spa.c revision 277300
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 25 * Copyright (c) 2013, 2014, Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 */ 28 29/* 30 * SPA: Storage Pool Allocator 31 * 32 * This file contains all the routines used when modifying on-disk SPA state. 33 * This includes opening, importing, destroying, exporting a pool, and syncing a 34 * pool. 35 */ 36 37#include <sys/zfs_context.h> 38#include <sys/fm/fs/zfs.h> 39#include <sys/spa_impl.h> 40#include <sys/zio.h> 41#include <sys/zio_checksum.h> 42#include <sys/dmu.h> 43#include <sys/dmu_tx.h> 44#include <sys/zap.h> 45#include <sys/zil.h> 46#include <sys/ddt.h> 47#include <sys/vdev_impl.h> 48#include <sys/metaslab.h> 49#include <sys/metaslab_impl.h> 50#include <sys/uberblock_impl.h> 51#include <sys/txg.h> 52#include <sys/avl.h> 53#include <sys/dmu_traverse.h> 54#include <sys/dmu_objset.h> 55#include <sys/unique.h> 56#include <sys/dsl_pool.h> 57#include <sys/dsl_dataset.h> 58#include <sys/dsl_dir.h> 59#include <sys/dsl_prop.h> 60#include <sys/dsl_synctask.h> 61#include <sys/fs/zfs.h> 62#include <sys/arc.h> 63#include <sys/callb.h> 64#include <sys/spa_boot.h> 65#include <sys/zfs_ioctl.h> 66#include <sys/dsl_scan.h> 67#include <sys/dmu_send.h> 68#include <sys/dsl_destroy.h> 69#include <sys/dsl_userhold.h> 70#include <sys/zfeature.h> 71#include <sys/zvol.h> 72#include <sys/trim_map.h> 73 74#ifdef _KERNEL 75#include <sys/callb.h> 76#include <sys/cpupart.h> 77#include <sys/zone.h> 78#endif /* _KERNEL */ 79 80#include "zfs_prop.h" 81#include "zfs_comutil.h" 82 83/* Check hostid on import? */ 84static int check_hostid = 1; 85 86/* 87 * The interval, in seconds, at which failed configuration cache file writes 88 * should be retried. 89 */ 90static int zfs_ccw_retry_interval = 300; 91 92SYSCTL_DECL(_vfs_zfs); 93SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0, 94 "Check hostid on import?"); 95TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval); 96SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW, 97 &zfs_ccw_retry_interval, 0, 98 "Configuration cache file write, retry after failure, interval (seconds)"); 99 100typedef enum zti_modes { 101 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 102 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 103 ZTI_MODE_NULL, /* don't create a taskq */ 104 ZTI_NMODES 105} zti_modes_t; 106 107#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 108#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 109#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 110 111#define ZTI_N(n) ZTI_P(n, 1) 112#define ZTI_ONE ZTI_N(1) 113 114typedef struct zio_taskq_info { 115 zti_modes_t zti_mode; 116 uint_t zti_value; 117 uint_t zti_count; 118} zio_taskq_info_t; 119 120static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 121 "issue", "issue_high", "intr", "intr_high" 122}; 123 124/* 125 * This table defines the taskq settings for each ZFS I/O type. When 126 * initializing a pool, we use this table to create an appropriately sized 127 * taskq. Some operations are low volume and therefore have a small, static 128 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 129 * macros. Other operations process a large amount of data; the ZTI_BATCH 130 * macro causes us to create a taskq oriented for throughput. Some operations 131 * are so high frequency and short-lived that the taskq itself can become a a 132 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 133 * additional degree of parallelism specified by the number of threads per- 134 * taskq and the number of taskqs; when dispatching an event in this case, the 135 * particular taskq is chosen at random. 136 * 137 * The different taskq priorities are to handle the different contexts (issue 138 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 139 * need to be handled with minimum delay. 140 */ 141const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 142 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 143 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 144 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */ 145 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 146 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 147 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 148 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 149}; 150 151static void spa_sync_version(void *arg, dmu_tx_t *tx); 152static void spa_sync_props(void *arg, dmu_tx_t *tx); 153static boolean_t spa_has_active_shared_spare(spa_t *spa); 154static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 155 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 156 char **ereport); 157static void spa_vdev_resilver_done(spa_t *spa); 158 159uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 160#ifdef PSRSET_BIND 161id_t zio_taskq_psrset_bind = PS_NONE; 162#endif 163#ifdef SYSDC 164boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 165#endif 166uint_t zio_taskq_basedc = 80; /* base duty cycle */ 167 168boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 169extern int zfs_sync_pass_deferred_free; 170 171#ifndef illumos 172extern void spa_deadman(void *arg); 173#endif 174 175/* 176 * This (illegal) pool name is used when temporarily importing a spa_t in order 177 * to get the vdev stats associated with the imported devices. 178 */ 179#define TRYIMPORT_NAME "$import" 180 181/* 182 * ========================================================================== 183 * SPA properties routines 184 * ========================================================================== 185 */ 186 187/* 188 * Add a (source=src, propname=propval) list to an nvlist. 189 */ 190static void 191spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 192 uint64_t intval, zprop_source_t src) 193{ 194 const char *propname = zpool_prop_to_name(prop); 195 nvlist_t *propval; 196 197 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 198 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 199 200 if (strval != NULL) 201 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 202 else 203 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 204 205 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 206 nvlist_free(propval); 207} 208 209/* 210 * Get property values from the spa configuration. 211 */ 212static void 213spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 214{ 215 vdev_t *rvd = spa->spa_root_vdev; 216 dsl_pool_t *pool = spa->spa_dsl_pool; 217 uint64_t size, alloc, cap, version; 218 zprop_source_t src = ZPROP_SRC_NONE; 219 spa_config_dirent_t *dp; 220 metaslab_class_t *mc = spa_normal_class(spa); 221 222 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 223 224 if (rvd != NULL) { 225 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 226 size = metaslab_class_get_space(spa_normal_class(spa)); 227 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 228 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 229 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 230 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 231 size - alloc, src); 232 233 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL, 234 metaslab_class_fragmentation(mc), src); 235 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, 236 metaslab_class_expandable_space(mc), src); 237 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 238 (spa_mode(spa) == FREAD), src); 239 240 cap = (size == 0) ? 0 : (alloc * 100 / size); 241 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 242 243 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 244 ddt_get_pool_dedup_ratio(spa), src); 245 246 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 247 rvd->vdev_state, src); 248 249 version = spa_version(spa); 250 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 251 src = ZPROP_SRC_DEFAULT; 252 else 253 src = ZPROP_SRC_LOCAL; 254 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 255 } 256 257 if (pool != NULL) { 258 /* 259 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 260 * when opening pools before this version freedir will be NULL. 261 */ 262 if (pool->dp_free_dir != NULL) { 263 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 264 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes, 265 src); 266 } else { 267 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 268 NULL, 0, src); 269 } 270 271 if (pool->dp_leak_dir != NULL) { 272 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL, 273 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes, 274 src); 275 } else { 276 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, 277 NULL, 0, src); 278 } 279 } 280 281 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 282 283 if (spa->spa_comment != NULL) { 284 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 285 0, ZPROP_SRC_LOCAL); 286 } 287 288 if (spa->spa_root != NULL) 289 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 290 0, ZPROP_SRC_LOCAL); 291 292 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { 293 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 294 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE); 295 } else { 296 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL, 297 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE); 298 } 299 300 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 301 if (dp->scd_path == NULL) { 302 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 303 "none", 0, ZPROP_SRC_LOCAL); 304 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 305 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 306 dp->scd_path, 0, ZPROP_SRC_LOCAL); 307 } 308 } 309} 310 311/* 312 * Get zpool property values. 313 */ 314int 315spa_prop_get(spa_t *spa, nvlist_t **nvp) 316{ 317 objset_t *mos = spa->spa_meta_objset; 318 zap_cursor_t zc; 319 zap_attribute_t za; 320 int err; 321 322 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 323 324 mutex_enter(&spa->spa_props_lock); 325 326 /* 327 * Get properties from the spa config. 328 */ 329 spa_prop_get_config(spa, nvp); 330 331 /* If no pool property object, no more prop to get. */ 332 if (mos == NULL || spa->spa_pool_props_object == 0) { 333 mutex_exit(&spa->spa_props_lock); 334 return (0); 335 } 336 337 /* 338 * Get properties from the MOS pool property object. 339 */ 340 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 341 (err = zap_cursor_retrieve(&zc, &za)) == 0; 342 zap_cursor_advance(&zc)) { 343 uint64_t intval = 0; 344 char *strval = NULL; 345 zprop_source_t src = ZPROP_SRC_DEFAULT; 346 zpool_prop_t prop; 347 348 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 349 continue; 350 351 switch (za.za_integer_length) { 352 case 8: 353 /* integer property */ 354 if (za.za_first_integer != 355 zpool_prop_default_numeric(prop)) 356 src = ZPROP_SRC_LOCAL; 357 358 if (prop == ZPOOL_PROP_BOOTFS) { 359 dsl_pool_t *dp; 360 dsl_dataset_t *ds = NULL; 361 362 dp = spa_get_dsl(spa); 363 dsl_pool_config_enter(dp, FTAG); 364 if (err = dsl_dataset_hold_obj(dp, 365 za.za_first_integer, FTAG, &ds)) { 366 dsl_pool_config_exit(dp, FTAG); 367 break; 368 } 369 370 strval = kmem_alloc( 371 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 372 KM_SLEEP); 373 dsl_dataset_name(ds, strval); 374 dsl_dataset_rele(ds, FTAG); 375 dsl_pool_config_exit(dp, FTAG); 376 } else { 377 strval = NULL; 378 intval = za.za_first_integer; 379 } 380 381 spa_prop_add_list(*nvp, prop, strval, intval, src); 382 383 if (strval != NULL) 384 kmem_free(strval, 385 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 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 check++; 612 } 613 if (strlen(strval) > ZPROP_MAX_COMMENT) 614 error = E2BIG; 615 break; 616 617 case ZPOOL_PROP_DEDUPDITTO: 618 if (spa_version(spa) < SPA_VERSION_DEDUP) 619 error = SET_ERROR(ENOTSUP); 620 else 621 error = nvpair_value_uint64(elem, &intval); 622 if (error == 0 && 623 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 624 error = SET_ERROR(EINVAL); 625 break; 626 } 627 628 if (error) 629 break; 630 } 631 632 if (!error && reset_bootfs) { 633 error = nvlist_remove(props, 634 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 635 636 if (!error) { 637 error = nvlist_add_uint64(props, 638 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 639 } 640 } 641 642 return (error); 643} 644 645void 646spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 647{ 648 char *cachefile; 649 spa_config_dirent_t *dp; 650 651 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 652 &cachefile) != 0) 653 return; 654 655 dp = kmem_alloc(sizeof (spa_config_dirent_t), 656 KM_SLEEP); 657 658 if (cachefile[0] == '\0') 659 dp->scd_path = spa_strdup(spa_config_path); 660 else if (strcmp(cachefile, "none") == 0) 661 dp->scd_path = NULL; 662 else 663 dp->scd_path = spa_strdup(cachefile); 664 665 list_insert_head(&spa->spa_config_list, dp); 666 if (need_sync) 667 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 668} 669 670int 671spa_prop_set(spa_t *spa, nvlist_t *nvp) 672{ 673 int error; 674 nvpair_t *elem = NULL; 675 boolean_t need_sync = B_FALSE; 676 677 if ((error = spa_prop_validate(spa, nvp)) != 0) 678 return (error); 679 680 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 681 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 682 683 if (prop == ZPOOL_PROP_CACHEFILE || 684 prop == ZPOOL_PROP_ALTROOT || 685 prop == ZPOOL_PROP_READONLY) 686 continue; 687 688 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 689 uint64_t ver; 690 691 if (prop == ZPOOL_PROP_VERSION) { 692 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 693 } else { 694 ASSERT(zpool_prop_feature(nvpair_name(elem))); 695 ver = SPA_VERSION_FEATURES; 696 need_sync = B_TRUE; 697 } 698 699 /* Save time if the version is already set. */ 700 if (ver == spa_version(spa)) 701 continue; 702 703 /* 704 * In addition to the pool directory object, we might 705 * create the pool properties object, the features for 706 * read object, the features for write object, or the 707 * feature descriptions object. 708 */ 709 error = dsl_sync_task(spa->spa_name, NULL, 710 spa_sync_version, &ver, 711 6, ZFS_SPACE_CHECK_RESERVED); 712 if (error) 713 return (error); 714 continue; 715 } 716 717 need_sync = B_TRUE; 718 break; 719 } 720 721 if (need_sync) { 722 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 723 nvp, 6, ZFS_SPACE_CHECK_RESERVED)); 724 } 725 726 return (0); 727} 728 729/* 730 * If the bootfs property value is dsobj, clear it. 731 */ 732void 733spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 734{ 735 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 736 VERIFY(zap_remove(spa->spa_meta_objset, 737 spa->spa_pool_props_object, 738 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 739 spa->spa_bootfs = 0; 740 } 741} 742 743/*ARGSUSED*/ 744static int 745spa_change_guid_check(void *arg, dmu_tx_t *tx) 746{ 747 uint64_t *newguid = arg; 748 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 749 vdev_t *rvd = spa->spa_root_vdev; 750 uint64_t vdev_state; 751 752 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 753 vdev_state = rvd->vdev_state; 754 spa_config_exit(spa, SCL_STATE, FTAG); 755 756 if (vdev_state != VDEV_STATE_HEALTHY) 757 return (SET_ERROR(ENXIO)); 758 759 ASSERT3U(spa_guid(spa), !=, *newguid); 760 761 return (0); 762} 763 764static void 765spa_change_guid_sync(void *arg, dmu_tx_t *tx) 766{ 767 uint64_t *newguid = arg; 768 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 769 uint64_t oldguid; 770 vdev_t *rvd = spa->spa_root_vdev; 771 772 oldguid = spa_guid(spa); 773 774 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 775 rvd->vdev_guid = *newguid; 776 rvd->vdev_guid_sum += (*newguid - oldguid); 777 vdev_config_dirty(rvd); 778 spa_config_exit(spa, SCL_STATE, FTAG); 779 780 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 781 oldguid, *newguid); 782} 783 784/* 785 * Change the GUID for the pool. This is done so that we can later 786 * re-import a pool built from a clone of our own vdevs. We will modify 787 * the root vdev's guid, our own pool guid, and then mark all of our 788 * vdevs dirty. Note that we must make sure that all our vdevs are 789 * online when we do this, or else any vdevs that weren't present 790 * would be orphaned from our pool. We are also going to issue a 791 * sysevent to update any watchers. 792 */ 793int 794spa_change_guid(spa_t *spa) 795{ 796 int error; 797 uint64_t guid; 798 799 mutex_enter(&spa->spa_vdev_top_lock); 800 mutex_enter(&spa_namespace_lock); 801 guid = spa_generate_guid(NULL); 802 803 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 804 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED); 805 806 if (error == 0) { 807 spa_config_sync(spa, B_FALSE, B_TRUE); 808 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 809 } 810 811 mutex_exit(&spa_namespace_lock); 812 mutex_exit(&spa->spa_vdev_top_lock); 813 814 return (error); 815} 816 817/* 818 * ========================================================================== 819 * SPA state manipulation (open/create/destroy/import/export) 820 * ========================================================================== 821 */ 822 823static int 824spa_error_entry_compare(const void *a, const void *b) 825{ 826 spa_error_entry_t *sa = (spa_error_entry_t *)a; 827 spa_error_entry_t *sb = (spa_error_entry_t *)b; 828 int ret; 829 830 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 831 sizeof (zbookmark_phys_t)); 832 833 if (ret < 0) 834 return (-1); 835 else if (ret > 0) 836 return (1); 837 else 838 return (0); 839} 840 841/* 842 * Utility function which retrieves copies of the current logs and 843 * re-initializes them in the process. 844 */ 845void 846spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 847{ 848 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 849 850 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 851 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 852 853 avl_create(&spa->spa_errlist_scrub, 854 spa_error_entry_compare, sizeof (spa_error_entry_t), 855 offsetof(spa_error_entry_t, se_avl)); 856 avl_create(&spa->spa_errlist_last, 857 spa_error_entry_compare, sizeof (spa_error_entry_t), 858 offsetof(spa_error_entry_t, se_avl)); 859} 860 861static void 862spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 863{ 864 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 865 enum zti_modes mode = ztip->zti_mode; 866 uint_t value = ztip->zti_value; 867 uint_t count = ztip->zti_count; 868 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 869 char name[32]; 870 uint_t flags = 0; 871 boolean_t batch = B_FALSE; 872 873 if (mode == ZTI_MODE_NULL) { 874 tqs->stqs_count = 0; 875 tqs->stqs_taskq = NULL; 876 return; 877 } 878 879 ASSERT3U(count, >, 0); 880 881 tqs->stqs_count = count; 882 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 883 884 switch (mode) { 885 case ZTI_MODE_FIXED: 886 ASSERT3U(value, >=, 1); 887 value = MAX(value, 1); 888 break; 889 890 case ZTI_MODE_BATCH: 891 batch = B_TRUE; 892 flags |= TASKQ_THREADS_CPU_PCT; 893 value = zio_taskq_batch_pct; 894 break; 895 896 default: 897 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 898 "spa_activate()", 899 zio_type_name[t], zio_taskq_types[q], mode, value); 900 break; 901 } 902 903 for (uint_t i = 0; i < count; i++) { 904 taskq_t *tq; 905 906 if (count > 1) { 907 (void) snprintf(name, sizeof (name), "%s_%s_%u", 908 zio_type_name[t], zio_taskq_types[q], i); 909 } else { 910 (void) snprintf(name, sizeof (name), "%s_%s", 911 zio_type_name[t], zio_taskq_types[q]); 912 } 913 914#ifdef SYSDC 915 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 916 if (batch) 917 flags |= TASKQ_DC_BATCH; 918 919 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 920 spa->spa_proc, zio_taskq_basedc, flags); 921 } else { 922#endif 923 pri_t pri = maxclsyspri; 924 /* 925 * The write issue taskq can be extremely CPU 926 * intensive. Run it at slightly lower priority 927 * than the other taskqs. 928 */ 929 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 930 pri--; 931 932 tq = taskq_create_proc(name, value, pri, 50, 933 INT_MAX, spa->spa_proc, flags); 934#ifdef SYSDC 935 } 936#endif 937 938 tqs->stqs_taskq[i] = tq; 939 } 940} 941 942static void 943spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 944{ 945 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 946 947 if (tqs->stqs_taskq == NULL) { 948 ASSERT0(tqs->stqs_count); 949 return; 950 } 951 952 for (uint_t i = 0; i < tqs->stqs_count; i++) { 953 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 954 taskq_destroy(tqs->stqs_taskq[i]); 955 } 956 957 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 958 tqs->stqs_taskq = NULL; 959} 960 961/* 962 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 963 * Note that a type may have multiple discrete taskqs to avoid lock contention 964 * on the taskq itself. In that case we choose which taskq at random by using 965 * the low bits of gethrtime(). 966 */ 967void 968spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 969 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 970{ 971 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 972 taskq_t *tq; 973 974 ASSERT3P(tqs->stqs_taskq, !=, NULL); 975 ASSERT3U(tqs->stqs_count, !=, 0); 976 977 if (tqs->stqs_count == 1) { 978 tq = tqs->stqs_taskq[0]; 979 } else { 980#ifdef _KERNEL 981 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count]; 982#else 983 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 984#endif 985 } 986 987 taskq_dispatch_ent(tq, func, arg, flags, ent); 988} 989 990static void 991spa_create_zio_taskqs(spa_t *spa) 992{ 993 for (int t = 0; t < ZIO_TYPES; t++) { 994 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 995 spa_taskqs_init(spa, t, q); 996 } 997 } 998} 999 1000#ifdef _KERNEL 1001#ifdef SPA_PROCESS 1002static void 1003spa_thread(void *arg) 1004{ 1005 callb_cpr_t cprinfo; 1006 1007 spa_t *spa = arg; 1008 user_t *pu = PTOU(curproc); 1009 1010 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 1011 spa->spa_name); 1012 1013 ASSERT(curproc != &p0); 1014 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 1015 "zpool-%s", spa->spa_name); 1016 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 1017 1018#ifdef PSRSET_BIND 1019 /* bind this thread to the requested psrset */ 1020 if (zio_taskq_psrset_bind != PS_NONE) { 1021 pool_lock(); 1022 mutex_enter(&cpu_lock); 1023 mutex_enter(&pidlock); 1024 mutex_enter(&curproc->p_lock); 1025 1026 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 1027 0, NULL, NULL) == 0) { 1028 curthread->t_bind_pset = zio_taskq_psrset_bind; 1029 } else { 1030 cmn_err(CE_WARN, 1031 "Couldn't bind process for zfs pool \"%s\" to " 1032 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1033 } 1034 1035 mutex_exit(&curproc->p_lock); 1036 mutex_exit(&pidlock); 1037 mutex_exit(&cpu_lock); 1038 pool_unlock(); 1039 } 1040#endif 1041 1042#ifdef SYSDC 1043 if (zio_taskq_sysdc) { 1044 sysdc_thread_enter(curthread, 100, 0); 1045 } 1046#endif 1047 1048 spa->spa_proc = curproc; 1049 spa->spa_did = curthread->t_did; 1050 1051 spa_create_zio_taskqs(spa); 1052 1053 mutex_enter(&spa->spa_proc_lock); 1054 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1055 1056 spa->spa_proc_state = SPA_PROC_ACTIVE; 1057 cv_broadcast(&spa->spa_proc_cv); 1058 1059 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1060 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1061 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1062 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1063 1064 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1065 spa->spa_proc_state = SPA_PROC_GONE; 1066 spa->spa_proc = &p0; 1067 cv_broadcast(&spa->spa_proc_cv); 1068 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1069 1070 mutex_enter(&curproc->p_lock); 1071 lwp_exit(); 1072} 1073#endif /* SPA_PROCESS */ 1074#endif 1075 1076/* 1077 * Activate an uninitialized pool. 1078 */ 1079static void 1080spa_activate(spa_t *spa, int mode) 1081{ 1082 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1083 1084 spa->spa_state = POOL_STATE_ACTIVE; 1085 spa->spa_mode = mode; 1086 1087 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1088 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1089 1090 /* Try to create a covering process */ 1091 mutex_enter(&spa->spa_proc_lock); 1092 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1093 ASSERT(spa->spa_proc == &p0); 1094 spa->spa_did = 0; 1095 1096#ifdef SPA_PROCESS 1097 /* Only create a process if we're going to be around a while. */ 1098 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1099 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1100 NULL, 0) == 0) { 1101 spa->spa_proc_state = SPA_PROC_CREATED; 1102 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1103 cv_wait(&spa->spa_proc_cv, 1104 &spa->spa_proc_lock); 1105 } 1106 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1107 ASSERT(spa->spa_proc != &p0); 1108 ASSERT(spa->spa_did != 0); 1109 } else { 1110#ifdef _KERNEL 1111 cmn_err(CE_WARN, 1112 "Couldn't create process for zfs pool \"%s\"\n", 1113 spa->spa_name); 1114#endif 1115 } 1116 } 1117#endif /* SPA_PROCESS */ 1118 mutex_exit(&spa->spa_proc_lock); 1119 1120 /* If we didn't create a process, we need to create our taskqs. */ 1121 ASSERT(spa->spa_proc == &p0); 1122 if (spa->spa_proc == &p0) { 1123 spa_create_zio_taskqs(spa); 1124 } 1125 1126 /* 1127 * Start TRIM thread. 