spa.c revision 258717
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2013 by Delphix. All rights reserved. 25 * Copyright 2013 Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 */ 28 29/* 30 * SPA: Storage Pool Allocator 31 * 32 * This file contains all the routines used when modifying on-disk SPA state. 33 * This includes opening, importing, destroying, exporting a pool, and syncing a 34 * pool. 35 */ 36 37#include <sys/zfs_context.h> 38#include <sys/fm/fs/zfs.h> 39#include <sys/spa_impl.h> 40#include <sys/zio.h> 41#include <sys/zio_checksum.h> 42#include <sys/dmu.h> 43#include <sys/dmu_tx.h> 44#include <sys/zap.h> 45#include <sys/zil.h> 46#include <sys/ddt.h> 47#include <sys/vdev_impl.h> 48#include <sys/metaslab.h> 49#include <sys/metaslab_impl.h> 50#include <sys/uberblock_impl.h> 51#include <sys/txg.h> 52#include <sys/avl.h> 53#include <sys/dmu_traverse.h> 54#include <sys/dmu_objset.h> 55#include <sys/unique.h> 56#include <sys/dsl_pool.h> 57#include <sys/dsl_dataset.h> 58#include <sys/dsl_dir.h> 59#include <sys/dsl_prop.h> 60#include <sys/dsl_synctask.h> 61#include <sys/fs/zfs.h> 62#include <sys/arc.h> 63#include <sys/callb.h> 64#include <sys/spa_boot.h> 65#include <sys/zfs_ioctl.h> 66#include <sys/dsl_scan.h> 67#include <sys/dmu_send.h> 68#include <sys/dsl_destroy.h> 69#include <sys/dsl_userhold.h> 70#include <sys/zfeature.h> 71#include <sys/zvol.h> 72#include <sys/trim_map.h> 73 74#ifdef _KERNEL 75#include <sys/callb.h> 76#include <sys/cpupart.h> 77#include <sys/zone.h> 78#endif /* _KERNEL */ 79 80#include "zfs_prop.h" 81#include "zfs_comutil.h" 82 83/* Check hostid on import? */ 84static int check_hostid = 1; 85 86SYSCTL_DECL(_vfs_zfs); 87TUNABLE_INT("vfs.zfs.check_hostid", &check_hostid); 88SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RW, &check_hostid, 0, 89 "Check hostid on import?"); 90 91/* 92 * The interval, in seconds, at which failed configuration cache file writes 93 * should be retried. 94 */ 95static int zfs_ccw_retry_interval = 300; 96 97typedef enum zti_modes { 98 ZTI_MODE_FIXED, /* value is # of threads (min 1) */ 99 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */ 100 ZTI_MODE_NULL, /* don't create a taskq */ 101 ZTI_NMODES 102} zti_modes_t; 103 104#define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) } 105#define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 } 106#define ZTI_NULL { ZTI_MODE_NULL, 0, 0 } 107 108#define ZTI_N(n) ZTI_P(n, 1) 109#define ZTI_ONE ZTI_N(1) 110 111typedef struct zio_taskq_info { 112 zti_modes_t zti_mode; 113 uint_t zti_value; 114 uint_t zti_count; 115} zio_taskq_info_t; 116 117static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = { 118 "issue", "issue_high", "intr", "intr_high" 119}; 120 121/* 122 * This table defines the taskq settings for each ZFS I/O type. When 123 * initializing a pool, we use this table to create an appropriately sized 124 * taskq. Some operations are low volume and therefore have a small, static 125 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE 126 * macros. Other operations process a large amount of data; the ZTI_BATCH 127 * macro causes us to create a taskq oriented for throughput. Some operations 128 * are so high frequency and short-lived that the taskq itself can become a a 129 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an 130 * additional degree of parallelism specified by the number of threads per- 131 * taskq and the number of taskqs; when dispatching an event in this case, the 132 * particular taskq is chosen at random. 133 * 134 * The different taskq priorities are to handle the different contexts (issue 135 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that 136 * need to be handled with minimum delay. 137 */ 138const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = { 139 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */ 140 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */ 141 { ZTI_N(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL }, /* READ */ 142 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */ 143 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */ 144 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */ 145 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */ 146}; 147 148static void spa_sync_version(void *arg, dmu_tx_t *tx); 149static void spa_sync_props(void *arg, dmu_tx_t *tx); 150static boolean_t spa_has_active_shared_spare(spa_t *spa); 151static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config, 152 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 153 char **ereport); 154static void spa_vdev_resilver_done(spa_t *spa); 155 156uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */ 157#ifdef PSRSET_BIND 158id_t zio_taskq_psrset_bind = PS_NONE; 159#endif 160#ifdef SYSDC 161boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */ 162#endif 163uint_t zio_taskq_basedc = 80; /* base duty cycle */ 164 165boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */ 166extern int zfs_sync_pass_deferred_free; 167 168#ifndef illumos 169extern void spa_deadman(void *arg); 170#endif 171 172/* 173 * This (illegal) pool name is used when temporarily importing a spa_t in order 174 * to get the vdev stats associated with the imported devices. 175 */ 176#define TRYIMPORT_NAME "$import" 177 178/* 179 * ========================================================================== 180 * SPA properties routines 181 * ========================================================================== 182 */ 183 184/* 185 * Add a (source=src, propname=propval) list to an nvlist. 186 */ 187static void 188spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval, 189 uint64_t intval, zprop_source_t src) 190{ 191 const char *propname = zpool_prop_to_name(prop); 192 nvlist_t *propval; 193 194 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0); 195 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0); 196 197 if (strval != NULL) 198 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0); 199 else 200 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0); 201 202 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0); 203 nvlist_free(propval); 204} 205 206/* 207 * Get property values from the spa configuration. 208 */ 209static void 210spa_prop_get_config(spa_t *spa, nvlist_t **nvp) 211{ 212 vdev_t *rvd = spa->spa_root_vdev; 213 dsl_pool_t *pool = spa->spa_dsl_pool; 214 uint64_t size; 215 uint64_t alloc; 216 uint64_t space; 217 uint64_t cap, version; 218 zprop_source_t src = ZPROP_SRC_NONE; 219 spa_config_dirent_t *dp; 220 221 ASSERT(MUTEX_HELD(&spa->spa_props_lock)); 222 223 if (rvd != NULL) { 224 alloc = metaslab_class_get_alloc(spa_normal_class(spa)); 225 size = metaslab_class_get_space(spa_normal_class(spa)); 226 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src); 227 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src); 228 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src); 229 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL, 230 size - alloc, src); 231 232 space = 0; 233 for (int c = 0; c < rvd->vdev_children; c++) { 234 vdev_t *tvd = rvd->vdev_child[c]; 235 space += tvd->vdev_max_asize - tvd->vdev_asize; 236 } 237 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space, 238 src); 239 240 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL, 241 (spa_mode(spa) == FREAD), src); 242 243 cap = (size == 0) ? 0 : (alloc * 100 / size); 244 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src); 245 246 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL, 247 ddt_get_pool_dedup_ratio(spa), src); 248 249 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL, 250 rvd->vdev_state, src); 251 252 version = spa_version(spa); 253 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) 254 src = ZPROP_SRC_DEFAULT; 255 else 256 src = ZPROP_SRC_LOCAL; 257 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src); 258 } 259 260 if (pool != NULL) { 261 dsl_dir_t *freedir = pool->dp_free_dir; 262 263 /* 264 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS, 265 * when opening pools before this version freedir will be NULL. 266 */ 267 if (freedir != NULL) { 268 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL, 269 freedir->dd_phys->dd_used_bytes, src); 270 } else { 271 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, 272 NULL, 0, src); 273 } 274 } 275 276 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src); 277 278 if (spa->spa_comment != NULL) { 279 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment, 280 0, ZPROP_SRC_LOCAL); 281 } 282 283 if (spa->spa_root != NULL) 284 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root, 285 0, ZPROP_SRC_LOCAL); 286 287 if ((dp = list_head(&spa->spa_config_list)) != NULL) { 288 if (dp->scd_path == NULL) { 289 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 290 "none", 0, ZPROP_SRC_LOCAL); 291 } else if (strcmp(dp->scd_path, spa_config_path) != 0) { 292 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE, 293 dp->scd_path, 0, ZPROP_SRC_LOCAL); 294 } 295 } 296} 297 298/* 299 * Get zpool property values. 300 */ 301int 302spa_prop_get(spa_t *spa, nvlist_t **nvp) 303{ 304 objset_t *mos = spa->spa_meta_objset; 305 zap_cursor_t zc; 306 zap_attribute_t za; 307 int err; 308 309 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0); 310 311 mutex_enter(&spa->spa_props_lock); 312 313 /* 314 * Get properties from the spa config. 315 */ 316 spa_prop_get_config(spa, nvp); 317 318 /* If no pool property object, no more prop to get. */ 319 if (mos == NULL || spa->spa_pool_props_object == 0) { 320 mutex_exit(&spa->spa_props_lock); 321 return (0); 322 } 323 324 /* 325 * Get properties from the MOS pool property object. 326 */ 327 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object); 328 (err = zap_cursor_retrieve(&zc, &za)) == 0; 329 zap_cursor_advance(&zc)) { 330 uint64_t intval = 0; 331 char *strval = NULL; 332 zprop_source_t src = ZPROP_SRC_DEFAULT; 333 zpool_prop_t prop; 334 335 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL) 336 continue; 337 338 switch (za.za_integer_length) { 339 case 8: 340 /* integer property */ 341 if (za.za_first_integer != 342 zpool_prop_default_numeric(prop)) 343 src = ZPROP_SRC_LOCAL; 344 345 if (prop == ZPOOL_PROP_BOOTFS) { 346 dsl_pool_t *dp; 347 dsl_dataset_t *ds = NULL; 348 349 dp = spa_get_dsl(spa); 350 dsl_pool_config_enter(dp, FTAG); 351 if (err = dsl_dataset_hold_obj(dp, 352 za.za_first_integer, FTAG, &ds)) { 353 dsl_pool_config_exit(dp, FTAG); 354 break; 355 } 356 357 strval = kmem_alloc( 358 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1, 359 KM_SLEEP); 360 dsl_dataset_name(ds, strval); 361 dsl_dataset_rele(ds, FTAG); 362 dsl_pool_config_exit(dp, FTAG); 363 } else { 364 strval = NULL; 365 intval = za.za_first_integer; 366 } 367 368 spa_prop_add_list(*nvp, prop, strval, intval, src); 369 370 if (strval != NULL) 371 kmem_free(strval, 372 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1); 373 374 break; 375 376 case 1: 377 /* string property */ 378 strval = kmem_alloc(za.za_num_integers, KM_SLEEP); 379 err = zap_lookup(mos, spa->spa_pool_props_object, 380 za.za_name, 1, za.za_num_integers, strval); 381 if (err) { 382 kmem_free(strval, za.za_num_integers); 383 break; 384 } 385 spa_prop_add_list(*nvp, prop, strval, 0, src); 386 kmem_free(strval, za.za_num_integers); 387 break; 388 389 default: 390 break; 391 } 392 } 393 zap_cursor_fini(&zc); 394 mutex_exit(&spa->spa_props_lock); 395out: 396 if (err && err != ENOENT) { 397 nvlist_free(*nvp); 398 *nvp = NULL; 399 return (err); 400 } 401 402 return (0); 403} 404 405/* 406 * Validate the given pool properties nvlist and modify the list 407 * for the property values to be set. 408 */ 409static int 410spa_prop_validate(spa_t *spa, nvlist_t *props) 411{ 412 nvpair_t *elem; 413 int error = 0, reset_bootfs = 0; 414 uint64_t objnum = 0; 415 boolean_t has_feature = B_FALSE; 416 417 elem = NULL; 418 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) { 419 uint64_t intval; 420 char *strval, *slash, *check, *fname; 421 const char *propname = nvpair_name(elem); 422 zpool_prop_t prop = zpool_name_to_prop(propname); 423 424 switch (prop) { 425 case ZPROP_INVAL: 426 if (!zpool_prop_feature(propname)) { 427 error = SET_ERROR(EINVAL); 428 break; 429 } 430 431 /* 432 * Sanitize the input. 433 */ 434 if (nvpair_type(elem) != DATA_TYPE_UINT64) { 435 error = SET_ERROR(EINVAL); 436 break; 437 } 438 439 if (nvpair_value_uint64(elem, &intval) != 0) { 440 error = SET_ERROR(EINVAL); 441 break; 442 } 443 444 if (intval != 0) { 445 error = SET_ERROR(EINVAL); 446 break; 447 } 448 449 fname = strchr(propname, '@') + 1; 450 if (zfeature_lookup_name(fname, NULL) != 0) { 451 error = SET_ERROR(EINVAL); 452 break; 453 } 454 455 has_feature = B_TRUE; 456 break; 457 458 case ZPOOL_PROP_VERSION: 459 error = nvpair_value_uint64(elem, &intval); 460 if (!error && 461 (intval < spa_version(spa) || 462 intval > SPA_VERSION_BEFORE_FEATURES || 463 has_feature)) 464 error = SET_ERROR(EINVAL); 465 break; 466 467 case ZPOOL_PROP_DELEGATION: 468 case ZPOOL_PROP_AUTOREPLACE: 469 case ZPOOL_PROP_LISTSNAPS: 470 case ZPOOL_PROP_AUTOEXPAND: 471 error = nvpair_value_uint64(elem, &intval); 472 if (!error && intval > 1) 473 error = SET_ERROR(EINVAL); 474 break; 475 476 case ZPOOL_PROP_BOOTFS: 477 /* 478 * If the pool version is less than SPA_VERSION_BOOTFS, 479 * or the pool is still being created (version == 0), 480 * the bootfs property cannot be set. 481 */ 482 if (spa_version(spa) < SPA_VERSION_BOOTFS) { 483 error = SET_ERROR(ENOTSUP); 484 break; 485 } 486 487 /* 488 * Make sure the vdev config is bootable 489 */ 490 if (!vdev_is_bootable(spa->spa_root_vdev)) { 491 error = SET_ERROR(ENOTSUP); 492 break; 493 } 494 495 reset_bootfs = 1; 496 497 error = nvpair_value_string(elem, &strval); 498 499 if (!error) { 500 objset_t *os; 501 uint64_t compress; 502 503 if (strval == NULL || strval[0] == '\0') { 504 objnum = zpool_prop_default_numeric( 505 ZPOOL_PROP_BOOTFS); 506 break; 507 } 508 509 if (error = dmu_objset_hold(strval, FTAG, &os)) 510 break; 511 512 /* Must be ZPL and not gzip compressed. */ 513 514 if (dmu_objset_type(os) != DMU_OST_ZFS) { 515 error = SET_ERROR(ENOTSUP); 516 } else if ((error = 517 dsl_prop_get_int_ds(dmu_objset_ds(os), 518 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 519 &compress)) == 0 && 520 !BOOTFS_COMPRESS_VALID(compress)) { 521 error = SET_ERROR(ENOTSUP); 522 } else { 523 objnum = dmu_objset_id(os); 524 } 525 dmu_objset_rele(os, FTAG); 526 } 527 break; 528 529 case ZPOOL_PROP_FAILUREMODE: 530 error = nvpair_value_uint64(elem, &intval); 531 if (!error && (intval < ZIO_FAILURE_MODE_WAIT || 532 intval > ZIO_FAILURE_MODE_PANIC)) 533 error = SET_ERROR(EINVAL); 534 535 /* 536 * This is a special case which only occurs when 537 * the pool has completely failed. This allows 538 * the user to change the in-core failmode property 539 * without syncing it out to disk (I/Os might 540 * currently be blocked). We do this by returning 541 * EIO to the caller (spa_prop_set) to trick it 542 * into thinking we encountered a property validation 543 * error. 544 */ 545 if (!error && spa_suspended(spa)) { 546 spa->spa_failmode = intval; 547 error = SET_ERROR(EIO); 548 } 549 break; 550 551 case ZPOOL_PROP_CACHEFILE: 552 if ((error = nvpair_value_string(elem, &strval)) != 0) 553 break; 554 555 if (strval[0] == '\0') 556 break; 557 558 if (strcmp(strval, "none") == 0) 559 break; 560 561 if (strval[0] != '/') { 562 error = SET_ERROR(EINVAL); 563 break; 564 } 565 566 slash = strrchr(strval, '/'); 567 ASSERT(slash != NULL); 568 569 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 || 570 strcmp(slash, "/..") == 0) 571 error = SET_ERROR(EINVAL); 572 break; 573 574 case ZPOOL_PROP_COMMENT: 575 if ((error = nvpair_value_string(elem, &strval)) != 0) 576 break; 577 for (check = strval; *check != '\0'; check++) { 578 /* 579 * The kernel doesn't have an easy isprint() 580 * check. For this kernel check, we merely 581 * check ASCII apart from DEL. Fix this if 582 * there is an easy-to-use kernel isprint(). 583 */ 584 if (*check >= 0x7f) { 585 error = SET_ERROR(EINVAL); 586 break; 587 } 588 check++; 589 } 590 if (strlen(strval) > ZPROP_MAX_COMMENT) 591 error = E2BIG; 592 break; 593 594 case ZPOOL_PROP_DEDUPDITTO: 595 if (spa_version(spa) < SPA_VERSION_DEDUP) 596 error = SET_ERROR(ENOTSUP); 597 else 598 error = nvpair_value_uint64(elem, &intval); 599 if (error == 0 && 600 intval != 0 && intval < ZIO_DEDUPDITTO_MIN) 601 error = SET_ERROR(EINVAL); 602 break; 603 } 604 605 if (error) 606 break; 607 } 608 609 if (!error && reset_bootfs) { 610 error = nvlist_remove(props, 611 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING); 612 613 if (!error) { 614 error = nvlist_add_uint64(props, 615 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum); 616 } 617 } 618 619 return (error); 620} 621 622void 623spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync) 624{ 625 char *cachefile; 626 spa_config_dirent_t *dp; 627 628 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), 629 &cachefile) != 0) 630 return; 631 632 dp = kmem_alloc(sizeof (spa_config_dirent_t), 633 KM_SLEEP); 634 635 if (cachefile[0] == '\0') 636 dp->scd_path = spa_strdup(spa_config_path); 637 else if (strcmp(cachefile, "none") == 0) 638 dp->scd_path = NULL; 639 else 640 dp->scd_path = spa_strdup(cachefile); 641 642 list_insert_head(&spa->spa_config_list, dp); 643 if (need_sync) 644 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 645} 646 647int 648spa_prop_set(spa_t *spa, nvlist_t *nvp) 649{ 650 int error; 651 nvpair_t *elem = NULL; 652 boolean_t need_sync = B_FALSE; 653 654 if ((error = spa_prop_validate(spa, nvp)) != 0) 655 return (error); 656 657 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) { 658 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem)); 659 660 if (prop == ZPOOL_PROP_CACHEFILE || 661 prop == ZPOOL_PROP_ALTROOT || 662 prop == ZPOOL_PROP_READONLY) 663 continue; 664 665 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) { 666 uint64_t ver; 667 668 if (prop == ZPOOL_PROP_VERSION) { 669 VERIFY(nvpair_value_uint64(elem, &ver) == 0); 670 } else { 671 ASSERT(zpool_prop_feature(nvpair_name(elem))); 672 ver = SPA_VERSION_FEATURES; 673 need_sync = B_TRUE; 674 } 675 676 /* Save time if the version is already set. */ 677 if (ver == spa_version(spa)) 678 continue; 679 680 /* 681 * In addition to the pool directory object, we might 682 * create the pool properties object, the features for 683 * read object, the features for write object, or the 684 * feature descriptions object. 685 */ 686 error = dsl_sync_task(spa->spa_name, NULL, 687 spa_sync_version, &ver, 6); 688 if (error) 689 return (error); 690 continue; 691 } 692 693 need_sync = B_TRUE; 694 break; 695 } 696 697 if (need_sync) { 698 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props, 699 nvp, 6)); 700 } 701 702 return (0); 703} 704 705/* 706 * If the bootfs property value is dsobj, clear it. 707 */ 708void 709spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx) 710{ 711 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) { 712 VERIFY(zap_remove(spa->spa_meta_objset, 713 spa->spa_pool_props_object, 714 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0); 715 spa->spa_bootfs = 0; 716 } 717} 718 719/*ARGSUSED*/ 720static int 721spa_change_guid_check(void *arg, dmu_tx_t *tx) 722{ 723 uint64_t *newguid = arg; 724 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 725 vdev_t *rvd = spa->spa_root_vdev; 726 uint64_t vdev_state; 727 728 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 729 vdev_state = rvd->vdev_state; 730 spa_config_exit(spa, SCL_STATE, FTAG); 731 732 if (vdev_state != VDEV_STATE_HEALTHY) 733 return (SET_ERROR(ENXIO)); 734 735 ASSERT3U(spa_guid(spa), !=, *newguid); 736 737 return (0); 738} 739 740static void 741spa_change_guid_sync(void *arg, dmu_tx_t *tx) 742{ 743 uint64_t *newguid = arg; 744 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 745 uint64_t oldguid; 746 vdev_t *rvd = spa->spa_root_vdev; 747 748 oldguid = spa_guid(spa); 749 750 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 751 rvd->vdev_guid = *newguid; 752 rvd->vdev_guid_sum += (*newguid - oldguid); 753 vdev_config_dirty(rvd); 754 spa_config_exit(spa, SCL_STATE, FTAG); 755 756 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu", 757 oldguid, *newguid); 758} 759 760/* 761 * Change the GUID for the pool. This is done so that we can later 762 * re-import a pool built from a clone of our own vdevs. We will modify 763 * the root vdev's guid, our own pool guid, and then mark all of our 764 * vdevs dirty. Note that we must make sure that all our vdevs are 765 * online when we do this, or else any vdevs that weren't present 766 * would be orphaned from our pool. We are also going to issue a 767 * sysevent to update any watchers. 768 */ 769int 770spa_change_guid(spa_t *spa) 771{ 772 int error; 773 uint64_t guid; 774 775 mutex_enter(&spa->spa_vdev_top_lock); 776 mutex_enter(&spa_namespace_lock); 777 guid = spa_generate_guid(NULL); 778 779 error = dsl_sync_task(spa->spa_name, spa_change_guid_check, 780 spa_change_guid_sync, &guid, 5); 781 782 if (error == 0) { 783 spa_config_sync(spa, B_FALSE, B_TRUE); 784 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID); 785 } 786 787 mutex_exit(&spa_namespace_lock); 788 mutex_exit(&spa->spa_vdev_top_lock); 789 790 return (error); 791} 792 793/* 794 * ========================================================================== 795 * SPA state manipulation (open/create/destroy/import/export) 796 * ========================================================================== 797 */ 798 799static int 800spa_error_entry_compare(const void *a, const void *b) 801{ 802 spa_error_entry_t *sa = (spa_error_entry_t *)a; 803 spa_error_entry_t *sb = (spa_error_entry_t *)b; 804 int ret; 805 806 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark, 807 sizeof (zbookmark_t)); 808 809 if (ret < 0) 810 return (-1); 811 else if (ret > 0) 812 return (1); 813 else 814 return (0); 815} 816 817/* 818 * Utility function which retrieves copies of the current logs and 819 * re-initializes them in the process. 820 */ 821void 822spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub) 823{ 824 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock)); 825 826 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t)); 827 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t)); 828 829 avl_create(&spa->spa_errlist_scrub, 830 spa_error_entry_compare, sizeof (spa_error_entry_t), 831 offsetof(spa_error_entry_t, se_avl)); 832 avl_create(&spa->spa_errlist_last, 833 spa_error_entry_compare, sizeof (spa_error_entry_t), 834 offsetof(spa_error_entry_t, se_avl)); 835} 836 837static void 838spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 839{ 840 const zio_taskq_info_t *ztip = &zio_taskqs[t][q]; 841 enum zti_modes mode = ztip->zti_mode; 842 uint_t value = ztip->zti_value; 843 uint_t count = ztip->zti_count; 844 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 845 char name[32]; 846 uint_t flags = 0; 847 boolean_t batch = B_FALSE; 848 849 if (mode == ZTI_MODE_NULL) { 850 tqs->stqs_count = 0; 851 tqs->stqs_taskq = NULL; 852 return; 853 } 854 855 ASSERT3U(count, >, 0); 856 857 tqs->stqs_count = count; 858 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP); 859 860 switch (mode) { 861 case ZTI_MODE_FIXED: 862 ASSERT3U(value, >=, 1); 863 value = MAX(value, 1); 864 break; 865 866 case ZTI_MODE_BATCH: 867 batch = B_TRUE; 868 flags |= TASKQ_THREADS_CPU_PCT; 869 value = zio_taskq_batch_pct; 870 break; 871 872 default: 873 panic("unrecognized mode for %s_%s taskq (%u:%u) in " 874 "spa_activate()", 875 zio_type_name[t], zio_taskq_types[q], mode, value); 876 break; 877 } 878 879 for (uint_t i = 0; i < count; i++) { 880 taskq_t *tq; 881 882 if (count > 1) { 883 (void) snprintf(name, sizeof (name), "%s_%s_%u", 884 zio_type_name[t], zio_taskq_types[q], i); 885 } else { 886 (void) snprintf(name, sizeof (name), "%s_%s", 887 zio_type_name[t], zio_taskq_types[q]); 888 } 889 890#ifdef SYSDC 891 if (zio_taskq_sysdc && spa->spa_proc != &p0) { 892 if (batch) 893 flags |= TASKQ_DC_BATCH; 894 895 tq = taskq_create_sysdc(name, value, 50, INT_MAX, 896 spa->spa_proc, zio_taskq_basedc, flags); 897 } else { 898#endif 899 pri_t pri = maxclsyspri; 900 /* 901 * The write issue taskq can be extremely CPU 902 * intensive. Run it at slightly lower priority 903 * than the other taskqs. 904 */ 905 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE) 906 pri--; 907 908 tq = taskq_create_proc(name, value, pri, 50, 909 INT_MAX, spa->spa_proc, flags); 910#ifdef SYSDC 911 } 912#endif 913 914 tqs->stqs_taskq[i] = tq; 915 } 916} 917 918static void 919spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q) 920{ 921 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 922 923 if (tqs->stqs_taskq == NULL) { 924 ASSERT0(tqs->stqs_count); 925 return; 926 } 927 928 for (uint_t i = 0; i < tqs->stqs_count; i++) { 929 ASSERT3P(tqs->stqs_taskq[i], !=, NULL); 930 taskq_destroy(tqs->stqs_taskq[i]); 931 } 932 933 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *)); 934 tqs->stqs_taskq = NULL; 935} 936 937/* 938 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority. 