spa.c revision 275781
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 */ 28 29/* 30 * SPA: Storage Pool Allocator 31 * 32 * This file contains all the routines used when modifying on-disk SPA state. 33 * This includes opening, importing, destroying, exporting a pool, and syncing a 34 * pool. 35 */ 36 37#include <sys/zfs_context.h> 38#include <sys/fm/fs/zfs.h> 39#include <sys/spa_impl.h> 40#include <sys/zio.h> 41#include <sys/zio_checksum.h> 42#include <sys/dmu.h> 43#include <sys/dmu_tx.h> 44#include <sys/zap.h> 45#include <sys/zil.h> 46#include <sys/ddt.h> 47#include <sys/vdev_impl.h> 48#include <sys/metaslab.h> 49#include <sys/metaslab_impl.h> 50#include <sys/uberblock_impl.h> 51#include <sys/txg.h> 52#include <sys/avl.h> 53#include <sys/dmu_traverse.h> 54#include <sys/dmu_objset.h> 55#include <sys/unique.h> 56#include <sys/dsl_pool.h> 57#include <sys/dsl_dataset.h> 58#include <sys/dsl_dir.h> 59#include <sys/dsl_prop.h> 60#include <sys/dsl_synctask.h> 61#include <sys/fs/zfs.h> 62#include <sys/arc.h> 63#include <sys/callb.h> 64#include <sys/spa_boot.h> 65#include <sys/zfs_ioctl.h> 66#include <sys/dsl_scan.h> 67#include <sys/dmu_send.h> 68#include <sys/dsl_destroy.h> 69#include <sys/dsl_userhold.h> 70#include <sys/zfeature.h> 71#include <sys/zvol.h> 72#include <sys/trim_map.h> 73 74#ifdef _KERNEL 75#include <sys/callb.h> 76#include <sys/cpupart.h> 77#include <sys/zone.h> 78#endif /* _KERNEL */ 79 80#include "zfs_prop.h" 81#include "zfs_comutil.h" 82 83/* Check hostid on import? */ 84static int check_hostid = 1; 85 86/* 87 * The interval, in seconds, at which failed configuration cache file writes 88 * should be retried. 89 */ 90static int zfs_ccw_retry_interval = 300; 91 92SYSCTL_DECL(_vfs_zfs); 93SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 94 "Check hostid on import?"); 95TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 96SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 97 &zfs_ccw_retry_interval, 0, 98 "Configuration cache file write, retry after failure, interval (seconds)"); 99 100typedef enum zti_modes { 101 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 102 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 103 ZTI_MODE_NULL, /* don't create a taskq */ 104 ZTI_NMODES 105} zti_modes_t; 106 107#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 108#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 109#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 110 111#define ZTI_N(n) ZTI_P(n, 1) 112#define ZTI_ONE ZTI_N(1) 113 114typedef struct zio_taskq_info { 115 zti_modes_t zti_mode; 116 uint_t zti_value; 117 uint_t zti_count; 118} zio_taskq_info_t; 119 120static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 121 "issue", "issue_high", "intr", "intr_high" 122}; 123 124/* 125 * This table defines the taskq settings for each ZFS I/O type. When 126 * initializing a pool, we use this table to create an appropriately sized 127 * taskq. Some operations are low volume and therefore have a small, static 128 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 129 * macros. Other operations process a large amount of data; the ZTI_BATCH 130 * macro causes us to create a taskq oriented for throughput. Some operations 131 * are so high frequency and short-lived that the taskq itself can become a a 132 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 133 * additional degree of parallelism specified by the number of threads per- 134 * taskq and the number of taskqs; when dispatching an event in this case, the 135 * particular taskq is chosen at random. 136 * 137 * The different taskq priorities are to handle the different contexts (issue 138 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 139 * need to be handled with minimum delay. 140 */ 141const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 142 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 143 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 144 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 145 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 146 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 149}; 150 151static void spa_sync_version(void *arg, dmu_tx_t *tx); 152static void spa_sync_props(void *arg, dmu_tx_t *tx); 153static boolean_t spa_has_active_shared_spare(spa_t *spa); 154static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 155 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 156 char **ereport); 157static void spa_vdev_resilver_done(spa_t *spa); 158 159uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 160#ifdef PSRSET_BIND 161id_t zio_taskq_psrset_bind = PS_NONE; 162#endif 163#ifdef SYSDC 164boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 165#endif 166uint_t zio_taskq_basedc = 80; /* base duty cycle */ 167 168boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 169extern int zfs_sync_pass_deferred_free; 170 171#ifndef illumos 172extern void spa_deadman(void *arg); 173#endif 174 175/* 176 * This (illegal) pool name is used when temporarily importing a spa_t in order 177 * to get the vdev stats associated with the imported devices. 178 */ 179#define TRYIMPORT_NAME "$import" 180 181/* 182 * ========================================================================== 183 * SPA properties routines 184 * ========================================================================== 185 */ 186 187/* 188 * Add a (source=src, propname=propval) list to an nvlist. 189 */ 190static void 191spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 192 uint64_t intval, zprop_source_t src) 193{ 194 const char *propname = zpool_prop_to_name(prop); 195 nvlist_t *propval; 196 197 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 198 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 199 200 if (strval != NULL) 201 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 202 else 203 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 204 205 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 206 nvlist_free(propval); 207} 208 209/* 210 * Get property values from the spa configuration. 211 */ 212static void 213spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 214{ 215 vdev_t *rvd = spa->spa_root_vdev; 216 dsl_pool_t *pool = spa->spa_dsl_pool; 217 uint64_t size, alloc, cap, version; 218 zprop_source_t src = ZPROP_SRC_NONE; 219 spa_config_dirent_t *dp; 220 metaslab_class_t *mc = spa_normal_class(spa); 221 222 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 223 224 if (rvd != NULL) { 225 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 226 size = metaslab_class_get_space(spa_normal_class(spa)); 227 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 228 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 229 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 230 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 231 size - alloc, src); 232 233 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 234 metaslab_class_fragmentation(mc), src); 235 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 236 metaslab_class_expandable_space(mc), src); 237 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 238 (spa_mode(spa) == FREAD), src); 239 240 cap = (size == 0) ? 0 : (alloc * 100 / size); 241 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 242 243 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 244 ddt_get_pool_dedup_ratio(spa), src); 245 246 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 247 rvd->vdev_state, src); 248 249 version = spa_version(spa); 250 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 251 src = ZPROP_SRC_DEFAULT; 252 else 253 src = ZPROP_SRC_LOCAL; 254 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 255 } 256 257 if (pool != NULL) { 258 /* 259 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 260 * when opening pools before this version freedir will be NULL. 261 */ 262 if (pool->dp_free_dir != NULL) { 263 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 264 pool->dp_free_dir->dd_phys->dd_used_bytes, src); 265 } else { 266 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 267 NULL, 0, src); 268 } 269 270 if (pool->dp_leak_dir != NULL) { 271 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 272 pool->dp_leak_dir->dd_phys->dd_used_bytes, 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_state_dirty_list, sizeof (vdev_t), 1132 offsetof(vdev_t, vdev_state_dirty_node)); 1133 1134 txg_list_create(&spa->spa_vdev_txg_list, 1135 offsetof(struct vdev, vdev_txg_node)); 1136 1137 avl_create(&spa->spa_errlist_scrub, 1138 spa_error_entry_compare, sizeof (spa_error_entry_t), 1139 offsetof(spa_error_entry_t, se_avl)); 1140 avl_create(&spa->spa_errlist_last, 1141 spa_error_entry_compare, sizeof (spa_error_entry_t), 1142 offsetof(spa_error_entry_t, se_avl)); 1143} 1144 1145/* 1146 * Opposite of spa_activate(). 1147 */ 1148static void 1149spa_deactivate(spa_t *spa) 1150{ 1151 ASSERT(spa->spa_sync_on == B_FALSE); 1152 ASSERT(spa->spa_dsl_pool == NULL); 1153 ASSERT(spa->spa_root_vdev == NULL); 1154 ASSERT(spa->spa_async_zio_root == NULL); 1155 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1156 1157 /* 1158 * Stop TRIM thread in case spa_unload() wasn't called directly 1159 * before spa_deactivate(). 1160 */ 1161 trim_thread_destroy(spa); 1162 1163 txg_list_destroy(&spa->spa_vdev_txg_list); 1164 1165 list_destroy(&spa->spa_config_dirty_list); 1166 list_destroy(&spa->spa_state_dirty_list); 1167 1168 for (int t = 0; t < ZIO_TYPES; t++) { 1169 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1170 spa_taskqs_fini(spa, t, q); 1171 } 1172 } 1173 1174 metaslab_class_destroy(spa->spa_normal_class); 1175 spa->spa_normal_class = NULL; 1176 1177 metaslab_class_destroy(spa->spa_log_class); 1178 spa->spa_log_class = NULL; 1179 1180 /* 1181 * If this was part of an import or the open otherwise failed, we may 1182 * still have errors left in the queues. Empty them just in case. 1183 */ 1184 spa_errlog_drain(spa); 1185 1186 avl_destroy(&spa->spa_errlist_scrub); 1187 avl_destroy(&spa->spa_errlist_last); 1188 1189 spa->spa_state = POOL_STATE_UNINITIALIZED; 1190 1191 mutex_enter(&spa->spa_proc_lock); 1192 if (spa->spa_proc_state != SPA_PROC_NONE) { 1193 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1194 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1195 cv_broadcast(&spa->spa_proc_cv); 1196 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1197 ASSERT(spa->spa_proc != &p0); 1198 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1199 } 1200 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1201 spa->spa_proc_state = SPA_PROC_NONE; 1202 } 1203 ASSERT(spa->spa_proc == &p0); 1204 mutex_exit(&spa->spa_proc_lock); 1205 1206#ifdef SPA_PROCESS 1207 /* 1208 * We want to make sure spa_thread() has actually exited the ZFS 1209 * module, so that the module can't be unloaded out from underneath 1210 * it. 1211 */ 1212 if (spa->spa_did != 0) { 1213 thread_join(spa->spa_did); 1214 spa->spa_did = 0; 1215 } 1216#endif /* SPA_PROCESS */ 1217} 1218 1219/* 1220 * Verify a pool configuration, and construct the vdev tree appropriately. This 1221 * will create all the necessary vdevs in the appropriate layout, with each vdev 1222 * in the CLOSED state. This will prep the pool before open/creation/import. 1223 * All vdev validation is done by the vdev_alloc() routine. 1224 */ 1225static int 1226spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1227 uint_t id, int atype) 1228{ 1229 nvlist_t **child; 1230 uint_t children; 1231 int error; 1232 1233 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1234 return (error); 1235 1236 if ((*vdp)->vdev_ops->vdev_op_leaf) 1237 return (0); 1238 1239 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1240 &child, &children); 1241 1242 if (error == ENOENT) 1243 return (0); 1244 1245 if (error) { 1246 vdev_free(*vdp); 1247 *vdp = NULL; 1248 return (SET_ERROR(EINVAL)); 1249 } 1250 1251 for (int c = 0; c < children; c++) { 1252 vdev_t *vd; 1253 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1254 atype)) != 0) { 1255 vdev_free(*vdp); 1256 *vdp = NULL; 1257 return (error); 1258 } 1259 } 1260 1261 ASSERT(*vdp != NULL); 1262 1263 return (0); 1264} 1265 1266/* 1267 * Opposite of spa_load(). 1268 */ 1269static void 1270spa_unload(spa_t *spa) 1271{ 1272 int i; 1273 1274 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1275 1276 /* 1277 * Stop TRIM thread. 1278 */ 1279 trim_thread_destroy(spa); 1280 1281 /* 1282 * Stop async tasks. 1283 */ 1284 spa_async_suspend(spa); 1285 1286 /* 1287 * Stop syncing. 1288 */ 1289 if (spa->spa_sync_on) { 1290 txg_sync_stop(spa->spa_dsl_pool); 1291 spa->spa_sync_on = B_FALSE; 1292 } 1293 1294 /* 1295 * Wait for any outstanding async I/O to complete. 1296 */ 1297 if (spa->spa_async_zio_root != NULL) { 1298 for (int i = 0; i < max_ncpus; i++) 1299 (void) zio_wait(spa->spa_async_zio_root[i]); 1300 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1301 spa->spa_async_zio_root = NULL; 1302 } 1303 1304 bpobj_close(&spa->spa_deferred_bpobj); 1305 1306 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1307 1308 /* 1309 * Close all vdevs. 1310 */ 1311 if (spa->spa_root_vdev) 1312 vdev_free(spa->spa_root_vdev); 1313 ASSERT(spa->spa_root_vdev == NULL); 1314 1315 /* 1316 * Close the dsl pool. 1317 */ 1318 if (spa->spa_dsl_pool) { 1319 dsl_pool_close(spa->spa_dsl_pool); 1320 spa->spa_dsl_pool = NULL; 1321 spa->spa_meta_objset = NULL; 1322 } 1323 1324 ddt_unload(spa); 1325 1326 1327 /* 1328 * Drop and purge level 2 cache 1329 */ 1330 spa_l2cache_drop(spa); 1331 1332 for (i = 0; i < spa->spa_spares.sav_count; i++) 1333 vdev_free(spa->spa_spares.sav_vdevs[i]); 1334 if (spa->spa_spares.sav_vdevs) { 1335 kmem_free(spa->spa_spares.sav_vdevs, 1336 spa->spa_spares.sav_count * sizeof (void *)); 1337 spa->spa_spares.sav_vdevs = NULL; 1338 } 1339 if (spa->spa_spares.sav_config) { 1340 nvlist_free(spa->spa_spares.sav_config); 1341 spa->spa_spares.sav_config = NULL; 1342 } 1343 spa->spa_spares.sav_count = 0; 1344 1345 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1346 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1347 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1348 } 1349 if (spa->spa_l2cache.sav_vdevs) { 1350 kmem_free(spa->spa_l2cache.sav_vdevs, 1351 spa->spa_l2cache.sav_count * sizeof (void *)); 1352 spa->spa_l2cache.sav_vdevs = NULL; 1353 } 1354 if (spa->spa_l2cache.sav_config) { 1355 nvlist_free(spa->spa_l2cache.sav_config); 1356 spa->spa_l2cache.sav_config = NULL; 1357 } 1358 spa->spa_l2cache.sav_count = 0; 1359 1360 spa->spa_async_suspended = 0; 1361 1362 if (spa->spa_comment != NULL) { 1363 spa_strfree(spa->spa_comment); 1364 spa->spa_comment = NULL; 1365 } 1366 1367 spa_config_exit(spa, SCL_ALL, FTAG); 1368} 1369 1370/* 1371 * Load (or re-load) the current list of vdevs describing the active spares for 1372 * this pool. When this is called, we have some form of basic information in 1373 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1374 * then re-generate a more complete list including status information. 1375 */ 1376static void 1377spa_load_spares(spa_t *spa) 1378{ 1379 nvlist_t **spares; 1380 uint_t nspares; 1381 int i; 1382 vdev_t *vd, *tvd; 1383 1384 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1385 1386 /* 1387 * First, close and free any existing spare vdevs. 1388 */ 1389 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1390 vd = spa->spa_spares.sav_vdevs[i]; 1391 1392 /* Undo the call to spa_activate() below */ 1393 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1394 B_FALSE)) != NULL && tvd->vdev_isspare) 1395 spa_spare_remove(tvd); 1396 vdev_close(vd); 1397 vdev_free(vd); 1398 } 1399 1400 if (spa->spa_spares.sav_vdevs) 1401 kmem_free(spa->spa_spares.sav_vdevs, 1402 spa->spa_spares.sav_count * sizeof (void *)); 1403 1404 if (spa->spa_spares.sav_config == NULL) 1405 nspares = 0; 1406 else 1407 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1408 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1409 1410 spa->spa_spares.sav_count = (int)nspares; 1411 spa->spa_spares.sav_vdevs = NULL; 1412 1413 if (nspares == 0) 1414 return; 1415 1416 /* 1417 * Construct the array of vdevs, opening them to get status in the 1418 * process. For each spare, there is potentially two different vdev_t 1419 * structures associated with it: one in the list of spares (used only 1420 * for basic validation purposes) and one in the active vdev 1421 * configuration (if it's spared in). During this phase we open and 1422 * validate each vdev on the spare list. If the vdev also exists in the 1423 * active configuration, then we also mark this vdev as an active spare. 1424 */ 1425 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1426 KM_SLEEP); 1427 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1428 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1429 VDEV_ALLOC_SPARE) == 0); 1430 ASSERT(vd != NULL); 1431 1432 spa->spa_spares.sav_vdevs[i] = vd; 1433 1434 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1435 B_FALSE)) != NULL) { 1436 if (!tvd->vdev_isspare) 1437 spa_spare_add(tvd); 1438 1439 /* 1440 * We only mark the spare active if we were successfully 1441 * able to load the vdev. Otherwise, importing a pool 1442 * with a bad active spare would result in strange 1443 * behavior, because multiple pool would think the spare 1444 * is actively in use. 1445 * 1446 * There is a vulnerability here to an equally bizarre 1447 * circumstance, where a dead active spare is later 1448 * brought back to life (onlined or otherwise). Given 1449 * the rarity of this scenario, and the extra complexity 1450 * it adds, we ignore the possibility. 1451 */ 1452 if (!vdev_is_dead(tvd)) 1453 spa_spare_activate(tvd); 1454 } 1455 1456 vd->vdev_top = vd; 1457 vd->vdev_aux = &spa->spa_spares; 1458 1459 if (vdev_open(vd) != 0) 1460 continue; 1461 1462 if (vdev_validate_aux(vd) == 0) 1463 spa_spare_add(vd); 1464 } 1465 1466 /* 1467 * Recompute the stashed list of spares, with status information 1468 * this time. 1469 */ 1470 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1471 DATA_TYPE_NVLIST_ARRAY) == 0); 1472 1473 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1474 KM_SLEEP); 1475 for (i = 0; i < spa->spa_spares.sav_count; i++) 1476 spares[i] = vdev_config_generate(spa, 1477 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1478 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1479 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1480 for (i = 0; i < spa->spa_spares.sav_count; i++) 1481 nvlist_free(spares[i]); 1482 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1483} 1484 1485/* 1486 * Load (or re-load) the current list of vdevs describing the active l2cache for 1487 * this pool. When this is called, we have some form of basic information in 1488 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1489 * then re-generate a more complete list including status information. 1490 * Devices which are already active have their details maintained, and are 1491 * not re-opened. 1492 */ 1493static void 1494spa_load_l2cache(spa_t *spa) 1495{ 1496 nvlist_t **l2cache; 1497 uint_t nl2cache; 1498 int i, j, oldnvdevs; 1499 uint64_t guid; 1500 vdev_t *vd, **oldvdevs, **newvdevs; 1501 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1502 1503 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1504 1505 if (sav->sav_config != NULL) { 1506 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1507 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1508 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1509 } else { 1510 nl2cache = 0; 1511 newvdevs = NULL; 1512 } 1513 1514 oldvdevs = sav->sav_vdevs; 1515 oldnvdevs = sav->sav_count; 1516 sav->sav_vdevs = NULL; 1517 sav->sav_count = 0; 1518 1519 /* 1520 * Process new nvlist of vdevs. 1521 */ 1522 for (i = 0; i < nl2cache; i++) { 1523 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1524 &guid) == 0); 1525 1526 newvdevs[i] = NULL; 1527 for (j = 0; j < oldnvdevs; j++) { 1528 vd = oldvdevs[j]; 1529 if (vd != NULL && guid == vd->vdev_guid) { 1530 /* 1531 * Retain previous vdev for add/remove ops. 1532 */ 1533 newvdevs[i] = vd; 1534 oldvdevs[j] = NULL; 1535 break; 1536 } 1537 } 1538 1539 if (newvdevs[i] == NULL) { 1540 /* 1541 * Create new vdev 1542 */ 1543 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1544 VDEV_ALLOC_L2CACHE) == 0); 1545 ASSERT(vd != NULL); 1546 newvdevs[i] = vd; 1547 1548 /* 1549 * Commit this vdev as an l2cache device, 1550 * even if it fails to open. 1551 */ 1552 spa_l2cache_add(vd); 1553 1554 vd->vdev_top = vd; 1555 vd->vdev_aux = sav; 1556 1557 spa_l2cache_activate(vd); 1558 1559 if (vdev_open(vd) != 0) 1560 continue; 1561 1562 (void) vdev_validate_aux(vd); 1563 1564 if (!vdev_is_dead(vd)) 1565 l2arc_add_vdev(spa, vd); 1566 } 1567 } 1568 1569 /* 1570 * Purge vdevs that were dropped 1571 */ 1572 for (i = 0; i < oldnvdevs; i++) { 1573 uint64_t pool; 1574 1575 vd = oldvdevs[i]; 1576 if (vd != NULL) { 1577 ASSERT(vd->vdev_isl2cache); 1578 1579 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1580 pool != 0ULL && l2arc_vdev_present(vd)) 1581 l2arc_remove_vdev(vd); 1582 vdev_clear_stats(vd); 1583 vdev_free(vd); 1584 } 1585 } 1586 1587 if (oldvdevs) 1588 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1589 1590 if (sav->sav_config == NULL) 1591 goto out; 1592 1593 sav->sav_vdevs = newvdevs; 1594 sav->sav_count = (int)nl2cache; 1595 1596 /* 1597 * Recompute the stashed list of l2cache devices, with status 1598 * information this time. 1599 */ 1600 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1601 DATA_TYPE_NVLIST_ARRAY) == 0); 1602 1603 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1604 for (i = 0; i < sav->sav_count; i++) 1605 l2cache[i] = vdev_config_generate(spa, 1606 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1607 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1608 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1609out: 1610 for (i = 0; i < sav->sav_count; i++) 1611 nvlist_free(l2cache[i]); 1612 if (sav->sav_count) 1613 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1614} 1615 1616static int 1617load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1618{ 1619 dmu_buf_t *db; 1620 char *packed = NULL; 1621 size_t nvsize = 0; 1622 int error; 1623 *value = NULL; 1624 1625 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1626 if (error != 0) 1627 return (error); 1628 nvsize = *(uint64_t *)db->db_data; 1629 dmu_buf_rele(db, FTAG); 1630 1631 packed = kmem_alloc(nvsize, KM_SLEEP); 1632 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1633 DMU_READ_PREFETCH); 1634 if (error == 0) 1635 error = nvlist_unpack(packed, nvsize, value, 0); 1636 kmem_free(packed, nvsize); 1637 1638 return (error); 1639} 1640 1641/* 1642 * Checks to see if the given vdev could not be opened, in which case we post a 1643 * sysevent to notify the autoreplace code that the device has been removed. 