spa.c revision 241286
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#include <sys/trim_map.h> 66 67#ifdef _KERNEL 68#include <sys/callb.h> 69#include <sys/cpupart.h> 70#include <sys/zone.h> 71#endif /* _KERNEL */ 72 73#include "zfs_prop.h" 74#include "zfs_comutil.h" 75 76/* Check hostid on import? */ 77static int check_hostid = 1; 78 79SYSCTL_DECL(_vfs_zfs); 80TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 81SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 82 "Check hostid on import?"); 83 84typedef enum zti_modes { 85 zti_mode_fixed, /* value is # of threads (min 1) */ 86 zti_mode_online_percent, /* value is % of online CPUs */ 87 zti_mode_batch, /* cpu-intensive; value is ignored */ 88 zti_mode_null, /* don't create a taskq */ 89 zti_nmodes 90} zti_modes_t; 91 92#define ZTI_FIX(n) { zti_mode_fixed, (n) } 93#define ZTI_PCT(n) { zti_mode_online_percent, (n) } 94#define ZTI_BATCH { zti_mode_batch, 0 } 95#define ZTI_NULL { zti_mode_null, 0 } 96 97#define ZTI_ONE ZTI_FIX(1) 98 99typedef struct zio_taskq_info { 100 enum zti_modes zti_mode; 101 uint_t zti_value; 102} zio_taskq_info_t; 103 104static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 105 "issue", "issue_high", "intr", "intr_high" 106}; 107 108/* 109 * Define the taskq threads for the following I/O types: 110 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL 111 */ 112const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 113 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 114 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 115 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, 116 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(8), ZTI_FIX(5) }, 117 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL }, 118 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 119 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, 120}; 121 122static dsl_syncfunc_t spa_sync_version; 123static dsl_syncfunc_t spa_sync_props; 124static dsl_checkfunc_t spa_change_guid_check; 125static dsl_syncfunc_t spa_change_guid_sync; 126static boolean_t spa_has_active_shared_spare(spa_t *spa); 127static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 128 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 129 char **ereport); 130static void spa_vdev_resilver_done(spa_t *spa); 131 132uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */ 133#ifdef PSRSET_BIND 134id_t zio_taskq_psrset_bind = PS_NONE; 135#endif 136#ifdef SYSDC 137boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 138#endif 139uint_t zio_taskq_basedc = 80; /* base duty cycle */ 140 141boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 142 143/* 144 * This (illegal) pool name is used when temporarily importing a spa_t in order 145 * to get the vdev stats associated with the imported devices. 146 */ 147#define TRYIMPORT_NAME "$import" 148 149/* 150 * ========================================================================== 151 * SPA properties routines 152 * ========================================================================== 153 */ 154 155/* 156 * Add a (source=src, propname=propval) list to an nvlist. 157 */ 158static void 159spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 160 uint64_t intval, zprop_source_t src) 161{ 162 const char *propname = zpool_prop_to_name(prop); 163 nvlist_t *propval; 164 165 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 166 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 167 168 if (strval != NULL) 169 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 170 else 171 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 172 173 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 174 nvlist_free(propval); 175} 176 177/* 178 * Get property values from the spa configuration. 179 */ 180static void 181spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 182{ 183 vdev_t *rvd = spa->spa_root_vdev; 184 dsl_pool_t *pool = spa->spa_dsl_pool; 185 uint64_t size; 186 uint64_t alloc; 187 uint64_t space; 188 uint64_t cap, version; 189 zprop_source_t src = ZPROP_SRC_NONE; 190 spa_config_dirent_t *dp; 191 192 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 193 194 if (rvd != NULL) { 195 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 196 size = metaslab_class_get_space(spa_normal_class(spa)); 197 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 198 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 199 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 200 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 201 size - alloc, src); 202 203 space = 0; 204 for (int c = 0; c < rvd->vdev_children; c++) { 205 vdev_t *tvd = rvd->vdev_child[c]; 206 space += tvd->vdev_max_asize - tvd->vdev_asize; 207 } 208 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space, 209 src); 210 211 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 212 (spa_mode(spa) == FREAD), src); 213 214 cap = (size == 0) ? 0 : (alloc * 100 / size); 215 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 216 217 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 218 ddt_get_pool_dedup_ratio(spa), src); 219 220 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 221 rvd->vdev_state, src); 222 223 version = spa_version(spa); 224 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 225 src = ZPROP_SRC_DEFAULT; 226 else 227 src = ZPROP_SRC_LOCAL; 228 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 229 } 230 231 if (pool != NULL) { 232 dsl_dir_t *freedir = pool->dp_free_dir; 233 234 /* 235 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 236 * when opening pools before this version freedir will be NULL. 237 */ 238 if (freedir != NULL) { 239 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 240 freedir->dd_phys->dd_used_bytes, src); 241 } else { 242 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 243 NULL, 0, src); 244 } 245 } 246 247 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 248 249 if (spa->spa_comment != NULL) { 250 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 251 0, ZPROP_SRC_LOCAL); 252 } 253 254 if (spa->spa_root != NULL) 255 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 256 0, ZPROP_SRC_LOCAL); 257 258 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 259 if (dp->scd_path == NULL) { 260 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 261 "none", 0, ZPROP_SRC_LOCAL); 262 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 263 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 264 dp->scd_path, 0, ZPROP_SRC_LOCAL); 265 } 266 } 267} 268 269/* 270 * Get zpool property values. 271 */ 272int 273spa_prop_get(spa_t *spa, nvlist_t **nvp) 274{ 275 objset_t *mos = spa->spa_meta_objset; 276 zap_cursor_t zc; 277 zap_attribute_t za; 278 int err; 279 280 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 281 282 mutex_enter(&spa->spa_props_lock); 283 284 /* 285 * Get properties from the spa config. 286 */ 287 spa_prop_get_config(spa, nvp); 288 289 /* If no pool property object, no more prop to get. */ 290 if (mos == NULL || spa->spa_pool_props_object == 0) { 291 mutex_exit(&spa->spa_props_lock); 292 return (0); 293 } 294 295 /* 296 * Get properties from the MOS pool property object. 297 */ 298 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 299 (err = zap_cursor_retrieve(&zc, &za)) == 0; 300 zap_cursor_advance(&zc)) { 301 uint64_t intval = 0; 302 char *strval = NULL; 303 zprop_source_t src = ZPROP_SRC_DEFAULT; 304 zpool_prop_t prop; 305 306 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 307 continue; 308 309 switch (za.za_integer_length) { 310 case 8: 311 /* integer property */ 312 if (za.za_first_integer != 313 zpool_prop_default_numeric(prop)) 314 src = ZPROP_SRC_LOCAL; 315 316 if (prop == ZPOOL_PROP_BOOTFS) { 317 dsl_pool_t *dp; 318 dsl_dataset_t *ds = NULL; 319 320 dp = spa_get_dsl(spa); 321 rw_enter(&dp->dp_config_rwlock, RW_READER); 322 if (err = dsl_dataset_hold_obj(dp, 323 za.za_first_integer, FTAG, &ds)) { 324 rw_exit(&dp->dp_config_rwlock); 325 break; 326 } 327 328 strval = kmem_alloc( 329 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 330 KM_SLEEP); 331 dsl_dataset_name(ds, strval); 332 dsl_dataset_rele(ds, FTAG); 333 rw_exit(&dp->dp_config_rwlock); 334 } else { 335 strval = NULL; 336 intval = za.za_first_integer; 337 } 338 339 spa_prop_add_list(*nvp, prop, strval, intval, src); 340 341 if (strval != NULL) 342 kmem_free(strval, 343 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 344 345 break; 346 347 case 1: 348 /* string property */ 349 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 350 err = zap_lookup(mos, spa->spa_pool_props_object, 351 za.za_name, 1, za.za_num_integers, strval); 352 if (err) { 353 kmem_free(strval, za.za_num_integers); 354 break; 355 } 356 spa_prop_add_list(*nvp, prop, strval, 0, src); 357 kmem_free(strval, za.za_num_integers); 358 break; 359 360 default: 361 break; 362 } 363 } 364 zap_cursor_fini(&zc); 365 mutex_exit(&spa->spa_props_lock); 366out: 367 if (err && err != ENOENT) { 368 nvlist_free(*nvp); 369 *nvp = NULL; 370 return (err); 371 } 372 373 return (0); 374} 375 376/* 377 * Validate the given pool properties nvlist and modify the list 378 * for the property values to be set. 379 */ 380static int 381spa_prop_validate(spa_t *spa, nvlist_t *props) 382{ 383 nvpair_t *elem; 384 int error = 0, reset_bootfs = 0; 385 uint64_t objnum; 386 boolean_t has_feature = B_FALSE; 387 388 elem = NULL; 389 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 390 uint64_t intval; 391 char *strval, *slash, *check, *fname; 392 const char *propname = nvpair_name(elem); 393 zpool_prop_t prop = zpool_name_to_prop(propname); 394 395 switch (prop) { 396 case ZPROP_INVAL: 397 if (!zpool_prop_feature(propname)) { 398 error = EINVAL; 399 break; 400 } 401 402 /* 403 * Sanitize the input. 404 */ 405 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 406 error = EINVAL; 407 break; 408 } 409 410 if (nvpair_value_uint64(elem, &intval) != 0) { 411 error = EINVAL; 412 break; 413 } 414 415 if (intval != 0) { 416 error = EINVAL; 417 break; 418 } 419 420 fname = strchr(propname, '@') + 1; 421 if (zfeature_lookup_name(fname, NULL) != 0) { 422 error = EINVAL; 423 break; 424 } 425 426 has_feature = B_TRUE; 427 break; 428 429 case ZPOOL_PROP_VERSION: 430 error = nvpair_value_uint64(elem, &intval); 431 if (!error && 432 (intval < spa_version(spa) || 433 intval > SPA_VERSION_BEFORE_FEATURES || 434 has_feature)) 435 error = EINVAL; 436 break; 437 438 case ZPOOL_PROP_DELEGATION: 439 case ZPOOL_PROP_AUTOREPLACE: 440 case ZPOOL_PROP_LISTSNAPS: 441 case ZPOOL_PROP_AUTOEXPAND: 442 error = nvpair_value_uint64(elem, &intval); 443 if (!error && intval > 1) 444 error = EINVAL; 445 break; 446 447 case ZPOOL_PROP_BOOTFS: 448 /* 449 * If the pool version is less than SPA_VERSION_BOOTFS, 450 * or the pool is still being created (version == 0), 451 * the bootfs property cannot be set. 452 */ 453 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 454 error = ENOTSUP; 455 break; 456 } 457 458 /* 459 * Make sure the vdev config is bootable 460 */ 461 if (!vdev_is_bootable(spa->spa_root_vdev)) { 462 error = ENOTSUP; 463 break; 464 } 465 466 reset_bootfs = 1; 467 468 error = nvpair_value_string(elem, &strval); 469 470 if (!error) { 471 objset_t *os; 472 uint64_t compress; 473 474 if (strval == NULL || strval[0] == '\0') { 475 objnum = zpool_prop_default_numeric( 476 ZPOOL_PROP_BOOTFS); 477 break; 478 } 479 480 if (error = dmu_objset_hold(strval, FTAG, &os)) 481 break; 482 483 /* Must be ZPL and not gzip compressed. */ 484 485 if (dmu_objset_type(os) != DMU_OST_ZFS) { 486 error = ENOTSUP; 487 } else if ((error = dsl_prop_get_integer(strval, 488 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 489 &compress, NULL)) == 0 && 490 !BOOTFS_COMPRESS_VALID(compress)) { 491 error = ENOTSUP; 492 } else { 493 objnum = dmu_objset_id(os); 494 } 495 dmu_objset_rele(os, FTAG); 496 } 497 break; 498 499 case ZPOOL_PROP_FAILUREMODE: 500 error = nvpair_value_uint64(elem, &intval); 501 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 502 intval > ZIO_FAILURE_MODE_PANIC)) 503 error = EINVAL; 504 505 /* 506 * This is a special case which only occurs when 507 * the pool has completely failed. This allows 508 * the user to change the in-core failmode property 509 * without syncing it out to disk (I/Os might 510 * currently be blocked). We do this by returning 511 * EIO to the caller (spa_prop_set) to trick it 512 * into thinking we encountered a property validation 513 * error. 514 */ 515 if (!error && spa_suspended(spa)) { 516 spa->spa_failmode = intval; 517 error = EIO; 518 } 519 break; 520 521 case ZPOOL_PROP_CACHEFILE: 522 if ((error = nvpair_value_string(elem, &strval)) != 0) 523 break; 524 525 if (strval[0] == '\0') 526 break; 527 528 if (strcmp(strval, "none") == 0) 529 break; 530 531 if (strval[0] != '/') { 532 error = EINVAL; 533 break; 534 } 535 536 slash = strrchr(strval, '/'); 537 ASSERT(slash != NULL); 538 539 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 540 strcmp(slash, "/..") == 0) 541 error = EINVAL; 542 break; 543 544 case ZPOOL_PROP_COMMENT: 545 if ((error = nvpair_value_string(elem, &strval)) != 0) 546 break; 547 for (check = strval; *check != '\0'; check++) { 548 /* 549 * The kernel doesn't have an easy isprint() 550 * check. For this kernel check, we merely 551 * check ASCII apart from DEL. Fix this if 552 * there is an easy-to-use kernel isprint(). 553 */ 554 if (*check >= 0x7f) { 555 error = EINVAL; 556 break; 557 } 558 check++; 559 } 560 if (strlen(strval) > ZPROP_MAX_COMMENT) 561 error = E2BIG; 562 break; 563 564 case ZPOOL_PROP_DEDUPDITTO: 565 if (spa_version(spa) < SPA_VERSION_DEDUP) 566 error = ENOTSUP; 567 else 568 error = nvpair_value_uint64(elem, &intval); 569 if (error == 0 && 570 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 571 error = EINVAL; 572 break; 573 } 574 575 if (error) 576 break; 577 } 578 579 if (!error && reset_bootfs) { 580 error = nvlist_remove(props, 581 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 582 583 if (!error) { 584 error = nvlist_add_uint64(props, 585 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 586 } 587 } 588 589 return (error); 590} 591 592void 593spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 594{ 595 char *cachefile; 596 spa_config_dirent_t *dp; 597 598 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 599 &cachefile) != 0) 600 return; 601 602 dp = kmem_alloc(sizeof (spa_config_dirent_t), 603 KM_SLEEP); 604 605 if (cachefile[0] == '\0') 606 dp->scd_path = spa_strdup(spa_config_path); 607 else if (strcmp(cachefile, "none") == 0) 608 dp->scd_path = NULL; 609 else 610 dp->scd_path = spa_strdup(cachefile); 611 612 list_insert_head(&spa->spa_config_list, dp); 613 if (need_sync) 614 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 615} 616 617int 618spa_prop_set(spa_t *spa, nvlist_t *nvp) 619{ 620 int error; 621 nvpair_t *elem = NULL; 622 boolean_t need_sync = B_FALSE; 623 624 if ((error = spa_prop_validate(spa, nvp)) != 0) 625 return (error); 626 627 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 628 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 629 630 if (prop == ZPOOL_PROP_CACHEFILE || 631 prop == ZPOOL_PROP_ALTROOT || 632 prop == ZPOOL_PROP_READONLY) 633 continue; 634 635 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 636 uint64_t ver; 637 638 if (prop == ZPOOL_PROP_VERSION) { 639 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 640 } else { 641 ASSERT(zpool_prop_feature(nvpair_name(elem))); 642 ver = SPA_VERSION_FEATURES; 643 need_sync = B_TRUE; 644 } 645 646 /* Save time if the version is already set. */ 647 if (ver == spa_version(spa)) 648 continue; 649 650 /* 651 * In addition to the pool directory object, we might 652 * create the pool properties object, the features for 653 * read object, the features for write object, or the 654 * feature descriptions object. 655 */ 656 error = dsl_sync_task_do(spa_get_dsl(spa), NULL, 657 spa_sync_version, spa, &ver, 6); 658 if (error) 659 return (error); 660 continue; 661 } 662 663 need_sync = B_TRUE; 664 break; 665 } 666 667 if (need_sync) { 668 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props, 669 spa, nvp, 6)); 670 } 671 672 return (0); 673} 674 675/* 676 * If the bootfs property value is dsobj, clear it. 677 */ 678void 679spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 680{ 681 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 682 VERIFY(zap_remove(spa->spa_meta_objset, 683 spa->spa_pool_props_object, 684 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 685 spa->spa_bootfs = 0; 686 } 687} 688 689/*ARGSUSED*/ 690static int 691spa_change_guid_check(void *arg1, void *arg2, dmu_tx_t *tx) 692{ 693 spa_t *spa = arg1; 694 uint64_t *newguid = arg2; 695 vdev_t *rvd = spa->spa_root_vdev; 696 uint64_t vdev_state; 697 698 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 699 vdev_state = rvd->vdev_state; 700 spa_config_exit(spa, SCL_STATE, FTAG); 701 702 if (vdev_state != VDEV_STATE_HEALTHY) 703 return (ENXIO); 704 705 ASSERT3U(spa_guid(spa), !=, *newguid); 706 707 return (0); 708} 709 710static void 711spa_change_guid_sync(void *arg1, void *arg2, dmu_tx_t *tx) 712{ 713 spa_t *spa = arg1; 714 uint64_t *newguid = arg2; 715 uint64_t oldguid; 716 vdev_t *rvd = spa->spa_root_vdev; 717 718 oldguid = spa_guid(spa); 719 720 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 721 rvd->vdev_guid = *newguid; 722 rvd->vdev_guid_sum += (*newguid - oldguid); 723 vdev_config_dirty(rvd); 724 spa_config_exit(spa, SCL_STATE, FTAG); 725 726#ifdef __FreeBSD__ 727 /* 728 * TODO: until recent illumos logging changes are merged 729 * log reguid as pool property change 730 */ 731 spa_history_log_internal(LOG_POOL_PROPSET, spa, tx, 732 "guid change old=%llu new=%llu", oldguid, *newguid); 733#else 734 spa_history_log_internal(spa, "guid change", tx, "old=%lld new=%lld", 735 oldguid, *newguid); 736#endif 737} 738 739/* 740 * Change the GUID for the pool. This is done so that we can later 741 * re-import a pool built from a clone of our own vdevs. We will modify 742 * the root vdev's guid, our own pool guid, and then mark all of our 743 * vdevs dirty. Note that we must make sure that all our vdevs are 744 * online when we do this, or else any vdevs that weren't present 745 * would be orphaned from our pool. We are also going to issue a 746 * sysevent to update any watchers. 747 */ 748int 749spa_change_guid(spa_t *spa) 750{ 751 int error; 752 uint64_t guid; 753 754 mutex_enter(&spa_namespace_lock); 755 guid = spa_generate_guid(NULL); 756 757 error = dsl_sync_task_do(spa_get_dsl(spa), spa_change_guid_check, 758 spa_change_guid_sync, spa, &guid, 5); 759 760 if (error == 0) { 761 spa_config_sync(spa, B_FALSE, B_TRUE); 762 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 763 } 764 765 mutex_exit(&spa_namespace_lock); 766 767 return (error); 768} 769 770/* 771 * ========================================================================== 772 * SPA state manipulation (open/create/destroy/import/export) 773 * ========================================================================== 774 */ 775 776static int 777spa_error_entry_compare(const void *a, const void *b) 778{ 779 spa_error_entry_t *sa = (spa_error_entry_t *)a; 780 spa_error_entry_t *sb = (spa_error_entry_t *)b; 781 int ret; 782 783 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 784 sizeof (zbookmark_t)); 785 786 if (ret < 0) 787 return (-1); 788 else if (ret > 0) 789 return (1); 790 else 791 return (0); 792} 793 794/* 795 * Utility function which retrieves copies of the current logs and 796 * re-initializes them in the process. 797 */ 798void 799spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 800{ 801 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 802 803 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 804 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 805 806 avl_create(&spa->spa_errlist_scrub, 807 spa_error_entry_compare, sizeof (spa_error_entry_t), 808 offsetof(spa_error_entry_t, se_avl)); 809 avl_create(&spa->spa_errlist_last, 810 spa_error_entry_compare, sizeof (spa_error_entry_t), 811 offsetof(spa_error_entry_t, se_avl)); 812} 813 814static taskq_t * 815spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode, 816 uint_t value) 817{ 818 uint_t flags = TASKQ_PREPOPULATE; 819 boolean_t batch = B_FALSE; 820 821 switch (mode) { 822 case zti_mode_null: 823 return (NULL); /* no taskq needed */ 824 825 case zti_mode_fixed: 826 ASSERT3U(value, >=, 1); 827 value = MAX(value, 1); 828 break; 829 830 case zti_mode_batch: 831 batch = B_TRUE; 832 flags |= TASKQ_THREADS_CPU_PCT; 833 value = zio_taskq_batch_pct; 834 break; 835 836 case zti_mode_online_percent: 837 flags |= TASKQ_THREADS_CPU_PCT; 838 break; 839 840 default: 841 panic("unrecognized mode for %s taskq (%u:%u) in " 842 "spa_activate()", 843 name, mode, value); 844 break; 845 } 846 847#ifdef SYSDC 848 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 849 if (batch) 850 flags |= TASKQ_DC_BATCH; 851 852 return (taskq_create_sysdc(name, value, 50, INT_MAX, 853 spa->spa_proc, zio_taskq_basedc, flags)); 854 } 855#endif 856 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX, 857 spa->spa_proc, flags)); 858} 859 860static void 861spa_create_zio_taskqs(spa_t *spa) 862{ 863 for (int t = 0; t < ZIO_TYPES; t++) { 864 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 865 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 866 enum zti_modes mode = ztip->zti_mode; 867 uint_t value = ztip->zti_value; 868 char name[32]; 869 870 (void) snprintf(name, sizeof (name), 871 "%s_%s", zio_type_name[t], zio_taskq_types[q]); 872 873 spa->spa_zio_taskq[t][q] = 874 spa_taskq_create(spa, name, mode, value); 875 } 876 } 877} 878 879#ifdef _KERNEL 880#ifdef SPA_PROCESS 881static void 882spa_thread(void *arg) 883{ 884 callb_cpr_t cprinfo; 885 886 spa_t *spa = arg; 887 user_t *pu = PTOU(curproc); 888 889 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 890 spa->spa_name); 891 892 ASSERT(curproc != &p0); 893 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 894 "zpool-%s", spa->spa_name); 895 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 896 897#ifdef PSRSET_BIND 898 /* bind this thread to the requested psrset */ 899 if (zio_taskq_psrset_bind != PS_NONE) { 900 pool_lock(); 901 mutex_enter(&cpu_lock); 902 mutex_enter(&pidlock); 903 mutex_enter(&curproc->p_lock); 904 905 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 906 0, NULL, NULL) == 0) { 907 curthread->t_bind_pset = zio_taskq_psrset_bind; 908 } else { 909 cmn_err(CE_WARN, 910 "Couldn't bind process for zfs pool \"%s\" to " 911 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 912 } 913 914 mutex_exit(&curproc->p_lock); 915 mutex_exit(&pidlock); 916 mutex_exit(&cpu_lock); 917 pool_unlock(); 918 } 919#endif 920 921#ifdef SYSDC 922 if (zio_taskq_sysdc) { 923 sysdc_thread_enter(curthread, 100, 0); 924 } 925#endif 926 927 spa->spa_proc = curproc; 928 spa->spa_did = curthread->t_did; 929 930 spa_create_zio_taskqs(spa); 931 932 mutex_enter(&spa->spa_proc_lock); 933 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 934 935 spa->spa_proc_state = SPA_PROC_ACTIVE; 936 cv_broadcast(&spa->spa_proc_cv); 937 938 CALLB_CPR_SAFE_BEGIN(&cprinfo); 939 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 940 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 941 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 942 943 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 944 spa->spa_proc_state = SPA_PROC_GONE; 945 spa->spa_proc = &p0; 946 cv_broadcast(&spa->spa_proc_cv); 947 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 948 949 mutex_enter(&curproc->p_lock); 950 lwp_exit(); 951} 952#endif /* SPA_PROCESS */ 953#endif 954 955/* 956 * Activate an uninitialized pool. 957 */ 958static void 959spa_activate(spa_t *spa, int mode) 960{ 961 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 962 963 spa->spa_state = POOL_STATE_ACTIVE; 964 spa->spa_mode = mode; 965 966 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 967 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 968 969 /* Try to create a covering process */ 970 mutex_enter(&spa->spa_proc_lock); 971 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 972 ASSERT(spa->spa_proc == &p0); 973 spa->spa_did = 0; 974 975#ifdef SPA_PROCESS 976 /* Only create a process if we're going to be around a while. */ 977 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 978 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 979 NULL, 0) == 0) { 980 spa->spa_proc_state = SPA_PROC_CREATED; 981 while (spa->spa_proc_state == SPA_PROC_CREATED) { 982 cv_wait(&spa->spa_proc_cv, 983 &spa->spa_proc_lock); 984 } 985 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 986 ASSERT(spa->spa_proc != &p0); 987 ASSERT(spa->spa_did != 0); 988 } else { 989#ifdef _KERNEL 990 cmn_err(CE_WARN, 991 "Couldn't create process for zfs pool \"%s\"\n", 992 spa->spa_name); 993#endif 994 } 995 } 996#endif /* SPA_PROCESS */ 997 mutex_exit(&spa->spa_proc_lock); 998 999 /* If we didn't create a process, we need to create our taskqs. */ 1000 ASSERT(spa->spa_proc == &p0); 1001 if (spa->spa_proc == &p0) { 1002 spa_create_zio_taskqs(spa); 1003 } 1004 1005 /* 1006 * Start TRIM thread. 