5278 5279 /* make sure there's nothing stopping the split */ 5280 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5281 vml[c]->vdev_islog || 5282 vml[c]->vdev_ishole || 5283 vml[c]->vdev_isspare || 5284 vml[c]->vdev_isl2cache || 5285 !vdev_writeable(vml[c]) || 5286 vml[c]->vdev_children != 0 || 5287 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5288 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5289 error = SET_ERROR(EINVAL); 5290 break; 5291 } 5292 5293 if (vdev_dtl_required(vml[c])) { 5294 error = SET_ERROR(EBUSY); 5295 break; 5296 } 5297 5298 /* we need certain info from the top level */ 5299 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5300 vml[c]->vdev_top->vdev_ms_array) == 0); 5301 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5302 vml[c]->vdev_top->vdev_ms_shift) == 0); 5303 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5304 vml[c]->vdev_top->vdev_asize) == 0); 5305 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5306 vml[c]->vdev_top->vdev_ashift) == 0); 5307 } 5308 5309 if (error != 0) { 5310 kmem_free(vml, children * sizeof (vdev_t *)); 5311 kmem_free(glist, children * sizeof (uint64_t)); 5312 return (spa_vdev_exit(spa, NULL, txg, error)); 5313 } 5314 5315 /* stop writers from using the disks */ 5316 for (c = 0; c < children; c++) { 5317 if (vml[c] != NULL) 5318 vml[c]->vdev_offline = B_TRUE; 5319 } 5320 vdev_reopen(spa->spa_root_vdev); 5321 5322 /* 5323 * Temporarily record the splitting vdevs in the spa config. This 5324 * will disappear once the config is regenerated. 5325 */ 5326 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5327 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5328 glist, children) == 0); 5329 kmem_free(glist, children * sizeof (uint64_t)); 5330 5331 mutex_enter(&spa->spa_props_lock); 5332 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5333 nvl) == 0); 5334 mutex_exit(&spa->spa_props_lock); 5335 spa->spa_config_splitting = nvl; 5336 vdev_config_dirty(spa->spa_root_vdev); 5337 5338 /* configure and create the new pool */ 5339 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5340 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5341 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5342 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5343 spa_version(spa)) == 0); 5344 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5345 spa->spa_config_txg) == 0); 5346 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5347 spa_generate_guid(NULL)) == 0); 5348 (void) nvlist_lookup_string(props, 5349 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5350 5351 /* add the new pool to the namespace */ 5352 newspa = spa_add(newname, config, altroot); 5353 newspa->spa_config_txg = spa->spa_config_txg; 5354 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5355 5356 /* release the spa config lock, retaining the namespace lock */ 5357 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5358 5359 if (zio_injection_enabled) 5360 zio_handle_panic_injection(spa, FTAG, 1); 5361 5362 spa_activate(newspa, spa_mode_global); 5363 spa_async_suspend(newspa); 5364 5365#ifndef illumos 5366 /* mark that we are creating new spa by splitting */ 5367 newspa->spa_splitting_newspa = B_TRUE; 5368#endif 5369 /* create the new pool from the disks of the original pool */ 5370 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5371#ifndef illumos 5372 newspa->spa_splitting_newspa = B_FALSE; 5373#endif 5374 if (error) 5375 goto out; 5376 5377 /* if that worked, generate a real config for the new pool */ 5378 if (newspa->spa_root_vdev != NULL) { 5379 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5380 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5381 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5382 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5383 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5384 B_TRUE)); 5385 } 5386 5387 /* set the props */ 5388 if (props != NULL) { 5389 spa_configfile_set(newspa, props, B_FALSE); 5390 error = spa_prop_set(newspa, props); 5391 if (error) 5392 goto out; 5393 } 5394 5395 /* flush everything */ 5396 txg = spa_vdev_config_enter(newspa); 5397 vdev_config_dirty(newspa->spa_root_vdev); 5398 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5399 5400 if (zio_injection_enabled) 5401 zio_handle_panic_injection(spa, FTAG, 2); 5402 5403 spa_async_resume(newspa); 5404 5405 /* finally, update the original pool's config */ 5406 txg = spa_vdev_config_enter(spa); 5407 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5408 error = dmu_tx_assign(tx, TXG_WAIT); 5409 if (error != 0) 5410 dmu_tx_abort(tx); 5411 for (c = 0; c < children; c++) { 5412 if (vml[c] != NULL) { 5413 vdev_split(vml[c]); 5414 if (error == 0) 5415 spa_history_log_internal(spa, "detach", tx, 5416 "vdev=%s", vml[c]->vdev_path); 5417 vdev_free(vml[c]); 5418 } 5419 } 5420 vdev_config_dirty(spa->spa_root_vdev); 5421 spa->spa_config_splitting = NULL; 5422 nvlist_free(nvl); 5423 if (error == 0) 5424 dmu_tx_commit(tx); 5425 (void) spa_vdev_exit(spa, NULL, txg, 0); 5426 5427 if (zio_injection_enabled) 5428 zio_handle_panic_injection(spa, FTAG, 3); 5429 5430 /* split is complete; log a history record */ 5431 spa_history_log_internal(newspa, "split", NULL, 5432 "from pool %s", spa_name(spa)); 5433 5434 kmem_free(vml, children * sizeof (vdev_t *)); 5435 5436 /* if we're not going to mount the filesystems in userland, export */ 5437 if (exp) 5438 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5439 B_FALSE, B_FALSE); 5440 5441 return (error); 5442 5443out: 5444 spa_unload(newspa); 5445 spa_deactivate(newspa); 5446 spa_remove(newspa); 5447 5448 txg = spa_vdev_config_enter(spa); 5449 5450 /* re-online all offlined disks */ 5451 for (c = 0; c < children; c++) { 5452 if (vml[c] != NULL) 5453 vml[c]->vdev_offline = B_FALSE; 5454 } 5455 vdev_reopen(spa->spa_root_vdev); 5456 5457 nvlist_free(spa->spa_config_splitting); 5458 spa->spa_config_splitting = NULL; 5459 (void) spa_vdev_exit(spa, NULL, txg, error); 5460 5461 kmem_free(vml, children * sizeof (vdev_t *)); 5462 return (error); 5463} 5464 5465static nvlist_t * 5466spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5467{ 5468 for (int i = 0; i < count; i++) { 5469 uint64_t guid; 5470 5471 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5472 &guid) == 0); 5473 5474 if (guid == target_guid) 5475 return (nvpp[i]); 5476 } 5477 5478 return (NULL); 5479} 5480 5481static void 5482spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5483 nvlist_t *dev_to_remove) 5484{ 5485 nvlist_t **newdev = NULL; 5486 5487 if (count > 1) 5488 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5489 5490 for (int i = 0, j = 0; i < count; i++) { 5491 if (dev[i] == dev_to_remove) 5492 continue; 5493 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5494 } 5495 5496 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5497 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5498 5499 for (int i = 0; i < count - 1; i++) 5500 nvlist_free(newdev[i]); 5501 5502 if (count > 1) 5503 kmem_free(newdev, (count - 1) * sizeof (void *)); 5504} 5505 5506/* 5507 * Evacuate the device. 