spa.c revision 288597
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 1634 nvsize = *(uint64_t *)db->db_data; 1635 dmu_buf_rele(db, FTAG); 1636 1637 packed = kmem_alloc(nvsize, KM_SLEEP); 1638 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1639 DMU_READ_PREFETCH); 1640 if (error == 0) 1641 error = nvlist_unpack(packed, nvsize, value, 0); 1642 kmem_free(packed, nvsize); 1643 1644 return (error); 1645} 1646 1647/* 1648 * Checks to see if the given vdev could not be opened, in which case we post a 1649 * sysevent to notify the autoreplace code that the device has been removed. 1650 */ 1651static void 1652spa_check_removed(vdev_t *vd) 1653{ 1654 for (int c = 0; c < vd->vdev_children; c++) 1655 spa_check_removed(vd->vdev_child[c]); 1656 1657 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1658 !vd->vdev_ishole) { 1659 zfs_post_autoreplace(vd->vdev_spa, vd); 1660 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1661 } 1662} 1663 1664/* 1665 * Validate the current config against the MOS config 1666 */ 1667static boolean_t 1668spa_config_valid(spa_t *spa, nvlist_t *config) 1669{ 1670 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1671 nvlist_t *nv; 1672 1673 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1674 1675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1676 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1677 1678 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1679 1680 /* 1681 * If we're doing a normal import, then build up any additional 1682 * diagnostic information about missing devices in this config. 1683 * We'll pass this up to the user for further processing. 1684 */ 1685 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1686 nvlist_t **child, *nv; 1687 uint64_t idx = 0; 1688 1689 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1690 KM_SLEEP); 1691 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1692 1693 for (int c = 0; c < rvd->vdev_children; c++) { 1694 vdev_t *tvd = rvd->vdev_child[c]; 1695 vdev_t *mtvd = mrvd->vdev_child[c]; 1696 1697 if (tvd->vdev_ops == &vdev_missing_ops && 1698 mtvd->vdev_ops != &vdev_missing_ops && 1699 mtvd->vdev_islog) 1700 child[idx++] = vdev_config_generate(spa, mtvd, 1701 B_FALSE, 0); 1702 } 1703 1704 if (idx) { 1705 VERIFY(nvlist_add_nvlist_array(nv, 1706 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1707 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1708 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1709 1710 for (int i = 0; i < idx; i++) 1711 nvlist_free(child[i]); 1712 } 1713 nvlist_free(nv); 1714 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1715 } 1716 1717 /* 1718 * Compare the root vdev tree with the information we have 1719 * from the MOS config (mrvd). Check each top-level vdev 1720 * with the corresponding MOS config top-level (mtvd). 1721 */ 1722 for (int c = 0; c < rvd->vdev_children; c++) { 1723 vdev_t *tvd = rvd->vdev_child[c]; 1724 vdev_t *mtvd = mrvd->vdev_child[c]; 1725 1726 /* 1727 * Resolve any "missing" vdevs in the current configuration. 1728 * If we find that the MOS config has more accurate information 1729 * about the top-level vdev then use that vdev instead. 1730 */ 1731 if (tvd->vdev_ops == &vdev_missing_ops && 1732 mtvd->vdev_ops != &vdev_missing_ops) { 1733 1734 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1735 continue; 1736 1737 /* 1738 * Device specific actions. 1739 */ 1740 if (mtvd->vdev_islog) { 1741 spa_set_log_state(spa, SPA_LOG_CLEAR); 1742 } else { 1743 /* 1744 * XXX - once we have 'readonly' pool 1745 * support we should be able to handle 1746 * missing data devices by transitioning 1747 * the pool to readonly. 1748 */ 1749 continue; 1750 } 1751 1752 /* 1753 * Swap the missing vdev with the data we were 1754 * able to obtain from the MOS config. 1755 */ 1756 vdev_remove_child(rvd, tvd); 1757 vdev_remove_child(mrvd, mtvd); 1758 1759 vdev_add_child(rvd, mtvd); 1760 vdev_add_child(mrvd, tvd); 1761 1762 spa_config_exit(spa, SCL_ALL, FTAG); 1763 vdev_load(mtvd); 1764 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1765 1766 vdev_reopen(rvd); 1767 } else if (mtvd->vdev_islog) { 1768 /* 1769 * Load the slog device's state from the MOS config 1770 * since it's possible that the label does not 1771 * contain the most up-to-date information. 1772 */ 1773 vdev_load_log_state(tvd, mtvd); 1774 vdev_reopen(tvd); 1775 } 1776 } 1777 vdev_free(mrvd); 1778 spa_config_exit(spa, SCL_ALL, FTAG); 1779 1780 /* 1781 * Ensure we were able to validate the config. 1782 */ 1783 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1784} 1785 1786/* 1787 * Check for missing log devices 1788 */ 1789static boolean_t 1790spa_check_logs(spa_t *spa) 1791{ 1792 boolean_t rv = B_FALSE; 1793 dsl_pool_t *dp = spa_get_dsl(spa); 1794 1795 switch (spa->spa_log_state) { 1796 case SPA_LOG_MISSING: 1797 /* need to recheck in case slog has been restored */ 1798 case SPA_LOG_UNKNOWN: 1799 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1800 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1801 if (rv) 1802 spa_set_log_state(spa, SPA_LOG_MISSING); 1803 break; 1804 } 1805 return (rv); 1806} 1807 1808static boolean_t 1809spa_passivate_log(spa_t *spa) 1810{ 1811 vdev_t *rvd = spa->spa_root_vdev; 1812 boolean_t slog_found = B_FALSE; 1813 1814 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1815 1816 if (!spa_has_slogs(spa)) 1817 return (B_FALSE); 1818 1819 for (int c = 0; c < rvd->vdev_children; c++) { 1820 vdev_t *tvd = rvd->vdev_child[c]; 1821 metaslab_group_t *mg = tvd->vdev_mg; 1822 1823 if (tvd->vdev_islog) { 1824 metaslab_group_passivate(mg); 1825 slog_found = B_TRUE; 1826 } 1827 } 1828 1829 return (slog_found); 1830} 1831 1832static void 1833spa_activate_log(spa_t *spa) 1834{ 1835 vdev_t *rvd = spa->spa_root_vdev; 1836 1837 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1838 1839 for (int c = 0; c < rvd->vdev_children; c++) { 1840 vdev_t *tvd = rvd->vdev_child[c]; 1841 metaslab_group_t *mg = tvd->vdev_mg; 1842 1843 if (tvd->vdev_islog) 1844 metaslab_group_activate(mg); 1845 } 1846} 1847 1848int 1849spa_offline_log(spa_t *spa) 1850{ 1851 int error; 1852 1853 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1854 NULL, DS_FIND_CHILDREN); 1855 if (error == 0) { 1856 /* 1857 * We successfully offlined the log device, sync out the 1858 * current txg so that the "stubby" block can be removed 1859 * by zil_sync(). 1860 */ 1861 txg_wait_synced(spa->spa_dsl_pool, 0); 1862 } 1863 return (error); 1864} 1865 1866static void 1867spa_aux_check_removed(spa_aux_vdev_t *sav) 1868{ 1869 int i; 1870 1871 for (i = 0; i < sav->sav_count; i++) 1872 spa_check_removed(sav->sav_vdevs[i]); 1873} 1874 1875void 1876spa_claim_notify(zio_t *zio) 1877{ 1878 spa_t *spa = zio->io_spa; 1879 1880 if (zio->io_error) 1881 return; 1882 1883 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1884 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1885 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1886 mutex_exit(&spa->spa_props_lock); 1887} 1888 1889typedef struct spa_load_error { 1890 uint64_t sle_meta_count; 1891 uint64_t sle_data_count; 1892} spa_load_error_t; 1893 1894static void 1895spa_load_verify_done(zio_t *zio) 1896{ 1897 blkptr_t *bp = zio->io_bp; 1898 spa_load_error_t *sle = zio->io_private; 1899 dmu_object_type_t type = BP_GET_TYPE(bp); 1900 int error = zio->io_error; 1901 spa_t *spa = zio->io_spa; 1902 1903 if (error) { 1904 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1905 type != DMU_OT_INTENT_LOG) 1906 atomic_inc_64(&sle->sle_meta_count); 1907 else 1908 atomic_inc_64(&sle->sle_data_count); 1909 } 1910 zio_data_buf_free(zio->io_data, zio->io_size); 1911 1912 mutex_enter(&spa->spa_scrub_lock); 1913 spa->spa_scrub_inflight--; 1914 cv_broadcast(&spa->spa_scrub_io_cv); 1915 mutex_exit(&spa->spa_scrub_lock); 1916} 1917 1918/* 1919 * Maximum number of concurrent scrub i/os to create while verifying 1920 * a pool while importing it. 1921 */ 1922int spa_load_verify_maxinflight = 10000; 1923boolean_t spa_load_verify_metadata = B_TRUE; 1924boolean_t spa_load_verify_data = B_TRUE; 1925 1926SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1927 &spa_load_verify_maxinflight, 0, 1928 "Maximum number of concurrent scrub I/Os to create while verifying a " 1929 "pool while importing it"); 1930 1931SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1932 &spa_load_verify_metadata, 0, 1933 "Check metadata on import?"); 1934 1935SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1936 &spa_load_verify_data, 0, 1937 "Check user data on import?"); 1938 1939/*ARGSUSED*/ 1940static int 1941spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1942 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1943{ 1944 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1945 return (0); 1946 /* 1947 * Note: normally this routine will not be called if 1948 * spa_load_verify_metadata is not set. However, it may be useful 1949 * to manually set the flag after the traversal has begun. 1950 */ 1951 if (!spa_load_verify_metadata) 1952 return (0); 1953 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1954 return (0); 1955 1956 zio_t *rio = arg; 1957 size_t size = BP_GET_PSIZE(bp); 1958 void *data = zio_data_buf_alloc(size); 1959 1960 mutex_enter(&spa->spa_scrub_lock); 1961 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1962 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1963 spa->spa_scrub_inflight++; 1964 mutex_exit(&spa->spa_scrub_lock); 1965 1966 zio_nowait(zio_read(rio, spa, bp, data, size, 1967 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1968 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1969 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1970 return (0); 1971} 1972 1973static int 1974spa_load_verify(spa_t *spa) 1975{ 1976 zio_t *rio; 1977 spa_load_error_t sle = { 0 }; 1978 zpool_rewind_policy_t policy; 1979 boolean_t verify_ok = B_FALSE; 1980 int error = 0; 1981 1982 zpool_get_rewind_policy(spa->spa_config, &policy); 1983 1984 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1985 return (0); 1986 1987 rio = zio_root(spa, NULL, &sle, 1988 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1989 1990 if (spa_load_verify_metadata) { 1991 error = traverse_pool(spa, spa->spa_verify_min_txg, 1992 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1993 spa_load_verify_cb, rio); 1994 } 1995 1996 (void) zio_wait(rio); 1997 1998 spa->spa_load_meta_errors = sle.sle_meta_count; 1999 spa->spa_load_data_errors = sle.sle_data_count; 2000 2001 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2002 sle.sle_data_count <= policy.zrp_maxdata) { 2003 int64_t loss = 0; 2004 2005 verify_ok = B_TRUE; 2006 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2007 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2008 2009 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2010 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2011 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2012 VERIFY(nvlist_add_int64(spa->spa_load_info, 2013 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2014 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2015 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2016 } else { 2017 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2018 } 2019 2020 if (error) { 2021 if (error != ENXIO && error != EIO) 2022 error = SET_ERROR(EIO); 2023 return (error); 2024 } 2025 2026 return (verify_ok ? 0 : EIO); 2027} 2028 2029/* 2030 * Find a value in the pool props object. 2031 */ 2032static void 2033spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2034{ 2035 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2036 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2037} 2038 2039/* 2040 * Find a value in the pool directory object. 2041 */ 2042static int 2043spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2044{ 2045 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2046 name, sizeof (uint64_t), 1, val)); 2047} 2048 2049static int 2050spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2051{ 2052 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2053 return (err); 2054} 2055 2056/* 2057 * Fix up config after a partly-completed split. This is done with the 2058 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2059 * pool have that entry in their config, but only the splitting one contains 2060 * a list of all the guids of the vdevs that are being split off. 2061 * 2062 * This function determines what to do with that list: either rejoin 2063 * all the disks to the pool, or complete the splitting process. To attempt 2064 * the rejoin, each disk that is offlined is marked online again, and 2065 * we do a reopen() call. If the vdev label for every disk that was 2066 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2067 * then we call vdev_split() on each disk, and complete the split. 2068 * 2069 * Otherwise we leave the config alone, with all the vdevs in place in 2070 * the original pool. 2071 */ 2072static void 2073spa_try_repair(spa_t *spa, nvlist_t *config) 2074{ 2075 uint_t extracted; 2076 uint64_t *glist; 2077 uint_t i, gcount; 2078 nvlist_t *nvl; 2079 vdev_t **vd; 2080 boolean_t attempt_reopen; 2081 2082 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2083 return; 2084 2085 /* check that the config is complete */ 2086 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2087 &glist, &gcount) != 0) 2088 return; 2089 2090 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2091 2092 /* attempt to online all the vdevs & validate */ 2093 attempt_reopen = B_TRUE; 2094 for (i = 0; i < gcount; i++) { 2095 if (glist[i] == 0) /* vdev is hole */ 2096 continue; 2097 2098 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2099 if (vd[i] == NULL) { 2100 /* 2101 * Don't bother attempting to reopen the disks; 2102 * just do the split. 2103 */ 2104 attempt_reopen = B_FALSE; 2105 } else { 2106 /* attempt to re-online it */ 2107 vd[i]->vdev_offline = B_FALSE; 2108 } 2109 } 2110 2111 if (attempt_reopen) { 2112 vdev_reopen(spa->spa_root_vdev); 2113 2114 /* check each device to see what state it's in */ 2115 for (extracted = 0, i = 0; i < gcount; i++) { 2116 if (vd[i] != NULL && 2117 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2118 break; 2119 ++extracted; 2120 } 2121 } 2122 2123 /* 2124 * If every disk has been moved to the new pool, or if we never 2125 * even attempted to look at them, then we split them off for 2126 * good. 2127 */ 2128 if (!attempt_reopen || gcount == extracted) { 2129 for (i = 0; i < gcount; i++) 2130 if (vd[i] != NULL) 2131 vdev_split(vd[i]); 2132 vdev_reopen(spa->spa_root_vdev); 2133 } 2134 2135 kmem_free(vd, gcount * sizeof (vdev_t *)); 2136} 2137 2138static int 2139spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2140 boolean_t mosconfig) 2141{ 2142 nvlist_t *config = spa->spa_config; 2143 char *ereport = FM_EREPORT_ZFS_POOL; 2144 char *comment; 2145 int error; 2146 uint64_t pool_guid; 2147 nvlist_t *nvl; 2148 2149 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2150 return (SET_ERROR(EINVAL)); 2151 2152 ASSERT(spa->spa_comment == NULL); 2153 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2154 spa->spa_comment = spa_strdup(comment); 2155 2156 /* 2157 * Versioning wasn't explicitly added to the label until later, so if 2158 * it's not present treat it as the initial version. 2159 */ 2160 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2161 &spa->spa_ubsync.ub_version) != 0) 2162 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2163 2164 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2165 &spa->spa_config_txg); 2166 2167 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2168 spa_guid_exists(pool_guid, 0)) { 2169 error = SET_ERROR(EEXIST); 2170 } else { 2171 spa->spa_config_guid = pool_guid; 2172 2173 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2174 &nvl) == 0) { 2175 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2176 KM_SLEEP) == 0); 2177 } 2178 2179 nvlist_free(spa->spa_load_info); 2180 spa->spa_load_info = fnvlist_alloc(); 2181 2182 gethrestime(&spa->spa_loaded_ts); 2183 error = spa_load_impl(spa, pool_guid, config, state, type, 2184 mosconfig, &ereport); 2185 } 2186 2187 /* 2188 * Don't count references from objsets that are already closed 2189 * and are making their way through the eviction process. 2190 */ 2191 spa_evicting_os_wait(spa); 2192 spa->spa_minref = refcount_count(&spa->spa_refcount); 2193 if (error) { 2194 if (error != EEXIST) { 2195 spa->spa_loaded_ts.tv_sec = 0; 2196 spa->spa_loaded_ts.tv_nsec = 0; 2197 } 2198 if (error != EBADF) { 2199 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2200 } 2201 } 2202 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2203 spa->spa_ena = 0; 2204 2205 return (error); 2206} 2207 2208/* 2209 * Load an existing storage pool, using the pool's builtin spa_config as a 2210 * source of configuration information. 2211 */ 2212static int 2213spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2214 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2215 char **ereport) 2216{ 2217 int error = 0; 2218 nvlist_t *nvroot = NULL; 2219 nvlist_t *label; 2220 vdev_t *rvd; 2221 uberblock_t *ub = &spa->spa_uberblock; 2222 uint64_t children, config_cache_txg = spa->spa_config_txg; 2223 int orig_mode = spa->spa_mode; 2224 int parse; 2225 uint64_t obj; 2226 boolean_t missing_feat_write = B_FALSE; 2227 2228 /* 2229 * If this is an untrusted config, access the pool in read-only mode. 2230 * This prevents things like resilvering recently removed devices. 2231 */ 2232 if (!mosconfig) 2233 spa->spa_mode = FREAD; 2234 2235 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2236 2237 spa->spa_load_state = state; 2238 2239 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2240 return (SET_ERROR(EINVAL)); 2241 2242 parse = (type == SPA_IMPORT_EXISTING ? 2243 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2244 2245 /* 2246 * Create "The Godfather" zio to hold all async IOs 2247 */ 2248 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2249 KM_SLEEP); 2250 for (int i = 0; i < max_ncpus; i++) { 2251 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2252 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2253 ZIO_FLAG_GODFATHER); 2254 } 2255 2256 /* 2257 * Parse the configuration into a vdev tree. We explicitly set the 2258 * value that will be returned by spa_version() since parsing the 2259 * configuration requires knowing the version number. 2260 */ 2261 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2262 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2263 spa_config_exit(spa, SCL_ALL, FTAG); 2264 2265 if (error != 0) 2266 return (error); 2267 2268 ASSERT(spa->spa_root_vdev == rvd); 2269 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2270 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2271 2272 if (type != SPA_IMPORT_ASSEMBLE) { 2273 ASSERT(spa_guid(spa) == pool_guid); 2274 } 2275 2276 /* 2277 * Try to open all vdevs, loading each label in the process. 2278 */ 2279 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2280 error = vdev_open(rvd); 2281 spa_config_exit(spa, SCL_ALL, FTAG); 2282 if (error != 0) 2283 return (error); 2284 2285 /* 2286 * We need to validate the vdev labels against the configuration that 2287 * we have in hand, which is dependent on the setting of mosconfig. If 2288 * mosconfig is true then we're validating the vdev labels based on 2289 * that config. Otherwise, we're validating against the cached config 2290 * (zpool.cache) that was read when we loaded the zfs module, and then 2291 * later we will recursively call spa_load() and validate against 2292 * the vdev config. 2293 * 2294 * If we're assembling a new pool that's been split off from an 2295 * existing pool, the labels haven't yet been updated so we skip 2296 * validation for now. 2297 */ 2298 if (type != SPA_IMPORT_ASSEMBLE) { 2299 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2300 error = vdev_validate(rvd, mosconfig); 2301 spa_config_exit(spa, SCL_ALL, FTAG); 2302 2303 if (error != 0) 2304 return (error); 2305 2306 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2307 return (SET_ERROR(ENXIO)); 2308 } 2309 2310 /* 2311 * Find the best uberblock. 