spa.c revision 331399
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved. 25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2013 Saso Kiselkov. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 * Copyright 2017 Joyent, Inc. 31 * Copyright (c) 2017 Datto Inc. 32 */ 33 34/* 35 * SPA: Storage Pool Allocator 36 * 37 * This file contains all the routines used when modifying on-disk SPA state. 38 * This includes opening, importing, destroying, exporting a pool, and syncing a 39 * pool. 40 */ 41 42#include <sys/zfs_context.h> 43#include <sys/fm/fs/zfs.h> 44#include <sys/spa_impl.h> 45#include <sys/zio.h> 46#include <sys/zio_checksum.h> 47#include <sys/dmu.h> 48#include <sys/dmu_tx.h> 49#include <sys/zap.h> 50#include <sys/zil.h> 51#include <sys/ddt.h> 52#include <sys/vdev_impl.h> 53#include <sys/metaslab.h> 54#include <sys/metaslab_impl.h> 55#include <sys/uberblock_impl.h> 56#include <sys/txg.h> 57#include <sys/avl.h> 58#include <sys/dmu_traverse.h> 59#include <sys/dmu_objset.h> 60#include <sys/unique.h> 61#include <sys/dsl_pool.h> 62#include <sys/dsl_dataset.h> 63#include <sys/dsl_dir.h> 64#include <sys/dsl_prop.h> 65#include <sys/dsl_synctask.h> 66#include <sys/fs/zfs.h> 67#include <sys/arc.h> 68#include <sys/callb.h> 69#include <sys/spa_boot.h> 70#include <sys/zfs_ioctl.h> 71#include <sys/dsl_scan.h> 72#include <sys/dmu_send.h> 73#include <sys/dsl_destroy.h> 74#include <sys/dsl_userhold.h> 75#include <sys/zfeature.h> 76#include <sys/zvol.h> 77#include <sys/trim_map.h> 78#include <sys/abd.h> 79 80#ifdef _KERNEL 81#include <sys/callb.h> 82#include <sys/cpupart.h> 83#include <sys/zone.h> 84#endif /* _KERNEL */ 85 86#include "zfs_prop.h" 87#include "zfs_comutil.h" 88 89/* Check hostid on import? */ 90static int check_hostid = 1; 91 92/* 93 * The interval, in seconds, at which failed configuration cache file writes 94 * should be retried. 95 */ 96static int zfs_ccw_retry_interval = 300; 97 98SYSCTL_DECL(_vfs_zfs); 99SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 100 "Check hostid on import?"); 101TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 102SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 103 &zfs_ccw_retry_interval, 0, 104 "Configuration cache file write, retry after failure, interval (seconds)"); 105 106typedef enum zti_modes { 107 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 108 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 109 ZTI_MODE_NULL, /* don't create a taskq */ 110 ZTI_NMODES 111} zti_modes_t; 112 113#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 114#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 115#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 116 117#define ZTI_N(n) ZTI_P(n, 1) 118#define ZTI_ONE ZTI_N(1) 119 120typedef struct zio_taskq_info { 121 zti_modes_t zti_mode; 122 uint_t zti_value; 123 uint_t zti_count; 124} zio_taskq_info_t; 125 126static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 127 "issue", "issue_high", "intr", "intr_high" 128}; 129 130/* 131 * This table defines the taskq settings for each ZFS I/O type. When 132 * initializing a pool, we use this table to create an appropriately sized 133 * taskq. Some operations are low volume and therefore have a small, static 134 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 135 * macros. Other operations process a large amount of data; the ZTI_BATCH 136 * macro causes us to create a taskq oriented for throughput. Some operations 137 * are so high frequency and short-lived that the taskq itself can become a a 138 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 139 * additional degree of parallelism specified by the number of threads per- 140 * taskq and the number of taskqs; when dispatching an event in this case, the 141 * particular taskq is chosen at random. 142 * 143 * The different taskq priorities are to handle the different contexts (issue 144 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 145 * need to be handled with minimum delay. 146 */ 147const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 148 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 149 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 150 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 151 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 152 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 153 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 154 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 155}; 156 157static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, 158 const char *name); 159static void spa_event_post(sysevent_t *ev); 160static void spa_sync_version(void *arg, dmu_tx_t *tx); 161static void spa_sync_props(void *arg, dmu_tx_t *tx); 162static boolean_t spa_has_active_shared_spare(spa_t *spa); 163static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 164 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 165 char **ereport); 166static void spa_vdev_resilver_done(spa_t *spa); 167 168uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 169#ifdef PSRSET_BIND 170id_t zio_taskq_psrset_bind = PS_NONE; 171#endif 172#ifdef SYSDC 173boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 174uint_t zio_taskq_basedc = 80; /* base duty cycle */ 175#endif 176 177boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 178extern int zfs_sync_pass_deferred_free; 179 180/* 181 * This (illegal) pool name is used when temporarily importing a spa_t in order 182 * to get the vdev stats associated with the imported devices. 183 */ 184#define TRYIMPORT_NAME "$import" 185 186/* 187 * ========================================================================== 188 * SPA properties routines 189 * ========================================================================== 190 */ 191 192/* 193 * Add a (source=src, propname=propval) list to an nvlist. 194 */ 195static void 196spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 197 uint64_t intval, zprop_source_t src) 198{ 199 const char *propname = zpool_prop_to_name(prop); 200 nvlist_t *propval; 201 202 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 203 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 204 205 if (strval != NULL) 206 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 207 else 208 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 209 210 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 211 nvlist_free(propval); 212} 213 214/* 215 * Get property values from the spa configuration. 216 */ 217static void 218spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 219{ 220 vdev_t *rvd = spa->spa_root_vdev; 221 dsl_pool_t *pool = spa->spa_dsl_pool; 222 uint64_t size, alloc, cap, version; 223 zprop_source_t src = ZPROP_SRC_NONE; 224 spa_config_dirent_t *dp; 225 metaslab_class_t *mc = spa_normal_class(spa); 226 227 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 228 229 if (rvd != NULL) { 230 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 231 size = metaslab_class_get_space(spa_normal_class(spa)); 232 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 233 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 234 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 235 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 236 size - alloc, src); 237 238 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 239 metaslab_class_fragmentation(mc), src); 240 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 241 metaslab_class_expandable_space(mc), src); 242 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 243 (spa_mode(spa) == FREAD), src); 244 245 cap = (size == 0) ? 0 : (alloc * 100 / size); 246 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 247 248 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 249 ddt_get_pool_dedup_ratio(spa), src); 250 251 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 252 rvd->vdev_state, src); 253 254 version = spa_version(spa); 255 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 256 src = ZPROP_SRC_DEFAULT; 257 else 258 src = ZPROP_SRC_LOCAL; 259 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 260 } 261 262 if (pool != NULL) { 263 /* 264 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 265 * when opening pools before this version freedir will be NULL. 266 */ 267 if (pool->dp_free_dir != NULL) { 268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 269 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 270 src); 271 } else { 272 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 273 NULL, 0, src); 274 } 275 276 if (pool->dp_leak_dir != NULL) { 277 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 278 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 279 src); 280 } else { 281 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 282 NULL, 0, src); 283 } 284 } 285 286 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 287 288 if (spa->spa_comment != NULL) { 289 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 290 0, ZPROP_SRC_LOCAL); 291 } 292 293 if (spa->spa_root != NULL) 294 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 295 0, ZPROP_SRC_LOCAL); 296 297 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 298 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 299 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 300 } else { 301 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 302 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 303 } 304 305 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 306 if (dp->scd_path == NULL) { 307 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 308 "none", 0, ZPROP_SRC_LOCAL); 309 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 310 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 311 dp->scd_path, 0, ZPROP_SRC_LOCAL); 312 } 313 } 314} 315 316/* 317 * Get zpool property values. 318 */ 319int 320spa_prop_get(spa_t *spa, nvlist_t **nvp) 321{ 322 objset_t *mos = spa->spa_meta_objset; 323 zap_cursor_t zc; 324 zap_attribute_t za; 325 int err; 326 327 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 328 329 mutex_enter(&spa->spa_props_lock); 330 331 /* 332 * Get properties from the spa config. 333 */ 334 spa_prop_get_config(spa, nvp); 335 336 /* If no pool property object, no more prop to get. */ 337 if (mos == NULL || spa->spa_pool_props_object == 0) { 338 mutex_exit(&spa->spa_props_lock); 339 return (0); 340 } 341 342 /* 343 * Get properties from the MOS pool property object. 344 */ 345 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 346 (err = zap_cursor_retrieve(&zc, &za)) == 0; 347 zap_cursor_advance(&zc)) { 348 uint64_t intval = 0; 349 char *strval = NULL; 350 zprop_source_t src = ZPROP_SRC_DEFAULT; 351 zpool_prop_t prop; 352 353 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL) 354 continue; 355 356 switch (za.za_integer_length) { 357 case 8: 358 /* integer property */ 359 if (za.za_first_integer != 360 zpool_prop_default_numeric(prop)) 361 src = ZPROP_SRC_LOCAL; 362 363 if (prop == ZPOOL_PROP_BOOTFS) { 364 dsl_pool_t *dp; 365 dsl_dataset_t *ds = NULL; 366 367 dp = spa_get_dsl(spa); 368 dsl_pool_config_enter(dp, FTAG); 369 if (err = dsl_dataset_hold_obj(dp, 370 za.za_first_integer, FTAG, &ds)) { 371 dsl_pool_config_exit(dp, FTAG); 372 break; 373 } 374 375 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 376 KM_SLEEP); 377 dsl_dataset_name(ds, strval); 378 dsl_dataset_rele(ds, FTAG); 379 dsl_pool_config_exit(dp, FTAG); 380 } else { 381 strval = NULL; 382 intval = za.za_first_integer; 383 } 384 385 spa_prop_add_list(*nvp, prop, strval, intval, src); 386 387 if (strval != NULL) 388 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 389 390 break; 391 392 case 1: 393 /* string property */ 394 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 395 err = zap_lookup(mos, spa->spa_pool_props_object, 396 za.za_name, 1, za.za_num_integers, strval); 397 if (err) { 398 kmem_free(strval, za.za_num_integers); 399 break; 400 } 401 spa_prop_add_list(*nvp, prop, strval, 0, src); 402 kmem_free(strval, za.za_num_integers); 403 break; 404 405 default: 406 break; 407 } 408 } 409 zap_cursor_fini(&zc); 410 mutex_exit(&spa->spa_props_lock); 411out: 412 if (err && err != ENOENT) { 413 nvlist_free(*nvp); 414 *nvp = NULL; 415 return (err); 416 } 417 418 return (0); 419} 420 421/* 422 * Validate the given pool properties nvlist and modify the list 423 * for the property values to be set. 424 */ 425static int 426spa_prop_validate(spa_t *spa, nvlist_t *props) 427{ 428 nvpair_t *elem; 429 int error = 0, reset_bootfs = 0; 430 uint64_t objnum = 0; 431 boolean_t has_feature = B_FALSE; 432 433 elem = NULL; 434 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 435 uint64_t intval; 436 char *strval, *slash, *check, *fname; 437 const char *propname = nvpair_name(elem); 438 zpool_prop_t prop = zpool_name_to_prop(propname); 439 440 switch (prop) { 441 case ZPOOL_PROP_INVAL: 442 if (!zpool_prop_feature(propname)) { 443 error = SET_ERROR(EINVAL); 444 break; 445 } 446 447 /* 448 * Sanitize the input. 449 */ 450 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 451 error = SET_ERROR(EINVAL); 452 break; 453 } 454 455 if (nvpair_value_uint64(elem, &intval) != 0) { 456 error = SET_ERROR(EINVAL); 457 break; 458 } 459 460 if (intval != 0) { 461 error = SET_ERROR(EINVAL); 462 break; 463 } 464 465 fname = strchr(propname, '@') + 1; 466 if (zfeature_lookup_name(fname, NULL) != 0) { 467 error = SET_ERROR(EINVAL); 468 break; 469 } 470 471 has_feature = B_TRUE; 472 break; 473 474 case ZPOOL_PROP_VERSION: 475 error = nvpair_value_uint64(elem, &intval); 476 if (!error && 477 (intval < spa_version(spa) || 478 intval > SPA_VERSION_BEFORE_FEATURES || 479 has_feature)) 480 error = SET_ERROR(EINVAL); 481 break; 482 483 case ZPOOL_PROP_DELEGATION: 484 case ZPOOL_PROP_AUTOREPLACE: 485 case ZPOOL_PROP_LISTSNAPS: 486 case ZPOOL_PROP_AUTOEXPAND: 487 error = nvpair_value_uint64(elem, &intval); 488 if (!error && intval > 1) 489 error = SET_ERROR(EINVAL); 490 break; 491 492 case ZPOOL_PROP_BOOTFS: 493 /* 494 * If the pool version is less than SPA_VERSION_BOOTFS, 495 * or the pool is still being created (version == 0), 496 * the bootfs property cannot be set. 497 */ 498 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 499 error = SET_ERROR(ENOTSUP); 500 break; 501 } 502 503 /* 504 * Make sure the vdev config is bootable 505 */ 506 if (!vdev_is_bootable(spa->spa_root_vdev)) { 507 error = SET_ERROR(ENOTSUP); 508 break; 509 } 510 511 reset_bootfs = 1; 512 513 error = nvpair_value_string(elem, &strval); 514 515 if (!error) { 516 objset_t *os; 517 uint64_t propval; 518 519 if (strval == NULL || strval[0] == '\0') { 520 objnum = zpool_prop_default_numeric( 521 ZPOOL_PROP_BOOTFS); 522 break; 523 } 524 525 if (error = dmu_objset_hold(strval, FTAG, &os)) 526 break; 527 528 /* 529 * Must be ZPL, and its property settings 530 * must be supported by GRUB (compression 531 * is not gzip, and large blocks are not used). 532 */ 533 534 if (dmu_objset_type(os) != DMU_OST_ZFS) { 535 error = SET_ERROR(ENOTSUP); 536 } else if ((error = 537 dsl_prop_get_int_ds(dmu_objset_ds(os), 538 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 539 &propval)) == 0 && 540 !BOOTFS_COMPRESS_VALID(propval)) { 541 error = SET_ERROR(ENOTSUP); 542 } else if ((error = 543 dsl_prop_get_int_ds(dmu_objset_ds(os), 544 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 545 &propval)) == 0 && 546 propval > SPA_OLD_MAXBLOCKSIZE) { 547 error = SET_ERROR(ENOTSUP); 548 } else { 549 objnum = dmu_objset_id(os); 550 } 551 dmu_objset_rele(os, FTAG); 552 } 553 break; 554 555 case ZPOOL_PROP_FAILUREMODE: 556 error = nvpair_value_uint64(elem, &intval); 557 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 558 intval > ZIO_FAILURE_MODE_PANIC)) 559 error = SET_ERROR(EINVAL); 560 561 /* 562 * This is a special case which only occurs when 563 * the pool has completely failed. This allows 564 * the user to change the in-core failmode property 565 * without syncing it out to disk (I/Os might 566 * currently be blocked). We do this by returning 567 * EIO to the caller (spa_prop_set) to trick it 568 * into thinking we encountered a property validation 569 * error. 570 */ 571 if (!error && spa_suspended(spa)) { 572 spa->spa_failmode = intval; 573 error = SET_ERROR(EIO); 574 } 575 break; 576 577 case ZPOOL_PROP_CACHEFILE: 578 if ((error = nvpair_value_string(elem, &strval)) != 0) 579 break; 580 581 if (strval[0] == '\0') 582 break; 583 584 if (strcmp(strval, "none") == 0) 585 break; 586 587 if (strval[0] != '/') { 588 error = SET_ERROR(EINVAL); 589 break; 590 } 591 592 slash = strrchr(strval, '/'); 593 ASSERT(slash != NULL); 594 595 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 596 strcmp(slash, "/..") == 0) 597 error = SET_ERROR(EINVAL); 598 break; 599 600 case ZPOOL_PROP_COMMENT: 601 if ((error = nvpair_value_string(elem, &strval)) != 0) 602 break; 603 for (check = strval; *check != '\0'; check++) { 604 /* 605 * The kernel doesn't have an easy isprint() 606 * check. For this kernel check, we merely 607 * check ASCII apart from DEL. Fix this if 608 * there is an easy-to-use kernel isprint(). 609 */ 610 if (*check >= 0x7f) { 611 error = SET_ERROR(EINVAL); 612 break; 613 } 614 } 615 if (strlen(strval) > ZPROP_MAX_COMMENT) 616 error = E2BIG; 617 break; 618 619 case ZPOOL_PROP_DEDUPDITTO: 620 if (spa_version(spa) < SPA_VERSION_DEDUP) 621 error = SET_ERROR(ENOTSUP); 622 else 623 error = nvpair_value_uint64(elem, &intval); 624 if (error == 0 && 625 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 626 error = SET_ERROR(EINVAL); 627 break; 628 } 629 630 if (error) 631 break; 632 } 633 634 if (!error && reset_bootfs) { 635 error = nvlist_remove(props, 636 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 637 638 if (!error) { 639 error = nvlist_add_uint64(props, 640 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 641 } 642 } 643 644 return (error); 645} 646 647void 648spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 649{ 650 char *cachefile; 651 spa_config_dirent_t *dp; 652 653 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 654 &cachefile) != 0) 655 return; 656 657 dp = kmem_alloc(sizeof (spa_config_dirent_t), 658 KM_SLEEP); 659 660 if (cachefile[0] == '\0') 661 dp->scd_path = spa_strdup(spa_config_path); 662 else if (strcmp(cachefile, "none") == 0) 663 dp->scd_path = NULL; 664 else 665 dp->scd_path = spa_strdup(cachefile); 666 667 list_insert_head(&spa->spa_config_list, dp); 668 if (need_sync) 669 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 670} 671 672int 673spa_prop_set(spa_t *spa, nvlist_t *nvp) 674{ 675 int error; 676 nvpair_t *elem = NULL; 677 boolean_t need_sync = B_FALSE; 678 679 if ((error = spa_prop_validate(spa, nvp)) != 0) 680 return (error); 681 682 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 683 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 684 685 if (prop == ZPOOL_PROP_CACHEFILE || 686 prop == ZPOOL_PROP_ALTROOT || 687 prop == ZPOOL_PROP_READONLY) 688 continue; 689 690 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) { 691 uint64_t ver; 692 693 if (prop == ZPOOL_PROP_VERSION) { 694 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 695 } else { 696 ASSERT(zpool_prop_feature(nvpair_name(elem))); 697 ver = SPA_VERSION_FEATURES; 698 need_sync = B_TRUE; 699 } 700 701 /* Save time if the version is already set. */ 702 if (ver == spa_version(spa)) 703 continue; 704 705 /* 706 * In addition to the pool directory object, we might 707 * create the pool properties object, the features for 708 * read object, the features for write object, or the 709 * feature descriptions object. 710 */ 711 error = dsl_sync_task(spa->spa_name, NULL, 712 spa_sync_version, &ver, 713 6, ZFS_SPACE_CHECK_RESERVED); 714 if (error) 715 return (error); 716 continue; 717 } 718 719 need_sync = B_TRUE; 720 break; 721 } 722 723 if (need_sync) { 724 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 725 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 726 } 727 728 return (0); 729} 730 731/* 732 * If the bootfs property value is dsobj, clear it. 733 */ 734void 735spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 736{ 737 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 738 VERIFY(zap_remove(spa->spa_meta_objset, 739 spa->spa_pool_props_object, 740 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 741 spa->spa_bootfs = 0; 742 } 743} 744 745/*ARGSUSED*/ 746static int 747spa_change_guid_check(void *arg, dmu_tx_t *tx) 748{ 749 uint64_t *newguid = arg; 750 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 751 vdev_t *rvd = spa->spa_root_vdev; 752 uint64_t vdev_state; 753 754 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 755 vdev_state = rvd->vdev_state; 756 spa_config_exit(spa, SCL_STATE, FTAG); 757 758 if (vdev_state != VDEV_STATE_HEALTHY) 759 return (SET_ERROR(ENXIO)); 760 761 ASSERT3U(spa_guid(spa), !=, *newguid); 762 763 return (0); 764} 765 766static void 767spa_change_guid_sync(void *arg, dmu_tx_t *tx) 768{ 769 uint64_t *newguid = arg; 770 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 771 uint64_t oldguid; 772 vdev_t *rvd = spa->spa_root_vdev; 773 774 oldguid = spa_guid(spa); 775 776 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 777 rvd->vdev_guid = *newguid; 778 rvd->vdev_guid_sum += (*newguid - oldguid); 779 vdev_config_dirty(rvd); 780 spa_config_exit(spa, SCL_STATE, FTAG); 781 782 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 783 oldguid, *newguid); 784} 785 786/* 787 * Change the GUID for the pool. This is done so that we can later 788 * re-import a pool built from a clone of our own vdevs. We will modify 789 * the root vdev's guid, our own pool guid, and then mark all of our 790 * vdevs dirty. Note that we must make sure that all our vdevs are 791 * online when we do this, or else any vdevs that weren't present 792 * would be orphaned from our pool. We are also going to issue a 793 * sysevent to update any watchers. 794 */ 795int 796spa_change_guid(spa_t *spa) 797{ 798 int error; 799 uint64_t guid; 800 801 mutex_enter(&spa->spa_vdev_top_lock); 802 mutex_enter(&spa_namespace_lock); 803 guid = spa_generate_guid(NULL); 804 805 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 806 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 807 808 if (error == 0) { 809 spa_config_sync(spa, B_FALSE, B_TRUE); 810 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID); 811 } 812 813 mutex_exit(&spa_namespace_lock); 814 mutex_exit(&spa->spa_vdev_top_lock); 815 816 return (error); 817} 818 819/* 820 * ========================================================================== 821 * SPA state manipulation (open/create/destroy/import/export) 822 * ========================================================================== 823 */ 824 825static int 826spa_error_entry_compare(const void *a, const void *b) 827{ 828 spa_error_entry_t *sa = (spa_error_entry_t *)a; 829 spa_error_entry_t *sb = (spa_error_entry_t *)b; 830 int ret; 831 832 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 833 sizeof (zbookmark_phys_t)); 834 835 if (ret < 0) 836 return (-1); 837 else if (ret > 0) 838 return (1); 839 else 840 return (0); 841} 842 843/* 844 * Utility function which retrieves copies of the current logs and 845 * re-initializes them in the process. 846 */ 847void 848spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 849{ 850 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 851 852 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 853 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 854 855 avl_create(&spa->spa_errlist_scrub, 856 spa_error_entry_compare, sizeof (spa_error_entry_t), 857 offsetof(spa_error_entry_t, se_avl)); 858 avl_create(&spa->spa_errlist_last, 859 spa_error_entry_compare, sizeof (spa_error_entry_t), 860 offsetof(spa_error_entry_t, se_avl)); 861} 862 863static void 864spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 865{ 866 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 867 enum zti_modes mode = ztip->zti_mode; 868 uint_t value = ztip->zti_value; 869 uint_t count = ztip->zti_count; 870 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 871 char name[32]; 872 uint_t flags = 0; 873 boolean_t batch = B_FALSE; 874 875 if (mode == ZTI_MODE_NULL) { 876 tqs->stqs_count = 0; 877 tqs->stqs_taskq = NULL; 878 return; 879 } 880 881 ASSERT3U(count, >, 0); 882 883 tqs->stqs_count = count; 884 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 885 886 switch (mode) { 887 case ZTI_MODE_FIXED: 888 ASSERT3U(value, >=, 1); 889 value = MAX(value, 1); 890 break; 891 892 case ZTI_MODE_BATCH: 893 batch = B_TRUE; 894 flags |= TASKQ_THREADS_CPU_PCT; 895 value = zio_taskq_batch_pct; 896 break; 897 898 default: 899 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 900 "spa_activate()", 901 zio_type_name[t], zio_taskq_types[q], mode, value); 902 break; 903 } 904 905 for (uint_t i = 0; i < count; i++) { 906 taskq_t *tq; 907 908 if (count > 1) { 909 (void) snprintf(name, sizeof (name), "%s_%s_%u", 910 zio_type_name[t], zio_taskq_types[q], i); 911 } else { 912 (void) snprintf(name, sizeof (name), "%s_%s", 913 zio_type_name[t], zio_taskq_types[q]); 914 } 915 916#ifdef SYSDC 917 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 918 if (batch) 919 flags |= TASKQ_DC_BATCH; 920 921 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 922 spa->spa_proc, zio_taskq_basedc, flags); 923 } else { 924#endif 925 pri_t pri = maxclsyspri; 926 /* 927 * The write issue taskq can be extremely CPU 928 * intensive. Run it at slightly lower priority 929 * than the other taskqs. 930 * FreeBSD notes: 931 * - numerically higher priorities are lower priorities; 932 * - if priorities divided by four (RQ_PPQ) are equal 933 * then a difference between them is insignificant. 934 */ 935 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 936#ifdef illumos 937 pri--; 938#else 939 pri += 4; 940#endif 941 942 tq = taskq_create_proc(name, value, pri, 50, 943 INT_MAX, spa->spa_proc, flags); 944#ifdef SYSDC 945 } 946#endif 947 948 tqs->stqs_taskq[i] = tq; 949 } 950} 951 952static void 953spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 954{ 955 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 956 957 if (tqs->stqs_taskq == NULL) { 958 ASSERT0(tqs->stqs_count); 959 return; 960 } 961 962 for (uint_t i = 0; i < tqs->stqs_count; i++) { 963 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 964 taskq_destroy(tqs->stqs_taskq[i]); 965 } 966 967 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 968 tqs->stqs_taskq = NULL; 969} 970 971/* 972 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 973 * Note that a type may have multiple discrete taskqs to avoid lock contention 974 * on the taskq itself. In that case we choose which taskq at random by using 975 * the low bits of gethrtime(). 976 */ 977void 978spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 979 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 980{ 981 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 982 taskq_t *tq; 983 984 ASSERT3P(tqs->stqs_taskq, !=, NULL); 985 ASSERT3U(tqs->stqs_count, !