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.
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.
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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
25 * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
26 * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright 2013 Saso Kiselkov. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2016 Toomas Soome <tsoome@me.com>
31 * Copyright 2018 Joyent, Inc.
32 * Copyright (c) 2017 Datto Inc.
33 * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
37 * SPA: Storage Pool Allocator
39 * This file contains all the routines used when modifying on-disk SPA state.
40 * This includes opening, importing, destroying, exporting a pool, and syncing a
44 #include <sys/zfs_context.h>
45 #include <sys/fm/fs/zfs.h>
46 #include <sys/spa_impl.h>
48 #include <sys/zio_checksum.h>
50 #include <sys/dmu_tx.h>
54 #include <sys/vdev_impl.h>
55 #include <sys/vdev_removal.h>
56 #include <sys/vdev_indirect_mapping.h>
57 #include <sys/vdev_indirect_births.h>
58 #include <sys/vdev_initialize.h>
59 #include <sys/metaslab.h>
60 #include <sys/metaslab_impl.h>
61 #include <sys/uberblock_impl.h>
64 #include <sys/bpobj.h>
65 #include <sys/dmu_traverse.h>
66 #include <sys/dmu_objset.h>
67 #include <sys/unique.h>
68 #include <sys/dsl_pool.h>
69 #include <sys/dsl_dataset.h>
70 #include <sys/dsl_dir.h>
71 #include <sys/dsl_prop.h>
72 #include <sys/dsl_synctask.h>
73 #include <sys/fs/zfs.h>
75 #include <sys/callb.h>
76 #include <sys/spa_boot.h>
77 #include <sys/zfs_ioctl.h>
78 #include <sys/dsl_scan.h>
79 #include <sys/dmu_send.h>
80 #include <sys/dsl_destroy.h>
81 #include <sys/dsl_userhold.h>
82 #include <sys/zfeature.h>
84 #include <sys/trim_map.h>
88 #include <sys/callb.h>
89 #include <sys/cpupart.h>
94 #include "zfs_comutil.h"
96 /* Check hostid on import? */
97 static int check_hostid = 1;
100 * The interval, in seconds, at which failed configuration cache file writes
103 int zfs_ccw_retry_interval = 300;
105 SYSCTL_DECL(_vfs_zfs);
106 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
107 "Check hostid on import?");
108 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
109 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
110 &zfs_ccw_retry_interval, 0,
111 "Configuration cache file write, retry after failure, interval (seconds)");
113 typedef enum zti_modes {
114 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
115 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
116 ZTI_MODE_NULL, /* don't create a taskq */
120 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
121 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
122 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
124 #define ZTI_N(n) ZTI_P(n, 1)
125 #define ZTI_ONE ZTI_N(1)
127 typedef struct zio_taskq_info {
128 zti_modes_t zti_mode;
133 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
134 "issue", "issue_high", "intr", "intr_high"
138 * This table defines the taskq settings for each ZFS I/O type. When
139 * initializing a pool, we use this table to create an appropriately sized
140 * taskq. Some operations are low volume and therefore have a small, static
141 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
142 * macros. Other operations process a large amount of data; the ZTI_BATCH
143 * macro causes us to create a taskq oriented for throughput. Some operations
144 * are so high frequency and short-lived that the taskq itself can become a a
145 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
146 * additional degree of parallelism specified by the number of threads per-
147 * taskq and the number of taskqs; when dispatching an event in this case, the
148 * particular taskq is chosen at random.
150 * The different taskq priorities are to handle the different contexts (issue
151 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
152 * need to be handled with minimum delay.
154 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
155 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
156 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
157 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
158 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
159 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
161 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
164 static void spa_sync_version(void *arg, dmu_tx_t *tx);
165 static void spa_sync_props(void *arg, dmu_tx_t *tx);
166 static boolean_t spa_has_active_shared_spare(spa_t *spa);
167 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
168 static void spa_vdev_resilver_done(spa_t *spa);
170 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
172 id_t zio_taskq_psrset_bind = PS_NONE;
175 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
176 uint_t zio_taskq_basedc = 80; /* base duty cycle */
179 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
180 extern int zfs_sync_pass_deferred_free;
183 * Report any spa_load_verify errors found, but do not fail spa_load.
184 * This is used by zdb to analyze non-idle pools.
186 boolean_t spa_load_verify_dryrun = B_FALSE;
189 * This (illegal) pool name is used when temporarily importing a spa_t in order
190 * to get the vdev stats associated with the imported devices.
192 #define TRYIMPORT_NAME "$import"
195 * For debugging purposes: print out vdev tree during pool import.
197 int spa_load_print_vdev_tree = B_FALSE;
200 * A non-zero value for zfs_max_missing_tvds means that we allow importing
201 * pools with missing top-level vdevs. This is strictly intended for advanced
202 * pool recovery cases since missing data is almost inevitable. Pools with
203 * missing devices can only be imported read-only for safety reasons, and their
204 * fail-mode will be automatically set to "continue".
206 * With 1 missing vdev we should be able to import the pool and mount all
207 * datasets. User data that was not modified after the missing device has been
208 * added should be recoverable. This means that snapshots created prior to the
209 * addition of that device should be completely intact.
211 * With 2 missing vdevs, some datasets may fail to mount since there are
212 * dataset statistics that are stored as regular metadata. Some data might be
213 * recoverable if those vdevs were added recently.
215 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
216 * may be missing entirely. Chances of data recovery are very low. Note that
217 * there are also risks of performing an inadvertent rewind as we might be
218 * missing all the vdevs with the latest uberblocks.
220 uint64_t zfs_max_missing_tvds = 0;
223 * The parameters below are similar to zfs_max_missing_tvds but are only
224 * intended for a preliminary open of the pool with an untrusted config which
225 * might be incomplete or out-dated.
227 * We are more tolerant for pools opened from a cachefile since we could have
228 * an out-dated cachefile where a device removal was not registered.
229 * We could have set the limit arbitrarily high but in the case where devices
230 * are really missing we would want to return the proper error codes; we chose
231 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
232 * and we get a chance to retrieve the trusted config.
234 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
237 * In the case where config was assembled by scanning device paths (/dev/dsks
238 * by default) we are less tolerant since all the existing devices should have
239 * been detected and we want spa_load to return the right error codes.
241 uint64_t zfs_max_missing_tvds_scan = 0;
244 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN,
245 &spa_load_print_vdev_tree, 0,
246 "print out vdev tree during pool import");
247 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN,
248 &zfs_max_missing_tvds, 0,
249 "allow importing pools with missing top-level vdevs");
250 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN,
251 &zfs_max_missing_tvds_cachefile, 0,
252 "allow importing pools with missing top-level vdevs in cache file");
253 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN,
254 &zfs_max_missing_tvds_scan, 0,
255 "allow importing pools with missing top-level vdevs during scan");
258 * Debugging aid that pauses spa_sync() towards the end.
260 boolean_t zfs_pause_spa_sync = B_FALSE;
263 * ==========================================================================
264 * SPA properties routines
265 * ==========================================================================
269 * Add a (source=src, propname=propval) list to an nvlist.
272 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
273 uint64_t intval, zprop_source_t src)
275 const char *propname = zpool_prop_to_name(prop);
278 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
279 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
282 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
284 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
286 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
287 nvlist_free(propval);
291 * Get property values from the spa configuration.
294 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
296 vdev_t *rvd = spa->spa_root_vdev;
297 dsl_pool_t *pool = spa->spa_dsl_pool;
298 uint64_t size, alloc, cap, version;
299 zprop_source_t src = ZPROP_SRC_NONE;
300 spa_config_dirent_t *dp;
301 metaslab_class_t *mc = spa_normal_class(spa);
303 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
306 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
307 size = metaslab_class_get_space(spa_normal_class(spa));
308 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
309 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
310 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
311 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
313 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
314 spa->spa_checkpoint_info.sci_dspace, src);
316 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
317 metaslab_class_fragmentation(mc), src);
318 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
319 metaslab_class_expandable_space(mc), src);
320 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
321 (spa_mode(spa) == FREAD), src);
323 cap = (size == 0) ? 0 : (alloc * 100 / size);
324 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
326 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
327 ddt_get_pool_dedup_ratio(spa), src);
329 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
330 rvd->vdev_state, src);
332 version = spa_version(spa);
333 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
334 src = ZPROP_SRC_DEFAULT;
336 src = ZPROP_SRC_LOCAL;
337 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
342 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
343 * when opening pools before this version freedir will be NULL.
345 if (pool->dp_free_dir != NULL) {
346 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
347 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
350 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
354 if (pool->dp_leak_dir != NULL) {
355 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
356 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
359 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
364 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
366 if (spa->spa_comment != NULL) {
367 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
371 if (spa->spa_root != NULL)
372 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
375 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
376 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
377 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
379 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
380 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
383 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
384 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
385 DNODE_MAX_SIZE, ZPROP_SRC_NONE);
387 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
388 DNODE_MIN_SIZE, ZPROP_SRC_NONE);
391 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
392 if (dp->scd_path == NULL) {
393 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
394 "none", 0, ZPROP_SRC_LOCAL);
395 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
396 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
397 dp->scd_path, 0, ZPROP_SRC_LOCAL);
403 * Get zpool property values.
406 spa_prop_get(spa_t *spa, nvlist_t **nvp)
408 objset_t *mos = spa->spa_meta_objset;
413 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
415 mutex_enter(&spa->spa_props_lock);
418 * Get properties from the spa config.
420 spa_prop_get_config(spa, nvp);
422 /* If no pool property object, no more prop to get. */
423 if (mos == NULL || spa->spa_pool_props_object == 0) {
424 mutex_exit(&spa->spa_props_lock);
429 * Get properties from the MOS pool property object.
431 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
432 (err = zap_cursor_retrieve(&zc, &za)) == 0;
433 zap_cursor_advance(&zc)) {
436 zprop_source_t src = ZPROP_SRC_DEFAULT;
439 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
442 switch (za.za_integer_length) {
444 /* integer property */
445 if (za.za_first_integer !=
446 zpool_prop_default_numeric(prop))
447 src = ZPROP_SRC_LOCAL;
449 if (prop == ZPOOL_PROP_BOOTFS) {
451 dsl_dataset_t *ds = NULL;
453 dp = spa_get_dsl(spa);
454 dsl_pool_config_enter(dp, FTAG);
455 err = dsl_dataset_hold_obj(dp,
456 za.za_first_integer, FTAG, &ds);
458 dsl_pool_config_exit(dp, FTAG);
462 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
464 dsl_dataset_name(ds, strval);
465 dsl_dataset_rele(ds, FTAG);
466 dsl_pool_config_exit(dp, FTAG);
469 intval = za.za_first_integer;
472 spa_prop_add_list(*nvp, prop, strval, intval, src);
475 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
480 /* string property */
481 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
482 err = zap_lookup(mos, spa->spa_pool_props_object,
483 za.za_name, 1, za.za_num_integers, strval);
485 kmem_free(strval, za.za_num_integers);
488 spa_prop_add_list(*nvp, prop, strval, 0, src);
489 kmem_free(strval, za.za_num_integers);
496 zap_cursor_fini(&zc);
497 mutex_exit(&spa->spa_props_lock);
499 if (err && err != ENOENT) {
509 * Validate the given pool properties nvlist and modify the list
510 * for the property values to be set.
513 spa_prop_validate(spa_t *spa, nvlist_t *props)
516 int error = 0, reset_bootfs = 0;
518 boolean_t has_feature = B_FALSE;
521 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
523 char *strval, *slash, *check, *fname;
524 const char *propname = nvpair_name(elem);
525 zpool_prop_t prop = zpool_name_to_prop(propname);
528 case ZPOOL_PROP_INVAL:
529 if (!zpool_prop_feature(propname)) {
530 error = SET_ERROR(EINVAL);
535 * Sanitize the input.
537 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
538 error = SET_ERROR(EINVAL);
542 if (nvpair_value_uint64(elem, &intval) != 0) {
543 error = SET_ERROR(EINVAL);
548 error = SET_ERROR(EINVAL);
552 fname = strchr(propname, '@') + 1;
553 if (zfeature_lookup_name(fname, NULL) != 0) {
554 error = SET_ERROR(EINVAL);
558 has_feature = B_TRUE;
561 case ZPOOL_PROP_VERSION:
562 error = nvpair_value_uint64(elem, &intval);
564 (intval < spa_version(spa) ||
565 intval > SPA_VERSION_BEFORE_FEATURES ||
567 error = SET_ERROR(EINVAL);
570 case ZPOOL_PROP_DELEGATION:
571 case ZPOOL_PROP_AUTOREPLACE:
572 case ZPOOL_PROP_LISTSNAPS:
573 case ZPOOL_PROP_AUTOEXPAND:
574 error = nvpair_value_uint64(elem, &intval);
575 if (!error && intval > 1)
576 error = SET_ERROR(EINVAL);
579 case ZPOOL_PROP_BOOTFS:
581 * If the pool version is less than SPA_VERSION_BOOTFS,
582 * or the pool is still being created (version == 0),
583 * the bootfs property cannot be set.
585 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
586 error = SET_ERROR(ENOTSUP);
591 * Make sure the vdev config is bootable
593 if (!vdev_is_bootable(spa->spa_root_vdev)) {
594 error = SET_ERROR(ENOTSUP);
600 error = nvpair_value_string(elem, &strval);
606 if (strval == NULL || strval[0] == '\0') {
607 objnum = zpool_prop_default_numeric(
612 error = dmu_objset_hold(strval, FTAG, &os);
617 * Must be ZPL, and its property settings
621 if (dmu_objset_type(os) != DMU_OST_ZFS) {
622 error = SET_ERROR(ENOTSUP);
624 dsl_prop_get_int_ds(dmu_objset_ds(os),
625 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
627 !BOOTFS_COMPRESS_VALID(propval)) {
628 error = SET_ERROR(ENOTSUP);
630 objnum = dmu_objset_id(os);
632 dmu_objset_rele(os, FTAG);
636 case ZPOOL_PROP_FAILUREMODE:
637 error = nvpair_value_uint64(elem, &intval);
638 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
639 intval > ZIO_FAILURE_MODE_PANIC))
640 error = SET_ERROR(EINVAL);
643 * This is a special case which only occurs when
644 * the pool has completely failed. This allows
645 * the user to change the in-core failmode property
646 * without syncing it out to disk (I/Os might
647 * currently be blocked). We do this by returning
648 * EIO to the caller (spa_prop_set) to trick it
649 * into thinking we encountered a property validation
652 if (!error && spa_suspended(spa)) {
653 spa->spa_failmode = intval;
654 error = SET_ERROR(EIO);
658 case ZPOOL_PROP_CACHEFILE:
659 if ((error = nvpair_value_string(elem, &strval)) != 0)
662 if (strval[0] == '\0')
665 if (strcmp(strval, "none") == 0)
668 if (strval[0] != '/') {
669 error = SET_ERROR(EINVAL);
673 slash = strrchr(strval, '/');
674 ASSERT(slash != NULL);
676 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
677 strcmp(slash, "/..") == 0)
678 error = SET_ERROR(EINVAL);
681 case ZPOOL_PROP_COMMENT:
682 if ((error = nvpair_value_string(elem, &strval)) != 0)
684 for (check = strval; *check != '\0'; check++) {
686 * The kernel doesn't have an easy isprint()
687 * check. For this kernel check, we merely
688 * check ASCII apart from DEL. Fix this if
689 * there is an easy-to-use kernel isprint().
691 if (*check >= 0x7f) {
692 error = SET_ERROR(EINVAL);
696 if (strlen(strval) > ZPROP_MAX_COMMENT)
700 case ZPOOL_PROP_DEDUPDITTO:
701 if (spa_version(spa) < SPA_VERSION_DEDUP)
702 error = SET_ERROR(ENOTSUP);
704 error = nvpair_value_uint64(elem, &intval);
706 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
707 error = SET_ERROR(EINVAL);
715 if (!error && reset_bootfs) {
716 error = nvlist_remove(props,
717 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
720 error = nvlist_add_uint64(props,
721 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
729 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
732 spa_config_dirent_t *dp;
734 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
738 dp = kmem_alloc(sizeof (spa_config_dirent_t),
741 if (cachefile[0] == '\0')
742 dp->scd_path = spa_strdup(spa_config_path);
743 else if (strcmp(cachefile, "none") == 0)
746 dp->scd_path = spa_strdup(cachefile);
748 list_insert_head(&spa->spa_config_list, dp);
750 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
754 spa_prop_set(spa_t *spa, nvlist_t *nvp)
757 nvpair_t *elem = NULL;
758 boolean_t need_sync = B_FALSE;
760 if ((error = spa_prop_validate(spa, nvp)) != 0)
763 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
764 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
766 if (prop == ZPOOL_PROP_CACHEFILE ||
767 prop == ZPOOL_PROP_ALTROOT ||
768 prop == ZPOOL_PROP_READONLY)
771 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
774 if (prop == ZPOOL_PROP_VERSION) {
775 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
777 ASSERT(zpool_prop_feature(nvpair_name(elem)));
778 ver = SPA_VERSION_FEATURES;
782 /* Save time if the version is already set. */
783 if (ver == spa_version(spa))
787 * In addition to the pool directory object, we might
788 * create the pool properties object, the features for
789 * read object, the features for write object, or the
790 * feature descriptions object.
792 error = dsl_sync_task(spa->spa_name, NULL,
793 spa_sync_version, &ver,
794 6, ZFS_SPACE_CHECK_RESERVED);
805 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
806 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
813 * If the bootfs property value is dsobj, clear it.
816 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
818 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
819 VERIFY(zap_remove(spa->spa_meta_objset,
820 spa->spa_pool_props_object,
821 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
828 spa_change_guid_check(void *arg, dmu_tx_t *tx)
830 uint64_t *newguid = arg;
831 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
832 vdev_t *rvd = spa->spa_root_vdev;
835 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
836 int error = (spa_has_checkpoint(spa)) ?
837 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
838 return (SET_ERROR(error));
841 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
842 vdev_state = rvd->vdev_state;
843 spa_config_exit(spa, SCL_STATE, FTAG);
845 if (vdev_state != VDEV_STATE_HEALTHY)
846 return (SET_ERROR(ENXIO));
848 ASSERT3U(spa_guid(spa), !=, *newguid);
854 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
856 uint64_t *newguid = arg;
857 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
859 vdev_t *rvd = spa->spa_root_vdev;
861 oldguid = spa_guid(spa);
863 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
864 rvd->vdev_guid = *newguid;
865 rvd->vdev_guid_sum += (*newguid - oldguid);
866 vdev_config_dirty(rvd);
867 spa_config_exit(spa, SCL_STATE, FTAG);
869 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
874 * Change the GUID for the pool. This is done so that we can later
875 * re-import a pool built from a clone of our own vdevs. We will modify
876 * the root vdev's guid, our own pool guid, and then mark all of our
877 * vdevs dirty. Note that we must make sure that all our vdevs are
878 * online when we do this, or else any vdevs that weren't present
879 * would be orphaned from our pool. We are also going to issue a
880 * sysevent to update any watchers.
883 spa_change_guid(spa_t *spa)
888 mutex_enter(&spa->spa_vdev_top_lock);
889 mutex_enter(&spa_namespace_lock);
890 guid = spa_generate_guid(NULL);
892 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
893 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
896 spa_write_cachefile(spa, B_FALSE, B_TRUE);
897 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
900 mutex_exit(&spa_namespace_lock);
901 mutex_exit(&spa->spa_vdev_top_lock);
907 * ==========================================================================
908 * SPA state manipulation (open/create/destroy/import/export)
909 * ==========================================================================
913 spa_error_entry_compare(const void *a, const void *b)
915 const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
916 const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
919 ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
920 sizeof (zbookmark_phys_t));
922 return (AVL_ISIGN(ret));
926 * Utility function which retrieves copies of the current logs and
927 * re-initializes them in the process.
930 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
932 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
934 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
935 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
937 avl_create(&spa->spa_errlist_scrub,
938 spa_error_entry_compare, sizeof (spa_error_entry_t),
939 offsetof(spa_error_entry_t, se_avl));
940 avl_create(&spa->spa_errlist_last,
941 spa_error_entry_compare, sizeof (spa_error_entry_t),
942 offsetof(spa_error_entry_t, se_avl));
946 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
948 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
949 enum zti_modes mode = ztip->zti_mode;
950 uint_t value = ztip->zti_value;
951 uint_t count = ztip->zti_count;
952 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
955 boolean_t batch = B_FALSE;
957 if (mode == ZTI_MODE_NULL) {
959 tqs->stqs_taskq = NULL;
963 ASSERT3U(count, >, 0);
965 tqs->stqs_count = count;
966 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
970 ASSERT3U(value, >=, 1);
971 value = MAX(value, 1);
976 flags |= TASKQ_THREADS_CPU_PCT;
977 value = zio_taskq_batch_pct;
981 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
983 zio_type_name[t], zio_taskq_types[q], mode, value);
987 for (uint_t i = 0; i < count; i++) {
991 (void) snprintf(name, sizeof (name), "%s_%s_%u",
992 zio_type_name[t], zio_taskq_types[q], i);
994 (void) snprintf(name, sizeof (name), "%s_%s",
995 zio_type_name[t], zio_taskq_types[q]);
999 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
1001 flags |= TASKQ_DC_BATCH;
1003 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
1004 spa->spa_proc, zio_taskq_basedc, flags);
1007 pri_t pri = maxclsyspri;
1009 * The write issue taskq can be extremely CPU
1010 * intensive. Run it at slightly lower priority
1011 * than the other taskqs.
1013 * - numerically higher priorities are lower priorities;
1014 * - if priorities divided by four (RQ_PPQ) are equal
1015 * then a difference between them is insignificant.
1017 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1024 tq = taskq_create_proc(name, value, pri, 50,
1025 INT_MAX, spa->spa_proc, flags);
1030 tqs->stqs_taskq[i] = tq;
1035 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1037 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1039 if (tqs->stqs_taskq == NULL) {
1040 ASSERT0(tqs->stqs_count);
1044 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1045 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1046 taskq_destroy(tqs->stqs_taskq[i]);
1049 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1050 tqs->stqs_taskq = NULL;
1054 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1055 * Note that a type may have multiple discrete taskqs to avoid lock contention
1056 * on the taskq itself. In that case we choose which taskq at random by using
1057 * the low bits of gethrtime().
1060 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1061 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1063 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1066 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1067 ASSERT3U(tqs->stqs_count, !=, 0);
1069 if (tqs->stqs_count == 1) {
1070 tq = tqs->stqs_taskq[0];
1073 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
1075 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1079 taskq_dispatch_ent(tq, func, arg, flags, ent);
1083 spa_create_zio_taskqs(spa_t *spa)
1085 for (int t = 0; t < ZIO_TYPES; t++) {
1086 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1087 spa_taskqs_init(spa, t, q);
1095 spa_thread(void *arg)
1097 callb_cpr_t cprinfo;
1100 user_t *pu = PTOU(curproc);
1102 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1105 ASSERT(curproc != &p0);
1106 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1107 "zpool-%s", spa->spa_name);
1108 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1111 /* bind this thread to the requested psrset */
1112 if (zio_taskq_psrset_bind != PS_NONE) {
1114 mutex_enter(&cpu_lock);
1115 mutex_enter(&pidlock);
1116 mutex_enter(&curproc->p_lock);
1118 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1119 0, NULL, NULL) == 0) {
1120 curthread->t_bind_pset = zio_taskq_psrset_bind;
1123 "Couldn't bind process for zfs pool \"%s\" to "
1124 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1127 mutex_exit(&curproc->p_lock);
1128 mutex_exit(&pidlock);
1129 mutex_exit(&cpu_lock);
1135 if (zio_taskq_sysdc) {
1136 sysdc_thread_enter(curthread, 100, 0);
1140 spa->spa_proc = curproc;
1141 spa->spa_did = curthread->t_did;
1143 spa_create_zio_taskqs(spa);
1145 mutex_enter(&spa->spa_proc_lock);
1146 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1148 spa->spa_proc_state = SPA_PROC_ACTIVE;
1149 cv_broadcast(&spa->spa_proc_cv);
1151 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1152 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1153 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1154 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1156 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1157 spa->spa_proc_state = SPA_PROC_GONE;
1158 spa->spa_proc = &p0;
1159 cv_broadcast(&spa->spa_proc_cv);
1160 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1162 mutex_enter(&curproc->p_lock);
1165 #endif /* SPA_PROCESS */
1169 * Activate an uninitialized pool.