2312 */ 2313 vdev_uberblock_load(rvd, ub, &label); 2314 2315 /* 2316 * If we weren't able to find a single valid uberblock, return failure. 2317 */ 2318 if (ub->ub_txg == 0) { 2319 nvlist_free(label); 2320 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2321 } 2322 2323 /* 2324 * If the pool has an unsupported version we can't open it. 2325 */ 2326 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2327 nvlist_free(label); 2328 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2329 } 2330 2331 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2332 nvlist_t *features; 2333 2334 /* 2335 * If we weren't able to find what's necessary for reading the 2336 * MOS in the label, return failure. 2337 */ 2338 if (label == NULL || nvlist_lookup_nvlist(label, 2339 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2340 nvlist_free(label); 2341 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2342 ENXIO)); 2343 } 2344 2345 /* 2346 * Update our in-core representation with the definitive values 2347 * from the label. 2348 */ 2349 nvlist_free(spa->spa_label_features); 2350 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2351 } 2352 2353 nvlist_free(label); 2354 2355 /* 2356 * Look through entries in the label nvlist's features_for_read. If 2357 * there is a feature listed there which we don't understand then we 2358 * cannot open a pool. 2359 */ 2360 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2361 nvlist_t *unsup_feat; 2362 2363 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2364 0); 2365 2366 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2367 NULL); nvp != NULL; 2368 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2369 if (!zfeature_is_supported(nvpair_name(nvp))) { 2370 VERIFY(nvlist_add_string(unsup_feat, 2371 nvpair_name(nvp), "") == 0); 2372 } 2373 } 2374 2375 if (!nvlist_empty(unsup_feat)) { 2376 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2377 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2378 nvlist_free(unsup_feat); 2379 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2380 ENOTSUP)); 2381 } 2382 2383 nvlist_free(unsup_feat); 2384 } 2385 2386 /* 2387 * If the vdev guid sum doesn't match the uberblock, we have an 2388 * incomplete configuration. We first check to see if the pool 2389 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2390 * If it is, defer the vdev_guid_sum check till later so we 2391 * can handle missing vdevs. 2392 */ 2393 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2394 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2395 rvd->vdev_guid_sum != ub->ub_guid_sum) 2396 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2397 2398 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2399 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2400 spa_try_repair(spa, config); 2401 spa_config_exit(spa, SCL_ALL, FTAG); 2402 nvlist_free(spa->spa_config_splitting); 2403 spa->spa_config_splitting = NULL; 2404 } 2405 2406 /* 2407 * Initialize internal SPA structures. 2408 */ 2409 spa->spa_state = POOL_STATE_ACTIVE; 2410 spa->spa_ubsync = spa->spa_uberblock; 2411 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2412 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2413 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2414 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2415 spa->spa_claim_max_txg = spa->spa_first_txg; 2416 spa->spa_prev_software_version = ub->ub_software_version; 2417 2418 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2419 if (error) 2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2421 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2422 2423 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2424 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2425 2426 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2427 boolean_t missing_feat_read = B_FALSE; 2428 nvlist_t *unsup_feat, *enabled_feat; 2429 2430 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2431 &spa->spa_feat_for_read_obj) != 0) { 2432 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2433 } 2434 2435 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2436 &spa->spa_feat_for_write_obj) != 0) { 2437 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2438 } 2439 2440 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2441 &spa->spa_feat_desc_obj) != 0) { 2442 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2443 } 2444 2445 enabled_feat = fnvlist_alloc(); 2446 unsup_feat = fnvlist_alloc(); 2447 2448 if (!spa_features_check(spa, B_FALSE, 2449 unsup_feat, enabled_feat)) 2450 missing_feat_read = B_TRUE; 2451 2452 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2453 if (!spa_features_check(spa, B_TRUE, 2454 unsup_feat, enabled_feat)) { 2455 missing_feat_write = B_TRUE; 2456 } 2457 } 2458 2459 fnvlist_add_nvlist(spa->spa_load_info, 2460 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2461 2462 if (!nvlist_empty(unsup_feat)) { 2463 fnvlist_add_nvlist(spa->spa_load_info, 2464 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2465 } 2466 2467 fnvlist_free(enabled_feat); 2468 fnvlist_free(unsup_feat); 2469 2470 if (!missing_feat_read) { 2471 fnvlist_add_boolean(spa->spa_load_info, 2472 ZPOOL_CONFIG_CAN_RDONLY); 2473 } 2474 2475 /* 2476 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2477 * twofold: to determine whether the pool is available for 2478 * import in read-write mode and (if it is not) whether the 2479 * pool is available for import in read-only mode. If the pool 2480 * is available for import in read-write mode, it is displayed 2481 * as available in userland; if it is not available for import 2482 * in read-only mode, it is displayed as unavailable in 2483 * userland. If the pool is available for import in read-only 2484 * mode but not read-write mode, it is displayed as unavailable 2485 * in userland with a special note that the pool is actually 2486 * available for open in read-only mode. 2487 * 2488 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2489 * missing a feature for write, we must first determine whether 2490 * the pool can be opened read-only before returning to 2491 * userland in order to know whether to display the 2492 * abovementioned note. 2493 */ 2494 if (missing_feat_read || (missing_feat_write && 2495 spa_writeable(spa))) { 2496 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2497 ENOTSUP)); 2498 } 2499 2500 /* 2501 * Load refcounts for ZFS features from disk into an in-memory 2502 * cache during SPA initialization. 2503 */ 2504 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2505 uint64_t refcount; 2506 2507 error = feature_get_refcount_from_disk(spa, 2508 &spa_feature_table[i], &refcount); 2509 if (error == 0) { 2510 spa->spa_feat_refcount_cache[i] = refcount; 2511 } else if (error == ENOTSUP) { 2512 spa->spa_feat_refcount_cache[i] = 2513 SPA_FEATURE_DISABLED; 2514 } else { 2515 return (spa_vdev_err(rvd, 2516 VDEV_AUX_CORRUPT_DATA, EIO)); 2517 } 2518 } 2519 } 2520 2521 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2522 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2523 &spa->spa_feat_enabled_txg_obj) != 0) 2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2525 } 2526 2527 spa->spa_is_initializing = B_TRUE; 2528 error = dsl_pool_open(spa->spa_dsl_pool); 2529 spa->spa_is_initializing = B_FALSE; 2530 if (error != 0) 2531 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2532 2533 if (!mosconfig) { 2534 uint64_t hostid; 2535 nvlist_t *policy = NULL, *nvconfig; 2536 2537 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2538 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2539 2540 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2541 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2542 char *hostname; 2543 unsigned long myhostid = 0; 2544 2545 VERIFY(nvlist_lookup_string(nvconfig, 2546 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2547 2548#ifdef _KERNEL 2549 myhostid = zone_get_hostid(NULL); 2550#else /* _KERNEL */ 2551 /* 2552 * We're emulating the system's hostid in userland, so 2553 * we can't use zone_get_hostid(). 2554 */ 2555 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2556#endif /* _KERNEL */ 2557 if (check_hostid && hostid != 0 && myhostid != 0 && 2558 hostid != myhostid) { 2559 nvlist_free(nvconfig); 2560 cmn_err(CE_WARN, "pool '%s' could not be " 2561 "loaded as it was last accessed by " 2562 "another system (host: %s hostid: 0x%lx). " 2563 "See: http://illumos.org/msg/ZFS-8000-EY", 2564 spa_name(spa), hostname, 2565 (unsigned long)hostid); 2566 return (SET_ERROR(EBADF)); 2567 } 2568 } 2569 if (nvlist_lookup_nvlist(spa->spa_config, 2570 ZPOOL_REWIND_POLICY, &policy) == 0) 2571 VERIFY(nvlist_add_nvlist(nvconfig, 2572 ZPOOL_REWIND_POLICY, policy) == 0); 2573 2574 spa_config_set(spa, nvconfig); 2575 spa_unload(spa); 2576 spa_deactivate(spa); 2577 spa_activate(spa, orig_mode); 2578 2579 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2580 } 2581 2582 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2583 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2584 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2585 if (error != 0) 2586 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2587 2588 /* 2589 * Load the bit that tells us to use the new accounting function 2590 * (raid-z deflation). If we have an older pool, this will not 2591 * be present. 2592 */ 2593 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2594 if (error != 0 && error != ENOENT) 2595 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2596 2597 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2598 &spa->spa_creation_version); 2599 if (error != 0 && error != ENOENT) 2600 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2601 2602 /* 2603 * Load the persistent error log. If we have an older pool, this will 2604 * not be present. 2605 */ 2606 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2607 if (error != 0 && error != ENOENT) 2608 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2609 2610 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2611 &spa->spa_errlog_scrub); 2612 if (error != 0 && error != ENOENT) 2613 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2614 2615 /* 2616 * Load the history object. If we have an older pool, this 2617 * will not be present. 2618 */ 2619 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2620 if (error != 0 && error != ENOENT) 2621 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2622 2623 /* 2624 * If we're assembling the pool from the split-off vdevs of 2625 * an existing pool, we don't want to attach the spares & cache 2626 * devices. 2627 */ 2628 2629 /* 2630 * Load any hot spares for this pool. 2631 */ 2632 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2633 if (error != 0 && error != ENOENT) 2634 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2635 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2636 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2637 if (load_nvlist(spa, spa->spa_spares.sav_object, 2638 &spa->spa_spares.sav_config) != 0) 2639 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2640 2641 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2642 spa_load_spares(spa); 2643 spa_config_exit(spa, SCL_ALL, FTAG); 2644 } else if (error == 0) { 2645 spa->spa_spares.sav_sync = B_TRUE; 2646 } 2647 2648 /* 2649 * Load any level 2 ARC devices for this pool. 2650 */ 2651 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2652 &spa->spa_l2cache.sav_object); 2653 if (error != 0 && error != ENOENT) 2654 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2655 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2656 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2657 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2658 &spa->spa_l2cache.sav_config) != 0) 2659 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2660 2661 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2662 spa_load_l2cache(spa); 2663 spa_config_exit(spa, SCL_ALL, FTAG); 2664 } else if (error == 0) { 2665 spa->spa_l2cache.sav_sync = B_TRUE; 2666 } 2667 2668 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2669 2670 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2671 if (error && error != ENOENT) 2672 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2673 2674 if (error == 0) { 2675 uint64_t autoreplace; 2676 2677 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2678 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2679 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2680 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2681 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2682 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2683 &spa->spa_dedup_ditto); 2684 2685 spa->spa_autoreplace = (autoreplace != 0); 2686 } 2687 2688 /* 2689 * If the 'autoreplace' property is set, then post a resource notifying 2690 * the ZFS DE that it should not issue any faults for unopenable 2691 * devices. We also iterate over the vdevs, and post a sysevent for any 2692 * unopenable vdevs so that the normal autoreplace handler can take 2693 * over. 2694 */ 2695 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2696 spa_check_removed(spa->spa_root_vdev); 2697 /* 2698 * For the import case, this is done in spa_import(), because 2699 * at this point we're using the spare definitions from 2700 * the MOS config, not necessarily from the userland config. 2701 */ 2702 if (state != SPA_LOAD_IMPORT) { 2703 spa_aux_check_removed(&spa->spa_spares); 2704 spa_aux_check_removed(&spa->spa_l2cache); 2705 } 2706 } 2707 2708 /* 2709 * Load the vdev state for all toplevel vdevs. 2710 */ 2711 vdev_load(rvd); 2712 2713 /* 2714 * Propagate the leaf DTLs we just loaded all the way up the tree. 2715 */ 2716 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2717 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2718 spa_config_exit(spa, SCL_ALL, FTAG); 2719 2720 /* 2721 * Load the DDTs (dedup tables). 2722 */ 2723 error = ddt_load(spa); 2724 if (error != 0) 2725 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2726 2727 spa_update_dspace(spa); 2728 2729 /* 2730 * Validate the config, using the MOS config to fill in any 2731 * information which might be missing. If we fail to validate 2732 * the config then declare the pool unfit for use. If we're 2733 * assembling a pool from a split, the log is not transferred 2734 * over. 2735 */ 2736 if (type != SPA_IMPORT_ASSEMBLE) { 2737 nvlist_t *nvconfig; 2738 2739 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2740 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2741 2742 if (!spa_config_valid(spa, nvconfig)) { 2743 nvlist_free(nvconfig); 2744 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2745 ENXIO)); 2746 } 2747 nvlist_free(nvconfig); 2748 2749 /* 2750 * Now that we've validated the config, check the state of the 2751 * root vdev. If it can't be opened, it indicates one or 2752 * more toplevel vdevs are faulted. 2753 */ 2754 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2755 return (SET_ERROR(ENXIO)); 2756 2757 if (spa_writeable(spa) && spa_check_logs(spa)) { 2758 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2759 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2760 } 2761 } 2762 2763 if (missing_feat_write) { 2764 ASSERT(state == SPA_LOAD_TRYIMPORT); 2765 2766 /* 2767 * At this point, we know that we can open the pool in 2768 * read-only mode but not read-write mode. We now have enough 2769 * information and can return to userland. 2770 */ 2771 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2772 } 2773 2774 /* 2775 * We've successfully opened the pool, verify that we're ready 2776 * to start pushing transactions. 2777 */ 2778 if (state != SPA_LOAD_TRYIMPORT) { 2779 if (error = spa_load_verify(spa)) 2780 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2781 error)); 2782 } 2783 2784 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2785 spa->spa_load_max_txg == UINT64_MAX)) { 2786 dmu_tx_t *tx; 2787 int need_update = B_FALSE; 2788 dsl_pool_t *dp = spa_get_dsl(spa); 2789 2790 ASSERT(state != SPA_LOAD_TRYIMPORT); 2791 2792 /* 2793 * Claim log blocks that haven't been committed yet. 2794 * This must all happen in a single txg. 2795 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2796 * invoked from zil_claim_log_block()'s i/o done callback. 2797 * Price of rollback is that we abandon the log. 2798 */ 2799 spa->spa_claiming = B_TRUE; 2800 2801 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2802 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2803 zil_claim, tx, DS_FIND_CHILDREN); 2804 dmu_tx_commit(tx); 2805 2806 spa->spa_claiming = B_FALSE; 2807 2808 spa_set_log_state(spa, SPA_LOG_GOOD); 2809 spa->spa_sync_on = B_TRUE; 2810 txg_sync_start(spa->spa_dsl_pool); 2811 2812 /* 2813 * Wait for all claims to sync. We sync up to the highest 2814 * claimed log block birth time so that claimed log blocks 2815 * don't appear to be from the future. spa_claim_max_txg 2816 * will have been set for us by either zil_check_log_chain() 2817 * (invoked from spa_check_logs()) or zil_claim() above. 2818 */ 2819 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2820 2821 /* 2822 * If the config cache is stale, or we have uninitialized 2823 * metaslabs (see spa_vdev_add()), then update the config. 2824 * 2825 * If this is a verbatim import, trust the current 2826 * in-core spa_config and update the disk labels. 2827 */ 2828 if (config_cache_txg != spa->spa_config_txg || 2829 state == SPA_LOAD_IMPORT || 2830 state == SPA_LOAD_RECOVER || 2831 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2832 need_update = B_TRUE; 2833 2834 for (int c = 0; c < rvd->vdev_children; c++) 2835 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2836 need_update = B_TRUE; 2837 2838 /* 2839 * Update the config cache asychronously in case we're the 2840 * root pool, in which case the config cache isn't writable yet. 2841 */ 2842 if (need_update) 2843 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2844 2845 /* 2846 * Check all DTLs to see if anything needs resilvering. 2847 */ 2848 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2849 vdev_resilver_needed(rvd, NULL, NULL)) 2850 spa_async_request(spa, SPA_ASYNC_RESILVER); 2851 2852 /* 2853 * Log the fact that we booted up (so that we can detect if 2854 * we rebooted in the middle of an operation). 2855 */ 2856 spa_history_log_version(spa, "open"); 2857 2858 /* 2859 * Delete any inconsistent datasets. 2860 */ 2861 (void) dmu_objset_find(spa_name(spa), 2862 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2863 2864 /* 2865 * Clean up any stale temporary dataset userrefs. 2866 */ 2867 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2868 } 2869 2870 return (0); 2871} 2872 2873static int 2874spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2875{ 2876 int mode = spa->spa_mode; 2877 2878 spa_unload(spa); 2879 spa_deactivate(spa); 2880 2881 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2882 2883 spa_activate(spa, mode); 2884 spa_async_suspend(spa); 2885 2886 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2887} 2888 2889/* 2890 * If spa_load() fails this function will try loading prior txg's. If 2891 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2892 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2893 * function will not rewind the pool and will return the same error as 2894 * spa_load(). 2895 */ 2896static int 2897spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2898 uint64_t max_request, int rewind_flags) 2899{ 2900 nvlist_t *loadinfo = NULL; 2901 nvlist_t *config = NULL; 2902 int load_error, rewind_error; 2903 uint64_t safe_rewind_txg; 2904 uint64_t min_txg; 2905 2906 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2907 spa->spa_load_max_txg = spa->spa_load_txg; 2908 spa_set_log_state(spa, SPA_LOG_CLEAR); 2909 } else { 2910 spa->spa_load_max_txg = max_request; 2911 if (max_request != UINT64_MAX) 2912 spa->spa_extreme_rewind = B_TRUE; 2913 } 2914 2915 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2916 mosconfig); 2917 if (load_error == 0) 2918 return (0); 2919 2920 if (spa->spa_root_vdev != NULL) 2921 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2922 2923 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2924 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2925 2926 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2927 nvlist_free(config); 2928 return (load_error); 2929 } 2930 2931 if (state == SPA_LOAD_RECOVER) { 2932 /* Price of rolling back is discarding txgs, including log */ 2933 spa_set_log_state(spa, SPA_LOG_CLEAR); 2934 } else { 2935 /* 2936 * If we aren't rolling back save the load info from our first 2937 * import attempt so that we can restore it after attempting 2938 * to rewind. 