1644 */ 1645static void 1646spa_check_removed(vdev_t *vd) 1647{ 1648 for (int c = 0; c < vd->vdev_children; c++) 1649 spa_check_removed(vd->vdev_child[c]); 1650 1651 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1652 !vd->vdev_ishole) { 1653 zfs_post_autoreplace(vd->vdev_spa, vd); 1654 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1655 } 1656} 1657 1658/* 1659 * Validate the current config against the MOS config 1660 */ 1661static boolean_t 1662spa_config_valid(spa_t *spa, nvlist_t *config) 1663{ 1664 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1665 nvlist_t *nv; 1666 1667 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1668 1669 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1670 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1671 1672 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1673 1674 /* 1675 * If we're doing a normal import, then build up any additional 1676 * diagnostic information about missing devices in this config. 1677 * We'll pass this up to the user for further processing. 1678 */ 1679 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1680 nvlist_t **child, *nv; 1681 uint64_t idx = 0; 1682 1683 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1684 KM_SLEEP); 1685 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1686 1687 for (int c = 0; c < rvd->vdev_children; c++) { 1688 vdev_t *tvd = rvd->vdev_child[c]; 1689 vdev_t *mtvd = mrvd->vdev_child[c]; 1690 1691 if (tvd->vdev_ops == &vdev_missing_ops && 1692 mtvd->vdev_ops != &vdev_missing_ops && 1693 mtvd->vdev_islog) 1694 child[idx++] = vdev_config_generate(spa, mtvd, 1695 B_FALSE, 0); 1696 } 1697 1698 if (idx) { 1699 VERIFY(nvlist_add_nvlist_array(nv, 1700 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1701 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1702 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1703 1704 for (int i = 0; i < idx; i++) 1705 nvlist_free(child[i]); 1706 } 1707 nvlist_free(nv); 1708 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1709 } 1710 1711 /* 1712 * Compare the root vdev tree with the information we have 1713 * from the MOS config (mrvd). Check each top-level vdev 1714 * with the corresponding MOS config top-level (mtvd). 1715 */ 1716 for (int c = 0; c < rvd->vdev_children; c++) { 1717 vdev_t *tvd = rvd->vdev_child[c]; 1718 vdev_t *mtvd = mrvd->vdev_child[c]; 1719 1720 /* 1721 * Resolve any "missing" vdevs in the current configuration. 1722 * If we find that the MOS config has more accurate information 1723 * about the top-level vdev then use that vdev instead. 1724 */ 1725 if (tvd->vdev_ops == &vdev_missing_ops && 1726 mtvd->vdev_ops != &vdev_missing_ops) { 1727 1728 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1729 continue; 1730 1731 /* 1732 * Device specific actions. 1733 */ 1734 if (mtvd->vdev_islog) { 1735 spa_set_log_state(spa, SPA_LOG_CLEAR); 1736 } else { 1737 /* 1738 * XXX - once we have 'readonly' pool 1739 * support we should be able to handle 1740 * missing data devices by transitioning 1741 * the pool to readonly. 1742 */ 1743 continue; 1744 } 1745 1746 /* 1747 * Swap the missing vdev with the data we were 1748 * able to obtain from the MOS config. 1749 */ 1750 vdev_remove_child(rvd, tvd); 1751 vdev_remove_child(mrvd, mtvd); 1752 1753 vdev_add_child(rvd, mtvd); 1754 vdev_add_child(mrvd, tvd); 1755 1756 spa_config_exit(spa, SCL_ALL, FTAG); 1757 vdev_load(mtvd); 1758 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1759 1760 vdev_reopen(rvd); 1761 } else if (mtvd->vdev_islog) { 1762 /* 1763 * Load the slog device's state from the MOS config 1764 * since it's possible that the label does not 1765 * contain the most up-to-date information. 1766 */ 1767 vdev_load_log_state(tvd, mtvd); 1768 vdev_reopen(tvd); 1769 } 1770 } 1771 vdev_free(mrvd); 1772 spa_config_exit(spa, SCL_ALL, FTAG); 1773 1774 /* 1775 * Ensure we were able to validate the config. 1776 */ 1777 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1778} 1779 1780/* 1781 * Check for missing log devices 1782 */ 1783static boolean_t 1784spa_check_logs(spa_t *spa) 1785{ 1786 boolean_t rv = B_FALSE; 1787 1788 switch (spa->spa_log_state) { 1789 case SPA_LOG_MISSING: 1790 /* need to recheck in case slog has been restored */ 1791 case SPA_LOG_UNKNOWN: 1792 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1793 NULL, DS_FIND_CHILDREN) != 0); 1794 if (rv) 1795 spa_set_log_state(spa, SPA_LOG_MISSING); 1796 break; 1797 } 1798 return (rv); 1799} 1800 1801static boolean_t 1802spa_passivate_log(spa_t *spa) 1803{ 1804 vdev_t *rvd = spa->spa_root_vdev; 1805 boolean_t slog_found = B_FALSE; 1806 1807 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1808 1809 if (!spa_has_slogs(spa)) 1810 return (B_FALSE); 1811 1812 for (int c = 0; c < rvd->vdev_children; c++) { 1813 vdev_t *tvd = rvd->vdev_child[c]; 1814 metaslab_group_t *mg = tvd->vdev_mg; 1815 1816 if (tvd->vdev_islog) { 1817 metaslab_group_passivate(mg); 1818 slog_found = B_TRUE; 1819 } 1820 } 1821 1822 return (slog_found); 1823} 1824 1825static void 1826spa_activate_log(spa_t *spa) 1827{ 1828 vdev_t *rvd = spa->spa_root_vdev; 1829 1830 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1831 1832 for (int c = 0; c < rvd->vdev_children; c++) { 1833 vdev_t *tvd = rvd->vdev_child[c]; 1834 metaslab_group_t *mg = tvd->vdev_mg; 1835 1836 if (tvd->vdev_islog) 1837 metaslab_group_activate(mg); 1838 } 1839} 1840 1841int 1842spa_offline_log(spa_t *spa) 1843{ 1844 int error; 1845 1846 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1847 NULL, DS_FIND_CHILDREN); 1848 if (error == 0) { 1849 /* 1850 * We successfully offlined the log device, sync out the 1851 * current txg so that the "stubby" block can be removed 1852 * by zil_sync(). 1853 */ 1854 txg_wait_synced(spa->spa_dsl_pool, 0); 1855 } 1856 return (error); 1857} 1858 1859static void 1860spa_aux_check_removed(spa_aux_vdev_t *sav) 1861{ 1862 int i; 1863 1864 for (i = 0; i < sav->sav_count; i++) 1865 spa_check_removed(sav->sav_vdevs[i]); 1866} 1867 1868void 1869spa_claim_notify(zio_t *zio) 1870{ 1871 spa_t *spa = zio->io_spa; 1872 1873 if (zio->io_error) 1874 return; 1875 1876 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1877 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1878 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1879 mutex_exit(&spa->spa_props_lock); 1880} 1881 1882typedef struct spa_load_error { 1883 uint64_t sle_meta_count; 1884 uint64_t sle_data_count; 1885} spa_load_error_t; 1886 1887static void 1888spa_load_verify_done(zio_t *zio) 1889{ 1890 blkptr_t *bp = zio->io_bp; 1891 spa_load_error_t *sle = zio->io_private; 1892 dmu_object_type_t type = BP_GET_TYPE(bp); 1893 int error = zio->io_error; 1894 spa_t *spa = zio->io_spa; 1895 1896 if (error) { 1897 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1898 type != DMU_OT_INTENT_LOG) 1899 atomic_inc_64(&sle->sle_meta_count); 1900 else 1901 atomic_inc_64(&sle->sle_data_count); 1902 } 1903 zio_data_buf_free(zio->io_data, zio->io_size); 1904 1905 mutex_enter(&spa->spa_scrub_lock); 1906 spa->spa_scrub_inflight--; 1907 cv_broadcast(&spa->spa_scrub_io_cv); 1908 mutex_exit(&spa->spa_scrub_lock); 1909} 1910 1911/* 1912 * Maximum number of concurrent scrub i/os to create while verifying 1913 * a pool while importing it. 1914 */ 1915int spa_load_verify_maxinflight = 10000; 1916boolean_t spa_load_verify_metadata = B_TRUE; 1917boolean_t spa_load_verify_data = B_TRUE; 1918 1919SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1920 &spa_load_verify_maxinflight, 0, 1921 "Maximum number of concurrent scrub I/Os to create while verifying a " 1922 "pool while importing it"); 1923 1924SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1925 &spa_load_verify_metadata, 0, 1926 "Check metadata on import?"); 1927 1928SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1929 &spa_load_verify_data, 0, 1930 "Check user data on import?"); 1931 1932/*ARGSUSED*/ 1933static int 1934spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1935 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1936{ 1937 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1938 return (0); 1939 /* 1940 * Note: normally this routine will not be called if 1941 * spa_load_verify_metadata is not set. However, it may be useful 1942 * to manually set the flag after the traversal has begun. 1943 */ 1944 if (!spa_load_verify_metadata) 1945 return (0); 1946 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1947 return (0); 1948 1949 zio_t *rio = arg; 1950 size_t size = BP_GET_PSIZE(bp); 1951 void *data = zio_data_buf_alloc(size); 1952 1953 mutex_enter(&spa->spa_scrub_lock); 1954 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1955 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1956 spa->spa_scrub_inflight++; 1957 mutex_exit(&spa->spa_scrub_lock); 1958 1959 zio_nowait(zio_read(rio, spa, bp, data, size, 1960 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1961 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1962 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1963 return (0); 1964} 1965 1966static int 1967spa_load_verify(spa_t *spa) 1968{ 1969 zio_t *rio; 1970 spa_load_error_t sle = { 0 }; 1971 zpool_rewind_policy_t policy; 1972 boolean_t verify_ok = B_FALSE; 1973 int error = 0; 1974 1975 zpool_get_rewind_policy(spa->spa_config, &policy); 1976 1977 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1978 return (0); 1979 1980 rio = zio_root(spa, NULL, &sle, 1981 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1982 1983 if (spa_load_verify_metadata) { 1984 error = traverse_pool(spa, spa->spa_verify_min_txg, 1985 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1986 spa_load_verify_cb, rio); 1987 } 1988 1989 (void) zio_wait(rio); 1990 1991 spa->spa_load_meta_errors = sle.sle_meta_count; 1992 spa->spa_load_data_errors = sle.sle_data_count; 1993 1994 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1995 sle.sle_data_count <= policy.zrp_maxdata) { 1996 int64_t loss = 0; 1997 1998 verify_ok = B_TRUE; 1999 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2000 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2001 2002 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2003 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2004 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2005 VERIFY(nvlist_add_int64(spa->spa_load_info, 2006 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2007 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2008 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2009 } else { 2010 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2011 } 2012 2013 if (error) { 2014 if (error != ENXIO && error != EIO) 2015 error = SET_ERROR(EIO); 2016 return (error); 2017 } 2018 2019 return (verify_ok ? 0 : EIO); 2020} 2021 2022/* 2023 * Find a value in the pool props object. 2024 */ 2025static void 2026spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2027{ 2028 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2029 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2030} 2031 2032/* 2033 * Find a value in the pool directory object. 2034 */ 2035static int 2036spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2037{ 2038 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2039 name, sizeof (uint64_t), 1, val)); 2040} 2041 2042static int 2043spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2044{ 2045 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2046 return (err); 2047} 2048 2049/* 2050 * Fix up config after a partly-completed split. This is done with the 2051 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2052 * pool have that entry in their config, but only the splitting one contains 2053 * a list of all the guids of the vdevs that are being split off. 2054 * 2055 * This function determines what to do with that list: either rejoin 2056 * all the disks to the pool, or complete the splitting process. To attempt 2057 * the rejoin, each disk that is offlined is marked online again, and 2058 * we do a reopen() call. If the vdev label for every disk that was 2059 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2060 * then we call vdev_split() on each disk, and complete the split. 2061 * 2062 * Otherwise we leave the config alone, with all the vdevs in place in 2063 * the original pool. 2064 */ 2065static void 2066spa_try_repair(spa_t *spa, nvlist_t *config) 2067{ 2068 uint_t extracted; 2069 uint64_t *glist; 2070 uint_t i, gcount; 2071 nvlist_t *nvl; 2072 vdev_t **vd; 2073 boolean_t attempt_reopen; 2074 2075 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2076 return; 2077 2078 /* check that the config is complete */ 2079 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2080 &glist, &gcount) != 0) 2081 return; 2082 2083 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2084 2085 /* attempt to online all the vdevs & validate */ 2086 attempt_reopen = B_TRUE; 2087 for (i = 0; i < gcount; i++) { 2088 if (glist[i] == 0) /* vdev is hole */ 2089 continue; 2090 2091 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2092 if (vd[i] == NULL) { 2093 /* 2094 * Don't bother attempting to reopen the disks; 2095 * just do the split. 2096 */ 2097 attempt_reopen = B_FALSE; 2098 } else { 2099 /* attempt to re-online it */ 2100 vd[i]->vdev_offline = B_FALSE; 2101 } 2102 } 2103 2104 if (attempt_reopen) { 2105 vdev_reopen(spa->spa_root_vdev); 2106 2107 /* check each device to see what state it's in */ 2108 for (extracted = 0, i = 0; i < gcount; i++) { 2109 if (vd[i] != NULL && 2110 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2111 break; 2112 ++extracted; 2113 } 2114 } 2115 2116 /* 2117 * If every disk has been moved to the new pool, or if we never 2118 * even attempted to look at them, then we split them off for 2119 * good. 2120 */ 2121 if (!attempt_reopen || gcount == extracted) { 2122 for (i = 0; i < gcount; i++) 2123 if (vd[i] != NULL) 2124 vdev_split(vd[i]); 2125 vdev_reopen(spa->spa_root_vdev); 2126 } 2127 2128 kmem_free(vd, gcount * sizeof (vdev_t *)); 2129} 2130 2131static int 2132spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2133 boolean_t mosconfig) 2134{ 2135 nvlist_t *config = spa->spa_config; 2136 char *ereport = FM_EREPORT_ZFS_POOL; 2137 char *comment; 2138 int error; 2139 uint64_t pool_guid; 2140 nvlist_t *nvl; 2141 2142 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2143 return (SET_ERROR(EINVAL)); 2144 2145 ASSERT(spa->spa_comment == NULL); 2146 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2147 spa->spa_comment = spa_strdup(comment); 2148 2149 /* 2150 * Versioning wasn't explicitly added to the label until later, so if 2151 * it's not present treat it as the initial version. 2152 */ 2153 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2154 &spa->spa_ubsync.ub_version) != 0) 2155 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2156 2157 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2158 &spa->spa_config_txg); 2159 2160 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2161 spa_guid_exists(pool_guid, 0)) { 2162 error = SET_ERROR(EEXIST); 2163 } else { 2164 spa->spa_config_guid = pool_guid; 2165 2166 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2167 &nvl) == 0) { 2168 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2169 KM_SLEEP) == 0); 2170 } 2171 2172 nvlist_free(spa->spa_load_info); 2173 spa->spa_load_info = fnvlist_alloc(); 2174 2175 gethrestime(&spa->spa_loaded_ts); 2176 error = spa_load_impl(spa, pool_guid, config, state, type, 2177 mosconfig, &ereport); 2178 } 2179 2180 spa->spa_minref = refcount_count(&spa->spa_refcount); 2181 if (error) { 2182 if (error != EEXIST) { 2183 spa->spa_loaded_ts.tv_sec = 0; 2184 spa->spa_loaded_ts.tv_nsec = 0; 2185 } 2186 if (error != EBADF) { 2187 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2188 } 2189 } 2190 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2191 spa->spa_ena = 0; 2192 2193 return (error); 2194} 2195 2196/* 2197 * Load an existing storage pool, using the pool's builtin spa_config as a 2198 * source of configuration information. 2199 */ 2200static int 2201spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2202 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2203 char **ereport) 2204{ 2205 int error = 0; 2206 nvlist_t *nvroot = NULL; 2207 nvlist_t *label; 2208 vdev_t *rvd; 2209 uberblock_t *ub = &spa->spa_uberblock; 2210 uint64_t children, config_cache_txg = spa->spa_config_txg; 2211 int orig_mode = spa->spa_mode; 2212 int parse; 2213 uint64_t obj; 2214 boolean_t missing_feat_write = B_FALSE; 2215 2216 /* 2217 * If this is an untrusted config, access the pool in read-only mode. 2218 * This prevents things like resilvering recently removed devices. 2219 */ 2220 if (!mosconfig) 2221 spa->spa_mode = FREAD; 2222 2223 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2224 2225 spa->spa_load_state = state; 2226 2227 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2228 return (SET_ERROR(EINVAL)); 2229 2230 parse = (type == SPA_IMPORT_EXISTING ? 2231 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2232 2233 /* 2234 * Create "The Godfather" zio to hold all async IOs 2235 */ 2236 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2237 KM_SLEEP); 2238 for (int i = 0; i < max_ncpus; i++) { 2239 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2240 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2241 ZIO_FLAG_GODFATHER); 2242 } 2243 2244 /* 2245 * Parse the configuration into a vdev tree. We explicitly set the 2246 * value that will be returned by spa_version() since parsing the 2247 * configuration requires knowing the version number. 2248 */ 2249 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2250 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2251 spa_config_exit(spa, SCL_ALL, FTAG); 2252 2253 if (error != 0) 2254 return (error); 2255 2256 ASSERT(spa->spa_root_vdev == rvd); 2257 2258 if (type != SPA_IMPORT_ASSEMBLE) { 2259 ASSERT(spa_guid(spa) == pool_guid); 2260 } 2261 2262 /* 2263 * Try to open all vdevs, loading each label in the process. 2264 */ 2265 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2266 error = vdev_open(rvd); 2267 spa_config_exit(spa, SCL_ALL, FTAG); 2268 if (error != 0) 2269 return (error); 2270 2271 /* 2272 * We need to validate the vdev labels against the configuration that 2273 * we have in hand, which is dependent on the setting of mosconfig. If 2274 * mosconfig is true then we're validating the vdev labels based on 2275 * that config. Otherwise, we're validating against the cached config 2276 * (zpool.cache) that was read when we loaded the zfs module, and then 2277 * later we will recursively call spa_load() and validate against 2278 * the vdev config. 2279 * 2280 * If we're assembling a new pool that's been split off from an 2281 * existing pool, the labels haven't yet been updated so we skip 2282 * validation for now. 2283 */ 2284 if (type != SPA_IMPORT_ASSEMBLE) { 2285 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2286 error = vdev_validate(rvd, mosconfig); 2287 spa_config_exit(spa, SCL_ALL, FTAG); 2288 2289 if (error != 0) 2290 return (error); 2291 2292 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2293 return (SET_ERROR(ENXIO)); 2294 } 2295 2296 /* 2297 * Find the best uberblock. 2298 */ 2299 vdev_uberblock_load(rvd, ub, &label); 2300 2301 /* 2302 * If we weren't able to find a single valid uberblock, return failure. 2303 */ 2304 if (ub->ub_txg == 0) { 2305 nvlist_free(label); 2306 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2307 } 2308 2309 /* 2310 * If the pool has an unsupported version we can't open it. 2311 */ 2312 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2313 nvlist_free(label); 2314 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2315 } 2316 2317 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2318 nvlist_t *features; 2319 2320 /* 2321 * If we weren't able to find what's necessary for reading the 2322 * MOS in the label, return failure. 2323 */ 2324 if (label == NULL || nvlist_lookup_nvlist(label, 2325 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2326 nvlist_free(label); 2327 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2328 ENXIO)); 2329 } 2330 2331 /* 2332 * Update our in-core representation with the definitive values 2333 * from the label. 2334 */ 2335 nvlist_free(spa->spa_label_features); 2336 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2337 } 2338 2339 nvlist_free(label); 2340 2341 /* 2342 * Look through entries in the label nvlist's features_for_read. If 2343 * there is a feature listed there which we don't understand then we 2344 * cannot open a pool. 2345 */ 2346 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2347 nvlist_t *unsup_feat; 2348 2349 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2350 0); 2351 2352 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2353 NULL); nvp != NULL; 2354 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2355 if (!zfeature_is_supported(nvpair_name(nvp))) { 2356 VERIFY(nvlist_add_string(unsup_feat, 2357 nvpair_name(nvp), "") == 0); 2358 } 2359 } 2360 2361 if (!nvlist_empty(unsup_feat)) { 2362 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2363 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2364 nvlist_free(unsup_feat); 2365 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2366 ENOTSUP)); 2367 } 2368 2369 nvlist_free(unsup_feat); 2370 } 2371 2372 /* 2373 * If the vdev guid sum doesn't match the uberblock, we have an 2374 * incomplete configuration. We first check to see if the pool 2375 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2376 * If it is, defer the vdev_guid_sum check till later so we 2377 * can handle missing vdevs. 2378 */ 2379 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2380 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2381 rvd->vdev_guid_sum != ub->ub_guid_sum) 2382 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2383 2384 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2385 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2386 spa_try_repair(spa, config); 2387 spa_config_exit(spa, SCL_ALL, FTAG); 2388 nvlist_free(spa->spa_config_splitting); 2389 spa->spa_config_splitting = NULL; 2390 } 2391 2392 /* 2393 * Initialize internal SPA structures. 