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[(u_int)(sbinuptime() + curcpu) %
1076 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1080 taskq_dispatch_ent(tq, func, arg, flags, ent);
1084 spa_create_zio_taskqs(spa_t *spa)
1086 for (int t = 0; t < ZIO_TYPES; t++) {
1087 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1088 spa_taskqs_init(spa, t, q);
1096 spa_thread(void *arg)
1098 callb_cpr_t cprinfo;
1101 user_t *pu = PTOU(curproc);
1103 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1106 ASSERT(curproc != &p0);
1107 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1108 "zpool-%s", spa->spa_name);
1109 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1112 /* bind this thread to the requested psrset */
1113 if (zio_taskq_psrset_bind != PS_NONE) {
1115 mutex_enter(&cpu_lock);
1116 mutex_enter(&pidlock);
1117 mutex_enter(&curproc->p_lock);
1119 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1120 0, NULL, NULL) == 0) {
1121 curthread->t_bind_pset = zio_taskq_psrset_bind;
1124 "Couldn't bind process for zfs pool \"%s\" to "
1125 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1128 mutex_exit(&curproc->p_lock);
1129 mutex_exit(&pidlock);
1130 mutex_exit(&cpu_lock);
1136 if (zio_taskq_sysdc) {
1137 sysdc_thread_enter(curthread, 100, 0);
1141 spa->spa_proc = curproc;
1142 spa->spa_did = curthread->t_did;
1144 spa_create_zio_taskqs(spa);
1146 mutex_enter(&spa->spa_proc_lock);
1147 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1149 spa->spa_proc_state = SPA_PROC_ACTIVE;
1150 cv_broadcast(&spa->spa_proc_cv);
1152 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1153 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1154 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1155 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1157 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1158 spa->spa_proc_state = SPA_PROC_GONE;
1159 spa->spa_proc = &p0;
1160 cv_broadcast(&spa->spa_proc_cv);
1161 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1163 mutex_enter(&curproc->p_lock);
1166 #endif /* SPA_PROCESS */
1170 * Activate an uninitialized pool.
1173 spa_activate(spa_t *spa, int mode)
1175 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1177 spa->spa_state = POOL_STATE_ACTIVE;
1178 spa->spa_mode = mode;
1180 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1181 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1183 /* Try to create a covering process */
1184 mutex_enter(&spa->spa_proc_lock);
1185 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1186 ASSERT(spa->spa_proc == &p0);
1190 /* Only create a process if we're going to be around a while. */
1191 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1192 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1194 spa->spa_proc_state = SPA_PROC_CREATED;
1195 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1196 cv_wait(&spa->spa_proc_cv,
1197 &spa->spa_proc_lock);
1199 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1200 ASSERT(spa->spa_proc != &p0);
1201 ASSERT(spa->spa_did != 0);
1205 "Couldn't create process for zfs pool \"%s\"\n",
1210 #endif /* SPA_PROCESS */
1211 mutex_exit(&spa->spa_proc_lock);
1213 /* If we didn't create a process, we need to create our taskqs. */
1214 ASSERT(spa->spa_proc == &p0);
1215 if (spa->spa_proc == &p0) {
1216 spa_create_zio_taskqs(spa);
1220 * Start TRIM thread.
1222 trim_thread_create(spa);
1224 for (size_t i = 0; i < TXG_SIZE; i++) {
1225 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1229 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1230 offsetof(vdev_t, vdev_config_dirty_node));
1231 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1232 offsetof(objset_t, os_evicting_node));
1233 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1234 offsetof(vdev_t, vdev_state_dirty_node));
1236 txg_list_create(&spa->spa_vdev_txg_list, spa,
1237 offsetof(struct vdev, vdev_txg_node));
1239 avl_create(&spa->spa_errlist_scrub,
1240 spa_error_entry_compare, sizeof (spa_error_entry_t),
1241 offsetof(spa_error_entry_t, se_avl));
1242 avl_create(&spa->spa_errlist_last,
1243 spa_error_entry_compare, sizeof (spa_error_entry_t),
1244 offsetof(spa_error_entry_t, se_avl));
1248 * Opposite of spa_activate().
1251 spa_deactivate(spa_t *spa)
1253 ASSERT(spa->spa_sync_on == B_FALSE);
1254 ASSERT(spa->spa_dsl_pool == NULL);
1255 ASSERT(spa->spa_root_vdev == NULL);
1256 ASSERT(spa->spa_async_zio_root == NULL);
1257 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1260 * Stop TRIM thread in case spa_unload() wasn't called directly
1261 * before spa_deactivate().
