spa.c revision 249047
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) 2012 by Delphix. All rights reserved. 25 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 26 */ 27 28/* 29 * This file contains all the routines used when modifying on-disk SPA state. 30 * This includes opening, importing, destroying, exporting a pool, and syncing a 31 * pool. 32 */ 33 34#include <sys/zfs_context.h> 35#include <sys/fm/fs/zfs.h> 36#include <sys/spa_impl.h> 37#include <sys/zio.h> 38#include <sys/zio_checksum.h> 39#include <sys/dmu.h> 40#include <sys/dmu_tx.h> 41#include <sys/zap.h> 42#include <sys/zil.h> 43#include <sys/ddt.h> 44#include <sys/vdev_impl.h> 45#include <sys/metaslab.h> 46#include <sys/metaslab_impl.h> 47#include <sys/uberblock_impl.h> 48#include <sys/txg.h> 49#include <sys/avl.h> 50#include <sys/dmu_traverse.h> 51#include <sys/dmu_objset.h> 52#include <sys/unique.h> 53#include <sys/dsl_pool.h> 54#include <sys/dsl_dataset.h> 55#include <sys/dsl_dir.h> 56#include <sys/dsl_prop.h> 57#include <sys/dsl_synctask.h> 58#include <sys/fs/zfs.h> 59#include <sys/arc.h> 60#include <sys/callb.h> 61#include <sys/spa_boot.h> 62#include <sys/zfs_ioctl.h> 63#include <sys/dsl_scan.h> 64#include <sys/dmu_send.h> 65#include <sys/dsl_destroy.h> 66#include <sys/dsl_userhold.h> 67#include <sys/zfeature.h> 68#include <sys/zvol.h> 69#include <sys/trim_map.h> 70 71#ifdef _KERNEL 72#include <sys/callb.h> 73#include <sys/cpupart.h> 74#include <sys/zone.h> 75#endif /* _KERNEL */ 76 77#include "zfs_prop.h" 78#include "zfs_comutil.h" 79 80/* Check hostid on import? */ 81static int check_hostid = 1; 82 83SYSCTL_DECL(_vfs_zfs); 84TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 85SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 86 "Check hostid on import?"); 87 88typedef enum zti_modes { 89 zti_mode_fixed, /* value is # of threads (min 1) */ 90 zti_mode_online_percent, /* value is % of online CPUs */ 91 zti_mode_batch, /* cpu-intensive; value is ignored */ 92 zti_mode_null, /* don't create a taskq */ 93 zti_nmodes 94} zti_modes_t; 95 96#define ZTI_FIX(n) { zti_mode_fixed, (n) } 97#define ZTI_PCT(n) { zti_mode_online_percent, (n) } 98#define ZTI_BATCH { zti_mode_batch, 0 } 99#define ZTI_NULL { zti_mode_null, 0 } 100 101#define ZTI_ONE ZTI_FIX(1) 102 103typedef struct zio_taskq_info { 104 enum zti_modes zti_mode; 105 uint_t zti_value; 106} zio_taskq_info_t; 107 108static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 109 "issue", "issue_high", "intr", "intr_high" 110}; 111 112/* 113 * Define the taskq threads for the following I/O types: 114 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL 115 */ 116const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 117 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 119 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, 120 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) }, 121 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL }, 122 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 123 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 124}; 125 126static void spa_sync_version(void *arg, dmu_tx_t *tx); 127static void spa_sync_props(void *arg, dmu_tx_t *tx); 128static boolean_t spa_has_active_shared_spare(spa_t *spa); 129static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 130 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 131 char **ereport); 132static void spa_vdev_resilver_done(spa_t *spa); 133 134uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */ 135#ifdef PSRSET_BIND 136id_t zio_taskq_psrset_bind = PS_NONE; 137#endif 138#ifdef SYSDC 139boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 140#endif 141uint_t zio_taskq_basedc = 80; /* base duty cycle */ 142 143boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 144extern int zfs_sync_pass_deferred_free; 145 146#ifndef illumos 147extern void spa_deadman(void *arg); 148#endif 149 150/* 151 * This (illegal) pool name is used when temporarily importing a spa_t in order 152 * to get the vdev stats associated with the imported devices. 153 */ 154#define TRYIMPORT_NAME "$import" 155 156/* 157 * ========================================================================== 158 * SPA properties routines 159 * ========================================================================== 160 */ 161 162/* 163 * Add a (source=src, propname=propval) list to an nvlist. 164 */ 165static void 166spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 167 uint64_t intval, zprop_source_t src) 168{ 169 const char *propname = zpool_prop_to_name(prop); 170 nvlist_t *propval; 171 172 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 173 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 174 175 if (strval != NULL) 176 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 177 else 178 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 179 180 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 181 nvlist_free(propval); 182} 183 184/* 185 * Get property values from the spa configuration. 186 */ 187static void 188spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 189{ 190 vdev_t *rvd = spa->spa_root_vdev; 191 dsl_pool_t *pool = spa->spa_dsl_pool; 192 uint64_t size; 193 uint64_t alloc; 194 uint64_t space; 195 uint64_t cap, version; 196 zprop_source_t src = ZPROP_SRC_NONE; 197 spa_config_dirent_t *dp; 198 199 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 200 201 if (rvd != NULL) { 202 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 203 size = metaslab_class_get_space(spa_normal_class(spa)); 204 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 205 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 206 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 207 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 208 size - alloc, src); 209 210 space = 0; 211 for (int c = 0; c < rvd->vdev_children; c++) { 212 vdev_t *tvd = rvd->vdev_child[c]; 213 space += tvd->vdev_max_asize - tvd->vdev_asize; 214 } 215 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space, 216 src); 217 218 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 219 (spa_mode(spa) == FREAD), src); 220 221 cap = (size == 0) ? 0 : (alloc * 100 / size); 222 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 223 224 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 225 ddt_get_pool_dedup_ratio(spa), src); 226 227 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 228 rvd->vdev_state, src); 229 230 version = spa_version(spa); 231 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 232 src = ZPROP_SRC_DEFAULT; 233 else 234 src = ZPROP_SRC_LOCAL; 235 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 236 } 237 238 if (pool != NULL) { 239 dsl_dir_t *freedir = pool->dp_free_dir; 240 241 /* 242 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 243 * when opening pools before this version freedir will be NULL. 244 */ 245 if (freedir != NULL) { 246 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 247 freedir->dd_phys->dd_used_bytes, src); 248 } else { 249 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 250 NULL, 0, src); 251 } 252 } 253 254 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 255 256 if (spa->spa_comment != NULL) { 257 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 258 0, ZPROP_SRC_LOCAL); 259 } 260 261 if (spa->spa_root != NULL) 262 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 263 0, ZPROP_SRC_LOCAL); 264 265 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 266 if (dp->scd_path == NULL) { 267 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 268 "none", 0, ZPROP_SRC_LOCAL); 269 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 270 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 271 dp->scd_path, 0, ZPROP_SRC_LOCAL); 272 } 273 } 274} 275 276/* 277 * Get zpool property values. 278 */ 279int 280spa_prop_get(spa_t *spa, nvlist_t **nvp) 281{ 282 objset_t *mos = spa->spa_meta_objset; 283 zap_cursor_t zc; 284 zap_attribute_t za; 285 int err; 286 287 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 288 289 mutex_enter(&spa->spa_props_lock); 290 291 /* 292 * Get properties from the spa config. 293 */ 294 spa_prop_get_config(spa, nvp); 295 296 /* If no pool property object, no more prop to get. */ 297 if (mos == NULL || spa->spa_pool_props_object == 0) { 298 mutex_exit(&spa->spa_props_lock); 299 return (0); 300 } 301 302 /* 303 * Get properties from the MOS pool property object. 304 */ 305 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 306 (err = zap_cursor_retrieve(&zc, &za)) == 0; 307 zap_cursor_advance(&zc)) { 308 uint64_t intval = 0; 309 char *strval = NULL; 310 zprop_source_t src = ZPROP_SRC_DEFAULT; 311 zpool_prop_t prop; 312 313 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 314 continue; 315 316 switch (za.za_integer_length) { 317 case 8: 318 /* integer property */ 319 if (za.za_first_integer != 320 zpool_prop_default_numeric(prop)) 321 src = ZPROP_SRC_LOCAL; 322 323 if (prop == ZPOOL_PROP_BOOTFS) { 324 dsl_pool_t *dp; 325 dsl_dataset_t *ds = NULL; 326 327 dp = spa_get_dsl(spa); 328 dsl_pool_config_enter(dp, FTAG); 329 if (err = dsl_dataset_hold_obj(dp, 330 za.za_first_integer, FTAG, &ds)) { 331 dsl_pool_config_exit(dp, FTAG); 332 break; 333 } 334 335 strval = kmem_alloc( 336 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 337 KM_SLEEP); 338 dsl_dataset_name(ds, strval); 339 dsl_dataset_rele(ds, FTAG); 340 dsl_pool_config_exit(dp, FTAG); 341 } else { 342 strval = NULL; 343 intval = za.za_first_integer; 344 } 345 346 spa_prop_add_list(*nvp, prop, strval, intval, src); 347 348 if (strval != NULL) 349 kmem_free(strval, 350 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 351 352 break; 353 354 case 1: 355 /* string property */ 356 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 357 err = zap_lookup(mos, spa->spa_pool_props_object, 358 za.za_name, 1, za.za_num_integers, strval); 359 if (err) { 360 kmem_free(strval, za.za_num_integers); 361 break; 362 } 363 spa_prop_add_list(*nvp, prop, strval, 0, src); 364 kmem_free(strval, za.za_num_integers); 365 break; 366 367 default: 368 break; 369 } 370 } 371 zap_cursor_fini(&zc); 372 mutex_exit(&spa->spa_props_lock); 373out: 374 if (err && err != ENOENT) { 375 nvlist_free(*nvp); 376 *nvp = NULL; 377 return (err); 378 } 379 380 return (0); 381} 382 383/* 384 * Validate the given pool properties nvlist and modify the list 385 * for the property values to be set. 386 */ 387static int 388spa_prop_validate(spa_t *spa, nvlist_t *props) 389{ 390 nvpair_t *elem; 391 int error = 0, reset_bootfs = 0; 392 uint64_t objnum = 0; 393 boolean_t has_feature = B_FALSE; 394 395 elem = NULL; 396 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 397 uint64_t intval; 398 char *strval, *slash, *check, *fname; 399 const char *propname = nvpair_name(elem); 400 zpool_prop_t prop = zpool_name_to_prop(propname); 401 402 switch (prop) { 403 case ZPROP_INVAL: 404 if (!zpool_prop_feature(propname)) { 405 error = EINVAL; 406 break; 407 } 408 409 /* 410 * Sanitize the input. 411 */ 412 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 413 error = EINVAL; 414 break; 415 } 416 417 if (nvpair_value_uint64(elem, &intval) != 0) { 418 error = EINVAL; 419 break; 420 } 421 422 if (intval != 0) { 423 error = EINVAL; 424 break; 425 } 426 427 fname = strchr(propname, '@') + 1; 428 if (zfeature_lookup_name(fname, NULL) != 0) { 429 error = EINVAL; 430 break; 431 } 432 433 has_feature = B_TRUE; 434 break; 435 436 case ZPOOL_PROP_VERSION: 437 error = nvpair_value_uint64(elem, &intval); 438 if (!error && 439 (intval < spa_version(spa) || 440 intval > SPA_VERSION_BEFORE_FEATURES || 441 has_feature)) 442 error = EINVAL; 443 break; 444 445 case ZPOOL_PROP_DELEGATION: 446 case ZPOOL_PROP_AUTOREPLACE: 447 case ZPOOL_PROP_LISTSNAPS: 448 case ZPOOL_PROP_AUTOEXPAND: 449 error = nvpair_value_uint64(elem, &intval); 450 if (!error && intval > 1) 451 error = EINVAL; 452 break; 453 454 case ZPOOL_PROP_BOOTFS: 455 /* 456 * If the pool version is less than SPA_VERSION_BOOTFS, 457 * or the pool is still being created (version == 0), 458 * the bootfs property cannot be set. 459 */ 460 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 461 error = ENOTSUP; 462 break; 463 } 464 465 /* 466 * Make sure the vdev config is bootable 467 */ 468 if (!vdev_is_bootable(spa->spa_root_vdev)) { 469 error = ENOTSUP; 470 break; 471 } 472 473 reset_bootfs = 1; 474 475 error = nvpair_value_string(elem, &strval); 476 477 if (!error) { 478 objset_t *os; 479 uint64_t compress; 480 481 if (strval == NULL || strval[0] == '\0') { 482 objnum = zpool_prop_default_numeric( 483 ZPOOL_PROP_BOOTFS); 484 break; 485 } 486 487 if (error = dmu_objset_hold(strval, FTAG, &os)) 488 break; 489 490 /* Must be ZPL and not gzip compressed. */ 491 492 if (dmu_objset_type(os) != DMU_OST_ZFS) { 493 error = ENOTSUP; 494 } else if ((error = 495 dsl_prop_get_int_ds(dmu_objset_ds(os), 496 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 497 &compress)) == 0 && 498 !BOOTFS_COMPRESS_VALID(compress)) { 499 error = ENOTSUP; 500 } else { 501 objnum = dmu_objset_id(os); 502 } 503 dmu_objset_rele(os, FTAG); 504 } 505 break; 506 507 case ZPOOL_PROP_FAILUREMODE: 508 error = nvpair_value_uint64(elem, &intval); 509 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 510 intval > ZIO_FAILURE_MODE_PANIC)) 511 error = EINVAL; 512 513 /* 514 * This is a special case which only occurs when 515 * the pool has completely failed. This allows 516 * the user to change the in-core failmode property 517 * without syncing it out to disk (I/Os might 518 * currently be blocked). We do this by returning 519 * EIO to the caller (spa_prop_set) to trick it 520 * into thinking we encountered a property validation 521 * error. 522 */ 523 if (!error && spa_suspended(spa)) { 524 spa->spa_failmode = intval; 525 error = EIO; 526 } 527 break; 528 529 case ZPOOL_PROP_CACHEFILE: 530 if ((error = nvpair_value_string(elem, &strval)) != 0) 531 break; 532 533 if (strval[0] == '\0') 534 break; 535 536 if (strcmp(strval, "none") == 0) 537 break; 538 539 if (strval[0] != '/') { 540 error = EINVAL; 541 break; 542 } 543 544 slash = strrchr(strval, '/'); 545 ASSERT(slash != NULL); 546 547 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 548 strcmp(slash, "/..") == 0) 549 error = EINVAL; 550 break; 551 552 case ZPOOL_PROP_COMMENT: 553 if ((error = nvpair_value_string(elem, &strval)) != 0) 554 break; 555 for (check = strval; *check != '\0'; check++) { 556 /* 557 * The kernel doesn't have an easy isprint() 558 * check. For this kernel check, we merely 559 * check ASCII apart from DEL. Fix this if 560 * there is an easy-to-use kernel isprint(). 561 */ 562 if (*check >= 0x7f) { 563 error = EINVAL; 564 break; 565 } 566 check++; 567 } 568 if (strlen(strval) > ZPROP_MAX_COMMENT) 569 error = E2BIG; 570 break; 571 572 case ZPOOL_PROP_DEDUPDITTO: 573 if (spa_version(spa) < SPA_VERSION_DEDUP) 574 error = ENOTSUP; 575 else 576 error = nvpair_value_uint64(elem, &intval); 577 if (error == 0 && 578 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 579 error = EINVAL; 580 break; 581 } 582 583 if (error) 584 break; 585 } 586 587 if (!error && reset_bootfs) { 588 error = nvlist_remove(props, 589 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 590 591 if (!error) { 592 error = nvlist_add_uint64(props, 593 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 594 } 595 } 596 597 return (error); 598} 599 600void 601spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 602{ 603 char *cachefile; 604 spa_config_dirent_t *dp; 605 606 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 607 &cachefile) != 0) 608 return; 609 610 dp = kmem_alloc(sizeof (spa_config_dirent_t), 611 KM_SLEEP); 612 613 if (cachefile[0] == '\0') 614 dp->scd_path = spa_strdup(spa_config_path); 615 else if (strcmp(cachefile, "none") == 0) 616 dp->scd_path = NULL; 617 else 618 dp->scd_path = spa_strdup(cachefile); 619 620 list_insert_head(&spa->spa_config_list, dp); 621 if (need_sync) 622 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 623} 624 625int 626spa_prop_set(spa_t *spa, nvlist_t *nvp) 627{ 628 int error; 629 nvpair_t *elem = NULL; 630 boolean_t need_sync = B_FALSE; 631 632 if ((error = spa_prop_validate(spa, nvp)) != 0) 633 return (error); 634 635 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 636 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 637 638 if (prop == ZPOOL_PROP_CACHEFILE || 639 prop == ZPOOL_PROP_ALTROOT || 640 prop == ZPOOL_PROP_READONLY) 641 continue; 642 643 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 644 uint64_t ver; 645 646 if (prop == ZPOOL_PROP_VERSION) { 647 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 648 } else { 649 ASSERT(zpool_prop_feature(nvpair_name(elem))); 650 ver = SPA_VERSION_FEATURES; 651 need_sync = B_TRUE; 652 } 653 654 /* Save time if the version is already set. */ 655 if (ver == spa_version(spa)) 656 continue; 657 658 /* 659 * In addition to the pool directory object, we might 660 * create the pool properties object, the features for 661 * read object, the features for write object, or the 662 * feature descriptions object. 663 */ 664 error = dsl_sync_task(spa->spa_name, NULL, 665 spa_sync_version, &ver, 6); 666 if (error) 667 return (error); 668 continue; 669 } 670 671 need_sync = B_TRUE; 672 break; 673 } 674 675 if (need_sync) { 676 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 677 nvp, 6)); 678 } 679 680 return (0); 681} 682 683/* 684 * If the bootfs property value is dsobj, clear it. 685 */ 686void 687spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 688{ 689 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 690 VERIFY(zap_remove(spa->spa_meta_objset, 691 spa->spa_pool_props_object, 692 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 693 spa->spa_bootfs = 0; 694 } 695} 696 697/*ARGSUSED*/ 698static int 699spa_change_guid_check(void *arg, dmu_tx_t *tx) 700{ 701 uint64_t *newguid = arg; 702 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 703 vdev_t *rvd = spa->spa_root_vdev; 704 uint64_t vdev_state; 705 706 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 707 vdev_state = rvd->vdev_state; 708 spa_config_exit(spa, SCL_STATE, FTAG); 709 710 if (vdev_state != VDEV_STATE_HEALTHY) 711 return (ENXIO); 712 713 ASSERT3U(spa_guid(spa), !=, *newguid); 714 715 return (0); 716} 717 718static void 719spa_change_guid_sync(void *arg, dmu_tx_t *tx) 720{ 721 uint64_t *newguid = arg; 722 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 723 uint64_t oldguid; 724 vdev_t *rvd = spa->spa_root_vdev; 725 726 oldguid = spa_guid(spa); 727 728 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 729 rvd->vdev_guid = *newguid; 730 rvd->vdev_guid_sum += (*newguid - oldguid); 731 vdev_config_dirty(rvd); 732 spa_config_exit(spa, SCL_STATE, FTAG); 733 734 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 735 oldguid, *newguid); 736} 737 738/* 739 * Change the GUID for the pool. This is done so that we can later 740 * re-import a pool built from a clone of our own vdevs. We will modify 741 * the root vdev's guid, our own pool guid, and then mark all of our 742 * vdevs dirty. Note that we must make sure that all our vdevs are 743 * online when we do this, or else any vdevs that weren't present 744 * would be orphaned from our pool. We are also going to issue a 745 * sysevent to update any watchers. 746 */ 747int 748spa_change_guid(spa_t *spa) 749{ 750 int error; 751 uint64_t guid; 752 753 mutex_enter(&spa_namespace_lock); 754 guid = spa_generate_guid(NULL); 755 756 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 757 spa_change_guid_sync, &guid, 5); 758 759 if (error == 0) { 760 spa_config_sync(spa, B_FALSE, B_TRUE); 761 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 762 } 763 764 mutex_exit(&spa_namespace_lock); 765 766 return (error); 767} 768 769/* 770 * ========================================================================== 771 * SPA state manipulation (open/create/destroy/import/export) 772 * ========================================================================== 773 */ 774 775static int 776spa_error_entry_compare(const void *a, const void *b) 777{ 778 spa_error_entry_t *sa = (spa_error_entry_t *)a; 779 spa_error_entry_t *sb = (spa_error_entry_t *)b; 780 int ret; 781 782 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 783 sizeof (zbookmark_t)); 784 785 if (ret < 0) 786 return (-1); 787 else if (ret > 0) 788 return (1); 789 else 790 return (0); 791} 792 793/* 794 * Utility function which retrieves copies of the current logs and 795 * re-initializes them in the process. 796 */ 797void 798spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 799{ 800 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 801 802 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 803 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 804 805 avl_create(&spa->spa_errlist_scrub, 806 spa_error_entry_compare, sizeof (spa_error_entry_t), 807 offsetof(spa_error_entry_t, se_avl)); 808 avl_create(&spa->spa_errlist_last, 809 spa_error_entry_compare, sizeof (spa_error_entry_t), 810 offsetof(spa_error_entry_t, se_avl)); 811} 812 813static taskq_t * 814spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode, 815 uint_t value) 816{ 817 uint_t flags = TASKQ_PREPOPULATE; 818 boolean_t batch = B_FALSE; 819 820 switch (mode) { 821 case zti_mode_null: 822 return (NULL); /* no taskq needed */ 823 824 case zti_mode_fixed: 825 ASSERT3U(value, >=, 1); 826 value = MAX(value, 1); 827 break; 828 829 case zti_mode_batch: 830 batch = B_TRUE; 831 flags |= TASKQ_THREADS_CPU_PCT; 832 value = zio_taskq_batch_pct; 833 break; 834 835 case zti_mode_online_percent: 836 flags |= TASKQ_THREADS_CPU_PCT; 837 break; 838 839 default: 840 panic("unrecognized mode for %s taskq (%u:%u) in " 841 "spa_activate()", 842 name, mode, value); 843 break; 844 } 845 846#ifdef SYSDC 847 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 848 if (batch) 849 flags |= TASKQ_DC_BATCH; 850 851 return (taskq_create_sysdc(name, value, 50, INT_MAX, 852 spa->spa_proc, zio_taskq_basedc, flags)); 853 } 854#endif 855 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX, 856 spa->spa_proc, flags)); 857} 858 859static void 860spa_create_zio_taskqs(spa_t *spa) 861{ 862 for (int t = 0; t < ZIO_TYPES; t++) { 863 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 864 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 865 enum zti_modes mode = ztip->zti_mode; 866 uint_t value = ztip->zti_value; 867 char name[32]; 868 869 (void) snprintf(name, sizeof (name), 870 "%s_%s", zio_type_name[t], zio_taskq_types[q]); 871 872 spa->spa_zio_taskq[t][q] = 873 spa_taskq_create(spa, name, mode, value); 874 } 875 } 876} 877 878#ifdef _KERNEL 879#ifdef SPA_PROCESS 880static void 881spa_thread(void *arg) 882{ 883 callb_cpr_t cprinfo; 884 885 spa_t *spa = arg; 886 user_t *pu = PTOU(curproc); 887 888 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 889 spa->spa_name); 890 891 ASSERT(curproc != &p0); 892 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 893 "zpool-%s", spa->spa_name); 894 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 895 896#ifdef PSRSET_BIND 897 /* bind this thread to the requested psrset */ 898 if (zio_taskq_psrset_bind != PS_NONE) { 899 pool_lock(); 900 mutex_enter(&cpu_lock); 901 mutex_enter(&pidlock); 902 mutex_enter(&curproc->p_lock); 903 904 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 905 0, NULL, NULL) == 0) { 906 curthread->t_bind_pset = zio_taskq_psrset_bind; 907 } else { 908 cmn_err(CE_WARN, 909 "Couldn't bind process for zfs pool \"%s\" to " 910 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 911 } 912 913 mutex_exit(&curproc->p_lock); 914 mutex_exit(&pidlock); 915 mutex_exit(&cpu_lock); 916 pool_unlock(); 917 } 918#endif 919 920#ifdef SYSDC 921 if (zio_taskq_sysdc) { 922 sysdc_thread_enter(curthread, 100, 0); 923 } 924#endif 925 926 spa->spa_proc = curproc; 927 spa->spa_did = curthread->t_did; 928 929 spa_create_zio_taskqs(spa); 930 931 mutex_enter(&spa->spa_proc_lock); 932 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 933 934 spa->spa_proc_state = SPA_PROC_ACTIVE; 935 cv_broadcast(&spa->spa_proc_cv); 936 937 CALLB_CPR_SAFE_BEGIN(&cprinfo); 938 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 939 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 940 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 941 942 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 943 spa->spa_proc_state = SPA_PROC_GONE; 944 spa->spa_proc = &p0; 945 cv_broadcast(&spa->spa_proc_cv); 946 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 947 948 mutex_enter(&curproc->p_lock); 949 lwp_exit(); 950} 951#endif /* SPA_PROCESS */ 952#endif 953 954/* 955 * Activate an uninitialized pool. 956 */ 957static void 958spa_activate(spa_t *spa, int mode) 959{ 960 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 961 962 spa->spa_state = POOL_STATE_ACTIVE; 963 spa->spa_mode = mode; 964 965 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 966 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 967 968 /* Try to create a covering process */ 969 mutex_enter(&spa->spa_proc_lock); 970 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 971 ASSERT(spa->spa_proc == &p0); 972 spa->spa_did = 0; 973 974#ifdef SPA_PROCESS 975 /* Only create a process if we're going to be around a while. */ 976 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 977 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 978 NULL, 0) == 0) { 979 spa->spa_proc_state = SPA_PROC_CREATED; 980 while (spa->spa_proc_state == SPA_PROC_CREATED) { 981 cv_wait(&spa->spa_proc_cv, 982 &spa->spa_proc_lock); 983 } 984 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 985 ASSERT(spa->spa_proc != &p0); 986 ASSERT(spa->spa_did != 0); 987 } else { 988#ifdef _KERNEL 989 cmn_err(CE_WARN, 990 "Couldn't create process for zfs pool \"%s\"\n", 991 spa->spa_name); 992#endif 993 } 994 } 995#endif /* SPA_PROCESS */ 996 mutex_exit(&spa->spa_proc_lock); 997 998 /* If we didn't create a process, we need to create our taskqs. */ 999 ASSERT(spa->spa_proc == &p0); 1000 if (spa->spa_proc == &p0) { 1001 spa_create_zio_taskqs(spa); 1002 } 1003 1004 /* 1005 * Start TRIM thread. 