spa.c revision 240415
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 */ 26 27/* 28 * This file contains all the routines used when modifying on-disk SPA state. 29 * This includes opening, importing, destroying, exporting a pool, and syncing a 30 * pool. 31 */ 32 33#include <sys/zfs_context.h> 34#include <sys/fm/fs/zfs.h> 35#include <sys/spa_impl.h> 36#include <sys/zio.h> 37#include <sys/zio_checksum.h> 38#include <sys/dmu.h> 39#include <sys/dmu_tx.h> 40#include <sys/zap.h> 41#include <sys/zil.h> 42#include <sys/ddt.h> 43#include <sys/vdev_impl.h> 44#include <sys/metaslab.h> 45#include <sys/metaslab_impl.h> 46#include <sys/uberblock_impl.h> 47#include <sys/txg.h> 48#include <sys/avl.h> 49#include <sys/dmu_traverse.h> 50#include <sys/dmu_objset.h> 51#include <sys/unique.h> 52#include <sys/dsl_pool.h> 53#include <sys/dsl_dataset.h> 54#include <sys/dsl_dir.h> 55#include <sys/dsl_prop.h> 56#include <sys/dsl_synctask.h> 57#include <sys/fs/zfs.h> 58#include <sys/arc.h> 59#include <sys/callb.h> 60#include <sys/spa_boot.h> 61#include <sys/zfs_ioctl.h> 62#include <sys/dsl_scan.h> 63#include <sys/zfeature.h> 64#include <sys/zvol.h> 65 66#ifdef _KERNEL 67#include <sys/callb.h> 68#include <sys/cpupart.h> 69#include <sys/zone.h> 70#endif /* _KERNEL */ 71 72#include "zfs_prop.h" 73#include "zfs_comutil.h" 74 75/* Check hostid on import? */ 76static int check_hostid = 1; 77 78SYSCTL_DECL(_vfs_zfs); 79TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 80SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 81 "Check hostid on import?"); 82 83typedef enum zti_modes { 84 zti_mode_fixed, /* value is # of threads (min 1) */ 85 zti_mode_online_percent, /* value is % of online CPUs */ 86 zti_mode_batch, /* cpu-intensive; value is ignored */ 87 zti_mode_null, /* don't create a taskq */ 88 zti_nmodes 89} zti_modes_t; 90 91#define ZTI_FIX(n) { zti_mode_fixed, (n) } 92#define ZTI_PCT(n) { zti_mode_online_percent, (n) } 93#define ZTI_BATCH { zti_mode_batch, 0 } 94#define ZTI_NULL { zti_mode_null, 0 } 95 96#define ZTI_ONE ZTI_FIX(1) 97 98typedef struct zio_taskq_info { 99 enum zti_modes zti_mode; 100 uint_t zti_value; 101} zio_taskq_info_t; 102 103static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 104 "issue", "issue_high", "intr", "intr_high" 105}; 106 107/* 108 * Define the taskq threads for the following I/O types: 109 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL 110 */ 111const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 112 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 113 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 114 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, 115 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) }, 116 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL }, 117 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 119}; 120 121static dsl_syncfunc_t spa_sync_version; 122static dsl_syncfunc_t spa_sync_props; 123static dsl_checkfunc_t spa_change_guid_check; 124static dsl_syncfunc_t spa_change_guid_sync; 125static boolean_t spa_has_active_shared_spare(spa_t *spa); 126static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 127 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 128 char **ereport); 129static void spa_vdev_resilver_done(spa_t *spa); 130 131uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */ 132#ifdef PSRSET_BIND 133id_t zio_taskq_psrset_bind = PS_NONE; 134#endif 135#ifdef SYSDC 136boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 137#endif 138uint_t zio_taskq_basedc = 80; /* base duty cycle */ 139 140boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 141 142/* 143 * This (illegal) pool name is used when temporarily importing a spa_t in order 144 * to get the vdev stats associated with the imported devices. 145 */ 146#define TRYIMPORT_NAME "$import" 147 148/* 149 * ========================================================================== 150 * SPA properties routines 151 * ========================================================================== 152 */ 153 154/* 155 * Add a (source=src, propname=propval) list to an nvlist. 156 */ 157static void 158spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 159 uint64_t intval, zprop_source_t src) 160{ 161 const char *propname = zpool_prop_to_name(prop); 162 nvlist_t *propval; 163 164 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 165 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 166 167 if (strval != NULL) 168 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 169 else 170 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 171 172 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 173 nvlist_free(propval); 174} 175 176/* 177 * Get property values from the spa configuration. 178 */ 179static void 180spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 181{ 182 vdev_t *rvd = spa->spa_root_vdev; 183 dsl_pool_t *pool = spa->spa_dsl_pool; 184 uint64_t size; 185 uint64_t alloc; 186 uint64_t space; 187 uint64_t cap, version; 188 zprop_source_t src = ZPROP_SRC_NONE; 189 spa_config_dirent_t *dp; 190 191 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 192 193 if (rvd != NULL) { 194 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 195 size = metaslab_class_get_space(spa_normal_class(spa)); 196 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 197 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 198 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 199 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 200 size - alloc, src); 201 202 space = 0; 203 for (int c = 0; c < rvd->vdev_children; c++) { 204 vdev_t *tvd = rvd->vdev_child[c]; 205 space += tvd->vdev_max_asize - tvd->vdev_asize; 206 } 207 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space, 208 src); 209 210 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 211 (spa_mode(spa) == FREAD), src); 212 213 cap = (size == 0) ? 0 : (alloc * 100 / size); 214 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 215 216 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 217 ddt_get_pool_dedup_ratio(spa), src); 218 219 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 220 rvd->vdev_state, src); 221 222 version = spa_version(spa); 223 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 224 src = ZPROP_SRC_DEFAULT; 225 else 226 src = ZPROP_SRC_LOCAL; 227 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 228 } 229 230 if (pool != NULL) { 231 dsl_dir_t *freedir = pool->dp_free_dir; 232 233 /* 234 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 235 * when opening pools before this version freedir will be NULL. 236 */ 237 if (freedir != NULL) { 238 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 239 freedir->dd_phys->dd_used_bytes, src); 240 } else { 241 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 242 NULL, 0, src); 243 } 244 } 245 246 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 247 248 if (spa->spa_comment != NULL) { 249 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 250 0, ZPROP_SRC_LOCAL); 251 } 252 253 if (spa->spa_root != NULL) 254 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 255 0, ZPROP_SRC_LOCAL); 256 257 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 258 if (dp->scd_path == NULL) { 259 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 260 "none", 0, ZPROP_SRC_LOCAL); 261 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 262 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 263 dp->scd_path, 0, ZPROP_SRC_LOCAL); 264 } 265 } 266} 267 268/* 269 * Get zpool property values. 270 */ 271int 272spa_prop_get(spa_t *spa, nvlist_t **nvp) 273{ 274 objset_t *mos = spa->spa_meta_objset; 275 zap_cursor_t zc; 276 zap_attribute_t za; 277 int err; 278 279 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 280 281 mutex_enter(&spa->spa_props_lock); 282 283 /* 284 * Get properties from the spa config. 285 */ 286 spa_prop_get_config(spa, nvp); 287 288 /* If no pool property object, no more prop to get. */ 289 if (mos == NULL || spa->spa_pool_props_object == 0) { 290 mutex_exit(&spa->spa_props_lock); 291 return (0); 292 } 293 294 /* 295 * Get properties from the MOS pool property object. 296 */ 297 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 298 (err = zap_cursor_retrieve(&zc, &za)) == 0; 299 zap_cursor_advance(&zc)) { 300 uint64_t intval = 0; 301 char *strval = NULL; 302 zprop_source_t src = ZPROP_SRC_DEFAULT; 303 zpool_prop_t prop; 304 305 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 306 continue; 307 308 switch (za.za_integer_length) { 309 case 8: 310 /* integer property */ 311 if (za.za_first_integer != 312 zpool_prop_default_numeric(prop)) 313 src = ZPROP_SRC_LOCAL; 314 315 if (prop == ZPOOL_PROP_BOOTFS) { 316 dsl_pool_t *dp; 317 dsl_dataset_t *ds = NULL; 318 319 dp = spa_get_dsl(spa); 320 rw_enter(&dp->dp_config_rwlock, RW_READER); 321 if (err = dsl_dataset_hold_obj(dp, 322 za.za_first_integer, FTAG, &ds)) { 323 rw_exit(&dp->dp_config_rwlock); 324 break; 325 } 326 327 strval = kmem_alloc( 328 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 329 KM_SLEEP); 330 dsl_dataset_name(ds, strval); 331 dsl_dataset_rele(ds, FTAG); 332 rw_exit(&dp->dp_config_rwlock); 333 } else { 334 strval = NULL; 335 intval = za.za_first_integer; 336 } 337 338 spa_prop_add_list(*nvp, prop, strval, intval, src); 339 340 if (strval != NULL) 341 kmem_free(strval, 342 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 343 344 break; 345 346 case 1: 347 /* string property */ 348 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 349 err = zap_lookup(mos, spa->spa_pool_props_object, 350 za.za_name, 1, za.za_num_integers, strval); 351 if (err) { 352 kmem_free(strval, za.za_num_integers); 353 break; 354 } 355 spa_prop_add_list(*nvp, prop, strval, 0, src); 356 kmem_free(strval, za.za_num_integers); 357 break; 358 359 default: 360 break; 361 } 362 } 363 zap_cursor_fini(&zc); 364 mutex_exit(&spa->spa_props_lock); 365out: 366 if (err && err != ENOENT) { 367 nvlist_free(*nvp); 368 *nvp = NULL; 369 return (err); 370 } 371 372 return (0); 373} 374 375/* 376 * Validate the given pool properties nvlist and modify the list 377 * for the property values to be set. 378 */ 379static int 380spa_prop_validate(spa_t *spa, nvlist_t *props) 381{ 382 nvpair_t *elem; 383 int error = 0, reset_bootfs = 0; 384 uint64_t objnum; 385 boolean_t has_feature = B_FALSE; 386 387 elem = NULL; 388 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 389 uint64_t intval; 390 char *strval, *slash, *check, *fname; 391 const char *propname = nvpair_name(elem); 392 zpool_prop_t prop = zpool_name_to_prop(propname); 393 394 switch (prop) { 395 case ZPROP_INVAL: 396 if (!zpool_prop_feature(propname)) { 397 error = EINVAL; 398 break; 399 } 400 401 /* 402 * Sanitize the input. 403 */ 404 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 405 error = EINVAL; 406 break; 407 } 408 409 if (nvpair_value_uint64(elem, &intval) != 0) { 410 error = EINVAL; 411 break; 412 } 413 414 if (intval != 0) { 415 error = EINVAL; 416 break; 417 } 418 419 fname = strchr(propname, '@') + 1; 420 if (zfeature_lookup_name(fname, NULL) != 0) { 421 error = EINVAL; 422 break; 423 } 424 425 has_feature = B_TRUE; 426 break; 427 428 case ZPOOL_PROP_VERSION: 429 error = nvpair_value_uint64(elem, &intval); 430 if (!error && 431 (intval < spa_version(spa) || 432 intval > SPA_VERSION_BEFORE_FEATURES || 433 has_feature)) 434 error = EINVAL; 435 break; 436 437 case ZPOOL_PROP_DELEGATION: 438 case ZPOOL_PROP_AUTOREPLACE: 439 case ZPOOL_PROP_LISTSNAPS: 440 case ZPOOL_PROP_AUTOEXPAND: 441 error = nvpair_value_uint64(elem, &intval); 442 if (!error && intval > 1) 443 error = EINVAL; 444 break; 445 446 case ZPOOL_PROP_BOOTFS: 447 /* 448 * If the pool version is less than SPA_VERSION_BOOTFS, 449 * or the pool is still being created (version == 0), 450 * the bootfs property cannot be set. 451 */ 452 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 453 error = ENOTSUP; 454 break; 455 } 456 457 /* 458 * Make sure the vdev config is bootable 459 */ 460 if (!vdev_is_bootable(spa->spa_root_vdev)) { 461 error = ENOTSUP; 462 break; 463 } 464 465 reset_bootfs = 1; 466 467 error = nvpair_value_string(elem, &strval); 468 469 if (!error) { 470 objset_t *os; 471 uint64_t compress; 472 473 if (strval == NULL || strval[0] == '\0') { 474 objnum = zpool_prop_default_numeric( 475 ZPOOL_PROP_BOOTFS); 476 break; 477 } 478 479 if (error = dmu_objset_hold(strval, FTAG, &os)) 480 break; 481 482 /* Must be ZPL and not gzip compressed. */ 483 484 if (dmu_objset_type(os) != DMU_OST_ZFS) { 485 error = ENOTSUP; 486 } else if ((error = dsl_prop_get_integer(strval, 487 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 488 &compress, NULL)) == 0 && 489 !BOOTFS_COMPRESS_VALID(compress)) { 490 error = ENOTSUP; 491 } else { 492 objnum = dmu_objset_id(os); 493 } 494 dmu_objset_rele(os, FTAG); 495 } 496 break; 497 498 case ZPOOL_PROP_FAILUREMODE: 499 error = nvpair_value_uint64(elem, &intval); 500 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 501 intval > ZIO_FAILURE_MODE_PANIC)) 502 error = EINVAL; 503 504 /* 505 * This is a special case which only occurs when 506 * the pool has completely failed. This allows 507 * the user to change the in-core failmode property 508 * without syncing it out to disk (I/Os might 509 * currently be blocked). We do this by returning 510 * EIO to the caller (spa_prop_set) to trick it 511 * into thinking we encountered a property validation 512 * error. 513 */ 514 if (!error && spa_suspended(spa)) { 515 spa->spa_failmode = intval; 516 error = EIO; 517 } 518 break; 519 520 case ZPOOL_PROP_CACHEFILE: 521 if ((error = nvpair_value_string(elem, &strval)) != 0) 522 break; 523 524 if (strval[0] == '\0') 525 break; 526 527 if (strcmp(strval, "none") == 0) 528 break; 529 530 if (strval[0] != '/') { 531 error = EINVAL; 532 break; 533 } 534 535 slash = strrchr(strval, '/'); 536 ASSERT(slash != NULL); 537 538 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 539 strcmp(slash, "/..") == 0) 540 error = EINVAL; 541 break; 542 543 case ZPOOL_PROP_COMMENT: 544 if ((error = nvpair_value_string(elem, &strval)) != 0) 545 break; 546 for (check = strval; *check != '\0'; check++) { 547 /* 548 * The kernel doesn't have an easy isprint() 549 * check. For this kernel check, we merely 550 * check ASCII apart from DEL. Fix this if 551 * there is an easy-to-use kernel isprint(). 552 */ 553 if (*check >= 0x7f) { 554 error = EINVAL; 555 break; 556 } 557 check++; 558 } 559 if (strlen(strval) > ZPROP_MAX_COMMENT) 560 error = E2BIG; 561 break; 562 563 case ZPOOL_PROP_DEDUPDITTO: 564 if (spa_version(spa) < SPA_VERSION_DEDUP) 565 error = ENOTSUP; 566 else 567 error = nvpair_value_uint64(elem, &intval); 568 if (error == 0 && 569 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 570 error = EINVAL; 571 break; 572 } 573 574 if (error) 575 break; 576 } 577 578 if (!error && reset_bootfs) { 579 error = nvlist_remove(props, 580 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 581 582 if (!error) { 583 error = nvlist_add_uint64(props, 584 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 585 } 586 } 587 588 return (error); 589} 590 591void 592spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 593{ 594 char *cachefile; 595 spa_config_dirent_t *dp; 596 597 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 598 &cachefile) != 0) 599 return; 600 601 dp = kmem_alloc(sizeof (spa_config_dirent_t), 602 KM_SLEEP); 603 604 if (cachefile[0] == '\0') 605 dp->scd_path = spa_strdup(spa_config_path); 606 else if (strcmp(cachefile, "none") == 0) 607 dp->scd_path = NULL; 608 else 609 dp->scd_path = spa_strdup(cachefile); 610 611 list_insert_head(&spa->spa_config_list, dp); 612 if (need_sync) 613 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 614} 615 616int 617spa_prop_set(spa_t *spa, nvlist_t *nvp) 618{ 619 int error; 620 nvpair_t *elem = NULL; 621 boolean_t need_sync = B_FALSE; 622 623 if ((error = spa_prop_validate(spa, nvp)) != 0) 624 return (error); 625 626 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 627 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 628 629 if (prop == ZPOOL_PROP_CACHEFILE || 630 prop == ZPOOL_PROP_ALTROOT || 631 prop == ZPOOL_PROP_READONLY) 632 continue; 633 634 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 635 uint64_t ver; 636 637 if (prop == ZPOOL_PROP_VERSION) { 638 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 639 } else { 640 ASSERT(zpool_prop_feature(nvpair_name(elem))); 641 ver = SPA_VERSION_FEATURES; 642 need_sync = B_TRUE; 643 } 644 645 /* Save time if the version is already set. */ 646 if (ver == spa_version(spa)) 647 continue; 648 649 /* 650 * In addition to the pool directory object, we might 651 * create the pool properties object, the features for 652 * read object, the features for write object, or the 653 * feature descriptions object. 654 */ 655 error = dsl_sync_task_do(spa_get_dsl(spa), NULL, 656 spa_sync_version, spa, &ver, 6); 657 if (error) 658 return (error); 659 continue; 660 } 661 662 need_sync = B_TRUE; 663 break; 664 } 665 666 if (need_sync) { 667 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, 668 spa, nvp, 6)); 669 } 670 671 return (0); 672} 673 674/* 675 * If the bootfs property value is dsobj, clear it. 676 */ 677void 678spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 679{ 680 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 681 VERIFY(zap_remove(spa->spa_meta_objset, 682 spa->spa_pool_props_object, 683 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 684 spa->spa_bootfs = 0; 685 } 686} 687 688/*ARGSUSED*/ 689static int 690spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx) 691{ 692 spa_t *spa = arg1; 693 uint64_t *newguid = arg2; 694 vdev_t *rvd = spa->spa_root_vdev; 695 uint64_t vdev_state; 696 697 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 698 vdev_state = rvd->vdev_state; 699 spa_config_exit(spa, SCL_STATE, FTAG); 700 701 if (vdev_state != VDEV_STATE_HEALTHY) 702 return (ENXIO); 703 704 ASSERT3U(spa_guid(spa), !=, *newguid); 705 706 return (0); 707} 708 709static void 710spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx) 711{ 712 spa_t *spa = arg1; 713 uint64_t *newguid = arg2; 714 uint64_t oldguid; 715 vdev_t *rvd = spa->spa_root_vdev; 716 717 oldguid = spa_guid(spa); 718 719 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 720 rvd->vdev_guid = *newguid; 721 rvd->vdev_guid_sum += (*newguid - oldguid); 722 vdev_config_dirty(rvd); 723 spa_config_exit(spa, SCL_STATE, FTAG); 724 725#ifdef __FreeBSD__ 726 /* 727 * TODO: until recent illumos logging changes are merged 728 * log reguid as pool property change 729 */ 730 spa_history_log_internal(LOG_POOL_PROPSET, spa, tx, 731 "guid change old=%llu new=%llu", oldguid, *newguid); 732#else 733 spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld", 734 oldguid, *newguid); 735#endif 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_do(spa_get_dsl(spa), spa_change_guid_check, 757 spa_change_guid_sync, spa, &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 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1005 offsetof(vdev_t, vdev_config_dirty_node)); 1006 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1007 offsetof(vdev_t, vdev_state_dirty_node)); 1008 1009 txg_list_create(&spa->spa_vdev_txg_list, 1010 offsetof(struct vdev, vdev_txg_node)); 1011 1012 avl_create(&spa->spa_errlist_scrub, 1013 spa_error_entry_compare, sizeof (spa_error_entry_t), 1014 offsetof(spa_error_entry_t, se_avl)); 1015 avl_create(&spa->spa_errlist_last, 1016 spa_error_entry_compare, sizeof (spa_error_entry_t), 1017 offsetof(spa_error_entry_t, se_avl)); 1018} 1019 1020/* 1021 * Opposite of spa_activate(). 1022 */ 1023static void 1024spa_deactivate(spa_t *spa) 1025{ 1026 ASSERT(spa->spa_sync_on == B_FALSE); 1027 ASSERT(spa->spa_dsl_pool == NULL); 1028 ASSERT(spa->spa_root_vdev == NULL); 1029 ASSERT(spa->spa_async_zio_root == NULL); 1030 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1031 1032 txg_list_destroy(&spa->spa_vdev_txg_list); 1033 1034 list_destroy(&spa->spa_config_dirty_list); 1035 list_destroy(&spa->spa_state_dirty_list); 1036 1037 for (int t = 0; t < ZIO_TYPES; t++) { 1038 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1039 if (spa->spa_zio_taskq[t][q] != NULL) 1040 taskq_destroy(spa->spa_zio_taskq[t][q]); 1041 spa->spa_zio_taskq[t][q] = NULL; 1042 } 1043 } 1044 1045 metaslab_class_destroy(spa->spa_normal_class); 1046 spa->spa_normal_class = NULL; 1047 1048 metaslab_class_destroy(spa->spa_log_class); 1049 spa->spa_log_class = NULL; 1050 1051 /* 1052 * If this was part of an import or the open otherwise failed, we may 1053 * still have errors left in the queues. Empty them just in case. 1054 */ 1055 spa_errlog_drain(spa); 1056 1057 avl_destroy(&spa->spa_errlist_scrub); 1058 avl_destroy(&spa->spa_errlist_last); 1059 1060 spa->spa_state = POOL_STATE_UNINITIALIZED; 1061 1062 mutex_enter(&spa->spa_proc_lock); 1063 if (spa->spa_proc_state != SPA_PROC_NONE) { 1064 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1065 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1066 cv_broadcast(&spa->spa_proc_cv); 1067 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1068 ASSERT(spa->spa_proc != &p0); 1069 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1070 } 1071 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1072 spa->spa_proc_state = SPA_PROC_NONE; 1073 } 1074 ASSERT(spa->spa_proc == &p0); 1075 mutex_exit(&spa->spa_proc_lock); 1076 1077#ifdef SPA_PROCESS 1078 /* 1079 * We want to make sure spa_thread() has actually exited the ZFS 1080 * module, so that the module can't be unloaded out from underneath 1081 * it. 1082 */ 1083 if (spa->spa_did != 0) { 1084 thread_join(spa->spa_did); 1085 spa->spa_did = 0; 1086 } 1087#endif /* SPA_PROCESS */ 1088} 1089 1090/* 1091 * Verify a pool configuration, and construct the vdev tree appropriately. This 1092 * will create all the necessary vdevs in the appropriate layout, with each vdev 1093 * in the CLOSED state. This will prep the pool before open/creation/import. 1094 * All vdev validation is done by the vdev_alloc() routine. 1095 */ 1096static int 1097spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1098 uint_t id, int atype) 1099{ 1100 nvlist_t **child; 1101 uint_t children; 1102 int error; 1103 1104 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1105 return (error); 1106 1107 if ((*vdp)->vdev_ops->vdev_op_leaf) 1108 return (0); 1109 1110 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1111 &child, &children); 1112 1113 if (error == ENOENT) 1114 return (0); 1115 1116 if (error) { 1117 vdev_free(*vdp); 1118 *vdp = NULL; 1119 return (EINVAL); 1120 } 1121 1122 for (int c = 0; c < children; c++) { 1123 vdev_t *vd; 1124 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1125 atype)) != 0) { 1126 vdev_free(*vdp); 1127 *vdp = NULL; 1128 return (error); 1129 } 1130 } 1131 1132 ASSERT(*vdp != NULL); 1133 1134 return (0); 1135} 1136 1137/* 1138 * Opposite of spa_load(). 