939 * Note that a type may have multiple discrete taskqs to avoid lock contention 940 * on the taskq itself. In that case we choose which taskq at random by using 941 * the low bits of gethrtime(). 942 */ 943void 944spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q, 945 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent) 946{ 947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; 948 taskq_t *tq; 949 950 ASSERT3P(tqs->stqs_taskq, !=, NULL); 951 ASSERT3U(tqs->stqs_count, !=, 0); 952 953 if (tqs->stqs_count == 1) { 954 tq = tqs->stqs_taskq[0]; 955 } else { 956 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count]; 957 } 958 959 taskq_dispatch_ent(tq, func, arg, flags, ent); 960} 961 962static void 963spa_create_zio_taskqs(spa_t *spa) 964{ 965 for (int t = 0; t < ZIO_TYPES; t++) { 966 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 967 spa_taskqs_init(spa, t, q); 968 } 969 } 970} 971 972#ifdef _KERNEL 973#ifdef SPA_PROCESS 974static void 975spa_thread(void *arg) 976{ 977 callb_cpr_t cprinfo; 978 979 spa_t *spa = arg; 980 user_t *pu = PTOU(curproc); 981 982 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr, 983 spa->spa_name); 984 985 ASSERT(curproc != &p0); 986 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs), 987 "zpool-%s", spa->spa_name); 988 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm)); 989 990#ifdef PSRSET_BIND 991 /* bind this thread to the requested psrset */ 992 if (zio_taskq_psrset_bind != PS_NONE) { 993 pool_lock(); 994 mutex_enter(&cpu_lock); 995 mutex_enter(&pidlock); 996 mutex_enter(&curproc->p_lock); 997 998 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind, 999 0, NULL, NULL) == 0) { 1000 curthread->t_bind_pset = zio_taskq_psrset_bind; 1001 } else { 1002 cmn_err(CE_WARN, 1003 "Couldn't bind process for zfs pool \"%s\" to " 1004 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind); 1005 } 1006 1007 mutex_exit(&curproc->p_lock); 1008 mutex_exit(&pidlock); 1009 mutex_exit(&cpu_lock); 1010 pool_unlock(); 1011 } 1012#endif 1013 1014#ifdef SYSDC 1015 if (zio_taskq_sysdc) { 1016 sysdc_thread_enter(curthread, 100, 0); 1017 } 1018#endif 1019 1020 spa->spa_proc = curproc; 1021 spa->spa_did = curthread->t_did; 1022 1023 spa_create_zio_taskqs(spa); 1024 1025 mutex_enter(&spa->spa_proc_lock); 1026 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED); 1027 1028 spa->spa_proc_state = SPA_PROC_ACTIVE; 1029 cv_broadcast(&spa->spa_proc_cv); 1030 1031 CALLB_CPR_SAFE_BEGIN(&cprinfo); 1032 while (spa->spa_proc_state == SPA_PROC_ACTIVE) 1033 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1034 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock); 1035 1036 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE); 1037 spa->spa_proc_state = SPA_PROC_GONE; 1038 spa->spa_proc = &p0; 1039 cv_broadcast(&spa->spa_proc_cv); 1040 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */ 1041 1042 mutex_enter(&curproc->p_lock); 1043 lwp_exit(); 1044} 1045#endif /* SPA_PROCESS */ 1046#endif 1047 1048/* 1049 * Activate an uninitialized pool. 1050 */ 1051static void 1052spa_activate(spa_t *spa, int mode) 1053{ 1054 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED); 1055 1056 spa->spa_state = POOL_STATE_ACTIVE; 1057 spa->spa_mode = mode; 1058 1059 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops); 1060 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops); 1061 1062 /* Try to create a covering process */ 1063 mutex_enter(&spa->spa_proc_lock); 1064 ASSERT(spa->spa_proc_state == SPA_PROC_NONE); 1065 ASSERT(spa->spa_proc == &p0); 1066 spa->spa_did = 0; 1067 1068#ifdef SPA_PROCESS 1069 /* Only create a process if we're going to be around a while. */ 1070 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) { 1071 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri, 1072 NULL, 0) == 0) { 1073 spa->spa_proc_state = SPA_PROC_CREATED; 1074 while (spa->spa_proc_state == SPA_PROC_CREATED) { 1075 cv_wait(&spa->spa_proc_cv, 1076 &spa->spa_proc_lock); 1077 } 1078 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1079 ASSERT(spa->spa_proc != &p0); 1080 ASSERT(spa->spa_did != 0); 1081 } else { 1082#ifdef _KERNEL 1083 cmn_err(CE_WARN, 1084 "Couldn't create process for zfs pool \"%s\"\n", 1085 spa->spa_name); 1086#endif 1087 } 1088 } 1089#endif /* SPA_PROCESS */ 1090 mutex_exit(&spa->spa_proc_lock); 1091 1092 /* If we didn't create a process, we need to create our taskqs. */ 1093 ASSERT(spa->spa_proc == &p0); 1094 if (spa->spa_proc == &p0) { 1095 spa_create_zio_taskqs(spa); 1096 } 1097 1098 /* 1099 * Start TRIM thread. 1100 */ 1101 trim_thread_create(spa); 1102 1103 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t), 1104 offsetof(vdev_t, vdev_config_dirty_node)); 1105 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t), 1106 offsetof(vdev_t, vdev_state_dirty_node)); 1107 1108 txg_list_create(&spa->spa_vdev_txg_list, 1109 offsetof(struct vdev, vdev_txg_node)); 1110 1111 avl_create(&spa->spa_errlist_scrub, 1112 spa_error_entry_compare, sizeof (spa_error_entry_t), 1113 offsetof(spa_error_entry_t, se_avl)); 1114 avl_create(&spa->spa_errlist_last, 1115 spa_error_entry_compare, sizeof (spa_error_entry_t), 1116 offsetof(spa_error_entry_t, se_avl)); 1117} 1118 1119/* 1120 * Opposite of spa_activate(). 1121 */ 1122static void 1123spa_deactivate(spa_t *spa) 1124{ 1125 ASSERT(spa->spa_sync_on == B_FALSE); 1126 ASSERT(spa->spa_dsl_pool == NULL); 1127 ASSERT(spa->spa_root_vdev == NULL); 1128 ASSERT(spa->spa_async_zio_root == NULL); 1129 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED); 1130 1131 /* 1132 * Stop TRIM thread in case spa_unload() wasn't called directly 1133 * before spa_deactivate(). 1134 */ 1135 trim_thread_destroy(spa); 1136 1137 txg_list_destroy(&spa->spa_vdev_txg_list); 1138 1139 list_destroy(&spa->spa_config_dirty_list); 1140 list_destroy(&spa->spa_state_dirty_list); 1141 1142 for (int t = 0; t < ZIO_TYPES; t++) { 1143 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) { 1144 spa_taskqs_fini(spa, t, q); 1145 } 1146 } 1147 1148 metaslab_class_destroy(spa->spa_normal_class); 1149 spa->spa_normal_class = NULL; 1150 1151 metaslab_class_destroy(spa->spa_log_class); 1152 spa->spa_log_class = NULL; 1153 1154 /* 1155 * If this was part of an import or the open otherwise failed, we may 1156 * still have errors left in the queues. Empty them just in case. 1157 */ 1158 spa_errlog_drain(spa); 1159 1160 avl_destroy(&spa->spa_errlist_scrub); 1161 avl_destroy(&spa->spa_errlist_last); 1162 1163 spa->spa_state = POOL_STATE_UNINITIALIZED; 1164 1165 mutex_enter(&spa->spa_proc_lock); 1166 if (spa->spa_proc_state != SPA_PROC_NONE) { 1167 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE); 1168 spa->spa_proc_state = SPA_PROC_DEACTIVATE; 1169 cv_broadcast(&spa->spa_proc_cv); 1170 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) { 1171 ASSERT(spa->spa_proc != &p0); 1172 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock); 1173 } 1174 ASSERT(spa->spa_proc_state == SPA_PROC_GONE); 1175 spa->spa_proc_state = SPA_PROC_NONE; 1176 } 1177 ASSERT(spa->spa_proc == &p0); 1178 mutex_exit(&spa->spa_proc_lock); 1179 1180#ifdef SPA_PROCESS 1181 /* 1182 * We want to make sure spa_thread() has actually exited the ZFS 1183 * module, so that the module can't be unloaded out from underneath 1184 * it. 1185 */ 1186 if (spa->spa_did != 0) { 1187 thread_join(spa->spa_did); 1188 spa->spa_did = 0; 1189 } 1190#endif /* SPA_PROCESS */ 1191} 1192 1193/* 1194 * Verify a pool configuration, and construct the vdev tree appropriately. This 1195 * will create all the necessary vdevs in the appropriate layout, with each vdev 1196 * in the CLOSED state. This will prep the pool before open/creation/import. 1197 * All vdev validation is done by the vdev_alloc() routine. 1198 */ 1199static int 1200spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, 1201 uint_t id, int atype) 1202{ 1203 nvlist_t **child; 1204 uint_t children; 1205 int error; 1206 1207 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0) 1208 return (error); 1209 1210 if ((*vdp)->vdev_ops->vdev_op_leaf) 1211 return (0); 1212 1213 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN, 1214 &child, &children); 1215 1216 if (error == ENOENT) 1217 return (0); 1218 1219 if (error) { 1220 vdev_free(*vdp); 1221 *vdp = NULL; 1222 return (SET_ERROR(EINVAL)); 1223 } 1224 1225 for (int c = 0; c < children; c++) { 1226 vdev_t *vd; 1227 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c, 1228 atype)) != 0) { 1229 vdev_free(*vdp); 1230 *vdp = NULL; 1231 return (error); 1232 } 1233 } 1234 1235 ASSERT(*vdp != NULL); 1236 1237 return (0); 1238} 1239 1240/* 1241 * Opposite of spa_load(). 1242 */ 1243static void 1244spa_unload(spa_t *spa) 1245{ 1246 int i; 1247 1248 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 1249 1250 /* 1251 * Stop TRIM thread. 1252 */ 1253 trim_thread_destroy(spa); 1254 1255 /* 1256 * Stop async tasks. 1257 */ 1258 spa_async_suspend(spa); 1259 1260 /* 1261 * Stop syncing. 1262 */ 1263 if (spa->spa_sync_on) { 1264 txg_sync_stop(spa->spa_dsl_pool); 1265 spa->spa_sync_on = B_FALSE; 1266 } 1267 1268 /* 1269 * Wait for any outstanding async I/O to complete. 1270 */ 1271 if (spa->spa_async_zio_root != NULL) { 1272 (void) zio_wait(spa->spa_async_zio_root); 1273 spa->spa_async_zio_root = NULL; 1274 } 1275 1276 bpobj_close(&spa->spa_deferred_bpobj); 1277 1278 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1279 1280 /* 1281 * Close all vdevs. 1282 */ 1283 if (spa->spa_root_vdev) 1284 vdev_free(spa->spa_root_vdev); 1285 ASSERT(spa->spa_root_vdev == NULL); 1286 1287 /* 1288 * Close the dsl pool. 1289 */ 1290 if (spa->spa_dsl_pool) { 1291 dsl_pool_close(spa->spa_dsl_pool); 1292 spa->spa_dsl_pool = NULL; 1293 spa->spa_meta_objset = NULL; 1294 } 1295 1296 ddt_unload(spa); 1297 1298 1299 /* 1300 * Drop and purge level 2 cache 1301 */ 1302 spa_l2cache_drop(spa); 1303 1304 for (i = 0; i < spa->spa_spares.sav_count; i++) 1305 vdev_free(spa->spa_spares.sav_vdevs[i]); 1306 if (spa->spa_spares.sav_vdevs) { 1307 kmem_free(spa->spa_spares.sav_vdevs, 1308 spa->spa_spares.sav_count * sizeof (void *)); 1309 spa->spa_spares.sav_vdevs = NULL; 1310 } 1311 if (spa->spa_spares.sav_config) { 1312 nvlist_free(spa->spa_spares.sav_config); 1313 spa->spa_spares.sav_config = NULL; 1314 } 1315 spa->spa_spares.sav_count = 0; 1316 1317 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1318 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1319 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1320 } 1321 if (spa->spa_l2cache.sav_vdevs) { 1322 kmem_free(spa->spa_l2cache.sav_vdevs, 1323 spa->spa_l2cache.sav_count * sizeof (void *)); 1324 spa->spa_l2cache.sav_vdevs = NULL; 1325 } 1326 if (spa->spa_l2cache.sav_config) { 1327 nvlist_free(spa->spa_l2cache.sav_config); 1328 spa->spa_l2cache.sav_config = NULL; 1329 } 1330 spa->spa_l2cache.sav_count = 0; 1331 1332 spa->spa_async_suspended = 0; 1333 1334 if (spa->spa_comment != NULL) { 1335 spa_strfree(spa->spa_comment); 1336 spa->spa_comment = NULL; 1337 } 1338 1339 spa_config_exit(spa, SCL_ALL, FTAG); 1340} 1341 1342/* 1343 * Load (or re-load) the current list of vdevs describing the active spares for 1344 * this pool. When this is called, we have some form of basic information in 1345 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1346 * then re-generate a more complete list including status information. 1347 */ 1348static void 1349spa_load_spares(spa_t *spa) 1350{ 1351 nvlist_t **spares; 1352 uint_t nspares; 1353 int i; 1354 vdev_t *vd, *tvd; 1355 1356 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1357 1358 /* 1359 * First, close and free any existing spare vdevs. 1360 */ 1361 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1362 vd = spa->spa_spares.sav_vdevs[i]; 1363 1364 /* Undo the call to spa_activate() below */ 1365 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1366 B_FALSE)) != NULL && tvd->vdev_isspare) 1367 spa_spare_remove(tvd); 1368 vdev_close(vd); 1369 vdev_free(vd); 1370 } 1371 1372 if (spa->spa_spares.sav_vdevs) 1373 kmem_free(spa->spa_spares.sav_vdevs, 1374 spa->spa_spares.sav_count * sizeof (void *)); 1375 1376 if (spa->spa_spares.sav_config == NULL) 1377 nspares = 0; 1378 else 1379 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1380 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1381 1382 spa->spa_spares.sav_count = (int)nspares; 1383 spa->spa_spares.sav_vdevs = NULL; 1384 1385 if (nspares == 0) 1386 return; 1387 1388 /* 1389 * Construct the array of vdevs, opening them to get status in the 1390 * process. For each spare, there is potentially two different vdev_t 1391 * structures associated with it: one in the list of spares (used only 1392 * for basic validation purposes) and one in the active vdev 1393 * configuration (if it's spared in). During this phase we open and 1394 * validate each vdev on the spare list. If the vdev also exists in the 1395 * active configuration, then we also mark this vdev as an active spare. 1396 */ 1397 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1398 KM_SLEEP); 1399 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1400 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1401 VDEV_ALLOC_SPARE) == 0); 1402 ASSERT(vd != NULL); 1403 1404 spa->spa_spares.sav_vdevs[i] = vd; 1405 1406 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1407 B_FALSE)) != NULL) { 1408 if (!tvd->vdev_isspare) 1409 spa_spare_add(tvd); 1410 1411 /* 1412 * We only mark the spare active if we were successfully 1413 * able to load the vdev. Otherwise, importing a pool 1414 * with a bad active spare would result in strange 1415 * behavior, because multiple pool would think the spare 1416 * is actively in use. 1417 * 1418 * There is a vulnerability here to an equally bizarre 1419 * circumstance, where a dead active spare is later 1420 * brought back to life (onlined or otherwise). Given 1421 * the rarity of this scenario, and the extra complexity 1422 * it adds, we ignore the possibility. 1423 */ 1424 if (!vdev_is_dead(tvd)) 1425 spa_spare_activate(tvd); 1426 } 1427 1428 vd->vdev_top = vd; 1429 vd->vdev_aux = &spa->spa_spares; 1430 1431 if (vdev_open(vd) != 0) 1432 continue; 1433 1434 if (vdev_validate_aux(vd) == 0) 1435 spa_spare_add(vd); 1436 } 1437 1438 /* 1439 * Recompute the stashed list of spares, with status information 1440 * this time. 1441 */ 1442 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1443 DATA_TYPE_NVLIST_ARRAY) == 0); 1444 1445 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1446 KM_SLEEP); 1447 for (i = 0; i < spa->spa_spares.sav_count; i++) 1448 spares[i] = vdev_config_generate(spa, 1449 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1450 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1451 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1452 for (i = 0; i < spa->spa_spares.sav_count; i++) 1453 nvlist_free(spares[i]); 1454 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1455} 1456 1457/* 1458 * Load (or re-load) the current list of vdevs describing the active l2cache for 1459 * this pool. When this is called, we have some form of basic information in 1460 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1461 * then re-generate a more complete list including status information. 1462 * Devices which are already active have their details maintained, and are 1463 * not re-opened. 1464 */ 1465static void 1466spa_load_l2cache(spa_t *spa) 1467{ 1468 nvlist_t **l2cache; 1469 uint_t nl2cache; 1470 int i, j, oldnvdevs; 1471 uint64_t guid; 1472 vdev_t *vd, **oldvdevs, **newvdevs; 1473 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1474 1475 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1476 1477 if (sav->sav_config != NULL) { 1478 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1479 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1480 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1481 } else { 1482 nl2cache = 0; 1483 newvdevs = NULL; 1484 } 1485 1486 oldvdevs = sav->sav_vdevs; 1487 oldnvdevs = sav->sav_count; 1488 sav->sav_vdevs = NULL; 1489 sav->sav_count = 0; 1490 1491 /* 1492 * Process new nvlist of vdevs. 1493 */ 1494 for (i = 0; i < nl2cache; i++) { 1495 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1496 &guid) == 0); 1497 1498 newvdevs[i] = NULL; 1499 for (j = 0; j < oldnvdevs; j++) { 1500 vd = oldvdevs[j]; 1501 if (vd != NULL && guid == vd->vdev_guid) { 1502 /* 1503 * Retain previous vdev for add/remove ops. 1504 */ 1505 newvdevs[i] = vd; 1506 oldvdevs[j] = NULL; 1507 break; 1508 } 1509 } 1510 1511 if (newvdevs[i] == NULL) { 1512 /* 1513 * Create new vdev 1514 */ 1515 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1516 VDEV_ALLOC_L2CACHE) == 0); 1517 ASSERT(vd != NULL); 1518 newvdevs[i] = vd; 1519 1520 /* 1521 * Commit this vdev as an l2cache device, 1522 * even if it fails to open. 1523 */ 1524 spa_l2cache_add(vd); 1525 1526 vd->vdev_top = vd; 1527 vd->vdev_aux = sav; 1528 1529 spa_l2cache_activate(vd); 1530 1531 if (vdev_open(vd) != 0) 1532 continue; 1533 1534 (void) vdev_validate_aux(vd); 1535 1536 if (!vdev_is_dead(vd)) 1537 l2arc_add_vdev(spa, vd); 1538 } 1539 } 1540 1541 /* 1542 * Purge vdevs that were dropped 1543 */ 1544 for (i = 0; i < oldnvdevs; i++) { 1545 uint64_t pool; 1546 1547 vd = oldvdevs[i]; 1548 if (vd != NULL) { 1549 ASSERT(vd->vdev_isl2cache); 1550 1551 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1552 pool != 0ULL && l2arc_vdev_present(vd)) 1553 l2arc_remove_vdev(vd); 1554 vdev_clear_stats(vd); 1555 vdev_free(vd); 1556 } 1557 } 1558 1559 if (oldvdevs) 1560 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1561 1562 if (sav->sav_config == NULL) 1563 goto out; 1564 1565 sav->sav_vdevs = newvdevs; 1566 sav->sav_count = (int)nl2cache; 1567 1568 /* 1569 * Recompute the stashed list of l2cache devices, with status 1570 * information this time. 1571 */ 1572 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1573 DATA_TYPE_NVLIST_ARRAY) == 0); 1574 1575 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1576 for (i = 0; i < sav->sav_count; i++) 1577 l2cache[i] = vdev_config_generate(spa, 1578 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1579 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1580 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1581out: 1582 for (i = 0; i < sav->sav_count; i++) 1583 nvlist_free(l2cache[i]); 1584 if (sav->sav_count) 1585 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1586} 1587 1588static int 1589load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1590{ 1591 dmu_buf_t *db; 1592 char *packed = NULL; 1593 size_t nvsize = 0; 1594 int error; 1595 *value = NULL; 1596 1597 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 1598 nvsize = *(uint64_t *)db->db_data; 1599 dmu_buf_rele(db, FTAG); 1600 1601 packed = kmem_alloc(nvsize, KM_SLEEP); 1602 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1603 DMU_READ_PREFETCH); 1604 if (error == 0) 1605 error = nvlist_unpack(packed, nvsize, value, 0); 1606 kmem_free(packed, nvsize); 1607 1608 return (error); 1609} 1610 1611/* 1612 * Checks to see if the given vdev could not be opened, in which case we post a 1613 * sysevent to notify the autoreplace code that the device has been removed. 1614 */ 1615static void 1616spa_check_removed(vdev_t *vd) 1617{ 1618 for (int c = 0; c < vd->vdev_children; c++) 1619 spa_check_removed(vd->vdev_child[c]); 1620 1621 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1622 !vd->vdev_ishole) { 1623 zfs_post_autoreplace(vd->vdev_spa, vd); 1624 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1625 } 1626} 1627 1628/* 1629 * Validate the current config against the MOS config 1630 */ 1631static boolean_t 1632spa_config_valid(spa_t *spa, nvlist_t *config) 1633{ 1634 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1635 nvlist_t *nv; 1636 1637 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1638 1639 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1640 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1641 1642 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1643 1644 /* 1645 * If we're doing a normal import, then build up any additional 1646 * diagnostic information about missing devices in this config. 1647 * We'll pass this up to the user for further processing. 1648 */ 1649 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1650 nvlist_t **child, *nv; 1651 uint64_t idx = 0; 1652 1653 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1654 KM_SLEEP); 1655 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1656 1657 for (int c = 0; c < rvd->vdev_children; c++) { 1658 vdev_t *tvd = rvd->vdev_child[c]; 1659 vdev_t *mtvd = mrvd->vdev_child[c]; 1660 1661 if (tvd->vdev_ops == &vdev_missing_ops && 1662 mtvd->vdev_ops != &vdev_missing_ops && 1663 mtvd->vdev_islog) 1664 child[idx++] = vdev_config_generate(spa, mtvd, 1665 B_FALSE, 0); 1666 } 1667 1668 if (idx) { 1669 VERIFY(nvlist_add_nvlist_array(nv, 1670 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1671 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1672 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1673 1674 for (int i = 0; i < idx; i++) 1675 nvlist_free(child[i]); 1676 } 1677 nvlist_free(nv); 1678 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1679 } 1680 1681 /* 1682 * Compare the root vdev tree with the information we have 1683 * from the MOS config (mrvd). Check each top-level vdev 1684 * with the corresponding MOS config top-level (mtvd). 1685 */ 1686 for (int c = 0; c < rvd->vdev_children; c++) { 1687 vdev_t *tvd = rvd->vdev_child[c]; 1688 vdev_t *mtvd = mrvd->vdev_child[c]; 1689 1690 /* 1691 * Resolve any "missing" vdevs in the current configuration. 1692 * If we find that the MOS config has more accurate information 1693 * about the top-level vdev then use that vdev instead. 1694 */ 1695 if (tvd->vdev_ops == &vdev_missing_ops && 1696 mtvd->vdev_ops != &vdev_missing_ops) { 1697 1698 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1699 continue; 1700 1701 /* 1702 * Device specific actions. 1703 */ 1704 if (mtvd->vdev_islog) { 1705 spa_set_log_state(spa, SPA_LOG_CLEAR); 1706 } else { 1707 /* 1708 * XXX - once we have 'readonly' pool 1709 * support we should be able to handle 1710 * missing data devices by transitioning 1711 * the pool to readonly. 1712 */ 1713 continue; 1714 } 1715 1716 /* 1717 * Swap the missing vdev with the data we were 1718 * able to obtain from the MOS config. 1719 */ 1720 vdev_remove_child(rvd, tvd); 1721 vdev_remove_child(mrvd, mtvd); 1722 1723 vdev_add_child(rvd, mtvd); 1724 vdev_add_child(mrvd, tvd); 1725 1726 spa_config_exit(spa, SCL_ALL, FTAG); 1727 vdev_load(mtvd); 1728 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1729 1730 vdev_reopen(rvd); 1731 } else if (mtvd->vdev_islog) { 1732 /* 1733 * Load the slog device's state from the MOS config 1734 * since it's possible that the label does not 1735 * contain the most up-to-date information. 1736 */ 1737 vdev_load_log_state(tvd, mtvd); 1738 vdev_reopen(tvd); 1739 } 1740 } 1741 vdev_free(mrvd); 1742 spa_config_exit(spa, SCL_ALL, FTAG); 1743 1744 /* 1745 * Ensure we were able to validate the config. 1746 */ 1747 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1748} 1749 1750/* 1751 * Check for missing log devices 1752 */ 1753static boolean_t 1754spa_check_logs(spa_t *spa) 1755{ 1756 boolean_t rv = B_FALSE; 1757 1758 switch (spa->spa_log_state) { 1759 case SPA_LOG_MISSING: 1760 /* need to recheck in case slog has been restored */ 1761 case SPA_LOG_UNKNOWN: 1762 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1763 NULL, DS_FIND_CHILDREN) != 0); 1764 if (rv) 1765 spa_set_log_state(spa, SPA_LOG_MISSING); 1766 break; 1767 } 1768 return (rv); 1769} 1770 1771static boolean_t 1772spa_passivate_log(spa_t *spa) 1773{ 1774 vdev_t *rvd = spa->spa_root_vdev; 1775 boolean_t slog_found = B_FALSE; 1776 1777 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1778 1779 if (!spa_has_slogs(spa)) 1780 return (B_FALSE); 1781 1782 for (int c = 0; c < rvd->vdev_children; c++) { 1783 vdev_t *tvd = rvd->vdev_child[c]; 1784 metaslab_group_t *mg = tvd->vdev_mg; 1785 1786 if (tvd->vdev_islog) { 1787 metaslab_group_passivate(mg); 1788 slog_found = B_TRUE; 1789 } 1790 } 1791 1792 return (slog_found); 1793} 1794 1795static void 1796spa_activate_log(spa_t *spa) 1797{ 1798 vdev_t *rvd = spa->spa_root_vdev; 1799 1800 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1801 1802 for (int c = 0; c < rvd->vdev_children; c++) { 1803 vdev_t *tvd = rvd->vdev_child[c]; 1804 metaslab_group_t *mg = tvd->vdev_mg; 1805 1806 if (tvd->vdev_islog) 1807 metaslab_group_activate(mg); 1808 } 1809} 1810 1811int 1812spa_offline_log(spa_t *spa) 1813{ 1814 int error; 1815 1816 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1817 NULL, DS_FIND_CHILDREN); 1818 if (error == 0) { 1819 /* 1820 * We successfully offlined the log device, sync out the 1821 * current txg so that the "stubby" block can be removed 1822 * by zil_sync(). 