1007 */ 1008 trim_thread_create(spa); 1009 1010 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1011 offsetof(vdev_t, vdev_config_dirty_node)); 1012 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1013 offsetof(vdev_t, vdev_state_dirty_node)); 1014 1015 txg_list_create(&spa->spa_vdev_txg_list, 1016 offsetof(struct vdev, vdev_txg_node)); 1017 1018 avl_create(&spa->spa_errlist_scrub, 1019 spa_error_entry_compare, sizeof (spa_error_entry_t), 1020 offsetof(spa_error_entry_t, se_avl)); 1021 avl_create(&spa->spa_errlist_last, 1022 spa_error_entry_compare, sizeof (spa_error_entry_t), 1023 offsetof(spa_error_entry_t, se_avl)); 1024} 1025 1026/* 1027 * Opposite of spa_activate(). 1028 */ 1029static void 1030spa_deactivate(spa_t *spa) 1031{ 1032 ASSERT(spa->spa_sync_on == B_FALSE); 1033 ASSERT(spa->spa_dsl_pool == NULL); 1034 ASSERT(spa->spa_root_vdev == NULL); 1035 ASSERT(spa->spa_async_zio_root == NULL); 1036 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1037 1038 /* 1039 * Stop TRIM thread in case spa_unload() wasn't called directly 1040 * before spa_deactivate(). 1041 */ 1042 trim_thread_destroy(spa); 1043 1044 txg_list_destroy(&spa->spa_vdev_txg_list); 1045 1046 list_destroy(&spa->spa_config_dirty_list); 1047 list_destroy(&spa->spa_state_dirty_list); 1048 1049 for (int t = 0; t < ZIO_TYPES; t++) { 1050 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1051 if (spa->spa_zio_taskq[t][q] != NULL) 1052 taskq_destroy(spa->spa_zio_taskq[t][q]); 1053 spa->spa_zio_taskq[t][q] = NULL; 1054 } 1055 } 1056 1057 metaslab_class_destroy(spa->spa_normal_class); 1058 spa->spa_normal_class = NULL; 1059 1060 metaslab_class_destroy(spa->spa_log_class); 1061 spa->spa_log_class = NULL; 1062 1063 /* 1064 * If this was part of an import or the open otherwise failed, we may 1065 * still have errors left in the queues. Empty them just in case. 1066 */ 1067 spa_errlog_drain(spa); 1068 1069 avl_destroy(&spa->spa_errlist_scrub); 1070 avl_destroy(&spa->spa_errlist_last); 1071 1072 spa->spa_state = POOL_STATE_UNINITIALIZED; 1073 1074 mutex_enter(&spa->spa_proc_lock); 1075 if (spa->spa_proc_state != SPA_PROC_NONE) { 1076 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1077 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1078 cv_broadcast(&spa->spa_proc_cv); 1079 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1080 ASSERT(spa->spa_proc != &p0); 1081 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1082 } 1083 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1084 spa->spa_proc_state = SPA_PROC_NONE; 1085 } 1086 ASSERT(spa->spa_proc == &p0); 1087 mutex_exit(&spa->spa_proc_lock); 1088 1089#ifdef SPA_PROCESS 1090 /* 1091 * We want to make sure spa_thread() has actually exited the ZFS 1092 * module, so that the module can't be unloaded out from underneath 1093 * it. 1094 */ 1095 if (spa->spa_did != 0) { 1096 thread_join(spa->spa_did); 1097 spa->spa_did = 0; 1098 } 1099#endif /* SPA_PROCESS */ 1100} 1101 1102/* 1103 * Verify a pool configuration, and construct the vdev tree appropriately. This 1104 * will create all the necessary vdevs in the appropriate layout, with each vdev 1105 * in the CLOSED state. This will prep the pool before open/creation/import. 1106 * All vdev validation is done by the vdev_alloc() routine. 1107 */ 1108static int 1109spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1110 uint_t id, int atype) 1111{ 1112 nvlist_t **child; 1113 uint_t children; 1114 int error; 1115 1116 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1117 return (error); 1118 1119 if ((*vdp)->vdev_ops->vdev_op_leaf) 1120 return (0); 1121 1122 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1123 &child, &children); 1124 1125 if (error == ENOENT) 1126 return (0); 1127 1128 if (error) { 1129 vdev_free(*vdp); 1130 *vdp = NULL; 1131 return (EINVAL); 1132 } 1133 1134 for (int c = 0; c < children; c++) { 1135 vdev_t *vd; 1136 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1137 atype)) != 0) { 1138 vdev_free(*vdp); 1139 *vdp = NULL; 1140 return (error); 1141 } 1142 } 1143 1144 ASSERT(*vdp != NULL); 1145 1146 return (0); 1147} 1148 1149/* 1150 * Opposite of spa_load(). 1151 */ 1152static void 1153spa_unload(spa_t *spa) 1154{ 1155 int i; 1156 1157 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1158 1159 /* 1160 * Stop TRIM thread. 1161 */ 1162 trim_thread_destroy(spa); 1163 1164 /* 1165 * Stop async tasks. 1166 */ 1167 spa_async_suspend(spa); 1168 1169 /* 1170 * Stop syncing. 1171 */ 1172 if (spa->spa_sync_on) { 1173 txg_sync_stop(spa->spa_dsl_pool); 1174 spa->spa_sync_on = B_FALSE; 1175 } 1176 1177 /* 1178 * Wait for any outstanding async I/O to complete. 1179 */ 1180 if (spa->spa_async_zio_root != NULL) { 1181 (void) zio_wait(spa->spa_async_zio_root); 1182 spa->spa_async_zio_root = NULL; 1183 } 1184 1185 bpobj_close(&spa->spa_deferred_bpobj); 1186 1187 /* 1188 * Close the dsl pool. 1189 */ 1190 if (spa->spa_dsl_pool) { 1191 dsl_pool_close(spa->spa_dsl_pool); 1192 spa->spa_dsl_pool = NULL; 1193 spa->spa_meta_objset = NULL; 1194 } 1195 1196 ddt_unload(spa); 1197 1198 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1199 1200 /* 1201 * Drop and purge level 2 cache 1202 */ 1203 spa_l2cache_drop(spa); 1204 1205 /* 1206 * Close all vdevs. 1207 */ 1208 if (spa->spa_root_vdev) 1209 vdev_free(spa->spa_root_vdev); 1210 ASSERT(spa->spa_root_vdev == NULL); 1211 1212 for (i = 0; i < spa->spa_spares.sav_count; i++) 1213 vdev_free(spa->spa_spares.sav_vdevs[i]); 1214 if (spa->spa_spares.sav_vdevs) { 1215 kmem_free(spa->spa_spares.sav_vdevs, 1216 spa->spa_spares.sav_count * sizeof (void *)); 1217 spa->spa_spares.sav_vdevs = NULL; 1218 } 1219 if (spa->spa_spares.sav_config) { 1220 nvlist_free(spa->spa_spares.sav_config); 1221 spa->spa_spares.sav_config = NULL; 1222 } 1223 spa->spa_spares.sav_count = 0; 1224 1225 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1226 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1227 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1228 } 1229 if (spa->spa_l2cache.sav_vdevs) { 1230 kmem_free(spa->spa_l2cache.sav_vdevs, 1231 spa->spa_l2cache.sav_count * sizeof (void *)); 1232 spa->spa_l2cache.sav_vdevs = NULL; 1233 } 1234 if (spa->spa_l2cache.sav_config) { 1235 nvlist_free(spa->spa_l2cache.sav_config); 1236 spa->spa_l2cache.sav_config = NULL; 1237 } 1238 spa->spa_l2cache.sav_count = 0; 1239 1240 spa->spa_async_suspended = 0; 1241 1242 if (spa->spa_comment != NULL) { 1243 spa_strfree(spa->spa_comment); 1244 spa->spa_comment = NULL; 1245 } 1246 1247 spa_config_exit(spa, SCL_ALL, FTAG); 1248} 1249 1250/* 1251 * Load (or re-load) the current list of vdevs describing the active spares for 1252 * this pool. When this is called, we have some form of basic information in 1253 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1254 * then re-generate a more complete list including status information. 1255 */ 1256static void 1257spa_load_spares(spa_t *spa) 1258{ 1259 nvlist_t **spares; 1260 uint_t nspares; 1261 int i; 1262 vdev_t *vd, *tvd; 1263 1264 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1265 1266 /* 1267 * First, close and free any existing spare vdevs. 1268 */ 1269 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1270 vd = spa->spa_spares.sav_vdevs[i]; 1271 1272 /* Undo the call to spa_activate() below */ 1273 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1274 B_FALSE)) != NULL && tvd->vdev_isspare) 1275 spa_spare_remove(tvd); 1276 vdev_close(vd); 1277 vdev_free(vd); 1278 } 1279 1280 if (spa->spa_spares.sav_vdevs) 1281 kmem_free(spa->spa_spares.sav_vdevs, 1282 spa->spa_spares.sav_count * sizeof (void *)); 1283 1284 if (spa->spa_spares.sav_config == NULL) 1285 nspares = 0; 1286 else 1287 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1288 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1289 1290 spa->spa_spares.sav_count = (int)nspares; 1291 spa->spa_spares.sav_vdevs = NULL; 1292 1293 if (nspares == 0) 1294 return; 1295 1296 /* 1297 * Construct the array of vdevs, opening them to get status in the 1298 * process. For each spare, there is potentially two different vdev_t 1299 * structures associated with it: one in the list of spares (used only 1300 * for basic validation purposes) and one in the active vdev 1301 * configuration (if it's spared in). During this phase we open and 1302 * validate each vdev on the spare list. If the vdev also exists in the 1303 * active configuration, then we also mark this vdev as an active spare. 1304 */ 1305 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1306 KM_SLEEP); 1307 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1308 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1309 VDEV_ALLOC_SPARE) == 0); 1310 ASSERT(vd != NULL); 1311 1312 spa->spa_spares.sav_vdevs[i] = vd; 1313 1314 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1315 B_FALSE)) != NULL) { 1316 if (!tvd->vdev_isspare) 1317 spa_spare_add(tvd); 1318 1319 /* 1320 * We only mark the spare active if we were successfully 1321 * able to load the vdev. Otherwise, importing a pool 1322 * with a bad active spare would result in strange 1323 * behavior, because multiple pool would think the spare 1324 * is actively in use. 1325 * 1326 * There is a vulnerability here to an equally bizarre 1327 * circumstance, where a dead active spare is later 1328 * brought back to life (onlined or otherwise). Given 1329 * the rarity of this scenario, and the extra complexity 1330 * it adds, we ignore the possibility. 1331 */ 1332 if (!vdev_is_dead(tvd)) 1333 spa_spare_activate(tvd); 1334 } 1335 1336 vd->vdev_top = vd; 1337 vd->vdev_aux = &spa->spa_spares; 1338 1339 if (vdev_open(vd) != 0) 1340 continue; 1341 1342 if (vdev_validate_aux(vd) == 0) 1343 spa_spare_add(vd); 1344 } 1345 1346 /* 1347 * Recompute the stashed list of spares, with status information 1348 * this time. 1349 */ 1350 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1351 DATA_TYPE_NVLIST_ARRAY) == 0); 1352 1353 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1354 KM_SLEEP); 1355 for (i = 0; i < spa->spa_spares.sav_count; i++) 1356 spares[i] = vdev_config_generate(spa, 1357 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1358 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1359 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1360 for (i = 0; i < spa->spa_spares.sav_count; i++) 1361 nvlist_free(spares[i]); 1362 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1363} 1364 1365/* 1366 * Load (or re-load) the current list of vdevs describing the active l2cache for 1367 * this pool. When this is called, we have some form of basic information in 1368 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1369 * then re-generate a more complete list including status information. 1370 * Devices which are already active have their details maintained, and are 1371 * not re-opened. 1372 */ 1373static void 1374spa_load_l2cache(spa_t *spa) 1375{ 1376 nvlist_t **l2cache; 1377 uint_t nl2cache; 1378 int i, j, oldnvdevs; 1379 uint64_t guid; 1380 vdev_t *vd, **oldvdevs, **newvdevs; 1381 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1382 1383 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1384 1385 if (sav->sav_config != NULL) { 1386 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1387 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1388 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1389 } else { 1390 nl2cache = 0; 1391 } 1392 1393 oldvdevs = sav->sav_vdevs; 1394 oldnvdevs = sav->sav_count; 1395 sav->sav_vdevs = NULL; 1396 sav->sav_count = 0; 1397 1398 /* 1399 * Process new nvlist of vdevs. 1400 */ 1401 for (i = 0; i < nl2cache; i++) { 1402 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1403 &guid) == 0); 1404 1405 newvdevs[i] = NULL; 1406 for (j = 0; j < oldnvdevs; j++) { 1407 vd = oldvdevs[j]; 1408 if (vd != NULL && guid == vd->vdev_guid) { 1409 /* 1410 * Retain previous vdev for add/remove ops. 1411 */ 1412 newvdevs[i] = vd; 1413 oldvdevs[j] = NULL; 1414 break; 1415 } 1416 } 1417 1418 if (newvdevs[i] == NULL) { 1419 /* 1420 * Create new vdev 1421 */ 1422 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1423 VDEV_ALLOC_L2CACHE) == 0); 1424 ASSERT(vd != NULL); 1425 newvdevs[i] = vd; 1426 1427 /* 1428 * Commit this vdev as an l2cache device, 1429 * even if it fails to open. 1430 */ 1431 spa_l2cache_add(vd); 1432 1433 vd->vdev_top = vd; 1434 vd->vdev_aux = sav; 1435 1436 spa_l2cache_activate(vd); 1437 1438 if (vdev_open(vd) != 0) 1439 continue; 1440 1441 (void) vdev_validate_aux(vd); 1442 1443 if (!vdev_is_dead(vd)) 1444 l2arc_add_vdev(spa, vd); 1445 } 1446 } 1447 1448 /* 1449 * Purge vdevs that were dropped 1450 */ 1451 for (i = 0; i < oldnvdevs; i++) { 1452 uint64_t pool; 1453 1454 vd = oldvdevs[i]; 1455 if (vd != NULL) { 1456 ASSERT(vd->vdev_isl2cache); 1457 1458 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1459 pool != 0ULL && l2arc_vdev_present(vd)) 1460 l2arc_remove_vdev(vd); 1461 vdev_clear_stats(vd); 1462 vdev_free(vd); 1463 } 1464 } 1465 1466 if (oldvdevs) 1467 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1468 1469 if (sav->sav_config == NULL) 1470 goto out; 1471 1472 sav->sav_vdevs = newvdevs; 1473 sav->sav_count = (int)nl2cache; 1474 1475 /* 1476 * Recompute the stashed list of l2cache devices, with status 1477 * information this time. 1478 */ 1479 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1480 DATA_TYPE_NVLIST_ARRAY) == 0); 1481 1482 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1483 for (i = 0; i < sav->sav_count; i++) 1484 l2cache[i] = vdev_config_generate(spa, 1485 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1486 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1487 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1488out: 1489 for (i = 0; i < sav->sav_count; i++) 1490 nvlist_free(l2cache[i]); 1491 if (sav->sav_count) 1492 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1493} 1494 1495static int 1496load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1497{ 1498 dmu_buf_t *db; 1499 char *packed = NULL; 1500 size_t nvsize = 0; 1501 int error; 1502 *value = NULL; 1503 1504 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 1505 nvsize = *(uint64_t *)db->db_data; 1506 dmu_buf_rele(db, FTAG); 1507 1508 packed = kmem_alloc(nvsize, KM_SLEEP); 1509 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1510 DMU_READ_PREFETCH); 1511 if (error == 0) 1512 error = nvlist_unpack(packed, nvsize, value, 0); 1513 kmem_free(packed, nvsize); 1514 1515 return (error); 1516} 1517 1518/* 1519 * Checks to see if the given vdev could not be opened, in which case we post a 1520 * sysevent to notify the autoreplace code that the device has been removed. 1521 */ 1522static void 1523spa_check_removed(vdev_t *vd) 1524{ 1525 for (int c = 0; c < vd->vdev_children; c++) 1526 spa_check_removed(vd->vdev_child[c]); 1527 1528 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) { 1529 zfs_post_autoreplace(vd->vdev_spa, vd); 1530 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1531 } 1532} 1533 1534/* 1535 * Validate the current config against the MOS config 1536 */ 1537static boolean_t 1538spa_config_valid(spa_t *spa, nvlist_t *config) 1539{ 1540 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1541 nvlist_t *nv; 1542 1543 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1544 1545 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1546 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1547 1548 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1549 1550 /* 1551 * If we're doing a normal import, then build up any additional 1552 * diagnostic information about missing devices in this config. 1553 * We'll pass this up to the user for further processing. 1554 */ 1555 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1556 nvlist_t **child, *nv; 1557 uint64_t idx = 0; 1558 1559 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1560 KM_SLEEP); 1561 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1562 1563 for (int c = 0; c < rvd->vdev_children; c++) { 1564 vdev_t *tvd = rvd->vdev_child[c]; 1565 vdev_t *mtvd = mrvd->vdev_child[c]; 1566 1567 if (tvd->vdev_ops == &vdev_missing_ops && 1568 mtvd->vdev_ops != &vdev_missing_ops && 1569 mtvd->vdev_islog) 1570 child[idx++] = vdev_config_generate(spa, mtvd, 1571 B_FALSE, 0); 1572 } 1573 1574 if (idx) { 1575 VERIFY(nvlist_add_nvlist_array(nv, 1576 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1577 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1578 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1579 1580 for (int i = 0; i < idx; i++) 1581 nvlist_free(child[i]); 1582 } 1583 nvlist_free(nv); 1584 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1585 } 1586 1587 /* 1588 * Compare the root vdev tree with the information we have 1589 * from the MOS config (mrvd). Check each top-level vdev 1590 * with the corresponding MOS config top-level (mtvd). 1591 */ 1592 for (int c = 0; c < rvd->vdev_children; c++) { 1593 vdev_t *tvd = rvd->vdev_child[c]; 1594 vdev_t *mtvd = mrvd->vdev_child[c]; 1595 1596 /* 1597 * Resolve any "missing" vdevs in the current configuration. 1598 * If we find that the MOS config has more accurate information 1599 * about the top-level vdev then use that vdev instead. 1600 */ 1601 if (tvd->vdev_ops == &vdev_missing_ops && 1602 mtvd->vdev_ops != &vdev_missing_ops) { 1603 1604 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1605 continue; 1606 1607 /* 1608 * Device specific actions. 1609 */ 1610 if (mtvd->vdev_islog) { 1611 spa_set_log_state(spa, SPA_LOG_CLEAR); 1612 } else { 1613 /* 1614 * XXX - once we have 'readonly' pool 1615 * support we should be able to handle 1616 * missing data devices by transitioning 1617 * the pool to readonly. 1618 */ 1619 continue; 1620 } 1621 1622 /* 1623 * Swap the missing vdev with the data we were 1624 * able to obtain from the MOS config. 1625 */ 1626 vdev_remove_child(rvd, tvd); 1627 vdev_remove_child(mrvd, mtvd); 1628 1629 vdev_add_child(rvd, mtvd); 1630 vdev_add_child(mrvd, tvd); 1631 1632 spa_config_exit(spa, SCL_ALL, FTAG); 1633 vdev_load(mtvd); 1634 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1635 1636 vdev_reopen(rvd); 1637 } else if (mtvd->vdev_islog) { 1638 /* 1639 * Load the slog device's state from the MOS config 1640 * since it's possible that the label does not 1641 * contain the most up-to-date information. 1642 */ 1643 vdev_load_log_state(tvd, mtvd); 1644 vdev_reopen(tvd); 1645 } 1646 } 1647 vdev_free(mrvd); 1648 spa_config_exit(spa, SCL_ALL, FTAG); 1649 1650 /* 1651 * Ensure we were able to validate the config. 1652 */ 1653 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1654} 1655 1656/* 1657 * Check for missing log devices 1658 */ 1659static int 1660spa_check_logs(spa_t *spa) 1661{ 1662 switch (spa->spa_log_state) { 1663 case SPA_LOG_MISSING: 1664 /* need to recheck in case slog has been restored */ 1665 case SPA_LOG_UNKNOWN: 1666 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL, 1667 DS_FIND_CHILDREN)) { 1668 spa_set_log_state(spa, SPA_LOG_MISSING); 1669 return (1); 1670 } 1671 break; 1672 } 1673 return (0); 1674} 1675 1676static boolean_t 1677spa_passivate_log(spa_t *spa) 1678{ 1679 vdev_t *rvd = spa->spa_root_vdev; 1680 boolean_t slog_found = B_FALSE; 1681 1682 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1683 1684 if (!spa_has_slogs(spa)) 1685 return (B_FALSE); 1686 1687 for (int c = 0; c < rvd->vdev_children; c++) { 1688 vdev_t *tvd = rvd->vdev_child[c]; 1689 metaslab_group_t *mg = tvd->vdev_mg; 1690 1691 if (tvd->vdev_islog) { 1692 metaslab_group_passivate(mg); 1693 slog_found = B_TRUE; 1694 } 1695 } 1696 1697 return (slog_found); 1698} 1699 1700static void 1701spa_activate_log(spa_t *spa) 1702{ 1703 vdev_t *rvd = spa->spa_root_vdev; 1704 1705 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1706 1707 for (int c = 0; c < rvd->vdev_children; c++) { 1708 vdev_t *tvd = rvd->vdev_child[c]; 1709 metaslab_group_t *mg = tvd->vdev_mg; 1710 1711 if (tvd->vdev_islog) 1712 metaslab_group_activate(mg); 1713 } 1714} 1715 1716int 1717spa_offline_log(spa_t *spa) 1718{ 1719 int error = 0; 1720 1721 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1722 NULL, DS_FIND_CHILDREN)) == 0) { 1723 1724 /* 1725 * We successfully offlined the log device, sync out the 1726 * current txg so that the "stubby" block can be removed 1727 * by zil_sync(). 1728 */ 1729 txg_wait_synced(spa->spa_dsl_pool, 0); 1730 } 1731 return (error); 1732} 1733 1734static void 1735spa_aux_check_removed(spa_aux_vdev_t *sav) 1736{ 1737 int i; 1738 1739 for (i = 0; i < sav->sav_count; i++) 1740 spa_check_removed(sav->sav_vdevs[i]); 1741} 1742 1743void 1744spa_claim_notify(zio_t *zio) 1745{ 1746 spa_t *spa = zio->io_spa; 1747 1748 if (zio->io_error) 1749 return; 1750 1751 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1752 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1753 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1754 mutex_exit(&spa->spa_props_lock); 1755} 1756 1757typedef struct spa_load_error { 1758 uint64_t sle_meta_count; 1759 uint64_t sle_data_count; 1760} spa_load_error_t; 1761 1762static void 1763spa_load_verify_done(zio_t *zio) 1764{ 1765 blkptr_t *bp = zio->io_bp; 1766 spa_load_error_t *sle = zio->io_private; 1767 dmu_object_type_t type = BP_GET_TYPE(bp); 1768 int error = zio->io_error; 1769 1770 if (error) { 1771 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1772 type != DMU_OT_INTENT_LOG) 1773 atomic_add_64(&sle->sle_meta_count, 1); 1774 else 1775 atomic_add_64(&sle->sle_data_count, 1); 1776 } 1777 zio_data_buf_free(zio->io_data, zio->io_size); 1778} 1779 1780/*ARGSUSED*/ 1781static int 1782spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1783 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1784{ 1785 if (bp != NULL) { 1786 zio_t *rio = arg; 1787 size_t size = BP_GET_PSIZE(bp); 1788 void *data = zio_data_buf_alloc(size); 1789 1790 zio_nowait(zio_read(rio, spa, bp, data, size, 1791 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1792 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1793 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1794 } 1795 return (0); 1796} 1797 1798static int 1799spa_load_verify(spa_t *spa) 1800{ 1801 zio_t *rio; 1802 spa_load_error_t sle = { 0 }; 1803 zpool_rewind_policy_t policy; 1804 boolean_t verify_ok = B_FALSE; 1805 int error; 1806 1807 zpool_get_rewind_policy(spa->spa_config, &policy); 1808 1809 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1810 return (0); 1811 1812 rio = zio_root(spa, NULL, &sle, 1813 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1814 1815 error = traverse_pool(spa, spa->spa_verify_min_txg, 1816 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1817 1818 (void) zio_wait(rio); 1819 1820 spa->spa_load_meta_errors = sle.sle_meta_count; 1821 spa->spa_load_data_errors = sle.sle_data_count; 1822 1823 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1824 sle.sle_data_count <= policy.zrp_maxdata) { 1825 int64_t loss = 0; 1826 1827 verify_ok = B_TRUE; 1828 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1829 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1830 1831 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1832 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1833 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1834 VERIFY(nvlist_add_int64(spa->spa_load_info, 1835 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1836 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1837 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1838 } else { 1839 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1840 } 1841 1842 if (error) { 1843 if (error != ENXIO && error != EIO) 1844 error = EIO; 1845 return (error); 1846 } 1847 1848 return (verify_ok ? 