1263 trim_thread_destroy(spa);
1265 spa_evicting_os_wait(spa);
1267 txg_list_destroy(&spa->spa_vdev_txg_list);
1269 list_destroy(&spa->spa_config_dirty_list);
1270 list_destroy(&spa->spa_evicting_os_list);
1271 list_destroy(&spa->spa_state_dirty_list);
1273 for (int t = 0; t < ZIO_TYPES; t++) {
1274 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1275 spa_taskqs_fini(spa, t, q);
1279 for (size_t i = 0; i < TXG_SIZE; i++) {
1280 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1281 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1282 spa->spa_txg_zio[i] = NULL;
1285 metaslab_class_destroy(spa->spa_normal_class);
1286 spa->spa_normal_class = NULL;
1288 metaslab_class_destroy(spa->spa_log_class);
1289 spa->spa_log_class = NULL;
1292 * If this was part of an import or the open otherwise failed, we may
1293 * still have errors left in the queues. Empty them just in case.
1295 spa_errlog_drain(spa);
1297 avl_destroy(&spa->spa_errlist_scrub);
1298 avl_destroy(&spa->spa_errlist_last);
1300 spa->spa_state = POOL_STATE_UNINITIALIZED;
1302 mutex_enter(&spa->spa_proc_lock);
1303 if (spa->spa_proc_state != SPA_PROC_NONE) {
1304 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1305 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1306 cv_broadcast(&spa->spa_proc_cv);
1307 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1308 ASSERT(spa->spa_proc != &p0);
1309 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1311 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1312 spa->spa_proc_state = SPA_PROC_NONE;
1314 ASSERT(spa->spa_proc == &p0);
1315 mutex_exit(&spa->spa_proc_lock);
1319 * We want to make sure spa_thread() has actually exited the ZFS
1320 * module, so that the module can't be unloaded out from underneath
1323 if (spa->spa_did != 0) {
1324 thread_join(spa->spa_did);
1327 #endif /* SPA_PROCESS */
1331 * Verify a pool configuration, and construct the vdev tree appropriately. This
1332 * will create all the necessary vdevs in the appropriate layout, with each vdev
1333 * in the CLOSED state. This will prep the pool before open/creation/import.
1334 * All vdev validation is done by the vdev_alloc() routine.
1337 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1338 uint_t id, int atype)
1344 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1347 if ((*vdp)->vdev_ops->vdev_op_leaf)
1350 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1353 if (error == ENOENT)
1359 return (SET_ERROR(EINVAL));
1362 for (int c = 0; c < children; c++) {
1364 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1372 ASSERT(*vdp != NULL);
1378 * Opposite of spa_load().
1381 spa_unload(spa_t *spa)
1385 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1387 spa_load_note(spa, "UNLOADING");
1392 trim_thread_destroy(spa);
1397 spa_async_suspend(spa);
1399 if (spa->spa_root_vdev) {
1400 vdev_initialize_stop_all(spa->spa_root_vdev,
1401 VDEV_INITIALIZE_ACTIVE);
1407 if (spa->spa_sync_on) {
1408 txg_sync_stop(spa->spa_dsl_pool);
1409 spa->spa_sync_on = B_FALSE;
1413 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1414 * to call it earlier, before we wait for async i/o to complete.
1415 * This ensures that there is no async metaslab prefetching, by
1416 * calling taskq_wait(mg_taskq).
1418 if (spa->spa_root_vdev != NULL) {
1419 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1420 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1421 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1422 spa_config_exit(spa, SCL_ALL, spa);
1426 * Wait for any outstanding async I/O to complete.
1428 if (spa->spa_async_zio_root != NULL) {
1429 for (int i = 0; i < max_ncpus; i++)
1430 (void) zio_wait(spa->spa_async_zio_root[i]);
1431 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1432 spa->spa_async_zio_root = NULL;
1435 if (spa->spa_vdev_removal != NULL) {
1436 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1437 spa->spa_vdev_removal = NULL;
1440 if (spa->spa_condense_zthr != NULL) {
1441 ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1442 zthr_destroy(spa->spa_condense_zthr);
1443 spa->spa_condense_zthr = NULL;
1446 if (spa->spa_checkpoint_discard_zthr != NULL) {
1447 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1448 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1449 spa->spa_checkpoint_discard_zthr = NULL;
1452 spa_condense_fini(spa);
1454 bpobj_close(&spa->spa_deferred_bpobj);
1456 spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1461 if (spa->spa_root_vdev)
1462 vdev_free(spa->spa_root_vdev);
1463 ASSERT(spa->spa_root_vdev == NULL);
1466 * Close the dsl pool.