1128 */ 1129 trim_thread_create(spa); 1130 1131 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1132 offsetof(vdev_t, vdev_config_dirty_node)); 1133 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1134 offsetof(vdev_t, vdev_state_dirty_node)); 1135 1136 txg_list_create(&spa->spa_vdev_txg_list, 1137 offsetof(struct vdev, vdev_txg_node)); 1138 1139 avl_create(&spa->spa_errlist_scrub, 1140 spa_error_entry_compare, sizeof (spa_error_entry_t), 1141 offsetof(spa_error_entry_t, se_avl)); 1142 avl_create(&spa->spa_errlist_last, 1143 spa_error_entry_compare, sizeof (spa_error_entry_t), 1144 offsetof(spa_error_entry_t, se_avl)); 1145} 1146 1147/* 1148 * Opposite of spa_activate(). 1149 */ 1150static void 1151spa_deactivate(spa_t *spa) 1152{ 1153 ASSERT(spa->spa_sync_on == B_FALSE); 1154 ASSERT(spa->spa_dsl_pool == NULL); 1155 ASSERT(spa->spa_root_vdev == NULL); 1156 ASSERT(spa->spa_async_zio_root == NULL); 1157 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1158 1159 /* 1160 * Stop TRIM thread in case spa_unload() wasn't called directly 1161 * before spa_deactivate(). 1162 */ 1163 trim_thread_destroy(spa); 1164 1165 txg_list_destroy(&spa->spa_vdev_txg_list); 1166 1167 list_destroy(&spa->spa_config_dirty_list); 1168 list_destroy(&spa->spa_state_dirty_list); 1169 1170 for (int t = 0; t < ZIO_TYPES; t++) { 1171 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1172 spa_taskqs_fini(spa, t, q); 1173 } 1174 } 1175 1176 metaslab_class_destroy(spa->spa_normal_class); 1177 spa->spa_normal_class = NULL; 1178 1179 metaslab_class_destroy(spa->spa_log_class); 1180 spa->spa_log_class = NULL; 1181 1182 /* 1183 * If this was part of an import or the open otherwise failed, we may 1184 * still have errors left in the queues. Empty them just in case. 1185 */ 1186 spa_errlog_drain(spa); 1187 1188 avl_destroy(&spa->spa_errlist_scrub); 1189 avl_destroy(&spa->spa_errlist_last); 1190 1191 spa->spa_state = POOL_STATE_UNINITIALIZED; 1192 1193 mutex_enter(&spa->spa_proc_lock); 1194 if (spa->spa_proc_state != SPA_PROC_NONE) { 1195 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1196 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1197 cv_broadcast(&spa->spa_proc_cv); 1198 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1199 ASSERT(spa->spa_proc != &p0); 1200 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1201 } 1202 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1203 spa->spa_proc_state = SPA_PROC_NONE; 1204 } 1205 ASSERT(spa->spa_proc == &p0); 1206 mutex_exit(&spa->spa_proc_lock); 1207 1208#ifdef SPA_PROCESS 1209 /* 1210 * We want to make sure spa_thread() has actually exited the ZFS 1211 * module, so that the module can't be unloaded out from underneath 1212 * it. 1213 */ 1214 if (spa->spa_did != 0) { 1215 thread_join(spa->spa_did); 1216 spa->spa_did = 0; 1217 } 1218#endif /* SPA_PROCESS */ 1219} 1220 1221/* 1222 * Verify a pool configuration, and construct the vdev tree appropriately. This 1223 * will create all the necessary vdevs in the appropriate layout, with each vdev 1224 * in the CLOSED state. This will prep the pool before open/creation/import. 1225 * All vdev validation is done by the vdev_alloc() routine. 1226 */ 1227static int 1228spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1229 uint_t id, int atype) 1230{ 1231 nvlist_t **child; 1232 uint_t children; 1233 int error; 1234 1235 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1236 return (error); 1237 1238 if ((*vdp)->vdev_ops->vdev_op_leaf) 1239 return (0); 1240 1241 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1242 &child, &children); 1243 1244 if (error == ENOENT) 1245 return (0); 1246 1247 if (error) { 1248 vdev_free(*vdp); 1249 *vdp = NULL; 1250 return (SET_ERROR(EINVAL)); 1251 } 1252 1253 for (int c = 0; c < children; c++) { 1254 vdev_t *vd; 1255 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1256 atype)) != 0) { 1257 vdev_free(*vdp); 1258 *vdp = NULL; 1259 return (error); 1260 } 1261 } 1262 1263 ASSERT(*vdp != NULL); 1264 1265 return (0); 1266} 1267 1268/* 1269 * Opposite of spa_load(). 1270 */ 1271static void 1272spa_unload(spa_t *spa) 1273{ 1274 int i; 1275 1276 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1277 1278 /* 1279 * Stop TRIM thread. 1280 */ 1281 trim_thread_destroy(spa); 1282 1283 /* 1284 * Stop async tasks. 1285 */ 1286 spa_async_suspend(spa); 1287 1288 /* 1289 * Stop syncing. 1290 */ 1291 if (spa->spa_sync_on) { 1292 txg_sync_stop(spa->spa_dsl_pool); 1293 spa->spa_sync_on = B_FALSE; 1294 } 1295 1296 /* 1297 * Wait for any outstanding async I/O to complete. 1298 */ 1299 if (spa->spa_async_zio_root != NULL) { 1300 for (int i = 0; i < max_ncpus; i++) 1301 (void) zio_wait(spa->spa_async_zio_root[i]); 1302 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *)); 1303 spa->spa_async_zio_root = NULL; 1304 } 1305 1306 bpobj_close(&spa->spa_deferred_bpobj); 1307 1308 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1309 1310 /* 1311 * Close all vdevs. 1312 */ 1313 if (spa->spa_root_vdev) 1314 vdev_free(spa->spa_root_vdev); 1315 ASSERT(spa->spa_root_vdev == NULL); 1316 1317 /* 1318 * Close the dsl pool. 1319 */ 1320 if (spa->spa_dsl_pool) { 1321 dsl_pool_close(spa->spa_dsl_pool); 1322 spa->spa_dsl_pool = NULL; 1323 spa->spa_meta_objset = NULL; 1324 } 1325 1326 ddt_unload(spa); 1327 1328 1329 /* 1330 * Drop and purge level 2 cache 1331 */ 1332 spa_l2cache_drop(spa); 1333 1334 for (i = 0; i < spa->spa_spares.sav_count; i++) 1335 vdev_free(spa->spa_spares.sav_vdevs[i]); 1336 if (spa->spa_spares.sav_vdevs) { 1337 kmem_free(spa->spa_spares.sav_vdevs, 1338 spa->spa_spares.sav_count * sizeof (void *)); 1339 spa->spa_spares.sav_vdevs = NULL; 1340 } 1341 if (spa->spa_spares.sav_config) { 1342 nvlist_free(spa->spa_spares.sav_config); 1343 spa->spa_spares.sav_config = NULL; 1344 } 1345 spa->spa_spares.sav_count = 0; 1346 1347 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1348 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1349 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1350 } 1351 if (spa->spa_l2cache.sav_vdevs) { 1352 kmem_free(spa->spa_l2cache.sav_vdevs, 1353 spa->spa_l2cache.sav_count * sizeof (void *)); 1354 spa->spa_l2cache.sav_vdevs = NULL; 1355 } 1356 if (spa->spa_l2cache.sav_config) { 1357 nvlist_free(spa->spa_l2cache.sav_config); 1358 spa->spa_l2cache.sav_config = NULL; 1359 } 1360 spa->spa_l2cache.sav_count = 0; 1361 1362 spa->spa_async_suspended = 0; 1363 1364 if (spa->spa_comment != NULL) { 1365 spa_strfree(spa->spa_comment); 1366 spa->spa_comment = NULL; 1367 } 1368 1369 spa_config_exit(spa, SCL_ALL, FTAG); 1370} 1371 1372/* 1373 * Load (or re-load) the current list of vdevs describing the active spares for 1374 * this pool. When this is called, we have some form of basic information in 1375 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1376 * then re-generate a more complete list including status information. 1377 */ 1378static void 1379spa_load_spares(spa_t *spa) 1380{ 1381 nvlist_t **spares; 1382 uint_t nspares; 1383 int i; 1384 vdev_t *vd, *tvd; 1385 1386 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1387 1388 /* 1389 * First, close and free any existing spare vdevs. 1390 */ 1391 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1392 vd = spa->spa_spares.sav_vdevs[i]; 1393 1394 /* Undo the call to spa_activate() below */ 1395 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1396 B_FALSE)) != NULL && tvd->vdev_isspare) 1397 spa_spare_remove(tvd); 1398 vdev_close(vd); 1399 vdev_free(vd); 1400 } 1401 1402 if (spa->spa_spares.sav_vdevs) 1403 kmem_free(spa->spa_spares.sav_vdevs, 1404 spa->spa_spares.sav_count * sizeof (void *)); 1405 1406 if (spa->spa_spares.sav_config == NULL) 1407 nspares = 0; 1408 else 1409 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1410 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1411 1412 spa->spa_spares.sav_count = (int)nspares; 1413 spa->spa_spares.sav_vdevs = NULL; 1414 1415 if (nspares == 0) 1416 return; 1417 1418 /* 1419 * Construct the array of vdevs, opening them to get status in the 1420 * process. For each spare, there is potentially two different vdev_t 1421 * structures associated with it: one in the list of spares (used only 1422 * for basic validation purposes) and one in the active vdev 1423 * configuration (if it's spared in). During this phase we open and 1424 * validate each vdev on the spare list. If the vdev also exists in the 1425 * active configuration, then we also mark this vdev as an active spare. 1426 */ 1427 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1428 KM_SLEEP); 1429 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1430 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1431 VDEV_ALLOC_SPARE) == 0); 1432 ASSERT(vd != NULL); 1433 1434 spa->spa_spares.sav_vdevs[i] = vd; 1435 1436 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1437 B_FALSE)) != NULL) { 1438 if (!tvd->vdev_isspare) 1439 spa_spare_add(tvd); 1440 1441 /* 1442 * We only mark the spare active if we were successfully 1443 * able to load the vdev. Otherwise, importing a pool 1444 * with a bad active spare would result in strange 1445 * behavior, because multiple pool would think the spare 1446 * is actively in use. 1447 * 1448 * There is a vulnerability here to an equally bizarre 1449 * circumstance, where a dead active spare is later 1450 * brought back to life (onlined or otherwise). Given 1451 * the rarity of this scenario, and the extra complexity 1452 * it adds, we ignore the possibility. 1453 */ 1454 if (!vdev_is_dead(tvd)) 1455 spa_spare_activate(tvd); 1456 } 1457 1458 vd->vdev_top = vd; 1459 vd->vdev_aux = &spa->spa_spares; 1460 1461 if (vdev_open(vd) != 0) 1462 continue; 1463 1464 if (vdev_validate_aux(vd) == 0) 1465 spa_spare_add(vd); 1466 } 1467 1468 /* 1469 * Recompute the stashed list of spares, with status information 1470 * this time. 1471 */ 1472 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1473 DATA_TYPE_NVLIST_ARRAY) == 0); 1474 1475 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1476 KM_SLEEP); 1477 for (i = 0; i < spa->spa_spares.sav_count; i++) 1478 spares[i] = vdev_config_generate(spa, 1479 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1480 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1481 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1482 for (i = 0; i < spa->spa_spares.sav_count; i++) 1483 nvlist_free(spares[i]); 1484 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1485} 1486 1487/* 1488 * Load (or re-load) the current list of vdevs describing the active l2cache for 1489 * this pool. When this is called, we have some form of basic information in 1490 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1491 * then re-generate a more complete list including status information. 1492 * Devices which are already active have their details maintained, and are 1493 * not re-opened. 1494 */ 1495static void 1496spa_load_l2cache(spa_t *spa) 1497{ 1498 nvlist_t **l2cache; 1499 uint_t nl2cache; 1500 int i, j, oldnvdevs; 1501 uint64_t guid; 1502 vdev_t *vd, **oldvdevs, **newvdevs; 1503 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1504 1505 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1506 1507 if (sav->sav_config != NULL) { 1508 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1509 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1510 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1511 } else { 1512 nl2cache = 0; 1513 newvdevs = NULL; 1514 } 1515 1516 oldvdevs = sav->sav_vdevs; 1517 oldnvdevs = sav->sav_count; 1518 sav->sav_vdevs = NULL; 1519 sav->sav_count = 0; 1520 1521 /* 1522 * Process new nvlist of vdevs. 1523 */ 1524 for (i = 0; i < nl2cache; i++) { 1525 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1526 &guid) == 0); 1527 1528 newvdevs[i] = NULL; 1529 for (j = 0; j < oldnvdevs; j++) { 1530 vd = oldvdevs[j]; 1531 if (vd != NULL && guid == vd->vdev_guid) { 1532 /* 1533 * Retain previous vdev for add/remove ops. 1534 */ 1535 newvdevs[i] = vd; 1536 oldvdevs[j] = NULL; 1537 break; 1538 } 1539 } 1540 1541 if (newvdevs[i] == NULL) { 1542 /* 1543 * Create new vdev 1544 */ 1545 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1546 VDEV_ALLOC_L2CACHE) == 0); 1547 ASSERT(vd != NULL); 1548 newvdevs[i] = vd; 1549 1550 /* 1551 * Commit this vdev as an l2cache device, 1552 * even if it fails to open. 1553 */ 1554 spa_l2cache_add(vd); 1555 1556 vd->vdev_top = vd; 1557 vd->vdev_aux = sav; 1558 1559 spa_l2cache_activate(vd); 1560 1561 if (vdev_open(vd) != 0) 1562 continue; 1563 1564 (void) vdev_validate_aux(vd); 1565 1566 if (!vdev_is_dead(vd)) 1567 l2arc_add_vdev(spa, vd); 1568 } 1569 } 1570 1571 /* 1572 * Purge vdevs that were dropped 1573 */ 1574 for (i = 0; i < oldnvdevs; i++) { 1575 uint64_t pool; 1576 1577 vd = oldvdevs[i]; 1578 if (vd != NULL) { 1579 ASSERT(vd->vdev_isl2cache); 1580 1581 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1582 pool != 0ULL && l2arc_vdev_present(vd)) 1583 l2arc_remove_vdev(vd); 1584 vdev_clear_stats(vd); 1585 vdev_free(vd); 1586 } 1587 } 1588 1589 if (oldvdevs) 1590 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1591 1592 if (sav->sav_config == NULL) 1593 goto out; 1594 1595 sav->sav_vdevs = newvdevs; 1596 sav->sav_count = (int)nl2cache; 1597 1598 /* 1599 * Recompute the stashed list of l2cache devices, with status 1600 * information this time. 1601 */ 1602 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1603 DATA_TYPE_NVLIST_ARRAY) == 0); 1604 1605 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1606 for (i = 0; i < sav->sav_count; i++) 1607 l2cache[i] = vdev_config_generate(spa, 1608 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1609 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1610 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1611out: 1612 for (i = 0; i < sav->sav_count; i++) 1613 nvlist_free(l2cache[i]); 1614 if (sav->sav_count) 1615 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1616} 1617 1618static int 1619load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1620{ 1621 dmu_buf_t *db; 1622 char *packed = NULL; 1623 size_t nvsize = 0; 1624 int error; 1625 *value = NULL; 1626 1627 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db); 1628 if (error != 0) 1629 return (error); 1630 nvsize = *(uint64_t *)db->db_data; 1631 dmu_buf_rele(db, FTAG); 1632 1633 packed = kmem_alloc(nvsize, KM_SLEEP); 1634 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1635 DMU_READ_PREFETCH); 1636 if (error == 0) 1637 error = nvlist_unpack(packed, nvsize, value, 0); 1638 kmem_free(packed, nvsize); 1639 1640 return (error); 1641} 1642 1643/* 1644 * Checks to see if the given vdev could not be opened, in which case we post a 1645 * sysevent to notify the autoreplace code that the device has been removed. 1646 */ 1647static void 1648spa_check_removed(vdev_t *vd) 1649{ 1650 for (int c = 0; c < vd->vdev_children; c++) 1651 spa_check_removed(vd->vdev_child[c]); 1652 1653 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1654 !vd->vdev_ishole) { 1655 zfs_post_autoreplace(vd->vdev_spa, vd); 1656 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1657 } 1658} 1659 1660/* 1661 * Validate the current config against the MOS config 1662 */ 1663static boolean_t 1664spa_config_valid(spa_t *spa, nvlist_t *config) 1665{ 1666 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1667 nvlist_t *nv; 1668 1669 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1670 1671 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1672 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1673 1674 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1675 1676 /* 1677 * If we're doing a normal import, then build up any additional 1678 * diagnostic information about missing devices in this config. 1679 * We'll pass this up to the user for further processing. 1680 */ 1681 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1682 nvlist_t **child, *nv; 1683 uint64_t idx = 0; 1684 1685 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1686 KM_SLEEP); 1687 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1688 1689 for (int c = 0; c < rvd->vdev_children; c++) { 1690 vdev_t *tvd = rvd->vdev_child[c]; 1691 vdev_t *mtvd = mrvd->vdev_child[c]; 1692 1693 if (tvd->vdev_ops == &vdev_missing_ops && 1694 mtvd->vdev_ops != &vdev_missing_ops && 1695 mtvd->vdev_islog) 1696 child[idx++] = vdev_config_generate(spa, mtvd, 1697 B_FALSE, 0); 1698 } 1699 1700 if (idx) { 1701 VERIFY(nvlist_add_nvlist_array(nv, 1702 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1703 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1704 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1705 1706 for (int i = 0; i < idx; i++) 1707 nvlist_free(child[i]); 1708 } 1709 nvlist_free(nv); 1710 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1711 } 1712 1713 /* 1714 * Compare the root vdev tree with the information we have 1715 * from the MOS config (mrvd). Check each top-level vdev 1716 * with the corresponding MOS config top-level (mtvd). 1717 */ 1718 for (int c = 0; c < rvd->vdev_children; c++) { 1719 vdev_t *tvd = rvd->vdev_child[c]; 1720 vdev_t *mtvd = mrvd->vdev_child[c]; 1721 1722 /* 1723 * Resolve any "missing" vdevs in the current configuration. 1724 * If we find that the MOS config has more accurate information 1725 * about the top-level vdev then use that vdev instead. 1726 */ 1727 if (tvd->vdev_ops == &vdev_missing_ops && 1728 mtvd->vdev_ops != &vdev_missing_ops) { 1729 1730 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1731 continue; 1732 1733 /* 1734 * Device specific actions. 1735 */ 1736 if (mtvd->vdev_islog) { 1737 spa_set_log_state(spa, SPA_LOG_CLEAR); 1738 } else { 1739 /* 1740 * XXX - once we have 'readonly' pool 1741 * support we should be able to handle 1742 * missing data devices by transitioning 1743 * the pool to readonly. 1744 */ 1745 continue; 1746 } 1747 1748 /* 1749 * Swap the missing vdev with the data we were 1750 * able to obtain from the MOS config. 1751 */ 1752 vdev_remove_child(rvd, tvd); 1753 vdev_remove_child(mrvd, mtvd); 1754 1755 vdev_add_child(rvd, mtvd); 1756 vdev_add_child(mrvd, tvd); 1757 1758 spa_config_exit(spa, SCL_ALL, FTAG); 1759 vdev_load(mtvd); 1760 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1761 1762 vdev_reopen(rvd); 1763 } else if (mtvd->vdev_islog) { 1764 /* 1765 * Load the slog device's state from the MOS config 1766 * since it's possible that the label does not 1767 * contain the most up-to-date information. 1768 */ 1769 vdev_load_log_state(tvd, mtvd); 1770 vdev_reopen(tvd); 1771 } 1772 } 1773 vdev_free(mrvd); 1774 spa_config_exit(spa, SCL_ALL, FTAG); 1775 1776 /* 1777 * Ensure we were able to validate the config. 1778 */ 1779 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1780} 1781 1782/* 1783 * Check for missing log devices 1784 */ 1785static boolean_t 1786spa_check_logs(spa_t *spa) 1787{ 1788 boolean_t rv = B_FALSE; 1789 1790 switch (spa->spa_log_state) { 1791 case SPA_LOG_MISSING: 1792 /* need to recheck in case slog has been restored */ 1793 case SPA_LOG_UNKNOWN: 1794 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1795 NULL, DS_FIND_CHILDREN) != 0); 1796 if (rv) 1797 spa_set_log_state(spa, SPA_LOG_MISSING); 1798 break; 1799 } 1800 return (rv); 1801} 1802 1803static boolean_t 1804spa_passivate_log(spa_t *spa) 1805{ 1806 vdev_t *rvd = spa->spa_root_vdev; 1807 boolean_t slog_found = B_FALSE; 1808 1809 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1810 1811 if (!spa_has_slogs(spa)) 1812 return (B_FALSE); 1813 1814 for (int c = 0; c < rvd->vdev_children; c++) { 1815 vdev_t *tvd = rvd->vdev_child[c]; 1816 metaslab_group_t *mg = tvd->vdev_mg; 1817 1818 if (tvd->vdev_islog) { 1819 metaslab_group_passivate(mg); 1820 slog_found = B_TRUE; 1821 } 1822 } 1823 1824 return (slog_found); 1825} 1826 1827static void 1828spa_activate_log(spa_t *spa) 1829{ 1830 vdev_t *rvd = spa->spa_root_vdev; 1831 1832 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1833 1834 for (int c = 0; c < rvd->vdev_children; c++) { 1835 vdev_t *tvd = rvd->vdev_child[c]; 1836 metaslab_group_t *mg = tvd->vdev_mg; 1837 1838 if (tvd->vdev_islog) 1839 metaslab_group_activate(mg); 1840 } 1841} 1842 1843int 1844spa_offline_log(spa_t *spa) 1845{ 1846 int error; 1847 1848 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1849 NULL, DS_FIND_CHILDREN); 1850 if (error == 0) { 1851 /* 1852 * We successfully offlined the log device, sync out the 1853 * current txg so that the "stubby" block can be removed 1854 * by zil_sync(). 1855 */ 1856 txg_wait_synced(spa->spa_dsl_pool, 0); 1857 } 1858 return (error); 1859} 1860 1861static void 1862spa_aux_check_removed(spa_aux_vdev_t *sav) 1863{ 1864 int i; 1865 1866 for (i = 0; i < sav->sav_count; i++) 1867 spa_check_removed(sav->sav_vdevs[i]); 1868} 1869 1870void 1871spa_claim_notify(zio_t *zio) 1872{ 1873 spa_t *spa = zio->io_spa; 1874 1875 if (zio->io_error) 1876 return; 1877 1878 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1879 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1880 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1881 mutex_exit(&spa->spa_props_lock); 1882} 1883 1884typedef struct spa_load_error { 1885 uint64_t sle_meta_count; 1886 uint64_t sle_data_count; 1887} spa_load_error_t; 1888 1889static void 1890spa_load_verify_done(zio_t *zio) 1891{ 1892 blkptr_t *bp = zio->io_bp; 1893 spa_load_error_t *sle = zio->io_private; 1894 dmu_object_type_t type = BP_GET_TYPE(bp); 1895 int error = zio->io_error; 1896 spa_t *spa = zio->io_spa; 1897 1898 if (error) { 1899 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1900 type != DMU_OT_INTENT_LOG) 1901 atomic_inc_64(&sle->sle_meta_count); 1902 else 1903 atomic_inc_64(&sle->sle_data_count); 1904 } 1905 zio_data_buf_free(zio->io_data, zio->io_size); 1906 1907 mutex_enter(&spa->spa_scrub_lock); 1908 spa->spa_scrub_inflight--; 1909 cv_broadcast(&spa->spa_scrub_io_cv); 1910 mutex_exit(&spa->spa_scrub_lock); 1911} 1912 1913/* 1914 * Maximum number of concurrent scrub i/os to create while verifying 1915 * a pool while importing it. 1916 */ 1917int spa_load_verify_maxinflight = 10000; 1918boolean_t spa_load_verify_metadata = B_TRUE; 1919boolean_t spa_load_verify_data = B_TRUE; 1920 1921SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN, 1922 &spa_load_verify_maxinflight, 0, 1923 "Maximum number of concurrent scrub I/Os to create while verifying a " 1924 "pool while importing it"); 1925 1926SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN, 1927 &spa_load_verify_metadata, 0, 1928 "Check metadata on import?"); 1929 1930SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN, 1931 &spa_load_verify_data, 0, 1932 "Check user data on import?"); 1933 1934/*ARGSUSED*/ 1935static int 1936spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1937 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg) 1938{ 1939 if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp)) 1940 return (0); 1941 /* 1942 * Note: normally this routine will not be called if 1943 * spa_load_verify_metadata is not set. However, it may be useful 1944 * to manually set the flag after the traversal has begun. 1945 */ 1946 if (!spa_load_verify_metadata) 1947 return (0); 1948 if (BP_GET_BUFC_TYPE(bp) == ARC_BUFC_DATA && !spa_load_verify_data) 1949 return (0); 1950 1951 zio_t *rio = arg; 1952 size_t size = BP_GET_PSIZE(bp); 1953 void *data = zio_data_buf_alloc(size); 1954 1955 mutex_enter(&spa->spa_scrub_lock); 1956 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight) 1957 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); 1958 spa->spa_scrub_inflight++; 1959 mutex_exit(&spa->spa_scrub_lock); 1960 1961 zio_nowait(zio_read(rio, spa, bp, data, size, 1962 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1963 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1964 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1965 return (0); 1966} 1967 1968static int 1969spa_load_verify(spa_t *spa) 1970{ 1971 zio_t *rio; 1972 spa_load_error_t sle = { 0 }; 1973 zpool_rewind_policy_t policy; 1974 boolean_t verify_ok = B_FALSE; 1975 int error = 0; 1976 1977 zpool_get_rewind_policy(spa->spa_config, &policy); 1978 1979 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1980 return (0); 1981 1982 rio = zio_root(spa, NULL, &sle, 1983 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1984 1985 if (spa_load_verify_metadata) { 1986 error = traverse_pool(spa, spa->spa_verify_min_txg, 1987 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, 1988 spa_load_verify_cb, rio); 1989 } 1990 1991 (void) zio_wait(rio); 1992 1993 spa->spa_load_meta_errors = sle.sle_meta_count; 1994 spa->spa_load_data_errors = sle.sle_data_count; 1995 1996 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1997 sle.sle_data_count <= policy.zrp_maxdata) { 1998 int64_t loss = 0; 1999 2000 verify_ok = B_TRUE; 2001 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 2002 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 2003 2004 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 2005 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2006 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 2007 VERIFY(nvlist_add_int64(spa->spa_load_info, 2008 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 2009 VERIFY(nvlist_add_uint64(spa->spa_load_info, 2010 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 2011 } else { 2012 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 2013 } 2014 2015 if (error) { 2016 if (error != ENXIO && error != EIO) 2017 error = SET_ERROR(EIO); 2018 return (error); 2019 } 2020 2021 return (verify_ok ? 