1172 spa_activate(spa_t *spa, int mode)
1174 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1176 spa->spa_state = POOL_STATE_ACTIVE;
1177 spa->spa_mode = mode;
1179 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1180 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1182 /* Try to create a covering process */
1183 mutex_enter(&spa->spa_proc_lock);
1184 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1185 ASSERT(spa->spa_proc == &p0);
1189 /* Only create a process if we're going to be around a while. */
1190 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1191 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1193 spa->spa_proc_state = SPA_PROC_CREATED;
1194 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1195 cv_wait(&spa->spa_proc_cv,
1196 &spa->spa_proc_lock);
1198 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1199 ASSERT(spa->spa_proc != &p0);
1200 ASSERT(spa->spa_did != 0);
1204 "Couldn't create process for zfs pool \"%s\"\n",
1209 #endif /* SPA_PROCESS */
1210 mutex_exit(&spa->spa_proc_lock);
1212 /* If we didn't create a process, we need to create our taskqs. */
1213 ASSERT(spa->spa_proc == &p0);
1214 if (spa->spa_proc == &p0) {
1215 spa_create_zio_taskqs(spa);
1219 * Start TRIM thread.
1221 trim_thread_create(spa);
1223 for (size_t i = 0; i < TXG_SIZE; i++) {
1224 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1228 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1229 offsetof(vdev_t, vdev_config_dirty_node));
1230 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1231 offsetof(objset_t, os_evicting_node));
1232 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1233 offsetof(vdev_t, vdev_state_dirty_node));
1235 txg_list_create(&spa->spa_vdev_txg_list, spa,
1236 offsetof(struct vdev, vdev_txg_node));
1238 avl_create(&spa->spa_errlist_scrub,
1239 spa_error_entry_compare, sizeof (spa_error_entry_t),
1240 offsetof(spa_error_entry_t, se_avl));
1241 avl_create(&spa->spa_errlist_last,
1242 spa_error_entry_compare, sizeof (spa_error_entry_t),
1243 offsetof(spa_error_entry_t, se_avl));
1247 * Opposite of spa_activate().
1250 spa_deactivate(spa_t *spa)
1252 ASSERT(spa->spa_sync_on == B_FALSE);
1253 ASSERT(spa->spa_dsl_pool == NULL);
1254 ASSERT(spa->spa_root_vdev == NULL);
1255 ASSERT(spa->spa_async_zio_root == NULL);
1256 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1259 * Stop TRIM thread in case spa_unload() wasn't called directly
1260 * before spa_deactivate().
1262 trim_thread_destroy(spa);
1264 spa_evicting_os_wait(spa);
1266 txg_list_destroy(&spa->spa_vdev_txg_list);
1268 list_destroy(&spa->spa_config_dirty_list);
1269 list_destroy(&spa->spa_evicting_os_list);
1270 list_destroy(&spa->spa_state_dirty_list);
1272 for (int t = 0; t < ZIO_TYPES; t++) {
1273 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1274 spa_taskqs_fini(spa, t, q);
1278 for (size_t i = 0; i < TXG_SIZE; i++) {
1279 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1280 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1281 spa->spa_txg_zio[i] = NULL;
1284 metaslab_class_destroy(spa->spa_normal_class);
1285 spa->spa_normal_class = NULL;
1287 metaslab_class_destroy(spa->spa_log_class);
1288 spa->spa_log_class = NULL;
1291 * If this was part of an import or the open otherwise failed, we may
1292 * still have errors left in the queues. Empty them just in case.
1294 spa_errlog_drain(spa);
1296 avl_destroy(&spa->spa_errlist_scrub);
1297 avl_destroy(&spa->spa_errlist_last);
1299 spa->spa_state = POOL_STATE_UNINITIALIZED;
1301 mutex_enter(&spa->spa_proc_lock);
1302 if (spa->spa_proc_state != SPA_PROC_NONE) {
1303 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1304 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1305 cv_broadcast(&spa->spa_proc_cv);
1306 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1307 ASSERT(spa->spa_proc != &p0);
1308 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1310 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1311 spa->spa_proc_state = SPA_PROC_NONE;
1313 ASSERT(spa->spa_proc == &p0);
1314 mutex_exit(&spa->spa_proc_lock);
1318 * We want to make sure spa_thread() has actually exited the ZFS
1319 * module, so that the module can't be unloaded out from underneath
1322 if (spa->spa_did != 0) {
1323 thread_join(spa->spa_did);
1326 #endif /* SPA_PROCESS */
1330 * Verify a pool configuration, and construct the vdev tree appropriately. This
1331 * will create all the necessary vdevs in the appropriate layout, with each vdev
1332 * in the CLOSED state. This will prep the pool before open/creation/import.
1333 * All vdev validation is done by the vdev_alloc() routine.
1336 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1337 uint_t id, int atype)
1343 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1346 if ((*vdp)->vdev_ops->vdev_op_leaf)
1349 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1352 if (error == ENOENT)
1358 return (SET_ERROR(EINVAL));
1361 for (int c = 0; c < children; c++) {
1363 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1371 ASSERT(*vdp != NULL);
1377 * Opposite of spa_load().
1380 spa_unload(spa_t *spa)
1384 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1386 spa_load_note(spa, "UNLOADING");
1391 trim_thread_destroy(spa);
1396 spa_async_suspend(spa);
1398 if (spa->spa_root_vdev) {
1399 vdev_initialize_stop_all(spa->spa_root_vdev,
1400 VDEV_INITIALIZE_ACTIVE);
1406 if (spa->spa_sync_on) {
1407 txg_sync_stop(spa->spa_dsl_pool);
1408 spa->spa_sync_on = B_FALSE;
1412 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1413 * to call it earlier, before we wait for async i/o to complete.
1414 * This ensures that there is no async metaslab prefetching, by
1415 * calling taskq_wait(mg_taskq).
1417 if (spa->spa_root_vdev != NULL) {
1418 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1419 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1420 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1421 spa_config_exit(spa, SCL_ALL, spa);
1425 * Wait for any outstanding async I/O to complete.
1427 if (spa->spa_async_zio_root != NULL) {
1428 for (int i = 0; i < max_ncpus; i++)
1429 (void) zio_wait(spa->spa_async_zio_root[i]);
1430 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1431 spa->spa_async_zio_root = NULL;
1434 if (spa->spa_vdev_removal != NULL) {
1435 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1436 spa->spa_vdev_removal = NULL;
1439 if (spa->spa_condense_zthr != NULL) {
1440 ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1441 zthr_destroy(spa->spa_condense_zthr);
1442 spa->spa_condense_zthr = NULL;
1445 if (spa->spa_checkpoint_discard_zthr != NULL) {
1446 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1447 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1448 spa->spa_checkpoint_discard_zthr = NULL;
1451 spa_condense_fini(spa);
1453 bpobj_close(&spa->spa_deferred_bpobj);
1455 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1460 if (spa->spa_root_vdev)
1461 vdev_free(spa->spa_root_vdev);
1462 ASSERT(spa->spa_root_vdev == NULL);
1465 * Close the dsl pool.
1467 if (spa->spa_dsl_pool) {
1468 dsl_pool_close(spa->spa_dsl_pool);
1469 spa->spa_dsl_pool = NULL;
1470 spa->spa_meta_objset = NULL;
1476 * Drop and purge level 2 cache
1478 spa_l2cache_drop(spa);
1480 for (i = 0; i < spa->spa_spares.sav_count; i++)
1481 vdev_free(spa->spa_spares.sav_vdevs[i]);
1482 if (spa->spa_spares.sav_vdevs) {
1483 kmem_free(spa->spa_spares.sav_vdevs,
1484 spa->spa_spares.sav_count * sizeof (void *));
1485 spa->spa_spares.sav_vdevs = NULL;
1487 if (spa->spa_spares.sav_config) {
1488 nvlist_free(spa->spa_spares.sav_config);
1489 spa->spa_spares.sav_config = NULL;
1491 spa->spa_spares.sav_count = 0;
1493 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1494 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1495 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1497 if (spa->spa_l2cache.sav_vdevs) {
1498 kmem_free(spa->spa_l2cache.sav_vdevs,
1499 spa->spa_l2cache.sav_count * sizeof (void *));
1500 spa->spa_l2cache.sav_vdevs = NULL;
1502 if (spa->spa_l2cache.sav_config) {
1503 nvlist_free(spa->spa_l2cache.sav_config);
1504 spa->spa_l2cache.sav_config = NULL;
1506 spa->spa_l2cache.sav_count = 0;
1508 spa->spa_async_suspended = 0;
1510 spa->spa_indirect_vdevs_loaded = B_FALSE;
1512 if (spa->spa_comment != NULL) {
1513 spa_strfree(spa->spa_comment);
1514 spa->spa_comment = NULL;
1517 spa_config_exit(spa, SCL_ALL, spa);
1521 * Load (or re-load) the current list of vdevs describing the active spares for
1522 * this pool. When this is called, we have some form of basic information in
1523 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1524 * then re-generate a more complete list including status information.
1527 spa_load_spares(spa_t *spa)
1536 * zdb opens both the current state of the pool and the
1537 * checkpointed state (if present), with a different spa_t.
1539 * As spare vdevs are shared among open pools, we skip loading
1540 * them when we load the checkpointed state of the pool.
1542 if (!spa_writeable(spa))
1546 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1549 * First, close and free any existing spare vdevs.
1551 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1552 vd = spa->spa_spares.sav_vdevs[i];
1554 /* Undo the call to spa_activate() below */
1555 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1556 B_FALSE)) != NULL && tvd->vdev_isspare)
1557 spa_spare_remove(tvd);
1562 if (spa->spa_spares.sav_vdevs)
1563 kmem_free(spa->spa_spares.sav_vdevs,
1564 spa->spa_spares.sav_count * sizeof (void *));
1566 if (spa->spa_spares.sav_config == NULL)
1569 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1570 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1572 spa->spa_spares.sav_count = (int)nspares;
1573 spa->spa_spares.sav_vdevs = NULL;
1579 * Construct the array of vdevs, opening them to get status in the
1580 * process. For each spare, there is potentially two different vdev_t
1581 * structures associated with it: one in the list of spares (used only
1582 * for basic validation purposes) and one in the active vdev
1583 * configuration (if it's spared in). During this phase we open and
1584 * validate each vdev on the spare list. If the vdev also exists in the
1585 * active configuration, then we also mark this vdev as an active spare.
1587 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1589 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1590 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1591 VDEV_ALLOC_SPARE) == 0);
1594 spa->spa_spares.sav_vdevs[i] = vd;
1596 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1597 B_FALSE)) != NULL) {
1598 if (!tvd->vdev_isspare)
1602 * We only mark the spare active if we were successfully
1603 * able to load the vdev. Otherwise, importing a pool
1604 * with a bad active spare would result in strange
1605 * behavior, because multiple pool would think the spare
1606 * is actively in use.
1608 * There is a vulnerability here to an equally bizarre
1609 * circumstance, where a dead active spare is later
1610 * brought back to life (onlined or otherwise). Given
1611 * the rarity of this scenario, and the extra complexity
1612 * it adds, we ignore the possibility.
1614 if (!vdev_is_dead(tvd))
1615 spa_spare_activate(tvd);
1619 vd->vdev_aux = &spa->spa_spares;
1621 if (vdev_open(vd) != 0)
1624 if (vdev_validate_aux(vd) == 0)
1629 * Recompute the stashed list of spares, with status information
1632 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1633 DATA_TYPE_NVLIST_ARRAY) == 0);
1635 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1637 for (i = 0; i < spa->spa_spares.sav_count; i++)
1638 spares[i] = vdev_config_generate(spa,
1639 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1640 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1641 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1642 for (i = 0; i < spa->spa_spares.sav_count; i++)
1643 nvlist_free(spares[i]);
1644 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1648 * Load (or re-load) the current list of vdevs describing the active l2cache for
1649 * this pool. When this is called, we have some form of basic information in
1650 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1651 * then re-generate a more complete list including status information.
1652 * Devices which are already active have their details maintained, and are
1656 spa_load_l2cache(spa_t *spa)
1660 int i, j, oldnvdevs;
1662 vdev_t *vd, **oldvdevs, **newvdevs;
1663 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1667 * zdb opens both the current state of the pool and the
1668 * checkpointed state (if present), with a different spa_t.
1670 * As L2 caches are part of the ARC which is shared among open
1671 * pools, we skip loading them when we load the checkpointed
1672 * state of the pool.
1674 if (!spa_writeable(spa))
1678 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1680 if (sav->sav_config != NULL) {
1681 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1682 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1683 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1689 oldvdevs = sav->sav_vdevs;
1690 oldnvdevs = sav->sav_count;
1691 sav->sav_vdevs = NULL;
1695 * Process new nvlist of vdevs.
1697 for (i = 0; i < nl2cache; i++) {
1698 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1702 for (j = 0; j < oldnvdevs; j++) {
1704 if (vd != NULL && guid == vd->vdev_guid) {
1706 * Retain previous vdev for add/remove ops.
1714 if (newvdevs[i] == NULL) {
1718 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1719 VDEV_ALLOC_L2CACHE) == 0);
1724 * Commit this vdev as an l2cache device,
1725 * even if it fails to open.
1727 spa_l2cache_add(vd);
1732 spa_l2cache_activate(vd);
1734 if (vdev_open(vd) != 0)
1737 (void) vdev_validate_aux(vd);
1739 if (!vdev_is_dead(vd))
1740 l2arc_add_vdev(spa, vd);
1745 * Purge vdevs that were dropped
1747 for (i = 0; i < oldnvdevs; i++) {
1752 ASSERT(vd->vdev_isl2cache);
1754 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1755 pool != 0ULL && l2arc_vdev_present(vd))
1756 l2arc_remove_vdev(vd);
1757 vdev_clear_stats(vd);
1763 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1765 if (sav->sav_config == NULL)
1768 sav->sav_vdevs = newvdevs;
1769 sav->sav_count = (int)nl2cache;
1772 * Recompute the stashed list of l2cache devices, with status
1773 * information this time.
1775 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1776 DATA_TYPE_NVLIST_ARRAY) == 0);
1778 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1779 for (i = 0; i < sav->sav_count; i++)
1780 l2cache[i] = vdev_config_generate(spa,
1781 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1782 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1783 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1785 for (i = 0; i < sav->sav_count; i++)
1786 nvlist_free(l2cache[i]);
1788 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1792 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1795 char *packed = NULL;
1800 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1804 nvsize = *(uint64_t *)db->db_data;
1805 dmu_buf_rele(db, FTAG);
1807 packed = kmem_alloc(nvsize, KM_SLEEP);
1808 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1811 error = nvlist_unpack(packed, nvsize, value, 0);
1812 kmem_free(packed, nvsize);
1818 * Concrete top-level vdevs that are not missing and are not logs. At every
1819 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1822 spa_healthy_core_tvds(spa_t *spa)
1824 vdev_t *rvd = spa->spa_root_vdev;
1827 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1828 vdev_t *vd = rvd->vdev_child[i];
1831 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1839 * Checks to see if the given vdev could not be opened, in which case we post a
1840 * sysevent to notify the autoreplace code that the device has been removed.
1843 spa_check_removed(vdev_t *vd)
1845 for (uint64_t c = 0; c < vd->vdev_children; c++)
1846 spa_check_removed(vd->vdev_child[c]);
1848 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1849 vdev_is_concrete(vd)) {
1850 zfs_post_autoreplace(vd->vdev_spa, vd);
1851 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1856 spa_check_for_missing_logs(spa_t *spa)
1858 vdev_t *rvd = spa->spa_root_vdev;
1861 * If we're doing a normal import, then build up any additional
1862 * diagnostic information about missing log devices.
1863 * We'll pass this up to the user for further processing.
1865 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1866 nvlist_t **child, *nv;
1869 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1871 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1873 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1874 vdev_t *tvd = rvd->vdev_child[c];
1877 * We consider a device as missing only if it failed
1878 * to open (i.e. offline or faulted is not considered
1881 if (tvd->vdev_islog &&
1882 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1883 child[idx++] = vdev_config_generate(spa, tvd,
1884 B_FALSE, VDEV_CONFIG_MISSING);
1889 fnvlist_add_nvlist_array(nv,
1890 ZPOOL_CONFIG_CHILDREN, child, idx);
1891 fnvlist_add_nvlist(spa->spa_load_info,
1892 ZPOOL_CONFIG_MISSING_DEVICES, nv);
1894 for (uint64_t i = 0; i < idx; i++)
1895 nvlist_free(child[i]);
1898 kmem_free(child, rvd->vdev_children * sizeof (char **));
1901 spa_load_failed(spa, "some log devices are missing");
1902 vdev_dbgmsg_print_tree(rvd, 2);
1903 return (SET_ERROR(ENXIO));
1906 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1907 vdev_t *tvd = rvd->vdev_child[c];
1909 if (tvd->vdev_islog &&
1910 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1911 spa_set_log_state(spa, SPA_LOG_CLEAR);
1912 spa_load_note(spa, "some log devices are "
1913 "missing, ZIL is dropped.");
1914 vdev_dbgmsg_print_tree(rvd, 2);
1924 * Check for missing log devices
1927 spa_check_logs(spa_t *spa)
1929 boolean_t rv = B_FALSE;
1930 dsl_pool_t *dp = spa_get_dsl(spa);
1932 switch (spa->spa_log_state) {
1933 case SPA_LOG_MISSING:
1934 /* need to recheck in case slog has been restored */
1935 case SPA_LOG_UNKNOWN:
1936 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1937 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1939 spa_set_log_state(spa, SPA_LOG_MISSING);
1946 spa_passivate_log(spa_t *spa)
1948 vdev_t *rvd = spa->spa_root_vdev;
1949 boolean_t slog_found = B_FALSE;
1951 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1953 if (!spa_has_slogs(spa))
1956 for (int c = 0; c < rvd->vdev_children; c++) {
1957 vdev_t *tvd = rvd->vdev_child[c];
1958 metaslab_group_t *mg = tvd->vdev_mg;
1960 if (tvd->vdev_islog) {
1961 metaslab_group_passivate(mg);
1962 slog_found = B_TRUE;
1966 return (slog_found);
1970 spa_activate_log(spa_t *spa)
1972 vdev_t *rvd = spa->spa_root_vdev;
1974 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1976 for (int c = 0; c < rvd->vdev_children; c++) {
1977 vdev_t *tvd = rvd->vdev_child[c];
1978 metaslab_group_t *mg = tvd->vdev_mg;
1980 if (tvd->vdev_islog)
1981 metaslab_group_activate(mg);
1986 spa_reset_logs(spa_t *spa)
1990 error = dmu_objset_find(spa_name(spa), zil_reset,
1991 NULL, DS_FIND_CHILDREN);
1994 * We successfully offlined the log device, sync out the
1995 * current txg so that the "stubby" block can be removed
1998 txg_wait_synced(spa->spa_dsl_pool, 0);
2004 spa_aux_check_removed(spa_aux_vdev_t *sav)
2008 for (i = 0; i < sav->sav_count; i++)
2009 spa_check_removed(sav->sav_vdevs[i]);
2013 spa_claim_notify(zio_t *zio)
2015 spa_t *spa = zio->io_spa;
2020 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2021 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2022 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2023 mutex_exit(&spa->spa_props_lock);
2026 typedef struct spa_load_error {
2027 uint64_t sle_meta_count;
2028 uint64_t sle_data_count;
2032 spa_load_verify_done(zio_t *zio)
2034 blkptr_t *bp = zio->io_bp;
2035 spa_load_error_t *sle = zio->io_private;
2036 dmu_object_type_t type = BP_GET_TYPE(bp);
2037 int error = zio->io_error;
2038 spa_t *spa = zio->io_spa;
2040 abd_free(zio->io_abd);
2042 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2043 type != DMU_OT_INTENT_LOG)
2044 atomic_inc_64(&sle->sle_meta_count);
2046 atomic_inc_64(&sle->sle_data_count);
2049 mutex_enter(&spa->spa_scrub_lock);
2050 spa->spa_load_verify_ios--;
2051 cv_broadcast(&spa->spa_scrub_io_cv);
2052 mutex_exit(&spa->spa_scrub_lock);
2056 * Maximum number of concurrent scrub i/os to create while verifying
2057 * a pool while importing it.
2059 int spa_load_verify_maxinflight = 10000;
2060 boolean_t spa_load_verify_metadata = B_TRUE;
2061 boolean_t spa_load_verify_data = B_TRUE;
2063 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
2064 &spa_load_verify_maxinflight, 0,
2065 "Maximum number of concurrent scrub I/Os to create while verifying a "
2066 "pool while importing it");
2068 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
2069 &spa_load_verify_metadata, 0,
2070 "Check metadata on import?");
2072 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
2073 &spa_load_verify_data, 0,
2074 "Check user data on import?");
2078 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2079 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2081 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2084 * Note: normally this routine will not be called if
2085 * spa_load_verify_metadata is not set. However, it may be useful
2086 * to manually set the flag after the traversal has begun.
2088 if (!spa_load_verify_metadata)
2090 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2094 size_t size = BP_GET_PSIZE(bp);
2096 mutex_enter(&spa->spa_scrub_lock);
2097 while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight)
2098 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2099 spa->spa_load_verify_ios++;
2100 mutex_exit(&spa->spa_scrub_lock);
2102 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2103 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2104 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2105 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2111 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2113 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2114 return (SET_ERROR(ENAMETOOLONG));
2120 spa_load_verify(spa_t *spa)
2123 spa_load_error_t sle = { 0 };
2124 zpool_load_policy_t policy;
2125 boolean_t verify_ok = B_FALSE;
2128 zpool_get_load_policy(spa->spa_config, &policy);
2130 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2133 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2134 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2135 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2137 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2141 rio = zio_root(spa, NULL, &sle,
2142 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2144 if (spa_load_verify_metadata) {
2145 if (spa->spa_extreme_rewind) {
2146 spa_load_note(spa, "performing a complete scan of the "
2147 "pool since extreme rewind is on. This may take "
2148 "a very long time.\n (spa_load_verify_data=%u, "
2149 "spa_load_verify_metadata=%u)",
2150 spa_load_verify_data, spa_load_verify_metadata);
2152 error = traverse_pool(spa, spa->spa_verify_min_txg,
2153 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2154 spa_load_verify_cb, rio);
2157 (void) zio_wait(rio);
2159 spa->spa_load_meta_errors = sle.sle_meta_count;
2160 spa->spa_load_data_errors = sle.sle_data_count;
2162 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2163 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2164 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2165 (u_longlong_t)sle.sle_data_count);
2168 if (spa_load_verify_dryrun ||
2169 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2170 sle.sle_data_count <= policy.zlp_maxdata)) {
2174 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2175 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2177 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2178 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2179 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2180 VERIFY(nvlist_add_int64(spa->spa_load_info,
2181 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2182 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2183 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2185 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2188 if (spa_load_verify_dryrun)
2192 if (error != ENXIO && error != EIO)
2193 error = SET_ERROR(EIO);
2197 return (verify_ok ? 0 : EIO);
2201 * Find a value in the pool props object.
2204 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2206 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2207 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2211 * Find a value in the pool directory object.
2214 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2216 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2217 name, sizeof (uint64_t), 1, val);
2219 if (error != 0 && (error != ENOENT || log_enoent)) {
2220 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2221 "[error=%d]", name, error);
2228 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2230 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2231 return (SET_ERROR(err));
2235 spa_spawn_aux_threads(spa_t *spa)
2237 ASSERT(spa_writeable(spa));
2239 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2241 spa_start_indirect_condensing_thread(spa);
2243 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2244 spa->spa_checkpoint_discard_zthr =
2245 zthr_create(spa_checkpoint_discard_thread_check,
2246 spa_checkpoint_discard_thread, spa);
2250 * Fix up config after a partly-completed split. This is done with the
2251 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2252 * pool have that entry in their config, but only the splitting one contains
2253 * a list of all the guids of the vdevs that are being split off.