=, 0); 986 987 if (tqs->stqs_count == 1) { 988 tq = tqs->stqs_taskq[0]; 989 } else { 990#ifdef _KERNEL 991 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 992#else 993 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 994#endif 995 } 996 997 taskq_dispatch_ent(tq, func, arg, flags, ent); 998} 999 1000static void 1001spa_create_zio_taskqs(spa_t *spa) 1002{ 1003 for (int t = 0; t < ZIO_TYPES; t++) { 1004 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1005 spa_taskqs_init(spa, t, q); 1006 } 1007 } 1008} 1009 1010#ifdef _KERNEL 1011#ifdef SPA_PROCESS 1012static void 1013spa_thread(void *arg) 1014{ 1015 callb_cpr_t cprinfo; 1016 1017 spa_t *spa = arg; 1018 user_t *pu = PTOU(curproc); 1019 1020 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1021 spa->spa_name); 1022 1023 ASSERT(curproc != &p0); 1024 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1025 "zpool-%s", spa->spa_name); 1026 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1027 1028#ifdef PSRSET_BIND 1029 /* bind this thread to the requested psrset */ 1030 if (zio_taskq_psrset_bind != PS_NONE) { 1031 pool_lock(); 1032 mutex_enter(&cpu_lock); 1033 mutex_enter(&pidlock); 1034 mutex_enter(&curproc->p_lock); 1035 1036 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1037 0, NULL, NULL) == 0) { 1038 curthread->t_bind_pset = zio_taskq_psrset_bind; 1039 } else { 1040 cmn_err(CE_WARN, 1041 "Couldn't bind process for zfs pool \"%s\" to " 1042 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1043 } 1044 1045 mutex_exit(&curproc->p_lock); 1046 mutex_exit(&pidlock); 1047 mutex_exit(&cpu_lock); 1048 pool_unlock(); 1049 } 1050#endif 1051 1052#ifdef SYSDC 1053 if (zio_taskq_sysdc) { 1054 sysdc_thread_enter(curthread, 100, 0); 1055 } 1056#endif 1057 1058 spa->spa_proc = curproc; 1059 spa->spa_did = curthread->t_did; 1060 1061 spa_create_zio_taskqs(spa); 1062 1063 mutex_enter(&spa->spa_proc_lock); 1064 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1065 1066 spa->spa_proc_state = SPA_PROC_ACTIVE; 1067 cv_broadcast(&spa->spa_proc_cv); 1068 1069 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1070 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1071 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1072 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1073 1074 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1075 spa->spa_proc_state = SPA_PROC_GONE; 1076 spa->spa_proc = &p0; 1077 cv_broadcast(&spa->spa_proc_cv); 1078 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1079 1080 mutex_enter(&curproc->p_lock); 1081 lwp_exit(); 1082} 1083#endif /* SPA_PROCESS */ 1084#endif 1085 1086/* 1087 * Activate an uninitialized pool. 1088 */ 1089static void 1090spa_activate(spa_t *spa, int mode) 1091{ 1092 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1093 1094 spa->spa_state = POOL_STATE_ACTIVE; 1095 spa->spa_mode = mode; 1096 1097 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1098 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1099 1100 /* Try to create a covering process */ 1101 mutex_enter(&spa->spa_proc_lock); 1102 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1103 ASSERT(spa->spa_proc == &p0); 1104 spa->spa_did = 0; 1105 1106#ifdef SPA_PROCESS 1107 /* Only create a process if we're going to be around a while. */ 1108 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1109 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1110 NULL, 0) == 0) { 1111 spa->spa_proc_state = SPA_PROC_CREATED; 1112 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1113 cv_wait(&spa->spa_proc_cv, 1114 &spa->spa_proc_lock); 1115 } 1116 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1117 ASSERT(spa->spa_proc != &p0); 1118 ASSERT(spa->spa_did != 0); 1119 } else { 1120#ifdef _KERNEL 1121 cmn_err(CE_WARN, 1122 "Couldn't create process for zfs pool \"%s\"\n", 1123 spa->spa_name); 1124#endif 1125 } 1126 } 1127#endif /* SPA_PROCESS */ 1128 mutex_exit(&spa->spa_proc_lock); 1129 1130 /* If we didn't create a process, we need to create our taskqs. */ 1131 ASSERT(spa->spa_proc == &p0); 1132 if (spa->spa_proc == &p0) { 1133 spa_create_zio_taskqs(spa); 1134 } 1135 1136 /* 1137 * Start TRIM thread. 1138 */ 1139 trim_thread_create(spa); 1140 1141 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1142 offsetof(vdev_t, vdev_config_dirty_node)); 1143 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1144 offsetof(objset_t, os_evicting_node)); 1145 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1146 offsetof(vdev_t, vdev_state_dirty_node)); 1147 1148 txg_list_create(&spa->spa_vdev_txg_list, spa, 1149 offsetof(struct vdev, vdev_txg_node)); 1150 1151 avl_create(&spa->spa_errlist_scrub, 1152 spa_error_entry_compare, sizeof (spa_error_entry_t), 1153 offsetof(spa_error_entry_t, se_avl)); 1154 avl_create(&spa->spa_errlist_last, 1155 spa_error_entry_compare, sizeof (spa_error_entry_t), 1156 offsetof(spa_error_entry_t, se_avl)); 1157} 1158 1159/* 1160 * Opposite of spa_activate(). 1161 */ 1162static void 1163spa_deactivate(spa_t *spa) 1164{ 1165 ASSERT(spa->spa_sync_on == B_FALSE); 1166 ASSERT(spa->spa_dsl_pool == NULL); 1167 ASSERT(spa->spa_root_vdev == NULL); 1168 ASSERT(spa->spa_async_zio_root == NULL); 1169 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1170 1171 /* 1172 * Stop TRIM thread in case spa_unload() wasn't called directly 1173 * before spa_deactivate(). 1174 */ 1175 trim_thread_destroy(spa); 1176 1177 spa_evicting_os_wait(spa); 1178 1179 txg_list_destroy(&spa->spa_vdev_txg_list); 1180 1181 list_destroy(&spa->spa_config_dirty_list); 1182 list_destroy(&spa->spa_evicting_os_list); 1183 list_destroy(&spa->spa_state_dirty_list); 1184 1185 for (int t = 0; t < ZIO_TYPES; t++) { 1186 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1187 spa_taskqs_fini(spa, t, q); 1188 } 1189 } 1190 1191 metaslab_class_destroy(spa->spa_normal_class); 1192 spa->spa_normal_class = NULL; 1193 1194 metaslab_class_destroy(spa->spa_log_class); 1195 spa->spa_log_class = NULL; 1196 1197 /* 1198 * If this was part of an import or the open otherwise failed, we may 1199 * still have errors left in the queues. Empty them just in case. 1200 */ 1201 spa_errlog_drain(spa); 1202 1203 avl_destroy(&spa->spa_errlist_scrub); 1204 avl_destroy(&spa->spa_errlist_last); 1205 1206 spa->spa_state = POOL_STATE_UNINITIALIZED; 1207 1208 mutex_enter(&spa->spa_proc_lock); 1209 if (spa->spa_proc_state != SPA_PROC_NONE) { 1210 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1211 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1212 cv_broadcast(&spa->spa_proc_cv); 1213 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1214 ASSERT(spa->spa_proc != &p0); 1215 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1216 } 1217 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1218 spa->spa_proc_state = SPA_PROC_NONE; 1219 } 1220 ASSERT(spa->spa_proc == &p0); 1221 mutex_exit(&spa->spa_proc_lock); 1222 1223#ifdef SPA_PROCESS 1224 /* 1225 * We want to make sure spa_thread() has actually exited the ZFS 1226 * module, so that the module can't be unloaded out from underneath 1227 * it. 1228 */ 1229 if (spa->spa_did != 0) { 1230 thread_join(spa->spa_did); 1231 spa->spa_did = 0; 1232 } 1233#endif /* SPA_PROCESS */ 1234} 1235 1236/* 1237 * Verify a pool configuration, and construct the vdev tree appropriately. This 1238 * will create all the necessary vdevs in the appropriate layout, with each vdev 1239 * in the CLOSED state. This will prep the pool before open/creation/import. 1240 * All vdev validation is done by the vdev_alloc() routine. 1241 */ 1242static int 1243spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1244 uint_t id, int atype) 1245{ 1246 nvlist_t **child; 1247 uint_t children; 1248 int error; 1249 1250 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1251 return (error); 1252 1253 if ((*vdp)->vdev_ops->vdev_op_leaf) 1254 return (0); 1255 1256 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1257 &child, &children); 1258 1259 if (error == ENOENT) 1260 return (0); 1261 1262 if (error) { 1263 vdev_free(*vdp); 1264 *vdp = NULL; 1265 return (SET_ERROR(EINVAL)); 1266 } 1267 1268 for (int c = 0; c < children; c++) { 1269 vdev_t *vd; 1270 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1271 atype)) != 0) { 1272 vdev_free(*vdp); 1273 *vdp = NULL; 1274 return (error); 1275 } 1276 } 1277 1278 ASSERT(*vdp != NULL); 1279 1280 return (0); 1281} 1282 1283/* 1284 * Opposite of spa_load(). 1285 */ 1286static void 1287spa_unload(spa_t *spa) 1288{ 1289 int i; 1290 1291 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1292 1293 /* 1294 * Stop TRIM thread. 1295 */ 1296 trim_thread_destroy(spa); 1297 1298 /* 1299 * Stop async tasks. 1300 */ 1301 spa_async_suspend(spa); 1302 1303 /* 1304 * Stop syncing. 1305 */ 1306 if (spa->spa_sync_on) { 1307 txg_sync_stop(spa->spa_dsl_pool); 1308 spa->spa_sync_on = B_FALSE; 1309 } 1310 1311 /* 1312 * Even though vdev_free() also calls vdev_metaslab_fini, we need 1313 * to call it earlier, before we wait for async i/o to complete. 1314 * This ensures that there is no async metaslab prefetching, by 1315 * calling taskq_wait(mg_taskq). 1316 */ 1317 if (spa->spa_root_vdev != NULL) { 1318 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1319 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) 1320 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]); 1321 spa_config_exit(spa, SCL_ALL, FTAG); 1322 } 1323 1324 /* 1325 * Wait for any outstanding async I/O to complete. 1326 */ 1327 if (spa->spa_async_zio_root != NULL) { 1328 for (int i = 0; i < max_ncpus; i++) 1329 (void) zio_wait(spa->spa_async_zio_root[i]); 1330 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1331 spa->spa_async_zio_root = NULL; 1332 } 1333 1334 bpobj_close(&spa->spa_deferred_bpobj); 1335 1336 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1337 1338 /* 1339 * Close all vdevs. 1340 */ 1341 if (spa->spa_root_vdev) 1342 vdev_free(spa->spa_root_vdev); 1343 ASSERT(spa->spa_root_vdev == NULL); 1344 1345 /* 1346 * Close the dsl pool. 1347 */ 1348 if (spa->spa_dsl_pool) { 1349 dsl_pool_close(spa->spa_dsl_pool); 1350 spa->spa_dsl_pool = NULL; 1351 spa->spa_meta_objset = NULL; 1352 } 1353 1354 ddt_unload(spa); 1355 1356 /* 1357 * Drop and purge level 2 cache 1358 */ 1359 spa_l2cache_drop(spa); 1360 1361 for (i = 0; i < spa->spa_spares.sav_count; i++) 1362 vdev_free(spa->spa_spares.sav_vdevs[i]); 1363 if (spa->spa_spares.sav_vdevs) { 1364 kmem_free(spa->spa_spares.sav_vdevs, 1365 spa->spa_spares.sav_count * sizeof (void *)); 1366 spa->spa_spares.sav_vdevs = NULL; 1367 } 1368 if (spa->spa_spares.sav_config) { 1369 nvlist_free(spa->spa_spares.sav_config); 1370 spa->spa_spares.sav_config = NULL; 1371 } 1372 spa->spa_spares.sav_count = 0; 1373 1374 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1375 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1376 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1377 } 1378 if (spa->spa_l2cache.sav_vdevs) { 1379 kmem_free(spa->spa_l2cache.sav_vdevs, 1380 spa->spa_l2cache.sav_count * sizeof (void *)); 1381 spa->spa_l2cache.sav_vdevs = NULL; 1382 } 1383 if (spa->spa_l2cache.sav_config) { 1384 nvlist_free(spa->spa_l2cache.sav_config); 1385 spa->spa_l2cache.sav_config = NULL; 1386 } 1387 spa->spa_l2cache.sav_count = 0; 1388 1389 spa->spa_async_suspended = 0; 1390 1391 if (spa->spa_comment != NULL) { 1392 spa_strfree(spa->spa_comment); 1393 spa->spa_comment = NULL; 1394 } 1395 1396 spa_config_exit(spa, SCL_ALL, FTAG); 1397} 1398 1399/* 1400 * Load (or re-load) the current list of vdevs describing the active spares for 1401 * this pool. When this is called, we have some form of basic information in 1402 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1403 * then re-generate a more complete list including status information. 1404 */ 1405static void 1406spa_load_spares(spa_t *spa) 1407{ 1408 nvlist_t **spares; 1409 uint_t nspares; 1410 int i; 1411 vdev_t *vd, *tvd; 1412 1413 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1414 1415 /* 1416 * First, close and free any existing spare vdevs. 1417 */ 1418 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1419 vd = spa->spa_spares.sav_vdevs[i]; 1420 1421 /* Undo the call to spa_activate() below */ 1422 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1423 B_FALSE)) != NULL && tvd->vdev_isspare) 1424 spa_spare_remove(tvd); 1425 vdev_close(vd); 1426 vdev_free(vd); 1427 } 1428 1429 if (spa->spa_spares.sav_vdevs) 1430 kmem_free(spa->spa_spares.sav_vdevs, 1431 spa->spa_spares.sav_count * sizeof (void *)); 1432 1433 if (spa->spa_spares.sav_config == NULL) 1434 nspares = 0; 1435 else 1436 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1437 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1438 1439 spa->spa_spares.sav_count = (int)nspares; 1440 spa->spa_spares.sav_vdevs = NULL; 1441 1442 if (nspares == 0) 1443 return; 1444 1445 /* 1446 * Construct the array of vdevs, opening them to get status in the 1447 * process. For each spare, there is potentially two different vdev_t 1448 * structures associated with it: one in the list of spares (used only 1449 * for basic validation purposes) and one in the active vdev 1450 * configuration (if it's spared in). During this phase we open and 1451 * validate each vdev on the spare list. If the vdev also exists in the 1452 * active configuration, then we also mark this vdev as an active spare. 1453 */ 1454 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1455 KM_SLEEP); 1456 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1457 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1458 VDEV_ALLOC_SPARE) == 0); 1459 ASSERT(vd != NULL); 1460 1461 spa->spa_spares.sav_vdevs[i] = vd; 1462 1463 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1464 B_FALSE)) != NULL) { 1465 if (!tvd->vdev_isspare) 1466 spa_spare_add(tvd); 1467 1468 /* 1469 * We only mark the spare active if we were successfully 1470 * able to load the vdev. Otherwise, importing a pool 1471 * with a bad active spare would result in strange 1472 * behavior, because multiple pool would think the spare 1473 * is actively in use. 1474 * 1475 * There is a vulnerability here to an equally bizarre 1476 * circumstance, where a dead active spare is later 1477 * brought back to life (onlined or otherwise). Given 1478 * the rarity of this scenario, and the extra complexity 1479 * it adds, we ignore the possibility. 1480 */ 1481 if (!vdev_is_dead(tvd)) 1482 spa_spare_activate(tvd); 1483 } 1484 1485 vd->vdev_top = vd; 1486 vd->vdev_aux = &spa->spa_spares; 1487 1488 if (vdev_open(vd) != 0) 1489 continue; 1490 1491 if (vdev_validate_aux(vd) == 0) 1492 spa_spare_add(vd); 1493 } 1494 1495 /* 1496 * Recompute the stashed list of spares, with status information 1497 * this time. 1498 */ 1499 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1500 DATA_TYPE_NVLIST_ARRAY) == 0); 1501 1502 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1503 KM_SLEEP); 1504 for (i = 0; i < spa->spa_spares.sav_count; i++) 1505 spares[i] = vdev_config_generate(spa, 1506 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1507 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1508 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1509 for (i = 0; i < spa->spa_spares.sav_count; i++) 1510 nvlist_free(spares[i]); 1511 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1512} 1513 1514/* 1515 * Load (or re-load) the current list of vdevs describing the active l2cache for 1516 * this pool. When this is called, we have some form of basic information in 1517 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1518 * then re-generate a more complete list including status information. 1519 * Devices which are already active have their details maintained, and are 1520 * not re-opened. 1521 */ 1522static void 1523spa_load_l2cache(spa_t *spa) 1524{ 1525 nvlist_t **l2cache; 1526 uint_t nl2cache; 1527 int i, j, oldnvdevs; 1528 uint64_t guid; 1529 vdev_t *vd, **oldvdevs, **newvdevs; 1530 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1531 1532 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1533 1534 if (sav->sav_config != NULL) { 1535 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1536 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1537 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1538 } else { 1539 nl2cache = 0; 1540 newvdevs = NULL; 1541 } 1542 1543 oldvdevs = sav->sav_vdevs; 1544 oldnvdevs = sav->sav_count; 1545 sav->sav_vdevs = NULL; 1546 sav->sav_count = 0; 1547 1548 /* 1549 * Process new nvlist of vdevs. 1550 */ 1551 for (i = 0; i < nl2cache; i++) { 1552 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1553 &guid) == 0); 1554 1555 newvdevs[i] = NULL; 1556 for (j = 0; j < oldnvdevs; j++) { 1557 vd = oldvdevs[j]; 1558 if (vd != NULL && guid == vd->vdev_guid) { 1559 /* 1560 * Retain previous vdev for add/remove ops. 1561 */ 1562 newvdevs[i] = vd; 1563 oldvdevs[j] = NULL; 1564 break; 1565 } 1566 } 1567 1568 if (newvdevs[i] == NULL) { 1569 /* 1570 * Create new vdev 1571 */ 1572 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1573 VDEV_ALLOC_L2CACHE) == 0); 1574 ASSERT(vd != NULL); 1575 newvdevs[i] = vd; 1576 1577 /* 1578 * Commit this vdev as an l2cache device, 1579 * even if it fails to open. 1580 */ 1581 spa_l2cache_add(vd); 1582 1583 vd->vdev_top = vd; 1584 vd->vdev_aux = sav; 1585 1586 spa_l2cache_activate(vd); 1587 1588 if (vdev_open(vd) != 0) 1589 continue; 1590 1591 (void) vdev_validate_aux(vd); 1592 1593 if (!vdev_is_dead(vd)) 1594 l2arc_add_vdev(spa, vd); 1595 } 1596 } 1597 1598 /* 1599 * Purge vdevs that were dropped 1600 */ 1601 for (i = 0; i < oldnvdevs; i++) { 1602 uint64_t pool; 1603 1604 vd = oldvdevs[i]; 1605 if (vd != NULL) { 1606 ASSERT(vd->vdev_isl2cache); 1607 1608 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1609 pool != 0ULL && l2arc_vdev_present(vd)) 1610 l2arc_remove_vdev(vd); 1611 vdev_clear_stats(vd); 1612 vdev_free(vd); 1613 } 1614 } 1615 1616 if (oldvdevs) 1617 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1618 1619 if (sav->sav_config == NULL) 1620 goto out; 1621 1622 sav->sav_vdevs = newvdevs; 1623 sav->sav_count = (int)nl2cache; 1624 1625 /* 1626 * Recompute the stashed list of l2cache devices, with status 1627 * information this time. 1628 */ 1629 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1630 DATA_TYPE_NVLIST_ARRAY) == 0); 1631 1632 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1633 for (i = 0; i < sav->sav_count; i++) 1634 l2cache[i] = vdev_config_generate(spa, 1635 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1636 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1637 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1638out: 1639 for (i = 0; i < sav->sav_count; i++) 1640 nvlist_free(l2cache[i]); 1641 if (sav->sav_count) 1642 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1643} 1644 1645static int 1646load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1647{ 1648 dmu_buf_t *db; 1649 char *packed = NULL; 1650 size_t nvsize = 0; 1651 int error; 1652 *value = NULL; 1653 1654 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1655 if (error != 0) 1656 return (error); 1657 1658 nvsize = *(uint64_t *)db->db_data; 1659 dmu_buf_rele(db, FTAG); 1660 1661 packed = kmem_alloc(nvsize, KM_SLEEP); 1662 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1663 DMU_READ_PREFETCH); 1664 if (error == 0) 1665 error = nvlist_unpack(packed, nvsize, value, 0); 1666 kmem_free(packed, nvsize); 1667 1668 return (error); 1669} 1670 1671/* 1672 * Checks to see if the given vdev could not be opened, in which case we post a 1673 * sysevent to notify the autoreplace code that the device has been removed. 1674 */ 1675static void 1676spa_check_removed(vdev_t *vd) 1677{ 1678 for (int c = 0; c < vd->vdev_children; c++) 1679 spa_check_removed(vd->vdev_child[c]); 1680 1681 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1682 !vd->vdev_ishole) { 1683 zfs_post_autoreplace(vd->vdev_spa, vd); 1684 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK); 1685 } 1686} 1687 1688static void 1689spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd) 1690{ 1691 ASSERT3U(vd->vdev_children, ==, mvd->vdev_children); 1692 1693 vd->vdev_top_zap = mvd->vdev_top_zap; 1694 vd->vdev_leaf_zap = mvd->vdev_leaf_zap; 1695 1696 for (uint64_t i = 0; i < vd->vdev_children; i++) { 1697 spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]); 1698 } 1699} 1700 1701/* 1702 * Validate the current config against the MOS config 1703 */ 1704static boolean_t 1705spa_config_valid(spa_t *spa, nvlist_t *config) 1706{ 1707 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1708 nvlist_t *nv; 1709 1710 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1711 1712 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1713 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1714 1715 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1716 1717 /* 1718 * If we're doing a normal import, then build up any additional 1719 * diagnostic information about missing devices in this config. 1720 * We'll pass this up to the user for further processing. 1721 */ 1722 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1723 nvlist_t **child, *nv; 1724 uint64_t idx = 0; 1725 1726 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1727 KM_SLEEP); 1728 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1729 1730 for (int c = 0; c < rvd->vdev_children; c++) { 1731 vdev_t *tvd = rvd->vdev_child[c]; 1732 vdev_t *mtvd = mrvd->vdev_child[c]; 1733 1734 if (tvd->vdev_ops == &vdev_missing_ops && 1735 mtvd->vdev_ops != &vdev_missing_ops && 1736 mtvd->vdev_islog) 1737 child[idx++] = vdev_config_generate(spa, mtvd, 1738 B_FALSE, 0); 1739 } 1740 1741 if (idx) { 1742 VERIFY(nvlist_add_nvlist_array(nv, 1743 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1744 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1745 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1746 1747 for (int i = 0; i < idx; i++) 1748 nvlist_free(child[i]); 1749 } 1750 nvlist_free(nv); 1751 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1752 } 1753 1754 /* 1755 * Compare the root vdev tree with the information we have 1756 * from the MOS config (mrvd). Check each top-level vdev 1757 * with the corresponding MOS config top-level (mtvd). 1758 */ 1759 for (int c = 0; c < rvd->vdev_children; c++) { 1760 vdev_t *tvd = rvd->vdev_child[c]; 1761 vdev_t *mtvd = mrvd->vdev_child[c]; 1762 1763 /* 1764 * Resolve any "missing" vdevs in the current configuration. 1765 * If we find that the MOS config has more accurate information 1766 * about the top-level vdev then use that vdev instead. 1767 */ 1768 if (tvd->vdev_ops == &vdev_missing_ops && 1769 mtvd->vdev_ops != &vdev_missing_ops) { 1770 1771 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1772 continue; 1773 1774 /* 1775 * Device specific actions. 1776 */ 1777 if (mtvd->vdev_islog) { 1778 spa_set_log_state(spa, SPA_LOG_CLEAR); 1779 } else { 1780 /* 1781 * XXX - once we have 'readonly' pool 1782 * support we should be able to handle 1783 * missing data devices by transitioning 1784 * the pool to readonly. 1785 */ 1786 continue; 1787 } 1788 1789 /* 1790 * Swap the missing vdev with the data we were 1791 * able to obtain from the MOS config. 1792 */ 1793 vdev_remove_child(rvd, tvd); 1794 vdev_remove_child(mrvd, mtvd); 1795 1796 vdev_add_child(rvd, mtvd); 1797 vdev_add_child(mrvd, tvd); 1798 1799 spa_config_exit(spa, SCL_ALL, FTAG); 1800 vdev_load(mtvd); 1801 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1802 1803 vdev_reopen(rvd); 1804 } else { 1805 if (mtvd->vdev_islog) { 1806 /* 1807 * Load the slog device's state from the MOS 1808 * config since it's possible that the label 1809 * does not contain the most up-to-date 1810 * information. 1811 */ 1812 vdev_load_log_state(tvd, mtvd); 1813 vdev_reopen(tvd); 1814 } 1815 1816 /* 1817 * Per-vdev ZAP info is stored exclusively in the MOS. 1818 */ 1819 spa_config_valid_zaps(tvd, mtvd); 1820 } 1821 } 1822 1823 vdev_free(mrvd); 1824 spa_config_exit(spa, SCL_ALL, FTAG); 1825 1826 /* 1827 * Ensure we were able to validate the config. 1828 */ 1829 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1830} 1831 1832/* 1833 * Check for missing log devices 1834 */ 1835static boolean_t 1836spa_check_logs(spa_t *spa) 1837{ 1838 boolean_t rv = B_FALSE; 1839 dsl_pool_t *dp = spa_get_dsl(spa); 1840 1841 switch (spa->spa_log_state) { 1842 case SPA_LOG_MISSING: 1843 /* need to recheck in case slog has been restored */ 1844 case SPA_LOG_UNKNOWN: 1845 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1846 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1847 if (rv) 1848 spa_set_log_state(spa, SPA_LOG_MISSING); 1849 break; 1850 } 1851 return (rv); 1852} 1853 1854static boolean_t 1855spa_passivate_log(spa_t *spa) 1856{ 1857 vdev_t *rvd = spa->spa_root_vdev; 1858 boolean_t slog_found = B_FALSE; 1859 1860 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1861 1862 if (!spa_has_slogs(spa)) 1863 return (B_FALSE); 1864 1865 for (int c = 0; c < rvd->vdev_children; c++) { 1866 vdev_t *tvd = rvd->vdev_child[c]; 1867 metaslab_group_t *mg = tvd->vdev_mg; 1868 1869 if (tvd->vdev_islog) { 1870 metaslab_group_passivate(mg); 1871 slog_found = B_TRUE; 1872 } 1873 } 1874 1875 return (slog_found); 1876} 1877 1878static void 1879spa_activate_log(spa_t *spa) 1880{ 1881 vdev_t *rvd = spa->spa_root_vdev; 1882 1883 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1884 1885 for (int c = 0; c < rvd->vdev_children; c++) { 1886 vdev_t *tvd = rvd->vdev_child[c]; 1887 metaslab_group_t *mg = tvd->vdev_mg; 1888 1889 if (tvd->vdev_islog) 1890 metaslab_group_activate(mg); 1891 } 1892} 1893 1894int 1895spa_offline_log(spa_t *spa) 1896{ 1897 int error; 1898 1899 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1900 NULL, DS_FIND_CHILDREN); 1901 if (error == 0) { 1902 /* 1903 * We successfully offlined the log device, sync out the 1904 * current txg so that the "stubby" block can be removed 1905 * by zil_sync(). 1906 */ 1907 txg_wait_synced(spa->spa_dsl_pool, 0); 1908 } 1909 return (error); 1910} 1911 1912static void 1913spa_aux_check_removed(spa_aux_vdev_t *sav) 1914{ 1915 int i; 1916 1917 for (i = 0; i < sav->sav_count; i++) 1918 spa_check_removed(sav->sav_vdevs[i]); 1919} 1920 1921void 1922spa_claim_notify(zio_t *zio) 1923{ 1924 spa_t *spa = zio->io_spa; 1925 1926 if (zio->io_error) 1927 return; 1928 1929 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1930 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1931 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1932 mutex_exit(&spa->spa_props_lock); 1933} 1934 1935typedef struct spa_load_error { 1936 uint64_t sle_meta_count; 1937 uint64_t sle_data_count; 1938} spa_load_error_t; 1939 1940static void 1941spa_load_verify_done(zio_t *zio) 1942{ 1943 blkptr_t *bp = zio->io_bp; 1944 spa_load_error_t *sle = zio->io_private; 1945 dmu_object_type_t type = BP_GET_TYPE(bp); 1946 int error = zio->io_error; 1947 spa_t *spa = zio->io_spa; 1948 1949 abd_free(zio->io_abd); 1950 if (error) { 1951 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1952 type != DMU_OT_INTENT_LOG) 1953 atomic_inc_64(&sle->sle_meta_count); 1954 else 1955 atomic_inc_64(&sle->sle_data_count); 1956 } 1957 1958 mutex_enter(&spa->spa_scrub_lock); 1959 spa->spa_scrub_inflight--; 1960 cv_broadcast(&spa->spa_scrub_io_cv); 1961 mutex_exit(&spa->spa_scrub_lock); 1962} 1963 1964/* 1965 * Maximum number of concurrent scrub i/os to create while verifying 1966 * a pool while importing it. 1967 */ 1968int spa_load_verify_maxinflight = 10000; 1969boolean_t spa_load_verify_metadata = B_TRUE; 1970boolean_t spa_load_verify_data = B_TRUE; 1971 1972SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1973 &spa_load_verify_maxinflight, 0, 1974 "Maximum number of concurrent scrub I/Os to create while verifying a " 1975 "pool while importing it"); 1976 1977SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1978 &spa_load_verify_metadata, 0, 1979 "Check metadata on import?"); 1980 1981SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1982 &spa_load_verify_data, 0, 1983 "Check user data on import?"); 1984 1985/*ARGSUSED*/ 1986static int 1987spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1988 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1989{ 1990 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1991 return (0); 1992 /* 1993 * Note: normally this routine will not be called if 1994 * spa_load_verify_metadata is not set. However, it may be useful 1995 * to manually set the flag after the traversal has begun. 