1468 if (spa->spa_dsl_pool) {
1469 dsl_pool_close(spa->spa_dsl_pool);
1470 spa->spa_dsl_pool = NULL;
1471 spa->spa_meta_objset = NULL;
1477 * Drop and purge level 2 cache
1479 spa_l2cache_drop(spa);
1481 for (i = 0; i < spa->spa_spares.sav_count; i++)
1482 vdev_free(spa->spa_spares.sav_vdevs[i]);
1483 if (spa->spa_spares.sav_vdevs) {
1484 kmem_free(spa->spa_spares.sav_vdevs,
1485 spa->spa_spares.sav_count * sizeof (void *));
1486 spa->spa_spares.sav_vdevs = NULL;
1488 if (spa->spa_spares.sav_config) {
1489 nvlist_free(spa->spa_spares.sav_config);
1490 spa->spa_spares.sav_config = NULL;
1492 spa->spa_spares.sav_count = 0;
1494 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1495 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1496 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1498 if (spa->spa_l2cache.sav_vdevs) {
1499 kmem_free(spa->spa_l2cache.sav_vdevs,
1500 spa->spa_l2cache.sav_count * sizeof (void *));
1501 spa->spa_l2cache.sav_vdevs = NULL;
1503 if (spa->spa_l2cache.sav_config) {
1504 nvlist_free(spa->spa_l2cache.sav_config);
1505 spa->spa_l2cache.sav_config = NULL;
1507 spa->spa_l2cache.sav_count = 0;
1509 spa->spa_async_suspended = 0;
1511 spa->spa_indirect_vdevs_loaded = B_FALSE;
1513 if (spa->spa_comment != NULL) {
1514 spa_strfree(spa->spa_comment);
1515 spa->spa_comment = NULL;
1518 spa_config_exit(spa, SCL_ALL, spa);
1522 * Load (or re-load) the current list of vdevs describing the active spares for
1523 * this pool. When this is called, we have some form of basic information in
1524 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1525 * then re-generate a more complete list including status information.
1528 spa_load_spares(spa_t *spa)
1537 * zdb opens both the current state of the pool and the
1538 * checkpointed state (if present), with a different spa_t.
1540 * As spare vdevs are shared among open pools, we skip loading
1541 * them when we load the checkpointed state of the pool.
1543 if (!spa_writeable(spa))
1547 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1550 * First, close and free any existing spare vdevs.
1552 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1553 vd = spa->spa_spares.sav_vdevs[i];
1555 /* Undo the call to spa_activate() below */
1556 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1557 B_FALSE)) != NULL && tvd->vdev_isspare)
1558 spa_spare_remove(tvd);
1563 if (spa->spa_spares.sav_vdevs)
1564 kmem_free(spa->spa_spares.sav_vdevs,
1565 spa->spa_spares.sav_count * sizeof (void *));
1567 if (spa->spa_spares.sav_config == NULL)
1570 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1571 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1573 spa->spa_spares.sav_count = (int)nspares;
1574 spa->spa_spares.sav_vdevs = NULL;
1580 * Construct the array of vdevs, opening them to get status in the
1581 * process. For each spare, there is potentially two different vdev_t
1582 * structures associated with it: one in the list of spares (used only
1583 * for basic validation purposes) and one in the active vdev
1584 * configuration (if it's spared in). During this phase we open and
1585 * validate each vdev on the spare list. If the vdev also exists in the
1586 * active configuration, then we also mark this vdev as an active spare.
1588 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1590 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1591 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1592 VDEV_ALLOC_SPARE) == 0);
1595 spa->spa_spares.sav_vdevs[i] = vd;
1597 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1598 B_FALSE)) != NULL) {
1599 if (!tvd->vdev_isspare)
1603 * We only mark the spare active if we were successfully
1604 * able to load the vdev. Otherwise, importing a pool
1605 * with a bad active spare would result in strange
1606 * behavior, because multiple pool would think the spare
1607 * is actively in use.
1609 * There is a vulnerability here to an equally bizarre
1610 * circumstance, where a dead active spare is later
1611 * brought back to life (onlined or otherwise). Given
1612 * the rarity of this scenario, and the extra complexity
1613 * it adds, we ignore the possibility.