5508 */ 5509static int 5510spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5511{ 5512 uint64_t txg; 5513 int error = 0; 5514 5515 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5516 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5517 ASSERT(vd == vd->vdev_top); 5518 5519 /* 5520 * Evacuate the device. We don't hold the config lock as writer 5521 * since we need to do I/O but we do keep the 5522 * spa_namespace_lock held. Once this completes the device 5523 * should no longer have any blocks allocated on it. 5524 */ 5525 if (vd->vdev_islog) { 5526 if (vd->vdev_stat.vs_alloc != 0) 5527 error = spa_offline_log(spa); 5528 } else { 5529 error = SET_ERROR(ENOTSUP); 5530 } 5531 5532 if (error) 5533 return (error); 5534 5535 /* 5536 * The evacuation succeeded. Remove any remaining MOS metadata 5537 * associated with this vdev, and wait for these changes to sync. 5538 */ 5539 ASSERT0(vd->vdev_stat.vs_alloc); 5540 txg = spa_vdev_config_enter(spa); 5541 vd->vdev_removing = B_TRUE; 5542 vdev_dirty_leaves(vd, VDD_DTL, txg); 5543 vdev_config_dirty(vd); 5544 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5545 5546 return (0); 5547} 5548 5549/* 5550 * Complete the removal by cleaning up the namespace. 5551 */ 5552static void 5553spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5554{ 5555 vdev_t *rvd = spa->spa_root_vdev; 5556 uint64_t id = vd->vdev_id; 5557 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5558 5559 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5560 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5561 ASSERT(vd == vd->vdev_top); 5562 5563 /* 5564 * Only remove any devices which are empty. 5565 */ 5566 if (vd->vdev_stat.vs_alloc != 0) 5567 return; 5568 5569 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5570 5571 if (list_link_active(&vd->vdev_state_dirty_node)) 5572 vdev_state_clean(vd); 5573 if (list_link_active(&vd->vdev_config_dirty_node)) 5574 vdev_config_clean(vd); 5575 5576 vdev_free(vd); 5577 5578 if (last_vdev) { 5579 vdev_compact_children(rvd); 5580 } else { 5581 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5582 vdev_add_child(rvd, vd); 5583 } 5584 vdev_config_dirty(rvd); 5585 5586 /* 5587 * Reassess the health of our root vdev. 5588 */ 5589 vdev_reopen(rvd); 5590} 5591 5592/* 5593 * Remove a device from the pool - 5594 * 5595 * Removing a device from the vdev namespace requires several steps 5596 * and can take a significant amount of time. As a result we use 5597 * the spa_vdev_config_[enter/exit] functions which allow us to 5598 * grab and release the spa_config_lock while still holding the namespace 5599 * lock. During each step the configuration is synced out. 5600 * 5601 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5602 * devices. 5603 */ 5604int 5605spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5606{ 5607 vdev_t *vd; 5608 metaslab_group_t *mg; 5609 nvlist_t **spares, **l2cache, *nv; 5610 uint64_t txg = 0; 5611 uint_t nspares, nl2cache; 5612 int error = 0; 5613 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5614 5615 ASSERT(spa_writeable(spa)); 5616 5617 if (!locked) 5618 txg = spa_vdev_enter(spa); 5619 5620 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5621 5622 if (spa->spa_spares.sav_vdevs != NULL && 5623 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5624 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5625 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5626 /* 5627 * Only remove the hot spare if it's not currently in use 5628 * in this pool. 5629 */ 5630 if (vd == NULL || unspare) { 5631 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5632 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5633 spa_load_spares(spa); 5634 spa->spa_spares.sav_sync = B_TRUE; 5635 } else { 5636 error = SET_ERROR(EBUSY); 5637 } 5638 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5639 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5640 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5641 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5642 /* 5643 * Cache devices can always be removed. 5644 */ 5645 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5646 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5647 spa_load_l2cache(spa); 5648 spa->spa_l2cache.sav_sync = B_TRUE; 5649 } else if (vd != NULL && vd->vdev_islog) { 5650 ASSERT(!locked); 5651 ASSERT(vd == vd->vdev_top); 5652 5653 mg = vd->vdev_mg; 5654 5655 /* 5656 * Stop allocating from this vdev. 5657 */ 5658 metaslab_group_passivate(mg); 5659 5660 /* 5661 * Wait for the youngest allocations and frees to sync, 5662 * and then wait for the deferral of those frees to finish. 5663 */ 5664 spa_vdev_config_exit(spa, NULL, 5665 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5666 5667 /* 5668 * Attempt to evacuate the vdev. 5669 */ 5670 error = spa_vdev_remove_evacuate(spa, vd); 5671 5672 txg = spa_vdev_config_enter(spa); 5673 5674 /* 5675 * If we couldn't evacuate the vdev, unwind. 5676 */ 5677 if (error) { 5678 metaslab_group_activate(mg); 5679 return (spa_vdev_exit(spa, NULL, txg, error)); 5680 } 5681 5682 /* 5683 * Clean up the vdev namespace. 5684 */ 5685 spa_vdev_remove_from_namespace(spa, vd); 5686 5687 } else if (vd != NULL) { 5688 /* 5689 * Normal vdevs cannot be removed (yet). 