2255 * This function determines what to do with that list: either rejoin
2256 * all the disks to the pool, or complete the splitting process. To attempt
2257 * the rejoin, each disk that is offlined is marked online again, and
2258 * we do a reopen() call. If the vdev label for every disk that was
2259 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2260 * then we call vdev_split() on each disk, and complete the split.
2262 * Otherwise we leave the config alone, with all the vdevs in place in
2263 * the original pool.
2266 spa_try_repair(spa_t *spa, nvlist_t *config)
2273 boolean_t attempt_reopen;
2275 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2278 /* check that the config is complete */
2279 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2280 &glist, &gcount) != 0)
2283 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2285 /* attempt to online all the vdevs & validate */
2286 attempt_reopen = B_TRUE;
2287 for (i = 0; i < gcount; i++) {
2288 if (glist[i] == 0) /* vdev is hole */
2291 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2292 if (vd[i] == NULL) {
2294 * Don't bother attempting to reopen the disks;
2295 * just do the split.
2297 attempt_reopen = B_FALSE;
2299 /* attempt to re-online it */
2300 vd[i]->vdev_offline = B_FALSE;
2304 if (attempt_reopen) {
2305 vdev_reopen(spa->spa_root_vdev);
2307 /* check each device to see what state it's in */
2308 for (extracted = 0, i = 0; i < gcount; i++) {
2309 if (vd[i] != NULL &&
2310 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2317 * If every disk has been moved to the new pool, or if we never
2318 * even attempted to look at them, then we split them off for
2321 if (!attempt_reopen || gcount == extracted) {
2322 for (i = 0; i < gcount; i++)
2325 vdev_reopen(spa->spa_root_vdev);
2328 kmem_free(vd, gcount * sizeof (vdev_t *));
2332 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2334 char *ereport = FM_EREPORT_ZFS_POOL;
2337 spa->spa_load_state = state;
2339 gethrestime(&spa->spa_loaded_ts);
2340 error = spa_load_impl(spa, type, &ereport);
2343 * Don't count references from objsets that are already closed
2344 * and are making their way through the eviction process.
2346 spa_evicting_os_wait(spa);
2347 spa->spa_minref = refcount_count(&spa->spa_refcount);
2349 if (error != EEXIST) {
2350 spa->spa_loaded_ts.tv_sec = 0;
2351 spa->spa_loaded_ts.tv_nsec = 0;
2353 if (error != EBADF) {
2354 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2357 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2364 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2365 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2366 * spa's per-vdev ZAP list.
2369 vdev_count_verify_zaps(vdev_t *vd)
2371 spa_t *spa = vd->vdev_spa;
2373 if (vd->vdev_top_zap != 0) {
2375 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2376 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2378 if (vd->vdev_leaf_zap != 0) {
2380 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2381 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2384 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2385 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2392 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2396 uint64_t myhostid = 0;
2398 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2399 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2400 hostname = fnvlist_lookup_string(mos_config,
2401 ZPOOL_CONFIG_HOSTNAME);
2403 myhostid = zone_get_hostid(NULL);
2405 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2406 cmn_err(CE_WARN, "pool '%s' could not be "
2407 "loaded as it was last accessed by "
2408 "another system (host: %s hostid: 0x%llx). "
2409 "See: http://illumos.org/msg/ZFS-8000-EY",
2410 spa_name(spa), hostname, (u_longlong_t)hostid);
2411 spa_load_failed(spa, "hostid verification failed: pool "
2412 "last accessed by host: %s (hostid: 0x%llx)",
2413 hostname, (u_longlong_t)hostid);
2414 return (SET_ERROR(EBADF));
2422 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2425 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2432 * Versioning wasn't explicitly added to the label until later, so if
2433 * it's not present treat it as the initial version.
2435 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2436 &spa->spa_ubsync.ub_version) != 0)
2437 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2439 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2440 spa_load_failed(spa, "invalid config provided: '%s' missing",
2441 ZPOOL_CONFIG_POOL_GUID);
2442 return (SET_ERROR(EINVAL));
2446 * If we are doing an import, ensure that the pool is not already
2447 * imported by checking if its pool guid already exists in the
2450 * The only case that we allow an already imported pool to be
2451 * imported again, is when the pool is checkpointed and we want to
2452 * look at its checkpointed state from userland tools like zdb.
2455 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2456 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2457 spa_guid_exists(pool_guid, 0)) {
2459 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2460 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2461 spa_guid_exists(pool_guid, 0) &&
2462 !spa_importing_readonly_checkpoint(spa)) {
2464 spa_load_failed(spa, "a pool with guid %llu is already open",
2465 (u_longlong_t)pool_guid);
2466 return (SET_ERROR(EEXIST));
2469 spa->spa_config_guid = pool_guid;
2471 nvlist_free(spa->spa_load_info);
2472 spa->spa_load_info = fnvlist_alloc();
2474 ASSERT(spa->spa_comment == NULL);
2475 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2476 spa->spa_comment = spa_strdup(comment);
2478 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2479 &spa->spa_config_txg);
2481 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2482 spa->spa_config_splitting = fnvlist_dup(nvl);
2484 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2485 spa_load_failed(spa, "invalid config provided: '%s' missing",
2486 ZPOOL_CONFIG_VDEV_TREE);
2487 return (SET_ERROR(EINVAL));
2491 * Create "The Godfather" zio to hold all async IOs
2493 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2495 for (int i = 0; i < max_ncpus; i++) {
2496 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2497 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2498 ZIO_FLAG_GODFATHER);
2502 * Parse the configuration into a vdev tree. We explicitly set the
2503 * value that will be returned by spa_version() since parsing the
2504 * configuration requires knowing the version number.
2506 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2507 parse = (type == SPA_IMPORT_EXISTING ?
2508 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2509 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2510 spa_config_exit(spa, SCL_ALL, FTAG);
2513 spa_load_failed(spa, "unable to parse config [error=%d]",
2518 ASSERT(spa->spa_root_vdev == rvd);
2519 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2520 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2522 if (type != SPA_IMPORT_ASSEMBLE) {
2523 ASSERT(spa_guid(spa) == pool_guid);
2530 * Recursively open all vdevs in the vdev tree. This function is called twice:
2531 * first with the untrusted config, then with the trusted config.
2534 spa_ld_open_vdevs(spa_t *spa)
2539 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2540 * missing/unopenable for the root vdev to be still considered openable.
2542 if (spa->spa_trust_config) {
2543 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2544 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2545 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2546 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2547 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2549 spa->spa_missing_tvds_allowed = 0;
2552 spa->spa_missing_tvds_allowed =
2553 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2555 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2556 error = vdev_open(spa->spa_root_vdev);
2557 spa_config_exit(spa, SCL_ALL, FTAG);
2559 if (spa->spa_missing_tvds != 0) {
2560 spa_load_note(spa, "vdev tree has %lld missing top-level "
2561 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2562 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2564 * Although theoretically we could allow users to open
2565 * incomplete pools in RW mode, we'd need to add a lot
2566 * of extra logic (e.g. adjust pool space to account
2567 * for missing vdevs).
2568 * This limitation also prevents users from accidentally
2569 * opening the pool in RW mode during data recovery and
2570 * damaging it further.
2572 spa_load_note(spa, "pools with missing top-level "
2573 "vdevs can only be opened in read-only mode.");
2574 error = SET_ERROR(ENXIO);
2576 spa_load_note(spa, "current settings allow for maximum "
2577 "%lld missing top-level vdevs at this stage.",
2578 (u_longlong_t)spa->spa_missing_tvds_allowed);
2582 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2585 if (spa->spa_missing_tvds != 0 || error != 0)
2586 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2592 * We need to validate the vdev labels against the configuration that
2593 * we have in hand. This function is called twice: first with an untrusted
2594 * config, then with a trusted config. The validation is more strict when the
2595 * config is trusted.
2598 spa_ld_validate_vdevs(spa_t *spa)
2601 vdev_t *rvd = spa->spa_root_vdev;
2603 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2604 error = vdev_validate(rvd);
2605 spa_config_exit(spa, SCL_ALL, FTAG);
2608 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2612 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2613 spa_load_failed(spa, "cannot open vdev tree after invalidating "
2615 vdev_dbgmsg_print_tree(rvd, 2);
2616 return (SET_ERROR(ENXIO));
2623 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2625 spa->spa_state = POOL_STATE_ACTIVE;
2626 spa->spa_ubsync = spa->spa_uberblock;
2627 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2628 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2629 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2630 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2631 spa->spa_claim_max_txg = spa->spa_first_txg;
2632 spa->spa_prev_software_version = ub->ub_software_version;
2636 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2638 vdev_t *rvd = spa->spa_root_vdev;
2640 uberblock_t *ub = &spa->spa_uberblock;
2643 * If we are opening the checkpointed state of the pool by
2644 * rewinding to it, at this point we will have written the
2645 * checkpointed uberblock to the vdev labels, so searching
2646 * the labels will find the right uberblock. However, if
2647 * we are opening the checkpointed state read-only, we have
2648 * not modified the labels. Therefore, we must ignore the
2649 * labels and continue using the spa_uberblock that was set
2650 * by spa_ld_checkpoint_rewind.
2652 * Note that it would be fine to ignore the labels when
2653 * rewinding (opening writeable) as well. However, if we
2654 * crash just after writing the labels, we will end up
2655 * searching the labels. Doing so in the common case means
2656 * that this code path gets exercised normally, rather than
2657 * just in the edge case.
2659 if (ub->ub_checkpoint_txg != 0 &&
2660 spa_importing_readonly_checkpoint(spa)) {
2661 spa_ld_select_uberblock_done(spa, ub);
2666 * Find the best uberblock.
2668 vdev_uberblock_load(rvd, ub, &label);
2671 * If we weren't able to find a single valid uberblock, return failure.
2673 if (ub->ub_txg == 0) {
2675 spa_load_failed(spa, "no valid uberblock found");
2676 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2679 spa_load_note(spa, "using uberblock with txg=%llu",
2680 (u_longlong_t)ub->ub_txg);
2683 * If the pool has an unsupported version we can't open it.
2685 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2687 spa_load_failed(spa, "version %llu is not supported",
2688 (u_longlong_t)ub->ub_version);
2689 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2692 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2696 * If we weren't able to find what's necessary for reading the
2697 * MOS in the label, return failure.
2699 if (label == NULL) {
2700 spa_load_failed(spa, "label config unavailable");
2701 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2705 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2708 spa_load_failed(spa, "invalid label: '%s' missing",
2709 ZPOOL_CONFIG_FEATURES_FOR_READ);
2710 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2715 * Update our in-core representation with the definitive values
2718 nvlist_free(spa->spa_label_features);
2719 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2725 * Look through entries in the label nvlist's features_for_read. If
2726 * there is a feature listed there which we don't understand then we
2727 * cannot open a pool.
2729 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2730 nvlist_t *unsup_feat;
2732 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2735 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2737 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2738 if (!zfeature_is_supported(nvpair_name(nvp))) {
2739 VERIFY(nvlist_add_string(unsup_feat,
2740 nvpair_name(nvp), "") == 0);
2744 if (!nvlist_empty(unsup_feat)) {
2745 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2746 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2747 nvlist_free(unsup_feat);
2748 spa_load_failed(spa, "some features are unsupported");
2749 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2753 nvlist_free(unsup_feat);
2756 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2757 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2758 spa_try_repair(spa, spa->spa_config);
2759 spa_config_exit(spa, SCL_ALL, FTAG);
2760 nvlist_free(spa->spa_config_splitting);
2761 spa->spa_config_splitting = NULL;
2765 * Initialize internal SPA structures.
2767 spa_ld_select_uberblock_done(spa, ub);
2773 spa_ld_open_rootbp(spa_t *spa)
2776 vdev_t *rvd = spa->spa_root_vdev;
2778 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2780 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2781 "[error=%d]", error);
2782 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2784 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2790 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2791 boolean_t reloading)
2793 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2794 nvlist_t *nv, *mos_config, *policy;
2795 int error = 0, copy_error;
2796 uint64_t healthy_tvds, healthy_tvds_mos;
2797 uint64_t mos_config_txg;
2799 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2801 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2804 * If we're assembling a pool from a split, the config provided is
2805 * already trusted so there is nothing to do.
2807 if (type == SPA_IMPORT_ASSEMBLE)
2810 healthy_tvds = spa_healthy_core_tvds(spa);
2812 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2814 spa_load_failed(spa, "unable to retrieve MOS config");
2815 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2819 * If we are doing an open, pool owner wasn't verified yet, thus do
2820 * the verification here.
2822 if (spa->spa_load_state == SPA_LOAD_OPEN) {
2823 error = spa_verify_host(spa, mos_config);
2825 nvlist_free(mos_config);
2830 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2832 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2835 * Build a new vdev tree from the trusted config
2837 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2840 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2841 * obtained by scanning /dev/dsk, then it will have the right vdev
2842 * paths. We update the trusted MOS config with this information.
2843 * We first try to copy the paths with vdev_copy_path_strict, which
2844 * succeeds only when both configs have exactly the same vdev tree.
2845 * If that fails, we fall back to a more flexible method that has a
2846 * best effort policy.
2848 copy_error = vdev_copy_path_strict(rvd, mrvd);
2849 if (copy_error != 0 || spa_load_print_vdev_tree) {
2850 spa_load_note(spa, "provided vdev tree:");
2851 vdev_dbgmsg_print_tree(rvd, 2);
2852 spa_load_note(spa, "MOS vdev tree:");
2853 vdev_dbgmsg_print_tree(mrvd, 2);
2855 if (copy_error != 0) {
2856 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2857 "back to vdev_copy_path_relaxed");
2858 vdev_copy_path_relaxed(rvd, mrvd);
2863 spa->spa_root_vdev = mrvd;
2865 spa_config_exit(spa, SCL_ALL, FTAG);
2868 * We will use spa_config if we decide to reload the spa or if spa_load
2869 * fails and we rewind. We must thus regenerate the config using the
2870 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
2871 * pass settings on how to load the pool and is not stored in the MOS.
2872 * We copy it over to our new, trusted config.
2874 mos_config_txg = fnvlist_lookup_uint64(mos_config,
2875 ZPOOL_CONFIG_POOL_TXG);
2876 nvlist_free(mos_config);
2877 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
2878 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
2880 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
2881 spa_config_set(spa, mos_config);
2882 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2885 * Now that we got the config from the MOS, we should be more strict
2886 * in checking blkptrs and can make assumptions about the consistency
2887 * of the vdev tree. spa_trust_config must be set to true before opening
2888 * vdevs in order for them to be writeable.
2890 spa->spa_trust_config = B_TRUE;
2893 * Open and validate the new vdev tree
2895 error = spa_ld_open_vdevs(spa);
2899 error = spa_ld_validate_vdevs(spa);
2903 if (copy_error != 0 || spa_load_print_vdev_tree) {
2904 spa_load_note(spa, "final vdev tree:");
2905 vdev_dbgmsg_print_tree(rvd, 2);
2908 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2909 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2911 * Sanity check to make sure that we are indeed loading the
2912 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2913 * in the config provided and they happened to be the only ones
2914 * to have the latest uberblock, we could involuntarily perform
2915 * an extreme rewind.
2917 healthy_tvds_mos = spa_healthy_core_tvds(spa);
2918 if (healthy_tvds_mos - healthy_tvds >=
2919 SPA_SYNC_MIN_VDEVS) {
2920 spa_load_note(spa, "config provided misses too many "
2921 "top-level vdevs compared to MOS (%lld vs %lld). ",
2922 (u_longlong_t)healthy_tvds,
2923 (u_longlong_t)healthy_tvds_mos);
2924 spa_load_note(spa, "vdev tree:");
2925 vdev_dbgmsg_print_tree(rvd, 2);
2927 spa_load_failed(spa, "config was already "
2928 "provided from MOS. Aborting.");
2929 return (spa_vdev_err(rvd,
2930 VDEV_AUX_CORRUPT_DATA, EIO));
2932 spa_load_note(spa, "spa must be reloaded using MOS "
2934 return (SET_ERROR(EAGAIN));
2938 error = spa_check_for_missing_logs(spa);
2940 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2942 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2943 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2944 "guid sum (%llu != %llu)",
2945 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2946 (u_longlong_t)rvd->vdev_guid_sum);
2947 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2955 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2958 vdev_t *rvd = spa->spa_root_vdev;
2961 * Everything that we read before spa_remove_init() must be stored
2962 * on concreted vdevs. Therefore we do this as early as possible.
2964 error = spa_remove_init(spa);
2966 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2968 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2972 * Retrieve information needed to condense indirect vdev mappings.
2974 error = spa_condense_init(spa);
2976 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2978 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2985 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2988 vdev_t *rvd = spa->spa_root_vdev;
2990 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2991 boolean_t missing_feat_read = B_FALSE;
2992 nvlist_t *unsup_feat, *enabled_feat;
2994 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2995 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2996 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2999 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3000 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3001 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3004 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3005 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3006 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3009 enabled_feat = fnvlist_alloc();
3010 unsup_feat = fnvlist_alloc();
3012 if (!spa_features_check(spa, B_FALSE,
3013 unsup_feat, enabled_feat))
3014 missing_feat_read = B_TRUE;
3016 if (spa_writeable(spa) ||
3017 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3018 if (!spa_features_check(spa, B_TRUE,
3019 unsup_feat, enabled_feat)) {
3020 *missing_feat_writep = B_TRUE;
3024 fnvlist_add_nvlist(spa->spa_load_info,
3025 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3027 if (!nvlist_empty(unsup_feat)) {
3028 fnvlist_add_nvlist(spa->spa_load_info,
3029 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3032 fnvlist_free(enabled_feat);
3033 fnvlist_free(unsup_feat);
3035 if (!missing_feat_read) {
3036 fnvlist_add_boolean(spa->spa_load_info,
3037 ZPOOL_CONFIG_CAN_RDONLY);
3041 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3042 * twofold: to determine whether the pool is available for
3043 * import in read-write mode and (if it is not) whether the
3044 * pool is available for import in read-only mode. If the pool
3045 * is available for import in read-write mode, it is displayed
3046 * as available in userland; if it is not available for import
3047 * in read-only mode, it is displayed as unavailable in
3048 * userland. If the pool is available for import in read-only
3049 * mode but not read-write mode, it is displayed as unavailable
3050 * in userland with a special note that the pool is actually
3051 * available for open in read-only mode.
3053 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3054 * missing a feature for write, we must first determine whether
3055 * the pool can be opened read-only before returning to
3056 * userland in order to know whether to display the
3057 * abovementioned note.
3059 if (missing_feat_read || (*missing_feat_writep &&
3060 spa_writeable(spa))) {
3061 spa_load_failed(spa, "pool uses unsupported features");
3062 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3067 * Load refcounts for ZFS features from disk into an in-memory
3068 * cache during SPA initialization.
3070 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3073 error = feature_get_refcount_from_disk(spa,
3074 &spa_feature_table[i], &refcount);
3076 spa->spa_feat_refcount_cache[i] = refcount;
3077 } else if (error == ENOTSUP) {
3078 spa->spa_feat_refcount_cache[i] =
3079 SPA_FEATURE_DISABLED;
3081 spa_load_failed(spa, "error getting refcount "
3082 "for feature %s [error=%d]",
3083 spa_feature_table[i].fi_guid, error);
3084 return (spa_vdev_err(rvd,
3085 VDEV_AUX_CORRUPT_DATA, EIO));
3090 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3091 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3092 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3093 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3100 spa_ld_load_special_directories(spa_t *spa)
3103 vdev_t *rvd = spa->spa_root_vdev;
3105 spa->spa_is_initializing = B_TRUE;
3106 error = dsl_pool_open(spa->spa_dsl_pool);
3107 spa->spa_is_initializing = B_FALSE;
3109 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3110 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3117 spa_ld_get_props(spa_t *spa)
3121 vdev_t *rvd = spa->spa_root_vdev;
3123 /* Grab the secret checksum salt from the MOS. */
3124 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3125 DMU_POOL_CHECKSUM_SALT, 1,
3126 sizeof (spa->spa_cksum_salt.zcs_bytes),
3127 spa->spa_cksum_salt.zcs_bytes);
3128 if (error == ENOENT) {
3129 /* Generate a new salt for subsequent use */
3130 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3131 sizeof (spa->spa_cksum_salt.zcs_bytes));
3132 } else if (error != 0) {
3133 spa_load_failed(spa, "unable to retrieve checksum salt from "
3134 "MOS [error=%d]", error);
3135 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3138 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3139 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3140 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3142 spa_load_failed(spa, "error opening deferred-frees bpobj "
3143 "[error=%d]", error);
3144 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3148 * Load the bit that tells us to use the new accounting function
3149 * (raid-z deflation). If we have an older pool, this will not
3152 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3153 if (error != 0 && error != ENOENT)
3154 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3156 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3157 &spa->spa_creation_version, B_FALSE);
3158 if (error != 0 && error != ENOENT)
3159 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3162 * Load the persistent error log. If we have an older pool, this will
3165 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3167 if (error != 0 && error != ENOENT)
3168 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3170 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3171 &spa->spa_errlog_scrub, B_FALSE);
3172 if (error != 0 && error != ENOENT)
3173 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3176 * Load the history object. If we have an older pool, this
3177 * will not be present.
3179 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3180 if (error != 0 && error != ENOENT)
3181 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3184 * Load the per-vdev ZAP map. If we have an older pool, this will not
3185 * be present; in this case, defer its creation to a later time to
3186 * avoid dirtying the MOS this early / out of sync context. See
3187 * spa_sync_config_object.
3190 /* The sentinel is only available in the MOS config. */
3191 nvlist_t *mos_config;
3192 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3193 spa_load_failed(spa, "unable to retrieve MOS config");
3194 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3197 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3198 &spa->spa_all_vdev_zaps, B_FALSE);
3200 if (error == ENOENT) {
3201 VERIFY(!nvlist_exists(mos_config,
3202 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3203 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3204 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3205 } else if (error != 0) {
3206 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3207 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3209 * An older version of ZFS overwrote the sentinel value, so
3210 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3211 * destruction to later; see spa_sync_config_object.
3213 spa->spa_avz_action = AVZ_ACTION_DESTROY;
3215 * We're assuming that no vdevs have had their ZAPs created
3216 * before this. Better be sure of it.
3218 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3220 nvlist_free(mos_config);
3222 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3224 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3226 if (error && error != ENOENT)
3227 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3230 uint64_t autoreplace;
3232 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3233 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3234 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3235 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3236 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3237 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3238 &spa->spa_dedup_ditto);
3240 spa->spa_autoreplace = (autoreplace != 0);
3244 * If we are importing a pool with missing top-level vdevs,
3245 * we enforce that the pool doesn't panic or get suspended on
3246 * error since the likelihood of missing data is extremely high.
3248 if (spa->spa_missing_tvds > 0 &&
3249 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3250 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3251 spa_load_note(spa, "forcing failmode to 'continue' "
3252 "as some top level vdevs are missing");
3253 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3260 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3263 vdev_t *rvd = spa->spa_root_vdev;
3266 * If we're assembling the pool from the split-off vdevs of
3267 * an existing pool, we don't want to attach the spares & cache
3272 * Load any hot spares for this pool.
3274 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3276 if (error != 0 && error != ENOENT)
3277 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3278 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3279 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3280 if (load_nvlist(spa, spa->spa_spares.sav_object,
3281 &spa->spa_spares.sav_config) != 0) {
3282 spa_load_failed(spa, "error loading spares nvlist");
3283 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3286 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3287 spa_load_spares(spa);
3288 spa_config_exit(spa, SCL_ALL, FTAG);
3289 } else if (error == 0) {
3290 spa->spa_spares.sav_sync = B_TRUE;
3294 * Load any level 2 ARC devices for this pool.