1006 */ 1007 trim_thread_create(spa); 1008 1009 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1010 offsetof(vdev_t, vdev_config_dirty_node)); 1011 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1012 offsetof(vdev_t, vdev_state_dirty_node)); 1013 1014 txg_list_create(&spa->spa_vdev_txg_list, 1015 offsetof(struct vdev, vdev_txg_node)); 1016 1017 avl_create(&spa->spa_errlist_scrub, 1018 spa_error_entry_compare, sizeof (spa_error_entry_t), 1019 offsetof(spa_error_entry_t, se_avl)); 1020 avl_create(&spa->spa_errlist_last, 1021 spa_error_entry_compare, sizeof (spa_error_entry_t), 1022 offsetof(spa_error_entry_t, se_avl)); 1023} 1024 1025/* 1026 * Opposite of spa_activate(). 1027 */ 1028static void 1029spa_deactivate(spa_t *spa) 1030{ 1031 ASSERT(spa->spa_sync_on == B_FALSE); 1032 ASSERT(spa->spa_dsl_pool == NULL); 1033 ASSERT(spa->spa_root_vdev == NULL); 1034 ASSERT(spa->spa_async_zio_root == NULL); 1035 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1036 1037 /* 1038 * Stop TRIM thread in case spa_unload() wasn't called directly 1039 * before spa_deactivate(). 1040 */ 1041 trim_thread_destroy(spa); 1042 1043 txg_list_destroy(&spa->spa_vdev_txg_list); 1044 1045 list_destroy(&spa->spa_config_dirty_list); 1046 list_destroy(&spa->spa_state_dirty_list); 1047 1048 for (int t = 0; t < ZIO_TYPES; t++) { 1049 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1050 if (spa->spa_zio_taskq[t][q] != NULL) 1051 taskq_destroy(spa->spa_zio_taskq[t][q]); 1052 spa->spa_zio_taskq[t][q] = NULL; 1053 } 1054 } 1055 1056 metaslab_class_destroy(spa->spa_normal_class); 1057 spa->spa_normal_class = NULL; 1058 1059 metaslab_class_destroy(spa->spa_log_class); 1060 spa->spa_log_class = NULL; 1061 1062 /* 1063 * If this was part of an import or the open otherwise failed, we may 1064 * still have errors left in the queues. Empty them just in case. 1065 */ 1066 spa_errlog_drain(spa); 1067 1068 avl_destroy(&spa->spa_errlist_scrub); 1069 avl_destroy(&spa->spa_errlist_last); 1070 1071 spa->spa_state = POOL_STATE_UNINITIALIZED; 1072 1073 mutex_enter(&spa->spa_proc_lock); 1074 if (spa->spa_proc_state != SPA_PROC_NONE) { 1075 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1076 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1077 cv_broadcast(&spa->spa_proc_cv); 1078 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1079 ASSERT(spa->spa_proc != &p0); 1080 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1081 } 1082 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1083 spa->spa_proc_state = SPA_PROC_NONE; 1084 } 1085 ASSERT(spa->spa_proc == &p0); 1086 mutex_exit(&spa->spa_proc_lock); 1087 1088#ifdef SPA_PROCESS 1089 /* 1090 * We want to make sure spa_thread() has actually exited the ZFS 1091 * module, so that the module can't be unloaded out from underneath 1092 * it. 1093 */ 1094 if (spa->spa_did != 0) { 1095 thread_join(spa->spa_did); 1096 spa->spa_did = 0; 1097 } 1098#endif /* SPA_PROCESS */ 1099} 1100 1101/* 1102 * Verify a pool configuration, and construct the vdev tree appropriately. This 1103 * will create all the necessary vdevs in the appropriate layout, with each vdev 1104 * in the CLOSED state. This will prep the pool before open/creation/import. 1105 * All vdev validation is done by the vdev_alloc() routine. 1106 */ 1107static int 1108spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1109 uint_t id, int atype) 1110{ 1111 nvlist_t **child; 1112 uint_t children; 1113 int error; 1114 1115 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1116 return (error); 1117 1118 if ((*vdp)->vdev_ops->vdev_op_leaf) 1119 return (0); 1120 1121 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1122 &child, &children); 1123 1124 if (error == ENOENT) 1125 return (0); 1126 1127 if (error) { 1128 vdev_free(*vdp); 1129 *vdp = NULL; 1130 return (EINVAL); 1131 } 1132 1133 for (int c = 0; c < children; c++) { 1134 vdev_t *vd; 1135 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1136 atype)) != 0) { 1137 vdev_free(*vdp); 1138 *vdp = NULL; 1139 return (error); 1140 } 1141 } 1142 1143 ASSERT(*vdp != NULL); 1144 1145 return (0); 1146} 1147 1148/* 1149 * Opposite of spa_load(). 1150 */ 1151static void 1152spa_unload(spa_t *spa) 1153{ 1154 int i; 1155 1156 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1157 1158 /* 1159 * Stop TRIM thread. 1160 */ 1161 trim_thread_destroy(spa); 1162 1163 /* 1164 * Stop async tasks. 1165 */ 1166 spa_async_suspend(spa); 1167 1168 /* 1169 * Stop syncing. 1170 */ 1171 if (spa->spa_sync_on) { 1172 txg_sync_stop(spa->spa_dsl_pool); 1173 spa->spa_sync_on = B_FALSE; 1174 } 1175 1176 /* 1177 * Wait for any outstanding async I/O to complete. 1178 */ 1179 if (spa->spa_async_zio_root != NULL) { 1180 (void) zio_wait(spa->spa_async_zio_root); 1181 spa->spa_async_zio_root = NULL; 1182 } 1183 1184 bpobj_close(&spa->spa_deferred_bpobj); 1185 1186 /* 1187 * Close the dsl pool. 1188 */ 1189 if (spa->spa_dsl_pool) { 1190 dsl_pool_close(spa->spa_dsl_pool); 1191 spa->spa_dsl_pool = NULL; 1192 spa->spa_meta_objset = NULL; 1193 } 1194 1195 ddt_unload(spa); 1196 1197 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1198 1199 /* 1200 * Drop and purge level 2 cache 1201 */ 1202 spa_l2cache_drop(spa); 1203 1204 /* 1205 * Close all vdevs. 1206 */ 1207 if (spa->spa_root_vdev) 1208 vdev_free(spa->spa_root_vdev); 1209 ASSERT(spa->spa_root_vdev == NULL); 1210 1211 for (i = 0; i < spa->spa_spares.sav_count; i++) 1212 vdev_free(spa->spa_spares.sav_vdevs[i]); 1213 if (spa->spa_spares.sav_vdevs) { 1214 kmem_free(spa->spa_spares.sav_vdevs, 1215 spa->spa_spares.sav_count * sizeof (void *)); 1216 spa->spa_spares.sav_vdevs = NULL; 1217 } 1218 if (spa->spa_spares.sav_config) { 1219 nvlist_free(spa->spa_spares.sav_config); 1220 spa->spa_spares.sav_config = NULL; 1221 } 1222 spa->spa_spares.sav_count = 0; 1223 1224 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1225 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1226 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1227 } 1228 if (spa->spa_l2cache.sav_vdevs) { 1229 kmem_free(spa->spa_l2cache.sav_vdevs, 1230 spa->spa_l2cache.sav_count * sizeof (void *)); 1231 spa->spa_l2cache.sav_vdevs = NULL; 1232 } 1233 if (spa->spa_l2cache.sav_config) { 1234 nvlist_free(spa->spa_l2cache.sav_config); 1235 spa->spa_l2cache.sav_config = NULL; 1236 } 1237 spa->spa_l2cache.sav_count = 0; 1238 1239 spa->spa_async_suspended = 0; 1240 1241 if (spa->spa_comment != NULL) { 1242 spa_strfree(spa->spa_comment); 1243 spa->spa_comment = NULL; 1244 } 1245 1246 spa_config_exit(spa, SCL_ALL, FTAG); 1247} 1248 1249/* 1250 * Load (or re-load) the current list of vdevs describing the active spares for 1251 * this pool. When this is called, we have some form of basic information in 1252 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1253 * then re-generate a more complete list including status information. 1254 */ 1255static void 1256spa_load_spares(spa_t *spa) 1257{ 1258 nvlist_t **spares; 1259 uint_t nspares; 1260 int i; 1261 vdev_t *vd, *tvd; 1262 1263 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1264 1265 /* 1266 * First, close and free any existing spare vdevs. 1267 */ 1268 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1269 vd = spa->spa_spares.sav_vdevs[i]; 1270 1271 /* Undo the call to spa_activate() below */ 1272 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1273 B_FALSE)) != NULL && tvd->vdev_isspare) 1274 spa_spare_remove(tvd); 1275 vdev_close(vd); 1276 vdev_free(vd); 1277 } 1278 1279 if (spa->spa_spares.sav_vdevs) 1280 kmem_free(spa->spa_spares.sav_vdevs, 1281 spa->spa_spares.sav_count * sizeof (void *)); 1282 1283 if (spa->spa_spares.sav_config == NULL) 1284 nspares = 0; 1285 else 1286 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1287 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1288 1289 spa->spa_spares.sav_count = (int)nspares; 1290 spa->spa_spares.sav_vdevs = NULL; 1291 1292 if (nspares == 0) 1293 return; 1294 1295 /* 1296 * Construct the array of vdevs, opening them to get status in the 1297 * process. For each spare, there is potentially two different vdev_t 1298 * structures associated with it: one in the list of spares (used only 1299 * for basic validation purposes) and one in the active vdev 1300 * configuration (if it's spared in). During this phase we open and 1301 * validate each vdev on the spare list. If the vdev also exists in the 1302 * active configuration, then we also mark this vdev as an active spare. 1303 */ 1304 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1305 KM_SLEEP); 1306 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1307 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1308 VDEV_ALLOC_SPARE) == 0); 1309 ASSERT(vd != NULL); 1310 1311 spa->spa_spares.sav_vdevs[i] = vd; 1312 1313 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1314 B_FALSE)) != NULL) { 1315 if (!tvd->vdev_isspare) 1316 spa_spare_add(tvd); 1317 1318 /* 1319 * We only mark the spare active if we were successfully 1320 * able to load the vdev. Otherwise, importing a pool 1321 * with a bad active spare would result in strange 1322 * behavior, because multiple pool would think the spare 1323 * is actively in use. 1324 * 1325 * There is a vulnerability here to an equally bizarre 1326 * circumstance, where a dead active spare is later 1327 * brought back to life (onlined or otherwise). Given 1328 * the rarity of this scenario, and the extra complexity 1329 * it adds, we ignore the possibility. 1330 */ 1331 if (!vdev_is_dead(tvd)) 1332 spa_spare_activate(tvd); 1333 } 1334 1335 vd->vdev_top = vd; 1336 vd->vdev_aux = &spa->spa_spares; 1337 1338 if (vdev_open(vd) != 0) 1339 continue; 1340 1341 if (vdev_validate_aux(vd) == 0) 1342 spa_spare_add(vd); 1343 } 1344 1345 /* 1346 * Recompute the stashed list of spares, with status information 1347 * this time. 1348 */ 1349 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1350 DATA_TYPE_NVLIST_ARRAY) == 0); 1351 1352 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1353 KM_SLEEP); 1354 for (i = 0; i < spa->spa_spares.sav_count; i++) 1355 spares[i] = vdev_config_generate(spa, 1356 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1357 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1358 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1359 for (i = 0; i < spa->spa_spares.sav_count; i++) 1360 nvlist_free(spares[i]); 1361 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1362} 1363 1364/* 1365 * Load (or re-load) the current list of vdevs describing the active l2cache for 1366 * this pool. When this is called, we have some form of basic information in 1367 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1368 * then re-generate a more complete list including status information. 1369 * Devices which are already active have their details maintained, and are 1370 * not re-opened. 1371 */ 1372static void 1373spa_load_l2cache(spa_t *spa) 1374{ 1375 nvlist_t **l2cache; 1376 uint_t nl2cache; 1377 int i, j, oldnvdevs; 1378 uint64_t guid; 1379 vdev_t *vd, **oldvdevs, **newvdevs; 1380 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1381 1382 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1383 1384 if (sav->sav_config != NULL) { 1385 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1386 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1387 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1388 } else { 1389 nl2cache = 0; 1390 newvdevs = NULL; 1391 } 1392 1393 oldvdevs = sav->sav_vdevs; 1394 oldnvdevs = sav->sav_count; 1395 sav->sav_vdevs = NULL; 1396 sav->sav_count = 0; 1397 1398 /* 1399 * Process new nvlist of vdevs. 1400 */ 1401 for (i = 0; i < nl2cache; i++) { 1402 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1403 &guid) == 0); 1404 1405 newvdevs[i] = NULL; 1406 for (j = 0; j < oldnvdevs; j++) { 1407 vd = oldvdevs[j]; 1408 if (vd != NULL && guid == vd->vdev_guid) { 1409 /* 1410 * Retain previous vdev for add/remove ops. 1411 */ 1412 newvdevs[i] = vd; 1413 oldvdevs[j] = NULL; 1414 break; 1415 } 1416 } 1417 1418 if (newvdevs[i] == NULL) { 1419 /* 1420 * Create new vdev 1421 */ 1422 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1423 VDEV_ALLOC_L2CACHE) == 0); 1424 ASSERT(vd != NULL); 1425 newvdevs[i] = vd; 1426 1427 /* 1428 * Commit this vdev as an l2cache device, 1429 * even if it fails to open. 1430 */ 1431 spa_l2cache_add(vd); 1432 1433 vd->vdev_top = vd; 1434 vd->vdev_aux = sav; 1435 1436 spa_l2cache_activate(vd); 1437 1438 if (vdev_open(vd) != 0) 1439 continue; 1440 1441 (void) vdev_validate_aux(vd); 1442 1443 if (!vdev_is_dead(vd)) 1444 l2arc_add_vdev(spa, vd); 1445 } 1446 } 1447 1448 /* 1449 * Purge vdevs that were dropped 1450 */ 1451 for (i = 0; i < oldnvdevs; i++) { 1452 uint64_t pool; 1453 1454 vd = oldvdevs[i]; 1455 if (vd != NULL) { 1456 ASSERT(vd->vdev_isl2cache); 1457 1458 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1459 pool != 0ULL && l2arc_vdev_present(vd)) 1460 l2arc_remove_vdev(vd); 1461 vdev_clear_stats(vd); 1462 vdev_free(vd); 1463 } 1464 } 1465 1466 if (oldvdevs) 1467 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1468 1469 if (sav->sav_config == NULL) 1470 goto out; 1471 1472 sav->sav_vdevs = newvdevs; 1473 sav->sav_count = (int)nl2cache; 1474 1475 /* 1476 * Recompute the stashed list of l2cache devices, with status 1477 * information this time. 1478 */ 1479 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1480 DATA_TYPE_NVLIST_ARRAY) == 0); 1481 1482 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1483 for (i = 0; i < sav->sav_count; i++) 1484 l2cache[i] = vdev_config_generate(spa, 1485 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1486 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1487 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1488out: 1489 for (i = 0; i < sav->sav_count; i++) 1490 nvlist_free(l2cache[i]); 1491 if (sav->sav_count) 1492 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1493} 1494 1495static int 1496load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1497{ 1498 dmu_buf_t *db; 1499 char *packed = NULL; 1500 size_t nvsize = 0; 1501 int error; 1502 *value = NULL; 1503 1504 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 1505 nvsize = *(uint64_t *)db->db_data; 1506 dmu_buf_rele(db, FTAG); 1507 1508 packed = kmem_alloc(nvsize, KM_SLEEP); 1509 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1510 DMU_READ_PREFETCH); 1511 if (error == 0) 1512 error = nvlist_unpack(packed, nvsize, value, 0); 1513 kmem_free(packed, nvsize); 1514 1515 return (error); 1516} 1517 1518/* 1519 * Checks to see if the given vdev could not be opened, in which case we post a 1520 * sysevent to notify the autoreplace code that the device has been removed. 1521 */ 1522static void 1523spa_check_removed(vdev_t *vd) 1524{ 1525 for (int c = 0; c < vd->vdev_children; c++) 1526 spa_check_removed(vd->vdev_child[c]); 1527 1528 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) { 1529 zfs_post_autoreplace(vd->vdev_spa, vd); 1530 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1531 } 1532} 1533 1534/* 1535 * Validate the current config against the MOS config 1536 */ 1537static boolean_t 1538spa_config_valid(spa_t *spa, nvlist_t *config) 1539{ 1540 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1541 nvlist_t *nv; 1542 1543 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1544 1545 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1546 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1547 1548 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1549 1550 /* 1551 * If we're doing a normal import, then build up any additional 1552 * diagnostic information about missing devices in this config. 1553 * We'll pass this up to the user for further processing. 1554 */ 1555 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1556 nvlist_t **child, *nv; 1557 uint64_t idx = 0; 1558 1559 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1560 KM_SLEEP); 1561 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1562 1563 for (int c = 0; c < rvd->vdev_children; c++) { 1564 vdev_t *tvd = rvd->vdev_child[c]; 1565 vdev_t *mtvd = mrvd->vdev_child[c]; 1566 1567 if (tvd->vdev_ops == &vdev_missing_ops && 1568 mtvd->vdev_ops != &vdev_missing_ops && 1569 mtvd->vdev_islog) 1570 child[idx++] = vdev_config_generate(spa, mtvd, 1571 B_FALSE, 0); 1572 } 1573 1574 if (idx) { 1575 VERIFY(nvlist_add_nvlist_array(nv, 1576 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1577 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1578 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1579 1580 for (int i = 0; i < idx; i++) 1581 nvlist_free(child[i]); 1582 } 1583 nvlist_free(nv); 1584 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1585 } 1586 1587 /* 1588 * Compare the root vdev tree with the information we have 1589 * from the MOS config (mrvd). Check each top-level vdev 1590 * with the corresponding MOS config top-level (mtvd). 1591 */ 1592 for (int c = 0; c < rvd->vdev_children; c++) { 1593 vdev_t *tvd = rvd->vdev_child[c]; 1594 vdev_t *mtvd = mrvd->vdev_child[c]; 1595 1596 /* 1597 * Resolve any "missing" vdevs in the current configuration. 1598 * If we find that the MOS config has more accurate information 1599 * about the top-level vdev then use that vdev instead. 1600 */ 1601 if (tvd->vdev_ops == &vdev_missing_ops && 1602 mtvd->vdev_ops != &vdev_missing_ops) { 1603 1604 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1605 continue; 1606 1607 /* 1608 * Device specific actions. 1609 */ 1610 if (mtvd->vdev_islog) { 1611 spa_set_log_state(spa, SPA_LOG_CLEAR); 1612 } else { 1613 /* 1614 * XXX - once we have 'readonly' pool 1615 * support we should be able to handle 1616 * missing data devices by transitioning 1617 * the pool to readonly. 1618 */ 1619 continue; 1620 } 1621 1622 /* 1623 * Swap the missing vdev with the data we were 1624 * able to obtain from the MOS config. 1625 */ 1626 vdev_remove_child(rvd, tvd); 1627 vdev_remove_child(mrvd, mtvd); 1628 1629 vdev_add_child(rvd, mtvd); 1630 vdev_add_child(mrvd, tvd); 1631 1632 spa_config_exit(spa, SCL_ALL, FTAG); 1633 vdev_load(mtvd); 1634 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1635 1636 vdev_reopen(rvd); 1637 } else if (mtvd->vdev_islog) { 1638 /* 1639 * Load the slog device's state from the MOS config 1640 * since it's possible that the label does not 1641 * contain the most up-to-date information. 1642 */ 1643 vdev_load_log_state(tvd, mtvd); 1644 vdev_reopen(tvd); 1645 } 1646 } 1647 vdev_free(mrvd); 1648 spa_config_exit(spa, SCL_ALL, FTAG); 1649 1650 /* 1651 * Ensure we were able to validate the config. 1652 */ 1653 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1654} 1655 1656/* 1657 * Check for missing log devices 1658 */ 1659static boolean_t 1660spa_check_logs(spa_t *spa) 1661{ 1662 boolean_t rv = B_FALSE; 1663 1664 switch (spa->spa_log_state) { 1665 case SPA_LOG_MISSING: 1666 /* need to recheck in case slog has been restored */ 1667 case SPA_LOG_UNKNOWN: 1668 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1669 NULL, DS_FIND_CHILDREN) != 0); 1670 if (rv) 1671 spa_set_log_state(spa, SPA_LOG_MISSING); 1672 break; 1673 } 1674 return (rv); 1675} 1676 1677static boolean_t 1678spa_passivate_log(spa_t *spa) 1679{ 1680 vdev_t *rvd = spa->spa_root_vdev; 1681 boolean_t slog_found = B_FALSE; 1682 1683 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1684 1685 if (!spa_has_slogs(spa)) 1686 return (B_FALSE); 1687 1688 for (int c = 0; c < rvd->vdev_children; c++) { 1689 vdev_t *tvd = rvd->vdev_child[c]; 1690 metaslab_group_t *mg = tvd->vdev_mg; 1691 1692 if (tvd->vdev_islog) { 1693 metaslab_group_passivate(mg); 1694 slog_found = B_TRUE; 1695 } 1696 } 1697 1698 return (slog_found); 1699} 1700 1701static void 1702spa_activate_log(spa_t *spa) 1703{ 1704 vdev_t *rvd = spa->spa_root_vdev; 1705 1706 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1707 1708 for (int c = 0; c < rvd->vdev_children; c++) { 1709 vdev_t *tvd = rvd->vdev_child[c]; 1710 metaslab_group_t *mg = tvd->vdev_mg; 1711 1712 if (tvd->vdev_islog) 1713 metaslab_group_activate(mg); 1714 } 1715} 1716 1717int 1718spa_offline_log(spa_t *spa) 1719{ 1720 int error; 1721 1722 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1723 NULL, DS_FIND_CHILDREN); 1724 if (error == 0) { 1725 /* 1726 * We successfully offlined the log device, sync out the 1727 * current txg so that the "stubby" block can be removed 1728 * by zil_sync(). 1729 */ 1730 txg_wait_synced(spa->spa_dsl_pool, 0); 1731 } 1732 return (error); 1733} 1734 1735static void 1736spa_aux_check_removed(spa_aux_vdev_t *sav) 1737{ 1738 int i; 1739 1740 for (i = 0; i < sav->sav_count; i++) 1741 spa_check_removed(sav->sav_vdevs[i]); 1742} 1743 1744void 1745spa_claim_notify(zio_t *zio) 1746{ 1747 spa_t *spa = zio->io_spa; 1748 1749 if (zio->io_error) 1750 return; 1751 1752 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1753 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1754 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1755 mutex_exit(&spa->spa_props_lock); 1756} 1757 1758typedef struct spa_load_error { 1759 uint64_t sle_meta_count; 1760 uint64_t sle_data_count; 1761} spa_load_error_t; 1762 1763static void 1764spa_load_verify_done(zio_t *zio) 1765{ 1766 blkptr_t *bp = zio->io_bp; 1767 spa_load_error_t *sle = zio->io_private; 1768 dmu_object_type_t type = BP_GET_TYPE(bp); 1769 int error = zio->io_error; 1770 1771 if (error) { 1772 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1773 type != DMU_OT_INTENT_LOG) 1774 atomic_add_64(&sle->sle_meta_count, 1); 1775 else 1776 atomic_add_64(&sle->sle_data_count, 1); 1777 } 1778 zio_data_buf_free(zio->io_data, zio->io_size); 1779} 1780 1781/*ARGSUSED*/ 1782static int 1783spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1784 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1785{ 1786 if (bp != NULL) { 1787 zio_t *rio = arg; 1788 size_t size = BP_GET_PSIZE(bp); 1789 void *data = zio_data_buf_alloc(size); 1790 1791 zio_nowait(zio_read(rio, spa, bp, data, size, 1792 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1793 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1794 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1795 } 1796 return (0); 1797} 1798 1799static int 1800spa_load_verify(spa_t *spa) 1801{ 1802 zio_t *rio; 1803 spa_load_error_t sle = { 0 }; 1804 zpool_rewind_policy_t policy; 1805 boolean_t verify_ok = B_FALSE; 1806 int error; 1807 1808 zpool_get_rewind_policy(spa->spa_config, &policy); 1809 1810 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1811 return (0); 1812 1813 rio = zio_root(spa, NULL, &sle, 1814 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1815 1816 error = traverse_pool(spa, spa->spa_verify_min_txg, 1817 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1818 1819 (void) zio_wait(rio); 1820 1821 spa->spa_load_meta_errors = sle.sle_meta_count; 1822 spa->spa_load_data_errors = sle.sle_data_count; 1823 1824 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1825 sle.sle_data_count <= policy.zrp_maxdata) { 1826 int64_t loss = 0; 1827 1828 verify_ok = B_TRUE; 1829 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1830 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1831 1832 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1833 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1834 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1835 VERIFY(nvlist_add_int64(spa->spa_load_info, 1836 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1837 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1838 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1839 } else { 1840 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1841 } 1842 1843 if (error) { 1844 if (error != ENXIO && error != EIO) 1845 error = EIO; 1846 return (error); 1847 } 1848 1849 return (verify_ok ? 