1996 */ 1997 if (!spa_load_verify_metadata) 1998 return (0); 1999 if (!BP_IS_METADATA(bp) && !spa_load_verify_data) 2000 return (0); 2001 2002 zio_t *rio = arg; 2003 size_t size = BP_GET_PSIZE(bp); 2004 2005 mutex_enter(&spa->spa_scrub_lock); 2006 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 2007 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 2008 spa->spa_scrub_inflight++; 2009 mutex_exit(&spa->spa_scrub_lock); 2010 2011 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size, 2012 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 2013 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 2014 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 2015 return (0); 2016} 2017 2018/* ARGSUSED */ 2019int 2020verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 2021{ 2022 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 2023 return (SET_ERROR(ENAMETOOLONG)); 2024 2025 return (0); 2026} 2027 2028static int 2029spa_load_verify(spa_t *spa) 2030{ 2031 zio_t *rio; 2032 spa_load_error_t sle = { 0 }; 2033 zpool_rewind_policy_t policy; 2034 boolean_t verify_ok = B_FALSE; 2035 int error = 0; 2036 2037 zpool_get_rewind_policy(spa->spa_config, &policy); 2038 2039 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 2040 return (0); 2041 2042 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2043 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2044 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2045 DS_FIND_CHILDREN); 2046 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2047 if (error != 0) 2048 return (error); 2049 2050 rio = zio_root(spa, NULL, &sle, 2051 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2052 2053 if (spa_load_verify_metadata) { 2054 error = traverse_pool(spa, spa->spa_verify_min_txg, 2055 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2056 spa_load_verify_cb, rio); 2057 } 2058 2059 (void) zio_wait(rio); 2060 2061 spa->spa_load_meta_errors = sle.sle_meta_count; 2062 spa->spa_load_data_errors = sle.sle_data_count; 2063 2064 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2065 sle.sle_data_count <= policy.zrp_maxdata) { 2066 int64_t loss = 0; 2067 2068 verify_ok = B_TRUE; 2069 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2070 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2071 2072 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2073 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2074 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2075 VERIFY(nvlist_add_int64(spa->spa_load_info, 2076 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2077 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2078 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2079 } else { 2080 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2081 } 2082 2083 if (error) { 2084 if (error != ENXIO && error != EIO) 2085 error = SET_ERROR(EIO); 2086 return (error); 2087 } 2088 2089 return (verify_ok ? 0 : EIO); 2090} 2091 2092/* 2093 * Find a value in the pool props object. 2094 */ 2095static void 2096spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2097{ 2098 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2099 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2100} 2101 2102/* 2103 * Find a value in the pool directory object. 2104 */ 2105static int 2106spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2107{ 2108 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2109 name, sizeof (uint64_t), 1, val)); 2110} 2111 2112static int 2113spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2114{ 2115 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2116 return (err); 2117} 2118 2119/* 2120 * Fix up config after a partly-completed split. This is done with the 2121 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2122 * pool have that entry in their config, but only the splitting one contains 2123 * a list of all the guids of the vdevs that are being split off. 2124 * 2125 * This function determines what to do with that list: either rejoin 2126 * all the disks to the pool, or complete the splitting process. To attempt 2127 * the rejoin, each disk that is offlined is marked online again, and 2128 * we do a reopen() call. If the vdev label for every disk that was 2129 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2130 * then we call vdev_split() on each disk, and complete the split. 2131 * 2132 * Otherwise we leave the config alone, with all the vdevs in place in 2133 * the original pool. 2134 */ 2135static void 2136spa_try_repair(spa_t *spa, nvlist_t *config) 2137{ 2138 uint_t extracted; 2139 uint64_t *glist; 2140 uint_t i, gcount; 2141 nvlist_t *nvl; 2142 vdev_t **vd; 2143 boolean_t attempt_reopen; 2144 2145 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2146 return; 2147 2148 /* check that the config is complete */ 2149 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2150 &glist, &gcount) != 0) 2151 return; 2152 2153 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2154 2155 /* attempt to online all the vdevs & validate */ 2156 attempt_reopen = B_TRUE; 2157 for (i = 0; i < gcount; i++) { 2158 if (glist[i] == 0) /* vdev is hole */ 2159 continue; 2160 2161 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2162 if (vd[i] == NULL) { 2163 /* 2164 * Don't bother attempting to reopen the disks; 2165 * just do the split. 2166 */ 2167 attempt_reopen = B_FALSE; 2168 } else { 2169 /* attempt to re-online it */ 2170 vd[i]->vdev_offline = B_FALSE; 2171 } 2172 } 2173 2174 if (attempt_reopen) { 2175 vdev_reopen(spa->spa_root_vdev); 2176 2177 /* check each device to see what state it's in */ 2178 for (extracted = 0, i = 0; i < gcount; i++) { 2179 if (vd[i] != NULL && 2180 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2181 break; 2182 ++extracted; 2183 } 2184 } 2185 2186 /* 2187 * If every disk has been moved to the new pool, or if we never 2188 * even attempted to look at them, then we split them off for 2189 * good. 2190 */ 2191 if (!attempt_reopen || gcount == extracted) { 2192 for (i = 0; i < gcount; i++) 2193 if (vd[i] != NULL) 2194 vdev_split(vd[i]); 2195 vdev_reopen(spa->spa_root_vdev); 2196 } 2197 2198 kmem_free(vd, gcount * sizeof (vdev_t *)); 2199} 2200 2201static int 2202spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2203 boolean_t mosconfig) 2204{ 2205 nvlist_t *config = spa->spa_config; 2206 char *ereport = FM_EREPORT_ZFS_POOL; 2207 char *comment; 2208 int error; 2209 uint64_t pool_guid; 2210 nvlist_t *nvl; 2211 2212 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2213 return (SET_ERROR(EINVAL)); 2214 2215 ASSERT(spa->spa_comment == NULL); 2216 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2217 spa->spa_comment = spa_strdup(comment); 2218 2219 /* 2220 * Versioning wasn't explicitly added to the label until later, so if 2221 * it's not present treat it as the initial version. 2222 */ 2223 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2224 &spa->spa_ubsync.ub_version) != 0) 2225 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2226 2227 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2228 &spa->spa_config_txg); 2229 2230 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2231 spa_guid_exists(pool_guid, 0)) { 2232 error = SET_ERROR(EEXIST); 2233 } else { 2234 spa->spa_config_guid = pool_guid; 2235 2236 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2237 &nvl) == 0) { 2238 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2239 KM_SLEEP) == 0); 2240 } 2241 2242 nvlist_free(spa->spa_load_info); 2243 spa->spa_load_info = fnvlist_alloc(); 2244 2245 gethrestime(&spa->spa_loaded_ts); 2246 error = spa_load_impl(spa, pool_guid, config, state, type, 2247 mosconfig, &ereport); 2248 } 2249 2250 /* 2251 * Don't count references from objsets that are already closed 2252 * and are making their way through the eviction process. 2253 */ 2254 spa_evicting_os_wait(spa); 2255 spa->spa_minref = refcount_count(&spa->spa_refcount); 2256 if (error) { 2257 if (error != EEXIST) { 2258 spa->spa_loaded_ts.tv_sec = 0; 2259 spa->spa_loaded_ts.tv_nsec = 0; 2260 } 2261 if (error != EBADF) { 2262 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2263 } 2264 } 2265 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2266 spa->spa_ena = 0; 2267 2268 return (error); 2269} 2270 2271/* 2272 * Count the number of per-vdev ZAPs associated with all of the vdevs in the 2273 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the 2274 * spa's per-vdev ZAP list. 2275 */ 2276static uint64_t 2277vdev_count_verify_zaps(vdev_t *vd) 2278{ 2279 spa_t *spa = vd->vdev_spa; 2280 uint64_t total = 0; 2281 if (vd->vdev_top_zap != 0) { 2282 total++; 2283 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2284 spa->spa_all_vdev_zaps, vd->vdev_top_zap)); 2285 } 2286 if (vd->vdev_leaf_zap != 0) { 2287 total++; 2288 ASSERT0(zap_lookup_int(spa->spa_meta_objset, 2289 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap)); 2290 } 2291 2292 for (uint64_t i = 0; i < vd->vdev_children; i++) { 2293 total += vdev_count_verify_zaps(vd->vdev_child[i]); 2294 } 2295 2296 return (total); 2297} 2298 2299/* 2300 * Load an existing storage pool, using the pool's builtin spa_config as a 2301 * source of configuration information. 2302 */ 2303static int 2304spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2305 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2306 char **ereport) 2307{ 2308 int error = 0; 2309 nvlist_t *nvroot = NULL; 2310 nvlist_t *label; 2311 vdev_t *rvd; 2312 uberblock_t *ub = &spa->spa_uberblock; 2313 uint64_t children, config_cache_txg = spa->spa_config_txg; 2314 int orig_mode = spa->spa_mode; 2315 int parse; 2316 uint64_t obj; 2317 boolean_t missing_feat_write = B_FALSE; 2318 2319 /* 2320 * If this is an untrusted config, access the pool in read-only mode. 2321 * This prevents things like resilvering recently removed devices. 2322 */ 2323 if (!mosconfig) 2324 spa->spa_mode = FREAD; 2325 2326 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2327 2328 spa->spa_load_state = state; 2329 2330 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2331 return (SET_ERROR(EINVAL)); 2332 2333 parse = (type == SPA_IMPORT_EXISTING ? 2334 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2335 2336 /* 2337 * Create "The Godfather" zio to hold all async IOs 2338 */ 2339 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2340 KM_SLEEP); 2341 for (int i = 0; i < max_ncpus; i++) { 2342 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2343 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2344 ZIO_FLAG_GODFATHER); 2345 } 2346 2347 /* 2348 * Parse the configuration into a vdev tree. We explicitly set the 2349 * value that will be returned by spa_version() since parsing the 2350 * configuration requires knowing the version number. 2351 */ 2352 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2353 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2354 spa_config_exit(spa, SCL_ALL, FTAG); 2355 2356 if (error != 0) 2357 return (error); 2358 2359 ASSERT(spa->spa_root_vdev == rvd); 2360 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2361 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2362 2363 if (type != SPA_IMPORT_ASSEMBLE) { 2364 ASSERT(spa_guid(spa) == pool_guid); 2365 } 2366 2367 /* 2368 * Try to open all vdevs, loading each label in the process. 2369 */ 2370 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2371 error = vdev_open(rvd); 2372 spa_config_exit(spa, SCL_ALL, FTAG); 2373 if (error != 0) 2374 return (error); 2375 2376 /* 2377 * We need to validate the vdev labels against the configuration that 2378 * we have in hand, which is dependent on the setting of mosconfig. If 2379 * mosconfig is true then we're validating the vdev labels based on 2380 * that config. Otherwise, we're validating against the cached config 2381 * (zpool.cache) that was read when we loaded the zfs module, and then 2382 * later we will recursively call spa_load() and validate against 2383 * the vdev config. 2384 * 2385 * If we're assembling a new pool that's been split off from an 2386 * existing pool, the labels haven't yet been updated so we skip 2387 * validation for now. 2388 */ 2389 if (type != SPA_IMPORT_ASSEMBLE) { 2390 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2391 error = vdev_validate(rvd, mosconfig); 2392 spa_config_exit(spa, SCL_ALL, FTAG); 2393 2394 if (error != 0) 2395 return (error); 2396 2397 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2398 return (SET_ERROR(ENXIO)); 2399 } 2400 2401 /* 2402 * Find the best uberblock. 2403 */ 2404 vdev_uberblock_load(rvd, ub, &label); 2405 2406 /* 2407 * If we weren't able to find a single valid uberblock, return failure. 2408 */ 2409 if (ub->ub_txg == 0) { 2410 nvlist_free(label); 2411 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2412 } 2413 2414 /* 2415 * If the pool has an unsupported version we can't open it. 2416 */ 2417 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2418 nvlist_free(label); 2419 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2420 } 2421 2422 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2423 nvlist_t *features; 2424 2425 /* 2426 * If we weren't able to find what's necessary for reading the 2427 * MOS in the label, return failure. 2428 */ 2429 if (label == NULL || nvlist_lookup_nvlist(label, 2430 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2431 nvlist_free(label); 2432 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2433 ENXIO)); 2434 } 2435 2436 /* 2437 * Update our in-core representation with the definitive values 2438 * from the label. 2439 */ 2440 nvlist_free(spa->spa_label_features); 2441 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2442 } 2443 2444 nvlist_free(label); 2445 2446 /* 2447 * Look through entries in the label nvlist's features_for_read. If 2448 * there is a feature listed there which we don't understand then we 2449 * cannot open a pool. 2450 */ 2451 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2452 nvlist_t *unsup_feat; 2453 2454 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2455 0); 2456 2457 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2458 NULL); nvp != NULL; 2459 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2460 if (!zfeature_is_supported(nvpair_name(nvp))) { 2461 VERIFY(nvlist_add_string(unsup_feat, 2462 nvpair_name(nvp), "") == 0); 2463 } 2464 } 2465 2466 if (!nvlist_empty(unsup_feat)) { 2467 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2468 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2469 nvlist_free(unsup_feat); 2470 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2471 ENOTSUP)); 2472 } 2473 2474 nvlist_free(unsup_feat); 2475 } 2476 2477 /* 2478 * If the vdev guid sum doesn't match the uberblock, we have an 2479 * incomplete configuration. We first check to see if the pool 2480 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2481 * If it is, defer the vdev_guid_sum check till later so we 2482 * can handle missing vdevs. 2483 */ 2484 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2485 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2486 rvd->vdev_guid_sum != ub->ub_guid_sum) 2487 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2488 2489 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2490 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2491 spa_try_repair(spa, config); 2492 spa_config_exit(spa, SCL_ALL, FTAG); 2493 nvlist_free(spa->spa_config_splitting); 2494 spa->spa_config_splitting = NULL; 2495 } 2496 2497 /* 2498 * Initialize internal SPA structures. 2499 */ 2500 spa->spa_state = POOL_STATE_ACTIVE; 2501 spa->spa_ubsync = spa->spa_uberblock; 2502 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2503 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2504 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2505 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2506 spa->spa_claim_max_txg = spa->spa_first_txg; 2507 spa->spa_prev_software_version = ub->ub_software_version; 2508 2509 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2510 if (error) 2511 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2512 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2513 2514 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2515 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2516 2517 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2518 boolean_t missing_feat_read = B_FALSE; 2519 nvlist_t *unsup_feat, *enabled_feat; 2520 2521 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2522 &spa->spa_feat_for_read_obj) != 0) { 2523 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2524 } 2525 2526 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2527 &spa->spa_feat_for_write_obj) != 0) { 2528 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2529 } 2530 2531 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2532 &spa->spa_feat_desc_obj) != 0) { 2533 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2534 } 2535 2536 enabled_feat = fnvlist_alloc(); 2537 unsup_feat = fnvlist_alloc(); 2538 2539 if (!spa_features_check(spa, B_FALSE, 2540 unsup_feat, enabled_feat)) 2541 missing_feat_read = B_TRUE; 2542 2543 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2544 if (!spa_features_check(spa, B_TRUE, 2545 unsup_feat, enabled_feat)) { 2546 missing_feat_write = B_TRUE; 2547 } 2548 } 2549 2550 fnvlist_add_nvlist(spa->spa_load_info, 2551 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2552 2553 if (!nvlist_empty(unsup_feat)) { 2554 fnvlist_add_nvlist(spa->spa_load_info, 2555 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2556 } 2557 2558 fnvlist_free(enabled_feat); 2559 fnvlist_free(unsup_feat); 2560 2561 if (!missing_feat_read) { 2562 fnvlist_add_boolean(spa->spa_load_info, 2563 ZPOOL_CONFIG_CAN_RDONLY); 2564 } 2565 2566 /* 2567 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2568 * twofold: to determine whether the pool is available for 2569 * import in read-write mode and (if it is not) whether the 2570 * pool is available for import in read-only mode. If the pool 2571 * is available for import in read-write mode, it is displayed 2572 * as available in userland; if it is not available for import 2573 * in read-only mode, it is displayed as unavailable in 2574 * userland. If the pool is available for import in read-only 2575 * mode but not read-write mode, it is displayed as unavailable 2576 * in userland with a special note that the pool is actually 2577 * available for open in read-only mode. 2578 * 2579 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2580 * missing a feature for write, we must first determine whether 2581 * the pool can be opened read-only before returning to 2582 * userland in order to know whether to display the 2583 * abovementioned note. 2584 */ 2585 if (missing_feat_read || (missing_feat_write && 2586 spa_writeable(spa))) { 2587 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2588 ENOTSUP)); 2589 } 2590 2591 /* 2592 * Load refcounts for ZFS features from disk into an in-memory 2593 * cache during SPA initialization. 2594 */ 2595 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2596 uint64_t refcount; 2597 2598 error = feature_get_refcount_from_disk(spa, 2599 &spa_feature_table[i], &refcount); 2600 if (error == 0) { 2601 spa->spa_feat_refcount_cache[i] = refcount; 2602 } else if (error == ENOTSUP) { 2603 spa->spa_feat_refcount_cache[i] = 2604 SPA_FEATURE_DISABLED; 2605 } else { 2606 return (spa_vdev_err(rvd, 2607 VDEV_AUX_CORRUPT_DATA, EIO)); 2608 } 2609 } 2610 } 2611 2612 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2613 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2614 &spa->spa_feat_enabled_txg_obj) != 0) 2615 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2616 } 2617 2618 spa->spa_is_initializing = B_TRUE; 2619 error = dsl_pool_open(spa->spa_dsl_pool); 2620 spa->spa_is_initializing = B_FALSE; 2621 if (error != 0) 2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2623 2624 if (!mosconfig) { 2625 uint64_t hostid; 2626 nvlist_t *policy = NULL, *nvconfig; 2627 2628 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2629 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2630 2631 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2632 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2633 char *hostname; 2634 unsigned long myhostid = 0; 2635 2636 VERIFY(nvlist_lookup_string(nvconfig, 2637 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2638 2639#ifdef _KERNEL 2640 myhostid = zone_get_hostid(NULL); 2641#else /* _KERNEL */ 2642 /* 2643 * We're emulating the system's hostid in userland, so 2644 * we can't use zone_get_hostid(). 2645 */ 2646 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2647#endif /* _KERNEL */ 2648 if (check_hostid && hostid != 0 && myhostid != 0 && 2649 hostid != myhostid) { 2650 nvlist_free(nvconfig); 2651 cmn_err(CE_WARN, "pool '%s' could not be " 2652 "loaded as it was last accessed by " 2653 "another system (host: %s hostid: 0x%lx). " 2654 "See: http://illumos.org/msg/ZFS-8000-EY", 2655 spa_name(spa), hostname, 2656 (unsigned long)hostid); 2657 return (SET_ERROR(EBADF)); 2658 } 2659 } 2660 if (nvlist_lookup_nvlist(spa->spa_config, 2661 ZPOOL_REWIND_POLICY, &policy) == 0) 2662 VERIFY(nvlist_add_nvlist(nvconfig, 2663 ZPOOL_REWIND_POLICY, policy) == 0); 2664 2665 spa_config_set(spa, nvconfig); 2666 spa_unload(spa); 2667 spa_deactivate(spa); 2668 spa_activate(spa, orig_mode); 2669 2670 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2671 } 2672 2673 /* Grab the secret checksum salt from the MOS. */ 2674 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2675 DMU_POOL_CHECKSUM_SALT, 1, 2676 sizeof (spa->spa_cksum_salt.zcs_bytes), 2677 spa->spa_cksum_salt.zcs_bytes); 2678 if (error == ENOENT) { 2679 /* Generate a new salt for subsequent use */ 2680 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 2681 sizeof (spa->spa_cksum_salt.zcs_bytes)); 2682 } else if (error != 0) { 2683 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2684 } 2685 2686 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2687 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2688 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2689 if (error != 0) 2690 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2691 2692 /* 2693 * Load the bit that tells us to use the new accounting function 2694 * (raid-z deflation). If we have an older pool, this will not 2695 * be present. 2696 */ 2697 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2698 if (error != 0 && error != ENOENT) 2699 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2700 2701 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2702 &spa->spa_creation_version); 2703 if (error != 0 && error != ENOENT) 2704 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2705 2706 /* 2707 * Load the persistent error log. If we have an older pool, this will 2708 * not be present. 2709 */ 2710 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2711 if (error != 0 && error != ENOENT) 2712 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2713 2714 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2715 &spa->spa_errlog_scrub); 2716 if (error != 0 && error != ENOENT) 2717 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2718 2719 /* 2720 * Load the history object. If we have an older pool, this 2721 * will not be present. 2722 */ 2723 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2724 if (error != 0 && error != ENOENT) 2725 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2726 2727 /* 2728 * Load the per-vdev ZAP map. If we have an older pool, this will not 2729 * be present; in this case, defer its creation to a later time to 2730 * avoid dirtying the MOS this early / out of sync context. See 2731 * spa_sync_config_object. 2732 */ 2733 2734 /* The sentinel is only available in the MOS config. */ 2735 nvlist_t *mos_config; 2736 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) 2737 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2738 2739 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP, 2740 &spa->spa_all_vdev_zaps); 2741 2742 if (error == ENOENT) { 2743 VERIFY(!nvlist_exists(mos_config, 2744 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 2745 spa->spa_avz_action = AVZ_ACTION_INITIALIZE; 2746 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 2747 } else if (error != 0) { 2748 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2749 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) { 2750 /* 2751 * An older version of ZFS overwrote the sentinel value, so 2752 * we have orphaned per-vdev ZAPs in the MOS. Defer their 2753 * destruction to later; see spa_sync_config_object. 2754 */ 2755 spa->spa_avz_action = AVZ_ACTION_DESTROY; 2756 /* 2757 * We're assuming that no vdevs have had their ZAPs created 2758 * before this. Better be sure of it. 2759 */ 2760 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev)); 2761 } 2762 nvlist_free(mos_config); 2763 2764 /* 2765 * If we're assembling the pool from the split-off vdevs of 2766 * an existing pool, we don't want to attach the spares & cache 2767 * devices. 2768 */ 2769 2770 /* 2771 * Load any hot spares for this pool. 2772 */ 2773 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2774 if (error != 0 && error != ENOENT) 2775 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2776 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2777 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2778 if (load_nvlist(spa, spa->spa_spares.sav_object, 2779 &spa->spa_spares.sav_config) != 0) 2780 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2781 2782 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2783 spa_load_spares(spa); 2784 spa_config_exit(spa, SCL_ALL, FTAG); 2785 } else if (error == 0) { 2786 spa->spa_spares.sav_sync = B_TRUE; 2787 } 2788 2789 /* 2790 * Load any level 2 ARC devices for this pool. 2791 */ 2792 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2793 &spa->spa_l2cache.sav_object); 2794 if (error != 0 && error != ENOENT) 2795 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2796 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2797 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2798 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2799 &spa->spa_l2cache.sav_config) != 0) 2800 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2801 2802 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2803 spa_load_l2cache(spa); 2804 spa_config_exit(spa, SCL_ALL, FTAG); 2805 } else if (error == 0) { 2806 spa->spa_l2cache.sav_sync = B_TRUE; 2807 } 2808 2809 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2810 2811 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2812 if (error && error != ENOENT) 2813 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2814 2815 if (error == 0) { 2816 uint64_t autoreplace; 2817 2818 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2819 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2820 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2821 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2822 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2823 spa_prop_find(spa, ZPOOL_PROP_BOOTSIZE, &spa->spa_bootsize); 2824 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2825 &spa->spa_dedup_ditto); 2826 2827 spa->spa_autoreplace = (autoreplace != 0); 2828 } 2829 2830 /* 2831 * If the 'autoreplace' property is set, then post a resource notifying 2832 * the ZFS DE that it should not issue any faults for unopenable 2833 * devices. We also iterate over the vdevs, and post a sysevent for any 2834 * unopenable vdevs so that the normal autoreplace handler can take 2835 * over. 2836 */ 2837 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2838 spa_check_removed(spa->spa_root_vdev); 2839 /* 2840 * For the import case, this is done in spa_import(), because 2841 * at this point we're using the spare definitions from 2842 * the MOS config, not necessarily from the userland config. 