5690 */ 5691 error = SET_ERROR(ENOTSUP); 5692 } else { 5693 /* 5694 * There is no vdev of any kind with the specified guid. 5695 */ 5696 error = SET_ERROR(ENOENT); 5697 } 5698 5699 if (!locked) 5700 return (spa_vdev_exit(spa, NULL, txg, error)); 5701 5702 return (error); 5703} 5704 5705/* 5706 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5707 * currently spared, so we can detach it. 5708 */ 5709static vdev_t * 5710spa_vdev_resilver_done_hunt(vdev_t *vd) 5711{ 5712 vdev_t *newvd, *oldvd; 5713 5714 for (int c = 0; c < vd->vdev_children; c++) { 5715 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5716 if (oldvd != NULL) 5717 return (oldvd); 5718 } 5719 5720 /* 5721 * Check for a completed replacement. We always consider the first 5722 * vdev in the list to be the oldest vdev, and the last one to be 5723 * the newest (see spa_vdev_attach() for how that works). In 5724 * the case where the newest vdev is faulted, we will not automatically 5725 * remove it after a resilver completes. This is OK as it will require 5726 * user intervention to determine which disk the admin wishes to keep. 5727 */ 5728 if (vd->vdev_ops == &vdev_replacing_ops) { 5729 ASSERT(vd->vdev_children > 1); 5730 5731 newvd = vd->vdev_child[vd->vdev_children - 1]; 5732 oldvd = vd->vdev_child[0]; 5733 5734 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5735 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5736 !vdev_dtl_required(oldvd)) 5737 return (oldvd); 5738 } 5739 5740 /* 5741 * Check for a completed resilver with the 'unspare' flag set. 5742 */ 5743 if (vd->vdev_ops == &vdev_spare_ops) { 5744 vdev_t *first = vd->vdev_child[0]; 5745 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5746 5747 if (last->vdev_unspare) { 5748 oldvd = first; 5749 newvd = last; 5750 } else if (first->vdev_unspare) { 5751 oldvd = last; 5752 newvd = first; 5753 } else { 5754 oldvd = NULL; 5755 } 5756 5757 if (oldvd != NULL && 5758 vdev_dtl_empty(newvd, DTL_MISSING) && 5759 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5760 !vdev_dtl_required(oldvd)) 5761 return (oldvd); 5762 5763 /* 5764 * If there are more than two spares attached to a disk, 5765 * and those spares are not required, then we want to 5766 * attempt to free them up now so that they can be used 5767 * by other pools. Once we're back down to a single 5768 * disk+spare, we stop removing them. 5769 */ 5770 if (vd->vdev_children > 2) { 5771 newvd = vd->vdev_child[1]; 5772 5773 if (newvd->vdev_isspare && last->vdev_isspare && 5774 vdev_dtl_empty(last, DTL_MISSING) && 5775 vdev_dtl_empty(last, DTL_OUTAGE) && 5776 !vdev_dtl_required(newvd)) 5777 return (newvd); 5778 } 5779 } 5780 5781 return (NULL); 5782} 5783 5784static void 5785spa_vdev_resilver_done(spa_t *spa) 5786{ 5787 vdev_t *vd, *pvd, *ppvd; 5788 uint64_t guid, sguid, pguid, ppguid; 5789 5790 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5791 5792 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5793 pvd = vd->vdev_parent; 5794 ppvd = pvd->vdev_parent; 5795 guid = vd->vdev_guid; 5796 pguid = pvd->vdev_guid; 5797 ppguid = ppvd->vdev_guid; 5798 sguid = 0; 5799 /* 5800 * If we have just finished replacing a hot spared device, then 5801 * we need to detach the parent's first child (the original hot 5802 * spare) as well. 5803 */ 5804 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5805 ppvd->vdev_children == 2) { 5806 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5807 sguid = ppvd->vdev_child[1]->vdev_guid; 5808 } 5809 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5810 5811 spa_config_exit(spa, SCL_ALL, FTAG); 5812 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5813 return; 5814 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5815 return; 5816 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5817 } 5818 5819 spa_config_exit(spa, SCL_ALL, FTAG); 5820} 5821 5822/* 5823 * Update the stored path or FRU for this vdev. 5824 */ 5825int 5826spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5827 boolean_t ispath) 5828{ 5829 vdev_t *vd; 5830 boolean_t sync = B_FALSE; 5831 5832 ASSERT(spa_writeable(spa)); 5833 5834 spa_vdev_state_enter(spa, SCL_ALL); 5835 5836 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5837 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5838 5839 if (!vd->vdev_ops->vdev_op_leaf) 5840 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5841 5842 if (ispath) { 5843 if (strcmp(value, vd->vdev_path) != 0) { 5844 spa_strfree(vd->vdev_path); 5845 vd->vdev_path = spa_strdup(value); 5846 sync = B_TRUE; 5847 } 5848 } else { 5849 if (vd->vdev_fru == NULL) { 5850 vd->vdev_fru = spa_strdup(value); 5851 sync = B_TRUE; 5852 } else if (strcmp(value, vd->vdev_fru) != 0) { 5853 spa_strfree(vd->vdev_fru); 5854 vd->vdev_fru = spa_strdup(value); 5855 sync = B_TRUE; 5856 } 5857 } 5858 5859 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5860} 5861 5862int 5863spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5864{ 5865 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5866} 5867 5868int 5869spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5870{ 5871 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5872} 5873 5874/* 5875 * ========================================================================== 5876 * SPA Scanning 5877 * ========================================================================== 5878 */ 5879 5880int 5881spa_scan_stop(spa_t *spa) 5882{ 5883 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5884 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5885 return (SET_ERROR(EBUSY)); 5886 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5887} 5888 5889int 5890spa_scan(spa_t *spa, pool_scan_func_t func) 5891{ 5892 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5893 5894 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5895 return (SET_ERROR(ENOTSUP)); 5896 5897 /* 5898 * If a resilver was requested, but there is no DTL on a 5899 * writeable leaf device, we have nothing to do. 5900 */ 5901 if (func == POOL_SCAN_RESILVER && 5902 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5903 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5904 return (0); 5905 } 5906 5907 return (dsl_scan(spa->spa_dsl_pool, func)); 5908} 5909 5910/* 5911 * ========================================================================== 5912 * SPA async task processing 5913 * ========================================================================== 5914 */ 5915 5916static void 5917spa_async_remove(spa_t *spa, vdev_t *vd) 5918{ 5919 if (vd->vdev_remove_wanted) { 5920 vd->vdev_remove_wanted = B_FALSE; 5921 vd->vdev_delayed_close = B_FALSE; 5922 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5923 5924 /* 5925 * We want to clear the stats, but we don't want to do a full 5926 * vdev_clear() as that will cause us to throw away 5927 * degraded/faulted state as well as attempt to reopen the 5928 * device, all of which is a waste. 