spa.c revision 272598
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 ((dp = list_head(&spa->spa_config_list)) != NULL) { 291 if (dp->scd_path == NULL) { 292 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 293 "none", 0, ZPROP_SRC_LOCAL); 294 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 295 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 296 dp->scd_path, 0, ZPROP_SRC_LOCAL); 297 } 298 } 299} 300 301/* 302 * Get zpool property values. 303 */ 304int 305spa_prop_get(spa_t *spa, nvlist_t **nvp) 306{ 307 objset_t *mos = spa->spa_meta_objset; 308 zap_cursor_t zc; 309 zap_attribute_t za; 310 int err; 311 312 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 313 314 mutex_enter(&spa->spa_props_lock); 315 316 /* 317 * Get properties from the spa config. 318 */ 319 spa_prop_get_config(spa, nvp); 320 321 /* If no pool property object, no more prop to get. */ 322 if (mos == NULL || spa->spa_pool_props_object == 0) { 323 mutex_exit(&spa->spa_props_lock); 324 return (0); 325 } 326 327 /* 328 * Get properties from the MOS pool property object. 329 */ 330 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 331 (err = zap_cursor_retrieve(&zc, &za)) == 0; 332 zap_cursor_advance(&zc)) { 333 uint64_t intval = 0; 334 char *strval = NULL; 335 zprop_source_t src = ZPROP_SRC_DEFAULT; 336 zpool_prop_t prop; 337 338 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 339 continue; 340 341 switch (za.za_integer_length) { 342 case 8: 343 /* integer property */ 344 if (za.za_first_integer != 345 zpool_prop_default_numeric(prop)) 346 src = ZPROP_SRC_LOCAL; 347 348 if (prop == ZPOOL_PROP_BOOTFS) { 349 dsl_pool_t *dp; 350 dsl_dataset_t *ds = NULL; 351 352 dp = spa_get_dsl(spa); 353 dsl_pool_config_enter(dp, FTAG); 354 if (err = dsl_dataset_hold_obj(dp, 355 za.za_first_integer, FTAG, &ds)) { 356 dsl_pool_config_exit(dp, FTAG); 357 break; 358 } 359 360 strval = kmem_alloc( 361 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 362 KM_SLEEP); 363 dsl_dataset_name(ds, strval); 364 dsl_dataset_rele(ds, FTAG); 365 dsl_pool_config_exit(dp, FTAG); 366 } else { 367 strval = NULL; 368 intval = za.za_first_integer; 369 } 370 371 spa_prop_add_list(*nvp, prop, strval, intval, src); 372 373 if (strval != NULL) 374 kmem_free(strval, 375 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 376 377 break; 378 379 case 1: 380 /* string property */ 381 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 382 err = zap_lookup(mos, spa->spa_pool_props_object, 383 za.za_name, 1, za.za_num_integers, strval); 384 if (err) { 385 kmem_free(strval, za.za_num_integers); 386 break; 387 } 388 spa_prop_add_list(*nvp, prop, strval, 0, src); 389 kmem_free(strval, za.za_num_integers); 390 break; 391 392 default: 393 break; 394 } 395 } 396 zap_cursor_fini(&zc); 397 mutex_exit(&spa->spa_props_lock); 398out: 399 if (err && err != ENOENT) { 400 nvlist_free(*nvp); 401 *nvp = NULL; 402 return (err); 403 } 404 405 return (0); 406} 407 408/* 409 * Validate the given pool properties nvlist and modify the list 410 * for the property values to be set. 411 */ 412static int 413spa_prop_validate(spa_t *spa, nvlist_t *props) 414{ 415 nvpair_t *elem; 416 int error = 0, reset_bootfs = 0; 417 uint64_t objnum = 0; 418 boolean_t has_feature = B_FALSE; 419 420 elem = NULL; 421 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 422 uint64_t intval; 423 char *strval, *slash, *check, *fname; 424 const char *propname = nvpair_name(elem); 425 zpool_prop_t prop = zpool_name_to_prop(propname); 426 427 switch (prop) { 428 case ZPROP_INVAL: 429 if (!zpool_prop_feature(propname)) { 430 error = SET_ERROR(EINVAL); 431 break; 432 } 433 434 /* 435 * Sanitize the input. 436 */ 437 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 438 error = SET_ERROR(EINVAL); 439 break; 440 } 441 442 if (nvpair_value_uint64(elem, &intval) != 0) { 443 error = SET_ERROR(EINVAL); 444 break; 445 } 446 447 if (intval != 0) { 448 error = SET_ERROR(EINVAL); 449 break; 450 } 451 452 fname = strchr(propname, '@') + 1; 453 if (zfeature_lookup_name(fname, NULL) != 0) { 454 error = SET_ERROR(EINVAL); 455 break; 456 } 457 458 has_feature = B_TRUE; 459 break; 460 461 case ZPOOL_PROP_VERSION: 462 error = nvpair_value_uint64(elem, &intval); 463 if (!error && 464 (intval < spa_version(spa) || 465 intval > SPA_VERSION_BEFORE_FEATURES || 466 has_feature)) 467 error = SET_ERROR(EINVAL); 468 break; 469 470 case ZPOOL_PROP_DELEGATION: 471 case ZPOOL_PROP_AUTOREPLACE: 472 case ZPOOL_PROP_LISTSNAPS: 473 case ZPOOL_PROP_AUTOEXPAND: 474 error = nvpair_value_uint64(elem, &intval); 475 if (!error && intval > 1) 476 error = SET_ERROR(EINVAL); 477 break; 478 479 case ZPOOL_PROP_BOOTFS: 480 /* 481 * If the pool version is less than SPA_VERSION_BOOTFS, 482 * or the pool is still being created (version == 0), 483 * the bootfs property cannot be set. 484 */ 485 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 486 error = SET_ERROR(ENOTSUP); 487 break; 488 } 489 490 /* 491 * Make sure the vdev config is bootable 492 */ 493 if (!vdev_is_bootable(spa->spa_root_vdev)) { 494 error = SET_ERROR(ENOTSUP); 495 break; 496 } 497 498 reset_bootfs = 1; 499 500 error = nvpair_value_string(elem, &strval); 501 502 if (!error) { 503 objset_t *os; 504 uint64_t compress; 505 506 if (strval == NULL || strval[0] == '\0') { 507 objnum = zpool_prop_default_numeric( 508 ZPOOL_PROP_BOOTFS); 509 break; 510 } 511 512 if (error = dmu_objset_hold(strval, FTAG, &os)) 513 break; 514 515 /* Must be ZPL and not gzip compressed. */ 516 517 if (dmu_objset_type(os) != DMU_OST_ZFS) { 518 error = SET_ERROR(ENOTSUP); 519 } else if ((error = 520 dsl_prop_get_int_ds(dmu_objset_ds(os), 521 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 522 &compress)) == 0 && 523 !BOOTFS_COMPRESS_VALID(compress)) { 524 error = SET_ERROR(ENOTSUP); 525 } else { 526 objnum = dmu_objset_id(os); 527 } 528 dmu_objset_rele(os, FTAG); 529 } 530 break; 531 532 case ZPOOL_PROP_FAILUREMODE: 533 error = nvpair_value_uint64(elem, &intval); 534 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 535 intval > ZIO_FAILURE_MODE_PANIC)) 536 error = SET_ERROR(EINVAL); 537 538 /* 539 * This is a special case which only occurs when 540 * the pool has completely failed. This allows 541 * the user to change the in-core failmode property 542 * without syncing it out to disk (I/Os might 543 * currently be blocked). We do this by returning 544 * EIO to the caller (spa_prop_set) to trick it 545 * into thinking we encountered a property validation 546 * error. 547 */ 548 if (!error && spa_suspended(spa)) { 549 spa->spa_failmode = intval; 550 error = SET_ERROR(EIO); 551 } 552 break; 553 554 case ZPOOL_PROP_CACHEFILE: 555 if ((error = nvpair_value_string(elem, &strval)) != 0) 556 break; 557 558 if (strval[0] == '\0') 559 break; 560 561 if (strcmp(strval, "none") == 0) 562 break; 563 564 if (strval[0] != '/') { 565 error = SET_ERROR(EINVAL); 566 break; 567 } 568 569 slash = strrchr(strval, '/'); 570 ASSERT(slash != NULL); 571 572 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 573 strcmp(slash, "/..") == 0) 574 error = SET_ERROR(EINVAL); 575 break; 576 577 case ZPOOL_PROP_COMMENT: 578 if ((error = nvpair_value_string(elem, &strval)) != 0) 579 break; 580 for (check = strval; *check != '\0'; check++) { 581 /* 582 * The kernel doesn't have an easy isprint() 583 * check. For this kernel check, we merely 584 * check ASCII apart from DEL. Fix this if 585 * there is an easy-to-use kernel isprint(). 586 */ 587 if (*check >= 0x7f) { 588 error = SET_ERROR(EINVAL); 589 break; 590 } 591 check++; 592 } 593 if (strlen(strval) > ZPROP_MAX_COMMENT) 594 error = E2BIG; 595 break; 596 597 case ZPOOL_PROP_DEDUPDITTO: 598 if (spa_version(spa) < SPA_VERSION_DEDUP) 599 error = SET_ERROR(ENOTSUP); 600 else 601 error = nvpair_value_uint64(elem, &intval); 602 if (error == 0 && 603 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 604 error = SET_ERROR(EINVAL); 605 break; 606 } 607 608 if (error) 609 break; 610 } 611 612 if (!error && reset_bootfs) { 613 error = nvlist_remove(props, 614 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 615 616 if (!error) { 617 error = nvlist_add_uint64(props, 618 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 619 } 620 } 621 622 return (error); 623} 624 625void 626spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 627{ 628 char *cachefile; 629 spa_config_dirent_t *dp; 630 631 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 632 &cachefile) != 0) 633 return; 634 635 dp = kmem_alloc(sizeof (spa_config_dirent_t), 636 KM_SLEEP); 637 638 if (cachefile[0] == '\0') 639 dp->scd_path = spa_strdup(spa_config_path); 640 else if (strcmp(cachefile, "none") == 0) 641 dp->scd_path = NULL; 642 else 643 dp->scd_path = spa_strdup(cachefile); 644 645 list_insert_head(&spa->spa_config_list, dp); 646 if (need_sync) 647 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 648} 649 650int 651spa_prop_set(spa_t *spa, nvlist_t *nvp) 652{ 653 int error; 654 nvpair_t *elem = NULL; 655 boolean_t need_sync = B_FALSE; 656 657 if ((error = spa_prop_validate(spa, nvp)) != 0) 658 return (error); 659 660 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 661 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 662 663 if (prop == ZPOOL_PROP_CACHEFILE || 664 prop == ZPOOL_PROP_ALTROOT || 665 prop == ZPOOL_PROP_READONLY) 666 continue; 667 668 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 669 uint64_t ver; 670 671 if (prop == ZPOOL_PROP_VERSION) { 672 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 673 } else { 674 ASSERT(zpool_prop_feature(nvpair_name(elem))); 675 ver = SPA_VERSION_FEATURES; 676 need_sync = B_TRUE; 677 } 678 679 /* Save time if the version is already set. */ 680 if (ver == spa_version(spa)) 681 continue; 682 683 /* 684 * In addition to the pool directory object, we might 685 * create the pool properties object, the features for 686 * read object, the features for write object, or the 687 * feature descriptions object. 688 */ 689 error = dsl_sync_task(spa->spa_name, NULL, 690 spa_sync_version, &ver, 691 6, ZFS_SPACE_CHECK_RESERVED); 692 if (error) 693 return (error); 694 continue; 695 } 696 697 need_sync = B_TRUE; 698 break; 699 } 700 701 if (need_sync) { 702 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 703 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 704 } 705 706 return (0); 707} 708 709/* 710 * If the bootfs property value is dsobj, clear it. 711 */ 712void 713spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 714{ 715 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 716 VERIFY(zap_remove(spa->spa_meta_objset, 717 spa->spa_pool_props_object, 718 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 719 spa->spa_bootfs = 0; 720 } 721} 722 723/*ARGSUSED*/ 724static int 725spa_change_guid_check(void *arg, dmu_tx_t *tx) 726{ 727 uint64_t *newguid = arg; 728 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 729 vdev_t *rvd = spa->spa_root_vdev; 730 uint64_t vdev_state; 731 732 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 733 vdev_state = rvd->vdev_state; 734 spa_config_exit(spa, SCL_STATE, FTAG); 735 736 if (vdev_state != VDEV_STATE_HEALTHY) 737 return (SET_ERROR(ENXIO)); 738 739 ASSERT3U(spa_guid(spa), !=, *newguid); 740 741 return (0); 742} 743 744static void 745spa_change_guid_sync(void *arg, dmu_tx_t *tx) 746{ 747 uint64_t *newguid = arg; 748 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 749 uint64_t oldguid; 750 vdev_t *rvd = spa->spa_root_vdev; 751 752 oldguid = spa_guid(spa); 753 754 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 755 rvd->vdev_guid = *newguid; 756 rvd->vdev_guid_sum += (*newguid - oldguid); 757 vdev_config_dirty(rvd); 758 spa_config_exit(spa, SCL_STATE, FTAG); 759 760 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 761 oldguid, *newguid); 762} 763 764/* 765 * Change the GUID for the pool. This is done so that we can later 766 * re-import a pool built from a clone of our own vdevs. We will modify 767 * the root vdev's guid, our own pool guid, and then mark all of our 768 * vdevs dirty. Note that we must make sure that all our vdevs are 769 * online when we do this, or else any vdevs that weren't present 770 * would be orphaned from our pool. We are also going to issue a 771 * sysevent to update any watchers. 772 */ 773int 774spa_change_guid(spa_t *spa) 775{ 776 int error; 777 uint64_t guid; 778 779 mutex_enter(&spa->spa_vdev_top_lock); 780 mutex_enter(&spa_namespace_lock); 781 guid = spa_generate_guid(NULL); 782 783 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 784 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 785 786 if (error == 0) { 787 spa_config_sync(spa, B_FALSE, B_TRUE); 788 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 789 } 790 791 mutex_exit(&spa_namespace_lock); 792 mutex_exit(&spa->spa_vdev_top_lock); 793 794 return (error); 795} 796 797/* 798 * ========================================================================== 799 * SPA state manipulation (open/create/destroy/import/export) 800 * ========================================================================== 801 */ 802 803static int 804spa_error_entry_compare(const void *a, const void *b) 805{ 806 spa_error_entry_t *sa = (spa_error_entry_t *)a; 807 spa_error_entry_t *sb = (spa_error_entry_t *)b; 808 int ret; 809 810 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 811 sizeof (zbookmark_phys_t)); 812 813 if (ret < 0) 814 return (-1); 815 else if (ret > 0) 816 return (1); 817 else 818 return (0); 819} 820 821/* 822 * Utility function which retrieves copies of the current logs and 823 * re-initializes them in the process. 824 */ 825void 826spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 827{ 828 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 829 830 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 831 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 832 833 avl_create(&spa->spa_errlist_scrub, 834 spa_error_entry_compare, sizeof (spa_error_entry_t), 835 offsetof(spa_error_entry_t, se_avl)); 836 avl_create(&spa->spa_errlist_last, 837 spa_error_entry_compare, sizeof (spa_error_entry_t), 838 offsetof(spa_error_entry_t, se_avl)); 839} 840 841static void 842spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 843{ 844 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 845 enum zti_modes mode = ztip->zti_mode; 846 uint_t value = ztip->zti_value; 847 uint_t count = ztip->zti_count; 848 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 849 char name[32]; 850 uint_t flags = 0; 851 boolean_t batch = B_FALSE; 852 853 if (mode == ZTI_MODE_NULL) { 854 tqs->stqs_count = 0; 855 tqs->stqs_taskq = NULL; 856 return; 857 } 858 859 ASSERT3U(count, >, 0); 860 861 tqs->stqs_count = count; 862 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 863 864 switch (mode) { 865 case ZTI_MODE_FIXED: 866 ASSERT3U(value, >=, 1); 867 value = MAX(value, 1); 868 break; 869 870 case ZTI_MODE_BATCH: 871 batch = B_TRUE; 872 flags |= TASKQ_THREADS_CPU_PCT; 873 value = zio_taskq_batch_pct; 874 break; 875 876 default: 877 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 878 "spa_activate()", 879 zio_type_name[t], zio_taskq_types[q], mode, value); 880 break; 881 } 882 883 for (uint_t i = 0; i < count; i++) { 884 taskq_t *tq; 885 886 if (count > 1) { 887 (void) snprintf(name, sizeof (name), "%s_%s_%u", 888 zio_type_name[t], zio_taskq_types[q], i); 889 } else { 890 (void) snprintf(name, sizeof (name), "%s_%s", 891 zio_type_name[t], zio_taskq_types[q]); 892 } 893 894#ifdef SYSDC 895 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 896 if (batch) 897 flags |= TASKQ_DC_BATCH; 898 899 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 900 spa->spa_proc, zio_taskq_basedc, flags); 901 } else { 902#endif 903 pri_t pri = maxclsyspri; 904 /* 905 * The write issue taskq can be extremely CPU 906 * intensive. Run it at slightly lower priority 907 * than the other taskqs. 908 */ 909 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 910 pri--; 911 912 tq = taskq_create_proc(name, value, pri, 50, 913 INT_MAX, spa->spa_proc, flags); 914#ifdef SYSDC 915 } 916#endif 917 918 tqs->stqs_taskq[i] = tq; 919 } 920} 921 922static void 923spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 924{ 925 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 926 927 if (tqs->stqs_taskq == NULL) { 928 ASSERT0(tqs->stqs_count); 929 return; 930 } 931 932 for (uint_t i = 0; i < tqs->stqs_count; i++) { 933 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 934 taskq_destroy(tqs->stqs_taskq[i]); 935 } 936 937 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 938 tqs->stqs_taskq = NULL; 939} 940 941/* 942 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 943 * Note that a type may have multiple discrete taskqs to avoid lock contention 944 * on the taskq itself. In that case we choose which taskq at random by using 945 * the low bits of gethrtime(). 946 */ 947void 948spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 949 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 950{ 951 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 952 taskq_t *tq; 953 954 ASSERT3P(tqs->stqs_taskq, !=, NULL); 955 ASSERT3U(tqs->stqs_count, !=, 0); 956 957 if (tqs->stqs_count == 1) { 958 tq = tqs->stqs_taskq[0]; 959 } else { 960#ifdef _KERNEL 961 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 962#else 963 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 964#endif 965 } 966 967 taskq_dispatch_ent(tq, func, arg, flags, ent); 968} 969 970static void 971spa_create_zio_taskqs(spa_t *spa) 972{ 973 for (int t = 0; t < ZIO_TYPES; t++) { 974 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 975 spa_taskqs_init(spa, t, q); 976 } 977 } 978} 979 980#ifdef _KERNEL 981#ifdef SPA_PROCESS 982static void 983spa_thread(void *arg) 984{ 985 callb_cpr_t cprinfo; 986 987 spa_t *spa = arg; 988 user_t *pu = PTOU(curproc); 989 990 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 991 spa->spa_name); 992 993 ASSERT(curproc != &p0); 994 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 995 "zpool-%s", spa->spa_name); 996 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 997 998#ifdef PSRSET_BIND 999 /* bind this thread to the requested psrset */ 1000 if (zio_taskq_psrset_bind != PS_NONE) { 1001 pool_lock(); 1002 mutex_enter(&cpu_lock); 1003 mutex_enter(&pidlock); 1004 mutex_enter(&curproc->p_lock); 1005 1006 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1007 0, NULL, NULL) == 0) { 1008 curthread->t_bind_pset = zio_taskq_psrset_bind; 1009 } else { 1010 cmn_err(CE_WARN, 1011 "Couldn't bind process for zfs pool \"%s\" to " 1012 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1013 } 1014 1015 mutex_exit(&curproc->p_lock); 1016 mutex_exit(&pidlock); 1017 mutex_exit(&cpu_lock); 1018 pool_unlock(); 1019 } 1020#endif 1021 1022#ifdef SYSDC 1023 if (zio_taskq_sysdc) { 1024 sysdc_thread_enter(curthread, 100, 0); 1025 } 1026#endif 1027 1028 spa->spa_proc = curproc; 1029 spa->spa_did = curthread->t_did; 1030 1031 spa_create_zio_taskqs(spa); 1032 1033 mutex_enter(&spa->spa_proc_lock); 1034 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1035 1036 spa->spa_proc_state = SPA_PROC_ACTIVE; 1037 cv_broadcast(&spa->spa_proc_cv); 1038 1039 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1040 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1041 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1042 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1043 1044 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1045 spa->spa_proc_state = SPA_PROC_GONE; 1046 spa->spa_proc = &p0; 1047 cv_broadcast(&spa->spa_proc_cv); 1048 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1049 1050 mutex_enter(&curproc->p_lock); 1051 lwp_exit(); 1052} 1053#endif /* SPA_PROCESS */ 1054#endif 1055 1056/* 1057 * Activate an uninitialized pool. 1058 */ 1059static void 1060spa_activate(spa_t *spa, int mode) 1061{ 1062 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1063 1064 spa->spa_state = POOL_STATE_ACTIVE; 1065 spa->spa_mode = mode; 1066 1067 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1068 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1069 1070 /* Try to create a covering process */ 1071 mutex_enter(&spa->spa_proc_lock); 1072 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1073 ASSERT(spa->spa_proc == &p0); 1074 spa->spa_did = 0; 1075 1076#ifdef SPA_PROCESS 1077 /* Only create a process if we're going to be around a while. */ 1078 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1079 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1080 NULL, 0) == 0) { 1081 spa->spa_proc_state = SPA_PROC_CREATED; 1082 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1083 cv_wait(&spa->spa_proc_cv, 1084 &spa->spa_proc_lock); 1085 } 1086 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1087 ASSERT(spa->spa_proc != &p0); 1088 ASSERT(spa->spa_did != 0); 1089 } else { 1090#ifdef _KERNEL 1091 cmn_err(CE_WARN, 1092 "Couldn't create process for zfs pool \"%s\"\n", 1093 spa->spa_name); 1094#endif 1095 } 1096 } 1097#endif /* SPA_PROCESS */ 1098 mutex_exit(&spa->spa_proc_lock); 1099 1100 /* If we didn't create a process, we need to create our taskqs. */ 1101 ASSERT(spa->spa_proc == &p0); 1102 if (spa->spa_proc == &p0) { 1103 spa_create_zio_taskqs(spa); 1104 } 1105 1106 /* 1107 * Start TRIM thread. 1108 */ 1109 trim_thread_create(spa); 1110 1111 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1112 offsetof(vdev_t, vdev_config_dirty_node)); 1113 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1114 offsetof(vdev_t, vdev_state_dirty_node)); 1115 1116 txg_list_create(&spa->spa_vdev_txg_list, 1117 offsetof(struct vdev, vdev_txg_node)); 1118 1119 avl_create(&spa->spa_errlist_scrub, 1120 spa_error_entry_compare, sizeof (spa_error_entry_t), 1121 offsetof(spa_error_entry_t, se_avl)); 1122 avl_create(&spa->spa_errlist_last, 1123 spa_error_entry_compare, sizeof (spa_error_entry_t), 1124 offsetof(spa_error_entry_t, se_avl)); 1125} 1126 1127/* 1128 * Opposite of spa_activate(). 1129 */ 1130static void 1131spa_deactivate(spa_t *spa) 1132{ 1133 ASSERT(spa->spa_sync_on == B_FALSE); 1134 ASSERT(spa->spa_dsl_pool == NULL); 1135 ASSERT(spa->spa_root_vdev == NULL); 1136 ASSERT(spa->spa_async_zio_root == NULL); 1137 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1138 1139 /* 1140 * Stop TRIM thread in case spa_unload() wasn't called directly 1141 * before spa_deactivate(). 1142 */ 1143 trim_thread_destroy(spa); 1144 1145 txg_list_destroy(&spa->spa_vdev_txg_list); 1146 1147 list_destroy(&spa->spa_config_dirty_list); 1148 list_destroy(&spa->spa_state_dirty_list); 1149 1150 for (int t = 0; t < ZIO_TYPES; t++) { 1151 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1152 spa_taskqs_fini(spa, t, q); 1153 } 1154 } 1155 1156 metaslab_class_destroy(spa->spa_normal_class); 1157 spa->spa_normal_class = NULL; 1158 1159 metaslab_class_destroy(spa->spa_log_class); 1160 spa->spa_log_class = NULL; 1161 1162 /* 1163 * If this was part of an import or the open otherwise failed, we may 1164 * still have errors left in the queues. Empty them just in case. 1165 */ 1166 spa_errlog_drain(spa); 1167 1168 avl_destroy(&spa->spa_errlist_scrub); 1169 avl_destroy(&spa->spa_errlist_last); 1170 1171 spa->spa_state = POOL_STATE_UNINITIALIZED; 1172 1173 mutex_enter(&spa->spa_proc_lock); 1174 if (spa->spa_proc_state != SPA_PROC_NONE) { 1175 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1176 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1177 cv_broadcast(&spa->spa_proc_cv); 1178 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1179 ASSERT(spa->spa_proc != &p0); 1180 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1181 } 1182 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1183 spa->spa_proc_state = SPA_PROC_NONE; 1184 } 1185 ASSERT(spa->spa_proc == &p0); 1186 mutex_exit(&spa->spa_proc_lock); 1187 1188#ifdef SPA_PROCESS 1189 /* 1190 * We want to make sure spa_thread() has actually exited the ZFS 1191 * module, so that the module can't be unloaded out from underneath 1192 * it. 1193 */ 1194 if (spa->spa_did != 0) { 1195 thread_join(spa->spa_did); 1196 spa->spa_did = 0; 1197 } 1198#endif /* SPA_PROCESS */ 1199} 1200 1201/* 1202 * Verify a pool configuration, and construct the vdev tree appropriately. This 1203 * will create all the necessary vdevs in the appropriate layout, with each vdev 1204 * in the CLOSED state. This will prep the pool before open/creation/import. 1205 * All vdev validation is done by the vdev_alloc() routine. 