2939 */ 2940 loadinfo = spa->spa_load_info; 2941 spa->spa_load_info = fnvlist_alloc(); 2942 } 2943 2944 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2945 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2946 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2947 TXG_INITIAL : safe_rewind_txg; 2948 2949 /* 2950 * Continue as long as we're finding errors, we're still within 2951 * the acceptable rewind range, and we're still finding uberblocks 2952 */ 2953 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2954 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2955 if (spa->spa_load_max_txg < safe_rewind_txg) 2956 spa->spa_extreme_rewind = B_TRUE; 2957 rewind_error = spa_load_retry(spa, state, mosconfig); 2958 } 2959 2960 spa->spa_extreme_rewind = B_FALSE; 2961 spa->spa_load_max_txg = UINT64_MAX; 2962 2963 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2964 spa_config_set(spa, config); 2965 2966 if (state == SPA_LOAD_RECOVER) { 2967 ASSERT3P(loadinfo, ==, NULL); 2968 return (rewind_error); 2969 } else { 2970 /* Store the rewind info as part of the initial load info */ 2971 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2972 spa->spa_load_info); 2973 2974 /* Restore the initial load info */ 2975 fnvlist_free(spa->spa_load_info); 2976 spa->spa_load_info = loadinfo; 2977 2978 return (load_error); 2979 } 2980} 2981 2982/* 2983 * Pool Open/Import 2984 * 2985 * The import case is identical to an open except that the configuration is sent 2986 * down from userland, instead of grabbed from the configuration cache. For the 2987 * case of an open, the pool configuration will exist in the 2988 * POOL_STATE_UNINITIALIZED state. 2989 * 2990 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2991 * the same time open the pool, without having to keep around the spa_t in some 2992 * ambiguous state. 2993 */ 2994static int 2995spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2996 nvlist_t **config) 2997{ 2998 spa_t *spa; 2999 spa_load_state_t state = SPA_LOAD_OPEN; 3000 int error; 3001 int locked = B_FALSE; 3002 int firstopen = B_FALSE; 3003 3004 *spapp = NULL; 3005 3006 /* 3007 * As disgusting as this is, we need to support recursive calls to this 3008 * function because dsl_dir_open() is called during spa_load(), and ends 3009 * up calling spa_open() again. The real fix is to figure out how to 3010 * avoid dsl_dir_open() calling this in the first place. 3011 */ 3012 if (mutex_owner(&spa_namespace_lock) != curthread) { 3013 mutex_enter(&spa_namespace_lock); 3014 locked = B_TRUE; 3015 } 3016 3017 if ((spa = spa_lookup(pool)) == NULL) { 3018 if (locked) 3019 mutex_exit(&spa_namespace_lock); 3020 return (SET_ERROR(ENOENT)); 3021 } 3022 3023 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3024 zpool_rewind_policy_t policy; 3025 3026 firstopen = B_TRUE; 3027 3028 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3029 &policy); 3030 if (policy.zrp_request & ZPOOL_DO_REWIND) 3031 state = SPA_LOAD_RECOVER; 3032 3033 spa_activate(spa, spa_mode_global); 3034 3035 if (state != SPA_LOAD_RECOVER) 3036 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3037 3038 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3039 policy.zrp_request); 3040 3041 if (error == EBADF) { 3042 /* 3043 * If vdev_validate() returns failure (indicated by 3044 * EBADF), it indicates that one of the vdevs indicates 3045 * that the pool has been exported or destroyed. If 3046 * this is the case, the config cache is out of sync and 3047 * we should remove the pool from the namespace. 3048 */ 3049 spa_unload(spa); 3050 spa_deactivate(spa); 3051 spa_config_sync(spa, B_TRUE, B_TRUE); 3052 spa_remove(spa); 3053 if (locked) 3054 mutex_exit(&spa_namespace_lock); 3055 return (SET_ERROR(ENOENT)); 3056 } 3057 3058 if (error) { 3059 /* 3060 * We can't open the pool, but we still have useful 3061 * information: the state of each vdev after the 3062 * attempted vdev_open(). Return this to the user. 3063 */ 3064 if (config != NULL && spa->spa_config) { 3065 VERIFY(nvlist_dup(spa->spa_config, config, 3066 KM_SLEEP) == 0); 3067 VERIFY(nvlist_add_nvlist(*config, 3068 ZPOOL_CONFIG_LOAD_INFO, 3069 spa->spa_load_info) == 0); 3070 } 3071 spa_unload(spa); 3072 spa_deactivate(spa); 3073 spa->spa_last_open_failed = error; 3074 if (locked) 3075 mutex_exit(&spa_namespace_lock); 3076 *spapp = NULL; 3077 return (error); 3078 } 3079 } 3080 3081 spa_open_ref(spa, tag); 3082 3083 if (config != NULL) 3084 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3085 3086 /* 3087 * If we've recovered the pool, pass back any information we 3088 * gathered while doing the load. 3089 */ 3090 if (state == SPA_LOAD_RECOVER) { 3091 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3092 spa->spa_load_info) == 0); 3093 } 3094 3095 if (locked) { 3096 spa->spa_last_open_failed = 0; 3097 spa->spa_last_ubsync_txg = 0; 3098 spa->spa_load_txg = 0; 3099 mutex_exit(&spa_namespace_lock); 3100#ifdef __FreeBSD__ 3101#ifdef _KERNEL 3102 if (firstopen) 3103 zvol_create_minors(spa->spa_name); 3104#endif 3105#endif 3106 } 3107 3108 *spapp = spa; 3109 3110 return (0); 3111} 3112 3113int 3114spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3115 nvlist_t **config) 3116{ 3117 return (spa_open_common(name, spapp, tag, policy, config)); 3118} 3119 3120int 3121spa_open(const char *name, spa_t **spapp, void *tag) 3122{ 3123 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3124} 3125 3126/* 3127 * Lookup the given spa_t, incrementing the inject count in the process, 3128 * preventing it from being exported or destroyed. 3129 */ 3130spa_t * 3131spa_inject_addref(char *name) 3132{ 3133 spa_t *spa; 3134 3135 mutex_enter(&spa_namespace_lock); 3136 if ((spa = spa_lookup(name)) == NULL) { 3137 mutex_exit(&spa_namespace_lock); 3138 return (NULL); 3139 } 3140 spa->spa_inject_ref++; 3141 mutex_exit(&spa_namespace_lock); 3142 3143 return (spa); 3144} 3145 3146void 3147spa_inject_delref(spa_t *spa) 3148{ 3149 mutex_enter(&spa_namespace_lock); 3150 spa->spa_inject_ref--; 3151 mutex_exit(&spa_namespace_lock); 3152} 3153 3154/* 3155 * Add spares device information to the nvlist. 3156 */ 3157static void 3158spa_add_spares(spa_t *spa, nvlist_t *config) 3159{ 3160 nvlist_t **spares; 3161 uint_t i, nspares; 3162 nvlist_t *nvroot; 3163 uint64_t guid; 3164 vdev_stat_t *vs; 3165 uint_t vsc; 3166 uint64_t pool; 3167 3168 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3169 3170 if (spa->spa_spares.sav_count == 0) 3171 return; 3172 3173 VERIFY(nvlist_lookup_nvlist(config, 3174 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3175 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3176 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3177 if (nspares != 0) { 3178 VERIFY(nvlist_add_nvlist_array(nvroot, 3179 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3180 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3181 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3182 3183 /* 3184 * Go through and find any spares which have since been 3185 * repurposed as an active spare. If this is the case, update 3186 * their status appropriately. 3187 */ 3188 for (i = 0; i < nspares; i++) { 3189 VERIFY(nvlist_lookup_uint64(spares[i], 3190 ZPOOL_CONFIG_GUID, &guid) == 0); 3191 if (spa_spare_exists(guid, &pool, NULL) && 3192 pool != 0ULL) { 3193 VERIFY(nvlist_lookup_uint64_array( 3194 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3195 (uint64_t **)&vs, &vsc) == 0); 3196 vs->vs_state = VDEV_STATE_CANT_OPEN; 3197 vs->vs_aux = VDEV_AUX_SPARED; 3198 } 3199 } 3200 } 3201} 3202 3203/* 3204 * Add l2cache device information to the nvlist, including vdev stats. 3205 */ 3206static void 3207spa_add_l2cache(spa_t *spa, nvlist_t *config) 3208{ 3209 nvlist_t **l2cache; 3210 uint_t i, j, nl2cache; 3211 nvlist_t *nvroot; 3212 uint64_t guid; 3213 vdev_t *vd; 3214 vdev_stat_t *vs; 3215 uint_t vsc; 3216 3217 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3218 3219 if (spa->spa_l2cache.sav_count == 0) 3220 return; 3221 3222 VERIFY(nvlist_lookup_nvlist(config, 3223 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3224 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3225 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3226 if (nl2cache != 0) { 3227 VERIFY(nvlist_add_nvlist_array(nvroot, 3228 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3229 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3230 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3231 3232 /* 3233 * Update level 2 cache device stats. 3234 */ 3235 3236 for (i = 0; i < nl2cache; i++) { 3237 VERIFY(nvlist_lookup_uint64(l2cache[i], 3238 ZPOOL_CONFIG_GUID, &guid) == 0); 3239 3240 vd = NULL; 3241 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3242 if (guid == 3243 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3244 vd = spa->spa_l2cache.sav_vdevs[j]; 3245 break; 3246 } 3247 } 3248 ASSERT(vd != NULL); 3249 3250 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3251 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3252 == 0); 3253 vdev_get_stats(vd, vs); 3254 } 3255 } 3256} 3257 3258static void 3259spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3260{ 3261 nvlist_t *features; 3262 zap_cursor_t zc; 3263 zap_attribute_t za; 3264 3265 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3266 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3267 3268 /* We may be unable to read features if pool is suspended. */ 3269 if (spa_suspended(spa)) 3270 goto out; 3271 3272 if (spa->spa_feat_for_read_obj != 0) { 3273 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3274 spa->spa_feat_for_read_obj); 3275 zap_cursor_retrieve(&zc, &za) == 0; 3276 zap_cursor_advance(&zc)) { 3277 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3278 za.za_num_integers == 1); 3279 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3280 za.za_first_integer)); 3281 } 3282 zap_cursor_fini(&zc); 3283 } 3284 3285 if (spa->spa_feat_for_write_obj != 0) { 3286 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3287 spa->spa_feat_for_write_obj); 3288 zap_cursor_retrieve(&zc, &za) == 0; 3289 zap_cursor_advance(&zc)) { 3290 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3291 za.za_num_integers == 1); 3292 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3293 za.za_first_integer)); 3294 } 3295 zap_cursor_fini(&zc); 3296 } 3297 3298out: 3299 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3300 features) == 0); 3301 nvlist_free(features); 3302} 3303 3304int 3305spa_get_stats(const char *name, nvlist_t **config, 3306 char *altroot, size_t buflen) 3307{ 3308 int error; 3309 spa_t *spa; 3310 3311 *config = NULL; 3312 error = spa_open_common(name, &spa, FTAG, NULL, config); 3313 3314 if (spa != NULL) { 3315 /* 3316 * This still leaves a window of inconsistency where the spares 3317 * or l2cache devices could change and the config would be 3318 * self-inconsistent. 3319 */ 3320 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3321 3322 if (*config != NULL) { 3323 uint64_t loadtimes[2]; 3324 3325 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3326 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3327 VERIFY(nvlist_add_uint64_array(*config, 3328 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3329 3330 VERIFY(nvlist_add_uint64(*config, 3331 ZPOOL_CONFIG_ERRCOUNT, 3332 spa_get_errlog_size(spa)) == 0); 3333 3334 if (spa_suspended(spa)) 3335 VERIFY(nvlist_add_uint64(*config, 3336 ZPOOL_CONFIG_SUSPENDED, 3337 spa->spa_failmode) == 0); 3338 3339 spa_add_spares(spa, *config); 3340 spa_add_l2cache(spa, *config); 3341 spa_add_feature_stats(spa, *config); 3342 } 3343 } 3344 3345 /* 3346 * We want to get the alternate root even for faulted pools, so we cheat 3347 * and call spa_lookup() directly. 3348 */ 3349 if (altroot) { 3350 if (spa == NULL) { 3351 mutex_enter(&spa_namespace_lock); 3352 spa = spa_lookup(name); 3353 if (spa) 3354 spa_altroot(spa, altroot, buflen); 3355 else 3356 altroot[0] = '\0'; 3357 spa = NULL; 3358 mutex_exit(&spa_namespace_lock); 3359 } else { 3360 spa_altroot(spa, altroot, buflen); 3361 } 3362 } 3363 3364 if (spa != NULL) { 3365 spa_config_exit(spa, SCL_CONFIG, FTAG); 3366 spa_close(spa, FTAG); 3367 } 3368 3369 return (error); 3370} 3371 3372/* 3373 * Validate that the auxiliary device array is well formed. We must have an 3374 * array of nvlists, each which describes a valid leaf vdev. If this is an 3375 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3376 * specified, as long as they are well-formed. 3377 */ 3378static int 3379spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3380 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3381 vdev_labeltype_t label) 3382{ 3383 nvlist_t **dev; 3384 uint_t i, ndev; 3385 vdev_t *vd; 3386 int error; 3387 3388 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3389 3390 /* 3391 * It's acceptable to have no devs specified. 3392 */ 3393 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3394 return (0); 3395 3396 if (ndev == 0) 3397 return (SET_ERROR(EINVAL)); 3398 3399 /* 3400 * Make sure the pool is formatted with a version that supports this 3401 * device type. 3402 */ 3403 if (spa_version(spa) < version) 3404 return (SET_ERROR(ENOTSUP)); 3405 3406 /* 3407 * Set the pending device list so we correctly handle device in-use 3408 * checking. 3409 */ 3410 sav->sav_pending = dev; 3411 sav->sav_npending = ndev; 3412 3413 for (i = 0; i < ndev; i++) { 3414 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3415 mode)) != 0) 3416 goto out; 3417 3418 if (!vd->vdev_ops->vdev_op_leaf) { 3419 vdev_free(vd); 3420 error = SET_ERROR(EINVAL); 3421 goto out; 3422 } 3423 3424 /* 3425 * The L2ARC currently only supports disk devices in 3426 * kernel context. For user-level testing, we allow it. 3427 */ 3428#ifdef _KERNEL 3429 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3430 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3431 error = SET_ERROR(ENOTBLK); 3432 vdev_free(vd); 3433 goto out; 3434 } 3435#endif 3436 vd->vdev_top = vd; 3437 3438 if ((error = vdev_open(vd)) == 0 && 3439 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3440 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3441 vd->vdev_guid) == 0); 3442 } 3443 3444 vdev_free(vd); 3445 3446 if (error && 3447 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3448 goto out; 3449 else 3450 error = 0; 3451 } 3452 3453out: 3454 sav->sav_pending = NULL; 3455 sav->sav_npending = 0; 3456 return (error); 3457} 3458 3459static int 3460spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3461{ 3462 int error; 3463 3464 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3465 3466 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3467 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3468 VDEV_LABEL_SPARE)) != 0) { 3469 return (error); 3470 } 3471 3472 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3473 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3474 VDEV_LABEL_L2CACHE)); 3475} 3476 3477static void 3478spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3479 const char *config) 3480{ 3481 int i; 3482 3483 if (sav->sav_config != NULL) { 3484 nvlist_t **olddevs; 3485 uint_t oldndevs; 3486 nvlist_t **newdevs; 3487 3488 /* 3489 * Generate new dev list by concatentating with the 3490 * current dev list. 3491 */ 3492 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3493 &olddevs, &oldndevs) == 0); 3494 3495 newdevs = kmem_alloc(sizeof (void *) * 3496 (ndevs + oldndevs), KM_SLEEP); 3497 for (i = 0; i < oldndevs; i++) 3498 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3499 KM_SLEEP) == 0); 3500 for (i = 0; i < ndevs; i++) 3501 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3502 KM_SLEEP) == 0); 3503 3504 VERIFY(nvlist_remove(sav->sav_config, config, 3505 DATA_TYPE_NVLIST_ARRAY) == 0); 3506 3507 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3508 config, newdevs, ndevs + oldndevs) == 0); 3509 for (i = 0; i < oldndevs + ndevs; i++) 3510 nvlist_free(newdevs[i]); 3511 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3512 } else { 3513 /* 3514 * Generate a new dev list. 3515 */ 3516 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3517 KM_SLEEP) == 0); 3518 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3519 devs, ndevs) == 0); 3520 } 3521} 3522 3523/* 3524 * Stop and drop level 2 ARC devices 3525 */ 3526void 3527spa_l2cache_drop(spa_t *spa) 3528{ 3529 vdev_t *vd; 3530 int i; 3531 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3532 3533 for (i = 0; i < sav->sav_count; i++) { 3534 uint64_t pool; 3535 3536 vd = sav->sav_vdevs[i]; 3537 ASSERT(vd != NULL); 3538 3539 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3540 pool != 0ULL && l2arc_vdev_present(vd)) 3541 l2arc_remove_vdev(vd); 3542 } 3543} 3544 3545/* 3546 * Pool Creation 3547 */ 3548int 3549spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3550 nvlist_t *zplprops) 3551{ 3552 spa_t *spa; 3553 char *altroot = NULL; 3554 vdev_t *rvd; 3555 dsl_pool_t *dp; 3556 dmu_tx_t *tx; 3557 int error = 0; 3558 uint64_t txg = TXG_INITIAL; 3559 nvlist_t **spares, **l2cache; 3560 uint_t nspares, nl2cache; 3561 uint64_t version, obj; 3562 boolean_t has_features; 3563 3564 /* 3565 * If this pool already exists, return failure. 3566 */ 3567 mutex_enter(&spa_namespace_lock); 3568 if (spa_lookup(pool) != NULL) { 3569 mutex_exit(&spa_namespace_lock); 3570 return (SET_ERROR(EEXIST)); 3571 } 3572 3573 /* 3574 * Allocate a new spa_t structure. 3575 */ 3576 (void) nvlist_lookup_string(props, 3577 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3578 spa = spa_add(pool, NULL, altroot); 3579 spa_activate(spa, spa_mode_global); 3580 3581 if (props && (error = spa_prop_validate(spa, props))) { 3582 spa_deactivate(spa); 3583 spa_remove(spa); 3584 mutex_exit(&spa_namespace_lock); 3585 return (error); 3586 } 3587 3588 has_features = B_FALSE; 3589 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3590 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3591 if (zpool_prop_feature(nvpair_name(elem))) 3592 has_features = B_TRUE; 3593 } 3594 3595 if (has_features || nvlist_lookup_uint64(props, 3596 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3597 version = SPA_VERSION; 3598 } 3599 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3600 3601 spa->spa_first_txg = txg; 3602 spa->spa_uberblock.ub_txg = txg - 1; 3603 spa->spa_uberblock.ub_version = version; 3604 spa->spa_ubsync = spa->spa_uberblock; 3605 3606 /* 3607 * Create "The Godfather" zio to hold all async IOs 3608 */ 3609 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3610 KM_SLEEP); 3611 for (int i = 0; i < max_ncpus; i++) { 3612 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3613 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3614 ZIO_FLAG_GODFATHER); 3615 } 3616 3617 /* 3618 * Create the root vdev. 3619 */ 3620 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3621 3622 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3623 3624 ASSERT(error != 0 || rvd != NULL); 3625 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3626 3627 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3628 error = SET_ERROR(EINVAL); 3629 3630 if (error == 0 && 3631 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3632 (error = spa_validate_aux(spa, nvroot, txg, 3633 VDEV_ALLOC_ADD)) == 0) { 3634 for (int c = 0; c < rvd->vdev_children; c++) { 3635 vdev_ashift_optimize(rvd->vdev_child[c]); 3636 vdev_metaslab_set_size(rvd->vdev_child[c]); 3637 vdev_expand(rvd->vdev_child[c], txg); 3638 } 3639 } 3640 3641 spa_config_exit(spa, SCL_ALL, FTAG); 3642 3643 if (error != 0) { 3644 spa_unload(spa); 3645 spa_deactivate(spa); 3646 spa_remove(spa); 3647 mutex_exit(&spa_namespace_lock); 3648 return (error); 3649 } 3650 3651 /* 3652 * Get the list of spares, if specified. 