spa.c revision 260763
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 /* 1279 * Close the dsl pool. 1280 */ 1281 if (spa->spa_dsl_pool) { 1282 dsl_pool_close(spa->spa_dsl_pool); 1283 spa->spa_dsl_pool = NULL; 1284 spa->spa_meta_objset = NULL; 1285 } 1286 1287 ddt_unload(spa); 1288 1289 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1290 1291 /* 1292 * Drop and purge level 2 cache 1293 */ 1294 spa_l2cache_drop(spa); 1295 1296 /* 1297 * Close all vdevs. 1298 */ 1299 if (spa->spa_root_vdev) 1300 vdev_free(spa->spa_root_vdev); 1301 ASSERT(spa->spa_root_vdev == NULL); 1302 1303 for (i = 0; i < spa->spa_spares.sav_count; i++) 1304 vdev_free(spa->spa_spares.sav_vdevs[i]); 1305 if (spa->spa_spares.sav_vdevs) { 1306 kmem_free(spa->spa_spares.sav_vdevs, 1307 spa->spa_spares.sav_count * sizeof (void *)); 1308 spa->spa_spares.sav_vdevs = NULL; 1309 } 1310 if (spa->spa_spares.sav_config) { 1311 nvlist_free(spa->spa_spares.sav_config); 1312 spa->spa_spares.sav_config = NULL; 1313 } 1314 spa->spa_spares.sav_count = 0; 1315 1316 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 1317 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]); 1318 vdev_free(spa->spa_l2cache.sav_vdevs[i]); 1319 } 1320 if (spa->spa_l2cache.sav_vdevs) { 1321 kmem_free(spa->spa_l2cache.sav_vdevs, 1322 spa->spa_l2cache.sav_count * sizeof (void *)); 1323 spa->spa_l2cache.sav_vdevs = NULL; 1324 } 1325 if (spa->spa_l2cache.sav_config) { 1326 nvlist_free(spa->spa_l2cache.sav_config); 1327 spa->spa_l2cache.sav_config = NULL; 1328 } 1329 spa->spa_l2cache.sav_count = 0; 1330 1331 spa->spa_async_suspended = 0; 1332 1333 if (spa->spa_comment != NULL) { 1334 spa_strfree(spa->spa_comment); 1335 spa->spa_comment = NULL; 1336 } 1337 1338 spa_config_exit(spa, SCL_ALL, FTAG); 1339} 1340 1341/* 1342 * Load (or re-load) the current list of vdevs describing the active spares for 1343 * this pool. When this is called, we have some form of basic information in 1344 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and 1345 * then re-generate a more complete list including status information. 1346 */ 1347static void 1348spa_load_spares(spa_t *spa) 1349{ 1350 nvlist_t **spares; 1351 uint_t nspares; 1352 int i; 1353 vdev_t *vd, *tvd; 1354 1355 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1356 1357 /* 1358 * First, close and free any existing spare vdevs. 1359 */ 1360 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1361 vd = spa->spa_spares.sav_vdevs[i]; 1362 1363 /* Undo the call to spa_activate() below */ 1364 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1365 B_FALSE)) != NULL && tvd->vdev_isspare) 1366 spa_spare_remove(tvd); 1367 vdev_close(vd); 1368 vdev_free(vd); 1369 } 1370 1371 if (spa->spa_spares.sav_vdevs) 1372 kmem_free(spa->spa_spares.sav_vdevs, 1373 spa->spa_spares.sav_count * sizeof (void *)); 1374 1375 if (spa->spa_spares.sav_config == NULL) 1376 nspares = 0; 1377 else 1378 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 1379 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 1380 1381 spa->spa_spares.sav_count = (int)nspares; 1382 spa->spa_spares.sav_vdevs = NULL; 1383 1384 if (nspares == 0) 1385 return; 1386 1387 /* 1388 * Construct the array of vdevs, opening them to get status in the 1389 * process. For each spare, there is potentially two different vdev_t 1390 * structures associated with it: one in the list of spares (used only 1391 * for basic validation purposes) and one in the active vdev 1392 * configuration (if it's spared in). During this phase we open and 1393 * validate each vdev on the spare list. If the vdev also exists in the 1394 * active configuration, then we also mark this vdev as an active spare. 1395 */ 1396 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *), 1397 KM_SLEEP); 1398 for (i = 0; i < spa->spa_spares.sav_count; i++) { 1399 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0, 1400 VDEV_ALLOC_SPARE) == 0); 1401 ASSERT(vd != NULL); 1402 1403 spa->spa_spares.sav_vdevs[i] = vd; 1404 1405 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid, 1406 B_FALSE)) != NULL) { 1407 if (!tvd->vdev_isspare) 1408 spa_spare_add(tvd); 1409 1410 /* 1411 * We only mark the spare active if we were successfully 1412 * able to load the vdev. Otherwise, importing a pool 1413 * with a bad active spare would result in strange 1414 * behavior, because multiple pool would think the spare 1415 * is actively in use. 1416 * 1417 * There is a vulnerability here to an equally bizarre 1418 * circumstance, where a dead active spare is later 1419 * brought back to life (onlined or otherwise). Given 1420 * the rarity of this scenario, and the extra complexity 1421 * it adds, we ignore the possibility. 1422 */ 1423 if (!vdev_is_dead(tvd)) 1424 spa_spare_activate(tvd); 1425 } 1426 1427 vd->vdev_top = vd; 1428 vd->vdev_aux = &spa->spa_spares; 1429 1430 if (vdev_open(vd) != 0) 1431 continue; 1432 1433 if (vdev_validate_aux(vd) == 0) 1434 spa_spare_add(vd); 1435 } 1436 1437 /* 1438 * Recompute the stashed list of spares, with status information 1439 * this time. 1440 */ 1441 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES, 1442 DATA_TYPE_NVLIST_ARRAY) == 0); 1443 1444 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *), 1445 KM_SLEEP); 1446 for (i = 0; i < spa->spa_spares.sav_count; i++) 1447 spares[i] = vdev_config_generate(spa, 1448 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE); 1449 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 1450 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0); 1451 for (i = 0; i < spa->spa_spares.sav_count; i++) 1452 nvlist_free(spares[i]); 1453 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *)); 1454} 1455 1456/* 1457 * Load (or re-load) the current list of vdevs describing the active l2cache for 1458 * this pool. When this is called, we have some form of basic information in 1459 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and 1460 * then re-generate a more complete list including status information. 1461 * Devices which are already active have their details maintained, and are 1462 * not re-opened. 1463 */ 1464static void 1465spa_load_l2cache(spa_t *spa) 1466{ 1467 nvlist_t **l2cache; 1468 uint_t nl2cache; 1469 int i, j, oldnvdevs; 1470 uint64_t guid; 1471 vdev_t *vd, **oldvdevs, **newvdevs; 1472 spa_aux_vdev_t *sav = &spa->spa_l2cache; 1473 1474 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 1475 1476 if (sav->sav_config != NULL) { 1477 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 1478 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 1479 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP); 1480 } else { 1481 nl2cache = 0; 1482 newvdevs = NULL; 1483 } 1484 1485 oldvdevs = sav->sav_vdevs; 1486 oldnvdevs = sav->sav_count; 1487 sav->sav_vdevs = NULL; 1488 sav->sav_count = 0; 1489 1490 /* 1491 * Process new nvlist of vdevs. 1492 */ 1493 for (i = 0; i < nl2cache; i++) { 1494 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID, 1495 &guid) == 0); 1496 1497 newvdevs[i] = NULL; 1498 for (j = 0; j < oldnvdevs; j++) { 1499 vd = oldvdevs[j]; 1500 if (vd != NULL && guid == vd->vdev_guid) { 1501 /* 1502 * Retain previous vdev for add/remove ops. 1503 */ 1504 newvdevs[i] = vd; 1505 oldvdevs[j] = NULL; 1506 break; 1507 } 1508 } 1509 1510 if (newvdevs[i] == NULL) { 1511 /* 1512 * Create new vdev 1513 */ 1514 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0, 1515 VDEV_ALLOC_L2CACHE) == 0); 1516 ASSERT(vd != NULL); 1517 newvdevs[i] = vd; 1518 1519 /* 1520 * Commit this vdev as an l2cache device, 1521 * even if it fails to open. 1522 */ 1523 spa_l2cache_add(vd); 1524 1525 vd->vdev_top = vd; 1526 vd->vdev_aux = sav; 1527 1528 spa_l2cache_activate(vd); 1529 1530 if (vdev_open(vd) != 0) 1531 continue; 1532 1533 (void) vdev_validate_aux(vd); 1534 1535 if (!vdev_is_dead(vd)) 1536 l2arc_add_vdev(spa, vd); 1537 } 1538 } 1539 1540 /* 1541 * Purge vdevs that were dropped 1542 */ 1543 for (i = 0; i < oldnvdevs; i++) { 1544 uint64_t pool; 1545 1546 vd = oldvdevs[i]; 1547 if (vd != NULL) { 1548 ASSERT(vd->vdev_isl2cache); 1549 1550 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 1551 pool != 0ULL && l2arc_vdev_present(vd)) 1552 l2arc_remove_vdev(vd); 1553 vdev_clear_stats(vd); 1554 vdev_free(vd); 1555 } 1556 } 1557 1558 if (oldvdevs) 1559 kmem_free(oldvdevs, oldnvdevs * sizeof (void *)); 1560 1561 if (sav->sav_config == NULL) 1562 goto out; 1563 1564 sav->sav_vdevs = newvdevs; 1565 sav->sav_count = (int)nl2cache; 1566 1567 /* 1568 * Recompute the stashed list of l2cache devices, with status 1569 * information this time. 1570 */ 1571 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE, 1572 DATA_TYPE_NVLIST_ARRAY) == 0); 1573 1574 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 1575 for (i = 0; i < sav->sav_count; i++) 1576 l2cache[i] = vdev_config_generate(spa, 1577 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE); 1578 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 1579 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0); 1580out: 1581 for (i = 0; i < sav->sav_count; i++) 1582 nvlist_free(l2cache[i]); 1583 if (sav->sav_count) 1584 kmem_free(l2cache, sav->sav_count * sizeof (void *)); 1585} 1586 1587static int 1588load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value) 1589{ 1590 dmu_buf_t *db; 1591 char *packed = NULL; 1592 size_t nvsize = 0; 1593 int error; 1594 *value = NULL; 1595 1596 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 1597 nvsize = *(uint64_t *)db->db_data; 1598 dmu_buf_rele(db, FTAG); 1599 1600 packed = kmem_alloc(nvsize, KM_SLEEP); 1601 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed, 1602 DMU_READ_PREFETCH); 1603 if (error == 0) 1604 error = nvlist_unpack(packed, nvsize, value, 0); 1605 kmem_free(packed, nvsize); 1606 1607 return (error); 1608} 1609 1610/* 1611 * Checks to see if the given vdev could not be opened, in which case we post a 1612 * sysevent to notify the autoreplace code that the device has been removed. 1613 */ 1614static void 1615spa_check_removed(vdev_t *vd) 1616{ 1617 for (int c = 0; c < vd->vdev_children; c++) 1618 spa_check_removed(vd->vdev_child[c]); 1619 1620 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) && 1621 !vd->vdev_ishole) { 1622 zfs_post_autoreplace(vd->vdev_spa, vd); 1623 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK); 1624 } 1625} 1626 1627/* 1628 * Validate the current config against the MOS config 1629 */ 1630static boolean_t 1631spa_config_valid(spa_t *spa, nvlist_t *config) 1632{ 1633 vdev_t *mrvd, *rvd = spa->spa_root_vdev; 1634 nvlist_t *nv; 1635 1636 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0); 1637 1638 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1639 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0); 1640 1641 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children); 1642 1643 /* 1644 * If we're doing a normal import, then build up any additional 1645 * diagnostic information about missing devices in this config. 1646 * We'll pass this up to the user for further processing. 1647 */ 1648 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) { 1649 nvlist_t **child, *nv; 1650 uint64_t idx = 0; 1651 1652 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **), 1653 KM_SLEEP); 1654 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0); 1655 1656 for (int c = 0; c < rvd->vdev_children; c++) { 1657 vdev_t *tvd = rvd->vdev_child[c]; 1658 vdev_t *mtvd = mrvd->vdev_child[c]; 1659 1660 if (tvd->vdev_ops == &vdev_missing_ops && 1661 mtvd->vdev_ops != &vdev_missing_ops && 1662 mtvd->vdev_islog) 1663 child[idx++] = vdev_config_generate(spa, mtvd, 1664 B_FALSE, 0); 1665 } 1666 1667 if (idx) { 1668 VERIFY(nvlist_add_nvlist_array(nv, 1669 ZPOOL_CONFIG_CHILDREN, child, idx) == 0); 1670 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 1671 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0); 1672 1673 for (int i = 0; i < idx; i++) 1674 nvlist_free(child[i]); 1675 } 1676 nvlist_free(nv); 1677 kmem_free(child, rvd->vdev_children * sizeof (char **)); 1678 } 1679 1680 /* 1681 * Compare the root vdev tree with the information we have 1682 * from the MOS config (mrvd). Check each top-level vdev 1683 * with the corresponding MOS config top-level (mtvd). 1684 */ 1685 for (int c = 0; c < rvd->vdev_children; c++) { 1686 vdev_t *tvd = rvd->vdev_child[c]; 1687 vdev_t *mtvd = mrvd->vdev_child[c]; 1688 1689 /* 1690 * Resolve any "missing" vdevs in the current configuration. 1691 * If we find that the MOS config has more accurate information 1692 * about the top-level vdev then use that vdev instead. 1693 */ 1694 if (tvd->vdev_ops == &vdev_missing_ops && 1695 mtvd->vdev_ops != &vdev_missing_ops) { 1696 1697 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) 1698 continue; 1699 1700 /* 1701 * Device specific actions. 1702 */ 1703 if (mtvd->vdev_islog) { 1704 spa_set_log_state(spa, SPA_LOG_CLEAR); 1705 } else { 1706 /* 1707 * XXX - once we have 'readonly' pool 1708 * support we should be able to handle 1709 * missing data devices by transitioning 1710 * the pool to readonly. 1711 */ 1712 continue; 1713 } 1714 1715 /* 1716 * Swap the missing vdev with the data we were 1717 * able to obtain from the MOS config. 1718 */ 1719 vdev_remove_child(rvd, tvd); 1720 vdev_remove_child(mrvd, mtvd); 1721 1722 vdev_add_child(rvd, mtvd); 1723 vdev_add_child(mrvd, tvd); 1724 1725 spa_config_exit(spa, SCL_ALL, FTAG); 1726 vdev_load(mtvd); 1727 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 1728 1729 vdev_reopen(rvd); 1730 } else if (mtvd->vdev_islog) { 1731 /* 1732 * Load the slog device's state from the MOS config 1733 * since it's possible that the label does not 1734 * contain the most up-to-date information. 1735 */ 1736 vdev_load_log_state(tvd, mtvd); 1737 vdev_reopen(tvd); 1738 } 1739 } 1740 vdev_free(mrvd); 1741 spa_config_exit(spa, SCL_ALL, FTAG); 1742 1743 /* 1744 * Ensure we were able to validate the config. 1745 */ 1746 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum); 1747} 1748 1749/* 1750 * Check for missing log devices 1751 */ 1752static boolean_t 1753spa_check_logs(spa_t *spa) 1754{ 1755 boolean_t rv = B_FALSE; 1756 1757 switch (spa->spa_log_state) { 1758 case SPA_LOG_MISSING: 1759 /* need to recheck in case slog has been restored */ 1760 case SPA_LOG_UNKNOWN: 1761 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain, 1762 NULL, DS_FIND_CHILDREN) != 0); 1763 if (rv) 1764 spa_set_log_state(spa, SPA_LOG_MISSING); 1765 break; 1766 } 1767 return (rv); 1768} 1769 1770static boolean_t 1771spa_passivate_log(spa_t *spa) 1772{ 1773 vdev_t *rvd = spa->spa_root_vdev; 1774 boolean_t slog_found = B_FALSE; 1775 1776 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1777 1778 if (!spa_has_slogs(spa)) 1779 return (B_FALSE); 1780 1781 for (int c = 0; c < rvd->vdev_children; c++) { 1782 vdev_t *tvd = rvd->vdev_child[c]; 1783 metaslab_group_t *mg = tvd->vdev_mg; 1784 1785 if (tvd->vdev_islog) { 1786 metaslab_group_passivate(mg); 1787 slog_found = B_TRUE; 1788 } 1789 } 1790 1791 return (slog_found); 1792} 1793 1794static void 1795spa_activate_log(spa_t *spa) 1796{ 1797 vdev_t *rvd = spa->spa_root_vdev; 1798 1799 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 1800 1801 for (int c = 0; c < rvd->vdev_children; c++) { 1802 vdev_t *tvd = rvd->vdev_child[c]; 1803 metaslab_group_t *mg = tvd->vdev_mg; 1804 1805 if (tvd->vdev_islog) 1806 metaslab_group_activate(mg); 1807 } 1808} 1809 1810int 1811spa_offline_log(spa_t *spa) 1812{ 1813 int error; 1814 1815 error = dmu_objset_find(spa_name(spa), zil_vdev_offline, 1816 NULL, DS_FIND_CHILDREN); 1817 if (error == 0) { 1818 /* 1819 * We successfully offlined the log device, sync out the 1820 * current txg so that the "stubby" block can be removed 1821 * by zil_sync(). 1822 */ 1823 txg_wait_synced(spa->spa_dsl_pool, 0); 1824 } 1825 return (error); 1826} 1827 1828static void 1829spa_aux_check_removed(spa_aux_vdev_t *sav) 1830{ 1831 int i; 1832 1833 for (i = 0; i < sav->sav_count; i++) 1834 spa_check_removed(sav->sav_vdevs[i]); 1835} 1836 1837void 1838spa_claim_notify(zio_t *zio) 1839{ 1840 spa_t *spa = zio->io_spa; 1841 1842 if (zio->io_error) 1843 return; 1844 1845 mutex_enter(&spa->spa_props_lock); /* any mutex will do */ 1846 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth) 1847 spa->spa_claim_max_txg = zio->io_bp->blk_birth; 1848 mutex_exit(&spa->spa_props_lock); 1849} 1850 1851typedef struct spa_load_error { 1852 uint64_t sle_meta_count; 1853 uint64_t sle_data_count; 1854} spa_load_error_t; 1855 1856static void 1857spa_load_verify_done(zio_t *zio) 1858{ 1859 blkptr_t *bp = zio->io_bp; 1860 spa_load_error_t *sle = zio->io_private; 1861 dmu_object_type_t type = BP_GET_TYPE(bp); 1862 int error = zio->io_error; 1863 1864 if (error) { 1865 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) && 1866 type != DMU_OT_INTENT_LOG) 1867 atomic_add_64(&sle->sle_meta_count, 1); 1868 else 1869 atomic_add_64(&sle->sle_data_count, 1); 1870 } 1871 zio_data_buf_free(zio->io_data, zio->io_size); 1872} 1873 1874/*ARGSUSED*/ 1875static int 1876spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, 1877 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) 1878{ 1879 if (bp != NULL) { 1880 zio_t *rio = arg; 1881 size_t size = BP_GET_PSIZE(bp); 1882 void *data = zio_data_buf_alloc(size); 1883 1884 zio_nowait(zio_read(rio, spa, bp, data, size, 1885 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB, 1886 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL | 1887 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb)); 1888 } 1889 return (0); 1890} 1891 1892static int 1893spa_load_verify(spa_t *spa) 1894{ 1895 zio_t *rio; 1896 spa_load_error_t sle = { 0 }; 1897 zpool_rewind_policy_t policy; 1898 boolean_t verify_ok = B_FALSE; 1899 int error; 1900 1901 zpool_get_rewind_policy(spa->spa_config, &policy); 1902 1903 if (policy.zrp_request & ZPOOL_NEVER_REWIND) 1904 return (0); 1905 1906 rio = zio_root(spa, NULL, &sle, 1907 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); 1908 1909 error = traverse_pool(spa, spa->spa_verify_min_txg, 1910 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio); 1911 1912 (void) zio_wait(rio); 1913 1914 spa->spa_load_meta_errors = sle.sle_meta_count; 1915 spa->spa_load_data_errors = sle.sle_data_count; 1916 1917 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta && 1918 sle.sle_data_count <= policy.zrp_maxdata) { 1919 int64_t loss = 0; 1920 1921 verify_ok = B_TRUE; 1922 spa->spa_load_txg = spa->spa_uberblock.ub_txg; 1923 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp; 1924 1925 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts; 1926 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1927 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0); 1928 VERIFY(nvlist_add_int64(spa->spa_load_info, 1929 ZPOOL_CONFIG_REWIND_TIME, loss) == 0); 1930 VERIFY(nvlist_add_uint64(spa->spa_load_info, 1931 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0); 1932 } else { 1933 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg; 1934 } 1935 1936 if (error) { 1937 if (error != ENXIO && error != EIO) 1938 error = SET_ERROR(EIO); 1939 return (error); 1940 } 1941 1942 return (verify_ok ? 0 : EIO); 1943} 1944 1945/* 1946 * Find a value in the pool props object. 1947 */ 1948static void 1949spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val) 1950{ 1951 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object, 1952 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val); 1953} 1954 1955/* 1956 * Find a value in the pool directory object. 1957 */ 1958static int 1959spa_dir_prop(spa_t *spa, const char *name, uint64_t *val) 1960{ 1961 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, 1962 name, sizeof (uint64_t), 1, val)); 1963} 1964 1965static int 1966spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err) 1967{ 1968 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux); 1969 return (err); 1970} 1971 1972/* 1973 * Fix up config after a partly-completed split. This is done with the 1974 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off 1975 * pool have that entry in their config, but only the splitting one contains 1976 * a list of all the guids of the vdevs that are being split off. 1977 * 1978 * This function determines what to do with that list: either rejoin 1979 * all the disks to the pool, or complete the splitting process. To attempt 1980 * the rejoin, each disk that is offlined is marked online again, and 1981 * we do a reopen() call. If the vdev label for every disk that was 1982 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL) 1983 * then we call vdev_split() on each disk, and complete the split. 