0 : EIO); 1850} 1851 1852/* 1853 * Find a value in the pool props object. 1854 */ 1855static void 1856spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1857{ 1858 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1859 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1860} 1861 1862/* 1863 * Find a value in the pool directory object. 1864 */ 1865static int 1866spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1867{ 1868 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1869 name, sizeof (uint64_t), 1, val)); 1870} 1871 1872static int 1873spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1874{ 1875 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1876 return (err); 1877} 1878 1879/* 1880 * Fix up config after a partly-completed split. This is done with the 1881 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1882 * pool have that entry in their config, but only the splitting one contains 1883 * a list of all the guids of the vdevs that are being split off. 1884 * 1885 * This function determines what to do with that list: either rejoin 1886 * all the disks to the pool, or complete the splitting process. To attempt 1887 * the rejoin, each disk that is offlined is marked online again, and 1888 * we do a reopen() call. If the vdev label for every disk that was 1889 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 1890 * then we call vdev_split() on each disk, and complete the split. 1891 * 1892 * Otherwise we leave the config alone, with all the vdevs in place in 1893 * the original pool. 1894 */ 1895static void 1896spa_try_repair(spa_t *spa, nvlist_t *config) 1897{ 1898 uint_t extracted; 1899 uint64_t *glist; 1900 uint_t i, gcount; 1901 nvlist_t *nvl; 1902 vdev_t **vd; 1903 boolean_t attempt_reopen; 1904 1905 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 1906 return; 1907 1908 /* check that the config is complete */ 1909 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 1910 &glist, &gcount) != 0) 1911 return; 1912 1913 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 1914 1915 /* attempt to online all the vdevs & validate */ 1916 attempt_reopen = B_TRUE; 1917 for (i = 0; i < gcount; i++) { 1918 if (glist[i] == 0) /* vdev is hole */ 1919 continue; 1920 1921 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 1922 if (vd[i] == NULL) { 1923 /* 1924 * Don't bother attempting to reopen the disks; 1925 * just do the split. 1926 */ 1927 attempt_reopen = B_FALSE; 1928 } else { 1929 /* attempt to re-online it */ 1930 vd[i]->vdev_offline = B_FALSE; 1931 } 1932 } 1933 1934 if (attempt_reopen) { 1935 vdev_reopen(spa->spa_root_vdev); 1936 1937 /* check each device to see what state it's in */ 1938 for (extracted = 0, i = 0; i < gcount; i++) { 1939 if (vd[i] != NULL && 1940 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 1941 break; 1942 ++extracted; 1943 } 1944 } 1945 1946 /* 1947 * If every disk has been moved to the new pool, or if we never 1948 * even attempted to look at them, then we split them off for 1949 * good. 1950 */ 1951 if (!attempt_reopen || gcount == extracted) { 1952 for (i = 0; i < gcount; i++) 1953 if (vd[i] != NULL) 1954 vdev_split(vd[i]); 1955 vdev_reopen(spa->spa_root_vdev); 1956 } 1957 1958 kmem_free(vd, gcount * sizeof (vdev_t *)); 1959} 1960 1961static int 1962spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 1963 boolean_t mosconfig) 1964{ 1965 nvlist_t *config = spa->spa_config; 1966 char *ereport = FM_EREPORT_ZFS_POOL; 1967 char *comment; 1968 int error; 1969 uint64_t pool_guid; 1970 nvlist_t *nvl; 1971 1972 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 1973 return (EINVAL); 1974 1975 ASSERT(spa->spa_comment == NULL); 1976 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 1977 spa->spa_comment = spa_strdup(comment); 1978 1979 /* 1980 * Versioning wasn't explicitly added to the label until later, so if 1981 * it's not present treat it as the initial version. 1982 */ 1983 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 1984 &spa->spa_ubsync.ub_version) != 0) 1985 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 1986 1987 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 1988 &spa->spa_config_txg); 1989 1990 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 1991 spa_guid_exists(pool_guid, 0)) { 1992 error = EEXIST; 1993 } else { 1994 spa->spa_config_guid = pool_guid; 1995 1996 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 1997 &nvl) == 0) { 1998 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 1999 KM_SLEEP) == 0); 2000 } 2001 2002 nvlist_free(spa->spa_load_info); 2003 spa->spa_load_info = fnvlist_alloc(); 2004 2005 gethrestime(&spa->spa_loaded_ts); 2006 error = spa_load_impl(spa, pool_guid, config, state, type, 2007 mosconfig, &ereport); 2008 } 2009 2010 spa->spa_minref = refcount_count(&spa->spa_refcount); 2011 if (error) { 2012 if (error != EEXIST) { 2013 spa->spa_loaded_ts.tv_sec = 0; 2014 spa->spa_loaded_ts.tv_nsec = 0; 2015 } 2016 if (error != EBADF) { 2017 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2018 } 2019 } 2020 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2021 spa->spa_ena = 0; 2022 2023 return (error); 2024} 2025 2026/* 2027 * Load an existing storage pool, using the pool's builtin spa_config as a 2028 * source of configuration information. 2029 */ 2030static int 2031spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2032 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2033 char **ereport) 2034{ 2035 int error = 0; 2036 nvlist_t *nvroot = NULL; 2037 nvlist_t *label; 2038 vdev_t *rvd; 2039 uberblock_t *ub = &spa->spa_uberblock; 2040 uint64_t children, config_cache_txg = spa->spa_config_txg; 2041 int orig_mode = spa->spa_mode; 2042 int parse; 2043 uint64_t obj; 2044 boolean_t missing_feat_write = B_FALSE; 2045 2046 /* 2047 * If this is an untrusted config, access the pool in read-only mode. 2048 * This prevents things like resilvering recently removed devices. 2049 */ 2050 if (!mosconfig) 2051 spa->spa_mode = FREAD; 2052 2053 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2054 2055 spa->spa_load_state = state; 2056 2057 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2058 return (EINVAL); 2059 2060 parse = (type == SPA_IMPORT_EXISTING ? 2061 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2062 2063 /* 2064 * Create "The Godfather" zio to hold all async IOs 2065 */ 2066 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 2067 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 2068 2069 /* 2070 * Parse the configuration into a vdev tree. We explicitly set the 2071 * value that will be returned by spa_version() since parsing the 2072 * configuration requires knowing the version number. 2073 */ 2074 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2075 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2076 spa_config_exit(spa, SCL_ALL, FTAG); 2077 2078 if (error != 0) 2079 return (error); 2080 2081 ASSERT(spa->spa_root_vdev == rvd); 2082 2083 if (type != SPA_IMPORT_ASSEMBLE) { 2084 ASSERT(spa_guid(spa) == pool_guid); 2085 } 2086 2087 /* 2088 * Try to open all vdevs, loading each label in the process. 2089 */ 2090 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2091 error = vdev_open(rvd); 2092 spa_config_exit(spa, SCL_ALL, FTAG); 2093 if (error != 0) 2094 return (error); 2095 2096 /* 2097 * We need to validate the vdev labels against the configuration that 2098 * we have in hand, which is dependent on the setting of mosconfig. If 2099 * mosconfig is true then we're validating the vdev labels based on 2100 * that config. Otherwise, we're validating against the cached config 2101 * (zpool.cache) that was read when we loaded the zfs module, and then 2102 * later we will recursively call spa_load() and validate against 2103 * the vdev config. 2104 * 2105 * If we're assembling a new pool that's been split off from an 2106 * existing pool, the labels haven't yet been updated so we skip 2107 * validation for now. 2108 */ 2109 if (type != SPA_IMPORT_ASSEMBLE) { 2110 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2111 error = vdev_validate(rvd, mosconfig); 2112 spa_config_exit(spa, SCL_ALL, FTAG); 2113 2114 if (error != 0) 2115 return (error); 2116 2117 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2118 return (ENXIO); 2119 } 2120 2121 /* 2122 * Find the best uberblock. 2123 */ 2124 vdev_uberblock_load(rvd, ub, &label); 2125 2126 /* 2127 * If we weren't able to find a single valid uberblock, return failure. 2128 */ 2129 if (ub->ub_txg == 0) { 2130 nvlist_free(label); 2131 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2132 } 2133 2134 /* 2135 * If the pool has an unsupported version we can't open it. 2136 */ 2137 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2138 nvlist_free(label); 2139 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2140 } 2141 2142 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2143 nvlist_t *features; 2144 2145 /* 2146 * If we weren't able to find what's necessary for reading the 2147 * MOS in the label, return failure. 2148 */ 2149 if (label == NULL || nvlist_lookup_nvlist(label, 2150 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2151 nvlist_free(label); 2152 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2153 ENXIO)); 2154 } 2155 2156 /* 2157 * Update our in-core representation with the definitive values 2158 * from the label. 2159 */ 2160 nvlist_free(spa->spa_label_features); 2161 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2162 } 2163 2164 nvlist_free(label); 2165 2166 /* 2167 * Look through entries in the label nvlist's features_for_read. If 2168 * there is a feature listed there which we don't understand then we 2169 * cannot open a pool. 2170 */ 2171 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2172 nvlist_t *unsup_feat; 2173 2174 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2175 0); 2176 2177 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2178 NULL); nvp != NULL; 2179 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2180 if (!zfeature_is_supported(nvpair_name(nvp))) { 2181 VERIFY(nvlist_add_string(unsup_feat, 2182 nvpair_name(nvp), "") == 0); 2183 } 2184 } 2185 2186 if (!nvlist_empty(unsup_feat)) { 2187 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2188 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2189 nvlist_free(unsup_feat); 2190 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2191 ENOTSUP)); 2192 } 2193 2194 nvlist_free(unsup_feat); 2195 } 2196 2197 /* 2198 * If the vdev guid sum doesn't match the uberblock, we have an 2199 * incomplete configuration. We first check to see if the pool 2200 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2201 * If it is, defer the vdev_guid_sum check till later so we 2202 * can handle missing vdevs. 2203 */ 2204 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2205 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2206 rvd->vdev_guid_sum != ub->ub_guid_sum) 2207 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2208 2209 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2210 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2211 spa_try_repair(spa, config); 2212 spa_config_exit(spa, SCL_ALL, FTAG); 2213 nvlist_free(spa->spa_config_splitting); 2214 spa->spa_config_splitting = NULL; 2215 } 2216 2217 /* 2218 * Initialize internal SPA structures. 2219 */ 2220 spa->spa_state = POOL_STATE_ACTIVE; 2221 spa->spa_ubsync = spa->spa_uberblock; 2222 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2223 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2224 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2225 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2226 spa->spa_claim_max_txg = spa->spa_first_txg; 2227 spa->spa_prev_software_version = ub->ub_software_version; 2228 2229 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2230 if (error) 2231 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2232 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2233 2234 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2235 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2236 2237 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2238 boolean_t missing_feat_read = B_FALSE; 2239 nvlist_t *unsup_feat, *enabled_feat; 2240 2241 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2242 &spa->spa_feat_for_read_obj) != 0) { 2243 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2244 } 2245 2246 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2247 &spa->spa_feat_for_write_obj) != 0) { 2248 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2249 } 2250 2251 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2252 &spa->spa_feat_desc_obj) != 0) { 2253 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2254 } 2255 2256 enabled_feat = fnvlist_alloc(); 2257 unsup_feat = fnvlist_alloc(); 2258 2259 if (!feature_is_supported(spa->spa_meta_objset, 2260 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, 2261 unsup_feat, enabled_feat)) 2262 missing_feat_read = B_TRUE; 2263 2264 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2265 if (!feature_is_supported(spa->spa_meta_objset, 2266 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, 2267 unsup_feat, enabled_feat)) { 2268 missing_feat_write = B_TRUE; 2269 } 2270 } 2271 2272 fnvlist_add_nvlist(spa->spa_load_info, 2273 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2274 2275 if (!nvlist_empty(unsup_feat)) { 2276 fnvlist_add_nvlist(spa->spa_load_info, 2277 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2278 } 2279 2280 fnvlist_free(enabled_feat); 2281 fnvlist_free(unsup_feat); 2282 2283 if (!missing_feat_read) { 2284 fnvlist_add_boolean(spa->spa_load_info, 2285 ZPOOL_CONFIG_CAN_RDONLY); 2286 } 2287 2288 /* 2289 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2290 * twofold: to determine whether the pool is available for 2291 * import in read-write mode and (if it is not) whether the 2292 * pool is available for import in read-only mode. If the pool 2293 * is available for import in read-write mode, it is displayed 2294 * as available in userland; if it is not available for import 2295 * in read-only mode, it is displayed as unavailable in 2296 * userland. If the pool is available for import in read-only 2297 * mode but not read-write mode, it is displayed as unavailable 2298 * in userland with a special note that the pool is actually 2299 * available for open in read-only mode. 2300 * 2301 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2302 * missing a feature for write, we must first determine whether 2303 * the pool can be opened read-only before returning to 2304 * userland in order to know whether to display the 2305 * abovementioned note. 2306 */ 2307 if (missing_feat_read || (missing_feat_write && 2308 spa_writeable(spa))) { 2309 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2310 ENOTSUP)); 2311 } 2312 } 2313 2314 spa->spa_is_initializing = B_TRUE; 2315 error = dsl_pool_open(spa->spa_dsl_pool); 2316 spa->spa_is_initializing = B_FALSE; 2317 if (error != 0) 2318 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2319 2320 if (!mosconfig) { 2321 uint64_t hostid; 2322 nvlist_t *policy = NULL, *nvconfig; 2323 2324 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2326 2327 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2328 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2329 char *hostname; 2330 unsigned long myhostid = 0; 2331 2332 VERIFY(nvlist_lookup_string(nvconfig, 2333 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2334 2335#ifdef _KERNEL 2336 myhostid = zone_get_hostid(NULL); 2337#else /* _KERNEL */ 2338 /* 2339 * We're emulating the system's hostid in userland, so 2340 * we can't use zone_get_hostid(). 2341 */ 2342 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2343#endif /* _KERNEL */ 2344 if (check_hostid && hostid != 0 && myhostid != 0 && 2345 hostid != myhostid) { 2346 nvlist_free(nvconfig); 2347 cmn_err(CE_WARN, "pool '%s' could not be " 2348 "loaded as it was last accessed by " 2349 "another system (host: %s hostid: 0x%lx). " 2350 "See: http://illumos.org/msg/ZFS-8000-EY", 2351 spa_name(spa), hostname, 2352 (unsigned long)hostid); 2353 return (EBADF); 2354 } 2355 } 2356 if (nvlist_lookup_nvlist(spa->spa_config, 2357 ZPOOL_REWIND_POLICY, &policy) == 0) 2358 VERIFY(nvlist_add_nvlist(nvconfig, 2359 ZPOOL_REWIND_POLICY, policy) == 0); 2360 2361 spa_config_set(spa, nvconfig); 2362 spa_unload(spa); 2363 spa_deactivate(spa); 2364 spa_activate(spa, orig_mode); 2365 2366 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2367 } 2368 2369 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2370 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2371 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2372 if (error != 0) 2373 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2374 2375 /* 2376 * Load the bit that tells us to use the new accounting function 2377 * (raid-z deflation). If we have an older pool, this will not 2378 * be present. 2379 */ 2380 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2381 if (error != 0 && error != ENOENT) 2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2383 2384 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2385 &spa->spa_creation_version); 2386 if (error != 0 && error != ENOENT) 2387 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2388 2389 /* 2390 * Load the persistent error log. If we have an older pool, this will 2391 * not be present. 2392 */ 2393 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2394 if (error != 0 && error != ENOENT) 2395 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2396 2397 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2398 &spa->spa_errlog_scrub); 2399 if (error != 0 && error != ENOENT) 2400 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2401 2402 /* 2403 * Load the history object. If we have an older pool, this 2404 * will not be present. 2405 */ 2406 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2407 if (error != 0 && error != ENOENT) 2408 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2409 2410 /* 2411 * If we're assembling the pool from the split-off vdevs of 2412 * an existing pool, we don't want to attach the spares & cache 2413 * devices. 2414 */ 2415 2416 /* 2417 * Load any hot spares for this pool. 2418 */ 2419 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2420 if (error != 0 && error != ENOENT) 2421 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2422 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2423 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2424 if (load_nvlist(spa, spa->spa_spares.sav_object, 2425 &spa->spa_spares.sav_config) != 0) 2426 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2427 2428 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2429 spa_load_spares(spa); 2430 spa_config_exit(spa, SCL_ALL, FTAG); 2431 } else if (error == 0) { 2432 spa->spa_spares.sav_sync = B_TRUE; 2433 } 2434 2435 /* 2436 * Load any level 2 ARC devices for this pool. 2437 */ 2438 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2439 &spa->spa_l2cache.sav_object); 2440 if (error != 0 && error != ENOENT) 2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2442 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2443 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2444 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2445 &spa->spa_l2cache.sav_config) != 0) 2446 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2447 2448 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2449 spa_load_l2cache(spa); 2450 spa_config_exit(spa, SCL_ALL, FTAG); 2451 } else if (error == 0) { 2452 spa->spa_l2cache.sav_sync = B_TRUE; 2453 } 2454 2455 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2456 2457 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2458 if (error && error != ENOENT) 2459 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2460 2461 if (error == 0) { 2462 uint64_t autoreplace; 2463 2464 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2465 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2466 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2467 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2468 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2469 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2470 &spa->spa_dedup_ditto); 2471 2472 spa->spa_autoreplace = (autoreplace != 0); 2473 } 2474 2475 /* 2476 * If the 'autoreplace' property is set, then post a resource notifying 2477 * the ZFS DE that it should not issue any faults for unopenable 2478 * devices. We also iterate over the vdevs, and post a sysevent for any 2479 * unopenable vdevs so that the normal autoreplace handler can take 2480 * over. 2481 */ 2482 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2483 spa_check_removed(spa->spa_root_vdev); 2484 /* 2485 * For the import case, this is done in spa_import(), because 2486 * at this point we're using the spare definitions from 2487 * the MOS config, not necessarily from the userland config. 2488 */ 2489 if (state != SPA_LOAD_IMPORT) { 2490 spa_aux_check_removed(&spa->spa_spares); 2491 spa_aux_check_removed(&spa->spa_l2cache); 2492 } 2493 } 2494 2495 /* 2496 * Load the vdev state for all toplevel vdevs. 2497 */ 2498 vdev_load(rvd); 2499 2500 /* 2501 * Propagate the leaf DTLs we just loaded all the way up the tree. 2502 */ 2503 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2504 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2505 spa_config_exit(spa, SCL_ALL, FTAG); 2506 2507 /* 2508 * Load the DDTs (dedup tables). 2509 */ 2510 error = ddt_load(spa); 2511 if (error != 0) 2512 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2513 2514 spa_update_dspace(spa); 2515 2516 /* 2517 * Validate the config, using the MOS config to fill in any 2518 * information which might be missing. If we fail to validate 2519 * the config then declare the pool unfit for use. If we're 2520 * assembling a pool from a split, the log is not transferred 2521 * over. 2522 */ 2523 if (type != SPA_IMPORT_ASSEMBLE) { 2524 nvlist_t *nvconfig; 2525 2526 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2527 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2528 2529 if (!spa_config_valid(spa, nvconfig)) { 2530 nvlist_free(nvconfig); 2531 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2532 ENXIO)); 2533 } 2534 nvlist_free(nvconfig); 2535 2536 /* 2537 * Now that we've validated the config, check the state of the 2538 * root vdev. If it can't be opened, it indicates one or 2539 * more toplevel vdevs are faulted. 2540 */ 2541 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2542 return (ENXIO); 2543 2544 if (spa_check_logs(spa)) { 2545 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2546 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2547 } 2548 } 2549 2550 if (missing_feat_write) { 2551 ASSERT(state == SPA_LOAD_TRYIMPORT); 2552 2553 /* 2554 * At this point, we know that we can open the pool in 2555 * read-only mode but not read-write mode. We now have enough 2556 * information and can return to userland. 