3653 */ 3654 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3655 &spares, &nspares) == 0) { 3656 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3657 KM_SLEEP) == 0); 3658 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3659 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3660 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3661 spa_load_spares(spa); 3662 spa_config_exit(spa, SCL_ALL, FTAG); 3663 spa->spa_spares.sav_sync = B_TRUE; 3664 } 3665 3666 /* 3667 * Get the list of level 2 cache devices, if specified. 3668 */ 3669 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3670 &l2cache, &nl2cache) == 0) { 3671 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3672 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3673 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3674 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3676 spa_load_l2cache(spa); 3677 spa_config_exit(spa, SCL_ALL, FTAG); 3678 spa->spa_l2cache.sav_sync = B_TRUE; 3679 } 3680 3681 spa->spa_is_initializing = B_TRUE; 3682 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3683 spa->spa_meta_objset = dp->dp_meta_objset; 3684 spa->spa_is_initializing = B_FALSE; 3685 3686 /* 3687 * Create DDTs (dedup tables). 3688 */ 3689 ddt_create(spa); 3690 3691 spa_update_dspace(spa); 3692 3693 tx = dmu_tx_create_assigned(dp, txg); 3694 3695 /* 3696 * Create the pool config object. 3697 */ 3698 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3699 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3700 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3701 3702 if (zap_add(spa->spa_meta_objset, 3703 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3704 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3705 cmn_err(CE_PANIC, "failed to add pool config"); 3706 } 3707 3708 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3709 spa_feature_create_zap_objects(spa, tx); 3710 3711 if (zap_add(spa->spa_meta_objset, 3712 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3713 sizeof (uint64_t), 1, &version, tx) != 0) { 3714 cmn_err(CE_PANIC, "failed to add pool version"); 3715 } 3716 3717 /* Newly created pools with the right version are always deflated. */ 3718 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3719 spa->spa_deflate = TRUE; 3720 if (zap_add(spa->spa_meta_objset, 3721 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3722 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3723 cmn_err(CE_PANIC, "failed to add deflate"); 3724 } 3725 } 3726 3727 /* 3728 * Create the deferred-free bpobj. Turn off compression 3729 * because sync-to-convergence takes longer if the blocksize 3730 * keeps changing. 3731 */ 3732 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3733 dmu_object_set_compress(spa->spa_meta_objset, obj, 3734 ZIO_COMPRESS_OFF, tx); 3735 if (zap_add(spa->spa_meta_objset, 3736 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3737 sizeof (uint64_t), 1, &obj, tx) != 0) { 3738 cmn_err(CE_PANIC, "failed to add bpobj"); 3739 } 3740 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3741 spa->spa_meta_objset, obj)); 3742 3743 /* 3744 * Create the pool's history object. 3745 */ 3746 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3747 spa_history_create_obj(spa, tx); 3748 3749 /* 3750 * Set pool properties. 3751 */ 3752 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3753 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3754 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3755 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3756 3757 if (props != NULL) { 3758 spa_configfile_set(spa, props, B_FALSE); 3759 spa_sync_props(props, tx); 3760 } 3761 3762 dmu_tx_commit(tx); 3763 3764 spa->spa_sync_on = B_TRUE; 3765 txg_sync_start(spa->spa_dsl_pool); 3766 3767 /* 3768 * We explicitly wait for the first transaction to complete so that our 3769 * bean counters are appropriately updated. 3770 */ 3771 txg_wait_synced(spa->spa_dsl_pool, txg); 3772 3773 spa_config_sync(spa, B_FALSE, B_TRUE); 3774 3775 spa_history_log_version(spa, "create"); 3776 3777 /* 3778 * Don't count references from objsets that are already closed 3779 * and are making their way through the eviction process. 3780 */ 3781 spa_evicting_os_wait(spa); 3782 spa->spa_minref = refcount_count(&spa->spa_refcount); 3783 3784 mutex_exit(&spa_namespace_lock); 3785 3786 return (0); 3787} 3788 3789#ifdef _KERNEL 3790#if defined(sun) 3791/* 3792 * Get the root pool information from the root disk, then import the root pool 3793 * during the system boot up time. 3794 */ 3795extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3796 3797static nvlist_t * 3798spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3799{ 3800 nvlist_t *config; 3801 nvlist_t *nvtop, *nvroot; 3802 uint64_t pgid; 3803 3804 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3805 return (NULL); 3806 3807 /* 3808 * Add this top-level vdev to the child array. 3809 */ 3810 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3811 &nvtop) == 0); 3812 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3813 &pgid) == 0); 3814 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3815 3816 /* 3817 * Put this pool's top-level vdevs into a root vdev. 3818 */ 3819 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3820 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3821 VDEV_TYPE_ROOT) == 0); 3822 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3823 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3824 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3825 &nvtop, 1) == 0); 3826 3827 /* 3828 * Replace the existing vdev_tree with the new root vdev in 3829 * this pool's configuration (remove the old, add the new). 3830 */ 3831 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3832 nvlist_free(nvroot); 3833 return (config); 3834} 3835 3836/* 3837 * Walk the vdev tree and see if we can find a device with "better" 3838 * configuration. A configuration is "better" if the label on that 3839 * device has a more recent txg. 3840 */ 3841static void 3842spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3843{ 3844 for (int c = 0; c < vd->vdev_children; c++) 3845 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3846 3847 if (vd->vdev_ops->vdev_op_leaf) { 3848 nvlist_t *label; 3849 uint64_t label_txg; 3850 3851 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3852 &label) != 0) 3853 return; 3854 3855 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3856 &label_txg) == 0); 3857 3858 /* 3859 * Do we have a better boot device? 3860 */ 3861 if (label_txg > *txg) { 3862 *txg = label_txg; 3863 *avd = vd; 3864 } 3865 nvlist_free(label); 3866 } 3867} 3868 3869/* 3870 * Import a root pool. 3871 * 3872 * For x86. devpath_list will consist of devid and/or physpath name of 3873 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3874 * The GRUB "findroot" command will return the vdev we should boot. 3875 * 3876 * For Sparc, devpath_list consists the physpath name of the booting device 3877 * no matter the rootpool is a single device pool or a mirrored pool. 3878 * e.g. 3879 * "/pci@1f,0/ide@d/disk@0,0:a" 3880 */ 3881int 3882spa_import_rootpool(char *devpath, char *devid) 3883{ 3884 spa_t *spa; 3885 vdev_t *rvd, *bvd, *avd = NULL; 3886 nvlist_t *config, *nvtop; 3887 uint64_t guid, txg; 3888 char *pname; 3889 int error; 3890 3891 /* 3892 * Read the label from the boot device and generate a configuration. 3893 */ 3894 config = spa_generate_rootconf(devpath, devid, &guid); 3895#if defined(_OBP) && defined(_KERNEL) 3896 if (config == NULL) { 3897 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3898 /* iscsi boot */ 3899 get_iscsi_bootpath_phy(devpath); 3900 config = spa_generate_rootconf(devpath, devid, &guid); 3901 } 3902 } 3903#endif 3904 if (config == NULL) { 3905 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3906 devpath); 3907 return (SET_ERROR(EIO)); 3908 } 3909 3910 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3911 &pname) == 0); 3912 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3913 3914 mutex_enter(&spa_namespace_lock); 3915 if ((spa = spa_lookup(pname)) != NULL) { 3916 /* 3917 * Remove the existing root pool from the namespace so that we 3918 * can replace it with the correct config we just read in. 3919 */ 3920 spa_remove(spa); 3921 } 3922 3923 spa = spa_add(pname, config, NULL); 3924 spa->spa_is_root = B_TRUE; 3925 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3926 3927 /* 3928 * Build up a vdev tree based on the boot device's label config. 3929 */ 3930 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3931 &nvtop) == 0); 3932 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3933 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3934 VDEV_ALLOC_ROOTPOOL); 3935 spa_config_exit(spa, SCL_ALL, FTAG); 3936 if (error) { 3937 mutex_exit(&spa_namespace_lock); 3938 nvlist_free(config); 3939 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3940 pname); 3941 return (error); 3942 } 3943 3944 /* 3945 * Get the boot vdev. 3946 */ 3947 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3948 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3949 (u_longlong_t)guid); 3950 error = SET_ERROR(ENOENT); 3951 goto out; 3952 } 3953 3954 /* 3955 * Determine if there is a better boot device. 3956 */ 3957 avd = bvd; 3958 spa_alt_rootvdev(rvd, &avd, &txg); 3959 if (avd != bvd) { 3960 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3961 "try booting from '%s'", avd->vdev_path); 3962 error = SET_ERROR(EINVAL); 3963 goto out; 3964 } 3965 3966 /* 3967 * If the boot device is part of a spare vdev then ensure that 3968 * we're booting off the active spare. 3969 */ 3970 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3971 !bvd->vdev_isspare) { 3972 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3973 "try booting from '%s'", 3974 bvd->vdev_parent-> 3975 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3976 error = SET_ERROR(EINVAL); 3977 goto out; 3978 } 3979 3980 error = 0; 3981out: 3982 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3983 vdev_free(rvd); 3984 spa_config_exit(spa, SCL_ALL, FTAG); 3985 mutex_exit(&spa_namespace_lock); 3986 3987 nvlist_free(config); 3988 return (error); 3989} 3990 3991#else 3992 3993extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3994 uint64_t *count); 3995 3996static nvlist_t * 3997spa_generate_rootconf(const char *name) 3998{ 3999 nvlist_t **configs, **tops; 4000 nvlist_t *config; 4001 nvlist_t *best_cfg, *nvtop, *nvroot; 4002 uint64_t *holes; 4003 uint64_t best_txg; 4004 uint64_t nchildren; 4005 uint64_t pgid; 4006 uint64_t count; 4007 uint64_t i; 4008 uint_t nholes; 4009 4010 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4011 return (NULL); 4012 4013 ASSERT3U(count, !=, 0); 4014 best_txg = 0; 4015 for (i = 0; i < count; i++) { 4016 uint64_t txg; 4017 4018 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4019 &txg) == 0); 4020 if (txg > best_txg) { 4021 best_txg = txg; 4022 best_cfg = configs[i]; 4023 } 4024 } 4025 4026 /* 4027 * Multi-vdev root pool configuration discovery is not supported yet. 4028 */ 4029 nchildren = 1; 4030 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4031 holes = NULL; 4032 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4033 &holes, &nholes); 4034 4035 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4036 for (i = 0; i < nchildren; i++) { 4037 if (i >= count) 4038 break; 4039 if (configs[i] == NULL) 4040 continue; 4041 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4042 &nvtop) == 0); 4043 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4044 } 4045 for (i = 0; holes != NULL && i < nholes; i++) { 4046 if (i >= nchildren) 4047 continue; 4048 if (tops[holes[i]] != NULL) 4049 continue; 4050 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4051 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4052 VDEV_TYPE_HOLE) == 0); 4053 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4054 holes[i]) == 0); 4055 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4056 0) == 0); 4057 } 4058 for (i = 0; i < nchildren; i++) { 4059 if (tops[i] != NULL) 4060 continue; 4061 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4062 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4063 VDEV_TYPE_MISSING) == 0); 4064 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4065 i) == 0); 4066 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4067 0) == 0); 4068 } 4069 4070 /* 4071 * Create pool config based on the best vdev config. 4072 */ 4073 nvlist_dup(best_cfg, &config, KM_SLEEP); 4074 4075 /* 4076 * Put this pool's top-level vdevs into a root vdev. 4077 */ 4078 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4079 &pgid) == 0); 4080 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4081 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4082 VDEV_TYPE_ROOT) == 0); 4083 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4084 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4085 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4086 tops, nchildren) == 0); 4087 4088 /* 4089 * Replace the existing vdev_tree with the new root vdev in 4090 * this pool's configuration (remove the old, add the new). 4091 */ 4092 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4093 4094 /* 4095 * Drop vdev config elements that should not be present at pool level. 4096 */ 4097 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4098 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4099 4100 for (i = 0; i < count; i++) 4101 nvlist_free(configs[i]); 4102 kmem_free(configs, count * sizeof(void *)); 4103 for (i = 0; i < nchildren; i++) 4104 nvlist_free(tops[i]); 4105 kmem_free(tops, nchildren * sizeof(void *)); 4106 nvlist_free(nvroot); 4107 return (config); 4108} 4109 4110int 4111spa_import_rootpool(const char *name) 4112{ 4113 spa_t *spa; 4114 vdev_t *rvd, *bvd, *avd = NULL; 4115 nvlist_t *config, *nvtop; 4116 uint64_t txg; 4117 char *pname; 4118 int error; 4119 4120 /* 4121 * Read the label from the boot device and generate a configuration. 4122 */ 4123 config = spa_generate_rootconf(name); 4124 4125 mutex_enter(&spa_namespace_lock); 4126 if (config != NULL) { 4127 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4128 &pname) == 0 && strcmp(name, pname) == 0); 4129 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4130 == 0); 4131 4132 if ((spa = spa_lookup(pname)) != NULL) { 4133 /* 4134 * Remove the existing root pool from the namespace so 4135 * that we can replace it with the correct config 4136 * we just read in. 4137 */ 4138 spa_remove(spa); 4139 } 4140 spa = spa_add(pname, config, NULL); 4141 4142 /* 4143 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4144 * via spa_version(). 4145 */ 4146 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4147 &spa->spa_ubsync.ub_version) != 0) 4148 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4149 } else if ((spa = spa_lookup(name)) == NULL) { 4150 mutex_exit(&spa_namespace_lock); 4151 nvlist_free(config); 4152 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4153 name); 4154 return (EIO); 4155 } else { 4156 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4157 } 4158 spa->spa_is_root = B_TRUE; 4159 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4160 4161 /* 4162 * Build up a vdev tree based on the boot device's label config. 4163 */ 4164 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4165 &nvtop) == 0); 4166 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4167 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4168 VDEV_ALLOC_ROOTPOOL); 4169 spa_config_exit(spa, SCL_ALL, FTAG); 4170 if (error) { 4171 mutex_exit(&spa_namespace_lock); 4172 nvlist_free(config); 4173 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4174 pname); 4175 return (error); 4176 } 4177 4178 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4179 vdev_free(rvd); 4180 spa_config_exit(spa, SCL_ALL, FTAG); 4181 mutex_exit(&spa_namespace_lock); 4182 4183 nvlist_free(config); 4184 return (0); 4185} 4186 4187#endif /* sun */ 4188#endif 4189 4190/* 4191 * Import a non-root pool into the system. 4192 */ 4193int 4194spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4195{ 4196 spa_t *spa; 4197 char *altroot = NULL; 4198 spa_load_state_t state = SPA_LOAD_IMPORT; 4199 zpool_rewind_policy_t policy; 4200 uint64_t mode = spa_mode_global; 4201 uint64_t readonly = B_FALSE; 4202 int error; 4203 nvlist_t *nvroot; 4204 nvlist_t **spares, **l2cache; 4205 uint_t nspares, nl2cache; 4206 4207 /* 4208 * If a pool with this name exists, return failure. 4209 */ 4210 mutex_enter(&spa_namespace_lock); 4211 if (spa_lookup(pool) != NULL) { 4212 mutex_exit(&spa_namespace_lock); 4213 return (SET_ERROR(EEXIST)); 4214 } 4215 4216 /* 4217 * Create and initialize the spa structure. 4218 */ 4219 (void) nvlist_lookup_string(props, 4220 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4221 (void) nvlist_lookup_uint64(props, 4222 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4223 if (readonly) 4224 mode = FREAD; 4225 spa = spa_add(pool, config, altroot); 4226 spa->spa_import_flags = flags; 4227 4228 /* 4229 * Verbatim import - Take a pool and insert it into the namespace 4230 * as if it had been loaded at boot. 4231 */ 4232 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4233 if (props != NULL) 4234 spa_configfile_set(spa, props, B_FALSE); 4235 4236 spa_config_sync(spa, B_FALSE, B_TRUE); 4237 4238 mutex_exit(&spa_namespace_lock); 4239 return (0); 4240 } 4241 4242 spa_activate(spa, mode); 4243 4244 /* 4245 * Don't start async tasks until we know everything is healthy. 4246 */ 4247 spa_async_suspend(spa); 4248 4249 zpool_get_rewind_policy(config, &policy); 4250 if (policy.zrp_request & ZPOOL_DO_REWIND) 4251 state = SPA_LOAD_RECOVER; 4252 4253 /* 4254 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4255 * because the user-supplied config is actually the one to trust when 4256 * doing an import. 4257 */ 4258 if (state != SPA_LOAD_RECOVER) 4259 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4260 4261 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4262 policy.zrp_request); 4263 4264 /* 4265 * Propagate anything learned while loading the pool and pass it 4266 * back to caller (i.e. rewind info, missing devices, etc). 4267 */ 4268 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4269 spa->spa_load_info) == 0); 4270 4271 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4272 /* 4273 * Toss any existing sparelist, as it doesn't have any validity 4274 * anymore, and conflicts with spa_has_spare(). 4275 */ 4276 if (spa->spa_spares.sav_config) { 4277 nvlist_free(spa->spa_spares.sav_config); 4278 spa->spa_spares.sav_config = NULL; 4279 spa_load_spares(spa); 4280 } 4281 if (spa->spa_l2cache.sav_config) { 4282 nvlist_free(spa->spa_l2cache.sav_config); 4283 spa->spa_l2cache.sav_config = NULL; 4284 spa_load_l2cache(spa); 4285 } 4286 4287 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4288 &nvroot) == 0); 4289 if (error == 0) 4290 error = spa_validate_aux(spa, nvroot, -1ULL, 4291 VDEV_ALLOC_SPARE); 4292 if (error == 0) 4293 error = spa_validate_aux(spa, nvroot, -1ULL, 4294 VDEV_ALLOC_L2CACHE); 4295 spa_config_exit(spa, SCL_ALL, FTAG); 4296 4297 if (props != NULL) 4298 spa_configfile_set(spa, props, B_FALSE); 4299 4300 if (error != 0 || (props && spa_writeable(spa) && 4301 (error = spa_prop_set(spa, props)))) { 4302 spa_unload(spa); 4303 spa_deactivate(spa); 4304 spa_remove(spa); 4305 mutex_exit(&spa_namespace_lock); 4306 return (error); 4307 } 4308 4309 spa_async_resume(spa); 4310 4311 /* 4312 * Override any spares and level 2 cache devices as specified by 4313 * the user, as these may have correct device names/devids, etc. 4314 */ 4315 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4316 &spares, &nspares) == 0) { 4317 if (spa->spa_spares.sav_config) 4318 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4319 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4320 else 4321 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4322 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4323 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4324 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4325 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4326 spa_load_spares(spa); 4327 spa_config_exit(spa, SCL_ALL, FTAG); 4328 spa->spa_spares.sav_sync = B_TRUE; 4329 } 4330 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4331 &l2cache, &nl2cache) == 0) { 4332 if (spa->spa_l2cache.sav_config) 4333 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4334 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4335 else 4336 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4337 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4338 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4339 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4340 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4341 spa_load_l2cache(spa); 4342 spa_config_exit(spa, SCL_ALL, FTAG); 4343 spa->spa_l2cache.sav_sync = B_TRUE; 4344 } 4345 4346 /* 4347 * Check for any removed devices. 