2394 */ 2395 spa->spa_state = POOL_STATE_ACTIVE; 2396 spa->spa_ubsync = spa->spa_uberblock; 2397 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2398 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2399 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2400 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2401 spa->spa_claim_max_txg = spa->spa_first_txg; 2402 spa->spa_prev_software_version = ub->ub_software_version; 2403 2404 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2405 if (error) 2406 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2407 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2408 2409 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2410 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2411 2412 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2413 boolean_t missing_feat_read = B_FALSE; 2414 nvlist_t *unsup_feat, *enabled_feat; 2415 2416 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2417 &spa->spa_feat_for_read_obj) != 0) { 2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2419 } 2420 2421 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2422 &spa->spa_feat_for_write_obj) != 0) { 2423 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2424 } 2425 2426 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2427 &spa->spa_feat_desc_obj) != 0) { 2428 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2429 } 2430 2431 enabled_feat = fnvlist_alloc(); 2432 unsup_feat = fnvlist_alloc(); 2433 2434 if (!spa_features_check(spa, B_FALSE, 2435 unsup_feat, enabled_feat)) 2436 missing_feat_read = B_TRUE; 2437 2438 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2439 if (!spa_features_check(spa, B_TRUE, 2440 unsup_feat, enabled_feat)) { 2441 missing_feat_write = B_TRUE; 2442 } 2443 } 2444 2445 fnvlist_add_nvlist(spa->spa_load_info, 2446 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2447 2448 if (!nvlist_empty(unsup_feat)) { 2449 fnvlist_add_nvlist(spa->spa_load_info, 2450 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2451 } 2452 2453 fnvlist_free(enabled_feat); 2454 fnvlist_free(unsup_feat); 2455 2456 if (!missing_feat_read) { 2457 fnvlist_add_boolean(spa->spa_load_info, 2458 ZPOOL_CONFIG_CAN_RDONLY); 2459 } 2460 2461 /* 2462 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2463 * twofold: to determine whether the pool is available for 2464 * import in read-write mode and (if it is not) whether the 2465 * pool is available for import in read-only mode. If the pool 2466 * is available for import in read-write mode, it is displayed 2467 * as available in userland; if it is not available for import 2468 * in read-only mode, it is displayed as unavailable in 2469 * userland. If the pool is available for import in read-only 2470 * mode but not read-write mode, it is displayed as unavailable 2471 * in userland with a special note that the pool is actually 2472 * available for open in read-only mode. 2473 * 2474 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2475 * missing a feature for write, we must first determine whether 2476 * the pool can be opened read-only before returning to 2477 * userland in order to know whether to display the 2478 * abovementioned note. 2479 */ 2480 if (missing_feat_read || (missing_feat_write && 2481 spa_writeable(spa))) { 2482 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2483 ENOTSUP)); 2484 } 2485 2486 /* 2487 * Load refcounts for ZFS features from disk into an in-memory 2488 * cache during SPA initialization. 2489 */ 2490 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2491 uint64_t refcount; 2492 2493 error = feature_get_refcount_from_disk(spa, 2494 &spa_feature_table[i], &refcount); 2495 if (error == 0) { 2496 spa->spa_feat_refcount_cache[i] = refcount; 2497 } else if (error == ENOTSUP) { 2498 spa->spa_feat_refcount_cache[i] = 2499 SPA_FEATURE_DISABLED; 2500 } else { 2501 return (spa_vdev_err(rvd, 2502 VDEV_AUX_CORRUPT_DATA, EIO)); 2503 } 2504 } 2505 } 2506 2507 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2508 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2509 &spa->spa_feat_enabled_txg_obj) != 0) 2510 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2511 } 2512 2513 spa->spa_is_initializing = B_TRUE; 2514 error = dsl_pool_open(spa->spa_dsl_pool); 2515 spa->spa_is_initializing = B_FALSE; 2516 if (error != 0) 2517 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2518 2519 if (!mosconfig) { 2520 uint64_t hostid; 2521 nvlist_t *policy = NULL, *nvconfig; 2522 2523 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2525 2526 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2527 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2528 char *hostname; 2529 unsigned long myhostid = 0; 2530 2531 VERIFY(nvlist_lookup_string(nvconfig, 2532 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2533 2534#ifdef _KERNEL 2535 myhostid = zone_get_hostid(NULL); 2536#else /* _KERNEL */ 2537 /* 2538 * We're emulating the system's hostid in userland, so 2539 * we can't use zone_get_hostid(). 2540 */ 2541 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2542#endif /* _KERNEL */ 2543 if (check_hostid && hostid != 0 && myhostid != 0 && 2544 hostid != myhostid) { 2545 nvlist_free(nvconfig); 2546 cmn_err(CE_WARN, "pool '%s' could not be " 2547 "loaded as it was last accessed by " 2548 "another system (host: %s hostid: 0x%lx). " 2549 "See: http://illumos.org/msg/ZFS-8000-EY", 2550 spa_name(spa), hostname, 2551 (unsigned long)hostid); 2552 return (SET_ERROR(EBADF)); 2553 } 2554 } 2555 if (nvlist_lookup_nvlist(spa->spa_config, 2556 ZPOOL_REWIND_POLICY, &policy) == 0) 2557 VERIFY(nvlist_add_nvlist(nvconfig, 2558 ZPOOL_REWIND_POLICY, policy) == 0); 2559 2560 spa_config_set(spa, nvconfig); 2561 spa_unload(spa); 2562 spa_deactivate(spa); 2563 spa_activate(spa, orig_mode); 2564 2565 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2566 } 2567 2568 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2569 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2570 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2571 if (error != 0) 2572 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2573 2574 /* 2575 * Load the bit that tells us to use the new accounting function 2576 * (raid-z deflation). If we have an older pool, this will not 2577 * be present. 2578 */ 2579 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2580 if (error != 0 && error != ENOENT) 2581 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2582 2583 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2584 &spa->spa_creation_version); 2585 if (error != 0 && error != ENOENT) 2586 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2587 2588 /* 2589 * Load the persistent error log. If we have an older pool, this will 2590 * not be present. 2591 */ 2592 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2593 if (error != 0 && error != ENOENT) 2594 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2595 2596 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2597 &spa->spa_errlog_scrub); 2598 if (error != 0 && error != ENOENT) 2599 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2600 2601 /* 2602 * Load the history object. If we have an older pool, this 2603 * will not be present. 2604 */ 2605 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2606 if (error != 0 && error != ENOENT) 2607 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2608 2609 /* 2610 * If we're assembling the pool from the split-off vdevs of 2611 * an existing pool, we don't want to attach the spares & cache 2612 * devices. 2613 */ 2614 2615 /* 2616 * Load any hot spares for this pool. 2617 */ 2618 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2619 if (error != 0 && error != ENOENT) 2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2621 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2622 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2623 if (load_nvlist(spa, spa->spa_spares.sav_object, 2624 &spa->spa_spares.sav_config) != 0) 2625 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2626 2627 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2628 spa_load_spares(spa); 2629 spa_config_exit(spa, SCL_ALL, FTAG); 2630 } else if (error == 0) { 2631 spa->spa_spares.sav_sync = B_TRUE; 2632 } 2633 2634 /* 2635 * Load any level 2 ARC devices for this pool. 2636 */ 2637 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2638 &spa->spa_l2cache.sav_object); 2639 if (error != 0 && error != ENOENT) 2640 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2641 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2642 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2643 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2644 &spa->spa_l2cache.sav_config) != 0) 2645 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2646 2647 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2648 spa_load_l2cache(spa); 2649 spa_config_exit(spa, SCL_ALL, FTAG); 2650 } else if (error == 0) { 2651 spa->spa_l2cache.sav_sync = B_TRUE; 2652 } 2653 2654 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2655 2656 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2657 if (error && error != ENOENT) 2658 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2659 2660 if (error == 0) { 2661 uint64_t autoreplace; 2662 2663 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2664 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2665 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2666 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2667 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2668 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2669 &spa->spa_dedup_ditto); 2670 2671 spa->spa_autoreplace = (autoreplace != 0); 2672 } 2673 2674 /* 2675 * If the 'autoreplace' property is set, then post a resource notifying 2676 * the ZFS DE that it should not issue any faults for unopenable 2677 * devices. We also iterate over the vdevs, and post a sysevent for any 2678 * unopenable vdevs so that the normal autoreplace handler can take 2679 * over. 2680 */ 2681 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2682 spa_check_removed(spa->spa_root_vdev); 2683 /* 2684 * For the import case, this is done in spa_import(), because 2685 * at this point we're using the spare definitions from 2686 * the MOS config, not necessarily from the userland config. 2687 */ 2688 if (state != SPA_LOAD_IMPORT) { 2689 spa_aux_check_removed(&spa->spa_spares); 2690 spa_aux_check_removed(&spa->spa_l2cache); 2691 } 2692 } 2693 2694 /* 2695 * Load the vdev state for all toplevel vdevs. 2696 */ 2697 vdev_load(rvd); 2698 2699 /* 2700 * Propagate the leaf DTLs we just loaded all the way up the tree. 2701 */ 2702 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2703 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2704 spa_config_exit(spa, SCL_ALL, FTAG); 2705 2706 /* 2707 * Load the DDTs (dedup tables). 2708 */ 2709 error = ddt_load(spa); 2710 if (error != 0) 2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2712 2713 spa_update_dspace(spa); 2714 2715 /* 2716 * Validate the config, using the MOS config to fill in any 2717 * information which might be missing. If we fail to validate 2718 * the config then declare the pool unfit for use. If we're 2719 * assembling a pool from a split, the log is not transferred 2720 * over. 2721 */ 2722 if (type != SPA_IMPORT_ASSEMBLE) { 2723 nvlist_t *nvconfig; 2724 2725 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2726 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2727 2728 if (!spa_config_valid(spa, nvconfig)) { 2729 nvlist_free(nvconfig); 2730 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2731 ENXIO)); 2732 } 2733 nvlist_free(nvconfig); 2734 2735 /* 2736 * Now that we've validated the config, check the state of the 2737 * root vdev. If it can't be opened, it indicates one or 2738 * more toplevel vdevs are faulted. 2739 */ 2740 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2741 return (SET_ERROR(ENXIO)); 2742 2743 if (spa_check_logs(spa)) { 2744 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2745 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2746 } 2747 } 2748 2749 if (missing_feat_write) { 2750 ASSERT(state == SPA_LOAD_TRYIMPORT); 2751 2752 /* 2753 * At this point, we know that we can open the pool in 2754 * read-only mode but not read-write mode. We now have enough 2755 * information and can return to userland. 2756 */ 2757 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2758 } 2759 2760 /* 2761 * We've successfully opened the pool, verify that we're ready 2762 * to start pushing transactions. 2763 */ 2764 if (state != SPA_LOAD_TRYIMPORT) { 2765 if (error = spa_load_verify(spa)) 2766 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2767 error)); 2768 } 2769 2770 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2771 spa->spa_load_max_txg == UINT64_MAX)) { 2772 dmu_tx_t *tx; 2773 int need_update = B_FALSE; 2774 2775 ASSERT(state != SPA_LOAD_TRYIMPORT); 2776 2777 /* 2778 * Claim log blocks that haven't been committed yet. 2779 * This must all happen in a single txg. 2780 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2781 * invoked from zil_claim_log_block()'s i/o done callback. 2782 * Price of rollback is that we abandon the log. 2783 */ 2784 spa->spa_claiming = B_TRUE; 2785 2786 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2787 spa_first_txg(spa)); 2788 (void) dmu_objset_find(spa_name(spa), 2789 zil_claim, tx, DS_FIND_CHILDREN); 2790 dmu_tx_commit(tx); 2791 2792 spa->spa_claiming = B_FALSE; 2793 2794 spa_set_log_state(spa, SPA_LOG_GOOD); 2795 spa->spa_sync_on = B_TRUE; 2796 txg_sync_start(spa->spa_dsl_pool); 2797 2798 /* 2799 * Wait for all claims to sync. We sync up to the highest 2800 * claimed log block birth time so that claimed log blocks 2801 * don't appear to be from the future. spa_claim_max_txg 2802 * will have been set for us by either zil_check_log_chain() 2803 * (invoked from spa_check_logs()) or zil_claim() above. 2804 */ 2805 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2806 2807 /* 2808 * If the config cache is stale, or we have uninitialized 2809 * metaslabs (see spa_vdev_add()), then update the config. 2810 * 2811 * If this is a verbatim import, trust the current 2812 * in-core spa_config and update the disk labels. 2813 */ 2814 if (config_cache_txg != spa->spa_config_txg || 2815 state == SPA_LOAD_IMPORT || 2816 state == SPA_LOAD_RECOVER || 2817 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2818 need_update = B_TRUE; 2819 2820 for (int c = 0; c < rvd->vdev_children; c++) 2821 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2822 need_update = B_TRUE; 2823 2824 /* 2825 * Update the config cache asychronously in case we're the 2826 * root pool, in which case the config cache isn't writable yet. 2827 */ 2828 if (need_update) 2829 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2830 2831 /* 2832 * Check all DTLs to see if anything needs resilvering. 2833 */ 2834 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2835 vdev_resilver_needed(rvd, NULL, NULL)) 2836 spa_async_request(spa, SPA_ASYNC_RESILVER); 2837 2838 /* 2839 * Log the fact that we booted up (so that we can detect if 2840 * we rebooted in the middle of an operation). 2841 */ 2842 spa_history_log_version(spa, "open"); 2843 2844 /* 2845 * Delete any inconsistent datasets. 2846 */ 2847 (void) dmu_objset_find(spa_name(spa), 2848 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2849 2850 /* 2851 * Clean up any stale temporary dataset userrefs. 2852 */ 2853 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2854 } 2855 2856 return (0); 2857} 2858 2859static int 2860spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2861{ 2862 int mode = spa->spa_mode; 2863 2864 spa_unload(spa); 2865 spa_deactivate(spa); 2866 2867 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2868 2869 spa_activate(spa, mode); 2870 spa_async_suspend(spa); 2871 2872 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2873} 2874 2875/* 2876 * If spa_load() fails this function will try loading prior txg's. If 2877 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2878 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2879 * function will not rewind the pool and will return the same error as 2880 * spa_load(). 2881 */ 2882static int 2883spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2884 uint64_t max_request, int rewind_flags) 2885{ 2886 nvlist_t *loadinfo = NULL; 2887 nvlist_t *config = NULL; 2888 int load_error, rewind_error; 2889 uint64_t safe_rewind_txg; 2890 uint64_t min_txg; 2891 2892 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2893 spa->spa_load_max_txg = spa->spa_load_txg; 2894 spa_set_log_state(spa, SPA_LOG_CLEAR); 2895 } else { 2896 spa->spa_load_max_txg = max_request; 2897 if (max_request != UINT64_MAX) 2898 spa->spa_extreme_rewind = B_TRUE; 2899 } 2900 2901 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2902 mosconfig); 2903 if (load_error == 0) 2904 return (0); 2905 2906 if (spa->spa_root_vdev != NULL) 2907 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2908 2909 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2910 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2911 2912 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2913 nvlist_free(config); 2914 return (load_error); 2915 } 2916 2917 if (state == SPA_LOAD_RECOVER) { 2918 /* Price of rolling back is discarding txgs, including log */ 2919 spa_set_log_state(spa, SPA_LOG_CLEAR); 2920 } else { 2921 /* 2922 * If we aren't rolling back save the load info from our first 2923 * import attempt so that we can restore it after attempting 2924 * to rewind. 2925 */ 2926 loadinfo = spa->spa_load_info; 2927 spa->spa_load_info = fnvlist_alloc(); 2928 } 2929 2930 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2931 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2932 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2933 TXG_INITIAL : safe_rewind_txg; 2934 2935 /* 2936 * Continue as long as we're finding errors, we're still within 2937 * the acceptable rewind range, and we're still finding uberblocks 2938 */ 2939 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2940 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2941 if (spa->spa_load_max_txg < safe_rewind_txg) 2942 spa->spa_extreme_rewind = B_TRUE; 2943 rewind_error = spa_load_retry(spa, state, mosconfig); 2944 } 2945 2946 spa->spa_extreme_rewind = B_FALSE; 2947 spa->spa_load_max_txg = UINT64_MAX; 2948 2949 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2950 spa_config_set(spa, config); 2951 2952 if (state == SPA_LOAD_RECOVER) { 2953 ASSERT3P(loadinfo, ==, NULL); 2954 return (rewind_error); 2955 } else { 2956 /* Store the rewind info as part of the initial load info */ 2957 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2958 spa->spa_load_info); 2959 2960 /* Restore the initial load info */ 2961 fnvlist_free(spa->spa_load_info); 2962 spa->spa_load_info = loadinfo; 2963 2964 return (load_error); 2965 } 2966} 2967 2968/* 2969 * Pool Open/Import 2970 * 2971 * The import case is identical to an open except that the configuration is sent 2972 * down from userland, instead of grabbed from the configuration cache. For the 2973 * case of an open, the pool configuration will exist in the 2974 * POOL_STATE_UNINITIALIZED state. 2975 * 2976 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2977 * the same time open the pool, without having to keep around the spa_t in some 2978 * ambiguous state. 2979 */ 2980static int 2981spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2982 nvlist_t **config) 2983{ 2984 spa_t *spa; 2985 spa_load_state_t state = SPA_LOAD_OPEN; 2986 int error; 2987 int locked = B_FALSE; 2988 int firstopen = B_FALSE; 2989 2990 *spapp = NULL; 2991 2992 /* 2993 * As disgusting as this is, we need to support recursive calls to this 2994 * function because dsl_dir_open() is called during spa_load(), and ends 2995 * up calling spa_open() again. The real fix is to figure out how to 2996 * avoid dsl_dir_open() calling this in the first place. 2997 */ 2998 if (mutex_owner(&spa_namespace_lock) != curthread) { 2999 mutex_enter(&spa_namespace_lock); 3000 locked = B_TRUE; 3001 } 3002 3003 if ((spa = spa_lookup(pool)) == NULL) { 3004 if (locked) 3005 mutex_exit(&spa_namespace_lock); 3006 return (SET_ERROR(ENOENT)); 3007 } 3008 3009 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3010 zpool_rewind_policy_t policy; 3011 3012 firstopen = B_TRUE; 3013 3014 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3015 &policy); 3016 if (policy.zrp_request & ZPOOL_DO_REWIND) 3017 state = SPA_LOAD_RECOVER; 3018 3019 spa_activate(spa, spa_mode_global); 3020 3021 if (state != SPA_LOAD_RECOVER) 3022 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3023 3024 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3025 policy.zrp_request); 3026 3027 if (error == EBADF) { 3028 /* 3029 * If vdev_validate() returns failure (indicated by 3030 * EBADF), it indicates that one of the vdevs indicates 3031 * that the pool has been exported or destroyed. If 3032 * this is the case, the config cache is out of sync and 3033 * we should remove the pool from the namespace. 3034 */ 3035 spa_unload(spa); 3036 spa_deactivate(spa); 3037 spa_config_sync(spa, B_TRUE, B_TRUE); 3038 spa_remove(spa); 3039 if (locked) 3040 mutex_exit(&spa_namespace_lock); 3041 return (SET_ERROR(ENOENT)); 3042 } 3043 3044 if (error) { 3045 /* 3046 * We can't open the pool, but we still have useful 3047 * information: the state of each vdev after the 3048 * attempted vdev_open(). Return this to the user. 3049 */ 3050 if (config != NULL && spa->spa_config) { 3051 VERIFY(nvlist_dup(spa->spa_config, config, 3052 KM_SLEEP) == 0); 3053 VERIFY(nvlist_add_nvlist(*config, 3054 ZPOOL_CONFIG_LOAD_INFO, 3055 spa->spa_load_info) == 0); 3056 } 3057 spa_unload(spa); 3058 spa_deactivate(spa); 3059 spa->spa_last_open_failed = error; 3060 if (locked) 3061 mutex_exit(&spa_namespace_lock); 3062 *spapp = NULL; 3063 return (error); 3064 } 3065 } 3066 3067 spa_open_ref(spa, tag); 3068 3069 if (config != NULL) 3070 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3071 3072 /* 3073 * If we've recovered the pool, pass back any information we 3074 * gathered while doing the load. 3075 */ 3076 if (state == SPA_LOAD_RECOVER) { 3077 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3078 spa->spa_load_info) == 0); 3079 } 3080 3081 if (locked) { 3082 spa->spa_last_open_failed = 0; 3083 spa->spa_last_ubsync_txg = 0; 3084 spa->spa_load_txg = 0; 3085 mutex_exit(&spa_namespace_lock); 3086#ifdef __FreeBSD__ 3087#ifdef _KERNEL 3088 if (firstopen) 3089 zvol_create_minors(spa->spa_name); 3090#endif 3091#endif 3092 } 3093 3094 *spapp = spa; 3095 3096 return (0); 3097} 3098 3099int 3100spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3101 nvlist_t **config) 3102{ 3103 return (spa_open_common(name, spapp, tag, policy, config)); 3104} 3105 3106int 3107spa_open(const char *name, spa_t **spapp, void *tag) 3108{ 3109 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3110} 3111 3112/* 3113 * Lookup the given spa_t, incrementing the inject count in the process, 3114 * preventing it from being exported or destroyed. 3115 */ 3116spa_t * 3117spa_inject_addref(char *name) 3118{ 3119 spa_t *spa; 3120 3121 mutex_enter(&spa_namespace_lock); 3122 if ((spa = spa_lookup(name)) == NULL) { 3123 mutex_exit(&spa_namespace_lock); 3124 return (NULL); 3125 } 3126 spa->spa_inject_ref++; 3127 mutex_exit(&spa_namespace_lock); 3128 3129 return (spa); 3130} 3131 3132void 3133spa_inject_delref(spa_t *spa) 3134{ 3135 mutex_enter(&spa_namespace_lock); 3136 spa->spa_inject_ref--; 3137 mutex_exit(&spa_namespace_lock); 3138} 3139 3140/* 3141 * Add spares device information to the nvlist. 