1206 */ 1207static int 1208spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1209 uint_t id, int atype) 1210{ 1211 nvlist_t **child; 1212 uint_t children; 1213 int error; 1214 1215 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1216 return (error); 1217 1218 if ((*vdp)->vdev_ops->vdev_op_leaf) 1219 return (0); 1220 1221 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1222 &child, &children); 1223 1224 if (error == ENOENT) 1225 return (0); 1226 1227 if (error) { 1228 vdev_free(*vdp); 1229 *vdp = NULL; 1230 return (SET_ERROR(EINVAL)); 1231 } 1232 1233 for (int c = 0; c < children; c++) { 1234 vdev_t *vd; 1235 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1236 atype)) != 0) { 1237 vdev_free(*vdp); 1238 *vdp = NULL; 1239 return (error); 1240 } 1241 } 1242 1243 ASSERT(*vdp != NULL); 1244 1245 return (0); 1246} 1247 1248/* 1249 * Opposite of spa_load(). 1250 */ 1251static void 1252spa_unload(spa_t *spa) 1253{ 1254 int i; 1255 1256 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1257 1258 /* 1259 * Stop TRIM thread. 1260 */ 1261 trim_thread_destroy(spa); 1262 1263 /* 1264 * Stop async tasks. 1265 */ 1266 spa_async_suspend(spa); 1267 1268 /* 1269 * Stop syncing. 1270 */ 1271 if (spa->spa_sync_on) { 1272 txg_sync_stop(spa->spa_dsl_pool); 1273 spa->spa_sync_on = B_FALSE; 1274 } 1275 1276 /* 1277 * Wait for any outstanding async I/O to complete. 1278 */ 1279 if (spa->spa_async_zio_root != NULL) { 1280 for (int i = 0; i < max_ncpus; i++) 1281 (void) zio_wait(spa->spa_async_zio_root[i]); 1282 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1283 spa->spa_async_zio_root = NULL; 1284 } 1285 1286 bpobj_close(&spa->spa_deferred_bpobj); 1287 1288 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1289 1290 /* 1291 * Close all vdevs. 1292 */ 1293 if (spa->spa_root_vdev) 1294 vdev_free(spa->spa_root_vdev); 1295 ASSERT(spa->spa_root_vdev == NULL); 1296 1297 /* 1298 * Close the dsl pool. 1299 */ 1300 if (spa->spa_dsl_pool) { 1301 dsl_pool_close(spa->spa_dsl_pool); 1302 spa->spa_dsl_pool = NULL; 1303 spa->spa_meta_objset = NULL; 1304 } 1305 1306 ddt_unload(spa); 1307 1308 1309 /* 1310 * Drop and purge level 2 cache 1311 */ 1312 spa_l2cache_drop(spa); 1313 1314 for (i = 0; i < spa->spa_spares.sav_count; i++) 1315 vdev_free(spa->spa_spares.sav_vdevs[i]); 1316 if (spa->spa_spares.sav_vdevs) { 1317 kmem_free(spa->spa_spares.sav_vdevs, 1318 spa->spa_spares.sav_count * sizeof (void *)); 1319 spa->spa_spares.sav_vdevs = NULL; 1320 } 1321 if (spa->spa_spares.sav_config) { 1322 nvlist_free(spa->spa_spares.sav_config); 1323 spa->spa_spares.sav_config = NULL; 1324 } 1325 spa->spa_spares.sav_count = 0; 1326 1327 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1328 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1329 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1330 } 1331 if (spa->spa_l2cache.sav_vdevs) { 1332 kmem_free(spa->spa_l2cache.sav_vdevs, 1333 spa->spa_l2cache.sav_count * sizeof (void *)); 1334 spa->spa_l2cache.sav_vdevs = NULL; 1335 } 1336 if (spa->spa_l2cache.sav_config) { 1337 nvlist_free(spa->spa_l2cache.sav_config); 1338 spa->spa_l2cache.sav_config = NULL; 1339 } 1340 spa->spa_l2cache.sav_count = 0; 1341 1342 spa->spa_async_suspended = 0; 1343 1344 if (spa->spa_comment != NULL) { 1345 spa_strfree(spa->spa_comment); 1346 spa->spa_comment = NULL; 1347 } 1348 1349 spa_config_exit(spa, SCL_ALL, FTAG); 1350} 1351 1352/* 1353 * Load (or re-load) the current list of vdevs describing the active spares for 1354 * this pool. When this is called, we have some form of basic information in 1355 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1356 * then re-generate a more complete list including status information. 1357 */ 1358static void 1359spa_load_spares(spa_t *spa) 1360{ 1361 nvlist_t **spares; 1362 uint_t nspares; 1363 int i; 1364 vdev_t *vd, *tvd; 1365 1366 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1367 1368 /* 1369 * First, close and free any existing spare vdevs. 1370 */ 1371 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1372 vd = spa->spa_spares.sav_vdevs[i]; 1373 1374 /* Undo the call to spa_activate() below */ 1375 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1376 B_FALSE)) != NULL && tvd->vdev_isspare) 1377 spa_spare_remove(tvd); 1378 vdev_close(vd); 1379 vdev_free(vd); 1380 } 1381 1382 if (spa->spa_spares.sav_vdevs) 1383 kmem_free(spa->spa_spares.sav_vdevs, 1384 spa->spa_spares.sav_count * sizeof (void *)); 1385 1386 if (spa->spa_spares.sav_config == NULL) 1387 nspares = 0; 1388 else 1389 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1390 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1391 1392 spa->spa_spares.sav_count = (int)nspares; 1393 spa->spa_spares.sav_vdevs = NULL; 1394 1395 if (nspares == 0) 1396 return; 1397 1398 /* 1399 * Construct the array of vdevs, opening them to get status in the 1400 * process. For each spare, there is potentially two different vdev_t 1401 * structures associated with it: one in the list of spares (used only 1402 * for basic validation purposes) and one in the active vdev 1403 * configuration (if it's spared in). During this phase we open and 1404 * validate each vdev on the spare list. If the vdev also exists in the 1405 * active configuration, then we also mark this vdev as an active spare. 1406 */ 1407 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1408 KM_SLEEP); 1409 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1410 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1411 VDEV_ALLOC_SPARE) == 0); 1412 ASSERT(vd != NULL); 1413 1414 spa->spa_spares.sav_vdevs[i] = vd; 1415 1416 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1417 B_FALSE)) != NULL) { 1418 if (!tvd->vdev_isspare) 1419 spa_spare_add(tvd); 1420 1421 /* 1422 * We only mark the spare active if we were successfully 1423 * able to load the vdev. Otherwise, importing a pool 1424 * with a bad active spare would result in strange 1425 * behavior, because multiple pool would think the spare 1426 * is actively in use. 1427 * 1428 * There is a vulnerability here to an equally bizarre 1429 * circumstance, where a dead active spare is later 1430 * brought back to life (onlined or otherwise). Given 1431 * the rarity of this scenario, and the extra complexity 1432 * it adds, we ignore the possibility. 1433 */ 1434 if (!vdev_is_dead(tvd)) 1435 spa_spare_activate(tvd); 1436 } 1437 1438 vd->vdev_top = vd; 1439 vd->vdev_aux = &spa->spa_spares; 1440 1441 if (vdev_open(vd) != 0) 1442 continue; 1443 1444 if (vdev_validate_aux(vd) == 0) 1445 spa_spare_add(vd); 1446 } 1447 1448 /* 1449 * Recompute the stashed list of spares, with status information 1450 * this time. 1451 */ 1452 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1453 DATA_TYPE_NVLIST_ARRAY) == 0); 1454 1455 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1456 KM_SLEEP); 1457 for (i = 0; i < spa->spa_spares.sav_count; i++) 1458 spares[i] = vdev_config_generate(spa, 1459 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1460 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1461 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1462 for (i = 0; i < spa->spa_spares.sav_count; i++) 1463 nvlist_free(spares[i]); 1464 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1465} 1466 1467/* 1468 * Load (or re-load) the current list of vdevs describing the active l2cache for 1469 * this pool. When this is called, we have some form of basic information in 1470 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1471 * then re-generate a more complete list including status information. 1472 * Devices which are already active have their details maintained, and are 1473 * not re-opened. 1474 */ 1475static void 1476spa_load_l2cache(spa_t *spa) 1477{ 1478 nvlist_t **l2cache; 1479 uint_t nl2cache; 1480 int i, j, oldnvdevs; 1481 uint64_t guid; 1482 vdev_t *vd, **oldvdevs, **newvdevs; 1483 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1484 1485 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1486 1487 if (sav->sav_config != NULL) { 1488 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1489 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1490 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1491 } else { 1492 nl2cache = 0; 1493 newvdevs = NULL; 1494 } 1495 1496 oldvdevs = sav->sav_vdevs; 1497 oldnvdevs = sav->sav_count; 1498 sav->sav_vdevs = NULL; 1499 sav->sav_count = 0; 1500 1501 /* 1502 * Process new nvlist of vdevs. 1503 */ 1504 for (i = 0; i < nl2cache; i++) { 1505 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1506 &guid) == 0); 1507 1508 newvdevs[i] = NULL; 1509 for (j = 0; j < oldnvdevs; j++) { 1510 vd = oldvdevs[j]; 1511 if (vd != NULL && guid == vd->vdev_guid) { 1512 /* 1513 * Retain previous vdev for add/remove ops. 1514 */ 1515 newvdevs[i] = vd; 1516 oldvdevs[j] = NULL; 1517 break; 1518 } 1519 } 1520 1521 if (newvdevs[i] == NULL) { 1522 /* 1523 * Create new vdev 1524 */ 1525 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1526 VDEV_ALLOC_L2CACHE) == 0); 1527 ASSERT(vd != NULL); 1528 newvdevs[i] = vd; 1529 1530 /* 1531 * Commit this vdev as an l2cache device, 1532 * even if it fails to open. 1533 */ 1534 spa_l2cache_add(vd); 1535 1536 vd->vdev_top = vd; 1537 vd->vdev_aux = sav; 1538 1539 spa_l2cache_activate(vd); 1540 1541 if (vdev_open(vd) != 0) 1542 continue; 1543 1544 (void) vdev_validate_aux(vd); 1545 1546 if (!vdev_is_dead(vd)) 1547 l2arc_add_vdev(spa, vd); 1548 } 1549 } 1550 1551 /* 1552 * Purge vdevs that were dropped 1553 */ 1554 for (i = 0; i < oldnvdevs; i++) { 1555 uint64_t pool; 1556 1557 vd = oldvdevs[i]; 1558 if (vd != NULL) { 1559 ASSERT(vd->vdev_isl2cache); 1560 1561 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1562 pool != 0ULL && l2arc_vdev_present(vd)) 1563 l2arc_remove_vdev(vd); 1564 vdev_clear_stats(vd); 1565 vdev_free(vd); 1566 } 1567 } 1568 1569 if (oldvdevs) 1570 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1571 1572 if (sav->sav_config == NULL) 1573 goto out; 1574 1575 sav->sav_vdevs = newvdevs; 1576 sav->sav_count = (int)nl2cache; 1577 1578 /* 1579 * Recompute the stashed list of l2cache devices, with status 1580 * information this time. 1581 */ 1582 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1583 DATA_TYPE_NVLIST_ARRAY) == 0); 1584 1585 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1586 for (i = 0; i < sav->sav_count; i++) 1587 l2cache[i] = vdev_config_generate(spa, 1588 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1589 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1590 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1591out: 1592 for (i = 0; i < sav->sav_count; i++) 1593 nvlist_free(l2cache[i]); 1594 if (sav->sav_count) 1595 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1596} 1597 1598static int 1599load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1600{ 1601 dmu_buf_t *db; 1602 char *packed = NULL; 1603 size_t nvsize = 0; 1604 int error; 1605 *value = NULL; 1606 1607 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1608 if (error != 0) 1609 return (error); 1610 nvsize = *(uint64_t *)db->db_data; 1611 dmu_buf_rele(db, FTAG); 1612 1613 packed = kmem_alloc(nvsize, KM_SLEEP); 1614 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1615 DMU_READ_PREFETCH); 1616 if (error == 0) 1617 error = nvlist_unpack(packed, nvsize, value, 0); 1618 kmem_free(packed, nvsize); 1619 1620 return (error); 1621} 1622 1623/* 1624 * Checks to see if the given vdev could not be opened, in which case we post a 1625 * sysevent to notify the autoreplace code that the device has been removed. 1626 */ 1627static void 1628spa_check_removed(vdev_t *vd) 1629{ 1630 for (int c = 0; c < vd->vdev_children; c++) 1631 spa_check_removed(vd->vdev_child[c]); 1632 1633 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1634 !vd->vdev_ishole) { 1635 zfs_post_autoreplace(vd->vdev_spa, vd); 1636 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1637 } 1638} 1639 1640/* 1641 * Validate the current config against the MOS config 1642 */ 1643static boolean_t 1644spa_config_valid(spa_t *spa, nvlist_t *config) 1645{ 1646 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1647 nvlist_t *nv; 1648 1649 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1650 1651 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1652 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1653 1654 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1655 1656 /* 1657 * If we're doing a normal import, then build up any additional 1658 * diagnostic information about missing devices in this config. 1659 * We'll pass this up to the user for further processing. 1660 */ 1661 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1662 nvlist_t **child, *nv; 1663 uint64_t idx = 0; 1664 1665 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1666 KM_SLEEP); 1667 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1668 1669 for (int c = 0; c < rvd->vdev_children; c++) { 1670 vdev_t *tvd = rvd->vdev_child[c]; 1671 vdev_t *mtvd = mrvd->vdev_child[c]; 1672 1673 if (tvd->vdev_ops == &vdev_missing_ops && 1674 mtvd->vdev_ops != &vdev_missing_ops && 1675 mtvd->vdev_islog) 1676 child[idx++] = vdev_config_generate(spa, mtvd, 1677 B_FALSE, 0); 1678 } 1679 1680 if (idx) { 1681 VERIFY(nvlist_add_nvlist_array(nv, 1682 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1683 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1684 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1685 1686 for (int i = 0; i < idx; i++) 1687 nvlist_free(child[i]); 1688 } 1689 nvlist_free(nv); 1690 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1691 } 1692 1693 /* 1694 * Compare the root vdev tree with the information we have 1695 * from the MOS config (mrvd). Check each top-level vdev 1696 * with the corresponding MOS config top-level (mtvd). 1697 */ 1698 for (int c = 0; c < rvd->vdev_children; c++) { 1699 vdev_t *tvd = rvd->vdev_child[c]; 1700 vdev_t *mtvd = mrvd->vdev_child[c]; 1701 1702 /* 1703 * Resolve any "missing" vdevs in the current configuration. 1704 * If we find that the MOS config has more accurate information 1705 * about the top-level vdev then use that vdev instead. 1706 */ 1707 if (tvd->vdev_ops == &vdev_missing_ops && 1708 mtvd->vdev_ops != &vdev_missing_ops) { 1709 1710 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1711 continue; 1712 1713 /* 1714 * Device specific actions. 1715 */ 1716 if (mtvd->vdev_islog) { 1717 spa_set_log_state(spa, SPA_LOG_CLEAR); 1718 } else { 1719 /* 1720 * XXX - once we have 'readonly' pool 1721 * support we should be able to handle 1722 * missing data devices by transitioning 1723 * the pool to readonly. 1724 */ 1725 continue; 1726 } 1727 1728 /* 1729 * Swap the missing vdev with the data we were 1730 * able to obtain from the MOS config. 1731 */ 1732 vdev_remove_child(rvd, tvd); 1733 vdev_remove_child(mrvd, mtvd); 1734 1735 vdev_add_child(rvd, mtvd); 1736 vdev_add_child(mrvd, tvd); 1737 1738 spa_config_exit(spa, SCL_ALL, FTAG); 1739 vdev_load(mtvd); 1740 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1741 1742 vdev_reopen(rvd); 1743 } else if (mtvd->vdev_islog) { 1744 /* 1745 * Load the slog device's state from the MOS config 1746 * since it's possible that the label does not 1747 * contain the most up-to-date information. 1748 */ 1749 vdev_load_log_state(tvd, mtvd); 1750 vdev_reopen(tvd); 1751 } 1752 } 1753 vdev_free(mrvd); 1754 spa_config_exit(spa, SCL_ALL, FTAG); 1755 1756 /* 1757 * Ensure we were able to validate the config. 1758 */ 1759 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1760} 1761 1762/* 1763 * Check for missing log devices 1764 */ 1765static boolean_t 1766spa_check_logs(spa_t *spa) 1767{ 1768 boolean_t rv = B_FALSE; 1769 1770 switch (spa->spa_log_state) { 1771 case SPA_LOG_MISSING: 1772 /* need to recheck in case slog has been restored */ 1773 case SPA_LOG_UNKNOWN: 1774 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1775 NULL, DS_FIND_CHILDREN) != 0); 1776 if (rv) 1777 spa_set_log_state(spa, SPA_LOG_MISSING); 1778 break; 1779 } 1780 return (rv); 1781} 1782 1783static boolean_t 1784spa_passivate_log(spa_t *spa) 1785{ 1786 vdev_t *rvd = spa->spa_root_vdev; 1787 boolean_t slog_found = B_FALSE; 1788 1789 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1790 1791 if (!spa_has_slogs(spa)) 1792 return (B_FALSE); 1793 1794 for (int c = 0; c < rvd->vdev_children; c++) { 1795 vdev_t *tvd = rvd->vdev_child[c]; 1796 metaslab_group_t *mg = tvd->vdev_mg; 1797 1798 if (tvd->vdev_islog) { 1799 metaslab_group_passivate(mg); 1800 slog_found = B_TRUE; 1801 } 1802 } 1803 1804 return (slog_found); 1805} 1806 1807static void 1808spa_activate_log(spa_t *spa) 1809{ 1810 vdev_t *rvd = spa->spa_root_vdev; 1811 1812 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1813 1814 for (int c = 0; c < rvd->vdev_children; c++) { 1815 vdev_t *tvd = rvd->vdev_child[c]; 1816 metaslab_group_t *mg = tvd->vdev_mg; 1817 1818 if (tvd->vdev_islog) 1819 metaslab_group_activate(mg); 1820 } 1821} 1822 1823int 1824spa_offline_log(spa_t *spa) 1825{ 1826 int error; 1827 1828 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1829 NULL, DS_FIND_CHILDREN); 1830 if (error == 0) { 1831 /* 1832 * We successfully offlined the log device, sync out the 1833 * current txg so that the "stubby" block can be removed 1834 * by zil_sync(). 1835 */ 1836 txg_wait_synced(spa->spa_dsl_pool, 0); 1837 } 1838 return (error); 1839} 1840 1841static void 1842spa_aux_check_removed(spa_aux_vdev_t *sav) 1843{ 1844 int i; 1845 1846 for (i = 0; i < sav->sav_count; i++) 1847 spa_check_removed(sav->sav_vdevs[i]); 1848} 1849 1850void 1851spa_claim_notify(zio_t *zio) 1852{ 1853 spa_t *spa = zio->io_spa; 1854 1855 if (zio->io_error) 1856 return; 1857 1858 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1859 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1860 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1861 mutex_exit(&spa->spa_props_lock); 1862} 1863 1864typedef struct spa_load_error { 1865 uint64_t sle_meta_count; 1866 uint64_t sle_data_count; 1867} spa_load_error_t; 1868 1869static void 1870spa_load_verify_done(zio_t *zio) 1871{ 1872 blkptr_t *bp = zio->io_bp; 1873 spa_load_error_t *sle = zio->io_private; 1874 dmu_object_type_t type = BP_GET_TYPE(bp); 1875 int error = zio->io_error; 1876 spa_t *spa = zio->io_spa; 1877 1878 if (error) { 1879 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1880 type != DMU_OT_INTENT_LOG) 1881 atomic_inc_64(&sle->sle_meta_count); 1882 else 1883 atomic_inc_64(&sle->sle_data_count); 1884 } 1885 zio_data_buf_free(zio->io_data, zio->io_size); 1886 1887 mutex_enter(&spa->spa_scrub_lock); 1888 spa->spa_scrub_inflight--; 1889 cv_broadcast(&spa->spa_scrub_io_cv); 1890 mutex_exit(&spa->spa_scrub_lock); 1891} 1892 1893/* 1894 * Maximum number of concurrent scrub i/os to create while verifying 1895 * a pool while importing it. 1896 */ 1897int spa_load_verify_maxinflight = 10000; 1898boolean_t spa_load_verify_metadata = B_TRUE; 1899boolean_t spa_load_verify_data = B_TRUE; 1900 1901SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1902 &spa_load_verify_maxinflight, 0, 1903 "Maximum number of concurrent scrub I/Os to create while verifying a " 1904 "pool while importing it"); 1905 1906SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1907 &spa_load_verify_metadata, 0, 1908 "Check metadata on import?"); 1909 1910SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1911 &spa_load_verify_data, 0, 1912 "Check user data on import?"); 1913 1914/*ARGSUSED*/ 1915static int 1916spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1917 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1918{ 1919 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1920 return (0); 1921 /* 1922 * Note: normally this routine will not be called if 1923 * spa_load_verify_metadata is not set. However, it may be useful 1924 * to manually set the flag after the traversal has begun. 1925 */ 1926 if (!spa_load_verify_metadata) 1927 return (0); 1928 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1929 return (0); 1930 1931 zio_t *rio = arg; 1932 size_t size = BP_GET_PSIZE(bp); 1933 void *data = zio_data_buf_alloc(size); 1934 1935 mutex_enter(&spa->spa_scrub_lock); 1936 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1937 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1938 spa->spa_scrub_inflight++; 1939 mutex_exit(&spa->spa_scrub_lock); 1940 1941 zio_nowait(zio_read(rio, spa, bp, data, size, 1942 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1943 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1944 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1945 return (0); 1946} 1947 1948static int 1949spa_load_verify(spa_t *spa) 1950{ 1951 zio_t *rio; 1952 spa_load_error_t sle = { 0 }; 1953 zpool_rewind_policy_t policy; 1954 boolean_t verify_ok = B_FALSE; 1955 int error = 0; 1956 1957 zpool_get_rewind_policy(spa->spa_config, &policy); 1958 1959 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1960 return (0); 1961 1962 rio = zio_root(spa, NULL, &sle, 1963 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1964 1965 if (spa_load_verify_metadata) { 1966 error = traverse_pool(spa, spa->spa_verify_min_txg, 1967 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1968 spa_load_verify_cb, rio); 1969 } 1970 1971 (void) zio_wait(rio); 1972 1973 spa->spa_load_meta_errors = sle.sle_meta_count; 1974 spa->spa_load_data_errors = sle.sle_data_count; 1975 1976 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1977 sle.sle_data_count <= policy.zrp_maxdata) { 1978 int64_t loss = 0; 1979 1980 verify_ok = B_TRUE; 1981 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1982 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1983 1984 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1985 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1986 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1987 VERIFY(nvlist_add_int64(spa->spa_load_info, 1988 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1989 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1990 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1991 } else { 1992 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1993 } 1994 1995 if (error) { 1996 if (error != ENXIO && error != EIO) 1997 error = SET_ERROR(EIO); 1998 return (error); 1999 } 2000 2001 return (verify_ok ? 0 : EIO); 2002} 2003 2004/* 2005 * Find a value in the pool props object. 2006 */ 2007static void 2008spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2009{ 2010 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2011 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2012} 2013 2014/* 2015 * Find a value in the pool directory object. 2016 */ 2017static int 2018spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2019{ 2020 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2021 name, sizeof (uint64_t), 1, val)); 2022} 2023 2024static int 2025spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2026{ 2027 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2028 return (err); 2029} 2030 2031/* 2032 * Fix up config after a partly-completed split. This is done with the 2033 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2034 * pool have that entry in their config, but only the splitting one contains 2035 * a list of all the guids of the vdevs that are being split off. 2036 * 2037 * This function determines what to do with that list: either rejoin 2038 * all the disks to the pool, or complete the splitting process. To attempt 2039 * the rejoin, each disk that is offlined is marked online again, and 2040 * we do a reopen() call. If the vdev label for every disk that was 2041 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2042 * then we call vdev_split() on each disk, and complete the split. 2043 * 2044 * Otherwise we leave the config alone, with all the vdevs in place in 2045 * the original pool. 2046 */ 2047static void 2048spa_try_repair(spa_t *spa, nvlist_t *config) 2049{ 2050 uint_t extracted; 2051 uint64_t *glist; 2052 uint_t i, gcount; 2053 nvlist_t *nvl; 2054 vdev_t **vd; 2055 boolean_t attempt_reopen; 2056 2057 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2058 return; 2059 2060 /* check that the config is complete */ 2061 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2062 &glist, &gcount) != 0) 2063 return; 2064 2065 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2066 2067 /* attempt to online all the vdevs & validate */ 2068 attempt_reopen = B_TRUE; 2069 for (i = 0; i < gcount; i++) { 2070 if (glist[i] == 0) /* vdev is hole */ 2071 continue; 2072 2073 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2074 if (vd[i] == NULL) { 2075 /* 2076 * Don't bother attempting to reopen the disks; 2077 * just do the split. 