spa.c revision 288571
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2013 by Delphix. All rights reserved. 25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 */ 29 30/* 31 * SPA: Storage Pool Allocator 32 * 33 * This file contains all the routines used when modifying on-disk SPA state. 34 * This includes opening, importing, destroying, exporting a pool, and syncing a 35 * pool. 36 */ 37 38#include <sys/zfs_context.h> 39#include <sys/fm/fs/zfs.h> 40#include <sys/spa_impl.h> 41#include <sys/zio.h> 42#include <sys/zio_checksum.h> 43#include <sys/dmu.h> 44#include <sys/dmu_tx.h> 45#include <sys/zap.h> 46#include <sys/zil.h> 47#include <sys/ddt.h> 48#include <sys/vdev_impl.h> 49#include <sys/metaslab.h> 50#include <sys/metaslab_impl.h> 51#include <sys/uberblock_impl.h> 52#include <sys/txg.h> 53#include <sys/avl.h> 54#include <sys/dmu_traverse.h> 55#include <sys/dmu_objset.h> 56#include <sys/unique.h> 57#include <sys/dsl_pool.h> 58#include <sys/dsl_dataset.h> 59#include <sys/dsl_dir.h> 60#include <sys/dsl_prop.h> 61#include <sys/dsl_synctask.h> 62#include <sys/fs/zfs.h> 63#include <sys/arc.h> 64#include <sys/callb.h> 65#include <sys/spa_boot.h> 66#include <sys/zfs_ioctl.h> 67#include <sys/dsl_scan.h> 68#include <sys/dmu_send.h> 69#include <sys/dsl_destroy.h> 70#include <sys/dsl_userhold.h> 71#include <sys/zfeature.h> 72#include <sys/zvol.h> 73#include <sys/trim_map.h> 74 75#ifdef _KERNEL 76#include <sys/callb.h> 77#include <sys/cpupart.h> 78#include <sys/zone.h> 79#endif /* _KERNEL */ 80 81#include "zfs_prop.h" 82#include "zfs_comutil.h" 83 84/* Check hostid on import? */ 85static int check_hostid = 1; 86 87SYSCTL_DECL(_vfs_zfs); 88TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 89SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 90 "Check hostid on import?"); 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 98typedef enum zti_modes { 99 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 100 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 101 ZTI_MODE_NULL, /* don't create a taskq */ 102 ZTI_NMODES 103} zti_modes_t; 104 105#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 106#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 107#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 108 109#define ZTI_N(n) ZTI_P(n, 1) 110#define ZTI_ONE ZTI_N(1) 111 112typedef struct zio_taskq_info { 113 zti_modes_t zti_mode; 114 uint_t zti_value; 115 uint_t zti_count; 116} zio_taskq_info_t; 117 118static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 119 "issue", "issue_high", "intr", "intr_high" 120}; 121 122/* 123 * This table defines the taskq settings for each ZFS I/O type. When 124 * initializing a pool, we use this table to create an appropriately sized 125 * taskq. Some operations are low volume and therefore have a small, static 126 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 127 * macros. Other operations process a large amount of data; the ZTI_BATCH 128 * macro causes us to create a taskq oriented for throughput. Some operations 129 * are so high frequency and short-lived that the taskq itself can become a a 130 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 131 * additional degree of parallelism specified by the number of threads per- 132 * taskq and the number of taskqs; when dispatching an event in this case, the 133 * particular taskq is chosen at random. 134 * 135 * The different taskq priorities are to handle the different contexts (issue 136 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 137 * need to be handled with minimum delay. 138 */ 139const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 140 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 141 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 142 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */ 143 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 144 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 146 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 147}; 148 149static void spa_sync_version(void *arg, dmu_tx_t *tx); 150static void spa_sync_props(void *arg, dmu_tx_t *tx); 151static boolean_t spa_has_active_shared_spare(spa_t *spa); 152static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 153 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 154 char **ereport); 155static void spa_vdev_resilver_done(spa_t *spa); 156 157uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 158#ifdef PSRSET_BIND 159id_t zio_taskq_psrset_bind = PS_NONE; 160#endif 161#ifdef SYSDC 162boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 163#endif 164uint_t zio_taskq_basedc = 80; /* base duty cycle */ 165 166boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 167extern int zfs_sync_pass_deferred_free; 168 169#ifndef illumos 170extern void spa_deadman(void *arg); 171#endif 172 173/* 174 * This (illegal) pool name is used when temporarily importing a spa_t in order 175 * to get the vdev stats associated with the imported devices. 176 */ 177#define TRYIMPORT_NAME "$import" 178 179/* 180 * ========================================================================== 181 * SPA properties routines 182 * ========================================================================== 183 */ 184 185/* 186 * Add a (source=src, propname=propval) list to an nvlist. 187 */ 188static void 189spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 190 uint64_t intval, zprop_source_t src) 191{ 192 const char *propname = zpool_prop_to_name(prop); 193 nvlist_t *propval; 194 195 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 196 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 197 198 if (strval != NULL) 199 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 200 else 201 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 202 203 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 204 nvlist_free(propval); 205} 206 207/* 208 * Get property values from the spa configuration. 209 */ 210static void 211spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 212{ 213 vdev_t *rvd = spa->spa_root_vdev; 214 dsl_pool_t *pool = spa->spa_dsl_pool; 215 uint64_t size, alloc, cap, version; 216 zprop_source_t src = ZPROP_SRC_NONE; 217 spa_config_dirent_t *dp; 218 metaslab_class_t *mc = spa_normal_class(spa); 219 220 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 221 222 if (rvd != NULL) { 223 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 224 size = metaslab_class_get_space(spa_normal_class(spa)); 225 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 226 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 227 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 228 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 229 size - alloc, src); 230 231 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 232 metaslab_class_fragmentation(mc), src); 233 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 234 metaslab_class_expandable_space(mc), src); 235 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 236 (spa_mode(spa) == FREAD), src); 237 238 cap = (size == 0) ? 0 : (alloc * 100 / size); 239 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 240 241 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 242 ddt_get_pool_dedup_ratio(spa), src); 243 244 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 245 rvd->vdev_state, src); 246 247 version = spa_version(spa); 248 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 249 src = ZPROP_SRC_DEFAULT; 250 else 251 src = ZPROP_SRC_LOCAL; 252 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 253 } 254 255 if (pool != NULL) { 256 /* 257 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 258 * when opening pools before this version freedir will be NULL. 259 */ 260 if (pool->dp_free_dir != NULL) { 261 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 262 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 263 src); 264 } else { 265 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 266 NULL, 0, src); 267 } 268 269 if (pool->dp_leak_dir != NULL) { 270 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 271 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 272 src); 273 } else { 274 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 275 NULL, 0, src); 276 } 277 } 278 279 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 280 281 if (spa->spa_comment != NULL) { 282 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 283 0, ZPROP_SRC_LOCAL); 284 } 285 286 if (spa->spa_root != NULL) 287 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 288 0, ZPROP_SRC_LOCAL); 289 290 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 291 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 292 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 293 } else { 294 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 295 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 296 } 297 298 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 299 if (dp->scd_path == NULL) { 300 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 301 "none", 0, ZPROP_SRC_LOCAL); 302 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 303 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 304 dp->scd_path, 0, ZPROP_SRC_LOCAL); 305 } 306 } 307} 308 309/* 310 * Get zpool property values. 311 */ 312int 313spa_prop_get(spa_t *spa, nvlist_t **nvp) 314{ 315 objset_t *mos = spa->spa_meta_objset; 316 zap_cursor_t zc; 317 zap_attribute_t za; 318 int err; 319 320 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 321 322 mutex_enter(&spa->spa_props_lock); 323 324 /* 325 * Get properties from the spa config. 326 */ 327 spa_prop_get_config(spa, nvp); 328 329 /* If no pool property object, no more prop to get. */ 330 if (mos == NULL || spa->spa_pool_props_object == 0) { 331 mutex_exit(&spa->spa_props_lock); 332 return (0); 333 } 334 335 /* 336 * Get properties from the MOS pool property object. 337 */ 338 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 339 (err = zap_cursor_retrieve(&zc, &za)) == 0; 340 zap_cursor_advance(&zc)) { 341 uint64_t intval = 0; 342 char *strval = NULL; 343 zprop_source_t src = ZPROP_SRC_DEFAULT; 344 zpool_prop_t prop; 345 346 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 347 continue; 348 349 switch (za.za_integer_length) { 350 case 8: 351 /* integer property */ 352 if (za.za_first_integer != 353 zpool_prop_default_numeric(prop)) 354 src = ZPROP_SRC_LOCAL; 355 356 if (prop == ZPOOL_PROP_BOOTFS) { 357 dsl_pool_t *dp; 358 dsl_dataset_t *ds = NULL; 359 360 dp = spa_get_dsl(spa); 361 dsl_pool_config_enter(dp, FTAG); 362 if (err = dsl_dataset_hold_obj(dp, 363 za.za_first_integer, FTAG, &ds)) { 364 dsl_pool_config_exit(dp, FTAG); 365 break; 366 } 367 368 strval = kmem_alloc( 369 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 370 KM_SLEEP); 371 dsl_dataset_name(ds, strval); 372 dsl_dataset_rele(ds, FTAG); 373 dsl_pool_config_exit(dp, FTAG); 374 } else { 375 strval = NULL; 376 intval = za.za_first_integer; 377 } 378 379 spa_prop_add_list(*nvp, prop, strval, intval, src); 380 381 if (strval != NULL) 382 kmem_free(strval, 383 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 384 385 break; 386 387 case 1: 388 /* string property */ 389 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 390 err = zap_lookup(mos, spa->spa_pool_props_object, 391 za.za_name, 1, za.za_num_integers, strval); 392 if (err) { 393 kmem_free(strval, za.za_num_integers); 394 break; 395 } 396 spa_prop_add_list(*nvp, prop, strval, 0, src); 397 kmem_free(strval, za.za_num_integers); 398 break; 399 400 default: 401 break; 402 } 403 } 404 zap_cursor_fini(&zc); 405 mutex_exit(&spa->spa_props_lock); 406out: 407 if (err && err != ENOENT) { 408 nvlist_free(*nvp); 409 *nvp = NULL; 410 return (err); 411 } 412 413 return (0); 414} 415 416/* 417 * Validate the given pool properties nvlist and modify the list 418 * for the property values to be set. 419 */ 420static int 421spa_prop_validate(spa_t *spa, nvlist_t *props) 422{ 423 nvpair_t *elem; 424 int error = 0, reset_bootfs = 0; 425 uint64_t objnum = 0; 426 boolean_t has_feature = B_FALSE; 427 428 elem = NULL; 429 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 430 uint64_t intval; 431 char *strval, *slash, *check, *fname; 432 const char *propname = nvpair_name(elem); 433 zpool_prop_t prop = zpool_name_to_prop(propname); 434 435 switch (prop) { 436 case ZPROP_INVAL: 437 if (!zpool_prop_feature(propname)) { 438 error = SET_ERROR(EINVAL); 439 break; 440 } 441 442 /* 443 * Sanitize the input. 444 */ 445 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 446 error = SET_ERROR(EINVAL); 447 break; 448 } 449 450 if (nvpair_value_uint64(elem, &intval) != 0) { 451 error = SET_ERROR(EINVAL); 452 break; 453 } 454 455 if (intval != 0) { 456 error = SET_ERROR(EINVAL); 457 break; 458 } 459 460 fname = strchr(propname, '@') + 1; 461 if (zfeature_lookup_name(fname, NULL) != 0) { 462 error = SET_ERROR(EINVAL); 463 break; 464 } 465 466 has_feature = B_TRUE; 467 break; 468 469 case ZPOOL_PROP_VERSION: 470 error = nvpair_value_uint64(elem, &intval); 471 if (!error && 472 (intval < spa_version(spa) || 473 intval > SPA_VERSION_BEFORE_FEATURES || 474 has_feature)) 475 error = SET_ERROR(EINVAL); 476 break; 477 478 case ZPOOL_PROP_DELEGATION: 479 case ZPOOL_PROP_AUTOREPLACE: 480 case ZPOOL_PROP_LISTSNAPS: 481 case ZPOOL_PROP_AUTOEXPAND: 482 error = nvpair_value_uint64(elem, &intval); 483 if (!error && intval > 1) 484 error = SET_ERROR(EINVAL); 485 break; 486 487 case ZPOOL_PROP_BOOTFS: 488 /* 489 * If the pool version is less than SPA_VERSION_BOOTFS, 490 * or the pool is still being created (version == 0), 491 * the bootfs property cannot be set. 492 */ 493 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 494 error = SET_ERROR(ENOTSUP); 495 break; 496 } 497 498 /* 499 * Make sure the vdev config is bootable 500 */ 501 if (!vdev_is_bootable(spa->spa_root_vdev)) { 502 error = SET_ERROR(ENOTSUP); 503 break; 504 } 505 506 reset_bootfs = 1; 507 508 error = nvpair_value_string(elem, &strval); 509 510 if (!error) { 511 objset_t *os; 512 uint64_t propval; 513 514 if (strval == NULL || strval[0] == '\0') { 515 objnum = zpool_prop_default_numeric( 516 ZPOOL_PROP_BOOTFS); 517 break; 518 } 519 520 if (error = dmu_objset_hold(strval, FTAG, &os)) 521 break; 522 523 /* 524 * Must be ZPL, and its property settings 525 * must be supported by GRUB (compression 526 * is not gzip, and large blocks are not used). 527 */ 528 529 if (dmu_objset_type(os) != DMU_OST_ZFS) { 530 error = SET_ERROR(ENOTSUP); 531 } else if ((error = 532 dsl_prop_get_int_ds(dmu_objset_ds(os), 533 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 534 &propval)) == 0 && 535 !BOOTFS_COMPRESS_VALID(propval)) { 536 error = SET_ERROR(ENOTSUP); 537 } else if ((error = 538 dsl_prop_get_int_ds(dmu_objset_ds(os), 539 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 540 &propval)) == 0 && 541 propval > SPA_OLD_MAXBLOCKSIZE) { 542 error = SET_ERROR(ENOTSUP); 543 } else { 544 objnum = dmu_objset_id(os); 545 } 546 dmu_objset_rele(os, FTAG); 547 } 548 break; 549 550 case ZPOOL_PROP_FAILUREMODE: 551 error = nvpair_value_uint64(elem, &intval); 552 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 553 intval > ZIO_FAILURE_MODE_PANIC)) 554 error = SET_ERROR(EINVAL); 555 556 /* 557 * This is a special case which only occurs when 558 * the pool has completely failed. This allows 559 * the user to change the in-core failmode property 560 * without syncing it out to disk (I/Os might 561 * currently be blocked). We do this by returning 562 * EIO to the caller (spa_prop_set) to trick it 563 * into thinking we encountered a property validation 564 * error. 565 */ 566 if (!error && spa_suspended(spa)) { 567 spa->spa_failmode = intval; 568 error = SET_ERROR(EIO); 569 } 570 break; 571 572 case ZPOOL_PROP_CACHEFILE: 573 if ((error = nvpair_value_string(elem, &strval)) != 0) 574 break; 575 576 if (strval[0] == '\0') 577 break; 578 579 if (strcmp(strval, "none") == 0) 580 break; 581 582 if (strval[0] != '/') { 583 error = SET_ERROR(EINVAL); 584 break; 585 } 586 587 slash = strrchr(strval, '/'); 588 ASSERT(slash != NULL); 589 590 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 591 strcmp(slash, "/..") == 0) 592 error = SET_ERROR(EINVAL); 593 break; 594 595 case ZPOOL_PROP_COMMENT: 596 if ((error = nvpair_value_string(elem, &strval)) != 0) 597 break; 598 for (check = strval; *check != '\0'; check++) { 599 /* 600 * The kernel doesn't have an easy isprint() 601 * check. For this kernel check, we merely 602 * check ASCII apart from DEL. Fix this if 603 * there is an easy-to-use kernel isprint(). 604 */ 605 if (*check >= 0x7f) { 606 error = SET_ERROR(EINVAL); 607 break; 608 } 609 check++; 610 } 611 if (strlen(strval) > ZPROP_MAX_COMMENT) 612 error = E2BIG; 613 break; 614 615 case ZPOOL_PROP_DEDUPDITTO: 616 if (spa_version(spa) < SPA_VERSION_DEDUP) 617 error = SET_ERROR(ENOTSUP); 618 else 619 error = nvpair_value_uint64(elem, &intval); 620 if (error == 0 && 621 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 622 error = SET_ERROR(EINVAL); 623 break; 624 } 625 626 if (error) 627 break; 628 } 629 630 if (!error && reset_bootfs) { 631 error = nvlist_remove(props, 632 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 633 634 if (!error) { 635 error = nvlist_add_uint64(props, 636 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 637 } 638 } 639 640 return (error); 641} 642 643void 644spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 645{ 646 char *cachefile; 647 spa_config_dirent_t *dp; 648 649 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 650 &cachefile) != 0) 651 return; 652 653 dp = kmem_alloc(sizeof (spa_config_dirent_t), 654 KM_SLEEP); 655 656 if (cachefile[0] == '\0') 657 dp->scd_path = spa_strdup(spa_config_path); 658 else if (strcmp(cachefile, "none") == 0) 659 dp->scd_path = NULL; 660 else 661 dp->scd_path = spa_strdup(cachefile); 662 663 list_insert_head(&spa->spa_config_list, dp); 664 if (need_sync) 665 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 666} 667 668int 669spa_prop_set(spa_t *spa, nvlist_t *nvp) 670{ 671 int error; 672 nvpair_t *elem = NULL; 673 boolean_t need_sync = B_FALSE; 674 675 if ((error = spa_prop_validate(spa, nvp)) != 0) 676 return (error); 677 678 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 679 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 680 681 if (prop == ZPOOL_PROP_CACHEFILE || 682 prop == ZPOOL_PROP_ALTROOT || 683 prop == ZPOOL_PROP_READONLY) 684 continue; 685 686 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 687 uint64_t ver; 688 689 if (prop == ZPOOL_PROP_VERSION) { 690 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 691 } else { 692 ASSERT(zpool_prop_feature(nvpair_name(elem))); 693 ver = SPA_VERSION_FEATURES; 694 need_sync = B_TRUE; 695 } 696 697 /* Save time if the version is already set. */ 698 if (ver == spa_version(spa)) 699 continue; 700 701 /* 702 * In addition to the pool directory object, we might 703 * create the pool properties object, the features for 704 * read object, the features for write object, or the 705 * feature descriptions object. 706 */ 707 error = dsl_sync_task(spa->spa_name, NULL, 708 spa_sync_version, &ver, 709 6, ZFS_SPACE_CHECK_RESERVED); 710 if (error) 711 return (error); 712 continue; 713 } 714 715 need_sync = B_TRUE; 716 break; 717 } 718 719 if (need_sync) { 720 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 721 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 722 } 723 724 return (0); 725} 726 727/* 728 * If the bootfs property value is dsobj, clear it. 729 */ 730void 731spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 732{ 733 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 734 VERIFY(zap_remove(spa->spa_meta_objset, 735 spa->spa_pool_props_object, 736 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 737 spa->spa_bootfs = 0; 738 } 739} 740 741/*ARGSUSED*/ 742static int 743spa_change_guid_check(void *arg, dmu_tx_t *tx) 744{ 745 uint64_t *newguid = arg; 746 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 747 vdev_t *rvd = spa->spa_root_vdev; 748 uint64_t vdev_state; 749 750 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 751 vdev_state = rvd->vdev_state; 752 spa_config_exit(spa, SCL_STATE, FTAG); 753 754 if (vdev_state != VDEV_STATE_HEALTHY) 755 return (SET_ERROR(ENXIO)); 756 757 ASSERT3U(spa_guid(spa), !=, *newguid); 758 759 return (0); 760} 761 762static void 763spa_change_guid_sync(void *arg, dmu_tx_t *tx) 764{ 765 uint64_t *newguid = arg; 766 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 767 uint64_t oldguid; 768 vdev_t *rvd = spa->spa_root_vdev; 769 770 oldguid = spa_guid(spa); 771 772 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 773 rvd->vdev_guid = *newguid; 774 rvd->vdev_guid_sum += (*newguid - oldguid); 775 vdev_config_dirty(rvd); 776 spa_config_exit(spa, SCL_STATE, FTAG); 777 778 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 779 oldguid, *newguid); 780} 781 782/* 783 * Change the GUID for the pool. This is done so that we can later 784 * re-import a pool built from a clone of our own vdevs. We will modify 785 * the root vdev's guid, our own pool guid, and then mark all of our 786 * vdevs dirty. Note that we must make sure that all our vdevs are 787 * online when we do this, or else any vdevs that weren't present 788 * would be orphaned from our pool. We are also going to issue a 789 * sysevent to update any watchers. 790 */ 791int 792spa_change_guid(spa_t *spa) 793{ 794 int error; 795 uint64_t guid; 796 797 mutex_enter(&spa->spa_vdev_top_lock); 798 mutex_enter(&spa_namespace_lock); 799 guid = spa_generate_guid(NULL); 800 801 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 802 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 803 804 if (error == 0) { 805 spa_config_sync(spa, B_FALSE, B_TRUE); 806 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 807 } 808 809 mutex_exit(&spa_namespace_lock); 810 mutex_exit(&spa->spa_vdev_top_lock); 811 812 return (error); 813} 814 815/* 816 * ========================================================================== 817 * SPA state manipulation (open/create/destroy/import/export) 818 * ========================================================================== 819 */ 820 821static int 822spa_error_entry_compare(const void *a, const void *b) 823{ 824 spa_error_entry_t *sa = (spa_error_entry_t *)a; 825 spa_error_entry_t *sb = (spa_error_entry_t *)b; 826 int ret; 827 828 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 829 sizeof (zbookmark_phys_t)); 830 831 if (ret < 0) 832 return (-1); 833 else if (ret > 0) 834 return (1); 835 else 836 return (0); 837} 838 839/* 840 * Utility function which retrieves copies of the current logs and 841 * re-initializes them in the process. 842 */ 843void 844spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 845{ 846 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 847 848 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 849 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 850 851 avl_create(&spa->spa_errlist_scrub, 852 spa_error_entry_compare, sizeof (spa_error_entry_t), 853 offsetof(spa_error_entry_t, se_avl)); 854 avl_create(&spa->spa_errlist_last, 855 spa_error_entry_compare, sizeof (spa_error_entry_t), 856 offsetof(spa_error_entry_t, se_avl)); 857} 858 859static void 860spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 861{ 862 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 863 enum zti_modes mode = ztip->zti_mode; 864 uint_t value = ztip->zti_value; 865 uint_t count = ztip->zti_count; 866 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 867 char name[32]; 868 uint_t flags = 0; 869 boolean_t batch = B_FALSE; 870 871 if (mode == ZTI_MODE_NULL) { 872 tqs->stqs_count = 0; 873 tqs->stqs_taskq = NULL; 874 return; 875 } 876 877 ASSERT3U(count, >, 0); 878 879 tqs->stqs_count = count; 880 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 881 882 switch (mode) { 883 case ZTI_MODE_FIXED: 884 ASSERT3U(value, >=, 1); 885 value = MAX(value, 1); 886 break; 887 888 case ZTI_MODE_BATCH: 889 batch = B_TRUE; 890 flags |= TASKQ_THREADS_CPU_PCT; 891 value = zio_taskq_batch_pct; 892 break; 893 894 default: 895 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 896 "spa_activate()", 897 zio_type_name[t], zio_taskq_types[q], mode, value); 898 break; 899 } 900 901 for (uint_t i = 0; i < count; i++) { 902 taskq_t *tq; 903 904 if (count > 1) { 905 (void) snprintf(name, sizeof (name), "%s_%s_%u", 906 zio_type_name[t], zio_taskq_types[q], i); 907 } else { 908 (void) snprintf(name, sizeof (name), "%s_%s", 909 zio_type_name[t], zio_taskq_types[q]); 910 } 911 912#ifdef SYSDC 913 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 914 if (batch) 915 flags |= TASKQ_DC_BATCH; 916 917 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 918 spa->spa_proc, zio_taskq_basedc, flags); 919 } else { 920#endif 921 pri_t pri = maxclsyspri; 922 /* 923 * The write issue taskq can be extremely CPU 924 * intensive. Run it at slightly lower priority 925 * than the other taskqs. 926 */ 927 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 928 pri--; 929 930 tq = taskq_create_proc(name, value, pri, 50, 931 INT_MAX, spa->spa_proc, flags); 932#ifdef SYSDC 933 } 934#endif 935 936 tqs->stqs_taskq[i] = tq; 937 } 938} 939 940static void 941spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 942{ 943 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 944 945 if (tqs->stqs_taskq == NULL) { 946 ASSERT0(tqs->stqs_count); 947 return; 948 } 949 950 for (uint_t i = 0; i < tqs->stqs_count; i++) { 951 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 952 taskq_destroy(tqs->stqs_taskq[i]); 953 } 954 955 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 956 tqs->stqs_taskq = NULL; 957} 958 959/* 960 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 961 * Note that a type may have multiple discrete taskqs to avoid lock contention 962 * on the taskq itself. In that case we choose which taskq at random by using 963 * the low bits of gethrtime(). 