5929 */ 5930 vd->vdev_stat.vs_read_errors = 0; 5931 vd->vdev_stat.vs_write_errors = 0; 5932 vd->vdev_stat.vs_checksum_errors = 0; 5933 5934 vdev_state_dirty(vd->vdev_top); 5935 } 5936 5937 for (int c = 0; c < vd->vdev_children; c++) 5938 spa_async_remove(spa, vd->vdev_child[c]); 5939} 5940 5941static void 5942spa_async_probe(spa_t *spa, vdev_t *vd) 5943{ 5944 if (vd->vdev_probe_wanted) { 5945 vd->vdev_probe_wanted = B_FALSE; 5946 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5947 } 5948 5949 for (int c = 0; c < vd->vdev_children; c++) 5950 spa_async_probe(spa, vd->vdev_child[c]); 5951} 5952 5953static void 5954spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5955{ 5956 sysevent_id_t eid; 5957 nvlist_t *attr; 5958 char *physpath; 5959 5960 if (!spa->spa_autoexpand) 5961 return; 5962 5963 for (int c = 0; c < vd->vdev_children; c++) { 5964 vdev_t *cvd = vd->vdev_child[c]; 5965 spa_async_autoexpand(spa, cvd); 5966 } 5967 5968 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5969 return; 5970 5971 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5972 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5973 5974 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5975 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5976 5977 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5978 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5979 5980 nvlist_free(attr); 5981 kmem_free(physpath, MAXPATHLEN); 5982} 5983 5984static void 5985spa_async_thread(void *arg) 5986{ 5987 spa_t *spa = arg; 5988 int tasks; 5989 5990 ASSERT(spa->spa_sync_on); 5991 5992 mutex_enter(&spa->spa_async_lock); 5993 tasks = spa->spa_async_tasks; 5994 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5995 mutex_exit(&spa->spa_async_lock); 5996 5997 /* 5998 * See if the config needs to be updated. 5999 */ 6000 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6001 uint64_t old_space, new_space; 6002 6003 mutex_enter(&spa_namespace_lock); 6004 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6005 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6006 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6007 mutex_exit(&spa_namespace_lock); 6008 6009 /* 6010 * If the pool grew as a result of the config update, 6011 * then log an internal history event. 6012 */ 6013 if (new_space != old_space) { 6014 spa_history_log_internal(spa, "vdev online", NULL, 6015 "pool '%s' size: %llu(+%llu)", 6016 spa_name(spa), new_space, new_space - old_space); 6017 } 6018 } 6019 6020 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6021 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6022 spa_async_autoexpand(spa, spa->spa_root_vdev); 6023 spa_config_exit(spa, SCL_CONFIG, FTAG); 6024 } 6025 6026 /* 6027 * See if any devices need to be probed. 6028 */ 6029 if (tasks & SPA_ASYNC_PROBE) { 6030 spa_vdev_state_enter(spa, SCL_NONE); 6031 spa_async_probe(spa, spa->spa_root_vdev); 6032 (void) spa_vdev_state_exit(spa, NULL, 0); 6033 } 6034 6035 /* 6036 * If any devices are done replacing, detach them. 6037 */ 6038 if (tasks & SPA_ASYNC_RESILVER_DONE) 6039 spa_vdev_resilver_done(spa); 6040 6041 /* 6042 * Kick off a resilver. 6043 */ 6044 if (tasks & SPA_ASYNC_RESILVER) 6045 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6046 6047 /* 6048 * Let the world know that we're done. 6049 */ 6050 mutex_enter(&spa->spa_async_lock); 6051 spa->spa_async_thread = NULL; 6052 cv_broadcast(&spa->spa_async_cv); 6053 mutex_exit(&spa->spa_async_lock); 6054 thread_exit(); 6055} 6056 6057static void 6058spa_async_thread_vd(void *arg) 6059{ 6060 spa_t *spa = arg; 6061 int tasks; 6062 6063 ASSERT(spa->spa_sync_on); 6064 6065 mutex_enter(&spa->spa_async_lock); 6066 tasks = spa->spa_async_tasks; 6067retry: 6068 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6069 mutex_exit(&spa->spa_async_lock); 6070 6071 /* 6072 * See if any devices need to be marked REMOVED. 6073 */ 6074 if (tasks & SPA_ASYNC_REMOVE) { 6075 spa_vdev_state_enter(spa, SCL_NONE); 6076 spa_async_remove(spa, spa->spa_root_vdev); 6077 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6078 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6079 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6080 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6081 (void) spa_vdev_state_exit(spa, NULL, 0); 6082 } 6083 6084 /* 6085 * Let the world know that we're done. 6086 */ 6087 mutex_enter(&spa->spa_async_lock); 6088 tasks = spa->spa_async_tasks; 6089 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6090 goto retry; 6091 spa->spa_async_thread_vd = NULL; 6092 cv_broadcast(&spa->spa_async_cv); 6093 mutex_exit(&spa->spa_async_lock); 6094 thread_exit(); 6095} 6096 6097void 6098spa_async_suspend(spa_t *spa) 6099{ 6100 mutex_enter(&spa->spa_async_lock); 6101 spa->spa_async_suspended++; 6102 while (spa->spa_async_thread != NULL && 6103 spa->spa_async_thread_vd != NULL) 6104 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6105 mutex_exit(&spa->spa_async_lock); 6106} 6107 6108void 6109spa_async_resume(spa_t *spa) 6110{ 6111 mutex_enter(&spa->spa_async_lock); 6112 ASSERT(spa->spa_async_suspended != 0); 6113 spa->spa_async_suspended--; 6114 mutex_exit(&spa->spa_async_lock); 6115} 6116 6117static boolean_t 6118spa_async_tasks_pending(spa_t *spa) 6119{ 6120 uint_t non_config_tasks; 6121 uint_t config_task; 6122 boolean_t config_task_suspended; 6123 6124 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6125 SPA_ASYNC_REMOVE); 6126 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6127 if (spa->spa_ccw_fail_time == 0) { 6128 config_task_suspended = B_FALSE; 6129 } else { 6130 config_task_suspended = 6131 (gethrtime() - spa->spa_ccw_fail_time) < 6132 (zfs_ccw_retry_interval * NANOSEC); 6133 } 6134 6135 return (non_config_tasks || (config_task && !config_task_suspended)); 6136} 6137 6138static void 6139spa_async_dispatch(spa_t *spa) 6140{ 6141 mutex_enter(&spa->spa_async_lock); 6142 if (spa_async_tasks_pending(spa) && 6143 !spa->spa_async_suspended && 6144 spa->spa_async_thread == NULL && 6145 rootdir != NULL) 6146 spa->spa_async_thread = thread_create(NULL, 0, 6147 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6148 mutex_exit(&spa->spa_async_lock); 6149} 6150 6151static void 6152spa_async_dispatch_vd(spa_t *spa) 6153{ 6154 mutex_enter(&spa->spa_async_lock); 6155 