1615 if (!vdev_is_dead(tvd))
1616 spa_spare_activate(tvd);
1620 vd->vdev_aux = &spa->spa_spares;
1622 if (vdev_open(vd) != 0)
1625 if (vdev_validate_aux(vd) == 0)
1630 * Recompute the stashed list of spares, with status information
1633 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1634 DATA_TYPE_NVLIST_ARRAY) == 0);
1636 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1638 for (i = 0; i < spa->spa_spares.sav_count; i++)
1639 spares[i] = vdev_config_generate(spa,
1640 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1641 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1642 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1643 for (i = 0; i < spa->spa_spares.sav_count; i++)
1644 nvlist_free(spares[i]);
1645 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1649 * Load (or re-load) the current list of vdevs describing the active l2cache for
1650 * this pool. When this is called, we have some form of basic information in
1651 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1652 * then re-generate a more complete list including status information.
1653 * Devices which are already active have their details maintained, and are
1657 spa_load_l2cache(spa_t *spa)
1661 int i, j, oldnvdevs;
1663 vdev_t *vd, **oldvdevs, **newvdevs;
1664 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1668 * zdb opens both the current state of the pool and the
1669 * checkpointed state (if present), with a different spa_t.
1671 * As L2 caches are part of the ARC which is shared among open
1672 * pools, we skip loading them when we load the checkpointed
1673 * state of the pool.
1675 if (!spa_writeable(spa))
1679 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1681 if (sav->sav_config != NULL) {
1682 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1683 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1684 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1690 oldvdevs = sav->sav_vdevs;
1691 oldnvdevs = sav->sav_count;
1692 sav->sav_vdevs = NULL;
1696 * Process new nvlist of vdevs.
1698 for (i = 0; i < nl2cache; i++) {
1699 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1703 for (j = 0; j < oldnvdevs; j++) {
1705 if (vd != NULL && guid == vd->vdev_guid) {
1707 * Retain previous vdev for add/remove ops.
1715 if (newvdevs[i] == NULL) {
1719 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1720 VDEV_ALLOC_L2CACHE) == 0);
1725 * Commit this vdev as an l2cache device,
1726 * even if it fails to open.
1728 spa_l2cache_add(vd);
1733 spa_l2cache_activate(vd);
1735 if (vdev_open(vd) != 0)
1738 (void) vdev_validate_aux(vd);
1740 if (!vdev_is_dead(vd))
1741 l2arc_add_vdev(spa, vd);
1746 * Purge vdevs that were dropped
1748 for (i = 0; i < oldnvdevs; i++) {
1753 ASSERT(vd->vdev_isl2cache);
1755 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1756 pool != 0ULL && l2arc_vdev_present(vd))
1757 l2arc_remove_vdev(vd);
1758 vdev_clear_stats(vd);
1764 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1766 if (sav->sav_config == NULL)
1769 sav->sav_vdevs = newvdevs;
1770 sav->sav_count = (int)nl2cache;
1773 * Recompute the stashed list of l2cache devices, with status
1774 * information this time.
1776 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1777 DATA_TYPE_NVLIST_ARRAY) == 0);
1779 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1780 for (i = 0; i < sav->sav_count; i++)
1781 l2cache[i] = vdev_config_generate(spa,
1782 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1783 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1784 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1786 for (i = 0; i < sav->sav_count; i++)
1787 nvlist_free(l2cache[i]);
1789 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1793 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1796 char *packed = NULL;
1801 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1805 nvsize = *(uint64_t *)db->db_data;
1806 dmu_buf_rele(db, FTAG);
1808 packed = kmem_alloc(nvsize, KM_SLEEP);
1809 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1812 error = nvlist_unpack(packed, nvsize, value, 0);
1813 kmem_free(packed, nvsize);
1819 * Concrete top-level vdevs that are not missing and are not logs. At every
1820 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1823 spa_healthy_core_tvds(spa_t *spa)
1825 vdev_t *rvd = spa->spa_root_vdev;
1828 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1829 vdev_t *vd = rvd->vdev_child[i];
1832 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1840 * Checks to see if the given vdev could not be opened, in which case we post a
1841 * sysevent to notify the autoreplace code that the device has been removed.
1844 spa_check_removed(vdev_t *vd)
1846 for (uint64_t c = 0; c < vd->vdev_children; c++)
1847 spa_check_removed(vd->vdev_child[c]);
1849 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1850 vdev_is_concrete(vd)) {
1851 zfs_post_autoreplace(vd->vdev_spa, vd);
1852 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1857 spa_check_for_missing_logs(spa_t *spa)
1859 vdev_t *rvd = spa->spa_root_vdev;
1862 * If we're doing a normal import, then build up any additional
1863 * diagnostic information about missing log devices.