1139 */ 1140static void 1141spa_unload(spa_t *spa) 1142{ 1143 int i; 1144 1145 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1146 1147 /* 1148 * Stop async tasks. 1149 */ 1150 spa_async_suspend(spa); 1151 1152 /* 1153 * Stop syncing. 1154 */ 1155 if (spa->spa_sync_on) { 1156 txg_sync_stop(spa->spa_dsl_pool); 1157 spa->spa_sync_on = B_FALSE; 1158 } 1159 1160 /* 1161 * Wait for any outstanding async I/O to complete. 1162 */ 1163 if (spa->spa_async_zio_root != NULL) { 1164 (void) zio_wait(spa->spa_async_zio_root); 1165 spa->spa_async_zio_root = NULL; 1166 } 1167 1168 bpobj_close(&spa->spa_deferred_bpobj); 1169 1170 /* 1171 * Close the dsl pool. 1172 */ 1173 if (spa->spa_dsl_pool) { 1174 dsl_pool_close(spa->spa_dsl_pool); 1175 spa->spa_dsl_pool = NULL; 1176 spa->spa_meta_objset = NULL; 1177 } 1178 1179 ddt_unload(spa); 1180 1181 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1182 1183 /* 1184 * Drop and purge level 2 cache 1185 */ 1186 spa_l2cache_drop(spa); 1187 1188 /* 1189 * Close all vdevs. 1190 */ 1191 if (spa->spa_root_vdev) 1192 vdev_free(spa->spa_root_vdev); 1193 ASSERT(spa->spa_root_vdev == NULL); 1194 1195 for (i = 0; i < spa->spa_spares.sav_count; i++) 1196 vdev_free(spa->spa_spares.sav_vdevs[i]); 1197 if (spa->spa_spares.sav_vdevs) { 1198 kmem_free(spa->spa_spares.sav_vdevs, 1199 spa->spa_spares.sav_count * sizeof (void *)); 1200 spa->spa_spares.sav_vdevs = NULL; 1201 } 1202 if (spa->spa_spares.sav_config) { 1203 nvlist_free(spa->spa_spares.sav_config); 1204 spa->spa_spares.sav_config = NULL; 1205 } 1206 spa->spa_spares.sav_count = 0; 1207 1208 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1209 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1210 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1211 } 1212 if (spa->spa_l2cache.sav_vdevs) { 1213 kmem_free(spa->spa_l2cache.sav_vdevs, 1214 spa->spa_l2cache.sav_count * sizeof (void *)); 1215 spa->spa_l2cache.sav_vdevs = NULL; 1216 } 1217 if (spa->spa_l2cache.sav_config) { 1218 nvlist_free(spa->spa_l2cache.sav_config); 1219 spa->spa_l2cache.sav_config = NULL; 1220 } 1221 spa->spa_l2cache.sav_count = 0; 1222 1223 spa->spa_async_suspended = 0; 1224 1225 if (spa->spa_comment != NULL) { 1226 spa_strfree(spa->spa_comment); 1227 spa->spa_comment = NULL; 1228 } 1229 1230 spa_config_exit(spa, SCL_ALL, FTAG); 1231} 1232 1233/* 1234 * Load (or re-load) the current list of vdevs describing the active spares for 1235 * this pool. When this is called, we have some form of basic information in 1236 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1237 * then re-generate a more complete list including status information. 1238 */ 1239static void 1240spa_load_spares(spa_t *spa) 1241{ 1242 nvlist_t **spares; 1243 uint_t nspares; 1244 int i; 1245 vdev_t *vd, *tvd; 1246 1247 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1248 1249 /* 1250 * First, close and free any existing spare vdevs. 1251 */ 1252 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1253 vd = spa->spa_spares.sav_vdevs[i]; 1254 1255 /* Undo the call to spa_activate() below */ 1256 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1257 B_FALSE)) != NULL && tvd->vdev_isspare) 1258 spa_spare_remove(tvd); 1259 vdev_close(vd); 1260 vdev_free(vd); 1261 } 1262 1263 if (spa->spa_spares.sav_vdevs) 1264 kmem_free(spa->spa_spares.sav_vdevs, 1265 spa->spa_spares.sav_count * sizeof (void *)); 1266 1267 if (spa->spa_spares.sav_config == NULL) 1268 nspares = 0; 1269 else 1270 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1271 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1272 1273 spa->spa_spares.sav_count = (int)nspares; 1274 spa->spa_spares.sav_vdevs = NULL; 1275 1276 if (nspares == 0) 1277 return; 1278 1279 /* 1280 * Construct the array of vdevs, opening them to get status in the 1281 * process. For each spare, there is potentially two different vdev_t 1282 * structures associated with it: one in the list of spares (used only 1283 * for basic validation purposes) and one in the active vdev 1284 * configuration (if it's spared in). During this phase we open and 1285 * validate each vdev on the spare list. If the vdev also exists in the 1286 * active configuration, then we also mark this vdev as an active spare. 1287 */ 1288 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1289 KM_SLEEP); 1290 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1291 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1292 VDEV_ALLOC_SPARE) == 0); 1293 ASSERT(vd != NULL); 1294 1295 spa->spa_spares.sav_vdevs[i] = vd; 1296 1297 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1298 B_FALSE)) != NULL) { 1299 if (!tvd->vdev_isspare) 1300 spa_spare_add(tvd); 1301 1302 /* 1303 * We only mark the spare active if we were successfully 1304 * able to load the vdev. Otherwise, importing a pool 1305 * with a bad active spare would result in strange 1306 * behavior, because multiple pool would think the spare 1307 * is actively in use. 1308 * 1309 * There is a vulnerability here to an equally bizarre 1310 * circumstance, where a dead active spare is later 1311 * brought back to life (onlined or otherwise). Given 1312 * the rarity of this scenario, and the extra complexity 1313 * it adds, we ignore the possibility. 1314 */ 1315 if (!vdev_is_dead(tvd)) 1316 spa_spare_activate(tvd); 1317 } 1318 1319 vd->vdev_top = vd; 1320 vd->vdev_aux = &spa->spa_spares; 1321 1322 if (vdev_open(vd) != 0) 1323 continue; 1324 1325 if (vdev_validate_aux(vd) == 0) 1326 spa_spare_add(vd); 1327 } 1328 1329 /* 1330 * Recompute the stashed list of spares, with status information 1331 * this time. 1332 */ 1333 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1334 DATA_TYPE_NVLIST_ARRAY) == 0); 1335 1336 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1337 KM_SLEEP); 1338 for (i = 0; i < spa->spa_spares.sav_count; i++) 1339 spares[i] = vdev_config_generate(spa, 1340 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1341 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1342 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1343 for (i = 0; i < spa->spa_spares.sav_count; i++) 1344 nvlist_free(spares[i]); 1345 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1346} 1347 1348/* 1349 * Load (or re-load) the current list of vdevs describing the active l2cache for 1350 * this pool. When this is called, we have some form of basic information in 1351 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1352 * then re-generate a more complete list including status information. 1353 * Devices which are already active have their details maintained, and are 1354 * not re-opened. 1355 */ 1356static void 1357spa_load_l2cache(spa_t *spa) 1358{ 1359 nvlist_t **l2cache; 1360 uint_t nl2cache; 1361 int i, j, oldnvdevs; 1362 uint64_t guid; 1363 vdev_t *vd, **oldvdevs, **newvdevs; 1364 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1365 1366 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1367 1368 if (sav->sav_config != NULL) { 1369 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1370 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1371 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1372 } else { 1373 nl2cache = 0; 1374 } 1375 1376 oldvdevs = sav->sav_vdevs; 1377 oldnvdevs = sav->sav_count; 1378 sav->sav_vdevs = NULL; 1379 sav->sav_count = 0; 1380 1381 /* 1382 * Process new nvlist of vdevs. 1383 */ 1384 for (i = 0; i < nl2cache; i++) { 1385 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1386 &guid) == 0); 1387 1388 newvdevs[i] = NULL; 1389 for (j = 0; j < oldnvdevs; j++) { 1390 vd = oldvdevs[j]; 1391 if (vd != NULL && guid == vd->vdev_guid) { 1392 /* 1393 * Retain previous vdev for add/remove ops. 1394 */ 1395 newvdevs[i] = vd; 1396 oldvdevs[j] = NULL; 1397 break; 1398 } 1399 } 1400 1401 if (newvdevs[i] == NULL) { 1402 /* 1403 * Create new vdev 1404 */ 1405 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1406 VDEV_ALLOC_L2CACHE) == 0); 1407 ASSERT(vd != NULL); 1408 newvdevs[i] = vd; 1409 1410 /* 1411 * Commit this vdev as an l2cache device, 1412 * even if it fails to open. 1413 */ 1414 spa_l2cache_add(vd); 1415 1416 vd->vdev_top = vd; 1417 vd->vdev_aux = sav; 1418 1419 spa_l2cache_activate(vd); 1420 1421 if (vdev_open(vd) != 0) 1422 continue; 1423 1424 (void) vdev_validate_aux(vd); 1425 1426 if (!vdev_is_dead(vd)) 1427 l2arc_add_vdev(spa, vd); 1428 } 1429 } 1430 1431 /* 1432 * Purge vdevs that were dropped 1433 */ 1434 for (i = 0; i < oldnvdevs; i++) { 1435 uint64_t pool; 1436 1437 vd = oldvdevs[i]; 1438 if (vd != NULL) { 1439 ASSERT(vd->vdev_isl2cache); 1440 1441 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1442 pool != 0ULL && l2arc_vdev_present(vd)) 1443 l2arc_remove_vdev(vd); 1444 vdev_clear_stats(vd); 1445 vdev_free(vd); 1446 } 1447 } 1448 1449 if (oldvdevs) 1450 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1451 1452 if (sav->sav_config == NULL) 1453 goto out; 1454 1455 sav->sav_vdevs = newvdevs; 1456 sav->sav_count = (int)nl2cache; 1457 1458 /* 1459 * Recompute the stashed list of l2cache devices, with status 1460 * information this time. 1461 */ 1462 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1463 DATA_TYPE_NVLIST_ARRAY) == 0); 1464 1465 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1466 for (i = 0; i < sav->sav_count; i++) 1467 l2cache[i] = vdev_config_generate(spa, 1468 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1469 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1470 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1471out: 1472 for (i = 0; i < sav->sav_count; i++) 1473 nvlist_free(l2cache[i]); 1474 if (sav->sav_count) 1475 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1476} 1477 1478static int 1479load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1480{ 1481 dmu_buf_t *db; 1482 char *packed = NULL; 1483 size_t nvsize = 0; 1484 int error; 1485 *value = NULL; 1486 1487 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 1488 nvsize = *(uint64_t *)db->db_data; 1489 dmu_buf_rele(db, FTAG); 1490 1491 packed = kmem_alloc(nvsize, KM_SLEEP); 1492 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1493 DMU_READ_PREFETCH); 1494 if (error == 0) 1495 error = nvlist_unpack(packed, nvsize, value, 0); 1496 kmem_free(packed, nvsize); 1497 1498 return (error); 1499} 1500 1501/* 1502 * Checks to see if the given vdev could not be opened, in which case we post a 1503 * sysevent to notify the autoreplace code that the device has been removed. 1504 */ 1505static void 1506spa_check_removed(vdev_t *vd) 1507{ 1508 for (int c = 0; c < vd->vdev_children; c++) 1509 spa_check_removed(vd->vdev_child[c]); 1510 1511 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) { 1512 zfs_post_autoreplace(vd->vdev_spa, vd); 1513 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1514 } 1515} 1516 1517/* 1518 * Validate the current config against the MOS config 1519 */ 1520static boolean_t 1521spa_config_valid(spa_t *spa, nvlist_t *config) 1522{ 1523 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1524 nvlist_t *nv; 1525 1526 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1527 1528 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1529 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1530 1531 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1532 1533 /* 1534 * If we're doing a normal import, then build up any additional 1535 * diagnostic information about missing devices in this config. 1536 * We'll pass this up to the user for further processing. 1537 */ 1538 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1539 nvlist_t **child, *nv; 1540 uint64_t idx = 0; 1541 1542 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1543 KM_SLEEP); 1544 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1545 1546 for (int c = 0; c < rvd->vdev_children; c++) { 1547 vdev_t *tvd = rvd->vdev_child[c]; 1548 vdev_t *mtvd = mrvd->vdev_child[c]; 1549 1550 if (tvd->vdev_ops == &vdev_missing_ops && 1551 mtvd->vdev_ops != &vdev_missing_ops && 1552 mtvd->vdev_islog) 1553 child[idx++] = vdev_config_generate(spa, mtvd, 1554 B_FALSE, 0); 1555 } 1556 1557 if (idx) { 1558 VERIFY(nvlist_add_nvlist_array(nv, 1559 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1560 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1561 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1562 1563 for (int i = 0; i < idx; i++) 1564 nvlist_free(child[i]); 1565 } 1566 nvlist_free(nv); 1567 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1568 } 1569 1570 /* 1571 * Compare the root vdev tree with the information we have 1572 * from the MOS config (mrvd). Check each top-level vdev 1573 * with the corresponding MOS config top-level (mtvd). 1574 */ 1575 for (int c = 0; c < rvd->vdev_children; c++) { 1576 vdev_t *tvd = rvd->vdev_child[c]; 1577 vdev_t *mtvd = mrvd->vdev_child[c]; 1578 1579 /* 1580 * Resolve any "missing" vdevs in the current configuration. 1581 * If we find that the MOS config has more accurate information 1582 * about the top-level vdev then use that vdev instead. 1583 */ 1584 if (tvd->vdev_ops == &vdev_missing_ops && 1585 mtvd->vdev_ops != &vdev_missing_ops) { 1586 1587 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1588 continue; 1589 1590 /* 1591 * Device specific actions. 1592 */ 1593 if (mtvd->vdev_islog) { 1594 spa_set_log_state(spa, SPA_LOG_CLEAR); 1595 } else { 1596 /* 1597 * XXX - once we have 'readonly' pool 1598 * support we should be able to handle 1599 * missing data devices by transitioning 1600 * the pool to readonly. 1601 */ 1602 continue; 1603 } 1604 1605 /* 1606 * Swap the missing vdev with the data we were 1607 * able to obtain from the MOS config. 1608 */ 1609 vdev_remove_child(rvd, tvd); 1610 vdev_remove_child(mrvd, mtvd); 1611 1612 vdev_add_child(rvd, mtvd); 1613 vdev_add_child(mrvd, tvd); 1614 1615 spa_config_exit(spa, SCL_ALL, FTAG); 1616 vdev_load(mtvd); 1617 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1618 1619 vdev_reopen(rvd); 1620 } else if (mtvd->vdev_islog) { 1621 /* 1622 * Load the slog device's state from the MOS config 1623 * since it's possible that the label does not 1624 * contain the most up-to-date information. 1625 */ 1626 vdev_load_log_state(tvd, mtvd); 1627 vdev_reopen(tvd); 1628 } 1629 } 1630 vdev_free(mrvd); 1631 spa_config_exit(spa, SCL_ALL, FTAG); 1632 1633 /* 1634 * Ensure we were able to validate the config. 1635 */ 1636 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1637} 1638 1639/* 1640 * Check for missing log devices 1641 */ 1642static int 1643spa_check_logs(spa_t *spa) 1644{ 1645 switch (spa->spa_log_state) { 1646 case SPA_LOG_MISSING: 1647 /* need to recheck in case slog has been restored */ 1648 case SPA_LOG_UNKNOWN: 1649 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL, 1650 DS_FIND_CHILDREN)) { 1651 spa_set_log_state(spa, SPA_LOG_MISSING); 1652 return (1); 1653 } 1654 break; 1655 } 1656 return (0); 1657} 1658 1659static boolean_t 1660spa_passivate_log(spa_t *spa) 1661{ 1662 vdev_t *rvd = spa->spa_root_vdev; 1663 boolean_t slog_found = B_FALSE; 1664 1665 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1666 1667 if (!spa_has_slogs(spa)) 1668 return (B_FALSE); 1669 1670 for (int c = 0; c < rvd->vdev_children; c++) { 1671 vdev_t *tvd = rvd->vdev_child[c]; 1672 metaslab_group_t *mg = tvd->vdev_mg; 1673 1674 if (tvd->vdev_islog) { 1675 metaslab_group_passivate(mg); 1676 slog_found = B_TRUE; 1677 } 1678 } 1679 1680 return (slog_found); 1681} 1682 1683static void 1684spa_activate_log(spa_t *spa) 1685{ 1686 vdev_t *rvd = spa->spa_root_vdev; 1687 1688 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1689 1690 for (int c = 0; c < rvd->vdev_children; c++) { 1691 vdev_t *tvd = rvd->vdev_child[c]; 1692 metaslab_group_t *mg = tvd->vdev_mg; 1693 1694 if (tvd->vdev_islog) 1695 metaslab_group_activate(mg); 1696 } 1697} 1698 1699int 1700spa_offline_log(spa_t *spa) 1701{ 1702 int error = 0; 1703 1704 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1705 NULL, DS_FIND_CHILDREN)) == 0) { 1706 1707 /* 1708 * We successfully offlined the log device, sync out the 1709 * current txg so that the "stubby" block can be removed 1710 * by zil_sync(). 1711 */ 1712 txg_wait_synced(spa->spa_dsl_pool, 0); 1713 } 1714 return (error); 1715} 1716 1717static void 1718spa_aux_check_removed(spa_aux_vdev_t *sav) 1719{ 1720 int i; 1721 1722 for (i = 0; i < sav->sav_count; i++) 1723 spa_check_removed(sav->sav_vdevs[i]); 1724} 1725 1726void 1727spa_claim_notify(zio_t *zio) 1728{ 1729 spa_t *spa = zio->io_spa; 1730 1731 if (zio->io_error) 1732 return; 1733 1734 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1735 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1736 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1737 mutex_exit(&spa->spa_props_lock); 1738} 1739 1740typedef struct spa_load_error { 1741 uint64_t sle_meta_count; 1742 uint64_t sle_data_count; 1743} spa_load_error_t; 1744 1745static void 1746spa_load_verify_done(zio_t *zio) 1747{ 1748 blkptr_t *bp = zio->io_bp; 1749 spa_load_error_t *sle = zio->io_private; 1750 dmu_object_type_t type = BP_GET_TYPE(bp); 1751 int error = zio->io_error; 1752 1753 if (error) { 1754 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1755 type != DMU_OT_INTENT_LOG) 1756 atomic_add_64(&sle->sle_meta_count, 1); 1757 else 1758 atomic_add_64(&sle->sle_data_count, 1); 1759 } 1760 zio_data_buf_free(zio->io_data, zio->io_size); 1761} 1762 1763/*ARGSUSED*/ 1764static int 1765spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1766 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1767{ 1768 if (bp != NULL) { 1769 zio_t *rio = arg; 1770 size_t size = BP_GET_PSIZE(bp); 1771 void *data = zio_data_buf_alloc(size); 1772 1773 zio_nowait(zio_read(rio, spa, bp, data, size, 1774 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1775 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1776 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1777 } 1778 return (0); 1779} 1780 1781static int 1782spa_load_verify(spa_t *spa) 1783{ 1784 zio_t *rio; 1785 spa_load_error_t sle = { 0 }; 1786 zpool_rewind_policy_t policy; 1787 boolean_t verify_ok = B_FALSE; 1788 int error; 1789 1790 zpool_get_rewind_policy(spa->spa_config, &policy); 1791 1792 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1793 return (0); 1794 1795 rio = zio_root(spa, NULL, &sle, 1796 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1797 1798 error = traverse_pool(spa, spa->spa_verify_min_txg, 1799 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1800 1801 (void) zio_wait(rio); 1802 1803 spa->spa_load_meta_errors = sle.sle_meta_count; 1804 spa->spa_load_data_errors = sle.sle_data_count; 1805 1806 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1807 sle.sle_data_count <= policy.zrp_maxdata) { 1808 int64_t loss = 0; 1809 1810 verify_ok = B_TRUE; 1811 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1812 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1813 1814 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1815 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1816 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1817 VERIFY(nvlist_add_int64(spa->spa_load_info, 1818 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1819 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1820 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1821 } else { 1822 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1823 } 1824 1825 if (error) { 1826 if (error != ENXIO && error != EIO) 1827 error = EIO; 1828 return (error); 1829 } 1830 1831 return (verify_ok ? 0 : EIO); 1832} 1833 1834/* 1835 * Find a value in the pool props object. 1836 */ 1837static void 1838spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1839{ 1840 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1841 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1842} 1843 1844/* 1845 * Find a value in the pool directory object. 1846 */ 1847static int 1848spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1849{ 1850 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1851 name, sizeof (uint64_t), 1, val)); 1852} 1853 1854static int 1855spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1856{ 1857 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1858 return (err); 1859} 1860 1861/* 1862 * Fix up config after a partly-completed split. This is done with the 1863 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1864 * pool have that entry in their config, but only the splitting one contains 1865 * a list of all the guids of the vdevs that are being split off. 1866 * 1867 * This function determines what to do with that list: either rejoin 1868 * all the disks to the pool, or complete the splitting process. To attempt 1869 * the rejoin, each disk that is offlined is marked online again, and 1870 * we do a reopen() call. If the vdev label for every disk that was 1871 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 1872 * then we call vdev_split() on each disk, and complete the split. 1873 * 1874 * Otherwise we leave the config alone, with all the vdevs in place in 1875 * the original pool. 1876 */ 1877static void 1878spa_try_repair(spa_t *spa, nvlist_t *config) 1879{ 1880 uint_t extracted; 1881 uint64_t *glist; 1882 uint_t i, gcount; 1883 nvlist_t *nvl; 1884 vdev_t **vd; 1885 boolean_t attempt_reopen; 1886 1887 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 1888 return; 1889 1890 /* check that the config is complete */ 1891 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 1892 &glist, &gcount) != 0) 1893 return; 1894 1895 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 1896 1897 /* attempt to online all the vdevs & validate */ 1898 attempt_reopen = B_TRUE; 1899 for (i = 0; i < gcount; i++) { 1900 if (glist[i] == 0) /* vdev is hole */ 1901 continue; 1902 1903 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 1904 if (vd[i] == NULL) { 1905 /* 1906 * Don't bother attempting to reopen the disks; 1907 * just do the split. 