3142 */ 3143static void 3144spa_add_spares(spa_t *spa, nvlist_t *config) 3145{ 3146 nvlist_t **spares; 3147 uint_t i, nspares; 3148 nvlist_t *nvroot; 3149 uint64_t guid; 3150 vdev_stat_t *vs; 3151 uint_t vsc; 3152 uint64_t pool; 3153 3154 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3155 3156 if (spa->spa_spares.sav_count == 0) 3157 return; 3158 3159 VERIFY(nvlist_lookup_nvlist(config, 3160 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3161 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3162 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3163 if (nspares != 0) { 3164 VERIFY(nvlist_add_nvlist_array(nvroot, 3165 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3166 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3167 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3168 3169 /* 3170 * Go through and find any spares which have since been 3171 * repurposed as an active spare. If this is the case, update 3172 * their status appropriately. 3173 */ 3174 for (i = 0; i < nspares; i++) { 3175 VERIFY(nvlist_lookup_uint64(spares[i], 3176 ZPOOL_CONFIG_GUID, &guid) == 0); 3177 if (spa_spare_exists(guid, &pool, NULL) && 3178 pool != 0ULL) { 3179 VERIFY(nvlist_lookup_uint64_array( 3180 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3181 (uint64_t **)&vs, &vsc) == 0); 3182 vs->vs_state = VDEV_STATE_CANT_OPEN; 3183 vs->vs_aux = VDEV_AUX_SPARED; 3184 } 3185 } 3186 } 3187} 3188 3189/* 3190 * Add l2cache device information to the nvlist, including vdev stats. 3191 */ 3192static void 3193spa_add_l2cache(spa_t *spa, nvlist_t *config) 3194{ 3195 nvlist_t **l2cache; 3196 uint_t i, j, nl2cache; 3197 nvlist_t *nvroot; 3198 uint64_t guid; 3199 vdev_t *vd; 3200 vdev_stat_t *vs; 3201 uint_t vsc; 3202 3203 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3204 3205 if (spa->spa_l2cache.sav_count == 0) 3206 return; 3207 3208 VERIFY(nvlist_lookup_nvlist(config, 3209 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3210 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3211 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3212 if (nl2cache != 0) { 3213 VERIFY(nvlist_add_nvlist_array(nvroot, 3214 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3215 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3216 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3217 3218 /* 3219 * Update level 2 cache device stats. 3220 */ 3221 3222 for (i = 0; i < nl2cache; i++) { 3223 VERIFY(nvlist_lookup_uint64(l2cache[i], 3224 ZPOOL_CONFIG_GUID, &guid) == 0); 3225 3226 vd = NULL; 3227 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3228 if (guid == 3229 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3230 vd = spa->spa_l2cache.sav_vdevs[j]; 3231 break; 3232 } 3233 } 3234 ASSERT(vd != NULL); 3235 3236 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3237 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3238 == 0); 3239 vdev_get_stats(vd, vs); 3240 } 3241 } 3242} 3243 3244static void 3245spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3246{ 3247 nvlist_t *features; 3248 zap_cursor_t zc; 3249 zap_attribute_t za; 3250 3251 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3252 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3253 3254 /* We may be unable to read features if pool is suspended. */ 3255 if (spa_suspended(spa)) 3256 goto out; 3257 3258 if (spa->spa_feat_for_read_obj != 0) { 3259 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3260 spa->spa_feat_for_read_obj); 3261 zap_cursor_retrieve(&zc, &za) == 0; 3262 zap_cursor_advance(&zc)) { 3263 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3264 za.za_num_integers == 1); 3265 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3266 za.za_first_integer)); 3267 } 3268 zap_cursor_fini(&zc); 3269 } 3270 3271 if (spa->spa_feat_for_write_obj != 0) { 3272 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3273 spa->spa_feat_for_write_obj); 3274 zap_cursor_retrieve(&zc, &za) == 0; 3275 zap_cursor_advance(&zc)) { 3276 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3277 za.za_num_integers == 1); 3278 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3279 za.za_first_integer)); 3280 } 3281 zap_cursor_fini(&zc); 3282 } 3283 3284out: 3285 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3286 features) == 0); 3287 nvlist_free(features); 3288} 3289 3290int 3291spa_get_stats(const char *name, nvlist_t **config, 3292 char *altroot, size_t buflen) 3293{ 3294 int error; 3295 spa_t *spa; 3296 3297 *config = NULL; 3298 error = spa_open_common(name, &spa, FTAG, NULL, config); 3299 3300 if (spa != NULL) { 3301 /* 3302 * This still leaves a window of inconsistency where the spares 3303 * or l2cache devices could change and the config would be 3304 * self-inconsistent. 3305 */ 3306 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3307 3308 if (*config != NULL) { 3309 uint64_t loadtimes[2]; 3310 3311 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3312 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3313 VERIFY(nvlist_add_uint64_array(*config, 3314 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3315 3316 VERIFY(nvlist_add_uint64(*config, 3317 ZPOOL_CONFIG_ERRCOUNT, 3318 spa_get_errlog_size(spa)) == 0); 3319 3320 if (spa_suspended(spa)) 3321 VERIFY(nvlist_add_uint64(*config, 3322 ZPOOL_CONFIG_SUSPENDED, 3323 spa->spa_failmode) == 0); 3324 3325 spa_add_spares(spa, *config); 3326 spa_add_l2cache(spa, *config); 3327 spa_add_feature_stats(spa, *config); 3328 } 3329 } 3330 3331 /* 3332 * We want to get the alternate root even for faulted pools, so we cheat 3333 * and call spa_lookup() directly. 3334 */ 3335 if (altroot) { 3336 if (spa == NULL) { 3337 mutex_enter(&spa_namespace_lock); 3338 spa = spa_lookup(name); 3339 if (spa) 3340 spa_altroot(spa, altroot, buflen); 3341 else 3342 altroot[0] = '\0'; 3343 spa = NULL; 3344 mutex_exit(&spa_namespace_lock); 3345 } else { 3346 spa_altroot(spa, altroot, buflen); 3347 } 3348 } 3349 3350 if (spa != NULL) { 3351 spa_config_exit(spa, SCL_CONFIG, FTAG); 3352 spa_close(spa, FTAG); 3353 } 3354 3355 return (error); 3356} 3357 3358/* 3359 * Validate that the auxiliary device array is well formed. We must have an 3360 * array of nvlists, each which describes a valid leaf vdev. If this is an 3361 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3362 * specified, as long as they are well-formed. 3363 */ 3364static int 3365spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3366 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3367 vdev_labeltype_t label) 3368{ 3369 nvlist_t **dev; 3370 uint_t i, ndev; 3371 vdev_t *vd; 3372 int error; 3373 3374 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3375 3376 /* 3377 * It's acceptable to have no devs specified. 3378 */ 3379 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3380 return (0); 3381 3382 if (ndev == 0) 3383 return (SET_ERROR(EINVAL)); 3384 3385 /* 3386 * Make sure the pool is formatted with a version that supports this 3387 * device type. 3388 */ 3389 if (spa_version(spa) < version) 3390 return (SET_ERROR(ENOTSUP)); 3391 3392 /* 3393 * Set the pending device list so we correctly handle device in-use 3394 * checking. 3395 */ 3396 sav->sav_pending = dev; 3397 sav->sav_npending = ndev; 3398 3399 for (i = 0; i < ndev; i++) { 3400 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3401 mode)) != 0) 3402 goto out; 3403 3404 if (!vd->vdev_ops->vdev_op_leaf) { 3405 vdev_free(vd); 3406 error = SET_ERROR(EINVAL); 3407 goto out; 3408 } 3409 3410 /* 3411 * The L2ARC currently only supports disk devices in 3412 * kernel context. For user-level testing, we allow it. 3413 */ 3414#ifdef _KERNEL 3415 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3416 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3417 error = SET_ERROR(ENOTBLK); 3418 vdev_free(vd); 3419 goto out; 3420 } 3421#endif 3422 vd->vdev_top = vd; 3423 3424 if ((error = vdev_open(vd)) == 0 && 3425 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3426 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3427 vd->vdev_guid) == 0); 3428 } 3429 3430 vdev_free(vd); 3431 3432 if (error && 3433 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3434 goto out; 3435 else 3436 error = 0; 3437 } 3438 3439out: 3440 sav->sav_pending = NULL; 3441 sav->sav_npending = 0; 3442 return (error); 3443} 3444 3445static int 3446spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3447{ 3448 int error; 3449 3450 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3451 3452 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3453 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3454 VDEV_LABEL_SPARE)) != 0) { 3455 return (error); 3456 } 3457 3458 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3459 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3460 VDEV_LABEL_L2CACHE)); 3461} 3462 3463static void 3464spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3465 const char *config) 3466{ 3467 int i; 3468 3469 if (sav->sav_config != NULL) { 3470 nvlist_t **olddevs; 3471 uint_t oldndevs; 3472 nvlist_t **newdevs; 3473 3474 /* 3475 * Generate new dev list by concatentating with the 3476 * current dev list. 3477 */ 3478 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3479 &olddevs, &oldndevs) == 0); 3480 3481 newdevs = kmem_alloc(sizeof (void *) * 3482 (ndevs + oldndevs), KM_SLEEP); 3483 for (i = 0; i < oldndevs; i++) 3484 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3485 KM_SLEEP) == 0); 3486 for (i = 0; i < ndevs; i++) 3487 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3488 KM_SLEEP) == 0); 3489 3490 VERIFY(nvlist_remove(sav->sav_config, config, 3491 DATA_TYPE_NVLIST_ARRAY) == 0); 3492 3493 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3494 config, newdevs, ndevs + oldndevs) == 0); 3495 for (i = 0; i < oldndevs + ndevs; i++) 3496 nvlist_free(newdevs[i]); 3497 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3498 } else { 3499 /* 3500 * Generate a new dev list. 3501 */ 3502 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3503 KM_SLEEP) == 0); 3504 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3505 devs, ndevs) == 0); 3506 } 3507} 3508 3509/* 3510 * Stop and drop level 2 ARC devices 3511 */ 3512void 3513spa_l2cache_drop(spa_t *spa) 3514{ 3515 vdev_t *vd; 3516 int i; 3517 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3518 3519 for (i = 0; i < sav->sav_count; i++) { 3520 uint64_t pool; 3521 3522 vd = sav->sav_vdevs[i]; 3523 ASSERT(vd != NULL); 3524 3525 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3526 pool != 0ULL && l2arc_vdev_present(vd)) 3527 l2arc_remove_vdev(vd); 3528 } 3529} 3530 3531/* 3532 * Pool Creation 3533 */ 3534int 3535spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3536 nvlist_t *zplprops) 3537{ 3538 spa_t *spa; 3539 char *altroot = NULL; 3540 vdev_t *rvd; 3541 dsl_pool_t *dp; 3542 dmu_tx_t *tx; 3543 int error = 0; 3544 uint64_t txg = TXG_INITIAL; 3545 nvlist_t **spares, **l2cache; 3546 uint_t nspares, nl2cache; 3547 uint64_t version, obj; 3548 boolean_t has_features; 3549 3550 /* 3551 * If this pool already exists, return failure. 3552 */ 3553 mutex_enter(&spa_namespace_lock); 3554 if (spa_lookup(pool) != NULL) { 3555 mutex_exit(&spa_namespace_lock); 3556 return (SET_ERROR(EEXIST)); 3557 } 3558 3559 /* 3560 * Allocate a new spa_t structure. 3561 */ 3562 (void) nvlist_lookup_string(props, 3563 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3564 spa = spa_add(pool, NULL, altroot); 3565 spa_activate(spa, spa_mode_global); 3566 3567 if (props && (error = spa_prop_validate(spa, props))) { 3568 spa_deactivate(spa); 3569 spa_remove(spa); 3570 mutex_exit(&spa_namespace_lock); 3571 return (error); 3572 } 3573 3574 has_features = B_FALSE; 3575 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3576 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3577 if (zpool_prop_feature(nvpair_name(elem))) 3578 has_features = B_TRUE; 3579 } 3580 3581 if (has_features || nvlist_lookup_uint64(props, 3582 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3583 version = SPA_VERSION; 3584 } 3585 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3586 3587 spa->spa_first_txg = txg; 3588 spa->spa_uberblock.ub_txg = txg - 1; 3589 spa->spa_uberblock.ub_version = version; 3590 spa->spa_ubsync = spa->spa_uberblock; 3591 3592 /* 3593 * Create "The Godfather" zio to hold all async IOs 3594 */ 3595 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3596 KM_SLEEP); 3597 for (int i = 0; i < max_ncpus; i++) { 3598 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3599 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3600 ZIO_FLAG_GODFATHER); 3601 } 3602 3603 /* 3604 * Create the root vdev. 3605 */ 3606 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3607 3608 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3609 3610 ASSERT(error != 0 || rvd != NULL); 3611 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3612 3613 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3614 error = SET_ERROR(EINVAL); 3615 3616 if (error == 0 && 3617 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3618 (error = spa_validate_aux(spa, nvroot, txg, 3619 VDEV_ALLOC_ADD)) == 0) { 3620 for (int c = 0; c < rvd->vdev_children; c++) { 3621 vdev_ashift_optimize(rvd->vdev_child[c]); 3622 vdev_metaslab_set_size(rvd->vdev_child[c]); 3623 vdev_expand(rvd->vdev_child[c], txg); 3624 } 3625 } 3626 3627 spa_config_exit(spa, SCL_ALL, FTAG); 3628 3629 if (error != 0) { 3630 spa_unload(spa); 3631 spa_deactivate(spa); 3632 spa_remove(spa); 3633 mutex_exit(&spa_namespace_lock); 3634 return (error); 3635 } 3636 3637 /* 3638 * Get the list of spares, if specified. 3639 */ 3640 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3641 &spares, &nspares) == 0) { 3642 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3643 KM_SLEEP) == 0); 3644 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3645 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3646 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3647 spa_load_spares(spa); 3648 spa_config_exit(spa, SCL_ALL, FTAG); 3649 spa->spa_spares.sav_sync = B_TRUE; 3650 } 3651 3652 /* 3653 * Get the list of level 2 cache devices, if specified. 3654 */ 3655 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3656 &l2cache, &nl2cache) == 0) { 3657 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3658 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3659 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3660 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3661 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3662 spa_load_l2cache(spa); 3663 spa_config_exit(spa, SCL_ALL, FTAG); 3664 spa->spa_l2cache.sav_sync = B_TRUE; 3665 } 3666 3667 spa->spa_is_initializing = B_TRUE; 3668 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3669 spa->spa_meta_objset = dp->dp_meta_objset; 3670 spa->spa_is_initializing = B_FALSE; 3671 3672 /* 3673 * Create DDTs (dedup tables). 3674 */ 3675 ddt_create(spa); 3676 3677 spa_update_dspace(spa); 3678 3679 tx = dmu_tx_create_assigned(dp, txg); 3680 3681 /* 3682 * Create the pool config object. 3683 */ 3684 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3685 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3686 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3687 3688 if (zap_add(spa->spa_meta_objset, 3689 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3690 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3691 cmn_err(CE_PANIC, "failed to add pool config"); 3692 } 3693 3694 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3695 spa_feature_create_zap_objects(spa, tx); 3696 3697 if (zap_add(spa->spa_meta_objset, 3698 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3699 sizeof (uint64_t), 1, &version, tx) != 0) { 3700 cmn_err(CE_PANIC, "failed to add pool version"); 3701 } 3702 3703 /* Newly created pools with the right version are always deflated. */ 3704 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3705 spa->spa_deflate = TRUE; 3706 if (zap_add(spa->spa_meta_objset, 3707 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3708 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3709 cmn_err(CE_PANIC, "failed to add deflate"); 3710 } 3711 } 3712 3713 /* 3714 * Create the deferred-free bpobj. Turn off compression 3715 * because sync-to-convergence takes longer if the blocksize 3716 * keeps changing. 3717 */ 3718 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3719 dmu_object_set_compress(spa->spa_meta_objset, obj, 3720 ZIO_COMPRESS_OFF, tx); 3721 if (zap_add(spa->spa_meta_objset, 3722 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3723 sizeof (uint64_t), 1, &obj, tx) != 0) { 3724 cmn_err(CE_PANIC, "failed to add bpobj"); 3725 } 3726 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3727 spa->spa_meta_objset, obj)); 3728 3729 /* 3730 * Create the pool's history object. 3731 */ 3732 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3733 spa_history_create_obj(spa, tx); 3734 3735 /* 3736 * Set pool properties. 3737 */ 3738 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3739 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3740 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3741 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3742 3743 if (props != NULL) { 3744 spa_configfile_set(spa, props, B_FALSE); 3745 spa_sync_props(props, tx); 3746 } 3747 3748 dmu_tx_commit(tx); 3749 3750 spa->spa_sync_on = B_TRUE; 3751 txg_sync_start(spa->spa_dsl_pool); 3752 3753 /* 3754 * We explicitly wait for the first transaction to complete so that our 3755 * bean counters are appropriately updated. 3756 */ 3757 txg_wait_synced(spa->spa_dsl_pool, txg); 3758 3759 spa_config_sync(spa, B_FALSE, B_TRUE); 3760 3761 spa_history_log_version(spa, "create"); 3762 3763 spa->spa_minref = refcount_count(&spa->spa_refcount); 3764 3765 mutex_exit(&spa_namespace_lock); 3766 3767 return (0); 3768} 3769 3770#ifdef _KERNEL 3771#if defined(sun) 3772/* 3773 * Get the root pool information from the root disk, then import the root pool 3774 * during the system boot up time. 3775 */ 3776extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3777 3778static nvlist_t * 3779spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3780{ 3781 nvlist_t *config; 3782 nvlist_t *nvtop, *nvroot; 3783 uint64_t pgid; 3784 3785 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3786 return (NULL); 3787 3788 /* 3789 * Add this top-level vdev to the child array. 3790 */ 3791 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3792 &nvtop) == 0); 3793 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3794 &pgid) == 0); 3795 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3796 3797 /* 3798 * Put this pool's top-level vdevs into a root vdev. 3799 */ 3800 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3801 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3802 VDEV_TYPE_ROOT) == 0); 3803 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3804 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3805 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3806 &nvtop, 1) == 0); 3807 3808 /* 3809 * Replace the existing vdev_tree with the new root vdev in 3810 * this pool's configuration (remove the old, add the new). 3811 */ 3812 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3813 nvlist_free(nvroot); 3814 return (config); 3815} 3816 3817/* 3818 * Walk the vdev tree and see if we can find a device with "better" 3819 * configuration. A configuration is "better" if the label on that 3820 * device has a more recent txg. 3821 */ 3822static void 3823spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3824{ 3825 for (int c = 0; c < vd->vdev_children; c++) 3826 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3827 3828 if (vd->vdev_ops->vdev_op_leaf) { 3829 nvlist_t *label; 3830 uint64_t label_txg; 3831 3832 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3833 &label) != 0) 3834 return; 3835 3836 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3837 &label_txg) == 0); 3838 3839 /* 3840 * Do we have a better boot device? 3841 */ 3842 if (label_txg > *txg) { 3843 *txg = label_txg; 3844 *avd = vd; 3845 } 3846 nvlist_free(label); 3847 } 3848} 3849 3850/* 3851 * Import a root pool. 3852 * 3853 * For x86. devpath_list will consist of devid and/or physpath name of 3854 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3855 * The GRUB "findroot" command will return the vdev we should boot. 3856 * 3857 * For Sparc, devpath_list consists the physpath name of the booting device 3858 * no matter the rootpool is a single device pool or a mirrored pool. 3859 * e.g. 3860 * "/pci@1f,0/ide@d/disk@0,0:a" 3861 */ 3862int 3863spa_import_rootpool(char *devpath, char *devid) 3864{ 3865 spa_t *spa; 3866 vdev_t *rvd, *bvd, *avd = NULL; 3867 nvlist_t *config, *nvtop; 3868 uint64_t guid, txg; 3869 char *pname; 3870 int error; 3871 3872 /* 3873 * Read the label from the boot device and generate a configuration. 3874 */ 3875 config = spa_generate_rootconf(devpath, devid, &guid); 3876#if defined(_OBP) && defined(_KERNEL) 3877 if (config == NULL) { 3878 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3879 /* iscsi boot */ 3880 get_iscsi_bootpath_phy(devpath); 3881 config = spa_generate_rootconf(devpath, devid, &guid); 3882 } 3883 } 3884#endif 3885 if (config == NULL) { 3886 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3887 devpath); 3888 return (SET_ERROR(EIO)); 3889 } 3890 3891 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3892 &pname) == 0); 3893 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3894 3895 mutex_enter(&spa_namespace_lock); 3896 if ((spa = spa_lookup(pname)) != NULL) { 3897 /* 3898 * Remove the existing root pool from the namespace so that we 3899 * can replace it with the correct config we just read in. 3900 */ 3901 spa_remove(spa); 3902 } 3903 3904 spa = spa_add(pname, config, NULL); 3905 spa->spa_is_root = B_TRUE; 3906 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3907 3908 /* 3909 * Build up a vdev tree based on the boot device's label config. 3910 */ 3911 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3912 &nvtop) == 0); 3913 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3914 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3915 VDEV_ALLOC_ROOTPOOL); 3916 spa_config_exit(spa, SCL_ALL, FTAG); 3917 if (error) { 3918 mutex_exit(&spa_namespace_lock); 3919 nvlist_free(config); 3920 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3921 pname); 3922 return (error); 3923 } 3924 3925 /* 3926 * Get the boot vdev. 3927 */ 3928 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3929 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3930 (u_longlong_t)guid); 3931 error = SET_ERROR(ENOENT); 3932 goto out; 3933 } 3934 3935 /* 3936 * Determine if there is a better boot device. 