964 */ 965void 966spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 967 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 968{ 969 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 970 taskq_t *tq; 971 972 ASSERT3P(tqs->stqs_taskq, !=, NULL); 973 ASSERT3U(tqs->stqs_count, !=, 0); 974 975 if (tqs->stqs_count == 1) { 976 tq = tqs->stqs_taskq[0]; 977 } else { 978#ifdef _KERNEL 979 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 980#else 981 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 982#endif 983 } 984 985 taskq_dispatch_ent(tq, func, arg, flags, ent); 986} 987 988static void 989spa_create_zio_taskqs(spa_t *spa) 990{ 991 for (int t = 0; t < ZIO_TYPES; t++) { 992 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 993 spa_taskqs_init(spa, t, q); 994 } 995 } 996} 997 998#ifdef _KERNEL 999#ifdef SPA_PROCESS 1000static void 1001spa_thread(void *arg) 1002{ 1003 callb_cpr_t cprinfo; 1004 1005 spa_t *spa = arg; 1006 user_t *pu = PTOU(curproc); 1007 1008 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1009 spa->spa_name); 1010 1011 ASSERT(curproc != &p0); 1012 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1013 "zpool-%s", spa->spa_name); 1014 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1015 1016#ifdef PSRSET_BIND 1017 /* bind this thread to the requested psrset */ 1018 if (zio_taskq_psrset_bind != PS_NONE) { 1019 pool_lock(); 1020 mutex_enter(&cpu_lock); 1021 mutex_enter(&pidlock); 1022 mutex_enter(&curproc->p_lock); 1023 1024 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1025 0, NULL, NULL) == 0) { 1026 curthread->t_bind_pset = zio_taskq_psrset_bind; 1027 } else { 1028 cmn_err(CE_WARN, 1029 "Couldn't bind process for zfs pool \"%s\" to " 1030 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1031 } 1032 1033 mutex_exit(&curproc->p_lock); 1034 mutex_exit(&pidlock); 1035 mutex_exit(&cpu_lock); 1036 pool_unlock(); 1037 } 1038#endif 1039 1040#ifdef SYSDC 1041 if (zio_taskq_sysdc) { 1042 sysdc_thread_enter(curthread, 100, 0); 1043 } 1044#endif 1045 1046 spa->spa_proc = curproc; 1047 spa->spa_did = curthread->t_did; 1048 1049 spa_create_zio_taskqs(spa); 1050 1051 mutex_enter(&spa->spa_proc_lock); 1052 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1053 1054 spa->spa_proc_state = SPA_PROC_ACTIVE; 1055 cv_broadcast(&spa->spa_proc_cv); 1056 1057 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1058 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1059 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1060 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1061 1062 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1063 spa->spa_proc_state = SPA_PROC_GONE; 1064 spa->spa_proc = &p0; 1065 cv_broadcast(&spa->spa_proc_cv); 1066 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1067 1068 mutex_enter(&curproc->p_lock); 1069 lwp_exit(); 1070} 1071#endif /* SPA_PROCESS */ 1072#endif 1073 1074/* 1075 * Activate an uninitialized pool. 1076 */ 1077static void 1078spa_activate(spa_t *spa, int mode) 1079{ 1080 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1081 1082 spa->spa_state = POOL_STATE_ACTIVE; 1083 spa->spa_mode = mode; 1084 1085 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1086 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1087 1088 /* Try to create a covering process */ 1089 mutex_enter(&spa->spa_proc_lock); 1090 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1091 ASSERT(spa->spa_proc == &p0); 1092 spa->spa_did = 0; 1093 1094#ifdef SPA_PROCESS 1095 /* Only create a process if we're going to be around a while. */ 1096 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1097 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1098 NULL, 0) == 0) { 1099 spa->spa_proc_state = SPA_PROC_CREATED; 1100 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1101 cv_wait(&spa->spa_proc_cv, 1102 &spa->spa_proc_lock); 1103 } 1104 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1105 ASSERT(spa->spa_proc != &p0); 1106 ASSERT(spa->spa_did != 0); 1107 } else { 1108#ifdef _KERNEL 1109 cmn_err(CE_WARN, 1110 "Couldn't create process for zfs pool \"%s\"\n", 1111 spa->spa_name); 1112#endif 1113 } 1114 } 1115#endif /* SPA_PROCESS */ 1116 mutex_exit(&spa->spa_proc_lock); 1117 1118 /* If we didn't create a process, we need to create our taskqs. */ 1119 ASSERT(spa->spa_proc == &p0); 1120 if (spa->spa_proc == &p0) { 1121 spa_create_zio_taskqs(spa); 1122 } 1123 1124 /* 1125 * Start TRIM thread. 1126 */ 1127 trim_thread_create(spa); 1128 1129 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1130 offsetof(vdev_t, vdev_config_dirty_node)); 1131 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1132 offsetof(objset_t, os_evicting_node)); 1133 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1134 offsetof(vdev_t, vdev_state_dirty_node)); 1135 1136 txg_list_create(&spa->spa_vdev_txg_list, 1137 offsetof(struct vdev, vdev_txg_node)); 1138 1139 avl_create(&spa->spa_errlist_scrub, 1140 spa_error_entry_compare, sizeof (spa_error_entry_t), 1141 offsetof(spa_error_entry_t, se_avl)); 1142 avl_create(&spa->spa_errlist_last, 1143 spa_error_entry_compare, sizeof (spa_error_entry_t), 1144 offsetof(spa_error_entry_t, se_avl)); 1145} 1146 1147/* 1148 * Opposite of spa_activate(). 1149 */ 1150static void 1151spa_deactivate(spa_t *spa) 1152{ 1153 ASSERT(spa->spa_sync_on == B_FALSE); 1154 ASSERT(spa->spa_dsl_pool == NULL); 1155 ASSERT(spa->spa_root_vdev == NULL); 1156 ASSERT(spa->spa_async_zio_root == NULL); 1157 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1158 1159 /* 1160 * Stop TRIM thread in case spa_unload() wasn't called directly 1161 * before spa_deactivate(). 1162 */ 1163 trim_thread_destroy(spa); 1164 1165 spa_evicting_os_wait(spa); 1166 1167 txg_list_destroy(&spa->spa_vdev_txg_list); 1168 1169 list_destroy(&spa->spa_config_dirty_list); 1170 list_destroy(&spa->spa_evicting_os_list); 1171 list_destroy(&spa->spa_state_dirty_list); 1172 1173 for (int t = 0; t < ZIO_TYPES; t++) { 1174 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1175 spa_taskqs_fini(spa, t, q); 1176 } 1177 } 1178 1179 metaslab_class_destroy(spa->spa_normal_class); 1180 spa->spa_normal_class = NULL; 1181 1182 metaslab_class_destroy(spa->spa_log_class); 1183 spa->spa_log_class = NULL; 1184 1185 /* 1186 * If this was part of an import or the open otherwise failed, we may 1187 * still have errors left in the queues. Empty them just in case. 1188 */ 1189 spa_errlog_drain(spa); 1190 1191 avl_destroy(&spa->spa_errlist_scrub); 1192 avl_destroy(&spa->spa_errlist_last); 1193 1194 spa->spa_state = POOL_STATE_UNINITIALIZED; 1195 1196 mutex_enter(&spa->spa_proc_lock); 1197 if (spa->spa_proc_state != SPA_PROC_NONE) { 1198 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1199 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1200 cv_broadcast(&spa->spa_proc_cv); 1201 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1202 ASSERT(spa->spa_proc != &p0); 1203 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1204 } 1205 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1206 spa->spa_proc_state = SPA_PROC_NONE; 1207 } 1208 ASSERT(spa->spa_proc == &p0); 1209 mutex_exit(&spa->spa_proc_lock); 1210 1211#ifdef SPA_PROCESS 1212 /* 1213 * We want to make sure spa_thread() has actually exited the ZFS 1214 * module, so that the module can't be unloaded out from underneath 1215 * it. 1216 */ 1217 if (spa->spa_did != 0) { 1218 thread_join(spa->spa_did); 1219 spa->spa_did = 0; 1220 } 1221#endif /* SPA_PROCESS */ 1222} 1223 1224/* 1225 * Verify a pool configuration, and construct the vdev tree appropriately. This 1226 * will create all the necessary vdevs in the appropriate layout, with each vdev 1227 * in the CLOSED state. This will prep the pool before open/creation/import. 1228 * All vdev validation is done by the vdev_alloc() routine. 1229 */ 1230static int 1231spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1232 uint_t id, int atype) 1233{ 1234 nvlist_t **child; 1235 uint_t children; 1236 int error; 1237 1238 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1239 return (error); 1240 1241 if ((*vdp)->vdev_ops->vdev_op_leaf) 1242 return (0); 1243 1244 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1245 &child, &children); 1246 1247 if (error == ENOENT) 1248 return (0); 1249 1250 if (error) { 1251 vdev_free(*vdp); 1252 *vdp = NULL; 1253 return (SET_ERROR(EINVAL)); 1254 } 1255 1256 for (int c = 0; c < children; c++) { 1257 vdev_t *vd; 1258 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1259 atype)) != 0) { 1260 vdev_free(*vdp); 1261 *vdp = NULL; 1262 return (error); 1263 } 1264 } 1265 1266 ASSERT(*vdp != NULL); 1267 1268 return (0); 1269} 1270 1271/* 1272 * Opposite of spa_load(). 1273 */ 1274static void 1275spa_unload(spa_t *spa) 1276{ 1277 int i; 1278 1279 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1280 1281 /* 1282 * Stop TRIM thread. 1283 */ 1284 trim_thread_destroy(spa); 1285 1286 /* 1287 * Stop async tasks. 1288 */ 1289 spa_async_suspend(spa); 1290 1291 /* 1292 * Stop syncing. 1293 */ 1294 if (spa->spa_sync_on) { 1295 txg_sync_stop(spa->spa_dsl_pool); 1296 spa->spa_sync_on = B_FALSE; 1297 } 1298 1299 /* 1300 * Wait for any outstanding async I/O to complete. 1301 */ 1302 if (spa->spa_async_zio_root != NULL) { 1303 for (int i = 0; i < max_ncpus; i++) 1304 (void) zio_wait(spa->spa_async_zio_root[i]); 1305 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1306 spa->spa_async_zio_root = NULL; 1307 } 1308 1309 bpobj_close(&spa->spa_deferred_bpobj); 1310 1311 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1312 1313 /* 1314 * Close all vdevs. 1315 */ 1316 if (spa->spa_root_vdev) 1317 vdev_free(spa->spa_root_vdev); 1318 ASSERT(spa->spa_root_vdev == NULL); 1319 1320 /* 1321 * Close the dsl pool. 1322 */ 1323 if (spa->spa_dsl_pool) { 1324 dsl_pool_close(spa->spa_dsl_pool); 1325 spa->spa_dsl_pool = NULL; 1326 spa->spa_meta_objset = NULL; 1327 } 1328 1329 ddt_unload(spa); 1330 1331 1332 /* 1333 * Drop and purge level 2 cache 1334 */ 1335 spa_l2cache_drop(spa); 1336 1337 for (i = 0; i < spa->spa_spares.sav_count; i++) 1338 vdev_free(spa->spa_spares.sav_vdevs[i]); 1339 if (spa->spa_spares.sav_vdevs) { 1340 kmem_free(spa->spa_spares.sav_vdevs, 1341 spa->spa_spares.sav_count * sizeof (void *)); 1342 spa->spa_spares.sav_vdevs = NULL; 1343 } 1344 if (spa->spa_spares.sav_config) { 1345 nvlist_free(spa->spa_spares.sav_config); 1346 spa->spa_spares.sav_config = NULL; 1347 } 1348 spa->spa_spares.sav_count = 0; 1349 1350 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1351 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1352 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1353 } 1354 if (spa->spa_l2cache.sav_vdevs) { 1355 kmem_free(spa->spa_l2cache.sav_vdevs, 1356 spa->spa_l2cache.sav_count * sizeof (void *)); 1357 spa->spa_l2cache.sav_vdevs = NULL; 1358 } 1359 if (spa->spa_l2cache.sav_config) { 1360 nvlist_free(spa->spa_l2cache.sav_config); 1361 spa->spa_l2cache.sav_config = NULL; 1362 } 1363 spa->spa_l2cache.sav_count = 0; 1364 1365 spa->spa_async_suspended = 0; 1366 1367 if (spa->spa_comment != NULL) { 1368 spa_strfree(spa->spa_comment); 1369 spa->spa_comment = NULL; 1370 } 1371 1372 spa_config_exit(spa, SCL_ALL, FTAG); 1373} 1374 1375/* 1376 * Load (or re-load) the current list of vdevs describing the active spares for 1377 * this pool. When this is called, we have some form of basic information in 1378 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1379 * then re-generate a more complete list including status information. 1380 */ 1381static void 1382spa_load_spares(spa_t *spa) 1383{ 1384 nvlist_t **spares; 1385 uint_t nspares; 1386 int i; 1387 vdev_t *vd, *tvd; 1388 1389 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1390 1391 /* 1392 * First, close and free any existing spare vdevs. 1393 */ 1394 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1395 vd = spa->spa_spares.sav_vdevs[i]; 1396 1397 /* Undo the call to spa_activate() below */ 1398 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1399 B_FALSE)) != NULL && tvd->vdev_isspare) 1400 spa_spare_remove(tvd); 1401 vdev_close(vd); 1402 vdev_free(vd); 1403 } 1404 1405 if (spa->spa_spares.sav_vdevs) 1406 kmem_free(spa->spa_spares.sav_vdevs, 1407 spa->spa_spares.sav_count * sizeof (void *)); 1408 1409 if (spa->spa_spares.sav_config == NULL) 1410 nspares = 0; 1411 else 1412 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1413 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1414 1415 spa->spa_spares.sav_count = (int)nspares; 1416 spa->spa_spares.sav_vdevs = NULL; 1417 1418 if (nspares == 0) 1419 return; 1420 1421 /* 1422 * Construct the array of vdevs, opening them to get status in the 1423 * process. For each spare, there is potentially two different vdev_t 1424 * structures associated with it: one in the list of spares (used only 1425 * for basic validation purposes) and one in the active vdev 1426 * configuration (if it's spared in). During this phase we open and 1427 * validate each vdev on the spare list. If the vdev also exists in the 1428 * active configuration, then we also mark this vdev as an active spare. 1429 */ 1430 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1431 KM_SLEEP); 1432 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1433 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1434 VDEV_ALLOC_SPARE) == 0); 1435 ASSERT(vd != NULL); 1436 1437 spa->spa_spares.sav_vdevs[i] = vd; 1438 1439 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1440 B_FALSE)) != NULL) { 1441 if (!tvd->vdev_isspare) 1442 spa_spare_add(tvd); 1443 1444 /* 1445 * We only mark the spare active if we were successfully 1446 * able to load the vdev. Otherwise, importing a pool 1447 * with a bad active spare would result in strange 1448 * behavior, because multiple pool would think the spare 1449 * is actively in use. 1450 * 1451 * There is a vulnerability here to an equally bizarre 1452 * circumstance, where a dead active spare is later 1453 * brought back to life (onlined or otherwise). Given 1454 * the rarity of this scenario, and the extra complexity 1455 * it adds, we ignore the possibility. 1456 */ 1457 if (!vdev_is_dead(tvd)) 1458 spa_spare_activate(tvd); 1459 } 1460 1461 vd->vdev_top = vd; 1462 vd->vdev_aux = &spa->spa_spares; 1463 1464 if (vdev_open(vd) != 0) 1465 continue; 1466 1467 if (vdev_validate_aux(vd) == 0) 1468 spa_spare_add(vd); 1469 } 1470 1471 /* 1472 * Recompute the stashed list of spares, with status information 1473 * this time. 1474 */ 1475 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1476 DATA_TYPE_NVLIST_ARRAY) == 0); 1477 1478 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1479 KM_SLEEP); 1480 for (i = 0; i < spa->spa_spares.sav_count; i++) 1481 spares[i] = vdev_config_generate(spa, 1482 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1483 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1484 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1485 for (i = 0; i < spa->spa_spares.sav_count; i++) 1486 nvlist_free(spares[i]); 1487 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1488} 1489 1490/* 1491 * Load (or re-load) the current list of vdevs describing the active l2cache for 1492 * this pool. When this is called, we have some form of basic information in 1493 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1494 * then re-generate a more complete list including status information. 1495 * Devices which are already active have their details maintained, and are 1496 * not re-opened. 1497 */ 1498static void 1499spa_load_l2cache(spa_t *spa) 1500{ 1501 nvlist_t **l2cache; 1502 uint_t nl2cache; 1503 int i, j, oldnvdevs; 1504 uint64_t guid; 1505 vdev_t *vd, **oldvdevs, **newvdevs; 1506 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1507 1508 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1509 1510 if (sav->sav_config != NULL) { 1511 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1512 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1513 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1514 } else { 1515 nl2cache = 0; 1516 newvdevs = NULL; 1517 } 1518 1519 oldvdevs = sav->sav_vdevs; 1520 oldnvdevs = sav->sav_count; 1521 sav->sav_vdevs = NULL; 1522 sav->sav_count = 0; 1523 1524 /* 1525 * Process new nvlist of vdevs. 1526 */ 1527 for (i = 0; i < nl2cache; i++) { 1528 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1529 &guid) == 0); 1530 1531 newvdevs[i] = NULL; 1532 for (j = 0; j < oldnvdevs; j++) { 1533 vd = oldvdevs[j]; 1534 if (vd != NULL && guid == vd->vdev_guid) { 1535 /* 1536 * Retain previous vdev for add/remove ops. 1537 */ 1538 newvdevs[i] = vd; 1539 oldvdevs[j] = NULL; 1540 break; 1541 } 1542 } 1543 1544 if (newvdevs[i] == NULL) { 1545 /* 1546 * Create new vdev 1547 */ 1548 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1549 VDEV_ALLOC_L2CACHE) == 0); 1550 ASSERT(vd != NULL); 1551 newvdevs[i] = vd; 1552 1553 /* 1554 * Commit this vdev as an l2cache device, 1555 * even if it fails to open. 1556 */ 1557 spa_l2cache_add(vd); 1558 1559 vd->vdev_top = vd; 1560 vd->vdev_aux = sav; 1561 1562 spa_l2cache_activate(vd); 1563 1564 if (vdev_open(vd) != 0) 1565 continue; 1566 1567 (void) vdev_validate_aux(vd); 1568 1569 if (!vdev_is_dead(vd)) 1570 l2arc_add_vdev(spa, vd); 1571 } 1572 } 1573 1574 /* 1575 * Purge vdevs that were dropped 1576 */ 1577 for (i = 0; i < oldnvdevs; i++) { 1578 uint64_t pool; 1579 1580 vd = oldvdevs[i]; 1581 if (vd != NULL) { 1582 ASSERT(vd->vdev_isl2cache); 1583 1584 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1585 pool != 0ULL && l2arc_vdev_present(vd)) 1586 l2arc_remove_vdev(vd); 1587 vdev_clear_stats(vd); 1588 vdev_free(vd); 1589 } 1590 } 1591 1592 if (oldvdevs) 1593 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1594 1595 if (sav->sav_config == NULL) 1596 goto out; 1597 1598 sav->sav_vdevs = newvdevs; 1599 sav->sav_count = (int)nl2cache; 1600 1601 /* 1602 * Recompute the stashed list of l2cache devices, with status 1603 * information this time. 1604 */ 1605 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1606 DATA_TYPE_NVLIST_ARRAY) == 0); 1607 1608 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1609 for (i = 0; i < sav->sav_count; i++) 1610 l2cache[i] = vdev_config_generate(spa, 1611 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1612 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1613 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1614out: 1615 for (i = 0; i < sav->sav_count; i++) 1616 nvlist_free(l2cache[i]); 1617 if (sav->sav_count) 1618 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1619} 1620 1621static int 1622load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1623{ 1624 dmu_buf_t *db; 1625 char *packed = NULL; 1626 size_t nvsize = 0; 1627 int error; 1628 *value = NULL; 1629 1630 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1631 if (error != 0) 1632 return (error); 1633 nvsize = *(uint64_t *)db->db_data; 1634 dmu_buf_rele(db, FTAG); 1635 1636 packed = kmem_alloc(nvsize, KM_SLEEP); 1637 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1638 DMU_READ_PREFETCH); 1639 if (error == 0) 1640 error = nvlist_unpack(packed, nvsize, value, 0); 1641 kmem_free(packed, nvsize); 1642 1643 return (error); 1644} 1645 1646/* 1647 * Checks to see if the given vdev could not be opened, in which case we post a 1648 * sysevent to notify the autoreplace code that the device has been removed. 1649 */ 1650static void 1651spa_check_removed(vdev_t *vd) 1652{ 1653 for (int c = 0; c < vd->vdev_children; c++) 1654 spa_check_removed(vd->vdev_child[c]); 1655 1656 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1657 !vd->vdev_ishole) { 1658 zfs_post_autoreplace(vd->vdev_spa, vd); 1659 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1660 } 1661} 1662 1663/* 1664 * Validate the current config against the MOS config 1665 */ 1666static boolean_t 1667spa_config_valid(spa_t *spa, nvlist_t *config) 1668{ 1669 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1670 nvlist_t *nv; 1671 1672 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1673 1674 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1675 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1676 1677 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1678 1679 /* 1680 * If we're doing a normal import, then build up any additional 1681 * diagnostic information about missing devices in this config. 1682 * We'll pass this up to the user for further processing. 1683 */ 1684 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1685 nvlist_t **child, *nv; 1686 uint64_t idx = 0; 1687 1688 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1689 KM_SLEEP); 1690 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1691 1692 for (int c = 0; c < rvd->vdev_children; c++) { 1693 vdev_t *tvd = rvd->vdev_child[c]; 1694 vdev_t *mtvd = mrvd->vdev_child[c]; 1695 1696 if (tvd->vdev_ops == &vdev_missing_ops && 1697 mtvd->vdev_ops != &vdev_missing_ops && 1698 mtvd->vdev_islog) 1699 child[idx++] = vdev_config_generate(spa, mtvd, 1700 B_FALSE, 0); 1701 } 1702 1703 if (idx) { 1704 VERIFY(nvlist_add_nvlist_array(nv, 1705 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1706 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1707 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1708 1709 for (int i = 0; i < idx; i++) 1710 nvlist_free(child[i]); 1711 } 1712 nvlist_free(nv); 1713 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1714 } 1715 1716 /* 1717 * Compare the root vdev tree with the information we have 1718 * from the MOS config (mrvd). Check each top-level vdev 1719 * with the corresponding MOS config top-level (mtvd). 1720 */ 1721 for (int c = 0; c < rvd->vdev_children; c++) { 1722 vdev_t *tvd = rvd->vdev_child[c]; 1723 vdev_t *mtvd = mrvd->vdev_child[c]; 1724 1725 /* 1726 * Resolve any "missing" vdevs in the current configuration. 1727 * If we find that the MOS config has more accurate information 1728 * about the top-level vdev then use that vdev instead. 1729 */ 1730 if (tvd->vdev_ops == &vdev_missing_ops && 1731 mtvd->vdev_ops != &vdev_missing_ops) { 1732 1733 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1734 continue; 1735 1736 /* 1737 * Device specific actions. 1738 */ 1739 if (mtvd->vdev_islog) { 1740 spa_set_log_state(spa, SPA_LOG_CLEAR); 1741 } else { 1742 /* 1743 * XXX - once we have 'readonly' pool 1744 * support we should be able to handle 1745 * missing data devices by transitioning 1746 * the pool to readonly. 1747 */ 1748 continue; 1749 } 1750 1751 /* 1752 * Swap the missing vdev with the data we were 1753 * able to obtain from the MOS config. 1754 */ 1755 vdev_remove_child(rvd, tvd); 1756 vdev_remove_child(mrvd, mtvd); 1757 1758 vdev_add_child(rvd, mtvd); 1759 vdev_add_child(mrvd, tvd); 1760 1761 spa_config_exit(spa, SCL_ALL, FTAG); 1762 vdev_load(mtvd); 1763 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1764 1765 vdev_reopen(rvd); 1766 } else if (mtvd->vdev_islog) { 1767 /* 1768 * Load the slog device's state from the MOS config 1769 * since it's possible that the label does not 1770 * contain the most up-to-date information. 1771 */ 1772 vdev_load_log_state(tvd, mtvd); 1773 vdev_reopen(tvd); 1774 } 1775 } 1776 vdev_free(mrvd); 1777 spa_config_exit(spa, SCL_ALL, FTAG); 1778 1779 /* 1780 * Ensure we were able to validate the config. 1781 */ 1782 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1783} 1784 1785/* 1786 * Check for missing log devices 1787 */ 1788static boolean_t 1789spa_check_logs(spa_t *spa) 1790{ 1791 boolean_t rv = B_FALSE; 1792 dsl_pool_t *dp = spa_get_dsl(spa); 1793 1794 switch (spa->spa_log_state) { 1795 case SPA_LOG_MISSING: 1796 /* need to recheck in case slog has been restored */ 1797 case SPA_LOG_UNKNOWN: 1798 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1799 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1800 if (rv) 1801 spa_set_log_state(spa, SPA_LOG_MISSING); 1802 break; 1803 } 1804 return (rv); 1805} 1806 1807static boolean_t 1808spa_passivate_log(spa_t *spa) 1809{ 1810 vdev_t *rvd = spa->spa_root_vdev; 1811 boolean_t slog_found = B_FALSE; 1812 1813 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1814 1815 if (!spa_has_slogs(spa)) 1816 return (B_FALSE); 1817 1818 for (int c = 0; c < rvd->vdev_children; c++) { 1819 vdev_t *tvd = rvd->vdev_child[c]; 1820 metaslab_group_t *mg = tvd->vdev_mg; 1821 1822 if (tvd->vdev_islog) { 1823 metaslab_group_passivate(mg); 1824 slog_found = B_TRUE; 1825 } 1826 } 1827 1828 return (slog_found); 1829} 1830 1831static void 1832spa_activate_log(spa_t *spa) 1833{ 1834 vdev_t *rvd = spa->spa_root_vdev; 1835 1836 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1837 1838 for (int c = 0; c < rvd->vdev_children; c++) { 1839 vdev_t *tvd = rvd->vdev_child[c]; 1840 metaslab_group_t *mg = tvd->vdev_mg; 1841 1842 if (tvd->vdev_islog) 1843 metaslab_group_activate(mg); 1844 } 1845} 1846 1847int 1848spa_offline_log(spa_t *spa) 1849{ 1850 int error; 1851 1852 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1853 NULL, DS_FIND_CHILDREN); 1854 if (error == 0) { 1855 /* 1856 * We successfully offlined the log device, sync out the 1857 * current txg so that the "stubby" block can be removed 1858 * by zil_sync(). 