1984 * 1985 * Otherwise we leave the config alone, with all the vdevs in place in 1986 * the original pool. 1987 */ 1988static void 1989spa_try_repair(spa_t *spa, nvlist_t *config) 1990{ 1991 uint_t extracted; 1992 uint64_t *glist; 1993 uint_t i, gcount; 1994 nvlist_t *nvl; 1995 vdev_t **vd; 1996 boolean_t attempt_reopen; 1997 1998 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0) 1999 return; 2000 2001 /* check that the config is complete */ 2002 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 2003 &glist, &gcount) != 0) 2004 return; 2005 2006 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP); 2007 2008 /* attempt to online all the vdevs & validate */ 2009 attempt_reopen = B_TRUE; 2010 for (i = 0; i < gcount; i++) { 2011 if (glist[i] == 0) /* vdev is hole */ 2012 continue; 2013 2014 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE); 2015 if (vd[i] == NULL) { 2016 /* 2017 * Don't bother attempting to reopen the disks; 2018 * just do the split. 2019 */ 2020 attempt_reopen = B_FALSE; 2021 } else { 2022 /* attempt to re-online it */ 2023 vd[i]->vdev_offline = B_FALSE; 2024 } 2025 } 2026 2027 if (attempt_reopen) { 2028 vdev_reopen(spa->spa_root_vdev); 2029 2030 /* check each device to see what state it's in */ 2031 for (extracted = 0, i = 0; i < gcount; i++) { 2032 if (vd[i] != NULL && 2033 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL) 2034 break; 2035 ++extracted; 2036 } 2037 } 2038 2039 /* 2040 * If every disk has been moved to the new pool, or if we never 2041 * even attempted to look at them, then we split them off for 2042 * good. 2043 */ 2044 if (!attempt_reopen || gcount == extracted) { 2045 for (i = 0; i < gcount; i++) 2046 if (vd[i] != NULL) 2047 vdev_split(vd[i]); 2048 vdev_reopen(spa->spa_root_vdev); 2049 } 2050 2051 kmem_free(vd, gcount * sizeof (vdev_t *)); 2052} 2053 2054static int 2055spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type, 2056 boolean_t mosconfig) 2057{ 2058 nvlist_t *config = spa->spa_config; 2059 char *ereport = FM_EREPORT_ZFS_POOL; 2060 char *comment; 2061 int error; 2062 uint64_t pool_guid; 2063 nvlist_t *nvl; 2064 2065 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) 2066 return (SET_ERROR(EINVAL)); 2067 2068 ASSERT(spa->spa_comment == NULL); 2069 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0) 2070 spa->spa_comment = spa_strdup(comment); 2071 2072 /* 2073 * Versioning wasn't explicitly added to the label until later, so if 2074 * it's not present treat it as the initial version. 2075 */ 2076 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 2077 &spa->spa_ubsync.ub_version) != 0) 2078 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 2079 2080 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, 2081 &spa->spa_config_txg); 2082 2083 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) && 2084 spa_guid_exists(pool_guid, 0)) { 2085 error = SET_ERROR(EEXIST); 2086 } else { 2087 spa->spa_config_guid = pool_guid; 2088 2089 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, 2090 &nvl) == 0) { 2091 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting, 2092 KM_SLEEP) == 0); 2093 } 2094 2095 nvlist_free(spa->spa_load_info); 2096 spa->spa_load_info = fnvlist_alloc(); 2097 2098 gethrestime(&spa->spa_loaded_ts); 2099 error = spa_load_impl(spa, pool_guid, config, state, type, 2100 mosconfig, &ereport); 2101 } 2102 2103 spa->spa_minref = refcount_count(&spa->spa_refcount); 2104 if (error) { 2105 if (error != EEXIST) { 2106 spa->spa_loaded_ts.tv_sec = 0; 2107 spa->spa_loaded_ts.tv_nsec = 0; 2108 } 2109 if (error != EBADF) { 2110 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0); 2111 } 2112 } 2113 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE; 2114 spa->spa_ena = 0; 2115 2116 return (error); 2117} 2118 2119/* 2120 * Load an existing storage pool, using the pool's builtin spa_config as a 2121 * source of configuration information. 2122 */ 2123static int 2124spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config, 2125 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig, 2126 char **ereport) 2127{ 2128 int error = 0; 2129 nvlist_t *nvroot = NULL; 2130 nvlist_t *label; 2131 vdev_t *rvd; 2132 uberblock_t *ub = &spa->spa_uberblock; 2133 uint64_t children, config_cache_txg = spa->spa_config_txg; 2134 int orig_mode = spa->spa_mode; 2135 int parse; 2136 uint64_t obj; 2137 boolean_t missing_feat_write = B_FALSE; 2138 2139 /* 2140 * If this is an untrusted config, access the pool in read-only mode. 2141 * This prevents things like resilvering recently removed devices. 2142 */ 2143 if (!mosconfig) 2144 spa->spa_mode = FREAD; 2145 2146 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 2147 2148 spa->spa_load_state = state; 2149 2150 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot)) 2151 return (SET_ERROR(EINVAL)); 2152 2153 parse = (type == SPA_IMPORT_EXISTING ? 2154 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT); 2155 2156 /* 2157 * Create "The Godfather" zio to hold all async IOs 2158 */ 2159 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 2160 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 2161 2162 /* 2163 * Parse the configuration into a vdev tree. We explicitly set the 2164 * value that will be returned by spa_version() since parsing the 2165 * configuration requires knowing the version number. 2166 */ 2167 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2168 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse); 2169 spa_config_exit(spa, SCL_ALL, FTAG); 2170 2171 if (error != 0) 2172 return (error); 2173 2174 ASSERT(spa->spa_root_vdev == rvd); 2175 2176 if (type != SPA_IMPORT_ASSEMBLE) { 2177 ASSERT(spa_guid(spa) == pool_guid); 2178 } 2179 2180 /* 2181 * Try to open all vdevs, loading each label in the process. 2182 */ 2183 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2184 error = vdev_open(rvd); 2185 spa_config_exit(spa, SCL_ALL, FTAG); 2186 if (error != 0) 2187 return (error); 2188 2189 /* 2190 * We need to validate the vdev labels against the configuration that 2191 * we have in hand, which is dependent on the setting of mosconfig. If 2192 * mosconfig is true then we're validating the vdev labels based on 2193 * that config. Otherwise, we're validating against the cached config 2194 * (zpool.cache) that was read when we loaded the zfs module, and then 2195 * later we will recursively call spa_load() and validate against 2196 * the vdev config. 2197 * 2198 * If we're assembling a new pool that's been split off from an 2199 * existing pool, the labels haven't yet been updated so we skip 2200 * validation for now. 2201 */ 2202 if (type != SPA_IMPORT_ASSEMBLE) { 2203 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2204 error = vdev_validate(rvd, mosconfig); 2205 spa_config_exit(spa, SCL_ALL, FTAG); 2206 2207 if (error != 0) 2208 return (error); 2209 2210 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2211 return (SET_ERROR(ENXIO)); 2212 } 2213 2214 /* 2215 * Find the best uberblock. 2216 */ 2217 vdev_uberblock_load(rvd, ub, &label); 2218 2219 /* 2220 * If we weren't able to find a single valid uberblock, return failure. 2221 */ 2222 if (ub->ub_txg == 0) { 2223 nvlist_free(label); 2224 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO)); 2225 } 2226 2227 /* 2228 * If the pool has an unsupported version we can't open it. 2229 */ 2230 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) { 2231 nvlist_free(label); 2232 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP)); 2233 } 2234 2235 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2236 nvlist_t *features; 2237 2238 /* 2239 * If we weren't able to find what's necessary for reading the 2240 * MOS in the label, return failure. 2241 */ 2242 if (label == NULL || nvlist_lookup_nvlist(label, 2243 ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) { 2244 nvlist_free(label); 2245 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2246 ENXIO)); 2247 } 2248 2249 /* 2250 * Update our in-core representation with the definitive values 2251 * from the label. 2252 */ 2253 nvlist_free(spa->spa_label_features); 2254 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0); 2255 } 2256 2257 nvlist_free(label); 2258 2259 /* 2260 * Look through entries in the label nvlist's features_for_read. If 2261 * there is a feature listed there which we don't understand then we 2262 * cannot open a pool. 2263 */ 2264 if (ub->ub_version >= SPA_VERSION_FEATURES) { 2265 nvlist_t *unsup_feat; 2266 2267 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) == 2268 0); 2269 2270 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features, 2271 NULL); nvp != NULL; 2272 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) { 2273 if (!zfeature_is_supported(nvpair_name(nvp))) { 2274 VERIFY(nvlist_add_string(unsup_feat, 2275 nvpair_name(nvp), "") == 0); 2276 } 2277 } 2278 2279 if (!nvlist_empty(unsup_feat)) { 2280 VERIFY(nvlist_add_nvlist(spa->spa_load_info, 2281 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0); 2282 nvlist_free(unsup_feat); 2283 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2284 ENOTSUP)); 2285 } 2286 2287 nvlist_free(unsup_feat); 2288 } 2289 2290 /* 2291 * If the vdev guid sum doesn't match the uberblock, we have an 2292 * incomplete configuration. We first check to see if the pool 2293 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN). 2294 * If it is, defer the vdev_guid_sum check till later so we 2295 * can handle missing vdevs. 2296 */ 2297 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN, 2298 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE && 2299 rvd->vdev_guid_sum != ub->ub_guid_sum) 2300 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO)); 2301 2302 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) { 2303 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2304 spa_try_repair(spa, config); 2305 spa_config_exit(spa, SCL_ALL, FTAG); 2306 nvlist_free(spa->spa_config_splitting); 2307 spa->spa_config_splitting = NULL; 2308 } 2309 2310 /* 2311 * Initialize internal SPA structures. 2312 */ 2313 spa->spa_state = POOL_STATE_ACTIVE; 2314 spa->spa_ubsync = spa->spa_uberblock; 2315 spa->spa_verify_min_txg = spa->spa_extreme_rewind ? 2316 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1; 2317 spa->spa_first_txg = spa->spa_last_ubsync_txg ? 2318 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1; 2319 spa->spa_claim_max_txg = spa->spa_first_txg; 2320 spa->spa_prev_software_version = ub->ub_software_version; 2321 2322 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool); 2323 if (error) 2324 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2325 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset; 2326 2327 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0) 2328 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2329 2330 if (spa_version(spa) >= SPA_VERSION_FEATURES) { 2331 boolean_t missing_feat_read = B_FALSE; 2332 nvlist_t *unsup_feat, *enabled_feat; 2333 2334 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ, 2335 &spa->spa_feat_for_read_obj) != 0) { 2336 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2337 } 2338 2339 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE, 2340 &spa->spa_feat_for_write_obj) != 0) { 2341 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2342 } 2343 2344 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS, 2345 &spa->spa_feat_desc_obj) != 0) { 2346 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2347 } 2348 2349 enabled_feat = fnvlist_alloc(); 2350 unsup_feat = fnvlist_alloc(); 2351 2352 if (!feature_is_supported(spa->spa_meta_objset, 2353 spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj, 2354 unsup_feat, enabled_feat)) 2355 missing_feat_read = B_TRUE; 2356 2357 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) { 2358 if (!feature_is_supported(spa->spa_meta_objset, 2359 spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj, 2360 unsup_feat, enabled_feat)) { 2361 missing_feat_write = B_TRUE; 2362 } 2363 } 2364 2365 fnvlist_add_nvlist(spa->spa_load_info, 2366 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat); 2367 2368 if (!nvlist_empty(unsup_feat)) { 2369 fnvlist_add_nvlist(spa->spa_load_info, 2370 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat); 2371 } 2372 2373 fnvlist_free(enabled_feat); 2374 fnvlist_free(unsup_feat); 2375 2376 if (!missing_feat_read) { 2377 fnvlist_add_boolean(spa->spa_load_info, 2378 ZPOOL_CONFIG_CAN_RDONLY); 2379 } 2380 2381 /* 2382 * If the state is SPA_LOAD_TRYIMPORT, our objective is 2383 * twofold: to determine whether the pool is available for 2384 * import in read-write mode and (if it is not) whether the 2385 * pool is available for import in read-only mode. If the pool 2386 * is available for import in read-write mode, it is displayed 2387 * as available in userland; if it is not available for import 2388 * in read-only mode, it is displayed as unavailable in 2389 * userland. If the pool is available for import in read-only 2390 * mode but not read-write mode, it is displayed as unavailable 2391 * in userland with a special note that the pool is actually 2392 * available for open in read-only mode. 2393 * 2394 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are 2395 * missing a feature for write, we must first determine whether 2396 * the pool can be opened read-only before returning to 2397 * userland in order to know whether to display the 2398 * abovementioned note. 2399 */ 2400 if (missing_feat_read || (missing_feat_write && 2401 spa_writeable(spa))) { 2402 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, 2403 ENOTSUP)); 2404 } 2405 } 2406 2407 spa->spa_is_initializing = B_TRUE; 2408 error = dsl_pool_open(spa->spa_dsl_pool); 2409 spa->spa_is_initializing = B_FALSE; 2410 if (error != 0) 2411 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2412 2413 if (!mosconfig) { 2414 uint64_t hostid; 2415 nvlist_t *policy = NULL, *nvconfig; 2416 2417 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2418 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2419 2420 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig, 2421 ZPOOL_CONFIG_HOSTID, &hostid) == 0) { 2422 char *hostname; 2423 unsigned long myhostid = 0; 2424 2425 VERIFY(nvlist_lookup_string(nvconfig, 2426 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0); 2427 2428#ifdef _KERNEL 2429 myhostid = zone_get_hostid(NULL); 2430#else /* _KERNEL */ 2431 /* 2432 * We're emulating the system's hostid in userland, so 2433 * we can't use zone_get_hostid(). 2434 */ 2435 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid); 2436#endif /* _KERNEL */ 2437 if (check_hostid && hostid != 0 && myhostid != 0 && 2438 hostid != myhostid) { 2439 nvlist_free(nvconfig); 2440 cmn_err(CE_WARN, "pool '%s' could not be " 2441 "loaded as it was last accessed by " 2442 "another system (host: %s hostid: 0x%lx). " 2443 "See: http://illumos.org/msg/ZFS-8000-EY", 2444 spa_name(spa), hostname, 2445 (unsigned long)hostid); 2446 return (SET_ERROR(EBADF)); 2447 } 2448 } 2449 if (nvlist_lookup_nvlist(spa->spa_config, 2450 ZPOOL_REWIND_POLICY, &policy) == 0) 2451 VERIFY(nvlist_add_nvlist(nvconfig, 2452 ZPOOL_REWIND_POLICY, policy) == 0); 2453 2454 spa_config_set(spa, nvconfig); 2455 spa_unload(spa); 2456 spa_deactivate(spa); 2457 spa_activate(spa, orig_mode); 2458 2459 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE)); 2460 } 2461 2462 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0) 2463 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2464 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj); 2465 if (error != 0) 2466 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2467 2468 /* 2469 * Load the bit that tells us to use the new accounting function 2470 * (raid-z deflation). If we have an older pool, this will not 2471 * be present. 2472 */ 2473 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate); 2474 if (error != 0 && error != ENOENT) 2475 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2476 2477 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION, 2478 &spa->spa_creation_version); 2479 if (error != 0 && error != ENOENT) 2480 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2481 2482 /* 2483 * Load the persistent error log. If we have an older pool, this will 2484 * not be present. 2485 */ 2486 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last); 2487 if (error != 0 && error != ENOENT) 2488 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2489 2490 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB, 2491 &spa->spa_errlog_scrub); 2492 if (error != 0 && error != ENOENT) 2493 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2494 2495 /* 2496 * Load the history object. If we have an older pool, this 2497 * will not be present. 2498 */ 2499 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history); 2500 if (error != 0 && error != ENOENT) 2501 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2502 2503 /* 2504 * If we're assembling the pool from the split-off vdevs of 2505 * an existing pool, we don't want to attach the spares & cache 2506 * devices. 2507 */ 2508 2509 /* 2510 * Load any hot spares for this pool. 2511 */ 2512 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object); 2513 if (error != 0 && error != ENOENT) 2514 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2515 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2516 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES); 2517 if (load_nvlist(spa, spa->spa_spares.sav_object, 2518 &spa->spa_spares.sav_config) != 0) 2519 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2520 2521 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2522 spa_load_spares(spa); 2523 spa_config_exit(spa, SCL_ALL, FTAG); 2524 } else if (error == 0) { 2525 spa->spa_spares.sav_sync = B_TRUE; 2526 } 2527 2528 /* 2529 * Load any level 2 ARC devices for this pool. 2530 */ 2531 error = spa_dir_prop(spa, DMU_POOL_L2CACHE, 2532 &spa->spa_l2cache.sav_object); 2533 if (error != 0 && error != ENOENT) 2534 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2535 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) { 2536 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE); 2537 if (load_nvlist(spa, spa->spa_l2cache.sav_object, 2538 &spa->spa_l2cache.sav_config) != 0) 2539 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2540 2541 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2542 spa_load_l2cache(spa); 2543 spa_config_exit(spa, SCL_ALL, FTAG); 2544 } else if (error == 0) { 2545 spa->spa_l2cache.sav_sync = B_TRUE; 2546 } 2547 2548 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 2549 2550 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object); 2551 if (error && error != ENOENT) 2552 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2553 2554 if (error == 0) { 2555 uint64_t autoreplace; 2556 2557 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs); 2558 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace); 2559 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation); 2560 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode); 2561 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand); 2562 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO, 2563 &spa->spa_dedup_ditto); 2564 2565 spa->spa_autoreplace = (autoreplace != 0); 2566 } 2567 2568 /* 2569 * If the 'autoreplace' property is set, then post a resource notifying 2570 * the ZFS DE that it should not issue any faults for unopenable 2571 * devices. We also iterate over the vdevs, and post a sysevent for any 2572 * unopenable vdevs so that the normal autoreplace handler can take 2573 * over. 2574 */ 2575 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) { 2576 spa_check_removed(spa->spa_root_vdev); 2577 /* 2578 * For the import case, this is done in spa_import(), because 2579 * at this point we're using the spare definitions from 2580 * the MOS config, not necessarily from the userland config. 2581 */ 2582 if (state != SPA_LOAD_IMPORT) { 2583 spa_aux_check_removed(&spa->spa_spares); 2584 spa_aux_check_removed(&spa->spa_l2cache); 2585 } 2586 } 2587 2588 /* 2589 * Load the vdev state for all toplevel vdevs. 2590 */ 2591 vdev_load(rvd); 2592 2593 /* 2594 * Propagate the leaf DTLs we just loaded all the way up the tree. 2595 */ 2596 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 2597 vdev_dtl_reassess(rvd, 0, 0, B_FALSE); 2598 spa_config_exit(spa, SCL_ALL, FTAG); 2599 2600 /* 2601 * Load the DDTs (dedup tables). 2602 */ 2603 error = ddt_load(spa); 2604 if (error != 0) 2605 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2606 2607 spa_update_dspace(spa); 2608 2609 /* 2610 * Validate the config, using the MOS config to fill in any 2611 * information which might be missing. If we fail to validate 2612 * the config then declare the pool unfit for use. If we're 2613 * assembling a pool from a split, the log is not transferred 2614 * over. 2615 */ 2616 if (type != SPA_IMPORT_ASSEMBLE) { 2617 nvlist_t *nvconfig; 2618 2619 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0) 2620 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO)); 2621 2622 if (!spa_config_valid(spa, nvconfig)) { 2623 nvlist_free(nvconfig); 2624 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, 2625 ENXIO)); 2626 } 2627 nvlist_free(nvconfig); 2628 2629 /* 2630 * Now that we've validated the config, check the state of the 2631 * root vdev. If it can't be opened, it indicates one or 2632 * more toplevel vdevs are faulted. 