1864 * We'll pass this up to the user for further processing.
1866 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1867 nvlist_t **child, *nv;
1870 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1872 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1874 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1875 vdev_t *tvd = rvd->vdev_child[c];
1878 * We consider a device as missing only if it failed
1879 * to open (i.e. offline or faulted is not considered
1882 if (tvd->vdev_islog &&
1883 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1884 child[idx++] = vdev_config_generate(spa, tvd,
1885 B_FALSE, VDEV_CONFIG_MISSING);
1890 fnvlist_add_nvlist_array(nv,
1891 ZPOOL_CONFIG_CHILDREN, child, idx);
1892 fnvlist_add_nvlist(spa->spa_load_info,
1893 ZPOOL_CONFIG_MISSING_DEVICES, nv);
1895 for (uint64_t i = 0; i < idx; i++)
1896 nvlist_free(child[i]);
1899 kmem_free(child, rvd->vdev_children * sizeof (char **));
1902 spa_load_failed(spa, "some log devices are missing");
1903 vdev_dbgmsg_print_tree(rvd, 2);
1904 return (SET_ERROR(ENXIO));
1907 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1908 vdev_t *tvd = rvd->vdev_child[c];
1910 if (tvd->vdev_islog &&
1911 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1912 spa_set_log_state(spa, SPA_LOG_CLEAR);
1913 spa_load_note(spa, "some log devices are "
1914 "missing, ZIL is dropped.");
1915 vdev_dbgmsg_print_tree(rvd, 2);
1925 * Check for missing log devices
1928 spa_check_logs(spa_t *spa)
1930 boolean_t rv = B_FALSE;
1931 dsl_pool_t *dp = spa_get_dsl(spa);
1933 switch (spa->spa_log_state) {
1934 case SPA_LOG_MISSING:
1935 /* need to recheck in case slog has been restored */
1936 case SPA_LOG_UNKNOWN:
1937 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1938 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1940 spa_set_log_state(spa, SPA_LOG_MISSING);
1947 spa_passivate_log(spa_t *spa)
1949 vdev_t *rvd = spa->spa_root_vdev;
1950 boolean_t slog_found = B_FALSE;
1952 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1954 if (!spa_has_slogs(spa))
1957 for (int c = 0; c < rvd->vdev_children; c++) {
1958 vdev_t *tvd = rvd->vdev_child[c];
1959 metaslab_group_t *mg = tvd->vdev_mg;
1961 if (tvd->vdev_islog) {
1962 metaslab_group_passivate(mg);
1963 slog_found = B_TRUE;
1967 return (slog_found);
1971 spa_activate_log(spa_t *spa)
1973 vdev_t *rvd = spa->spa_root_vdev;
1975 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1977 for (int c = 0; c < rvd->vdev_children; c++) {
1978 vdev_t *tvd = rvd->vdev_child[c];
1979 metaslab_group_t *mg = tvd->vdev_mg;
1981 if (tvd->vdev_islog)
1982 metaslab_group_activate(mg);
1987 spa_reset_logs(spa_t *spa)
1991 error = dmu_objset_find(spa_name(spa), zil_reset,
1992 NULL, DS_FIND_CHILDREN);
1995 * We successfully offlined the log device, sync out the
1996 * current txg so that the "stubby" block can be removed
1999 txg_wait_synced(spa->spa_dsl_pool, 0);
2005 spa_aux_check_removed(spa_aux_vdev_t *sav)
2009 for (i = 0; i < sav->sav_count; i++)
2010 spa_check_removed(sav->sav_vdevs[i]);
2014 spa_claim_notify(zio_t *zio)
2016 spa_t *spa = zio->io_spa;
2021 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2022 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2023 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2024 mutex_exit(&spa->spa_props_lock);
2027 typedef struct spa_load_error {
2028 uint64_t sle_meta_count;
2029 uint64_t sle_data_count;
2033 spa_load_verify_done(zio_t *zio)
2035 blkptr_t *bp = zio->io_bp;
2036 spa_load_error_t *sle = zio->io_private;
2037 dmu_object_type_t type = BP_GET_TYPE(bp);
2038 int error = zio->io_error;
2039 spa_t *spa = zio->io_spa;
2041 abd_free(zio->io_abd);
2043 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2044 type != DMU_OT_INTENT_LOG)
2045 atomic_inc_64(&sle->sle_meta_count);
2047 atomic_inc_64(&sle->sle_data_count);
2050 mutex_enter(&spa->spa_scrub_lock);
2051 spa->spa_load_verify_ios--;
2052 cv_broadcast(&spa->spa_scrub_io_cv);
2053 mutex_exit(&spa->spa_scrub_lock);
2057 * Maximum number of concurrent scrub i/os to create while verifying
2058 * a pool while importing it.