2078 */ 2079 attempt_reopen = B_FALSE; 2080 } else { 2081 /* attempt to re-online it */ 2082 vd[i]->vdev_offline = B_FALSE; 2083 } 2084 } 2085 2086 if (attempt_reopen) { 2087 vdev_reopen(spa->spa_root_vdev); 2088 2089 /* check each device to see what state it's in */ 2090 for (extracted = 0, i = 0; i < gcount; i++) { 2091 if (vd[i] != NULL && 2092 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2093 break; 2094 ++extracted; 2095 } 2096 } 2097 2098 /* 2099 * If every disk has been moved to the new pool, or if we never 2100 * even attempted to look at them, then we split them off for 2101 * good. 2102 */ 2103 if (!attempt_reopen || gcount == extracted) { 2104 for (i = 0; i < gcount; i++) 2105 if (vd[i] != NULL) 2106 vdev_split(vd[i]); 2107 vdev_reopen(spa->spa_root_vdev); 2108 } 2109 2110 kmem_free(vd, gcount * sizeof (vdev_t *)); 2111} 2112 2113static int 2114spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2115 boolean_t mosconfig) 2116{ 2117 nvlist_t *config = spa->spa_config; 2118 char *ereport = FM_EREPORT_ZFS_POOL; 2119 char *comment; 2120 int error; 2121 uint64_t pool_guid; 2122 nvlist_t *nvl; 2123 2124 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2125 return (SET_ERROR(EINVAL)); 2126 2127 ASSERT(spa->spa_comment == NULL); 2128 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2129 spa->spa_comment = spa_strdup(comment); 2130 2131 /* 2132 * Versioning wasn't explicitly added to the label until later, so if 2133 * it's not present treat it as the initial version. 2134 */ 2135 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2136 &spa->spa_ubsync.ub_version) != 0) 2137 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2138 2139 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2140 &spa->spa_config_txg); 2141 2142 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2143 spa_guid_exists(pool_guid, 0)) { 2144 error = SET_ERROR(EEXIST); 2145 } else { 2146 spa->spa_config_guid = pool_guid; 2147 2148 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2149 &nvl) == 0) { 2150 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2151 KM_SLEEP) == 0); 2152 } 2153 2154 nvlist_free(spa->spa_load_info); 2155 spa->spa_load_info = fnvlist_alloc(); 2156 2157 gethrestime(&spa->spa_loaded_ts); 2158 error = spa_load_impl(spa, pool_guid, config, state, type, 2159 mosconfig, &ereport); 2160 } 2161 2162 spa->spa_minref = refcount_count(&spa->spa_refcount); 2163 if (error) { 2164 if (error != EEXIST) { 2165 spa->spa_loaded_ts.tv_sec = 0; 2166 spa->spa_loaded_ts.tv_nsec = 0; 2167 } 2168 if (error != EBADF) { 2169 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2170 } 2171 } 2172 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2173 spa->spa_ena = 0; 2174 2175 return (error); 2176} 2177 2178/* 2179 * Load an existing storage pool, using the pool's builtin spa_config as a 2180 * source of configuration information. 2181 */ 2182static int 2183spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2184 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2185 char **ereport) 2186{ 2187 int error = 0; 2188 nvlist_t *nvroot = NULL; 2189 nvlist_t *label; 2190 vdev_t *rvd; 2191 uberblock_t *ub = &spa->spa_uberblock; 2192 uint64_t children, config_cache_txg = spa->spa_config_txg; 2193 int orig_mode = spa->spa_mode; 2194 int parse; 2195 uint64_t obj; 2196 boolean_t missing_feat_write = B_FALSE; 2197 2198 /* 2199 * If this is an untrusted config, access the pool in read-only mode. 2200 * This prevents things like resilvering recently removed devices. 2201 */ 2202 if (!mosconfig) 2203 spa->spa_mode = FREAD; 2204 2205 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2206 2207 spa->spa_load_state = state; 2208 2209 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2210 return (SET_ERROR(EINVAL)); 2211 2212 parse = (type == SPA_IMPORT_EXISTING ? 2213 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2214 2215 /* 2216 * Create "The Godfather" zio to hold all async IOs 2217 */ 2218 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2219 KM_SLEEP); 2220 for (int i = 0; i < max_ncpus; i++) { 2221 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2222 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2223 ZIO_FLAG_GODFATHER); 2224 } 2225 2226 /* 2227 * Parse the configuration into a vdev tree. We explicitly set the 2228 * value that will be returned by spa_version() since parsing the 2229 * configuration requires knowing the version number. 2230 */ 2231 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2232 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2233 spa_config_exit(spa, SCL_ALL, FTAG); 2234 2235 if (error != 0) 2236 return (error); 2237 2238 ASSERT(spa->spa_root_vdev == rvd); 2239 2240 if (type != SPA_IMPORT_ASSEMBLE) { 2241 ASSERT(spa_guid(spa) == pool_guid); 2242 } 2243 2244 /* 2245 * Try to open all vdevs, loading each label in the process. 2246 */ 2247 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2248 error = vdev_open(rvd); 2249 spa_config_exit(spa, SCL_ALL, FTAG); 2250 if (error != 0) 2251 return (error); 2252 2253 /* 2254 * We need to validate the vdev labels against the configuration that 2255 * we have in hand, which is dependent on the setting of mosconfig. If 2256 * mosconfig is true then we're validating the vdev labels based on 2257 * that config. Otherwise, we're validating against the cached config 2258 * (zpool.cache) that was read when we loaded the zfs module, and then 2259 * later we will recursively call spa_load() and validate against 2260 * the vdev config. 2261 * 2262 * If we're assembling a new pool that's been split off from an 2263 * existing pool, the labels haven't yet been updated so we skip 2264 * validation for now. 2265 */ 2266 if (type != SPA_IMPORT_ASSEMBLE) { 2267 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2268 error = vdev_validate(rvd, mosconfig); 2269 spa_config_exit(spa, SCL_ALL, FTAG); 2270 2271 if (error != 0) 2272 return (error); 2273 2274 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2275 return (SET_ERROR(ENXIO)); 2276 } 2277 2278 /* 2279 * Find the best uberblock. 2280 */ 2281 vdev_uberblock_load(rvd, ub, &label); 2282 2283 /* 2284 * If we weren't able to find a single valid uberblock, return failure. 2285 */ 2286 if (ub->ub_txg == 0) { 2287 nvlist_free(label); 2288 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2289 } 2290 2291 /* 2292 * If the pool has an unsupported version we can't open it. 2293 */ 2294 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2295 nvlist_free(label); 2296 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2297 } 2298 2299 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2300 nvlist_t *features; 2301 2302 /* 2303 * If we weren't able to find what's necessary for reading the 2304 * MOS in the label, return failure. 2305 */ 2306 if (label == NULL || nvlist_lookup_nvlist(label, 2307 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2308 nvlist_free(label); 2309 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2310 ENXIO)); 2311 } 2312 2313 /* 2314 * Update our in-core representation with the definitive values 2315 * from the label. 2316 */ 2317 nvlist_free(spa->spa_label_features); 2318 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2319 } 2320 2321 nvlist_free(label); 2322 2323 /* 2324 * Look through entries in the label nvlist's features_for_read. If 2325 * there is a feature listed there which we don't understand then we 2326 * cannot open a pool. 2327 */ 2328 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2329 nvlist_t *unsup_feat; 2330 2331 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2332 0); 2333 2334 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2335 NULL); nvp != NULL; 2336 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2337 if (!zfeature_is_supported(nvpair_name(nvp))) { 2338 VERIFY(nvlist_add_string(unsup_feat, 2339 nvpair_name(nvp), "") == 0); 2340 } 2341 } 2342 2343 if (!nvlist_empty(unsup_feat)) { 2344 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2345 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2346 nvlist_free(unsup_feat); 2347 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2348 ENOTSUP)); 2349 } 2350 2351 nvlist_free(unsup_feat); 2352 } 2353 2354 /* 2355 * If the vdev guid sum doesn't match the uberblock, we have an 2356 * incomplete configuration. We first check to see if the pool 2357 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2358 * If it is, defer the vdev_guid_sum check till later so we 2359 * can handle missing vdevs. 2360 */ 2361 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2362 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2363 rvd->vdev_guid_sum != ub->ub_guid_sum) 2364 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2365 2366 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2367 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2368 spa_try_repair(spa, config); 2369 spa_config_exit(spa, SCL_ALL, FTAG); 2370 nvlist_free(spa->spa_config_splitting); 2371 spa->spa_config_splitting = NULL; 2372 } 2373 2374 /* 2375 * Initialize internal SPA structures. 2376 */ 2377 spa->spa_state = POOL_STATE_ACTIVE; 2378 spa->spa_ubsync = spa->spa_uberblock; 2379 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2380 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2381 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2382 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2383 spa->spa_claim_max_txg = spa->spa_first_txg; 2384 spa->spa_prev_software_version = ub->ub_software_version; 2385 2386 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2387 if (error) 2388 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2389 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2390 2391 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2392 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2393 2394 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2395 boolean_t missing_feat_read = B_FALSE; 2396 nvlist_t *unsup_feat, *enabled_feat; 2397 2398 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2399 &spa->spa_feat_for_read_obj) != 0) { 2400 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2401 } 2402 2403 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2404 &spa->spa_feat_for_write_obj) != 0) { 2405 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2406 } 2407 2408 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2409 &spa->spa_feat_desc_obj) != 0) { 2410 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2411 } 2412 2413 enabled_feat = fnvlist_alloc(); 2414 unsup_feat = fnvlist_alloc(); 2415 2416 if (!spa_features_check(spa, B_FALSE, 2417 unsup_feat, enabled_feat)) 2418 missing_feat_read = B_TRUE; 2419 2420 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2421 if (!spa_features_check(spa, B_TRUE, 2422 unsup_feat, enabled_feat)) { 2423 missing_feat_write = B_TRUE; 2424 } 2425 } 2426 2427 fnvlist_add_nvlist(spa->spa_load_info, 2428 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2429 2430 if (!nvlist_empty(unsup_feat)) { 2431 fnvlist_add_nvlist(spa->spa_load_info, 2432 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2433 } 2434 2435 fnvlist_free(enabled_feat); 2436 fnvlist_free(unsup_feat); 2437 2438 if (!missing_feat_read) { 2439 fnvlist_add_boolean(spa->spa_load_info, 2440 ZPOOL_CONFIG_CAN_RDONLY); 2441 } 2442 2443 /* 2444 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2445 * twofold: to determine whether the pool is available for 2446 * import in read-write mode and (if it is not) whether the 2447 * pool is available for import in read-only mode. If the pool 2448 * is available for import in read-write mode, it is displayed 2449 * as available in userland; if it is not available for import 2450 * in read-only mode, it is displayed as unavailable in 2451 * userland. If the pool is available for import in read-only 2452 * mode but not read-write mode, it is displayed as unavailable 2453 * in userland with a special note that the pool is actually 2454 * available for open in read-only mode. 2455 * 2456 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2457 * missing a feature for write, we must first determine whether 2458 * the pool can be opened read-only before returning to 2459 * userland in order to know whether to display the 2460 * abovementioned note. 2461 */ 2462 if (missing_feat_read || (missing_feat_write && 2463 spa_writeable(spa))) { 2464 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2465 ENOTSUP)); 2466 } 2467 2468 /* 2469 * Load refcounts for ZFS features from disk into an in-memory 2470 * cache during SPA initialization. 2471 */ 2472 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2473 uint64_t refcount; 2474 2475 error = feature_get_refcount_from_disk(spa, 2476 &spa_feature_table[i], &refcount); 2477 if (error == 0) { 2478 spa->spa_feat_refcount_cache[i] = refcount; 2479 } else if (error == ENOTSUP) { 2480 spa->spa_feat_refcount_cache[i] = 2481 SPA_FEATURE_DISABLED; 2482 } else { 2483 return (spa_vdev_err(rvd, 2484 VDEV_AUX_CORRUPT_DATA, EIO)); 2485 } 2486 } 2487 } 2488 2489 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2490 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2491 &spa->spa_feat_enabled_txg_obj) != 0) 2492 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2493 } 2494 2495 spa->spa_is_initializing = B_TRUE; 2496 error = dsl_pool_open(spa->spa_dsl_pool); 2497 spa->spa_is_initializing = B_FALSE; 2498 if (error != 0) 2499 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2500 2501 if (!mosconfig) { 2502 uint64_t hostid; 2503 nvlist_t *policy = NULL, *nvconfig; 2504 2505 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2506 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2507 2508 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2509 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2510 char *hostname; 2511 unsigned long myhostid = 0; 2512 2513 VERIFY(nvlist_lookup_string(nvconfig, 2514 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2515 2516#ifdef _KERNEL 2517 myhostid = zone_get_hostid(NULL); 2518#else /* _KERNEL */ 2519 /* 2520 * We're emulating the system's hostid in userland, so 2521 * we can't use zone_get_hostid(). 2522 */ 2523 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2524#endif /* _KERNEL */ 2525 if (check_hostid && hostid != 0 && myhostid != 0 && 2526 hostid != myhostid) { 2527 nvlist_free(nvconfig); 2528 cmn_err(CE_WARN, "pool '%s' could not be " 2529 "loaded as it was last accessed by " 2530 "another system (host: %s hostid: 0x%lx). " 2531 "See: http://illumos.org/msg/ZFS-8000-EY", 2532 spa_name(spa), hostname, 2533 (unsigned long)hostid); 2534 return (SET_ERROR(EBADF)); 2535 } 2536 } 2537 if (nvlist_lookup_nvlist(spa->spa_config, 2538 ZPOOL_REWIND_POLICY, &policy) == 0) 2539 VERIFY(nvlist_add_nvlist(nvconfig, 2540 ZPOOL_REWIND_POLICY, policy) == 0); 2541 2542 spa_config_set(spa, nvconfig); 2543 spa_unload(spa); 2544 spa_deactivate(spa); 2545 spa_activate(spa, orig_mode); 2546 2547 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2548 } 2549 2550 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2551 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2552 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2553 if (error != 0) 2554 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2555 2556 /* 2557 * Load the bit that tells us to use the new accounting function 2558 * (raid-z deflation). If we have an older pool, this will not 2559 * be present. 2560 */ 2561 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2562 if (error != 0 && error != ENOENT) 2563 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2564 2565 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2566 &spa->spa_creation_version); 2567 if (error != 0 && error != ENOENT) 2568 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2569 2570 /* 2571 * Load the persistent error log. If we have an older pool, this will 2572 * not be present. 2573 */ 2574 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2575 if (error != 0 && error != ENOENT) 2576 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2577 2578 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2579 &spa->spa_errlog_scrub); 2580 if (error != 0 && error != ENOENT) 2581 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2582 2583 /* 2584 * Load the history object. If we have an older pool, this 2585 * will not be present. 2586 */ 2587 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2588 if (error != 0 && error != ENOENT) 2589 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2590 2591 /* 2592 * If we're assembling the pool from the split-off vdevs of 2593 * an existing pool, we don't want to attach the spares & cache 2594 * devices. 2595 */ 2596 2597 /* 2598 * Load any hot spares for this pool. 2599 */ 2600 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2601 if (error != 0 && error != ENOENT) 2602 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2603 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2604 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2605 if (load_nvlist(spa, spa->spa_spares.sav_object, 2606 &spa->spa_spares.sav_config) != 0) 2607 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2608 2609 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2610 spa_load_spares(spa); 2611 spa_config_exit(spa, SCL_ALL, FTAG); 2612 } else if (error == 0) { 2613 spa->spa_spares.sav_sync = B_TRUE; 2614 } 2615 2616 /* 2617 * Load any level 2 ARC devices for this pool. 2618 */ 2619 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2620 &spa->spa_l2cache.sav_object); 2621 if (error != 0 && error != ENOENT) 2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2623 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2624 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2625 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2626 &spa->spa_l2cache.sav_config) != 0) 2627 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2628 2629 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2630 spa_load_l2cache(spa); 2631 spa_config_exit(spa, SCL_ALL, FTAG); 2632 } else if (error == 0) { 2633 spa->spa_l2cache.sav_sync = B_TRUE; 2634 } 2635 2636 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2637 2638 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2639 if (error && error != ENOENT) 2640 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2641 2642 if (error == 0) { 2643 uint64_t autoreplace; 2644 2645 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2646 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2647 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2648 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2649 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2650 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2651 &spa->spa_dedup_ditto); 2652 2653 spa->spa_autoreplace = (autoreplace != 0); 2654 } 2655 2656 /* 2657 * If the 'autoreplace' property is set, then post a resource notifying 2658 * the ZFS DE that it should not issue any faults for unopenable 2659 * devices. We also iterate over the vdevs, and post a sysevent for any 2660 * unopenable vdevs so that the normal autoreplace handler can take 2661 * over. 2662 */ 2663 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2664 spa_check_removed(spa->spa_root_vdev); 2665 /* 2666 * For the import case, this is done in spa_import(), because 2667 * at this point we're using the spare definitions from 2668 * the MOS config, not necessarily from the userland config. 2669 */ 2670 if (state != SPA_LOAD_IMPORT) { 2671 spa_aux_check_removed(&spa->spa_spares); 2672 spa_aux_check_removed(&spa->spa_l2cache); 2673 } 2674 } 2675 2676 /* 2677 * Load the vdev state for all toplevel vdevs. 2678 */ 2679 vdev_load(rvd); 2680 2681 /* 2682 * Propagate the leaf DTLs we just loaded all the way up the tree. 2683 */ 2684 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2685 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2686 spa_config_exit(spa, SCL_ALL, FTAG); 2687 2688 /* 2689 * Load the DDTs (dedup tables). 2690 */ 2691 error = ddt_load(spa); 2692 if (error != 0) 2693 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2694 2695 spa_update_dspace(spa); 2696 2697 /* 2698 * Validate the config, using the MOS config to fill in any 2699 * information which might be missing. If we fail to validate 2700 * the config then declare the pool unfit for use. If we're 2701 * assembling a pool from a split, the log is not transferred 2702 * over. 2703 */ 2704 if (type != SPA_IMPORT_ASSEMBLE) { 2705 nvlist_t *nvconfig; 2706 2707 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2708 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2709 2710 if (!spa_config_valid(spa, nvconfig)) { 2711 nvlist_free(nvconfig); 2712 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2713 ENXIO)); 2714 } 2715 nvlist_free(nvconfig); 2716 2717 /* 2718 * Now that we've validated the config, check the state of the 2719 * root vdev. If it can't be opened, it indicates one or 2720 * more toplevel vdevs are faulted. 2721 */ 2722 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2723 return (SET_ERROR(ENXIO)); 2724 2725 if (spa_check_logs(spa)) { 2726 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2727 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2728 } 2729 } 2730 2731 if (missing_feat_write) { 2732 ASSERT(state == SPA_LOAD_TRYIMPORT); 2733 2734 /* 2735 * At this point, we know that we can open the pool in 2736 * read-only mode but not read-write mode. We now have enough 2737 * information and can return to userland. 2738 */ 2739 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2740 } 2741 2742 /* 2743 * We've successfully opened the pool, verify that we're ready 2744 * to start pushing transactions. 2745 */ 2746 if (state != SPA_LOAD_TRYIMPORT) { 2747 if (error = spa_load_verify(spa)) 2748 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2749 error)); 2750 } 2751 2752 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2753 spa->spa_load_max_txg == UINT64_MAX)) { 2754 dmu_tx_t *tx; 2755 int need_update = B_FALSE; 2756 2757 ASSERT(state != SPA_LOAD_TRYIMPORT); 2758 2759 /* 2760 * Claim log blocks that haven't been committed yet. 2761 * This must all happen in a single txg. 2762 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2763 * invoked from zil_claim_log_block()'s i/o done callback. 2764 * Price of rollback is that we abandon the log. 2765 */ 2766 spa->spa_claiming = B_TRUE; 2767 2768 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2769 spa_first_txg(spa)); 2770 (void) dmu_objset_find(spa_name(spa), 2771 zil_claim, tx, DS_FIND_CHILDREN); 2772 dmu_tx_commit(tx); 2773 2774 spa->spa_claiming = B_FALSE; 2775 2776 spa_set_log_state(spa, SPA_LOG_GOOD); 2777 spa->spa_sync_on = B_TRUE; 2778 txg_sync_start(spa->spa_dsl_pool); 2779 2780 /* 2781 * Wait for all claims to sync. We sync up to the highest 2782 * claimed log block birth time so that claimed log blocks 2783 * don't appear to be from the future. spa_claim_max_txg 2784 * will have been set for us by either zil_check_log_chain() 2785 * (invoked from spa_check_logs()) or zil_claim() above. 2786 */ 2787 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2788 2789 /* 2790 * If the config cache is stale, or we have uninitialized 2791 * metaslabs (see spa_vdev_add()), then update the config. 2792 * 2793 * If this is a verbatim import, trust the current 2794 * in-core spa_config and update the disk labels. 2795 */ 2796 if (config_cache_txg != spa->spa_config_txg || 2797 state == SPA_LOAD_IMPORT || 2798 state == SPA_LOAD_RECOVER || 2799 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2800 need_update = B_TRUE; 2801 2802 for (int c = 0; c < rvd->vdev_children; c++) 2803 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2804 need_update = B_TRUE; 2805 2806 /* 2807 * Update the config cache asychronously in case we're the 2808 * root pool, in which case the config cache isn't writable yet. 2809 */ 2810 if (need_update) 2811 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2812 2813 /* 2814 * Check all DTLs to see if anything needs resilvering. 