2633 */ 2634 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) 2635 return (SET_ERROR(ENXIO)); 2636 2637 if (spa_check_logs(spa)) { 2638 *ereport = FM_EREPORT_ZFS_LOG_REPLAY; 2639 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO)); 2640 } 2641 } 2642 2643 if (missing_feat_write) { 2644 ASSERT(state == SPA_LOAD_TRYIMPORT); 2645 2646 /* 2647 * At this point, we know that we can open the pool in 2648 * read-only mode but not read-write mode. We now have enough 2649 * information and can return to userland. 2650 */ 2651 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP)); 2652 } 2653 2654 /* 2655 * We've successfully opened the pool, verify that we're ready 2656 * to start pushing transactions. 2657 */ 2658 if (state != SPA_LOAD_TRYIMPORT) { 2659 if (error = spa_load_verify(spa)) 2660 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, 2661 error)); 2662 } 2663 2664 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER || 2665 spa->spa_load_max_txg == UINT64_MAX)) { 2666 dmu_tx_t *tx; 2667 int need_update = B_FALSE; 2668 2669 ASSERT(state != SPA_LOAD_TRYIMPORT); 2670 2671 /* 2672 * Claim log blocks that haven't been committed yet. 2673 * This must all happen in a single txg. 2674 * Note: spa_claim_max_txg is updated by spa_claim_notify(), 2675 * invoked from zil_claim_log_block()'s i/o done callback. 2676 * Price of rollback is that we abandon the log. 2677 */ 2678 spa->spa_claiming = B_TRUE; 2679 2680 tx = dmu_tx_create_assigned(spa_get_dsl(spa), 2681 spa_first_txg(spa)); 2682 (void) dmu_objset_find(spa_name(spa), 2683 zil_claim, tx, DS_FIND_CHILDREN); 2684 dmu_tx_commit(tx); 2685 2686 spa->spa_claiming = B_FALSE; 2687 2688 spa_set_log_state(spa, SPA_LOG_GOOD); 2689 spa->spa_sync_on = B_TRUE; 2690 txg_sync_start(spa->spa_dsl_pool); 2691 2692 /* 2693 * Wait for all claims to sync. We sync up to the highest 2694 * claimed log block birth time so that claimed log blocks 2695 * don't appear to be from the future. spa_claim_max_txg 2696 * will have been set for us by either zil_check_log_chain() 2697 * (invoked from spa_check_logs()) or zil_claim() above. 2698 */ 2699 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg); 2700 2701 /* 2702 * If the config cache is stale, or we have uninitialized 2703 * metaslabs (see spa_vdev_add()), then update the config. 2704 * 2705 * If this is a verbatim import, trust the current 2706 * in-core spa_config and update the disk labels. 2707 */ 2708 if (config_cache_txg != spa->spa_config_txg || 2709 state == SPA_LOAD_IMPORT || 2710 state == SPA_LOAD_RECOVER || 2711 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM)) 2712 need_update = B_TRUE; 2713 2714 for (int c = 0; c < rvd->vdev_children; c++) 2715 if (rvd->vdev_child[c]->vdev_ms_array == 0) 2716 need_update = B_TRUE; 2717 2718 /* 2719 * Update the config cache asychronously in case we're the 2720 * root pool, in which case the config cache isn't writable yet. 2721 */ 2722 if (need_update) 2723 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2724 2725 /* 2726 * Check all DTLs to see if anything needs resilvering. 2727 */ 2728 if (!dsl_scan_resilvering(spa->spa_dsl_pool) && 2729 vdev_resilver_needed(rvd, NULL, NULL)) 2730 spa_async_request(spa, SPA_ASYNC_RESILVER); 2731 2732 /* 2733 * Log the fact that we booted up (so that we can detect if 2734 * we rebooted in the middle of an operation). 2735 */ 2736 spa_history_log_version(spa, "open"); 2737 2738 /* 2739 * Delete any inconsistent datasets. 2740 */ 2741 (void) dmu_objset_find(spa_name(spa), 2742 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN); 2743 2744 /* 2745 * Clean up any stale temporary dataset userrefs. 2746 */ 2747 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool); 2748 } 2749 2750 return (0); 2751} 2752 2753static int 2754spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig) 2755{ 2756 int mode = spa->spa_mode; 2757 2758 spa_unload(spa); 2759 spa_deactivate(spa); 2760 2761 spa->spa_load_max_txg--; 2762 2763 spa_activate(spa, mode); 2764 spa_async_suspend(spa); 2765 2766 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig)); 2767} 2768 2769/* 2770 * If spa_load() fails this function will try loading prior txg's. If 2771 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool 2772 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this 2773 * function will not rewind the pool and will return the same error as 2774 * spa_load(). 2775 */ 2776static int 2777spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig, 2778 uint64_t max_request, int rewind_flags) 2779{ 2780 nvlist_t *loadinfo = NULL; 2781 nvlist_t *config = NULL; 2782 int load_error, rewind_error; 2783 uint64_t safe_rewind_txg; 2784 uint64_t min_txg; 2785 2786 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) { 2787 spa->spa_load_max_txg = spa->spa_load_txg; 2788 spa_set_log_state(spa, SPA_LOG_CLEAR); 2789 } else { 2790 spa->spa_load_max_txg = max_request; 2791 } 2792 2793 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING, 2794 mosconfig); 2795 if (load_error == 0) 2796 return (0); 2797 2798 if (spa->spa_root_vdev != NULL) 2799 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2800 2801 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg; 2802 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp; 2803 2804 if (rewind_flags & ZPOOL_NEVER_REWIND) { 2805 nvlist_free(config); 2806 return (load_error); 2807 } 2808 2809 if (state == SPA_LOAD_RECOVER) { 2810 /* Price of rolling back is discarding txgs, including log */ 2811 spa_set_log_state(spa, SPA_LOG_CLEAR); 2812 } else { 2813 /* 2814 * If we aren't rolling back save the load info from our first 2815 * import attempt so that we can restore it after attempting 2816 * to rewind. 2817 */ 2818 loadinfo = spa->spa_load_info; 2819 spa->spa_load_info = fnvlist_alloc(); 2820 } 2821 2822 spa->spa_load_max_txg = spa->spa_last_ubsync_txg; 2823 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE; 2824 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ? 2825 TXG_INITIAL : safe_rewind_txg; 2826 2827 /* 2828 * Continue as long as we're finding errors, we're still within 2829 * the acceptable rewind range, and we're still finding uberblocks 2830 */ 2831 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg && 2832 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) { 2833 if (spa->spa_load_max_txg < safe_rewind_txg) 2834 spa->spa_extreme_rewind = B_TRUE; 2835 rewind_error = spa_load_retry(spa, state, mosconfig); 2836 } 2837 2838 spa->spa_extreme_rewind = B_FALSE; 2839 spa->spa_load_max_txg = UINT64_MAX; 2840 2841 if (config && (rewind_error || state != SPA_LOAD_RECOVER)) 2842 spa_config_set(spa, config); 2843 2844 if (state == SPA_LOAD_RECOVER) { 2845 ASSERT3P(loadinfo, ==, NULL); 2846 return (rewind_error); 2847 } else { 2848 /* Store the rewind info as part of the initial load info */ 2849 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO, 2850 spa->spa_load_info); 2851 2852 /* Restore the initial load info */ 2853 fnvlist_free(spa->spa_load_info); 2854 spa->spa_load_info = loadinfo; 2855 2856 return (load_error); 2857 } 2858} 2859 2860/* 2861 * Pool Open/Import 2862 * 2863 * The import case is identical to an open except that the configuration is sent 2864 * down from userland, instead of grabbed from the configuration cache. For the 2865 * case of an open, the pool configuration will exist in the 2866 * POOL_STATE_UNINITIALIZED state. 2867 * 2868 * The stats information (gen/count/ustats) is used to gather vdev statistics at 2869 * the same time open the pool, without having to keep around the spa_t in some 2870 * ambiguous state. 2871 */ 2872static int 2873spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy, 2874 nvlist_t **config) 2875{ 2876 spa_t *spa; 2877 spa_load_state_t state = SPA_LOAD_OPEN; 2878 int error; 2879 int locked = B_FALSE; 2880 int firstopen = B_FALSE; 2881 2882 *spapp = NULL; 2883 2884 /* 2885 * As disgusting as this is, we need to support recursive calls to this 2886 * function because dsl_dir_open() is called during spa_load(), and ends 2887 * up calling spa_open() again. The real fix is to figure out how to 2888 * avoid dsl_dir_open() calling this in the first place. 2889 */ 2890 if (mutex_owner(&spa_namespace_lock) != curthread) { 2891 mutex_enter(&spa_namespace_lock); 2892 locked = B_TRUE; 2893 } 2894 2895 if ((spa = spa_lookup(pool)) == NULL) { 2896 if (locked) 2897 mutex_exit(&spa_namespace_lock); 2898 return (SET_ERROR(ENOENT)); 2899 } 2900 2901 if (spa->spa_state == POOL_STATE_UNINITIALIZED) { 2902 zpool_rewind_policy_t policy; 2903 2904 firstopen = B_TRUE; 2905 2906 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config, 2907 &policy); 2908 if (policy.zrp_request & ZPOOL_DO_REWIND) 2909 state = SPA_LOAD_RECOVER; 2910 2911 spa_activate(spa, spa_mode_global); 2912 2913 if (state != SPA_LOAD_RECOVER) 2914 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 2915 2916 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg, 2917 policy.zrp_request); 2918 2919 if (error == EBADF) { 2920 /* 2921 * If vdev_validate() returns failure (indicated by 2922 * EBADF), it indicates that one of the vdevs indicates 2923 * that the pool has been exported or destroyed. If 2924 * this is the case, the config cache is out of sync and 2925 * we should remove the pool from the namespace. 2926 */ 2927 spa_unload(spa); 2928 spa_deactivate(spa); 2929 spa_config_sync(spa, B_TRUE, B_TRUE); 2930 spa_remove(spa); 2931 if (locked) 2932 mutex_exit(&spa_namespace_lock); 2933 return (SET_ERROR(ENOENT)); 2934 } 2935 2936 if (error) { 2937 /* 2938 * We can't open the pool, but we still have useful 2939 * information: the state of each vdev after the 2940 * attempted vdev_open(). Return this to the user. 2941 */ 2942 if (config != NULL && spa->spa_config) { 2943 VERIFY(nvlist_dup(spa->spa_config, config, 2944 KM_SLEEP) == 0); 2945 VERIFY(nvlist_add_nvlist(*config, 2946 ZPOOL_CONFIG_LOAD_INFO, 2947 spa->spa_load_info) == 0); 2948 } 2949 spa_unload(spa); 2950 spa_deactivate(spa); 2951 spa->spa_last_open_failed = error; 2952 if (locked) 2953 mutex_exit(&spa_namespace_lock); 2954 *spapp = NULL; 2955 return (error); 2956 } 2957 } 2958 2959 spa_open_ref(spa, tag); 2960 2961 if (config != NULL) 2962 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 2963 2964 /* 2965 * If we've recovered the pool, pass back any information we 2966 * gathered while doing the load. 2967 */ 2968 if (state == SPA_LOAD_RECOVER) { 2969 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO, 2970 spa->spa_load_info) == 0); 2971 } 2972 2973 if (locked) { 2974 spa->spa_last_open_failed = 0; 2975 spa->spa_last_ubsync_txg = 0; 2976 spa->spa_load_txg = 0; 2977 mutex_exit(&spa_namespace_lock); 2978#ifdef __FreeBSD__ 2979#ifdef _KERNEL 2980 if (firstopen) 2981 zvol_create_minors(spa->spa_name); 2982#endif 2983#endif 2984 } 2985 2986 *spapp = spa; 2987 2988 return (0); 2989} 2990 2991int 2992spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy, 2993 nvlist_t **config) 2994{ 2995 return (spa_open_common(name, spapp, tag, policy, config)); 2996} 2997 2998int 2999spa_open(const char *name, spa_t **spapp, void *tag) 3000{ 3001 return (spa_open_common(name, spapp, tag, NULL, NULL)); 3002} 3003 3004/* 3005 * Lookup the given spa_t, incrementing the inject count in the process, 3006 * preventing it from being exported or destroyed. 3007 */ 3008spa_t * 3009spa_inject_addref(char *name) 3010{ 3011 spa_t *spa; 3012 3013 mutex_enter(&spa_namespace_lock); 3014 if ((spa = spa_lookup(name)) == NULL) { 3015 mutex_exit(&spa_namespace_lock); 3016 return (NULL); 3017 } 3018 spa->spa_inject_ref++; 3019 mutex_exit(&spa_namespace_lock); 3020 3021 return (spa); 3022} 3023 3024void 3025spa_inject_delref(spa_t *spa) 3026{ 3027 mutex_enter(&spa_namespace_lock); 3028 spa->spa_inject_ref--; 3029 mutex_exit(&spa_namespace_lock); 3030} 3031 3032/* 3033 * Add spares device information to the nvlist. 3034 */ 3035static void 3036spa_add_spares(spa_t *spa, nvlist_t *config) 3037{ 3038 nvlist_t **spares; 3039 uint_t i, nspares; 3040 nvlist_t *nvroot; 3041 uint64_t guid; 3042 vdev_stat_t *vs; 3043 uint_t vsc; 3044 uint64_t pool; 3045 3046 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3047 3048 if (spa->spa_spares.sav_count == 0) 3049 return; 3050 3051 VERIFY(nvlist_lookup_nvlist(config, 3052 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3053 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 3054 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3055 if (nspares != 0) { 3056 VERIFY(nvlist_add_nvlist_array(nvroot, 3057 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3058 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3059 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0); 3060 3061 /* 3062 * Go through and find any spares which have since been 3063 * repurposed as an active spare. If this is the case, update 3064 * their status appropriately. 3065 */ 3066 for (i = 0; i < nspares; i++) { 3067 VERIFY(nvlist_lookup_uint64(spares[i], 3068 ZPOOL_CONFIG_GUID, &guid) == 0); 3069 if (spa_spare_exists(guid, &pool, NULL) && 3070 pool != 0ULL) { 3071 VERIFY(nvlist_lookup_uint64_array( 3072 spares[i], ZPOOL_CONFIG_VDEV_STATS, 3073 (uint64_t **)&vs, &vsc) == 0); 3074 vs->vs_state = VDEV_STATE_CANT_OPEN; 3075 vs->vs_aux = VDEV_AUX_SPARED; 3076 } 3077 } 3078 } 3079} 3080 3081/* 3082 * Add l2cache device information to the nvlist, including vdev stats. 3083 */ 3084static void 3085spa_add_l2cache(spa_t *spa, nvlist_t *config) 3086{ 3087 nvlist_t **l2cache; 3088 uint_t i, j, nl2cache; 3089 nvlist_t *nvroot; 3090 uint64_t guid; 3091 vdev_t *vd; 3092 vdev_stat_t *vs; 3093 uint_t vsc; 3094 3095 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3096 3097 if (spa->spa_l2cache.sav_count == 0) 3098 return; 3099 3100 VERIFY(nvlist_lookup_nvlist(config, 3101 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0); 3102 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 3103 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3104 if (nl2cache != 0) { 3105 VERIFY(nvlist_add_nvlist_array(nvroot, 3106 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3107 VERIFY(nvlist_lookup_nvlist_array(nvroot, 3108 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0); 3109 3110 /* 3111 * Update level 2 cache device stats. 3112 */ 3113 3114 for (i = 0; i < nl2cache; i++) { 3115 VERIFY(nvlist_lookup_uint64(l2cache[i], 3116 ZPOOL_CONFIG_GUID, &guid) == 0); 3117 3118 vd = NULL; 3119 for (j = 0; j < spa->spa_l2cache.sav_count; j++) { 3120 if (guid == 3121 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) { 3122 vd = spa->spa_l2cache.sav_vdevs[j]; 3123 break; 3124 } 3125 } 3126 ASSERT(vd != NULL); 3127 3128 VERIFY(nvlist_lookup_uint64_array(l2cache[i], 3129 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc) 3130 == 0); 3131 vdev_get_stats(vd, vs); 3132 } 3133 } 3134} 3135 3136static void 3137spa_add_feature_stats(spa_t *spa, nvlist_t *config) 3138{ 3139 nvlist_t *features; 3140 zap_cursor_t zc; 3141 zap_attribute_t za; 3142 3143 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER)); 3144 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3145 3146 /* We may be unable to read features if pool is suspended. */ 3147 if (spa_suspended(spa)) 3148 goto out; 3149 3150 if (spa->spa_feat_for_read_obj != 0) { 3151 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3152 spa->spa_feat_for_read_obj); 3153 zap_cursor_retrieve(&zc, &za) == 0; 3154 zap_cursor_advance(&zc)) { 3155 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3156 za.za_num_integers == 1); 3157 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3158 za.za_first_integer)); 3159 } 3160 zap_cursor_fini(&zc); 3161 } 3162 3163 if (spa->spa_feat_for_write_obj != 0) { 3164 for (zap_cursor_init(&zc, spa->spa_meta_objset, 3165 spa->spa_feat_for_write_obj); 3166 zap_cursor_retrieve(&zc, &za) == 0; 3167 zap_cursor_advance(&zc)) { 3168 ASSERT(za.za_integer_length == sizeof (uint64_t) && 3169 za.za_num_integers == 1); 3170 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name, 3171 za.za_first_integer)); 3172 } 3173 zap_cursor_fini(&zc); 3174 } 3175 3176out: 3177 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS, 3178 features) == 0); 3179 nvlist_free(features); 3180} 3181 3182int 3183spa_get_stats(const char *name, nvlist_t **config, 3184 char *altroot, size_t buflen) 3185{ 3186 int error; 3187 spa_t *spa; 3188 3189 *config = NULL; 3190 error = spa_open_common(name, &spa, FTAG, NULL, config); 3191 3192 if (spa != NULL) { 3193 /* 3194 * This still leaves a window of inconsistency where the spares 3195 * or l2cache devices could change and the config would be 3196 * self-inconsistent. 3197 */ 3198 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 3199 3200 if (*config != NULL) { 3201 uint64_t loadtimes[2]; 3202 3203 loadtimes[0] = spa->spa_loaded_ts.tv_sec; 3204 loadtimes[1] = spa->spa_loaded_ts.tv_nsec; 3205 VERIFY(nvlist_add_uint64_array(*config, 3206 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0); 3207 3208 VERIFY(nvlist_add_uint64(*config, 3209 ZPOOL_CONFIG_ERRCOUNT, 3210 spa_get_errlog_size(spa)) == 0); 3211 3212 if (spa_suspended(spa)) 3213 VERIFY(nvlist_add_uint64(*config, 3214 ZPOOL_CONFIG_SUSPENDED, 3215 spa->spa_failmode) == 0); 3216 3217 spa_add_spares(spa, *config); 3218 spa_add_l2cache(spa, *config); 3219 spa_add_feature_stats(spa, *config); 3220 } 3221 } 3222 3223 /* 3224 * We want to get the alternate root even for faulted pools, so we cheat 3225 * and call spa_lookup() directly. 3226 */ 3227 if (altroot) { 3228 if (spa == NULL) { 3229 mutex_enter(&spa_namespace_lock); 3230 spa = spa_lookup(name); 3231 if (spa) 3232 spa_altroot(spa, altroot, buflen); 3233 else 3234 altroot[0] = '\0'; 3235 spa = NULL; 3236 mutex_exit(&spa_namespace_lock); 3237 } else { 3238 spa_altroot(spa, altroot, buflen); 3239 } 3240 } 3241 3242 if (spa != NULL) { 3243 spa_config_exit(spa, SCL_CONFIG, FTAG); 3244 spa_close(spa, FTAG); 3245 } 3246 3247 return (error); 3248} 3249 3250/* 3251 * Validate that the auxiliary device array is well formed. We must have an 3252 * array of nvlists, each which describes a valid leaf vdev. If this is an 3253 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be 3254 * specified, as long as they are well-formed. 3255 */ 3256static int 3257spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode, 3258 spa_aux_vdev_t *sav, const char *config, uint64_t version, 3259 vdev_labeltype_t label) 3260{ 3261 nvlist_t **dev; 3262 uint_t i, ndev; 3263 vdev_t *vd; 3264 int error; 3265 3266 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3267 3268 /* 3269 * It's acceptable to have no devs specified. 3270 */ 3271 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0) 3272 return (0); 3273 3274 if (ndev == 0) 3275 return (SET_ERROR(EINVAL)); 3276 3277 /* 3278 * Make sure the pool is formatted with a version that supports this 3279 * device type. 3280 */ 3281 if (spa_version(spa) < version) 3282 return (SET_ERROR(ENOTSUP)); 3283 3284 /* 3285 * Set the pending device list so we correctly handle device in-use 3286 * checking. 