0 : EIO); 1849} 1850 1851/* 1852 * Find a value in the pool props object. 1853 */ 1854static void 1855spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1856{ 1857 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1858 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1859} 1860 1861/* 1862 * Find a value in the pool directory object. 1863 */ 1864static int 1865spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1866{ 1867 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1868 name, sizeof (uint64_t), 1, val)); 1869} 1870 1871static int 1872spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1873{ 1874 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1875 return (err); 1876} 1877 1878/* 1879 * Fix up config after a partly-completed split. This is done with the 1880 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1881 * pool have that entry in their config, but only the splitting one contains 1882 * a list of all the guids of the vdevs that are being split off. 1883 * 1884 * This function determines what to do with that list: either rejoin 1885 * all the disks to the pool, or complete the splitting process. To attempt 1886 * the rejoin, each disk that is offlined is marked online again, and 1887 * we do a reopen() call. If the vdev label for every disk that was 1888 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 1889 * then we call vdev_split() on each disk, and complete the split. 1890 * 1891 * Otherwise we leave the config alone, with all the vdevs in place in 1892 * the original pool. 1893 */ 1894static void 1895spa_try_repair(spa_t *spa, nvlist_t *config) 1896{ 1897 uint_t extracted; 1898 uint64_t *glist; 1899 uint_t i, gcount; 1900 nvlist_t *nvl; 1901 vdev_t **vd; 1902 boolean_t attempt_reopen; 1903 1904 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 1905 return; 1906 1907 /* check that the config is complete */ 1908 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 1909 &glist, &gcount) != 0) 1910 return; 1911 1912 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 1913 1914 /* attempt to online all the vdevs & validate */ 1915 attempt_reopen = B_TRUE; 1916 for (i = 0; i < gcount; i++) { 1917 if (glist[i] == 0) /* vdev is hole */ 1918 continue; 1919 1920 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 1921 if (vd[i] == NULL) { 1922 /* 1923 * Don't bother attempting to reopen the disks; 1924 * just do the split. 1925 */ 1926 attempt_reopen = B_FALSE; 1927 } else { 1928 /* attempt to re-online it */ 1929 vd[i]->vdev_offline = B_FALSE; 1930 } 1931 } 1932 1933 if (attempt_reopen) { 1934 vdev_reopen(spa->spa_root_vdev); 1935 1936 /* check each device to see what state it's in */ 1937 for (extracted = 0, i = 0; i < gcount; i++) { 1938 if (vd[i] != NULL && 1939 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 1940 break; 1941 ++extracted; 1942 } 1943 } 1944 1945 /* 1946 * If every disk has been moved to the new pool, or if we never 1947 * even attempted to look at them, then we split them off for 1948 * good. 1949 */ 1950 if (!attempt_reopen || gcount == extracted) { 1951 for (i = 0; i < gcount; i++) 1952 if (vd[i] != NULL) 1953 vdev_split(vd[i]); 1954 vdev_reopen(spa->spa_root_vdev); 1955 } 1956 1957 kmem_free(vd, gcount * sizeof (vdev_t *)); 1958} 1959 1960static int 1961spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 1962 boolean_t mosconfig) 1963{ 1964 nvlist_t *config = spa->spa_config; 1965 char *ereport = FM_EREPORT_ZFS_POOL; 1966 char *comment; 1967 int error; 1968 uint64_t pool_guid; 1969 nvlist_t *nvl; 1970 1971 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 1972 return (EINVAL); 1973 1974 ASSERT(spa->spa_comment == NULL); 1975 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 1976 spa->spa_comment = spa_strdup(comment); 1977 1978 /* 1979 * Versioning wasn't explicitly added to the label until later, so if 1980 * it's not present treat it as the initial version. 1981 */ 1982 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 1983 &spa->spa_ubsync.ub_version) != 0) 1984 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 1985 1986 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 1987 &spa->spa_config_txg); 1988 1989 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 1990 spa_guid_exists(pool_guid, 0)) { 1991 error = EEXIST; 1992 } else { 1993 spa->spa_config_guid = pool_guid; 1994 1995 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 1996 &nvl) == 0) { 1997 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 1998 KM_SLEEP) == 0); 1999 } 2000 2001 nvlist_free(spa->spa_load_info); 2002 spa->spa_load_info = fnvlist_alloc(); 2003 2004 gethrestime(&spa->spa_loaded_ts); 2005 error = spa_load_impl(spa, pool_guid, config, state, type, 2006 mosconfig, &ereport); 2007 } 2008 2009 spa->spa_minref = refcount_count(&spa->spa_refcount); 2010 if (error) { 2011 if (error != EEXIST) { 2012 spa->spa_loaded_ts.tv_sec = 0; 2013 spa->spa_loaded_ts.tv_nsec = 0; 2014 } 2015 if (error != EBADF) { 2016 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2017 } 2018 } 2019 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2020 spa->spa_ena = 0; 2021 2022 return (error); 2023} 2024 2025/* 2026 * Load an existing storage pool, using the pool's builtin spa_config as a 2027 * source of configuration information. 2028 */ 2029static int 2030spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2031 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2032 char **ereport) 2033{ 2034 int error = 0; 2035 nvlist_t *nvroot = NULL; 2036 nvlist_t *label; 2037 vdev_t *rvd; 2038 uberblock_t *ub = &spa->spa_uberblock; 2039 uint64_t children, config_cache_txg = spa->spa_config_txg; 2040 int orig_mode = spa->spa_mode; 2041 int parse; 2042 uint64_t obj; 2043 boolean_t missing_feat_write = B_FALSE; 2044 2045 /* 2046 * If this is an untrusted config, access the pool in read-only mode. 2047 * This prevents things like resilvering recently removed devices. 2048 */ 2049 if (!mosconfig) 2050 spa->spa_mode = FREAD; 2051 2052 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2053 2054 spa->spa_load_state = state; 2055 2056 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2057 return (EINVAL); 2058 2059 parse = (type == SPA_IMPORT_EXISTING ? 2060 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2061 2062 /* 2063 * Create "The Godfather" zio to hold all async IOs 2064 */ 2065 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 2066 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 2067 2068 /* 2069 * Parse the configuration into a vdev tree. We explicitly set the 2070 * value that will be returned by spa_version() since parsing the 2071 * configuration requires knowing the version number. 2072 */ 2073 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2074 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2075 spa_config_exit(spa, SCL_ALL, FTAG); 2076 2077 if (error != 0) 2078 return (error); 2079 2080 ASSERT(spa->spa_root_vdev == rvd); 2081 2082 if (type != SPA_IMPORT_ASSEMBLE) { 2083 ASSERT(spa_guid(spa) == pool_guid); 2084 } 2085 2086 /* 2087 * Try to open all vdevs, loading each label in the process. 2088 */ 2089 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2090 error = vdev_open(rvd); 2091 spa_config_exit(spa, SCL_ALL, FTAG); 2092 if (error != 0) 2093 return (error); 2094 2095 /* 2096 * We need to validate the vdev labels against the configuration that 2097 * we have in hand, which is dependent on the setting of mosconfig. If 2098 * mosconfig is true then we're validating the vdev labels based on 2099 * that config. Otherwise, we're validating against the cached config 2100 * (zpool.cache) that was read when we loaded the zfs module, and then 2101 * later we will recursively call spa_load() and validate against 2102 * the vdev config. 2103 * 2104 * If we're assembling a new pool that's been split off from an 2105 * existing pool, the labels haven't yet been updated so we skip 2106 * validation for now. 2107 */ 2108 if (type != SPA_IMPORT_ASSEMBLE) { 2109 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2110 error = vdev_validate(rvd, mosconfig); 2111 spa_config_exit(spa, SCL_ALL, FTAG); 2112 2113 if (error != 0) 2114 return (error); 2115 2116 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2117 return (ENXIO); 2118 } 2119 2120 /* 2121 * Find the best uberblock. 2122 */ 2123 vdev_uberblock_load(rvd, ub, &label); 2124 2125 /* 2126 * If we weren't able to find a single valid uberblock, return failure. 2127 */ 2128 if (ub->ub_txg == 0) { 2129 nvlist_free(label); 2130 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2131 } 2132 2133 /* 2134 * If the pool has an unsupported version we can't open it. 2135 */ 2136 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2137 nvlist_free(label); 2138 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2139 } 2140 2141 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2142 nvlist_t *features; 2143 2144 /* 2145 * If we weren't able to find what's necessary for reading the 2146 * MOS in the label, return failure. 2147 */ 2148 if (label == NULL || nvlist_lookup_nvlist(label, 2149 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2150 nvlist_free(label); 2151 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2152 ENXIO)); 2153 } 2154 2155 /* 2156 * Update our in-core representation with the definitive values 2157 * from the label. 2158 */ 2159 nvlist_free(spa->spa_label_features); 2160 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2161 } 2162 2163 nvlist_free(label); 2164 2165 /* 2166 * Look through entries in the label nvlist's features_for_read. If 2167 * there is a feature listed there which we don't understand then we 2168 * cannot open a pool. 2169 */ 2170 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2171 nvlist_t *unsup_feat; 2172 2173 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2174 0); 2175 2176 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2177 NULL); nvp != NULL; 2178 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2179 if (!zfeature_is_supported(nvpair_name(nvp))) { 2180 VERIFY(nvlist_add_string(unsup_feat, 2181 nvpair_name(nvp), "") == 0); 2182 } 2183 } 2184 2185 if (!nvlist_empty(unsup_feat)) { 2186 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2187 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2188 nvlist_free(unsup_feat); 2189 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2190 ENOTSUP)); 2191 } 2192 2193 nvlist_free(unsup_feat); 2194 } 2195 2196 /* 2197 * If the vdev guid sum doesn't match the uberblock, we have an 2198 * incomplete configuration. We first check to see if the pool 2199 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2200 * If it is, defer the vdev_guid_sum check till later so we 2201 * can handle missing vdevs. 2202 */ 2203 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2204 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2205 rvd->vdev_guid_sum != ub->ub_guid_sum) 2206 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2207 2208 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2209 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2210 spa_try_repair(spa, config); 2211 spa_config_exit(spa, SCL_ALL, FTAG); 2212 nvlist_free(spa->spa_config_splitting); 2213 spa->spa_config_splitting = NULL; 2214 } 2215 2216 /* 2217 * Initialize internal SPA structures. 2218 */ 2219 spa->spa_state = POOL_STATE_ACTIVE; 2220 spa->spa_ubsync = spa->spa_uberblock; 2221 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2222 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2223 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2224 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2225 spa->spa_claim_max_txg = spa->spa_first_txg; 2226 spa->spa_prev_software_version = ub->ub_software_version; 2227 2228 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2229 if (error) 2230 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2231 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2232 2233 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2234 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2235 2236 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2237 boolean_t missing_feat_read = B_FALSE; 2238 nvlist_t *unsup_feat, *enabled_feat; 2239 2240 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2241 &spa->spa_feat_for_read_obj) != 0) { 2242 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2243 } 2244 2245 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2246 &spa->spa_feat_for_write_obj) != 0) { 2247 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2248 } 2249 2250 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2251 &spa->spa_feat_desc_obj) != 0) { 2252 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2253 } 2254 2255 enabled_feat = fnvlist_alloc(); 2256 unsup_feat = fnvlist_alloc(); 2257 2258 if (!feature_is_supported(spa->spa_meta_objset, 2259 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, 2260 unsup_feat, enabled_feat)) 2261 missing_feat_read = B_TRUE; 2262 2263 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2264 if (!feature_is_supported(spa->spa_meta_objset, 2265 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, 2266 unsup_feat, enabled_feat)) { 2267 missing_feat_write = B_TRUE; 2268 } 2269 } 2270 2271 fnvlist_add_nvlist(spa->spa_load_info, 2272 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2273 2274 if (!nvlist_empty(unsup_feat)) { 2275 fnvlist_add_nvlist(spa->spa_load_info, 2276 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2277 } 2278 2279 fnvlist_free(enabled_feat); 2280 fnvlist_free(unsup_feat); 2281 2282 if (!missing_feat_read) { 2283 fnvlist_add_boolean(spa->spa_load_info, 2284 ZPOOL_CONFIG_CAN_RDONLY); 2285 } 2286 2287 /* 2288 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2289 * twofold: to determine whether the pool is available for 2290 * import in read-write mode and (if it is not) whether the 2291 * pool is available for import in read-only mode. If the pool 2292 * is available for import in read-write mode, it is displayed 2293 * as available in userland; if it is not available for import 2294 * in read-only mode, it is displayed as unavailable in 2295 * userland. If the pool is available for import in read-only 2296 * mode but not read-write mode, it is displayed as unavailable 2297 * in userland with a special note that the pool is actually 2298 * available for open in read-only mode. 2299 * 2300 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2301 * missing a feature for write, we must first determine whether 2302 * the pool can be opened read-only before returning to 2303 * userland in order to know whether to display the 2304 * abovementioned note. 2305 */ 2306 if (missing_feat_read || (missing_feat_write && 2307 spa_writeable(spa))) { 2308 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2309 ENOTSUP)); 2310 } 2311 } 2312 2313 spa->spa_is_initializing = B_TRUE; 2314 error = dsl_pool_open(spa->spa_dsl_pool); 2315 spa->spa_is_initializing = B_FALSE; 2316 if (error != 0) 2317 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2318 2319 if (!mosconfig) { 2320 uint64_t hostid; 2321 nvlist_t *policy = NULL, *nvconfig; 2322 2323 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2325 2326 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2327 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2328 char *hostname; 2329 unsigned long myhostid = 0; 2330 2331 VERIFY(nvlist_lookup_string(nvconfig, 2332 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2333 2334#ifdef _KERNEL 2335 myhostid = zone_get_hostid(NULL); 2336#else /* _KERNEL */ 2337 /* 2338 * We're emulating the system's hostid in userland, so 2339 * we can't use zone_get_hostid(). 2340 */ 2341 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2342#endif /* _KERNEL */ 2343 if (check_hostid && hostid != 0 && myhostid != 0 && 2344 hostid != myhostid) { 2345 nvlist_free(nvconfig); 2346 cmn_err(CE_WARN, "pool '%s' could not be " 2347 "loaded as it was last accessed by " 2348 "another system (host: %s hostid: 0x%lx). " 2349 "See: http://illumos.org/msg/ZFS-8000-EY", 2350 spa_name(spa), hostname, 2351 (unsigned long)hostid); 2352 return (EBADF); 2353 } 2354 } 2355 if (nvlist_lookup_nvlist(spa->spa_config, 2356 ZPOOL_REWIND_POLICY, &policy) == 0) 2357 VERIFY(nvlist_add_nvlist(nvconfig, 2358 ZPOOL_REWIND_POLICY, policy) == 0); 2359 2360 spa_config_set(spa, nvconfig); 2361 spa_unload(spa); 2362 spa_deactivate(spa); 2363 spa_activate(spa, orig_mode); 2364 2365 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2366 } 2367 2368 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2369 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2370 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2371 if (error != 0) 2372 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2373 2374 /* 2375 * Load the bit that tells us to use the new accounting function 2376 * (raid-z deflation). If we have an older pool, this will not 2377 * be present. 2378 */ 2379 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2380 if (error != 0 && error != ENOENT) 2381 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2382 2383 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2384 &spa->spa_creation_version); 2385 if (error != 0 && error != ENOENT) 2386 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2387 2388 /* 2389 * Load the persistent error log. If we have an older pool, this will 2390 * not be present. 2391 */ 2392 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2393 if (error != 0 && error != ENOENT) 2394 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2395 2396 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2397 &spa->spa_errlog_scrub); 2398 if (error != 0 && error != ENOENT) 2399 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2400 2401 /* 2402 * Load the history object. If we have an older pool, this 2403 * will not be present. 2404 */ 2405 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2406 if (error != 0 && error != ENOENT) 2407 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2408 2409 /* 2410 * If we're assembling the pool from the split-off vdevs of 2411 * an existing pool, we don't want to attach the spares & cache 2412 * devices. 2413 */ 2414 2415 /* 2416 * Load any hot spares for this pool. 2417 */ 2418 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2419 if (error != 0 && error != ENOENT) 2420 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2421 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2422 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2423 if (load_nvlist(spa, spa->spa_spares.sav_object, 2424 &spa->spa_spares.sav_config) != 0) 2425 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2426 2427 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2428 spa_load_spares(spa); 2429 spa_config_exit(spa, SCL_ALL, FTAG); 2430 } else if (error == 0) { 2431 spa->spa_spares.sav_sync = B_TRUE; 2432 } 2433 2434 /* 2435 * Load any level 2 ARC devices for this pool. 2436 */ 2437 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2438 &spa->spa_l2cache.sav_object); 2439 if (error != 0 && error != ENOENT) 2440 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2441 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2442 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2443 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2444 &spa->spa_l2cache.sav_config) != 0) 2445 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2446 2447 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2448 spa_load_l2cache(spa); 2449 spa_config_exit(spa, SCL_ALL, FTAG); 2450 } else if (error == 0) { 2451 spa->spa_l2cache.sav_sync = B_TRUE; 2452 } 2453 2454 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2455 2456 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2457 if (error && error != ENOENT) 2458 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2459 2460 if (error == 0) { 2461 uint64_t autoreplace; 2462 2463 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2464 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2465 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2466 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2467 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2468 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2469 &spa->spa_dedup_ditto); 2470 2471 spa->spa_autoreplace = (autoreplace != 0); 2472 } 2473 2474 /* 2475 * If the 'autoreplace' property is set, then post a resource notifying 2476 * the ZFS DE that it should not issue any faults for unopenable 2477 * devices. We also iterate over the vdevs, and post a sysevent for any 2478 * unopenable vdevs so that the normal autoreplace handler can take 2479 * over. 2480 */ 2481 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2482 spa_check_removed(spa->spa_root_vdev); 2483 /* 2484 * For the import case, this is done in spa_import(), because 2485 * at this point we're using the spare definitions from 2486 * the MOS config, not necessarily from the userland config. 2487 */ 2488 if (state != SPA_LOAD_IMPORT) { 2489 spa_aux_check_removed(&spa->spa_spares); 2490 spa_aux_check_removed(&spa->spa_l2cache); 2491 } 2492 } 2493 2494 /* 2495 * Load the vdev state for all toplevel vdevs. 2496 */ 2497 vdev_load(rvd); 2498 2499 /* 2500 * Propagate the leaf DTLs we just loaded all the way up the tree. 2501 */ 2502 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2503 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2504 spa_config_exit(spa, SCL_ALL, FTAG); 2505 2506 /* 2507 * Load the DDTs (dedup tables). 2508 */ 2509 error = ddt_load(spa); 2510 if (error != 0) 2511 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2512 2513 spa_update_dspace(spa); 2514 2515 /* 2516 * Validate the config, using the MOS config to fill in any 2517 * information which might be missing. If we fail to validate 2518 * the config then declare the pool unfit for use. If we're 2519 * assembling a pool from a split, the log is not transferred 2520 * over. 2521 */ 2522 if (type != SPA_IMPORT_ASSEMBLE) { 2523 nvlist_t *nvconfig; 2524 2525 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2526 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2527 2528 if (!spa_config_valid(spa, nvconfig)) { 2529 nvlist_free(nvconfig); 2530 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2531 ENXIO)); 2532 } 2533 nvlist_free(nvconfig); 2534 2535 /* 2536 * Now that we've validated the config, check the state of the 2537 * root vdev. If it can't be opened, it indicates one or 2538 * more toplevel vdevs are faulted. 