2815 */ 2816 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2817 vdev_resilver_needed(rvd, NULL, NULL)) 2818 spa_async_request(spa, SPA_ASYNC_RESILVER); 2819 2820 /* 2821 * Log the fact that we booted up (so that we can detect if 2822 * we rebooted in the middle of an operation). 2823 */ 2824 spa_history_log_version(spa, "open"); 2825 2826 /* 2827 * Delete any inconsistent datasets. 2828 */ 2829 (void) dmu_objset_find(spa_name(spa), 2830 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2831 2832 /* 2833 * Clean up any stale temporary dataset userrefs. 2834 */ 2835 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2836 } 2837 2838 return (0); 2839} 2840 2841static int 2842spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2843{ 2844 int mode = spa->spa_mode; 2845 2846 spa_unload(spa); 2847 spa_deactivate(spa); 2848 2849 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2850 2851 spa_activate(spa, mode); 2852 spa_async_suspend(spa); 2853 2854 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2855} 2856 2857/* 2858 * If spa_load() fails this function will try loading prior txg's. If 2859 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2860 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2861 * function will not rewind the pool and will return the same error as 2862 * spa_load(). 2863 */ 2864static int 2865spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2866 uint64_t max_request, int rewind_flags) 2867{ 2868 nvlist_t *loadinfo = NULL; 2869 nvlist_t *config = NULL; 2870 int load_error, rewind_error; 2871 uint64_t safe_rewind_txg; 2872 uint64_t min_txg; 2873 2874 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2875 spa->spa_load_max_txg = spa->spa_load_txg; 2876 spa_set_log_state(spa, SPA_LOG_CLEAR); 2877 } else { 2878 spa->spa_load_max_txg = max_request; 2879 if (max_request != UINT64_MAX) 2880 spa->spa_extreme_rewind = B_TRUE; 2881 } 2882 2883 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2884 mosconfig); 2885 if (load_error == 0) 2886 return (0); 2887 2888 if (spa->spa_root_vdev != NULL) 2889 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2890 2891 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2892 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2893 2894 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2895 nvlist_free(config); 2896 return (load_error); 2897 } 2898 2899 if (state == SPA_LOAD_RECOVER) { 2900 /* Price of rolling back is discarding txgs, including log */ 2901 spa_set_log_state(spa, SPA_LOG_CLEAR); 2902 } else { 2903 /* 2904 * If we aren't rolling back save the load info from our first 2905 * import attempt so that we can restore it after attempting 2906 * to rewind. 2907 */ 2908 loadinfo = spa->spa_load_info; 2909 spa->spa_load_info = fnvlist_alloc(); 2910 } 2911 2912 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2913 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2914 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2915 TXG_INITIAL : safe_rewind_txg; 2916 2917 /* 2918 * Continue as long as we're finding errors, we're still within 2919 * the acceptable rewind range, and we're still finding uberblocks 2920 */ 2921 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2922 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2923 if (spa->spa_load_max_txg < safe_rewind_txg) 2924 spa->spa_extreme_rewind = B_TRUE; 2925 rewind_error = spa_load_retry(spa, state, mosconfig); 2926 } 2927 2928 spa->spa_extreme_rewind = B_FALSE; 2929 spa->spa_load_max_txg = UINT64_MAX; 2930 2931 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2932 spa_config_set(spa, config); 2933 2934 if (state == SPA_LOAD_RECOVER) { 2935 ASSERT3P(loadinfo, ==, NULL); 2936 return (rewind_error); 2937 } else { 2938 /* Store the rewind info as part of the initial load info */ 2939 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2940 spa->spa_load_info); 2941 2942 /* Restore the initial load info */ 2943 fnvlist_free(spa->spa_load_info); 2944 spa->spa_load_info = loadinfo; 2945 2946 return (load_error); 2947 } 2948} 2949 2950/* 2951 * Pool Open/Import 2952 * 2953 * The import case is identical to an open except that the configuration is sent 2954 * down from userland, instead of grabbed from the configuration cache. For the 2955 * case of an open, the pool configuration will exist in the 2956 * POOL_STATE_UNINITIALIZED state. 2957 * 2958 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2959 * the same time open the pool, without having to keep around the spa_t in some 2960 * ambiguous state. 2961 */ 2962static int 2963spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2964 nvlist_t **config) 2965{ 2966 spa_t *spa; 2967 spa_load_state_t state = SPA_LOAD_OPEN; 2968 int error; 2969 int locked = B_FALSE; 2970 int firstopen = B_FALSE; 2971 2972 *spapp = NULL; 2973 2974 /* 2975 * As disgusting as this is, we need to support recursive calls to this 2976 * function because dsl_dir_open() is called during spa_load(), and ends 2977 * up calling spa_open() again. The real fix is to figure out how to 2978 * avoid dsl_dir_open() calling this in the first place. 2979 */ 2980 if (mutex_owner(&spa_namespace_lock) != curthread) { 2981 mutex_enter(&spa_namespace_lock); 2982 locked = B_TRUE; 2983 } 2984 2985 if ((spa = spa_lookup(pool)) == NULL) { 2986 if (locked) 2987 mutex_exit(&spa_namespace_lock); 2988 return (SET_ERROR(ENOENT)); 2989 } 2990 2991 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2992 zpool_rewind_policy_t policy; 2993 2994 firstopen = B_TRUE; 2995 2996 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2997 &policy); 2998 if (policy.zrp_request & ZPOOL_DO_REWIND) 2999 state = SPA_LOAD_RECOVER; 3000 3001 spa_activate(spa, spa_mode_global); 3002 3003 if (state != SPA_LOAD_RECOVER) 3004 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3005 3006 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3007 policy.zrp_request); 3008 3009 if (error == EBADF) { 3010 /* 3011 * If vdev_validate() returns failure (indicated by 3012 * EBADF), it indicates that one of the vdevs indicates 3013 * that the pool has been exported or destroyed. If 3014 * this is the case, the config cache is out of sync and 3015 * we should remove the pool from the namespace. 3016 */ 3017 spa_unload(spa); 3018 spa_deactivate(spa); 3019 spa_config_sync(spa, B_TRUE, B_TRUE); 3020 spa_remove(spa); 3021 if (locked) 3022 mutex_exit(&spa_namespace_lock); 3023 return (SET_ERROR(ENOENT)); 3024 } 3025 3026 if (error) { 3027 /* 3028 * We can't open the pool, but we still have useful 3029 * information: the state of each vdev after the 3030 * attempted vdev_open(). Return this to the user. 3031 */ 3032 if (config != NULL && spa->spa_config) { 3033 VERIFY(nvlist_dup(spa->spa_config, config, 3034 KM_SLEEP) == 0); 3035 VERIFY(nvlist_add_nvlist(*config, 3036 ZPOOL_CONFIG_LOAD_INFO, 3037 spa->spa_load_info) == 0); 3038 } 3039 spa_unload(spa); 3040 spa_deactivate(spa); 3041 spa->spa_last_open_failed = error; 3042 if (locked) 3043 mutex_exit(&spa_namespace_lock); 3044 *spapp = NULL; 3045 return (error); 3046 } 3047 } 3048 3049 spa_open_ref(spa, tag); 3050 3051 if (config != NULL) 3052 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3053 3054 /* 3055 * If we've recovered the pool, pass back any information we 3056 * gathered while doing the load. 3057 */ 3058 if (state == SPA_LOAD_RECOVER) { 3059 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3060 spa->spa_load_info) == 0); 3061 } 3062 3063 if (locked) { 3064 spa->spa_last_open_failed = 0; 3065 spa->spa_last_ubsync_txg = 0; 3066 spa->spa_load_txg = 0; 3067 mutex_exit(&spa_namespace_lock); 3068#ifdef __FreeBSD__ 3069#ifdef _KERNEL 3070 if (firstopen) 3071 zvol_create_minors(spa->spa_name); 3072#endif 3073#endif 3074 } 3075 3076 *spapp = spa; 3077 3078 return (0); 3079} 3080 3081int 3082spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3083 nvlist_t **config) 3084{ 3085 return (spa_open_common(name, spapp, tag, policy, config)); 3086} 3087 3088int 3089spa_open(const char *name, spa_t **spapp, void *tag) 3090{ 3091 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3092} 3093 3094/* 3095 * Lookup the given spa_t, incrementing the inject count in the process, 3096 * preventing it from being exported or destroyed. 3097 */ 3098spa_t * 3099spa_inject_addref(char *name) 3100{ 3101 spa_t *spa; 3102 3103 mutex_enter(&spa_namespace_lock); 3104 if ((spa = spa_lookup(name)) == NULL) { 3105 mutex_exit(&spa_namespace_lock); 3106 return (NULL); 3107 } 3108 spa->spa_inject_ref++; 3109 mutex_exit(&spa_namespace_lock); 3110 3111 return (spa); 3112} 3113 3114void 3115spa_inject_delref(spa_t *spa) 3116{ 3117 mutex_enter(&spa_namespace_lock); 3118 spa->spa_inject_ref--; 3119 mutex_exit(&spa_namespace_lock); 3120} 3121 3122/* 3123 * Add spares device information to the nvlist. 3124 */ 3125static void 3126spa_add_spares(spa_t *spa, nvlist_t *config) 3127{ 3128 nvlist_t **spares; 3129 uint_t i, nspares; 3130 nvlist_t *nvroot; 3131 uint64_t guid; 3132 vdev_stat_t *vs; 3133 uint_t vsc; 3134 uint64_t pool; 3135 3136 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3137 3138 if (spa->spa_spares.sav_count == 0) 3139 return; 3140 3141 VERIFY(nvlist_lookup_nvlist(config, 3142 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3143 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3144 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3145 if (nspares != 0) { 3146 VERIFY(nvlist_add_nvlist_array(nvroot, 3147 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3148 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3149 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3150 3151 /* 3152 * Go through and find any spares which have since been 3153 * repurposed as an active spare. If this is the case, update 3154 * their status appropriately. 3155 */ 3156 for (i = 0; i < nspares; i++) { 3157 VERIFY(nvlist_lookup_uint64(spares[i], 3158 ZPOOL_CONFIG_GUID, &guid) == 0); 3159 if (spa_spare_exists(guid, &pool, NULL) && 3160 pool != 0ULL) { 3161 VERIFY(nvlist_lookup_uint64_array( 3162 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3163 (uint64_t **)&vs, &vsc) == 0); 3164 vs->vs_state = VDEV_STATE_CANT_OPEN; 3165 vs->vs_aux = VDEV_AUX_SPARED; 3166 } 3167 } 3168 } 3169} 3170 3171/* 3172 * Add l2cache device information to the nvlist, including vdev stats. 3173 */ 3174static void 3175spa_add_l2cache(spa_t *spa, nvlist_t *config) 3176{ 3177 nvlist_t **l2cache; 3178 uint_t i, j, nl2cache; 3179 nvlist_t *nvroot; 3180 uint64_t guid; 3181 vdev_t *vd; 3182 vdev_stat_t *vs; 3183 uint_t vsc; 3184 3185 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3186 3187 if (spa->spa_l2cache.sav_count == 0) 3188 return; 3189 3190 VERIFY(nvlist_lookup_nvlist(config, 3191 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3192 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3193 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3194 if (nl2cache != 0) { 3195 VERIFY(nvlist_add_nvlist_array(nvroot, 3196 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3197 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3198 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3199 3200 /* 3201 * Update level 2 cache device stats. 3202 */ 3203 3204 for (i = 0; i < nl2cache; i++) { 3205 VERIFY(nvlist_lookup_uint64(l2cache[i], 3206 ZPOOL_CONFIG_GUID, &guid) == 0); 3207 3208 vd = NULL; 3209 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3210 if (guid == 3211 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3212 vd = spa->spa_l2cache.sav_vdevs[j]; 3213 break; 3214 } 3215 } 3216 ASSERT(vd != NULL); 3217 3218 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3219 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3220 == 0); 3221 vdev_get_stats(vd, vs); 3222 } 3223 } 3224} 3225 3226static void 3227spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3228{ 3229 nvlist_t *features; 3230 zap_cursor_t zc; 3231 zap_attribute_t za; 3232 3233 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3234 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3235 3236 /* We may be unable to read features if pool is suspended. */ 3237 if (spa_suspended(spa)) 3238 goto out; 3239 3240 if (spa->spa_feat_for_read_obj != 0) { 3241 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3242 spa->spa_feat_for_read_obj); 3243 zap_cursor_retrieve(&zc, &za) == 0; 3244 zap_cursor_advance(&zc)) { 3245 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3246 za.za_num_integers == 1); 3247 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3248 za.za_first_integer)); 3249 } 3250 zap_cursor_fini(&zc); 3251 } 3252 3253 if (spa->spa_feat_for_write_obj != 0) { 3254 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3255 spa->spa_feat_for_write_obj); 3256 zap_cursor_retrieve(&zc, &za) == 0; 3257 zap_cursor_advance(&zc)) { 3258 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3259 za.za_num_integers == 1); 3260 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3261 za.za_first_integer)); 3262 } 3263 zap_cursor_fini(&zc); 3264 } 3265 3266out: 3267 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3268 features) == 0); 3269 nvlist_free(features); 3270} 3271 3272int 3273spa_get_stats(const char *name, nvlist_t **config, 3274 char *altroot, size_t buflen) 3275{ 3276 int error; 3277 spa_t *spa; 3278 3279 *config = NULL; 3280 error = spa_open_common(name, &spa, FTAG, NULL, config); 3281 3282 if (spa != NULL) { 3283 /* 3284 * This still leaves a window of inconsistency where the spares 3285 * or l2cache devices could change and the config would be 3286 * self-inconsistent. 3287 */ 3288 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3289 3290 if (*config != NULL) { 3291 uint64_t loadtimes[2]; 3292 3293 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3294 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3295 VERIFY(nvlist_add_uint64_array(*config, 3296 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3297 3298 VERIFY(nvlist_add_uint64(*config, 3299 ZPOOL_CONFIG_ERRCOUNT, 3300 spa_get_errlog_size(spa)) == 0); 3301 3302 if (spa_suspended(spa)) 3303 VERIFY(nvlist_add_uint64(*config, 3304 ZPOOL_CONFIG_SUSPENDED, 3305 spa->spa_failmode) == 0); 3306 3307 spa_add_spares(spa, *config); 3308 spa_add_l2cache(spa, *config); 3309 spa_add_feature_stats(spa, *config); 3310 } 3311 } 3312 3313 /* 3314 * We want to get the alternate root even for faulted pools, so we cheat 3315 * and call spa_lookup() directly. 3316 */ 3317 if (altroot) { 3318 if (spa == NULL) { 3319 mutex_enter(&spa_namespace_lock); 3320 spa = spa_lookup(name); 3321 if (spa) 3322 spa_altroot(spa, altroot, buflen); 3323 else 3324 altroot[0] = '\0'; 3325 spa = NULL; 3326 mutex_exit(&spa_namespace_lock); 3327 } else { 3328 spa_altroot(spa, altroot, buflen); 3329 } 3330 } 3331 3332 if (spa != NULL) { 3333 spa_config_exit(spa, SCL_CONFIG, FTAG); 3334 spa_close(spa, FTAG); 3335 } 3336 3337 return (error); 3338} 3339 3340/* 3341 * Validate that the auxiliary device array is well formed. We must have an 3342 * array of nvlists, each which describes a valid leaf vdev. If this is an 3343 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3344 * specified, as long as they are well-formed. 3345 */ 3346static int 3347spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3348 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3349 vdev_labeltype_t label) 3350{ 3351 nvlist_t **dev; 3352 uint_t i, ndev; 3353 vdev_t *vd; 3354 int error; 3355 3356 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3357 3358 /* 3359 * It's acceptable to have no devs specified. 3360 */ 3361 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3362 return (0); 3363 3364 if (ndev == 0) 3365 return (SET_ERROR(EINVAL)); 3366 3367 /* 3368 * Make sure the pool is formatted with a version that supports this 3369 * device type. 3370 */ 3371 if (spa_version(spa) < version) 3372 return (SET_ERROR(ENOTSUP)); 3373 3374 /* 3375 * Set the pending device list so we correctly handle device in-use 3376 * checking. 3377 */ 3378 sav->sav_pending = dev; 3379 sav->sav_npending = ndev; 3380 3381 for (i = 0; i < ndev; i++) { 3382 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3383 mode)) != 0) 3384 goto out; 3385 3386 if (!vd->vdev_ops->vdev_op_leaf) { 3387 vdev_free(vd); 3388 error = SET_ERROR(EINVAL); 3389 goto out; 3390 } 3391 3392 /* 3393 * The L2ARC currently only supports disk devices in 3394 * kernel context. For user-level testing, we allow it. 3395 */ 3396#ifdef _KERNEL 3397 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3398 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3399 error = SET_ERROR(ENOTBLK); 3400 vdev_free(vd); 3401 goto out; 3402 } 3403#endif 3404 vd->vdev_top = vd; 3405 3406 if ((error = vdev_open(vd)) == 0 && 3407 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3408 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3409 vd->vdev_guid) == 0); 3410 } 3411 3412 vdev_free(vd); 3413 3414 if (error && 3415 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3416 goto out; 3417 else 3418 error = 0; 3419 } 3420 3421out: 3422 sav->sav_pending = NULL; 3423 sav->sav_npending = 0; 3424 return (error); 3425} 3426 3427static int 3428spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3429{ 3430 int error; 3431 3432 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3433 3434 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3435 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3436 VDEV_LABEL_SPARE)) != 0) { 3437 return (error); 3438 } 3439 3440 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3441 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3442 VDEV_LABEL_L2CACHE)); 3443} 3444 3445static void 3446spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3447 const char *config) 3448{ 3449 int i; 3450 3451 if (sav->sav_config != NULL) { 3452 nvlist_t **olddevs; 3453 uint_t oldndevs; 3454 nvlist_t **newdevs; 3455 3456 /* 3457 * Generate new dev list by concatentating with the 3458 * current dev list. 3459 */ 3460 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3461 &olddevs, &oldndevs) == 0); 3462 3463 newdevs = kmem_alloc(sizeof (void *) * 3464 (ndevs + oldndevs), KM_SLEEP); 3465 for (i = 0; i < oldndevs; i++) 3466 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3467 KM_SLEEP) == 0); 3468 for (i = 0; i < ndevs; i++) 3469 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3470 KM_SLEEP) == 0); 3471 3472 VERIFY(nvlist_remove(sav->sav_config, config, 3473 DATA_TYPE_NVLIST_ARRAY) == 0); 3474 3475 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3476 config, newdevs, ndevs + oldndevs) == 0); 3477 for (i = 0; i < oldndevs + ndevs; i++) 3478 nvlist_free(newdevs[i]); 3479 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3480 } else { 3481 /* 3482 * Generate a new dev list. 3483 */ 3484 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3485 KM_SLEEP) == 0); 3486 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3487 devs, ndevs) == 0); 3488 } 3489} 3490 3491/* 3492 * Stop and drop level 2 ARC devices 3493 */ 3494void 3495spa_l2cache_drop(spa_t *spa) 3496{ 3497 vdev_t *vd; 3498 int i; 3499 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3500 3501 for (i = 0; i < sav->sav_count; i++) { 3502 uint64_t pool; 3503 3504 vd = sav->sav_vdevs[i]; 3505 ASSERT(vd != NULL); 3506 3507 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3508 pool != 0ULL && l2arc_vdev_present(vd)) 3509 l2arc_remove_vdev(vd); 3510 } 3511} 3512 3513/* 3514 * Pool Creation 3515 */ 3516int 3517spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3518 nvlist_t *zplprops) 3519{ 3520 spa_t *spa; 3521 char *altroot = NULL; 3522 vdev_t *rvd; 3523 dsl_pool_t *dp; 3524 dmu_tx_t *tx; 3525 int error = 0; 3526 uint64_t txg = TXG_INITIAL; 3527 nvlist_t **spares, **l2cache; 3528 uint_t nspares, nl2cache; 3529 uint64_t version, obj; 3530 boolean_t has_features; 3531 3532 /* 3533 * If this pool already exists, return failure. 3534 */ 3535 mutex_enter(&spa_namespace_lock); 3536 if (spa_lookup(pool) != NULL) { 3537 mutex_exit(&spa_namespace_lock); 3538 return (SET_ERROR(EEXIST)); 3539 } 3540 3541 /* 3542 * Allocate a new spa_t structure. 3543 */ 3544 (void) nvlist_lookup_string(props, 3545 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3546 spa = spa_add(pool, NULL, altroot); 3547 spa_activate(spa, spa_mode_global); 3548 3549 if (props && (error = spa_prop_validate(spa, props))) { 3550 spa_deactivate(spa); 3551 spa_remove(spa); 3552 mutex_exit(&spa_namespace_lock); 3553 return (error); 3554 } 3555 3556 has_features = B_FALSE; 3557 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3558 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3559 if (zpool_prop_feature(nvpair_name(elem))) 3560 has_features = B_TRUE; 3561 } 3562 3563 if (has_features || nvlist_lookup_uint64(props, 3564 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3565 version = SPA_VERSION; 3566 } 3567 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3568 3569 spa->spa_first_txg = txg; 3570 spa->spa_uberblock.ub_txg = txg - 1; 3571 spa->spa_uberblock.ub_version = version; 3572 spa->spa_ubsync = spa->spa_uberblock; 3573 3574 /* 3575 * Create "The Godfather" zio to hold all async IOs 3576 */ 3577 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3578 KM_SLEEP); 3579 for (int i = 0; i < max_ncpus; i++) { 3580 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3581 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3582 ZIO_FLAG_GODFATHER); 3583 } 3584 3585 /* 3586 * Create the root vdev. 3587 */ 3588 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3589 3590 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3591 3592 ASSERT(error != 0 || rvd != NULL); 3593 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3594 3595 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3596 error = SET_ERROR(EINVAL); 3597 3598 if (error == 0 && 3599 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3600 (error = spa_validate_aux(spa, nvroot, txg, 3601 VDEV_ALLOC_ADD)) == 0) { 3602 for (int c = 0; c < rvd->vdev_children; c++) { 3603 vdev_ashift_optimize(rvd->vdev_child[c]); 3604 vdev_metaslab_set_size(rvd->vdev_child[c]); 3605 vdev_expand(rvd->vdev_child[c], txg); 3606 } 3607 } 3608 3609 spa_config_exit(spa, SCL_ALL, FTAG); 3610 3611 if (error != 0) { 3612 spa_unload(spa); 3613 spa_deactivate(spa); 3614 spa_remove(spa); 3615 mutex_exit(&spa_namespace_lock); 3616 return (error); 3617 } 3618 3619 /* 3620 * Get the list of spares, if specified. 3621 */ 3622 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3623 &spares, &nspares) == 0) { 3624 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3625 KM_SLEEP) == 0); 3626 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3627 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3628 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3629 spa_load_spares(spa); 3630 spa_config_exit(spa, SCL_ALL, FTAG); 3631 spa->spa_spares.sav_sync = B_TRUE; 3632 } 3633 3634 /* 3635 * Get the list of level 2 cache devices, if specified. 3636 */ 3637 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3638 &l2cache, &nl2cache) == 0) { 3639 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3640 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3641 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3642 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3643 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3644 spa_load_l2cache(spa); 3645 spa_config_exit(spa, SCL_ALL, FTAG); 3646 spa->spa_l2cache.sav_sync = B_TRUE; 3647 } 3648 3649 spa->spa_is_initializing = B_TRUE; 3650 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3651 spa->spa_meta_objset = dp->dp_meta_objset; 3652 spa->spa_is_initializing = B_FALSE; 3653 3654 /* 3655 * Create DDTs (dedup tables). 3656 */ 3657 ddt_create(spa); 3658 3659 spa_update_dspace(spa); 3660 3661 tx = dmu_tx_create_assigned(dp, txg); 3662 3663 /* 3664 * Create the pool config object. 3665 */ 3666 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3667 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3668 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3669 3670 if (zap_add(spa->spa_meta_objset, 3671 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3672 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3673 cmn_err(CE_PANIC, "failed to add pool config"); 3674 } 3675 3676 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3677 spa_feature_create_zap_objects(spa, tx); 3678 3679 if (zap_add(spa->spa_meta_objset, 3680 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3681 sizeof (uint64_t), 1, &version, tx) != 0) { 3682 cmn_err(CE_PANIC, "failed to add pool version"); 3683 } 3684 3685 /* Newly created pools with the right version are always deflated. */ 3686 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3687 spa->spa_deflate = TRUE; 3688 if (zap_add(spa->spa_meta_objset, 3689 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3690 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3691 cmn_err(CE_PANIC, "failed to add deflate"); 3692 } 3693 } 3694 3695 /* 3696 * Create the deferred-free bpobj. Turn off compression 3697 * because sync-to-convergence takes longer if the blocksize 3698 * keeps changing. 