4348 */ 4349 if (spa->spa_autoreplace) { 4350 spa_aux_check_removed(&spa->spa_spares); 4351 spa_aux_check_removed(&spa->spa_l2cache); 4352 } 4353 4354 if (spa_writeable(spa)) { 4355 /* 4356 * Update the config cache to include the newly-imported pool. 4357 */ 4358 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4359 } 4360 4361 /* 4362 * It's possible that the pool was expanded while it was exported. 4363 * We kick off an async task to handle this for us. 4364 */ 4365 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4366 4367 mutex_exit(&spa_namespace_lock); 4368 spa_history_log_version(spa, "import"); 4369 4370#ifdef __FreeBSD__ 4371#ifdef _KERNEL 4372 zvol_create_minors(pool); 4373#endif 4374#endif 4375 return (0); 4376} 4377 4378nvlist_t * 4379spa_tryimport(nvlist_t *tryconfig) 4380{ 4381 nvlist_t *config = NULL; 4382 char *poolname; 4383 spa_t *spa; 4384 uint64_t state; 4385 int error; 4386 4387 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4388 return (NULL); 4389 4390 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4391 return (NULL); 4392 4393 /* 4394 * Create and initialize the spa structure. 4395 */ 4396 mutex_enter(&spa_namespace_lock); 4397 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4398 spa_activate(spa, FREAD); 4399 4400 /* 4401 * Pass off the heavy lifting to spa_load(). 4402 * Pass TRUE for mosconfig because the user-supplied config 4403 * is actually the one to trust when doing an import. 4404 */ 4405 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4406 4407 /* 4408 * If 'tryconfig' was at least parsable, return the current config. 4409 */ 4410 if (spa->spa_root_vdev != NULL) { 4411 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4412 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4413 poolname) == 0); 4414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4415 state) == 0); 4416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4417 spa->spa_uberblock.ub_timestamp) == 0); 4418 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4419 spa->spa_load_info) == 0); 4420 4421 /* 4422 * If the bootfs property exists on this pool then we 4423 * copy it out so that external consumers can tell which 4424 * pools are bootable. 4425 */ 4426 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4427 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4428 4429 /* 4430 * We have to play games with the name since the 4431 * pool was opened as TRYIMPORT_NAME. 4432 */ 4433 if (dsl_dsobj_to_dsname(spa_name(spa), 4434 spa->spa_bootfs, tmpname) == 0) { 4435 char *cp; 4436 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4437 4438 cp = strchr(tmpname, '/'); 4439 if (cp == NULL) { 4440 (void) strlcpy(dsname, tmpname, 4441 MAXPATHLEN); 4442 } else { 4443 (void) snprintf(dsname, MAXPATHLEN, 4444 "%s/%s", poolname, ++cp); 4445 } 4446 VERIFY(nvlist_add_string(config, 4447 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4448 kmem_free(dsname, MAXPATHLEN); 4449 } 4450 kmem_free(tmpname, MAXPATHLEN); 4451 } 4452 4453 /* 4454 * Add the list of hot spares and level 2 cache devices. 4455 */ 4456 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4457 spa_add_spares(spa, config); 4458 spa_add_l2cache(spa, config); 4459 spa_config_exit(spa, SCL_CONFIG, FTAG); 4460 } 4461 4462 spa_unload(spa); 4463 spa_deactivate(spa); 4464 spa_remove(spa); 4465 mutex_exit(&spa_namespace_lock); 4466 4467 return (config); 4468} 4469 4470/* 4471 * Pool export/destroy 4472 * 4473 * The act of destroying or exporting a pool is very simple. We make sure there 4474 * is no more pending I/O and any references to the pool are gone. Then, we 4475 * update the pool state and sync all the labels to disk, removing the 4476 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4477 * we don't sync the labels or remove the configuration cache. 4478 */ 4479static int 4480spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4481 boolean_t force, boolean_t hardforce) 4482{ 4483 spa_t *spa; 4484 4485 if (oldconfig) 4486 *oldconfig = NULL; 4487 4488 if (!(spa_mode_global & FWRITE)) 4489 return (SET_ERROR(EROFS)); 4490 4491 mutex_enter(&spa_namespace_lock); 4492 if ((spa = spa_lookup(pool)) == NULL) { 4493 mutex_exit(&spa_namespace_lock); 4494 return (SET_ERROR(ENOENT)); 4495 } 4496 4497 /* 4498 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4499 * reacquire the namespace lock, and see if we can export. 4500 */ 4501 spa_open_ref(spa, FTAG); 4502 mutex_exit(&spa_namespace_lock); 4503 spa_async_suspend(spa); 4504 mutex_enter(&spa_namespace_lock); 4505 spa_close(spa, FTAG); 4506 4507 /* 4508 * The pool will be in core if it's openable, 4509 * in which case we can modify its state. 4510 */ 4511 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4512 /* 4513 * Objsets may be open only because they're dirty, so we 4514 * have to force it to sync before checking spa_refcnt. 4515 */ 4516 txg_wait_synced(spa->spa_dsl_pool, 0); 4517 spa_evicting_os_wait(spa); 4518 4519 /* 4520 * A pool cannot be exported or destroyed if there are active 4521 * references. If we are resetting a pool, allow references by 4522 * fault injection handlers. 4523 */ 4524 if (!spa_refcount_zero(spa) || 4525 (spa->spa_inject_ref != 0 && 4526 new_state != POOL_STATE_UNINITIALIZED)) { 4527 spa_async_resume(spa); 4528 mutex_exit(&spa_namespace_lock); 4529 return (SET_ERROR(EBUSY)); 4530 } 4531 4532 /* 4533 * A pool cannot be exported if it has an active shared spare. 4534 * This is to prevent other pools stealing the active spare 4535 * from an exported pool. At user's own will, such pool can 4536 * be forcedly exported. 4537 */ 4538 if (!force && new_state == POOL_STATE_EXPORTED && 4539 spa_has_active_shared_spare(spa)) { 4540 spa_async_resume(spa); 4541 mutex_exit(&spa_namespace_lock); 4542 return (SET_ERROR(EXDEV)); 4543 } 4544 4545 /* 4546 * We want this to be reflected on every label, 4547 * so mark them all dirty. spa_unload() will do the 4548 * final sync that pushes these changes out. 4549 */ 4550 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4551 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4552 spa->spa_state = new_state; 4553 spa->spa_final_txg = spa_last_synced_txg(spa) + 4554 TXG_DEFER_SIZE + 1; 4555 vdev_config_dirty(spa->spa_root_vdev); 4556 spa_config_exit(spa, SCL_ALL, FTAG); 4557 } 4558 } 4559 4560 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4561 4562 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4563 spa_unload(spa); 4564 spa_deactivate(spa); 4565 } 4566 4567 if (oldconfig && spa->spa_config) 4568 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4569 4570 if (new_state != POOL_STATE_UNINITIALIZED) { 4571 if (!hardforce) 4572 spa_config_sync(spa, B_TRUE, B_TRUE); 4573 spa_remove(spa); 4574 } 4575 mutex_exit(&spa_namespace_lock); 4576 4577 return (0); 4578} 4579 4580/* 4581 * Destroy a storage pool. 4582 */ 4583int 4584spa_destroy(char *pool) 4585{ 4586 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4587 B_FALSE, B_FALSE)); 4588} 4589 4590/* 4591 * Export a storage pool. 4592 */ 4593int 4594spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4595 boolean_t hardforce) 4596{ 4597 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4598 force, hardforce)); 4599} 4600 4601/* 4602 * Similar to spa_export(), this unloads the spa_t without actually removing it 4603 * from the namespace in any way. 4604 */ 4605int 4606spa_reset(char *pool) 4607{ 4608 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4609 B_FALSE, B_FALSE)); 4610} 4611 4612/* 4613 * ========================================================================== 4614 * Device manipulation 4615 * ========================================================================== 4616 */ 4617 4618/* 4619 * Add a device to a storage pool. 4620 */ 4621int 4622spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4623{ 4624 uint64_t txg, id; 4625 int error; 4626 vdev_t *rvd = spa->spa_root_vdev; 4627 vdev_t *vd, *tvd; 4628 nvlist_t **spares, **l2cache; 4629 uint_t nspares, nl2cache; 4630 4631 ASSERT(spa_writeable(spa)); 4632 4633 txg = spa_vdev_enter(spa); 4634 4635 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4636 VDEV_ALLOC_ADD)) != 0) 4637 return (spa_vdev_exit(spa, NULL, txg, error)); 4638 4639 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4640 4641 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4642 &nspares) != 0) 4643 nspares = 0; 4644 4645 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4646 &nl2cache) != 0) 4647 nl2cache = 0; 4648 4649 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4650 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4651 4652 if (vd->vdev_children != 0 && 4653 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4654 return (spa_vdev_exit(spa, vd, txg, error)); 4655 4656 /* 4657 * We must validate the spares and l2cache devices after checking the 4658 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4659 */ 4660 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4661 return (spa_vdev_exit(spa, vd, txg, error)); 4662 4663 /* 4664 * Transfer each new top-level vdev from vd to rvd. 4665 */ 4666 for (int c = 0; c < vd->vdev_children; c++) { 4667 4668 /* 4669 * Set the vdev id to the first hole, if one exists. 4670 */ 4671 for (id = 0; id < rvd->vdev_children; id++) { 4672 if (rvd->vdev_child[id]->vdev_ishole) { 4673 vdev_free(rvd->vdev_child[id]); 4674 break; 4675 } 4676 } 4677 tvd = vd->vdev_child[c]; 4678 vdev_remove_child(vd, tvd); 4679 tvd->vdev_id = id; 4680 vdev_add_child(rvd, tvd); 4681 vdev_config_dirty(tvd); 4682 } 4683 4684 if (nspares != 0) { 4685 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4686 ZPOOL_CONFIG_SPARES); 4687 spa_load_spares(spa); 4688 spa->spa_spares.sav_sync = B_TRUE; 4689 } 4690 4691 if (nl2cache != 0) { 4692 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4693 ZPOOL_CONFIG_L2CACHE); 4694 spa_load_l2cache(spa); 4695 spa->spa_l2cache.sav_sync = B_TRUE; 4696 } 4697 4698 /* 4699 * We have to be careful when adding new vdevs to an existing pool. 4700 * If other threads start allocating from these vdevs before we 4701 * sync the config cache, and we lose power, then upon reboot we may 4702 * fail to open the pool because there are DVAs that the config cache 4703 * can't translate. Therefore, we first add the vdevs without 4704 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4705 * and then let spa_config_update() initialize the new metaslabs. 4706 * 4707 * spa_load() checks for added-but-not-initialized vdevs, so that 4708 * if we lose power at any point in this sequence, the remaining 4709 * steps will be completed the next time we load the pool. 4710 */ 4711 (void) spa_vdev_exit(spa, vd, txg, 0); 4712 4713 mutex_enter(&spa_namespace_lock); 4714 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4715 mutex_exit(&spa_namespace_lock); 4716 4717 return (0); 4718} 4719 4720/* 4721 * Attach a device to a mirror. The arguments are the path to any device 4722 * in the mirror, and the nvroot for the new device. If the path specifies 4723 * a device that is not mirrored, we automatically insert the mirror vdev. 4724 * 4725 * If 'replacing' is specified, the new device is intended to replace the 4726 * existing device; in this case the two devices are made into their own 4727 * mirror using the 'replacing' vdev, which is functionally identical to 4728 * the mirror vdev (it actually reuses all the same ops) but has a few 4729 * extra rules: you can't attach to it after it's been created, and upon 4730 * completion of resilvering, the first disk (the one being replaced) 4731 * is automatically detached. 4732 */ 4733int 4734spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4735{ 4736 uint64_t txg, dtl_max_txg; 4737 vdev_t *rvd = spa->spa_root_vdev; 4738 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4739 vdev_ops_t *pvops; 4740 char *oldvdpath, *newvdpath; 4741 int newvd_isspare; 4742 int error; 4743 4744 ASSERT(spa_writeable(spa)); 4745 4746 txg = spa_vdev_enter(spa); 4747 4748 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4749 4750 if (oldvd == NULL) 4751 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4752 4753 if (!oldvd->vdev_ops->vdev_op_leaf) 4754 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4755 4756 pvd = oldvd->vdev_parent; 4757 4758 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4759 VDEV_ALLOC_ATTACH)) != 0) 4760 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4761 4762 if (newrootvd->vdev_children != 1) 4763 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4764 4765 newvd = newrootvd->vdev_child[0]; 4766 4767 if (!newvd->vdev_ops->vdev_op_leaf) 4768 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4769 4770 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4771 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4772 4773 /* 4774 * Spares can't replace logs 4775 */ 4776 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4777 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4778 4779 if (!replacing) { 4780 /* 4781 * For attach, the only allowable parent is a mirror or the root 4782 * vdev. 4783 */ 4784 if (pvd->vdev_ops != &vdev_mirror_ops && 4785 pvd->vdev_ops != &vdev_root_ops) 4786 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4787 4788 pvops = &vdev_mirror_ops; 4789 } else { 4790 /* 4791 * Active hot spares can only be replaced by inactive hot 4792 * spares. 4793 */ 4794 if (pvd->vdev_ops == &vdev_spare_ops && 4795 oldvd->vdev_isspare && 4796 !spa_has_spare(spa, newvd->vdev_guid)) 4797 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4798 4799 /* 4800 * If the source is a hot spare, and the parent isn't already a 4801 * spare, then we want to create a new hot spare. Otherwise, we 4802 * want to create a replacing vdev. The user is not allowed to 4803 * attach to a spared vdev child unless the 'isspare' state is 4804 * the same (spare replaces spare, non-spare replaces 4805 * non-spare). 4806 */ 4807 if (pvd->vdev_ops == &vdev_replacing_ops && 4808 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4809 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4810 } else if (pvd->vdev_ops == &vdev_spare_ops && 4811 newvd->vdev_isspare != oldvd->vdev_isspare) { 4812 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4813 } 4814 4815 if (newvd->vdev_isspare) 4816 pvops = &vdev_spare_ops; 4817 else 4818 pvops = &vdev_replacing_ops; 4819 } 4820 4821 /* 4822 * Make sure the new device is big enough. 4823 */ 4824 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4825 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4826 4827 /* 4828 * The new device cannot have a higher alignment requirement 4829 * than the top-level vdev. 4830 */ 4831 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4832 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4833 4834 /* 4835 * If this is an in-place replacement, update oldvd's path and devid 4836 * to make it distinguishable from newvd, and unopenable from now on. 4837 */ 4838 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4839 spa_strfree(oldvd->vdev_path); 4840 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4841 KM_SLEEP); 4842 (void) sprintf(oldvd->vdev_path, "%s/%s", 4843 newvd->vdev_path, "old"); 4844 if (oldvd->vdev_devid != NULL) { 4845 spa_strfree(oldvd->vdev_devid); 4846 oldvd->vdev_devid = NULL; 4847 } 4848 } 4849 4850 /* mark the device being resilvered */ 4851 newvd->vdev_resilver_txg = txg; 4852 4853 /* 4854 * If the parent is not a mirror, or if we're replacing, insert the new 4855 * mirror/replacing/spare vdev above oldvd. 4856 */ 4857 if (pvd->vdev_ops != pvops) 4858 pvd = vdev_add_parent(oldvd, pvops); 4859 4860 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4861 ASSERT(pvd->vdev_ops == pvops); 4862 ASSERT(oldvd->vdev_parent == pvd); 4863 4864 /* 4865 * Extract the new device from its root and add it to pvd. 4866 */ 4867 vdev_remove_child(newrootvd, newvd); 4868 newvd->vdev_id = pvd->vdev_children; 4869 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4870 vdev_add_child(pvd, newvd); 4871 4872 tvd = newvd->vdev_top; 4873 ASSERT(pvd->vdev_top == tvd); 4874 ASSERT(tvd->vdev_parent == rvd); 4875 4876 vdev_config_dirty(tvd); 4877 4878 /* 4879 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4880 * for any dmu_sync-ed blocks. It will propagate upward when 4881 * spa_vdev_exit() calls vdev_dtl_reassess(). 4882 */ 4883 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4884 4885 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4886 dtl_max_txg - TXG_INITIAL); 4887 4888 if (newvd->vdev_isspare) { 4889 spa_spare_activate(newvd); 4890 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4891 } 4892 4893 oldvdpath = spa_strdup(oldvd->vdev_path); 4894 newvdpath = spa_strdup(newvd->vdev_path); 4895 newvd_isspare = newvd->vdev_isspare; 4896 4897 /* 4898 * Mark newvd's DTL dirty in this txg. 4899 */ 4900 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4901 4902 /* 4903 * Schedule the resilver to restart in the future. We do this to 4904 * ensure that dmu_sync-ed blocks have been stitched into the 4905 * respective datasets. 4906 */ 4907 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4908 4909 /* 4910 * Commit the config 4911 */ 4912 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4913 4914 spa_history_log_internal(spa, "vdev attach", NULL, 4915 "%s vdev=%s %s vdev=%s", 4916 replacing && newvd_isspare ? "spare in" : 4917 replacing ? "replace" : "attach", newvdpath, 4918 replacing ? "for" : "to", oldvdpath); 4919 4920 spa_strfree(oldvdpath); 4921 spa_strfree(newvdpath); 4922 4923 if (spa->spa_bootfs) 4924 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4925 4926 return (0); 4927} 4928 4929/* 4930 * Detach a device from a mirror or replacing vdev. 4931 * 4932 * If 'replace_done' is specified, only detach if the parent 4933 * is a replacing vdev. 4934 */ 4935int 4936spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4937{ 4938 uint64_t txg; 4939 int error; 4940 vdev_t *rvd = spa->spa_root_vdev; 4941 vdev_t *vd, *pvd, *cvd, *tvd; 4942 boolean_t unspare = B_FALSE; 4943 uint64_t unspare_guid = 0; 4944 char *vdpath; 4945 4946 ASSERT(spa_writeable(spa)); 4947 4948 txg = spa_vdev_enter(spa); 4949 4950 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4951 4952 if (vd == NULL) 4953 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4954 4955 if (!vd->vdev_ops->vdev_op_leaf) 4956 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4957 4958 pvd = vd->vdev_parent; 4959 4960 /* 4961 * If the parent/child relationship is not as expected, don't do it. 4962 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4963 * vdev that's replacing B with C. The user's intent in replacing 4964 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4965 * the replace by detaching C, the expected behavior is to end up 4966 * M(A,B). But suppose that right after deciding to detach C, 4967 * the replacement of B completes. We would have M(A,C), and then 4968 * ask to detach C, which would leave us with just A -- not what 4969 * the user wanted. To prevent this, we make sure that the 4970 * parent/child relationship hasn't changed -- in this example, 4971 * that C's parent is still the replacing vdev R. 4972 */ 4973 if (pvd->vdev_guid != pguid && pguid != 0) 4974 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4975 4976 /* 4977 * Only 'replacing' or 'spare' vdevs can be replaced. 