2843 */ 2844 if (state != SPA_LOAD_IMPORT) { 2845 spa_aux_check_removed(&spa->spa_spares); 2846 spa_aux_check_removed(&spa->spa_l2cache); 2847 } 2848 } 2849 2850 /* 2851 * Load the vdev state for all toplevel vdevs. 2852 */ 2853 vdev_load(rvd); 2854 2855 /* 2856 * Propagate the leaf DTLs we just loaded all the way up the tree. 2857 */ 2858 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2859 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2860 spa_config_exit(spa, SCL_ALL, FTAG); 2861 2862 /* 2863 * Load the DDTs (dedup tables). 2864 */ 2865 error = ddt_load(spa); 2866 if (error != 0) 2867 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2868 2869 spa_update_dspace(spa); 2870 2871 /* 2872 * Validate the config, using the MOS config to fill in any 2873 * information which might be missing. If we fail to validate 2874 * the config then declare the pool unfit for use. If we're 2875 * assembling a pool from a split, the log is not transferred 2876 * over. 2877 */ 2878 if (type != SPA_IMPORT_ASSEMBLE) { 2879 nvlist_t *nvconfig; 2880 2881 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2882 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2883 2884 if (!spa_config_valid(spa, nvconfig)) { 2885 nvlist_free(nvconfig); 2886 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2887 ENXIO)); 2888 } 2889 nvlist_free(nvconfig); 2890 2891 /* 2892 * Now that we've validated the config, check the state of the 2893 * root vdev. If it can't be opened, it indicates one or 2894 * more toplevel vdevs are faulted. 2895 */ 2896 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2897 return (SET_ERROR(ENXIO)); 2898 2899 if (spa_writeable(spa) && spa_check_logs(spa)) { 2900 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2901 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2902 } 2903 } 2904 2905 if (missing_feat_write) { 2906 ASSERT(state == SPA_LOAD_TRYIMPORT); 2907 2908 /* 2909 * At this point, we know that we can open the pool in 2910 * read-only mode but not read-write mode. We now have enough 2911 * information and can return to userland. 2912 */ 2913 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2914 } 2915 2916 /* 2917 * We've successfully opened the pool, verify that we're ready 2918 * to start pushing transactions. 2919 */ 2920 if (state != SPA_LOAD_TRYIMPORT) { 2921 if (error = spa_load_verify(spa)) 2922 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2923 error)); 2924 } 2925 2926 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2927 spa->spa_load_max_txg == UINT64_MAX)) { 2928 dmu_tx_t *tx; 2929 int need_update = B_FALSE; 2930 dsl_pool_t *dp = spa_get_dsl(spa); 2931 2932 ASSERT(state != SPA_LOAD_TRYIMPORT); 2933 2934 /* 2935 * Claim log blocks that haven't been committed yet. 2936 * This must all happen in a single txg. 2937 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2938 * invoked from zil_claim_log_block()'s i/o done callback. 2939 * Price of rollback is that we abandon the log. 2940 */ 2941 spa->spa_claiming = B_TRUE; 2942 2943 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2944 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2945 zil_claim, tx, DS_FIND_CHILDREN); 2946 dmu_tx_commit(tx); 2947 2948 spa->spa_claiming = B_FALSE; 2949 2950 spa_set_log_state(spa, SPA_LOG_GOOD); 2951 spa->spa_sync_on = B_TRUE; 2952 txg_sync_start(spa->spa_dsl_pool); 2953 2954 /* 2955 * Wait for all claims to sync. We sync up to the highest 2956 * claimed log block birth time so that claimed log blocks 2957 * don't appear to be from the future. spa_claim_max_txg 2958 * will have been set for us by either zil_check_log_chain() 2959 * (invoked from spa_check_logs()) or zil_claim() above. 2960 */ 2961 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2962 2963 /* 2964 * If the config cache is stale, or we have uninitialized 2965 * metaslabs (see spa_vdev_add()), then update the config. 2966 * 2967 * If this is a verbatim import, trust the current 2968 * in-core spa_config and update the disk labels. 2969 */ 2970 if (config_cache_txg != spa->spa_config_txg || 2971 state == SPA_LOAD_IMPORT || 2972 state == SPA_LOAD_RECOVER || 2973 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2974 need_update = B_TRUE; 2975 2976 for (int c = 0; c < rvd->vdev_children; c++) 2977 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2978 need_update = B_TRUE; 2979 2980 /* 2981 * Update the config cache asychronously in case we're the 2982 * root pool, in which case the config cache isn't writable yet. 2983 */ 2984 if (need_update) 2985 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2986 2987 /* 2988 * Check all DTLs to see if anything needs resilvering. 2989 */ 2990 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2991 vdev_resilver_needed(rvd, NULL, NULL)) 2992 spa_async_request(spa, SPA_ASYNC_RESILVER); 2993 2994 /* 2995 * Log the fact that we booted up (so that we can detect if 2996 * we rebooted in the middle of an operation). 2997 */ 2998 spa_history_log_version(spa, "open"); 2999 3000 /* 3001 * Delete any inconsistent datasets. 3002 */ 3003 (void) dmu_objset_find(spa_name(spa), 3004 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 3005 3006 /* 3007 * Clean up any stale temporary dataset userrefs. 3008 */ 3009 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 3010 } 3011 3012 return (0); 3013} 3014 3015static int 3016spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 3017{ 3018 int mode = spa->spa_mode; 3019 3020 spa_unload(spa); 3021 spa_deactivate(spa); 3022 3023 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 3024 3025 spa_activate(spa, mode); 3026 spa_async_suspend(spa); 3027 3028 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 3029} 3030 3031/* 3032 * If spa_load() fails this function will try loading prior txg's. If 3033 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 3034 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 3035 * function will not rewind the pool and will return the same error as 3036 * spa_load(). 3037 */ 3038static int 3039spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 3040 uint64_t max_request, int rewind_flags) 3041{ 3042 nvlist_t *loadinfo = NULL; 3043 nvlist_t *config = NULL; 3044 int load_error, rewind_error; 3045 uint64_t safe_rewind_txg; 3046 uint64_t min_txg; 3047 3048 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 3049 spa->spa_load_max_txg = spa->spa_load_txg; 3050 spa_set_log_state(spa, SPA_LOG_CLEAR); 3051 } else { 3052 spa->spa_load_max_txg = max_request; 3053 if (max_request != UINT64_MAX) 3054 spa->spa_extreme_rewind = B_TRUE; 3055 } 3056 3057 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 3058 mosconfig); 3059 if (load_error == 0) 3060 return (0); 3061 3062 if (spa->spa_root_vdev != NULL) 3063 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3064 3065 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 3066 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 3067 3068 if (rewind_flags & ZPOOL_NEVER_REWIND) { 3069 nvlist_free(config); 3070 return (load_error); 3071 } 3072 3073 if (state == SPA_LOAD_RECOVER) { 3074 /* Price of rolling back is discarding txgs, including log */ 3075 spa_set_log_state(spa, SPA_LOG_CLEAR); 3076 } else { 3077 /* 3078 * If we aren't rolling back save the load info from our first 3079 * import attempt so that we can restore it after attempting 3080 * to rewind. 3081 */ 3082 loadinfo = spa->spa_load_info; 3083 spa->spa_load_info = fnvlist_alloc(); 3084 } 3085 3086 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 3087 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 3088 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 3089 TXG_INITIAL : safe_rewind_txg; 3090 3091 /* 3092 * Continue as long as we're finding errors, we're still within 3093 * the acceptable rewind range, and we're still finding uberblocks 3094 */ 3095 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 3096 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 3097 if (spa->spa_load_max_txg < safe_rewind_txg) 3098 spa->spa_extreme_rewind = B_TRUE; 3099 rewind_error = spa_load_retry(spa, state, mosconfig); 3100 } 3101 3102 spa->spa_extreme_rewind = B_FALSE; 3103 spa->spa_load_max_txg = UINT64_MAX; 3104 3105 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 3106 spa_config_set(spa, config); 3107 else 3108 nvlist_free(config); 3109 3110 if (state == SPA_LOAD_RECOVER) { 3111 ASSERT3P(loadinfo, ==, NULL); 3112 return (rewind_error); 3113 } else { 3114 /* Store the rewind info as part of the initial load info */ 3115 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3116 spa->spa_load_info); 3117 3118 /* Restore the initial load info */ 3119 fnvlist_free(spa->spa_load_info); 3120 spa->spa_load_info = loadinfo; 3121 3122 return (load_error); 3123 } 3124} 3125 3126/* 3127 * Pool Open/Import 3128 * 3129 * The import case is identical to an open except that the configuration is sent 3130 * down from userland, instead of grabbed from the configuration cache. For the 3131 * case of an open, the pool configuration will exist in the 3132 * POOL_STATE_UNINITIALIZED state. 3133 * 3134 * The stats information (gen/count/ustats) is used to gather vdev statistics at 3135 * the same time open the pool, without having to keep around the spa_t in some 3136 * ambiguous state. 3137 */ 3138static int 3139spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3140 nvlist_t **config) 3141{ 3142 spa_t *spa; 3143 spa_load_state_t state = SPA_LOAD_OPEN; 3144 int error; 3145 int locked = B_FALSE; 3146 int firstopen = B_FALSE; 3147 3148 *spapp = NULL; 3149 3150 /* 3151 * As disgusting as this is, we need to support recursive calls to this 3152 * function because dsl_dir_open() is called during spa_load(), and ends 3153 * up calling spa_open() again. The real fix is to figure out how to 3154 * avoid dsl_dir_open() calling this in the first place. 3155 */ 3156 if (mutex_owner(&spa_namespace_lock) != curthread) { 3157 mutex_enter(&spa_namespace_lock); 3158 locked = B_TRUE; 3159 } 3160 3161 if ((spa = spa_lookup(pool)) == NULL) { 3162 if (locked) 3163 mutex_exit(&spa_namespace_lock); 3164 return (SET_ERROR(ENOENT)); 3165 } 3166 3167 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3168 zpool_rewind_policy_t policy; 3169 3170 firstopen = B_TRUE; 3171 3172 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3173 &policy); 3174 if (policy.zrp_request & ZPOOL_DO_REWIND) 3175 state = SPA_LOAD_RECOVER; 3176 3177 spa_activate(spa, spa_mode_global); 3178 3179 if (state != SPA_LOAD_RECOVER) 3180 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3181 3182 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3183 policy.zrp_request); 3184 3185 if (error == EBADF) { 3186 /* 3187 * If vdev_validate() returns failure (indicated by 3188 * EBADF), it indicates that one of the vdevs indicates 3189 * that the pool has been exported or destroyed. If 3190 * this is the case, the config cache is out of sync and 3191 * we should remove the pool from the namespace. 3192 */ 3193 spa_unload(spa); 3194 spa_deactivate(spa); 3195 spa_config_sync(spa, B_TRUE, B_TRUE); 3196 spa_remove(spa); 3197 if (locked) 3198 mutex_exit(&spa_namespace_lock); 3199 return (SET_ERROR(ENOENT)); 3200 } 3201 3202 if (error) { 3203 /* 3204 * We can't open the pool, but we still have useful 3205 * information: the state of each vdev after the 3206 * attempted vdev_open(). Return this to the user. 3207 */ 3208 if (config != NULL && spa->spa_config) { 3209 VERIFY(nvlist_dup(spa->spa_config, config, 3210 KM_SLEEP) == 0); 3211 VERIFY(nvlist_add_nvlist(*config, 3212 ZPOOL_CONFIG_LOAD_INFO, 3213 spa->spa_load_info) == 0); 3214 } 3215 spa_unload(spa); 3216 spa_deactivate(spa); 3217 spa->spa_last_open_failed = error; 3218 if (locked) 3219 mutex_exit(&spa_namespace_lock); 3220 *spapp = NULL; 3221 return (error); 3222 } 3223 } 3224 3225 spa_open_ref(spa, tag); 3226 3227 if (config != NULL) 3228 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3229 3230 /* 3231 * If we've recovered the pool, pass back any information we 3232 * gathered while doing the load. 3233 */ 3234 if (state == SPA_LOAD_RECOVER) { 3235 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3236 spa->spa_load_info) == 0); 3237 } 3238 3239 if (locked) { 3240 spa->spa_last_open_failed = 0; 3241 spa->spa_last_ubsync_txg = 0; 3242 spa->spa_load_txg = 0; 3243 mutex_exit(&spa_namespace_lock); 3244#ifdef __FreeBSD__ 3245#ifdef _KERNEL 3246 if (firstopen) 3247 zvol_create_minors(spa->spa_name); 3248#endif 3249#endif 3250 } 3251 3252 *spapp = spa; 3253 3254 return (0); 3255} 3256 3257int 3258spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3259 nvlist_t **config) 3260{ 3261 return (spa_open_common(name, spapp, tag, policy, config)); 3262} 3263 3264int 3265spa_open(const char *name, spa_t **spapp, void *tag) 3266{ 3267 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3268} 3269 3270/* 3271 * Lookup the given spa_t, incrementing the inject count in the process, 3272 * preventing it from being exported or destroyed. 3273 */ 3274spa_t * 3275spa_inject_addref(char *name) 3276{ 3277 spa_t *spa; 3278 3279 mutex_enter(&spa_namespace_lock); 3280 if ((spa = spa_lookup(name)) == NULL) { 3281 mutex_exit(&spa_namespace_lock); 3282 return (NULL); 3283 } 3284 spa->spa_inject_ref++; 3285 mutex_exit(&spa_namespace_lock); 3286 3287 return (spa); 3288} 3289 3290void 3291spa_inject_delref(spa_t *spa) 3292{ 3293 mutex_enter(&spa_namespace_lock); 3294 spa->spa_inject_ref--; 3295 mutex_exit(&spa_namespace_lock); 3296} 3297 3298/* 3299 * Add spares device information to the nvlist. 3300 */ 3301static void 3302spa_add_spares(spa_t *spa, nvlist_t *config) 3303{ 3304 nvlist_t **spares; 3305 uint_t i, nspares; 3306 nvlist_t *nvroot; 3307 uint64_t guid; 3308 vdev_stat_t *vs; 3309 uint_t vsc; 3310 uint64_t pool; 3311 3312 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3313 3314 if (spa->spa_spares.sav_count == 0) 3315 return; 3316 3317 VERIFY(nvlist_lookup_nvlist(config, 3318 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3319 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3320 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3321 if (nspares != 0) { 3322 VERIFY(nvlist_add_nvlist_array(nvroot, 3323 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3324 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3325 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3326 3327 /* 3328 * Go through and find any spares which have since been 3329 * repurposed as an active spare. If this is the case, update 3330 * their status appropriately. 3331 */ 3332 for (i = 0; i < nspares; i++) { 3333 VERIFY(nvlist_lookup_uint64(spares[i], 3334 ZPOOL_CONFIG_GUID, &guid) == 0); 3335 if (spa_spare_exists(guid, &pool, NULL) && 3336 pool != 0ULL) { 3337 VERIFY(nvlist_lookup_uint64_array( 3338 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3339 (uint64_t **)&vs, &vsc) == 0); 3340 vs->vs_state = VDEV_STATE_CANT_OPEN; 3341 vs->vs_aux = VDEV_AUX_SPARED; 3342 } 3343 } 3344 } 3345} 3346 3347/* 3348 * Add l2cache device information to the nvlist, including vdev stats. 3349 */ 3350static void 3351spa_add_l2cache(spa_t *spa, nvlist_t *config) 3352{ 3353 nvlist_t **l2cache; 3354 uint_t i, j, nl2cache; 3355 nvlist_t *nvroot; 3356 uint64_t guid; 3357 vdev_t *vd; 3358 vdev_stat_t *vs; 3359 uint_t vsc; 3360 3361 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3362 3363 if (spa->spa_l2cache.sav_count == 0) 3364 return; 3365 3366 VERIFY(nvlist_lookup_nvlist(config, 3367 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3368 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3369 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3370 if (nl2cache != 0) { 3371 VERIFY(nvlist_add_nvlist_array(nvroot, 3372 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3373 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3374 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3375 3376 /* 3377 * Update level 2 cache device stats. 3378 */ 3379 3380 for (i = 0; i < nl2cache; i++) { 3381 VERIFY(nvlist_lookup_uint64(l2cache[i], 3382 ZPOOL_CONFIG_GUID, &guid) == 0); 3383 3384 vd = NULL; 3385 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3386 if (guid == 3387 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3388 vd = spa->spa_l2cache.sav_vdevs[j]; 3389 break; 3390 } 3391 } 3392 ASSERT(vd != NULL); 3393 3394 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3395 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3396 == 0); 3397 vdev_get_stats(vd, vs); 3398 } 3399 } 3400} 3401 3402static void 3403spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3404{ 3405 nvlist_t *features; 3406 zap_cursor_t zc; 3407 zap_attribute_t za; 3408 3409 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3410 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3411 3412 /* We may be unable to read features if pool is suspended. */ 3413 if (spa_suspended(spa)) 3414 goto out; 3415 3416 if (spa->spa_feat_for_read_obj != 0) { 3417 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3418 spa->spa_feat_for_read_obj); 3419 zap_cursor_retrieve(&zc, &za) == 0; 3420 zap_cursor_advance(&zc)) { 3421 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3422 za.za_num_integers == 1); 3423 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3424 za.za_first_integer)); 3425 } 3426 zap_cursor_fini(&zc); 3427 } 3428 3429 if (spa->spa_feat_for_write_obj != 0) { 3430 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3431 spa->spa_feat_for_write_obj); 3432 zap_cursor_retrieve(&zc, &za) == 0; 3433 zap_cursor_advance(&zc)) { 3434 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3435 za.za_num_integers == 1); 3436 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3437 za.za_first_integer)); 3438 } 3439 zap_cursor_fini(&zc); 3440 } 3441 3442out: 3443 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3444 features) == 0); 3445 nvlist_free(features); 3446} 3447 3448int 3449spa_get_stats(const char *name, nvlist_t **config, 3450 char *altroot, size_t buflen) 3451{ 3452 int error; 3453 spa_t *spa; 3454 3455 *config = NULL; 3456 error = spa_open_common(name, &spa, FTAG, NULL, config); 3457 3458 if (spa != NULL) { 3459 /* 3460 * This still leaves a window of inconsistency where the spares 3461 * or l2cache devices could change and the config would be 3462 * self-inconsistent. 3463 */ 3464 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3465 3466 if (*config != NULL) { 3467 uint64_t loadtimes[2]; 3468 3469 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3470 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3471 VERIFY(nvlist_add_uint64_array(*config, 3472 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3473 3474 VERIFY(nvlist_add_uint64(*config, 3475 ZPOOL_CONFIG_ERRCOUNT, 3476 spa_get_errlog_size(spa)) == 0); 3477 3478 if (spa_suspended(spa)) 3479 VERIFY(nvlist_add_uint64(*config, 3480 ZPOOL_CONFIG_SUSPENDED, 3481 spa->spa_failmode) == 0); 3482 3483 spa_add_spares(spa, *config); 3484 spa_add_l2cache(spa, *config); 3485 spa_add_feature_stats(spa, *config); 3486 } 3487 } 3488 3489 /* 3490 * We want to get the alternate root even for faulted pools, so we cheat 3491 * and call spa_lookup() directly. 3492 */ 3493 if (altroot) { 3494 if (spa == NULL) { 3495 mutex_enter(&spa_namespace_lock); 3496 spa = spa_lookup(name); 3497 if (spa) 3498 spa_altroot(spa, altroot, buflen); 3499 else 3500 altroot[0] = '\0'; 3501 spa = NULL; 3502 mutex_exit(&spa_namespace_lock); 3503 } else { 3504 spa_altroot(spa, altroot, buflen); 3505 } 3506 } 3507 3508 if (spa != NULL) { 3509 spa_config_exit(spa, SCL_CONFIG, FTAG); 3510 spa_close(spa, FTAG); 3511 } 3512 3513 return (error); 3514} 3515 3516/* 3517 * Validate that the auxiliary device array is well formed. We must have an 3518 * array of nvlists, each which describes a valid leaf vdev. If this is an 3519 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3520 * specified, as long as they are well-formed. 3521 */ 3522static int 3523spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3524 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3525 vdev_labeltype_t label) 3526{ 3527 nvlist_t **dev; 3528 uint_t i, ndev; 3529 vdev_t *vd; 3530 int error; 3531 3532 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3533 3534 /* 3535 * It's acceptable to have no devs specified. 3536 */ 3537 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3538 return (0); 3539 3540 if (ndev == 0) 3541 return (SET_ERROR(EINVAL)); 3542 3543 /* 3544 * Make sure the pool is formatted with a version that supports this 3545 * device type. 3546 */ 3547 if (spa_version(spa) < version) 3548 return (SET_ERROR(ENOTSUP)); 3549 3550 /* 3551 * Set the pending device list so we correctly handle device in-use 3552 * checking. 3553 */ 3554 sav->sav_pending = dev; 3555 sav->sav_npending = ndev; 3556 3557 for (i = 0; i < ndev; i++) { 3558 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3559 mode)) != 0) 3560 goto out; 3561 3562 if (!vd->vdev_ops->vdev_op_leaf) { 3563 vdev_free(vd); 3564 error = SET_ERROR(EINVAL); 3565 goto out; 3566 } 3567 3568 /* 3569 * The L2ARC currently only supports disk devices in 3570 * kernel context. For user-level testing, we allow it. 3571 */ 3572#ifdef _KERNEL 3573 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3574 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3575 error = SET_ERROR(ENOTBLK); 3576 vdev_free(vd); 3577 goto out; 3578 } 3579#endif 3580 vd->vdev_top = vd; 3581 3582 if ((error = vdev_open(vd)) == 0 && 3583 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3584 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3585 vd->vdev_guid) == 0); 3586 } 3587 3588 vdev_free(vd); 3589 3590 if (error && 3591 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3592 goto out; 3593 else 3594 error = 0; 3595 } 3596 3597out: 3598 sav->sav_pending = NULL; 3599 sav->sav_npending = 0; 3600 return (error); 3601} 3602 3603static int 3604spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3605{ 3606 int error; 3607 3608 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3609 3610 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3611 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3612 VDEV_LABEL_SPARE)) != 0) { 3613 return (error); 3614 } 3615 3616 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3617 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3618 VDEV_LABEL_L2CACHE)); 3619} 3620 3621static void 3622spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3623 const char *config) 3624{ 3625 int i; 3626 3627 if (sav->sav_config != NULL) { 3628 nvlist_t **olddevs; 3629 uint_t oldndevs; 3630 nvlist_t **newdevs; 3631 3632 /* 3633 * Generate new dev list by concatentating with the 3634 * current dev list. 3635 */ 3636 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3637 &olddevs, &oldndevs) == 0); 3638 3639 newdevs = kmem_alloc(sizeof (void *) * 3640 (ndevs + oldndevs), KM_SLEEP); 3641 for (i = 0; i < oldndevs; i++) 3642 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3643 KM_SLEEP) == 0); 3644 for (i = 0; i < ndevs; i++) 3645 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3646 KM_SLEEP) == 0); 3647 3648 VERIFY(nvlist_remove(sav->sav_config, config, 3649 DATA_TYPE_NVLIST_ARRAY) == 0); 3650 3651 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3652 config, newdevs, ndevs + oldndevs) == 0); 3653 for (i = 0; i < oldndevs + ndevs; i++) 3654 nvlist_free(newdevs[i]); 3655 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3656 } else { 3657 /* 3658 * Generate a new dev list. 3659 */ 3660 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3661 KM_SLEEP) == 0); 3662 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3663 devs, ndevs) == 0); 3664 } 3665} 3666 3667/* 3668 * Stop and drop level 2 ARC devices 3669 */ 3670void 3671spa_l2cache_drop(spa_t *spa) 3672{ 3673 vdev_t *vd; 3674 int i; 3675 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3676 3677 for (i = 0; i < sav->sav_count; i++) { 3678 uint64_t pool; 3679 3680 vd = sav->sav_vdevs[i]; 3681 ASSERT(vd != NULL); 3682 3683 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3684 pool != 0ULL && l2arc_vdev_present(vd)) 3685 l2arc_remove_vdev(vd); 3686 } 3687} 3688 3689/* 3690 * Pool Creation 3691 */ 3692int 3693spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3694 nvlist_t *zplprops) 3695{ 3696 spa_t *spa; 3697 char *altroot = NULL; 3698 vdev_t *rvd; 3699 dsl_pool_t *dp; 3700 dmu_tx_t *tx; 3701 int error = 0; 3702 uint64_t txg = TXG_INITIAL; 3703 nvlist_t **spares, **l2cache; 3704 uint_t nspares, nl2cache; 3705 uint64_t version, obj; 3706 boolean_t has_features; 3707 3708 /* 3709 * If this pool already exists, return failure. 3710 */ 3711 mutex_enter(&spa_namespace_lock); 3712 if (spa_lookup(pool) != NULL) { 3713 mutex_exit(&spa_namespace_lock); 3714 return (SET_ERROR(EEXIST)); 3715 } 3716 3717 /* 3718 * Allocate a new spa_t structure. 3719 */ 3720 (void) nvlist_lookup_string(props, 3721 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3722 spa = spa_add(pool, NULL, altroot); 3723 spa_activate(spa, spa_mode_global); 3724 3725 if (props && (error = spa_prop_validate(spa, props))) { 3726 spa_deactivate(spa); 3727 spa_remove(spa); 3728 mutex_exit(&spa_namespace_lock); 3729 return (error); 3730 } 3731 3732 has_features = B_FALSE; 3733 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3734 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3735 if (zpool_prop_feature(nvpair_name(elem))) 3736 has_features = B_TRUE; 3737 } 3738 3739 if (has_features || nvlist_lookup_uint64(props, 3740 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3741 version = SPA_VERSION; 3742 } 3743 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3744 3745 spa->spa_first_txg = txg; 3746 spa->spa_uberblock.ub_txg = txg - 1; 3747 spa->spa_uberblock.ub_version = version; 3748 spa->spa_ubsync = spa->spa_uberblock; 3749 spa->spa_load_state = SPA_LOAD_CREATE; 3750 3751 /* 3752 * Create "The Godfather" zio to hold all async IOs 3753 */ 3754 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3755 KM_SLEEP); 3756 for (int i = 0; i < max_ncpus; i++) { 3757 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3758 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3759 ZIO_FLAG_GODFATHER); 3760 } 3761 3762 /* 3763 * Create the root vdev. 3764 */ 3765 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3766 3767 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3768 3769 ASSERT(error != 0 || rvd != NULL); 3770 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3771 3772 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3773 error = SET_ERROR(EINVAL); 3774 3775 if (error == 0 && 3776 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3777 (error = spa_validate_aux(spa, nvroot, txg, 3778 VDEV_ALLOC_ADD)) == 0) { 3779 for (int c = 0; c < rvd->vdev_children; c++) { 3780 vdev_ashift_optimize(rvd->vdev_child[c]); 3781 vdev_metaslab_set_size(rvd->vdev_child[c]); 3782 vdev_expand(rvd->vdev_child[c], txg); 3783 } 3784 } 3785 3786 spa_config_exit(spa, SCL_ALL, FTAG); 3787 3788 if (error != 0) { 3789 spa_unload(spa); 3790 spa_deactivate(spa); 3791 spa_remove(spa); 3792 mutex_exit(&spa_namespace_lock); 3793 return (error); 3794 } 3795 3796 /* 3797 * Get the list of spares, if specified. 