3699 */ 3700 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3701 dmu_object_set_compress(spa->spa_meta_objset, obj, 3702 ZIO_COMPRESS_OFF, tx); 3703 if (zap_add(spa->spa_meta_objset, 3704 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3705 sizeof (uint64_t), 1, &obj, tx) != 0) { 3706 cmn_err(CE_PANIC, "failed to add bpobj"); 3707 } 3708 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3709 spa->spa_meta_objset, obj)); 3710 3711 /* 3712 * Create the pool's history object. 3713 */ 3714 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3715 spa_history_create_obj(spa, tx); 3716 3717 /* 3718 * Set pool properties. 3719 */ 3720 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3721 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3722 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3723 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3724 3725 if (props != NULL) { 3726 spa_configfile_set(spa, props, B_FALSE); 3727 spa_sync_props(props, tx); 3728 } 3729 3730 dmu_tx_commit(tx); 3731 3732 spa->spa_sync_on = B_TRUE; 3733 txg_sync_start(spa->spa_dsl_pool); 3734 3735 /* 3736 * We explicitly wait for the first transaction to complete so that our 3737 * bean counters are appropriately updated. 3738 */ 3739 txg_wait_synced(spa->spa_dsl_pool, txg); 3740 3741 spa_config_sync(spa, B_FALSE, B_TRUE); 3742 3743 spa_history_log_version(spa, "create"); 3744 3745 spa->spa_minref = refcount_count(&spa->spa_refcount); 3746 3747 mutex_exit(&spa_namespace_lock); 3748 3749 return (0); 3750} 3751 3752#ifdef _KERNEL 3753#if defined(sun) 3754/* 3755 * Get the root pool information from the root disk, then import the root pool 3756 * during the system boot up time. 3757 */ 3758extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3759 3760static nvlist_t * 3761spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3762{ 3763 nvlist_t *config; 3764 nvlist_t *nvtop, *nvroot; 3765 uint64_t pgid; 3766 3767 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3768 return (NULL); 3769 3770 /* 3771 * Add this top-level vdev to the child array. 3772 */ 3773 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3774 &nvtop) == 0); 3775 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3776 &pgid) == 0); 3777 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3778 3779 /* 3780 * Put this pool's top-level vdevs into a root vdev. 3781 */ 3782 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3783 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3784 VDEV_TYPE_ROOT) == 0); 3785 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3786 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3787 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3788 &nvtop, 1) == 0); 3789 3790 /* 3791 * Replace the existing vdev_tree with the new root vdev in 3792 * this pool's configuration (remove the old, add the new). 3793 */ 3794 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3795 nvlist_free(nvroot); 3796 return (config); 3797} 3798 3799/* 3800 * Walk the vdev tree and see if we can find a device with "better" 3801 * configuration. A configuration is "better" if the label on that 3802 * device has a more recent txg. 3803 */ 3804static void 3805spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3806{ 3807 for (int c = 0; c < vd->vdev_children; c++) 3808 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3809 3810 if (vd->vdev_ops->vdev_op_leaf) { 3811 nvlist_t *label; 3812 uint64_t label_txg; 3813 3814 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3815 &label) != 0) 3816 return; 3817 3818 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3819 &label_txg) == 0); 3820 3821 /* 3822 * Do we have a better boot device? 3823 */ 3824 if (label_txg > *txg) { 3825 *txg = label_txg; 3826 *avd = vd; 3827 } 3828 nvlist_free(label); 3829 } 3830} 3831 3832/* 3833 * Import a root pool. 3834 * 3835 * For x86. devpath_list will consist of devid and/or physpath name of 3836 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3837 * The GRUB "findroot" command will return the vdev we should boot. 3838 * 3839 * For Sparc, devpath_list consists the physpath name of the booting device 3840 * no matter the rootpool is a single device pool or a mirrored pool. 3841 * e.g. 3842 * "/pci@1f,0/ide@d/disk@0,0:a" 3843 */ 3844int 3845spa_import_rootpool(char *devpath, char *devid) 3846{ 3847 spa_t *spa; 3848 vdev_t *rvd, *bvd, *avd = NULL; 3849 nvlist_t *config, *nvtop; 3850 uint64_t guid, txg; 3851 char *pname; 3852 int error; 3853 3854 /* 3855 * Read the label from the boot device and generate a configuration. 3856 */ 3857 config = spa_generate_rootconf(devpath, devid, &guid); 3858#if defined(_OBP) && defined(_KERNEL) 3859 if (config == NULL) { 3860 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3861 /* iscsi boot */ 3862 get_iscsi_bootpath_phy(devpath); 3863 config = spa_generate_rootconf(devpath, devid, &guid); 3864 } 3865 } 3866#endif 3867 if (config == NULL) { 3868 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3869 devpath); 3870 return (SET_ERROR(EIO)); 3871 } 3872 3873 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3874 &pname) == 0); 3875 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3876 3877 mutex_enter(&spa_namespace_lock); 3878 if ((spa = spa_lookup(pname)) != NULL) { 3879 /* 3880 * Remove the existing root pool from the namespace so that we 3881 * can replace it with the correct config we just read in. 3882 */ 3883 spa_remove(spa); 3884 } 3885 3886 spa = spa_add(pname, config, NULL); 3887 spa->spa_is_root = B_TRUE; 3888 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3889 3890 /* 3891 * Build up a vdev tree based on the boot device's label config. 3892 */ 3893 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3894 &nvtop) == 0); 3895 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3896 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3897 VDEV_ALLOC_ROOTPOOL); 3898 spa_config_exit(spa, SCL_ALL, FTAG); 3899 if (error) { 3900 mutex_exit(&spa_namespace_lock); 3901 nvlist_free(config); 3902 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3903 pname); 3904 return (error); 3905 } 3906 3907 /* 3908 * Get the boot vdev. 3909 */ 3910 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3911 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3912 (u_longlong_t)guid); 3913 error = SET_ERROR(ENOENT); 3914 goto out; 3915 } 3916 3917 /* 3918 * Determine if there is a better boot device. 3919 */ 3920 avd = bvd; 3921 spa_alt_rootvdev(rvd, &avd, &txg); 3922 if (avd != bvd) { 3923 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3924 "try booting from '%s'", avd->vdev_path); 3925 error = SET_ERROR(EINVAL); 3926 goto out; 3927 } 3928 3929 /* 3930 * If the boot device is part of a spare vdev then ensure that 3931 * we're booting off the active spare. 3932 */ 3933 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3934 !bvd->vdev_isspare) { 3935 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3936 "try booting from '%s'", 3937 bvd->vdev_parent-> 3938 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3939 error = SET_ERROR(EINVAL); 3940 goto out; 3941 } 3942 3943 error = 0; 3944out: 3945 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3946 vdev_free(rvd); 3947 spa_config_exit(spa, SCL_ALL, FTAG); 3948 mutex_exit(&spa_namespace_lock); 3949 3950 nvlist_free(config); 3951 return (error); 3952} 3953 3954#else 3955 3956extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3957 uint64_t *count); 3958 3959static nvlist_t * 3960spa_generate_rootconf(const char *name) 3961{ 3962 nvlist_t **configs, **tops; 3963 nvlist_t *config; 3964 nvlist_t *best_cfg, *nvtop, *nvroot; 3965 uint64_t *holes; 3966 uint64_t best_txg; 3967 uint64_t nchildren; 3968 uint64_t pgid; 3969 uint64_t count; 3970 uint64_t i; 3971 uint_t nholes; 3972 3973 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3974 return (NULL); 3975 3976 ASSERT3U(count, !=, 0); 3977 best_txg = 0; 3978 for (i = 0; i < count; i++) { 3979 uint64_t txg; 3980 3981 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 3982 &txg) == 0); 3983 if (txg > best_txg) { 3984 best_txg = txg; 3985 best_cfg = configs[i]; 3986 } 3987 } 3988 3989 nchildren = 1; 3990 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 3991 holes = NULL; 3992 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 3993 &holes, &nholes); 3994 3995 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 3996 for (i = 0; i < nchildren; i++) { 3997 if (i >= count) 3998 break; 3999 if (configs[i] == NULL) 4000 continue; 4001 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4002 &nvtop) == 0); 4003 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4004 } 4005 for (i = 0; holes != NULL && i < nholes; i++) { 4006 if (i >= nchildren) 4007 continue; 4008 if (tops[holes[i]] != NULL) 4009 continue; 4010 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4011 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4012 VDEV_TYPE_HOLE) == 0); 4013 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4014 holes[i]) == 0); 4015 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4016 0) == 0); 4017 } 4018 for (i = 0; i < nchildren; i++) { 4019 if (tops[i] != NULL) 4020 continue; 4021 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4022 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4023 VDEV_TYPE_MISSING) == 0); 4024 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4025 i) == 0); 4026 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4027 0) == 0); 4028 } 4029 4030 /* 4031 * Create pool config based on the best vdev config. 4032 */ 4033 nvlist_dup(best_cfg, &config, KM_SLEEP); 4034 4035 /* 4036 * Put this pool's top-level vdevs into a root vdev. 4037 */ 4038 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4039 &pgid) == 0); 4040 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4041 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4042 VDEV_TYPE_ROOT) == 0); 4043 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4044 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4045 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4046 tops, nchildren) == 0); 4047 4048 /* 4049 * Replace the existing vdev_tree with the new root vdev in 4050 * this pool's configuration (remove the old, add the new). 4051 */ 4052 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4053 4054 /* 4055 * Drop vdev config elements that should not be present at pool level. 4056 */ 4057 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4058 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4059 4060 for (i = 0; i < count; i++) 4061 nvlist_free(configs[i]); 4062 kmem_free(configs, count * sizeof(void *)); 4063 for (i = 0; i < nchildren; i++) 4064 nvlist_free(tops[i]); 4065 kmem_free(tops, nchildren * sizeof(void *)); 4066 nvlist_free(nvroot); 4067 return (config); 4068} 4069 4070int 4071spa_import_rootpool(const char *name) 4072{ 4073 spa_t *spa; 4074 vdev_t *rvd, *bvd, *avd = NULL; 4075 nvlist_t *config, *nvtop; 4076 uint64_t txg; 4077 char *pname; 4078 int error; 4079 4080 /* 4081 * Read the label from the boot device and generate a configuration. 4082 */ 4083 config = spa_generate_rootconf(name); 4084 4085 mutex_enter(&spa_namespace_lock); 4086 if (config != NULL) { 4087 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4088 &pname) == 0 && strcmp(name, pname) == 0); 4089 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4090 == 0); 4091 4092 if ((spa = spa_lookup(pname)) != NULL) { 4093 /* 4094 * Remove the existing root pool from the namespace so 4095 * that we can replace it with the correct config 4096 * we just read in. 4097 */ 4098 spa_remove(spa); 4099 } 4100 spa = spa_add(pname, config, NULL); 4101 4102 /* 4103 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4104 * via spa_version(). 4105 */ 4106 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4107 &spa->spa_ubsync.ub_version) != 0) 4108 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4109 } else if ((spa = spa_lookup(name)) == NULL) { 4110 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4111 name); 4112 return (EIO); 4113 } else { 4114 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4115 } 4116 spa->spa_is_root = B_TRUE; 4117 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4118 4119 /* 4120 * Build up a vdev tree based on the boot device's label config. 4121 */ 4122 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4123 &nvtop) == 0); 4124 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4125 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4126 VDEV_ALLOC_ROOTPOOL); 4127 spa_config_exit(spa, SCL_ALL, FTAG); 4128 if (error) { 4129 mutex_exit(&spa_namespace_lock); 4130 nvlist_free(config); 4131 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4132 pname); 4133 return (error); 4134 } 4135 4136 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4137 vdev_free(rvd); 4138 spa_config_exit(spa, SCL_ALL, FTAG); 4139 mutex_exit(&spa_namespace_lock); 4140 4141 nvlist_free(config); 4142 return (0); 4143} 4144 4145#endif /* sun */ 4146#endif 4147 4148/* 4149 * Import a non-root pool into the system. 4150 */ 4151int 4152spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4153{ 4154 spa_t *spa; 4155 char *altroot = NULL; 4156 spa_load_state_t state = SPA_LOAD_IMPORT; 4157 zpool_rewind_policy_t policy; 4158 uint64_t mode = spa_mode_global; 4159 uint64_t readonly = B_FALSE; 4160 int error; 4161 nvlist_t *nvroot; 4162 nvlist_t **spares, **l2cache; 4163 uint_t nspares, nl2cache; 4164 4165 /* 4166 * If a pool with this name exists, return failure. 4167 */ 4168 mutex_enter(&spa_namespace_lock); 4169 if (spa_lookup(pool) != NULL) { 4170 mutex_exit(&spa_namespace_lock); 4171 return (SET_ERROR(EEXIST)); 4172 } 4173 4174 /* 4175 * Create and initialize the spa structure. 4176 */ 4177 (void) nvlist_lookup_string(props, 4178 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4179 (void) nvlist_lookup_uint64(props, 4180 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4181 if (readonly) 4182 mode = FREAD; 4183 spa = spa_add(pool, config, altroot); 4184 spa->spa_import_flags = flags; 4185 4186 /* 4187 * Verbatim import - Take a pool and insert it into the namespace 4188 * as if it had been loaded at boot. 4189 */ 4190 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4191 if (props != NULL) 4192 spa_configfile_set(spa, props, B_FALSE); 4193 4194 spa_config_sync(spa, B_FALSE, B_TRUE); 4195 4196 mutex_exit(&spa_namespace_lock); 4197 return (0); 4198 } 4199 4200 spa_activate(spa, mode); 4201 4202 /* 4203 * Don't start async tasks until we know everything is healthy. 4204 */ 4205 spa_async_suspend(spa); 4206 4207 zpool_get_rewind_policy(config, &policy); 4208 if (policy.zrp_request & ZPOOL_DO_REWIND) 4209 state = SPA_LOAD_RECOVER; 4210 4211 /* 4212 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4213 * because the user-supplied config is actually the one to trust when 4214 * doing an import. 4215 */ 4216 if (state != SPA_LOAD_RECOVER) 4217 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4218 4219 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4220 policy.zrp_request); 4221 4222 /* 4223 * Propagate anything learned while loading the pool and pass it 4224 * back to caller (i.e. rewind info, missing devices, etc). 4225 */ 4226 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4227 spa->spa_load_info) == 0); 4228 4229 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4230 /* 4231 * Toss any existing sparelist, as it doesn't have any validity 4232 * anymore, and conflicts with spa_has_spare(). 4233 */ 4234 if (spa->spa_spares.sav_config) { 4235 nvlist_free(spa->spa_spares.sav_config); 4236 spa->spa_spares.sav_config = NULL; 4237 spa_load_spares(spa); 4238 } 4239 if (spa->spa_l2cache.sav_config) { 4240 nvlist_free(spa->spa_l2cache.sav_config); 4241 spa->spa_l2cache.sav_config = NULL; 4242 spa_load_l2cache(spa); 4243 } 4244 4245 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4246 &nvroot) == 0); 4247 if (error == 0) 4248 error = spa_validate_aux(spa, nvroot, -1ULL, 4249 VDEV_ALLOC_SPARE); 4250 if (error == 0) 4251 error = spa_validate_aux(spa, nvroot, -1ULL, 4252 VDEV_ALLOC_L2CACHE); 4253 spa_config_exit(spa, SCL_ALL, FTAG); 4254 4255 if (props != NULL) 4256 spa_configfile_set(spa, props, B_FALSE); 4257 4258 if (error != 0 || (props && spa_writeable(spa) && 4259 (error = spa_prop_set(spa, props)))) { 4260 spa_unload(spa); 4261 spa_deactivate(spa); 4262 spa_remove(spa); 4263 mutex_exit(&spa_namespace_lock); 4264 return (error); 4265 } 4266 4267 spa_async_resume(spa); 4268 4269 /* 4270 * Override any spares and level 2 cache devices as specified by 4271 * the user, as these may have correct device names/devids, etc. 4272 */ 4273 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4274 &spares, &nspares) == 0) { 4275 if (spa->spa_spares.sav_config) 4276 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4277 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4278 else 4279 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4280 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4281 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4282 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4283 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4284 spa_load_spares(spa); 4285 spa_config_exit(spa, SCL_ALL, FTAG); 4286 spa->spa_spares.sav_sync = B_TRUE; 4287 } 4288 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4289 &l2cache, &nl2cache) == 0) { 4290 if (spa->spa_l2cache.sav_config) 4291 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4292 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4293 else 4294 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4295 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4296 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4297 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4298 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4299 spa_load_l2cache(spa); 4300 spa_config_exit(spa, SCL_ALL, FTAG); 4301 spa->spa_l2cache.sav_sync = B_TRUE; 4302 } 4303 4304 /* 4305 * Check for any removed devices. 4306 */ 4307 if (spa->spa_autoreplace) { 4308 spa_aux_check_removed(&spa->spa_spares); 4309 spa_aux_check_removed(&spa->spa_l2cache); 4310 } 4311 4312 if (spa_writeable(spa)) { 4313 /* 4314 * Update the config cache to include the newly-imported pool. 4315 */ 4316 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4317 } 4318 4319 /* 4320 * It's possible that the pool was expanded while it was exported. 4321 * We kick off an async task to handle this for us. 4322 */ 4323 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4324 4325 mutex_exit(&spa_namespace_lock); 4326 spa_history_log_version(spa, "import"); 4327 4328#ifdef __FreeBSD__ 4329#ifdef _KERNEL 4330 zvol_create_minors(pool); 4331#endif 4332#endif 4333 return (0); 4334} 4335 4336nvlist_t * 4337spa_tryimport(nvlist_t *tryconfig) 4338{ 4339 nvlist_t *config = NULL; 4340 char *poolname; 4341 spa_t *spa; 4342 uint64_t state; 4343 int error; 4344 4345 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4346 return (NULL); 4347 4348 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4349 return (NULL); 4350 4351 /* 4352 * Create and initialize the spa structure. 4353 */ 4354 mutex_enter(&spa_namespace_lock); 4355 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4356 spa_activate(spa, FREAD); 4357 4358 /* 4359 * Pass off the heavy lifting to spa_load(). 4360 * Pass TRUE for mosconfig because the user-supplied config 4361 * is actually the one to trust when doing an import. 4362 */ 4363 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4364 4365 /* 4366 * If 'tryconfig' was at least parsable, return the current config. 4367 */ 4368 if (spa->spa_root_vdev != NULL) { 4369 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4370 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4371 poolname) == 0); 4372 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4373 state) == 0); 4374 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4375 spa->spa_uberblock.ub_timestamp) == 0); 4376 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4377 spa->spa_load_info) == 0); 4378 4379 /* 4380 * If the bootfs property exists on this pool then we 4381 * copy it out so that external consumers can tell which 4382 * pools are bootable. 4383 */ 4384 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4385 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4386 4387 /* 4388 * We have to play games with the name since the 4389 * pool was opened as TRYIMPORT_NAME. 4390 */ 4391 if (dsl_dsobj_to_dsname(spa_name(spa), 4392 spa->spa_bootfs, tmpname) == 0) { 4393 char *cp; 4394 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4395 4396 cp = strchr(tmpname, '/'); 4397 if (cp == NULL) { 4398 (void) strlcpy(dsname, tmpname, 4399 MAXPATHLEN); 4400 } else { 4401 (void) snprintf(dsname, MAXPATHLEN, 4402 "%s/%s", poolname, ++cp); 4403 } 4404 VERIFY(nvlist_add_string(config, 4405 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4406 kmem_free(dsname, MAXPATHLEN); 4407 } 4408 kmem_free(tmpname, MAXPATHLEN); 4409 } 4410 4411 /* 4412 * Add the list of hot spares and level 2 cache devices. 4413 */ 4414 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4415 spa_add_spares(spa, config); 4416 spa_add_l2cache(spa, config); 4417 spa_config_exit(spa, SCL_CONFIG, FTAG); 4418 } 4419 4420 spa_unload(spa); 4421 spa_deactivate(spa); 4422 spa_remove(spa); 4423 mutex_exit(&spa_namespace_lock); 4424 4425 return (config); 4426} 4427 4428/* 4429 * Pool export/destroy 4430 * 4431 * The act of destroying or exporting a pool is very simple. We make sure there 4432 * is no more pending I/O and any references to the pool are gone. Then, we 4433 * update the pool state and sync all the labels to disk, removing the 4434 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4435 * we don't sync the labels or remove the configuration cache. 4436 */ 4437static int 4438spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4439 boolean_t force, boolean_t hardforce) 4440{ 4441 spa_t *spa; 4442 4443 if (oldconfig) 4444 *oldconfig = NULL; 4445 4446 if (!(spa_mode_global & FWRITE)) 4447 return (SET_ERROR(EROFS)); 4448 4449 mutex_enter(&spa_namespace_lock); 4450 if ((spa = spa_lookup(pool)) == NULL) { 4451 mutex_exit(&spa_namespace_lock); 4452 return (SET_ERROR(ENOENT)); 4453 } 4454 4455 /* 4456 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4457 * reacquire the namespace lock, and see if we can export. 4458 */ 4459 spa_open_ref(spa, FTAG); 4460 mutex_exit(&spa_namespace_lock); 4461 spa_async_suspend(spa); 4462 mutex_enter(&spa_namespace_lock); 4463 spa_close(spa, FTAG); 4464 4465 /* 4466 * The pool will be in core if it's openable, 4467 * in which case we can modify its state. 4468 */ 4469 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4470 /* 4471 * Objsets may be open only because they're dirty, so we 4472 * have to force it to sync before checking spa_refcnt. 4473 */ 4474 txg_wait_synced(spa->spa_dsl_pool, 0); 4475 4476 /* 4477 * A pool cannot be exported or destroyed if there are active 4478 * references. If we are resetting a pool, allow references by 4479 * fault injection handlers. 4480 */ 4481 if (!spa_refcount_zero(spa) || 4482 (spa->spa_inject_ref != 0 && 4483 new_state != POOL_STATE_UNINITIALIZED)) { 4484 spa_async_resume(spa); 4485 mutex_exit(&spa_namespace_lock); 4486 return (SET_ERROR(EBUSY)); 4487 } 4488 4489 /* 4490 * A pool cannot be exported if it has an active shared spare. 4491 * This is to prevent other pools stealing the active spare 4492 * from an exported pool. At user's own will, such pool can 4493 * be forcedly exported. 