1908 */ 1909 attempt_reopen = B_FALSE; 1910 } else { 1911 /* attempt to re-online it */ 1912 vd[i]->vdev_offline = B_FALSE; 1913 } 1914 } 1915 1916 if (attempt_reopen) { 1917 vdev_reopen(spa->spa_root_vdev); 1918 1919 /* check each device to see what state it's in */ 1920 for (extracted = 0, i = 0; i < gcount; i++) { 1921 if (vd[i] != NULL && 1922 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 1923 break; 1924 ++extracted; 1925 } 1926 } 1927 1928 /* 1929 * If every disk has been moved to the new pool, or if we never 1930 * even attempted to look at them, then we split them off for 1931 * good. 1932 */ 1933 if (!attempt_reopen || gcount == extracted) { 1934 for (i = 0; i < gcount; i++) 1935 if (vd[i] != NULL) 1936 vdev_split(vd[i]); 1937 vdev_reopen(spa->spa_root_vdev); 1938 } 1939 1940 kmem_free(vd, gcount * sizeof (vdev_t *)); 1941} 1942 1943static int 1944spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 1945 boolean_t mosconfig) 1946{ 1947 nvlist_t *config = spa->spa_config; 1948 char *ereport = FM_EREPORT_ZFS_POOL; 1949 char *comment; 1950 int error; 1951 uint64_t pool_guid; 1952 nvlist_t *nvl; 1953 1954 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 1955 return (EINVAL); 1956 1957 ASSERT(spa->spa_comment == NULL); 1958 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 1959 spa->spa_comment = spa_strdup(comment); 1960 1961 /* 1962 * Versioning wasn't explicitly added to the label until later, so if 1963 * it's not present treat it as the initial version. 1964 */ 1965 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 1966 &spa->spa_ubsync.ub_version) != 0) 1967 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 1968 1969 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 1970 &spa->spa_config_txg); 1971 1972 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 1973 spa_guid_exists(pool_guid, 0)) { 1974 error = EEXIST; 1975 } else { 1976 spa->spa_config_guid = pool_guid; 1977 1978 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 1979 &nvl) == 0) { 1980 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 1981 KM_SLEEP) == 0); 1982 } 1983 1984 nvlist_free(spa->spa_load_info); 1985 spa->spa_load_info = fnvlist_alloc(); 1986 1987 gethrestime(&spa->spa_loaded_ts); 1988 error = spa_load_impl(spa, pool_guid, config, state, type, 1989 mosconfig, &ereport); 1990 } 1991 1992 spa->spa_minref = refcount_count(&spa->spa_refcount); 1993 if (error) { 1994 if (error != EEXIST) { 1995 spa->spa_loaded_ts.tv_sec = 0; 1996 spa->spa_loaded_ts.tv_nsec = 0; 1997 } 1998 if (error != EBADF) { 1999 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2000 } 2001 } 2002 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2003 spa->spa_ena = 0; 2004 2005 return (error); 2006} 2007 2008/* 2009 * Load an existing storage pool, using the pool's builtin spa_config as a 2010 * source of configuration information. 2011 */ 2012static int 2013spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2014 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2015 char **ereport) 2016{ 2017 int error = 0; 2018 nvlist_t *nvroot = NULL; 2019 nvlist_t *label; 2020 vdev_t *rvd; 2021 uberblock_t *ub = &spa->spa_uberblock; 2022 uint64_t children, config_cache_txg = spa->spa_config_txg; 2023 int orig_mode = spa->spa_mode; 2024 int parse; 2025 uint64_t obj; 2026 boolean_t missing_feat_write = B_FALSE; 2027 2028 /* 2029 * If this is an untrusted config, access the pool in read-only mode. 2030 * This prevents things like resilvering recently removed devices. 2031 */ 2032 if (!mosconfig) 2033 spa->spa_mode = FREAD; 2034 2035 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2036 2037 spa->spa_load_state = state; 2038 2039 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2040 return (EINVAL); 2041 2042 parse = (type == SPA_IMPORT_EXISTING ? 2043 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2044 2045 /* 2046 * Create "The Godfather" zio to hold all async IOs 2047 */ 2048 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 2049 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 2050 2051 /* 2052 * Parse the configuration into a vdev tree. We explicitly set the 2053 * value that will be returned by spa_version() since parsing the 2054 * configuration requires knowing the version number. 2055 */ 2056 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2057 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2058 spa_config_exit(spa, SCL_ALL, FTAG); 2059 2060 if (error != 0) 2061 return (error); 2062 2063 ASSERT(spa->spa_root_vdev == rvd); 2064 2065 if (type != SPA_IMPORT_ASSEMBLE) { 2066 ASSERT(spa_guid(spa) == pool_guid); 2067 } 2068 2069 /* 2070 * Try to open all vdevs, loading each label in the process. 2071 */ 2072 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2073 error = vdev_open(rvd); 2074 spa_config_exit(spa, SCL_ALL, FTAG); 2075 if (error != 0) 2076 return (error); 2077 2078 /* 2079 * We need to validate the vdev labels against the configuration that 2080 * we have in hand, which is dependent on the setting of mosconfig. If 2081 * mosconfig is true then we're validating the vdev labels based on 2082 * that config. Otherwise, we're validating against the cached config 2083 * (zpool.cache) that was read when we loaded the zfs module, and then 2084 * later we will recursively call spa_load() and validate against 2085 * the vdev config. 2086 * 2087 * If we're assembling a new pool that's been split off from an 2088 * existing pool, the labels haven't yet been updated so we skip 2089 * validation for now. 2090 */ 2091 if (type != SPA_IMPORT_ASSEMBLE) { 2092 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2093 error = vdev_validate(rvd, mosconfig); 2094 spa_config_exit(spa, SCL_ALL, FTAG); 2095 2096 if (error != 0) 2097 return (error); 2098 2099 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2100 return (ENXIO); 2101 } 2102 2103 /* 2104 * Find the best uberblock. 2105 */ 2106 vdev_uberblock_load(rvd, ub, &label); 2107 2108 /* 2109 * If we weren't able to find a single valid uberblock, return failure. 2110 */ 2111 if (ub->ub_txg == 0) { 2112 nvlist_free(label); 2113 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2114 } 2115 2116 /* 2117 * If the pool has an unsupported version we can't open it. 2118 */ 2119 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2120 nvlist_free(label); 2121 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2122 } 2123 2124 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2125 nvlist_t *features; 2126 2127 /* 2128 * If we weren't able to find what's necessary for reading the 2129 * MOS in the label, return failure. 2130 */ 2131 if (label == NULL || nvlist_lookup_nvlist(label, 2132 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2133 nvlist_free(label); 2134 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2135 ENXIO)); 2136 } 2137 2138 /* 2139 * Update our in-core representation with the definitive values 2140 * from the label. 2141 */ 2142 nvlist_free(spa->spa_label_features); 2143 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2144 } 2145 2146 nvlist_free(label); 2147 2148 /* 2149 * Look through entries in the label nvlist's features_for_read. If 2150 * there is a feature listed there which we don't understand then we 2151 * cannot open a pool. 2152 */ 2153 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2154 nvlist_t *unsup_feat; 2155 2156 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2157 0); 2158 2159 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2160 NULL); nvp != NULL; 2161 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2162 if (!zfeature_is_supported(nvpair_name(nvp))) { 2163 VERIFY(nvlist_add_string(unsup_feat, 2164 nvpair_name(nvp), "") == 0); 2165 } 2166 } 2167 2168 if (!nvlist_empty(unsup_feat)) { 2169 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2170 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2171 nvlist_free(unsup_feat); 2172 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2173 ENOTSUP)); 2174 } 2175 2176 nvlist_free(unsup_feat); 2177 } 2178 2179 /* 2180 * If the vdev guid sum doesn't match the uberblock, we have an 2181 * incomplete configuration. We first check to see if the pool 2182 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2183 * If it is, defer the vdev_guid_sum check till later so we 2184 * can handle missing vdevs. 2185 */ 2186 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2187 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2188 rvd->vdev_guid_sum != ub->ub_guid_sum) 2189 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2190 2191 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2192 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2193 spa_try_repair(spa, config); 2194 spa_config_exit(spa, SCL_ALL, FTAG); 2195 nvlist_free(spa->spa_config_splitting); 2196 spa->spa_config_splitting = NULL; 2197 } 2198 2199 /* 2200 * Initialize internal SPA structures. 2201 */ 2202 spa->spa_state = POOL_STATE_ACTIVE; 2203 spa->spa_ubsync = spa->spa_uberblock; 2204 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2205 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2206 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2207 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2208 spa->spa_claim_max_txg = spa->spa_first_txg; 2209 spa->spa_prev_software_version = ub->ub_software_version; 2210 2211 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2212 if (error) 2213 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2214 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2215 2216 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2217 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2218 2219 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2220 boolean_t missing_feat_read = B_FALSE; 2221 nvlist_t *unsup_feat, *enabled_feat; 2222 2223 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2224 &spa->spa_feat_for_read_obj) != 0) { 2225 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2226 } 2227 2228 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2229 &spa->spa_feat_for_write_obj) != 0) { 2230 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2231 } 2232 2233 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2234 &spa->spa_feat_desc_obj) != 0) { 2235 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2236 } 2237 2238 enabled_feat = fnvlist_alloc(); 2239 unsup_feat = fnvlist_alloc(); 2240 2241 if (!feature_is_supported(spa->spa_meta_objset, 2242 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, 2243 unsup_feat, enabled_feat)) 2244 missing_feat_read = B_TRUE; 2245 2246 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2247 if (!feature_is_supported(spa->spa_meta_objset, 2248 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, 2249 unsup_feat, enabled_feat)) { 2250 missing_feat_write = B_TRUE; 2251 } 2252 } 2253 2254 fnvlist_add_nvlist(spa->spa_load_info, 2255 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2256 2257 if (!nvlist_empty(unsup_feat)) { 2258 fnvlist_add_nvlist(spa->spa_load_info, 2259 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2260 } 2261 2262 fnvlist_free(enabled_feat); 2263 fnvlist_free(unsup_feat); 2264 2265 if (!missing_feat_read) { 2266 fnvlist_add_boolean(spa->spa_load_info, 2267 ZPOOL_CONFIG_CAN_RDONLY); 2268 } 2269 2270 /* 2271 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2272 * twofold: to determine whether the pool is available for 2273 * import in read-write mode and (if it is not) whether the 2274 * pool is available for import in read-only mode. If the pool 2275 * is available for import in read-write mode, it is displayed 2276 * as available in userland; if it is not available for import 2277 * in read-only mode, it is displayed as unavailable in 2278 * userland. If the pool is available for import in read-only 2279 * mode but not read-write mode, it is displayed as unavailable 2280 * in userland with a special note that the pool is actually 2281 * available for open in read-only mode. 2282 * 2283 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2284 * missing a feature for write, we must first determine whether 2285 * the pool can be opened read-only before returning to 2286 * userland in order to know whether to display the 2287 * abovementioned note. 2288 */ 2289 if (missing_feat_read || (missing_feat_write && 2290 spa_writeable(spa))) { 2291 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2292 ENOTSUP)); 2293 } 2294 } 2295 2296 spa->spa_is_initializing = B_TRUE; 2297 error = dsl_pool_open(spa->spa_dsl_pool); 2298 spa->spa_is_initializing = B_FALSE; 2299 if (error != 0) 2300 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2301 2302 if (!mosconfig) { 2303 uint64_t hostid; 2304 nvlist_t *policy = NULL, *nvconfig; 2305 2306 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2307 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2308 2309 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2310 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2311 char *hostname; 2312 unsigned long myhostid = 0; 2313 2314 VERIFY(nvlist_lookup_string(nvconfig, 2315 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2316 2317#ifdef _KERNEL 2318 myhostid = zone_get_hostid(NULL); 2319#else /* _KERNEL */ 2320 /* 2321 * We're emulating the system's hostid in userland, so 2322 * we can't use zone_get_hostid(). 2323 */ 2324 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2325#endif /* _KERNEL */ 2326 if (check_hostid && hostid != 0 && myhostid != 0 && 2327 hostid != myhostid) { 2328 nvlist_free(nvconfig); 2329 cmn_err(CE_WARN, "pool '%s' could not be " 2330 "loaded as it was last accessed by " 2331 "another system (host: %s hostid: 0x%lx). " 2332 "See: http://illumos.org/msg/ZFS-8000-EY", 2333 spa_name(spa), hostname, 2334 (unsigned long)hostid); 2335 return (EBADF); 2336 } 2337 } 2338 if (nvlist_lookup_nvlist(spa->spa_config, 2339 ZPOOL_REWIND_POLICY, &policy) == 0) 2340 VERIFY(nvlist_add_nvlist(nvconfig, 2341 ZPOOL_REWIND_POLICY, policy) == 0); 2342 2343 spa_config_set(spa, nvconfig); 2344 spa_unload(spa); 2345 spa_deactivate(spa); 2346 spa_activate(spa, orig_mode); 2347 2348 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2349 } 2350 2351 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2352 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2353 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2354 if (error != 0) 2355 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2356 2357 /* 2358 * Load the bit that tells us to use the new accounting function 2359 * (raid-z deflation). If we have an older pool, this will not 2360 * be present. 2361 */ 2362 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2363 if (error != 0 && error != ENOENT) 2364 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2365 2366 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2367 &spa->spa_creation_version); 2368 if (error != 0 && error != ENOENT) 2369 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2370 2371 /* 2372 * Load the persistent error log. If we have an older pool, this will 2373 * not be present. 2374 */ 2375 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2376 if (error != 0 && error != ENOENT) 2377 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2378 2379 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2380 &spa->spa_errlog_scrub); 2381 if (error != 0 && error != ENOENT) 2382 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2383 2384 /* 2385 * Load the history object. If we have an older pool, this 2386 * will not be present. 2387 */ 2388 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2389 if (error != 0 && error != ENOENT) 2390 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2391 2392 /* 2393 * If we're assembling the pool from the split-off vdevs of 2394 * an existing pool, we don't want to attach the spares & cache 2395 * devices. 2396 */ 2397 2398 /* 2399 * Load any hot spares for this pool. 2400 */ 2401 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2402 if (error != 0 && error != ENOENT) 2403 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2404 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2405 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2406 if (load_nvlist(spa, spa->spa_spares.sav_object, 2407 &spa->spa_spares.sav_config) != 0) 2408 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2409 2410 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2411 spa_load_spares(spa); 2412 spa_config_exit(spa, SCL_ALL, FTAG); 2413 } else if (error == 0) { 2414 spa->spa_spares.sav_sync = B_TRUE; 2415 } 2416 2417 /* 2418 * Load any level 2 ARC devices for this pool. 2419 */ 2420 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2421 &spa->spa_l2cache.sav_object); 2422 if (error != 0 && error != ENOENT) 2423 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2424 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2425 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2426 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2427 &spa->spa_l2cache.sav_config) != 0) 2428 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2429 2430 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2431 spa_load_l2cache(spa); 2432 spa_config_exit(spa, SCL_ALL, FTAG); 2433 } else if (error == 0) { 2434 spa->spa_l2cache.sav_sync = B_TRUE; 2435 } 2436 2437 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2438 2439 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2440 if (error && error != ENOENT) 2441 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2442 2443 if (error == 0) { 2444 uint64_t autoreplace; 2445 2446 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2447 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2448 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2449 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2450 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2451 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2452 &spa->spa_dedup_ditto); 2453 2454 spa->spa_autoreplace = (autoreplace != 0); 2455 } 2456 2457 /* 2458 * If the 'autoreplace' property is set, then post a resource notifying 2459 * the ZFS DE that it should not issue any faults for unopenable 2460 * devices. We also iterate over the vdevs, and post a sysevent for any 2461 * unopenable vdevs so that the normal autoreplace handler can take 2462 * over. 2463 */ 2464 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2465 spa_check_removed(spa->spa_root_vdev); 2466 /* 2467 * For the import case, this is done in spa_import(), because 2468 * at this point we're using the spare definitions from 2469 * the MOS config, not necessarily from the userland config. 2470 */ 2471 if (state != SPA_LOAD_IMPORT) { 2472 spa_aux_check_removed(&spa->spa_spares); 2473 spa_aux_check_removed(&spa->spa_l2cache); 2474 } 2475 } 2476 2477 /* 2478 * Load the vdev state for all toplevel vdevs. 2479 */ 2480 vdev_load(rvd); 2481 2482 /* 2483 * Propagate the leaf DTLs we just loaded all the way up the tree. 2484 */ 2485 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2486 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2487 spa_config_exit(spa, SCL_ALL, FTAG); 2488 2489 /* 2490 * Load the DDTs (dedup tables). 2491 */ 2492 error = ddt_load(spa); 2493 if (error != 0) 2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2495 2496 spa_update_dspace(spa); 2497 2498 /* 2499 * Validate the config, using the MOS config to fill in any 2500 * information which might be missing. If we fail to validate 2501 * the config then declare the pool unfit for use. If we're 2502 * assembling a pool from a split, the log is not transferred 2503 * over. 2504 */ 2505 if (type != SPA_IMPORT_ASSEMBLE) { 2506 nvlist_t *nvconfig; 2507 2508 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2509 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2510 2511 if (!spa_config_valid(spa, nvconfig)) { 2512 nvlist_free(nvconfig); 2513 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2514 ENXIO)); 2515 } 2516 nvlist_free(nvconfig); 2517 2518 /* 2519 * Now that we've validated the config, check the state of the 2520 * root vdev. If it can't be opened, it indicates one or 2521 * more toplevel vdevs are faulted. 2522 */ 2523 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2524 return (ENXIO); 2525 2526 if (spa_check_logs(spa)) { 2527 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2528 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2529 } 2530 } 2531 2532 if (missing_feat_write) { 2533 ASSERT(state == SPA_LOAD_TRYIMPORT); 2534 2535 /* 2536 * At this point, we know that we can open the pool in 2537 * read-only mode but not read-write mode. We now have enough 2538 * information and can return to userland. 2539 */ 2540 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2541 } 2542 2543 /* 2544 * We've successfully opened the pool, verify that we're ready 2545 * to start pushing transactions. 2546 */ 2547 if (state != SPA_LOAD_TRYIMPORT) { 2548 if (error = spa_load_verify(spa)) 2549 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2550 error)); 2551 } 2552 2553 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2554 spa->spa_load_max_txg == UINT64_MAX)) { 2555 dmu_tx_t *tx; 2556 int need_update = B_FALSE; 2557 2558 ASSERT(state != SPA_LOAD_TRYIMPORT); 2559 2560 /* 2561 * Claim log blocks that haven't been committed yet. 2562 * This must all happen in a single txg. 2563 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2564 * invoked from zil_claim_log_block()'s i/o done callback. 2565 * Price of rollback is that we abandon the log. 2566 */ 2567 spa->spa_claiming = B_TRUE; 2568 2569 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2570 spa_first_txg(spa)); 2571 (void) dmu_objset_find(spa_name(spa), 2572 zil_claim, tx, DS_FIND_CHILDREN); 2573 dmu_tx_commit(tx); 2574 2575 spa->spa_claiming = B_FALSE; 2576 2577 spa_set_log_state(spa, SPA_LOG_GOOD); 2578 spa->spa_sync_on = B_TRUE; 2579 txg_sync_start(spa->spa_dsl_pool); 2580 2581 /* 2582 * Wait for all claims to sync. We sync up to the highest 2583 * claimed log block birth time so that claimed log blocks 2584 * don't appear to be from the future. spa_claim_max_txg 2585 * will have been set for us by either zil_check_log_chain() 2586 * (invoked from spa_check_logs()) or zil_claim() above. 2587 */ 2588 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2589 2590 /* 2591 * If the config cache is stale, or we have uninitialized 2592 * metaslabs (see spa_vdev_add()), then update the config. 2593 * 2594 * If this is a verbatim import, trust the current 2595 * in-core spa_config and update the disk labels. 