2060 int spa_load_verify_maxinflight = 10000;
2061 boolean_t spa_load_verify_metadata = B_TRUE;
2062 boolean_t spa_load_verify_data = B_TRUE;
2064 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
2065 &spa_load_verify_maxinflight, 0,
2066 "Maximum number of concurrent scrub I/Os to create while verifying a "
2067 "pool while importing it");
2069 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
2070 &spa_load_verify_metadata, 0,
2071 "Check metadata on import?");
2073 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
2074 &spa_load_verify_data, 0,
2075 "Check user data on import?");
2079 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2080 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2082 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2085 * Note: normally this routine will not be called if
2086 * spa_load_verify_metadata is not set. However, it may be useful
2087 * to manually set the flag after the traversal has begun.
2089 if (!spa_load_verify_metadata)
2091 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2095 size_t size = BP_GET_PSIZE(bp);
2097 mutex_enter(&spa->spa_scrub_lock);
2098 while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight)
2099 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2100 spa->spa_load_verify_ios++;
2101 mutex_exit(&spa->spa_scrub_lock);
2103 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2104 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2105 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2106 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2112 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2114 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2115 return (SET_ERROR(ENAMETOOLONG));
2121 spa_load_verify(spa_t *spa)
2124 spa_load_error_t sle = { 0 };
2125 zpool_load_policy_t policy;
2126 boolean_t verify_ok = B_FALSE;
2129 zpool_get_load_policy(spa->spa_config, &policy);
2131 if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2134 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2135 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2136 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2138 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2142 rio = zio_root(spa, NULL, &sle,
2143 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2145 if (spa_load_verify_metadata) {
2146 if (spa->spa_extreme_rewind) {
2147 spa_load_note(spa, "performing a complete scan of the "
2148 "pool since extreme rewind is on. This may take "
2149 "a very long time.\n (spa_load_verify_data=%u, "
2150 "spa_load_verify_metadata=%u)",
2151 spa_load_verify_data, spa_load_verify_metadata);
2153 error = traverse_pool(spa, spa->spa_verify_min_txg,
2154 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2155 spa_load_verify_cb, rio);
2158 (void) zio_wait(rio);
2160 spa->spa_load_meta_errors = sle.sle_meta_count;
2161 spa->spa_load_data_errors = sle.sle_data_count;
2163 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2164 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2165 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2166 (u_longlong_t)sle.sle_data_count);
2169 if (spa_load_verify_dryrun ||
2170 (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2171 sle.sle_data_count <= policy.zlp_maxdata)) {
2175 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2176 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2178 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2179 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2180 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2181 VERIFY(nvlist_add_int64(spa->spa_load_info,
2182 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2183 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2184 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2186 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2189 if (spa_load_verify_dryrun)
2193 if (error != ENXIO && error != EIO)
2194 error = SET_ERROR(EIO);
2198 return (verify_ok ? 0 : EIO);
2202 * Find a value in the pool props object.
2205 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2207 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2208 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2212 * Find a value in the pool directory object.
2215 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2217 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2218 name, sizeof (uint64_t), 1, val);
2220 if (error != 0 && (error != ENOENT || log_enoent)) {
2221 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2222 "[error=%d]", name, error);
2229 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2231 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2232 return (SET_ERROR(err));
2236 spa_spawn_aux_threads(spa_t *spa)
2238 ASSERT(spa_writeable(spa));
2240 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2242 spa_start_indirect_condensing_thread(spa);
2244 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2245 spa->spa_checkpoint_discard_zthr =
2246 zthr_create(spa_checkpoint_discard_thread_check,
2247 spa_checkpoint_discard_thread, spa);
2251 * Fix up config after a partly-completed split. This is done with the
2252 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2253 * pool have that entry in their config, but only the splitting one contains
2254 * a list of all the guids of the vdevs that are being split off.
