spa.c revision 276081
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) 2013 by Delphix. All rights reserved. 25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 */ 28 29/* 30 * SPA: Storage Pool Allocator 31 * 32 * This file contains all the routines used when modifying on-disk SPA state. 33 * This includes opening, importing, destroying, exporting a pool, and syncing a 34 * pool. 35 */ 36 37#include <sys/zfs_context.h> 38#include <sys/fm/fs/zfs.h> 39#include <sys/spa_impl.h> 40#include <sys/zio.h> 41#include <sys/zio_checksum.h> 42#include <sys/dmu.h> 43#include <sys/dmu_tx.h> 44#include <sys/zap.h> 45#include <sys/zil.h> 46#include <sys/ddt.h> 47#include <sys/vdev_impl.h> 48#include <sys/metaslab.h> 49#include <sys/metaslab_impl.h> 50#include <sys/uberblock_impl.h> 51#include <sys/txg.h> 52#include <sys/avl.h> 53#include <sys/dmu_traverse.h> 54#include <sys/dmu_objset.h> 55#include <sys/unique.h> 56#include <sys/dsl_pool.h> 57#include <sys/dsl_dataset.h> 58#include <sys/dsl_dir.h> 59#include <sys/dsl_prop.h> 60#include <sys/dsl_synctask.h> 61#include <sys/fs/zfs.h> 62#include <sys/arc.h> 63#include <sys/callb.h> 64#include <sys/spa_boot.h> 65#include <sys/zfs_ioctl.h> 66#include <sys/dsl_scan.h> 67#include <sys/dmu_send.h> 68#include <sys/dsl_destroy.h> 69#include <sys/dsl_userhold.h> 70#include <sys/zfeature.h> 71#include <sys/zvol.h> 72#include <sys/trim_map.h> 73 74#ifdef _KERNEL 75#include <sys/callb.h> 76#include <sys/cpupart.h> 77#include <sys/zone.h> 78#endif /* _KERNEL */ 79 80#include "zfs_prop.h" 81#include "zfs_comutil.h" 82 83/* Check hostid on import? */ 84static int check_hostid = 1; 85 86SYSCTL_DECL(_vfs_zfs); 87TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 88SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 89 "Check hostid on import?"); 90 91/* 92 * The interval, in seconds, at which failed configuration cache file writes 93 * should be retried. 94 */ 95static int zfs_ccw_retry_interval = 300; 96 97typedef enum zti_modes { 98 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 99 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 100 ZTI_MODE_NULL, /* don't create a taskq */ 101 ZTI_NMODES 102} zti_modes_t; 103 104#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 105#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 106#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 107 108#define ZTI_N(n) ZTI_P(n, 1) 109#define ZTI_ONE ZTI_N(1) 110 111typedef struct zio_taskq_info { 112 zti_modes_t zti_mode; 113 uint_t zti_value; 114 uint_t zti_count; 115} zio_taskq_info_t; 116 117static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 118 "issue", "issue_high", "intr", "intr_high" 119}; 120 121/* 122 * This table defines the taskq settings for each ZFS I/O type. When 123 * initializing a pool, we use this table to create an appropriately sized 124 * taskq. Some operations are low volume and therefore have a small, static 125 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 126 * macros. Other operations process a large amount of data; the ZTI_BATCH 127 * macro causes us to create a taskq oriented for throughput. Some operations 128 * are so high frequency and short-lived that the taskq itself can become a a 129 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 130 * additional degree of parallelism specified by the number of threads per- 131 * taskq and the number of taskqs; when dispatching an event in this case, the 132 * particular taskq is chosen at random. 133 * 134 * The different taskq priorities are to handle the different contexts (issue 135 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 136 * need to be handled with minimum delay. 137 */ 138const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 139 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 140 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 141 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */ 142 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 143 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 146}; 147 148static void spa_sync_version(void *arg, dmu_tx_t *tx); 149static void spa_sync_props(void *arg, dmu_tx_t *tx); 150static boolean_t spa_has_active_shared_spare(spa_t *spa); 151static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 152 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 153 char **ereport); 154static void spa_vdev_resilver_done(spa_t *spa); 155 156uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 157#ifdef PSRSET_BIND 158id_t zio_taskq_psrset_bind = PS_NONE; 159#endif 160#ifdef SYSDC 161boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 162#endif 163uint_t zio_taskq_basedc = 80; /* base duty cycle */ 164 165boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 166extern int zfs_sync_pass_deferred_free; 167 168#ifndef illumos 169extern void spa_deadman(void *arg); 170#endif 171 172/* 173 * This (illegal) pool name is used when temporarily importing a spa_t in order 174 * to get the vdev stats associated with the imported devices. 175 */ 176#define TRYIMPORT_NAME "$import" 177 178/* 179 * ========================================================================== 180 * SPA properties routines 181 * ========================================================================== 182 */ 183 184/* 185 * Add a (source=src, propname=propval) list to an nvlist. 186 */ 187static void 188spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 189 uint64_t intval, zprop_source_t src) 190{ 191 const char *propname = zpool_prop_to_name(prop); 192 nvlist_t *propval; 193 194 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 195 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 196 197 if (strval != NULL) 198 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 199 else 200 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 201 202 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 203 nvlist_free(propval); 204} 205 206/* 207 * Get property values from the spa configuration. 208 */ 209static void 210spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 211{ 212 vdev_t *rvd = spa->spa_root_vdev; 213 dsl_pool_t *pool = spa->spa_dsl_pool; 214 uint64_t size, alloc, cap, version; 215 zprop_source_t src = ZPROP_SRC_NONE; 216 spa_config_dirent_t *dp; 217 metaslab_class_t *mc = spa_normal_class(spa); 218 219 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 220 221 if (rvd != NULL) { 222 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 223 size = metaslab_class_get_space(spa_normal_class(spa)); 224 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 225 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 226 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 227 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 228 size - alloc, src); 229 230 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 231 metaslab_class_fragmentation(mc), src); 232 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 233 metaslab_class_expandable_space(mc), src); 234 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 235 (spa_mode(spa) == FREAD), src); 236 237 cap = (size == 0) ? 0 : (alloc * 100 / size); 238 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 239 240 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 241 ddt_get_pool_dedup_ratio(spa), src); 242 243 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 244 rvd->vdev_state, src); 245 246 version = spa_version(spa); 247 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 248 src = ZPROP_SRC_DEFAULT; 249 else 250 src = ZPROP_SRC_LOCAL; 251 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 252 } 253 254 if (pool != NULL) { 255 /* 256 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 257 * when opening pools before this version freedir will be NULL. 258 */ 259 if (pool->dp_free_dir != NULL) { 260 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 261 pool->dp_free_dir->dd_phys->dd_used_bytes, src); 262 } else { 263 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 264 NULL, 0, src); 265 } 266 267 if (pool->dp_leak_dir != NULL) { 268 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 269 pool->dp_leak_dir->dd_phys->dd_used_bytes, src); 270 } else { 271 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 272 NULL, 0, src); 273 } 274 } 275 276 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 277 278 if (spa->spa_comment != NULL) { 279 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 280 0, ZPROP_SRC_LOCAL); 281 } 282 283 if (spa->spa_root != NULL) 284 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 285 0, ZPROP_SRC_LOCAL); 286 287 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 288 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 289 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 290 } else { 291 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 292 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 293 } 294 295 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 296 if (dp->scd_path == NULL) { 297 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 298 "none", 0, ZPROP_SRC_LOCAL); 299 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 300 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 301 dp->scd_path, 0, ZPROP_SRC_LOCAL); 302 } 303 } 304} 305 306/* 307 * Get zpool property values. 308 */ 309int 310spa_prop_get(spa_t *spa, nvlist_t **nvp) 311{ 312 objset_t *mos = spa->spa_meta_objset; 313 zap_cursor_t zc; 314 zap_attribute_t za; 315 int err; 316 317 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 318 319 mutex_enter(&spa->spa_props_lock); 320 321 /* 322 * Get properties from the spa config. 323 */ 324 spa_prop_get_config(spa, nvp); 325 326 /* If no pool property object, no more prop to get. */ 327 if (mos == NULL || spa->spa_pool_props_object == 0) { 328 mutex_exit(&spa->spa_props_lock); 329 return (0); 330 } 331 332 /* 333 * Get properties from the MOS pool property object. 334 */ 335 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 336 (err = zap_cursor_retrieve(&zc, &za)) == 0; 337 zap_cursor_advance(&zc)) { 338 uint64_t intval = 0; 339 char *strval = NULL; 340 zprop_source_t src = ZPROP_SRC_DEFAULT; 341 zpool_prop_t prop; 342 343 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 344 continue; 345 346 switch (za.za_integer_length) { 347 case 8: 348 /* integer property */ 349 if (za.za_first_integer != 350 zpool_prop_default_numeric(prop)) 351 src = ZPROP_SRC_LOCAL; 352 353 if (prop == ZPOOL_PROP_BOOTFS) { 354 dsl_pool_t *dp; 355 dsl_dataset_t *ds = NULL; 356 357 dp = spa_get_dsl(spa); 358 dsl_pool_config_enter(dp, FTAG); 359 if (err = dsl_dataset_hold_obj(dp, 360 za.za_first_integer, FTAG, &ds)) { 361 dsl_pool_config_exit(dp, FTAG); 362 break; 363 } 364 365 strval = kmem_alloc( 366 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 367 KM_SLEEP); 368 dsl_dataset_name(ds, strval); 369 dsl_dataset_rele(ds, FTAG); 370 dsl_pool_config_exit(dp, FTAG); 371 } else { 372 strval = NULL; 373 intval = za.za_first_integer; 374 } 375 376 spa_prop_add_list(*nvp, prop, strval, intval, src); 377 378 if (strval != NULL) 379 kmem_free(strval, 380 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 381 382 break; 383 384 case 1: 385 /* string property */ 386 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 387 err = zap_lookup(mos, spa->spa_pool_props_object, 388 za.za_name, 1, za.za_num_integers, strval); 389 if (err) { 390 kmem_free(strval, za.za_num_integers); 391 break; 392 } 393 spa_prop_add_list(*nvp, prop, strval, 0, src); 394 kmem_free(strval, za.za_num_integers); 395 break; 396 397 default: 398 break; 399 } 400 } 401 zap_cursor_fini(&zc); 402 mutex_exit(&spa->spa_props_lock); 403out: 404 if (err && err != ENOENT) { 405 nvlist_free(*nvp); 406 *nvp = NULL; 407 return (err); 408 } 409 410 return (0); 411} 412 413/* 414 * Validate the given pool properties nvlist and modify the list 415 * for the property values to be set. 416 */ 417static int 418spa_prop_validate(spa_t *spa, nvlist_t *props) 419{ 420 nvpair_t *elem; 421 int error = 0, reset_bootfs = 0; 422 uint64_t objnum = 0; 423 boolean_t has_feature = B_FALSE; 424 425 elem = NULL; 426 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 427 uint64_t intval; 428 char *strval, *slash, *check, *fname; 429 const char *propname = nvpair_name(elem); 430 zpool_prop_t prop = zpool_name_to_prop(propname); 431 432 switch (prop) { 433 case ZPROP_INVAL: 434 if (!zpool_prop_feature(propname)) { 435 error = SET_ERROR(EINVAL); 436 break; 437 } 438 439 /* 440 * Sanitize the input. 441 */ 442 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 443 error = SET_ERROR(EINVAL); 444 break; 445 } 446 447 if (nvpair_value_uint64(elem, &intval) != 0) { 448 error = SET_ERROR(EINVAL); 449 break; 450 } 451 452 if (intval != 0) { 453 error = SET_ERROR(EINVAL); 454 break; 455 } 456 457 fname = strchr(propname, '@') + 1; 458 if (zfeature_lookup_name(fname, NULL) != 0) { 459 error = SET_ERROR(EINVAL); 460 break; 461 } 462 463 has_feature = B_TRUE; 464 break; 465 466 case ZPOOL_PROP_VERSION: 467 error = nvpair_value_uint64(elem, &intval); 468 if (!error && 469 (intval < spa_version(spa) || 470 intval > SPA_VERSION_BEFORE_FEATURES || 471 has_feature)) 472 error = SET_ERROR(EINVAL); 473 break; 474 475 case ZPOOL_PROP_DELEGATION: 476 case ZPOOL_PROP_AUTOREPLACE: 477 case ZPOOL_PROP_LISTSNAPS: 478 case ZPOOL_PROP_AUTOEXPAND: 479 error = nvpair_value_uint64(elem, &intval); 480 if (!error && intval > 1) 481 error = SET_ERROR(EINVAL); 482 break; 483 484 case ZPOOL_PROP_BOOTFS: 485 /* 486 * If the pool version is less than SPA_VERSION_BOOTFS, 487 * or the pool is still being created (version == 0), 488 * the bootfs property cannot be set. 489 */ 490 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 491 error = SET_ERROR(ENOTSUP); 492 break; 493 } 494 495 /* 496 * Make sure the vdev config is bootable 497 */ 498 if (!vdev_is_bootable(spa->spa_root_vdev)) { 499 error = SET_ERROR(ENOTSUP); 500 break; 501 } 502 503 reset_bootfs = 1; 504 505 error = nvpair_value_string(elem, &strval); 506 507 if (!error) { 508 objset_t *os; 509 uint64_t propval; 510 511 if (strval == NULL || strval[0] == '\0') { 512 objnum = zpool_prop_default_numeric( 513 ZPOOL_PROP_BOOTFS); 514 break; 515 } 516 517 if (error = dmu_objset_hold(strval, FTAG, &os)) 518 break; 519 520 /* 521 * Must be ZPL, and its property settings 522 * must be supported by GRUB (compression 523 * is not gzip, and large blocks are not used). 524 */ 525 526 if (dmu_objset_type(os) != DMU_OST_ZFS) { 527 error = SET_ERROR(ENOTSUP); 528 } else if ((error = 529 dsl_prop_get_int_ds(dmu_objset_ds(os), 530 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 531 &propval)) == 0 && 532 !BOOTFS_COMPRESS_VALID(propval)) { 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_RECORDSIZE), 537 &propval)) == 0 && 538 propval > SPA_OLD_MAXBLOCKSIZE) { 539 error = SET_ERROR(ENOTSUP); 540 } else { 541 objnum = dmu_objset_id(os); 542 } 543 dmu_objset_rele(os, FTAG); 544 } 545 break; 546 547 case ZPOOL_PROP_FAILUREMODE: 548 error = nvpair_value_uint64(elem, &intval); 549 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 550 intval > ZIO_FAILURE_MODE_PANIC)) 551 error = SET_ERROR(EINVAL); 552 553 /* 554 * This is a special case which only occurs when 555 * the pool has completely failed. This allows 556 * the user to change the in-core failmode property 557 * without syncing it out to disk (I/Os might 558 * currently be blocked). We do this by returning 559 * EIO to the caller (spa_prop_set) to trick it 560 * into thinking we encountered a property validation 561 * error. 562 */ 563 if (!error && spa_suspended(spa)) { 564 spa->spa_failmode = intval; 565 error = SET_ERROR(EIO); 566 } 567 break; 568 569 case ZPOOL_PROP_CACHEFILE: 570 if ((error = nvpair_value_string(elem, &strval)) != 0) 571 break; 572 573 if (strval[0] == '\0') 574 break; 575 576 if (strcmp(strval, "none") == 0) 577 break; 578 579 if (strval[0] != '/') { 580 error = SET_ERROR(EINVAL); 581 break; 582 } 583 584 slash = strrchr(strval, '/'); 585 ASSERT(slash != NULL); 586 587 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 588 strcmp(slash, "/..") == 0) 589 error = SET_ERROR(EINVAL); 590 break; 591 592 case ZPOOL_PROP_COMMENT: 593 if ((error = nvpair_value_string(elem, &strval)) != 0) 594 break; 595 for (check = strval; *check != '\0'; check++) { 596 /* 597 * The kernel doesn't have an easy isprint() 598 * check. For this kernel check, we merely 599 * check ASCII apart from DEL. Fix this if 600 * there is an easy-to-use kernel isprint(). 601 */ 602 if (*check >= 0x7f) { 603 error = SET_ERROR(EINVAL); 604 break; 605 } 606 check++; 607 } 608 if (strlen(strval) > ZPROP_MAX_COMMENT) 609 error = E2BIG; 610 break; 611 612 case ZPOOL_PROP_DEDUPDITTO: 613 if (spa_version(spa) < SPA_VERSION_DEDUP) 614 error = SET_ERROR(ENOTSUP); 615 else 616 error = nvpair_value_uint64(elem, &intval); 617 if (error == 0 && 618 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 619 error = SET_ERROR(EINVAL); 620 break; 621 } 622 623 if (error) 624 break; 625 } 626 627 if (!error && reset_bootfs) { 628 error = nvlist_remove(props, 629 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 630 631 if (!error) { 632 error = nvlist_add_uint64(props, 633 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 634 } 635 } 636 637 return (error); 638} 639 640void 641spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 642{ 643 char *cachefile; 644 spa_config_dirent_t *dp; 645 646 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 647 &cachefile) != 0) 648 return; 649 650 dp = kmem_alloc(sizeof (spa_config_dirent_t), 651 KM_SLEEP); 652 653 if (cachefile[0] == '\0') 654 dp->scd_path = spa_strdup(spa_config_path); 655 else if (strcmp(cachefile, "none") == 0) 656 dp->scd_path = NULL; 657 else 658 dp->scd_path = spa_strdup(cachefile); 659 660 list_insert_head(&spa->spa_config_list, dp); 661 if (need_sync) 662 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 663} 664 665int 666spa_prop_set(spa_t *spa, nvlist_t *nvp) 667{ 668 int error; 669 nvpair_t *elem = NULL; 670 boolean_t need_sync = B_FALSE; 671 672 if ((error = spa_prop_validate(spa, nvp)) != 0) 673 return (error); 674 675 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 676 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 677 678 if (prop == ZPOOL_PROP_CACHEFILE || 679 prop == ZPOOL_PROP_ALTROOT || 680 prop == ZPOOL_PROP_READONLY) 681 continue; 682 683 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 684 uint64_t ver; 685 686 if (prop == ZPOOL_PROP_VERSION) { 687 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 688 } else { 689 ASSERT(zpool_prop_feature(nvpair_name(elem))); 690 ver = SPA_VERSION_FEATURES; 691 need_sync = B_TRUE; 692 } 693 694 /* Save time if the version is already set. */ 695 if (ver == spa_version(spa)) 696 continue; 697 698 /* 699 * In addition to the pool directory object, we might 700 * create the pool properties object, the features for 701 * read object, the features for write object, or the 702 * feature descriptions object. 703 */ 704 error = dsl_sync_task(spa->spa_name, NULL, 705 spa_sync_version, &ver, 706 6, ZFS_SPACE_CHECK_RESERVED); 707 if (error) 708 return (error); 709 continue; 710 } 711 712 need_sync = B_TRUE; 713 break; 714 } 715 716 if (need_sync) { 717 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 718 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 719 } 720 721 return (0); 722} 723 724/* 725 * If the bootfs property value is dsobj, clear it. 726 */ 727void 728spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 729{ 730 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 731 VERIFY(zap_remove(spa->spa_meta_objset, 732 spa->spa_pool_props_object, 733 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 734 spa->spa_bootfs = 0; 735 } 736} 737 738/*ARGSUSED*/ 739static int 740spa_change_guid_check(void *arg, dmu_tx_t *tx) 741{ 742 uint64_t *newguid = arg; 743 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 744 vdev_t *rvd = spa->spa_root_vdev; 745 uint64_t vdev_state; 746 747 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 748 vdev_state = rvd->vdev_state; 749 spa_config_exit(spa, SCL_STATE, FTAG); 750 751 if (vdev_state != VDEV_STATE_HEALTHY) 752 return (SET_ERROR(ENXIO)); 753 754 ASSERT3U(spa_guid(spa), !=, *newguid); 755 756 return (0); 757} 758 759static void 760spa_change_guid_sync(void *arg, dmu_tx_t *tx) 761{ 762 uint64_t *newguid = arg; 763 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 764 uint64_t oldguid; 765 vdev_t *rvd = spa->spa_root_vdev; 766 767 oldguid = spa_guid(spa); 768 769 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 770 rvd->vdev_guid = *newguid; 771 rvd->vdev_guid_sum += (*newguid - oldguid); 772 vdev_config_dirty(rvd); 773 spa_config_exit(spa, SCL_STATE, FTAG); 774 775 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 776 oldguid, *newguid); 777} 778 779/* 780 * Change the GUID for the pool. This is done so that we can later 781 * re-import a pool built from a clone of our own vdevs. We will modify 782 * the root vdev's guid, our own pool guid, and then mark all of our 783 * vdevs dirty. Note that we must make sure that all our vdevs are 784 * online when we do this, or else any vdevs that weren't present 785 * would be orphaned from our pool. We are also going to issue a 786 * sysevent to update any watchers. 787 */ 788int 789spa_change_guid(spa_t *spa) 790{ 791 int error; 792 uint64_t guid; 793 794 mutex_enter(&spa->spa_vdev_top_lock); 795 mutex_enter(&spa_namespace_lock); 796 guid = spa_generate_guid(NULL); 797 798 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 799 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 800 801 if (error == 0) { 802 spa_config_sync(spa, B_FALSE, B_TRUE); 803 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 804 } 805 806 mutex_exit(&spa_namespace_lock); 807 mutex_exit(&spa->spa_vdev_top_lock); 808 809 return (error); 810} 811 812/* 813 * ========================================================================== 814 * SPA state manipulation (open/create/destroy/import/export) 815 * ========================================================================== 816 */ 817 818static int 819spa_error_entry_compare(const void *a, const void *b) 820{ 821 spa_error_entry_t *sa = (spa_error_entry_t *)a; 822 spa_error_entry_t *sb = (spa_error_entry_t *)b; 823 int ret; 824 825 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 826 sizeof (zbookmark_phys_t)); 827 828 if (ret < 0) 829 return (-1); 830 else if (ret > 0) 831 return (1); 832 else 833 return (0); 834} 835 836/* 837 * Utility function which retrieves copies of the current logs and 838 * re-initializes them in the process. 839 */ 840void 841spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 842{ 843 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 844 845 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 846 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 847 848 avl_create(&spa->spa_errlist_scrub, 849 spa_error_entry_compare, sizeof (spa_error_entry_t), 850 offsetof(spa_error_entry_t, se_avl)); 851 avl_create(&spa->spa_errlist_last, 852 spa_error_entry_compare, sizeof (spa_error_entry_t), 853 offsetof(spa_error_entry_t, se_avl)); 854} 855 856static void 857spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 858{ 859 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 860 enum zti_modes mode = ztip->zti_mode; 861 uint_t value = ztip->zti_value; 862 uint_t count = ztip->zti_count; 863 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 864 char name[32]; 865 uint_t flags = 0; 866 boolean_t batch = B_FALSE; 867 868 if (mode == ZTI_MODE_NULL) { 869 tqs->stqs_count = 0; 870 tqs->stqs_taskq = NULL; 871 return; 872 } 873 874 ASSERT3U(count, >, 0); 875 876 tqs->stqs_count = count; 877 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 878 879 switch (mode) { 880 case ZTI_MODE_FIXED: 881 ASSERT3U(value, >=, 1); 882 value = MAX(value, 1); 883 break; 884 885 case ZTI_MODE_BATCH: 886 batch = B_TRUE; 887 flags |= TASKQ_THREADS_CPU_PCT; 888 value = zio_taskq_batch_pct; 889 break; 890 891 default: 892 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 893 "spa_activate()", 894 zio_type_name[t], zio_taskq_types[q], mode, value); 895 break; 896 } 897 898 for (uint_t i = 0; i < count; i++) { 899 taskq_t *tq; 900 901 if (count > 1) { 902 (void) snprintf(name, sizeof (name), "%s_%s_%u", 903 zio_type_name[t], zio_taskq_types[q], i); 904 } else { 905 (void) snprintf(name, sizeof (name), "%s_%s", 906 zio_type_name[t], zio_taskq_types[q]); 907 } 908 909#ifdef SYSDC 910 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 911 if (batch) 912 flags |= TASKQ_DC_BATCH; 913 914 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 915 spa->spa_proc, zio_taskq_basedc, flags); 916 } else { 917#endif 918 pri_t pri = maxclsyspri; 919 /* 920 * The write issue taskq can be extremely CPU 921 * intensive. Run it at slightly lower priority 922 * than the other taskqs. 923 */ 924 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 925 pri--; 926 927 tq = taskq_create_proc(name, value, pri, 50, 928 INT_MAX, spa->spa_proc, flags); 929#ifdef SYSDC 930 } 931#endif 932 933 tqs->stqs_taskq[i] = tq; 934 } 935} 936 937static void 938spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 939{ 940 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 941 942 if (tqs->stqs_taskq == NULL) { 943 ASSERT0(tqs->stqs_count); 944 return; 945 } 946 947 for (uint_t i = 0; i < tqs->stqs_count; i++) { 948 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 949 taskq_destroy(tqs->stqs_taskq[i]); 950 } 951 952 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 953 tqs->stqs_taskq = NULL; 954} 955 956/* 957 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 958 * Note that a type may have multiple discrete taskqs to avoid lock contention 959 * on the taskq itself. In that case we choose which taskq at random by using 960 * the low bits of gethrtime(). 