spa.c revision 260617
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2013 by Delphix. All rights reserved. 25 * Copyright 2013 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 86SYSCTL_DECL(_vfs_zfs); 87TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 88SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 89 "Check hostid on import?"); 90 91/* 92 * The interval, in seconds, at which failed configuration cache file writes 93 * should be retried. 94 */ 95static int zfs_ccw_retry_interval = 300; 96 97typedef enum zti_modes { 98 zti_mode_fixed, /* value is # of threads (min 1) */ 99 zti_mode_online_percent, /* value is % of online CPUs */ 100 zti_mode_batch, /* cpu-intensive; value is ignored */ 101 zti_mode_null, /* don't create a taskq */ 102 zti_nmodes 103} zti_modes_t; 104 105#define ZTI_FIX(n) { zti_mode_fixed, (n) } 106#define ZTI_PCT(n) { zti_mode_online_percent, (n) } 107#define ZTI_BATCH { zti_mode_batch, 0 } 108#define ZTI_NULL { zti_mode_null, 0 } 109 110#define ZTI_ONE ZTI_FIX(1) 111 112typedef struct zio_taskq_info { 113 enum zti_modes zti_mode; 114 uint_t zti_value; 115} zio_taskq_info_t; 116 117static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 118 "issue", "issue_high", "intr", "intr_high" 119}; 120 121/* 122 * Define the taskq threads for the following I/O types: 123 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL 124 */ 125const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 126 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 127 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 128 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, 129 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) }, 130 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL }, 131 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 132 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 133}; 134 135static void spa_sync_version(void *arg, dmu_tx_t *tx); 136static void spa_sync_props(void *arg, dmu_tx_t *tx); 137static boolean_t spa_has_active_shared_spare(spa_t *spa); 138static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 139 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 140 char **ereport); 141static void spa_vdev_resilver_done(spa_t *spa); 142 143uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */ 144#ifdef PSRSET_BIND 145id_t zio_taskq_psrset_bind = PS_NONE; 146#endif 147#ifdef SYSDC 148boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 149#endif 150uint_t zio_taskq_basedc = 80; /* base duty cycle */ 151 152boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 153extern int zfs_sync_pass_deferred_free; 154 155#ifndef illumos 156extern void spa_deadman(void *arg); 157#endif 158 159/* 160 * This (illegal) pool name is used when temporarily importing a spa_t in order 161 * to get the vdev stats associated with the imported devices. 162 */ 163#define TRYIMPORT_NAME "$import" 164 165/* 166 * ========================================================================== 167 * SPA properties routines 168 * ========================================================================== 169 */ 170 171/* 172 * Add a (source=src, propname=propval) list to an nvlist. 173 */ 174static void 175spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 176 uint64_t intval, zprop_source_t src) 177{ 178 const char *propname = zpool_prop_to_name(prop); 179 nvlist_t *propval; 180 181 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 182 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 183 184 if (strval != NULL) 185 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 186 else 187 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 188 189 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 190 nvlist_free(propval); 191} 192 193/* 194 * Get property values from the spa configuration. 195 */ 196static void 197spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 198{ 199 vdev_t *rvd = spa->spa_root_vdev; 200 dsl_pool_t *pool = spa->spa_dsl_pool; 201 uint64_t size; 202 uint64_t alloc; 203 uint64_t space; 204 uint64_t cap, version; 205 zprop_source_t src = ZPROP_SRC_NONE; 206 spa_config_dirent_t *dp; 207 208 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 209 210 if (rvd != NULL) { 211 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 212 size = metaslab_class_get_space(spa_normal_class(spa)); 213 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 214 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 215 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 216 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 217 size - alloc, src); 218 219 space = 0; 220 for (int c = 0; c < rvd->vdev_children; c++) { 221 vdev_t *tvd = rvd->vdev_child[c]; 222 space += tvd->vdev_max_asize - tvd->vdev_asize; 223 } 224 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space, 225 src); 226 227 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 228 (spa_mode(spa) == FREAD), src); 229 230 cap = (size == 0) ? 0 : (alloc * 100 / size); 231 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 232 233 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 234 ddt_get_pool_dedup_ratio(spa), src); 235 236 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 237 rvd->vdev_state, src); 238 239 version = spa_version(spa); 240 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 241 src = ZPROP_SRC_DEFAULT; 242 else 243 src = ZPROP_SRC_LOCAL; 244 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 245 } 246 247 if (pool != NULL) { 248 dsl_dir_t *freedir = pool->dp_free_dir; 249 250 /* 251 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 252 * when opening pools before this version freedir will be NULL. 253 */ 254 if (freedir != NULL) { 255 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 256 freedir->dd_phys->dd_used_bytes, src); 257 } else { 258 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 259 NULL, 0, src); 260 } 261 } 262 263 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 264 265 if (spa->spa_comment != NULL) { 266 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 267 0, ZPROP_SRC_LOCAL); 268 } 269 270 if (spa->spa_root != NULL) 271 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 272 0, ZPROP_SRC_LOCAL); 273 274 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 275 if (dp->scd_path == NULL) { 276 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 277 "none", 0, ZPROP_SRC_LOCAL); 278 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 279 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 280 dp->scd_path, 0, ZPROP_SRC_LOCAL); 281 } 282 } 283} 284 285/* 286 * Get zpool property values. 287 */ 288int 289spa_prop_get(spa_t *spa, nvlist_t **nvp) 290{ 291 objset_t *mos = spa->spa_meta_objset; 292 zap_cursor_t zc; 293 zap_attribute_t za; 294 int err; 295 296 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 297 298 mutex_enter(&spa->spa_props_lock); 299 300 /* 301 * Get properties from the spa config. 302 */ 303 spa_prop_get_config(spa, nvp); 304 305 /* If no pool property object, no more prop to get. */ 306 if (mos == NULL || spa->spa_pool_props_object == 0) { 307 mutex_exit(&spa->spa_props_lock); 308 return (0); 309 } 310 311 /* 312 * Get properties from the MOS pool property object. 313 */ 314 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 315 (err = zap_cursor_retrieve(&zc, &za)) == 0; 316 zap_cursor_advance(&zc)) { 317 uint64_t intval = 0; 318 char *strval = NULL; 319 zprop_source_t src = ZPROP_SRC_DEFAULT; 320 zpool_prop_t prop; 321 322 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 323 continue; 324 325 switch (za.za_integer_length) { 326 case 8: 327 /* integer property */ 328 if (za.za_first_integer != 329 zpool_prop_default_numeric(prop)) 330 src = ZPROP_SRC_LOCAL; 331 332 if (prop == ZPOOL_PROP_BOOTFS) { 333 dsl_pool_t *dp; 334 dsl_dataset_t *ds = NULL; 335 336 dp = spa_get_dsl(spa); 337 dsl_pool_config_enter(dp, FTAG); 338 if (err = dsl_dataset_hold_obj(dp, 339 za.za_first_integer, FTAG, &ds)) { 340 dsl_pool_config_exit(dp, FTAG); 341 break; 342 } 343 344 strval = kmem_alloc( 345 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 346 KM_SLEEP); 347 dsl_dataset_name(ds, strval); 348 dsl_dataset_rele(ds, FTAG); 349 dsl_pool_config_exit(dp, FTAG); 350 } else { 351 strval = NULL; 352 intval = za.za_first_integer; 353 } 354 355 spa_prop_add_list(*nvp, prop, strval, intval, src); 356 357 if (strval != NULL) 358 kmem_free(strval, 359 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 360 361 break; 362 363 case 1: 364 /* string property */ 365 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 366 err = zap_lookup(mos, spa->spa_pool_props_object, 367 za.za_name, 1, za.za_num_integers, strval); 368 if (err) { 369 kmem_free(strval, za.za_num_integers); 370 break; 371 } 372 spa_prop_add_list(*nvp, prop, strval, 0, src); 373 kmem_free(strval, za.za_num_integers); 374 break; 375 376 default: 377 break; 378 } 379 } 380 zap_cursor_fini(&zc); 381 mutex_exit(&spa->spa_props_lock); 382out: 383 if (err && err != ENOENT) { 384 nvlist_free(*nvp); 385 *nvp = NULL; 386 return (err); 387 } 388 389 return (0); 390} 391 392/* 393 * Validate the given pool properties nvlist and modify the list 394 * for the property values to be set. 395 */ 396static int 397spa_prop_validate(spa_t *spa, nvlist_t *props) 398{ 399 nvpair_t *elem; 400 int error = 0, reset_bootfs = 0; 401 uint64_t objnum = 0; 402 boolean_t has_feature = B_FALSE; 403 404 elem = NULL; 405 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 406 uint64_t intval; 407 char *strval, *slash, *check, *fname; 408 const char *propname = nvpair_name(elem); 409 zpool_prop_t prop = zpool_name_to_prop(propname); 410 411 switch (prop) { 412 case ZPROP_INVAL: 413 if (!zpool_prop_feature(propname)) { 414 error = SET_ERROR(EINVAL); 415 break; 416 } 417 418 /* 419 * Sanitize the input. 420 */ 421 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 422 error = SET_ERROR(EINVAL); 423 break; 424 } 425 426 if (nvpair_value_uint64(elem, &intval) != 0) { 427 error = SET_ERROR(EINVAL); 428 break; 429 } 430 431 if (intval != 0) { 432 error = SET_ERROR(EINVAL); 433 break; 434 } 435 436 fname = strchr(propname, '@') + 1; 437 if (zfeature_lookup_name(fname, NULL) != 0) { 438 error = SET_ERROR(EINVAL); 439 break; 440 } 441 442 has_feature = B_TRUE; 443 break; 444 445 case ZPOOL_PROP_VERSION: 446 error = nvpair_value_uint64(elem, &intval); 447 if (!error && 448 (intval < spa_version(spa) || 449 intval > SPA_VERSION_BEFORE_FEATURES || 450 has_feature)) 451 error = SET_ERROR(EINVAL); 452 break; 453 454 case ZPOOL_PROP_DELEGATION: 455 case ZPOOL_PROP_AUTOREPLACE: 456 case ZPOOL_PROP_LISTSNAPS: 457 case ZPOOL_PROP_AUTOEXPAND: 458 error = nvpair_value_uint64(elem, &intval); 459 if (!error && intval > 1) 460 error = SET_ERROR(EINVAL); 461 break; 462 463 case ZPOOL_PROP_BOOTFS: 464 /* 465 * If the pool version is less than SPA_VERSION_BOOTFS, 466 * or the pool is still being created (version == 0), 467 * the bootfs property cannot be set. 468 */ 469 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 470 error = SET_ERROR(ENOTSUP); 471 break; 472 } 473 474 /* 475 * Make sure the vdev config is bootable 476 */ 477 if (!vdev_is_bootable(spa->spa_root_vdev)) { 478 error = SET_ERROR(ENOTSUP); 479 break; 480 } 481 482 reset_bootfs = 1; 483 484 error = nvpair_value_string(elem, &strval); 485 486 if (!error) { 487 objset_t *os; 488 uint64_t compress; 489 490 if (strval == NULL || strval[0] == '\0') { 491 objnum = zpool_prop_default_numeric( 492 ZPOOL_PROP_BOOTFS); 493 break; 494 } 495 496 if (error = dmu_objset_hold(strval, FTAG, &os)) 497 break; 498 499 /* Must be ZPL and not gzip compressed. */ 500 501 if (dmu_objset_type(os) != DMU_OST_ZFS) { 502 error = SET_ERROR(ENOTSUP); 503 } else if ((error = 504 dsl_prop_get_int_ds(dmu_objset_ds(os), 505 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 506 &compress)) == 0 && 507 !BOOTFS_COMPRESS_VALID(compress)) { 508 error = SET_ERROR(ENOTSUP); 509 } else { 510 objnum = dmu_objset_id(os); 511 } 512 dmu_objset_rele(os, FTAG); 513 } 514 break; 515 516 case ZPOOL_PROP_FAILUREMODE: 517 error = nvpair_value_uint64(elem, &intval); 518 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 519 intval > ZIO_FAILURE_MODE_PANIC)) 520 error = SET_ERROR(EINVAL); 521 522 /* 523 * This is a special case which only occurs when 524 * the pool has completely failed. This allows 525 * the user to change the in-core failmode property 526 * without syncing it out to disk (I/Os might 527 * currently be blocked). We do this by returning 528 * EIO to the caller (spa_prop_set) to trick it 529 * into thinking we encountered a property validation 530 * error. 531 */ 532 if (!error && spa_suspended(spa)) { 533 spa->spa_failmode = intval; 534 error = SET_ERROR(EIO); 535 } 536 break; 537 538 case ZPOOL_PROP_CACHEFILE: 539 if ((error = nvpair_value_string(elem, &strval)) != 0) 540 break; 541 542 if (strval[0] == '\0') 543 break; 544 545 if (strcmp(strval, "none") == 0) 546 break; 547 548 if (strval[0] != '/') { 549 error = SET_ERROR(EINVAL); 550 break; 551 } 552 553 slash = strrchr(strval, '/'); 554 ASSERT(slash != NULL); 555 556 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 557 strcmp(slash, "/..") == 0) 558 error = SET_ERROR(EINVAL); 559 break; 560 561 case ZPOOL_PROP_COMMENT: 562 if ((error = nvpair_value_string(elem, &strval)) != 0) 563 break; 564 for (check = strval; *check != '\0'; check++) { 565 /* 566 * The kernel doesn't have an easy isprint() 567 * check. For this kernel check, we merely 568 * check ASCII apart from DEL. Fix this if 569 * there is an easy-to-use kernel isprint(). 570 */ 571 if (*check >= 0x7f) { 572 error = SET_ERROR(EINVAL); 573 break; 574 } 575 check++; 576 } 577 if (strlen(strval) > ZPROP_MAX_COMMENT) 578 error = E2BIG; 579 break; 580 581 case ZPOOL_PROP_DEDUPDITTO: 582 if (spa_version(spa) < SPA_VERSION_DEDUP) 583 error = SET_ERROR(ENOTSUP); 584 else 585 error = nvpair_value_uint64(elem, &intval); 586 if (error == 0 && 587 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 588 error = SET_ERROR(EINVAL); 589 break; 590 } 591 592 if (error) 593 break; 594 } 595 596 if (!error && reset_bootfs) { 597 error = nvlist_remove(props, 598 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 599 600 if (!error) { 601 error = nvlist_add_uint64(props, 602 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 603 } 604 } 605 606 return (error); 607} 608 609void 610spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 611{ 612 char *cachefile; 613 spa_config_dirent_t *dp; 614 615 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 616 &cachefile) != 0) 617 return; 618 619 dp = kmem_alloc(sizeof (spa_config_dirent_t), 620 KM_SLEEP); 621 622 if (cachefile[0] == '\0') 623 dp->scd_path = spa_strdup(spa_config_path); 624 else if (strcmp(cachefile, "none") == 0) 625 dp->scd_path = NULL; 626 else 627 dp->scd_path = spa_strdup(cachefile); 628 629 list_insert_head(&spa->spa_config_list, dp); 630 if (need_sync) 631 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 632} 633 634int 635spa_prop_set(spa_t *spa, nvlist_t *nvp) 636{ 637 int error; 638 nvpair_t *elem = NULL; 639 boolean_t need_sync = B_FALSE; 640 641 if ((error = spa_prop_validate(spa, nvp)) != 0) 642 return (error); 643 644 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 645 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 646 647 if (prop == ZPOOL_PROP_CACHEFILE || 648 prop == ZPOOL_PROP_ALTROOT || 649 prop == ZPOOL_PROP_READONLY) 650 continue; 651 652 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 653 uint64_t ver; 654 655 if (prop == ZPOOL_PROP_VERSION) { 656 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 657 } else { 658 ASSERT(zpool_prop_feature(nvpair_name(elem))); 659 ver = SPA_VERSION_FEATURES; 660 need_sync = B_TRUE; 661 } 662 663 /* Save time if the version is already set. */ 664 if (ver == spa_version(spa)) 665 continue; 666 667 /* 668 * In addition to the pool directory object, we might 669 * create the pool properties object, the features for 670 * read object, the features for write object, or the 671 * feature descriptions object. 672 */ 673 error = dsl_sync_task(spa->spa_name, NULL, 674 spa_sync_version, &ver, 6); 675 if (error) 676 return (error); 677 continue; 678 } 679 680 need_sync = B_TRUE; 681 break; 682 } 683 684 if (need_sync) { 685 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 686 nvp, 6)); 687 } 688 689 return (0); 690} 691 692/* 693 * If the bootfs property value is dsobj, clear it. 694 */ 695void 696spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 697{ 698 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 699 VERIFY(zap_remove(spa->spa_meta_objset, 700 spa->spa_pool_props_object, 701 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 702 spa->spa_bootfs = 0; 703 } 704} 705 706/*ARGSUSED*/ 707static int 708spa_change_guid_check(void *arg, dmu_tx_t *tx) 709{ 710 uint64_t *newguid = arg; 711 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 712 vdev_t *rvd = spa->spa_root_vdev; 713 uint64_t vdev_state; 714 715 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 716 vdev_state = rvd->vdev_state; 717 spa_config_exit(spa, SCL_STATE, FTAG); 718 719 if (vdev_state != VDEV_STATE_HEALTHY) 720 return (SET_ERROR(ENXIO)); 721 722 ASSERT3U(spa_guid(spa), !=, *newguid); 723 724 return (0); 725} 726 727static void 728spa_change_guid_sync(void *arg, dmu_tx_t *tx) 729{ 730 uint64_t *newguid = arg; 731 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 732 uint64_t oldguid; 733 vdev_t *rvd = spa->spa_root_vdev; 734 735 oldguid = spa_guid(spa); 736 737 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 738 rvd->vdev_guid = *newguid; 739 rvd->vdev_guid_sum += (*newguid - oldguid); 740 vdev_config_dirty(rvd); 741 spa_config_exit(spa, SCL_STATE, FTAG); 742 743 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 744 oldguid, *newguid); 745} 746 747/* 748 * Change the GUID for the pool. This is done so that we can later 749 * re-import a pool built from a clone of our own vdevs. We will modify 750 * the root vdev's guid, our own pool guid, and then mark all of our 751 * vdevs dirty. Note that we must make sure that all our vdevs are 752 * online when we do this, or else any vdevs that weren't present 753 * would be orphaned from our pool. We are also going to issue a 754 * sysevent to update any watchers. 755 */ 756int 757spa_change_guid(spa_t *spa) 758{ 759 int error; 760 uint64_t guid; 761 762 mutex_enter(&spa->spa_vdev_top_lock); 763 mutex_enter(&spa_namespace_lock); 764 guid = spa_generate_guid(NULL); 765 766 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 767 spa_change_guid_sync, &guid, 5); 768 769 if (error == 0) { 770 spa_config_sync(spa, B_FALSE, B_TRUE); 771 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 772 } 773 774 mutex_exit(&spa_namespace_lock); 775 mutex_exit(&spa->spa_vdev_top_lock); 776 777 return (error); 778} 779 780/* 781 * ========================================================================== 782 * SPA state manipulation (open/create/destroy/import/export) 783 * ========================================================================== 784 */ 785 786static int 787spa_error_entry_compare(const void *a, const void *b) 788{ 789 spa_error_entry_t *sa = (spa_error_entry_t *)a; 790 spa_error_entry_t *sb = (spa_error_entry_t *)b; 791 int ret; 792 793 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 794 sizeof (zbookmark_t)); 795 796 if (ret < 0) 797 return (-1); 798 else if (ret > 0) 799 return (1); 800 else 801 return (0); 802} 803 804/* 805 * Utility function which retrieves copies of the current logs and 806 * re-initializes them in the process. 807 */ 808void 809spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 810{ 811 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 812 813 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 814 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 815 816 avl_create(&spa->spa_errlist_scrub, 817 spa_error_entry_compare, sizeof (spa_error_entry_t), 818 offsetof(spa_error_entry_t, se_avl)); 819 avl_create(&spa->spa_errlist_last, 820 spa_error_entry_compare, sizeof (spa_error_entry_t), 821 offsetof(spa_error_entry_t, se_avl)); 822} 823 824static taskq_t * 825spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode, 826 uint_t value) 827{ 828 uint_t flags = TASKQ_PREPOPULATE; 829 boolean_t batch = B_FALSE; 830 831 switch (mode) { 832 case zti_mode_null: 833 return (NULL); /* no taskq needed */ 834 835 case zti_mode_fixed: 836 ASSERT3U(value, >=, 1); 837 value = MAX(value, 1); 838 break; 839 840 case zti_mode_batch: 841 batch = B_TRUE; 842 flags |= TASKQ_THREADS_CPU_PCT; 843 value = zio_taskq_batch_pct; 844 break; 845 846 case zti_mode_online_percent: 847 flags |= TASKQ_THREADS_CPU_PCT; 848 break; 849 850 default: 851 panic("unrecognized mode for %s taskq (%u:%u) in " 852 "spa_activate()", 853 name, mode, value); 854 break; 855 } 856 857#ifdef SYSDC 858 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 859 if (batch) 860 flags |= TASKQ_DC_BATCH; 861 862 return (taskq_create_sysdc(name, value, 50, INT_MAX, 863 spa->spa_proc, zio_taskq_basedc, flags)); 864 } 865#endif 866 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX, 867 spa->spa_proc, flags)); 868} 869 870static void 871spa_create_zio_taskqs(spa_t *spa) 872{ 873 for (int t = 0; t < ZIO_TYPES; t++) { 874 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 875 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 876 enum zti_modes mode = ztip->zti_mode; 877 uint_t value = ztip->zti_value; 878 char name[32]; 879 880 (void) snprintf(name, sizeof (name), 881 "%s_%s", zio_type_name[t], zio_taskq_types[q]); 882 883 spa->spa_zio_taskq[t][q] = 884 spa_taskq_create(spa, name, mode, value); 885 } 886 } 887} 888 889#ifdef _KERNEL 890#ifdef SPA_PROCESS 891static void 892spa_thread(void *arg) 893{ 894 callb_cpr_t cprinfo; 895 896 spa_t *spa = arg; 897 user_t *pu = PTOU(curproc); 898 899 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 900 spa->spa_name); 901 902 ASSERT(curproc != &p0); 903 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 904 "zpool-%s", spa->spa_name); 905 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 906 907#ifdef PSRSET_BIND 908 /* bind this thread to the requested psrset */ 909 if (zio_taskq_psrset_bind != PS_NONE) { 910 pool_lock(); 911 mutex_enter(&cpu_lock); 912 mutex_enter(&pidlock); 913 mutex_enter(&curproc->p_lock); 914 915 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 916 0, NULL, NULL) == 0) { 917 curthread->t_bind_pset = zio_taskq_psrset_bind; 918 } else { 919 cmn_err(CE_WARN, 920 "Couldn't bind process for zfs pool \"%s\" to " 921 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 922 } 923 924 mutex_exit(&curproc->p_lock); 925 mutex_exit(&pidlock); 926 mutex_exit(&cpu_lock); 927 pool_unlock(); 928 } 929#endif 930 931#ifdef SYSDC 932 if (zio_taskq_sysdc) { 933 sysdc_thread_enter(curthread, 100, 0); 934 } 935#endif 936 937 spa->spa_proc = curproc; 938 spa->spa_did = curthread->t_did; 939 940 spa_create_zio_taskqs(spa); 941 942 mutex_enter(&spa->spa_proc_lock); 943 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 944 945 spa->spa_proc_state = SPA_PROC_ACTIVE; 946 cv_broadcast(&spa->spa_proc_cv); 947 948 CALLB_CPR_SAFE_BEGIN(&cprinfo); 949 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 950 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 951 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 952 953 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 954 spa->spa_proc_state = SPA_PROC_GONE; 955 spa->spa_proc = &p0; 956 cv_broadcast(&spa->spa_proc_cv); 957 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 958 959 mutex_enter(&curproc->p_lock); 960 lwp_exit(); 961} 962#endif /* SPA_PROCESS */ 963#endif 964 965/* 966 * Activate an uninitialized pool. 967 */ 968static void 969spa_activate(spa_t *spa, int mode) 970{ 971 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 972 973 spa->spa_state = POOL_STATE_ACTIVE; 974 spa->spa_mode = mode; 975 976 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 977 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 978 979 /* Try to create a covering process */ 980 mutex_enter(&spa->spa_proc_lock); 981 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 982 ASSERT(spa->spa_proc == &p0); 983 spa->spa_did = 0; 984 985#ifdef SPA_PROCESS 986 /* Only create a process if we're going to be around a while. */ 987 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 988 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 989 NULL, 0) == 0) { 990 spa->spa_proc_state = SPA_PROC_CREATED; 991 while (spa->spa_proc_state == SPA_PROC_CREATED) { 992 cv_wait(&spa->spa_proc_cv, 993 &spa->spa_proc_lock); 994 } 995 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 996 ASSERT(spa->spa_proc != &p0); 997 ASSERT(spa->spa_did != 0); 998 } else { 999#ifdef _KERNEL 1000 cmn_err(CE_WARN, 1001 "Couldn't create process for zfs pool \"%s\"\n", 1002 spa->spa_name); 1003#endif 1004 } 1005 } 1006#endif /* SPA_PROCESS */ 1007 mutex_exit(&spa->spa_proc_lock); 1008 1009 /* If we didn't create a process, we need to create our taskqs. */ 1010 ASSERT(spa->spa_proc == &p0); 1011 if (spa->spa_proc == &p0) { 1012 spa_create_zio_taskqs(spa); 1013 } 1014 1015 /* 1016 * Start TRIM thread. 