spa.c revision 333196
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) 2015, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 28 * Copyright 2013 Saso Kiselkov. All rights reserved. 29 * Copyright (c) 2014 Integros [integros.com] 30 */ 31 32/* 33 * SPA: Storage Pool Allocator 34 * 35 * This file contains all the routines used when modifying on-disk SPA state. 36 * This includes opening, importing, destroying, exporting a pool, and syncing a 37 * pool. 38 */ 39 40#include <sys/zfs_context.h> 41#include <sys/fm/fs/zfs.h> 42#include <sys/spa_impl.h> 43#include <sys/zio.h> 44#include <sys/zio_checksum.h> 45#include <sys/dmu.h> 46#include <sys/dmu_tx.h> 47#include <sys/zap.h> 48#include <sys/zil.h> 49#include <sys/ddt.h> 50#include <sys/vdev_impl.h> 51#include <sys/metaslab.h> 52#include <sys/metaslab_impl.h> 53#include <sys/uberblock_impl.h> 54#include <sys/txg.h> 55#include <sys/avl.h> 56#include <sys/dmu_traverse.h> 57#include <sys/dmu_objset.h> 58#include <sys/unique.h> 59#include <sys/dsl_pool.h> 60#include <sys/dsl_dataset.h> 61#include <sys/dsl_dir.h> 62#include <sys/dsl_prop.h> 63#include <sys/dsl_synctask.h> 64#include <sys/fs/zfs.h> 65#include <sys/arc.h> 66#include <sys/callb.h> 67#include <sys/spa_boot.h> 68#include <sys/zfs_ioctl.h> 69#include <sys/dsl_scan.h> 70#include <sys/dmu_send.h> 71#include <sys/dsl_destroy.h> 72#include <sys/dsl_userhold.h> 73#include <sys/zfeature.h> 74#include <sys/zvol.h> 75#include <sys/trim_map.h> 76 77#ifdef _KERNEL 78#include <sys/callb.h> 79#include <sys/cpupart.h> 80#include <sys/zone.h> 81#endif /* _KERNEL */ 82 83#include "zfs_prop.h" 84#include "zfs_comutil.h" 85 86/* Check hostid on import? */ 87static int check_hostid = 1; 88 89/* 90 * The interval, in seconds, at which failed configuration cache file writes 91 * should be retried. 92 */ 93static int zfs_ccw_retry_interval = 300; 94 95SYSCTL_DECL(_vfs_zfs); 96TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 97SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 98 "Check hostid on import?"); 99TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 100SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 101 &zfs_ccw_retry_interval, 0, 102 "Configuration cache file write, retry after failure, interval (seconds)"); 103 104typedef enum zti_modes { 105 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 106 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 107 ZTI_MODE_NULL, /* don't create a taskq */ 108 ZTI_NMODES 109} zti_modes_t; 110 111#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 112#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 113#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 114 115#define ZTI_N(n) ZTI_P(n, 1) 116#define ZTI_ONE ZTI_N(1) 117 118typedef struct zio_taskq_info { 119 zti_modes_t zti_mode; 120 uint_t zti_value; 121 uint_t zti_count; 122} zio_taskq_info_t; 123 124static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 125 "issue", "issue_high", "intr", "intr_high" 126}; 127 128/* 129 * This table defines the taskq settings for each ZFS I/O type. When 130 * initializing a pool, we use this table to create an appropriately sized 131 * taskq. Some operations are low volume and therefore have a small, static 132 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 133 * macros. Other operations process a large amount of data; the ZTI_BATCH 134 * macro causes us to create a taskq oriented for throughput. Some operations 135 * are so high frequency and short-lived that the taskq itself can become a a 136 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 137 * additional degree of parallelism specified by the number of threads per- 138 * taskq and the number of taskqs; when dispatching an event in this case, the 139 * particular taskq is chosen at random. 140 * 141 * The different taskq priorities are to handle the different contexts (issue 142 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 143 * need to be handled with minimum delay. 144 */ 145const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 146 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 148 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 149 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 150 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 151 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 152 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 153}; 154 155static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, const char *name); 156static void spa_event_post(sysevent_t *ev); 157static void spa_sync_version(void *arg, dmu_tx_t *tx); 158static void spa_sync_props(void *arg, dmu_tx_t *tx); 159static boolean_t spa_has_active_shared_spare(spa_t *spa); 160static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 161 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 162 char **ereport); 163static void spa_vdev_resilver_done(spa_t *spa); 164 165uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 166#ifdef PSRSET_BIND 167id_t zio_taskq_psrset_bind = PS_NONE; 168#endif 169#ifdef SYSDC 170boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 171uint_t zio_taskq_basedc = 80; /* base duty cycle */ 172#endif 173 174boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 175extern int zfs_sync_pass_deferred_free; 176 177/* 178 * This (illegal) pool name is used when temporarily importing a spa_t in order 179 * to get the vdev stats associated with the imported devices. 180 */ 181#define TRYIMPORT_NAME "$import" 182 183/* 184 * ========================================================================== 185 * SPA properties routines 186 * ========================================================================== 187 */ 188 189/* 190 * Add a (source=src, propname=propval) list to an nvlist. 191 */ 192static void 193spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 194 uint64_t intval, zprop_source_t src) 195{ 196 const char *propname = zpool_prop_to_name(prop); 197 nvlist_t *propval; 198 199 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 200 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 201 202 if (strval != NULL) 203 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 204 else 205 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 206 207 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 208 nvlist_free(propval); 209} 210 211/* 212 * Get property values from the spa configuration. 213 */ 214static void 215spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 216{ 217 vdev_t *rvd = spa->spa_root_vdev; 218 dsl_pool_t *pool = spa->spa_dsl_pool; 219 uint64_t size, alloc, cap, version; 220 zprop_source_t src = ZPROP_SRC_NONE; 221 spa_config_dirent_t *dp; 222 metaslab_class_t *mc = spa_normal_class(spa); 223 224 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 225 226 if (rvd != NULL) { 227 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 228 size = metaslab_class_get_space(spa_normal_class(spa)); 229 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 230 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 231 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 232 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 233 size - alloc, src); 234 235 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 236 metaslab_class_fragmentation(mc), src); 237 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 238 metaslab_class_expandable_space(mc), src); 239 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 240 (spa_mode(spa) == FREAD), src); 241 242 cap = (size == 0) ? 0 : (alloc * 100 / size); 243 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 244 245 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 246 ddt_get_pool_dedup_ratio(spa), src); 247 248 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 249 rvd->vdev_state, src); 250 251 version = spa_version(spa); 252 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 253 src = ZPROP_SRC_DEFAULT; 254 else 255 src = ZPROP_SRC_LOCAL; 256 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 257 } 258 259 if (pool != NULL) { 260 /* 261 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 262 * when opening pools before this version freedir will be NULL. 263 */ 264 if (pool->dp_free_dir != NULL) { 265 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 266 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 267 src); 268 } else { 269 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 270 NULL, 0, src); 271 } 272 273 if (pool->dp_leak_dir != NULL) { 274 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 275 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 276 src); 277 } else { 278 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 279 NULL, 0, src); 280 } 281 } 282 283 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 284 285 if (spa->spa_comment != NULL) { 286 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 287 0, ZPROP_SRC_LOCAL); 288 } 289 290 if (spa->spa_root != NULL) 291 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 292 0, ZPROP_SRC_LOCAL); 293 294 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 295 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 296 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 297 } else { 298 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 299 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 300 } 301 302 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 303 if (dp->scd_path == NULL) { 304 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 305 "none", 0, ZPROP_SRC_LOCAL); 306 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 307 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 308 dp->scd_path, 0, ZPROP_SRC_LOCAL); 309 } 310 } 311} 312 313/* 314 * Get zpool property values. 315 */ 316int 317spa_prop_get(spa_t *spa, nvlist_t **nvp) 318{ 319 objset_t *mos = spa->spa_meta_objset; 320 zap_cursor_t zc; 321 zap_attribute_t za; 322 int err; 323 324 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 325 326 mutex_enter(&spa->spa_props_lock); 327 328 /* 329 * Get properties from the spa config. 330 */ 331 spa_prop_get_config(spa, nvp); 332 333 /* If no pool property object, no more prop to get. */ 334 if (mos == NULL || spa->spa_pool_props_object == 0) { 335 mutex_exit(&spa->spa_props_lock); 336 return (0); 337 } 338 339 /* 340 * Get properties from the MOS pool property object. 341 */ 342 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 343 (err = zap_cursor_retrieve(&zc, &za)) == 0; 344 zap_cursor_advance(&zc)) { 345 uint64_t intval = 0; 346 char *strval = NULL; 347 zprop_source_t src = ZPROP_SRC_DEFAULT; 348 zpool_prop_t prop; 349 350 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 351 continue; 352 353 switch (za.za_integer_length) { 354 case 8: 355 /* integer property */ 356 if (za.za_first_integer != 357 zpool_prop_default_numeric(prop)) 358 src = ZPROP_SRC_LOCAL; 359 360 if (prop == ZPOOL_PROP_BOOTFS) { 361 dsl_pool_t *dp; 362 dsl_dataset_t *ds = NULL; 363 364 dp = spa_get_dsl(spa); 365 dsl_pool_config_enter(dp, FTAG); 366 if (err = dsl_dataset_hold_obj(dp, 367 za.za_first_integer, FTAG, &ds)) { 368 dsl_pool_config_exit(dp, FTAG); 369 break; 370 } 371 372 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, 373 KM_SLEEP); 374 dsl_dataset_name(ds, strval); 375 dsl_dataset_rele(ds, FTAG); 376 dsl_pool_config_exit(dp, FTAG); 377 } else { 378 strval = NULL; 379 intval = za.za_first_integer; 380 } 381 382 spa_prop_add_list(*nvp, prop, strval, intval, src); 383 384 if (strval != NULL) 385 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN); 386 387 break; 388 389 case 1: 390 /* string property */ 391 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 392 err = zap_lookup(mos, spa->spa_pool_props_object, 393 za.za_name, 1, za.za_num_integers, strval); 394 if (err) { 395 kmem_free(strval, za.za_num_integers); 396 break; 397 } 398 spa_prop_add_list(*nvp, prop, strval, 0, src); 399 kmem_free(strval, za.za_num_integers); 400 break; 401 402 default: 403 break; 404 } 405 } 406 zap_cursor_fini(&zc); 407 mutex_exit(&spa->spa_props_lock); 408out: 409 if (err && err != ENOENT) { 410 nvlist_free(*nvp); 411 *nvp = NULL; 412 return (err); 413 } 414 415 return (0); 416} 417 418/* 419 * Validate the given pool properties nvlist and modify the list 420 * for the property values to be set. 421 */ 422static int 423spa_prop_validate(spa_t *spa, nvlist_t *props) 424{ 425 nvpair_t *elem; 426 int error = 0, reset_bootfs = 0; 427 uint64_t objnum = 0; 428 boolean_t has_feature = B_FALSE; 429 430 elem = NULL; 431 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 432 uint64_t intval; 433 char *strval, *slash, *check, *fname; 434 const char *propname = nvpair_name(elem); 435 zpool_prop_t prop = zpool_name_to_prop(propname); 436 437 switch (prop) { 438 case ZPROP_INVAL: 439 if (!zpool_prop_feature(propname)) { 440 error = SET_ERROR(EINVAL); 441 break; 442 } 443 444 /* 445 * Sanitize the input. 446 */ 447 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 448 error = SET_ERROR(EINVAL); 449 break; 450 } 451 452 if (nvpair_value_uint64(elem, &intval) != 0) { 453 error = SET_ERROR(EINVAL); 454 break; 455 } 456 457 if (intval != 0) { 458 error = SET_ERROR(EINVAL); 459 break; 460 } 461 462 fname = strchr(propname, '@') + 1; 463 if (zfeature_lookup_name(fname, NULL) != 0) { 464 error = SET_ERROR(EINVAL); 465 break; 466 } 467 468 has_feature = B_TRUE; 469 break; 470 471 case ZPOOL_PROP_VERSION: 472 error = nvpair_value_uint64(elem, &intval); 473 if (!error && 474 (intval < spa_version(spa) || 475 intval > SPA_VERSION_BEFORE_FEATURES || 476 has_feature)) 477 error = SET_ERROR(EINVAL); 478 break; 479 480 case ZPOOL_PROP_DELEGATION: 481 case ZPOOL_PROP_AUTOREPLACE: 482 case ZPOOL_PROP_LISTSNAPS: 483 case ZPOOL_PROP_AUTOEXPAND: 484 error = nvpair_value_uint64(elem, &intval); 485 if (!error && intval > 1) 486 error = SET_ERROR(EINVAL); 487 break; 488 489 case ZPOOL_PROP_BOOTFS: 490 /* 491 * If the pool version is less than SPA_VERSION_BOOTFS, 492 * or the pool is still being created (version == 0), 493 * the bootfs property cannot be set. 494 */ 495 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 496 error = SET_ERROR(ENOTSUP); 497 break; 498 } 499 500 /* 501 * Make sure the vdev config is bootable 502 */ 503 if (!vdev_is_bootable(spa->spa_root_vdev)) { 504 error = SET_ERROR(ENOTSUP); 505 break; 506 } 507 508 reset_bootfs = 1; 509 510 error = nvpair_value_string(elem, &strval); 511 512 if (!error) { 513 objset_t *os; 514 uint64_t propval; 515 516 if (strval == NULL || strval[0] == '\0') { 517 objnum = zpool_prop_default_numeric( 518 ZPOOL_PROP_BOOTFS); 519 break; 520 } 521 522 if (error = dmu_objset_hold(strval, FTAG, &os)) 523 break; 524 525 /* 526 * Must be ZPL, and its property settings 527 * must be supported by GRUB (compression 528 * is not gzip, and large blocks are not used). 529 */ 530 531 if (dmu_objset_type(os) != DMU_OST_ZFS) { 532 error = SET_ERROR(ENOTSUP); 533 } else if ((error = 534 dsl_prop_get_int_ds(dmu_objset_ds(os), 535 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 536 &propval)) == 0 && 537 !BOOTFS_COMPRESS_VALID(propval)) { 538 error = SET_ERROR(ENOTSUP); 539 } else if ((error = 540 dsl_prop_get_int_ds(dmu_objset_ds(os), 541 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 542 &propval)) == 0 && 543 propval > SPA_OLD_MAXBLOCKSIZE) { 544 error = SET_ERROR(ENOTSUP); 545 } else { 546 objnum = dmu_objset_id(os); 547 } 548 dmu_objset_rele(os, FTAG); 549 } 550 break; 551 552 case ZPOOL_PROP_FAILUREMODE: 553 error = nvpair_value_uint64(elem, &intval); 554 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 555 intval > ZIO_FAILURE_MODE_PANIC)) 556 error = SET_ERROR(EINVAL); 557 558 /* 559 * This is a special case which only occurs when 560 * the pool has completely failed. This allows 561 * the user to change the in-core failmode property 562 * without syncing it out to disk (I/Os might 563 * currently be blocked). We do this by returning 564 * EIO to the caller (spa_prop_set) to trick it 565 * into thinking we encountered a property validation 566 * error. 567 */ 568 if (!error && spa_suspended(spa)) { 569 spa->spa_failmode = intval; 570 error = SET_ERROR(EIO); 571 } 572 break; 573 574 case ZPOOL_PROP_CACHEFILE: 575 if ((error = nvpair_value_string(elem, &strval)) != 0) 576 break; 577 578 if (strval[0] == '\0') 579 break; 580 581 if (strcmp(strval, "none") == 0) 582 break; 583 584 if (strval[0] != '/') { 585 error = SET_ERROR(EINVAL); 586 break; 587 } 588 589 slash = strrchr(strval, '/'); 590 ASSERT(slash != NULL); 591 592 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 593 strcmp(slash, "/..") == 0) 594 error = SET_ERROR(EINVAL); 595 break; 596 597 case ZPOOL_PROP_COMMENT: 598 if ((error = nvpair_value_string(elem, &strval)) != 0) 599 break; 600 for (check = strval; *check != '\0'; check++) { 601 /* 602 * The kernel doesn't have an easy isprint() 603 * check. For this kernel check, we merely 604 * check ASCII apart from DEL. Fix this if 605 * there is an easy-to-use kernel isprint(). 606 */ 607 if (*check >= 0x7f) { 608 error = SET_ERROR(EINVAL); 609 break; 610 } 611 } 612 if (strlen(strval) > ZPROP_MAX_COMMENT) 613 error = E2BIG; 614 break; 615 616 case ZPOOL_PROP_DEDUPDITTO: 617 if (spa_version(spa) < SPA_VERSION_DEDUP) 618 error = SET_ERROR(ENOTSUP); 619 else 620 error = nvpair_value_uint64(elem, &intval); 621 if (error == 0 && 622 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 623 error = SET_ERROR(EINVAL); 624 break; 625 } 626 627 if (error) 628 break; 629 } 630 631 if (!error && reset_bootfs) { 632 error = nvlist_remove(props, 633 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 634 635 if (!error) { 636 error = nvlist_add_uint64(props, 637 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 638 } 639 } 640 641 return (error); 642} 643 644void 645spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 646{ 647 char *cachefile; 648 spa_config_dirent_t *dp; 649 650 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 651 &cachefile) != 0) 652 return; 653 654 dp = kmem_alloc(sizeof (spa_config_dirent_t), 655 KM_SLEEP); 656 657 if (cachefile[0] == '\0') 658 dp->scd_path = spa_strdup(spa_config_path); 659 else if (strcmp(cachefile, "none") == 0) 660 dp->scd_path = NULL; 661 else 662 dp->scd_path = spa_strdup(cachefile); 663 664 list_insert_head(&spa->spa_config_list, dp); 665 if (need_sync) 666 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 667} 668 669int 670spa_prop_set(spa_t *spa, nvlist_t *nvp) 671{ 672 int error; 673 nvpair_t *elem = NULL; 674 boolean_t need_sync = B_FALSE; 675 676 if ((error = spa_prop_validate(spa, nvp)) != 0) 677 return (error); 678 679 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 680 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 681 682 if (prop == ZPOOL_PROP_CACHEFILE || 683 prop == ZPOOL_PROP_ALTROOT || 684 prop == ZPOOL_PROP_READONLY) 685 continue; 686 687 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 688 uint64_t ver; 689 690 if (prop == ZPOOL_PROP_VERSION) { 691 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 692 } else { 693 ASSERT(zpool_prop_feature(nvpair_name(elem))); 694 ver = SPA_VERSION_FEATURES; 695 need_sync = B_TRUE; 696 } 697 698 /* Save time if the version is already set. */ 699 if (ver == spa_version(spa)) 700 continue; 701 702 /* 703 * In addition to the pool directory object, we might 704 * create the pool properties object, the features for 705 * read object, the features for write object, or the 706 * feature descriptions object. 707 */ 708 error = dsl_sync_task(spa->spa_name, NULL, 709 spa_sync_version, &ver, 710 6, ZFS_SPACE_CHECK_RESERVED); 711 if (error) 712 return (error); 713 continue; 714 } 715 716 need_sync = B_TRUE; 717 break; 718 } 719 720 if (need_sync) { 721 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 722 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 723 } 724 725 return (0); 726} 727 728/* 729 * If the bootfs property value is dsobj, clear it. 730 */ 731void 732spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 733{ 734 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 735 VERIFY(zap_remove(spa->spa_meta_objset, 736 spa->spa_pool_props_object, 737 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 738 spa->spa_bootfs = 0; 739 } 740} 741 742/*ARGSUSED*/ 743static int 744spa_change_guid_check(void *arg, dmu_tx_t *tx) 745{ 746 uint64_t *newguid = arg; 747 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 748 vdev_t *rvd = spa->spa_root_vdev; 749 uint64_t vdev_state; 750 751 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 752 vdev_state = rvd->vdev_state; 753 spa_config_exit(spa, SCL_STATE, FTAG); 754 755 if (vdev_state != VDEV_STATE_HEALTHY) 756 return (SET_ERROR(ENXIO)); 757 758 ASSERT3U(spa_guid(spa), !=, *newguid); 759 760 return (0); 761} 762 763static void 764spa_change_guid_sync(void *arg, dmu_tx_t *tx) 765{ 766 uint64_t *newguid = arg; 767 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 768 uint64_t oldguid; 769 vdev_t *rvd = spa->spa_root_vdev; 770 771 oldguid = spa_guid(spa); 772 773 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 774 rvd->vdev_guid = *newguid; 775 rvd->vdev_guid_sum += (*newguid - oldguid); 776 vdev_config_dirty(rvd); 777 spa_config_exit(spa, SCL_STATE, FTAG); 778 779 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 780 oldguid, *newguid); 781} 782 783/* 784 * Change the GUID for the pool. This is done so that we can later 785 * re-import a pool built from a clone of our own vdevs. We will modify 786 * the root vdev's guid, our own pool guid, and then mark all of our 787 * vdevs dirty. Note that we must make sure that all our vdevs are 788 * online when we do this, or else any vdevs that weren't present 789 * would be orphaned from our pool. We are also going to issue a 790 * sysevent to update any watchers. 791 */ 792int 793spa_change_guid(spa_t *spa) 794{ 795 int error; 796 uint64_t guid; 797 798 mutex_enter(&spa->spa_vdev_top_lock); 799 mutex_enter(&spa_namespace_lock); 800 guid = spa_generate_guid(NULL); 801 802 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 803 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 804 805 if (error == 0) { 806 spa_config_sync(spa, B_FALSE, B_TRUE); 807 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 808 } 809 810 mutex_exit(&spa_namespace_lock); 811 mutex_exit(&spa->spa_vdev_top_lock); 812 813 return (error); 814} 815 816/* 817 * ========================================================================== 818 * SPA state manipulation (open/create/destroy/import/export) 819 * ========================================================================== 820 */ 821 822static int 823spa_error_entry_compare(const void *a, const void *b) 824{ 825 spa_error_entry_t *sa = (spa_error_entry_t *)a; 826 spa_error_entry_t *sb = (spa_error_entry_t *)b; 827 int ret; 828 829 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 830 sizeof (zbookmark_phys_t)); 831 832 if (ret < 0) 833 return (-1); 834 else if (ret > 0) 835 return (1); 836 else 837 return (0); 838} 839 840/* 841 * Utility function which retrieves copies of the current logs and 842 * re-initializes them in the process. 843 */ 844void 845spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 846{ 847 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 848 849 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 850 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 851 852 avl_create(&spa->spa_errlist_scrub, 853 spa_error_entry_compare, sizeof (spa_error_entry_t), 854 offsetof(spa_error_entry_t, se_avl)); 855 avl_create(&spa->spa_errlist_last, 856 spa_error_entry_compare, sizeof (spa_error_entry_t), 857 offsetof(spa_error_entry_t, se_avl)); 858} 859 860static void 861spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 862{ 863 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 864 enum zti_modes mode = ztip->zti_mode; 865 uint_t value = ztip->zti_value; 866 uint_t count = ztip->zti_count; 867 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 868 char name[32]; 869 uint_t flags = 0; 870 boolean_t batch = B_FALSE; 871 872 if (mode == ZTI_MODE_NULL) { 873 tqs->stqs_count = 0; 874 tqs->stqs_taskq = NULL; 875 return; 876 } 877 878 ASSERT3U(count, >, 0); 879 880 tqs->stqs_count = count; 881 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 882 883 switch (mode) { 884 case ZTI_MODE_FIXED: 885 ASSERT3U(value, >=, 1); 886 value = MAX(value, 1); 887 break; 888 889 case ZTI_MODE_BATCH: 890 batch = B_TRUE; 891 flags |= TASKQ_THREADS_CPU_PCT; 892 value = zio_taskq_batch_pct; 893 break; 894 895 default: 896 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 897 "spa_activate()", 898 zio_type_name[t], zio_taskq_types[q], mode, value); 899 break; 900 } 901 902 for (uint_t i = 0; i < count; i++) { 903 taskq_t *tq; 904 905 if (count > 1) { 906 (void) snprintf(name, sizeof (name), "%s_%s_%u", 907 zio_type_name[t], zio_taskq_types[q], i); 908 } else { 909 (void) snprintf(name, sizeof (name), "%s_%s", 910 zio_type_name[t], zio_taskq_types[q]); 911 } 912 913#ifdef SYSDC 914 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 915 if (batch) 916 flags |= TASKQ_DC_BATCH; 917 918 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 919 spa->spa_proc, zio_taskq_basedc, flags); 920 } else { 921#endif 922 pri_t pri = maxclsyspri; 923 /* 924 * The write issue taskq can be extremely CPU 925 * intensive. Run it at slightly lower priority 926 * than the other taskqs. 927 * FreeBSD notes: 928 * - numerically higher priorities are lower priorities; 929 * - if priorities divided by four (RQ_PPQ) are equal 930 * then a difference between them is insignificant. 931 */ 932 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 933#ifdef illumos 934 pri--; 935#else 936 pri += 4; 937#endif 938 939 tq = taskq_create_proc(name, value, pri, 50, 940 INT_MAX, spa->spa_proc, flags); 941#ifdef SYSDC 942 } 943#endif 944 945 tqs->stqs_taskq[i] = tq; 946 } 947} 948 949static void 950spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 951{ 952 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 953 954 if (tqs->stqs_taskq == NULL) { 955 ASSERT0(tqs->stqs_count); 956 return; 957 } 958 959 for (uint_t i = 0; i < tqs->stqs_count; i++) { 960 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 961 taskq_destroy(tqs->stqs_taskq[i]); 962 } 963 964 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 965 tqs->stqs_taskq = NULL; 966} 967 968/* 969 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 970 * Note that a type may have multiple discrete taskqs to avoid lock contention 971 * on the taskq itself. In that case we choose which taskq at random by using 972 * the low bits of gethrtime(). 