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6156 !spa->spa_async_suspended && 6157 spa->spa_async_thread_vd == NULL && 6158 rootdir != NULL) 6159 spa->spa_async_thread_vd = thread_create(NULL, 0, 6160 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6161 mutex_exit(&spa->spa_async_lock); 6162} 6163 6164void 6165spa_async_request(spa_t *spa, int task) 6166{ 6167 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6168 mutex_enter(&spa->spa_async_lock); 6169 spa->spa_async_tasks |= task; 6170 mutex_exit(&spa->spa_async_lock); 6171 spa_async_dispatch_vd(spa); 6172} 6173 6174/* 6175 * ========================================================================== 6176 * SPA syncing routines 6177 * ========================================================================== 6178 */ 6179 6180static int 6181bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6182{ 6183 bpobj_t *bpo = arg; 6184 bpobj_enqueue(bpo, bp, tx); 6185 return (0); 6186} 6187 6188static int 6189spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6190{ 6191 zio_t *zio = arg; 6192 6193 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6194 BP_GET_PSIZE(bp), zio->io_flags)); 6195 return (0); 6196} 6197 6198/* 6199 * Note: this simple function is not inlined to make it easier to dtrace the 6200 * amount of time spent syncing frees. 6201 */ 6202static void 6203spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6204{ 6205 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6206 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6207 VERIFY(zio_wait(zio) == 0); 6208} 6209 6210/* 6211 * Note: this simple function is not inlined to make it easier to dtrace the 6212 * amount of time spent syncing deferred frees. 6213 */ 6214static void 6215spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6216{ 6217 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6218 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6219 spa_free_sync_cb, zio, tx), ==, 0); 6220 VERIFY0(zio_wait(zio)); 6221} 6222 6223 6224static void 6225spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6226{ 6227 char *packed = NULL; 6228 size_t bufsize; 6229 size_t nvsize = 0; 6230 dmu_buf_t *db; 6231 6232 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6233 6234 /* 6235 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6236 * information. This avoids the dmu_buf_will_dirty() path and 6237 * saves us a pre-read to get data we don't actually care about. 6238 */ 6239 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6240 packed = kmem_alloc(bufsize, KM_SLEEP); 6241 6242 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6243 KM_SLEEP) == 0); 6244 bzero(packed + nvsize, bufsize - nvsize); 6245 6246 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6247 6248 kmem_free(packed, bufsize); 6249 6250 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6251 dmu_buf_will_dirty(db, tx); 6252 *(uint64_t *)db->db_data = nvsize; 6253 dmu_buf_rele(db, FTAG); 6254} 6255 6256static void 6257spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6258 const char *config, const char *entry) 6259{ 6260 nvlist_t *nvroot; 6261 nvlist_t **list; 6262 int i; 6263 6264 if (!sav->sav_sync) 6265 return; 6266 6267 /* 6268 * Update the MOS nvlist describing the list of available devices. 6269 * spa_validate_aux() will have already made sure this nvlist is 6270 * valid and the vdevs are labeled appropriately. 6271 */ 6272 if (sav->sav_object == 0) { 6273 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6274 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6275 sizeof (uint64_t), tx); 6276 VERIFY(zap_update(spa->spa_meta_objset, 6277 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6278 &sav->sav_object, tx) == 0); 6279 } 6280 6281 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6282 if (sav->sav_count == 0) { 6283 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6284 } else { 6285 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6286 for (i = 0; i < sav->sav_count; i++) 6287 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6288 B_FALSE, VDEV_CONFIG_L2CACHE); 6289 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6290 sav->sav_count) == 0); 6291 for (i = 0; i < sav->sav_count; i++) 6292 nvlist_free(list[i]); 6293 kmem_free(list, sav->sav_count * sizeof (void *)); 6294 } 6295 6296 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6297 nvlist_free(nvroot); 6298 6299 sav->sav_sync = B_FALSE; 6300} 6301 6302static void 6303spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6304{ 6305 nvlist_t *config; 6306 6307 if (list_is_empty(&spa->spa_config_dirty_list)) 6308 return; 6309 6310 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6311 6312 config = spa_config_generate(spa, spa->spa_root_vdev, 6313 dmu_tx_get_txg(tx), B_FALSE); 6314 6315 /* 6316 * If we're upgrading the spa version then make sure that 6317 * the config object gets updated with the correct version. 6318 */ 6319 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6320 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6321 spa->spa_uberblock.ub_version); 6322 6323 spa_config_exit(spa, SCL_STATE, FTAG); 6324 6325 if (spa->spa_config_syncing) 6326 nvlist_free(spa->spa_config_syncing); 6327 spa->spa_config_syncing = config; 6328 6329 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6330} 6331 6332static void 6333spa_sync_version(void *arg, dmu_tx_t *tx) 6334{ 6335 uint64_t *versionp = arg; 6336 uint64_t version = *versionp; 6337 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6338 6339 /* 6340 * Setting the version is special cased when first creating the pool. 6341 */ 6342 ASSERT(tx->tx_txg != TXG_INITIAL); 6343 6344 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6345 ASSERT(version >= spa_version(spa)); 6346 6347 spa->spa_uberblock.ub_version = version; 6348 vdev_config_dirty(spa->spa_root_vdev); 6349 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6350} 6351 6352/* 6353 * Set zpool properties. 6354 */ 6355static void 6356spa_sync_props(void *arg, dmu_tx_t *tx) 6357{ 6358 nvlist_t *nvp = arg; 6359 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6360 objset_t *mos = spa->spa_meta_objset; 6361 nvpair_t *elem = NULL; 6362 6363 mutex_enter(&spa->spa_props_lock); 6364 6365 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6366 uint64_t intval; 6367 char *strval, *fname; 6368 zpool_prop_t prop; 6369 const char *propname; 6370 zprop_type_t proptype; 6371 spa_feature_t fid; 6372 6373 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6374 case ZPROP_INVAL: 6375 /* 6376 * We checked this earlier in spa_prop_validate(). 