1017 */ 1018 trim_thread_create(spa); 1019 1020 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1021 offsetof(vdev_t, vdev_config_dirty_node)); 1022 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1023 offsetof(vdev_t, vdev_state_dirty_node)); 1024 1025 txg_list_create(&spa->spa_vdev_txg_list, 1026 offsetof(struct vdev, vdev_txg_node)); 1027 1028 avl_create(&spa->spa_errlist_scrub, 1029 spa_error_entry_compare, sizeof (spa_error_entry_t), 1030 offsetof(spa_error_entry_t, se_avl)); 1031 avl_create(&spa->spa_errlist_last, 1032 spa_error_entry_compare, sizeof (spa_error_entry_t), 1033 offsetof(spa_error_entry_t, se_avl)); 1034} 1035 1036/* 1037 * Opposite of spa_activate(). 1038 */ 1039static void 1040spa_deactivate(spa_t *spa) 1041{ 1042 ASSERT(spa->spa_sync_on == B_FALSE); 1043 ASSERT(spa->spa_dsl_pool == NULL); 1044 ASSERT(spa->spa_root_vdev == NULL); 1045 ASSERT(spa->spa_async_zio_root == NULL); 1046 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1047 1048 /* 1049 * Stop TRIM thread in case spa_unload() wasn't called directly 1050 * before spa_deactivate(). 1051 */ 1052 trim_thread_destroy(spa); 1053 1054 txg_list_destroy(&spa->spa_vdev_txg_list); 1055 1056 list_destroy(&spa->spa_config_dirty_list); 1057 list_destroy(&spa->spa_state_dirty_list); 1058 1059 for (int t = 0; t < ZIO_TYPES; t++) { 1060 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1061 if (spa->spa_zio_taskq[t][q] != NULL) 1062 taskq_destroy(spa->spa_zio_taskq[t][q]); 1063 spa->spa_zio_taskq[t][q] = NULL; 1064 } 1065 } 1066 1067 metaslab_class_destroy(spa->spa_normal_class); 1068 spa->spa_normal_class = NULL; 1069 1070 metaslab_class_destroy(spa->spa_log_class); 1071 spa->spa_log_class = NULL; 1072 1073 /* 1074 * If this was part of an import or the open otherwise failed, we may 1075 * still have errors left in the queues. Empty them just in case. 1076 */ 1077 spa_errlog_drain(spa); 1078 1079 avl_destroy(&spa->spa_errlist_scrub); 1080 avl_destroy(&spa->spa_errlist_last); 1081 1082 spa->spa_state = POOL_STATE_UNINITIALIZED; 1083 1084 mutex_enter(&spa->spa_proc_lock); 1085 if (spa->spa_proc_state != SPA_PROC_NONE) { 1086 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1087 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1088 cv_broadcast(&spa->spa_proc_cv); 1089 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1090 ASSERT(spa->spa_proc != &p0); 1091 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1092 } 1093 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1094 spa->spa_proc_state = SPA_PROC_NONE; 1095 } 1096 ASSERT(spa->spa_proc == &p0); 1097 mutex_exit(&spa->spa_proc_lock); 1098 1099#ifdef SPA_PROCESS 1100 /* 1101 * We want to make sure spa_thread() has actually exited the ZFS 1102 * module, so that the module can't be unloaded out from underneath 1103 * it. 1104 */ 1105 if (spa->spa_did != 0) { 1106 thread_join(spa->spa_did); 1107 spa->spa_did = 0; 1108 } 1109#endif /* SPA_PROCESS */ 1110} 1111 1112/* 1113 * Verify a pool configuration, and construct the vdev tree appropriately. This 1114 * will create all the necessary vdevs in the appropriate layout, with each vdev 1115 * in the CLOSED state. This will prep the pool before open/creation/import. 1116 * All vdev validation is done by the vdev_alloc() routine. 1117 */ 1118static int 1119spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1120 uint_t id, int atype) 1121{ 1122 nvlist_t **child; 1123 uint_t children; 1124 int error; 1125 1126 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1127 return (error); 1128 1129 if ((*vdp)->vdev_ops->vdev_op_leaf) 1130 return (0); 1131 1132 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1133 &child, &children); 1134 1135 if (error == ENOENT) 1136 return (0); 1137 1138 if (error) { 1139 vdev_free(*vdp); 1140 *vdp = NULL; 1141 return (SET_ERROR(EINVAL)); 1142 } 1143 1144 for (int c = 0; c < children; c++) { 1145 vdev_t *vd; 1146 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1147 atype)) != 0) { 1148 vdev_free(*vdp); 1149 *vdp = NULL; 1150 return (error); 1151 } 1152 } 1153 1154 ASSERT(*vdp != NULL); 1155 1156 return (0); 1157} 1158 1159/* 1160 * Opposite of spa_load(). 1161 */ 1162static void 1163spa_unload(spa_t *spa) 1164{ 1165 int i; 1166 1167 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1168 1169 /* 1170 * Stop TRIM thread. 1171 */ 1172 trim_thread_destroy(spa); 1173 1174 /* 1175 * Stop async tasks. 1176 */ 1177 spa_async_suspend(spa); 1178 1179 /* 1180 * Stop syncing. 1181 */ 1182 if (spa->spa_sync_on) { 1183 txg_sync_stop(spa->spa_dsl_pool); 1184 spa->spa_sync_on = B_FALSE; 1185 } 1186 1187 /* 1188 * Wait for any outstanding async I/O to complete. 1189 */ 1190 if (spa->spa_async_zio_root != NULL) { 1191 (void) zio_wait(spa->spa_async_zio_root); 1192 spa->spa_async_zio_root = NULL; 1193 } 1194 1195 bpobj_close(&spa->spa_deferred_bpobj); 1196 1197 /* 1198 * Close the dsl pool. 1199 */ 1200 if (spa->spa_dsl_pool) { 1201 dsl_pool_close(spa->spa_dsl_pool); 1202 spa->spa_dsl_pool = NULL; 1203 spa->spa_meta_objset = NULL; 1204 } 1205 1206 ddt_unload(spa); 1207 1208 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1209 1210 /* 1211 * Drop and purge level 2 cache 1212 */ 1213 spa_l2cache_drop(spa); 1214 1215 /* 1216 * Close all vdevs. 1217 */ 1218 if (spa->spa_root_vdev) 1219 vdev_free(spa->spa_root_vdev); 1220 ASSERT(spa->spa_root_vdev == NULL); 1221 1222 for (i = 0; i < spa->spa_spares.sav_count; i++) 1223 vdev_free(spa->spa_spares.sav_vdevs[i]); 1224 if (spa->spa_spares.sav_vdevs) { 1225 kmem_free(spa->spa_spares.sav_vdevs, 1226 spa->spa_spares.sav_count * sizeof (void *)); 1227 spa->spa_spares.sav_vdevs = NULL; 1228 } 1229 if (spa->spa_spares.sav_config) { 1230 nvlist_free(spa->spa_spares.sav_config); 1231 spa->spa_spares.sav_config = NULL; 1232 } 1233 spa->spa_spares.sav_count = 0; 1234 1235 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1236 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1237 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1238 } 1239 if (spa->spa_l2cache.sav_vdevs) { 1240 kmem_free(spa->spa_l2cache.sav_vdevs, 1241 spa->spa_l2cache.sav_count * sizeof (void *)); 1242 spa->spa_l2cache.sav_vdevs = NULL; 1243 } 1244 if (spa->spa_l2cache.sav_config) { 1245 nvlist_free(spa->spa_l2cache.sav_config); 1246 spa->spa_l2cache.sav_config = NULL; 1247 } 1248 spa->spa_l2cache.sav_count = 0; 1249 1250 spa->spa_async_suspended = 0; 1251 1252 if (spa->spa_comment != NULL) { 1253 spa_strfree(spa->spa_comment); 1254 spa->spa_comment = NULL; 1255 } 1256 1257 spa_config_exit(spa, SCL_ALL, FTAG); 1258} 1259 1260/* 1261 * Load (or re-load) the current list of vdevs describing the active spares for 1262 * this pool. When this is called, we have some form of basic information in 1263 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1264 * then re-generate a more complete list including status information. 1265 */ 1266static void 1267spa_load_spares(spa_t *spa) 1268{ 1269 nvlist_t **spares; 1270 uint_t nspares; 1271 int i; 1272 vdev_t *vd, *tvd; 1273 1274 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1275 1276 /* 1277 * First, close and free any existing spare vdevs. 1278 */ 1279 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1280 vd = spa->spa_spares.sav_vdevs[i]; 1281 1282 /* Undo the call to spa_activate() below */ 1283 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1284 B_FALSE)) != NULL && tvd->vdev_isspare) 1285 spa_spare_remove(tvd); 1286 vdev_close(vd); 1287 vdev_free(vd); 1288 } 1289 1290 if (spa->spa_spares.sav_vdevs) 1291 kmem_free(spa->spa_spares.sav_vdevs, 1292 spa->spa_spares.sav_count * sizeof (void *)); 1293 1294 if (spa->spa_spares.sav_config == NULL) 1295 nspares = 0; 1296 else 1297 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1298 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1299 1300 spa->spa_spares.sav_count = (int)nspares; 1301 spa->spa_spares.sav_vdevs = NULL; 1302 1303 if (nspares == 0) 1304 return; 1305 1306 /* 1307 * Construct the array of vdevs, opening them to get status in the 1308 * process. For each spare, there is potentially two different vdev_t 1309 * structures associated with it: one in the list of spares (used only 1310 * for basic validation purposes) and one in the active vdev 1311 * configuration (if it's spared in). During this phase we open and 1312 * validate each vdev on the spare list. If the vdev also exists in the 1313 * active configuration, then we also mark this vdev as an active spare. 1314 */ 1315 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1316 KM_SLEEP); 1317 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1318 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1319 VDEV_ALLOC_SPARE) == 0); 1320 ASSERT(vd != NULL); 1321 1322 spa->spa_spares.sav_vdevs[i] = vd; 1323 1324 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1325 B_FALSE)) != NULL) { 1326 if (!tvd->vdev_isspare) 1327 spa_spare_add(tvd); 1328 1329 /* 1330 * We only mark the spare active if we were successfully 1331 * able to load the vdev. Otherwise, importing a pool 1332 * with a bad active spare would result in strange 1333 * behavior, because multiple pool would think the spare 1334 * is actively in use. 1335 * 1336 * There is a vulnerability here to an equally bizarre 1337 * circumstance, where a dead active spare is later 1338 * brought back to life (onlined or otherwise). Given 1339 * the rarity of this scenario, and the extra complexity 1340 * it adds, we ignore the possibility. 1341 */ 1342 if (!vdev_is_dead(tvd)) 1343 spa_spare_activate(tvd); 1344 } 1345 1346 vd->vdev_top = vd; 1347 vd->vdev_aux = &spa->spa_spares; 1348 1349 if (vdev_open(vd) != 0) 1350 continue; 1351 1352 if (vdev_validate_aux(vd) == 0) 1353 spa_spare_add(vd); 1354 } 1355 1356 /* 1357 * Recompute the stashed list of spares, with status information 1358 * this time. 1359 */ 1360 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1361 DATA_TYPE_NVLIST_ARRAY) == 0); 1362 1363 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1364 KM_SLEEP); 1365 for (i = 0; i < spa->spa_spares.sav_count; i++) 1366 spares[i] = vdev_config_generate(spa, 1367 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1368 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1369 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1370 for (i = 0; i < spa->spa_spares.sav_count; i++) 1371 nvlist_free(spares[i]); 1372 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1373} 1374 1375/* 1376 * Load (or re-load) the current list of vdevs describing the active l2cache for 1377 * this pool. When this is called, we have some form of basic information in 1378 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1379 * then re-generate a more complete list including status information. 1380 * Devices which are already active have their details maintained, and are 1381 * not re-opened. 1382 */ 1383static void 1384spa_load_l2cache(spa_t *spa) 1385{ 1386 nvlist_t **l2cache; 1387 uint_t nl2cache; 1388 int i, j, oldnvdevs; 1389 uint64_t guid; 1390 vdev_t *vd, **oldvdevs, **newvdevs; 1391 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1392 1393 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1394 1395 if (sav->sav_config != NULL) { 1396 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1397 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1398 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1399 } else { 1400 nl2cache = 0; 1401 newvdevs = NULL; 1402 } 1403 1404 oldvdevs = sav->sav_vdevs; 1405 oldnvdevs = sav->sav_count; 1406 sav->sav_vdevs = NULL; 1407 sav->sav_count = 0; 1408 1409 /* 1410 * Process new nvlist of vdevs. 1411 */ 1412 for (i = 0; i < nl2cache; i++) { 1413 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1414 &guid) == 0); 1415 1416 newvdevs[i] = NULL; 1417 for (j = 0; j < oldnvdevs; j++) { 1418 vd = oldvdevs[j]; 1419 if (vd != NULL && guid == vd->vdev_guid) { 1420 /* 1421 * Retain previous vdev for add/remove ops. 1422 */ 1423 newvdevs[i] = vd; 1424 oldvdevs[j] = NULL; 1425 break; 1426 } 1427 } 1428 1429 if (newvdevs[i] == NULL) { 1430 /* 1431 * Create new vdev 1432 */ 1433 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1434 VDEV_ALLOC_L2CACHE) == 0); 1435 ASSERT(vd != NULL); 1436 newvdevs[i] = vd; 1437 1438 /* 1439 * Commit this vdev as an l2cache device, 1440 * even if it fails to open. 1441 */ 1442 spa_l2cache_add(vd); 1443 1444 vd->vdev_top = vd; 1445 vd->vdev_aux = sav; 1446 1447 spa_l2cache_activate(vd); 1448 1449 if (vdev_open(vd) != 0) 1450 continue; 1451 1452 (void) vdev_validate_aux(vd); 1453 1454 if (!vdev_is_dead(vd)) 1455 l2arc_add_vdev(spa, vd); 1456 } 1457 } 1458 1459 /* 1460 * Purge vdevs that were dropped 1461 */ 1462 for (i = 0; i < oldnvdevs; i++) { 1463 uint64_t pool; 1464 1465 vd = oldvdevs[i]; 1466 if (vd != NULL) { 1467 ASSERT(vd->vdev_isl2cache); 1468 1469 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1470 pool != 0ULL && l2arc_vdev_present(vd)) 1471 l2arc_remove_vdev(vd); 1472 vdev_clear_stats(vd); 1473 vdev_free(vd); 1474 } 1475 } 1476 1477 if (oldvdevs) 1478 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1479 1480 if (sav->sav_config == NULL) 1481 goto out; 1482 1483 sav->sav_vdevs = newvdevs; 1484 sav->sav_count = (int)nl2cache; 1485 1486 /* 1487 * Recompute the stashed list of l2cache devices, with status 1488 * information this time. 1489 */ 1490 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1491 DATA_TYPE_NVLIST_ARRAY) == 0); 1492 1493 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1494 for (i = 0; i < sav->sav_count; i++) 1495 l2cache[i] = vdev_config_generate(spa, 1496 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1497 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1498 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1499out: 1500 for (i = 0; i < sav->sav_count; i++) 1501 nvlist_free(l2cache[i]); 1502 if (sav->sav_count) 1503 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1504} 1505 1506static int 1507load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1508{ 1509 dmu_buf_t *db; 1510 char *packed = NULL; 1511 size_t nvsize = 0; 1512 int error; 1513 *value = NULL; 1514 1515 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 1516 nvsize = *(uint64_t *)db->db_data; 1517 dmu_buf_rele(db, FTAG); 1518 1519 packed = kmem_alloc(nvsize, KM_SLEEP); 1520 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1521 DMU_READ_PREFETCH); 1522 if (error == 0) 1523 error = nvlist_unpack(packed, nvsize, value, 0); 1524 kmem_free(packed, nvsize); 1525 1526 return (error); 1527} 1528 1529/* 1530 * Checks to see if the given vdev could not be opened, in which case we post a 1531 * sysevent to notify the autoreplace code that the device has been removed. 1532 */ 1533static void 1534spa_check_removed(vdev_t *vd) 1535{ 1536 for (int c = 0; c < vd->vdev_children; c++) 1537 spa_check_removed(vd->vdev_child[c]); 1538 1539 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1540 !vd->vdev_ishole) { 1541 zfs_post_autoreplace(vd->vdev_spa, vd); 1542 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1543 } 1544} 1545 1546/* 1547 * Validate the current config against the MOS config 1548 */ 1549static boolean_t 1550spa_config_valid(spa_t *spa, nvlist_t *config) 1551{ 1552 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1553 nvlist_t *nv; 1554 1555 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1556 1557 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1558 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1559 1560 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1561 1562 /* 1563 * If we're doing a normal import, then build up any additional 1564 * diagnostic information about missing devices in this config. 1565 * We'll pass this up to the user for further processing. 1566 */ 1567 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1568 nvlist_t **child, *nv; 1569 uint64_t idx = 0; 1570 1571 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1572 KM_SLEEP); 1573 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1574 1575 for (int c = 0; c < rvd->vdev_children; c++) { 1576 vdev_t *tvd = rvd->vdev_child[c]; 1577 vdev_t *mtvd = mrvd->vdev_child[c]; 1578 1579 if (tvd->vdev_ops == &vdev_missing_ops && 1580 mtvd->vdev_ops != &vdev_missing_ops && 1581 mtvd->vdev_islog) 1582 child[idx++] = vdev_config_generate(spa, mtvd, 1583 B_FALSE, 0); 1584 } 1585 1586 if (idx) { 1587 VERIFY(nvlist_add_nvlist_array(nv, 1588 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1589 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1590 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1591 1592 for (int i = 0; i < idx; i++) 1593 nvlist_free(child[i]); 1594 } 1595 nvlist_free(nv); 1596 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1597 } 1598 1599 /* 1600 * Compare the root vdev tree with the information we have 1601 * from the MOS config (mrvd). Check each top-level vdev 1602 * with the corresponding MOS config top-level (mtvd). 1603 */ 1604 for (int c = 0; c < rvd->vdev_children; c++) { 1605 vdev_t *tvd = rvd->vdev_child[c]; 1606 vdev_t *mtvd = mrvd->vdev_child[c]; 1607 1608 /* 1609 * Resolve any "missing" vdevs in the current configuration. 1610 * If we find that the MOS config has more accurate information 1611 * about the top-level vdev then use that vdev instead. 1612 */ 1613 if (tvd->vdev_ops == &vdev_missing_ops && 1614 mtvd->vdev_ops != &vdev_missing_ops) { 1615 1616 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1617 continue; 1618 1619 /* 1620 * Device specific actions. 1621 */ 1622 if (mtvd->vdev_islog) { 1623 spa_set_log_state(spa, SPA_LOG_CLEAR); 1624 } else { 1625 /* 1626 * XXX - once we have 'readonly' pool 1627 * support we should be able to handle 1628 * missing data devices by transitioning 1629 * the pool to readonly. 1630 */ 1631 continue; 1632 } 1633 1634 /* 1635 * Swap the missing vdev with the data we were 1636 * able to obtain from the MOS config. 1637 */ 1638 vdev_remove_child(rvd, tvd); 1639 vdev_remove_child(mrvd, mtvd); 1640 1641 vdev_add_child(rvd, mtvd); 1642 vdev_add_child(mrvd, tvd); 1643 1644 spa_config_exit(spa, SCL_ALL, FTAG); 1645 vdev_load(mtvd); 1646 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1647 1648 vdev_reopen(rvd); 1649 } else if (mtvd->vdev_islog) { 1650 /* 1651 * Load the slog device's state from the MOS config 1652 * since it's possible that the label does not 1653 * contain the most up-to-date information. 1654 */ 1655 vdev_load_log_state(tvd, mtvd); 1656 vdev_reopen(tvd); 1657 } 1658 } 1659 vdev_free(mrvd); 1660 spa_config_exit(spa, SCL_ALL, FTAG); 1661 1662 /* 1663 * Ensure we were able to validate the config. 1664 */ 1665 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1666} 1667 1668/* 1669 * Check for missing log devices 1670 */ 1671static boolean_t 1672spa_check_logs(spa_t *spa) 1673{ 1674 boolean_t rv = B_FALSE; 1675 1676 switch (spa->spa_log_state) { 1677 case SPA_LOG_MISSING: 1678 /* need to recheck in case slog has been restored */ 1679 case SPA_LOG_UNKNOWN: 1680 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1681 NULL, DS_FIND_CHILDREN) != 0); 1682 if (rv) 1683 spa_set_log_state(spa, SPA_LOG_MISSING); 1684 break; 1685 } 1686 return (rv); 1687} 1688 1689static boolean_t 1690spa_passivate_log(spa_t *spa) 1691{ 1692 vdev_t *rvd = spa->spa_root_vdev; 1693 boolean_t slog_found = B_FALSE; 1694 1695 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1696 1697 if (!spa_has_slogs(spa)) 1698 return (B_FALSE); 1699 1700 for (int c = 0; c < rvd->vdev_children; c++) { 1701 vdev_t *tvd = rvd->vdev_child[c]; 1702 metaslab_group_t *mg = tvd->vdev_mg; 1703 1704 if (tvd->vdev_islog) { 1705 metaslab_group_passivate(mg); 1706 slog_found = B_TRUE; 1707 } 1708 } 1709 1710 return (slog_found); 1711} 1712 1713static void 1714spa_activate_log(spa_t *spa) 1715{ 1716 vdev_t *rvd = spa->spa_root_vdev; 1717 1718 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1719 1720 for (int c = 0; c < rvd->vdev_children; c++) { 1721 vdev_t *tvd = rvd->vdev_child[c]; 1722 metaslab_group_t *mg = tvd->vdev_mg; 1723 1724 if (tvd->vdev_islog) 1725 metaslab_group_activate(mg); 1726 } 1727} 1728 1729int 1730spa_offline_log(spa_t *spa) 1731{ 1732 int error; 1733 1734 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1735 NULL, DS_FIND_CHILDREN); 1736 if (error == 0) { 1737 /* 1738 * We successfully offlined the log device, sync out the 1739 * current txg so that the "stubby" block can be removed 1740 * by zil_sync(). 1741 */ 1742 txg_wait_synced(spa->spa_dsl_pool, 0); 1743 } 1744 return (error); 1745} 1746 1747static void 1748spa_aux_check_removed(spa_aux_vdev_t *sav) 1749{ 1750 int i; 1751 1752 for (i = 0; i < sav->sav_count; i++) 1753 spa_check_removed(sav->sav_vdevs[i]); 1754} 1755 1756void 1757spa_claim_notify(zio_t *zio) 1758{ 1759 spa_t *spa = zio->io_spa; 1760 1761 if (zio->io_error) 1762 return; 1763 1764 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1765 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1766 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1767 mutex_exit(&spa->spa_props_lock); 1768} 1769 1770typedef struct spa_load_error { 1771 uint64_t sle_meta_count; 1772 uint64_t sle_data_count; 1773} spa_load_error_t; 1774 1775static void 1776spa_load_verify_done(zio_t *zio) 1777{ 1778 blkptr_t *bp = zio->io_bp; 1779 spa_load_error_t *sle = zio->io_private; 1780 dmu_object_type_t type = BP_GET_TYPE(bp); 1781 int error = zio->io_error; 1782 1783 if (error) { 1784 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1785 type != DMU_OT_INTENT_LOG) 1786 atomic_add_64(&sle->sle_meta_count, 1); 1787 else 1788 atomic_add_64(&sle->sle_data_count, 1); 1789 } 1790 zio_data_buf_free(zio->io_data, zio->io_size); 1791} 1792 1793/*ARGSUSED*/ 1794static int 1795spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1796 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1797{ 1798 if (bp != NULL) { 1799 zio_t *rio = arg; 1800 size_t size = BP_GET_PSIZE(bp); 1801 void *data = zio_data_buf_alloc(size); 1802 1803 zio_nowait(zio_read(rio, spa, bp, data, size, 1804 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1805 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1806 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1807 } 1808 return (0); 1809} 1810 1811static int 1812spa_load_verify(spa_t *spa) 1813{ 1814 zio_t *rio; 1815 spa_load_error_t sle = { 0 }; 1816 zpool_rewind_policy_t policy; 1817 boolean_t verify_ok = B_FALSE; 1818 int error; 1819 1820 zpool_get_rewind_policy(spa->spa_config, &policy); 1821 1822 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1823 return (0); 1824 1825 rio = zio_root(spa, NULL, &sle, 1826 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1827 1828 error = traverse_pool(spa, spa->spa_verify_min_txg, 1829 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1830 1831 (void) zio_wait(rio); 1832 1833 spa->spa_load_meta_errors = sle.sle_meta_count; 1834 spa->spa_load_data_errors = sle.sle_data_count; 1835 1836 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1837 sle.sle_data_count <= policy.zrp_maxdata) { 1838 int64_t loss = 0; 1839 1840 verify_ok = B_TRUE; 1841 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1842 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1843 1844 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1845 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1846 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1847 VERIFY(nvlist_add_int64(spa->spa_load_info, 1848 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1849 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1850 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1851 } else { 1852 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1853 } 1854 1855 if (error) { 1856 if (error != ENXIO && error != EIO) 1857 error = SET_ERROR(EIO); 1858 return (error); 1859 } 1860 1861 return (verify_ok ? 