4494 */ 4495 if (!force && new_state == POOL_STATE_EXPORTED && 4496 spa_has_active_shared_spare(spa)) { 4497 spa_async_resume(spa); 4498 mutex_exit(&spa_namespace_lock); 4499 return (SET_ERROR(EXDEV)); 4500 } 4501 4502 /* 4503 * We want this to be reflected on every label, 4504 * so mark them all dirty. spa_unload() will do the 4505 * final sync that pushes these changes out. 4506 */ 4507 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4508 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4509 spa->spa_state = new_state; 4510 spa->spa_final_txg = spa_last_synced_txg(spa) + 4511 TXG_DEFER_SIZE + 1; 4512 vdev_config_dirty(spa->spa_root_vdev); 4513 spa_config_exit(spa, SCL_ALL, FTAG); 4514 } 4515 } 4516 4517 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4518 4519 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4520 spa_unload(spa); 4521 spa_deactivate(spa); 4522 } 4523 4524 if (oldconfig && spa->spa_config) 4525 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4526 4527 if (new_state != POOL_STATE_UNINITIALIZED) { 4528 if (!hardforce) 4529 spa_config_sync(spa, B_TRUE, B_TRUE); 4530 spa_remove(spa); 4531 } 4532 mutex_exit(&spa_namespace_lock); 4533 4534 return (0); 4535} 4536 4537/* 4538 * Destroy a storage pool. 4539 */ 4540int 4541spa_destroy(char *pool) 4542{ 4543 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4544 B_FALSE, B_FALSE)); 4545} 4546 4547/* 4548 * Export a storage pool. 4549 */ 4550int 4551spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4552 boolean_t hardforce) 4553{ 4554 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4555 force, hardforce)); 4556} 4557 4558/* 4559 * Similar to spa_export(), this unloads the spa_t without actually removing it 4560 * from the namespace in any way. 4561 */ 4562int 4563spa_reset(char *pool) 4564{ 4565 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4566 B_FALSE, B_FALSE)); 4567} 4568 4569/* 4570 * ========================================================================== 4571 * Device manipulation 4572 * ========================================================================== 4573 */ 4574 4575/* 4576 * Add a device to a storage pool. 4577 */ 4578int 4579spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4580{ 4581 uint64_t txg, id; 4582 int error; 4583 vdev_t *rvd = spa->spa_root_vdev; 4584 vdev_t *vd, *tvd; 4585 nvlist_t **spares, **l2cache; 4586 uint_t nspares, nl2cache; 4587 4588 ASSERT(spa_writeable(spa)); 4589 4590 txg = spa_vdev_enter(spa); 4591 4592 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4593 VDEV_ALLOC_ADD)) != 0) 4594 return (spa_vdev_exit(spa, NULL, txg, error)); 4595 4596 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4597 4598 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4599 &nspares) != 0) 4600 nspares = 0; 4601 4602 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4603 &nl2cache) != 0) 4604 nl2cache = 0; 4605 4606 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4607 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4608 4609 if (vd->vdev_children != 0 && 4610 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4611 return (spa_vdev_exit(spa, vd, txg, error)); 4612 4613 /* 4614 * We must validate the spares and l2cache devices after checking the 4615 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4616 */ 4617 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4618 return (spa_vdev_exit(spa, vd, txg, error)); 4619 4620 /* 4621 * Transfer each new top-level vdev from vd to rvd. 4622 */ 4623 for (int c = 0; c < vd->vdev_children; c++) { 4624 4625 /* 4626 * Set the vdev id to the first hole, if one exists. 4627 */ 4628 for (id = 0; id < rvd->vdev_children; id++) { 4629 if (rvd->vdev_child[id]->vdev_ishole) { 4630 vdev_free(rvd->vdev_child[id]); 4631 break; 4632 } 4633 } 4634 tvd = vd->vdev_child[c]; 4635 vdev_remove_child(vd, tvd); 4636 tvd->vdev_id = id; 4637 vdev_add_child(rvd, tvd); 4638 vdev_config_dirty(tvd); 4639 } 4640 4641 if (nspares != 0) { 4642 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4643 ZPOOL_CONFIG_SPARES); 4644 spa_load_spares(spa); 4645 spa->spa_spares.sav_sync = B_TRUE; 4646 } 4647 4648 if (nl2cache != 0) { 4649 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4650 ZPOOL_CONFIG_L2CACHE); 4651 spa_load_l2cache(spa); 4652 spa->spa_l2cache.sav_sync = B_TRUE; 4653 } 4654 4655 /* 4656 * We have to be careful when adding new vdevs to an existing pool. 4657 * If other threads start allocating from these vdevs before we 4658 * sync the config cache, and we lose power, then upon reboot we may 4659 * fail to open the pool because there are DVAs that the config cache 4660 * can't translate. Therefore, we first add the vdevs without 4661 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4662 * and then let spa_config_update() initialize the new metaslabs. 4663 * 4664 * spa_load() checks for added-but-not-initialized vdevs, so that 4665 * if we lose power at any point in this sequence, the remaining 4666 * steps will be completed the next time we load the pool. 4667 */ 4668 (void) spa_vdev_exit(spa, vd, txg, 0); 4669 4670 mutex_enter(&spa_namespace_lock); 4671 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4672 mutex_exit(&spa_namespace_lock); 4673 4674 return (0); 4675} 4676 4677/* 4678 * Attach a device to a mirror. The arguments are the path to any device 4679 * in the mirror, and the nvroot for the new device. If the path specifies 4680 * a device that is not mirrored, we automatically insert the mirror vdev. 4681 * 4682 * If 'replacing' is specified, the new device is intended to replace the 4683 * existing device; in this case the two devices are made into their own 4684 * mirror using the 'replacing' vdev, which is functionally identical to 4685 * the mirror vdev (it actually reuses all the same ops) but has a few 4686 * extra rules: you can't attach to it after it's been created, and upon 4687 * completion of resilvering, the first disk (the one being replaced) 4688 * is automatically detached. 4689 */ 4690int 4691spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4692{ 4693 uint64_t txg, dtl_max_txg; 4694 vdev_t *rvd = spa->spa_root_vdev; 4695 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4696 vdev_ops_t *pvops; 4697 char *oldvdpath, *newvdpath; 4698 int newvd_isspare; 4699 int error; 4700 4701 ASSERT(spa_writeable(spa)); 4702 4703 txg = spa_vdev_enter(spa); 4704 4705 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4706 4707 if (oldvd == NULL) 4708 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4709 4710 if (!oldvd->vdev_ops->vdev_op_leaf) 4711 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4712 4713 pvd = oldvd->vdev_parent; 4714 4715 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4716 VDEV_ALLOC_ATTACH)) != 0) 4717 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4718 4719 if (newrootvd->vdev_children != 1) 4720 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4721 4722 newvd = newrootvd->vdev_child[0]; 4723 4724 if (!newvd->vdev_ops->vdev_op_leaf) 4725 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4726 4727 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4728 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4729 4730 /* 4731 * Spares can't replace logs 4732 */ 4733 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4734 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4735 4736 if (!replacing) { 4737 /* 4738 * For attach, the only allowable parent is a mirror or the root 4739 * vdev. 4740 */ 4741 if (pvd->vdev_ops != &vdev_mirror_ops && 4742 pvd->vdev_ops != &vdev_root_ops) 4743 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4744 4745 pvops = &vdev_mirror_ops; 4746 } else { 4747 /* 4748 * Active hot spares can only be replaced by inactive hot 4749 * spares. 4750 */ 4751 if (pvd->vdev_ops == &vdev_spare_ops && 4752 oldvd->vdev_isspare && 4753 !spa_has_spare(spa, newvd->vdev_guid)) 4754 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4755 4756 /* 4757 * If the source is a hot spare, and the parent isn't already a 4758 * spare, then we want to create a new hot spare. Otherwise, we 4759 * want to create a replacing vdev. The user is not allowed to 4760 * attach to a spared vdev child unless the 'isspare' state is 4761 * the same (spare replaces spare, non-spare replaces 4762 * non-spare). 4763 */ 4764 if (pvd->vdev_ops == &vdev_replacing_ops && 4765 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4766 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4767 } else if (pvd->vdev_ops == &vdev_spare_ops && 4768 newvd->vdev_isspare != oldvd->vdev_isspare) { 4769 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4770 } 4771 4772 if (newvd->vdev_isspare) 4773 pvops = &vdev_spare_ops; 4774 else 4775 pvops = &vdev_replacing_ops; 4776 } 4777 4778 /* 4779 * Make sure the new device is big enough. 4780 */ 4781 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4782 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4783 4784 /* 4785 * The new device cannot have a higher alignment requirement 4786 * than the top-level vdev. 4787 */ 4788 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4789 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4790 4791 /* 4792 * If this is an in-place replacement, update oldvd's path and devid 4793 * to make it distinguishable from newvd, and unopenable from now on. 4794 */ 4795 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4796 spa_strfree(oldvd->vdev_path); 4797 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4798 KM_SLEEP); 4799 (void) sprintf(oldvd->vdev_path, "%s/%s", 4800 newvd->vdev_path, "old"); 4801 if (oldvd->vdev_devid != NULL) { 4802 spa_strfree(oldvd->vdev_devid); 4803 oldvd->vdev_devid = NULL; 4804 } 4805 } 4806 4807 /* mark the device being resilvered */ 4808 newvd->vdev_resilver_txg = txg; 4809 4810 /* 4811 * If the parent is not a mirror, or if we're replacing, insert the new 4812 * mirror/replacing/spare vdev above oldvd. 4813 */ 4814 if (pvd->vdev_ops != pvops) 4815 pvd = vdev_add_parent(oldvd, pvops); 4816 4817 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4818 ASSERT(pvd->vdev_ops == pvops); 4819 ASSERT(oldvd->vdev_parent == pvd); 4820 4821 /* 4822 * Extract the new device from its root and add it to pvd. 4823 */ 4824 vdev_remove_child(newrootvd, newvd); 4825 newvd->vdev_id = pvd->vdev_children; 4826 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4827 vdev_add_child(pvd, newvd); 4828 4829 tvd = newvd->vdev_top; 4830 ASSERT(pvd->vdev_top == tvd); 4831 ASSERT(tvd->vdev_parent == rvd); 4832 4833 vdev_config_dirty(tvd); 4834 4835 /* 4836 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4837 * for any dmu_sync-ed blocks. It will propagate upward when 4838 * spa_vdev_exit() calls vdev_dtl_reassess(). 4839 */ 4840 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4841 4842 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4843 dtl_max_txg - TXG_INITIAL); 4844 4845 if (newvd->vdev_isspare) { 4846 spa_spare_activate(newvd); 4847 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4848 } 4849 4850 oldvdpath = spa_strdup(oldvd->vdev_path); 4851 newvdpath = spa_strdup(newvd->vdev_path); 4852 newvd_isspare = newvd->vdev_isspare; 4853 4854 /* 4855 * Mark newvd's DTL dirty in this txg. 4856 */ 4857 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4858 4859 /* 4860 * Schedule the resilver to restart in the future. We do this to 4861 * ensure that dmu_sync-ed blocks have been stitched into the 4862 * respective datasets. 4863 */ 4864 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4865 4866 /* 4867 * Commit the config 4868 */ 4869 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4870 4871 spa_history_log_internal(spa, "vdev attach", NULL, 4872 "%s vdev=%s %s vdev=%s", 4873 replacing && newvd_isspare ? "spare in" : 4874 replacing ? "replace" : "attach", newvdpath, 4875 replacing ? "for" : "to", oldvdpath); 4876 4877 spa_strfree(oldvdpath); 4878 spa_strfree(newvdpath); 4879 4880 if (spa->spa_bootfs) 4881 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4882 4883 return (0); 4884} 4885 4886/* 4887 * Detach a device from a mirror or replacing vdev. 4888 * 4889 * If 'replace_done' is specified, only detach if the parent 4890 * is a replacing vdev. 4891 */ 4892int 4893spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4894{ 4895 uint64_t txg; 4896 int error; 4897 vdev_t *rvd = spa->spa_root_vdev; 4898 vdev_t *vd, *pvd, *cvd, *tvd; 4899 boolean_t unspare = B_FALSE; 4900 uint64_t unspare_guid = 0; 4901 char *vdpath; 4902 4903 ASSERT(spa_writeable(spa)); 4904 4905 txg = spa_vdev_enter(spa); 4906 4907 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4908 4909 if (vd == NULL) 4910 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4911 4912 if (!vd->vdev_ops->vdev_op_leaf) 4913 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4914 4915 pvd = vd->vdev_parent; 4916 4917 /* 4918 * If the parent/child relationship is not as expected, don't do it. 4919 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4920 * vdev that's replacing B with C. The user's intent in replacing 4921 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4922 * the replace by detaching C, the expected behavior is to end up 4923 * M(A,B). But suppose that right after deciding to detach C, 4924 * the replacement of B completes. We would have M(A,C), and then 4925 * ask to detach C, which would leave us with just A -- not what 4926 * the user wanted. To prevent this, we make sure that the 4927 * parent/child relationship hasn't changed -- in this example, 4928 * that C's parent is still the replacing vdev R. 4929 */ 4930 if (pvd->vdev_guid != pguid && pguid != 0) 4931 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4932 4933 /* 4934 * Only 'replacing' or 'spare' vdevs can be replaced. 4935 */ 4936 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4937 pvd->vdev_ops != &vdev_spare_ops) 4938 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4939 4940 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4941 spa_version(spa) >= SPA_VERSION_SPARES); 4942 4943 /* 4944 * Only mirror, replacing, and spare vdevs support detach. 4945 */ 4946 if (pvd->vdev_ops != &vdev_replacing_ops && 4947 pvd->vdev_ops != &vdev_mirror_ops && 4948 pvd->vdev_ops != &vdev_spare_ops) 4949 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4950 4951 /* 4952 * If this device has the only valid copy of some data, 4953 * we cannot safely detach it. 4954 */ 4955 if (vdev_dtl_required(vd)) 4956 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4957 4958 ASSERT(pvd->vdev_children >= 2); 4959 4960 /* 4961 * If we are detaching the second disk from a replacing vdev, then 4962 * check to see if we changed the original vdev's path to have "/old" 4963 * at the end in spa_vdev_attach(). If so, undo that change now. 4964 */ 4965 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4966 vd->vdev_path != NULL) { 4967 size_t len = strlen(vd->vdev_path); 4968 4969 for (int c = 0; c < pvd->vdev_children; c++) { 4970 cvd = pvd->vdev_child[c]; 4971 4972 if (cvd == vd || cvd->vdev_path == NULL) 4973 continue; 4974 4975 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4976 strcmp(cvd->vdev_path + len, "/old") == 0) { 4977 spa_strfree(cvd->vdev_path); 4978 cvd->vdev_path = spa_strdup(vd->vdev_path); 4979 break; 4980 } 4981 } 4982 } 4983 4984 /* 4985 * If we are detaching the original disk from a spare, then it implies 4986 * that the spare should become a real disk, and be removed from the 4987 * active spare list for the pool. 4988 */ 4989 if (pvd->vdev_ops == &vdev_spare_ops && 4990 vd->vdev_id == 0 && 4991 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4992 unspare = B_TRUE; 4993 4994 /* 4995 * Erase the disk labels so the disk can be used for other things. 4996 * This must be done after all other error cases are handled, 4997 * but before we disembowel vd (so we can still do I/O to it). 4998 * But if we can't do it, don't treat the error as fatal -- 4999 * it may be that the unwritability of the disk is the reason 5000 * it's being detached! 5001 */ 5002 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5003 5004 /* 5005 * Remove vd from its parent and compact the parent's children. 5006 */ 5007 vdev_remove_child(pvd, vd); 5008 vdev_compact_children(pvd); 5009 5010 /* 5011 * Remember one of the remaining children so we can get tvd below. 5012 */ 5013 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5014 5015 /* 5016 * If we need to remove the remaining child from the list of hot spares, 5017 * do it now, marking the vdev as no longer a spare in the process. 5018 * We must do this before vdev_remove_parent(), because that can 5019 * change the GUID if it creates a new toplevel GUID. For a similar 5020 * reason, we must remove the spare now, in the same txg as the detach; 5021 * otherwise someone could attach a new sibling, change the GUID, and 5022 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5023 */ 5024 if (unspare) { 5025 ASSERT(cvd->vdev_isspare); 5026 spa_spare_remove(cvd); 5027 unspare_guid = cvd->vdev_guid; 5028 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5029 cvd->vdev_unspare = B_TRUE; 5030 } 5031 5032 /* 5033 * If the parent mirror/replacing vdev only has one child, 5034 * the parent is no longer needed. Remove it from the tree. 5035 */ 5036 if (pvd->vdev_children == 1) { 5037 if (pvd->vdev_ops == &vdev_spare_ops) 5038 cvd->vdev_unspare = B_FALSE; 5039 vdev_remove_parent(cvd); 5040 } 5041 5042 5043 /* 5044 * We don't set tvd until now because the parent we just removed 5045 * may have been the previous top-level vdev. 5046 */ 5047 tvd = cvd->vdev_top; 5048 ASSERT(tvd->vdev_parent == rvd); 5049 5050 /* 5051 * Reevaluate the parent vdev state. 5052 */ 5053 vdev_propagate_state(cvd); 5054 5055 /* 5056 * If the 'autoexpand' property is set on the pool then automatically 5057 * try to expand the size of the pool. For example if the device we 5058 * just detached was smaller than the others, it may be possible to 5059 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5060 * first so that we can obtain the updated sizes of the leaf vdevs. 5061 */ 5062 if (spa->spa_autoexpand) { 5063 vdev_reopen(tvd); 5064 vdev_expand(tvd, txg); 5065 } 5066 5067 vdev_config_dirty(tvd); 5068 5069 /* 5070 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5071 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5072 * But first make sure we're not on any *other* txg's DTL list, to 5073 * prevent vd from being accessed after it's freed. 5074 */ 5075 vdpath = spa_strdup(vd->vdev_path); 5076 for (int t = 0; t < TXG_SIZE; t++) 5077 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5078 vd->vdev_detached = B_TRUE; 5079 vdev_dirty(tvd, VDD_DTL, vd, txg); 5080 5081 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5082 5083 /* hang on to the spa before we release the lock */ 5084 spa_open_ref(spa, FTAG); 5085 5086 error = spa_vdev_exit(spa, vd, txg, 0); 5087 5088 spa_history_log_internal(spa, "detach", NULL, 5089 "vdev=%s", vdpath); 5090 spa_strfree(vdpath); 5091 5092 /* 5093 * If this was the removal of the original device in a hot spare vdev, 5094 * then we want to go through and remove the device from the hot spare 5095 * list of every other pool. 5096 */ 5097 if (unspare) { 5098 spa_t *altspa = NULL; 5099 5100 mutex_enter(&spa_namespace_lock); 5101 while ((altspa = spa_next(altspa)) != NULL) { 5102 if (altspa->spa_state != POOL_STATE_ACTIVE || 5103 altspa == spa) 5104 continue; 5105 5106 spa_open_ref(altspa, FTAG); 5107 mutex_exit(&spa_namespace_lock); 5108 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5109 mutex_enter(&spa_namespace_lock); 5110 spa_close(altspa, FTAG); 5111 } 5112 mutex_exit(&spa_namespace_lock); 5113 5114 /* search the rest of the vdevs for spares to remove */ 5115 spa_vdev_resilver_done(spa); 5116 } 5117 5118 /* all done with the spa; OK to release */ 5119 mutex_enter(&spa_namespace_lock); 5120 spa_close(spa, FTAG); 5121 mutex_exit(&spa_namespace_lock); 5122 5123 return (error); 5124} 5125 5126/* 5127 * Split a set of devices from their mirrors, and create a new pool from them. 5128 */ 5129int 5130spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5131 nvlist_t *props, boolean_t exp) 5132{ 5133 int error = 0; 5134 uint64_t txg, *glist; 5135 spa_t *newspa; 5136 uint_t c, children, lastlog; 5137 nvlist_t **child, *nvl, *tmp; 5138 dmu_tx_t *tx; 5139 char *altroot = NULL; 5140 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5141 boolean_t activate_slog; 5142 5143 ASSERT(spa_writeable(spa)); 5144 5145 txg = spa_vdev_enter(spa); 5146 5147 /* clear the log and flush everything up to now */ 5148 activate_slog = spa_passivate_log(spa); 5149 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5150 error = spa_offline_log(spa); 5151 txg = spa_vdev_config_enter(spa); 5152 5153 if (activate_slog) 5154 spa_activate_log(spa); 5155 5156 if (error != 0) 5157 return (spa_vdev_exit(spa, NULL, txg, error)); 5158 5159 /* check new spa name before going any further */ 5160 if (spa_lookup(newname) != NULL) 5161 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5162 5163 /* 5164 * scan through all the children to ensure they're all mirrors 5165 */ 5166 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5167 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5168 &children) != 0) 5169 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5170 5171 /* first, check to ensure we've got the right child count */ 5172 rvd = spa->spa_root_vdev; 5173 lastlog = 0; 5174 for (c = 0; c < rvd->vdev_children; c++) { 5175 vdev_t *vd = rvd->vdev_child[c]; 5176 5177 /* don't count the holes & logs as children */ 5178 if (vd->vdev_islog || vd->vdev_ishole) { 5179 if (lastlog == 0) 5180 lastlog = c; 5181 continue; 5182 } 5183 5184 lastlog = 0; 5185 } 5186 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5187 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5188 5189 /* next, ensure no spare or cache devices are part of the split */ 5190 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5191 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5192 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5193 5194 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5195 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5196 5197 /* then, loop over each vdev and validate it */ 5198 for (c = 0; c < children; c++) { 5199 uint64_t is_hole = 0; 5200 5201 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5202 &is_hole); 5203 5204 if (is_hole != 0) { 5205 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5206 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5207 continue; 5208 } else { 5209 error = SET_ERROR(EINVAL); 5210 break; 5211 } 5212 } 5213 5214 /* which disk is going to be split? */ 5215 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5216 &glist[c]) != 0) { 5217 error = SET_ERROR(EINVAL); 5218 break; 5219 } 5220 5221 /* look it up in the spa */ 5222 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5223 if (vml[c] == NULL) { 5224 error = SET_ERROR(ENODEV); 5225 break; 5226 } 5227 5228 /* make sure there's nothing stopping the split */ 5229 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5230 vml[c]->vdev_islog || 5231 vml[c]->vdev_ishole || 5232 vml[c]->vdev_isspare || 5233 vml[c]->vdev_isl2cache || 5234 !vdev_writeable(vml[c]) || 5235 vml[c]->vdev_children != 0 || 5236 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5237 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5238 error = SET_ERROR(EINVAL); 5239 break; 5240 } 5241 5242 if (vdev_dtl_required(vml[c])) { 5243 error = SET_ERROR(EBUSY); 5244 break; 5245 } 5246 5247 /* we need certain info from the top level */ 5248 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5249 vml[c]->vdev_top->vdev_ms_array) == 0); 5250 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5251 vml[c]->vdev_top->vdev_ms_shift) == 0); 5252 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5253 vml[c]->vdev_top->vdev_asize) == 0); 5254 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5255 vml[c]->vdev_top->vdev_ashift) == 0); 5256 } 5257 5258 if (error != 0) { 5259 kmem_free(vml, children * sizeof (vdev_t *)); 5260 kmem_free(glist, children * sizeof (uint64_t)); 5261 return (spa_vdev_exit(spa, NULL, txg, error)); 5262 } 5263 5264 /* stop writers from using the disks */ 5265 for (c = 0; c < children; c++) { 5266 if (vml[c] != NULL) 5267 vml[c]->vdev_offline = B_TRUE; 5268 } 5269 vdev_reopen(spa->spa_root_vdev); 5270 5271 /* 5272 * Temporarily record the splitting vdevs in the spa config. This 5273 * will disappear once the config is regenerated. 