973 */ 974void 975spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 976 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 977{ 978 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 979 taskq_t *tq; 980 981 ASSERT3P(tqs->stqs_taskq, !=, NULL); 982 ASSERT3U(tqs->stqs_count, !=, 0); 983 984 if (tqs->stqs_count == 1) { 985 tq = tqs->stqs_taskq[0]; 986 } else { 987#ifdef _KERNEL 988 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 989#else 990 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 991#endif 992 } 993 994 taskq_dispatch_ent(tq, func, arg, flags, ent); 995} 996 997static void 998spa_create_zio_taskqs(spa_t *spa) 999{ 1000 for (int t = 0; t < ZIO_TYPES; t++) { 1001 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1002 spa_taskqs_init(spa, t, q); 1003 } 1004 } 1005} 1006 1007#ifdef _KERNEL 1008#ifdef SPA_PROCESS 1009static void 1010spa_thread(void *arg) 1011{ 1012 callb_cpr_t cprinfo; 1013 1014 spa_t *spa = arg; 1015 user_t *pu = PTOU(curproc); 1016 1017 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1018 spa->spa_name); 1019 1020 ASSERT(curproc != &p0); 1021 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1022 "zpool-%s", spa->spa_name); 1023 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1024 1025#ifdef PSRSET_BIND 1026 /* bind this thread to the requested psrset */ 1027 if (zio_taskq_psrset_bind != PS_NONE) { 1028 pool_lock(); 1029 mutex_enter(&cpu_lock); 1030 mutex_enter(&pidlock); 1031 mutex_enter(&curproc->p_lock); 1032 1033 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1034 0, NULL, NULL) == 0) { 1035 curthread->t_bind_pset = zio_taskq_psrset_bind; 1036 } else { 1037 cmn_err(CE_WARN, 1038 "Couldn't bind process for zfs pool \"%s\" to " 1039 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1040 } 1041 1042 mutex_exit(&curproc->p_lock); 1043 mutex_exit(&pidlock); 1044 mutex_exit(&cpu_lock); 1045 pool_unlock(); 1046 } 1047#endif 1048 1049#ifdef SYSDC 1050 if (zio_taskq_sysdc) { 1051 sysdc_thread_enter(curthread, 100, 0); 1052 } 1053#endif 1054 1055 spa->spa_proc = curproc; 1056 spa->spa_did = curthread->t_did; 1057 1058 spa_create_zio_taskqs(spa); 1059 1060 mutex_enter(&spa->spa_proc_lock); 1061 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1062 1063 spa->spa_proc_state = SPA_PROC_ACTIVE; 1064 cv_broadcast(&spa->spa_proc_cv); 1065 1066 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1067 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1068 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1069 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1070 1071 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1072 spa->spa_proc_state = SPA_PROC_GONE; 1073 spa->spa_proc = &p0; 1074 cv_broadcast(&spa->spa_proc_cv); 1075 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1076 1077 mutex_enter(&curproc->p_lock); 1078 lwp_exit(); 1079} 1080#endif /* SPA_PROCESS */ 1081#endif 1082 1083/* 1084 * Activate an uninitialized pool. 1085 */ 1086static void 1087spa_activate(spa_t *spa, int mode) 1088{ 1089 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1090 1091 spa->spa_state = POOL_STATE_ACTIVE; 1092 spa->spa_mode = mode; 1093 1094 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1095 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1096 1097 /* Try to create a covering process */ 1098 mutex_enter(&spa->spa_proc_lock); 1099 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1100 ASSERT(spa->spa_proc == &p0); 1101 spa->spa_did = 0; 1102 1103#ifdef SPA_PROCESS 1104 /* Only create a process if we're going to be around a while. */ 1105 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1106 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1107 NULL, 0) == 0) { 1108 spa->spa_proc_state = SPA_PROC_CREATED; 1109 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1110 cv_wait(&spa->spa_proc_cv, 1111 &spa->spa_proc_lock); 1112 } 1113 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1114 ASSERT(spa->spa_proc != &p0); 1115 ASSERT(spa->spa_did != 0); 1116 } else { 1117#ifdef _KERNEL 1118 cmn_err(CE_WARN, 1119 "Couldn't create process for zfs pool \"%s\"\n", 1120 spa->spa_name); 1121#endif 1122 } 1123 } 1124#endif /* SPA_PROCESS */ 1125 mutex_exit(&spa->spa_proc_lock); 1126 1127 /* If we didn't create a process, we need to create our taskqs. */ 1128 ASSERT(spa->spa_proc == &p0); 1129 if (spa->spa_proc == &p0) { 1130 spa_create_zio_taskqs(spa); 1131 } 1132 1133 /* 1134 * Start TRIM thread. 1135 */ 1136 trim_thread_create(spa); 1137 1138 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1139 offsetof(vdev_t, vdev_config_dirty_node)); 1140 list_create(&spa->spa_evicting_os_list, sizeof (objset_t), 1141 offsetof(objset_t, os_evicting_node)); 1142 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1143 offsetof(vdev_t, vdev_state_dirty_node)); 1144 1145 txg_list_create(&spa->spa_vdev_txg_list, 1146 offsetof(struct vdev, vdev_txg_node)); 1147 1148 avl_create(&spa->spa_errlist_scrub, 1149 spa_error_entry_compare, sizeof (spa_error_entry_t), 1150 offsetof(spa_error_entry_t, se_avl)); 1151 avl_create(&spa->spa_errlist_last, 1152 spa_error_entry_compare, sizeof (spa_error_entry_t), 1153 offsetof(spa_error_entry_t, se_avl)); 1154} 1155 1156/* 1157 * Opposite of spa_activate(). 1158 */ 1159static void 1160spa_deactivate(spa_t *spa) 1161{ 1162 ASSERT(spa->spa_sync_on == B_FALSE); 1163 ASSERT(spa->spa_dsl_pool == NULL); 1164 ASSERT(spa->spa_root_vdev == NULL); 1165 ASSERT(spa->spa_async_zio_root == NULL); 1166 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1167 1168 /* 1169 * Stop TRIM thread in case spa_unload() wasn't called directly 1170 * before spa_deactivate(). 1171 */ 1172 trim_thread_destroy(spa); 1173 1174 spa_evicting_os_wait(spa); 1175 1176 txg_list_destroy(&spa->spa_vdev_txg_list); 1177 1178 list_destroy(&spa->spa_config_dirty_list); 1179 list_destroy(&spa->spa_evicting_os_list); 1180 list_destroy(&spa->spa_state_dirty_list); 1181 1182 for (int t = 0; t < ZIO_TYPES; t++) { 1183 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1184 spa_taskqs_fini(spa, t, q); 1185 } 1186 } 1187 1188 metaslab_class_destroy(spa->spa_normal_class); 1189 spa->spa_normal_class = NULL; 1190 1191 metaslab_class_destroy(spa->spa_log_class); 1192 spa->spa_log_class = NULL; 1193 1194 /* 1195 * If this was part of an import or the open otherwise failed, we may 1196 * still have errors left in the queues. Empty them just in case. 1197 */ 1198 spa_errlog_drain(spa); 1199 1200 avl_destroy(&spa->spa_errlist_scrub); 1201 avl_destroy(&spa->spa_errlist_last); 1202 1203 spa->spa_state = POOL_STATE_UNINITIALIZED; 1204 1205 mutex_enter(&spa->spa_proc_lock); 1206 if (spa->spa_proc_state != SPA_PROC_NONE) { 1207 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1208 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1209 cv_broadcast(&spa->spa_proc_cv); 1210 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1211 ASSERT(spa->spa_proc != &p0); 1212 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1213 } 1214 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1215 spa->spa_proc_state = SPA_PROC_NONE; 1216 } 1217 ASSERT(spa->spa_proc == &p0); 1218 mutex_exit(&spa->spa_proc_lock); 1219 1220#ifdef SPA_PROCESS 1221 /* 1222 * We want to make sure spa_thread() has actually exited the ZFS 1223 * module, so that the module can't be unloaded out from underneath 1224 * it. 1225 */ 1226 if (spa->spa_did != 0) { 1227 thread_join(spa->spa_did); 1228 spa->spa_did = 0; 1229 } 1230#endif /* SPA_PROCESS */ 1231} 1232 1233/* 1234 * Verify a pool configuration, and construct the vdev tree appropriately. This 1235 * will create all the necessary vdevs in the appropriate layout, with each vdev 1236 * in the CLOSED state. This will prep the pool before open/creation/import. 1237 * All vdev validation is done by the vdev_alloc() routine. 1238 */ 1239static int 1240spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1241 uint_t id, int atype) 1242{ 1243 nvlist_t **child; 1244 uint_t children; 1245 int error; 1246 1247 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1248 return (error); 1249 1250 if ((*vdp)->vdev_ops->vdev_op_leaf) 1251 return (0); 1252 1253 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1254 &child, &children); 1255 1256 if (error == ENOENT) 1257 return (0); 1258 1259 if (error) { 1260 vdev_free(*vdp); 1261 *vdp = NULL; 1262 return (SET_ERROR(EINVAL)); 1263 } 1264 1265 for (int c = 0; c < children; c++) { 1266 vdev_t *vd; 1267 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1268 atype)) != 0) { 1269 vdev_free(*vdp); 1270 *vdp = NULL; 1271 return (error); 1272 } 1273 } 1274 1275 ASSERT(*vdp != NULL); 1276 1277 return (0); 1278} 1279 1280/* 1281 * Opposite of spa_load(). 1282 */ 1283static void 1284spa_unload(spa_t *spa) 1285{ 1286 int i; 1287 1288 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1289 1290 /* 1291 * Stop TRIM thread. 1292 */ 1293 trim_thread_destroy(spa); 1294 1295 /* 1296 * Stop async tasks. 1297 */ 1298 spa_async_suspend(spa); 1299 1300 /* 1301 * Stop syncing. 1302 */ 1303 if (spa->spa_sync_on) { 1304 txg_sync_stop(spa->spa_dsl_pool); 1305 spa->spa_sync_on = B_FALSE; 1306 } 1307 1308 /* 1309 * Wait for any outstanding async I/O to complete. 1310 */ 1311 if (spa->spa_async_zio_root != NULL) { 1312 for (int i = 0; i < max_ncpus; i++) 1313 (void) zio_wait(spa->spa_async_zio_root[i]); 1314 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1315 spa->spa_async_zio_root = NULL; 1316 } 1317 1318 bpobj_close(&spa->spa_deferred_bpobj); 1319 1320 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1321 1322 /* 1323 * Close all vdevs. 1324 */ 1325 if (spa->spa_root_vdev) 1326 vdev_free(spa->spa_root_vdev); 1327 ASSERT(spa->spa_root_vdev == NULL); 1328 1329 /* 1330 * Close the dsl pool. 1331 */ 1332 if (spa->spa_dsl_pool) { 1333 dsl_pool_close(spa->spa_dsl_pool); 1334 spa->spa_dsl_pool = NULL; 1335 spa->spa_meta_objset = NULL; 1336 } 1337 1338 ddt_unload(spa); 1339 1340 /* 1341 * Drop and purge level 2 cache 1342 */ 1343 spa_l2cache_drop(spa); 1344 1345 for (i = 0; i < spa->spa_spares.sav_count; i++) 1346 vdev_free(spa->spa_spares.sav_vdevs[i]); 1347 if (spa->spa_spares.sav_vdevs) { 1348 kmem_free(spa->spa_spares.sav_vdevs, 1349 spa->spa_spares.sav_count * sizeof (void *)); 1350 spa->spa_spares.sav_vdevs = NULL; 1351 } 1352 if (spa->spa_spares.sav_config) { 1353 nvlist_free(spa->spa_spares.sav_config); 1354 spa->spa_spares.sav_config = NULL; 1355 } 1356 spa->spa_spares.sav_count = 0; 1357 1358 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1359 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1360 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1361 } 1362 if (spa->spa_l2cache.sav_vdevs) { 1363 kmem_free(spa->spa_l2cache.sav_vdevs, 1364 spa->spa_l2cache.sav_count * sizeof (void *)); 1365 spa->spa_l2cache.sav_vdevs = NULL; 1366 } 1367 if (spa->spa_l2cache.sav_config) { 1368 nvlist_free(spa->spa_l2cache.sav_config); 1369 spa->spa_l2cache.sav_config = NULL; 1370 } 1371 spa->spa_l2cache.sav_count = 0; 1372 1373 spa->spa_async_suspended = 0; 1374 1375 if (spa->spa_comment != NULL) { 1376 spa_strfree(spa->spa_comment); 1377 spa->spa_comment = NULL; 1378 } 1379 1380 spa_config_exit(spa, SCL_ALL, FTAG); 1381} 1382 1383/* 1384 * Load (or re-load) the current list of vdevs describing the active spares for 1385 * this pool. When this is called, we have some form of basic information in 1386 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1387 * then re-generate a more complete list including status information. 1388 */ 1389static void 1390spa_load_spares(spa_t *spa) 1391{ 1392 nvlist_t **spares; 1393 uint_t nspares; 1394 int i; 1395 vdev_t *vd, *tvd; 1396 1397 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1398 1399 /* 1400 * First, close and free any existing spare vdevs. 1401 */ 1402 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1403 vd = spa->spa_spares.sav_vdevs[i]; 1404 1405 /* Undo the call to spa_activate() below */ 1406 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1407 B_FALSE)) != NULL && tvd->vdev_isspare) 1408 spa_spare_remove(tvd); 1409 vdev_close(vd); 1410 vdev_free(vd); 1411 } 1412 1413 if (spa->spa_spares.sav_vdevs) 1414 kmem_free(spa->spa_spares.sav_vdevs, 1415 spa->spa_spares.sav_count * sizeof (void *)); 1416 1417 if (spa->spa_spares.sav_config == NULL) 1418 nspares = 0; 1419 else 1420 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1421 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1422 1423 spa->spa_spares.sav_count = (int)nspares; 1424 spa->spa_spares.sav_vdevs = NULL; 1425 1426 if (nspares == 0) 1427 return; 1428 1429 /* 1430 * Construct the array of vdevs, opening them to get status in the 1431 * process. For each spare, there is potentially two different vdev_t 1432 * structures associated with it: one in the list of spares (used only 1433 * for basic validation purposes) and one in the active vdev 1434 * configuration (if it's spared in). During this phase we open and 1435 * validate each vdev on the spare list. If the vdev also exists in the 1436 * active configuration, then we also mark this vdev as an active spare. 1437 */ 1438 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1439 KM_SLEEP); 1440 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1441 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1442 VDEV_ALLOC_SPARE) == 0); 1443 ASSERT(vd != NULL); 1444 1445 spa->spa_spares.sav_vdevs[i] = vd; 1446 1447 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1448 B_FALSE)) != NULL) { 1449 if (!tvd->vdev_isspare) 1450 spa_spare_add(tvd); 1451 1452 /* 1453 * We only mark the spare active if we were successfully 1454 * able to load the vdev. Otherwise, importing a pool 1455 * with a bad active spare would result in strange 1456 * behavior, because multiple pool would think the spare 1457 * is actively in use. 1458 * 1459 * There is a vulnerability here to an equally bizarre 1460 * circumstance, where a dead active spare is later 1461 * brought back to life (onlined or otherwise). Given 1462 * the rarity of this scenario, and the extra complexity 1463 * it adds, we ignore the possibility. 1464 */ 1465 if (!vdev_is_dead(tvd)) 1466 spa_spare_activate(tvd); 1467 } 1468 1469 vd->vdev_top = vd; 1470 vd->vdev_aux = &spa->spa_spares; 1471 1472 if (vdev_open(vd) != 0) 1473 continue; 1474 1475 if (vdev_validate_aux(vd) == 0) 1476 spa_spare_add(vd); 1477 } 1478 1479 /* 1480 * Recompute the stashed list of spares, with status information 1481 * this time. 1482 */ 1483 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1484 DATA_TYPE_NVLIST_ARRAY) == 0); 1485 1486 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1487 KM_SLEEP); 1488 for (i = 0; i < spa->spa_spares.sav_count; i++) 1489 spares[i] = vdev_config_generate(spa, 1490 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1491 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1492 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1493 for (i = 0; i < spa->spa_spares.sav_count; i++) 1494 nvlist_free(spares[i]); 1495 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1496} 1497 1498/* 1499 * Load (or re-load) the current list of vdevs describing the active l2cache for 1500 * this pool. When this is called, we have some form of basic information in 1501 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1502 * then re-generate a more complete list including status information. 1503 * Devices which are already active have their details maintained, and are 1504 * not re-opened. 1505 */ 1506static void 1507spa_load_l2cache(spa_t *spa) 1508{ 1509 nvlist_t **l2cache; 1510 uint_t nl2cache; 1511 int i, j, oldnvdevs; 1512 uint64_t guid; 1513 vdev_t *vd, **oldvdevs, **newvdevs; 1514 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1515 1516 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1517 1518 if (sav->sav_config != NULL) { 1519 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1520 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1521 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1522 } else { 1523 nl2cache = 0; 1524 newvdevs = NULL; 1525 } 1526 1527 oldvdevs = sav->sav_vdevs; 1528 oldnvdevs = sav->sav_count; 1529 sav->sav_vdevs = NULL; 1530 sav->sav_count = 0; 1531 1532 /* 1533 * Process new nvlist of vdevs. 1534 */ 1535 for (i = 0; i < nl2cache; i++) { 1536 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1537 &guid) == 0); 1538 1539 newvdevs[i] = NULL; 1540 for (j = 0; j < oldnvdevs; j++) { 1541 vd = oldvdevs[j]; 1542 if (vd != NULL && guid == vd->vdev_guid) { 1543 /* 1544 * Retain previous vdev for add/remove ops. 1545 */ 1546 newvdevs[i] = vd; 1547 oldvdevs[j] = NULL; 1548 break; 1549 } 1550 } 1551 1552 if (newvdevs[i] == NULL) { 1553 /* 1554 * Create new vdev 1555 */ 1556 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1557 VDEV_ALLOC_L2CACHE) == 0); 1558 ASSERT(vd != NULL); 1559 newvdevs[i] = vd; 1560 1561 /* 1562 * Commit this vdev as an l2cache device, 1563 * even if it fails to open. 1564 */ 1565 spa_l2cache_add(vd); 1566 1567 vd->vdev_top = vd; 1568 vd->vdev_aux = sav; 1569 1570 spa_l2cache_activate(vd); 1571 1572 if (vdev_open(vd) != 0) 1573 continue; 1574 1575 (void) vdev_validate_aux(vd); 1576 1577 if (!vdev_is_dead(vd)) 1578 l2arc_add_vdev(spa, vd); 1579 } 1580 } 1581 1582 /* 1583 * Purge vdevs that were dropped 1584 */ 1585 for (i = 0; i < oldnvdevs; i++) { 1586 uint64_t pool; 1587 1588 vd = oldvdevs[i]; 1589 if (vd != NULL) { 1590 ASSERT(vd->vdev_isl2cache); 1591 1592 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1593 pool != 0ULL && l2arc_vdev_present(vd)) 1594 l2arc_remove_vdev(vd); 1595 vdev_clear_stats(vd); 1596 vdev_free(vd); 1597 } 1598 } 1599 1600 if (oldvdevs) 1601 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1602 1603 if (sav->sav_config == NULL) 1604 goto out; 1605 1606 sav->sav_vdevs = newvdevs; 1607 sav->sav_count = (int)nl2cache; 1608 1609 /* 1610 * Recompute the stashed list of l2cache devices, with status 1611 * information this time. 1612 */ 1613 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1614 DATA_TYPE_NVLIST_ARRAY) == 0); 1615 1616 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1617 for (i = 0; i < sav->sav_count; i++) 1618 l2cache[i] = vdev_config_generate(spa, 1619 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1620 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1621 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1622out: 1623 for (i = 0; i < sav->sav_count; i++) 1624 nvlist_free(l2cache[i]); 1625 if (sav->sav_count) 1626 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1627} 1628 1629static int 1630load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1631{ 1632 dmu_buf_t *db; 1633 char *packed = NULL; 1634 size_t nvsize = 0; 1635 int error; 1636 *value = NULL; 1637 1638 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1639 if (error != 0) 1640 return (error); 1641 1642 nvsize = *(uint64_t *)db->db_data; 1643 dmu_buf_rele(db, FTAG); 1644 1645 packed = kmem_alloc(nvsize, KM_SLEEP); 1646 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1647 DMU_READ_PREFETCH); 1648 if (error == 0) 1649 error = nvlist_unpack(packed, nvsize, value, 0); 1650 kmem_free(packed, nvsize); 1651 1652 return (error); 1653} 1654 1655/* 1656 * Checks to see if the given vdev could not be opened, in which case we post a 1657 * sysevent to notify the autoreplace code that the device has been removed. 1658 */ 1659static void 1660spa_check_removed(vdev_t *vd) 1661{ 1662 for (int c = 0; c < vd->vdev_children; c++) 1663 spa_check_removed(vd->vdev_child[c]); 1664 1665 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1666 !vd->vdev_ishole) { 1667 zfs_post_autoreplace(vd->vdev_spa, vd); 1668 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1669 } 1670} 1671 1672/* 1673 * Validate the current config against the MOS config 1674 */ 1675static boolean_t 1676spa_config_valid(spa_t *spa, nvlist_t *config) 1677{ 1678 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1679 nvlist_t *nv; 1680 1681 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1682 1683 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1684 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1685 1686 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1687 1688 /* 1689 * If we're doing a normal import, then build up any additional 1690 * diagnostic information about missing devices in this config. 1691 * We'll pass this up to the user for further processing. 1692 */ 1693 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1694 nvlist_t **child, *nv; 1695 uint64_t idx = 0; 1696 1697 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1698 KM_SLEEP); 1699 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1700 1701 for (int c = 0; c < rvd->vdev_children; c++) { 1702 vdev_t *tvd = rvd->vdev_child[c]; 1703 vdev_t *mtvd = mrvd->vdev_child[c]; 1704 1705 if (tvd->vdev_ops == &vdev_missing_ops && 1706 mtvd->vdev_ops != &vdev_missing_ops && 1707 mtvd->vdev_islog) 1708 child[idx++] = vdev_config_generate(spa, mtvd, 1709 B_FALSE, 0); 1710 } 1711 1712 if (idx) { 1713 VERIFY(nvlist_add_nvlist_array(nv, 1714 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1715 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1716 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1717 1718 for (int i = 0; i < idx; i++) 1719 nvlist_free(child[i]); 1720 } 1721 nvlist_free(nv); 1722 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1723 } 1724 1725 /* 1726 * Compare the root vdev tree with the information we have 1727 * from the MOS config (mrvd). Check each top-level vdev 1728 * with the corresponding MOS config top-level (mtvd). 1729 */ 1730 for (int c = 0; c < rvd->vdev_children; c++) { 1731 vdev_t *tvd = rvd->vdev_child[c]; 1732 vdev_t *mtvd = mrvd->vdev_child[c]; 1733 1734 /* 1735 * Resolve any "missing" vdevs in the current configuration. 1736 * If we find that the MOS config has more accurate information 1737 * about the top-level vdev then use that vdev instead. 1738 */ 1739 if (tvd->vdev_ops == &vdev_missing_ops && 1740 mtvd->vdev_ops != &vdev_missing_ops) { 1741 1742 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1743 continue; 1744 1745 /* 1746 * Device specific actions. 1747 */ 1748 if (mtvd->vdev_islog) { 1749 spa_set_log_state(spa, SPA_LOG_CLEAR); 1750 } else { 1751 /* 1752 * XXX - once we have 'readonly' pool 1753 * support we should be able to handle 1754 * missing data devices by transitioning 1755 * the pool to readonly. 1756 */ 1757 continue; 1758 } 1759 1760 /* 1761 * Swap the missing vdev with the data we were 1762 * able to obtain from the MOS config. 1763 */ 1764 vdev_remove_child(rvd, tvd); 1765 vdev_remove_child(mrvd, mtvd); 1766 1767 vdev_add_child(rvd, mtvd); 1768 vdev_add_child(mrvd, tvd); 1769 1770 spa_config_exit(spa, SCL_ALL, FTAG); 1771 vdev_load(mtvd); 1772 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1773 1774 vdev_reopen(rvd); 1775 } else if (mtvd->vdev_islog) { 1776 /* 1777 * Load the slog device's state from the MOS config 1778 * since it's possible that the label does not 1779 * contain the most up-to-date information. 1780 */ 1781 vdev_load_log_state(tvd, mtvd); 1782 vdev_reopen(tvd); 1783 } 1784 } 1785 vdev_free(mrvd); 1786 spa_config_exit(spa, SCL_ALL, FTAG); 1787 1788 /* 1789 * Ensure we were able to validate the config. 1790 */ 1791 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1792} 1793 1794/* 1795 * Check for missing log devices 1796 */ 1797static boolean_t 1798spa_check_logs(spa_t *spa) 1799{ 1800 boolean_t rv = B_FALSE; 1801 dsl_pool_t *dp = spa_get_dsl(spa); 1802 1803 switch (spa->spa_log_state) { 1804 case SPA_LOG_MISSING: 1805 /* need to recheck in case slog has been restored */ 1806 case SPA_LOG_UNKNOWN: 1807 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 1808 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0); 1809 if (rv) 1810 spa_set_log_state(spa, SPA_LOG_MISSING); 1811 break; 1812 } 1813 return (rv); 1814} 1815 1816static boolean_t 1817spa_passivate_log(spa_t *spa) 1818{ 1819 vdev_t *rvd = spa->spa_root_vdev; 1820 boolean_t slog_found = B_FALSE; 1821 1822 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1823 1824 if (!spa_has_slogs(spa)) 1825 return (B_FALSE); 1826 1827 for (int c = 0; c < rvd->vdev_children; c++) { 1828 vdev_t *tvd = rvd->vdev_child[c]; 1829 metaslab_group_t *mg = tvd->vdev_mg; 1830 1831 if (tvd->vdev_islog) { 1832 metaslab_group_passivate(mg); 1833 slog_found = B_TRUE; 1834 } 1835 } 1836 1837 return (slog_found); 1838} 1839 1840static void 1841spa_activate_log(spa_t *spa) 1842{ 1843 vdev_t *rvd = spa->spa_root_vdev; 1844 1845 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1846 1847 for (int c = 0; c < rvd->vdev_children; c++) { 1848 vdev_t *tvd = rvd->vdev_child[c]; 1849 metaslab_group_t *mg = tvd->vdev_mg; 1850 1851 if (tvd->vdev_islog) 1852 metaslab_group_activate(mg); 1853 } 1854} 1855 1856int 1857spa_offline_log(spa_t *spa) 1858{ 1859 int error; 1860 1861 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1862 NULL, DS_FIND_CHILDREN); 1863 if (error == 0) { 1864 /* 1865 * We successfully offlined the log device, sync out the 1866 * current txg so that the "stubby" block can be removed 1867 * by zil_sync(). 