0 : EIO); 1862} 1863 1864/* 1865 * Find a value in the pool props object. 1866 */ 1867static void 1868spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1869{ 1870 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1871 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1872} 1873 1874/* 1875 * Find a value in the pool directory object. 1876 */ 1877static int 1878spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1879{ 1880 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1881 name, sizeof (uint64_t), 1, val)); 1882} 1883 1884static int 1885spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1886{ 1887 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1888 return (err); 1889} 1890 1891/* 1892 * Fix up config after a partly-completed split. This is done with the 1893 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1894 * pool have that entry in their config, but only the splitting one contains 1895 * a list of all the guids of the vdevs that are being split off. 1896 * 1897 * This function determines what to do with that list: either rejoin 1898 * all the disks to the pool, or complete the splitting process. To attempt 1899 * the rejoin, each disk that is offlined is marked online again, and 1900 * we do a reopen() call. If the vdev label for every disk that was 1901 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 1902 * then we call vdev_split() on each disk, and complete the split. 1903 * 1904 * Otherwise we leave the config alone, with all the vdevs in place in 1905 * the original pool. 1906 */ 1907static void 1908spa_try_repair(spa_t *spa, nvlist_t *config) 1909{ 1910 uint_t extracted; 1911 uint64_t *glist; 1912 uint_t i, gcount; 1913 nvlist_t *nvl; 1914 vdev_t **vd; 1915 boolean_t attempt_reopen; 1916 1917 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 1918 return; 1919 1920 /* check that the config is complete */ 1921 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 1922 &glist, &gcount) != 0) 1923 return; 1924 1925 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 1926 1927 /* attempt to online all the vdevs & validate */ 1928 attempt_reopen = B_TRUE; 1929 for (i = 0; i < gcount; i++) { 1930 if (glist[i] == 0) /* vdev is hole */ 1931 continue; 1932 1933 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 1934 if (vd[i] == NULL) { 1935 /* 1936 * Don't bother attempting to reopen the disks; 1937 * just do the split. 1938 */ 1939 attempt_reopen = B_FALSE; 1940 } else { 1941 /* attempt to re-online it */ 1942 vd[i]->vdev_offline = B_FALSE; 1943 } 1944 } 1945 1946 if (attempt_reopen) { 1947 vdev_reopen(spa->spa_root_vdev); 1948 1949 /* check each device to see what state it's in */ 1950 for (extracted = 0, i = 0; i < gcount; i++) { 1951 if (vd[i] != NULL && 1952 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 1953 break; 1954 ++extracted; 1955 } 1956 } 1957 1958 /* 1959 * If every disk has been moved to the new pool, or if we never 1960 * even attempted to look at them, then we split them off for 1961 * good. 1962 */ 1963 if (!attempt_reopen || gcount == extracted) { 1964 for (i = 0; i < gcount; i++) 1965 if (vd[i] != NULL) 1966 vdev_split(vd[i]); 1967 vdev_reopen(spa->spa_root_vdev); 1968 } 1969 1970 kmem_free(vd, gcount * sizeof (vdev_t *)); 1971} 1972 1973static int 1974spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 1975 boolean_t mosconfig) 1976{ 1977 nvlist_t *config = spa->spa_config; 1978 char *ereport = FM_EREPORT_ZFS_POOL; 1979 char *comment; 1980 int error; 1981 uint64_t pool_guid; 1982 nvlist_t *nvl; 1983 1984 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 1985 return (SET_ERROR(EINVAL)); 1986 1987 ASSERT(spa->spa_comment == NULL); 1988 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 1989 spa->spa_comment = spa_strdup(comment); 1990 1991 /* 1992 * Versioning wasn't explicitly added to the label until later, so if 1993 * it's not present treat it as the initial version. 1994 */ 1995 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 1996 &spa->spa_ubsync.ub_version) != 0) 1997 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 1998 1999 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2000 &spa->spa_config_txg); 2001 2002 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2003 spa_guid_exists(pool_guid, 0)) { 2004 error = SET_ERROR(EEXIST); 2005 } else { 2006 spa->spa_config_guid = pool_guid; 2007 2008 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2009 &nvl) == 0) { 2010 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2011 KM_SLEEP) == 0); 2012 } 2013 2014 nvlist_free(spa->spa_load_info); 2015 spa->spa_load_info = fnvlist_alloc(); 2016 2017 gethrestime(&spa->spa_loaded_ts); 2018 error = spa_load_impl(spa, pool_guid, config, state, type, 2019 mosconfig, &ereport); 2020 } 2021 2022 spa->spa_minref = refcount_count(&spa->spa_refcount); 2023 if (error) { 2024 if (error != EEXIST) { 2025 spa->spa_loaded_ts.tv_sec = 0; 2026 spa->spa_loaded_ts.tv_nsec = 0; 2027 } 2028 if (error != EBADF) { 2029 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2030 } 2031 } 2032 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2033 spa->spa_ena = 0; 2034 2035 return (error); 2036} 2037 2038/* 2039 * Load an existing storage pool, using the pool's builtin spa_config as a 2040 * source of configuration information. 2041 */ 2042static int 2043spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2044 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2045 char **ereport) 2046{ 2047 int error = 0; 2048 nvlist_t *nvroot = NULL; 2049 nvlist_t *label; 2050 vdev_t *rvd; 2051 uberblock_t *ub = &spa->spa_uberblock; 2052 uint64_t children, config_cache_txg = spa->spa_config_txg; 2053 int orig_mode = spa->spa_mode; 2054 int parse; 2055 uint64_t obj; 2056 boolean_t missing_feat_write = B_FALSE; 2057 2058 /* 2059 * If this is an untrusted config, access the pool in read-only mode. 2060 * This prevents things like resilvering recently removed devices. 2061 */ 2062 if (!mosconfig) 2063 spa->spa_mode = FREAD; 2064 2065 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2066 2067 spa->spa_load_state = state; 2068 2069 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2070 return (SET_ERROR(EINVAL)); 2071 2072 parse = (type == SPA_IMPORT_EXISTING ? 2073 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2074 2075 /* 2076 * Create "The Godfather" zio to hold all async IOs 2077 */ 2078 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 2079 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 2080 2081 /* 2082 * Parse the configuration into a vdev tree. We explicitly set the 2083 * value that will be returned by spa_version() since parsing the 2084 * configuration requires knowing the version number. 2085 */ 2086 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2087 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2088 spa_config_exit(spa, SCL_ALL, FTAG); 2089 2090 if (error != 0) 2091 return (error); 2092 2093 ASSERT(spa->spa_root_vdev == rvd); 2094 2095 if (type != SPA_IMPORT_ASSEMBLE) { 2096 ASSERT(spa_guid(spa) == pool_guid); 2097 } 2098 2099 /* 2100 * Try to open all vdevs, loading each label in the process. 2101 */ 2102 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2103 error = vdev_open(rvd); 2104 spa_config_exit(spa, SCL_ALL, FTAG); 2105 if (error != 0) 2106 return (error); 2107 2108 /* 2109 * We need to validate the vdev labels against the configuration that 2110 * we have in hand, which is dependent on the setting of mosconfig. If 2111 * mosconfig is true then we're validating the vdev labels based on 2112 * that config. Otherwise, we're validating against the cached config 2113 * (zpool.cache) that was read when we loaded the zfs module, and then 2114 * later we will recursively call spa_load() and validate against 2115 * the vdev config. 2116 * 2117 * If we're assembling a new pool that's been split off from an 2118 * existing pool, the labels haven't yet been updated so we skip 2119 * validation for now. 2120 */ 2121 if (type != SPA_IMPORT_ASSEMBLE) { 2122 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2123 error = vdev_validate(rvd, mosconfig); 2124 spa_config_exit(spa, SCL_ALL, FTAG); 2125 2126 if (error != 0) 2127 return (error); 2128 2129 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2130 return (SET_ERROR(ENXIO)); 2131 } 2132 2133 /* 2134 * Find the best uberblock. 2135 */ 2136 vdev_uberblock_load(rvd, ub, &label); 2137 2138 /* 2139 * If we weren't able to find a single valid uberblock, return failure. 2140 */ 2141 if (ub->ub_txg == 0) { 2142 nvlist_free(label); 2143 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2144 } 2145 2146 /* 2147 * If the pool has an unsupported version we can't open it. 2148 */ 2149 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2150 nvlist_free(label); 2151 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2152 } 2153 2154 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2155 nvlist_t *features; 2156 2157 /* 2158 * If we weren't able to find what's necessary for reading the 2159 * MOS in the label, return failure. 2160 */ 2161 if (label == NULL || nvlist_lookup_nvlist(label, 2162 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2163 nvlist_free(label); 2164 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2165 ENXIO)); 2166 } 2167 2168 /* 2169 * Update our in-core representation with the definitive values 2170 * from the label. 2171 */ 2172 nvlist_free(spa->spa_label_features); 2173 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2174 } 2175 2176 nvlist_free(label); 2177 2178 /* 2179 * Look through entries in the label nvlist's features_for_read. If 2180 * there is a feature listed there which we don't understand then we 2181 * cannot open a pool. 2182 */ 2183 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2184 nvlist_t *unsup_feat; 2185 2186 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2187 0); 2188 2189 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2190 NULL); nvp != NULL; 2191 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2192 if (!zfeature_is_supported(nvpair_name(nvp))) { 2193 VERIFY(nvlist_add_string(unsup_feat, 2194 nvpair_name(nvp), "") == 0); 2195 } 2196 } 2197 2198 if (!nvlist_empty(unsup_feat)) { 2199 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2200 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2201 nvlist_free(unsup_feat); 2202 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2203 ENOTSUP)); 2204 } 2205 2206 nvlist_free(unsup_feat); 2207 } 2208 2209 /* 2210 * If the vdev guid sum doesn't match the uberblock, we have an 2211 * incomplete configuration. We first check to see if the pool 2212 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2213 * If it is, defer the vdev_guid_sum check till later so we 2214 * can handle missing vdevs. 2215 */ 2216 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2217 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2218 rvd->vdev_guid_sum != ub->ub_guid_sum) 2219 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2220 2221 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2222 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2223 spa_try_repair(spa, config); 2224 spa_config_exit(spa, SCL_ALL, FTAG); 2225 nvlist_free(spa->spa_config_splitting); 2226 spa->spa_config_splitting = NULL; 2227 } 2228 2229 /* 2230 * Initialize internal SPA structures. 2231 */ 2232 spa->spa_state = POOL_STATE_ACTIVE; 2233 spa->spa_ubsync = spa->spa_uberblock; 2234 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2235 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2236 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2237 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2238 spa->spa_claim_max_txg = spa->spa_first_txg; 2239 spa->spa_prev_software_version = ub->ub_software_version; 2240 2241 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2242 if (error) 2243 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2244 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2245 2246 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2247 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2248 2249 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2250 boolean_t missing_feat_read = B_FALSE; 2251 nvlist_t *unsup_feat, *enabled_feat; 2252 2253 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2254 &spa->spa_feat_for_read_obj) != 0) { 2255 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2256 } 2257 2258 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2259 &spa->spa_feat_for_write_obj) != 0) { 2260 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2261 } 2262 2263 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2264 &spa->spa_feat_desc_obj) != 0) { 2265 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2266 } 2267 2268 enabled_feat = fnvlist_alloc(); 2269 unsup_feat = fnvlist_alloc(); 2270 2271 if (!feature_is_supported(spa->spa_meta_objset, 2272 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, 2273 unsup_feat, enabled_feat)) 2274 missing_feat_read = B_TRUE; 2275 2276 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2277 if (!feature_is_supported(spa->spa_meta_objset, 2278 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, 2279 unsup_feat, enabled_feat)) { 2280 missing_feat_write = B_TRUE; 2281 } 2282 } 2283 2284 fnvlist_add_nvlist(spa->spa_load_info, 2285 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2286 2287 if (!nvlist_empty(unsup_feat)) { 2288 fnvlist_add_nvlist(spa->spa_load_info, 2289 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2290 } 2291 2292 fnvlist_free(enabled_feat); 2293 fnvlist_free(unsup_feat); 2294 2295 if (!missing_feat_read) { 2296 fnvlist_add_boolean(spa->spa_load_info, 2297 ZPOOL_CONFIG_CAN_RDONLY); 2298 } 2299 2300 /* 2301 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2302 * twofold: to determine whether the pool is available for 2303 * import in read-write mode and (if it is not) whether the 2304 * pool is available for import in read-only mode. If the pool 2305 * is available for import in read-write mode, it is displayed 2306 * as available in userland; if it is not available for import 2307 * in read-only mode, it is displayed as unavailable in 2308 * userland. If the pool is available for import in read-only 2309 * mode but not read-write mode, it is displayed as unavailable 2310 * in userland with a special note that the pool is actually 2311 * available for open in read-only mode. 2312 * 2313 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2314 * missing a feature for write, we must first determine whether 2315 * the pool can be opened read-only before returning to 2316 * userland in order to know whether to display the 2317 * abovementioned note. 2318 */ 2319 if (missing_feat_read || (missing_feat_write && 2320 spa_writeable(spa))) { 2321 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2322 ENOTSUP)); 2323 } 2324 } 2325 2326 spa->spa_is_initializing = B_TRUE; 2327 error = dsl_pool_open(spa->spa_dsl_pool); 2328 spa->spa_is_initializing = B_FALSE; 2329 if (error != 0) 2330 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2331 2332 if (!mosconfig) { 2333 uint64_t hostid; 2334 nvlist_t *policy = NULL, *nvconfig; 2335 2336 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2337 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2338 2339 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2340 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2341 char *hostname; 2342 unsigned long myhostid = 0; 2343 2344 VERIFY(nvlist_lookup_string(nvconfig, 2345 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2346 2347#ifdef _KERNEL 2348 myhostid = zone_get_hostid(NULL); 2349#else /* _KERNEL */ 2350 /* 2351 * We're emulating the system's hostid in userland, so 2352 * we can't use zone_get_hostid(). 2353 */ 2354 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2355#endif /* _KERNEL */ 2356 if (check_hostid && hostid != 0 && myhostid != 0 && 2357 hostid != myhostid) { 2358 nvlist_free(nvconfig); 2359 cmn_err(CE_WARN, "pool '%s' could not be " 2360 "loaded as it was last accessed by " 2361 "another system (host: %s hostid: 0x%lx). " 2362 "See: http://illumos.org/msg/ZFS-8000-EY", 2363 spa_name(spa), hostname, 2364 (unsigned long)hostid); 2365 return (SET_ERROR(EBADF)); 2366 } 2367 } 2368 if (nvlist_lookup_nvlist(spa->spa_config, 2369 ZPOOL_REWIND_POLICY, &policy) == 0) 2370 VERIFY(nvlist_add_nvlist(nvconfig, 2371 ZPOOL_REWIND_POLICY, policy) == 0); 2372 2373 spa_config_set(spa, nvconfig); 2374 spa_unload(spa); 2375 spa_deactivate(spa); 2376 spa_activate(spa, orig_mode); 2377 2378 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2379 } 2380 2381 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2383 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2384 if (error != 0) 2385 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2386 2387 /* 2388 * Load the bit that tells us to use the new accounting function 2389 * (raid-z deflation). If we have an older pool, this will not 2390 * be present. 2391 */ 2392 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2393 if (error != 0 && error != ENOENT) 2394 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2395 2396 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2397 &spa->spa_creation_version); 2398 if (error != 0 && error != ENOENT) 2399 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2400 2401 /* 2402 * Load the persistent error log. If we have an older pool, this will 2403 * not be present. 2404 */ 2405 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2406 if (error != 0 && error != ENOENT) 2407 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2408 2409 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2410 &spa->spa_errlog_scrub); 2411 if (error != 0 && error != ENOENT) 2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2413 2414 /* 2415 * Load the history object. If we have an older pool, this 2416 * will not be present. 2417 */ 2418 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2419 if (error != 0 && error != ENOENT) 2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2421 2422 /* 2423 * If we're assembling the pool from the split-off vdevs of 2424 * an existing pool, we don't want to attach the spares & cache 2425 * devices. 2426 */ 2427 2428 /* 2429 * Load any hot spares for this pool. 2430 */ 2431 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2432 if (error != 0 && error != ENOENT) 2433 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2434 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2435 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2436 if (load_nvlist(spa, spa->spa_spares.sav_object, 2437 &spa->spa_spares.sav_config) != 0) 2438 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2439 2440 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2441 spa_load_spares(spa); 2442 spa_config_exit(spa, SCL_ALL, FTAG); 2443 } else if (error == 0) { 2444 spa->spa_spares.sav_sync = B_TRUE; 2445 } 2446 2447 /* 2448 * Load any level 2 ARC devices for this pool. 2449 */ 2450 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2451 &spa->spa_l2cache.sav_object); 2452 if (error != 0 && error != ENOENT) 2453 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2454 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2455 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2456 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2457 &spa->spa_l2cache.sav_config) != 0) 2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2459 2460 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2461 spa_load_l2cache(spa); 2462 spa_config_exit(spa, SCL_ALL, FTAG); 2463 } else if (error == 0) { 2464 spa->spa_l2cache.sav_sync = B_TRUE; 2465 } 2466 2467 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2468 2469 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2470 if (error && error != ENOENT) 2471 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2472 2473 if (error == 0) { 2474 uint64_t autoreplace; 2475 2476 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2477 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2478 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2479 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2480 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2481 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2482 &spa->spa_dedup_ditto); 2483 2484 spa->spa_autoreplace = (autoreplace != 0); 2485 } 2486 2487 /* 2488 * If the 'autoreplace' property is set, then post a resource notifying 2489 * the ZFS DE that it should not issue any faults for unopenable 2490 * devices. We also iterate over the vdevs, and post a sysevent for any 2491 * unopenable vdevs so that the normal autoreplace handler can take 2492 * over. 2493 */ 2494 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2495 spa_check_removed(spa->spa_root_vdev); 2496 /* 2497 * For the import case, this is done in spa_import(), because 2498 * at this point we're using the spare definitions from 2499 * the MOS config, not necessarily from the userland config. 2500 */ 2501 if (state != SPA_LOAD_IMPORT) { 2502 spa_aux_check_removed(&spa->spa_spares); 2503 spa_aux_check_removed(&spa->spa_l2cache); 2504 } 2505 } 2506 2507 /* 2508 * Load the vdev state for all toplevel vdevs. 2509 */ 2510 vdev_load(rvd); 2511 2512 /* 2513 * Propagate the leaf DTLs we just loaded all the way up the tree. 2514 */ 2515 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2516 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2517 spa_config_exit(spa, SCL_ALL, FTAG); 2518 2519 /* 2520 * Load the DDTs (dedup tables). 2521 */ 2522 error = ddt_load(spa); 2523 if (error != 0) 2524 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2525 2526 spa_update_dspace(spa); 2527 2528 /* 2529 * Validate the config, using the MOS config to fill in any 2530 * information which might be missing. If we fail to validate 2531 * the config then declare the pool unfit for use. If we're 2532 * assembling a pool from a split, the log is not transferred 2533 * over. 2534 */ 2535 if (type != SPA_IMPORT_ASSEMBLE) { 2536 nvlist_t *nvconfig; 2537 2538 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2539 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2540 2541 if (!spa_config_valid(spa, nvconfig)) { 2542 nvlist_free(nvconfig); 2543 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2544 ENXIO)); 2545 } 2546 nvlist_free(nvconfig); 2547 2548 /* 2549 * Now that we've validated the config, check the state of the 2550 * root vdev. If it can't be opened, it indicates one or 2551 * more toplevel vdevs are faulted. 2552 */ 2553 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2554 return (SET_ERROR(ENXIO)); 2555 2556 if (spa_check_logs(spa)) { 2557 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2558 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2559 } 2560 } 2561 2562 if (missing_feat_write) { 2563 ASSERT(state == SPA_LOAD_TRYIMPORT); 2564 2565 /* 2566 * At this point, we know that we can open the pool in 2567 * read-only mode but not read-write mode. We now have enough 2568 * information and can return to userland. 2569 */ 2570 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2571 } 2572 2573 /* 2574 * We've successfully opened the pool, verify that we're ready 2575 * to start pushing transactions. 2576 */ 2577 if (state != SPA_LOAD_TRYIMPORT) { 2578 if (error = spa_load_verify(spa)) 2579 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2580 error)); 2581 } 2582 2583 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2584 spa->spa_load_max_txg == UINT64_MAX)) { 2585 dmu_tx_t *tx; 2586 int need_update = B_FALSE; 2587 2588 ASSERT(state != SPA_LOAD_TRYIMPORT); 2589 2590 /* 2591 * Claim log blocks that haven't been committed yet. 2592 * This must all happen in a single txg. 2593 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2594 * invoked from zil_claim_log_block()'s i/o done callback. 2595 * Price of rollback is that we abandon the log. 2596 */ 2597 spa->spa_claiming = B_TRUE; 2598 2599 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2600 spa_first_txg(spa)); 2601 (void) dmu_objset_find(spa_name(spa), 2602 zil_claim, tx, DS_FIND_CHILDREN); 2603 dmu_tx_commit(tx); 2604 2605 spa->spa_claiming = B_FALSE; 2606 2607 spa_set_log_state(spa, SPA_LOG_GOOD); 2608 spa->spa_sync_on = B_TRUE; 2609 txg_sync_start(spa->spa_dsl_pool); 2610 2611 /* 2612 * Wait for all claims to sync. We sync up to the highest 2613 * claimed log block birth time so that claimed log blocks 2614 * don't appear to be from the future. spa_claim_max_txg 2615 * will have been set for us by either zil_check_log_chain() 2616 * (invoked from spa_check_logs()) or zil_claim() above. 