3798 */ 3799 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3800 &spares, &nspares) == 0) { 3801 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3802 KM_SLEEP) == 0); 3803 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3804 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3805 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3806 spa_load_spares(spa); 3807 spa_config_exit(spa, SCL_ALL, FTAG); 3808 spa->spa_spares.sav_sync = B_TRUE; 3809 } 3810 3811 /* 3812 * Get the list of level 2 cache devices, if specified. 3813 */ 3814 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3815 &l2cache, &nl2cache) == 0) { 3816 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3817 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3818 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3819 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3820 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3821 spa_load_l2cache(spa); 3822 spa_config_exit(spa, SCL_ALL, FTAG); 3823 spa->spa_l2cache.sav_sync = B_TRUE; 3824 } 3825 3826 spa->spa_is_initializing = B_TRUE; 3827 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3828 spa->spa_meta_objset = dp->dp_meta_objset; 3829 spa->spa_is_initializing = B_FALSE; 3830 3831 /* 3832 * Create DDTs (dedup tables). 3833 */ 3834 ddt_create(spa); 3835 3836 spa_update_dspace(spa); 3837 3838 tx = dmu_tx_create_assigned(dp, txg); 3839 3840 /* 3841 * Create the pool config object. 3842 */ 3843 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3844 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3845 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3846 3847 if (zap_add(spa->spa_meta_objset, 3848 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3849 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3850 cmn_err(CE_PANIC, "failed to add pool config"); 3851 } 3852 3853 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3854 spa_feature_create_zap_objects(spa, tx); 3855 3856 if (zap_add(spa->spa_meta_objset, 3857 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3858 sizeof (uint64_t), 1, &version, tx) != 0) { 3859 cmn_err(CE_PANIC, "failed to add pool version"); 3860 } 3861 3862 /* Newly created pools with the right version are always deflated. */ 3863 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3864 spa->spa_deflate = TRUE; 3865 if (zap_add(spa->spa_meta_objset, 3866 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3867 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3868 cmn_err(CE_PANIC, "failed to add deflate"); 3869 } 3870 } 3871 3872 /* 3873 * Create the deferred-free bpobj. Turn off compression 3874 * because sync-to-convergence takes longer if the blocksize 3875 * keeps changing. 3876 */ 3877 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3878 dmu_object_set_compress(spa->spa_meta_objset, obj, 3879 ZIO_COMPRESS_OFF, tx); 3880 if (zap_add(spa->spa_meta_objset, 3881 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3882 sizeof (uint64_t), 1, &obj, tx) != 0) { 3883 cmn_err(CE_PANIC, "failed to add bpobj"); 3884 } 3885 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3886 spa->spa_meta_objset, obj)); 3887 3888 /* 3889 * Create the pool's history object. 3890 */ 3891 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3892 spa_history_create_obj(spa, tx); 3893 3894 /* 3895 * Generate some random noise for salted checksums to operate on. 3896 */ 3897 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3898 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3899 3900 /* 3901 * Set pool properties. 3902 */ 3903 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3904 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3905 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3906 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3907 3908 if (props != NULL) { 3909 spa_configfile_set(spa, props, B_FALSE); 3910 spa_sync_props(props, tx); 3911 } 3912 3913 dmu_tx_commit(tx); 3914 3915 spa->spa_sync_on = B_TRUE; 3916 txg_sync_start(spa->spa_dsl_pool); 3917 3918 /* 3919 * We explicitly wait for the first transaction to complete so that our 3920 * bean counters are appropriately updated. 3921 */ 3922 txg_wait_synced(spa->spa_dsl_pool, txg); 3923 3924 spa_config_sync(spa, B_FALSE, B_TRUE); 3925 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE); 3926 3927 spa_history_log_version(spa, "create"); 3928 3929 /* 3930 * Don't count references from objsets that are already closed 3931 * and are making their way through the eviction process. 3932 */ 3933 spa_evicting_os_wait(spa); 3934 spa->spa_minref = refcount_count(&spa->spa_refcount); 3935 spa->spa_load_state = SPA_LOAD_NONE; 3936 3937 mutex_exit(&spa_namespace_lock); 3938 3939 return (0); 3940} 3941 3942#ifdef _KERNEL 3943#ifdef illumos 3944/* 3945 * Get the root pool information from the root disk, then import the root pool 3946 * during the system boot up time. 3947 */ 3948extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3949 3950static nvlist_t * 3951spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3952{ 3953 nvlist_t *config; 3954 nvlist_t *nvtop, *nvroot; 3955 uint64_t pgid; 3956 3957 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3958 return (NULL); 3959 3960 /* 3961 * Add this top-level vdev to the child array. 3962 */ 3963 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3964 &nvtop) == 0); 3965 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3966 &pgid) == 0); 3967 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3968 3969 /* 3970 * Put this pool's top-level vdevs into a root vdev. 3971 */ 3972 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3973 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3974 VDEV_TYPE_ROOT) == 0); 3975 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3976 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3977 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3978 &nvtop, 1) == 0); 3979 3980 /* 3981 * Replace the existing vdev_tree with the new root vdev in 3982 * this pool's configuration (remove the old, add the new). 3983 */ 3984 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3985 nvlist_free(nvroot); 3986 return (config); 3987} 3988 3989/* 3990 * Walk the vdev tree and see if we can find a device with "better" 3991 * configuration. A configuration is "better" if the label on that 3992 * device has a more recent txg. 3993 */ 3994static void 3995spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3996{ 3997 for (int c = 0; c < vd->vdev_children; c++) 3998 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3999 4000 if (vd->vdev_ops->vdev_op_leaf) { 4001 nvlist_t *label; 4002 uint64_t label_txg; 4003 4004 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 4005 &label) != 0) 4006 return; 4007 4008 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 4009 &label_txg) == 0); 4010 4011 /* 4012 * Do we have a better boot device? 4013 */ 4014 if (label_txg > *txg) { 4015 *txg = label_txg; 4016 *avd = vd; 4017 } 4018 nvlist_free(label); 4019 } 4020} 4021 4022/* 4023 * Import a root pool. 4024 * 4025 * For x86. devpath_list will consist of devid and/or physpath name of 4026 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 4027 * The GRUB "findroot" command will return the vdev we should boot. 4028 * 4029 * For Sparc, devpath_list consists the physpath name of the booting device 4030 * no matter the rootpool is a single device pool or a mirrored pool. 4031 * e.g. 4032 * "/pci@1f,0/ide@d/disk@0,0:a" 4033 */ 4034int 4035spa_import_rootpool(char *devpath, char *devid) 4036{ 4037 spa_t *spa; 4038 vdev_t *rvd, *bvd, *avd = NULL; 4039 nvlist_t *config, *nvtop; 4040 uint64_t guid, txg; 4041 char *pname; 4042 int error; 4043 4044 /* 4045 * Read the label from the boot device and generate a configuration. 4046 */ 4047 config = spa_generate_rootconf(devpath, devid, &guid); 4048#if defined(_OBP) && defined(_KERNEL) 4049 if (config == NULL) { 4050 if (strstr(devpath, "/iscsi/ssd") != NULL) { 4051 /* iscsi boot */ 4052 get_iscsi_bootpath_phy(devpath); 4053 config = spa_generate_rootconf(devpath, devid, &guid); 4054 } 4055 } 4056#endif 4057 if (config == NULL) { 4058 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 4059 devpath); 4060 return (SET_ERROR(EIO)); 4061 } 4062 4063 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4064 &pname) == 0); 4065 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 4066 4067 mutex_enter(&spa_namespace_lock); 4068 if ((spa = spa_lookup(pname)) != NULL) { 4069 /* 4070 * Remove the existing root pool from the namespace so that we 4071 * can replace it with the correct config we just read in. 4072 */ 4073 spa_remove(spa); 4074 } 4075 4076 spa = spa_add(pname, config, NULL); 4077 spa->spa_is_root = B_TRUE; 4078 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4079 4080 /* 4081 * Build up a vdev tree based on the boot device's label config. 4082 */ 4083 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4084 &nvtop) == 0); 4085 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4086 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4087 VDEV_ALLOC_ROOTPOOL); 4088 spa_config_exit(spa, SCL_ALL, FTAG); 4089 if (error) { 4090 mutex_exit(&spa_namespace_lock); 4091 nvlist_free(config); 4092 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4093 pname); 4094 return (error); 4095 } 4096 4097 /* 4098 * Get the boot vdev. 4099 */ 4100 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 4101 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 4102 (u_longlong_t)guid); 4103 error = SET_ERROR(ENOENT); 4104 goto out; 4105 } 4106 4107 /* 4108 * Determine if there is a better boot device. 4109 */ 4110 avd = bvd; 4111 spa_alt_rootvdev(rvd, &avd, &txg); 4112 if (avd != bvd) { 4113 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 4114 "try booting from '%s'", avd->vdev_path); 4115 error = SET_ERROR(EINVAL); 4116 goto out; 4117 } 4118 4119 /* 4120 * If the boot device is part of a spare vdev then ensure that 4121 * we're booting off the active spare. 4122 */ 4123 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 4124 !bvd->vdev_isspare) { 4125 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 4126 "try booting from '%s'", 4127 bvd->vdev_parent-> 4128 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4129 error = SET_ERROR(EINVAL); 4130 goto out; 4131 } 4132 4133 error = 0; 4134out: 4135 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4136 vdev_free(rvd); 4137 spa_config_exit(spa, SCL_ALL, FTAG); 4138 mutex_exit(&spa_namespace_lock); 4139 4140 nvlist_free(config); 4141 return (error); 4142} 4143 4144#else /* !illumos */ 4145 4146extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 4147 uint64_t *count); 4148 4149static nvlist_t * 4150spa_generate_rootconf(const char *name) 4151{ 4152 nvlist_t **configs, **tops; 4153 nvlist_t *config; 4154 nvlist_t *best_cfg, *nvtop, *nvroot; 4155 uint64_t *holes; 4156 uint64_t best_txg; 4157 uint64_t nchildren; 4158 uint64_t pgid; 4159 uint64_t count; 4160 uint64_t i; 4161 uint_t nholes; 4162 4163 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4164 return (NULL); 4165 4166 ASSERT3U(count, !=, 0); 4167 best_txg = 0; 4168 for (i = 0; i < count; i++) { 4169 uint64_t txg; 4170 4171 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4172 &txg) == 0); 4173 if (txg > best_txg) { 4174 best_txg = txg; 4175 best_cfg = configs[i]; 4176 } 4177 } 4178 4179 nchildren = 1; 4180 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4181 holes = NULL; 4182 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4183 &holes, &nholes); 4184 4185 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4186 for (i = 0; i < nchildren; i++) { 4187 if (i >= count) 4188 break; 4189 if (configs[i] == NULL) 4190 continue; 4191 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4192 &nvtop) == 0); 4193 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4194 } 4195 for (i = 0; holes != NULL && i < nholes; i++) { 4196 if (i >= nchildren) 4197 continue; 4198 if (tops[holes[i]] != NULL) 4199 continue; 4200 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4201 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4202 VDEV_TYPE_HOLE) == 0); 4203 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4204 holes[i]) == 0); 4205 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4206 0) == 0); 4207 } 4208 for (i = 0; i < nchildren; i++) { 4209 if (tops[i] != NULL) 4210 continue; 4211 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4212 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4213 VDEV_TYPE_MISSING) == 0); 4214 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4215 i) == 0); 4216 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4217 0) == 0); 4218 } 4219 4220 /* 4221 * Create pool config based on the best vdev config. 4222 */ 4223 nvlist_dup(best_cfg, &config, KM_SLEEP); 4224 4225 /* 4226 * Put this pool's top-level vdevs into a root vdev. 4227 */ 4228 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4229 &pgid) == 0); 4230 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4231 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4232 VDEV_TYPE_ROOT) == 0); 4233 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4234 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4235 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4236 tops, nchildren) == 0); 4237 4238 /* 4239 * Replace the existing vdev_tree with the new root vdev in 4240 * this pool's configuration (remove the old, add the new). 4241 */ 4242 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4243 4244 /* 4245 * Drop vdev config elements that should not be present at pool level. 4246 */ 4247 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4248 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4249 4250 for (i = 0; i < count; i++) 4251 nvlist_free(configs[i]); 4252 kmem_free(configs, count * sizeof(void *)); 4253 for (i = 0; i < nchildren; i++) 4254 nvlist_free(tops[i]); 4255 kmem_free(tops, nchildren * sizeof(void *)); 4256 nvlist_free(nvroot); 4257 return (config); 4258} 4259 4260int 4261spa_import_rootpool(const char *name) 4262{ 4263 spa_t *spa; 4264 vdev_t *rvd, *bvd, *avd = NULL; 4265 nvlist_t *config, *nvtop; 4266 uint64_t txg; 4267 char *pname; 4268 int error; 4269 4270 /* 4271 * Read the label from the boot device and generate a configuration. 4272 */ 4273 config = spa_generate_rootconf(name); 4274 4275 mutex_enter(&spa_namespace_lock); 4276 if (config != NULL) { 4277 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4278 &pname) == 0 && strcmp(name, pname) == 0); 4279 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4280 == 0); 4281 4282 if ((spa = spa_lookup(pname)) != NULL) { 4283 /* 4284 * The pool could already be imported, 4285 * e.g., after reboot -r. 4286 */ 4287 if (spa->spa_state == POOL_STATE_ACTIVE) { 4288 mutex_exit(&spa_namespace_lock); 4289 nvlist_free(config); 4290 return (0); 4291 } 4292 4293 /* 4294 * Remove the existing root pool from the namespace so 4295 * that we can replace it with the correct config 4296 * we just read in. 4297 */ 4298 spa_remove(spa); 4299 } 4300 spa = spa_add(pname, config, NULL); 4301 4302 /* 4303 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4304 * via spa_version(). 4305 */ 4306 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4307 &spa->spa_ubsync.ub_version) != 0) 4308 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4309 } else if ((spa = spa_lookup(name)) == NULL) { 4310 mutex_exit(&spa_namespace_lock); 4311 nvlist_free(config); 4312 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4313 name); 4314 return (EIO); 4315 } else { 4316 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4317 } 4318 spa->spa_is_root = B_TRUE; 4319 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4320 4321 /* 4322 * Build up a vdev tree based on the boot device's label config. 4323 */ 4324 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4325 &nvtop) == 0); 4326 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4327 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4328 VDEV_ALLOC_ROOTPOOL); 4329 spa_config_exit(spa, SCL_ALL, FTAG); 4330 if (error) { 4331 mutex_exit(&spa_namespace_lock); 4332 nvlist_free(config); 4333 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4334 pname); 4335 return (error); 4336 } 4337 4338 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4339 vdev_free(rvd); 4340 spa_config_exit(spa, SCL_ALL, FTAG); 4341 mutex_exit(&spa_namespace_lock); 4342 4343 nvlist_free(config); 4344 return (0); 4345} 4346 4347#endif /* illumos */ 4348#endif /* _KERNEL */ 4349 4350/* 4351 * Import a non-root pool into the system. 4352 */ 4353int 4354spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4355{ 4356 spa_t *spa; 4357 char *altroot = NULL; 4358 spa_load_state_t state = SPA_LOAD_IMPORT; 4359 zpool_rewind_policy_t policy; 4360 uint64_t mode = spa_mode_global; 4361 uint64_t readonly = B_FALSE; 4362 int error; 4363 nvlist_t *nvroot; 4364 nvlist_t **spares, **l2cache; 4365 uint_t nspares, nl2cache; 4366 4367 /* 4368 * If a pool with this name exists, return failure. 4369 */ 4370 mutex_enter(&spa_namespace_lock); 4371 if (spa_lookup(pool) != NULL) { 4372 mutex_exit(&spa_namespace_lock); 4373 return (SET_ERROR(EEXIST)); 4374 } 4375 4376 /* 4377 * Create and initialize the spa structure. 4378 */ 4379 (void) nvlist_lookup_string(props, 4380 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4381 (void) nvlist_lookup_uint64(props, 4382 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4383 if (readonly) 4384 mode = FREAD; 4385 spa = spa_add(pool, config, altroot); 4386 spa->spa_import_flags = flags; 4387 4388 /* 4389 * Verbatim import - Take a pool and insert it into the namespace 4390 * as if it had been loaded at boot. 4391 */ 4392 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4393 if (props != NULL) 4394 spa_configfile_set(spa, props, B_FALSE); 4395 4396 spa_config_sync(spa, B_FALSE, B_TRUE); 4397 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 4398 4399 mutex_exit(&spa_namespace_lock); 4400 return (0); 4401 } 4402 4403 spa_activate(spa, mode); 4404 4405 /* 4406 * Don't start async tasks until we know everything is healthy. 4407 */ 4408 spa_async_suspend(spa); 4409 4410 zpool_get_rewind_policy(config, &policy); 4411 if (policy.zrp_request & ZPOOL_DO_REWIND) 4412 state = SPA_LOAD_RECOVER; 4413 4414 /* 4415 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4416 * because the user-supplied config is actually the one to trust when 4417 * doing an import. 4418 */ 4419 if (state != SPA_LOAD_RECOVER) 4420 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4421 4422 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4423 policy.zrp_request); 4424 4425 /* 4426 * Propagate anything learned while loading the pool and pass it 4427 * back to caller (i.e. rewind info, missing devices, etc). 4428 */ 4429 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4430 spa->spa_load_info) == 0); 4431 4432 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4433 /* 4434 * Toss any existing sparelist, as it doesn't have any validity 4435 * anymore, and conflicts with spa_has_spare(). 4436 */ 4437 if (spa->spa_spares.sav_config) { 4438 nvlist_free(spa->spa_spares.sav_config); 4439 spa->spa_spares.sav_config = NULL; 4440 spa_load_spares(spa); 4441 } 4442 if (spa->spa_l2cache.sav_config) { 4443 nvlist_free(spa->spa_l2cache.sav_config); 4444 spa->spa_l2cache.sav_config = NULL; 4445 spa_load_l2cache(spa); 4446 } 4447 4448 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4449 &nvroot) == 0); 4450 if (error == 0) 4451 error = spa_validate_aux(spa, nvroot, -1ULL, 4452 VDEV_ALLOC_SPARE); 4453 if (error == 0) 4454 error = spa_validate_aux(spa, nvroot, -1ULL, 4455 VDEV_ALLOC_L2CACHE); 4456 spa_config_exit(spa, SCL_ALL, FTAG); 4457 4458 if (props != NULL) 4459 spa_configfile_set(spa, props, B_FALSE); 4460 4461 if (error != 0 || (props && spa_writeable(spa) && 4462 (error = spa_prop_set(spa, props)))) { 4463 spa_unload(spa); 4464 spa_deactivate(spa); 4465 spa_remove(spa); 4466 mutex_exit(&spa_namespace_lock); 4467 return (error); 4468 } 4469 4470 spa_async_resume(spa); 4471 4472 /* 4473 * Override any spares and level 2 cache devices as specified by 4474 * the user, as these may have correct device names/devids, etc. 4475 */ 4476 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4477 &spares, &nspares) == 0) { 4478 if (spa->spa_spares.sav_config) 4479 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4480 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4481 else 4482 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4483 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4484 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4485 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4486 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4487 spa_load_spares(spa); 4488 spa_config_exit(spa, SCL_ALL, FTAG); 4489 spa->spa_spares.sav_sync = B_TRUE; 4490 } 4491 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4492 &l2cache, &nl2cache) == 0) { 4493 if (spa->spa_l2cache.sav_config) 4494 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4495 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4496 else 4497 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4498 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4499 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4500 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4501 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4502 spa_load_l2cache(spa); 4503 spa_config_exit(spa, SCL_ALL, FTAG); 4504 spa->spa_l2cache.sav_sync = B_TRUE; 4505 } 4506 4507 /* 4508 * Check for any removed devices. 4509 */ 4510 if (spa->spa_autoreplace) { 4511 spa_aux_check_removed(&spa->spa_spares); 4512 spa_aux_check_removed(&spa->spa_l2cache); 4513 } 4514 4515 if (spa_writeable(spa)) { 4516 /* 4517 * Update the config cache to include the newly-imported pool. 4518 */ 4519 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4520 } 4521 4522 /* 4523 * It's possible that the pool was expanded while it was exported. 4524 * We kick off an async task to handle this for us. 4525 */ 4526 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4527 4528 spa_history_log_version(spa, "import"); 4529 4530 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT); 4531 4532 mutex_exit(&spa_namespace_lock); 4533 4534#ifdef __FreeBSD__ 4535#ifdef _KERNEL 4536 zvol_create_minors(pool); 4537#endif 4538#endif 4539 return (0); 4540} 4541 4542nvlist_t * 4543spa_tryimport(nvlist_t *tryconfig) 4544{ 4545 nvlist_t *config = NULL; 4546 char *poolname; 4547 spa_t *spa; 4548 uint64_t state; 4549 int error; 4550 4551 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4552 return (NULL); 4553 4554 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4555 return (NULL); 4556 4557 /* 4558 * Create and initialize the spa structure. 4559 */ 4560 mutex_enter(&spa_namespace_lock); 4561 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4562 spa_activate(spa, FREAD); 4563 4564 /* 4565 * Pass off the heavy lifting to spa_load(). 4566 * Pass TRUE for mosconfig because the user-supplied config 4567 * is actually the one to trust when doing an import. 4568 */ 4569 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4570 4571 /* 4572 * If 'tryconfig' was at least parsable, return the current config. 4573 */ 4574 if (spa->spa_root_vdev != NULL) { 4575 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4576 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4577 poolname) == 0); 4578 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4579 state) == 0); 4580 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4581 spa->spa_uberblock.ub_timestamp) == 0); 4582 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4583 spa->spa_load_info) == 0); 4584 4585 /* 4586 * If the bootfs property exists on this pool then we 4587 * copy it out so that external consumers can tell which 4588 * pools are bootable. 4589 */ 4590 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4591 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4592 4593 /* 4594 * We have to play games with the name since the 4595 * pool was opened as TRYIMPORT_NAME. 4596 */ 4597 if (dsl_dsobj_to_dsname(spa_name(spa), 4598 spa->spa_bootfs, tmpname) == 0) { 4599 char *cp; 4600 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4601 4602 cp = strchr(tmpname, '/'); 4603 if (cp == NULL) { 4604 (void) strlcpy(dsname, tmpname, 4605 MAXPATHLEN); 4606 } else { 4607 (void) snprintf(dsname, MAXPATHLEN, 4608 "%s/%s", poolname, ++cp); 4609 } 4610 VERIFY(nvlist_add_string(config, 4611 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4612 kmem_free(dsname, MAXPATHLEN); 4613 } 4614 kmem_free(tmpname, MAXPATHLEN); 4615 } 4616 4617 /* 4618 * Add the list of hot spares and level 2 cache devices. 4619 */ 4620 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4621 spa_add_spares(spa, config); 4622 spa_add_l2cache(spa, config); 4623 spa_config_exit(spa, SCL_CONFIG, FTAG); 4624 } 4625 4626 spa_unload(spa); 4627 spa_deactivate(spa); 4628 spa_remove(spa); 4629 mutex_exit(&spa_namespace_lock); 4630 4631 return (config); 4632} 4633 4634/* 4635 * Pool export/destroy 4636 * 4637 * The act of destroying or exporting a pool is very simple. We make sure there 4638 * is no more pending I/O and any references to the pool are gone. Then, we 4639 * update the pool state and sync all the labels to disk, removing the 4640 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4641 * we don't sync the labels or remove the configuration cache. 4642 */ 4643static int 4644spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4645 boolean_t force, boolean_t hardforce) 4646{ 4647 spa_t *spa; 4648 4649 if (oldconfig) 4650 *oldconfig = NULL; 4651 4652 if (!(spa_mode_global & FWRITE)) 4653 return (SET_ERROR(EROFS)); 4654 4655 mutex_enter(&spa_namespace_lock); 4656 if ((spa = spa_lookup(pool)) == NULL) { 4657 mutex_exit(&spa_namespace_lock); 4658 return (SET_ERROR(ENOENT)); 4659 } 4660 4661 /* 4662 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4663 * reacquire the namespace lock, and see if we can export. 4664 */ 4665 spa_open_ref(spa, FTAG); 4666 mutex_exit(&spa_namespace_lock); 4667 spa_async_suspend(spa); 4668 mutex_enter(&spa_namespace_lock); 4669 spa_close(spa, FTAG); 4670 4671 /* 4672 * The pool will be in core if it's openable, 4673 * in which case we can modify its state. 4674 */ 4675 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4676 /* 4677 * Objsets may be open only because they're dirty, so we 4678 * have to force it to sync before checking spa_refcnt. 4679 */ 4680 txg_wait_synced(spa->spa_dsl_pool, 0); 4681 spa_evicting_os_wait(spa); 4682 4683 /* 4684 * A pool cannot be exported or destroyed if there are active 4685 * references. If we are resetting a pool, allow references by 4686 * fault injection handlers. 4687 */ 4688 if (!spa_refcount_zero(spa) || 4689 (spa->spa_inject_ref != 0 && 4690 new_state != POOL_STATE_UNINITIALIZED)) { 4691 spa_async_resume(spa); 4692 mutex_exit(&spa_namespace_lock); 4693 return (SET_ERROR(EBUSY)); 4694 } 4695 4696 /* 4697 * A pool cannot be exported if it has an active shared spare. 