6377 */ 6378 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6379 6380 fname = strchr(nvpair_name(elem), '@') + 1; 6381 VERIFY0(zfeature_lookup_name(fname, &fid)); 6382 6383 spa_feature_enable(spa, fid, tx); 6384 spa_history_log_internal(spa, "set", tx, 6385 "%s=enabled", nvpair_name(elem)); 6386 break; 6387 6388 case ZPOOL_PROP_VERSION: 6389 intval = fnvpair_value_uint64(elem); 6390 /* 6391 * The version is synced seperatly before other 6392 * properties and should be correct by now. 6393 */ 6394 ASSERT3U(spa_version(spa), >=, intval); 6395 break; 6396 6397 case ZPOOL_PROP_ALTROOT: 6398 /* 6399 * 'altroot' is a non-persistent property. It should 6400 * have been set temporarily at creation or import time. 6401 */ 6402 ASSERT(spa->spa_root != NULL); 6403 break; 6404 6405 case ZPOOL_PROP_READONLY: 6406 case ZPOOL_PROP_CACHEFILE: 6407 /* 6408 * 'readonly' and 'cachefile' are also non-persisitent 6409 * properties. 6410 */ 6411 break; 6412 case ZPOOL_PROP_COMMENT: 6413 strval = fnvpair_value_string(elem); 6414 if (spa->spa_comment != NULL) 6415 spa_strfree(spa->spa_comment); 6416 spa->spa_comment = spa_strdup(strval); 6417 /* 6418 * We need to dirty the configuration on all the vdevs 6419 * so that their labels get updated. It's unnecessary 6420 * to do this for pool creation since the vdev's 6421 * configuratoin has already been dirtied. 6422 */ 6423 if (tx->tx_txg != TXG_INITIAL) 6424 vdev_config_dirty(spa->spa_root_vdev); 6425 spa_history_log_internal(spa, "set", tx, 6426 "%s=%s", nvpair_name(elem), strval); 6427 break; 6428 default: 6429 /* 6430 * Set pool property values in the poolprops mos object. 6431 */ 6432 if (spa->spa_pool_props_object == 0) { 6433 spa->spa_pool_props_object = 6434 zap_create_link(mos, DMU_OT_POOL_PROPS, 6435 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6436 tx); 6437 } 6438 6439 /* normalize the property name */ 6440 propname = zpool_prop_to_name(prop); 6441 proptype = zpool_prop_get_type(prop); 6442 6443 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6444 ASSERT(proptype == PROP_TYPE_STRING); 6445 strval = fnvpair_value_string(elem); 6446 VERIFY0(zap_update(mos, 6447 spa->spa_pool_props_object, propname, 6448 1, strlen(strval) + 1, strval, tx)); 6449 spa_history_log_internal(spa, "set", tx, 6450 "%s=%s", nvpair_name(elem), strval); 6451 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6452 intval = fnvpair_value_uint64(elem); 6453 6454 if (proptype == PROP_TYPE_INDEX) { 6455 const char *unused; 6456 VERIFY0(zpool_prop_index_to_string( 6457 prop, intval, &unused)); 6458 } 6459 VERIFY0(zap_update(mos, 6460 spa->spa_pool_props_object, propname, 6461 8, 1, &intval, tx)); 6462 spa_history_log_internal(spa, "set", tx, 6463 "%s=%lld", nvpair_name(elem), intval); 6464 } else { 6465 ASSERT(0); /* not allowed */ 6466 } 6467 6468 switch (prop) { 6469 case ZPOOL_PROP_DELEGATION: 6470 spa->spa_delegation = intval; 6471 break; 6472 case ZPOOL_PROP_BOOTFS: 6473 spa->spa_bootfs = intval; 6474 break; 6475 case ZPOOL_PROP_FAILUREMODE: 6476 spa->spa_failmode = intval; 6477 break; 6478 case ZPOOL_PROP_AUTOEXPAND: 6479 spa->spa_autoexpand = intval; 6480 if (tx->tx_txg != TXG_INITIAL) 6481 spa_async_request(spa, 6482 SPA_ASYNC_AUTOEXPAND); 6483 break; 6484 case ZPOOL_PROP_DEDUPDITTO: 6485 spa->spa_dedup_ditto = intval; 6486 break; 6487 default: 6488 break; 6489 } 6490 } 6491 6492 } 6493 6494 mutex_exit(&spa->spa_props_lock); 6495} 6496 6497/* 6498 * Perform one-time upgrade on-disk changes. spa_version() does not 6499 * reflect the new version this txg, so there must be no changes this 6500 * txg to anything that the upgrade code depends on after it executes. 6501 * Therefore this must be called after dsl_pool_sync() does the sync 6502 * tasks. 6503 */ 6504static void 6505spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6506{ 6507 dsl_pool_t *dp = spa->spa_dsl_pool; 6508 6509 ASSERT(spa->spa_sync_pass == 1); 6510 6511 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6512 6513 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6514 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6515 dsl_pool_create_origin(dp, tx); 6516 6517 /* Keeping the origin open increases spa_minref */ 6518 spa->spa_minref += 3; 6519 } 6520 6521 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6522 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6523 dsl_pool_upgrade_clones(dp, tx); 6524 } 6525 6526 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6527 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6528 dsl_pool_upgrade_dir_clones(dp, tx); 6529 6530 /* Keeping the freedir open increases spa_minref */ 6531 spa->spa_minref += 3; 6532 } 6533 6534 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6535 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6536 spa_feature_create_zap_objects(spa, tx); 6537 } 6538 6539 /* 6540 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6541 * when possibility to use lz4 compression for metadata was added 6542 * Old pools that have this feature enabled must be upgraded to have 6543 * this feature active 6544 */ 6545 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6546 boolean_t lz4_en = spa_feature_is_enabled(spa, 6547 SPA_FEATURE_LZ4_COMPRESS); 6548 boolean_t lz4_ac = spa_feature_is_active(spa, 6549 SPA_FEATURE_LZ4_COMPRESS); 6550 6551 if (lz4_en && !lz4_ac) 6552 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6553 } 6554 rrw_exit(&dp->dp_config_rwlock, FTAG); 6555} 6556 6557/* 6558 * Sync the specified transaction group. New blocks may be dirtied as 6559 * part of the process, so we iterate until it converges. 6560 */ 6561void 6562spa_sync(spa_t *spa, uint64_t txg) 6563{ 6564 dsl_pool_t *dp = spa->spa_dsl_pool; 6565 objset_t *mos = spa->spa_meta_objset; 6566 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6567 vdev_t *rvd = spa->spa_root_vdev; 6568 vdev_t *vd; 6569 dmu_tx_t *tx; 6570 int error; 6571 6572 VERIFY(spa_writeable(spa)); 6573 6574 /* 6575 * Lock out configuration changes. 6576 */ 6577 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6578 6579 spa->spa_syncing_txg = txg; 6580 spa->spa_sync_pass = 0; 6581 6582 /* 6583 * If there are any pending vdev state changes, convert them 6584 * into config changes that go out with this transaction group. 