961 */ 962void 963spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 964 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 965{ 966 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 967 taskq_t *tq; 968 969 ASSERT3P(tqs->stqs_taskq, !=, NULL); 970 ASSERT3U(tqs->stqs_count, !=, 0); 971 972 if (tqs->stqs_count == 1) { 973 tq = tqs->stqs_taskq[0]; 974 } else { 975#ifdef _KERNEL 976 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 977#else 978 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 979#endif 980 } 981 982 taskq_dispatch_ent(tq, func, arg, flags, ent); 983} 984 985static void 986spa_create_zio_taskqs(spa_t *spa) 987{ 988 for (int t = 0; t < ZIO_TYPES; t++) { 989 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 990 spa_taskqs_init(spa, t, q); 991 } 992 } 993} 994 995#ifdef _KERNEL 996#ifdef SPA_PROCESS 997static void 998spa_thread(void *arg) 999{ 1000 callb_cpr_t cprinfo; 1001 1002 spa_t *spa = arg; 1003 user_t *pu = PTOU(curproc); 1004 1005 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1006 spa->spa_name); 1007 1008 ASSERT(curproc != &p0); 1009 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1010 "zpool-%s", spa->spa_name); 1011 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1012 1013#ifdef PSRSET_BIND 1014 /* bind this thread to the requested psrset */ 1015 if (zio_taskq_psrset_bind != PS_NONE) { 1016 pool_lock(); 1017 mutex_enter(&cpu_lock); 1018 mutex_enter(&pidlock); 1019 mutex_enter(&curproc->p_lock); 1020 1021 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1022 0, NULL, NULL) == 0) { 1023 curthread->t_bind_pset = zio_taskq_psrset_bind; 1024 } else { 1025 cmn_err(CE_WARN, 1026 "Couldn't bind process for zfs pool \"%s\" to " 1027 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1028 } 1029 1030 mutex_exit(&curproc->p_lock); 1031 mutex_exit(&pidlock); 1032 mutex_exit(&cpu_lock); 1033 pool_unlock(); 1034 } 1035#endif 1036 1037#ifdef SYSDC 1038 if (zio_taskq_sysdc) { 1039 sysdc_thread_enter(curthread, 100, 0); 1040 } 1041#endif 1042 1043 spa->spa_proc = curproc; 1044 spa->spa_did = curthread->t_did; 1045 1046 spa_create_zio_taskqs(spa); 1047 1048 mutex_enter(&spa->spa_proc_lock); 1049 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1050 1051 spa->spa_proc_state = SPA_PROC_ACTIVE; 1052 cv_broadcast(&spa->spa_proc_cv); 1053 1054 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1055 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1056 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1057 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1058 1059 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1060 spa->spa_proc_state = SPA_PROC_GONE; 1061 spa->spa_proc = &p0; 1062 cv_broadcast(&spa->spa_proc_cv); 1063 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1064 1065 mutex_enter(&curproc->p_lock); 1066 lwp_exit(); 1067} 1068#endif /* SPA_PROCESS */ 1069#endif 1070 1071/* 1072 * Activate an uninitialized pool. 1073 */ 1074static void 1075spa_activate(spa_t *spa, int mode) 1076{ 1077 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1078 1079 spa->spa_state = POOL_STATE_ACTIVE; 1080 spa->spa_mode = mode; 1081 1082 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1083 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1084 1085 /* Try to create a covering process */ 1086 mutex_enter(&spa->spa_proc_lock); 1087 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1088 ASSERT(spa->spa_proc == &p0); 1089 spa->spa_did = 0; 1090 1091#ifdef SPA_PROCESS 1092 /* Only create a process if we're going to be around a while. */ 1093 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1094 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1095 NULL, 0) == 0) { 1096 spa->spa_proc_state = SPA_PROC_CREATED; 1097 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1098 cv_wait(&spa->spa_proc_cv, 1099 &spa->spa_proc_lock); 1100 } 1101 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1102 ASSERT(spa->spa_proc != &p0); 1103 ASSERT(spa->spa_did != 0); 1104 } else { 1105#ifdef _KERNEL 1106 cmn_err(CE_WARN, 1107 "Couldn't create process for zfs pool \"%s\"\n", 1108 spa->spa_name); 1109#endif 1110 } 1111 } 1112#endif /* SPA_PROCESS */ 1113 mutex_exit(&spa->spa_proc_lock); 1114 1115 /* If we didn't create a process, we need to create our taskqs. */ 1116 ASSERT(spa->spa_proc == &p0); 1117 if (spa->spa_proc == &p0) { 1118 spa_create_zio_taskqs(spa); 1119 } 1120 1121 /* 1122 * Start TRIM thread. 1123 */ 1124 trim_thread_create(spa); 1125 1126 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1127 offsetof(vdev_t, vdev_config_dirty_node)); 1128 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1129 offsetof(vdev_t, vdev_state_dirty_node)); 1130 1131 txg_list_create(&spa->spa_vdev_txg_list, 1132 offsetof(struct vdev, vdev_txg_node)); 1133 1134 avl_create(&spa->spa_errlist_scrub, 1135 spa_error_entry_compare, sizeof (spa_error_entry_t), 1136 offsetof(spa_error_entry_t, se_avl)); 1137 avl_create(&spa->spa_errlist_last, 1138 spa_error_entry_compare, sizeof (spa_error_entry_t), 1139 offsetof(spa_error_entry_t, se_avl)); 1140} 1141 1142/* 1143 * Opposite of spa_activate(). 1144 */ 1145static void 1146spa_deactivate(spa_t *spa) 1147{ 1148 ASSERT(spa->spa_sync_on == B_FALSE); 1149 ASSERT(spa->spa_dsl_pool == NULL); 1150 ASSERT(spa->spa_root_vdev == NULL); 1151 ASSERT(spa->spa_async_zio_root == NULL); 1152 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1153 1154 /* 1155 * Stop TRIM thread in case spa_unload() wasn't called directly 1156 * before spa_deactivate(). 1157 */ 1158 trim_thread_destroy(spa); 1159 1160 txg_list_destroy(&spa->spa_vdev_txg_list); 1161 1162 list_destroy(&spa->spa_config_dirty_list); 1163 list_destroy(&spa->spa_state_dirty_list); 1164 1165 for (int t = 0; t < ZIO_TYPES; t++) { 1166 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1167 spa_taskqs_fini(spa, t, q); 1168 } 1169 } 1170 1171 metaslab_class_destroy(spa->spa_normal_class); 1172 spa->spa_normal_class = NULL; 1173 1174 metaslab_class_destroy(spa->spa_log_class); 1175 spa->spa_log_class = NULL; 1176 1177 /* 1178 * If this was part of an import or the open otherwise failed, we may 1179 * still have errors left in the queues. Empty them just in case. 1180 */ 1181 spa_errlog_drain(spa); 1182 1183 avl_destroy(&spa->spa_errlist_scrub); 1184 avl_destroy(&spa->spa_errlist_last); 1185 1186 spa->spa_state = POOL_STATE_UNINITIALIZED; 1187 1188 mutex_enter(&spa->spa_proc_lock); 1189 if (spa->spa_proc_state != SPA_PROC_NONE) { 1190 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1191 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1192 cv_broadcast(&spa->spa_proc_cv); 1193 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1194 ASSERT(spa->spa_proc != &p0); 1195 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1196 } 1197 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1198 spa->spa_proc_state = SPA_PROC_NONE; 1199 } 1200 ASSERT(spa->spa_proc == &p0); 1201 mutex_exit(&spa->spa_proc_lock); 1202 1203#ifdef SPA_PROCESS 1204 /* 1205 * We want to make sure spa_thread() has actually exited the ZFS 1206 * module, so that the module can't be unloaded out from underneath 1207 * it. 1208 */ 1209 if (spa->spa_did != 0) { 1210 thread_join(spa->spa_did); 1211 spa->spa_did = 0; 1212 } 1213#endif /* SPA_PROCESS */ 1214} 1215 1216/* 1217 * Verify a pool configuration, and construct the vdev tree appropriately. This 1218 * will create all the necessary vdevs in the appropriate layout, with each vdev 1219 * in the CLOSED state. This will prep the pool before open/creation/import. 1220 * All vdev validation is done by the vdev_alloc() routine. 1221 */ 1222static int 1223spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1224 uint_t id, int atype) 1225{ 1226 nvlist_t **child; 1227 uint_t children; 1228 int error; 1229 1230 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1231 return (error); 1232 1233 if ((*vdp)->vdev_ops->vdev_op_leaf) 1234 return (0); 1235 1236 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1237 &child, &children); 1238 1239 if (error == ENOENT) 1240 return (0); 1241 1242 if (error) { 1243 vdev_free(*vdp); 1244 *vdp = NULL; 1245 return (SET_ERROR(EINVAL)); 1246 } 1247 1248 for (int c = 0; c < children; c++) { 1249 vdev_t *vd; 1250 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1251 atype)) != 0) { 1252 vdev_free(*vdp); 1253 *vdp = NULL; 1254 return (error); 1255 } 1256 } 1257 1258 ASSERT(*vdp != NULL); 1259 1260 return (0); 1261} 1262 1263/* 1264 * Opposite of spa_load(). 1265 */ 1266static void 1267spa_unload(spa_t *spa) 1268{ 1269 int i; 1270 1271 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1272 1273 /* 1274 * Stop TRIM thread. 1275 */ 1276 trim_thread_destroy(spa); 1277 1278 /* 1279 * Stop async tasks. 1280 */ 1281 spa_async_suspend(spa); 1282 1283 /* 1284 * Stop syncing. 1285 */ 1286 if (spa->spa_sync_on) { 1287 txg_sync_stop(spa->spa_dsl_pool); 1288 spa->spa_sync_on = B_FALSE; 1289 } 1290 1291 /* 1292 * Wait for any outstanding async I/O to complete. 1293 */ 1294 if (spa->spa_async_zio_root != NULL) { 1295 for (int i = 0; i < max_ncpus; i++) 1296 (void) zio_wait(spa->spa_async_zio_root[i]); 1297 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1298 spa->spa_async_zio_root = NULL; 1299 } 1300 1301 bpobj_close(&spa->spa_deferred_bpobj); 1302 1303 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1304 1305 /* 1306 * Close all vdevs. 1307 */ 1308 if (spa->spa_root_vdev) 1309 vdev_free(spa->spa_root_vdev); 1310 ASSERT(spa->spa_root_vdev == NULL); 1311 1312 /* 1313 * Close the dsl pool. 1314 */ 1315 if (spa->spa_dsl_pool) { 1316 dsl_pool_close(spa->spa_dsl_pool); 1317 spa->spa_dsl_pool = NULL; 1318 spa->spa_meta_objset = NULL; 1319 } 1320 1321 ddt_unload(spa); 1322 1323 1324 /* 1325 * Drop and purge level 2 cache 1326 */ 1327 spa_l2cache_drop(spa); 1328 1329 for (i = 0; i < spa->spa_spares.sav_count; i++) 1330 vdev_free(spa->spa_spares.sav_vdevs[i]); 1331 if (spa->spa_spares.sav_vdevs) { 1332 kmem_free(spa->spa_spares.sav_vdevs, 1333 spa->spa_spares.sav_count * sizeof (void *)); 1334 spa->spa_spares.sav_vdevs = NULL; 1335 } 1336 if (spa->spa_spares.sav_config) { 1337 nvlist_free(spa->spa_spares.sav_config); 1338 spa->spa_spares.sav_config = NULL; 1339 } 1340 spa->spa_spares.sav_count = 0; 1341 1342 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1343 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1344 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1345 } 1346 if (spa->spa_l2cache.sav_vdevs) { 1347 kmem_free(spa->spa_l2cache.sav_vdevs, 1348 spa->spa_l2cache.sav_count * sizeof (void *)); 1349 spa->spa_l2cache.sav_vdevs = NULL; 1350 } 1351 if (spa->spa_l2cache.sav_config) { 1352 nvlist_free(spa->spa_l2cache.sav_config); 1353 spa->spa_l2cache.sav_config = NULL; 1354 } 1355 spa->spa_l2cache.sav_count = 0; 1356 1357 spa->spa_async_suspended = 0; 1358 1359 if (spa->spa_comment != NULL) { 1360 spa_strfree(spa->spa_comment); 1361 spa->spa_comment = NULL; 1362 } 1363 1364 spa_config_exit(spa, SCL_ALL, FTAG); 1365} 1366 1367/* 1368 * Load (or re-load) the current list of vdevs describing the active spares for 1369 * this pool. When this is called, we have some form of basic information in 1370 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1371 * then re-generate a more complete list including status information. 1372 */ 1373static void 1374spa_load_spares(spa_t *spa) 1375{ 1376 nvlist_t **spares; 1377 uint_t nspares; 1378 int i; 1379 vdev_t *vd, *tvd; 1380 1381 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1382 1383 /* 1384 * First, close and free any existing spare vdevs. 1385 */ 1386 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1387 vd = spa->spa_spares.sav_vdevs[i]; 1388 1389 /* Undo the call to spa_activate() below */ 1390 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1391 B_FALSE)) != NULL && tvd->vdev_isspare) 1392 spa_spare_remove(tvd); 1393 vdev_close(vd); 1394 vdev_free(vd); 1395 } 1396 1397 if (spa->spa_spares.sav_vdevs) 1398 kmem_free(spa->spa_spares.sav_vdevs, 1399 spa->spa_spares.sav_count * sizeof (void *)); 1400 1401 if (spa->spa_spares.sav_config == NULL) 1402 nspares = 0; 1403 else 1404 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1405 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1406 1407 spa->spa_spares.sav_count = (int)nspares; 1408 spa->spa_spares.sav_vdevs = NULL; 1409 1410 if (nspares == 0) 1411 return; 1412 1413 /* 1414 * Construct the array of vdevs, opening them to get status in the 1415 * process. For each spare, there is potentially two different vdev_t 1416 * structures associated with it: one in the list of spares (used only 1417 * for basic validation purposes) and one in the active vdev 1418 * configuration (if it's spared in). During this phase we open and 1419 * validate each vdev on the spare list. If the vdev also exists in the 1420 * active configuration, then we also mark this vdev as an active spare. 1421 */ 1422 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1423 KM_SLEEP); 1424 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1425 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1426 VDEV_ALLOC_SPARE) == 0); 1427 ASSERT(vd != NULL); 1428 1429 spa->spa_spares.sav_vdevs[i] = vd; 1430 1431 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1432 B_FALSE)) != NULL) { 1433 if (!tvd->vdev_isspare) 1434 spa_spare_add(tvd); 1435 1436 /* 1437 * We only mark the spare active if we were successfully 1438 * able to load the vdev. Otherwise, importing a pool 1439 * with a bad active spare would result in strange 1440 * behavior, because multiple pool would think the spare 1441 * is actively in use. 1442 * 1443 * There is a vulnerability here to an equally bizarre 1444 * circumstance, where a dead active spare is later 1445 * brought back to life (onlined or otherwise). Given 1446 * the rarity of this scenario, and the extra complexity 1447 * it adds, we ignore the possibility. 1448 */ 1449 if (!vdev_is_dead(tvd)) 1450 spa_spare_activate(tvd); 1451 } 1452 1453 vd->vdev_top = vd; 1454 vd->vdev_aux = &spa->spa_spares; 1455 1456 if (vdev_open(vd) != 0) 1457 continue; 1458 1459 if (vdev_validate_aux(vd) == 0) 1460 spa_spare_add(vd); 1461 } 1462 1463 /* 1464 * Recompute the stashed list of spares, with status information 1465 * this time. 1466 */ 1467 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1468 DATA_TYPE_NVLIST_ARRAY) == 0); 1469 1470 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1471 KM_SLEEP); 1472 for (i = 0; i < spa->spa_spares.sav_count; i++) 1473 spares[i] = vdev_config_generate(spa, 1474 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1475 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1476 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1477 for (i = 0; i < spa->spa_spares.sav_count; i++) 1478 nvlist_free(spares[i]); 1479 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1480} 1481 1482/* 1483 * Load (or re-load) the current list of vdevs describing the active l2cache for 1484 * this pool. When this is called, we have some form of basic information in 1485 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1486 * then re-generate a more complete list including status information. 1487 * Devices which are already active have their details maintained, and are 1488 * not re-opened. 1489 */ 1490static void 1491spa_load_l2cache(spa_t *spa) 1492{ 1493 nvlist_t **l2cache; 1494 uint_t nl2cache; 1495 int i, j, oldnvdevs; 1496 uint64_t guid; 1497 vdev_t *vd, **oldvdevs, **newvdevs; 1498 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1499 1500 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1501 1502 if (sav->sav_config != NULL) { 1503 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1504 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1505 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1506 } else { 1507 nl2cache = 0; 1508 newvdevs = NULL; 1509 } 1510 1511 oldvdevs = sav->sav_vdevs; 1512 oldnvdevs = sav->sav_count; 1513 sav->sav_vdevs = NULL; 1514 sav->sav_count = 0; 1515 1516 /* 1517 * Process new nvlist of vdevs. 1518 */ 1519 for (i = 0; i < nl2cache; i++) { 1520 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1521 &guid) == 0); 1522 1523 newvdevs[i] = NULL; 1524 for (j = 0; j < oldnvdevs; j++) { 1525 vd = oldvdevs[j]; 1526 if (vd != NULL && guid == vd->vdev_guid) { 1527 /* 1528 * Retain previous vdev for add/remove ops. 1529 */ 1530 newvdevs[i] = vd; 1531 oldvdevs[j] = NULL; 1532 break; 1533 } 1534 } 1535 1536 if (newvdevs[i] == NULL) { 1537 /* 1538 * Create new vdev 1539 */ 1540 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1541 VDEV_ALLOC_L2CACHE) == 0); 1542 ASSERT(vd != NULL); 1543 newvdevs[i] = vd; 1544 1545 /* 1546 * Commit this vdev as an l2cache device, 1547 * even if it fails to open. 1548 */ 1549 spa_l2cache_add(vd); 1550 1551 vd->vdev_top = vd; 1552 vd->vdev_aux = sav; 1553 1554 spa_l2cache_activate(vd); 1555 1556 if (vdev_open(vd) != 0) 1557 continue; 1558 1559 (void) vdev_validate_aux(vd); 1560 1561 if (!vdev_is_dead(vd)) 1562 l2arc_add_vdev(spa, vd); 1563 } 1564 } 1565 1566 /* 1567 * Purge vdevs that were dropped 1568 */ 1569 for (i = 0; i < oldnvdevs; i++) { 1570 uint64_t pool; 1571 1572 vd = oldvdevs[i]; 1573 if (vd != NULL) { 1574 ASSERT(vd->vdev_isl2cache); 1575 1576 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1577 pool != 0ULL && l2arc_vdev_present(vd)) 1578 l2arc_remove_vdev(vd); 1579 vdev_clear_stats(vd); 1580 vdev_free(vd); 1581 } 1582 } 1583 1584 if (oldvdevs) 1585 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1586 1587 if (sav->sav_config == NULL) 1588 goto out; 1589 1590 sav->sav_vdevs = newvdevs; 1591 sav->sav_count = (int)nl2cache; 1592 1593 /* 1594 * Recompute the stashed list of l2cache devices, with status 1595 * information this time. 1596 */ 1597 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1598 DATA_TYPE_NVLIST_ARRAY) == 0); 1599 1600 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1601 for (i = 0; i < sav->sav_count; i++) 1602 l2cache[i] = vdev_config_generate(spa, 1603 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1604 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1605 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1606out: 1607 for (i = 0; i < sav->sav_count; i++) 1608 nvlist_free(l2cache[i]); 1609 if (sav->sav_count) 1610 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1611} 1612 1613static int 1614load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1615{ 1616 dmu_buf_t *db; 1617 char *packed = NULL; 1618 size_t nvsize = 0; 1619 int error; 1620 *value = NULL; 1621 1622 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1623 if (error != 0) 1624 return (error); 1625 nvsize = *(uint64_t *)db->db_data; 1626 dmu_buf_rele(db, FTAG); 1627 1628 packed = kmem_alloc(nvsize, KM_SLEEP); 1629 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1630 DMU_READ_PREFETCH); 1631 if (error == 0) 1632 error = nvlist_unpack(packed, nvsize, value, 0); 1633 kmem_free(packed, nvsize); 1634 1635 return (error); 1636} 1637 1638/* 1639 * Checks to see if the given vdev could not be opened, in which case we post a 1640 * sysevent to notify the autoreplace code that the device has been removed. 1641 */ 1642static void 1643spa_check_removed(vdev_t *vd) 1644{ 1645 for (int c = 0; c < vd->vdev_children; c++) 1646 spa_check_removed(vd->vdev_child[c]); 1647 1648 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1649 !vd->vdev_ishole) { 1650 zfs_post_autoreplace(vd->vdev_spa, vd); 1651 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1652 } 1653} 1654 1655/* 1656 * Validate the current config against the MOS config 1657 */ 1658static boolean_t 1659spa_config_valid(spa_t *spa, nvlist_t *config) 1660{ 1661 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1662 nvlist_t *nv; 1663 1664 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1665 1666 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1667 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1668 1669 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1670 1671 /* 1672 * If we're doing a normal import, then build up any additional 1673 * diagnostic information about missing devices in this config. 1674 * We'll pass this up to the user for further processing. 1675 */ 1676 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1677 nvlist_t **child, *nv; 1678 uint64_t idx = 0; 1679 1680 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1681 KM_SLEEP); 1682 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1683 1684 for (int c = 0; c < rvd->vdev_children; c++) { 1685 vdev_t *tvd = rvd->vdev_child[c]; 1686 vdev_t *mtvd = mrvd->vdev_child[c]; 1687 1688 if (tvd->vdev_ops == &vdev_missing_ops && 1689 mtvd->vdev_ops != &vdev_missing_ops && 1690 mtvd->vdev_islog) 1691 child[idx++] = vdev_config_generate(spa, mtvd, 1692 B_FALSE, 0); 1693 } 1694 1695 if (idx) { 1696 VERIFY(nvlist_add_nvlist_array(nv, 1697 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1698 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1699 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1700 1701 for (int i = 0; i < idx; i++) 1702 nvlist_free(child[i]); 1703 } 1704 nvlist_free(nv); 1705 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1706 } 1707 1708 /* 1709 * Compare the root vdev tree with the information we have 1710 * from the MOS config (mrvd). Check each top-level vdev 1711 * with the corresponding MOS config top-level (mtvd). 1712 */ 1713 for (int c = 0; c < rvd->vdev_children; c++) { 1714 vdev_t *tvd = rvd->vdev_child[c]; 1715 vdev_t *mtvd = mrvd->vdev_child[c]; 1716 1717 /* 1718 * Resolve any "missing" vdevs in the current configuration. 1719 * If we find that the MOS config has more accurate information 1720 * about the top-level vdev then use that vdev instead. 1721 */ 1722 if (tvd->vdev_ops == &vdev_missing_ops && 1723 mtvd->vdev_ops != &vdev_missing_ops) { 1724 1725 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1726 continue; 1727 1728 /* 1729 * Device specific actions. 1730 */ 1731 if (mtvd->vdev_islog) { 1732 spa_set_log_state(spa, SPA_LOG_CLEAR); 1733 } else { 1734 /* 1735 * XXX - once we have 'readonly' pool 1736 * support we should be able to handle 1737 * missing data devices by transitioning 1738 * the pool to readonly. 1739 */ 1740 continue; 1741 } 1742 1743 /* 1744 * Swap the missing vdev with the data we were 1745 * able to obtain from the MOS config. 1746 */ 1747 vdev_remove_child(rvd, tvd); 1748 vdev_remove_child(mrvd, mtvd); 1749 1750 vdev_add_child(rvd, mtvd); 1751 vdev_add_child(mrvd, tvd); 1752 1753 spa_config_exit(spa, SCL_ALL, FTAG); 1754 vdev_load(mtvd); 1755 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1756 1757 vdev_reopen(rvd); 1758 } else if (mtvd->vdev_islog) { 1759 /* 1760 * Load the slog device's state from the MOS config 1761 * since it's possible that the label does not 1762 * contain the most up-to-date information. 1763 */ 1764 vdev_load_log_state(tvd, mtvd); 1765 vdev_reopen(tvd); 1766 } 1767 } 1768 vdev_free(mrvd); 1769 spa_config_exit(spa, SCL_ALL, FTAG); 1770 1771 /* 1772 * Ensure we were able to validate the config. 1773 */ 1774 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1775} 1776 1777/* 1778 * Check for missing log devices 1779 */ 1780static boolean_t 1781spa_check_logs(spa_t *spa) 1782{ 1783 boolean_t rv = B_FALSE; 1784 1785 switch (spa->spa_log_state) { 1786 case SPA_LOG_MISSING: 1787 /* need to recheck in case slog has been restored */ 1788 case SPA_LOG_UNKNOWN: 1789 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1790 NULL, DS_FIND_CHILDREN) != 0); 1791 if (rv) 1792 spa_set_log_state(spa, SPA_LOG_MISSING); 1793 break; 1794 } 1795 return (rv); 1796} 1797 1798static boolean_t 1799spa_passivate_log(spa_t *spa) 1800{ 1801 vdev_t *rvd = spa->spa_root_vdev; 1802 boolean_t slog_found = B_FALSE; 1803 1804 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1805 1806 if (!spa_has_slogs(spa)) 1807 return (B_FALSE); 1808 1809 for (int c = 0; c < rvd->vdev_children; c++) { 1810 vdev_t *tvd = rvd->vdev_child[c]; 1811 metaslab_group_t *mg = tvd->vdev_mg; 1812 1813 if (tvd->vdev_islog) { 1814 metaslab_group_passivate(mg); 1815 slog_found = B_TRUE; 1816 } 1817 } 1818 1819 return (slog_found); 1820} 1821 1822static void 1823spa_activate_log(spa_t *spa) 1824{ 1825 vdev_t *rvd = spa->spa_root_vdev; 1826 1827 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1828 1829 for (int c = 0; c < rvd->vdev_children; c++) { 1830 vdev_t *tvd = rvd->vdev_child[c]; 1831 metaslab_group_t *mg = tvd->vdev_mg; 1832 1833 if (tvd->vdev_islog) 1834 metaslab_group_activate(mg); 1835 } 1836} 1837 1838int 1839spa_offline_log(spa_t *spa) 1840{ 1841 int error; 1842 1843 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1844 NULL, DS_FIND_CHILDREN); 1845 if (error == 0) { 1846 /* 1847 * We successfully offlined the log device, sync out the 1848 * current txg so that the "stubby" block can be removed 1849 * by zil_sync(). 