3937 */ 3938 avd = bvd; 3939 spa_alt_rootvdev(rvd, &avd, &txg); 3940 if (avd != bvd) { 3941 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3942 "try booting from '%s'", avd->vdev_path); 3943 error = SET_ERROR(EINVAL); 3944 goto out; 3945 } 3946 3947 /* 3948 * If the boot device is part of a spare vdev then ensure that 3949 * we're booting off the active spare. 3950 */ 3951 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3952 !bvd->vdev_isspare) { 3953 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3954 "try booting from '%s'", 3955 bvd->vdev_parent-> 3956 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3957 error = SET_ERROR(EINVAL); 3958 goto out; 3959 } 3960 3961 error = 0; 3962out: 3963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3964 vdev_free(rvd); 3965 spa_config_exit(spa, SCL_ALL, FTAG); 3966 mutex_exit(&spa_namespace_lock); 3967 3968 nvlist_free(config); 3969 return (error); 3970} 3971 3972#else 3973 3974extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3975 uint64_t *count); 3976 3977static nvlist_t * 3978spa_generate_rootconf(const char *name) 3979{ 3980 nvlist_t **configs, **tops; 3981 nvlist_t *config; 3982 nvlist_t *best_cfg, *nvtop, *nvroot; 3983 uint64_t *holes; 3984 uint64_t best_txg; 3985 uint64_t nchildren; 3986 uint64_t pgid; 3987 uint64_t count; 3988 uint64_t i; 3989 uint_t nholes; 3990 3991 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3992 return (NULL); 3993 3994 ASSERT3U(count, !=, 0); 3995 best_txg = 0; 3996 for (i = 0; i < count; i++) { 3997 uint64_t txg; 3998 3999 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4000 &txg) == 0); 4001 if (txg > best_txg) { 4002 best_txg = txg; 4003 best_cfg = configs[i]; 4004 } 4005 } 4006 4007 nchildren = 1; 4008 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4009 holes = NULL; 4010 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4011 &holes, &nholes); 4012 4013 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4014 for (i = 0; i < nchildren; i++) { 4015 if (i >= count) 4016 break; 4017 if (configs[i] == NULL) 4018 continue; 4019 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4020 &nvtop) == 0); 4021 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4022 } 4023 for (i = 0; holes != NULL && i < nholes; i++) { 4024 if (i >= nchildren) 4025 continue; 4026 if (tops[holes[i]] != NULL) 4027 continue; 4028 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4029 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4030 VDEV_TYPE_HOLE) == 0); 4031 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4032 holes[i]) == 0); 4033 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4034 0) == 0); 4035 } 4036 for (i = 0; i < nchildren; i++) { 4037 if (tops[i] != NULL) 4038 continue; 4039 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4040 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4041 VDEV_TYPE_MISSING) == 0); 4042 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4043 i) == 0); 4044 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4045 0) == 0); 4046 } 4047 4048 /* 4049 * Create pool config based on the best vdev config. 4050 */ 4051 nvlist_dup(best_cfg, &config, KM_SLEEP); 4052 4053 /* 4054 * Put this pool's top-level vdevs into a root vdev. 4055 */ 4056 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4057 &pgid) == 0); 4058 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4059 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4060 VDEV_TYPE_ROOT) == 0); 4061 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4062 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4063 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4064 tops, nchildren) == 0); 4065 4066 /* 4067 * Replace the existing vdev_tree with the new root vdev in 4068 * this pool's configuration (remove the old, add the new). 4069 */ 4070 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4071 4072 /* 4073 * Drop vdev config elements that should not be present at pool level. 4074 */ 4075 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4076 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4077 4078 for (i = 0; i < count; i++) 4079 nvlist_free(configs[i]); 4080 kmem_free(configs, count * sizeof(void *)); 4081 for (i = 0; i < nchildren; i++) 4082 nvlist_free(tops[i]); 4083 kmem_free(tops, nchildren * sizeof(void *)); 4084 nvlist_free(nvroot); 4085 return (config); 4086} 4087 4088int 4089spa_import_rootpool(const char *name) 4090{ 4091 spa_t *spa; 4092 vdev_t *rvd, *bvd, *avd = NULL; 4093 nvlist_t *config, *nvtop; 4094 uint64_t txg; 4095 char *pname; 4096 int error; 4097 4098 /* 4099 * Read the label from the boot device and generate a configuration. 4100 */ 4101 config = spa_generate_rootconf(name); 4102 4103 mutex_enter(&spa_namespace_lock); 4104 if (config != NULL) { 4105 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4106 &pname) == 0 && strcmp(name, pname) == 0); 4107 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4108 == 0); 4109 4110 if ((spa = spa_lookup(pname)) != NULL) { 4111 /* 4112 * Remove the existing root pool from the namespace so 4113 * that we can replace it with the correct config 4114 * we just read in. 4115 */ 4116 spa_remove(spa); 4117 } 4118 spa = spa_add(pname, config, NULL); 4119 4120 /* 4121 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4122 * via spa_version(). 4123 */ 4124 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4125 &spa->spa_ubsync.ub_version) != 0) 4126 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4127 } else if ((spa = spa_lookup(name)) == NULL) { 4128 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4129 name); 4130 return (EIO); 4131 } else { 4132 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4133 } 4134 spa->spa_is_root = B_TRUE; 4135 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4136 4137 /* 4138 * Build up a vdev tree based on the boot device's label config. 4139 */ 4140 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4141 &nvtop) == 0); 4142 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4143 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4144 VDEV_ALLOC_ROOTPOOL); 4145 spa_config_exit(spa, SCL_ALL, FTAG); 4146 if (error) { 4147 mutex_exit(&spa_namespace_lock); 4148 nvlist_free(config); 4149 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4150 pname); 4151 return (error); 4152 } 4153 4154 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4155 vdev_free(rvd); 4156 spa_config_exit(spa, SCL_ALL, FTAG); 4157 mutex_exit(&spa_namespace_lock); 4158 4159 nvlist_free(config); 4160 return (0); 4161} 4162 4163#endif /* sun */ 4164#endif 4165 4166/* 4167 * Import a non-root pool into the system. 4168 */ 4169int 4170spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4171{ 4172 spa_t *spa; 4173 char *altroot = NULL; 4174 spa_load_state_t state = SPA_LOAD_IMPORT; 4175 zpool_rewind_policy_t policy; 4176 uint64_t mode = spa_mode_global; 4177 uint64_t readonly = B_FALSE; 4178 int error; 4179 nvlist_t *nvroot; 4180 nvlist_t **spares, **l2cache; 4181 uint_t nspares, nl2cache; 4182 4183 /* 4184 * If a pool with this name exists, return failure. 4185 */ 4186 mutex_enter(&spa_namespace_lock); 4187 if (spa_lookup(pool) != NULL) { 4188 mutex_exit(&spa_namespace_lock); 4189 return (SET_ERROR(EEXIST)); 4190 } 4191 4192 /* 4193 * Create and initialize the spa structure. 4194 */ 4195 (void) nvlist_lookup_string(props, 4196 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4197 (void) nvlist_lookup_uint64(props, 4198 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4199 if (readonly) 4200 mode = FREAD; 4201 spa = spa_add(pool, config, altroot); 4202 spa->spa_import_flags = flags; 4203 4204 /* 4205 * Verbatim import - Take a pool and insert it into the namespace 4206 * as if it had been loaded at boot. 4207 */ 4208 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4209 if (props != NULL) 4210 spa_configfile_set(spa, props, B_FALSE); 4211 4212 spa_config_sync(spa, B_FALSE, B_TRUE); 4213 4214 mutex_exit(&spa_namespace_lock); 4215 return (0); 4216 } 4217 4218 spa_activate(spa, mode); 4219 4220 /* 4221 * Don't start async tasks until we know everything is healthy. 4222 */ 4223 spa_async_suspend(spa); 4224 4225 zpool_get_rewind_policy(config, &policy); 4226 if (policy.zrp_request & ZPOOL_DO_REWIND) 4227 state = SPA_LOAD_RECOVER; 4228 4229 /* 4230 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4231 * because the user-supplied config is actually the one to trust when 4232 * doing an import. 4233 */ 4234 if (state != SPA_LOAD_RECOVER) 4235 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4236 4237 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4238 policy.zrp_request); 4239 4240 /* 4241 * Propagate anything learned while loading the pool and pass it 4242 * back to caller (i.e. rewind info, missing devices, etc). 4243 */ 4244 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4245 spa->spa_load_info) == 0); 4246 4247 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4248 /* 4249 * Toss any existing sparelist, as it doesn't have any validity 4250 * anymore, and conflicts with spa_has_spare(). 4251 */ 4252 if (spa->spa_spares.sav_config) { 4253 nvlist_free(spa->spa_spares.sav_config); 4254 spa->spa_spares.sav_config = NULL; 4255 spa_load_spares(spa); 4256 } 4257 if (spa->spa_l2cache.sav_config) { 4258 nvlist_free(spa->spa_l2cache.sav_config); 4259 spa->spa_l2cache.sav_config = NULL; 4260 spa_load_l2cache(spa); 4261 } 4262 4263 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4264 &nvroot) == 0); 4265 if (error == 0) 4266 error = spa_validate_aux(spa, nvroot, -1ULL, 4267 VDEV_ALLOC_SPARE); 4268 if (error == 0) 4269 error = spa_validate_aux(spa, nvroot, -1ULL, 4270 VDEV_ALLOC_L2CACHE); 4271 spa_config_exit(spa, SCL_ALL, FTAG); 4272 4273 if (props != NULL) 4274 spa_configfile_set(spa, props, B_FALSE); 4275 4276 if (error != 0 || (props && spa_writeable(spa) && 4277 (error = spa_prop_set(spa, props)))) { 4278 spa_unload(spa); 4279 spa_deactivate(spa); 4280 spa_remove(spa); 4281 mutex_exit(&spa_namespace_lock); 4282 return (error); 4283 } 4284 4285 spa_async_resume(spa); 4286 4287 /* 4288 * Override any spares and level 2 cache devices as specified by 4289 * the user, as these may have correct device names/devids, etc. 4290 */ 4291 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4292 &spares, &nspares) == 0) { 4293 if (spa->spa_spares.sav_config) 4294 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4295 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4296 else 4297 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4298 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4299 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4300 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4301 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4302 spa_load_spares(spa); 4303 spa_config_exit(spa, SCL_ALL, FTAG); 4304 spa->spa_spares.sav_sync = B_TRUE; 4305 } 4306 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4307 &l2cache, &nl2cache) == 0) { 4308 if (spa->spa_l2cache.sav_config) 4309 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4310 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4311 else 4312 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4313 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4314 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4315 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4316 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4317 spa_load_l2cache(spa); 4318 spa_config_exit(spa, SCL_ALL, FTAG); 4319 spa->spa_l2cache.sav_sync = B_TRUE; 4320 } 4321 4322 /* 4323 * Check for any removed devices. 4324 */ 4325 if (spa->spa_autoreplace) { 4326 spa_aux_check_removed(&spa->spa_spares); 4327 spa_aux_check_removed(&spa->spa_l2cache); 4328 } 4329 4330 if (spa_writeable(spa)) { 4331 /* 4332 * Update the config cache to include the newly-imported pool. 4333 */ 4334 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4335 } 4336 4337 /* 4338 * It's possible that the pool was expanded while it was exported. 4339 * We kick off an async task to handle this for us. 4340 */ 4341 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4342 4343 mutex_exit(&spa_namespace_lock); 4344 spa_history_log_version(spa, "import"); 4345 4346#ifdef __FreeBSD__ 4347#ifdef _KERNEL 4348 zvol_create_minors(pool); 4349#endif 4350#endif 4351 return (0); 4352} 4353 4354nvlist_t * 4355spa_tryimport(nvlist_t *tryconfig) 4356{ 4357 nvlist_t *config = NULL; 4358 char *poolname; 4359 spa_t *spa; 4360 uint64_t state; 4361 int error; 4362 4363 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4364 return (NULL); 4365 4366 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4367 return (NULL); 4368 4369 /* 4370 * Create and initialize the spa structure. 4371 */ 4372 mutex_enter(&spa_namespace_lock); 4373 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4374 spa_activate(spa, FREAD); 4375 4376 /* 4377 * Pass off the heavy lifting to spa_load(). 4378 * Pass TRUE for mosconfig because the user-supplied config 4379 * is actually the one to trust when doing an import. 4380 */ 4381 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4382 4383 /* 4384 * If 'tryconfig' was at least parsable, return the current config. 4385 */ 4386 if (spa->spa_root_vdev != NULL) { 4387 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4388 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4389 poolname) == 0); 4390 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4391 state) == 0); 4392 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4393 spa->spa_uberblock.ub_timestamp) == 0); 4394 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4395 spa->spa_load_info) == 0); 4396 4397 /* 4398 * If the bootfs property exists on this pool then we 4399 * copy it out so that external consumers can tell which 4400 * pools are bootable. 4401 */ 4402 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4403 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4404 4405 /* 4406 * We have to play games with the name since the 4407 * pool was opened as TRYIMPORT_NAME. 4408 */ 4409 if (dsl_dsobj_to_dsname(spa_name(spa), 4410 spa->spa_bootfs, tmpname) == 0) { 4411 char *cp; 4412 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4413 4414 cp = strchr(tmpname, '/'); 4415 if (cp == NULL) { 4416 (void) strlcpy(dsname, tmpname, 4417 MAXPATHLEN); 4418 } else { 4419 (void) snprintf(dsname, MAXPATHLEN, 4420 "%s/%s", poolname, ++cp); 4421 } 4422 VERIFY(nvlist_add_string(config, 4423 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4424 kmem_free(dsname, MAXPATHLEN); 4425 } 4426 kmem_free(tmpname, MAXPATHLEN); 4427 } 4428 4429 /* 4430 * Add the list of hot spares and level 2 cache devices. 4431 */ 4432 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4433 spa_add_spares(spa, config); 4434 spa_add_l2cache(spa, config); 4435 spa_config_exit(spa, SCL_CONFIG, FTAG); 4436 } 4437 4438 spa_unload(spa); 4439 spa_deactivate(spa); 4440 spa_remove(spa); 4441 mutex_exit(&spa_namespace_lock); 4442 4443 return (config); 4444} 4445 4446/* 4447 * Pool export/destroy 4448 * 4449 * The act of destroying or exporting a pool is very simple. We make sure there 4450 * is no more pending I/O and any references to the pool are gone. Then, we 4451 * update the pool state and sync all the labels to disk, removing the 4452 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4453 * we don't sync the labels or remove the configuration cache. 4454 */ 4455static int 4456spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4457 boolean_t force, boolean_t hardforce) 4458{ 4459 spa_t *spa; 4460 4461 if (oldconfig) 4462 *oldconfig = NULL; 4463 4464 if (!(spa_mode_global & FWRITE)) 4465 return (SET_ERROR(EROFS)); 4466 4467 mutex_enter(&spa_namespace_lock); 4468 if ((spa = spa_lookup(pool)) == NULL) { 4469 mutex_exit(&spa_namespace_lock); 4470 return (SET_ERROR(ENOENT)); 4471 } 4472 4473 /* 4474 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4475 * reacquire the namespace lock, and see if we can export. 4476 */ 4477 spa_open_ref(spa, FTAG); 4478 mutex_exit(&spa_namespace_lock); 4479 spa_async_suspend(spa); 4480 mutex_enter(&spa_namespace_lock); 4481 spa_close(spa, FTAG); 4482 4483 /* 4484 * The pool will be in core if it's openable, 4485 * in which case we can modify its state. 4486 */ 4487 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4488 /* 4489 * Objsets may be open only because they're dirty, so we 4490 * have to force it to sync before checking spa_refcnt. 4491 */ 4492 txg_wait_synced(spa->spa_dsl_pool, 0); 4493 4494 /* 4495 * A pool cannot be exported or destroyed if there are active 4496 * references. If we are resetting a pool, allow references by 4497 * fault injection handlers. 4498 */ 4499 if (!spa_refcount_zero(spa) || 4500 (spa->spa_inject_ref != 0 && 4501 new_state != POOL_STATE_UNINITIALIZED)) { 4502 spa_async_resume(spa); 4503 mutex_exit(&spa_namespace_lock); 4504 return (SET_ERROR(EBUSY)); 4505 } 4506 4507 /* 4508 * A pool cannot be exported if it has an active shared spare. 4509 * This is to prevent other pools stealing the active spare 4510 * from an exported pool. At user's own will, such pool can 4511 * be forcedly exported. 4512 */ 4513 if (!force && new_state == POOL_STATE_EXPORTED && 4514 spa_has_active_shared_spare(spa)) { 4515 spa_async_resume(spa); 4516 mutex_exit(&spa_namespace_lock); 4517 return (SET_ERROR(EXDEV)); 4518 } 4519 4520 /* 4521 * We want this to be reflected on every label, 4522 * so mark them all dirty. spa_unload() will do the 4523 * final sync that pushes these changes out. 4524 */ 4525 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4526 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4527 spa->spa_state = new_state; 4528 spa->spa_final_txg = spa_last_synced_txg(spa) + 4529 TXG_DEFER_SIZE + 1; 4530 vdev_config_dirty(spa->spa_root_vdev); 4531 spa_config_exit(spa, SCL_ALL, FTAG); 4532 } 4533 } 4534 4535 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4536 4537 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4538 spa_unload(spa); 4539 spa_deactivate(spa); 4540 } 4541 4542 if (oldconfig && spa->spa_config) 4543 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4544 4545 if (new_state != POOL_STATE_UNINITIALIZED) { 4546 if (!hardforce) 4547 spa_config_sync(spa, B_TRUE, B_TRUE); 4548 spa_remove(spa); 4549 } 4550 mutex_exit(&spa_namespace_lock); 4551 4552 return (0); 4553} 4554 4555/* 4556 * Destroy a storage pool. 4557 */ 4558int 4559spa_destroy(char *pool) 4560{ 4561 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4562 B_FALSE, B_FALSE)); 4563} 4564 4565/* 4566 * Export a storage pool. 4567 */ 4568int 4569spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4570 boolean_t hardforce) 4571{ 4572 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4573 force, hardforce)); 4574} 4575 4576/* 4577 * Similar to spa_export(), this unloads the spa_t without actually removing it 4578 * from the namespace in any way. 4579 */ 4580int 4581spa_reset(char *pool) 4582{ 4583 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4584 B_FALSE, B_FALSE)); 4585} 4586 4587/* 4588 * ========================================================================== 4589 * Device manipulation 4590 * ========================================================================== 4591 */ 4592 4593/* 4594 * Add a device to a storage pool. 4595 */ 4596int 4597spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4598{ 4599 uint64_t txg, id; 4600 int error; 4601 vdev_t *rvd = spa->spa_root_vdev; 4602 vdev_t *vd, *tvd; 4603 nvlist_t **spares, **l2cache; 4604 uint_t nspares, nl2cache; 4605 4606 ASSERT(spa_writeable(spa)); 4607 4608 txg = spa_vdev_enter(spa); 4609 4610 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4611 VDEV_ALLOC_ADD)) != 0) 4612 return (spa_vdev_exit(spa, NULL, txg, error)); 4613 4614 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4615 4616 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4617 &nspares) != 0) 4618 nspares = 0; 4619 4620 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4621 &nl2cache) != 0) 4622 nl2cache = 0; 4623 4624 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4625 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4626 4627 if (vd->vdev_children != 0 && 4628 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4629 return (spa_vdev_exit(spa, vd, txg, error)); 4630 4631 /* 4632 * We must validate the spares and l2cache devices after checking the 4633 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4634 */ 4635 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4636 return (spa_vdev_exit(spa, vd, txg, error)); 4637 4638 /* 4639 * Transfer each new top-level vdev from vd to rvd. 4640 */ 4641 for (int c = 0; c < vd->vdev_children; c++) { 4642 4643 /* 4644 * Set the vdev id to the first hole, if one exists. 4645 */ 4646 for (id = 0; id < rvd->vdev_children; id++) { 4647 if (rvd->vdev_child[id]->vdev_ishole) { 4648 vdev_free(rvd->vdev_child[id]); 4649 break; 4650 } 4651 } 4652 tvd = vd->vdev_child[c]; 4653 vdev_remove_child(vd, tvd); 4654 tvd->vdev_id = id; 4655 vdev_add_child(rvd, tvd); 4656 vdev_config_dirty(tvd); 4657 } 4658 4659 if (nspares != 0) { 4660 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4661 ZPOOL_CONFIG_SPARES); 4662 spa_load_spares(spa); 4663 spa->spa_spares.sav_sync = B_TRUE; 4664 } 4665 4666 if (nl2cache != 0) { 4667 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4668 ZPOOL_CONFIG_L2CACHE); 4669 spa_load_l2cache(spa); 4670 spa->spa_l2cache.sav_sync = B_TRUE; 4671 } 4672 4673 /* 4674 * We have to be careful when adding new vdevs to an existing pool. 4675 * If other threads start allocating from these vdevs before we 4676 * sync the config cache, and we lose power, then upon reboot we may 4677 * fail to open the pool because there are DVAs that the config cache 4678 * can't translate. Therefore, we first add the vdevs without 4679 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4680 * and then let spa_config_update() initialize the new metaslabs. 4681 * 4682 * spa_load() checks for added-but-not-initialized vdevs, so that 4683 * if we lose power at any point in this sequence, the remaining 4684 * steps will be completed the next time we load the pool. 4685 */ 4686 (void) spa_vdev_exit(spa, vd, txg, 0); 4687 4688 mutex_enter(&spa_namespace_lock); 4689 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4690 mutex_exit(&spa_namespace_lock); 4691 4692 return (0); 4693} 4694 4695/* 4696 * Attach a device to a mirror. The arguments are the path to any device 4697 * in the mirror, and the nvroot for the new device. If the path specifies 4698 * a device that is not mirrored, we automatically insert the mirror vdev. 4699 * 4700 * If 'replacing' is specified, the new device is intended to replace the 4701 * existing device; in this case the two devices are made into their own 4702 * mirror using the 'replacing' vdev, which is functionally identical to 4703 * the mirror vdev (it actually reuses all the same ops) but has a few 4704 * extra rules: you can't attach to it after it's been created, and upon 4705 * completion of resilvering, the first disk (the one being replaced) 4706 * is automatically detached. 