1868 */ 1869 txg_wait_synced(spa->spa_dsl_pool, 0); 1870 } 1871 return (error); 1872} 1873 1874static void 1875spa_aux_check_removed(spa_aux_vdev_t *sav) 1876{ 1877 int i; 1878 1879 for (i = 0; i < sav->sav_count; i++) 1880 spa_check_removed(sav->sav_vdevs[i]); 1881} 1882 1883void 1884spa_claim_notify(zio_t *zio) 1885{ 1886 spa_t *spa = zio->io_spa; 1887 1888 if (zio->io_error) 1889 return; 1890 1891 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1892 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1893 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1894 mutex_exit(&spa->spa_props_lock); 1895} 1896 1897typedef struct spa_load_error { 1898 uint64_t sle_meta_count; 1899 uint64_t sle_data_count; 1900} spa_load_error_t; 1901 1902static void 1903spa_load_verify_done(zio_t *zio) 1904{ 1905 blkptr_t *bp = zio->io_bp; 1906 spa_load_error_t *sle = zio->io_private; 1907 dmu_object_type_t type = BP_GET_TYPE(bp); 1908 int error = zio->io_error; 1909 spa_t *spa = zio->io_spa; 1910 1911 if (error) { 1912 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1913 type != DMU_OT_INTENT_LOG) 1914 atomic_inc_64(&sle->sle_meta_count); 1915 else 1916 atomic_inc_64(&sle->sle_data_count); 1917 } 1918 zio_data_buf_free(zio->io_data, zio->io_size); 1919 1920 mutex_enter(&spa->spa_scrub_lock); 1921 spa->spa_scrub_inflight--; 1922 cv_broadcast(&spa->spa_scrub_io_cv); 1923 mutex_exit(&spa->spa_scrub_lock); 1924} 1925 1926/* 1927 * Maximum number of concurrent scrub i/os to create while verifying 1928 * a pool while importing it. 1929 */ 1930int spa_load_verify_maxinflight = 10000; 1931boolean_t spa_load_verify_metadata = B_TRUE; 1932boolean_t spa_load_verify_data = B_TRUE; 1933 1934SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1935 &spa_load_verify_maxinflight, 0, 1936 "Maximum number of concurrent scrub I/Os to create while verifying a " 1937 "pool while importing it"); 1938 1939SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1940 &spa_load_verify_metadata, 0, 1941 "Check metadata on import?"); 1942 1943SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1944 &spa_load_verify_data, 0, 1945 "Check user data on import?"); 1946 1947/*ARGSUSED*/ 1948static int 1949spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1950 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1951{ 1952 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1953 return (0); 1954 /* 1955 * Note: normally this routine will not be called if 1956 * spa_load_verify_metadata is not set. However, it may be useful 1957 * to manually set the flag after the traversal has begun. 1958 */ 1959 if (!spa_load_verify_metadata) 1960 return (0); 1961 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1962 return (0); 1963 1964 zio_t *rio = arg; 1965 size_t size = BP_GET_PSIZE(bp); 1966 void *data = zio_data_buf_alloc(size); 1967 1968 mutex_enter(&spa->spa_scrub_lock); 1969 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1970 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1971 spa->spa_scrub_inflight++; 1972 mutex_exit(&spa->spa_scrub_lock); 1973 1974 zio_nowait(zio_read(rio, spa, bp, data, size, 1975 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1976 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1977 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1978 return (0); 1979} 1980 1981/* ARGSUSED */ 1982int 1983verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) 1984{ 1985 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN) 1986 return (SET_ERROR(ENAMETOOLONG)); 1987 1988 return (0); 1989} 1990 1991static int 1992spa_load_verify(spa_t *spa) 1993{ 1994 zio_t *rio; 1995 spa_load_error_t sle = { 0 }; 1996 zpool_rewind_policy_t policy; 1997 boolean_t verify_ok = B_FALSE; 1998 int error = 0; 1999 2000 zpool_get_rewind_policy(spa->spa_config, &policy); 2001 2002 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 2003 return (0); 2004 2005 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG); 2006 error = dmu_objset_find_dp(spa->spa_dsl_pool, 2007 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL, 2008 DS_FIND_CHILDREN); 2009 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG); 2010 if (error != 0) 2011 return (error); 2012 2013 rio = zio_root(spa, NULL, &sle, 2014 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 2015 2016 if (spa_load_verify_metadata) { 2017 error = traverse_pool(spa, spa->spa_verify_min_txg, 2018 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 2019 spa_load_verify_cb, rio); 2020 } 2021 2022 (void) zio_wait(rio); 2023 2024 spa->spa_load_meta_errors = sle.sle_meta_count; 2025 spa->spa_load_data_errors = sle.sle_data_count; 2026 2027 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 2028 sle.sle_data_count <= policy.zrp_maxdata) { 2029 int64_t loss = 0; 2030 2031 verify_ok = B_TRUE; 2032 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2033 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2034 2035 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2036 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2037 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2038 VERIFY(nvlist_add_int64(spa->spa_load_info, 2039 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2040 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2041 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2042 } else { 2043 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2044 } 2045 2046 if (error) { 2047 if (error != ENXIO && error != EIO) 2048 error = SET_ERROR(EIO); 2049 return (error); 2050 } 2051 2052 return (verify_ok ? 0 : EIO); 2053} 2054 2055/* 2056 * Find a value in the pool props object. 2057 */ 2058static void 2059spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2060{ 2061 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2062 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2063} 2064 2065/* 2066 * Find a value in the pool directory object. 2067 */ 2068static int 2069spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2070{ 2071 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2072 name, sizeof (uint64_t), 1, val)); 2073} 2074 2075static int 2076spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2077{ 2078 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2079 return (err); 2080} 2081 2082/* 2083 * Fix up config after a partly-completed split. This is done with the 2084 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2085 * pool have that entry in their config, but only the splitting one contains 2086 * a list of all the guids of the vdevs that are being split off. 2087 * 2088 * This function determines what to do with that list: either rejoin 2089 * all the disks to the pool, or complete the splitting process. To attempt 2090 * the rejoin, each disk that is offlined is marked online again, and 2091 * we do a reopen() call. If the vdev label for every disk that was 2092 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2093 * then we call vdev_split() on each disk, and complete the split. 2094 * 2095 * Otherwise we leave the config alone, with all the vdevs in place in 2096 * the original pool. 2097 */ 2098static void 2099spa_try_repair(spa_t *spa, nvlist_t *config) 2100{ 2101 uint_t extracted; 2102 uint64_t *glist; 2103 uint_t i, gcount; 2104 nvlist_t *nvl; 2105 vdev_t **vd; 2106 boolean_t attempt_reopen; 2107 2108 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2109 return; 2110 2111 /* check that the config is complete */ 2112 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2113 &glist, &gcount) != 0) 2114 return; 2115 2116 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2117 2118 /* attempt to online all the vdevs & validate */ 2119 attempt_reopen = B_TRUE; 2120 for (i = 0; i < gcount; i++) { 2121 if (glist[i] == 0) /* vdev is hole */ 2122 continue; 2123 2124 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2125 if (vd[i] == NULL) { 2126 /* 2127 * Don't bother attempting to reopen the disks; 2128 * just do the split. 2129 */ 2130 attempt_reopen = B_FALSE; 2131 } else { 2132 /* attempt to re-online it */ 2133 vd[i]->vdev_offline = B_FALSE; 2134 } 2135 } 2136 2137 if (attempt_reopen) { 2138 vdev_reopen(spa->spa_root_vdev); 2139 2140 /* check each device to see what state it's in */ 2141 for (extracted = 0, i = 0; i < gcount; i++) { 2142 if (vd[i] != NULL && 2143 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2144 break; 2145 ++extracted; 2146 } 2147 } 2148 2149 /* 2150 * If every disk has been moved to the new pool, or if we never 2151 * even attempted to look at them, then we split them off for 2152 * good. 2153 */ 2154 if (!attempt_reopen || gcount == extracted) { 2155 for (i = 0; i < gcount; i++) 2156 if (vd[i] != NULL) 2157 vdev_split(vd[i]); 2158 vdev_reopen(spa->spa_root_vdev); 2159 } 2160 2161 kmem_free(vd, gcount * sizeof (vdev_t *)); 2162} 2163 2164static int 2165spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2166 boolean_t mosconfig) 2167{ 2168 nvlist_t *config = spa->spa_config; 2169 char *ereport = FM_EREPORT_ZFS_POOL; 2170 char *comment; 2171 int error; 2172 uint64_t pool_guid; 2173 nvlist_t *nvl; 2174 2175 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2176 return (SET_ERROR(EINVAL)); 2177 2178 ASSERT(spa->spa_comment == NULL); 2179 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2180 spa->spa_comment = spa_strdup(comment); 2181 2182 /* 2183 * Versioning wasn't explicitly added to the label until later, so if 2184 * it's not present treat it as the initial version. 2185 */ 2186 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2187 &spa->spa_ubsync.ub_version) != 0) 2188 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2189 2190 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2191 &spa->spa_config_txg); 2192 2193 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2194 spa_guid_exists(pool_guid, 0)) { 2195 error = SET_ERROR(EEXIST); 2196 } else { 2197 spa->spa_config_guid = pool_guid; 2198 2199 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2200 &nvl) == 0) { 2201 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2202 KM_SLEEP) == 0); 2203 } 2204 2205 nvlist_free(spa->spa_load_info); 2206 spa->spa_load_info = fnvlist_alloc(); 2207 2208 gethrestime(&spa->spa_loaded_ts); 2209 error = spa_load_impl(spa, pool_guid, config, state, type, 2210 mosconfig, &ereport); 2211 } 2212 2213 /* 2214 * Don't count references from objsets that are already closed 2215 * and are making their way through the eviction process. 2216 */ 2217 spa_evicting_os_wait(spa); 2218 spa->spa_minref = refcount_count(&spa->spa_refcount); 2219 if (error) { 2220 if (error != EEXIST) { 2221 spa->spa_loaded_ts.tv_sec = 0; 2222 spa->spa_loaded_ts.tv_nsec = 0; 2223 } 2224 if (error != EBADF) { 2225 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2226 } 2227 } 2228 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2229 spa->spa_ena = 0; 2230 2231 return (error); 2232} 2233 2234/* 2235 * Load an existing storage pool, using the pool's builtin spa_config as a 2236 * source of configuration information. 2237 */ 2238static int 2239spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2240 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2241 char **ereport) 2242{ 2243 int error = 0; 2244 nvlist_t *nvroot = NULL; 2245 nvlist_t *label; 2246 vdev_t *rvd; 2247 uberblock_t *ub = &spa->spa_uberblock; 2248 uint64_t children, config_cache_txg = spa->spa_config_txg; 2249 int orig_mode = spa->spa_mode; 2250 int parse; 2251 uint64_t obj; 2252 boolean_t missing_feat_write = B_FALSE; 2253 2254 /* 2255 * If this is an untrusted config, access the pool in read-only mode. 2256 * This prevents things like resilvering recently removed devices. 2257 */ 2258 if (!mosconfig) 2259 spa->spa_mode = FREAD; 2260 2261 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2262 2263 spa->spa_load_state = state; 2264 2265 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2266 return (SET_ERROR(EINVAL)); 2267 2268 parse = (type == SPA_IMPORT_EXISTING ? 2269 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2270 2271 /* 2272 * Create "The Godfather" zio to hold all async IOs 2273 */ 2274 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2275 KM_SLEEP); 2276 for (int i = 0; i < max_ncpus; i++) { 2277 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2278 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2279 ZIO_FLAG_GODFATHER); 2280 } 2281 2282 /* 2283 * Parse the configuration into a vdev tree. We explicitly set the 2284 * value that will be returned by spa_version() since parsing the 2285 * configuration requires knowing the version number. 2286 */ 2287 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2288 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2289 spa_config_exit(spa, SCL_ALL, FTAG); 2290 2291 if (error != 0) 2292 return (error); 2293 2294 ASSERT(spa->spa_root_vdev == rvd); 2295 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT); 2296 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT); 2297 2298 if (type != SPA_IMPORT_ASSEMBLE) { 2299 ASSERT(spa_guid(spa) == pool_guid); 2300 } 2301 2302 /* 2303 * Try to open all vdevs, loading each label in the process. 2304 */ 2305 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2306 error = vdev_open(rvd); 2307 spa_config_exit(spa, SCL_ALL, FTAG); 2308 if (error != 0) 2309 return (error); 2310 2311 /* 2312 * We need to validate the vdev labels against the configuration that 2313 * we have in hand, which is dependent on the setting of mosconfig. If 2314 * mosconfig is true then we're validating the vdev labels based on 2315 * that config. Otherwise, we're validating against the cached config 2316 * (zpool.cache) that was read when we loaded the zfs module, and then 2317 * later we will recursively call spa_load() and validate against 2318 * the vdev config. 2319 * 2320 * If we're assembling a new pool that's been split off from an 2321 * existing pool, the labels haven't yet been updated so we skip 2322 * validation for now. 2323 */ 2324 if (type != SPA_IMPORT_ASSEMBLE) { 2325 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2326 error = vdev_validate(rvd, mosconfig); 2327 spa_config_exit(spa, SCL_ALL, FTAG); 2328 2329 if (error != 0) 2330 return (error); 2331 2332 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2333 return (SET_ERROR(ENXIO)); 2334 } 2335 2336 /* 2337 * Find the best uberblock. 2338 */ 2339 vdev_uberblock_load(rvd, ub, &label); 2340 2341 /* 2342 * If we weren't able to find a single valid uberblock, return failure. 2343 */ 2344 if (ub->ub_txg == 0) { 2345 nvlist_free(label); 2346 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2347 } 2348 2349 /* 2350 * If the pool has an unsupported version we can't open it. 2351 */ 2352 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2353 nvlist_free(label); 2354 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2355 } 2356 2357 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2358 nvlist_t *features; 2359 2360 /* 2361 * If we weren't able to find what's necessary for reading the 2362 * MOS in the label, return failure. 2363 */ 2364 if (label == NULL || nvlist_lookup_nvlist(label, 2365 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2366 nvlist_free(label); 2367 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2368 ENXIO)); 2369 } 2370 2371 /* 2372 * Update our in-core representation with the definitive values 2373 * from the label. 2374 */ 2375 nvlist_free(spa->spa_label_features); 2376 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2377 } 2378 2379 nvlist_free(label); 2380 2381 /* 2382 * Look through entries in the label nvlist's features_for_read. If 2383 * there is a feature listed there which we don't understand then we 2384 * cannot open a pool. 2385 */ 2386 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2387 nvlist_t *unsup_feat; 2388 2389 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2390 0); 2391 2392 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2393 NULL); nvp != NULL; 2394 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2395 if (!zfeature_is_supported(nvpair_name(nvp))) { 2396 VERIFY(nvlist_add_string(unsup_feat, 2397 nvpair_name(nvp), "") == 0); 2398 } 2399 } 2400 2401 if (!nvlist_empty(unsup_feat)) { 2402 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2403 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2404 nvlist_free(unsup_feat); 2405 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2406 ENOTSUP)); 2407 } 2408 2409 nvlist_free(unsup_feat); 2410 } 2411 2412 /* 2413 * If the vdev guid sum doesn't match the uberblock, we have an 2414 * incomplete configuration. We first check to see if the pool 2415 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2416 * If it is, defer the vdev_guid_sum check till later so we 2417 * can handle missing vdevs. 2418 */ 2419 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2420 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2421 rvd->vdev_guid_sum != ub->ub_guid_sum) 2422 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2423 2424 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2425 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2426 spa_try_repair(spa, config); 2427 spa_config_exit(spa, SCL_ALL, FTAG); 2428 nvlist_free(spa->spa_config_splitting); 2429 spa->spa_config_splitting = NULL; 2430 } 2431 2432 /* 2433 * Initialize internal SPA structures. 2434 */ 2435 spa->spa_state = POOL_STATE_ACTIVE; 2436 spa->spa_ubsync = spa->spa_uberblock; 2437 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2438 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2439 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2440 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2441 spa->spa_claim_max_txg = spa->spa_first_txg; 2442 spa->spa_prev_software_version = ub->ub_software_version; 2443 2444 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2445 if (error) 2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2447 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2448 2449 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2450 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2451 2452 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2453 boolean_t missing_feat_read = B_FALSE; 2454 nvlist_t *unsup_feat, *enabled_feat; 2455 2456 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2457 &spa->spa_feat_for_read_obj) != 0) { 2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2459 } 2460 2461 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2462 &spa->spa_feat_for_write_obj) != 0) { 2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2464 } 2465 2466 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2467 &spa->spa_feat_desc_obj) != 0) { 2468 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2469 } 2470 2471 enabled_feat = fnvlist_alloc(); 2472 unsup_feat = fnvlist_alloc(); 2473 2474 if (!spa_features_check(spa, B_FALSE, 2475 unsup_feat, enabled_feat)) 2476 missing_feat_read = B_TRUE; 2477 2478 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2479 if (!spa_features_check(spa, B_TRUE, 2480 unsup_feat, enabled_feat)) { 2481 missing_feat_write = B_TRUE; 2482 } 2483 } 2484 2485 fnvlist_add_nvlist(spa->spa_load_info, 2486 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2487 2488 if (!nvlist_empty(unsup_feat)) { 2489 fnvlist_add_nvlist(spa->spa_load_info, 2490 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2491 } 2492 2493 fnvlist_free(enabled_feat); 2494 fnvlist_free(unsup_feat); 2495 2496 if (!missing_feat_read) { 2497 fnvlist_add_boolean(spa->spa_load_info, 2498 ZPOOL_CONFIG_CAN_RDONLY); 2499 } 2500 2501 /* 2502 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2503 * twofold: to determine whether the pool is available for 2504 * import in read-write mode and (if it is not) whether the 2505 * pool is available for import in read-only mode. If the pool 2506 * is available for import in read-write mode, it is displayed 2507 * as available in userland; if it is not available for import 2508 * in read-only mode, it is displayed as unavailable in 2509 * userland. If the pool is available for import in read-only 2510 * mode but not read-write mode, it is displayed as unavailable 2511 * in userland with a special note that the pool is actually 2512 * available for open in read-only mode. 2513 * 2514 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2515 * missing a feature for write, we must first determine whether 2516 * the pool can be opened read-only before returning to 2517 * userland in order to know whether to display the 2518 * abovementioned note. 2519 */ 2520 if (missing_feat_read || (missing_feat_write && 2521 spa_writeable(spa))) { 2522 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2523 ENOTSUP)); 2524 } 2525 2526 /* 2527 * Load refcounts for ZFS features from disk into an in-memory 2528 * cache during SPA initialization. 2529 */ 2530 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2531 uint64_t refcount; 2532 2533 error = feature_get_refcount_from_disk(spa, 2534 &spa_feature_table[i], &refcount); 2535 if (error == 0) { 2536 spa->spa_feat_refcount_cache[i] = refcount; 2537 } else if (error == ENOTSUP) { 2538 spa->spa_feat_refcount_cache[i] = 2539 SPA_FEATURE_DISABLED; 2540 } else { 2541 return (spa_vdev_err(rvd, 2542 VDEV_AUX_CORRUPT_DATA, EIO)); 2543 } 2544 } 2545 } 2546 2547 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2548 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2549 &spa->spa_feat_enabled_txg_obj) != 0) 2550 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2551 } 2552 2553 spa->spa_is_initializing = B_TRUE; 2554 error = dsl_pool_open(spa->spa_dsl_pool); 2555 spa->spa_is_initializing = B_FALSE; 2556 if (error != 0) 2557 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2558 2559 if (!mosconfig) { 2560 uint64_t hostid; 2561 nvlist_t *policy = NULL, *nvconfig; 2562 2563 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2564 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2565 2566 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2567 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2568 char *hostname; 2569 unsigned long myhostid = 0; 2570 2571 VERIFY(nvlist_lookup_string(nvconfig, 2572 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2573 2574#ifdef _KERNEL 2575 myhostid = zone_get_hostid(NULL); 2576#else /* _KERNEL */ 2577 /* 2578 * We're emulating the system's hostid in userland, so 2579 * we can't use zone_get_hostid(). 2580 */ 2581 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2582#endif /* _KERNEL */ 2583 if (check_hostid && hostid != 0 && myhostid != 0 && 2584 hostid != myhostid) { 2585 nvlist_free(nvconfig); 2586 cmn_err(CE_WARN, "pool '%s' could not be " 2587 "loaded as it was last accessed by " 2588 "another system (host: %s hostid: 0x%lx). " 2589 "See: http://illumos.org/msg/ZFS-8000-EY", 2590 spa_name(spa), hostname, 2591 (unsigned long)hostid); 2592 return (SET_ERROR(EBADF)); 2593 } 2594 } 2595 if (nvlist_lookup_nvlist(spa->spa_config, 2596 ZPOOL_REWIND_POLICY, &policy) == 0) 2597 VERIFY(nvlist_add_nvlist(nvconfig, 2598 ZPOOL_REWIND_POLICY, policy) == 0); 2599 2600 spa_config_set(spa, nvconfig); 2601 spa_unload(spa); 2602 spa_deactivate(spa); 2603 spa_activate(spa, orig_mode); 2604 2605 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2606 } 2607 2608 /* Grab the secret checksum salt from the MOS. */ 2609 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2610 DMU_POOL_CHECKSUM_SALT, 1, 2611 sizeof (spa->spa_cksum_salt.zcs_bytes), 2612 spa->spa_cksum_salt.zcs_bytes); 2613 if (error == ENOENT) { 2614 /* Generate a new salt for subsequent use */ 2615 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 2616 sizeof (spa->spa_cksum_salt.zcs_bytes)); 2617 } else if (error != 0) { 2618 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2619 } 2620 2621 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2623 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2624 if (error != 0) 2625 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2626 2627 /* 2628 * Load the bit that tells us to use the new accounting function 2629 * (raid-z deflation). If we have an older pool, this will not 2630 * be present. 2631 */ 2632 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2633 if (error != 0 && error != ENOENT) 2634 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2635 2636 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2637 &spa->spa_creation_version); 2638 if (error != 0 && error != ENOENT) 2639 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2640 2641 /* 2642 * Load the persistent error log. If we have an older pool, this will 2643 * not be present. 2644 */ 2645 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2646 if (error != 0 && error != ENOENT) 2647 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2648 2649 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2650 &spa->spa_errlog_scrub); 2651 if (error != 0 && error != ENOENT) 2652 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2653 2654 /* 2655 * Load the history object. If we have an older pool, this 2656 * will not be present. 2657 */ 2658 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2659 if (error != 0 && error != ENOENT) 2660 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2661 2662 /* 2663 * If we're assembling the pool from the split-off vdevs of 2664 * an existing pool, we don't want to attach the spares & cache 2665 * devices. 2666 */ 2667 2668 /* 2669 * Load any hot spares for this pool. 2670 */ 2671 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2672 if (error != 0 && error != ENOENT) 2673 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2674 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2675 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2676 if (load_nvlist(spa, spa->spa_spares.sav_object, 2677 &spa->spa_spares.sav_config) != 0) 2678 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2679 2680 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2681 spa_load_spares(spa); 2682 spa_config_exit(spa, SCL_ALL, FTAG); 2683 } else if (error == 0) { 2684 spa->spa_spares.sav_sync = B_TRUE; 2685 } 2686 2687 /* 2688 * Load any level 2 ARC devices for this pool. 2689 */ 2690 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2691 &spa->spa_l2cache.sav_object); 2692 if (error != 0 && error != ENOENT) 2693 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2694 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2695 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2696 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2697 &spa->spa_l2cache.sav_config) != 0) 2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2699 2700 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2701 spa_load_l2cache(spa); 2702 spa_config_exit(spa, SCL_ALL, FTAG); 2703 } else if (error == 0) { 2704 spa->spa_l2cache.sav_sync = B_TRUE; 2705 } 2706 2707 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2708 2709 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2710 if (error && error != ENOENT) 2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2712 2713 if (error == 0) { 2714 uint64_t autoreplace; 2715 2716 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2717 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2718 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2719 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2720 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2721 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2722 &spa->spa_dedup_ditto); 2723 2724 spa->spa_autoreplace = (autoreplace != 0); 2725 } 2726 2727 /* 2728 * If the 'autoreplace' property is set, then post a resource notifying 2729 * the ZFS DE that it should not issue any faults for unopenable 2730 * devices. We also iterate over the vdevs, and post a sysevent for any 2731 * unopenable vdevs so that the normal autoreplace handler can take 2732 * over. 2733 */ 2734 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2735 spa_check_removed(spa->spa_root_vdev); 2736 /* 2737 * For the import case, this is done in spa_import(), because 2738 * at this point we're using the spare definitions from 2739 * the MOS config, not necessarily from the userland config. 