1850 */ 1851 txg_wait_synced(spa->spa_dsl_pool, 0); 1852 } 1853 return (error); 1854} 1855 1856static void 1857spa_aux_check_removed(spa_aux_vdev_t *sav) 1858{ 1859 int i; 1860 1861 for (i = 0; i < sav->sav_count; i++) 1862 spa_check_removed(sav->sav_vdevs[i]); 1863} 1864 1865void 1866spa_claim_notify(zio_t *zio) 1867{ 1868 spa_t *spa = zio->io_spa; 1869 1870 if (zio->io_error) 1871 return; 1872 1873 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1874 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1875 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1876 mutex_exit(&spa->spa_props_lock); 1877} 1878 1879typedef struct spa_load_error { 1880 uint64_t sle_meta_count; 1881 uint64_t sle_data_count; 1882} spa_load_error_t; 1883 1884static void 1885spa_load_verify_done(zio_t *zio) 1886{ 1887 blkptr_t *bp = zio->io_bp; 1888 spa_load_error_t *sle = zio->io_private; 1889 dmu_object_type_t type = BP_GET_TYPE(bp); 1890 int error = zio->io_error; 1891 spa_t *spa = zio->io_spa; 1892 1893 if (error) { 1894 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1895 type != DMU_OT_INTENT_LOG) 1896 atomic_inc_64(&sle->sle_meta_count); 1897 else 1898 atomic_inc_64(&sle->sle_data_count); 1899 } 1900 zio_data_buf_free(zio->io_data, zio->io_size); 1901 1902 mutex_enter(&spa->spa_scrub_lock); 1903 spa->spa_scrub_inflight--; 1904 cv_broadcast(&spa->spa_scrub_io_cv); 1905 mutex_exit(&spa->spa_scrub_lock); 1906} 1907 1908/* 1909 * Maximum number of concurrent scrub i/os to create while verifying 1910 * a pool while importing it. 1911 */ 1912int spa_load_verify_maxinflight = 10000; 1913boolean_t spa_load_verify_metadata = B_TRUE; 1914boolean_t spa_load_verify_data = B_TRUE; 1915 1916SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1917 &spa_load_verify_maxinflight, 0, 1918 "Maximum number of concurrent scrub I/Os to create while verifying a " 1919 "pool while importing it"); 1920 1921SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1922 &spa_load_verify_metadata, 0, 1923 "Check metadata on import?"); 1924 1925SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1926 &spa_load_verify_data, 0, 1927 "Check user data on import?"); 1928 1929/*ARGSUSED*/ 1930static int 1931spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1932 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1933{ 1934 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1935 return (0); 1936 /* 1937 * Note: normally this routine will not be called if 1938 * spa_load_verify_metadata is not set. However, it may be useful 1939 * to manually set the flag after the traversal has begun. 1940 */ 1941 if (!spa_load_verify_metadata) 1942 return (0); 1943 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1944 return (0); 1945 1946 zio_t *rio = arg; 1947 size_t size = BP_GET_PSIZE(bp); 1948 void *data = zio_data_buf_alloc(size); 1949 1950 mutex_enter(&spa->spa_scrub_lock); 1951 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1952 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1953 spa->spa_scrub_inflight++; 1954 mutex_exit(&spa->spa_scrub_lock); 1955 1956 zio_nowait(zio_read(rio, spa, bp, data, size, 1957 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1958 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1959 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1960 return (0); 1961} 1962 1963static int 1964spa_load_verify(spa_t *spa) 1965{ 1966 zio_t *rio; 1967 spa_load_error_t sle = { 0 }; 1968 zpool_rewind_policy_t policy; 1969 boolean_t verify_ok = B_FALSE; 1970 int error = 0; 1971 1972 zpool_get_rewind_policy(spa->spa_config, &policy); 1973 1974 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1975 return (0); 1976 1977 rio = zio_root(spa, NULL, &sle, 1978 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1979 1980 if (spa_load_verify_metadata) { 1981 error = traverse_pool(spa, spa->spa_verify_min_txg, 1982 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1983 spa_load_verify_cb, rio); 1984 } 1985 1986 (void) zio_wait(rio); 1987 1988 spa->spa_load_meta_errors = sle.sle_meta_count; 1989 spa->spa_load_data_errors = sle.sle_data_count; 1990 1991 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1992 sle.sle_data_count <= policy.zrp_maxdata) { 1993 int64_t loss = 0; 1994 1995 verify_ok = B_TRUE; 1996 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1997 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1998 1999 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2000 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2001 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2002 VERIFY(nvlist_add_int64(spa->spa_load_info, 2003 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2004 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2005 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2006 } else { 2007 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2008 } 2009 2010 if (error) { 2011 if (error != ENXIO && error != EIO) 2012 error = SET_ERROR(EIO); 2013 return (error); 2014 } 2015 2016 return (verify_ok ? 0 : EIO); 2017} 2018 2019/* 2020 * Find a value in the pool props object. 2021 */ 2022static void 2023spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2024{ 2025 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2026 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2027} 2028 2029/* 2030 * Find a value in the pool directory object. 2031 */ 2032static int 2033spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2034{ 2035 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2036 name, sizeof (uint64_t), 1, val)); 2037} 2038 2039static int 2040spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2041{ 2042 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2043 return (err); 2044} 2045 2046/* 2047 * Fix up config after a partly-completed split. This is done with the 2048 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2049 * pool have that entry in their config, but only the splitting one contains 2050 * a list of all the guids of the vdevs that are being split off. 2051 * 2052 * This function determines what to do with that list: either rejoin 2053 * all the disks to the pool, or complete the splitting process. To attempt 2054 * the rejoin, each disk that is offlined is marked online again, and 2055 * we do a reopen() call. If the vdev label for every disk that was 2056 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2057 * then we call vdev_split() on each disk, and complete the split. 2058 * 2059 * Otherwise we leave the config alone, with all the vdevs in place in 2060 * the original pool. 2061 */ 2062static void 2063spa_try_repair(spa_t *spa, nvlist_t *config) 2064{ 2065 uint_t extracted; 2066 uint64_t *glist; 2067 uint_t i, gcount; 2068 nvlist_t *nvl; 2069 vdev_t **vd; 2070 boolean_t attempt_reopen; 2071 2072 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2073 return; 2074 2075 /* check that the config is complete */ 2076 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2077 &glist, &gcount) != 0) 2078 return; 2079 2080 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2081 2082 /* attempt to online all the vdevs & validate */ 2083 attempt_reopen = B_TRUE; 2084 for (i = 0; i < gcount; i++) { 2085 if (glist[i] == 0) /* vdev is hole */ 2086 continue; 2087 2088 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2089 if (vd[i] == NULL) { 2090 /* 2091 * Don't bother attempting to reopen the disks; 2092 * just do the split. 2093 */ 2094 attempt_reopen = B_FALSE; 2095 } else { 2096 /* attempt to re-online it */ 2097 vd[i]->vdev_offline = B_FALSE; 2098 } 2099 } 2100 2101 if (attempt_reopen) { 2102 vdev_reopen(spa->spa_root_vdev); 2103 2104 /* check each device to see what state it's in */ 2105 for (extracted = 0, i = 0; i < gcount; i++) { 2106 if (vd[i] != NULL && 2107 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2108 break; 2109 ++extracted; 2110 } 2111 } 2112 2113 /* 2114 * If every disk has been moved to the new pool, or if we never 2115 * even attempted to look at them, then we split them off for 2116 * good. 2117 */ 2118 if (!attempt_reopen || gcount == extracted) { 2119 for (i = 0; i < gcount; i++) 2120 if (vd[i] != NULL) 2121 vdev_split(vd[i]); 2122 vdev_reopen(spa->spa_root_vdev); 2123 } 2124 2125 kmem_free(vd, gcount * sizeof (vdev_t *)); 2126} 2127 2128static int 2129spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2130 boolean_t mosconfig) 2131{ 2132 nvlist_t *config = spa->spa_config; 2133 char *ereport = FM_EREPORT_ZFS_POOL; 2134 char *comment; 2135 int error; 2136 uint64_t pool_guid; 2137 nvlist_t *nvl; 2138 2139 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2140 return (SET_ERROR(EINVAL)); 2141 2142 ASSERT(spa->spa_comment == NULL); 2143 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2144 spa->spa_comment = spa_strdup(comment); 2145 2146 /* 2147 * Versioning wasn't explicitly added to the label until later, so if 2148 * it's not present treat it as the initial version. 2149 */ 2150 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2151 &spa->spa_ubsync.ub_version) != 0) 2152 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2153 2154 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2155 &spa->spa_config_txg); 2156 2157 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2158 spa_guid_exists(pool_guid, 0)) { 2159 error = SET_ERROR(EEXIST); 2160 } else { 2161 spa->spa_config_guid = pool_guid; 2162 2163 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2164 &nvl) == 0) { 2165 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2166 KM_SLEEP) == 0); 2167 } 2168 2169 nvlist_free(spa->spa_load_info); 2170 spa->spa_load_info = fnvlist_alloc(); 2171 2172 gethrestime(&spa->spa_loaded_ts); 2173 error = spa_load_impl(spa, pool_guid, config, state, type, 2174 mosconfig, &ereport); 2175 } 2176 2177 spa->spa_minref = refcount_count(&spa->spa_refcount); 2178 if (error) { 2179 if (error != EEXIST) { 2180 spa->spa_loaded_ts.tv_sec = 0; 2181 spa->spa_loaded_ts.tv_nsec = 0; 2182 } 2183 if (error != EBADF) { 2184 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2185 } 2186 } 2187 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2188 spa->spa_ena = 0; 2189 2190 return (error); 2191} 2192 2193/* 2194 * Load an existing storage pool, using the pool's builtin spa_config as a 2195 * source of configuration information. 2196 */ 2197static int 2198spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2199 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2200 char **ereport) 2201{ 2202 int error = 0; 2203 nvlist_t *nvroot = NULL; 2204 nvlist_t *label; 2205 vdev_t *rvd; 2206 uberblock_t *ub = &spa->spa_uberblock; 2207 uint64_t children, config_cache_txg = spa->spa_config_txg; 2208 int orig_mode = spa->spa_mode; 2209 int parse; 2210 uint64_t obj; 2211 boolean_t missing_feat_write = B_FALSE; 2212 2213 /* 2214 * If this is an untrusted config, access the pool in read-only mode. 2215 * This prevents things like resilvering recently removed devices. 2216 */ 2217 if (!mosconfig) 2218 spa->spa_mode = FREAD; 2219 2220 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2221 2222 spa->spa_load_state = state; 2223 2224 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2225 return (SET_ERROR(EINVAL)); 2226 2227 parse = (type == SPA_IMPORT_EXISTING ? 2228 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2229 2230 /* 2231 * Create "The Godfather" zio to hold all async IOs 2232 */ 2233 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2234 KM_SLEEP); 2235 for (int i = 0; i < max_ncpus; i++) { 2236 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2237 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2238 ZIO_FLAG_GODFATHER); 2239 } 2240 2241 /* 2242 * Parse the configuration into a vdev tree. We explicitly set the 2243 * value that will be returned by spa_version() since parsing the 2244 * configuration requires knowing the version number. 2245 */ 2246 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2247 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2248 spa_config_exit(spa, SCL_ALL, FTAG); 2249 2250 if (error != 0) 2251 return (error); 2252 2253 ASSERT(spa->spa_root_vdev == rvd); 2254 2255 if (type != SPA_IMPORT_ASSEMBLE) { 2256 ASSERT(spa_guid(spa) == pool_guid); 2257 } 2258 2259 /* 2260 * Try to open all vdevs, loading each label in the process. 2261 */ 2262 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2263 error = vdev_open(rvd); 2264 spa_config_exit(spa, SCL_ALL, FTAG); 2265 if (error != 0) 2266 return (error); 2267 2268 /* 2269 * We need to validate the vdev labels against the configuration that 2270 * we have in hand, which is dependent on the setting of mosconfig. If 2271 * mosconfig is true then we're validating the vdev labels based on 2272 * that config. Otherwise, we're validating against the cached config 2273 * (zpool.cache) that was read when we loaded the zfs module, and then 2274 * later we will recursively call spa_load() and validate against 2275 * the vdev config. 2276 * 2277 * If we're assembling a new pool that's been split off from an 2278 * existing pool, the labels haven't yet been updated so we skip 2279 * validation for now. 2280 */ 2281 if (type != SPA_IMPORT_ASSEMBLE) { 2282 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2283 error = vdev_validate(rvd, mosconfig); 2284 spa_config_exit(spa, SCL_ALL, FTAG); 2285 2286 if (error != 0) 2287 return (error); 2288 2289 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2290 return (SET_ERROR(ENXIO)); 2291 } 2292 2293 /* 2294 * Find the best uberblock. 2295 */ 2296 vdev_uberblock_load(rvd, ub, &label); 2297 2298 /* 2299 * If we weren't able to find a single valid uberblock, return failure. 2300 */ 2301 if (ub->ub_txg == 0) { 2302 nvlist_free(label); 2303 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2304 } 2305 2306 /* 2307 * If the pool has an unsupported version we can't open it. 2308 */ 2309 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2310 nvlist_free(label); 2311 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2312 } 2313 2314 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2315 nvlist_t *features; 2316 2317 /* 2318 * If we weren't able to find what's necessary for reading the 2319 * MOS in the label, return failure. 2320 */ 2321 if (label == NULL || nvlist_lookup_nvlist(label, 2322 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2323 nvlist_free(label); 2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2325 ENXIO)); 2326 } 2327 2328 /* 2329 * Update our in-core representation with the definitive values 2330 * from the label. 2331 */ 2332 nvlist_free(spa->spa_label_features); 2333 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2334 } 2335 2336 nvlist_free(label); 2337 2338 /* 2339 * Look through entries in the label nvlist's features_for_read. If 2340 * there is a feature listed there which we don't understand then we 2341 * cannot open a pool. 2342 */ 2343 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2344 nvlist_t *unsup_feat; 2345 2346 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2347 0); 2348 2349 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2350 NULL); nvp != NULL; 2351 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2352 if (!zfeature_is_supported(nvpair_name(nvp))) { 2353 VERIFY(nvlist_add_string(unsup_feat, 2354 nvpair_name(nvp), "") == 0); 2355 } 2356 } 2357 2358 if (!nvlist_empty(unsup_feat)) { 2359 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2360 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2361 nvlist_free(unsup_feat); 2362 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2363 ENOTSUP)); 2364 } 2365 2366 nvlist_free(unsup_feat); 2367 } 2368 2369 /* 2370 * If the vdev guid sum doesn't match the uberblock, we have an 2371 * incomplete configuration. We first check to see if the pool 2372 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2373 * If it is, defer the vdev_guid_sum check till later so we 2374 * can handle missing vdevs. 2375 */ 2376 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2377 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2378 rvd->vdev_guid_sum != ub->ub_guid_sum) 2379 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2380 2381 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2382 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2383 spa_try_repair(spa, config); 2384 spa_config_exit(spa, SCL_ALL, FTAG); 2385 nvlist_free(spa->spa_config_splitting); 2386 spa->spa_config_splitting = NULL; 2387 } 2388 2389 /* 2390 * Initialize internal SPA structures. 2391 */ 2392 spa->spa_state = POOL_STATE_ACTIVE; 2393 spa->spa_ubsync = spa->spa_uberblock; 2394 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2395 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2396 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2397 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2398 spa->spa_claim_max_txg = spa->spa_first_txg; 2399 spa->spa_prev_software_version = ub->ub_software_version; 2400 2401 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2402 if (error) 2403 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2404 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2405 2406 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2407 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2408 2409 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2410 boolean_t missing_feat_read = B_FALSE; 2411 nvlist_t *unsup_feat, *enabled_feat; 2412 2413 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2414 &spa->spa_feat_for_read_obj) != 0) { 2415 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2416 } 2417 2418 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2419 &spa->spa_feat_for_write_obj) != 0) { 2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2421 } 2422 2423 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2424 &spa->spa_feat_desc_obj) != 0) { 2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2426 } 2427 2428 enabled_feat = fnvlist_alloc(); 2429 unsup_feat = fnvlist_alloc(); 2430 2431 if (!spa_features_check(spa, B_FALSE, 2432 unsup_feat, enabled_feat)) 2433 missing_feat_read = B_TRUE; 2434 2435 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2436 if (!spa_features_check(spa, B_TRUE, 2437 unsup_feat, enabled_feat)) { 2438 missing_feat_write = B_TRUE; 2439 } 2440 } 2441 2442 fnvlist_add_nvlist(spa->spa_load_info, 2443 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2444 2445 if (!nvlist_empty(unsup_feat)) { 2446 fnvlist_add_nvlist(spa->spa_load_info, 2447 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2448 } 2449 2450 fnvlist_free(enabled_feat); 2451 fnvlist_free(unsup_feat); 2452 2453 if (!missing_feat_read) { 2454 fnvlist_add_boolean(spa->spa_load_info, 2455 ZPOOL_CONFIG_CAN_RDONLY); 2456 } 2457 2458 /* 2459 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2460 * twofold: to determine whether the pool is available for 2461 * import in read-write mode and (if it is not) whether the 2462 * pool is available for import in read-only mode. If the pool 2463 * is available for import in read-write mode, it is displayed 2464 * as available in userland; if it is not available for import 2465 * in read-only mode, it is displayed as unavailable in 2466 * userland. If the pool is available for import in read-only 2467 * mode but not read-write mode, it is displayed as unavailable 2468 * in userland with a special note that the pool is actually 2469 * available for open in read-only mode. 2470 * 2471 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2472 * missing a feature for write, we must first determine whether 2473 * the pool can be opened read-only before returning to 2474 * userland in order to know whether to display the 2475 * abovementioned note. 2476 */ 2477 if (missing_feat_read || (missing_feat_write && 2478 spa_writeable(spa))) { 2479 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2480 ENOTSUP)); 2481 } 2482 2483 /* 2484 * Load refcounts for ZFS features from disk into an in-memory 2485 * cache during SPA initialization. 2486 */ 2487 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2488 uint64_t refcount; 2489 2490 error = feature_get_refcount_from_disk(spa, 2491 &spa_feature_table[i], &refcount); 2492 if (error == 0) { 2493 spa->spa_feat_refcount_cache[i] = refcount; 2494 } else if (error == ENOTSUP) { 2495 spa->spa_feat_refcount_cache[i] = 2496 SPA_FEATURE_DISABLED; 2497 } else { 2498 return (spa_vdev_err(rvd, 2499 VDEV_AUX_CORRUPT_DATA, EIO)); 2500 } 2501 } 2502 } 2503 2504 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2505 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2506 &spa->spa_feat_enabled_txg_obj) != 0) 2507 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2508 } 2509 2510 spa->spa_is_initializing = B_TRUE; 2511 error = dsl_pool_open(spa->spa_dsl_pool); 2512 spa->spa_is_initializing = B_FALSE; 2513 if (error != 0) 2514 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2515 2516 if (!mosconfig) { 2517 uint64_t hostid; 2518 nvlist_t *policy = NULL, *nvconfig; 2519 2520 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2521 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2522 2523 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2524 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2525 char *hostname; 2526 unsigned long myhostid = 0; 2527 2528 VERIFY(nvlist_lookup_string(nvconfig, 2529 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2530 2531#ifdef _KERNEL 2532 myhostid = zone_get_hostid(NULL); 2533#else /* _KERNEL */ 2534 /* 2535 * We're emulating the system's hostid in userland, so 2536 * we can't use zone_get_hostid(). 2537 */ 2538 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2539#endif /* _KERNEL */ 2540 if (check_hostid && hostid != 0 && myhostid != 0 && 2541 hostid != myhostid) { 2542 nvlist_free(nvconfig); 2543 cmn_err(CE_WARN, "pool '%s' could not be " 2544 "loaded as it was last accessed by " 2545 "another system (host: %s hostid: 0x%lx). " 2546 "See: http://illumos.org/msg/ZFS-8000-EY", 2547 spa_name(spa), hostname, 2548 (unsigned long)hostid); 2549 return (SET_ERROR(EBADF)); 2550 } 2551 } 2552 if (nvlist_lookup_nvlist(spa->spa_config, 2553 ZPOOL_REWIND_POLICY, &policy) == 0) 2554 VERIFY(nvlist_add_nvlist(nvconfig, 2555 ZPOOL_REWIND_POLICY, policy) == 0); 2556 2557 spa_config_set(spa, nvconfig); 2558 spa_unload(spa); 2559 spa_deactivate(spa); 2560 spa_activate(spa, orig_mode); 2561 2562 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2563 } 2564 2565 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2566 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2567 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2568 if (error != 0) 2569 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2570 2571 /* 2572 * Load the bit that tells us to use the new accounting function 2573 * (raid-z deflation). If we have an older pool, this will not 2574 * be present. 2575 */ 2576 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2577 if (error != 0 && error != ENOENT) 2578 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2579 2580 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2581 &spa->spa_creation_version); 2582 if (error != 0 && error != ENOENT) 2583 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2584 2585 /* 2586 * Load the persistent error log. If we have an older pool, this will 2587 * not be present. 2588 */ 2589 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2590 if (error != 0 && error != ENOENT) 2591 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2592 2593 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2594 &spa->spa_errlog_scrub); 2595 if (error != 0 && error != ENOENT) 2596 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2597 2598 /* 2599 * Load the history object. If we have an older pool, this 2600 * will not be present. 2601 */ 2602 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2603 if (error != 0 && error != ENOENT) 2604 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2605 2606 /* 2607 * If we're assembling the pool from the split-off vdevs of 2608 * an existing pool, we don't want to attach the spares & cache 2609 * devices. 2610 */ 2611 2612 /* 2613 * Load any hot spares for this pool. 2614 */ 2615 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2616 if (error != 0 && error != ENOENT) 2617 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2618 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2619 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2620 if (load_nvlist(spa, spa->spa_spares.sav_object, 2621 &spa->spa_spares.sav_config) != 0) 2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2623 2624 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2625 spa_load_spares(spa); 2626 spa_config_exit(spa, SCL_ALL, FTAG); 2627 } else if (error == 0) { 2628 spa->spa_spares.sav_sync = B_TRUE; 2629 } 2630 2631 /* 2632 * Load any level 2 ARC devices for this pool. 2633 */ 2634 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2635 &spa->spa_l2cache.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_L2CACHE); 2640 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2641 &spa->spa_l2cache.