3287 */ 3288 sav->sav_pending = dev; 3289 sav->sav_npending = ndev; 3290 3291 for (i = 0; i < ndev; i++) { 3292 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0, 3293 mode)) != 0) 3294 goto out; 3295 3296 if (!vd->vdev_ops->vdev_op_leaf) { 3297 vdev_free(vd); 3298 error = SET_ERROR(EINVAL); 3299 goto out; 3300 } 3301 3302 /* 3303 * The L2ARC currently only supports disk devices in 3304 * kernel context. For user-level testing, we allow it. 3305 */ 3306#ifdef _KERNEL 3307 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) && 3308 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) { 3309 error = SET_ERROR(ENOTBLK); 3310 vdev_free(vd); 3311 goto out; 3312 } 3313#endif 3314 vd->vdev_top = vd; 3315 3316 if ((error = vdev_open(vd)) == 0 && 3317 (error = vdev_label_init(vd, crtxg, label)) == 0) { 3318 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID, 3319 vd->vdev_guid) == 0); 3320 } 3321 3322 vdev_free(vd); 3323 3324 if (error && 3325 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE)) 3326 goto out; 3327 else 3328 error = 0; 3329 } 3330 3331out: 3332 sav->sav_pending = NULL; 3333 sav->sav_npending = 0; 3334 return (error); 3335} 3336 3337static int 3338spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode) 3339{ 3340 int error; 3341 3342 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3343 3344 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3345 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES, 3346 VDEV_LABEL_SPARE)) != 0) { 3347 return (error); 3348 } 3349 3350 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode, 3351 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE, 3352 VDEV_LABEL_L2CACHE)); 3353} 3354 3355static void 3356spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs, 3357 const char *config) 3358{ 3359 int i; 3360 3361 if (sav->sav_config != NULL) { 3362 nvlist_t **olddevs; 3363 uint_t oldndevs; 3364 nvlist_t **newdevs; 3365 3366 /* 3367 * Generate new dev list by concatentating with the 3368 * current dev list. 3369 */ 3370 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config, 3371 &olddevs, &oldndevs) == 0); 3372 3373 newdevs = kmem_alloc(sizeof (void *) * 3374 (ndevs + oldndevs), KM_SLEEP); 3375 for (i = 0; i < oldndevs; i++) 3376 VERIFY(nvlist_dup(olddevs[i], &newdevs[i], 3377 KM_SLEEP) == 0); 3378 for (i = 0; i < ndevs; i++) 3379 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs], 3380 KM_SLEEP) == 0); 3381 3382 VERIFY(nvlist_remove(sav->sav_config, config, 3383 DATA_TYPE_NVLIST_ARRAY) == 0); 3384 3385 VERIFY(nvlist_add_nvlist_array(sav->sav_config, 3386 config, newdevs, ndevs + oldndevs) == 0); 3387 for (i = 0; i < oldndevs + ndevs; i++) 3388 nvlist_free(newdevs[i]); 3389 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *)); 3390 } else { 3391 /* 3392 * Generate a new dev list. 3393 */ 3394 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME, 3395 KM_SLEEP) == 0); 3396 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config, 3397 devs, ndevs) == 0); 3398 } 3399} 3400 3401/* 3402 * Stop and drop level 2 ARC devices 3403 */ 3404void 3405spa_l2cache_drop(spa_t *spa) 3406{ 3407 vdev_t *vd; 3408 int i; 3409 spa_aux_vdev_t *sav = &spa->spa_l2cache; 3410 3411 for (i = 0; i < sav->sav_count; i++) { 3412 uint64_t pool; 3413 3414 vd = sav->sav_vdevs[i]; 3415 ASSERT(vd != NULL); 3416 3417 if (spa_l2cache_exists(vd->vdev_guid, &pool) && 3418 pool != 0ULL && l2arc_vdev_present(vd)) 3419 l2arc_remove_vdev(vd); 3420 } 3421} 3422 3423/* 3424 * Pool Creation 3425 */ 3426int 3427spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props, 3428 nvlist_t *zplprops) 3429{ 3430 spa_t *spa; 3431 char *altroot = NULL; 3432 vdev_t *rvd; 3433 dsl_pool_t *dp; 3434 dmu_tx_t *tx; 3435 int error = 0; 3436 uint64_t txg = TXG_INITIAL; 3437 nvlist_t **spares, **l2cache; 3438 uint_t nspares, nl2cache; 3439 uint64_t version, obj; 3440 boolean_t has_features; 3441 3442 /* 3443 * If this pool already exists, return failure. 3444 */ 3445 mutex_enter(&spa_namespace_lock); 3446 if (spa_lookup(pool) != NULL) { 3447 mutex_exit(&spa_namespace_lock); 3448 return (SET_ERROR(EEXIST)); 3449 } 3450 3451 /* 3452 * Allocate a new spa_t structure. 3453 */ 3454 (void) nvlist_lookup_string(props, 3455 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 3456 spa = spa_add(pool, NULL, altroot); 3457 spa_activate(spa, spa_mode_global); 3458 3459 if (props && (error = spa_prop_validate(spa, props))) { 3460 spa_deactivate(spa); 3461 spa_remove(spa); 3462 mutex_exit(&spa_namespace_lock); 3463 return (error); 3464 } 3465 3466 has_features = B_FALSE; 3467 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL); 3468 elem != NULL; elem = nvlist_next_nvpair(props, elem)) { 3469 if (zpool_prop_feature(nvpair_name(elem))) 3470 has_features = B_TRUE; 3471 } 3472 3473 if (has_features || nvlist_lookup_uint64(props, 3474 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) { 3475 version = SPA_VERSION; 3476 } 3477 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 3478 3479 spa->spa_first_txg = txg; 3480 spa->spa_uberblock.ub_txg = txg - 1; 3481 spa->spa_uberblock.ub_version = version; 3482 spa->spa_ubsync = spa->spa_uberblock; 3483 3484 /* 3485 * Create "The Godfather" zio to hold all async IOs 3486 */ 3487 spa->spa_async_zio_root = zio_root(spa, NULL, NULL, 3488 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER); 3489 3490 /* 3491 * Create the root vdev. 3492 */ 3493 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3494 3495 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD); 3496 3497 ASSERT(error != 0 || rvd != NULL); 3498 ASSERT(error != 0 || spa->spa_root_vdev == rvd); 3499 3500 if (error == 0 && !zfs_allocatable_devs(nvroot)) 3501 error = SET_ERROR(EINVAL); 3502 3503 if (error == 0 && 3504 (error = vdev_create(rvd, txg, B_FALSE)) == 0 && 3505 (error = spa_validate_aux(spa, nvroot, txg, 3506 VDEV_ALLOC_ADD)) == 0) { 3507 for (int c = 0; c < rvd->vdev_children; c++) { 3508 vdev_ashift_optimize(rvd->vdev_child[c]); 3509 vdev_metaslab_set_size(rvd->vdev_child[c]); 3510 vdev_expand(rvd->vdev_child[c], txg); 3511 } 3512 } 3513 3514 spa_config_exit(spa, SCL_ALL, FTAG); 3515 3516 if (error != 0) { 3517 spa_unload(spa); 3518 spa_deactivate(spa); 3519 spa_remove(spa); 3520 mutex_exit(&spa_namespace_lock); 3521 return (error); 3522 } 3523 3524 /* 3525 * Get the list of spares, if specified. 3526 */ 3527 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 3528 &spares, &nspares) == 0) { 3529 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME, 3530 KM_SLEEP) == 0); 3531 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 3532 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 3533 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3534 spa_load_spares(spa); 3535 spa_config_exit(spa, SCL_ALL, FTAG); 3536 spa->spa_spares.sav_sync = B_TRUE; 3537 } 3538 3539 /* 3540 * Get the list of level 2 cache devices, if specified. 3541 */ 3542 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 3543 &l2cache, &nl2cache) == 0) { 3544 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 3545 NV_UNIQUE_NAME, KM_SLEEP) == 0); 3546 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 3547 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 3548 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3549 spa_load_l2cache(spa); 3550 spa_config_exit(spa, SCL_ALL, FTAG); 3551 spa->spa_l2cache.sav_sync = B_TRUE; 3552 } 3553 3554 spa->spa_is_initializing = B_TRUE; 3555 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg); 3556 spa->spa_meta_objset = dp->dp_meta_objset; 3557 spa->spa_is_initializing = B_FALSE; 3558 3559 /* 3560 * Create DDTs (dedup tables). 3561 */ 3562 ddt_create(spa); 3563 3564 spa_update_dspace(spa); 3565 3566 tx = dmu_tx_create_assigned(dp, txg); 3567 3568 /* 3569 * Create the pool config object. 3570 */ 3571 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset, 3572 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE, 3573 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx); 3574 3575 if (zap_add(spa->spa_meta_objset, 3576 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG, 3577 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) { 3578 cmn_err(CE_PANIC, "failed to add pool config"); 3579 } 3580 3581 if (spa_version(spa) >= SPA_VERSION_FEATURES) 3582 spa_feature_create_zap_objects(spa, tx); 3583 3584 if (zap_add(spa->spa_meta_objset, 3585 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION, 3586 sizeof (uint64_t), 1, &version, tx) != 0) { 3587 cmn_err(CE_PANIC, "failed to add pool version"); 3588 } 3589 3590 /* Newly created pools with the right version are always deflated. */ 3591 if (version >= SPA_VERSION_RAIDZ_DEFLATE) { 3592 spa->spa_deflate = TRUE; 3593 if (zap_add(spa->spa_meta_objset, 3594 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 3595 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) { 3596 cmn_err(CE_PANIC, "failed to add deflate"); 3597 } 3598 } 3599 3600 /* 3601 * Create the deferred-free bpobj. Turn off compression 3602 * because sync-to-convergence takes longer if the blocksize 3603 * keeps changing. 3604 */ 3605 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx); 3606 dmu_object_set_compress(spa->spa_meta_objset, obj, 3607 ZIO_COMPRESS_OFF, tx); 3608 if (zap_add(spa->spa_meta_objset, 3609 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ, 3610 sizeof (uint64_t), 1, &obj, tx) != 0) { 3611 cmn_err(CE_PANIC, "failed to add bpobj"); 3612 } 3613 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj, 3614 spa->spa_meta_objset, obj)); 3615 3616 /* 3617 * Create the pool's history object. 3618 */ 3619 if (version >= SPA_VERSION_ZPOOL_HISTORY) 3620 spa_history_create_obj(spa, tx); 3621 3622 /* 3623 * Set pool properties. 3624 */ 3625 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS); 3626 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION); 3627 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE); 3628 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND); 3629 3630 if (props != NULL) { 3631 spa_configfile_set(spa, props, B_FALSE); 3632 spa_sync_props(props, tx); 3633 } 3634 3635 dmu_tx_commit(tx); 3636 3637 spa->spa_sync_on = B_TRUE; 3638 txg_sync_start(spa->spa_dsl_pool); 3639 3640 /* 3641 * We explicitly wait for the first transaction to complete so that our 3642 * bean counters are appropriately updated. 3643 */ 3644 txg_wait_synced(spa->spa_dsl_pool, txg); 3645 3646 spa_config_sync(spa, B_FALSE, B_TRUE); 3647 3648 spa_history_log_version(spa, "create"); 3649 3650 spa->spa_minref = refcount_count(&spa->spa_refcount); 3651 3652 mutex_exit(&spa_namespace_lock); 3653 3654 return (0); 3655} 3656 3657#ifdef _KERNEL 3658#if defined(sun) 3659/* 3660 * Get the root pool information from the root disk, then import the root pool 3661 * during the system boot up time. 3662 */ 3663extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **); 3664 3665static nvlist_t * 3666spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid) 3667{ 3668 nvlist_t *config; 3669 nvlist_t *nvtop, *nvroot; 3670 uint64_t pgid; 3671 3672 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0) 3673 return (NULL); 3674 3675 /* 3676 * Add this top-level vdev to the child array. 3677 */ 3678 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3679 &nvtop) == 0); 3680 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3681 &pgid) == 0); 3682 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0); 3683 3684 /* 3685 * Put this pool's top-level vdevs into a root vdev. 3686 */ 3687 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3688 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3689 VDEV_TYPE_ROOT) == 0); 3690 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3691 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3692 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3693 &nvtop, 1) == 0); 3694 3695 /* 3696 * Replace the existing vdev_tree with the new root vdev in 3697 * this pool's configuration (remove the old, add the new). 3698 */ 3699 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3700 nvlist_free(nvroot); 3701 return (config); 3702} 3703 3704/* 3705 * Walk the vdev tree and see if we can find a device with "better" 3706 * configuration. A configuration is "better" if the label on that 3707 * device has a more recent txg. 3708 */ 3709static void 3710spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg) 3711{ 3712 for (int c = 0; c < vd->vdev_children; c++) 3713 spa_alt_rootvdev(vd->vdev_child[c], avd, txg); 3714 3715 if (vd->vdev_ops->vdev_op_leaf) { 3716 nvlist_t *label; 3717 uint64_t label_txg; 3718 3719 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid, 3720 &label) != 0) 3721 return; 3722 3723 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG, 3724 &label_txg) == 0); 3725 3726 /* 3727 * Do we have a better boot device? 3728 */ 3729 if (label_txg > *txg) { 3730 *txg = label_txg; 3731 *avd = vd; 3732 } 3733 nvlist_free(label); 3734 } 3735} 3736 3737/* 3738 * Import a root pool. 3739 * 3740 * For x86. devpath_list will consist of devid and/or physpath name of 3741 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a"). 3742 * The GRUB "findroot" command will return the vdev we should boot. 3743 * 3744 * For Sparc, devpath_list consists the physpath name of the booting device 3745 * no matter the rootpool is a single device pool or a mirrored pool. 3746 * e.g. 3747 * "/pci@1f,0/ide@d/disk@0,0:a" 3748 */ 3749int 3750spa_import_rootpool(char *devpath, char *devid) 3751{ 3752 spa_t *spa; 3753 vdev_t *rvd, *bvd, *avd = NULL; 3754 nvlist_t *config, *nvtop; 3755 uint64_t guid, txg; 3756 char *pname; 3757 int error; 3758 3759 /* 3760 * Read the label from the boot device and generate a configuration. 3761 */ 3762 config = spa_generate_rootconf(devpath, devid, &guid); 3763#if defined(_OBP) && defined(_KERNEL) 3764 if (config == NULL) { 3765 if (strstr(devpath, "/iscsi/ssd") != NULL) { 3766 /* iscsi boot */ 3767 get_iscsi_bootpath_phy(devpath); 3768 config = spa_generate_rootconf(devpath, devid, &guid); 3769 } 3770 } 3771#endif 3772 if (config == NULL) { 3773 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'", 3774 devpath); 3775 return (SET_ERROR(EIO)); 3776 } 3777 3778 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3779 &pname) == 0); 3780 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0); 3781 3782 mutex_enter(&spa_namespace_lock); 3783 if ((spa = spa_lookup(pname)) != NULL) { 3784 /* 3785 * Remove the existing root pool from the namespace so that we 3786 * can replace it with the correct config we just read in. 3787 */ 3788 spa_remove(spa); 3789 } 3790 3791 spa = spa_add(pname, config, NULL); 3792 spa->spa_is_root = B_TRUE; 3793 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 3794 3795 /* 3796 * Build up a vdev tree based on the boot device's label config. 3797 */ 3798 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 3799 &nvtop) == 0); 3800 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3801 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 3802 VDEV_ALLOC_ROOTPOOL); 3803 spa_config_exit(spa, SCL_ALL, FTAG); 3804 if (error) { 3805 mutex_exit(&spa_namespace_lock); 3806 nvlist_free(config); 3807 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 3808 pname); 3809 return (error); 3810 } 3811 3812 /* 3813 * Get the boot vdev. 3814 */ 3815 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) { 3816 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu", 3817 (u_longlong_t)guid); 3818 error = SET_ERROR(ENOENT); 3819 goto out; 3820 } 3821 3822 /* 3823 * Determine if there is a better boot device. 3824 */ 3825 avd = bvd; 3826 spa_alt_rootvdev(rvd, &avd, &txg); 3827 if (avd != bvd) { 3828 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please " 3829 "try booting from '%s'", avd->vdev_path); 3830 error = SET_ERROR(EINVAL); 3831 goto out; 3832 } 3833 3834 /* 3835 * If the boot device is part of a spare vdev then ensure that 3836 * we're booting off the active spare. 3837 */ 3838 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops && 3839 !bvd->vdev_isspare) { 3840 cmn_err(CE_NOTE, "The boot device is currently spared. Please " 3841 "try booting from '%s'", 3842 bvd->vdev_parent-> 3843 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path); 3844 error = SET_ERROR(EINVAL); 3845 goto out; 3846 } 3847 3848 error = 0; 3849out: 3850 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 3851 vdev_free(rvd); 3852 spa_config_exit(spa, SCL_ALL, FTAG); 3853 mutex_exit(&spa_namespace_lock); 3854 3855 nvlist_free(config); 3856 return (error); 3857} 3858 3859#else 3860 3861extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs, 3862 uint64_t *count); 3863 3864static nvlist_t * 3865spa_generate_rootconf(const char *name) 3866{ 3867 nvlist_t **configs, **tops; 3868 nvlist_t *config; 3869 nvlist_t *best_cfg, *nvtop, *nvroot; 3870 uint64_t *holes; 3871 uint64_t best_txg; 3872 uint64_t nchildren; 3873 uint64_t pgid; 3874 uint64_t count; 3875 uint64_t i; 3876 uint_t nholes; 3877 3878 if (vdev_geom_read_pool_label(name, &configs, &count) != 0) 3879 return (NULL); 3880 3881 ASSERT3U(count, !=, 0); 3882 best_txg = 0; 3883 for (i = 0; i < count; i++) { 3884 uint64_t txg; 3885 3886 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG, 3887 &txg) == 0); 3888 if (txg > best_txg) { 3889 best_txg = txg; 3890 best_cfg = configs[i]; 3891 } 3892 } 3893 3894 /* 3895 * Multi-vdev root pool configuration discovery is not supported yet. 3896 */ 3897 nchildren = 1; 3898 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren); 3899 holes = NULL; 3900 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY, 3901 &holes, &nholes); 3902 3903 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP); 3904 for (i = 0; i < nchildren; i++) { 3905 if (i >= count) 3906 break; 3907 if (configs[i] == NULL) 3908 continue; 3909 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE, 3910 &nvtop) == 0); 3911 nvlist_dup(nvtop, &tops[i], KM_SLEEP); 3912 } 3913 for (i = 0; holes != NULL && i < nholes; i++) { 3914 if (i >= nchildren) 3915 continue; 3916 if (tops[holes[i]] != NULL) 3917 continue; 3918 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP); 3919 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE, 3920 VDEV_TYPE_HOLE) == 0); 3921 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, 3922 holes[i]) == 0); 3923 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 3924 0) == 0); 3925 } 3926 for (i = 0; i < nchildren; i++) { 3927 if (tops[i] != NULL) 3928 continue; 3929 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP); 3930 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE, 3931 VDEV_TYPE_MISSING) == 0); 3932 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, 3933 i) == 0); 3934 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 3935 0) == 0); 3936 } 3937 3938 /* 3939 * Create pool config based on the best vdev config. 3940 */ 3941 nvlist_dup(best_cfg, &config, KM_SLEEP); 3942 3943 /* 3944 * Put this pool's top-level vdevs into a root vdev. 3945 */ 3946 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, 3947 &pgid) == 0); 3948 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 3949 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, 3950 VDEV_TYPE_ROOT) == 0); 3951 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0); 3952 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0); 3953 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, 3954 tops, nchildren) == 0); 3955 3956 /* 3957 * Replace the existing vdev_tree with the new root vdev in 3958 * this pool's configuration (remove the old, add the new). 3959 */ 3960 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 3961 3962 /* 3963 * Drop vdev config elements that should not be present at pool level. 3964 */ 3965 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64); 3966 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64); 3967 3968 for (i = 0; i < count; i++) 3969 nvlist_free(configs[i]); 3970 kmem_free(configs, count * sizeof(void *)); 3971 for (i = 0; i < nchildren; i++) 3972 nvlist_free(tops[i]); 3973 kmem_free(tops, nchildren * sizeof(void *)); 3974 nvlist_free(nvroot); 3975 return (config); 3976} 3977 3978int 3979spa_import_rootpool(const char *name) 3980{ 3981 spa_t *spa; 3982 vdev_t *rvd, *bvd, *avd = NULL; 3983 nvlist_t *config, *nvtop; 3984 uint64_t txg; 3985 char *pname; 3986 int error; 3987 3988 /* 3989 * Read the label from the boot device and generate a configuration. 3990 */ 3991 config = spa_generate_rootconf(name); 3992 3993 mutex_enter(&spa_namespace_lock); 3994 if (config != NULL) { 3995 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME, 3996 &pname) == 0 && strcmp(name, pname) == 0); 3997 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) 3998 == 0); 3999 4000 if ((spa = spa_lookup(pname)) != NULL) { 4001 /* 4002 * Remove the existing root pool from the namespace so 4003 * that we can replace it with the correct config 4004 * we just read in. 4005 */ 4006 spa_remove(spa); 4007 } 4008 spa = spa_add(pname, config, NULL); 4009 4010 /* 4011 * Set spa_ubsync.ub_version as it can be used in vdev_alloc() 4012 * via spa_version(). 4013 */ 4014 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, 4015 &spa->spa_ubsync.ub_version) != 0) 4016 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL; 4017 } else if ((spa = spa_lookup(name)) == NULL) { 4018 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'", 4019 name); 4020 return (EIO); 4021 } else { 4022 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0); 4023 } 4024 spa->spa_is_root = B_TRUE; 4025 spa->spa_import_flags = ZFS_IMPORT_VERBATIM; 4026 4027 /* 4028 * Build up a vdev tree based on the boot device's label config. 4029 */ 4030 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4031 &nvtop) == 0); 4032 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4033 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0, 4034 VDEV_ALLOC_ROOTPOOL); 4035 spa_config_exit(spa, SCL_ALL, FTAG); 4036 if (error) { 4037 mutex_exit(&spa_namespace_lock); 4038 nvlist_free(config); 4039 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'", 4040 pname); 4041 return (error); 4042 } 4043 4044 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4045 vdev_free(rvd); 4046 spa_config_exit(spa, SCL_ALL, FTAG); 4047 mutex_exit(&spa_namespace_lock); 4048 4049 nvlist_free(config); 4050 return (0); 4051} 4052 4053#endif /* sun */ 4054#endif 4055 4056/* 4057 * Import a non-root pool into the system. 4058 */ 4059int 4060spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags) 4061{ 4062 spa_t *spa; 4063 char *altroot = NULL; 4064 spa_load_state_t state = SPA_LOAD_IMPORT; 4065 zpool_rewind_policy_t policy; 4066 uint64_t mode = spa_mode_global; 4067 uint64_t readonly = B_FALSE; 4068 int error; 4069 nvlist_t *nvroot; 4070 nvlist_t **spares, **l2cache; 4071 uint_t nspares, nl2cache; 4072 4073 /* 4074 * If a pool with this name exists, return failure. 