2617 */ 2618 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2619 2620 /* 2621 * If the config cache is stale, or we have uninitialized 2622 * metaslabs (see spa_vdev_add()), then update the config. 2623 * 2624 * If this is a verbatim import, trust the current 2625 * in-core spa_config and update the disk labels. 2626 */ 2627 if (config_cache_txg != spa->spa_config_txg || 2628 state == SPA_LOAD_IMPORT || 2629 state == SPA_LOAD_RECOVER || 2630 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2631 need_update = B_TRUE; 2632 2633 for (int c = 0; c < rvd->vdev_children; c++) 2634 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2635 need_update = B_TRUE; 2636 2637 /* 2638 * Update the config cache asychronously in case we're the 2639 * root pool, in which case the config cache isn't writable yet. 2640 */ 2641 if (need_update) 2642 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2643 2644 /* 2645 * Check all DTLs to see if anything needs resilvering. 2646 */ 2647 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2648 vdev_resilver_needed(rvd, NULL, NULL)) 2649 spa_async_request(spa, SPA_ASYNC_RESILVER); 2650 2651 /* 2652 * Log the fact that we booted up (so that we can detect if 2653 * we rebooted in the middle of an operation). 2654 */ 2655 spa_history_log_version(spa, "open"); 2656 2657 /* 2658 * Delete any inconsistent datasets. 2659 */ 2660 (void) dmu_objset_find(spa_name(spa), 2661 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2662 2663 /* 2664 * Clean up any stale temporary dataset userrefs. 2665 */ 2666 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2667 } 2668 2669 return (0); 2670} 2671 2672static int 2673spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2674{ 2675 int mode = spa->spa_mode; 2676 2677 spa_unload(spa); 2678 spa_deactivate(spa); 2679 2680 spa->spa_load_max_txg--; 2681 2682 spa_activate(spa, mode); 2683 spa_async_suspend(spa); 2684 2685 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2686} 2687 2688/* 2689 * If spa_load() fails this function will try loading prior txg's. If 2690 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2691 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2692 * function will not rewind the pool and will return the same error as 2693 * spa_load(). 2694 */ 2695static int 2696spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2697 uint64_t max_request, int rewind_flags) 2698{ 2699 nvlist_t *loadinfo = NULL; 2700 nvlist_t *config = NULL; 2701 int load_error, rewind_error; 2702 uint64_t safe_rewind_txg; 2703 uint64_t min_txg; 2704 2705 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2706 spa->spa_load_max_txg = spa->spa_load_txg; 2707 spa_set_log_state(spa, SPA_LOG_CLEAR); 2708 } else { 2709 spa->spa_load_max_txg = max_request; 2710 } 2711 2712 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2713 mosconfig); 2714 if (load_error == 0) 2715 return (0); 2716 2717 if (spa->spa_root_vdev != NULL) 2718 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2719 2720 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2721 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2722 2723 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2724 nvlist_free(config); 2725 return (load_error); 2726 } 2727 2728 if (state == SPA_LOAD_RECOVER) { 2729 /* Price of rolling back is discarding txgs, including log */ 2730 spa_set_log_state(spa, SPA_LOG_CLEAR); 2731 } else { 2732 /* 2733 * If we aren't rolling back save the load info from our first 2734 * import attempt so that we can restore it after attempting 2735 * to rewind. 2736 */ 2737 loadinfo = spa->spa_load_info; 2738 spa->spa_load_info = fnvlist_alloc(); 2739 } 2740 2741 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2742 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2743 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2744 TXG_INITIAL : safe_rewind_txg; 2745 2746 /* 2747 * Continue as long as we're finding errors, we're still within 2748 * the acceptable rewind range, and we're still finding uberblocks 2749 */ 2750 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2751 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2752 if (spa->spa_load_max_txg < safe_rewind_txg) 2753 spa->spa_extreme_rewind = B_TRUE; 2754 rewind_error = spa_load_retry(spa, state, mosconfig); 2755 } 2756 2757 spa->spa_extreme_rewind = B_FALSE; 2758 spa->spa_load_max_txg = UINT64_MAX; 2759 2760 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2761 spa_config_set(spa, config); 2762 2763 if (state == SPA_LOAD_RECOVER) { 2764 ASSERT3P(loadinfo, ==, NULL); 2765 return (rewind_error); 2766 } else { 2767 /* Store the rewind info as part of the initial load info */ 2768 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2769 spa->spa_load_info); 2770 2771 /* Restore the initial load info */ 2772 fnvlist_free(spa->spa_load_info); 2773 spa->spa_load_info = loadinfo; 2774 2775 return (load_error); 2776 } 2777} 2778 2779/* 2780 * Pool Open/Import 2781 * 2782 * The import case is identical to an open except that the configuration is sent 2783 * down from userland, instead of grabbed from the configuration cache. For the 2784 * case of an open, the pool configuration will exist in the 2785 * POOL_STATE_UNINITIALIZED state. 2786 * 2787 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2788 * the same time open the pool, without having to keep around the spa_t in some 2789 * ambiguous state. 2790 */ 2791static int 2792spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2793 nvlist_t **config) 2794{ 2795 spa_t *spa; 2796 spa_load_state_t state = SPA_LOAD_OPEN; 2797 int error; 2798 int locked = B_FALSE; 2799 int firstopen = B_FALSE; 2800 2801 *spapp = NULL; 2802 2803 /* 2804 * As disgusting as this is, we need to support recursive calls to this 2805 * function because dsl_dir_open() is called during spa_load(), and ends 2806 * up calling spa_open() again. The real fix is to figure out how to 2807 * avoid dsl_dir_open() calling this in the first place. 2808 */ 2809 if (mutex_owner(&spa_namespace_lock) != curthread) { 2810 mutex_enter(&spa_namespace_lock); 2811 locked = B_TRUE; 2812 } 2813 2814 if ((spa = spa_lookup(pool)) == NULL) { 2815 if (locked) 2816 mutex_exit(&spa_namespace_lock); 2817 return (SET_ERROR(ENOENT)); 2818 } 2819 2820 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2821 zpool_rewind_policy_t policy; 2822 2823 firstopen = B_TRUE; 2824 2825 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2826 &policy); 2827 if (policy.zrp_request & ZPOOL_DO_REWIND) 2828 state = SPA_LOAD_RECOVER; 2829 2830 spa_activate(spa, spa_mode_global); 2831 2832 if (state != SPA_LOAD_RECOVER) 2833 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2834 2835 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2836 policy.zrp_request); 2837 2838 if (error == EBADF) { 2839 /* 2840 * If vdev_validate() returns failure (indicated by 2841 * EBADF), it indicates that one of the vdevs indicates 2842 * that the pool has been exported or destroyed. If 2843 * this is the case, the config cache is out of sync and 2844 * we should remove the pool from the namespace. 2845 */ 2846 spa_unload(spa); 2847 spa_deactivate(spa); 2848 spa_config_sync(spa, B_TRUE, B_TRUE); 2849 spa_remove(spa); 2850 if (locked) 2851 mutex_exit(&spa_namespace_lock); 2852 return (SET_ERROR(ENOENT)); 2853 } 2854 2855 if (error) { 2856 /* 2857 * We can't open the pool, but we still have useful 2858 * information: the state of each vdev after the 2859 * attempted vdev_open(). Return this to the user. 2860 */ 2861 if (config != NULL && spa->spa_config) { 2862 VERIFY(nvlist_dup(spa->spa_config, config, 2863 KM_SLEEP) == 0); 2864 VERIFY(nvlist_add_nvlist(*config, 2865 ZPOOL_CONFIG_LOAD_INFO, 2866 spa->spa_load_info) == 0); 2867 } 2868 spa_unload(spa); 2869 spa_deactivate(spa); 2870 spa->spa_last_open_failed = error; 2871 if (locked) 2872 mutex_exit(&spa_namespace_lock); 2873 *spapp = NULL; 2874 return (error); 2875 } 2876 } 2877 2878 spa_open_ref(spa, tag); 2879 2880 if (config != NULL) 2881 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2882 2883 /* 2884 * If we've recovered the pool, pass back any information we 2885 * gathered while doing the load. 2886 */ 2887 if (state == SPA_LOAD_RECOVER) { 2888 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 2889 spa->spa_load_info) == 0); 2890 } 2891 2892 if (locked) { 2893 spa->spa_last_open_failed = 0; 2894 spa->spa_last_ubsync_txg = 0; 2895 spa->spa_load_txg = 0; 2896 mutex_exit(&spa_namespace_lock); 2897#ifdef __FreeBSD__ 2898#ifdef _KERNEL 2899 if (firstopen) 2900 zvol_create_minors(spa->spa_name); 2901#endif 2902#endif 2903 } 2904 2905 *spapp = spa; 2906 2907 return (0); 2908} 2909 2910int 2911spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2912 nvlist_t **config) 2913{ 2914 return (spa_open_common(name, spapp, tag, policy, config)); 2915} 2916 2917int 2918spa_open(const char *name, spa_t **spapp, void *tag) 2919{ 2920 return (spa_open_common(name, spapp, tag, NULL, NULL)); 2921} 2922 2923/* 2924 * Lookup the given spa_t, incrementing the inject count in the process, 2925 * preventing it from being exported or destroyed. 2926 */ 2927spa_t * 2928spa_inject_addref(char *name) 2929{ 2930 spa_t *spa; 2931 2932 mutex_enter(&spa_namespace_lock); 2933 if ((spa = spa_lookup(name)) == NULL) { 2934 mutex_exit(&spa_namespace_lock); 2935 return (NULL); 2936 } 2937 spa->spa_inject_ref++; 2938 mutex_exit(&spa_namespace_lock); 2939 2940 return (spa); 2941} 2942 2943void 2944spa_inject_delref(spa_t *spa) 2945{ 2946 mutex_enter(&spa_namespace_lock); 2947 spa->spa_inject_ref--; 2948 mutex_exit(&spa_namespace_lock); 2949} 2950 2951/* 2952 * Add spares device information to the nvlist. 2953 */ 2954static void 2955spa_add_spares(spa_t *spa, nvlist_t *config) 2956{ 2957 nvlist_t **spares; 2958 uint_t i, nspares; 2959 nvlist_t *nvroot; 2960 uint64_t guid; 2961 vdev_stat_t *vs; 2962 uint_t vsc; 2963 uint64_t pool; 2964 2965 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 2966 2967 if (spa->spa_spares.sav_count == 0) 2968 return; 2969 2970 VERIFY(nvlist_lookup_nvlist(config, 2971 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2972 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 2973 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2974 if (nspares != 0) { 2975 VERIFY(nvlist_add_nvlist_array(nvroot, 2976 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 2977 VERIFY(nvlist_lookup_nvlist_array(nvroot, 2978 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2979 2980 /* 2981 * Go through and find any spares which have since been 2982 * repurposed as an active spare. If this is the case, update 2983 * their status appropriately. 2984 */ 2985 for (i = 0; i < nspares; i++) { 2986 VERIFY(nvlist_lookup_uint64(spares[i], 2987 ZPOOL_CONFIG_GUID, &guid) == 0); 2988 if (spa_spare_exists(guid, &pool, NULL) && 2989 pool != 0ULL) { 2990 VERIFY(nvlist_lookup_uint64_array( 2991 spares[i], ZPOOL_CONFIG_VDEV_STATS, 2992 (uint64_t **)&vs, &vsc) == 0); 2993 vs->vs_state = VDEV_STATE_CANT_OPEN; 2994 vs->vs_aux = VDEV_AUX_SPARED; 2995 } 2996 } 2997 } 2998} 2999 3000/* 3001 * Add l2cache device information to the nvlist, including vdev stats. 3002 */ 3003static void 3004spa_add_l2cache(spa_t *spa, nvlist_t *config) 3005{ 3006 nvlist_t **l2cache; 3007 uint_t i, j, nl2cache; 3008 nvlist_t *nvroot; 3009 uint64_t guid; 3010 vdev_t *vd; 3011 vdev_stat_t *vs; 3012 uint_t vsc; 3013 3014 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3015 3016 if (spa->spa_l2cache.sav_count == 0) 3017 return; 3018 3019 VERIFY(nvlist_lookup_nvlist(config, 3020 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3021 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3022 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3023 if (nl2cache != 0) { 3024 VERIFY(nvlist_add_nvlist_array(nvroot, 3025 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3026 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3027 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3028 3029 /* 3030 * Update level 2 cache device stats. 3031 */ 3032 3033 for (i = 0; i < nl2cache; i++) { 3034 VERIFY(nvlist_lookup_uint64(l2cache[i], 3035 ZPOOL_CONFIG_GUID, &guid) == 0); 3036 3037 vd = NULL; 3038 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3039 if (guid == 3040 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3041 vd = spa->spa_l2cache.sav_vdevs[j]; 3042 break; 3043 } 3044 } 3045 ASSERT(vd != NULL); 3046 3047 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3048 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3049 == 0); 3050 vdev_get_stats(vd, vs); 3051 } 3052 } 3053} 3054 3055static void 3056spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3057{ 3058 nvlist_t *features; 3059 zap_cursor_t zc; 3060 zap_attribute_t za; 3061 3062 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3063 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3064 3065 /* We may be unable to read features if pool is suspended. */ 3066 if (spa_suspended(spa)) 3067 goto out; 3068 3069 if (spa->spa_feat_for_read_obj != 0) { 3070 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3071 spa->spa_feat_for_read_obj); 3072 zap_cursor_retrieve(&zc, &za) == 0; 3073 zap_cursor_advance(&zc)) { 3074 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3075 za.za_num_integers == 1); 3076 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3077 za.za_first_integer)); 3078 } 3079 zap_cursor_fini(&zc); 3080 } 3081 3082 if (spa->spa_feat_for_write_obj != 0) { 3083 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3084 spa->spa_feat_for_write_obj); 3085 zap_cursor_retrieve(&zc, &za) == 0; 3086 zap_cursor_advance(&zc)) { 3087 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3088 za.za_num_integers == 1); 3089 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3090 za.za_first_integer)); 3091 } 3092 zap_cursor_fini(&zc); 3093 } 3094 3095out: 3096 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3097 features) == 0); 3098 nvlist_free(features); 3099} 3100 3101int 3102spa_get_stats(const char *name, nvlist_t **config, 3103 char *altroot, size_t buflen) 3104{ 3105 int error; 3106 spa_t *spa; 3107 3108 *config = NULL; 3109 error = spa_open_common(name, &spa, FTAG, NULL, config); 3110 3111 if (spa != NULL) { 3112 /* 3113 * This still leaves a window of inconsistency where the spares 3114 * or l2cache devices could change and the config would be 3115 * self-inconsistent. 3116 */ 3117 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3118 3119 if (*config != NULL) { 3120 uint64_t loadtimes[2]; 3121 3122 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3123 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3124 VERIFY(nvlist_add_uint64_array(*config, 3125 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3126 3127 VERIFY(nvlist_add_uint64(*config, 3128 ZPOOL_CONFIG_ERRCOUNT, 3129 spa_get_errlog_size(spa)) == 0); 3130 3131 if (spa_suspended(spa)) 3132 VERIFY(nvlist_add_uint64(*config, 3133 ZPOOL_CONFIG_SUSPENDED, 3134 spa->spa_failmode) == 0); 3135 3136 spa_add_spares(spa, *config); 3137 spa_add_l2cache(spa, *config); 3138 spa_add_feature_stats(spa, *config); 3139 } 3140 } 3141 3142 /* 3143 * We want to get the alternate root even for faulted pools, so we cheat 3144 * and call spa_lookup() directly. 3145 */ 3146 if (altroot) { 3147 if (spa == NULL) { 3148 mutex_enter(&spa_namespace_lock); 3149 spa = spa_lookup(name); 3150 if (spa) 3151 spa_altroot(spa, altroot, buflen); 3152 else 3153 altroot[0] = '\0'; 3154 spa = NULL; 3155 mutex_exit(&spa_namespace_lock); 3156 } else { 3157 spa_altroot(spa, altroot, buflen); 3158 } 3159 } 3160 3161 if (spa != NULL) { 3162 spa_config_exit(spa, SCL_CONFIG, FTAG); 3163 spa_close(spa, FTAG); 3164 } 3165 3166 return (error); 3167} 3168 3169/* 3170 * Validate that the auxiliary device array is well formed. We must have an 3171 * array of nvlists, each which describes a valid leaf vdev. If this is an 3172 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3173 * specified, as long as they are well-formed. 3174 */ 3175static int 3176spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3177 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3178 vdev_labeltype_t label) 3179{ 3180 nvlist_t **dev; 3181 uint_t i, ndev; 3182 vdev_t *vd; 3183 int error; 3184 3185 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3186 3187 /* 3188 * It's acceptable to have no devs specified. 3189 */ 3190 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3191 return (0); 3192 3193 if (ndev == 0) 3194 return (SET_ERROR(EINVAL)); 3195 3196 /* 3197 * Make sure the pool is formatted with a version that supports this 3198 * device type. 3199 */ 3200 if (spa_version(spa) < version) 3201 return (SET_ERROR(ENOTSUP)); 3202 3203 /* 3204 * Set the pending device list so we correctly handle device in-use 3205 * checking. 3206 */ 3207 sav->sav_pending = dev; 3208 sav->sav_npending = ndev; 3209 3210 for (i = 0; i < ndev; i++) { 3211 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3212 mode)) != 0) 3213 goto out; 3214 3215 if (!vd->vdev_ops->vdev_op_leaf) { 3216 vdev_free(vd); 3217 error = SET_ERROR(EINVAL); 3218 goto out; 3219 } 3220 3221 /* 3222 * The L2ARC currently only supports disk devices in 3223 * kernel context. For user-level testing, we allow it. 3224 */ 3225#ifdef _KERNEL 3226 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3227 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3228 error = SET_ERROR(ENOTBLK); 3229 vdev_free(vd); 3230 goto out; 3231 } 3232#endif 3233 vd->vdev_top = vd; 3234 3235 if ((error = vdev_open(vd)) == 0 && 3236 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3237 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3238 vd->vdev_guid) == 0); 3239 } 3240 3241 vdev_free(vd); 3242 3243 if (error && 3244 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3245 goto out; 3246 else 3247 error = 0; 3248 } 3249 3250out: 3251 sav->sav_pending = NULL; 3252 sav->sav_npending = 0; 3253 return (error); 3254} 3255 3256static int 3257spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3258{ 3259 int error; 3260 3261 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3262 3263 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3264 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3265 VDEV_LABEL_SPARE)) != 0) { 3266 return (error); 3267 } 3268 3269 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3270 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3271 VDEV_LABEL_L2CACHE)); 3272} 3273 3274static void 3275spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3276 const char *config) 3277{ 3278 int i; 3279 3280 if (sav->sav_config != NULL) { 3281 nvlist_t **olddevs; 3282 uint_t oldndevs; 3283 nvlist_t **newdevs; 3284 3285 /* 3286 * Generate new dev list by concatentating with the 3287 * current dev list. 3288 */ 3289 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3290 &olddevs, &oldndevs) == 0); 3291 3292 newdevs = kmem_alloc(sizeof (void *) * 3293 (ndevs + oldndevs), KM_SLEEP); 3294 for (i = 0; i < oldndevs; i++) 3295 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3296 KM_SLEEP) == 0); 3297 for (i = 0; i < ndevs; i++) 3298 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3299 KM_SLEEP) == 0); 3300 3301 VERIFY(nvlist_remove(sav->sav_config, config, 3302 DATA_TYPE_NVLIST_ARRAY) == 0); 3303 3304 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3305 config, newdevs, ndevs + oldndevs) == 0); 3306 for (i = 0; i < oldndevs + ndevs; i++) 3307 nvlist_free(newdevs[i]); 3308 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3309 } else { 3310 /* 3311 * Generate a new dev list. 3312 */ 3313 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3314 KM_SLEEP) == 0); 3315 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3316 devs, ndevs) == 0); 3317 } 3318} 3319 3320/* 3321 * Stop and drop level 2 ARC devices 3322 */ 3323void 3324spa_l2cache_drop(spa_t *spa) 3325{ 3326 vdev_t *vd; 3327 int i; 3328 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3329 3330 for (i = 0; i < sav->sav_count; i++) { 3331 uint64_t pool; 3332 3333 vd = sav->sav_vdevs[i]; 3334 ASSERT(vd != NULL); 3335 3336 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3337 pool != 0ULL && l2arc_vdev_present(vd)) 3338 l2arc_remove_vdev(vd); 3339 } 3340} 3341 3342/* 3343 * Pool Creation 3344 */ 3345int 3346spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3347 nvlist_t *zplprops) 3348{ 3349 spa_t *spa; 3350 char *altroot = NULL; 3351 vdev_t *rvd; 3352 dsl_pool_t *dp; 3353 dmu_tx_t *tx; 3354 int error = 0; 3355 uint64_t txg = TXG_INITIAL; 3356 nvlist_t **spares, **l2cache; 3357 uint_t nspares, nl2cache; 3358 uint64_t version, obj; 3359 boolean_t has_features; 3360 3361 /* 3362 * If this pool already exists, return failure. 3363 */ 3364 mutex_enter(&spa_namespace_lock); 3365 if (spa_lookup(pool) != NULL) { 3366 mutex_exit(&spa_namespace_lock); 3367 return (SET_ERROR(EEXIST)); 3368 } 3369 3370 /* 3371 * Allocate a new spa_t structure. 3372 */ 3373 (void) nvlist_lookup_string(props, 3374 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3375 spa = spa_add(pool, NULL, altroot); 3376 spa_activate(spa, spa_mode_global); 3377 3378 if (props && (error = spa_prop_validate(spa, props))) { 3379 spa_deactivate(spa); 3380 spa_remove(spa); 3381 mutex_exit(&spa_namespace_lock); 3382 return (error); 3383 } 3384 3385 has_features = B_FALSE; 3386 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3387 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3388 if (zpool_prop_feature(nvpair_name(elem))) 3389 has_features = B_TRUE; 3390 } 3391 3392 if (has_features || nvlist_lookup_uint64(props, 3393 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3394 version = SPA_VERSION; 3395 } 3396 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3397 3398 spa->spa_first_txg = txg; 3399 spa->spa_uberblock.ub_txg = txg - 1; 3400 spa->spa_uberblock.ub_version = version; 3401 spa->spa_ubsync = spa->spa_uberblock; 3402 3403 /* 3404 * Create "The Godfather" zio to hold all async IOs 3405 */ 3406 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 3407 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 3408 3409 /* 3410 * Create the root vdev. 3411 */ 3412 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3413 3414 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3415 3416 ASSERT(error != 0 || rvd != NULL); 3417 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3418 3419 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3420 error = SET_ERROR(EINVAL); 3421 3422 if (error == 0 && 3423 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3424 (error = spa_validate_aux(spa, nvroot, txg, 3425 VDEV_ALLOC_ADD)) == 0) { 3426 for (int c = 0; c < rvd->vdev_children; c++) { 3427 vdev_ashift_optimize(rvd->vdev_child[c]); 3428 vdev_metaslab_set_size(rvd->vdev_child[c]); 3429 vdev_expand(rvd->vdev_child[c], txg); 3430 } 3431 } 3432 3433 spa_config_exit(spa, SCL_ALL, FTAG); 3434 3435 if (error != 0) { 3436 spa_unload(spa); 3437 spa_deactivate(spa); 3438 spa_remove(spa); 3439 mutex_exit(&spa_namespace_lock); 3440 return (error); 3441 } 3442 3443 /* 3444 * Get the list of spares, if specified. 3445 */ 3446 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3447 &spares, &nspares) == 0) { 3448 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3449 KM_SLEEP) == 0); 3450 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3451 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3452 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3453 spa_load_spares(spa); 3454 spa_config_exit(spa, SCL_ALL, FTAG); 3455 spa->spa_spares.sav_sync = B_TRUE; 3456 } 3457 3458 /* 3459 * Get the list of level 2 cache devices, if specified. 3460 */ 3461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3462 &l2cache, &nl2cache) == 0) { 3463 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3464 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3465 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3466 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3467 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3468 spa_load_l2cache(spa); 3469 spa_config_exit(spa, SCL_ALL, FTAG); 3470 spa->spa_l2cache.sav_sync = B_TRUE; 3471 } 3472 3473 spa->spa_is_initializing = B_TRUE; 3474 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3475 spa->spa_meta_objset = dp->dp_meta_objset; 3476 spa->spa_is_initializing = B_FALSE; 3477 3478 /* 3479 * Create DDTs (dedup tables). 