2539 */ 2540 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2541 return (ENXIO); 2542 2543 if (spa_check_logs(spa)) { 2544 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2545 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2546 } 2547 } 2548 2549 if (missing_feat_write) { 2550 ASSERT(state == SPA_LOAD_TRYIMPORT); 2551 2552 /* 2553 * At this point, we know that we can open the pool in 2554 * read-only mode but not read-write mode. We now have enough 2555 * information and can return to userland. 2556 */ 2557 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2558 } 2559 2560 /* 2561 * We've successfully opened the pool, verify that we're ready 2562 * to start pushing transactions. 2563 */ 2564 if (state != SPA_LOAD_TRYIMPORT) { 2565 if (error = spa_load_verify(spa)) 2566 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2567 error)); 2568 } 2569 2570 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2571 spa->spa_load_max_txg == UINT64_MAX)) { 2572 dmu_tx_t *tx; 2573 int need_update = B_FALSE; 2574 2575 ASSERT(state != SPA_LOAD_TRYIMPORT); 2576 2577 /* 2578 * Claim log blocks that haven't been committed yet. 2579 * This must all happen in a single txg. 2580 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2581 * invoked from zil_claim_log_block()'s i/o done callback. 2582 * Price of rollback is that we abandon the log. 2583 */ 2584 spa->spa_claiming = B_TRUE; 2585 2586 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2587 spa_first_txg(spa)); 2588 (void) dmu_objset_find(spa_name(spa), 2589 zil_claim, tx, DS_FIND_CHILDREN); 2590 dmu_tx_commit(tx); 2591 2592 spa->spa_claiming = B_FALSE; 2593 2594 spa_set_log_state(spa, SPA_LOG_GOOD); 2595 spa->spa_sync_on = B_TRUE; 2596 txg_sync_start(spa->spa_dsl_pool); 2597 2598 /* 2599 * Wait for all claims to sync. We sync up to the highest 2600 * claimed log block birth time so that claimed log blocks 2601 * don't appear to be from the future. spa_claim_max_txg 2602 * will have been set for us by either zil_check_log_chain() 2603 * (invoked from spa_check_logs()) or zil_claim() above. 2604 */ 2605 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2606 2607 /* 2608 * If the config cache is stale, or we have uninitialized 2609 * metaslabs (see spa_vdev_add()), then update the config. 2610 * 2611 * If this is a verbatim import, trust the current 2612 * in-core spa_config and update the disk labels. 2613 */ 2614 if (config_cache_txg != spa->spa_config_txg || 2615 state == SPA_LOAD_IMPORT || 2616 state == SPA_LOAD_RECOVER || 2617 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2618 need_update = B_TRUE; 2619 2620 for (int c = 0; c < rvd->vdev_children; c++) 2621 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2622 need_update = B_TRUE; 2623 2624 /* 2625 * Update the config cache asychronously in case we're the 2626 * root pool, in which case the config cache isn't writable yet. 2627 */ 2628 if (need_update) 2629 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2630 2631 /* 2632 * Check all DTLs to see if anything needs resilvering. 2633 */ 2634 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2635 vdev_resilver_needed(rvd, NULL, NULL)) 2636 spa_async_request(spa, SPA_ASYNC_RESILVER); 2637 2638 /* 2639 * Delete any inconsistent datasets. 2640 */ 2641 (void) dmu_objset_find(spa_name(spa), 2642 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2643 2644 /* 2645 * Clean up any stale temporary dataset userrefs. 2646 */ 2647 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2648 } 2649 2650 return (0); 2651} 2652 2653static int 2654spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2655{ 2656 int mode = spa->spa_mode; 2657 2658 spa_unload(spa); 2659 spa_deactivate(spa); 2660 2661 spa->spa_load_max_txg--; 2662 2663 spa_activate(spa, mode); 2664 spa_async_suspend(spa); 2665 2666 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2667} 2668 2669/* 2670 * If spa_load() fails this function will try loading prior txg's. If 2671 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2672 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2673 * function will not rewind the pool and will return the same error as 2674 * spa_load(). 2675 */ 2676static int 2677spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2678 uint64_t max_request, int rewind_flags) 2679{ 2680 nvlist_t *loadinfo = NULL; 2681 nvlist_t *config = NULL; 2682 int load_error, rewind_error; 2683 uint64_t safe_rewind_txg; 2684 uint64_t min_txg; 2685 2686 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2687 spa->spa_load_max_txg = spa->spa_load_txg; 2688 spa_set_log_state(spa, SPA_LOG_CLEAR); 2689 } else { 2690 spa->spa_load_max_txg = max_request; 2691 } 2692 2693 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2694 mosconfig); 2695 if (load_error == 0) 2696 return (0); 2697 2698 if (spa->spa_root_vdev != NULL) 2699 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2700 2701 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2702 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2703 2704 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2705 nvlist_free(config); 2706 return (load_error); 2707 } 2708 2709 if (state == SPA_LOAD_RECOVER) { 2710 /* Price of rolling back is discarding txgs, including log */ 2711 spa_set_log_state(spa, SPA_LOG_CLEAR); 2712 } else { 2713 /* 2714 * If we aren't rolling back save the load info from our first 2715 * import attempt so that we can restore it after attempting 2716 * to rewind. 2717 */ 2718 loadinfo = spa->spa_load_info; 2719 spa->spa_load_info = fnvlist_alloc(); 2720 } 2721 2722 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2723 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2724 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2725 TXG_INITIAL : safe_rewind_txg; 2726 2727 /* 2728 * Continue as long as we're finding errors, we're still within 2729 * the acceptable rewind range, and we're still finding uberblocks 2730 */ 2731 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2732 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2733 if (spa->spa_load_max_txg < safe_rewind_txg) 2734 spa->spa_extreme_rewind = B_TRUE; 2735 rewind_error = spa_load_retry(spa, state, mosconfig); 2736 } 2737 2738 spa->spa_extreme_rewind = B_FALSE; 2739 spa->spa_load_max_txg = UINT64_MAX; 2740 2741 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2742 spa_config_set(spa, config); 2743 2744 if (state == SPA_LOAD_RECOVER) { 2745 ASSERT3P(loadinfo, ==, NULL); 2746 return (rewind_error); 2747 } else { 2748 /* Store the rewind info as part of the initial load info */ 2749 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2750 spa->spa_load_info); 2751 2752 /* Restore the initial load info */ 2753 fnvlist_free(spa->spa_load_info); 2754 spa->spa_load_info = loadinfo; 2755 2756 return (load_error); 2757 } 2758} 2759 2760/* 2761 * Pool Open/Import 2762 * 2763 * The import case is identical to an open except that the configuration is sent 2764 * down from userland, instead of grabbed from the configuration cache. For the 2765 * case of an open, the pool configuration will exist in the 2766 * POOL_STATE_UNINITIALIZED state. 2767 * 2768 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2769 * the same time open the pool, without having to keep around the spa_t in some 2770 * ambiguous state. 2771 */ 2772static int 2773spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2774 nvlist_t **config) 2775{ 2776 spa_t *spa; 2777 spa_load_state_t state = SPA_LOAD_OPEN; 2778 int error; 2779 int locked = B_FALSE; 2780 int firstopen = B_FALSE; 2781 2782 *spapp = NULL; 2783 2784 /* 2785 * As disgusting as this is, we need to support recursive calls to this 2786 * function because dsl_dir_open() is called during spa_load(), and ends 2787 * up calling spa_open() again. The real fix is to figure out how to 2788 * avoid dsl_dir_open() calling this in the first place. 2789 */ 2790 if (mutex_owner(&spa_namespace_lock) != curthread) { 2791 mutex_enter(&spa_namespace_lock); 2792 locked = B_TRUE; 2793 } 2794 2795 if ((spa = spa_lookup(pool)) == NULL) { 2796 if (locked) 2797 mutex_exit(&spa_namespace_lock); 2798 return (ENOENT); 2799 } 2800 2801 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2802 zpool_rewind_policy_t policy; 2803 2804 firstopen = B_TRUE; 2805 2806 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2807 &policy); 2808 if (policy.zrp_request & ZPOOL_DO_REWIND) 2809 state = SPA_LOAD_RECOVER; 2810 2811 spa_activate(spa, spa_mode_global); 2812 2813 if (state != SPA_LOAD_RECOVER) 2814 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2815 2816 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2817 policy.zrp_request); 2818 2819 if (error == EBADF) { 2820 /* 2821 * If vdev_validate() returns failure (indicated by 2822 * EBADF), it indicates that one of the vdevs indicates 2823 * that the pool has been exported or destroyed. If 2824 * this is the case, the config cache is out of sync and 2825 * we should remove the pool from the namespace. 2826 */ 2827 spa_unload(spa); 2828 spa_deactivate(spa); 2829 spa_config_sync(spa, B_TRUE, B_TRUE); 2830 spa_remove(spa); 2831 if (locked) 2832 mutex_exit(&spa_namespace_lock); 2833 return (ENOENT); 2834 } 2835 2836 if (error) { 2837 /* 2838 * We can't open the pool, but we still have useful 2839 * information: the state of each vdev after the 2840 * attempted vdev_open(). Return this to the user. 2841 */ 2842 if (config != NULL && spa->spa_config) { 2843 VERIFY(nvlist_dup(spa->spa_config, config, 2844 KM_SLEEP) == 0); 2845 VERIFY(nvlist_add_nvlist(*config, 2846 ZPOOL_CONFIG_LOAD_INFO, 2847 spa->spa_load_info) == 0); 2848 } 2849 spa_unload(spa); 2850 spa_deactivate(spa); 2851 spa->spa_last_open_failed = error; 2852 if (locked) 2853 mutex_exit(&spa_namespace_lock); 2854 *spapp = NULL; 2855 return (error); 2856 } 2857 } 2858 2859 spa_open_ref(spa, tag); 2860 2861 if (config != NULL) 2862 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2863 2864 /* 2865 * If we've recovered the pool, pass back any information we 2866 * gathered while doing the load. 2867 */ 2868 if (state == SPA_LOAD_RECOVER) { 2869 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 2870 spa->spa_load_info) == 0); 2871 } 2872 2873 if (locked) { 2874 spa->spa_last_open_failed = 0; 2875 spa->spa_last_ubsync_txg = 0; 2876 spa->spa_load_txg = 0; 2877 mutex_exit(&spa_namespace_lock); 2878#ifdef __FreeBSD__ 2879#ifdef _KERNEL 2880 if (firstopen) 2881 zvol_create_minors(pool); 2882#endif 2883#endif 2884 } 2885 2886 *spapp = spa; 2887 2888 return (0); 2889} 2890 2891int 2892spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2893 nvlist_t **config) 2894{ 2895 return (spa_open_common(name, spapp, tag, policy, config)); 2896} 2897 2898int 2899spa_open(const char *name, spa_t **spapp, void *tag) 2900{ 2901 return (spa_open_common(name, spapp, tag, NULL, NULL)); 2902} 2903 2904/* 2905 * Lookup the given spa_t, incrementing the inject count in the process, 2906 * preventing it from being exported or destroyed. 2907 */ 2908spa_t * 2909spa_inject_addref(char *name) 2910{ 2911 spa_t *spa; 2912 2913 mutex_enter(&spa_namespace_lock); 2914 if ((spa = spa_lookup(name)) == NULL) { 2915 mutex_exit(&spa_namespace_lock); 2916 return (NULL); 2917 } 2918 spa->spa_inject_ref++; 2919 mutex_exit(&spa_namespace_lock); 2920 2921 return (spa); 2922} 2923 2924void 2925spa_inject_delref(spa_t *spa) 2926{ 2927 mutex_enter(&spa_namespace_lock); 2928 spa->spa_inject_ref--; 2929 mutex_exit(&spa_namespace_lock); 2930} 2931 2932/* 2933 * Add spares device information to the nvlist. 2934 */ 2935static void 2936spa_add_spares(spa_t *spa, nvlist_t *config) 2937{ 2938 nvlist_t **spares; 2939 uint_t i, nspares; 2940 nvlist_t *nvroot; 2941 uint64_t guid; 2942 vdev_stat_t *vs; 2943 uint_t vsc; 2944 uint64_t pool; 2945 2946 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 2947 2948 if (spa->spa_spares.sav_count == 0) 2949 return; 2950 2951 VERIFY(nvlist_lookup_nvlist(config, 2952 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 2953 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 2954 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2955 if (nspares != 0) { 2956 VERIFY(nvlist_add_nvlist_array(nvroot, 2957 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 2958 VERIFY(nvlist_lookup_nvlist_array(nvroot, 2959 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 2960 2961 /* 2962 * Go through and find any spares which have since been 2963 * repurposed as an active spare. If this is the case, update 2964 * their status appropriately. 2965 */ 2966 for (i = 0; i < nspares; i++) { 2967 VERIFY(nvlist_lookup_uint64(spares[i], 2968 ZPOOL_CONFIG_GUID, &guid) == 0); 2969 if (spa_spare_exists(guid, &pool, NULL) && 2970 pool != 0ULL) { 2971 VERIFY(nvlist_lookup_uint64_array( 2972 spares[i], ZPOOL_CONFIG_VDEV_STATS, 2973 (uint64_t **)&vs, &vsc) == 0); 2974 vs->vs_state = VDEV_STATE_CANT_OPEN; 2975 vs->vs_aux = VDEV_AUX_SPARED; 2976 } 2977 } 2978 } 2979} 2980 2981/* 2982 * Add l2cache device information to the nvlist, including vdev stats. 2983 */ 2984static void 2985spa_add_l2cache(spa_t *spa, nvlist_t *config) 2986{ 2987 nvlist_t **l2cache; 2988 uint_t i, j, nl2cache; 2989 nvlist_t *nvroot; 2990 uint64_t guid; 2991 vdev_t *vd; 2992 vdev_stat_t *vs; 2993 uint_t vsc; 2994 2995 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 2996 2997 if (spa->spa_l2cache.sav_count == 0) 2998 return; 2999 3000 VERIFY(nvlist_lookup_nvlist(config, 3001 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3002 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3003 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3004 if (nl2cache != 0) { 3005 VERIFY(nvlist_add_nvlist_array(nvroot, 3006 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3007 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3008 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3009 3010 /* 3011 * Update level 2 cache device stats. 3012 */ 3013 3014 for (i = 0; i < nl2cache; i++) { 3015 VERIFY(nvlist_lookup_uint64(l2cache[i], 3016 ZPOOL_CONFIG_GUID, &guid) == 0); 3017 3018 vd = NULL; 3019 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3020 if (guid == 3021 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3022 vd = spa->spa_l2cache.sav_vdevs[j]; 3023 break; 3024 } 3025 } 3026 ASSERT(vd != NULL); 3027 3028 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3029 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3030 == 0); 3031 vdev_get_stats(vd, vs); 3032 } 3033 } 3034} 3035 3036static void 3037spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3038{ 3039 nvlist_t *features; 3040 zap_cursor_t zc; 3041 zap_attribute_t za; 3042 3043 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3044 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3045 3046 if (spa->spa_feat_for_read_obj != 0) { 3047 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3048 spa->spa_feat_for_read_obj); 3049 zap_cursor_retrieve(&zc, &za) == 0; 3050 zap_cursor_advance(&zc)) { 3051 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3052 za.za_num_integers == 1); 3053 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3054 za.za_first_integer)); 3055 } 3056 zap_cursor_fini(&zc); 3057 } 3058 3059 if (spa->spa_feat_for_write_obj != 0) { 3060 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3061 spa->spa_feat_for_write_obj); 3062 zap_cursor_retrieve(&zc, &za) == 0; 3063 zap_cursor_advance(&zc)) { 3064 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3065 za.za_num_integers == 1); 3066 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3067 za.za_first_integer)); 3068 } 3069 zap_cursor_fini(&zc); 3070 } 3071 3072 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3073 features) == 0); 3074 nvlist_free(features); 3075} 3076 3077int 3078spa_get_stats(const char *name, nvlist_t **config, 3079 char *altroot, size_t buflen) 3080{ 3081 int error; 3082 spa_t *spa; 3083 3084 *config = NULL; 3085 error = spa_open_common(name, &spa, FTAG, NULL, config); 3086 3087 if (spa != NULL) { 3088 /* 3089 * This still leaves a window of inconsistency where the spares 3090 * or l2cache devices could change and the config would be 3091 * self-inconsistent. 3092 */ 3093 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3094 3095 if (*config != NULL) { 3096 uint64_t loadtimes[2]; 3097 3098 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3099 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3100 VERIFY(nvlist_add_uint64_array(*config, 3101 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3102 3103 VERIFY(nvlist_add_uint64(*config, 3104 ZPOOL_CONFIG_ERRCOUNT, 3105 spa_get_errlog_size(spa)) == 0); 3106 3107 if (spa_suspended(spa)) 3108 VERIFY(nvlist_add_uint64(*config, 3109 ZPOOL_CONFIG_SUSPENDED, 3110 spa->spa_failmode) == 0); 3111 3112 spa_add_spares(spa, *config); 3113 spa_add_l2cache(spa, *config); 3114 spa_add_feature_stats(spa, *config); 3115 } 3116 } 3117 3118 /* 3119 * We want to get the alternate root even for faulted pools, so we cheat 3120 * and call spa_lookup() directly. 3121 */ 3122 if (altroot) { 3123 if (spa == NULL) { 3124 mutex_enter(&spa_namespace_lock); 3125 spa = spa_lookup(name); 3126 if (spa) 3127 spa_altroot(spa, altroot, buflen); 3128 else 3129 altroot[0] = '\0'; 3130 spa = NULL; 3131 mutex_exit(&spa_namespace_lock); 3132 } else { 3133 spa_altroot(spa, altroot, buflen); 3134 } 3135 } 3136 3137 if (spa != NULL) { 3138 spa_config_exit(spa, SCL_CONFIG, FTAG); 3139 spa_close(spa, FTAG); 3140 } 3141 3142 return (error); 3143} 3144 3145/* 3146 * Validate that the auxiliary device array is well formed. We must have an 3147 * array of nvlists, each which describes a valid leaf vdev. If this is an 3148 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3149 * specified, as long as they are well-formed. 3150 */ 3151static int 3152spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3153 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3154 vdev_labeltype_t label) 3155{ 3156 nvlist_t **dev; 3157 uint_t i, ndev; 3158 vdev_t *vd; 3159 int error; 3160 3161 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3162 3163 /* 3164 * It's acceptable to have no devs specified. 3165 */ 3166 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3167 return (0); 3168 3169 if (ndev == 0) 3170 return (EINVAL); 3171 3172 /* 3173 * Make sure the pool is formatted with a version that supports this 3174 * device type. 3175 */ 3176 if (spa_version(spa) < version) 3177 return (ENOTSUP); 3178 3179 /* 3180 * Set the pending device list so we correctly handle device in-use 3181 * checking. 3182 */ 3183 sav->sav_pending = dev; 3184 sav->sav_npending = ndev; 3185 3186 for (i = 0; i < ndev; i++) { 3187 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3188 mode)) != 0) 3189 goto out; 3190 3191 if (!vd->vdev_ops->vdev_op_leaf) { 3192 vdev_free(vd); 3193 error = EINVAL; 3194 goto out; 3195 } 3196 3197 /* 3198 * The L2ARC currently only supports disk devices in 3199 * kernel context. For user-level testing, we allow it. 3200 */ 3201#ifdef _KERNEL 3202 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3203 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3204 error = ENOTBLK; 3205 vdev_free(vd); 3206 goto out; 3207 } 3208#endif 3209 vd->vdev_top = vd; 3210 3211 if ((error = vdev_open(vd)) == 0 && 3212 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3213 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3214 vd->vdev_guid) == 0); 3215 } 3216 3217 vdev_free(vd); 3218 3219 if (error && 3220 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3221 goto out; 3222 else 3223 error = 0; 3224 } 3225 3226out: 3227 sav->sav_pending = NULL; 3228 sav->sav_npending = 0; 3229 return (error); 3230} 3231 3232static int 3233spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3234{ 3235 int error; 3236 3237 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3238 3239 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3240 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3241 VDEV_LABEL_SPARE)) != 0) { 3242 return (error); 3243 } 3244 3245 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3246 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3247 VDEV_LABEL_L2CACHE)); 3248} 3249 3250static void 3251spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3252 const char *config) 3253{ 3254 int i; 3255 3256 if (sav->sav_config != NULL) { 3257 nvlist_t **olddevs; 3258 uint_t oldndevs; 3259 nvlist_t **newdevs; 3260 3261 /* 3262 * Generate new dev list by concatentating with the 3263 * current dev list. 3264 */ 3265 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3266 &olddevs, &oldndevs) == 0); 3267 3268 newdevs = kmem_alloc(sizeof (void *) * 3269 (ndevs + oldndevs), KM_SLEEP); 3270 for (i = 0; i < oldndevs; i++) 3271 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3272 KM_SLEEP) == 0); 3273 for (i = 0; i < ndevs; i++) 3274 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3275 KM_SLEEP) == 0); 3276 3277 VERIFY(nvlist_remove(sav->sav_config, config, 3278 DATA_TYPE_NVLIST_ARRAY) == 0); 3279 3280 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3281 config, newdevs, ndevs + oldndevs) == 0); 3282 for (i = 0; i < oldndevs + ndevs; i++) 3283 nvlist_free(newdevs[i]); 3284 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3285 } else { 3286 /* 3287 * Generate a new dev list. 3288 */ 3289 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3290 KM_SLEEP) == 0); 3291 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3292 devs, ndevs) == 0); 3293 } 3294} 3295 3296/* 3297 * Stop and drop level 2 ARC devices 3298 */ 3299void 3300spa_l2cache_drop(spa_t *spa) 3301{ 3302 vdev_t *vd; 3303 int i; 3304 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3305 3306 for (i = 0; i < sav->sav_count; i++) { 3307 uint64_t pool; 3308 3309 vd = sav->sav_vdevs[i]; 3310 ASSERT(vd != NULL); 3311 3312 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3313 pool != 0ULL && l2arc_vdev_present(vd)) 3314 l2arc_remove_vdev(vd); 3315 } 3316} 3317 3318/* 3319 * Pool Creation 3320 */ 3321int 3322spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3323 const char *history_str, nvlist_t *zplprops) 3324{ 3325 spa_t *spa; 3326 char *altroot = NULL; 3327 vdev_t *rvd; 3328 dsl_pool_t *dp; 3329 dmu_tx_t *tx; 3330 int error = 0; 3331 uint64_t txg = TXG_INITIAL; 3332 nvlist_t **spares, **l2cache; 3333 uint_t nspares, nl2cache; 3334 uint64_t version, obj; 3335 boolean_t has_features; 3336 3337 /* 3338 * If this pool already exists, return failure. 3339 */ 3340 mutex_enter(&spa_namespace_lock); 3341 if (spa_lookup(pool) != NULL) { 3342 mutex_exit(&spa_namespace_lock); 3343 return (EEXIST); 3344 } 3345 3346 /* 3347 * Allocate a new spa_t structure. 