1859 */ 1860 txg_wait_synced(spa->spa_dsl_pool, 0); 1861 } 1862 return (error); 1863} 1864 1865static void 1866spa_aux_check_removed(spa_aux_vdev_t *sav) 1867{ 1868 int i; 1869 1870 for (i = 0; i < sav->sav_count; i++) 1871 spa_check_removed(sav->sav_vdevs[i]); 1872} 1873 1874void 1875spa_claim_notify(zio_t *zio) 1876{ 1877 spa_t *spa = zio->io_spa; 1878 1879 if (zio->io_error) 1880 return; 1881 1882 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1883 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1884 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1885 mutex_exit(&spa->spa_props_lock); 1886} 1887 1888typedef struct spa_load_error { 1889 uint64_t sle_meta_count; 1890 uint64_t sle_data_count; 1891} spa_load_error_t; 1892 1893static void 1894spa_load_verify_done(zio_t *zio) 1895{ 1896 blkptr_t *bp = zio->io_bp; 1897 spa_load_error_t *sle = zio->io_private; 1898 dmu_object_type_t type = BP_GET_TYPE(bp); 1899 int error = zio->io_error; 1900 spa_t *spa = zio->io_spa; 1901 1902 if (error) { 1903 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1904 type != DMU_OT_INTENT_LOG) 1905 atomic_inc_64(&sle->sle_meta_count); 1906 else 1907 atomic_inc_64(&sle->sle_data_count); 1908 } 1909 zio_data_buf_free(zio->io_data, zio->io_size); 1910 1911 mutex_enter(&spa->spa_scrub_lock); 1912 spa->spa_scrub_inflight--; 1913 cv_broadcast(&spa->spa_scrub_io_cv); 1914 mutex_exit(&spa->spa_scrub_lock); 1915} 1916 1917/* 1918 * Maximum number of concurrent scrub i/os to create while verifying 1919 * a pool while importing it. 1920 */ 1921int spa_load_verify_maxinflight = 10000; 1922boolean_t spa_load_verify_metadata = B_TRUE; 1923boolean_t spa_load_verify_data = B_TRUE; 1924 1925SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1926 &spa_load_verify_maxinflight, 0, 1927 "Maximum number of concurrent scrub I/Os to create while verifying a " 1928 "pool while importing it"); 1929 1930SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1931 &spa_load_verify_metadata, 0, 1932 "Check metadata on import?"); 1933 1934SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1935 &spa_load_verify_data, 0, 1936 "Check user data on import?"); 1937 1938/*ARGSUSED*/ 1939static int 1940spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1941 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1942{ 1943 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1944 return (0); 1945 /* 1946 * Note: normally this routine will not be called if 1947 * spa_load_verify_metadata is not set. However, it may be useful 1948 * to manually set the flag after the traversal has begun. 1949 */ 1950 if (!spa_load_verify_metadata) 1951 return (0); 1952 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1953 return (0); 1954 1955 zio_t *rio = arg; 1956 size_t size = BP_GET_PSIZE(bp); 1957 void *data = zio_data_buf_alloc(size); 1958 1959 mutex_enter(&spa->spa_scrub_lock); 1960 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1961 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1962 spa->spa_scrub_inflight++; 1963 mutex_exit(&spa->spa_scrub_lock); 1964 1965 zio_nowait(zio_read(rio, spa, bp, data, size, 1966 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1967 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1968 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1969 return (0); 1970} 1971 1972static int 1973spa_load_verify(spa_t *spa) 1974{ 1975 zio_t *rio; 1976 spa_load_error_t sle = { 0 }; 1977 zpool_rewind_policy_t policy; 1978 boolean_t verify_ok = B_FALSE; 1979 int error = 0; 1980 1981 zpool_get_rewind_policy(spa->spa_config, &policy); 1982 1983 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1984 return (0); 1985 1986 rio = zio_root(spa, NULL, &sle, 1987 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1988 1989 if (spa_load_verify_metadata) { 1990 error = traverse_pool(spa, spa->spa_verify_min_txg, 1991 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1992 spa_load_verify_cb, rio); 1993 } 1994 1995 (void) zio_wait(rio); 1996 1997 spa->spa_load_meta_errors = sle.sle_meta_count; 1998 spa->spa_load_data_errors = sle.sle_data_count; 1999 2000 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2001 sle.sle_data_count <= policy.zrp_maxdata) { 2002 int64_t loss = 0; 2003 2004 verify_ok = B_TRUE; 2005 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2006 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2007 2008 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2009 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2010 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2011 VERIFY(nvlist_add_int64(spa->spa_load_info, 2012 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2013 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2014 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2015 } else { 2016 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2017 } 2018 2019 if (error) { 2020 if (error != ENXIO && error != EIO) 2021 error = SET_ERROR(EIO); 2022 return (error); 2023 } 2024 2025 return (verify_ok ? 0 : EIO); 2026} 2027 2028/* 2029 * Find a value in the pool props object. 2030 */ 2031static void 2032spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2033{ 2034 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2035 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2036} 2037 2038/* 2039 * Find a value in the pool directory object. 2040 */ 2041static int 2042spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2043{ 2044 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2045 name, sizeof (uint64_t), 1, val)); 2046} 2047 2048static int 2049spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2050{ 2051 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2052 return (err); 2053} 2054 2055/* 2056 * Fix up config after a partly-completed split. This is done with the 2057 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2058 * pool have that entry in their config, but only the splitting one contains 2059 * a list of all the guids of the vdevs that are being split off. 2060 * 2061 * This function determines what to do with that list: either rejoin 2062 * all the disks to the pool, or complete the splitting process. To attempt 2063 * the rejoin, each disk that is offlined is marked online again, and 2064 * we do a reopen() call. If the vdev label for every disk that was 2065 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2066 * then we call vdev_split() on each disk, and complete the split. 2067 * 2068 * Otherwise we leave the config alone, with all the vdevs in place in 2069 * the original pool. 2070 */ 2071static void 2072spa_try_repair(spa_t *spa, nvlist_t *config) 2073{ 2074 uint_t extracted; 2075 uint64_t *glist; 2076 uint_t i, gcount; 2077 nvlist_t *nvl; 2078 vdev_t **vd; 2079 boolean_t attempt_reopen; 2080 2081 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2082 return; 2083 2084 /* check that the config is complete */ 2085 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2086 &glist, &gcount) != 0) 2087 return; 2088 2089 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2090 2091 /* attempt to online all the vdevs & validate */ 2092 attempt_reopen = B_TRUE; 2093 for (i = 0; i < gcount; i++) { 2094 if (glist[i] == 0) /* vdev is hole */ 2095 continue; 2096 2097 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2098 if (vd[i] == NULL) { 2099 /* 2100 * Don't bother attempting to reopen the disks; 2101 * just do the split. 2102 */ 2103 attempt_reopen = B_FALSE; 2104 } else { 2105 /* attempt to re-online it */ 2106 vd[i]->vdev_offline = B_FALSE; 2107 } 2108 } 2109 2110 if (attempt_reopen) { 2111 vdev_reopen(spa->spa_root_vdev); 2112 2113 /* check each device to see what state it's in */ 2114 for (extracted = 0, i = 0; i < gcount; i++) { 2115 if (vd[i] != NULL && 2116 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2117 break; 2118 ++extracted; 2119 } 2120 } 2121 2122 /* 2123 * If every disk has been moved to the new pool, or if we never 2124 * even attempted to look at them, then we split them off for 2125 * good. 2126 */ 2127 if (!attempt_reopen || gcount == extracted) { 2128 for (i = 0; i < gcount; i++) 2129 if (vd[i] != NULL) 2130 vdev_split(vd[i]); 2131 vdev_reopen(spa->spa_root_vdev); 2132 } 2133 2134 kmem_free(vd, gcount * sizeof (vdev_t *)); 2135} 2136 2137static int 2138spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2139 boolean_t mosconfig) 2140{ 2141 nvlist_t *config = spa->spa_config; 2142 char *ereport = FM_EREPORT_ZFS_POOL; 2143 char *comment; 2144 int error; 2145 uint64_t pool_guid; 2146 nvlist_t *nvl; 2147 2148 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2149 return (SET_ERROR(EINVAL)); 2150 2151 ASSERT(spa->spa_comment == NULL); 2152 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2153 spa->spa_comment = spa_strdup(comment); 2154 2155 /* 2156 * Versioning wasn't explicitly added to the label until later, so if 2157 * it's not present treat it as the initial version. 2158 */ 2159 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2160 &spa->spa_ubsync.ub_version) != 0) 2161 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2162 2163 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2164 &spa->spa_config_txg); 2165 2166 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2167 spa_guid_exists(pool_guid, 0)) { 2168 error = SET_ERROR(EEXIST); 2169 } else { 2170 spa->spa_config_guid = pool_guid; 2171 2172 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2173 &nvl) == 0) { 2174 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2175 KM_SLEEP) == 0); 2176 } 2177 2178 nvlist_free(spa->spa_load_info); 2179 spa->spa_load_info = fnvlist_alloc(); 2180 2181 gethrestime(&spa->spa_loaded_ts); 2182 error = spa_load_impl(spa, pool_guid, config, state, type, 2183 mosconfig, &ereport); 2184 } 2185 2186 /* 2187 * Don't count references from objsets that are already closed 2188 * and are making their way through the eviction process. 2189 */ 2190 spa_evicting_os_wait(spa); 2191 spa->spa_minref = refcount_count(&spa->spa_refcount); 2192 if (error) { 2193 if (error != EEXIST) { 2194 spa->spa_loaded_ts.tv_sec = 0; 2195 spa->spa_loaded_ts.tv_nsec = 0; 2196 } 2197 if (error != EBADF) { 2198 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2199 } 2200 } 2201 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2202 spa->spa_ena = 0; 2203 2204 return (error); 2205} 2206 2207/* 2208 * Load an existing storage pool, using the pool's builtin spa_config as a 2209 * source of configuration information. 2210 */ 2211static int 2212spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2213 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2214 char **ereport) 2215{ 2216 int error = 0; 2217 nvlist_t *nvroot = NULL; 2218 nvlist_t *label; 2219 vdev_t *rvd; 2220 uberblock_t *ub = &spa->spa_uberblock; 2221 uint64_t children, config_cache_txg = spa->spa_config_txg; 2222 int orig_mode = spa->spa_mode; 2223 int parse; 2224 uint64_t obj; 2225 boolean_t missing_feat_write = B_FALSE; 2226 2227 /* 2228 * If this is an untrusted config, access the pool in read-only mode. 2229 * This prevents things like resilvering recently removed devices. 2230 */ 2231 if (!mosconfig) 2232 spa->spa_mode = FREAD; 2233 2234 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2235 2236 spa->spa_load_state = state; 2237 2238 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2239 return (SET_ERROR(EINVAL)); 2240 2241 parse = (type == SPA_IMPORT_EXISTING ? 2242 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2243 2244 /* 2245 * Create "The Godfather" zio to hold all async IOs 2246 */ 2247 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2248 KM_SLEEP); 2249 for (int i = 0; i < max_ncpus; i++) { 2250 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2251 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2252 ZIO_FLAG_GODFATHER); 2253 } 2254 2255 /* 2256 * Parse the configuration into a vdev tree. We explicitly set the 2257 * value that will be returned by spa_version() since parsing the 2258 * configuration requires knowing the version number. 2259 */ 2260 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2261 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2262 spa_config_exit(spa, SCL_ALL, FTAG); 2263 2264 if (error != 0) 2265 return (error); 2266 2267 ASSERT(spa->spa_root_vdev == rvd); 2268 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2269 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2270 2271 if (type != SPA_IMPORT_ASSEMBLE) { 2272 ASSERT(spa_guid(spa) == pool_guid); 2273 } 2274 2275 /* 2276 * Try to open all vdevs, loading each label in the process. 2277 */ 2278 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2279 error = vdev_open(rvd); 2280 spa_config_exit(spa, SCL_ALL, FTAG); 2281 if (error != 0) 2282 return (error); 2283 2284 /* 2285 * We need to validate the vdev labels against the configuration that 2286 * we have in hand, which is dependent on the setting of mosconfig. If 2287 * mosconfig is true then we're validating the vdev labels based on 2288 * that config. Otherwise, we're validating against the cached config 2289 * (zpool.cache) that was read when we loaded the zfs module, and then 2290 * later we will recursively call spa_load() and validate against 2291 * the vdev config. 2292 * 2293 * If we're assembling a new pool that's been split off from an 2294 * existing pool, the labels haven't yet been updated so we skip 2295 * validation for now. 2296 */ 2297 if (type != SPA_IMPORT_ASSEMBLE) { 2298 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2299 error = vdev_validate(rvd, mosconfig); 2300 spa_config_exit(spa, SCL_ALL, FTAG); 2301 2302 if (error != 0) 2303 return (error); 2304 2305 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2306 return (SET_ERROR(ENXIO)); 2307 } 2308 2309 /* 2310 * Find the best uberblock. 2311 */ 2312 vdev_uberblock_load(rvd, ub, &label); 2313 2314 /* 2315 * If we weren't able to find a single valid uberblock, return failure. 2316 */ 2317 if (ub->ub_txg == 0) { 2318 nvlist_free(label); 2319 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2320 } 2321 2322 /* 2323 * If the pool has an unsupported version we can't open it. 2324 */ 2325 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2326 nvlist_free(label); 2327 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2328 } 2329 2330 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2331 nvlist_t *features; 2332 2333 /* 2334 * If we weren't able to find what's necessary for reading the 2335 * MOS in the label, return failure. 2336 */ 2337 if (label == NULL || nvlist_lookup_nvlist(label, 2338 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2339 nvlist_free(label); 2340 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2341 ENXIO)); 2342 } 2343 2344 /* 2345 * Update our in-core representation with the definitive values 2346 * from the label. 2347 */ 2348 nvlist_free(spa->spa_label_features); 2349 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2350 } 2351 2352 nvlist_free(label); 2353 2354 /* 2355 * Look through entries in the label nvlist's features_for_read. If 2356 * there is a feature listed there which we don't understand then we 2357 * cannot open a pool. 2358 */ 2359 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2360 nvlist_t *unsup_feat; 2361 2362 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2363 0); 2364 2365 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2366 NULL); nvp != NULL; 2367 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2368 if (!zfeature_is_supported(nvpair_name(nvp))) { 2369 VERIFY(nvlist_add_string(unsup_feat, 2370 nvpair_name(nvp), "") == 0); 2371 } 2372 } 2373 2374 if (!nvlist_empty(unsup_feat)) { 2375 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2376 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2377 nvlist_free(unsup_feat); 2378 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2379 ENOTSUP)); 2380 } 2381 2382 nvlist_free(unsup_feat); 2383 } 2384 2385 /* 2386 * If the vdev guid sum doesn't match the uberblock, we have an 2387 * incomplete configuration. We first check to see if the pool 2388 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2389 * If it is, defer the vdev_guid_sum check till later so we 2390 * can handle missing vdevs. 2391 */ 2392 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2393 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2394 rvd->vdev_guid_sum != ub->ub_guid_sum) 2395 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2396 2397 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2398 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2399 spa_try_repair(spa, config); 2400 spa_config_exit(spa, SCL_ALL, FTAG); 2401 nvlist_free(spa->spa_config_splitting); 2402 spa->spa_config_splitting = NULL; 2403 } 2404 2405 /* 2406 * Initialize internal SPA structures. 2407 */ 2408 spa->spa_state = POOL_STATE_ACTIVE; 2409 spa->spa_ubsync = spa->spa_uberblock; 2410 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2411 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2412 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2413 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2414 spa->spa_claim_max_txg = spa->spa_first_txg; 2415 spa->spa_prev_software_version = ub->ub_software_version; 2416 2417 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2418 if (error) 2419 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2420 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2421 2422 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2423 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2424 2425 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2426 boolean_t missing_feat_read = B_FALSE; 2427 nvlist_t *unsup_feat, *enabled_feat; 2428 2429 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2430 &spa->spa_feat_for_read_obj) != 0) { 2431 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2432 } 2433 2434 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2435 &spa->spa_feat_for_write_obj) != 0) { 2436 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2437 } 2438 2439 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2440 &spa->spa_feat_desc_obj) != 0) { 2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2442 } 2443 2444 enabled_feat = fnvlist_alloc(); 2445 unsup_feat = fnvlist_alloc(); 2446 2447 if (!spa_features_check(spa, B_FALSE, 2448 unsup_feat, enabled_feat)) 2449 missing_feat_read = B_TRUE; 2450 2451 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2452 if (!spa_features_check(spa, B_TRUE, 2453 unsup_feat, enabled_feat)) { 2454 missing_feat_write = B_TRUE; 2455 } 2456 } 2457 2458 fnvlist_add_nvlist(spa->spa_load_info, 2459 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2460 2461 if (!nvlist_empty(unsup_feat)) { 2462 fnvlist_add_nvlist(spa->spa_load_info, 2463 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2464 } 2465 2466 fnvlist_free(enabled_feat); 2467 fnvlist_free(unsup_feat); 2468 2469 if (!missing_feat_read) { 2470 fnvlist_add_boolean(spa->spa_load_info, 2471 ZPOOL_CONFIG_CAN_RDONLY); 2472 } 2473 2474 /* 2475 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2476 * twofold: to determine whether the pool is available for 2477 * import in read-write mode and (if it is not) whether the 2478 * pool is available for import in read-only mode. If the pool 2479 * is available for import in read-write mode, it is displayed 2480 * as available in userland; if it is not available for import 2481 * in read-only mode, it is displayed as unavailable in 2482 * userland. If the pool is available for import in read-only 2483 * mode but not read-write mode, it is displayed as unavailable 2484 * in userland with a special note that the pool is actually 2485 * available for open in read-only mode. 2486 * 2487 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2488 * missing a feature for write, we must first determine whether 2489 * the pool can be opened read-only before returning to 2490 * userland in order to know whether to display the 2491 * abovementioned note. 2492 */ 2493 if (missing_feat_read || (missing_feat_write && 2494 spa_writeable(spa))) { 2495 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2496 ENOTSUP)); 2497 } 2498 2499 /* 2500 * Load refcounts for ZFS features from disk into an in-memory 2501 * cache during SPA initialization. 2502 */ 2503 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2504 uint64_t refcount; 2505 2506 error = feature_get_refcount_from_disk(spa, 2507 &spa_feature_table[i], &refcount); 2508 if (error == 0) { 2509 spa->spa_feat_refcount_cache[i] = refcount; 2510 } else if (error == ENOTSUP) { 2511 spa->spa_feat_refcount_cache[i] = 2512 SPA_FEATURE_DISABLED; 2513 } else { 2514 return (spa_vdev_err(rvd, 2515 VDEV_AUX_CORRUPT_DATA, EIO)); 2516 } 2517 } 2518 } 2519 2520 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2521 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2522 &spa->spa_feat_enabled_txg_obj) != 0) 2523 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2524 } 2525 2526 spa->spa_is_initializing = B_TRUE; 2527 error = dsl_pool_open(spa->spa_dsl_pool); 2528 spa->spa_is_initializing = B_FALSE; 2529 if (error != 0) 2530 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2531 2532 if (!mosconfig) { 2533 uint64_t hostid; 2534 nvlist_t *policy = NULL, *nvconfig; 2535 2536 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2537 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2538 2539 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2540 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2541 char *hostname; 2542 unsigned long myhostid = 0; 2543 2544 VERIFY(nvlist_lookup_string(nvconfig, 2545 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2546 2547#ifdef _KERNEL 2548 myhostid = zone_get_hostid(NULL); 2549#else /* _KERNEL */ 2550 /* 2551 * We're emulating the system's hostid in userland, so 2552 * we can't use zone_get_hostid(). 2553 */ 2554 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2555#endif /* _KERNEL */ 2556 if (check_hostid && hostid != 0 && myhostid != 0 && 2557 hostid != myhostid) { 2558 nvlist_free(nvconfig); 2559 cmn_err(CE_WARN, "pool '%s' could not be " 2560 "loaded as it was last accessed by " 2561 "another system (host: %s hostid: 0x%lx). " 2562 "See: http://illumos.org/msg/ZFS-8000-EY", 2563 spa_name(spa), hostname, 2564 (unsigned long)hostid); 2565 return (SET_ERROR(EBADF)); 2566 } 2567 } 2568 if (nvlist_lookup_nvlist(spa->spa_config, 2569 ZPOOL_REWIND_POLICY, &policy) == 0) 2570 VERIFY(nvlist_add_nvlist(nvconfig, 2571 ZPOOL_REWIND_POLICY, policy) == 0); 2572 2573 spa_config_set(spa, nvconfig); 2574 spa_unload(spa); 2575 spa_deactivate(spa); 2576 spa_activate(spa, orig_mode); 2577 2578 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2579 } 2580 2581 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2582 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2583 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2584 if (error != 0) 2585 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2586 2587 /* 2588 * Load the bit that tells us to use the new accounting function 2589 * (raid-z deflation). If we have an older pool, this will not 2590 * be present. 2591 */ 2592 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2593 if (error != 0 && error != ENOENT) 2594 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2595 2596 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2597 &spa->spa_creation_version); 2598 if (error != 0 && error != ENOENT) 2599 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2600 2601 /* 2602 * Load the persistent error log. If we have an older pool, this will 2603 * not be present. 2604 */ 2605 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2606 if (error != 0 && error != ENOENT) 2607 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2608 2609 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2610 &spa->spa_errlog_scrub); 2611 if (error != 0 && error != ENOENT) 2612 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2613 2614 /* 2615 * Load the history object. If we have an older pool, this 2616 * will not be present. 2617 */ 2618 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2619 if (error != 0 && error != ENOENT) 2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2621 2622 /* 2623 * If we're assembling the pool from the split-off vdevs of 2624 * an existing pool, we don't want to attach the spares & cache 2625 * devices. 2626 */ 2627 2628 /* 2629 * Load any hot spares for this pool. 2630 */ 2631 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2632 if (error != 0 && error != ENOENT) 2633 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2634 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2635 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2636 if (load_nvlist(spa, spa->spa_spares.sav_object, 2637 &spa->spa_spares.sav_config) != 0) 2638 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2639 2640 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2641 spa_load_spares(spa); 2642 spa_config_exit(spa, SCL_ALL, FTAG); 2643 } else if (error == 0) { 2644 spa->spa_spares.sav_sync = B_TRUE; 2645 } 2646 2647 /* 2648 * Load any level 2 ARC devices for this pool. 2649 */ 2650 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2651 &spa->spa_l2cache.sav_object); 2652 if (error != 0 && error != ENOENT) 2653 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2654 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2655 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2656 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2657 &spa->spa_l2cache.sav_config) != 0) 2658 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2659 2660 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2661 spa_load_l2cache(spa); 2662 spa_config_exit(spa, SCL_ALL, FTAG); 2663 } else if (error == 0) { 2664 spa->spa_l2cache.sav_sync = B_TRUE; 2665 } 2666 2667 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2668 2669 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2670 if (error && error != ENOENT) 2671 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2672 2673 if (error == 0) { 2674 uint64_t autoreplace; 2675 2676 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2677 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2678 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2679 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2680 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2681 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2682 &spa->spa_dedup_ditto); 2683 2684 spa->spa_autoreplace = (autoreplace != 0); 2685 } 2686 2687 /* 2688 * If the 'autoreplace' property is set, then post a resource notifying 2689 * the ZFS DE that it should not issue any faults for unopenable 2690 * devices. We also iterate over the vdevs, and post a sysevent for any 2691 * unopenable vdevs so that the normal autoreplace handler can take 2692 * over. 2693 */ 2694 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2695 spa_check_removed(spa->spa_root_vdev); 2696 /* 2697 * For the import case, this is done in spa_import(), because 2698 * at this point we're using the spare definitions from 2699 * the MOS config, not necessarily from the userland config. 