3480 */ 3481 ddt_create(spa); 3482 3483 spa_update_dspace(spa); 3484 3485 tx = dmu_tx_create_assigned(dp, txg); 3486 3487 /* 3488 * Create the pool config object. 3489 */ 3490 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3491 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3492 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3493 3494 if (zap_add(spa->spa_meta_objset, 3495 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3496 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3497 cmn_err(CE_PANIC, "failed to add pool config"); 3498 } 3499 3500 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3501 spa_feature_create_zap_objects(spa, tx); 3502 3503 if (zap_add(spa->spa_meta_objset, 3504 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3505 sizeof (uint64_t), 1, &version, tx) != 0) { 3506 cmn_err(CE_PANIC, "failed to add pool version"); 3507 } 3508 3509 /* Newly created pools with the right version are always deflated. */ 3510 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3511 spa->spa_deflate = TRUE; 3512 if (zap_add(spa->spa_meta_objset, 3513 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3514 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3515 cmn_err(CE_PANIC, "failed to add deflate"); 3516 } 3517 } 3518 3519 /* 3520 * Create the deferred-free bpobj. Turn off compression 3521 * because sync-to-convergence takes longer if the blocksize 3522 * keeps changing. 3523 */ 3524 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3525 dmu_object_set_compress(spa->spa_meta_objset, obj, 3526 ZIO_COMPRESS_OFF, tx); 3527 if (zap_add(spa->spa_meta_objset, 3528 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3529 sizeof (uint64_t), 1, &obj, tx) != 0) { 3530 cmn_err(CE_PANIC, "failed to add bpobj"); 3531 } 3532 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3533 spa->spa_meta_objset, obj)); 3534 3535 /* 3536 * Create the pool's history object. 3537 */ 3538 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3539 spa_history_create_obj(spa, tx); 3540 3541 /* 3542 * Set pool properties. 3543 */ 3544 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3545 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3546 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3547 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3548 3549 if (props != NULL) { 3550 spa_configfile_set(spa, props, B_FALSE); 3551 spa_sync_props(props, tx); 3552 } 3553 3554 dmu_tx_commit(tx); 3555 3556 spa->spa_sync_on = B_TRUE; 3557 txg_sync_start(spa->spa_dsl_pool); 3558 3559 /* 3560 * We explicitly wait for the first transaction to complete so that our 3561 * bean counters are appropriately updated. 3562 */ 3563 txg_wait_synced(spa->spa_dsl_pool, txg); 3564 3565 spa_config_sync(spa, B_FALSE, B_TRUE); 3566 3567 spa_history_log_version(spa, "create"); 3568 3569 spa->spa_minref = refcount_count(&spa->spa_refcount); 3570 3571 mutex_exit(&spa_namespace_lock); 3572 3573 return (0); 3574} 3575 3576#ifdef _KERNEL 3577#if defined(sun) 3578/* 3579 * Get the root pool information from the root disk, then import the root pool 3580 * during the system boot up time. 3581 */ 3582extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3583 3584static nvlist_t * 3585spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3586{ 3587 nvlist_t *config; 3588 nvlist_t *nvtop, *nvroot; 3589 uint64_t pgid; 3590 3591 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3592 return (NULL); 3593 3594 /* 3595 * Add this top-level vdev to the child array. 3596 */ 3597 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3598 &nvtop) == 0); 3599 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3600 &pgid) == 0); 3601 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3602 3603 /* 3604 * Put this pool's top-level vdevs into a root vdev. 3605 */ 3606 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3607 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3608 VDEV_TYPE_ROOT) == 0); 3609 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3610 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3611 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3612 &nvtop, 1) == 0); 3613 3614 /* 3615 * Replace the existing vdev_tree with the new root vdev in 3616 * this pool's configuration (remove the old, add the new). 3617 */ 3618 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3619 nvlist_free(nvroot); 3620 return (config); 3621} 3622 3623/* 3624 * Walk the vdev tree and see if we can find a device with "better" 3625 * configuration. A configuration is "better" if the label on that 3626 * device has a more recent txg. 3627 */ 3628static void 3629spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3630{ 3631 for (int c = 0; c < vd->vdev_children; c++) 3632 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3633 3634 if (vd->vdev_ops->vdev_op_leaf) { 3635 nvlist_t *label; 3636 uint64_t label_txg; 3637 3638 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3639 &label) != 0) 3640 return; 3641 3642 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3643 &label_txg) == 0); 3644 3645 /* 3646 * Do we have a better boot device? 3647 */ 3648 if (label_txg > *txg) { 3649 *txg = label_txg; 3650 *avd = vd; 3651 } 3652 nvlist_free(label); 3653 } 3654} 3655 3656/* 3657 * Import a root pool. 3658 * 3659 * For x86. devpath_list will consist of devid and/or physpath name of 3660 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3661 * The GRUB "findroot" command will return the vdev we should boot. 3662 * 3663 * For Sparc, devpath_list consists the physpath name of the booting device 3664 * no matter the rootpool is a single device pool or a mirrored pool. 3665 * e.g. 3666 * "/pci@1f,0/ide@d/disk@0,0:a" 3667 */ 3668int 3669spa_import_rootpool(char *devpath, char *devid) 3670{ 3671 spa_t *spa; 3672 vdev_t *rvd, *bvd, *avd = NULL; 3673 nvlist_t *config, *nvtop; 3674 uint64_t guid, txg; 3675 char *pname; 3676 int error; 3677 3678 /* 3679 * Read the label from the boot device and generate a configuration. 3680 */ 3681 config = spa_generate_rootconf(devpath, devid, &guid); 3682#if defined(_OBP) && defined(_KERNEL) 3683 if (config == NULL) { 3684 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3685 /* iscsi boot */ 3686 get_iscsi_bootpath_phy(devpath); 3687 config = spa_generate_rootconf(devpath, devid, &guid); 3688 } 3689 } 3690#endif 3691 if (config == NULL) { 3692 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3693 devpath); 3694 return (SET_ERROR(EIO)); 3695 } 3696 3697 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3698 &pname) == 0); 3699 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3700 3701 mutex_enter(&spa_namespace_lock); 3702 if ((spa = spa_lookup(pname)) != NULL) { 3703 /* 3704 * Remove the existing root pool from the namespace so that we 3705 * can replace it with the correct config we just read in. 3706 */ 3707 spa_remove(spa); 3708 } 3709 3710 spa = spa_add(pname, config, NULL); 3711 spa->spa_is_root = B_TRUE; 3712 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3713 3714 /* 3715 * Build up a vdev tree based on the boot device's label config. 3716 */ 3717 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3718 &nvtop) == 0); 3719 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3720 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3721 VDEV_ALLOC_ROOTPOOL); 3722 spa_config_exit(spa, SCL_ALL, FTAG); 3723 if (error) { 3724 mutex_exit(&spa_namespace_lock); 3725 nvlist_free(config); 3726 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3727 pname); 3728 return (error); 3729 } 3730 3731 /* 3732 * Get the boot vdev. 3733 */ 3734 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3735 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3736 (u_longlong_t)guid); 3737 error = SET_ERROR(ENOENT); 3738 goto out; 3739 } 3740 3741 /* 3742 * Determine if there is a better boot device. 3743 */ 3744 avd = bvd; 3745 spa_alt_rootvdev(rvd, &avd, &txg); 3746 if (avd != bvd) { 3747 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3748 "try booting from '%s'", avd->vdev_path); 3749 error = SET_ERROR(EINVAL); 3750 goto out; 3751 } 3752 3753 /* 3754 * If the boot device is part of a spare vdev then ensure that 3755 * we're booting off the active spare. 3756 */ 3757 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3758 !bvd->vdev_isspare) { 3759 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3760 "try booting from '%s'", 3761 bvd->vdev_parent-> 3762 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3763 error = SET_ERROR(EINVAL); 3764 goto out; 3765 } 3766 3767 error = 0; 3768out: 3769 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3770 vdev_free(rvd); 3771 spa_config_exit(spa, SCL_ALL, FTAG); 3772 mutex_exit(&spa_namespace_lock); 3773 3774 nvlist_free(config); 3775 return (error); 3776} 3777 3778#else 3779 3780extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3781 uint64_t *count); 3782 3783static nvlist_t * 3784spa_generate_rootconf(const char *name) 3785{ 3786 nvlist_t **configs, **tops; 3787 nvlist_t *config; 3788 nvlist_t *best_cfg, *nvtop, *nvroot; 3789 uint64_t *holes; 3790 uint64_t best_txg; 3791 uint64_t nchildren; 3792 uint64_t pgid; 3793 uint64_t count; 3794 uint64_t i; 3795 uint_t nholes; 3796 3797 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3798 return (NULL); 3799 3800 ASSERT3U(count, !=, 0); 3801 best_txg = 0; 3802 for (i = 0; i < count; i++) { 3803 uint64_t txg; 3804 3805 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 3806 &txg) == 0); 3807 if (txg > best_txg) { 3808 best_txg = txg; 3809 best_cfg = configs[i]; 3810 } 3811 } 3812 3813 /* 3814 * Multi-vdev root pool configuration discovery is not supported yet. 3815 */ 3816 nchildren = 1; 3817 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 3818 holes = NULL; 3819 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 3820 &holes, &nholes); 3821 3822 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 3823 for (i = 0; i < nchildren; i++) { 3824 if (i >= count) 3825 break; 3826 if (configs[i] == NULL) 3827 continue; 3828 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 3829 &nvtop) == 0); 3830 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 3831 } 3832 for (i = 0; holes != NULL && i < nholes; i++) { 3833 if (i >= nchildren) 3834 continue; 3835 if (tops[holes[i]] != NULL) 3836 continue; 3837 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 3838 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 3839 VDEV_TYPE_HOLE) == 0); 3840 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 3841 holes[i]) == 0); 3842 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 3843 0) == 0); 3844 } 3845 for (i = 0; i < nchildren; i++) { 3846 if (tops[i] != NULL) 3847 continue; 3848 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 3849 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 3850 VDEV_TYPE_MISSING) == 0); 3851 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 3852 i) == 0); 3853 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 3854 0) == 0); 3855 } 3856 3857 /* 3858 * Create pool config based on the best vdev config. 3859 */ 3860 nvlist_dup(best_cfg, &config, KM_SLEEP); 3861 3862 /* 3863 * Put this pool's top-level vdevs into a root vdev. 3864 */ 3865 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3866 &pgid) == 0); 3867 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3868 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3869 VDEV_TYPE_ROOT) == 0); 3870 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3871 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3872 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3873 tops, nchildren) == 0); 3874 3875 /* 3876 * Replace the existing vdev_tree with the new root vdev in 3877 * this pool's configuration (remove the old, add the new). 3878 */ 3879 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3880 3881 /* 3882 * Drop vdev config elements that should not be present at pool level. 3883 */ 3884 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 3885 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 3886 3887 for (i = 0; i < count; i++) 3888 nvlist_free(configs[i]); 3889 kmem_free(configs, count * sizeof(void *)); 3890 for (i = 0; i < nchildren; i++) 3891 nvlist_free(tops[i]); 3892 kmem_free(tops, nchildren * sizeof(void *)); 3893 nvlist_free(nvroot); 3894 return (config); 3895} 3896 3897int 3898spa_import_rootpool(const char *name) 3899{ 3900 spa_t *spa; 3901 vdev_t *rvd, *bvd, *avd = NULL; 3902 nvlist_t *config, *nvtop; 3903 uint64_t txg; 3904 char *pname; 3905 int error; 3906 3907 /* 3908 * Read the label from the boot device and generate a configuration. 3909 */ 3910 config = spa_generate_rootconf(name); 3911 3912 mutex_enter(&spa_namespace_lock); 3913 if (config != NULL) { 3914 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3915 &pname) == 0 && strcmp(name, pname) == 0); 3916 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 3917 == 0); 3918 3919 if ((spa = spa_lookup(pname)) != NULL) { 3920 /* 3921 * Remove the existing root pool from the namespace so 3922 * that we can replace it with the correct config 3923 * we just read in. 3924 */ 3925 spa_remove(spa); 3926 } 3927 spa = spa_add(pname, config, NULL); 3928 3929 /* 3930 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 3931 * via spa_version(). 3932 */ 3933 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 3934 &spa->spa_ubsync.ub_version) != 0) 3935 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 3936 } else if ((spa = spa_lookup(name)) == NULL) { 3937 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 3938 name); 3939 return (EIO); 3940 } else { 3941 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 3942 } 3943 spa->spa_is_root = B_TRUE; 3944 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3945 3946 /* 3947 * Build up a vdev tree based on the boot device's label config. 3948 */ 3949 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3950 &nvtop) == 0); 3951 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3952 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3953 VDEV_ALLOC_ROOTPOOL); 3954 spa_config_exit(spa, SCL_ALL, FTAG); 3955 if (error) { 3956 mutex_exit(&spa_namespace_lock); 3957 nvlist_free(config); 3958 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3959 pname); 3960 return (error); 3961 } 3962 3963 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3964 vdev_free(rvd); 3965 spa_config_exit(spa, SCL_ALL, FTAG); 3966 mutex_exit(&spa_namespace_lock); 3967 3968 nvlist_free(config); 3969 return (0); 3970} 3971 3972#endif /* sun */ 3973#endif 3974 3975/* 3976 * Import a non-root pool into the system. 3977 */ 3978int 3979spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3980{ 3981 spa_t *spa; 3982 char *altroot = NULL; 3983 spa_load_state_t state = SPA_LOAD_IMPORT; 3984 zpool_rewind_policy_t policy; 3985 uint64_t mode = spa_mode_global; 3986 uint64_t readonly = B_FALSE; 3987 int error; 3988 nvlist_t *nvroot; 3989 nvlist_t **spares, **l2cache; 3990 uint_t nspares, nl2cache; 3991 3992 /* 3993 * If a pool with this name exists, return failure. 3994 */ 3995 mutex_enter(&spa_namespace_lock); 3996 if (spa_lookup(pool) != NULL) { 3997 mutex_exit(&spa_namespace_lock); 3998 return (SET_ERROR(EEXIST)); 3999 } 4000 4001 /* 4002 * Create and initialize the spa structure. 4003 */ 4004 (void) nvlist_lookup_string(props, 4005 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4006 (void) nvlist_lookup_uint64(props, 4007 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4008 if (readonly) 4009 mode = FREAD; 4010 spa = spa_add(pool, config, altroot); 4011 spa->spa_import_flags = flags; 4012 4013 /* 4014 * Verbatim import - Take a pool and insert it into the namespace 4015 * as if it had been loaded at boot. 4016 */ 4017 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4018 if (props != NULL) 4019 spa_configfile_set(spa, props, B_FALSE); 4020 4021 spa_config_sync(spa, B_FALSE, B_TRUE); 4022 4023 mutex_exit(&spa_namespace_lock); 4024 return (0); 4025 } 4026 4027 spa_activate(spa, mode); 4028 4029 /* 4030 * Don't start async tasks until we know everything is healthy. 4031 */ 4032 spa_async_suspend(spa); 4033 4034 zpool_get_rewind_policy(config, &policy); 4035 if (policy.zrp_request & ZPOOL_DO_REWIND) 4036 state = SPA_LOAD_RECOVER; 4037 4038 /* 4039 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4040 * because the user-supplied config is actually the one to trust when 4041 * doing an import. 4042 */ 4043 if (state != SPA_LOAD_RECOVER) 4044 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4045 4046 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4047 policy.zrp_request); 4048 4049 /* 4050 * Propagate anything learned while loading the pool and pass it 4051 * back to caller (i.e. rewind info, missing devices, etc). 4052 */ 4053 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4054 spa->spa_load_info) == 0); 4055 4056 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4057 /* 4058 * Toss any existing sparelist, as it doesn't have any validity 4059 * anymore, and conflicts with spa_has_spare(). 4060 */ 4061 if (spa->spa_spares.sav_config) { 4062 nvlist_free(spa->spa_spares.sav_config); 4063 spa->spa_spares.sav_config = NULL; 4064 spa_load_spares(spa); 4065 } 4066 if (spa->spa_l2cache.sav_config) { 4067 nvlist_free(spa->spa_l2cache.sav_config); 4068 spa->spa_l2cache.sav_config = NULL; 4069 spa_load_l2cache(spa); 4070 } 4071 4072 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4073 &nvroot) == 0); 4074 if (error == 0) 4075 error = spa_validate_aux(spa, nvroot, -1ULL, 4076 VDEV_ALLOC_SPARE); 4077 if (error == 0) 4078 error = spa_validate_aux(spa, nvroot, -1ULL, 4079 VDEV_ALLOC_L2CACHE); 4080 spa_config_exit(spa, SCL_ALL, FTAG); 4081 4082 if (props != NULL) 4083 spa_configfile_set(spa, props, B_FALSE); 4084 4085 if (error != 0 || (props && spa_writeable(spa) && 4086 (error = spa_prop_set(spa, props)))) { 4087 spa_unload(spa); 4088 spa_deactivate(spa); 4089 spa_remove(spa); 4090 mutex_exit(&spa_namespace_lock); 4091 return (error); 4092 } 4093 4094 spa_async_resume(spa); 4095 4096 /* 4097 * Override any spares and level 2 cache devices as specified by 4098 * the user, as these may have correct device names/devids, etc. 4099 */ 4100 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4101 &spares, &nspares) == 0) { 4102 if (spa->spa_spares.sav_config) 4103 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4104 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4105 else 4106 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4107 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4108 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4109 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4111 spa_load_spares(spa); 4112 spa_config_exit(spa, SCL_ALL, FTAG); 4113 spa->spa_spares.sav_sync = B_TRUE; 4114 } 4115 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4116 &l2cache, &nl2cache) == 0) { 4117 if (spa->spa_l2cache.sav_config) 4118 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4119 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4120 else 4121 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4122 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4123 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4124 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4125 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4126 spa_load_l2cache(spa); 4127 spa_config_exit(spa, SCL_ALL, FTAG); 4128 spa->spa_l2cache.sav_sync = B_TRUE; 4129 } 4130 4131 /* 4132 * Check for any removed devices. 4133 */ 4134 if (spa->spa_autoreplace) { 4135 spa_aux_check_removed(&spa->spa_spares); 4136 spa_aux_check_removed(&spa->spa_l2cache); 4137 } 4138 4139 if (spa_writeable(spa)) { 4140 /* 4141 * Update the config cache to include the newly-imported pool. 4142 */ 4143 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4144 } 4145 4146 /* 4147 * It's possible that the pool was expanded while it was exported. 4148 * We kick off an async task to handle this for us. 4149 */ 4150 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4151 4152 mutex_exit(&spa_namespace_lock); 4153 spa_history_log_version(spa, "import"); 4154 4155#ifdef __FreeBSD__ 4156#ifdef _KERNEL 4157 zvol_create_minors(pool); 4158#endif 4159#endif 4160 return (0); 4161} 4162 4163nvlist_t * 4164spa_tryimport(nvlist_t *tryconfig) 4165{ 4166 nvlist_t *config = NULL; 4167 char *poolname; 4168 spa_t *spa; 4169 uint64_t state; 4170 int error; 4171 4172 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4173 return (NULL); 4174 4175 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4176 return (NULL); 4177 4178 /* 4179 * Create and initialize the spa structure. 4180 */ 4181 mutex_enter(&spa_namespace_lock); 4182 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4183 spa_activate(spa, FREAD); 4184 4185 /* 4186 * Pass off the heavy lifting to spa_load(). 4187 * Pass TRUE for mosconfig because the user-supplied config 4188 * is actually the one to trust when doing an import. 4189 */ 4190 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4191 4192 /* 4193 * If 'tryconfig' was at least parsable, return the current config. 4194 */ 4195 if (spa->spa_root_vdev != NULL) { 4196 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4197 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4198 poolname) == 0); 4199 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4200 state) == 0); 4201 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4202 spa->spa_uberblock.ub_timestamp) == 0); 4203 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4204 spa->spa_load_info) == 0); 4205 4206 /* 4207 * If the bootfs property exists on this pool then we 4208 * copy it out so that external consumers can tell which 4209 * pools are bootable. 4210 */ 4211 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4212 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4213 4214 /* 4215 * We have to play games with the name since the 4216 * pool was opened as TRYIMPORT_NAME. 4217 */ 4218 if (dsl_dsobj_to_dsname(spa_name(spa), 4219 spa->spa_bootfs, tmpname) == 0) { 4220 char *cp; 4221 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4222 4223 cp = strchr(tmpname, '/'); 4224 if (cp == NULL) { 4225 (void) strlcpy(dsname, tmpname, 4226 MAXPATHLEN); 4227 } else { 4228 (void) snprintf(dsname, MAXPATHLEN, 4229 "%s/%s", poolname, ++cp); 4230 } 4231 VERIFY(nvlist_add_string(config, 4232 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4233 kmem_free(dsname, MAXPATHLEN); 4234 } 4235 kmem_free(tmpname, MAXPATHLEN); 4236 } 4237 4238 /* 4239 * Add the list of hot spares and level 2 cache devices. 4240 */ 4241 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4242 spa_add_spares(spa, config); 4243 spa_add_l2cache(spa, config); 4244 spa_config_exit(spa, SCL_CONFIG, FTAG); 4245 } 4246 4247 spa_unload(spa); 4248 spa_deactivate(spa); 4249 spa_remove(spa); 4250 mutex_exit(&spa_namespace_lock); 4251 4252 return (config); 4253} 4254 4255/* 4256 * Pool export/destroy 4257 * 4258 * The act of destroying or exporting a pool is very simple. We make sure there 4259 * is no more pending I/O and any references to the pool are gone. Then, we 4260 * update the pool state and sync all the labels to disk, removing the 4261 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4262 * we don't sync the labels or remove the configuration cache. 4263 */ 4264static int 4265spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4266 boolean_t force, boolean_t hardforce) 4267{ 4268 spa_t *spa; 4269 4270 if (oldconfig) 4271 *oldconfig = NULL; 4272 4273 if (!(spa_mode_global & FWRITE)) 4274 return (SET_ERROR(EROFS)); 4275 4276 mutex_enter(&spa_namespace_lock); 4277 if ((spa = spa_lookup(pool)) == NULL) { 4278 mutex_exit(&spa_namespace_lock); 4279 return (SET_ERROR(ENOENT)); 4280 } 4281 4282 /* 4283 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4284 * reacquire the namespace lock, and see if we can export. 4285 */ 4286 spa_open_ref(spa, FTAG); 4287 mutex_exit(&spa_namespace_lock); 4288 spa_async_suspend(spa); 4289 mutex_enter(&spa_namespace_lock); 4290 spa_close(spa, FTAG); 4291 4292 /* 4293 * The pool will be in core if it's openable, 4294 * in which case we can modify its state. 