2740 */ 2741 if (state != SPA_LOAD_IMPORT) { 2742 spa_aux_check_removed(&spa->spa_spares); 2743 spa_aux_check_removed(&spa->spa_l2cache); 2744 } 2745 } 2746 2747 /* 2748 * Load the vdev state for all toplevel vdevs. 2749 */ 2750 vdev_load(rvd); 2751 2752 /* 2753 * Propagate the leaf DTLs we just loaded all the way up the tree. 2754 */ 2755 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2756 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2757 spa_config_exit(spa, SCL_ALL, FTAG); 2758 2759 /* 2760 * Load the DDTs (dedup tables). 2761 */ 2762 error = ddt_load(spa); 2763 if (error != 0) 2764 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2765 2766 spa_update_dspace(spa); 2767 2768 /* 2769 * Validate the config, using the MOS config to fill in any 2770 * information which might be missing. If we fail to validate 2771 * the config then declare the pool unfit for use. If we're 2772 * assembling a pool from a split, the log is not transferred 2773 * over. 2774 */ 2775 if (type != SPA_IMPORT_ASSEMBLE) { 2776 nvlist_t *nvconfig; 2777 2778 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2779 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2780 2781 if (!spa_config_valid(spa, nvconfig)) { 2782 nvlist_free(nvconfig); 2783 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2784 ENXIO)); 2785 } 2786 nvlist_free(nvconfig); 2787 2788 /* 2789 * Now that we've validated the config, check the state of the 2790 * root vdev. If it can't be opened, it indicates one or 2791 * more toplevel vdevs are faulted. 2792 */ 2793 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2794 return (SET_ERROR(ENXIO)); 2795 2796 if (spa_writeable(spa) && spa_check_logs(spa)) { 2797 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2798 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2799 } 2800 } 2801 2802 if (missing_feat_write) { 2803 ASSERT(state == SPA_LOAD_TRYIMPORT); 2804 2805 /* 2806 * At this point, we know that we can open the pool in 2807 * read-only mode but not read-write mode. We now have enough 2808 * information and can return to userland. 2809 */ 2810 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2811 } 2812 2813 /* 2814 * We've successfully opened the pool, verify that we're ready 2815 * to start pushing transactions. 2816 */ 2817 if (state != SPA_LOAD_TRYIMPORT) { 2818 if (error = spa_load_verify(spa)) 2819 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2820 error)); 2821 } 2822 2823 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2824 spa->spa_load_max_txg == UINT64_MAX)) { 2825 dmu_tx_t *tx; 2826 int need_update = B_FALSE; 2827 dsl_pool_t *dp = spa_get_dsl(spa); 2828 2829 ASSERT(state != SPA_LOAD_TRYIMPORT); 2830 2831 /* 2832 * Claim log blocks that haven't been committed yet. 2833 * This must all happen in a single txg. 2834 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2835 * invoked from zil_claim_log_block()'s i/o done callback. 2836 * Price of rollback is that we abandon the log. 2837 */ 2838 spa->spa_claiming = B_TRUE; 2839 2840 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa)); 2841 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj, 2842 zil_claim, tx, DS_FIND_CHILDREN); 2843 dmu_tx_commit(tx); 2844 2845 spa->spa_claiming = B_FALSE; 2846 2847 spa_set_log_state(spa, SPA_LOG_GOOD); 2848 spa->spa_sync_on = B_TRUE; 2849 txg_sync_start(spa->spa_dsl_pool); 2850 2851 /* 2852 * Wait for all claims to sync. We sync up to the highest 2853 * claimed log block birth time so that claimed log blocks 2854 * don't appear to be from the future. spa_claim_max_txg 2855 * will have been set for us by either zil_check_log_chain() 2856 * (invoked from spa_check_logs()) or zil_claim() above. 2857 */ 2858 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2859 2860 /* 2861 * If the config cache is stale, or we have uninitialized 2862 * metaslabs (see spa_vdev_add()), then update the config. 2863 * 2864 * If this is a verbatim import, trust the current 2865 * in-core spa_config and update the disk labels. 2866 */ 2867 if (config_cache_txg != spa->spa_config_txg || 2868 state == SPA_LOAD_IMPORT || 2869 state == SPA_LOAD_RECOVER || 2870 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2871 need_update = B_TRUE; 2872 2873 for (int c = 0; c < rvd->vdev_children; c++) 2874 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2875 need_update = B_TRUE; 2876 2877 /* 2878 * Update the config cache asychronously in case we're the 2879 * root pool, in which case the config cache isn't writable yet. 2880 */ 2881 if (need_update) 2882 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2883 2884 /* 2885 * Check all DTLs to see if anything needs resilvering. 2886 */ 2887 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2888 vdev_resilver_needed(rvd, NULL, NULL)) 2889 spa_async_request(spa, SPA_ASYNC_RESILVER); 2890 2891 /* 2892 * Log the fact that we booted up (so that we can detect if 2893 * we rebooted in the middle of an operation). 2894 */ 2895 spa_history_log_version(spa, "open"); 2896 2897 /* 2898 * Delete any inconsistent datasets. 2899 */ 2900 (void) dmu_objset_find(spa_name(spa), 2901 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2902 2903 /* 2904 * Clean up any stale temporary dataset userrefs. 2905 */ 2906 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2907 } 2908 2909 return (0); 2910} 2911 2912static int 2913spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2914{ 2915 int mode = spa->spa_mode; 2916 2917 spa_unload(spa); 2918 spa_deactivate(spa); 2919 2920 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2921 2922 spa_activate(spa, mode); 2923 spa_async_suspend(spa); 2924 2925 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2926} 2927 2928/* 2929 * If spa_load() fails this function will try loading prior txg's. If 2930 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2931 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2932 * function will not rewind the pool and will return the same error as 2933 * spa_load(). 2934 */ 2935static int 2936spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2937 uint64_t max_request, int rewind_flags) 2938{ 2939 nvlist_t *loadinfo = NULL; 2940 nvlist_t *config = NULL; 2941 int load_error, rewind_error; 2942 uint64_t safe_rewind_txg; 2943 uint64_t min_txg; 2944 2945 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2946 spa->spa_load_max_txg = spa->spa_load_txg; 2947 spa_set_log_state(spa, SPA_LOG_CLEAR); 2948 } else { 2949 spa->spa_load_max_txg = max_request; 2950 if (max_request != UINT64_MAX) 2951 spa->spa_extreme_rewind = B_TRUE; 2952 } 2953 2954 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2955 mosconfig); 2956 if (load_error == 0) 2957 return (0); 2958 2959 if (spa->spa_root_vdev != NULL) 2960 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2961 2962 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2963 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2964 2965 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2966 nvlist_free(config); 2967 return (load_error); 2968 } 2969 2970 if (state == SPA_LOAD_RECOVER) { 2971 /* Price of rolling back is discarding txgs, including log */ 2972 spa_set_log_state(spa, SPA_LOG_CLEAR); 2973 } else { 2974 /* 2975 * If we aren't rolling back save the load info from our first 2976 * import attempt so that we can restore it after attempting 2977 * to rewind. 2978 */ 2979 loadinfo = spa->spa_load_info; 2980 spa->spa_load_info = fnvlist_alloc(); 2981 } 2982 2983 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2984 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2985 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2986 TXG_INITIAL : safe_rewind_txg; 2987 2988 /* 2989 * Continue as long as we're finding errors, we're still within 2990 * the acceptable rewind range, and we're still finding uberblocks 2991 */ 2992 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2993 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2994 if (spa->spa_load_max_txg < safe_rewind_txg) 2995 spa->spa_extreme_rewind = B_TRUE; 2996 rewind_error = spa_load_retry(spa, state, mosconfig); 2997 } 2998 2999 spa->spa_extreme_rewind = B_FALSE; 3000 spa->spa_load_max_txg = UINT64_MAX; 3001 3002 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 3003 spa_config_set(spa, config); 3004 3005 if (state == SPA_LOAD_RECOVER) { 3006 ASSERT3P(loadinfo, ==, NULL); 3007 return (rewind_error); 3008 } else { 3009 /* Store the rewind info as part of the initial load info */ 3010 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 3011 spa->spa_load_info); 3012 3013 /* Restore the initial load info */ 3014 fnvlist_free(spa->spa_load_info); 3015 spa->spa_load_info = loadinfo; 3016 3017 return (load_error); 3018 } 3019} 3020 3021/* 3022 * Pool Open/Import 3023 * 3024 * The import case is identical to an open except that the configuration is sent 3025 * down from userland, instead of grabbed from the configuration cache. For the 3026 * case of an open, the pool configuration will exist in the 3027 * POOL_STATE_UNINITIALIZED state. 3028 * 3029 * The stats information (gen/count/ustats) is used to gather vdev statistics at 3030 * the same time open the pool, without having to keep around the spa_t in some 3031 * ambiguous state. 3032 */ 3033static int 3034spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 3035 nvlist_t **config) 3036{ 3037 spa_t *spa; 3038 spa_load_state_t state = SPA_LOAD_OPEN; 3039 int error; 3040 int locked = B_FALSE; 3041 int firstopen = B_FALSE; 3042 3043 *spapp = NULL; 3044 3045 /* 3046 * As disgusting as this is, we need to support recursive calls to this 3047 * function because dsl_dir_open() is called during spa_load(), and ends 3048 * up calling spa_open() again. The real fix is to figure out how to 3049 * avoid dsl_dir_open() calling this in the first place. 3050 */ 3051 if (mutex_owner(&spa_namespace_lock) != curthread) { 3052 mutex_enter(&spa_namespace_lock); 3053 locked = B_TRUE; 3054 } 3055 3056 if ((spa = spa_lookup(pool)) == NULL) { 3057 if (locked) 3058 mutex_exit(&spa_namespace_lock); 3059 return (SET_ERROR(ENOENT)); 3060 } 3061 3062 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3063 zpool_rewind_policy_t policy; 3064 3065 firstopen = B_TRUE; 3066 3067 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3068 &policy); 3069 if (policy.zrp_request & ZPOOL_DO_REWIND) 3070 state = SPA_LOAD_RECOVER; 3071 3072 spa_activate(spa, spa_mode_global); 3073 3074 if (state != SPA_LOAD_RECOVER) 3075 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3076 3077 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3078 policy.zrp_request); 3079 3080 if (error == EBADF) { 3081 /* 3082 * If vdev_validate() returns failure (indicated by 3083 * EBADF), it indicates that one of the vdevs indicates 3084 * that the pool has been exported or destroyed. If 3085 * this is the case, the config cache is out of sync and 3086 * we should remove the pool from the namespace. 3087 */ 3088 spa_unload(spa); 3089 spa_deactivate(spa); 3090 spa_config_sync(spa, B_TRUE, B_TRUE); 3091 spa_remove(spa); 3092 if (locked) 3093 mutex_exit(&spa_namespace_lock); 3094 return (SET_ERROR(ENOENT)); 3095 } 3096 3097 if (error) { 3098 /* 3099 * We can't open the pool, but we still have useful 3100 * information: the state of each vdev after the 3101 * attempted vdev_open(). Return this to the user. 3102 */ 3103 if (config != NULL && spa->spa_config) { 3104 VERIFY(nvlist_dup(spa->spa_config, config, 3105 KM_SLEEP) == 0); 3106 VERIFY(nvlist_add_nvlist(*config, 3107 ZPOOL_CONFIG_LOAD_INFO, 3108 spa->spa_load_info) == 0); 3109 } 3110 spa_unload(spa); 3111 spa_deactivate(spa); 3112 spa->spa_last_open_failed = error; 3113 if (locked) 3114 mutex_exit(&spa_namespace_lock); 3115 *spapp = NULL; 3116 return (error); 3117 } 3118 } 3119 3120 spa_open_ref(spa, tag); 3121 3122 if (config != NULL) 3123 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3124 3125 /* 3126 * If we've recovered the pool, pass back any information we 3127 * gathered while doing the load. 3128 */ 3129 if (state == SPA_LOAD_RECOVER) { 3130 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3131 spa->spa_load_info) == 0); 3132 } 3133 3134 if (locked) { 3135 spa->spa_last_open_failed = 0; 3136 spa->spa_last_ubsync_txg = 0; 3137 spa->spa_load_txg = 0; 3138 mutex_exit(&spa_namespace_lock); 3139#ifdef __FreeBSD__ 3140#ifdef _KERNEL 3141 if (firstopen) 3142 zvol_create_minors(spa->spa_name); 3143#endif 3144#endif 3145 } 3146 3147 *spapp = spa; 3148 3149 return (0); 3150} 3151 3152int 3153spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3154 nvlist_t **config) 3155{ 3156 return (spa_open_common(name, spapp, tag, policy, config)); 3157} 3158 3159int 3160spa_open(const char *name, spa_t **spapp, void *tag) 3161{ 3162 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3163} 3164 3165/* 3166 * Lookup the given spa_t, incrementing the inject count in the process, 3167 * preventing it from being exported or destroyed. 3168 */ 3169spa_t * 3170spa_inject_addref(char *name) 3171{ 3172 spa_t *spa; 3173 3174 mutex_enter(&spa_namespace_lock); 3175 if ((spa = spa_lookup(name)) == NULL) { 3176 mutex_exit(&spa_namespace_lock); 3177 return (NULL); 3178 } 3179 spa->spa_inject_ref++; 3180 mutex_exit(&spa_namespace_lock); 3181 3182 return (spa); 3183} 3184 3185void 3186spa_inject_delref(spa_t *spa) 3187{ 3188 mutex_enter(&spa_namespace_lock); 3189 spa->spa_inject_ref--; 3190 mutex_exit(&spa_namespace_lock); 3191} 3192 3193/* 3194 * Add spares device information to the nvlist. 3195 */ 3196static void 3197spa_add_spares(spa_t *spa, nvlist_t *config) 3198{ 3199 nvlist_t **spares; 3200 uint_t i, nspares; 3201 nvlist_t *nvroot; 3202 uint64_t guid; 3203 vdev_stat_t *vs; 3204 uint_t vsc; 3205 uint64_t pool; 3206 3207 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3208 3209 if (spa->spa_spares.sav_count == 0) 3210 return; 3211 3212 VERIFY(nvlist_lookup_nvlist(config, 3213 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3214 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3215 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3216 if (nspares != 0) { 3217 VERIFY(nvlist_add_nvlist_array(nvroot, 3218 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3219 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3220 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3221 3222 /* 3223 * Go through and find any spares which have since been 3224 * repurposed as an active spare. If this is the case, update 3225 * their status appropriately. 3226 */ 3227 for (i = 0; i < nspares; i++) { 3228 VERIFY(nvlist_lookup_uint64(spares[i], 3229 ZPOOL_CONFIG_GUID, &guid) == 0); 3230 if (spa_spare_exists(guid, &pool, NULL) && 3231 pool != 0ULL) { 3232 VERIFY(nvlist_lookup_uint64_array( 3233 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3234 (uint64_t **)&vs, &vsc) == 0); 3235 vs->vs_state = VDEV_STATE_CANT_OPEN; 3236 vs->vs_aux = VDEV_AUX_SPARED; 3237 } 3238 } 3239 } 3240} 3241 3242/* 3243 * Add l2cache device information to the nvlist, including vdev stats. 3244 */ 3245static void 3246spa_add_l2cache(spa_t *spa, nvlist_t *config) 3247{ 3248 nvlist_t **l2cache; 3249 uint_t i, j, nl2cache; 3250 nvlist_t *nvroot; 3251 uint64_t guid; 3252 vdev_t *vd; 3253 vdev_stat_t *vs; 3254 uint_t vsc; 3255 3256 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3257 3258 if (spa->spa_l2cache.sav_count == 0) 3259 return; 3260 3261 VERIFY(nvlist_lookup_nvlist(config, 3262 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3263 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3264 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3265 if (nl2cache != 0) { 3266 VERIFY(nvlist_add_nvlist_array(nvroot, 3267 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3268 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3269 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3270 3271 /* 3272 * Update level 2 cache device stats. 3273 */ 3274 3275 for (i = 0; i < nl2cache; i++) { 3276 VERIFY(nvlist_lookup_uint64(l2cache[i], 3277 ZPOOL_CONFIG_GUID, &guid) == 0); 3278 3279 vd = NULL; 3280 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3281 if (guid == 3282 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3283 vd = spa->spa_l2cache.sav_vdevs[j]; 3284 break; 3285 } 3286 } 3287 ASSERT(vd != NULL); 3288 3289 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3290 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3291 == 0); 3292 vdev_get_stats(vd, vs); 3293 } 3294 } 3295} 3296 3297static void 3298spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3299{ 3300 nvlist_t *features; 3301 zap_cursor_t zc; 3302 zap_attribute_t za; 3303 3304 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3305 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3306 3307 /* We may be unable to read features if pool is suspended. */ 3308 if (spa_suspended(spa)) 3309 goto out; 3310 3311 if (spa->spa_feat_for_read_obj != 0) { 3312 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3313 spa->spa_feat_for_read_obj); 3314 zap_cursor_retrieve(&zc, &za) == 0; 3315 zap_cursor_advance(&zc)) { 3316 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3317 za.za_num_integers == 1); 3318 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3319 za.za_first_integer)); 3320 } 3321 zap_cursor_fini(&zc); 3322 } 3323 3324 if (spa->spa_feat_for_write_obj != 0) { 3325 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3326 spa->spa_feat_for_write_obj); 3327 zap_cursor_retrieve(&zc, &za) == 0; 3328 zap_cursor_advance(&zc)) { 3329 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3330 za.za_num_integers == 1); 3331 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3332 za.za_first_integer)); 3333 } 3334 zap_cursor_fini(&zc); 3335 } 3336 3337out: 3338 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3339 features) == 0); 3340 nvlist_free(features); 3341} 3342 3343int 3344spa_get_stats(const char *name, nvlist_t **config, 3345 char *altroot, size_t buflen) 3346{ 3347 int error; 3348 spa_t *spa; 3349 3350 *config = NULL; 3351 error = spa_open_common(name, &spa, FTAG, NULL, config); 3352 3353 if (spa != NULL) { 3354 /* 3355 * This still leaves a window of inconsistency where the spares 3356 * or l2cache devices could change and the config would be 3357 * self-inconsistent. 3358 */ 3359 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3360 3361 if (*config != NULL) { 3362 uint64_t loadtimes[2]; 3363 3364 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3365 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3366 VERIFY(nvlist_add_uint64_array(*config, 3367 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3368 3369 VERIFY(nvlist_add_uint64(*config, 3370 ZPOOL_CONFIG_ERRCOUNT, 3371 spa_get_errlog_size(spa)) == 0); 3372 3373 if (spa_suspended(spa)) 3374 VERIFY(nvlist_add_uint64(*config, 3375 ZPOOL_CONFIG_SUSPENDED, 3376 spa->spa_failmode) == 0); 3377 3378 spa_add_spares(spa, *config); 3379 spa_add_l2cache(spa, *config); 3380 spa_add_feature_stats(spa, *config); 3381 } 3382 } 3383 3384 /* 3385 * We want to get the alternate root even for faulted pools, so we cheat 3386 * and call spa_lookup() directly. 3387 */ 3388 if (altroot) { 3389 if (spa == NULL) { 3390 mutex_enter(&spa_namespace_lock); 3391 spa = spa_lookup(name); 3392 if (spa) 3393 spa_altroot(spa, altroot, buflen); 3394 else 3395 altroot[0] = '\0'; 3396 spa = NULL; 3397 mutex_exit(&spa_namespace_lock); 3398 } else { 3399 spa_altroot(spa, altroot, buflen); 3400 } 3401 } 3402 3403 if (spa != NULL) { 3404 spa_config_exit(spa, SCL_CONFIG, FTAG); 3405 spa_close(spa, FTAG); 3406 } 3407 3408 return (error); 3409} 3410 3411/* 3412 * Validate that the auxiliary device array is well formed. We must have an 3413 * array of nvlists, each which describes a valid leaf vdev. If this is an 3414 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3415 * specified, as long as they are well-formed. 3416 */ 3417static int 3418spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3419 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3420 vdev_labeltype_t label) 3421{ 3422 nvlist_t **dev; 3423 uint_t i, ndev; 3424 vdev_t *vd; 3425 int error; 3426 3427 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3428 3429 /* 3430 * It's acceptable to have no devs specified. 3431 */ 3432 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3433 return (0); 3434 3435 if (ndev == 0) 3436 return (SET_ERROR(EINVAL)); 3437 3438 /* 3439 * Make sure the pool is formatted with a version that supports this 3440 * device type. 3441 */ 3442 if (spa_version(spa) < version) 3443 return (SET_ERROR(ENOTSUP)); 3444 3445 /* 3446 * Set the pending device list so we correctly handle device in-use 3447 * checking. 3448 */ 3449 sav->sav_pending = dev; 3450 sav->sav_npending = ndev; 3451 3452 for (i = 0; i < ndev; i++) { 3453 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3454 mode)) != 0) 3455 goto out; 3456 3457 if (!vd->vdev_ops->vdev_op_leaf) { 3458 vdev_free(vd); 3459 error = SET_ERROR(EINVAL); 3460 goto out; 3461 } 3462 3463 /* 3464 * The L2ARC currently only supports disk devices in 3465 * kernel context. For user-level testing, we allow it. 3466 */ 3467#ifdef _KERNEL 3468 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3469 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3470 error = SET_ERROR(ENOTBLK); 3471 vdev_free(vd); 3472 goto out; 3473 } 3474#endif 3475 vd->vdev_top = vd; 3476 3477 if ((error = vdev_open(vd)) == 0 && 3478 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3479 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3480 vd->vdev_guid) == 0); 3481 } 3482 3483 vdev_free(vd); 3484 3485 if (error && 3486 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3487 goto out; 3488 else 3489 error = 0; 3490 } 3491 3492out: 3493 sav->sav_pending = NULL; 3494 sav->sav_npending = 0; 3495 return (error); 3496} 3497 3498static int 3499spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3500{ 3501 int error; 3502 3503 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3504 3505 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3506 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3507 VDEV_LABEL_SPARE)) != 0) { 3508 return (error); 3509 } 3510 3511 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3512 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3513 VDEV_LABEL_L2CACHE)); 3514} 3515 3516static void 3517spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3518 const char *config) 3519{ 3520 int i; 3521 3522 if (sav->sav_config != NULL) { 3523 nvlist_t **olddevs; 3524 uint_t oldndevs; 3525 nvlist_t **newdevs; 3526 3527 /* 3528 * Generate new dev list by concatentating with the 3529 * current dev list. 3530 */ 3531 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3532 &olddevs, &oldndevs) == 0); 3533 3534 newdevs = kmem_alloc(sizeof (void *) * 3535 (ndevs + oldndevs), KM_SLEEP); 3536 for (i = 0; i < oldndevs; i++) 3537 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3538 KM_SLEEP) == 0); 3539 for (i = 0; i < ndevs; i++) 3540 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3541 KM_SLEEP) == 0); 3542 3543 VERIFY(nvlist_remove(sav->sav_config, config, 3544 DATA_TYPE_NVLIST_ARRAY) == 0); 3545 3546 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3547 config, newdevs, ndevs + oldndevs) == 0); 3548 for (i = 0; i < oldndevs + ndevs; i++) 3549 nvlist_free(newdevs[i]); 3550 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3551 } else { 3552 /* 3553 * Generate a new dev list. 3554 */ 3555 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3556 KM_SLEEP) == 0); 3557 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3558 devs, ndevs) == 0); 3559 } 3560} 3561 3562/* 3563 * Stop and drop level 2 ARC devices 3564 */ 3565void 3566spa_l2cache_drop(spa_t *spa) 3567{ 3568 vdev_t *vd; 3569 int i; 3570 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3571 3572 for (i = 0; i < sav->sav_count; i++) { 3573 uint64_t pool; 3574 3575 vd = sav->sav_vdevs[i]; 3576 ASSERT(vd != NULL); 3577 3578 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3579 pool != 0ULL && l2arc_vdev_present(vd)) 3580 l2arc_remove_vdev(vd); 3581 } 3582} 3583 3584/* 3585 * Pool Creation 3586 */ 3587int 3588spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3589 nvlist_t *zplprops) 3590{ 3591 spa_t *spa; 3592 char *altroot = NULL; 3593 vdev_t *rvd; 3594 dsl_pool_t *dp; 3595 dmu_tx_t *tx; 3596 int error = 0; 3597 uint64_t txg = TXG_INITIAL; 3598 nvlist_t **spares, **l2cache; 3599 uint_t nspares, nl2cache; 3600 uint64_t version, obj; 3601 boolean_t has_features; 3602 char *poolname; 3603 nvlist_t *nvl; 3604 3605 if (nvlist_lookup_string(props, 3606 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0) 3607 poolname = (char *)pool; 3608 3609 /* 3610 * If this pool already exists, return failure. 3611 */ 3612 mutex_enter(&spa_namespace_lock); 3613 if (spa_lookup(poolname) != NULL) { 3614 mutex_exit(&spa_namespace_lock); 3615 return (SET_ERROR(EEXIST)); 3616 } 3617 3618 /* 3619 * Allocate a new spa_t structure. 