4707 */ 4708int 4709spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4710{ 4711 uint64_t txg, dtl_max_txg; 4712 vdev_t *rvd = spa->spa_root_vdev; 4713 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4714 vdev_ops_t *pvops; 4715 char *oldvdpath, *newvdpath; 4716 int newvd_isspare; 4717 int error; 4718 4719 ASSERT(spa_writeable(spa)); 4720 4721 txg = spa_vdev_enter(spa); 4722 4723 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4724 4725 if (oldvd == NULL) 4726 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4727 4728 if (!oldvd->vdev_ops->vdev_op_leaf) 4729 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4730 4731 pvd = oldvd->vdev_parent; 4732 4733 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4734 VDEV_ALLOC_ATTACH)) != 0) 4735 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4736 4737 if (newrootvd->vdev_children != 1) 4738 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4739 4740 newvd = newrootvd->vdev_child[0]; 4741 4742 if (!newvd->vdev_ops->vdev_op_leaf) 4743 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4744 4745 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4746 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4747 4748 /* 4749 * Spares can't replace logs 4750 */ 4751 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4752 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4753 4754 if (!replacing) { 4755 /* 4756 * For attach, the only allowable parent is a mirror or the root 4757 * vdev. 4758 */ 4759 if (pvd->vdev_ops != &vdev_mirror_ops && 4760 pvd->vdev_ops != &vdev_root_ops) 4761 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4762 4763 pvops = &vdev_mirror_ops; 4764 } else { 4765 /* 4766 * Active hot spares can only be replaced by inactive hot 4767 * spares. 4768 */ 4769 if (pvd->vdev_ops == &vdev_spare_ops && 4770 oldvd->vdev_isspare && 4771 !spa_has_spare(spa, newvd->vdev_guid)) 4772 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4773 4774 /* 4775 * If the source is a hot spare, and the parent isn't already a 4776 * spare, then we want to create a new hot spare. Otherwise, we 4777 * want to create a replacing vdev. The user is not allowed to 4778 * attach to a spared vdev child unless the 'isspare' state is 4779 * the same (spare replaces spare, non-spare replaces 4780 * non-spare). 4781 */ 4782 if (pvd->vdev_ops == &vdev_replacing_ops && 4783 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4784 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4785 } else if (pvd->vdev_ops == &vdev_spare_ops && 4786 newvd->vdev_isspare != oldvd->vdev_isspare) { 4787 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4788 } 4789 4790 if (newvd->vdev_isspare) 4791 pvops = &vdev_spare_ops; 4792 else 4793 pvops = &vdev_replacing_ops; 4794 } 4795 4796 /* 4797 * Make sure the new device is big enough. 4798 */ 4799 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4800 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4801 4802 /* 4803 * The new device cannot have a higher alignment requirement 4804 * than the top-level vdev. 4805 */ 4806 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4807 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4808 4809 /* 4810 * If this is an in-place replacement, update oldvd's path and devid 4811 * to make it distinguishable from newvd, and unopenable from now on. 4812 */ 4813 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4814 spa_strfree(oldvd->vdev_path); 4815 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4816 KM_SLEEP); 4817 (void) sprintf(oldvd->vdev_path, "%s/%s", 4818 newvd->vdev_path, "old"); 4819 if (oldvd->vdev_devid != NULL) { 4820 spa_strfree(oldvd->vdev_devid); 4821 oldvd->vdev_devid = NULL; 4822 } 4823 } 4824 4825 /* mark the device being resilvered */ 4826 newvd->vdev_resilver_txg = txg; 4827 4828 /* 4829 * If the parent is not a mirror, or if we're replacing, insert the new 4830 * mirror/replacing/spare vdev above oldvd. 4831 */ 4832 if (pvd->vdev_ops != pvops) 4833 pvd = vdev_add_parent(oldvd, pvops); 4834 4835 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4836 ASSERT(pvd->vdev_ops == pvops); 4837 ASSERT(oldvd->vdev_parent == pvd); 4838 4839 /* 4840 * Extract the new device from its root and add it to pvd. 4841 */ 4842 vdev_remove_child(newrootvd, newvd); 4843 newvd->vdev_id = pvd->vdev_children; 4844 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4845 vdev_add_child(pvd, newvd); 4846 4847 tvd = newvd->vdev_top; 4848 ASSERT(pvd->vdev_top == tvd); 4849 ASSERT(tvd->vdev_parent == rvd); 4850 4851 vdev_config_dirty(tvd); 4852 4853 /* 4854 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4855 * for any dmu_sync-ed blocks. It will propagate upward when 4856 * spa_vdev_exit() calls vdev_dtl_reassess(). 4857 */ 4858 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4859 4860 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4861 dtl_max_txg - TXG_INITIAL); 4862 4863 if (newvd->vdev_isspare) { 4864 spa_spare_activate(newvd); 4865 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4866 } 4867 4868 oldvdpath = spa_strdup(oldvd->vdev_path); 4869 newvdpath = spa_strdup(newvd->vdev_path); 4870 newvd_isspare = newvd->vdev_isspare; 4871 4872 /* 4873 * Mark newvd's DTL dirty in this txg. 4874 */ 4875 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4876 4877 /* 4878 * Schedule the resilver to restart in the future. We do this to 4879 * ensure that dmu_sync-ed blocks have been stitched into the 4880 * respective datasets. 4881 */ 4882 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4883 4884 /* 4885 * Commit the config 4886 */ 4887 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4888 4889 spa_history_log_internal(spa, "vdev attach", NULL, 4890 "%s vdev=%s %s vdev=%s", 4891 replacing && newvd_isspare ? "spare in" : 4892 replacing ? "replace" : "attach", newvdpath, 4893 replacing ? "for" : "to", oldvdpath); 4894 4895 spa_strfree(oldvdpath); 4896 spa_strfree(newvdpath); 4897 4898 if (spa->spa_bootfs) 4899 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4900 4901 return (0); 4902} 4903 4904/* 4905 * Detach a device from a mirror or replacing vdev. 4906 * 4907 * If 'replace_done' is specified, only detach if the parent 4908 * is a replacing vdev. 4909 */ 4910int 4911spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4912{ 4913 uint64_t txg; 4914 int error; 4915 vdev_t *rvd = spa->spa_root_vdev; 4916 vdev_t *vd, *pvd, *cvd, *tvd; 4917 boolean_t unspare = B_FALSE; 4918 uint64_t unspare_guid = 0; 4919 char *vdpath; 4920 4921 ASSERT(spa_writeable(spa)); 4922 4923 txg = spa_vdev_enter(spa); 4924 4925 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4926 4927 if (vd == NULL) 4928 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4929 4930 if (!vd->vdev_ops->vdev_op_leaf) 4931 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4932 4933 pvd = vd->vdev_parent; 4934 4935 /* 4936 * If the parent/child relationship is not as expected, don't do it. 4937 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4938 * vdev that's replacing B with C. The user's intent in replacing 4939 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4940 * the replace by detaching C, the expected behavior is to end up 4941 * M(A,B). But suppose that right after deciding to detach C, 4942 * the replacement of B completes. We would have M(A,C), and then 4943 * ask to detach C, which would leave us with just A -- not what 4944 * the user wanted. To prevent this, we make sure that the 4945 * parent/child relationship hasn't changed -- in this example, 4946 * that C's parent is still the replacing vdev R. 4947 */ 4948 if (pvd->vdev_guid != pguid && pguid != 0) 4949 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4950 4951 /* 4952 * Only 'replacing' or 'spare' vdevs can be replaced. 4953 */ 4954 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4955 pvd->vdev_ops != &vdev_spare_ops) 4956 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4957 4958 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4959 spa_version(spa) >= SPA_VERSION_SPARES); 4960 4961 /* 4962 * Only mirror, replacing, and spare vdevs support detach. 4963 */ 4964 if (pvd->vdev_ops != &vdev_replacing_ops && 4965 pvd->vdev_ops != &vdev_mirror_ops && 4966 pvd->vdev_ops != &vdev_spare_ops) 4967 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4968 4969 /* 4970 * If this device has the only valid copy of some data, 4971 * we cannot safely detach it. 4972 */ 4973 if (vdev_dtl_required(vd)) 4974 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4975 4976 ASSERT(pvd->vdev_children >= 2); 4977 4978 /* 4979 * If we are detaching the second disk from a replacing vdev, then 4980 * check to see if we changed the original vdev's path to have "/old" 4981 * at the end in spa_vdev_attach(). If so, undo that change now. 4982 */ 4983 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4984 vd->vdev_path != NULL) { 4985 size_t len = strlen(vd->vdev_path); 4986 4987 for (int c = 0; c < pvd->vdev_children; c++) { 4988 cvd = pvd->vdev_child[c]; 4989 4990 if (cvd == vd || cvd->vdev_path == NULL) 4991 continue; 4992 4993 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4994 strcmp(cvd->vdev_path + len, "/old") == 0) { 4995 spa_strfree(cvd->vdev_path); 4996 cvd->vdev_path = spa_strdup(vd->vdev_path); 4997 break; 4998 } 4999 } 5000 } 5001 5002 /* 5003 * If we are detaching the original disk from a spare, then it implies 5004 * that the spare should become a real disk, and be removed from the 5005 * active spare list for the pool. 5006 */ 5007 if (pvd->vdev_ops == &vdev_spare_ops && 5008 vd->vdev_id == 0 && 5009 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5010 unspare = B_TRUE; 5011 5012 /* 5013 * Erase the disk labels so the disk can be used for other things. 5014 * This must be done after all other error cases are handled, 5015 * but before we disembowel vd (so we can still do I/O to it). 5016 * But if we can't do it, don't treat the error as fatal -- 5017 * it may be that the unwritability of the disk is the reason 5018 * it's being detached! 5019 */ 5020 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5021 5022 /* 5023 * Remove vd from its parent and compact the parent's children. 5024 */ 5025 vdev_remove_child(pvd, vd); 5026 vdev_compact_children(pvd); 5027 5028 /* 5029 * Remember one of the remaining children so we can get tvd below. 5030 */ 5031 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5032 5033 /* 5034 * If we need to remove the remaining child from the list of hot spares, 5035 * do it now, marking the vdev as no longer a spare in the process. 5036 * We must do this before vdev_remove_parent(), because that can 5037 * change the GUID if it creates a new toplevel GUID. For a similar 5038 * reason, we must remove the spare now, in the same txg as the detach; 5039 * otherwise someone could attach a new sibling, change the GUID, and 5040 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5041 */ 5042 if (unspare) { 5043 ASSERT(cvd->vdev_isspare); 5044 spa_spare_remove(cvd); 5045 unspare_guid = cvd->vdev_guid; 5046 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5047 cvd->vdev_unspare = B_TRUE; 5048 } 5049 5050 /* 5051 * If the parent mirror/replacing vdev only has one child, 5052 * the parent is no longer needed. Remove it from the tree. 5053 */ 5054 if (pvd->vdev_children == 1) { 5055 if (pvd->vdev_ops == &vdev_spare_ops) 5056 cvd->vdev_unspare = B_FALSE; 5057 vdev_remove_parent(cvd); 5058 } 5059 5060 5061 /* 5062 * We don't set tvd until now because the parent we just removed 5063 * may have been the previous top-level vdev. 5064 */ 5065 tvd = cvd->vdev_top; 5066 ASSERT(tvd->vdev_parent == rvd); 5067 5068 /* 5069 * Reevaluate the parent vdev state. 5070 */ 5071 vdev_propagate_state(cvd); 5072 5073 /* 5074 * If the 'autoexpand' property is set on the pool then automatically 5075 * try to expand the size of the pool. For example if the device we 5076 * just detached was smaller than the others, it may be possible to 5077 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5078 * first so that we can obtain the updated sizes of the leaf vdevs. 5079 */ 5080 if (spa->spa_autoexpand) { 5081 vdev_reopen(tvd); 5082 vdev_expand(tvd, txg); 5083 } 5084 5085 vdev_config_dirty(tvd); 5086 5087 /* 5088 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5089 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5090 * But first make sure we're not on any *other* txg's DTL list, to 5091 * prevent vd from being accessed after it's freed. 5092 */ 5093 vdpath = spa_strdup(vd->vdev_path); 5094 for (int t = 0; t < TXG_SIZE; t++) 5095 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5096 vd->vdev_detached = B_TRUE; 5097 vdev_dirty(tvd, VDD_DTL, vd, txg); 5098 5099 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5100 5101 /* hang on to the spa before we release the lock */ 5102 spa_open_ref(spa, FTAG); 5103 5104 error = spa_vdev_exit(spa, vd, txg, 0); 5105 5106 spa_history_log_internal(spa, "detach", NULL, 5107 "vdev=%s", vdpath); 5108 spa_strfree(vdpath); 5109 5110 /* 5111 * If this was the removal of the original device in a hot spare vdev, 5112 * then we want to go through and remove the device from the hot spare 5113 * list of every other pool. 5114 */ 5115 if (unspare) { 5116 spa_t *altspa = NULL; 5117 5118 mutex_enter(&spa_namespace_lock); 5119 while ((altspa = spa_next(altspa)) != NULL) { 5120 if (altspa->spa_state != POOL_STATE_ACTIVE || 5121 altspa == spa) 5122 continue; 5123 5124 spa_open_ref(altspa, FTAG); 5125 mutex_exit(&spa_namespace_lock); 5126 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5127 mutex_enter(&spa_namespace_lock); 5128 spa_close(altspa, FTAG); 5129 } 5130 mutex_exit(&spa_namespace_lock); 5131 5132 /* search the rest of the vdevs for spares to remove */ 5133 spa_vdev_resilver_done(spa); 5134 } 5135 5136 /* all done with the spa; OK to release */ 5137 mutex_enter(&spa_namespace_lock); 5138 spa_close(spa, FTAG); 5139 mutex_exit(&spa_namespace_lock); 5140 5141 return (error); 5142} 5143 5144/* 5145 * Split a set of devices from their mirrors, and create a new pool from them. 5146 */ 5147int 5148spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5149 nvlist_t *props, boolean_t exp) 5150{ 5151 int error = 0; 5152 uint64_t txg, *glist; 5153 spa_t *newspa; 5154 uint_t c, children, lastlog; 5155 nvlist_t **child, *nvl, *tmp; 5156 dmu_tx_t *tx; 5157 char *altroot = NULL; 5158 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5159 boolean_t activate_slog; 5160 5161 ASSERT(spa_writeable(spa)); 5162 5163 txg = spa_vdev_enter(spa); 5164 5165 /* clear the log and flush everything up to now */ 5166 activate_slog = spa_passivate_log(spa); 5167 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5168 error = spa_offline_log(spa); 5169 txg = spa_vdev_config_enter(spa); 5170 5171 if (activate_slog) 5172 spa_activate_log(spa); 5173 5174 if (error != 0) 5175 return (spa_vdev_exit(spa, NULL, txg, error)); 5176 5177 /* check new spa name before going any further */ 5178 if (spa_lookup(newname) != NULL) 5179 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5180 5181 /* 5182 * scan through all the children to ensure they're all mirrors 5183 */ 5184 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5185 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5186 &children) != 0) 5187 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5188 5189 /* first, check to ensure we've got the right child count */ 5190 rvd = spa->spa_root_vdev; 5191 lastlog = 0; 5192 for (c = 0; c < rvd->vdev_children; c++) { 5193 vdev_t *vd = rvd->vdev_child[c]; 5194 5195 /* don't count the holes & logs as children */ 5196 if (vd->vdev_islog || vd->vdev_ishole) { 5197 if (lastlog == 0) 5198 lastlog = c; 5199 continue; 5200 } 5201 5202 lastlog = 0; 5203 } 5204 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5205 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5206 5207 /* next, ensure no spare or cache devices are part of the split */ 5208 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5209 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5210 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5211 5212 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5213 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5214 5215 /* then, loop over each vdev and validate it */ 5216 for (c = 0; c < children; c++) { 5217 uint64_t is_hole = 0; 5218 5219 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5220 &is_hole); 5221 5222 if (is_hole != 0) { 5223 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5224 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5225 continue; 5226 } else { 5227 error = SET_ERROR(EINVAL); 5228 break; 5229 } 5230 } 5231 5232 /* which disk is going to be split? */ 5233 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5234 &glist[c]) != 0) { 5235 error = SET_ERROR(EINVAL); 5236 break; 5237 } 5238 5239 /* look it up in the spa */ 5240 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5241 if (vml[c] == NULL) { 5242 error = SET_ERROR(ENODEV); 5243 break; 5244 } 5245 5246 /* make sure there's nothing stopping the split */ 5247 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5248 vml[c]->vdev_islog || 5249 vml[c]->vdev_ishole || 5250 vml[c]->vdev_isspare || 5251 vml[c]->vdev_isl2cache || 5252 !vdev_writeable(vml[c]) || 5253 vml[c]->vdev_children != 0 || 5254 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5255 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5256 error = SET_ERROR(EINVAL); 5257 break; 5258 } 5259 5260 if (vdev_dtl_required(vml[c])) { 5261 error = SET_ERROR(EBUSY); 5262 break; 5263 } 5264 5265 /* we need certain info from the top level */ 5266 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5267 vml[c]->vdev_top->vdev_ms_array) == 0); 5268 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5269 vml[c]->vdev_top->vdev_ms_shift) == 0); 5270 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5271 vml[c]->vdev_top->vdev_asize) == 0); 5272 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5273 vml[c]->vdev_top->vdev_ashift) == 0); 5274 } 5275 5276 if (error != 0) { 5277 kmem_free(vml, children * sizeof (vdev_t *)); 5278 kmem_free(glist, children * sizeof (uint64_t)); 5279 return (spa_vdev_exit(spa, NULL, txg, error)); 5280 } 5281 5282 /* stop writers from using the disks */ 5283 for (c = 0; c < children; c++) { 5284 if (vml[c] != NULL) 5285 vml[c]->vdev_offline = B_TRUE; 5286 } 5287 vdev_reopen(spa->spa_root_vdev); 5288 5289 /* 5290 * Temporarily record the splitting vdevs in the spa config. This 5291 * will disappear once the config is regenerated. 5292 */ 5293 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5294 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5295 glist, children) == 0); 5296 kmem_free(glist, children * sizeof (uint64_t)); 5297 5298 mutex_enter(&spa->spa_props_lock); 5299 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5300 nvl) == 0); 5301 mutex_exit(&spa->spa_props_lock); 5302 spa->spa_config_splitting = nvl; 5303 vdev_config_dirty(spa->spa_root_vdev); 5304 5305 /* configure and create the new pool */ 5306 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5307 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5308 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5309 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5310 spa_version(spa)) == 0); 5311 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5312 spa->spa_config_txg) == 0); 5313 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5314 spa_generate_guid(NULL)) == 0); 5315 (void) nvlist_lookup_string(props, 5316 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5317 5318 /* add the new pool to the namespace */ 5319 newspa = spa_add(newname, config, altroot); 5320 newspa->spa_config_txg = spa->spa_config_txg; 5321 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5322 5323 /* release the spa config lock, retaining the namespace lock */ 5324 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5325 5326 if (zio_injection_enabled) 5327 zio_handle_panic_injection(spa, FTAG, 1); 5328 5329 spa_activate(newspa, spa_mode_global); 5330 spa_async_suspend(newspa); 5331 5332#ifndef sun 5333 /* mark that we are creating new spa by splitting */ 5334 newspa->spa_splitting_newspa = B_TRUE; 5335#endif 5336 /* create the new pool from the disks of the original pool */ 5337 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5338#ifndef sun 5339 newspa->spa_splitting_newspa = B_FALSE; 5340#endif 5341 if (error) 5342 goto out; 5343 5344 /* if that worked, generate a real config for the new pool */ 5345 if (newspa->spa_root_vdev != NULL) { 5346 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5347 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5348 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5349 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5350 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5351 B_TRUE)); 5352 } 5353 5354 /* set the props */ 5355 if (props != NULL) { 5356 spa_configfile_set(newspa, props, B_FALSE); 5357 error = spa_prop_set(newspa, props); 5358 if (error) 5359 goto out; 5360 } 5361 5362 /* flush everything */ 5363 txg = spa_vdev_config_enter(newspa); 5364 vdev_config_dirty(newspa->spa_root_vdev); 5365 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5366 5367 if (zio_injection_enabled) 5368 zio_handle_panic_injection(spa, FTAG, 2); 5369 5370 spa_async_resume(newspa); 5371 5372 /* finally, update the original pool's config */ 5373 txg = spa_vdev_config_enter(spa); 5374 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5375 error = dmu_tx_assign(tx, TXG_WAIT); 5376 if (error != 0) 5377 dmu_tx_abort(tx); 5378 for (c = 0; c < children; c++) { 5379 if (vml[c] != NULL) { 5380 vdev_split(vml[c]); 5381 if (error == 0) 5382 spa_history_log_internal(spa, "detach", tx, 5383 "vdev=%s", vml[c]->vdev_path); 5384 vdev_free(vml[c]); 5385 } 5386 } 5387 vdev_config_dirty(spa->spa_root_vdev); 5388 spa->spa_config_splitting = NULL; 5389 nvlist_free(nvl); 5390 if (error == 0) 5391 dmu_tx_commit(tx); 5392 (void) spa_vdev_exit(spa, NULL, txg, 0); 5393 5394 if (zio_injection_enabled) 5395 zio_handle_panic_injection(spa, FTAG, 3); 5396 5397 /* split is complete; log a history record */ 5398 spa_history_log_internal(newspa, "split", NULL, 5399 "from pool %s", spa_name(spa)); 5400 5401 kmem_free(vml, children * sizeof (vdev_t *)); 5402 5403 /* if we're not going to mount the filesystems in userland, export */ 5404 if (exp) 5405 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5406 B_FALSE, B_FALSE); 5407 5408 return (error); 5409 5410out: 5411 spa_unload(newspa); 5412 spa_deactivate(newspa); 5413 spa_remove(newspa); 5414 5415 txg = spa_vdev_config_enter(spa); 5416 5417 /* re-online all offlined disks */ 5418 for (c = 0; c < children; c++) { 5419 if (vml[c] != NULL) 5420 vml[c]->vdev_offline = B_FALSE; 5421 } 5422 vdev_reopen(spa->spa_root_vdev); 5423 5424 nvlist_free(spa->spa_config_splitting); 5425 spa->spa_config_splitting = NULL; 5426 (void) spa_vdev_exit(spa, NULL, txg, error); 5427 5428 kmem_free(vml, children * sizeof (vdev_t *)); 5429 return (error); 5430} 5431 5432static nvlist_t * 5433spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5434{ 5435 for (int i = 0; i < count; i++) { 5436 uint64_t guid; 5437 5438 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5439 &guid) == 0); 5440 5441 if (guid == target_guid) 5442 return (nvpp[i]); 5443 } 5444 5445 return (NULL); 5446} 5447 5448static void 5449spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5450 nvlist_t *dev_to_remove) 5451{ 5452 nvlist_t **newdev = NULL; 5453 5454 if (count > 1) 5455 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5456 5457 for (int i = 0, j = 0; i < count; i++) { 5458 if (dev[i] == dev_to_remove) 5459 continue; 5460 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5461 } 5462 5463 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5464 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5465 5466 for (int i = 0; i < count - 1; i++) 5467 nvlist_free(newdev[i]); 5468 5469 if (count > 1) 5470 kmem_free(newdev, (count - 1) * sizeof (void *)); 5471} 5472 5473/* 5474 * Evacuate the device. 