3348 */ 3349 (void) nvlist_lookup_string(props, 3350 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3351 spa = spa_add(pool, NULL, altroot); 3352 spa_activate(spa, spa_mode_global); 3353 3354 if (props && (error = spa_prop_validate(spa, props))) { 3355 spa_deactivate(spa); 3356 spa_remove(spa); 3357 mutex_exit(&spa_namespace_lock); 3358 return (error); 3359 } 3360 3361 has_features = B_FALSE; 3362 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3363 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3364 if (zpool_prop_feature(nvpair_name(elem))) 3365 has_features = B_TRUE; 3366 } 3367 3368 if (has_features || nvlist_lookup_uint64(props, 3369 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3370 version = SPA_VERSION; 3371 } 3372 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3373 3374 spa->spa_first_txg = txg; 3375 spa->spa_uberblock.ub_txg = txg - 1; 3376 spa->spa_uberblock.ub_version = version; 3377 spa->spa_ubsync = spa->spa_uberblock; 3378 3379 /* 3380 * Create "The Godfather" zio to hold all async IOs 3381 */ 3382 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 3383 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 3384 3385 /* 3386 * Create the root vdev. 3387 */ 3388 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3389 3390 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3391 3392 ASSERT(error != 0 || rvd != NULL); 3393 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3394 3395 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3396 error = EINVAL; 3397 3398 if (error == 0 && 3399 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3400 (error = spa_validate_aux(spa, nvroot, txg, 3401 VDEV_ALLOC_ADD)) == 0) { 3402 for (int c = 0; c < rvd->vdev_children; c++) { 3403 vdev_metaslab_set_size(rvd->vdev_child[c]); 3404 vdev_expand(rvd->vdev_child[c], txg); 3405 } 3406 } 3407 3408 spa_config_exit(spa, SCL_ALL, FTAG); 3409 3410 if (error != 0) { 3411 spa_unload(spa); 3412 spa_deactivate(spa); 3413 spa_remove(spa); 3414 mutex_exit(&spa_namespace_lock); 3415 return (error); 3416 } 3417 3418 /* 3419 * Get the list of spares, if specified. 3420 */ 3421 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3422 &spares, &nspares) == 0) { 3423 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3424 KM_SLEEP) == 0); 3425 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3426 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3427 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3428 spa_load_spares(spa); 3429 spa_config_exit(spa, SCL_ALL, FTAG); 3430 spa->spa_spares.sav_sync = B_TRUE; 3431 } 3432 3433 /* 3434 * Get the list of level 2 cache devices, if specified. 3435 */ 3436 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3437 &l2cache, &nl2cache) == 0) { 3438 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3439 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3440 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3441 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3442 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3443 spa_load_l2cache(spa); 3444 spa_config_exit(spa, SCL_ALL, FTAG); 3445 spa->spa_l2cache.sav_sync = B_TRUE; 3446 } 3447 3448 spa->spa_is_initializing = B_TRUE; 3449 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3450 spa->spa_meta_objset = dp->dp_meta_objset; 3451 spa->spa_is_initializing = B_FALSE; 3452 3453 /* 3454 * Create DDTs (dedup tables). 3455 */ 3456 ddt_create(spa); 3457 3458 spa_update_dspace(spa); 3459 3460 tx = dmu_tx_create_assigned(dp, txg); 3461 3462 /* 3463 * Create the pool config object. 3464 */ 3465 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3466 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3467 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3468 3469 if (zap_add(spa->spa_meta_objset, 3470 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3471 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3472 cmn_err(CE_PANIC, "failed to add pool config"); 3473 } 3474 3475 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3476 spa_feature_create_zap_objects(spa, tx); 3477 3478 if (zap_add(spa->spa_meta_objset, 3479 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3480 sizeof (uint64_t), 1, &version, tx) != 0) { 3481 cmn_err(CE_PANIC, "failed to add pool version"); 3482 } 3483 3484 /* Newly created pools with the right version are always deflated. */ 3485 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3486 spa->spa_deflate = TRUE; 3487 if (zap_add(spa->spa_meta_objset, 3488 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3489 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3490 cmn_err(CE_PANIC, "failed to add deflate"); 3491 } 3492 } 3493 3494 /* 3495 * Create the deferred-free bpobj. Turn off compression 3496 * because sync-to-convergence takes longer if the blocksize 3497 * keeps changing. 3498 */ 3499 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3500 dmu_object_set_compress(spa->spa_meta_objset, obj, 3501 ZIO_COMPRESS_OFF, tx); 3502 if (zap_add(spa->spa_meta_objset, 3503 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3504 sizeof (uint64_t), 1, &obj, tx) != 0) { 3505 cmn_err(CE_PANIC, "failed to add bpobj"); 3506 } 3507 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3508 spa->spa_meta_objset, obj)); 3509 3510 /* 3511 * Create the pool's history object. 3512 */ 3513 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3514 spa_history_create_obj(spa, tx); 3515 3516 /* 3517 * Set pool properties. 3518 */ 3519 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3520 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3521 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3522 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3523 3524 if (props != NULL) { 3525 spa_configfile_set(spa, props, B_FALSE); 3526 spa_sync_props(spa, props, tx); 3527 } 3528 3529 dmu_tx_commit(tx); 3530 3531 spa->spa_sync_on = B_TRUE; 3532 txg_sync_start(spa->spa_dsl_pool); 3533 3534 /* 3535 * We explicitly wait for the first transaction to complete so that our 3536 * bean counters are appropriately updated. 3537 */ 3538 txg_wait_synced(spa->spa_dsl_pool, txg); 3539 3540 spa_config_sync(spa, B_FALSE, B_TRUE); 3541 3542 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL) 3543 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE); 3544 spa_history_log_version(spa, LOG_POOL_CREATE); 3545 3546 spa->spa_minref = refcount_count(&spa->spa_refcount); 3547 3548 mutex_exit(&spa_namespace_lock); 3549 3550 return (0); 3551} 3552 3553#ifdef _KERNEL 3554#if defined(sun) 3555/* 3556 * Get the root pool information from the root disk, then import the root pool 3557 * during the system boot up time. 3558 */ 3559extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3560 3561static nvlist_t * 3562spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3563{ 3564 nvlist_t *config; 3565 nvlist_t *nvtop, *nvroot; 3566 uint64_t pgid; 3567 3568 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3569 return (NULL); 3570 3571 /* 3572 * Add this top-level vdev to the child array. 3573 */ 3574 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3575 &nvtop) == 0); 3576 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3577 &pgid) == 0); 3578 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3579 3580 /* 3581 * Put this pool's top-level vdevs into a root vdev. 3582 */ 3583 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3584 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3585 VDEV_TYPE_ROOT) == 0); 3586 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3587 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3588 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3589 &nvtop, 1) == 0); 3590 3591 /* 3592 * Replace the existing vdev_tree with the new root vdev in 3593 * this pool's configuration (remove the old, add the new). 3594 */ 3595 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3596 nvlist_free(nvroot); 3597 return (config); 3598} 3599 3600/* 3601 * Walk the vdev tree and see if we can find a device with "better" 3602 * configuration. A configuration is "better" if the label on that 3603 * device has a more recent txg. 3604 */ 3605static void 3606spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3607{ 3608 for (int c = 0; c < vd->vdev_children; c++) 3609 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3610 3611 if (vd->vdev_ops->vdev_op_leaf) { 3612 nvlist_t *label; 3613 uint64_t label_txg; 3614 3615 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3616 &label) != 0) 3617 return; 3618 3619 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3620 &label_txg) == 0); 3621 3622 /* 3623 * Do we have a better boot device? 3624 */ 3625 if (label_txg > *txg) { 3626 *txg = label_txg; 3627 *avd = vd; 3628 } 3629 nvlist_free(label); 3630 } 3631} 3632 3633/* 3634 * Import a root pool. 3635 * 3636 * For x86. devpath_list will consist of devid and/or physpath name of 3637 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3638 * The GRUB "findroot" command will return the vdev we should boot. 3639 * 3640 * For Sparc, devpath_list consists the physpath name of the booting device 3641 * no matter the rootpool is a single device pool or a mirrored pool. 3642 * e.g. 3643 * "/pci@1f,0/ide@d/disk@0,0:a" 3644 */ 3645int 3646spa_import_rootpool(char *devpath, char *devid) 3647{ 3648 spa_t *spa; 3649 vdev_t *rvd, *bvd, *avd = NULL; 3650 nvlist_t *config, *nvtop; 3651 uint64_t guid, txg; 3652 char *pname; 3653 int error; 3654 3655 /* 3656 * Read the label from the boot device and generate a configuration. 3657 */ 3658 config = spa_generate_rootconf(devpath, devid, &guid); 3659#if defined(_OBP) && defined(_KERNEL) 3660 if (config == NULL) { 3661 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3662 /* iscsi boot */ 3663 get_iscsi_bootpath_phy(devpath); 3664 config = spa_generate_rootconf(devpath, devid, &guid); 3665 } 3666 } 3667#endif 3668 if (config == NULL) { 3669 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3670 devpath); 3671 return (EIO); 3672 } 3673 3674 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3675 &pname) == 0); 3676 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3677 3678 mutex_enter(&spa_namespace_lock); 3679 if ((spa = spa_lookup(pname)) != NULL) { 3680 /* 3681 * Remove the existing root pool from the namespace so that we 3682 * can replace it with the correct config we just read in. 3683 */ 3684 spa_remove(spa); 3685 } 3686 3687 spa = spa_add(pname, config, NULL); 3688 spa->spa_is_root = B_TRUE; 3689 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3690 3691 /* 3692 * Build up a vdev tree based on the boot device's label config. 3693 */ 3694 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3695 &nvtop) == 0); 3696 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3697 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3698 VDEV_ALLOC_ROOTPOOL); 3699 spa_config_exit(spa, SCL_ALL, FTAG); 3700 if (error) { 3701 mutex_exit(&spa_namespace_lock); 3702 nvlist_free(config); 3703 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3704 pname); 3705 return (error); 3706 } 3707 3708 /* 3709 * Get the boot vdev. 3710 */ 3711 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3712 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3713 (u_longlong_t)guid); 3714 error = ENOENT; 3715 goto out; 3716 } 3717 3718 /* 3719 * Determine if there is a better boot device. 3720 */ 3721 avd = bvd; 3722 spa_alt_rootvdev(rvd, &avd, &txg); 3723 if (avd != bvd) { 3724 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3725 "try booting from '%s'", avd->vdev_path); 3726 error = EINVAL; 3727 goto out; 3728 } 3729 3730 /* 3731 * If the boot device is part of a spare vdev then ensure that 3732 * we're booting off the active spare. 3733 */ 3734 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3735 !bvd->vdev_isspare) { 3736 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3737 "try booting from '%s'", 3738 bvd->vdev_parent-> 3739 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3740 error = EINVAL; 3741 goto out; 3742 } 3743 3744 error = 0; 3745 spa_history_log_version(spa, LOG_POOL_IMPORT); 3746out: 3747 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3748 vdev_free(rvd); 3749 spa_config_exit(spa, SCL_ALL, FTAG); 3750 mutex_exit(&spa_namespace_lock); 3751 3752 nvlist_free(config); 3753 return (error); 3754} 3755 3756#else 3757 3758extern int 3759vdev_geom_read_pool_label(const char *name, nvlist_t **config); 3760 3761static nvlist_t * 3762spa_generate_rootconf(const char *name) 3763{ 3764 nvlist_t *config; 3765 nvlist_t *nvtop, *nvroot; 3766 uint64_t pgid; 3767 3768 if (vdev_geom_read_pool_label(name, &config) != 0) 3769 return (NULL); 3770 3771 /* 3772 * Add this top-level vdev to the child array. 3773 */ 3774 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3775 &nvtop) == 0); 3776 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3777 &pgid) == 0); 3778 3779 /* 3780 * Put this pool's top-level vdevs into a root vdev. 3781 */ 3782 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3783 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3784 VDEV_TYPE_ROOT) == 0); 3785 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3786 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3787 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3788 &nvtop, 1) == 0); 3789 3790 /* 3791 * Replace the existing vdev_tree with the new root vdev in 3792 * this pool's configuration (remove the old, add the new). 3793 */ 3794 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3795 nvlist_free(nvroot); 3796 return (config); 3797} 3798 3799int 3800spa_import_rootpool(const char *name) 3801{ 3802 spa_t *spa; 3803 vdev_t *rvd, *bvd, *avd = NULL; 3804 nvlist_t *config, *nvtop; 3805 uint64_t txg; 3806 char *pname; 3807 int error; 3808 3809 /* 3810 * Read the label from the boot device and generate a configuration. 3811 */ 3812 config = spa_generate_rootconf(name); 3813 if (config == NULL) { 3814 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 3815 name); 3816 return (EIO); 3817 } 3818 3819 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3820 &pname) == 0 && strcmp(name, pname) == 0); 3821 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3822 3823 mutex_enter(&spa_namespace_lock); 3824 if ((spa = spa_lookup(pname)) != NULL) { 3825 /* 3826 * Remove the existing root pool from the namespace so that we 3827 * can replace it with the correct config we just read in. 3828 */ 3829 spa_remove(spa); 3830 } 3831 spa = spa_add(pname, config, NULL); 3832 spa->spa_is_root = B_TRUE; 3833 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3834 3835 /* 3836 * Build up a vdev tree based on the boot device's label config. 3837 */ 3838 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3839 &nvtop) == 0); 3840 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3841 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3842 VDEV_ALLOC_ROOTPOOL); 3843 spa_config_exit(spa, SCL_ALL, FTAG); 3844 if (error) { 3845 mutex_exit(&spa_namespace_lock); 3846 nvlist_free(config); 3847 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3848 pname); 3849 return (error); 3850 } 3851 3852 error = 0; 3853 spa_history_log_version(spa, LOG_POOL_IMPORT); 3854out: 3855 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3856 vdev_free(rvd); 3857 spa_config_exit(spa, SCL_ALL, FTAG); 3858 mutex_exit(&spa_namespace_lock); 3859 3860 return (error); 3861} 3862 3863#endif /* sun */ 3864#endif 3865 3866/* 3867 * Import a non-root pool into the system. 3868 */ 3869int 3870spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 3871{ 3872 spa_t *spa; 3873 char *altroot = NULL; 3874 spa_load_state_t state = SPA_LOAD_IMPORT; 3875 zpool_rewind_policy_t policy; 3876 uint64_t mode = spa_mode_global; 3877 uint64_t readonly = B_FALSE; 3878 int error; 3879 nvlist_t *nvroot; 3880 nvlist_t **spares, **l2cache; 3881 uint_t nspares, nl2cache; 3882 3883 /* 3884 * If a pool with this name exists, return failure. 3885 */ 3886 mutex_enter(&spa_namespace_lock); 3887 if (spa_lookup(pool) != NULL) { 3888 mutex_exit(&spa_namespace_lock); 3889 return (EEXIST); 3890 } 3891 3892 /* 3893 * Create and initialize the spa structure. 3894 */ 3895 (void) nvlist_lookup_string(props, 3896 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3897 (void) nvlist_lookup_uint64(props, 3898 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 3899 if (readonly) 3900 mode = FREAD; 3901 spa = spa_add(pool, config, altroot); 3902 spa->spa_import_flags = flags; 3903 3904 /* 3905 * Verbatim import - Take a pool and insert it into the namespace 3906 * as if it had been loaded at boot. 3907 */ 3908 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 3909 if (props != NULL) 3910 spa_configfile_set(spa, props, B_FALSE); 3911 3912 spa_config_sync(spa, B_FALSE, B_TRUE); 3913 3914 mutex_exit(&spa_namespace_lock); 3915 spa_history_log_version(spa, LOG_POOL_IMPORT); 3916 3917 return (0); 3918 } 3919 3920 spa_activate(spa, mode); 3921 3922 /* 3923 * Don't start async tasks until we know everything is healthy. 3924 */ 3925 spa_async_suspend(spa); 3926 3927 zpool_get_rewind_policy(config, &policy); 3928 if (policy.zrp_request & ZPOOL_DO_REWIND) 3929 state = SPA_LOAD_RECOVER; 3930 3931 /* 3932 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 3933 * because the user-supplied config is actually the one to trust when 3934 * doing an import. 3935 */ 3936 if (state != SPA_LOAD_RECOVER) 3937 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 3938 3939 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 3940 policy.zrp_request); 3941 3942 /* 3943 * Propagate anything learned while loading the pool and pass it 3944 * back to caller (i.e. rewind info, missing devices, etc). 3945 */ 3946 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 3947 spa->spa_load_info) == 0); 3948 3949 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3950 /* 3951 * Toss any existing sparelist, as it doesn't have any validity 3952 * anymore, and conflicts with spa_has_spare(). 3953 */ 3954 if (spa->spa_spares.sav_config) { 3955 nvlist_free(spa->spa_spares.sav_config); 3956 spa->spa_spares.sav_config = NULL; 3957 spa_load_spares(spa); 3958 } 3959 if (spa->spa_l2cache.sav_config) { 3960 nvlist_free(spa->spa_l2cache.sav_config); 3961 spa->spa_l2cache.sav_config = NULL; 3962 spa_load_l2cache(spa); 3963 } 3964 3965 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3966 &nvroot) == 0); 3967 if (error == 0) 3968 error = spa_validate_aux(spa, nvroot, -1ULL, 3969 VDEV_ALLOC_SPARE); 3970 if (error == 0) 3971 error = spa_validate_aux(spa, nvroot, -1ULL, 3972 VDEV_ALLOC_L2CACHE); 3973 spa_config_exit(spa, SCL_ALL, FTAG); 3974 3975 if (props != NULL) 3976 spa_configfile_set(spa, props, B_FALSE); 3977 3978 if (error != 0 || (props && spa_writeable(spa) && 3979 (error = spa_prop_set(spa, props)))) { 3980 spa_unload(spa); 3981 spa_deactivate(spa); 3982 spa_remove(spa); 3983 mutex_exit(&spa_namespace_lock); 3984 return (error); 3985 } 3986 3987 spa_async_resume(spa); 3988 3989 /* 3990 * Override any spares and level 2 cache devices as specified by 3991 * the user, as these may have correct device names/devids, etc. 3992 */ 3993 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3994 &spares, &nspares) == 0) { 3995 if (spa->spa_spares.sav_config) 3996 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 3997 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 3998 else 3999 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4000 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4001 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4002 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4003 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4004 spa_load_spares(spa); 4005 spa_config_exit(spa, SCL_ALL, FTAG); 4006 spa->spa_spares.sav_sync = B_TRUE; 4007 } 4008 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4009 &l2cache, &nl2cache) == 0) { 4010 if (spa->spa_l2cache.sav_config) 4011 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4012 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4013 else 4014 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4015 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4016 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4017 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4018 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4019 spa_load_l2cache(spa); 4020 spa_config_exit(spa, SCL_ALL, FTAG); 4021 spa->spa_l2cache.sav_sync = B_TRUE; 4022 } 4023 4024 /* 4025 * Check for any removed devices. 4026 */ 4027 if (spa->spa_autoreplace) { 4028 spa_aux_check_removed(&spa->spa_spares); 4029 spa_aux_check_removed(&spa->spa_l2cache); 4030 } 4031 4032 if (spa_writeable(spa)) { 4033 /* 4034 * Update the config cache to include the newly-imported pool. 4035 */ 4036 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4037 } 4038 4039 /* 4040 * It's possible that the pool was expanded while it was exported. 4041 * We kick off an async task to handle this for us. 4042 */ 4043 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4044 4045 mutex_exit(&spa_namespace_lock); 4046 spa_history_log_version(spa, LOG_POOL_IMPORT); 4047 4048#ifdef __FreeBSD__ 4049#ifdef _KERNEL 4050 zvol_create_minors(pool); 4051#endif 4052#endif 4053 return (0); 4054} 4055 4056nvlist_t * 4057spa_tryimport(nvlist_t *tryconfig) 4058{ 4059 nvlist_t *config = NULL; 4060 char *poolname; 4061 spa_t *spa; 4062 uint64_t state; 4063 int error; 4064 4065 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4066 return (NULL); 4067 4068 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4069 return (NULL); 4070 4071 /* 4072 * Create and initialize the spa structure. 4073 */ 4074 mutex_enter(&spa_namespace_lock); 4075 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4076 spa_activate(spa, FREAD); 4077 4078 /* 4079 * Pass off the heavy lifting to spa_load(). 4080 * Pass TRUE for mosconfig because the user-supplied config 4081 * is actually the one to trust when doing an import. 4082 */ 4083 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4084 4085 /* 4086 * If 'tryconfig' was at least parsable, return the current config. 4087 */ 4088 if (spa->spa_root_vdev != NULL) { 4089 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4090 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4091 poolname) == 0); 4092 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4093 state) == 0); 4094 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4095 spa->spa_uberblock.ub_timestamp) == 0); 4096 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4097 spa->spa_load_info) == 0); 4098 4099 /* 4100 * If the bootfs property exists on this pool then we 4101 * copy it out so that external consumers can tell which 4102 * pools are bootable. 4103 */ 4104 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4105 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4106 4107 /* 4108 * We have to play games with the name since the 4109 * pool was opened as TRYIMPORT_NAME. 4110 */ 4111 if (dsl_dsobj_to_dsname(spa_name(spa), 4112 spa->spa_bootfs, tmpname) == 0) { 4113 char *cp; 4114 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4115 4116 cp = strchr(tmpname, '/'); 4117 if (cp == NULL) { 4118 (void) strlcpy(dsname, tmpname, 4119 MAXPATHLEN); 4120 } else { 4121 (void) snprintf(dsname, MAXPATHLEN, 4122 "%s/%s", poolname, ++cp); 4123 } 4124 VERIFY(nvlist_add_string(config, 4125 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4126 kmem_free(dsname, MAXPATHLEN); 4127 } 4128 kmem_free(tmpname, MAXPATHLEN); 4129 } 4130 4131 /* 4132 * Add the list of hot spares and level 2 cache devices. 4133 */ 4134 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4135 spa_add_spares(spa, config); 4136 spa_add_l2cache(spa, config); 4137 spa_config_exit(spa, SCL_CONFIG, FTAG); 4138 } 4139 4140 spa_unload(spa); 4141 spa_deactivate(spa); 4142 spa_remove(spa); 4143 mutex_exit(&spa_namespace_lock); 4144 4145 return (config); 4146} 4147 4148/* 4149 * Pool export/destroy 4150 * 4151 * The act of destroying or exporting a pool is very simple. We make sure there 4152 * is no more pending I/O and any references to the pool are gone. Then, we 4153 * update the pool state and sync all the labels to disk, removing the 4154 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4155 * we don't sync the labels or remove the configuration cache. 4156 */ 4157static int 4158spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4159 boolean_t force, boolean_t hardforce) 4160{ 4161 spa_t *spa; 4162 4163 if (oldconfig) 4164 *oldconfig = NULL; 4165 4166 if (!