5274 */ 5275 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5276 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5277 glist, children) == 0); 5278 kmem_free(glist, children * sizeof (uint64_t)); 5279 5280 mutex_enter(&spa->spa_props_lock); 5281 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5282 nvl) == 0); 5283 mutex_exit(&spa->spa_props_lock); 5284 spa->spa_config_splitting = nvl; 5285 vdev_config_dirty(spa->spa_root_vdev); 5286 5287 /* configure and create the new pool */ 5288 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5289 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5290 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5291 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5292 spa_version(spa)) == 0); 5293 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5294 spa->spa_config_txg) == 0); 5295 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5296 spa_generate_guid(NULL)) == 0); 5297 (void) nvlist_lookup_string(props, 5298 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5299 5300 /* add the new pool to the namespace */ 5301 newspa = spa_add(newname, config, altroot); 5302 newspa->spa_config_txg = spa->spa_config_txg; 5303 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5304 5305 /* release the spa config lock, retaining the namespace lock */ 5306 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5307 5308 if (zio_injection_enabled) 5309 zio_handle_panic_injection(spa, FTAG, 1); 5310 5311 spa_activate(newspa, spa_mode_global); 5312 spa_async_suspend(newspa); 5313 5314#ifndef sun 5315 /* mark that we are creating new spa by splitting */ 5316 newspa->spa_splitting_newspa = B_TRUE; 5317#endif 5318 /* create the new pool from the disks of the original pool */ 5319 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5320#ifndef sun 5321 newspa->spa_splitting_newspa = B_FALSE; 5322#endif 5323 if (error) 5324 goto out; 5325 5326 /* if that worked, generate a real config for the new pool */ 5327 if (newspa->spa_root_vdev != NULL) { 5328 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5329 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5330 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5331 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5332 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5333 B_TRUE)); 5334 } 5335 5336 /* set the props */ 5337 if (props != NULL) { 5338 spa_configfile_set(newspa, props, B_FALSE); 5339 error = spa_prop_set(newspa, props); 5340 if (error) 5341 goto out; 5342 } 5343 5344 /* flush everything */ 5345 txg = spa_vdev_config_enter(newspa); 5346 vdev_config_dirty(newspa->spa_root_vdev); 5347 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5348 5349 if (zio_injection_enabled) 5350 zio_handle_panic_injection(spa, FTAG, 2); 5351 5352 spa_async_resume(newspa); 5353 5354 /* finally, update the original pool's config */ 5355 txg = spa_vdev_config_enter(spa); 5356 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5357 error = dmu_tx_assign(tx, TXG_WAIT); 5358 if (error != 0) 5359 dmu_tx_abort(tx); 5360 for (c = 0; c < children; c++) { 5361 if (vml[c] != NULL) { 5362 vdev_split(vml[c]); 5363 if (error == 0) 5364 spa_history_log_internal(spa, "detach", tx, 5365 "vdev=%s", vml[c]->vdev_path); 5366 vdev_free(vml[c]); 5367 } 5368 } 5369 vdev_config_dirty(spa->spa_root_vdev); 5370 spa->spa_config_splitting = NULL; 5371 nvlist_free(nvl); 5372 if (error == 0) 5373 dmu_tx_commit(tx); 5374 (void) spa_vdev_exit(spa, NULL, txg, 0); 5375 5376 if (zio_injection_enabled) 5377 zio_handle_panic_injection(spa, FTAG, 3); 5378 5379 /* split is complete; log a history record */ 5380 spa_history_log_internal(newspa, "split", NULL, 5381 "from pool %s", spa_name(spa)); 5382 5383 kmem_free(vml, children * sizeof (vdev_t *)); 5384 5385 /* if we're not going to mount the filesystems in userland, export */ 5386 if (exp) 5387 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5388 B_FALSE, B_FALSE); 5389 5390 return (error); 5391 5392out: 5393 spa_unload(newspa); 5394 spa_deactivate(newspa); 5395 spa_remove(newspa); 5396 5397 txg = spa_vdev_config_enter(spa); 5398 5399 /* re-online all offlined disks */ 5400 for (c = 0; c < children; c++) { 5401 if (vml[c] != NULL) 5402 vml[c]->vdev_offline = B_FALSE; 5403 } 5404 vdev_reopen(spa->spa_root_vdev); 5405 5406 nvlist_free(spa->spa_config_splitting); 5407 spa->spa_config_splitting = NULL; 5408 (void) spa_vdev_exit(spa, NULL, txg, error); 5409 5410 kmem_free(vml, children * sizeof (vdev_t *)); 5411 return (error); 5412} 5413 5414static nvlist_t * 5415spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5416{ 5417 for (int i = 0; i < count; i++) { 5418 uint64_t guid; 5419 5420 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5421 &guid) == 0); 5422 5423 if (guid == target_guid) 5424 return (nvpp[i]); 5425 } 5426 5427 return (NULL); 5428} 5429 5430static void 5431spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5432 nvlist_t *dev_to_remove) 5433{ 5434 nvlist_t **newdev = NULL; 5435 5436 if (count > 1) 5437 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5438 5439 for (int i = 0, j = 0; i < count; i++) { 5440 if (dev[i] == dev_to_remove) 5441 continue; 5442 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5443 } 5444 5445 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5446 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5447 5448 for (int i = 0; i < count - 1; i++) 5449 nvlist_free(newdev[i]); 5450 5451 if (count > 1) 5452 kmem_free(newdev, (count - 1) * sizeof (void *)); 5453} 5454 5455/* 5456 * Evacuate the device. 5457 */ 5458static int 5459spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5460{ 5461 uint64_t txg; 5462 int error = 0; 5463 5464 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5465 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5466 ASSERT(vd == vd->vdev_top); 5467 5468 /* 5469 * Evacuate the device. We don't hold the config lock as writer 5470 * since we need to do I/O but we do keep the 5471 * spa_namespace_lock held. Once this completes the device 5472 * should no longer have any blocks allocated on it. 5473 */ 5474 if (vd->vdev_islog) { 5475 if (vd->vdev_stat.vs_alloc != 0) 5476 error = spa_offline_log(spa); 5477 } else { 5478 error = SET_ERROR(ENOTSUP); 5479 } 5480 5481 if (error) 5482 return (error); 5483 5484 /* 5485 * The evacuation succeeded. Remove any remaining MOS metadata 5486 * associated with this vdev, and wait for these changes to sync. 5487 */ 5488 ASSERT0(vd->vdev_stat.vs_alloc); 5489 txg = spa_vdev_config_enter(spa); 5490 vd->vdev_removing = B_TRUE; 5491 vdev_dirty_leaves(vd, VDD_DTL, txg); 5492 vdev_config_dirty(vd); 5493 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5494 5495 return (0); 5496} 5497 5498/* 5499 * Complete the removal by cleaning up the namespace. 5500 */ 5501static void 5502spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5503{ 5504 vdev_t *rvd = spa->spa_root_vdev; 5505 uint64_t id = vd->vdev_id; 5506 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5507 5508 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5509 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5510 ASSERT(vd == vd->vdev_top); 5511 5512 /* 5513 * Only remove any devices which are empty. 5514 */ 5515 if (vd->vdev_stat.vs_alloc != 0) 5516 return; 5517 5518 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5519 5520 if (list_link_active(&vd->vdev_state_dirty_node)) 5521 vdev_state_clean(vd); 5522 if (list_link_active(&vd->vdev_config_dirty_node)) 5523 vdev_config_clean(vd); 5524 5525 vdev_free(vd); 5526 5527 if (last_vdev) { 5528 vdev_compact_children(rvd); 5529 } else { 5530 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5531 vdev_add_child(rvd, vd); 5532 } 5533 vdev_config_dirty(rvd); 5534 5535 /* 5536 * Reassess the health of our root vdev. 5537 */ 5538 vdev_reopen(rvd); 5539} 5540 5541/* 5542 * Remove a device from the pool - 5543 * 5544 * Removing a device from the vdev namespace requires several steps 5545 * and can take a significant amount of time. As a result we use 5546 * the spa_vdev_config_[enter/exit] functions which allow us to 5547 * grab and release the spa_config_lock while still holding the namespace 5548 * lock. During each step the configuration is synced out. 5549 * 5550 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5551 * devices. 5552 */ 5553int 5554spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5555{ 5556 vdev_t *vd; 5557 metaslab_group_t *mg; 5558 nvlist_t **spares, **l2cache, *nv; 5559 uint64_t txg = 0; 5560 uint_t nspares, nl2cache; 5561 int error = 0; 5562 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5563 5564 ASSERT(spa_writeable(spa)); 5565 5566 if (!locked) 5567 txg = spa_vdev_enter(spa); 5568 5569 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5570 5571 if (spa->spa_spares.sav_vdevs != NULL && 5572 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5573 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5574 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5575 /* 5576 * Only remove the hot spare if it's not currently in use 5577 * in this pool. 5578 */ 5579 if (vd == NULL || unspare) { 5580 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5581 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5582 spa_load_spares(spa); 5583 spa->spa_spares.sav_sync = B_TRUE; 5584 } else { 5585 error = SET_ERROR(EBUSY); 5586 } 5587 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5588 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5589 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5590 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5591 /* 5592 * Cache devices can always be removed. 5593 */ 5594 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5595 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5596 spa_load_l2cache(spa); 5597 spa->spa_l2cache.sav_sync = B_TRUE; 5598 } else if (vd != NULL && vd->vdev_islog) { 5599 ASSERT(!locked); 5600 ASSERT(vd == vd->vdev_top); 5601 5602 mg = vd->vdev_mg; 5603 5604 /* 5605 * Stop allocating from this vdev. 5606 */ 5607 metaslab_group_passivate(mg); 5608 5609 /* 5610 * Wait for the youngest allocations and frees to sync, 5611 * and then wait for the deferral of those frees to finish. 5612 */ 5613 spa_vdev_config_exit(spa, NULL, 5614 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5615 5616 /* 5617 * Attempt to evacuate the vdev. 5618 */ 5619 error = spa_vdev_remove_evacuate(spa, vd); 5620 5621 txg = spa_vdev_config_enter(spa); 5622 5623 /* 5624 * If we couldn't evacuate the vdev, unwind. 5625 */ 5626 if (error) { 5627 metaslab_group_activate(mg); 5628 return (spa_vdev_exit(spa, NULL, txg, error)); 5629 } 5630 5631 /* 5632 * Clean up the vdev namespace. 5633 */ 5634 spa_vdev_remove_from_namespace(spa, vd); 5635 5636 } else if (vd != NULL) { 5637 /* 5638 * Normal vdevs cannot be removed (yet). 5639 */ 5640 error = SET_ERROR(ENOTSUP); 5641 } else { 5642 /* 5643 * There is no vdev of any kind with the specified guid. 5644 */ 5645 error = SET_ERROR(ENOENT); 5646 } 5647 5648 if (!locked) 5649 return (spa_vdev_exit(spa, NULL, txg, error)); 5650 5651 return (error); 5652} 5653 5654/* 5655 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5656 * currently spared, so we can detach it. 5657 */ 5658static vdev_t * 5659spa_vdev_resilver_done_hunt(vdev_t *vd) 5660{ 5661 vdev_t *newvd, *oldvd; 5662 5663 for (int c = 0; c < vd->vdev_children; c++) { 5664 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5665 if (oldvd != NULL) 5666 return (oldvd); 5667 } 5668 5669 /* 5670 * Check for a completed replacement. We always consider the first 5671 * vdev in the list to be the oldest vdev, and the last one to be 5672 * the newest (see spa_vdev_attach() for how that works). In 5673 * the case where the newest vdev is faulted, we will not automatically 5674 * remove it after a resilver completes. This is OK as it will require 5675 * user intervention to determine which disk the admin wishes to keep. 5676 */ 5677 if (vd->vdev_ops == &vdev_replacing_ops) { 5678 ASSERT(vd->vdev_children > 1); 5679 5680 newvd = vd->vdev_child[vd->vdev_children - 1]; 5681 oldvd = vd->vdev_child[0]; 5682 5683 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5684 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5685 !vdev_dtl_required(oldvd)) 5686 return (oldvd); 5687 } 5688 5689 /* 5690 * Check for a completed resilver with the 'unspare' flag set. 5691 */ 5692 if (vd->vdev_ops == &vdev_spare_ops) { 5693 vdev_t *first = vd->vdev_child[0]; 5694 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5695 5696 if (last->vdev_unspare) { 5697 oldvd = first; 5698 newvd = last; 5699 } else if (first->vdev_unspare) { 5700 oldvd = last; 5701 newvd = first; 5702 } else { 5703 oldvd = NULL; 5704 } 5705 5706 if (oldvd != NULL && 5707 vdev_dtl_empty(newvd, DTL_MISSING) && 5708 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5709 !vdev_dtl_required(oldvd)) 5710 return (oldvd); 5711 5712 /* 5713 * If there are more than two spares attached to a disk, 5714 * and those spares are not required, then we want to 5715 * attempt to free them up now so that they can be used 5716 * by other pools. Once we're back down to a single 5717 * disk+spare, we stop removing them. 5718 */ 5719 if (vd->vdev_children > 2) { 5720 newvd = vd->vdev_child[1]; 5721 5722 if (newvd->vdev_isspare && last->vdev_isspare && 5723 vdev_dtl_empty(last, DTL_MISSING) && 5724 vdev_dtl_empty(last, DTL_OUTAGE) && 5725 !vdev_dtl_required(newvd)) 5726 return (newvd); 5727 } 5728 } 5729 5730 return (NULL); 5731} 5732 5733static void 5734spa_vdev_resilver_done(spa_t *spa) 5735{ 5736 vdev_t *vd, *pvd, *ppvd; 5737 uint64_t guid, sguid, pguid, ppguid; 5738 5739 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5740 5741 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5742 pvd = vd->vdev_parent; 5743 ppvd = pvd->vdev_parent; 5744 guid = vd->vdev_guid; 5745 pguid = pvd->vdev_guid; 5746 ppguid = ppvd->vdev_guid; 5747 sguid = 0; 5748 /* 5749 * If we have just finished replacing a hot spared device, then 5750 * we need to detach the parent's first child (the original hot 5751 * spare) as well. 5752 */ 5753 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5754 ppvd->vdev_children == 2) { 5755 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5756 sguid = ppvd->vdev_child[1]->vdev_guid; 5757 } 5758 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5759 5760 spa_config_exit(spa, SCL_ALL, FTAG); 5761 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5762 return; 5763 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5764 return; 5765 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5766 } 5767 5768 spa_config_exit(spa, SCL_ALL, FTAG); 5769} 5770 5771/* 5772 * Update the stored path or FRU for this vdev. 5773 */ 5774int 5775spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5776 boolean_t ispath) 5777{ 5778 vdev_t *vd; 5779 boolean_t sync = B_FALSE; 5780 5781 ASSERT(spa_writeable(spa)); 5782 5783 spa_vdev_state_enter(spa, SCL_ALL); 5784 5785 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5786 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5787 5788 if (!vd->vdev_ops->vdev_op_leaf) 5789 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5790 5791 if (ispath) { 5792 if (strcmp(value, vd->vdev_path) != 0) { 5793 spa_strfree(vd->vdev_path); 5794 vd->vdev_path = spa_strdup(value); 5795 sync = B_TRUE; 5796 } 5797 } else { 5798 if (vd->vdev_fru == NULL) { 5799 vd->vdev_fru = spa_strdup(value); 5800 sync = B_TRUE; 5801 } else if (strcmp(value, vd->vdev_fru) != 0) { 5802 spa_strfree(vd->vdev_fru); 5803 vd->vdev_fru = spa_strdup(value); 5804 sync = B_TRUE; 5805 } 5806 } 5807 5808 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5809} 5810 5811int 5812spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5813{ 5814 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5815} 5816 5817int 5818spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5819{ 5820 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5821} 5822 5823/* 5824 * ========================================================================== 5825 * SPA Scanning 5826 * ========================================================================== 5827 */ 5828 5829int 5830spa_scan_stop(spa_t *spa) 5831{ 5832 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5833 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5834 return (SET_ERROR(EBUSY)); 5835 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5836} 5837 5838int 5839spa_scan(spa_t *spa, pool_scan_func_t func) 5840{ 5841 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5842 5843 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5844 return (SET_ERROR(ENOTSUP)); 5845 5846 /* 5847 * If a resilver was requested, but there is no DTL on a 5848 * writeable leaf device, we have nothing to do. 5849 */ 5850 if (func == POOL_SCAN_RESILVER && 5851 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5852 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5853 return (0); 5854 } 5855 5856 return (dsl_scan(spa->spa_dsl_pool, func)); 5857} 5858 5859/* 5860 * ========================================================================== 5861 * SPA async task processing 5862 * ========================================================================== 5863 */ 5864 5865static void 5866spa_async_remove(spa_t *spa, vdev_t *vd) 5867{ 5868 if (vd->vdev_remove_wanted) { 5869 vd->vdev_remove_wanted = B_FALSE; 5870 vd->vdev_delayed_close = B_FALSE; 5871 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5872 5873 /* 5874 * We want to clear the stats, but we don't want to do a full 5875 * vdev_clear() as that will cause us to throw away 5876 * degraded/faulted state as well as attempt to reopen the 5877 * device, all of which is a waste. 5878 */ 5879 vd->vdev_stat.vs_read_errors = 0; 5880 vd->vdev_stat.vs_write_errors = 0; 5881 vd->vdev_stat.vs_checksum_errors = 0; 5882 5883 vdev_state_dirty(vd->vdev_top); 5884 } 5885 5886 for (int c = 0; c < vd->vdev_children; c++) 5887 spa_async_remove(spa, vd->vdev_child[c]); 5888} 5889 5890static void 5891spa_async_probe(spa_t *spa, vdev_t *vd) 5892{ 5893 if (vd->vdev_probe_wanted) { 5894 vd->vdev_probe_wanted = B_FALSE; 5895 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5896 } 5897 5898 for (int c = 0; c < vd->vdev_children; c++) 5899 spa_async_probe(spa, vd->vdev_child[c]); 5900} 5901 5902static void 5903spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5904{ 5905 sysevent_id_t eid; 5906 nvlist_t *attr; 5907 char *physpath; 5908 5909 if (!spa->spa_autoexpand) 5910 return; 5911 5912 for (int c = 0; c < vd->vdev_children; c++) { 5913 vdev_t *cvd = vd->vdev_child[c]; 5914 spa_async_autoexpand(spa, cvd); 5915 } 5916 5917 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5918 return; 5919 5920 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5921 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5922 5923 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5924 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5925 5926 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5927 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5928 5929 nvlist_free(attr); 5930 kmem_free(physpath, MAXPATHLEN); 5931} 5932 5933static void 5934spa_async_thread(void *arg) 5935{ 5936 spa_t *spa = arg; 5937 int tasks; 5938 5939 ASSERT(spa->spa_sync_on); 5940 5941 mutex_enter(&spa->spa_async_lock); 5942 tasks = spa->spa_async_tasks; 5943 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5944 mutex_exit(&spa->spa_async_lock); 5945 5946 /* 5947 * See if the config needs to be updated. 5948 */ 5949 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5950 uint64_t old_space, new_space; 5951 5952 mutex_enter(&spa_namespace_lock); 5953 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5954 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5955 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5956 mutex_exit(&spa_namespace_lock); 5957 5958 /* 5959 * If the pool grew as a result of the config update, 5960 * then log an internal history event. 5961 */ 5962 if (new_space != old_space) { 5963 spa_history_log_internal(spa, "vdev online", NULL, 5964 "pool '%s' size: %llu(+%llu)", 5965 spa_name(spa), new_space, new_space - old_space); 5966 } 5967 } 5968 5969 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5970 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5971 spa_async_autoexpand(spa, spa->spa_root_vdev); 5972 spa_config_exit(spa, SCL_CONFIG, FTAG); 5973 } 5974 5975 /* 5976 * See if any devices need to be probed. 5977 */ 5978 if (tasks & SPA_ASYNC_PROBE) { 5979 spa_vdev_state_enter(spa, SCL_NONE); 5980 spa_async_probe(spa, spa->spa_root_vdev); 5981 (void) spa_vdev_state_exit(spa, NULL, 0); 5982 } 5983 5984 /* 5985 * If any devices are done replacing, detach them. 5986 */ 5987 if (tasks & SPA_ASYNC_RESILVER_DONE) 5988 spa_vdev_resilver_done(spa); 5989 5990 /* 5991 * Kick off a resilver. 5992 */ 5993 if (tasks & SPA_ASYNC_RESILVER) 5994 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5995 5996 /* 5997 * Let the world know that we're done. 5998 */ 5999 mutex_enter(&spa->spa_async_lock); 6000 spa->spa_async_thread = NULL; 6001 cv_broadcast(&spa->spa_async_cv); 6002 mutex_exit(&spa->spa_async_lock); 6003 thread_exit(); 6004} 6005 6006static void 6007spa_async_thread_vd(void *arg) 6008{ 6009 spa_t *spa = arg; 6010 int tasks; 6011 6012 ASSERT(spa->spa_sync_on); 6013 6014 mutex_enter(&spa->spa_async_lock); 6015 tasks = spa->spa_async_tasks; 6016retry: 6017 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6018 mutex_exit(&spa->spa_async_lock); 6019 6020 /* 6021 * See if any devices need to be marked REMOVED. 6022 */ 6023 if (tasks & SPA_ASYNC_REMOVE) { 6024 spa_vdev_state_enter(spa, SCL_NONE); 6025 spa_async_remove(spa, spa->spa_root_vdev); 6026 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6027 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6028 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6029 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6030 (void) spa_vdev_state_exit(spa, NULL, 0); 6031 } 6032 6033 /* 6034 * Let the world know that we're done. 6035 */ 6036 mutex_enter(&spa->spa_async_lock); 6037 tasks = spa->spa_async_tasks; 6038 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6039 goto retry; 6040 spa->spa_async_thread_vd = NULL; 6041 cv_broadcast(&spa->spa_async_cv); 6042 mutex_exit(&spa->spa_async_lock); 6043 thread_exit(); 6044} 6045 6046void 6047spa_async_suspend(spa_t *spa) 6048{ 6049 mutex_enter(&spa->spa_async_lock); 6050 spa->spa_async_suspended++; 6051 while (spa->spa_async_thread != NULL && 6052 spa->spa_async_thread_vd != NULL) 6053 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6054 mutex_exit(&spa->spa_async_lock); 6055} 6056 6057void 6058spa_async_resume(spa_t *spa) 6059{ 6060 mutex_enter(&spa->spa_async_lock); 6061 ASSERT(spa->spa_async_suspended != 0); 6062 spa->spa_async_suspended--; 6063 mutex_exit(&spa->spa_async_lock); 6064} 6065 6066static boolean_t 6067spa_async_tasks_pending(spa_t *spa) 6068{ 6069 uint_t non_config_tasks; 6070 uint_t config_task; 6071 boolean_t config_task_suspended; 6072 6073 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6074 SPA_ASYNC_REMOVE); 6075 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6076 if (spa->spa_ccw_fail_time == 0) { 6077 config_task_suspended = B_FALSE; 6078 } else { 6079 config_task_suspended = 6080 (gethrtime() - spa->spa_ccw_fail_time) < 6081 (zfs_ccw_retry_interval * NANOSEC); 6082 } 6083 6084 return (non_config_tasks || (config_task && !config_task_suspended)); 6085} 6086 6087static void 6088spa_async_dispatch(spa_t *spa) 6089{ 6090 mutex_enter(&spa->spa_async_lock); 6091 if (spa_async_tasks_pending(spa) && 6092 !spa->spa_async_suspended && 6093 spa->spa_async_thread == NULL && 6094 rootdir != NULL) 6095 spa->spa_async_thread = thread_create(NULL, 0, 6096 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6097 mutex_exit(&spa->spa_async_lock); 6098} 6099 6100static void 6101spa_async_dispatch_vd(spa_t *spa) 6102{ 6103 mutex_enter(&spa->spa_async_lock); 6104 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6105 !spa->spa_async_suspended && 6106 spa->spa_async_thread_vd == NULL && 6107 rootdir != NULL) 6108 spa->spa_async_thread_vd = thread_create(NULL, 0, 6109 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6110 mutex_exit(&spa->spa_async_lock); 6111} 6112 6113void 6114spa_async_request(spa_t *spa, int task) 6115{ 6116 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6117 mutex_enter(&spa->spa_async_lock); 6118 spa->spa_async_tasks |= task; 6119 mutex_exit(&spa->spa_async_lock); 6120 spa_async_dispatch_vd(spa); 6121} 6122 6123/* 6124 * ========================================================================== 6125 * SPA syncing routines 6126 * ========================================================================== 6127 */ 6128 6129static int 6130bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6131{ 6132 bpobj_t *bpo = arg; 6133 bpobj_enqueue(bpo, bp, tx); 6134 return (0); 6135} 6136 6137static int 6138spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6139{ 6140 zio_t *zio = arg; 6141 6142 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6143 BP_GET_PSIZE(bp), zio->io_flags)); 6144 return (0); 6145} 6146 6147/* 6148 * Note: this simple function is not inlined to make it easier to dtrace the 6149 * amount of time spent syncing frees. 