4698 * This is to prevent other pools stealing the active spare 4699 * from an exported pool. At user's own will, such pool can 4700 * be forcedly exported. 4701 */ 4702 if (!force && new_state == POOL_STATE_EXPORTED && 4703 spa_has_active_shared_spare(spa)) { 4704 spa_async_resume(spa); 4705 mutex_exit(&spa_namespace_lock); 4706 return (SET_ERROR(EXDEV)); 4707 } 4708 4709 /* 4710 * We want this to be reflected on every label, 4711 * so mark them all dirty. spa_unload() will do the 4712 * final sync that pushes these changes out. 4713 */ 4714 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4715 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4716 spa->spa_state = new_state; 4717 spa->spa_final_txg = spa_last_synced_txg(spa) + 4718 TXG_DEFER_SIZE + 1; 4719 vdev_config_dirty(spa->spa_root_vdev); 4720 spa_config_exit(spa, SCL_ALL, FTAG); 4721 } 4722 } 4723 4724 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY); 4725 4726 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4727 spa_unload(spa); 4728 spa_deactivate(spa); 4729 } 4730 4731 if (oldconfig && spa->spa_config) 4732 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4733 4734 if (new_state != POOL_STATE_UNINITIALIZED) { 4735 if (!hardforce) 4736 spa_config_sync(spa, B_TRUE, B_TRUE); 4737 spa_remove(spa); 4738 } 4739 mutex_exit(&spa_namespace_lock); 4740 4741 return (0); 4742} 4743 4744/* 4745 * Destroy a storage pool. 4746 */ 4747int 4748spa_destroy(char *pool) 4749{ 4750 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4751 B_FALSE, B_FALSE)); 4752} 4753 4754/* 4755 * Export a storage pool. 4756 */ 4757int 4758spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4759 boolean_t hardforce) 4760{ 4761 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4762 force, hardforce)); 4763} 4764 4765/* 4766 * Similar to spa_export(), this unloads the spa_t without actually removing it 4767 * from the namespace in any way. 4768 */ 4769int 4770spa_reset(char *pool) 4771{ 4772 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4773 B_FALSE, B_FALSE)); 4774} 4775 4776/* 4777 * ========================================================================== 4778 * Device manipulation 4779 * ========================================================================== 4780 */ 4781 4782/* 4783 * Add a device to a storage pool. 4784 */ 4785int 4786spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4787{ 4788 uint64_t txg, id; 4789 int error; 4790 vdev_t *rvd = spa->spa_root_vdev; 4791 vdev_t *vd, *tvd; 4792 nvlist_t **spares, **l2cache; 4793 uint_t nspares, nl2cache; 4794 4795 ASSERT(spa_writeable(spa)); 4796 4797 txg = spa_vdev_enter(spa); 4798 4799 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4800 VDEV_ALLOC_ADD)) != 0) 4801 return (spa_vdev_exit(spa, NULL, txg, error)); 4802 4803 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4804 4805 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4806 &nspares) != 0) 4807 nspares = 0; 4808 4809 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4810 &nl2cache) != 0) 4811 nl2cache = 0; 4812 4813 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4814 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4815 4816 if (vd->vdev_children != 0 && 4817 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4818 return (spa_vdev_exit(spa, vd, txg, error)); 4819 4820 /* 4821 * We must validate the spares and l2cache devices after checking the 4822 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4823 */ 4824 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4825 return (spa_vdev_exit(spa, vd, txg, error)); 4826 4827 /* 4828 * Transfer each new top-level vdev from vd to rvd. 4829 */ 4830 for (int c = 0; c < vd->vdev_children; c++) { 4831 4832 /* 4833 * Set the vdev id to the first hole, if one exists. 4834 */ 4835 for (id = 0; id < rvd->vdev_children; id++) { 4836 if (rvd->vdev_child[id]->vdev_ishole) { 4837 vdev_free(rvd->vdev_child[id]); 4838 break; 4839 } 4840 } 4841 tvd = vd->vdev_child[c]; 4842 vdev_remove_child(vd, tvd); 4843 tvd->vdev_id = id; 4844 vdev_add_child(rvd, tvd); 4845 vdev_config_dirty(tvd); 4846 } 4847 4848 if (nspares != 0) { 4849 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4850 ZPOOL_CONFIG_SPARES); 4851 spa_load_spares(spa); 4852 spa->spa_spares.sav_sync = B_TRUE; 4853 } 4854 4855 if (nl2cache != 0) { 4856 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4857 ZPOOL_CONFIG_L2CACHE); 4858 spa_load_l2cache(spa); 4859 spa->spa_l2cache.sav_sync = B_TRUE; 4860 } 4861 4862 /* 4863 * We have to be careful when adding new vdevs to an existing pool. 4864 * If other threads start allocating from these vdevs before we 4865 * sync the config cache, and we lose power, then upon reboot we may 4866 * fail to open the pool because there are DVAs that the config cache 4867 * can't translate. Therefore, we first add the vdevs without 4868 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4869 * and then let spa_config_update() initialize the new metaslabs. 4870 * 4871 * spa_load() checks for added-but-not-initialized vdevs, so that 4872 * if we lose power at any point in this sequence, the remaining 4873 * steps will be completed the next time we load the pool. 4874 */ 4875 (void) spa_vdev_exit(spa, vd, txg, 0); 4876 4877 mutex_enter(&spa_namespace_lock); 4878 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4879 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD); 4880 mutex_exit(&spa_namespace_lock); 4881 4882 return (0); 4883} 4884 4885/* 4886 * Attach a device to a mirror. The arguments are the path to any device 4887 * in the mirror, and the nvroot for the new device. If the path specifies 4888 * a device that is not mirrored, we automatically insert the mirror vdev. 4889 * 4890 * If 'replacing' is specified, the new device is intended to replace the 4891 * existing device; in this case the two devices are made into their own 4892 * mirror using the 'replacing' vdev, which is functionally identical to 4893 * the mirror vdev (it actually reuses all the same ops) but has a few 4894 * extra rules: you can't attach to it after it's been created, and upon 4895 * completion of resilvering, the first disk (the one being replaced) 4896 * is automatically detached. 4897 */ 4898int 4899spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4900{ 4901 uint64_t txg, dtl_max_txg; 4902 vdev_t *rvd = spa->spa_root_vdev; 4903 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4904 vdev_ops_t *pvops; 4905 char *oldvdpath, *newvdpath; 4906 int newvd_isspare; 4907 int error; 4908 4909 ASSERT(spa_writeable(spa)); 4910 4911 txg = spa_vdev_enter(spa); 4912 4913 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4914 4915 if (oldvd == NULL) 4916 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4917 4918 if (!oldvd->vdev_ops->vdev_op_leaf) 4919 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4920 4921 pvd = oldvd->vdev_parent; 4922 4923 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4924 VDEV_ALLOC_ATTACH)) != 0) 4925 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4926 4927 if (newrootvd->vdev_children != 1) 4928 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4929 4930 newvd = newrootvd->vdev_child[0]; 4931 4932 if (!newvd->vdev_ops->vdev_op_leaf) 4933 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4934 4935 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4936 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4937 4938 /* 4939 * Spares can't replace logs 4940 */ 4941 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4942 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4943 4944 if (!replacing) { 4945 /* 4946 * For attach, the only allowable parent is a mirror or the root 4947 * vdev. 4948 */ 4949 if (pvd->vdev_ops != &vdev_mirror_ops && 4950 pvd->vdev_ops != &vdev_root_ops) 4951 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4952 4953 pvops = &vdev_mirror_ops; 4954 } else { 4955 /* 4956 * Active hot spares can only be replaced by inactive hot 4957 * spares. 4958 */ 4959 if (pvd->vdev_ops == &vdev_spare_ops && 4960 oldvd->vdev_isspare && 4961 !spa_has_spare(spa, newvd->vdev_guid)) 4962 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4963 4964 /* 4965 * If the source is a hot spare, and the parent isn't already a 4966 * spare, then we want to create a new hot spare. Otherwise, we 4967 * want to create a replacing vdev. The user is not allowed to 4968 * attach to a spared vdev child unless the 'isspare' state is 4969 * the same (spare replaces spare, non-spare replaces 4970 * non-spare). 4971 */ 4972 if (pvd->vdev_ops == &vdev_replacing_ops && 4973 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4974 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4975 } else if (pvd->vdev_ops == &vdev_spare_ops && 4976 newvd->vdev_isspare != oldvd->vdev_isspare) { 4977 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4978 } 4979 4980 if (newvd->vdev_isspare) 4981 pvops = &vdev_spare_ops; 4982 else 4983 pvops = &vdev_replacing_ops; 4984 } 4985 4986 /* 4987 * Make sure the new device is big enough. 4988 */ 4989 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4990 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4991 4992 /* 4993 * The new device cannot have a higher alignment requirement 4994 * than the top-level vdev. 4995 */ 4996 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4997 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4998 4999 /* 5000 * If this is an in-place replacement, update oldvd's path and devid 5001 * to make it distinguishable from newvd, and unopenable from now on. 5002 */ 5003 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 5004 spa_strfree(oldvd->vdev_path); 5005 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 5006 KM_SLEEP); 5007 (void) sprintf(oldvd->vdev_path, "%s/%s", 5008 newvd->vdev_path, "old"); 5009 if (oldvd->vdev_devid != NULL) { 5010 spa_strfree(oldvd->vdev_devid); 5011 oldvd->vdev_devid = NULL; 5012 } 5013 } 5014 5015 /* mark the device being resilvered */ 5016 newvd->vdev_resilver_txg = txg; 5017 5018 /* 5019 * If the parent is not a mirror, or if we're replacing, insert the new 5020 * mirror/replacing/spare vdev above oldvd. 5021 */ 5022 if (pvd->vdev_ops != pvops) 5023 pvd = vdev_add_parent(oldvd, pvops); 5024 5025 ASSERT(pvd->vdev_top->vdev_parent == rvd); 5026 ASSERT(pvd->vdev_ops == pvops); 5027 ASSERT(oldvd->vdev_parent == pvd); 5028 5029 /* 5030 * Extract the new device from its root and add it to pvd. 5031 */ 5032 vdev_remove_child(newrootvd, newvd); 5033 newvd->vdev_id = pvd->vdev_children; 5034 newvd->vdev_crtxg = oldvd->vdev_crtxg; 5035 vdev_add_child(pvd, newvd); 5036 5037 tvd = newvd->vdev_top; 5038 ASSERT(pvd->vdev_top == tvd); 5039 ASSERT(tvd->vdev_parent == rvd); 5040 5041 vdev_config_dirty(tvd); 5042 5043 /* 5044 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 5045 * for any dmu_sync-ed blocks. It will propagate upward when 5046 * spa_vdev_exit() calls vdev_dtl_reassess(). 5047 */ 5048 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 5049 5050 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 5051 dtl_max_txg - TXG_INITIAL); 5052 5053 if (newvd->vdev_isspare) { 5054 spa_spare_activate(newvd); 5055 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE); 5056 } 5057 5058 oldvdpath = spa_strdup(oldvd->vdev_path); 5059 newvdpath = spa_strdup(newvd->vdev_path); 5060 newvd_isspare = newvd->vdev_isspare; 5061 5062 /* 5063 * Mark newvd's DTL dirty in this txg. 5064 */ 5065 vdev_dirty(tvd, VDD_DTL, newvd, txg); 5066 5067 /* 5068 * Schedule the resilver to restart in the future. We do this to 5069 * ensure that dmu_sync-ed blocks have been stitched into the 5070 * respective datasets. 5071 */ 5072 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 5073 5074 if (spa->spa_bootfs) 5075 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH); 5076 5077 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH); 5078 5079 /* 5080 * Commit the config 5081 */ 5082 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 5083 5084 spa_history_log_internal(spa, "vdev attach", NULL, 5085 "%s vdev=%s %s vdev=%s", 5086 replacing && newvd_isspare ? "spare in" : 5087 replacing ? "replace" : "attach", newvdpath, 5088 replacing ? "for" : "to", oldvdpath); 5089 5090 spa_strfree(oldvdpath); 5091 spa_strfree(newvdpath); 5092 5093 return (0); 5094} 5095 5096/* 5097 * Detach a device from a mirror or replacing vdev. 5098 * 5099 * If 'replace_done' is specified, only detach if the parent 5100 * is a replacing vdev. 5101 */ 5102int 5103spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 5104{ 5105 uint64_t txg; 5106 int error; 5107 vdev_t *rvd = spa->spa_root_vdev; 5108 vdev_t *vd, *pvd, *cvd, *tvd; 5109 boolean_t unspare = B_FALSE; 5110 uint64_t unspare_guid = 0; 5111 char *vdpath; 5112 5113 ASSERT(spa_writeable(spa)); 5114 5115 txg = spa_vdev_enter(spa); 5116 5117 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5118 5119 if (vd == NULL) 5120 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5121 5122 if (!vd->vdev_ops->vdev_op_leaf) 5123 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5124 5125 pvd = vd->vdev_parent; 5126 5127 /* 5128 * If the parent/child relationship is not as expected, don't do it. 5129 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 5130 * vdev that's replacing B with C. The user's intent in replacing 5131 * is to go from M(A,B) to M(A,C). If the user decides to cancel 5132 * the replace by detaching C, the expected behavior is to end up 5133 * M(A,B). But suppose that right after deciding to detach C, 5134 * the replacement of B completes. We would have M(A,C), and then 5135 * ask to detach C, which would leave us with just A -- not what 5136 * the user wanted. To prevent this, we make sure that the 5137 * parent/child relationship hasn't changed -- in this example, 5138 * that C's parent is still the replacing vdev R. 5139 */ 5140 if (pvd->vdev_guid != pguid && pguid != 0) 5141 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5142 5143 /* 5144 * Only 'replacing' or 'spare' vdevs can be replaced. 5145 */ 5146 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5147 pvd->vdev_ops != &vdev_spare_ops) 5148 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5149 5150 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5151 spa_version(spa) >= SPA_VERSION_SPARES); 5152 5153 /* 5154 * Only mirror, replacing, and spare vdevs support detach. 5155 */ 5156 if (pvd->vdev_ops != &vdev_replacing_ops && 5157 pvd->vdev_ops != &vdev_mirror_ops && 5158 pvd->vdev_ops != &vdev_spare_ops) 5159 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5160 5161 /* 5162 * If this device has the only valid copy of some data, 5163 * we cannot safely detach it. 5164 */ 5165 if (vdev_dtl_required(vd)) 5166 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5167 5168 ASSERT(pvd->vdev_children >= 2); 5169 5170 /* 5171 * If we are detaching the second disk from a replacing vdev, then 5172 * check to see if we changed the original vdev's path to have "/old" 5173 * at the end in spa_vdev_attach(). If so, undo that change now. 5174 */ 5175 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5176 vd->vdev_path != NULL) { 5177 size_t len = strlen(vd->vdev_path); 5178 5179 for (int c = 0; c < pvd->vdev_children; c++) { 5180 cvd = pvd->vdev_child[c]; 5181 5182 if (cvd == vd || cvd->vdev_path == NULL) 5183 continue; 5184 5185 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5186 strcmp(cvd->vdev_path + len, "/old") == 0) { 5187 spa_strfree(cvd->vdev_path); 5188 cvd->vdev_path = spa_strdup(vd->vdev_path); 5189 break; 5190 } 5191 } 5192 } 5193 5194 /* 5195 * If we are detaching the original disk from a spare, then it implies 5196 * that the spare should become a real disk, and be removed from the 5197 * active spare list for the pool. 5198 */ 5199 if (pvd->vdev_ops == &vdev_spare_ops && 5200 vd->vdev_id == 0 && 5201 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5202 unspare = B_TRUE; 5203 5204 /* 5205 * Erase the disk labels so the disk can be used for other things. 5206 * This must be done after all other error cases are handled, 5207 * but before we disembowel vd (so we can still do I/O to it). 5208 * But if we can't do it, don't treat the error as fatal -- 5209 * it may be that the unwritability of the disk is the reason 5210 * it's being detached! 5211 */ 5212 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5213 5214 /* 5215 * Remove vd from its parent and compact the parent's children. 5216 */ 5217 vdev_remove_child(pvd, vd); 5218 vdev_compact_children(pvd); 5219 5220 /* 5221 * Remember one of the remaining children so we can get tvd below. 5222 */ 5223 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5224 5225 /* 5226 * If we need to remove the remaining child from the list of hot spares, 5227 * do it now, marking the vdev as no longer a spare in the process. 5228 * We must do this before vdev_remove_parent(), because that can 5229 * change the GUID if it creates a new toplevel GUID. For a similar 5230 * reason, we must remove the spare now, in the same txg as the detach; 5231 * otherwise someone could attach a new sibling, change the GUID, and 5232 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5233 */ 5234 if (unspare) { 5235 ASSERT(cvd->vdev_isspare); 5236 spa_spare_remove(cvd); 5237 unspare_guid = cvd->vdev_guid; 5238 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5239 cvd->vdev_unspare = B_TRUE; 5240 } 5241 5242 /* 5243 * If the parent mirror/replacing vdev only has one child, 5244 * the parent is no longer needed. Remove it from the tree. 5245 */ 5246 if (pvd->vdev_children == 1) { 5247 if (pvd->vdev_ops == &vdev_spare_ops) 5248 cvd->vdev_unspare = B_FALSE; 5249 vdev_remove_parent(cvd); 5250 } 5251 5252 5253 /* 5254 * We don't set tvd until now because the parent we just removed 5255 * may have been the previous top-level vdev. 5256 */ 5257 tvd = cvd->vdev_top; 5258 ASSERT(tvd->vdev_parent == rvd); 5259 5260 /* 5261 * Reevaluate the parent vdev state. 5262 */ 5263 vdev_propagate_state(cvd); 5264 5265 /* 5266 * If the 'autoexpand' property is set on the pool then automatically 5267 * try to expand the size of the pool. For example if the device we 5268 * just detached was smaller than the others, it may be possible to 5269 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5270 * first so that we can obtain the updated sizes of the leaf vdevs. 5271 */ 5272 if (spa->spa_autoexpand) { 5273 vdev_reopen(tvd); 5274 vdev_expand(tvd, txg); 5275 } 5276 5277 vdev_config_dirty(tvd); 5278 5279 /* 5280 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5281 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5282 * But first make sure we're not on any *other* txg's DTL list, to 5283 * prevent vd from being accessed after it's freed. 5284 */ 5285 vdpath = spa_strdup(vd->vdev_path); 5286 for (int t = 0; t < TXG_SIZE; t++) 5287 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5288 vd->vdev_detached = B_TRUE; 5289 vdev_dirty(tvd, VDD_DTL, vd, txg); 5290 5291 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE); 5292 5293 /* hang on to the spa before we release the lock */ 5294 spa_open_ref(spa, FTAG); 5295 5296 error = spa_vdev_exit(spa, vd, txg, 0); 5297 5298 spa_history_log_internal(spa, "detach", NULL, 5299 "vdev=%s", vdpath); 5300 spa_strfree(vdpath); 5301 5302 /* 5303 * If this was the removal of the original device in a hot spare vdev, 5304 * then we want to go through and remove the device from the hot spare 5305 * list of every other pool. 5306 */ 5307 if (unspare) { 5308 spa_t *altspa = NULL; 5309 5310 mutex_enter(&spa_namespace_lock); 5311 while ((altspa = spa_next(altspa)) != NULL) { 5312 if (altspa->spa_state != POOL_STATE_ACTIVE || 5313 altspa == spa) 5314 continue; 5315 5316 spa_open_ref(altspa, FTAG); 5317 mutex_exit(&spa_namespace_lock); 5318 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5319 mutex_enter(&spa_namespace_lock); 5320 spa_close(altspa, FTAG); 5321 } 5322 mutex_exit(&spa_namespace_lock); 5323 5324 /* search the rest of the vdevs for spares to remove */ 5325 spa_vdev_resilver_done(spa); 5326 } 5327 5328 /* all done with the spa; OK to release */ 5329 mutex_enter(&spa_namespace_lock); 5330 spa_close(spa, FTAG); 5331 mutex_exit(&spa_namespace_lock); 5332 5333 return (error); 5334} 5335 5336/* 5337 * Split a set of devices from their mirrors, and create a new pool from them. 5338 */ 5339int 5340spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5341 nvlist_t *props, boolean_t exp) 5342{ 5343 int error = 0; 5344 uint64_t txg, *glist; 5345 spa_t *newspa; 5346 uint_t c, children, lastlog; 5347 nvlist_t **child, *nvl, *tmp; 5348 dmu_tx_t *tx; 5349 char *altroot = NULL; 5350 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5351 boolean_t activate_slog; 5352 5353 ASSERT(spa_writeable(spa)); 5354 5355 txg = spa_vdev_enter(spa); 5356 5357 /* clear the log and flush everything up to now */ 5358 activate_slog = spa_passivate_log(spa); 5359 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5360 error = spa_offline_log(spa); 5361 txg = spa_vdev_config_enter(spa); 5362 5363 if (activate_slog) 5364 spa_activate_log(spa); 5365 5366 if (error != 0) 5367 return (spa_vdev_exit(spa, NULL, txg, error)); 5368 5369 /* check new spa name before going any further */ 5370 if (spa_lookup(newname) != NULL) 5371 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5372 5373 /* 5374 * scan through all the children to ensure they're all mirrors 5375 */ 5376 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5377 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5378 &children) != 0) 5379 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5380 5381 /* first, check to ensure we've got the right child count */ 5382 rvd = spa->spa_root_vdev; 5383 lastlog = 0; 5384 for (c = 0; c < rvd->vdev_children; c++) { 5385 vdev_t *vd = rvd->vdev_child[c]; 5386 5387 /* don't count the holes & logs as children */ 5388 if (vd->vdev_islog || vd->vdev_ishole) { 5389 if (lastlog == 0) 5390 lastlog = c; 5391 continue; 5392 } 5393 5394 lastlog = 0; 5395 } 5396 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5397 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5398 5399 /* next, ensure no spare or cache devices are part of the split */ 5400 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5401 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5402 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5403 5404 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5405 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5406 5407 /* then, loop over each vdev and validate it */ 5408 for (c = 0; c < children; c++) { 5409 uint64_t is_hole = 0; 5410 5411 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5412 &is_hole); 5413 5414 if (is_hole != 0) { 5415 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5416 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5417 continue; 5418 } else { 5419 error = SET_ERROR(EINVAL); 5420 break; 5421 } 5422 } 5423 5424 /* which disk is going to be split? */ 5425 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5426 &glist[c]) != 0) { 5427 error = SET_ERROR(EINVAL); 5428 break; 5429 } 5430 5431 /* look it up in the spa */ 5432 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5433 if (vml[c] == NULL) { 5434 error = SET_ERROR(ENODEV); 5435 break; 5436 } 5437 5438 /* make sure there's nothing stopping the split */ 5439 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5440 vml[c]->vdev_islog || 5441 vml[c]->vdev_ishole || 5442 vml[c]->vdev_isspare || 5443 vml[c]->vdev_isl2cache || 5444 !vdev_writeable(vml[c]) || 5445 vml[c]->vdev_children != 0 || 5446 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5447 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5448 error = SET_ERROR(EINVAL); 5449 break; 5450 } 5451 5452 if (vdev_dtl_required(vml[c])) { 5453 error = SET_ERROR(EBUSY); 5454 break; 5455 } 5456 5457 /* we need certain info from the top level */ 5458 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5459 vml[c]->vdev_top->vdev_ms_array) == 0); 5460 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5461 vml[c]->vdev_top->vdev_ms_shift) == 0); 5462 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5463 vml[c]->vdev_top->vdev_asize) == 0); 5464 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5465 vml[c]->vdev_top->vdev_ashift) == 0); 5466 5467 /* transfer per-vdev ZAPs */ 5468 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0); 5469 VERIFY0(nvlist_add_uint64(child[c], 5470 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap)); 5471 5472 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0); 5473 VERIFY0(nvlist_add_uint64(child[c], 5474 ZPOOL_CONFIG_VDEV_TOP_ZAP, 5475 vml[c]->vdev_parent->vdev_top_zap)); 5476 } 5477 5478 if (error != 0) { 5479 kmem_free(vml, children * sizeof (vdev_t *)); 5480 kmem_free(glist, children * sizeof (uint64_t)); 5481 return (spa_vdev_exit(spa, NULL, txg, error)); 5482 } 5483 5484 /* stop writers from using the disks */ 5485 for (c = 0; c < children; c++) { 5486 if (vml[c] != NULL) 5487 vml[c]->vdev_offline = B_TRUE; 5488 } 5489 vdev_reopen(spa->spa_root_vdev); 5490 5491 /* 5492 * Temporarily record the splitting vdevs in the spa config. This 5493 * will disappear once the config is regenerated. 