4075 */ 4076 mutex_enter(&spa_namespace_lock); 4077 if (spa_lookup(pool) != NULL) { 4078 mutex_exit(&spa_namespace_lock); 4079 return (SET_ERROR(EEXIST)); 4080 } 4081 4082 /* 4083 * Create and initialize the spa structure. 4084 */ 4085 (void) nvlist_lookup_string(props, 4086 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 4087 (void) nvlist_lookup_uint64(props, 4088 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly); 4089 if (readonly) 4090 mode = FREAD; 4091 spa = spa_add(pool, config, altroot); 4092 spa->spa_import_flags = flags; 4093 4094 /* 4095 * Verbatim import - Take a pool and insert it into the namespace 4096 * as if it had been loaded at boot. 4097 */ 4098 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) { 4099 if (props != NULL) 4100 spa_configfile_set(spa, props, B_FALSE); 4101 4102 spa_config_sync(spa, B_FALSE, B_TRUE); 4103 4104 mutex_exit(&spa_namespace_lock); 4105 return (0); 4106 } 4107 4108 spa_activate(spa, mode); 4109 4110 /* 4111 * Don't start async tasks until we know everything is healthy. 4112 */ 4113 spa_async_suspend(spa); 4114 4115 zpool_get_rewind_policy(config, &policy); 4116 if (policy.zrp_request & ZPOOL_DO_REWIND) 4117 state = SPA_LOAD_RECOVER; 4118 4119 /* 4120 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig 4121 * because the user-supplied config is actually the one to trust when 4122 * doing an import. 4123 */ 4124 if (state != SPA_LOAD_RECOVER) 4125 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0; 4126 4127 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg, 4128 policy.zrp_request); 4129 4130 /* 4131 * Propagate anything learned while loading the pool and pass it 4132 * back to caller (i.e. rewind info, missing devices, etc). 4133 */ 4134 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4135 spa->spa_load_info) == 0); 4136 4137 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4138 /* 4139 * Toss any existing sparelist, as it doesn't have any validity 4140 * anymore, and conflicts with spa_has_spare(). 4141 */ 4142 if (spa->spa_spares.sav_config) { 4143 nvlist_free(spa->spa_spares.sav_config); 4144 spa->spa_spares.sav_config = NULL; 4145 spa_load_spares(spa); 4146 } 4147 if (spa->spa_l2cache.sav_config) { 4148 nvlist_free(spa->spa_l2cache.sav_config); 4149 spa->spa_l2cache.sav_config = NULL; 4150 spa_load_l2cache(spa); 4151 } 4152 4153 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, 4154 &nvroot) == 0); 4155 if (error == 0) 4156 error = spa_validate_aux(spa, nvroot, -1ULL, 4157 VDEV_ALLOC_SPARE); 4158 if (error == 0) 4159 error = spa_validate_aux(spa, nvroot, -1ULL, 4160 VDEV_ALLOC_L2CACHE); 4161 spa_config_exit(spa, SCL_ALL, FTAG); 4162 4163 if (props != NULL) 4164 spa_configfile_set(spa, props, B_FALSE); 4165 4166 if (error != 0 || (props && spa_writeable(spa) && 4167 (error = spa_prop_set(spa, props)))) { 4168 spa_unload(spa); 4169 spa_deactivate(spa); 4170 spa_remove(spa); 4171 mutex_exit(&spa_namespace_lock); 4172 return (error); 4173 } 4174 4175 spa_async_resume(spa); 4176 4177 /* 4178 * Override any spares and level 2 cache devices as specified by 4179 * the user, as these may have correct device names/devids, etc. 4180 */ 4181 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, 4182 &spares, &nspares) == 0) { 4183 if (spa->spa_spares.sav_config) 4184 VERIFY(nvlist_remove(spa->spa_spares.sav_config, 4185 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0); 4186 else 4187 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, 4188 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4189 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config, 4190 ZPOOL_CONFIG_SPARES, spares, nspares) == 0); 4191 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4192 spa_load_spares(spa); 4193 spa_config_exit(spa, SCL_ALL, FTAG); 4194 spa->spa_spares.sav_sync = B_TRUE; 4195 } 4196 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, 4197 &l2cache, &nl2cache) == 0) { 4198 if (spa->spa_l2cache.sav_config) 4199 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config, 4200 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0); 4201 else 4202 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config, 4203 NV_UNIQUE_NAME, KM_SLEEP) == 0); 4204 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config, 4205 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0); 4206 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4207 spa_load_l2cache(spa); 4208 spa_config_exit(spa, SCL_ALL, FTAG); 4209 spa->spa_l2cache.sav_sync = B_TRUE; 4210 } 4211 4212 /* 4213 * Check for any removed devices. 4214 */ 4215 if (spa->spa_autoreplace) { 4216 spa_aux_check_removed(&spa->spa_spares); 4217 spa_aux_check_removed(&spa->spa_l2cache); 4218 } 4219 4220 if (spa_writeable(spa)) { 4221 /* 4222 * Update the config cache to include the newly-imported pool. 4223 */ 4224 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4225 } 4226 4227 /* 4228 * It's possible that the pool was expanded while it was exported. 4229 * We kick off an async task to handle this for us. 4230 */ 4231 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND); 4232 4233 mutex_exit(&spa_namespace_lock); 4234 spa_history_log_version(spa, "import"); 4235 4236#ifdef __FreeBSD__ 4237#ifdef _KERNEL 4238 zvol_create_minors(pool); 4239#endif 4240#endif 4241 return (0); 4242} 4243 4244nvlist_t * 4245spa_tryimport(nvlist_t *tryconfig) 4246{ 4247 nvlist_t *config = NULL; 4248 char *poolname; 4249 spa_t *spa; 4250 uint64_t state; 4251 int error; 4252 4253 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname)) 4254 return (NULL); 4255 4256 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state)) 4257 return (NULL); 4258 4259 /* 4260 * Create and initialize the spa structure. 4261 */ 4262 mutex_enter(&spa_namespace_lock); 4263 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL); 4264 spa_activate(spa, FREAD); 4265 4266 /* 4267 * Pass off the heavy lifting to spa_load(). 4268 * Pass TRUE for mosconfig because the user-supplied config 4269 * is actually the one to trust when doing an import. 4270 */ 4271 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE); 4272 4273 /* 4274 * If 'tryconfig' was at least parsable, return the current config. 4275 */ 4276 if (spa->spa_root_vdev != NULL) { 4277 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE); 4278 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 4279 poolname) == 0); 4280 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 4281 state) == 0); 4282 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP, 4283 spa->spa_uberblock.ub_timestamp) == 0); 4284 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, 4285 spa->spa_load_info) == 0); 4286 4287 /* 4288 * If the bootfs property exists on this pool then we 4289 * copy it out so that external consumers can tell which 4290 * pools are bootable. 4291 */ 4292 if ((!error || error == EEXIST) && spa->spa_bootfs) { 4293 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4294 4295 /* 4296 * We have to play games with the name since the 4297 * pool was opened as TRYIMPORT_NAME. 4298 */ 4299 if (dsl_dsobj_to_dsname(spa_name(spa), 4300 spa->spa_bootfs, tmpname) == 0) { 4301 char *cp; 4302 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 4303 4304 cp = strchr(tmpname, '/'); 4305 if (cp == NULL) { 4306 (void) strlcpy(dsname, tmpname, 4307 MAXPATHLEN); 4308 } else { 4309 (void) snprintf(dsname, MAXPATHLEN, 4310 "%s/%s", poolname, ++cp); 4311 } 4312 VERIFY(nvlist_add_string(config, 4313 ZPOOL_CONFIG_BOOTFS, dsname) == 0); 4314 kmem_free(dsname, MAXPATHLEN); 4315 } 4316 kmem_free(tmpname, MAXPATHLEN); 4317 } 4318 4319 /* 4320 * Add the list of hot spares and level 2 cache devices. 4321 */ 4322 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 4323 spa_add_spares(spa, config); 4324 spa_add_l2cache(spa, config); 4325 spa_config_exit(spa, SCL_CONFIG, FTAG); 4326 } 4327 4328 spa_unload(spa); 4329 spa_deactivate(spa); 4330 spa_remove(spa); 4331 mutex_exit(&spa_namespace_lock); 4332 4333 return (config); 4334} 4335 4336/* 4337 * Pool export/destroy 4338 * 4339 * The act of destroying or exporting a pool is very simple. We make sure there 4340 * is no more pending I/O and any references to the pool are gone. Then, we 4341 * update the pool state and sync all the labels to disk, removing the 4342 * configuration from the cache afterwards. If the 'hardforce' flag is set, then 4343 * we don't sync the labels or remove the configuration cache. 4344 */ 4345static int 4346spa_export_common(char *pool, int new_state, nvlist_t **oldconfig, 4347 boolean_t force, boolean_t hardforce) 4348{ 4349 spa_t *spa; 4350 4351 if (oldconfig) 4352 *oldconfig = NULL; 4353 4354 if (!(spa_mode_global & FWRITE)) 4355 return (SET_ERROR(EROFS)); 4356 4357 mutex_enter(&spa_namespace_lock); 4358 if ((spa = spa_lookup(pool)) == NULL) { 4359 mutex_exit(&spa_namespace_lock); 4360 return (SET_ERROR(ENOENT)); 4361 } 4362 4363 /* 4364 * Put a hold on the pool, drop the namespace lock, stop async tasks, 4365 * reacquire the namespace lock, and see if we can export. 4366 */ 4367 spa_open_ref(spa, FTAG); 4368 mutex_exit(&spa_namespace_lock); 4369 spa_async_suspend(spa); 4370 mutex_enter(&spa_namespace_lock); 4371 spa_close(spa, FTAG); 4372 4373 /* 4374 * The pool will be in core if it's openable, 4375 * in which case we can modify its state. 4376 */ 4377 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) { 4378 /* 4379 * Objsets may be open only because they're dirty, so we 4380 * have to force it to sync before checking spa_refcnt. 4381 */ 4382 txg_wait_synced(spa->spa_dsl_pool, 0); 4383 4384 /* 4385 * A pool cannot be exported or destroyed if there are active 4386 * references. If we are resetting a pool, allow references by 4387 * fault injection handlers. 4388 */ 4389 if (!spa_refcount_zero(spa) || 4390 (spa->spa_inject_ref != 0 && 4391 new_state != POOL_STATE_UNINITIALIZED)) { 4392 spa_async_resume(spa); 4393 mutex_exit(&spa_namespace_lock); 4394 return (SET_ERROR(EBUSY)); 4395 } 4396 4397 /* 4398 * A pool cannot be exported if it has an active shared spare. 4399 * This is to prevent other pools stealing the active spare 4400 * from an exported pool. At user's own will, such pool can 4401 * be forcedly exported. 4402 */ 4403 if (!force && new_state == POOL_STATE_EXPORTED && 4404 spa_has_active_shared_spare(spa)) { 4405 spa_async_resume(spa); 4406 mutex_exit(&spa_namespace_lock); 4407 return (SET_ERROR(EXDEV)); 4408 } 4409 4410 /* 4411 * We want this to be reflected on every label, 4412 * so mark them all dirty. spa_unload() will do the 4413 * final sync that pushes these changes out. 4414 */ 4415 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) { 4416 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 4417 spa->spa_state = new_state; 4418 spa->spa_final_txg = spa_last_synced_txg(spa) + 4419 TXG_DEFER_SIZE + 1; 4420 vdev_config_dirty(spa->spa_root_vdev); 4421 spa_config_exit(spa, SCL_ALL, FTAG); 4422 } 4423 } 4424 4425 spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY); 4426 4427 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 4428 spa_unload(spa); 4429 spa_deactivate(spa); 4430 } 4431 4432 if (oldconfig && spa->spa_config) 4433 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0); 4434 4435 if (new_state != POOL_STATE_UNINITIALIZED) { 4436 if (!hardforce) 4437 spa_config_sync(spa, B_TRUE, B_TRUE); 4438 spa_remove(spa); 4439 } 4440 mutex_exit(&spa_namespace_lock); 4441 4442 return (0); 4443} 4444 4445/* 4446 * Destroy a storage pool. 4447 */ 4448int 4449spa_destroy(char *pool) 4450{ 4451 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL, 4452 B_FALSE, B_FALSE)); 4453} 4454 4455/* 4456 * Export a storage pool. 4457 */ 4458int 4459spa_export(char *pool, nvlist_t **oldconfig, boolean_t force, 4460 boolean_t hardforce) 4461{ 4462 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig, 4463 force, hardforce)); 4464} 4465 4466/* 4467 * Similar to spa_export(), this unloads the spa_t without actually removing it 4468 * from the namespace in any way. 4469 */ 4470int 4471spa_reset(char *pool) 4472{ 4473 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL, 4474 B_FALSE, B_FALSE)); 4475} 4476 4477/* 4478 * ========================================================================== 4479 * Device manipulation 4480 * ========================================================================== 4481 */ 4482 4483/* 4484 * Add a device to a storage pool. 4485 */ 4486int 4487spa_vdev_add(spa_t *spa, nvlist_t *nvroot) 4488{ 4489 uint64_t txg, id; 4490 int error; 4491 vdev_t *rvd = spa->spa_root_vdev; 4492 vdev_t *vd, *tvd; 4493 nvlist_t **spares, **l2cache; 4494 uint_t nspares, nl2cache; 4495 4496 ASSERT(spa_writeable(spa)); 4497 4498 txg = spa_vdev_enter(spa); 4499 4500 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0, 4501 VDEV_ALLOC_ADD)) != 0) 4502 return (spa_vdev_exit(spa, NULL, txg, error)); 4503 4504 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */ 4505 4506 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares, 4507 &nspares) != 0) 4508 nspares = 0; 4509 4510 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache, 4511 &nl2cache) != 0) 4512 nl2cache = 0; 4513 4514 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0) 4515 return (spa_vdev_exit(spa, vd, txg, EINVAL)); 4516 4517 if (vd->vdev_children != 0 && 4518 (error = vdev_create(vd, txg, B_FALSE)) != 0) 4519 return (spa_vdev_exit(spa, vd, txg, error)); 4520 4521 /* 4522 * We must validate the spares and l2cache devices after checking the 4523 * children. Otherwise, vdev_inuse() will blindly overwrite the spare. 4524 */ 4525 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0) 4526 return (spa_vdev_exit(spa, vd, txg, error)); 4527 4528 /* 4529 * Transfer each new top-level vdev from vd to rvd. 4530 */ 4531 for (int c = 0; c < vd->vdev_children; c++) { 4532 4533 /* 4534 * Set the vdev id to the first hole, if one exists. 4535 */ 4536 for (id = 0; id < rvd->vdev_children; id++) { 4537 if (rvd->vdev_child[id]->vdev_ishole) { 4538 vdev_free(rvd->vdev_child[id]); 4539 break; 4540 } 4541 } 4542 tvd = vd->vdev_child[c]; 4543 vdev_remove_child(vd, tvd); 4544 tvd->vdev_id = id; 4545 vdev_add_child(rvd, tvd); 4546 vdev_config_dirty(tvd); 4547 } 4548 4549 if (nspares != 0) { 4550 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares, 4551 ZPOOL_CONFIG_SPARES); 4552 spa_load_spares(spa); 4553 spa->spa_spares.sav_sync = B_TRUE; 4554 } 4555 4556 if (nl2cache != 0) { 4557 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache, 4558 ZPOOL_CONFIG_L2CACHE); 4559 spa_load_l2cache(spa); 4560 spa->spa_l2cache.sav_sync = B_TRUE; 4561 } 4562 4563 /* 4564 * We have to be careful when adding new vdevs to an existing pool. 4565 * If other threads start allocating from these vdevs before we 4566 * sync the config cache, and we lose power, then upon reboot we may 4567 * fail to open the pool because there are DVAs that the config cache 4568 * can't translate. Therefore, we first add the vdevs without 4569 * initializing metaslabs; sync the config cache (via spa_vdev_exit()); 4570 * and then let spa_config_update() initialize the new metaslabs. 4571 * 4572 * spa_load() checks for added-but-not-initialized vdevs, so that 4573 * if we lose power at any point in this sequence, the remaining 4574 * steps will be completed the next time we load the pool. 4575 */ 4576 (void) spa_vdev_exit(spa, vd, txg, 0); 4577 4578 mutex_enter(&spa_namespace_lock); 4579 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 4580 mutex_exit(&spa_namespace_lock); 4581 4582 return (0); 4583} 4584 4585/* 4586 * Attach a device to a mirror. The arguments are the path to any device 4587 * in the mirror, and the nvroot for the new device. If the path specifies 4588 * a device that is not mirrored, we automatically insert the mirror vdev. 4589 * 4590 * If 'replacing' is specified, the new device is intended to replace the 4591 * existing device; in this case the two devices are made into their own 4592 * mirror using the 'replacing' vdev, which is functionally identical to 4593 * the mirror vdev (it actually reuses all the same ops) but has a few 4594 * extra rules: you can't attach to it after it's been created, and upon 4595 * completion of resilvering, the first disk (the one being replaced) 4596 * is automatically detached. 4597 */ 4598int 4599spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing) 4600{ 4601 uint64_t txg, dtl_max_txg; 4602 vdev_t *rvd = spa->spa_root_vdev; 4603 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd; 4604 vdev_ops_t *pvops; 4605 char *oldvdpath, *newvdpath; 4606 int newvd_isspare; 4607 int error; 4608 4609 ASSERT(spa_writeable(spa)); 4610 4611 txg = spa_vdev_enter(spa); 4612 4613 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE); 4614 4615 if (oldvd == NULL) 4616 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4617 4618 if (!oldvd->vdev_ops->vdev_op_leaf) 4619 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4620 4621 pvd = oldvd->vdev_parent; 4622 4623 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0, 4624 VDEV_ALLOC_ATTACH)) != 0) 4625 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 4626 4627 if (newrootvd->vdev_children != 1) 4628 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4629 4630 newvd = newrootvd->vdev_child[0]; 4631 4632 if (!newvd->vdev_ops->vdev_op_leaf) 4633 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL)); 4634 4635 if ((error = vdev_create(newrootvd, txg, replacing)) != 0) 4636 return (spa_vdev_exit(spa, newrootvd, txg, error)); 4637 4638 /* 4639 * Spares can't replace logs 4640 */ 4641 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare) 4642 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4643 4644 if (!replacing) { 4645 /* 4646 * For attach, the only allowable parent is a mirror or the root 4647 * vdev. 4648 */ 4649 if (pvd->vdev_ops != &vdev_mirror_ops && 4650 pvd->vdev_ops != &vdev_root_ops) 4651 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4652 4653 pvops = &vdev_mirror_ops; 4654 } else { 4655 /* 4656 * Active hot spares can only be replaced by inactive hot 4657 * spares. 4658 */ 4659 if (pvd->vdev_ops == &vdev_spare_ops && 4660 oldvd->vdev_isspare && 4661 !spa_has_spare(spa, newvd->vdev_guid)) 4662 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4663 4664 /* 4665 * If the source is a hot spare, and the parent isn't already a 4666 * spare, then we want to create a new hot spare. Otherwise, we 4667 * want to create a replacing vdev. The user is not allowed to 4668 * attach to a spared vdev child unless the 'isspare' state is 4669 * the same (spare replaces spare, non-spare replaces 4670 * non-spare). 4671 */ 4672 if (pvd->vdev_ops == &vdev_replacing_ops && 4673 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) { 4674 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4675 } else if (pvd->vdev_ops == &vdev_spare_ops && 4676 newvd->vdev_isspare != oldvd->vdev_isspare) { 4677 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP)); 4678 } 4679 4680 if (newvd->vdev_isspare) 4681 pvops = &vdev_spare_ops; 4682 else 4683 pvops = &vdev_replacing_ops; 4684 } 4685 4686 /* 4687 * Make sure the new device is big enough. 4688 */ 4689 if (newvd->vdev_asize < vdev_get_min_asize(oldvd)) 4690 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW)); 4691 4692 /* 4693 * The new device cannot have a higher alignment requirement 4694 * than the top-level vdev. 4695 */ 4696 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift) 4697 return (spa_vdev_exit(spa, newrootvd, txg, EDOM)); 4698 4699 /* 4700 * If this is an in-place replacement, update oldvd's path and devid 4701 * to make it distinguishable from newvd, and unopenable from now on. 4702 */ 4703 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) { 4704 spa_strfree(oldvd->vdev_path); 4705 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5, 4706 KM_SLEEP); 4707 (void) sprintf(oldvd->vdev_path, "%s/%s", 4708 newvd->vdev_path, "old"); 4709 if (oldvd->vdev_devid != NULL) { 4710 spa_strfree(oldvd->vdev_devid); 4711 oldvd->vdev_devid = NULL; 4712 } 4713 } 4714 4715 /* mark the device being resilvered */ 4716 newvd->vdev_resilver_txg = txg; 4717 4718 /* 4719 * If the parent is not a mirror, or if we're replacing, insert the new 4720 * mirror/replacing/spare vdev above oldvd. 4721 */ 4722 if (pvd->vdev_ops != pvops) 4723 pvd = vdev_add_parent(oldvd, pvops); 4724 4725 ASSERT(pvd->vdev_top->vdev_parent == rvd); 4726 ASSERT(pvd->vdev_ops == pvops); 4727 ASSERT(oldvd->vdev_parent == pvd); 4728 4729 /* 4730 * Extract the new device from its root and add it to pvd. 4731 */ 4732 vdev_remove_child(newrootvd, newvd); 4733 newvd->vdev_id = pvd->vdev_children; 4734 newvd->vdev_crtxg = oldvd->vdev_crtxg; 4735 vdev_add_child(pvd, newvd); 4736 4737 tvd = newvd->vdev_top; 4738 ASSERT(pvd->vdev_top == tvd); 4739 ASSERT(tvd->vdev_parent == rvd); 4740 4741 vdev_config_dirty(tvd); 4742 4743 /* 4744 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account 4745 * for any dmu_sync-ed blocks. It will propagate upward when 4746 * spa_vdev_exit() calls vdev_dtl_reassess(). 