3296 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3297 &spa->spa_l2cache.sav_object, B_FALSE);
3298 if (error != 0 && error != ENOENT)
3299 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3300 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3301 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3302 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3303 &spa->spa_l2cache.sav_config) != 0) {
3304 spa_load_failed(spa, "error loading l2cache nvlist");
3305 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3308 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3309 spa_load_l2cache(spa);
3310 spa_config_exit(spa, SCL_ALL, FTAG);
3311 } else if (error == 0) {
3312 spa->spa_l2cache.sav_sync = B_TRUE;
3319 spa_ld_load_vdev_metadata(spa_t *spa)
3322 vdev_t *rvd = spa->spa_root_vdev;
3325 * If the 'autoreplace' property is set, then post a resource notifying
3326 * the ZFS DE that it should not issue any faults for unopenable
3327 * devices. We also iterate over the vdevs, and post a sysevent for any
3328 * unopenable vdevs so that the normal autoreplace handler can take
3331 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3332 spa_check_removed(spa->spa_root_vdev);
3334 * For the import case, this is done in spa_import(), because
3335 * at this point we're using the spare definitions from
3336 * the MOS config, not necessarily from the userland config.
3338 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3339 spa_aux_check_removed(&spa->spa_spares);
3340 spa_aux_check_removed(&spa->spa_l2cache);
3345 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3347 error = vdev_load(rvd);
3349 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3350 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3354 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3356 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3357 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3358 spa_config_exit(spa, SCL_ALL, FTAG);
3364 spa_ld_load_dedup_tables(spa_t *spa)
3367 vdev_t *rvd = spa->spa_root_vdev;
3369 error = ddt_load(spa);
3371 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3372 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3379 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3381 vdev_t *rvd = spa->spa_root_vdev;
3383 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3384 boolean_t missing = spa_check_logs(spa);
3386 if (spa->spa_missing_tvds != 0) {
3387 spa_load_note(spa, "spa_check_logs failed "
3388 "so dropping the logs");
3390 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3391 spa_load_failed(spa, "spa_check_logs failed");
3392 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3402 spa_ld_verify_pool_data(spa_t *spa)
3405 vdev_t *rvd = spa->spa_root_vdev;
3408 * We've successfully opened the pool, verify that we're ready
3409 * to start pushing transactions.
3411 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3412 error = spa_load_verify(spa);
3414 spa_load_failed(spa, "spa_load_verify failed "
3415 "[error=%d]", error);
3416 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3425 spa_ld_claim_log_blocks(spa_t *spa)
3428 dsl_pool_t *dp = spa_get_dsl(spa);
3431 * Claim log blocks that haven't been committed yet.
3432 * This must all happen in a single txg.
3433 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3434 * invoked from zil_claim_log_block()'s i/o done callback.
3435 * Price of rollback is that we abandon the log.
3437 spa->spa_claiming = B_TRUE;
3439 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3440 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3441 zil_claim, tx, DS_FIND_CHILDREN);
3444 spa->spa_claiming = B_FALSE;
3446 spa_set_log_state(spa, SPA_LOG_GOOD);
3450 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3451 boolean_t update_config_cache)
3453 vdev_t *rvd = spa->spa_root_vdev;
3454 int need_update = B_FALSE;
3457 * If the config cache is stale, or we have uninitialized
3458 * metaslabs (see spa_vdev_add()), then update the config.
3460 * If this is a verbatim import, trust the current
3461 * in-core spa_config and update the disk labels.
3463 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3464 spa->spa_load_state == SPA_LOAD_IMPORT ||
3465 spa->spa_load_state == SPA_LOAD_RECOVER ||
3466 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3467 need_update = B_TRUE;
3469 for (int c = 0; c < rvd->vdev_children; c++)
3470 if (rvd->vdev_child[c]->vdev_ms_array == 0)
3471 need_update = B_TRUE;
3474 * Update the config cache asychronously in case we're the
3475 * root pool, in which case the config cache isn't writable yet.
3478 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3482 spa_ld_prepare_for_reload(spa_t *spa)
3484 int mode = spa->spa_mode;
3485 int async_suspended = spa->spa_async_suspended;
3488 spa_deactivate(spa);
3489 spa_activate(spa, mode);
3492 * We save the value of spa_async_suspended as it gets reset to 0 by
3493 * spa_unload(). We want to restore it back to the original value before
3494 * returning as we might be calling spa_async_resume() later.
3496 spa->spa_async_suspended = async_suspended;
3500 spa_ld_read_checkpoint_txg(spa_t *spa)
3502 uberblock_t checkpoint;
3505 ASSERT0(spa->spa_checkpoint_txg);
3506 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3508 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3509 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3510 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3512 if (error == ENOENT)
3518 ASSERT3U(checkpoint.ub_txg, !=, 0);
3519 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3520 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3521 spa->spa_checkpoint_txg = checkpoint.ub_txg;
3522 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3528 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3532 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3533 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3536 * Never trust the config that is provided unless we are assembling
3537 * a pool following a split.
3538 * This means don't trust blkptrs and the vdev tree in general. This
3539 * also effectively puts the spa in read-only mode since
3540 * spa_writeable() checks for spa_trust_config to be true.
3541 * We will later load a trusted config from the MOS.
3543 if (type != SPA_IMPORT_ASSEMBLE)
3544 spa->spa_trust_config = B_FALSE;
3547 * Parse the config provided to create a vdev tree.
3549 error = spa_ld_parse_config(spa, type);
3554 * Now that we have the vdev tree, try to open each vdev. This involves
3555 * opening the underlying physical device, retrieving its geometry and
3556 * probing the vdev with a dummy I/O. The state of each vdev will be set
3557 * based on the success of those operations. After this we'll be ready
3558 * to read from the vdevs.
3560 error = spa_ld_open_vdevs(spa);
3565 * Read the label of each vdev and make sure that the GUIDs stored
3566 * there match the GUIDs in the config provided.
3567 * If we're assembling a new pool that's been split off from an
3568 * existing pool, the labels haven't yet been updated so we skip
3569 * validation for now.
3571 if (type != SPA_IMPORT_ASSEMBLE) {
3572 error = spa_ld_validate_vdevs(spa);
3578 * Read all vdev labels to find the best uberblock (i.e. latest,
3579 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3580 * get the list of features required to read blkptrs in the MOS from
3581 * the vdev label with the best uberblock and verify that our version
3582 * of zfs supports them all.
3584 error = spa_ld_select_uberblock(spa, type);
3589 * Pass that uberblock to the dsl_pool layer which will open the root
3590 * blkptr. This blkptr points to the latest version of the MOS and will
3591 * allow us to read its contents.
3593 error = spa_ld_open_rootbp(spa);
3601 spa_ld_checkpoint_rewind(spa_t *spa)
3603 uberblock_t checkpoint;
3606 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3607 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3609 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3610 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3611 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3614 spa_load_failed(spa, "unable to retrieve checkpointed "
3615 "uberblock from the MOS config [error=%d]", error);
3617 if (error == ENOENT)
3618 error = ZFS_ERR_NO_CHECKPOINT;
3623 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3624 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3627 * We need to update the txg and timestamp of the checkpointed
3628 * uberblock to be higher than the latest one. This ensures that
3629 * the checkpointed uberblock is selected if we were to close and
3630 * reopen the pool right after we've written it in the vdev labels.
3631 * (also see block comment in vdev_uberblock_compare)
3633 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3634 checkpoint.ub_timestamp = gethrestime_sec();
3637 * Set current uberblock to be the checkpointed uberblock.
3639 spa->spa_uberblock = checkpoint;
3642 * If we are doing a normal rewind, then the pool is open for
3643 * writing and we sync the "updated" checkpointed uberblock to
3644 * disk. Once this is done, we've basically rewound the whole
3645 * pool and there is no way back.
3647 * There are cases when we don't want to attempt and sync the
3648 * checkpointed uberblock to disk because we are opening a
3649 * pool as read-only. Specifically, verifying the checkpointed
3650 * state with zdb, and importing the checkpointed state to get
3651 * a "preview" of its content.
3653 if (spa_writeable(spa)) {
3654 vdev_t *rvd = spa->spa_root_vdev;
3656 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3657 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3659 int children = rvd->vdev_children;
3660 int c0 = spa_get_random(children);
3662 for (int c = 0; c < children; c++) {
3663 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3665 /* Stop when revisiting the first vdev */
3666 if (c > 0 && svd[0] == vd)
3669 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3670 !vdev_is_concrete(vd))
3673 svd[svdcount++] = vd;
3674 if (svdcount == SPA_SYNC_MIN_VDEVS)
3677 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3679 spa->spa_last_synced_guid = rvd->vdev_guid;
3680 spa_config_exit(spa, SCL_ALL, FTAG);
3683 spa_load_failed(spa, "failed to write checkpointed "
3684 "uberblock to the vdev labels [error=%d]", error);
3693 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3694 boolean_t *update_config_cache)
3699 * Parse the config for pool, open and validate vdevs,
3700 * select an uberblock, and use that uberblock to open
3703 error = spa_ld_mos_init(spa, type);
3708 * Retrieve the trusted config stored in the MOS and use it to create
3709 * a new, exact version of the vdev tree, then reopen all vdevs.
3711 error = spa_ld_trusted_config(spa, type, B_FALSE);
3712 if (error == EAGAIN) {
3713 if (update_config_cache != NULL)
3714 *update_config_cache = B_TRUE;
3717 * Redo the loading process with the trusted config if it is
3718 * too different from the untrusted config.
3720 spa_ld_prepare_for_reload(spa);
3721 spa_load_note(spa, "RELOADING");
3722 error = spa_ld_mos_init(spa, type);
3726 error = spa_ld_trusted_config(spa, type, B_TRUE);
3730 } else if (error != 0) {
3738 * Load an existing storage pool, using the config provided. This config
3739 * describes which vdevs are part of the pool and is later validated against
3740 * partial configs present in each vdev's label and an entire copy of the
3741 * config stored in the MOS.
3744 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3747 boolean_t missing_feat_write = B_FALSE;
3748 boolean_t checkpoint_rewind =
3749 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3750 boolean_t update_config_cache = B_FALSE;
3752 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3753 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3755 spa_load_note(spa, "LOADING");
3757 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3762 * If we are rewinding to the checkpoint then we need to repeat
3763 * everything we've done so far in this function but this time
3764 * selecting the checkpointed uberblock and using that to open
3767 if (checkpoint_rewind) {
3769 * If we are rewinding to the checkpoint update config cache
3772 update_config_cache = B_TRUE;
3775 * Extract the checkpointed uberblock from the current MOS
3776 * and use this as the pool's uberblock from now on. If the
3777 * pool is imported as writeable we also write the checkpoint
3778 * uberblock to the labels, making the rewind permanent.
3780 error = spa_ld_checkpoint_rewind(spa);
3785 * Redo the loading process process again with the
3786 * checkpointed uberblock.
3788 spa_ld_prepare_for_reload(spa);
3789 spa_load_note(spa, "LOADING checkpointed uberblock");
3790 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3796 * Retrieve the checkpoint txg if the pool has a checkpoint.
3798 error = spa_ld_read_checkpoint_txg(spa);
3803 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3804 * from the pool and their contents were re-mapped to other vdevs. Note
3805 * that everything that we read before this step must have been
3806 * rewritten on concrete vdevs after the last device removal was
3807 * initiated. Otherwise we could be reading from indirect vdevs before
3808 * we have loaded their mappings.
3810 error = spa_ld_open_indirect_vdev_metadata(spa);
3815 * Retrieve the full list of active features from the MOS and check if
3816 * they are all supported.
3818 error = spa_ld_check_features(spa, &missing_feat_write);
3823 * Load several special directories from the MOS needed by the dsl_pool
3826 error = spa_ld_load_special_directories(spa);
3831 * Retrieve pool properties from the MOS.
3833 error = spa_ld_get_props(spa);
3838 * Retrieve the list of auxiliary devices - cache devices and spares -
3841 error = spa_ld_open_aux_vdevs(spa, type);
3846 * Load the metadata for all vdevs. Also check if unopenable devices
3847 * should be autoreplaced.
3849 error = spa_ld_load_vdev_metadata(spa);
3853 error = spa_ld_load_dedup_tables(spa);
3858 * Verify the logs now to make sure we don't have any unexpected errors
3859 * when we claim log blocks later.
3861 error = spa_ld_verify_logs(spa, type, ereport);
3865 if (missing_feat_write) {
3866 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3869 * At this point, we know that we can open the pool in
3870 * read-only mode but not read-write mode. We now have enough
3871 * information and can return to userland.
3873 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3878 * Traverse the last txgs to make sure the pool was left off in a safe
3879 * state. When performing an extreme rewind, we verify the whole pool,
3880 * which can take a very long time.
3882 error = spa_ld_verify_pool_data(spa);
3887 * Calculate the deflated space for the pool. This must be done before
3888 * we write anything to the pool because we'd need to update the space
3889 * accounting using the deflated sizes.
3891 spa_update_dspace(spa);
3894 * We have now retrieved all the information we needed to open the
3895 * pool. If we are importing the pool in read-write mode, a few
3896 * additional steps must be performed to finish the import.
3898 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3899 spa->spa_load_max_txg == UINT64_MAX)) {
3900 uint64_t config_cache_txg = spa->spa_config_txg;
3902 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3905 * In case of a checkpoint rewind, log the original txg
3906 * of the checkpointed uberblock.
3908 if (checkpoint_rewind) {
3909 spa_history_log_internal(spa, "checkpoint rewind",
3910 NULL, "rewound state to txg=%llu",
3911 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3915 * Traverse the ZIL and claim all blocks.
3917 spa_ld_claim_log_blocks(spa);
3920 * Kick-off the syncing thread.
3922 spa->spa_sync_on = B_TRUE;
3923 txg_sync_start(spa->spa_dsl_pool);
3926 * Wait for all claims to sync. We sync up to the highest
3927 * claimed log block birth time so that claimed log blocks
3928 * don't appear to be from the future. spa_claim_max_txg
3929 * will have been set for us by ZIL traversal operations
3932 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3935 * Check if we need to request an update of the config. On the
3936 * next sync, we would update the config stored in vdev labels
3937 * and the cachefile (by default /etc/zfs/zpool.cache).
3939 spa_ld_check_for_config_update(spa, config_cache_txg,
3940 update_config_cache);
3943 * Check all DTLs to see if anything needs resilvering.
3945 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3946 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3947 spa_async_request(spa, SPA_ASYNC_RESILVER);
3950 * Log the fact that we booted up (so that we can detect if
3951 * we rebooted in the middle of an operation).
3953 spa_history_log_version(spa, "open");
3956 * Delete any inconsistent datasets.
3958 (void) dmu_objset_find(spa_name(spa),
3959 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3962 * Clean up any stale temporary dataset userrefs.
3964 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3966 spa_restart_removal(spa);
3968 spa_spawn_aux_threads(spa);
3970 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3971 vdev_initialize_restart(spa->spa_root_vdev);
3972 spa_config_exit(spa, SCL_CONFIG, FTAG);
3975 spa_load_note(spa, "LOADED");
3981 spa_load_retry(spa_t *spa, spa_load_state_t state)
3983 int mode = spa->spa_mode;
3986 spa_deactivate(spa);
3988 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3990 spa_activate(spa, mode);
3991 spa_async_suspend(spa);
3993 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3994 (u_longlong_t)spa->spa_load_max_txg);
3996 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4000 * If spa_load() fails this function will try loading prior txg's. If
4001 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4002 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4003 * function will not rewind the pool and will return the same error as
4007 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
4010 nvlist_t *loadinfo = NULL;
4011 nvlist_t *config = NULL;
4012 int load_error, rewind_error;
4013 uint64_t safe_rewind_txg;
4016 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4017 spa->spa_load_max_txg = spa->spa_load_txg;
4018 spa_set_log_state(spa, SPA_LOG_CLEAR);
4020 spa->spa_load_max_txg = max_request;
4021 if (max_request != UINT64_MAX)
4022 spa->spa_extreme_rewind = B_TRUE;
4025 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4026 if (load_error == 0)
4028 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4030 * When attempting checkpoint-rewind on a pool with no
4031 * checkpoint, we should not attempt to load uberblocks
4032 * from previous txgs when spa_load fails.
4034 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4035 return (load_error);
4038 if (spa->spa_root_vdev != NULL)
4039 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4041 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4042 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4044 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4045 nvlist_free(config);
4046 return (load_error);
4049 if (state == SPA_LOAD_RECOVER) {
4050 /* Price of rolling back is discarding txgs, including log */
4051 spa_set_log_state(spa, SPA_LOG_CLEAR);
4054 * If we aren't rolling back save the load info from our first
4055 * import attempt so that we can restore it after attempting
4058 loadinfo = spa->spa_load_info;
4059 spa->spa_load_info = fnvlist_alloc();
4062 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4063 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4064 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4065 TXG_INITIAL : safe_rewind_txg;
4068 * Continue as long as we're finding errors, we're still within
4069 * the acceptable rewind range, and we're still finding uberblocks
4071 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4072 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4073 if (spa->spa_load_max_txg < safe_rewind_txg)
4074 spa->spa_extreme_rewind = B_TRUE;
4075 rewind_error = spa_load_retry(spa, state);
4078 spa->spa_extreme_rewind = B_FALSE;
4079 spa->spa_load_max_txg = UINT64_MAX;
4081 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4082 spa_config_set(spa, config);
4084 nvlist_free(config);
4086 if (state == SPA_LOAD_RECOVER) {
4087 ASSERT3P(loadinfo, ==, NULL);
4088 return (rewind_error);
4090 /* Store the rewind info as part of the initial load info */
4091 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4092 spa->spa_load_info);
4094 /* Restore the initial load info */
4095 fnvlist_free(spa->spa_load_info);
4096 spa->spa_load_info = loadinfo;
4098 return (load_error);
4105 * The import case is identical to an open except that the configuration is sent
4106 * down from userland, instead of grabbed from the configuration cache. For the
4107 * case of an open, the pool configuration will exist in the
4108 * POOL_STATE_UNINITIALIZED state.
4110 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4111 * the same time open the pool, without having to keep around the spa_t in some
4115 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4119 spa_load_state_t state = SPA_LOAD_OPEN;
4121 int locked = B_FALSE;
4122 int firstopen = B_FALSE;
4127 * As disgusting as this is, we need to support recursive calls to this
4128 * function because dsl_dir_open() is called during spa_load(), and ends
4129 * up calling spa_open() again. The real fix is to figure out how to
4130 * avoid dsl_dir_open() calling this in the first place.
4132 if (mutex_owner(&spa_namespace_lock) != curthread) {
4133 mutex_enter(&spa_namespace_lock);
4137 if ((spa = spa_lookup(pool)) == NULL) {
4139 mutex_exit(&spa_namespace_lock);
4140 return (SET_ERROR(ENOENT));
4143 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
4144 zpool_load_policy_t policy;
4148 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
4150 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
4151 state = SPA_LOAD_RECOVER;
4153 spa_activate(spa, spa_mode_global);
4155 if (state != SPA_LOAD_RECOVER)
4156 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4157 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4159 zfs_dbgmsg("spa_open_common: opening %s", pool);
4160 error = spa_load_best(spa, state, policy.zlp_txg,
4163 if (error == EBADF) {
4165 * If vdev_validate() returns failure (indicated by
4166 * EBADF), it indicates that one of the vdevs indicates
4167 * that the pool has been exported or destroyed. If
4168 * this is the case, the config cache is out of sync and
4169 * we should remove the pool from the namespace.
4172 spa_deactivate(spa);
4173 spa_write_cachefile(spa, B_TRUE, B_TRUE);
4176 mutex_exit(&spa_namespace_lock);
4177 return (SET_ERROR(ENOENT));
4182 * We can't open the pool, but we still have useful
4183 * information: the state of each vdev after the
4184 * attempted vdev_open(). Return this to the user.
4186 if (config != NULL && spa->spa_config) {
4187 VERIFY(nvlist_dup(spa->spa_config, config,
4189 VERIFY(nvlist_add_nvlist(*config,
4190 ZPOOL_CONFIG_LOAD_INFO,
4191 spa->spa_load_info) == 0);
4194 spa_deactivate(spa);
4195 spa->spa_last_open_failed = error;
4197 mutex_exit(&spa_namespace_lock);
4203 spa_open_ref(spa, tag);
4206 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4209 * If we've recovered the pool, pass back any information we
4210 * gathered while doing the load.
4212 if (state == SPA_LOAD_RECOVER) {
4213 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4214 spa->spa_load_info) == 0);
4218 spa->spa_last_open_failed = 0;
4219 spa->spa_last_ubsync_txg = 0;
4220 spa->spa_load_txg = 0;
4221 mutex_exit(&spa_namespace_lock);
4225 zvol_create_minors(spa->spa_name);
4236 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4239 return (spa_open_common(name, spapp, tag, policy, config));
4243 spa_open(const char *name, spa_t **spapp, void *tag)
4245 return (spa_open_common(name, spapp, tag, NULL, NULL));
4249 * Lookup the given spa_t, incrementing the inject count in the process,
4250 * preventing it from being exported or destroyed.
4253 spa_inject_addref(char *name)
4257 mutex_enter(&spa_namespace_lock);
4258 if ((spa = spa_lookup(name)) == NULL) {
4259 mutex_exit(&spa_namespace_lock);
4262 spa->spa_inject_ref++;
4263 mutex_exit(&spa_namespace_lock);
4269 spa_inject_delref(spa_t *spa)
4271 mutex_enter(&spa_namespace_lock);
4272 spa->spa_inject_ref--;
4273 mutex_exit(&spa_namespace_lock);
4277 * Add spares device information to the nvlist.
4280 spa_add_spares(spa_t *spa, nvlist_t *config)
4290 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4292 if (spa->spa_spares.sav_count == 0)
4295 VERIFY(nvlist_lookup_nvlist(config,
4296 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4297 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4298 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4300 VERIFY(nvlist_add_nvlist_array(nvroot,
4301 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4302 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4303 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4306 * Go through and find any spares which have since been
4307 * repurposed as an active spare. If this is the case, update
4308 * their status appropriately.
4310 for (i = 0; i < nspares; i++) {
4311 VERIFY(nvlist_lookup_uint64(spares[i],
4312 ZPOOL_CONFIG_GUID, &guid) == 0);
4313 if (spa_spare_exists(guid, &pool, NULL) &&
4315 VERIFY(nvlist_lookup_uint64_array(
4316 spares[i], ZPOOL_CONFIG_VDEV_STATS,
4317 (uint64_t **)&vs, &vsc) == 0);
4318 vs->vs_state = VDEV_STATE_CANT_OPEN;
4319 vs->vs_aux = VDEV_AUX_SPARED;
4326 * Add l2cache device information to the nvlist, including vdev stats.
4329 spa_add_l2cache(spa_t *spa, nvlist_t *config)
4332 uint_t i, j, nl2cache;
4339 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4341 if (spa->spa_l2cache.sav_count == 0)
4344 VERIFY(nvlist_lookup_nvlist(config,
4345 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4346 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4347 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4348 if (nl2cache != 0) {
4349 VERIFY(nvlist_add_nvlist_array(nvroot,
4350 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4351 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4352 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4355 * Update level 2 cache device stats.
4358 for (i = 0; i < nl2cache; i++) {
4359 VERIFY(nvlist_lookup_uint64(l2cache[i],
4360 ZPOOL_CONFIG_GUID, &guid) == 0);
4363 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4365 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4366 vd = spa->spa_l2cache.sav_vdevs[j];
4372 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4373 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4375 vdev_get_stats(vd, vs);
4381 spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
4386 /* We may be unable to read features if pool is suspended. */
4387 if (spa_suspended(spa))
4390 if (spa->spa_feat_for_read_obj != 0) {
4391 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4392 spa->spa_feat_for_read_obj);
4393 zap_cursor_retrieve(&zc, &za) == 0;
4394 zap_cursor_advance(&zc)) {
4395 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4396 za.za_num_integers == 1);
4397 VERIFY0(nvlist_add_uint64(features, za.za_name,
4398 za.za_first_integer));
4400 zap_cursor_fini(&zc);
4403 if (spa->spa_feat_for_write_obj != 0) {
4404 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4405 spa->spa_feat_for_write_obj);
4406 zap_cursor_retrieve(&zc, &za) == 0;
4407 zap_cursor_advance(&zc)) {
4408 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4409 za.za_num_integers == 1);
4410 VERIFY0(nvlist_add_uint64(features, za.za_name,
4411 za.za_first_integer));
4413 zap_cursor_fini(&zc);
4418 spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
4422 for (i = 0; i < SPA_FEATURES; i++) {
4423 zfeature_info_t feature = spa_feature_table[i];
4426 if (feature_get_refcount(spa, &feature, &refcount) != 0)
4429 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
4434 * Store a list of pool features and their reference counts in the
4437 * The first time this is called on a spa, allocate a new nvlist, fetch
4438 * the pool features and reference counts from disk, then save the list
4439 * in the spa. In subsequent calls on the same spa use the saved nvlist
4440 * and refresh its values from the cached reference counts. This
4441 * ensures we don't block here on I/O on a suspended pool so 'zpool
4442 * clear' can resume the pool.