0 : EIO); 2022} 2023 2024/* 2025 * Find a value in the pool props object. 2026 */ 2027static void 2028spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 2029{ 2030 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 2031 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 2032} 2033 2034/* 2035 * Find a value in the pool directory object. 2036 */ 2037static int 2038spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 2039{ 2040 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 2041 name, sizeof (uint64_t), 1, val)); 2042} 2043 2044static int 2045spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 2046{ 2047 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 2048 return (err); 2049} 2050 2051/* 2052 * Fix up config after a partly-completed split. This is done with the 2053 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 2054 * pool have that entry in their config, but only the splitting one contains 2055 * a list of all the guids of the vdevs that are being split off. 2056 * 2057 * This function determines what to do with that list: either rejoin 2058 * all the disks to the pool, or complete the splitting process. To attempt 2059 * the rejoin, each disk that is offlined is marked online again, and 2060 * we do a reopen() call. If the vdev label for every disk that was 2061 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 2062 * then we call vdev_split() on each disk, and complete the split. 2063 * 2064 * Otherwise we leave the config alone, with all the vdevs in place in 2065 * the original pool. 2066 */ 2067static void 2068spa_try_repair(spa_t *spa, nvlist_t *config) 2069{ 2070 uint_t extracted; 2071 uint64_t *glist; 2072 uint_t i, gcount; 2073 nvlist_t *nvl; 2074 vdev_t **vd; 2075 boolean_t attempt_reopen; 2076 2077 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2078 return; 2079 2080 /* check that the config is complete */ 2081 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2082 &glist, &gcount) != 0) 2083 return; 2084 2085 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2086 2087 /* attempt to online all the vdevs & validate */ 2088 attempt_reopen = B_TRUE; 2089 for (i = 0; i < gcount; i++) { 2090 if (glist[i] == 0) /* vdev is hole */ 2091 continue; 2092 2093 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2094 if (vd[i] == NULL) { 2095 /* 2096 * Don't bother attempting to reopen the disks; 2097 * just do the split. 2098 */ 2099 attempt_reopen = B_FALSE; 2100 } else { 2101 /* attempt to re-online it */ 2102 vd[i]->vdev_offline = B_FALSE; 2103 } 2104 } 2105 2106 if (attempt_reopen) { 2107 vdev_reopen(spa->spa_root_vdev); 2108 2109 /* check each device to see what state it's in */ 2110 for (extracted = 0, i = 0; i < gcount; i++) { 2111 if (vd[i] != NULL && 2112 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2113 break; 2114 ++extracted; 2115 } 2116 } 2117 2118 /* 2119 * If every disk has been moved to the new pool, or if we never 2120 * even attempted to look at them, then we split them off for 2121 * good. 2122 */ 2123 if (!attempt_reopen || gcount == extracted) { 2124 for (i = 0; i < gcount; i++) 2125 if (vd[i] != NULL) 2126 vdev_split(vd[i]); 2127 vdev_reopen(spa->spa_root_vdev); 2128 } 2129 2130 kmem_free(vd, gcount * sizeof (vdev_t *)); 2131} 2132 2133static int 2134spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2135 boolean_t mosconfig) 2136{ 2137 nvlist_t *config = spa->spa_config; 2138 char *ereport = FM_EREPORT_ZFS_POOL; 2139 char *comment; 2140 int error; 2141 uint64_t pool_guid; 2142 nvlist_t *nvl; 2143 2144 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2145 return (SET_ERROR(EINVAL)); 2146 2147 ASSERT(spa->spa_comment == NULL); 2148 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2149 spa->spa_comment = spa_strdup(comment); 2150 2151 /* 2152 * Versioning wasn't explicitly added to the label until later, so if 2153 * it's not present treat it as the initial version. 2154 */ 2155 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2156 &spa->spa_ubsync.ub_version) != 0) 2157 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2158 2159 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2160 &spa->spa_config_txg); 2161 2162 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2163 spa_guid_exists(pool_guid, 0)) { 2164 error = SET_ERROR(EEXIST); 2165 } else { 2166 spa->spa_config_guid = pool_guid; 2167 2168 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2169 &nvl) == 0) { 2170 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2171 KM_SLEEP) == 0); 2172 } 2173 2174 nvlist_free(spa->spa_load_info); 2175 spa->spa_load_info = fnvlist_alloc(); 2176 2177 gethrestime(&spa->spa_loaded_ts); 2178 error = spa_load_impl(spa, pool_guid, config, state, type, 2179 mosconfig, &ereport); 2180 } 2181 2182 spa->spa_minref = refcount_count(&spa->spa_refcount); 2183 if (error) { 2184 if (error != EEXIST) { 2185 spa->spa_loaded_ts.tv_sec = 0; 2186 spa->spa_loaded_ts.tv_nsec = 0; 2187 } 2188 if (error != EBADF) { 2189 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2190 } 2191 } 2192 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2193 spa->spa_ena = 0; 2194 2195 return (error); 2196} 2197 2198/* 2199 * Load an existing storage pool, using the pool's builtin spa_config as a 2200 * source of configuration information. 2201 */ 2202static int 2203spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2204 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2205 char **ereport) 2206{ 2207 int error = 0; 2208 nvlist_t *nvroot = NULL; 2209 nvlist_t *label; 2210 vdev_t *rvd; 2211 uberblock_t *ub = &spa->spa_uberblock; 2212 uint64_t children, config_cache_txg = spa->spa_config_txg; 2213 int orig_mode = spa->spa_mode; 2214 int parse; 2215 uint64_t obj; 2216 boolean_t missing_feat_write = B_FALSE; 2217 2218 /* 2219 * If this is an untrusted config, access the pool in read-only mode. 2220 * This prevents things like resilvering recently removed devices. 2221 */ 2222 if (!mosconfig) 2223 spa->spa_mode = FREAD; 2224 2225 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2226 2227 spa->spa_load_state = state; 2228 2229 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2230 return (SET_ERROR(EINVAL)); 2231 2232 parse = (type == SPA_IMPORT_EXISTING ? 2233 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2234 2235 /* 2236 * Create "The Godfather" zio to hold all async IOs 2237 */ 2238 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 2239 KM_SLEEP); 2240 for (int i = 0; i < max_ncpus; i++) { 2241 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 2242 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 2243 ZIO_FLAG_GODFATHER); 2244 } 2245 2246 /* 2247 * Parse the configuration into a vdev tree. We explicitly set the 2248 * value that will be returned by spa_version() since parsing the 2249 * configuration requires knowing the version number. 2250 */ 2251 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2252 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2253 spa_config_exit(spa, SCL_ALL, FTAG); 2254 2255 if (error != 0) 2256 return (error); 2257 2258 ASSERT(spa->spa_root_vdev == rvd); 2259 2260 if (type != SPA_IMPORT_ASSEMBLE) { 2261 ASSERT(spa_guid(spa) == pool_guid); 2262 } 2263 2264 /* 2265 * Try to open all vdevs, loading each label in the process. 2266 */ 2267 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2268 error = vdev_open(rvd); 2269 spa_config_exit(spa, SCL_ALL, FTAG); 2270 if (error != 0) 2271 return (error); 2272 2273 /* 2274 * We need to validate the vdev labels against the configuration that 2275 * we have in hand, which is dependent on the setting of mosconfig. If 2276 * mosconfig is true then we're validating the vdev labels based on 2277 * that config. Otherwise, we're validating against the cached config 2278 * (zpool.cache) that was read when we loaded the zfs module, and then 2279 * later we will recursively call spa_load() and validate against 2280 * the vdev config. 2281 * 2282 * If we're assembling a new pool that's been split off from an 2283 * existing pool, the labels haven't yet been updated so we skip 2284 * validation for now. 2285 */ 2286 if (type != SPA_IMPORT_ASSEMBLE) { 2287 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2288 error = vdev_validate(rvd, mosconfig); 2289 spa_config_exit(spa, SCL_ALL, FTAG); 2290 2291 if (error != 0) 2292 return (error); 2293 2294 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2295 return (SET_ERROR(ENXIO)); 2296 } 2297 2298 /* 2299 * Find the best uberblock. 2300 */ 2301 vdev_uberblock_load(rvd, ub, &label); 2302 2303 /* 2304 * If we weren't able to find a single valid uberblock, return failure. 2305 */ 2306 if (ub->ub_txg == 0) { 2307 nvlist_free(label); 2308 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2309 } 2310 2311 /* 2312 * If the pool has an unsupported version we can't open it. 2313 */ 2314 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2315 nvlist_free(label); 2316 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2317 } 2318 2319 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2320 nvlist_t *features; 2321 2322 /* 2323 * If we weren't able to find what's necessary for reading the 2324 * MOS in the label, return failure. 2325 */ 2326 if (label == NULL || nvlist_lookup_nvlist(label, 2327 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2328 nvlist_free(label); 2329 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2330 ENXIO)); 2331 } 2332 2333 /* 2334 * Update our in-core representation with the definitive values 2335 * from the label. 2336 */ 2337 nvlist_free(spa->spa_label_features); 2338 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2339 } 2340 2341 nvlist_free(label); 2342 2343 /* 2344 * Look through entries in the label nvlist's features_for_read. If 2345 * there is a feature listed there which we don't understand then we 2346 * cannot open a pool. 2347 */ 2348 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2349 nvlist_t *unsup_feat; 2350 2351 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2352 0); 2353 2354 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2355 NULL); nvp != NULL; 2356 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2357 if (!zfeature_is_supported(nvpair_name(nvp))) { 2358 VERIFY(nvlist_add_string(unsup_feat, 2359 nvpair_name(nvp), "") == 0); 2360 } 2361 } 2362 2363 if (!nvlist_empty(unsup_feat)) { 2364 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2365 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2366 nvlist_free(unsup_feat); 2367 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2368 ENOTSUP)); 2369 } 2370 2371 nvlist_free(unsup_feat); 2372 } 2373 2374 /* 2375 * If the vdev guid sum doesn't match the uberblock, we have an 2376 * incomplete configuration. We first check to see if the pool 2377 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2378 * If it is, defer the vdev_guid_sum check till later so we 2379 * can handle missing vdevs. 2380 */ 2381 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2382 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2383 rvd->vdev_guid_sum != ub->ub_guid_sum) 2384 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2385 2386 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2387 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2388 spa_try_repair(spa, config); 2389 spa_config_exit(spa, SCL_ALL, FTAG); 2390 nvlist_free(spa->spa_config_splitting); 2391 spa->spa_config_splitting = NULL; 2392 } 2393 2394 /* 2395 * Initialize internal SPA structures. 2396 */ 2397 spa->spa_state = POOL_STATE_ACTIVE; 2398 spa->spa_ubsync = spa->spa_uberblock; 2399 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2400 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2401 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2402 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2403 spa->spa_claim_max_txg = spa->spa_first_txg; 2404 spa->spa_prev_software_version = ub->ub_software_version; 2405 2406 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2407 if (error) 2408 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2409 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2410 2411 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2413 2414 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2415 boolean_t missing_feat_read = B_FALSE; 2416 nvlist_t *unsup_feat, *enabled_feat; 2417 2418 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2419 &spa->spa_feat_for_read_obj) != 0) { 2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2421 } 2422 2423 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2424 &spa->spa_feat_for_write_obj) != 0) { 2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2426 } 2427 2428 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2429 &spa->spa_feat_desc_obj) != 0) { 2430 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2431 } 2432 2433 enabled_feat = fnvlist_alloc(); 2434 unsup_feat = fnvlist_alloc(); 2435 2436 if (!spa_features_check(spa, B_FALSE, 2437 unsup_feat, enabled_feat)) 2438 missing_feat_read = B_TRUE; 2439 2440 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2441 if (!spa_features_check(spa, B_TRUE, 2442 unsup_feat, enabled_feat)) { 2443 missing_feat_write = B_TRUE; 2444 } 2445 } 2446 2447 fnvlist_add_nvlist(spa->spa_load_info, 2448 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2449 2450 if (!nvlist_empty(unsup_feat)) { 2451 fnvlist_add_nvlist(spa->spa_load_info, 2452 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2453 } 2454 2455 fnvlist_free(enabled_feat); 2456 fnvlist_free(unsup_feat); 2457 2458 if (!missing_feat_read) { 2459 fnvlist_add_boolean(spa->spa_load_info, 2460 ZPOOL_CONFIG_CAN_RDONLY); 2461 } 2462 2463 /* 2464 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2465 * twofold: to determine whether the pool is available for 2466 * import in read-write mode and (if it is not) whether the 2467 * pool is available for import in read-only mode. If the pool 2468 * is available for import in read-write mode, it is displayed 2469 * as available in userland; if it is not available for import 2470 * in read-only mode, it is displayed as unavailable in 2471 * userland. If the pool is available for import in read-only 2472 * mode but not read-write mode, it is displayed as unavailable 2473 * in userland with a special note that the pool is actually 2474 * available for open in read-only mode. 2475 * 2476 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2477 * missing a feature for write, we must first determine whether 2478 * the pool can be opened read-only before returning to 2479 * userland in order to know whether to display the 2480 * abovementioned note. 2481 */ 2482 if (missing_feat_read || (missing_feat_write && 2483 spa_writeable(spa))) { 2484 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2485 ENOTSUP)); 2486 } 2487 2488 /* 2489 * Load refcounts for ZFS features from disk into an in-memory 2490 * cache during SPA initialization. 2491 */ 2492 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) { 2493 uint64_t refcount; 2494 2495 error = feature_get_refcount_from_disk(spa, 2496 &spa_feature_table[i], &refcount); 2497 if (error == 0) { 2498 spa->spa_feat_refcount_cache[i] = refcount; 2499 } else if (error == ENOTSUP) { 2500 spa->spa_feat_refcount_cache[i] = 2501 SPA_FEATURE_DISABLED; 2502 } else { 2503 return (spa_vdev_err(rvd, 2504 VDEV_AUX_CORRUPT_DATA, EIO)); 2505 } 2506 } 2507 } 2508 2509 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) { 2510 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG, 2511 &spa->spa_feat_enabled_txg_obj) != 0) 2512 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2513 } 2514 2515 spa->spa_is_initializing = B_TRUE; 2516 error = dsl_pool_open(spa->spa_dsl_pool); 2517 spa->spa_is_initializing = B_FALSE; 2518 if (error != 0) 2519 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2520 2521 if (!mosconfig) { 2522 uint64_t hostid; 2523 nvlist_t *policy = NULL, *nvconfig; 2524 2525 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2526 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2527 2528 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2529 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2530 char *hostname; 2531 unsigned long myhostid = 0; 2532 2533 VERIFY(nvlist_lookup_string(nvconfig, 2534 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2535 2536#ifdef _KERNEL 2537 myhostid = zone_get_hostid(NULL); 2538#else /* _KERNEL */ 2539 /* 2540 * We're emulating the system's hostid in userland, so 2541 * we can't use zone_get_hostid(). 2542 */ 2543 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2544#endif /* _KERNEL */ 2545 if (check_hostid && hostid != 0 && myhostid != 0 && 2546 hostid != myhostid) { 2547 nvlist_free(nvconfig); 2548 cmn_err(CE_WARN, "pool '%s' could not be " 2549 "loaded as it was last accessed by " 2550 "another system (host: %s hostid: 0x%lx). " 2551 "See: http://illumos.org/msg/ZFS-8000-EY", 2552 spa_name(spa), hostname, 2553 (unsigned long)hostid); 2554 return (SET_ERROR(EBADF)); 2555 } 2556 } 2557 if (nvlist_lookup_nvlist(spa->spa_config, 2558 ZPOOL_REWIND_POLICY, &policy) == 0) 2559 VERIFY(nvlist_add_nvlist(nvconfig, 2560 ZPOOL_REWIND_POLICY, policy) == 0); 2561 2562 spa_config_set(spa, nvconfig); 2563 spa_unload(spa); 2564 spa_deactivate(spa); 2565 spa_activate(spa, orig_mode); 2566 2567 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2568 } 2569 2570 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2571 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2572 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2573 if (error != 0) 2574 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2575 2576 /* 2577 * Load the bit that tells us to use the new accounting function 2578 * (raid-z deflation). If we have an older pool, this will not 2579 * be present. 2580 */ 2581 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2582 if (error != 0 && error != ENOENT) 2583 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2584 2585 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2586 &spa->spa_creation_version); 2587 if (error != 0 && error != ENOENT) 2588 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2589 2590 /* 2591 * Load the persistent error log. If we have an older pool, this will 2592 * not be present. 2593 */ 2594 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2595 if (error != 0 && error != ENOENT) 2596 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2597 2598 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2599 &spa->spa_errlog_scrub); 2600 if (error != 0 && error != ENOENT) 2601 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2602 2603 /* 2604 * Load the history object. If we have an older pool, this 2605 * will not be present. 2606 */ 2607 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2608 if (error != 0 && error != ENOENT) 2609 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2610 2611 /* 2612 * If we're assembling the pool from the split-off vdevs of 2613 * an existing pool, we don't want to attach the spares & cache 2614 * devices. 2615 */ 2616 2617 /* 2618 * Load any hot spares for this pool. 2619 */ 2620 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2621 if (error != 0 && error != ENOENT) 2622 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2623 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2624 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2625 if (load_nvlist(spa, spa->spa_spares.sav_object, 2626 &spa->spa_spares.sav_config) != 0) 2627 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2628 2629 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2630 spa_load_spares(spa); 2631 spa_config_exit(spa, SCL_ALL, FTAG); 2632 } else if (error == 0) { 2633 spa->spa_spares.sav_sync = B_TRUE; 2634 } 2635 2636 /* 2637 * Load any level 2 ARC devices for this pool. 2638 */ 2639 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2640 &spa->spa_l2cache.sav_object); 2641 if (error != 0 && error != ENOENT) 2642 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2643 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2644 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2645 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2646 &spa->spa_l2cache.sav_config) != 0) 2647 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2648 2649 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2650 spa_load_l2cache(spa); 2651 spa_config_exit(spa, SCL_ALL, FTAG); 2652 } else if (error == 0) { 2653 spa->spa_l2cache.sav_sync = B_TRUE; 2654 } 2655 2656 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2657 2658 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2659 if (error && error != ENOENT) 2660 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2661 2662 if (error == 0) { 2663 uint64_t autoreplace; 2664 2665 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2666 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2667 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2668 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2669 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2670 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2671 &spa->spa_dedup_ditto); 2672 2673 spa->spa_autoreplace = (autoreplace != 0); 2674 } 2675 2676 /* 2677 * If the 'autoreplace' property is set, then post a resource notifying 2678 * the ZFS DE that it should not issue any faults for unopenable 2679 * devices. We also iterate over the vdevs, and post a sysevent for any 2680 * unopenable vdevs so that the normal autoreplace handler can take 2681 * over. 2682 */ 2683 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2684 spa_check_removed(spa->spa_root_vdev); 2685 /* 2686 * For the import case, this is done in spa_import(), because 2687 * at this point we're using the spare definitions from 2688 * the MOS config, not necessarily from the userland config. 2689 */ 2690 if (state != SPA_LOAD_IMPORT) { 2691 spa_aux_check_removed(&spa->spa_spares); 2692 spa_aux_check_removed(&spa->spa_l2cache); 2693 } 2694 } 2695 2696 /* 2697 * Load the vdev state for all toplevel vdevs. 2698 */ 2699 vdev_load(rvd); 2700 2701 /* 2702 * Propagate the leaf DTLs we just loaded all the way up the tree. 2703 */ 2704 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2705 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2706 spa_config_exit(spa, SCL_ALL, FTAG); 2707 2708 /* 2709 * Load the DDTs (dedup tables). 2710 */ 2711 error = ddt_load(spa); 2712 if (error != 0) 2713 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2714 2715 spa_update_dspace(spa); 2716 2717 /* 2718 * Validate the config, using the MOS config to fill in any 2719 * information which might be missing. If we fail to validate 2720 * the config then declare the pool unfit for use. If we're 2721 * assembling a pool from a split, the log is not transferred 2722 * over. 2723 */ 2724 if (type != SPA_IMPORT_ASSEMBLE) { 2725 nvlist_t *nvconfig; 2726 2727 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2728 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2729 2730 if (!spa_config_valid(spa, nvconfig)) { 2731 nvlist_free(nvconfig); 2732 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2733 ENXIO)); 2734 } 2735 nvlist_free(nvconfig); 2736 2737 /* 2738 * Now that we've validated the config, check the state of the 2739 * root vdev. If it can't be opened, it indicates one or 2740 * more toplevel vdevs are faulted. 2741 */ 2742 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2743 return (SET_ERROR(ENXIO)); 2744 2745 if (spa_check_logs(spa)) { 2746 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2747 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2748 } 2749 } 2750 2751 if (missing_feat_write) { 2752 ASSERT(state == SPA_LOAD_TRYIMPORT); 2753 2754 /* 2755 * At this point, we know that we can open the pool in 2756 * read-only mode but not read-write mode. We now have enough 2757 * information and can return to userland. 2758 */ 2759 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2760 } 2761 2762 /* 2763 * We've successfully opened the pool, verify that we're ready 2764 * to start pushing transactions. 