2256 * This function determines what to do with that list: either rejoin
2257 * all the disks to the pool, or complete the splitting process. To attempt
2258 * the rejoin, each disk that is offlined is marked online again, and
2259 * we do a reopen() call. If the vdev label for every disk that was
2260 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2261 * then we call vdev_split() on each disk, and complete the split.
2263 * Otherwise we leave the config alone, with all the vdevs in place in
2264 * the original pool.
2267 spa_try_repair(spa_t *spa, nvlist_t *config)
2274 boolean_t attempt_reopen;
2276 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2279 /* check that the config is complete */
2280 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2281 &glist, &gcount) != 0)
2284 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2286 /* attempt to online all the vdevs & validate */
2287 attempt_reopen = B_TRUE;
2288 for (i = 0; i < gcount; i++) {
2289 if (glist[i] == 0) /* vdev is hole */
2292 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2293 if (vd[i] == NULL) {
2295 * Don't bother attempting to reopen the disks;
2296 * just do the split.
2298 attempt_reopen = B_FALSE;
2300 /* attempt to re-online it */
2301 vd[i]->vdev_offline = B_FALSE;
2305 if (attempt_reopen) {
2306 vdev_reopen(spa->spa_root_vdev);
2308 /* check each device to see what state it's in */
2309 for (extracted = 0, i = 0; i < gcount; i++) {
2310 if (vd[i] != NULL &&
2311 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2318 * If every disk has been moved to the new pool, or if we never
2319 * even attempted to look at them, then we split them off for
2322 if (!attempt_reopen || gcount == extracted) {
2323 for (i = 0; i < gcount; i++)
2326 vdev_reopen(spa->spa_root_vdev);
2329 kmem_free(vd, gcount * sizeof (vdev_t *));
2333 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2335 char *ereport = FM_EREPORT_ZFS_POOL;
2338 spa->spa_load_state = state;
2340 gethrestime(&spa->spa_loaded_ts);
2341 error = spa_load_impl(spa, type, &ereport);
2344 * Don't count references from objsets that are already closed
2345 * and are making their way through the eviction process.
2347 spa_evicting_os_wait(spa);
2348 spa->spa_minref = refcount_count(&spa->spa_refcount);
2350 if (error != EEXIST) {
2351 spa->spa_loaded_ts.tv_sec = 0;
2352 spa->spa_loaded_ts.tv_nsec = 0;
2354 if (error != EBADF) {
2355 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2358 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2365 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2366 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2367 * spa's per-vdev ZAP list.
2370 vdev_count_verify_zaps(vdev_t *vd)
2372 spa_t *spa = vd->vdev_spa;
2374 if (vd->vdev_top_zap != 0) {
2376 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2377 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2379 if (vd->vdev_leaf_zap != 0) {
2381 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2382 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2385 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2386 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2393 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2397 uint64_t myhostid = 0;
2399 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2400 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2401 hostname = fnvlist_lookup_string(mos_config,
2402 ZPOOL_CONFIG_HOSTNAME);
2404 myhostid = zone_get_hostid(NULL);
2406 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2407 cmn_err(CE_WARN, "pool '%s' could not be "
2408 "loaded as it was last accessed by "
2409 "another system (host: %s hostid: 0x%llx). "
2410 "See: http://illumos.org/msg/ZFS-8000-EY",
2411 spa_name(spa), hostname, (u_longlong_t)hostid);
2412 spa_load_failed(spa, "hostid verification failed: pool "
2413 "last accessed by host: %s (hostid: 0x%llx)",
2414 hostname, (u_longlong_t)hostid);
2415 return (SET_ERROR(EBADF));
2423 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2426 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2433 * Versioning wasn't explicitly added to the label until later, so if
2434 * it's not present treat it as the initial version.
2436 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2437 &spa->spa_ubsync.ub_version) != 0)
2438 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2440 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2441 spa_load_failed(spa, "invalid config provided: '%s' missing",
2442 ZPOOL_CONFIG_POOL_GUID);
2443 return (SET_ERROR(EINVAL));
2447 * If we are doing an import, ensure that the pool is not already
2448 * imported by checking if its pool guid already exists in the
2451 * The only case that we allow an already imported pool to be
2452 * imported again, is when the pool is checkpointed and we want to
2453 * look at its checkpointed state from userland tools like zdb.