2557 */ 2558 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2559 } 2560 2561 /* 2562 * We've successfully opened the pool, verify that we're ready 2563 * to start pushing transactions. 2564 */ 2565 if (state != SPA_LOAD_TRYIMPORT) { 2566 if (error = spa_load_verify(spa)) 2567 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2568 error)); 2569 } 2570 2571 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2572 spa->spa_load_max_txg == UINT64_MAX)) { 2573 dmu_tx_t *tx; 2574 int need_update = B_FALSE; 2575 2576 ASSERT(state != SPA_LOAD_TRYIMPORT); 2577 2578 /* 2579 * Claim log blocks that haven't been committed yet. 2580 * This must all happen in a single txg. 2581 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2582 * invoked from zil_claim_log_block()'s i/o done callback. 2583 * Price of rollback is that we abandon the log. 2584 */ 2585 spa->spa_claiming = B_TRUE; 2586 2587 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2588 spa_first_txg(spa)); 2589 (void) dmu_objset_find(spa_name(spa), 2590 zil_claim, tx, DS_FIND_CHILDREN); 2591 dmu_tx_commit(tx); 2592 2593 spa->spa_claiming = B_FALSE; 2594 2595 spa_set_log_state(spa, SPA_LOG_GOOD); 2596 spa->spa_sync_on = B_TRUE; 2597 txg_sync_start(spa->spa_dsl_pool); 2598 2599 /* 2600 * Wait for all claims to sync. We sync up to the highest 2601 * claimed log block birth time so that claimed log blocks 2602 * don't appear to be from the future. spa_claim_max_txg 2603 * will have been set for us by either zil_check_log_chain() 2604 * (invoked from spa_check_logs()) or zil_claim() above. 2605 */ 2606 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2607 2608 /* 2609 * If the config cache is stale, or we have uninitialized 2610 * metaslabs (see spa_vdev_add()), then update the config. 2611 * 2612 * If this is a verbatim import, trust the current 2613 * in-core spa_config and update the disk labels. 2614 */ 2615 if (config_cache_txg != spa->spa_config_txg || 2616 state == SPA_LOAD_IMPORT || 2617 state == SPA_LOAD_RECOVER || 2618 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2619 need_update = B_TRUE; 2620 2621 for (int c = 0; c < rvd->vdev_children; c++) 2622 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2623 need_update = B_TRUE; 2624 2625 /* 2626 * Update the config cache asychronously in case we're the 2627 * root pool, in which case the config cache isn't writable yet. 2628 */ 2629 if (need_update) 2630 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2631 2632 /* 2633 * Check all DTLs to see if anything needs resilvering. 2634 */ 2635 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2636 vdev_resilver_needed(rvd, NULL, NULL)) 2637 spa_async_request(spa, SPA_ASYNC_RESILVER); 2638 2639 /* 2640 * Log the fact that we booted up (so that we can detect if 2641 * we rebooted in the middle of an operation). 2642 */ 2643 spa_history_log_version(spa, "open"); 2644 2645 /* 2646 * Delete any inconsistent datasets. 2647 */ 2648 (void) dmu_objset_find(spa_name(spa), 2649 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2650 2651 /* 2652 * Clean up any stale temporary dataset userrefs. 2653 */ 2654 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2655 } 2656 2657 return (0); 2658} 2659 2660static int 2661spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2662{ 2663 int mode = spa->spa_mode; 2664 2665 spa_unload(spa); 2666 spa_deactivate(spa); 2667 2668 spa->spa_load_max_txg--; 2669 2670 spa_activate(spa, mode); 2671 spa_async_suspend(spa); 2672 2673 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2674} 2675 2676/* 2677 * If spa_load() fails this function will try loading prior txg's. If 2678 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2679 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2680 * function will not rewind the pool and will return the same error as 2681 * spa_load(). 2682 */ 2683static int 2684spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2685 uint64_t max_request, int rewind_flags) 2686{ 2687 nvlist_t *loadinfo = NULL; 2688 nvlist_t *config = NULL; 2689 int load_error, rewind_error; 2690 uint64_t safe_rewind_txg; 2691 uint64_t min_txg; 2692 2693 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2694 spa->spa_load_max_txg = spa->spa_load_txg; 2695 spa_set_log_state(spa, SPA_LOG_CLEAR); 2696 } else { 2697 spa->spa_load_max_txg = max_request; 2698 } 2699 2700 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2701 mosconfig); 2702 if (load_error == 0) 2703 return (0); 2704 2705 if (spa->spa_root_vdev != NULL) 2706 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2707 2708 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2709 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2710 2711 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2712 nvlist_free(config); 2713 return (load_error); 2714 } 2715 2716 if (state == SPA_LOAD_RECOVER) { 2717 /* Price of rolling back is discarding txgs, including log */ 2718 spa_set_log_state(spa, SPA_LOG_CLEAR); 2719 } else { 2720 /* 2721 * If we aren't rolling back save the load info from our first 2722 * import attempt so that we can restore it after attempting 2723 * to rewind. 2724 */ 2725 loadinfo = spa->spa_load_info; 2726 spa->spa_load_info = fnvlist_alloc(); 2727 } 2728 2729 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2730 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2731 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2732 TXG_INITIAL : safe_rewind_txg; 2733 2734 /* 2735 * Continue as long as we're finding errors, we're still within 2736 * the acceptable rewind range, and we're still finding uberblocks 2737 */ 2738 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2739 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2740 if (spa->spa_load_max_txg < safe_rewind_txg) 2741 spa->spa_extreme_rewind = B_TRUE; 2742 rewind_error = spa_load_retry(spa, state, mosconfig); 2743 } 2744 2745 spa->spa_extreme_rewind = B_FALSE; 2746 spa->spa_load_max_txg = UINT64_MAX; 2747 2748 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2749 spa_config_set(spa, config); 2750 2751 if (state == SPA_LOAD_RECOVER) { 2752 ASSERT3P(loadinfo, ==, NULL); 2753 return (rewind_error); 2754 } else { 2755 /* Store the rewind info as part of the initial load info */ 2756 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2757 spa->spa_load_info); 2758 2759 /* Restore the initial load info */ 2760 fnvlist_free(spa->spa_load_info); 2761 spa->spa_load_info = loadinfo; 2762 2763 return (load_error); 2764 } 2765} 2766 2767/* 2768 * Pool Open/Import 2769 * 2770 * The import case is identical to an open except that the configuration is sent 2771 * down from userland, instead of grabbed from the configuration cache. For the 2772 * case of an open, the pool configuration will exist in the 2773 * POOL_STATE_UNINITIALIZED state. 2774 * 2775 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2776 * the same time open the pool, without having to keep around the spa_t in some 2777 * ambiguous state. 2778 */ 2779static int 2780spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2781 nvlist_t **config) 2782{ 2783 spa_t *spa; 2784 spa_load_state_t state = SPA_LOAD_OPEN; 2785 int error; 2786 int locked = B_FALSE; 2787 int firstopen = B_FALSE; 2788 2789 *spapp = NULL; 2790 2791 /* 2792 * As disgusting as this is, we need to support recursive calls to this 2793 * function because dsl_dir_open() is called during spa_load(), and ends 2794 * up calling spa_open() again. The real fix is to figure out how to 2795 * avoid dsl_dir_open() calling this in the first place. 2796 */ 2797 if (mutex_owner(&spa_namespace_lock) != curthread) { 2798 mutex_enter(&spa_namespace_lock); 2799 locked = B_TRUE; 2800 } 2801 2802 if ((spa = spa_lookup(pool)) == NULL) { 2803 if (locked) 2804 mutex_exit(&spa_namespace_lock); 2805 return (ENOENT); 2806 } 2807 2808 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2809 zpool_rewind_policy_t policy; 2810 2811 firstopen = B_TRUE; 2812 2813 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2814 &policy); 2815 if (policy.zrp_request & ZPOOL_DO_REWIND) 2816 state = SPA_LOAD_RECOVER; 2817 2818 spa_activate(spa, spa_mode_global); 2819 2820 if (state != SPA_LOAD_RECOVER) 2821 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2822 2823 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2824 policy.zrp_request); 2825 2826 if (error == EBADF) { 2827 /* 2828 * If vdev_validate() returns failure (indicated by 2829 * EBADF), it indicates that one of the vdevs indicates 2830 * that the pool has been exported or destroyed. If 2831 * this is the case, the config cache is out of sync and 2832 * we should remove the pool from the namespace. 2833 */ 2834 spa_unload(spa); 2835 spa_deactivate(spa); 2836 spa_config_sync(spa, B_TRUE, B_TRUE); 2837 spa_remove(spa); 2838 if (locked) 2839 mutex_exit(&spa_namespace_lock); 2840 return (ENOENT); 2841 } 2842 2843 if (error) { 2844 /* 2845 * We can't open the pool, but we still have useful 2846 * information: the state of each vdev after the 2847 * attempted vdev_open(). Return this to the user. 2848 */ 2849 if (config != NULL && spa->spa_config) { 2850 VERIFY(nvlist_dup(spa->spa_config, config, 2851 KM_SLEEP) == 0); 2852 VERIFY(nvlist_add_nvlist(*config, 2853 ZPOOL_CONFIG_LOAD_INFO, 2854 spa->spa_load_info) == 0); 2855 } 2856 spa_unload(spa); 2857 spa_deactivate(spa); 2858 spa->spa_last_open_failed = error; 2859 if (locked) 2860 mutex_exit(&spa_namespace_lock); 2861 *spapp = NULL; 2862 return (error); 2863 } 2864 } 2865 2866 spa_open_ref(spa, tag); 2867 2868 if (config != NULL) 2869 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2870 2871 /* 2872 * If we've recovered the pool, pass back any information we 2873 * gathered while doing the load. 2874 */ 2875 if (state == SPA_LOAD_RECOVER) { 2876 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 2877 spa->spa_load_info) == 0); 2878 } 2879 2880 if (locked) { 2881 spa->spa_last_open_failed = 0; 2882 spa->spa_last_ubsync_txg = 0; 2883 spa->spa_load_txg = 0; 2884 mutex_exit(&spa_namespace_lock); 2885#ifdef __FreeBSD__ 2886#ifdef _KERNEL 2887 if (firstopen) 2888 zvol_create_minors(spa->spa_name); 2889#endif 2890#endif 2891 } 2892 2893 *spapp = spa; 2894 2895 return (0); 2896} 2897 2898int 2899spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2900 nvlist_t **config) 2901{ 2902 return (spa_open_common(name, spapp, tag, policy, config)); 2903} 2904 2905int 2906spa_open(const char *name, spa_t **spapp, void *tag) 2907{ 2908 return (spa_open_common(name, spapp, tag, NULL, NULL)); 2909} 2910 2911/* 2912 * Lookup the given spa_t, incrementing the inject count in the process, 2913 * preventing it from being exported or destroyed. 2914 */ 2915spa_t * 2916spa_inject_addref(char *name) 2917{ 2918 spa_t *spa; 2919 2920 mutex_enter(&spa_namespace_lock); 2921 if ((spa = spa_lookup(name)) == NULL) { 2922 mutex_exit(&spa_namespace_lock); 2923 return (NULL); 2924 } 2925 spa->spa_inject_ref++; 2926 mutex_exit(&spa_namespace_lock); 2927 2928 return (spa); 2929} 2930 2931void 2932spa_inject_delref(spa_t *spa) 2933{ 2934 mutex_enter(&spa_namespace_lock); 2935 spa->spa_inject_ref--; 2936 mutex_exit(&spa_namespace_lock); 2937} 2938 2939/* 2940 * Add spares device information to the nvlist. 2941 */ 2942static void 2943spa_add_spares(spa_t *spa, nvlist_t *config) 2944{ 2945 nvlist_t **spares; 2946 uint_t i, nspares; 2947 nvlist_t *nvroot; 2948 uint64_t guid; 2949 vdev_stat_t *vs; 2950 uint_t vsc; 2951 uint64_t pool; 2952 2953 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 2954 2955 if (spa->spa_spares.sav_count == 0) 2956 return; 2957 2958 VERIFY(nvlist_lookup_nvlist(config, 2959 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2960 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 2961 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2962 if (nspares != 0) { 2963 VERIFY(nvlist_add_nvlist_array(nvroot, 2964 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 2965 VERIFY(nvlist_lookup_nvlist_array(nvroot, 2966 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2967 2968 /* 2969 * Go through and find any spares which have since been 2970 * repurposed as an active spare. If this is the case, update 2971 * their status appropriately. 2972 */ 2973 for (i = 0; i < nspares; i++) { 2974 VERIFY(nvlist_lookup_uint64(spares[i], 2975 ZPOOL_CONFIG_GUID, &guid) == 0); 2976 if (spa_spare_exists(guid, &pool, NULL) && 2977 pool != 0ULL) { 2978 VERIFY(nvlist_lookup_uint64_array( 2979 spares[i], ZPOOL_CONFIG_VDEV_STATS, 2980 (uint64_t **)&vs, &vsc) == 0); 2981 vs->vs_state = VDEV_STATE_CANT_OPEN; 2982 vs->vs_aux = VDEV_AUX_SPARED; 2983 } 2984 } 2985 } 2986} 2987 2988/* 2989 * Add l2cache device information to the nvlist, including vdev stats. 2990 */ 2991static void 2992spa_add_l2cache(spa_t *spa, nvlist_t *config) 2993{ 2994 nvlist_t **l2cache; 2995 uint_t i, j, nl2cache; 2996 nvlist_t *nvroot; 2997 uint64_t guid; 2998 vdev_t *vd; 2999 vdev_stat_t *vs; 3000 uint_t vsc; 3001 3002 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3003 3004 if (spa->spa_l2cache.sav_count == 0) 3005 return; 3006 3007 VERIFY(nvlist_lookup_nvlist(config, 3008 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3009 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3010 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3011 if (nl2cache != 0) { 3012 VERIFY(nvlist_add_nvlist_array(nvroot, 3013 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3014 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3015 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3016 3017 /* 3018 * Update level 2 cache device stats. 3019 */ 3020 3021 for (i = 0; i < nl2cache; i++) { 3022 VERIFY(nvlist_lookup_uint64(l2cache[i], 3023 ZPOOL_CONFIG_GUID, &guid) == 0); 3024 3025 vd = NULL; 3026 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3027 if (guid == 3028 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3029 vd = spa->spa_l2cache.sav_vdevs[j]; 3030 break; 3031 } 3032 } 3033 ASSERT(vd != NULL); 3034 3035 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3036 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3037 == 0); 3038 vdev_get_stats(vd, vs); 3039 } 3040 } 3041} 3042 3043static void 3044spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3045{ 3046 nvlist_t *features; 3047 zap_cursor_t zc; 3048 zap_attribute_t za; 3049 3050 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3051 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3052 3053 if (spa->spa_feat_for_read_obj != 0) { 3054 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3055 spa->spa_feat_for_read_obj); 3056 zap_cursor_retrieve(&zc, &za) == 0; 3057 zap_cursor_advance(&zc)) { 3058 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3059 za.za_num_integers == 1); 3060 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3061 za.za_first_integer)); 3062 } 3063 zap_cursor_fini(&zc); 3064 } 3065 3066 if (spa->spa_feat_for_write_obj != 0) { 3067 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3068 spa->spa_feat_for_write_obj); 3069 zap_cursor_retrieve(&zc, &za) == 0; 3070 zap_cursor_advance(&zc)) { 3071 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3072 za.za_num_integers == 1); 3073 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3074 za.za_first_integer)); 3075 } 3076 zap_cursor_fini(&zc); 3077 } 3078 3079 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3080 features) == 0); 3081 nvlist_free(features); 3082} 3083 3084int 3085spa_get_stats(const char *name, nvlist_t **config, 3086 char *altroot, size_t buflen) 3087{ 3088 int error; 3089 spa_t *spa; 3090 3091 *config = NULL; 3092 error = spa_open_common(name, &spa, FTAG, NULL, config); 3093 3094 if (spa != NULL) { 3095 /* 3096 * This still leaves a window of inconsistency where the spares 3097 * or l2cache devices could change and the config would be 3098 * self-inconsistent. 3099 */ 3100 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3101 3102 if (*config != NULL) { 3103 uint64_t loadtimes[2]; 3104 3105 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3106 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3107 VERIFY(nvlist_add_uint64_array(*config, 3108 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3109 3110 VERIFY(nvlist_add_uint64(*config, 3111 ZPOOL_CONFIG_ERRCOUNT, 3112 spa_get_errlog_size(spa)) == 0); 3113 3114 if (spa_suspended(spa)) 3115 VERIFY(nvlist_add_uint64(*config, 3116 ZPOOL_CONFIG_SUSPENDED, 3117 spa->spa_failmode) == 0); 3118 3119 spa_add_spares(spa, *config); 3120 spa_add_l2cache(spa, *config); 3121 spa_add_feature_stats(spa, *config); 3122 } 3123 } 3124 3125 /* 3126 * We want to get the alternate root even for faulted pools, so we cheat 3127 * and call spa_lookup() directly. 3128 */ 3129 if (altroot) { 3130 if (spa == NULL) { 3131 mutex_enter(&spa_namespace_lock); 3132 spa = spa_lookup(name); 3133 if (spa) 3134 spa_altroot(spa, altroot, buflen); 3135 else 3136 altroot[0] = '\0'; 3137 spa = NULL; 3138 mutex_exit(&spa_namespace_lock); 3139 } else { 3140 spa_altroot(spa, altroot, buflen); 3141 } 3142 } 3143 3144 if (spa != NULL) { 3145 spa_config_exit(spa, SCL_CONFIG, FTAG); 3146 spa_close(spa, FTAG); 3147 } 3148 3149 return (error); 3150} 3151 3152/* 3153 * Validate that the auxiliary device array is well formed. We must have an 3154 * array of nvlists, each which describes a valid leaf vdev. If this is an 3155 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3156 * specified, as long as they are well-formed. 3157 */ 3158static int 3159spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3160 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3161 vdev_labeltype_t label) 3162{ 3163 nvlist_t **dev; 3164 uint_t i, ndev; 3165 vdev_t *vd; 3166 int error; 3167 3168 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3169 3170 /* 3171 * It's acceptable to have no devs specified. 3172 */ 3173 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3174 return (0); 3175 3176 if (ndev == 0) 3177 return (EINVAL); 3178 3179 /* 3180 * Make sure the pool is formatted with a version that supports this 3181 * device type. 3182 */ 3183 if (spa_version(spa) < version) 3184 return (ENOTSUP); 3185 3186 /* 3187 * Set the pending device list so we correctly handle device in-use 3188 * checking. 3189 */ 3190 sav->sav_pending = dev; 3191 sav->sav_npending = ndev; 3192 3193 for (i = 0; i < ndev; i++) { 3194 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3195 mode)) != 0) 3196 goto out; 3197 3198 if (!vd->vdev_ops->vdev_op_leaf) { 3199 vdev_free(vd); 3200 error = EINVAL; 3201 goto out; 3202 } 3203 3204 /* 3205 * The L2ARC currently only supports disk devices in 3206 * kernel context. For user-level testing, we allow it. 3207 */ 3208#ifdef _KERNEL 3209 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3210 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3211 error = ENOTBLK; 3212 vdev_free(vd); 3213 goto out; 3214 } 3215#endif 3216 vd->vdev_top = vd; 3217 3218 if ((error = vdev_open(vd)) == 0 && 3219 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3220 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3221 vd->vdev_guid) == 0); 3222 } 3223 3224 vdev_free(vd); 3225 3226 if (error && 3227 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3228 goto out; 3229 else 3230 error = 0; 3231 } 3232 3233out: 3234 sav->sav_pending = NULL; 3235 sav->sav_npending = 0; 3236 return (error); 3237} 3238 3239static int 3240spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3241{ 3242 int error; 3243 3244 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3245 3246 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3247 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3248 VDEV_LABEL_SPARE)) != 0) { 3249 return (error); 3250 } 3251 3252 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3253 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3254 VDEV_LABEL_L2CACHE)); 3255} 3256 3257static void 3258spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3259 const char *config) 3260{ 3261 int i; 3262 3263 if (sav->sav_config != NULL) { 3264 nvlist_t **olddevs; 3265 uint_t oldndevs; 3266 nvlist_t **newdevs; 3267 3268 /* 3269 * Generate new dev list by concatentating with the 3270 * current dev list. 3271 */ 3272 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3273 &olddevs, &oldndevs) == 0); 3274 3275 newdevs = kmem_alloc(sizeof (void *) * 3276 (ndevs + oldndevs), KM_SLEEP); 3277 for (i = 0; i < oldndevs; i++) 3278 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3279 KM_SLEEP) == 0); 3280 for (i = 0; i < ndevs; i++) 3281 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3282 KM_SLEEP) == 0); 3283 3284 VERIFY(nvlist_remove(sav->sav_config, config, 3285 DATA_TYPE_NVLIST_ARRAY) == 0); 3286 3287 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3288 config, newdevs, ndevs + oldndevs) == 0); 3289 for (i = 0; i < oldndevs + ndevs; i++) 3290 nvlist_free(newdevs[i]); 3291 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3292 } else { 3293 /* 3294 * Generate a new dev list. 3295 */ 3296 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3297 KM_SLEEP) == 0); 3298 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3299 devs, ndevs) == 0); 3300 } 3301} 3302 3303/* 3304 * Stop and drop level 2 ARC devices 3305 */ 3306void 3307spa_l2cache_drop(spa_t *spa) 3308{ 3309 vdev_t *vd; 3310 int i; 3311 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3312 3313 for (i = 0; i < sav->sav_count; i++) { 3314 uint64_t pool; 3315 3316 vd = sav->sav_vdevs[i]; 3317 ASSERT(vd != NULL); 3318 3319 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3320 pool != 0ULL && l2arc_vdev_present(vd)) 3321 l2arc_remove_vdev(vd); 3322 } 3323} 3324 3325/* 3326 * Pool Creation 3327 */ 3328int 3329spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3330 nvlist_t *zplprops) 3331{ 3332 spa_t *spa; 3333 char *altroot = NULL; 3334 vdev_t *rvd; 3335 dsl_pool_t *dp; 3336 dmu_tx_t *tx; 3337 int error = 0; 3338 uint64_t txg = TXG_INITIAL; 3339 nvlist_t **spares, **l2cache; 3340 uint_t nspares, nl2cache; 3341 uint64_t version, obj; 3342 boolean_t has_features; 3343 3344 /* 3345 * If this pool already exists, return failure. 3346 */ 3347 mutex_enter(&spa_namespace_lock); 3348 if (spa_lookup(pool) != NULL) { 3349 mutex_exit(&spa_namespace_lock); 3350 return (EEXIST); 3351 } 3352 3353 /* 3354 * Allocate a new spa_t structure. 3355 */ 3356 (void) nvlist_lookup_string(props, 3357 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3358 spa = spa_add(pool, NULL, altroot); 3359 spa_activate(spa, spa_mode_global); 3360 3361 if (props && (error = spa_prop_validate(spa, props))) { 3362 spa_deactivate(spa); 3363 spa_remove(spa); 3364 mutex_exit(&spa_namespace_lock); 3365 return (error); 3366 } 3367 3368 has_features = B_FALSE; 3369 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3370 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3371 if (zpool_prop_feature(nvpair_name(elem))) 3372 has_features = B_TRUE; 3373 } 3374 3375 if (has_features || nvlist_lookup_uint64(props, 3376 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3377 version = SPA_VERSION; 3378 } 3379 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3380 3381 spa->spa_first_txg = txg; 3382 spa->spa_uberblock.ub_txg = txg - 1; 3383 spa->spa_uberblock.ub_version = version; 3384 spa->spa_ubsync = spa->spa_uberblock; 3385 3386 /* 3387 * Create "The Godfather" zio to hold all async IOs 3388 */ 3389 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 3390 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 3391 3392 /* 3393 * Create the root vdev. 3394 */ 3395 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3396 3397 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3398 3399 ASSERT(error != 0 || rvd != NULL); 3400 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3401 3402 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3403 error = EINVAL; 3404 3405 if (error == 0 && 3406 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3407 (error = spa_validate_aux(spa, nvroot, txg, 3408 VDEV_ALLOC_ADD)) == 0) { 3409 for (int c = 0; c < rvd->vdev_children; c++) { 3410 vdev_metaslab_set_size(rvd->vdev_child[c]); 3411 vdev_expand(rvd->vdev_child[c], txg); 3412 } 3413 } 3414 3415 spa_config_exit(spa, SCL_ALL, FTAG); 3416 3417 if (error != 0) { 3418 spa_unload(spa); 3419 spa_deactivate(spa); 3420 spa_remove(spa); 3421 mutex_exit(&spa_namespace_lock); 3422 return (error); 3423 } 3424 3425 /* 3426 * Get the list of spares, if specified. 