2765 */ 2766 if (state != SPA_LOAD_TRYIMPORT) { 2767 if (error = spa_load_verify(spa)) 2768 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2769 error)); 2770 } 2771 2772 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2773 spa->spa_load_max_txg == UINT64_MAX)) { 2774 dmu_tx_t *tx; 2775 int need_update = B_FALSE; 2776 2777 ASSERT(state != SPA_LOAD_TRYIMPORT); 2778 2779 /* 2780 * Claim log blocks that haven't been committed yet. 2781 * This must all happen in a single txg. 2782 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2783 * invoked from zil_claim_log_block()'s i/o done callback. 2784 * Price of rollback is that we abandon the log. 2785 */ 2786 spa->spa_claiming = B_TRUE; 2787 2788 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2789 spa_first_txg(spa)); 2790 (void) dmu_objset_find(spa_name(spa), 2791 zil_claim, tx, DS_FIND_CHILDREN); 2792 dmu_tx_commit(tx); 2793 2794 spa->spa_claiming = B_FALSE; 2795 2796 spa_set_log_state(spa, SPA_LOG_GOOD); 2797 spa->spa_sync_on = B_TRUE; 2798 txg_sync_start(spa->spa_dsl_pool); 2799 2800 /* 2801 * Wait for all claims to sync. We sync up to the highest 2802 * claimed log block birth time so that claimed log blocks 2803 * don't appear to be from the future. spa_claim_max_txg 2804 * will have been set for us by either zil_check_log_chain() 2805 * (invoked from spa_check_logs()) or zil_claim() above. 2806 */ 2807 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2808 2809 /* 2810 * If the config cache is stale, or we have uninitialized 2811 * metaslabs (see spa_vdev_add()), then update the config. 2812 * 2813 * If this is a verbatim import, trust the current 2814 * in-core spa_config and update the disk labels. 2815 */ 2816 if (config_cache_txg != spa->spa_config_txg || 2817 state == SPA_LOAD_IMPORT || 2818 state == SPA_LOAD_RECOVER || 2819 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2820 need_update = B_TRUE; 2821 2822 for (int c = 0; c < rvd->vdev_children; c++) 2823 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2824 need_update = B_TRUE; 2825 2826 /* 2827 * Update the config cache asychronously in case we're the 2828 * root pool, in which case the config cache isn't writable yet. 2829 */ 2830 if (need_update) 2831 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2832 2833 /* 2834 * Check all DTLs to see if anything needs resilvering. 2835 */ 2836 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2837 vdev_resilver_needed(rvd, NULL, NULL)) 2838 spa_async_request(spa, SPA_ASYNC_RESILVER); 2839 2840 /* 2841 * Log the fact that we booted up (so that we can detect if 2842 * we rebooted in the middle of an operation). 2843 */ 2844 spa_history_log_version(spa, "open"); 2845 2846 /* 2847 * Delete any inconsistent datasets. 2848 */ 2849 (void) dmu_objset_find(spa_name(spa), 2850 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2851 2852 /* 2853 * Clean up any stale temporary dataset userrefs. 2854 */ 2855 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2856 } 2857 2858 return (0); 2859} 2860 2861static int 2862spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2863{ 2864 int mode = spa->spa_mode; 2865 2866 spa_unload(spa); 2867 spa_deactivate(spa); 2868 2869 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1; 2870 2871 spa_activate(spa, mode); 2872 spa_async_suspend(spa); 2873 2874 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2875} 2876 2877/* 2878 * If spa_load() fails this function will try loading prior txg's. If 2879 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2880 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2881 * function will not rewind the pool and will return the same error as 2882 * spa_load(). 2883 */ 2884static int 2885spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2886 uint64_t max_request, int rewind_flags) 2887{ 2888 nvlist_t *loadinfo = NULL; 2889 nvlist_t *config = NULL; 2890 int load_error, rewind_error; 2891 uint64_t safe_rewind_txg; 2892 uint64_t min_txg; 2893 2894 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2895 spa->spa_load_max_txg = spa->spa_load_txg; 2896 spa_set_log_state(spa, SPA_LOG_CLEAR); 2897 } else { 2898 spa->spa_load_max_txg = max_request; 2899 if (max_request != UINT64_MAX) 2900 spa->spa_extreme_rewind = B_TRUE; 2901 } 2902 2903 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2904 mosconfig); 2905 if (load_error == 0) 2906 return (0); 2907 2908 if (spa->spa_root_vdev != NULL) 2909 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2910 2911 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2912 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2913 2914 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2915 nvlist_free(config); 2916 return (load_error); 2917 } 2918 2919 if (state == SPA_LOAD_RECOVER) { 2920 /* Price of rolling back is discarding txgs, including log */ 2921 spa_set_log_state(spa, SPA_LOG_CLEAR); 2922 } else { 2923 /* 2924 * If we aren't rolling back save the load info from our first 2925 * import attempt so that we can restore it after attempting 2926 * to rewind. 2927 */ 2928 loadinfo = spa->spa_load_info; 2929 spa->spa_load_info = fnvlist_alloc(); 2930 } 2931 2932 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2933 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2934 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2935 TXG_INITIAL : safe_rewind_txg; 2936 2937 /* 2938 * Continue as long as we're finding errors, we're still within 2939 * the acceptable rewind range, and we're still finding uberblocks 2940 */ 2941 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2942 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2943 if (spa->spa_load_max_txg < safe_rewind_txg) 2944 spa->spa_extreme_rewind = B_TRUE; 2945 rewind_error = spa_load_retry(spa, state, mosconfig); 2946 } 2947 2948 spa->spa_extreme_rewind = B_FALSE; 2949 spa->spa_load_max_txg = UINT64_MAX; 2950 2951 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2952 spa_config_set(spa, config); 2953 2954 if (state == SPA_LOAD_RECOVER) { 2955 ASSERT3P(loadinfo, ==, NULL); 2956 return (rewind_error); 2957 } else { 2958 /* Store the rewind info as part of the initial load info */ 2959 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2960 spa->spa_load_info); 2961 2962 /* Restore the initial load info */ 2963 fnvlist_free(spa->spa_load_info); 2964 spa->spa_load_info = loadinfo; 2965 2966 return (load_error); 2967 } 2968} 2969 2970/* 2971 * Pool Open/Import 2972 * 2973 * The import case is identical to an open except that the configuration is sent 2974 * down from userland, instead of grabbed from the configuration cache. For the 2975 * case of an open, the pool configuration will exist in the 2976 * POOL_STATE_UNINITIALIZED state. 2977 * 2978 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2979 * the same time open the pool, without having to keep around the spa_t in some 2980 * ambiguous state. 2981 */ 2982static int 2983spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2984 nvlist_t **config) 2985{ 2986 spa_t *spa; 2987 spa_load_state_t state = SPA_LOAD_OPEN; 2988 int error; 2989 int locked = B_FALSE; 2990 int firstopen = B_FALSE; 2991 2992 *spapp = NULL; 2993 2994 /* 2995 * As disgusting as this is, we need to support recursive calls to this 2996 * function because dsl_dir_open() is called during spa_load(), and ends 2997 * up calling spa_open() again. The real fix is to figure out how to 2998 * avoid dsl_dir_open() calling this in the first place. 2999 */ 3000 if (mutex_owner(&spa_namespace_lock) != curthread) { 3001 mutex_enter(&spa_namespace_lock); 3002 locked = B_TRUE; 3003 } 3004 3005 if ((spa = spa_lookup(pool)) == NULL) { 3006 if (locked) 3007 mutex_exit(&spa_namespace_lock); 3008 return (SET_ERROR(ENOENT)); 3009 } 3010 3011 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 3012 zpool_rewind_policy_t policy; 3013 3014 firstopen = B_TRUE; 3015 3016 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 3017 &policy); 3018 if (policy.zrp_request & ZPOOL_DO_REWIND) 3019 state = SPA_LOAD_RECOVER; 3020 3021 spa_activate(spa, spa_mode_global); 3022 3023 if (state != SPA_LOAD_RECOVER) 3024 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3025 3026 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 3027 policy.zrp_request); 3028 3029 if (error == EBADF) { 3030 /* 3031 * If vdev_validate() returns failure (indicated by 3032 * EBADF), it indicates that one of the vdevs indicates 3033 * that the pool has been exported or destroyed. If 3034 * this is the case, the config cache is out of sync and 3035 * we should remove the pool from the namespace. 3036 */ 3037 spa_unload(spa); 3038 spa_deactivate(spa); 3039 spa_config_sync(spa, B_TRUE, B_TRUE); 3040 spa_remove(spa); 3041 if (locked) 3042 mutex_exit(&spa_namespace_lock); 3043 return (SET_ERROR(ENOENT)); 3044 } 3045 3046 if (error) { 3047 /* 3048 * We can't open the pool, but we still have useful 3049 * information: the state of each vdev after the 3050 * attempted vdev_open(). Return this to the user. 3051 */ 3052 if (config != NULL && spa->spa_config) { 3053 VERIFY(nvlist_dup(spa->spa_config, config, 3054 KM_SLEEP) == 0); 3055 VERIFY(nvlist_add_nvlist(*config, 3056 ZPOOL_CONFIG_LOAD_INFO, 3057 spa->spa_load_info) == 0); 3058 } 3059 spa_unload(spa); 3060 spa_deactivate(spa); 3061 spa->spa_last_open_failed = error; 3062 if (locked) 3063 mutex_exit(&spa_namespace_lock); 3064 *spapp = NULL; 3065 return (error); 3066 } 3067 } 3068 3069 spa_open_ref(spa, tag); 3070 3071 if (config != NULL) 3072 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3073 3074 /* 3075 * If we've recovered the pool, pass back any information we 3076 * gathered while doing the load. 3077 */ 3078 if (state == SPA_LOAD_RECOVER) { 3079 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 3080 spa->spa_load_info) == 0); 3081 } 3082 3083 if (locked) { 3084 spa->spa_last_open_failed = 0; 3085 spa->spa_last_ubsync_txg = 0; 3086 spa->spa_load_txg = 0; 3087 mutex_exit(&spa_namespace_lock); 3088#ifdef __FreeBSD__ 3089#ifdef _KERNEL 3090 if (firstopen) 3091 zvol_create_minors(spa->spa_name); 3092#endif 3093#endif 3094 } 3095 3096 *spapp = spa; 3097 3098 return (0); 3099} 3100 3101int 3102spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 3103 nvlist_t **config) 3104{ 3105 return (spa_open_common(name, spapp, tag, policy, config)); 3106} 3107 3108int 3109spa_open(const char *name, spa_t **spapp, void *tag) 3110{ 3111 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3112} 3113 3114/* 3115 * Lookup the given spa_t, incrementing the inject count in the process, 3116 * preventing it from being exported or destroyed. 3117 */ 3118spa_t * 3119spa_inject_addref(char *name) 3120{ 3121 spa_t *spa; 3122 3123 mutex_enter(&spa_namespace_lock); 3124 if ((spa = spa_lookup(name)) == NULL) { 3125 mutex_exit(&spa_namespace_lock); 3126 return (NULL); 3127 } 3128 spa->spa_inject_ref++; 3129 mutex_exit(&spa_namespace_lock); 3130 3131 return (spa); 3132} 3133 3134void 3135spa_inject_delref(spa_t *spa) 3136{ 3137 mutex_enter(&spa_namespace_lock); 3138 spa->spa_inject_ref--; 3139 mutex_exit(&spa_namespace_lock); 3140} 3141 3142/* 3143 * Add spares device information to the nvlist. 3144 */ 3145static void 3146spa_add_spares(spa_t *spa, nvlist_t *config) 3147{ 3148 nvlist_t **spares; 3149 uint_t i, nspares; 3150 nvlist_t *nvroot; 3151 uint64_t guid; 3152 vdev_stat_t *vs; 3153 uint_t vsc; 3154 uint64_t pool; 3155 3156 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3157 3158 if (spa->spa_spares.sav_count == 0) 3159 return; 3160 3161 VERIFY(nvlist_lookup_nvlist(config, 3162 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3163 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3164 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3165 if (nspares != 0) { 3166 VERIFY(nvlist_add_nvlist_array(nvroot, 3167 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3168 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3169 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3170 3171 /* 3172 * Go through and find any spares which have since been 3173 * repurposed as an active spare. If this is the case, update 3174 * their status appropriately. 3175 */ 3176 for (i = 0; i < nspares; i++) { 3177 VERIFY(nvlist_lookup_uint64(spares[i], 3178 ZPOOL_CONFIG_GUID, &guid) == 0); 3179 if (spa_spare_exists(guid, &pool, NULL) && 3180 pool != 0ULL) { 3181 VERIFY(nvlist_lookup_uint64_array( 3182 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3183 (uint64_t **)&vs, &vsc) == 0); 3184 vs->vs_state = VDEV_STATE_CANT_OPEN; 3185 vs->vs_aux = VDEV_AUX_SPARED; 3186 } 3187 } 3188 } 3189} 3190 3191/* 3192 * Add l2cache device information to the nvlist, including vdev stats. 3193 */ 3194static void 3195spa_add_l2cache(spa_t *spa, nvlist_t *config) 3196{ 3197 nvlist_t **l2cache; 3198 uint_t i, j, nl2cache; 3199 nvlist_t *nvroot; 3200 uint64_t guid; 3201 vdev_t *vd; 3202 vdev_stat_t *vs; 3203 uint_t vsc; 3204 3205 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3206 3207 if (spa->spa_l2cache.sav_count == 0) 3208 return; 3209 3210 VERIFY(nvlist_lookup_nvlist(config, 3211 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3212 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3213 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3214 if (nl2cache != 0) { 3215 VERIFY(nvlist_add_nvlist_array(nvroot, 3216 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3217 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3218 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3219 3220 /* 3221 * Update level 2 cache device stats. 3222 */ 3223 3224 for (i = 0; i < nl2cache; i++) { 3225 VERIFY(nvlist_lookup_uint64(l2cache[i], 3226 ZPOOL_CONFIG_GUID, &guid) == 0); 3227 3228 vd = NULL; 3229 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3230 if (guid == 3231 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3232 vd = spa->spa_l2cache.sav_vdevs[j]; 3233 break; 3234 } 3235 } 3236 ASSERT(vd != NULL); 3237 3238 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3239 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3240 == 0); 3241 vdev_get_stats(vd, vs); 3242 } 3243 } 3244} 3245 3246static void 3247spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3248{ 3249 nvlist_t *features; 3250 zap_cursor_t zc; 3251 zap_attribute_t za; 3252 3253 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3254 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3255 3256 /* We may be unable to read features if pool is suspended. */ 3257 if (spa_suspended(spa)) 3258 goto out; 3259 3260 if (spa->spa_feat_for_read_obj != 0) { 3261 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3262 spa->spa_feat_for_read_obj); 3263 zap_cursor_retrieve(&zc, &za) == 0; 3264 zap_cursor_advance(&zc)) { 3265 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3266 za.za_num_integers == 1); 3267 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3268 za.za_first_integer)); 3269 } 3270 zap_cursor_fini(&zc); 3271 } 3272 3273 if (spa->spa_feat_for_write_obj != 0) { 3274 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3275 spa->spa_feat_for_write_obj); 3276 zap_cursor_retrieve(&zc, &za) == 0; 3277 zap_cursor_advance(&zc)) { 3278 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3279 za.za_num_integers == 1); 3280 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3281 za.za_first_integer)); 3282 } 3283 zap_cursor_fini(&zc); 3284 } 3285 3286out: 3287 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3288 features) == 0); 3289 nvlist_free(features); 3290} 3291 3292int 3293spa_get_stats(const char *name, nvlist_t **config, 3294 char *altroot, size_t buflen) 3295{ 3296 int error; 3297 spa_t *spa; 3298 3299 *config = NULL; 3300 error = spa_open_common(name, &spa, FTAG, NULL, config); 3301 3302 if (spa != NULL) { 3303 /* 3304 * This still leaves a window of inconsistency where the spares 3305 * or l2cache devices could change and the config would be 3306 * self-inconsistent. 3307 */ 3308 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3309 3310 if (*config != NULL) { 3311 uint64_t loadtimes[2]; 3312 3313 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3314 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3315 VERIFY(nvlist_add_uint64_array(*config, 3316 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3317 3318 VERIFY(nvlist_add_uint64(*config, 3319 ZPOOL_CONFIG_ERRCOUNT, 3320 spa_get_errlog_size(spa)) == 0); 3321 3322 if (spa_suspended(spa)) 3323 VERIFY(nvlist_add_uint64(*config, 3324 ZPOOL_CONFIG_SUSPENDED, 3325 spa->spa_failmode) == 0); 3326 3327 spa_add_spares(spa, *config); 3328 spa_add_l2cache(spa, *config); 3329 spa_add_feature_stats(spa, *config); 3330 } 3331 } 3332 3333 /* 3334 * We want to get the alternate root even for faulted pools, so we cheat 3335 * and call spa_lookup() directly. 3336 */ 3337 if (altroot) { 3338 if (spa == NULL) { 3339 mutex_enter(&spa_namespace_lock); 3340 spa = spa_lookup(name); 3341 if (spa) 3342 spa_altroot(spa, altroot, buflen); 3343 else 3344 altroot[0] = '\0'; 3345 spa = NULL; 3346 mutex_exit(&spa_namespace_lock); 3347 } else { 3348 spa_altroot(spa, altroot, buflen); 3349 } 3350 } 3351 3352 if (spa != NULL) { 3353 spa_config_exit(spa, SCL_CONFIG, FTAG); 3354 spa_close(spa, FTAG); 3355 } 3356 3357 return (error); 3358} 3359 3360/* 3361 * Validate that the auxiliary device array is well formed. We must have an 3362 * array of nvlists, each which describes a valid leaf vdev. If this is an 3363 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3364 * specified, as long as they are well-formed. 3365 */ 3366static int 3367spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3368 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3369 vdev_labeltype_t label) 3370{ 3371 nvlist_t **dev; 3372 uint_t i, ndev; 3373 vdev_t *vd; 3374 int error; 3375 3376 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3377 3378 /* 3379 * It's acceptable to have no devs specified. 3380 */ 3381 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3382 return (0); 3383 3384 if (ndev == 0) 3385 return (SET_ERROR(EINVAL)); 3386 3387 /* 3388 * Make sure the pool is formatted with a version that supports this 3389 * device type. 3390 */ 3391 if (spa_version(spa) < version) 3392 return (SET_ERROR(ENOTSUP)); 3393 3394 /* 3395 * Set the pending device list so we correctly handle device in-use 3396 * checking. 3397 */ 3398 sav->sav_pending = dev; 3399 sav->sav_npending = ndev; 3400 3401 for (i = 0; i < ndev; i++) { 3402 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3403 mode)) != 0) 3404 goto out; 3405 3406 if (!vd->vdev_ops->vdev_op_leaf) { 3407 vdev_free(vd); 3408 error = SET_ERROR(EINVAL); 3409 goto out; 3410 } 3411 3412 /* 3413 * The L2ARC currently only supports disk devices in 3414 * kernel context. For user-level testing, we allow it. 3415 */ 3416#ifdef _KERNEL 3417 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3418 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3419 error = SET_ERROR(ENOTBLK); 3420 vdev_free(vd); 3421 goto out; 3422 } 3423#endif 3424 vd->vdev_top = vd; 3425 3426 if ((error = vdev_open(vd)) == 0 && 3427 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3428 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3429 vd->vdev_guid) == 0); 3430 } 3431 3432 vdev_free(vd); 3433 3434 if (error && 3435 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3436 goto out; 3437 else 3438 error = 0; 3439 } 3440 3441out: 3442 sav->sav_pending = NULL; 3443 sav->sav_npending = 0; 3444 return (error); 3445} 3446 3447static int 3448spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3449{ 3450 int error; 3451 3452 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3453 3454 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3455 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3456 VDEV_LABEL_SPARE)) != 0) { 3457 return (error); 3458 } 3459 3460 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3461 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3462 VDEV_LABEL_L2CACHE)); 3463} 3464 3465static void 3466spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3467 const char *config) 3468{ 3469 int i; 3470 3471 if (sav->sav_config != NULL) { 3472 nvlist_t **olddevs; 3473 uint_t oldndevs; 3474 nvlist_t **newdevs; 3475 3476 /* 3477 * Generate new dev list by concatentating with the 3478 * current dev list. 3479 */ 3480 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3481 &olddevs, &oldndevs) == 0); 3482 3483 newdevs = kmem_alloc(sizeof (void *) * 3484 (ndevs + oldndevs), KM_SLEEP); 3485 for (i = 0; i < oldndevs; i++) 3486 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3487 KM_SLEEP) == 0); 3488 for (i = 0; i < ndevs; i++) 3489 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3490 KM_SLEEP) == 0); 3491 3492 VERIFY(nvlist_remove(sav->sav_config, config, 3493 DATA_TYPE_NVLIST_ARRAY) == 0); 3494 3495 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3496 config, newdevs, ndevs + oldndevs) == 0); 3497 for (i = 0; i < oldndevs + ndevs; i++) 3498 nvlist_free(newdevs[i]); 3499 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3500 } else { 3501 /* 3502 * Generate a new dev list. 3503 */ 3504 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3505 KM_SLEEP) == 0); 3506 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3507 devs, ndevs) == 0); 3508 } 3509} 3510 3511/* 3512 * Stop and drop level 2 ARC devices 3513 */ 3514void 3515spa_l2cache_drop(spa_t *spa) 3516{ 3517 vdev_t *vd; 3518 int i; 3519 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3520 3521 for (i = 0; i < sav->sav_count; i++) { 3522 uint64_t pool; 3523 3524 vd = sav->sav_vdevs[i]; 3525 ASSERT(vd != NULL); 3526 3527 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3528 pool != 0ULL && l2arc_vdev_present(vd)) 3529 l2arc_remove_vdev(vd); 3530 } 3531} 3532 3533/* 3534 * Pool Creation 3535 */ 3536int 3537spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3538 nvlist_t *zplprops) 3539{ 3540 spa_t *spa; 3541 char *altroot = NULL; 3542 vdev_t *rvd; 3543 dsl_pool_t *dp; 3544 dmu_tx_t *tx; 3545 int error = 0; 3546 uint64_t txg = TXG_INITIAL; 3547 nvlist_t **spares, **l2cache; 3548 uint_t nspares, nl2cache; 3549 uint64_t version, obj; 3550 boolean_t has_features; 3551 3552 /* 3553 * If this pool already exists, return failure. 3554 */ 3555 mutex_enter(&spa_namespace_lock); 3556 if (spa_lookup(pool) != NULL) { 3557 mutex_exit(&spa_namespace_lock); 3558 return (SET_ERROR(EEXIST)); 3559 } 3560 3561 /* 3562 * Allocate a new spa_t structure. 3563 */ 3564 (void) nvlist_lookup_string(props, 3565 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3566 spa = spa_add(pool, NULL, altroot); 3567 spa_activate(spa, spa_mode_global); 3568 3569 if (props && (error = spa_prop_validate(spa, props))) { 3570 spa_deactivate(spa); 3571 spa_remove(spa); 3572 mutex_exit(&spa_namespace_lock); 3573 return (error); 3574 } 3575 3576 has_features = B_FALSE; 3577 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3578 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3579 if (zpool_prop_feature(nvpair_name(elem))) 3580 has_features = B_TRUE; 3581 } 3582 3583 if (has_features || nvlist_lookup_uint64(props, 3584 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3585 version = SPA_VERSION; 3586 } 3587 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3588 3589 spa->spa_first_txg = txg; 3590 spa->spa_uberblock.ub_txg = txg - 1; 3591 spa->spa_uberblock.ub_version = version; 3592 spa->spa_ubsync = spa->spa_uberblock; 3593 3594 /* 3595 * Create "The Godfather" zio to hold all async IOs 3596 */ 3597 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *), 3598 KM_SLEEP); 3599 for (int i = 0; i < max_ncpus; i++) { 3600 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL, 3601 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | 3602 ZIO_FLAG_GODFATHER); 3603 } 3604 3605 /* 3606 * Create the root vdev. 3607 */ 3608 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3609 3610 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3611 3612 ASSERT(error != 0 || rvd != NULL); 3613 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3614 3615 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3616 error = SET_ERROR(EINVAL); 3617 3618 if (error == 0 && 3619 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3620 (error = spa_validate_aux(spa, nvroot, txg, 3621 VDEV_ALLOC_ADD)) == 0) { 3622 for (int c = 0; c < rvd->vdev_children; c++) { 3623 vdev_ashift_optimize(rvd->vdev_child[c]); 3624 vdev_metaslab_set_size(rvd->vdev_child[c]); 3625 vdev_expand(rvd->vdev_child[c], txg); 3626 } 3627 } 3628 3629 spa_config_exit(spa, SCL_ALL, FTAG); 3630 3631 if (error != 0) { 3632 spa_unload(spa); 3633 spa_deactivate(spa); 3634 spa_remove(spa); 3635 mutex_exit(&spa_namespace_lock); 3636 return (error); 3637 } 3638 3639 /* 3640 * Get the list of spares, if specified. 3641 */ 3642 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3643 &spares, &nspares) == 0) { 3644 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3645 KM_SLEEP) == 0); 3646 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3647 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3648 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3649 spa_load_spares(spa); 3650 spa_config_exit(spa, SCL_ALL, FTAG); 3651 spa->spa_spares.sav_sync = B_TRUE; 3652 } 3653 3654 /* 3655 * Get the list of level 2 cache devices, if specified. 