2700 */ 2701 if (state != SPA_LOAD_IMPORT) { 2702 spa_aux_check_removed(&spa->spa_spares); 2703 spa_aux_check_removed(&spa->spa_l2cache); 2704 } 2705 } 2706 2707 /* 2708 * Load the vdev state for all toplevel vdevs. 2709 */ 2710 vdev_load(rvd); 2711 2712 /* 2713 * Propagate the leaf DTLs we just loaded all the way up the tree. 2714 */ 2715 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2716 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2717 spa_config_exit(spa, SCL_ALL, FTAG); 2718 2719 /* 2720 * Load the DDTs (dedup tables). 2721 */ 2722 error = ddt_load(spa); 2723 if (error != 0) 2724 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2725 2726 spa_update_dspace(spa); 2727 2728 /* 2729 * Validate the config, using the MOS config to fill in any 2730 * information which might be missing. If we fail to validate 2731 * the config then declare the pool unfit for use. If we're 2732 * assembling a pool from a split, the log is not transferred 2733 * over. 2734 */ 2735 if (type != SPA_IMPORT_ASSEMBLE) { 2736 nvlist_t *nvconfig; 2737 2738 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2739 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2740 2741 if (!spa_config_valid(spa, nvconfig)) { 2742 nvlist_free(nvconfig); 2743 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2744 ENXIO)); 2745 } 2746 nvlist_free(nvconfig); 2747 2748 /* 2749 * Now that we've validated the config, check the state of the 2750 * root vdev. If it can't be opened, it indicates one or 2751 * more toplevel vdevs are faulted. 2752 */ 2753 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2754 return (SET_ERROR(ENXIO)); 2755 2756 if (spa_writeable(spa) && spa_check_logs(spa)) { 2757 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2758 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2759 } 2760 } 2761 2762 if (missing_feat_write) { 2763 ASSERT(state == SPA_LOAD_TRYIMPORT); 2764 2765 /* 2766 * At this point, we know that we can open the pool in 2767 * read-only mode but not read-write mode. We now have enough 2768 * information and can return to userland. 2769 */ 2770 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2771 } 2772 2773 /* 2774 * We've successfully opened the pool, verify that we're ready 2775 * to start pushing transactions. 2776 */ 2777 if (state != SPA_LOAD_TRYIMPORT) { 2778 if (error = spa_load_verify(spa)) 2779 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2780 error)); 2781 } 2782 2783 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2784 spa->spa_load_max_txg == UINT64_MAX)) { 2785 dmu_tx_t *tx; 2786 int need_update = B_FALSE; 2787 dsl_pool_t *dp = spa_get_dsl(spa); 2788 2789 ASSERT(state != SPA_LOAD_TRYIMPORT); 2790 2791 /* 2792 * Claim log blocks that haven't been committed yet. 2793 * This must all happen in a single txg. 2794 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2795 * invoked from zil_claim_log_block()'s i/o done callback. 2796 * Price of rollback is that we abandon the log. 2797 */ 2798 spa->spa_claiming = B_TRUE; 2799 2800 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2801 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2802 zil_claim, tx, DS_FIND_CHILDREN); 2803 dmu_tx_commit(tx); 2804 2805 spa->spa_claiming = B_FALSE; 2806 2807 spa_set_log_state(spa, SPA_LOG_GOOD); 2808 spa->spa_sync_on = B_TRUE; 2809 txg_sync_start(spa->spa_dsl_pool); 2810 2811 /* 2812 * Wait for all claims to sync. We sync up to the highest 2813 * claimed log block birth time so that claimed log blocks 2814 * don't appear to be from the future. spa_claim_max_txg 2815 * will have been set for us by either zil_check_log_chain() 2816 * (invoked from spa_check_logs()) or zil_claim() above. 2817 */ 2818 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2819 2820 /* 2821 * If the config cache is stale, or we have uninitialized 2822 * metaslabs (see spa_vdev_add()), then update the config. 2823 * 2824 * If this is a verbatim import, trust the current 2825 * in-core spa_config and update the disk labels. 2826 */ 2827 if (config_cache_txg != spa->spa_config_txg || 2828 state == SPA_LOAD_IMPORT || 2829 state == SPA_LOAD_RECOVER || 2830 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2831 need_update = B_TRUE; 2832 2833 for (int c = 0; c < rvd->vdev_children; c++) 2834 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2835 need_update = B_TRUE; 2836 2837 /* 2838 * Update the config cache asychronously in case we're the 2839 * root pool, in which case the config cache isn't writable yet. 2840 */ 2841 if (need_update) 2842 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2843 2844 /* 2845 * Check all DTLs to see if anything needs resilvering. 2846 */ 2847 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2848 vdev_resilver_needed(rvd, NULL, NULL)) 2849 spa_async_request(spa, SPA_ASYNC_RESILVER); 2850 2851 /* 2852 * Log the fact that we booted up (so that we can detect if 2853 * we rebooted in the middle of an operation). 2854 */ 2855 spa_history_log_version(spa, "open"); 2856 2857 /* 2858 * Delete any inconsistent datasets. 2859 */ 2860 (void) dmu_objset_find(spa_name(spa), 2861 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2862 2863 /* 2864 * Clean up any stale temporary dataset userrefs. 2865 */ 2866 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2867 } 2868 2869 return (0); 2870} 2871 2872static int 2873spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2874{ 2875 int mode = spa->spa_mode; 2876 2877 spa_unload(spa); 2878 spa_deactivate(spa); 2879 2880 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2881 2882 spa_activate(spa, mode); 2883 spa_async_suspend(spa); 2884 2885 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2886} 2887 2888/* 2889 * If spa_load() fails this function will try loading prior txg's. If 2890 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2891 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2892 * function will not rewind the pool and will return the same error as 2893 * spa_load(). 2894 */ 2895static int 2896spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2897 uint64_t max_request, int rewind_flags) 2898{ 2899 nvlist_t *loadinfo = NULL; 2900 nvlist_t *config = NULL; 2901 int load_error, rewind_error; 2902 uint64_t safe_rewind_txg; 2903 uint64_t min_txg; 2904 2905 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2906 spa->spa_load_max_txg = spa->spa_load_txg; 2907 spa_set_log_state(spa, SPA_LOG_CLEAR); 2908 } else { 2909 spa->spa_load_max_txg = max_request; 2910 if (max_request != UINT64_MAX) 2911 spa->spa_extreme_rewind = B_TRUE; 2912 } 2913 2914 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2915 mosconfig); 2916 if (load_error == 0) 2917 return (0); 2918 2919 if (spa->spa_root_vdev != NULL) 2920 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2921 2922 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2923 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2924 2925 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2926 nvlist_free(config); 2927 return (load_error); 2928 } 2929 2930 if (state == SPA_LOAD_RECOVER) { 2931 /* Price of rolling back is discarding txgs, including log */ 2932 spa_set_log_state(spa, SPA_LOG_CLEAR); 2933 } else { 2934 /* 2935 * If we aren't rolling back save the load info from our first 2936 * import attempt so that we can restore it after attempting 2937 * to rewind. 2938 */ 2939 loadinfo = spa->spa_load_info; 2940 spa->spa_load_info = fnvlist_alloc(); 2941 } 2942 2943 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2944 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2945 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2946 TXG_INITIAL : safe_rewind_txg; 2947 2948 /* 2949 * Continue as long as we're finding errors, we're still within 2950 * the acceptable rewind range, and we're still finding uberblocks 2951 */ 2952 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2953 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2954 if (spa->spa_load_max_txg < safe_rewind_txg) 2955 spa->spa_extreme_rewind = B_TRUE; 2956 rewind_error = spa_load_retry(spa, state, mosconfig); 2957 } 2958 2959 spa->spa_extreme_rewind = B_FALSE; 2960 spa->spa_load_max_txg = UINT64_MAX; 2961 2962 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2963 spa_config_set(spa, config); 2964 2965 if (state == SPA_LOAD_RECOVER) { 2966 ASSERT3P(loadinfo, ==, NULL); 2967 return (rewind_error); 2968 } else { 2969 /* Store the rewind info as part of the initial load info */ 2970 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2971 spa->spa_load_info); 2972 2973 /* Restore the initial load info */ 2974 fnvlist_free(spa->spa_load_info); 2975 spa->spa_load_info = loadinfo; 2976 2977 return (load_error); 2978 } 2979} 2980 2981/* 2982 * Pool Open/Import 2983 * 2984 * The import case is identical to an open except that the configuration is sent 2985 * down from userland, instead of grabbed from the configuration cache. For the 2986 * case of an open, the pool configuration will exist in the 2987 * POOL_STATE_UNINITIALIZED state. 2988 * 2989 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2990 * the same time open the pool, without having to keep around the spa_t in some 2991 * ambiguous state. 2992 */ 2993static int 2994spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2995 nvlist_t **config) 2996{ 2997 spa_t *spa; 2998 spa_load_state_t state = SPA_LOAD_OPEN; 2999 int error; 3000 int locked = B_FALSE; 3001 int firstopen = B_FALSE; 3002 3003 *spapp = NULL; 3004 3005 /* 3006 * As disgusting as this is, we need to support recursive calls to this 3007 * function because dsl_dir_open() is called during spa_load(), and ends 3008 * up calling spa_open() again. The real fix is to figure out how to 3009 * avoid dsl_dir_open() calling this in the first place. 3010 */ 3011 if (mutex_owner(&spa_namespace_lock) != curthread) { 3012 mutex_enter(&spa_namespace_lock); 3013 locked = B_TRUE; 3014 } 3015 3016 if ((spa = spa_lookup(pool)) == NULL) { 3017 if (locked) 3018 mutex_exit(&spa_namespace_lock); 3019 return (SET_ERROR(ENOENT)); 3020 } 3021 3022 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3023 zpool_rewind_policy_t policy; 3024 3025 firstopen = B_TRUE; 3026 3027 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3028 &policy); 3029 if (policy.zrp_request & ZPOOL_DO_REWIND) 3030 state = SPA_LOAD_RECOVER; 3031 3032 spa_activate(spa, spa_mode_global); 3033 3034 if (state != SPA_LOAD_RECOVER) 3035 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3036 3037 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3038 policy.zrp_request); 3039 3040 if (error == EBADF) { 3041 /* 3042 * If vdev_validate() returns failure (indicated by 3043 * EBADF), it indicates that one of the vdevs indicates 3044 * that the pool has been exported or destroyed. If 3045 * this is the case, the config cache is out of sync and 3046 * we should remove the pool from the namespace. 3047 */ 3048 spa_unload(spa); 3049 spa_deactivate(spa); 3050 spa_config_sync(spa, B_TRUE, B_TRUE); 3051 spa_remove(spa); 3052 if (locked) 3053 mutex_exit(&spa_namespace_lock); 3054 return (SET_ERROR(ENOENT)); 3055 } 3056 3057 if (error) { 3058 /* 3059 * We can't open the pool, but we still have useful 3060 * information: the state of each vdev after the 3061 * attempted vdev_open(). Return this to the user. 3062 */ 3063 if (config != NULL && spa->spa_config) { 3064 VERIFY(nvlist_dup(spa->spa_config, config, 3065 KM_SLEEP) == 0); 3066 VERIFY(nvlist_add_nvlist(*config, 3067 ZPOOL_CONFIG_LOAD_INFO, 3068 spa->spa_load_info) == 0); 3069 } 3070 spa_unload(spa); 3071 spa_deactivate(spa); 3072 spa->spa_last_open_failed = error; 3073 if (locked) 3074 mutex_exit(&spa_namespace_lock); 3075 *spapp = NULL; 3076 return (error); 3077 } 3078 } 3079 3080 spa_open_ref(spa, tag); 3081 3082 if (config != NULL) 3083 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3084 3085 /* 3086 * If we've recovered the pool, pass back any information we 3087 * gathered while doing the load. 3088 */ 3089 if (state == SPA_LOAD_RECOVER) { 3090 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3091 spa->spa_load_info) == 0); 3092 } 3093 3094 if (locked) { 3095 spa->spa_last_open_failed = 0; 3096 spa->spa_last_ubsync_txg = 0; 3097 spa->spa_load_txg = 0; 3098 mutex_exit(&spa_namespace_lock); 3099#ifdef __FreeBSD__ 3100#ifdef _KERNEL 3101 if (firstopen) 3102 zvol_create_minors(spa->spa_name); 3103#endif 3104#endif 3105 } 3106 3107 *spapp = spa; 3108 3109 return (0); 3110} 3111 3112int 3113spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3114 nvlist_t **config) 3115{ 3116 return (spa_open_common(name, spapp, tag, policy, config)); 3117} 3118 3119int 3120spa_open(const char *name, spa_t **spapp, void *tag) 3121{ 3122 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3123} 3124 3125/* 3126 * Lookup the given spa_t, incrementing the inject count in the process, 3127 * preventing it from being exported or destroyed. 3128 */ 3129spa_t * 3130spa_inject_addref(char *name) 3131{ 3132 spa_t *spa; 3133 3134 mutex_enter(&spa_namespace_lock); 3135 if ((spa = spa_lookup(name)) == NULL) { 3136 mutex_exit(&spa_namespace_lock); 3137 return (NULL); 3138 } 3139 spa->spa_inject_ref++; 3140 mutex_exit(&spa_namespace_lock); 3141 3142 return (spa); 3143} 3144 3145void 3146spa_inject_delref(spa_t *spa) 3147{ 3148 mutex_enter(&spa_namespace_lock); 3149 spa->spa_inject_ref--; 3150 mutex_exit(&spa_namespace_lock); 3151} 3152 3153/* 3154 * Add spares device information to the nvlist. 3155 */ 3156static void 3157spa_add_spares(spa_t *spa, nvlist_t *config) 3158{ 3159 nvlist_t **spares; 3160 uint_t i, nspares; 3161 nvlist_t *nvroot; 3162 uint64_t guid; 3163 vdev_stat_t *vs; 3164 uint_t vsc; 3165 uint64_t pool; 3166 3167 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3168 3169 if (spa->spa_spares.sav_count == 0) 3170 return; 3171 3172 VERIFY(nvlist_lookup_nvlist(config, 3173 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3174 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3175 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3176 if (nspares != 0) { 3177 VERIFY(nvlist_add_nvlist_array(nvroot, 3178 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3179 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3180 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3181 3182 /* 3183 * Go through and find any spares which have since been 3184 * repurposed as an active spare. If this is the case, update 3185 * their status appropriately. 3186 */ 3187 for (i = 0; i < nspares; i++) { 3188 VERIFY(nvlist_lookup_uint64(spares[i], 3189 ZPOOL_CONFIG_GUID, &guid) == 0); 3190 if (spa_spare_exists(guid, &pool, NULL) && 3191 pool != 0ULL) { 3192 VERIFY(nvlist_lookup_uint64_array( 3193 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3194 (uint64_t **)&vs, &vsc) == 0); 3195 vs->vs_state = VDEV_STATE_CANT_OPEN; 3196 vs->vs_aux = VDEV_AUX_SPARED; 3197 } 3198 } 3199 } 3200} 3201 3202/* 3203 * Add l2cache device information to the nvlist, including vdev stats. 3204 */ 3205static void 3206spa_add_l2cache(spa_t *spa, nvlist_t *config) 3207{ 3208 nvlist_t **l2cache; 3209 uint_t i, j, nl2cache; 3210 nvlist_t *nvroot; 3211 uint64_t guid; 3212 vdev_t *vd; 3213 vdev_stat_t *vs; 3214 uint_t vsc; 3215 3216 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3217 3218 if (spa->spa_l2cache.sav_count == 0) 3219 return; 3220 3221 VERIFY(nvlist_lookup_nvlist(config, 3222 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3223 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3224 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3225 if (nl2cache != 0) { 3226 VERIFY(nvlist_add_nvlist_array(nvroot, 3227 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3228 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3229 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3230 3231 /* 3232 * Update level 2 cache device stats. 3233 */ 3234 3235 for (i = 0; i < nl2cache; i++) { 3236 VERIFY(nvlist_lookup_uint64(l2cache[i], 3237 ZPOOL_CONFIG_GUID, &guid) == 0); 3238 3239 vd = NULL; 3240 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3241 if (guid == 3242 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3243 vd = spa->spa_l2cache.sav_vdevs[j]; 3244 break; 3245 } 3246 } 3247 ASSERT(vd != NULL); 3248 3249 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3250 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3251 == 0); 3252 vdev_get_stats(vd, vs); 3253 } 3254 } 3255} 3256 3257static void 3258spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3259{ 3260 nvlist_t *features; 3261 zap_cursor_t zc; 3262 zap_attribute_t za; 3263 3264 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3265 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3266 3267 /* We may be unable to read features if pool is suspended. */ 3268 if (spa_suspended(spa)) 3269 goto out; 3270 3271 if (spa->spa_feat_for_read_obj != 0) { 3272 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3273 spa->spa_feat_for_read_obj); 3274 zap_cursor_retrieve(&zc, &za) == 0; 3275 zap_cursor_advance(&zc)) { 3276 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3277 za.za_num_integers == 1); 3278 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3279 za.za_first_integer)); 3280 } 3281 zap_cursor_fini(&zc); 3282 } 3283 3284 if (spa->spa_feat_for_write_obj != 0) { 3285 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3286 spa->spa_feat_for_write_obj); 3287 zap_cursor_retrieve(&zc, &za) == 0; 3288 zap_cursor_advance(&zc)) { 3289 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3290 za.za_num_integers == 1); 3291 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3292 za.za_first_integer)); 3293 } 3294 zap_cursor_fini(&zc); 3295 } 3296 3297out: 3298 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3299 features) == 0); 3300 nvlist_free(features); 3301} 3302 3303int 3304spa_get_stats(const char *name, nvlist_t **config, 3305 char *altroot, size_t buflen) 3306{ 3307 int error; 3308 spa_t *spa; 3309 3310 *config = NULL; 3311 error = spa_open_common(name, &spa, FTAG, NULL, config); 3312 3313 if (spa != NULL) { 3314 /* 3315 * This still leaves a window of inconsistency where the spares 3316 * or l2cache devices could change and the config would be 3317 * self-inconsistent. 3318 */ 3319 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3320 3321 if (*config != NULL) { 3322 uint64_t loadtimes[2]; 3323 3324 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3325 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3326 VERIFY(nvlist_add_uint64_array(*config, 3327 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3328 3329 VERIFY(nvlist_add_uint64(*config, 3330 ZPOOL_CONFIG_ERRCOUNT, 3331 spa_get_errlog_size(spa)) == 0); 3332 3333 if (spa_suspended(spa)) 3334 VERIFY(nvlist_add_uint64(*config, 3335 ZPOOL_CONFIG_SUSPENDED, 3336 spa->spa_failmode) == 0); 3337 3338 spa_add_spares(spa, *config); 3339 spa_add_l2cache(spa, *config); 3340 spa_add_feature_stats(spa, *config); 3341 } 3342 } 3343 3344 /* 3345 * We want to get the alternate root even for faulted pools, so we cheat 3346 * and call spa_lookup() directly. 3347 */ 3348 if (altroot) { 3349 if (spa == NULL) { 3350 mutex_enter(&spa_namespace_lock); 3351 spa = spa_lookup(name); 3352 if (spa) 3353 spa_altroot(spa, altroot, buflen); 3354 else 3355 altroot[0] = '\0'; 3356 spa = NULL; 3357 mutex_exit(&spa_namespace_lock); 3358 } else { 3359 spa_altroot(spa, altroot, buflen); 3360 } 3361 } 3362 3363 if (spa != NULL) { 3364 spa_config_exit(spa, SCL_CONFIG, FTAG); 3365 spa_close(spa, FTAG); 3366 } 3367 3368 return (error); 3369} 3370 3371/* 3372 * Validate that the auxiliary device array is well formed. We must have an 3373 * array of nvlists, each which describes a valid leaf vdev. If this is an 3374 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3375 * specified, as long as they are well-formed. 3376 */ 3377static int 3378spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3379 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3380 vdev_labeltype_t label) 3381{ 3382 nvlist_t **dev; 3383 uint_t i, ndev; 3384 vdev_t *vd; 3385 int error; 3386 3387 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3388 3389 /* 3390 * It's acceptable to have no devs specified. 3391 */ 3392 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3393 return (0); 3394 3395 if (ndev == 0) 3396 return (SET_ERROR(EINVAL)); 3397 3398 /* 3399 * Make sure the pool is formatted with a version that supports this 3400 * device type. 3401 */ 3402 if (spa_version(spa) < version) 3403 return (SET_ERROR(ENOTSUP)); 3404 3405 /* 3406 * Set the pending device list so we correctly handle device in-use 3407 * checking. 3408 */ 3409 sav->sav_pending = dev; 3410 sav->sav_npending = ndev; 3411 3412 for (i = 0; i < ndev; i++) { 3413 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3414 mode)) != 0) 3415 goto out; 3416 3417 if (!vd->vdev_ops->vdev_op_leaf) { 3418 vdev_free(vd); 3419 error = SET_ERROR(EINVAL); 3420 goto out; 3421 } 3422 3423 /* 3424 * The L2ARC currently only supports disk devices in 3425 * kernel context. For user-level testing, we allow it. 3426 */ 3427#ifdef _KERNEL 3428 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3429 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3430 error = SET_ERROR(ENOTBLK); 3431 vdev_free(vd); 3432 goto out; 3433 } 3434#endif 3435 vd->vdev_top = vd; 3436 3437 if ((error = vdev_open(vd)) == 0 && 3438 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3439 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3440 vd->vdev_guid) == 0); 3441 } 3442 3443 vdev_free(vd); 3444 3445 if (error && 3446 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3447 goto out; 3448 else 3449 error = 0; 3450 } 3451 3452out: 3453 sav->sav_pending = NULL; 3454 sav->sav_npending = 0; 3455 return (error); 3456} 3457 3458static int 3459spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3460{ 3461 int error; 3462 3463 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3464 3465 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3466 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3467 VDEV_LABEL_SPARE)) != 0) { 3468 return (error); 3469 } 3470 3471 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3472 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3473 VDEV_LABEL_L2CACHE)); 3474} 3475 3476static void 3477spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3478 const char *config) 3479{ 3480 int i; 3481 3482 if (sav->sav_config != NULL) { 3483 nvlist_t **olddevs; 3484 uint_t oldndevs; 3485 nvlist_t **newdevs; 3486 3487 /* 3488 * Generate new dev list by concatentating with the 3489 * current dev list. 3490 */ 3491 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3492 &olddevs, &oldndevs) == 0); 3493 3494 newdevs = kmem_alloc(sizeof (void *) * 3495 (ndevs + oldndevs), KM_SLEEP); 3496 for (i = 0; i < oldndevs; i++) 3497 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3498 KM_SLEEP) == 0); 3499 for (i = 0; i < ndevs; i++) 3500 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3501 KM_SLEEP) == 0); 3502 3503 VERIFY(nvlist_remove(sav->sav_config, config, 3504 DATA_TYPE_NVLIST_ARRAY) == 0); 3505 3506 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3507 config, newdevs, ndevs + oldndevs) == 0); 3508 for (i = 0; i < oldndevs + ndevs; i++) 3509 nvlist_free(newdevs[i]); 3510 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3511 } else { 3512 /* 3513 * Generate a new dev list. 3514 */ 3515 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3516 KM_SLEEP) == 0); 3517 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3518 devs, ndevs) == 0); 3519 } 3520} 3521 3522/* 3523 * Stop and drop level 2 ARC devices 3524 */ 3525void 3526spa_l2cache_drop(spa_t *spa) 3527{ 3528 vdev_t *vd; 3529 int i; 3530 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3531 3532 for (i = 0; i < sav->sav_count; i++) { 3533 uint64_t pool; 3534 3535 vd = sav->sav_vdevs[i]; 3536 ASSERT(vd != NULL); 3537 3538 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3539 pool != 0ULL && l2arc_vdev_present(vd)) 3540 l2arc_remove_vdev(vd); 3541 } 3542} 3543 3544/* 3545 * Pool Creation 3546 */ 3547int 3548spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3549 nvlist_t *zplprops) 3550{ 3551 spa_t *spa; 3552 char *altroot = NULL; 3553 vdev_t *rvd; 3554 dsl_pool_t *dp; 3555 dmu_tx_t *tx; 3556 int error = 0; 3557 uint64_t txg = TXG_INITIAL; 3558 nvlist_t **spares, **l2cache; 3559 uint_t nspares, nl2cache; 3560 uint64_t version, obj; 3561 boolean_t has_features; 3562 3563 /* 3564 * If this pool already exists, return failure. 3565 */ 3566 mutex_enter(&spa_namespace_lock); 3567 if (spa_lookup(pool) != NULL) { 3568 mutex_exit(&spa_namespace_lock); 3569 return (SET_ERROR(EEXIST)); 3570 } 3571 3572 /* 3573 * Allocate a new spa_t structure. 