2596 */ 2597 if (config_cache_txg != spa->spa_config_txg || 2598 state == SPA_LOAD_IMPORT || 2599 state == SPA_LOAD_RECOVER || 2600 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2601 need_update = B_TRUE; 2602 2603 for (int c = 0; c < rvd->vdev_children; c++) 2604 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2605 need_update = B_TRUE; 2606 2607 /* 2608 * Update the config cache asychronously in case we're the 2609 * root pool, in which case the config cache isn't writable yet. 2610 */ 2611 if (need_update) 2612 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2613 2614 /* 2615 * Check all DTLs to see if anything needs resilvering. 2616 */ 2617 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2618 vdev_resilver_needed(rvd, NULL, NULL)) 2619 spa_async_request(spa, SPA_ASYNC_RESILVER); 2620 2621 /* 2622 * Delete any inconsistent datasets. 2623 */ 2624 (void) dmu_objset_find(spa_name(spa), 2625 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2626 2627 /* 2628 * Clean up any stale temporary dataset userrefs. 2629 */ 2630 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2631 } 2632 2633 return (0); 2634} 2635 2636static int 2637spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2638{ 2639 int mode = spa->spa_mode; 2640 2641 spa_unload(spa); 2642 spa_deactivate(spa); 2643 2644 spa->spa_load_max_txg--; 2645 2646 spa_activate(spa, mode); 2647 spa_async_suspend(spa); 2648 2649 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2650} 2651 2652/* 2653 * If spa_load() fails this function will try loading prior txg's. If 2654 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2655 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2656 * function will not rewind the pool and will return the same error as 2657 * spa_load(). 2658 */ 2659static int 2660spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2661 uint64_t max_request, int rewind_flags) 2662{ 2663 nvlist_t *loadinfo = NULL; 2664 nvlist_t *config = NULL; 2665 int load_error, rewind_error; 2666 uint64_t safe_rewind_txg; 2667 uint64_t min_txg; 2668 2669 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2670 spa->spa_load_max_txg = spa->spa_load_txg; 2671 spa_set_log_state(spa, SPA_LOG_CLEAR); 2672 } else { 2673 spa->spa_load_max_txg = max_request; 2674 } 2675 2676 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2677 mosconfig); 2678 if (load_error == 0) 2679 return (0); 2680 2681 if (spa->spa_root_vdev != NULL) 2682 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2683 2684 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2685 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2686 2687 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2688 nvlist_free(config); 2689 return (load_error); 2690 } 2691 2692 if (state == SPA_LOAD_RECOVER) { 2693 /* Price of rolling back is discarding txgs, including log */ 2694 spa_set_log_state(spa, SPA_LOG_CLEAR); 2695 } else { 2696 /* 2697 * If we aren't rolling back save the load info from our first 2698 * import attempt so that we can restore it after attempting 2699 * to rewind. 2700 */ 2701 loadinfo = spa->spa_load_info; 2702 spa->spa_load_info = fnvlist_alloc(); 2703 } 2704 2705 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2706 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2707 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2708 TXG_INITIAL : safe_rewind_txg; 2709 2710 /* 2711 * Continue as long as we're finding errors, we're still within 2712 * the acceptable rewind range, and we're still finding uberblocks 2713 */ 2714 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2715 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2716 if (spa->spa_load_max_txg < safe_rewind_txg) 2717 spa->spa_extreme_rewind = B_TRUE; 2718 rewind_error = spa_load_retry(spa, state, mosconfig); 2719 } 2720 2721 spa->spa_extreme_rewind = B_FALSE; 2722 spa->spa_load_max_txg = UINT64_MAX; 2723 2724 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2725 spa_config_set(spa, config); 2726 2727 if (state == SPA_LOAD_RECOVER) { 2728 ASSERT3P(loadinfo, ==, NULL); 2729 return (rewind_error); 2730 } else { 2731 /* Store the rewind info as part of the initial load info */ 2732 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2733 spa->spa_load_info); 2734 2735 /* Restore the initial load info */ 2736 fnvlist_free(spa->spa_load_info); 2737 spa->spa_load_info = loadinfo; 2738 2739 return (load_error); 2740 } 2741} 2742 2743/* 2744 * Pool Open/Import 2745 * 2746 * The import case is identical to an open except that the configuration is sent 2747 * down from userland, instead of grabbed from the configuration cache. For the 2748 * case of an open, the pool configuration will exist in the 2749 * POOL_STATE_UNINITIALIZED state. 2750 * 2751 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2752 * the same time open the pool, without having to keep around the spa_t in some 2753 * ambiguous state. 2754 */ 2755static int 2756spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2757 nvlist_t **config) 2758{ 2759 spa_t *spa; 2760 spa_load_state_t state = SPA_LOAD_OPEN; 2761 int error; 2762 int locked = B_FALSE; 2763 int firstopen = B_FALSE; 2764 2765 *spapp = NULL; 2766 2767 /* 2768 * As disgusting as this is, we need to support recursive calls to this 2769 * function because dsl_dir_open() is called during spa_load(), and ends 2770 * up calling spa_open() again. The real fix is to figure out how to 2771 * avoid dsl_dir_open() calling this in the first place. 2772 */ 2773 if (mutex_owner(&spa_namespace_lock) != curthread) { 2774 mutex_enter(&spa_namespace_lock); 2775 locked = B_TRUE; 2776 } 2777 2778 if ((spa = spa_lookup(pool)) == NULL) { 2779 if (locked) 2780 mutex_exit(&spa_namespace_lock); 2781 return (ENOENT); 2782 } 2783 2784 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2785 zpool_rewind_policy_t policy; 2786 2787 firstopen = B_TRUE; 2788 2789 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2790 &policy); 2791 if (policy.zrp_request & ZPOOL_DO_REWIND) 2792 state = SPA_LOAD_RECOVER; 2793 2794 spa_activate(spa, spa_mode_global); 2795 2796 if (state != SPA_LOAD_RECOVER) 2797 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2798 2799 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2800 policy.zrp_request); 2801 2802 if (error == EBADF) { 2803 /* 2804 * If vdev_validate() returns failure (indicated by 2805 * EBADF), it indicates that one of the vdevs indicates 2806 * that the pool has been exported or destroyed. If 2807 * this is the case, the config cache is out of sync and 2808 * we should remove the pool from the namespace. 2809 */ 2810 spa_unload(spa); 2811 spa_deactivate(spa); 2812 spa_config_sync(spa, B_TRUE, B_TRUE); 2813 spa_remove(spa); 2814 if (locked) 2815 mutex_exit(&spa_namespace_lock); 2816 return (ENOENT); 2817 } 2818 2819 if (error) { 2820 /* 2821 * We can't open the pool, but we still have useful 2822 * information: the state of each vdev after the 2823 * attempted vdev_open(). Return this to the user. 2824 */ 2825 if (config != NULL && spa->spa_config) { 2826 VERIFY(nvlist_dup(spa->spa_config, config, 2827 KM_SLEEP) == 0); 2828 VERIFY(nvlist_add_nvlist(*config, 2829 ZPOOL_CONFIG_LOAD_INFO, 2830 spa->spa_load_info) == 0); 2831 } 2832 spa_unload(spa); 2833 spa_deactivate(spa); 2834 spa->spa_last_open_failed = error; 2835 if (locked) 2836 mutex_exit(&spa_namespace_lock); 2837 *spapp = NULL; 2838 return (error); 2839 } 2840 } 2841 2842 spa_open_ref(spa, tag); 2843 2844 if (config != NULL) 2845 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2846 2847 /* 2848 * If we've recovered the pool, pass back any information we 2849 * gathered while doing the load. 2850 */ 2851 if (state == SPA_LOAD_RECOVER) { 2852 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 2853 spa->spa_load_info) == 0); 2854 } 2855 2856 if (locked) { 2857 spa->spa_last_open_failed = 0; 2858 spa->spa_last_ubsync_txg = 0; 2859 spa->spa_load_txg = 0; 2860 mutex_exit(&spa_namespace_lock); 2861#ifdef __FreeBSD__ 2862#ifdef _KERNEL 2863 if (firstopen) 2864 zvol_create_minors(pool); 2865#endif 2866#endif 2867 } 2868 2869 *spapp = spa; 2870 2871 return (0); 2872} 2873 2874int 2875spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2876 nvlist_t **config) 2877{ 2878 return (spa_open_common(name, spapp, tag, policy, config)); 2879} 2880 2881int 2882spa_open(const char *name, spa_t **spapp, void *tag) 2883{ 2884 return (spa_open_common(name, spapp, tag, NULL, NULL)); 2885} 2886 2887/* 2888 * Lookup the given spa_t, incrementing the inject count in the process, 2889 * preventing it from being exported or destroyed. 2890 */ 2891spa_t * 2892spa_inject_addref(char *name) 2893{ 2894 spa_t *spa; 2895 2896 mutex_enter(&spa_namespace_lock); 2897 if ((spa = spa_lookup(name)) == NULL) { 2898 mutex_exit(&spa_namespace_lock); 2899 return (NULL); 2900 } 2901 spa->spa_inject_ref++; 2902 mutex_exit(&spa_namespace_lock); 2903 2904 return (spa); 2905} 2906 2907void 2908spa_inject_delref(spa_t *spa) 2909{ 2910 mutex_enter(&spa_namespace_lock); 2911 spa->spa_inject_ref--; 2912 mutex_exit(&spa_namespace_lock); 2913} 2914 2915/* 2916 * Add spares device information to the nvlist. 2917 */ 2918static void 2919spa_add_spares(spa_t *spa, nvlist_t *config) 2920{ 2921 nvlist_t **spares; 2922 uint_t i, nspares; 2923 nvlist_t *nvroot; 2924 uint64_t guid; 2925 vdev_stat_t *vs; 2926 uint_t vsc; 2927 uint64_t pool; 2928 2929 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 2930 2931 if (spa->spa_spares.sav_count == 0) 2932 return; 2933 2934 VERIFY(nvlist_lookup_nvlist(config, 2935 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2936 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 2937 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2938 if (nspares != 0) { 2939 VERIFY(nvlist_add_nvlist_array(nvroot, 2940 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 2941 VERIFY(nvlist_lookup_nvlist_array(nvroot, 2942 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2943 2944 /* 2945 * Go through and find any spares which have since been 2946 * repurposed as an active spare. If this is the case, update 2947 * their status appropriately. 2948 */ 2949 for (i = 0; i < nspares; i++) { 2950 VERIFY(nvlist_lookup_uint64(spares[i], 2951 ZPOOL_CONFIG_GUID, &guid) == 0); 2952 if (spa_spare_exists(guid, &pool, NULL) && 2953 pool != 0ULL) { 2954 VERIFY(nvlist_lookup_uint64_array( 2955 spares[i], ZPOOL_CONFIG_VDEV_STATS, 2956 (uint64_t **)&vs, &vsc) == 0); 2957 vs->vs_state = VDEV_STATE_CANT_OPEN; 2958 vs->vs_aux = VDEV_AUX_SPARED; 2959 } 2960 } 2961 } 2962} 2963 2964/* 2965 * Add l2cache device information to the nvlist, including vdev stats. 2966 */ 2967static void 2968spa_add_l2cache(spa_t *spa, nvlist_t *config) 2969{ 2970 nvlist_t **l2cache; 2971 uint_t i, j, nl2cache; 2972 nvlist_t *nvroot; 2973 uint64_t guid; 2974 vdev_t *vd; 2975 vdev_stat_t *vs; 2976 uint_t vsc; 2977 2978 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 2979 2980 if (spa->spa_l2cache.sav_count == 0) 2981 return; 2982 2983 VERIFY(nvlist_lookup_nvlist(config, 2984 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2985 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 2986 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 2987 if (nl2cache != 0) { 2988 VERIFY(nvlist_add_nvlist_array(nvroot, 2989 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 2990 VERIFY(nvlist_lookup_nvlist_array(nvroot, 2991 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 2992 2993 /* 2994 * Update level 2 cache device stats. 2995 */ 2996 2997 for (i = 0; i < nl2cache; i++) { 2998 VERIFY(nvlist_lookup_uint64(l2cache[i], 2999 ZPOOL_CONFIG_GUID, &guid) == 0); 3000 3001 vd = NULL; 3002 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3003 if (guid == 3004 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3005 vd = spa->spa_l2cache.sav_vdevs[j]; 3006 break; 3007 } 3008 } 3009 ASSERT(vd != NULL); 3010 3011 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3012 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3013 == 0); 3014 vdev_get_stats(vd, vs); 3015 } 3016 } 3017} 3018 3019static void 3020spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3021{ 3022 nvlist_t *features; 3023 zap_cursor_t zc; 3024 zap_attribute_t za; 3025 3026 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3027 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3028 3029 if (spa->spa_feat_for_read_obj != 0) { 3030 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3031 spa->spa_feat_for_read_obj); 3032 zap_cursor_retrieve(&zc, &za) == 0; 3033 zap_cursor_advance(&zc)) { 3034 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3035 za.za_num_integers == 1); 3036 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3037 za.za_first_integer)); 3038 } 3039 zap_cursor_fini(&zc); 3040 } 3041 3042 if (spa->spa_feat_for_write_obj != 0) { 3043 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3044 spa->spa_feat_for_write_obj); 3045 zap_cursor_retrieve(&zc, &za) == 0; 3046 zap_cursor_advance(&zc)) { 3047 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3048 za.za_num_integers == 1); 3049 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3050 za.za_first_integer)); 3051 } 3052 zap_cursor_fini(&zc); 3053 } 3054 3055 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3056 features) == 0); 3057 nvlist_free(features); 3058} 3059 3060int 3061spa_get_stats(const char *name, nvlist_t **config, 3062 char *altroot, size_t buflen) 3063{ 3064 int error; 3065 spa_t *spa; 3066 3067 *config = NULL; 3068 error = spa_open_common(name, &spa, FTAG, NULL, config); 3069 3070 if (spa != NULL) { 3071 /* 3072 * This still leaves a window of inconsistency where the spares 3073 * or l2cache devices could change and the config would be 3074 * self-inconsistent. 3075 */ 3076 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3077 3078 if (*config != NULL) { 3079 uint64_t loadtimes[2]; 3080 3081 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3082 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3083 VERIFY(nvlist_add_uint64_array(*config, 3084 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3085 3086 VERIFY(nvlist_add_uint64(*config, 3087 ZPOOL_CONFIG_ERRCOUNT, 3088 spa_get_errlog_size(spa)) == 0); 3089 3090 if (spa_suspended(spa)) 3091 VERIFY(nvlist_add_uint64(*config, 3092 ZPOOL_CONFIG_SUSPENDED, 3093 spa->spa_failmode) == 0); 3094 3095 spa_add_spares(spa, *config); 3096 spa_add_l2cache(spa, *config); 3097 spa_add_feature_stats(spa, *config); 3098 } 3099 } 3100 3101 /* 3102 * We want to get the alternate root even for faulted pools, so we cheat 3103 * and call spa_lookup() directly. 3104 */ 3105 if (altroot) { 3106 if (spa == NULL) { 3107 mutex_enter(&spa_namespace_lock); 3108 spa = spa_lookup(name); 3109 if (spa) 3110 spa_altroot(spa, altroot, buflen); 3111 else 3112 altroot[0] = '\0'; 3113 spa = NULL; 3114 mutex_exit(&spa_namespace_lock); 3115 } else { 3116 spa_altroot(spa, altroot, buflen); 3117 } 3118 } 3119 3120 if (spa != NULL) { 3121 spa_config_exit(spa, SCL_CONFIG, FTAG); 3122 spa_close(spa, FTAG); 3123 } 3124 3125 return (error); 3126} 3127 3128/* 3129 * Validate that the auxiliary device array is well formed. We must have an 3130 * array of nvlists, each which describes a valid leaf vdev. If this is an 3131 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3132 * specified, as long as they are well-formed. 3133 */ 3134static int 3135spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3136 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3137 vdev_labeltype_t label) 3138{ 3139 nvlist_t **dev; 3140 uint_t i, ndev; 3141 vdev_t *vd; 3142 int error; 3143 3144 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3145 3146 /* 3147 * It's acceptable to have no devs specified. 3148 */ 3149 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3150 return (0); 3151 3152 if (ndev == 0) 3153 return (EINVAL); 3154 3155 /* 3156 * Make sure the pool is formatted with a version that supports this 3157 * device type. 3158 */ 3159 if (spa_version(spa) < version) 3160 return (ENOTSUP); 3161 3162 /* 3163 * Set the pending device list so we correctly handle device in-use 3164 * checking. 3165 */ 3166 sav->sav_pending = dev; 3167 sav->sav_npending = ndev; 3168 3169 for (i = 0; i < ndev; i++) { 3170 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3171 mode)) != 0) 3172 goto out; 3173 3174 if (!vd->vdev_ops->vdev_op_leaf) { 3175 vdev_free(vd); 3176 error = EINVAL; 3177 goto out; 3178 } 3179 3180 /* 3181 * The L2ARC currently only supports disk devices in 3182 * kernel context. For user-level testing, we allow it. 3183 */ 3184#ifdef _KERNEL 3185 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3186 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3187 error = ENOTBLK; 3188 vdev_free(vd); 3189 goto out; 3190 } 3191#endif 3192 vd->vdev_top = vd; 3193 3194 if ((error = vdev_open(vd)) == 0 && 3195 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3196 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3197 vd->vdev_guid) == 0); 3198 } 3199 3200 vdev_free(vd); 3201 3202 if (error && 3203 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3204 goto out; 3205 else 3206 error = 0; 3207 } 3208 3209out: 3210 sav->sav_pending = NULL; 3211 sav->sav_npending = 0; 3212 return (error); 3213} 3214 3215static int 3216spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3217{ 3218 int error; 3219 3220 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3221 3222 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3223 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3224 VDEV_LABEL_SPARE)) != 0) { 3225 return (error); 3226 } 3227 3228 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3229 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3230 VDEV_LABEL_L2CACHE)); 3231} 3232 3233static void 3234spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3235 const char *config) 3236{ 3237 int i; 3238 3239 if (sav->sav_config != NULL) { 3240 nvlist_t **olddevs; 3241 uint_t oldndevs; 3242 nvlist_t **newdevs; 3243 3244 /* 3245 * Generate new dev list by concatentating with the 3246 * current dev list. 3247 */ 3248 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3249 &olddevs, &oldndevs) == 0); 3250 3251 newdevs = kmem_alloc(sizeof (void *) * 3252 (ndevs + oldndevs), KM_SLEEP); 3253 for (i = 0; i < oldndevs; i++) 3254 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3255 KM_SLEEP) == 0); 3256 for (i = 0; i < ndevs; i++) 3257 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3258 KM_SLEEP) == 0); 3259 3260 VERIFY(nvlist_remove(sav->sav_config, config, 3261 DATA_TYPE_NVLIST_ARRAY) == 0); 3262 3263 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3264 config, newdevs, ndevs + oldndevs) == 0); 3265 for (i = 0; i < oldndevs + ndevs; i++) 3266 nvlist_free(newdevs[i]); 3267 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3268 } else { 3269 /* 3270 * Generate a new dev list. 3271 */ 3272 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3273 KM_SLEEP) == 0); 3274 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3275 devs, ndevs) == 0); 3276 } 3277} 3278 3279/* 3280 * Stop and drop level 2 ARC devices 3281 */ 3282void 3283spa_l2cache_drop(spa_t *spa) 3284{ 3285 vdev_t *vd; 3286 int i; 3287 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3288 3289 for (i = 0; i < sav->sav_count; i++) { 3290 uint64_t pool; 3291 3292 vd = sav->sav_vdevs[i]; 3293 ASSERT(vd != NULL); 3294 3295 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3296 pool != 0ULL && l2arc_vdev_present(vd)) 3297 l2arc_remove_vdev(vd); 3298 } 3299} 3300 3301/* 3302 * Pool Creation 3303 */ 3304int 3305spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3306 const char *history_str, nvlist_t *zplprops) 3307{ 3308 spa_t *spa; 3309 char *altroot = NULL; 3310 vdev_t *rvd; 3311 dsl_pool_t *dp; 3312 dmu_tx_t *tx; 3313 int error = 0; 3314 uint64_t txg = TXG_INITIAL; 3315 nvlist_t **spares, **l2cache; 3316 uint_t nspares, nl2cache; 3317 uint64_t version, obj; 3318 boolean_t has_features; 3319 3320 /* 3321 * If this pool already exists, return failure. 3322 */ 3323 mutex_enter(&spa_namespace_lock); 3324 if (spa_lookup(pool) != NULL) { 3325 mutex_exit(&spa_namespace_lock); 3326 return (EEXIST); 3327 } 3328 3329 /* 3330 * Allocate a new spa_t structure. 3331 */ 3332 (void) nvlist_lookup_string(props, 3333 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3334 spa = spa_add(pool, NULL, altroot); 3335 spa_activate(spa, spa_mode_global); 3336 3337 if (props && (error = spa_prop_validate(spa, props))) { 3338 spa_deactivate(spa); 3339 spa_remove(spa); 3340 mutex_exit(&spa_namespace_lock); 3341 return (error); 3342 } 3343 3344 has_features = B_FALSE; 3345 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3346 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3347 if (zpool_prop_feature(nvpair_name(elem))) 3348 has_features = B_TRUE; 3349 } 3350 3351 if (has_features || nvlist_lookup_uint64(props, 3352 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3353 version = SPA_VERSION; 3354 } 3355 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3356 3357 spa->spa_first_txg = txg; 3358 spa->spa_uberblock.ub_txg = txg - 1; 3359 spa->spa_uberblock.ub_version = version; 3360 spa->spa_ubsync = spa->spa_uberblock; 3361 3362 /* 3363 * Create "The Godfather" zio to hold all async IOs 3364 */ 3365 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 3366 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 3367 3368 /* 3369 * Create the root vdev. 3370 */ 3371 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3372 3373 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3374 3375 ASSERT(error != 0 || rvd != NULL); 3376 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3377 3378 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3379 error = EINVAL; 3380 3381 if (error == 0 && 3382 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3383 (error = spa_validate_aux(spa, nvroot, txg, 3384 VDEV_ALLOC_ADD)) == 0) { 3385 for (int c = 0; c < rvd->vdev_children; c++) { 3386 vdev_metaslab_set_size(rvd->vdev_child[c]); 3387 vdev_expand(rvd->vdev_child[c], txg); 3388 } 3389 } 3390 3391 spa_config_exit(spa, SCL_ALL, FTAG); 3392 3393 if (error != 0) { 3394 spa_unload(spa); 3395 spa_deactivate(spa); 3396 spa_remove(spa); 3397 mutex_exit(&spa_namespace_lock); 3398 return (error); 3399 } 3400 3401 /* 3402 * Get the list of spares, if specified. 3403 */ 3404 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3405 &spares, &nspares) == 0) { 3406 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3407 KM_SLEEP) == 0); 3408 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3409 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3410 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3411 spa_load_spares(spa); 3412 spa_config_exit(spa, SCL_ALL, FTAG); 3413 spa->spa_spares.sav_sync = B_TRUE; 3414 } 3415 3416 /* 3417 * Get the list of level 2 cache devices, if specified. 