3656 */ 3657 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3658 &l2cache, &nl2cache) == 0) { 3659 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3660 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3661 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3662 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3663 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3664 spa_load_l2cache(spa); 3665 spa_config_exit(spa, SCL_ALL, FTAG); 3666 spa->spa_l2cache.sav_sync = B_TRUE; 3667 } 3668 3669 spa->spa_is_initializing = B_TRUE; 3670 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3671 spa->spa_meta_objset = dp->dp_meta_objset; 3672 spa->spa_is_initializing = B_FALSE; 3673 3674 /* 3675 * Create DDTs (dedup tables). 3676 */ 3677 ddt_create(spa); 3678 3679 spa_update_dspace(spa); 3680 3681 tx = dmu_tx_create_assigned(dp, txg); 3682 3683 /* 3684 * Create the pool config object. 3685 */ 3686 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3687 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3688 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3689 3690 if (zap_add(spa->spa_meta_objset, 3691 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3692 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3693 cmn_err(CE_PANIC, "failed to add pool config"); 3694 } 3695 3696 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3697 spa_feature_create_zap_objects(spa, tx); 3698 3699 if (zap_add(spa->spa_meta_objset, 3700 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3701 sizeof (uint64_t), 1, &version, tx) != 0) { 3702 cmn_err(CE_PANIC, "failed to add pool version"); 3703 } 3704 3705 /* Newly created pools with the right version are always deflated. */ 3706 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3707 spa->spa_deflate = TRUE; 3708 if (zap_add(spa->spa_meta_objset, 3709 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3710 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3711 cmn_err(CE_PANIC, "failed to add deflate"); 3712 } 3713 } 3714 3715 /* 3716 * Create the deferred-free bpobj. Turn off compression 3717 * because sync-to-convergence takes longer if the blocksize 3718 * keeps changing. 3719 */ 3720 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3721 dmu_object_set_compress(spa->spa_meta_objset, obj, 3722 ZIO_COMPRESS_OFF, tx); 3723 if (zap_add(spa->spa_meta_objset, 3724 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3725 sizeof (uint64_t), 1, &obj, tx) != 0) { 3726 cmn_err(CE_PANIC, "failed to add bpobj"); 3727 } 3728 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3729 spa->spa_meta_objset, obj)); 3730 3731 /* 3732 * Create the pool's history object. 3733 */ 3734 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3735 spa_history_create_obj(spa, tx); 3736 3737 /* 3738 * Set pool properties. 3739 */ 3740 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3741 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3742 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3743 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3744 3745 if (props != NULL) { 3746 spa_configfile_set(spa, props, B_FALSE); 3747 spa_sync_props(props, tx); 3748 } 3749 3750 dmu_tx_commit(tx); 3751 3752 spa->spa_sync_on = B_TRUE; 3753 txg_sync_start(spa->spa_dsl_pool); 3754 3755 /* 3756 * We explicitly wait for the first transaction to complete so that our 3757 * bean counters are appropriately updated. 3758 */ 3759 txg_wait_synced(spa->spa_dsl_pool, txg); 3760 3761 spa_config_sync(spa, B_FALSE, B_TRUE); 3762 3763 spa_history_log_version(spa, "create"); 3764 3765 spa->spa_minref = refcount_count(&spa->spa_refcount); 3766 3767 mutex_exit(&spa_namespace_lock); 3768 3769 return (0); 3770} 3771 3772#ifdef _KERNEL 3773#ifdef illumos 3774/* 3775 * Get the root pool information from the root disk, then import the root pool 3776 * during the system boot up time. 3777 */ 3778extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3779 3780static nvlist_t * 3781spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3782{ 3783 nvlist_t *config; 3784 nvlist_t *nvtop, *nvroot; 3785 uint64_t pgid; 3786 3787 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3788 return (NULL); 3789 3790 /* 3791 * Add this top-level vdev to the child array. 3792 */ 3793 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3794 &nvtop) == 0); 3795 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3796 &pgid) == 0); 3797 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3798 3799 /* 3800 * Put this pool's top-level vdevs into a root vdev. 3801 */ 3802 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3803 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3804 VDEV_TYPE_ROOT) == 0); 3805 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3806 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3807 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3808 &nvtop, 1) == 0); 3809 3810 /* 3811 * Replace the existing vdev_tree with the new root vdev in 3812 * this pool's configuration (remove the old, add the new). 3813 */ 3814 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3815 nvlist_free(nvroot); 3816 return (config); 3817} 3818 3819/* 3820 * Walk the vdev tree and see if we can find a device with "better" 3821 * configuration. A configuration is "better" if the label on that 3822 * device has a more recent txg. 3823 */ 3824static void 3825spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3826{ 3827 for (int c = 0; c < vd->vdev_children; c++) 3828 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3829 3830 if (vd->vdev_ops->vdev_op_leaf) { 3831 nvlist_t *label; 3832 uint64_t label_txg; 3833 3834 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3835 &label) != 0) 3836 return; 3837 3838 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3839 &label_txg) == 0); 3840 3841 /* 3842 * Do we have a better boot device? 3843 */ 3844 if (label_txg > *txg) { 3845 *txg = label_txg; 3846 *avd = vd; 3847 } 3848 nvlist_free(label); 3849 } 3850} 3851 3852/* 3853 * Import a root pool. 3854 * 3855 * For x86. devpath_list will consist of devid and/or physpath name of 3856 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3857 * The GRUB "findroot" command will return the vdev we should boot. 3858 * 3859 * For Sparc, devpath_list consists the physpath name of the booting device 3860 * no matter the rootpool is a single device pool or a mirrored pool. 3861 * e.g. 3862 * "/pci@1f,0/ide@d/disk@0,0:a" 3863 */ 3864int 3865spa_import_rootpool(char *devpath, char *devid) 3866{ 3867 spa_t *spa; 3868 vdev_t *rvd, *bvd, *avd = NULL; 3869 nvlist_t *config, *nvtop; 3870 uint64_t guid, txg; 3871 char *pname; 3872 int error; 3873 3874 /* 3875 * Read the label from the boot device and generate a configuration. 3876 */ 3877 config = spa_generate_rootconf(devpath, devid, &guid); 3878#if defined(_OBP) && defined(_KERNEL) 3879 if (config == NULL) { 3880 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3881 /* iscsi boot */ 3882 get_iscsi_bootpath_phy(devpath); 3883 config = spa_generate_rootconf(devpath, devid, &guid); 3884 } 3885 } 3886#endif 3887 if (config == NULL) { 3888 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3889 devpath); 3890 return (SET_ERROR(EIO)); 3891 } 3892 3893 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3894 &pname) == 0); 3895 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3896 3897 mutex_enter(&spa_namespace_lock); 3898 if ((spa = spa_lookup(pname)) != NULL) { 3899 /* 3900 * Remove the existing root pool from the namespace so that we 3901 * can replace it with the correct config we just read in. 3902 */ 3903 spa_remove(spa); 3904 } 3905 3906 spa = spa_add(pname, config, NULL); 3907 spa->spa_is_root = B_TRUE; 3908 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3909 3910 /* 3911 * Build up a vdev tree based on the boot device's label config. 3912 */ 3913 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3914 &nvtop) == 0); 3915 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3916 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3917 VDEV_ALLOC_ROOTPOOL); 3918 spa_config_exit(spa, SCL_ALL, FTAG); 3919 if (error) { 3920 mutex_exit(&spa_namespace_lock); 3921 nvlist_free(config); 3922 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3923 pname); 3924 return (error); 3925 } 3926 3927 /* 3928 * Get the boot vdev. 3929 */ 3930 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3931 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3932 (u_longlong_t)guid); 3933 error = SET_ERROR(ENOENT); 3934 goto out; 3935 } 3936 3937 /* 3938 * Determine if there is a better boot device. 3939 */ 3940 avd = bvd; 3941 spa_alt_rootvdev(rvd, &avd, &txg); 3942 if (avd != bvd) { 3943 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3944 "try booting from '%s'", avd->vdev_path); 3945 error = SET_ERROR(EINVAL); 3946 goto out; 3947 } 3948 3949 /* 3950 * If the boot device is part of a spare vdev then ensure that 3951 * we're booting off the active spare. 3952 */ 3953 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3954 !bvd->vdev_isspare) { 3955 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3956 "try booting from '%s'", 3957 bvd->vdev_parent-> 3958 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3959 error = SET_ERROR(EINVAL); 3960 goto out; 3961 } 3962 3963 error = 0; 3964out: 3965 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3966 vdev_free(rvd); 3967 spa_config_exit(spa, SCL_ALL, FTAG); 3968 mutex_exit(&spa_namespace_lock); 3969 3970 nvlist_free(config); 3971 return (error); 3972} 3973 3974#else /* !illumos */ 3975 3976extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3977 uint64_t *count); 3978 3979static nvlist_t * 3980spa_generate_rootconf(const char *name) 3981{ 3982 nvlist_t **configs, **tops; 3983 nvlist_t *config; 3984 nvlist_t *best_cfg, *nvtop, *nvroot; 3985 uint64_t *holes; 3986 uint64_t best_txg; 3987 uint64_t nchildren; 3988 uint64_t pgid; 3989 uint64_t count; 3990 uint64_t i; 3991 uint_t nholes; 3992 3993 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3994 return (NULL); 3995 3996 ASSERT3U(count, !=, 0); 3997 best_txg = 0; 3998 for (i = 0; i < count; i++) { 3999 uint64_t txg; 4000 4001 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 4002 &txg) == 0); 4003 if (txg > best_txg) { 4004 best_txg = txg; 4005 best_cfg = configs[i]; 4006 } 4007 } 4008 4009 nchildren = 1; 4010 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 4011 holes = NULL; 4012 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 4013 &holes, &nholes); 4014 4015 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 4016 for (i = 0; i < nchildren; i++) { 4017 if (i >= count) 4018 break; 4019 if (configs[i] == NULL) 4020 continue; 4021 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 4022 &nvtop) == 0); 4023 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 4024 } 4025 for (i = 0; holes != NULL && i < nholes; i++) { 4026 if (i >= nchildren) 4027 continue; 4028 if (tops[holes[i]] != NULL) 4029 continue; 4030 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 4031 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 4032 VDEV_TYPE_HOLE) == 0); 4033 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 4034 holes[i]) == 0); 4035 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 4036 0) == 0); 4037 } 4038 for (i = 0; i < nchildren; i++) { 4039 if (tops[i] != NULL) 4040 continue; 4041 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 4042 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 4043 VDEV_TYPE_MISSING) == 0); 4044 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 4045 i) == 0); 4046 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 4047 0) == 0); 4048 } 4049 4050 /* 4051 * Create pool config based on the best vdev config. 4052 */ 4053 nvlist_dup(best_cfg, &config, KM_SLEEP); 4054 4055 /* 4056 * Put this pool's top-level vdevs into a root vdev. 4057 */ 4058 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4059 &pgid) == 0); 4060 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4061 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 4062 VDEV_TYPE_ROOT) == 0); 4063 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 4064 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 4065 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 4066 tops, nchildren) == 0); 4067 4068 /* 4069 * Replace the existing vdev_tree with the new root vdev in 4070 * this pool's configuration (remove the old, add the new). 4071 */ 4072 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 4073 4074 /* 4075 * Drop vdev config elements that should not be present at pool level. 4076 */ 4077 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 4078 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 4079 4080 for (i = 0; i < count; i++) 4081 nvlist_free(configs[i]); 4082 kmem_free(configs, count * sizeof(void *)); 4083 for (i = 0; i < nchildren; i++) 4084 nvlist_free(tops[i]); 4085 kmem_free(tops, nchildren * sizeof(void *)); 4086 nvlist_free(nvroot); 4087 return (config); 4088} 4089 4090int 4091spa_import_rootpool(const char *name) 4092{ 4093 spa_t *spa; 4094 vdev_t *rvd, *bvd, *avd = NULL; 4095 nvlist_t *config, *nvtop; 4096 uint64_t txg; 4097 char *pname; 4098 int error; 4099 4100 /* 4101 * Read the label from the boot device and generate a configuration. 4102 */ 4103 config = spa_generate_rootconf(name); 4104 4105 mutex_enter(&spa_namespace_lock); 4106 if (config != NULL) { 4107 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 4108 &pname) == 0 && strcmp(name, pname) == 0); 4109 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 4110 == 0); 4111 4112 if ((spa = spa_lookup(pname)) != NULL) { 4113 /* 4114 * Remove the existing root pool from the namespace so 4115 * that we can replace it with the correct config 4116 * we just read in. 4117 */ 4118 spa_remove(spa); 4119 } 4120 spa = spa_add(pname, config, NULL); 4121 4122 /* 4123 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4124 * via spa_version(). 4125 */ 4126 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4127 &spa->spa_ubsync.ub_version) != 0) 4128 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4129 } else if ((spa = spa_lookup(name)) == NULL) { 4130 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4131 name); 4132 return (EIO); 4133 } else { 4134 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4135 } 4136 spa->spa_is_root = B_TRUE; 4137 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4138 4139 /* 4140 * Build up a vdev tree based on the boot device's label config. 4141 */ 4142 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4143 &nvtop) == 0); 4144 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4145 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4146 VDEV_ALLOC_ROOTPOOL); 4147 spa_config_exit(spa, SCL_ALL, FTAG); 4148 if (error) { 4149 mutex_exit(&spa_namespace_lock); 4150 nvlist_free(config); 4151 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4152 pname); 4153 return (error); 4154 } 4155 4156 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4157 vdev_free(rvd); 4158 spa_config_exit(spa, SCL_ALL, FTAG); 4159 mutex_exit(&spa_namespace_lock); 4160 4161 nvlist_free(config); 4162 return (0); 4163} 4164 4165#endif /* illumos */ 4166#endif /* _KERNEL */ 4167 4168/* 4169 * Import a non-root pool into the system. 4170 */ 4171int 4172spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4173{ 4174 spa_t *spa; 4175 char *altroot = NULL; 4176 spa_load_state_t state = SPA_LOAD_IMPORT; 4177 zpool_rewind_policy_t policy; 4178 uint64_t mode = spa_mode_global; 4179 uint64_t readonly = B_FALSE; 4180 int error; 4181 nvlist_t *nvroot; 4182 nvlist_t **spares, **l2cache; 4183 uint_t nspares, nl2cache; 4184 4185 /* 4186 * If a pool with this name exists, return failure. 4187 */ 4188 mutex_enter(&spa_namespace_lock); 4189 if (spa_lookup(pool) != NULL) { 4190 mutex_exit(&spa_namespace_lock); 4191 return (SET_ERROR(EEXIST)); 4192 } 4193 4194 /* 4195 * Create and initialize the spa structure. 4196 */ 4197 (void) nvlist_lookup_string(props, 4198 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4199 (void) nvlist_lookup_uint64(props, 4200 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4201 if (readonly) 4202 mode = FREAD; 4203 spa = spa_add(pool, config, altroot); 4204 spa->spa_import_flags = flags; 4205 4206 /* 4207 * Verbatim import - Take a pool and insert it into the namespace 4208 * as if it had been loaded at boot. 4209 */ 4210 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4211 if (props != NULL) 4212 spa_configfile_set(spa, props, B_FALSE); 4213 4214 spa_config_sync(spa, B_FALSE, B_TRUE); 4215 4216 mutex_exit(&spa_namespace_lock); 4217 return (0); 4218 } 4219 4220 spa_activate(spa, mode); 4221 4222 /* 4223 * Don't start async tasks until we know everything is healthy. 4224 */ 4225 spa_async_suspend(spa); 4226 4227 zpool_get_rewind_policy(config, &policy); 4228 if (policy.zrp_request & ZPOOL_DO_REWIND) 4229 state = SPA_LOAD_RECOVER; 4230 4231 /* 4232 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4233 * because the user-supplied config is actually the one to trust when 4234 * doing an import. 4235 */ 4236 if (state != SPA_LOAD_RECOVER) 4237 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4238 4239 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4240 policy.zrp_request); 4241 4242 /* 4243 * Propagate anything learned while loading the pool and pass it 4244 * back to caller (i.e. rewind info, missing devices, etc). 4245 */ 4246 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4247 spa->spa_load_info) == 0); 4248 4249 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4250 /* 4251 * Toss any existing sparelist, as it doesn't have any validity 4252 * anymore, and conflicts with spa_has_spare(). 4253 */ 4254 if (spa->spa_spares.sav_config) { 4255 nvlist_free(spa->spa_spares.sav_config); 4256 spa->spa_spares.sav_config = NULL; 4257 spa_load_spares(spa); 4258 } 4259 if (spa->spa_l2cache.sav_config) { 4260 nvlist_free(spa->spa_l2cache.sav_config); 4261 spa->spa_l2cache.sav_config = NULL; 4262 spa_load_l2cache(spa); 4263 } 4264 4265 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4266 &nvroot) == 0); 4267 if (error == 0) 4268 error = spa_validate_aux(spa, nvroot, -1ULL, 4269 VDEV_ALLOC_SPARE); 4270 if (error == 0) 4271 error = spa_validate_aux(spa, nvroot, -1ULL, 4272 VDEV_ALLOC_L2CACHE); 4273 spa_config_exit(spa, SCL_ALL, FTAG); 4274 4275 if (props != NULL) 4276 spa_configfile_set(spa, props, B_FALSE); 4277 4278 if (error != 0 || (props && spa_writeable(spa) && 4279 (error = spa_prop_set(spa, props)))) { 4280 spa_unload(spa); 4281 spa_deactivate(spa); 4282 spa_remove(spa); 4283 mutex_exit(&spa_namespace_lock); 4284 return (error); 4285 } 4286 4287 spa_async_resume(spa); 4288 4289 /* 4290 * Override any spares and level 2 cache devices as specified by 4291 * the user, as these may have correct device names/devids, etc. 4292 */ 4293 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4294 &spares, &nspares) == 0) { 4295 if (spa->spa_spares.sav_config) 4296 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4297 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4298 else 4299 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4300 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4301 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4302 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4303 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4304 spa_load_spares(spa); 4305 spa_config_exit(spa, SCL_ALL, FTAG); 4306 spa->spa_spares.sav_sync = B_TRUE; 4307 } 4308 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4309 &l2cache, &nl2cache) == 0) { 4310 if (spa->spa_l2cache.sav_config) 4311 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4312 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4313 else 4314 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4315 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4316 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4317 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4318 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4319 spa_load_l2cache(spa); 4320 spa_config_exit(spa, SCL_ALL, FTAG); 4321 spa->spa_l2cache.sav_sync = B_TRUE; 4322 } 4323 4324 /* 4325 * Check for any removed devices. 4326 */ 4327 if (spa->spa_autoreplace) { 4328 spa_aux_check_removed(&spa->spa_spares); 4329 spa_aux_check_removed(&spa->spa_l2cache); 4330 } 4331 4332 if (spa_writeable(spa)) { 4333 /* 4334 * Update the config cache to include the newly-imported pool. 4335 */ 4336 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4337 } 4338 4339 /* 4340 * It's possible that the pool was expanded while it was exported. 4341 * We kick off an async task to handle this for us. 4342 */ 4343 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4344 4345 mutex_exit(&spa_namespace_lock); 4346 spa_history_log_version(spa, "import"); 4347 4348#ifdef __FreeBSD__ 4349#ifdef _KERNEL 4350 zvol_create_minors(pool); 4351#endif 4352#endif 4353 return (0); 4354} 4355 4356nvlist_t * 4357spa_tryimport(nvlist_t *tryconfig) 4358{ 4359 nvlist_t *config = NULL; 4360 char *poolname; 4361 spa_t *spa; 4362 uint64_t state; 4363 int error; 4364 4365 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4366 return (NULL); 4367 4368 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4369 return (NULL); 4370 4371 /* 4372 * Create and initialize the spa structure. 4373 */ 4374 mutex_enter(&spa_namespace_lock); 4375 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4376 spa_activate(spa, FREAD); 4377 4378 /* 4379 * Pass off the heavy lifting to spa_load(). 4380 * Pass TRUE for mosconfig because the user-supplied config 4381 * is actually the one to trust when doing an import. 4382 */ 4383 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4384 4385 /* 4386 * If 'tryconfig' was at least parsable, return the current config. 4387 */ 4388 if (spa->spa_root_vdev != NULL) { 4389 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4390 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4391 poolname) == 0); 4392 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4393 state) == 0); 4394 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4395 spa->spa_uberblock.ub_timestamp) == 0); 4396 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4397 spa->spa_load_info) == 0); 4398 4399 /* 4400 * If the bootfs property exists on this pool then we 4401 * copy it out so that external consumers can tell which 4402 * pools are bootable. 4403 */ 4404 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4405 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4406 4407 /* 4408 * We have to play games with the name since the 4409 * pool was opened as TRYIMPORT_NAME. 4410 */ 4411 if (dsl_dsobj_to_dsname(spa_name(spa), 4412 spa->spa_bootfs, tmpname) == 0) { 4413 char *cp; 4414 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4415 4416 cp = strchr(tmpname, '/'); 4417 if (cp == NULL) { 4418 (void) strlcpy(dsname, tmpname, 4419 MAXPATHLEN); 4420 } else { 4421 (void) snprintf(dsname, MAXPATHLEN, 4422 "%s/%s", poolname, ++cp); 4423 } 4424 VERIFY(nvlist_add_string(config, 4425 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4426 kmem_free(dsname, MAXPATHLEN); 4427 } 4428 kmem_free(tmpname, MAXPATHLEN); 4429 } 4430 4431 /* 4432 * Add the list of hot spares and level 2 cache devices. 4433 */ 4434 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4435 spa_add_spares(spa, config); 4436 spa_add_l2cache(spa, config); 4437 spa_config_exit(spa, SCL_CONFIG, FTAG); 4438 } 4439 4440 spa_unload(spa); 4441 spa_deactivate(spa); 4442 spa_remove(spa); 4443 mutex_exit(&spa_namespace_lock); 4444 4445 return (config); 4446} 4447 4448/* 4449 * Pool export/destroy 4450 * 4451 * The act of destroying or exporting a pool is very simple. We make sure there 4452 * is no more pending I/O and any references to the pool are gone. Then, we 4453 * update the pool state and sync all the labels to disk, removing the 4454 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4455 * we don't sync the labels or remove the configuration cache. 4456 */ 4457static int 4458spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4459 boolean_t force, boolean_t hardforce) 4460{ 4461 spa_t *spa; 4462 4463 if (oldconfig) 4464 *oldconfig = NULL; 4465 4466 if (!(spa_mode_global & FWRITE)) 4467 return (SET_ERROR(EROFS)); 4468 4469 mutex_enter(&spa_namespace_lock); 4470 if ((spa = spa_lookup(pool)) == NULL) { 4471 mutex_exit(&spa_namespace_lock); 4472 return (SET_ERROR(ENOENT)); 4473 } 4474 4475 /* 4476 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4477 * reacquire the namespace lock, and see if we can export. 4478 */ 4479 spa_open_ref(spa, FTAG); 4480 mutex_exit(&spa_namespace_lock); 4481 spa_async_suspend(spa); 4482 mutex_enter(&spa_namespace_lock); 4483 spa_close(spa, FTAG); 4484 4485 /* 4486 * The pool will be in core if it's openable, 4487 * in which case we can modify its state. 