3427 */ 3428 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3429 &spares, &nspares) == 0) { 3430 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3431 KM_SLEEP) == 0); 3432 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3433 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3434 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3435 spa_load_spares(spa); 3436 spa_config_exit(spa, SCL_ALL, FTAG); 3437 spa->spa_spares.sav_sync = B_TRUE; 3438 } 3439 3440 /* 3441 * Get the list of level 2 cache devices, if specified. 3442 */ 3443 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3444 &l2cache, &nl2cache) == 0) { 3445 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3446 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3447 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3448 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3449 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3450 spa_load_l2cache(spa); 3451 spa_config_exit(spa, SCL_ALL, FTAG); 3452 spa->spa_l2cache.sav_sync = B_TRUE; 3453 } 3454 3455 spa->spa_is_initializing = B_TRUE; 3456 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3457 spa->spa_meta_objset = dp->dp_meta_objset; 3458 spa->spa_is_initializing = B_FALSE; 3459 3460 /* 3461 * Create DDTs (dedup tables). 3462 */ 3463 ddt_create(spa); 3464 3465 spa_update_dspace(spa); 3466 3467 tx = dmu_tx_create_assigned(dp, txg); 3468 3469 /* 3470 * Create the pool config object. 3471 */ 3472 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3473 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3474 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3475 3476 if (zap_add(spa->spa_meta_objset, 3477 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3478 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3479 cmn_err(CE_PANIC, "failed to add pool config"); 3480 } 3481 3482 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3483 spa_feature_create_zap_objects(spa, tx); 3484 3485 if (zap_add(spa->spa_meta_objset, 3486 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3487 sizeof (uint64_t), 1, &version, tx) != 0) { 3488 cmn_err(CE_PANIC, "failed to add pool version"); 3489 } 3490 3491 /* Newly created pools with the right version are always deflated. */ 3492 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3493 spa->spa_deflate = TRUE; 3494 if (zap_add(spa->spa_meta_objset, 3495 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3496 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3497 cmn_err(CE_PANIC, "failed to add deflate"); 3498 } 3499 } 3500 3501 /* 3502 * Create the deferred-free bpobj. Turn off compression 3503 * because sync-to-convergence takes longer if the blocksize 3504 * keeps changing. 3505 */ 3506 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3507 dmu_object_set_compress(spa->spa_meta_objset, obj, 3508 ZIO_COMPRESS_OFF, tx); 3509 if (zap_add(spa->spa_meta_objset, 3510 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3511 sizeof (uint64_t), 1, &obj, tx) != 0) { 3512 cmn_err(CE_PANIC, "failed to add bpobj"); 3513 } 3514 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3515 spa->spa_meta_objset, obj)); 3516 3517 /* 3518 * Create the pool's history object. 3519 */ 3520 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3521 spa_history_create_obj(spa, tx); 3522 3523 /* 3524 * Set pool properties. 3525 */ 3526 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3527 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3528 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3529 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3530 3531 if (props != NULL) { 3532 spa_configfile_set(spa, props, B_FALSE); 3533 spa_sync_props(props, tx); 3534 } 3535 3536 dmu_tx_commit(tx); 3537 3538 spa->spa_sync_on = B_TRUE; 3539 txg_sync_start(spa->spa_dsl_pool); 3540 3541 /* 3542 * We explicitly wait for the first transaction to complete so that our 3543 * bean counters are appropriately updated. 3544 */ 3545 txg_wait_synced(spa->spa_dsl_pool, txg); 3546 3547 spa_config_sync(spa, B_FALSE, B_TRUE); 3548 3549 spa_history_log_version(spa, "create"); 3550 3551 spa->spa_minref = refcount_count(&spa->spa_refcount); 3552 3553 mutex_exit(&spa_namespace_lock); 3554 3555 return (0); 3556} 3557 3558#ifdef _KERNEL 3559#if defined(sun) 3560/* 3561 * Get the root pool information from the root disk, then import the root pool 3562 * during the system boot up time. 3563 */ 3564extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3565 3566static nvlist_t * 3567spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3568{ 3569 nvlist_t *config; 3570 nvlist_t *nvtop, *nvroot; 3571 uint64_t pgid; 3572 3573 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3574 return (NULL); 3575 3576 /* 3577 * Add this top-level vdev to the child array. 3578 */ 3579 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3580 &nvtop) == 0); 3581 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3582 &pgid) == 0); 3583 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3584 3585 /* 3586 * Put this pool's top-level vdevs into a root vdev. 3587 */ 3588 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3589 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3590 VDEV_TYPE_ROOT) == 0); 3591 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3592 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3593 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3594 &nvtop, 1) == 0); 3595 3596 /* 3597 * Replace the existing vdev_tree with the new root vdev in 3598 * this pool's configuration (remove the old, add the new). 3599 */ 3600 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3601 nvlist_free(nvroot); 3602 return (config); 3603} 3604 3605/* 3606 * Walk the vdev tree and see if we can find a device with "better" 3607 * configuration. A configuration is "better" if the label on that 3608 * device has a more recent txg. 3609 */ 3610static void 3611spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3612{ 3613 for (int c = 0; c < vd->vdev_children; c++) 3614 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3615 3616 if (vd->vdev_ops->vdev_op_leaf) { 3617 nvlist_t *label; 3618 uint64_t label_txg; 3619 3620 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3621 &label) != 0) 3622 return; 3623 3624 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3625 &label_txg) == 0); 3626 3627 /* 3628 * Do we have a better boot device? 3629 */ 3630 if (label_txg > *txg) { 3631 *txg = label_txg; 3632 *avd = vd; 3633 } 3634 nvlist_free(label); 3635 } 3636} 3637 3638/* 3639 * Import a root pool. 3640 * 3641 * For x86. devpath_list will consist of devid and/or physpath name of 3642 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3643 * The GRUB "findroot" command will return the vdev we should boot. 3644 * 3645 * For Sparc, devpath_list consists the physpath name of the booting device 3646 * no matter the rootpool is a single device pool or a mirrored pool. 3647 * e.g. 3648 * "/pci@1f,0/ide@d/disk@0,0:a" 3649 */ 3650int 3651spa_import_rootpool(char *devpath, char *devid) 3652{ 3653 spa_t *spa; 3654 vdev_t *rvd, *bvd, *avd = NULL; 3655 nvlist_t *config, *nvtop; 3656 uint64_t guid, txg; 3657 char *pname; 3658 int error; 3659 3660 /* 3661 * Read the label from the boot device and generate a configuration. 3662 */ 3663 config = spa_generate_rootconf(devpath, devid, &guid); 3664#if defined(_OBP) && defined(_KERNEL) 3665 if (config == NULL) { 3666 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3667 /* iscsi boot */ 3668 get_iscsi_bootpath_phy(devpath); 3669 config = spa_generate_rootconf(devpath, devid, &guid); 3670 } 3671 } 3672#endif 3673 if (config == NULL) { 3674 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3675 devpath); 3676 return (EIO); 3677 } 3678 3679 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3680 &pname) == 0); 3681 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3682 3683 mutex_enter(&spa_namespace_lock); 3684 if ((spa = spa_lookup(pname)) != NULL) { 3685 /* 3686 * Remove the existing root pool from the namespace so that we 3687 * can replace it with the correct config we just read in. 3688 */ 3689 spa_remove(spa); 3690 } 3691 3692 spa = spa_add(pname, config, NULL); 3693 spa->spa_is_root = B_TRUE; 3694 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3695 3696 /* 3697 * Build up a vdev tree based on the boot device's label config. 3698 */ 3699 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3700 &nvtop) == 0); 3701 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3702 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3703 VDEV_ALLOC_ROOTPOOL); 3704 spa_config_exit(spa, SCL_ALL, FTAG); 3705 if (error) { 3706 mutex_exit(&spa_namespace_lock); 3707 nvlist_free(config); 3708 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3709 pname); 3710 return (error); 3711 } 3712 3713 /* 3714 * Get the boot vdev. 3715 */ 3716 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3717 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3718 (u_longlong_t)guid); 3719 error = ENOENT; 3720 goto out; 3721 } 3722 3723 /* 3724 * Determine if there is a better boot device. 3725 */ 3726 avd = bvd; 3727 spa_alt_rootvdev(rvd, &avd, &txg); 3728 if (avd != bvd) { 3729 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3730 "try booting from '%s'", avd->vdev_path); 3731 error = EINVAL; 3732 goto out; 3733 } 3734 3735 /* 3736 * If the boot device is part of a spare vdev then ensure that 3737 * we're booting off the active spare. 3738 */ 3739 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3740 !bvd->vdev_isspare) { 3741 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3742 "try booting from '%s'", 3743 bvd->vdev_parent-> 3744 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3745 error = EINVAL; 3746 goto out; 3747 } 3748 3749 error = 0; 3750out: 3751 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3752 vdev_free(rvd); 3753 spa_config_exit(spa, SCL_ALL, FTAG); 3754 mutex_exit(&spa_namespace_lock); 3755 3756 nvlist_free(config); 3757 return (error); 3758} 3759 3760#else 3761 3762extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3763 uint64_t *count); 3764 3765static nvlist_t * 3766spa_generate_rootconf(const char *name) 3767{ 3768 nvlist_t **configs, **tops; 3769 nvlist_t *config; 3770 nvlist_t *best_cfg, *nvtop, *nvroot; 3771 uint64_t *holes; 3772 uint64_t best_txg; 3773 uint64_t nchildren; 3774 uint64_t pgid; 3775 uint64_t count; 3776 uint64_t i; 3777 uint_t nholes; 3778 3779 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3780 return (NULL); 3781 3782 ASSERT3U(count, !=, 0); 3783 best_txg = 0; 3784 for (i = 0; i < count; i++) { 3785 uint64_t txg; 3786 3787 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 3788 &txg) == 0); 3789 if (txg > best_txg) { 3790 best_txg = txg; 3791 best_cfg = configs[i]; 3792 } 3793 } 3794 3795 /* 3796 * Multi-vdev root pool configuration discovery is not supported yet. 3797 */ 3798 nchildren = 1; 3799 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 3800 holes = NULL; 3801 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 3802 &holes, &nholes); 3803 3804 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 3805 for (i = 0; i < nchildren; i++) { 3806 if (i >= count) 3807 break; 3808 if (configs[i] == NULL) 3809 continue; 3810 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 3811 &nvtop) == 0); 3812 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 3813 } 3814 for (i = 0; holes != NULL && i < nholes; i++) { 3815 if (i >= nchildren) 3816 continue; 3817 if (tops[holes[i]] != NULL) 3818 continue; 3819 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 3820 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 3821 VDEV_TYPE_HOLE) == 0); 3822 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 3823 holes[i]) == 0); 3824 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 3825 0) == 0); 3826 } 3827 for (i = 0; i < nchildren; i++) { 3828 if (tops[i] != NULL) 3829 continue; 3830 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 3831 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 3832 VDEV_TYPE_MISSING) == 0); 3833 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 3834 i) == 0); 3835 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 3836 0) == 0); 3837 } 3838 3839 /* 3840 * Create pool config based on the best vdev config. 3841 */ 3842 nvlist_dup(best_cfg, &config, KM_SLEEP); 3843 3844 /* 3845 * Put this pool's top-level vdevs into a root vdev. 3846 */ 3847 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3848 &pgid) == 0); 3849 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3850 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3851 VDEV_TYPE_ROOT) == 0); 3852 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3853 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3854 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3855 tops, nchildren) == 0); 3856 3857 /* 3858 * Replace the existing vdev_tree with the new root vdev in 3859 * this pool's configuration (remove the old, add the new). 3860 */ 3861 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3862 3863 /* 3864 * Drop vdev config elements that should not be present at pool level. 3865 */ 3866 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 3867 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 3868 3869 for (i = 0; i < count; i++) 3870 nvlist_free(configs[i]); 3871 kmem_free(configs, count * sizeof(void *)); 3872 for (i = 0; i < nchildren; i++) 3873 nvlist_free(tops[i]); 3874 kmem_free(tops, nchildren * sizeof(void *)); 3875 nvlist_free(nvroot); 3876 return (config); 3877} 3878 3879int 3880spa_import_rootpool(const char *name) 3881{ 3882 spa_t *spa; 3883 vdev_t *rvd, *bvd, *avd = NULL; 3884 nvlist_t *config, *nvtop; 3885 uint64_t txg; 3886 char *pname; 3887 int error; 3888 3889 /* 3890 * Read the label from the boot device and generate a configuration. 3891 */ 3892 config = spa_generate_rootconf(name); 3893 3894 mutex_enter(&spa_namespace_lock); 3895 if (config != NULL) { 3896 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3897 &pname) == 0 && strcmp(name, pname) == 0); 3898 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 3899 == 0); 3900 3901 if ((spa = spa_lookup(pname)) != NULL) { 3902 /* 3903 * Remove the existing root pool from the namespace so 3904 * that we can replace it with the correct config 3905 * we just read in. 3906 */ 3907 spa_remove(spa); 3908 } 3909 spa = spa_add(pname, config, NULL); 3910 3911 /* 3912 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 3913 * via spa_version(). 3914 */ 3915 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 3916 &spa->spa_ubsync.ub_version) != 0) 3917 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 3918 } else if ((spa = spa_lookup(name)) == NULL) { 3919 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 3920 name); 3921 return (EIO); 3922 } else { 3923 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 3924 } 3925 spa->spa_is_root = B_TRUE; 3926 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3927 3928 /* 3929 * Build up a vdev tree based on the boot device's label config. 3930 */ 3931 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3932 &nvtop) == 0); 3933 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3934 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3935 VDEV_ALLOC_ROOTPOOL); 3936 spa_config_exit(spa, SCL_ALL, FTAG); 3937 if (error) { 3938 mutex_exit(&spa_namespace_lock); 3939 nvlist_free(config); 3940 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3941 pname); 3942 return (error); 3943 } 3944 3945 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3946 vdev_free(rvd); 3947 spa_config_exit(spa, SCL_ALL, FTAG); 3948 mutex_exit(&spa_namespace_lock); 3949 3950 nvlist_free(config); 3951 return (0); 3952} 3953 3954#endif /* sun */ 3955#endif 3956 3957/* 3958 * Import a non-root pool into the system. 3959 */ 3960int 3961spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3962{ 3963 spa_t *spa; 3964 char *altroot = NULL; 3965 spa_load_state_t state = SPA_LOAD_IMPORT; 3966 zpool_rewind_policy_t policy; 3967 uint64_t mode = spa_mode_global; 3968 uint64_t readonly = B_FALSE; 3969 int error; 3970 nvlist_t *nvroot; 3971 nvlist_t **spares, **l2cache; 3972 uint_t nspares, nl2cache; 3973 3974 /* 3975 * If a pool with this name exists, return failure. 3976 */ 3977 mutex_enter(&spa_namespace_lock); 3978 if (spa_lookup(pool) != NULL) { 3979 mutex_exit(&spa_namespace_lock); 3980 return (EEXIST); 3981 } 3982 3983 /* 3984 * Create and initialize the spa structure. 3985 */ 3986 (void) nvlist_lookup_string(props, 3987 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3988 (void) nvlist_lookup_uint64(props, 3989 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 3990 if (readonly) 3991 mode = FREAD; 3992 spa = spa_add(pool, config, altroot); 3993 spa->spa_import_flags = flags; 3994 3995 /* 3996 * Verbatim import - Take a pool and insert it into the namespace 3997 * as if it had been loaded at boot. 3998 */ 3999 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4000 if (props != NULL) 4001 spa_configfile_set(spa, props, B_FALSE); 4002 4003 spa_config_sync(spa, B_FALSE, B_TRUE); 4004 4005 mutex_exit(&spa_namespace_lock); 4006 spa_history_log_version(spa, "import"); 4007 4008 return (0); 4009 } 4010 4011 spa_activate(spa, mode); 4012 4013 /* 4014 * Don't start async tasks until we know everything is healthy. 4015 */ 4016 spa_async_suspend(spa); 4017 4018 zpool_get_rewind_policy(config, &policy); 4019 if (policy.zrp_request & ZPOOL_DO_REWIND) 4020 state = SPA_LOAD_RECOVER; 4021 4022 /* 4023 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4024 * because the user-supplied config is actually the one to trust when 4025 * doing an import. 4026 */ 4027 if (state != SPA_LOAD_RECOVER) 4028 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4029 4030 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4031 policy.zrp_request); 4032 4033 /* 4034 * Propagate anything learned while loading the pool and pass it 4035 * back to caller (i.e. rewind info, missing devices, etc). 4036 */ 4037 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4038 spa->spa_load_info) == 0); 4039 4040 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4041 /* 4042 * Toss any existing sparelist, as it doesn't have any validity 4043 * anymore, and conflicts with spa_has_spare(). 4044 */ 4045 if (spa->spa_spares.sav_config) { 4046 nvlist_free(spa->spa_spares.sav_config); 4047 spa->spa_spares.sav_config = NULL; 4048 spa_load_spares(spa); 4049 } 4050 if (spa->spa_l2cache.sav_config) { 4051 nvlist_free(spa->spa_l2cache.sav_config); 4052 spa->spa_l2cache.sav_config = NULL; 4053 spa_load_l2cache(spa); 4054 } 4055 4056 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4057 &nvroot) == 0); 4058 if (error == 0) 4059 error = spa_validate_aux(spa, nvroot, -1ULL, 4060 VDEV_ALLOC_SPARE); 4061 if (error == 0) 4062 error = spa_validate_aux(spa, nvroot, -1ULL, 4063 VDEV_ALLOC_L2CACHE); 4064 spa_config_exit(spa, SCL_ALL, FTAG); 4065 4066 if (props != NULL) 4067 spa_configfile_set(spa, props, B_FALSE); 4068 4069 if (error != 0 || (props && spa_writeable(spa) && 4070 (error = spa_prop_set(spa, props)))) { 4071 spa_unload(spa); 4072 spa_deactivate(spa); 4073 spa_remove(spa); 4074 mutex_exit(&spa_namespace_lock); 4075 return (error); 4076 } 4077 4078 spa_async_resume(spa); 4079 4080 /* 4081 * Override any spares and level 2 cache devices as specified by 4082 * the user, as these may have correct device names/devids, etc. 4083 */ 4084 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4085 &spares, &nspares) == 0) { 4086 if (spa->spa_spares.sav_config) 4087 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4088 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4089 else 4090 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4091 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4092 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4093 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4094 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4095 spa_load_spares(spa); 4096 spa_config_exit(spa, SCL_ALL, FTAG); 4097 spa->spa_spares.sav_sync = B_TRUE; 4098 } 4099 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4100 &l2cache, &nl2cache) == 0) { 4101 if (spa->spa_l2cache.sav_config) 4102 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4103 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4104 else 4105 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4106 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4107 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4108 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4109 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4110 spa_load_l2cache(spa); 4111 spa_config_exit(spa, SCL_ALL, FTAG); 4112 spa->spa_l2cache.sav_sync = B_TRUE; 4113 } 4114 4115 /* 4116 * Check for any removed devices. 4117 */ 4118 if (spa->spa_autoreplace) { 4119 spa_aux_check_removed(&spa->spa_spares); 4120 spa_aux_check_removed(&spa->spa_l2cache); 4121 } 4122 4123 if (spa_writeable(spa)) { 4124 /* 4125 * Update the config cache to include the newly-imported pool. 4126 */ 4127 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4128 } 4129 4130 /* 4131 * It's possible that the pool was expanded while it was exported. 4132 * We kick off an async task to handle this for us. 4133 */ 4134 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4135 4136 mutex_exit(&spa_namespace_lock); 4137 spa_history_log_version(spa, "import"); 4138 4139#ifdef __FreeBSD__ 4140#ifdef _KERNEL 4141 zvol_create_minors(pool); 4142#endif 4143#endif 4144 return (0); 4145} 4146 4147nvlist_t * 4148spa_tryimport(nvlist_t *tryconfig) 4149{ 4150 nvlist_t *config = NULL; 4151 char *poolname; 4152 spa_t *spa; 4153 uint64_t state; 4154 int error; 4155 4156 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4157 return (NULL); 4158 4159 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4160 return (NULL); 4161 4162 /* 4163 * Create and initialize the spa structure. 4164 */ 4165 mutex_enter(&spa_namespace_lock); 4166 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4167 spa_activate(spa, FREAD); 4168 4169 /* 4170 * Pass off the heavy lifting to spa_load(). 4171 * Pass TRUE for mosconfig because the user-supplied config 4172 * is actually the one to trust when doing an import. 4173 */ 4174 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4175 4176 /* 4177 * If 'tryconfig' was at least parsable, return the current config. 4178 */ 4179 if (spa->spa_root_vdev != NULL) { 4180 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4181 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4182 poolname) == 0); 4183 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4184 state) == 0); 4185 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4186 spa->spa_uberblock.ub_timestamp) == 0); 4187 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4188 spa->spa_load_info) == 0); 4189 4190 /* 4191 * If the bootfs property exists on this pool then we 4192 * copy it out so that external consumers can tell which 4193 * pools are bootable. 4194 */ 4195 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4196 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4197 4198 /* 4199 * We have to play games with the name since the 4200 * pool was opened as TRYIMPORT_NAME. 4201 */ 4202 if (dsl_dsobj_to_dsname(spa_name(spa), 4203 spa->spa_bootfs, tmpname) == 0) { 4204 char *cp; 4205 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4206 4207 cp = strchr(tmpname, '/'); 4208 if (cp == NULL) { 4209 (void) strlcpy(dsname, tmpname, 4210 MAXPATHLEN); 4211 } else { 4212 (void) snprintf(dsname, MAXPATHLEN, 4213 "%s/%s", poolname, ++cp); 4214 } 4215 VERIFY(nvlist_add_string(config, 4216 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4217 kmem_free(dsname, MAXPATHLEN); 4218 } 4219 kmem_free(tmpname, MAXPATHLEN); 4220 } 4221 4222 /* 4223 * Add the list of hot spares and level 2 cache devices. 