4978 */ 4979 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4980 pvd->vdev_ops != &vdev_spare_ops) 4981 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4982 4983 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4984 spa_version(spa) >= SPA_VERSION_SPARES); 4985 4986 /* 4987 * Only mirror, replacing, and spare vdevs support detach. 4988 */ 4989 if (pvd->vdev_ops != &vdev_replacing_ops && 4990 pvd->vdev_ops != &vdev_mirror_ops && 4991 pvd->vdev_ops != &vdev_spare_ops) 4992 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4993 4994 /* 4995 * If this device has the only valid copy of some data, 4996 * we cannot safely detach it. 4997 */ 4998 if (vdev_dtl_required(vd)) 4999 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5000 5001 ASSERT(pvd->vdev_children >= 2); 5002 5003 /* 5004 * If we are detaching the second disk from a replacing vdev, then 5005 * check to see if we changed the original vdev's path to have "/old" 5006 * at the end in spa_vdev_attach(). If so, undo that change now. 5007 */ 5008 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5009 vd->vdev_path != NULL) { 5010 size_t len = strlen(vd->vdev_path); 5011 5012 for (int c = 0; c < pvd->vdev_children; c++) { 5013 cvd = pvd->vdev_child[c]; 5014 5015 if (cvd == vd || cvd->vdev_path == NULL) 5016 continue; 5017 5018 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5019 strcmp(cvd->vdev_path + len, "/old") == 0) { 5020 spa_strfree(cvd->vdev_path); 5021 cvd->vdev_path = spa_strdup(vd->vdev_path); 5022 break; 5023 } 5024 } 5025 } 5026 5027 /* 5028 * If we are detaching the original disk from a spare, then it implies 5029 * that the spare should become a real disk, and be removed from the 5030 * active spare list for the pool. 5031 */ 5032 if (pvd->vdev_ops == &vdev_spare_ops && 5033 vd->vdev_id == 0 && 5034 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5035 unspare = B_TRUE; 5036 5037 /* 5038 * Erase the disk labels so the disk can be used for other things. 5039 * This must be done after all other error cases are handled, 5040 * but before we disembowel vd (so we can still do I/O to it). 5041 * But if we can't do it, don't treat the error as fatal -- 5042 * it may be that the unwritability of the disk is the reason 5043 * it's being detached! 5044 */ 5045 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5046 5047 /* 5048 * Remove vd from its parent and compact the parent's children. 5049 */ 5050 vdev_remove_child(pvd, vd); 5051 vdev_compact_children(pvd); 5052 5053 /* 5054 * Remember one of the remaining children so we can get tvd below. 5055 */ 5056 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5057 5058 /* 5059 * If we need to remove the remaining child from the list of hot spares, 5060 * do it now, marking the vdev as no longer a spare in the process. 5061 * We must do this before vdev_remove_parent(), because that can 5062 * change the GUID if it creates a new toplevel GUID. For a similar 5063 * reason, we must remove the spare now, in the same txg as the detach; 5064 * otherwise someone could attach a new sibling, change the GUID, and 5065 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5066 */ 5067 if (unspare) { 5068 ASSERT(cvd->vdev_isspare); 5069 spa_spare_remove(cvd); 5070 unspare_guid = cvd->vdev_guid; 5071 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5072 cvd->vdev_unspare = B_TRUE; 5073 } 5074 5075 /* 5076 * If the parent mirror/replacing vdev only has one child, 5077 * the parent is no longer needed. Remove it from the tree. 5078 */ 5079 if (pvd->vdev_children == 1) { 5080 if (pvd->vdev_ops == &vdev_spare_ops) 5081 cvd->vdev_unspare = B_FALSE; 5082 vdev_remove_parent(cvd); 5083 } 5084 5085 5086 /* 5087 * We don't set tvd until now because the parent we just removed 5088 * may have been the previous top-level vdev. 5089 */ 5090 tvd = cvd->vdev_top; 5091 ASSERT(tvd->vdev_parent == rvd); 5092 5093 /* 5094 * Reevaluate the parent vdev state. 5095 */ 5096 vdev_propagate_state(cvd); 5097 5098 /* 5099 * If the 'autoexpand' property is set on the pool then automatically 5100 * try to expand the size of the pool. For example if the device we 5101 * just detached was smaller than the others, it may be possible to 5102 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5103 * first so that we can obtain the updated sizes of the leaf vdevs. 5104 */ 5105 if (spa->spa_autoexpand) { 5106 vdev_reopen(tvd); 5107 vdev_expand(tvd, txg); 5108 } 5109 5110 vdev_config_dirty(tvd); 5111 5112 /* 5113 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5114 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5115 * But first make sure we're not on any *other* txg's DTL list, to 5116 * prevent vd from being accessed after it's freed. 5117 */ 5118 vdpath = spa_strdup(vd->vdev_path); 5119 for (int t = 0; t < TXG_SIZE; t++) 5120 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5121 vd->vdev_detached = B_TRUE; 5122 vdev_dirty(tvd, VDD_DTL, vd, txg); 5123 5124 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5125 5126 /* hang on to the spa before we release the lock */ 5127 spa_open_ref(spa, FTAG); 5128 5129 error = spa_vdev_exit(spa, vd, txg, 0); 5130 5131 spa_history_log_internal(spa, "detach", NULL, 5132 "vdev=%s", vdpath); 5133 spa_strfree(vdpath); 5134 5135 /* 5136 * If this was the removal of the original device in a hot spare vdev, 5137 * then we want to go through and remove the device from the hot spare 5138 * list of every other pool. 5139 */ 5140 if (unspare) { 5141 spa_t *altspa = NULL; 5142 5143 mutex_enter(&spa_namespace_lock); 5144 while ((altspa = spa_next(altspa)) != NULL) { 5145 if (altspa->spa_state != POOL_STATE_ACTIVE || 5146 altspa == spa) 5147 continue; 5148 5149 spa_open_ref(altspa, FTAG); 5150 mutex_exit(&spa_namespace_lock); 5151 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5152 mutex_enter(&spa_namespace_lock); 5153 spa_close(altspa, FTAG); 5154 } 5155 mutex_exit(&spa_namespace_lock); 5156 5157 /* search the rest of the vdevs for spares to remove */ 5158 spa_vdev_resilver_done(spa); 5159 } 5160 5161 /* all done with the spa; OK to release */ 5162 mutex_enter(&spa_namespace_lock); 5163 spa_close(spa, FTAG); 5164 mutex_exit(&spa_namespace_lock); 5165 5166 return (error); 5167} 5168 5169/* 5170 * Split a set of devices from their mirrors, and create a new pool from them. 5171 */ 5172int 5173spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5174 nvlist_t *props, boolean_t exp) 5175{ 5176 int error = 0; 5177 uint64_t txg, *glist; 5178 spa_t *newspa; 5179 uint_t c, children, lastlog; 5180 nvlist_t **child, *nvl, *tmp; 5181 dmu_tx_t *tx; 5182 char *altroot = NULL; 5183 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5184 boolean_t activate_slog; 5185 5186 ASSERT(spa_writeable(spa)); 5187 5188 txg = spa_vdev_enter(spa); 5189 5190 /* clear the log and flush everything up to now */ 5191 activate_slog = spa_passivate_log(spa); 5192 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5193 error = spa_offline_log(spa); 5194 txg = spa_vdev_config_enter(spa); 5195 5196 if (activate_slog) 5197 spa_activate_log(spa); 5198 5199 if (error != 0) 5200 return (spa_vdev_exit(spa, NULL, txg, error)); 5201 5202 /* check new spa name before going any further */ 5203 if (spa_lookup(newname) != NULL) 5204 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5205 5206 /* 5207 * scan through all the children to ensure they're all mirrors 5208 */ 5209 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5210 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5211 &children) != 0) 5212 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5213 5214 /* first, check to ensure we've got the right child count */ 5215 rvd = spa->spa_root_vdev; 5216 lastlog = 0; 5217 for (c = 0; c < rvd->vdev_children; c++) { 5218 vdev_t *vd = rvd->vdev_child[c]; 5219 5220 /* don't count the holes & logs as children */ 5221 if (vd->vdev_islog || vd->vdev_ishole) { 5222 if (lastlog == 0) 5223 lastlog = c; 5224 continue; 5225 } 5226 5227 lastlog = 0; 5228 } 5229 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5230 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5231 5232 /* next, ensure no spare or cache devices are part of the split */ 5233 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5234 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5235 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5236 5237 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5238 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5239 5240 /* then, loop over each vdev and validate it */ 5241 for (c = 0; c < children; c++) { 5242 uint64_t is_hole = 0; 5243 5244 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5245 &is_hole); 5246 5247 if (is_hole != 0) { 5248 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5249 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5250 continue; 5251 } else { 5252 error = SET_ERROR(EINVAL); 5253 break; 5254 } 5255 } 5256 5257 /* which disk is going to be split? */ 5258 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5259 &glist[c]) != 0) { 5260 error = SET_ERROR(EINVAL); 5261 break; 5262 } 5263 5264 /* look it up in the spa */ 5265 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5266 if (vml[c] == NULL) { 5267 error = SET_ERROR(ENODEV); 5268 break; 5269 } 5270 5271 /* make sure there's nothing stopping the split */ 5272 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5273 vml[c]->vdev_islog || 5274 vml[c]->vdev_ishole || 5275 vml[c]->vdev_isspare || 5276 vml[c]->vdev_isl2cache || 5277 !vdev_writeable(vml[c]) || 5278 vml[c]->vdev_children != 0 || 5279 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5280 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5281 error = SET_ERROR(EINVAL); 5282 break; 5283 } 5284 5285 if (vdev_dtl_required(vml[c])) { 5286 error = SET_ERROR(EBUSY); 5287 break; 5288 } 5289 5290 /* we need certain info from the top level */ 5291 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5292 vml[c]->vdev_top->vdev_ms_array) == 0); 5293 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5294 vml[c]->vdev_top->vdev_ms_shift) == 0); 5295 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5296 vml[c]->vdev_top->vdev_asize) == 0); 5297 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5298 vml[c]->vdev_top->vdev_ashift) == 0); 5299 } 5300 5301 if (error != 0) { 5302 kmem_free(vml, children * sizeof (vdev_t *)); 5303 kmem_free(glist, children * sizeof (uint64_t)); 5304 return (spa_vdev_exit(spa, NULL, txg, error)); 5305 } 5306 5307 /* stop writers from using the disks */ 5308 for (c = 0; c < children; c++) { 5309 if (vml[c] != NULL) 5310 vml[c]->vdev_offline = B_TRUE; 5311 } 5312 vdev_reopen(spa->spa_root_vdev); 5313 5314 /* 5315 * Temporarily record the splitting vdevs in the spa config. This 5316 * will disappear once the config is regenerated. 5317 */ 5318 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5319 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5320 glist, children) == 0); 5321 kmem_free(glist, children * sizeof (uint64_t)); 5322 5323 mutex_enter(&spa->spa_props_lock); 5324 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5325 nvl) == 0); 5326 mutex_exit(&spa->spa_props_lock); 5327 spa->spa_config_splitting = nvl; 5328 vdev_config_dirty(spa->spa_root_vdev); 5329 5330 /* configure and create the new pool */ 5331 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5332 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5333 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5334 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5335 spa_version(spa)) == 0); 5336 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5337 spa->spa_config_txg) == 0); 5338 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5339 spa_generate_guid(NULL)) == 0); 5340 (void) nvlist_lookup_string(props, 5341 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5342 5343 /* add the new pool to the namespace */ 5344 newspa = spa_add(newname, config, altroot); 5345 newspa->spa_config_txg = spa->spa_config_txg; 5346 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5347 5348 /* release the spa config lock, retaining the namespace lock */ 5349 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5350 5351 if (zio_injection_enabled) 5352 zio_handle_panic_injection(spa, FTAG, 1); 5353 5354 spa_activate(newspa, spa_mode_global); 5355 spa_async_suspend(newspa); 5356 5357#ifndef sun 5358 /* mark that we are creating new spa by splitting */ 5359 newspa->spa_splitting_newspa = B_TRUE; 5360#endif 5361 /* create the new pool from the disks of the original pool */ 5362 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5363#ifndef sun 5364 newspa->spa_splitting_newspa = B_FALSE; 5365#endif 5366 if (error) 5367 goto out; 5368 5369 /* if that worked, generate a real config for the new pool */ 5370 if (newspa->spa_root_vdev != NULL) { 5371 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5372 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5373 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5374 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5375 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5376 B_TRUE)); 5377 } 5378 5379 /* set the props */ 5380 if (props != NULL) { 5381 spa_configfile_set(newspa, props, B_FALSE); 5382 error = spa_prop_set(newspa, props); 5383 if (error) 5384 goto out; 5385 } 5386 5387 /* flush everything */ 5388 txg = spa_vdev_config_enter(newspa); 5389 vdev_config_dirty(newspa->spa_root_vdev); 5390 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5391 5392 if (zio_injection_enabled) 5393 zio_handle_panic_injection(spa, FTAG, 2); 5394 5395 spa_async_resume(newspa); 5396 5397 /* finally, update the original pool's config */ 5398 txg = spa_vdev_config_enter(spa); 5399 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5400 error = dmu_tx_assign(tx, TXG_WAIT); 5401 if (error != 0) 5402 dmu_tx_abort(tx); 5403 for (c = 0; c < children; c++) { 5404 if (vml[c] != NULL) { 5405 vdev_split(vml[c]); 5406 if (error == 0) 5407 spa_history_log_internal(spa, "detach", tx, 5408 "vdev=%s", vml[c]->vdev_path); 5409 vdev_free(vml[c]); 5410 } 5411 } 5412 vdev_config_dirty(spa->spa_root_vdev); 5413 spa->spa_config_splitting = NULL; 5414 nvlist_free(nvl); 5415 if (error == 0) 5416 dmu_tx_commit(tx); 5417 (void) spa_vdev_exit(spa, NULL, txg, 0); 5418 5419 if (zio_injection_enabled) 5420 zio_handle_panic_injection(spa, FTAG, 3); 5421 5422 /* split is complete; log a history record */ 5423 spa_history_log_internal(newspa, "split", NULL, 5424 "from pool %s", spa_name(spa)); 5425 5426 kmem_free(vml, children * sizeof (vdev_t *)); 5427 5428 /* if we're not going to mount the filesystems in userland, export */ 5429 if (exp) 5430 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5431 B_FALSE, B_FALSE); 5432 5433 return (error); 5434 5435out: 5436 spa_unload(newspa); 5437 spa_deactivate(newspa); 5438 spa_remove(newspa); 5439 5440 txg = spa_vdev_config_enter(spa); 5441 5442 /* re-online all offlined disks */ 5443 for (c = 0; c < children; c++) { 5444 if (vml[c] != NULL) 5445 vml[c]->vdev_offline = B_FALSE; 5446 } 5447 vdev_reopen(spa->spa_root_vdev); 5448 5449 nvlist_free(spa->spa_config_splitting); 5450 spa->spa_config_splitting = NULL; 5451 (void) spa_vdev_exit(spa, NULL, txg, error); 5452 5453 kmem_free(vml, children * sizeof (vdev_t *)); 5454 return (error); 5455} 5456 5457static nvlist_t * 5458spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5459{ 5460 for (int i = 0; i < count; i++) { 5461 uint64_t guid; 5462 5463 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5464 &guid) == 0); 5465 5466 if (guid == target_guid) 5467 return (nvpp[i]); 5468 } 5469 5470 return (NULL); 5471} 5472 5473static void 5474spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5475 nvlist_t *dev_to_remove) 5476{ 5477 nvlist_t **newdev = NULL; 5478 5479 if (count > 1) 5480 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5481 5482 for (int i = 0, j = 0; i < count; i++) { 5483 if (dev[i] == dev_to_remove) 5484 continue; 5485 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5486 } 5487 5488 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5489 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5490 5491 for (int i = 0; i < count - 1; i++) 5492 nvlist_free(newdev[i]); 5493 5494 if (count > 1) 5495 kmem_free(newdev, (count - 1) * sizeof (void *)); 5496} 5497 5498/* 5499 * Evacuate the device. 5500 */ 5501static int 5502spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5503{ 5504 uint64_t txg; 5505 int error = 0; 5506 5507 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5508 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5509 ASSERT(vd == vd->vdev_top); 5510 5511 /* 5512 * Evacuate the device. We don't hold the config lock as writer 5513 * since we need to do I/O but we do keep the 5514 * spa_namespace_lock held. Once this completes the device 5515 * should no longer have any blocks allocated on it. 5516 */ 5517 if (vd->vdev_islog) { 5518 if (vd->vdev_stat.vs_alloc != 0) 5519 error = spa_offline_log(spa); 5520 } else { 5521 error = SET_ERROR(ENOTSUP); 5522 } 5523 5524 if (error) 5525 return (error); 5526 5527 /* 5528 * The evacuation succeeded. Remove any remaining MOS metadata 5529 * associated with this vdev, and wait for these changes to sync. 5530 */ 5531 ASSERT0(vd->vdev_stat.vs_alloc); 5532 txg = spa_vdev_config_enter(spa); 5533 vd->vdev_removing = B_TRUE; 5534 vdev_dirty_leaves(vd, VDD_DTL, txg); 5535 vdev_config_dirty(vd); 5536 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5537 5538 return (0); 5539} 5540 5541/* 5542 * Complete the removal by cleaning up the namespace. 5543 */ 5544static void 5545spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5546{ 5547 vdev_t *rvd = spa->spa_root_vdev; 5548 uint64_t id = vd->vdev_id; 5549 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5550 5551 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5552 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5553 ASSERT(vd == vd->vdev_top); 5554 5555 /* 5556 * Only remove any devices which are empty. 5557 */ 5558 if (vd->vdev_stat.vs_alloc != 0) 5559 return; 5560 5561 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5562 5563 if (list_link_active(&vd->vdev_state_dirty_node)) 5564 vdev_state_clean(vd); 5565 if (list_link_active(&vd->vdev_config_dirty_node)) 5566 vdev_config_clean(vd); 5567 5568 vdev_free(vd); 5569 5570 if (last_vdev) { 5571 vdev_compact_children(rvd); 5572 } else { 5573 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5574 vdev_add_child(rvd, vd); 5575 } 5576 vdev_config_dirty(rvd); 5577 5578 /* 5579 * Reassess the health of our root vdev. 5580 */ 5581 vdev_reopen(rvd); 5582} 5583 5584/* 5585 * Remove a device from the pool - 5586 * 5587 * Removing a device from the vdev namespace requires several steps 5588 * and can take a significant amount of time. As a result we use 5589 * the spa_vdev_config_[enter/exit] functions which allow us to 5590 * grab and release the spa_config_lock while still holding the namespace 5591 * lock. During each step the configuration is synced out. 5592 * 5593 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5594 * devices. 5595 */ 5596int 5597spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5598{ 5599 vdev_t *vd; 5600 metaslab_group_t *mg; 5601 nvlist_t **spares, **l2cache, *nv; 5602 uint64_t txg = 0; 5603 uint_t nspares, nl2cache; 5604 int error = 0; 5605 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5606 5607 ASSERT(spa_writeable(spa)); 5608 5609 if (!locked) 5610 txg = spa_vdev_enter(spa); 5611 5612 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5613 5614 if (spa->spa_spares.sav_vdevs != NULL && 5615 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5616 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5617 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5618 /* 5619 * Only remove the hot spare if it's not currently in use 5620 * in this pool. 5621 */ 5622 if (vd == NULL || unspare) { 5623 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5624 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5625 spa_load_spares(spa); 5626 spa->spa_spares.sav_sync = B_TRUE; 5627 } else { 5628 error = SET_ERROR(EBUSY); 5629 } 5630 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5631 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5632 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5633 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5634 /* 5635 * Cache devices can always be removed. 