4445 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4449 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4451 mutex_enter(&spa->spa_feat_stats_lock);
4452 features = spa->spa_feat_stats;
4454 if (features != NULL) {
4455 spa_feature_stats_from_cache(spa, features);
4457 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
4458 spa->spa_feat_stats = features;
4459 spa_feature_stats_from_disk(spa, features);
4462 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4465 mutex_exit(&spa->spa_feat_stats_lock);
4469 spa_get_stats(const char *name, nvlist_t **config,
4470 char *altroot, size_t buflen)
4476 error = spa_open_common(name, &spa, FTAG, NULL, config);
4480 * This still leaves a window of inconsistency where the spares
4481 * or l2cache devices could change and the config would be
4482 * self-inconsistent.
4484 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4486 if (*config != NULL) {
4487 uint64_t loadtimes[2];
4489 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4490 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4491 VERIFY(nvlist_add_uint64_array(*config,
4492 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4494 VERIFY(nvlist_add_uint64(*config,
4495 ZPOOL_CONFIG_ERRCOUNT,
4496 spa_get_errlog_size(spa)) == 0);
4498 if (spa_suspended(spa))
4499 VERIFY(nvlist_add_uint64(*config,
4500 ZPOOL_CONFIG_SUSPENDED,
4501 spa->spa_failmode) == 0);
4503 spa_add_spares(spa, *config);
4504 spa_add_l2cache(spa, *config);
4505 spa_add_feature_stats(spa, *config);
4510 * We want to get the alternate root even for faulted pools, so we cheat
4511 * and call spa_lookup() directly.
4515 mutex_enter(&spa_namespace_lock);
4516 spa = spa_lookup(name);
4518 spa_altroot(spa, altroot, buflen);
4522 mutex_exit(&spa_namespace_lock);
4524 spa_altroot(spa, altroot, buflen);
4529 spa_config_exit(spa, SCL_CONFIG, FTAG);
4530 spa_close(spa, FTAG);
4537 * Validate that the auxiliary device array is well formed. We must have an
4538 * array of nvlists, each which describes a valid leaf vdev. If this is an
4539 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4540 * specified, as long as they are well-formed.
4543 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4544 spa_aux_vdev_t *sav, const char *config, uint64_t version,
4545 vdev_labeltype_t label)
4552 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4555 * It's acceptable to have no devs specified.
4557 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4561 return (SET_ERROR(EINVAL));
4564 * Make sure the pool is formatted with a version that supports this
4567 if (spa_version(spa) < version)
4568 return (SET_ERROR(ENOTSUP));
4571 * Set the pending device list so we correctly handle device in-use
4574 sav->sav_pending = dev;
4575 sav->sav_npending = ndev;
4577 for (i = 0; i < ndev; i++) {
4578 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4582 if (!vd->vdev_ops->vdev_op_leaf) {
4584 error = SET_ERROR(EINVAL);
4589 * The L2ARC currently only supports disk devices in
4590 * kernel context. For user-level testing, we allow it.
4593 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4594 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4595 error = SET_ERROR(ENOTBLK);
4602 if ((error = vdev_open(vd)) == 0 &&
4603 (error = vdev_label_init(vd, crtxg, label)) == 0) {
4604 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4605 vd->vdev_guid) == 0);
4611 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4618 sav->sav_pending = NULL;
4619 sav->sav_npending = 0;
4624 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4628 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4630 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4631 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4632 VDEV_LABEL_SPARE)) != 0) {
4636 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4637 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4638 VDEV_LABEL_L2CACHE));
4642 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4647 if (sav->sav_config != NULL) {
4653 * Generate new dev list by concatentating with the
4656 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4657 &olddevs, &oldndevs) == 0);
4659 newdevs = kmem_alloc(sizeof (void *) *
4660 (ndevs + oldndevs), KM_SLEEP);
4661 for (i = 0; i < oldndevs; i++)
4662 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4664 for (i = 0; i < ndevs; i++)
4665 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4668 VERIFY(nvlist_remove(sav->sav_config, config,
4669 DATA_TYPE_NVLIST_ARRAY) == 0);
4671 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4672 config, newdevs, ndevs + oldndevs) == 0);
4673 for (i = 0; i < oldndevs + ndevs; i++)
4674 nvlist_free(newdevs[i]);
4675 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4678 * Generate a new dev list.
4680 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4682 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4688 * Stop and drop level 2 ARC devices
4691 spa_l2cache_drop(spa_t *spa)
4695 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4697 for (i = 0; i < sav->sav_count; i++) {
4700 vd = sav->sav_vdevs[i];
4703 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4704 pool != 0ULL && l2arc_vdev_present(vd))
4705 l2arc_remove_vdev(vd);
4713 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4717 char *altroot = NULL;
4722 uint64_t txg = TXG_INITIAL;
4723 nvlist_t **spares, **l2cache;
4724 uint_t nspares, nl2cache;
4725 uint64_t version, obj;
4726 boolean_t has_features;
4730 if (nvlist_lookup_string(props,
4731 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0)
4732 poolname = (char *)pool;
4735 * If this pool already exists, return failure.
4737 mutex_enter(&spa_namespace_lock);
4738 if (spa_lookup(poolname) != NULL) {
4739 mutex_exit(&spa_namespace_lock);
4740 return (SET_ERROR(EEXIST));
4744 * Allocate a new spa_t structure.
4746 nvl = fnvlist_alloc();
4747 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
4748 (void) nvlist_lookup_string(props,
4749 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4750 spa = spa_add(poolname, nvl, altroot);
4752 spa_activate(spa, spa_mode_global);
4754 if (props && (error = spa_prop_validate(spa, props))) {
4755 spa_deactivate(spa);
4757 mutex_exit(&spa_namespace_lock);
4762 * Temporary pool names should never be written to disk.
4764 if (poolname != pool)
4765 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
4767 has_features = B_FALSE;
4768 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4769 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4770 if (zpool_prop_feature(nvpair_name(elem)))
4771 has_features = B_TRUE;
4774 if (has_features || nvlist_lookup_uint64(props,
4775 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4776 version = SPA_VERSION;
4778 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4780 spa->spa_first_txg = txg;
4781 spa->spa_uberblock.ub_txg = txg - 1;
4782 spa->spa_uberblock.ub_version = version;
4783 spa->spa_ubsync = spa->spa_uberblock;
4784 spa->spa_load_state = SPA_LOAD_CREATE;
4785 spa->spa_removing_phys.sr_state = DSS_NONE;
4786 spa->spa_removing_phys.sr_removing_vdev = -1;
4787 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4788 spa->spa_indirect_vdevs_loaded = B_TRUE;
4791 * Create "The Godfather" zio to hold all async IOs
4793 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4795 for (int i = 0; i < max_ncpus; i++) {
4796 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4797 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4798 ZIO_FLAG_GODFATHER);
4802 * Create the root vdev.
4804 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4806 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4808 ASSERT(error != 0 || rvd != NULL);
4809 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4811 if (error == 0 && !zfs_allocatable_devs(nvroot))
4812 error = SET_ERROR(EINVAL);
4815 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4816 (error = spa_validate_aux(spa, nvroot, txg,
4817 VDEV_ALLOC_ADD)) == 0) {
4818 for (int c = 0; c < rvd->vdev_children; c++) {
4819 vdev_ashift_optimize(rvd->vdev_child[c]);
4820 vdev_metaslab_set_size(rvd->vdev_child[c]);
4821 vdev_expand(rvd->vdev_child[c], txg);
4825 spa_config_exit(spa, SCL_ALL, FTAG);
4829 spa_deactivate(spa);
4831 mutex_exit(&spa_namespace_lock);
4836 * Get the list of spares, if specified.
4838 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4839 &spares, &nspares) == 0) {
4840 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4842 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4843 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4844 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4845 spa_load_spares(spa);
4846 spa_config_exit(spa, SCL_ALL, FTAG);
4847 spa->spa_spares.sav_sync = B_TRUE;
4851 * Get the list of level 2 cache devices, if specified.
4853 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4854 &l2cache, &nl2cache) == 0) {
4855 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4856 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4857 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4858 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4859 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4860 spa_load_l2cache(spa);
4861 spa_config_exit(spa, SCL_ALL, FTAG);
4862 spa->spa_l2cache.sav_sync = B_TRUE;
4865 spa->spa_is_initializing = B_TRUE;
4866 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4867 spa->spa_meta_objset = dp->dp_meta_objset;
4868 spa->spa_is_initializing = B_FALSE;
4871 * Create DDTs (dedup tables).
4875 spa_update_dspace(spa);
4877 tx = dmu_tx_create_assigned(dp, txg);
4880 * Create the pool config object.
4882 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4883 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4884 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4886 if (zap_add(spa->spa_meta_objset,
4887 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4888 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4889 cmn_err(CE_PANIC, "failed to add pool config");
4892 if (spa_version(spa) >= SPA_VERSION_FEATURES)
4893 spa_feature_create_zap_objects(spa, tx);
4895 if (zap_add(spa->spa_meta_objset,
4896 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4897 sizeof (uint64_t), 1, &version, tx) != 0) {
4898 cmn_err(CE_PANIC, "failed to add pool version");
4901 /* Newly created pools with the right version are always deflated. */
4902 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4903 spa->spa_deflate = TRUE;
4904 if (zap_add(spa->spa_meta_objset,
4905 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4906 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4907 cmn_err(CE_PANIC, "failed to add deflate");
4912 * Create the deferred-free bpobj. Turn off compression
4913 * because sync-to-convergence takes longer if the blocksize
4916 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4917 dmu_object_set_compress(spa->spa_meta_objset, obj,
4918 ZIO_COMPRESS_OFF, tx);
4919 if (zap_add(spa->spa_meta_objset,
4920 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4921 sizeof (uint64_t), 1, &obj, tx) != 0) {
4922 cmn_err(CE_PANIC, "failed to add bpobj");
4924 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4925 spa->spa_meta_objset, obj));
4928 * Create the pool's history object.
4930 if (version >= SPA_VERSION_ZPOOL_HISTORY)
4931 spa_history_create_obj(spa, tx);
4934 * Generate some random noise for salted checksums to operate on.
4936 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4937 sizeof (spa->spa_cksum_salt.zcs_bytes));
4940 * Set pool properties.
4942 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4943 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4944 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4945 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4947 if (props != NULL) {
4948 spa_configfile_set(spa, props, B_FALSE);
4949 spa_sync_props(props, tx);
4954 spa->spa_sync_on = B_TRUE;
4955 txg_sync_start(spa->spa_dsl_pool);
4958 * We explicitly wait for the first transaction to complete so that our
4959 * bean counters are appropriately updated.
4961 txg_wait_synced(spa->spa_dsl_pool, txg);
4963 spa_spawn_aux_threads(spa);
4965 spa_write_cachefile(spa, B_FALSE, B_TRUE);
4966 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4968 spa_history_log_version(spa, "create");
4971 * Don't count references from objsets that are already closed
4972 * and are making their way through the eviction process.
4974 spa_evicting_os_wait(spa);
4975 spa->spa_minref = refcount_count(&spa->spa_refcount);
4976 spa->spa_load_state = SPA_LOAD_NONE;
4978 mutex_exit(&spa_namespace_lock);
4986 * Get the root pool information from the root disk, then import the root pool
4987 * during the system boot up time.
4989 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4992 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4995 nvlist_t *nvtop, *nvroot;
4998 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
5002 * Add this top-level vdev to the child array.
5004 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5006 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5008 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
5011 * Put this pool's top-level vdevs into a root vdev.
5013 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5014 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5015 VDEV_TYPE_ROOT) == 0);
5016 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5017 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5018 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5022 * Replace the existing vdev_tree with the new root vdev in
5023 * this pool's configuration (remove the old, add the new).
5025 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5026 nvlist_free(nvroot);
5031 * Walk the vdev tree and see if we can find a device with "better"
5032 * configuration. A configuration is "better" if the label on that
5033 * device has a more recent txg.
5036 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
5038 for (int c = 0; c < vd->vdev_children; c++)
5039 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
5041 if (vd->vdev_ops->vdev_op_leaf) {
5045 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
5049 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
5053 * Do we have a better boot device?
5055 if (label_txg > *txg) {
5064 * Import a root pool.
5066 * For x86. devpath_list will consist of devid and/or physpath name of
5067 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
5068 * The GRUB "findroot" command will return the vdev we should boot.
5070 * For Sparc, devpath_list consists the physpath name of the booting device
5071 * no matter the rootpool is a single device pool or a mirrored pool.
5073 * "/pci@1f,0/ide@d/disk@0,0:a"
5076 spa_import_rootpool(char *devpath, char *devid)
5079 vdev_t *rvd, *bvd, *avd = NULL;
5080 nvlist_t *config, *nvtop;
5086 * Read the label from the boot device and generate a configuration.
5088 config = spa_generate_rootconf(devpath, devid, &guid);
5089 #if defined(_OBP) && defined(_KERNEL)
5090 if (config == NULL) {
5091 if (strstr(devpath, "/iscsi/ssd") != NULL) {
5093 get_iscsi_bootpath_phy(devpath);
5094 config = spa_generate_rootconf(devpath, devid, &guid);
5098 if (config == NULL) {
5099 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
5101 return (SET_ERROR(EIO));
5104 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5106 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
5108 mutex_enter(&spa_namespace_lock);
5109 if ((spa = spa_lookup(pname)) != NULL) {
5111 * Remove the existing root pool from the namespace so that we
5112 * can replace it with the correct config we just read in.
5117 spa = spa_add(pname, config, NULL);
5118 spa->spa_is_root = B_TRUE;
5119 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5120 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5121 &spa->spa_ubsync.ub_version) != 0)
5122 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5125 * Build up a vdev tree based on the boot device's label config.
5127 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5129 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5130 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5131 VDEV_ALLOC_ROOTPOOL);
5132 spa_config_exit(spa, SCL_ALL, FTAG);
5134 mutex_exit(&spa_namespace_lock);
5135 nvlist_free(config);
5136 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5142 * Get the boot vdev.
5144 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5145 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5146 (u_longlong_t)guid);
5147 error = SET_ERROR(ENOENT);
5152 * Determine if there is a better boot device.
5155 spa_alt_rootvdev(rvd, &avd, &txg);
5157 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5158 "try booting from '%s'", avd->vdev_path);
5159 error = SET_ERROR(EINVAL);
5164 * If the boot device is part of a spare vdev then ensure that
5165 * we're booting off the active spare.
5167 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5168 !bvd->vdev_isspare) {
5169 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5170 "try booting from '%s'",
5172 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5173 error = SET_ERROR(EINVAL);
5179 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5181 spa_config_exit(spa, SCL_ALL, FTAG);
5182 mutex_exit(&spa_namespace_lock);
5184 nvlist_free(config);
5188 #else /* !illumos */
5190 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
5194 spa_generate_rootconf(const char *name)
5196 nvlist_t **configs, **tops;
5198 nvlist_t *best_cfg, *nvtop, *nvroot;
5207 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
5210 ASSERT3U(count, !=, 0);
5212 for (i = 0; i < count; i++) {
5215 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
5217 if (txg > best_txg) {
5219 best_cfg = configs[i];
5224 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
5226 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
5229 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
5230 for (i = 0; i < nchildren; i++) {
5233 if (configs[i] == NULL)
5235 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
5237 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
5239 for (i = 0; holes != NULL && i < nholes; i++) {
5242 if (tops[holes[i]] != NULL)
5244 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
5245 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
5246 VDEV_TYPE_HOLE) == 0);
5247 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
5249 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
5252 for (i = 0; i < nchildren; i++) {
5253 if (tops[i] != NULL)
5255 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
5256 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
5257 VDEV_TYPE_MISSING) == 0);
5258 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
5260 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
5265 * Create pool config based on the best vdev config.
5267 nvlist_dup(best_cfg, &config, KM_SLEEP);
5270 * Put this pool's top-level vdevs into a root vdev.
5272 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5274 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5275 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5276 VDEV_TYPE_ROOT) == 0);
5277 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5278 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5279 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5280 tops, nchildren) == 0);
5283 * Replace the existing vdev_tree with the new root vdev in
5284 * this pool's configuration (remove the old, add the new).
5286 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5289 * Drop vdev config elements that should not be present at pool level.
5291 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
5292 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
5294 for (i = 0; i < count; i++)
5295 nvlist_free(configs[i]);
5296 kmem_free(configs, count * sizeof(void *));
5297 for (i = 0; i < nchildren; i++)
5298 nvlist_free(tops[i]);
5299 kmem_free(tops, nchildren * sizeof(void *));
5300 nvlist_free(nvroot);
5305 spa_import_rootpool(const char *name)
5308 vdev_t *rvd, *bvd, *avd = NULL;
5309 nvlist_t *config, *nvtop;
5315 * Read the label from the boot device and generate a configuration.
5317 config = spa_generate_rootconf(name);
5319 mutex_enter(&spa_namespace_lock);
5320 if (config != NULL) {
5321 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5322 &pname) == 0 && strcmp(name, pname) == 0);
5323 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
5326 if ((spa = spa_lookup(pname)) != NULL) {
5328 * The pool could already be imported,
5329 * e.g., after reboot -r.
5331 if (spa->spa_state == POOL_STATE_ACTIVE) {
5332 mutex_exit(&spa_namespace_lock);
5333 nvlist_free(config);
5338 * Remove the existing root pool from the namespace so
5339 * that we can replace it with the correct config
5344 spa = spa_add(pname, config, NULL);
5347 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
5348 * via spa_version().
5350 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5351 &spa->spa_ubsync.ub_version) != 0)
5352 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5353 } else if ((spa = spa_lookup(name)) == NULL) {
5354 mutex_exit(&spa_namespace_lock);
5355 nvlist_free(config);
5356 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
5360 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
5362 spa->spa_is_root = B_TRUE;
5363 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5366 * Build up a vdev tree based on the boot device's label config.
5368 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5370 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5371 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5372 VDEV_ALLOC_ROOTPOOL);
5373 spa_config_exit(spa, SCL_ALL, FTAG);
5375 mutex_exit(&spa_namespace_lock);
5376 nvlist_free(config);
5377 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5382 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5384 spa_config_exit(spa, SCL_ALL, FTAG);
5385 mutex_exit(&spa_namespace_lock);
5387 nvlist_free(config);
5391 #endif /* illumos */
5392 #endif /* _KERNEL */
5395 * Import a non-root pool into the system.
5398 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5401 char *altroot = NULL;
5402 spa_load_state_t state = SPA_LOAD_IMPORT;
5403 zpool_load_policy_t policy;
5404 uint64_t mode = spa_mode_global;
5405 uint64_t readonly = B_FALSE;
5408 nvlist_t **spares, **l2cache;
5409 uint_t nspares, nl2cache;
5412 * If a pool with this name exists, return failure.
5414 mutex_enter(&spa_namespace_lock);
5415 if (spa_lookup(pool) != NULL) {
5416 mutex_exit(&spa_namespace_lock);
5417 return (SET_ERROR(EEXIST));
5421 * Create and initialize the spa structure.
5423 (void) nvlist_lookup_string(props,
5424 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5425 (void) nvlist_lookup_uint64(props,
5426 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5429 spa = spa_add(pool, config, altroot);
5430 spa->spa_import_flags = flags;
5433 * Verbatim import - Take a pool and insert it into the namespace
5434 * as if it had been loaded at boot.
5436 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5438 spa_configfile_set(spa, props, B_FALSE);
5440 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5441 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5442 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5443 mutex_exit(&spa_namespace_lock);
5447 spa_activate(spa, mode);
5450 * Don't start async tasks until we know everything is healthy.
5452 spa_async_suspend(spa);
5454 zpool_get_load_policy(config, &policy);
5455 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5456 state = SPA_LOAD_RECOVER;
5458 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5460 if (state != SPA_LOAD_RECOVER) {
5461 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5462 zfs_dbgmsg("spa_import: importing %s", pool);
5464 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5465 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5467 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5470 * Propagate anything learned while loading the pool and pass it
5471 * back to caller (i.e. rewind info, missing devices, etc).
5473 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5474 spa->spa_load_info) == 0);
5476 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5478 * Toss any existing sparelist, as it doesn't have any validity
5479 * anymore, and conflicts with spa_has_spare().
5481 if (spa->spa_spares.sav_config) {
5482 nvlist_free(spa->spa_spares.sav_config);
5483 spa->spa_spares.sav_config = NULL;
5484 spa_load_spares(spa);
5486 if (spa->spa_l2cache.sav_config) {
5487 nvlist_free(spa->spa_l2cache.sav_config);
5488 spa->spa_l2cache.sav_config = NULL;
5489 spa_load_l2cache(spa);
5492 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5495 error = spa_validate_aux(spa, nvroot, -1ULL,
5498 error = spa_validate_aux(spa, nvroot, -1ULL,
5499 VDEV_ALLOC_L2CACHE);
5500 spa_config_exit(spa, SCL_ALL, FTAG);
5503 spa_configfile_set(spa, props, B_FALSE);
5505 if (error != 0 || (props && spa_writeable(spa) &&
5506 (error = spa_prop_set(spa, props)))) {
5508 spa_deactivate(spa);
5510 mutex_exit(&spa_namespace_lock);
5514 spa_async_resume(spa);
5517 * Override any spares and level 2 cache devices as specified by
5518 * the user, as these may have correct device names/devids, etc.
5520 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5521 &spares, &nspares) == 0) {
5522 if (spa->spa_spares.sav_config)
5523 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5524 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5526 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5527 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5528 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5529 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5530 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5531 spa_load_spares(spa);
5532 spa_config_exit(spa, SCL_ALL, FTAG);
5533 spa->spa_spares.sav_sync = B_TRUE;
5535 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5536 &l2cache, &nl2cache) == 0) {
5537 if (spa->spa_l2cache.sav_config)
5538 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5539 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5541 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5542 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5543 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5544 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5545 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5546 spa_load_l2cache(spa);
5547 spa_config_exit(spa, SCL_ALL, FTAG);
5548 spa->spa_l2cache.sav_sync = B_TRUE;
5552 * Check for any removed devices.
5554 if (spa->spa_autoreplace) {
5555 spa_aux_check_removed(&spa->spa_spares);
5556 spa_aux_check_removed(&spa->spa_l2cache);
5559 if (spa_writeable(spa)) {
5561 * Update the config cache to include the newly-imported pool.
5563 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5567 * It's possible that the pool was expanded while it was exported.
5568 * We kick off an async task to handle this for us.
5570 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5572 spa_history_log_version(spa, "import");
5574 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5576 mutex_exit(&spa_namespace_lock);
5580 zvol_create_minors(pool);
5587 spa_tryimport(nvlist_t *tryconfig)
5589 nvlist_t *config = NULL;
5590 char *poolname, *cachefile;
5594 zpool_load_policy_t policy;
5596 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5599 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5603 * Create and initialize the spa structure.
5605 mutex_enter(&spa_namespace_lock);
5606 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5607 spa_activate(spa, FREAD);
5610 * Rewind pool if a max txg was provided.
5612 zpool_get_load_policy(spa->spa_config, &policy);
5613 if (policy.zlp_txg != UINT64_MAX) {
5614 spa->spa_load_max_txg = policy.zlp_txg;
5615 spa->spa_extreme_rewind = B_TRUE;
5616 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5617 poolname, (longlong_t)policy.zlp_txg);
5619 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5622 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5624 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5625 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5627 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5630 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5633 * If 'tryconfig' was at least parsable, return the current config.