4295 */ 4296 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4297 /* 4298 * Objsets may be open only because they're dirty, so we 4299 * have to force it to sync before checking spa_refcnt. 4300 */ 4301 txg_wait_synced(spa->spa_dsl_pool, 0); 4302 4303 /* 4304 * A pool cannot be exported or destroyed if there are active 4305 * references. If we are resetting a pool, allow references by 4306 * fault injection handlers. 4307 */ 4308 if (!spa_refcount_zero(spa) || 4309 (spa->spa_inject_ref != 0 && 4310 new_state != POOL_STATE_UNINITIALIZED)) { 4311 spa_async_resume(spa); 4312 mutex_exit(&spa_namespace_lock); 4313 return (SET_ERROR(EBUSY)); 4314 } 4315 4316 /* 4317 * A pool cannot be exported if it has an active shared spare. 4318 * This is to prevent other pools stealing the active spare 4319 * from an exported pool. At user's own will, such pool can 4320 * be forcedly exported. 4321 */ 4322 if (!force && new_state == POOL_STATE_EXPORTED && 4323 spa_has_active_shared_spare(spa)) { 4324 spa_async_resume(spa); 4325 mutex_exit(&spa_namespace_lock); 4326 return (SET_ERROR(EXDEV)); 4327 } 4328 4329 /* 4330 * We want this to be reflected on every label, 4331 * so mark them all dirty. spa_unload() will do the 4332 * final sync that pushes these changes out. 4333 */ 4334 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4335 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4336 spa->spa_state = new_state; 4337 spa->spa_final_txg = spa_last_synced_txg(spa) + 4338 TXG_DEFER_SIZE + 1; 4339 vdev_config_dirty(spa->spa_root_vdev); 4340 spa_config_exit(spa, SCL_ALL, FTAG); 4341 } 4342 } 4343 4344 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4345 4346 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4347 spa_unload(spa); 4348 spa_deactivate(spa); 4349 } 4350 4351 if (oldconfig && spa->spa_config) 4352 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4353 4354 if (new_state != POOL_STATE_UNINITIALIZED) { 4355 if (!hardforce) 4356 spa_config_sync(spa, B_TRUE, B_TRUE); 4357 spa_remove(spa); 4358 } 4359 mutex_exit(&spa_namespace_lock); 4360 4361 return (0); 4362} 4363 4364/* 4365 * Destroy a storage pool. 4366 */ 4367int 4368spa_destroy(char *pool) 4369{ 4370 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4371 B_FALSE, B_FALSE)); 4372} 4373 4374/* 4375 * Export a storage pool. 4376 */ 4377int 4378spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4379 boolean_t hardforce) 4380{ 4381 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4382 force, hardforce)); 4383} 4384 4385/* 4386 * Similar to spa_export(), this unloads the spa_t without actually removing it 4387 * from the namespace in any way. 4388 */ 4389int 4390spa_reset(char *pool) 4391{ 4392 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4393 B_FALSE, B_FALSE)); 4394} 4395 4396/* 4397 * ========================================================================== 4398 * Device manipulation 4399 * ========================================================================== 4400 */ 4401 4402/* 4403 * Add a device to a storage pool. 4404 */ 4405int 4406spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4407{ 4408 uint64_t txg, id; 4409 int error; 4410 vdev_t *rvd = spa->spa_root_vdev; 4411 vdev_t *vd, *tvd; 4412 nvlist_t **spares, **l2cache; 4413 uint_t nspares, nl2cache; 4414 4415 ASSERT(spa_writeable(spa)); 4416 4417 txg = spa_vdev_enter(spa); 4418 4419 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4420 VDEV_ALLOC_ADD)) != 0) 4421 return (spa_vdev_exit(spa, NULL, txg, error)); 4422 4423 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4424 4425 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4426 &nspares) != 0) 4427 nspares = 0; 4428 4429 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4430 &nl2cache) != 0) 4431 nl2cache = 0; 4432 4433 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4434 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4435 4436 if (vd->vdev_children != 0 && 4437 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4438 return (spa_vdev_exit(spa, vd, txg, error)); 4439 4440 /* 4441 * We must validate the spares and l2cache devices after checking the 4442 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4443 */ 4444 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4445 return (spa_vdev_exit(spa, vd, txg, error)); 4446 4447 /* 4448 * Transfer each new top-level vdev from vd to rvd. 4449 */ 4450 for (int c = 0; c < vd->vdev_children; c++) { 4451 4452 /* 4453 * Set the vdev id to the first hole, if one exists. 4454 */ 4455 for (id = 0; id < rvd->vdev_children; id++) { 4456 if (rvd->vdev_child[id]->vdev_ishole) { 4457 vdev_free(rvd->vdev_child[id]); 4458 break; 4459 } 4460 } 4461 tvd = vd->vdev_child[c]; 4462 vdev_remove_child(vd, tvd); 4463 tvd->vdev_id = id; 4464 vdev_add_child(rvd, tvd); 4465 vdev_config_dirty(tvd); 4466 } 4467 4468 if (nspares != 0) { 4469 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4470 ZPOOL_CONFIG_SPARES); 4471 spa_load_spares(spa); 4472 spa->spa_spares.sav_sync = B_TRUE; 4473 } 4474 4475 if (nl2cache != 0) { 4476 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4477 ZPOOL_CONFIG_L2CACHE); 4478 spa_load_l2cache(spa); 4479 spa->spa_l2cache.sav_sync = B_TRUE; 4480 } 4481 4482 /* 4483 * We have to be careful when adding new vdevs to an existing pool. 4484 * If other threads start allocating from these vdevs before we 4485 * sync the config cache, and we lose power, then upon reboot we may 4486 * fail to open the pool because there are DVAs that the config cache 4487 * can't translate. Therefore, we first add the vdevs without 4488 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4489 * and then let spa_config_update() initialize the new metaslabs. 4490 * 4491 * spa_load() checks for added-but-not-initialized vdevs, so that 4492 * if we lose power at any point in this sequence, the remaining 4493 * steps will be completed the next time we load the pool. 4494 */ 4495 (void) spa_vdev_exit(spa, vd, txg, 0); 4496 4497 mutex_enter(&spa_namespace_lock); 4498 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4499 mutex_exit(&spa_namespace_lock); 4500 4501 return (0); 4502} 4503 4504/* 4505 * Attach a device to a mirror. The arguments are the path to any device 4506 * in the mirror, and the nvroot for the new device. If the path specifies 4507 * a device that is not mirrored, we automatically insert the mirror vdev. 4508 * 4509 * If 'replacing' is specified, the new device is intended to replace the 4510 * existing device; in this case the two devices are made into their own 4511 * mirror using the 'replacing' vdev, which is functionally identical to 4512 * the mirror vdev (it actually reuses all the same ops) but has a few 4513 * extra rules: you can't attach to it after it's been created, and upon 4514 * completion of resilvering, the first disk (the one being replaced) 4515 * is automatically detached. 4516 */ 4517int 4518spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4519{ 4520 uint64_t txg, dtl_max_txg; 4521 vdev_t *rvd = spa->spa_root_vdev; 4522 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4523 vdev_ops_t *pvops; 4524 char *oldvdpath, *newvdpath; 4525 int newvd_isspare; 4526 int error; 4527 4528 ASSERT(spa_writeable(spa)); 4529 4530 txg = spa_vdev_enter(spa); 4531 4532 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4533 4534 if (oldvd == NULL) 4535 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4536 4537 if (!oldvd->vdev_ops->vdev_op_leaf) 4538 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4539 4540 pvd = oldvd->vdev_parent; 4541 4542 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4543 VDEV_ALLOC_ATTACH)) != 0) 4544 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4545 4546 if (newrootvd->vdev_children != 1) 4547 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4548 4549 newvd = newrootvd->vdev_child[0]; 4550 4551 if (!newvd->vdev_ops->vdev_op_leaf) 4552 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4553 4554 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4555 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4556 4557 /* 4558 * Spares can't replace logs 4559 */ 4560 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4561 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4562 4563 if (!replacing) { 4564 /* 4565 * For attach, the only allowable parent is a mirror or the root 4566 * vdev. 4567 */ 4568 if (pvd->vdev_ops != &vdev_mirror_ops && 4569 pvd->vdev_ops != &vdev_root_ops) 4570 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4571 4572 pvops = &vdev_mirror_ops; 4573 } else { 4574 /* 4575 * Active hot spares can only be replaced by inactive hot 4576 * spares. 4577 */ 4578 if (pvd->vdev_ops == &vdev_spare_ops && 4579 oldvd->vdev_isspare && 4580 !spa_has_spare(spa, newvd->vdev_guid)) 4581 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4582 4583 /* 4584 * If the source is a hot spare, and the parent isn't already a 4585 * spare, then we want to create a new hot spare. Otherwise, we 4586 * want to create a replacing vdev. The user is not allowed to 4587 * attach to a spared vdev child unless the 'isspare' state is 4588 * the same (spare replaces spare, non-spare replaces 4589 * non-spare). 4590 */ 4591 if (pvd->vdev_ops == &vdev_replacing_ops && 4592 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4593 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4594 } else if (pvd->vdev_ops == &vdev_spare_ops && 4595 newvd->vdev_isspare != oldvd->vdev_isspare) { 4596 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4597 } 4598 4599 if (newvd->vdev_isspare) 4600 pvops = &vdev_spare_ops; 4601 else 4602 pvops = &vdev_replacing_ops; 4603 } 4604 4605 /* 4606 * Make sure the new device is big enough. 4607 */ 4608 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4609 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4610 4611 /* 4612 * The new device cannot have a higher alignment requirement 4613 * than the top-level vdev. 4614 */ 4615 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4616 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4617 4618 /* 4619 * If this is an in-place replacement, update oldvd's path and devid 4620 * to make it distinguishable from newvd, and unopenable from now on. 4621 */ 4622 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4623 spa_strfree(oldvd->vdev_path); 4624 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4625 KM_SLEEP); 4626 (void) sprintf(oldvd->vdev_path, "%s/%s", 4627 newvd->vdev_path, "old"); 4628 if (oldvd->vdev_devid != NULL) { 4629 spa_strfree(oldvd->vdev_devid); 4630 oldvd->vdev_devid = NULL; 4631 } 4632 } 4633 4634 /* mark the device being resilvered */ 4635 newvd->vdev_resilver_txg = txg; 4636 4637 /* 4638 * If the parent is not a mirror, or if we're replacing, insert the new 4639 * mirror/replacing/spare vdev above oldvd. 4640 */ 4641 if (pvd->vdev_ops != pvops) 4642 pvd = vdev_add_parent(oldvd, pvops); 4643 4644 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4645 ASSERT(pvd->vdev_ops == pvops); 4646 ASSERT(oldvd->vdev_parent == pvd); 4647 4648 /* 4649 * Extract the new device from its root and add it to pvd. 4650 */ 4651 vdev_remove_child(newrootvd, newvd); 4652 newvd->vdev_id = pvd->vdev_children; 4653 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4654 vdev_add_child(pvd, newvd); 4655 4656 tvd = newvd->vdev_top; 4657 ASSERT(pvd->vdev_top == tvd); 4658 ASSERT(tvd->vdev_parent == rvd); 4659 4660 vdev_config_dirty(tvd); 4661 4662 /* 4663 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4664 * for any dmu_sync-ed blocks. It will propagate upward when 4665 * spa_vdev_exit() calls vdev_dtl_reassess(). 4666 */ 4667 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4668 4669 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4670 dtl_max_txg - TXG_INITIAL); 4671 4672 if (newvd->vdev_isspare) { 4673 spa_spare_activate(newvd); 4674 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4675 } 4676 4677 oldvdpath = spa_strdup(oldvd->vdev_path); 4678 newvdpath = spa_strdup(newvd->vdev_path); 4679 newvd_isspare = newvd->vdev_isspare; 4680 4681 /* 4682 * Mark newvd's DTL dirty in this txg. 4683 */ 4684 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4685 4686 /* 4687 * Restart the resilver 4688 */ 4689 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4690 4691 /* 4692 * Commit the config 4693 */ 4694 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4695 4696 spa_history_log_internal(spa, "vdev attach", NULL, 4697 "%s vdev=%s %s vdev=%s", 4698 replacing && newvd_isspare ? "spare in" : 4699 replacing ? "replace" : "attach", newvdpath, 4700 replacing ? "for" : "to", oldvdpath); 4701 4702 spa_strfree(oldvdpath); 4703 spa_strfree(newvdpath); 4704 4705 if (spa->spa_bootfs) 4706 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4707 4708 return (0); 4709} 4710 4711/* 4712 * Detach a device from a mirror or replacing vdev. 4713 * 4714 * If 'replace_done' is specified, only detach if the parent 4715 * is a replacing vdev. 4716 */ 4717int 4718spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4719{ 4720 uint64_t txg; 4721 int error; 4722 vdev_t *rvd = spa->spa_root_vdev; 4723 vdev_t *vd, *pvd, *cvd, *tvd; 4724 boolean_t unspare = B_FALSE; 4725 uint64_t unspare_guid = 0; 4726 char *vdpath; 4727 4728 ASSERT(spa_writeable(spa)); 4729 4730 txg = spa_vdev_enter(spa); 4731 4732 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4733 4734 if (vd == NULL) 4735 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4736 4737 if (!vd->vdev_ops->vdev_op_leaf) 4738 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4739 4740 pvd = vd->vdev_parent; 4741 4742 /* 4743 * If the parent/child relationship is not as expected, don't do it. 4744 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4745 * vdev that's replacing B with C. The user's intent in replacing 4746 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4747 * the replace by detaching C, the expected behavior is to end up 4748 * M(A,B). But suppose that right after deciding to detach C, 4749 * the replacement of B completes. We would have M(A,C), and then 4750 * ask to detach C, which would leave us with just A -- not what 4751 * the user wanted. To prevent this, we make sure that the 4752 * parent/child relationship hasn't changed -- in this example, 4753 * that C's parent is still the replacing vdev R. 4754 */ 4755 if (pvd->vdev_guid != pguid && pguid != 0) 4756 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4757 4758 /* 4759 * Only 'replacing' or 'spare' vdevs can be replaced. 4760 */ 4761 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4762 pvd->vdev_ops != &vdev_spare_ops) 4763 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4764 4765 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4766 spa_version(spa) >= SPA_VERSION_SPARES); 4767 4768 /* 4769 * Only mirror, replacing, and spare vdevs support detach. 4770 */ 4771 if (pvd->vdev_ops != &vdev_replacing_ops && 4772 pvd->vdev_ops != &vdev_mirror_ops && 4773 pvd->vdev_ops != &vdev_spare_ops) 4774 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4775 4776 /* 4777 * If this device has the only valid copy of some data, 4778 * we cannot safely detach it. 4779 */ 4780 if (vdev_dtl_required(vd)) 4781 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4782 4783 ASSERT(pvd->vdev_children >= 2); 4784 4785 /* 4786 * If we are detaching the second disk from a replacing vdev, then 4787 * check to see if we changed the original vdev's path to have "/old" 4788 * at the end in spa_vdev_attach(). If so, undo that change now. 4789 */ 4790 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4791 vd->vdev_path != NULL) { 4792 size_t len = strlen(vd->vdev_path); 4793 4794 for (int c = 0; c < pvd->vdev_children; c++) { 4795 cvd = pvd->vdev_child[c]; 4796 4797 if (cvd == vd || cvd->vdev_path == NULL) 4798 continue; 4799 4800 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4801 strcmp(cvd->vdev_path + len, "/old") == 0) { 4802 spa_strfree(cvd->vdev_path); 4803 cvd->vdev_path = spa_strdup(vd->vdev_path); 4804 break; 4805 } 4806 } 4807 } 4808 4809 /* 4810 * If we are detaching the original disk from a spare, then it implies 4811 * that the spare should become a real disk, and be removed from the 4812 * active spare list for the pool. 4813 */ 4814 if (pvd->vdev_ops == &vdev_spare_ops && 4815 vd->vdev_id == 0 && 4816 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4817 unspare = B_TRUE; 4818 4819 /* 4820 * Erase the disk labels so the disk can be used for other things. 4821 * This must be done after all other error cases are handled, 4822 * but before we disembowel vd (so we can still do I/O to it). 4823 * But if we can't do it, don't treat the error as fatal -- 4824 * it may be that the unwritability of the disk is the reason 4825 * it's being detached! 4826 */ 4827 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4828 4829 /* 4830 * Remove vd from its parent and compact the parent's children. 4831 */ 4832 vdev_remove_child(pvd, vd); 4833 vdev_compact_children(pvd); 4834 4835 /* 4836 * Remember one of the remaining children so we can get tvd below. 4837 */ 4838 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4839 4840 /* 4841 * If we need to remove the remaining child from the list of hot spares, 4842 * do it now, marking the vdev as no longer a spare in the process. 4843 * We must do this before vdev_remove_parent(), because that can 4844 * change the GUID if it creates a new toplevel GUID. For a similar 4845 * reason, we must remove the spare now, in the same txg as the detach; 4846 * otherwise someone could attach a new sibling, change the GUID, and 4847 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4848 */ 4849 if (unspare) { 4850 ASSERT(cvd->vdev_isspare); 4851 spa_spare_remove(cvd); 4852 unspare_guid = cvd->vdev_guid; 4853 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4854 cvd->vdev_unspare = B_TRUE; 4855 } 4856 4857 /* 4858 * If the parent mirror/replacing vdev only has one child, 4859 * the parent is no longer needed. Remove it from the tree. 4860 */ 4861 if (pvd->vdev_children == 1) { 4862 if (pvd->vdev_ops == &vdev_spare_ops) 4863 cvd->vdev_unspare = B_FALSE; 4864 vdev_remove_parent(cvd); 4865 } 4866 4867 4868 /* 4869 * We don't set tvd until now because the parent we just removed 4870 * may have been the previous top-level vdev. 4871 */ 4872 tvd = cvd->vdev_top; 4873 ASSERT(tvd->vdev_parent == rvd); 4874 4875 /* 4876 * Reevaluate the parent vdev state. 4877 */ 4878 vdev_propagate_state(cvd); 4879 4880 /* 4881 * If the 'autoexpand' property is set on the pool then automatically 4882 * try to expand the size of the pool. For example if the device we 4883 * just detached was smaller than the others, it may be possible to 4884 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4885 * first so that we can obtain the updated sizes of the leaf vdevs. 4886 */ 4887 if (spa->spa_autoexpand) { 4888 vdev_reopen(tvd); 4889 vdev_expand(tvd, txg); 4890 } 4891 4892 vdev_config_dirty(tvd); 4893 4894 /* 4895 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4896 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4897 * But first make sure we're not on any *other* txg's DTL list, to 4898 * prevent vd from being accessed after it's freed. 4899 */ 4900 vdpath = spa_strdup(vd->vdev_path); 4901 for (int t = 0; t < TXG_SIZE; t++) 4902 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4903 vd->vdev_detached = B_TRUE; 4904 vdev_dirty(tvd, VDD_DTL, vd, txg); 4905 4906 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4907 4908 /* hang on to the spa before we release the lock */ 4909 spa_open_ref(spa, FTAG); 4910 4911 error = spa_vdev_exit(spa, vd, txg, 0); 4912 4913 spa_history_log_internal(spa, "detach", NULL, 4914 "vdev=%s", vdpath); 4915 spa_strfree(vdpath); 4916 4917 /* 4918 * If this was the removal of the original device in a hot spare vdev, 4919 * then we want to go through and remove the device from the hot spare 4920 * list of every other pool. 4921 */ 4922 if (unspare) { 4923 spa_t *altspa = NULL; 4924 4925 mutex_enter(&spa_namespace_lock); 4926 while ((altspa = spa_next(altspa)) != NULL) { 4927 if (altspa->spa_state != POOL_STATE_ACTIVE || 4928 altspa == spa) 4929 continue; 4930 4931 spa_open_ref(altspa, FTAG); 4932 mutex_exit(&spa_namespace_lock); 4933 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 4934 mutex_enter(&spa_namespace_lock); 4935 spa_close(altspa, FTAG); 4936 } 4937 mutex_exit(&spa_namespace_lock); 4938 4939 /* search the rest of the vdevs for spares to remove */ 4940 spa_vdev_resilver_done(spa); 4941 } 4942 4943 /* all done with the spa; OK to release */ 4944 mutex_enter(&spa_namespace_lock); 4945 spa_close(spa, FTAG); 4946 mutex_exit(&spa_namespace_lock); 4947 4948 return (error); 4949} 4950 4951/* 4952 * Split a set of devices from their mirrors, and create a new pool from them. 4953 */ 4954int 4955spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 4956 nvlist_t *props, boolean_t exp) 4957{ 4958 int error = 0; 4959 uint64_t txg, *glist; 4960 spa_t *newspa; 4961 uint_t c, children, lastlog; 4962 nvlist_t **child, *nvl, *tmp; 4963 dmu_tx_t *tx; 4964 char *altroot = NULL; 4965 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 4966 boolean_t activate_slog; 4967 4968 ASSERT(spa_writeable(spa)); 4969 4970 txg = spa_vdev_enter(spa); 4971 4972 /* clear the log and flush everything up to now */ 4973 activate_slog = spa_passivate_log(spa); 4974 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 4975 error = spa_offline_log(spa); 4976 txg = spa_vdev_config_enter(spa); 4977 4978 if (activate_slog) 4979 spa_activate_log(spa); 4980 4981 if (error != 0) 4982 return (spa_vdev_exit(spa, NULL, txg, error)); 4983 4984 /* check new spa name before going any further */ 4985 if (spa_lookup(newname) != NULL) 4986 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 4987 4988 /* 4989 * scan through all the children to ensure they're all mirrors 4990 */ 4991 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 4992 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 4993 &children) != 0) 4994 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4995 4996 /* first, check to ensure we've got the right child count */ 4997 rvd = spa->spa_root_vdev; 4998 lastlog = 0; 4999 for (c = 0; c < rvd->vdev_children; c++) { 5000 vdev_t *vd = rvd->vdev_child[c]; 5001 5002 /* don't count the holes & logs as children */ 5003 if (vd->vdev_islog || vd->vdev_ishole) { 5004 if (lastlog == 0) 5005 lastlog = c; 5006 continue; 5007 } 5008 5009 lastlog = 0; 5010 } 5011 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5012 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5013 5014 /* next, ensure no spare or cache devices are part of the split */ 5015 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5016 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5017 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5018 5019 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5020 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5021 5022 /* then, loop over each vdev and validate it */ 5023 for (c = 0; c < children; c++) { 5024 uint64_t is_hole = 0; 5025 5026 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5027 &is_hole); 5028 5029 if (is_hole != 0) { 5030 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5031 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5032 continue; 5033 } else { 5034 error = SET_ERROR(EINVAL); 5035 break; 5036 } 5037 } 5038 5039 /* which disk is going to be split? */ 5040 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5041 &glist[c]) != 0) { 5042 error = SET_ERROR(EINVAL); 5043 break; 5044 } 5045 5046 /* look it up in the spa */ 5047 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5048 if (vml[c] == NULL) { 5049 error = SET_ERROR(ENODEV); 5050 break; 5051 } 5052 5053 /* make sure there's nothing stopping the split */ 5054 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5055 vml[c]->vdev_islog || 5056 vml[c]->vdev_ishole || 5057 vml[c]->vdev_isspare || 5058 vml[c]->vdev_isl2cache || 5059 !vdev_writeable(vml[c]) || 5060 vml[c]->vdev_children != 0 || 5061 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5062 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5063 error = SET_ERROR(EINVAL); 5064 break; 5065 } 5066 5067 if (vdev_dtl_required(vml[c])) { 5068 error = SET_ERROR(EBUSY); 5069 break; 5070 } 5071 5072 /* we need certain info from the top level */ 5073 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5074 vml[c]->vdev_top->vdev_ms_array) == 0); 5075 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5076 vml[c]->vdev_top->vdev_ms_shift) == 0); 5077 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5078 vml[c]->vdev_top->vdev_asize) == 0); 5079 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5080 vml[c]->vdev_top->vdev_ashift) == 0); 5081 } 5082 5083 if (error != 0) { 5084 kmem_free(vml, children * sizeof (vdev_t *)); 5085 kmem_free(glist, children * sizeof (uint64_t)); 5086 return (spa_vdev_exit(spa, NULL, txg, error)); 5087 } 5088 5089 /* stop writers from using the disks */ 5090 for (c = 0; c < children; c++) { 5091 if (vml[c] != NULL) 5092 vml[c]->vdev_offline = B_TRUE; 5093 } 5094 vdev_reopen(spa->spa_root_vdev); 5095 5096 /* 5097 * Temporarily record the splitting vdevs in the spa config. This 5098 * will disappear once the config is regenerated. 