5475 */ 5476static int 5477spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5478{ 5479 uint64_t txg; 5480 int error = 0; 5481 5482 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5483 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5484 ASSERT(vd == vd->vdev_top); 5485 5486 /* 5487 * Evacuate the device. We don't hold the config lock as writer 5488 * since we need to do I/O but we do keep the 5489 * spa_namespace_lock held. Once this completes the device 5490 * should no longer have any blocks allocated on it. 5491 */ 5492 if (vd->vdev_islog) { 5493 if (vd->vdev_stat.vs_alloc != 0) 5494 error = spa_offline_log(spa); 5495 } else { 5496 error = SET_ERROR(ENOTSUP); 5497 } 5498 5499 if (error) 5500 return (error); 5501 5502 /* 5503 * The evacuation succeeded. Remove any remaining MOS metadata 5504 * associated with this vdev, and wait for these changes to sync. 5505 */ 5506 ASSERT0(vd->vdev_stat.vs_alloc); 5507 txg = spa_vdev_config_enter(spa); 5508 vd->vdev_removing = B_TRUE; 5509 vdev_dirty_leaves(vd, VDD_DTL, txg); 5510 vdev_config_dirty(vd); 5511 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5512 5513 return (0); 5514} 5515 5516/* 5517 * Complete the removal by cleaning up the namespace. 5518 */ 5519static void 5520spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5521{ 5522 vdev_t *rvd = spa->spa_root_vdev; 5523 uint64_t id = vd->vdev_id; 5524 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5525 5526 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5527 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5528 ASSERT(vd == vd->vdev_top); 5529 5530 /* 5531 * Only remove any devices which are empty. 5532 */ 5533 if (vd->vdev_stat.vs_alloc != 0) 5534 return; 5535 5536 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5537 5538 if (list_link_active(&vd->vdev_state_dirty_node)) 5539 vdev_state_clean(vd); 5540 if (list_link_active(&vd->vdev_config_dirty_node)) 5541 vdev_config_clean(vd); 5542 5543 vdev_free(vd); 5544 5545 if (last_vdev) { 5546 vdev_compact_children(rvd); 5547 } else { 5548 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5549 vdev_add_child(rvd, vd); 5550 } 5551 vdev_config_dirty(rvd); 5552 5553 /* 5554 * Reassess the health of our root vdev. 5555 */ 5556 vdev_reopen(rvd); 5557} 5558 5559/* 5560 * Remove a device from the pool - 5561 * 5562 * Removing a device from the vdev namespace requires several steps 5563 * and can take a significant amount of time. As a result we use 5564 * the spa_vdev_config_[enter/exit] functions which allow us to 5565 * grab and release the spa_config_lock while still holding the namespace 5566 * lock. During each step the configuration is synced out. 5567 * 5568 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5569 * devices. 5570 */ 5571int 5572spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5573{ 5574 vdev_t *vd; 5575 metaslab_group_t *mg; 5576 nvlist_t **spares, **l2cache, *nv; 5577 uint64_t txg = 0; 5578 uint_t nspares, nl2cache; 5579 int error = 0; 5580 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5581 5582 ASSERT(spa_writeable(spa)); 5583 5584 if (!locked) 5585 txg = spa_vdev_enter(spa); 5586 5587 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5588 5589 if (spa->spa_spares.sav_vdevs != NULL && 5590 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5591 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5592 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5593 /* 5594 * Only remove the hot spare if it's not currently in use 5595 * in this pool. 5596 */ 5597 if (vd == NULL || unspare) { 5598 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5599 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5600 spa_load_spares(spa); 5601 spa->spa_spares.sav_sync = B_TRUE; 5602 } else { 5603 error = SET_ERROR(EBUSY); 5604 } 5605 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5606 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5607 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5608 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5609 /* 5610 * Cache devices can always be removed. 5611 */ 5612 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5613 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5614 spa_load_l2cache(spa); 5615 spa->spa_l2cache.sav_sync = B_TRUE; 5616 } else if (vd != NULL && vd->vdev_islog) { 5617 ASSERT(!locked); 5618 ASSERT(vd == vd->vdev_top); 5619 5620 mg = vd->vdev_mg; 5621 5622 /* 5623 * Stop allocating from this vdev. 5624 */ 5625 metaslab_group_passivate(mg); 5626 5627 /* 5628 * Wait for the youngest allocations and frees to sync, 5629 * and then wait for the deferral of those frees to finish. 5630 */ 5631 spa_vdev_config_exit(spa, NULL, 5632 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5633 5634 /* 5635 * Attempt to evacuate the vdev. 5636 */ 5637 error = spa_vdev_remove_evacuate(spa, vd); 5638 5639 txg = spa_vdev_config_enter(spa); 5640 5641 /* 5642 * If we couldn't evacuate the vdev, unwind. 5643 */ 5644 if (error) { 5645 metaslab_group_activate(mg); 5646 return (spa_vdev_exit(spa, NULL, txg, error)); 5647 } 5648 5649 /* 5650 * Clean up the vdev namespace. 5651 */ 5652 spa_vdev_remove_from_namespace(spa, vd); 5653 5654 } else if (vd != NULL) { 5655 /* 5656 * Normal vdevs cannot be removed (yet). 5657 */ 5658 error = SET_ERROR(ENOTSUP); 5659 } else { 5660 /* 5661 * There is no vdev of any kind with the specified guid. 5662 */ 5663 error = SET_ERROR(ENOENT); 5664 } 5665 5666 if (!locked) 5667 return (spa_vdev_exit(spa, NULL, txg, error)); 5668 5669 return (error); 5670} 5671 5672/* 5673 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5674 * currently spared, so we can detach it. 5675 */ 5676static vdev_t * 5677spa_vdev_resilver_done_hunt(vdev_t *vd) 5678{ 5679 vdev_t *newvd, *oldvd; 5680 5681 for (int c = 0; c < vd->vdev_children; c++) { 5682 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5683 if (oldvd != NULL) 5684 return (oldvd); 5685 } 5686 5687 /* 5688 * Check for a completed replacement. We always consider the first 5689 * vdev in the list to be the oldest vdev, and the last one to be 5690 * the newest (see spa_vdev_attach() for how that works). In 5691 * the case where the newest vdev is faulted, we will not automatically 5692 * remove it after a resilver completes. This is OK as it will require 5693 * user intervention to determine which disk the admin wishes to keep. 5694 */ 5695 if (vd->vdev_ops == &vdev_replacing_ops) { 5696 ASSERT(vd->vdev_children > 1); 5697 5698 newvd = vd->vdev_child[vd->vdev_children - 1]; 5699 oldvd = vd->vdev_child[0]; 5700 5701 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5702 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5703 !vdev_dtl_required(oldvd)) 5704 return (oldvd); 5705 } 5706 5707 /* 5708 * Check for a completed resilver with the 'unspare' flag set. 5709 */ 5710 if (vd->vdev_ops == &vdev_spare_ops) { 5711 vdev_t *first = vd->vdev_child[0]; 5712 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5713 5714 if (last->vdev_unspare) { 5715 oldvd = first; 5716 newvd = last; 5717 } else if (first->vdev_unspare) { 5718 oldvd = last; 5719 newvd = first; 5720 } else { 5721 oldvd = NULL; 5722 } 5723 5724 if (oldvd != NULL && 5725 vdev_dtl_empty(newvd, DTL_MISSING) && 5726 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5727 !vdev_dtl_required(oldvd)) 5728 return (oldvd); 5729 5730 /* 5731 * If there are more than two spares attached to a disk, 5732 * and those spares are not required, then we want to 5733 * attempt to free them up now so that they can be used 5734 * by other pools. Once we're back down to a single 5735 * disk+spare, we stop removing them. 5736 */ 5737 if (vd->vdev_children > 2) { 5738 newvd = vd->vdev_child[1]; 5739 5740 if (newvd->vdev_isspare && last->vdev_isspare && 5741 vdev_dtl_empty(last, DTL_MISSING) && 5742 vdev_dtl_empty(last, DTL_OUTAGE) && 5743 !vdev_dtl_required(newvd)) 5744 return (newvd); 5745 } 5746 } 5747 5748 return (NULL); 5749} 5750 5751static void 5752spa_vdev_resilver_done(spa_t *spa) 5753{ 5754 vdev_t *vd, *pvd, *ppvd; 5755 uint64_t guid, sguid, pguid, ppguid; 5756 5757 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5758 5759 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5760 pvd = vd->vdev_parent; 5761 ppvd = pvd->vdev_parent; 5762 guid = vd->vdev_guid; 5763 pguid = pvd->vdev_guid; 5764 ppguid = ppvd->vdev_guid; 5765 sguid = 0; 5766 /* 5767 * If we have just finished replacing a hot spared device, then 5768 * we need to detach the parent's first child (the original hot 5769 * spare) as well. 5770 */ 5771 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5772 ppvd->vdev_children == 2) { 5773 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5774 sguid = ppvd->vdev_child[1]->vdev_guid; 5775 } 5776 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5777 5778 spa_config_exit(spa, SCL_ALL, FTAG); 5779 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5780 return; 5781 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5782 return; 5783 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5784 } 5785 5786 spa_config_exit(spa, SCL_ALL, FTAG); 5787} 5788 5789/* 5790 * Update the stored path or FRU for this vdev. 5791 */ 5792int 5793spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5794 boolean_t ispath) 5795{ 5796 vdev_t *vd; 5797 boolean_t sync = B_FALSE; 5798 5799 ASSERT(spa_writeable(spa)); 5800 5801 spa_vdev_state_enter(spa, SCL_ALL); 5802 5803 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5804 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5805 5806 if (!vd->vdev_ops->vdev_op_leaf) 5807 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5808 5809 if (ispath) { 5810 if (strcmp(value, vd->vdev_path) != 0) { 5811 spa_strfree(vd->vdev_path); 5812 vd->vdev_path = spa_strdup(value); 5813 sync = B_TRUE; 5814 } 5815 } else { 5816 if (vd->vdev_fru == NULL) { 5817 vd->vdev_fru = spa_strdup(value); 5818 sync = B_TRUE; 5819 } else if (strcmp(value, vd->vdev_fru) != 0) { 5820 spa_strfree(vd->vdev_fru); 5821 vd->vdev_fru = spa_strdup(value); 5822 sync = B_TRUE; 5823 } 5824 } 5825 5826 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5827} 5828 5829int 5830spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5831{ 5832 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5833} 5834 5835int 5836spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5837{ 5838 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5839} 5840 5841/* 5842 * ========================================================================== 5843 * SPA Scanning 5844 * ========================================================================== 5845 */ 5846 5847int 5848spa_scan_stop(spa_t *spa) 5849{ 5850 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5851 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5852 return (SET_ERROR(EBUSY)); 5853 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5854} 5855 5856int 5857spa_scan(spa_t *spa, pool_scan_func_t func) 5858{ 5859 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5860 5861 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5862 return (SET_ERROR(ENOTSUP)); 5863 5864 /* 5865 * If a resilver was requested, but there is no DTL on a 5866 * writeable leaf device, we have nothing to do. 5867 */ 5868 if (func == POOL_SCAN_RESILVER && 5869 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5870 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5871 return (0); 5872 } 5873 5874 return (dsl_scan(spa->spa_dsl_pool, func)); 5875} 5876 5877/* 5878 * ========================================================================== 5879 * SPA async task processing 5880 * ========================================================================== 5881 */ 5882 5883static void 5884spa_async_remove(spa_t *spa, vdev_t *vd) 5885{ 5886 if (vd->vdev_remove_wanted) { 5887 vd->vdev_remove_wanted = B_FALSE; 5888 vd->vdev_delayed_close = B_FALSE; 5889 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5890 5891 /* 5892 * We want to clear the stats, but we don't want to do a full 5893 * vdev_clear() as that will cause us to throw away 5894 * degraded/faulted state as well as attempt to reopen the 5895 * device, all of which is a waste. 5896 */ 5897 vd->vdev_stat.vs_read_errors = 0; 5898 vd->vdev_stat.vs_write_errors = 0; 5899 vd->vdev_stat.vs_checksum_errors = 0; 5900 5901 vdev_state_dirty(vd->vdev_top); 5902 } 5903 5904 for (int c = 0; c < vd->vdev_children; c++) 5905 spa_async_remove(spa, vd->vdev_child[c]); 5906} 5907 5908static void 5909spa_async_probe(spa_t *spa, vdev_t *vd) 5910{ 5911 if (vd->vdev_probe_wanted) { 5912 vd->vdev_probe_wanted = B_FALSE; 5913 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5914 } 5915 5916 for (int c = 0; c < vd->vdev_children; c++) 5917 spa_async_probe(spa, vd->vdev_child[c]); 5918} 5919 5920static void 5921spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5922{ 5923 sysevent_id_t eid; 5924 nvlist_t *attr; 5925 char *physpath; 5926 5927 if (!spa->spa_autoexpand) 5928 return; 5929 5930 for (int c = 0; c < vd->vdev_children; c++) { 5931 vdev_t *cvd = vd->vdev_child[c]; 5932 spa_async_autoexpand(spa, cvd); 5933 } 5934 5935 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5936 return; 5937 5938 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5939 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5940 5941 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5942 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5943 5944 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5945 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5946 5947 nvlist_free(attr); 5948 kmem_free(physpath, MAXPATHLEN); 5949} 5950 5951static void 5952spa_async_thread(void *arg) 5953{ 5954 spa_t *spa = arg; 5955 int tasks; 5956 5957 ASSERT(spa->spa_sync_on); 5958 5959 mutex_enter(&spa->spa_async_lock); 5960 tasks = spa->spa_async_tasks; 5961 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5962 mutex_exit(&spa->spa_async_lock); 5963 5964 /* 5965 * See if the config needs to be updated. 5966 */ 5967 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5968 uint64_t old_space, new_space; 5969 5970 mutex_enter(&spa_namespace_lock); 5971 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5972 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5973 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5974 mutex_exit(&spa_namespace_lock); 5975 5976 /* 5977 * If the pool grew as a result of the config update, 5978 * then log an internal history event. 5979 */ 5980 if (new_space != old_space) { 5981 spa_history_log_internal(spa, "vdev online", NULL, 5982 "pool '%s' size: %llu(+%llu)", 5983 spa_name(spa), new_space, new_space - old_space); 5984 } 5985 } 5986 5987 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5988 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5989 spa_async_autoexpand(spa, spa->spa_root_vdev); 5990 spa_config_exit(spa, SCL_CONFIG, FTAG); 5991 } 5992 5993 /* 5994 * See if any devices need to be probed. 5995 */ 5996 if (tasks & SPA_ASYNC_PROBE) { 5997 spa_vdev_state_enter(spa, SCL_NONE); 5998 spa_async_probe(spa, spa->spa_root_vdev); 5999 (void) spa_vdev_state_exit(spa, NULL, 0); 6000 } 6001 6002 /* 6003 * If any devices are done replacing, detach them. 6004 */ 6005 if (tasks & SPA_ASYNC_RESILVER_DONE) 6006 spa_vdev_resilver_done(spa); 6007 6008 /* 6009 * Kick off a resilver. 6010 */ 6011 if (tasks & SPA_ASYNC_RESILVER) 6012 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6013 6014 /* 6015 * Let the world know that we're done. 6016 */ 6017 mutex_enter(&spa->spa_async_lock); 6018 spa->spa_async_thread = NULL; 6019 cv_broadcast(&spa->spa_async_cv); 6020 mutex_exit(&spa->spa_async_lock); 6021 thread_exit(); 6022} 6023 6024static void 6025spa_async_thread_vd(void *arg) 6026{ 6027 spa_t *spa = arg; 6028 int tasks; 6029 6030 ASSERT(spa->spa_sync_on); 6031 6032 mutex_enter(&spa->spa_async_lock); 6033 tasks = spa->spa_async_tasks; 6034retry: 6035 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6036 mutex_exit(&spa->spa_async_lock); 6037 6038 /* 6039 * See if any devices need to be marked REMOVED. 6040 */ 6041 if (tasks & SPA_ASYNC_REMOVE) { 6042 spa_vdev_state_enter(spa, SCL_NONE); 6043 spa_async_remove(spa, spa->spa_root_vdev); 6044 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6045 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6046 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6047 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6048 (void) spa_vdev_state_exit(spa, NULL, 0); 6049 } 6050 6051 /* 6052 * Let the world know that we're done. 6053 */ 6054 mutex_enter(&spa->spa_async_lock); 6055 tasks = spa->spa_async_tasks; 6056 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6057 goto retry; 6058 spa->spa_async_thread_vd = NULL; 6059 cv_broadcast(&spa->spa_async_cv); 6060 mutex_exit(&spa->spa_async_lock); 6061 thread_exit(); 6062} 6063 6064void 6065spa_async_suspend(spa_t *spa) 6066{ 6067 mutex_enter(&spa->spa_async_lock); 6068 spa->spa_async_suspended++; 6069 while (spa->spa_async_thread != NULL && 6070 spa->spa_async_thread_vd != NULL) 6071 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6072 mutex_exit(&spa->spa_async_lock); 6073} 6074 6075void 6076spa_async_resume(spa_t *spa) 6077{ 6078 mutex_enter(&spa->spa_async_lock); 6079 ASSERT(spa->spa_async_suspended != 0); 6080 spa->spa_async_suspended--; 6081 mutex_exit(&spa->spa_async_lock); 6082} 6083 6084static boolean_t 6085spa_async_tasks_pending(spa_t *spa) 6086{ 6087 uint_t non_config_tasks; 6088 uint_t config_task; 6089 boolean_t config_task_suspended; 6090 6091 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6092 SPA_ASYNC_REMOVE); 6093 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6094 if (spa->spa_ccw_fail_time == 0) { 6095 config_task_suspended = B_FALSE; 6096 } else { 6097 config_task_suspended = 6098 (gethrtime() - spa->spa_ccw_fail_time) < 6099 (zfs_ccw_retry_interval * NANOSEC); 6100 } 6101 6102 return (non_config_tasks || (config_task && !config_task_suspended)); 6103} 6104 6105static void 6106spa_async_dispatch(spa_t *spa) 6107{ 6108 mutex_enter(&spa->spa_async_lock); 6109 if (spa_async_tasks_pending(spa) && 6110 !spa->spa_async_suspended && 6111 spa->spa_async_thread == NULL && 6112 rootdir != NULL) 6113 spa->spa_async_thread = thread_create(NULL, 0, 6114 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6115 mutex_exit(&spa->spa_async_lock); 6116} 6117 6118static void 6119spa_async_dispatch_vd(spa_t *spa) 6120{ 6121 mutex_enter(&spa->spa_async_lock); 6122 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6123 !spa->spa_async_suspended && 6124 spa->spa_async_thread_vd == NULL && 6125 rootdir != NULL) 6126 spa->spa_async_thread_vd = thread_create(NULL, 0, 6127 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6128 mutex_exit(&spa->spa_async_lock); 6129} 6130 6131void 6132spa_async_request(spa_t *spa, int task) 6133{ 6134 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6135 mutex_enter(&spa->spa_async_lock); 6136 spa->spa_async_tasks |= task; 6137 mutex_exit(&spa->spa_async_lock); 6138 spa_async_dispatch_vd(spa); 6139} 6140 6141/* 6142 * ========================================================================== 6143 * SPA syncing routines 6144 * ========================================================================== 6145 */ 6146 6147static int 6148bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6149{ 6150 bpobj_t *bpo = arg; 6151 bpobj_enqueue(bpo, bp, tx); 6152 return (0); 6153} 6154 6155static int 6156spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6157{ 6158 zio_t *zio = arg; 6159 6160 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6161 BP_GET_PSIZE(bp), zio->io_flags)); 6162 return (0); 6163} 6164 6165/* 6166 * Note: this simple function is not inlined to make it easier to dtrace the 6167 * amount of time spent syncing frees. 6168 */ 6169static void 6170spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6171{ 6172 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6173 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6174 VERIFY(zio_wait(zio) == 0); 6175} 6176 6177/* 6178 * Note: this simple function is not inlined to make it easier to dtrace the 6179 * amount of time spent syncing deferred frees. 6180 */ 6181static void 6182spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6183{ 6184 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6185 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6186 spa_free_sync_cb, zio, tx), ==, 0); 6187 VERIFY0(zio_wait(zio)); 6188} 6189 6190 6191static void 6192spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6193{ 6194 char *packed = NULL; 6195 size_t bufsize; 6196 size_t nvsize = 0; 6197 dmu_buf_t *db; 6198 6199 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6200 6201 /* 6202 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6203 * information. This avoids the dmu_buf_will_dirty() path and 6204 * saves us a pre-read to get data we don't actually care about. 