5494 */ 5495 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5496 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5497 glist, children) == 0); 5498 kmem_free(glist, children * sizeof (uint64_t)); 5499 5500 mutex_enter(&spa->spa_props_lock); 5501 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5502 nvl) == 0); 5503 mutex_exit(&spa->spa_props_lock); 5504 spa->spa_config_splitting = nvl; 5505 vdev_config_dirty(spa->spa_root_vdev); 5506 5507 /* configure and create the new pool */ 5508 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5509 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5510 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5511 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5512 spa_version(spa)) == 0); 5513 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5514 spa->spa_config_txg) == 0); 5515 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5516 spa_generate_guid(NULL)) == 0); 5517 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)); 5518 (void) nvlist_lookup_string(props, 5519 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5520 5521 /* add the new pool to the namespace */ 5522 newspa = spa_add(newname, config, altroot); 5523 newspa->spa_avz_action = AVZ_ACTION_REBUILD; 5524 newspa->spa_config_txg = spa->spa_config_txg; 5525 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5526 5527 /* release the spa config lock, retaining the namespace lock */ 5528 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5529 5530 if (zio_injection_enabled) 5531 zio_handle_panic_injection(spa, FTAG, 1); 5532 5533 spa_activate(newspa, spa_mode_global); 5534 spa_async_suspend(newspa); 5535 5536#ifndef illumos 5537 /* mark that we are creating new spa by splitting */ 5538 newspa->spa_splitting_newspa = B_TRUE; 5539#endif 5540 /* create the new pool from the disks of the original pool */ 5541 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5542#ifndef illumos 5543 newspa->spa_splitting_newspa = B_FALSE; 5544#endif 5545 if (error) 5546 goto out; 5547 5548 /* if that worked, generate a real config for the new pool */ 5549 if (newspa->spa_root_vdev != NULL) { 5550 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5551 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5552 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5553 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5554 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5555 B_TRUE)); 5556 } 5557 5558 /* set the props */ 5559 if (props != NULL) { 5560 spa_configfile_set(newspa, props, B_FALSE); 5561 error = spa_prop_set(newspa, props); 5562 if (error) 5563 goto out; 5564 } 5565 5566 /* flush everything */ 5567 txg = spa_vdev_config_enter(newspa); 5568 vdev_config_dirty(newspa->spa_root_vdev); 5569 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5570 5571 if (zio_injection_enabled) 5572 zio_handle_panic_injection(spa, FTAG, 2); 5573 5574 spa_async_resume(newspa); 5575 5576 /* finally, update the original pool's config */ 5577 txg = spa_vdev_config_enter(spa); 5578 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5579 error = dmu_tx_assign(tx, TXG_WAIT); 5580 if (error != 0) 5581 dmu_tx_abort(tx); 5582 for (c = 0; c < children; c++) { 5583 if (vml[c] != NULL) { 5584 vdev_split(vml[c]); 5585 if (error == 0) 5586 spa_history_log_internal(spa, "detach", tx, 5587 "vdev=%s", vml[c]->vdev_path); 5588 5589 vdev_free(vml[c]); 5590 } 5591 } 5592 spa->spa_avz_action = AVZ_ACTION_REBUILD; 5593 vdev_config_dirty(spa->spa_root_vdev); 5594 spa->spa_config_splitting = NULL; 5595 nvlist_free(nvl); 5596 if (error == 0) 5597 dmu_tx_commit(tx); 5598 (void) spa_vdev_exit(spa, NULL, txg, 0); 5599 5600 if (zio_injection_enabled) 5601 zio_handle_panic_injection(spa, FTAG, 3); 5602 5603 /* split is complete; log a history record */ 5604 spa_history_log_internal(newspa, "split", NULL, 5605 "from pool %s", spa_name(spa)); 5606 5607 kmem_free(vml, children * sizeof (vdev_t *)); 5608 5609 /* if we're not going to mount the filesystems in userland, export */ 5610 if (exp) 5611 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5612 B_FALSE, B_FALSE); 5613 5614 return (error); 5615 5616out: 5617 spa_unload(newspa); 5618 spa_deactivate(newspa); 5619 spa_remove(newspa); 5620 5621 txg = spa_vdev_config_enter(spa); 5622 5623 /* re-online all offlined disks */ 5624 for (c = 0; c < children; c++) { 5625 if (vml[c] != NULL) 5626 vml[c]->vdev_offline = B_FALSE; 5627 } 5628 vdev_reopen(spa->spa_root_vdev); 5629 5630 nvlist_free(spa->spa_config_splitting); 5631 spa->spa_config_splitting = NULL; 5632 (void) spa_vdev_exit(spa, NULL, txg, error); 5633 5634 kmem_free(vml, children * sizeof (vdev_t *)); 5635 return (error); 5636} 5637 5638static nvlist_t * 5639spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5640{ 5641 for (int i = 0; i < count; i++) { 5642 uint64_t guid; 5643 5644 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5645 &guid) == 0); 5646 5647 if (guid == target_guid) 5648 return (nvpp[i]); 5649 } 5650 5651 return (NULL); 5652} 5653 5654static void 5655spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5656 nvlist_t *dev_to_remove) 5657{ 5658 nvlist_t **newdev = NULL; 5659 5660 if (count > 1) 5661 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5662 5663 for (int i = 0, j = 0; i < count; i++) { 5664 if (dev[i] == dev_to_remove) 5665 continue; 5666 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5667 } 5668 5669 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5670 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5671 5672 for (int i = 0; i < count - 1; i++) 5673 nvlist_free(newdev[i]); 5674 5675 if (count > 1) 5676 kmem_free(newdev, (count - 1) * sizeof (void *)); 5677} 5678 5679/* 5680 * Evacuate the device. 5681 */ 5682static int 5683spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5684{ 5685 uint64_t txg; 5686 int error = 0; 5687 5688 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5689 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5690 ASSERT(vd == vd->vdev_top); 5691 5692 /* 5693 * Evacuate the device. We don't hold the config lock as writer 5694 * since we need to do I/O but we do keep the 5695 * spa_namespace_lock held. Once this completes the device 5696 * should no longer have any blocks allocated on it. 5697 */ 5698 if (vd->vdev_islog) { 5699 if (vd->vdev_stat.vs_alloc != 0) 5700 error = spa_offline_log(spa); 5701 } else { 5702 error = SET_ERROR(ENOTSUP); 5703 } 5704 5705 if (error) 5706 return (error); 5707 5708 /* 5709 * The evacuation succeeded. Remove any remaining MOS metadata 5710 * associated with this vdev, and wait for these changes to sync. 5711 */ 5712 ASSERT0(vd->vdev_stat.vs_alloc); 5713 txg = spa_vdev_config_enter(spa); 5714 vd->vdev_removing = B_TRUE; 5715 vdev_dirty_leaves(vd, VDD_DTL, txg); 5716 vdev_config_dirty(vd); 5717 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5718 5719 return (0); 5720} 5721 5722/* 5723 * Complete the removal by cleaning up the namespace. 5724 */ 5725static void 5726spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5727{ 5728 vdev_t *rvd = spa->spa_root_vdev; 5729 uint64_t id = vd->vdev_id; 5730 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5731 5732 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5733 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5734 ASSERT(vd == vd->vdev_top); 5735 5736 /* 5737 * Only remove any devices which are empty. 5738 */ 5739 if (vd->vdev_stat.vs_alloc != 0) 5740 return; 5741 5742 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5743 5744 if (list_link_active(&vd->vdev_state_dirty_node)) 5745 vdev_state_clean(vd); 5746 if (list_link_active(&vd->vdev_config_dirty_node)) 5747 vdev_config_clean(vd); 5748 5749 vdev_free(vd); 5750 5751 if (last_vdev) { 5752 vdev_compact_children(rvd); 5753 } else { 5754 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5755 vdev_add_child(rvd, vd); 5756 } 5757 vdev_config_dirty(rvd); 5758 5759 /* 5760 * Reassess the health of our root vdev. 5761 */ 5762 vdev_reopen(rvd); 5763} 5764 5765/* 5766 * Remove a device from the pool - 5767 * 5768 * Removing a device from the vdev namespace requires several steps 5769 * and can take a significant amount of time. As a result we use 5770 * the spa_vdev_config_[enter/exit] functions which allow us to 5771 * grab and release the spa_config_lock while still holding the namespace 5772 * lock. During each step the configuration is synced out. 5773 * 5774 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5775 * devices. 5776 */ 5777int 5778spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5779{ 5780 vdev_t *vd; 5781 sysevent_t *ev = NULL; 5782 metaslab_group_t *mg; 5783 nvlist_t **spares, **l2cache, *nv; 5784 uint64_t txg = 0; 5785 uint_t nspares, nl2cache; 5786 int error = 0; 5787 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5788 5789 ASSERT(spa_writeable(spa)); 5790 5791 if (!locked) 5792 txg = spa_vdev_enter(spa); 5793 5794 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5795 5796 if (spa->spa_spares.sav_vdevs != NULL && 5797 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5798 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5799 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5800 /* 5801 * Only remove the hot spare if it's not currently in use 5802 * in this pool. 5803 */ 5804 if (vd == NULL || unspare) { 5805 if (vd == NULL) 5806 vd = spa_lookup_by_guid(spa, guid, B_TRUE); 5807 ev = spa_event_create(spa, vd, NULL, 5808 ESC_ZFS_VDEV_REMOVE_AUX); 5809 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5810 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5811 spa_load_spares(spa); 5812 spa->spa_spares.sav_sync = B_TRUE; 5813 } else { 5814 error = SET_ERROR(EBUSY); 5815 } 5816 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5817 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5818 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5819 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5820 /* 5821 * Cache devices can always be removed. 5822 */ 5823 vd = spa_lookup_by_guid(spa, guid, B_TRUE); 5824 ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_AUX); 5825 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5826 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5827 spa_load_l2cache(spa); 5828 spa->spa_l2cache.sav_sync = B_TRUE; 5829 } else if (vd != NULL && vd->vdev_islog) { 5830 ASSERT(!locked); 5831 ASSERT(vd == vd->vdev_top); 5832 5833 mg = vd->vdev_mg; 5834 5835 /* 5836 * Stop allocating from this vdev. 5837 */ 5838 metaslab_group_passivate(mg); 5839 5840 /* 5841 * Wait for the youngest allocations and frees to sync, 5842 * and then wait for the deferral of those frees to finish. 5843 */ 5844 spa_vdev_config_exit(spa, NULL, 5845 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5846 5847 /* 5848 * Attempt to evacuate the vdev. 5849 */ 5850 error = spa_vdev_remove_evacuate(spa, vd); 5851 5852 txg = spa_vdev_config_enter(spa); 5853 5854 /* 5855 * If we couldn't evacuate the vdev, unwind. 5856 */ 5857 if (error) { 5858 metaslab_group_activate(mg); 5859 return (spa_vdev_exit(spa, NULL, txg, error)); 5860 } 5861 5862 /* 5863 * Clean up the vdev namespace. 5864 */ 5865 ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_DEV); 5866 spa_vdev_remove_from_namespace(spa, vd); 5867 5868 } else if (vd != NULL) { 5869 /* 5870 * Normal vdevs cannot be removed (yet). 5871 */ 5872 error = SET_ERROR(ENOTSUP); 5873 } else { 5874 /* 5875 * There is no vdev of any kind with the specified guid. 5876 */ 5877 error = SET_ERROR(ENOENT); 5878 } 5879 5880 if (!locked) 5881 error = spa_vdev_exit(spa, NULL, txg, error); 5882 5883 if (ev) 5884 spa_event_post(ev); 5885 5886 return (error); 5887} 5888 5889/* 5890 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5891 * currently spared, so we can detach it. 5892 */ 5893static vdev_t * 5894spa_vdev_resilver_done_hunt(vdev_t *vd) 5895{ 5896 vdev_t *newvd, *oldvd; 5897 5898 for (int c = 0; c < vd->vdev_children; c++) { 5899 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5900 if (oldvd != NULL) 5901 return (oldvd); 5902 } 5903 5904 /* 5905 * Check for a completed replacement. We always consider the first 5906 * vdev in the list to be the oldest vdev, and the last one to be 5907 * the newest (see spa_vdev_attach() for how that works). In 5908 * the case where the newest vdev is faulted, we will not automatically 5909 * remove it after a resilver completes. This is OK as it will require 5910 * user intervention to determine which disk the admin wishes to keep. 5911 */ 5912 if (vd->vdev_ops == &vdev_replacing_ops) { 5913 ASSERT(vd->vdev_children > 1); 5914 5915 newvd = vd->vdev_child[vd->vdev_children - 1]; 5916 oldvd = vd->vdev_child[0]; 5917 5918 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5919 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5920 !vdev_dtl_required(oldvd)) 5921 return (oldvd); 5922 } 5923 5924 /* 5925 * Check for a completed resilver with the 'unspare' flag set. 5926 */ 5927 if (vd->vdev_ops == &vdev_spare_ops) { 5928 vdev_t *first = vd->vdev_child[0]; 5929 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5930 5931 if (last->vdev_unspare) { 5932 oldvd = first; 5933 newvd = last; 5934 } else if (first->vdev_unspare) { 5935 oldvd = last; 5936 newvd = first; 5937 } else { 5938 oldvd = NULL; 5939 } 5940 5941 if (oldvd != NULL && 5942 vdev_dtl_empty(newvd, DTL_MISSING) && 5943 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5944 !vdev_dtl_required(oldvd)) 5945 return (oldvd); 5946 5947 /* 5948 * If there are more than two spares attached to a disk, 5949 * and those spares are not required, then we want to 5950 * attempt to free them up now so that they can be used 5951 * by other pools. Once we're back down to a single 5952 * disk+spare, we stop removing them. 5953 */ 5954 if (vd->vdev_children > 2) { 5955 newvd = vd->vdev_child[1]; 5956 5957 if (newvd->vdev_isspare && last->vdev_isspare && 5958 vdev_dtl_empty(last, DTL_MISSING) && 5959 vdev_dtl_empty(last, DTL_OUTAGE) && 5960 !vdev_dtl_required(newvd)) 5961 return (newvd); 5962 } 5963 } 5964 5965 return (NULL); 5966} 5967 5968static void 5969spa_vdev_resilver_done(spa_t *spa) 5970{ 5971 vdev_t *vd, *pvd, *ppvd; 5972 uint64_t guid, sguid, pguid, ppguid; 5973 5974 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5975 5976 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5977 pvd = vd->vdev_parent; 5978 ppvd = pvd->vdev_parent; 5979 guid = vd->vdev_guid; 5980 pguid = pvd->vdev_guid; 5981 ppguid = ppvd->vdev_guid; 5982 sguid = 0; 5983 /* 5984 * If we have just finished replacing a hot spared device, then 5985 * we need to detach the parent's first child (the original hot 5986 * spare) as well. 5987 */ 5988 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5989 ppvd->vdev_children == 2) { 5990 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5991 sguid = ppvd->vdev_child[1]->vdev_guid; 5992 } 5993 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5994 5995 spa_config_exit(spa, SCL_ALL, FTAG); 5996 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5997 return; 5998 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5999 return; 6000 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6001 } 6002 6003 spa_config_exit(spa, SCL_ALL, FTAG); 6004} 6005 6006/* 6007 * Update the stored path or FRU for this vdev. 6008 */ 6009int 6010spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 6011 boolean_t ispath) 6012{ 6013 vdev_t *vd; 6014 boolean_t sync = B_FALSE; 6015 6016 ASSERT(spa_writeable(spa)); 6017 6018 spa_vdev_state_enter(spa, SCL_ALL); 6019 6020 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 6021 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 6022 6023 if (!vd->vdev_ops->vdev_op_leaf) 6024 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 6025 6026 if (ispath) { 6027 if (strcmp(value, vd->vdev_path) != 0) { 6028 spa_strfree(vd->vdev_path); 6029 vd->vdev_path = spa_strdup(value); 6030 sync = B_TRUE; 6031 } 6032 } else { 6033 if (vd->vdev_fru == NULL) { 6034 vd->vdev_fru = spa_strdup(value); 6035 sync = B_TRUE; 6036 } else if (strcmp(value, vd->vdev_fru) != 0) { 6037 spa_strfree(vd->vdev_fru); 6038 vd->vdev_fru = spa_strdup(value); 6039 sync = B_TRUE; 6040 } 6041 } 6042 6043 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 6044} 6045 6046int 6047spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 6048{ 6049 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 6050} 6051 6052int 6053spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 6054{ 6055 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 6056} 6057 6058/* 6059 * ========================================================================== 6060 * SPA Scanning 6061 * ========================================================================== 6062 */ 6063int 6064spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd) 6065{ 6066 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6067 6068 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6069 return (SET_ERROR(EBUSY)); 6070 6071 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd)); 6072} 6073 6074int 6075spa_scan_stop(spa_t *spa) 6076{ 6077 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6078 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 6079 return (SET_ERROR(EBUSY)); 6080 return (dsl_scan_cancel(spa->spa_dsl_pool)); 6081} 6082 6083int 6084spa_scan(spa_t *spa, pool_scan_func_t func) 6085{ 6086 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 6087 6088 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 6089 return (SET_ERROR(ENOTSUP)); 6090 6091 /* 6092 * If a resilver was requested, but there is no DTL on a 6093 * writeable leaf device, we have nothing to do. 6094 */ 6095 if (func == POOL_SCAN_RESILVER && 6096 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 6097 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 6098 return (0); 6099 } 6100 6101 return (dsl_scan(spa->spa_dsl_pool, func)); 6102} 6103 6104/* 6105 * ========================================================================== 6106 * SPA async task processing 6107 * ========================================================================== 6108 */ 6109 6110static void 6111spa_async_remove(spa_t *spa, vdev_t *vd) 6112{ 6113 if (vd->vdev_remove_wanted) { 6114 vd->vdev_remove_wanted = B_FALSE; 6115 vd->vdev_delayed_close = B_FALSE; 6116 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 6117 6118 /* 6119 * We want to clear the stats, but we don't want to do a full 6120 * vdev_clear() as that will cause us to throw away 6121 * degraded/faulted state as well as attempt to reopen the 6122 * device, all of which is a waste. 6123 */ 6124 vd->vdev_stat.vs_read_errors = 0; 6125 vd->vdev_stat.vs_write_errors = 0; 6126 vd->vdev_stat.vs_checksum_errors = 0; 6127 6128 vdev_state_dirty(vd->vdev_top); 6129 /* Tell userspace that the vdev is gone. */ 6130 zfs_post_remove(spa, vd); 6131 } 6132 6133 for (int c = 0; c < vd->vdev_children; c++) 6134 spa_async_remove(spa, vd->vdev_child[c]); 6135} 6136 6137static void 6138spa_async_probe(spa_t *spa, vdev_t *vd) 6139{ 6140 if (vd->vdev_probe_wanted) { 6141 vd->vdev_probe_wanted = B_FALSE; 6142 vdev_reopen(vd); /* vdev_open() does the actual probe */ 6143 } 6144 6145 for (int c = 0; c < vd->vdev_children; c++) 6146 spa_async_probe(spa, vd->vdev_child[c]); 6147} 6148 6149static void 6150spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6151{ 6152 sysevent_id_t eid; 6153 nvlist_t *attr; 6154 char *physpath; 6155 6156 if (!spa->spa_autoexpand) 6157 return; 6158 6159 for (int c = 0; c < vd->vdev_children; c++) { 6160 vdev_t *cvd = vd->vdev_child[c]; 6161 spa_async_autoexpand(spa, cvd); 6162 } 6163 6164 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6165 return; 6166 6167 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6168 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6169 6170 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6171 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6172 6173 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6174 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 6175 6176 nvlist_free(attr); 6177 kmem_free(physpath, MAXPATHLEN); 6178} 6179 6180static void 6181spa_async_thread(void *arg) 6182{ 6183 spa_t *spa = (spa_t *)arg; 6184 int tasks; 6185 6186 ASSERT(spa->spa_sync_on); 6187 6188 mutex_enter(&spa->spa_async_lock); 6189 tasks = spa->spa_async_tasks; 6190 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 6191 mutex_exit(&spa->spa_async_lock); 6192 6193 /* 6194 * See if the config needs to be updated. 6195 */ 6196 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6197 uint64_t old_space, new_space; 6198 6199 mutex_enter(&spa_namespace_lock); 6200 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6201 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6202 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6203 mutex_exit(&spa_namespace_lock); 6204 6205 /* 6206 * If the pool grew as a result of the config update, 6207 * then log an internal history event. 6208 */ 6209 if (new_space != old_space) { 6210 spa_history_log_internal(spa, "vdev online", NULL, 6211 "pool '%s' size: %llu(+%llu)", 6212 spa_name(spa), new_space, new_space - old_space); 6213 } 6214 } 6215 6216 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6217 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6218 spa_async_autoexpand(spa, spa->spa_root_vdev); 6219 spa_config_exit(spa, SCL_CONFIG, FTAG); 6220 } 6221 6222 /* 6223 * See if any devices need to be probed. 6224 */ 6225 if (tasks & SPA_ASYNC_PROBE) { 6226 spa_vdev_state_enter(spa, SCL_NONE); 6227 spa_async_probe(spa, spa->spa_root_vdev); 6228 (void) spa_vdev_state_exit(spa, NULL, 0); 6229 } 6230 6231 /* 6232 * If any devices are done replacing, detach them. 6233 */ 6234 if (tasks & SPA_ASYNC_RESILVER_DONE) 6235 spa_vdev_resilver_done(spa); 6236 6237 /* 6238 * Kick off a resilver. 6239 */ 6240 if (tasks & SPA_ASYNC_RESILVER) 6241 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6242 6243 /* 6244 * Let the world know that we're done. 6245 */ 6246 mutex_enter(&spa->spa_async_lock); 6247 spa->spa_async_thread = NULL; 6248 cv_broadcast(&spa->spa_async_cv); 6249 mutex_exit(&spa->spa_async_lock); 6250 thread_exit(); 6251} 6252 6253static void 6254spa_async_thread_vd(void *arg) 6255{ 6256 spa_t *spa = arg; 6257 int tasks; 6258 6259 mutex_enter(&spa->spa_async_lock); 6260 tasks = spa->spa_async_tasks; 6261retry: 6262 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6263 mutex_exit(&spa->spa_async_lock); 6264 6265 /* 6266 * See if any devices need to be marked REMOVED. 6267 */ 6268 if (tasks & SPA_ASYNC_REMOVE) { 6269 spa_vdev_state_enter(spa, SCL_NONE); 6270 spa_async_remove(spa, spa->spa_root_vdev); 6271 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6272 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6273 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6274 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6275 (void) spa_vdev_state_exit(spa, NULL, 0); 6276 } 6277 6278 /* 6279 * Let the world know that we're done. 6280 */ 6281 mutex_enter(&spa->spa_async_lock); 6282 tasks = spa->spa_async_tasks; 6283 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6284 goto retry; 6285 spa->spa_async_thread_vd = NULL; 6286 cv_broadcast(&spa->spa_async_cv); 6287 mutex_exit(&spa->spa_async_lock); 6288 thread_exit(); 6289} 6290 6291void 6292spa_async_suspend(spa_t *spa) 6293{ 6294 mutex_enter(&spa->spa_async_lock); 6295 spa->spa_async_suspended++; 6296 while (spa->spa_async_thread != NULL && 6297 spa->spa_async_thread_vd != NULL) 6298 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6299 mutex_exit(&spa->spa_async_lock); 6300} 6301 6302void 6303spa_async_resume(spa_t *spa) 6304{ 6305 mutex_enter(&spa->spa_async_lock); 6306 ASSERT(spa->spa_async_suspended != 0); 6307 spa->spa_async_suspended--; 6308 mutex_exit(&spa->spa_async_lock); 6309} 6310 6311static boolean_t 6312spa_async_tasks_pending(spa_t *spa) 6313{ 6314 uint_t non_config_tasks; 6315 uint_t config_task; 6316 boolean_t config_task_suspended; 6317 6318 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6319 SPA_ASYNC_REMOVE); 6320 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6321 if (spa->spa_ccw_fail_time == 0) { 6322 config_task_suspended = B_FALSE; 6323 } else { 6324 config_task_suspended = 6325 (gethrtime() - spa->spa_ccw_fail_time) < 6326 (zfs_ccw_retry_interval * NANOSEC); 6327 } 6328 6329 return (non_config_tasks || (config_task && !config_task_suspended)); 6330} 6331 6332static void 6333spa_async_dispatch(spa_t *spa) 6334{ 6335 mutex_enter(&spa->spa_async_lock); 6336 if (spa_async_tasks_pending(spa) && 6337 !spa->spa_async_suspended && 6338 spa->spa_async_thread == NULL && 6339 rootdir != NULL) 6340 spa->spa_async_thread = thread_create(NULL, 0, 6341 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6342 mutex_exit(&spa->spa_async_lock); 6343} 6344 6345static void 6346spa_async_dispatch_vd(spa_t *spa) 6347{ 6348 mutex_enter(&spa->spa_async_lock); 6349 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6350 !spa->spa_async_suspended && 6351 spa->spa_async_thread_vd == NULL && 6352 rootdir != NULL) 6353 spa->spa_async_thread_vd = thread_create(NULL, 0, 6354 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6355 mutex_exit(&spa->spa_async_lock); 6356} 6357 6358void 6359spa_async_request(spa_t *spa, int task) 6360{ 6361 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6362 mutex_enter(&spa->spa_async_lock); 6363 spa->spa_async_tasks |= task; 6364 mutex_exit(&spa->spa_async_lock); 6365 spa_async_dispatch_vd(spa); 6366} 6367 6368/* 6369 * ========================================================================== 6370 * SPA syncing routines 6371 * ========================================================================== 6372 */ 6373 6374static int 6375bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6376{ 6377 bpobj_t *bpo = arg; 6378 bpobj_enqueue(bpo, bp, tx); 6379 return (0); 6380} 6381 6382static int 6383spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6384{ 6385 zio_t *zio = arg; 6386 6387 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6388 BP_GET_PSIZE(bp), zio->io_flags)); 6389 return (0); 6390} 6391 6392/* 6393 * Note: this simple function is not inlined to make it easier to dtrace the 6394 * amount of time spent syncing frees. 6395 */ 6396static void 6397spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6398{ 6399 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6400 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6401 VERIFY(zio_wait(zio) == 0); 6402} 6403 6404/* 6405 * Note: this simple function is not inlined to make it easier to dtrace the 6406 * amount of time spent syncing deferred frees. 6407 */ 6408static void 6409spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6410{ 6411 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6412 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6413 spa_free_sync_cb, zio, tx), ==, 0); 6414 VERIFY0(zio_wait(zio)); 6415} 6416 6417 6418static void 6419spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6420{ 6421 char *packed = NULL; 6422 size_t bufsize; 6423 size_t nvsize = 0; 6424 dmu_buf_t *db; 6425 6426 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6427 6428 /* 6429 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6430 * information. This avoids the dmu_buf_will_dirty() path and 6431 * saves us a pre-read to get data we don't actually care about. 6432 */ 6433 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6434 packed = kmem_alloc(bufsize, KM_SLEEP); 6435 6436 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6437 KM_SLEEP) == 0); 6438 bzero(packed + nvsize, bufsize - nvsize); 6439 6440 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6441 6442 kmem_free(packed, bufsize); 6443 6444 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6445 dmu_buf_will_dirty(db, tx); 6446 *(uint64_t *)db->db_data = nvsize; 6447 dmu_buf_rele(db, FTAG); 6448} 6449 6450static void 6451spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6452 const char *config, const char *entry) 6453{ 6454 nvlist_t *nvroot; 6455 nvlist_t **list; 6456 int i; 6457 6458 if (!sav->sav_sync) 6459 return; 6460 6461 /* 6462 * Update the MOS nvlist describing the list of available devices. 6463 * spa_validate_aux() will have already made sure this nvlist is 6464 * valid and the vdevs are labeled appropriately. 6465 */ 6466 if (sav->sav_object == 0) { 6467 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6468 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6469 sizeof (uint64_t), tx); 6470 VERIFY(zap_update(spa->spa_meta_objset, 6471 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6472 &sav->sav_object, tx) == 0); 6473 } 6474 6475 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6476 if (sav->sav_count == 0) { 6477 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6478 } else { 6479 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6480 for (i = 0; i < sav->sav_count; i++) 6481 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6482 B_FALSE, VDEV_CONFIG_L2CACHE); 6483 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6484 sav->sav_count) == 0); 6485 for (i = 0; i < sav->sav_count; i++) 6486 nvlist_free(list[i]); 6487 kmem_free(list, sav->sav_count * sizeof (void *)); 6488 } 6489 6490 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6491 nvlist_free(nvroot); 6492 6493 sav->sav_sync = B_FALSE; 6494} 6495 6496/* 6497 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t. 6498 * The all-vdev ZAP must be empty. 