5635 if (spa->spa_root_vdev != NULL) {
5636 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5637 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5639 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5641 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5642 spa->spa_uberblock.ub_timestamp) == 0);
5643 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5644 spa->spa_load_info) == 0);
5647 * If the bootfs property exists on this pool then we
5648 * copy it out so that external consumers can tell which
5649 * pools are bootable.
5651 if ((!error || error == EEXIST) && spa->spa_bootfs) {
5652 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5655 * We have to play games with the name since the
5656 * pool was opened as TRYIMPORT_NAME.
5658 if (dsl_dsobj_to_dsname(spa_name(spa),
5659 spa->spa_bootfs, tmpname) == 0) {
5661 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5663 cp = strchr(tmpname, '/');
5665 (void) strlcpy(dsname, tmpname,
5668 (void) snprintf(dsname, MAXPATHLEN,
5669 "%s/%s", poolname, ++cp);
5671 VERIFY(nvlist_add_string(config,
5672 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5673 kmem_free(dsname, MAXPATHLEN);
5675 kmem_free(tmpname, MAXPATHLEN);
5679 * Add the list of hot spares and level 2 cache devices.
5681 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5682 spa_add_spares(spa, config);
5683 spa_add_l2cache(spa, config);
5684 spa_config_exit(spa, SCL_CONFIG, FTAG);
5688 spa_deactivate(spa);
5690 mutex_exit(&spa_namespace_lock);
5696 * Pool export/destroy
5698 * The act of destroying or exporting a pool is very simple. We make sure there
5699 * is no more pending I/O and any references to the pool are gone. Then, we
5700 * update the pool state and sync all the labels to disk, removing the
5701 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5702 * we don't sync the labels or remove the configuration cache.
5705 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5706 boolean_t force, boolean_t hardforce)
5713 if (!(spa_mode_global & FWRITE))
5714 return (SET_ERROR(EROFS));
5716 mutex_enter(&spa_namespace_lock);
5717 if ((spa = spa_lookup(pool)) == NULL) {
5718 mutex_exit(&spa_namespace_lock);
5719 return (SET_ERROR(ENOENT));
5723 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5724 * reacquire the namespace lock, and see if we can export.
5726 spa_open_ref(spa, FTAG);
5727 mutex_exit(&spa_namespace_lock);
5728 spa_async_suspend(spa);
5729 mutex_enter(&spa_namespace_lock);
5730 spa_close(spa, FTAG);
5733 * The pool will be in core if it's openable,
5734 * in which case we can modify its state.
5736 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5739 * Objsets may be open only because they're dirty, so we
5740 * have to force it to sync before checking spa_refcnt.
5742 txg_wait_synced(spa->spa_dsl_pool, 0);
5743 spa_evicting_os_wait(spa);
5746 * A pool cannot be exported or destroyed if there are active
5747 * references. If we are resetting a pool, allow references by
5748 * fault injection handlers.
5750 if (!spa_refcount_zero(spa) ||
5751 (spa->spa_inject_ref != 0 &&
5752 new_state != POOL_STATE_UNINITIALIZED)) {
5753 spa_async_resume(spa);
5754 mutex_exit(&spa_namespace_lock);
5755 return (SET_ERROR(EBUSY));
5759 * A pool cannot be exported if it has an active shared spare.
5760 * This is to prevent other pools stealing the active spare
5761 * from an exported pool. At user's own will, such pool can
5762 * be forcedly exported.
5764 if (!force && new_state == POOL_STATE_EXPORTED &&
5765 spa_has_active_shared_spare(spa)) {
5766 spa_async_resume(spa);
5767 mutex_exit(&spa_namespace_lock);
5768 return (SET_ERROR(EXDEV));
5772 * We're about to export or destroy this pool. Make sure
5773 * we stop all initializtion activity here before we
5774 * set the spa_final_txg. This will ensure that all
5775 * dirty data resulting from the initialization is
5776 * committed to disk before we unload the pool.
5778 if (spa->spa_root_vdev != NULL) {
5779 vdev_initialize_stop_all(spa->spa_root_vdev,
5780 VDEV_INITIALIZE_ACTIVE);
5784 * We want this to be reflected on every label,
5785 * so mark them all dirty. spa_unload() will do the
5786 * final sync that pushes these changes out.
5788 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5789 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5790 spa->spa_state = new_state;
5791 spa->spa_final_txg = spa_last_synced_txg(spa) +
5793 vdev_config_dirty(spa->spa_root_vdev);
5794 spa_config_exit(spa, SCL_ALL, FTAG);
5798 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5800 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5802 spa_deactivate(spa);
5805 if (oldconfig && spa->spa_config)
5806 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5808 if (new_state != POOL_STATE_UNINITIALIZED) {
5810 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5813 mutex_exit(&spa_namespace_lock);
5819 * Destroy a storage pool.
5822 spa_destroy(char *pool)
5824 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5829 * Export a storage pool.
5832 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5833 boolean_t hardforce)
5835 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5840 * Similar to spa_export(), this unloads the spa_t without actually removing it
5841 * from the namespace in any way.
5844 spa_reset(char *pool)
5846 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5851 * ==========================================================================
5852 * Device manipulation
5853 * ==========================================================================
5857 * Add a device to a storage pool.
5860 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5864 vdev_t *rvd = spa->spa_root_vdev;
5866 nvlist_t **spares, **l2cache;
5867 uint_t nspares, nl2cache;
5869 ASSERT(spa_writeable(spa));
5871 txg = spa_vdev_enter(spa);
5873 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5874 VDEV_ALLOC_ADD)) != 0)
5875 return (spa_vdev_exit(spa, NULL, txg, error));
5877 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
5879 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5883 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5887 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5888 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5890 if (vd->vdev_children != 0 &&
5891 (error = vdev_create(vd, txg, B_FALSE)) != 0)
5892 return (spa_vdev_exit(spa, vd, txg, error));
5895 * We must validate the spares and l2cache devices after checking the
5896 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5898 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5899 return (spa_vdev_exit(spa, vd, txg, error));
5902 * If we are in the middle of a device removal, we can only add
5903 * devices which match the existing devices in the pool.
5904 * If we are in the middle of a removal, or have some indirect
5905 * vdevs, we can not add raidz toplevels.
5907 if (spa->spa_vdev_removal != NULL ||
5908 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5909 for (int c = 0; c < vd->vdev_children; c++) {
5910 tvd = vd->vdev_child[c];
5911 if (spa->spa_vdev_removal != NULL &&
5912 tvd->vdev_ashift != spa->spa_max_ashift) {
5913 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5915 /* Fail if top level vdev is raidz */
5916 if (tvd->vdev_ops == &vdev_raidz_ops) {
5917 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5920 * Need the top level mirror to be
5921 * a mirror of leaf vdevs only
5923 if (tvd->vdev_ops == &vdev_mirror_ops) {
5924 for (uint64_t cid = 0;
5925 cid < tvd->vdev_children; cid++) {
5926 vdev_t *cvd = tvd->vdev_child[cid];
5927 if (!cvd->vdev_ops->vdev_op_leaf) {
5928 return (spa_vdev_exit(spa, vd,
5936 for (int c = 0; c < vd->vdev_children; c++) {
5939 * Set the vdev id to the first hole, if one exists.
5941 for (id = 0; id < rvd->vdev_children; id++) {
5942 if (rvd->vdev_child[id]->vdev_ishole) {
5943 vdev_free(rvd->vdev_child[id]);
5947 tvd = vd->vdev_child[c];
5948 vdev_remove_child(vd, tvd);
5950 vdev_add_child(rvd, tvd);
5951 vdev_config_dirty(tvd);
5955 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5956 ZPOOL_CONFIG_SPARES);
5957 spa_load_spares(spa);
5958 spa->spa_spares.sav_sync = B_TRUE;
5961 if (nl2cache != 0) {
5962 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5963 ZPOOL_CONFIG_L2CACHE);
5964 spa_load_l2cache(spa);
5965 spa->spa_l2cache.sav_sync = B_TRUE;
5969 * We have to be careful when adding new vdevs to an existing pool.
5970 * If other threads start allocating from these vdevs before we
5971 * sync the config cache, and we lose power, then upon reboot we may
5972 * fail to open the pool because there are DVAs that the config cache
5973 * can't translate. Therefore, we first add the vdevs without
5974 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5975 * and then let spa_config_update() initialize the new metaslabs.
5977 * spa_load() checks for added-but-not-initialized vdevs, so that
5978 * if we lose power at any point in this sequence, the remaining
5979 * steps will be completed the next time we load the pool.
5981 (void) spa_vdev_exit(spa, vd, txg, 0);
5983 mutex_enter(&spa_namespace_lock);
5984 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5985 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5986 mutex_exit(&spa_namespace_lock);
5992 * Attach a device to a mirror. The arguments are the path to any device
5993 * in the mirror, and the nvroot for the new device. If the path specifies
5994 * a device that is not mirrored, we automatically insert the mirror vdev.
5996 * If 'replacing' is specified, the new device is intended to replace the
5997 * existing device; in this case the two devices are made into their own
5998 * mirror using the 'replacing' vdev, which is functionally identical to
5999 * the mirror vdev (it actually reuses all the same ops) but has a few
6000 * extra rules: you can't attach to it after it's been created, and upon
6001 * completion of resilvering, the first disk (the one being replaced)
6002 * is automatically detached.
6005 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
6007 uint64_t txg, dtl_max_txg;
6008 vdev_t *rvd = spa->spa_root_vdev;
6009 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
6011 char *oldvdpath, *newvdpath;
6015 ASSERT(spa_writeable(spa));
6017 txg = spa_vdev_enter(spa);
6019 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
6021 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6022 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6023 error = (spa_has_checkpoint(spa)) ?
6024 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6025 return (spa_vdev_exit(spa, NULL, txg, error));
6028 if (spa->spa_vdev_removal != NULL)
6029 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6032 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6034 if (!oldvd->vdev_ops->vdev_op_leaf)
6035 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6037 pvd = oldvd->vdev_parent;
6039 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6040 VDEV_ALLOC_ATTACH)) != 0)
6041 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6043 if (newrootvd->vdev_children != 1)
6044 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6046 newvd = newrootvd->vdev_child[0];
6048 if (!newvd->vdev_ops->vdev_op_leaf)
6049 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6051 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6052 return (spa_vdev_exit(spa, newrootvd, txg, error));
6055 * Spares can't replace logs
6057 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6058 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6062 * For attach, the only allowable parent is a mirror or the root
6065 if (pvd->vdev_ops != &vdev_mirror_ops &&
6066 pvd->vdev_ops != &vdev_root_ops)
6067 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6069 pvops = &vdev_mirror_ops;
6072 * Active hot spares can only be replaced by inactive hot
6075 if (pvd->vdev_ops == &vdev_spare_ops &&
6076 oldvd->vdev_isspare &&
6077 !spa_has_spare(spa, newvd->vdev_guid))
6078 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6081 * If the source is a hot spare, and the parent isn't already a
6082 * spare, then we want to create a new hot spare. Otherwise, we
6083 * want to create a replacing vdev. The user is not allowed to
6084 * attach to a spared vdev child unless the 'isspare' state is
6085 * the same (spare replaces spare, non-spare replaces
6088 if (pvd->vdev_ops == &vdev_replacing_ops &&
6089 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6090 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6091 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6092 newvd->vdev_isspare != oldvd->vdev_isspare) {
6093 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6096 if (newvd->vdev_isspare)
6097 pvops = &vdev_spare_ops;
6099 pvops = &vdev_replacing_ops;
6103 * Make sure the new device is big enough.
6105 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6106 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6109 * The new device cannot have a higher alignment requirement
6110 * than the top-level vdev.
6112 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6113 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
6116 * If this is an in-place replacement, update oldvd's path and devid
6117 * to make it distinguishable from newvd, and unopenable from now on.
6119 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6120 spa_strfree(oldvd->vdev_path);
6121 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6123 (void) sprintf(oldvd->vdev_path, "%s/%s",
6124 newvd->vdev_path, "old");
6125 if (oldvd->vdev_devid != NULL) {
6126 spa_strfree(oldvd->vdev_devid);
6127 oldvd->vdev_devid = NULL;
6131 /* mark the device being resilvered */
6132 newvd->vdev_resilver_txg = txg;
6135 * If the parent is not a mirror, or if we're replacing, insert the new
6136 * mirror/replacing/spare vdev above oldvd.
6138 if (pvd->vdev_ops != pvops)
6139 pvd = vdev_add_parent(oldvd, pvops);
6141 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6142 ASSERT(pvd->vdev_ops == pvops);
6143 ASSERT(oldvd->vdev_parent == pvd);
6146 * Extract the new device from its root and add it to pvd.
6148 vdev_remove_child(newrootvd, newvd);
6149 newvd->vdev_id = pvd->vdev_children;
6150 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6151 vdev_add_child(pvd, newvd);
6153 tvd = newvd->vdev_top;
6154 ASSERT(pvd->vdev_top == tvd);
6155 ASSERT(tvd->vdev_parent == rvd);
6157 vdev_config_dirty(tvd);
6160 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6161 * for any dmu_sync-ed blocks. It will propagate upward when
6162 * spa_vdev_exit() calls vdev_dtl_reassess().
6164 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6166 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
6167 dtl_max_txg - TXG_INITIAL);
6169 if (newvd->vdev_isspare) {
6170 spa_spare_activate(newvd);
6171 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6174 oldvdpath = spa_strdup(oldvd->vdev_path);
6175 newvdpath = spa_strdup(newvd->vdev_path);
6176 newvd_isspare = newvd->vdev_isspare;
6179 * Mark newvd's DTL dirty in this txg.
6181 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6184 * Schedule the resilver to restart in the future. We do this to
6185 * ensure that dmu_sync-ed blocks have been stitched into the
6186 * respective datasets.
6188 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
6190 if (spa->spa_bootfs)
6191 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6193 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6198 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6200 spa_history_log_internal(spa, "vdev attach", NULL,
6201 "%s vdev=%s %s vdev=%s",
6202 replacing && newvd_isspare ? "spare in" :
6203 replacing ? "replace" : "attach", newvdpath,
6204 replacing ? "for" : "to", oldvdpath);
6206 spa_strfree(oldvdpath);
6207 spa_strfree(newvdpath);
6213 * Detach a device from a mirror or replacing vdev.
6215 * If 'replace_done' is specified, only detach if the parent
6216 * is a replacing vdev.
6219 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6223 vdev_t *rvd = spa->spa_root_vdev;
6224 vdev_t *vd, *pvd, *cvd, *tvd;
6225 boolean_t unspare = B_FALSE;
6226 uint64_t unspare_guid = 0;
6229 ASSERT(spa_writeable(spa));
6231 txg = spa_vdev_enter(spa);
6233 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6236 * Besides being called directly from the userland through the
6237 * ioctl interface, spa_vdev_detach() can be potentially called
6238 * at the end of spa_vdev_resilver_done().
6240 * In the regular case, when we have a checkpoint this shouldn't
6241 * happen as we never empty the DTLs of a vdev during the scrub
6242 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6243 * should never get here when we have a checkpoint.
6245 * That said, even in a case when we checkpoint the pool exactly
6246 * as spa_vdev_resilver_done() calls this function everything
6247 * should be fine as the resilver will return right away.
6249 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6250 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6251 error = (spa_has_checkpoint(spa)) ?
6252 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6253 return (spa_vdev_exit(spa, NULL, txg, error));
6257 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6259 if (!vd->vdev_ops->vdev_op_leaf)
6260 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6262 pvd = vd->vdev_parent;
6265 * If the parent/child relationship is not as expected, don't do it.
6266 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6267 * vdev that's replacing B with C. The user's intent in replacing
6268 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6269 * the replace by detaching C, the expected behavior is to end up
6270 * M(A,B). But suppose that right after deciding to detach C,
6271 * the replacement of B completes. We would have M(A,C), and then
6272 * ask to detach C, which would leave us with just A -- not what
6273 * the user wanted. To prevent this, we make sure that the
6274 * parent/child relationship hasn't changed -- in this example,
6275 * that C's parent is still the replacing vdev R.
6277 if (pvd->vdev_guid != pguid && pguid != 0)
6278 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6281 * Only 'replacing' or 'spare' vdevs can be replaced.
6283 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6284 pvd->vdev_ops != &vdev_spare_ops)
6285 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6287 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6288 spa_version(spa) >= SPA_VERSION_SPARES);
6291 * Only mirror, replacing, and spare vdevs support detach.
6293 if (pvd->vdev_ops != &vdev_replacing_ops &&
6294 pvd->vdev_ops != &vdev_mirror_ops &&
6295 pvd->vdev_ops != &vdev_spare_ops)
6296 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6299 * If this device has the only valid copy of some data,
6300 * we cannot safely detach it.
6302 if (vdev_dtl_required(vd))
6303 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6305 ASSERT(pvd->vdev_children >= 2);
6308 * If we are detaching the second disk from a replacing vdev, then
6309 * check to see if we changed the original vdev's path to have "/old"
6310 * at the end in spa_vdev_attach(). If so, undo that change now.
6312 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6313 vd->vdev_path != NULL) {
6314 size_t len = strlen(vd->vdev_path);
6316 for (int c = 0; c < pvd->vdev_children; c++) {
6317 cvd = pvd->vdev_child[c];
6319 if (cvd == vd || cvd->vdev_path == NULL)
6322 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6323 strcmp(cvd->vdev_path + len, "/old") == 0) {
6324 spa_strfree(cvd->vdev_path);
6325 cvd->vdev_path = spa_strdup(vd->vdev_path);
6332 * If we are detaching the original disk from a spare, then it implies
6333 * that the spare should become a real disk, and be removed from the
6334 * active spare list for the pool.
6336 if (pvd->vdev_ops == &vdev_spare_ops &&
6338 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6342 * Erase the disk labels so the disk can be used for other things.
6343 * This must be done after all other error cases are handled,
6344 * but before we disembowel vd (so we can still do I/O to it).
6345 * But if we can't do it, don't treat the error as fatal --
6346 * it may be that the unwritability of the disk is the reason
6347 * it's being detached!
6349 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6352 * Remove vd from its parent and compact the parent's children.
6354 vdev_remove_child(pvd, vd);
6355 vdev_compact_children(pvd);
6358 * Remember one of the remaining children so we can get tvd below.
6360 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6363 * If we need to remove the remaining child from the list of hot spares,
6364 * do it now, marking the vdev as no longer a spare in the process.
6365 * We must do this before vdev_remove_parent(), because that can
6366 * change the GUID if it creates a new toplevel GUID. For a similar
6367 * reason, we must remove the spare now, in the same txg as the detach;
6368 * otherwise someone could attach a new sibling, change the GUID, and
6369 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6372 ASSERT(cvd->vdev_isspare);
6373 spa_spare_remove(cvd);
6374 unspare_guid = cvd->vdev_guid;
6375 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6376 cvd->vdev_unspare = B_TRUE;
6380 * If the parent mirror/replacing vdev only has one child,
6381 * the parent is no longer needed. Remove it from the tree.
6383 if (pvd->vdev_children == 1) {
6384 if (pvd->vdev_ops == &vdev_spare_ops)
6385 cvd->vdev_unspare = B_FALSE;
6386 vdev_remove_parent(cvd);
6391 * We don't set tvd until now because the parent we just removed
6392 * may have been the previous top-level vdev.
6394 tvd = cvd->vdev_top;
6395 ASSERT(tvd->vdev_parent == rvd);
6398 * Reevaluate the parent vdev state.
6400 vdev_propagate_state(cvd);
6403 * If the 'autoexpand' property is set on the pool then automatically
6404 * try to expand the size of the pool. For example if the device we
6405 * just detached was smaller than the others, it may be possible to
6406 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6407 * first so that we can obtain the updated sizes of the leaf vdevs.
6409 if (spa->spa_autoexpand) {
6411 vdev_expand(tvd, txg);
6414 vdev_config_dirty(tvd);
6417 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6418 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6419 * But first make sure we're not on any *other* txg's DTL list, to
6420 * prevent vd from being accessed after it's freed.
6422 vdpath = spa_strdup(vd->vdev_path);
6423 for (int t = 0; t < TXG_SIZE; t++)
6424 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6425 vd->vdev_detached = B_TRUE;
6426 vdev_dirty(tvd, VDD_DTL, vd, txg);
6428 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6430 /* hang on to the spa before we release the lock */
6431 spa_open_ref(spa, FTAG);
6433 error = spa_vdev_exit(spa, vd, txg, 0);
6435 spa_history_log_internal(spa, "detach", NULL,
6437 spa_strfree(vdpath);
6440 * If this was the removal of the original device in a hot spare vdev,
6441 * then we want to go through and remove the device from the hot spare
6442 * list of every other pool.
6445 spa_t *altspa = NULL;
6447 mutex_enter(&spa_namespace_lock);
6448 while ((altspa = spa_next(altspa)) != NULL) {
6449 if (altspa->spa_state != POOL_STATE_ACTIVE ||
6453 spa_open_ref(altspa, FTAG);
6454 mutex_exit(&spa_namespace_lock);
6455 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6456 mutex_enter(&spa_namespace_lock);
6457 spa_close(altspa, FTAG);
6459 mutex_exit(&spa_namespace_lock);
6461 /* search the rest of the vdevs for spares to remove */
6462 spa_vdev_resilver_done(spa);
6465 /* all done with the spa; OK to release */
6466 mutex_enter(&spa_namespace_lock);
6467 spa_close(spa, FTAG);
6468 mutex_exit(&spa_namespace_lock);
6474 spa_vdev_initialize(spa_t *spa, uint64_t guid, uint64_t cmd_type)
6477 * We hold the namespace lock through the whole function
6478 * to prevent any changes to the pool while we're starting or
6479 * stopping initialization. The config and state locks are held so that
6480 * we can properly assess the vdev state before we commit to
6481 * the initializing operation.
6483 mutex_enter(&spa_namespace_lock);
6484 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6486 /* Look up vdev and ensure it's a leaf. */
6487 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6488 if (vd == NULL || vd->vdev_detached) {
6489 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6490 mutex_exit(&spa_namespace_lock);
6491 return (SET_ERROR(ENODEV));
6492 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
6493 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6494 mutex_exit(&spa_namespace_lock);
6495 return (SET_ERROR(EINVAL));
6496 } else if (!vdev_writeable(vd)) {
6497 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6498 mutex_exit(&spa_namespace_lock);
6499 return (SET_ERROR(EROFS));
6501 mutex_enter(&vd->vdev_initialize_lock);
6502 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6505 * When we activate an initialize action we check to see
6506 * if the vdev_initialize_thread is NULL. We do this instead
6507 * of using the vdev_initialize_state since there might be
6508 * a previous initialization process which has completed but
6509 * the thread is not exited.
6511 if (cmd_type == POOL_INITIALIZE_DO &&
6512 (vd->vdev_initialize_thread != NULL ||
6513 vd->vdev_top->vdev_removing)) {
6514 mutex_exit(&vd->vdev_initialize_lock);
6515 mutex_exit(&spa_namespace_lock);
6516 return (SET_ERROR(EBUSY));
6517 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
6518 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
6519 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
6520 mutex_exit(&vd->vdev_initialize_lock);
6521 mutex_exit(&spa_namespace_lock);
6522 return (SET_ERROR(ESRCH));
6523 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
6524 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
6525 mutex_exit(&vd->vdev_initialize_lock);
6526 mutex_exit(&spa_namespace_lock);
6527 return (SET_ERROR(ESRCH));
6531 case POOL_INITIALIZE_DO:
6532 vdev_initialize(vd);
6534 case POOL_INITIALIZE_CANCEL:
6535 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
6537 case POOL_INITIALIZE_SUSPEND:
6538 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED);
6541 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
6543 mutex_exit(&vd->vdev_initialize_lock);
6545 /* Sync out the initializing state */
6546 txg_wait_synced(spa->spa_dsl_pool, 0);
6547 mutex_exit(&spa_namespace_lock);
6554 * Split a set of devices from their mirrors, and create a new pool from them.