4747 */ 4748 dtl_max_txg = txg + TXG_CONCURRENT_STATES; 4749 4750 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL, 4751 dtl_max_txg - TXG_INITIAL); 4752 4753 if (newvd->vdev_isspare) { 4754 spa_spare_activate(newvd); 4755 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE); 4756 } 4757 4758 oldvdpath = spa_strdup(oldvd->vdev_path); 4759 newvdpath = spa_strdup(newvd->vdev_path); 4760 newvd_isspare = newvd->vdev_isspare; 4761 4762 /* 4763 * Mark newvd's DTL dirty in this txg. 4764 */ 4765 vdev_dirty(tvd, VDD_DTL, newvd, txg); 4766 4767 /* 4768 * Restart the resilver 4769 */ 4770 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg); 4771 4772 /* 4773 * Commit the config 4774 */ 4775 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0); 4776 4777 spa_history_log_internal(spa, "vdev attach", NULL, 4778 "%s vdev=%s %s vdev=%s", 4779 replacing && newvd_isspare ? "spare in" : 4780 replacing ? "replace" : "attach", newvdpath, 4781 replacing ? "for" : "to", oldvdpath); 4782 4783 spa_strfree(oldvdpath); 4784 spa_strfree(newvdpath); 4785 4786 if (spa->spa_bootfs) 4787 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH); 4788 4789 return (0); 4790} 4791 4792/* 4793 * Detach a device from a mirror or replacing vdev. 4794 * 4795 * If 'replace_done' is specified, only detach if the parent 4796 * is a replacing vdev. 4797 */ 4798int 4799spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done) 4800{ 4801 uint64_t txg; 4802 int error; 4803 vdev_t *rvd = spa->spa_root_vdev; 4804 vdev_t *vd, *pvd, *cvd, *tvd; 4805 boolean_t unspare = B_FALSE; 4806 uint64_t unspare_guid = 0; 4807 char *vdpath; 4808 4809 ASSERT(spa_writeable(spa)); 4810 4811 txg = spa_vdev_enter(spa); 4812 4813 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 4814 4815 if (vd == NULL) 4816 return (spa_vdev_exit(spa, NULL, txg, ENODEV)); 4817 4818 if (!vd->vdev_ops->vdev_op_leaf) 4819 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4820 4821 pvd = vd->vdev_parent; 4822 4823 /* 4824 * If the parent/child relationship is not as expected, don't do it. 4825 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing 4826 * vdev that's replacing B with C. The user's intent in replacing 4827 * is to go from M(A,B) to M(A,C). If the user decides to cancel 4828 * the replace by detaching C, the expected behavior is to end up 4829 * M(A,B). But suppose that right after deciding to detach C, 4830 * the replacement of B completes. We would have M(A,C), and then 4831 * ask to detach C, which would leave us with just A -- not what 4832 * the user wanted. To prevent this, we make sure that the 4833 * parent/child relationship hasn't changed -- in this example, 4834 * that C's parent is still the replacing vdev R. 4835 */ 4836 if (pvd->vdev_guid != pguid && pguid != 0) 4837 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4838 4839 /* 4840 * Only 'replacing' or 'spare' vdevs can be replaced. 4841 */ 4842 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops && 4843 pvd->vdev_ops != &vdev_spare_ops) 4844 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4845 4846 ASSERT(pvd->vdev_ops != &vdev_spare_ops || 4847 spa_version(spa) >= SPA_VERSION_SPARES); 4848 4849 /* 4850 * Only mirror, replacing, and spare vdevs support detach. 4851 */ 4852 if (pvd->vdev_ops != &vdev_replacing_ops && 4853 pvd->vdev_ops != &vdev_mirror_ops && 4854 pvd->vdev_ops != &vdev_spare_ops) 4855 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP)); 4856 4857 /* 4858 * If this device has the only valid copy of some data, 4859 * we cannot safely detach it. 4860 */ 4861 if (vdev_dtl_required(vd)) 4862 return (spa_vdev_exit(spa, NULL, txg, EBUSY)); 4863 4864 ASSERT(pvd->vdev_children >= 2); 4865 4866 /* 4867 * If we are detaching the second disk from a replacing vdev, then 4868 * check to see if we changed the original vdev's path to have "/old" 4869 * at the end in spa_vdev_attach(). If so, undo that change now. 4870 */ 4871 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 && 4872 vd->vdev_path != NULL) { 4873 size_t len = strlen(vd->vdev_path); 4874 4875 for (int c = 0; c < pvd->vdev_children; c++) { 4876 cvd = pvd->vdev_child[c]; 4877 4878 if (cvd == vd || cvd->vdev_path == NULL) 4879 continue; 4880 4881 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 && 4882 strcmp(cvd->vdev_path + len, "/old") == 0) { 4883 spa_strfree(cvd->vdev_path); 4884 cvd->vdev_path = spa_strdup(vd->vdev_path); 4885 break; 4886 } 4887 } 4888 } 4889 4890 /* 4891 * If we are detaching the original disk from a spare, then it implies 4892 * that the spare should become a real disk, and be removed from the 4893 * active spare list for the pool. 4894 */ 4895 if (pvd->vdev_ops == &vdev_spare_ops && 4896 vd->vdev_id == 0 && 4897 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare) 4898 unspare = B_TRUE; 4899 4900 /* 4901 * Erase the disk labels so the disk can be used for other things. 4902 * This must be done after all other error cases are handled, 4903 * but before we disembowel vd (so we can still do I/O to it). 4904 * But if we can't do it, don't treat the error as fatal -- 4905 * it may be that the unwritability of the disk is the reason 4906 * it's being detached! 4907 */ 4908 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 4909 4910 /* 4911 * Remove vd from its parent and compact the parent's children. 4912 */ 4913 vdev_remove_child(pvd, vd); 4914 vdev_compact_children(pvd); 4915 4916 /* 4917 * Remember one of the remaining children so we can get tvd below. 4918 */ 4919 cvd = pvd->vdev_child[pvd->vdev_children - 1]; 4920 4921 /* 4922 * If we need to remove the remaining child from the list of hot spares, 4923 * do it now, marking the vdev as no longer a spare in the process. 4924 * We must do this before vdev_remove_parent(), because that can 4925 * change the GUID if it creates a new toplevel GUID. For a similar 4926 * reason, we must remove the spare now, in the same txg as the detach; 4927 * otherwise someone could attach a new sibling, change the GUID, and 4928 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail. 4929 */ 4930 if (unspare) { 4931 ASSERT(cvd->vdev_isspare); 4932 spa_spare_remove(cvd); 4933 unspare_guid = cvd->vdev_guid; 4934 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE); 4935 cvd->vdev_unspare = B_TRUE; 4936 } 4937 4938 /* 4939 * If the parent mirror/replacing vdev only has one child, 4940 * the parent is no longer needed. Remove it from the tree. 4941 */ 4942 if (pvd->vdev_children == 1) { 4943 if (pvd->vdev_ops == &vdev_spare_ops) 4944 cvd->vdev_unspare = B_FALSE; 4945 vdev_remove_parent(cvd); 4946 } 4947 4948 4949 /* 4950 * We don't set tvd until now because the parent we just removed 4951 * may have been the previous top-level vdev. 4952 */ 4953 tvd = cvd->vdev_top; 4954 ASSERT(tvd->vdev_parent == rvd); 4955 4956 /* 4957 * Reevaluate the parent vdev state. 4958 */ 4959 vdev_propagate_state(cvd); 4960 4961 /* 4962 * If the 'autoexpand' property is set on the pool then automatically 4963 * try to expand the size of the pool. For example if the device we 4964 * just detached was smaller than the others, it may be possible to 4965 * add metaslabs (i.e. grow the pool). We need to reopen the vdev 4966 * first so that we can obtain the updated sizes of the leaf vdevs. 4967 */ 4968 if (spa->spa_autoexpand) { 4969 vdev_reopen(tvd); 4970 vdev_expand(tvd, txg); 4971 } 4972 4973 vdev_config_dirty(tvd); 4974 4975 /* 4976 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that 4977 * vd->vdev_detached is set and free vd's DTL object in syncing context. 4978 * But first make sure we're not on any *other* txg's DTL list, to 4979 * prevent vd from being accessed after it's freed. 4980 */ 4981 vdpath = spa_strdup(vd->vdev_path); 4982 for (int t = 0; t < TXG_SIZE; t++) 4983 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t); 4984 vd->vdev_detached = B_TRUE; 4985 vdev_dirty(tvd, VDD_DTL, vd, txg); 4986 4987 spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE); 4988 4989 /* hang on to the spa before we release the lock */ 4990 spa_open_ref(spa, FTAG); 4991 4992 error = spa_vdev_exit(spa, vd, txg, 0); 4993 4994 spa_history_log_internal(spa, "detach", NULL, 4995 "vdev=%s", vdpath); 4996 spa_strfree(vdpath); 4997 4998 /* 4999 * If this was the removal of the original device in a hot spare vdev, 5000 * then we want to go through and remove the device from the hot spare 5001 * list of every other pool. 5002 */ 5003 if (unspare) { 5004 spa_t *altspa = NULL; 5005 5006 mutex_enter(&spa_namespace_lock); 5007 while ((altspa = spa_next(altspa)) != NULL) { 5008 if (altspa->spa_state != POOL_STATE_ACTIVE || 5009 altspa == spa) 5010 continue; 5011 5012 spa_open_ref(altspa, FTAG); 5013 mutex_exit(&spa_namespace_lock); 5014 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE); 5015 mutex_enter(&spa_namespace_lock); 5016 spa_close(altspa, FTAG); 5017 } 5018 mutex_exit(&spa_namespace_lock); 5019 5020 /* search the rest of the vdevs for spares to remove */ 5021 spa_vdev_resilver_done(spa); 5022 } 5023 5024 /* all done with the spa; OK to release */ 5025 mutex_enter(&spa_namespace_lock); 5026 spa_close(spa, FTAG); 5027 mutex_exit(&spa_namespace_lock); 5028 5029 return (error); 5030} 5031 5032/* 5033 * Split a set of devices from their mirrors, and create a new pool from them. 5034 */ 5035int 5036spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config, 5037 nvlist_t *props, boolean_t exp) 5038{ 5039 int error = 0; 5040 uint64_t txg, *glist; 5041 spa_t *newspa; 5042 uint_t c, children, lastlog; 5043 nvlist_t **child, *nvl, *tmp; 5044 dmu_tx_t *tx; 5045 char *altroot = NULL; 5046 vdev_t *rvd, **vml = NULL; /* vdev modify list */ 5047 boolean_t activate_slog; 5048 5049 ASSERT(spa_writeable(spa)); 5050 5051 txg = spa_vdev_enter(spa); 5052 5053 /* clear the log and flush everything up to now */ 5054 activate_slog = spa_passivate_log(spa); 5055 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5056 error = spa_offline_log(spa); 5057 txg = spa_vdev_config_enter(spa); 5058 5059 if (activate_slog) 5060 spa_activate_log(spa); 5061 5062 if (error != 0) 5063 return (spa_vdev_exit(spa, NULL, txg, error)); 5064 5065 /* check new spa name before going any further */ 5066 if (spa_lookup(newname) != NULL) 5067 return (spa_vdev_exit(spa, NULL, txg, EEXIST)); 5068 5069 /* 5070 * scan through all the children to ensure they're all mirrors 5071 */ 5072 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 || 5073 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child, 5074 &children) != 0) 5075 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5076 5077 /* first, check to ensure we've got the right child count */ 5078 rvd = spa->spa_root_vdev; 5079 lastlog = 0; 5080 for (c = 0; c < rvd->vdev_children; c++) { 5081 vdev_t *vd = rvd->vdev_child[c]; 5082 5083 /* don't count the holes & logs as children */ 5084 if (vd->vdev_islog || vd->vdev_ishole) { 5085 if (lastlog == 0) 5086 lastlog = c; 5087 continue; 5088 } 5089 5090 lastlog = 0; 5091 } 5092 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children)) 5093 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5094 5095 /* next, ensure no spare or cache devices are part of the split */ 5096 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 || 5097 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0) 5098 return (spa_vdev_exit(spa, NULL, txg, EINVAL)); 5099 5100 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP); 5101 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP); 5102 5103 /* then, loop over each vdev and validate it */ 5104 for (c = 0; c < children; c++) { 5105 uint64_t is_hole = 0; 5106 5107 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE, 5108 &is_hole); 5109 5110 if (is_hole != 0) { 5111 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole || 5112 spa->spa_root_vdev->vdev_child[c]->vdev_islog) { 5113 continue; 5114 } else { 5115 error = SET_ERROR(EINVAL); 5116 break; 5117 } 5118 } 5119 5120 /* which disk is going to be split? */ 5121 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID, 5122 &glist[c]) != 0) { 5123 error = SET_ERROR(EINVAL); 5124 break; 5125 } 5126 5127 /* look it up in the spa */ 5128 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE); 5129 if (vml[c] == NULL) { 5130 error = SET_ERROR(ENODEV); 5131 break; 5132 } 5133 5134 /* make sure there's nothing stopping the split */ 5135 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops || 5136 vml[c]->vdev_islog || 5137 vml[c]->vdev_ishole || 5138 vml[c]->vdev_isspare || 5139 vml[c]->vdev_isl2cache || 5140 !vdev_writeable(vml[c]) || 5141 vml[c]->vdev_children != 0 || 5142 vml[c]->vdev_state != VDEV_STATE_HEALTHY || 5143 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) { 5144 error = SET_ERROR(EINVAL); 5145 break; 5146 } 5147 5148 if (vdev_dtl_required(vml[c])) { 5149 error = SET_ERROR(EBUSY); 5150 break; 5151 } 5152 5153 /* we need certain info from the top level */ 5154 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY, 5155 vml[c]->vdev_top->vdev_ms_array) == 0); 5156 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT, 5157 vml[c]->vdev_top->vdev_ms_shift) == 0); 5158 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE, 5159 vml[c]->vdev_top->vdev_asize) == 0); 5160 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT, 5161 vml[c]->vdev_top->vdev_ashift) == 0); 5162 } 5163 5164 if (error != 0) { 5165 kmem_free(vml, children * sizeof (vdev_t *)); 5166 kmem_free(glist, children * sizeof (uint64_t)); 5167 return (spa_vdev_exit(spa, NULL, txg, error)); 5168 } 5169 5170 /* stop writers from using the disks */ 5171 for (c = 0; c < children; c++) { 5172 if (vml[c] != NULL) 5173 vml[c]->vdev_offline = B_TRUE; 5174 } 5175 vdev_reopen(spa->spa_root_vdev); 5176 5177 /* 5178 * Temporarily record the splitting vdevs in the spa config. This 5179 * will disappear once the config is regenerated. 5180 */ 5181 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5182 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST, 5183 glist, children) == 0); 5184 kmem_free(glist, children * sizeof (uint64_t)); 5185 5186 mutex_enter(&spa->spa_props_lock); 5187 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT, 5188 nvl) == 0); 5189 mutex_exit(&spa->spa_props_lock); 5190 spa->spa_config_splitting = nvl; 5191 vdev_config_dirty(spa->spa_root_vdev); 5192 5193 /* configure and create the new pool */ 5194 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0); 5195 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 5196 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0); 5197 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 5198 spa_version(spa)) == 0); 5199 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 5200 spa->spa_config_txg) == 0); 5201 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 5202 spa_generate_guid(NULL)) == 0); 5203 (void) nvlist_lookup_string(props, 5204 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot); 5205 5206 /* add the new pool to the namespace */ 5207 newspa = spa_add(newname, config, altroot); 5208 newspa->spa_config_txg = spa->spa_config_txg; 5209 spa_set_log_state(newspa, SPA_LOG_CLEAR); 5210 5211 /* release the spa config lock, retaining the namespace lock */ 5212 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5213 5214 if (zio_injection_enabled) 5215 zio_handle_panic_injection(spa, FTAG, 1); 5216 5217 spa_activate(newspa, spa_mode_global); 5218 spa_async_suspend(newspa); 5219 5220#ifndef sun 5221 /* mark that we are creating new spa by splitting */ 5222 newspa->spa_splitting_newspa = B_TRUE; 5223#endif 5224 /* create the new pool from the disks of the original pool */ 5225 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE); 5226#ifndef sun 5227 newspa->spa_splitting_newspa = B_FALSE; 5228#endif 5229 if (error) 5230 goto out; 5231 5232 /* if that worked, generate a real config for the new pool */ 5233 if (newspa->spa_root_vdev != NULL) { 5234 VERIFY(nvlist_alloc(&newspa->spa_config_splitting, 5235 NV_UNIQUE_NAME, KM_SLEEP) == 0); 5236 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting, 5237 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0); 5238 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL, 5239 B_TRUE)); 5240 } 5241 5242 /* set the props */ 5243 if (props != NULL) { 5244 spa_configfile_set(newspa, props, B_FALSE); 5245 error = spa_prop_set(newspa, props); 5246 if (error) 5247 goto out; 5248 } 5249 5250 /* flush everything */ 5251 txg = spa_vdev_config_enter(newspa); 5252 vdev_config_dirty(newspa->spa_root_vdev); 5253 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG); 5254 5255 if (zio_injection_enabled) 5256 zio_handle_panic_injection(spa, FTAG, 2); 5257 5258 spa_async_resume(newspa); 5259 5260 /* finally, update the original pool's config */ 5261 txg = spa_vdev_config_enter(spa); 5262 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir); 5263 error = dmu_tx_assign(tx, TXG_WAIT); 5264 if (error != 0) 5265 dmu_tx_abort(tx); 5266 for (c = 0; c < children; c++) { 5267 if (vml[c] != NULL) { 5268 vdev_split(vml[c]); 5269 if (error == 0) 5270 spa_history_log_internal(spa, "detach", tx, 5271 "vdev=%s", vml[c]->vdev_path); 5272 vdev_free(vml[c]); 5273 } 5274 } 5275 vdev_config_dirty(spa->spa_root_vdev); 5276 spa->spa_config_splitting = NULL; 5277 nvlist_free(nvl); 5278 if (error == 0) 5279 dmu_tx_commit(tx); 5280 (void) spa_vdev_exit(spa, NULL, txg, 0); 5281 5282 if (zio_injection_enabled) 5283 zio_handle_panic_injection(spa, FTAG, 3); 5284 5285 /* split is complete; log a history record */ 5286 spa_history_log_internal(newspa, "split", NULL, 5287 "from pool %s", spa_name(spa)); 5288 5289 kmem_free(vml, children * sizeof (vdev_t *)); 5290 5291 /* if we're not going to mount the filesystems in userland, export */ 5292 if (exp) 5293 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL, 5294 B_FALSE, B_FALSE); 5295 5296 return (error); 5297 5298out: 5299 spa_unload(newspa); 5300 spa_deactivate(newspa); 5301 spa_remove(newspa); 5302 5303 txg = spa_vdev_config_enter(spa); 5304 5305 /* re-online all offlined disks */ 5306 for (c = 0; c < children; c++) { 5307 if (vml[c] != NULL) 5308 vml[c]->vdev_offline = B_FALSE; 5309 } 5310 vdev_reopen(spa->spa_root_vdev); 5311 5312 nvlist_free(spa->spa_config_splitting); 5313 spa->spa_config_splitting = NULL; 5314 (void) spa_vdev_exit(spa, NULL, txg, error); 5315 5316 kmem_free(vml, children * sizeof (vdev_t *)); 5317 return (error); 5318} 5319 5320static nvlist_t * 5321spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid) 5322{ 5323 for (int i = 0; i < count; i++) { 5324 uint64_t guid; 5325 5326 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID, 5327 &guid) == 0); 5328 5329 if (guid == target_guid) 5330 return (nvpp[i]); 5331 } 5332 5333 return (NULL); 5334} 5335 5336static void 5337spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count, 5338 nvlist_t *dev_to_remove) 5339{ 5340 nvlist_t **newdev = NULL; 5341 5342 if (count > 1) 5343 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP); 5344 5345 for (int i = 0, j = 0; i < count; i++) { 5346 if (dev[i] == dev_to_remove) 5347 continue; 5348 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0); 5349 } 5350 5351 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0); 5352 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0); 5353 5354 for (int i = 0; i < count - 1; i++) 5355 nvlist_free(newdev[i]); 5356 5357 if (count > 1) 5358 kmem_free(newdev, (count - 1) * sizeof (void *)); 5359} 5360 5361/* 5362 * Evacuate the device. 5363 */ 5364static int 5365spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd) 5366{ 5367 uint64_t txg; 5368 int error = 0; 5369 5370 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5371 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5372 ASSERT(vd == vd->vdev_top); 5373 5374 /* 5375 * Evacuate the device. We don't hold the config lock as writer 5376 * since we need to do I/O but we do keep the 5377 * spa_namespace_lock held. Once this completes the device 5378 * should no longer have any blocks allocated on it. 5379 */ 5380 if (vd->vdev_islog) { 5381 if (vd->vdev_stat.vs_alloc != 0) 5382 error = spa_offline_log(spa); 5383 } else { 5384 error = SET_ERROR(ENOTSUP); 5385 } 5386 5387 if (error) 5388 return (error); 5389 5390 /* 5391 * The evacuation succeeded. Remove any remaining MOS metadata 5392 * associated with this vdev, and wait for these changes to sync. 5393 */ 5394 ASSERT0(vd->vdev_stat.vs_alloc); 5395 txg = spa_vdev_config_enter(spa); 5396 vd->vdev_removing = B_TRUE; 5397 vdev_dirty(vd, 0, NULL, txg); 5398 vdev_config_dirty(vd); 5399 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG); 5400 5401 return (0); 5402} 5403 5404/* 5405 * Complete the removal by cleaning up the namespace. 5406 */ 5407static void 5408spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd) 5409{ 5410 vdev_t *rvd = spa->spa_root_vdev; 5411 uint64_t id = vd->vdev_id; 5412 boolean_t last_vdev = (id == (rvd->vdev_children - 1)); 5413 5414 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 5415 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 5416 ASSERT(vd == vd->vdev_top); 5417 5418 /* 5419 * Only remove any devices which are empty. 