4488 */ 4489 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4490 /* 4491 * Objsets may be open only because they're dirty, so we 4492 * have to force it to sync before checking spa_refcnt. 4493 */ 4494 txg_wait_synced(spa->spa_dsl_pool, 0); 4495 4496 /* 4497 * A pool cannot be exported or destroyed if there are active 4498 * references. If we are resetting a pool, allow references by 4499 * fault injection handlers. 4500 */ 4501 if (!spa_refcount_zero(spa) || 4502 (spa->spa_inject_ref != 0 && 4503 new_state != POOL_STATE_UNINITIALIZED)) { 4504 spa_async_resume(spa); 4505 mutex_exit(&spa_namespace_lock); 4506 return (SET_ERROR(EBUSY)); 4507 } 4508 4509 /* 4510 * A pool cannot be exported if it has an active shared spare. 4511 * This is to prevent other pools stealing the active spare 4512 * from an exported pool. At user's own will, such pool can 4513 * be forcedly exported. 4514 */ 4515 if (!force && new_state == POOL_STATE_EXPORTED && 4516 spa_has_active_shared_spare(spa)) { 4517 spa_async_resume(spa); 4518 mutex_exit(&spa_namespace_lock); 4519 return (SET_ERROR(EXDEV)); 4520 } 4521 4522 /* 4523 * We want this to be reflected on every label, 4524 * so mark them all dirty. spa_unload() will do the 4525 * final sync that pushes these changes out. 4526 */ 4527 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4528 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4529 spa->spa_state = new_state; 4530 spa->spa_final_txg = spa_last_synced_txg(spa) + 4531 TXG_DEFER_SIZE + 1; 4532 vdev_config_dirty(spa->spa_root_vdev); 4533 spa_config_exit(spa, SCL_ALL, FTAG); 4534 } 4535 } 4536 4537 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4538 4539 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4540 spa_unload(spa); 4541 spa_deactivate(spa); 4542 } 4543 4544 if (oldconfig && spa->spa_config) 4545 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4546 4547 if (new_state != POOL_STATE_UNINITIALIZED) { 4548 if (!hardforce) 4549 spa_config_sync(spa, B_TRUE, B_TRUE); 4550 spa_remove(spa); 4551 } 4552 mutex_exit(&spa_namespace_lock); 4553 4554 return (0); 4555} 4556 4557/* 4558 * Destroy a storage pool. 4559 */ 4560int 4561spa_destroy(char *pool) 4562{ 4563 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4564 B_FALSE, B_FALSE)); 4565} 4566 4567/* 4568 * Export a storage pool. 4569 */ 4570int 4571spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4572 boolean_t hardforce) 4573{ 4574 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4575 force, hardforce)); 4576} 4577 4578/* 4579 * Similar to spa_export(), this unloads the spa_t without actually removing it 4580 * from the namespace in any way. 4581 */ 4582int 4583spa_reset(char *pool) 4584{ 4585 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4586 B_FALSE, B_FALSE)); 4587} 4588 4589/* 4590 * ========================================================================== 4591 * Device manipulation 4592 * ========================================================================== 4593 */ 4594 4595/* 4596 * Add a device to a storage pool. 4597 */ 4598int 4599spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4600{ 4601 uint64_t txg, id; 4602 int error; 4603 vdev_t *rvd = spa->spa_root_vdev; 4604 vdev_t *vd, *tvd; 4605 nvlist_t **spares, **l2cache; 4606 uint_t nspares, nl2cache; 4607 4608 ASSERT(spa_writeable(spa)); 4609 4610 txg = spa_vdev_enter(spa); 4611 4612 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4613 VDEV_ALLOC_ADD)) != 0) 4614 return (spa_vdev_exit(spa, NULL, txg, error)); 4615 4616 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4617 4618 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4619 &nspares) != 0) 4620 nspares = 0; 4621 4622 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4623 &nl2cache) != 0) 4624 nl2cache = 0; 4625 4626 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4627 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4628 4629 if (vd->vdev_children != 0 && 4630 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4631 return (spa_vdev_exit(spa, vd, txg, error)); 4632 4633 /* 4634 * We must validate the spares and l2cache devices after checking the 4635 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4636 */ 4637 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4638 return (spa_vdev_exit(spa, vd, txg, error)); 4639 4640 /* 4641 * Transfer each new top-level vdev from vd to rvd. 4642 */ 4643 for (int c = 0; c < vd->vdev_children; c++) { 4644 4645 /* 4646 * Set the vdev id to the first hole, if one exists. 4647 */ 4648 for (id = 0; id < rvd->vdev_children; id++) { 4649 if (rvd->vdev_child[id]->vdev_ishole) { 4650 vdev_free(rvd->vdev_child[id]); 4651 break; 4652 } 4653 } 4654 tvd = vd->vdev_child[c]; 4655 vdev_remove_child(vd, tvd); 4656 tvd->vdev_id = id; 4657 vdev_add_child(rvd, tvd); 4658 vdev_config_dirty(tvd); 4659 } 4660 4661 if (nspares != 0) { 4662 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4663 ZPOOL_CONFIG_SPARES); 4664 spa_load_spares(spa); 4665 spa->spa_spares.sav_sync = B_TRUE; 4666 } 4667 4668 if (nl2cache != 0) { 4669 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4670 ZPOOL_CONFIG_L2CACHE); 4671 spa_load_l2cache(spa); 4672 spa->spa_l2cache.sav_sync = B_TRUE; 4673 } 4674 4675 /* 4676 * We have to be careful when adding new vdevs to an existing pool. 4677 * If other threads start allocating from these vdevs before we 4678 * sync the config cache, and we lose power, then upon reboot we may 4679 * fail to open the pool because there are DVAs that the config cache 4680 * can't translate. Therefore, we first add the vdevs without 4681 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4682 * and then let spa_config_update() initialize the new metaslabs. 4683 * 4684 * spa_load() checks for added-but-not-initialized vdevs, so that 4685 * if we lose power at any point in this sequence, the remaining 4686 * steps will be completed the next time we load the pool. 4687 */ 4688 (void) spa_vdev_exit(spa, vd, txg, 0); 4689 4690 mutex_enter(&spa_namespace_lock); 4691 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4692 mutex_exit(&spa_namespace_lock); 4693 4694 return (0); 4695} 4696 4697/* 4698 * Attach a device to a mirror. The arguments are the path to any device 4699 * in the mirror, and the nvroot for the new device. If the path specifies 4700 * a device that is not mirrored, we automatically insert the mirror vdev. 4701 * 4702 * If 'replacing' is specified, the new device is intended to replace the 4703 * existing device; in this case the two devices are made into their own 4704 * mirror using the 'replacing' vdev, which is functionally identical to 4705 * the mirror vdev (it actually reuses all the same ops) but has a few 4706 * extra rules: you can't attach to it after it's been created, and upon 4707 * completion of resilvering, the first disk (the one being replaced) 4708 * is automatically detached. 4709 */ 4710int 4711spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4712{ 4713 uint64_t txg, dtl_max_txg; 4714 vdev_t *rvd = spa->spa_root_vdev; 4715 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4716 vdev_ops_t *pvops; 4717 char *oldvdpath, *newvdpath; 4718 int newvd_isspare; 4719 int error; 4720 4721 ASSERT(spa_writeable(spa)); 4722 4723 txg = spa_vdev_enter(spa); 4724 4725 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4726 4727 if (oldvd == NULL) 4728 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4729 4730 if (!oldvd->vdev_ops->vdev_op_leaf) 4731 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4732 4733 pvd = oldvd->vdev_parent; 4734 4735 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4736 VDEV_ALLOC_ATTACH)) != 0) 4737 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4738 4739 if (newrootvd->vdev_children != 1) 4740 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4741 4742 newvd = newrootvd->vdev_child[0]; 4743 4744 if (!newvd->vdev_ops->vdev_op_leaf) 4745 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4746 4747 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4748 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4749 4750 /* 4751 * Spares can't replace logs 4752 */ 4753 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4754 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4755 4756 if (!replacing) { 4757 /* 4758 * For attach, the only allowable parent is a mirror or the root 4759 * vdev. 4760 */ 4761 if (pvd->vdev_ops != &vdev_mirror_ops && 4762 pvd->vdev_ops != &vdev_root_ops) 4763 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4764 4765 pvops = &vdev_mirror_ops; 4766 } else { 4767 /* 4768 * Active hot spares can only be replaced by inactive hot 4769 * spares. 4770 */ 4771 if (pvd->vdev_ops == &vdev_spare_ops && 4772 oldvd->vdev_isspare && 4773 !spa_has_spare(spa, newvd->vdev_guid)) 4774 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4775 4776 /* 4777 * If the source is a hot spare, and the parent isn't already a 4778 * spare, then we want to create a new hot spare. Otherwise, we 4779 * want to create a replacing vdev. The user is not allowed to 4780 * attach to a spared vdev child unless the 'isspare' state is 4781 * the same (spare replaces spare, non-spare replaces 4782 * non-spare). 4783 */ 4784 if (pvd->vdev_ops == &vdev_replacing_ops && 4785 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4786 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4787 } else if (pvd->vdev_ops == &vdev_spare_ops && 4788 newvd->vdev_isspare != oldvd->vdev_isspare) { 4789 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4790 } 4791 4792 if (newvd->vdev_isspare) 4793 pvops = &vdev_spare_ops; 4794 else 4795 pvops = &vdev_replacing_ops; 4796 } 4797 4798 /* 4799 * Make sure the new device is big enough. 4800 */ 4801 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4802 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4803 4804 /* 4805 * The new device cannot have a higher alignment requirement 4806 * than the top-level vdev. 4807 */ 4808 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4809 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4810 4811 /* 4812 * If this is an in-place replacement, update oldvd's path and devid 4813 * to make it distinguishable from newvd, and unopenable from now on. 4814 */ 4815 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4816 spa_strfree(oldvd->vdev_path); 4817 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4818 KM_SLEEP); 4819 (void) sprintf(oldvd->vdev_path, "%s/%s", 4820 newvd->vdev_path, "old"); 4821 if (oldvd->vdev_devid != NULL) { 4822 spa_strfree(oldvd->vdev_devid); 4823 oldvd->vdev_devid = NULL; 4824 } 4825 } 4826 4827 /* mark the device being resilvered */ 4828 newvd->vdev_resilver_txg = txg; 4829 4830 /* 4831 * If the parent is not a mirror, or if we're replacing, insert the new 4832 * mirror/replacing/spare vdev above oldvd. 4833 */ 4834 if (pvd->vdev_ops != pvops) 4835 pvd = vdev_add_parent(oldvd, pvops); 4836 4837 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4838 ASSERT(pvd->vdev_ops == pvops); 4839 ASSERT(oldvd->vdev_parent == pvd); 4840 4841 /* 4842 * Extract the new device from its root and add it to pvd. 4843 */ 4844 vdev_remove_child(newrootvd, newvd); 4845 newvd->vdev_id = pvd->vdev_children; 4846 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4847 vdev_add_child(pvd, newvd); 4848 4849 tvd = newvd->vdev_top; 4850 ASSERT(pvd->vdev_top == tvd); 4851 ASSERT(tvd->vdev_parent == rvd); 4852 4853 vdev_config_dirty(tvd); 4854 4855 /* 4856 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4857 * for any dmu_sync-ed blocks. It will propagate upward when 4858 * spa_vdev_exit() calls vdev_dtl_reassess(). 4859 */ 4860 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4861 4862 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4863 dtl_max_txg - TXG_INITIAL); 4864 4865 if (newvd->vdev_isspare) { 4866 spa_spare_activate(newvd); 4867 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4868 } 4869 4870 oldvdpath = spa_strdup(oldvd->vdev_path); 4871 newvdpath = spa_strdup(newvd->vdev_path); 4872 newvd_isspare = newvd->vdev_isspare; 4873 4874 /* 4875 * Mark newvd's DTL dirty in this txg. 4876 */ 4877 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4878 4879 /* 4880 * Schedule the resilver to restart in the future. We do this to 4881 * ensure that dmu_sync-ed blocks have been stitched into the 4882 * respective datasets. 4883 */ 4884 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4885 4886 /* 4887 * Commit the config 4888 */ 4889 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4890 4891 spa_history_log_internal(spa, "vdev attach", NULL, 4892 "%s vdev=%s %s vdev=%s", 4893 replacing && newvd_isspare ? "spare in" : 4894 replacing ? "replace" : "attach", newvdpath, 4895 replacing ? "for" : "to", oldvdpath); 4896 4897 spa_strfree(oldvdpath); 4898 spa_strfree(newvdpath); 4899 4900 if (spa->spa_bootfs) 4901 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4902 4903 return (0); 4904} 4905 4906/* 4907 * Detach a device from a mirror or replacing vdev. 4908 * 4909 * If 'replace_done' is specified, only detach if the parent 4910 * is a replacing vdev. 4911 */ 4912int 4913spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4914{ 4915 uint64_t txg; 4916 int error; 4917 vdev_t *rvd = spa->spa_root_vdev; 4918 vdev_t *vd, *pvd, *cvd, *tvd; 4919 boolean_t unspare = B_FALSE; 4920 uint64_t unspare_guid = 0; 4921 char *vdpath; 4922 4923 ASSERT(spa_writeable(spa)); 4924 4925 txg = spa_vdev_enter(spa); 4926 4927 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4928 4929 if (vd == NULL) 4930 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4931 4932 if (!vd->vdev_ops->vdev_op_leaf) 4933 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4934 4935 pvd = vd->vdev_parent; 4936 4937 /* 4938 * If the parent/child relationship is not as expected, don't do it. 4939 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4940 * vdev that's replacing B with C. The user's intent in replacing 4941 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4942 * the replace by detaching C, the expected behavior is to end up 4943 * M(A,B). But suppose that right after deciding to detach C, 4944 * the replacement of B completes. We would have M(A,C), and then 4945 * ask to detach C, which would leave us with just A -- not what 4946 * the user wanted. To prevent this, we make sure that the 4947 * parent/child relationship hasn't changed -- in this example, 4948 * that C's parent is still the replacing vdev R. 4949 */ 4950 if (pvd->vdev_guid != pguid && pguid != 0) 4951 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4952 4953 /* 4954 * Only 'replacing' or 'spare' vdevs can be replaced. 4955 */ 4956 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4957 pvd->vdev_ops != &vdev_spare_ops) 4958 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4959 4960 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4961 spa_version(spa) >= SPA_VERSION_SPARES); 4962 4963 /* 4964 * Only mirror, replacing, and spare vdevs support detach. 4965 */ 4966 if (pvd->vdev_ops != &vdev_replacing_ops && 4967 pvd->vdev_ops != &vdev_mirror_ops && 4968 pvd->vdev_ops != &vdev_spare_ops) 4969 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4970 4971 /* 4972 * If this device has the only valid copy of some data, 4973 * we cannot safely detach it. 4974 */ 4975 if (vdev_dtl_required(vd)) 4976 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4977 4978 ASSERT(pvd->vdev_children >= 2); 4979 4980 /* 4981 * If we are detaching the second disk from a replacing vdev, then 4982 * check to see if we changed the original vdev's path to have "/old" 4983 * at the end in spa_vdev_attach(). If so, undo that change now. 4984 */ 4985 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4986 vd->vdev_path != NULL) { 4987 size_t len = strlen(vd->vdev_path); 4988 4989 for (int c = 0; c < pvd->vdev_children; c++) { 4990 cvd = pvd->vdev_child[c]; 4991 4992 if (cvd == vd || cvd->vdev_path == NULL) 4993 continue; 4994 4995 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4996 strcmp(cvd->vdev_path + len, "/old") == 0) { 4997 spa_strfree(cvd->vdev_path); 4998 cvd->vdev_path = spa_strdup(vd->vdev_path); 4999 break; 5000 } 5001 } 5002 } 5003 5004 /* 5005 * If we are detaching the original disk from a spare, then it implies 5006 * that the spare should become a real disk, and be removed from the 5007 * active spare list for the pool. 5008 */ 5009 if (pvd->vdev_ops == &vdev_spare_ops && 5010 vd->vdev_id == 0 && 5011 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 5012 unspare = B_TRUE; 5013 5014 /* 5015 * Erase the disk labels so the disk can be used for other things. 5016 * This must be done after all other error cases are handled, 5017 * but before we disembowel vd (so we can still do I/O to it). 5018 * But if we can't do it, don't treat the error as fatal -- 5019 * it may be that the unwritability of the disk is the reason 5020 * it's being detached! 5021 */ 5022 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5023 5024 /* 5025 * Remove vd from its parent and compact the parent's children. 5026 */ 5027 vdev_remove_child(pvd, vd); 5028 vdev_compact_children(pvd); 5029 5030 /* 5031 * Remember one of the remaining children so we can get tvd below. 5032 */ 5033 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 5034 5035 /* 5036 * If we need to remove the remaining child from the list of hot spares, 5037 * do it now, marking the vdev as no longer a spare in the process. 5038 * We must do this before vdev_remove_parent(), because that can 5039 * change the GUID if it creates a new toplevel GUID. For a similar 5040 * reason, we must remove the spare now, in the same txg as the detach; 5041 * otherwise someone could attach a new sibling, change the GUID, and 5042 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 5043 */ 5044 if (unspare) { 5045 ASSERT(cvd->vdev_isspare); 5046 spa_spare_remove(cvd); 5047 unspare_guid = cvd->vdev_guid; 5048 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 5049 cvd->vdev_unspare = B_TRUE; 5050 } 5051 5052 /* 5053 * If the parent mirror/replacing vdev only has one child, 5054 * the parent is no longer needed. Remove it from the tree. 5055 */ 5056 if (pvd->vdev_children == 1) { 5057 if (pvd->vdev_ops == &vdev_spare_ops) 5058 cvd->vdev_unspare = B_FALSE; 5059 vdev_remove_parent(cvd); 5060 } 5061 5062 5063 /* 5064 * We don't set tvd until now because the parent we just removed 5065 * may have been the previous top-level vdev. 5066 */ 5067 tvd = cvd->vdev_top; 5068 ASSERT(tvd->vdev_parent == rvd); 5069 5070 /* 5071 * Reevaluate the parent vdev state. 5072 */ 5073 vdev_propagate_state(cvd); 5074 5075 /* 5076 * If the 'autoexpand' property is set on the pool then automatically 5077 * try to expand the size of the pool. For example if the device we 5078 * just detached was smaller than the others, it may be possible to 5079 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 5080 * first so that we can obtain the updated sizes of the leaf vdevs. 5081 */ 5082 if (spa->spa_autoexpand) { 5083 vdev_reopen(tvd); 5084 vdev_expand(tvd, txg); 5085 } 5086 5087 vdev_config_dirty(tvd); 5088 5089 /* 5090 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 5091 * vd->vdev_detached is set and free vd's DTL object in syncing context. 5092 * But first make sure we're not on any *other* txg's DTL list, to 5093 * prevent vd from being accessed after it's freed. 5094 */ 5095 vdpath = spa_strdup(vd->vdev_path); 5096 for (int t = 0; t < TXG_SIZE; t++) 5097 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 5098 vd->vdev_detached = B_TRUE; 5099 vdev_dirty(tvd, VDD_DTL, vd, txg); 5100 5101 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 5102 5103 /* hang on to the spa before we release the lock */ 5104 spa_open_ref(spa, FTAG); 5105 5106 error = spa_vdev_exit(spa, vd, txg, 0); 5107 5108 spa_history_log_internal(spa, "detach", NULL, 5109 "vdev=%s", vdpath); 5110 spa_strfree(vdpath); 5111 5112 /* 5113 * If this was the removal of the original device in a hot spare vdev, 5114 * then we want to go through and remove the device from the hot spare 5115 * list of every other pool. 5116 */ 5117 if (unspare) { 5118 spa_t *altspa = NULL; 5119 5120 mutex_enter(&spa_namespace_lock); 5121 while ((altspa = spa_next(altspa)) != NULL) { 5122 if (altspa->spa_state != POOL_STATE_ACTIVE || 5123 altspa == spa) 5124 continue; 5125 5126 spa_open_ref(altspa, FTAG); 5127 mutex_exit(&spa_namespace_lock); 5128 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5129 mutex_enter(&spa_namespace_lock); 5130 spa_close(altspa, FTAG); 5131 } 5132 mutex_exit(&spa_namespace_lock); 5133 5134 /* search the rest of the vdevs for spares to remove */ 5135 spa_vdev_resilver_done(spa); 5136 } 5137 5138 /* all done with the spa; OK to release */ 5139 mutex_enter(&spa_namespace_lock); 5140 spa_close(spa, FTAG); 5141 mutex_exit(&spa_namespace_lock); 5142 5143 return (error); 5144} 5145 5146/* 5147 * Split a set of devices from their mirrors, and create a new pool from them. 5148 */ 5149int 5150spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5151 nvlist_t *props, boolean_t exp) 5152{ 5153 int error = 0; 5154 uint64_t txg, *glist; 5155 spa_t *newspa; 5156 uint_t c, children, lastlog; 5157 nvlist_t **child, *nvl, *tmp; 5158 dmu_tx_t *tx; 5159 char *altroot = NULL; 5160 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5161 boolean_t activate_slog; 5162 5163 ASSERT(spa_writeable(spa)); 5164 5165 txg = spa_vdev_enter(spa); 5166 5167 /* clear the log and flush everything up to now */ 5168 activate_slog = spa_passivate_log(spa); 5169 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5170 error = spa_offline_log(spa); 5171 txg = spa_vdev_config_enter(spa); 5172 5173 if (activate_slog) 5174 spa_activate_log(spa); 5175 5176 if (error != 0) 5177 return (spa_vdev_exit(spa, NULL, txg, error)); 5178 5179 /* check new spa name before going any further */ 5180 if (spa_lookup(newname) != NULL) 5181 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5182 5183 /* 5184 * scan through all the children to ensure they're all mirrors 5185 */ 5186 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5187 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5188 &children) != 0) 5189 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5190 5191 /* first, check to ensure we've got the right child count */ 5192 rvd = spa->spa_root_vdev; 5193 lastlog = 0; 5194 for (c = 0; c < rvd->vdev_children; c++) { 5195 vdev_t *vd = rvd->vdev_child[c]; 5196 5197 /* don't count the holes & logs as children */ 5198 if (vd->vdev_islog || vd->vdev_ishole) { 5199 if (lastlog == 0) 5200 lastlog = c; 5201 continue; 5202 } 5203 5204 lastlog = 0; 5205 } 5206 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5207 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5208 5209 /* next, ensure no spare or cache devices are part of the split */ 5210 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5211 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5212 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5213 5214 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5215 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5216 5217 /* then, loop over each vdev and validate it */ 5218 for (c = 0; c < children; c++) { 5219 uint64_t is_hole = 0; 5220 5221 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5222 &is_hole); 5223 5224 if (is_hole != 0) { 5225 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5226 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5227 continue; 5228 } else { 5229 error = SET_ERROR(EINVAL); 5230 break; 5231 } 5232 } 5233 5234 /* which disk is going to be split? */ 5235 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5236 &glist[c]) != 0) { 5237 error = SET_ERROR(EINVAL); 5238 break; 5239 } 5240 5241 /* look it up in the spa */ 5242 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5243 if (vml[c] == NULL) { 5244 error = SET_ERROR(ENODEV); 5245 break; 5246 } 5247 5248 /* make sure there's nothing stopping the split */ 5249 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5250 vml[c]->vdev_islog || 5251 vml[c]->vdev_ishole || 5252 vml[c]->vdev_isspare || 5253 vml[c]->vdev_isl2cache || 5254 !vdev_writeable(vml[c]) || 5255 vml[c]->vdev_children != 0 || 5256 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5257 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5258 error = SET_ERROR(EINVAL); 5259 break; 5260 } 5261 5262 if (vdev_dtl_required(vml[c])) { 5263 error = SET_ERROR(EBUSY); 5264 break; 5265 } 5266 5267 /* we need certain info from the top level */ 5268 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5269 vml[c]->vdev_top->vdev_ms_array) == 0); 5270 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5271 vml[c]->vdev_top->vdev_ms_shift) == 0); 5272 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5273 vml[c]->vdev_top->vdev_asize) == 0); 5274 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5275 vml[c]->vdev_top->vdev_ashift) == 0); 5276 } 5277 5278 if (error != 0) { 5279 kmem_free(vml, children * sizeof (vdev_t *)); 5280 kmem_free(glist, children * sizeof (uint64_t)); 5281 return (spa_vdev_exit(spa, NULL, txg, error)); 5282 } 5283 5284 /* stop writers from using the disks */ 5285 for (c = 0; c < children; c++) { 5286 if (vml[c] != NULL) 5287 vml[c]->vdev_offline = B_TRUE; 5288 } 5289 vdev_reopen(spa->spa_root_vdev); 5290 5291 /* 5292 * Temporarily record the splitting vdevs in the spa config. This 5293 * will disappear once the config is regenerated. 