6585 */ 6586 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6587 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6588 /* 6589 * We need the write lock here because, for aux vdevs, 6590 * calling vdev_config_dirty() modifies sav_config. 6591 * This is ugly and will become unnecessary when we 6592 * eliminate the aux vdev wart by integrating all vdevs 6593 * into the root vdev tree. 6594 */ 6595 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6596 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6597 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6598 vdev_state_clean(vd); 6599 vdev_config_dirty(vd); 6600 } 6601 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6602 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6603 } 6604 spa_config_exit(spa, SCL_STATE, FTAG); 6605 6606 tx = dmu_tx_create_assigned(dp, txg); 6607 6608 spa->spa_sync_starttime = gethrtime(); 6609#ifdef illumos 6610 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6611 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6612#else /* !illumos */ 6613#ifdef _KERNEL 6614 callout_reset(&spa->spa_deadman_cycid, 6615 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6616#endif 6617#endif /* illumos */ 6618 6619 /* 6620 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6621 * set spa_deflate if we have no raid-z vdevs. 6622 */ 6623 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6624 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6625 int i; 6626 6627 for (i = 0; i < rvd->vdev_children; i++) { 6628 vd = rvd->vdev_child[i]; 6629 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6630 break; 6631 } 6632 if (i == rvd->vdev_children) { 6633 spa->spa_deflate = TRUE; 6634 VERIFY(0 == zap_add(spa->spa_meta_objset, 6635 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6636 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6637 } 6638 } 6639 6640 /* 6641 * Iterate to convergence. 6642 */ 6643 do { 6644 int pass = ++spa->spa_sync_pass; 6645 6646 spa_sync_config_object(spa, tx); 6647 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6648 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6649 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6650 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6651 spa_errlog_sync(spa, txg); 6652 dsl_pool_sync(dp, txg); 6653 6654 if (pass < zfs_sync_pass_deferred_free) { 6655 spa_sync_frees(spa, free_bpl, tx); 6656 } else { 6657 /* 6658 * We can not defer frees in pass 1, because 6659 * we sync the deferred frees later in pass 1. 6660 */ 6661 ASSERT3U(pass, >, 1); 6662 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6663 &spa->spa_deferred_bpobj, tx); 6664 } 6665 6666 ddt_sync(spa, txg); 6667 dsl_scan_sync(dp, tx); 6668 6669 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6670 vdev_sync(vd, txg); 6671 6672 if (pass == 1) { 6673 spa_sync_upgrades(spa, tx); 6674 ASSERT3U(txg, >=, 6675 spa->spa_uberblock.ub_rootbp.blk_birth); 6676 /* 6677 * Note: We need to check if the MOS is dirty 6678 * because we could have marked the MOS dirty 6679 * without updating the uberblock (e.g. if we 6680 * have sync tasks but no dirty user data). We 6681 * need to check the uberblock's rootbp because 6682 * it is updated if we have synced out dirty 6683 * data (though in this case the MOS will most 6684 * likely also be dirty due to second order 6685 * effects, we don't want to rely on that here). 6686 */ 6687 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6688 !dmu_objset_is_dirty(mos, txg)) { 6689 /* 6690 * Nothing changed on the first pass, 6691 * therefore this TXG is a no-op. Avoid 6692 * syncing deferred frees, so that we 6693 * can keep this TXG as a no-op. 6694 */ 6695 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6696 txg)); 6697 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6698 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6699 break; 6700 } 6701 spa_sync_deferred_frees(spa, tx); 6702 } 6703 6704 } while (dmu_objset_is_dirty(mos, txg)); 6705 6706 /* 6707 * Rewrite the vdev configuration (which includes the uberblock) 6708 * to commit the transaction group. 6709 * 6710 * If there are no dirty vdevs, we sync the uberblock to a few 6711 * random top-level vdevs that are known to be visible in the 6712 * config cache (see spa_vdev_add() for a complete description). 6713 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6714 */ 6715 for (;;) { 6716 /* 6717 * We hold SCL_STATE to prevent vdev open/close/etc. 6718 * while we're attempting to write the vdev labels. 6719 */ 6720 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6721 6722 if (list_is_empty(&spa->spa_config_dirty_list)) { 6723 vdev_t *svd[SPA_DVAS_PER_BP]; 6724 int svdcount = 0; 6725 int children = rvd->vdev_children; 6726 int c0 = spa_get_random(children); 6727 6728 for (int c = 0; c < children; c++) { 6729 vd = rvd->vdev_child[(c0 + c) % children]; 6730 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6731 continue; 6732 svd[svdcount++] = vd; 6733 if (svdcount == SPA_DVAS_PER_BP) 6734 break; 6735 } 6736 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6737 if (error != 0) 6738 error = vdev_config_sync(svd, svdcount, txg, 6739 B_TRUE); 6740 } else { 6741 error = vdev_config_sync(rvd->vdev_child, 6742 rvd->vdev_children, txg, B_FALSE); 6743 if (error != 0) 6744 error = vdev_config_sync(rvd->vdev_child, 6745 rvd->vdev_children, txg, B_TRUE); 6746 } 6747 6748 if (error == 0) 6749 spa->spa_last_synced_guid = rvd->vdev_guid; 6750 6751 spa_config_exit(spa, SCL_STATE, FTAG); 6752 6753 if (error == 0) 6754 break; 6755 zio_suspend(spa, NULL); 6756 zio_resume_wait(spa); 6757 } 6758 dmu_tx_commit(tx); 6759 6760#ifdef illumos 6761 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6762#else /* !illumos */ 6763#ifdef _KERNEL 6764 callout_drain(&spa->spa_deadman_cycid); 6765#endif 6766#endif /* illumos */ 6767 6768 /* 6769 * Clear the dirty config list. 6770 */ 6771 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6772 vdev_config_clean(vd); 6773 6774 /* 6775 * Now that the new config has synced transactionally, 6776 * let it become visible to the config cache. 6777 */ 6778 if (spa->spa_config_syncing != NULL) { 6779 spa_config_set(spa, spa->spa_config_syncing); 6780 spa->spa_config_txg = txg; 6781 spa->spa_config_syncing = NULL; 6782 } 6783 6784 spa->spa_ubsync = spa->spa_uberblock; 6785 6786 dsl_pool_sync_done(dp, txg); 6787 6788 /* 6789 * Update usable space statistics. 