3574 */ 3575 (void) nvlist_lookup_string(props, 3576 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3577 spa = spa_add(pool, NULL, altroot); 3578 spa_activate(spa, spa_mode_global); 3579 3580 if (props && (error = spa_prop_validate(spa, props))) { 3581 spa_deactivate(spa); 3582 spa_remove(spa); 3583 mutex_exit(&spa_namespace_lock); 3584 return (error); 3585 } 3586 3587 has_features = B_FALSE; 3588 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3589 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3590 if (zpool_prop_feature(nvpair_name(elem))) 3591 has_features = B_TRUE; 3592 } 3593 3594 if (has_features || nvlist_lookup_uint64(props, 3595 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3596 version = SPA_VERSION; 3597 } 3598 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3599 3600 spa->spa_first_txg = txg; 3601 spa->spa_uberblock.ub_txg = txg - 1; 3602 spa->spa_uberblock.ub_version = version; 3603 spa->spa_ubsync = spa->spa_uberblock; 3604 3605 /* 3606 * Create "The Godfather" zio to hold all async IOs 3607 */ 3608 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3609 KM_SLEEP); 3610 for (int i = 0; i < max_ncpus; i++) { 3611 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3612 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3613 ZIO_FLAG_GODFATHER); 3614 } 3615 3616 /* 3617 * Create the root vdev. 3618 */ 3619 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3620 3621 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3622 3623 ASSERT(error != 0 || rvd != NULL); 3624 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3625 3626 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3627 error = SET_ERROR(EINVAL); 3628 3629 if (error == 0 && 3630 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3631 (error = spa_validate_aux(spa, nvroot, txg, 3632 VDEV_ALLOC_ADD)) == 0) { 3633 for (int c = 0; c < rvd->vdev_children; c++) { 3634 vdev_ashift_optimize(rvd->vdev_child[c]); 3635 vdev_metaslab_set_size(rvd->vdev_child[c]); 3636 vdev_expand(rvd->vdev_child[c], txg); 3637 } 3638 } 3639 3640 spa_config_exit(spa, SCL_ALL, FTAG); 3641 3642 if (error != 0) { 3643 spa_unload(spa); 3644 spa_deactivate(spa); 3645 spa_remove(spa); 3646 mutex_exit(&spa_namespace_lock); 3647 return (error); 3648 } 3649 3650 /* 3651 * Get the list of spares, if specified. 3652 */ 3653 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3654 &spares, &nspares) == 0) { 3655 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3656 KM_SLEEP) == 0); 3657 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3658 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3659 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3660 spa_load_spares(spa); 3661 spa_config_exit(spa, SCL_ALL, FTAG); 3662 spa->spa_spares.sav_sync = B_TRUE; 3663 } 3664 3665 /* 3666 * Get the list of level 2 cache devices, if specified. 3667 */ 3668 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3669 &l2cache, &nl2cache) == 0) { 3670 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3671 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3672 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3673 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3674 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3675 spa_load_l2cache(spa); 3676 spa_config_exit(spa, SCL_ALL, FTAG); 3677 spa->spa_l2cache.sav_sync = B_TRUE; 3678 } 3679 3680 spa->spa_is_initializing = B_TRUE; 3681 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3682 spa->spa_meta_objset = dp->dp_meta_objset; 3683 spa->spa_is_initializing = B_FALSE; 3684 3685 /* 3686 * Create DDTs (dedup tables). 3687 */ 3688 ddt_create(spa); 3689 3690 spa_update_dspace(spa); 3691 3692 tx = dmu_tx_create_assigned(dp, txg); 3693 3694 /* 3695 * Create the pool config object. 3696 */ 3697 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3698 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3699 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3700 3701 if (zap_add(spa->spa_meta_objset, 3702 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3703 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3704 cmn_err(CE_PANIC, "failed to add pool config"); 3705 } 3706 3707 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3708 spa_feature_create_zap_objects(spa, tx); 3709 3710 if (zap_add(spa->spa_meta_objset, 3711 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3712 sizeof (uint64_t), 1, &version, tx) != 0) { 3713 cmn_err(CE_PANIC, "failed to add pool version"); 3714 } 3715 3716 /* Newly created pools with the right version are always deflated. */ 3717 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3718 spa->spa_deflate = TRUE; 3719 if (zap_add(spa->spa_meta_objset, 3720 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3721 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3722 cmn_err(CE_PANIC, "failed to add deflate"); 3723 } 3724 } 3725 3726 /* 3727 * Create the deferred-free bpobj. Turn off compression 3728 * because sync-to-convergence takes longer if the blocksize 3729 * keeps changing. 3730 */ 3731 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3732 dmu_object_set_compress(spa->spa_meta_objset, obj, 3733 ZIO_COMPRESS_OFF, tx); 3734 if (zap_add(spa->spa_meta_objset, 3735 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3736 sizeof (uint64_t), 1, &obj, tx) != 0) { 3737 cmn_err(CE_PANIC, "failed to add bpobj"); 3738 } 3739 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3740 spa->spa_meta_objset, obj)); 3741 3742 /* 3743 * Create the pool's history object. 3744 */ 3745 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3746 spa_history_create_obj(spa, tx); 3747 3748 /* 3749 * Set pool properties. 3750 */ 3751 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3752 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3753 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3754 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3755 3756 if (props != NULL) { 3757 spa_configfile_set(spa, props, B_FALSE); 3758 spa_sync_props(props, tx); 3759 } 3760 3761 dmu_tx_commit(tx); 3762 3763 spa->spa_sync_on = B_TRUE; 3764 txg_sync_start(spa->spa_dsl_pool); 3765 3766 /* 3767 * We explicitly wait for the first transaction to complete so that our 3768 * bean counters are appropriately updated. 3769 */ 3770 txg_wait_synced(spa->spa_dsl_pool, txg); 3771 3772 spa_config_sync(spa, B_FALSE, B_TRUE); 3773 3774 spa_history_log_version(spa, "create"); 3775 3776 /* 3777 * Don't count references from objsets that are already closed 3778 * and are making their way through the eviction process. 3779 */ 3780 spa_evicting_os_wait(spa); 3781 spa->spa_minref = refcount_count(&spa->spa_refcount); 3782 3783 mutex_exit(&spa_namespace_lock); 3784 3785 return (0); 3786} 3787 3788#ifdef _KERNEL 3789#if defined(sun) 3790/* 3791 * Get the root pool information from the root disk, then import the root pool 3792 * during the system boot up time. 3793 */ 3794extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3795 3796static nvlist_t * 3797spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3798{ 3799 nvlist_t *config; 3800 nvlist_t *nvtop, *nvroot; 3801 uint64_t pgid; 3802 3803 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3804 return (NULL); 3805 3806 /* 3807 * Add this top-level vdev to the child array. 3808 */ 3809 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3810 &nvtop) == 0); 3811 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3812 &pgid) == 0); 3813 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3814 3815 /* 3816 * Put this pool's top-level vdevs into a root vdev. 3817 */ 3818 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3819 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3820 VDEV_TYPE_ROOT) == 0); 3821 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3822 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3823 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3824 &nvtop, 1) == 0); 3825 3826 /* 3827 * Replace the existing vdev_tree with the new root vdev in 3828 * this pool's configuration (remove the old, add the new). 3829 */ 3830 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3831 nvlist_free(nvroot); 3832 return (config); 3833} 3834 3835/* 3836 * Walk the vdev tree and see if we can find a device with "better" 3837 * configuration. A configuration is "better" if the label on that 3838 * device has a more recent txg. 3839 */ 3840static void 3841spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3842{ 3843 for (int c = 0; c < vd->vdev_children; c++) 3844 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3845 3846 if (vd->vdev_ops->vdev_op_leaf) { 3847 nvlist_t *label; 3848 uint64_t label_txg; 3849 3850 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3851 &label) != 0) 3852 return; 3853 3854 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3855 &label_txg) == 0); 3856 3857 /* 3858 * Do we have a better boot device? 3859 */ 3860 if (label_txg > *txg) { 3861 *txg = label_txg; 3862 *avd = vd; 3863 } 3864 nvlist_free(label); 3865 } 3866} 3867 3868/* 3869 * Import a root pool. 3870 * 3871 * For x86. devpath_list will consist of devid and/or physpath name of 3872 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3873 * The GRUB "findroot" command will return the vdev we should boot. 3874 * 3875 * For Sparc, devpath_list consists the physpath name of the booting device 3876 * no matter the rootpool is a single device pool or a mirrored pool. 3877 * e.g. 3878 * "/pci@1f,0/ide@d/disk@0,0:a" 3879 */ 3880int 3881spa_import_rootpool(char *devpath, char *devid) 3882{ 3883 spa_t *spa; 3884 vdev_t *rvd, *bvd, *avd = NULL; 3885 nvlist_t *config, *nvtop; 3886 uint64_t guid, txg; 3887 char *pname; 3888 int error; 3889 3890 /* 3891 * Read the label from the boot device and generate a configuration. 3892 */ 3893 config = spa_generate_rootconf(devpath, devid, &guid); 3894#if defined(_OBP) && defined(_KERNEL) 3895 if (config == NULL) { 3896 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3897 /* iscsi boot */ 3898 get_iscsi_bootpath_phy(devpath); 3899 config = spa_generate_rootconf(devpath, devid, &guid); 3900 } 3901 } 3902#endif 3903 if (config == NULL) { 3904 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3905 devpath); 3906 return (SET_ERROR(EIO)); 3907 } 3908 3909 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3910 &pname) == 0); 3911 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3912 3913 mutex_enter(&spa_namespace_lock); 3914 if ((spa = spa_lookup(pname)) != NULL) { 3915 /* 3916 * Remove the existing root pool from the namespace so that we 3917 * can replace it with the correct config we just read in. 3918 */ 3919 spa_remove(spa); 3920 } 3921 3922 spa = spa_add(pname, config, NULL); 3923 spa->spa_is_root = B_TRUE; 3924 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3925 3926 /* 3927 * Build up a vdev tree based on the boot device's label config. 3928 */ 3929 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3930 &nvtop) == 0); 3931 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3932 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3933 VDEV_ALLOC_ROOTPOOL); 3934 spa_config_exit(spa, SCL_ALL, FTAG); 3935 if (error) { 3936 mutex_exit(&spa_namespace_lock); 3937 nvlist_free(config); 3938 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3939 pname); 3940 return (error); 3941 } 3942 3943 /* 3944 * Get the boot vdev. 3945 */ 3946 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3947 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3948 (u_longlong_t)guid); 3949 error = SET_ERROR(ENOENT); 3950 goto out; 3951 } 3952 3953 /* 3954 * Determine if there is a better boot device. 3955 */ 3956 avd = bvd; 3957 spa_alt_rootvdev(rvd, &avd, &txg); 3958 if (avd != bvd) { 3959 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3960 "try booting from '%s'", avd->vdev_path); 3961 error = SET_ERROR(EINVAL); 3962 goto out; 3963 } 3964 3965 /* 3966 * If the boot device is part of a spare vdev then ensure that 3967 * we're booting off the active spare. 3968 */ 3969 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3970 !bvd->vdev_isspare) { 3971 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3972 "try booting from '%s'", 3973 bvd->vdev_parent-> 3974 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3975 error = SET_ERROR(EINVAL); 3976 goto out; 3977 } 3978 3979 error = 0; 3980out: 3981 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3982 vdev_free(rvd); 3983 spa_config_exit(spa, SCL_ALL, FTAG); 3984 mutex_exit(&spa_namespace_lock); 3985 3986 nvlist_free(config); 3987 return (error); 3988} 3989 3990#else 3991 3992extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3993 uint64_t *count); 3994 3995static nvlist_t * 3996spa_generate_rootconf(const char *name) 3997{ 3998 nvlist_t **configs, **tops; 3999 nvlist_t *config; 4000 nvlist_t *best_cfg, *nvtop, *nvroot; 4001 uint64_t *holes; 4002 uint64_t best_txg; 4003 uint64_t nchildren; 4004 uint64_t pgid; 4005 uint64_t count; 4006 uint64_t i; 4007 uint_t nholes; 4008 4009 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4010 return (NULL); 4011 4012 ASSERT3U(count, !=, 0); 4013 best_txg = 0; 4014 for (i = 0; i < count; i++) { 4015 uint64_t txg; 4016 4017 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4018 &txg) == 0); 4019 if (txg > best_txg) { 4020 best_txg = txg; 4021 best_cfg = configs[i]; 4022 } 4023 } 4024 4025 /* 4026 * Multi-vdev root pool configuration discovery is not supported yet. 4027 */ 4028 nchildren = 1; 4029 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4030 holes = NULL; 4031 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4032 &holes, &nholes); 4033 4034 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4035 for (i = 0; i < nchildren; i++) { 4036 if (i >= count) 4037 break; 4038 if (configs[i] == NULL) 4039 continue; 4040 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4041 &nvtop) == 0); 4042 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4043 } 4044 for (i = 0; holes != NULL && i < nholes; i++) { 4045 if (i >= nchildren) 4046 continue; 4047 if (tops[holes[i]] != NULL) 4048 continue; 4049 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4050 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4051 VDEV_TYPE_HOLE) == 0); 4052 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4053 holes[i]) == 0); 4054 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4055 0) == 0); 4056 } 4057 for (i = 0; i < nchildren; i++) { 4058 if (tops[i] != NULL) 4059 continue; 4060 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4061 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4062 VDEV_TYPE_MISSING) == 0); 4063 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4064 i) == 0); 4065 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4066 0) == 0); 4067 } 4068 4069 /* 4070 * Create pool config based on the best vdev config. 4071 */ 4072 nvlist_dup(best_cfg, &config, KM_SLEEP); 4073 4074 /* 4075 * Put this pool's top-level vdevs into a root vdev. 4076 */ 4077 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4078 &pgid) == 0); 4079 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4080 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4081 VDEV_TYPE_ROOT) == 0); 4082 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4083 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4084 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4085 tops, nchildren) == 0); 4086 4087 /* 4088 * Replace the existing vdev_tree with the new root vdev in 4089 * this pool's configuration (remove the old, add the new). 4090 */ 4091 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4092 4093 /* 4094 * Drop vdev config elements that should not be present at pool level. 4095 */ 4096 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4097 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4098 4099 for (i = 0; i < count; i++) 4100 nvlist_free(configs[i]); 4101 kmem_free(configs, count * sizeof(void *)); 4102 for (i = 0; i < nchildren; i++) 4103 nvlist_free(tops[i]); 4104 kmem_free(tops, nchildren * sizeof(void *)); 4105 nvlist_free(nvroot); 4106 return (config); 4107} 4108 4109int 4110spa_import_rootpool(const char *name) 4111{ 4112 spa_t *spa; 4113 vdev_t *rvd, *bvd, *avd = NULL; 4114 nvlist_t *config, *nvtop; 4115 uint64_t txg; 4116 char *pname; 4117 int error; 4118 4119 /* 4120 * Read the label from the boot device and generate a configuration. 4121 */ 4122 config = spa_generate_rootconf(name); 4123 4124 mutex_enter(&spa_namespace_lock); 4125 if (config != NULL) { 4126 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4127 &pname) == 0 && strcmp(name, pname) == 0); 4128 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4129 == 0); 4130 4131 if ((spa = spa_lookup(pname)) != NULL) { 4132 /* 4133 * Remove the existing root pool from the namespace so 4134 * that we can replace it with the correct config 4135 * we just read in. 4136 */ 4137 spa_remove(spa); 4138 } 4139 spa = spa_add(pname, config, NULL); 4140 4141 /* 4142 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4143 * via spa_version(). 4144 */ 4145 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4146 &spa->spa_ubsync.ub_version) != 0) 4147 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4148 } else if ((spa = spa_lookup(name)) == NULL) { 4149 mutex_exit(&spa_namespace_lock); 4150 nvlist_free(config); 4151 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4152 name); 4153 return (EIO); 4154 } else { 4155 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4156 } 4157 spa->spa_is_root = B_TRUE; 4158 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4159 4160 /* 4161 * Build up a vdev tree based on the boot device's label config. 4162 */ 4163 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4164 &nvtop) == 0); 4165 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4166 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4167 VDEV_ALLOC_ROOTPOOL); 4168 spa_config_exit(spa, SCL_ALL, FTAG); 4169 if (error) { 4170 mutex_exit(&spa_namespace_lock); 4171 nvlist_free(config); 4172 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4173 pname); 4174 return (error); 4175 } 4176 4177 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4178 vdev_free(rvd); 4179 spa_config_exit(spa, SCL_ALL, FTAG); 4180 mutex_exit(&spa_namespace_lock); 4181 4182 nvlist_free(config); 4183 return (0); 4184} 4185 4186#endif /* sun */ 4187#endif 4188 4189/* 4190 * Import a non-root pool into the system. 4191 */ 4192int 4193spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4194{ 4195 spa_t *spa; 4196 char *altroot = NULL; 4197 spa_load_state_t state = SPA_LOAD_IMPORT; 4198 zpool_rewind_policy_t policy; 4199 uint64_t mode = spa_mode_global; 4200 uint64_t readonly = B_FALSE; 4201 int error; 4202 nvlist_t *nvroot; 4203 nvlist_t **spares, **l2cache; 4204 uint_t nspares, nl2cache; 4205 4206 /* 4207 * If a pool with this name exists, return failure. 4208 */ 4209 mutex_enter(&spa_namespace_lock); 4210 if (spa_lookup(pool) != NULL) { 4211 mutex_exit(&spa_namespace_lock); 4212 return (SET_ERROR(EEXIST)); 4213 } 4214 4215 /* 4216 * Create and initialize the spa structure. 4217 */ 4218 (void) nvlist_lookup_string(props, 4219 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4220 (void) nvlist_lookup_uint64(props, 4221 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4222 if (readonly) 4223 mode = FREAD; 4224 spa = spa_add(pool, config, altroot); 4225 spa->spa_import_flags = flags; 4226 4227 /* 4228 * Verbatim import - Take a pool and insert it into the namespace 4229 * as if it had been loaded at boot. 4230 */ 4231 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4232 if (props != NULL) 4233 spa_configfile_set(spa, props, B_FALSE); 4234 4235 spa_config_sync(spa, B_FALSE, B_TRUE); 4236 4237 mutex_exit(&spa_namespace_lock); 4238 return (0); 4239 } 4240 4241 spa_activate(spa, mode); 4242 4243 /* 4244 * Don't start async tasks until we know everything is healthy. 4245 */ 4246 spa_async_suspend(spa); 4247 4248 zpool_get_rewind_policy(config, &policy); 4249 if (policy.zrp_request & ZPOOL_DO_REWIND) 4250 state = SPA_LOAD_RECOVER; 4251 4252 /* 4253 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4254 * because the user-supplied config is actually the one to trust when 4255 * doing an import. 4256 */ 4257 if (state != SPA_LOAD_RECOVER) 4258 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4259 4260 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4261 policy.zrp_request); 4262 4263 /* 4264 * Propagate anything learned while loading the pool and pass it 4265 * back to caller (i.e. rewind info, missing devices, etc). 4266 */ 4267 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4268 spa->spa_load_info) == 0); 4269 4270 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4271 /* 4272 * Toss any existing sparelist, as it doesn't have any validity 4273 * anymore, and conflicts with spa_has_spare(). 4274 */ 4275 if (spa->spa_spares.sav_config) { 4276 nvlist_free(spa->spa_spares.sav_config); 4277 spa->spa_spares.sav_config = NULL; 4278 spa_load_spares(spa); 4279 } 4280 if (spa->spa_l2cache.sav_config) { 4281 nvlist_free(spa->spa_l2cache.sav_config); 4282 spa->spa_l2cache.sav_config = NULL; 4283 spa_load_l2cache(spa); 4284 } 4285 4286 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4287 &nvroot) == 0); 4288 if (error == 0) 4289 error = spa_validate_aux(spa, nvroot, -1ULL, 4290 VDEV_ALLOC_SPARE); 4291 if (error == 0) 4292 error = spa_validate_aux(spa, nvroot, -1ULL, 4293 VDEV_ALLOC_L2CACHE); 4294 spa_config_exit(spa, SCL_ALL, FTAG); 4295 4296 if (props != NULL) 4297 spa_configfile_set(spa, props, B_FALSE); 4298 4299 if (error != 0 || (props && spa_writeable(spa) && 4300 (error = spa_prop_set(spa, props)))) { 4301 spa_unload(spa); 4302 spa_deactivate(spa); 4303 spa_remove(spa); 4304 mutex_exit(&spa_namespace_lock); 4305 return (error); 4306 } 4307 4308 spa_async_resume(spa); 4309 4310 /* 4311 * Override any spares and level 2 cache devices as specified by 4312 * the user, as these may have correct device names/devids, etc. 4313 */ 4314 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4315 &spares, &nspares) == 0) { 4316 if (spa->spa_spares.sav_config) 4317 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4318 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4319 else 4320 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4321 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4322 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4323 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4324 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4325 spa_load_spares(spa); 4326 spa_config_exit(spa, SCL_ALL, FTAG); 4327 spa->spa_spares.sav_sync = B_TRUE; 4328 } 4329 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4330 &l2cache, &nl2cache) == 0) { 4331 if (spa->spa_l2cache.sav_config) 4332 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4333 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4334 else 4335 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4336 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4337 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4338 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4339 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4340 spa_load_l2cache(spa); 4341 spa_config_exit(spa, SCL_ALL, FTAG); 4342 spa->spa_l2cache.sav_sync = B_TRUE; 4343 } 4344 4345 /* 4346 * Check for any removed devices. 4347 */ 4348 if (spa->spa_autoreplace) { 4349 spa_aux_check_removed(&spa->spa_spares); 4350 spa_aux_check_removed(&spa->spa_l2cache); 4351 } 4352 4353 if (spa_writeable(spa)) { 4354 /* 4355 * Update the config cache to include the newly-imported pool. 4356 */ 4357 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4358 } 4359 4360 /* 4361 * It's possible that the pool was expanded while it was exported. 4362 * We kick off an async task to handle this for us. 4363 */ 4364 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4365 4366 mutex_exit(&spa_namespace_lock); 4367 spa_history_log_version(spa, "import"); 4368 4369#ifdef __FreeBSD__ 4370#ifdef _KERNEL 4371 zvol_create_minors(pool); 4372#endif 4373#endif 4374 return (0); 4375} 4376 4377nvlist_t * 4378spa_tryimport(nvlist_t *tryconfig) 4379{ 4380 nvlist_t *config = NULL; 4381 char *poolname; 4382 spa_t *spa; 4383 uint64_t state; 4384 int error; 4385 4386 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4387 return (NULL); 4388 4389 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4390 return (NULL); 4391 4392 /* 4393 * Create and initialize the spa structure. 4394 */ 4395 mutex_enter(&spa_namespace_lock); 4396 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4397 spa_activate(spa, FREAD); 4398 4399 /* 4400 * Pass off the heavy lifting to spa_load(). 4401 * Pass TRUE for mosconfig because the user-supplied config 4402 * is actually the one to trust when doing an import. 4403 */ 4404 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4405 4406 /* 4407 * If 'tryconfig' was at least parsable, return the current config. 4408 */ 4409 if (spa->spa_root_vdev != NULL) { 4410 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4411 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4412 poolname) == 0); 4413 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4414 state) == 0); 4415 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4416 spa->spa_uberblock.ub_timestamp) == 0); 4417 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4418 spa->spa_load_info) == 0); 4419 4420 /* 4421 * If the bootfs property exists on this pool then we 4422 * copy it out so that external consumers can tell which 4423 * pools are bootable. 