5099 */ 5100 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5101 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5102 glist, children) == 0); 5103 kmem_free(glist, children * sizeof (uint64_t)); 5104 5105 mutex_enter(&spa->spa_props_lock); 5106 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5107 nvl) == 0); 5108 mutex_exit(&spa->spa_props_lock); 5109 spa->spa_config_splitting = nvl; 5110 vdev_config_dirty(spa->spa_root_vdev); 5111 5112 /* configure and create the new pool */ 5113 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5114 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5115 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5116 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5117 spa_version(spa)) == 0); 5118 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5119 spa->spa_config_txg) == 0); 5120 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5121 spa_generate_guid(NULL)) == 0); 5122 (void) nvlist_lookup_string(props, 5123 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5124 5125 /* add the new pool to the namespace */ 5126 newspa = spa_add(newname, config, altroot); 5127 newspa->spa_config_txg = spa->spa_config_txg; 5128 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5129 5130 /* release the spa config lock, retaining the namespace lock */ 5131 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5132 5133 if (zio_injection_enabled) 5134 zio_handle_panic_injection(spa, FTAG, 1); 5135 5136 spa_activate(newspa, spa_mode_global); 5137 spa_async_suspend(newspa); 5138 5139#ifndef sun 5140 /* mark that we are creating new spa by splitting */ 5141 newspa->spa_splitting_newspa = B_TRUE; 5142#endif 5143 /* create the new pool from the disks of the original pool */ 5144 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5145#ifndef sun 5146 newspa->spa_splitting_newspa = B_FALSE; 5147#endif 5148 if (error) 5149 goto out; 5150 5151 /* if that worked, generate a real config for the new pool */ 5152 if (newspa->spa_root_vdev != NULL) { 5153 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5154 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5155 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5156 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5157 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5158 B_TRUE)); 5159 } 5160 5161 /* set the props */ 5162 if (props != NULL) { 5163 spa_configfile_set(newspa, props, B_FALSE); 5164 error = spa_prop_set(newspa, props); 5165 if (error) 5166 goto out; 5167 } 5168 5169 /* flush everything */ 5170 txg = spa_vdev_config_enter(newspa); 5171 vdev_config_dirty(newspa->spa_root_vdev); 5172 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5173 5174 if (zio_injection_enabled) 5175 zio_handle_panic_injection(spa, FTAG, 2); 5176 5177 spa_async_resume(newspa); 5178 5179 /* finally, update the original pool's config */ 5180 txg = spa_vdev_config_enter(spa); 5181 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5182 error = dmu_tx_assign(tx, TXG_WAIT); 5183 if (error != 0) 5184 dmu_tx_abort(tx); 5185 for (c = 0; c < children; c++) { 5186 if (vml[c] != NULL) { 5187 vdev_split(vml[c]); 5188 if (error == 0) 5189 spa_history_log_internal(spa, "detach", tx, 5190 "vdev=%s", vml[c]->vdev_path); 5191 vdev_free(vml[c]); 5192 } 5193 } 5194 vdev_config_dirty(spa->spa_root_vdev); 5195 spa->spa_config_splitting = NULL; 5196 nvlist_free(nvl); 5197 if (error == 0) 5198 dmu_tx_commit(tx); 5199 (void) spa_vdev_exit(spa, NULL, txg, 0); 5200 5201 if (zio_injection_enabled) 5202 zio_handle_panic_injection(spa, FTAG, 3); 5203 5204 /* split is complete; log a history record */ 5205 spa_history_log_internal(newspa, "split", NULL, 5206 "from pool %s", spa_name(spa)); 5207 5208 kmem_free(vml, children * sizeof (vdev_t *)); 5209 5210 /* if we're not going to mount the filesystems in userland, export */ 5211 if (exp) 5212 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5213 B_FALSE, B_FALSE); 5214 5215 return (error); 5216 5217out: 5218 spa_unload(newspa); 5219 spa_deactivate(newspa); 5220 spa_remove(newspa); 5221 5222 txg = spa_vdev_config_enter(spa); 5223 5224 /* re-online all offlined disks */ 5225 for (c = 0; c < children; c++) { 5226 if (vml[c] != NULL) 5227 vml[c]->vdev_offline = B_FALSE; 5228 } 5229 vdev_reopen(spa->spa_root_vdev); 5230 5231 nvlist_free(spa->spa_config_splitting); 5232 spa->spa_config_splitting = NULL; 5233 (void) spa_vdev_exit(spa, NULL, txg, error); 5234 5235 kmem_free(vml, children * sizeof (vdev_t *)); 5236 return (error); 5237} 5238 5239static nvlist_t * 5240spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5241{ 5242 for (int i = 0; i < count; i++) { 5243 uint64_t guid; 5244 5245 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5246 &guid) == 0); 5247 5248 if (guid == target_guid) 5249 return (nvpp[i]); 5250 } 5251 5252 return (NULL); 5253} 5254 5255static void 5256spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5257 nvlist_t *dev_to_remove) 5258{ 5259 nvlist_t **newdev = NULL; 5260 5261 if (count > 1) 5262 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5263 5264 for (int i = 0, j = 0; i < count; i++) { 5265 if (dev[i] == dev_to_remove) 5266 continue; 5267 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5268 } 5269 5270 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5271 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5272 5273 for (int i = 0; i < count - 1; i++) 5274 nvlist_free(newdev[i]); 5275 5276 if (count > 1) 5277 kmem_free(newdev, (count - 1) * sizeof (void *)); 5278} 5279 5280/* 5281 * Evacuate the device. 5282 */ 5283static int 5284spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5285{ 5286 uint64_t txg; 5287 int error = 0; 5288 5289 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5290 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5291 ASSERT(vd == vd->vdev_top); 5292 5293 /* 5294 * Evacuate the device. We don't hold the config lock as writer 5295 * since we need to do I/O but we do keep the 5296 * spa_namespace_lock held. Once this completes the device 5297 * should no longer have any blocks allocated on it. 5298 */ 5299 if (vd->vdev_islog) { 5300 if (vd->vdev_stat.vs_alloc != 0) 5301 error = spa_offline_log(spa); 5302 } else { 5303 error = SET_ERROR(ENOTSUP); 5304 } 5305 5306 if (error) 5307 return (error); 5308 5309 /* 5310 * The evacuation succeeded. Remove any remaining MOS metadata 5311 * associated with this vdev, and wait for these changes to sync. 5312 */ 5313 ASSERT0(vd->vdev_stat.vs_alloc); 5314 txg = spa_vdev_config_enter(spa); 5315 vd->vdev_removing = B_TRUE; 5316 vdev_dirty(vd, 0, NULL, txg); 5317 vdev_config_dirty(vd); 5318 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5319 5320 return (0); 5321} 5322 5323/* 5324 * Complete the removal by cleaning up the namespace. 5325 */ 5326static void 5327spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5328{ 5329 vdev_t *rvd = spa->spa_root_vdev; 5330 uint64_t id = vd->vdev_id; 5331 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5332 5333 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5334 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5335 ASSERT(vd == vd->vdev_top); 5336 5337 /* 5338 * Only remove any devices which are empty. 5339 */ 5340 if (vd->vdev_stat.vs_alloc != 0) 5341 return; 5342 5343 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5344 5345 if (list_link_active(&vd->vdev_state_dirty_node)) 5346 vdev_state_clean(vd); 5347 if (list_link_active(&vd->vdev_config_dirty_node)) 5348 vdev_config_clean(vd); 5349 5350 vdev_free(vd); 5351 5352 if (last_vdev) { 5353 vdev_compact_children(rvd); 5354 } else { 5355 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5356 vdev_add_child(rvd, vd); 5357 } 5358 vdev_config_dirty(rvd); 5359 5360 /* 5361 * Reassess the health of our root vdev. 5362 */ 5363 vdev_reopen(rvd); 5364} 5365 5366/* 5367 * Remove a device from the pool - 5368 * 5369 * Removing a device from the vdev namespace requires several steps 5370 * and can take a significant amount of time. As a result we use 5371 * the spa_vdev_config_[enter/exit] functions which allow us to 5372 * grab and release the spa_config_lock while still holding the namespace 5373 * lock. During each step the configuration is synced out. 5374 * 5375 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5376 * devices. 5377 */ 5378int 5379spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5380{ 5381 vdev_t *vd; 5382 metaslab_group_t *mg; 5383 nvlist_t **spares, **l2cache, *nv; 5384 uint64_t txg = 0; 5385 uint_t nspares, nl2cache; 5386 int error = 0; 5387 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5388 5389 ASSERT(spa_writeable(spa)); 5390 5391 if (!locked) 5392 txg = spa_vdev_enter(spa); 5393 5394 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5395 5396 if (spa->spa_spares.sav_vdevs != NULL && 5397 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5398 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5399 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5400 /* 5401 * Only remove the hot spare if it's not currently in use 5402 * in this pool. 5403 */ 5404 if (vd == NULL || unspare) { 5405 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5406 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5407 spa_load_spares(spa); 5408 spa->spa_spares.sav_sync = B_TRUE; 5409 } else { 5410 error = SET_ERROR(EBUSY); 5411 } 5412 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5413 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5414 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5415 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5416 /* 5417 * Cache devices can always be removed. 5418 */ 5419 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5420 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5421 spa_load_l2cache(spa); 5422 spa->spa_l2cache.sav_sync = B_TRUE; 5423 } else if (vd != NULL && vd->vdev_islog) { 5424 ASSERT(!locked); 5425 ASSERT(vd == vd->vdev_top); 5426 5427 /* 5428 * XXX - Once we have bp-rewrite this should 5429 * become the common case. 5430 */ 5431 5432 mg = vd->vdev_mg; 5433 5434 /* 5435 * Stop allocating from this vdev. 5436 */ 5437 metaslab_group_passivate(mg); 5438 5439 /* 5440 * Wait for the youngest allocations and frees to sync, 5441 * and then wait for the deferral of those frees to finish. 5442 */ 5443 spa_vdev_config_exit(spa, NULL, 5444 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5445 5446 /* 5447 * Attempt to evacuate the vdev. 5448 */ 5449 error = spa_vdev_remove_evacuate(spa, vd); 5450 5451 txg = spa_vdev_config_enter(spa); 5452 5453 /* 5454 * If we couldn't evacuate the vdev, unwind. 5455 */ 5456 if (error) { 5457 metaslab_group_activate(mg); 5458 return (spa_vdev_exit(spa, NULL, txg, error)); 5459 } 5460 5461 /* 5462 * Clean up the vdev namespace. 5463 */ 5464 spa_vdev_remove_from_namespace(spa, vd); 5465 5466 } else if (vd != NULL) { 5467 /* 5468 * Normal vdevs cannot be removed (yet). 5469 */ 5470 error = SET_ERROR(ENOTSUP); 5471 } else { 5472 /* 5473 * There is no vdev of any kind with the specified guid. 5474 */ 5475 error = SET_ERROR(ENOENT); 5476 } 5477 5478 if (!locked) 5479 return (spa_vdev_exit(spa, NULL, txg, error)); 5480 5481 return (error); 5482} 5483 5484/* 5485 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5486 * currently spared, so we can detach it. 5487 */ 5488static vdev_t * 5489spa_vdev_resilver_done_hunt(vdev_t *vd) 5490{ 5491 vdev_t *newvd, *oldvd; 5492 5493 for (int c = 0; c < vd->vdev_children; c++) { 5494 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5495 if (oldvd != NULL) 5496 return (oldvd); 5497 } 5498 5499 /* 5500 * Check for a completed replacement. We always consider the first 5501 * vdev in the list to be the oldest vdev, and the last one to be 5502 * the newest (see spa_vdev_attach() for how that works). In 5503 * the case where the newest vdev is faulted, we will not automatically 5504 * remove it after a resilver completes. This is OK as it will require 5505 * user intervention to determine which disk the admin wishes to keep. 5506 */ 5507 if (vd->vdev_ops == &vdev_replacing_ops) { 5508 ASSERT(vd->vdev_children > 1); 5509 5510 newvd = vd->vdev_child[vd->vdev_children - 1]; 5511 oldvd = vd->vdev_child[0]; 5512 5513 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5514 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5515 !vdev_dtl_required(oldvd)) 5516 return (oldvd); 5517 } 5518 5519 /* 5520 * Check for a completed resilver with the 'unspare' flag set. 5521 */ 5522 if (vd->vdev_ops == &vdev_spare_ops) { 5523 vdev_t *first = vd->vdev_child[0]; 5524 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5525 5526 if (last->vdev_unspare) { 5527 oldvd = first; 5528 newvd = last; 5529 } else if (first->vdev_unspare) { 5530 oldvd = last; 5531 newvd = first; 5532 } else { 5533 oldvd = NULL; 5534 } 5535 5536 if (oldvd != NULL && 5537 vdev_dtl_empty(newvd, DTL_MISSING) && 5538 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5539 !vdev_dtl_required(oldvd)) 5540 return (oldvd); 5541 5542 /* 5543 * If there are more than two spares attached to a disk, 5544 * and those spares are not required, then we want to 5545 * attempt to free them up now so that they can be used 5546 * by other pools. Once we're back down to a single 5547 * disk+spare, we stop removing them. 5548 */ 5549 if (vd->vdev_children > 2) { 5550 newvd = vd->vdev_child[1]; 5551 5552 if (newvd->vdev_isspare && last->vdev_isspare && 5553 vdev_dtl_empty(last, DTL_MISSING) && 5554 vdev_dtl_empty(last, DTL_OUTAGE) && 5555 !vdev_dtl_required(newvd)) 5556 return (newvd); 5557 } 5558 } 5559 5560 return (NULL); 5561} 5562 5563static void 5564spa_vdev_resilver_done(spa_t *spa) 5565{ 5566 vdev_t *vd, *pvd, *ppvd; 5567 uint64_t guid, sguid, pguid, ppguid; 5568 5569 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5570 5571 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5572 pvd = vd->vdev_parent; 5573 ppvd = pvd->vdev_parent; 5574 guid = vd->vdev_guid; 5575 pguid = pvd->vdev_guid; 5576 ppguid = ppvd->vdev_guid; 5577 sguid = 0; 5578 /* 5579 * If we have just finished replacing a hot spared device, then 5580 * we need to detach the parent's first child (the original hot 5581 * spare) as well. 5582 */ 5583 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5584 ppvd->vdev_children == 2) { 5585 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5586 sguid = ppvd->vdev_child[1]->vdev_guid; 5587 } 5588 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5589 5590 spa_config_exit(spa, SCL_ALL, FTAG); 5591 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5592 return; 5593 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5594 return; 5595 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5596 } 5597 5598 spa_config_exit(spa, SCL_ALL, FTAG); 5599} 5600 5601/* 5602 * Update the stored path or FRU for this vdev. 5603 */ 5604int 5605spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5606 boolean_t ispath) 5607{ 5608 vdev_t *vd; 5609 boolean_t sync = B_FALSE; 5610 5611 ASSERT(spa_writeable(spa)); 5612 5613 spa_vdev_state_enter(spa, SCL_ALL); 5614 5615 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5616 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5617 5618 if (!vd->vdev_ops->vdev_op_leaf) 5619 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5620 5621 if (ispath) { 5622 if (strcmp(value, vd->vdev_path) != 0) { 5623 spa_strfree(vd->vdev_path); 5624 vd->vdev_path = spa_strdup(value); 5625 sync = B_TRUE; 5626 } 5627 } else { 5628 if (vd->vdev_fru == NULL) { 5629 vd->vdev_fru = spa_strdup(value); 5630 sync = B_TRUE; 5631 } else if (strcmp(value, vd->vdev_fru) != 0) { 5632 spa_strfree(vd->vdev_fru); 5633 vd->vdev_fru = spa_strdup(value); 5634 sync = B_TRUE; 5635 } 5636 } 5637 5638 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5639} 5640 5641int 5642spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5643{ 5644 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5645} 5646 5647int 5648spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5649{ 5650 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5651} 5652 5653/* 5654 * ========================================================================== 5655 * SPA Scanning 5656 * ========================================================================== 5657 */ 5658 5659int 5660spa_scan_stop(spa_t *spa) 5661{ 5662 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5663 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5664 return (SET_ERROR(EBUSY)); 5665 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5666} 5667 5668int 5669spa_scan(spa_t *spa, pool_scan_func_t func) 5670{ 5671 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5672 5673 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5674 return (SET_ERROR(ENOTSUP)); 5675 5676 /* 5677 * If a resilver was requested, but there is no DTL on a 5678 * writeable leaf device, we have nothing to do. 5679 */ 5680 if (func == POOL_SCAN_RESILVER && 5681 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5682 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5683 return (0); 5684 } 5685 5686 return (dsl_scan(spa->spa_dsl_pool, func)); 5687} 5688 5689/* 5690 * ========================================================================== 5691 * SPA async task processing 5692 * ========================================================================== 5693 */ 5694 5695static void 5696spa_async_remove(spa_t *spa, vdev_t *vd) 5697{ 5698 if (vd->vdev_remove_wanted) { 5699 vd->vdev_remove_wanted = B_FALSE; 5700 vd->vdev_delayed_close = B_FALSE; 5701 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5702 5703 /* 5704 * We want to clear the stats, but we don't want to do a full 5705 * vdev_clear() as that will cause us to throw away 5706 * degraded/faulted state as well as attempt to reopen the 5707 * device, all of which is a waste. 5708 */ 5709 vd->vdev_stat.vs_read_errors = 0; 5710 vd->vdev_stat.vs_write_errors = 0; 5711 vd->vdev_stat.vs_checksum_errors = 0; 5712 5713 vdev_state_dirty(vd->vdev_top); 5714 } 5715 5716 for (int c = 0; c < vd->vdev_children; c++) 5717 spa_async_remove(spa, vd->vdev_child[c]); 5718} 5719 5720static void 5721spa_async_probe(spa_t *spa, vdev_t *vd) 5722{ 5723 if (vd->vdev_probe_wanted) { 5724 vd->vdev_probe_wanted = B_FALSE; 5725 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5726 } 5727 5728 for (int c = 0; c < vd->vdev_children; c++) 5729 spa_async_probe(spa, vd->vdev_child[c]); 5730} 5731 5732static void 5733spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5734{ 5735 sysevent_id_t eid; 5736 nvlist_t *attr; 5737 char *physpath; 5738 5739 if (!spa->spa_autoexpand) 5740 return; 5741 5742 for (int c = 0; c < vd->vdev_children; c++) { 5743 vdev_t *cvd = vd->vdev_child[c]; 5744 spa_async_autoexpand(spa, cvd); 5745 } 5746 5747 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5748 return; 5749 5750 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5751 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5752 5753 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5754 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5755 5756 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5757 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5758 5759 nvlist_free(attr); 5760 kmem_free(physpath, MAXPATHLEN); 5761} 5762 5763static void 5764spa_async_thread(void *arg) 5765{ 5766 spa_t *spa = arg; 5767 int tasks; 5768 5769 ASSERT(spa->spa_sync_on); 5770 5771 mutex_enter(&spa->spa_async_lock); 5772 tasks = spa->spa_async_tasks; 5773 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5774 mutex_exit(&spa->spa_async_lock); 5775 5776 /* 5777 * See if the config needs to be updated. 5778 */ 5779 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5780 uint64_t old_space, new_space; 5781 5782 mutex_enter(&spa_namespace_lock); 5783 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5784 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5785 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5786 mutex_exit(&spa_namespace_lock); 5787 5788 /* 5789 * If the pool grew as a result of the config update, 5790 * then log an internal history event. 5791 */ 5792 if (new_space != old_space) { 5793 spa_history_log_internal(spa, "vdev online", NULL, 5794 "pool '%s' size: %llu(+%llu)", 5795 spa_name(spa), new_space, new_space - old_space); 5796 } 5797 } 5798 5799 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5800 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5801 spa_async_autoexpand(spa, spa->spa_root_vdev); 5802 spa_config_exit(spa, SCL_CONFIG, FTAG); 5803 } 5804 5805 /* 5806 * See if any devices need to be probed. 5807 */ 5808 if (tasks & SPA_ASYNC_PROBE) { 5809 spa_vdev_state_enter(spa, SCL_NONE); 5810 spa_async_probe(spa, spa->spa_root_vdev); 5811 (void) spa_vdev_state_exit(spa, NULL, 0); 5812 } 5813 5814 /* 5815 * If any devices are done replacing, detach them. 5816 */ 5817 if (tasks & SPA_ASYNC_RESILVER_DONE) 5818 spa_vdev_resilver_done(spa); 5819 5820 /* 5821 * Kick off a resilver. 5822 */ 5823 if (tasks & SPA_ASYNC_RESILVER) 5824 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5825 5826 /* 5827 * Let the world know that we're done. 5828 */ 5829 mutex_enter(&spa->spa_async_lock); 5830 spa->spa_async_thread = NULL; 5831 cv_broadcast(&spa->spa_async_cv); 5832 mutex_exit(&spa->spa_async_lock); 5833 thread_exit(); 5834} 5835 5836static void 5837spa_async_thread_vd(void *arg) 5838{ 5839 spa_t *spa = arg; 5840 int tasks; 5841 5842 ASSERT(spa->spa_sync_on); 5843 5844 mutex_enter(&spa->spa_async_lock); 5845 tasks = spa->spa_async_tasks; 5846retry: 5847 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 5848 mutex_exit(&spa->spa_async_lock); 5849 5850 /* 5851 * See if any devices need to be marked REMOVED. 