6557 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6558 nvlist_t *props, boolean_t exp)
6561 uint64_t txg, *glist;
6563 uint_t c, children, lastlog;
6564 nvlist_t **child, *nvl, *tmp;
6566 char *altroot = NULL;
6567 vdev_t *rvd, **vml = NULL; /* vdev modify list */
6568 boolean_t activate_slog;
6570 ASSERT(spa_writeable(spa));
6572 txg = spa_vdev_enter(spa);
6574 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6575 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6576 error = (spa_has_checkpoint(spa)) ?
6577 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6578 return (spa_vdev_exit(spa, NULL, txg, error));
6581 /* clear the log and flush everything up to now */
6582 activate_slog = spa_passivate_log(spa);
6583 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6584 error = spa_reset_logs(spa);
6585 txg = spa_vdev_config_enter(spa);
6588 spa_activate_log(spa);
6591 return (spa_vdev_exit(spa, NULL, txg, error));
6593 /* check new spa name before going any further */
6594 if (spa_lookup(newname) != NULL)
6595 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
6598 * scan through all the children to ensure they're all mirrors
6600 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
6601 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
6603 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6605 /* first, check to ensure we've got the right child count */
6606 rvd = spa->spa_root_vdev;
6608 for (c = 0; c < rvd->vdev_children; c++) {
6609 vdev_t *vd = rvd->vdev_child[c];
6611 /* don't count the holes & logs as children */
6612 if (vd->vdev_islog || !vdev_is_concrete(vd)) {
6620 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
6621 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6623 /* next, ensure no spare or cache devices are part of the split */
6624 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
6625 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
6626 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6628 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
6629 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
6631 /* then, loop over each vdev and validate it */
6632 for (c = 0; c < children; c++) {
6633 uint64_t is_hole = 0;
6635 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
6639 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
6640 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
6643 error = SET_ERROR(EINVAL);
6648 /* which disk is going to be split? */
6649 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
6651 error = SET_ERROR(EINVAL);
6655 /* look it up in the spa */
6656 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
6657 if (vml[c] == NULL) {
6658 error = SET_ERROR(ENODEV);
6662 /* make sure there's nothing stopping the split */
6663 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
6664 vml[c]->vdev_islog ||
6665 !vdev_is_concrete(vml[c]) ||
6666 vml[c]->vdev_isspare ||
6667 vml[c]->vdev_isl2cache ||
6668 !vdev_writeable(vml[c]) ||
6669 vml[c]->vdev_children != 0 ||
6670 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
6671 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
6672 error = SET_ERROR(EINVAL);
6676 if (vdev_dtl_required(vml[c])) {
6677 error = SET_ERROR(EBUSY);
6681 /* we need certain info from the top level */
6682 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
6683 vml[c]->vdev_top->vdev_ms_array) == 0);
6684 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
6685 vml[c]->vdev_top->vdev_ms_shift) == 0);
6686 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
6687 vml[c]->vdev_top->vdev_asize) == 0);
6688 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
6689 vml[c]->vdev_top->vdev_ashift) == 0);
6691 /* transfer per-vdev ZAPs */
6692 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
6693 VERIFY0(nvlist_add_uint64(child[c],
6694 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
6696 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
6697 VERIFY0(nvlist_add_uint64(child[c],
6698 ZPOOL_CONFIG_VDEV_TOP_ZAP,
6699 vml[c]->vdev_parent->vdev_top_zap));
6703 kmem_free(vml, children * sizeof (vdev_t *));
6704 kmem_free(glist, children * sizeof (uint64_t));
6705 return (spa_vdev_exit(spa, NULL, txg, error));
6708 /* stop writers from using the disks */
6709 for (c = 0; c < children; c++) {
6711 vml[c]->vdev_offline = B_TRUE;
6713 vdev_reopen(spa->spa_root_vdev);
6716 * Temporarily record the splitting vdevs in the spa config. This
6717 * will disappear once the config is regenerated.
6719 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6720 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
6721 glist, children) == 0);
6722 kmem_free(glist, children * sizeof (uint64_t));
6724 mutex_enter(&spa->spa_props_lock);
6725 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
6727 mutex_exit(&spa->spa_props_lock);
6728 spa->spa_config_splitting = nvl;
6729 vdev_config_dirty(spa->spa_root_vdev);
6731 /* configure and create the new pool */
6732 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
6733 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6734 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
6735 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6736 spa_version(spa)) == 0);
6737 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
6738 spa->spa_config_txg) == 0);
6739 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
6740 spa_generate_guid(NULL)) == 0);
6741 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
6742 (void) nvlist_lookup_string(props,
6743 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6745 /* add the new pool to the namespace */
6746 newspa = spa_add(newname, config, altroot);
6747 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
6748 newspa->spa_config_txg = spa->spa_config_txg;
6749 spa_set_log_state(newspa, SPA_LOG_CLEAR);
6751 /* release the spa config lock, retaining the namespace lock */
6752 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6754 if (zio_injection_enabled)
6755 zio_handle_panic_injection(spa, FTAG, 1);
6757 spa_activate(newspa, spa_mode_global);
6758 spa_async_suspend(newspa);
6760 for (c = 0; c < children; c++) {
6761 if (vml[c] != NULL) {
6763 * Temporarily stop the initializing activity. We set
6764 * the state to ACTIVE so that we know to resume
6765 * the initializing once the split has completed.
6767 mutex_enter(&vml[c]->vdev_initialize_lock);
6768 vdev_initialize_stop(vml[c], VDEV_INITIALIZE_ACTIVE);
6769 mutex_exit(&vml[c]->vdev_initialize_lock);
6774 /* mark that we are creating new spa by splitting */
6775 newspa->spa_splitting_newspa = B_TRUE;
6777 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
6779 /* create the new pool from the disks of the original pool */
6780 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
6782 newspa->spa_splitting_newspa = B_FALSE;
6787 /* if that worked, generate a real config for the new pool */
6788 if (newspa->spa_root_vdev != NULL) {
6789 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
6790 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6791 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
6792 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
6793 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
6798 if (props != NULL) {
6799 spa_configfile_set(newspa, props, B_FALSE);
6800 error = spa_prop_set(newspa, props);
6805 /* flush everything */
6806 txg = spa_vdev_config_enter(newspa);
6807 vdev_config_dirty(newspa->spa_root_vdev);
6808 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
6810 if (zio_injection_enabled)
6811 zio_handle_panic_injection(spa, FTAG, 2);
6813 spa_async_resume(newspa);
6815 /* finally, update the original pool's config */
6816 txg = spa_vdev_config_enter(spa);
6817 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
6818 error = dmu_tx_assign(tx, TXG_WAIT);
6821 for (c = 0; c < children; c++) {
6822 if (vml[c] != NULL) {
6825 spa_history_log_internal(spa, "detach", tx,
6826 "vdev=%s", vml[c]->vdev_path);
6831 spa->spa_avz_action = AVZ_ACTION_REBUILD;
6832 vdev_config_dirty(spa->spa_root_vdev);
6833 spa->spa_config_splitting = NULL;
6837 (void) spa_vdev_exit(spa, NULL, txg, 0);
6839 if (zio_injection_enabled)
6840 zio_handle_panic_injection(spa, FTAG, 3);
6842 /* split is complete; log a history record */
6843 spa_history_log_internal(newspa, "split", NULL,
6844 "from pool %s", spa_name(spa));
6846 kmem_free(vml, children * sizeof (vdev_t *));
6848 /* if we're not going to mount the filesystems in userland, export */
6850 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
6857 spa_deactivate(newspa);
6860 txg = spa_vdev_config_enter(spa);
6862 /* re-online all offlined disks */
6863 for (c = 0; c < children; c++) {
6865 vml[c]->vdev_offline = B_FALSE;
6868 /* restart initializing disks as necessary */
6869 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
6871 vdev_reopen(spa->spa_root_vdev);
6873 nvlist_free(spa->spa_config_splitting);
6874 spa->spa_config_splitting = NULL;
6875 (void) spa_vdev_exit(spa, NULL, txg, error);
6877 kmem_free(vml, children * sizeof (vdev_t *));
6882 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6883 * currently spared, so we can detach it.
6886 spa_vdev_resilver_done_hunt(vdev_t *vd)
6888 vdev_t *newvd, *oldvd;
6890 for (int c = 0; c < vd->vdev_children; c++) {
6891 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
6897 * Check for a completed replacement. We always consider the first
6898 * vdev in the list to be the oldest vdev, and the last one to be
6899 * the newest (see spa_vdev_attach() for how that works). In
6900 * the case where the newest vdev is faulted, we will not automatically
6901 * remove it after a resilver completes. This is OK as it will require
6902 * user intervention to determine which disk the admin wishes to keep.
6904 if (vd->vdev_ops == &vdev_replacing_ops) {
6905 ASSERT(vd->vdev_children > 1);
6907 newvd = vd->vdev_child[vd->vdev_children - 1];
6908 oldvd = vd->vdev_child[0];
6910 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
6911 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6912 !vdev_dtl_required(oldvd))
6917 * Check for a completed resilver with the 'unspare' flag set.
6918 * Also potentially update faulted state.
6920 if (vd->vdev_ops == &vdev_spare_ops) {
6921 vdev_t *first = vd->vdev_child[0];
6922 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
6924 if (last->vdev_unspare) {
6927 } else if (first->vdev_unspare) {
6934 if (oldvd != NULL &&
6935 vdev_dtl_empty(newvd, DTL_MISSING) &&
6936 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6937 !vdev_dtl_required(oldvd))
6940 vdev_propagate_state(vd);
6943 * If there are more than two spares attached to a disk,
6944 * and those spares are not required, then we want to
6945 * attempt to free them up now so that they can be used
6946 * by other pools. Once we're back down to a single
6947 * disk+spare, we stop removing them.
6949 if (vd->vdev_children > 2) {
6950 newvd = vd->vdev_child[1];
6952 if (newvd->vdev_isspare && last->vdev_isspare &&
6953 vdev_dtl_empty(last, DTL_MISSING) &&
6954 vdev_dtl_empty(last, DTL_OUTAGE) &&
6955 !vdev_dtl_required(newvd))
6964 spa_vdev_resilver_done(spa_t *spa)
6966 vdev_t *vd, *pvd, *ppvd;
6967 uint64_t guid, sguid, pguid, ppguid;
6969 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6971 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
6972 pvd = vd->vdev_parent;
6973 ppvd = pvd->vdev_parent;
6974 guid = vd->vdev_guid;
6975 pguid = pvd->vdev_guid;
6976 ppguid = ppvd->vdev_guid;
6979 * If we have just finished replacing a hot spared device, then
6980 * we need to detach the parent's first child (the original hot
6983 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
6984 ppvd->vdev_children == 2) {
6985 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
6986 sguid = ppvd->vdev_child[1]->vdev_guid;
6988 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
6990 spa_config_exit(spa, SCL_ALL, FTAG);
6991 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
6993 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
6995 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6998 spa_config_exit(spa, SCL_ALL, FTAG);
7002 * Update the stored path or FRU for this vdev.
7005 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
7009 boolean_t sync = B_FALSE;
7011 ASSERT(spa_writeable(spa));
7013 spa_vdev_state_enter(spa, SCL_ALL);
7015 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
7016 return (spa_vdev_state_exit(spa, NULL, ENOENT));
7018 if (!vd->vdev_ops->vdev_op_leaf)
7019 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
7022 if (strcmp(value, vd->vdev_path) != 0) {
7023 spa_strfree(vd->vdev_path);
7024 vd->vdev_path = spa_strdup(value);
7028 if (vd->vdev_fru == NULL) {
7029 vd->vdev_fru = spa_strdup(value);
7031 } else if (strcmp(value, vd->vdev_fru) != 0) {
7032 spa_strfree(vd->vdev_fru);
7033 vd->vdev_fru = spa_strdup(value);
7038 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
7042 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
7044 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
7048 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
7050 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
7054 * ==========================================================================
7056 * ==========================================================================
7059 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
7061 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7063 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7064 return (SET_ERROR(EBUSY));
7066 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
7070 spa_scan_stop(spa_t *spa)
7072 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7073 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7074 return (SET_ERROR(EBUSY));
7075 return (dsl_scan_cancel(spa->spa_dsl_pool));
7079 spa_scan(spa_t *spa, pool_scan_func_t func)
7081 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7083 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
7084 return (SET_ERROR(ENOTSUP));
7087 * If a resilver was requested, but there is no DTL on a
7088 * writeable leaf device, we have nothing to do.
7090 if (func == POOL_SCAN_RESILVER &&
7091 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
7092 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
7096 return (dsl_scan(spa->spa_dsl_pool, func));
7100 * ==========================================================================
7101 * SPA async task processing
7102 * ==========================================================================
7106 spa_async_remove(spa_t *spa, vdev_t *vd)
7108 if (vd->vdev_remove_wanted) {
7109 vd->vdev_remove_wanted = B_FALSE;
7110 vd->vdev_delayed_close = B_FALSE;
7111 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
7114 * We want to clear the stats, but we don't want to do a full
7115 * vdev_clear() as that will cause us to throw away
7116 * degraded/faulted state as well as attempt to reopen the
7117 * device, all of which is a waste.
7119 vd->vdev_stat.vs_read_errors = 0;
7120 vd->vdev_stat.vs_write_errors = 0;
7121 vd->vdev_stat.vs_checksum_errors = 0;
7123 vdev_state_dirty(vd->vdev_top);
7124 /* Tell userspace that the vdev is gone. */
7125 zfs_post_remove(spa, vd);
7128 for (int c = 0; c < vd->vdev_children; c++)
7129 spa_async_remove(spa, vd->vdev_child[c]);
7133 spa_async_probe(spa_t *spa, vdev_t *vd)
7135 if (vd->vdev_probe_wanted) {
7136 vd->vdev_probe_wanted = B_FALSE;
7137 vdev_reopen(vd); /* vdev_open() does the actual probe */
7140 for (int c = 0; c < vd->vdev_children; c++)
7141 spa_async_probe(spa, vd->vdev_child[c]);
7145 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
7151 if (!spa->spa_autoexpand)
7154 for (int c = 0; c < vd->vdev_children; c++) {
7155 vdev_t *cvd = vd->vdev_child[c];
7156 spa_async_autoexpand(spa, cvd);
7159 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
7162 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
7163 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
7165 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7166 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
7168 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
7169 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
7172 kmem_free(physpath, MAXPATHLEN);
7176 spa_async_thread(void *arg)
7178 spa_t *spa = (spa_t *)arg;
7181 ASSERT(spa->spa_sync_on);
7183 mutex_enter(&spa->spa_async_lock);
7184 tasks = spa->spa_async_tasks;
7185 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
7186 mutex_exit(&spa->spa_async_lock);
7189 * See if the config needs to be updated.
7191 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7192 uint64_t old_space, new_space;
7194 mutex_enter(&spa_namespace_lock);
7195 old_space = metaslab_class_get_space(spa_normal_class(spa));
7196 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7197 new_space = metaslab_class_get_space(spa_normal_class(spa));
7198 mutex_exit(&spa_namespace_lock);
7201 * If the pool grew as a result of the config update,
7202 * then log an internal history event.
7204 if (new_space != old_space) {
7205 spa_history_log_internal(spa, "vdev online", NULL,
7206 "pool '%s' size: %llu(+%llu)",
7207 spa_name(spa), new_space, new_space - old_space);
7211 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7212 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7213 spa_async_autoexpand(spa, spa->spa_root_vdev);
7214 spa_config_exit(spa, SCL_CONFIG, FTAG);
7218 * See if any devices need to be probed.
7220 if (tasks & SPA_ASYNC_PROBE) {
7221 spa_vdev_state_enter(spa, SCL_NONE);
7222 spa_async_probe(spa, spa->spa_root_vdev);
7223 (void) spa_vdev_state_exit(spa, NULL, 0);
7227 * If any devices are done replacing, detach them.
7229 if (tasks & SPA_ASYNC_RESILVER_DONE)
7230 spa_vdev_resilver_done(spa);
7233 * Kick off a resilver.
7235 if (tasks & SPA_ASYNC_RESILVER)
7236 dsl_resilver_restart(spa->spa_dsl_pool, 0);
7238 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
7239 mutex_enter(&spa_namespace_lock);
7240 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7241 vdev_initialize_restart(spa->spa_root_vdev);
7242 spa_config_exit(spa, SCL_CONFIG, FTAG);
7243 mutex_exit(&spa_namespace_lock);
7247 * Let the world know that we're done.
7249 mutex_enter(&spa->spa_async_lock);
7250 spa->spa_async_thread = NULL;
7251 cv_broadcast(&spa->spa_async_cv);
7252 mutex_exit(&spa->spa_async_lock);
7257 spa_async_thread_vd(void *arg)
7262 mutex_enter(&spa->spa_async_lock);
7263 tasks = spa->spa_async_tasks;
7265 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
7266 mutex_exit(&spa->spa_async_lock);
7269 * See if any devices need to be marked REMOVED.
7271 if (tasks & SPA_ASYNC_REMOVE) {
7272 spa_vdev_state_enter(spa, SCL_NONE);
7273 spa_async_remove(spa, spa->spa_root_vdev);
7274 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7275 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7276 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7277 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7278 (void) spa_vdev_state_exit(spa, NULL, 0);
7282 * Let the world know that we're done.
7284 mutex_enter(&spa->spa_async_lock);
7285 tasks = spa->spa_async_tasks;
7286 if ((tasks & SPA_ASYNC_REMOVE) != 0)
7288 spa->spa_async_thread_vd = NULL;
7289 cv_broadcast(&spa->spa_async_cv);
7290 mutex_exit(&spa->spa_async_lock);
7295 spa_async_suspend(spa_t *spa)
7297 mutex_enter(&spa->spa_async_lock);
7298 spa->spa_async_suspended++;
7299 while (spa->spa_async_thread != NULL ||
7300 spa->spa_async_thread_vd != NULL)
7301 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
7302 mutex_exit(&spa->spa_async_lock);
7304 spa_vdev_remove_suspend(spa);
7306 zthr_t *condense_thread = spa->spa_condense_zthr;
7307 if (condense_thread != NULL && zthr_isrunning(condense_thread))
7308 VERIFY0(zthr_cancel(condense_thread));
7310 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7311 if (discard_thread != NULL && zthr_isrunning(discard_thread))
7312 VERIFY0(zthr_cancel(discard_thread));
7316 spa_async_resume(spa_t *spa)
7318 mutex_enter(&spa->spa_async_lock);
7319 ASSERT(spa->spa_async_suspended != 0);
7320 spa->spa_async_suspended--;
7321 mutex_exit(&spa->spa_async_lock);
7322 spa_restart_removal(spa);
7324 zthr_t *condense_thread = spa->spa_condense_zthr;
7325 if (condense_thread != NULL && !zthr_isrunning(condense_thread))
7326 zthr_resume(condense_thread);
7328 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7329 if (discard_thread != NULL && !zthr_isrunning(discard_thread))
7330 zthr_resume(discard_thread);
7334 spa_async_tasks_pending(spa_t *spa)
7336 uint_t non_config_tasks;
7338 boolean_t config_task_suspended;
7340 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
7342 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
7343 if (spa->spa_ccw_fail_time == 0) {
7344 config_task_suspended = B_FALSE;
7346 config_task_suspended =
7347 (gethrtime() - spa->spa_ccw_fail_time) <
7348 (zfs_ccw_retry_interval * NANOSEC);
7351 return (non_config_tasks || (config_task && !config_task_suspended));
7355 spa_async_dispatch(spa_t *spa)
7357 mutex_enter(&spa->spa_async_lock);
7358 if (spa_async_tasks_pending(spa) &&
7359 !spa->spa_async_suspended &&
7360 spa->spa_async_thread == NULL &&
7362 spa->spa_async_thread = thread_create(NULL, 0,
7363 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
7364 mutex_exit(&spa->spa_async_lock);
7368 spa_async_dispatch_vd(spa_t *spa)
7370 mutex_enter(&spa->spa_async_lock);
7371 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
7372 !spa->spa_async_suspended &&
7373 spa->spa_async_thread_vd == NULL &&
7375 spa->spa_async_thread_vd = thread_create(NULL, 0,
7376 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
7377 mutex_exit(&spa->spa_async_lock);
7381 spa_async_request(spa_t *spa, int task)
7383 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
7384 mutex_enter(&spa->spa_async_lock);
7385 spa->spa_async_tasks |= task;
7386 mutex_exit(&spa->spa_async_lock);
7387 spa_async_dispatch_vd(spa);
7391 * ==========================================================================
7392 * SPA syncing routines
7393 * ==========================================================================
7397 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7400 bpobj_enqueue(bpo, bp, tx);
7405 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7409 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
7410 BP_GET_PSIZE(bp), zio->io_flags));
7415 * Note: this simple function is not inlined to make it easier to dtrace the
7416 * amount of time spent syncing frees.
7419 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
7421 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7422 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
7423 VERIFY(zio_wait(zio) == 0);
7427 * Note: this simple function is not inlined to make it easier to dtrace the
7428 * amount of time spent syncing deferred frees.
7431 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
7433 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7434 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
7435 spa_free_sync_cb, zio, tx), ==, 0);
7436 VERIFY0(zio_wait(zio));
7441 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
7443 char *packed = NULL;
7448 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
7451 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
7452 * information. This avoids the dmu_buf_will_dirty() path and
7453 * saves us a pre-read to get data we don't actually care about.
7455 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
7456 packed = kmem_alloc(bufsize, KM_SLEEP);
7458 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
7460 bzero(packed + nvsize, bufsize - nvsize);
7462 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
7464 kmem_free(packed, bufsize);
7466 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
7467 dmu_buf_will_dirty(db, tx);
7468 *(uint64_t *)db->db_data = nvsize;
7469 dmu_buf_rele(db, FTAG);
7473 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
7474 const char *config, const char *entry)
7484 * Update the MOS nvlist describing the list of available devices.
7485 * spa_validate_aux() will have already made sure this nvlist is
7486 * valid and the vdevs are labeled appropriately.
7488 if (sav->sav_object == 0) {
7489 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
7490 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
7491 sizeof (uint64_t), tx);
7492 VERIFY(zap_update(spa->spa_meta_objset,
7493 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
7494 &sav->sav_object, tx) == 0);
7497 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7498 if (sav->sav_count == 0) {
7499 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
7501 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
7502 for (i = 0; i < sav->sav_count; i++)
7503 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
7504 B_FALSE, VDEV_CONFIG_L2CACHE);
7505 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
7506 sav->sav_count) == 0);
7507 for (i = 0; i < sav->sav_count; i++)
7508 nvlist_free(list[i]);
7509 kmem_free(list, sav->sav_count * sizeof (void *));
7512 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
7513 nvlist_free(nvroot);
7515 sav->sav_sync = B_FALSE;
7519 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
7520 * The all-vdev ZAP must be empty.
7523 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
7525 spa_t *spa = vd->vdev_spa;
7526 if (vd->vdev_top_zap != 0) {
7527 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7528 vd->vdev_top_zap, tx));
7530 if (vd->vdev_leaf_zap != 0) {
7531 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7532 vd->vdev_leaf_zap, tx));
7534 for (uint64_t i = 0; i < vd->vdev_children; i++) {
7535 spa_avz_build(vd->vdev_child[i], avz, tx);
7540 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
7545 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
7546 * its config may not be dirty but we still need to build per-vdev ZAPs.
7547 * Similarly, if the pool is being assembled (e.g. after a split), we
7548 * need to rebuild the AVZ although the config may not be dirty.