5636 */ 5637 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5638 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5639 spa_load_l2cache(spa); 5640 spa->spa_l2cache.sav_sync = B_TRUE; 5641 } else if (vd != NULL && vd->vdev_islog) { 5642 ASSERT(!locked); 5643 ASSERT(vd == vd->vdev_top); 5644 5645 mg = vd->vdev_mg; 5646 5647 /* 5648 * Stop allocating from this vdev. 5649 */ 5650 metaslab_group_passivate(mg); 5651 5652 /* 5653 * Wait for the youngest allocations and frees to sync, 5654 * and then wait for the deferral of those frees to finish. 5655 */ 5656 spa_vdev_config_exit(spa, NULL, 5657 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5658 5659 /* 5660 * Attempt to evacuate the vdev. 5661 */ 5662 error = spa_vdev_remove_evacuate(spa, vd); 5663 5664 txg = spa_vdev_config_enter(spa); 5665 5666 /* 5667 * If we couldn't evacuate the vdev, unwind. 5668 */ 5669 if (error) { 5670 metaslab_group_activate(mg); 5671 return (spa_vdev_exit(spa, NULL, txg, error)); 5672 } 5673 5674 /* 5675 * Clean up the vdev namespace. 5676 */ 5677 spa_vdev_remove_from_namespace(spa, vd); 5678 5679 } else if (vd != NULL) { 5680 /* 5681 * Normal vdevs cannot be removed (yet). 5682 */ 5683 error = SET_ERROR(ENOTSUP); 5684 } else { 5685 /* 5686 * There is no vdev of any kind with the specified guid. 5687 */ 5688 error = SET_ERROR(ENOENT); 5689 } 5690 5691 if (!locked) 5692 return (spa_vdev_exit(spa, NULL, txg, error)); 5693 5694 return (error); 5695} 5696 5697/* 5698 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5699 * currently spared, so we can detach it. 5700 */ 5701static vdev_t * 5702spa_vdev_resilver_done_hunt(vdev_t *vd) 5703{ 5704 vdev_t *newvd, *oldvd; 5705 5706 for (int c = 0; c < vd->vdev_children; c++) { 5707 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5708 if (oldvd != NULL) 5709 return (oldvd); 5710 } 5711 5712 /* 5713 * Check for a completed replacement. We always consider the first 5714 * vdev in the list to be the oldest vdev, and the last one to be 5715 * the newest (see spa_vdev_attach() for how that works). In 5716 * the case where the newest vdev is faulted, we will not automatically 5717 * remove it after a resilver completes. This is OK as it will require 5718 * user intervention to determine which disk the admin wishes to keep. 5719 */ 5720 if (vd->vdev_ops == &vdev_replacing_ops) { 5721 ASSERT(vd->vdev_children > 1); 5722 5723 newvd = vd->vdev_child[vd->vdev_children - 1]; 5724 oldvd = vd->vdev_child[0]; 5725 5726 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5727 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5728 !vdev_dtl_required(oldvd)) 5729 return (oldvd); 5730 } 5731 5732 /* 5733 * Check for a completed resilver with the 'unspare' flag set. 5734 */ 5735 if (vd->vdev_ops == &vdev_spare_ops) { 5736 vdev_t *first = vd->vdev_child[0]; 5737 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5738 5739 if (last->vdev_unspare) { 5740 oldvd = first; 5741 newvd = last; 5742 } else if (first->vdev_unspare) { 5743 oldvd = last; 5744 newvd = first; 5745 } else { 5746 oldvd = NULL; 5747 } 5748 5749 if (oldvd != NULL && 5750 vdev_dtl_empty(newvd, DTL_MISSING) && 5751 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5752 !vdev_dtl_required(oldvd)) 5753 return (oldvd); 5754 5755 /* 5756 * If there are more than two spares attached to a disk, 5757 * and those spares are not required, then we want to 5758 * attempt to free them up now so that they can be used 5759 * by other pools. Once we're back down to a single 5760 * disk+spare, we stop removing them. 5761 */ 5762 if (vd->vdev_children > 2) { 5763 newvd = vd->vdev_child[1]; 5764 5765 if (newvd->vdev_isspare && last->vdev_isspare && 5766 vdev_dtl_empty(last, DTL_MISSING) && 5767 vdev_dtl_empty(last, DTL_OUTAGE) && 5768 !vdev_dtl_required(newvd)) 5769 return (newvd); 5770 } 5771 } 5772 5773 return (NULL); 5774} 5775 5776static void 5777spa_vdev_resilver_done(spa_t *spa) 5778{ 5779 vdev_t *vd, *pvd, *ppvd; 5780 uint64_t guid, sguid, pguid, ppguid; 5781 5782 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5783 5784 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5785 pvd = vd->vdev_parent; 5786 ppvd = pvd->vdev_parent; 5787 guid = vd->vdev_guid; 5788 pguid = pvd->vdev_guid; 5789 ppguid = ppvd->vdev_guid; 5790 sguid = 0; 5791 /* 5792 * If we have just finished replacing a hot spared device, then 5793 * we need to detach the parent's first child (the original hot 5794 * spare) as well. 5795 */ 5796 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5797 ppvd->vdev_children == 2) { 5798 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5799 sguid = ppvd->vdev_child[1]->vdev_guid; 5800 } 5801 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5802 5803 spa_config_exit(spa, SCL_ALL, FTAG); 5804 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5805 return; 5806 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5807 return; 5808 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5809 } 5810 5811 spa_config_exit(spa, SCL_ALL, FTAG); 5812} 5813 5814/* 5815 * Update the stored path or FRU for this vdev. 5816 */ 5817int 5818spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5819 boolean_t ispath) 5820{ 5821 vdev_t *vd; 5822 boolean_t sync = B_FALSE; 5823 5824 ASSERT(spa_writeable(spa)); 5825 5826 spa_vdev_state_enter(spa, SCL_ALL); 5827 5828 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5829 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5830 5831 if (!vd->vdev_ops->vdev_op_leaf) 5832 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5833 5834 if (ispath) { 5835 if (strcmp(value, vd->vdev_path) != 0) { 5836 spa_strfree(vd->vdev_path); 5837 vd->vdev_path = spa_strdup(value); 5838 sync = B_TRUE; 5839 } 5840 } else { 5841 if (vd->vdev_fru == NULL) { 5842 vd->vdev_fru = spa_strdup(value); 5843 sync = B_TRUE; 5844 } else if (strcmp(value, vd->vdev_fru) != 0) { 5845 spa_strfree(vd->vdev_fru); 5846 vd->vdev_fru = spa_strdup(value); 5847 sync = B_TRUE; 5848 } 5849 } 5850 5851 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5852} 5853 5854int 5855spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5856{ 5857 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5858} 5859 5860int 5861spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5862{ 5863 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5864} 5865 5866/* 5867 * ========================================================================== 5868 * SPA Scanning 5869 * ========================================================================== 5870 */ 5871 5872int 5873spa_scan_stop(spa_t *spa) 5874{ 5875 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5876 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5877 return (SET_ERROR(EBUSY)); 5878 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5879} 5880 5881int 5882spa_scan(spa_t *spa, pool_scan_func_t func) 5883{ 5884 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5885 5886 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5887 return (SET_ERROR(ENOTSUP)); 5888 5889 /* 5890 * If a resilver was requested, but there is no DTL on a 5891 * writeable leaf device, we have nothing to do. 5892 */ 5893 if (func == POOL_SCAN_RESILVER && 5894 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5895 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5896 return (0); 5897 } 5898 5899 return (dsl_scan(spa->spa_dsl_pool, func)); 5900} 5901 5902/* 5903 * ========================================================================== 5904 * SPA async task processing 5905 * ========================================================================== 5906 */ 5907 5908static void 5909spa_async_remove(spa_t *spa, vdev_t *vd) 5910{ 5911 if (vd->vdev_remove_wanted) { 5912 vd->vdev_remove_wanted = B_FALSE; 5913 vd->vdev_delayed_close = B_FALSE; 5914 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5915 5916 /* 5917 * We want to clear the stats, but we don't want to do a full 5918 * vdev_clear() as that will cause us to throw away 5919 * degraded/faulted state as well as attempt to reopen the 5920 * device, all of which is a waste. 5921 */ 5922 vd->vdev_stat.vs_read_errors = 0; 5923 vd->vdev_stat.vs_write_errors = 0; 5924 vd->vdev_stat.vs_checksum_errors = 0; 5925 5926 vdev_state_dirty(vd->vdev_top); 5927 } 5928 5929 for (int c = 0; c < vd->vdev_children; c++) 5930 spa_async_remove(spa, vd->vdev_child[c]); 5931} 5932 5933static void 5934spa_async_probe(spa_t *spa, vdev_t *vd) 5935{ 5936 if (vd->vdev_probe_wanted) { 5937 vd->vdev_probe_wanted = B_FALSE; 5938 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5939 } 5940 5941 for (int c = 0; c < vd->vdev_children; c++) 5942 spa_async_probe(spa, vd->vdev_child[c]); 5943} 5944 5945static void 5946spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5947{ 5948 sysevent_id_t eid; 5949 nvlist_t *attr; 5950 char *physpath; 5951 5952 if (!spa->spa_autoexpand) 5953 return; 5954 5955 for (int c = 0; c < vd->vdev_children; c++) { 5956 vdev_t *cvd = vd->vdev_child[c]; 5957 spa_async_autoexpand(spa, cvd); 5958 } 5959 5960 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5961 return; 5962 5963 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5964 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5965 5966 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5967 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5968 5969 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5970 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5971 5972 nvlist_free(attr); 5973 kmem_free(physpath, MAXPATHLEN); 5974} 5975 5976static void 5977spa_async_thread(void *arg) 5978{ 5979 spa_t *spa = arg; 5980 int tasks; 5981 5982 ASSERT(spa->spa_sync_on); 5983 5984 mutex_enter(&spa->spa_async_lock); 5985 tasks = spa->spa_async_tasks; 5986 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5987 mutex_exit(&spa->spa_async_lock); 5988 5989 /* 5990 * See if the config needs to be updated. 5991 */ 5992 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5993 uint64_t old_space, new_space; 5994 5995 mutex_enter(&spa_namespace_lock); 5996 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5997 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5998 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5999 mutex_exit(&spa_namespace_lock); 6000 6001 /* 6002 * If the pool grew as a result of the config update, 6003 * then log an internal history event. 6004 */ 6005 if (new_space != old_space) { 6006 spa_history_log_internal(spa, "vdev online", NULL, 6007 "pool '%s' size: %llu(+%llu)", 6008 spa_name(spa), new_space, new_space - old_space); 6009 } 6010 } 6011 6012 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6013 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6014 spa_async_autoexpand(spa, spa->spa_root_vdev); 6015 spa_config_exit(spa, SCL_CONFIG, FTAG); 6016 } 6017 6018 /* 6019 * See if any devices need to be probed. 6020 */ 6021 if (tasks & SPA_ASYNC_PROBE) { 6022 spa_vdev_state_enter(spa, SCL_NONE); 6023 spa_async_probe(spa, spa->spa_root_vdev); 6024 (void) spa_vdev_state_exit(spa, NULL, 0); 6025 } 6026 6027 /* 6028 * If any devices are done replacing, detach them. 6029 */ 6030 if (tasks & SPA_ASYNC_RESILVER_DONE) 6031 spa_vdev_resilver_done(spa); 6032 6033 /* 6034 * Kick off a resilver. 6035 */ 6036 if (tasks & SPA_ASYNC_RESILVER) 6037 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6038 6039 /* 6040 * Let the world know that we're done. 6041 */ 6042 mutex_enter(&spa->spa_async_lock); 6043 spa->spa_async_thread = NULL; 6044 cv_broadcast(&spa->spa_async_cv); 6045 mutex_exit(&spa->spa_async_lock); 6046 thread_exit(); 6047} 6048 6049static void 6050spa_async_thread_vd(void *arg) 6051{ 6052 spa_t *spa = arg; 6053 int tasks; 6054 6055 ASSERT(spa->spa_sync_on); 6056 6057 mutex_enter(&spa->spa_async_lock); 6058 tasks = spa->spa_async_tasks; 6059retry: 6060 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6061 mutex_exit(&spa->spa_async_lock); 6062 6063 /* 6064 * See if any devices need to be marked REMOVED. 6065 */ 6066 if (tasks & SPA_ASYNC_REMOVE) { 6067 spa_vdev_state_enter(spa, SCL_NONE); 6068 spa_async_remove(spa, spa->spa_root_vdev); 6069 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6070 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6071 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6072 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6073 (void) spa_vdev_state_exit(spa, NULL, 0); 6074 } 6075 6076 /* 6077 * Let the world know that we're done. 6078 */ 6079 mutex_enter(&spa->spa_async_lock); 6080 tasks = spa->spa_async_tasks; 6081 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6082 goto retry; 6083 spa->spa_async_thread_vd = NULL; 6084 cv_broadcast(&spa->spa_async_cv); 6085 mutex_exit(&spa->spa_async_lock); 6086 thread_exit(); 6087} 6088 6089void 6090spa_async_suspend(spa_t *spa) 6091{ 6092 mutex_enter(&spa->spa_async_lock); 6093 spa->spa_async_suspended++; 6094 while (spa->spa_async_thread != NULL && 6095 spa->spa_async_thread_vd != NULL) 6096 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6097 mutex_exit(&spa->spa_async_lock); 6098} 6099 6100void 6101spa_async_resume(spa_t *spa) 6102{ 6103 mutex_enter(&spa->spa_async_lock); 6104 ASSERT(spa->spa_async_suspended != 0); 6105 spa->spa_async_suspended--; 6106 mutex_exit(&spa->spa_async_lock); 6107} 6108 6109static boolean_t 6110spa_async_tasks_pending(spa_t *spa) 6111{ 6112 uint_t non_config_tasks; 6113 uint_t config_task; 6114 boolean_t config_task_suspended; 6115 6116 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6117 SPA_ASYNC_REMOVE); 6118 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6119 if (spa->spa_ccw_fail_time == 0) { 6120 config_task_suspended = B_FALSE; 6121 } else { 6122 config_task_suspended = 6123 (gethrtime() - spa->spa_ccw_fail_time) < 6124 (zfs_ccw_retry_interval * NANOSEC); 6125 } 6126 6127 return (non_config_tasks || (config_task && !config_task_suspended)); 6128} 6129 6130static void 6131spa_async_dispatch(spa_t *spa) 6132{ 6133 mutex_enter(&spa->spa_async_lock); 6134 if (spa_async_tasks_pending(spa) && 6135 !spa->spa_async_suspended && 6136 spa->spa_async_thread == NULL && 6137 rootdir != NULL) 6138 spa->spa_async_thread = thread_create(NULL, 0, 6139 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6140 mutex_exit(&spa->spa_async_lock); 6141} 6142 6143static void 6144spa_async_dispatch_vd(spa_t *spa) 6145{ 6146 mutex_enter(&spa->spa_async_lock); 6147 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6148 !spa->spa_async_suspended && 6149 spa->spa_async_thread_vd == NULL && 6150 rootdir != NULL) 6151 spa->spa_async_thread_vd = thread_create(NULL, 0, 6152 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6153 mutex_exit(&spa->spa_async_lock); 6154} 6155 6156void 6157spa_async_request(spa_t *spa, int task) 6158{ 6159 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6160 mutex_enter(&spa->spa_async_lock); 6161 spa->spa_async_tasks |= task; 6162 mutex_exit(&spa->spa_async_lock); 6163 spa_async_dispatch_vd(spa); 6164} 6165 6166/* 6167 * ========================================================================== 6168 * SPA syncing routines 6169 * ========================================================================== 6170 */ 6171 6172static int 6173bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6174{ 6175 bpobj_t *bpo = arg; 6176 bpobj_enqueue(bpo, bp, tx); 6177 return (0); 6178} 6179 6180static int 6181spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6182{ 6183 zio_t *zio = arg; 6184 6185 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6186 BP_GET_PSIZE(bp), zio->io_flags)); 6187 return (0); 6188} 6189 6190/* 6191 * Note: this simple function is not inlined to make it easier to dtrace the 6192 * amount of time spent syncing frees. 6193 */ 6194static void 6195spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6196{ 6197 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6198 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6199 VERIFY(zio_wait(zio) == 0); 6200} 6201 6202/* 6203 * Note: this simple function is not inlined to make it easier to dtrace the 6204 * amount of time spent syncing deferred frees. 6205 */ 6206static void 6207spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6208{ 6209 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6210 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6211 spa_free_sync_cb, zio, tx), ==, 0); 6212 VERIFY0(zio_wait(zio)); 6213} 6214 6215 6216static void 6217spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6218{ 6219 char *packed = NULL; 6220 size_t bufsize; 6221 size_t nvsize = 0; 6222 dmu_buf_t *db; 6223 6224 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6225 6226 /* 6227 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6228 * information. This avoids the dmu_buf_will_dirty() path and 6229 * saves us a pre-read to get data we don't actually care about. 6230 */ 6231 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6232 packed = kmem_alloc(bufsize, KM_SLEEP); 6233 6234 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6235 KM_SLEEP) == 0); 6236 bzero(packed + nvsize, bufsize - nvsize); 6237 6238 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6239 6240 kmem_free(packed, bufsize); 6241 6242 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6243 dmu_buf_will_dirty(db, tx); 6244 *(uint64_t *)db->db_data = nvsize; 6245 dmu_buf_rele(db, FTAG); 6246} 6247 6248static void 6249spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6250 const char *config, const char *entry) 6251{ 6252 nvlist_t *nvroot; 6253 nvlist_t **list; 6254 int i; 6255 6256 if (!sav->sav_sync) 6257 return; 6258 6259 /* 6260 * Update the MOS nvlist describing the list of available devices. 6261 * spa_validate_aux() will have already made sure this nvlist is 6262 * valid and the vdevs are labeled appropriately. 6263 */ 6264 if (sav->sav_object == 0) { 6265 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6266 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6267 sizeof (uint64_t), tx); 6268 VERIFY(zap_update(spa->spa_meta_objset, 6269 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6270 &sav->sav_object, tx) == 0); 6271 } 6272 6273 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6274 if (sav->sav_count == 0) { 6275 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6276 } else { 6277 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6278 for (i = 0; i < sav->sav_count; i++) 6279 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6280 B_FALSE, VDEV_CONFIG_L2CACHE); 6281 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6282 sav->sav_count) == 0); 6283 for (i = 0; i < sav->sav_count; i++) 6284 nvlist_free(list[i]); 6285 kmem_free(list, sav->sav_count * sizeof (void *)); 6286 } 6287 6288 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6289 nvlist_free(nvroot); 6290 6291 sav->sav_sync = B_FALSE; 6292} 6293 6294static void 6295spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6296{ 6297 nvlist_t *config; 6298 6299 if (list_is_empty(&spa->spa_config_dirty_list)) 6300 return; 6301 6302 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6303 6304 config = spa_config_generate(spa, spa->spa_root_vdev, 6305 dmu_tx_get_txg(tx), B_FALSE); 6306 6307 /* 6308 * If we're upgrading the spa version then make sure that 6309 * the config object gets updated with the correct version. 