4224 */ 4225 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4226 spa_add_spares(spa, config); 4227 spa_add_l2cache(spa, config); 4228 spa_config_exit(spa, SCL_CONFIG, FTAG); 4229 } 4230 4231 spa_unload(spa); 4232 spa_deactivate(spa); 4233 spa_remove(spa); 4234 mutex_exit(&spa_namespace_lock); 4235 4236 return (config); 4237} 4238 4239/* 4240 * Pool export/destroy 4241 * 4242 * The act of destroying or exporting a pool is very simple. We make sure there 4243 * is no more pending I/O and any references to the pool are gone. Then, we 4244 * update the pool state and sync all the labels to disk, removing the 4245 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4246 * we don't sync the labels or remove the configuration cache. 4247 */ 4248static int 4249spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4250 boolean_t force, boolean_t hardforce) 4251{ 4252 spa_t *spa; 4253 4254 if (oldconfig) 4255 *oldconfig = NULL; 4256 4257 if (!(spa_mode_global & FWRITE)) 4258 return (EROFS); 4259 4260 mutex_enter(&spa_namespace_lock); 4261 if ((spa = spa_lookup(pool)) == NULL) { 4262 mutex_exit(&spa_namespace_lock); 4263 return (ENOENT); 4264 } 4265 4266 /* 4267 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4268 * reacquire the namespace lock, and see if we can export. 4269 */ 4270 spa_open_ref(spa, FTAG); 4271 mutex_exit(&spa_namespace_lock); 4272 spa_async_suspend(spa); 4273 mutex_enter(&spa_namespace_lock); 4274 spa_close(spa, FTAG); 4275 4276 /* 4277 * The pool will be in core if it's openable, 4278 * in which case we can modify its state. 4279 */ 4280 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4281 /* 4282 * Objsets may be open only because they're dirty, so we 4283 * have to force it to sync before checking spa_refcnt. 4284 */ 4285 txg_wait_synced(spa->spa_dsl_pool, 0); 4286 4287 /* 4288 * A pool cannot be exported or destroyed if there are active 4289 * references. If we are resetting a pool, allow references by 4290 * fault injection handlers. 4291 */ 4292 if (!spa_refcount_zero(spa) || 4293 (spa->spa_inject_ref != 0 && 4294 new_state != POOL_STATE_UNINITIALIZED)) { 4295 spa_async_resume(spa); 4296 mutex_exit(&spa_namespace_lock); 4297 return (EBUSY); 4298 } 4299 4300 /* 4301 * A pool cannot be exported if it has an active shared spare. 4302 * This is to prevent other pools stealing the active spare 4303 * from an exported pool. At user's own will, such pool can 4304 * be forcedly exported. 4305 */ 4306 if (!force && new_state == POOL_STATE_EXPORTED && 4307 spa_has_active_shared_spare(spa)) { 4308 spa_async_resume(spa); 4309 mutex_exit(&spa_namespace_lock); 4310 return (EXDEV); 4311 } 4312 4313 /* 4314 * We want this to be reflected on every label, 4315 * so mark them all dirty. spa_unload() will do the 4316 * final sync that pushes these changes out. 4317 */ 4318 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4319 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4320 spa->spa_state = new_state; 4321 spa->spa_final_txg = spa_last_synced_txg(spa) + 4322 TXG_DEFER_SIZE + 1; 4323 vdev_config_dirty(spa->spa_root_vdev); 4324 spa_config_exit(spa, SCL_ALL, FTAG); 4325 } 4326 } 4327 4328 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4329 4330 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4331 spa_unload(spa); 4332 spa_deactivate(spa); 4333 } 4334 4335 if (oldconfig && spa->spa_config) 4336 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4337 4338 if (new_state != POOL_STATE_UNINITIALIZED) { 4339 if (!hardforce) 4340 spa_config_sync(spa, B_TRUE, B_TRUE); 4341 spa_remove(spa); 4342 } 4343 mutex_exit(&spa_namespace_lock); 4344 4345 return (0); 4346} 4347 4348/* 4349 * Destroy a storage pool. 4350 */ 4351int 4352spa_destroy(char *pool) 4353{ 4354 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4355 B_FALSE, B_FALSE)); 4356} 4357 4358/* 4359 * Export a storage pool. 4360 */ 4361int 4362spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4363 boolean_t hardforce) 4364{ 4365 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4366 force, hardforce)); 4367} 4368 4369/* 4370 * Similar to spa_export(), this unloads the spa_t without actually removing it 4371 * from the namespace in any way. 4372 */ 4373int 4374spa_reset(char *pool) 4375{ 4376 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4377 B_FALSE, B_FALSE)); 4378} 4379 4380/* 4381 * ========================================================================== 4382 * Device manipulation 4383 * ========================================================================== 4384 */ 4385 4386/* 4387 * Add a device to a storage pool. 4388 */ 4389int 4390spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4391{ 4392 uint64_t txg, id; 4393 int error; 4394 vdev_t *rvd = spa->spa_root_vdev; 4395 vdev_t *vd, *tvd; 4396 nvlist_t **spares, **l2cache; 4397 uint_t nspares, nl2cache; 4398 4399 ASSERT(spa_writeable(spa)); 4400 4401 txg = spa_vdev_enter(spa); 4402 4403 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4404 VDEV_ALLOC_ADD)) != 0) 4405 return (spa_vdev_exit(spa, NULL, txg, error)); 4406 4407 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4408 4409 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4410 &nspares) != 0) 4411 nspares = 0; 4412 4413 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4414 &nl2cache) != 0) 4415 nl2cache = 0; 4416 4417 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4418 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4419 4420 if (vd->vdev_children != 0 && 4421 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4422 return (spa_vdev_exit(spa, vd, txg, error)); 4423 4424 /* 4425 * We must validate the spares and l2cache devices after checking the 4426 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4427 */ 4428 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4429 return (spa_vdev_exit(spa, vd, txg, error)); 4430 4431 /* 4432 * Transfer each new top-level vdev from vd to rvd. 4433 */ 4434 for (int c = 0; c < vd->vdev_children; c++) { 4435 4436 /* 4437 * Set the vdev id to the first hole, if one exists. 4438 */ 4439 for (id = 0; id < rvd->vdev_children; id++) { 4440 if (rvd->vdev_child[id]->vdev_ishole) { 4441 vdev_free(rvd->vdev_child[id]); 4442 break; 4443 } 4444 } 4445 tvd = vd->vdev_child[c]; 4446 vdev_remove_child(vd, tvd); 4447 tvd->vdev_id = id; 4448 vdev_add_child(rvd, tvd); 4449 vdev_config_dirty(tvd); 4450 } 4451 4452 if (nspares != 0) { 4453 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4454 ZPOOL_CONFIG_SPARES); 4455 spa_load_spares(spa); 4456 spa->spa_spares.sav_sync = B_TRUE; 4457 } 4458 4459 if (nl2cache != 0) { 4460 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4461 ZPOOL_CONFIG_L2CACHE); 4462 spa_load_l2cache(spa); 4463 spa->spa_l2cache.sav_sync = B_TRUE; 4464 } 4465 4466 /* 4467 * We have to be careful when adding new vdevs to an existing pool. 4468 * If other threads start allocating from these vdevs before we 4469 * sync the config cache, and we lose power, then upon reboot we may 4470 * fail to open the pool because there are DVAs that the config cache 4471 * can't translate. Therefore, we first add the vdevs without 4472 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4473 * and then let spa_config_update() initialize the new metaslabs. 4474 * 4475 * spa_load() checks for added-but-not-initialized vdevs, so that 4476 * if we lose power at any point in this sequence, the remaining 4477 * steps will be completed the next time we load the pool. 4478 */ 4479 (void) spa_vdev_exit(spa, vd, txg, 0); 4480 4481 mutex_enter(&spa_namespace_lock); 4482 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4483 mutex_exit(&spa_namespace_lock); 4484 4485 return (0); 4486} 4487 4488/* 4489 * Attach a device to a mirror. The arguments are the path to any device 4490 * in the mirror, and the nvroot for the new device. If the path specifies 4491 * a device that is not mirrored, we automatically insert the mirror vdev. 4492 * 4493 * If 'replacing' is specified, the new device is intended to replace the 4494 * existing device; in this case the two devices are made into their own 4495 * mirror using the 'replacing' vdev, which is functionally identical to 4496 * the mirror vdev (it actually reuses all the same ops) but has a few 4497 * extra rules: you can't attach to it after it's been created, and upon 4498 * completion of resilvering, the first disk (the one being replaced) 4499 * is automatically detached. 4500 */ 4501int 4502spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4503{ 4504 uint64_t txg, dtl_max_txg; 4505 vdev_t *rvd = spa->spa_root_vdev; 4506 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4507 vdev_ops_t *pvops; 4508 char *oldvdpath, *newvdpath; 4509 int newvd_isspare; 4510 int error; 4511 4512 ASSERT(spa_writeable(spa)); 4513 4514 txg = spa_vdev_enter(spa); 4515 4516 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4517 4518 if (oldvd == NULL) 4519 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4520 4521 if (!oldvd->vdev_ops->vdev_op_leaf) 4522 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4523 4524 pvd = oldvd->vdev_parent; 4525 4526 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4527 VDEV_ALLOC_ATTACH)) != 0) 4528 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4529 4530 if (newrootvd->vdev_children != 1) 4531 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4532 4533 newvd = newrootvd->vdev_child[0]; 4534 4535 if (!newvd->vdev_ops->vdev_op_leaf) 4536 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4537 4538 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4539 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4540 4541 /* 4542 * Spares can't replace logs 4543 */ 4544 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4545 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4546 4547 if (!replacing) { 4548 /* 4549 * For attach, the only allowable parent is a mirror or the root 4550 * vdev. 4551 */ 4552 if (pvd->vdev_ops != &vdev_mirror_ops && 4553 pvd->vdev_ops != &vdev_root_ops) 4554 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4555 4556 pvops = &vdev_mirror_ops; 4557 } else { 4558 /* 4559 * Active hot spares can only be replaced by inactive hot 4560 * spares. 4561 */ 4562 if (pvd->vdev_ops == &vdev_spare_ops && 4563 oldvd->vdev_isspare && 4564 !spa_has_spare(spa, newvd->vdev_guid)) 4565 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4566 4567 /* 4568 * If the source is a hot spare, and the parent isn't already a 4569 * spare, then we want to create a new hot spare. Otherwise, we 4570 * want to create a replacing vdev. The user is not allowed to 4571 * attach to a spared vdev child unless the 'isspare' state is 4572 * the same (spare replaces spare, non-spare replaces 4573 * non-spare). 4574 */ 4575 if (pvd->vdev_ops == &vdev_replacing_ops && 4576 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4577 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4578 } else if (pvd->vdev_ops == &vdev_spare_ops && 4579 newvd->vdev_isspare != oldvd->vdev_isspare) { 4580 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4581 } 4582 4583 if (newvd->vdev_isspare) 4584 pvops = &vdev_spare_ops; 4585 else 4586 pvops = &vdev_replacing_ops; 4587 } 4588 4589 /* 4590 * Make sure the new device is big enough. 4591 */ 4592 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4593 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4594 4595 /* 4596 * The new device cannot have a higher alignment requirement 4597 * than the top-level vdev. 4598 */ 4599 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4600 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4601 4602 /* 4603 * If this is an in-place replacement, update oldvd's path and devid 4604 * to make it distinguishable from newvd, and unopenable from now on. 4605 */ 4606 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4607 spa_strfree(oldvd->vdev_path); 4608 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4609 KM_SLEEP); 4610 (void) sprintf(oldvd->vdev_path, "%s/%s", 4611 newvd->vdev_path, "old"); 4612 if (oldvd->vdev_devid != NULL) { 4613 spa_strfree(oldvd->vdev_devid); 4614 oldvd->vdev_devid = NULL; 4615 } 4616 } 4617 4618 /* mark the device being resilvered */ 4619 newvd->vdev_resilvering = B_TRUE; 4620 4621 /* 4622 * If the parent is not a mirror, or if we're replacing, insert the new 4623 * mirror/replacing/spare vdev above oldvd. 4624 */ 4625 if (pvd->vdev_ops != pvops) 4626 pvd = vdev_add_parent(oldvd, pvops); 4627 4628 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4629 ASSERT(pvd->vdev_ops == pvops); 4630 ASSERT(oldvd->vdev_parent == pvd); 4631 4632 /* 4633 * Extract the new device from its root and add it to pvd. 4634 */ 4635 vdev_remove_child(newrootvd, newvd); 4636 newvd->vdev_id = pvd->vdev_children; 4637 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4638 vdev_add_child(pvd, newvd); 4639 4640 tvd = newvd->vdev_top; 4641 ASSERT(pvd->vdev_top == tvd); 4642 ASSERT(tvd->vdev_parent == rvd); 4643 4644 vdev_config_dirty(tvd); 4645 4646 /* 4647 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4648 * for any dmu_sync-ed blocks. It will propagate upward when 4649 * spa_vdev_exit() calls vdev_dtl_reassess(). 4650 */ 4651 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4652 4653 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4654 dtl_max_txg - TXG_INITIAL); 4655 4656 if (newvd->vdev_isspare) { 4657 spa_spare_activate(newvd); 4658 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4659 } 4660 4661 oldvdpath = spa_strdup(oldvd->vdev_path); 4662 newvdpath = spa_strdup(newvd->vdev_path); 4663 newvd_isspare = newvd->vdev_isspare; 4664 4665 /* 4666 * Mark newvd's DTL dirty in this txg. 4667 */ 4668 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4669 4670 /* 4671 * Restart the resilver 4672 */ 4673 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4674 4675 /* 4676 * Commit the config 4677 */ 4678 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4679 4680 spa_history_log_internal(spa, "vdev attach", NULL, 4681 "%s vdev=%s %s vdev=%s", 4682 replacing && newvd_isspare ? "spare in" : 4683 replacing ? "replace" : "attach", newvdpath, 4684 replacing ? "for" : "to", oldvdpath); 4685 4686 spa_strfree(oldvdpath); 4687 spa_strfree(newvdpath); 4688 4689 if (spa->spa_bootfs) 4690 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4691 4692 return (0); 4693} 4694 4695/* 4696 * Detach a device from a mirror or replacing vdev. 4697 * If 'replace_done' is specified, only detach if the parent 4698 * is a replacing vdev. 4699 */ 4700int 4701spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4702{ 4703 uint64_t txg; 4704 int error; 4705 vdev_t *rvd = spa->spa_root_vdev; 4706 vdev_t *vd, *pvd, *cvd, *tvd; 4707 boolean_t unspare = B_FALSE; 4708 uint64_t unspare_guid = 0; 4709 char *vdpath; 4710 4711 ASSERT(spa_writeable(spa)); 4712 4713 txg = spa_vdev_enter(spa); 4714 4715 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4716 4717 if (vd == NULL) 4718 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4719 4720 if (!vd->vdev_ops->vdev_op_leaf) 4721 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4722 4723 pvd = vd->vdev_parent; 4724 4725 /* 4726 * If the parent/child relationship is not as expected, don't do it. 4727 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4728 * vdev that's replacing B with C. The user's intent in replacing 4729 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4730 * the replace by detaching C, the expected behavior is to end up 4731 * M(A,B). But suppose that right after deciding to detach C, 4732 * the replacement of B completes. We would have M(A,C), and then 4733 * ask to detach C, which would leave us with just A -- not what 4734 * the user wanted. To prevent this, we make sure that the 4735 * parent/child relationship hasn't changed -- in this example, 4736 * that C's parent is still the replacing vdev R. 4737 */ 4738 if (pvd->vdev_guid != pguid && pguid != 0) 4739 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4740 4741 /* 4742 * Only 'replacing' or 'spare' vdevs can be replaced. 4743 */ 4744 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4745 pvd->vdev_ops != &vdev_spare_ops) 4746 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4747 4748 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4749 spa_version(spa) >= SPA_VERSION_SPARES); 4750 4751 /* 4752 * Only mirror, replacing, and spare vdevs support detach. 4753 */ 4754 if (pvd->vdev_ops != &vdev_replacing_ops && 4755 pvd->vdev_ops != &vdev_mirror_ops && 4756 pvd->vdev_ops != &vdev_spare_ops) 4757 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4758 4759 /* 4760 * If this device has the only valid copy of some data, 4761 * we cannot safely detach it. 4762 */ 4763 if (vdev_dtl_required(vd)) 4764 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4765 4766 ASSERT(pvd->vdev_children >= 2); 4767 4768 /* 4769 * If we are detaching the second disk from a replacing vdev, then 4770 * check to see if we changed the original vdev's path to have "/old" 4771 * at the end in spa_vdev_attach(). If so, undo that change now. 4772 */ 4773 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4774 vd->vdev_path != NULL) { 4775 size_t len = strlen(vd->vdev_path); 4776 4777 for (int c = 0; c < pvd->vdev_children; c++) { 4778 cvd = pvd->vdev_child[c]; 4779 4780 if (cvd == vd || cvd->vdev_path == NULL) 4781 continue; 4782 4783 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4784 strcmp(cvd->vdev_path + len, "/old") == 0) { 4785 spa_strfree(cvd->vdev_path); 4786 cvd->vdev_path = spa_strdup(vd->vdev_path); 4787 break; 4788 } 4789 } 4790 } 4791 4792 /* 4793 * If we are detaching the original disk from a spare, then it implies 4794 * that the spare should become a real disk, and be removed from the 4795 * active spare list for the pool. 4796 */ 4797 if (pvd->vdev_ops == &vdev_spare_ops && 4798 vd->vdev_id == 0 && 4799 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4800 unspare = B_TRUE; 4801 4802 /* 4803 * Erase the disk labels so the disk can be used for other things. 4804 * This must be done after all other error cases are handled, 4805 * but before we disembowel vd (so we can still do I/O to it). 4806 * But if we can't do it, don't treat the error as fatal -- 4807 * it may be that the unwritability of the disk is the reason 4808 * it's being detached! 4809 */ 4810 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4811 4812 /* 4813 * Remove vd from its parent and compact the parent's children. 4814 */ 4815 vdev_remove_child(pvd, vd); 4816 vdev_compact_children(pvd); 4817 4818 /* 4819 * Remember one of the remaining children so we can get tvd below. 4820 */ 4821 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4822 4823 /* 4824 * If we need to remove the remaining child from the list of hot spares, 4825 * do it now, marking the vdev as no longer a spare in the process. 4826 * We must do this before vdev_remove_parent(), because that can 4827 * change the GUID if it creates a new toplevel GUID. For a similar 4828 * reason, we must remove the spare now, in the same txg as the detach; 4829 * otherwise someone could attach a new sibling, change the GUID, and 4830 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4831 */ 4832 if (unspare) { 4833 ASSERT(cvd->vdev_isspare); 4834 spa_spare_remove(cvd); 4835 unspare_guid = cvd->vdev_guid; 4836 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4837 cvd->vdev_unspare = B_TRUE; 4838 } 4839 4840 /* 4841 * If the parent mirror/replacing vdev only has one child, 4842 * the parent is no longer needed. Remove it from the tree. 4843 */ 4844 if (pvd->vdev_children == 1) { 4845 if (pvd->vdev_ops == &vdev_spare_ops) 4846 cvd->vdev_unspare = B_FALSE; 4847 vdev_remove_parent(cvd); 4848 cvd->vdev_resilvering = B_FALSE; 4849 } 4850 4851 4852 /* 4853 * We don't set tvd until now because the parent we just removed 4854 * may have been the previous top-level vdev. 4855 */ 4856 tvd = cvd->vdev_top; 4857 ASSERT(tvd->vdev_parent == rvd); 4858 4859 /* 4860 * Reevaluate the parent vdev state. 4861 */ 4862 vdev_propagate_state(cvd); 4863 4864 /* 4865 * If the 'autoexpand' property is set on the pool then automatically 4866 * try to expand the size of the pool. For example if the device we 4867 * just detached was smaller than the others, it may be possible to 4868 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4869 * first so that we can obtain the updated sizes of the leaf vdevs. 4870 */ 4871 if (spa->spa_autoexpand) { 4872 vdev_reopen(tvd); 4873 vdev_expand(tvd, txg); 4874 } 4875 4876 vdev_config_dirty(tvd); 4877 4878 /* 4879 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4880 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4881 * But first make sure we're not on any *other* txg's DTL list, to 4882 * prevent vd from being accessed after it's freed. 4883 */ 4884 vdpath = spa_strdup(vd->vdev_path); 4885 for (int t = 0; t < TXG_SIZE; t++) 4886 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4887 vd->vdev_detached = B_TRUE; 4888 vdev_dirty(tvd, VDD_DTL, vd, txg); 4889 4890 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4891 4892 /* hang on to the spa before we release the lock */ 4893 spa_open_ref(spa, FTAG); 4894 4895 error = spa_vdev_exit(spa, vd, txg, 0); 4896 4897 spa_history_log_internal(spa, "detach", NULL, 4898 "vdev=%s", vdpath); 4899 spa_strfree(vdpath); 4900 4901 /* 4902 * If this was the removal of the original device in a hot spare vdev, 4903 * then we want to go through and remove the device from the hot spare 4904 * list of every other pool. 4905 */ 4906 if (unspare) { 4907 spa_t *altspa = NULL; 4908 4909 mutex_enter(&spa_namespace_lock); 4910 while ((altspa = spa_next(altspa)) != NULL) { 4911 if (altspa->spa_state != POOL_STATE_ACTIVE || 4912 altspa == spa) 4913 continue; 4914 4915 spa_open_ref(altspa, FTAG); 4916 mutex_exit(&spa_namespace_lock); 4917 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 4918 mutex_enter(&spa_namespace_lock); 4919 spa_close(altspa, FTAG); 4920 } 4921 mutex_exit(&spa_namespace_lock); 4922 4923 /* search the rest of the vdevs for spares to remove */ 4924 spa_vdev_resilver_done(spa); 4925 } 4926 4927 /* all done with the spa; OK to release */ 4928 mutex_enter(&spa_namespace_lock); 4929 spa_close(spa, FTAG); 4930 mutex_exit(&spa_namespace_lock); 4931 4932 return (error); 4933} 4934 4935/* 4936 * Split a set of devices from their mirrors, and create a new pool from them. 4937 */ 4938int 4939spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 4940 nvlist_t *props, boolean_t exp) 4941{ 4942 int error = 0; 4943 uint64_t txg, *glist; 4944 spa_t *newspa; 4945 uint_t c, children, lastlog; 4946 nvlist_t **child, *nvl, *tmp; 4947 dmu_tx_t *tx; 4948 char *altroot = NULL; 4949 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 4950 boolean_t activate_slog; 4951 4952 ASSERT(spa_writeable(spa)); 4953 4954 txg = spa_vdev_enter(spa); 4955 4956 /* clear the log and flush everything up to now */ 4957 activate_slog = spa_passivate_log(spa); 4958 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 4959 error = spa_offline_log(spa); 4960 txg = spa_vdev_config_enter(spa); 4961 4962 if (activate_slog) 4963 spa_activate_log(spa); 4964 4965 if (error != 0) 4966 return (spa_vdev_exit(spa, NULL, txg, error)); 4967 4968 /* check new spa name before going any further */ 4969 if (spa_lookup(newname) != NULL) 4970 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 4971 4972 /* 4973 * scan through all the children to ensure they're all mirrors 4974 */ 4975 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 4976 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 4977 &children) != 0) 4978 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4979 4980 /* first, check to ensure we've got the right child count */ 4981 rvd = spa->spa_root_vdev; 4982 lastlog = 0; 4983 for (c = 0; c < rvd->vdev_children; c++) { 4984 vdev_t *vd = rvd->vdev_child[c]; 4985 4986 /* don't count the holes & logs as children */ 4987 if (vd->vdev_islog || vd->vdev_ishole) { 4988 if (lastlog == 0) 4989 lastlog = c; 4990 continue; 4991 } 4992 4993 lastlog = 0; 4994 } 4995 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 4996 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4997 4998 /* next, ensure no spare or cache devices are part of the split */ 4999 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5000 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5001 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5002 5003 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5004 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5005 5006 /* then, loop over each vdev and validate it */ 5007 for (c = 0; c < children; c++) { 5008 uint64_t is_hole = 0; 5009 5010 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5011 &is_hole); 5012 5013 if (is_hole != 0) { 5014 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5015 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5016 continue; 5017 } else { 5018 error = EINVAL; 5019 break; 5020 } 5021 } 5022 5023 /* which disk is going to be split? */ 5024 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5025 &glist[c]) != 0) { 5026 error = EINVAL; 5027 break; 5028 } 5029 5030 /* look it up in the spa */ 5031 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5032 if (vml[c] == NULL) { 5033 error = ENODEV; 5034 break; 5035 } 5036 5037 /* make sure there's nothing stopping the split */ 5038 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5039 vml[c]->vdev_islog || 5040 vml[c]->vdev_ishole || 5041 vml[c]->vdev_isspare || 5042 vml[c]->vdev_isl2cache || 5043 !vdev_writeable(vml[c]) || 5044 vml[c]->vdev_children != 0 || 5045 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5046 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5047 error = EINVAL; 5048 break; 5049 } 5050 5051 if (vdev_dtl_required(vml[c])) { 5052 error = EBUSY; 5053 break; 5054 } 5055 5056 /* we need certain info from the top level */ 5057 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5058 vml[c]->vdev_top->vdev_ms_array) == 0); 5059 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5060 vml[c]->vdev_top->vdev_ms_shift) == 0); 5061 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5062 vml[c]->vdev_top->vdev_asize) == 0); 5063 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5064 vml[c]->vdev_top->vdev_ashift) == 0); 5065 } 5066 5067 if (error != 0) { 5068 kmem_free(vml, children * sizeof (vdev_t *)); 5069 kmem_free(glist, children * sizeof (uint64_t)); 5070 return (spa_vdev_exit(spa, NULL, txg, error)); 5071 } 5072 5073 /* stop writers from using the disks */ 5074 for (c = 0; c < children; c++) { 5075 if (vml[c] != NULL) 5076 vml[c]->vdev_offline = B_TRUE; 5077 } 5078 vdev_reopen(spa->spa_root_vdev); 5079 5080 /* 5081 * Temporarily record the splitting vdevs in the spa config. This 5082 * will disappear once the config is regenerated. 