7550 if (list_is_empty(&spa->spa_config_dirty_list) &&
7551 spa->spa_avz_action == AVZ_ACTION_NONE)
7554 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7556 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
7557 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
7558 spa->spa_all_vdev_zaps != 0);
7560 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
7561 /* Make and build the new AVZ */
7562 uint64_t new_avz = zap_create(spa->spa_meta_objset,
7563 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
7564 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
7566 /* Diff old AVZ with new one */
7570 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7571 spa->spa_all_vdev_zaps);
7572 zap_cursor_retrieve(&zc, &za) == 0;
7573 zap_cursor_advance(&zc)) {
7574 uint64_t vdzap = za.za_first_integer;
7575 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
7578 * ZAP is listed in old AVZ but not in new one;
7581 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
7586 zap_cursor_fini(&zc);
7588 /* Destroy the old AVZ */
7589 VERIFY0(zap_destroy(spa->spa_meta_objset,
7590 spa->spa_all_vdev_zaps, tx));
7592 /* Replace the old AVZ in the dir obj with the new one */
7593 VERIFY0(zap_update(spa->spa_meta_objset,
7594 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
7595 sizeof (new_avz), 1, &new_avz, tx));
7597 spa->spa_all_vdev_zaps = new_avz;
7598 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
7602 /* Walk through the AVZ and destroy all listed ZAPs */
7603 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7604 spa->spa_all_vdev_zaps);
7605 zap_cursor_retrieve(&zc, &za) == 0;
7606 zap_cursor_advance(&zc)) {
7607 uint64_t zap = za.za_first_integer;
7608 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
7611 zap_cursor_fini(&zc);
7613 /* Destroy and unlink the AVZ itself */
7614 VERIFY0(zap_destroy(spa->spa_meta_objset,
7615 spa->spa_all_vdev_zaps, tx));
7616 VERIFY0(zap_remove(spa->spa_meta_objset,
7617 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
7618 spa->spa_all_vdev_zaps = 0;
7621 if (spa->spa_all_vdev_zaps == 0) {
7622 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
7623 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
7624 DMU_POOL_VDEV_ZAP_MAP, tx);
7626 spa->spa_avz_action = AVZ_ACTION_NONE;
7628 /* Create ZAPs for vdevs that don't have them. */
7629 vdev_construct_zaps(spa->spa_root_vdev, tx);
7631 config = spa_config_generate(spa, spa->spa_root_vdev,
7632 dmu_tx_get_txg(tx), B_FALSE);
7635 * If we're upgrading the spa version then make sure that
7636 * the config object gets updated with the correct version.
7638 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
7639 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7640 spa->spa_uberblock.ub_version);
7642 spa_config_exit(spa, SCL_STATE, FTAG);
7644 nvlist_free(spa->spa_config_syncing);
7645 spa->spa_config_syncing = config;
7647 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
7651 spa_sync_version(void *arg, dmu_tx_t *tx)
7653 uint64_t *versionp = arg;
7654 uint64_t version = *versionp;
7655 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7658 * Setting the version is special cased when first creating the pool.
7660 ASSERT(tx->tx_txg != TXG_INITIAL);
7662 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
7663 ASSERT(version >= spa_version(spa));
7665 spa->spa_uberblock.ub_version = version;
7666 vdev_config_dirty(spa->spa_root_vdev);
7667 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
7671 * Set zpool properties.
7674 spa_sync_props(void *arg, dmu_tx_t *tx)
7676 nvlist_t *nvp = arg;
7677 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7678 objset_t *mos = spa->spa_meta_objset;
7679 nvpair_t *elem = NULL;
7681 mutex_enter(&spa->spa_props_lock);
7683 while ((elem = nvlist_next_nvpair(nvp, elem))) {
7685 char *strval, *fname;
7687 const char *propname;
7688 zprop_type_t proptype;
7691 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
7692 case ZPOOL_PROP_INVAL:
7694 * We checked this earlier in spa_prop_validate().
7696 ASSERT(zpool_prop_feature(nvpair_name(elem)));
7698 fname = strchr(nvpair_name(elem), '@') + 1;
7699 VERIFY0(zfeature_lookup_name(fname, &fid));
7701 spa_feature_enable(spa, fid, tx);
7702 spa_history_log_internal(spa, "set", tx,
7703 "%s=enabled", nvpair_name(elem));
7706 case ZPOOL_PROP_VERSION:
7707 intval = fnvpair_value_uint64(elem);
7709 * The version is synced seperatly before other
7710 * properties and should be correct by now.
7712 ASSERT3U(spa_version(spa), >=, intval);
7715 case ZPOOL_PROP_ALTROOT:
7717 * 'altroot' is a non-persistent property. It should
7718 * have been set temporarily at creation or import time.
7720 ASSERT(spa->spa_root != NULL);
7723 case ZPOOL_PROP_READONLY:
7724 case ZPOOL_PROP_CACHEFILE:
7726 * 'readonly' and 'cachefile' are also non-persisitent
7730 case ZPOOL_PROP_COMMENT:
7731 strval = fnvpair_value_string(elem);
7732 if (spa->spa_comment != NULL)
7733 spa_strfree(spa->spa_comment);
7734 spa->spa_comment = spa_strdup(strval);
7736 * We need to dirty the configuration on all the vdevs
7737 * so that their labels get updated. It's unnecessary
7738 * to do this for pool creation since the vdev's
7739 * configuratoin has already been dirtied.
7741 if (tx->tx_txg != TXG_INITIAL)
7742 vdev_config_dirty(spa->spa_root_vdev);
7743 spa_history_log_internal(spa, "set", tx,
7744 "%s=%s", nvpair_name(elem), strval);
7748 * Set pool property values in the poolprops mos object.
7750 if (spa->spa_pool_props_object == 0) {
7751 spa->spa_pool_props_object =
7752 zap_create_link(mos, DMU_OT_POOL_PROPS,
7753 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
7757 /* normalize the property name */
7758 propname = zpool_prop_to_name(prop);
7759 proptype = zpool_prop_get_type(prop);
7761 if (nvpair_type(elem) == DATA_TYPE_STRING) {
7762 ASSERT(proptype == PROP_TYPE_STRING);
7763 strval = fnvpair_value_string(elem);
7764 VERIFY0(zap_update(mos,
7765 spa->spa_pool_props_object, propname,
7766 1, strlen(strval) + 1, strval, tx));
7767 spa_history_log_internal(spa, "set", tx,
7768 "%s=%s", nvpair_name(elem), strval);
7769 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
7770 intval = fnvpair_value_uint64(elem);
7772 if (proptype == PROP_TYPE_INDEX) {
7774 VERIFY0(zpool_prop_index_to_string(
7775 prop, intval, &unused));
7777 VERIFY0(zap_update(mos,
7778 spa->spa_pool_props_object, propname,
7779 8, 1, &intval, tx));
7780 spa_history_log_internal(spa, "set", tx,
7781 "%s=%lld", nvpair_name(elem), intval);
7783 ASSERT(0); /* not allowed */
7787 case ZPOOL_PROP_DELEGATION:
7788 spa->spa_delegation = intval;
7790 case ZPOOL_PROP_BOOTFS:
7791 spa->spa_bootfs = intval;
7793 case ZPOOL_PROP_FAILUREMODE:
7794 spa->spa_failmode = intval;
7796 case ZPOOL_PROP_AUTOEXPAND:
7797 spa->spa_autoexpand = intval;
7798 if (tx->tx_txg != TXG_INITIAL)
7799 spa_async_request(spa,
7800 SPA_ASYNC_AUTOEXPAND);
7802 case ZPOOL_PROP_DEDUPDITTO:
7803 spa->spa_dedup_ditto = intval;
7812 mutex_exit(&spa->spa_props_lock);
7816 * Perform one-time upgrade on-disk changes. spa_version() does not
7817 * reflect the new version this txg, so there must be no changes this
7818 * txg to anything that the upgrade code depends on after it executes.
7819 * Therefore this must be called after dsl_pool_sync() does the sync
7823 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
7825 dsl_pool_t *dp = spa->spa_dsl_pool;
7827 ASSERT(spa->spa_sync_pass == 1);
7829 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7831 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
7832 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
7833 dsl_pool_create_origin(dp, tx);
7835 /* Keeping the origin open increases spa_minref */
7836 spa->spa_minref += 3;
7839 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
7840 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
7841 dsl_pool_upgrade_clones(dp, tx);
7844 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
7845 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
7846 dsl_pool_upgrade_dir_clones(dp, tx);
7848 /* Keeping the freedir open increases spa_minref */
7849 spa->spa_minref += 3;
7852 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
7853 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7854 spa_feature_create_zap_objects(spa, tx);
7858 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7859 * when possibility to use lz4 compression for metadata was added
7860 * Old pools that have this feature enabled must be upgraded to have
7861 * this feature active
7863 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7864 boolean_t lz4_en = spa_feature_is_enabled(spa,
7865 SPA_FEATURE_LZ4_COMPRESS);
7866 boolean_t lz4_ac = spa_feature_is_active(spa,
7867 SPA_FEATURE_LZ4_COMPRESS);
7869 if (lz4_en && !lz4_ac)
7870 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
7874 * If we haven't written the salt, do so now. Note that the
7875 * feature may not be activated yet, but that's fine since
7876 * the presence of this ZAP entry is backwards compatible.
7878 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7879 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
7880 VERIFY0(zap_add(spa->spa_meta_objset,
7881 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
7882 sizeof (spa->spa_cksum_salt.zcs_bytes),
7883 spa->spa_cksum_salt.zcs_bytes, tx));
7886 rrw_exit(&dp->dp_config_rwlock, FTAG);
7890 vdev_indirect_state_sync_verify(vdev_t *vd)
7892 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7893 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
7895 if (vd->vdev_ops == &vdev_indirect_ops) {
7896 ASSERT(vim != NULL);
7897 ASSERT(vib != NULL);
7900 if (vdev_obsolete_sm_object(vd) != 0) {
7901 ASSERT(vd->vdev_obsolete_sm != NULL);
7902 ASSERT(vd->vdev_removing ||
7903 vd->vdev_ops == &vdev_indirect_ops);
7904 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
7905 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
7907 ASSERT3U(vdev_obsolete_sm_object(vd), ==,
7908 space_map_object(vd->vdev_obsolete_sm));
7909 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
7910 space_map_allocated(vd->vdev_obsolete_sm));
7912 ASSERT(vd->vdev_obsolete_segments != NULL);
7915 * Since frees / remaps to an indirect vdev can only
7916 * happen in syncing context, the obsolete segments
7917 * tree must be empty when we start syncing.
7919 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
7923 * Sync the specified transaction group. New blocks may be dirtied as
7924 * part of the process, so we iterate until it converges.
7927 spa_sync(spa_t *spa, uint64_t txg)
7929 dsl_pool_t *dp = spa->spa_dsl_pool;
7930 objset_t *mos = spa->spa_meta_objset;
7931 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
7932 vdev_t *rvd = spa->spa_root_vdev;
7936 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
7937 zfs_vdev_queue_depth_pct / 100;
7939 VERIFY(spa_writeable(spa));
7942 * Wait for i/os issued in open context that need to complete
7943 * before this txg syncs.
7945 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
7946 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
7950 * Lock out configuration changes.
7952 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7954 spa->spa_syncing_txg = txg;
7955 spa->spa_sync_pass = 0;
7957 for (int i = 0; i < spa->spa_alloc_count; i++) {
7958 mutex_enter(&spa->spa_alloc_locks[i]);
7959 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
7960 mutex_exit(&spa->spa_alloc_locks[i]);
7964 * If there are any pending vdev state changes, convert them
7965 * into config changes that go out with this transaction group.
7967 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7968 while (list_head(&spa->spa_state_dirty_list) != NULL) {
7970 * We need the write lock here because, for aux vdevs,
7971 * calling vdev_config_dirty() modifies sav_config.
7972 * This is ugly and will become unnecessary when we
7973 * eliminate the aux vdev wart by integrating all vdevs
7974 * into the root vdev tree.
7976 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7977 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
7978 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
7979 vdev_state_clean(vd);
7980 vdev_config_dirty(vd);
7982 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7983 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7985 spa_config_exit(spa, SCL_STATE, FTAG);
7987 tx = dmu_tx_create_assigned(dp, txg);
7989 spa->spa_sync_starttime = gethrtime();
7991 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
7992 spa->spa_sync_starttime + spa->spa_deadman_synctime));
7993 #else /* !illumos */
7995 callout_schedule(&spa->spa_deadman_cycid,
7996 hz * spa->spa_deadman_synctime / NANOSEC);
7998 #endif /* illumos */
8001 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
8002 * set spa_deflate if we have no raid-z vdevs.
8004 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
8005 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
8008 for (i = 0; i < rvd->vdev_children; i++) {
8009 vd = rvd->vdev_child[i];
8010 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
8013 if (i == rvd->vdev_children) {
8014 spa->spa_deflate = TRUE;
8015 VERIFY(0 == zap_add(spa->spa_meta_objset,
8016 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
8017 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
8022 * Set the top-level vdev's max queue depth. Evaluate each
8023 * top-level's async write queue depth in case it changed.
8024 * The max queue depth will not change in the middle of syncing
8027 uint64_t slots_per_allocator = 0;
8028 for (int c = 0; c < rvd->vdev_children; c++) {
8029 vdev_t *tvd = rvd->vdev_child[c];
8030 metaslab_group_t *mg = tvd->vdev_mg;
8032 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
8033 !metaslab_group_initialized(mg))
8037 * It is safe to do a lock-free check here because only async
8038 * allocations look at mg_max_alloc_queue_depth, and async
8039 * allocations all happen from spa_sync().
8041 for (int i = 0; i < spa->spa_alloc_count; i++)
8042 ASSERT0(refcount_count(&(mg->mg_alloc_queue_depth[i])));
8043 mg->mg_max_alloc_queue_depth = max_queue_depth;
8045 for (int i = 0; i < spa->spa_alloc_count; i++) {
8046 mg->mg_cur_max_alloc_queue_depth[i] =
8047 zfs_vdev_def_queue_depth;
8049 slots_per_allocator += zfs_vdev_def_queue_depth;
8051 metaslab_class_t *mc = spa_normal_class(spa);
8052 for (int i = 0; i < spa->spa_alloc_count; i++) {
8053 ASSERT0(refcount_count(&mc->mc_alloc_slots[i]));
8054 mc->mc_alloc_max_slots[i] = slots_per_allocator;
8056 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8058 for (int c = 0; c < rvd->vdev_children; c++) {
8059 vdev_t *vd = rvd->vdev_child[c];
8060 vdev_indirect_state_sync_verify(vd);
8062 if (vdev_indirect_should_condense(vd)) {
8063 spa_condense_indirect_start_sync(vd, tx);
8069 * Iterate to convergence.
8072 int pass = ++spa->spa_sync_pass;
8074 spa_sync_config_object(spa, tx);
8075 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
8076 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
8077 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
8078 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
8079 spa_errlog_sync(spa, txg);
8080 dsl_pool_sync(dp, txg);
8082 if (pass < zfs_sync_pass_deferred_free) {
8083 spa_sync_frees(spa, free_bpl, tx);
8086 * We can not defer frees in pass 1, because
8087 * we sync the deferred frees later in pass 1.
8089 ASSERT3U(pass, >, 1);
8090 bplist_iterate(free_bpl, bpobj_enqueue_cb,
8091 &spa->spa_deferred_bpobj, tx);
8095 dsl_scan_sync(dp, tx);
8097 if (spa->spa_vdev_removal != NULL)
8100 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
8105 spa_sync_upgrades(spa, tx);
8107 spa->spa_uberblock.ub_rootbp.blk_birth);
8109 * Note: We need to check if the MOS is dirty
8110 * because we could have marked the MOS dirty
8111 * without updating the uberblock (e.g. if we
8112 * have sync tasks but no dirty user data). We
8113 * need to check the uberblock's rootbp because
8114 * it is updated if we have synced out dirty
8115 * data (though in this case the MOS will most
8116 * likely also be dirty due to second order
8117 * effects, we don't want to rely on that here).
8119 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
8120 !dmu_objset_is_dirty(mos, txg)) {
8122 * Nothing changed on the first pass,
8123 * therefore this TXG is a no-op. Avoid
8124 * syncing deferred frees, so that we
8125 * can keep this TXG as a no-op.
8127 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
8129 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8130 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
8131 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks,
8135 spa_sync_deferred_frees(spa, tx);
8138 } while (dmu_objset_is_dirty(mos, txg));
8140 if (!list_is_empty(&spa->spa_config_dirty_list)) {
8142 * Make sure that the number of ZAPs for all the vdevs matches
8143 * the number of ZAPs in the per-vdev ZAP list. This only gets
8144 * called if the config is dirty; otherwise there may be
8145 * outstanding AVZ operations that weren't completed in
8146 * spa_sync_config_object.
8148 uint64_t all_vdev_zap_entry_count;
8149 ASSERT0(zap_count(spa->spa_meta_objset,
8150 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
8151 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
8152 all_vdev_zap_entry_count);
8155 if (spa->spa_vdev_removal != NULL) {
8156 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
8160 * Rewrite the vdev configuration (which includes the uberblock)
8161 * to commit the transaction group.
8163 * If there are no dirty vdevs, we sync the uberblock to a few
8164 * random top-level vdevs that are known to be visible in the
8165 * config cache (see spa_vdev_add() for a complete description).
8166 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
8170 * We hold SCL_STATE to prevent vdev open/close/etc.
8171 * while we're attempting to write the vdev labels.
8173 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8175 if (list_is_empty(&spa->spa_config_dirty_list)) {
8176 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
8178 int children = rvd->vdev_children;
8179 int c0 = spa_get_random(children);
8181 for (int c = 0; c < children; c++) {
8182 vd = rvd->vdev_child[(c0 + c) % children];
8184 /* Stop when revisiting the first vdev */
8185 if (c > 0 && svd[0] == vd)
8188 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
8189 !vdev_is_concrete(vd))
8192 svd[svdcount++] = vd;
8193 if (svdcount == SPA_SYNC_MIN_VDEVS)
8196 error = vdev_config_sync(svd, svdcount, txg);
8198 error = vdev_config_sync(rvd->vdev_child,
8199 rvd->vdev_children, txg);
8203 spa->spa_last_synced_guid = rvd->vdev_guid;
8205 spa_config_exit(spa, SCL_STATE, FTAG);
8209 zio_suspend(spa, NULL);
8210 zio_resume_wait(spa);
8215 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
8216 #else /* !illumos */
8218 callout_drain(&spa->spa_deadman_cycid);
8220 #endif /* illumos */
8223 * Clear the dirty config list.
8225 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
8226 vdev_config_clean(vd);
8229 * Now that the new config has synced transactionally,
8230 * let it become visible to the config cache.
8232 if (spa->spa_config_syncing != NULL) {
8233 spa_config_set(spa, spa->spa_config_syncing);
8234 spa->spa_config_txg = txg;
8235 spa->spa_config_syncing = NULL;
8238 dsl_pool_sync_done(dp, txg);
8240 for (int i = 0; i < spa->spa_alloc_count; i++) {
8241 mutex_enter(&spa->spa_alloc_locks[i]);
8242 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
8243 mutex_exit(&spa->spa_alloc_locks[i]);
8247 * Update usable space statistics.
8249 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
8251 vdev_sync_done(vd, txg);
8253 spa_update_dspace(spa);
8256 * It had better be the case that we didn't dirty anything
8257 * since vdev_config_sync().
8259 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8260 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8261 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
8263 while (zfs_pause_spa_sync)
8266 spa->spa_sync_pass = 0;
8269 * Update the last synced uberblock here. We want to do this at
8270 * the end of spa_sync() so that consumers of spa_last_synced_txg()
8271 * will be guaranteed that all the processing associated with
8272 * that txg has been completed.
8274 spa->spa_ubsync = spa->spa_uberblock;
8275 spa_config_exit(spa, SCL_CONFIG, FTAG);
8277 spa_handle_ignored_writes(spa);
8280 * If any async tasks have been requested, kick them off.
8282 spa_async_dispatch(spa);
8283 spa_async_dispatch_vd(spa);
8287 * Sync all pools. We don't want to hold the namespace lock across these
8288 * operations, so we take a reference on the spa_t and drop the lock during the
8292 spa_sync_allpools(void)
8295 mutex_enter(&spa_namespace_lock);
8296 while ((spa = spa_next(spa)) != NULL) {
8297 if (spa_state(spa) != POOL_STATE_ACTIVE ||
8298 !spa_writeable(spa) || spa_suspended(spa))
8300 spa_open_ref(spa, FTAG);
8301 mutex_exit(&spa_namespace_lock);
8302 txg_wait_synced(spa_get_dsl(spa), 0);
8303 mutex_enter(&spa_namespace_lock);
8304 spa_close(spa, FTAG);
8306 mutex_exit(&spa_namespace_lock);
8310 * ==========================================================================
8311 * Miscellaneous routines
8312 * ==========================================================================
8316 * Remove all pools in the system.
8324 * Remove all cached state. All pools should be closed now,
8325 * so every spa in the AVL tree should be unreferenced.
8327 mutex_enter(&spa_namespace_lock);
8328 while ((spa = spa_next(NULL)) != NULL) {
8330 * Stop async tasks. The async thread may need to detach
8331 * a device that's been replaced, which requires grabbing
8332 * spa_namespace_lock, so we must drop it here.
8334 spa_open_ref(spa, FTAG);
8335 mutex_exit(&spa_namespace_lock);
8336 spa_async_suspend(spa);
8337 mutex_enter(&spa_namespace_lock);
8338 spa_close(spa, FTAG);
8340 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
8342 spa_deactivate(spa);
8346 mutex_exit(&spa_namespace_lock);
8350 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
8355 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
8359 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
8360 vd = spa->spa_l2cache.sav_vdevs[i];
8361 if (vd->vdev_guid == guid)
8365 for (i = 0; i < spa->spa_spares.sav_count; i++) {
8366 vd = spa->spa_spares.sav_vdevs[i];
8367 if (vd->vdev_guid == guid)
8376 spa_upgrade(spa_t *spa, uint64_t version)
8378 ASSERT(spa_writeable(spa));
8380 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
8383 * This should only be called for a non-faulted pool, and since a
8384 * future version would result in an unopenable pool, this shouldn't be
8387 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
8388 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
8390 spa->spa_uberblock.ub_version = version;
8391 vdev_config_dirty(spa->spa_root_vdev);
8393 spa_config_exit(spa, SCL_ALL, FTAG);
8395 txg_wait_synced(spa_get_dsl(spa), 0);
8399 spa_has_spare(spa_t *spa, uint64_t guid)
8403 spa_aux_vdev_t *sav = &spa->spa_spares;
8405 for (i = 0; i < sav->sav_count; i++)
8406 if (sav->sav_vdevs[i]->vdev_guid == guid)
8409 for (i = 0; i < sav->sav_npending; i++) {
8410 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
8411 &spareguid) == 0 && spareguid == guid)
8419 * Check if a pool has an active shared spare device.
8420 * Note: reference count of an active spare is 2, as a spare and as a replace
8423 spa_has_active_shared_spare(spa_t *spa)
8427 spa_aux_vdev_t *sav = &spa->spa_spares;
8429 for (i = 0; i < sav->sav_count; i++) {
8430 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
8431 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
8440 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8442 sysevent_t *ev = NULL;
8444 sysevent_attr_list_t *attr = NULL;
8445 sysevent_value_t value;
8447 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
8451 value.value_type = SE_DATA_TYPE_STRING;
8452 value.value.sv_string = spa_name(spa);
8453 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
8456 value.value_type = SE_DATA_TYPE_UINT64;
8457 value.value.sv_uint64 = spa_guid(spa);
8458 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
8462 value.value_type = SE_DATA_TYPE_UINT64;
8463 value.value.sv_uint64 = vd->vdev_guid;
8464 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
8468 if (vd->vdev_path) {
8469 value.value_type = SE_DATA_TYPE_STRING;
8470 value.value.sv_string = vd->vdev_path;
8471 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
8472 &value, SE_SLEEP) != 0)
8477 if (hist_nvl != NULL) {
8478 fnvlist_merge((nvlist_t *)attr, hist_nvl);
8481 if (sysevent_attach_attributes(ev, attr) != 0)
8487 sysevent_free_attr(attr);
8494 spa_event_post(sysevent_t *ev)
8499 (void) log_sysevent(ev, SE_SLEEP, &eid);
8505 spa_event_discard(sysevent_t *ev)
8513 * Post a sysevent corresponding to the given event. The 'name' must be one of
8514 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
8515 * filled in from the spa and (optionally) the vdev and history nvl. This
8516 * doesn't do anything in the userland libzpool, as we don't want consumers to
8517 * misinterpret ztest or zdb as real changes.
8520 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8522 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));