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_l2cache(spa); 2646 spa_config_exit(spa, SCL_ALL, FTAG); 2647 } else if (error == 0) { 2648 spa->spa_l2cache.sav_sync = B_TRUE; 2649 } 2650 2651 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2652 2653 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2654 if (error && error != ENOENT) 2655 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2656 2657 if (error == 0) { 2658 uint64_t autoreplace; 2659 2660 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2661 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2662 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2663 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2664 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2665 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2666 &spa->spa_dedup_ditto); 2667 2668 spa->spa_autoreplace = (autoreplace != 0); 2669 } 2670 2671 /* 2672 * If the 'autoreplace' property is set, then post a resource notifying 2673 * the ZFS DE that it should not issue any faults for unopenable 2674 * devices. We also iterate over the vdevs, and post a sysevent for any 2675 * unopenable vdevs so that the normal autoreplace handler can take 2676 * over. 2677 */ 2678 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2679 spa_check_removed(spa->spa_root_vdev); 2680 /* 2681 * For the import case, this is done in spa_import(), because 2682 * at this point we're using the spare definitions from 2683 * the MOS config, not necessarily from the userland config. 2684 */ 2685 if (state != SPA_LOAD_IMPORT) { 2686 spa_aux_check_removed(&spa->spa_spares); 2687 spa_aux_check_removed(&spa->spa_l2cache); 2688 } 2689 } 2690 2691 /* 2692 * Load the vdev state for all toplevel vdevs. 2693 */ 2694 vdev_load(rvd); 2695 2696 /* 2697 * Propagate the leaf DTLs we just loaded all the way up the tree. 2698 */ 2699 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2700 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2701 spa_config_exit(spa, SCL_ALL, FTAG); 2702 2703 /* 2704 * Load the DDTs (dedup tables). 2705 */ 2706 error = ddt_load(spa); 2707 if (error != 0) 2708 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2709 2710 spa_update_dspace(spa); 2711 2712 /* 2713 * Validate the config, using the MOS config to fill in any 2714 * information which might be missing. If we fail to validate 2715 * the config then declare the pool unfit for use. If we're 2716 * assembling a pool from a split, the log is not transferred 2717 * over. 2718 */ 2719 if (type != SPA_IMPORT_ASSEMBLE) { 2720 nvlist_t *nvconfig; 2721 2722 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2723 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2724 2725 if (!spa_config_valid(spa, nvconfig)) { 2726 nvlist_free(nvconfig); 2727 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2728 ENXIO)); 2729 } 2730 nvlist_free(nvconfig); 2731 2732 /* 2733 * Now that we've validated the config, check the state of the 2734 * root vdev. If it can't be opened, it indicates one or 2735 * more toplevel vdevs are faulted. 2736 */ 2737 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2738 return (SET_ERROR(ENXIO)); 2739 2740 if (spa_check_logs(spa)) { 2741 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2742 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2743 } 2744 } 2745 2746 if (missing_feat_write) { 2747 ASSERT(state == SPA_LOAD_TRYIMPORT); 2748 2749 /* 2750 * At this point, we know that we can open the pool in 2751 * read-only mode but not read-write mode. We now have enough 2752 * information and can return to userland. 2753 */ 2754 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2755 } 2756 2757 /* 2758 * We've successfully opened the pool, verify that we're ready 2759 * to start pushing transactions. 2760 */ 2761 if (state != SPA_LOAD_TRYIMPORT) { 2762 if (error = spa_load_verify(spa)) 2763 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2764 error)); 2765 } 2766 2767 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2768 spa->spa_load_max_txg == UINT64_MAX)) { 2769 dmu_tx_t *tx; 2770 int need_update = B_FALSE; 2771 2772 ASSERT(state != SPA_LOAD_TRYIMPORT); 2773 2774 /* 2775 * Claim log blocks that haven't been committed yet. 2776 * This must all happen in a single txg. 2777 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2778 * invoked from zil_claim_log_block()'s i/o done callback. 2779 * Price of rollback is that we abandon the log. 2780 */ 2781 spa->spa_claiming = B_TRUE; 2782 2783 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2784 spa_first_txg(spa)); 2785 (void) dmu_objset_find(spa_name(spa), 2786 zil_claim, tx, DS_FIND_CHILDREN); 2787 dmu_tx_commit(tx); 2788 2789 spa->spa_claiming = B_FALSE; 2790 2791 spa_set_log_state(spa, SPA_LOG_GOOD); 2792 spa->spa_sync_on = B_TRUE; 2793 txg_sync_start(spa->spa_dsl_pool); 2794 2795 /* 2796 * Wait for all claims to sync. We sync up to the highest 2797 * claimed log block birth time so that claimed log blocks 2798 * don't appear to be from the future. spa_claim_max_txg 2799 * will have been set for us by either zil_check_log_chain() 2800 * (invoked from spa_check_logs()) or zil_claim() above. 2801 */ 2802 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2803 2804 /* 2805 * If the config cache is stale, or we have uninitialized 2806 * metaslabs (see spa_vdev_add()), then update the config. 2807 * 2808 * If this is a verbatim import, trust the current 2809 * in-core spa_config and update the disk labels. 2810 */ 2811 if (config_cache_txg != spa->spa_config_txg || 2812 state == SPA_LOAD_IMPORT || 2813 state == SPA_LOAD_RECOVER || 2814 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2815 need_update = B_TRUE; 2816 2817 for (int c = 0; c < rvd->vdev_children; c++) 2818 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2819 need_update = B_TRUE; 2820 2821 /* 2822 * Update the config cache asychronously in case we're the 2823 * root pool, in which case the config cache isn't writable yet. 2824 */ 2825 if (need_update) 2826 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2827 2828 /* 2829 * Check all DTLs to see if anything needs resilvering. 2830 */ 2831 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2832 vdev_resilver_needed(rvd, NULL, NULL)) 2833 spa_async_request(spa, SPA_ASYNC_RESILVER); 2834 2835 /* 2836 * Log the fact that we booted up (so that we can detect if 2837 * we rebooted in the middle of an operation). 2838 */ 2839 spa_history_log_version(spa, "open"); 2840 2841 /* 2842 * Delete any inconsistent datasets. 2843 */ 2844 (void) dmu_objset_find(spa_name(spa), 2845 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2846 2847 /* 2848 * Clean up any stale temporary dataset userrefs. 2849 */ 2850 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2851 } 2852 2853 return (0); 2854} 2855 2856static int 2857spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2858{ 2859 int mode = spa->spa_mode; 2860 2861 spa_unload(spa); 2862 spa_deactivate(spa); 2863 2864 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2865 2866 spa_activate(spa, mode); 2867 spa_async_suspend(spa); 2868 2869 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2870} 2871 2872/* 2873 * If spa_load() fails this function will try loading prior txg's. If 2874 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2875 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2876 * function will not rewind the pool and will return the same error as 2877 * spa_load(). 2878 */ 2879static int 2880spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2881 uint64_t max_request, int rewind_flags) 2882{ 2883 nvlist_t *loadinfo = NULL; 2884 nvlist_t *config = NULL; 2885 int load_error, rewind_error; 2886 uint64_t safe_rewind_txg; 2887 uint64_t min_txg; 2888 2889 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2890 spa->spa_load_max_txg = spa->spa_load_txg; 2891 spa_set_log_state(spa, SPA_LOG_CLEAR); 2892 } else { 2893 spa->spa_load_max_txg = max_request; 2894 if (max_request != UINT64_MAX) 2895 spa->spa_extreme_rewind = B_TRUE; 2896 } 2897 2898 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2899 mosconfig); 2900 if (load_error == 0) 2901 return (0); 2902 2903 if (spa->spa_root_vdev != NULL) 2904 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2905 2906 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2907 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2908 2909 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2910 nvlist_free(config); 2911 return (load_error); 2912 } 2913 2914 if (state == SPA_LOAD_RECOVER) { 2915 /* Price of rolling back is discarding txgs, including log */ 2916 spa_set_log_state(spa, SPA_LOG_CLEAR); 2917 } else { 2918 /* 2919 * If we aren't rolling back save the load info from our first 2920 * import attempt so that we can restore it after attempting 2921 * to rewind. 2922 */ 2923 loadinfo = spa->spa_load_info; 2924 spa->spa_load_info = fnvlist_alloc(); 2925 } 2926 2927 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2928 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2929 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2930 TXG_INITIAL : safe_rewind_txg; 2931 2932 /* 2933 * Continue as long as we're finding errors, we're still within 2934 * the acceptable rewind range, and we're still finding uberblocks 2935 */ 2936 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2937 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2938 if (spa->spa_load_max_txg < safe_rewind_txg) 2939 spa->spa_extreme_rewind = B_TRUE; 2940 rewind_error = spa_load_retry(spa, state, mosconfig); 2941 } 2942 2943 spa->spa_extreme_rewind = B_FALSE; 2944 spa->spa_load_max_txg = UINT64_MAX; 2945 2946 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2947 spa_config_set(spa, config); 2948 2949 if (state == SPA_LOAD_RECOVER) { 2950 ASSERT3P(loadinfo, ==, NULL); 2951 return (rewind_error); 2952 } else { 2953 /* Store the rewind info as part of the initial load info */ 2954 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2955 spa->spa_load_info); 2956 2957 /* Restore the initial load info */ 2958 fnvlist_free(spa->spa_load_info); 2959 spa->spa_load_info = loadinfo; 2960 2961 return (load_error); 2962 } 2963} 2964 2965/* 2966 * Pool Open/Import 2967 * 2968 * The import case is identical to an open except that the configuration is sent 2969 * down from userland, instead of grabbed from the configuration cache. For the 2970 * case of an open, the pool configuration will exist in the 2971 * POOL_STATE_UNINITIALIZED state. 2972 * 2973 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2974 * the same time open the pool, without having to keep around the spa_t in some 2975 * ambiguous state. 2976 */ 2977static int 2978spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2979 nvlist_t **config) 2980{ 2981 spa_t *spa; 2982 spa_load_state_t state = SPA_LOAD_OPEN; 2983 int error; 2984 int locked = B_FALSE; 2985 int firstopen = B_FALSE; 2986 2987 *spapp = NULL; 2988 2989 /* 2990 * As disgusting as this is, we need to support recursive calls to this 2991 * function because dsl_dir_open() is called during spa_load(), and ends 2992 * up calling spa_open() again. The real fix is to figure out how to 2993 * avoid dsl_dir_open() calling this in the first place. 2994 */ 2995 if (mutex_owner(&spa_namespace_lock) != curthread) { 2996 mutex_enter(&spa_namespace_lock); 2997 locked = B_TRUE; 2998 } 2999 3000 if ((spa = spa_lookup(pool)) == NULL) { 3001 if (locked) 3002 mutex_exit(&spa_namespace_lock); 3003 return (SET_ERROR(ENOENT)); 3004 } 3005 3006 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3007 zpool_rewind_policy_t policy; 3008 3009 firstopen = B_TRUE; 3010 3011 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3012 &policy); 3013 if (policy.zrp_request & ZPOOL_DO_REWIND) 3014 state = SPA_LOAD_RECOVER; 3015 3016 spa_activate(spa, spa_mode_global); 3017 3018 if (state != SPA_LOAD_RECOVER) 3019 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3020 3021 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3022 policy.zrp_request); 3023 3024 if (error == EBADF) { 3025 /* 3026 * If vdev_validate() returns failure (indicated by 3027 * EBADF), it indicates that one of the vdevs indicates 3028 * that the pool has been exported or destroyed. If 3029 * this is the case, the config cache is out of sync and 3030 * we should remove the pool from the namespace. 3031 */ 3032 spa_unload(spa); 3033 spa_deactivate(spa); 3034 spa_config_sync(spa, B_TRUE, B_TRUE); 3035 spa_remove(spa); 3036 if (locked) 3037 mutex_exit(&spa_namespace_lock); 3038 return (SET_ERROR(ENOENT)); 3039 } 3040 3041 if (error) { 3042 /* 3043 * We can't open the pool, but we still have useful 3044 * information: the state of each vdev after the 3045 * attempted vdev_open(). Return this to the user. 3046 */ 3047 if (config != NULL && spa->spa_config) { 3048 VERIFY(nvlist_dup(spa->spa_config, config, 3049 KM_SLEEP) == 0); 3050 VERIFY(nvlist_add_nvlist(*config, 3051 ZPOOL_CONFIG_LOAD_INFO, 3052 spa->spa_load_info) == 0); 3053 } 3054 spa_unload(spa); 3055 spa_deactivate(spa); 3056 spa->spa_last_open_failed = error; 3057 if (locked) 3058 mutex_exit(&spa_namespace_lock); 3059 *spapp = NULL; 3060 return (error); 3061 } 3062 } 3063 3064 spa_open_ref(spa, tag); 3065 3066 if (config != NULL) 3067 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3068 3069 /* 3070 * If we've recovered the pool, pass back any information we 3071 * gathered while doing the load. 3072 */ 3073 if (state == SPA_LOAD_RECOVER) { 3074 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3075 spa->spa_load_info) == 0); 3076 } 3077 3078 if (locked) { 3079 spa->spa_last_open_failed = 0; 3080 spa->spa_last_ubsync_txg = 0; 3081 spa->spa_load_txg = 0; 3082 mutex_exit(&spa_namespace_lock); 3083#ifdef __FreeBSD__ 3084#ifdef _KERNEL 3085 if (firstopen) 3086 zvol_create_minors(spa->spa_name); 3087#endif 3088#endif 3089 } 3090 3091 *spapp = spa; 3092 3093 return (0); 3094} 3095 3096int 3097spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3098 nvlist_t **config) 3099{ 3100 return (spa_open_common(name, spapp, tag, policy, config)); 3101} 3102 3103int 3104spa_open(const char *name, spa_t **spapp, void *tag) 3105{ 3106 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3107} 3108 3109/* 3110 * Lookup the given spa_t, incrementing the inject count in the process, 3111 * preventing it from being exported or destroyed. 3112 */ 3113spa_t * 3114spa_inject_addref(char *name) 3115{ 3116 spa_t *spa; 3117 3118 mutex_enter(&spa_namespace_lock); 3119 if ((spa = spa_lookup(name)) == NULL) { 3120 mutex_exit(&spa_namespace_lock); 3121 return (NULL); 3122 } 3123 spa->spa_inject_ref++; 3124 mutex_exit(&spa_namespace_lock); 3125 3126 return (spa); 3127} 3128 3129void 3130spa_inject_delref(spa_t *spa) 3131{ 3132 mutex_enter(&spa_namespace_lock); 3133 spa->spa_inject_ref--; 3134 mutex_exit(&spa_namespace_lock); 3135} 3136 3137/* 3138 * Add spares device information to the nvlist. 3139 */ 3140static void 3141spa_add_spares(spa_t *spa, nvlist_t *config) 3142{ 3143 nvlist_t **spares; 3144 uint_t i, nspares; 3145 nvlist_t *nvroot; 3146 uint64_t guid; 3147 vdev_stat_t *vs; 3148 uint_t vsc; 3149 uint64_t pool; 3150 3151 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3152 3153 if (spa->spa_spares.sav_count == 0) 3154 return; 3155 3156 VERIFY(nvlist_lookup_nvlist(config, 3157 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3158 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3159 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3160 if (nspares != 0) { 3161 VERIFY(nvlist_add_nvlist_array(nvroot, 3162 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3163 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3164 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3165 3166 /* 3167 * Go through and find any spares which have since been 3168 * repurposed as an active spare. If this is the case, update 3169 * their status appropriately. 3170 */ 3171 for (i = 0; i < nspares; i++) { 3172 VERIFY(nvlist_lookup_uint64(spares[i], 3173 ZPOOL_CONFIG_GUID, &guid) == 0); 3174 if (spa_spare_exists(guid, &pool, NULL) && 3175 pool != 0ULL) { 3176 VERIFY(nvlist_lookup_uint64_array( 3177 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3178 (uint64_t **)&vs, &vsc) == 0); 3179 vs->vs_state = VDEV_STATE_CANT_OPEN; 3180 vs->vs_aux = VDEV_AUX_SPARED; 3181 } 3182 } 3183 } 3184} 3185 3186/* 3187 * Add l2cache device information to the nvlist, including vdev stats. 3188 */ 3189static void 3190spa_add_l2cache(spa_t *spa, nvlist_t *config) 3191{ 3192 nvlist_t **l2cache; 3193 uint_t i, j, nl2cache; 3194 nvlist_t *nvroot; 3195 uint64_t guid; 3196 vdev_t *vd; 3197 vdev_stat_t *vs; 3198 uint_t vsc; 3199 3200 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3201 3202 if (spa->spa_l2cache.sav_count == 0) 3203 return; 3204 3205 VERIFY(nvlist_lookup_nvlist(config, 3206 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3207 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3208 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3209 if (nl2cache != 0) { 3210 VERIFY(nvlist_add_nvlist_array(nvroot, 3211 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3212 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3213 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3214 3215 /* 3216 * Update level 2 cache device stats. 3217 */ 3218 3219 for (i = 0; i < nl2cache; i++) { 3220 VERIFY(nvlist_lookup_uint64(l2cache[i], 3221 ZPOOL_CONFIG_GUID, &guid) == 0); 3222 3223 vd = NULL; 3224 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3225 if (guid == 3226 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3227 vd = spa->spa_l2cache.sav_vdevs[j]; 3228 break; 3229 } 3230 } 3231 ASSERT(vd != NULL); 3232 3233 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3234 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3235 == 0); 3236 vdev_get_stats(vd, vs); 3237 } 3238 } 3239} 3240 3241static void 3242spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3243{ 3244 nvlist_t *features; 3245 zap_cursor_t zc; 3246 zap_attribute_t za; 3247 3248 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3249 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3250 3251 /* We may be unable to read features if pool is suspended. */ 3252 if (spa_suspended(spa)) 3253 goto out; 3254 3255 if (spa->spa_feat_for_read_obj != 0) { 3256 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3257 spa->spa_feat_for_read_obj); 3258 zap_cursor_retrieve(&zc, &za) == 0; 3259 zap_cursor_advance(&zc)) { 3260 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3261 za.za_num_integers == 1); 3262 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3263 za.za_first_integer)); 3264 } 3265 zap_cursor_fini(&zc); 3266 } 3267 3268 if (spa->spa_feat_for_write_obj != 0) { 3269 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3270 spa->spa_feat_for_write_obj); 3271 zap_cursor_retrieve(&zc, &za) == 0; 3272 zap_cursor_advance(&zc)) { 3273 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3274 za.za_num_integers == 1); 3275 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3276 za.za_first_integer)); 3277 } 3278 zap_cursor_fini(&zc); 3279 } 3280 3281out: 3282 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3283 features) == 0); 3284 nvlist_free(features); 3285} 3286 3287int 3288spa_get_stats(const char *name, nvlist_t **config, 3289 char *altroot, size_t buflen) 3290{ 3291 int error; 3292 spa_t *spa; 3293 3294 *config = NULL; 3295 error = spa_open_common(name, &spa, FTAG, NULL, config); 3296 3297 if (spa != NULL) { 3298 /* 3299 * This still leaves a window of inconsistency where the spares 3300 * or l2cache devices could change and the config would be 3301 * self-inconsistent. 3302 */ 3303 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3304 3305 if (*config != NULL) { 3306 uint64_t loadtimes[2]; 3307 3308 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3309 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3310 VERIFY(nvlist_add_uint64_array(*config, 3311 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3312 3313 VERIFY(nvlist_add_uint64(*config, 3314 ZPOOL_CONFIG_ERRCOUNT, 3315 spa_get_errlog_size(spa)) == 0); 3316 3317 if (spa_suspended(spa)) 3318 VERIFY(nvlist_add_uint64(*config, 3319 ZPOOL_CONFIG_SUSPENDED, 3320 spa->spa_failmode) == 0); 3321 3322 spa_add_spares(spa, *config); 3323 spa_add_l2cache(spa, *config); 3324 spa_add_feature_stats(spa, *config); 3325 } 3326 } 3327 3328 /* 3329 * We want to get the alternate root even for faulted pools, so we cheat 3330 * and call spa_lookup() directly. 3331 */ 3332 if (altroot) { 3333 if (spa == NULL) { 3334 mutex_enter(&spa_namespace_lock); 3335 spa = spa_lookup(name); 3336 if (spa) 3337 spa_altroot(spa, altroot, buflen); 3338 else 3339 altroot[0] = '\0'; 3340 spa = NULL; 3341 mutex_exit(&spa_namespace_lock); 3342 } else { 3343 spa_altroot(spa, altroot, buflen); 3344 } 3345 } 3346 3347 if (spa != NULL) { 3348 spa_config_exit(spa, SCL_CONFIG, FTAG); 3349 spa_close(spa, FTAG); 3350 } 3351 3352 return (error); 3353} 3354 3355/* 3356 * Validate that the auxiliary device array is well formed. We must have an 3357 * array of nvlists, each which describes a valid leaf vdev. If this is an 3358 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3359 * specified, as long as they are well-formed. 3360 */ 3361static int 3362spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3363 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3364 vdev_labeltype_t label) 3365{ 3366 nvlist_t **dev; 3367 uint_t i, ndev; 3368 vdev_t *vd; 3369 int error; 3370 3371 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3372 3373 /* 3374 * It's acceptable to have no devs specified. 3375 */ 3376 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3377 return (0); 3378 3379 if (ndev == 0) 3380 return (SET_ERROR(EINVAL)); 3381 3382 /* 3383 * Make sure the pool is formatted with a version that supports this 3384 * device type. 3385 */ 3386 if (spa_version(spa) < version) 3387 return (SET_ERROR(ENOTSUP)); 3388 3389 /* 3390 * Set the pending device list so we correctly handle device in-use 3391 * checking. 3392 */ 3393 sav->sav_pending = dev; 3394 sav->sav_npending = ndev; 3395 3396 for (i = 0; i < ndev; i++) { 3397 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3398 mode)) != 0) 3399 goto out; 3400 3401 if (!vd->vdev_ops->vdev_op_leaf) { 3402 vdev_free(vd); 3403 error = SET_ERROR(EINVAL); 3404 goto out; 3405 } 3406 3407 /* 3408 * The L2ARC currently only supports disk devices in 3409 * kernel context. For user-level testing, we allow it. 3410 */ 3411#ifdef _KERNEL 3412 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3413 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3414 error = SET_ERROR(ENOTBLK); 3415 vdev_free(vd); 3416 goto out; 3417 } 3418#endif 3419 vd->vdev_top = vd; 3420 3421 if ((error = vdev_open(vd)) == 0 && 3422 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3423 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3424 vd->vdev_guid) == 0); 3425 } 3426 3427 vdev_free(vd); 3428 3429 if (error && 3430 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3431 goto out; 3432 else 3433 error = 0; 3434 } 3435 3436out: 3437 sav->sav_pending = NULL; 3438 sav->sav_npending = 0; 3439 return (error); 3440} 3441 3442static int 3443spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3444{ 3445 int error; 3446 3447 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3448 3449 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3450 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3451 VDEV_LABEL_SPARE)) != 0) { 3452 return (error); 3453 } 3454 3455 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3456 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3457 VDEV_LABEL_L2CACHE)); 3458} 3459 3460static void 3461spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3462 const char *config) 3463{ 3464 int i; 3465 3466 if (sav->sav_config != NULL) { 3467 nvlist_t **olddevs; 3468 uint_t oldndevs; 3469 nvlist_t **newdevs; 3470 3471 /* 3472 * Generate new dev list by concatentating with the 3473 * current dev list. 3474 */ 3475 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3476 &olddevs, &oldndevs) == 0); 3477 3478 newdevs = kmem_alloc(sizeof (void *) * 3479 (ndevs + oldndevs), KM_SLEEP); 3480 for (i = 0; i < oldndevs; i++) 3481 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3482 KM_SLEEP) == 0); 3483 for (i = 0; i < ndevs; i++) 3484 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3485 KM_SLEEP) == 0); 3486 3487 VERIFY(nvlist_remove(sav->sav_config, config, 3488 DATA_TYPE_NVLIST_ARRAY) == 0); 3489 3490 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3491 config, newdevs, ndevs + oldndevs) == 0); 3492 for (i = 0; i < oldndevs + ndevs; i++) 3493 nvlist_free(newdevs[i]); 3494 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3495 } else { 3496 /* 3497 * Generate a new dev list. 3498 */ 3499 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3500 KM_SLEEP) == 0); 3501 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3502 devs, ndevs) == 0); 3503 } 3504} 3505 3506/* 3507 * Stop and drop level 2 ARC devices 3508 */ 3509void 3510spa_l2cache_drop(spa_t *spa) 3511{ 3512 vdev_t *vd; 3513 int i; 3514 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3515 3516 for (i = 0; i < sav->sav_count; i++) { 3517 uint64_t pool; 3518 3519 vd = sav->sav_vdevs[i]; 3520 ASSERT(vd != NULL); 3521 3522 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3523 pool != 0ULL && l2arc_vdev_present(vd)) 3524 l2arc_remove_vdev(vd); 3525 } 3526} 3527 3528/* 3529 * Pool Creation 3530 */ 3531int 3532spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3533 nvlist_t *zplprops) 3534{ 3535 spa_t *spa; 3536 char *altroot = NULL; 3537 vdev_t *rvd; 3538 dsl_pool_t *dp; 3539 dmu_tx_t *tx; 3540 int error = 0; 3541 uint64_t txg = TXG_INITIAL; 3542 nvlist_t **spares, **l2cache; 3543 uint_t nspares, nl2cache; 3544 uint64_t version, obj; 3545 boolean_t has_features; 3546 3547 /* 3548 * If this pool already exists, return failure. 3549 */ 3550 mutex_enter(&spa_namespace_lock); 3551 if (spa_lookup(pool) != NULL) { 3552 mutex_exit(&spa_namespace_lock); 3553 return (SET_ERROR(EEXIST)); 3554 } 3555 3556 /* 3557 * Allocate a new spa_t structure. 