5083 */ 5084 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5085 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5086 glist, children) == 0); 5087 kmem_free(glist, children * sizeof (uint64_t)); 5088 5089 mutex_enter(&spa->spa_props_lock); 5090 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5091 nvl) == 0); 5092 mutex_exit(&spa->spa_props_lock); 5093 spa->spa_config_splitting = nvl; 5094 vdev_config_dirty(spa->spa_root_vdev); 5095 5096 /* configure and create the new pool */ 5097 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5098 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5099 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5100 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5101 spa_version(spa)) == 0); 5102 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5103 spa->spa_config_txg) == 0); 5104 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5105 spa_generate_guid(NULL)) == 0); 5106 (void) nvlist_lookup_string(props, 5107 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5108 5109 /* add the new pool to the namespace */ 5110 newspa = spa_add(newname, config, altroot); 5111 newspa->spa_config_txg = spa->spa_config_txg; 5112 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5113 5114 /* release the spa config lock, retaining the namespace lock */ 5115 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5116 5117 if (zio_injection_enabled) 5118 zio_handle_panic_injection(spa, FTAG, 1); 5119 5120 spa_activate(newspa, spa_mode_global); 5121 spa_async_suspend(newspa); 5122 5123#ifndef sun 5124 /* mark that we are creating new spa by splitting */ 5125 newspa->spa_splitting_newspa = B_TRUE; 5126#endif 5127 /* create the new pool from the disks of the original pool */ 5128 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5129#ifndef sun 5130 newspa->spa_splitting_newspa = B_FALSE; 5131#endif 5132 if (error) 5133 goto out; 5134 5135 /* if that worked, generate a real config for the new pool */ 5136 if (newspa->spa_root_vdev != NULL) { 5137 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5138 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5139 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5140 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5141 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5142 B_TRUE)); 5143 } 5144 5145 /* set the props */ 5146 if (props != NULL) { 5147 spa_configfile_set(newspa, props, B_FALSE); 5148 error = spa_prop_set(newspa, props); 5149 if (error) 5150 goto out; 5151 } 5152 5153 /* flush everything */ 5154 txg = spa_vdev_config_enter(newspa); 5155 vdev_config_dirty(newspa->spa_root_vdev); 5156 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5157 5158 if (zio_injection_enabled) 5159 zio_handle_panic_injection(spa, FTAG, 2); 5160 5161 spa_async_resume(newspa); 5162 5163 /* finally, update the original pool's config */ 5164 txg = spa_vdev_config_enter(spa); 5165 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5166 error = dmu_tx_assign(tx, TXG_WAIT); 5167 if (error != 0) 5168 dmu_tx_abort(tx); 5169 for (c = 0; c < children; c++) { 5170 if (vml[c] != NULL) { 5171 vdev_split(vml[c]); 5172 if (error == 0) 5173 spa_history_log_internal(spa, "detach", tx, 5174 "vdev=%s", vml[c]->vdev_path); 5175 vdev_free(vml[c]); 5176 } 5177 } 5178 vdev_config_dirty(spa->spa_root_vdev); 5179 spa->spa_config_splitting = NULL; 5180 nvlist_free(nvl); 5181 if (error == 0) 5182 dmu_tx_commit(tx); 5183 (void) spa_vdev_exit(spa, NULL, txg, 0); 5184 5185 if (zio_injection_enabled) 5186 zio_handle_panic_injection(spa, FTAG, 3); 5187 5188 /* split is complete; log a history record */ 5189 spa_history_log_internal(newspa, "split", NULL, 5190 "from pool %s", spa_name(spa)); 5191 5192 kmem_free(vml, children * sizeof (vdev_t *)); 5193 5194 /* if we're not going to mount the filesystems in userland, export */ 5195 if (exp) 5196 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5197 B_FALSE, B_FALSE); 5198 5199 return (error); 5200 5201out: 5202 spa_unload(newspa); 5203 spa_deactivate(newspa); 5204 spa_remove(newspa); 5205 5206 txg = spa_vdev_config_enter(spa); 5207 5208 /* re-online all offlined disks */ 5209 for (c = 0; c < children; c++) { 5210 if (vml[c] != NULL) 5211 vml[c]->vdev_offline = B_FALSE; 5212 } 5213 vdev_reopen(spa->spa_root_vdev); 5214 5215 nvlist_free(spa->spa_config_splitting); 5216 spa->spa_config_splitting = NULL; 5217 (void) spa_vdev_exit(spa, NULL, txg, error); 5218 5219 kmem_free(vml, children * sizeof (vdev_t *)); 5220 return (error); 5221} 5222 5223static nvlist_t * 5224spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5225{ 5226 for (int i = 0; i < count; i++) { 5227 uint64_t guid; 5228 5229 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5230 &guid) == 0); 5231 5232 if (guid == target_guid) 5233 return (nvpp[i]); 5234 } 5235 5236 return (NULL); 5237} 5238 5239static void 5240spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5241 nvlist_t *dev_to_remove) 5242{ 5243 nvlist_t **newdev = NULL; 5244 5245 if (count > 1) 5246 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5247 5248 for (int i = 0, j = 0; i < count; i++) { 5249 if (dev[i] == dev_to_remove) 5250 continue; 5251 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5252 } 5253 5254 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5255 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5256 5257 for (int i = 0; i < count - 1; i++) 5258 nvlist_free(newdev[i]); 5259 5260 if (count > 1) 5261 kmem_free(newdev, (count - 1) * sizeof (void *)); 5262} 5263 5264/* 5265 * Evacuate the device. 5266 */ 5267static int 5268spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5269{ 5270 uint64_t txg; 5271 int error = 0; 5272 5273 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5274 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5275 ASSERT(vd == vd->vdev_top); 5276 5277 /* 5278 * Evacuate the device. We don't hold the config lock as writer 5279 * since we need to do I/O but we do keep the 5280 * spa_namespace_lock held. Once this completes the device 5281 * should no longer have any blocks allocated on it. 5282 */ 5283 if (vd->vdev_islog) { 5284 if (vd->vdev_stat.vs_alloc != 0) 5285 error = spa_offline_log(spa); 5286 } else { 5287 error = ENOTSUP; 5288 } 5289 5290 if (error) 5291 return (error); 5292 5293 /* 5294 * The evacuation succeeded. Remove any remaining MOS metadata 5295 * associated with this vdev, and wait for these changes to sync. 5296 */ 5297 ASSERT0(vd->vdev_stat.vs_alloc); 5298 txg = spa_vdev_config_enter(spa); 5299 vd->vdev_removing = B_TRUE; 5300 vdev_dirty(vd, 0, NULL, txg); 5301 vdev_config_dirty(vd); 5302 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5303 5304 return (0); 5305} 5306 5307/* 5308 * Complete the removal by cleaning up the namespace. 5309 */ 5310static void 5311spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5312{ 5313 vdev_t *rvd = spa->spa_root_vdev; 5314 uint64_t id = vd->vdev_id; 5315 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5316 5317 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5318 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5319 ASSERT(vd == vd->vdev_top); 5320 5321 /* 5322 * Only remove any devices which are empty. 5323 */ 5324 if (vd->vdev_stat.vs_alloc != 0) 5325 return; 5326 5327 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5328 5329 if (list_link_active(&vd->vdev_state_dirty_node)) 5330 vdev_state_clean(vd); 5331 if (list_link_active(&vd->vdev_config_dirty_node)) 5332 vdev_config_clean(vd); 5333 5334 vdev_free(vd); 5335 5336 if (last_vdev) { 5337 vdev_compact_children(rvd); 5338 } else { 5339 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5340 vdev_add_child(rvd, vd); 5341 } 5342 vdev_config_dirty(rvd); 5343 5344 /* 5345 * Reassess the health of our root vdev. 5346 */ 5347 vdev_reopen(rvd); 5348} 5349 5350/* 5351 * Remove a device from the pool - 5352 * 5353 * Removing a device from the vdev namespace requires several steps 5354 * and can take a significant amount of time. As a result we use 5355 * the spa_vdev_config_[enter/exit] functions which allow us to 5356 * grab and release the spa_config_lock while still holding the namespace 5357 * lock. During each step the configuration is synced out. 5358 */ 5359 5360/* 5361 * Remove a device from the pool. Currently, this supports removing only hot 5362 * spares, slogs, and level 2 ARC devices. 5363 */ 5364int 5365spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5366{ 5367 vdev_t *vd; 5368 metaslab_group_t *mg; 5369 nvlist_t **spares, **l2cache, *nv; 5370 uint64_t txg = 0; 5371 uint_t nspares, nl2cache; 5372 int error = 0; 5373 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5374 5375 ASSERT(spa_writeable(spa)); 5376 5377 if (!locked) 5378 txg = spa_vdev_enter(spa); 5379 5380 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5381 5382 if (spa->spa_spares.sav_vdevs != NULL && 5383 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5384 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5385 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5386 /* 5387 * Only remove the hot spare if it's not currently in use 5388 * in this pool. 5389 */ 5390 if (vd == NULL || unspare) { 5391 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5392 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5393 spa_load_spares(spa); 5394 spa->spa_spares.sav_sync = B_TRUE; 5395 } else { 5396 error = EBUSY; 5397 } 5398 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5399 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5400 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5401 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5402 /* 5403 * Cache devices can always be removed. 5404 */ 5405 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5406 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5407 spa_load_l2cache(spa); 5408 spa->spa_l2cache.sav_sync = B_TRUE; 5409 } else if (vd != NULL && vd->vdev_islog) { 5410 ASSERT(!locked); 5411 ASSERT(vd == vd->vdev_top); 5412 5413 /* 5414 * XXX - Once we have bp-rewrite this should 5415 * become the common case. 5416 */ 5417 5418 mg = vd->vdev_mg; 5419 5420 /* 5421 * Stop allocating from this vdev. 5422 */ 5423 metaslab_group_passivate(mg); 5424 5425 /* 5426 * Wait for the youngest allocations and frees to sync, 5427 * and then wait for the deferral of those frees to finish. 5428 */ 5429 spa_vdev_config_exit(spa, NULL, 5430 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5431 5432 /* 5433 * Attempt to evacuate the vdev. 5434 */ 5435 error = spa_vdev_remove_evacuate(spa, vd); 5436 5437 txg = spa_vdev_config_enter(spa); 5438 5439 /* 5440 * If we couldn't evacuate the vdev, unwind. 5441 */ 5442 if (error) { 5443 metaslab_group_activate(mg); 5444 return (spa_vdev_exit(spa, NULL, txg, error)); 5445 } 5446 5447 /* 5448 * Clean up the vdev namespace. 5449 */ 5450 spa_vdev_remove_from_namespace(spa, vd); 5451 5452 } else if (vd != NULL) { 5453 /* 5454 * Normal vdevs cannot be removed (yet). 5455 */ 5456 error = ENOTSUP; 5457 } else { 5458 /* 5459 * There is no vdev of any kind with the specified guid. 5460 */ 5461 error = ENOENT; 5462 } 5463 5464 if (!locked) 5465 return (spa_vdev_exit(spa, NULL, txg, error)); 5466 5467 return (error); 5468} 5469 5470/* 5471 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5472 * current spared, so we can detach it. 5473 */ 5474static vdev_t * 5475spa_vdev_resilver_done_hunt(vdev_t *vd) 5476{ 5477 vdev_t *newvd, *oldvd; 5478 5479 for (int c = 0; c < vd->vdev_children; c++) { 5480 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5481 if (oldvd != NULL) 5482 return (oldvd); 5483 } 5484 5485 /* 5486 * Check for a completed replacement. We always consider the first 5487 * vdev in the list to be the oldest vdev, and the last one to be 5488 * the newest (see spa_vdev_attach() for how that works). In 5489 * the case where the newest vdev is faulted, we will not automatically 5490 * remove it after a resilver completes. This is OK as it will require 5491 * user intervention to determine which disk the admin wishes to keep. 5492 */ 5493 if (vd->vdev_ops == &vdev_replacing_ops) { 5494 ASSERT(vd->vdev_children > 1); 5495 5496 newvd = vd->vdev_child[vd->vdev_children - 1]; 5497 oldvd = vd->vdev_child[0]; 5498 5499 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5500 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5501 !vdev_dtl_required(oldvd)) 5502 return (oldvd); 5503 } 5504 5505 /* 5506 * Check for a completed resilver with the 'unspare' flag set. 5507 */ 5508 if (vd->vdev_ops == &vdev_spare_ops) { 5509 vdev_t *first = vd->vdev_child[0]; 5510 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5511 5512 if (last->vdev_unspare) { 5513 oldvd = first; 5514 newvd = last; 5515 } else if (first->vdev_unspare) { 5516 oldvd = last; 5517 newvd = first; 5518 } else { 5519 oldvd = NULL; 5520 } 5521 5522 if (oldvd != NULL && 5523 vdev_dtl_empty(newvd, DTL_MISSING) && 5524 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5525 !vdev_dtl_required(oldvd)) 5526 return (oldvd); 5527 5528 /* 5529 * If there are more than two spares attached to a disk, 5530 * and those spares are not required, then we want to 5531 * attempt to free them up now so that they can be used 5532 * by other pools. Once we're back down to a single 5533 * disk+spare, we stop removing them. 5534 */ 5535 if (vd->vdev_children > 2) { 5536 newvd = vd->vdev_child[1]; 5537 5538 if (newvd->vdev_isspare && last->vdev_isspare && 5539 vdev_dtl_empty(last, DTL_MISSING) && 5540 vdev_dtl_empty(last, DTL_OUTAGE) && 5541 !vdev_dtl_required(newvd)) 5542 return (newvd); 5543 } 5544 } 5545 5546 return (NULL); 5547} 5548 5549static void 5550spa_vdev_resilver_done(spa_t *spa) 5551{ 5552 vdev_t *vd, *pvd, *ppvd; 5553 uint64_t guid, sguid, pguid, ppguid; 5554 5555 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5556 5557 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5558 pvd = vd->vdev_parent; 5559 ppvd = pvd->vdev_parent; 5560 guid = vd->vdev_guid; 5561 pguid = pvd->vdev_guid; 5562 ppguid = ppvd->vdev_guid; 5563 sguid = 0; 5564 /* 5565 * If we have just finished replacing a hot spared device, then 5566 * we need to detach the parent's first child (the original hot 5567 * spare) as well. 5568 */ 5569 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5570 ppvd->vdev_children == 2) { 5571 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5572 sguid = ppvd->vdev_child[1]->vdev_guid; 5573 } 5574 spa_config_exit(spa, SCL_ALL, FTAG); 5575 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5576 return; 5577 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5578 return; 5579 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5580 } 5581 5582 spa_config_exit(spa, SCL_ALL, FTAG); 5583} 5584 5585/* 5586 * Update the stored path or FRU for this vdev. 5587 */ 5588int 5589spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5590 boolean_t ispath) 5591{ 5592 vdev_t *vd; 5593 boolean_t sync = B_FALSE; 5594 5595 ASSERT(spa_writeable(spa)); 5596 5597 spa_vdev_state_enter(spa, SCL_ALL); 5598 5599 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5600 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5601 5602 if (!vd->vdev_ops->vdev_op_leaf) 5603 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5604 5605 if (ispath) { 5606 if (strcmp(value, vd->vdev_path) != 0) { 5607 spa_strfree(vd->vdev_path); 5608 vd->vdev_path = spa_strdup(value); 5609 sync = B_TRUE; 5610 } 5611 } else { 5612 if (vd->vdev_fru == NULL) { 5613 vd->vdev_fru = spa_strdup(value); 5614 sync = B_TRUE; 5615 } else if (strcmp(value, vd->vdev_fru) != 0) { 5616 spa_strfree(vd->vdev_fru); 5617 vd->vdev_fru = spa_strdup(value); 5618 sync = B_TRUE; 5619 } 5620 } 5621 5622 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5623} 5624 5625int 5626spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5627{ 5628 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5629} 5630 5631int 5632spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5633{ 5634 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5635} 5636 5637/* 5638 * ========================================================================== 5639 * SPA Scanning 5640 * ========================================================================== 5641 */ 5642 5643int 5644spa_scan_stop(spa_t *spa) 5645{ 5646 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5647 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5648 return (EBUSY); 5649 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5650} 5651 5652int 5653spa_scan(spa_t *spa, pool_scan_func_t func) 5654{ 5655 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5656 5657 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5658 return (ENOTSUP); 5659 5660 /* 5661 * If a resilver was requested, but there is no DTL on a 5662 * writeable leaf device, we have nothing to do. 5663 */ 5664 if (func == POOL_SCAN_RESILVER && 5665 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5666 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5667 return (0); 5668 } 5669 5670 return (dsl_scan(spa->spa_dsl_pool, func)); 5671} 5672 5673/* 5674 * ========================================================================== 5675 * SPA async task processing 5676 * ========================================================================== 5677 */ 5678 5679static void 5680spa_async_remove(spa_t *spa, vdev_t *vd) 5681{ 5682 if (vd->vdev_remove_wanted) { 5683 vd->vdev_remove_wanted = B_FALSE; 5684 vd->vdev_delayed_close = B_FALSE; 5685 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5686 5687 /* 5688 * We want to clear the stats, but we don't want to do a full 5689 * vdev_clear() as that will cause us to throw away 5690 * degraded/faulted state as well as attempt to reopen the 5691 * device, all of which is a waste. 5692 */ 5693 vd->vdev_stat.vs_read_errors = 0; 5694 vd->vdev_stat.vs_write_errors = 0; 5695 vd->vdev_stat.vs_checksum_errors = 0; 5696 5697 vdev_state_dirty(vd->vdev_top); 5698 } 5699 5700 for (int c = 0; c < vd->vdev_children; c++) 5701 spa_async_remove(spa, vd->vdev_child[c]); 5702} 5703 5704static void 5705spa_async_probe(spa_t *spa, vdev_t *vd) 5706{ 5707 if (vd->vdev_probe_wanted) { 5708 vd->vdev_probe_wanted = B_FALSE; 5709 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5710 } 5711 5712 for (int c = 0; c < vd->vdev_children; c++) 5713 spa_async_probe(spa, vd->vdev_child[c]); 5714} 5715 5716static void 5717spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5718{ 5719 sysevent_id_t eid; 5720 nvlist_t *attr; 5721 char *physpath; 5722 5723 if (!spa->spa_autoexpand) 5724 return; 5725 5726 for (int c = 0; c < vd->vdev_children; c++) { 5727 vdev_t *cvd = vd->vdev_child[c]; 5728 spa_async_autoexpand(spa, cvd); 5729 } 5730 5731 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5732 return; 5733 5734 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5735 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5736 5737 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5738 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5739 5740 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5741 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5742 5743 nvlist_free(attr); 5744 kmem_free(physpath, MAXPATHLEN); 5745} 5746 5747static void 5748spa_async_thread(void *arg) 5749{ 5750 spa_t *spa = arg; 5751 int tasks; 5752 5753 ASSERT(spa->spa_sync_on); 5754 5755 mutex_enter(&spa->spa_async_lock); 5756 tasks = spa->spa_async_tasks; 5757 spa->spa_async_tasks = 0; 5758 mutex_exit(&spa->spa_async_lock); 5759 5760 /* 5761 * See if the config needs to be updated. 5762 */ 5763 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5764 uint64_t old_space, new_space; 5765 5766 mutex_enter(&spa_namespace_lock); 5767 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5768 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5769 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5770 mutex_exit(&spa_namespace_lock); 5771 5772 /* 5773 * If the pool grew as a result of the config update, 5774 * then log an internal history event. 5775 */ 5776 if (new_space != old_space) { 5777 spa_history_log_internal(spa, "vdev online", NULL, 5778 "pool '%s' size: %llu(+%llu)", 5779 spa_name(spa), new_space, new_space - old_space); 5780 } 5781 } 5782 5783 /* 5784 * See if any devices need to be marked REMOVED. 5785 */ 5786 if (tasks & SPA_ASYNC_REMOVE) { 5787 spa_vdev_state_enter(spa, SCL_NONE); 5788 spa_async_remove(spa, spa->spa_root_vdev); 5789 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5790 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5791 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5792 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5793 (void) spa_vdev_state_exit(spa, NULL, 0); 5794 } 5795 5796 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5797 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5798 spa_async_autoexpand(spa, spa->spa_root_vdev); 5799 spa_config_exit(spa, SCL_CONFIG, FTAG); 5800 } 5801 5802 /* 5803 * See if any devices need to be probed. 5804 */ 5805 if (tasks & SPA_ASYNC_PROBE) { 5806 spa_vdev_state_enter(spa, SCL_NONE); 5807 spa_async_probe(spa, spa->spa_root_vdev); 5808 (void) spa_vdev_state_exit(spa, NULL, 0); 5809 } 5810 5811 /* 5812 * If any devices are done replacing, detach them. 5813 */ 5814 if (tasks & SPA_ASYNC_RESILVER_DONE) 5815 spa_vdev_resilver_done(spa); 5816 5817 /* 5818 * Kick off a resilver. 