6499 */ 6500static void 6501spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx) 6502{ 6503 spa_t *spa = vd->vdev_spa; 6504 if (vd->vdev_top_zap != 0) { 6505 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6506 vd->vdev_top_zap, tx)); 6507 } 6508 if (vd->vdev_leaf_zap != 0) { 6509 VERIFY0(zap_add_int(spa->spa_meta_objset, avz, 6510 vd->vdev_leaf_zap, tx)); 6511 } 6512 for (uint64_t i = 0; i < vd->vdev_children; i++) { 6513 spa_avz_build(vd->vdev_child[i], avz, tx); 6514 } 6515} 6516 6517static void 6518spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6519{ 6520 nvlist_t *config; 6521 6522 /* 6523 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS, 6524 * its config may not be dirty but we still need to build per-vdev ZAPs. 6525 * Similarly, if the pool is being assembled (e.g. after a split), we 6526 * need to rebuild the AVZ although the config may not be dirty. 6527 */ 6528 if (list_is_empty(&spa->spa_config_dirty_list) && 6529 spa->spa_avz_action == AVZ_ACTION_NONE) 6530 return; 6531 6532 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6533 6534 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE || 6535 spa->spa_avz_action == AVZ_ACTION_INITIALIZE || 6536 spa->spa_all_vdev_zaps != 0); 6537 6538 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) { 6539 /* Make and build the new AVZ */ 6540 uint64_t new_avz = zap_create(spa->spa_meta_objset, 6541 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx); 6542 spa_avz_build(spa->spa_root_vdev, new_avz, tx); 6543 6544 /* Diff old AVZ with new one */ 6545 zap_cursor_t zc; 6546 zap_attribute_t za; 6547 6548 for (zap_cursor_init(&zc, spa->spa_meta_objset, 6549 spa->spa_all_vdev_zaps); 6550 zap_cursor_retrieve(&zc, &za) == 0; 6551 zap_cursor_advance(&zc)) { 6552 uint64_t vdzap = za.za_first_integer; 6553 if (zap_lookup_int(spa->spa_meta_objset, new_avz, 6554 vdzap) == ENOENT) { 6555 /* 6556 * ZAP is listed in old AVZ but not in new one; 6557 * destroy it 6558 */ 6559 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap, 6560 tx)); 6561 } 6562 } 6563 6564 zap_cursor_fini(&zc); 6565 6566 /* Destroy the old AVZ */ 6567 VERIFY0(zap_destroy(spa->spa_meta_objset, 6568 spa->spa_all_vdev_zaps, tx)); 6569 6570 /* Replace the old AVZ in the dir obj with the new one */ 6571 VERIFY0(zap_update(spa->spa_meta_objset, 6572 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, 6573 sizeof (new_avz), 1, &new_avz, tx)); 6574 6575 spa->spa_all_vdev_zaps = new_avz; 6576 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) { 6577 zap_cursor_t zc; 6578 zap_attribute_t za; 6579 6580 /* Walk through the AVZ and destroy all listed ZAPs */ 6581 for (zap_cursor_init(&zc, spa->spa_meta_objset, 6582 spa->spa_all_vdev_zaps); 6583 zap_cursor_retrieve(&zc, &za) == 0; 6584 zap_cursor_advance(&zc)) { 6585 uint64_t zap = za.za_first_integer; 6586 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx)); 6587 } 6588 6589 zap_cursor_fini(&zc); 6590 6591 /* Destroy and unlink the AVZ itself */ 6592 VERIFY0(zap_destroy(spa->spa_meta_objset, 6593 spa->spa_all_vdev_zaps, tx)); 6594 VERIFY0(zap_remove(spa->spa_meta_objset, 6595 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx)); 6596 spa->spa_all_vdev_zaps = 0; 6597 } 6598 6599 if (spa->spa_all_vdev_zaps == 0) { 6600 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset, 6601 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT, 6602 DMU_POOL_VDEV_ZAP_MAP, tx); 6603 } 6604 spa->spa_avz_action = AVZ_ACTION_NONE; 6605 6606 /* Create ZAPs for vdevs that don't have them. */ 6607 vdev_construct_zaps(spa->spa_root_vdev, tx); 6608 6609 config = spa_config_generate(spa, spa->spa_root_vdev, 6610 dmu_tx_get_txg(tx), B_FALSE); 6611 6612 /* 6613 * If we're upgrading the spa version then make sure that 6614 * the config object gets updated with the correct version. 6615 */ 6616 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6617 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6618 spa->spa_uberblock.ub_version); 6619 6620 spa_config_exit(spa, SCL_STATE, FTAG); 6621 6622 nvlist_free(spa->spa_config_syncing); 6623 spa->spa_config_syncing = config; 6624 6625 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6626} 6627 6628static void 6629spa_sync_version(void *arg, dmu_tx_t *tx) 6630{ 6631 uint64_t *versionp = arg; 6632 uint64_t version = *versionp; 6633 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6634 6635 /* 6636 * Setting the version is special cased when first creating the pool. 6637 */ 6638 ASSERT(tx->tx_txg != TXG_INITIAL); 6639 6640 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6641 ASSERT(version >= spa_version(spa)); 6642 6643 spa->spa_uberblock.ub_version = version; 6644 vdev_config_dirty(spa->spa_root_vdev); 6645 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6646} 6647 6648/* 6649 * Set zpool properties. 6650 */ 6651static void 6652spa_sync_props(void *arg, dmu_tx_t *tx) 6653{ 6654 nvlist_t *nvp = arg; 6655 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6656 objset_t *mos = spa->spa_meta_objset; 6657 nvpair_t *elem = NULL; 6658 6659 mutex_enter(&spa->spa_props_lock); 6660 6661 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6662 uint64_t intval; 6663 char *strval, *fname; 6664 zpool_prop_t prop; 6665 const char *propname; 6666 zprop_type_t proptype; 6667 spa_feature_t fid; 6668 6669 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6670 case ZPOOL_PROP_INVAL: 6671 /* 6672 * We checked this earlier in spa_prop_validate(). 6673 */ 6674 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6675 6676 fname = strchr(nvpair_name(elem), '@') + 1; 6677 VERIFY0(zfeature_lookup_name(fname, &fid)); 6678 6679 spa_feature_enable(spa, fid, tx); 6680 spa_history_log_internal(spa, "set", tx, 6681 "%s=enabled", nvpair_name(elem)); 6682 break; 6683 6684 case ZPOOL_PROP_VERSION: 6685 intval = fnvpair_value_uint64(elem); 6686 /* 6687 * The version is synced seperatly before other 6688 * properties and should be correct by now. 6689 */ 6690 ASSERT3U(spa_version(spa), >=, intval); 6691 break; 6692 6693 case ZPOOL_PROP_ALTROOT: 6694 /* 6695 * 'altroot' is a non-persistent property. It should 6696 * have been set temporarily at creation or import time. 6697 */ 6698 ASSERT(spa->spa_root != NULL); 6699 break; 6700 6701 case ZPOOL_PROP_READONLY: 6702 case ZPOOL_PROP_CACHEFILE: 6703 /* 6704 * 'readonly' and 'cachefile' are also non-persisitent 6705 * properties. 6706 */ 6707 break; 6708 case ZPOOL_PROP_COMMENT: 6709 strval = fnvpair_value_string(elem); 6710 if (spa->spa_comment != NULL) 6711 spa_strfree(spa->spa_comment); 6712 spa->spa_comment = spa_strdup(strval); 6713 /* 6714 * We need to dirty the configuration on all the vdevs 6715 * so that their labels get updated. It's unnecessary 6716 * to do this for pool creation since the vdev's 6717 * configuratoin has already been dirtied. 6718 */ 6719 if (tx->tx_txg != TXG_INITIAL) 6720 vdev_config_dirty(spa->spa_root_vdev); 6721 spa_history_log_internal(spa, "set", tx, 6722 "%s=%s", nvpair_name(elem), strval); 6723 break; 6724 default: 6725 /* 6726 * Set pool property values in the poolprops mos object. 6727 */ 6728 if (spa->spa_pool_props_object == 0) { 6729 spa->spa_pool_props_object = 6730 zap_create_link(mos, DMU_OT_POOL_PROPS, 6731 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6732 tx); 6733 } 6734 6735 /* normalize the property name */ 6736 propname = zpool_prop_to_name(prop); 6737 proptype = zpool_prop_get_type(prop); 6738 6739 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6740 ASSERT(proptype == PROP_TYPE_STRING); 6741 strval = fnvpair_value_string(elem); 6742 VERIFY0(zap_update(mos, 6743 spa->spa_pool_props_object, propname, 6744 1, strlen(strval) + 1, strval, tx)); 6745 spa_history_log_internal(spa, "set", tx, 6746 "%s=%s", nvpair_name(elem), strval); 6747 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6748 intval = fnvpair_value_uint64(elem); 6749 6750 if (proptype == PROP_TYPE_INDEX) { 6751 const char *unused; 6752 VERIFY0(zpool_prop_index_to_string( 6753 prop, intval, &unused)); 6754 } 6755 VERIFY0(zap_update(mos, 6756 spa->spa_pool_props_object, propname, 6757 8, 1, &intval, tx)); 6758 spa_history_log_internal(spa, "set", tx, 6759 "%s=%lld", nvpair_name(elem), intval); 6760 } else { 6761 ASSERT(0); /* not allowed */ 6762 } 6763 6764 switch (prop) { 6765 case ZPOOL_PROP_DELEGATION: 6766 spa->spa_delegation = intval; 6767 break; 6768 case ZPOOL_PROP_BOOTFS: 6769 spa->spa_bootfs = intval; 6770 break; 6771 case ZPOOL_PROP_FAILUREMODE: 6772 spa->spa_failmode = intval; 6773 break; 6774 case ZPOOL_PROP_AUTOEXPAND: 6775 spa->spa_autoexpand = intval; 6776 if (tx->tx_txg != TXG_INITIAL) 6777 spa_async_request(spa, 6778 SPA_ASYNC_AUTOEXPAND); 6779 break; 6780 case ZPOOL_PROP_DEDUPDITTO: 6781 spa->spa_dedup_ditto = intval; 6782 break; 6783 default: 6784 break; 6785 } 6786 } 6787 6788 } 6789 6790 mutex_exit(&spa->spa_props_lock); 6791} 6792 6793/* 6794 * Perform one-time upgrade on-disk changes. spa_version() does not 6795 * reflect the new version this txg, so there must be no changes this 6796 * txg to anything that the upgrade code depends on after it executes. 6797 * Therefore this must be called after dsl_pool_sync() does the sync 6798 * tasks. 6799 */ 6800static void 6801spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6802{ 6803 dsl_pool_t *dp = spa->spa_dsl_pool; 6804 6805 ASSERT(spa->spa_sync_pass == 1); 6806 6807 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6808 6809 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6810 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6811 dsl_pool_create_origin(dp, tx); 6812 6813 /* Keeping the origin open increases spa_minref */ 6814 spa->spa_minref += 3; 6815 } 6816 6817 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6818 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6819 dsl_pool_upgrade_clones(dp, tx); 6820 } 6821 6822 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6823 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6824 dsl_pool_upgrade_dir_clones(dp, tx); 6825 6826 /* Keeping the freedir open increases spa_minref */ 6827 spa->spa_minref += 3; 6828 } 6829 6830 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6831 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6832 spa_feature_create_zap_objects(spa, tx); 6833 } 6834 6835 /* 6836 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6837 * when possibility to use lz4 compression for metadata was added 6838 * Old pools that have this feature enabled must be upgraded to have 6839 * this feature active 6840 */ 6841 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6842 boolean_t lz4_en = spa_feature_is_enabled(spa, 6843 SPA_FEATURE_LZ4_COMPRESS); 6844 boolean_t lz4_ac = spa_feature_is_active(spa, 6845 SPA_FEATURE_LZ4_COMPRESS); 6846 6847 if (lz4_en && !lz4_ac) 6848 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6849 } 6850 6851 /* 6852 * If we haven't written the salt, do so now. Note that the 6853 * feature may not be activated yet, but that's fine since 6854 * the presence of this ZAP entry is backwards compatible. 6855 */ 6856 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 6857 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 6858 VERIFY0(zap_add(spa->spa_meta_objset, 6859 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 6860 sizeof (spa->spa_cksum_salt.zcs_bytes), 6861 spa->spa_cksum_salt.zcs_bytes, tx)); 6862 } 6863 6864 rrw_exit(&dp->dp_config_rwlock, FTAG); 6865} 6866 6867/* 6868 * Sync the specified transaction group. New blocks may be dirtied as 6869 * part of the process, so we iterate until it converges. 6870 */ 6871void 6872spa_sync(spa_t *spa, uint64_t txg) 6873{ 6874 dsl_pool_t *dp = spa->spa_dsl_pool; 6875 objset_t *mos = spa->spa_meta_objset; 6876 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6877 vdev_t *rvd = spa->spa_root_vdev; 6878 vdev_t *vd; 6879 dmu_tx_t *tx; 6880 int error; 6881 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 6882 zfs_vdev_queue_depth_pct / 100; 6883 6884 VERIFY(spa_writeable(spa)); 6885 6886 /* 6887 * Lock out configuration changes. 6888 */ 6889 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6890 6891 spa->spa_syncing_txg = txg; 6892 spa->spa_sync_pass = 0; 6893 6894 mutex_enter(&spa->spa_alloc_lock); 6895 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 6896 mutex_exit(&spa->spa_alloc_lock); 6897 6898 /* 6899 * If there are any pending vdev state changes, convert them 6900 * into config changes that go out with this transaction group. 6901 */ 6902 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6903 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6904 /* 6905 * We need the write lock here because, for aux vdevs, 6906 * calling vdev_config_dirty() modifies sav_config. 6907 * This is ugly and will become unnecessary when we 6908 * eliminate the aux vdev wart by integrating all vdevs 6909 * into the root vdev tree. 6910 */ 6911 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6912 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6913 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6914 vdev_state_clean(vd); 6915 vdev_config_dirty(vd); 6916 } 6917 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6918 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6919 } 6920 spa_config_exit(spa, SCL_STATE, FTAG); 6921 6922 tx = dmu_tx_create_assigned(dp, txg); 6923 6924 spa->spa_sync_starttime = gethrtime(); 6925#ifdef illumos 6926 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6927 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6928#else /* !illumos */ 6929#ifdef _KERNEL 6930 callout_schedule(&spa->spa_deadman_cycid, 6931 hz * spa->spa_deadman_synctime / NANOSEC); 6932#endif 6933#endif /* illumos */ 6934 6935 /* 6936 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6937 * set spa_deflate if we have no raid-z vdevs. 6938 */ 6939 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6940 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6941 int i; 6942 6943 for (i = 0; i < rvd->vdev_children; i++) { 6944 vd = rvd->vdev_child[i]; 6945 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6946 break; 6947 } 6948 if (i == rvd->vdev_children) { 6949 spa->spa_deflate = TRUE; 6950 VERIFY(0 == zap_add(spa->spa_meta_objset, 6951 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6952 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6953 } 6954 } 6955 6956 /* 6957 * Set the top-level vdev's max queue depth. Evaluate each 6958 * top-level's async write queue depth in case it changed. 6959 * The max queue depth will not change in the middle of syncing 6960 * out this txg. 6961 */ 6962 uint64_t queue_depth_total = 0; 6963 for (int c = 0; c < rvd->vdev_children; c++) { 6964 vdev_t *tvd = rvd->vdev_child[c]; 6965 metaslab_group_t *mg = tvd->vdev_mg; 6966 6967 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 6968 !metaslab_group_initialized(mg)) 6969 continue; 6970 6971 /* 6972 * It is safe to do a lock-free check here because only async 6973 * allocations look at mg_max_alloc_queue_depth, and async 6974 * allocations all happen from spa_sync(). 6975 */ 6976 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); 6977 mg->mg_max_alloc_queue_depth = max_queue_depth; 6978 queue_depth_total += mg->mg_max_alloc_queue_depth; 6979 } 6980 metaslab_class_t *mc = spa_normal_class(spa); 6981 ASSERT0(refcount_count(&mc->mc_alloc_slots)); 6982 mc->mc_alloc_max_slots = queue_depth_total; 6983 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 6984 6985 ASSERT3U(mc->mc_alloc_max_slots, <=, 6986 max_queue_depth * rvd->vdev_children); 6987 6988 /* 6989 * Iterate to convergence. 6990 */ 6991 do { 6992 int pass = ++spa->spa_sync_pass; 6993 6994 spa_sync_config_object(spa, tx); 6995 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6996 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6997 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6998 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6999 spa_errlog_sync(spa, txg); 7000 dsl_pool_sync(dp, txg); 7001 7002 if (pass < zfs_sync_pass_deferred_free) { 7003 spa_sync_frees(spa, free_bpl, tx); 7004 } else { 7005 /* 7006 * We can not defer frees in pass 1, because 7007 * we sync the deferred frees later in pass 1. 7008 */ 7009 ASSERT3U(pass, >, 1); 7010 bplist_iterate(free_bpl, bpobj_enqueue_cb, 7011 &spa->spa_deferred_bpobj, tx); 7012 } 7013 7014 ddt_sync(spa, txg); 7015 dsl_scan_sync(dp, tx); 7016 7017 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 7018 vdev_sync(vd, txg); 7019 7020 if (pass == 1) { 7021 spa_sync_upgrades(spa, tx); 7022 ASSERT3U(txg, >=, 7023 spa->spa_uberblock.ub_rootbp.blk_birth); 7024 /* 7025 * Note: We need to check if the MOS is dirty 7026 * because we could have marked the MOS dirty 7027 * without updating the uberblock (e.g. if we 7028 * have sync tasks but no dirty user data). We 7029 * need to check the uberblock's rootbp because 7030 * it is updated if we have synced out dirty 7031 * data (though in this case the MOS will most 7032 * likely also be dirty due to second order 7033 * effects, we don't want to rely on that here). 7034 */ 7035 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 7036 !dmu_objset_is_dirty(mos, txg)) { 7037 /* 7038 * Nothing changed on the first pass, 7039 * therefore this TXG is a no-op. Avoid 7040 * syncing deferred frees, so that we 7041 * can keep this TXG as a no-op. 7042 */ 7043 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 7044 txg)); 7045 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7046 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 7047 break; 7048 } 7049 spa_sync_deferred_frees(spa, tx); 7050 } 7051 7052 } while (dmu_objset_is_dirty(mos, txg)); 7053 7054 if (!list_is_empty(&spa->spa_config_dirty_list)) { 7055 /* 7056 * Make sure that the number of ZAPs for all the vdevs matches 7057 * the number of ZAPs in the per-vdev ZAP list. This only gets 7058 * called if the config is dirty; otherwise there may be 7059 * outstanding AVZ operations that weren't completed in 7060 * spa_sync_config_object. 7061 */ 7062 uint64_t all_vdev_zap_entry_count; 7063 ASSERT0(zap_count(spa->spa_meta_objset, 7064 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count)); 7065 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==, 7066 all_vdev_zap_entry_count); 7067 } 7068 7069 /* 7070 * Rewrite the vdev configuration (which includes the uberblock) 7071 * to commit the transaction group. 7072 * 7073 * If there are no dirty vdevs, we sync the uberblock to a few 7074 * random top-level vdevs that are known to be visible in the 7075 * config cache (see spa_vdev_add() for a complete description). 7076 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 7077 */ 7078 for (;;) { 7079 /* 7080 * We hold SCL_STATE to prevent vdev open/close/etc. 7081 * while we're attempting to write the vdev labels. 7082 */ 7083 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 7084 7085 if (list_is_empty(&spa->spa_config_dirty_list)) { 7086 vdev_t *svd[SPA_DVAS_PER_BP]; 7087 int svdcount = 0; 7088 int children = rvd->vdev_children; 7089 int c0 = spa_get_random(children); 7090 7091 for (int c = 0; c < children; c++) { 7092 vd = rvd->vdev_child[(c0 + c) % children]; 7093 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 7094 continue; 7095 svd[svdcount++] = vd; 7096 if (svdcount == SPA_DVAS_PER_BP) 7097 break; 7098 } 7099 error = vdev_config_sync(svd, svdcount, txg); 7100 } else { 7101 error = vdev_config_sync(rvd->vdev_child, 7102 rvd->vdev_children, txg); 7103 } 7104 7105 if (error == 0) 7106 spa->spa_last_synced_guid = rvd->vdev_guid; 7107 7108 spa_config_exit(spa, SCL_STATE, FTAG); 7109 7110 if (error == 0) 7111 break; 7112 zio_suspend(spa, NULL); 7113 zio_resume_wait(spa); 7114 } 7115 dmu_tx_commit(tx); 7116 7117#ifdef illumos 7118 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 7119#else /* !illumos */ 7120#ifdef _KERNEL 7121 callout_drain(&spa->spa_deadman_cycid); 7122#endif 7123#endif /* illumos */ 7124 7125 /* 7126 * Clear the dirty config list. 7127 */ 7128 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 7129 vdev_config_clean(vd); 7130 7131 /* 7132 * Now that the new config has synced transactionally, 7133 * let it become visible to the config cache. 7134 */ 7135 if (spa->spa_config_syncing != NULL) { 7136 spa_config_set(spa, spa->spa_config_syncing); 7137 spa->spa_config_txg = txg; 7138 spa->spa_config_syncing = NULL; 7139 } 7140 7141 dsl_pool_sync_done(dp, txg); 7142 7143 mutex_enter(&spa->spa_alloc_lock); 7144 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 7145 mutex_exit(&spa->spa_alloc_lock); 7146 7147 /* 7148 * Update usable space statistics. 7149 */ 7150 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 7151 vdev_sync_done(vd, txg); 7152 7153 spa_update_dspace(spa); 7154 7155 /* 7156 * It had better be the case that we didn't dirty anything 7157 * since vdev_config_sync(). 7158 */ 7159 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 7160 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 7161 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 7162 7163 spa->spa_sync_pass = 0; 7164 7165 /* 7166 * Update the last synced uberblock here. We want to do this at 7167 * the end of spa_sync() so that consumers of spa_last_synced_txg() 7168 * will be guaranteed that all the processing associated with 7169 * that txg has been completed. 7170 */ 7171 spa->spa_ubsync = spa->spa_uberblock; 7172 spa_config_exit(spa, SCL_CONFIG, FTAG); 7173 7174 spa_handle_ignored_writes(spa); 7175 7176 /* 7177 * If any async tasks have been requested, kick them off. 7178 */ 7179 spa_async_dispatch(spa); 7180 spa_async_dispatch_vd(spa); 7181} 7182 7183/* 7184 * Sync all pools. We don't want to hold the namespace lock across these 7185 * operations, so we take a reference on the spa_t and drop the lock during the 7186 * sync. 7187 */ 7188void 7189spa_sync_allpools(void) 7190{ 7191 spa_t *spa = NULL; 7192 mutex_enter(&spa_namespace_lock); 7193 while ((spa = spa_next(spa)) != NULL) { 7194 if (spa_state(spa) != POOL_STATE_ACTIVE || 7195 !spa_writeable(spa) || spa_suspended(spa)) 7196 continue; 7197 spa_open_ref(spa, FTAG); 7198 mutex_exit(&spa_namespace_lock); 7199 txg_wait_synced(spa_get_dsl(spa), 0); 7200 mutex_enter(&spa_namespace_lock); 7201 spa_close(spa, FTAG); 7202 } 7203 mutex_exit(&spa_namespace_lock); 7204} 7205 7206/* 7207 * ========================================================================== 7208 * Miscellaneous routines 7209 * ========================================================================== 7210 */ 7211 7212/* 7213 * Remove all pools in the system. 7214 */ 7215void 7216spa_evict_all(void) 7217{ 7218 spa_t *spa; 7219 7220 /* 7221 * Remove all cached state. All pools should be closed now, 7222 * so every spa in the AVL tree should be unreferenced. 7223 */ 7224 mutex_enter(&spa_namespace_lock); 7225 while ((spa = spa_next(NULL)) != NULL) { 7226 /* 7227 * Stop async tasks. The async thread may need to detach 7228 * a device that's been replaced, which requires grabbing 7229 * spa_namespace_lock, so we must drop it here. 7230 */ 7231 spa_open_ref(spa, FTAG); 7232 mutex_exit(&spa_namespace_lock); 7233 spa_async_suspend(spa); 7234 mutex_enter(&spa_namespace_lock); 7235 spa_close(spa, FTAG); 7236 7237 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 7238 spa_unload(spa); 7239 spa_deactivate(spa); 7240 } 7241 spa_remove(spa); 7242 } 7243 mutex_exit(&spa_namespace_lock); 7244} 7245 7246vdev_t * 7247spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 7248{ 7249 vdev_t *vd; 7250 int i; 7251 7252 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 7253 return (vd); 7254 7255 if (aux) { 7256 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 7257 vd = spa->spa_l2cache.sav_vdevs[i]; 7258 if (vd->vdev_guid == guid) 7259 return (vd); 7260 } 7261 7262 for (i = 0; i < spa->spa_spares.sav_count; i++) { 7263 vd = spa->spa_spares.sav_vdevs[i]; 7264 if (vd->vdev_guid == guid) 7265 return (vd); 7266 } 7267 } 7268 7269 return (NULL); 7270} 7271 7272void 7273spa_upgrade(spa_t *spa, uint64_t version) 7274{ 7275 ASSERT(spa_writeable(spa)); 7276 7277 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 7278 7279 /* 7280 * This should only be called for a non-faulted pool, and since a 7281 * future version would result in an unopenable pool, this shouldn't be 7282 * possible. 7283 */ 7284 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 7285 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 7286 7287 spa->spa_uberblock.ub_version = version; 7288 vdev_config_dirty(spa->spa_root_vdev); 7289 7290 spa_config_exit(spa, SCL_ALL, FTAG); 7291 7292 txg_wait_synced(spa_get_dsl(spa), 0); 7293} 7294 7295boolean_t 7296spa_has_spare(spa_t *spa, uint64_t guid) 7297{ 7298 int i; 7299 uint64_t spareguid; 7300 spa_aux_vdev_t *sav = &spa->spa_spares; 7301 7302 for (i = 0; i < sav->sav_count; i++) 7303 if (sav->sav_vdevs[i]->vdev_guid == guid) 7304 return (B_TRUE); 7305 7306 for (i = 0; i < sav->sav_npending; i++) { 7307 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 7308 &spareguid) == 0 && spareguid == guid) 7309 return (B_TRUE); 7310 } 7311 7312 return (B_FALSE); 7313} 7314 7315/* 7316 * Check if a pool has an active shared spare device. 7317 * Note: reference count of an active spare is 2, as a spare and as a replace 7318 */ 7319static boolean_t 7320spa_has_active_shared_spare(spa_t *spa) 7321{ 7322 int i, refcnt; 7323 uint64_t pool; 7324 spa_aux_vdev_t *sav = &spa->spa_spares; 7325 7326 for (i = 0; i < sav->sav_count; i++) { 7327 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 7328 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 7329 refcnt > 2) 7330 return (B_TRUE); 7331 } 7332 7333 return (B_FALSE); 7334} 7335 7336static sysevent_t * 7337spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 7338{ 7339 sysevent_t *ev = NULL; 7340#ifdef _KERNEL 7341 sysevent_attr_list_t *attr = NULL; 7342 sysevent_value_t value; 7343 7344 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7345 SE_SLEEP); 7346 ASSERT(ev != NULL); 7347 7348 value.value_type = SE_DATA_TYPE_STRING; 7349 value.value.sv_string = spa_name(spa); 7350 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7351 goto done; 7352 7353 value.value_type = SE_DATA_TYPE_UINT64; 7354 value.value.sv_uint64 = spa_guid(spa); 7355 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7356 goto done; 7357 7358 if (vd) { 7359 value.value_type = SE_DATA_TYPE_UINT64; 7360 value.value.sv_uint64 = vd->vdev_guid; 7361 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7362 SE_SLEEP) != 0) 7363 goto done; 7364 7365 if (vd->vdev_path) { 7366 value.value_type = SE_DATA_TYPE_STRING; 7367 value.value.sv_string = vd->vdev_path; 7368 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7369 &value, SE_SLEEP) != 0) 7370 goto done; 7371 } 7372 } 7373 7374 if (hist_nvl != NULL) { 7375 fnvlist_merge((nvlist_t *)attr, hist_nvl); 7376 } 7377 7378 if (sysevent_attach_attributes(ev, attr) != 0) 7379 goto done; 7380 attr = NULL; 7381 7382done: 7383 if (attr) 7384 sysevent_free_attr(attr); 7385 7386#endif 7387 return (ev); 7388} 7389 7390static void 7391spa_event_post(sysevent_t *ev) 7392{ 7393#ifdef _KERNEL 7394 sysevent_id_t eid; 7395 7396 (void) log_sysevent(ev, SE_SLEEP, &eid); 7397 sysevent_free(ev); 7398#endif 7399} 7400 7401/* 7402 * Post a sysevent corresponding to the given event. The 'name' must be one of 7403 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7404 * filled in from the spa and (optionally) the vdev and history nvl. This 7405 * doesn't do anything in the userland libzpool, as we don't want consumers to 7406 * misinterpret ztest or zdb as real changes. 7407 */ 7408void 7409spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name) 7410{ 7411 spa_event_post(spa_event_create(spa, vd, hist_nvl, name)); 7412} 7413