6205 */ 6206 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6207 packed = kmem_alloc(bufsize, KM_SLEEP); 6208 6209 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6210 KM_SLEEP) == 0); 6211 bzero(packed + nvsize, bufsize - nvsize); 6212 6213 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6214 6215 kmem_free(packed, bufsize); 6216 6217 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6218 dmu_buf_will_dirty(db, tx); 6219 *(uint64_t *)db->db_data = nvsize; 6220 dmu_buf_rele(db, FTAG); 6221} 6222 6223static void 6224spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6225 const char *config, const char *entry) 6226{ 6227 nvlist_t *nvroot; 6228 nvlist_t **list; 6229 int i; 6230 6231 if (!sav->sav_sync) 6232 return; 6233 6234 /* 6235 * Update the MOS nvlist describing the list of available devices. 6236 * spa_validate_aux() will have already made sure this nvlist is 6237 * valid and the vdevs are labeled appropriately. 6238 */ 6239 if (sav->sav_object == 0) { 6240 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6241 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6242 sizeof (uint64_t), tx); 6243 VERIFY(zap_update(spa->spa_meta_objset, 6244 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6245 &sav->sav_object, tx) == 0); 6246 } 6247 6248 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6249 if (sav->sav_count == 0) { 6250 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6251 } else { 6252 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6253 for (i = 0; i < sav->sav_count; i++) 6254 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6255 B_FALSE, VDEV_CONFIG_L2CACHE); 6256 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6257 sav->sav_count) == 0); 6258 for (i = 0; i < sav->sav_count; i++) 6259 nvlist_free(list[i]); 6260 kmem_free(list, sav->sav_count * sizeof (void *)); 6261 } 6262 6263 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6264 nvlist_free(nvroot); 6265 6266 sav->sav_sync = B_FALSE; 6267} 6268 6269static void 6270spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6271{ 6272 nvlist_t *config; 6273 6274 if (list_is_empty(&spa->spa_config_dirty_list)) 6275 return; 6276 6277 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6278 6279 config = spa_config_generate(spa, spa->spa_root_vdev, 6280 dmu_tx_get_txg(tx), B_FALSE); 6281 6282 /* 6283 * If we're upgrading the spa version then make sure that 6284 * the config object gets updated with the correct version. 6285 */ 6286 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6287 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6288 spa->spa_uberblock.ub_version); 6289 6290 spa_config_exit(spa, SCL_STATE, FTAG); 6291 6292 if (spa->spa_config_syncing) 6293 nvlist_free(spa->spa_config_syncing); 6294 spa->spa_config_syncing = config; 6295 6296 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6297} 6298 6299static void 6300spa_sync_version(void *arg, dmu_tx_t *tx) 6301{ 6302 uint64_t *versionp = arg; 6303 uint64_t version = *versionp; 6304 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6305 6306 /* 6307 * Setting the version is special cased when first creating the pool. 6308 */ 6309 ASSERT(tx->tx_txg != TXG_INITIAL); 6310 6311 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6312 ASSERT(version >= spa_version(spa)); 6313 6314 spa->spa_uberblock.ub_version = version; 6315 vdev_config_dirty(spa->spa_root_vdev); 6316 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6317} 6318 6319/* 6320 * Set zpool properties. 6321 */ 6322static void 6323spa_sync_props(void *arg, dmu_tx_t *tx) 6324{ 6325 nvlist_t *nvp = arg; 6326 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6327 objset_t *mos = spa->spa_meta_objset; 6328 nvpair_t *elem = NULL; 6329 6330 mutex_enter(&spa->spa_props_lock); 6331 6332 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6333 uint64_t intval; 6334 char *strval, *fname; 6335 zpool_prop_t prop; 6336 const char *propname; 6337 zprop_type_t proptype; 6338 spa_feature_t fid; 6339 6340 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6341 case ZPROP_INVAL: 6342 /* 6343 * We checked this earlier in spa_prop_validate(). 6344 */ 6345 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6346 6347 fname = strchr(nvpair_name(elem), '@') + 1; 6348 VERIFY0(zfeature_lookup_name(fname, &fid)); 6349 6350 spa_feature_enable(spa, fid, tx); 6351 spa_history_log_internal(spa, "set", tx, 6352 "%s=enabled", nvpair_name(elem)); 6353 break; 6354 6355 case ZPOOL_PROP_VERSION: 6356 intval = fnvpair_value_uint64(elem); 6357 /* 6358 * The version is synced seperatly before other 6359 * properties and should be correct by now. 6360 */ 6361 ASSERT3U(spa_version(spa), >=, intval); 6362 break; 6363 6364 case ZPOOL_PROP_ALTROOT: 6365 /* 6366 * 'altroot' is a non-persistent property. It should 6367 * have been set temporarily at creation or import time. 6368 */ 6369 ASSERT(spa->spa_root != NULL); 6370 break; 6371 6372 case ZPOOL_PROP_READONLY: 6373 case ZPOOL_PROP_CACHEFILE: 6374 /* 6375 * 'readonly' and 'cachefile' are also non-persisitent 6376 * properties. 6377 */ 6378 break; 6379 case ZPOOL_PROP_COMMENT: 6380 strval = fnvpair_value_string(elem); 6381 if (spa->spa_comment != NULL) 6382 spa_strfree(spa->spa_comment); 6383 spa->spa_comment = spa_strdup(strval); 6384 /* 6385 * We need to dirty the configuration on all the vdevs 6386 * so that their labels get updated. It's unnecessary 6387 * to do this for pool creation since the vdev's 6388 * configuratoin has already been dirtied. 6389 */ 6390 if (tx->tx_txg != TXG_INITIAL) 6391 vdev_config_dirty(spa->spa_root_vdev); 6392 spa_history_log_internal(spa, "set", tx, 6393 "%s=%s", nvpair_name(elem), strval); 6394 break; 6395 default: 6396 /* 6397 * Set pool property values in the poolprops mos object. 6398 */ 6399 if (spa->spa_pool_props_object == 0) { 6400 spa->spa_pool_props_object = 6401 zap_create_link(mos, DMU_OT_POOL_PROPS, 6402 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6403 tx); 6404 } 6405 6406 /* normalize the property name */ 6407 propname = zpool_prop_to_name(prop); 6408 proptype = zpool_prop_get_type(prop); 6409 6410 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6411 ASSERT(proptype == PROP_TYPE_STRING); 6412 strval = fnvpair_value_string(elem); 6413 VERIFY0(zap_update(mos, 6414 spa->spa_pool_props_object, propname, 6415 1, strlen(strval) + 1, strval, tx)); 6416 spa_history_log_internal(spa, "set", tx, 6417 "%s=%s", nvpair_name(elem), strval); 6418 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6419 intval = fnvpair_value_uint64(elem); 6420 6421 if (proptype == PROP_TYPE_INDEX) { 6422 const char *unused; 6423 VERIFY0(zpool_prop_index_to_string( 6424 prop, intval, &unused)); 6425 } 6426 VERIFY0(zap_update(mos, 6427 spa->spa_pool_props_object, propname, 6428 8, 1, &intval, tx)); 6429 spa_history_log_internal(spa, "set", tx, 6430 "%s=%lld", nvpair_name(elem), intval); 6431 } else { 6432 ASSERT(0); /* not allowed */ 6433 } 6434 6435 switch (prop) { 6436 case ZPOOL_PROP_DELEGATION: 6437 spa->spa_delegation = intval; 6438 break; 6439 case ZPOOL_PROP_BOOTFS: 6440 spa->spa_bootfs = intval; 6441 break; 6442 case ZPOOL_PROP_FAILUREMODE: 6443 spa->spa_failmode = intval; 6444 break; 6445 case ZPOOL_PROP_AUTOEXPAND: 6446 spa->spa_autoexpand = intval; 6447 if (tx->tx_txg != TXG_INITIAL) 6448 spa_async_request(spa, 6449 SPA_ASYNC_AUTOEXPAND); 6450 break; 6451 case ZPOOL_PROP_DEDUPDITTO: 6452 spa->spa_dedup_ditto = intval; 6453 break; 6454 default: 6455 break; 6456 } 6457 } 6458 6459 } 6460 6461 mutex_exit(&spa->spa_props_lock); 6462} 6463 6464/* 6465 * Perform one-time upgrade on-disk changes. spa_version() does not 6466 * reflect the new version this txg, so there must be no changes this 6467 * txg to anything that the upgrade code depends on after it executes. 6468 * Therefore this must be called after dsl_pool_sync() does the sync 6469 * tasks. 6470 */ 6471static void 6472spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6473{ 6474 dsl_pool_t *dp = spa->spa_dsl_pool; 6475 6476 ASSERT(spa->spa_sync_pass == 1); 6477 6478 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6479 6480 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6481 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6482 dsl_pool_create_origin(dp, tx); 6483 6484 /* Keeping the origin open increases spa_minref */ 6485 spa->spa_minref += 3; 6486 } 6487 6488 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6489 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6490 dsl_pool_upgrade_clones(dp, tx); 6491 } 6492 6493 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6494 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6495 dsl_pool_upgrade_dir_clones(dp, tx); 6496 6497 /* Keeping the freedir open increases spa_minref */ 6498 spa->spa_minref += 3; 6499 } 6500 6501 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6502 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6503 spa_feature_create_zap_objects(spa, tx); 6504 } 6505 6506 /* 6507 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6508 * when possibility to use lz4 compression for metadata was added 6509 * Old pools that have this feature enabled must be upgraded to have 6510 * this feature active 6511 */ 6512 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6513 boolean_t lz4_en = spa_feature_is_enabled(spa, 6514 SPA_FEATURE_LZ4_COMPRESS); 6515 boolean_t lz4_ac = spa_feature_is_active(spa, 6516 SPA_FEATURE_LZ4_COMPRESS); 6517 6518 if (lz4_en && !lz4_ac) 6519 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6520 } 6521 rrw_exit(&dp->dp_config_rwlock, FTAG); 6522} 6523 6524/* 6525 * Sync the specified transaction group. New blocks may be dirtied as 6526 * part of the process, so we iterate until it converges. 6527 */ 6528void 6529spa_sync(spa_t *spa, uint64_t txg) 6530{ 6531 dsl_pool_t *dp = spa->spa_dsl_pool; 6532 objset_t *mos = spa->spa_meta_objset; 6533 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6534 vdev_t *rvd = spa->spa_root_vdev; 6535 vdev_t *vd; 6536 dmu_tx_t *tx; 6537 int error; 6538 6539 VERIFY(spa_writeable(spa)); 6540 6541 /* 6542 * Lock out configuration changes. 6543 */ 6544 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6545 6546 spa->spa_syncing_txg = txg; 6547 spa->spa_sync_pass = 0; 6548 6549 /* 6550 * If there are any pending vdev state changes, convert them 6551 * into config changes that go out with this transaction group. 6552 */ 6553 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6554 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6555 /* 6556 * We need the write lock here because, for aux vdevs, 6557 * calling vdev_config_dirty() modifies sav_config. 6558 * This is ugly and will become unnecessary when we 6559 * eliminate the aux vdev wart by integrating all vdevs 6560 * into the root vdev tree. 6561 */ 6562 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6563 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6564 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6565 vdev_state_clean(vd); 6566 vdev_config_dirty(vd); 6567 } 6568 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6569 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6570 } 6571 spa_config_exit(spa, SCL_STATE, FTAG); 6572 6573 tx = dmu_tx_create_assigned(dp, txg); 6574 6575 spa->spa_sync_starttime = gethrtime(); 6576#ifdef illumos 6577 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6578 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6579#else /* FreeBSD */ 6580#ifdef _KERNEL 6581 callout_reset(&spa->spa_deadman_cycid, 6582 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6583#endif 6584#endif 6585 6586 /* 6587 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6588 * set spa_deflate if we have no raid-z vdevs. 6589 */ 6590 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6591 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6592 int i; 6593 6594 for (i = 0; i < rvd->vdev_children; i++) { 6595 vd = rvd->vdev_child[i]; 6596 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6597 break; 6598 } 6599 if (i == rvd->vdev_children) { 6600 spa->spa_deflate = TRUE; 6601 VERIFY(0 == zap_add(spa->spa_meta_objset, 6602 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6603 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6604 } 6605 } 6606 6607 /* 6608 * Iterate to convergence. 6609 */ 6610 do { 6611 int pass = ++spa->spa_sync_pass; 6612 6613 spa_sync_config_object(spa, tx); 6614 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6615 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6616 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6617 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6618 spa_errlog_sync(spa, txg); 6619 dsl_pool_sync(dp, txg); 6620 6621 if (pass < zfs_sync_pass_deferred_free) { 6622 spa_sync_frees(spa, free_bpl, tx); 6623 } else { 6624 /* 6625 * We can not defer frees in pass 1, because 6626 * we sync the deferred frees later in pass 1. 6627 */ 6628 ASSERT3U(pass, >, 1); 6629 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6630 &spa->spa_deferred_bpobj, tx); 6631 } 6632 6633 ddt_sync(spa, txg); 6634 dsl_scan_sync(dp, tx); 6635 6636 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6637 vdev_sync(vd, txg); 6638 6639 if (pass == 1) { 6640 spa_sync_upgrades(spa, tx); 6641 ASSERT3U(txg, >=, 6642 spa->spa_uberblock.ub_rootbp.blk_birth); 6643 /* 6644 * Note: We need to check if the MOS is dirty 6645 * because we could have marked the MOS dirty 6646 * without updating the uberblock (e.g. if we 6647 * have sync tasks but no dirty user data). We 6648 * need to check the uberblock's rootbp because 6649 * it is updated if we have synced out dirty 6650 * data (though in this case the MOS will most 6651 * likely also be dirty due to second order 6652 * effects, we don't want to rely on that here). 6653 */ 6654 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6655 !dmu_objset_is_dirty(mos, txg)) { 6656 /* 6657 * Nothing changed on the first pass, 6658 * therefore this TXG is a no-op. Avoid 6659 * syncing deferred frees, so that we 6660 * can keep this TXG as a no-op. 6661 */ 6662 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6663 txg)); 6664 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6665 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6666 break; 6667 } 6668 spa_sync_deferred_frees(spa, tx); 6669 } 6670 6671 } while (dmu_objset_is_dirty(mos, txg)); 6672 6673 /* 6674 * Rewrite the vdev configuration (which includes the uberblock) 6675 * to commit the transaction group. 6676 * 6677 * If there are no dirty vdevs, we sync the uberblock to a few 6678 * random top-level vdevs that are known to be visible in the 6679 * config cache (see spa_vdev_add() for a complete description). 6680 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6681 */ 6682 for (;;) { 6683 /* 6684 * We hold SCL_STATE to prevent vdev open/close/etc. 6685 * while we're attempting to write the vdev labels. 6686 */ 6687 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6688 6689 if (list_is_empty(&spa->spa_config_dirty_list)) { 6690 vdev_t *svd[SPA_DVAS_PER_BP]; 6691 int svdcount = 0; 6692 int children = rvd->vdev_children; 6693 int c0 = spa_get_random(children); 6694 6695 for (int c = 0; c < children; c++) { 6696 vd = rvd->vdev_child[(c0 + c) % children]; 6697 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6698 continue; 6699 svd[svdcount++] = vd; 6700 if (svdcount == SPA_DVAS_PER_BP) 6701 break; 6702 } 6703 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6704 if (error != 0) 6705 error = vdev_config_sync(svd, svdcount, txg, 6706 B_TRUE); 6707 } else { 6708 error = vdev_config_sync(rvd->vdev_child, 6709 rvd->vdev_children, txg, B_FALSE); 6710 if (error != 0) 6711 error = vdev_config_sync(rvd->vdev_child, 6712 rvd->vdev_children, txg, B_TRUE); 6713 } 6714 6715 if (error == 0) 6716 spa->spa_last_synced_guid = rvd->vdev_guid; 6717 6718 spa_config_exit(spa, SCL_STATE, FTAG); 6719 6720 if (error == 0) 6721 break; 6722 zio_suspend(spa, NULL); 6723 zio_resume_wait(spa); 6724 } 6725 dmu_tx_commit(tx); 6726 6727#ifdef illumos 6728 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6729#else /* FreeBSD */ 6730#ifdef _KERNEL 6731 callout_drain(&spa->spa_deadman_cycid); 6732#endif 6733#endif 6734 6735 /* 6736 * Clear the dirty config list. 6737 */ 6738 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6739 vdev_config_clean(vd); 6740 6741 /* 6742 * Now that the new config has synced transactionally, 6743 * let it become visible to the config cache. 6744 */ 6745 if (spa->spa_config_syncing != NULL) { 6746 spa_config_set(spa, spa->spa_config_syncing); 6747 spa->spa_config_txg = txg; 6748 spa->spa_config_syncing = NULL; 6749 } 6750 6751 spa->spa_ubsync = spa->spa_uberblock; 6752 6753 dsl_pool_sync_done(dp, txg); 6754 6755 /* 6756 * Update usable space statistics. 6757 */ 6758 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6759 vdev_sync_done(vd, txg); 6760 6761 spa_update_dspace(spa); 6762 6763 /* 6764 * It had better be the case that we didn't dirty anything 6765 * since vdev_config_sync(). 6766 */ 6767 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6768 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6769 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6770 6771 spa->spa_sync_pass = 0; 6772 6773 spa_config_exit(spa, SCL_CONFIG, FTAG); 6774 6775 spa_handle_ignored_writes(spa); 6776 6777 /* 6778 * If any async tasks have been requested, kick them off. 6779 */ 6780 spa_async_dispatch(spa); 6781 spa_async_dispatch_vd(spa); 6782} 6783 6784/* 6785 * Sync all pools. We don't want to hold the namespace lock across these 6786 * operations, so we take a reference on the spa_t and drop the lock during the 6787 * sync. 6788 */ 6789void 6790spa_sync_allpools(void) 6791{ 6792 spa_t *spa = NULL; 6793 mutex_enter(&spa_namespace_lock); 6794 while ((spa = spa_next(spa)) != NULL) { 6795 if (spa_state(spa) != POOL_STATE_ACTIVE || 6796 !spa_writeable(spa) || spa_suspended(spa)) 6797 continue; 6798 spa_open_ref(spa, FTAG); 6799 mutex_exit(&spa_namespace_lock); 6800 txg_wait_synced(spa_get_dsl(spa), 0); 6801 mutex_enter(&spa_namespace_lock); 6802 spa_close(spa, FTAG); 6803 } 6804 mutex_exit(&spa_namespace_lock); 6805} 6806 6807/* 6808 * ========================================================================== 6809 * Miscellaneous routines 6810 * ========================================================================== 6811 */ 6812 6813/* 6814 * Remove all pools in the system. 6815 */ 6816void 6817spa_evict_all(void) 6818{ 6819 spa_t *spa; 6820 6821 /* 6822 * Remove all cached state. All pools should be closed now, 6823 * so every spa in the AVL tree should be unreferenced. 6824 */ 6825 mutex_enter(&spa_namespace_lock); 6826 while ((spa = spa_next(NULL)) != NULL) { 6827 /* 6828 * Stop async tasks. The async thread may need to detach 6829 * a device that's been replaced, which requires grabbing 6830 * spa_namespace_lock, so we must drop it here. 6831 */ 6832 spa_open_ref(spa, FTAG); 6833 mutex_exit(&spa_namespace_lock); 6834 spa_async_suspend(spa); 6835 mutex_enter(&spa_namespace_lock); 6836 spa_close(spa, FTAG); 6837 6838 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6839 spa_unload(spa); 6840 spa_deactivate(spa); 6841 } 6842 spa_remove(spa); 6843 } 6844 mutex_exit(&spa_namespace_lock); 6845} 6846 6847vdev_t * 6848spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6849{ 6850 vdev_t *vd; 6851 int i; 6852 6853 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6854 return (vd); 6855 6856 if (aux) { 6857 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6858 vd = spa->spa_l2cache.sav_vdevs[i]; 6859 if (vd->vdev_guid == guid) 6860 return (vd); 6861 } 6862 6863 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6864 vd = spa->spa_spares.sav_vdevs[i]; 6865 if (vd->vdev_guid == guid) 6866 return (vd); 6867 } 6868 } 6869 6870 return (NULL); 6871} 6872 6873void 6874spa_upgrade(spa_t *spa, uint64_t version) 6875{ 6876 ASSERT(spa_writeable(spa)); 6877 6878 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6879 6880 /* 6881 * This should only be called for a non-faulted pool, and since a 6882 * future version would result in an unopenable pool, this shouldn't be 6883 * possible. 6884 */ 6885 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6886 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6887 6888 spa->spa_uberblock.ub_version = version; 6889 vdev_config_dirty(spa->spa_root_vdev); 6890 6891 spa_config_exit(spa, SCL_ALL, FTAG); 6892 6893 txg_wait_synced(spa_get_dsl(spa), 0); 6894} 6895 6896boolean_t 6897spa_has_spare(spa_t *spa, uint64_t guid) 6898{ 6899 int i; 6900 uint64_t spareguid; 6901 spa_aux_vdev_t *sav = &spa->spa_spares; 6902 6903 for (i = 0; i < sav->sav_count; i++) 6904 if (sav->sav_vdevs[i]->vdev_guid == guid) 6905 return (B_TRUE); 6906 6907 for (i = 0; i < sav->sav_npending; i++) { 6908 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6909 &spareguid) == 0 && spareguid == guid) 6910 return (B_TRUE); 6911 } 6912 6913 return (B_FALSE); 6914} 6915 6916/* 6917 * Check if a pool has an active shared spare device. 6918 * Note: reference count of an active spare is 2, as a spare and as a replace 6919 */ 6920static boolean_t 6921spa_has_active_shared_spare(spa_t *spa) 6922{ 6923 int i, refcnt; 6924 uint64_t pool; 6925 spa_aux_vdev_t *sav = &spa->spa_spares; 6926 6927 for (i = 0; i < sav->sav_count; i++) { 6928 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6929 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6930 refcnt > 2) 6931 return (B_TRUE); 6932 } 6933 6934 return (B_FALSE); 6935} 6936 6937/* 6938 * Post a sysevent corresponding to the given event. The 'name' must be one of 6939 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6940 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6941 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6942 * or zdb as real changes. 6943 */ 6944void 6945spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6946{ 6947#ifdef _KERNEL 6948 sysevent_t *ev; 6949 sysevent_attr_list_t *attr = NULL; 6950 sysevent_value_t value; 6951 sysevent_id_t eid; 6952 6953 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6954 SE_SLEEP); 6955 6956 value.value_type = SE_DATA_TYPE_STRING; 6957 value.value.sv_string = spa_name(spa); 6958 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6959 goto done; 6960 6961 value.value_type = SE_DATA_TYPE_UINT64; 6962 value.value.sv_uint64 = spa_guid(spa); 6963 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6964 goto done; 6965 6966 if (vd) { 6967 value.value_type = SE_DATA_TYPE_UINT64; 6968 value.value.sv_uint64 = vd->vdev_guid; 6969 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6970 SE_SLEEP) != 0) 6971 goto done; 6972 6973 if (vd->vdev_path) { 6974 value.value_type = SE_DATA_TYPE_STRING; 6975 value.value.sv_string = vd->vdev_path; 6976 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6977 &value, SE_SLEEP) != 0) 6978 goto done; 6979 } 6980 } 6981 6982 if (sysevent_attach_attributes(ev, attr) != 0) 6983 goto done; 6984 attr = NULL; 6985 6986 (void) log_sysevent(ev, SE_SLEEP, &eid); 6987 6988done: 6989 if (attr) 6990 sysevent_free_attr(attr); 6991 sysevent_free(ev); 6992#endif 6993} 6994