5819 */ 5820 if (tasks & SPA_ASYNC_RESILVER) 5821 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5822 5823 /* 5824 * Let the world know that we're done. 5825 */ 5826 mutex_enter(&spa->spa_async_lock); 5827 spa->spa_async_thread = NULL; 5828 cv_broadcast(&spa->spa_async_cv); 5829 mutex_exit(&spa->spa_async_lock); 5830 thread_exit(); 5831} 5832 5833void 5834spa_async_suspend(spa_t *spa) 5835{ 5836 mutex_enter(&spa->spa_async_lock); 5837 spa->spa_async_suspended++; 5838 while (spa->spa_async_thread != NULL) 5839 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5840 mutex_exit(&spa->spa_async_lock); 5841} 5842 5843void 5844spa_async_resume(spa_t *spa) 5845{ 5846 mutex_enter(&spa->spa_async_lock); 5847 ASSERT(spa->spa_async_suspended != 0); 5848 spa->spa_async_suspended--; 5849 mutex_exit(&spa->spa_async_lock); 5850} 5851 5852static void 5853spa_async_dispatch(spa_t *spa) 5854{ 5855 mutex_enter(&spa->spa_async_lock); 5856 if (spa->spa_async_tasks && !spa->spa_async_suspended && 5857 spa->spa_async_thread == NULL && 5858 rootdir != NULL && !vn_is_readonly(rootdir)) 5859 spa->spa_async_thread = thread_create(NULL, 0, 5860 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5861 mutex_exit(&spa->spa_async_lock); 5862} 5863 5864void 5865spa_async_request(spa_t *spa, int task) 5866{ 5867 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5868 mutex_enter(&spa->spa_async_lock); 5869 spa->spa_async_tasks |= task; 5870 mutex_exit(&spa->spa_async_lock); 5871} 5872 5873/* 5874 * ========================================================================== 5875 * SPA syncing routines 5876 * ========================================================================== 5877 */ 5878 5879static int 5880bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5881{ 5882 bpobj_t *bpo = arg; 5883 bpobj_enqueue(bpo, bp, tx); 5884 return (0); 5885} 5886 5887static int 5888spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5889{ 5890 zio_t *zio = arg; 5891 5892 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5893 BP_GET_PSIZE(bp), zio->io_flags)); 5894 return (0); 5895} 5896 5897static void 5898spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5899{ 5900 char *packed = NULL; 5901 size_t bufsize; 5902 size_t nvsize = 0; 5903 dmu_buf_t *db; 5904 5905 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5906 5907 /* 5908 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 5909 * information. This avoids the dbuf_will_dirty() path and 5910 * saves us a pre-read to get data we don't actually care about. 5911 */ 5912 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 5913 packed = kmem_alloc(bufsize, KM_SLEEP); 5914 5915 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5916 KM_SLEEP) == 0); 5917 bzero(packed + nvsize, bufsize - nvsize); 5918 5919 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 5920 5921 kmem_free(packed, bufsize); 5922 5923 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 5924 dmu_buf_will_dirty(db, tx); 5925 *(uint64_t *)db->db_data = nvsize; 5926 dmu_buf_rele(db, FTAG); 5927} 5928 5929static void 5930spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 5931 const char *config, const char *entry) 5932{ 5933 nvlist_t *nvroot; 5934 nvlist_t **list; 5935 int i; 5936 5937 if (!sav->sav_sync) 5938 return; 5939 5940 /* 5941 * Update the MOS nvlist describing the list of available devices. 5942 * spa_validate_aux() will have already made sure this nvlist is 5943 * valid and the vdevs are labeled appropriately. 5944 */ 5945 if (sav->sav_object == 0) { 5946 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 5947 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 5948 sizeof (uint64_t), tx); 5949 VERIFY(zap_update(spa->spa_meta_objset, 5950 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 5951 &sav->sav_object, tx) == 0); 5952 } 5953 5954 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5955 if (sav->sav_count == 0) { 5956 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 5957 } else { 5958 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 5959 for (i = 0; i < sav->sav_count; i++) 5960 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 5961 B_FALSE, VDEV_CONFIG_L2CACHE); 5962 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 5963 sav->sav_count) == 0); 5964 for (i = 0; i < sav->sav_count; i++) 5965 nvlist_free(list[i]); 5966 kmem_free(list, sav->sav_count * sizeof (void *)); 5967 } 5968 5969 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 5970 nvlist_free(nvroot); 5971 5972 sav->sav_sync = B_FALSE; 5973} 5974 5975static void 5976spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 5977{ 5978 nvlist_t *config; 5979 5980 if (list_is_empty(&spa->spa_config_dirty_list)) 5981 return; 5982 5983 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5984 5985 config = spa_config_generate(spa, spa->spa_root_vdev, 5986 dmu_tx_get_txg(tx), B_FALSE); 5987 5988 /* 5989 * If we're upgrading the spa version then make sure that 5990 * the config object gets updated with the correct version. 5991 */ 5992 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 5993 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5994 spa->spa_uberblock.ub_version); 5995 5996 spa_config_exit(spa, SCL_STATE, FTAG); 5997 5998 if (spa->spa_config_syncing) 5999 nvlist_free(spa->spa_config_syncing); 6000 spa->spa_config_syncing = config; 6001 6002 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6003} 6004 6005static void 6006spa_sync_version(void *arg, dmu_tx_t *tx) 6007{ 6008 uint64_t *versionp = arg; 6009 uint64_t version = *versionp; 6010 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6011 6012 /* 6013 * Setting the version is special cased when first creating the pool. 6014 */ 6015 ASSERT(tx->tx_txg != TXG_INITIAL); 6016 6017 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6018 ASSERT(version >= spa_version(spa)); 6019 6020 spa->spa_uberblock.ub_version = version; 6021 vdev_config_dirty(spa->spa_root_vdev); 6022 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6023} 6024 6025/* 6026 * Set zpool properties. 6027 */ 6028static void 6029spa_sync_props(void *arg, dmu_tx_t *tx) 6030{ 6031 nvlist_t *nvp = arg; 6032 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6033 objset_t *mos = spa->spa_meta_objset; 6034 nvpair_t *elem = NULL; 6035 6036 mutex_enter(&spa->spa_props_lock); 6037 6038 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6039 uint64_t intval; 6040 char *strval, *fname; 6041 zpool_prop_t prop; 6042 const char *propname; 6043 zprop_type_t proptype; 6044 zfeature_info_t *feature; 6045 6046 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6047 case ZPROP_INVAL: 6048 /* 6049 * We checked this earlier in spa_prop_validate(). 6050 */ 6051 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6052 6053 fname = strchr(nvpair_name(elem), '@') + 1; 6054 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature)); 6055 6056 spa_feature_enable(spa, feature, tx); 6057 spa_history_log_internal(spa, "set", tx, 6058 "%s=enabled", nvpair_name(elem)); 6059 break; 6060 6061 case ZPOOL_PROP_VERSION: 6062 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6063 /* 6064 * The version is synced seperatly before other 6065 * properties and should be correct by now. 6066 */ 6067 ASSERT3U(spa_version(spa), >=, intval); 6068 break; 6069 6070 case ZPOOL_PROP_ALTROOT: 6071 /* 6072 * 'altroot' is a non-persistent property. It should 6073 * have been set temporarily at creation or import time. 6074 */ 6075 ASSERT(spa->spa_root != NULL); 6076 break; 6077 6078 case ZPOOL_PROP_READONLY: 6079 case ZPOOL_PROP_CACHEFILE: 6080 /* 6081 * 'readonly' and 'cachefile' are also non-persisitent 6082 * properties. 6083 */ 6084 break; 6085 case ZPOOL_PROP_COMMENT: 6086 VERIFY(nvpair_value_string(elem, &strval) == 0); 6087 if (spa->spa_comment != NULL) 6088 spa_strfree(spa->spa_comment); 6089 spa->spa_comment = spa_strdup(strval); 6090 /* 6091 * We need to dirty the configuration on all the vdevs 6092 * so that their labels get updated. It's unnecessary 6093 * to do this for pool creation since the vdev's 6094 * configuratoin has already been dirtied. 6095 */ 6096 if (tx->tx_txg != TXG_INITIAL) 6097 vdev_config_dirty(spa->spa_root_vdev); 6098 spa_history_log_internal(spa, "set", tx, 6099 "%s=%s", nvpair_name(elem), strval); 6100 break; 6101 default: 6102 /* 6103 * Set pool property values in the poolprops mos object. 6104 */ 6105 if (spa->spa_pool_props_object == 0) { 6106 spa->spa_pool_props_object = 6107 zap_create_link(mos, DMU_OT_POOL_PROPS, 6108 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6109 tx); 6110 } 6111 6112 /* normalize the property name */ 6113 propname = zpool_prop_to_name(prop); 6114 proptype = zpool_prop_get_type(prop); 6115 6116 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6117 ASSERT(proptype == PROP_TYPE_STRING); 6118 VERIFY(nvpair_value_string(elem, &strval) == 0); 6119 VERIFY(zap_update(mos, 6120 spa->spa_pool_props_object, propname, 6121 1, strlen(strval) + 1, strval, tx) == 0); 6122 spa_history_log_internal(spa, "set", tx, 6123 "%s=%s", nvpair_name(elem), strval); 6124 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6125 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6126 6127 if (proptype == PROP_TYPE_INDEX) { 6128 const char *unused; 6129 VERIFY(zpool_prop_index_to_string( 6130 prop, intval, &unused) == 0); 6131 } 6132 VERIFY(zap_update(mos, 6133 spa->spa_pool_props_object, propname, 6134 8, 1, &intval, tx) == 0); 6135 spa_history_log_internal(spa, "set", tx, 6136 "%s=%lld", nvpair_name(elem), intval); 6137 } else { 6138 ASSERT(0); /* not allowed */ 6139 } 6140 6141 switch (prop) { 6142 case ZPOOL_PROP_DELEGATION: 6143 spa->spa_delegation = intval; 6144 break; 6145 case ZPOOL_PROP_BOOTFS: 6146 spa->spa_bootfs = intval; 6147 break; 6148 case ZPOOL_PROP_FAILUREMODE: 6149 spa->spa_failmode = intval; 6150 break; 6151 case ZPOOL_PROP_AUTOEXPAND: 6152 spa->spa_autoexpand = intval; 6153 if (tx->tx_txg != TXG_INITIAL) 6154 spa_async_request(spa, 6155 SPA_ASYNC_AUTOEXPAND); 6156 break; 6157 case ZPOOL_PROP_DEDUPDITTO: 6158 spa->spa_dedup_ditto = intval; 6159 break; 6160 default: 6161 break; 6162 } 6163 } 6164 6165 } 6166 6167 mutex_exit(&spa->spa_props_lock); 6168} 6169 6170/* 6171 * Perform one-time upgrade on-disk changes. spa_version() does not 6172 * reflect the new version this txg, so there must be no changes this 6173 * txg to anything that the upgrade code depends on after it executes. 6174 * Therefore this must be called after dsl_pool_sync() does the sync 6175 * tasks. 6176 */ 6177static void 6178spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6179{ 6180 dsl_pool_t *dp = spa->spa_dsl_pool; 6181 6182 ASSERT(spa->spa_sync_pass == 1); 6183 6184 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6185 6186 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6187 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6188 dsl_pool_create_origin(dp, tx); 6189 6190 /* Keeping the origin open increases spa_minref */ 6191 spa->spa_minref += 3; 6192 } 6193 6194 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6195 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6196 dsl_pool_upgrade_clones(dp, tx); 6197 } 6198 6199 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6200 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6201 dsl_pool_upgrade_dir_clones(dp, tx); 6202 6203 /* Keeping the freedir open increases spa_minref */ 6204 spa->spa_minref += 3; 6205 } 6206 6207 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6208 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6209 spa_feature_create_zap_objects(spa, tx); 6210 } 6211 rrw_exit(&dp->dp_config_rwlock, FTAG); 6212} 6213 6214/* 6215 * Sync the specified transaction group. New blocks may be dirtied as 6216 * part of the process, so we iterate until it converges. 6217 */ 6218void 6219spa_sync(spa_t *spa, uint64_t txg) 6220{ 6221 dsl_pool_t *dp = spa->spa_dsl_pool; 6222 objset_t *mos = spa->spa_meta_objset; 6223 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj; 6224 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6225 vdev_t *rvd = spa->spa_root_vdev; 6226 vdev_t *vd; 6227 dmu_tx_t *tx; 6228 int error; 6229 6230 VERIFY(spa_writeable(spa)); 6231 6232 /* 6233 * Lock out configuration changes. 6234 */ 6235 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6236 6237 spa->spa_syncing_txg = txg; 6238 spa->spa_sync_pass = 0; 6239 6240 /* 6241 * If there are any pending vdev state changes, convert them 6242 * into config changes that go out with this transaction group. 6243 */ 6244 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6245 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6246 /* 6247 * We need the write lock here because, for aux vdevs, 6248 * calling vdev_config_dirty() modifies sav_config. 6249 * This is ugly and will become unnecessary when we 6250 * eliminate the aux vdev wart by integrating all vdevs 6251 * into the root vdev tree. 6252 */ 6253 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6254 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6255 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6256 vdev_state_clean(vd); 6257 vdev_config_dirty(vd); 6258 } 6259 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6260 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6261 } 6262 spa_config_exit(spa, SCL_STATE, FTAG); 6263 6264 tx = dmu_tx_create_assigned(dp, txg); 6265 6266 spa->spa_sync_starttime = gethrtime(); 6267#ifdef illumos 6268 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6269 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6270#else /* FreeBSD */ 6271#ifdef _KERNEL 6272 callout_reset(&spa->spa_deadman_cycid, 6273 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6274#endif 6275#endif 6276 6277 /* 6278 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6279 * set spa_deflate if we have no raid-z vdevs. 6280 */ 6281 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6282 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6283 int i; 6284 6285 for (i = 0; i < rvd->vdev_children; i++) { 6286 vd = rvd->vdev_child[i]; 6287 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6288 break; 6289 } 6290 if (i == rvd->vdev_children) { 6291 spa->spa_deflate = TRUE; 6292 VERIFY(0 == zap_add(spa->spa_meta_objset, 6293 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6294 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6295 } 6296 } 6297 6298 /* 6299 * If anything has changed in this txg, or if someone is waiting 6300 * for this txg to sync (eg, spa_vdev_remove()), push the 6301 * deferred frees from the previous txg. If not, leave them 6302 * alone so that we don't generate work on an otherwise idle 6303 * system. 6304 */ 6305 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6306 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6307 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6308 ((dsl_scan_active(dp->dp_scan) || 6309 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6310 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6311 VERIFY3U(bpobj_iterate(defer_bpo, 6312 spa_free_sync_cb, zio, tx), ==, 0); 6313 VERIFY0(zio_wait(zio)); 6314 } 6315 6316 /* 6317 * Iterate to convergence. 6318 */ 6319 do { 6320 int pass = ++spa->spa_sync_pass; 6321 6322 spa_sync_config_object(spa, tx); 6323 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6324 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6325 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6326 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6327 spa_errlog_sync(spa, txg); 6328 dsl_pool_sync(dp, txg); 6329 6330 if (pass < zfs_sync_pass_deferred_free) { 6331 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6332 bplist_iterate(free_bpl, spa_free_sync_cb, 6333 zio, tx); 6334 VERIFY(zio_wait(zio) == 0); 6335 } else { 6336 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6337 defer_bpo, tx); 6338 } 6339 6340 ddt_sync(spa, txg); 6341 dsl_scan_sync(dp, tx); 6342 6343 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6344 vdev_sync(vd, txg); 6345 6346 if (pass == 1) 6347 spa_sync_upgrades(spa, tx); 6348 6349 } while (dmu_objset_is_dirty(mos, txg)); 6350 6351 /* 6352 * Rewrite the vdev configuration (which includes the uberblock) 6353 * to commit the transaction group. 6354 * 6355 * If there are no dirty vdevs, we sync the uberblock to a few 6356 * random top-level vdevs that are known to be visible in the 6357 * config cache (see spa_vdev_add() for a complete description). 6358 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6359 */ 6360 for (;;) { 6361 /* 6362 * We hold SCL_STATE to prevent vdev open/close/etc. 6363 * while we're attempting to write the vdev labels. 6364 */ 6365 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6366 6367 if (list_is_empty(&spa->spa_config_dirty_list)) { 6368 vdev_t *svd[SPA_DVAS_PER_BP]; 6369 int svdcount = 0; 6370 int children = rvd->vdev_children; 6371 int c0 = spa_get_random(children); 6372 6373 for (int c = 0; c < children; c++) { 6374 vd = rvd->vdev_child[(c0 + c) % children]; 6375 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6376 continue; 6377 svd[svdcount++] = vd; 6378 if (svdcount == SPA_DVAS_PER_BP) 6379 break; 6380 } 6381 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6382 if (error != 0) 6383 error = vdev_config_sync(svd, svdcount, txg, 6384 B_TRUE); 6385 } else { 6386 error = vdev_config_sync(rvd->vdev_child, 6387 rvd->vdev_children, txg, B_FALSE); 6388 if (error != 0) 6389 error = vdev_config_sync(rvd->vdev_child, 6390 rvd->vdev_children, txg, B_TRUE); 6391 } 6392 6393 if (error == 0) 6394 spa->spa_last_synced_guid = rvd->vdev_guid; 6395 6396 spa_config_exit(spa, SCL_STATE, FTAG); 6397 6398 if (error == 0) 6399 break; 6400 zio_suspend(spa, NULL); 6401 zio_resume_wait(spa); 6402 } 6403 dmu_tx_commit(tx); 6404 6405#ifdef illumos 6406 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6407#else /* FreeBSD */ 6408#ifdef _KERNEL 6409 callout_drain(&spa->spa_deadman_cycid); 6410#endif 6411#endif 6412 6413 /* 6414 * Clear the dirty config list. 6415 */ 6416 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6417 vdev_config_clean(vd); 6418 6419 /* 6420 * Now that the new config has synced transactionally, 6421 * let it become visible to the config cache. 6422 */ 6423 if (spa->spa_config_syncing != NULL) { 6424 spa_config_set(spa, spa->spa_config_syncing); 6425 spa->spa_config_txg = txg; 6426 spa->spa_config_syncing = NULL; 6427 } 6428 6429 spa->spa_ubsync = spa->spa_uberblock; 6430 6431 dsl_pool_sync_done(dp, txg); 6432 6433 /* 6434 * Update usable space statistics. 6435 */ 6436 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6437 vdev_sync_done(vd, txg); 6438 6439 spa_update_dspace(spa); 6440 6441 /* 6442 * It had better be the case that we didn't dirty anything 6443 * since vdev_config_sync(). 6444 */ 6445 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6446 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6447 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6448 6449 spa->spa_sync_pass = 0; 6450 6451 spa_config_exit(spa, SCL_CONFIG, FTAG); 6452 6453 spa_handle_ignored_writes(spa); 6454 6455 /* 6456 * If any async tasks have been requested, kick them off. 6457 */ 6458 spa_async_dispatch(spa); 6459} 6460 6461/* 6462 * Sync all pools. We don't want to hold the namespace lock across these 6463 * operations, so we take a reference on the spa_t and drop the lock during the 6464 * sync. 6465 */ 6466void 6467spa_sync_allpools(void) 6468{ 6469 spa_t *spa = NULL; 6470 mutex_enter(&spa_namespace_lock); 6471 while ((spa = spa_next(spa)) != NULL) { 6472 if (spa_state(spa) != POOL_STATE_ACTIVE || 6473 !spa_writeable(spa) || spa_suspended(spa)) 6474 continue; 6475 spa_open_ref(spa, FTAG); 6476 mutex_exit(&spa_namespace_lock); 6477 txg_wait_synced(spa_get_dsl(spa), 0); 6478 mutex_enter(&spa_namespace_lock); 6479 spa_close(spa, FTAG); 6480 } 6481 mutex_exit(&spa_namespace_lock); 6482} 6483 6484/* 6485 * ========================================================================== 6486 * Miscellaneous routines 6487 * ========================================================================== 6488 */ 6489 6490/* 6491 * Remove all pools in the system. 6492 */ 6493void 6494spa_evict_all(void) 6495{ 6496 spa_t *spa; 6497 6498 /* 6499 * Remove all cached state. All pools should be closed now, 6500 * so every spa in the AVL tree should be unreferenced. 6501 */ 6502 mutex_enter(&spa_namespace_lock); 6503 while ((spa = spa_next(NULL)) != NULL) { 6504 /* 6505 * Stop async tasks. The async thread may need to detach 6506 * a device that's been replaced, which requires grabbing 6507 * spa_namespace_lock, so we must drop it here. 6508 */ 6509 spa_open_ref(spa, FTAG); 6510 mutex_exit(&spa_namespace_lock); 6511 spa_async_suspend(spa); 6512 mutex_enter(&spa_namespace_lock); 6513 spa_close(spa, FTAG); 6514 6515 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6516 spa_unload(spa); 6517 spa_deactivate(spa); 6518 } 6519 spa_remove(spa); 6520 } 6521 mutex_exit(&spa_namespace_lock); 6522} 6523 6524vdev_t * 6525spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6526{ 6527 vdev_t *vd; 6528 int i; 6529 6530 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6531 return (vd); 6532 6533 if (aux) { 6534 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6535 vd = spa->spa_l2cache.sav_vdevs[i]; 6536 if (vd->vdev_guid == guid) 6537 return (vd); 6538 } 6539 6540 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6541 vd = spa->spa_spares.sav_vdevs[i]; 6542 if (vd->vdev_guid == guid) 6543 return (vd); 6544 } 6545 } 6546 6547 return (NULL); 6548} 6549 6550void 6551spa_upgrade(spa_t *spa, uint64_t version) 6552{ 6553 ASSERT(spa_writeable(spa)); 6554 6555 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6556 6557 /* 6558 * This should only be called for a non-faulted pool, and since a 6559 * future version would result in an unopenable pool, this shouldn't be 6560 * possible. 6561 */ 6562 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6563 ASSERT(version >= spa->spa_uberblock.ub_version); 6564 6565 spa->spa_uberblock.ub_version = version; 6566 vdev_config_dirty(spa->spa_root_vdev); 6567 6568 spa_config_exit(spa, SCL_ALL, FTAG); 6569 6570 txg_wait_synced(spa_get_dsl(spa), 0); 6571} 6572 6573boolean_t 6574spa_has_spare(spa_t *spa, uint64_t guid) 6575{ 6576 int i; 6577 uint64_t spareguid; 6578 spa_aux_vdev_t *sav = &spa->spa_spares; 6579 6580 for (i = 0; i < sav->sav_count; i++) 6581 if (sav->sav_vdevs[i]->vdev_guid == guid) 6582 return (B_TRUE); 6583 6584 for (i = 0; i < sav->sav_npending; i++) { 6585 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6586 &spareguid) == 0 && spareguid == guid) 6587 return (B_TRUE); 6588 } 6589 6590 return (B_FALSE); 6591} 6592 6593/* 6594 * Check if a pool has an active shared spare device. 6595 * Note: reference count of an active spare is 2, as a spare and as a replace 6596 */ 6597static boolean_t 6598spa_has_active_shared_spare(spa_t *spa) 6599{ 6600 int i, refcnt; 6601 uint64_t pool; 6602 spa_aux_vdev_t *sav = &spa->spa_spares; 6603 6604 for (i = 0; i < sav->sav_count; i++) { 6605 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6606 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6607 refcnt > 2) 6608 return (B_TRUE); 6609 } 6610 6611 return (B_FALSE); 6612} 6613 6614/* 6615 * Post a sysevent corresponding to the given event. The 'name' must be one of 6616 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6617 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6618 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6619 * or zdb as real changes. 6620 */ 6621void 6622spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6623{ 6624#ifdef _KERNEL 6625 sysevent_t *ev; 6626 sysevent_attr_list_t *attr = NULL; 6627 sysevent_value_t value; 6628 sysevent_id_t eid; 6629 6630 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6631 SE_SLEEP); 6632 6633 value.value_type = SE_DATA_TYPE_STRING; 6634 value.value.sv_string = spa_name(spa); 6635 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6636 goto done; 6637 6638 value.value_type = SE_DATA_TYPE_UINT64; 6639 value.value.sv_uint64 = spa_guid(spa); 6640 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6641 goto done; 6642 6643 if (vd) { 6644 value.value_type = SE_DATA_TYPE_UINT64; 6645 value.value.sv_uint64 = vd->vdev_guid; 6646 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6647 SE_SLEEP) != 0) 6648 goto done; 6649 6650 if (vd->vdev_path) { 6651 value.value_type = SE_DATA_TYPE_STRING; 6652 value.value.sv_string = vd->vdev_path; 6653 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6654 &value, SE_SLEEP) != 0) 6655 goto done; 6656 } 6657 } 6658 6659 if (sysevent_attach_attributes(ev, attr) != 0) 6660 goto done; 6661 attr = NULL; 6662 6663 (void) log_sysevent(ev, SE_SLEEP, &eid); 6664 6665done: 6666 if (attr) 6667 sysevent_free_attr(attr); 6668 sysevent_free(ev); 6669#endif 6670} 6671