6150 */ 6151static void 6152spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6153{ 6154 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6155 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6156 VERIFY(zio_wait(zio) == 0); 6157} 6158 6159/* 6160 * Note: this simple function is not inlined to make it easier to dtrace the 6161 * amount of time spent syncing deferred frees. 6162 */ 6163static void 6164spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6165{ 6166 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6167 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6168 spa_free_sync_cb, zio, tx), ==, 0); 6169 VERIFY0(zio_wait(zio)); 6170} 6171 6172 6173static void 6174spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6175{ 6176 char *packed = NULL; 6177 size_t bufsize; 6178 size_t nvsize = 0; 6179 dmu_buf_t *db; 6180 6181 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6182 6183 /* 6184 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6185 * information. This avoids the dmu_buf_will_dirty() path and 6186 * saves us a pre-read to get data we don't actually care about. 6187 */ 6188 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6189 packed = kmem_alloc(bufsize, KM_SLEEP); 6190 6191 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6192 KM_SLEEP) == 0); 6193 bzero(packed + nvsize, bufsize - nvsize); 6194 6195 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6196 6197 kmem_free(packed, bufsize); 6198 6199 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6200 dmu_buf_will_dirty(db, tx); 6201 *(uint64_t *)db->db_data = nvsize; 6202 dmu_buf_rele(db, FTAG); 6203} 6204 6205static void 6206spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6207 const char *config, const char *entry) 6208{ 6209 nvlist_t *nvroot; 6210 nvlist_t **list; 6211 int i; 6212 6213 if (!sav->sav_sync) 6214 return; 6215 6216 /* 6217 * Update the MOS nvlist describing the list of available devices. 6218 * spa_validate_aux() will have already made sure this nvlist is 6219 * valid and the vdevs are labeled appropriately. 6220 */ 6221 if (sav->sav_object == 0) { 6222 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6223 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6224 sizeof (uint64_t), tx); 6225 VERIFY(zap_update(spa->spa_meta_objset, 6226 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6227 &sav->sav_object, tx) == 0); 6228 } 6229 6230 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6231 if (sav->sav_count == 0) { 6232 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6233 } else { 6234 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6235 for (i = 0; i < sav->sav_count; i++) 6236 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6237 B_FALSE, VDEV_CONFIG_L2CACHE); 6238 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6239 sav->sav_count) == 0); 6240 for (i = 0; i < sav->sav_count; i++) 6241 nvlist_free(list[i]); 6242 kmem_free(list, sav->sav_count * sizeof (void *)); 6243 } 6244 6245 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6246 nvlist_free(nvroot); 6247 6248 sav->sav_sync = B_FALSE; 6249} 6250 6251static void 6252spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6253{ 6254 nvlist_t *config; 6255 6256 if (list_is_empty(&spa->spa_config_dirty_list)) 6257 return; 6258 6259 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6260 6261 config = spa_config_generate(spa, spa->spa_root_vdev, 6262 dmu_tx_get_txg(tx), B_FALSE); 6263 6264 /* 6265 * If we're upgrading the spa version then make sure that 6266 * the config object gets updated with the correct version. 6267 */ 6268 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6269 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6270 spa->spa_uberblock.ub_version); 6271 6272 spa_config_exit(spa, SCL_STATE, FTAG); 6273 6274 if (spa->spa_config_syncing) 6275 nvlist_free(spa->spa_config_syncing); 6276 spa->spa_config_syncing = config; 6277 6278 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6279} 6280 6281static void 6282spa_sync_version(void *arg, dmu_tx_t *tx) 6283{ 6284 uint64_t *versionp = arg; 6285 uint64_t version = *versionp; 6286 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6287 6288 /* 6289 * Setting the version is special cased when first creating the pool. 6290 */ 6291 ASSERT(tx->tx_txg != TXG_INITIAL); 6292 6293 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6294 ASSERT(version >= spa_version(spa)); 6295 6296 spa->spa_uberblock.ub_version = version; 6297 vdev_config_dirty(spa->spa_root_vdev); 6298 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6299} 6300 6301/* 6302 * Set zpool properties. 6303 */ 6304static void 6305spa_sync_props(void *arg, dmu_tx_t *tx) 6306{ 6307 nvlist_t *nvp = arg; 6308 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6309 objset_t *mos = spa->spa_meta_objset; 6310 nvpair_t *elem = NULL; 6311 6312 mutex_enter(&spa->spa_props_lock); 6313 6314 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6315 uint64_t intval; 6316 char *strval, *fname; 6317 zpool_prop_t prop; 6318 const char *propname; 6319 zprop_type_t proptype; 6320 spa_feature_t fid; 6321 6322 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6323 case ZPROP_INVAL: 6324 /* 6325 * We checked this earlier in spa_prop_validate(). 6326 */ 6327 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6328 6329 fname = strchr(nvpair_name(elem), '@') + 1; 6330 VERIFY0(zfeature_lookup_name(fname, &fid)); 6331 6332 spa_feature_enable(spa, fid, tx); 6333 spa_history_log_internal(spa, "set", tx, 6334 "%s=enabled", nvpair_name(elem)); 6335 break; 6336 6337 case ZPOOL_PROP_VERSION: 6338 intval = fnvpair_value_uint64(elem); 6339 /* 6340 * The version is synced seperatly before other 6341 * properties and should be correct by now. 6342 */ 6343 ASSERT3U(spa_version(spa), >=, intval); 6344 break; 6345 6346 case ZPOOL_PROP_ALTROOT: 6347 /* 6348 * 'altroot' is a non-persistent property. It should 6349 * have been set temporarily at creation or import time. 6350 */ 6351 ASSERT(spa->spa_root != NULL); 6352 break; 6353 6354 case ZPOOL_PROP_READONLY: 6355 case ZPOOL_PROP_CACHEFILE: 6356 /* 6357 * 'readonly' and 'cachefile' are also non-persisitent 6358 * properties. 6359 */ 6360 break; 6361 case ZPOOL_PROP_COMMENT: 6362 strval = fnvpair_value_string(elem); 6363 if (spa->spa_comment != NULL) 6364 spa_strfree(spa->spa_comment); 6365 spa->spa_comment = spa_strdup(strval); 6366 /* 6367 * We need to dirty the configuration on all the vdevs 6368 * so that their labels get updated. It's unnecessary 6369 * to do this for pool creation since the vdev's 6370 * configuratoin has already been dirtied. 6371 */ 6372 if (tx->tx_txg != TXG_INITIAL) 6373 vdev_config_dirty(spa->spa_root_vdev); 6374 spa_history_log_internal(spa, "set", tx, 6375 "%s=%s", nvpair_name(elem), strval); 6376 break; 6377 default: 6378 /* 6379 * Set pool property values in the poolprops mos object. 6380 */ 6381 if (spa->spa_pool_props_object == 0) { 6382 spa->spa_pool_props_object = 6383 zap_create_link(mos, DMU_OT_POOL_PROPS, 6384 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6385 tx); 6386 } 6387 6388 /* normalize the property name */ 6389 propname = zpool_prop_to_name(prop); 6390 proptype = zpool_prop_get_type(prop); 6391 6392 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6393 ASSERT(proptype == PROP_TYPE_STRING); 6394 strval = fnvpair_value_string(elem); 6395 VERIFY0(zap_update(mos, 6396 spa->spa_pool_props_object, propname, 6397 1, strlen(strval) + 1, strval, tx)); 6398 spa_history_log_internal(spa, "set", tx, 6399 "%s=%s", nvpair_name(elem), strval); 6400 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6401 intval = fnvpair_value_uint64(elem); 6402 6403 if (proptype == PROP_TYPE_INDEX) { 6404 const char *unused; 6405 VERIFY0(zpool_prop_index_to_string( 6406 prop, intval, &unused)); 6407 } 6408 VERIFY0(zap_update(mos, 6409 spa->spa_pool_props_object, propname, 6410 8, 1, &intval, tx)); 6411 spa_history_log_internal(spa, "set", tx, 6412 "%s=%lld", nvpair_name(elem), intval); 6413 } else { 6414 ASSERT(0); /* not allowed */ 6415 } 6416 6417 switch (prop) { 6418 case ZPOOL_PROP_DELEGATION: 6419 spa->spa_delegation = intval; 6420 break; 6421 case ZPOOL_PROP_BOOTFS: 6422 spa->spa_bootfs = intval; 6423 break; 6424 case ZPOOL_PROP_FAILUREMODE: 6425 spa->spa_failmode = intval; 6426 break; 6427 case ZPOOL_PROP_AUTOEXPAND: 6428 spa->spa_autoexpand = intval; 6429 if (tx->tx_txg != TXG_INITIAL) 6430 spa_async_request(spa, 6431 SPA_ASYNC_AUTOEXPAND); 6432 break; 6433 case ZPOOL_PROP_DEDUPDITTO: 6434 spa->spa_dedup_ditto = intval; 6435 break; 6436 default: 6437 break; 6438 } 6439 } 6440 6441 } 6442 6443 mutex_exit(&spa->spa_props_lock); 6444} 6445 6446/* 6447 * Perform one-time upgrade on-disk changes. spa_version() does not 6448 * reflect the new version this txg, so there must be no changes this 6449 * txg to anything that the upgrade code depends on after it executes. 6450 * Therefore this must be called after dsl_pool_sync() does the sync 6451 * tasks. 6452 */ 6453static void 6454spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6455{ 6456 dsl_pool_t *dp = spa->spa_dsl_pool; 6457 6458 ASSERT(spa->spa_sync_pass == 1); 6459 6460 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6461 6462 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6463 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6464 dsl_pool_create_origin(dp, tx); 6465 6466 /* Keeping the origin open increases spa_minref */ 6467 spa->spa_minref += 3; 6468 } 6469 6470 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6471 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6472 dsl_pool_upgrade_clones(dp, tx); 6473 } 6474 6475 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6476 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6477 dsl_pool_upgrade_dir_clones(dp, tx); 6478 6479 /* Keeping the freedir open increases spa_minref */ 6480 spa->spa_minref += 3; 6481 } 6482 6483 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6484 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6485 spa_feature_create_zap_objects(spa, tx); 6486 } 6487 6488 /* 6489 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6490 * when possibility to use lz4 compression for metadata was added 6491 * Old pools that have this feature enabled must be upgraded to have 6492 * this feature active 6493 */ 6494 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6495 boolean_t lz4_en = spa_feature_is_enabled(spa, 6496 SPA_FEATURE_LZ4_COMPRESS); 6497 boolean_t lz4_ac = spa_feature_is_active(spa, 6498 SPA_FEATURE_LZ4_COMPRESS); 6499 6500 if (lz4_en && !lz4_ac) 6501 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6502 } 6503 rrw_exit(&dp->dp_config_rwlock, FTAG); 6504} 6505 6506/* 6507 * Sync the specified transaction group. New blocks may be dirtied as 6508 * part of the process, so we iterate until it converges. 6509 */ 6510void 6511spa_sync(spa_t *spa, uint64_t txg) 6512{ 6513 dsl_pool_t *dp = spa->spa_dsl_pool; 6514 objset_t *mos = spa->spa_meta_objset; 6515 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6516 vdev_t *rvd = spa->spa_root_vdev; 6517 vdev_t *vd; 6518 dmu_tx_t *tx; 6519 int error; 6520 6521 VERIFY(spa_writeable(spa)); 6522 6523 /* 6524 * Lock out configuration changes. 6525 */ 6526 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6527 6528 spa->spa_syncing_txg = txg; 6529 spa->spa_sync_pass = 0; 6530 6531 /* 6532 * If there are any pending vdev state changes, convert them 6533 * into config changes that go out with this transaction group. 6534 */ 6535 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6536 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6537 /* 6538 * We need the write lock here because, for aux vdevs, 6539 * calling vdev_config_dirty() modifies sav_config. 6540 * This is ugly and will become unnecessary when we 6541 * eliminate the aux vdev wart by integrating all vdevs 6542 * into the root vdev tree. 6543 */ 6544 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6545 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6546 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6547 vdev_state_clean(vd); 6548 vdev_config_dirty(vd); 6549 } 6550 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6551 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6552 } 6553 spa_config_exit(spa, SCL_STATE, FTAG); 6554 6555 tx = dmu_tx_create_assigned(dp, txg); 6556 6557 spa->spa_sync_starttime = gethrtime(); 6558#ifdef illumos 6559 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6560 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6561#else /* FreeBSD */ 6562#ifdef _KERNEL 6563 callout_reset(&spa->spa_deadman_cycid, 6564 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6565#endif 6566#endif 6567 6568 /* 6569 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6570 * set spa_deflate if we have no raid-z vdevs. 6571 */ 6572 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6573 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6574 int i; 6575 6576 for (i = 0; i < rvd->vdev_children; i++) { 6577 vd = rvd->vdev_child[i]; 6578 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6579 break; 6580 } 6581 if (i == rvd->vdev_children) { 6582 spa->spa_deflate = TRUE; 6583 VERIFY(0 == zap_add(spa->spa_meta_objset, 6584 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6585 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6586 } 6587 } 6588 6589 /* 6590 * If anything has changed in this txg, or if someone is waiting 6591 * for this txg to sync (eg, spa_vdev_remove()), push the 6592 * deferred frees from the previous txg. If not, leave them 6593 * alone so that we don't generate work on an otherwise idle 6594 * system. 6595 */ 6596 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6597 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6598 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6599 ((dsl_scan_active(dp->dp_scan) || 6600 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6601 spa_sync_deferred_frees(spa, tx); 6602 } 6603 6604 /* 6605 * Iterate to convergence. 6606 */ 6607 do { 6608 int pass = ++spa->spa_sync_pass; 6609 6610 spa_sync_config_object(spa, tx); 6611 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6612 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6613 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6614 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6615 spa_errlog_sync(spa, txg); 6616 dsl_pool_sync(dp, txg); 6617 6618 if (pass < zfs_sync_pass_deferred_free) { 6619 spa_sync_frees(spa, free_bpl, tx); 6620 } else { 6621 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6622 &spa->spa_deferred_bpobj, tx); 6623 } 6624 6625 ddt_sync(spa, txg); 6626 dsl_scan_sync(dp, tx); 6627 6628 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6629 vdev_sync(vd, txg); 6630 6631 if (pass == 1) 6632 spa_sync_upgrades(spa, tx); 6633 6634 } while (dmu_objset_is_dirty(mos, txg)); 6635 6636 /* 6637 * Rewrite the vdev configuration (which includes the uberblock) 6638 * to commit the transaction group. 6639 * 6640 * If there are no dirty vdevs, we sync the uberblock to a few 6641 * random top-level vdevs that are known to be visible in the 6642 * config cache (see spa_vdev_add() for a complete description). 6643 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6644 */ 6645 for (;;) { 6646 /* 6647 * We hold SCL_STATE to prevent vdev open/close/etc. 6648 * while we're attempting to write the vdev labels. 6649 */ 6650 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6651 6652 if (list_is_empty(&spa->spa_config_dirty_list)) { 6653 vdev_t *svd[SPA_DVAS_PER_BP]; 6654 int svdcount = 0; 6655 int children = rvd->vdev_children; 6656 int c0 = spa_get_random(children); 6657 6658 for (int c = 0; c < children; c++) { 6659 vd = rvd->vdev_child[(c0 + c) % children]; 6660 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6661 continue; 6662 svd[svdcount++] = vd; 6663 if (svdcount == SPA_DVAS_PER_BP) 6664 break; 6665 } 6666 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6667 if (error != 0) 6668 error = vdev_config_sync(svd, svdcount, txg, 6669 B_TRUE); 6670 } else { 6671 error = vdev_config_sync(rvd->vdev_child, 6672 rvd->vdev_children, txg, B_FALSE); 6673 if (error != 0) 6674 error = vdev_config_sync(rvd->vdev_child, 6675 rvd->vdev_children, txg, B_TRUE); 6676 } 6677 6678 if (error == 0) 6679 spa->spa_last_synced_guid = rvd->vdev_guid; 6680 6681 spa_config_exit(spa, SCL_STATE, FTAG); 6682 6683 if (error == 0) 6684 break; 6685 zio_suspend(spa, NULL); 6686 zio_resume_wait(spa); 6687 } 6688 dmu_tx_commit(tx); 6689 6690#ifdef illumos 6691 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6692#else /* FreeBSD */ 6693#ifdef _KERNEL 6694 callout_drain(&spa->spa_deadman_cycid); 6695#endif 6696#endif 6697 6698 /* 6699 * Clear the dirty config list. 6700 */ 6701 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6702 vdev_config_clean(vd); 6703 6704 /* 6705 * Now that the new config has synced transactionally, 6706 * let it become visible to the config cache. 6707 */ 6708 if (spa->spa_config_syncing != NULL) { 6709 spa_config_set(spa, spa->spa_config_syncing); 6710 spa->spa_config_txg = txg; 6711 spa->spa_config_syncing = NULL; 6712 } 6713 6714 spa->spa_ubsync = spa->spa_uberblock; 6715 6716 dsl_pool_sync_done(dp, txg); 6717 6718 /* 6719 * Update usable space statistics. 6720 */ 6721 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6722 vdev_sync_done(vd, txg); 6723 6724 spa_update_dspace(spa); 6725 6726 /* 6727 * It had better be the case that we didn't dirty anything 6728 * since vdev_config_sync(). 6729 */ 6730 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6731 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6732 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6733 6734 spa->spa_sync_pass = 0; 6735 6736 spa_config_exit(spa, SCL_CONFIG, FTAG); 6737 6738 spa_handle_ignored_writes(spa); 6739 6740 /* 6741 * If any async tasks have been requested, kick them off. 6742 */ 6743 spa_async_dispatch(spa); 6744 spa_async_dispatch_vd(spa); 6745} 6746 6747/* 6748 * Sync all pools. We don't want to hold the namespace lock across these 6749 * operations, so we take a reference on the spa_t and drop the lock during the 6750 * sync. 6751 */ 6752void 6753spa_sync_allpools(void) 6754{ 6755 spa_t *spa = NULL; 6756 mutex_enter(&spa_namespace_lock); 6757 while ((spa = spa_next(spa)) != NULL) { 6758 if (spa_state(spa) != POOL_STATE_ACTIVE || 6759 !spa_writeable(spa) || spa_suspended(spa)) 6760 continue; 6761 spa_open_ref(spa, FTAG); 6762 mutex_exit(&spa_namespace_lock); 6763 txg_wait_synced(spa_get_dsl(spa), 0); 6764 mutex_enter(&spa_namespace_lock); 6765 spa_close(spa, FTAG); 6766 } 6767 mutex_exit(&spa_namespace_lock); 6768} 6769 6770/* 6771 * ========================================================================== 6772 * Miscellaneous routines 6773 * ========================================================================== 6774 */ 6775 6776/* 6777 * Remove all pools in the system. 6778 */ 6779void 6780spa_evict_all(void) 6781{ 6782 spa_t *spa; 6783 6784 /* 6785 * Remove all cached state. All pools should be closed now, 6786 * so every spa in the AVL tree should be unreferenced. 6787 */ 6788 mutex_enter(&spa_namespace_lock); 6789 while ((spa = spa_next(NULL)) != NULL) { 6790 /* 6791 * Stop async tasks. The async thread may need to detach 6792 * a device that's been replaced, which requires grabbing 6793 * spa_namespace_lock, so we must drop it here. 6794 */ 6795 spa_open_ref(spa, FTAG); 6796 mutex_exit(&spa_namespace_lock); 6797 spa_async_suspend(spa); 6798 mutex_enter(&spa_namespace_lock); 6799 spa_close(spa, FTAG); 6800 6801 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6802 spa_unload(spa); 6803 spa_deactivate(spa); 6804 } 6805 spa_remove(spa); 6806 } 6807 mutex_exit(&spa_namespace_lock); 6808} 6809 6810vdev_t * 6811spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6812{ 6813 vdev_t *vd; 6814 int i; 6815 6816 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6817 return (vd); 6818 6819 if (aux) { 6820 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6821 vd = spa->spa_l2cache.sav_vdevs[i]; 6822 if (vd->vdev_guid == guid) 6823 return (vd); 6824 } 6825 6826 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6827 vd = spa->spa_spares.sav_vdevs[i]; 6828 if (vd->vdev_guid == guid) 6829 return (vd); 6830 } 6831 } 6832 6833 return (NULL); 6834} 6835 6836void 6837spa_upgrade(spa_t *spa, uint64_t version) 6838{ 6839 ASSERT(spa_writeable(spa)); 6840 6841 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6842 6843 /* 6844 * This should only be called for a non-faulted pool, and since a 6845 * future version would result in an unopenable pool, this shouldn't be 6846 * possible. 6847 */ 6848 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6849 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6850 6851 spa->spa_uberblock.ub_version = version; 6852 vdev_config_dirty(spa->spa_root_vdev); 6853 6854 spa_config_exit(spa, SCL_ALL, FTAG); 6855 6856 txg_wait_synced(spa_get_dsl(spa), 0); 6857} 6858 6859boolean_t 6860spa_has_spare(spa_t *spa, uint64_t guid) 6861{ 6862 int i; 6863 uint64_t spareguid; 6864 spa_aux_vdev_t *sav = &spa->spa_spares; 6865 6866 for (i = 0; i < sav->sav_count; i++) 6867 if (sav->sav_vdevs[i]->vdev_guid == guid) 6868 return (B_TRUE); 6869 6870 for (i = 0; i < sav->sav_npending; i++) { 6871 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6872 &spareguid) == 0 && spareguid == guid) 6873 return (B_TRUE); 6874 } 6875 6876 return (B_FALSE); 6877} 6878 6879/* 6880 * Check if a pool has an active shared spare device. 6881 * Note: reference count of an active spare is 2, as a spare and as a replace 6882 */ 6883static boolean_t 6884spa_has_active_shared_spare(spa_t *spa) 6885{ 6886 int i, refcnt; 6887 uint64_t pool; 6888 spa_aux_vdev_t *sav = &spa->spa_spares; 6889 6890 for (i = 0; i < sav->sav_count; i++) { 6891 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6892 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6893 refcnt > 2) 6894 return (B_TRUE); 6895 } 6896 6897 return (B_FALSE); 6898} 6899 6900/* 6901 * Post a sysevent corresponding to the given event. The 'name' must be one of 6902 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6903 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6904 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6905 * or zdb as real changes. 6906 */ 6907void 6908spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6909{ 6910#ifdef _KERNEL 6911 sysevent_t *ev; 6912 sysevent_attr_list_t *attr = NULL; 6913 sysevent_value_t value; 6914 sysevent_id_t eid; 6915 6916 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6917 SE_SLEEP); 6918 6919 value.value_type = SE_DATA_TYPE_STRING; 6920 value.value.sv_string = spa_name(spa); 6921 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6922 goto done; 6923 6924 value.value_type = SE_DATA_TYPE_UINT64; 6925 value.value.sv_uint64 = spa_guid(spa); 6926 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6927 goto done; 6928 6929 if (vd) { 6930 value.value_type = SE_DATA_TYPE_UINT64; 6931 value.value.sv_uint64 = vd->vdev_guid; 6932 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6933 SE_SLEEP) != 0) 6934 goto done; 6935 6936 if (vd->vdev_path) { 6937 value.value_type = SE_DATA_TYPE_STRING; 6938 value.value.sv_string = vd->vdev_path; 6939 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6940 &value, SE_SLEEP) != 0) 6941 goto done; 6942 } 6943 } 6944 6945 if (sysevent_attach_attributes(ev, attr) != 0) 6946 goto done; 6947 attr = NULL; 6948 6949 (void) log_sysevent(ev, SE_SLEEP, &eid); 6950 6951done: 6952 if (attr) 6953 sysevent_free_attr(attr); 6954 sysevent_free(ev); 6955#endif 6956} 6957