5294 */ 5295 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5296 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5297 glist, children) == 0); 5298 kmem_free(glist, children * sizeof (uint64_t)); 5299 5300 mutex_enter(&spa->spa_props_lock); 5301 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5302 nvl) == 0); 5303 mutex_exit(&spa->spa_props_lock); 5304 spa->spa_config_splitting = nvl; 5305 vdev_config_dirty(spa->spa_root_vdev); 5306 5307 /* configure and create the new pool */ 5308 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5309 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5310 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5311 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5312 spa_version(spa)) == 0); 5313 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5314 spa->spa_config_txg) == 0); 5315 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5316 spa_generate_guid(NULL)) == 0); 5317 (void) nvlist_lookup_string(props, 5318 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5319 5320 /* add the new pool to the namespace */ 5321 newspa = spa_add(newname, config, altroot); 5322 newspa->spa_config_txg = spa->spa_config_txg; 5323 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5324 5325 /* release the spa config lock, retaining the namespace lock */ 5326 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5327 5328 if (zio_injection_enabled) 5329 zio_handle_panic_injection(spa, FTAG, 1); 5330 5331 spa_activate(newspa, spa_mode_global); 5332 spa_async_suspend(newspa); 5333 5334#ifndef illumos 5335 /* mark that we are creating new spa by splitting */ 5336 newspa->spa_splitting_newspa = B_TRUE; 5337#endif 5338 /* create the new pool from the disks of the original pool */ 5339 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5340#ifndef illumos 5341 newspa->spa_splitting_newspa = B_FALSE; 5342#endif 5343 if (error) 5344 goto out; 5345 5346 /* if that worked, generate a real config for the new pool */ 5347 if (newspa->spa_root_vdev != NULL) { 5348 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5349 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5350 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5351 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5352 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5353 B_TRUE)); 5354 } 5355 5356 /* set the props */ 5357 if (props != NULL) { 5358 spa_configfile_set(newspa, props, B_FALSE); 5359 error = spa_prop_set(newspa, props); 5360 if (error) 5361 goto out; 5362 } 5363 5364 /* flush everything */ 5365 txg = spa_vdev_config_enter(newspa); 5366 vdev_config_dirty(newspa->spa_root_vdev); 5367 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5368 5369 if (zio_injection_enabled) 5370 zio_handle_panic_injection(spa, FTAG, 2); 5371 5372 spa_async_resume(newspa); 5373 5374 /* finally, update the original pool's config */ 5375 txg = spa_vdev_config_enter(spa); 5376 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5377 error = dmu_tx_assign(tx, TXG_WAIT); 5378 if (error != 0) 5379 dmu_tx_abort(tx); 5380 for (c = 0; c < children; c++) { 5381 if (vml[c] != NULL) { 5382 vdev_split(vml[c]); 5383 if (error == 0) 5384 spa_history_log_internal(spa, "detach", tx, 5385 "vdev=%s", vml[c]->vdev_path); 5386 vdev_free(vml[c]); 5387 } 5388 } 5389 vdev_config_dirty(spa->spa_root_vdev); 5390 spa->spa_config_splitting = NULL; 5391 nvlist_free(nvl); 5392 if (error == 0) 5393 dmu_tx_commit(tx); 5394 (void) spa_vdev_exit(spa, NULL, txg, 0); 5395 5396 if (zio_injection_enabled) 5397 zio_handle_panic_injection(spa, FTAG, 3); 5398 5399 /* split is complete; log a history record */ 5400 spa_history_log_internal(newspa, "split", NULL, 5401 "from pool %s", spa_name(spa)); 5402 5403 kmem_free(vml, children * sizeof (vdev_t *)); 5404 5405 /* if we're not going to mount the filesystems in userland, export */ 5406 if (exp) 5407 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5408 B_FALSE, B_FALSE); 5409 5410 return (error); 5411 5412out: 5413 spa_unload(newspa); 5414 spa_deactivate(newspa); 5415 spa_remove(newspa); 5416 5417 txg = spa_vdev_config_enter(spa); 5418 5419 /* re-online all offlined disks */ 5420 for (c = 0; c < children; c++) { 5421 if (vml[c] != NULL) 5422 vml[c]->vdev_offline = B_FALSE; 5423 } 5424 vdev_reopen(spa->spa_root_vdev); 5425 5426 nvlist_free(spa->spa_config_splitting); 5427 spa->spa_config_splitting = NULL; 5428 (void) spa_vdev_exit(spa, NULL, txg, error); 5429 5430 kmem_free(vml, children * sizeof (vdev_t *)); 5431 return (error); 5432} 5433 5434static nvlist_t * 5435spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5436{ 5437 for (int i = 0; i < count; i++) { 5438 uint64_t guid; 5439 5440 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5441 &guid) == 0); 5442 5443 if (guid == target_guid) 5444 return (nvpp[i]); 5445 } 5446 5447 return (NULL); 5448} 5449 5450static void 5451spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5452 nvlist_t *dev_to_remove) 5453{ 5454 nvlist_t **newdev = NULL; 5455 5456 if (count > 1) 5457 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5458 5459 for (int i = 0, j = 0; i < count; i++) { 5460 if (dev[i] == dev_to_remove) 5461 continue; 5462 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5463 } 5464 5465 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5466 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5467 5468 for (int i = 0; i < count - 1; i++) 5469 nvlist_free(newdev[i]); 5470 5471 if (count > 1) 5472 kmem_free(newdev, (count - 1) * sizeof (void *)); 5473} 5474 5475/* 5476 * Evacuate the device. 5477 */ 5478static int 5479spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5480{ 5481 uint64_t txg; 5482 int error = 0; 5483 5484 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5485 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5486 ASSERT(vd == vd->vdev_top); 5487 5488 /* 5489 * Evacuate the device. We don't hold the config lock as writer 5490 * since we need to do I/O but we do keep the 5491 * spa_namespace_lock held. Once this completes the device 5492 * should no longer have any blocks allocated on it. 5493 */ 5494 if (vd->vdev_islog) { 5495 if (vd->vdev_stat.vs_alloc != 0) 5496 error = spa_offline_log(spa); 5497 } else { 5498 error = SET_ERROR(ENOTSUP); 5499 } 5500 5501 if (error) 5502 return (error); 5503 5504 /* 5505 * The evacuation succeeded. Remove any remaining MOS metadata 5506 * associated with this vdev, and wait for these changes to sync. 5507 */ 5508 ASSERT0(vd->vdev_stat.vs_alloc); 5509 txg = spa_vdev_config_enter(spa); 5510 vd->vdev_removing = B_TRUE; 5511 vdev_dirty_leaves(vd, VDD_DTL, txg); 5512 vdev_config_dirty(vd); 5513 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5514 5515 return (0); 5516} 5517 5518/* 5519 * Complete the removal by cleaning up the namespace. 5520 */ 5521static void 5522spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5523{ 5524 vdev_t *rvd = spa->spa_root_vdev; 5525 uint64_t id = vd->vdev_id; 5526 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5527 5528 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5529 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5530 ASSERT(vd == vd->vdev_top); 5531 5532 /* 5533 * Only remove any devices which are empty. 5534 */ 5535 if (vd->vdev_stat.vs_alloc != 0) 5536 return; 5537 5538 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5539 5540 if (list_link_active(&vd->vdev_state_dirty_node)) 5541 vdev_state_clean(vd); 5542 if (list_link_active(&vd->vdev_config_dirty_node)) 5543 vdev_config_clean(vd); 5544 5545 vdev_free(vd); 5546 5547 if (last_vdev) { 5548 vdev_compact_children(rvd); 5549 } else { 5550 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5551 vdev_add_child(rvd, vd); 5552 } 5553 vdev_config_dirty(rvd); 5554 5555 /* 5556 * Reassess the health of our root vdev. 5557 */ 5558 vdev_reopen(rvd); 5559} 5560 5561/* 5562 * Remove a device from the pool - 5563 * 5564 * Removing a device from the vdev namespace requires several steps 5565 * and can take a significant amount of time. As a result we use 5566 * the spa_vdev_config_[enter/exit] functions which allow us to 5567 * grab and release the spa_config_lock while still holding the namespace 5568 * lock. During each step the configuration is synced out. 5569 * 5570 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5571 * devices. 5572 */ 5573int 5574spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5575{ 5576 vdev_t *vd; 5577 metaslab_group_t *mg; 5578 nvlist_t **spares, **l2cache, *nv; 5579 uint64_t txg = 0; 5580 uint_t nspares, nl2cache; 5581 int error = 0; 5582 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5583 5584 ASSERT(spa_writeable(spa)); 5585 5586 if (!locked) 5587 txg = spa_vdev_enter(spa); 5588 5589 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5590 5591 if (spa->spa_spares.sav_vdevs != NULL && 5592 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5593 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5594 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5595 /* 5596 * Only remove the hot spare if it's not currently in use 5597 * in this pool. 5598 */ 5599 if (vd == NULL || unspare) { 5600 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5601 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5602 spa_load_spares(spa); 5603 spa->spa_spares.sav_sync = B_TRUE; 5604 } else { 5605 error = SET_ERROR(EBUSY); 5606 } 5607 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5608 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5609 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5610 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5611 /* 5612 * Cache devices can always be removed. 5613 */ 5614 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5615 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5616 spa_load_l2cache(spa); 5617 spa->spa_l2cache.sav_sync = B_TRUE; 5618 } else if (vd != NULL && vd->vdev_islog) { 5619 ASSERT(!locked); 5620 ASSERT(vd == vd->vdev_top); 5621 5622 mg = vd->vdev_mg; 5623 5624 /* 5625 * Stop allocating from this vdev. 5626 */ 5627 metaslab_group_passivate(mg); 5628 5629 /* 5630 * Wait for the youngest allocations and frees to sync, 5631 * and then wait for the deferral of those frees to finish. 5632 */ 5633 spa_vdev_config_exit(spa, NULL, 5634 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5635 5636 /* 5637 * Attempt to evacuate the vdev. 5638 */ 5639 error = spa_vdev_remove_evacuate(spa, vd); 5640 5641 txg = spa_vdev_config_enter(spa); 5642 5643 /* 5644 * If we couldn't evacuate the vdev, unwind. 5645 */ 5646 if (error) { 5647 metaslab_group_activate(mg); 5648 return (spa_vdev_exit(spa, NULL, txg, error)); 5649 } 5650 5651 /* 5652 * Clean up the vdev namespace. 5653 */ 5654 spa_vdev_remove_from_namespace(spa, vd); 5655 5656 } else if (vd != NULL) { 5657 /* 5658 * Normal vdevs cannot be removed (yet). 5659 */ 5660 error = SET_ERROR(ENOTSUP); 5661 } else { 5662 /* 5663 * There is no vdev of any kind with the specified guid. 5664 */ 5665 error = SET_ERROR(ENOENT); 5666 } 5667 5668 if (!locked) 5669 return (spa_vdev_exit(spa, NULL, txg, error)); 5670 5671 return (error); 5672} 5673 5674/* 5675 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5676 * currently spared, so we can detach it. 5677 */ 5678static vdev_t * 5679spa_vdev_resilver_done_hunt(vdev_t *vd) 5680{ 5681 vdev_t *newvd, *oldvd; 5682 5683 for (int c = 0; c < vd->vdev_children; c++) { 5684 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5685 if (oldvd != NULL) 5686 return (oldvd); 5687 } 5688 5689 /* 5690 * Check for a completed replacement. We always consider the first 5691 * vdev in the list to be the oldest vdev, and the last one to be 5692 * the newest (see spa_vdev_attach() for how that works). In 5693 * the case where the newest vdev is faulted, we will not automatically 5694 * remove it after a resilver completes. This is OK as it will require 5695 * user intervention to determine which disk the admin wishes to keep. 5696 */ 5697 if (vd->vdev_ops == &vdev_replacing_ops) { 5698 ASSERT(vd->vdev_children > 1); 5699 5700 newvd = vd->vdev_child[vd->vdev_children - 1]; 5701 oldvd = vd->vdev_child[0]; 5702 5703 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5704 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5705 !vdev_dtl_required(oldvd)) 5706 return (oldvd); 5707 } 5708 5709 /* 5710 * Check for a completed resilver with the 'unspare' flag set. 5711 */ 5712 if (vd->vdev_ops == &vdev_spare_ops) { 5713 vdev_t *first = vd->vdev_child[0]; 5714 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5715 5716 if (last->vdev_unspare) { 5717 oldvd = first; 5718 newvd = last; 5719 } else if (first->vdev_unspare) { 5720 oldvd = last; 5721 newvd = first; 5722 } else { 5723 oldvd = NULL; 5724 } 5725 5726 if (oldvd != NULL && 5727 vdev_dtl_empty(newvd, DTL_MISSING) && 5728 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5729 !vdev_dtl_required(oldvd)) 5730 return (oldvd); 5731 5732 /* 5733 * If there are more than two spares attached to a disk, 5734 * and those spares are not required, then we want to 5735 * attempt to free them up now so that they can be used 5736 * by other pools. Once we're back down to a single 5737 * disk+spare, we stop removing them. 5738 */ 5739 if (vd->vdev_children > 2) { 5740 newvd = vd->vdev_child[1]; 5741 5742 if (newvd->vdev_isspare && last->vdev_isspare && 5743 vdev_dtl_empty(last, DTL_MISSING) && 5744 vdev_dtl_empty(last, DTL_OUTAGE) && 5745 !vdev_dtl_required(newvd)) 5746 return (newvd); 5747 } 5748 } 5749 5750 return (NULL); 5751} 5752 5753static void 5754spa_vdev_resilver_done(spa_t *spa) 5755{ 5756 vdev_t *vd, *pvd, *ppvd; 5757 uint64_t guid, sguid, pguid, ppguid; 5758 5759 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5760 5761 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5762 pvd = vd->vdev_parent; 5763 ppvd = pvd->vdev_parent; 5764 guid = vd->vdev_guid; 5765 pguid = pvd->vdev_guid; 5766 ppguid = ppvd->vdev_guid; 5767 sguid = 0; 5768 /* 5769 * If we have just finished replacing a hot spared device, then 5770 * we need to detach the parent's first child (the original hot 5771 * spare) as well. 5772 */ 5773 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5774 ppvd->vdev_children == 2) { 5775 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5776 sguid = ppvd->vdev_child[1]->vdev_guid; 5777 } 5778 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5779 5780 spa_config_exit(spa, SCL_ALL, FTAG); 5781 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5782 return; 5783 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5784 return; 5785 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5786 } 5787 5788 spa_config_exit(spa, SCL_ALL, FTAG); 5789} 5790 5791/* 5792 * Update the stored path or FRU for this vdev. 5793 */ 5794int 5795spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5796 boolean_t ispath) 5797{ 5798 vdev_t *vd; 5799 boolean_t sync = B_FALSE; 5800 5801 ASSERT(spa_writeable(spa)); 5802 5803 spa_vdev_state_enter(spa, SCL_ALL); 5804 5805 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5806 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5807 5808 if (!vd->vdev_ops->vdev_op_leaf) 5809 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5810 5811 if (ispath) { 5812 if (strcmp(value, vd->vdev_path) != 0) { 5813 spa_strfree(vd->vdev_path); 5814 vd->vdev_path = spa_strdup(value); 5815 sync = B_TRUE; 5816 } 5817 } else { 5818 if (vd->vdev_fru == NULL) { 5819 vd->vdev_fru = spa_strdup(value); 5820 sync = B_TRUE; 5821 } else if (strcmp(value, vd->vdev_fru) != 0) { 5822 spa_strfree(vd->vdev_fru); 5823 vd->vdev_fru = spa_strdup(value); 5824 sync = B_TRUE; 5825 } 5826 } 5827 5828 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5829} 5830 5831int 5832spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5833{ 5834 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5835} 5836 5837int 5838spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5839{ 5840 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5841} 5842 5843/* 5844 * ========================================================================== 5845 * SPA Scanning 5846 * ========================================================================== 5847 */ 5848 5849int 5850spa_scan_stop(spa_t *spa) 5851{ 5852 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5853 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5854 return (SET_ERROR(EBUSY)); 5855 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5856} 5857 5858int 5859spa_scan(spa_t *spa, pool_scan_func_t func) 5860{ 5861 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5862 5863 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5864 return (SET_ERROR(ENOTSUP)); 5865 5866 /* 5867 * If a resilver was requested, but there is no DTL on a 5868 * writeable leaf device, we have nothing to do. 5869 */ 5870 if (func == POOL_SCAN_RESILVER && 5871 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5872 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5873 return (0); 5874 } 5875 5876 return (dsl_scan(spa->spa_dsl_pool, func)); 5877} 5878 5879/* 5880 * ========================================================================== 5881 * SPA async task processing 5882 * ========================================================================== 5883 */ 5884 5885static void 5886spa_async_remove(spa_t *spa, vdev_t *vd) 5887{ 5888 if (vd->vdev_remove_wanted) { 5889 vd->vdev_remove_wanted = B_FALSE; 5890 vd->vdev_delayed_close = B_FALSE; 5891 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5892 5893 /* 5894 * We want to clear the stats, but we don't want to do a full 5895 * vdev_clear() as that will cause us to throw away 5896 * degraded/faulted state as well as attempt to reopen the 5897 * device, all of which is a waste. 5898 */ 5899 vd->vdev_stat.vs_read_errors = 0; 5900 vd->vdev_stat.vs_write_errors = 0; 5901 vd->vdev_stat.vs_checksum_errors = 0; 5902 5903 vdev_state_dirty(vd->vdev_top); 5904 } 5905 5906 for (int c = 0; c < vd->vdev_children; c++) 5907 spa_async_remove(spa, vd->vdev_child[c]); 5908} 5909 5910static void 5911spa_async_probe(spa_t *spa, vdev_t *vd) 5912{ 5913 if (vd->vdev_probe_wanted) { 5914 vd->vdev_probe_wanted = B_FALSE; 5915 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5916 } 5917 5918 for (int c = 0; c < vd->vdev_children; c++) 5919 spa_async_probe(spa, vd->vdev_child[c]); 5920} 5921 5922static void 5923spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5924{ 5925 sysevent_id_t eid; 5926 nvlist_t *attr; 5927 char *physpath; 5928 5929 if (!spa->spa_autoexpand) 5930 return; 5931 5932 for (int c = 0; c < vd->vdev_children; c++) { 5933 vdev_t *cvd = vd->vdev_child[c]; 5934 spa_async_autoexpand(spa, cvd); 5935 } 5936 5937 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5938 return; 5939 5940 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5941 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5942 5943 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5944 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5945 5946 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5947 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5948 5949 nvlist_free(attr); 5950 kmem_free(physpath, MAXPATHLEN); 5951} 5952 5953static void 5954spa_async_thread(void *arg) 5955{ 5956 spa_t *spa = arg; 5957 int tasks; 5958 5959 ASSERT(spa->spa_sync_on); 5960 5961 mutex_enter(&spa->spa_async_lock); 5962 tasks = spa->spa_async_tasks; 5963 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5964 mutex_exit(&spa->spa_async_lock); 5965 5966 /* 5967 * See if the config needs to be updated. 5968 */ 5969 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5970 uint64_t old_space, new_space; 5971 5972 mutex_enter(&spa_namespace_lock); 5973 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5974 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5975 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5976 mutex_exit(&spa_namespace_lock); 5977 5978 /* 5979 * If the pool grew as a result of the config update, 5980 * then log an internal history event. 5981 */ 5982 if (new_space != old_space) { 5983 spa_history_log_internal(spa, "vdev online", NULL, 5984 "pool '%s' size: %llu(+%llu)", 5985 spa_name(spa), new_space, new_space - old_space); 5986 } 5987 } 5988 5989 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5990 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5991 spa_async_autoexpand(spa, spa->spa_root_vdev); 5992 spa_config_exit(spa, SCL_CONFIG, FTAG); 5993 } 5994 5995 /* 5996 * See if any devices need to be probed. 5997 */ 5998 if (tasks & SPA_ASYNC_PROBE) { 5999 spa_vdev_state_enter(spa, SCL_NONE); 6000 spa_async_probe(spa, spa->spa_root_vdev); 6001 (void) spa_vdev_state_exit(spa, NULL, 0); 6002 } 6003 6004 /* 6005 * If any devices are done replacing, detach them. 6006 */ 6007 if (tasks & SPA_ASYNC_RESILVER_DONE) 6008 spa_vdev_resilver_done(spa); 6009 6010 /* 6011 * Kick off a resilver. 6012 */ 6013 if (tasks & SPA_ASYNC_RESILVER) 6014 dsl_resilver_restart(spa->spa_dsl_pool, 0); 6015 6016 /* 6017 * Let the world know that we're done. 6018 */ 6019 mutex_enter(&spa->spa_async_lock); 6020 spa->spa_async_thread = NULL; 6021 cv_broadcast(&spa->spa_async_cv); 6022 mutex_exit(&spa->spa_async_lock); 6023 thread_exit(); 6024} 6025 6026static void 6027spa_async_thread_vd(void *arg) 6028{ 6029 spa_t *spa = arg; 6030 int tasks; 6031 6032 ASSERT(spa->spa_sync_on); 6033 6034 mutex_enter(&spa->spa_async_lock); 6035 tasks = spa->spa_async_tasks; 6036retry: 6037 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 6038 mutex_exit(&spa->spa_async_lock); 6039 6040 /* 6041 * See if any devices need to be marked REMOVED. 6042 */ 6043 if (tasks & SPA_ASYNC_REMOVE) { 6044 spa_vdev_state_enter(spa, SCL_NONE); 6045 spa_async_remove(spa, spa->spa_root_vdev); 6046 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 6047 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 6048 for (int i = 0; i < spa->spa_spares.sav_count; i++) 6049 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 6050 (void) spa_vdev_state_exit(spa, NULL, 0); 6051 } 6052 6053 /* 6054 * Let the world know that we're done. 6055 */ 6056 mutex_enter(&spa->spa_async_lock); 6057 tasks = spa->spa_async_tasks; 6058 if ((tasks & SPA_ASYNC_REMOVE) != 0) 6059 goto retry; 6060 spa->spa_async_thread_vd = NULL; 6061 cv_broadcast(&spa->spa_async_cv); 6062 mutex_exit(&spa->spa_async_lock); 6063 thread_exit(); 6064} 6065 6066void 6067spa_async_suspend(spa_t *spa) 6068{ 6069 mutex_enter(&spa->spa_async_lock); 6070 spa->spa_async_suspended++; 6071 while (spa->spa_async_thread != NULL && 6072 spa->spa_async_thread_vd != NULL) 6073 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 6074 mutex_exit(&spa->spa_async_lock); 6075} 6076 6077void 6078spa_async_resume(spa_t *spa) 6079{ 6080 mutex_enter(&spa->spa_async_lock); 6081 ASSERT(spa->spa_async_suspended != 0); 6082 spa->spa_async_suspended--; 6083 mutex_exit(&spa->spa_async_lock); 6084} 6085 6086static boolean_t 6087spa_async_tasks_pending(spa_t *spa) 6088{ 6089 uint_t non_config_tasks; 6090 uint_t config_task; 6091 boolean_t config_task_suspended; 6092 6093 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 6094 SPA_ASYNC_REMOVE); 6095 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 6096 if (spa->spa_ccw_fail_time == 0) { 6097 config_task_suspended = B_FALSE; 6098 } else { 6099 config_task_suspended = 6100 (gethrtime() - spa->spa_ccw_fail_time) < 6101 (zfs_ccw_retry_interval * NANOSEC); 6102 } 6103 6104 return (non_config_tasks || (config_task && !config_task_suspended)); 6105} 6106 6107static void 6108spa_async_dispatch(spa_t *spa) 6109{ 6110 mutex_enter(&spa->spa_async_lock); 6111 if (spa_async_tasks_pending(spa) && 6112 !spa->spa_async_suspended && 6113 spa->spa_async_thread == NULL && 6114 rootdir != NULL) 6115 spa->spa_async_thread = thread_create(NULL, 0, 6116 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6117 mutex_exit(&spa->spa_async_lock); 6118} 6119 6120static void 6121spa_async_dispatch_vd(spa_t *spa) 6122{ 6123 mutex_enter(&spa->spa_async_lock); 6124 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6125 !spa->spa_async_suspended && 6126 spa->spa_async_thread_vd == NULL && 6127 rootdir != NULL) 6128 spa->spa_async_thread_vd = thread_create(NULL, 0, 6129 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6130 mutex_exit(&spa->spa_async_lock); 6131} 6132 6133void 6134spa_async_request(spa_t *spa, int task) 6135{ 6136 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6137 mutex_enter(&spa->spa_async_lock); 6138 spa->spa_async_tasks |= task; 6139 mutex_exit(&spa->spa_async_lock); 6140 spa_async_dispatch_vd(spa); 6141} 6142 6143/* 6144 * ========================================================================== 6145 * SPA syncing routines 6146 * ========================================================================== 6147 */ 6148 6149static int 6150bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6151{ 6152 bpobj_t *bpo = arg; 6153 bpobj_enqueue(bpo, bp, tx); 6154 return (0); 6155} 6156 6157static int 6158spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6159{ 6160 zio_t *zio = arg; 6161 6162 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6163 BP_GET_PSIZE(bp), zio->io_flags)); 6164 return (0); 6165} 6166 6167/* 6168 * Note: this simple function is not inlined to make it easier to dtrace the 6169 * amount of time spent syncing frees. 