6310 */ 6311 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6312 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6313 spa->spa_uberblock.ub_version); 6314 6315 spa_config_exit(spa, SCL_STATE, FTAG); 6316 6317 if (spa->spa_config_syncing) 6318 nvlist_free(spa->spa_config_syncing); 6319 spa->spa_config_syncing = config; 6320 6321 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6322} 6323 6324static void 6325spa_sync_version(void *arg, dmu_tx_t *tx) 6326{ 6327 uint64_t *versionp = arg; 6328 uint64_t version = *versionp; 6329 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6330 6331 /* 6332 * Setting the version is special cased when first creating the pool. 6333 */ 6334 ASSERT(tx->tx_txg != TXG_INITIAL); 6335 6336 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6337 ASSERT(version >= spa_version(spa)); 6338 6339 spa->spa_uberblock.ub_version = version; 6340 vdev_config_dirty(spa->spa_root_vdev); 6341 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6342} 6343 6344/* 6345 * Set zpool properties. 6346 */ 6347static void 6348spa_sync_props(void *arg, dmu_tx_t *tx) 6349{ 6350 nvlist_t *nvp = arg; 6351 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6352 objset_t *mos = spa->spa_meta_objset; 6353 nvpair_t *elem = NULL; 6354 6355 mutex_enter(&spa->spa_props_lock); 6356 6357 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6358 uint64_t intval; 6359 char *strval, *fname; 6360 zpool_prop_t prop; 6361 const char *propname; 6362 zprop_type_t proptype; 6363 spa_feature_t fid; 6364 6365 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6366 case ZPROP_INVAL: 6367 /* 6368 * We checked this earlier in spa_prop_validate(). 6369 */ 6370 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6371 6372 fname = strchr(nvpair_name(elem), '@') + 1; 6373 VERIFY0(zfeature_lookup_name(fname, &fid)); 6374 6375 spa_feature_enable(spa, fid, tx); 6376 spa_history_log_internal(spa, "set", tx, 6377 "%s=enabled", nvpair_name(elem)); 6378 break; 6379 6380 case ZPOOL_PROP_VERSION: 6381 intval = fnvpair_value_uint64(elem); 6382 /* 6383 * The version is synced seperatly before other 6384 * properties and should be correct by now. 6385 */ 6386 ASSERT3U(spa_version(spa), >=, intval); 6387 break; 6388 6389 case ZPOOL_PROP_ALTROOT: 6390 /* 6391 * 'altroot' is a non-persistent property. It should 6392 * have been set temporarily at creation or import time. 6393 */ 6394 ASSERT(spa->spa_root != NULL); 6395 break; 6396 6397 case ZPOOL_PROP_READONLY: 6398 case ZPOOL_PROP_CACHEFILE: 6399 /* 6400 * 'readonly' and 'cachefile' are also non-persisitent 6401 * properties. 6402 */ 6403 break; 6404 case ZPOOL_PROP_COMMENT: 6405 strval = fnvpair_value_string(elem); 6406 if (spa->spa_comment != NULL) 6407 spa_strfree(spa->spa_comment); 6408 spa->spa_comment = spa_strdup(strval); 6409 /* 6410 * We need to dirty the configuration on all the vdevs 6411 * so that their labels get updated. It's unnecessary 6412 * to do this for pool creation since the vdev's 6413 * configuratoin has already been dirtied. 6414 */ 6415 if (tx->tx_txg != TXG_INITIAL) 6416 vdev_config_dirty(spa->spa_root_vdev); 6417 spa_history_log_internal(spa, "set", tx, 6418 "%s=%s", nvpair_name(elem), strval); 6419 break; 6420 default: 6421 /* 6422 * Set pool property values in the poolprops mos object. 6423 */ 6424 if (spa->spa_pool_props_object == 0) { 6425 spa->spa_pool_props_object = 6426 zap_create_link(mos, DMU_OT_POOL_PROPS, 6427 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6428 tx); 6429 } 6430 6431 /* normalize the property name */ 6432 propname = zpool_prop_to_name(prop); 6433 proptype = zpool_prop_get_type(prop); 6434 6435 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6436 ASSERT(proptype == PROP_TYPE_STRING); 6437 strval = fnvpair_value_string(elem); 6438 VERIFY0(zap_update(mos, 6439 spa->spa_pool_props_object, propname, 6440 1, strlen(strval) + 1, strval, tx)); 6441 spa_history_log_internal(spa, "set", tx, 6442 "%s=%s", nvpair_name(elem), strval); 6443 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6444 intval = fnvpair_value_uint64(elem); 6445 6446 if (proptype == PROP_TYPE_INDEX) { 6447 const char *unused; 6448 VERIFY0(zpool_prop_index_to_string( 6449 prop, intval, &unused)); 6450 } 6451 VERIFY0(zap_update(mos, 6452 spa->spa_pool_props_object, propname, 6453 8, 1, &intval, tx)); 6454 spa_history_log_internal(spa, "set", tx, 6455 "%s=%lld", nvpair_name(elem), intval); 6456 } else { 6457 ASSERT(0); /* not allowed */ 6458 } 6459 6460 switch (prop) { 6461 case ZPOOL_PROP_DELEGATION: 6462 spa->spa_delegation = intval; 6463 break; 6464 case ZPOOL_PROP_BOOTFS: 6465 spa->spa_bootfs = intval; 6466 break; 6467 case ZPOOL_PROP_FAILUREMODE: 6468 spa->spa_failmode = intval; 6469 break; 6470 case ZPOOL_PROP_AUTOEXPAND: 6471 spa->spa_autoexpand = intval; 6472 if (tx->tx_txg != TXG_INITIAL) 6473 spa_async_request(spa, 6474 SPA_ASYNC_AUTOEXPAND); 6475 break; 6476 case ZPOOL_PROP_DEDUPDITTO: 6477 spa->spa_dedup_ditto = intval; 6478 break; 6479 default: 6480 break; 6481 } 6482 } 6483 6484 } 6485 6486 mutex_exit(&spa->spa_props_lock); 6487} 6488 6489/* 6490 * Perform one-time upgrade on-disk changes. spa_version() does not 6491 * reflect the new version this txg, so there must be no changes this 6492 * txg to anything that the upgrade code depends on after it executes. 6493 * Therefore this must be called after dsl_pool_sync() does the sync 6494 * tasks. 6495 */ 6496static void 6497spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6498{ 6499 dsl_pool_t *dp = spa->spa_dsl_pool; 6500 6501 ASSERT(spa->spa_sync_pass == 1); 6502 6503 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6504 6505 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6506 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6507 dsl_pool_create_origin(dp, tx); 6508 6509 /* Keeping the origin open increases spa_minref */ 6510 spa->spa_minref += 3; 6511 } 6512 6513 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6514 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6515 dsl_pool_upgrade_clones(dp, tx); 6516 } 6517 6518 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6519 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6520 dsl_pool_upgrade_dir_clones(dp, tx); 6521 6522 /* Keeping the freedir open increases spa_minref */ 6523 spa->spa_minref += 3; 6524 } 6525 6526 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6527 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6528 spa_feature_create_zap_objects(spa, tx); 6529 } 6530 6531 /* 6532 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6533 * when possibility to use lz4 compression for metadata was added 6534 * Old pools that have this feature enabled must be upgraded to have 6535 * this feature active 6536 */ 6537 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6538 boolean_t lz4_en = spa_feature_is_enabled(spa, 6539 SPA_FEATURE_LZ4_COMPRESS); 6540 boolean_t lz4_ac = spa_feature_is_active(spa, 6541 SPA_FEATURE_LZ4_COMPRESS); 6542 6543 if (lz4_en && !lz4_ac) 6544 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6545 } 6546 rrw_exit(&dp->dp_config_rwlock, FTAG); 6547} 6548 6549/* 6550 * Sync the specified transaction group. New blocks may be dirtied as 6551 * part of the process, so we iterate until it converges. 6552 */ 6553void 6554spa_sync(spa_t *spa, uint64_t txg) 6555{ 6556 dsl_pool_t *dp = spa->spa_dsl_pool; 6557 objset_t *mos = spa->spa_meta_objset; 6558 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6559 vdev_t *rvd = spa->spa_root_vdev; 6560 vdev_t *vd; 6561 dmu_tx_t *tx; 6562 int error; 6563 6564 VERIFY(spa_writeable(spa)); 6565 6566 /* 6567 * Lock out configuration changes. 6568 */ 6569 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6570 6571 spa->spa_syncing_txg = txg; 6572 spa->spa_sync_pass = 0; 6573 6574 /* 6575 * If there are any pending vdev state changes, convert them 6576 * into config changes that go out with this transaction group. 6577 */ 6578 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6579 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6580 /* 6581 * We need the write lock here because, for aux vdevs, 6582 * calling vdev_config_dirty() modifies sav_config. 6583 * This is ugly and will become unnecessary when we 6584 * eliminate the aux vdev wart by integrating all vdevs 6585 * into the root vdev tree. 6586 */ 6587 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6588 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6589 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6590 vdev_state_clean(vd); 6591 vdev_config_dirty(vd); 6592 } 6593 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6594 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6595 } 6596 spa_config_exit(spa, SCL_STATE, FTAG); 6597 6598 tx = dmu_tx_create_assigned(dp, txg); 6599 6600 spa->spa_sync_starttime = gethrtime(); 6601#ifdef illumos 6602 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6603 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6604#else /* FreeBSD */ 6605#ifdef _KERNEL 6606 callout_reset(&spa->spa_deadman_cycid, 6607 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6608#endif 6609#endif 6610 6611 /* 6612 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6613 * set spa_deflate if we have no raid-z vdevs. 6614 */ 6615 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6616 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6617 int i; 6618 6619 for (i = 0; i < rvd->vdev_children; i++) { 6620 vd = rvd->vdev_child[i]; 6621 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6622 break; 6623 } 6624 if (i == rvd->vdev_children) { 6625 spa->spa_deflate = TRUE; 6626 VERIFY(0 == zap_add(spa->spa_meta_objset, 6627 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6628 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6629 } 6630 } 6631 6632 /* 6633 * Iterate to convergence. 6634 */ 6635 do { 6636 int pass = ++spa->spa_sync_pass; 6637 6638 spa_sync_config_object(spa, tx); 6639 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6640 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6641 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6642 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6643 spa_errlog_sync(spa, txg); 6644 dsl_pool_sync(dp, txg); 6645 6646 if (pass < zfs_sync_pass_deferred_free) { 6647 spa_sync_frees(spa, free_bpl, tx); 6648 } else { 6649 /* 6650 * We can not defer frees in pass 1, because 6651 * we sync the deferred frees later in pass 1. 6652 */ 6653 ASSERT3U(pass, >, 1); 6654 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6655 &spa->spa_deferred_bpobj, tx); 6656 } 6657 6658 ddt_sync(spa, txg); 6659 dsl_scan_sync(dp, tx); 6660 6661 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6662 vdev_sync(vd, txg); 6663 6664 if (pass == 1) { 6665 spa_sync_upgrades(spa, tx); 6666 ASSERT3U(txg, >=, 6667 spa->spa_uberblock.ub_rootbp.blk_birth); 6668 /* 6669 * Note: We need to check if the MOS is dirty 6670 * because we could have marked the MOS dirty 6671 * without updating the uberblock (e.g. if we 6672 * have sync tasks but no dirty user data). We 6673 * need to check the uberblock's rootbp because 6674 * it is updated if we have synced out dirty 6675 * data (though in this case the MOS will most 6676 * likely also be dirty due to second order 6677 * effects, we don't want to rely on that here). 6678 */ 6679 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6680 !dmu_objset_is_dirty(mos, txg)) { 6681 /* 6682 * Nothing changed on the first pass, 6683 * therefore this TXG is a no-op. Avoid 6684 * syncing deferred frees, so that we 6685 * can keep this TXG as a no-op. 6686 */ 6687 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6688 txg)); 6689 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6690 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6691 break; 6692 } 6693 spa_sync_deferred_frees(spa, tx); 6694 } 6695 6696 } while (dmu_objset_is_dirty(mos, txg)); 6697 6698 /* 6699 * Rewrite the vdev configuration (which includes the uberblock) 6700 * to commit the transaction group. 6701 * 6702 * If there are no dirty vdevs, we sync the uberblock to a few 6703 * random top-level vdevs that are known to be visible in the 6704 * config cache (see spa_vdev_add() for a complete description). 6705 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6706 */ 6707 for (;;) { 6708 /* 6709 * We hold SCL_STATE to prevent vdev open/close/etc. 6710 * while we're attempting to write the vdev labels. 6711 */ 6712 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6713 6714 if (list_is_empty(&spa->spa_config_dirty_list)) { 6715 vdev_t *svd[SPA_DVAS_PER_BP]; 6716 int svdcount = 0; 6717 int children = rvd->vdev_children; 6718 int c0 = spa_get_random(children); 6719 6720 for (int c = 0; c < children; c++) { 6721 vd = rvd->vdev_child[(c0 + c) % children]; 6722 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6723 continue; 6724 svd[svdcount++] = vd; 6725 if (svdcount == SPA_DVAS_PER_BP) 6726 break; 6727 } 6728 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6729 if (error != 0) 6730 error = vdev_config_sync(svd, svdcount, txg, 6731 B_TRUE); 6732 } else { 6733 error = vdev_config_sync(rvd->vdev_child, 6734 rvd->vdev_children, txg, B_FALSE); 6735 if (error != 0) 6736 error = vdev_config_sync(rvd->vdev_child, 6737 rvd->vdev_children, txg, B_TRUE); 6738 } 6739 6740 if (error == 0) 6741 spa->spa_last_synced_guid = rvd->vdev_guid; 6742 6743 spa_config_exit(spa, SCL_STATE, FTAG); 6744 6745 if (error == 0) 6746 break; 6747 zio_suspend(spa, NULL); 6748 zio_resume_wait(spa); 6749 } 6750 dmu_tx_commit(tx); 6751 6752#ifdef illumos 6753 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6754#else /* FreeBSD */ 6755#ifdef _KERNEL 6756 callout_drain(&spa->spa_deadman_cycid); 6757#endif 6758#endif 6759 6760 /* 6761 * Clear the dirty config list. 6762 */ 6763 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6764 vdev_config_clean(vd); 6765 6766 /* 6767 * Now that the new config has synced transactionally, 6768 * let it become visible to the config cache. 6769 */ 6770 if (spa->spa_config_syncing != NULL) { 6771 spa_config_set(spa, spa->spa_config_syncing); 6772 spa->spa_config_txg = txg; 6773 spa->spa_config_syncing = NULL; 6774 } 6775 6776 spa->spa_ubsync = spa->spa_uberblock; 6777 6778 dsl_pool_sync_done(dp, txg); 6779 6780 /* 6781 * Update usable space statistics. 6782 */ 6783 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6784 vdev_sync_done(vd, txg); 6785 6786 spa_update_dspace(spa); 6787 6788 /* 6789 * It had better be the case that we didn't dirty anything 6790 * since vdev_config_sync(). 6791 */ 6792 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6793 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6794 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6795 6796 spa->spa_sync_pass = 0; 6797 6798 spa_config_exit(spa, SCL_CONFIG, FTAG); 6799 6800 spa_handle_ignored_writes(spa); 6801 6802 /* 6803 * If any async tasks have been requested, kick them off. 6804 */ 6805 spa_async_dispatch(spa); 6806 spa_async_dispatch_vd(spa); 6807} 6808 6809/* 6810 * Sync all pools. We don't want to hold the namespace lock across these 6811 * operations, so we take a reference on the spa_t and drop the lock during the 6812 * sync. 6813 */ 6814void 6815spa_sync_allpools(void) 6816{ 6817 spa_t *spa = NULL; 6818 mutex_enter(&spa_namespace_lock); 6819 while ((spa = spa_next(spa)) != NULL) { 6820 if (spa_state(spa) != POOL_STATE_ACTIVE || 6821 !spa_writeable(spa) || spa_suspended(spa)) 6822 continue; 6823 spa_open_ref(spa, FTAG); 6824 mutex_exit(&spa_namespace_lock); 6825 txg_wait_synced(spa_get_dsl(spa), 0); 6826 mutex_enter(&spa_namespace_lock); 6827 spa_close(spa, FTAG); 6828 } 6829 mutex_exit(&spa_namespace_lock); 6830} 6831 6832/* 6833 * ========================================================================== 6834 * Miscellaneous routines 6835 * ========================================================================== 6836 */ 6837 6838/* 6839 * Remove all pools in the system. 6840 */ 6841void 6842spa_evict_all(void) 6843{ 6844 spa_t *spa; 6845 6846 /* 6847 * Remove all cached state. All pools should be closed now, 6848 * so every spa in the AVL tree should be unreferenced. 6849 */ 6850 mutex_enter(&spa_namespace_lock); 6851 while ((spa = spa_next(NULL)) != NULL) { 6852 /* 6853 * Stop async tasks. The async thread may need to detach 6854 * a device that's been replaced, which requires grabbing 6855 * spa_namespace_lock, so we must drop it here. 6856 */ 6857 spa_open_ref(spa, FTAG); 6858 mutex_exit(&spa_namespace_lock); 6859 spa_async_suspend(spa); 6860 mutex_enter(&spa_namespace_lock); 6861 spa_close(spa, FTAG); 6862 6863 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6864 spa_unload(spa); 6865 spa_deactivate(spa); 6866 } 6867 spa_remove(spa); 6868 } 6869 mutex_exit(&spa_namespace_lock); 6870} 6871 6872vdev_t * 6873spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6874{ 6875 vdev_t *vd; 6876 int i; 6877 6878 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6879 return (vd); 6880 6881 if (aux) { 6882 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6883 vd = spa->spa_l2cache.sav_vdevs[i]; 6884 if (vd->vdev_guid == guid) 6885 return (vd); 6886 } 6887 6888 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6889 vd = spa->spa_spares.sav_vdevs[i]; 6890 if (vd->vdev_guid == guid) 6891 return (vd); 6892 } 6893 } 6894 6895 return (NULL); 6896} 6897 6898void 6899spa_upgrade(spa_t *spa, uint64_t version) 6900{ 6901 ASSERT(spa_writeable(spa)); 6902 6903 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6904 6905 /* 6906 * This should only be called for a non-faulted pool, and since a 6907 * future version would result in an unopenable pool, this shouldn't be 6908 * possible. 6909 */ 6910 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6911 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6912 6913 spa->spa_uberblock.ub_version = version; 6914 vdev_config_dirty(spa->spa_root_vdev); 6915 6916 spa_config_exit(spa, SCL_ALL, FTAG); 6917 6918 txg_wait_synced(spa_get_dsl(spa), 0); 6919} 6920 6921boolean_t 6922spa_has_spare(spa_t *spa, uint64_t guid) 6923{ 6924 int i; 6925 uint64_t spareguid; 6926 spa_aux_vdev_t *sav = &spa->spa_spares; 6927 6928 for (i = 0; i < sav->sav_count; i++) 6929 if (sav->sav_vdevs[i]->vdev_guid == guid) 6930 return (B_TRUE); 6931 6932 for (i = 0; i < sav->sav_npending; i++) { 6933 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6934 &spareguid) == 0 && spareguid == guid) 6935 return (B_TRUE); 6936 } 6937 6938 return (B_FALSE); 6939} 6940 6941/* 6942 * Check if a pool has an active shared spare device. 6943 * Note: reference count of an active spare is 2, as a spare and as a replace 6944 */ 6945static boolean_t 6946spa_has_active_shared_spare(spa_t *spa) 6947{ 6948 int i, refcnt; 6949 uint64_t pool; 6950 spa_aux_vdev_t *sav = &spa->spa_spares; 6951 6952 for (i = 0; i < sav->sav_count; i++) { 6953 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6954 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6955 refcnt > 2) 6956 return (B_TRUE); 6957 } 6958 6959 return (B_FALSE); 6960} 6961 6962/* 6963 * Post a sysevent corresponding to the given event. The 'name' must be one of 6964 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6965 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6966 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6967 * or zdb as real changes. 6968 */ 6969void 6970spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6971{ 6972#ifdef _KERNEL 6973 sysevent_t *ev; 6974 sysevent_attr_list_t *attr = NULL; 6975 sysevent_value_t value; 6976 sysevent_id_t eid; 6977 6978 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6979 SE_SLEEP); 6980 6981 value.value_type = SE_DATA_TYPE_STRING; 6982 value.value.sv_string = spa_name(spa); 6983 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6984 goto done; 6985 6986 value.value_type = SE_DATA_TYPE_UINT64; 6987 value.value.sv_uint64 = spa_guid(spa); 6988 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6989 goto done; 6990 6991 if (vd) { 6992 value.value_type = SE_DATA_TYPE_UINT64; 6993 value.value.sv_uint64 = vd->vdev_guid; 6994 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6995 SE_SLEEP) != 0) 6996 goto done; 6997 6998 if (vd->vdev_path) { 6999 value.value_type = SE_DATA_TYPE_STRING; 7000 value.value.sv_string = vd->vdev_path; 7001 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7002 &value, SE_SLEEP) != 0) 7003 goto done; 7004 } 7005 } 7006 7007 if (sysevent_attach_attributes(ev, attr) != 0) 7008 goto done; 7009 attr = NULL; 7010 7011 (void) log_sysevent(ev, SE_SLEEP, &eid); 7012 7013done: 7014 if (attr) 7015 sysevent_free_attr(attr); 7016 sysevent_free(ev); 7017#endif 7018} 7019