3620 */ 3621 nvl = fnvlist_alloc(); 3622 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool); 3623 (void) nvlist_lookup_string(props, 3624 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3625 spa = spa_add(poolname, nvl, altroot); 3626 fnvlist_free(nvl); 3627 spa_activate(spa, spa_mode_global); 3628 3629 if (props && (error = spa_prop_validate(spa, props))) { 3630 spa_deactivate(spa); 3631 spa_remove(spa); 3632 mutex_exit(&spa_namespace_lock); 3633 return (error); 3634 } 3635 3636 /* 3637 * Temporary pool names should never be written to disk. 3638 */ 3639 if (poolname != pool) 3640 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME; 3641 3642 has_features = B_FALSE; 3643 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3644 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3645 if (zpool_prop_feature(nvpair_name(elem))) 3646 has_features = B_TRUE; 3647 } 3648 3649 if (has_features || nvlist_lookup_uint64(props, 3650 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3651 version = SPA_VERSION; 3652 } 3653 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3654 3655 spa->spa_first_txg = txg; 3656 spa->spa_uberblock.ub_txg = txg - 1; 3657 spa->spa_uberblock.ub_version = version; 3658 spa->spa_ubsync = spa->spa_uberblock; 3659 spa->spa_load_state = SPA_LOAD_CREATE; 3660 3661 /* 3662 * Create "The Godfather" zio to hold all async IOs 3663 */ 3664 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3665 KM_SLEEP); 3666 for (int i = 0; i < max_ncpus; i++) { 3667 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3668 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3669 ZIO_FLAG_GODFATHER); 3670 } 3671 3672 /* 3673 * Create the root vdev. 3674 */ 3675 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3676 3677 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3678 3679 ASSERT(error != 0 || rvd != NULL); 3680 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3681 3682 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3683 error = SET_ERROR(EINVAL); 3684 3685 if (error == 0 && 3686 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3687 (error = spa_validate_aux(spa, nvroot, txg, 3688 VDEV_ALLOC_ADD)) == 0) { 3689 for (int c = 0; c < rvd->vdev_children; c++) { 3690 vdev_ashift_optimize(rvd->vdev_child[c]); 3691 vdev_metaslab_set_size(rvd->vdev_child[c]); 3692 vdev_expand(rvd->vdev_child[c], txg); 3693 } 3694 } 3695 3696 spa_config_exit(spa, SCL_ALL, FTAG); 3697 3698 if (error != 0) { 3699 spa_unload(spa); 3700 spa_deactivate(spa); 3701 spa_remove(spa); 3702 mutex_exit(&spa_namespace_lock); 3703 return (error); 3704 } 3705 3706 /* 3707 * Get the list of spares, if specified. 3708 */ 3709 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3710 &spares, &nspares) == 0) { 3711 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3712 KM_SLEEP) == 0); 3713 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3714 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3715 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3716 spa_load_spares(spa); 3717 spa_config_exit(spa, SCL_ALL, FTAG); 3718 spa->spa_spares.sav_sync = B_TRUE; 3719 } 3720 3721 /* 3722 * Get the list of level 2 cache devices, if specified. 3723 */ 3724 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3725 &l2cache, &nl2cache) == 0) { 3726 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3727 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3728 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3729 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3731 spa_load_l2cache(spa); 3732 spa_config_exit(spa, SCL_ALL, FTAG); 3733 spa->spa_l2cache.sav_sync = B_TRUE; 3734 } 3735 3736 spa->spa_is_initializing = B_TRUE; 3737 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3738 spa->spa_meta_objset = dp->dp_meta_objset; 3739 spa->spa_is_initializing = B_FALSE; 3740 3741 /* 3742 * Create DDTs (dedup tables). 3743 */ 3744 ddt_create(spa); 3745 3746 spa_update_dspace(spa); 3747 3748 tx = dmu_tx_create_assigned(dp, txg); 3749 3750 /* 3751 * Create the pool config object. 3752 */ 3753 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3754 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3755 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3756 3757 if (zap_add(spa->spa_meta_objset, 3758 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3759 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3760 cmn_err(CE_PANIC, "failed to add pool config"); 3761 } 3762 3763 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3764 spa_feature_create_zap_objects(spa, tx); 3765 3766 if (zap_add(spa->spa_meta_objset, 3767 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3768 sizeof (uint64_t), 1, &version, tx) != 0) { 3769 cmn_err(CE_PANIC, "failed to add pool version"); 3770 } 3771 3772 /* Newly created pools with the right version are always deflated. */ 3773 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3774 spa->spa_deflate = TRUE; 3775 if (zap_add(spa->spa_meta_objset, 3776 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3777 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3778 cmn_err(CE_PANIC, "failed to add deflate"); 3779 } 3780 } 3781 3782 /* 3783 * Create the deferred-free bpobj. Turn off compression 3784 * because sync-to-convergence takes longer if the blocksize 3785 * keeps changing. 3786 */ 3787 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3788 dmu_object_set_compress(spa->spa_meta_objset, obj, 3789 ZIO_COMPRESS_OFF, tx); 3790 if (zap_add(spa->spa_meta_objset, 3791 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3792 sizeof (uint64_t), 1, &obj, tx) != 0) { 3793 cmn_err(CE_PANIC, "failed to add bpobj"); 3794 } 3795 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3796 spa->spa_meta_objset, obj)); 3797 3798 /* 3799 * Create the pool's history object. 3800 */ 3801 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3802 spa_history_create_obj(spa, tx); 3803 3804 /* 3805 * Generate some random noise for salted checksums to operate on. 3806 */ 3807 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes, 3808 sizeof (spa->spa_cksum_salt.zcs_bytes)); 3809 3810 /* 3811 * Set pool properties. 3812 */ 3813 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3814 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3815 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3816 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3817 3818 if (props != NULL) { 3819 spa_configfile_set(spa, props, B_FALSE); 3820 spa_sync_props(props, tx); 3821 } 3822 3823 dmu_tx_commit(tx); 3824 3825 spa->spa_sync_on = B_TRUE; 3826 txg_sync_start(spa->spa_dsl_pool); 3827 3828 /* 3829 * We explicitly wait for the first transaction to complete so that our 3830 * bean counters are appropriately updated. 3831 */ 3832 txg_wait_synced(spa->spa_dsl_pool, txg); 3833 3834 spa_config_sync(spa, B_FALSE, B_TRUE); 3835 spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE); 3836 3837 spa_history_log_version(spa, "create"); 3838 3839 /* 3840 * Don't count references from objsets that are already closed 3841 * and are making their way through the eviction process. 3842 */ 3843 spa_evicting_os_wait(spa); 3844 spa->spa_minref = refcount_count(&spa->spa_refcount); 3845 spa->spa_load_state = SPA_LOAD_NONE; 3846 3847 mutex_exit(&spa_namespace_lock); 3848 3849 return (0); 3850} 3851 3852#ifdef _KERNEL 3853#ifdef illumos 3854/* 3855 * Get the root pool information from the root disk, then import the root pool 3856 * during the system boot up time. 3857 */ 3858extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3859 3860static nvlist_t * 3861spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3862{ 3863 nvlist_t *config; 3864 nvlist_t *nvtop, *nvroot; 3865 uint64_t pgid; 3866 3867 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3868 return (NULL); 3869 3870 /* 3871 * Add this top-level vdev to the child array. 3872 */ 3873 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3874 &nvtop) == 0); 3875 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3876 &pgid) == 0); 3877 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3878 3879 /* 3880 * Put this pool's top-level vdevs into a root vdev. 3881 */ 3882 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3883 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3884 VDEV_TYPE_ROOT) == 0); 3885 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3886 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3887 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3888 &nvtop, 1) == 0); 3889 3890 /* 3891 * Replace the existing vdev_tree with the new root vdev in 3892 * this pool's configuration (remove the old, add the new). 3893 */ 3894 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3895 nvlist_free(nvroot); 3896 return (config); 3897} 3898 3899/* 3900 * Walk the vdev tree and see if we can find a device with "better" 3901 * configuration. A configuration is "better" if the label on that 3902 * device has a more recent txg. 3903 */ 3904static void 3905spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3906{ 3907 for (int c = 0; c < vd->vdev_children; c++) 3908 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3909 3910 if (vd->vdev_ops->vdev_op_leaf) { 3911 nvlist_t *label; 3912 uint64_t label_txg; 3913 3914 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3915 &label) != 0) 3916 return; 3917 3918 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3919 &label_txg) == 0); 3920 3921 /* 3922 * Do we have a better boot device? 3923 */ 3924 if (label_txg > *txg) { 3925 *txg = label_txg; 3926 *avd = vd; 3927 } 3928 nvlist_free(label); 3929 } 3930} 3931 3932/* 3933 * Import a root pool. 3934 * 3935 * For x86. devpath_list will consist of devid and/or physpath name of 3936 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3937 * The GRUB "findroot" command will return the vdev we should boot. 3938 * 3939 * For Sparc, devpath_list consists the physpath name of the booting device 3940 * no matter the rootpool is a single device pool or a mirrored pool. 3941 * e.g. 3942 * "/pci@1f,0/ide@d/disk@0,0:a" 3943 */ 3944int 3945spa_import_rootpool(char *devpath, char *devid) 3946{ 3947 spa_t *spa; 3948 vdev_t *rvd, *bvd, *avd = NULL; 3949 nvlist_t *config, *nvtop; 3950 uint64_t guid, txg; 3951 char *pname; 3952 int error; 3953 3954 /* 3955 * Read the label from the boot device and generate a configuration. 3956 */ 3957 config = spa_generate_rootconf(devpath, devid, &guid); 3958#if defined(_OBP) && defined(_KERNEL) 3959 if (config == NULL) { 3960 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3961 /* iscsi boot */ 3962 get_iscsi_bootpath_phy(devpath); 3963 config = spa_generate_rootconf(devpath, devid, &guid); 3964 } 3965 } 3966#endif 3967 if (config == NULL) { 3968 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3969 devpath); 3970 return (SET_ERROR(EIO)); 3971 } 3972 3973 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3974 &pname) == 0); 3975 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3976 3977 mutex_enter(&spa_namespace_lock); 3978 if ((spa = spa_lookup(pname)) != NULL) { 3979 /* 3980 * Remove the existing root pool from the namespace so that we 3981 * can replace it with the correct config we just read in. 3982 */ 3983 spa_remove(spa); 3984 } 3985 3986 spa = spa_add(pname, config, NULL); 3987 spa->spa_is_root = B_TRUE; 3988 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3989 3990 /* 3991 * Build up a vdev tree based on the boot device's label config. 3992 */ 3993 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3994 &nvtop) == 0); 3995 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3996 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3997 VDEV_ALLOC_ROOTPOOL); 3998 spa_config_exit(spa, SCL_ALL, FTAG); 3999 if (error) { 4000 mutex_exit(&spa_namespace_lock); 4001 nvlist_free(config); 4002 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4003 pname); 4004 return (error); 4005 } 4006 4007 /* 4008 * Get the boot vdev. 4009 */ 4010 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 4011 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 4012 (u_longlong_t)guid); 4013 error = SET_ERROR(ENOENT); 4014 goto out; 4015 } 4016 4017 /* 4018 * Determine if there is a better boot device. 4019 */ 4020 avd = bvd; 4021 spa_alt_rootvdev(rvd, &avd, &txg); 4022 if (avd != bvd) { 4023 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 4024 "try booting from '%s'", avd->vdev_path); 4025 error = SET_ERROR(EINVAL); 4026 goto out; 4027 } 4028 4029 /* 4030 * If the boot device is part of a spare vdev then ensure that 4031 * we're booting off the active spare. 4032 */ 4033 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 4034 !bvd->vdev_isspare) { 4035 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 4036 "try booting from '%s'", 4037 bvd->vdev_parent-> 4038 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 4039 error = SET_ERROR(EINVAL); 4040 goto out; 4041 } 4042 4043 error = 0; 4044out: 4045 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4046 vdev_free(rvd); 4047 spa_config_exit(spa, SCL_ALL, FTAG); 4048 mutex_exit(&spa_namespace_lock); 4049 4050 nvlist_free(config); 4051 return (error); 4052} 4053 4054#else /* !illumos */ 4055 4056extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 4057 uint64_t *count); 4058 4059static nvlist_t * 4060spa_generate_rootconf(const char *name) 4061{ 4062 nvlist_t **configs, **tops; 4063 nvlist_t *config; 4064 nvlist_t *best_cfg, *nvtop, *nvroot; 4065 uint64_t *holes; 4066 uint64_t best_txg; 4067 uint64_t nchildren; 4068 uint64_t pgid; 4069 uint64_t count; 4070 uint64_t i; 4071 uint_t nholes; 4072 4073 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 4074 return (NULL); 4075 4076 ASSERT3U(count, !=, 0); 4077 best_txg = 0; 4078 for (i = 0; i < count; i++) { 4079 uint64_t txg; 4080 4081 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4082 &txg) == 0); 4083 if (txg > best_txg) { 4084 best_txg = txg; 4085 best_cfg = configs[i]; 4086 } 4087 } 4088 4089 /* 4090 * Multi-vdev root pool configuration discovery is not supported yet. 4091 */ 4092 nchildren = 1; 4093 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4094 holes = NULL; 4095 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4096 &holes, &nholes); 4097 4098 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4099 for (i = 0; i < nchildren; i++) { 4100 if (i >= count) 4101 break; 4102 if (configs[i] == NULL) 4103 continue; 4104 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4105 &nvtop) == 0); 4106 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4107 } 4108 for (i = 0; holes != NULL && i < nholes; i++) { 4109 if (i >= nchildren) 4110 continue; 4111 if (tops[holes[i]] != NULL) 4112 continue; 4113 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4114 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4115 VDEV_TYPE_HOLE) == 0); 4116 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4117 holes[i]) == 0); 4118 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4119 0) == 0); 4120 } 4121 for (i = 0; i < nchildren; i++) { 4122 if (tops[i] != NULL) 4123 continue; 4124 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4125 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4126 VDEV_TYPE_MISSING) == 0); 4127 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4128 i) == 0); 4129 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4130 0) == 0); 4131 } 4132 4133 /* 4134 * Create pool config based on the best vdev config. 4135 */ 4136 nvlist_dup(best_cfg, &config, KM_SLEEP); 4137 4138 /* 4139 * Put this pool's top-level vdevs into a root vdev. 4140 */ 4141 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4142 &pgid) == 0); 4143 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4144 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4145 VDEV_TYPE_ROOT) == 0); 4146 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4147 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4148 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4149 tops, nchildren) == 0); 4150 4151 /* 4152 * Replace the existing vdev_tree with the new root vdev in 4153 * this pool's configuration (remove the old, add the new). 4154 */ 4155 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4156 4157 /* 4158 * Drop vdev config elements that should not be present at pool level. 4159 */ 4160 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4161 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4162 4163 for (i = 0; i < count; i++) 4164 nvlist_free(configs[i]); 4165 kmem_free(configs, count * sizeof(void *)); 4166 for (i = 0; i < nchildren; i++) 4167 nvlist_free(tops[i]); 4168 kmem_free(tops, nchildren * sizeof(void *)); 4169 nvlist_free(nvroot); 4170 return (config); 4171} 4172 4173int 4174spa_import_rootpool(const char *name) 4175{ 4176 spa_t *spa; 4177 vdev_t *rvd, *bvd, *avd = NULL; 4178 nvlist_t *config, *nvtop; 4179 uint64_t txg; 4180 char *pname; 4181 int error; 4182 4183 /* 4184 * Read the label from the boot device and generate a configuration. 4185 */ 4186 config = spa_generate_rootconf(name); 4187 4188 mutex_enter(&spa_namespace_lock); 4189 if (config != NULL) { 4190 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4191 &pname) == 0 && strcmp(name, pname) == 0); 4192 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4193 == 0); 4194 4195 if ((spa = spa_lookup(pname)) != NULL) { 4196 /* 4197 * The pool could already be imported, 4198 * e.g., after reboot -r. 4199 */ 4200 if (spa->spa_state == POOL_STATE_ACTIVE) { 4201 mutex_exit(&spa_namespace_lock); 4202 nvlist_free(config); 4203 return (0); 4204 } 4205 4206 /* 4207 * Remove the existing root pool from the namespace so 4208 * that we can replace it with the correct config 4209 * we just read in. 4210 */ 4211 spa_remove(spa); 4212 } 4213 spa = spa_add(pname, config, NULL); 4214 4215 /* 4216 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4217 * via spa_version(). 4218 */ 4219 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4220 &spa->spa_ubsync.ub_version) != 0) 4221 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4222 } else if ((spa = spa_lookup(name)) == NULL) { 4223 mutex_exit(&spa_namespace_lock); 4224 nvlist_free(config); 4225 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4226 name); 4227 return (EIO); 4228 } else { 4229 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4230 } 4231 spa->spa_is_root = B_TRUE; 4232 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4233 4234 /* 4235 * Build up a vdev tree based on the boot device's label config. 4236 */ 4237 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4238 &nvtop) == 0); 4239 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4240 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4241 VDEV_ALLOC_ROOTPOOL); 4242 spa_config_exit(spa, SCL_ALL, FTAG); 4243 if (error) { 4244 mutex_exit(&spa_namespace_lock); 4245 nvlist_free(config); 4246 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4247 pname); 4248 return (error); 4249 } 4250 4251 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4252 vdev_free(rvd); 4253 spa_config_exit(spa, SCL_ALL, FTAG); 4254 mutex_exit(&spa_namespace_lock); 4255 4256 nvlist_free(config); 4257 return (0); 4258} 4259 4260#endif /* illumos */ 4261#endif /* _KERNEL */ 4262 4263/* 4264 * Import a non-root pool into the system. 4265 */ 4266int 4267spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4268{ 4269 spa_t *spa; 4270 char *altroot = NULL; 4271 spa_load_state_t state = SPA_LOAD_IMPORT; 4272 zpool_rewind_policy_t policy; 4273 uint64_t mode = spa_mode_global; 4274 uint64_t readonly = B_FALSE; 4275 int error; 4276 nvlist_t *nvroot; 4277 nvlist_t **spares, **l2cache; 4278 uint_t nspares, nl2cache; 4279 4280 /* 4281 * If a pool with this name exists, return failure. 4282 */ 4283 mutex_enter(&spa_namespace_lock); 4284 if (spa_lookup(pool) != NULL) { 4285 mutex_exit(&spa_namespace_lock); 4286 return (SET_ERROR(EEXIST)); 4287 } 4288 4289 /* 4290 * Create and initialize the spa structure. 4291 */ 4292 (void) nvlist_lookup_string(props, 4293 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4294 (void) nvlist_lookup_uint64(props, 4295 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4296 if (readonly) 4297 mode = FREAD; 4298 spa = spa_add(pool, config, altroot); 4299 spa->spa_import_flags = flags; 4300 4301 /* 4302 * Verbatim import - Take a pool and insert it into the namespace 4303 * as if it had been loaded at boot. 4304 */ 4305 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4306 if (props != NULL) 4307 spa_configfile_set(spa, props, B_FALSE); 4308 4309 spa_config_sync(spa, B_FALSE, B_TRUE); 4310 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4311 4312 mutex_exit(&spa_namespace_lock); 4313 return (0); 4314 } 4315 4316 spa_activate(spa, mode); 4317 4318 /* 4319 * Don't start async tasks until we know everything is healthy. 4320 */ 4321 spa_async_suspend(spa); 4322 4323 zpool_get_rewind_policy(config, &policy); 4324 if (policy.zrp_request & ZPOOL_DO_REWIND) 4325 state = SPA_LOAD_RECOVER; 4326 4327 /* 4328 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4329 * because the user-supplied config is actually the one to trust when 4330 * doing an import. 4331 */ 4332 if (state != SPA_LOAD_RECOVER) 4333 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4334 4335 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4336 policy.zrp_request); 4337 4338 /* 4339 * Propagate anything learned while loading the pool and pass it 4340 * back to caller (i.e. rewind info, missing devices, etc). 4341 */ 4342 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4343 spa->spa_load_info) == 0); 4344 4345 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4346 /* 4347 * Toss any existing sparelist, as it doesn't have any validity 4348 * anymore, and conflicts with spa_has_spare(). 4349 */ 4350 if (spa->spa_spares.sav_config) { 4351 nvlist_free(spa->spa_spares.sav_config); 4352 spa->spa_spares.sav_config = NULL; 4353 spa_load_spares(spa); 4354 } 4355 if (spa->spa_l2cache.sav_config) { 4356 nvlist_free(spa->spa_l2cache.sav_config); 4357 spa->spa_l2cache.sav_config = NULL; 4358 spa_load_l2cache(spa); 4359 } 4360 4361 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4362 &nvroot) == 0); 4363 if (error == 0) 4364 error = spa_validate_aux(spa, nvroot, -1ULL, 4365 VDEV_ALLOC_SPARE); 4366 if (error == 0) 4367 error = spa_validate_aux(spa, nvroot, -1ULL, 4368 VDEV_ALLOC_L2CACHE); 4369 spa_config_exit(spa, SCL_ALL, FTAG); 4370 4371 if (props != NULL) 4372 spa_configfile_set(spa, props, B_FALSE); 4373 4374 if (error != 0 || (props && spa_writeable(spa) && 4375 (error = spa_prop_set(spa, props)))) { 4376 spa_unload(spa); 4377 spa_deactivate(spa); 4378 spa_remove(spa); 4379 mutex_exit(&spa_namespace_lock); 4380 return (error); 4381 } 4382 4383 spa_async_resume(spa); 4384 4385 /* 4386 * Override any spares and level 2 cache devices as specified by 4387 * the user, as these may have correct device names/devids, etc. 4388 */ 4389 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4390 &spares, &nspares) == 0) { 4391 if (spa->spa_spares.sav_config) 4392 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4393 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4394 else 4395 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4396 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4397 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4398 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4399 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4400 spa_load_spares(spa); 4401 spa_config_exit(spa, SCL_ALL, FTAG); 4402 spa->spa_spares.sav_sync = B_TRUE; 4403 } 4404 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4405 &l2cache, &nl2cache) == 0) { 4406 if (spa->spa_l2cache.sav_config) 4407 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4408 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4409 else 4410 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4411 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4412 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4413 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4414 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4415 spa_load_l2cache(spa); 4416 spa_config_exit(spa, SCL_ALL, FTAG); 4417 spa->spa_l2cache.sav_sync = B_TRUE; 4418 } 4419 4420 /* 4421 * Check for any removed devices. 4422 */ 4423 if (spa->spa_autoreplace) { 4424 spa_aux_check_removed(&spa->spa_spares); 4425 spa_aux_check_removed(&spa->spa_l2cache); 4426 } 4427 4428 if (spa_writeable(spa)) { 4429 /* 4430 * Update the config cache to include the newly-imported pool. 4431 */ 4432 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4433 } 4434 4435 /* 4436 * It's possible that the pool was expanded while it was exported. 4437 * We kick off an async task to handle this for us. 4438 */ 4439 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4440 4441 spa_history_log_version(spa, "import"); 4442 4443 spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT); 4444 4445 mutex_exit(&spa_namespace_lock); 4446 4447#ifdef __FreeBSD__ 4448#ifdef _KERNEL 4449 zvol_create_minors(pool); 4450#endif 4451#endif 4452 return (0); 4453} 4454 4455nvlist_t * 4456spa_tryimport(nvlist_t *tryconfig) 4457{ 4458 nvlist_t *config = NULL; 4459 char *poolname; 4460 spa_t *spa; 4461 uint64_t state; 4462 int error; 4463 4464 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4465 return (NULL); 4466 4467 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4468 return (NULL); 4469 4470 /* 4471 * Create and initialize the spa structure. 4472 */ 4473 mutex_enter(&spa_namespace_lock); 4474 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4475 spa_activate(spa, FREAD); 4476 4477 /* 4478 * Pass off the heavy lifting to spa_load(). 4479 * Pass TRUE for mosconfig because the user-supplied config 4480 * is actually the one to trust when doing an import. 4481 */ 4482 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4483 4484 /* 4485 * If 'tryconfig' was at least parsable, return the current config. 4486 */ 4487 if (spa->spa_root_vdev != NULL) { 4488 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4489 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4490 poolname) == 0); 4491 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4492 state) == 0); 4493 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4494 spa->spa_uberblock.ub_timestamp) == 0); 4495 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4496 spa->spa_load_info) == 0); 4497 4498 /* 4499 * If the bootfs property exists on this pool then we 4500 * copy it out so that external consumers can tell which 4501 * pools are bootable. 4502 */ 4503 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4504 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4505 4506 /* 4507 * We have to play games with the name since the 4508 * pool was opened as TRYIMPORT_NAME. 4509 */ 4510 if (dsl_dsobj_to_dsname(spa_name(spa), 4511 spa->spa_bootfs, tmpname) == 0) { 4512 char *cp; 4513 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4514 4515 cp = strchr(tmpname, '/'); 4516 if (cp == NULL) { 4517 (void) strlcpy(dsname, tmpname, 4518 MAXPATHLEN); 4519 } else { 4520 (void) snprintf(dsname, MAXPATHLEN, 4521 "%s/%s", poolname, ++cp); 4522 } 4523 VERIFY(nvlist_add_string(config, 4524 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4525 kmem_free(dsname, MAXPATHLEN); 4526 } 4527 kmem_free(tmpname, MAXPATHLEN); 4528 } 4529 4530 /* 4531 * Add the list of hot spares and level 2 cache devices. 