5420 */ 5421 if (vd->vdev_stat.vs_alloc != 0) 5422 return; 5423 5424 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE); 5425 5426 if (list_link_active(&vd->vdev_state_dirty_node)) 5427 vdev_state_clean(vd); 5428 if (list_link_active(&vd->vdev_config_dirty_node)) 5429 vdev_config_clean(vd); 5430 5431 vdev_free(vd); 5432 5433 if (last_vdev) { 5434 vdev_compact_children(rvd); 5435 } else { 5436 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops); 5437 vdev_add_child(rvd, vd); 5438 } 5439 vdev_config_dirty(rvd); 5440 5441 /* 5442 * Reassess the health of our root vdev. 5443 */ 5444 vdev_reopen(rvd); 5445} 5446 5447/* 5448 * Remove a device from the pool - 5449 * 5450 * Removing a device from the vdev namespace requires several steps 5451 * and can take a significant amount of time. As a result we use 5452 * the spa_vdev_config_[enter/exit] functions which allow us to 5453 * grab and release the spa_config_lock while still holding the namespace 5454 * lock. During each step the configuration is synced out. 5455 * 5456 * Currently, this supports removing only hot spares, slogs, and level 2 ARC 5457 * devices. 5458 */ 5459int 5460spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare) 5461{ 5462 vdev_t *vd; 5463 metaslab_group_t *mg; 5464 nvlist_t **spares, **l2cache, *nv; 5465 uint64_t txg = 0; 5466 uint_t nspares, nl2cache; 5467 int error = 0; 5468 boolean_t locked = MUTEX_HELD(&spa_namespace_lock); 5469 5470 ASSERT(spa_writeable(spa)); 5471 5472 if (!locked) 5473 txg = spa_vdev_enter(spa); 5474 5475 vd = spa_lookup_by_guid(spa, guid, B_FALSE); 5476 5477 if (spa->spa_spares.sav_vdevs != NULL && 5478 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config, 5479 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 && 5480 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) { 5481 /* 5482 * Only remove the hot spare if it's not currently in use 5483 * in this pool. 5484 */ 5485 if (vd == NULL || unspare) { 5486 spa_vdev_remove_aux(spa->spa_spares.sav_config, 5487 ZPOOL_CONFIG_SPARES, spares, nspares, nv); 5488 spa_load_spares(spa); 5489 spa->spa_spares.sav_sync = B_TRUE; 5490 } else { 5491 error = SET_ERROR(EBUSY); 5492 } 5493 } else if (spa->spa_l2cache.sav_vdevs != NULL && 5494 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config, 5495 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 && 5496 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) { 5497 /* 5498 * Cache devices can always be removed. 5499 */ 5500 spa_vdev_remove_aux(spa->spa_l2cache.sav_config, 5501 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv); 5502 spa_load_l2cache(spa); 5503 spa->spa_l2cache.sav_sync = B_TRUE; 5504 } else if (vd != NULL && vd->vdev_islog) { 5505 ASSERT(!locked); 5506 ASSERT(vd == vd->vdev_top); 5507 5508 /* 5509 * XXX - Once we have bp-rewrite this should 5510 * become the common case. 5511 */ 5512 5513 mg = vd->vdev_mg; 5514 5515 /* 5516 * Stop allocating from this vdev. 5517 */ 5518 metaslab_group_passivate(mg); 5519 5520 /* 5521 * Wait for the youngest allocations and frees to sync, 5522 * and then wait for the deferral of those frees to finish. 5523 */ 5524 spa_vdev_config_exit(spa, NULL, 5525 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG); 5526 5527 /* 5528 * Attempt to evacuate the vdev. 5529 */ 5530 error = spa_vdev_remove_evacuate(spa, vd); 5531 5532 txg = spa_vdev_config_enter(spa); 5533 5534 /* 5535 * If we couldn't evacuate the vdev, unwind. 5536 */ 5537 if (error) { 5538 metaslab_group_activate(mg); 5539 return (spa_vdev_exit(spa, NULL, txg, error)); 5540 } 5541 5542 /* 5543 * Clean up the vdev namespace. 5544 */ 5545 spa_vdev_remove_from_namespace(spa, vd); 5546 5547 } else if (vd != NULL) { 5548 /* 5549 * Normal vdevs cannot be removed (yet). 5550 */ 5551 error = SET_ERROR(ENOTSUP); 5552 } else { 5553 /* 5554 * There is no vdev of any kind with the specified guid. 5555 */ 5556 error = SET_ERROR(ENOENT); 5557 } 5558 5559 if (!locked) 5560 return (spa_vdev_exit(spa, NULL, txg, error)); 5561 5562 return (error); 5563} 5564 5565/* 5566 * Find any device that's done replacing, or a vdev marked 'unspare' that's 5567 * currently spared, so we can detach it. 5568 */ 5569static vdev_t * 5570spa_vdev_resilver_done_hunt(vdev_t *vd) 5571{ 5572 vdev_t *newvd, *oldvd; 5573 5574 for (int c = 0; c < vd->vdev_children; c++) { 5575 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]); 5576 if (oldvd != NULL) 5577 return (oldvd); 5578 } 5579 5580 /* 5581 * Check for a completed replacement. We always consider the first 5582 * vdev in the list to be the oldest vdev, and the last one to be 5583 * the newest (see spa_vdev_attach() for how that works). In 5584 * the case where the newest vdev is faulted, we will not automatically 5585 * remove it after a resilver completes. This is OK as it will require 5586 * user intervention to determine which disk the admin wishes to keep. 5587 */ 5588 if (vd->vdev_ops == &vdev_replacing_ops) { 5589 ASSERT(vd->vdev_children > 1); 5590 5591 newvd = vd->vdev_child[vd->vdev_children - 1]; 5592 oldvd = vd->vdev_child[0]; 5593 5594 if (vdev_dtl_empty(newvd, DTL_MISSING) && 5595 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5596 !vdev_dtl_required(oldvd)) 5597 return (oldvd); 5598 } 5599 5600 /* 5601 * Check for a completed resilver with the 'unspare' flag set. 5602 */ 5603 if (vd->vdev_ops == &vdev_spare_ops) { 5604 vdev_t *first = vd->vdev_child[0]; 5605 vdev_t *last = vd->vdev_child[vd->vdev_children - 1]; 5606 5607 if (last->vdev_unspare) { 5608 oldvd = first; 5609 newvd = last; 5610 } else if (first->vdev_unspare) { 5611 oldvd = last; 5612 newvd = first; 5613 } else { 5614 oldvd = NULL; 5615 } 5616 5617 if (oldvd != NULL && 5618 vdev_dtl_empty(newvd, DTL_MISSING) && 5619 vdev_dtl_empty(newvd, DTL_OUTAGE) && 5620 !vdev_dtl_required(oldvd)) 5621 return (oldvd); 5622 5623 /* 5624 * If there are more than two spares attached to a disk, 5625 * and those spares are not required, then we want to 5626 * attempt to free them up now so that they can be used 5627 * by other pools. Once we're back down to a single 5628 * disk+spare, we stop removing them. 5629 */ 5630 if (vd->vdev_children > 2) { 5631 newvd = vd->vdev_child[1]; 5632 5633 if (newvd->vdev_isspare && last->vdev_isspare && 5634 vdev_dtl_empty(last, DTL_MISSING) && 5635 vdev_dtl_empty(last, DTL_OUTAGE) && 5636 !vdev_dtl_required(newvd)) 5637 return (newvd); 5638 } 5639 } 5640 5641 return (NULL); 5642} 5643 5644static void 5645spa_vdev_resilver_done(spa_t *spa) 5646{ 5647 vdev_t *vd, *pvd, *ppvd; 5648 uint64_t guid, sguid, pguid, ppguid; 5649 5650 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5651 5652 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) { 5653 pvd = vd->vdev_parent; 5654 ppvd = pvd->vdev_parent; 5655 guid = vd->vdev_guid; 5656 pguid = pvd->vdev_guid; 5657 ppguid = ppvd->vdev_guid; 5658 sguid = 0; 5659 /* 5660 * If we have just finished replacing a hot spared device, then 5661 * we need to detach the parent's first child (the original hot 5662 * spare) as well. 5663 */ 5664 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 && 5665 ppvd->vdev_children == 2) { 5666 ASSERT(pvd->vdev_ops == &vdev_replacing_ops); 5667 sguid = ppvd->vdev_child[1]->vdev_guid; 5668 } 5669 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd)); 5670 5671 spa_config_exit(spa, SCL_ALL, FTAG); 5672 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0) 5673 return; 5674 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0) 5675 return; 5676 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 5677 } 5678 5679 spa_config_exit(spa, SCL_ALL, FTAG); 5680} 5681 5682/* 5683 * Update the stored path or FRU for this vdev. 5684 */ 5685int 5686spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value, 5687 boolean_t ispath) 5688{ 5689 vdev_t *vd; 5690 boolean_t sync = B_FALSE; 5691 5692 ASSERT(spa_writeable(spa)); 5693 5694 spa_vdev_state_enter(spa, SCL_ALL); 5695 5696 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 5697 return (spa_vdev_state_exit(spa, NULL, ENOENT)); 5698 5699 if (!vd->vdev_ops->vdev_op_leaf) 5700 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 5701 5702 if (ispath) { 5703 if (strcmp(value, vd->vdev_path) != 0) { 5704 spa_strfree(vd->vdev_path); 5705 vd->vdev_path = spa_strdup(value); 5706 sync = B_TRUE; 5707 } 5708 } else { 5709 if (vd->vdev_fru == NULL) { 5710 vd->vdev_fru = spa_strdup(value); 5711 sync = B_TRUE; 5712 } else if (strcmp(value, vd->vdev_fru) != 0) { 5713 spa_strfree(vd->vdev_fru); 5714 vd->vdev_fru = spa_strdup(value); 5715 sync = B_TRUE; 5716 } 5717 } 5718 5719 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0)); 5720} 5721 5722int 5723spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath) 5724{ 5725 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE)); 5726} 5727 5728int 5729spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru) 5730{ 5731 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE)); 5732} 5733 5734/* 5735 * ========================================================================== 5736 * SPA Scanning 5737 * ========================================================================== 5738 */ 5739 5740int 5741spa_scan_stop(spa_t *spa) 5742{ 5743 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5744 if (dsl_scan_resilvering(spa->spa_dsl_pool)) 5745 return (SET_ERROR(EBUSY)); 5746 return (dsl_scan_cancel(spa->spa_dsl_pool)); 5747} 5748 5749int 5750spa_scan(spa_t *spa, pool_scan_func_t func) 5751{ 5752 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0); 5753 5754 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE) 5755 return (SET_ERROR(ENOTSUP)); 5756 5757 /* 5758 * If a resilver was requested, but there is no DTL on a 5759 * writeable leaf device, we have nothing to do. 5760 */ 5761 if (func == POOL_SCAN_RESILVER && 5762 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) { 5763 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); 5764 return (0); 5765 } 5766 5767 return (dsl_scan(spa->spa_dsl_pool, func)); 5768} 5769 5770/* 5771 * ========================================================================== 5772 * SPA async task processing 5773 * ========================================================================== 5774 */ 5775 5776static void 5777spa_async_remove(spa_t *spa, vdev_t *vd) 5778{ 5779 if (vd->vdev_remove_wanted) { 5780 vd->vdev_remove_wanted = B_FALSE; 5781 vd->vdev_delayed_close = B_FALSE; 5782 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE); 5783 5784 /* 5785 * We want to clear the stats, but we don't want to do a full 5786 * vdev_clear() as that will cause us to throw away 5787 * degraded/faulted state as well as attempt to reopen the 5788 * device, all of which is a waste. 5789 */ 5790 vd->vdev_stat.vs_read_errors = 0; 5791 vd->vdev_stat.vs_write_errors = 0; 5792 vd->vdev_stat.vs_checksum_errors = 0; 5793 5794 vdev_state_dirty(vd->vdev_top); 5795 } 5796 5797 for (int c = 0; c < vd->vdev_children; c++) 5798 spa_async_remove(spa, vd->vdev_child[c]); 5799} 5800 5801static void 5802spa_async_probe(spa_t *spa, vdev_t *vd) 5803{ 5804 if (vd->vdev_probe_wanted) { 5805 vd->vdev_probe_wanted = B_FALSE; 5806 vdev_reopen(vd); /* vdev_open() does the actual probe */ 5807 } 5808 5809 for (int c = 0; c < vd->vdev_children; c++) 5810 spa_async_probe(spa, vd->vdev_child[c]); 5811} 5812 5813static void 5814spa_async_autoexpand(spa_t *spa, vdev_t *vd) 5815{ 5816 sysevent_id_t eid; 5817 nvlist_t *attr; 5818 char *physpath; 5819 5820 if (!spa->spa_autoexpand) 5821 return; 5822 5823 for (int c = 0; c < vd->vdev_children; c++) { 5824 vdev_t *cvd = vd->vdev_child[c]; 5825 spa_async_autoexpand(spa, cvd); 5826 } 5827 5828 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL) 5829 return; 5830 5831 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 5832 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath); 5833 5834 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0); 5835 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0); 5836 5837 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS, 5838 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP); 5839 5840 nvlist_free(attr); 5841 kmem_free(physpath, MAXPATHLEN); 5842} 5843 5844static void 5845spa_async_thread(void *arg) 5846{ 5847 spa_t *spa = arg; 5848 int tasks; 5849 5850 ASSERT(spa->spa_sync_on); 5851 5852 mutex_enter(&spa->spa_async_lock); 5853 tasks = spa->spa_async_tasks; 5854 spa->spa_async_tasks &= SPA_ASYNC_REMOVE; 5855 mutex_exit(&spa->spa_async_lock); 5856 5857 /* 5858 * See if the config needs to be updated. 5859 */ 5860 if (tasks & SPA_ASYNC_CONFIG_UPDATE) { 5861 uint64_t old_space, new_space; 5862 5863 mutex_enter(&spa_namespace_lock); 5864 old_space = metaslab_class_get_space(spa_normal_class(spa)); 5865 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL); 5866 new_space = metaslab_class_get_space(spa_normal_class(spa)); 5867 mutex_exit(&spa_namespace_lock); 5868 5869 /* 5870 * If the pool grew as a result of the config update, 5871 * then log an internal history event. 5872 */ 5873 if (new_space != old_space) { 5874 spa_history_log_internal(spa, "vdev online", NULL, 5875 "pool '%s' size: %llu(+%llu)", 5876 spa_name(spa), new_space, new_space - old_space); 5877 } 5878 } 5879 5880 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) { 5881 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 5882 spa_async_autoexpand(spa, spa->spa_root_vdev); 5883 spa_config_exit(spa, SCL_CONFIG, FTAG); 5884 } 5885 5886 /* 5887 * See if any devices need to be probed. 5888 */ 5889 if (tasks & SPA_ASYNC_PROBE) { 5890 spa_vdev_state_enter(spa, SCL_NONE); 5891 spa_async_probe(spa, spa->spa_root_vdev); 5892 (void) spa_vdev_state_exit(spa, NULL, 0); 5893 } 5894 5895 /* 5896 * If any devices are done replacing, detach them. 5897 */ 5898 if (tasks & SPA_ASYNC_RESILVER_DONE) 5899 spa_vdev_resilver_done(spa); 5900 5901 /* 5902 * Kick off a resilver. 5903 */ 5904 if (tasks & SPA_ASYNC_RESILVER) 5905 dsl_resilver_restart(spa->spa_dsl_pool, 0); 5906 5907 /* 5908 * Let the world know that we're done. 5909 */ 5910 mutex_enter(&spa->spa_async_lock); 5911 spa->spa_async_thread = NULL; 5912 cv_broadcast(&spa->spa_async_cv); 5913 mutex_exit(&spa->spa_async_lock); 5914 thread_exit(); 5915} 5916 5917static void 5918spa_async_thread_vd(void *arg) 5919{ 5920 spa_t *spa = arg; 5921 int tasks; 5922 5923 ASSERT(spa->spa_sync_on); 5924 5925 mutex_enter(&spa->spa_async_lock); 5926 tasks = spa->spa_async_tasks; 5927retry: 5928 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE; 5929 mutex_exit(&spa->spa_async_lock); 5930 5931 /* 5932 * See if any devices need to be marked REMOVED. 5933 */ 5934 if (tasks & SPA_ASYNC_REMOVE) { 5935 spa_vdev_state_enter(spa, SCL_NONE); 5936 spa_async_remove(spa, spa->spa_root_vdev); 5937 for (int i = 0; i < spa->spa_l2cache.sav_count; i++) 5938 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]); 5939 for (int i = 0; i < spa->spa_spares.sav_count; i++) 5940 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]); 5941 (void) spa_vdev_state_exit(spa, NULL, 0); 5942 } 5943 5944 /* 5945 * Let the world know that we're done. 5946 */ 5947 mutex_enter(&spa->spa_async_lock); 5948 tasks = spa->spa_async_tasks; 5949 if ((tasks & SPA_ASYNC_REMOVE) != 0) 5950 goto retry; 5951 spa->spa_async_thread_vd = NULL; 5952 cv_broadcast(&spa->spa_async_cv); 5953 mutex_exit(&spa->spa_async_lock); 5954 thread_exit(); 5955} 5956 5957void 5958spa_async_suspend(spa_t *spa) 5959{ 5960 mutex_enter(&spa->spa_async_lock); 5961 spa->spa_async_suspended++; 5962 while (spa->spa_async_thread != NULL && 5963 spa->spa_async_thread_vd != NULL) 5964 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock); 5965 mutex_exit(&spa->spa_async_lock); 5966} 5967 5968void 5969spa_async_resume(spa_t *spa) 5970{ 5971 mutex_enter(&spa->spa_async_lock); 5972 ASSERT(spa->spa_async_suspended != 0); 5973 spa->spa_async_suspended--; 5974 mutex_exit(&spa->spa_async_lock); 5975} 5976 5977static boolean_t 5978spa_async_tasks_pending(spa_t *spa) 5979{ 5980 uint_t non_config_tasks; 5981 uint_t config_task; 5982 boolean_t config_task_suspended; 5983 5984 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE | 5985 SPA_ASYNC_REMOVE); 5986 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE; 5987 if (spa->spa_ccw_fail_time == 0) { 5988 config_task_suspended = B_FALSE; 5989 } else { 5990 config_task_suspended = 5991 (gethrtime() - spa->spa_ccw_fail_time) < 5992 (zfs_ccw_retry_interval * NANOSEC); 5993 } 5994 5995 return (non_config_tasks || (config_task && !config_task_suspended)); 5996} 5997 5998static void 5999spa_async_dispatch(spa_t *spa) 6000{ 6001 mutex_enter(&spa->spa_async_lock); 6002 if (spa_async_tasks_pending(spa) && 6003 !spa->spa_async_suspended && 6004 spa->spa_async_thread == NULL && 6005 rootdir != NULL) 6006 spa->spa_async_thread = thread_create(NULL, 0, 6007 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri); 6008 mutex_exit(&spa->spa_async_lock); 6009} 6010 6011static void 6012spa_async_dispatch_vd(spa_t *spa) 6013{ 6014 mutex_enter(&spa->spa_async_lock); 6015 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 && 6016 !spa->spa_async_suspended && 6017 spa->spa_async_thread_vd == NULL && 6018 rootdir != NULL) 6019 spa->spa_async_thread_vd = thread_create(NULL, 0, 6020 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri); 6021 mutex_exit(&spa->spa_async_lock); 6022} 6023 6024void 6025spa_async_request(spa_t *spa, int task) 6026{ 6027 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task); 6028 mutex_enter(&spa->spa_async_lock); 6029 spa->spa_async_tasks |= task; 6030 mutex_exit(&spa->spa_async_lock); 6031 spa_async_dispatch_vd(spa); 6032} 6033 6034/* 6035 * ========================================================================== 6036 * SPA syncing routines 6037 * ========================================================================== 6038 */ 6039 6040static int 6041bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6042{ 6043 bpobj_t *bpo = arg; 6044 bpobj_enqueue(bpo, bp, tx); 6045 return (0); 6046} 6047 6048static int 6049spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) 6050{ 6051 zio_t *zio = arg; 6052 6053 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp, 6054 BP_GET_PSIZE(bp), zio->io_flags)); 6055 return (0); 6056} 6057 6058/* 6059 * Note: this simple function is not inlined to make it easier to dtrace the 6060 * amount of time spent syncing frees. 6061 */ 6062static void 6063spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx) 6064{ 6065 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6066 bplist_iterate(bpl, spa_free_sync_cb, zio, tx); 6067 VERIFY(zio_wait(zio) == 0); 6068} 6069 6070/* 6071 * Note: this simple function is not inlined to make it easier to dtrace the 6072 * amount of time spent syncing deferred frees. 6073 */ 6074static void 6075spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx) 6076{ 6077 zio_t *zio = zio_root(spa, NULL, NULL, 0); 6078 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj, 6079 spa_free_sync_cb, zio, tx), ==, 0); 6080 VERIFY0(zio_wait(zio)); 6081} 6082 6083 6084static void 6085spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx) 6086{ 6087 char *packed = NULL; 6088 size_t bufsize; 6089 size_t nvsize = 0; 6090 dmu_buf_t *db; 6091 6092 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0); 6093 6094 /* 6095 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration 6096 * information. This avoids the dbuf_will_dirty() path and 6097 * saves us a pre-read to get data we don't actually care about. 6098 */ 6099 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE); 6100 packed = kmem_alloc(bufsize, KM_SLEEP); 6101 6102 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR, 6103 KM_SLEEP) == 0); 6104 bzero(packed + nvsize, bufsize - nvsize); 6105 6106 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx); 6107 6108 kmem_free(packed, bufsize); 6109 6110 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db)); 6111 dmu_buf_will_dirty(db, tx); 6112 *(uint64_t *)db->db_data = nvsize; 6113 dmu_buf_rele(db, FTAG); 6114} 6115 6116static void 6117spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx, 6118 const char *config, const char *entry) 6119{ 6120 nvlist_t *nvroot; 6121 nvlist_t **list; 6122 int i; 6123 6124 if (!sav->sav_sync) 6125 return; 6126 6127 /* 6128 * Update the MOS nvlist describing the list of available devices. 6129 * spa_validate_aux() will have already made sure this nvlist is 6130 * valid and the vdevs are labeled appropriately. 