1823 */ 1824 txg_wait_synced(spa->spa_dsl_pool, 0); 1825 } 1826 return (error); 1827} 1828 1829static void 1830spa_aux_check_removed(spa_aux_vdev_t *sav) 1831{ 1832 int i; 1833 1834 for (i = 0; i < sav->sav_count; i++) 1835 spa_check_removed(sav->sav_vdevs[i]); 1836} 1837 1838void 1839spa_claim_notify(zio_t *zio) 1840{ 1841 spa_t *spa = zio->io_spa; 1842 1843 if (zio->io_error) 1844 return; 1845 1846 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1847 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1848 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1849 mutex_exit(&spa->spa_props_lock); 1850} 1851 1852typedef struct spa_load_error { 1853 uint64_t sle_meta_count; 1854 uint64_t sle_data_count; 1855} spa_load_error_t; 1856 1857static void 1858spa_load_verify_done(zio_t *zio) 1859{ 1860 blkptr_t *bp = zio->io_bp; 1861 spa_load_error_t *sle = zio->io_private; 1862 dmu_object_type_t type = BP_GET_TYPE(bp); 1863 int error = zio->io_error; 1864 1865 if (error) { 1866 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1867 type != DMU_OT_INTENT_LOG) 1868 atomic_add_64(&sle->sle_meta_count, 1); 1869 else 1870 atomic_add_64(&sle->sle_data_count, 1); 1871 } 1872 zio_data_buf_free(zio->io_data, zio->io_size); 1873} 1874 1875/*ARGSUSED*/ 1876static int 1877spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1878 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1879{ 1880 if (bp != NULL) { 1881 zio_t *rio = arg; 1882 size_t size = BP_GET_PSIZE(bp); 1883 void *data = zio_data_buf_alloc(size); 1884 1885 zio_nowait(zio_read(rio, spa, bp, data, size, 1886 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1887 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1888 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1889 } 1890 return (0); 1891} 1892 1893static int 1894spa_load_verify(spa_t *spa) 1895{ 1896 zio_t *rio; 1897 spa_load_error_t sle = { 0 }; 1898 zpool_rewind_policy_t policy; 1899 boolean_t verify_ok = B_FALSE; 1900 int error; 1901 1902 zpool_get_rewind_policy(spa->spa_config, &policy); 1903 1904 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1905 return (0); 1906 1907 rio = zio_root(spa, NULL, &sle, 1908 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1909 1910 error = traverse_pool(spa, spa->spa_verify_min_txg, 1911 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1912 1913 (void) zio_wait(rio); 1914 1915 spa->spa_load_meta_errors = sle.sle_meta_count; 1916 spa->spa_load_data_errors = sle.sle_data_count; 1917 1918 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1919 sle.sle_data_count <= policy.zrp_maxdata) { 1920 int64_t loss = 0; 1921 1922 verify_ok = B_TRUE; 1923 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1924 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1925 1926 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1927 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1928 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1929 VERIFY(nvlist_add_int64(spa->spa_load_info, 1930 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1931 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1932 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1933 } else { 1934 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1935 } 1936 1937 if (error) { 1938 if (error != ENXIO && error != EIO) 1939 error = SET_ERROR(EIO); 1940 return (error); 1941 } 1942 1943 return (verify_ok ? 0 : EIO); 1944} 1945 1946/* 1947 * Find a value in the pool props object. 1948 */ 1949static void 1950spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1951{ 1952 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1953 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1954} 1955 1956/* 1957 * Find a value in the pool directory object. 1958 */ 1959static int 1960spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1961{ 1962 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1963 name, sizeof (uint64_t), 1, val)); 1964} 1965 1966static int 1967spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1968{ 1969 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1970 return (err); 1971} 1972 1973/* 1974 * Fix up config after a partly-completed split. This is done with the 1975 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1976 * pool have that entry in their config, but only the splitting one contains 1977 * a list of all the guids of the vdevs that are being split off. 1978 * 1979 * This function determines what to do with that list: either rejoin 1980 * all the disks to the pool, or complete the splitting process. To attempt 1981 * the rejoin, each disk that is offlined is marked online again, and 1982 * we do a reopen() call. If the vdev label for every disk that was 1983 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 1984 * then we call vdev_split() on each disk, and complete the split. 1985 * 1986 * Otherwise we leave the config alone, with all the vdevs in place in 1987 * the original pool. 1988 */ 1989static void 1990spa_try_repair(spa_t *spa, nvlist_t *config) 1991{ 1992 uint_t extracted; 1993 uint64_t *glist; 1994 uint_t i, gcount; 1995 nvlist_t *nvl; 1996 vdev_t **vd; 1997 boolean_t attempt_reopen; 1998 1999 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 2000 return; 2001 2002 /* check that the config is complete */ 2003 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2004 &glist, &gcount) != 0) 2005 return; 2006 2007 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2008 2009 /* attempt to online all the vdevs & validate */ 2010 attempt_reopen = B_TRUE; 2011 for (i = 0; i < gcount; i++) { 2012 if (glist[i] == 0) /* vdev is hole */ 2013 continue; 2014 2015 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2016 if (vd[i] == NULL) { 2017 /* 2018 * Don't bother attempting to reopen the disks; 2019 * just do the split. 2020 */ 2021 attempt_reopen = B_FALSE; 2022 } else { 2023 /* attempt to re-online it */ 2024 vd[i]->vdev_offline = B_FALSE; 2025 } 2026 } 2027 2028 if (attempt_reopen) { 2029 vdev_reopen(spa->spa_root_vdev); 2030 2031 /* check each device to see what state it's in */ 2032 for (extracted = 0, i = 0; i < gcount; i++) { 2033 if (vd[i] != NULL && 2034 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2035 break; 2036 ++extracted; 2037 } 2038 } 2039 2040 /* 2041 * If every disk has been moved to the new pool, or if we never 2042 * even attempted to look at them, then we split them off for 2043 * good. 2044 */ 2045 if (!attempt_reopen || gcount == extracted) { 2046 for (i = 0; i < gcount; i++) 2047 if (vd[i] != NULL) 2048 vdev_split(vd[i]); 2049 vdev_reopen(spa->spa_root_vdev); 2050 } 2051 2052 kmem_free(vd, gcount * sizeof (vdev_t *)); 2053} 2054 2055static int 2056spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2057 boolean_t mosconfig) 2058{ 2059 nvlist_t *config = spa->spa_config; 2060 char *ereport = FM_EREPORT_ZFS_POOL; 2061 char *comment; 2062 int error; 2063 uint64_t pool_guid; 2064 nvlist_t *nvl; 2065 2066 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2067 return (SET_ERROR(EINVAL)); 2068 2069 ASSERT(spa->spa_comment == NULL); 2070 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2071 spa->spa_comment = spa_strdup(comment); 2072 2073 /* 2074 * Versioning wasn't explicitly added to the label until later, so if 2075 * it's not present treat it as the initial version. 2076 */ 2077 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2078 &spa->spa_ubsync.ub_version) != 0) 2079 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2080 2081 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2082 &spa->spa_config_txg); 2083 2084 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2085 spa_guid_exists(pool_guid, 0)) { 2086 error = SET_ERROR(EEXIST); 2087 } else { 2088 spa->spa_config_guid = pool_guid; 2089 2090 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2091 &nvl) == 0) { 2092 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2093 KM_SLEEP) == 0); 2094 } 2095 2096 nvlist_free(spa->spa_load_info); 2097 spa->spa_load_info = fnvlist_alloc(); 2098 2099 gethrestime(&spa->spa_loaded_ts); 2100 error = spa_load_impl(spa, pool_guid, config, state, type, 2101 mosconfig, &ereport); 2102 } 2103 2104 spa->spa_minref = refcount_count(&spa->spa_refcount); 2105 if (error) { 2106 if (error != EEXIST) { 2107 spa->spa_loaded_ts.tv_sec = 0; 2108 spa->spa_loaded_ts.tv_nsec = 0; 2109 } 2110 if (error != EBADF) { 2111 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2112 } 2113 } 2114 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2115 spa->spa_ena = 0; 2116 2117 return (error); 2118} 2119 2120/* 2121 * Load an existing storage pool, using the pool's builtin spa_config as a 2122 * source of configuration information. 2123 */ 2124static int 2125spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2126 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2127 char **ereport) 2128{ 2129 int error = 0; 2130 nvlist_t *nvroot = NULL; 2131 nvlist_t *label; 2132 vdev_t *rvd; 2133 uberblock_t *ub = &spa->spa_uberblock; 2134 uint64_t children, config_cache_txg = spa->spa_config_txg; 2135 int orig_mode = spa->spa_mode; 2136 int parse; 2137 uint64_t obj; 2138 boolean_t missing_feat_write = B_FALSE; 2139 2140 /* 2141 * If this is an untrusted config, access the pool in read-only mode. 2142 * This prevents things like resilvering recently removed devices. 2143 */ 2144 if (!mosconfig) 2145 spa->spa_mode = FREAD; 2146 2147 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2148 2149 spa->spa_load_state = state; 2150 2151 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2152 return (SET_ERROR(EINVAL)); 2153 2154 parse = (type == SPA_IMPORT_EXISTING ? 2155 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2156 2157 /* 2158 * Create "The Godfather" zio to hold all async IOs 2159 */ 2160 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 2161 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 2162 2163 /* 2164 * Parse the configuration into a vdev tree. We explicitly set the 2165 * value that will be returned by spa_version() since parsing the 2166 * configuration requires knowing the version number. 2167 */ 2168 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2169 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2170 spa_config_exit(spa, SCL_ALL, FTAG); 2171 2172 if (error != 0) 2173 return (error); 2174 2175 ASSERT(spa->spa_root_vdev == rvd); 2176 2177 if (type != SPA_IMPORT_ASSEMBLE) { 2178 ASSERT(spa_guid(spa) == pool_guid); 2179 } 2180 2181 /* 2182 * Try to open all vdevs, loading each label in the process. 2183 */ 2184 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2185 error = vdev_open(rvd); 2186 spa_config_exit(spa, SCL_ALL, FTAG); 2187 if (error != 0) 2188 return (error); 2189 2190 /* 2191 * We need to validate the vdev labels against the configuration that 2192 * we have in hand, which is dependent on the setting of mosconfig. If 2193 * mosconfig is true then we're validating the vdev labels based on 2194 * that config. Otherwise, we're validating against the cached config 2195 * (zpool.cache) that was read when we loaded the zfs module, and then 2196 * later we will recursively call spa_load() and validate against 2197 * the vdev config. 2198 * 2199 * If we're assembling a new pool that's been split off from an 2200 * existing pool, the labels haven't yet been updated so we skip 2201 * validation for now. 2202 */ 2203 if (type != SPA_IMPORT_ASSEMBLE) { 2204 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2205 error = vdev_validate(rvd, mosconfig); 2206 spa_config_exit(spa, SCL_ALL, FTAG); 2207 2208 if (error != 0) 2209 return (error); 2210 2211 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2212 return (SET_ERROR(ENXIO)); 2213 } 2214 2215 /* 2216 * Find the best uberblock. 2217 */ 2218 vdev_uberblock_load(rvd, ub, &label); 2219 2220 /* 2221 * If we weren't able to find a single valid uberblock, return failure. 2222 */ 2223 if (ub->ub_txg == 0) { 2224 nvlist_free(label); 2225 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2226 } 2227 2228 /* 2229 * If the pool has an unsupported version we can't open it. 2230 */ 2231 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2232 nvlist_free(label); 2233 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2234 } 2235 2236 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2237 nvlist_t *features; 2238 2239 /* 2240 * If we weren't able to find what's necessary for reading the 2241 * MOS in the label, return failure. 2242 */ 2243 if (label == NULL || nvlist_lookup_nvlist(label, 2244 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2245 nvlist_free(label); 2246 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2247 ENXIO)); 2248 } 2249 2250 /* 2251 * Update our in-core representation with the definitive values 2252 * from the label. 2253 */ 2254 nvlist_free(spa->spa_label_features); 2255 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2256 } 2257 2258 nvlist_free(label); 2259 2260 /* 2261 * Look through entries in the label nvlist's features_for_read. If 2262 * there is a feature listed there which we don't understand then we 2263 * cannot open a pool. 2264 */ 2265 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2266 nvlist_t *unsup_feat; 2267 2268 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2269 0); 2270 2271 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2272 NULL); nvp != NULL; 2273 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2274 if (!zfeature_is_supported(nvpair_name(nvp))) { 2275 VERIFY(nvlist_add_string(unsup_feat, 2276 nvpair_name(nvp), "") == 0); 2277 } 2278 } 2279 2280 if (!nvlist_empty(unsup_feat)) { 2281 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2282 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2283 nvlist_free(unsup_feat); 2284 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2285 ENOTSUP)); 2286 } 2287 2288 nvlist_free(unsup_feat); 2289 } 2290 2291 /* 2292 * If the vdev guid sum doesn't match the uberblock, we have an 2293 * incomplete configuration. We first check to see if the pool 2294 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2295 * If it is, defer the vdev_guid_sum check till later so we 2296 * can handle missing vdevs. 2297 */ 2298 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2299 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2300 rvd->vdev_guid_sum != ub->ub_guid_sum) 2301 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2302 2303 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2304 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2305 spa_try_repair(spa, config); 2306 spa_config_exit(spa, SCL_ALL, FTAG); 2307 nvlist_free(spa->spa_config_splitting); 2308 spa->spa_config_splitting = NULL; 2309 } 2310 2311 /* 2312 * Initialize internal SPA structures. 2313 */ 2314 spa->spa_state = POOL_STATE_ACTIVE; 2315 spa->spa_ubsync = spa->spa_uberblock; 2316 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2317 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2318 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2319 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2320 spa->spa_claim_max_txg = spa->spa_first_txg; 2321 spa->spa_prev_software_version = ub->ub_software_version; 2322 2323 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2324 if (error) 2325 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2326 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2327 2328 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2329 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2330 2331 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2332 boolean_t missing_feat_read = B_FALSE; 2333 nvlist_t *unsup_feat, *enabled_feat; 2334 2335 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2336 &spa->spa_feat_for_read_obj) != 0) { 2337 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2338 } 2339 2340 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2341 &spa->spa_feat_for_write_obj) != 0) { 2342 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2343 } 2344 2345 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2346 &spa->spa_feat_desc_obj) != 0) { 2347 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2348 } 2349 2350 enabled_feat = fnvlist_alloc(); 2351 unsup_feat = fnvlist_alloc(); 2352 2353 if (!feature_is_supported(spa->spa_meta_objset, 2354 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, 2355 unsup_feat, enabled_feat)) 2356 missing_feat_read = B_TRUE; 2357 2358 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2359 if (!feature_is_supported(spa->spa_meta_objset, 2360 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, 2361 unsup_feat, enabled_feat)) { 2362 missing_feat_write = B_TRUE; 2363 } 2364 } 2365 2366 fnvlist_add_nvlist(spa->spa_load_info, 2367 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2368 2369 if (!nvlist_empty(unsup_feat)) { 2370 fnvlist_add_nvlist(spa->spa_load_info, 2371 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2372 } 2373 2374 fnvlist_free(enabled_feat); 2375 fnvlist_free(unsup_feat); 2376 2377 if (!missing_feat_read) { 2378 fnvlist_add_boolean(spa->spa_load_info, 2379 ZPOOL_CONFIG_CAN_RDONLY); 2380 } 2381 2382 /* 2383 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2384 * twofold: to determine whether the pool is available for 2385 * import in read-write mode and (if it is not) whether the 2386 * pool is available for import in read-only mode. If the pool 2387 * is available for import in read-write mode, it is displayed 2388 * as available in userland; if it is not available for import 2389 * in read-only mode, it is displayed as unavailable in 2390 * userland. If the pool is available for import in read-only 2391 * mode but not read-write mode, it is displayed as unavailable 2392 * in userland with a special note that the pool is actually 2393 * available for open in read-only mode. 2394 * 2395 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2396 * missing a feature for write, we must first determine whether 2397 * the pool can be opened read-only before returning to 2398 * userland in order to know whether to display the 2399 * abovementioned note. 2400 */ 2401 if (missing_feat_read || (missing_feat_write && 2402 spa_writeable(spa))) { 2403 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2404 ENOTSUP)); 2405 } 2406 } 2407 2408 spa->spa_is_initializing = B_TRUE; 2409 error = dsl_pool_open(spa->spa_dsl_pool); 2410 spa->spa_is_initializing = B_FALSE; 2411 if (error != 0) 2412 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2413 2414 if (!mosconfig) { 2415 uint64_t hostid; 2416 nvlist_t *policy = NULL, *nvconfig; 2417 2418 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2419 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2420 2421 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2422 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2423 char *hostname; 2424 unsigned long myhostid = 0; 2425 2426 VERIFY(nvlist_lookup_string(nvconfig, 2427 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2428 2429#ifdef _KERNEL 2430 myhostid = zone_get_hostid(NULL); 2431#else /* _KERNEL */ 2432 /* 2433 * We're emulating the system's hostid in userland, so 2434 * we can't use zone_get_hostid(). 2435 */ 2436 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2437#endif /* _KERNEL */ 2438 if (check_hostid && hostid != 0 && myhostid != 0 && 2439 hostid != myhostid) { 2440 nvlist_free(nvconfig); 2441 cmn_err(CE_WARN, "pool '%s' could not be " 2442 "loaded as it was last accessed by " 2443 "another system (host: %s hostid: 0x%lx). " 2444 "See: http://illumos.org/msg/ZFS-8000-EY", 2445 spa_name(spa), hostname, 2446 (unsigned long)hostid); 2447 return (SET_ERROR(EBADF)); 2448 } 2449 } 2450 if (nvlist_lookup_nvlist(spa->spa_config, 2451 ZPOOL_REWIND_POLICY, &policy) == 0) 2452 VERIFY(nvlist_add_nvlist(nvconfig, 2453 ZPOOL_REWIND_POLICY, policy) == 0); 2454 2455 spa_config_set(spa, nvconfig); 2456 spa_unload(spa); 2457 spa_deactivate(spa); 2458 spa_activate(spa, orig_mode); 2459 2460 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2461 } 2462 2463 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2464 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2465 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2466 if (error != 0) 2467 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2468 2469 /* 2470 * Load the bit that tells us to use the new accounting function 2471 * (raid-z deflation). If we have an older pool, this will not 2472 * be present. 2473 */ 2474 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2475 if (error != 0 && error != ENOENT) 2476 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2477 2478 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2479 &spa->spa_creation_version); 2480 if (error != 0 && error != ENOENT) 2481 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2482 2483 /* 2484 * Load the persistent error log. If we have an older pool, this will 2485 * not be present. 2486 */ 2487 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2488 if (error != 0 && error != ENOENT) 2489 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2490 2491 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2492 &spa->spa_errlog_scrub); 2493 if (error != 0 && error != ENOENT) 2494 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2495 2496 /* 2497 * Load the history object. If we have an older pool, this 2498 * will not be present. 2499 */ 2500 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2501 if (error != 0 && error != ENOENT) 2502 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2503 2504 /* 2505 * If we're assembling the pool from the split-off vdevs of 2506 * an existing pool, we don't want to attach the spares & cache 2507 * devices. 2508 */ 2509 2510 /* 2511 * Load any hot spares for this pool. 2512 */ 2513 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2514 if (error != 0 && error != ENOENT) 2515 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2516 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2517 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2518 if (load_nvlist(spa, spa->spa_spares.sav_object, 2519 &spa->spa_spares.sav_config) != 0) 2520 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2521 2522 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2523 spa_load_spares(spa); 2524 spa_config_exit(spa, SCL_ALL, FTAG); 2525 } else if (error == 0) { 2526 spa->spa_spares.sav_sync = B_TRUE; 2527 } 2528 2529 /* 2530 * Load any level 2 ARC devices for this pool. 2531 */ 2532 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2533 &spa->spa_l2cache.sav_object); 2534 if (error != 0 && error != ENOENT) 2535 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2536 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2537 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2538 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2539 &spa->spa_l2cache.sav_config) != 0) 2540 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2541 2542 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2543 spa_load_l2cache(spa); 2544 spa_config_exit(spa, SCL_ALL, FTAG); 2545 } else if (error == 0) { 2546 spa->spa_l2cache.sav_sync = B_TRUE; 2547 } 2548 2549 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2550 2551 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2552 if (error && error != ENOENT) 2553 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2554 2555 if (error == 0) { 2556 uint64_t autoreplace; 2557 2558 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2559 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2560 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2561 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2562 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2563 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2564 &spa->spa_dedup_ditto); 2565 2566 spa->spa_autoreplace = (autoreplace != 0); 2567 } 2568 2569 /* 2570 * If the 'autoreplace' property is set, then post a resource notifying 2571 * the ZFS DE that it should not issue any faults for unopenable 2572 * devices. We also iterate over the vdevs, and post a sysevent for any 2573 * unopenable vdevs so that the normal autoreplace handler can take 2574 * over. 2575 */ 2576 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2577 spa_check_removed(spa->spa_root_vdev); 2578 /* 2579 * For the import case, this is done in spa_import(), because 2580 * at this point we're using the spare definitions from 2581 * the MOS config, not necessarily from the userland config. 2582 */ 2583 if (state != SPA_LOAD_IMPORT) { 2584 spa_aux_check_removed(&spa->spa_spares); 2585 spa_aux_check_removed(&spa->spa_l2cache); 2586 } 2587 } 2588 2589 /* 2590 * Load the vdev state for all toplevel vdevs. 2591 */ 2592 vdev_load(rvd); 2593 2594 /* 2595 * Propagate the leaf DTLs we just loaded all the way up the tree. 2596 */ 2597 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2598 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2599 spa_config_exit(spa, SCL_ALL, FTAG); 2600 2601 /* 2602 * Load the DDTs (dedup tables). 