4424 */ 4425 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4426 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4427 4428 /* 4429 * We have to play games with the name since the 4430 * pool was opened as TRYIMPORT_NAME. 4431 */ 4432 if (dsl_dsobj_to_dsname(spa_name(spa), 4433 spa->spa_bootfs, tmpname) == 0) { 4434 char *cp; 4435 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4436 4437 cp = strchr(tmpname, '/'); 4438 if (cp == NULL) { 4439 (void) strlcpy(dsname, tmpname, 4440 MAXPATHLEN); 4441 } else { 4442 (void) snprintf(dsname, MAXPATHLEN, 4443 "%s/%s", poolname, ++cp); 4444 } 4445 VERIFY(nvlist_add_string(config, 4446 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4447 kmem_free(dsname, MAXPATHLEN); 4448 } 4449 kmem_free(tmpname, MAXPATHLEN); 4450 } 4451 4452 /* 4453 * Add the list of hot spares and level 2 cache devices. 4454 */ 4455 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4456 spa_add_spares(spa, config); 4457 spa_add_l2cache(spa, config); 4458 spa_config_exit(spa, SCL_CONFIG, FTAG); 4459 } 4460 4461 spa_unload(spa); 4462 spa_deactivate(spa); 4463 spa_remove(spa); 4464 mutex_exit(&spa_namespace_lock); 4465 4466 return (config); 4467} 4468 4469/* 4470 * Pool export/destroy 4471 * 4472 * The act of destroying or exporting a pool is very simple. We make sure there 4473 * is no more pending I/O and any references to the pool are gone. Then, we 4474 * update the pool state and sync all the labels to disk, removing the 4475 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4476 * we don't sync the labels or remove the configuration cache. 4477 */ 4478static int 4479spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4480 boolean_t force, boolean_t hardforce) 4481{ 4482 spa_t *spa; 4483 4484 if (oldconfig) 4485 *oldconfig = NULL; 4486 4487 if (!(spa_mode_global & FWRITE)) 4488 return (SET_ERROR(EROFS)); 4489 4490 mutex_enter(&spa_namespace_lock); 4491 if ((spa = spa_lookup(pool)) == NULL) { 4492 mutex_exit(&spa_namespace_lock); 4493 return (SET_ERROR(ENOENT)); 4494 } 4495 4496 /* 4497 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4498 * reacquire the namespace lock, and see if we can export. 4499 */ 4500 spa_open_ref(spa, FTAG); 4501 mutex_exit(&spa_namespace_lock); 4502 spa_async_suspend(spa); 4503 mutex_enter(&spa_namespace_lock); 4504 spa_close(spa, FTAG); 4505 4506 /* 4507 * The pool will be in core if it's openable, 4508 * in which case we can modify its state. 4509 */ 4510 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4511 /* 4512 * Objsets may be open only because they're dirty, so we 4513 * have to force it to sync before checking spa_refcnt. 4514 */ 4515 txg_wait_synced(spa->spa_dsl_pool, 0); 4516 spa_evicting_os_wait(spa); 4517 4518 /* 4519 * A pool cannot be exported or destroyed if there are active 4520 * references. If we are resetting a pool, allow references by 4521 * fault injection handlers. 4522 */ 4523 if (!spa_refcount_zero(spa) || 4524 (spa->spa_inject_ref != 0 && 4525 new_state != POOL_STATE_UNINITIALIZED)) { 4526 spa_async_resume(spa); 4527 mutex_exit(&spa_namespace_lock); 4528 return (SET_ERROR(EBUSY)); 4529 } 4530 4531 /* 4532 * A pool cannot be exported if it has an active shared spare. 4533 * This is to prevent other pools stealing the active spare 4534 * from an exported pool. At user's own will, such pool can 4535 * be forcedly exported. 4536 */ 4537 if (!force && new_state == POOL_STATE_EXPORTED && 4538 spa_has_active_shared_spare(spa)) { 4539 spa_async_resume(spa); 4540 mutex_exit(&spa_namespace_lock); 4541 return (SET_ERROR(EXDEV)); 4542 } 4543 4544 /* 4545 * We want this to be reflected on every label, 4546 * so mark them all dirty. spa_unload() will do the 4547 * final sync that pushes these changes out. 4548 */ 4549 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4550 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4551 spa->spa_state = new_state; 4552 spa->spa_final_txg = spa_last_synced_txg(spa) + 4553 TXG_DEFER_SIZE + 1; 4554 vdev_config_dirty(spa->spa_root_vdev); 4555 spa_config_exit(spa, SCL_ALL, FTAG); 4556 } 4557 } 4558 4559 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4560 4561 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4562 spa_unload(spa); 4563 spa_deactivate(spa); 4564 } 4565 4566 if (oldconfig && spa->spa_config) 4567 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4568 4569 if (new_state != POOL_STATE_UNINITIALIZED) { 4570 if (!hardforce) 4571 spa_config_sync(spa, B_TRUE, B_TRUE); 4572 spa_remove(spa); 4573 } 4574 mutex_exit(&spa_namespace_lock); 4575 4576 return (0); 4577} 4578 4579/* 4580 * Destroy a storage pool. 4581 */ 4582int 4583spa_destroy(char *pool) 4584{ 4585 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4586 B_FALSE, B_FALSE)); 4587} 4588 4589/* 4590 * Export a storage pool. 4591 */ 4592int 4593spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4594 boolean_t hardforce) 4595{ 4596 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4597 force, hardforce)); 4598} 4599 4600/* 4601 * Similar to spa_export(), this unloads the spa_t without actually removing it 4602 * from the namespace in any way. 4603 */ 4604int 4605spa_reset(char *pool) 4606{ 4607 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4608 B_FALSE, B_FALSE)); 4609} 4610 4611/* 4612 * ========================================================================== 4613 * Device manipulation 4614 * ========================================================================== 4615 */ 4616 4617/* 4618 * Add a device to a storage pool. 4619 */ 4620int 4621spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4622{ 4623 uint64_t txg, id; 4624 int error; 4625 vdev_t *rvd = spa->spa_root_vdev; 4626 vdev_t *vd, *tvd; 4627 nvlist_t **spares, **l2cache; 4628 uint_t nspares, nl2cache; 4629 4630 ASSERT(spa_writeable(spa)); 4631 4632 txg = spa_vdev_enter(spa); 4633 4634 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4635 VDEV_ALLOC_ADD)) != 0) 4636 return (spa_vdev_exit(spa, NULL, txg, error)); 4637 4638 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4639 4640 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4641 &nspares) != 0) 4642 nspares = 0; 4643 4644 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4645 &nl2cache) != 0) 4646 nl2cache = 0; 4647 4648 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4649 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4650 4651 if (vd->vdev_children != 0 && 4652 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4653 return (spa_vdev_exit(spa, vd, txg, error)); 4654 4655 /* 4656 * We must validate the spares and l2cache devices after checking the 4657 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4658 */ 4659 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4660 return (spa_vdev_exit(spa, vd, txg, error)); 4661 4662 /* 4663 * Transfer each new top-level vdev from vd to rvd. 4664 */ 4665 for (int c = 0; c < vd->vdev_children; c++) { 4666 4667 /* 4668 * Set the vdev id to the first hole, if one exists. 4669 */ 4670 for (id = 0; id < rvd->vdev_children; id++) { 4671 if (rvd->vdev_child[id]->vdev_ishole) { 4672 vdev_free(rvd->vdev_child[id]); 4673 break; 4674 } 4675 } 4676 tvd = vd->vdev_child[c]; 4677 vdev_remove_child(vd, tvd); 4678 tvd->vdev_id = id; 4679 vdev_add_child(rvd, tvd); 4680 vdev_config_dirty(tvd); 4681 } 4682 4683 if (nspares != 0) { 4684 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4685 ZPOOL_CONFIG_SPARES); 4686 spa_load_spares(spa); 4687 spa->spa_spares.sav_sync = B_TRUE; 4688 } 4689 4690 if (nl2cache != 0) { 4691 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4692 ZPOOL_CONFIG_L2CACHE); 4693 spa_load_l2cache(spa); 4694 spa->spa_l2cache.sav_sync = B_TRUE; 4695 } 4696 4697 /* 4698 * We have to be careful when adding new vdevs to an existing pool. 4699 * If other threads start allocating from these vdevs before we 4700 * sync the config cache, and we lose power, then upon reboot we may 4701 * fail to open the pool because there are DVAs that the config cache 4702 * can't translate. Therefore, we first add the vdevs without 4703 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4704 * and then let spa_config_update() initialize the new metaslabs. 4705 * 4706 * spa_load() checks for added-but-not-initialized vdevs, so that 4707 * if we lose power at any point in this sequence, the remaining 4708 * steps will be completed the next time we load the pool. 4709 */ 4710 (void) spa_vdev_exit(spa, vd, txg, 0); 4711 4712 mutex_enter(&spa_namespace_lock); 4713 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4714 mutex_exit(&spa_namespace_lock); 4715 4716 return (0); 4717} 4718 4719/* 4720 * Attach a device to a mirror. The arguments are the path to any device 4721 * in the mirror, and the nvroot for the new device. If the path specifies 4722 * a device that is not mirrored, we automatically insert the mirror vdev. 4723 * 4724 * If 'replacing' is specified, the new device is intended to replace the 4725 * existing device; in this case the two devices are made into their own 4726 * mirror using the 'replacing' vdev, which is functionally identical to 4727 * the mirror vdev (it actually reuses all the same ops) but has a few 4728 * extra rules: you can't attach to it after it's been created, and upon 4729 * completion of resilvering, the first disk (the one being replaced) 4730 * is automatically detached. 4731 */ 4732int 4733spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4734{ 4735 uint64_t txg, dtl_max_txg; 4736 vdev_t *rvd = spa->spa_root_vdev; 4737 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4738 vdev_ops_t *pvops; 4739 char *oldvdpath, *newvdpath; 4740 int newvd_isspare; 4741 int error; 4742 4743 ASSERT(spa_writeable(spa)); 4744 4745 txg = spa_vdev_enter(spa); 4746 4747 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4748 4749 if (oldvd == NULL) 4750 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4751 4752 if (!oldvd->vdev_ops->vdev_op_leaf) 4753 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4754 4755 pvd = oldvd->vdev_parent; 4756 4757 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4758 VDEV_ALLOC_ATTACH)) != 0) 4759 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4760 4761 if (newrootvd->vdev_children != 1) 4762 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4763 4764 newvd = newrootvd->vdev_child[0]; 4765 4766 if (!newvd->vdev_ops->vdev_op_leaf) 4767 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4768 4769 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4770 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4771 4772 /* 4773 * Spares can't replace logs 4774 */ 4775 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4776 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4777 4778 if (!replacing) { 4779 /* 4780 * For attach, the only allowable parent is a mirror or the root 4781 * vdev. 4782 */ 4783 if (pvd->vdev_ops != &vdev_mirror_ops && 4784 pvd->vdev_ops != &vdev_root_ops) 4785 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4786 4787 pvops = &vdev_mirror_ops; 4788 } else { 4789 /* 4790 * Active hot spares can only be replaced by inactive hot 4791 * spares. 4792 */ 4793 if (pvd->vdev_ops == &vdev_spare_ops && 4794 oldvd->vdev_isspare && 4795 !spa_has_spare(spa, newvd->vdev_guid)) 4796 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4797 4798 /* 4799 * If the source is a hot spare, and the parent isn't already a 4800 * spare, then we want to create a new hot spare. Otherwise, we 4801 * want to create a replacing vdev. The user is not allowed to 4802 * attach to a spared vdev child unless the 'isspare' state is 4803 * the same (spare replaces spare, non-spare replaces 4804 * non-spare). 4805 */ 4806 if (pvd->vdev_ops == &vdev_replacing_ops && 4807 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4808 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4809 } else if (pvd->vdev_ops == &vdev_spare_ops && 4810 newvd->vdev_isspare != oldvd->vdev_isspare) { 4811 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4812 } 4813 4814 if (newvd->vdev_isspare) 4815 pvops = &vdev_spare_ops; 4816 else 4817 pvops = &vdev_replacing_ops; 4818 } 4819 4820 /* 4821 * Make sure the new device is big enough. 4822 */ 4823 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4824 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4825 4826 /* 4827 * The new device cannot have a higher alignment requirement 4828 * than the top-level vdev. 4829 */ 4830 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4831 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4832 4833 /* 4834 * If this is an in-place replacement, update oldvd's path and devid 4835 * to make it distinguishable from newvd, and unopenable from now on. 4836 */ 4837 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4838 spa_strfree(oldvd->vdev_path); 4839 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4840 KM_SLEEP); 4841 (void) sprintf(oldvd->vdev_path, "%s/%s", 4842 newvd->vdev_path, "old"); 4843 if (oldvd->vdev_devid != NULL) { 4844 spa_strfree(oldvd->vdev_devid); 4845 oldvd->vdev_devid = NULL; 4846 } 4847 } 4848 4849 /* mark the device being resilvered */ 4850 newvd->vdev_resilver_txg = txg; 4851 4852 /* 4853 * If the parent is not a mirror, or if we're replacing, insert the new 4854 * mirror/replacing/spare vdev above oldvd. 4855 */ 4856 if (pvd->vdev_ops != pvops) 4857 pvd = vdev_add_parent(oldvd, pvops); 4858 4859 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4860 ASSERT(pvd->vdev_ops == pvops); 4861 ASSERT(oldvd->vdev_parent == pvd); 4862 4863 /* 4864 * Extract the new device from its root and add it to pvd. 4865 */ 4866 vdev_remove_child(newrootvd, newvd); 4867 newvd->vdev_id = pvd->vdev_children; 4868 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4869 vdev_add_child(pvd, newvd); 4870 4871 tvd = newvd->vdev_top; 4872 ASSERT(pvd->vdev_top == tvd); 4873 ASSERT(tvd->vdev_parent == rvd); 4874 4875 vdev_config_dirty(tvd); 4876 4877 /* 4878 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4879 * for any dmu_sync-ed blocks. It will propagate upward when 4880 * spa_vdev_exit() calls vdev_dtl_reassess(). 4881 */ 4882 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4883 4884 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4885 dtl_max_txg - TXG_INITIAL); 4886 4887 if (newvd->vdev_isspare) { 4888 spa_spare_activate(newvd); 4889 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4890 } 4891 4892 oldvdpath = spa_strdup(oldvd->vdev_path); 4893 newvdpath = spa_strdup(newvd->vdev_path); 4894 newvd_isspare = newvd->vdev_isspare; 4895 4896 /* 4897 * Mark newvd's DTL dirty in this txg. 4898 */ 4899 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4900 4901 /* 4902 * Schedule the resilver to restart in the future. We do this to 4903 * ensure that dmu_sync-ed blocks have been stitched into the 4904 * respective datasets. 4905 */ 4906 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4907 4908 /* 4909 * Commit the config 4910 */ 4911 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4912 4913 spa_history_log_internal(spa, "vdev attach", NULL, 4914 "%s vdev=%s %s vdev=%s", 4915 replacing && newvd_isspare ? "spare in" : 4916 replacing ? "replace" : "attach", newvdpath, 4917 replacing ? "for" : "to", oldvdpath); 4918 4919 spa_strfree(oldvdpath); 4920 spa_strfree(newvdpath); 4921 4922 if (spa->spa_bootfs) 4923 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4924 4925 return (0); 4926} 4927 4928/* 4929 * Detach a device from a mirror or replacing vdev. 4930 * 4931 * If 'replace_done' is specified, only detach if the parent 4932 * is a replacing vdev. 4933 */ 4934int 4935spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4936{ 4937 uint64_t txg; 4938 int error; 4939 vdev_t *rvd = spa->spa_root_vdev; 4940 vdev_t *vd, *pvd, *cvd, *tvd; 4941 boolean_t unspare = B_FALSE; 4942 uint64_t unspare_guid = 0; 4943 char *vdpath; 4944 4945 ASSERT(spa_writeable(spa)); 4946 4947 txg = spa_vdev_enter(spa); 4948 4949 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4950 4951 if (vd == NULL) 4952 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4953 4954 if (!vd->vdev_ops->vdev_op_leaf) 4955 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4956 4957 pvd = vd->vdev_parent; 4958 4959 /* 4960 * If the parent/child relationship is not as expected, don't do it. 4961 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4962 * vdev that's replacing B with C. The user's intent in replacing 4963 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4964 * the replace by detaching C, the expected behavior is to end up 4965 * M(A,B). But suppose that right after deciding to detach C, 4966 * the replacement of B completes. We would have M(A,C), and then 4967 * ask to detach C, which would leave us with just A -- not what 4968 * the user wanted. To prevent this, we make sure that the 4969 * parent/child relationship hasn't changed -- in this example, 4970 * that C's parent is still the replacing vdev R. 4971 */ 4972 if (pvd->vdev_guid != pguid && pguid != 0) 4973 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4974 4975 /* 4976 * Only 'replacing' or 'spare' vdevs can be replaced. 4977 */ 4978 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4979 pvd->vdev_ops != &vdev_spare_ops) 4980 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4981 4982 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4983 spa_version(spa) >= SPA_VERSION_SPARES); 4984 4985 /* 4986 * Only mirror, replacing, and spare vdevs support detach. 4987 */ 4988 if (pvd->vdev_ops != &vdev_replacing_ops && 4989 pvd->vdev_ops != &vdev_mirror_ops && 4990 pvd->vdev_ops != &vdev_spare_ops) 4991 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4992 4993 /* 4994 * If this device has the only valid copy of some data, 4995 * we cannot safely detach it. 4996 */ 4997 if (vdev_dtl_required(vd)) 4998 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4999 5000 ASSERT(pvd->vdev_children >= 2); 5001 5002 /* 5003 * If we are detaching the second disk from a replacing vdev, then 5004 * check to see if we changed the original vdev's path to have "/old" 5005 * at the end in spa_vdev_attach(). If so, undo that change now. 5006 */ 5007 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5008 vd->vdev_path != NULL) { 5009 size_t len = strlen(vd->vdev_path); 5010 5011 for (int c = 0; c < pvd->vdev_children; c++) { 5012 cvd = pvd->vdev_child[c]; 5013 5014 if (cvd == vd || cvd->vdev_path == NULL) 5015 continue; 5016 5017 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5018 strcmp(cvd->vdev_path + len, "/old") == 0) { 5019 spa_strfree(cvd->vdev_path); 5020 cvd->vdev_path = spa_strdup(vd->vdev_path); 5021 break; 5022 } 5023 } 5024 } 5025 5026 /* 5027 * If we are detaching the original disk from a spare, then it implies 5028 * that the spare should become a real disk, and be removed from the 5029 * active spare list for the pool. 5030 */ 5031 if (pvd->vdev_ops == &vdev_spare_ops && 5032 vd->vdev_id == 0 && 5033 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5034 unspare = B_TRUE; 5035 5036 /* 5037 * Erase the disk labels so the disk can be used for other things. 5038 * This must be done after all other error cases are handled, 5039 * but before we disembowel vd (so we can still do I/O to it). 5040 * But if we can't do it, don't treat the error as fatal -- 5041 * it may be that the unwritability of the disk is the reason 5042 * it's being detached! 5043 */ 5044 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5045 5046 /* 5047 * Remove vd from its parent and compact the parent's children. 5048 */ 5049 vdev_remove_child(pvd, vd); 5050 vdev_compact_children(pvd); 5051 5052 /* 5053 * Remember one of the remaining children so we can get tvd below. 5054 */ 5055 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5056 5057 /* 5058 * If we need to remove the remaining child from the list of hot spares, 5059 * do it now, marking the vdev as no longer a spare in the process. 5060 * We must do this before vdev_remove_parent(), because that can 5061 * change the GUID if it creates a new toplevel GUID. For a similar 5062 * reason, we must remove the spare now, in the same txg as the detach; 5063 * otherwise someone could attach a new sibling, change the GUID, and 5064 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5065 */ 5066 if (unspare) { 5067 ASSERT(cvd->vdev_isspare); 5068 spa_spare_remove(cvd); 5069 unspare_guid = cvd->vdev_guid; 5070 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5071 cvd->vdev_unspare = B_TRUE; 5072 } 5073 5074 /* 5075 * If the parent mirror/replacing vdev only has one child, 5076 * the parent is no longer needed. Remove it from the tree. 5077 */ 5078 if (pvd->vdev_children == 1) { 5079 if (pvd->vdev_ops == &vdev_spare_ops) 5080 cvd->vdev_unspare = B_FALSE; 5081 vdev_remove_parent(cvd); 5082 } 5083 5084 5085 /* 5086 * We don't set tvd until now because the parent we just removed 5087 * may have been the previous top-level vdev. 5088 */ 5089 tvd = cvd->vdev_top; 5090 ASSERT(tvd->vdev_parent == rvd); 5091 5092 /* 5093 * Reevaluate the parent vdev state. 5094 */ 5095 vdev_propagate_state(cvd); 5096 5097 /* 5098 * If the 'autoexpand' property is set on the pool then automatically 5099 * try to expand the size of the pool. For example if the device we 5100 * just detached was smaller than the others, it may be possible to 5101 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5102 * first so that we can obtain the updated sizes of the leaf vdevs. 5103 */ 5104 if (spa->spa_autoexpand) { 5105 vdev_reopen(tvd); 5106 vdev_expand(tvd, txg); 5107 } 5108 5109 vdev_config_dirty(tvd); 5110 5111 /* 5112 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5113 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5114 * But first make sure we're not on any *other* txg's DTL list, to 5115 * prevent vd from being accessed after it's freed. 5116 */ 5117 vdpath = spa_strdup(vd->vdev_path); 5118 for (int t = 0; t < TXG_SIZE; t++) 5119 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5120 vd->vdev_detached = B_TRUE; 5121 vdev_dirty(tvd, VDD_DTL, vd, txg); 5122 5123 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5124 5125 /* hang on to the spa before we release the lock */ 5126 spa_open_ref(spa, FTAG); 5127 5128 error = spa_vdev_exit(spa, vd, txg, 0); 5129 5130 spa_history_log_internal(spa, "detach", NULL, 5131 "vdev=%s", vdpath); 5132 spa_strfree(vdpath); 5133 5134 /* 5135 * If this was the removal of the original device in a hot spare vdev, 5136 * then we want to go through and remove the device from the hot spare 5137 * list of every other pool. 5138 */ 5139 if (unspare) { 5140 spa_t *altspa = NULL; 5141 5142 mutex_enter(&spa_namespace_lock); 5143 while ((altspa = spa_next(altspa)) != NULL) { 5144 if (altspa->spa_state != POOL_STATE_ACTIVE || 5145 altspa == spa) 5146 continue; 5147 5148 spa_open_ref(altspa, FTAG); 5149 mutex_exit(&spa_namespace_lock); 5150 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5151 mutex_enter(&spa_namespace_lock); 5152 spa_close(altspa, FTAG); 5153 } 5154 mutex_exit(&spa_namespace_lock); 5155 5156 /* search the rest of the vdevs for spares to remove */ 5157 spa_vdev_resilver_done(spa); 5158 } 5159 5160 /* all done with the spa; OK to release */ 5161 mutex_enter(&spa_namespace_lock); 5162 spa_close(spa, FTAG); 5163 mutex_exit(&spa_namespace_lock); 5164 5165 return (error); 5166} 5167 5168/* 5169 * Split a set of devices from their mirrors, and create a new pool from them. 5170 */ 5171int 5172spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5173 nvlist_t *props, boolean_t exp) 5174{ 5175 int error = 0; 5176 uint64_t txg, *glist; 5177 spa_t *newspa; 5178 uint_t c, children, lastlog; 5179 nvlist_t **child, *nvl, *tmp; 5180 dmu_tx_t *tx; 5181 char *altroot = NULL; 5182 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5183 boolean_t activate_slog; 5184 5185 ASSERT(spa_writeable(spa)); 5186 5187 txg = spa_vdev_enter(spa); 5188 5189 /* clear the log and flush everything up to now */ 5190 activate_slog = spa_passivate_log(spa); 5191 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5192 error = spa_offline_log(spa); 5193 txg = spa_vdev_config_enter(spa); 5194 5195 if (activate_slog) 5196 spa_activate_log(spa); 5197 5198 if (error != 0) 5199 return (spa_vdev_exit(spa, NULL, txg, error)); 5200 5201 /* check new spa name before going any further */ 5202 if (spa_lookup(newname) != NULL) 5203 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5204 5205 /* 5206 * scan through all the children to ensure they're all mirrors 5207 */ 5208 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5209 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5210 &children) != 0) 5211 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5212 5213 /* first, check to ensure we've got the right child count */ 5214 rvd = spa->spa_root_vdev; 5215 lastlog = 0; 5216 for (c = 0; c < rvd->vdev_children; c++) { 5217 vdev_t *vd = rvd->vdev_child[c]; 5218 5219 /* don't count the holes & logs as children */ 5220 if (vd->vdev_islog || vd->vdev_ishole) { 5221 if (lastlog == 0) 5222 lastlog = c; 5223 continue; 5224 } 5225 5226 lastlog = 0; 5227 } 5228 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5229 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5230 5231 /* next, ensure no spare or cache devices are part of the split */ 5232 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5233 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5234 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5235 5236 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5237 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5238 5239 /* then, loop over each vdev and validate it */ 5240 for (c = 0; c < children; c++) { 5241 uint64_t is_hole = 0; 5242 5243 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5244 &is_hole); 5245 5246 if (is_hole != 0) { 5247 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5248 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5249 continue; 5250 } else { 5251 error = SET_ERROR(EINVAL); 5252 break; 5253 } 5254 } 5255 5256 /* which disk is going to be split? */ 5257 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5258 &glist[c]) != 0) { 5259 error = SET_ERROR(EINVAL); 5260 break; 5261 } 5262 5263 /* look it up in the spa */ 5264 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5265 if (vml[c] == NULL) { 5266 error = SET_ERROR(ENODEV); 5267 break; 5268 } 5269 5270 /* make sure there's nothing stopping the split */ 5271 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5272 vml[c]->vdev_islog || 5273 vml[c]->vdev_ishole || 5274 vml[c]->vdev_isspare || 5275 vml[c]->vdev_isl2cache || 5276 !vdev_writeable(vml[c]) || 5277 vml[c]->vdev_children != 0 || 5278 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5279 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5280 error = SET_ERROR(EINVAL); 5281 break; 5282 } 5283 5284 if (vdev_dtl_required(vml[c])) { 5285 error = SET_ERROR(EBUSY); 5286 break; 5287 } 5288 5289 /* we need certain info from the top level */ 5290 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5291 vml[c]->vdev_top->vdev_ms_array) == 0); 5292 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5293 vml[c]->vdev_top->vdev_ms_shift) == 0); 5294 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5295 vml[c]->vdev_top->vdev_asize) == 0); 5296 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5297 vml[c]->vdev_top->vdev_ashift) == 0); 5298 } 5299 5300 if (error != 0) { 5301 kmem_free(vml, children * sizeof (vdev_t *)); 5302 kmem_free(glist, children * sizeof (uint64_t)); 5303 return (spa_vdev_exit(spa, NULL, txg, error)); 5304 } 5305 5306 /* stop writers from using the disks */ 5307 for (c = 0; c < children; c++) { 5308 if (vml[c] != NULL) 5309 vml[c]->vdev_offline = B_TRUE; 5310 } 5311 vdev_reopen(spa->spa_root_vdev); 5312 5313 /* 5314 * Temporarily record the splitting vdevs in the spa config. This 5315 * will disappear once the config is regenerated. 