(spa_mode_global & FWRITE)) 4167 return (EROFS); 4168 4169 mutex_enter(&spa_namespace_lock); 4170 if ((spa = spa_lookup(pool)) == NULL) { 4171 mutex_exit(&spa_namespace_lock); 4172 return (ENOENT); 4173 } 4174 4175 /* 4176 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4177 * reacquire the namespace lock, and see if we can export. 4178 */ 4179 spa_open_ref(spa, FTAG); 4180 mutex_exit(&spa_namespace_lock); 4181 spa_async_suspend(spa); 4182 mutex_enter(&spa_namespace_lock); 4183 spa_close(spa, FTAG); 4184 4185 /* 4186 * The pool will be in core if it's openable, 4187 * in which case we can modify its state. 4188 */ 4189 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4190 /* 4191 * Objsets may be open only because they're dirty, so we 4192 * have to force it to sync before checking spa_refcnt. 4193 */ 4194 txg_wait_synced(spa->spa_dsl_pool, 0); 4195 4196 /* 4197 * A pool cannot be exported or destroyed if there are active 4198 * references. If we are resetting a pool, allow references by 4199 * fault injection handlers. 4200 */ 4201 if (!spa_refcount_zero(spa) || 4202 (spa->spa_inject_ref != 0 && 4203 new_state != POOL_STATE_UNINITIALIZED)) { 4204 spa_async_resume(spa); 4205 mutex_exit(&spa_namespace_lock); 4206 return (EBUSY); 4207 } 4208 4209 /* 4210 * A pool cannot be exported if it has an active shared spare. 4211 * This is to prevent other pools stealing the active spare 4212 * from an exported pool. At user's own will, such pool can 4213 * be forcedly exported. 4214 */ 4215 if (!force && new_state == POOL_STATE_EXPORTED && 4216 spa_has_active_shared_spare(spa)) { 4217 spa_async_resume(spa); 4218 mutex_exit(&spa_namespace_lock); 4219 return (EXDEV); 4220 } 4221 4222 /* 4223 * We want this to be reflected on every label, 4224 * so mark them all dirty. spa_unload() will do the 4225 * final sync that pushes these changes out. 4226 */ 4227 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4228 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4229 spa->spa_state = new_state; 4230 spa->spa_final_txg = spa_last_synced_txg(spa) + 4231 TXG_DEFER_SIZE + 1; 4232 vdev_config_dirty(spa->spa_root_vdev); 4233 spa_config_exit(spa, SCL_ALL, FTAG); 4234 } 4235 } 4236 4237 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4238 4239 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4240 spa_unload(spa); 4241 spa_deactivate(spa); 4242 } 4243 4244 if (oldconfig && spa->spa_config) 4245 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4246 4247 if (new_state != POOL_STATE_UNINITIALIZED) { 4248 if (!hardforce) 4249 spa_config_sync(spa, B_TRUE, B_TRUE); 4250 spa_remove(spa); 4251 } 4252 mutex_exit(&spa_namespace_lock); 4253 4254 return (0); 4255} 4256 4257/* 4258 * Destroy a storage pool. 4259 */ 4260int 4261spa_destroy(char *pool) 4262{ 4263 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4264 B_FALSE, B_FALSE)); 4265} 4266 4267/* 4268 * Export a storage pool. 4269 */ 4270int 4271spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4272 boolean_t hardforce) 4273{ 4274 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4275 force, hardforce)); 4276} 4277 4278/* 4279 * Similar to spa_export(), this unloads the spa_t without actually removing it 4280 * from the namespace in any way. 4281 */ 4282int 4283spa_reset(char *pool) 4284{ 4285 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4286 B_FALSE, B_FALSE)); 4287} 4288 4289/* 4290 * ========================================================================== 4291 * Device manipulation 4292 * ========================================================================== 4293 */ 4294 4295/* 4296 * Add a device to a storage pool. 4297 */ 4298int 4299spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4300{ 4301 uint64_t txg, id; 4302 int error; 4303 vdev_t *rvd = spa->spa_root_vdev; 4304 vdev_t *vd, *tvd; 4305 nvlist_t **spares, **l2cache; 4306 uint_t nspares, nl2cache; 4307 4308 ASSERT(spa_writeable(spa)); 4309 4310 txg = spa_vdev_enter(spa); 4311 4312 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4313 VDEV_ALLOC_ADD)) != 0) 4314 return (spa_vdev_exit(spa, NULL, txg, error)); 4315 4316 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4317 4318 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4319 &nspares) != 0) 4320 nspares = 0; 4321 4322 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4323 &nl2cache) != 0) 4324 nl2cache = 0; 4325 4326 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4327 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4328 4329 if (vd->vdev_children != 0 && 4330 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4331 return (spa_vdev_exit(spa, vd, txg, error)); 4332 4333 /* 4334 * We must validate the spares and l2cache devices after checking the 4335 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4336 */ 4337 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4338 return (spa_vdev_exit(spa, vd, txg, error)); 4339 4340 /* 4341 * Transfer each new top-level vdev from vd to rvd. 4342 */ 4343 for (int c = 0; c < vd->vdev_children; c++) { 4344 4345 /* 4346 * Set the vdev id to the first hole, if one exists. 4347 */ 4348 for (id = 0; id < rvd->vdev_children; id++) { 4349 if (rvd->vdev_child[id]->vdev_ishole) { 4350 vdev_free(rvd->vdev_child[id]); 4351 break; 4352 } 4353 } 4354 tvd = vd->vdev_child[c]; 4355 vdev_remove_child(vd, tvd); 4356 tvd->vdev_id = id; 4357 vdev_add_child(rvd, tvd); 4358 vdev_config_dirty(tvd); 4359 } 4360 4361 if (nspares != 0) { 4362 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4363 ZPOOL_CONFIG_SPARES); 4364 spa_load_spares(spa); 4365 spa->spa_spares.sav_sync = B_TRUE; 4366 } 4367 4368 if (nl2cache != 0) { 4369 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4370 ZPOOL_CONFIG_L2CACHE); 4371 spa_load_l2cache(spa); 4372 spa->spa_l2cache.sav_sync = B_TRUE; 4373 } 4374 4375 /* 4376 * We have to be careful when adding new vdevs to an existing pool. 4377 * If other threads start allocating from these vdevs before we 4378 * sync the config cache, and we lose power, then upon reboot we may 4379 * fail to open the pool because there are DVAs that the config cache 4380 * can't translate. Therefore, we first add the vdevs without 4381 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4382 * and then let spa_config_update() initialize the new metaslabs. 4383 * 4384 * spa_load() checks for added-but-not-initialized vdevs, so that 4385 * if we lose power at any point in this sequence, the remaining 4386 * steps will be completed the next time we load the pool. 4387 */ 4388 (void) spa_vdev_exit(spa, vd, txg, 0); 4389 4390 mutex_enter(&spa_namespace_lock); 4391 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4392 mutex_exit(&spa_namespace_lock); 4393 4394 return (0); 4395} 4396 4397/* 4398 * Attach a device to a mirror. The arguments are the path to any device 4399 * in the mirror, and the nvroot for the new device. If the path specifies 4400 * a device that is not mirrored, we automatically insert the mirror vdev. 4401 * 4402 * If 'replacing' is specified, the new device is intended to replace the 4403 * existing device; in this case the two devices are made into their own 4404 * mirror using the 'replacing' vdev, which is functionally identical to 4405 * the mirror vdev (it actually reuses all the same ops) but has a few 4406 * extra rules: you can't attach to it after it's been created, and upon 4407 * completion of resilvering, the first disk (the one being replaced) 4408 * is automatically detached. 4409 */ 4410int 4411spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4412{ 4413 uint64_t txg, dtl_max_txg; 4414 vdev_t *rvd = spa->spa_root_vdev; 4415 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4416 vdev_ops_t *pvops; 4417 char *oldvdpath, *newvdpath; 4418 int newvd_isspare; 4419 int error; 4420 4421 ASSERT(spa_writeable(spa)); 4422 4423 txg = spa_vdev_enter(spa); 4424 4425 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4426 4427 if (oldvd == NULL) 4428 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4429 4430 if (!oldvd->vdev_ops->vdev_op_leaf) 4431 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4432 4433 pvd = oldvd->vdev_parent; 4434 4435 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4436 VDEV_ALLOC_ATTACH)) != 0) 4437 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4438 4439 if (newrootvd->vdev_children != 1) 4440 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4441 4442 newvd = newrootvd->vdev_child[0]; 4443 4444 if (!newvd->vdev_ops->vdev_op_leaf) 4445 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4446 4447 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4448 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4449 4450 /* 4451 * Spares can't replace logs 4452 */ 4453 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4454 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4455 4456 if (!replacing) { 4457 /* 4458 * For attach, the only allowable parent is a mirror or the root 4459 * vdev. 4460 */ 4461 if (pvd->vdev_ops != &vdev_mirror_ops && 4462 pvd->vdev_ops != &vdev_root_ops) 4463 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4464 4465 pvops = &vdev_mirror_ops; 4466 } else { 4467 /* 4468 * Active hot spares can only be replaced by inactive hot 4469 * spares. 4470 */ 4471 if (pvd->vdev_ops == &vdev_spare_ops && 4472 oldvd->vdev_isspare && 4473 !spa_has_spare(spa, newvd->vdev_guid)) 4474 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4475 4476 /* 4477 * If the source is a hot spare, and the parent isn't already a 4478 * spare, then we want to create a new hot spare. Otherwise, we 4479 * want to create a replacing vdev. The user is not allowed to 4480 * attach to a spared vdev child unless the 'isspare' state is 4481 * the same (spare replaces spare, non-spare replaces 4482 * non-spare). 4483 */ 4484 if (pvd->vdev_ops == &vdev_replacing_ops && 4485 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4486 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4487 } else if (pvd->vdev_ops == &vdev_spare_ops && 4488 newvd->vdev_isspare != oldvd->vdev_isspare) { 4489 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4490 } 4491 4492 if (newvd->vdev_isspare) 4493 pvops = &vdev_spare_ops; 4494 else 4495 pvops = &vdev_replacing_ops; 4496 } 4497 4498 /* 4499 * Make sure the new device is big enough. 4500 */ 4501 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4502 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4503 4504 /* 4505 * The new device cannot have a higher alignment requirement 4506 * than the top-level vdev. 4507 */ 4508 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4509 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4510 4511 /* 4512 * If this is an in-place replacement, update oldvd's path and devid 4513 * to make it distinguishable from newvd, and unopenable from now on. 4514 */ 4515 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4516 spa_strfree(oldvd->vdev_path); 4517 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4518 KM_SLEEP); 4519 (void) sprintf(oldvd->vdev_path, "%s/%s", 4520 newvd->vdev_path, "old"); 4521 if (oldvd->vdev_devid != NULL) { 4522 spa_strfree(oldvd->vdev_devid); 4523 oldvd->vdev_devid = NULL; 4524 } 4525 } 4526 4527 /* mark the device being resilvered */ 4528 newvd->vdev_resilvering = B_TRUE; 4529 4530 /* 4531 * If the parent is not a mirror, or if we're replacing, insert the new 4532 * mirror/replacing/spare vdev above oldvd. 4533 */ 4534 if (pvd->vdev_ops != pvops) 4535 pvd = vdev_add_parent(oldvd, pvops); 4536 4537 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4538 ASSERT(pvd->vdev_ops == pvops); 4539 ASSERT(oldvd->vdev_parent == pvd); 4540 4541 /* 4542 * Extract the new device from its root and add it to pvd. 4543 */ 4544 vdev_remove_child(newrootvd, newvd); 4545 newvd->vdev_id = pvd->vdev_children; 4546 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4547 vdev_add_child(pvd, newvd); 4548 4549 tvd = newvd->vdev_top; 4550 ASSERT(pvd->vdev_top == tvd); 4551 ASSERT(tvd->vdev_parent == rvd); 4552 4553 vdev_config_dirty(tvd); 4554 4555 /* 4556 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4557 * for any dmu_sync-ed blocks. It will propagate upward when 4558 * spa_vdev_exit() calls vdev_dtl_reassess(). 4559 */ 4560 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4561 4562 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4563 dtl_max_txg - TXG_INITIAL); 4564 4565 if (newvd->vdev_isspare) { 4566 spa_spare_activate(newvd); 4567 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4568 } 4569 4570 oldvdpath = spa_strdup(oldvd->vdev_path); 4571 newvdpath = spa_strdup(newvd->vdev_path); 4572 newvd_isspare = newvd->vdev_isspare; 4573 4574 /* 4575 * Mark newvd's DTL dirty in this txg. 4576 */ 4577 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4578 4579 /* 4580 * Restart the resilver 4581 */ 4582 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4583 4584 /* 4585 * Commit the config 4586 */ 4587 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4588 4589 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL, 4590 "%s vdev=%s %s vdev=%s", 4591 replacing && newvd_isspare ? "spare in" : 4592 replacing ? "replace" : "attach", newvdpath, 4593 replacing ? "for" : "to", oldvdpath); 4594 4595 spa_strfree(oldvdpath); 4596 spa_strfree(newvdpath); 4597 4598 if (spa->spa_bootfs) 4599 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4600 4601 return (0); 4602} 4603 4604/* 4605 * Detach a device from a mirror or replacing vdev. 4606 * If 'replace_done' is specified, only detach if the parent 4607 * is a replacing vdev. 4608 */ 4609int 4610spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4611{ 4612 uint64_t txg; 4613 int error; 4614 vdev_t *rvd = spa->spa_root_vdev; 4615 vdev_t *vd, *pvd, *cvd, *tvd; 4616 boolean_t unspare = B_FALSE; 4617 uint64_t unspare_guid; 4618 char *vdpath; 4619 4620 ASSERT(spa_writeable(spa)); 4621 4622 txg = spa_vdev_enter(spa); 4623 4624 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4625 4626 if (vd == NULL) 4627 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4628 4629 if (!vd->vdev_ops->vdev_op_leaf) 4630 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4631 4632 pvd = vd->vdev_parent; 4633 4634 /* 4635 * If the parent/child relationship is not as expected, don't do it. 4636 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4637 * vdev that's replacing B with C. The user's intent in replacing 4638 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4639 * the replace by detaching C, the expected behavior is to end up 4640 * M(A,B). But suppose that right after deciding to detach C, 4641 * the replacement of B completes. We would have M(A,C), and then 4642 * ask to detach C, which would leave us with just A -- not what 4643 * the user wanted. To prevent this, we make sure that the 4644 * parent/child relationship hasn't changed -- in this example, 4645 * that C's parent is still the replacing vdev R. 4646 */ 4647 if (pvd->vdev_guid != pguid && pguid != 0) 4648 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4649 4650 /* 4651 * Only 'replacing' or 'spare' vdevs can be replaced. 4652 */ 4653 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4654 pvd->vdev_ops != &vdev_spare_ops) 4655 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4656 4657 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4658 spa_version(spa) >= SPA_VERSION_SPARES); 4659 4660 /* 4661 * Only mirror, replacing, and spare vdevs support detach. 4662 */ 4663 if (pvd->vdev_ops != &vdev_replacing_ops && 4664 pvd->vdev_ops != &vdev_mirror_ops && 4665 pvd->vdev_ops != &vdev_spare_ops) 4666 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4667 4668 /* 4669 * If this device has the only valid copy of some data, 4670 * we cannot safely detach it. 4671 */ 4672 if (vdev_dtl_required(vd)) 4673 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4674 4675 ASSERT(pvd->vdev_children >= 2); 4676 4677 /* 4678 * If we are detaching the second disk from a replacing vdev, then 4679 * check to see if we changed the original vdev's path to have "/old" 4680 * at the end in spa_vdev_attach(). If so, undo that change now. 4681 */ 4682 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4683 vd->vdev_path != NULL) { 4684 size_t len = strlen(vd->vdev_path); 4685 4686 for (int c = 0; c < pvd->vdev_children; c++) { 4687 cvd = pvd->vdev_child[c]; 4688 4689 if (cvd == vd || cvd->vdev_path == NULL) 4690 continue; 4691 4692 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4693 strcmp(cvd->vdev_path + len, "/old") == 0) { 4694 spa_strfree(cvd->vdev_path); 4695 cvd->vdev_path = spa_strdup(vd->vdev_path); 4696 break; 4697 } 4698 } 4699 } 4700 4701 /* 4702 * If we are detaching the original disk from a spare, then it implies 4703 * that the spare should become a real disk, and be removed from the 4704 * active spare list for the pool. 4705 */ 4706 if (pvd->vdev_ops == &vdev_spare_ops && 4707 vd->vdev_id == 0 && 4708 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4709 unspare = B_TRUE; 4710 4711 /* 4712 * Erase the disk labels so the disk can be used for other things. 4713 * This must be done after all other error cases are handled, 4714 * but before we disembowel vd (so we can still do I/O to it). 4715 * But if we can't do it, don't treat the error as fatal -- 4716 * it may be that the unwritability of the disk is the reason 4717 * it's being detached! 4718 */ 4719 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4720 4721 /* 4722 * Remove vd from its parent and compact the parent's children. 4723 */ 4724 vdev_remove_child(pvd, vd); 4725 vdev_compact_children(pvd); 4726 4727 /* 4728 * Remember one of the remaining children so we can get tvd below. 4729 */ 4730 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4731 4732 /* 4733 * If we need to remove the remaining child from the list of hot spares, 4734 * do it now, marking the vdev as no longer a spare in the process. 4735 * We must do this before vdev_remove_parent(), because that can 4736 * change the GUID if it creates a new toplevel GUID. For a similar 4737 * reason, we must remove the spare now, in the same txg as the detach; 4738 * otherwise someone could attach a new sibling, change the GUID, and 4739 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4740 */ 4741 if (unspare) { 4742 ASSERT(cvd->vdev_isspare); 4743 spa_spare_remove(cvd); 4744 unspare_guid = cvd->vdev_guid; 4745 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4746 cvd->vdev_unspare = B_TRUE; 4747 } 4748 4749 /* 4750 * If the parent mirror/replacing vdev only has one child, 4751 * the parent is no longer needed. Remove it from the tree. 4752 */ 4753 if (pvd->vdev_children == 1) { 4754 if (pvd->vdev_ops == &vdev_spare_ops) 4755 cvd->vdev_unspare = B_FALSE; 4756 vdev_remove_parent(cvd); 4757 cvd->vdev_resilvering = B_FALSE; 4758 } 4759 4760 4761 /* 4762 * We don't set tvd until now because the parent we just removed 4763 * may have been the previous top-level vdev. 4764 */ 4765 tvd = cvd->vdev_top; 4766 ASSERT(tvd->vdev_parent == rvd); 4767 4768 /* 4769 * Reevaluate the parent vdev state. 4770 */ 4771 vdev_propagate_state(cvd); 4772 4773 /* 4774 * If the 'autoexpand' property is set on the pool then automatically 4775 * try to expand the size of the pool. For example if the device we 4776 * just detached was smaller than the others, it may be possible to 4777 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4778 * first so that we can obtain the updated sizes of the leaf vdevs. 4779 */ 4780 if (spa->spa_autoexpand) { 4781 vdev_reopen(tvd); 4782 vdev_expand(tvd, txg); 4783 } 4784 4785 vdev_config_dirty(tvd); 4786 4787 /* 4788 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4789 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4790 * But first make sure we're not on any *other* txg's DTL list, to 4791 * prevent vd from being accessed after it's freed. 4792 */ 4793 vdpath = spa_strdup(vd->vdev_path); 4794 for (int t = 0; t < TXG_SIZE; t++) 4795 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4796 vd->vdev_detached = B_TRUE; 4797 vdev_dirty(tvd, VDD_DTL, vd, txg); 4798 4799 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4800 4801 /* hang on to the spa before we release the lock */ 4802 spa_open_ref(spa, FTAG); 4803 4804 error = spa_vdev_exit(spa, vd, txg, 0); 4805 4806 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL, 4807 "vdev=%s", vdpath); 4808 spa_strfree(vdpath); 4809 4810 /* 4811 * If this was the removal of the original device in a hot spare vdev, 4812 * then we want to go through and remove the device from the hot spare 4813 * list of every other pool. 4814 */ 4815 if (unspare) { 4816 spa_t *altspa = NULL; 4817 4818 mutex_enter(&spa_namespace_lock); 4819 while ((altspa = spa_next(altspa)) != NULL) { 4820 if (altspa->spa_state != POOL_STATE_ACTIVE || 4821 altspa == spa) 4822 continue; 4823 4824 spa_open_ref(altspa, FTAG); 4825 mutex_exit(&spa_namespace_lock); 4826 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 4827 mutex_enter(&spa_namespace_lock); 4828 spa_close(altspa, FTAG); 4829 } 4830 mutex_exit(&spa_namespace_lock); 4831 4832 /* search the rest of the vdevs for spares to remove */ 4833 spa_vdev_resilver_done(spa); 4834 } 4835 4836 /* all done with the spa; OK to release */ 4837 mutex_enter(&spa_namespace_lock); 4838 spa_close(spa, FTAG); 4839 mutex_exit(&spa_namespace_lock); 4840 4841 return (error); 4842} 4843 4844/* 4845 * Split a set of devices from their mirrors, and create a new pool from them. 4846 */ 4847int 4848spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 4849 nvlist_t *props, boolean_t exp) 4850{ 4851 int error = 0; 4852 uint64_t txg, *glist; 4853 spa_t *newspa; 4854 uint_t c, children, lastlog; 4855 nvlist_t **child, *nvl, *tmp; 4856 dmu_tx_t *tx; 4857 char *altroot = NULL; 4858 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 4859 boolean_t activate_slog; 4860 4861 ASSERT(spa_writeable(spa)); 4862 4863 txg = spa_vdev_enter(spa); 4864 4865 /* clear the log and flush everything up to now */ 4866 activate_slog = spa_passivate_log(spa); 4867 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 4868 error = spa_offline_log(spa); 4869 txg = spa_vdev_config_enter(spa); 4870 4871 if (activate_slog) 4872 spa_activate_log(spa); 4873 4874 if (error != 0) 4875 return (spa_vdev_exit(spa, NULL, txg, error)); 4876 4877 /* check new spa name before going any further */ 4878 if (spa_lookup(newname) != NULL) 4879 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 4880 4881 /* 4882 * scan through all the children to ensure they're all mirrors 4883 */ 4884 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 4885 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 4886 &children) != 0) 4887 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4888 4889 /* first, check to ensure we've got the right child count */ 4890 rvd = spa->spa_root_vdev; 4891 lastlog = 0; 4892 for (c = 0; c < rvd->vdev_children; c++) { 4893 vdev_t *vd = rvd->vdev_child[c]; 4894 4895 /* don't count the holes & logs as children */ 4896 if (vd->vdev_islog || vd->vdev_ishole) { 4897 if (lastlog == 0) 4898 lastlog = c; 4899 continue; 4900 } 4901 4902 lastlog = 0; 4903 } 4904 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 4905 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4906 4907 /* next, ensure no spare or cache devices are part of the split */ 4908 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 4909 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 4910 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4911 4912 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 4913 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 4914 4915 /* then, loop over each vdev and validate it */ 4916 for (c = 0; c < children; c++) { 4917 uint64_t is_hole = 0; 4918 4919 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 4920 &is_hole); 4921 4922 if (is_hole != 0) { 4923 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 4924 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 4925 continue; 4926 } else { 4927 error = EINVAL; 4928 break; 4929 } 4930 } 4931 4932 /* which disk is going to be split? */ 4933 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 4934 &glist[c]) != 0) { 4935 error = EINVAL; 4936 break; 4937 } 4938 4939 /* look it up in the spa */ 4940 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 4941 if (vml[c] == NULL) { 4942 error = ENODEV; 4943 break; 4944 } 4945 4946 /* make sure there's nothing stopping the split */ 4947 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 4948 vml[c]->vdev_islog || 4949 vml[c]->vdev_ishole || 4950 vml[c]->vdev_isspare || 4951 vml[c]->vdev_isl2cache || 4952 !vdev_writeable(vml[c]) || 4953 vml[c]->vdev_children != 0 || 4954 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 4955 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 4956 error = EINVAL; 4957 break; 4958 } 4959 4960 if (vdev_dtl_required(vml[c])) { 4961 error = EBUSY; 4962 break; 4963 } 4964 4965 /* we need certain info from the top level */ 4966 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 4967 vml[c]->vdev_top->vdev_ms_array) == 0); 4968 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 4969 vml[c]->vdev_top->vdev_ms_shift) == 0); 4970 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 4971 vml[c]->vdev_top->vdev_asize) == 0); 4972 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 4973 vml[c]->vdev_top->vdev_ashift) == 0); 4974 } 4975 4976 if (error != 0) { 4977 kmem_free(vml, children * sizeof (vdev_t *)); 4978 kmem_free(glist, children * sizeof (uint64_t)); 4979 return (spa_vdev_exit(spa, NULL, txg, error)); 4980 } 4981 4982 /* stop writers from using the disks */ 4983 for (c = 0; c < children; c++) { 4984 if (vml[c] != NULL) 4985 vml[c]->vdev_offline = B_TRUE; 4986 } 4987 vdev_reopen(spa->spa_root_vdev); 4988 4989 /* 4990 * Temporarily record the splitting vdevs in the spa config. This 4991 * will disappear once the config is regenerated. 