2603 */ 2604 error = ddt_load(spa); 2605 if (error != 0) 2606 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2607 2608 spa_update_dspace(spa); 2609 2610 /* 2611 * Validate the config, using the MOS config to fill in any 2612 * information which might be missing. If we fail to validate 2613 * the config then declare the pool unfit for use. If we're 2614 * assembling a pool from a split, the log is not transferred 2615 * over. 2616 */ 2617 if (type != SPA_IMPORT_ASSEMBLE) { 2618 nvlist_t *nvconfig; 2619 2620 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2621 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2622 2623 if (!spa_config_valid(spa, nvconfig)) { 2624 nvlist_free(nvconfig); 2625 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2626 ENXIO)); 2627 } 2628 nvlist_free(nvconfig); 2629 2630 /* 2631 * Now that we've validated the config, check the state of the 2632 * root vdev. If it can't be opened, it indicates one or 2633 * more toplevel vdevs are faulted. 2634 */ 2635 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2636 return (SET_ERROR(ENXIO)); 2637 2638 if (spa_check_logs(spa)) { 2639 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2640 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2641 } 2642 } 2643 2644 if (missing_feat_write) { 2645 ASSERT(state == SPA_LOAD_TRYIMPORT); 2646 2647 /* 2648 * At this point, we know that we can open the pool in 2649 * read-only mode but not read-write mode. We now have enough 2650 * information and can return to userland. 2651 */ 2652 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2653 } 2654 2655 /* 2656 * We've successfully opened the pool, verify that we're ready 2657 * to start pushing transactions. 2658 */ 2659 if (state != SPA_LOAD_TRYIMPORT) { 2660 if (error = spa_load_verify(spa)) 2661 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2662 error)); 2663 } 2664 2665 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2666 spa->spa_load_max_txg == UINT64_MAX)) { 2667 dmu_tx_t *tx; 2668 int need_update = B_FALSE; 2669 2670 ASSERT(state != SPA_LOAD_TRYIMPORT); 2671 2672 /* 2673 * Claim log blocks that haven't been committed yet. 2674 * This must all happen in a single txg. 2675 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2676 * invoked from zil_claim_log_block()'s i/o done callback. 2677 * Price of rollback is that we abandon the log. 2678 */ 2679 spa->spa_claiming = B_TRUE; 2680 2681 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2682 spa_first_txg(spa)); 2683 (void) dmu_objset_find(spa_name(spa), 2684 zil_claim, tx, DS_FIND_CHILDREN); 2685 dmu_tx_commit(tx); 2686 2687 spa->spa_claiming = B_FALSE; 2688 2689 spa_set_log_state(spa, SPA_LOG_GOOD); 2690 spa->spa_sync_on = B_TRUE; 2691 txg_sync_start(spa->spa_dsl_pool); 2692 2693 /* 2694 * Wait for all claims to sync. We sync up to the highest 2695 * claimed log block birth time so that claimed log blocks 2696 * don't appear to be from the future. spa_claim_max_txg 2697 * will have been set for us by either zil_check_log_chain() 2698 * (invoked from spa_check_logs()) or zil_claim() above. 2699 */ 2700 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2701 2702 /* 2703 * If the config cache is stale, or we have uninitialized 2704 * metaslabs (see spa_vdev_add()), then update the config. 2705 * 2706 * If this is a verbatim import, trust the current 2707 * in-core spa_config and update the disk labels. 2708 */ 2709 if (config_cache_txg != spa->spa_config_txg || 2710 state == SPA_LOAD_IMPORT || 2711 state == SPA_LOAD_RECOVER || 2712 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2713 need_update = B_TRUE; 2714 2715 for (int c = 0; c < rvd->vdev_children; c++) 2716 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2717 need_update = B_TRUE; 2718 2719 /* 2720 * Update the config cache asychronously in case we're the 2721 * root pool, in which case the config cache isn't writable yet. 2722 */ 2723 if (need_update) 2724 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2725 2726 /* 2727 * Check all DTLs to see if anything needs resilvering. 2728 */ 2729 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2730 vdev_resilver_needed(rvd, NULL, NULL)) 2731 spa_async_request(spa, SPA_ASYNC_RESILVER); 2732 2733 /* 2734 * Log the fact that we booted up (so that we can detect if 2735 * we rebooted in the middle of an operation). 2736 */ 2737 spa_history_log_version(spa, "open"); 2738 2739 /* 2740 * Delete any inconsistent datasets. 2741 */ 2742 (void) dmu_objset_find(spa_name(spa), 2743 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2744 2745 /* 2746 * Clean up any stale temporary dataset userrefs. 2747 */ 2748 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2749 } 2750 2751 return (0); 2752} 2753 2754static int 2755spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2756{ 2757 int mode = spa->spa_mode; 2758 2759 spa_unload(spa); 2760 spa_deactivate(spa); 2761 2762 spa->spa_load_max_txg--; 2763 2764 spa_activate(spa, mode); 2765 spa_async_suspend(spa); 2766 2767 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2768} 2769 2770/* 2771 * If spa_load() fails this function will try loading prior txg's. If 2772 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2773 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2774 * function will not rewind the pool and will return the same error as 2775 * spa_load(). 2776 */ 2777static int 2778spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2779 uint64_t max_request, int rewind_flags) 2780{ 2781 nvlist_t *loadinfo = NULL; 2782 nvlist_t *config = NULL; 2783 int load_error, rewind_error; 2784 uint64_t safe_rewind_txg; 2785 uint64_t min_txg; 2786 2787 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2788 spa->spa_load_max_txg = spa->spa_load_txg; 2789 spa_set_log_state(spa, SPA_LOG_CLEAR); 2790 } else { 2791 spa->spa_load_max_txg = max_request; 2792 } 2793 2794 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2795 mosconfig); 2796 if (load_error == 0) 2797 return (0); 2798 2799 if (spa->spa_root_vdev != NULL) 2800 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2801 2802 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2803 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2804 2805 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2806 nvlist_free(config); 2807 return (load_error); 2808 } 2809 2810 if (state == SPA_LOAD_RECOVER) { 2811 /* Price of rolling back is discarding txgs, including log */ 2812 spa_set_log_state(spa, SPA_LOG_CLEAR); 2813 } else { 2814 /* 2815 * If we aren't rolling back save the load info from our first 2816 * import attempt so that we can restore it after attempting 2817 * to rewind. 2818 */ 2819 loadinfo = spa->spa_load_info; 2820 spa->spa_load_info = fnvlist_alloc(); 2821 } 2822 2823 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2824 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2825 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2826 TXG_INITIAL : safe_rewind_txg; 2827 2828 /* 2829 * Continue as long as we're finding errors, we're still within 2830 * the acceptable rewind range, and we're still finding uberblocks 2831 */ 2832 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2833 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2834 if (spa->spa_load_max_txg < safe_rewind_txg) 2835 spa->spa_extreme_rewind = B_TRUE; 2836 rewind_error = spa_load_retry(spa, state, mosconfig); 2837 } 2838 2839 spa->spa_extreme_rewind = B_FALSE; 2840 spa->spa_load_max_txg = UINT64_MAX; 2841 2842 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2843 spa_config_set(spa, config); 2844 2845 if (state == SPA_LOAD_RECOVER) { 2846 ASSERT3P(loadinfo, ==, NULL); 2847 return (rewind_error); 2848 } else { 2849 /* Store the rewind info as part of the initial load info */ 2850 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2851 spa->spa_load_info); 2852 2853 /* Restore the initial load info */ 2854 fnvlist_free(spa->spa_load_info); 2855 spa->spa_load_info = loadinfo; 2856 2857 return (load_error); 2858 } 2859} 2860 2861/* 2862 * Pool Open/Import 2863 * 2864 * The import case is identical to an open except that the configuration is sent 2865 * down from userland, instead of grabbed from the configuration cache. For the 2866 * case of an open, the pool configuration will exist in the 2867 * POOL_STATE_UNINITIALIZED state. 2868 * 2869 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2870 * the same time open the pool, without having to keep around the spa_t in some 2871 * ambiguous state. 2872 */ 2873static int 2874spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2875 nvlist_t **config) 2876{ 2877 spa_t *spa; 2878 spa_load_state_t state = SPA_LOAD_OPEN; 2879 int error; 2880 int locked = B_FALSE; 2881 int firstopen = B_FALSE; 2882 2883 *spapp = NULL; 2884 2885 /* 2886 * As disgusting as this is, we need to support recursive calls to this 2887 * function because dsl_dir_open() is called during spa_load(), and ends 2888 * up calling spa_open() again. The real fix is to figure out how to 2889 * avoid dsl_dir_open() calling this in the first place. 2890 */ 2891 if (mutex_owner(&spa_namespace_lock) != curthread) { 2892 mutex_enter(&spa_namespace_lock); 2893 locked = B_TRUE; 2894 } 2895 2896 if ((spa = spa_lookup(pool)) == NULL) { 2897 if (locked) 2898 mutex_exit(&spa_namespace_lock); 2899 return (SET_ERROR(ENOENT)); 2900 } 2901 2902 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2903 zpool_rewind_policy_t policy; 2904 2905 firstopen = B_TRUE; 2906 2907 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2908 &policy); 2909 if (policy.zrp_request & ZPOOL_DO_REWIND) 2910 state = SPA_LOAD_RECOVER; 2911 2912 spa_activate(spa, spa_mode_global); 2913 2914 if (state != SPA_LOAD_RECOVER) 2915 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2916 2917 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2918 policy.zrp_request); 2919 2920 if (error == EBADF) { 2921 /* 2922 * If vdev_validate() returns failure (indicated by 2923 * EBADF), it indicates that one of the vdevs indicates 2924 * that the pool has been exported or destroyed. If 2925 * this is the case, the config cache is out of sync and 2926 * we should remove the pool from the namespace. 2927 */ 2928 spa_unload(spa); 2929 spa_deactivate(spa); 2930 spa_config_sync(spa, B_TRUE, B_TRUE); 2931 spa_remove(spa); 2932 if (locked) 2933 mutex_exit(&spa_namespace_lock); 2934 return (SET_ERROR(ENOENT)); 2935 } 2936 2937 if (error) { 2938 /* 2939 * We can't open the pool, but we still have useful 2940 * information: the state of each vdev after the 2941 * attempted vdev_open(). Return this to the user. 2942 */ 2943 if (config != NULL && spa->spa_config) { 2944 VERIFY(nvlist_dup(spa->spa_config, config, 2945 KM_SLEEP) == 0); 2946 VERIFY(nvlist_add_nvlist(*config, 2947 ZPOOL_CONFIG_LOAD_INFO, 2948 spa->spa_load_info) == 0); 2949 } 2950 spa_unload(spa); 2951 spa_deactivate(spa); 2952 spa->spa_last_open_failed = error; 2953 if (locked) 2954 mutex_exit(&spa_namespace_lock); 2955 *spapp = NULL; 2956 return (error); 2957 } 2958 } 2959 2960 spa_open_ref(spa, tag); 2961 2962 if (config != NULL) 2963 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2964 2965 /* 2966 * If we've recovered the pool, pass back any information we 2967 * gathered while doing the load. 2968 */ 2969 if (state == SPA_LOAD_RECOVER) { 2970 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 2971 spa->spa_load_info) == 0); 2972 } 2973 2974 if (locked) { 2975 spa->spa_last_open_failed = 0; 2976 spa->spa_last_ubsync_txg = 0; 2977 spa->spa_load_txg = 0; 2978 mutex_exit(&spa_namespace_lock); 2979#ifdef __FreeBSD__ 2980#ifdef _KERNEL 2981 if (firstopen) 2982 zvol_create_minors(spa->spa_name); 2983#endif 2984#endif 2985 } 2986 2987 *spapp = spa; 2988 2989 return (0); 2990} 2991 2992int 2993spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2994 nvlist_t **config) 2995{ 2996 return (spa_open_common(name, spapp, tag, policy, config)); 2997} 2998 2999int 3000spa_open(const char *name, spa_t **spapp, void *tag) 3001{ 3002 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3003} 3004 3005/* 3006 * Lookup the given spa_t, incrementing the inject count in the process, 3007 * preventing it from being exported or destroyed. 3008 */ 3009spa_t * 3010spa_inject_addref(char *name) 3011{ 3012 spa_t *spa; 3013 3014 mutex_enter(&spa_namespace_lock); 3015 if ((spa = spa_lookup(name)) == NULL) { 3016 mutex_exit(&spa_namespace_lock); 3017 return (NULL); 3018 } 3019 spa->spa_inject_ref++; 3020 mutex_exit(&spa_namespace_lock); 3021 3022 return (spa); 3023} 3024 3025void 3026spa_inject_delref(spa_t *spa) 3027{ 3028 mutex_enter(&spa_namespace_lock); 3029 spa->spa_inject_ref--; 3030 mutex_exit(&spa_namespace_lock); 3031} 3032 3033/* 3034 * Add spares device information to the nvlist. 3035 */ 3036static void 3037spa_add_spares(spa_t *spa, nvlist_t *config) 3038{ 3039 nvlist_t **spares; 3040 uint_t i, nspares; 3041 nvlist_t *nvroot; 3042 uint64_t guid; 3043 vdev_stat_t *vs; 3044 uint_t vsc; 3045 uint64_t pool; 3046 3047 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3048 3049 if (spa->spa_spares.sav_count == 0) 3050 return; 3051 3052 VERIFY(nvlist_lookup_nvlist(config, 3053 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3054 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3055 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3056 if (nspares != 0) { 3057 VERIFY(nvlist_add_nvlist_array(nvroot, 3058 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3059 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3060 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3061 3062 /* 3063 * Go through and find any spares which have since been 3064 * repurposed as an active spare. If this is the case, update 3065 * their status appropriately. 3066 */ 3067 for (i = 0; i < nspares; i++) { 3068 VERIFY(nvlist_lookup_uint64(spares[i], 3069 ZPOOL_CONFIG_GUID, &guid) == 0); 3070 if (spa_spare_exists(guid, &pool, NULL) && 3071 pool != 0ULL) { 3072 VERIFY(nvlist_lookup_uint64_array( 3073 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3074 (uint64_t **)&vs, &vsc) == 0); 3075 vs->vs_state = VDEV_STATE_CANT_OPEN; 3076 vs->vs_aux = VDEV_AUX_SPARED; 3077 } 3078 } 3079 } 3080} 3081 3082/* 3083 * Add l2cache device information to the nvlist, including vdev stats. 3084 */ 3085static void 3086spa_add_l2cache(spa_t *spa, nvlist_t *config) 3087{ 3088 nvlist_t **l2cache; 3089 uint_t i, j, nl2cache; 3090 nvlist_t *nvroot; 3091 uint64_t guid; 3092 vdev_t *vd; 3093 vdev_stat_t *vs; 3094 uint_t vsc; 3095 3096 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3097 3098 if (spa->spa_l2cache.sav_count == 0) 3099 return; 3100 3101 VERIFY(nvlist_lookup_nvlist(config, 3102 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3103 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3104 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3105 if (nl2cache != 0) { 3106 VERIFY(nvlist_add_nvlist_array(nvroot, 3107 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3108 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3109 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3110 3111 /* 3112 * Update level 2 cache device stats. 3113 */ 3114 3115 for (i = 0; i < nl2cache; i++) { 3116 VERIFY(nvlist_lookup_uint64(l2cache[i], 3117 ZPOOL_CONFIG_GUID, &guid) == 0); 3118 3119 vd = NULL; 3120 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3121 if (guid == 3122 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3123 vd = spa->spa_l2cache.sav_vdevs[j]; 3124 break; 3125 } 3126 } 3127 ASSERT(vd != NULL); 3128 3129 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3130 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3131 == 0); 3132 vdev_get_stats(vd, vs); 3133 } 3134 } 3135} 3136 3137static void 3138spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3139{ 3140 nvlist_t *features; 3141 zap_cursor_t zc; 3142 zap_attribute_t za; 3143 3144 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3145 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3146 3147 /* We may be unable to read features if pool is suspended. */ 3148 if (spa_suspended(spa)) 3149 goto out; 3150 3151 if (spa->spa_feat_for_read_obj != 0) { 3152 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3153 spa->spa_feat_for_read_obj); 3154 zap_cursor_retrieve(&zc, &za) == 0; 3155 zap_cursor_advance(&zc)) { 3156 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3157 za.za_num_integers == 1); 3158 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3159 za.za_first_integer)); 3160 } 3161 zap_cursor_fini(&zc); 3162 } 3163 3164 if (spa->spa_feat_for_write_obj != 0) { 3165 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3166 spa->spa_feat_for_write_obj); 3167 zap_cursor_retrieve(&zc, &za) == 0; 3168 zap_cursor_advance(&zc)) { 3169 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3170 za.za_num_integers == 1); 3171 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3172 za.za_first_integer)); 3173 } 3174 zap_cursor_fini(&zc); 3175 } 3176 3177out: 3178 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3179 features) == 0); 3180 nvlist_free(features); 3181} 3182 3183int 3184spa_get_stats(const char *name, nvlist_t **config, 3185 char *altroot, size_t buflen) 3186{ 3187 int error; 3188 spa_t *spa; 3189 3190 *config = NULL; 3191 error = spa_open_common(name, &spa, FTAG, NULL, config); 3192 3193 if (spa != NULL) { 3194 /* 3195 * This still leaves a window of inconsistency where the spares 3196 * or l2cache devices could change and the config would be 3197 * self-inconsistent. 3198 */ 3199 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3200 3201 if (*config != NULL) { 3202 uint64_t loadtimes[2]; 3203 3204 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3205 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3206 VERIFY(nvlist_add_uint64_array(*config, 3207 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3208 3209 VERIFY(nvlist_add_uint64(*config, 3210 ZPOOL_CONFIG_ERRCOUNT, 3211 spa_get_errlog_size(spa)) == 0); 3212 3213 if (spa_suspended(spa)) 3214 VERIFY(nvlist_add_uint64(*config, 3215 ZPOOL_CONFIG_SUSPENDED, 3216 spa->spa_failmode) == 0); 3217 3218 spa_add_spares(spa, *config); 3219 spa_add_l2cache(spa, *config); 3220 spa_add_feature_stats(spa, *config); 3221 } 3222 } 3223 3224 /* 3225 * We want to get the alternate root even for faulted pools, so we cheat 3226 * and call spa_lookup() directly. 3227 */ 3228 if (altroot) { 3229 if (spa == NULL) { 3230 mutex_enter(&spa_namespace_lock); 3231 spa = spa_lookup(name); 3232 if (spa) 3233 spa_altroot(spa, altroot, buflen); 3234 else 3235 altroot[0] = '\0'; 3236 spa = NULL; 3237 mutex_exit(&spa_namespace_lock); 3238 } else { 3239 spa_altroot(spa, altroot, buflen); 3240 } 3241 } 3242 3243 if (spa != NULL) { 3244 spa_config_exit(spa, SCL_CONFIG, FTAG); 3245 spa_close(spa, FTAG); 3246 } 3247 3248 return (error); 3249} 3250 3251/* 3252 * Validate that the auxiliary device array is well formed. We must have an 3253 * array of nvlists, each which describes a valid leaf vdev. If this is an 3254 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3255 * specified, as long as they are well-formed. 3256 */ 3257static int 3258spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3259 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3260 vdev_labeltype_t label) 3261{ 3262 nvlist_t **dev; 3263 uint_t i, ndev; 3264 vdev_t *vd; 3265 int error; 3266 3267 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3268 3269 /* 3270 * It's acceptable to have no devs specified. 3271 */ 3272 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3273 return (0); 3274 3275 if (ndev == 0) 3276 return (SET_ERROR(EINVAL)); 3277 3278 /* 3279 * Make sure the pool is formatted with a version that supports this 3280 * device type. 3281 */ 3282 if (spa_version(spa) < version) 3283 return (SET_ERROR(ENOTSUP)); 3284 3285 /* 3286 * Set the pending device list so we correctly handle device in-use 3287 * checking. 3288 */ 3289 sav->sav_pending = dev; 3290 sav->sav_npending = ndev; 3291 3292 for (i = 0; i < ndev; i++) { 3293 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3294 mode)) != 0) 3295 goto out; 3296 3297 if (!vd->vdev_ops->vdev_op_leaf) { 3298 vdev_free(vd); 3299 error = SET_ERROR(EINVAL); 3300 goto out; 3301 } 3302 3303 /* 3304 * The L2ARC currently only supports disk devices in 3305 * kernel context. For user-level testing, we allow it. 3306 */ 3307#ifdef _KERNEL 3308 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3309 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3310 error = SET_ERROR(ENOTBLK); 3311 vdev_free(vd); 3312 goto out; 3313 } 3314#endif 3315 vd->vdev_top = vd; 3316 3317 if ((error = vdev_open(vd)) == 0 && 3318 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3319 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3320 vd->vdev_guid) == 0); 3321 } 3322 3323 vdev_free(vd); 3324 3325 if (error && 3326 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3327 goto out; 3328 else 3329 error = 0; 3330 } 3331 3332out: 3333 sav->sav_pending = NULL; 3334 sav->sav_npending = 0; 3335 return (error); 3336} 3337 3338static int 3339spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3340{ 3341 int error; 3342 3343 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3344 3345 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3346 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3347 VDEV_LABEL_SPARE)) != 0) { 3348 return (error); 3349 } 3350 3351 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3352 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3353 VDEV_LABEL_L2CACHE)); 3354} 3355 3356static void 3357spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3358 const char *config) 3359{ 3360 int i; 3361 3362 if (sav->sav_config != NULL) { 3363 nvlist_t **olddevs; 3364 uint_t oldndevs; 3365 nvlist_t **newdevs; 3366 3367 /* 3368 * Generate new dev list by concatentating with the 3369 * current dev list. 3370 */ 3371 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3372 &olddevs, &oldndevs) == 0); 3373 3374 newdevs = kmem_alloc(sizeof (void *) * 3375 (ndevs + oldndevs), KM_SLEEP); 3376 for (i = 0; i < oldndevs; i++) 3377 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3378 KM_SLEEP) == 0); 3379 for (i = 0; i < ndevs; i++) 3380 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3381 KM_SLEEP) == 0); 3382 3383 VERIFY(nvlist_remove(sav->sav_config, config, 3384 DATA_TYPE_NVLIST_ARRAY) == 0); 3385 3386 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3387 config, newdevs, ndevs + oldndevs) == 0); 3388 for (i = 0; i < oldndevs + ndevs; i++) 3389 nvlist_free(newdevs[i]); 3390 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3391 } else { 3392 /* 3393 * Generate a new dev list. 3394 */ 3395 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3396 KM_SLEEP) == 0); 3397 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3398 devs, ndevs) == 0); 3399 } 3400} 3401 3402/* 3403 * Stop and drop level 2 ARC devices 3404 */ 3405void 3406spa_l2cache_drop(spa_t *spa) 3407{ 3408 vdev_t *vd; 3409 int i; 3410 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3411 3412 for (i = 0; i < sav->sav_count; i++) { 3413 uint64_t pool; 3414 3415 vd = sav->sav_vdevs[i]; 3416 ASSERT(vd != NULL); 3417 3418 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3419 pool != 0ULL && l2arc_vdev_present(vd)) 3420 l2arc_remove_vdev(vd); 3421 } 3422} 3423 3424/* 3425 * Pool Creation 3426 */ 3427int 3428spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3429 nvlist_t *zplprops) 3430{ 3431 spa_t *spa; 3432 char *altroot = NULL; 3433 vdev_t *rvd; 3434 dsl_pool_t *dp; 3435 dmu_tx_t *tx; 3436 int error = 0; 3437 uint64_t txg = TXG_INITIAL; 3438 nvlist_t **spares, **l2cache; 3439 uint_t nspares, nl2cache; 3440 uint64_t version, obj; 3441 boolean_t has_features; 3442 3443 /* 3444 * If this pool already exists, return failure. 3445 */ 3446 mutex_enter(&spa_namespace_lock); 3447 if (spa_lookup(pool) != NULL) { 3448 mutex_exit(&spa_namespace_lock); 3449 return (SET_ERROR(EEXIST)); 3450 } 3451 3452 /* 3453 * Allocate a new spa_t structure. 3454 */ 3455 (void) nvlist_lookup_string(props, 3456 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3457 spa = spa_add(pool, NULL, altroot); 3458 spa_activate(spa, spa_mode_global); 3459 3460 if (props && (error = spa_prop_validate(spa, props))) { 3461 spa_deactivate(spa); 3462 spa_remove(spa); 3463 mutex_exit(&spa_namespace_lock); 3464 return (error); 3465 } 3466 3467 has_features = B_FALSE; 3468 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3469 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3470 if (zpool_prop_feature(nvpair_name(elem))) 3471 has_features = B_TRUE; 3472 } 3473 3474 if (has_features || nvlist_lookup_uint64(props, 3475 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3476 version = SPA_VERSION; 3477 } 3478 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3479 3480 spa->spa_first_txg = txg; 3481 spa->spa_uberblock.ub_txg = txg - 1; 3482 spa->spa_uberblock.ub_version = version; 3483 spa->spa_ubsync = spa->spa_uberblock; 3484 3485 /* 3486 * Create "The Godfather" zio to hold all async IOs 3487 */ 3488 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 3489 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 3490 3491 /* 3492 * Create the root vdev. 3493 */ 3494 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3495 3496 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3497 3498 ASSERT(error != 0 || rvd != NULL); 3499 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3500 3501 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3502 error = SET_ERROR(EINVAL); 3503 3504 if (error == 0 && 3505 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3506 (error = spa_validate_aux(spa, nvroot, txg, 3507 VDEV_ALLOC_ADD)) == 0) { 3508 for (int c = 0; c < rvd->vdev_children; c++) { 3509 vdev_ashift_optimize(rvd->vdev_child[c]); 3510 vdev_metaslab_set_size(rvd->vdev_child[c]); 3511 vdev_expand(rvd->vdev_child[c], txg); 3512 } 3513 } 3514 3515 spa_config_exit(spa, SCL_ALL, FTAG); 3516 3517 if (error != 0) { 3518 spa_unload(spa); 3519 spa_deactivate(spa); 3520 spa_remove(spa); 3521 mutex_exit(&spa_namespace_lock); 3522 return (error); 3523 } 3524 3525 /* 3526 * Get the list of spares, if specified. 