5316 */ 5317 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5318 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5319 glist, children) == 0); 5320 kmem_free(glist, children * sizeof (uint64_t)); 5321 5322 mutex_enter(&spa->spa_props_lock); 5323 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5324 nvl) == 0); 5325 mutex_exit(&spa->spa_props_lock); 5326 spa->spa_config_splitting = nvl; 5327 vdev_config_dirty(spa->spa_root_vdev); 5328 5329 /* configure and create the new pool */ 5330 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5331 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5332 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5333 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5334 spa_version(spa)) == 0); 5335 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5336 spa->spa_config_txg) == 0); 5337 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5338 spa_generate_guid(NULL)) == 0); 5339 (void) nvlist_lookup_string(props, 5340 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5341 5342 /* add the new pool to the namespace */ 5343 newspa = spa_add(newname, config, altroot); 5344 newspa->spa_config_txg = spa->spa_config_txg; 5345 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5346 5347 /* release the spa config lock, retaining the namespace lock */ 5348 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5349 5350 if (zio_injection_enabled) 5351 zio_handle_panic_injection(spa, FTAG, 1); 5352 5353 spa_activate(newspa, spa_mode_global); 5354 spa_async_suspend(newspa); 5355 5356#ifndef sun 5357 /* mark that we are creating new spa by splitting */ 5358 newspa->spa_splitting_newspa = B_TRUE; 5359#endif 5360 /* create the new pool from the disks of the original pool */ 5361 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5362#ifndef sun 5363 newspa->spa_splitting_newspa = B_FALSE; 5364#endif 5365 if (error) 5366 goto out; 5367 5368 /* if that worked, generate a real config for the new pool */ 5369 if (newspa->spa_root_vdev != NULL) { 5370 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5371 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5372 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5373 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5374 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5375 B_TRUE)); 5376 } 5377 5378 /* set the props */ 5379 if (props != NULL) { 5380 spa_configfile_set(newspa, props, B_FALSE); 5381 error = spa_prop_set(newspa, props); 5382 if (error) 5383 goto out; 5384 } 5385 5386 /* flush everything */ 5387 txg = spa_vdev_config_enter(newspa); 5388 vdev_config_dirty(newspa->spa_root_vdev); 5389 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5390 5391 if (zio_injection_enabled) 5392 zio_handle_panic_injection(spa, FTAG, 2); 5393 5394 spa_async_resume(newspa); 5395 5396 /* finally, update the original pool's config */ 5397 txg = spa_vdev_config_enter(spa); 5398 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5399 error = dmu_tx_assign(tx, TXG_WAIT); 5400 if (error != 0) 5401 dmu_tx_abort(tx); 5402 for (c = 0; c < children; c++) { 5403 if (vml[c] != NULL) { 5404 vdev_split(vml[c]); 5405 if (error == 0) 5406 spa_history_log_internal(spa, "detach", tx, 5407 "vdev=%s", vml[c]->vdev_path); 5408 vdev_free(vml[c]); 5409 } 5410 } 5411 vdev_config_dirty(spa->spa_root_vdev); 5412 spa->spa_config_splitting = NULL; 5413 nvlist_free(nvl); 5414 if (error == 0) 5415 dmu_tx_commit(tx); 5416 (void) spa_vdev_exit(spa, NULL, txg, 0); 5417 5418 if (zio_injection_enabled) 5419 zio_handle_panic_injection(spa, FTAG, 3); 5420 5421 /* split is complete; log a history record */ 5422 spa_history_log_internal(newspa, "split", NULL, 5423 "from pool %s", spa_name(spa)); 5424 5425 kmem_free(vml, children * sizeof (vdev_t *)); 5426 5427 /* if we're not going to mount the filesystems in userland, export */ 5428 if (exp) 5429 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5430 B_FALSE, B_FALSE); 5431 5432 return (error); 5433 5434out: 5435 spa_unload(newspa); 5436 spa_deactivate(newspa); 5437 spa_remove(newspa); 5438 5439 txg = spa_vdev_config_enter(spa); 5440 5441 /* re-online all offlined disks */ 5442 for (c = 0; c < children; c++) { 5443 if (vml[c] != NULL) 5444 vml[c]->vdev_offline = B_FALSE; 5445 } 5446 vdev_reopen(spa->spa_root_vdev); 5447 5448 nvlist_free(spa->spa_config_splitting); 5449 spa->spa_config_splitting = NULL; 5450 (void) spa_vdev_exit(spa, NULL, txg, error); 5451 5452 kmem_free(vml, children * sizeof (vdev_t *)); 5453 return (error); 5454} 5455 5456static nvlist_t * 5457spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5458{ 5459 for (int i = 0; i < count; i++) { 5460 uint64_t guid; 5461 5462 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5463 &guid) == 0); 5464 5465 if (guid == target_guid) 5466 return (nvpp[i]); 5467 } 5468 5469 return (NULL); 5470} 5471 5472static void 5473spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5474 nvlist_t *dev_to_remove) 5475{ 5476 nvlist_t **newdev = NULL; 5477 5478 if (count > 1) 5479 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5480 5481 for (int i = 0, j = 0; i < count; i++) { 5482 if (dev[i] == dev_to_remove) 5483 continue; 5484 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5485 } 5486 5487 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5488 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5489 5490 for (int i = 0; i < count - 1; i++) 5491 nvlist_free(newdev[i]); 5492 5493 if (count > 1) 5494 kmem_free(newdev, (count - 1) * sizeof (void *)); 5495} 5496 5497/* 5498 * Evacuate the device. 5499 */ 5500static int 5501spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5502{ 5503 uint64_t txg; 5504 int error = 0; 5505 5506 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5507 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5508 ASSERT(vd == vd->vdev_top); 5509 5510 /* 5511 * Evacuate the device. We don't hold the config lock as writer 5512 * since we need to do I/O but we do keep the 5513 * spa_namespace_lock held. Once this completes the device 5514 * should no longer have any blocks allocated on it. 5515 */ 5516 if (vd->vdev_islog) { 5517 if (vd->vdev_stat.vs_alloc != 0) 5518 error = spa_offline_log(spa); 5519 } else { 5520 error = SET_ERROR(ENOTSUP); 5521 } 5522 5523 if (error) 5524 return (error); 5525 5526 /* 5527 * The evacuation succeeded. Remove any remaining MOS metadata 5528 * associated with this vdev, and wait for these changes to sync. 5529 */ 5530 ASSERT0(vd->vdev_stat.vs_alloc); 5531 txg = spa_vdev_config_enter(spa); 5532 vd->vdev_removing = B_TRUE; 5533 vdev_dirty_leaves(vd, VDD_DTL, txg); 5534 vdev_config_dirty(vd); 5535 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5536 5537 return (0); 5538} 5539 5540/* 5541 * Complete the removal by cleaning up the namespace. 5542 */ 5543static void 5544spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5545{ 5546 vdev_t *rvd = spa->spa_root_vdev; 5547 uint64_t id = vd->vdev_id; 5548 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5549 5550 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5551 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5552 ASSERT(vd == vd->vdev_top); 5553 5554 /* 5555 * Only remove any devices which are empty. 5556 */ 5557 if (vd->vdev_stat.vs_alloc != 0) 5558 return; 5559 5560 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5561 5562 if (list_link_active(&vd->vdev_state_dirty_node)) 5563 vdev_state_clean(vd); 5564 if (list_link_active(&vd->vdev_config_dirty_node)) 5565 vdev_config_clean(vd); 5566 5567 vdev_free(vd); 5568 5569 if (last_vdev) { 5570 vdev_compact_children(rvd); 5571 } else { 5572 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5573 vdev_add_child(rvd, vd); 5574 } 5575 vdev_config_dirty(rvd); 5576 5577 /* 5578 * Reassess the health of our root vdev. 5579 */ 5580 vdev_reopen(rvd); 5581} 5582 5583/* 5584 * Remove a device from the pool - 5585 * 5586 * Removing a device from the vdev namespace requires several steps 5587 * and can take a significant amount of time. As a result we use 5588 * the spa_vdev_config_[enter/exit] functions which allow us to 5589 * grab and release the spa_config_lock while still holding the namespace 5590 * lock. During each step the configuration is synced out. 5591 * 5592 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5593 * devices. 5594 */ 5595int 5596spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5597{ 5598 vdev_t *vd; 5599 metaslab_group_t *mg; 5600 nvlist_t **spares, **l2cache, *nv; 5601 uint64_t txg = 0; 5602 uint_t nspares, nl2cache; 5603 int error = 0; 5604 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5605 5606 ASSERT(spa_writeable(spa)); 5607 5608 if (!locked) 5609 txg = spa_vdev_enter(spa); 5610 5611 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5612 5613 if (spa->spa_spares.sav_vdevs != NULL && 5614 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5615 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5616 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5617 /* 5618 * Only remove the hot spare if it's not currently in use 5619 * in this pool. 5620 */ 5621 if (vd == NULL || unspare) { 5622 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5623 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5624 spa_load_spares(spa); 5625 spa->spa_spares.sav_sync = B_TRUE; 5626 } else { 5627 error = SET_ERROR(EBUSY); 5628 } 5629 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5630 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5631 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5632 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5633 /* 5634 * Cache devices can always be removed. 5635 */ 5636 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5637 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5638 spa_load_l2cache(spa); 5639 spa->spa_l2cache.sav_sync = B_TRUE; 5640 } else if (vd != NULL && vd->vdev_islog) { 5641 ASSERT(!locked); 5642 ASSERT(vd == vd->vdev_top); 5643 5644 mg = vd->vdev_mg; 5645 5646 /* 5647 * Stop allocating from this vdev. 5648 */ 5649 metaslab_group_passivate(mg); 5650 5651 /* 5652 * Wait for the youngest allocations and frees to sync, 5653 * and then wait for the deferral of those frees to finish. 5654 */ 5655 spa_vdev_config_exit(spa, NULL, 5656 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5657 5658 /* 5659 * Attempt to evacuate the vdev. 5660 */ 5661 error = spa_vdev_remove_evacuate(spa, vd); 5662 5663 txg = spa_vdev_config_enter(spa); 5664 5665 /* 5666 * If we couldn't evacuate the vdev, unwind. 5667 */ 5668 if (error) { 5669 metaslab_group_activate(mg); 5670 return (spa_vdev_exit(spa, NULL, txg, error)); 5671 } 5672 5673 /* 5674 * Clean up the vdev namespace. 5675 */ 5676 spa_vdev_remove_from_namespace(spa, vd); 5677 5678 } else if (vd != NULL) { 5679 /* 5680 * Normal vdevs cannot be removed (yet). 5681 */ 5682 error = SET_ERROR(ENOTSUP); 5683 } else { 5684 /* 5685 * There is no vdev of any kind with the specified guid. 5686 */ 5687 error = SET_ERROR(ENOENT); 5688 } 5689 5690 if (!locked) 5691 return (spa_vdev_exit(spa, NULL, txg, error)); 5692 5693 return (error); 5694} 5695 5696/* 5697 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5698 * currently spared, so we can detach it. 5699 */ 5700static vdev_t * 5701spa_vdev_resilver_done_hunt(vdev_t *vd) 5702{ 5703 vdev_t *newvd, *oldvd; 5704 5705 for (int c = 0; c < vd->vdev_children; c++) { 5706 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5707 if (oldvd != NULL) 5708 return (oldvd); 5709 } 5710 5711 /* 5712 * Check for a completed replacement. We always consider the first 5713 * vdev in the list to be the oldest vdev, and the last one to be 5714 * the newest (see spa_vdev_attach() for how that works). In 5715 * the case where the newest vdev is faulted, we will not automatically 5716 * remove it after a resilver completes. This is OK as it will require 5717 * user intervention to determine which disk the admin wishes to keep. 5718 */ 5719 if (vd->vdev_ops == &vdev_replacing_ops) { 5720 ASSERT(vd->vdev_children > 1); 5721 5722 newvd = vd->vdev_child[vd->vdev_children - 1]; 5723 oldvd = vd->vdev_child[0]; 5724 5725 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5726 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5727 !vdev_dtl_required(oldvd)) 5728 return (oldvd); 5729 } 5730 5731 /* 5732 * Check for a completed resilver with the 'unspare' flag set. 5733 */ 5734 if (vd->vdev_ops == &vdev_spare_ops) { 5735 vdev_t *first = vd->vdev_child[0]; 5736 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5737 5738 if (last->vdev_unspare) { 5739 oldvd = first; 5740 newvd = last; 5741 } else if (first->vdev_unspare) { 5742 oldvd = last; 5743 newvd = first; 5744 } else { 5745 oldvd = NULL; 5746 } 5747 5748 if (oldvd != NULL && 5749 vdev_dtl_empty(newvd, DTL_MISSING) && 5750 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5751 !vdev_dtl_required(oldvd)) 5752 return (oldvd); 5753 5754 /* 5755 * If there are more than two spares attached to a disk, 5756 * and those spares are not required, then we want to 5757 * attempt to free them up now so that they can be used 5758 * by other pools. Once we're back down to a single 5759 * disk+spare, we stop removing them. 5760 */ 5761 if (vd->vdev_children > 2) { 5762 newvd = vd->vdev_child[1]; 5763 5764 if (newvd->vdev_isspare && last->vdev_isspare && 5765 vdev_dtl_empty(last, DTL_MISSING) && 5766 vdev_dtl_empty(last, DTL_OUTAGE) && 5767 !vdev_dtl_required(newvd)) 5768 return (newvd); 5769 } 5770 } 5771 5772 return (NULL); 5773} 5774 5775static void 5776spa_vdev_resilver_done(spa_t *spa) 5777{ 5778 vdev_t *vd, *pvd, *ppvd; 5779 uint64_t guid, sguid, pguid, ppguid; 5780 5781 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5782 5783 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5784 pvd = vd->vdev_parent; 5785 ppvd = pvd->vdev_parent; 5786 guid = vd->vdev_guid; 5787 pguid = pvd->vdev_guid; 5788 ppguid = ppvd->vdev_guid; 5789 sguid = 0; 5790 /* 5791 * If we have just finished replacing a hot spared device, then 5792 * we need to detach the parent's first child (the original hot 5793 * spare) as well. 5794 */ 5795 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5796 ppvd->vdev_children == 2) { 5797 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5798 sguid = ppvd->vdev_child[1]->vdev_guid; 5799 } 5800 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5801 5802 spa_config_exit(spa, SCL_ALL, FTAG); 5803 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5804 return; 5805 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5806 return; 5807 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5808 } 5809 5810 spa_config_exit(spa, SCL_ALL, FTAG); 5811} 5812 5813/* 5814 * Update the stored path or FRU for this vdev. 5815 */ 5816int 5817spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5818 boolean_t ispath) 5819{ 5820 vdev_t *vd; 5821 boolean_t sync = B_FALSE; 5822 5823 ASSERT(spa_writeable(spa)); 5824 5825 spa_vdev_state_enter(spa, SCL_ALL); 5826 5827 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5828 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5829 5830 if (!vd->vdev_ops->vdev_op_leaf) 5831 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5832 5833 if (ispath) { 5834 if (strcmp(value, vd->vdev_path) != 0) { 5835 spa_strfree(vd->vdev_path); 5836 vd->vdev_path = spa_strdup(value); 5837 sync = B_TRUE; 5838 } 5839 } else { 5840 if (vd->vdev_fru == NULL) { 5841 vd->vdev_fru = spa_strdup(value); 5842 sync = B_TRUE; 5843 } else if (strcmp(value, vd->vdev_fru) != 0) { 5844 spa_strfree(vd->vdev_fru); 5845 vd->vdev_fru = spa_strdup(value); 5846 sync = B_TRUE; 5847 } 5848 } 5849 5850 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5851} 5852 5853int 5854spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5855{ 5856 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5857} 5858 5859int 5860spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5861{ 5862 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5863} 5864 5865/* 5866 * ========================================================================== 5867 * SPA Scanning 5868 * ========================================================================== 5869 */ 5870 5871int 5872spa_scan_stop(spa_t *spa) 5873{ 5874 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5875 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5876 return (SET_ERROR(EBUSY)); 5877 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5878} 5879 5880int 5881spa_scan(spa_t *spa, pool_scan_func_t func) 5882{ 5883 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5884 5885 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5886 return (SET_ERROR(ENOTSUP)); 5887 5888 /* 5889 * If a resilver was requested, but there is no DTL on a 5890 * writeable leaf device, we have nothing to do. 5891 */ 5892 if (func == POOL_SCAN_RESILVER && 5893 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5894 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5895 return (0); 5896 } 5897 5898 return (dsl_scan(spa->spa_dsl_pool, func)); 5899} 5900 5901/* 5902 * ========================================================================== 5903 * SPA async task processing 5904 * ========================================================================== 5905 */ 5906 5907static void 5908spa_async_remove(spa_t *spa, vdev_t *vd) 5909{ 5910 if (vd->vdev_remove_wanted) { 5911 vd->vdev_remove_wanted = B_FALSE; 5912 vd->vdev_delayed_close = B_FALSE; 5913 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5914 5915 /* 5916 * We want to clear the stats, but we don't want to do a full 5917 * vdev_clear() as that will cause us to throw away 5918 * degraded/faulted state as well as attempt to reopen the 5919 * device, all of which is a waste. 5920 */ 5921 vd->vdev_stat.vs_read_errors = 0; 5922 vd->vdev_stat.vs_write_errors = 0; 5923 vd->vdev_stat.vs_checksum_errors = 0; 5924 5925 vdev_state_dirty(vd->vdev_top); 5926 } 5927 5928 for (int c = 0; c < vd->vdev_children; c++) 5929 spa_async_remove(spa, vd->vdev_child[c]); 5930} 5931 5932static void 5933spa_async_probe(spa_t *spa, vdev_t *vd) 5934{ 5935 if (vd->vdev_probe_wanted) { 5936 vd->vdev_probe_wanted = B_FALSE; 5937 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5938 } 5939 5940 for (int c = 0; c < vd->vdev_children; c++) 5941 spa_async_probe(spa, vd->vdev_child[c]); 5942} 5943 5944static void 5945spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5946{ 5947 sysevent_id_t eid; 5948 nvlist_t *attr; 5949 char *physpath; 5950 5951 if (!spa->spa_autoexpand) 5952 return; 5953 5954 for (int c = 0; c < vd->vdev_children; c++) { 5955 vdev_t *cvd = vd->vdev_child[c]; 5956 spa_async_autoexpand(spa, cvd); 5957 } 5958 5959 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5960 return; 5961 5962 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5963 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5964 5965 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5966 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5967 5968 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5969 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5970 5971 nvlist_free(attr); 5972 kmem_free(physpath, MAXPATHLEN); 5973} 5974 5975static void 5976spa_async_thread(void *arg) 5977{ 5978 spa_t *spa = arg; 5979 int tasks; 5980 5981 ASSERT(spa->spa_sync_on); 5982 5983 mutex_enter(&spa->spa_async_lock); 5984 tasks = spa->spa_async_tasks; 5985 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5986 mutex_exit(&spa->spa_async_lock); 5987 5988 /* 5989 * See if the config needs to be updated. 5990 */ 5991 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5992 uint64_t old_space, new_space; 5993 5994 mutex_enter(&spa_namespace_lock); 5995 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5996 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5997 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5998 mutex_exit(&spa_namespace_lock); 5999 6000 /* 6001 * If the pool grew as a result of the config update, 6002 * then log an internal history event. 6003 */ 6004 if (new_space != old_space) { 6005 spa_history_log_internal(spa, "vdev online", NULL, 6006 "pool '%s' size: %llu(+%llu)", 6007 spa_name(spa), new_space, new_space - old_space); 6008 } 6009 } 6010 6011 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6012 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6013 spa_async_autoexpand(spa, spa->spa_root_vdev); 6014 spa_config_exit(spa, SCL_CONFIG, FTAG); 6015 } 6016 6017 /* 6018 * See if any devices need to be probed. 6019 */ 6020 if (tasks & SPA_ASYNC_PROBE) { 6021 spa_vdev_state_enter(spa, SCL_NONE); 6022 spa_async_probe(spa, spa->spa_root_vdev); 6023 (void) spa_vdev_state_exit(spa, NULL, 0); 6024 } 6025 6026 /* 6027 * If any devices are done replacing, detach them. 6028 */ 6029 if (tasks & SPA_ASYNC_RESILVER_DONE) 6030 spa_vdev_resilver_done(spa); 6031 6032 /* 6033 * Kick off a resilver. 6034 */ 6035 if (tasks & SPA_ASYNC_RESILVER) 6036 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6037 6038 /* 6039 * Let the world know that we're done. 6040 */ 6041 mutex_enter(&spa->spa_async_lock); 6042 spa->spa_async_thread = NULL; 6043 cv_broadcast(&spa->spa_async_cv); 6044 mutex_exit(&spa->spa_async_lock); 6045 thread_exit(); 6046} 6047 6048static void 6049spa_async_thread_vd(void *arg) 6050{ 6051 spa_t *spa = arg; 6052 int tasks; 6053 6054 ASSERT(spa->spa_sync_on); 6055 6056 mutex_enter(&spa->spa_async_lock); 6057 tasks = spa->spa_async_tasks; 6058retry: 6059 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6060 mutex_exit(&spa->spa_async_lock); 6061 6062 /* 6063 * See if any devices need to be marked REMOVED. 6064 */ 6065 if (tasks & SPA_ASYNC_REMOVE) { 6066 spa_vdev_state_enter(spa, SCL_NONE); 6067 spa_async_remove(spa, spa->spa_root_vdev); 6068 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6069 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6070 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6071 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6072 (void) spa_vdev_state_exit(spa, NULL, 0); 6073 } 6074 6075 /* 6076 * Let the world know that we're done. 6077 */ 6078 mutex_enter(&spa->spa_async_lock); 6079 tasks = spa->spa_async_tasks; 6080 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6081 goto retry; 6082 spa->spa_async_thread_vd = NULL; 6083 cv_broadcast(&spa->spa_async_cv); 6084 mutex_exit(&spa->spa_async_lock); 6085 thread_exit(); 6086} 6087 6088void 6089spa_async_suspend(spa_t *spa) 6090{ 6091 mutex_enter(&spa->spa_async_lock); 6092 spa->spa_async_suspended++; 6093 while (spa->spa_async_thread != NULL && 6094 spa->spa_async_thread_vd != NULL) 6095 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6096 mutex_exit(&spa->spa_async_lock); 6097} 6098 6099void 6100spa_async_resume(spa_t *spa) 6101{ 6102 mutex_enter(&spa->spa_async_lock); 6103 ASSERT(spa->spa_async_suspended != 0); 6104 spa->spa_async_suspended--; 6105 mutex_exit(&spa->spa_async_lock); 6106} 6107 6108static boolean_t 6109spa_async_tasks_pending(spa_t *spa) 6110{ 6111 uint_t non_config_tasks; 6112 uint_t config_task; 6113 boolean_t config_task_suspended; 6114 6115 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6116 SPA_ASYNC_REMOVE); 6117 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6118 if (spa->spa_ccw_fail_time == 0) { 6119 config_task_suspended = B_FALSE; 6120 } else { 6121 config_task_suspended = 6122 (gethrtime() - spa->spa_ccw_fail_time) < 6123 (zfs_ccw_retry_interval * NANOSEC); 6124 } 6125 6126 return (non_config_tasks || (config_task && !config_task_suspended)); 6127} 6128 6129static void 6130spa_async_dispatch(spa_t *spa) 6131{ 6132 mutex_enter(&spa->spa_async_lock); 6133 if (spa_async_tasks_pending(spa) && 6134 !spa->spa_async_suspended && 6135 spa->spa_async_thread == NULL && 6136 rootdir != NULL) 6137 spa->spa_async_thread = thread_create(NULL, 0, 6138 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6139 mutex_exit(&spa->spa_async_lock); 6140} 6141 6142static void 6143spa_async_dispatch_vd(spa_t *spa) 6144{ 6145 mutex_enter(&spa->spa_async_lock); 6146 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6147 !spa->spa_async_suspended && 6148 spa->spa_async_thread_vd == NULL && 6149 rootdir != NULL) 6150 spa->spa_async_thread_vd = thread_create(NULL, 0, 6151 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6152 mutex_exit(&spa->spa_async_lock); 6153} 6154 6155void 6156spa_async_request(spa_t *spa, int task) 6157{ 6158 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6159 mutex_enter(&spa->spa_async_lock); 6160 spa->spa_async_tasks |= task; 6161 mutex_exit(&spa->spa_async_lock); 6162 spa_async_dispatch_vd(spa); 6163} 6164 6165/* 6166 * ========================================================================== 6167 * SPA syncing routines 6168 * ========================================================================== 6169 */ 6170 6171static int 6172bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6173{ 6174 bpobj_t *bpo = arg; 6175 bpobj_enqueue(bpo, bp, tx); 6176 return (0); 6177} 6178 6179static int 6180spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6181{ 6182 zio_t *zio = arg; 6183 6184 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6185 BP_GET_PSIZE(bp), zio->io_flags)); 6186 return (0); 6187} 6188 6189/* 6190 * Note: this simple function is not inlined to make it easier to dtrace the 6191 * amount of time spent syncing frees. 