4992 */ 4993 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 4994 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 4995 glist, children) == 0); 4996 kmem_free(glist, children * sizeof (uint64_t)); 4997 4998 mutex_enter(&spa->spa_props_lock); 4999 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5000 nvl) == 0); 5001 mutex_exit(&spa->spa_props_lock); 5002 spa->spa_config_splitting = nvl; 5003 vdev_config_dirty(spa->spa_root_vdev); 5004 5005 /* configure and create the new pool */ 5006 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5007 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5008 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5009 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5010 spa_version(spa)) == 0); 5011 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5012 spa->spa_config_txg) == 0); 5013 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5014 spa_generate_guid(NULL)) == 0); 5015 (void) nvlist_lookup_string(props, 5016 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5017 5018 /* add the new pool to the namespace */ 5019 newspa = spa_add(newname, config, altroot); 5020 newspa->spa_config_txg = spa->spa_config_txg; 5021 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5022 5023 /* release the spa config lock, retaining the namespace lock */ 5024 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5025 5026 if (zio_injection_enabled) 5027 zio_handle_panic_injection(spa, FTAG, 1); 5028 5029 spa_activate(newspa, spa_mode_global); 5030 spa_async_suspend(newspa); 5031 5032#ifndef sun 5033 /* mark that we are creating new spa by splitting */ 5034 newspa->spa_splitting_newspa = B_TRUE; 5035#endif 5036 /* create the new pool from the disks of the original pool */ 5037 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5038#ifndef sun 5039 newspa->spa_splitting_newspa = B_FALSE; 5040#endif 5041 if (error) 5042 goto out; 5043 5044 /* if that worked, generate a real config for the new pool */ 5045 if (newspa->spa_root_vdev != NULL) { 5046 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5047 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5048 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5049 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5050 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5051 B_TRUE)); 5052 } 5053 5054 /* set the props */ 5055 if (props != NULL) { 5056 spa_configfile_set(newspa, props, B_FALSE); 5057 error = spa_prop_set(newspa, props); 5058 if (error) 5059 goto out; 5060 } 5061 5062 /* flush everything */ 5063 txg = spa_vdev_config_enter(newspa); 5064 vdev_config_dirty(newspa->spa_root_vdev); 5065 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5066 5067 if (zio_injection_enabled) 5068 zio_handle_panic_injection(spa, FTAG, 2); 5069 5070 spa_async_resume(newspa); 5071 5072 /* finally, update the original pool's config */ 5073 txg = spa_vdev_config_enter(spa); 5074 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5075 error = dmu_tx_assign(tx, TXG_WAIT); 5076 if (error != 0) 5077 dmu_tx_abort(tx); 5078 for (c = 0; c < children; c++) { 5079 if (vml[c] != NULL) { 5080 vdev_split(vml[c]); 5081 if (error == 0) 5082 spa_history_log_internal(LOG_POOL_VDEV_DETACH, 5083 spa, tx, "vdev=%s", 5084 vml[c]->vdev_path); 5085 vdev_free(vml[c]); 5086 } 5087 } 5088 vdev_config_dirty(spa->spa_root_vdev); 5089 spa->spa_config_splitting = NULL; 5090 nvlist_free(nvl); 5091 if (error == 0) 5092 dmu_tx_commit(tx); 5093 (void) spa_vdev_exit(spa, NULL, txg, 0); 5094 5095 if (zio_injection_enabled) 5096 zio_handle_panic_injection(spa, FTAG, 3); 5097 5098 /* split is complete; log a history record */ 5099 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL, 5100 "split new pool %s from pool %s", newname, spa_name(spa)); 5101 5102 kmem_free(vml, children * sizeof (vdev_t *)); 5103 5104 /* if we're not going to mount the filesystems in userland, export */ 5105 if (exp) 5106 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5107 B_FALSE, B_FALSE); 5108 5109 return (error); 5110 5111out: 5112 spa_unload(newspa); 5113 spa_deactivate(newspa); 5114 spa_remove(newspa); 5115 5116 txg = spa_vdev_config_enter(spa); 5117 5118 /* re-online all offlined disks */ 5119 for (c = 0; c < children; c++) { 5120 if (vml[c] != NULL) 5121 vml[c]->vdev_offline = B_FALSE; 5122 } 5123 vdev_reopen(spa->spa_root_vdev); 5124 5125 nvlist_free(spa->spa_config_splitting); 5126 spa->spa_config_splitting = NULL; 5127 (void) spa_vdev_exit(spa, NULL, txg, error); 5128 5129 kmem_free(vml, children * sizeof (vdev_t *)); 5130 return (error); 5131} 5132 5133static nvlist_t * 5134spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5135{ 5136 for (int i = 0; i < count; i++) { 5137 uint64_t guid; 5138 5139 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5140 &guid) == 0); 5141 5142 if (guid == target_guid) 5143 return (nvpp[i]); 5144 } 5145 5146 return (NULL); 5147} 5148 5149static void 5150spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5151 nvlist_t *dev_to_remove) 5152{ 5153 nvlist_t **newdev = NULL; 5154 5155 if (count > 1) 5156 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5157 5158 for (int i = 0, j = 0; i < count; i++) { 5159 if (dev[i] == dev_to_remove) 5160 continue; 5161 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5162 } 5163 5164 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5165 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5166 5167 for (int i = 0; i < count - 1; i++) 5168 nvlist_free(newdev[i]); 5169 5170 if (count > 1) 5171 kmem_free(newdev, (count - 1) * sizeof (void *)); 5172} 5173 5174/* 5175 * Evacuate the device. 5176 */ 5177static int 5178spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5179{ 5180 uint64_t txg; 5181 int error = 0; 5182 5183 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5184 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5185 ASSERT(vd == vd->vdev_top); 5186 5187 /* 5188 * Evacuate the device. We don't hold the config lock as writer 5189 * since we need to do I/O but we do keep the 5190 * spa_namespace_lock held. Once this completes the device 5191 * should no longer have any blocks allocated on it. 5192 */ 5193 if (vd->vdev_islog) { 5194 if (vd->vdev_stat.vs_alloc != 0) 5195 error = spa_offline_log(spa); 5196 } else { 5197 error = ENOTSUP; 5198 } 5199 5200 if (error) 5201 return (error); 5202 5203 /* 5204 * The evacuation succeeded. Remove any remaining MOS metadata 5205 * associated with this vdev, and wait for these changes to sync. 5206 */ 5207 ASSERT0(vd->vdev_stat.vs_alloc); 5208 txg = spa_vdev_config_enter(spa); 5209 vd->vdev_removing = B_TRUE; 5210 vdev_dirty(vd, 0, NULL, txg); 5211 vdev_config_dirty(vd); 5212 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5213 5214 return (0); 5215} 5216 5217/* 5218 * Complete the removal by cleaning up the namespace. 5219 */ 5220static void 5221spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5222{ 5223 vdev_t *rvd = spa->spa_root_vdev; 5224 uint64_t id = vd->vdev_id; 5225 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5226 5227 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5228 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5229 ASSERT(vd == vd->vdev_top); 5230 5231 /* 5232 * Only remove any devices which are empty. 5233 */ 5234 if (vd->vdev_stat.vs_alloc != 0) 5235 return; 5236 5237 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5238 5239 if (list_link_active(&vd->vdev_state_dirty_node)) 5240 vdev_state_clean(vd); 5241 if (list_link_active(&vd->vdev_config_dirty_node)) 5242 vdev_config_clean(vd); 5243 5244 vdev_free(vd); 5245 5246 if (last_vdev) { 5247 vdev_compact_children(rvd); 5248 } else { 5249 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5250 vdev_add_child(rvd, vd); 5251 } 5252 vdev_config_dirty(rvd); 5253 5254 /* 5255 * Reassess the health of our root vdev. 5256 */ 5257 vdev_reopen(rvd); 5258} 5259 5260/* 5261 * Remove a device from the pool - 5262 * 5263 * Removing a device from the vdev namespace requires several steps 5264 * and can take a significant amount of time. As a result we use 5265 * the spa_vdev_config_[enter/exit] functions which allow us to 5266 * grab and release the spa_config_lock while still holding the namespace 5267 * lock. During each step the configuration is synced out. 5268 */ 5269 5270/* 5271 * Remove a device from the pool. Currently, this supports removing only hot 5272 * spares, slogs, and level 2 ARC devices. 5273 */ 5274int 5275spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5276{ 5277 vdev_t *vd; 5278 metaslab_group_t *mg; 5279 nvlist_t **spares, **l2cache, *nv; 5280 uint64_t txg = 0; 5281 uint_t nspares, nl2cache; 5282 int error = 0; 5283 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5284 5285 ASSERT(spa_writeable(spa)); 5286 5287 if (!locked) 5288 txg = spa_vdev_enter(spa); 5289 5290 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5291 5292 if (spa->spa_spares.sav_vdevs != NULL && 5293 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5294 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5295 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5296 /* 5297 * Only remove the hot spare if it's not currently in use 5298 * in this pool. 5299 */ 5300 if (vd == NULL || unspare) { 5301 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5302 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5303 spa_load_spares(spa); 5304 spa->spa_spares.sav_sync = B_TRUE; 5305 } else { 5306 error = EBUSY; 5307 } 5308 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5309 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5310 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5311 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5312 /* 5313 * Cache devices can always be removed. 5314 */ 5315 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5316 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5317 spa_load_l2cache(spa); 5318 spa->spa_l2cache.sav_sync = B_TRUE; 5319 } else if (vd != NULL && vd->vdev_islog) { 5320 ASSERT(!locked); 5321 ASSERT(vd == vd->vdev_top); 5322 5323 /* 5324 * XXX - Once we have bp-rewrite this should 5325 * become the common case. 5326 */ 5327 5328 mg = vd->vdev_mg; 5329 5330 /* 5331 * Stop allocating from this vdev. 5332 */ 5333 metaslab_group_passivate(mg); 5334 5335 /* 5336 * Wait for the youngest allocations and frees to sync, 5337 * and then wait for the deferral of those frees to finish. 5338 */ 5339 spa_vdev_config_exit(spa, NULL, 5340 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5341 5342 /* 5343 * Attempt to evacuate the vdev. 5344 */ 5345 error = spa_vdev_remove_evacuate(spa, vd); 5346 5347 txg = spa_vdev_config_enter(spa); 5348 5349 /* 5350 * If we couldn't evacuate the vdev, unwind. 5351 */ 5352 if (error) { 5353 metaslab_group_activate(mg); 5354 return (spa_vdev_exit(spa, NULL, txg, error)); 5355 } 5356 5357 /* 5358 * Clean up the vdev namespace. 5359 */ 5360 spa_vdev_remove_from_namespace(spa, vd); 5361 5362 } else if (vd != NULL) { 5363 /* 5364 * Normal vdevs cannot be removed (yet). 5365 */ 5366 error = ENOTSUP; 5367 } else { 5368 /* 5369 * There is no vdev of any kind with the specified guid. 5370 */ 5371 error = ENOENT; 5372 } 5373 5374 if (!locked) 5375 return (spa_vdev_exit(spa, NULL, txg, error)); 5376 5377 return (error); 5378} 5379 5380/* 5381 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5382 * current spared, so we can detach it. 5383 */ 5384static vdev_t * 5385spa_vdev_resilver_done_hunt(vdev_t *vd) 5386{ 5387 vdev_t *newvd, *oldvd; 5388 5389 for (int c = 0; c < vd->vdev_children; c++) { 5390 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5391 if (oldvd != NULL) 5392 return (oldvd); 5393 } 5394 5395 /* 5396 * Check for a completed replacement. We always consider the first 5397 * vdev in the list to be the oldest vdev, and the last one to be 5398 * the newest (see spa_vdev_attach() for how that works). In 5399 * the case where the newest vdev is faulted, we will not automatically 5400 * remove it after a resilver completes. This is OK as it will require 5401 * user intervention to determine which disk the admin wishes to keep. 5402 */ 5403 if (vd->vdev_ops == &vdev_replacing_ops) { 5404 ASSERT(vd->vdev_children > 1); 5405 5406 newvd = vd->vdev_child[vd->vdev_children - 1]; 5407 oldvd = vd->vdev_child[0]; 5408 5409 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5410 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5411 !vdev_dtl_required(oldvd)) 5412 return (oldvd); 5413 } 5414 5415 /* 5416 * Check for a completed resilver with the 'unspare' flag set. 5417 */ 5418 if (vd->vdev_ops == &vdev_spare_ops) { 5419 vdev_t *first = vd->vdev_child[0]; 5420 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5421 5422 if (last->vdev_unspare) { 5423 oldvd = first; 5424 newvd = last; 5425 } else if (first->vdev_unspare) { 5426 oldvd = last; 5427 newvd = first; 5428 } else { 5429 oldvd = NULL; 5430 } 5431 5432 if (oldvd != NULL && 5433 vdev_dtl_empty(newvd, DTL_MISSING) && 5434 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5435 !vdev_dtl_required(oldvd)) 5436 return (oldvd); 5437 5438 /* 5439 * If there are more than two spares attached to a disk, 5440 * and those spares are not required, then we want to 5441 * attempt to free them up now so that they can be used 5442 * by other pools. Once we're back down to a single 5443 * disk+spare, we stop removing them. 5444 */ 5445 if (vd->vdev_children > 2) { 5446 newvd = vd->vdev_child[1]; 5447 5448 if (newvd->vdev_isspare && last->vdev_isspare && 5449 vdev_dtl_empty(last, DTL_MISSING) && 5450 vdev_dtl_empty(last, DTL_OUTAGE) && 5451 !vdev_dtl_required(newvd)) 5452 return (newvd); 5453 } 5454 } 5455 5456 return (NULL); 5457} 5458 5459static void 5460spa_vdev_resilver_done(spa_t *spa) 5461{ 5462 vdev_t *vd, *pvd, *ppvd; 5463 uint64_t guid, sguid, pguid, ppguid; 5464 5465 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5466 5467 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5468 pvd = vd->vdev_parent; 5469 ppvd = pvd->vdev_parent; 5470 guid = vd->vdev_guid; 5471 pguid = pvd->vdev_guid; 5472 ppguid = ppvd->vdev_guid; 5473 sguid = 0; 5474 /* 5475 * If we have just finished replacing a hot spared device, then 5476 * we need to detach the parent's first child (the original hot 5477 * spare) as well. 5478 */ 5479 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5480 ppvd->vdev_children == 2) { 5481 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5482 sguid = ppvd->vdev_child[1]->vdev_guid; 5483 } 5484 spa_config_exit(spa, SCL_ALL, FTAG); 5485 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5486 return; 5487 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5488 return; 5489 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5490 } 5491 5492 spa_config_exit(spa, SCL_ALL, FTAG); 5493} 5494 5495/* 5496 * Update the stored path or FRU for this vdev. 5497 */ 5498int 5499spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5500 boolean_t ispath) 5501{ 5502 vdev_t *vd; 5503 boolean_t sync = B_FALSE; 5504 5505 ASSERT(spa_writeable(spa)); 5506 5507 spa_vdev_state_enter(spa, SCL_ALL); 5508 5509 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5510 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5511 5512 if (!vd->vdev_ops->vdev_op_leaf) 5513 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5514 5515 if (ispath) { 5516 if (strcmp(value, vd->vdev_path) != 0) { 5517 spa_strfree(vd->vdev_path); 5518 vd->vdev_path = spa_strdup(value); 5519 sync = B_TRUE; 5520 } 5521 } else { 5522 if (vd->vdev_fru == NULL) { 5523 vd->vdev_fru = spa_strdup(value); 5524 sync = B_TRUE; 5525 } else if (strcmp(value, vd->vdev_fru) != 0) { 5526 spa_strfree(vd->vdev_fru); 5527 vd->vdev_fru = spa_strdup(value); 5528 sync = B_TRUE; 5529 } 5530 } 5531 5532 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5533} 5534 5535int 5536spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5537{ 5538 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5539} 5540 5541int 5542spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5543{ 5544 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5545} 5546 5547/* 5548 * ========================================================================== 5549 * SPA Scanning 5550 * ========================================================================== 5551 */ 5552 5553int 5554spa_scan_stop(spa_t *spa) 5555{ 5556 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5557 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5558 return (EBUSY); 5559 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5560} 5561 5562int 5563spa_scan(spa_t *spa, pool_scan_func_t func) 5564{ 5565 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5566 5567 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5568 return (ENOTSUP); 5569 5570 /* 5571 * If a resilver was requested, but there is no DTL on a 5572 * writeable leaf device, we have nothing to do. 5573 */ 5574 if (func == POOL_SCAN_RESILVER && 5575 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5576 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5577 return (0); 5578 } 5579 5580 return (dsl_scan(spa->spa_dsl_pool, func)); 5581} 5582 5583/* 5584 * ========================================================================== 5585 * SPA async task processing 5586 * ========================================================================== 5587 */ 5588 5589static void 5590spa_async_remove(spa_t *spa, vdev_t *vd) 5591{ 5592 if (vd->vdev_remove_wanted) { 5593 vd->vdev_remove_wanted = B_FALSE; 5594 vd->vdev_delayed_close = B_FALSE; 5595 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5596 5597 /* 5598 * We want to clear the stats, but we don't want to do a full 5599 * vdev_clear() as that will cause us to throw away 5600 * degraded/faulted state as well as attempt to reopen the 5601 * device, all of which is a waste. 5602 */ 5603 vd->vdev_stat.vs_read_errors = 0; 5604 vd->vdev_stat.vs_write_errors = 0; 5605 vd->vdev_stat.vs_checksum_errors = 0; 5606 5607 vdev_state_dirty(vd->vdev_top); 5608 } 5609 5610 for (int c = 0; c < vd->vdev_children; c++) 5611 spa_async_remove(spa, vd->vdev_child[c]); 5612} 5613 5614static void 5615spa_async_probe(spa_t *spa, vdev_t *vd) 5616{ 5617 if (vd->vdev_probe_wanted) { 5618 vd->vdev_probe_wanted = B_FALSE; 5619 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5620 } 5621 5622 for (int c = 0; c < vd->vdev_children; c++) 5623 spa_async_probe(spa, vd->vdev_child[c]); 5624} 5625 5626static void 5627spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5628{ 5629 sysevent_id_t eid; 5630 nvlist_t *attr; 5631 char *physpath; 5632 5633 if (!spa->spa_autoexpand) 5634 return; 5635 5636 for (int c = 0; c < vd->vdev_children; c++) { 5637 vdev_t *cvd = vd->vdev_child[c]; 5638 spa_async_autoexpand(spa, cvd); 5639 } 5640 5641 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5642 return; 5643 5644 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5645 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5646 5647 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5648 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5649 5650 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5651 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5652 5653 nvlist_free(attr); 5654 kmem_free(physpath, MAXPATHLEN); 5655} 5656 5657static void 5658spa_async_thread(void *arg) 5659{ 5660 spa_t *spa = arg; 5661 int tasks; 5662 5663 ASSERT(spa->spa_sync_on); 5664 5665 mutex_enter(&spa->spa_async_lock); 5666 tasks = spa->spa_async_tasks; 5667 spa->spa_async_tasks = 0; 5668 mutex_exit(&spa->spa_async_lock); 5669 5670 /* 5671 * See if the config needs to be updated. 5672 */ 5673 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5674 uint64_t old_space, new_space; 5675 5676 mutex_enter(&spa_namespace_lock); 5677 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5678 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5679 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5680 mutex_exit(&spa_namespace_lock); 5681 5682 /* 5683 * If the pool grew as a result of the config update, 5684 * then log an internal history event. 5685 */ 5686 if (new_space != old_space) { 5687 spa_history_log_internal(LOG_POOL_VDEV_ONLINE, 5688 spa, NULL, 5689 "pool '%s' size: %llu(+%llu)", 5690 spa_name(spa), new_space, new_space - old_space); 5691 } 5692 } 5693 5694 /* 5695 * See if any devices need to be marked REMOVED. 5696 */ 5697 if (tasks & SPA_ASYNC_REMOVE) { 5698 spa_vdev_state_enter(spa, SCL_NONE); 5699 spa_async_remove(spa, spa->spa_root_vdev); 5700 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5701 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5702 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5703 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5704 (void) spa_vdev_state_exit(spa, NULL, 0); 5705 } 5706 5707 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5708 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5709 spa_async_autoexpand(spa, spa->spa_root_vdev); 5710 spa_config_exit(spa, SCL_CONFIG, FTAG); 5711 } 5712 5713 /* 5714 * See if any devices need to be probed. 5715 */ 5716 if (tasks & SPA_ASYNC_PROBE) { 5717 spa_vdev_state_enter(spa, SCL_NONE); 5718 spa_async_probe(spa, spa->spa_root_vdev); 5719 (void) spa_vdev_state_exit(spa, NULL, 0); 5720 } 5721 5722 /* 5723 * If any devices are done replacing, detach them. 5724 */ 5725 if (tasks & SPA_ASYNC_RESILVER_DONE) 5726 spa_vdev_resilver_done(spa); 5727 5728 /* 5729 * Kick off a resilver. 