3418 */ 3419 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3420 &l2cache, &nl2cache) == 0) { 3421 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3422 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3423 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3424 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3425 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3426 spa_load_l2cache(spa); 3427 spa_config_exit(spa, SCL_ALL, FTAG); 3428 spa->spa_l2cache.sav_sync = B_TRUE; 3429 } 3430 3431 spa->spa_is_initializing = B_TRUE; 3432 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3433 spa->spa_meta_objset = dp->dp_meta_objset; 3434 spa->spa_is_initializing = B_FALSE; 3435 3436 /* 3437 * Create DDTs (dedup tables). 3438 */ 3439 ddt_create(spa); 3440 3441 spa_update_dspace(spa); 3442 3443 tx = dmu_tx_create_assigned(dp, txg); 3444 3445 /* 3446 * Create the pool config object. 3447 */ 3448 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3449 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3450 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3451 3452 if (zap_add(spa->spa_meta_objset, 3453 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3454 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3455 cmn_err(CE_PANIC, "failed to add pool config"); 3456 } 3457 3458 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3459 spa_feature_create_zap_objects(spa, tx); 3460 3461 if (zap_add(spa->spa_meta_objset, 3462 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3463 sizeof (uint64_t), 1, &version, tx) != 0) { 3464 cmn_err(CE_PANIC, "failed to add pool version"); 3465 } 3466 3467 /* Newly created pools with the right version are always deflated. */ 3468 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3469 spa->spa_deflate = TRUE; 3470 if (zap_add(spa->spa_meta_objset, 3471 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3472 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3473 cmn_err(CE_PANIC, "failed to add deflate"); 3474 } 3475 } 3476 3477 /* 3478 * Create the deferred-free bpobj. Turn off compression 3479 * because sync-to-convergence takes longer if the blocksize 3480 * keeps changing. 3481 */ 3482 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3483 dmu_object_set_compress(spa->spa_meta_objset, obj, 3484 ZIO_COMPRESS_OFF, tx); 3485 if (zap_add(spa->spa_meta_objset, 3486 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3487 sizeof (uint64_t), 1, &obj, tx) != 0) { 3488 cmn_err(CE_PANIC, "failed to add bpobj"); 3489 } 3490 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3491 spa->spa_meta_objset, obj)); 3492 3493 /* 3494 * Create the pool's history object. 3495 */ 3496 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3497 spa_history_create_obj(spa, tx); 3498 3499 /* 3500 * Set pool properties. 3501 */ 3502 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3503 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3504 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3505 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3506 3507 if (props != NULL) { 3508 spa_configfile_set(spa, props, B_FALSE); 3509 spa_sync_props(spa, props, tx); 3510 } 3511 3512 dmu_tx_commit(tx); 3513 3514 spa->spa_sync_on = B_TRUE; 3515 txg_sync_start(spa->spa_dsl_pool); 3516 3517 /* 3518 * We explicitly wait for the first transaction to complete so that our 3519 * bean counters are appropriately updated. 3520 */ 3521 txg_wait_synced(spa->spa_dsl_pool, txg); 3522 3523 spa_config_sync(spa, B_FALSE, B_TRUE); 3524 3525 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL) 3526 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE); 3527 spa_history_log_version(spa, LOG_POOL_CREATE); 3528 3529 spa->spa_minref = refcount_count(&spa->spa_refcount); 3530 3531 mutex_exit(&spa_namespace_lock); 3532 3533 return (0); 3534} 3535 3536#if defined(sun) 3537#ifdef _KERNEL 3538/* 3539 * Get the root pool information from the root disk, then import the root pool 3540 * during the system boot up time. 3541 */ 3542extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3543 3544static nvlist_t * 3545spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3546{ 3547 nvlist_t *config; 3548 nvlist_t *nvtop, *nvroot; 3549 uint64_t pgid; 3550 3551 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3552 return (NULL); 3553 3554 /* 3555 * Add this top-level vdev to the child array. 3556 */ 3557 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3558 &nvtop) == 0); 3559 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3560 &pgid) == 0); 3561 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3562 3563 /* 3564 * Put this pool's top-level vdevs into a root vdev. 3565 */ 3566 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3567 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3568 VDEV_TYPE_ROOT) == 0); 3569 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3570 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3571 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3572 &nvtop, 1) == 0); 3573 3574 /* 3575 * Replace the existing vdev_tree with the new root vdev in 3576 * this pool's configuration (remove the old, add the new). 3577 */ 3578 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3579 nvlist_free(nvroot); 3580 return (config); 3581} 3582 3583/* 3584 * Walk the vdev tree and see if we can find a device with "better" 3585 * configuration. A configuration is "better" if the label on that 3586 * device has a more recent txg. 3587 */ 3588static void 3589spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3590{ 3591 for (int c = 0; c < vd->vdev_children; c++) 3592 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3593 3594 if (vd->vdev_ops->vdev_op_leaf) { 3595 nvlist_t *label; 3596 uint64_t label_txg; 3597 3598 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3599 &label) != 0) 3600 return; 3601 3602 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3603 &label_txg) == 0); 3604 3605 /* 3606 * Do we have a better boot device? 3607 */ 3608 if (label_txg > *txg) { 3609 *txg = label_txg; 3610 *avd = vd; 3611 } 3612 nvlist_free(label); 3613 } 3614} 3615 3616/* 3617 * Import a root pool. 3618 * 3619 * For x86. devpath_list will consist of devid and/or physpath name of 3620 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3621 * The GRUB "findroot" command will return the vdev we should boot. 3622 * 3623 * For Sparc, devpath_list consists the physpath name of the booting device 3624 * no matter the rootpool is a single device pool or a mirrored pool. 3625 * e.g. 3626 * "/pci@1f,0/ide@d/disk@0,0:a" 3627 */ 3628int 3629spa_import_rootpool(char *devpath, char *devid) 3630{ 3631 spa_t *spa; 3632 vdev_t *rvd, *bvd, *avd = NULL; 3633 nvlist_t *config, *nvtop; 3634 uint64_t guid, txg; 3635 char *pname; 3636 int error; 3637 3638 /* 3639 * Read the label from the boot device and generate a configuration. 3640 */ 3641 config = spa_generate_rootconf(devpath, devid, &guid); 3642#if defined(_OBP) && defined(_KERNEL) 3643 if (config == NULL) { 3644 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3645 /* iscsi boot */ 3646 get_iscsi_bootpath_phy(devpath); 3647 config = spa_generate_rootconf(devpath, devid, &guid); 3648 } 3649 } 3650#endif 3651 if (config == NULL) { 3652 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3653 devpath); 3654 return (EIO); 3655 } 3656 3657 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3658 &pname) == 0); 3659 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3660 3661 mutex_enter(&spa_namespace_lock); 3662 if ((spa = spa_lookup(pname)) != NULL) { 3663 /* 3664 * Remove the existing root pool from the namespace so that we 3665 * can replace it with the correct config we just read in. 3666 */ 3667 spa_remove(spa); 3668 } 3669 3670 spa = spa_add(pname, config, NULL); 3671 spa->spa_is_root = B_TRUE; 3672 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3673 3674 /* 3675 * Build up a vdev tree based on the boot device's label config. 3676 */ 3677 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3678 &nvtop) == 0); 3679 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3680 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3681 VDEV_ALLOC_ROOTPOOL); 3682 spa_config_exit(spa, SCL_ALL, FTAG); 3683 if (error) { 3684 mutex_exit(&spa_namespace_lock); 3685 nvlist_free(config); 3686 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3687 pname); 3688 return (error); 3689 } 3690 3691 /* 3692 * Get the boot vdev. 3693 */ 3694 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3695 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3696 (u_longlong_t)guid); 3697 error = ENOENT; 3698 goto out; 3699 } 3700 3701 /* 3702 * Determine if there is a better boot device. 3703 */ 3704 avd = bvd; 3705 spa_alt_rootvdev(rvd, &avd, &txg); 3706 if (avd != bvd) { 3707 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3708 "try booting from '%s'", avd->vdev_path); 3709 error = EINVAL; 3710 goto out; 3711 } 3712 3713 /* 3714 * If the boot device is part of a spare vdev then ensure that 3715 * we're booting off the active spare. 3716 */ 3717 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3718 !bvd->vdev_isspare) { 3719 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3720 "try booting from '%s'", 3721 bvd->vdev_parent-> 3722 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3723 error = EINVAL; 3724 goto out; 3725 } 3726 3727 error = 0; 3728 spa_history_log_version(spa, LOG_POOL_IMPORT); 3729out: 3730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3731 vdev_free(rvd); 3732 spa_config_exit(spa, SCL_ALL, FTAG); 3733 mutex_exit(&spa_namespace_lock); 3734 3735 nvlist_free(config); 3736 return (error); 3737} 3738 3739#endif 3740#endif /* sun */ 3741 3742/* 3743 * Import a non-root pool into the system. 3744 */ 3745int 3746spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3747{ 3748 spa_t *spa; 3749 char *altroot = NULL; 3750 spa_load_state_t state = SPA_LOAD_IMPORT; 3751 zpool_rewind_policy_t policy; 3752 uint64_t mode = spa_mode_global; 3753 uint64_t readonly = B_FALSE; 3754 int error; 3755 nvlist_t *nvroot; 3756 nvlist_t **spares, **l2cache; 3757 uint_t nspares, nl2cache; 3758 3759 /* 3760 * If a pool with this name exists, return failure. 3761 */ 3762 mutex_enter(&spa_namespace_lock); 3763 if (spa_lookup(pool) != NULL) { 3764 mutex_exit(&spa_namespace_lock); 3765 return (EEXIST); 3766 } 3767 3768 /* 3769 * Create and initialize the spa structure. 3770 */ 3771 (void) nvlist_lookup_string(props, 3772 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3773 (void) nvlist_lookup_uint64(props, 3774 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 3775 if (readonly) 3776 mode = FREAD; 3777 spa = spa_add(pool, config, altroot); 3778 spa->spa_import_flags = flags; 3779 3780 /* 3781 * Verbatim import - Take a pool and insert it into the namespace 3782 * as if it had been loaded at boot. 3783 */ 3784 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 3785 if (props != NULL) 3786 spa_configfile_set(spa, props, B_FALSE); 3787 3788 spa_config_sync(spa, B_FALSE, B_TRUE); 3789 3790 mutex_exit(&spa_namespace_lock); 3791 spa_history_log_version(spa, LOG_POOL_IMPORT); 3792 3793 return (0); 3794 } 3795 3796 spa_activate(spa, mode); 3797 3798 /* 3799 * Don't start async tasks until we know everything is healthy. 3800 */ 3801 spa_async_suspend(spa); 3802 3803 zpool_get_rewind_policy(config, &policy); 3804 if (policy.zrp_request & ZPOOL_DO_REWIND) 3805 state = SPA_LOAD_RECOVER; 3806 3807 /* 3808 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 3809 * because the user-supplied config is actually the one to trust when 3810 * doing an import. 3811 */ 3812 if (state != SPA_LOAD_RECOVER) 3813 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3814 3815 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 3816 policy.zrp_request); 3817 3818 /* 3819 * Propagate anything learned while loading the pool and pass it 3820 * back to caller (i.e. rewind info, missing devices, etc). 3821 */ 3822 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 3823 spa->spa_load_info) == 0); 3824 3825 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3826 /* 3827 * Toss any existing sparelist, as it doesn't have any validity 3828 * anymore, and conflicts with spa_has_spare(). 3829 */ 3830 if (spa->spa_spares.sav_config) { 3831 nvlist_free(spa->spa_spares.sav_config); 3832 spa->spa_spares.sav_config = NULL; 3833 spa_load_spares(spa); 3834 } 3835 if (spa->spa_l2cache.sav_config) { 3836 nvlist_free(spa->spa_l2cache.sav_config); 3837 spa->spa_l2cache.sav_config = NULL; 3838 spa_load_l2cache(spa); 3839 } 3840 3841 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3842 &nvroot) == 0); 3843 if (error == 0) 3844 error = spa_validate_aux(spa, nvroot, -1ULL, 3845 VDEV_ALLOC_SPARE); 3846 if (error == 0) 3847 error = spa_validate_aux(spa, nvroot, -1ULL, 3848 VDEV_ALLOC_L2CACHE); 3849 spa_config_exit(spa, SCL_ALL, FTAG); 3850 3851 if (props != NULL) 3852 spa_configfile_set(spa, props, B_FALSE); 3853 3854 if (error != 0 || (props && spa_writeable(spa) && 3855 (error = spa_prop_set(spa, props)))) { 3856 spa_unload(spa); 3857 spa_deactivate(spa); 3858 spa_remove(spa); 3859 mutex_exit(&spa_namespace_lock); 3860 return (error); 3861 } 3862 3863 spa_async_resume(spa); 3864 3865 /* 3866 * Override any spares and level 2 cache devices as specified by 3867 * the user, as these may have correct device names/devids, etc. 3868 */ 3869 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3870 &spares, &nspares) == 0) { 3871 if (spa->spa_spares.sav_config) 3872 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 3873 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 3874 else 3875 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 3876 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3877 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3878 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3879 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3880 spa_load_spares(spa); 3881 spa_config_exit(spa, SCL_ALL, FTAG); 3882 spa->spa_spares.sav_sync = B_TRUE; 3883 } 3884 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3885 &l2cache, &nl2cache) == 0) { 3886 if (spa->spa_l2cache.sav_config) 3887 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 3888 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 3889 else 3890 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3891 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3892 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3893 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3894 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3895 spa_load_l2cache(spa); 3896 spa_config_exit(spa, SCL_ALL, FTAG); 3897 spa->spa_l2cache.sav_sync = B_TRUE; 3898 } 3899 3900 /* 3901 * Check for any removed devices. 3902 */ 3903 if (spa->spa_autoreplace) { 3904 spa_aux_check_removed(&spa->spa_spares); 3905 spa_aux_check_removed(&spa->spa_l2cache); 3906 } 3907 3908 if (spa_writeable(spa)) { 3909 /* 3910 * Update the config cache to include the newly-imported pool. 3911 */ 3912 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 3913 } 3914 3915 /* 3916 * It's possible that the pool was expanded while it was exported. 3917 * We kick off an async task to handle this for us. 3918 */ 3919 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 3920 3921 mutex_exit(&spa_namespace_lock); 3922 spa_history_log_version(spa, LOG_POOL_IMPORT); 3923 3924#ifdef __FreeBSD__ 3925#ifdef _KERNEL 3926 zvol_create_minors(pool); 3927#endif 3928#endif 3929 return (0); 3930} 3931 3932nvlist_t * 3933spa_tryimport(nvlist_t *tryconfig) 3934{ 3935 nvlist_t *config = NULL; 3936 char *poolname; 3937 spa_t *spa; 3938 uint64_t state; 3939 int error; 3940 3941 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 3942 return (NULL); 3943 3944 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 3945 return (NULL); 3946 3947 /* 3948 * Create and initialize the spa structure. 3949 */ 3950 mutex_enter(&spa_namespace_lock); 3951 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 3952 spa_activate(spa, FREAD); 3953 3954 /* 3955 * Pass off the heavy lifting to spa_load(). 3956 * Pass TRUE for mosconfig because the user-supplied config 3957 * is actually the one to trust when doing an import. 3958 */ 3959 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 3960 3961 /* 3962 * If 'tryconfig' was at least parsable, return the current config. 3963 */ 3964 if (spa->spa_root_vdev != NULL) { 3965 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 3966 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 3967 poolname) == 0); 3968 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 3969 state) == 0); 3970 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 3971 spa->spa_uberblock.ub_timestamp) == 0); 3972 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 3973 spa->spa_load_info) == 0); 3974 3975 /* 3976 * If the bootfs property exists on this pool then we 3977 * copy it out so that external consumers can tell which 3978 * pools are bootable. 3979 */ 3980 if ((!error || error == EEXIST) && spa->spa_bootfs) { 3981 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3982 3983 /* 3984 * We have to play games with the name since the 3985 * pool was opened as TRYIMPORT_NAME. 3986 */ 3987 if (dsl_dsobj_to_dsname(spa_name(spa), 3988 spa->spa_bootfs, tmpname) == 0) { 3989 char *cp; 3990 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 3991 3992 cp = strchr(tmpname, '/'); 3993 if (cp == NULL) { 3994 (void) strlcpy(dsname, tmpname, 3995 MAXPATHLEN); 3996 } else { 3997 (void) snprintf(dsname, MAXPATHLEN, 3998 "%s/%s", poolname, ++cp); 3999 } 4000 VERIFY(nvlist_add_string(config, 4001 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4002 kmem_free(dsname, MAXPATHLEN); 4003 } 4004 kmem_free(tmpname, MAXPATHLEN); 4005 } 4006 4007 /* 4008 * Add the list of hot spares and level 2 cache devices. 4009 */ 4010 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4011 spa_add_spares(spa, config); 4012 spa_add_l2cache(spa, config); 4013 spa_config_exit(spa, SCL_CONFIG, FTAG); 4014 } 4015 4016 spa_unload(spa); 4017 spa_deactivate(spa); 4018 spa_remove(spa); 4019 mutex_exit(&spa_namespace_lock); 4020 4021 return (config); 4022} 4023 4024/* 4025 * Pool export/destroy 4026 * 4027 * The act of destroying or exporting a pool is very simple. We make sure there 4028 * is no more pending I/O and any references to the pool are gone. Then, we 4029 * update the pool state and sync all the labels to disk, removing the 4030 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4031 * we don't sync the labels or remove the configuration cache. 4032 */ 4033static int 4034spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4035 boolean_t force, boolean_t hardforce) 4036{ 4037 spa_t *spa; 4038 4039 if (oldconfig) 4040 *oldconfig = NULL; 4041 4042 if (!(spa_mode_global & FWRITE)) 4043 return (EROFS); 4044 4045 mutex_enter(&spa_namespace_lock); 4046 if ((spa = spa_lookup(pool)) == NULL) { 4047 mutex_exit(&spa_namespace_lock); 4048 return (ENOENT); 4049 } 4050 4051 /* 4052 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4053 * reacquire the namespace lock, and see if we can export. 4054 */ 4055 spa_open_ref(spa, FTAG); 4056 mutex_exit(&spa_namespace_lock); 4057 spa_async_suspend(spa); 4058 mutex_enter(&spa_namespace_lock); 4059 spa_close(spa, FTAG); 4060 4061 /* 4062 * The pool will be in core if it's openable, 4063 * in which case we can modify its state. 4064 */ 4065 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4066 /* 4067 * Objsets may be open only because they're dirty, so we 4068 * have to force it to sync before checking spa_refcnt. 4069 */ 4070 txg_wait_synced(spa->spa_dsl_pool, 0); 4071 4072 /* 4073 * A pool cannot be exported or destroyed if there are active 4074 * references. If we are resetting a pool, allow references by 4075 * fault injection handlers. 4076 */ 4077 if (!spa_refcount_zero(spa) || 4078 (spa->spa_inject_ref != 0 && 4079 new_state != POOL_STATE_UNINITIALIZED)) { 4080 spa_async_resume(spa); 4081 mutex_exit(&spa_namespace_lock); 4082 return (EBUSY); 4083 } 4084 4085 /* 4086 * A pool cannot be exported if it has an active shared spare. 4087 * This is to prevent other pools stealing the active spare 4088 * from an exported pool. At user's own will, such pool can 4089 * be forcedly exported. 4090 */ 4091 if (!force && new_state == POOL_STATE_EXPORTED && 4092 spa_has_active_shared_spare(spa)) { 4093 spa_async_resume(spa); 4094 mutex_exit(&spa_namespace_lock); 4095 return (EXDEV); 4096 } 4097 4098 /* 4099 * We want this to be reflected on every label, 4100 * so mark them all dirty. spa_unload() will do the 4101 * final sync that pushes these changes out. 4102 */ 4103 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4104 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4105 spa->spa_state = new_state; 4106 spa->spa_final_txg = spa_last_synced_txg(spa) + 4107 TXG_DEFER_SIZE + 1; 4108 vdev_config_dirty(spa->spa_root_vdev); 4109 spa_config_exit(spa, SCL_ALL, FTAG); 4110 } 4111 } 4112 4113 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4114 4115 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4116 spa_unload(spa); 4117 spa_deactivate(spa); 4118 } 4119 4120 if (oldconfig && spa->spa_config) 4121 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4122 4123 if (new_state != POOL_STATE_UNINITIALIZED) { 4124 if (!hardforce) 4125 spa_config_sync(spa, B_TRUE, B_TRUE); 4126 spa_remove(spa); 4127 } 4128 mutex_exit(&spa_namespace_lock); 4129 4130 return (0); 4131} 4132 4133/* 4134 * Destroy a storage pool. 4135 */ 4136int 4137spa_destroy(char *pool) 4138{ 4139 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4140 B_FALSE, B_FALSE)); 4141} 4142 4143/* 4144 * Export a storage pool. 4145 */ 4146int 4147spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4148 boolean_t hardforce) 4149{ 4150 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4151 force, hardforce)); 4152} 4153 4154/* 4155 * Similar to spa_export(), this unloads the spa_t without actually removing it 4156 * from the namespace in any way. 4157 */ 4158int 4159spa_reset(char *pool) 4160{ 4161 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4162 B_FALSE, B_FALSE)); 4163} 4164 4165/* 4166 * ========================================================================== 4167 * Device manipulation 4168 * ========================================================================== 4169 */ 4170 4171/* 4172 * Add a device to a storage pool. 4173 */ 4174int 4175spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4176{ 4177 uint64_t txg, id; 4178 int error; 4179 vdev_t *rvd = spa->spa_root_vdev; 4180 vdev_t *vd, *tvd; 4181 nvlist_t **spares, **l2cache; 4182 uint_t nspares, nl2cache; 4183 4184 ASSERT(spa_writeable(spa)); 4185 4186 txg = spa_vdev_enter(spa); 4187 4188 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4189 VDEV_ALLOC_ADD)) != 0) 4190 return (spa_vdev_exit(spa, NULL, txg, error)); 4191 4192 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4193 4194 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4195 &nspares) != 0) 4196 nspares = 0; 4197 4198 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4199 &nl2cache) != 0) 4200 nl2cache = 0; 4201 4202 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4203 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4204 4205 if (vd->vdev_children != 0 && 4206 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4207 return (spa_vdev_exit(spa, vd, txg, error)); 4208 4209 /* 4210 * We must validate the spares and l2cache devices after checking the 4211 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4212 */ 4213 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4214 return (spa_vdev_exit(spa, vd, txg, error)); 4215 4216 /* 4217 * Transfer each new top-level vdev from vd to rvd. 4218 */ 4219 for (int c = 0; c < vd->vdev_children; c++) { 4220 4221 /* 4222 * Set the vdev id to the first hole, if one exists. 