4532 */ 4533 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4534 spa_add_spares(spa, config); 4535 spa_add_l2cache(spa, config); 4536 spa_config_exit(spa, SCL_CONFIG, FTAG); 4537 } 4538 4539 spa_unload(spa); 4540 spa_deactivate(spa); 4541 spa_remove(spa); 4542 mutex_exit(&spa_namespace_lock); 4543 4544 return (config); 4545} 4546 4547/* 4548 * Pool export/destroy 4549 * 4550 * The act of destroying or exporting a pool is very simple. We make sure there 4551 * is no more pending I/O and any references to the pool are gone. Then, we 4552 * update the pool state and sync all the labels to disk, removing the 4553 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4554 * we don't sync the labels or remove the configuration cache. 4555 */ 4556static int 4557spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4558 boolean_t force, boolean_t hardforce) 4559{ 4560 spa_t *spa; 4561 4562 if (oldconfig) 4563 *oldconfig = NULL; 4564 4565 if (!(spa_mode_global & FWRITE)) 4566 return (SET_ERROR(EROFS)); 4567 4568 mutex_enter(&spa_namespace_lock); 4569 if ((spa = spa_lookup(pool)) == NULL) { 4570 mutex_exit(&spa_namespace_lock); 4571 return (SET_ERROR(ENOENT)); 4572 } 4573 4574 /* 4575 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4576 * reacquire the namespace lock, and see if we can export. 4577 */ 4578 spa_open_ref(spa, FTAG); 4579 mutex_exit(&spa_namespace_lock); 4580 spa_async_suspend(spa); 4581 mutex_enter(&spa_namespace_lock); 4582 spa_close(spa, FTAG); 4583 4584 /* 4585 * The pool will be in core if it's openable, 4586 * in which case we can modify its state. 4587 */ 4588 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4589 /* 4590 * Objsets may be open only because they're dirty, so we 4591 * have to force it to sync before checking spa_refcnt. 4592 */ 4593 txg_wait_synced(spa->spa_dsl_pool, 0); 4594 spa_evicting_os_wait(spa); 4595 4596 /* 4597 * A pool cannot be exported or destroyed if there are active 4598 * references. If we are resetting a pool, allow references by 4599 * fault injection handlers. 4600 */ 4601 if (!spa_refcount_zero(spa) || 4602 (spa->spa_inject_ref != 0 && 4603 new_state != POOL_STATE_UNINITIALIZED)) { 4604 spa_async_resume(spa); 4605 mutex_exit(&spa_namespace_lock); 4606 return (SET_ERROR(EBUSY)); 4607 } 4608 4609 /* 4610 * A pool cannot be exported if it has an active shared spare. 4611 * This is to prevent other pools stealing the active spare 4612 * from an exported pool. At user's own will, such pool can 4613 * be forcedly exported. 4614 */ 4615 if (!force && new_state == POOL_STATE_EXPORTED && 4616 spa_has_active_shared_spare(spa)) { 4617 spa_async_resume(spa); 4618 mutex_exit(&spa_namespace_lock); 4619 return (SET_ERROR(EXDEV)); 4620 } 4621 4622 /* 4623 * We want this to be reflected on every label, 4624 * so mark them all dirty. spa_unload() will do the 4625 * final sync that pushes these changes out. 4626 */ 4627 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4628 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4629 spa->spa_state = new_state; 4630 spa->spa_final_txg = spa_last_synced_txg(spa) + 4631 TXG_DEFER_SIZE + 1; 4632 vdev_config_dirty(spa->spa_root_vdev); 4633 spa_config_exit(spa, SCL_ALL, FTAG); 4634 } 4635 } 4636 4637 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4638 4639 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4640 spa_unload(spa); 4641 spa_deactivate(spa); 4642 } 4643 4644 if (oldconfig && spa->spa_config) 4645 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4646 4647 if (new_state != POOL_STATE_UNINITIALIZED) { 4648 if (!hardforce) 4649 spa_config_sync(spa, B_TRUE, B_TRUE); 4650 spa_remove(spa); 4651 } 4652 mutex_exit(&spa_namespace_lock); 4653 4654 return (0); 4655} 4656 4657/* 4658 * Destroy a storage pool. 4659 */ 4660int 4661spa_destroy(char *pool) 4662{ 4663 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4664 B_FALSE, B_FALSE)); 4665} 4666 4667/* 4668 * Export a storage pool. 4669 */ 4670int 4671spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4672 boolean_t hardforce) 4673{ 4674 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4675 force, hardforce)); 4676} 4677 4678/* 4679 * Similar to spa_export(), this unloads the spa_t without actually removing it 4680 * from the namespace in any way. 4681 */ 4682int 4683spa_reset(char *pool) 4684{ 4685 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4686 B_FALSE, B_FALSE)); 4687} 4688 4689/* 4690 * ========================================================================== 4691 * Device manipulation 4692 * ========================================================================== 4693 */ 4694 4695/* 4696 * Add a device to a storage pool. 4697 */ 4698int 4699spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4700{ 4701 uint64_t txg, id; 4702 int error; 4703 vdev_t *rvd = spa->spa_root_vdev; 4704 vdev_t *vd, *tvd; 4705 nvlist_t **spares, **l2cache; 4706 uint_t nspares, nl2cache; 4707 4708 ASSERT(spa_writeable(spa)); 4709 4710 txg = spa_vdev_enter(spa); 4711 4712 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4713 VDEV_ALLOC_ADD)) != 0) 4714 return (spa_vdev_exit(spa, NULL, txg, error)); 4715 4716 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4717 4718 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4719 &nspares) != 0) 4720 nspares = 0; 4721 4722 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4723 &nl2cache) != 0) 4724 nl2cache = 0; 4725 4726 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4727 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4728 4729 if (vd->vdev_children != 0 && 4730 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4731 return (spa_vdev_exit(spa, vd, txg, error)); 4732 4733 /* 4734 * We must validate the spares and l2cache devices after checking the 4735 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4736 */ 4737 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4738 return (spa_vdev_exit(spa, vd, txg, error)); 4739 4740 /* 4741 * Transfer each new top-level vdev from vd to rvd. 4742 */ 4743 for (int c = 0; c < vd->vdev_children; c++) { 4744 4745 /* 4746 * Set the vdev id to the first hole, if one exists. 4747 */ 4748 for (id = 0; id < rvd->vdev_children; id++) { 4749 if (rvd->vdev_child[id]->vdev_ishole) { 4750 vdev_free(rvd->vdev_child[id]); 4751 break; 4752 } 4753 } 4754 tvd = vd->vdev_child[c]; 4755 vdev_remove_child(vd, tvd); 4756 tvd->vdev_id = id; 4757 vdev_add_child(rvd, tvd); 4758 vdev_config_dirty(tvd); 4759 } 4760 4761 if (nspares != 0) { 4762 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4763 ZPOOL_CONFIG_SPARES); 4764 spa_load_spares(spa); 4765 spa->spa_spares.sav_sync = B_TRUE; 4766 } 4767 4768 if (nl2cache != 0) { 4769 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4770 ZPOOL_CONFIG_L2CACHE); 4771 spa_load_l2cache(spa); 4772 spa->spa_l2cache.sav_sync = B_TRUE; 4773 } 4774 4775 /* 4776 * We have to be careful when adding new vdevs to an existing pool. 4777 * If other threads start allocating from these vdevs before we 4778 * sync the config cache, and we lose power, then upon reboot we may 4779 * fail to open the pool because there are DVAs that the config cache 4780 * can't translate. Therefore, we first add the vdevs without 4781 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4782 * and then let spa_config_update() initialize the new metaslabs. 4783 * 4784 * spa_load() checks for added-but-not-initialized vdevs, so that 4785 * if we lose power at any point in this sequence, the remaining 4786 * steps will be completed the next time we load the pool. 4787 */ 4788 (void) spa_vdev_exit(spa, vd, txg, 0); 4789 4790 mutex_enter(&spa_namespace_lock); 4791 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4792 spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD); 4793 mutex_exit(&spa_namespace_lock); 4794 4795 return (0); 4796} 4797 4798/* 4799 * Attach a device to a mirror. The arguments are the path to any device 4800 * in the mirror, and the nvroot for the new device. If the path specifies 4801 * a device that is not mirrored, we automatically insert the mirror vdev. 4802 * 4803 * If 'replacing' is specified, the new device is intended to replace the 4804 * existing device; in this case the two devices are made into their own 4805 * mirror using the 'replacing' vdev, which is functionally identical to 4806 * the mirror vdev (it actually reuses all the same ops) but has a few 4807 * extra rules: you can't attach to it after it's been created, and upon 4808 * completion of resilvering, the first disk (the one being replaced) 4809 * is automatically detached. 4810 */ 4811int 4812spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4813{ 4814 uint64_t txg, dtl_max_txg; 4815 vdev_t *rvd = spa->spa_root_vdev; 4816 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4817 vdev_ops_t *pvops; 4818 char *oldvdpath, *newvdpath; 4819 int newvd_isspare; 4820 int error; 4821 4822 ASSERT(spa_writeable(spa)); 4823 4824 txg = spa_vdev_enter(spa); 4825 4826 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4827 4828 if (oldvd == NULL) 4829 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4830 4831 if (!oldvd->vdev_ops->vdev_op_leaf) 4832 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4833 4834 pvd = oldvd->vdev_parent; 4835 4836 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4837 VDEV_ALLOC_ATTACH)) != 0) 4838 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4839 4840 if (newrootvd->vdev_children != 1) 4841 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4842 4843 newvd = newrootvd->vdev_child[0]; 4844 4845 if (!newvd->vdev_ops->vdev_op_leaf) 4846 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4847 4848 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4849 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4850 4851 /* 4852 * Spares can't replace logs 4853 */ 4854 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4855 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4856 4857 if (!replacing) { 4858 /* 4859 * For attach, the only allowable parent is a mirror or the root 4860 * vdev. 4861 */ 4862 if (pvd->vdev_ops != &vdev_mirror_ops && 4863 pvd->vdev_ops != &vdev_root_ops) 4864 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4865 4866 pvops = &vdev_mirror_ops; 4867 } else { 4868 /* 4869 * Active hot spares can only be replaced by inactive hot 4870 * spares. 4871 */ 4872 if (pvd->vdev_ops == &vdev_spare_ops && 4873 oldvd->vdev_isspare && 4874 !spa_has_spare(spa, newvd->vdev_guid)) 4875 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4876 4877 /* 4878 * If the source is a hot spare, and the parent isn't already a 4879 * spare, then we want to create a new hot spare. Otherwise, we 4880 * want to create a replacing vdev. The user is not allowed to 4881 * attach to a spared vdev child unless the 'isspare' state is 4882 * the same (spare replaces spare, non-spare replaces 4883 * non-spare). 4884 */ 4885 if (pvd->vdev_ops == &vdev_replacing_ops && 4886 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4887 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4888 } else if (pvd->vdev_ops == &vdev_spare_ops && 4889 newvd->vdev_isspare != oldvd->vdev_isspare) { 4890 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4891 } 4892 4893 if (newvd->vdev_isspare) 4894 pvops = &vdev_spare_ops; 4895 else 4896 pvops = &vdev_replacing_ops; 4897 } 4898 4899 /* 4900 * Make sure the new device is big enough. 4901 */ 4902 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4903 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4904 4905 /* 4906 * The new device cannot have a higher alignment requirement 4907 * than the top-level vdev. 4908 */ 4909 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4910 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4911 4912 /* 4913 * If this is an in-place replacement, update oldvd's path and devid 4914 * to make it distinguishable from newvd, and unopenable from now on. 4915 */ 4916 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4917 spa_strfree(oldvd->vdev_path); 4918 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4919 KM_SLEEP); 4920 (void) sprintf(oldvd->vdev_path, "%s/%s", 4921 newvd->vdev_path, "old"); 4922 if (oldvd->vdev_devid != NULL) { 4923 spa_strfree(oldvd->vdev_devid); 4924 oldvd->vdev_devid = NULL; 4925 } 4926 } 4927 4928 /* mark the device being resilvered */ 4929 newvd->vdev_resilver_txg = txg; 4930 4931 /* 4932 * If the parent is not a mirror, or if we're replacing, insert the new 4933 * mirror/replacing/spare vdev above oldvd. 4934 */ 4935 if (pvd->vdev_ops != pvops) 4936 pvd = vdev_add_parent(oldvd, pvops); 4937 4938 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4939 ASSERT(pvd->vdev_ops == pvops); 4940 ASSERT(oldvd->vdev_parent == pvd); 4941 4942 /* 4943 * Extract the new device from its root and add it to pvd. 4944 */ 4945 vdev_remove_child(newrootvd, newvd); 4946 newvd->vdev_id = pvd->vdev_children; 4947 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4948 vdev_add_child(pvd, newvd); 4949 4950 tvd = newvd->vdev_top; 4951 ASSERT(pvd->vdev_top == tvd); 4952 ASSERT(tvd->vdev_parent == rvd); 4953 4954 vdev_config_dirty(tvd); 4955 4956 /* 4957 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4958 * for any dmu_sync-ed blocks. It will propagate upward when 4959 * spa_vdev_exit() calls vdev_dtl_reassess(). 4960 */ 4961 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4962 4963 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4964 dtl_max_txg - TXG_INITIAL); 4965 4966 if (newvd->vdev_isspare) { 4967 spa_spare_activate(newvd); 4968 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4969 } 4970 4971 oldvdpath = spa_strdup(oldvd->vdev_path); 4972 newvdpath = spa_strdup(newvd->vdev_path); 4973 newvd_isspare = newvd->vdev_isspare; 4974 4975 /* 4976 * Mark newvd's DTL dirty in this txg. 4977 */ 4978 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4979 4980 /* 4981 * Schedule the resilver to restart in the future. We do this to 4982 * ensure that dmu_sync-ed blocks have been stitched into the 4983 * respective datasets. 4984 */ 4985 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4986 4987 if (spa->spa_bootfs) 4988 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4989 4990 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH); 4991 4992 /* 4993 * Commit the config 4994 */ 4995 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4996 4997 spa_history_log_internal(spa, "vdev attach", NULL, 4998 "%s vdev=%s %s vdev=%s", 4999 replacing && newvd_isspare ? "spare in" : 5000 replacing ? "replace" : "attach", newvdpath, 5001 replacing ? "for" : "to", oldvdpath); 5002 5003 spa_strfree(oldvdpath); 5004 spa_strfree(newvdpath); 5005 5006 return (0); 5007} 5008 5009/* 5010 * Detach a device from a mirror or replacing vdev. 5011 * 5012 * If 'replace_done' is specified, only detach if the parent 5013 * is a replacing vdev. 5014 */ 5015int 5016spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 5017{ 5018 uint64_t txg; 5019 int error; 5020 vdev_t *rvd = spa->spa_root_vdev; 5021 vdev_t *vd, *pvd, *cvd, *tvd; 5022 boolean_t unspare = B_FALSE; 5023 uint64_t unspare_guid = 0; 5024 char *vdpath; 5025 5026 ASSERT(spa_writeable(spa)); 5027 5028 txg = spa_vdev_enter(spa); 5029 5030 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5031 5032 if (vd == NULL) 5033 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 5034 5035 if (!vd->vdev_ops->vdev_op_leaf) 5036 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5037 5038 pvd = vd->vdev_parent; 5039 5040 /* 5041 * If the parent/child relationship is not as expected, don't do it. 5042 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 5043 * vdev that's replacing B with C. The user's intent in replacing 5044 * is to go from M(A,B) to M(A,C). If the user decides to cancel 5045 * the replace by detaching C, the expected behavior is to end up 5046 * M(A,B). But suppose that right after deciding to detach C, 5047 * the replacement of B completes. We would have M(A,C), and then 5048 * ask to detach C, which would leave us with just A -- not what 5049 * the user wanted. To prevent this, we make sure that the 5050 * parent/child relationship hasn't changed -- in this example, 5051 * that C's parent is still the replacing vdev R. 5052 */ 5053 if (pvd->vdev_guid != pguid && pguid != 0) 5054 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5055 5056 /* 5057 * Only 'replacing' or 'spare' vdevs can be replaced. 5058 */ 5059 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 5060 pvd->vdev_ops != &vdev_spare_ops) 5061 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5062 5063 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 5064 spa_version(spa) >= SPA_VERSION_SPARES); 5065 5066 /* 5067 * Only mirror, replacing, and spare vdevs support detach. 5068 */ 5069 if (pvd->vdev_ops != &vdev_replacing_ops && 5070 pvd->vdev_ops != &vdev_mirror_ops && 5071 pvd->vdev_ops != &vdev_spare_ops) 5072 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 5073 5074 /* 5075 * If this device has the only valid copy of some data, 5076 * we cannot safely detach it. 5077 */ 5078 if (vdev_dtl_required(vd)) 5079 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 5080 5081 ASSERT(pvd->vdev_children >= 2); 5082 5083 /* 5084 * If we are detaching the second disk from a replacing vdev, then 5085 * check to see if we changed the original vdev's path to have "/old" 5086 * at the end in spa_vdev_attach(). If so, undo that change now. 5087 */ 5088 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 5089 vd->vdev_path != NULL) { 5090 size_t len = strlen(vd->vdev_path); 5091 5092 for (int c = 0; c < pvd->vdev_children; c++) { 5093 cvd = pvd->vdev_child[c]; 5094 5095 if (cvd == vd || cvd->vdev_path == NULL) 5096 continue; 5097 5098 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 5099 strcmp(cvd->vdev_path + len, "/old") == 0) { 5100 spa_strfree(cvd->vdev_path); 5101 cvd->vdev_path = spa_strdup(vd->vdev_path); 5102 break; 5103 } 5104 } 5105 } 5106 5107 /* 5108 * If we are detaching the original disk from a spare, then it implies 5109 * that the spare should become a real disk, and be removed from the 5110 * active spare list for the pool. 5111 */ 5112 if (pvd->vdev_ops == &vdev_spare_ops && 5113 vd->vdev_id == 0 && 5114 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5115 unspare = B_TRUE; 5116 5117 /* 5118 * Erase the disk labels so the disk can be used for other things. 5119 * This must be done after all other error cases are handled, 5120 * but before we disembowel vd (so we can still do I/O to it). 5121 * But if we can't do it, don't treat the error as fatal -- 5122 * it may be that the unwritability of the disk is the reason 5123 * it's being detached! 5124 */ 5125 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5126 5127 /* 5128 * Remove vd from its parent and compact the parent's children. 5129 */ 5130 vdev_remove_child(pvd, vd); 5131 vdev_compact_children(pvd); 5132 5133 /* 5134 * Remember one of the remaining children so we can get tvd below. 5135 */ 5136 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5137 5138 /* 5139 * If we need to remove the remaining child from the list of hot spares, 5140 * do it now, marking the vdev as no longer a spare in the process. 5141 * We must do this before vdev_remove_parent(), because that can 5142 * change the GUID if it creates a new toplevel GUID. For a similar 5143 * reason, we must remove the spare now, in the same txg as the detach; 5144 * otherwise someone could attach a new sibling, change the GUID, and 5145 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5146 */ 5147 if (unspare) { 5148 ASSERT(cvd->vdev_isspare); 5149 spa_spare_remove(cvd); 5150 unspare_guid = cvd->vdev_guid; 5151 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5152 cvd->vdev_unspare = B_TRUE; 5153 } 5154 5155 /* 5156 * If the parent mirror/replacing vdev only has one child, 5157 * the parent is no longer needed. Remove it from the tree. 5158 */ 5159 if (pvd->vdev_children == 1) { 5160 if (pvd->vdev_ops == &vdev_spare_ops) 5161 cvd->vdev_unspare = B_FALSE; 5162 vdev_remove_parent(cvd); 5163 } 5164 5165 5166 /* 5167 * We don't set tvd until now because the parent we just removed 5168 * may have been the previous top-level vdev. 5169 */ 5170 tvd = cvd->vdev_top; 5171 ASSERT(tvd->vdev_parent == rvd); 5172 5173 /* 5174 * Reevaluate the parent vdev state. 5175 */ 5176 vdev_propagate_state(cvd); 5177 5178 /* 5179 * If the 'autoexpand' property is set on the pool then automatically 5180 * try to expand the size of the pool. For example if the device we 5181 * just detached was smaller than the others, it may be possible to 5182 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5183 * first so that we can obtain the updated sizes of the leaf vdevs. 5184 */ 5185 if (spa->spa_autoexpand) { 5186 vdev_reopen(tvd); 5187 vdev_expand(tvd, txg); 5188 } 5189 5190 vdev_config_dirty(tvd); 5191 5192 /* 5193 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5194 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5195 * But first make sure we're not on any *other* txg's DTL list, to 5196 * prevent vd from being accessed after it's freed. 5197 */ 5198 vdpath = spa_strdup(vd->vdev_path); 5199 for (int t = 0; t < TXG_SIZE; t++) 5200 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5201 vd->vdev_detached = B_TRUE; 5202 vdev_dirty(tvd, VDD_DTL, vd, txg); 5203 5204 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5205 5206 /* hang on to the spa before we release the lock */ 5207 spa_open_ref(spa, FTAG); 5208 5209 error = spa_vdev_exit(spa, vd, txg, 0); 5210 5211 spa_history_log_internal(spa, "detach", NULL, 5212 "vdev=%s", vdpath); 5213 spa_strfree(vdpath); 5214 5215 /* 5216 * If this was the removal of the original device in a hot spare vdev, 5217 * then we want to go through and remove the device from the hot spare 5218 * list of every other pool. 5219 */ 5220 if (unspare) { 5221 spa_t *altspa = NULL; 5222 5223 mutex_enter(&spa_namespace_lock); 5224 while ((altspa = spa_next(altspa)) != NULL) { 5225 if (altspa->spa_state != POOL_STATE_ACTIVE || 5226 altspa == spa) 5227 continue; 5228 5229 spa_open_ref(altspa, FTAG); 5230 mutex_exit(&spa_namespace_lock); 5231 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5232 mutex_enter(&spa_namespace_lock); 5233 spa_close(altspa, FTAG); 5234 } 5235 mutex_exit(&spa_namespace_lock); 5236 5237 /* search the rest of the vdevs for spares to remove */ 5238 spa_vdev_resilver_done(spa); 5239 } 5240 5241 /* all done with the spa; OK to release */ 5242 mutex_enter(&spa_namespace_lock); 5243 spa_close(spa, FTAG); 5244 mutex_exit(&spa_namespace_lock); 5245 5246 return (error); 5247} 5248 5249/* 5250 * Split a set of devices from their mirrors, and create a new pool from them. 5251 */ 5252int 5253spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5254 nvlist_t *props, boolean_t exp) 5255{ 5256 int error = 0; 5257 uint64_t txg, *glist; 5258 spa_t *newspa; 5259 uint_t c, children, lastlog; 5260 nvlist_t **child, *nvl, *tmp; 5261 dmu_tx_t *tx; 5262 char *altroot = NULL; 5263 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5264 boolean_t activate_slog; 5265 5266 ASSERT(spa_writeable(spa)); 5267 5268 txg = spa_vdev_enter(spa); 5269 5270 /* clear the log and flush everything up to now */ 5271 activate_slog = spa_passivate_log(spa); 5272 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5273 error = spa_offline_log(spa); 5274 txg = spa_vdev_config_enter(spa); 5275 5276 if (activate_slog) 5277 spa_activate_log(spa); 5278 5279 if (error != 0) 5280 return (spa_vdev_exit(spa, NULL, txg, error)); 5281 5282 /* check new spa name before going any further */ 5283 if (spa_lookup(newname) != NULL) 5284 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5285 5286 /* 5287 * scan through all the children to ensure they're all mirrors 5288 */ 5289 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5290 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5291 &children) != 0) 5292 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5293 5294 /* first, check to ensure we've got the right child count */ 5295 rvd = spa->spa_root_vdev; 5296 lastlog = 0; 5297 for (c = 0; c < rvd->vdev_children; c++) { 5298 vdev_t *vd = rvd->vdev_child[c]; 5299 5300 /* don't count the holes & logs as children */ 5301 if (vd->vdev_islog || vd->vdev_ishole) { 5302 if (lastlog == 0) 5303 lastlog = c; 5304 continue; 5305 } 5306 5307 lastlog = 0; 5308 } 5309 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5310 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5311 5312 /* next, ensure no spare or cache devices are part of the split */ 5313 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5314 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5315 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5316 5317 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5318 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5319 5320 /* then, loop over each vdev and validate it */ 5321 for (c = 0; c < children; c++) { 5322 uint64_t is_hole = 0; 5323 5324 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5325 &is_hole); 5326 5327 if (is_hole != 0) { 5328 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5329 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5330 continue; 5331 } else { 5332 error = SET_ERROR(EINVAL); 5333 break; 5334 } 5335 } 5336 5337 /* which disk is going to be split? */ 5338 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5339 &glist[c]) != 0) { 5340 error = SET_ERROR(EINVAL); 5341 break; 5342 } 5343 5344 /* look it up in the spa */ 5345 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5346 if (vml[c] == NULL) { 5347 error = SET_ERROR(ENODEV); 5348 break; 5349 } 5350 5351 /* make sure there's nothing stopping the split */ 5352 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5353 vml[c]->vdev_islog || 5354 vml[c]->vdev_ishole || 5355 vml[c]->vdev_isspare || 5356 vml[c]->vdev_isl2cache || 5357 !vdev_writeable(vml[c]) || 5358 vml[c]->vdev_children != 0 || 5359 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5360 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5361 error = SET_ERROR(EINVAL); 5362 break; 5363 } 5364 5365 if (vdev_dtl_required(vml[c])) { 5366 error = SET_ERROR(EBUSY); 5367 break; 5368 } 5369 5370 /* we need certain info from the top level */ 5371 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5372 vml[c]->vdev_top->vdev_ms_array) == 0); 5373 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5374 vml[c]->vdev_top->vdev_ms_shift) == 0); 5375 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5376 vml[c]->vdev_top->vdev_asize) == 0); 5377 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5378 vml[c]->vdev_top->vdev_ashift) == 0); 5379 } 5380 5381 if (error != 0) { 5382 kmem_free(vml, children * sizeof (vdev_t *)); 5383 kmem_free(glist, children * sizeof (uint64_t)); 5384 return (spa_vdev_exit(spa, NULL, txg, error)); 5385 } 5386 5387 /* stop writers from using the disks */ 5388 for (c = 0; c < children; c++) { 5389 if (vml[c] != NULL) 5390 vml[c]->vdev_offline = B_TRUE; 5391 } 5392 vdev_reopen(spa->spa_root_vdev); 5393 5394 /* 5395 * Temporarily record the splitting vdevs in the spa config. This 5396 * will disappear once the config is regenerated. 