3558 */ 3559 (void) nvlist_lookup_string(props, 3560 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3561 spa = spa_add(pool, NULL, altroot); 3562 spa_activate(spa, spa_mode_global); 3563 3564 if (props && (error = spa_prop_validate(spa, props))) { 3565 spa_deactivate(spa); 3566 spa_remove(spa); 3567 mutex_exit(&spa_namespace_lock); 3568 return (error); 3569 } 3570 3571 has_features = B_FALSE; 3572 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3573 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3574 if (zpool_prop_feature(nvpair_name(elem))) 3575 has_features = B_TRUE; 3576 } 3577 3578 if (has_features || nvlist_lookup_uint64(props, 3579 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3580 version = SPA_VERSION; 3581 } 3582 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3583 3584 spa->spa_first_txg = txg; 3585 spa->spa_uberblock.ub_txg = txg - 1; 3586 spa->spa_uberblock.ub_version = version; 3587 spa->spa_ubsync = spa->spa_uberblock; 3588 3589 /* 3590 * Create "The Godfather" zio to hold all async IOs 3591 */ 3592 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3593 KM_SLEEP); 3594 for (int i = 0; i < max_ncpus; i++) { 3595 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3596 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3597 ZIO_FLAG_GODFATHER); 3598 } 3599 3600 /* 3601 * Create the root vdev. 3602 */ 3603 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3604 3605 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3606 3607 ASSERT(error != 0 || rvd != NULL); 3608 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3609 3610 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3611 error = SET_ERROR(EINVAL); 3612 3613 if (error == 0 && 3614 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3615 (error = spa_validate_aux(spa, nvroot, txg, 3616 VDEV_ALLOC_ADD)) == 0) { 3617 for (int c = 0; c < rvd->vdev_children; c++) { 3618 vdev_ashift_optimize(rvd->vdev_child[c]); 3619 vdev_metaslab_set_size(rvd->vdev_child[c]); 3620 vdev_expand(rvd->vdev_child[c], txg); 3621 } 3622 } 3623 3624 spa_config_exit(spa, SCL_ALL, FTAG); 3625 3626 if (error != 0) { 3627 spa_unload(spa); 3628 spa_deactivate(spa); 3629 spa_remove(spa); 3630 mutex_exit(&spa_namespace_lock); 3631 return (error); 3632 } 3633 3634 /* 3635 * Get the list of spares, if specified. 3636 */ 3637 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3638 &spares, &nspares) == 0) { 3639 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3640 KM_SLEEP) == 0); 3641 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3642 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3643 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3644 spa_load_spares(spa); 3645 spa_config_exit(spa, SCL_ALL, FTAG); 3646 spa->spa_spares.sav_sync = B_TRUE; 3647 } 3648 3649 /* 3650 * Get the list of level 2 cache devices, if specified. 3651 */ 3652 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3653 &l2cache, &nl2cache) == 0) { 3654 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3655 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3656 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3657 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3658 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3659 spa_load_l2cache(spa); 3660 spa_config_exit(spa, SCL_ALL, FTAG); 3661 spa->spa_l2cache.sav_sync = B_TRUE; 3662 } 3663 3664 spa->spa_is_initializing = B_TRUE; 3665 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3666 spa->spa_meta_objset = dp->dp_meta_objset; 3667 spa->spa_is_initializing = B_FALSE; 3668 3669 /* 3670 * Create DDTs (dedup tables). 3671 */ 3672 ddt_create(spa); 3673 3674 spa_update_dspace(spa); 3675 3676 tx = dmu_tx_create_assigned(dp, txg); 3677 3678 /* 3679 * Create the pool config object. 3680 */ 3681 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3682 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3683 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3684 3685 if (zap_add(spa->spa_meta_objset, 3686 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3687 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3688 cmn_err(CE_PANIC, "failed to add pool config"); 3689 } 3690 3691 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3692 spa_feature_create_zap_objects(spa, tx); 3693 3694 if (zap_add(spa->spa_meta_objset, 3695 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3696 sizeof (uint64_t), 1, &version, tx) != 0) { 3697 cmn_err(CE_PANIC, "failed to add pool version"); 3698 } 3699 3700 /* Newly created pools with the right version are always deflated. */ 3701 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3702 spa->spa_deflate = TRUE; 3703 if (zap_add(spa->spa_meta_objset, 3704 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3705 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3706 cmn_err(CE_PANIC, "failed to add deflate"); 3707 } 3708 } 3709 3710 /* 3711 * Create the deferred-free bpobj. Turn off compression 3712 * because sync-to-convergence takes longer if the blocksize 3713 * keeps changing. 3714 */ 3715 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3716 dmu_object_set_compress(spa->spa_meta_objset, obj, 3717 ZIO_COMPRESS_OFF, tx); 3718 if (zap_add(spa->spa_meta_objset, 3719 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3720 sizeof (uint64_t), 1, &obj, tx) != 0) { 3721 cmn_err(CE_PANIC, "failed to add bpobj"); 3722 } 3723 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3724 spa->spa_meta_objset, obj)); 3725 3726 /* 3727 * Create the pool's history object. 3728 */ 3729 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3730 spa_history_create_obj(spa, tx); 3731 3732 /* 3733 * Set pool properties. 3734 */ 3735 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3736 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3737 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3738 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3739 3740 if (props != NULL) { 3741 spa_configfile_set(spa, props, B_FALSE); 3742 spa_sync_props(props, tx); 3743 } 3744 3745 dmu_tx_commit(tx); 3746 3747 spa->spa_sync_on = B_TRUE; 3748 txg_sync_start(spa->spa_dsl_pool); 3749 3750 /* 3751 * We explicitly wait for the first transaction to complete so that our 3752 * bean counters are appropriately updated. 3753 */ 3754 txg_wait_synced(spa->spa_dsl_pool, txg); 3755 3756 spa_config_sync(spa, B_FALSE, B_TRUE); 3757 3758 spa_history_log_version(spa, "create"); 3759 3760 spa->spa_minref = refcount_count(&spa->spa_refcount); 3761 3762 mutex_exit(&spa_namespace_lock); 3763 3764 return (0); 3765} 3766 3767#ifdef _KERNEL 3768#if defined(sun) 3769/* 3770 * Get the root pool information from the root disk, then import the root pool 3771 * during the system boot up time. 3772 */ 3773extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3774 3775static nvlist_t * 3776spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3777{ 3778 nvlist_t *config; 3779 nvlist_t *nvtop, *nvroot; 3780 uint64_t pgid; 3781 3782 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3783 return (NULL); 3784 3785 /* 3786 * Add this top-level vdev to the child array. 3787 */ 3788 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3789 &nvtop) == 0); 3790 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3791 &pgid) == 0); 3792 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3793 3794 /* 3795 * Put this pool's top-level vdevs into a root vdev. 3796 */ 3797 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3798 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3799 VDEV_TYPE_ROOT) == 0); 3800 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3801 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3802 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3803 &nvtop, 1) == 0); 3804 3805 /* 3806 * Replace the existing vdev_tree with the new root vdev in 3807 * this pool's configuration (remove the old, add the new). 3808 */ 3809 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3810 nvlist_free(nvroot); 3811 return (config); 3812} 3813 3814/* 3815 * Walk the vdev tree and see if we can find a device with "better" 3816 * configuration. A configuration is "better" if the label on that 3817 * device has a more recent txg. 3818 */ 3819static void 3820spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3821{ 3822 for (int c = 0; c < vd->vdev_children; c++) 3823 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3824 3825 if (vd->vdev_ops->vdev_op_leaf) { 3826 nvlist_t *label; 3827 uint64_t label_txg; 3828 3829 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3830 &label) != 0) 3831 return; 3832 3833 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3834 &label_txg) == 0); 3835 3836 /* 3837 * Do we have a better boot device? 3838 */ 3839 if (label_txg > *txg) { 3840 *txg = label_txg; 3841 *avd = vd; 3842 } 3843 nvlist_free(label); 3844 } 3845} 3846 3847/* 3848 * Import a root pool. 3849 * 3850 * For x86. devpath_list will consist of devid and/or physpath name of 3851 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3852 * The GRUB "findroot" command will return the vdev we should boot. 3853 * 3854 * For Sparc, devpath_list consists the physpath name of the booting device 3855 * no matter the rootpool is a single device pool or a mirrored pool. 3856 * e.g. 3857 * "/pci@1f,0/ide@d/disk@0,0:a" 3858 */ 3859int 3860spa_import_rootpool(char *devpath, char *devid) 3861{ 3862 spa_t *spa; 3863 vdev_t *rvd, *bvd, *avd = NULL; 3864 nvlist_t *config, *nvtop; 3865 uint64_t guid, txg; 3866 char *pname; 3867 int error; 3868 3869 /* 3870 * Read the label from the boot device and generate a configuration. 3871 */ 3872 config = spa_generate_rootconf(devpath, devid, &guid); 3873#if defined(_OBP) && defined(_KERNEL) 3874 if (config == NULL) { 3875 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3876 /* iscsi boot */ 3877 get_iscsi_bootpath_phy(devpath); 3878 config = spa_generate_rootconf(devpath, devid, &guid); 3879 } 3880 } 3881#endif 3882 if (config == NULL) { 3883 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3884 devpath); 3885 return (SET_ERROR(EIO)); 3886 } 3887 3888 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3889 &pname) == 0); 3890 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3891 3892 mutex_enter(&spa_namespace_lock); 3893 if ((spa = spa_lookup(pname)) != NULL) { 3894 /* 3895 * Remove the existing root pool from the namespace so that we 3896 * can replace it with the correct config we just read in. 3897 */ 3898 spa_remove(spa); 3899 } 3900 3901 spa = spa_add(pname, config, NULL); 3902 spa->spa_is_root = B_TRUE; 3903 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3904 3905 /* 3906 * Build up a vdev tree based on the boot device's label config. 3907 */ 3908 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3909 &nvtop) == 0); 3910 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3911 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3912 VDEV_ALLOC_ROOTPOOL); 3913 spa_config_exit(spa, SCL_ALL, FTAG); 3914 if (error) { 3915 mutex_exit(&spa_namespace_lock); 3916 nvlist_free(config); 3917 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3918 pname); 3919 return (error); 3920 } 3921 3922 /* 3923 * Get the boot vdev. 3924 */ 3925 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3926 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3927 (u_longlong_t)guid); 3928 error = SET_ERROR(ENOENT); 3929 goto out; 3930 } 3931 3932 /* 3933 * Determine if there is a better boot device. 3934 */ 3935 avd = bvd; 3936 spa_alt_rootvdev(rvd, &avd, &txg); 3937 if (avd != bvd) { 3938 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3939 "try booting from '%s'", avd->vdev_path); 3940 error = SET_ERROR(EINVAL); 3941 goto out; 3942 } 3943 3944 /* 3945 * If the boot device is part of a spare vdev then ensure that 3946 * we're booting off the active spare. 3947 */ 3948 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3949 !bvd->vdev_isspare) { 3950 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3951 "try booting from '%s'", 3952 bvd->vdev_parent-> 3953 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3954 error = SET_ERROR(EINVAL); 3955 goto out; 3956 } 3957 3958 error = 0; 3959out: 3960 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3961 vdev_free(rvd); 3962 spa_config_exit(spa, SCL_ALL, FTAG); 3963 mutex_exit(&spa_namespace_lock); 3964 3965 nvlist_free(config); 3966 return (error); 3967} 3968 3969#else 3970 3971extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3972 uint64_t *count); 3973 3974static nvlist_t * 3975spa_generate_rootconf(const char *name) 3976{ 3977 nvlist_t **configs, **tops; 3978 nvlist_t *config; 3979 nvlist_t *best_cfg, *nvtop, *nvroot; 3980 uint64_t *holes; 3981 uint64_t best_txg; 3982 uint64_t nchildren; 3983 uint64_t pgid; 3984 uint64_t count; 3985 uint64_t i; 3986 uint_t nholes; 3987 3988 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3989 return (NULL); 3990 3991 ASSERT3U(count, !=, 0); 3992 best_txg = 0; 3993 for (i = 0; i < count; i++) { 3994 uint64_t txg; 3995 3996 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 3997 &txg) == 0); 3998 if (txg > best_txg) { 3999 best_txg = txg; 4000 best_cfg = configs[i]; 4001 } 4002 } 4003 4004 /* 4005 * Multi-vdev root pool configuration discovery is not supported yet. 4006 */ 4007 nchildren = 1; 4008 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4009 holes = NULL; 4010 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4011 &holes, &nholes); 4012 4013 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4014 for (i = 0; i < nchildren; i++) { 4015 if (i >= count) 4016 break; 4017 if (configs[i] == NULL) 4018 continue; 4019 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4020 &nvtop) == 0); 4021 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4022 } 4023 for (i = 0; holes != NULL && i < nholes; i++) { 4024 if (i >= nchildren) 4025 continue; 4026 if (tops[holes[i]] != NULL) 4027 continue; 4028 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4029 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4030 VDEV_TYPE_HOLE) == 0); 4031 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4032 holes[i]) == 0); 4033 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4034 0) == 0); 4035 } 4036 for (i = 0; i < nchildren; i++) { 4037 if (tops[i] != NULL) 4038 continue; 4039 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4040 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4041 VDEV_TYPE_MISSING) == 0); 4042 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4043 i) == 0); 4044 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4045 0) == 0); 4046 } 4047 4048 /* 4049 * Create pool config based on the best vdev config. 4050 */ 4051 nvlist_dup(best_cfg, &config, KM_SLEEP); 4052 4053 /* 4054 * Put this pool's top-level vdevs into a root vdev. 4055 */ 4056 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4057 &pgid) == 0); 4058 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4059 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4060 VDEV_TYPE_ROOT) == 0); 4061 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4062 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4063 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4064 tops, nchildren) == 0); 4065 4066 /* 4067 * Replace the existing vdev_tree with the new root vdev in 4068 * this pool's configuration (remove the old, add the new). 4069 */ 4070 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4071 4072 /* 4073 * Drop vdev config elements that should not be present at pool level. 4074 */ 4075 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4076 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4077 4078 for (i = 0; i < count; i++) 4079 nvlist_free(configs[i]); 4080 kmem_free(configs, count * sizeof(void *)); 4081 for (i = 0; i < nchildren; i++) 4082 nvlist_free(tops[i]); 4083 kmem_free(tops, nchildren * sizeof(void *)); 4084 nvlist_free(nvroot); 4085 return (config); 4086} 4087 4088int 4089spa_import_rootpool(const char *name) 4090{ 4091 spa_t *spa; 4092 vdev_t *rvd, *bvd, *avd = NULL; 4093 nvlist_t *config, *nvtop; 4094 uint64_t txg; 4095 char *pname; 4096 int error; 4097 4098 /* 4099 * Read the label from the boot device and generate a configuration. 4100 */ 4101 config = spa_generate_rootconf(name); 4102 4103 mutex_enter(&spa_namespace_lock); 4104 if (config != NULL) { 4105 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4106 &pname) == 0 && strcmp(name, pname) == 0); 4107 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4108 == 0); 4109 4110 if ((spa = spa_lookup(pname)) != NULL) { 4111 /* 4112 * Remove the existing root pool from the namespace so 4113 * that we can replace it with the correct config 4114 * we just read in. 4115 */ 4116 spa_remove(spa); 4117 } 4118 spa = spa_add(pname, config, NULL); 4119 4120 /* 4121 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4122 * via spa_version(). 4123 */ 4124 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4125 &spa->spa_ubsync.ub_version) != 0) 4126 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4127 } else if ((spa = spa_lookup(name)) == NULL) { 4128 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4129 name); 4130 return (EIO); 4131 } else { 4132 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4133 } 4134 spa->spa_is_root = B_TRUE; 4135 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4136 4137 /* 4138 * Build up a vdev tree based on the boot device's label config. 4139 */ 4140 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4141 &nvtop) == 0); 4142 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4143 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4144 VDEV_ALLOC_ROOTPOOL); 4145 spa_config_exit(spa, SCL_ALL, FTAG); 4146 if (error) { 4147 mutex_exit(&spa_namespace_lock); 4148 nvlist_free(config); 4149 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4150 pname); 4151 return (error); 4152 } 4153 4154 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4155 vdev_free(rvd); 4156 spa_config_exit(spa, SCL_ALL, FTAG); 4157 mutex_exit(&spa_namespace_lock); 4158 4159 nvlist_free(config); 4160 return (0); 4161} 4162 4163#endif /* sun */ 4164#endif 4165 4166/* 4167 * Import a non-root pool into the system. 4168 */ 4169int 4170spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4171{ 4172 spa_t *spa; 4173 char *altroot = NULL; 4174 spa_load_state_t state = SPA_LOAD_IMPORT; 4175 zpool_rewind_policy_t policy; 4176 uint64_t mode = spa_mode_global; 4177 uint64_t readonly = B_FALSE; 4178 int error; 4179 nvlist_t *nvroot; 4180 nvlist_t **spares, **l2cache; 4181 uint_t nspares, nl2cache; 4182 4183 /* 4184 * If a pool with this name exists, return failure. 4185 */ 4186 mutex_enter(&spa_namespace_lock); 4187 if (spa_lookup(pool) != NULL) { 4188 mutex_exit(&spa_namespace_lock); 4189 return (SET_ERROR(EEXIST)); 4190 } 4191 4192 /* 4193 * Create and initialize the spa structure. 4194 */ 4195 (void) nvlist_lookup_string(props, 4196 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4197 (void) nvlist_lookup_uint64(props, 4198 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4199 if (readonly) 4200 mode = FREAD; 4201 spa = spa_add(pool, config, altroot); 4202 spa->spa_import_flags = flags; 4203 4204 /* 4205 * Verbatim import - Take a pool and insert it into the namespace 4206 * as if it had been loaded at boot. 4207 */ 4208 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4209 if (props != NULL) 4210 spa_configfile_set(spa, props, B_FALSE); 4211 4212 spa_config_sync(spa, B_FALSE, B_TRUE); 4213 4214 mutex_exit(&spa_namespace_lock); 4215 return (0); 4216 } 4217 4218 spa_activate(spa, mode); 4219 4220 /* 4221 * Don't start async tasks until we know everything is healthy. 4222 */ 4223 spa_async_suspend(spa); 4224 4225 zpool_get_rewind_policy(config, &policy); 4226 if (policy.zrp_request & ZPOOL_DO_REWIND) 4227 state = SPA_LOAD_RECOVER; 4228 4229 /* 4230 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4231 * because the user-supplied config is actually the one to trust when 4232 * doing an import. 4233 */ 4234 if (state != SPA_LOAD_RECOVER) 4235 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4236 4237 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4238 policy.zrp_request); 4239 4240 /* 4241 * Propagate anything learned while loading the pool and pass it 4242 * back to caller (i.e. rewind info, missing devices, etc). 4243 */ 4244 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4245 spa->spa_load_info) == 0); 4246 4247 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4248 /* 4249 * Toss any existing sparelist, as it doesn't have any validity 4250 * anymore, and conflicts with spa_has_spare(). 4251 */ 4252 if (spa->spa_spares.sav_config) { 4253 nvlist_free(spa->spa_spares.sav_config); 4254 spa->spa_spares.sav_config = NULL; 4255 spa_load_spares(spa); 4256 } 4257 if (spa->spa_l2cache.sav_config) { 4258 nvlist_free(spa->spa_l2cache.sav_config); 4259 spa->spa_l2cache.sav_config = NULL; 4260 spa_load_l2cache(spa); 4261 } 4262 4263 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4264 &nvroot) == 0); 4265 if (error == 0) 4266 error = spa_validate_aux(spa, nvroot, -1ULL, 4267 VDEV_ALLOC_SPARE); 4268 if (error == 0) 4269 error = spa_validate_aux(spa, nvroot, -1ULL, 4270 VDEV_ALLOC_L2CACHE); 4271 spa_config_exit(spa, SCL_ALL, FTAG); 4272 4273 if (props != NULL) 4274 spa_configfile_set(spa, props, B_FALSE); 4275 4276 if (error != 0 || (props && spa_writeable(spa) && 4277 (error = spa_prop_set(spa, props)))) { 4278 spa_unload(spa); 4279 spa_deactivate(spa); 4280 spa_remove(spa); 4281 mutex_exit(&spa_namespace_lock); 4282 return (error); 4283 } 4284 4285 spa_async_resume(spa); 4286 4287 /* 4288 * Override any spares and level 2 cache devices as specified by 4289 * the user, as these may have correct device names/devids, etc. 4290 */ 4291 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4292 &spares, &nspares) == 0) { 4293 if (spa->spa_spares.sav_config) 4294 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4295 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4296 else 4297 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4298 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4299 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4300 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4302 spa_load_spares(spa); 4303 spa_config_exit(spa, SCL_ALL, FTAG); 4304 spa->spa_spares.sav_sync = B_TRUE; 4305 } 4306 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4307 &l2cache, &nl2cache) == 0) { 4308 if (spa->spa_l2cache.sav_config) 4309 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4310 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4311 else 4312 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4313 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4314 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4315 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4317 spa_load_l2cache(spa); 4318 spa_config_exit(spa, SCL_ALL, FTAG); 4319 spa->spa_l2cache.sav_sync = B_TRUE; 4320 } 4321 4322 /* 4323 * Check for any removed devices. 4324 */ 4325 if (spa->spa_autoreplace) { 4326 spa_aux_check_removed(&spa->spa_spares); 4327 spa_aux_check_removed(&spa->spa_l2cache); 4328 } 4329 4330 if (spa_writeable(spa)) { 4331 /* 4332 * Update the config cache to include the newly-imported pool. 4333 */ 4334 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4335 } 4336 4337 /* 4338 * It's possible that the pool was expanded while it was exported. 4339 * We kick off an async task to handle this for us. 4340 */ 4341 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4342 4343 mutex_exit(&spa_namespace_lock); 4344 spa_history_log_version(spa, "import"); 4345 4346#ifdef __FreeBSD__ 4347#ifdef _KERNEL 4348 zvol_create_minors(pool); 4349#endif 4350#endif 4351 return (0); 4352} 4353 4354nvlist_t * 4355spa_tryimport(nvlist_t *tryconfig) 4356{ 4357 nvlist_t *config = NULL; 4358 char *poolname; 4359 spa_t *spa; 4360 uint64_t state; 4361 int error; 4362 4363 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4364 return (NULL); 4365 4366 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4367 return (NULL); 4368 4369 /* 4370 * Create and initialize the spa structure. 4371 */ 4372 mutex_enter(&spa_namespace_lock); 4373 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4374 spa_activate(spa, FREAD); 4375 4376 /* 4377 * Pass off the heavy lifting to spa_load(). 4378 * Pass TRUE for mosconfig because the user-supplied config 4379 * is actually the one to trust when doing an import. 4380 */ 4381 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4382 4383 /* 4384 * If 'tryconfig' was at least parsable, return the current config. 4385 */ 4386 if (spa->spa_root_vdev != NULL) { 4387 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4388 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4389 poolname) == 0); 4390 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4391 state) == 0); 4392 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4393 spa->spa_uberblock.ub_timestamp) == 0); 4394 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4395 spa->spa_load_info) == 0); 4396 4397 /* 4398 * If the bootfs property exists on this pool then we 4399 * copy it out so that external consumers can tell which 4400 * pools are bootable. 