3527 */ 3528 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3529 &spares, &nspares) == 0) { 3530 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3531 KM_SLEEP) == 0); 3532 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3533 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3534 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3535 spa_load_spares(spa); 3536 spa_config_exit(spa, SCL_ALL, FTAG); 3537 spa->spa_spares.sav_sync = B_TRUE; 3538 } 3539 3540 /* 3541 * Get the list of level 2 cache devices, if specified. 3542 */ 3543 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3544 &l2cache, &nl2cache) == 0) { 3545 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3546 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3547 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3548 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3549 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3550 spa_load_l2cache(spa); 3551 spa_config_exit(spa, SCL_ALL, FTAG); 3552 spa->spa_l2cache.sav_sync = B_TRUE; 3553 } 3554 3555 spa->spa_is_initializing = B_TRUE; 3556 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3557 spa->spa_meta_objset = dp->dp_meta_objset; 3558 spa->spa_is_initializing = B_FALSE; 3559 3560 /* 3561 * Create DDTs (dedup tables). 3562 */ 3563 ddt_create(spa); 3564 3565 spa_update_dspace(spa); 3566 3567 tx = dmu_tx_create_assigned(dp, txg); 3568 3569 /* 3570 * Create the pool config object. 3571 */ 3572 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3573 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3574 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3575 3576 if (zap_add(spa->spa_meta_objset, 3577 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3578 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3579 cmn_err(CE_PANIC, "failed to add pool config"); 3580 } 3581 3582 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3583 spa_feature_create_zap_objects(spa, tx); 3584 3585 if (zap_add(spa->spa_meta_objset, 3586 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3587 sizeof (uint64_t), 1, &version, tx) != 0) { 3588 cmn_err(CE_PANIC, "failed to add pool version"); 3589 } 3590 3591 /* Newly created pools with the right version are always deflated. */ 3592 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3593 spa->spa_deflate = TRUE; 3594 if (zap_add(spa->spa_meta_objset, 3595 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3596 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3597 cmn_err(CE_PANIC, "failed to add deflate"); 3598 } 3599 } 3600 3601 /* 3602 * Create the deferred-free bpobj. Turn off compression 3603 * because sync-to-convergence takes longer if the blocksize 3604 * keeps changing. 3605 */ 3606 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3607 dmu_object_set_compress(spa->spa_meta_objset, obj, 3608 ZIO_COMPRESS_OFF, tx); 3609 if (zap_add(spa->spa_meta_objset, 3610 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3611 sizeof (uint64_t), 1, &obj, tx) != 0) { 3612 cmn_err(CE_PANIC, "failed to add bpobj"); 3613 } 3614 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3615 spa->spa_meta_objset, obj)); 3616 3617 /* 3618 * Create the pool's history object. 3619 */ 3620 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3621 spa_history_create_obj(spa, tx); 3622 3623 /* 3624 * Set pool properties. 3625 */ 3626 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3627 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3628 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3629 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3630 3631 if (props != NULL) { 3632 spa_configfile_set(spa, props, B_FALSE); 3633 spa_sync_props(props, tx); 3634 } 3635 3636 dmu_tx_commit(tx); 3637 3638 spa->spa_sync_on = B_TRUE; 3639 txg_sync_start(spa->spa_dsl_pool); 3640 3641 /* 3642 * We explicitly wait for the first transaction to complete so that our 3643 * bean counters are appropriately updated. 3644 */ 3645 txg_wait_synced(spa->spa_dsl_pool, txg); 3646 3647 spa_config_sync(spa, B_FALSE, B_TRUE); 3648 3649 spa_history_log_version(spa, "create"); 3650 3651 spa->spa_minref = refcount_count(&spa->spa_refcount); 3652 3653 mutex_exit(&spa_namespace_lock); 3654 3655 return (0); 3656} 3657 3658#ifdef _KERNEL 3659#if defined(sun) 3660/* 3661 * Get the root pool information from the root disk, then import the root pool 3662 * during the system boot up time. 3663 */ 3664extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3665 3666static nvlist_t * 3667spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3668{ 3669 nvlist_t *config; 3670 nvlist_t *nvtop, *nvroot; 3671 uint64_t pgid; 3672 3673 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3674 return (NULL); 3675 3676 /* 3677 * Add this top-level vdev to the child array. 3678 */ 3679 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3680 &nvtop) == 0); 3681 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3682 &pgid) == 0); 3683 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3684 3685 /* 3686 * Put this pool's top-level vdevs into a root vdev. 3687 */ 3688 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3689 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3690 VDEV_TYPE_ROOT) == 0); 3691 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3692 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3693 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3694 &nvtop, 1) == 0); 3695 3696 /* 3697 * Replace the existing vdev_tree with the new root vdev in 3698 * this pool's configuration (remove the old, add the new). 3699 */ 3700 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3701 nvlist_free(nvroot); 3702 return (config); 3703} 3704 3705/* 3706 * Walk the vdev tree and see if we can find a device with "better" 3707 * configuration. A configuration is "better" if the label on that 3708 * device has a more recent txg. 3709 */ 3710static void 3711spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3712{ 3713 for (int c = 0; c < vd->vdev_children; c++) 3714 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3715 3716 if (vd->vdev_ops->vdev_op_leaf) { 3717 nvlist_t *label; 3718 uint64_t label_txg; 3719 3720 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3721 &label) != 0) 3722 return; 3723 3724 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3725 &label_txg) == 0); 3726 3727 /* 3728 * Do we have a better boot device? 3729 */ 3730 if (label_txg > *txg) { 3731 *txg = label_txg; 3732 *avd = vd; 3733 } 3734 nvlist_free(label); 3735 } 3736} 3737 3738/* 3739 * Import a root pool. 3740 * 3741 * For x86. devpath_list will consist of devid and/or physpath name of 3742 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3743 * The GRUB "findroot" command will return the vdev we should boot. 3744 * 3745 * For Sparc, devpath_list consists the physpath name of the booting device 3746 * no matter the rootpool is a single device pool or a mirrored pool. 3747 * e.g. 3748 * "/pci@1f,0/ide@d/disk@0,0:a" 3749 */ 3750int 3751spa_import_rootpool(char *devpath, char *devid) 3752{ 3753 spa_t *spa; 3754 vdev_t *rvd, *bvd, *avd = NULL; 3755 nvlist_t *config, *nvtop; 3756 uint64_t guid, txg; 3757 char *pname; 3758 int error; 3759 3760 /* 3761 * Read the label from the boot device and generate a configuration. 3762 */ 3763 config = spa_generate_rootconf(devpath, devid, &guid); 3764#if defined(_OBP) && defined(_KERNEL) 3765 if (config == NULL) { 3766 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3767 /* iscsi boot */ 3768 get_iscsi_bootpath_phy(devpath); 3769 config = spa_generate_rootconf(devpath, devid, &guid); 3770 } 3771 } 3772#endif 3773 if (config == NULL) { 3774 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3775 devpath); 3776 return (SET_ERROR(EIO)); 3777 } 3778 3779 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3780 &pname) == 0); 3781 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3782 3783 mutex_enter(&spa_namespace_lock); 3784 if ((spa = spa_lookup(pname)) != NULL) { 3785 /* 3786 * Remove the existing root pool from the namespace so that we 3787 * can replace it with the correct config we just read in. 3788 */ 3789 spa_remove(spa); 3790 } 3791 3792 spa = spa_add(pname, config, NULL); 3793 spa->spa_is_root = B_TRUE; 3794 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3795 3796 /* 3797 * Build up a vdev tree based on the boot device's label config. 3798 */ 3799 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3800 &nvtop) == 0); 3801 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3802 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3803 VDEV_ALLOC_ROOTPOOL); 3804 spa_config_exit(spa, SCL_ALL, FTAG); 3805 if (error) { 3806 mutex_exit(&spa_namespace_lock); 3807 nvlist_free(config); 3808 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3809 pname); 3810 return (error); 3811 } 3812 3813 /* 3814 * Get the boot vdev. 3815 */ 3816 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3817 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3818 (u_longlong_t)guid); 3819 error = SET_ERROR(ENOENT); 3820 goto out; 3821 } 3822 3823 /* 3824 * Determine if there is a better boot device. 3825 */ 3826 avd = bvd; 3827 spa_alt_rootvdev(rvd, &avd, &txg); 3828 if (avd != bvd) { 3829 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3830 "try booting from '%s'", avd->vdev_path); 3831 error = SET_ERROR(EINVAL); 3832 goto out; 3833 } 3834 3835 /* 3836 * If the boot device is part of a spare vdev then ensure that 3837 * we're booting off the active spare. 3838 */ 3839 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3840 !bvd->vdev_isspare) { 3841 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3842 "try booting from '%s'", 3843 bvd->vdev_parent-> 3844 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3845 error = SET_ERROR(EINVAL); 3846 goto out; 3847 } 3848 3849 error = 0; 3850out: 3851 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3852 vdev_free(rvd); 3853 spa_config_exit(spa, SCL_ALL, FTAG); 3854 mutex_exit(&spa_namespace_lock); 3855 3856 nvlist_free(config); 3857 return (error); 3858} 3859 3860#else 3861 3862extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3863 uint64_t *count); 3864 3865static nvlist_t * 3866spa_generate_rootconf(const char *name) 3867{ 3868 nvlist_t **configs, **tops; 3869 nvlist_t *config; 3870 nvlist_t *best_cfg, *nvtop, *nvroot; 3871 uint64_t *holes; 3872 uint64_t best_txg; 3873 uint64_t nchildren; 3874 uint64_t pgid; 3875 uint64_t count; 3876 uint64_t i; 3877 uint_t nholes; 3878 3879 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3880 return (NULL); 3881 3882 ASSERT3U(count, !=, 0); 3883 best_txg = 0; 3884 for (i = 0; i < count; i++) { 3885 uint64_t txg; 3886 3887 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 3888 &txg) == 0); 3889 if (txg > best_txg) { 3890 best_txg = txg; 3891 best_cfg = configs[i]; 3892 } 3893 } 3894 3895 /* 3896 * Multi-vdev root pool configuration discovery is not supported yet. 3897 */ 3898 nchildren = 1; 3899 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 3900 holes = NULL; 3901 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 3902 &holes, &nholes); 3903 3904 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 3905 for (i = 0; i < nchildren; i++) { 3906 if (i >= count) 3907 break; 3908 if (configs[i] == NULL) 3909 continue; 3910 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 3911 &nvtop) == 0); 3912 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 3913 } 3914 for (i = 0; holes != NULL && i < nholes; i++) { 3915 if (i >= nchildren) 3916 continue; 3917 if (tops[holes[i]] != NULL) 3918 continue; 3919 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 3920 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 3921 VDEV_TYPE_HOLE) == 0); 3922 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 3923 holes[i]) == 0); 3924 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 3925 0) == 0); 3926 } 3927 for (i = 0; i < nchildren; i++) { 3928 if (tops[i] != NULL) 3929 continue; 3930 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 3931 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 3932 VDEV_TYPE_MISSING) == 0); 3933 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 3934 i) == 0); 3935 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 3936 0) == 0); 3937 } 3938 3939 /* 3940 * Create pool config based on the best vdev config. 3941 */ 3942 nvlist_dup(best_cfg, &config, KM_SLEEP); 3943 3944 /* 3945 * Put this pool's top-level vdevs into a root vdev. 3946 */ 3947 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3948 &pgid) == 0); 3949 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3950 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3951 VDEV_TYPE_ROOT) == 0); 3952 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3953 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3954 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3955 tops, nchildren) == 0); 3956 3957 /* 3958 * Replace the existing vdev_tree with the new root vdev in 3959 * this pool's configuration (remove the old, add the new). 3960 */ 3961 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3962 3963 /* 3964 * Drop vdev config elements that should not be present at pool level. 3965 */ 3966 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 3967 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 3968 3969 for (i = 0; i < count; i++) 3970 nvlist_free(configs[i]); 3971 kmem_free(configs, count * sizeof(void *)); 3972 for (i = 0; i < nchildren; i++) 3973 nvlist_free(tops[i]); 3974 kmem_free(tops, nchildren * sizeof(void *)); 3975 nvlist_free(nvroot); 3976 return (config); 3977} 3978 3979int 3980spa_import_rootpool(const char *name) 3981{ 3982 spa_t *spa; 3983 vdev_t *rvd, *bvd, *avd = NULL; 3984 nvlist_t *config, *nvtop; 3985 uint64_t txg; 3986 char *pname; 3987 int error; 3988 3989 /* 3990 * Read the label from the boot device and generate a configuration. 3991 */ 3992 config = spa_generate_rootconf(name); 3993 3994 mutex_enter(&spa_namespace_lock); 3995 if (config != NULL) { 3996 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3997 &pname) == 0 && strcmp(name, pname) == 0); 3998 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 3999 == 0); 4000 4001 if ((spa = spa_lookup(pname)) != NULL) { 4002 /* 4003 * Remove the existing root pool from the namespace so 4004 * that we can replace it with the correct config 4005 * we just read in. 4006 */ 4007 spa_remove(spa); 4008 } 4009 spa = spa_add(pname, config, NULL); 4010 4011 /* 4012 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4013 * via spa_version(). 4014 */ 4015 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4016 &spa->spa_ubsync.ub_version) != 0) 4017 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4018 } else if ((spa = spa_lookup(name)) == NULL) { 4019 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4020 name); 4021 return (EIO); 4022 } else { 4023 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4024 } 4025 spa->spa_is_root = B_TRUE; 4026 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4027 4028 /* 4029 * Build up a vdev tree based on the boot device's label config. 4030 */ 4031 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4032 &nvtop) == 0); 4033 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4034 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4035 VDEV_ALLOC_ROOTPOOL); 4036 spa_config_exit(spa, SCL_ALL, FTAG); 4037 if (error) { 4038 mutex_exit(&spa_namespace_lock); 4039 nvlist_free(config); 4040 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4041 pname); 4042 return (error); 4043 } 4044 4045 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4046 vdev_free(rvd); 4047 spa_config_exit(spa, SCL_ALL, FTAG); 4048 mutex_exit(&spa_namespace_lock); 4049 4050 nvlist_free(config); 4051 return (0); 4052} 4053 4054#endif /* sun */ 4055#endif 4056 4057/* 4058 * Import a non-root pool into the system. 4059 */ 4060int 4061spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4062{ 4063 spa_t *spa; 4064 char *altroot = NULL; 4065 spa_load_state_t state = SPA_LOAD_IMPORT; 4066 zpool_rewind_policy_t policy; 4067 uint64_t mode = spa_mode_global; 4068 uint64_t readonly = B_FALSE; 4069 int error; 4070 nvlist_t *nvroot; 4071 nvlist_t **spares, **l2cache; 4072 uint_t nspares, nl2cache; 4073 4074 /* 4075 * If a pool with this name exists, return failure. 4076 */ 4077 mutex_enter(&spa_namespace_lock); 4078 if (spa_lookup(pool) != NULL) { 4079 mutex_exit(&spa_namespace_lock); 4080 return (SET_ERROR(EEXIST)); 4081 } 4082 4083 /* 4084 * Create and initialize the spa structure. 4085 */ 4086 (void) nvlist_lookup_string(props, 4087 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4088 (void) nvlist_lookup_uint64(props, 4089 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4090 if (readonly) 4091 mode = FREAD; 4092 spa = spa_add(pool, config, altroot); 4093 spa->spa_import_flags = flags; 4094 4095 /* 4096 * Verbatim import - Take a pool and insert it into the namespace 4097 * as if it had been loaded at boot. 4098 */ 4099 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4100 if (props != NULL) 4101 spa_configfile_set(spa, props, B_FALSE); 4102 4103 spa_config_sync(spa, B_FALSE, B_TRUE); 4104 4105 mutex_exit(&spa_namespace_lock); 4106 spa_history_log_version(spa, "import"); 4107 4108 return (0); 4109 } 4110 4111 spa_activate(spa, mode); 4112 4113 /* 4114 * Don't start async tasks until we know everything is healthy. 4115 */ 4116 spa_async_suspend(spa); 4117 4118 zpool_get_rewind_policy(config, &policy); 4119 if (policy.zrp_request & ZPOOL_DO_REWIND) 4120 state = SPA_LOAD_RECOVER; 4121 4122 /* 4123 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4124 * because the user-supplied config is actually the one to trust when 4125 * doing an import. 4126 */ 4127 if (state != SPA_LOAD_RECOVER) 4128 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4129 4130 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4131 policy.zrp_request); 4132 4133 /* 4134 * Propagate anything learned while loading the pool and pass it 4135 * back to caller (i.e. rewind info, missing devices, etc). 4136 */ 4137 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4138 spa->spa_load_info) == 0); 4139 4140 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4141 /* 4142 * Toss any existing sparelist, as it doesn't have any validity 4143 * anymore, and conflicts with spa_has_spare(). 4144 */ 4145 if (spa->spa_spares.sav_config) { 4146 nvlist_free(spa->spa_spares.sav_config); 4147 spa->spa_spares.sav_config = NULL; 4148 spa_load_spares(spa); 4149 } 4150 if (spa->spa_l2cache.sav_config) { 4151 nvlist_free(spa->spa_l2cache.sav_config); 4152 spa->spa_l2cache.sav_config = NULL; 4153 spa_load_l2cache(spa); 4154 } 4155 4156 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4157 &nvroot) == 0); 4158 if (error == 0) 4159 error = spa_validate_aux(spa, nvroot, -1ULL, 4160 VDEV_ALLOC_SPARE); 4161 if (error == 0) 4162 error = spa_validate_aux(spa, nvroot, -1ULL, 4163 VDEV_ALLOC_L2CACHE); 4164 spa_config_exit(spa, SCL_ALL, FTAG); 4165 4166 if (props != NULL) 4167 spa_configfile_set(spa, props, B_FALSE); 4168 4169 if (error != 0 || (props && spa_writeable(spa) && 4170 (error = spa_prop_set(spa, props)))) { 4171 spa_unload(spa); 4172 spa_deactivate(spa); 4173 spa_remove(spa); 4174 mutex_exit(&spa_namespace_lock); 4175 return (error); 4176 } 4177 4178 spa_async_resume(spa); 4179 4180 /* 4181 * Override any spares and level 2 cache devices as specified by 4182 * the user, as these may have correct device names/devids, etc. 4183 */ 4184 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4185 &spares, &nspares) == 0) { 4186 if (spa->spa_spares.sav_config) 4187 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4188 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4189 else 4190 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4191 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4192 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4193 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4194 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4195 spa_load_spares(spa); 4196 spa_config_exit(spa, SCL_ALL, FTAG); 4197 spa->spa_spares.sav_sync = B_TRUE; 4198 } 4199 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4200 &l2cache, &nl2cache) == 0) { 4201 if (spa->spa_l2cache.sav_config) 4202 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4203 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4204 else 4205 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4206 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4207 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4208 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4209 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4210 spa_load_l2cache(spa); 4211 spa_config_exit(spa, SCL_ALL, FTAG); 4212 spa->spa_l2cache.sav_sync = B_TRUE; 4213 } 4214 4215 /* 4216 * Check for any removed devices. 4217 */ 4218 if (spa->spa_autoreplace) { 4219 spa_aux_check_removed(&spa->spa_spares); 4220 spa_aux_check_removed(&spa->spa_l2cache); 4221 } 4222 4223 if (spa_writeable(spa)) { 4224 /* 4225 * Update the config cache to include the newly-imported pool. 4226 */ 4227 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4228 } 4229 4230 /* 4231 * It's possible that the pool was expanded while it was exported. 4232 * We kick off an async task to handle this for us. 4233 */ 4234 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4235 4236 mutex_exit(&spa_namespace_lock); 4237 spa_history_log_version(spa, "import"); 4238 4239#ifdef __FreeBSD__ 4240#ifdef _KERNEL 4241 zvol_create_minors(pool); 4242#endif 4243#endif 4244 return (0); 4245} 4246 4247nvlist_t * 4248spa_tryimport(nvlist_t *tryconfig) 4249{ 4250 nvlist_t *config = NULL; 4251 char *poolname; 4252 spa_t *spa; 4253 uint64_t state; 4254 int error; 4255 4256 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4257 return (NULL); 4258 4259 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4260 return (NULL); 4261 4262 /* 4263 * Create and initialize the spa structure. 4264 */ 4265 mutex_enter(&spa_namespace_lock); 4266 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4267 spa_activate(spa, FREAD); 4268 4269 /* 4270 * Pass off the heavy lifting to spa_load(). 4271 * Pass TRUE for mosconfig because the user-supplied config 4272 * is actually the one to trust when doing an import. 4273 */ 4274 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4275 4276 /* 4277 * If 'tryconfig' was at least parsable, return the current config. 4278 */ 4279 if (spa->spa_root_vdev != NULL) { 4280 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4281 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4282 poolname) == 0); 4283 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4284 state) == 0); 4285 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4286 spa->spa_uberblock.ub_timestamp) == 0); 4287 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4288 spa->spa_load_info) == 0); 4289 4290 /* 4291 * If the bootfs property exists on this pool then we 4292 * copy it out so that external consumers can tell which 4293 * pools are bootable. 4294 */ 4295 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4296 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4297 4298 /* 4299 * We have to play games with the name since the 4300 * pool was opened as TRYIMPORT_NAME. 4301 */ 4302 if (dsl_dsobj_to_dsname(spa_name(spa), 4303 spa->spa_bootfs, tmpname) == 0) { 4304 char *cp; 4305 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4306 4307 cp = strchr(tmpname, '/'); 4308 if (cp == NULL) { 4309 (void) strlcpy(dsname, tmpname, 4310 MAXPATHLEN); 4311 } else { 4312 (void) snprintf(dsname, MAXPATHLEN, 4313 "%s/%s", poolname, ++cp); 4314 } 4315 VERIFY(nvlist_add_string(config, 4316 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4317 kmem_free(dsname, MAXPATHLEN); 4318 } 4319 kmem_free(tmpname, MAXPATHLEN); 4320 } 4321 4322 /* 4323 * Add the list of hot spares and level 2 cache devices. 