5397 */ 5398 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5399 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5400 glist, children) == 0); 5401 kmem_free(glist, children * sizeof (uint64_t)); 5402 5403 mutex_enter(&spa->spa_props_lock); 5404 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5405 nvl) == 0); 5406 mutex_exit(&spa->spa_props_lock); 5407 spa->spa_config_splitting = nvl; 5408 vdev_config_dirty(spa->spa_root_vdev); 5409 5410 /* configure and create the new pool */ 5411 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5413 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5415 spa_version(spa)) == 0); 5416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5417 spa->spa_config_txg) == 0); 5418 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5419 spa_generate_guid(NULL)) == 0); 5420 (void) nvlist_lookup_string(props, 5421 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5422 5423 /* add the new pool to the namespace */ 5424 newspa = spa_add(newname, config, altroot); 5425 newspa->spa_config_txg = spa->spa_config_txg; 5426 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5427 5428 /* release the spa config lock, retaining the namespace lock */ 5429 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5430 5431 if (zio_injection_enabled) 5432 zio_handle_panic_injection(spa, FTAG, 1); 5433 5434 spa_activate(newspa, spa_mode_global); 5435 spa_async_suspend(newspa); 5436 5437#ifndef illumos 5438 /* mark that we are creating new spa by splitting */ 5439 newspa->spa_splitting_newspa = B_TRUE; 5440#endif 5441 /* create the new pool from the disks of the original pool */ 5442 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5443#ifndef illumos 5444 newspa->spa_splitting_newspa = B_FALSE; 5445#endif 5446 if (error) 5447 goto out; 5448 5449 /* if that worked, generate a real config for the new pool */ 5450 if (newspa->spa_root_vdev != NULL) { 5451 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5452 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5453 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5454 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5455 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5456 B_TRUE)); 5457 } 5458 5459 /* set the props */ 5460 if (props != NULL) { 5461 spa_configfile_set(newspa, props, B_FALSE); 5462 error = spa_prop_set(newspa, props); 5463 if (error) 5464 goto out; 5465 } 5466 5467 /* flush everything */ 5468 txg = spa_vdev_config_enter(newspa); 5469 vdev_config_dirty(newspa->spa_root_vdev); 5470 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5471 5472 if (zio_injection_enabled) 5473 zio_handle_panic_injection(spa, FTAG, 2); 5474 5475 spa_async_resume(newspa); 5476 5477 /* finally, update the original pool's config */ 5478 txg = spa_vdev_config_enter(spa); 5479 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5480 error = dmu_tx_assign(tx, TXG_WAIT); 5481 if (error != 0) 5482 dmu_tx_abort(tx); 5483 for (c = 0; c < children; c++) { 5484 if (vml[c] != NULL) { 5485 vdev_split(vml[c]); 5486 if (error == 0) 5487 spa_history_log_internal(spa, "detach", tx, 5488 "vdev=%s", vml[c]->vdev_path); 5489 vdev_free(vml[c]); 5490 } 5491 } 5492 vdev_config_dirty(spa->spa_root_vdev); 5493 spa->spa_config_splitting = NULL; 5494 nvlist_free(nvl); 5495 if (error == 0) 5496 dmu_tx_commit(tx); 5497 (void) spa_vdev_exit(spa, NULL, txg, 0); 5498 5499 if (zio_injection_enabled) 5500 zio_handle_panic_injection(spa, FTAG, 3); 5501 5502 /* split is complete; log a history record */ 5503 spa_history_log_internal(newspa, "split", NULL, 5504 "from pool %s", spa_name(spa)); 5505 5506 kmem_free(vml, children * sizeof (vdev_t *)); 5507 5508 /* if we're not going to mount the filesystems in userland, export */ 5509 if (exp) 5510 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5511 B_FALSE, B_FALSE); 5512 5513 return (error); 5514 5515out: 5516 spa_unload(newspa); 5517 spa_deactivate(newspa); 5518 spa_remove(newspa); 5519 5520 txg = spa_vdev_config_enter(spa); 5521 5522 /* re-online all offlined disks */ 5523 for (c = 0; c < children; c++) { 5524 if (vml[c] != NULL) 5525 vml[c]->vdev_offline = B_FALSE; 5526 } 5527 vdev_reopen(spa->spa_root_vdev); 5528 5529 nvlist_free(spa->spa_config_splitting); 5530 spa->spa_config_splitting = NULL; 5531 (void) spa_vdev_exit(spa, NULL, txg, error); 5532 5533 kmem_free(vml, children * sizeof (vdev_t *)); 5534 return (error); 5535} 5536 5537static nvlist_t * 5538spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5539{ 5540 for (int i = 0; i < count; i++) { 5541 uint64_t guid; 5542 5543 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5544 &guid) == 0); 5545 5546 if (guid == target_guid) 5547 return (nvpp[i]); 5548 } 5549 5550 return (NULL); 5551} 5552 5553static void 5554spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5555 nvlist_t *dev_to_remove) 5556{ 5557 nvlist_t **newdev = NULL; 5558 5559 if (count > 1) 5560 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5561 5562 for (int i = 0, j = 0; i < count; i++) { 5563 if (dev[i] == dev_to_remove) 5564 continue; 5565 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5566 } 5567 5568 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5569 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5570 5571 for (int i = 0; i < count - 1; i++) 5572 nvlist_free(newdev[i]); 5573 5574 if (count > 1) 5575 kmem_free(newdev, (count - 1) * sizeof (void *)); 5576} 5577 5578/* 5579 * Evacuate the device. 5580 */ 5581static int 5582spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5583{ 5584 uint64_t txg; 5585 int error = 0; 5586 5587 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5588 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5589 ASSERT(vd == vd->vdev_top); 5590 5591 /* 5592 * Evacuate the device. We don't hold the config lock as writer 5593 * since we need to do I/O but we do keep the 5594 * spa_namespace_lock held. Once this completes the device 5595 * should no longer have any blocks allocated on it. 5596 */ 5597 if (vd->vdev_islog) { 5598 if (vd->vdev_stat.vs_alloc != 0) 5599 error = spa_offline_log(spa); 5600 } else { 5601 error = SET_ERROR(ENOTSUP); 5602 } 5603 5604 if (error) 5605 return (error); 5606 5607 /* 5608 * The evacuation succeeded. Remove any remaining MOS metadata 5609 * associated with this vdev, and wait for these changes to sync. 5610 */ 5611 ASSERT0(vd->vdev_stat.vs_alloc); 5612 txg = spa_vdev_config_enter(spa); 5613 vd->vdev_removing = B_TRUE; 5614 vdev_dirty_leaves(vd, VDD_DTL, txg); 5615 vdev_config_dirty(vd); 5616 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5617 5618 return (0); 5619} 5620 5621/* 5622 * Complete the removal by cleaning up the namespace. 5623 */ 5624static void 5625spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5626{ 5627 vdev_t *rvd = spa->spa_root_vdev; 5628 uint64_t id = vd->vdev_id; 5629 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5630 5631 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5632 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5633 ASSERT(vd == vd->vdev_top); 5634 5635 /* 5636 * Only remove any devices which are empty. 5637 */ 5638 if (vd->vdev_stat.vs_alloc != 0) 5639 return; 5640 5641 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5642 5643 if (list_link_active(&vd->vdev_state_dirty_node)) 5644 vdev_state_clean(vd); 5645 if (list_link_active(&vd->vdev_config_dirty_node)) 5646 vdev_config_clean(vd); 5647 5648 vdev_free(vd); 5649 5650 if (last_vdev) { 5651 vdev_compact_children(rvd); 5652 } else { 5653 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5654 vdev_add_child(rvd, vd); 5655 } 5656 vdev_config_dirty(rvd); 5657 5658 /* 5659 * Reassess the health of our root vdev. 5660 */ 5661 vdev_reopen(rvd); 5662} 5663 5664/* 5665 * Remove a device from the pool - 5666 * 5667 * Removing a device from the vdev namespace requires several steps 5668 * and can take a significant amount of time. As a result we use 5669 * the spa_vdev_config_[enter/exit] functions which allow us to 5670 * grab and release the spa_config_lock while still holding the namespace 5671 * lock. During each step the configuration is synced out. 5672 * 5673 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5674 * devices. 5675 */ 5676int 5677spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5678{ 5679 vdev_t *vd; 5680 sysevent_t *ev = NULL; 5681 metaslab_group_t *mg; 5682 nvlist_t **spares, **l2cache, *nv; 5683 uint64_t txg = 0; 5684 uint_t nspares, nl2cache; 5685 int error = 0; 5686 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5687 5688 ASSERT(spa_writeable(spa)); 5689 5690 if (!locked) 5691 txg = spa_vdev_enter(spa); 5692 5693 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5694 5695 if (spa->spa_spares.sav_vdevs != NULL && 5696 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5697 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5698 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5699 /* 5700 * Only remove the hot spare if it's not currently in use 5701 * in this pool. 5702 */ 5703 if (vd == NULL || unspare) { 5704 if (vd == NULL) 5705 vd = spa_lookup_by_guid(spa, guid, B_TRUE); 5706 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5707 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5708 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5709 spa_load_spares(spa); 5710 spa->spa_spares.sav_sync = B_TRUE; 5711 } else { 5712 error = SET_ERROR(EBUSY); 5713 } 5714 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5715 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5716 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5717 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5718 /* 5719 * Cache devices can always be removed. 5720 */ 5721 vd = spa_lookup_by_guid(spa, guid, B_TRUE); 5722 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX); 5723 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5724 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5725 spa_load_l2cache(spa); 5726 spa->spa_l2cache.sav_sync = B_TRUE; 5727 } else if (vd != NULL && vd->vdev_islog) { 5728 ASSERT(!locked); 5729 ASSERT(vd == vd->vdev_top); 5730 5731 mg = vd->vdev_mg; 5732 5733 /* 5734 * Stop allocating from this vdev. 5735 */ 5736 metaslab_group_passivate(mg); 5737 5738 /* 5739 * Wait for the youngest allocations and frees to sync, 5740 * and then wait for the deferral of those frees to finish. 5741 */ 5742 spa_vdev_config_exit(spa, NULL, 5743 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5744 5745 /* 5746 * Attempt to evacuate the vdev. 5747 */ 5748 error = spa_vdev_remove_evacuate(spa, vd); 5749 5750 txg = spa_vdev_config_enter(spa); 5751 5752 /* 5753 * If we couldn't evacuate the vdev, unwind. 5754 */ 5755 if (error) { 5756 metaslab_group_activate(mg); 5757 return (spa_vdev_exit(spa, NULL, txg, error)); 5758 } 5759 5760 /* 5761 * Clean up the vdev namespace. 5762 */ 5763 ev = spa_event_create(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV); 5764 spa_vdev_remove_from_namespace(spa, vd); 5765 5766 } else if (vd != NULL) { 5767 /* 5768 * Normal vdevs cannot be removed (yet). 5769 */ 5770 error = SET_ERROR(ENOTSUP); 5771 } else { 5772 /* 5773 * There is no vdev of any kind with the specified guid. 5774 */ 5775 error = SET_ERROR(ENOENT); 5776 } 5777 5778 if (!locked) 5779 error = spa_vdev_exit(spa, NULL, txg, error); 5780 5781 if (ev) 5782 spa_event_post(ev); 5783 5784 return (error); 5785} 5786 5787/* 5788 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5789 * currently spared, so we can detach it. 5790 */ 5791static vdev_t * 5792spa_vdev_resilver_done_hunt(vdev_t *vd) 5793{ 5794 vdev_t *newvd, *oldvd; 5795 5796 for (int c = 0; c < vd->vdev_children; c++) { 5797 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5798 if (oldvd != NULL) 5799 return (oldvd); 5800 } 5801 5802 /* 5803 * Check for a completed replacement. We always consider the first 5804 * vdev in the list to be the oldest vdev, and the last one to be 5805 * the newest (see spa_vdev_attach() for how that works). In 5806 * the case where the newest vdev is faulted, we will not automatically 5807 * remove it after a resilver completes. This is OK as it will require 5808 * user intervention to determine which disk the admin wishes to keep. 5809 */ 5810 if (vd->vdev_ops == &vdev_replacing_ops) { 5811 ASSERT(vd->vdev_children > 1); 5812 5813 newvd = vd->vdev_child[vd->vdev_children - 1]; 5814 oldvd = vd->vdev_child[0]; 5815 5816 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5817 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5818 !vdev_dtl_required(oldvd)) 5819 return (oldvd); 5820 } 5821 5822 /* 5823 * Check for a completed resilver with the 'unspare' flag set. 5824 */ 5825 if (vd->vdev_ops == &vdev_spare_ops) { 5826 vdev_t *first = vd->vdev_child[0]; 5827 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5828 5829 if (last->vdev_unspare) { 5830 oldvd = first; 5831 newvd = last; 5832 } else if (first->vdev_unspare) { 5833 oldvd = last; 5834 newvd = first; 5835 } else { 5836 oldvd = NULL; 5837 } 5838 5839 if (oldvd != NULL && 5840 vdev_dtl_empty(newvd, DTL_MISSING) && 5841 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5842 !vdev_dtl_required(oldvd)) 5843 return (oldvd); 5844 5845 /* 5846 * If there are more than two spares attached to a disk, 5847 * and those spares are not required, then we want to 5848 * attempt to free them up now so that they can be used 5849 * by other pools. Once we're back down to a single 5850 * disk+spare, we stop removing them. 5851 */ 5852 if (vd->vdev_children > 2) { 5853 newvd = vd->vdev_child[1]; 5854 5855 if (newvd->vdev_isspare && last->vdev_isspare && 5856 vdev_dtl_empty(last, DTL_MISSING) && 5857 vdev_dtl_empty(last, DTL_OUTAGE) && 5858 !vdev_dtl_required(newvd)) 5859 return (newvd); 5860 } 5861 } 5862 5863 return (NULL); 5864} 5865 5866static void 5867spa_vdev_resilver_done(spa_t *spa) 5868{ 5869 vdev_t *vd, *pvd, *ppvd; 5870 uint64_t guid, sguid, pguid, ppguid; 5871 5872 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5873 5874 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5875 pvd = vd->vdev_parent; 5876 ppvd = pvd->vdev_parent; 5877 guid = vd->vdev_guid; 5878 pguid = pvd->vdev_guid; 5879 ppguid = ppvd->vdev_guid; 5880 sguid = 0; 5881 /* 5882 * If we have just finished replacing a hot spared device, then 5883 * we need to detach the parent's first child (the original hot 5884 * spare) as well. 5885 */ 5886 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5887 ppvd->vdev_children == 2) { 5888 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5889 sguid = ppvd->vdev_child[1]->vdev_guid; 5890 } 5891 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5892 5893 spa_config_exit(spa, SCL_ALL, FTAG); 5894 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5895 return; 5896 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5897 return; 5898 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5899 } 5900 5901 spa_config_exit(spa, SCL_ALL, FTAG); 5902} 5903 5904/* 5905 * Update the stored path or FRU for this vdev. 5906 */ 5907int 5908spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5909 boolean_t ispath) 5910{ 5911 vdev_t *vd; 5912 boolean_t sync = B_FALSE; 5913 5914 ASSERT(spa_writeable(spa)); 5915 5916 spa_vdev_state_enter(spa, SCL_ALL); 5917 5918 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5919 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5920 5921 if (!vd->vdev_ops->vdev_op_leaf) 5922 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5923 5924 if (ispath) { 5925 if (strcmp(value, vd->vdev_path) != 0) { 5926 spa_strfree(vd->vdev_path); 5927 vd->vdev_path = spa_strdup(value); 5928 sync = B_TRUE; 5929 } 5930 } else { 5931 if (vd->vdev_fru == NULL) { 5932 vd->vdev_fru = spa_strdup(value); 5933 sync = B_TRUE; 5934 } else if (strcmp(value, vd->vdev_fru) != 0) { 5935 spa_strfree(vd->vdev_fru); 5936 vd->vdev_fru = spa_strdup(value); 5937 sync = B_TRUE; 5938 } 5939 } 5940 5941 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5942} 5943 5944int 5945spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5946{ 5947 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5948} 5949 5950int 5951spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5952{ 5953 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5954} 5955 5956/* 5957 * ========================================================================== 5958 * SPA Scanning 5959 * ========================================================================== 5960 */ 5961 5962int 5963spa_scan_stop(spa_t *spa) 5964{ 5965 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5966 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5967 return (SET_ERROR(EBUSY)); 5968 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5969} 5970 5971int 5972spa_scan(spa_t *spa, pool_scan_func_t func) 5973{ 5974 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5975 5976 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5977 return (SET_ERROR(ENOTSUP)); 5978 5979 /* 5980 * If a resilver was requested, but there is no DTL on a 5981 * writeable leaf device, we have nothing to do. 5982 */ 5983 if (func == POOL_SCAN_RESILVER && 5984 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5985 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5986 return (0); 5987 } 5988 5989 return (dsl_scan(spa->spa_dsl_pool, func)); 5990} 5991 5992/* 5993 * ========================================================================== 5994 * SPA async task processing 5995 * ========================================================================== 5996 */ 5997 5998static void 5999spa_async_remove(spa_t *spa, vdev_t *vd) 6000{ 6001 if (vd->vdev_remove_wanted) { 6002 vd->vdev_remove_wanted = B_FALSE; 6003 vd->vdev_delayed_close = B_FALSE; 6004 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 6005 6006 /* 6007 * We want to clear the stats, but we don't want to do a full 6008 * vdev_clear() as that will cause us to throw away 6009 * degraded/faulted state as well as attempt to reopen the 6010 * device, all of which is a waste. 6011 */ 6012 vd->vdev_stat.vs_read_errors = 0; 6013 vd->vdev_stat.vs_write_errors = 0; 6014 vd->vdev_stat.vs_checksum_errors = 0; 6015 6016 vdev_state_dirty(vd->vdev_top); 6017 /* Tell userspace that the vdev is gone. */ 6018 zfs_post_remove(spa, vd); 6019 } 6020 6021 for (int c = 0; c < vd->vdev_children; c++) 6022 spa_async_remove(spa, vd->vdev_child[c]); 6023} 6024 6025static void 6026spa_async_probe(spa_t *spa, vdev_t *vd) 6027{ 6028 if (vd->vdev_probe_wanted) { 6029 vd->vdev_probe_wanted = B_FALSE; 6030 vdev_reopen(vd); /* vdev_open() does the actual probe */ 6031 } 6032 6033 for (int c = 0; c < vd->vdev_children; c++) 6034 spa_async_probe(spa, vd->vdev_child[c]); 6035} 6036 6037static void 6038spa_async_autoexpand(spa_t *spa, vdev_t *vd) 6039{ 6040 sysevent_id_t eid; 6041 nvlist_t *attr; 6042 char *physpath; 6043 6044 if (!spa->spa_autoexpand) 6045 return; 6046 6047 for (int c = 0; c < vd->vdev_children; c++) { 6048 vdev_t *cvd = vd->vdev_child[c]; 6049 spa_async_autoexpand(spa, cvd); 6050 } 6051 6052 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 6053 return; 6054 6055 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 6056 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 6057 6058 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6059 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 6060 6061 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 6062 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 6063 6064 nvlist_free(attr); 6065 kmem_free(physpath, MAXPATHLEN); 6066} 6067 6068static void 6069spa_async_thread(void *arg) 6070{ 6071 spa_t *spa = arg; 6072 int tasks; 6073 6074 ASSERT(spa->spa_sync_on); 6075 6076 mutex_enter(&spa->spa_async_lock); 6077 tasks = spa->spa_async_tasks; 6078 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 6079 mutex_exit(&spa->spa_async_lock); 6080 6081 /* 6082 * See if the config needs to be updated. 6083 */ 6084 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 6085 uint64_t old_space, new_space; 6086 6087 mutex_enter(&spa_namespace_lock); 6088 old_space = metaslab_class_get_space(spa_normal_class(spa)); 6089 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 6090 new_space = metaslab_class_get_space(spa_normal_class(spa)); 6091 mutex_exit(&spa_namespace_lock); 6092 6093 /* 6094 * If the pool grew as a result of the config update, 6095 * then log an internal history event. 6096 */ 6097 if (new_space != old_space) { 6098 spa_history_log_internal(spa, "vdev online", NULL, 6099 "pool '%s' size: %llu(+%llu)", 6100 spa_name(spa), new_space, new_space - old_space); 6101 } 6102 } 6103 6104 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 6105 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6106 spa_async_autoexpand(spa, spa->spa_root_vdev); 6107 spa_config_exit(spa, SCL_CONFIG, FTAG); 6108 } 6109 6110 /* 6111 * See if any devices need to be probed. 6112 */ 6113 if (tasks & SPA_ASYNC_PROBE) { 6114 spa_vdev_state_enter(spa, SCL_NONE); 6115 spa_async_probe(spa, spa->spa_root_vdev); 6116 (void) spa_vdev_state_exit(spa, NULL, 0); 6117 } 6118 6119 /* 6120 * If any devices are done replacing, detach them. 6121 */ 6122 if (tasks & SPA_ASYNC_RESILVER_DONE) 6123 spa_vdev_resilver_done(spa); 6124 6125 /* 6126 * Kick off a resilver. 6127 */ 6128 if (tasks & SPA_ASYNC_RESILVER) 6129 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6130 6131 /* 6132 * Let the world know that we're done. 6133 */ 6134 mutex_enter(&spa->spa_async_lock); 6135 spa->spa_async_thread = NULL; 6136 cv_broadcast(&spa->spa_async_cv); 6137 mutex_exit(&spa->spa_async_lock); 6138 thread_exit(); 6139} 6140 6141static void 6142spa_async_thread_vd(void *arg) 6143{ 6144 spa_t *spa = arg; 6145 int tasks; 6146 6147 mutex_enter(&spa->spa_async_lock); 6148 tasks = spa->spa_async_tasks; 6149retry: 6150 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6151 mutex_exit(&spa->spa_async_lock); 6152 6153 /* 6154 * See if any devices need to be marked REMOVED. 6155 */ 6156 if (tasks & SPA_ASYNC_REMOVE) { 6157 spa_vdev_state_enter(spa, SCL_NONE); 6158 spa_async_remove(spa, spa->spa_root_vdev); 6159 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6160 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6161 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6162 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6163 (void) spa_vdev_state_exit(spa, NULL, 0); 6164 } 6165 6166 /* 6167 * Let the world know that we're done. 6168 */ 6169 mutex_enter(&spa->spa_async_lock); 6170 tasks = spa->spa_async_tasks; 6171 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6172 goto retry; 6173 spa->spa_async_thread_vd = NULL; 6174 cv_broadcast(&spa->spa_async_cv); 6175 mutex_exit(&spa->spa_async_lock); 6176 thread_exit(); 6177} 6178 6179void 6180spa_async_suspend(spa_t *spa) 6181{ 6182 mutex_enter(&spa->spa_async_lock); 6183 spa->spa_async_suspended++; 6184 while (spa->spa_async_thread != NULL && 6185 spa->spa_async_thread_vd != NULL) 6186 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6187 mutex_exit(&spa->spa_async_lock); 6188} 6189 6190void 6191spa_async_resume(spa_t *spa) 6192{ 6193 mutex_enter(&spa->spa_async_lock); 6194 ASSERT(spa->spa_async_suspended != 0); 6195 spa->spa_async_suspended--; 6196 mutex_exit(&spa->spa_async_lock); 6197} 6198 6199static boolean_t 6200spa_async_tasks_pending(spa_t *spa) 6201{ 6202 uint_t non_config_tasks; 6203 uint_t config_task; 6204 boolean_t config_task_suspended; 6205 6206 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6207 SPA_ASYNC_REMOVE); 6208 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6209 if (spa->spa_ccw_fail_time == 0) { 6210 config_task_suspended = B_FALSE; 6211 } else { 6212 config_task_suspended = 6213 (gethrtime() - spa->spa_ccw_fail_time) < 6214 (zfs_ccw_retry_interval * NANOSEC); 6215 } 6216 6217 return (non_config_tasks || (config_task && !config_task_suspended)); 6218} 6219 6220static void 6221spa_async_dispatch(spa_t *spa) 6222{ 6223 mutex_enter(&spa->spa_async_lock); 6224 if (spa_async_tasks_pending(spa) && 6225 !spa->spa_async_suspended && 6226 spa->spa_async_thread == NULL && 6227 rootdir != NULL) 6228 spa->spa_async_thread = thread_create(NULL, 0, 6229 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6230 mutex_exit(&spa->spa_async_lock); 6231} 6232 6233static void 6234spa_async_dispatch_vd(spa_t *spa) 6235{ 6236 mutex_enter(&spa->spa_async_lock); 6237 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6238 !spa->spa_async_suspended && 6239 spa->spa_async_thread_vd == NULL && 6240 rootdir != NULL) 6241 spa->spa_async_thread_vd = thread_create(NULL, 0, 6242 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6243 mutex_exit(&spa->spa_async_lock); 6244} 6245 6246void 6247spa_async_request(spa_t *spa, int task) 6248{ 6249 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6250 mutex_enter(&spa->spa_async_lock); 6251 spa->spa_async_tasks |= task; 6252 mutex_exit(&spa->spa_async_lock); 6253 spa_async_dispatch_vd(spa); 6254} 6255 6256/* 6257 * ========================================================================== 6258 * SPA syncing routines 6259 * ========================================================================== 6260 */ 6261 6262static int 6263bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6264{ 6265 bpobj_t *bpo = arg; 6266 bpobj_enqueue(bpo, bp, tx); 6267 return (0); 6268} 6269 6270static int 6271spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6272{ 6273 zio_t *zio = arg; 6274 6275 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6276 BP_GET_PSIZE(bp), zio->io_flags)); 6277 return (0); 6278} 6279 6280/* 6281 * Note: this simple function is not inlined to make it easier to dtrace the 6282 * amount of time spent syncing frees. 6283 */ 6284static void 6285spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6286{ 6287 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6288 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6289 VERIFY(zio_wait(zio) == 0); 6290} 6291 6292/* 6293 * Note: this simple function is not inlined to make it easier to dtrace the 6294 * amount of time spent syncing deferred frees. 