4401 */ 4402 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4403 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4404 4405 /* 4406 * We have to play games with the name since the 4407 * pool was opened as TRYIMPORT_NAME. 4408 */ 4409 if (dsl_dsobj_to_dsname(spa_name(spa), 4410 spa->spa_bootfs, tmpname) == 0) { 4411 char *cp; 4412 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4413 4414 cp = strchr(tmpname, '/'); 4415 if (cp == NULL) { 4416 (void) strlcpy(dsname, tmpname, 4417 MAXPATHLEN); 4418 } else { 4419 (void) snprintf(dsname, MAXPATHLEN, 4420 "%s/%s", poolname, ++cp); 4421 } 4422 VERIFY(nvlist_add_string(config, 4423 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4424 kmem_free(dsname, MAXPATHLEN); 4425 } 4426 kmem_free(tmpname, MAXPATHLEN); 4427 } 4428 4429 /* 4430 * Add the list of hot spares and level 2 cache devices. 4431 */ 4432 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4433 spa_add_spares(spa, config); 4434 spa_add_l2cache(spa, config); 4435 spa_config_exit(spa, SCL_CONFIG, FTAG); 4436 } 4437 4438 spa_unload(spa); 4439 spa_deactivate(spa); 4440 spa_remove(spa); 4441 mutex_exit(&spa_namespace_lock); 4442 4443 return (config); 4444} 4445 4446/* 4447 * Pool export/destroy 4448 * 4449 * The act of destroying or exporting a pool is very simple. We make sure there 4450 * is no more pending I/O and any references to the pool are gone. Then, we 4451 * update the pool state and sync all the labels to disk, removing the 4452 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4453 * we don't sync the labels or remove the configuration cache. 4454 */ 4455static int 4456spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4457 boolean_t force, boolean_t hardforce) 4458{ 4459 spa_t *spa; 4460 4461 if (oldconfig) 4462 *oldconfig = NULL; 4463 4464 if (!(spa_mode_global & FWRITE)) 4465 return (SET_ERROR(EROFS)); 4466 4467 mutex_enter(&spa_namespace_lock); 4468 if ((spa = spa_lookup(pool)) == NULL) { 4469 mutex_exit(&spa_namespace_lock); 4470 return (SET_ERROR(ENOENT)); 4471 } 4472 4473 /* 4474 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4475 * reacquire the namespace lock, and see if we can export. 4476 */ 4477 spa_open_ref(spa, FTAG); 4478 mutex_exit(&spa_namespace_lock); 4479 spa_async_suspend(spa); 4480 mutex_enter(&spa_namespace_lock); 4481 spa_close(spa, FTAG); 4482 4483 /* 4484 * The pool will be in core if it's openable, 4485 * in which case we can modify its state. 4486 */ 4487 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4488 /* 4489 * Objsets may be open only because they're dirty, so we 4490 * have to force it to sync before checking spa_refcnt. 4491 */ 4492 txg_wait_synced(spa->spa_dsl_pool, 0); 4493 4494 /* 4495 * A pool cannot be exported or destroyed if there are active 4496 * references. If we are resetting a pool, allow references by 4497 * fault injection handlers. 4498 */ 4499 if (!spa_refcount_zero(spa) || 4500 (spa->spa_inject_ref != 0 && 4501 new_state != POOL_STATE_UNINITIALIZED)) { 4502 spa_async_resume(spa); 4503 mutex_exit(&spa_namespace_lock); 4504 return (SET_ERROR(EBUSY)); 4505 } 4506 4507 /* 4508 * A pool cannot be exported if it has an active shared spare. 4509 * This is to prevent other pools stealing the active spare 4510 * from an exported pool. At user's own will, such pool can 4511 * be forcedly exported. 4512 */ 4513 if (!force && new_state == POOL_STATE_EXPORTED && 4514 spa_has_active_shared_spare(spa)) { 4515 spa_async_resume(spa); 4516 mutex_exit(&spa_namespace_lock); 4517 return (SET_ERROR(EXDEV)); 4518 } 4519 4520 /* 4521 * We want this to be reflected on every label, 4522 * so mark them all dirty. spa_unload() will do the 4523 * final sync that pushes these changes out. 4524 */ 4525 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4526 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4527 spa->spa_state = new_state; 4528 spa->spa_final_txg = spa_last_synced_txg(spa) + 4529 TXG_DEFER_SIZE + 1; 4530 vdev_config_dirty(spa->spa_root_vdev); 4531 spa_config_exit(spa, SCL_ALL, FTAG); 4532 } 4533 } 4534 4535 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4536 4537 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4538 spa_unload(spa); 4539 spa_deactivate(spa); 4540 } 4541 4542 if (oldconfig && spa->spa_config) 4543 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4544 4545 if (new_state != POOL_STATE_UNINITIALIZED) { 4546 if (!hardforce) 4547 spa_config_sync(spa, B_TRUE, B_TRUE); 4548 spa_remove(spa); 4549 } 4550 mutex_exit(&spa_namespace_lock); 4551 4552 return (0); 4553} 4554 4555/* 4556 * Destroy a storage pool. 4557 */ 4558int 4559spa_destroy(char *pool) 4560{ 4561 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4562 B_FALSE, B_FALSE)); 4563} 4564 4565/* 4566 * Export a storage pool. 4567 */ 4568int 4569spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4570 boolean_t hardforce) 4571{ 4572 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4573 force, hardforce)); 4574} 4575 4576/* 4577 * Similar to spa_export(), this unloads the spa_t without actually removing it 4578 * from the namespace in any way. 4579 */ 4580int 4581spa_reset(char *pool) 4582{ 4583 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4584 B_FALSE, B_FALSE)); 4585} 4586 4587/* 4588 * ========================================================================== 4589 * Device manipulation 4590 * ========================================================================== 4591 */ 4592 4593/* 4594 * Add a device to a storage pool. 4595 */ 4596int 4597spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4598{ 4599 uint64_t txg, id; 4600 int error; 4601 vdev_t *rvd = spa->spa_root_vdev; 4602 vdev_t *vd, *tvd; 4603 nvlist_t **spares, **l2cache; 4604 uint_t nspares, nl2cache; 4605 4606 ASSERT(spa_writeable(spa)); 4607 4608 txg = spa_vdev_enter(spa); 4609 4610 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4611 VDEV_ALLOC_ADD)) != 0) 4612 return (spa_vdev_exit(spa, NULL, txg, error)); 4613 4614 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4615 4616 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4617 &nspares) != 0) 4618 nspares = 0; 4619 4620 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4621 &nl2cache) != 0) 4622 nl2cache = 0; 4623 4624 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4625 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4626 4627 if (vd->vdev_children != 0 && 4628 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4629 return (spa_vdev_exit(spa, vd, txg, error)); 4630 4631 /* 4632 * We must validate the spares and l2cache devices after checking the 4633 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4634 */ 4635 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4636 return (spa_vdev_exit(spa, vd, txg, error)); 4637 4638 /* 4639 * Transfer each new top-level vdev from vd to rvd. 4640 */ 4641 for (int c = 0; c < vd->vdev_children; c++) { 4642 4643 /* 4644 * Set the vdev id to the first hole, if one exists. 4645 */ 4646 for (id = 0; id < rvd->vdev_children; id++) { 4647 if (rvd->vdev_child[id]->vdev_ishole) { 4648 vdev_free(rvd->vdev_child[id]); 4649 break; 4650 } 4651 } 4652 tvd = vd->vdev_child[c]; 4653 vdev_remove_child(vd, tvd); 4654 tvd->vdev_id = id; 4655 vdev_add_child(rvd, tvd); 4656 vdev_config_dirty(tvd); 4657 } 4658 4659 if (nspares != 0) { 4660 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4661 ZPOOL_CONFIG_SPARES); 4662 spa_load_spares(spa); 4663 spa->spa_spares.sav_sync = B_TRUE; 4664 } 4665 4666 if (nl2cache != 0) { 4667 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4668 ZPOOL_CONFIG_L2CACHE); 4669 spa_load_l2cache(spa); 4670 spa->spa_l2cache.sav_sync = B_TRUE; 4671 } 4672 4673 /* 4674 * We have to be careful when adding new vdevs to an existing pool. 4675 * If other threads start allocating from these vdevs before we 4676 * sync the config cache, and we lose power, then upon reboot we may 4677 * fail to open the pool because there are DVAs that the config cache 4678 * can't translate. Therefore, we first add the vdevs without 4679 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4680 * and then let spa_config_update() initialize the new metaslabs. 4681 * 4682 * spa_load() checks for added-but-not-initialized vdevs, so that 4683 * if we lose power at any point in this sequence, the remaining 4684 * steps will be completed the next time we load the pool. 4685 */ 4686 (void) spa_vdev_exit(spa, vd, txg, 0); 4687 4688 mutex_enter(&spa_namespace_lock); 4689 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4690 mutex_exit(&spa_namespace_lock); 4691 4692 return (0); 4693} 4694 4695/* 4696 * Attach a device to a mirror. The arguments are the path to any device 4697 * in the mirror, and the nvroot for the new device. If the path specifies 4698 * a device that is not mirrored, we automatically insert the mirror vdev. 4699 * 4700 * If 'replacing' is specified, the new device is intended to replace the 4701 * existing device; in this case the two devices are made into their own 4702 * mirror using the 'replacing' vdev, which is functionally identical to 4703 * the mirror vdev (it actually reuses all the same ops) but has a few 4704 * extra rules: you can't attach to it after it's been created, and upon 4705 * completion of resilvering, the first disk (the one being replaced) 4706 * is automatically detached. 4707 */ 4708int 4709spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4710{ 4711 uint64_t txg, dtl_max_txg; 4712 vdev_t *rvd = spa->spa_root_vdev; 4713 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4714 vdev_ops_t *pvops; 4715 char *oldvdpath, *newvdpath; 4716 int newvd_isspare; 4717 int error; 4718 4719 ASSERT(spa_writeable(spa)); 4720 4721 txg = spa_vdev_enter(spa); 4722 4723 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4724 4725 if (oldvd == NULL) 4726 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4727 4728 if (!oldvd->vdev_ops->vdev_op_leaf) 4729 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4730 4731 pvd = oldvd->vdev_parent; 4732 4733 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4734 VDEV_ALLOC_ATTACH)) != 0) 4735 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4736 4737 if (newrootvd->vdev_children != 1) 4738 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4739 4740 newvd = newrootvd->vdev_child[0]; 4741 4742 if (!newvd->vdev_ops->vdev_op_leaf) 4743 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4744 4745 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4746 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4747 4748 /* 4749 * Spares can't replace logs 4750 */ 4751 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4752 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4753 4754 if (!replacing) { 4755 /* 4756 * For attach, the only allowable parent is a mirror or the root 4757 * vdev. 4758 */ 4759 if (pvd->vdev_ops != &vdev_mirror_ops && 4760 pvd->vdev_ops != &vdev_root_ops) 4761 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4762 4763 pvops = &vdev_mirror_ops; 4764 } else { 4765 /* 4766 * Active hot spares can only be replaced by inactive hot 4767 * spares. 4768 */ 4769 if (pvd->vdev_ops == &vdev_spare_ops && 4770 oldvd->vdev_isspare && 4771 !spa_has_spare(spa, newvd->vdev_guid)) 4772 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4773 4774 /* 4775 * If the source is a hot spare, and the parent isn't already a 4776 * spare, then we want to create a new hot spare. Otherwise, we 4777 * want to create a replacing vdev. The user is not allowed to 4778 * attach to a spared vdev child unless the 'isspare' state is 4779 * the same (spare replaces spare, non-spare replaces 4780 * non-spare). 4781 */ 4782 if (pvd->vdev_ops == &vdev_replacing_ops && 4783 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4784 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4785 } else if (pvd->vdev_ops == &vdev_spare_ops && 4786 newvd->vdev_isspare != oldvd->vdev_isspare) { 4787 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4788 } 4789 4790 if (newvd->vdev_isspare) 4791 pvops = &vdev_spare_ops; 4792 else 4793 pvops = &vdev_replacing_ops; 4794 } 4795 4796 /* 4797 * Make sure the new device is big enough. 4798 */ 4799 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4800 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4801 4802 /* 4803 * The new device cannot have a higher alignment requirement 4804 * than the top-level vdev. 4805 */ 4806 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4807 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4808 4809 /* 4810 * If this is an in-place replacement, update oldvd's path and devid 4811 * to make it distinguishable from newvd, and unopenable from now on. 4812 */ 4813 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4814 spa_strfree(oldvd->vdev_path); 4815 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4816 KM_SLEEP); 4817 (void) sprintf(oldvd->vdev_path, "%s/%s", 4818 newvd->vdev_path, "old"); 4819 if (oldvd->vdev_devid != NULL) { 4820 spa_strfree(oldvd->vdev_devid); 4821 oldvd->vdev_devid = NULL; 4822 } 4823 } 4824 4825 /* mark the device being resilvered */ 4826 newvd->vdev_resilver_txg = txg; 4827 4828 /* 4829 * If the parent is not a mirror, or if we're replacing, insert the new 4830 * mirror/replacing/spare vdev above oldvd. 4831 */ 4832 if (pvd->vdev_ops != pvops) 4833 pvd = vdev_add_parent(oldvd, pvops); 4834 4835 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4836 ASSERT(pvd->vdev_ops == pvops); 4837 ASSERT(oldvd->vdev_parent == pvd); 4838 4839 /* 4840 * Extract the new device from its root and add it to pvd. 4841 */ 4842 vdev_remove_child(newrootvd, newvd); 4843 newvd->vdev_id = pvd->vdev_children; 4844 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4845 vdev_add_child(pvd, newvd); 4846 4847 tvd = newvd->vdev_top; 4848 ASSERT(pvd->vdev_top == tvd); 4849 ASSERT(tvd->vdev_parent == rvd); 4850 4851 vdev_config_dirty(tvd); 4852 4853 /* 4854 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4855 * for any dmu_sync-ed blocks. It will propagate upward when 4856 * spa_vdev_exit() calls vdev_dtl_reassess(). 4857 */ 4858 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4859 4860 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4861 dtl_max_txg - TXG_INITIAL); 4862 4863 if (newvd->vdev_isspare) { 4864 spa_spare_activate(newvd); 4865 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4866 } 4867 4868 oldvdpath = spa_strdup(oldvd->vdev_path); 4869 newvdpath = spa_strdup(newvd->vdev_path); 4870 newvd_isspare = newvd->vdev_isspare; 4871 4872 /* 4873 * Mark newvd's DTL dirty in this txg. 4874 */ 4875 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4876 4877 /* 4878 * Schedule the resilver to restart in the future. We do this to 4879 * ensure that dmu_sync-ed blocks have been stitched into the 4880 * respective datasets. 4881 */ 4882 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4883 4884 /* 4885 * Commit the config 4886 */ 4887 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4888 4889 spa_history_log_internal(spa, "vdev attach", NULL, 4890 "%s vdev=%s %s vdev=%s", 4891 replacing && newvd_isspare ? "spare in" : 4892 replacing ? "replace" : "attach", newvdpath, 4893 replacing ? "for" : "to", oldvdpath); 4894 4895 spa_strfree(oldvdpath); 4896 spa_strfree(newvdpath); 4897 4898 if (spa->spa_bootfs) 4899 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4900 4901 return (0); 4902} 4903 4904/* 4905 * Detach a device from a mirror or replacing vdev. 4906 * 4907 * If 'replace_done' is specified, only detach if the parent 4908 * is a replacing vdev. 4909 */ 4910int 4911spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4912{ 4913 uint64_t txg; 4914 int error; 4915 vdev_t *rvd = spa->spa_root_vdev; 4916 vdev_t *vd, *pvd, *cvd, *tvd; 4917 boolean_t unspare = B_FALSE; 4918 uint64_t unspare_guid = 0; 4919 char *vdpath; 4920 4921 ASSERT(spa_writeable(spa)); 4922 4923 txg = spa_vdev_enter(spa); 4924 4925 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4926 4927 if (vd == NULL) 4928 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4929 4930 if (!vd->vdev_ops->vdev_op_leaf) 4931 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4932 4933 pvd = vd->vdev_parent; 4934 4935 /* 4936 * If the parent/child relationship is not as expected, don't do it. 4937 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4938 * vdev that's replacing B with C. The user's intent in replacing 4939 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4940 * the replace by detaching C, the expected behavior is to end up 4941 * M(A,B). But suppose that right after deciding to detach C, 4942 * the replacement of B completes. We would have M(A,C), and then 4943 * ask to detach C, which would leave us with just A -- not what 4944 * the user wanted. To prevent this, we make sure that the 4945 * parent/child relationship hasn't changed -- in this example, 4946 * that C's parent is still the replacing vdev R. 4947 */ 4948 if (pvd->vdev_guid != pguid && pguid != 0) 4949 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4950 4951 /* 4952 * Only 'replacing' or 'spare' vdevs can be replaced. 4953 */ 4954 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4955 pvd->vdev_ops != &vdev_spare_ops) 4956 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4957 4958 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4959 spa_version(spa) >= SPA_VERSION_SPARES); 4960 4961 /* 4962 * Only mirror, replacing, and spare vdevs support detach. 4963 */ 4964 if (pvd->vdev_ops != &vdev_replacing_ops && 4965 pvd->vdev_ops != &vdev_mirror_ops && 4966 pvd->vdev_ops != &vdev_spare_ops) 4967 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4968 4969 /* 4970 * If this device has the only valid copy of some data, 4971 * we cannot safely detach it. 4972 */ 4973 if (vdev_dtl_required(vd)) 4974 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4975 4976 ASSERT(pvd->vdev_children >= 2); 4977 4978 /* 4979 * If we are detaching the second disk from a replacing vdev, then 4980 * check to see if we changed the original vdev's path to have "/old" 4981 * at the end in spa_vdev_attach(). If so, undo that change now. 4982 */ 4983 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4984 vd->vdev_path != NULL) { 4985 size_t len = strlen(vd->vdev_path); 4986 4987 for (int c = 0; c < pvd->vdev_children; c++) { 4988 cvd = pvd->vdev_child[c]; 4989 4990 if (cvd == vd || cvd->vdev_path == NULL) 4991 continue; 4992 4993 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4994 strcmp(cvd->vdev_path + len, "/old") == 0) { 4995 spa_strfree(cvd->vdev_path); 4996 cvd->vdev_path = spa_strdup(vd->vdev_path); 4997 break; 4998 } 4999 } 5000 } 5001 5002 /* 5003 * If we are detaching the original disk from a spare, then it implies 5004 * that the spare should become a real disk, and be removed from the 5005 * active spare list for the pool. 5006 */ 5007 if (pvd->vdev_ops == &vdev_spare_ops && 5008 vd->vdev_id == 0 && 5009 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5010 unspare = B_TRUE; 5011 5012 /* 5013 * Erase the disk labels so the disk can be used for other things. 5014 * This must be done after all other error cases are handled, 5015 * but before we disembowel vd (so we can still do I/O to it). 5016 * But if we can't do it, don't treat the error as fatal -- 5017 * it may be that the unwritability of the disk is the reason 5018 * it's being detached! 5019 */ 5020 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5021 5022 /* 5023 * Remove vd from its parent and compact the parent's children. 5024 */ 5025 vdev_remove_child(pvd, vd); 5026 vdev_compact_children(pvd); 5027 5028 /* 5029 * Remember one of the remaining children so we can get tvd below. 5030 */ 5031 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5032 5033 /* 5034 * If we need to remove the remaining child from the list of hot spares, 5035 * do it now, marking the vdev as no longer a spare in the process. 5036 * We must do this before vdev_remove_parent(), because that can 5037 * change the GUID if it creates a new toplevel GUID. For a similar 5038 * reason, we must remove the spare now, in the same txg as the detach; 5039 * otherwise someone could attach a new sibling, change the GUID, and 5040 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5041 */ 5042 if (unspare) { 5043 ASSERT(cvd->vdev_isspare); 5044 spa_spare_remove(cvd); 5045 unspare_guid = cvd->vdev_guid; 5046 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5047 cvd->vdev_unspare = B_TRUE; 5048 } 5049 5050 /* 5051 * If the parent mirror/replacing vdev only has one child, 5052 * the parent is no longer needed. Remove it from the tree. 5053 */ 5054 if (pvd->vdev_children == 1) { 5055 if (pvd->vdev_ops == &vdev_spare_ops) 5056 cvd->vdev_unspare = B_FALSE; 5057 vdev_remove_parent(cvd); 5058 } 5059 5060 5061 /* 5062 * We don't set tvd until now because the parent we just removed 5063 * may have been the previous top-level vdev. 5064 */ 5065 tvd = cvd->vdev_top; 5066 ASSERT(tvd->vdev_parent == rvd); 5067 5068 /* 5069 * Reevaluate the parent vdev state. 5070 */ 5071 vdev_propagate_state(cvd); 5072 5073 /* 5074 * If the 'autoexpand' property is set on the pool then automatically 5075 * try to expand the size of the pool. For example if the device we 5076 * just detached was smaller than the others, it may be possible to 5077 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5078 * first so that we can obtain the updated sizes of the leaf vdevs. 5079 */ 5080 if (spa->spa_autoexpand) { 5081 vdev_reopen(tvd); 5082 vdev_expand(tvd, txg); 5083 } 5084 5085 vdev_config_dirty(tvd); 5086 5087 /* 5088 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5089 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5090 * But first make sure we're not on any *other* txg's DTL list, to 5091 * prevent vd from being accessed after it's freed. 5092 */ 5093 vdpath = spa_strdup(vd->vdev_path); 5094 for (int t = 0; t < TXG_SIZE; t++) 5095 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5096 vd->vdev_detached = B_TRUE; 5097 vdev_dirty(tvd, VDD_DTL, vd, txg); 5098 5099 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5100 5101 /* hang on to the spa before we release the lock */ 5102 spa_open_ref(spa, FTAG); 5103 5104 error = spa_vdev_exit(spa, vd, txg, 0); 5105 5106 spa_history_log_internal(spa, "detach", NULL, 5107 "vdev=%s", vdpath); 5108 spa_strfree(vdpath); 5109 5110 /* 5111 * If this was the removal of the original device in a hot spare vdev, 5112 * then we want to go through and remove the device from the hot spare 5113 * list of every other pool. 5114 */ 5115 if (unspare) { 5116 spa_t *altspa = NULL; 5117 5118 mutex_enter(&spa_namespace_lock); 5119 while ((altspa = spa_next(altspa)) != NULL) { 5120 if (altspa->spa_state != POOL_STATE_ACTIVE || 5121 altspa == spa) 5122 continue; 5123 5124 spa_open_ref(altspa, FTAG); 5125 mutex_exit(&spa_namespace_lock); 5126 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5127 mutex_enter(&spa_namespace_lock); 5128 spa_close(altspa, FTAG); 5129 } 5130 mutex_exit(&spa_namespace_lock); 5131 5132 /* search the rest of the vdevs for spares to remove */ 5133 spa_vdev_resilver_done(spa); 5134 } 5135 5136 /* all done with the spa; OK to release */ 5137 mutex_enter(&spa_namespace_lock); 5138 spa_close(spa, FTAG); 5139 mutex_exit(&spa_namespace_lock); 5140 5141 return (error); 5142} 5143 5144/* 5145 * Split a set of devices from their mirrors, and create a new pool from them. 5146 */ 5147int 5148spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5149 nvlist_t *props, boolean_t exp) 5150{ 5151 int error = 0; 5152 uint64_t txg, *glist; 5153 spa_t *newspa; 5154 uint_t c, children, lastlog; 5155 nvlist_t **child, *nvl, *tmp; 5156 dmu_tx_t *tx; 5157 char *altroot = NULL; 5158 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5159 boolean_t activate_slog; 5160 5161 ASSERT(spa_writeable(spa)); 5162 5163 txg = spa_vdev_enter(spa); 5164 5165 /* clear the log and flush everything up to now */ 5166 activate_slog = spa_passivate_log(spa); 5167 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5168 error = spa_offline_log(spa); 5169 txg = spa_vdev_config_enter(spa); 5170 5171 if (activate_slog) 5172 spa_activate_log(spa); 5173 5174 if (error != 0) 5175 return (spa_vdev_exit(spa, NULL, txg, error)); 5176 5177 /* check new spa name before going any further */ 5178 if (spa_lookup(newname) != NULL) 5179 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5180 5181 /* 5182 * scan through all the children to ensure they're all mirrors 5183 */ 5184 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5185 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5186 &children) != 0) 5187 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5188 5189 /* first, check to ensure we've got the right child count */ 5190 rvd = spa->spa_root_vdev; 5191 lastlog = 0; 5192 for (c = 0; c < rvd->vdev_children; c++) { 5193 vdev_t *vd = rvd->vdev_child[c]; 5194 5195 /* don't count the holes & logs as children */ 5196 if (vd->vdev_islog || vd->vdev_ishole) { 5197 if (lastlog == 0) 5198 lastlog = c; 5199 continue; 5200 } 5201 5202 lastlog = 0; 5203 } 5204 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5205 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5206 5207 /* next, ensure no spare or cache devices are part of the split */ 5208 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5209 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5210 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5211 5212 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5213 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5214 5215 /* then, loop over each vdev and validate it */ 5216 for (c = 0; c < children; c++) { 5217 uint64_t is_hole = 0; 5218 5219 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5220 &is_hole); 5221 5222 if (is_hole != 0) { 5223 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5224 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5225 continue; 5226 } else { 5227 error = SET_ERROR(EINVAL); 5228 break; 5229 } 5230 } 5231 5232 /* which disk is going to be split? */ 5233 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5234 &glist[c]) != 0) { 5235 error = SET_ERROR(EINVAL); 5236 break; 5237 } 5238 5239 /* look it up in the spa */ 5240 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5241 if (vml[c] == NULL) { 5242 error = SET_ERROR(ENODEV); 5243 break; 5244 } 5245 5246 /* make sure there's nothing stopping the split */ 5247 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5248 vml[c]->vdev_islog || 5249 vml[c]->vdev_ishole || 5250 vml[c]->vdev_isspare || 5251 vml[c]->vdev_isl2cache || 5252 !vdev_writeable(vml[c]) || 5253 vml[c]->vdev_children != 0 || 5254 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5255 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5256 error = SET_ERROR(EINVAL); 5257 break; 5258 } 5259 5260 if (vdev_dtl_required(vml[c])) { 5261 error = SET_ERROR(EBUSY); 5262 break; 5263 } 5264 5265 /* we need certain info from the top level */ 5266 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5267 vml[c]->vdev_top->vdev_ms_array) == 0); 5268 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5269 vml[c]->vdev_top->vdev_ms_shift) == 0); 5270 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5271 vml[c]->vdev_top->vdev_asize) == 0); 5272 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5273 vml[c]->vdev_top->vdev_ashift) == 0); 5274 } 5275 5276 if (error != 0) { 5277 kmem_free(vml, children * sizeof (vdev_t *)); 5278 kmem_free(glist, children * sizeof (uint64_t)); 5279 return (spa_vdev_exit(spa, NULL, txg, error)); 5280 } 5281 5282 /* stop writers from using the disks */ 5283 for (c = 0; c < children; c++) { 5284 if (vml[c] != NULL) 5285 vml[c]->vdev_offline = B_TRUE; 5286 } 5287 vdev_reopen(spa->spa_root_vdev); 5288 5289 /* 5290 * Temporarily record the splitting vdevs in the spa config. This 5291 * will disappear once the config is regenerated. 