6170 */ 6171static void 6172spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6173{ 6174 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6175 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6176 VERIFY(zio_wait(zio) == 0); 6177} 6178 6179/* 6180 * Note: this simple function is not inlined to make it easier to dtrace the 6181 * amount of time spent syncing deferred frees. 6182 */ 6183static void 6184spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6185{ 6186 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6187 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6188 spa_free_sync_cb, zio, tx), ==, 0); 6189 VERIFY0(zio_wait(zio)); 6190} 6191 6192 6193static void 6194spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6195{ 6196 char *packed = NULL; 6197 size_t bufsize; 6198 size_t nvsize = 0; 6199 dmu_buf_t *db; 6200 6201 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6202 6203 /* 6204 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6205 * information. This avoids the dmu_buf_will_dirty() path and 6206 * saves us a pre-read to get data we don't actually care about. 6207 */ 6208 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6209 packed = kmem_alloc(bufsize, KM_SLEEP); 6210 6211 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6212 KM_SLEEP) == 0); 6213 bzero(packed + nvsize, bufsize - nvsize); 6214 6215 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6216 6217 kmem_free(packed, bufsize); 6218 6219 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6220 dmu_buf_will_dirty(db, tx); 6221 *(uint64_t *)db->db_data = nvsize; 6222 dmu_buf_rele(db, FTAG); 6223} 6224 6225static void 6226spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6227 const char *config, const char *entry) 6228{ 6229 nvlist_t *nvroot; 6230 nvlist_t **list; 6231 int i; 6232 6233 if (!sav->sav_sync) 6234 return; 6235 6236 /* 6237 * Update the MOS nvlist describing the list of available devices. 6238 * spa_validate_aux() will have already made sure this nvlist is 6239 * valid and the vdevs are labeled appropriately. 6240 */ 6241 if (sav->sav_object == 0) { 6242 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6243 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6244 sizeof (uint64_t), tx); 6245 VERIFY(zap_update(spa->spa_meta_objset, 6246 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6247 &sav->sav_object, tx) == 0); 6248 } 6249 6250 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6251 if (sav->sav_count == 0) { 6252 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6253 } else { 6254 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6255 for (i = 0; i < sav->sav_count; i++) 6256 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6257 B_FALSE, VDEV_CONFIG_L2CACHE); 6258 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6259 sav->sav_count) == 0); 6260 for (i = 0; i < sav->sav_count; i++) 6261 nvlist_free(list[i]); 6262 kmem_free(list, sav->sav_count * sizeof (void *)); 6263 } 6264 6265 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6266 nvlist_free(nvroot); 6267 6268 sav->sav_sync = B_FALSE; 6269} 6270 6271static void 6272spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6273{ 6274 nvlist_t *config; 6275 6276 if (list_is_empty(&spa->spa_config_dirty_list)) 6277 return; 6278 6279 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6280 6281 config = spa_config_generate(spa, spa->spa_root_vdev, 6282 dmu_tx_get_txg(tx), B_FALSE); 6283 6284 /* 6285 * If we're upgrading the spa version then make sure that 6286 * the config object gets updated with the correct version. 6287 */ 6288 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6289 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6290 spa->spa_uberblock.ub_version); 6291 6292 spa_config_exit(spa, SCL_STATE, FTAG); 6293 6294 if (spa->spa_config_syncing) 6295 nvlist_free(spa->spa_config_syncing); 6296 spa->spa_config_syncing = config; 6297 6298 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6299} 6300 6301static void 6302spa_sync_version(void *arg, dmu_tx_t *tx) 6303{ 6304 uint64_t *versionp = arg; 6305 uint64_t version = *versionp; 6306 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6307 6308 /* 6309 * Setting the version is special cased when first creating the pool. 6310 */ 6311 ASSERT(tx->tx_txg != TXG_INITIAL); 6312 6313 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6314 ASSERT(version >= spa_version(spa)); 6315 6316 spa->spa_uberblock.ub_version = version; 6317 vdev_config_dirty(spa->spa_root_vdev); 6318 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6319} 6320 6321/* 6322 * Set zpool properties. 6323 */ 6324static void 6325spa_sync_props(void *arg, dmu_tx_t *tx) 6326{ 6327 nvlist_t *nvp = arg; 6328 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6329 objset_t *mos = spa->spa_meta_objset; 6330 nvpair_t *elem = NULL; 6331 6332 mutex_enter(&spa->spa_props_lock); 6333 6334 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6335 uint64_t intval; 6336 char *strval, *fname; 6337 zpool_prop_t prop; 6338 const char *propname; 6339 zprop_type_t proptype; 6340 spa_feature_t fid; 6341 6342 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6343 case ZPROP_INVAL: 6344 /* 6345 * We checked this earlier in spa_prop_validate(). 6346 */ 6347 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6348 6349 fname = strchr(nvpair_name(elem), '@') + 1; 6350 VERIFY0(zfeature_lookup_name(fname, &fid)); 6351 6352 spa_feature_enable(spa, fid, tx); 6353 spa_history_log_internal(spa, "set", tx, 6354 "%s=enabled", nvpair_name(elem)); 6355 break; 6356 6357 case ZPOOL_PROP_VERSION: 6358 intval = fnvpair_value_uint64(elem); 6359 /* 6360 * The version is synced seperatly before other 6361 * properties and should be correct by now. 6362 */ 6363 ASSERT3U(spa_version(spa), >=, intval); 6364 break; 6365 6366 case ZPOOL_PROP_ALTROOT: 6367 /* 6368 * 'altroot' is a non-persistent property. It should 6369 * have been set temporarily at creation or import time. 6370 */ 6371 ASSERT(spa->spa_root != NULL); 6372 break; 6373 6374 case ZPOOL_PROP_READONLY: 6375 case ZPOOL_PROP_CACHEFILE: 6376 /* 6377 * 'readonly' and 'cachefile' are also non-persisitent 6378 * properties. 6379 */ 6380 break; 6381 case ZPOOL_PROP_COMMENT: 6382 strval = fnvpair_value_string(elem); 6383 if (spa->spa_comment != NULL) 6384 spa_strfree(spa->spa_comment); 6385 spa->spa_comment = spa_strdup(strval); 6386 /* 6387 * We need to dirty the configuration on all the vdevs 6388 * so that their labels get updated. It's unnecessary 6389 * to do this for pool creation since the vdev's 6390 * configuratoin has already been dirtied. 6391 */ 6392 if (tx->tx_txg != TXG_INITIAL) 6393 vdev_config_dirty(spa->spa_root_vdev); 6394 spa_history_log_internal(spa, "set", tx, 6395 "%s=%s", nvpair_name(elem), strval); 6396 break; 6397 default: 6398 /* 6399 * Set pool property values in the poolprops mos object. 6400 */ 6401 if (spa->spa_pool_props_object == 0) { 6402 spa->spa_pool_props_object = 6403 zap_create_link(mos, DMU_OT_POOL_PROPS, 6404 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6405 tx); 6406 } 6407 6408 /* normalize the property name */ 6409 propname = zpool_prop_to_name(prop); 6410 proptype = zpool_prop_get_type(prop); 6411 6412 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6413 ASSERT(proptype == PROP_TYPE_STRING); 6414 strval = fnvpair_value_string(elem); 6415 VERIFY0(zap_update(mos, 6416 spa->spa_pool_props_object, propname, 6417 1, strlen(strval) + 1, strval, tx)); 6418 spa_history_log_internal(spa, "set", tx, 6419 "%s=%s", nvpair_name(elem), strval); 6420 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6421 intval = fnvpair_value_uint64(elem); 6422 6423 if (proptype == PROP_TYPE_INDEX) { 6424 const char *unused; 6425 VERIFY0(zpool_prop_index_to_string( 6426 prop, intval, &unused)); 6427 } 6428 VERIFY0(zap_update(mos, 6429 spa->spa_pool_props_object, propname, 6430 8, 1, &intval, tx)); 6431 spa_history_log_internal(spa, "set", tx, 6432 "%s=%lld", nvpair_name(elem), intval); 6433 } else { 6434 ASSERT(0); /* not allowed */ 6435 } 6436 6437 switch (prop) { 6438 case ZPOOL_PROP_DELEGATION: 6439 spa->spa_delegation = intval; 6440 break; 6441 case ZPOOL_PROP_BOOTFS: 6442 spa->spa_bootfs = intval; 6443 break; 6444 case ZPOOL_PROP_FAILUREMODE: 6445 spa->spa_failmode = intval; 6446 break; 6447 case ZPOOL_PROP_AUTOEXPAND: 6448 spa->spa_autoexpand = intval; 6449 if (tx->tx_txg != TXG_INITIAL) 6450 spa_async_request(spa, 6451 SPA_ASYNC_AUTOEXPAND); 6452 break; 6453 case ZPOOL_PROP_DEDUPDITTO: 6454 spa->spa_dedup_ditto = intval; 6455 break; 6456 default: 6457 break; 6458 } 6459 } 6460 6461 } 6462 6463 mutex_exit(&spa->spa_props_lock); 6464} 6465 6466/* 6467 * Perform one-time upgrade on-disk changes. spa_version() does not 6468 * reflect the new version this txg, so there must be no changes this 6469 * txg to anything that the upgrade code depends on after it executes. 6470 * Therefore this must be called after dsl_pool_sync() does the sync 6471 * tasks. 6472 */ 6473static void 6474spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6475{ 6476 dsl_pool_t *dp = spa->spa_dsl_pool; 6477 6478 ASSERT(spa->spa_sync_pass == 1); 6479 6480 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6481 6482 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6483 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6484 dsl_pool_create_origin(dp, tx); 6485 6486 /* Keeping the origin open increases spa_minref */ 6487 spa->spa_minref += 3; 6488 } 6489 6490 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6491 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6492 dsl_pool_upgrade_clones(dp, tx); 6493 } 6494 6495 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6496 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6497 dsl_pool_upgrade_dir_clones(dp, tx); 6498 6499 /* Keeping the freedir open increases spa_minref */ 6500 spa->spa_minref += 3; 6501 } 6502 6503 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6504 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6505 spa_feature_create_zap_objects(spa, tx); 6506 } 6507 6508 /* 6509 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable 6510 * when possibility to use lz4 compression for metadata was added 6511 * Old pools that have this feature enabled must be upgraded to have 6512 * this feature active 6513 */ 6514 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6515 boolean_t lz4_en = spa_feature_is_enabled(spa, 6516 SPA_FEATURE_LZ4_COMPRESS); 6517 boolean_t lz4_ac = spa_feature_is_active(spa, 6518 SPA_FEATURE_LZ4_COMPRESS); 6519 6520 if (lz4_en && !lz4_ac) 6521 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx); 6522 } 6523 rrw_exit(&dp->dp_config_rwlock, FTAG); 6524} 6525 6526/* 6527 * Sync the specified transaction group. New blocks may be dirtied as 6528 * part of the process, so we iterate until it converges. 6529 */ 6530void 6531spa_sync(spa_t *spa, uint64_t txg) 6532{ 6533 dsl_pool_t *dp = spa->spa_dsl_pool; 6534 objset_t *mos = spa->spa_meta_objset; 6535 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6536 vdev_t *rvd = spa->spa_root_vdev; 6537 vdev_t *vd; 6538 dmu_tx_t *tx; 6539 int error; 6540 6541 VERIFY(spa_writeable(spa)); 6542 6543 /* 6544 * Lock out configuration changes. 6545 */ 6546 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6547 6548 spa->spa_syncing_txg = txg; 6549 spa->spa_sync_pass = 0; 6550 6551 /* 6552 * If there are any pending vdev state changes, convert them 6553 * into config changes that go out with this transaction group. 6554 */ 6555 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6556 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6557 /* 6558 * We need the write lock here because, for aux vdevs, 6559 * calling vdev_config_dirty() modifies sav_config. 6560 * This is ugly and will become unnecessary when we 6561 * eliminate the aux vdev wart by integrating all vdevs 6562 * into the root vdev tree. 6563 */ 6564 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6565 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6566 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6567 vdev_state_clean(vd); 6568 vdev_config_dirty(vd); 6569 } 6570 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6571 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6572 } 6573 spa_config_exit(spa, SCL_STATE, FTAG); 6574 6575 tx = dmu_tx_create_assigned(dp, txg); 6576 6577 spa->spa_sync_starttime = gethrtime(); 6578#ifdef illumos 6579 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6580 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6581#else /* !illumos */ 6582#ifdef _KERNEL 6583 callout_reset(&spa->spa_deadman_cycid, 6584 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6585#endif 6586#endif /* illumos */ 6587 6588 /* 6589 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6590 * set spa_deflate if we have no raid-z vdevs. 6591 */ 6592 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6593 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6594 int i; 6595 6596 for (i = 0; i < rvd->vdev_children; i++) { 6597 vd = rvd->vdev_child[i]; 6598 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6599 break; 6600 } 6601 if (i == rvd->vdev_children) { 6602 spa->spa_deflate = TRUE; 6603 VERIFY(0 == zap_add(spa->spa_meta_objset, 6604 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6605 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6606 } 6607 } 6608 6609 /* 6610 * Iterate to convergence. 6611 */ 6612 do { 6613 int pass = ++spa->spa_sync_pass; 6614 6615 spa_sync_config_object(spa, tx); 6616 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6617 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6618 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6619 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6620 spa_errlog_sync(spa, txg); 6621 dsl_pool_sync(dp, txg); 6622 6623 if (pass < zfs_sync_pass_deferred_free) { 6624 spa_sync_frees(spa, free_bpl, tx); 6625 } else { 6626 /* 6627 * We can not defer frees in pass 1, because 6628 * we sync the deferred frees later in pass 1. 6629 */ 6630 ASSERT3U(pass, >, 1); 6631 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6632 &spa->spa_deferred_bpobj, tx); 6633 } 6634 6635 ddt_sync(spa, txg); 6636 dsl_scan_sync(dp, tx); 6637 6638 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6639 vdev_sync(vd, txg); 6640 6641 if (pass == 1) { 6642 spa_sync_upgrades(spa, tx); 6643 ASSERT3U(txg, >=, 6644 spa->spa_uberblock.ub_rootbp.blk_birth); 6645 /* 6646 * Note: We need to check if the MOS is dirty 6647 * because we could have marked the MOS dirty 6648 * without updating the uberblock (e.g. if we 6649 * have sync tasks but no dirty user data). We 6650 * need to check the uberblock's rootbp because 6651 * it is updated if we have synced out dirty 6652 * data (though in this case the MOS will most 6653 * likely also be dirty due to second order 6654 * effects, we don't want to rely on that here). 6655 */ 6656 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg && 6657 !dmu_objset_is_dirty(mos, txg)) { 6658 /* 6659 * Nothing changed on the first pass, 6660 * therefore this TXG is a no-op. Avoid 6661 * syncing deferred frees, so that we 6662 * can keep this TXG as a no-op. 6663 */ 6664 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, 6665 txg)); 6666 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6667 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg)); 6668 break; 6669 } 6670 spa_sync_deferred_frees(spa, tx); 6671 } 6672 6673 } while (dmu_objset_is_dirty(mos, txg)); 6674 6675 /* 6676 * Rewrite the vdev configuration (which includes the uberblock) 6677 * to commit the transaction group. 6678 * 6679 * If there are no dirty vdevs, we sync the uberblock to a few 6680 * random top-level vdevs that are known to be visible in the 6681 * config cache (see spa_vdev_add() for a complete description). 6682 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6683 */ 6684 for (;;) { 6685 /* 6686 * We hold SCL_STATE to prevent vdev open/close/etc. 6687 * while we're attempting to write the vdev labels. 6688 */ 6689 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6690 6691 if (list_is_empty(&spa->spa_config_dirty_list)) { 6692 vdev_t *svd[SPA_DVAS_PER_BP]; 6693 int svdcount = 0; 6694 int children = rvd->vdev_children; 6695 int c0 = spa_get_random(children); 6696 6697 for (int c = 0; c < children; c++) { 6698 vd = rvd->vdev_child[(c0 + c) % children]; 6699 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6700 continue; 6701 svd[svdcount++] = vd; 6702 if (svdcount == SPA_DVAS_PER_BP) 6703 break; 6704 } 6705 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6706 if (error != 0) 6707 error = vdev_config_sync(svd, svdcount, txg, 6708 B_TRUE); 6709 } else { 6710 error = vdev_config_sync(rvd->vdev_child, 6711 rvd->vdev_children, txg, B_FALSE); 6712 if (error != 0) 6713 error = vdev_config_sync(rvd->vdev_child, 6714 rvd->vdev_children, txg, B_TRUE); 6715 } 6716 6717 if (error == 0) 6718 spa->spa_last_synced_guid = rvd->vdev_guid; 6719 6720 spa_config_exit(spa, SCL_STATE, FTAG); 6721 6722 if (error == 0) 6723 break; 6724 zio_suspend(spa, NULL); 6725 zio_resume_wait(spa); 6726 } 6727 dmu_tx_commit(tx); 6728 6729#ifdef illumos 6730 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6731#else /* !illumos */ 6732#ifdef _KERNEL 6733 callout_drain(&spa->spa_deadman_cycid); 6734#endif 6735#endif /* illumos */ 6736 6737 /* 6738 * Clear the dirty config list. 6739 */ 6740 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6741 vdev_config_clean(vd); 6742 6743 /* 6744 * Now that the new config has synced transactionally, 6745 * let it become visible to the config cache. 6746 */ 6747 if (spa->spa_config_syncing != NULL) { 6748 spa_config_set(spa, spa->spa_config_syncing); 6749 spa->spa_config_txg = txg; 6750 spa->spa_config_syncing = NULL; 6751 } 6752 6753 spa->spa_ubsync = spa->spa_uberblock; 6754 6755 dsl_pool_sync_done(dp, txg); 6756 6757 /* 6758 * Update usable space statistics. 6759 */ 6760 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6761 vdev_sync_done(vd, txg); 6762 6763 spa_update_dspace(spa); 6764 6765 /* 6766 * It had better be the case that we didn't dirty anything 6767 * since vdev_config_sync(). 6768 */ 6769 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6770 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6771 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6772 6773 spa->spa_sync_pass = 0; 6774 6775 spa_config_exit(spa, SCL_CONFIG, FTAG); 6776 6777 spa_handle_ignored_writes(spa); 6778 6779 /* 6780 * If any async tasks have been requested, kick them off. 6781 */ 6782 spa_async_dispatch(spa); 6783 spa_async_dispatch_vd(spa); 6784} 6785 6786/* 6787 * Sync all pools. We don't want to hold the namespace lock across these 6788 * operations, so we take a reference on the spa_t and drop the lock during the 6789 * sync. 6790 */ 6791void 6792spa_sync_allpools(void) 6793{ 6794 spa_t *spa = NULL; 6795 mutex_enter(&spa_namespace_lock); 6796 while ((spa = spa_next(spa)) != NULL) { 6797 if (spa_state(spa) != POOL_STATE_ACTIVE || 6798 !spa_writeable(spa) || spa_suspended(spa)) 6799 continue; 6800 spa_open_ref(spa, FTAG); 6801 mutex_exit(&spa_namespace_lock); 6802 txg_wait_synced(spa_get_dsl(spa), 0); 6803 mutex_enter(&spa_namespace_lock); 6804 spa_close(spa, FTAG); 6805 } 6806 mutex_exit(&spa_namespace_lock); 6807} 6808 6809/* 6810 * ========================================================================== 6811 * Miscellaneous routines 6812 * ========================================================================== 6813 */ 6814 6815/* 6816 * Remove all pools in the system. 6817 */ 6818void 6819spa_evict_all(void) 6820{ 6821 spa_t *spa; 6822 6823 /* 6824 * Remove all cached state. All pools should be closed now, 6825 * so every spa in the AVL tree should be unreferenced. 6826 */ 6827 mutex_enter(&spa_namespace_lock); 6828 while ((spa = spa_next(NULL)) != NULL) { 6829 /* 6830 * Stop async tasks. The async thread may need to detach 6831 * a device that's been replaced, which requires grabbing 6832 * spa_namespace_lock, so we must drop it here. 6833 */ 6834 spa_open_ref(spa, FTAG); 6835 mutex_exit(&spa_namespace_lock); 6836 spa_async_suspend(spa); 6837 mutex_enter(&spa_namespace_lock); 6838 spa_close(spa, FTAG); 6839 6840 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6841 spa_unload(spa); 6842 spa_deactivate(spa); 6843 } 6844 spa_remove(spa); 6845 } 6846 mutex_exit(&spa_namespace_lock); 6847} 6848 6849vdev_t * 6850spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6851{ 6852 vdev_t *vd; 6853 int i; 6854 6855 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6856 return (vd); 6857 6858 if (aux) { 6859 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6860 vd = spa->spa_l2cache.sav_vdevs[i]; 6861 if (vd->vdev_guid == guid) 6862 return (vd); 6863 } 6864 6865 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6866 vd = spa->spa_spares.sav_vdevs[i]; 6867 if (vd->vdev_guid == guid) 6868 return (vd); 6869 } 6870 } 6871 6872 return (NULL); 6873} 6874 6875void 6876spa_upgrade(spa_t *spa, uint64_t version) 6877{ 6878 ASSERT(spa_writeable(spa)); 6879 6880 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6881 6882 /* 6883 * This should only be called for a non-faulted pool, and since a 6884 * future version would result in an unopenable pool, this shouldn't be 6885 * possible. 6886 */ 6887 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6888 ASSERT3U(version, >=, spa->spa_uberblock.ub_version); 6889 6890 spa->spa_uberblock.ub_version = version; 6891 vdev_config_dirty(spa->spa_root_vdev); 6892 6893 spa_config_exit(spa, SCL_ALL, FTAG); 6894 6895 txg_wait_synced(spa_get_dsl(spa), 0); 6896} 6897 6898boolean_t 6899spa_has_spare(spa_t *spa, uint64_t guid) 6900{ 6901 int i; 6902 uint64_t spareguid; 6903 spa_aux_vdev_t *sav = &spa->spa_spares; 6904 6905 for (i = 0; i < sav->sav_count; i++) 6906 if (sav->sav_vdevs[i]->vdev_guid == guid) 6907 return (B_TRUE); 6908 6909 for (i = 0; i < sav->sav_npending; i++) { 6910 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6911 &spareguid) == 0 && spareguid == guid) 6912 return (B_TRUE); 6913 } 6914 6915 return (B_FALSE); 6916} 6917 6918/* 6919 * Check if a pool has an active shared spare device. 6920 * Note: reference count of an active spare is 2, as a spare and as a replace 6921 */ 6922static boolean_t 6923spa_has_active_shared_spare(spa_t *spa) 6924{ 6925 int i, refcnt; 6926 uint64_t pool; 6927 spa_aux_vdev_t *sav = &spa->spa_spares; 6928 6929 for (i = 0; i < sav->sav_count; i++) { 6930 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6931 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6932 refcnt > 2) 6933 return (B_TRUE); 6934 } 6935 6936 return (B_FALSE); 6937} 6938 6939/* 6940 * Post a sysevent corresponding to the given event. The 'name' must be one of 6941 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6942 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6943 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6944 * or zdb as real changes. 6945 */ 6946void 6947spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6948{ 6949#ifdef _KERNEL 6950 sysevent_t *ev; 6951 sysevent_attr_list_t *attr = NULL; 6952 sysevent_value_t value; 6953 sysevent_id_t eid; 6954 6955 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6956 SE_SLEEP); 6957 6958 value.value_type = SE_DATA_TYPE_STRING; 6959 value.value.sv_string = spa_name(spa); 6960 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6961 goto done; 6962 6963 value.value_type = SE_DATA_TYPE_UINT64; 6964 value.value.sv_uint64 = spa_guid(spa); 6965 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6966 goto done; 6967 6968 if (vd) { 6969 value.value_type = SE_DATA_TYPE_UINT64; 6970 value.value.sv_uint64 = vd->vdev_guid; 6971 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6972 SE_SLEEP) != 0) 6973 goto done; 6974 6975 if (vd->vdev_path) { 6976 value.value_type = SE_DATA_TYPE_STRING; 6977 value.value.sv_string = vd->vdev_path; 6978 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6979 &value, SE_SLEEP) != 0) 6980 goto done; 6981 } 6982 } 6983 6984 if (sysevent_attach_attributes(ev, attr) != 0) 6985 goto done; 6986 attr = NULL; 6987 6988 (void) log_sysevent(ev, SE_SLEEP, &eid); 6989 6990done: 6991 if (attr) 6992 sysevent_free_attr(attr); 6993 sysevent_free(ev); 6994#endif 6995} 6996