6295 */ 6296static void 6297spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6298{ 6299 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6300 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6301 spa_free_sync_cb, zio, tx), ==, 0); 6302 VERIFY0(zio_wait(zio)); 6303} 6304 6305 6306static void 6307spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6308{ 6309 char *packed = NULL; 6310 size_t bufsize; 6311 size_t nvsize = 0; 6312 dmu_buf_t *db; 6313 6314 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6315 6316 /* 6317 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6318 * information. This avoids the dmu_buf_will_dirty() path and 6319 * saves us a pre-read to get data we don't actually care about. 6320 */ 6321 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6322 packed = kmem_alloc(bufsize, KM_SLEEP); 6323 6324 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6325 KM_SLEEP) == 0); 6326 bzero(packed + nvsize, bufsize - nvsize); 6327 6328 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6329 6330 kmem_free(packed, bufsize); 6331 6332 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6333 dmu_buf_will_dirty(db, tx); 6334 *(uint64_t *)db->db_data = nvsize; 6335 dmu_buf_rele(db, FTAG); 6336} 6337 6338static void 6339spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6340 const char *config, const char *entry) 6341{ 6342 nvlist_t *nvroot; 6343 nvlist_t **list; 6344 int i; 6345 6346 if (!sav->sav_sync) 6347 return; 6348 6349 /* 6350 * Update the MOS nvlist describing the list of available devices. 6351 * spa_validate_aux() will have already made sure this nvlist is 6352 * valid and the vdevs are labeled appropriately. 6353 */ 6354 if (sav->sav_object == 0) { 6355 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6356 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6357 sizeof (uint64_t), tx); 6358 VERIFY(zap_update(spa->spa_meta_objset, 6359 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6360 &sav->sav_object, tx) == 0); 6361 } 6362 6363 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6364 if (sav->sav_count == 0) { 6365 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6366 } else { 6367 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6368 for (i = 0; i < sav->sav_count; i++) 6369 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6370 B_FALSE, VDEV_CONFIG_L2CACHE); 6371 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6372 sav->sav_count) == 0); 6373 for (i = 0; i < sav->sav_count; i++) 6374 nvlist_free(list[i]); 6375 kmem_free(list, sav->sav_count * sizeof (void *)); 6376 } 6377 6378 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6379 nvlist_free(nvroot); 6380 6381 sav->sav_sync = B_FALSE; 6382} 6383 6384static void 6385spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6386{ 6387 nvlist_t *config; 6388 6389 if (list_is_empty(&spa->spa_config_dirty_list)) 6390 return; 6391 6392 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6393 6394 config = spa_config_generate(spa, spa->spa_root_vdev, 6395 dmu_tx_get_txg(tx), B_FALSE); 6396 6397 /* 6398 * If we're upgrading the spa version then make sure that 6399 * the config object gets updated with the correct version. 6400 */ 6401 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6402 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6403 spa->spa_uberblock.ub_version); 6404 6405 spa_config_exit(spa, SCL_STATE, FTAG); 6406 6407 nvlist_free(spa->spa_config_syncing); 6408 spa->spa_config_syncing = config; 6409 6410 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6411} 6412 6413static void 6414spa_sync_version(void *arg, dmu_tx_t *tx) 6415{ 6416 uint64_t *versionp = arg; 6417 uint64_t version = *versionp; 6418 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6419 6420 /* 6421 * Setting the version is special cased when first creating the pool. 6422 */ 6423 ASSERT(tx->tx_txg != TXG_INITIAL); 6424 6425 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6426 ASSERT(version >= spa_version(spa)); 6427 6428 spa->spa_uberblock.ub_version = version; 6429 vdev_config_dirty(spa->spa_root_vdev); 6430 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6431} 6432 6433/* 6434 * Set zpool properties. 6435 */ 6436static void 6437spa_sync_props(void *arg, dmu_tx_t *tx) 6438{ 6439 nvlist_t *nvp = arg; 6440 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6441 objset_t *mos = spa->spa_meta_objset; 6442 nvpair_t *elem = NULL; 6443 6444 mutex_enter(&spa->spa_props_lock); 6445 6446 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6447 uint64_t intval; 6448 char *strval, *fname; 6449 zpool_prop_t prop; 6450 const char *propname; 6451 zprop_type_t proptype; 6452 spa_feature_t fid; 6453 6454 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6455 case ZPROP_INVAL: 6456 /* 6457 * We checked this earlier in spa_prop_validate(). 6458 */ 6459 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6460 6461 fname = strchr(nvpair_name(elem), '@') + 1; 6462 VERIFY0(zfeature_lookup_name(fname, &fid)); 6463 6464 spa_feature_enable(spa, fid, tx); 6465 spa_history_log_internal(spa, "set", tx, 6466 "%s=enabled", nvpair_name(elem)); 6467 break; 6468 6469 case ZPOOL_PROP_VERSION: 6470 intval = fnvpair_value_uint64(elem); 6471 /* 6472 * The version is synced seperatly before other 6473 * properties and should be correct by now. 6474 */ 6475 ASSERT3U(spa_version(spa), >=, intval); 6476 break; 6477 6478 case ZPOOL_PROP_ALTROOT: 6479 /* 6480 * 'altroot' is a non-persistent property. It should 6481 * have been set temporarily at creation or import time. 6482 */ 6483 ASSERT(spa->spa_root != NULL); 6484 break; 6485 6486 case ZPOOL_PROP_READONLY: 6487 case ZPOOL_PROP_CACHEFILE: 6488 /* 6489 * 'readonly' and 'cachefile' are also non-persisitent 6490 * properties. 6491 */ 6492 break; 6493 case ZPOOL_PROP_COMMENT: 6494 strval = fnvpair_value_string(elem); 6495 if (spa->spa_comment != NULL) 6496 spa_strfree(spa->spa_comment); 6497 spa->spa_comment = spa_strdup(strval); 6498 /* 6499 * We need to dirty the configuration on all the vdevs 6500 * so that their labels get updated. It's unnecessary 6501 * to do this for pool creation since the vdev's 6502 * configuratoin has already been dirtied. 6503 */ 6504 if (tx->tx_txg != TXG_INITIAL) 6505 vdev_config_dirty(spa->spa_root_vdev); 6506 spa_history_log_internal(spa, "set", tx, 6507 "%s=%s", nvpair_name(elem), strval); 6508 break; 6509 default: 6510 /* 6511 * Set pool property values in the poolprops mos object. 6512 */ 6513 if (spa->spa_pool_props_object == 0) { 6514 spa->spa_pool_props_object = 6515 zap_create_link(mos, DMU_OT_POOL_PROPS, 6516 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6517 tx); 6518 } 6519 6520 /* normalize the property name */ 6521 propname = zpool_prop_to_name(prop); 6522 proptype = zpool_prop_get_type(prop); 6523 6524 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6525 ASSERT(proptype == PROP_TYPE_STRING); 6526 strval = fnvpair_value_string(elem); 6527 VERIFY0(zap_update(mos, 6528 spa->spa_pool_props_object, propname, 6529 1, strlen(strval) + 1, strval, tx)); 6530 spa_history_log_internal(spa, "set", tx, 6531 "%s=%s", nvpair_name(elem), strval); 6532 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6533 intval = fnvpair_value_uint64(elem); 6534 6535 if (proptype == PROP_TYPE_INDEX) { 6536 const char *unused; 6537 VERIFY0(zpool_prop_index_to_string( 6538 prop, intval, &unused)); 6539 } 6540 VERIFY0(zap_update(mos, 6541 spa->spa_pool_props_object, propname, 6542 8, 1, &intval, tx)); 6543 spa_history_log_internal(spa, "set", tx, 6544 "%s=%lld", nvpair_name(elem), intval); 6545 } else { 6546 ASSERT(0); /* not allowed */ 6547 } 6548 6549 switch (prop) { 6550 case ZPOOL_PROP_DELEGATION: 6551 spa->spa_delegation = intval; 6552 break; 6553 case ZPOOL_PROP_BOOTFS: 6554 spa->spa_bootfs = intval; 6555 break; 6556 case ZPOOL_PROP_FAILUREMODE: 6557 spa->spa_failmode = intval; 6558 break; 6559 case ZPOOL_PROP_AUTOEXPAND: 6560 spa->spa_autoexpand = intval; 6561 if (tx->tx_txg != TXG_INITIAL) 6562 spa_async_request(spa, 6563 SPA_ASYNC_AUTOEXPAND); 6564 break; 6565 case ZPOOL_PROP_DEDUPDITTO: 6566 spa->spa_dedup_ditto = intval; 6567 break; 6568 default: 6569 break; 6570 } 6571 } 6572 6573 } 6574 6575 mutex_exit(&spa->spa_props_lock); 6576} 6577 6578/* 6579 * Perform one-time upgrade on-disk changes. spa_version() does not 6580 * reflect the new version this txg, so there must be no changes this 6581 * txg to anything that the upgrade code depends on after it executes. 6582 * Therefore this must be called after dsl_pool_sync() does the sync 6583 * tasks. 6584 */ 6585static void 6586spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6587{ 6588 dsl_pool_t *dp = spa->spa_dsl_pool; 6589 6590 ASSERT(spa->spa_sync_pass == 1); 6591 6592 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6593 6594 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6595 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6596 dsl_pool_create_origin(dp, tx); 6597 6598 /* Keeping the origin open increases spa_minref */ 6599 spa->spa_minref += 3; 6600 } 6601 6602 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6603 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6604 dsl_pool_upgrade_clones(dp, tx); 6605 } 6606 6607 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6608 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6609 dsl_pool_upgrade_dir_clones(dp, tx); 6610 6611 /* Keeping the freedir open increases spa_minref */ 6612 spa->spa_minref += 3; 6613 } 6614 6615 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6616 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6617 spa_feature_create_zap_objects(spa, tx); 6618 } 6619 6620 /* 6621 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6622 * when possibility to use lz4 compression for metadata was added 6623 * Old pools that have this feature enabled must be upgraded to have 6624 * this feature active 6625 */ 6626 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6627 boolean_t lz4_en = spa_feature_is_enabled(spa, 6628 SPA_FEATURE_LZ4_COMPRESS); 6629 boolean_t lz4_ac = spa_feature_is_active(spa, 6630 SPA_FEATURE_LZ4_COMPRESS); 6631 6632 if (lz4_en && !lz4_ac) 6633 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6634 } 6635 6636 /* 6637 * If we haven't written the salt, do so now. Note that the 6638 * feature may not be activated yet, but that's fine since 6639 * the presence of this ZAP entry is backwards compatible. 6640 */ 6641 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 6642 DMU_POOL_CHECKSUM_SALT) == ENOENT) { 6643 VERIFY0(zap_add(spa->spa_meta_objset, 6644 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1, 6645 sizeof (spa->spa_cksum_salt.zcs_bytes), 6646 spa->spa_cksum_salt.zcs_bytes, tx)); 6647 } 6648 6649 rrw_exit(&dp->dp_config_rwlock, FTAG); 6650} 6651 6652/* 6653 * Sync the specified transaction group. New blocks may be dirtied as 6654 * part of the process, so we iterate until it converges. 6655 */ 6656void 6657spa_sync(spa_t *spa, uint64_t txg) 6658{ 6659 dsl_pool_t *dp = spa->spa_dsl_pool; 6660 objset_t *mos = spa->spa_meta_objset; 6661 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6662 vdev_t *rvd = spa->spa_root_vdev; 6663 vdev_t *vd; 6664 dmu_tx_t *tx; 6665 int error; 6666 uint32_t max_queue_depth = zfs_vdev_async_write_max_active * 6667 zfs_vdev_queue_depth_pct / 100; 6668 6669 VERIFY(spa_writeable(spa)); 6670 6671 /* 6672 * Lock out configuration changes. 6673 */ 6674 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6675 6676 spa->spa_syncing_txg = txg; 6677 spa->spa_sync_pass = 0; 6678 6679 mutex_enter(&spa->spa_alloc_lock); 6680 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 6681 mutex_exit(&spa->spa_alloc_lock); 6682 6683 /* 6684 * If there are any pending vdev state changes, convert them 6685 * into config changes that go out with this transaction group. 6686 */ 6687 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6688 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6689 /* 6690 * We need the write lock here because, for aux vdevs, 6691 * calling vdev_config_dirty() modifies sav_config. 6692 * This is ugly and will become unnecessary when we 6693 * eliminate the aux vdev wart by integrating all vdevs 6694 * into the root vdev tree. 6695 */ 6696 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6697 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6698 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6699 vdev_state_clean(vd); 6700 vdev_config_dirty(vd); 6701 } 6702 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6703 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6704 } 6705 spa_config_exit(spa, SCL_STATE, FTAG); 6706 6707 tx = dmu_tx_create_assigned(dp, txg); 6708 6709 spa->spa_sync_starttime = gethrtime(); 6710#ifdef illumos 6711 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6712 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6713#else /* !illumos */ 6714#ifdef _KERNEL 6715 callout_schedule(&spa->spa_deadman_cycid, 6716 hz * spa->spa_deadman_synctime / NANOSEC); 6717#endif 6718#endif /* illumos */ 6719 6720 /* 6721 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6722 * set spa_deflate if we have no raid-z vdevs. 6723 */ 6724 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6725 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6726 int i; 6727 6728 for (i = 0; i < rvd->vdev_children; i++) { 6729 vd = rvd->vdev_child[i]; 6730 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6731 break; 6732 } 6733 if (i == rvd->vdev_children) { 6734 spa->spa_deflate = TRUE; 6735 VERIFY(0 == zap_add(spa->spa_meta_objset, 6736 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6737 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6738 } 6739 } 6740 6741 /* 6742 * Set the top-level vdev's max queue depth. Evaluate each 6743 * top-level's async write queue depth in case it changed. 6744 * The max queue depth will not change in the middle of syncing 6745 * out this txg. 6746 */ 6747 uint64_t queue_depth_total = 0; 6748 for (int c = 0; c < rvd->vdev_children; c++) { 6749 vdev_t *tvd = rvd->vdev_child[c]; 6750 metaslab_group_t *mg = tvd->vdev_mg; 6751 6752 if (mg == NULL || mg->mg_class != spa_normal_class(spa) || 6753 !metaslab_group_initialized(mg)) 6754 continue; 6755 6756 /* 6757 * It is safe to do a lock-free check here because only async 6758 * allocations look at mg_max_alloc_queue_depth, and async 6759 * allocations all happen from spa_sync(). 6760 */ 6761 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth)); 6762 mg->mg_max_alloc_queue_depth = max_queue_depth; 6763 queue_depth_total += mg->mg_max_alloc_queue_depth; 6764 } 6765 metaslab_class_t *mc = spa_normal_class(spa); 6766 ASSERT0(refcount_count(&mc->mc_alloc_slots)); 6767 mc->mc_alloc_max_slots = queue_depth_total; 6768 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled; 6769 6770 ASSERT3U(mc->mc_alloc_max_slots, <=, 6771 max_queue_depth * rvd->vdev_children); 6772 6773 /* 6774 * Iterate to convergence. 6775 */ 6776 do { 6777 int pass = ++spa->spa_sync_pass; 6778 6779 spa_sync_config_object(spa, tx); 6780 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6781 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6782 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6783 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6784 spa_errlog_sync(spa, txg); 6785 dsl_pool_sync(dp, txg); 6786 6787 if (pass < zfs_sync_pass_deferred_free) { 6788 spa_sync_frees(spa, free_bpl, tx); 6789 } else { 6790 /* 6791 * We can not defer frees in pass 1, because 6792 * we sync the deferred frees later in pass 1. 6793 */ 6794 ASSERT3U(pass, >, 1); 6795 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6796 &spa->spa_deferred_bpobj, tx); 6797 } 6798 6799 ddt_sync(spa, txg); 6800 dsl_scan_sync(dp, tx); 6801 6802 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6803 vdev_sync(vd, txg); 6804 6805 if (pass == 1) { 6806 spa_sync_upgrades(spa, tx); 6807 ASSERT3U(txg, >=, 6808 spa->spa_uberblock.ub_rootbp.blk_birth); 6809 /* 6810 * Note: We need to check if the MOS is dirty 6811 * because we could have marked the MOS dirty 6812 * without updating the uberblock (e.g. if we 6813 * have sync tasks but no dirty user data). We 6814 * need to check the uberblock's rootbp because 6815 * it is updated if we have synced out dirty 6816 * data (though in this case the MOS will most 6817 * likely also be dirty due to second order 6818 * effects, we don't want to rely on that here). 6819 */ 6820 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6821 !dmu_objset_is_dirty(mos, txg)) { 6822 /* 6823 * Nothing changed on the first pass, 6824 * therefore this TXG is a no-op. Avoid 6825 * syncing deferred frees, so that we 6826 * can keep this TXG as a no-op. 6827 */ 6828 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6829 txg)); 6830 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6831 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6832 break; 6833 } 6834 spa_sync_deferred_frees(spa, tx); 6835 } 6836 6837 } while (dmu_objset_is_dirty(mos, txg)); 6838 6839 /* 6840 * Rewrite the vdev configuration (which includes the uberblock) 6841 * to commit the transaction group. 6842 * 6843 * If there are no dirty vdevs, we sync the uberblock to a few 6844 * random top-level vdevs that are known to be visible in the 6845 * config cache (see spa_vdev_add() for a complete description). 6846 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6847 */ 6848 for (;;) { 6849 /* 6850 * We hold SCL_STATE to prevent vdev open/close/etc. 6851 * while we're attempting to write the vdev labels. 6852 */ 6853 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6854 6855 if (list_is_empty(&spa->spa_config_dirty_list)) { 6856 vdev_t *svd[SPA_DVAS_PER_BP]; 6857 int svdcount = 0; 6858 int children = rvd->vdev_children; 6859 int c0 = spa_get_random(children); 6860 6861 for (int c = 0; c < children; c++) { 6862 vd = rvd->vdev_child[(c0 + c) % children]; 6863 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6864 continue; 6865 svd[svdcount++] = vd; 6866 if (svdcount == SPA_DVAS_PER_BP) 6867 break; 6868 } 6869 error = vdev_config_sync(svd, svdcount, txg); 6870 } else { 6871 error = vdev_config_sync(rvd->vdev_child, 6872 rvd->vdev_children, txg); 6873 } 6874 6875 if (error == 0) 6876 spa->spa_last_synced_guid = rvd->vdev_guid; 6877 6878 spa_config_exit(spa, SCL_STATE, FTAG); 6879 6880 if (error == 0) 6881 break; 6882 zio_suspend(spa, NULL); 6883 zio_resume_wait(spa); 6884 } 6885 dmu_tx_commit(tx); 6886 6887#ifdef illumos 6888 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6889#else /* !illumos */ 6890#ifdef _KERNEL 6891 callout_drain(&spa->spa_deadman_cycid); 6892#endif 6893#endif /* illumos */ 6894 6895 /* 6896 * Clear the dirty config list. 6897 */ 6898 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6899 vdev_config_clean(vd); 6900 6901 /* 6902 * Now that the new config has synced transactionally, 6903 * let it become visible to the config cache. 6904 */ 6905 if (spa->spa_config_syncing != NULL) { 6906 spa_config_set(spa, spa->spa_config_syncing); 6907 spa->spa_config_txg = txg; 6908 spa->spa_config_syncing = NULL; 6909 } 6910 6911 dsl_pool_sync_done(dp, txg); 6912 6913 mutex_enter(&spa->spa_alloc_lock); 6914 VERIFY0(avl_numnodes(&spa->spa_alloc_tree)); 6915 mutex_exit(&spa->spa_alloc_lock); 6916 6917 /* 6918 * Update usable space statistics. 6919 */ 6920 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6921 vdev_sync_done(vd, txg); 6922 6923 spa_update_dspace(spa); 6924 6925 /* 6926 * It had better be the case that we didn't dirty anything 6927 * since vdev_config_sync(). 6928 */ 6929 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6930 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6931 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6932 6933 spa->spa_sync_pass = 0; 6934 6935 /* 6936 * Update the last synced uberblock here. We want to do this at 6937 * the end of spa_sync() so that consumers of spa_last_synced_txg() 6938 * will be guaranteed that all the processing associated with 6939 * that txg has been completed. 6940 */ 6941 spa->spa_ubsync = spa->spa_uberblock; 6942 spa_config_exit(spa, SCL_CONFIG, FTAG); 6943 6944 spa_handle_ignored_writes(spa); 6945 6946 /* 6947 * If any async tasks have been requested, kick them off. 6948 */ 6949 spa_async_dispatch(spa); 6950 spa_async_dispatch_vd(spa); 6951} 6952 6953/* 6954 * Sync all pools. We don't want to hold the namespace lock across these 6955 * operations, so we take a reference on the spa_t and drop the lock during the 6956 * sync. 6957 */ 6958void 6959spa_sync_allpools(void) 6960{ 6961 spa_t *spa = NULL; 6962 mutex_enter(&spa_namespace_lock); 6963 while ((spa = spa_next(spa)) != NULL) { 6964 if (spa_state(spa) != POOL_STATE_ACTIVE || 6965 !spa_writeable(spa) || spa_suspended(spa)) 6966 continue; 6967 spa_open_ref(spa, FTAG); 6968 mutex_exit(&spa_namespace_lock); 6969 txg_wait_synced(spa_get_dsl(spa), 0); 6970 mutex_enter(&spa_namespace_lock); 6971 spa_close(spa, FTAG); 6972 } 6973 mutex_exit(&spa_namespace_lock); 6974} 6975 6976/* 6977 * ========================================================================== 6978 * Miscellaneous routines 6979 * ========================================================================== 6980 */ 6981 6982/* 6983 * Remove all pools in the system. 6984 */ 6985void 6986spa_evict_all(void) 6987{ 6988 spa_t *spa; 6989 6990 /* 6991 * Remove all cached state. All pools should be closed now, 6992 * so every spa in the AVL tree should be unreferenced. 6993 */ 6994 mutex_enter(&spa_namespace_lock); 6995 while ((spa = spa_next(NULL)) != NULL) { 6996 /* 6997 * Stop async tasks. The async thread may need to detach 6998 * a device that's been replaced, which requires grabbing 6999 * spa_namespace_lock, so we must drop it here. 7000 */ 7001 spa_open_ref(spa, FTAG); 7002 mutex_exit(&spa_namespace_lock); 7003 spa_async_suspend(spa); 7004 mutex_enter(&spa_namespace_lock); 7005 spa_close(spa, FTAG); 7006 7007 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 7008 spa_unload(spa); 7009 spa_deactivate(spa); 7010 } 7011 spa_remove(spa); 7012 } 7013 mutex_exit(&spa_namespace_lock); 7014} 7015 7016vdev_t * 7017spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 7018{ 7019 vdev_t *vd; 7020 int i; 7021 7022 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 7023 return (vd); 7024 7025 if (aux) { 7026 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 7027 vd = spa->spa_l2cache.sav_vdevs[i]; 7028 if (vd->vdev_guid == guid) 7029 return (vd); 7030 } 7031 7032 for (i = 0; i < spa->spa_spares.sav_count; i++) { 7033 vd = spa->spa_spares.sav_vdevs[i]; 7034 if (vd->vdev_guid == guid) 7035 return (vd); 7036 } 7037 } 7038 7039 return (NULL); 7040} 7041 7042void 7043spa_upgrade(spa_t *spa, uint64_t version) 7044{ 7045 ASSERT(spa_writeable(spa)); 7046 7047 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 7048 7049 /* 7050 * This should only be called for a non-faulted pool, and since a 7051 * future version would result in an unopenable pool, this shouldn't be 7052 * possible. 7053 */ 7054 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 7055 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 7056 7057 spa->spa_uberblock.ub_version = version; 7058 vdev_config_dirty(spa->spa_root_vdev); 7059 7060 spa_config_exit(spa, SCL_ALL, FTAG); 7061 7062 txg_wait_synced(spa_get_dsl(spa), 0); 7063} 7064 7065boolean_t 7066spa_has_spare(spa_t *spa, uint64_t guid) 7067{ 7068 int i; 7069 uint64_t spareguid; 7070 spa_aux_vdev_t *sav = &spa->spa_spares; 7071 7072 for (i = 0; i < sav->sav_count; i++) 7073 if (sav->sav_vdevs[i]->vdev_guid == guid) 7074 return (B_TRUE); 7075 7076 for (i = 0; i < sav->sav_npending; i++) { 7077 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 7078 &spareguid) == 0 && spareguid == guid) 7079 return (B_TRUE); 7080 } 7081 7082 return (B_FALSE); 7083} 7084 7085/* 7086 * Check if a pool has an active shared spare device. 7087 * Note: reference count of an active spare is 2, as a spare and as a replace 7088 */ 7089static boolean_t 7090spa_has_active_shared_spare(spa_t *spa) 7091{ 7092 int i, refcnt; 7093 uint64_t pool; 7094 spa_aux_vdev_t *sav = &spa->spa_spares; 7095 7096 for (i = 0; i < sav->sav_count; i++) { 7097 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 7098 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 7099 refcnt > 2) 7100 return (B_TRUE); 7101 } 7102 7103 return (B_FALSE); 7104} 7105 7106static sysevent_t * 7107spa_event_create(spa_t *spa, vdev_t *vd, const char *name) 7108{ 7109 sysevent_t *ev = NULL; 7110#ifdef _KERNEL 7111 sysevent_attr_list_t *attr = NULL; 7112 sysevent_value_t value; 7113 7114 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 7115 SE_SLEEP); 7116 ASSERT(ev != NULL); 7117 7118 value.value_type = SE_DATA_TYPE_STRING; 7119 value.value.sv_string = spa_name(spa); 7120 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 7121 goto done; 7122 7123 value.value_type = SE_DATA_TYPE_UINT64; 7124 value.value.sv_uint64 = spa_guid(spa); 7125 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 7126 goto done; 7127 7128 if (vd) { 7129 value.value_type = SE_DATA_TYPE_UINT64; 7130 value.value.sv_uint64 = vd->vdev_guid; 7131 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 7132 SE_SLEEP) != 0) 7133 goto done; 7134 7135 if (vd->vdev_path) { 7136 value.value_type = SE_DATA_TYPE_STRING; 7137 value.value.sv_string = vd->vdev_path; 7138 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 7139 &value, SE_SLEEP) != 0) 7140 goto done; 7141 } 7142 } 7143 7144 if (sysevent_attach_attributes(ev, attr) != 0) 7145 goto done; 7146 attr = NULL; 7147 7148done: 7149 if (attr) 7150 sysevent_free_attr(attr); 7151 7152#endif 7153 return (ev); 7154} 7155 7156static void 7157spa_event_post(sysevent_t *ev) 7158{ 7159#ifdef _KERNEL 7160 sysevent_id_t eid; 7161 7162 (void) log_sysevent(ev, SE_SLEEP, &eid); 7163 sysevent_free(ev); 7164#endif 7165} 7166 7167/* 7168 * Post a sysevent corresponding to the given event. The 'name' must be one of 7169 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 7170 * filled in from the spa and (optionally) the vdev. This doesn't do anything 7171 * in the userland libzpool, as we don't want consumers to misinterpret ztest 7172 * or zdb as real changes. 7173 */ 7174void 7175spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 7176{ 7177 spa_event_post(spa_event_create(spa, vd, name)); 7178} 7179