5852 */ 5853 if (tasks & SPA_ASYNC_REMOVE) { 5854 spa_vdev_state_enter(spa, SCL_NONE); 5855 spa_async_remove(spa, spa->spa_root_vdev); 5856 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5857 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5858 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5859 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5860 (void) spa_vdev_state_exit(spa, NULL, 0); 5861 } 5862 5863 /* 5864 * Let the world know that we're done. 5865 */ 5866 mutex_enter(&spa->spa_async_lock); 5867 tasks = spa->spa_async_tasks; 5868 if ((tasks & SPA_ASYNC_REMOVE) != 0) 5869 goto retry; 5870 spa->spa_async_thread_vd = NULL; 5871 cv_broadcast(&spa->spa_async_cv); 5872 mutex_exit(&spa->spa_async_lock); 5873 thread_exit(); 5874} 5875 5876void 5877spa_async_suspend(spa_t *spa) 5878{ 5879 mutex_enter(&spa->spa_async_lock); 5880 spa->spa_async_suspended++; 5881 while (spa->spa_async_thread != NULL && 5882 spa->spa_async_thread_vd != NULL) 5883 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5884 mutex_exit(&spa->spa_async_lock); 5885} 5886 5887void 5888spa_async_resume(spa_t *spa) 5889{ 5890 mutex_enter(&spa->spa_async_lock); 5891 ASSERT(spa->spa_async_suspended != 0); 5892 spa->spa_async_suspended--; 5893 mutex_exit(&spa->spa_async_lock); 5894} 5895 5896static boolean_t 5897spa_async_tasks_pending(spa_t *spa) 5898{ 5899 uint_t non_config_tasks; 5900 uint_t config_task; 5901 boolean_t config_task_suspended; 5902 5903 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 5904 SPA_ASYNC_REMOVE); 5905 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 5906 if (spa->spa_ccw_fail_time == 0) { 5907 config_task_suspended = B_FALSE; 5908 } else { 5909 config_task_suspended = 5910 (gethrtime() - spa->spa_ccw_fail_time) < 5911 (zfs_ccw_retry_interval * NANOSEC); 5912 } 5913 5914 return (non_config_tasks || (config_task && !config_task_suspended)); 5915} 5916 5917static void 5918spa_async_dispatch(spa_t *spa) 5919{ 5920 mutex_enter(&spa->spa_async_lock); 5921 if (spa_async_tasks_pending(spa) && 5922 !spa->spa_async_suspended && 5923 spa->spa_async_thread == NULL && 5924 rootdir != NULL) 5925 spa->spa_async_thread = thread_create(NULL, 0, 5926 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5927 mutex_exit(&spa->spa_async_lock); 5928} 5929 5930static void 5931spa_async_dispatch_vd(spa_t *spa) 5932{ 5933 mutex_enter(&spa->spa_async_lock); 5934 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 5935 !spa->spa_async_suspended && 5936 spa->spa_async_thread_vd == NULL && 5937 rootdir != NULL) 5938 spa->spa_async_thread_vd = thread_create(NULL, 0, 5939 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 5940 mutex_exit(&spa->spa_async_lock); 5941} 5942 5943void 5944spa_async_request(spa_t *spa, int task) 5945{ 5946 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5947 mutex_enter(&spa->spa_async_lock); 5948 spa->spa_async_tasks |= task; 5949 mutex_exit(&spa->spa_async_lock); 5950 spa_async_dispatch_vd(spa); 5951} 5952 5953/* 5954 * ========================================================================== 5955 * SPA syncing routines 5956 * ========================================================================== 5957 */ 5958 5959static int 5960bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5961{ 5962 bpobj_t *bpo = arg; 5963 bpobj_enqueue(bpo, bp, tx); 5964 return (0); 5965} 5966 5967static int 5968spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5969{ 5970 zio_t *zio = arg; 5971 5972 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5973 BP_GET_PSIZE(bp), zio->io_flags)); 5974 return (0); 5975} 5976 5977static void 5978spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5979{ 5980 char *packed = NULL; 5981 size_t bufsize; 5982 size_t nvsize = 0; 5983 dmu_buf_t *db; 5984 5985 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5986 5987 /* 5988 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 5989 * information. This avoids the dbuf_will_dirty() path and 5990 * saves us a pre-read to get data we don't actually care about. 5991 */ 5992 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 5993 packed = kmem_alloc(bufsize, KM_SLEEP); 5994 5995 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5996 KM_SLEEP) == 0); 5997 bzero(packed + nvsize, bufsize - nvsize); 5998 5999 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6000 6001 kmem_free(packed, bufsize); 6002 6003 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6004 dmu_buf_will_dirty(db, tx); 6005 *(uint64_t *)db->db_data = nvsize; 6006 dmu_buf_rele(db, FTAG); 6007} 6008 6009static void 6010spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6011 const char *config, const char *entry) 6012{ 6013 nvlist_t *nvroot; 6014 nvlist_t **list; 6015 int i; 6016 6017 if (!sav->sav_sync) 6018 return; 6019 6020 /* 6021 * Update the MOS nvlist describing the list of available devices. 6022 * spa_validate_aux() will have already made sure this nvlist is 6023 * valid and the vdevs are labeled appropriately. 6024 */ 6025 if (sav->sav_object == 0) { 6026 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6027 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6028 sizeof (uint64_t), tx); 6029 VERIFY(zap_update(spa->spa_meta_objset, 6030 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6031 &sav->sav_object, tx) == 0); 6032 } 6033 6034 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6035 if (sav->sav_count == 0) { 6036 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6037 } else { 6038 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6039 for (i = 0; i < sav->sav_count; i++) 6040 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6041 B_FALSE, VDEV_CONFIG_L2CACHE); 6042 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6043 sav->sav_count) == 0); 6044 for (i = 0; i < sav->sav_count; i++) 6045 nvlist_free(list[i]); 6046 kmem_free(list, sav->sav_count * sizeof (void *)); 6047 } 6048 6049 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6050 nvlist_free(nvroot); 6051 6052 sav->sav_sync = B_FALSE; 6053} 6054 6055static void 6056spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6057{ 6058 nvlist_t *config; 6059 6060 if (list_is_empty(&spa->spa_config_dirty_list)) 6061 return; 6062 6063 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6064 6065 config = spa_config_generate(spa, spa->spa_root_vdev, 6066 dmu_tx_get_txg(tx), B_FALSE); 6067 6068 /* 6069 * If we're upgrading the spa version then make sure that 6070 * the config object gets updated with the correct version. 6071 */ 6072 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6073 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6074 spa->spa_uberblock.ub_version); 6075 6076 spa_config_exit(spa, SCL_STATE, FTAG); 6077 6078 if (spa->spa_config_syncing) 6079 nvlist_free(spa->spa_config_syncing); 6080 spa->spa_config_syncing = config; 6081 6082 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6083} 6084 6085static void 6086spa_sync_version(void *arg, dmu_tx_t *tx) 6087{ 6088 uint64_t *versionp = arg; 6089 uint64_t version = *versionp; 6090 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6091 6092 /* 6093 * Setting the version is special cased when first creating the pool. 6094 */ 6095 ASSERT(tx->tx_txg != TXG_INITIAL); 6096 6097 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6098 ASSERT(version >= spa_version(spa)); 6099 6100 spa->spa_uberblock.ub_version = version; 6101 vdev_config_dirty(spa->spa_root_vdev); 6102 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6103} 6104 6105/* 6106 * Set zpool properties. 6107 */ 6108static void 6109spa_sync_props(void *arg, dmu_tx_t *tx) 6110{ 6111 nvlist_t *nvp = arg; 6112 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6113 objset_t *mos = spa->spa_meta_objset; 6114 nvpair_t *elem = NULL; 6115 6116 mutex_enter(&spa->spa_props_lock); 6117 6118 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6119 uint64_t intval; 6120 char *strval, *fname; 6121 zpool_prop_t prop; 6122 const char *propname; 6123 zprop_type_t proptype; 6124 zfeature_info_t *feature; 6125 6126 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6127 case ZPROP_INVAL: 6128 /* 6129 * We checked this earlier in spa_prop_validate(). 6130 */ 6131 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6132 6133 fname = strchr(nvpair_name(elem), '@') + 1; 6134 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature)); 6135 6136 spa_feature_enable(spa, feature, tx); 6137 spa_history_log_internal(spa, "set", tx, 6138 "%s=enabled", nvpair_name(elem)); 6139 break; 6140 6141 case ZPOOL_PROP_VERSION: 6142 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6143 /* 6144 * The version is synced seperatly before other 6145 * properties and should be correct by now. 6146 */ 6147 ASSERT3U(spa_version(spa), >=, intval); 6148 break; 6149 6150 case ZPOOL_PROP_ALTROOT: 6151 /* 6152 * 'altroot' is a non-persistent property. It should 6153 * have been set temporarily at creation or import time. 6154 */ 6155 ASSERT(spa->spa_root != NULL); 6156 break; 6157 6158 case ZPOOL_PROP_READONLY: 6159 case ZPOOL_PROP_CACHEFILE: 6160 /* 6161 * 'readonly' and 'cachefile' are also non-persisitent 6162 * properties. 6163 */ 6164 break; 6165 case ZPOOL_PROP_COMMENT: 6166 VERIFY(nvpair_value_string(elem, &strval) == 0); 6167 if (spa->spa_comment != NULL) 6168 spa_strfree(spa->spa_comment); 6169 spa->spa_comment = spa_strdup(strval); 6170 /* 6171 * We need to dirty the configuration on all the vdevs 6172 * so that their labels get updated. It's unnecessary 6173 * to do this for pool creation since the vdev's 6174 * configuratoin has already been dirtied. 6175 */ 6176 if (tx->tx_txg != TXG_INITIAL) 6177 vdev_config_dirty(spa->spa_root_vdev); 6178 spa_history_log_internal(spa, "set", tx, 6179 "%s=%s", nvpair_name(elem), strval); 6180 break; 6181 default: 6182 /* 6183 * Set pool property values in the poolprops mos object. 6184 */ 6185 if (spa->spa_pool_props_object == 0) { 6186 spa->spa_pool_props_object = 6187 zap_create_link(mos, DMU_OT_POOL_PROPS, 6188 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6189 tx); 6190 } 6191 6192 /* normalize the property name */ 6193 propname = zpool_prop_to_name(prop); 6194 proptype = zpool_prop_get_type(prop); 6195 6196 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6197 ASSERT(proptype == PROP_TYPE_STRING); 6198 VERIFY(nvpair_value_string(elem, &strval) == 0); 6199 VERIFY(zap_update(mos, 6200 spa->spa_pool_props_object, propname, 6201 1, strlen(strval) + 1, strval, tx) == 0); 6202 spa_history_log_internal(spa, "set", tx, 6203 "%s=%s", nvpair_name(elem), strval); 6204 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6205 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6206 6207 if (proptype == PROP_TYPE_INDEX) { 6208 const char *unused; 6209 VERIFY(zpool_prop_index_to_string( 6210 prop, intval, &unused) == 0); 6211 } 6212 VERIFY(zap_update(mos, 6213 spa->spa_pool_props_object, propname, 6214 8, 1, &intval, tx) == 0); 6215 spa_history_log_internal(spa, "set", tx, 6216 "%s=%lld", nvpair_name(elem), intval); 6217 } else { 6218 ASSERT(0); /* not allowed */ 6219 } 6220 6221 switch (prop) { 6222 case ZPOOL_PROP_DELEGATION: 6223 spa->spa_delegation = intval; 6224 break; 6225 case ZPOOL_PROP_BOOTFS: 6226 spa->spa_bootfs = intval; 6227 break; 6228 case ZPOOL_PROP_FAILUREMODE: 6229 spa->spa_failmode = intval; 6230 break; 6231 case ZPOOL_PROP_AUTOEXPAND: 6232 spa->spa_autoexpand = intval; 6233 if (tx->tx_txg != TXG_INITIAL) 6234 spa_async_request(spa, 6235 SPA_ASYNC_AUTOEXPAND); 6236 break; 6237 case ZPOOL_PROP_DEDUPDITTO: 6238 spa->spa_dedup_ditto = intval; 6239 break; 6240 default: 6241 break; 6242 } 6243 } 6244 6245 } 6246 6247 mutex_exit(&spa->spa_props_lock); 6248} 6249 6250/* 6251 * Perform one-time upgrade on-disk changes. spa_version() does not 6252 * reflect the new version this txg, so there must be no changes this 6253 * txg to anything that the upgrade code depends on after it executes. 6254 * Therefore this must be called after dsl_pool_sync() does the sync 6255 * tasks. 6256 */ 6257static void 6258spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6259{ 6260 dsl_pool_t *dp = spa->spa_dsl_pool; 6261 6262 ASSERT(spa->spa_sync_pass == 1); 6263 6264 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6265 6266 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6267 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6268 dsl_pool_create_origin(dp, tx); 6269 6270 /* Keeping the origin open increases spa_minref */ 6271 spa->spa_minref += 3; 6272 } 6273 6274 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6275 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6276 dsl_pool_upgrade_clones(dp, tx); 6277 } 6278 6279 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6280 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6281 dsl_pool_upgrade_dir_clones(dp, tx); 6282 6283 /* Keeping the freedir open increases spa_minref */ 6284 spa->spa_minref += 3; 6285 } 6286 6287 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6288 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6289 spa_feature_create_zap_objects(spa, tx); 6290 } 6291 rrw_exit(&dp->dp_config_rwlock, FTAG); 6292} 6293 6294/* 6295 * Sync the specified transaction group. New blocks may be dirtied as 6296 * part of the process, so we iterate until it converges. 6297 */ 6298void 6299spa_sync(spa_t *spa, uint64_t txg) 6300{ 6301 dsl_pool_t *dp = spa->spa_dsl_pool; 6302 objset_t *mos = spa->spa_meta_objset; 6303 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj; 6304 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6305 vdev_t *rvd = spa->spa_root_vdev; 6306 vdev_t *vd; 6307 dmu_tx_t *tx; 6308 int error; 6309 6310 VERIFY(spa_writeable(spa)); 6311 6312 /* 6313 * Lock out configuration changes. 6314 */ 6315 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6316 6317 spa->spa_syncing_txg = txg; 6318 spa->spa_sync_pass = 0; 6319 6320 /* 6321 * If there are any pending vdev state changes, convert them 6322 * into config changes that go out with this transaction group. 6323 */ 6324 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6325 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6326 /* 6327 * We need the write lock here because, for aux vdevs, 6328 * calling vdev_config_dirty() modifies sav_config. 6329 * This is ugly and will become unnecessary when we 6330 * eliminate the aux vdev wart by integrating all vdevs 6331 * into the root vdev tree. 6332 */ 6333 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6334 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6335 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6336 vdev_state_clean(vd); 6337 vdev_config_dirty(vd); 6338 } 6339 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6340 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6341 } 6342 spa_config_exit(spa, SCL_STATE, FTAG); 6343 6344 tx = dmu_tx_create_assigned(dp, txg); 6345 6346 spa->spa_sync_starttime = gethrtime(); 6347#ifdef illumos 6348 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6349 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6350#else /* FreeBSD */ 6351#ifdef _KERNEL 6352 callout_reset(&spa->spa_deadman_cycid, 6353 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6354#endif 6355#endif 6356 6357 /* 6358 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6359 * set spa_deflate if we have no raid-z vdevs. 6360 */ 6361 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6362 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6363 int i; 6364 6365 for (i = 0; i < rvd->vdev_children; i++) { 6366 vd = rvd->vdev_child[i]; 6367 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6368 break; 6369 } 6370 if (i == rvd->vdev_children) { 6371 spa->spa_deflate = TRUE; 6372 VERIFY(0 == zap_add(spa->spa_meta_objset, 6373 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6374 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6375 } 6376 } 6377 6378 /* 6379 * If anything has changed in this txg, or if someone is waiting 6380 * for this txg to sync (eg, spa_vdev_remove()), push the 6381 * deferred frees from the previous txg. If not, leave them 6382 * alone so that we don't generate work on an otherwise idle 6383 * system. 6384 */ 6385 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6386 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6387 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6388 ((dsl_scan_active(dp->dp_scan) || 6389 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6390 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6391 VERIFY3U(bpobj_iterate(defer_bpo, 6392 spa_free_sync_cb, zio, tx), ==, 0); 6393 VERIFY0(zio_wait(zio)); 6394 } 6395 6396 /* 6397 * Iterate to convergence. 6398 */ 6399 do { 6400 int pass = ++spa->spa_sync_pass; 6401 6402 spa_sync_config_object(spa, tx); 6403 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6404 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6405 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6406 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6407 spa_errlog_sync(spa, txg); 6408 dsl_pool_sync(dp, txg); 6409 6410 if (pass < zfs_sync_pass_deferred_free) { 6411 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6412 bplist_iterate(free_bpl, spa_free_sync_cb, 6413 zio, tx); 6414 VERIFY(zio_wait(zio) == 0); 6415 } else { 6416 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6417 defer_bpo, tx); 6418 } 6419 6420 ddt_sync(spa, txg); 6421 dsl_scan_sync(dp, tx); 6422 6423 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6424 vdev_sync(vd, txg); 6425 6426 if (pass == 1) 6427 spa_sync_upgrades(spa, tx); 6428 6429 } while (dmu_objset_is_dirty(mos, txg)); 6430 6431 /* 6432 * Rewrite the vdev configuration (which includes the uberblock) 6433 * to commit the transaction group. 6434 * 6435 * If there are no dirty vdevs, we sync the uberblock to a few 6436 * random top-level vdevs that are known to be visible in the 6437 * config cache (see spa_vdev_add() for a complete description). 6438 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6439 */ 6440 for (;;) { 6441 /* 6442 * We hold SCL_STATE to prevent vdev open/close/etc. 6443 * while we're attempting to write the vdev labels. 6444 */ 6445 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6446 6447 if (list_is_empty(&spa->spa_config_dirty_list)) { 6448 vdev_t *svd[SPA_DVAS_PER_BP]; 6449 int svdcount = 0; 6450 int children = rvd->vdev_children; 6451 int c0 = spa_get_random(children); 6452 6453 for (int c = 0; c < children; c++) { 6454 vd = rvd->vdev_child[(c0 + c) % children]; 6455 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6456 continue; 6457 svd[svdcount++] = vd; 6458 if (svdcount == SPA_DVAS_PER_BP) 6459 break; 6460 } 6461 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6462 if (error != 0) 6463 error = vdev_config_sync(svd, svdcount, txg, 6464 B_TRUE); 6465 } else { 6466 error = vdev_config_sync(rvd->vdev_child, 6467 rvd->vdev_children, txg, B_FALSE); 6468 if (error != 0) 6469 error = vdev_config_sync(rvd->vdev_child, 6470 rvd->vdev_children, txg, B_TRUE); 6471 } 6472 6473 if (error == 0) 6474 spa->spa_last_synced_guid = rvd->vdev_guid; 6475 6476 spa_config_exit(spa, SCL_STATE, FTAG); 6477 6478 if (error == 0) 6479 break; 6480 zio_suspend(spa, NULL); 6481 zio_resume_wait(spa); 6482 } 6483 dmu_tx_commit(tx); 6484 6485#ifdef illumos 6486 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6487#else /* FreeBSD */ 6488#ifdef _KERNEL 6489 callout_drain(&spa->spa_deadman_cycid); 6490#endif 6491#endif 6492 6493 /* 6494 * Clear the dirty config list. 6495 */ 6496 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6497 vdev_config_clean(vd); 6498 6499 /* 6500 * Now that the new config has synced transactionally, 6501 * let it become visible to the config cache. 6502 */ 6503 if (spa->spa_config_syncing != NULL) { 6504 spa_config_set(spa, spa->spa_config_syncing); 6505 spa->spa_config_txg = txg; 6506 spa->spa_config_syncing = NULL; 6507 } 6508 6509 spa->spa_ubsync = spa->spa_uberblock; 6510 6511 dsl_pool_sync_done(dp, txg); 6512 6513 /* 6514 * Update usable space statistics. 6515 */ 6516 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6517 vdev_sync_done(vd, txg); 6518 6519 spa_update_dspace(spa); 6520 6521 /* 6522 * It had better be the case that we didn't dirty anything 6523 * since vdev_config_sync(). 6524 */ 6525 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6526 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6527 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6528 6529 spa->spa_sync_pass = 0; 6530 6531 spa_config_exit(spa, SCL_CONFIG, FTAG); 6532 6533 spa_handle_ignored_writes(spa); 6534 6535 /* 6536 * If any async tasks have been requested, kick them off. 6537 */ 6538 spa_async_dispatch(spa); 6539 spa_async_dispatch_vd(spa); 6540} 6541 6542/* 6543 * Sync all pools. We don't want to hold the namespace lock across these 6544 * operations, so we take a reference on the spa_t and drop the lock during the 6545 * sync. 6546 */ 6547void 6548spa_sync_allpools(void) 6549{ 6550 spa_t *spa = NULL; 6551 mutex_enter(&spa_namespace_lock); 6552 while ((spa = spa_next(spa)) != NULL) { 6553 if (spa_state(spa) != POOL_STATE_ACTIVE || 6554 !spa_writeable(spa) || spa_suspended(spa)) 6555 continue; 6556 spa_open_ref(spa, FTAG); 6557 mutex_exit(&spa_namespace_lock); 6558 txg_wait_synced(spa_get_dsl(spa), 0); 6559 mutex_enter(&spa_namespace_lock); 6560 spa_close(spa, FTAG); 6561 } 6562 mutex_exit(&spa_namespace_lock); 6563} 6564 6565/* 6566 * ========================================================================== 6567 * Miscellaneous routines 6568 * ========================================================================== 6569 */ 6570 6571/* 6572 * Remove all pools in the system. 6573 */ 6574void 6575spa_evict_all(void) 6576{ 6577 spa_t *spa; 6578 6579 /* 6580 * Remove all cached state. All pools should be closed now, 6581 * so every spa in the AVL tree should be unreferenced. 6582 */ 6583 mutex_enter(&spa_namespace_lock); 6584 while ((spa = spa_next(NULL)) != NULL) { 6585 /* 6586 * Stop async tasks. The async thread may need to detach 6587 * a device that's been replaced, which requires grabbing 6588 * spa_namespace_lock, so we must drop it here. 6589 */ 6590 spa_open_ref(spa, FTAG); 6591 mutex_exit(&spa_namespace_lock); 6592 spa_async_suspend(spa); 6593 mutex_enter(&spa_namespace_lock); 6594 spa_close(spa, FTAG); 6595 6596 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6597 spa_unload(spa); 6598 spa_deactivate(spa); 6599 } 6600 spa_remove(spa); 6601 } 6602 mutex_exit(&spa_namespace_lock); 6603} 6604 6605vdev_t * 6606spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6607{ 6608 vdev_t *vd; 6609 int i; 6610 6611 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6612 return (vd); 6613 6614 if (aux) { 6615 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6616 vd = spa->spa_l2cache.sav_vdevs[i]; 6617 if (vd->vdev_guid == guid) 6618 return (vd); 6619 } 6620 6621 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6622 vd = spa->spa_spares.sav_vdevs[i]; 6623 if (vd->vdev_guid == guid) 6624 return (vd); 6625 } 6626 } 6627 6628 return (NULL); 6629} 6630 6631void 6632spa_upgrade(spa_t *spa, uint64_t version) 6633{ 6634 ASSERT(spa_writeable(spa)); 6635 6636 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6637 6638 /* 6639 * This should only be called for a non-faulted pool, and since a 6640 * future version would result in an unopenable pool, this shouldn't be 6641 * possible. 6642 */ 6643 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6644 ASSERT(version >= spa->spa_uberblock.ub_version); 6645 6646 spa->spa_uberblock.ub_version = version; 6647 vdev_config_dirty(spa->spa_root_vdev); 6648 6649 spa_config_exit(spa, SCL_ALL, FTAG); 6650 6651 txg_wait_synced(spa_get_dsl(spa), 0); 6652} 6653 6654boolean_t 6655spa_has_spare(spa_t *spa, uint64_t guid) 6656{ 6657 int i; 6658 uint64_t spareguid; 6659 spa_aux_vdev_t *sav = &spa->spa_spares; 6660 6661 for (i = 0; i < sav->sav_count; i++) 6662 if (sav->sav_vdevs[i]->vdev_guid == guid) 6663 return (B_TRUE); 6664 6665 for (i = 0; i < sav->sav_npending; i++) { 6666 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6667 &spareguid) == 0 && spareguid == guid) 6668 return (B_TRUE); 6669 } 6670 6671 return (B_FALSE); 6672} 6673 6674/* 6675 * Check if a pool has an active shared spare device. 6676 * Note: reference count of an active spare is 2, as a spare and as a replace 6677 */ 6678static boolean_t 6679spa_has_active_shared_spare(spa_t *spa) 6680{ 6681 int i, refcnt; 6682 uint64_t pool; 6683 spa_aux_vdev_t *sav = &spa->spa_spares; 6684 6685 for (i = 0; i < sav->sav_count; i++) { 6686 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6687 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6688 refcnt > 2) 6689 return (B_TRUE); 6690 } 6691 6692 return (B_FALSE); 6693} 6694 6695/* 6696 * Post a sysevent corresponding to the given event. The 'name' must be one of 6697 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6698 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6699 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6700 * or zdb as real changes. 6701 */ 6702void 6703spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6704{ 6705#ifdef _KERNEL 6706 sysevent_t *ev; 6707 sysevent_attr_list_t *attr = NULL; 6708 sysevent_value_t value; 6709 sysevent_id_t eid; 6710 6711 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6712 SE_SLEEP); 6713 6714 value.value_type = SE_DATA_TYPE_STRING; 6715 value.value.sv_string = spa_name(spa); 6716 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6717 goto done; 6718 6719 value.value_type = SE_DATA_TYPE_UINT64; 6720 value.value.sv_uint64 = spa_guid(spa); 6721 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6722 goto done; 6723 6724 if (vd) { 6725 value.value_type = SE_DATA_TYPE_UINT64; 6726 value.value.sv_uint64 = vd->vdev_guid; 6727 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6728 SE_SLEEP) != 0) 6729 goto done; 6730 6731 if (vd->vdev_path) { 6732 value.value_type = SE_DATA_TYPE_STRING; 6733 value.value.sv_string = vd->vdev_path; 6734 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6735 &value, SE_SLEEP) != 0) 6736 goto done; 6737 } 6738 } 6739 6740 if (sysevent_attach_attributes(ev, attr) != 0) 6741 goto done; 6742 attr = NULL; 6743 6744 (void) log_sysevent(ev, SE_SLEEP, &eid); 6745 6746done: 6747 if (attr) 6748 sysevent_free_attr(attr); 6749 sysevent_free(ev); 6750#endif 6751} 6752