4223 */ 4224 for (id = 0; id < rvd->vdev_children; id++) { 4225 if (rvd->vdev_child[id]->vdev_ishole) { 4226 vdev_free(rvd->vdev_child[id]); 4227 break; 4228 } 4229 } 4230 tvd = vd->vdev_child[c]; 4231 vdev_remove_child(vd, tvd); 4232 tvd->vdev_id = id; 4233 vdev_add_child(rvd, tvd); 4234 vdev_config_dirty(tvd); 4235 } 4236 4237 if (nspares != 0) { 4238 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4239 ZPOOL_CONFIG_SPARES); 4240 spa_load_spares(spa); 4241 spa->spa_spares.sav_sync = B_TRUE; 4242 } 4243 4244 if (nl2cache != 0) { 4245 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4246 ZPOOL_CONFIG_L2CACHE); 4247 spa_load_l2cache(spa); 4248 spa->spa_l2cache.sav_sync = B_TRUE; 4249 } 4250 4251 /* 4252 * We have to be careful when adding new vdevs to an existing pool. 4253 * If other threads start allocating from these vdevs before we 4254 * sync the config cache, and we lose power, then upon reboot we may 4255 * fail to open the pool because there are DVAs that the config cache 4256 * can't translate. Therefore, we first add the vdevs without 4257 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4258 * and then let spa_config_update() initialize the new metaslabs. 4259 * 4260 * spa_load() checks for added-but-not-initialized vdevs, so that 4261 * if we lose power at any point in this sequence, the remaining 4262 * steps will be completed the next time we load the pool. 4263 */ 4264 (void) spa_vdev_exit(spa, vd, txg, 0); 4265 4266 mutex_enter(&spa_namespace_lock); 4267 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4268 mutex_exit(&spa_namespace_lock); 4269 4270 return (0); 4271} 4272 4273/* 4274 * Attach a device to a mirror. The arguments are the path to any device 4275 * in the mirror, and the nvroot for the new device. If the path specifies 4276 * a device that is not mirrored, we automatically insert the mirror vdev. 4277 * 4278 * If 'replacing' is specified, the new device is intended to replace the 4279 * existing device; in this case the two devices are made into their own 4280 * mirror using the 'replacing' vdev, which is functionally identical to 4281 * the mirror vdev (it actually reuses all the same ops) but has a few 4282 * extra rules: you can't attach to it after it's been created, and upon 4283 * completion of resilvering, the first disk (the one being replaced) 4284 * is automatically detached. 4285 */ 4286int 4287spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4288{ 4289 uint64_t txg, dtl_max_txg; 4290 vdev_t *rvd = spa->spa_root_vdev; 4291 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4292 vdev_ops_t *pvops; 4293 char *oldvdpath, *newvdpath; 4294 int newvd_isspare; 4295 int error; 4296 4297 ASSERT(spa_writeable(spa)); 4298 4299 txg = spa_vdev_enter(spa); 4300 4301 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4302 4303 if (oldvd == NULL) 4304 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4305 4306 if (!oldvd->vdev_ops->vdev_op_leaf) 4307 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4308 4309 pvd = oldvd->vdev_parent; 4310 4311 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4312 VDEV_ALLOC_ATTACH)) != 0) 4313 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4314 4315 if (newrootvd->vdev_children != 1) 4316 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4317 4318 newvd = newrootvd->vdev_child[0]; 4319 4320 if (!newvd->vdev_ops->vdev_op_leaf) 4321 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4322 4323 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4324 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4325 4326 /* 4327 * Spares can't replace logs 4328 */ 4329 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4330 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4331 4332 if (!replacing) { 4333 /* 4334 * For attach, the only allowable parent is a mirror or the root 4335 * vdev. 4336 */ 4337 if (pvd->vdev_ops != &vdev_mirror_ops && 4338 pvd->vdev_ops != &vdev_root_ops) 4339 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4340 4341 pvops = &vdev_mirror_ops; 4342 } else { 4343 /* 4344 * Active hot spares can only be replaced by inactive hot 4345 * spares. 4346 */ 4347 if (pvd->vdev_ops == &vdev_spare_ops && 4348 oldvd->vdev_isspare && 4349 !spa_has_spare(spa, newvd->vdev_guid)) 4350 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4351 4352 /* 4353 * If the source is a hot spare, and the parent isn't already a 4354 * spare, then we want to create a new hot spare. Otherwise, we 4355 * want to create a replacing vdev. The user is not allowed to 4356 * attach to a spared vdev child unless the 'isspare' state is 4357 * the same (spare replaces spare, non-spare replaces 4358 * non-spare). 4359 */ 4360 if (pvd->vdev_ops == &vdev_replacing_ops && 4361 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4362 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4363 } else if (pvd->vdev_ops == &vdev_spare_ops && 4364 newvd->vdev_isspare != oldvd->vdev_isspare) { 4365 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4366 } 4367 4368 if (newvd->vdev_isspare) 4369 pvops = &vdev_spare_ops; 4370 else 4371 pvops = &vdev_replacing_ops; 4372 } 4373 4374 /* 4375 * Make sure the new device is big enough. 4376 */ 4377 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4378 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4379 4380 /* 4381 * The new device cannot have a higher alignment requirement 4382 * than the top-level vdev. 4383 */ 4384 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4385 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4386 4387 /* 4388 * If this is an in-place replacement, update oldvd's path and devid 4389 * to make it distinguishable from newvd, and unopenable from now on. 4390 */ 4391 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4392 spa_strfree(oldvd->vdev_path); 4393 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4394 KM_SLEEP); 4395 (void) sprintf(oldvd->vdev_path, "%s/%s", 4396 newvd->vdev_path, "old"); 4397 if (oldvd->vdev_devid != NULL) { 4398 spa_strfree(oldvd->vdev_devid); 4399 oldvd->vdev_devid = NULL; 4400 } 4401 } 4402 4403 /* mark the device being resilvered */ 4404 newvd->vdev_resilvering = B_TRUE; 4405 4406 /* 4407 * If the parent is not a mirror, or if we're replacing, insert the new 4408 * mirror/replacing/spare vdev above oldvd. 4409 */ 4410 if (pvd->vdev_ops != pvops) 4411 pvd = vdev_add_parent(oldvd, pvops); 4412 4413 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4414 ASSERT(pvd->vdev_ops == pvops); 4415 ASSERT(oldvd->vdev_parent == pvd); 4416 4417 /* 4418 * Extract the new device from its root and add it to pvd. 4419 */ 4420 vdev_remove_child(newrootvd, newvd); 4421 newvd->vdev_id = pvd->vdev_children; 4422 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4423 vdev_add_child(pvd, newvd); 4424 4425 tvd = newvd->vdev_top; 4426 ASSERT(pvd->vdev_top == tvd); 4427 ASSERT(tvd->vdev_parent == rvd); 4428 4429 vdev_config_dirty(tvd); 4430 4431 /* 4432 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4433 * for any dmu_sync-ed blocks. It will propagate upward when 4434 * spa_vdev_exit() calls vdev_dtl_reassess(). 4435 */ 4436 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4437 4438 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4439 dtl_max_txg - TXG_INITIAL); 4440 4441 if (newvd->vdev_isspare) { 4442 spa_spare_activate(newvd); 4443 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4444 } 4445 4446 oldvdpath = spa_strdup(oldvd->vdev_path); 4447 newvdpath = spa_strdup(newvd->vdev_path); 4448 newvd_isspare = newvd->vdev_isspare; 4449 4450 /* 4451 * Mark newvd's DTL dirty in this txg. 4452 */ 4453 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4454 4455 /* 4456 * Restart the resilver 4457 */ 4458 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4459 4460 /* 4461 * Commit the config 4462 */ 4463 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4464 4465 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL, 4466 "%s vdev=%s %s vdev=%s", 4467 replacing && newvd_isspare ? "spare in" : 4468 replacing ? "replace" : "attach", newvdpath, 4469 replacing ? "for" : "to", oldvdpath); 4470 4471 spa_strfree(oldvdpath); 4472 spa_strfree(newvdpath); 4473 4474 if (spa->spa_bootfs) 4475 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4476 4477 return (0); 4478} 4479 4480/* 4481 * Detach a device from a mirror or replacing vdev. 4482 * If 'replace_done' is specified, only detach if the parent 4483 * is a replacing vdev. 4484 */ 4485int 4486spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4487{ 4488 uint64_t txg; 4489 int error; 4490 vdev_t *rvd = spa->spa_root_vdev; 4491 vdev_t *vd, *pvd, *cvd, *tvd; 4492 boolean_t unspare = B_FALSE; 4493 uint64_t unspare_guid; 4494 char *vdpath; 4495 4496 ASSERT(spa_writeable(spa)); 4497 4498 txg = spa_vdev_enter(spa); 4499 4500 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4501 4502 if (vd == NULL) 4503 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4504 4505 if (!vd->vdev_ops->vdev_op_leaf) 4506 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4507 4508 pvd = vd->vdev_parent; 4509 4510 /* 4511 * If the parent/child relationship is not as expected, don't do it. 4512 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4513 * vdev that's replacing B with C. The user's intent in replacing 4514 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4515 * the replace by detaching C, the expected behavior is to end up 4516 * M(A,B). But suppose that right after deciding to detach C, 4517 * the replacement of B completes. We would have M(A,C), and then 4518 * ask to detach C, which would leave us with just A -- not what 4519 * the user wanted. To prevent this, we make sure that the 4520 * parent/child relationship hasn't changed -- in this example, 4521 * that C's parent is still the replacing vdev R. 4522 */ 4523 if (pvd->vdev_guid != pguid && pguid != 0) 4524 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4525 4526 /* 4527 * Only 'replacing' or 'spare' vdevs can be replaced. 4528 */ 4529 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4530 pvd->vdev_ops != &vdev_spare_ops) 4531 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4532 4533 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4534 spa_version(spa) >= SPA_VERSION_SPARES); 4535 4536 /* 4537 * Only mirror, replacing, and spare vdevs support detach. 4538 */ 4539 if (pvd->vdev_ops != &vdev_replacing_ops && 4540 pvd->vdev_ops != &vdev_mirror_ops && 4541 pvd->vdev_ops != &vdev_spare_ops) 4542 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4543 4544 /* 4545 * If this device has the only valid copy of some data, 4546 * we cannot safely detach it. 4547 */ 4548 if (vdev_dtl_required(vd)) 4549 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4550 4551 ASSERT(pvd->vdev_children >= 2); 4552 4553 /* 4554 * If we are detaching the second disk from a replacing vdev, then 4555 * check to see if we changed the original vdev's path to have "/old" 4556 * at the end in spa_vdev_attach(). If so, undo that change now. 4557 */ 4558 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4559 vd->vdev_path != NULL) { 4560 size_t len = strlen(vd->vdev_path); 4561 4562 for (int c = 0; c < pvd->vdev_children; c++) { 4563 cvd = pvd->vdev_child[c]; 4564 4565 if (cvd == vd || cvd->vdev_path == NULL) 4566 continue; 4567 4568 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4569 strcmp(cvd->vdev_path + len, "/old") == 0) { 4570 spa_strfree(cvd->vdev_path); 4571 cvd->vdev_path = spa_strdup(vd->vdev_path); 4572 break; 4573 } 4574 } 4575 } 4576 4577 /* 4578 * If we are detaching the original disk from a spare, then it implies 4579 * that the spare should become a real disk, and be removed from the 4580 * active spare list for the pool. 4581 */ 4582 if (pvd->vdev_ops == &vdev_spare_ops && 4583 vd->vdev_id == 0 && 4584 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4585 unspare = B_TRUE; 4586 4587 /* 4588 * Erase the disk labels so the disk can be used for other things. 4589 * This must be done after all other error cases are handled, 4590 * but before we disembowel vd (so we can still do I/O to it). 4591 * But if we can't do it, don't treat the error as fatal -- 4592 * it may be that the unwritability of the disk is the reason 4593 * it's being detached! 4594 */ 4595 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4596 4597 /* 4598 * Remove vd from its parent and compact the parent's children. 4599 */ 4600 vdev_remove_child(pvd, vd); 4601 vdev_compact_children(pvd); 4602 4603 /* 4604 * Remember one of the remaining children so we can get tvd below. 4605 */ 4606 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4607 4608 /* 4609 * If we need to remove the remaining child from the list of hot spares, 4610 * do it now, marking the vdev as no longer a spare in the process. 4611 * We must do this before vdev_remove_parent(), because that can 4612 * change the GUID if it creates a new toplevel GUID. For a similar 4613 * reason, we must remove the spare now, in the same txg as the detach; 4614 * otherwise someone could attach a new sibling, change the GUID, and 4615 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4616 */ 4617 if (unspare) { 4618 ASSERT(cvd->vdev_isspare); 4619 spa_spare_remove(cvd); 4620 unspare_guid = cvd->vdev_guid; 4621 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4622 cvd->vdev_unspare = B_TRUE; 4623 } 4624 4625 /* 4626 * If the parent mirror/replacing vdev only has one child, 4627 * the parent is no longer needed. Remove it from the tree. 4628 */ 4629 if (pvd->vdev_children == 1) { 4630 if (pvd->vdev_ops == &vdev_spare_ops) 4631 cvd->vdev_unspare = B_FALSE; 4632 vdev_remove_parent(cvd); 4633 cvd->vdev_resilvering = B_FALSE; 4634 } 4635 4636 4637 /* 4638 * We don't set tvd until now because the parent we just removed 4639 * may have been the previous top-level vdev. 4640 */ 4641 tvd = cvd->vdev_top; 4642 ASSERT(tvd->vdev_parent == rvd); 4643 4644 /* 4645 * Reevaluate the parent vdev state. 4646 */ 4647 vdev_propagate_state(cvd); 4648 4649 /* 4650 * If the 'autoexpand' property is set on the pool then automatically 4651 * try to expand the size of the pool. For example if the device we 4652 * just detached was smaller than the others, it may be possible to 4653 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4654 * first so that we can obtain the updated sizes of the leaf vdevs. 4655 */ 4656 if (spa->spa_autoexpand) { 4657 vdev_reopen(tvd); 4658 vdev_expand(tvd, txg); 4659 } 4660 4661 vdev_config_dirty(tvd); 4662 4663 /* 4664 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4665 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4666 * But first make sure we're not on any *other* txg's DTL list, to 4667 * prevent vd from being accessed after it's freed. 4668 */ 4669 vdpath = spa_strdup(vd->vdev_path); 4670 for (int t = 0; t < TXG_SIZE; t++) 4671 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4672 vd->vdev_detached = B_TRUE; 4673 vdev_dirty(tvd, VDD_DTL, vd, txg); 4674 4675 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4676 4677 /* hang on to the spa before we release the lock */ 4678 spa_open_ref(spa, FTAG); 4679 4680 error = spa_vdev_exit(spa, vd, txg, 0); 4681 4682 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL, 4683 "vdev=%s", vdpath); 4684 spa_strfree(vdpath); 4685 4686 /* 4687 * If this was the removal of the original device in a hot spare vdev, 4688 * then we want to go through and remove the device from the hot spare 4689 * list of every other pool. 4690 */ 4691 if (unspare) { 4692 spa_t *altspa = NULL; 4693 4694 mutex_enter(&spa_namespace_lock); 4695 while ((altspa = spa_next(altspa)) != NULL) { 4696 if (altspa->spa_state != POOL_STATE_ACTIVE || 4697 altspa == spa) 4698 continue; 4699 4700 spa_open_ref(altspa, FTAG); 4701 mutex_exit(&spa_namespace_lock); 4702 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 4703 mutex_enter(&spa_namespace_lock); 4704 spa_close(altspa, FTAG); 4705 } 4706 mutex_exit(&spa_namespace_lock); 4707 4708 /* search the rest of the vdevs for spares to remove */ 4709 spa_vdev_resilver_done(spa); 4710 } 4711 4712 /* all done with the spa; OK to release */ 4713 mutex_enter(&spa_namespace_lock); 4714 spa_close(spa, FTAG); 4715 mutex_exit(&spa_namespace_lock); 4716 4717 return (error); 4718} 4719 4720/* 4721 * Split a set of devices from their mirrors, and create a new pool from them. 4722 */ 4723int 4724spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 4725 nvlist_t *props, boolean_t exp) 4726{ 4727 int error = 0; 4728 uint64_t txg, *glist; 4729 spa_t *newspa; 4730 uint_t c, children, lastlog; 4731 nvlist_t **child, *nvl, *tmp; 4732 dmu_tx_t *tx; 4733 char *altroot = NULL; 4734 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 4735 boolean_t activate_slog; 4736 4737 ASSERT(spa_writeable(spa)); 4738 4739 txg = spa_vdev_enter(spa); 4740 4741 /* clear the log and flush everything up to now */ 4742 activate_slog = spa_passivate_log(spa); 4743 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 4744 error = spa_offline_log(spa); 4745 txg = spa_vdev_config_enter(spa); 4746 4747 if (activate_slog) 4748 spa_activate_log(spa); 4749 4750 if (error != 0) 4751 return (spa_vdev_exit(spa, NULL, txg, error)); 4752 4753 /* check new spa name before going any further */ 4754 if (spa_lookup(newname) != NULL) 4755 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 4756 4757 /* 4758 * scan through all the children to ensure they're all mirrors 4759 */ 4760 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 4761 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 4762 &children) != 0) 4763 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4764 4765 /* first, check to ensure we've got the right child count */ 4766 rvd = spa->spa_root_vdev; 4767 lastlog = 0; 4768 for (c = 0; c < rvd->vdev_children; c++) { 4769 vdev_t *vd = rvd->vdev_child[c]; 4770 4771 /* don't count the holes & logs as children */ 4772 if (vd->vdev_islog || vd->vdev_ishole) { 4773 if (lastlog == 0) 4774 lastlog = c; 4775 continue; 4776 } 4777 4778 lastlog = 0; 4779 } 4780 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 4781 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4782 4783 /* next, ensure no spare or cache devices are part of the split */ 4784 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 4785 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 4786 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4787 4788 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 4789 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 4790 4791 /* then, loop over each vdev and validate it */ 4792 for (c = 0; c < children; c++) { 4793 uint64_t is_hole = 0; 4794 4795 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 4796 &is_hole); 4797 4798 if (is_hole != 0) { 4799 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 4800 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 4801 continue; 4802 } else { 4803 error = EINVAL; 4804 break; 4805 } 4806 } 4807 4808 /* which disk is going to be split? */ 4809 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 4810 &glist[c]) != 0) { 4811 error = EINVAL; 4812 break; 4813 } 4814 4815 /* look it up in the spa */ 4816 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 4817 if (vml[c] == NULL) { 4818 error = ENODEV; 4819 break; 4820 } 4821 4822 /* make sure there's nothing stopping the split */ 4823 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 4824 vml[c]->vdev_islog || 4825 vml[c]->vdev_ishole || 4826 vml[c]->vdev_isspare || 4827 vml[c]->vdev_isl2cache || 4828 !vdev_writeable(vml[c]) || 4829 vml[c]->vdev_children != 0 || 4830 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 4831 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 4832 error = EINVAL; 4833 break; 4834 } 4835 4836 if (vdev_dtl_required(vml[c])) { 4837 error = EBUSY; 4838 break; 4839 } 4840 4841 /* we need certain info from the top level */ 4842 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 4843 vml[c]->vdev_top->vdev_ms_array) == 0); 4844 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 4845 vml[c]->vdev_top->vdev_ms_shift) == 0); 4846 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 4847 vml[c]->vdev_top->vdev_asize) == 0); 4848 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 4849 vml[c]->vdev_top->vdev_ashift) == 0); 4850 } 4851 4852 if (error != 0) { 4853 kmem_free(vml, children * sizeof (vdev_t *)); 4854 kmem_free(glist, children * sizeof (uint64_t)); 4855 return (spa_vdev_exit(spa, NULL, txg, error)); 4856 } 4857 4858 /* stop writers from using the disks */ 4859 for (c = 0; c < children; c++) { 4860 if (vml[c] != NULL) 4861 vml[c]->vdev_offline = B_TRUE; 4862 } 4863 vdev_reopen(spa->spa_root_vdev); 4864 4865 /* 4866 * Temporarily record the splitting vdevs in the spa config. This 4867 * will disappear once the config is regenerated. 