4324 */ 4325 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4326 spa_add_spares(spa, config); 4327 spa_add_l2cache(spa, config); 4328 spa_config_exit(spa, SCL_CONFIG, FTAG); 4329 } 4330 4331 spa_unload(spa); 4332 spa_deactivate(spa); 4333 spa_remove(spa); 4334 mutex_exit(&spa_namespace_lock); 4335 4336 return (config); 4337} 4338 4339/* 4340 * Pool export/destroy 4341 * 4342 * The act of destroying or exporting a pool is very simple. We make sure there 4343 * is no more pending I/O and any references to the pool are gone. Then, we 4344 * update the pool state and sync all the labels to disk, removing the 4345 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4346 * we don't sync the labels or remove the configuration cache. 4347 */ 4348static int 4349spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4350 boolean_t force, boolean_t hardforce) 4351{ 4352 spa_t *spa; 4353 4354 if (oldconfig) 4355 *oldconfig = NULL; 4356 4357 if (!(spa_mode_global & FWRITE)) 4358 return (SET_ERROR(EROFS)); 4359 4360 mutex_enter(&spa_namespace_lock); 4361 if ((spa = spa_lookup(pool)) == NULL) { 4362 mutex_exit(&spa_namespace_lock); 4363 return (SET_ERROR(ENOENT)); 4364 } 4365 4366 /* 4367 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4368 * reacquire the namespace lock, and see if we can export. 4369 */ 4370 spa_open_ref(spa, FTAG); 4371 mutex_exit(&spa_namespace_lock); 4372 spa_async_suspend(spa); 4373 mutex_enter(&spa_namespace_lock); 4374 spa_close(spa, FTAG); 4375 4376 /* 4377 * The pool will be in core if it's openable, 4378 * in which case we can modify its state. 4379 */ 4380 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4381 /* 4382 * Objsets may be open only because they're dirty, so we 4383 * have to force it to sync before checking spa_refcnt. 4384 */ 4385 txg_wait_synced(spa->spa_dsl_pool, 0); 4386 4387 /* 4388 * A pool cannot be exported or destroyed if there are active 4389 * references. If we are resetting a pool, allow references by 4390 * fault injection handlers. 4391 */ 4392 if (!spa_refcount_zero(spa) || 4393 (spa->spa_inject_ref != 0 && 4394 new_state != POOL_STATE_UNINITIALIZED)) { 4395 spa_async_resume(spa); 4396 mutex_exit(&spa_namespace_lock); 4397 return (SET_ERROR(EBUSY)); 4398 } 4399 4400 /* 4401 * A pool cannot be exported if it has an active shared spare. 4402 * This is to prevent other pools stealing the active spare 4403 * from an exported pool. At user's own will, such pool can 4404 * be forcedly exported. 4405 */ 4406 if (!force && new_state == POOL_STATE_EXPORTED && 4407 spa_has_active_shared_spare(spa)) { 4408 spa_async_resume(spa); 4409 mutex_exit(&spa_namespace_lock); 4410 return (SET_ERROR(EXDEV)); 4411 } 4412 4413 /* 4414 * We want this to be reflected on every label, 4415 * so mark them all dirty. spa_unload() will do the 4416 * final sync that pushes these changes out. 4417 */ 4418 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4419 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4420 spa->spa_state = new_state; 4421 spa->spa_final_txg = spa_last_synced_txg(spa) + 4422 TXG_DEFER_SIZE + 1; 4423 vdev_config_dirty(spa->spa_root_vdev); 4424 spa_config_exit(spa, SCL_ALL, FTAG); 4425 } 4426 } 4427 4428 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4429 4430 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4431 spa_unload(spa); 4432 spa_deactivate(spa); 4433 } 4434 4435 if (oldconfig && spa->spa_config) 4436 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4437 4438 if (new_state != POOL_STATE_UNINITIALIZED) { 4439 if (!hardforce) 4440 spa_config_sync(spa, B_TRUE, B_TRUE); 4441 spa_remove(spa); 4442 } 4443 mutex_exit(&spa_namespace_lock); 4444 4445 return (0); 4446} 4447 4448/* 4449 * Destroy a storage pool. 4450 */ 4451int 4452spa_destroy(char *pool) 4453{ 4454 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4455 B_FALSE, B_FALSE)); 4456} 4457 4458/* 4459 * Export a storage pool. 4460 */ 4461int 4462spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4463 boolean_t hardforce) 4464{ 4465 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4466 force, hardforce)); 4467} 4468 4469/* 4470 * Similar to spa_export(), this unloads the spa_t without actually removing it 4471 * from the namespace in any way. 4472 */ 4473int 4474spa_reset(char *pool) 4475{ 4476 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4477 B_FALSE, B_FALSE)); 4478} 4479 4480/* 4481 * ========================================================================== 4482 * Device manipulation 4483 * ========================================================================== 4484 */ 4485 4486/* 4487 * Add a device to a storage pool. 4488 */ 4489int 4490spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4491{ 4492 uint64_t txg, id; 4493 int error; 4494 vdev_t *rvd = spa->spa_root_vdev; 4495 vdev_t *vd, *tvd; 4496 nvlist_t **spares, **l2cache; 4497 uint_t nspares, nl2cache; 4498 4499 ASSERT(spa_writeable(spa)); 4500 4501 txg = spa_vdev_enter(spa); 4502 4503 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4504 VDEV_ALLOC_ADD)) != 0) 4505 return (spa_vdev_exit(spa, NULL, txg, error)); 4506 4507 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4508 4509 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4510 &nspares) != 0) 4511 nspares = 0; 4512 4513 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4514 &nl2cache) != 0) 4515 nl2cache = 0; 4516 4517 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4518 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4519 4520 if (vd->vdev_children != 0 && 4521 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4522 return (spa_vdev_exit(spa, vd, txg, error)); 4523 4524 /* 4525 * We must validate the spares and l2cache devices after checking the 4526 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4527 */ 4528 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4529 return (spa_vdev_exit(spa, vd, txg, error)); 4530 4531 /* 4532 * Transfer each new top-level vdev from vd to rvd. 4533 */ 4534 for (int c = 0; c < vd->vdev_children; c++) { 4535 4536 /* 4537 * Set the vdev id to the first hole, if one exists. 4538 */ 4539 for (id = 0; id < rvd->vdev_children; id++) { 4540 if (rvd->vdev_child[id]->vdev_ishole) { 4541 vdev_free(rvd->vdev_child[id]); 4542 break; 4543 } 4544 } 4545 tvd = vd->vdev_child[c]; 4546 vdev_remove_child(vd, tvd); 4547 tvd->vdev_id = id; 4548 vdev_add_child(rvd, tvd); 4549 vdev_config_dirty(tvd); 4550 } 4551 4552 if (nspares != 0) { 4553 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4554 ZPOOL_CONFIG_SPARES); 4555 spa_load_spares(spa); 4556 spa->spa_spares.sav_sync = B_TRUE; 4557 } 4558 4559 if (nl2cache != 0) { 4560 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4561 ZPOOL_CONFIG_L2CACHE); 4562 spa_load_l2cache(spa); 4563 spa->spa_l2cache.sav_sync = B_TRUE; 4564 } 4565 4566 /* 4567 * We have to be careful when adding new vdevs to an existing pool. 4568 * If other threads start allocating from these vdevs before we 4569 * sync the config cache, and we lose power, then upon reboot we may 4570 * fail to open the pool because there are DVAs that the config cache 4571 * can't translate. Therefore, we first add the vdevs without 4572 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4573 * and then let spa_config_update() initialize the new metaslabs. 4574 * 4575 * spa_load() checks for added-but-not-initialized vdevs, so that 4576 * if we lose power at any point in this sequence, the remaining 4577 * steps will be completed the next time we load the pool. 4578 */ 4579 (void) spa_vdev_exit(spa, vd, txg, 0); 4580 4581 mutex_enter(&spa_namespace_lock); 4582 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4583 mutex_exit(&spa_namespace_lock); 4584 4585 return (0); 4586} 4587 4588/* 4589 * Attach a device to a mirror. The arguments are the path to any device 4590 * in the mirror, and the nvroot for the new device. If the path specifies 4591 * a device that is not mirrored, we automatically insert the mirror vdev. 4592 * 4593 * If 'replacing' is specified, the new device is intended to replace the 4594 * existing device; in this case the two devices are made into their own 4595 * mirror using the 'replacing' vdev, which is functionally identical to 4596 * the mirror vdev (it actually reuses all the same ops) but has a few 4597 * extra rules: you can't attach to it after it's been created, and upon 4598 * completion of resilvering, the first disk (the one being replaced) 4599 * is automatically detached. 4600 */ 4601int 4602spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4603{ 4604 uint64_t txg, dtl_max_txg; 4605 vdev_t *rvd = spa->spa_root_vdev; 4606 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4607 vdev_ops_t *pvops; 4608 char *oldvdpath, *newvdpath; 4609 int newvd_isspare; 4610 int error; 4611 4612 ASSERT(spa_writeable(spa)); 4613 4614 txg = spa_vdev_enter(spa); 4615 4616 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4617 4618 if (oldvd == NULL) 4619 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4620 4621 if (!oldvd->vdev_ops->vdev_op_leaf) 4622 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4623 4624 pvd = oldvd->vdev_parent; 4625 4626 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4627 VDEV_ALLOC_ATTACH)) != 0) 4628 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4629 4630 if (newrootvd->vdev_children != 1) 4631 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4632 4633 newvd = newrootvd->vdev_child[0]; 4634 4635 if (!newvd->vdev_ops->vdev_op_leaf) 4636 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4637 4638 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4639 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4640 4641 /* 4642 * Spares can't replace logs 4643 */ 4644 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4645 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4646 4647 if (!replacing) { 4648 /* 4649 * For attach, the only allowable parent is a mirror or the root 4650 * vdev. 4651 */ 4652 if (pvd->vdev_ops != &vdev_mirror_ops && 4653 pvd->vdev_ops != &vdev_root_ops) 4654 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4655 4656 pvops = &vdev_mirror_ops; 4657 } else { 4658 /* 4659 * Active hot spares can only be replaced by inactive hot 4660 * spares. 4661 */ 4662 if (pvd->vdev_ops == &vdev_spare_ops && 4663 oldvd->vdev_isspare && 4664 !spa_has_spare(spa, newvd->vdev_guid)) 4665 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4666 4667 /* 4668 * If the source is a hot spare, and the parent isn't already a 4669 * spare, then we want to create a new hot spare. Otherwise, we 4670 * want to create a replacing vdev. The user is not allowed to 4671 * attach to a spared vdev child unless the 'isspare' state is 4672 * the same (spare replaces spare, non-spare replaces 4673 * non-spare). 4674 */ 4675 if (pvd->vdev_ops == &vdev_replacing_ops && 4676 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4677 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4678 } else if (pvd->vdev_ops == &vdev_spare_ops && 4679 newvd->vdev_isspare != oldvd->vdev_isspare) { 4680 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4681 } 4682 4683 if (newvd->vdev_isspare) 4684 pvops = &vdev_spare_ops; 4685 else 4686 pvops = &vdev_replacing_ops; 4687 } 4688 4689 /* 4690 * Make sure the new device is big enough. 4691 */ 4692 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4693 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4694 4695 /* 4696 * The new device cannot have a higher alignment requirement 4697 * than the top-level vdev. 4698 */ 4699 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4700 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4701 4702 /* 4703 * If this is an in-place replacement, update oldvd's path and devid 4704 * to make it distinguishable from newvd, and unopenable from now on. 4705 */ 4706 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4707 spa_strfree(oldvd->vdev_path); 4708 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4709 KM_SLEEP); 4710 (void) sprintf(oldvd->vdev_path, "%s/%s", 4711 newvd->vdev_path, "old"); 4712 if (oldvd->vdev_devid != NULL) { 4713 spa_strfree(oldvd->vdev_devid); 4714 oldvd->vdev_devid = NULL; 4715 } 4716 } 4717 4718 /* mark the device being resilvered */ 4719 newvd->vdev_resilver_txg = txg; 4720 4721 /* 4722 * If the parent is not a mirror, or if we're replacing, insert the new 4723 * mirror/replacing/spare vdev above oldvd. 4724 */ 4725 if (pvd->vdev_ops != pvops) 4726 pvd = vdev_add_parent(oldvd, pvops); 4727 4728 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4729 ASSERT(pvd->vdev_ops == pvops); 4730 ASSERT(oldvd->vdev_parent == pvd); 4731 4732 /* 4733 * Extract the new device from its root and add it to pvd. 4734 */ 4735 vdev_remove_child(newrootvd, newvd); 4736 newvd->vdev_id = pvd->vdev_children; 4737 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4738 vdev_add_child(pvd, newvd); 4739 4740 tvd = newvd->vdev_top; 4741 ASSERT(pvd->vdev_top == tvd); 4742 ASSERT(tvd->vdev_parent == rvd); 4743 4744 vdev_config_dirty(tvd); 4745 4746 /* 4747 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4748 * for any dmu_sync-ed blocks. It will propagate upward when 4749 * spa_vdev_exit() calls vdev_dtl_reassess(). 4750 */ 4751 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4752 4753 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4754 dtl_max_txg - TXG_INITIAL); 4755 4756 if (newvd->vdev_isspare) { 4757 spa_spare_activate(newvd); 4758 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4759 } 4760 4761 oldvdpath = spa_strdup(oldvd->vdev_path); 4762 newvdpath = spa_strdup(newvd->vdev_path); 4763 newvd_isspare = newvd->vdev_isspare; 4764 4765 /* 4766 * Mark newvd's DTL dirty in this txg. 4767 */ 4768 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4769 4770 /* 4771 * Schedule the resilver to restart in the future. We do this to 4772 * ensure that dmu_sync-ed blocks have been stitched into the 4773 * respective datasets. 4774 */ 4775 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4776 4777 /* 4778 * Commit the config 4779 */ 4780 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4781 4782 spa_history_log_internal(spa, "vdev attach", NULL, 4783 "%s vdev=%s %s vdev=%s", 4784 replacing && newvd_isspare ? "spare in" : 4785 replacing ? "replace" : "attach", newvdpath, 4786 replacing ? "for" : "to", oldvdpath); 4787 4788 spa_strfree(oldvdpath); 4789 spa_strfree(newvdpath); 4790 4791 if (spa->spa_bootfs) 4792 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4793 4794 return (0); 4795} 4796 4797/* 4798 * Detach a device from a mirror or replacing vdev. 4799 * 4800 * If 'replace_done' is specified, only detach if the parent 4801 * is a replacing vdev. 4802 */ 4803int 4804spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4805{ 4806 uint64_t txg; 4807 int error; 4808 vdev_t *rvd = spa->spa_root_vdev; 4809 vdev_t *vd, *pvd, *cvd, *tvd; 4810 boolean_t unspare = B_FALSE; 4811 uint64_t unspare_guid = 0; 4812 char *vdpath; 4813 4814 ASSERT(spa_writeable(spa)); 4815 4816 txg = spa_vdev_enter(spa); 4817 4818 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4819 4820 if (vd == NULL) 4821 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4822 4823 if (!vd->vdev_ops->vdev_op_leaf) 4824 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4825 4826 pvd = vd->vdev_parent; 4827 4828 /* 4829 * If the parent/child relationship is not as expected, don't do it. 4830 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4831 * vdev that's replacing B with C. The user's intent in replacing 4832 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4833 * the replace by detaching C, the expected behavior is to end up 4834 * M(A,B). But suppose that right after deciding to detach C, 4835 * the replacement of B completes. We would have M(A,C), and then 4836 * ask to detach C, which would leave us with just A -- not what 4837 * the user wanted. To prevent this, we make sure that the 4838 * parent/child relationship hasn't changed -- in this example, 4839 * that C's parent is still the replacing vdev R. 4840 */ 4841 if (pvd->vdev_guid != pguid && pguid != 0) 4842 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4843 4844 /* 4845 * Only 'replacing' or 'spare' vdevs can be replaced. 4846 */ 4847 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4848 pvd->vdev_ops != &vdev_spare_ops) 4849 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4850 4851 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4852 spa_version(spa) >= SPA_VERSION_SPARES); 4853 4854 /* 4855 * Only mirror, replacing, and spare vdevs support detach. 4856 */ 4857 if (pvd->vdev_ops != &vdev_replacing_ops && 4858 pvd->vdev_ops != &vdev_mirror_ops && 4859 pvd->vdev_ops != &vdev_spare_ops) 4860 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4861 4862 /* 4863 * If this device has the only valid copy of some data, 4864 * we cannot safely detach it. 4865 */ 4866 if (vdev_dtl_required(vd)) 4867 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4868 4869 ASSERT(pvd->vdev_children >= 2); 4870 4871 /* 4872 * If we are detaching the second disk from a replacing vdev, then 4873 * check to see if we changed the original vdev's path to have "/old" 4874 * at the end in spa_vdev_attach(). If so, undo that change now. 4875 */ 4876 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4877 vd->vdev_path != NULL) { 4878 size_t len = strlen(vd->vdev_path); 4879 4880 for (int c = 0; c < pvd->vdev_children; c++) { 4881 cvd = pvd->vdev_child[c]; 4882 4883 if (cvd == vd || cvd->vdev_path == NULL) 4884 continue; 4885 4886 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4887 strcmp(cvd->vdev_path + len, "/old") == 0) { 4888 spa_strfree(cvd->vdev_path); 4889 cvd->vdev_path = spa_strdup(vd->vdev_path); 4890 break; 4891 } 4892 } 4893 } 4894 4895 /* 4896 * If we are detaching the original disk from a spare, then it implies 4897 * that the spare should become a real disk, and be removed from the 4898 * active spare list for the pool. 4899 */ 4900 if (pvd->vdev_ops == &vdev_spare_ops && 4901 vd->vdev_id == 0 && 4902 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4903 unspare = B_TRUE; 4904 4905 /* 4906 * Erase the disk labels so the disk can be used for other things. 4907 * This must be done after all other error cases are handled, 4908 * but before we disembowel vd (so we can still do I/O to it). 4909 * But if we can't do it, don't treat the error as fatal -- 4910 * it may be that the unwritability of the disk is the reason 4911 * it's being detached! 4912 */ 4913 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4914 4915 /* 4916 * Remove vd from its parent and compact the parent's children. 4917 */ 4918 vdev_remove_child(pvd, vd); 4919 vdev_compact_children(pvd); 4920 4921 /* 4922 * Remember one of the remaining children so we can get tvd below. 4923 */ 4924 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4925 4926 /* 4927 * If we need to remove the remaining child from the list of hot spares, 4928 * do it now, marking the vdev as no longer a spare in the process. 4929 * We must do this before vdev_remove_parent(), because that can 4930 * change the GUID if it creates a new toplevel GUID. For a similar 4931 * reason, we must remove the spare now, in the same txg as the detach; 4932 * otherwise someone could attach a new sibling, change the GUID, and 4933 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4934 */ 4935 if (unspare) { 4936 ASSERT(cvd->vdev_isspare); 4937 spa_spare_remove(cvd); 4938 unspare_guid = cvd->vdev_guid; 4939 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4940 cvd->vdev_unspare = B_TRUE; 4941 } 4942 4943 /* 4944 * If the parent mirror/replacing vdev only has one child, 4945 * the parent is no longer needed. Remove it from the tree. 4946 */ 4947 if (pvd->vdev_children == 1) { 4948 if (pvd->vdev_ops == &vdev_spare_ops) 4949 cvd->vdev_unspare = B_FALSE; 4950 vdev_remove_parent(cvd); 4951 } 4952 4953 4954 /* 4955 * We don't set tvd until now because the parent we just removed 4956 * may have been the previous top-level vdev. 4957 */ 4958 tvd = cvd->vdev_top; 4959 ASSERT(tvd->vdev_parent == rvd); 4960 4961 /* 4962 * Reevaluate the parent vdev state. 4963 */ 4964 vdev_propagate_state(cvd); 4965 4966 /* 4967 * If the 'autoexpand' property is set on the pool then automatically 4968 * try to expand the size of the pool. For example if the device we 4969 * just detached was smaller than the others, it may be possible to 4970 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4971 * first so that we can obtain the updated sizes of the leaf vdevs. 4972 */ 4973 if (spa->spa_autoexpand) { 4974 vdev_reopen(tvd); 4975 vdev_expand(tvd, txg); 4976 } 4977 4978 vdev_config_dirty(tvd); 4979 4980 /* 4981 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4982 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4983 * But first make sure we're not on any *other* txg's DTL list, to 4984 * prevent vd from being accessed after it's freed. 4985 */ 4986 vdpath = spa_strdup(vd->vdev_path); 4987 for (int t = 0; t < TXG_SIZE; t++) 4988 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4989 vd->vdev_detached = B_TRUE; 4990 vdev_dirty(tvd, VDD_DTL, vd, txg); 4991 4992 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4993 4994 /* hang on to the spa before we release the lock */ 4995 spa_open_ref(spa, FTAG); 4996 4997 error = spa_vdev_exit(spa, vd, txg, 0); 4998 4999 spa_history_log_internal(spa, "detach", NULL, 5000 "vdev=%s", vdpath); 5001 spa_strfree(vdpath); 5002 5003 /* 5004 * If this was the removal of the original device in a hot spare vdev, 5005 * then we want to go through and remove the device from the hot spare 5006 * list of every other pool. 5007 */ 5008 if (unspare) { 5009 spa_t *altspa = NULL; 5010 5011 mutex_enter(&spa_namespace_lock); 5012 while ((altspa = spa_next(altspa)) != NULL) { 5013 if (altspa->spa_state != POOL_STATE_ACTIVE || 5014 altspa == spa) 5015 continue; 5016 5017 spa_open_ref(altspa, FTAG); 5018 mutex_exit(&spa_namespace_lock); 5019 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5020 mutex_enter(&spa_namespace_lock); 5021 spa_close(altspa, FTAG); 5022 } 5023 mutex_exit(&spa_namespace_lock); 5024 5025 /* search the rest of the vdevs for spares to remove */ 5026 spa_vdev_resilver_done(spa); 5027 } 5028 5029 /* all done with the spa; OK to release */ 5030 mutex_enter(&spa_namespace_lock); 5031 spa_close(spa, FTAG); 5032 mutex_exit(&spa_namespace_lock); 5033 5034 return (error); 5035} 5036 5037/* 5038 * Split a set of devices from their mirrors, and create a new pool from them. 5039 */ 5040int 5041spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5042 nvlist_t *props, boolean_t exp) 5043{ 5044 int error = 0; 5045 uint64_t txg, *glist; 5046 spa_t *newspa; 5047 uint_t c, children, lastlog; 5048 nvlist_t **child, *nvl, *tmp; 5049 dmu_tx_t *tx; 5050 char *altroot = NULL; 5051 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5052 boolean_t activate_slog; 5053 5054 ASSERT(spa_writeable(spa)); 5055 5056 txg = spa_vdev_enter(spa); 5057 5058 /* clear the log and flush everything up to now */ 5059 activate_slog = spa_passivate_log(spa); 5060 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5061 error = spa_offline_log(spa); 5062 txg = spa_vdev_config_enter(spa); 5063 5064 if (activate_slog) 5065 spa_activate_log(spa); 5066 5067 if (error != 0) 5068 return (spa_vdev_exit(spa, NULL, txg, error)); 5069 5070 /* check new spa name before going any further */ 5071 if (spa_lookup(newname) != NULL) 5072 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5073 5074 /* 5075 * scan through all the children to ensure they're all mirrors 5076 */ 5077 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5078 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5079 &children) != 0) 5080 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5081 5082 /* first, check to ensure we've got the right child count */ 5083 rvd = spa->spa_root_vdev; 5084 lastlog = 0; 5085 for (c = 0; c < rvd->vdev_children; c++) { 5086 vdev_t *vd = rvd->vdev_child[c]; 5087 5088 /* don't count the holes & logs as children */ 5089 if (vd->vdev_islog || vd->vdev_ishole) { 5090 if (lastlog == 0) 5091 lastlog = c; 5092 continue; 5093 } 5094 5095 lastlog = 0; 5096 } 5097 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5098 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5099 5100 /* next, ensure no spare or cache devices are part of the split */ 5101 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5102 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5103 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5104 5105 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5106 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5107 5108 /* then, loop over each vdev and validate it */ 5109 for (c = 0; c < children; c++) { 5110 uint64_t is_hole = 0; 5111 5112 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5113 &is_hole); 5114 5115 if (is_hole != 0) { 5116 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5117 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5118 continue; 5119 } else { 5120 error = SET_ERROR(EINVAL); 5121 break; 5122 } 5123 } 5124 5125 /* which disk is going to be split? */ 5126 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5127 &glist[c]) != 0) { 5128 error = SET_ERROR(EINVAL); 5129 break; 5130 } 5131 5132 /* look it up in the spa */ 5133 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5134 if (vml[c] == NULL) { 5135 error = SET_ERROR(ENODEV); 5136 break; 5137 } 5138 5139 /* make sure there's nothing stopping the split */ 5140 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5141 vml[c]->vdev_islog || 5142 vml[c]->vdev_ishole || 5143 vml[c]->vdev_isspare || 5144 vml[c]->vdev_isl2cache || 5145 !vdev_writeable(vml[c]) || 5146 vml[c]->vdev_children != 0 || 5147 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5148 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5149 error = SET_ERROR(EINVAL); 5150 break; 5151 } 5152 5153 if (vdev_dtl_required(vml[c])) { 5154 error = SET_ERROR(EBUSY); 5155 break; 5156 } 5157 5158 /* we need certain info from the top level */ 5159 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5160 vml[c]->vdev_top->vdev_ms_array) == 0); 5161 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5162 vml[c]->vdev_top->vdev_ms_shift) == 0); 5163 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5164 vml[c]->vdev_top->vdev_asize) == 0); 5165 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5166 vml[c]->vdev_top->vdev_ashift) == 0); 5167 } 5168 5169 if (error != 0) { 5170 kmem_free(vml, children * sizeof (vdev_t *)); 5171 kmem_free(glist, children * sizeof (uint64_t)); 5172 return (spa_vdev_exit(spa, NULL, txg, error)); 5173 } 5174 5175 /* stop writers from using the disks */ 5176 for (c = 0; c < children; c++) { 5177 if (vml[c] != NULL) 5178 vml[c]->vdev_offline = B_TRUE; 5179 } 5180 vdev_reopen(spa->spa_root_vdev); 5181 5182 /* 5183 * Temporarily record the splitting vdevs in the spa config. This 5184 * will disappear once the config is regenerated. 