6131 */ 6132 if (sav->sav_object == 0) { 6133 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset, 6134 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE, 6135 sizeof (uint64_t), tx); 6136 VERIFY(zap_update(spa->spa_meta_objset, 6137 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1, 6138 &sav->sav_object, tx) == 0); 6139 } 6140 6141 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0); 6142 if (sav->sav_count == 0) { 6143 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0); 6144 } else { 6145 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP); 6146 for (i = 0; i < sav->sav_count; i++) 6147 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i], 6148 B_FALSE, VDEV_CONFIG_L2CACHE); 6149 VERIFY(nvlist_add_nvlist_array(nvroot, config, list, 6150 sav->sav_count) == 0); 6151 for (i = 0; i < sav->sav_count; i++) 6152 nvlist_free(list[i]); 6153 kmem_free(list, sav->sav_count * sizeof (void *)); 6154 } 6155 6156 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx); 6157 nvlist_free(nvroot); 6158 6159 sav->sav_sync = B_FALSE; 6160} 6161 6162static void 6163spa_sync_config_object(spa_t *spa, dmu_tx_t *tx) 6164{ 6165 nvlist_t *config; 6166 6167 if (list_is_empty(&spa->spa_config_dirty_list)) 6168 return; 6169 6170 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6171 6172 config = spa_config_generate(spa, spa->spa_root_vdev, 6173 dmu_tx_get_txg(tx), B_FALSE); 6174 6175 /* 6176 * If we're upgrading the spa version then make sure that 6177 * the config object gets updated with the correct version. 6178 */ 6179 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version) 6180 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 6181 spa->spa_uberblock.ub_version); 6182 6183 spa_config_exit(spa, SCL_STATE, FTAG); 6184 6185 if (spa->spa_config_syncing) 6186 nvlist_free(spa->spa_config_syncing); 6187 spa->spa_config_syncing = config; 6188 6189 spa_sync_nvlist(spa, spa->spa_config_object, config, tx); 6190} 6191 6192static void 6193spa_sync_version(void *arg, dmu_tx_t *tx) 6194{ 6195 uint64_t *versionp = arg; 6196 uint64_t version = *versionp; 6197 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6198 6199 /* 6200 * Setting the version is special cased when first creating the pool. 6201 */ 6202 ASSERT(tx->tx_txg != TXG_INITIAL); 6203 6204 ASSERT(SPA_VERSION_IS_SUPPORTED(version)); 6205 ASSERT(version >= spa_version(spa)); 6206 6207 spa->spa_uberblock.ub_version = version; 6208 vdev_config_dirty(spa->spa_root_vdev); 6209 spa_history_log_internal(spa, "set", tx, "version=%lld", version); 6210} 6211 6212/* 6213 * Set zpool properties. 6214 */ 6215static void 6216spa_sync_props(void *arg, dmu_tx_t *tx) 6217{ 6218 nvlist_t *nvp = arg; 6219 spa_t *spa = dmu_tx_pool(tx)->dp_spa; 6220 objset_t *mos = spa->spa_meta_objset; 6221 nvpair_t *elem = NULL; 6222 6223 mutex_enter(&spa->spa_props_lock); 6224 6225 while ((elem = nvlist_next_nvpair(nvp, elem))) { 6226 uint64_t intval; 6227 char *strval, *fname; 6228 zpool_prop_t prop; 6229 const char *propname; 6230 zprop_type_t proptype; 6231 zfeature_info_t *feature; 6232 6233 switch (prop = zpool_name_to_prop(nvpair_name(elem))) { 6234 case ZPROP_INVAL: 6235 /* 6236 * We checked this earlier in spa_prop_validate(). 6237 */ 6238 ASSERT(zpool_prop_feature(nvpair_name(elem))); 6239 6240 fname = strchr(nvpair_name(elem), '@') + 1; 6241 VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature)); 6242 6243 spa_feature_enable(spa, feature, tx); 6244 spa_history_log_internal(spa, "set", tx, 6245 "%s=enabled", nvpair_name(elem)); 6246 break; 6247 6248 case ZPOOL_PROP_VERSION: 6249 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6250 /* 6251 * The version is synced seperatly before other 6252 * properties and should be correct by now. 6253 */ 6254 ASSERT3U(spa_version(spa), >=, intval); 6255 break; 6256 6257 case ZPOOL_PROP_ALTROOT: 6258 /* 6259 * 'altroot' is a non-persistent property. It should 6260 * have been set temporarily at creation or import time. 6261 */ 6262 ASSERT(spa->spa_root != NULL); 6263 break; 6264 6265 case ZPOOL_PROP_READONLY: 6266 case ZPOOL_PROP_CACHEFILE: 6267 /* 6268 * 'readonly' and 'cachefile' are also non-persisitent 6269 * properties. 6270 */ 6271 break; 6272 case ZPOOL_PROP_COMMENT: 6273 VERIFY(nvpair_value_string(elem, &strval) == 0); 6274 if (spa->spa_comment != NULL) 6275 spa_strfree(spa->spa_comment); 6276 spa->spa_comment = spa_strdup(strval); 6277 /* 6278 * We need to dirty the configuration on all the vdevs 6279 * so that their labels get updated. It's unnecessary 6280 * to do this for pool creation since the vdev's 6281 * configuratoin has already been dirtied. 6282 */ 6283 if (tx->tx_txg != TXG_INITIAL) 6284 vdev_config_dirty(spa->spa_root_vdev); 6285 spa_history_log_internal(spa, "set", tx, 6286 "%s=%s", nvpair_name(elem), strval); 6287 break; 6288 default: 6289 /* 6290 * Set pool property values in the poolprops mos object. 6291 */ 6292 if (spa->spa_pool_props_object == 0) { 6293 spa->spa_pool_props_object = 6294 zap_create_link(mos, DMU_OT_POOL_PROPS, 6295 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS, 6296 tx); 6297 } 6298 6299 /* normalize the property name */ 6300 propname = zpool_prop_to_name(prop); 6301 proptype = zpool_prop_get_type(prop); 6302 6303 if (nvpair_type(elem) == DATA_TYPE_STRING) { 6304 ASSERT(proptype == PROP_TYPE_STRING); 6305 VERIFY(nvpair_value_string(elem, &strval) == 0); 6306 VERIFY(zap_update(mos, 6307 spa->spa_pool_props_object, propname, 6308 1, strlen(strval) + 1, strval, tx) == 0); 6309 spa_history_log_internal(spa, "set", tx, 6310 "%s=%s", nvpair_name(elem), strval); 6311 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) { 6312 VERIFY(nvpair_value_uint64(elem, &intval) == 0); 6313 6314 if (proptype == PROP_TYPE_INDEX) { 6315 const char *unused; 6316 VERIFY(zpool_prop_index_to_string( 6317 prop, intval, &unused) == 0); 6318 } 6319 VERIFY(zap_update(mos, 6320 spa->spa_pool_props_object, propname, 6321 8, 1, &intval, tx) == 0); 6322 spa_history_log_internal(spa, "set", tx, 6323 "%s=%lld", nvpair_name(elem), intval); 6324 } else { 6325 ASSERT(0); /* not allowed */ 6326 } 6327 6328 switch (prop) { 6329 case ZPOOL_PROP_DELEGATION: 6330 spa->spa_delegation = intval; 6331 break; 6332 case ZPOOL_PROP_BOOTFS: 6333 spa->spa_bootfs = intval; 6334 break; 6335 case ZPOOL_PROP_FAILUREMODE: 6336 spa->spa_failmode = intval; 6337 break; 6338 case ZPOOL_PROP_AUTOEXPAND: 6339 spa->spa_autoexpand = intval; 6340 if (tx->tx_txg != TXG_INITIAL) 6341 spa_async_request(spa, 6342 SPA_ASYNC_AUTOEXPAND); 6343 break; 6344 case ZPOOL_PROP_DEDUPDITTO: 6345 spa->spa_dedup_ditto = intval; 6346 break; 6347 default: 6348 break; 6349 } 6350 } 6351 6352 } 6353 6354 mutex_exit(&spa->spa_props_lock); 6355} 6356 6357/* 6358 * Perform one-time upgrade on-disk changes. spa_version() does not 6359 * reflect the new version this txg, so there must be no changes this 6360 * txg to anything that the upgrade code depends on after it executes. 6361 * Therefore this must be called after dsl_pool_sync() does the sync 6362 * tasks. 6363 */ 6364static void 6365spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx) 6366{ 6367 dsl_pool_t *dp = spa->spa_dsl_pool; 6368 6369 ASSERT(spa->spa_sync_pass == 1); 6370 6371 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 6372 6373 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN && 6374 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) { 6375 dsl_pool_create_origin(dp, tx); 6376 6377 /* Keeping the origin open increases spa_minref */ 6378 spa->spa_minref += 3; 6379 } 6380 6381 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES && 6382 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) { 6383 dsl_pool_upgrade_clones(dp, tx); 6384 } 6385 6386 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES && 6387 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) { 6388 dsl_pool_upgrade_dir_clones(dp, tx); 6389 6390 /* Keeping the freedir open increases spa_minref */ 6391 spa->spa_minref += 3; 6392 } 6393 6394 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES && 6395 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) { 6396 spa_feature_create_zap_objects(spa, tx); 6397 } 6398 rrw_exit(&dp->dp_config_rwlock, FTAG); 6399} 6400 6401/* 6402 * Sync the specified transaction group. New blocks may be dirtied as 6403 * part of the process, so we iterate until it converges. 6404 */ 6405void 6406spa_sync(spa_t *spa, uint64_t txg) 6407{ 6408 dsl_pool_t *dp = spa->spa_dsl_pool; 6409 objset_t *mos = spa->spa_meta_objset; 6410 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK]; 6411 vdev_t *rvd = spa->spa_root_vdev; 6412 vdev_t *vd; 6413 dmu_tx_t *tx; 6414 int error; 6415 6416 VERIFY(spa_writeable(spa)); 6417 6418 /* 6419 * Lock out configuration changes. 6420 */ 6421 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER); 6422 6423 spa->spa_syncing_txg = txg; 6424 spa->spa_sync_pass = 0; 6425 6426 /* 6427 * If there are any pending vdev state changes, convert them 6428 * into config changes that go out with this transaction group. 6429 */ 6430 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6431 while (list_head(&spa->spa_state_dirty_list) != NULL) { 6432 /* 6433 * We need the write lock here because, for aux vdevs, 6434 * calling vdev_config_dirty() modifies sav_config. 6435 * This is ugly and will become unnecessary when we 6436 * eliminate the aux vdev wart by integrating all vdevs 6437 * into the root vdev tree. 6438 */ 6439 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6440 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER); 6441 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) { 6442 vdev_state_clean(vd); 6443 vdev_config_dirty(vd); 6444 } 6445 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 6446 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 6447 } 6448 spa_config_exit(spa, SCL_STATE, FTAG); 6449 6450 tx = dmu_tx_create_assigned(dp, txg); 6451 6452 spa->spa_sync_starttime = gethrtime(); 6453#ifdef illumos 6454 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, 6455 spa->spa_sync_starttime + spa->spa_deadman_synctime)); 6456#else /* FreeBSD */ 6457#ifdef _KERNEL 6458 callout_reset(&spa->spa_deadman_cycid, 6459 hz * spa->spa_deadman_synctime / NANOSEC, spa_deadman, spa); 6460#endif 6461#endif 6462 6463 /* 6464 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg, 6465 * set spa_deflate if we have no raid-z vdevs. 6466 */ 6467 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE && 6468 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) { 6469 int i; 6470 6471 for (i = 0; i < rvd->vdev_children; i++) { 6472 vd = rvd->vdev_child[i]; 6473 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE) 6474 break; 6475 } 6476 if (i == rvd->vdev_children) { 6477 spa->spa_deflate = TRUE; 6478 VERIFY(0 == zap_add(spa->spa_meta_objset, 6479 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE, 6480 sizeof (uint64_t), 1, &spa->spa_deflate, tx)); 6481 } 6482 } 6483 6484 /* 6485 * If anything has changed in this txg, or if someone is waiting 6486 * for this txg to sync (eg, spa_vdev_remove()), push the 6487 * deferred frees from the previous txg. If not, leave them 6488 * alone so that we don't generate work on an otherwise idle 6489 * system. 6490 */ 6491 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) || 6492 !txg_list_empty(&dp->dp_dirty_dirs, txg) || 6493 !txg_list_empty(&dp->dp_sync_tasks, txg) || 6494 ((dsl_scan_active(dp->dp_scan) || 6495 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) { 6496 spa_sync_deferred_frees(spa, tx); 6497 } 6498 6499 /* 6500 * Iterate to convergence. 6501 */ 6502 do { 6503 int pass = ++spa->spa_sync_pass; 6504 6505 spa_sync_config_object(spa, tx); 6506 spa_sync_aux_dev(spa, &spa->spa_spares, tx, 6507 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES); 6508 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx, 6509 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE); 6510 spa_errlog_sync(spa, txg); 6511 dsl_pool_sync(dp, txg); 6512 6513 if (pass < zfs_sync_pass_deferred_free) { 6514 spa_sync_frees(spa, free_bpl, tx); 6515 } else { 6516 bplist_iterate(free_bpl, bpobj_enqueue_cb, 6517 &spa->spa_deferred_bpobj, tx); 6518 } 6519 6520 ddt_sync(spa, txg); 6521 dsl_scan_sync(dp, tx); 6522 6523 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)) 6524 vdev_sync(vd, txg); 6525 6526 if (pass == 1) 6527 spa_sync_upgrades(spa, tx); 6528 6529 } while (dmu_objset_is_dirty(mos, txg)); 6530 6531 /* 6532 * Rewrite the vdev configuration (which includes the uberblock) 6533 * to commit the transaction group. 6534 * 6535 * If there are no dirty vdevs, we sync the uberblock to a few 6536 * random top-level vdevs that are known to be visible in the 6537 * config cache (see spa_vdev_add() for a complete description). 6538 * If there *are* dirty vdevs, sync the uberblock to all vdevs. 6539 */ 6540 for (;;) { 6541 /* 6542 * We hold SCL_STATE to prevent vdev open/close/etc. 6543 * while we're attempting to write the vdev labels. 6544 */ 6545 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER); 6546 6547 if (list_is_empty(&spa->spa_config_dirty_list)) { 6548 vdev_t *svd[SPA_DVAS_PER_BP]; 6549 int svdcount = 0; 6550 int children = rvd->vdev_children; 6551 int c0 = spa_get_random(children); 6552 6553 for (int c = 0; c < children; c++) { 6554 vd = rvd->vdev_child[(c0 + c) % children]; 6555 if (vd->vdev_ms_array == 0 || vd->vdev_islog) 6556 continue; 6557 svd[svdcount++] = vd; 6558 if (svdcount == SPA_DVAS_PER_BP) 6559 break; 6560 } 6561 error = vdev_config_sync(svd, svdcount, txg, B_FALSE); 6562 if (error != 0) 6563 error = vdev_config_sync(svd, svdcount, txg, 6564 B_TRUE); 6565 } else { 6566 error = vdev_config_sync(rvd->vdev_child, 6567 rvd->vdev_children, txg, B_FALSE); 6568 if (error != 0) 6569 error = vdev_config_sync(rvd->vdev_child, 6570 rvd->vdev_children, txg, B_TRUE); 6571 } 6572 6573 if (error == 0) 6574 spa->spa_last_synced_guid = rvd->vdev_guid; 6575 6576 spa_config_exit(spa, SCL_STATE, FTAG); 6577 6578 if (error == 0) 6579 break; 6580 zio_suspend(spa, NULL); 6581 zio_resume_wait(spa); 6582 } 6583 dmu_tx_commit(tx); 6584 6585#ifdef illumos 6586 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY)); 6587#else /* FreeBSD */ 6588#ifdef _KERNEL 6589 callout_drain(&spa->spa_deadman_cycid); 6590#endif 6591#endif 6592 6593 /* 6594 * Clear the dirty config list. 6595 */ 6596 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL) 6597 vdev_config_clean(vd); 6598 6599 /* 6600 * Now that the new config has synced transactionally, 6601 * let it become visible to the config cache. 6602 */ 6603 if (spa->spa_config_syncing != NULL) { 6604 spa_config_set(spa, spa->spa_config_syncing); 6605 spa->spa_config_txg = txg; 6606 spa->spa_config_syncing = NULL; 6607 } 6608 6609 spa->spa_ubsync = spa->spa_uberblock; 6610 6611 dsl_pool_sync_done(dp, txg); 6612 6613 /* 6614 * Update usable space statistics. 6615 */ 6616 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))) 6617 vdev_sync_done(vd, txg); 6618 6619 spa_update_dspace(spa); 6620 6621 /* 6622 * It had better be the case that we didn't dirty anything 6623 * since vdev_config_sync(). 6624 */ 6625 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg)); 6626 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg)); 6627 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg)); 6628 6629 spa->spa_sync_pass = 0; 6630 6631 spa_config_exit(spa, SCL_CONFIG, FTAG); 6632 6633 spa_handle_ignored_writes(spa); 6634 6635 /* 6636 * If any async tasks have been requested, kick them off. 6637 */ 6638 spa_async_dispatch(spa); 6639 spa_async_dispatch_vd(spa); 6640} 6641 6642/* 6643 * Sync all pools. We don't want to hold the namespace lock across these 6644 * operations, so we take a reference on the spa_t and drop the lock during the 6645 * sync. 6646 */ 6647void 6648spa_sync_allpools(void) 6649{ 6650 spa_t *spa = NULL; 6651 mutex_enter(&spa_namespace_lock); 6652 while ((spa = spa_next(spa)) != NULL) { 6653 if (spa_state(spa) != POOL_STATE_ACTIVE || 6654 !spa_writeable(spa) || spa_suspended(spa)) 6655 continue; 6656 spa_open_ref(spa, FTAG); 6657 mutex_exit(&spa_namespace_lock); 6658 txg_wait_synced(spa_get_dsl(spa), 0); 6659 mutex_enter(&spa_namespace_lock); 6660 spa_close(spa, FTAG); 6661 } 6662 mutex_exit(&spa_namespace_lock); 6663} 6664 6665/* 6666 * ========================================================================== 6667 * Miscellaneous routines 6668 * ========================================================================== 6669 */ 6670 6671/* 6672 * Remove all pools in the system. 6673 */ 6674void 6675spa_evict_all(void) 6676{ 6677 spa_t *spa; 6678 6679 /* 6680 * Remove all cached state. All pools should be closed now, 6681 * so every spa in the AVL tree should be unreferenced. 6682 */ 6683 mutex_enter(&spa_namespace_lock); 6684 while ((spa = spa_next(NULL)) != NULL) { 6685 /* 6686 * Stop async tasks. The async thread may need to detach 6687 * a device that's been replaced, which requires grabbing 6688 * spa_namespace_lock, so we must drop it here. 6689 */ 6690 spa_open_ref(spa, FTAG); 6691 mutex_exit(&spa_namespace_lock); 6692 spa_async_suspend(spa); 6693 mutex_enter(&spa_namespace_lock); 6694 spa_close(spa, FTAG); 6695 6696 if (spa->spa_state != POOL_STATE_UNINITIALIZED) { 6697 spa_unload(spa); 6698 spa_deactivate(spa); 6699 } 6700 spa_remove(spa); 6701 } 6702 mutex_exit(&spa_namespace_lock); 6703} 6704 6705vdev_t * 6706spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux) 6707{ 6708 vdev_t *vd; 6709 int i; 6710 6711 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL) 6712 return (vd); 6713 6714 if (aux) { 6715 for (i = 0; i < spa->spa_l2cache.sav_count; i++) { 6716 vd = spa->spa_l2cache.sav_vdevs[i]; 6717 if (vd->vdev_guid == guid) 6718 return (vd); 6719 } 6720 6721 for (i = 0; i < spa->spa_spares.sav_count; i++) { 6722 vd = spa->spa_spares.sav_vdevs[i]; 6723 if (vd->vdev_guid == guid) 6724 return (vd); 6725 } 6726 } 6727 6728 return (NULL); 6729} 6730 6731void 6732spa_upgrade(spa_t *spa, uint64_t version) 6733{ 6734 ASSERT(spa_writeable(spa)); 6735 6736 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 6737 6738 /* 6739 * This should only be called for a non-faulted pool, and since a 6740 * future version would result in an unopenable pool, this shouldn't be 6741 * possible. 6742 */ 6743 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version)); 6744 ASSERT(version >= spa->spa_uberblock.ub_version); 6745 6746 spa->spa_uberblock.ub_version = version; 6747 vdev_config_dirty(spa->spa_root_vdev); 6748 6749 spa_config_exit(spa, SCL_ALL, FTAG); 6750 6751 txg_wait_synced(spa_get_dsl(spa), 0); 6752} 6753 6754boolean_t 6755spa_has_spare(spa_t *spa, uint64_t guid) 6756{ 6757 int i; 6758 uint64_t spareguid; 6759 spa_aux_vdev_t *sav = &spa->spa_spares; 6760 6761 for (i = 0; i < sav->sav_count; i++) 6762 if (sav->sav_vdevs[i]->vdev_guid == guid) 6763 return (B_TRUE); 6764 6765 for (i = 0; i < sav->sav_npending; i++) { 6766 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID, 6767 &spareguid) == 0 && spareguid == guid) 6768 return (B_TRUE); 6769 } 6770 6771 return (B_FALSE); 6772} 6773 6774/* 6775 * Check if a pool has an active shared spare device. 6776 * Note: reference count of an active spare is 2, as a spare and as a replace 6777 */ 6778static boolean_t 6779spa_has_active_shared_spare(spa_t *spa) 6780{ 6781 int i, refcnt; 6782 uint64_t pool; 6783 spa_aux_vdev_t *sav = &spa->spa_spares; 6784 6785 for (i = 0; i < sav->sav_count; i++) { 6786 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool, 6787 &refcnt) && pool != 0ULL && pool == spa_guid(spa) && 6788 refcnt > 2) 6789 return (B_TRUE); 6790 } 6791 6792 return (B_FALSE); 6793} 6794 6795/* 6796 * Post a sysevent corresponding to the given event. The 'name' must be one of 6797 * the event definitions in sys/sysevent/eventdefs.h. The payload will be 6798 * filled in from the spa and (optionally) the vdev. This doesn't do anything 6799 * in the userland libzpool, as we don't want consumers to misinterpret ztest 6800 * or zdb as real changes. 6801 */ 6802void 6803spa_event_notify(spa_t *spa, vdev_t *vd, const char *name) 6804{ 6805#ifdef _KERNEL 6806 sysevent_t *ev; 6807 sysevent_attr_list_t *attr = NULL; 6808 sysevent_value_t value; 6809 sysevent_id_t eid; 6810 6811 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs", 6812 SE_SLEEP); 6813 6814 value.value_type = SE_DATA_TYPE_STRING; 6815 value.value.sv_string = spa_name(spa); 6816 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0) 6817 goto done; 6818 6819 value.value_type = SE_DATA_TYPE_UINT64; 6820 value.value.sv_uint64 = spa_guid(spa); 6821 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0) 6822 goto done; 6823 6824 if (vd) { 6825 value.value_type = SE_DATA_TYPE_UINT64; 6826 value.value.sv_uint64 = vd->vdev_guid; 6827 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value, 6828 SE_SLEEP) != 0) 6829 goto done; 6830 6831 if (vd->vdev_path) { 6832 value.value_type = SE_DATA_TYPE_STRING; 6833 value.value.sv_string = vd->vdev_path; 6834 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH, 6835 &value, SE_SLEEP) != 0) 6836 goto done; 6837 } 6838 } 6839 6840 if (sysevent_attach_attributes(ev, attr) != 0) 6841 goto done; 6842 attr = NULL; 6843 6844 (void) log_sysevent(ev, SE_SLEEP, &eid); 6845 6846done: 6847 if (attr) 6848 sysevent_free_attr(attr); 6849 sysevent_free(ev); 6850#endif 6851} 6852