4868 */ 4869 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4870 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 4871 glist, children) == 0); 4872 kmem_free(glist, children * sizeof (uint64_t)); 4873 4874 mutex_enter(&spa->spa_props_lock); 4875 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 4876 nvl) == 0); 4877 mutex_exit(&spa->spa_props_lock); 4878 spa->spa_config_splitting = nvl; 4879 vdev_config_dirty(spa->spa_root_vdev); 4880 4881 /* configure and create the new pool */ 4882 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 4883 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4884 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 4885 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 4886 spa_version(spa)) == 0); 4887 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 4888 spa->spa_config_txg) == 0); 4889 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 4890 spa_generate_guid(NULL)) == 0); 4891 (void) nvlist_lookup_string(props, 4892 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4893 4894 /* add the new pool to the namespace */ 4895 newspa = spa_add(newname, config, altroot); 4896 newspa->spa_config_txg = spa->spa_config_txg; 4897 spa_set_log_state(newspa, SPA_LOG_CLEAR); 4898 4899 /* release the spa config lock, retaining the namespace lock */ 4900 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 4901 4902 if (zio_injection_enabled) 4903 zio_handle_panic_injection(spa, FTAG, 1); 4904 4905 spa_activate(newspa, spa_mode_global); 4906 spa_async_suspend(newspa); 4907 4908#ifndef sun 4909 /* mark that we are creating new spa by splitting */ 4910 newspa->spa_splitting_newspa = B_TRUE; 4911#endif 4912 /* create the new pool from the disks of the original pool */ 4913 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 4914#ifndef sun 4915 newspa->spa_splitting_newspa = B_FALSE; 4916#endif 4917 if (error) 4918 goto out; 4919 4920 /* if that worked, generate a real config for the new pool */ 4921 if (newspa->spa_root_vdev != NULL) { 4922 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 4923 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4924 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 4925 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 4926 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 4927 B_TRUE)); 4928 } 4929 4930 /* set the props */ 4931 if (props != NULL) { 4932 spa_configfile_set(newspa, props, B_FALSE); 4933 error = spa_prop_set(newspa, props); 4934 if (error) 4935 goto out; 4936 } 4937 4938 /* flush everything */ 4939 txg = spa_vdev_config_enter(newspa); 4940 vdev_config_dirty(newspa->spa_root_vdev); 4941 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 4942 4943 if (zio_injection_enabled) 4944 zio_handle_panic_injection(spa, FTAG, 2); 4945 4946 spa_async_resume(newspa); 4947 4948 /* finally, update the original pool's config */ 4949 txg = spa_vdev_config_enter(spa); 4950 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 4951 error = dmu_tx_assign(tx, TXG_WAIT); 4952 if (error != 0) 4953 dmu_tx_abort(tx); 4954 for (c = 0; c < children; c++) { 4955 if (vml[c] != NULL) { 4956 vdev_split(vml[c]); 4957 if (error == 0) 4958 spa_history_log_internal(LOG_POOL_VDEV_DETACH, 4959 spa, tx, "vdev=%s", 4960 vml[c]->vdev_path); 4961 vdev_free(vml[c]); 4962 } 4963 } 4964 vdev_config_dirty(spa->spa_root_vdev); 4965 spa->spa_config_splitting = NULL; 4966 nvlist_free(nvl); 4967 if (error == 0) 4968 dmu_tx_commit(tx); 4969 (void) spa_vdev_exit(spa, NULL, txg, 0); 4970 4971 if (zio_injection_enabled) 4972 zio_handle_panic_injection(spa, FTAG, 3); 4973 4974 /* split is complete; log a history record */ 4975 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL, 4976 "split new pool %s from pool %s", newname, spa_name(spa)); 4977 4978 kmem_free(vml, children * sizeof (vdev_t *)); 4979 4980 /* if we're not going to mount the filesystems in userland, export */ 4981 if (exp) 4982 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 4983 B_FALSE, B_FALSE); 4984 4985 return (error); 4986 4987out: 4988 spa_unload(newspa); 4989 spa_deactivate(newspa); 4990 spa_remove(newspa); 4991 4992 txg = spa_vdev_config_enter(spa); 4993 4994 /* re-online all offlined disks */ 4995 for (c = 0; c < children; c++) { 4996 if (vml[c] != NULL) 4997 vml[c]->vdev_offline = B_FALSE; 4998 } 4999 vdev_reopen(spa->spa_root_vdev); 5000 5001 nvlist_free(spa->spa_config_splitting); 5002 spa->spa_config_splitting = NULL; 5003 (void) spa_vdev_exit(spa, NULL, txg, error); 5004 5005 kmem_free(vml, children * sizeof (vdev_t *)); 5006 return (error); 5007} 5008 5009static nvlist_t * 5010spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5011{ 5012 for (int i = 0; i < count; i++) { 5013 uint64_t guid; 5014 5015 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5016 &guid) == 0); 5017 5018 if (guid == target_guid) 5019 return (nvpp[i]); 5020 } 5021 5022 return (NULL); 5023} 5024 5025static void 5026spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5027 nvlist_t *dev_to_remove) 5028{ 5029 nvlist_t **newdev = NULL; 5030 5031 if (count > 1) 5032 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5033 5034 for (int i = 0, j = 0; i < count; i++) { 5035 if (dev[i] == dev_to_remove) 5036 continue; 5037 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5038 } 5039 5040 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5041 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5042 5043 for (int i = 0; i < count - 1; i++) 5044 nvlist_free(newdev[i]); 5045 5046 if (count > 1) 5047 kmem_free(newdev, (count - 1) * sizeof (void *)); 5048} 5049 5050/* 5051 * Evacuate the device. 5052 */ 5053static int 5054spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5055{ 5056 uint64_t txg; 5057 int error = 0; 5058 5059 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5060 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5061 ASSERT(vd == vd->vdev_top); 5062 5063 /* 5064 * Evacuate the device. We don't hold the config lock as writer 5065 * since we need to do I/O but we do keep the 5066 * spa_namespace_lock held. Once this completes the device 5067 * should no longer have any blocks allocated on it. 5068 */ 5069 if (vd->vdev_islog) { 5070 if (vd->vdev_stat.vs_alloc != 0) 5071 error = spa_offline_log(spa); 5072 } else { 5073 error = ENOTSUP; 5074 } 5075 5076 if (error) 5077 return (error); 5078 5079 /* 5080 * The evacuation succeeded. Remove any remaining MOS metadata 5081 * associated with this vdev, and wait for these changes to sync. 5082 */ 5083 ASSERT0(vd->vdev_stat.vs_alloc); 5084 txg = spa_vdev_config_enter(spa); 5085 vd->vdev_removing = B_TRUE; 5086 vdev_dirty(vd, 0, NULL, txg); 5087 vdev_config_dirty(vd); 5088 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5089 5090 return (0); 5091} 5092 5093/* 5094 * Complete the removal by cleaning up the namespace. 5095 */ 5096static void 5097spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5098{ 5099 vdev_t *rvd = spa->spa_root_vdev; 5100 uint64_t id = vd->vdev_id; 5101 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5102 5103 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5104 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5105 ASSERT(vd == vd->vdev_top); 5106 5107 /* 5108 * Only remove any devices which are empty. 5109 */ 5110 if (vd->vdev_stat.vs_alloc != 0) 5111 return; 5112 5113 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5114 5115 if (list_link_active(&vd->vdev_state_dirty_node)) 5116 vdev_state_clean(vd); 5117 if (list_link_active(&vd->vdev_config_dirty_node)) 5118 vdev_config_clean(vd); 5119 5120 vdev_free(vd); 5121 5122 if (last_vdev) { 5123 vdev_compact_children(rvd); 5124 } else { 5125 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5126 vdev_add_child(rvd, vd); 5127 } 5128 vdev_config_dirty(rvd); 5129 5130 /* 5131 * Reassess the health of our root vdev. 5132 */ 5133 vdev_reopen(rvd); 5134} 5135 5136/* 5137 * Remove a device from the pool - 5138 * 5139 * Removing a device from the vdev namespace requires several steps 5140 * and can take a significant amount of time. As a result we use 5141 * the spa_vdev_config_[enter/exit] functions which allow us to 5142 * grab and release the spa_config_lock while still holding the namespace 5143 * lock. During each step the configuration is synced out. 5144 */ 5145 5146/* 5147 * Remove a device from the pool. Currently, this supports removing only hot 5148 * spares, slogs, and level 2 ARC devices. 5149 */ 5150int 5151spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5152{ 5153 vdev_t *vd; 5154 metaslab_group_t *mg; 5155 nvlist_t **spares, **l2cache, *nv; 5156 uint64_t txg = 0; 5157 uint_t nspares, nl2cache; 5158 int error = 0; 5159 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5160 5161 ASSERT(spa_writeable(spa)); 5162 5163 if (!locked) 5164 txg = spa_vdev_enter(spa); 5165 5166 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5167 5168 if (spa->spa_spares.sav_vdevs != NULL && 5169 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5170 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5171 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5172 /* 5173 * Only remove the hot spare if it's not currently in use 5174 * in this pool. 5175 */ 5176 if (vd == NULL || unspare) { 5177 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5178 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5179 spa_load_spares(spa); 5180 spa->spa_spares.sav_sync = B_TRUE; 5181 } else { 5182 error = EBUSY; 5183 } 5184 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5185 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5186 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5187 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5188 /* 5189 * Cache devices can always be removed. 5190 */ 5191 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5192 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5193 spa_load_l2cache(spa); 5194 spa->spa_l2cache.sav_sync = B_TRUE; 5195 } else if (vd != NULL && vd->vdev_islog) { 5196 ASSERT(!locked); 5197 ASSERT(vd == vd->vdev_top); 5198 5199 /* 5200 * XXX - Once we have bp-rewrite this should 5201 * become the common case. 5202 */ 5203 5204 mg = vd->vdev_mg; 5205 5206 /* 5207 * Stop allocating from this vdev. 5208 */ 5209 metaslab_group_passivate(mg); 5210 5211 /* 5212 * Wait for the youngest allocations and frees to sync, 5213 * and then wait for the deferral of those frees to finish. 5214 */ 5215 spa_vdev_config_exit(spa, NULL, 5216 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5217 5218 /* 5219 * Attempt to evacuate the vdev. 5220 */ 5221 error = spa_vdev_remove_evacuate(spa, vd); 5222 5223 txg = spa_vdev_config_enter(spa); 5224 5225 /* 5226 * If we couldn't evacuate the vdev, unwind. 5227 */ 5228 if (error) { 5229 metaslab_group_activate(mg); 5230 return (spa_vdev_exit(spa, NULL, txg, error)); 5231 } 5232 5233 /* 5234 * Clean up the vdev namespace. 5235 */ 5236 spa_vdev_remove_from_namespace(spa, vd); 5237 5238 } else if (vd != NULL) { 5239 /* 5240 * Normal vdevs cannot be removed (yet). 5241 */ 5242 error = ENOTSUP; 5243 } else { 5244 /* 5245 * There is no vdev of any kind with the specified guid. 5246 */ 5247 error = ENOENT; 5248 } 5249 5250 if (!locked) 5251 return (spa_vdev_exit(spa, NULL, txg, error)); 5252 5253 return (error); 5254} 5255 5256/* 5257 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5258 * current spared, so we can detach it. 5259 */ 5260static vdev_t * 5261spa_vdev_resilver_done_hunt(vdev_t *vd) 5262{ 5263 vdev_t *newvd, *oldvd; 5264 5265 for (int c = 0; c < vd->vdev_children; c++) { 5266 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5267 if (oldvd != NULL) 5268 return (oldvd); 5269 } 5270 5271 /* 5272 * Check for a completed replacement. We always consider the first 5273 * vdev in the list to be the oldest vdev, and the last one to be 5274 * the newest (see spa_vdev_attach() for how that works). In 5275 * the case where the newest vdev is faulted, we will not automatically 5276 * remove it after a resilver completes. This is OK as it will require 5277 * user intervention to determine which disk the admin wishes to keep. 5278 */ 5279 if (vd->vdev_ops == &vdev_replacing_ops) { 5280 ASSERT(vd->vdev_children > 1); 5281 5282 newvd = vd->vdev_child[vd->vdev_children - 1]; 5283 oldvd = vd->vdev_child[0]; 5284 5285 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5286 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5287 !vdev_dtl_required(oldvd)) 5288 return (oldvd); 5289 } 5290 5291 /* 5292 * Check for a completed resilver with the 'unspare' flag set. 5293 */ 5294 if (vd->vdev_ops == &vdev_spare_ops) { 5295 vdev_t *first = vd->vdev_child[0]; 5296 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5297 5298 if (last->vdev_unspare) { 5299 oldvd = first; 5300 newvd = last; 5301 } else if (first->vdev_unspare) { 5302 oldvd = last; 5303 newvd = first; 5304 } else { 5305 oldvd = NULL; 5306 } 5307 5308 if (oldvd != NULL && 5309 vdev_dtl_empty(newvd, DTL_MISSING) && 5310 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5311 !vdev_dtl_required(oldvd)) 5312 return (oldvd); 5313 5314 /* 5315 * If there are more than two spares attached to a disk, 5316 * and those spares are not required, then we want to 5317 * attempt to free them up now so that they can be used 5318 * by other pools. Once we're back down to a single 5319 * disk+spare, we stop removing them. 5320 */ 5321 if (vd->vdev_children > 2) { 5322 newvd = vd->vdev_child[1]; 5323 5324 if (newvd->vdev_isspare && last->vdev_isspare && 5325 vdev_dtl_empty(last, DTL_MISSING) && 5326 vdev_dtl_empty(last, DTL_OUTAGE) && 5327 !vdev_dtl_required(newvd)) 5328 return (newvd); 5329 } 5330 } 5331 5332 return (NULL); 5333} 5334 5335static void 5336spa_vdev_resilver_done(spa_t *spa) 5337{ 5338 vdev_t *vd, *pvd, *ppvd; 5339 uint64_t guid, sguid, pguid, ppguid; 5340 5341 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5342 5343 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5344 pvd = vd->vdev_parent; 5345 ppvd = pvd->vdev_parent; 5346 guid = vd->vdev_guid; 5347 pguid = pvd->vdev_guid; 5348 ppguid = ppvd->vdev_guid; 5349 sguid = 0; 5350 /* 5351 * If we have just finished replacing a hot spared device, then 5352 * we need to detach the parent's first child (the original hot 5353 * spare) as well. 5354 */ 5355 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5356 ppvd->vdev_children == 2) { 5357 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5358 sguid = ppvd->vdev_child[1]->vdev_guid; 5359 } 5360 spa_config_exit(spa, SCL_ALL, FTAG); 5361 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5362 return; 5363 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5364 return; 5365 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5366 } 5367 5368 spa_config_exit(spa, SCL_ALL, FTAG); 5369} 5370 5371/* 5372 * Update the stored path or FRU for this vdev. 5373 */ 5374int 5375spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5376 boolean_t ispath) 5377{ 5378 vdev_t *vd; 5379 boolean_t sync = B_FALSE; 5380 5381 ASSERT(spa_writeable(spa)); 5382 5383 spa_vdev_state_enter(spa, SCL_ALL); 5384 5385 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5386 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5387 5388 if (!vd->vdev_ops->vdev_op_leaf) 5389 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5390 5391 if (ispath) { 5392 if (strcmp(value, vd->vdev_path) != 0) { 5393 spa_strfree(vd->vdev_path); 5394 vd->vdev_path = spa_strdup(value); 5395 sync = B_TRUE; 5396 } 5397 } else { 5398 if (vd->vdev_fru == NULL) { 5399 vd->vdev_fru = spa_strdup(value); 5400 sync = B_TRUE; 5401 } else if (strcmp(value, vd->vdev_fru) != 0) { 5402 spa_strfree(vd->vdev_fru); 5403 vd->vdev_fru = spa_strdup(value); 5404 sync = B_TRUE; 5405 } 5406 } 5407 5408 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5409} 5410 5411int 5412spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5413{ 5414 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5415} 5416 5417int 5418spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5419{ 5420 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5421} 5422 5423/* 5424 * ========================================================================== 5425 * SPA Scanning 5426 * ========================================================================== 5427 */ 5428 5429int 5430spa_scan_stop(spa_t *spa) 5431{ 5432 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5433 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5434 return (EBUSY); 5435 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5436} 5437 5438int 5439spa_scan(spa_t *spa, pool_scan_func_t func) 5440{ 5441 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5442 5443 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5444 return (ENOTSUP); 5445 5446 /* 5447 * If a resilver was requested, but there is no DTL on a 5448 * writeable leaf device, we have nothing to do. 5449 */ 5450 if (func == POOL_SCAN_RESILVER && 5451 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5452 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5453 return (0); 5454 } 5455 5456 return (dsl_scan(spa->spa_dsl_pool, func)); 5457} 5458 5459/* 5460 * ========================================================================== 5461 * SPA async task processing 5462 * ========================================================================== 5463 */ 5464 5465static void 5466spa_async_remove(spa_t *spa, vdev_t *vd) 5467{ 5468 if (vd->vdev_remove_wanted) { 5469 vd->vdev_remove_wanted = B_FALSE; 5470 vd->vdev_delayed_close = B_FALSE; 5471 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5472 5473 /* 5474 * We want to clear the stats, but we don't want to do a full 5475 * vdev_clear() as that will cause us to throw away 5476 * degraded/faulted state as well as attempt to reopen the 5477 * device, all of which is a waste. 5478 */ 5479 vd->vdev_stat.vs_read_errors = 0; 5480 vd->vdev_stat.vs_write_errors = 0; 5481 vd->vdev_stat.vs_checksum_errors = 0; 5482 5483 vdev_state_dirty(vd->vdev_top); 5484 } 5485 5486 for (int c = 0; c < vd->vdev_children; c++) 5487 spa_async_remove(spa, vd->vdev_child[c]); 5488} 5489 5490static void 5491spa_async_probe(spa_t *spa, vdev_t *vd) 5492{ 5493 if (vd->vdev_probe_wanted) { 5494 vd->vdev_probe_wanted = B_FALSE; 5495 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5496 } 5497 5498 for (int c = 0; c < vd->vdev_children; c++) 5499 spa_async_probe(spa, vd->vdev_child[c]); 5500} 5501 5502static void 5503spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5504{ 5505 sysevent_id_t eid; 5506 nvlist_t *attr; 5507 char *physpath; 5508 5509 if (!spa->spa_autoexpand) 5510 return; 5511 5512 for (int c = 0; c < vd->vdev_children; c++) { 5513 vdev_t *cvd = vd->vdev_child[c]; 5514 spa_async_autoexpand(spa, cvd); 5515 } 5516 5517 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5518 return; 5519 5520 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5521 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5522 5523 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5524 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5525 5526 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5527 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5528 5529 nvlist_free(attr); 5530 kmem_free(physpath, MAXPATHLEN); 5531} 5532 5533static void 5534spa_async_thread(void *arg) 5535{ 5536 spa_t *spa = arg; 5537 int tasks; 5538 5539 ASSERT(spa->spa_sync_on); 5540 5541 mutex_enter(&spa->spa_async_lock); 5542 tasks = spa->spa_async_tasks; 5543 spa->spa_async_tasks = 0; 5544 mutex_exit(&spa->spa_async_lock); 5545 5546 /* 5547 * See if the config needs to be updated. 5548 */ 5549 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5550 uint64_t old_space, new_space; 5551 5552 mutex_enter(&spa_namespace_lock); 5553 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5554 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5555 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5556 mutex_exit(&spa_namespace_lock); 5557 5558 /* 5559 * If the pool grew as a result of the config update, 5560 * then log an internal history event. 5561 */ 5562 if (new_space != old_space) { 5563 spa_history_log_internal(LOG_POOL_VDEV_ONLINE, 5564 spa, NULL, 5565 "pool '%s' size: %llu(+%llu)", 5566 spa_name(spa), new_space, new_space - old_space); 5567 } 5568 } 5569 5570 /* 5571 * See if any devices need to be marked REMOVED. 5572 */ 5573 if (tasks & SPA_ASYNC_REMOVE) { 5574 spa_vdev_state_enter(spa, SCL_NONE); 5575 spa_async_remove(spa, spa->spa_root_vdev); 5576 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5577 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5578 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5579 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5580 (void) spa_vdev_state_exit(spa, NULL, 0); 5581 } 5582 5583 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5584 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5585 spa_async_autoexpand(spa, spa->spa_root_vdev); 5586 spa_config_exit(spa, SCL_CONFIG, FTAG); 5587 } 5588 5589 /* 5590 * See if any devices need to be probed. 5591 */ 5592 if (tasks & SPA_ASYNC_PROBE) { 5593 spa_vdev_state_enter(spa, SCL_NONE); 5594 spa_async_probe(spa, spa->spa_root_vdev); 5595 (void) spa_vdev_state_exit(spa, NULL, 0); 5596 } 5597 5598 /* 5599 * If any devices are done replacing, detach them. 5600 */ 5601 if (tasks & SPA_ASYNC_RESILVER_DONE) 5602 spa_vdev_resilver_done(spa); 5603 5604 /* 5605 * Kick off a resilver. 