5185 */ 5186 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5187 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5188 glist, children) == 0); 5189 kmem_free(glist, children * sizeof (uint64_t)); 5190 5191 mutex_enter(&spa->spa_props_lock); 5192 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5193 nvl) == 0); 5194 mutex_exit(&spa->spa_props_lock); 5195 spa->spa_config_splitting = nvl; 5196 vdev_config_dirty(spa->spa_root_vdev); 5197 5198 /* configure and create the new pool */ 5199 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5200 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5201 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5202 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5203 spa_version(spa)) == 0); 5204 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5205 spa->spa_config_txg) == 0); 5206 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5207 spa_generate_guid(NULL)) == 0); 5208 (void) nvlist_lookup_string(props, 5209 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5210 5211 /* add the new pool to the namespace */ 5212 newspa = spa_add(newname, config, altroot); 5213 newspa->spa_config_txg = spa->spa_config_txg; 5214 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5215 5216 /* release the spa config lock, retaining the namespace lock */ 5217 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5218 5219 if (zio_injection_enabled) 5220 zio_handle_panic_injection(spa, FTAG, 1); 5221 5222 spa_activate(newspa, spa_mode_global); 5223 spa_async_suspend(newspa); 5224 5225#ifndef sun 5226 /* mark that we are creating new spa by splitting */ 5227 newspa->spa_splitting_newspa = B_TRUE; 5228#endif 5229 /* create the new pool from the disks of the original pool */ 5230 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5231#ifndef sun 5232 newspa->spa_splitting_newspa = B_FALSE; 5233#endif 5234 if (error) 5235 goto out; 5236 5237 /* if that worked, generate a real config for the new pool */ 5238 if (newspa->spa_root_vdev != NULL) { 5239 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5240 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5241 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5242 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5243 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5244 B_TRUE)); 5245 } 5246 5247 /* set the props */ 5248 if (props != NULL) { 5249 spa_configfile_set(newspa, props, B_FALSE); 5250 error = spa_prop_set(newspa, props); 5251 if (error) 5252 goto out; 5253 } 5254 5255 /* flush everything */ 5256 txg = spa_vdev_config_enter(newspa); 5257 vdev_config_dirty(newspa->spa_root_vdev); 5258 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5259 5260 if (zio_injection_enabled) 5261 zio_handle_panic_injection(spa, FTAG, 2); 5262 5263 spa_async_resume(newspa); 5264 5265 /* finally, update the original pool's config */ 5266 txg = spa_vdev_config_enter(spa); 5267 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5268 error = dmu_tx_assign(tx, TXG_WAIT); 5269 if (error != 0) 5270 dmu_tx_abort(tx); 5271 for (c = 0; c < children; c++) { 5272 if (vml[c] != NULL) { 5273 vdev_split(vml[c]); 5274 if (error == 0) 5275 spa_history_log_internal(spa, "detach", tx, 5276 "vdev=%s", vml[c]->vdev_path); 5277 vdev_free(vml[c]); 5278 } 5279 } 5280 vdev_config_dirty(spa->spa_root_vdev); 5281 spa->spa_config_splitting = NULL; 5282 nvlist_free(nvl); 5283 if (error == 0) 5284 dmu_tx_commit(tx); 5285 (void) spa_vdev_exit(spa, NULL, txg, 0); 5286 5287 if (zio_injection_enabled) 5288 zio_handle_panic_injection(spa, FTAG, 3); 5289 5290 /* split is complete; log a history record */ 5291 spa_history_log_internal(newspa, "split", NULL, 5292 "from pool %s", spa_name(spa)); 5293 5294 kmem_free(vml, children * sizeof (vdev_t *)); 5295 5296 /* if we're not going to mount the filesystems in userland, export */ 5297 if (exp) 5298 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5299 B_FALSE, B_FALSE); 5300 5301 return (error); 5302 5303out: 5304 spa_unload(newspa); 5305 spa_deactivate(newspa); 5306 spa_remove(newspa); 5307 5308 txg = spa_vdev_config_enter(spa); 5309 5310 /* re-online all offlined disks */ 5311 for (c = 0; c < children; c++) { 5312 if (vml[c] != NULL) 5313 vml[c]->vdev_offline = B_FALSE; 5314 } 5315 vdev_reopen(spa->spa_root_vdev); 5316 5317 nvlist_free(spa->spa_config_splitting); 5318 spa->spa_config_splitting = NULL; 5319 (void) spa_vdev_exit(spa, NULL, txg, error); 5320 5321 kmem_free(vml, children * sizeof (vdev_t *)); 5322 return (error); 5323} 5324 5325static nvlist_t * 5326spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5327{ 5328 for (int i = 0; i < count; i++) { 5329 uint64_t guid; 5330 5331 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5332 &guid) == 0); 5333 5334 if (guid == target_guid) 5335 return (nvpp[i]); 5336 } 5337 5338 return (NULL); 5339} 5340 5341static void 5342spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5343 nvlist_t *dev_to_remove) 5344{ 5345 nvlist_t **newdev = NULL; 5346 5347 if (count > 1) 5348 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5349 5350 for (int i = 0, j = 0; i < count; i++) { 5351 if (dev[i] == dev_to_remove) 5352 continue; 5353 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5354 } 5355 5356 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5357 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5358 5359 for (int i = 0; i < count - 1; i++) 5360 nvlist_free(newdev[i]); 5361 5362 if (count > 1) 5363 kmem_free(newdev, (count - 1) * sizeof (void *)); 5364} 5365 5366/* 5367 * Evacuate the device. 5368 */ 5369static int 5370spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5371{ 5372 uint64_t txg; 5373 int error = 0; 5374 5375 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5376 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5377 ASSERT(vd == vd->vdev_top); 5378 5379 /* 5380 * Evacuate the device. We don't hold the config lock as writer 5381 * since we need to do I/O but we do keep the 5382 * spa_namespace_lock held. Once this completes the device 5383 * should no longer have any blocks allocated on it. 5384 */ 5385 if (vd->vdev_islog) { 5386 if (vd->vdev_stat.vs_alloc != 0) 5387 error = spa_offline_log(spa); 5388 } else { 5389 error = SET_ERROR(ENOTSUP); 5390 } 5391 5392 if (error) 5393 return (error); 5394 5395 /* 5396 * The evacuation succeeded. Remove any remaining MOS metadata 5397 * associated with this vdev, and wait for these changes to sync. 5398 */ 5399 ASSERT0(vd->vdev_stat.vs_alloc); 5400 txg = spa_vdev_config_enter(spa); 5401 vd->vdev_removing = B_TRUE; 5402 vdev_dirty_leaves(vd, VDD_DTL, txg); 5403 vdev_config_dirty(vd); 5404 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5405 5406 return (0); 5407} 5408 5409/* 5410 * Complete the removal by cleaning up the namespace. 5411 */ 5412static void 5413spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5414{ 5415 vdev_t *rvd = spa->spa_root_vdev; 5416 uint64_t id = vd->vdev_id; 5417 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5418 5419 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5420 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5421 ASSERT(vd == vd->vdev_top); 5422 5423 /* 5424 * Only remove any devices which are empty. 5425 */ 5426 if (vd->vdev_stat.vs_alloc != 0) 5427 return; 5428 5429 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5430 5431 if (list_link_active(&vd->vdev_state_dirty_node)) 5432 vdev_state_clean(vd); 5433 if (list_link_active(&vd->vdev_config_dirty_node)) 5434 vdev_config_clean(vd); 5435 5436 vdev_free(vd); 5437 5438 if (last_vdev) { 5439 vdev_compact_children(rvd); 5440 } else { 5441 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5442 vdev_add_child(rvd, vd); 5443 } 5444 vdev_config_dirty(rvd); 5445 5446 /* 5447 * Reassess the health of our root vdev. 5448 */ 5449 vdev_reopen(rvd); 5450} 5451 5452/* 5453 * Remove a device from the pool - 5454 * 5455 * Removing a device from the vdev namespace requires several steps 5456 * and can take a significant amount of time. As a result we use 5457 * the spa_vdev_config_[enter/exit] functions which allow us to 5458 * grab and release the spa_config_lock while still holding the namespace 5459 * lock. During each step the configuration is synced out. 5460 * 5461 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5462 * devices. 5463 */ 5464int 5465spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5466{ 5467 vdev_t *vd; 5468 metaslab_group_t *mg; 5469 nvlist_t **spares, **l2cache, *nv; 5470 uint64_t txg = 0; 5471 uint_t nspares, nl2cache; 5472 int error = 0; 5473 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5474 5475 ASSERT(spa_writeable(spa)); 5476 5477 if (!locked) 5478 txg = spa_vdev_enter(spa); 5479 5480 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5481 5482 if (spa->spa_spares.sav_vdevs != NULL && 5483 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5484 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5485 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5486 /* 5487 * Only remove the hot spare if it's not currently in use 5488 * in this pool. 5489 */ 5490 if (vd == NULL || unspare) { 5491 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5492 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5493 spa_load_spares(spa); 5494 spa->spa_spares.sav_sync = B_TRUE; 5495 } else { 5496 error = SET_ERROR(EBUSY); 5497 } 5498 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5499 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5500 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5501 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5502 /* 5503 * Cache devices can always be removed. 5504 */ 5505 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5506 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5507 spa_load_l2cache(spa); 5508 spa->spa_l2cache.sav_sync = B_TRUE; 5509 } else if (vd != NULL && vd->vdev_islog) { 5510 ASSERT(!locked); 5511 ASSERT(vd == vd->vdev_top); 5512 5513 /* 5514 * XXX - Once we have bp-rewrite this should 5515 * become the common case. 5516 */ 5517 5518 mg = vd->vdev_mg; 5519 5520 /* 5521 * Stop allocating from this vdev. 5522 */ 5523 metaslab_group_passivate(mg); 5524 5525 /* 5526 * Wait for the youngest allocations and frees to sync, 5527 * and then wait for the deferral of those frees to finish. 5528 */ 5529 spa_vdev_config_exit(spa, NULL, 5530 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5531 5532 /* 5533 * Attempt to evacuate the vdev. 5534 */ 5535 error = spa_vdev_remove_evacuate(spa, vd); 5536 5537 txg = spa_vdev_config_enter(spa); 5538 5539 /* 5540 * If we couldn't evacuate the vdev, unwind. 5541 */ 5542 if (error) { 5543 metaslab_group_activate(mg); 5544 return (spa_vdev_exit(spa, NULL, txg, error)); 5545 } 5546 5547 /* 5548 * Clean up the vdev namespace. 5549 */ 5550 spa_vdev_remove_from_namespace(spa, vd); 5551 5552 } else if (vd != NULL) { 5553 /* 5554 * Normal vdevs cannot be removed (yet). 5555 */ 5556 error = SET_ERROR(ENOTSUP); 5557 } else { 5558 /* 5559 * There is no vdev of any kind with the specified guid. 5560 */ 5561 error = SET_ERROR(ENOENT); 5562 } 5563 5564 if (!locked) 5565 return (spa_vdev_exit(spa, NULL, txg, error)); 5566 5567 return (error); 5568} 5569 5570/* 5571 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5572 * currently spared, so we can detach it. 5573 */ 5574static vdev_t * 5575spa_vdev_resilver_done_hunt(vdev_t *vd) 5576{ 5577 vdev_t *newvd, *oldvd; 5578 5579 for (int c = 0; c < vd->vdev_children; c++) { 5580 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5581 if (oldvd != NULL) 5582 return (oldvd); 5583 } 5584 5585 /* 5586 * Check for a completed replacement. We always consider the first 5587 * vdev in the list to be the oldest vdev, and the last one to be 5588 * the newest (see spa_vdev_attach() for how that works). In 5589 * the case where the newest vdev is faulted, we will not automatically 5590 * remove it after a resilver completes. This is OK as it will require 5591 * user intervention to determine which disk the admin wishes to keep. 5592 */ 5593 if (vd->vdev_ops == &vdev_replacing_ops) { 5594 ASSERT(vd->vdev_children > 1); 5595 5596 newvd = vd->vdev_child[vd->vdev_children - 1]; 5597 oldvd = vd->vdev_child[0]; 5598 5599 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5600 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5601 !vdev_dtl_required(oldvd)) 5602 return (oldvd); 5603 } 5604 5605 /* 5606 * Check for a completed resilver with the 'unspare' flag set. 5607 */ 5608 if (vd->vdev_ops == &vdev_spare_ops) { 5609 vdev_t *first = vd->vdev_child[0]; 5610 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5611 5612 if (last->vdev_unspare) { 5613 oldvd = first; 5614 newvd = last; 5615 } else if (first->vdev_unspare) { 5616 oldvd = last; 5617 newvd = first; 5618 } else { 5619 oldvd = NULL; 5620 } 5621 5622 if (oldvd != NULL && 5623 vdev_dtl_empty(newvd, DTL_MISSING) && 5624 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5625 !vdev_dtl_required(oldvd)) 5626 return (oldvd); 5627 5628 /* 5629 * If there are more than two spares attached to a disk, 5630 * and those spares are not required, then we want to 5631 * attempt to free them up now so that they can be used 5632 * by other pools. Once we're back down to a single 5633 * disk+spare, we stop removing them. 5634 */ 5635 if (vd->vdev_children > 2) { 5636 newvd = vd->vdev_child[1]; 5637 5638 if (newvd->vdev_isspare && last->vdev_isspare && 5639 vdev_dtl_empty(last, DTL_MISSING) && 5640 vdev_dtl_empty(last, DTL_OUTAGE) && 5641 !vdev_dtl_required(newvd)) 5642 return (newvd); 5643 } 5644 } 5645 5646 return (NULL); 5647} 5648 5649static void 5650spa_vdev_resilver_done(spa_t *spa) 5651{ 5652 vdev_t *vd, *pvd, *ppvd; 5653 uint64_t guid, sguid, pguid, ppguid; 5654 5655 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5656 5657 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5658 pvd = vd->vdev_parent; 5659 ppvd = pvd->vdev_parent; 5660 guid = vd->vdev_guid; 5661 pguid = pvd->vdev_guid; 5662 ppguid = ppvd->vdev_guid; 5663 sguid = 0; 5664 /* 5665 * If we have just finished replacing a hot spared device, then 5666 * we need to detach the parent's first child (the original hot 5667 * spare) as well. 5668 */ 5669 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5670 ppvd->vdev_children == 2) { 5671 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5672 sguid = ppvd->vdev_child[1]->vdev_guid; 5673 } 5674 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5675 5676 spa_config_exit(spa, SCL_ALL, FTAG); 5677 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5678 return; 5679 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5680 return; 5681 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5682 } 5683 5684 spa_config_exit(spa, SCL_ALL, FTAG); 5685} 5686 5687/* 5688 * Update the stored path or FRU for this vdev. 5689 */ 5690int 5691spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5692 boolean_t ispath) 5693{ 5694 vdev_t *vd; 5695 boolean_t sync = B_FALSE; 5696 5697 ASSERT(spa_writeable(spa)); 5698 5699 spa_vdev_state_enter(spa, SCL_ALL); 5700 5701 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5702 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5703 5704 if (!vd->vdev_ops->vdev_op_leaf) 5705 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5706 5707 if (ispath) { 5708 if (strcmp(value, vd->vdev_path) != 0) { 5709 spa_strfree(vd->vdev_path); 5710 vd->vdev_path = spa_strdup(value); 5711 sync = B_TRUE; 5712 } 5713 } else { 5714 if (vd->vdev_fru == NULL) { 5715 vd->vdev_fru = spa_strdup(value); 5716 sync = B_TRUE; 5717 } else if (strcmp(value, vd->vdev_fru) != 0) { 5718 spa_strfree(vd->vdev_fru); 5719 vd->vdev_fru = spa_strdup(value); 5720 sync = B_TRUE; 5721 } 5722 } 5723 5724 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5725} 5726 5727int 5728spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5729{ 5730 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5731} 5732 5733int 5734spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5735{ 5736 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5737} 5738 5739/* 5740 * ========================================================================== 5741 * SPA Scanning 5742 * ========================================================================== 5743 */ 5744 5745int 5746spa_scan_stop(spa_t *spa) 5747{ 5748 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5749 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5750 return (SET_ERROR(EBUSY)); 5751 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5752} 5753 5754int 5755spa_scan(spa_t *spa, pool_scan_func_t func) 5756{ 5757 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5758 5759 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5760 return (SET_ERROR(ENOTSUP)); 5761 5762 /* 5763 * If a resilver was requested, but there is no DTL on a 5764 * writeable leaf device, we have nothing to do. 5765 */ 5766 if (func == POOL_SCAN_RESILVER && 5767 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5768 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5769 return (0); 5770 } 5771 5772 return (dsl_scan(spa->spa_dsl_pool, func)); 5773} 5774 5775/* 5776 * ========================================================================== 5777 * SPA async task processing 5778 * ========================================================================== 5779 */ 5780 5781static void 5782spa_async_remove(spa_t *spa, vdev_t *vd) 5783{ 5784 if (vd->vdev_remove_wanted) { 5785 vd->vdev_remove_wanted = B_FALSE; 5786 vd->vdev_delayed_close = B_FALSE; 5787 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5788 5789 /* 5790 * We want to clear the stats, but we don't want to do a full 5791 * vdev_clear() as that will cause us to throw away 5792 * degraded/faulted state as well as attempt to reopen the 5793 * device, all of which is a waste. 5794 */ 5795 vd->vdev_stat.vs_read_errors = 0; 5796 vd->vdev_stat.vs_write_errors = 0; 5797 vd->vdev_stat.vs_checksum_errors = 0; 5798 5799 vdev_state_dirty(vd->vdev_top); 5800 } 5801 5802 for (int c = 0; c < vd->vdev_children; c++) 5803 spa_async_remove(spa, vd->vdev_child[c]); 5804} 5805 5806static void 5807spa_async_probe(spa_t *spa, vdev_t *vd) 5808{ 5809 if (vd->vdev_probe_wanted) { 5810 vd->vdev_probe_wanted = B_FALSE; 5811 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5812 } 5813 5814 for (int c = 0; c < vd->vdev_children; c++) 5815 spa_async_probe(spa, vd->vdev_child[c]); 5816} 5817 5818static void 5819spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5820{ 5821 sysevent_id_t eid; 5822 nvlist_t *attr; 5823 char *physpath; 5824 5825 if (!spa->spa_autoexpand) 5826 return; 5827 5828 for (int c = 0; c < vd->vdev_children; c++) { 5829 vdev_t *cvd = vd->vdev_child[c]; 5830 spa_async_autoexpand(spa, cvd); 5831 } 5832 5833 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5834 return; 5835 5836 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5837 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5838 5839 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5840 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5841 5842 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5843 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5844 5845 nvlist_free(attr); 5846 kmem_free(physpath, MAXPATHLEN); 5847} 5848 5849static void 5850spa_async_thread(void *arg) 5851{ 5852 spa_t *spa = arg; 5853 int tasks; 5854 5855 ASSERT(spa->spa_sync_on); 5856 5857 mutex_enter(&spa->spa_async_lock); 5858 tasks = spa->spa_async_tasks; 5859 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5860 mutex_exit(&spa->spa_async_lock); 5861 5862 /* 5863 * See if the config needs to be updated. 5864 */ 5865 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5866 uint64_t old_space, new_space; 5867 5868 mutex_enter(&spa_namespace_lock); 5869 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5870 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5871 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5872 mutex_exit(&spa_namespace_lock); 5873 5874 /* 5875 * If the pool grew as a result of the config update, 5876 * then log an internal history event. 5877 */ 5878 if (new_space != old_space) { 5879 spa_history_log_internal(spa, "vdev online", NULL, 5880 "pool '%s' size: %llu(+%llu)", 5881 spa_name(spa), new_space, new_space - old_space); 5882 } 5883 } 5884 5885 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5886 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5887 spa_async_autoexpand(spa, spa->spa_root_vdev); 5888 spa_config_exit(spa, SCL_CONFIG, FTAG); 5889 } 5890 5891 /* 5892 * See if any devices need to be probed. 5893 */ 5894 if (tasks & SPA_ASYNC_PROBE) { 5895 spa_vdev_state_enter(spa, SCL_NONE); 5896 spa_async_probe(spa, spa->spa_root_vdev); 5897 (void) spa_vdev_state_exit(spa, NULL, 0); 5898 } 5899 5900 /* 5901 * If any devices are done replacing, detach them. 5902 */ 5903 if (tasks & SPA_ASYNC_RESILVER_DONE) 5904 spa_vdev_resilver_done(spa); 5905 5906 /* 5907 * Kick off a resilver. 5908 */ 5909 if (tasks & SPA_ASYNC_RESILVER) 5910 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5911 5912 /* 5913 * Let the world know that we're done. 5914 */ 5915 mutex_enter(&spa->spa_async_lock); 5916 spa->spa_async_thread = NULL; 5917 cv_broadcast(&spa->spa_async_cv); 5918 mutex_exit(&spa->spa_async_lock); 5919 thread_exit(); 5920} 5921 5922static void 5923spa_async_thread_vd(void *arg) 5924{ 5925 spa_t *spa = arg; 5926 int tasks; 5927 5928 ASSERT(spa->spa_sync_on); 5929 5930 mutex_enter(&spa->spa_async_lock); 5931 tasks = spa->spa_async_tasks; 5932retry: 5933 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 5934 mutex_exit(&spa->spa_async_lock); 5935 5936 /* 5937 * See if any devices need to be marked REMOVED. 