5292 */ 5293 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5294 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5295 glist, children) == 0); 5296 kmem_free(glist, children * sizeof (uint64_t)); 5297 5298 mutex_enter(&spa->spa_props_lock); 5299 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5300 nvl) == 0); 5301 mutex_exit(&spa->spa_props_lock); 5302 spa->spa_config_splitting = nvl; 5303 vdev_config_dirty(spa->spa_root_vdev); 5304 5305 /* configure and create the new pool */ 5306 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5307 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5308 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5309 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5310 spa_version(spa)) == 0); 5311 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5312 spa->spa_config_txg) == 0); 5313 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5314 spa_generate_guid(NULL)) == 0); 5315 (void) nvlist_lookup_string(props, 5316 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5317 5318 /* add the new pool to the namespace */ 5319 newspa = spa_add(newname, config, altroot); 5320 newspa->spa_config_txg = spa->spa_config_txg; 5321 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5322 5323 /* release the spa config lock, retaining the namespace lock */ 5324 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5325 5326 if (zio_injection_enabled) 5327 zio_handle_panic_injection(spa, FTAG, 1); 5328 5329 spa_activate(newspa, spa_mode_global); 5330 spa_async_suspend(newspa); 5331 5332#ifndef sun 5333 /* mark that we are creating new spa by splitting */ 5334 newspa->spa_splitting_newspa = B_TRUE; 5335#endif 5336 /* create the new pool from the disks of the original pool */ 5337 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5338#ifndef sun 5339 newspa->spa_splitting_newspa = B_FALSE; 5340#endif 5341 if (error) 5342 goto out; 5343 5344 /* if that worked, generate a real config for the new pool */ 5345 if (newspa->spa_root_vdev != NULL) { 5346 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5347 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5348 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5349 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5350 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5351 B_TRUE)); 5352 } 5353 5354 /* set the props */ 5355 if (props != NULL) { 5356 spa_configfile_set(newspa, props, B_FALSE); 5357 error = spa_prop_set(newspa, props); 5358 if (error) 5359 goto out; 5360 } 5361 5362 /* flush everything */ 5363 txg = spa_vdev_config_enter(newspa); 5364 vdev_config_dirty(newspa->spa_root_vdev); 5365 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5366 5367 if (zio_injection_enabled) 5368 zio_handle_panic_injection(spa, FTAG, 2); 5369 5370 spa_async_resume(newspa); 5371 5372 /* finally, update the original pool's config */ 5373 txg = spa_vdev_config_enter(spa); 5374 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5375 error = dmu_tx_assign(tx, TXG_WAIT); 5376 if (error != 0) 5377 dmu_tx_abort(tx); 5378 for (c = 0; c < children; c++) { 5379 if (vml[c] != NULL) { 5380 vdev_split(vml[c]); 5381 if (error == 0) 5382 spa_history_log_internal(spa, "detach", tx, 5383 "vdev=%s", vml[c]->vdev_path); 5384 vdev_free(vml[c]); 5385 } 5386 } 5387 vdev_config_dirty(spa->spa_root_vdev); 5388 spa->spa_config_splitting = NULL; 5389 nvlist_free(nvl); 5390 if (error == 0) 5391 dmu_tx_commit(tx); 5392 (void) spa_vdev_exit(spa, NULL, txg, 0); 5393 5394 if (zio_injection_enabled) 5395 zio_handle_panic_injection(spa, FTAG, 3); 5396 5397 /* split is complete; log a history record */ 5398 spa_history_log_internal(newspa, "split", NULL, 5399 "from pool %s", spa_name(spa)); 5400 5401 kmem_free(vml, children * sizeof (vdev_t *)); 5402 5403 /* if we're not going to mount the filesystems in userland, export */ 5404 if (exp) 5405 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5406 B_FALSE, B_FALSE); 5407 5408 return (error); 5409 5410out: 5411 spa_unload(newspa); 5412 spa_deactivate(newspa); 5413 spa_remove(newspa); 5414 5415 txg = spa_vdev_config_enter(spa); 5416 5417 /* re-online all offlined disks */ 5418 for (c = 0; c < children; c++) { 5419 if (vml[c] != NULL) 5420 vml[c]->vdev_offline = B_FALSE; 5421 } 5422 vdev_reopen(spa->spa_root_vdev); 5423 5424 nvlist_free(spa->spa_config_splitting); 5425 spa->spa_config_splitting = NULL; 5426 (void) spa_vdev_exit(spa, NULL, txg, error); 5427 5428 kmem_free(vml, children * sizeof (vdev_t *)); 5429 return (error); 5430} 5431 5432static nvlist_t * 5433spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5434{ 5435 for (int i = 0; i < count; i++) { 5436 uint64_t guid; 5437 5438 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5439 &guid) == 0); 5440 5441 if (guid == target_guid) 5442 return (nvpp[i]); 5443 } 5444 5445 return (NULL); 5446} 5447 5448static void 5449spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5450 nvlist_t *dev_to_remove) 5451{ 5452 nvlist_t **newdev = NULL; 5453 5454 if (count > 1) 5455 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5456 5457 for (int i = 0, j = 0; i < count; i++) { 5458 if (dev[i] == dev_to_remove) 5459 continue; 5460 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5461 } 5462 5463 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5464 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5465 5466 for (int i = 0; i < count - 1; i++) 5467 nvlist_free(newdev[i]); 5468 5469 if (count > 1) 5470 kmem_free(newdev, (count - 1) * sizeof (void *)); 5471} 5472 5473/* 5474 * Evacuate the device. 5475 */ 5476static int 5477spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5478{ 5479 uint64_t txg; 5480 int error = 0; 5481 5482 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5483 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5484 ASSERT(vd == vd->vdev_top); 5485 5486 /* 5487 * Evacuate the device. We don't hold the config lock as writer 5488 * since we need to do I/O but we do keep the 5489 * spa_namespace_lock held. Once this completes the device 5490 * should no longer have any blocks allocated on it. 5491 */ 5492 if (vd->vdev_islog) { 5493 if (vd->vdev_stat.vs_alloc != 0) 5494 error = spa_offline_log(spa); 5495 } else { 5496 error = SET_ERROR(ENOTSUP); 5497 } 5498 5499 if (error) 5500 return (error); 5501 5502 /* 5503 * The evacuation succeeded. Remove any remaining MOS metadata 5504 * associated with this vdev, and wait for these changes to sync. 5505 */ 5506 ASSERT0(vd->vdev_stat.vs_alloc); 5507 txg = spa_vdev_config_enter(spa); 5508 vd->vdev_removing = B_TRUE; 5509 vdev_dirty_leaves(vd, VDD_DTL, txg); 5510 vdev_config_dirty(vd); 5511 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5512 5513 return (0); 5514} 5515 5516/* 5517 * Complete the removal by cleaning up the namespace. 5518 */ 5519static void 5520spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5521{ 5522 vdev_t *rvd = spa->spa_root_vdev; 5523 uint64_t id = vd->vdev_id; 5524 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5525 5526 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5527 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5528 ASSERT(vd == vd->vdev_top); 5529 5530 /* 5531 * Only remove any devices which are empty. 5532 */ 5533 if (vd->vdev_stat.vs_alloc != 0) 5534 return; 5535 5536 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5537 5538 if (list_link_active(&vd->vdev_state_dirty_node)) 5539 vdev_state_clean(vd); 5540 if (list_link_active(&vd->vdev_config_dirty_node)) 5541 vdev_config_clean(vd); 5542 5543 vdev_free(vd); 5544 5545 if (last_vdev) { 5546 vdev_compact_children(rvd); 5547 } else { 5548 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5549 vdev_add_child(rvd, vd); 5550 } 5551 vdev_config_dirty(rvd); 5552 5553 /* 5554 * Reassess the health of our root vdev. 5555 */ 5556 vdev_reopen(rvd); 5557} 5558 5559/* 5560 * Remove a device from the pool - 5561 * 5562 * Removing a device from the vdev namespace requires several steps 5563 * and can take a significant amount of time. As a result we use 5564 * the spa_vdev_config_[enter/exit] functions which allow us to 5565 * grab and release the spa_config_lock while still holding the namespace 5566 * lock. During each step the configuration is synced out. 5567 * 5568 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5569 * devices. 5570 */ 5571int 5572spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5573{ 5574 vdev_t *vd; 5575 metaslab_group_t *mg; 5576 nvlist_t **spares, **l2cache, *nv; 5577 uint64_t txg = 0; 5578 uint_t nspares, nl2cache; 5579 int error = 0; 5580 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5581 5582 ASSERT(spa_writeable(spa)); 5583 5584 if (!locked) 5585 txg = spa_vdev_enter(spa); 5586 5587 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5588 5589 if (spa->spa_spares.sav_vdevs != NULL && 5590 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5591 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5592 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5593 /* 5594 * Only remove the hot spare if it's not currently in use 5595 * in this pool. 5596 */ 5597 if (vd == NULL || unspare) { 5598 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5599 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5600 spa_load_spares(spa); 5601 spa->spa_spares.sav_sync = B_TRUE; 5602 } else { 5603 error = SET_ERROR(EBUSY); 5604 } 5605 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5606 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5607 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5608 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5609 /* 5610 * Cache devices can always be removed. 5611 */ 5612 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5613 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5614 spa_load_l2cache(spa); 5615 spa->spa_l2cache.sav_sync = B_TRUE; 5616 } else if (vd != NULL && vd->vdev_islog) { 5617 ASSERT(!locked); 5618 ASSERT(vd == vd->vdev_top); 5619 5620 mg = vd->vdev_mg; 5621 5622 /* 5623 * Stop allocating from this vdev. 5624 */ 5625 metaslab_group_passivate(mg); 5626 5627 /* 5628 * Wait for the youngest allocations and frees to sync, 5629 * and then wait for the deferral of those frees to finish. 5630 */ 5631 spa_vdev_config_exit(spa, NULL, 5632 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5633 5634 /* 5635 * Attempt to evacuate the vdev. 5636 */ 5637 error = spa_vdev_remove_evacuate(spa, vd); 5638 5639 txg = spa_vdev_config_enter(spa); 5640 5641 /* 5642 * If we couldn't evacuate the vdev, unwind. 5643 */ 5644 if (error) { 5645 metaslab_group_activate(mg); 5646 return (spa_vdev_exit(spa, NULL, txg, error)); 5647 } 5648 5649 /* 5650 * Clean up the vdev namespace. 5651 */ 5652 spa_vdev_remove_from_namespace(spa, vd); 5653 5654 } else if (vd != NULL) { 5655 /* 5656 * Normal vdevs cannot be removed (yet). 5657 */ 5658 error = SET_ERROR(ENOTSUP); 5659 } else { 5660 /* 5661 * There is no vdev of any kind with the specified guid. 5662 */ 5663 error = SET_ERROR(ENOENT); 5664 } 5665 5666 if (!locked) 5667 return (spa_vdev_exit(spa, NULL, txg, error)); 5668 5669 return (error); 5670} 5671 5672/* 5673 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5674 * currently spared, so we can detach it. 5675 */ 5676static vdev_t * 5677spa_vdev_resilver_done_hunt(vdev_t *vd) 5678{ 5679 vdev_t *newvd, *oldvd; 5680 5681 for (int c = 0; c < vd->vdev_children; c++) { 5682 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5683 if (oldvd != NULL) 5684 return (oldvd); 5685 } 5686 5687 /* 5688 * Check for a completed replacement. We always consider the first 5689 * vdev in the list to be the oldest vdev, and the last one to be 5690 * the newest (see spa_vdev_attach() for how that works). In 5691 * the case where the newest vdev is faulted, we will not automatically 5692 * remove it after a resilver completes. This is OK as it will require 5693 * user intervention to determine which disk the admin wishes to keep. 5694 */ 5695 if (vd->vdev_ops == &vdev_replacing_ops) { 5696 ASSERT(vd->vdev_children > 1); 5697 5698 newvd = vd->vdev_child[vd->vdev_children - 1]; 5699 oldvd = vd->vdev_child[0]; 5700 5701 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5702 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5703 !vdev_dtl_required(oldvd)) 5704 return (oldvd); 5705 } 5706 5707 /* 5708 * Check for a completed resilver with the 'unspare' flag set. 5709 */ 5710 if (vd->vdev_ops == &vdev_spare_ops) { 5711 vdev_t *first = vd->vdev_child[0]; 5712 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5713 5714 if (last->vdev_unspare) { 5715 oldvd = first; 5716 newvd = last; 5717 } else if (first->vdev_unspare) { 5718 oldvd = last; 5719 newvd = first; 5720 } else { 5721 oldvd = NULL; 5722 } 5723 5724 if (oldvd != NULL && 5725 vdev_dtl_empty(newvd, DTL_MISSING) && 5726 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5727 !vdev_dtl_required(oldvd)) 5728 return (oldvd); 5729 5730 /* 5731 * If there are more than two spares attached to a disk, 5732 * and those spares are not required, then we want to 5733 * attempt to free them up now so that they can be used 5734 * by other pools. Once we're back down to a single 5735 * disk+spare, we stop removing them. 5736 */ 5737 if (vd->vdev_children > 2) { 5738 newvd = vd->vdev_child[1]; 5739 5740 if (newvd->vdev_isspare && last->vdev_isspare && 5741 vdev_dtl_empty(last, DTL_MISSING) && 5742 vdev_dtl_empty(last, DTL_OUTAGE) && 5743 !vdev_dtl_required(newvd)) 5744 return (newvd); 5745 } 5746 } 5747 5748 return (NULL); 5749} 5750 5751static void 5752spa_vdev_resilver_done(spa_t *spa) 5753{ 5754 vdev_t *vd, *pvd, *ppvd; 5755 uint64_t guid, sguid, pguid, ppguid; 5756 5757 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5758 5759 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5760 pvd = vd->vdev_parent; 5761 ppvd = pvd->vdev_parent; 5762 guid = vd->vdev_guid; 5763 pguid = pvd->vdev_guid; 5764 ppguid = ppvd->vdev_guid; 5765 sguid = 0; 5766 /* 5767 * If we have just finished replacing a hot spared device, then 5768 * we need to detach the parent's first child (the original hot 5769 * spare) as well. 5770 */ 5771 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5772 ppvd->vdev_children == 2) { 5773 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5774 sguid = ppvd->vdev_child[1]->vdev_guid; 5775 } 5776 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5777 5778 spa_config_exit(spa, SCL_ALL, FTAG); 5779 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5780 return; 5781 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5782 return; 5783 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5784 } 5785 5786 spa_config_exit(spa, SCL_ALL, FTAG); 5787} 5788 5789/* 5790 * Update the stored path or FRU for this vdev. 5791 */ 5792int 5793spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5794 boolean_t ispath) 5795{ 5796 vdev_t *vd; 5797 boolean_t sync = B_FALSE; 5798 5799 ASSERT(spa_writeable(spa)); 5800 5801 spa_vdev_state_enter(spa, SCL_ALL); 5802 5803 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5804 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5805 5806 if (!vd->vdev_ops->vdev_op_leaf) 5807 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5808 5809 if (ispath) { 5810 if (strcmp(value, vd->vdev_path) != 0) { 5811 spa_strfree(vd->vdev_path); 5812 vd->vdev_path = spa_strdup(value); 5813 sync = B_TRUE; 5814 } 5815 } else { 5816 if (vd->vdev_fru == NULL) { 5817 vd->vdev_fru = spa_strdup(value); 5818 sync = B_TRUE; 5819 } else if (strcmp(value, vd->vdev_fru) != 0) { 5820 spa_strfree(vd->vdev_fru); 5821 vd->vdev_fru = spa_strdup(value); 5822 sync = B_TRUE; 5823 } 5824 } 5825 5826 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5827} 5828 5829int 5830spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5831{ 5832 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5833} 5834 5835int 5836spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5837{ 5838 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5839} 5840 5841/* 5842 * ========================================================================== 5843 * SPA Scanning 5844 * ========================================================================== 5845 */ 5846 5847int 5848spa_scan_stop(spa_t *spa) 5849{ 5850 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5851 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5852 return (SET_ERROR(EBUSY)); 5853 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5854} 5855 5856int 5857spa_scan(spa_t *spa, pool_scan_func_t func) 5858{ 5859 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5860 5861 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5862 return (SET_ERROR(ENOTSUP)); 5863 5864 /* 5865 * If a resilver was requested, but there is no DTL on a 5866 * writeable leaf device, we have nothing to do. 5867 */ 5868 if (func == POOL_SCAN_RESILVER && 5869 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5870 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5871 return (0); 5872 } 5873 5874 return (dsl_scan(spa->spa_dsl_pool, func)); 5875} 5876 5877/* 5878 * ========================================================================== 5879 * SPA async task processing 5880 * ========================================================================== 5881 */ 5882 5883static void 5884spa_async_remove(spa_t *spa, vdev_t *vd) 5885{ 5886 if (vd->vdev_remove_wanted) { 5887 vd->vdev_remove_wanted = B_FALSE; 5888 vd->vdev_delayed_close = B_FALSE; 5889 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5890 5891 /* 5892 * We want to clear the stats, but we don't want to do a full 5893 * vdev_clear() as that will cause us to throw away 5894 * degraded/faulted state as well as attempt to reopen the 5895 * device, all of which is a waste. 5896 */ 5897 vd->vdev_stat.vs_read_errors = 0; 5898 vd->vdev_stat.vs_write_errors = 0; 5899 vd->vdev_stat.vs_checksum_errors = 0; 5900 5901 vdev_state_dirty(vd->vdev_top); 5902 } 5903 5904 for (int c = 0; c < vd->vdev_children; c++) 5905 spa_async_remove(spa, vd->vdev_child[c]); 5906} 5907 5908static void 5909spa_async_probe(spa_t *spa, vdev_t *vd) 5910{ 5911 if (vd->vdev_probe_wanted) { 5912 vd->vdev_probe_wanted = B_FALSE; 5913 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5914 } 5915 5916 for (int c = 0; c < vd->vdev_children; c++) 5917 spa_async_probe(spa, vd->vdev_child[c]); 5918} 5919 5920static void 5921spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5922{ 5923 sysevent_id_t eid; 5924 nvlist_t *attr; 5925 char *physpath; 5926 5927 if (!spa->spa_autoexpand) 5928 return; 5929 5930 for (int c = 0; c < vd->vdev_children; c++) { 5931 vdev_t *cvd = vd->vdev_child[c]; 5932 spa_async_autoexpand(spa, cvd); 5933 } 5934 5935 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5936 return; 5937 5938 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5939 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5940 5941 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5942 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5943 5944 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5945 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5946 5947 nvlist_free(attr); 5948 kmem_free(physpath, MAXPATHLEN); 5949} 5950 5951static void 5952spa_async_thread(void *arg) 5953{ 5954 spa_t *spa = arg; 5955 int tasks; 5956 5957 ASSERT(spa->spa_sync_on); 5958 5959 mutex_enter(&spa->spa_async_lock); 5960 tasks = spa->spa_async_tasks; 5961 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5962 mutex_exit(&spa->spa_async_lock); 5963 5964 /* 5965 * See if the config needs to be updated. 5966 */ 5967 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5968 uint64_t old_space, new_space; 5969 5970 mutex_enter(&spa_namespace_lock); 5971 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5972 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5973 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5974 mutex_exit(&spa_namespace_lock); 5975 5976 /* 5977 * If the pool grew as a result of the config update, 5978 * then log an internal history event. 5979 */ 5980 if (new_space != old_space) { 5981 spa_history_log_internal(spa, "vdev online", NULL, 5982 "pool '%s' size: %llu(+%llu)", 5983 spa_name(spa), new_space, new_space - old_space); 5984 } 5985 } 5986 5987 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5988 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5989 spa_async_autoexpand(spa, spa->spa_root_vdev); 5990 spa_config_exit(spa, SCL_CONFIG, FTAG); 5991 } 5992 5993 /* 5994 * See if any devices need to be probed. 5995 */ 5996 if (tasks & SPA_ASYNC_PROBE) { 5997 spa_vdev_state_enter(spa, SCL_NONE); 5998 spa_async_probe(spa, spa->spa_root_vdev); 5999 (void) spa_vdev_state_exit(spa, NULL, 0); 6000 } 6001 6002 /* 6003 * If any devices are done replacing, detach them. 6004 */ 6005 if (tasks & SPA_ASYNC_RESILVER_DONE) 6006 spa_vdev_resilver_done(spa); 6007 6008 /* 6009 * Kick off a resilver. 6010 */ 6011 if (tasks & SPA_ASYNC_RESILVER) 6012 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6013 6014 /* 6015 * Let the world know that we're done. 6016 */ 6017 mutex_enter(&spa->spa_async_lock); 6018 spa->spa_async_thread = NULL; 6019 cv_broadcast(&spa->spa_async_cv); 6020 mutex_exit(&spa->spa_async_lock); 6021 thread_exit(); 6022} 6023 6024static void 6025spa_async_thread_vd(void *arg) 6026{ 6027 spa_t *spa = arg; 6028 int tasks; 6029 6030 ASSERT(spa->spa_sync_on); 6031 6032 mutex_enter(&spa->spa_async_lock); 6033 tasks = spa->spa_async_tasks; 6034retry: 6035 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6036 mutex_exit(&spa->spa_async_lock); 6037 6038 /* 6039 * See if any devices need to be marked REMOVED. 6040 */ 6041 if (tasks & SPA_ASYNC_REMOVE) { 6042 spa_vdev_state_enter(spa, SCL_NONE); 6043 spa_async_remove(spa, spa->spa_root_vdev); 6044 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6045 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6046 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6047 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6048 (void) spa_vdev_state_exit(spa, NULL, 0); 6049 } 6050 6051 /* 6052 * Let the world know that we're done. 6053 */ 6054 mutex_enter(&spa->spa_async_lock); 6055 tasks = spa->spa_async_tasks; 6056 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6057 goto retry; 6058 spa->spa_async_thread_vd = NULL; 6059 cv_broadcast(&spa->spa_async_cv); 6060 mutex_exit(&spa->spa_async_lock); 6061 thread_exit(); 6062} 6063 6064void 6065spa_async_suspend(spa_t *spa) 6066{ 6067 mutex_enter(&spa->spa_async_lock); 6068 spa->spa_async_suspended++; 6069 while (spa->spa_async_thread != NULL && 6070 spa->spa_async_thread_vd != NULL) 6071 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6072 mutex_exit(&spa->spa_async_lock); 6073} 6074 6075void 6076spa_async_resume(spa_t *spa) 6077{ 6078 mutex_enter(&spa->spa_async_lock); 6079 ASSERT(spa->spa_async_suspended != 0); 6080 spa->spa_async_suspended--; 6081 mutex_exit(&spa->spa_async_lock); 6082} 6083 6084static boolean_t 6085spa_async_tasks_pending(spa_t *spa) 6086{ 6087 uint_t non_config_tasks; 6088 uint_t config_task; 6089 boolean_t config_task_suspended; 6090 6091 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6092 SPA_ASYNC_REMOVE); 6093 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6094 if (spa->spa_ccw_fail_time == 0) { 6095 config_task_suspended = B_FALSE; 6096 } else { 6097 config_task_suspended = 6098 (gethrtime() - spa->spa_ccw_fail_time) < 6099 (zfs_ccw_retry_interval * NANOSEC); 6100 } 6101 6102 return (non_config_tasks || (config_task && !config_task_suspended)); 6103} 6104 6105static void 6106spa_async_dispatch(spa_t *spa) 6107{ 6108 mutex_enter(&spa->spa_async_lock); 6109 if (spa_async_tasks_pending(spa) && 6110 !spa->spa_async_suspended && 6111 spa->spa_async_thread == NULL && 6112 rootdir != NULL) 6113 spa->spa_async_thread = thread_create(NULL, 0, 6114 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6115 mutex_exit(&spa->spa_async_lock); 6116} 6117 6118static void 6119spa_async_dispatch_vd(spa_t *spa) 6120{ 6121 mutex_enter(&spa->spa_async_lock); 6122 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6123 !spa->spa_async_suspended && 6124 spa->spa_async_thread_vd == NULL && 6125 rootdir != NULL) 6126 spa->spa_async_thread_vd = thread_create(NULL, 0, 6127 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6128 mutex_exit(&spa->spa_async_lock); 6129} 6130 6131void 6132spa_async_request(spa_t *spa, int task) 6133{ 6134 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6135 mutex_enter(&spa->spa_async_lock); 6136 spa->spa_async_tasks |= task; 6137 mutex_exit(&spa->spa_async_lock); 6138 spa_async_dispatch_vd(spa); 6139} 6140 6141/* 6142 * ========================================================================== 6143 * SPA syncing routines 6144 * ========================================================================== 6145 */ 6146 6147static int 6148bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6149{ 6150 bpobj_t *bpo = arg; 6151 bpobj_enqueue(bpo, bp, tx); 6152 return (0); 6153} 6154 6155static int 6156spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6157{ 6158 zio_t *zio = arg; 6159 6160 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6161 BP_GET_PSIZE(bp), zio->io_flags)); 6162 return (0); 6163} 6164 6165/* 6166 * Note: this simple function is not inlined to make it easier to dtrace the 6167 * amount of time spent syncing frees. 