5606 */ 5607 if (tasks & SPA_ASYNC_RESILVER) 5608 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5609 5610 /* 5611 * Let the world know that we're done. 5612 */ 5613 mutex_enter(&spa->spa_async_lock); 5614 spa->spa_async_thread = NULL; 5615 cv_broadcast(&spa->spa_async_cv); 5616 mutex_exit(&spa->spa_async_lock); 5617 thread_exit(); 5618} 5619 5620void 5621spa_async_suspend(spa_t *spa) 5622{ 5623 mutex_enter(&spa->spa_async_lock); 5624 spa->spa_async_suspended++; 5625 while (spa->spa_async_thread != NULL) 5626 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5627 mutex_exit(&spa->spa_async_lock); 5628} 5629 5630void 5631spa_async_resume(spa_t *spa) 5632{ 5633 mutex_enter(&spa->spa_async_lock); 5634 ASSERT(spa->spa_async_suspended != 0); 5635 spa->spa_async_suspended--; 5636 mutex_exit(&spa->spa_async_lock); 5637} 5638 5639static void 5640spa_async_dispatch(spa_t *spa) 5641{ 5642 mutex_enter(&spa->spa_async_lock); 5643 if (spa->spa_async_tasks && !spa->spa_async_suspended && 5644 spa->spa_async_thread == NULL && 5645 rootdir != NULL && !vn_is_readonly(rootdir)) 5646 spa->spa_async_thread = thread_create(NULL, 0, 5647 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5648 mutex_exit(&spa->spa_async_lock); 5649} 5650 5651void 5652spa_async_request(spa_t *spa, int task) 5653{ 5654 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5655 mutex_enter(&spa->spa_async_lock); 5656 spa->spa_async_tasks |= task; 5657 mutex_exit(&spa->spa_async_lock); 5658} 5659 5660/* 5661 * ========================================================================== 5662 * SPA syncing routines 5663 * ========================================================================== 5664 */ 5665 5666static int 5667bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5668{ 5669 bpobj_t *bpo = arg; 5670 bpobj_enqueue(bpo, bp, tx); 5671 return (0); 5672} 5673 5674static int 5675spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5676{ 5677 zio_t *zio = arg; 5678 5679 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5680 zio->io_flags)); 5681 return (0); 5682} 5683 5684static void 5685spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5686{ 5687 char *packed = NULL; 5688 size_t bufsize; 5689 size_t nvsize = 0; 5690 dmu_buf_t *db; 5691 5692 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5693 5694 /* 5695 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 5696 * information. This avoids the dbuf_will_dirty() path and 5697 * saves us a pre-read to get data we don't actually care about. 5698 */ 5699 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 5700 packed = kmem_alloc(bufsize, KM_SLEEP); 5701 5702 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5703 KM_SLEEP) == 0); 5704 bzero(packed + nvsize, bufsize - nvsize); 5705 5706 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 5707 5708 kmem_free(packed, bufsize); 5709 5710 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 5711 dmu_buf_will_dirty(db, tx); 5712 *(uint64_t *)db->db_data = nvsize; 5713 dmu_buf_rele(db, FTAG); 5714} 5715 5716static void 5717spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 5718 const char *config, const char *entry) 5719{ 5720 nvlist_t *nvroot; 5721 nvlist_t **list; 5722 int i; 5723 5724 if (!sav->sav_sync) 5725 return; 5726 5727 /* 5728 * Update the MOS nvlist describing the list of available devices. 5729 * spa_validate_aux() will have already made sure this nvlist is 5730 * valid and the vdevs are labeled appropriately. 5731 */ 5732 if (sav->sav_object == 0) { 5733 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 5734 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 5735 sizeof (uint64_t), tx); 5736 VERIFY(zap_update(spa->spa_meta_objset, 5737 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 5738 &sav->sav_object, tx) == 0); 5739 } 5740 5741 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5742 if (sav->sav_count == 0) { 5743 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 5744 } else { 5745 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 5746 for (i = 0; i < sav->sav_count; i++) 5747 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 5748 B_FALSE, VDEV_CONFIG_L2CACHE); 5749 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 5750 sav->sav_count) == 0); 5751 for (i = 0; i < sav->sav_count; i++) 5752 nvlist_free(list[i]); 5753 kmem_free(list, sav->sav_count * sizeof (void *)); 5754 } 5755 5756 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 5757 nvlist_free(nvroot); 5758 5759 sav->sav_sync = B_FALSE; 5760} 5761 5762static void 5763spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 5764{ 5765 nvlist_t *config; 5766 5767 if (list_is_empty(&spa->spa_config_dirty_list)) 5768 return; 5769 5770 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5771 5772 config = spa_config_generate(spa, spa->spa_root_vdev, 5773 dmu_tx_get_txg(tx), B_FALSE); 5774 5775 spa_config_exit(spa, SCL_STATE, FTAG); 5776 5777 if (spa->spa_config_syncing) 5778 nvlist_free(spa->spa_config_syncing); 5779 spa->spa_config_syncing = config; 5780 5781 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 5782} 5783 5784static void 5785spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx) 5786{ 5787 spa_t *spa = arg1; 5788 uint64_t version = *(uint64_t *)arg2; 5789 5790 /* 5791 * Setting the version is special cased when first creating the pool. 5792 */ 5793 ASSERT(tx->tx_txg != TXG_INITIAL); 5794 5795 ASSERT(version <= SPA_VERSION); 5796 ASSERT(version >= spa_version(spa)); 5797 5798 spa->spa_uberblock.ub_version = version; 5799 vdev_config_dirty(spa->spa_root_vdev); 5800} 5801 5802/* 5803 * Set zpool properties. 5804 */ 5805static void 5806spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx) 5807{ 5808 spa_t *spa = arg1; 5809 objset_t *mos = spa->spa_meta_objset; 5810 nvlist_t *nvp = arg2; 5811 nvpair_t *elem = NULL; 5812 5813 mutex_enter(&spa->spa_props_lock); 5814 5815 while ((elem = nvlist_next_nvpair(nvp, elem))) { 5816 uint64_t intval; 5817 char *strval, *fname; 5818 zpool_prop_t prop; 5819 const char *propname; 5820 zprop_type_t proptype; 5821 zfeature_info_t *feature; 5822 5823 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 5824 case ZPROP_INVAL: 5825 /* 5826 * We checked this earlier in spa_prop_validate(). 5827 */ 5828 ASSERT(zpool_prop_feature(nvpair_name(elem))); 5829 5830 fname = strchr(nvpair_name(elem), '@') + 1; 5831 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature)); 5832 5833 spa_feature_enable(spa, feature, tx); 5834 break; 5835 5836 case ZPOOL_PROP_VERSION: 5837 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 5838 /* 5839 * The version is synced seperatly before other 5840 * properties and should be correct by now. 5841 */ 5842 ASSERT3U(spa_version(spa), >=, intval); 5843 break; 5844 5845 case ZPOOL_PROP_ALTROOT: 5846 /* 5847 * 'altroot' is a non-persistent property. It should 5848 * have been set temporarily at creation or import time. 5849 */ 5850 ASSERT(spa->spa_root != NULL); 5851 break; 5852 5853 case ZPOOL_PROP_READONLY: 5854 case ZPOOL_PROP_CACHEFILE: 5855 /* 5856 * 'readonly' and 'cachefile' are also non-persisitent 5857 * properties. 5858 */ 5859 break; 5860 case ZPOOL_PROP_COMMENT: 5861 VERIFY(nvpair_value_string(elem, &strval) == 0); 5862 if (spa->spa_comment != NULL) 5863 spa_strfree(spa->spa_comment); 5864 spa->spa_comment = spa_strdup(strval); 5865 /* 5866 * We need to dirty the configuration on all the vdevs 5867 * so that their labels get updated. It's unnecessary 5868 * to do this for pool creation since the vdev's 5869 * configuratoin has already been dirtied. 5870 */ 5871 if (tx->tx_txg != TXG_INITIAL) 5872 vdev_config_dirty(spa->spa_root_vdev); 5873 break; 5874 default: 5875 /* 5876 * Set pool property values in the poolprops mos object. 5877 */ 5878 if (spa->spa_pool_props_object == 0) { 5879 spa->spa_pool_props_object = 5880 zap_create_link(mos, DMU_OT_POOL_PROPS, 5881 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 5882 tx); 5883 } 5884 5885 /* normalize the property name */ 5886 propname = zpool_prop_to_name(prop); 5887 proptype = zpool_prop_get_type(prop); 5888 5889 if (nvpair_type(elem) == DATA_TYPE_STRING) { 5890 ASSERT(proptype == PROP_TYPE_STRING); 5891 VERIFY(nvpair_value_string(elem, &strval) == 0); 5892 VERIFY(zap_update(mos, 5893 spa->spa_pool_props_object, propname, 5894 1, strlen(strval) + 1, strval, tx) == 0); 5895 5896 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 5897 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 5898 5899 if (proptype == PROP_TYPE_INDEX) { 5900 const char *unused; 5901 VERIFY(zpool_prop_index_to_string( 5902 prop, intval, &unused) == 0); 5903 } 5904 VERIFY(zap_update(mos, 5905 spa->spa_pool_props_object, propname, 5906 8, 1, &intval, tx) == 0); 5907 } else { 5908 ASSERT(0); /* not allowed */ 5909 } 5910 5911 switch (prop) { 5912 case ZPOOL_PROP_DELEGATION: 5913 spa->spa_delegation = intval; 5914 break; 5915 case ZPOOL_PROP_BOOTFS: 5916 spa->spa_bootfs = intval; 5917 break; 5918 case ZPOOL_PROP_FAILUREMODE: 5919 spa->spa_failmode = intval; 5920 break; 5921 case ZPOOL_PROP_AUTOEXPAND: 5922 spa->spa_autoexpand = intval; 5923 if (tx->tx_txg != TXG_INITIAL) 5924 spa_async_request(spa, 5925 SPA_ASYNC_AUTOEXPAND); 5926 break; 5927 case ZPOOL_PROP_DEDUPDITTO: 5928 spa->spa_dedup_ditto = intval; 5929 break; 5930 default: 5931 break; 5932 } 5933 } 5934 5935 /* log internal history if this is not a zpool create */ 5936 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY && 5937 tx->tx_txg != TXG_INITIAL) { 5938 spa_history_log_internal(LOG_POOL_PROPSET, 5939 spa, tx, "%s %lld %s", 5940 nvpair_name(elem), intval, spa_name(spa)); 5941 } 5942 } 5943 5944 mutex_exit(&spa->spa_props_lock); 5945} 5946 5947/* 5948 * Perform one-time upgrade on-disk changes. spa_version() does not 5949 * reflect the new version this txg, so there must be no changes this 5950 * txg to anything that the upgrade code depends on after it executes. 5951 * Therefore this must be called after dsl_pool_sync() does the sync 5952 * tasks. 5953 */ 5954static void 5955spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 5956{ 5957 dsl_pool_t *dp = spa->spa_dsl_pool; 5958 5959 ASSERT(spa->spa_sync_pass == 1); 5960 5961 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 5962 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 5963 dsl_pool_create_origin(dp, tx); 5964 5965 /* Keeping the origin open increases spa_minref */ 5966 spa->spa_minref += 3; 5967 } 5968 5969 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 5970 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 5971 dsl_pool_upgrade_clones(dp, tx); 5972 } 5973 5974 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 5975 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 5976 dsl_pool_upgrade_dir_clones(dp, tx); 5977 5978 /* Keeping the freedir open increases spa_minref */ 5979 spa->spa_minref += 3; 5980 } 5981 5982 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 5983 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 5984 spa_feature_create_zap_objects(spa, tx); 5985 } 5986} 5987 5988/* 5989 * Sync the specified transaction group. New blocks may be dirtied as 5990 * part of the process, so we iterate until it converges. 5991 */ 5992void 5993spa_sync(spa_t *spa, uint64_t txg) 5994{ 5995 dsl_pool_t *dp = spa->spa_dsl_pool; 5996 objset_t *mos = spa->spa_meta_objset; 5997 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj; 5998 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 5999 vdev_t *rvd = spa->spa_root_vdev; 6000 vdev_t *vd; 6001 dmu_tx_t *tx; 6002 int error; 6003 6004 VERIFY(spa_writeable(spa)); 6005 6006 /* 6007 * Lock out configuration changes. 6008 */ 6009 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6010 6011 spa->spa_syncing_txg = txg; 6012 spa->spa_sync_pass = 0; 6013 6014 /* 6015 * If there are any pending vdev state changes, convert them 6016 * into config changes that go out with this transaction group. 6017 */ 6018 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6019 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6020 /* 6021 * We need the write lock here because, for aux vdevs, 6022 * calling vdev_config_dirty() modifies sav_config. 6023 * This is ugly and will become unnecessary when we 6024 * eliminate the aux vdev wart by integrating all vdevs 6025 * into the root vdev tree. 6026 */ 6027 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6028 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6029 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6030 vdev_state_clean(vd); 6031 vdev_config_dirty(vd); 6032 } 6033 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6034 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6035 } 6036 spa_config_exit(spa, SCL_STATE, FTAG); 6037 6038 tx = dmu_tx_create_assigned(dp, txg); 6039 6040 /* 6041 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6042 * set spa_deflate if we have no raid-z vdevs. 6043 */ 6044 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6045 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6046 int i; 6047 6048 for (i = 0; i < rvd->vdev_children; i++) { 6049 vd = rvd->vdev_child[i]; 6050 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6051 break; 6052 } 6053 if (i == rvd->vdev_children) { 6054 spa->spa_deflate = TRUE; 6055 VERIFY(0 == zap_add(spa->spa_meta_objset, 6056 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6057 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6058 } 6059 } 6060 6061 /* 6062 * If anything has changed in this txg, or if someone is waiting 6063 * for this txg to sync (eg, spa_vdev_remove()), push the 6064 * deferred frees from the previous txg. If not, leave them 6065 * alone so that we don't generate work on an otherwise idle 6066 * system. 6067 */ 6068 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6069 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6070 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6071 ((dsl_scan_active(dp->dp_scan) || 6072 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6073 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6074 VERIFY3U(bpobj_iterate(defer_bpo, 6075 spa_free_sync_cb, zio, tx), ==, 0); 6076 VERIFY0(zio_wait(zio)); 6077 } 6078 6079 /* 6080 * Iterate to convergence. 6081 */ 6082 do { 6083 int pass = ++spa->spa_sync_pass; 6084 6085 spa_sync_config_object(spa, tx); 6086 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6087 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6088 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6089 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6090 spa_errlog_sync(spa, txg); 6091 dsl_pool_sync(dp, txg); 6092 6093 if (pass <= SYNC_PASS_DEFERRED_FREE) { 6094 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6095 bplist_iterate(free_bpl, spa_free_sync_cb, 6096 zio, tx); 6097 VERIFY(zio_wait(zio) == 0); 6098 } else { 6099 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6100 defer_bpo, tx); 6101 } 6102 6103 ddt_sync(spa, txg); 6104 dsl_scan_sync(dp, tx); 6105 6106 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6107 vdev_sync(vd, txg); 6108 6109 if (pass == 1) 6110 spa_sync_upgrades(spa, tx); 6111 6112 } while (dmu_objset_is_dirty(mos, txg)); 6113 6114 /* 6115 * Rewrite the vdev configuration (which includes the uberblock) 6116 * to commit the transaction group. 6117 * 6118 * If there are no dirty vdevs, we sync the uberblock to a few 6119 * random top-level vdevs that are known to be visible in the 6120 * config cache (see spa_vdev_add() for a complete description). 6121 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6122 */ 6123 for (;;) { 6124 /* 6125 * We hold SCL_STATE to prevent vdev open/close/etc. 6126 * while we're attempting to write the vdev labels. 6127 */ 6128 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6129 6130 if (list_is_empty(&spa->spa_config_dirty_list)) { 6131 vdev_t *svd[SPA_DVAS_PER_BP]; 6132 int svdcount = 0; 6133 int children = rvd->vdev_children; 6134 int c0 = spa_get_random(children); 6135 6136 for (int c = 0; c < children; c++) { 6137 vd = rvd->vdev_child[(c0 + c) % children]; 6138 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6139 continue; 6140 svd[svdcount++] = vd; 6141 if (svdcount == SPA_DVAS_PER_BP) 6142 break; 6143 } 6144 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6145 if (error != 0) 6146 error = vdev_config_sync(svd, svdcount, txg, 6147 B_TRUE); 6148 } else { 6149 error = vdev_config_sync(rvd->vdev_child, 6150 rvd->vdev_children, txg, B_FALSE); 6151 if (error != 0) 6152 error = vdev_config_sync(rvd->vdev_child, 6153 rvd->vdev_children, txg, B_TRUE); 6154 } 6155 6156 if (error == 0) 6157 spa->spa_last_synced_guid = rvd->vdev_guid; 6158 6159 spa_config_exit(spa, SCL_STATE, FTAG); 6160 6161 if (error == 0) 6162 break; 6163 zio_suspend(spa, NULL); 6164 zio_resume_wait(spa); 6165 } 6166 dmu_tx_commit(tx); 6167 6168 /* 6169 * Clear the dirty config list. 6170 */ 6171 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6172 vdev_config_clean(vd); 6173 6174 /* 6175 * Now that the new config has synced transactionally, 6176 * let it become visible to the config cache. 6177 */ 6178 if (spa->spa_config_syncing != NULL) { 6179 spa_config_set(spa, spa->spa_config_syncing); 6180 spa->spa_config_txg = txg; 6181 spa->spa_config_syncing = NULL; 6182 } 6183 6184 spa->spa_ubsync = spa->spa_uberblock; 6185 6186 dsl_pool_sync_done(dp, txg); 6187 6188 /* 6189 * Update usable space statistics. 6190 */ 6191 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6192 vdev_sync_done(vd, txg); 6193 6194 spa_update_dspace(spa); 6195 6196 /* 6197 * It had better be the case that we didn't dirty anything 6198 * since vdev_config_sync(). 6199 */ 6200 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6201 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6202 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6203 6204 spa->spa_sync_pass = 0; 6205 6206 spa_config_exit(spa, SCL_CONFIG, FTAG); 6207 6208 spa_handle_ignored_writes(spa); 6209 6210 /* 6211 * If any async tasks have been requested, kick them off. 6212 */ 6213 spa_async_dispatch(spa); 6214} 6215 6216/* 6217 * Sync all pools. We don't want to hold the namespace lock across these 6218 * operations, so we take a reference on the spa_t and drop the lock during the 6219 * sync. 6220 */ 6221void 6222spa_sync_allpools(void) 6223{ 6224 spa_t *spa = NULL; 6225 mutex_enter(&spa_namespace_lock); 6226 while ((spa = spa_next(spa)) != NULL) { 6227 if (spa_state(spa) != POOL_STATE_ACTIVE || 6228 !spa_writeable(spa) || spa_suspended(spa)) 6229 continue; 6230 spa_open_ref(spa, FTAG); 6231 mutex_exit(&spa_namespace_lock); 6232 txg_wait_synced(spa_get_dsl(spa), 0); 6233 mutex_enter(&spa_namespace_lock); 6234 spa_close(spa, FTAG); 6235 } 6236 mutex_exit(&spa_namespace_lock); 6237} 6238 6239/* 6240 * ========================================================================== 6241 * Miscellaneous routines 6242 * ========================================================================== 6243 */ 6244 6245/* 6246 * Remove all pools in the system. 6247 */ 6248void 6249spa_evict_all(void) 6250{ 6251 spa_t *spa; 6252 6253 /* 6254 * Remove all cached state. All pools should be closed now, 6255 * so every spa in the AVL tree should be unreferenced. 6256 */ 6257 mutex_enter(&spa_namespace_lock); 6258 while ((spa = spa_next(NULL)) != NULL) { 6259 /* 6260 * Stop async tasks. The async thread may need to detach 6261 * a device that's been replaced, which requires grabbing 6262 * spa_namespace_lock, so we must drop it here. 6263 */ 6264 spa_open_ref(spa, FTAG); 6265 mutex_exit(&spa_namespace_lock); 6266 spa_async_suspend(spa); 6267 mutex_enter(&spa_namespace_lock); 6268 spa_close(spa, FTAG); 6269 6270 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6271 spa_unload(spa); 6272 spa_deactivate(spa); 6273 } 6274 spa_remove(spa); 6275 } 6276 mutex_exit(&spa_namespace_lock); 6277} 6278 6279vdev_t * 6280spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6281{ 6282 vdev_t *vd; 6283 int i; 6284 6285 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6286 return (vd); 6287 6288 if (aux) { 6289 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6290 vd = spa->spa_l2cache.sav_vdevs[i]; 6291 if (vd->vdev_guid == guid) 6292 return (vd); 6293 } 6294 6295 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6296 vd = spa->spa_spares.sav_vdevs[i]; 6297 if (vd->vdev_guid == guid) 6298 return (vd); 6299 } 6300 } 6301 6302 return (NULL); 6303} 6304 6305void 6306spa_upgrade(spa_t *spa, uint64_t version) 6307{ 6308 ASSERT(spa_writeable(spa)); 6309 6310 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6311 6312 /* 6313 * This should only be called for a non-faulted pool, and since a 6314 * future version would result in an unopenable pool, this shouldn't be 6315 * possible. 6316 */ 6317 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION); 6318 ASSERT(version >= spa->spa_uberblock.ub_version); 6319 6320 spa->spa_uberblock.ub_version = version; 6321 vdev_config_dirty(spa->spa_root_vdev); 6322 6323 spa_config_exit(spa, SCL_ALL, FTAG); 6324 6325 txg_wait_synced(spa_get_dsl(spa), 0); 6326} 6327 6328boolean_t 6329spa_has_spare(spa_t *spa, uint64_t guid) 6330{ 6331 int i; 6332 uint64_t spareguid; 6333 spa_aux_vdev_t *sav = &spa->spa_spares; 6334 6335 for (i = 0; i < sav->sav_count; i++) 6336 if (sav->sav_vdevs[i]->vdev_guid == guid) 6337 return (B_TRUE); 6338 6339 for (i = 0; i < sav->sav_npending; i++) { 6340 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6341 &spareguid) == 0 && spareguid == guid) 6342 return (B_TRUE); 6343 } 6344 6345 return (B_FALSE); 6346} 6347 6348/* 6349 * Check if a pool has an active shared spare device. 6350 * Note: reference count of an active spare is 2, as a spare and as a replace 6351 */ 6352static boolean_t 6353spa_has_active_shared_spare(spa_t *spa) 6354{ 6355 int i, refcnt; 6356 uint64_t pool; 6357 spa_aux_vdev_t *sav = &spa->spa_spares; 6358 6359 for (i = 0; i < sav->sav_count; i++) { 6360 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6361 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6362 refcnt > 2) 6363 return (B_TRUE); 6364 } 6365 6366 return (B_FALSE); 6367} 6368 6369/* 6370 * Post a sysevent corresponding to the given event. The 'name' must be one of 6371 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6372 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6373 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6374 * or zdb as real changes. 6375 */ 6376void 6377spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6378{ 6379#ifdef _KERNEL 6380 sysevent_t *ev; 6381 sysevent_attr_list_t *attr = NULL; 6382 sysevent_value_t value; 6383 sysevent_id_t eid; 6384 6385 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6386 SE_SLEEP); 6387 6388 value.value_type = SE_DATA_TYPE_STRING; 6389 value.value.sv_string = spa_name(spa); 6390 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6391 goto done; 6392 6393 value.value_type = SE_DATA_TYPE_UINT64; 6394 value.value.sv_uint64 = spa_guid(spa); 6395 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6396 goto done; 6397 6398 if (vd) { 6399 value.value_type = SE_DATA_TYPE_UINT64; 6400 value.value.sv_uint64 = vd->vdev_guid; 6401 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6402 SE_SLEEP) != 0) 6403 goto done; 6404 6405 if (vd->vdev_path) { 6406 value.value_type = SE_DATA_TYPE_STRING; 6407 value.value.sv_string = vd->vdev_path; 6408 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6409 &value, SE_SLEEP) != 0) 6410 goto done; 6411 } 6412 } 6413 6414 if (sysevent_attach_attributes(ev, attr) != 0) 6415 goto done; 6416 attr = NULL; 6417 6418 (void) log_sysevent(ev, SE_SLEEP, &eid); 6419 6420done: 6421 if (attr) 6422 sysevent_free_attr(attr); 6423 sysevent_free(ev); 6424#endif 6425} 6426