5938 */ 5939 if (tasks & SPA_ASYNC_REMOVE) { 5940 spa_vdev_state_enter(spa, SCL_NONE); 5941 spa_async_remove(spa, spa->spa_root_vdev); 5942 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5943 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5944 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5945 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5946 (void) spa_vdev_state_exit(spa, NULL, 0); 5947 } 5948 5949 /* 5950 * Let the world know that we're done. 5951 */ 5952 mutex_enter(&spa->spa_async_lock); 5953 tasks = spa->spa_async_tasks; 5954 if ((tasks & SPA_ASYNC_REMOVE) != 0) 5955 goto retry; 5956 spa->spa_async_thread_vd = NULL; 5957 cv_broadcast(&spa->spa_async_cv); 5958 mutex_exit(&spa->spa_async_lock); 5959 thread_exit(); 5960} 5961 5962void 5963spa_async_suspend(spa_t *spa) 5964{ 5965 mutex_enter(&spa->spa_async_lock); 5966 spa->spa_async_suspended++; 5967 while (spa->spa_async_thread != NULL && 5968 spa->spa_async_thread_vd != NULL) 5969 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5970 mutex_exit(&spa->spa_async_lock); 5971} 5972 5973void 5974spa_async_resume(spa_t *spa) 5975{ 5976 mutex_enter(&spa->spa_async_lock); 5977 ASSERT(spa->spa_async_suspended != 0); 5978 spa->spa_async_suspended--; 5979 mutex_exit(&spa->spa_async_lock); 5980} 5981 5982static boolean_t 5983spa_async_tasks_pending(spa_t *spa) 5984{ 5985 uint_t non_config_tasks; 5986 uint_t config_task; 5987 boolean_t config_task_suspended; 5988 5989 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 5990 SPA_ASYNC_REMOVE); 5991 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 5992 if (spa->spa_ccw_fail_time == 0) { 5993 config_task_suspended = B_FALSE; 5994 } else { 5995 config_task_suspended = 5996 (gethrtime() - spa->spa_ccw_fail_time) < 5997 (zfs_ccw_retry_interval * NANOSEC); 5998 } 5999 6000 return (non_config_tasks || (config_task && !config_task_suspended)); 6001} 6002 6003static void 6004spa_async_dispatch(spa_t *spa) 6005{ 6006 mutex_enter(&spa->spa_async_lock); 6007 if (spa_async_tasks_pending(spa) && 6008 !spa->spa_async_suspended && 6009 spa->spa_async_thread == NULL && 6010 rootdir != NULL) 6011 spa->spa_async_thread = thread_create(NULL, 0, 6012 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6013 mutex_exit(&spa->spa_async_lock); 6014} 6015 6016static void 6017spa_async_dispatch_vd(spa_t *spa) 6018{ 6019 mutex_enter(&spa->spa_async_lock); 6020 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6021 !spa->spa_async_suspended && 6022 spa->spa_async_thread_vd == NULL && 6023 rootdir != NULL) 6024 spa->spa_async_thread_vd = thread_create(NULL, 0, 6025 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6026 mutex_exit(&spa->spa_async_lock); 6027} 6028 6029void 6030spa_async_request(spa_t *spa, int task) 6031{ 6032 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6033 mutex_enter(&spa->spa_async_lock); 6034 spa->spa_async_tasks |= task; 6035 mutex_exit(&spa->spa_async_lock); 6036 spa_async_dispatch_vd(spa); 6037} 6038 6039/* 6040 * ========================================================================== 6041 * SPA syncing routines 6042 * ========================================================================== 6043 */ 6044 6045static int 6046bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6047{ 6048 bpobj_t *bpo = arg; 6049 bpobj_enqueue(bpo, bp, tx); 6050 return (0); 6051} 6052 6053static int 6054spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6055{ 6056 zio_t *zio = arg; 6057 6058 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6059 BP_GET_PSIZE(bp), zio->io_flags)); 6060 return (0); 6061} 6062 6063/* 6064 * Note: this simple function is not inlined to make it easier to dtrace the 6065 * amount of time spent syncing frees. 6066 */ 6067static void 6068spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6069{ 6070 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6071 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6072 VERIFY(zio_wait(zio) == 0); 6073} 6074 6075/* 6076 * Note: this simple function is not inlined to make it easier to dtrace the 6077 * amount of time spent syncing deferred frees. 6078 */ 6079static void 6080spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6081{ 6082 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6083 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6084 spa_free_sync_cb, zio, tx), ==, 0); 6085 VERIFY0(zio_wait(zio)); 6086} 6087 6088 6089static void 6090spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6091{ 6092 char *packed = NULL; 6093 size_t bufsize; 6094 size_t nvsize = 0; 6095 dmu_buf_t *db; 6096 6097 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6098 6099 /* 6100 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6101 * information. This avoids the dbuf_will_dirty() path and 6102 * saves us a pre-read to get data we don't actually care about. 6103 */ 6104 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6105 packed = kmem_alloc(bufsize, KM_SLEEP); 6106 6107 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6108 KM_SLEEP) == 0); 6109 bzero(packed + nvsize, bufsize - nvsize); 6110 6111 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6112 6113 kmem_free(packed, bufsize); 6114 6115 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6116 dmu_buf_will_dirty(db, tx); 6117 *(uint64_t *)db->db_data = nvsize; 6118 dmu_buf_rele(db, FTAG); 6119} 6120 6121static void 6122spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6123 const char *config, const char *entry) 6124{ 6125 nvlist_t *nvroot; 6126 nvlist_t **list; 6127 int i; 6128 6129 if (!sav->sav_sync) 6130 return; 6131 6132 /* 6133 * Update the MOS nvlist describing the list of available devices. 6134 * spa_validate_aux() will have already made sure this nvlist is 6135 * valid and the vdevs are labeled appropriately. 6136 */ 6137 if (sav->sav_object == 0) { 6138 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6139 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6140 sizeof (uint64_t), tx); 6141 VERIFY(zap_update(spa->spa_meta_objset, 6142 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6143 &sav->sav_object, tx) == 0); 6144 } 6145 6146 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6147 if (sav->sav_count == 0) { 6148 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6149 } else { 6150 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6151 for (i = 0; i < sav->sav_count; i++) 6152 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6153 B_FALSE, VDEV_CONFIG_L2CACHE); 6154 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6155 sav->sav_count) == 0); 6156 for (i = 0; i < sav->sav_count; i++) 6157 nvlist_free(list[i]); 6158 kmem_free(list, sav->sav_count * sizeof (void *)); 6159 } 6160 6161 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6162 nvlist_free(nvroot); 6163 6164 sav->sav_sync = B_FALSE; 6165} 6166 6167static void 6168spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6169{ 6170 nvlist_t *config; 6171 6172 if (list_is_empty(&spa->spa_config_dirty_list)) 6173 return; 6174 6175 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6176 6177 config = spa_config_generate(spa, spa->spa_root_vdev, 6178 dmu_tx_get_txg(tx), B_FALSE); 6179 6180 /* 6181 * If we're upgrading the spa version then make sure that 6182 * the config object gets updated with the correct version. 6183 */ 6184 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6185 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6186 spa->spa_uberblock.ub_version); 6187 6188 spa_config_exit(spa, SCL_STATE, FTAG); 6189 6190 if (spa->spa_config_syncing) 6191 nvlist_free(spa->spa_config_syncing); 6192 spa->spa_config_syncing = config; 6193 6194 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6195} 6196 6197static void 6198spa_sync_version(void *arg, dmu_tx_t *tx) 6199{ 6200 uint64_t *versionp = arg; 6201 uint64_t version = *versionp; 6202 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6203 6204 /* 6205 * Setting the version is special cased when first creating the pool. 6206 */ 6207 ASSERT(tx->tx_txg != TXG_INITIAL); 6208 6209 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6210 ASSERT(version >= spa_version(spa)); 6211 6212 spa->spa_uberblock.ub_version = version; 6213 vdev_config_dirty(spa->spa_root_vdev); 6214 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6215} 6216 6217/* 6218 * Set zpool properties. 6219 */ 6220static void 6221spa_sync_props(void *arg, dmu_tx_t *tx) 6222{ 6223 nvlist_t *nvp = arg; 6224 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6225 objset_t *mos = spa->spa_meta_objset; 6226 nvpair_t *elem = NULL; 6227 6228 mutex_enter(&spa->spa_props_lock); 6229 6230 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6231 uint64_t intval; 6232 char *strval, *fname; 6233 zpool_prop_t prop; 6234 const char *propname; 6235 zprop_type_t proptype; 6236 zfeature_info_t *feature; 6237 6238 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6239 case ZPROP_INVAL: 6240 /* 6241 * We checked this earlier in spa_prop_validate(). 6242 */ 6243 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6244 6245 fname = strchr(nvpair_name(elem), '@') + 1; 6246 VERIFY0(zfeature_lookup_name(fname, &feature)); 6247 6248 spa_feature_enable(spa, feature, tx); 6249 spa_history_log_internal(spa, "set", tx, 6250 "%s=enabled", nvpair_name(elem)); 6251 break; 6252 6253 case ZPOOL_PROP_VERSION: 6254 intval = fnvpair_value_uint64(elem); 6255 /* 6256 * The version is synced seperatly before other 6257 * properties and should be correct by now. 6258 */ 6259 ASSERT3U(spa_version(spa), >=, intval); 6260 break; 6261 6262 case ZPOOL_PROP_ALTROOT: 6263 /* 6264 * 'altroot' is a non-persistent property. It should 6265 * have been set temporarily at creation or import time. 6266 */ 6267 ASSERT(spa->spa_root != NULL); 6268 break; 6269 6270 case ZPOOL_PROP_READONLY: 6271 case ZPOOL_PROP_CACHEFILE: 6272 /* 6273 * 'readonly' and 'cachefile' are also non-persisitent 6274 * properties. 6275 */ 6276 break; 6277 case ZPOOL_PROP_COMMENT: 6278 strval = fnvpair_value_string(elem); 6279 if (spa->spa_comment != NULL) 6280 spa_strfree(spa->spa_comment); 6281 spa->spa_comment = spa_strdup(strval); 6282 /* 6283 * We need to dirty the configuration on all the vdevs 6284 * so that their labels get updated. It's unnecessary 6285 * to do this for pool creation since the vdev's 6286 * configuratoin has already been dirtied. 6287 */ 6288 if (tx->tx_txg != TXG_INITIAL) 6289 vdev_config_dirty(spa->spa_root_vdev); 6290 spa_history_log_internal(spa, "set", tx, 6291 "%s=%s", nvpair_name(elem), strval); 6292 break; 6293 default: 6294 /* 6295 * Set pool property values in the poolprops mos object. 6296 */ 6297 if (spa->spa_pool_props_object == 0) { 6298 spa->spa_pool_props_object = 6299 zap_create_link(mos, DMU_OT_POOL_PROPS, 6300 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6301 tx); 6302 } 6303 6304 /* normalize the property name */ 6305 propname = zpool_prop_to_name(prop); 6306 proptype = zpool_prop_get_type(prop); 6307 6308 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6309 ASSERT(proptype == PROP_TYPE_STRING); 6310 strval = fnvpair_value_string(elem); 6311 VERIFY0(zap_update(mos, 6312 spa->spa_pool_props_object, propname, 6313 1, strlen(strval) + 1, strval, tx)); 6314 spa_history_log_internal(spa, "set", tx, 6315 "%s=%s", nvpair_name(elem), strval); 6316 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6317 intval = fnvpair_value_uint64(elem); 6318 6319 if (proptype == PROP_TYPE_INDEX) { 6320 const char *unused; 6321 VERIFY0(zpool_prop_index_to_string( 6322 prop, intval, &unused)); 6323 } 6324 VERIFY0(zap_update(mos, 6325 spa->spa_pool_props_object, propname, 6326 8, 1, &intval, tx)); 6327 spa_history_log_internal(spa, "set", tx, 6328 "%s=%lld", nvpair_name(elem), intval); 6329 } else { 6330 ASSERT(0); /* not allowed */ 6331 } 6332 6333 switch (prop) { 6334 case ZPOOL_PROP_DELEGATION: 6335 spa->spa_delegation = intval; 6336 break; 6337 case ZPOOL_PROP_BOOTFS: 6338 spa->spa_bootfs = intval; 6339 break; 6340 case ZPOOL_PROP_FAILUREMODE: 6341 spa->spa_failmode = intval; 6342 break; 6343 case ZPOOL_PROP_AUTOEXPAND: 6344 spa->spa_autoexpand = intval; 6345 if (tx->tx_txg != TXG_INITIAL) 6346 spa_async_request(spa, 6347 SPA_ASYNC_AUTOEXPAND); 6348 break; 6349 case ZPOOL_PROP_DEDUPDITTO: 6350 spa->spa_dedup_ditto = intval; 6351 break; 6352 default: 6353 break; 6354 } 6355 } 6356 6357 } 6358 6359 mutex_exit(&spa->spa_props_lock); 6360} 6361 6362/* 6363 * Perform one-time upgrade on-disk changes. spa_version() does not 6364 * reflect the new version this txg, so there must be no changes this 6365 * txg to anything that the upgrade code depends on after it executes. 6366 * Therefore this must be called after dsl_pool_sync() does the sync 6367 * tasks. 6368 */ 6369static void 6370spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6371{ 6372 dsl_pool_t *dp = spa->spa_dsl_pool; 6373 6374 ASSERT(spa->spa_sync_pass == 1); 6375 6376 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6377 6378 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6379 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6380 dsl_pool_create_origin(dp, tx); 6381 6382 /* Keeping the origin open increases spa_minref */ 6383 spa->spa_minref += 3; 6384 } 6385 6386 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6387 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6388 dsl_pool_upgrade_clones(dp, tx); 6389 } 6390 6391 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6392 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6393 dsl_pool_upgrade_dir_clones(dp, tx); 6394 6395 /* Keeping the freedir open increases spa_minref */ 6396 spa->spa_minref += 3; 6397 } 6398 6399 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6400 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6401 spa_feature_create_zap_objects(spa, tx); 6402 } 6403 rrw_exit(&dp->dp_config_rwlock, FTAG); 6404} 6405 6406/* 6407 * Sync the specified transaction group. New blocks may be dirtied as 6408 * part of the process, so we iterate until it converges. 6409 */ 6410void 6411spa_sync(spa_t *spa, uint64_t txg) 6412{ 6413 dsl_pool_t *dp = spa->spa_dsl_pool; 6414 objset_t *mos = spa->spa_meta_objset; 6415 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6416 vdev_t *rvd = spa->spa_root_vdev; 6417 vdev_t *vd; 6418 dmu_tx_t *tx; 6419 int error; 6420 6421 VERIFY(spa_writeable(spa)); 6422 6423 /* 6424 * Lock out configuration changes. 6425 */ 6426 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6427 6428 spa->spa_syncing_txg = txg; 6429 spa->spa_sync_pass = 0; 6430 6431 /* 6432 * If there are any pending vdev state changes, convert them 6433 * into config changes that go out with this transaction group. 6434 */ 6435 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6436 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6437 /* 6438 * We need the write lock here because, for aux vdevs, 6439 * calling vdev_config_dirty() modifies sav_config. 6440 * This is ugly and will become unnecessary when we 6441 * eliminate the aux vdev wart by integrating all vdevs 6442 * into the root vdev tree. 6443 */ 6444 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6445 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6446 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6447 vdev_state_clean(vd); 6448 vdev_config_dirty(vd); 6449 } 6450 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6451 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6452 } 6453 spa_config_exit(spa, SCL_STATE, FTAG); 6454 6455 tx = dmu_tx_create_assigned(dp, txg); 6456 6457 spa->spa_sync_starttime = gethrtime(); 6458#ifdef illumos 6459 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6460 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6461#else /* FreeBSD */ 6462#ifdef _KERNEL 6463 callout_reset(&spa->spa_deadman_cycid, 6464 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6465#endif 6466#endif 6467 6468 /* 6469 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6470 * set spa_deflate if we have no raid-z vdevs. 6471 */ 6472 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6473 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6474 int i; 6475 6476 for (i = 0; i < rvd->vdev_children; i++) { 6477 vd = rvd->vdev_child[i]; 6478 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6479 break; 6480 } 6481 if (i == rvd->vdev_children) { 6482 spa->spa_deflate = TRUE; 6483 VERIFY(0 == zap_add(spa->spa_meta_objset, 6484 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6485 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6486 } 6487 } 6488 6489 /* 6490 * If anything has changed in this txg, or if someone is waiting 6491 * for this txg to sync (eg, spa_vdev_remove()), push the 6492 * deferred frees from the previous txg. If not, leave them 6493 * alone so that we don't generate work on an otherwise idle 6494 * system. 6495 */ 6496 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6497 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6498 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6499 ((dsl_scan_active(dp->dp_scan) || 6500 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6501 spa_sync_deferred_frees(spa, tx); 6502 } 6503 6504 /* 6505 * Iterate to convergence. 6506 */ 6507 do { 6508 int pass = ++spa->spa_sync_pass; 6509 6510 spa_sync_config_object(spa, tx); 6511 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6512 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6513 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6514 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6515 spa_errlog_sync(spa, txg); 6516 dsl_pool_sync(dp, txg); 6517 6518 if (pass < zfs_sync_pass_deferred_free) { 6519 spa_sync_frees(spa, free_bpl, tx); 6520 } else { 6521 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6522 &spa->spa_deferred_bpobj, tx); 6523 } 6524 6525 ddt_sync(spa, txg); 6526 dsl_scan_sync(dp, tx); 6527 6528 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6529 vdev_sync(vd, txg); 6530 6531 if (pass == 1) 6532 spa_sync_upgrades(spa, tx); 6533 6534 } while (dmu_objset_is_dirty(mos, txg)); 6535 6536 /* 6537 * Rewrite the vdev configuration (which includes the uberblock) 6538 * to commit the transaction group. 6539 * 6540 * If there are no dirty vdevs, we sync the uberblock to a few 6541 * random top-level vdevs that are known to be visible in the 6542 * config cache (see spa_vdev_add() for a complete description). 6543 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6544 */ 6545 for (;;) { 6546 /* 6547 * We hold SCL_STATE to prevent vdev open/close/etc. 6548 * while we're attempting to write the vdev labels. 6549 */ 6550 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6551 6552 if (list_is_empty(&spa->spa_config_dirty_list)) { 6553 vdev_t *svd[SPA_DVAS_PER_BP]; 6554 int svdcount = 0; 6555 int children = rvd->vdev_children; 6556 int c0 = spa_get_random(children); 6557 6558 for (int c = 0; c < children; c++) { 6559 vd = rvd->vdev_child[(c0 + c) % children]; 6560 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6561 continue; 6562 svd[svdcount++] = vd; 6563 if (svdcount == SPA_DVAS_PER_BP) 6564 break; 6565 } 6566 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6567 if (error != 0) 6568 error = vdev_config_sync(svd, svdcount, txg, 6569 B_TRUE); 6570 } else { 6571 error = vdev_config_sync(rvd->vdev_child, 6572 rvd->vdev_children, txg, B_FALSE); 6573 if (error != 0) 6574 error = vdev_config_sync(rvd->vdev_child, 6575 rvd->vdev_children, txg, B_TRUE); 6576 } 6577 6578 if (error == 0) 6579 spa->spa_last_synced_guid = rvd->vdev_guid; 6580 6581 spa_config_exit(spa, SCL_STATE, FTAG); 6582 6583 if (error == 0) 6584 break; 6585 zio_suspend(spa, NULL); 6586 zio_resume_wait(spa); 6587 } 6588 dmu_tx_commit(tx); 6589 6590#ifdef illumos 6591 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6592#else /* FreeBSD */ 6593#ifdef _KERNEL 6594 callout_drain(&spa->spa_deadman_cycid); 6595#endif 6596#endif 6597 6598 /* 6599 * Clear the dirty config list. 6600 */ 6601 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6602 vdev_config_clean(vd); 6603 6604 /* 6605 * Now that the new config has synced transactionally, 6606 * let it become visible to the config cache. 6607 */ 6608 if (spa->spa_config_syncing != NULL) { 6609 spa_config_set(spa, spa->spa_config_syncing); 6610 spa->spa_config_txg = txg; 6611 spa->spa_config_syncing = NULL; 6612 } 6613 6614 spa->spa_ubsync = spa->spa_uberblock; 6615 6616 dsl_pool_sync_done(dp, txg); 6617 6618 /* 6619 * Update usable space statistics. 6620 */ 6621 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6622 vdev_sync_done(vd, txg); 6623 6624 spa_update_dspace(spa); 6625 6626 /* 6627 * It had better be the case that we didn't dirty anything 6628 * since vdev_config_sync(). 6629 */ 6630 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6631 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6632 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6633 6634 spa->spa_sync_pass = 0; 6635 6636 spa_config_exit(spa, SCL_CONFIG, FTAG); 6637 6638 spa_handle_ignored_writes(spa); 6639 6640 /* 6641 * If any async tasks have been requested, kick them off. 6642 */ 6643 spa_async_dispatch(spa); 6644 spa_async_dispatch_vd(spa); 6645} 6646 6647/* 6648 * Sync all pools. We don't want to hold the namespace lock across these 6649 * operations, so we take a reference on the spa_t and drop the lock during the 6650 * sync. 6651 */ 6652void 6653spa_sync_allpools(void) 6654{ 6655 spa_t *spa = NULL; 6656 mutex_enter(&spa_namespace_lock); 6657 while ((spa = spa_next(spa)) != NULL) { 6658 if (spa_state(spa) != POOL_STATE_ACTIVE || 6659 !spa_writeable(spa) || spa_suspended(spa)) 6660 continue; 6661 spa_open_ref(spa, FTAG); 6662 mutex_exit(&spa_namespace_lock); 6663 txg_wait_synced(spa_get_dsl(spa), 0); 6664 mutex_enter(&spa_namespace_lock); 6665 spa_close(spa, FTAG); 6666 } 6667 mutex_exit(&spa_namespace_lock); 6668} 6669 6670/* 6671 * ========================================================================== 6672 * Miscellaneous routines 6673 * ========================================================================== 6674 */ 6675 6676/* 6677 * Remove all pools in the system. 6678 */ 6679void 6680spa_evict_all(void) 6681{ 6682 spa_t *spa; 6683 6684 /* 6685 * Remove all cached state. All pools should be closed now, 6686 * so every spa in the AVL tree should be unreferenced. 6687 */ 6688 mutex_enter(&spa_namespace_lock); 6689 while ((spa = spa_next(NULL)) != NULL) { 6690 /* 6691 * Stop async tasks. The async thread may need to detach 6692 * a device that's been replaced, which requires grabbing 6693 * spa_namespace_lock, so we must drop it here. 6694 */ 6695 spa_open_ref(spa, FTAG); 6696 mutex_exit(&spa_namespace_lock); 6697 spa_async_suspend(spa); 6698 mutex_enter(&spa_namespace_lock); 6699 spa_close(spa, FTAG); 6700 6701 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6702 spa_unload(spa); 6703 spa_deactivate(spa); 6704 } 6705 spa_remove(spa); 6706 } 6707 mutex_exit(&spa_namespace_lock); 6708} 6709 6710vdev_t * 6711spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6712{ 6713 vdev_t *vd; 6714 int i; 6715 6716 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6717 return (vd); 6718 6719 if (aux) { 6720 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6721 vd = spa->spa_l2cache.sav_vdevs[i]; 6722 if (vd->vdev_guid == guid) 6723 return (vd); 6724 } 6725 6726 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6727 vd = spa->spa_spares.sav_vdevs[i]; 6728 if (vd->vdev_guid == guid) 6729 return (vd); 6730 } 6731 } 6732 6733 return (NULL); 6734} 6735 6736void 6737spa_upgrade(spa_t *spa, uint64_t version) 6738{ 6739 ASSERT(spa_writeable(spa)); 6740 6741 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6742 6743 /* 6744 * This should only be called for a non-faulted pool, and since a 6745 * future version would result in an unopenable pool, this shouldn't be 6746 * possible. 6747 */ 6748 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6749 ASSERT(version >= spa->spa_uberblock.ub_version); 6750 6751 spa->spa_uberblock.ub_version = version; 6752 vdev_config_dirty(spa->spa_root_vdev); 6753 6754 spa_config_exit(spa, SCL_ALL, FTAG); 6755 6756 txg_wait_synced(spa_get_dsl(spa), 0); 6757} 6758 6759boolean_t 6760spa_has_spare(spa_t *spa, uint64_t guid) 6761{ 6762 int i; 6763 uint64_t spareguid; 6764 spa_aux_vdev_t *sav = &spa->spa_spares; 6765 6766 for (i = 0; i < sav->sav_count; i++) 6767 if (sav->sav_vdevs[i]->vdev_guid == guid) 6768 return (B_TRUE); 6769 6770 for (i = 0; i < sav->sav_npending; i++) { 6771 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6772 &spareguid) == 0 && spareguid == guid) 6773 return (B_TRUE); 6774 } 6775 6776 return (B_FALSE); 6777} 6778 6779/* 6780 * Check if a pool has an active shared spare device. 6781 * Note: reference count of an active spare is 2, as a spare and as a replace 6782 */ 6783static boolean_t 6784spa_has_active_shared_spare(spa_t *spa) 6785{ 6786 int i, refcnt; 6787 uint64_t pool; 6788 spa_aux_vdev_t *sav = &spa->spa_spares; 6789 6790 for (i = 0; i < sav->sav_count; i++) { 6791 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6792 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6793 refcnt > 2) 6794 return (B_TRUE); 6795 } 6796 6797 return (B_FALSE); 6798} 6799 6800/* 6801 * Post a sysevent corresponding to the given event. The 'name' must be one of 6802 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6803 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6804 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6805 * or zdb as real changes. 6806 */ 6807void 6808spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6809{ 6810#ifdef _KERNEL 6811 sysevent_t *ev; 6812 sysevent_attr_list_t *attr = NULL; 6813 sysevent_value_t value; 6814 sysevent_id_t eid; 6815 6816 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6817 SE_SLEEP); 6818 6819 value.value_type = SE_DATA_TYPE_STRING; 6820 value.value.sv_string = spa_name(spa); 6821 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6822 goto done; 6823 6824 value.value_type = SE_DATA_TYPE_UINT64; 6825 value.value.sv_uint64 = spa_guid(spa); 6826 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6827 goto done; 6828 6829 if (vd) { 6830 value.value_type = SE_DATA_TYPE_UINT64; 6831 value.value.sv_uint64 = vd->vdev_guid; 6832 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6833 SE_SLEEP) != 0) 6834 goto done; 6835 6836 if (vd->vdev_path) { 6837 value.value_type = SE_DATA_TYPE_STRING; 6838 value.value.sv_string = vd->vdev_path; 6839 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6840 &value, SE_SLEEP) != 0) 6841 goto done; 6842 } 6843 } 6844 6845 if (sysevent_attach_attributes(ev, attr) != 0) 6846 goto done; 6847 attr = NULL; 6848 6849 (void) log_sysevent(ev, SE_SLEEP, &eid); 6850 6851done: 6852 if (attr) 6853 sysevent_free_attr(attr); 6854 sysevent_free(ev); 6855#endif 6856} 6857