6790 */ 6791 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6792 vdev_sync_done(vd, txg); 6793 6794 spa_update_dspace(spa); 6795 6796 /* 6797 * It had better be the case that we didn't dirty anything 6798 * since vdev_config_sync(). 6799 */ 6800 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6801 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6802 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6803 6804 spa->spa_sync_pass = 0; 6805 6806 spa_config_exit(spa, SCL_CONFIG, FTAG); 6807 6808 spa_handle_ignored_writes(spa); 6809 6810 /* 6811 * If any async tasks have been requested, kick them off. 6812 */ 6813 spa_async_dispatch(spa); 6814 spa_async_dispatch_vd(spa); 6815} 6816 6817/* 6818 * Sync all pools. We don't want to hold the namespace lock across these 6819 * operations, so we take a reference on the spa_t and drop the lock during the 6820 * sync. 6821 */ 6822void 6823spa_sync_allpools(void) 6824{ 6825 spa_t *spa = NULL; 6826 mutex_enter(&spa_namespace_lock); 6827 while ((spa = spa_next(spa)) != NULL) { 6828 if (spa_state(spa) != POOL_STATE_ACTIVE || 6829 !spa_writeable(spa) || spa_suspended(spa)) 6830 continue; 6831 spa_open_ref(spa, FTAG); 6832 mutex_exit(&spa_namespace_lock); 6833 txg_wait_synced(spa_get_dsl(spa), 0); 6834 mutex_enter(&spa_namespace_lock); 6835 spa_close(spa, FTAG); 6836 } 6837 mutex_exit(&spa_namespace_lock); 6838} 6839 6840/* 6841 * ========================================================================== 6842 * Miscellaneous routines 6843 * ========================================================================== 6844 */ 6845 6846/* 6847 * Remove all pools in the system. 6848 */ 6849void 6850spa_evict_all(void) 6851{ 6852 spa_t *spa; 6853 6854 /* 6855 * Remove all cached state. All pools should be closed now, 6856 * so every spa in the AVL tree should be unreferenced. 6857 */ 6858 mutex_enter(&spa_namespace_lock); 6859 while ((spa = spa_next(NULL)) != NULL) { 6860 /* 6861 * Stop async tasks. The async thread may need to detach 6862 * a device that's been replaced, which requires grabbing 6863 * spa_namespace_lock, so we must drop it here. 6864 */ 6865 spa_open_ref(spa, FTAG); 6866 mutex_exit(&spa_namespace_lock); 6867 spa_async_suspend(spa); 6868 mutex_enter(&spa_namespace_lock); 6869 spa_close(spa, FTAG); 6870 6871 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6872 spa_unload(spa); 6873 spa_deactivate(spa); 6874 } 6875 spa_remove(spa); 6876 } 6877 mutex_exit(&spa_namespace_lock); 6878} 6879 6880vdev_t * 6881spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6882{ 6883 vdev_t *vd; 6884 int i; 6885 6886 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6887 return (vd); 6888 6889 if (aux) { 6890 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6891 vd = spa->spa_l2cache.sav_vdevs[i]; 6892 if (vd->vdev_guid == guid) 6893 return (vd); 6894 } 6895 6896 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6897 vd = spa->spa_spares.sav_vdevs[i]; 6898 if (vd->vdev_guid == guid) 6899 return (vd); 6900 } 6901 } 6902 6903 return (NULL); 6904} 6905 6906void 6907spa_upgrade(spa_t *spa, uint64_t version) 6908{ 6909 ASSERT(spa_writeable(spa)); 6910 6911 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6912 6913 /* 6914 * This should only be called for a non-faulted pool, and since a 6915 * future version would result in an unopenable pool, this shouldn't be 6916 * possible. 6917 */ 6918 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6919 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6920 6921 spa->spa_uberblock.ub_version = version; 6922 vdev_config_dirty(spa->spa_root_vdev); 6923 6924 spa_config_exit(spa, SCL_ALL, FTAG); 6925 6926 txg_wait_synced(spa_get_dsl(spa), 0); 6927} 6928 6929boolean_t 6930spa_has_spare(spa_t *spa, uint64_t guid) 6931{ 6932 int i; 6933 uint64_t spareguid; 6934 spa_aux_vdev_t *sav = &spa->spa_spares; 6935 6936 for (i = 0; i < sav->sav_count; i++) 6937 if (sav->sav_vdevs[i]->vdev_guid == guid) 6938 return (B_TRUE); 6939 6940 for (i = 0; i < sav->sav_npending; i++) { 6941 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6942 &spareguid) == 0 && spareguid == guid) 6943 return (B_TRUE); 6944 } 6945 6946 return (B_FALSE); 6947} 6948 6949/* 6950 * Check if a pool has an active shared spare device. 6951 * Note: reference count of an active spare is 2, as a spare and as a replace 6952 */ 6953static boolean_t 6954spa_has_active_shared_spare(spa_t *spa) 6955{ 6956 int i, refcnt; 6957 uint64_t pool; 6958 spa_aux_vdev_t *sav = &spa->spa_spares; 6959 6960 for (i = 0; i < sav->sav_count; i++) { 6961 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6962 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6963 refcnt > 2) 6964 return (B_TRUE); 6965 } 6966 6967 return (B_FALSE); 6968} 6969 6970/* 6971 * Post a sysevent corresponding to the given event. The 'name' must be one of 6972 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6973 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6974 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6975 * or zdb as real changes. 6976 */ 6977void 6978spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6979{ 6980#ifdef _KERNEL 6981 sysevent_t *ev; 6982 sysevent_attr_list_t *attr = NULL; 6983 sysevent_value_t value; 6984 sysevent_id_t eid; 6985 6986 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6987 SE_SLEEP); 6988 6989 value.value_type = SE_DATA_TYPE_STRING; 6990 value.value.sv_string = spa_name(spa); 6991 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6992 goto done; 6993 6994 value.value_type = SE_DATA_TYPE_UINT64; 6995 value.value.sv_uint64 = spa_guid(spa); 6996 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6997 goto done; 6998 6999 if (vd) { 7000 value.value_type = SE_DATA_TYPE_UINT64; 7001 value.value.sv_uint64 = vd->vdev_guid; 7002 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7003 SE_SLEEP) != 0) 7004 goto done; 7005 7006 if (vd->vdev_path) { 7007 value.value_type = SE_DATA_TYPE_STRING; 7008 value.value.sv_string = vd->vdev_path; 7009 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7010 &value, SE_SLEEP) != 0) 7011 goto done; 7012 } 7013 } 7014 7015 if (sysevent_attach_attributes(ev, attr) != 0) 7016 goto done; 7017 attr = NULL; 7018 7019 (void) log_sysevent(ev, SE_SLEEP, &eid); 7020 7021done: 7022 if (attr) 7023 sysevent_free_attr(attr); 7024 sysevent_free(ev); 7025#endif 7026} 7027