5730 */ 5731 if (tasks & SPA_ASYNC_RESILVER) 5732 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5733 5734 /* 5735 * Let the world know that we're done. 5736 */ 5737 mutex_enter(&spa->spa_async_lock); 5738 spa->spa_async_thread = NULL; 5739 cv_broadcast(&spa->spa_async_cv); 5740 mutex_exit(&spa->spa_async_lock); 5741 thread_exit(); 5742} 5743 5744void 5745spa_async_suspend(spa_t *spa) 5746{ 5747 mutex_enter(&spa->spa_async_lock); 5748 spa->spa_async_suspended++; 5749 while (spa->spa_async_thread != NULL) 5750 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5751 mutex_exit(&spa->spa_async_lock); 5752} 5753 5754void 5755spa_async_resume(spa_t *spa) 5756{ 5757 mutex_enter(&spa->spa_async_lock); 5758 ASSERT(spa->spa_async_suspended != 0); 5759 spa->spa_async_suspended--; 5760 mutex_exit(&spa->spa_async_lock); 5761} 5762 5763static void 5764spa_async_dispatch(spa_t *spa) 5765{ 5766 mutex_enter(&spa->spa_async_lock); 5767 if (spa->spa_async_tasks && !spa->spa_async_suspended && 5768 spa->spa_async_thread == NULL && 5769 rootdir != NULL && !vn_is_readonly(rootdir)) 5770 spa->spa_async_thread = thread_create(NULL, 0, 5771 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 5772 mutex_exit(&spa->spa_async_lock); 5773} 5774 5775void 5776spa_async_request(spa_t *spa, int task) 5777{ 5778 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 5779 mutex_enter(&spa->spa_async_lock); 5780 spa->spa_async_tasks |= task; 5781 mutex_exit(&spa->spa_async_lock); 5782} 5783 5784/* 5785 * ========================================================================== 5786 * SPA syncing routines 5787 * ========================================================================== 5788 */ 5789 5790static int 5791bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5792{ 5793 bpobj_t *bpo = arg; 5794 bpobj_enqueue(bpo, bp, tx); 5795 return (0); 5796} 5797 5798static int 5799spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 5800{ 5801 zio_t *zio = arg; 5802 5803 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 5804 BP_GET_PSIZE(bp), zio->io_flags)); 5805 return (0); 5806} 5807 5808static void 5809spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 5810{ 5811 char *packed = NULL; 5812 size_t bufsize; 5813 size_t nvsize = 0; 5814 dmu_buf_t *db; 5815 5816 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 5817 5818 /* 5819 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 5820 * information. This avoids the dbuf_will_dirty() path and 5821 * saves us a pre-read to get data we don't actually care about. 5822 */ 5823 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 5824 packed = kmem_alloc(bufsize, KM_SLEEP); 5825 5826 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 5827 KM_SLEEP) == 0); 5828 bzero(packed + nvsize, bufsize - nvsize); 5829 5830 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 5831 5832 kmem_free(packed, bufsize); 5833 5834 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 5835 dmu_buf_will_dirty(db, tx); 5836 *(uint64_t *)db->db_data = nvsize; 5837 dmu_buf_rele(db, FTAG); 5838} 5839 5840static void 5841spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 5842 const char *config, const char *entry) 5843{ 5844 nvlist_t *nvroot; 5845 nvlist_t **list; 5846 int i; 5847 5848 if (!sav->sav_sync) 5849 return; 5850 5851 /* 5852 * Update the MOS nvlist describing the list of available devices. 5853 * spa_validate_aux() will have already made sure this nvlist is 5854 * valid and the vdevs are labeled appropriately. 5855 */ 5856 if (sav->sav_object == 0) { 5857 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 5858 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 5859 sizeof (uint64_t), tx); 5860 VERIFY(zap_update(spa->spa_meta_objset, 5861 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 5862 &sav->sav_object, tx) == 0); 5863 } 5864 5865 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5866 if (sav->sav_count == 0) { 5867 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 5868 } else { 5869 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 5870 for (i = 0; i < sav->sav_count; i++) 5871 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 5872 B_FALSE, VDEV_CONFIG_L2CACHE); 5873 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 5874 sav->sav_count) == 0); 5875 for (i = 0; i < sav->sav_count; i++) 5876 nvlist_free(list[i]); 5877 kmem_free(list, sav->sav_count * sizeof (void *)); 5878 } 5879 5880 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 5881 nvlist_free(nvroot); 5882 5883 sav->sav_sync = B_FALSE; 5884} 5885 5886static void 5887spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 5888{ 5889 nvlist_t *config; 5890 5891 if (list_is_empty(&spa->spa_config_dirty_list)) 5892 return; 5893 5894 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 5895 5896 config = spa_config_generate(spa, spa->spa_root_vdev, 5897 dmu_tx_get_txg(tx), B_FALSE); 5898 5899 spa_config_exit(spa, SCL_STATE, FTAG); 5900 5901 if (spa->spa_config_syncing) 5902 nvlist_free(spa->spa_config_syncing); 5903 spa->spa_config_syncing = config; 5904 5905 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 5906} 5907 5908static void 5909spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx) 5910{ 5911 spa_t *spa = arg1; 5912 uint64_t version = *(uint64_t *)arg2; 5913 5914 /* 5915 * Setting the version is special cased when first creating the pool. 5916 */ 5917 ASSERT(tx->tx_txg != TXG_INITIAL); 5918 5919 ASSERT(version <= SPA_VERSION); 5920 ASSERT(version >= spa_version(spa)); 5921 5922 spa->spa_uberblock.ub_version = version; 5923 vdev_config_dirty(spa->spa_root_vdev); 5924} 5925 5926/* 5927 * Set zpool properties. 5928 */ 5929static void 5930spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx) 5931{ 5932 spa_t *spa = arg1; 5933 objset_t *mos = spa->spa_meta_objset; 5934 nvlist_t *nvp = arg2; 5935 nvpair_t *elem = NULL; 5936 5937 mutex_enter(&spa->spa_props_lock); 5938 5939 while ((elem = nvlist_next_nvpair(nvp, elem))) { 5940 uint64_t intval; 5941 char *strval, *fname; 5942 zpool_prop_t prop; 5943 const char *propname; 5944 zprop_type_t proptype; 5945 zfeature_info_t *feature; 5946 5947 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 5948 case ZPROP_INVAL: 5949 /* 5950 * We checked this earlier in spa_prop_validate(). 5951 */ 5952 ASSERT(zpool_prop_feature(nvpair_name(elem))); 5953 5954 fname = strchr(nvpair_name(elem), '@') + 1; 5955 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature)); 5956 5957 spa_feature_enable(spa, feature, tx); 5958 break; 5959 5960 case ZPOOL_PROP_VERSION: 5961 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 5962 /* 5963 * The version is synced seperatly before other 5964 * properties and should be correct by now. 5965 */ 5966 ASSERT3U(spa_version(spa), >=, intval); 5967 break; 5968 5969 case ZPOOL_PROP_ALTROOT: 5970 /* 5971 * 'altroot' is a non-persistent property. It should 5972 * have been set temporarily at creation or import time. 5973 */ 5974 ASSERT(spa->spa_root != NULL); 5975 break; 5976 5977 case ZPOOL_PROP_READONLY: 5978 case ZPOOL_PROP_CACHEFILE: 5979 /* 5980 * 'readonly' and 'cachefile' are also non-persisitent 5981 * properties. 5982 */ 5983 break; 5984 case ZPOOL_PROP_COMMENT: 5985 VERIFY(nvpair_value_string(elem, &strval) == 0); 5986 if (spa->spa_comment != NULL) 5987 spa_strfree(spa->spa_comment); 5988 spa->spa_comment = spa_strdup(strval); 5989 /* 5990 * We need to dirty the configuration on all the vdevs 5991 * so that their labels get updated. It's unnecessary 5992 * to do this for pool creation since the vdev's 5993 * configuratoin has already been dirtied. 5994 */ 5995 if (tx->tx_txg != TXG_INITIAL) 5996 vdev_config_dirty(spa->spa_root_vdev); 5997 break; 5998 default: 5999 /* 6000 * Set pool property values in the poolprops mos object. 6001 */ 6002 if (spa->spa_pool_props_object == 0) { 6003 spa->spa_pool_props_object = 6004 zap_create_link(mos, DMU_OT_POOL_PROPS, 6005 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6006 tx); 6007 } 6008 6009 /* normalize the property name */ 6010 propname = zpool_prop_to_name(prop); 6011 proptype = zpool_prop_get_type(prop); 6012 6013 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6014 ASSERT(proptype == PROP_TYPE_STRING); 6015 VERIFY(nvpair_value_string(elem, &strval) == 0); 6016 VERIFY(zap_update(mos, 6017 spa->spa_pool_props_object, propname, 6018 1, strlen(strval) + 1, strval, tx) == 0); 6019 6020 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6021 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6022 6023 if (proptype == PROP_TYPE_INDEX) { 6024 const char *unused; 6025 VERIFY(zpool_prop_index_to_string( 6026 prop, intval, &unused) == 0); 6027 } 6028 VERIFY(zap_update(mos, 6029 spa->spa_pool_props_object, propname, 6030 8, 1, &intval, tx) == 0); 6031 } else { 6032 ASSERT(0); /* not allowed */ 6033 } 6034 6035 switch (prop) { 6036 case ZPOOL_PROP_DELEGATION: 6037 spa->spa_delegation = intval; 6038 break; 6039 case ZPOOL_PROP_BOOTFS: 6040 spa->spa_bootfs = intval; 6041 break; 6042 case ZPOOL_PROP_FAILUREMODE: 6043 spa->spa_failmode = intval; 6044 break; 6045 case ZPOOL_PROP_AUTOEXPAND: 6046 spa->spa_autoexpand = intval; 6047 if (tx->tx_txg != TXG_INITIAL) 6048 spa_async_request(spa, 6049 SPA_ASYNC_AUTOEXPAND); 6050 break; 6051 case ZPOOL_PROP_DEDUPDITTO: 6052 spa->spa_dedup_ditto = intval; 6053 break; 6054 default: 6055 break; 6056 } 6057 } 6058 6059 /* log internal history if this is not a zpool create */ 6060 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY && 6061 tx->tx_txg != TXG_INITIAL) { 6062 spa_history_log_internal(LOG_POOL_PROPSET, 6063 spa, tx, "%s %lld %s", 6064 nvpair_name(elem), intval, spa_name(spa)); 6065 } 6066 } 6067 6068 mutex_exit(&spa->spa_props_lock); 6069} 6070 6071/* 6072 * Perform one-time upgrade on-disk changes. spa_version() does not 6073 * reflect the new version this txg, so there must be no changes this 6074 * txg to anything that the upgrade code depends on after it executes. 6075 * Therefore this must be called after dsl_pool_sync() does the sync 6076 * tasks. 6077 */ 6078static void 6079spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6080{ 6081 dsl_pool_t *dp = spa->spa_dsl_pool; 6082 6083 ASSERT(spa->spa_sync_pass == 1); 6084 6085 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6086 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6087 dsl_pool_create_origin(dp, tx); 6088 6089 /* Keeping the origin open increases spa_minref */ 6090 spa->spa_minref += 3; 6091 } 6092 6093 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6094 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6095 dsl_pool_upgrade_clones(dp, tx); 6096 } 6097 6098 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6099 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6100 dsl_pool_upgrade_dir_clones(dp, tx); 6101 6102 /* Keeping the freedir open increases spa_minref */ 6103 spa->spa_minref += 3; 6104 } 6105 6106 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6107 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6108 spa_feature_create_zap_objects(spa, tx); 6109 } 6110} 6111 6112/* 6113 * Sync the specified transaction group. New blocks may be dirtied as 6114 * part of the process, so we iterate until it converges. 6115 */ 6116void 6117spa_sync(spa_t *spa, uint64_t txg) 6118{ 6119 dsl_pool_t *dp = spa->spa_dsl_pool; 6120 objset_t *mos = spa->spa_meta_objset; 6121 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj; 6122 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6123 vdev_t *rvd = spa->spa_root_vdev; 6124 vdev_t *vd; 6125 dmu_tx_t *tx; 6126 int error; 6127 6128 VERIFY(spa_writeable(spa)); 6129 6130 /* 6131 * Lock out configuration changes. 6132 */ 6133 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6134 6135 spa->spa_syncing_txg = txg; 6136 spa->spa_sync_pass = 0; 6137 6138 /* 6139 * If there are any pending vdev state changes, convert them 6140 * into config changes that go out with this transaction group. 6141 */ 6142 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6143 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6144 /* 6145 * We need the write lock here because, for aux vdevs, 6146 * calling vdev_config_dirty() modifies sav_config. 6147 * This is ugly and will become unnecessary when we 6148 * eliminate the aux vdev wart by integrating all vdevs 6149 * into the root vdev tree. 6150 */ 6151 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6152 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6153 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6154 vdev_state_clean(vd); 6155 vdev_config_dirty(vd); 6156 } 6157 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6158 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6159 } 6160 spa_config_exit(spa, SCL_STATE, FTAG); 6161 6162 tx = dmu_tx_create_assigned(dp, txg); 6163 6164 /* 6165 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6166 * set spa_deflate if we have no raid-z vdevs. 6167 */ 6168 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6169 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6170 int i; 6171 6172 for (i = 0; i < rvd->vdev_children; i++) { 6173 vd = rvd->vdev_child[i]; 6174 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6175 break; 6176 } 6177 if (i == rvd->vdev_children) { 6178 spa->spa_deflate = TRUE; 6179 VERIFY(0 == zap_add(spa->spa_meta_objset, 6180 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6181 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6182 } 6183 } 6184 6185 /* 6186 * If anything has changed in this txg, or if someone is waiting 6187 * for this txg to sync (eg, spa_vdev_remove()), push the 6188 * deferred frees from the previous txg. If not, leave them 6189 * alone so that we don't generate work on an otherwise idle 6190 * system. 6191 */ 6192 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6193 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6194 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6195 ((dsl_scan_active(dp->dp_scan) || 6196 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6197 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6198 VERIFY3U(bpobj_iterate(defer_bpo, 6199 spa_free_sync_cb, zio, tx), ==, 0); 6200 VERIFY0(zio_wait(zio)); 6201 } 6202 6203 /* 6204 * Iterate to convergence. 6205 */ 6206 do { 6207 int pass = ++spa->spa_sync_pass; 6208 6209 spa_sync_config_object(spa, tx); 6210 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6211 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6212 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6213 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6214 spa_errlog_sync(spa, txg); 6215 dsl_pool_sync(dp, txg); 6216 6217 if (pass <= SYNC_PASS_DEFERRED_FREE) { 6218 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6219 bplist_iterate(free_bpl, spa_free_sync_cb, 6220 zio, tx); 6221 VERIFY(zio_wait(zio) == 0); 6222 } else { 6223 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6224 defer_bpo, tx); 6225 } 6226 6227 ddt_sync(spa, txg); 6228 dsl_scan_sync(dp, tx); 6229 6230 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6231 vdev_sync(vd, txg); 6232 6233 if (pass == 1) 6234 spa_sync_upgrades(spa, tx); 6235 6236 } while (dmu_objset_is_dirty(mos, txg)); 6237 6238 /* 6239 * Rewrite the vdev configuration (which includes the uberblock) 6240 * to commit the transaction group. 6241 * 6242 * If there are no dirty vdevs, we sync the uberblock to a few 6243 * random top-level vdevs that are known to be visible in the 6244 * config cache (see spa_vdev_add() for a complete description). 6245 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6246 */ 6247 for (;;) { 6248 /* 6249 * We hold SCL_STATE to prevent vdev open/close/etc. 6250 * while we're attempting to write the vdev labels. 6251 */ 6252 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6253 6254 if (list_is_empty(&spa->spa_config_dirty_list)) { 6255 vdev_t *svd[SPA_DVAS_PER_BP]; 6256 int svdcount = 0; 6257 int children = rvd->vdev_children; 6258 int c0 = spa_get_random(children); 6259 6260 for (int c = 0; c < children; c++) { 6261 vd = rvd->vdev_child[(c0 + c) % children]; 6262 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6263 continue; 6264 svd[svdcount++] = vd; 6265 if (svdcount == SPA_DVAS_PER_BP) 6266 break; 6267 } 6268 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6269 if (error != 0) 6270 error = vdev_config_sync(svd, svdcount, txg, 6271 B_TRUE); 6272 } else { 6273 error = vdev_config_sync(rvd->vdev_child, 6274 rvd->vdev_children, txg, B_FALSE); 6275 if (error != 0) 6276 error = vdev_config_sync(rvd->vdev_child, 6277 rvd->vdev_children, txg, B_TRUE); 6278 } 6279 6280 if (error == 0) 6281 spa->spa_last_synced_guid = rvd->vdev_guid; 6282 6283 spa_config_exit(spa, SCL_STATE, FTAG); 6284 6285 if (error == 0) 6286 break; 6287 zio_suspend(spa, NULL); 6288 zio_resume_wait(spa); 6289 } 6290 dmu_tx_commit(tx); 6291 6292 /* 6293 * Clear the dirty config list. 6294 */ 6295 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6296 vdev_config_clean(vd); 6297 6298 /* 6299 * Now that the new config has synced transactionally, 6300 * let it become visible to the config cache. 6301 */ 6302 if (spa->spa_config_syncing != NULL) { 6303 spa_config_set(spa, spa->spa_config_syncing); 6304 spa->spa_config_txg = txg; 6305 spa->spa_config_syncing = NULL; 6306 } 6307 6308 spa->spa_ubsync = spa->spa_uberblock; 6309 6310 dsl_pool_sync_done(dp, txg); 6311 6312 /* 6313 * Update usable space statistics. 6314 */ 6315 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6316 vdev_sync_done(vd, txg); 6317 6318 spa_update_dspace(spa); 6319 6320 /* 6321 * It had better be the case that we didn't dirty anything 6322 * since vdev_config_sync(). 6323 */ 6324 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6325 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6326 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6327 6328 spa->spa_sync_pass = 0; 6329 6330 spa_config_exit(spa, SCL_CONFIG, FTAG); 6331 6332 spa_handle_ignored_writes(spa); 6333 6334 /* 6335 * If any async tasks have been requested, kick them off. 6336 */ 6337 spa_async_dispatch(spa); 6338} 6339 6340/* 6341 * Sync all pools. We don't want to hold the namespace lock across these 6342 * operations, so we take a reference on the spa_t and drop the lock during the 6343 * sync. 6344 */ 6345void 6346spa_sync_allpools(void) 6347{ 6348 spa_t *spa = NULL; 6349 mutex_enter(&spa_namespace_lock); 6350 while ((spa = spa_next(spa)) != NULL) { 6351 if (spa_state(spa) != POOL_STATE_ACTIVE || 6352 !spa_writeable(spa) || spa_suspended(spa)) 6353 continue; 6354 spa_open_ref(spa, FTAG); 6355 mutex_exit(&spa_namespace_lock); 6356 txg_wait_synced(spa_get_dsl(spa), 0); 6357 mutex_enter(&spa_namespace_lock); 6358 spa_close(spa, FTAG); 6359 } 6360 mutex_exit(&spa_namespace_lock); 6361} 6362 6363/* 6364 * ========================================================================== 6365 * Miscellaneous routines 6366 * ========================================================================== 6367 */ 6368 6369/* 6370 * Remove all pools in the system. 6371 */ 6372void 6373spa_evict_all(void) 6374{ 6375 spa_t *spa; 6376 6377 /* 6378 * Remove all cached state. All pools should be closed now, 6379 * so every spa in the AVL tree should be unreferenced. 6380 */ 6381 mutex_enter(&spa_namespace_lock); 6382 while ((spa = spa_next(NULL)) != NULL) { 6383 /* 6384 * Stop async tasks. The async thread may need to detach 6385 * a device that's been replaced, which requires grabbing 6386 * spa_namespace_lock, so we must drop it here. 6387 */ 6388 spa_open_ref(spa, FTAG); 6389 mutex_exit(&spa_namespace_lock); 6390 spa_async_suspend(spa); 6391 mutex_enter(&spa_namespace_lock); 6392 spa_close(spa, FTAG); 6393 6394 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6395 spa_unload(spa); 6396 spa_deactivate(spa); 6397 } 6398 spa_remove(spa); 6399 } 6400 mutex_exit(&spa_namespace_lock); 6401} 6402 6403vdev_t * 6404spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6405{ 6406 vdev_t *vd; 6407 int i; 6408 6409 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6410 return (vd); 6411 6412 if (aux) { 6413 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6414 vd = spa->spa_l2cache.sav_vdevs[i]; 6415 if (vd->vdev_guid == guid) 6416 return (vd); 6417 } 6418 6419 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6420 vd = spa->spa_spares.sav_vdevs[i]; 6421 if (vd->vdev_guid == guid) 6422 return (vd); 6423 } 6424 } 6425 6426 return (NULL); 6427} 6428 6429void 6430spa_upgrade(spa_t *spa, uint64_t version) 6431{ 6432 ASSERT(spa_writeable(spa)); 6433 6434 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6435 6436 /* 6437 * This should only be called for a non-faulted pool, and since a 6438 * future version would result in an unopenable pool, this shouldn't be 6439 * possible. 6440 */ 6441 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION); 6442 ASSERT(version >= spa->spa_uberblock.ub_version); 6443 6444 spa->spa_uberblock.ub_version = version; 6445 vdev_config_dirty(spa->spa_root_vdev); 6446 6447 spa_config_exit(spa, SCL_ALL, FTAG); 6448 6449 txg_wait_synced(spa_get_dsl(spa), 0); 6450} 6451 6452boolean_t 6453spa_has_spare(spa_t *spa, uint64_t guid) 6454{ 6455 int i; 6456 uint64_t spareguid; 6457 spa_aux_vdev_t *sav = &spa->spa_spares; 6458 6459 for (i = 0; i < sav->sav_count; i++) 6460 if (sav->sav_vdevs[i]->vdev_guid == guid) 6461 return (B_TRUE); 6462 6463 for (i = 0; i < sav->sav_npending; i++) { 6464 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6465 &spareguid) == 0 && spareguid == guid) 6466 return (B_TRUE); 6467 } 6468 6469 return (B_FALSE); 6470} 6471 6472/* 6473 * Check if a pool has an active shared spare device. 6474 * Note: reference count of an active spare is 2, as a spare and as a replace 6475 */ 6476static boolean_t 6477spa_has_active_shared_spare(spa_t *spa) 6478{ 6479 int i, refcnt; 6480 uint64_t pool; 6481 spa_aux_vdev_t *sav = &spa->spa_spares; 6482 6483 for (i = 0; i < sav->sav_count; i++) { 6484 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6485 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6486 refcnt > 2) 6487 return (B_TRUE); 6488 } 6489 6490 return (B_FALSE); 6491} 6492 6493/* 6494 * Post a sysevent corresponding to the given event. The 'name' must be one of 6495 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6496 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6497 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6498 * or zdb as real changes. 6499 */ 6500void 6501spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6502{ 6503#ifdef _KERNEL 6504 sysevent_t *ev; 6505 sysevent_attr_list_t *attr = NULL; 6506 sysevent_value_t value; 6507 sysevent_id_t eid; 6508 6509 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6510 SE_SLEEP); 6511 6512 value.value_type = SE_DATA_TYPE_STRING; 6513 value.value.sv_string = spa_name(spa); 6514 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6515 goto done; 6516 6517 value.value_type = SE_DATA_TYPE_UINT64; 6518 value.value.sv_uint64 = spa_guid(spa); 6519 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6520 goto done; 6521 6522 if (vd) { 6523 value.value_type = SE_DATA_TYPE_UINT64; 6524 value.value.sv_uint64 = vd->vdev_guid; 6525 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6526 SE_SLEEP) != 0) 6527 goto done; 6528 6529 if (vd->vdev_path) { 6530 value.value_type = SE_DATA_TYPE_STRING; 6531 value.value.sv_string = vd->vdev_path; 6532 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6533 &value, SE_SLEEP) != 0) 6534 goto done; 6535 } 6536 } 6537 6538 if (sysevent_attach_attributes(ev, attr) != 0) 6539 goto done; 6540 attr = NULL; 6541 6542 (void) log_sysevent(ev, SE_SLEEP, &eid); 6543 6544done: 6545 if (attr) 6546 sysevent_free_attr(attr); 6547 sysevent_free(ev); 6548#endif 6549} 6550