6168 */ 6169static void 6170spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6171{ 6172 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6173 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6174 VERIFY(zio_wait(zio) == 0); 6175} 6176 6177/* 6178 * Note: this simple function is not inlined to make it easier to dtrace the 6179 * amount of time spent syncing deferred frees. 6180 */ 6181static void 6182spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6183{ 6184 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6185 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6186 spa_free_sync_cb, zio, tx), ==, 0); 6187 VERIFY0(zio_wait(zio)); 6188} 6189 6190 6191static void 6192spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6193{ 6194 char *packed = NULL; 6195 size_t bufsize; 6196 size_t nvsize = 0; 6197 dmu_buf_t *db; 6198 6199 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6200 6201 /* 6202 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6203 * information. This avoids the dmu_buf_will_dirty() path and 6204 * saves us a pre-read to get data we don't actually care about. 6205 */ 6206 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6207 packed = kmem_alloc(bufsize, KM_SLEEP); 6208 6209 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6210 KM_SLEEP) == 0); 6211 bzero(packed + nvsize, bufsize - nvsize); 6212 6213 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6214 6215 kmem_free(packed, bufsize); 6216 6217 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6218 dmu_buf_will_dirty(db, tx); 6219 *(uint64_t *)db->db_data = nvsize; 6220 dmu_buf_rele(db, FTAG); 6221} 6222 6223static void 6224spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6225 const char *config, const char *entry) 6226{ 6227 nvlist_t *nvroot; 6228 nvlist_t **list; 6229 int i; 6230 6231 if (!sav->sav_sync) 6232 return; 6233 6234 /* 6235 * Update the MOS nvlist describing the list of available devices. 6236 * spa_validate_aux() will have already made sure this nvlist is 6237 * valid and the vdevs are labeled appropriately. 6238 */ 6239 if (sav->sav_object == 0) { 6240 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6241 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6242 sizeof (uint64_t), tx); 6243 VERIFY(zap_update(spa->spa_meta_objset, 6244 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6245 &sav->sav_object, tx) == 0); 6246 } 6247 6248 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6249 if (sav->sav_count == 0) { 6250 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6251 } else { 6252 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6253 for (i = 0; i < sav->sav_count; i++) 6254 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6255 B_FALSE, VDEV_CONFIG_L2CACHE); 6256 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6257 sav->sav_count) == 0); 6258 for (i = 0; i < sav->sav_count; i++) 6259 nvlist_free(list[i]); 6260 kmem_free(list, sav->sav_count * sizeof (void *)); 6261 } 6262 6263 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6264 nvlist_free(nvroot); 6265 6266 sav->sav_sync = B_FALSE; 6267} 6268 6269static void 6270spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6271{ 6272 nvlist_t *config; 6273 6274 if (list_is_empty(&spa->spa_config_dirty_list)) 6275 return; 6276 6277 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6278 6279 config = spa_config_generate(spa, spa->spa_root_vdev, 6280 dmu_tx_get_txg(tx), B_FALSE); 6281 6282 /* 6283 * If we're upgrading the spa version then make sure that 6284 * the config object gets updated with the correct version. 6285 */ 6286 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6287 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6288 spa->spa_uberblock.ub_version); 6289 6290 spa_config_exit(spa, SCL_STATE, FTAG); 6291 6292 if (spa->spa_config_syncing) 6293 nvlist_free(spa->spa_config_syncing); 6294 spa->spa_config_syncing = config; 6295 6296 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6297} 6298 6299static void 6300spa_sync_version(void *arg, dmu_tx_t *tx) 6301{ 6302 uint64_t *versionp = arg; 6303 uint64_t version = *versionp; 6304 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6305 6306 /* 6307 * Setting the version is special cased when first creating the pool. 6308 */ 6309 ASSERT(tx->tx_txg != TXG_INITIAL); 6310 6311 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6312 ASSERT(version >= spa_version(spa)); 6313 6314 spa->spa_uberblock.ub_version = version; 6315 vdev_config_dirty(spa->spa_root_vdev); 6316 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6317} 6318 6319/* 6320 * Set zpool properties. 6321 */ 6322static void 6323spa_sync_props(void *arg, dmu_tx_t *tx) 6324{ 6325 nvlist_t *nvp = arg; 6326 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6327 objset_t *mos = spa->spa_meta_objset; 6328 nvpair_t *elem = NULL; 6329 6330 mutex_enter(&spa->spa_props_lock); 6331 6332 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6333 uint64_t intval; 6334 char *strval, *fname; 6335 zpool_prop_t prop; 6336 const char *propname; 6337 zprop_type_t proptype; 6338 spa_feature_t fid; 6339 6340 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6341 case ZPROP_INVAL: 6342 /* 6343 * We checked this earlier in spa_prop_validate(). 6344 */ 6345 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6346 6347 fname = strchr(nvpair_name(elem), '@') + 1; 6348 VERIFY0(zfeature_lookup_name(fname, &fid)); 6349 6350 spa_feature_enable(spa, fid, tx); 6351 spa_history_log_internal(spa, "set", tx, 6352 "%s=enabled", nvpair_name(elem)); 6353 break; 6354 6355 case ZPOOL_PROP_VERSION: 6356 intval = fnvpair_value_uint64(elem); 6357 /* 6358 * The version is synced seperatly before other 6359 * properties and should be correct by now. 6360 */ 6361 ASSERT3U(spa_version(spa), >=, intval); 6362 break; 6363 6364 case ZPOOL_PROP_ALTROOT: 6365 /* 6366 * 'altroot' is a non-persistent property. It should 6367 * have been set temporarily at creation or import time. 6368 */ 6369 ASSERT(spa->spa_root != NULL); 6370 break; 6371 6372 case ZPOOL_PROP_READONLY: 6373 case ZPOOL_PROP_CACHEFILE: 6374 /* 6375 * 'readonly' and 'cachefile' are also non-persisitent 6376 * properties. 6377 */ 6378 break; 6379 case ZPOOL_PROP_COMMENT: 6380 strval = fnvpair_value_string(elem); 6381 if (spa->spa_comment != NULL) 6382 spa_strfree(spa->spa_comment); 6383 spa->spa_comment = spa_strdup(strval); 6384 /* 6385 * We need to dirty the configuration on all the vdevs 6386 * so that their labels get updated. It's unnecessary 6387 * to do this for pool creation since the vdev's 6388 * configuratoin has already been dirtied. 6389 */ 6390 if (tx->tx_txg != TXG_INITIAL) 6391 vdev_config_dirty(spa->spa_root_vdev); 6392 spa_history_log_internal(spa, "set", tx, 6393 "%s=%s", nvpair_name(elem), strval); 6394 break; 6395 default: 6396 /* 6397 * Set pool property values in the poolprops mos object. 6398 */ 6399 if (spa->spa_pool_props_object == 0) { 6400 spa->spa_pool_props_object = 6401 zap_create_link(mos, DMU_OT_POOL_PROPS, 6402 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6403 tx); 6404 } 6405 6406 /* normalize the property name */ 6407 propname = zpool_prop_to_name(prop); 6408 proptype = zpool_prop_get_type(prop); 6409 6410 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6411 ASSERT(proptype == PROP_TYPE_STRING); 6412 strval = fnvpair_value_string(elem); 6413 VERIFY0(zap_update(mos, 6414 spa->spa_pool_props_object, propname, 6415 1, strlen(strval) + 1, strval, tx)); 6416 spa_history_log_internal(spa, "set", tx, 6417 "%s=%s", nvpair_name(elem), strval); 6418 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6419 intval = fnvpair_value_uint64(elem); 6420 6421 if (proptype == PROP_TYPE_INDEX) { 6422 const char *unused; 6423 VERIFY0(zpool_prop_index_to_string( 6424 prop, intval, &unused)); 6425 } 6426 VERIFY0(zap_update(mos, 6427 spa->spa_pool_props_object, propname, 6428 8, 1, &intval, tx)); 6429 spa_history_log_internal(spa, "set", tx, 6430 "%s=%lld", nvpair_name(elem), intval); 6431 } else { 6432 ASSERT(0); /* not allowed */ 6433 } 6434 6435 switch (prop) { 6436 case ZPOOL_PROP_DELEGATION: 6437 spa->spa_delegation = intval; 6438 break; 6439 case ZPOOL_PROP_BOOTFS: 6440 spa->spa_bootfs = intval; 6441 break; 6442 case ZPOOL_PROP_FAILUREMODE: 6443 spa->spa_failmode = intval; 6444 break; 6445 case ZPOOL_PROP_AUTOEXPAND: 6446 spa->spa_autoexpand = intval; 6447 if (tx->tx_txg != TXG_INITIAL) 6448 spa_async_request(spa, 6449 SPA_ASYNC_AUTOEXPAND); 6450 break; 6451 case ZPOOL_PROP_DEDUPDITTO: 6452 spa->spa_dedup_ditto = intval; 6453 break; 6454 default: 6455 break; 6456 } 6457 } 6458 6459 } 6460 6461 mutex_exit(&spa->spa_props_lock); 6462} 6463 6464/* 6465 * Perform one-time upgrade on-disk changes. spa_version() does not 6466 * reflect the new version this txg, so there must be no changes this 6467 * txg to anything that the upgrade code depends on after it executes. 6468 * Therefore this must be called after dsl_pool_sync() does the sync 6469 * tasks. 6470 */ 6471static void 6472spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6473{ 6474 dsl_pool_t *dp = spa->spa_dsl_pool; 6475 6476 ASSERT(spa->spa_sync_pass == 1); 6477 6478 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6479 6480 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6481 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6482 dsl_pool_create_origin(dp, tx); 6483 6484 /* Keeping the origin open increases spa_minref */ 6485 spa->spa_minref += 3; 6486 } 6487 6488 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6489 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6490 dsl_pool_upgrade_clones(dp, tx); 6491 } 6492 6493 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6494 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6495 dsl_pool_upgrade_dir_clones(dp, tx); 6496 6497 /* Keeping the freedir open increases spa_minref */ 6498 spa->spa_minref += 3; 6499 } 6500 6501 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6502 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6503 spa_feature_create_zap_objects(spa, tx); 6504 } 6505 6506 /* 6507 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6508 * when possibility to use lz4 compression for metadata was added 6509 * Old pools that have this feature enabled must be upgraded to have 6510 * this feature active 6511 */ 6512 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6513 boolean_t lz4_en = spa_feature_is_enabled(spa, 6514 SPA_FEATURE_LZ4_COMPRESS); 6515 boolean_t lz4_ac = spa_feature_is_active(spa, 6516 SPA_FEATURE_LZ4_COMPRESS); 6517 6518 if (lz4_en && !lz4_ac) 6519 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6520 } 6521 rrw_exit(&dp->dp_config_rwlock, FTAG); 6522} 6523 6524/* 6525 * Sync the specified transaction group. New blocks may be dirtied as 6526 * part of the process, so we iterate until it converges. 6527 */ 6528void 6529spa_sync(spa_t *spa, uint64_t txg) 6530{ 6531 dsl_pool_t *dp = spa->spa_dsl_pool; 6532 objset_t *mos = spa->spa_meta_objset; 6533 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6534 vdev_t *rvd = spa->spa_root_vdev; 6535 vdev_t *vd; 6536 dmu_tx_t *tx; 6537 int error; 6538 6539 VERIFY(spa_writeable(spa)); 6540 6541 /* 6542 * Lock out configuration changes. 6543 */ 6544 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6545 6546 spa->spa_syncing_txg = txg; 6547 spa->spa_sync_pass = 0; 6548 6549 /* 6550 * If there are any pending vdev state changes, convert them 6551 * into config changes that go out with this transaction group. 6552 */ 6553 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6554 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6555 /* 6556 * We need the write lock here because, for aux vdevs, 6557 * calling vdev_config_dirty() modifies sav_config. 6558 * This is ugly and will become unnecessary when we 6559 * eliminate the aux vdev wart by integrating all vdevs 6560 * into the root vdev tree. 6561 */ 6562 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6563 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6564 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6565 vdev_state_clean(vd); 6566 vdev_config_dirty(vd); 6567 } 6568 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6569 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6570 } 6571 spa_config_exit(spa, SCL_STATE, FTAG); 6572 6573 tx = dmu_tx_create_assigned(dp, txg); 6574 6575 spa->spa_sync_starttime = gethrtime(); 6576#ifdef illumos 6577 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6578 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6579#else /* FreeBSD */ 6580#ifdef _KERNEL 6581 callout_reset(&spa->spa_deadman_cycid, 6582 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6583#endif 6584#endif 6585 6586 /* 6587 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6588 * set spa_deflate if we have no raid-z vdevs. 6589 */ 6590 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6591 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6592 int i; 6593 6594 for (i = 0; i < rvd->vdev_children; i++) { 6595 vd = rvd->vdev_child[i]; 6596 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6597 break; 6598 } 6599 if (i == rvd->vdev_children) { 6600 spa->spa_deflate = TRUE; 6601 VERIFY(0 == zap_add(spa->spa_meta_objset, 6602 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6603 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6604 } 6605 } 6606 6607 /* 6608 * If anything has changed in this txg, or if someone is waiting 6609 * for this txg to sync (eg, spa_vdev_remove()), push the 6610 * deferred frees from the previous txg. If not, leave them 6611 * alone so that we don't generate work on an otherwise idle 6612 * system. 6613 */ 6614 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6615 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6616 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6617 ((dsl_scan_active(dp->dp_scan) || 6618 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6619 spa_sync_deferred_frees(spa, tx); 6620 } 6621 6622 /* 6623 * Iterate to convergence. 6624 */ 6625 do { 6626 int pass = ++spa->spa_sync_pass; 6627 6628 spa_sync_config_object(spa, tx); 6629 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6630 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6631 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6632 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6633 spa_errlog_sync(spa, txg); 6634 dsl_pool_sync(dp, txg); 6635 6636 if (pass < zfs_sync_pass_deferred_free) { 6637 spa_sync_frees(spa, free_bpl, tx); 6638 } else { 6639 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6640 &spa->spa_deferred_bpobj, tx); 6641 } 6642 6643 ddt_sync(spa, txg); 6644 dsl_scan_sync(dp, tx); 6645 6646 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6647 vdev_sync(vd, txg); 6648 6649 if (pass == 1) 6650 spa_sync_upgrades(spa, tx); 6651 6652 } while (dmu_objset_is_dirty(mos, txg)); 6653 6654 /* 6655 * Rewrite the vdev configuration (which includes the uberblock) 6656 * to commit the transaction group. 6657 * 6658 * If there are no dirty vdevs, we sync the uberblock to a few 6659 * random top-level vdevs that are known to be visible in the 6660 * config cache (see spa_vdev_add() for a complete description). 6661 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6662 */ 6663 for (;;) { 6664 /* 6665 * We hold SCL_STATE to prevent vdev open/close/etc. 6666 * while we're attempting to write the vdev labels. 6667 */ 6668 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6669 6670 if (list_is_empty(&spa->spa_config_dirty_list)) { 6671 vdev_t *svd[SPA_DVAS_PER_BP]; 6672 int svdcount = 0; 6673 int children = rvd->vdev_children; 6674 int c0 = spa_get_random(children); 6675 6676 for (int c = 0; c < children; c++) { 6677 vd = rvd->vdev_child[(c0 + c) % children]; 6678 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6679 continue; 6680 svd[svdcount++] = vd; 6681 if (svdcount == SPA_DVAS_PER_BP) 6682 break; 6683 } 6684 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6685 if (error != 0) 6686 error = vdev_config_sync(svd, svdcount, txg, 6687 B_TRUE); 6688 } else { 6689 error = vdev_config_sync(rvd->vdev_child, 6690 rvd->vdev_children, txg, B_FALSE); 6691 if (error != 0) 6692 error = vdev_config_sync(rvd->vdev_child, 6693 rvd->vdev_children, txg, B_TRUE); 6694 } 6695 6696 if (error == 0) 6697 spa->spa_last_synced_guid = rvd->vdev_guid; 6698 6699 spa_config_exit(spa, SCL_STATE, FTAG); 6700 6701 if (error == 0) 6702 break; 6703 zio_suspend(spa, NULL); 6704 zio_resume_wait(spa); 6705 } 6706 dmu_tx_commit(tx); 6707 6708#ifdef illumos 6709 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6710#else /* FreeBSD */ 6711#ifdef _KERNEL 6712 callout_drain(&spa->spa_deadman_cycid); 6713#endif 6714#endif 6715 6716 /* 6717 * Clear the dirty config list. 6718 */ 6719 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6720 vdev_config_clean(vd); 6721 6722 /* 6723 * Now that the new config has synced transactionally, 6724 * let it become visible to the config cache. 6725 */ 6726 if (spa->spa_config_syncing != NULL) { 6727 spa_config_set(spa, spa->spa_config_syncing); 6728 spa->spa_config_txg = txg; 6729 spa->spa_config_syncing = NULL; 6730 } 6731 6732 spa->spa_ubsync = spa->spa_uberblock; 6733 6734 dsl_pool_sync_done(dp, txg); 6735 6736 /* 6737 * Update usable space statistics. 6738 */ 6739 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6740 vdev_sync_done(vd, txg); 6741 6742 spa_update_dspace(spa); 6743 6744 /* 6745 * It had better be the case that we didn't dirty anything 6746 * since vdev_config_sync(). 6747 */ 6748 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6749 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6750 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6751 6752 spa->spa_sync_pass = 0; 6753 6754 spa_config_exit(spa, SCL_CONFIG, FTAG); 6755 6756 spa_handle_ignored_writes(spa); 6757 6758 /* 6759 * If any async tasks have been requested, kick them off. 6760 */ 6761 spa_async_dispatch(spa); 6762 spa_async_dispatch_vd(spa); 6763} 6764 6765/* 6766 * Sync all pools. We don't want to hold the namespace lock across these 6767 * operations, so we take a reference on the spa_t and drop the lock during the 6768 * sync. 6769 */ 6770void 6771spa_sync_allpools(void) 6772{ 6773 spa_t *spa = NULL; 6774 mutex_enter(&spa_namespace_lock); 6775 while ((spa = spa_next(spa)) != NULL) { 6776 if (spa_state(spa) != POOL_STATE_ACTIVE || 6777 !spa_writeable(spa) || spa_suspended(spa)) 6778 continue; 6779 spa_open_ref(spa, FTAG); 6780 mutex_exit(&spa_namespace_lock); 6781 txg_wait_synced(spa_get_dsl(spa), 0); 6782 mutex_enter(&spa_namespace_lock); 6783 spa_close(spa, FTAG); 6784 } 6785 mutex_exit(&spa_namespace_lock); 6786} 6787 6788/* 6789 * ========================================================================== 6790 * Miscellaneous routines 6791 * ========================================================================== 6792 */ 6793 6794/* 6795 * Remove all pools in the system. 6796 */ 6797void 6798spa_evict_all(void) 6799{ 6800 spa_t *spa; 6801 6802 /* 6803 * Remove all cached state. All pools should be closed now, 6804 * so every spa in the AVL tree should be unreferenced. 6805 */ 6806 mutex_enter(&spa_namespace_lock); 6807 while ((spa = spa_next(NULL)) != NULL) { 6808 /* 6809 * Stop async tasks. The async thread may need to detach 6810 * a device that's been replaced, which requires grabbing 6811 * spa_namespace_lock, so we must drop it here. 6812 */ 6813 spa_open_ref(spa, FTAG); 6814 mutex_exit(&spa_namespace_lock); 6815 spa_async_suspend(spa); 6816 mutex_enter(&spa_namespace_lock); 6817 spa_close(spa, FTAG); 6818 6819 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6820 spa_unload(spa); 6821 spa_deactivate(spa); 6822 } 6823 spa_remove(spa); 6824 } 6825 mutex_exit(&spa_namespace_lock); 6826} 6827 6828vdev_t * 6829spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6830{ 6831 vdev_t *vd; 6832 int i; 6833 6834 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6835 return (vd); 6836 6837 if (aux) { 6838 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6839 vd = spa->spa_l2cache.sav_vdevs[i]; 6840 if (vd->vdev_guid == guid) 6841 return (vd); 6842 } 6843 6844 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6845 vd = spa->spa_spares.sav_vdevs[i]; 6846 if (vd->vdev_guid == guid) 6847 return (vd); 6848 } 6849 } 6850 6851 return (NULL); 6852} 6853 6854void 6855spa_upgrade(spa_t *spa, uint64_t version) 6856{ 6857 ASSERT(spa_writeable(spa)); 6858 6859 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6860 6861 /* 6862 * This should only be called for a non-faulted pool, and since a 6863 * future version would result in an unopenable pool, this shouldn't be 6864 * possible. 6865 */ 6866 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6867 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6868 6869 spa->spa_uberblock.ub_version = version; 6870 vdev_config_dirty(spa->spa_root_vdev); 6871 6872 spa_config_exit(spa, SCL_ALL, FTAG); 6873 6874 txg_wait_synced(spa_get_dsl(spa), 0); 6875} 6876 6877boolean_t 6878spa_has_spare(spa_t *spa, uint64_t guid) 6879{ 6880 int i; 6881 uint64_t spareguid; 6882 spa_aux_vdev_t *sav = &spa->spa_spares; 6883 6884 for (i = 0; i < sav->sav_count; i++) 6885 if (sav->sav_vdevs[i]->vdev_guid == guid) 6886 return (B_TRUE); 6887 6888 for (i = 0; i < sav->sav_npending; i++) { 6889 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6890 &spareguid) == 0 && spareguid == guid) 6891 return (B_TRUE); 6892 } 6893 6894 return (B_FALSE); 6895} 6896 6897/* 6898 * Check if a pool has an active shared spare device. 6899 * Note: reference count of an active spare is 2, as a spare and as a replace 6900 */ 6901static boolean_t 6902spa_has_active_shared_spare(spa_t *spa) 6903{ 6904 int i, refcnt; 6905 uint64_t pool; 6906 spa_aux_vdev_t *sav = &spa->spa_spares; 6907 6908 for (i = 0; i < sav->sav_count; i++) { 6909 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6910 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6911 refcnt > 2) 6912 return (B_TRUE); 6913 } 6914 6915 return (B_FALSE); 6916} 6917 6918/* 6919 * Post a sysevent corresponding to the given event. The 'name' must be one of 6920 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6921 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6922 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6923 * or zdb as real changes. 6924 */ 6925void 6926spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6927{ 6928#ifdef _KERNEL 6929 sysevent_t *ev; 6930 sysevent_attr_list_t *attr = NULL; 6931 sysevent_value_t value; 6932 sysevent_id_t eid; 6933 6934 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6935 SE_SLEEP); 6936 6937 value.value_type = SE_DATA_TYPE_STRING; 6938 value.value.sv_string = spa_name(spa); 6939 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6940 goto done; 6941 6942 value.value_type = SE_DATA_TYPE_UINT64; 6943 value.value.sv_uint64 = spa_guid(spa); 6944 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6945 goto done; 6946 6947 if (vd) { 6948 value.value_type = SE_DATA_TYPE_UINT64; 6949 value.value.sv_uint64 = vd->vdev_guid; 6950 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6951 SE_SLEEP) != 0) 6952 goto done; 6953 6954 if (vd->vdev_path) { 6955 value.value_type = SE_DATA_TYPE_STRING; 6956 value.value.sv_string = vd->vdev_path; 6957 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6958 &value, SE_SLEEP) != 0) 6959 goto done; 6960 } 6961 } 6962 6963 if (sysevent_attach_attributes(ev, attr) != 0) 6964 goto done; 6965 attr = NULL; 6966 6967 (void) log_sysevent(ev, SE_SLEEP, &eid); 6968 6969done: 6970 if (attr) 6971 sysevent_free_attr(attr); 6972 sysevent_free(ev); 6973#endif 6974} 6975