6192 */ 6193static void 6194spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6195{ 6196 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6197 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6198 VERIFY(zio_wait(zio) == 0); 6199} 6200 6201/* 6202 * Note: this simple function is not inlined to make it easier to dtrace the 6203 * amount of time spent syncing deferred frees. 6204 */ 6205static void 6206spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6207{ 6208 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6209 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6210 spa_free_sync_cb, zio, tx), ==, 0); 6211 VERIFY0(zio_wait(zio)); 6212} 6213 6214 6215static void 6216spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6217{ 6218 char *packed = NULL; 6219 size_t bufsize; 6220 size_t nvsize = 0; 6221 dmu_buf_t *db; 6222 6223 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6224 6225 /* 6226 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6227 * information. This avoids the dmu_buf_will_dirty() path and 6228 * saves us a pre-read to get data we don't actually care about. 6229 */ 6230 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6231 packed = kmem_alloc(bufsize, KM_SLEEP); 6232 6233 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6234 KM_SLEEP) == 0); 6235 bzero(packed + nvsize, bufsize - nvsize); 6236 6237 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6238 6239 kmem_free(packed, bufsize); 6240 6241 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6242 dmu_buf_will_dirty(db, tx); 6243 *(uint64_t *)db->db_data = nvsize; 6244 dmu_buf_rele(db, FTAG); 6245} 6246 6247static void 6248spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6249 const char *config, const char *entry) 6250{ 6251 nvlist_t *nvroot; 6252 nvlist_t **list; 6253 int i; 6254 6255 if (!sav->sav_sync) 6256 return; 6257 6258 /* 6259 * Update the MOS nvlist describing the list of available devices. 6260 * spa_validate_aux() will have already made sure this nvlist is 6261 * valid and the vdevs are labeled appropriately. 6262 */ 6263 if (sav->sav_object == 0) { 6264 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6265 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6266 sizeof (uint64_t), tx); 6267 VERIFY(zap_update(spa->spa_meta_objset, 6268 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6269 &sav->sav_object, tx) == 0); 6270 } 6271 6272 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6273 if (sav->sav_count == 0) { 6274 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6275 } else { 6276 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6277 for (i = 0; i < sav->sav_count; i++) 6278 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6279 B_FALSE, VDEV_CONFIG_L2CACHE); 6280 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6281 sav->sav_count) == 0); 6282 for (i = 0; i < sav->sav_count; i++) 6283 nvlist_free(list[i]); 6284 kmem_free(list, sav->sav_count * sizeof (void *)); 6285 } 6286 6287 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6288 nvlist_free(nvroot); 6289 6290 sav->sav_sync = B_FALSE; 6291} 6292 6293static void 6294spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6295{ 6296 nvlist_t *config; 6297 6298 if (list_is_empty(&spa->spa_config_dirty_list)) 6299 return; 6300 6301 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6302 6303 config = spa_config_generate(spa, spa->spa_root_vdev, 6304 dmu_tx_get_txg(tx), B_FALSE); 6305 6306 /* 6307 * If we're upgrading the spa version then make sure that 6308 * the config object gets updated with the correct version. 6309 */ 6310 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6311 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6312 spa->spa_uberblock.ub_version); 6313 6314 spa_config_exit(spa, SCL_STATE, FTAG); 6315 6316 if (spa->spa_config_syncing) 6317 nvlist_free(spa->spa_config_syncing); 6318 spa->spa_config_syncing = config; 6319 6320 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6321} 6322 6323static void 6324spa_sync_version(void *arg, dmu_tx_t *tx) 6325{ 6326 uint64_t *versionp = arg; 6327 uint64_t version = *versionp; 6328 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6329 6330 /* 6331 * Setting the version is special cased when first creating the pool. 6332 */ 6333 ASSERT(tx->tx_txg != TXG_INITIAL); 6334 6335 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6336 ASSERT(version >= spa_version(spa)); 6337 6338 spa->spa_uberblock.ub_version = version; 6339 vdev_config_dirty(spa->spa_root_vdev); 6340 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6341} 6342 6343/* 6344 * Set zpool properties. 6345 */ 6346static void 6347spa_sync_props(void *arg, dmu_tx_t *tx) 6348{ 6349 nvlist_t *nvp = arg; 6350 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6351 objset_t *mos = spa->spa_meta_objset; 6352 nvpair_t *elem = NULL; 6353 6354 mutex_enter(&spa->spa_props_lock); 6355 6356 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6357 uint64_t intval; 6358 char *strval, *fname; 6359 zpool_prop_t prop; 6360 const char *propname; 6361 zprop_type_t proptype; 6362 spa_feature_t fid; 6363 6364 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6365 case ZPROP_INVAL: 6366 /* 6367 * We checked this earlier in spa_prop_validate(). 6368 */ 6369 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6370 6371 fname = strchr(nvpair_name(elem), '@') + 1; 6372 VERIFY0(zfeature_lookup_name(fname, &fid)); 6373 6374 spa_feature_enable(spa, fid, tx); 6375 spa_history_log_internal(spa, "set", tx, 6376 "%s=enabled", nvpair_name(elem)); 6377 break; 6378 6379 case ZPOOL_PROP_VERSION: 6380 intval = fnvpair_value_uint64(elem); 6381 /* 6382 * The version is synced seperatly before other 6383 * properties and should be correct by now. 6384 */ 6385 ASSERT3U(spa_version(spa), >=, intval); 6386 break; 6387 6388 case ZPOOL_PROP_ALTROOT: 6389 /* 6390 * 'altroot' is a non-persistent property. It should 6391 * have been set temporarily at creation or import time. 6392 */ 6393 ASSERT(spa->spa_root != NULL); 6394 break; 6395 6396 case ZPOOL_PROP_READONLY: 6397 case ZPOOL_PROP_CACHEFILE: 6398 /* 6399 * 'readonly' and 'cachefile' are also non-persisitent 6400 * properties. 6401 */ 6402 break; 6403 case ZPOOL_PROP_COMMENT: 6404 strval = fnvpair_value_string(elem); 6405 if (spa->spa_comment != NULL) 6406 spa_strfree(spa->spa_comment); 6407 spa->spa_comment = spa_strdup(strval); 6408 /* 6409 * We need to dirty the configuration on all the vdevs 6410 * so that their labels get updated. It's unnecessary 6411 * to do this for pool creation since the vdev's 6412 * configuratoin has already been dirtied. 6413 */ 6414 if (tx->tx_txg != TXG_INITIAL) 6415 vdev_config_dirty(spa->spa_root_vdev); 6416 spa_history_log_internal(spa, "set", tx, 6417 "%s=%s", nvpair_name(elem), strval); 6418 break; 6419 default: 6420 /* 6421 * Set pool property values in the poolprops mos object. 6422 */ 6423 if (spa->spa_pool_props_object == 0) { 6424 spa->spa_pool_props_object = 6425 zap_create_link(mos, DMU_OT_POOL_PROPS, 6426 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6427 tx); 6428 } 6429 6430 /* normalize the property name */ 6431 propname = zpool_prop_to_name(prop); 6432 proptype = zpool_prop_get_type(prop); 6433 6434 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6435 ASSERT(proptype == PROP_TYPE_STRING); 6436 strval = fnvpair_value_string(elem); 6437 VERIFY0(zap_update(mos, 6438 spa->spa_pool_props_object, propname, 6439 1, strlen(strval) + 1, strval, tx)); 6440 spa_history_log_internal(spa, "set", tx, 6441 "%s=%s", nvpair_name(elem), strval); 6442 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6443 intval = fnvpair_value_uint64(elem); 6444 6445 if (proptype == PROP_TYPE_INDEX) { 6446 const char *unused; 6447 VERIFY0(zpool_prop_index_to_string( 6448 prop, intval, &unused)); 6449 } 6450 VERIFY0(zap_update(mos, 6451 spa->spa_pool_props_object, propname, 6452 8, 1, &intval, tx)); 6453 spa_history_log_internal(spa, "set", tx, 6454 "%s=%lld", nvpair_name(elem), intval); 6455 } else { 6456 ASSERT(0); /* not allowed */ 6457 } 6458 6459 switch (prop) { 6460 case ZPOOL_PROP_DELEGATION: 6461 spa->spa_delegation = intval; 6462 break; 6463 case ZPOOL_PROP_BOOTFS: 6464 spa->spa_bootfs = intval; 6465 break; 6466 case ZPOOL_PROP_FAILUREMODE: 6467 spa->spa_failmode = intval; 6468 break; 6469 case ZPOOL_PROP_AUTOEXPAND: 6470 spa->spa_autoexpand = intval; 6471 if (tx->tx_txg != TXG_INITIAL) 6472 spa_async_request(spa, 6473 SPA_ASYNC_AUTOEXPAND); 6474 break; 6475 case ZPOOL_PROP_DEDUPDITTO: 6476 spa->spa_dedup_ditto = intval; 6477 break; 6478 default: 6479 break; 6480 } 6481 } 6482 6483 } 6484 6485 mutex_exit(&spa->spa_props_lock); 6486} 6487 6488/* 6489 * Perform one-time upgrade on-disk changes. spa_version() does not 6490 * reflect the new version this txg, so there must be no changes this 6491 * txg to anything that the upgrade code depends on after it executes. 6492 * Therefore this must be called after dsl_pool_sync() does the sync 6493 * tasks. 6494 */ 6495static void 6496spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6497{ 6498 dsl_pool_t *dp = spa->spa_dsl_pool; 6499 6500 ASSERT(spa->spa_sync_pass == 1); 6501 6502 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6503 6504 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6505 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6506 dsl_pool_create_origin(dp, tx); 6507 6508 /* Keeping the origin open increases spa_minref */ 6509 spa->spa_minref += 3; 6510 } 6511 6512 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6513 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6514 dsl_pool_upgrade_clones(dp, tx); 6515 } 6516 6517 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6518 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6519 dsl_pool_upgrade_dir_clones(dp, tx); 6520 6521 /* Keeping the freedir open increases spa_minref */ 6522 spa->spa_minref += 3; 6523 } 6524 6525 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6526 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6527 spa_feature_create_zap_objects(spa, tx); 6528 } 6529 6530 /* 6531 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6532 * when possibility to use lz4 compression for metadata was added 6533 * Old pools that have this feature enabled must be upgraded to have 6534 * this feature active 6535 */ 6536 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6537 boolean_t lz4_en = spa_feature_is_enabled(spa, 6538 SPA_FEATURE_LZ4_COMPRESS); 6539 boolean_t lz4_ac = spa_feature_is_active(spa, 6540 SPA_FEATURE_LZ4_COMPRESS); 6541 6542 if (lz4_en && !lz4_ac) 6543 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6544 } 6545 rrw_exit(&dp->dp_config_rwlock, FTAG); 6546} 6547 6548/* 6549 * Sync the specified transaction group. New blocks may be dirtied as 6550 * part of the process, so we iterate until it converges. 6551 */ 6552void 6553spa_sync(spa_t *spa, uint64_t txg) 6554{ 6555 dsl_pool_t *dp = spa->spa_dsl_pool; 6556 objset_t *mos = spa->spa_meta_objset; 6557 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6558 vdev_t *rvd = spa->spa_root_vdev; 6559 vdev_t *vd; 6560 dmu_tx_t *tx; 6561 int error; 6562 6563 VERIFY(spa_writeable(spa)); 6564 6565 /* 6566 * Lock out configuration changes. 6567 */ 6568 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6569 6570 spa->spa_syncing_txg = txg; 6571 spa->spa_sync_pass = 0; 6572 6573 /* 6574 * If there are any pending vdev state changes, convert them 6575 * into config changes that go out with this transaction group. 6576 */ 6577 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6578 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6579 /* 6580 * We need the write lock here because, for aux vdevs, 6581 * calling vdev_config_dirty() modifies sav_config. 6582 * This is ugly and will become unnecessary when we 6583 * eliminate the aux vdev wart by integrating all vdevs 6584 * into the root vdev tree. 6585 */ 6586 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6587 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6588 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6589 vdev_state_clean(vd); 6590 vdev_config_dirty(vd); 6591 } 6592 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6593 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6594 } 6595 spa_config_exit(spa, SCL_STATE, FTAG); 6596 6597 tx = dmu_tx_create_assigned(dp, txg); 6598 6599 spa->spa_sync_starttime = gethrtime(); 6600#ifdef illumos 6601 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6602 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6603#else /* FreeBSD */ 6604#ifdef _KERNEL 6605 callout_reset(&spa->spa_deadman_cycid, 6606 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6607#endif 6608#endif 6609 6610 /* 6611 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6612 * set spa_deflate if we have no raid-z vdevs. 6613 */ 6614 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6615 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6616 int i; 6617 6618 for (i = 0; i < rvd->vdev_children; i++) { 6619 vd = rvd->vdev_child[i]; 6620 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6621 break; 6622 } 6623 if (i == rvd->vdev_children) { 6624 spa->spa_deflate = TRUE; 6625 VERIFY(0 == zap_add(spa->spa_meta_objset, 6626 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6627 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6628 } 6629 } 6630 6631 /* 6632 * Iterate to convergence. 6633 */ 6634 do { 6635 int pass = ++spa->spa_sync_pass; 6636 6637 spa_sync_config_object(spa, tx); 6638 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6639 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6640 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6641 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6642 spa_errlog_sync(spa, txg); 6643 dsl_pool_sync(dp, txg); 6644 6645 if (pass < zfs_sync_pass_deferred_free) { 6646 spa_sync_frees(spa, free_bpl, tx); 6647 } else { 6648 /* 6649 * We can not defer frees in pass 1, because 6650 * we sync the deferred frees later in pass 1. 6651 */ 6652 ASSERT3U(pass, >, 1); 6653 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6654 &spa->spa_deferred_bpobj, tx); 6655 } 6656 6657 ddt_sync(spa, txg); 6658 dsl_scan_sync(dp, tx); 6659 6660 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6661 vdev_sync(vd, txg); 6662 6663 if (pass == 1) { 6664 spa_sync_upgrades(spa, tx); 6665 ASSERT3U(txg, >=, 6666 spa->spa_uberblock.ub_rootbp.blk_birth); 6667 /* 6668 * Note: We need to check if the MOS is dirty 6669 * because we could have marked the MOS dirty 6670 * without updating the uberblock (e.g. if we 6671 * have sync tasks but no dirty user data). We 6672 * need to check the uberblock's rootbp because 6673 * it is updated if we have synced out dirty 6674 * data (though in this case the MOS will most 6675 * likely also be dirty due to second order 6676 * effects, we don't want to rely on that here). 6677 */ 6678 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6679 !dmu_objset_is_dirty(mos, txg)) { 6680 /* 6681 * Nothing changed on the first pass, 6682 * therefore this TXG is a no-op. Avoid 6683 * syncing deferred frees, so that we 6684 * can keep this TXG as a no-op. 6685 */ 6686 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6687 txg)); 6688 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6689 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6690 break; 6691 } 6692 spa_sync_deferred_frees(spa, tx); 6693 } 6694 6695 } while (dmu_objset_is_dirty(mos, txg)); 6696 6697 /* 6698 * Rewrite the vdev configuration (which includes the uberblock) 6699 * to commit the transaction group. 6700 * 6701 * If there are no dirty vdevs, we sync the uberblock to a few 6702 * random top-level vdevs that are known to be visible in the 6703 * config cache (see spa_vdev_add() for a complete description). 6704 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6705 */ 6706 for (;;) { 6707 /* 6708 * We hold SCL_STATE to prevent vdev open/close/etc. 6709 * while we're attempting to write the vdev labels. 6710 */ 6711 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6712 6713 if (list_is_empty(&spa->spa_config_dirty_list)) { 6714 vdev_t *svd[SPA_DVAS_PER_BP]; 6715 int svdcount = 0; 6716 int children = rvd->vdev_children; 6717 int c0 = spa_get_random(children); 6718 6719 for (int c = 0; c < children; c++) { 6720 vd = rvd->vdev_child[(c0 + c) % children]; 6721 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6722 continue; 6723 svd[svdcount++] = vd; 6724 if (svdcount == SPA_DVAS_PER_BP) 6725 break; 6726 } 6727 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6728 if (error != 0) 6729 error = vdev_config_sync(svd, svdcount, txg, 6730 B_TRUE); 6731 } else { 6732 error = vdev_config_sync(rvd->vdev_child, 6733 rvd->vdev_children, txg, B_FALSE); 6734 if (error != 0) 6735 error = vdev_config_sync(rvd->vdev_child, 6736 rvd->vdev_children, txg, B_TRUE); 6737 } 6738 6739 if (error == 0) 6740 spa->spa_last_synced_guid = rvd->vdev_guid; 6741 6742 spa_config_exit(spa, SCL_STATE, FTAG); 6743 6744 if (error == 0) 6745 break; 6746 zio_suspend(spa, NULL); 6747 zio_resume_wait(spa); 6748 } 6749 dmu_tx_commit(tx); 6750 6751#ifdef illumos 6752 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6753#else /* FreeBSD */ 6754#ifdef _KERNEL 6755 callout_drain(&spa->spa_deadman_cycid); 6756#endif 6757#endif 6758 6759 /* 6760 * Clear the dirty config list. 6761 */ 6762 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6763 vdev_config_clean(vd); 6764 6765 /* 6766 * Now that the new config has synced transactionally, 6767 * let it become visible to the config cache. 6768 */ 6769 if (spa->spa_config_syncing != NULL) { 6770 spa_config_set(spa, spa->spa_config_syncing); 6771 spa->spa_config_txg = txg; 6772 spa->spa_config_syncing = NULL; 6773 } 6774 6775 spa->spa_ubsync = spa->spa_uberblock; 6776 6777 dsl_pool_sync_done(dp, txg); 6778 6779 /* 6780 * Update usable space statistics. 6781 */ 6782 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6783 vdev_sync_done(vd, txg); 6784 6785 spa_update_dspace(spa); 6786 6787 /* 6788 * It had better be the case that we didn't dirty anything 6789 * since vdev_config_sync(). 6790 */ 6791 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6792 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6793 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6794 6795 spa->spa_sync_pass = 0; 6796 6797 spa_config_exit(spa, SCL_CONFIG, FTAG); 6798 6799 spa_handle_ignored_writes(spa); 6800 6801 /* 6802 * If any async tasks have been requested, kick them off. 6803 */ 6804 spa_async_dispatch(spa); 6805 spa_async_dispatch_vd(spa); 6806} 6807 6808/* 6809 * Sync all pools. We don't want to hold the namespace lock across these 6810 * operations, so we take a reference on the spa_t and drop the lock during the 6811 * sync. 6812 */ 6813void 6814spa_sync_allpools(void) 6815{ 6816 spa_t *spa = NULL; 6817 mutex_enter(&spa_namespace_lock); 6818 while ((spa = spa_next(spa)) != NULL) { 6819 if (spa_state(spa) != POOL_STATE_ACTIVE || 6820 !spa_writeable(spa) || spa_suspended(spa)) 6821 continue; 6822 spa_open_ref(spa, FTAG); 6823 mutex_exit(&spa_namespace_lock); 6824 txg_wait_synced(spa_get_dsl(spa), 0); 6825 mutex_enter(&spa_namespace_lock); 6826 spa_close(spa, FTAG); 6827 } 6828 mutex_exit(&spa_namespace_lock); 6829} 6830 6831/* 6832 * ========================================================================== 6833 * Miscellaneous routines 6834 * ========================================================================== 6835 */ 6836 6837/* 6838 * Remove all pools in the system. 6839 */ 6840void 6841spa_evict_all(void) 6842{ 6843 spa_t *spa; 6844 6845 /* 6846 * Remove all cached state. All pools should be closed now, 6847 * so every spa in the AVL tree should be unreferenced. 6848 */ 6849 mutex_enter(&spa_namespace_lock); 6850 while ((spa = spa_next(NULL)) != NULL) { 6851 /* 6852 * Stop async tasks. The async thread may need to detach 6853 * a device that's been replaced, which requires grabbing 6854 * spa_namespace_lock, so we must drop it here. 6855 */ 6856 spa_open_ref(spa, FTAG); 6857 mutex_exit(&spa_namespace_lock); 6858 spa_async_suspend(spa); 6859 mutex_enter(&spa_namespace_lock); 6860 spa_close(spa, FTAG); 6861 6862 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6863 spa_unload(spa); 6864 spa_deactivate(spa); 6865 } 6866 spa_remove(spa); 6867 } 6868 mutex_exit(&spa_namespace_lock); 6869} 6870 6871vdev_t * 6872spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6873{ 6874 vdev_t *vd; 6875 int i; 6876 6877 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6878 return (vd); 6879 6880 if (aux) { 6881 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6882 vd = spa->spa_l2cache.sav_vdevs[i]; 6883 if (vd->vdev_guid == guid) 6884 return (vd); 6885 } 6886 6887 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6888 vd = spa->spa_spares.sav_vdevs[i]; 6889 if (vd->vdev_guid == guid) 6890 return (vd); 6891 } 6892 } 6893 6894 return (NULL); 6895} 6896 6897void 6898spa_upgrade(spa_t *spa, uint64_t version) 6899{ 6900 ASSERT(spa_writeable(spa)); 6901 6902 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6903 6904 /* 6905 * This should only be called for a non-faulted pool, and since a 6906 * future version would result in an unopenable pool, this shouldn't be 6907 * possible. 6908 */ 6909 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6910 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6911 6912 spa->spa_uberblock.ub_version = version; 6913 vdev_config_dirty(spa->spa_root_vdev); 6914 6915 spa_config_exit(spa, SCL_ALL, FTAG); 6916 6917 txg_wait_synced(spa_get_dsl(spa), 0); 6918} 6919 6920boolean_t 6921spa_has_spare(spa_t *spa, uint64_t guid) 6922{ 6923 int i; 6924 uint64_t spareguid; 6925 spa_aux_vdev_t *sav = &spa->spa_spares; 6926 6927 for (i = 0; i < sav->sav_count; i++) 6928 if (sav->sav_vdevs[i]->vdev_guid == guid) 6929 return (B_TRUE); 6930 6931 for (i = 0; i < sav->sav_npending; i++) { 6932 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6933 &spareguid) == 0 && spareguid == guid) 6934 return (B_TRUE); 6935 } 6936 6937 return (B_FALSE); 6938} 6939 6940/* 6941 * Check if a pool has an active shared spare device. 6942 * Note: reference count of an active spare is 2, as a spare and as a replace 6943 */ 6944static boolean_t 6945spa_has_active_shared_spare(spa_t *spa) 6946{ 6947 int i, refcnt; 6948 uint64_t pool; 6949 spa_aux_vdev_t *sav = &spa->spa_spares; 6950 6951 for (i = 0; i < sav->sav_count; i++) { 6952 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6953 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6954 refcnt > 2) 6955 return (B_TRUE); 6956 } 6957 6958 return (B_FALSE); 6959} 6960 6961/* 6962 * Post a sysevent corresponding to the given event. The 'name' must be one of 6963 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6964 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6965 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6966 * or zdb as real changes. 6967 */ 6968void 6969spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6970{ 6971#ifdef _KERNEL 6972 sysevent_t *ev; 6973 sysevent_attr_list_t *attr = NULL; 6974 sysevent_value_t value; 6975 sysevent_id_t eid; 6976 6977 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6978 SE_SLEEP); 6979 6980 value.value_type = SE_DATA_TYPE_STRING; 6981 value.value.sv_string = spa_name(spa); 6982 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6983 goto done; 6984 6985 value.value_type = SE_DATA_TYPE_UINT64; 6986 value.value.sv_uint64 = spa_guid(spa); 6987 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6988 goto done; 6989 6990 if (vd) { 6991 value.value_type = SE_DATA_TYPE_UINT64; 6992 value.value.sv_uint64 = vd->vdev_guid; 6993 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6994 SE_SLEEP) != 0) 6995 goto done; 6996 6997 if (vd->vdev_path) { 6998 value.value_type = SE_DATA_TYPE_STRING; 6999 value.value.sv_string = vd->vdev_path; 7000 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7001 &value, SE_SLEEP) != 0) 7002 goto done; 7003 } 7004 } 7005 7006 if (sysevent_attach_attributes(ev, attr) != 0) 7007 goto done; 7008 attr = NULL; 7009 7010 (void) log_sysevent(ev, SE_SLEEP, &eid); 7011 7012done: 7013 if (attr) 7014 sysevent_free_attr(attr); 7015 sysevent_free(ev); 7016#endif 7017} 7018