2456 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2457 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2458 spa_guid_exists(pool_guid, 0)) {
2460 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2461 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2462 spa_guid_exists(pool_guid, 0) &&
2463 !spa_importing_readonly_checkpoint(spa)) {
2465 spa_load_failed(spa, "a pool with guid %llu is already open",
2466 (u_longlong_t)pool_guid);
2467 return (SET_ERROR(EEXIST));
2470 spa->spa_config_guid = pool_guid;
2472 nvlist_free(spa->spa_load_info);
2473 spa->spa_load_info = fnvlist_alloc();
2475 ASSERT(spa->spa_comment == NULL);
2476 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2477 spa->spa_comment = spa_strdup(comment);
2479 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2480 &spa->spa_config_txg);
2482 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2483 spa->spa_config_splitting = fnvlist_dup(nvl);
2485 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2486 spa_load_failed(spa, "invalid config provided: '%s' missing",
2487 ZPOOL_CONFIG_VDEV_TREE);
2488 return (SET_ERROR(EINVAL));
2492 * Create "The Godfather" zio to hold all async IOs
2494 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2496 for (int i = 0; i < max_ncpus; i++) {
2497 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2498 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2499 ZIO_FLAG_GODFATHER);
2503 * Parse the configuration into a vdev tree. We explicitly set the
2504 * value that will be returned by spa_version() since parsing the
2505 * configuration requires knowing the version number.
2507 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2508 parse = (type == SPA_IMPORT_EXISTING ?
2509 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2510 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2511 spa_config_exit(spa, SCL_ALL, FTAG);
2514 spa_load_failed(spa, "unable to parse config [error=%d]",
2519 ASSERT(spa->spa_root_vdev == rvd);
2520 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2521 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2523 if (type != SPA_IMPORT_ASSEMBLE) {
2524 ASSERT(spa_guid(spa) == pool_guid);
2531 * Recursively open all vdevs in the vdev tree. This function is called twice:
2532 * first with the untrusted config, then with the trusted config.
2535 spa_ld_open_vdevs(spa_t *spa)
2540 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2541 * missing/unopenable for the root vdev to be still considered openable.
2543 if (spa->spa_trust_config) {
2544 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2545 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2546 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2547 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2548 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2550 spa->spa_missing_tvds_allowed = 0;
2553 spa->spa_missing_tvds_allowed =
2554 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2556 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2557 error = vdev_open(spa->spa_root_vdev);
2558 spa_config_exit(spa, SCL_ALL, FTAG);
2560 if (spa->spa_missing_tvds != 0) {
2561 spa_load_note(spa, "vdev tree has %lld missing top-level "
2562 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2563 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2565 * Although theoretically we could allow users to open
2566 * incomplete pools in RW mode, we'd need to add a lot
2567 * of extra logic (e.g. adjust pool space to account
2568 * for missing vdevs).
2569 * This limitation also prevents users from accidentally
2570 * opening the pool in RW mode during data recovery and
2571 * damaging it further.
2573 spa_load_note(spa, "pools with missing top-level "
2574 "vdevs can only be opened in read-only mode.");
2575 error = SET_ERROR(ENXIO);
2577 spa_load_note(spa, "current settings allow for maximum "
2578 "%lld missing top-level vdevs at this stage.",
2579 (u_longlong_t)spa->spa_missing_tvds_allowed);
2583 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2586 if (spa->spa_missing_tvds != 0 || error != 0)
2587 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2593 * We need to validate the vdev labels against the configuration that
2594 * we have in hand. This function is called twice: first with an untrusted
2595 * config, then with a trusted config. The validation is more strict when the
2596 * config is trusted.
2599 spa_ld_validate_vdevs(spa_t *spa)
2602 vdev_t *rvd = spa->spa_root_vdev;
2604 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2605 error = vdev_validate(rvd);
2606 spa_config_exit(spa, SCL_ALL, FTAG);
2609 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2613 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2614 spa_load_failed(spa, "cannot open vdev tree after invalidating "
2616 vdev_dbgmsg_print_tree(rvd, 2);
2617 return (SET_ERROR(ENXIO));
2624 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2626 spa->spa_state = POOL_STATE_ACTIVE;
2627 spa->spa_ubsync = spa->spa_uberblock;
2628 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2629 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2630 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2631 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2632 spa->spa_claim_max_txg = spa->spa_first_txg;
2633 spa->spa_prev_software_version = ub->ub_software_version;
2637 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2639 vdev_t *rvd = spa->spa_root_vdev;
2641 uberblock_t *ub = &spa->spa_uberblock;
2644 * If we are opening the checkpointed state of the pool by
2645 * rewinding to it, at this point we will have written the
2646 * checkpointed uberblock to the vdev labels, so searching
2647 * the labels will find the right uberblock. However, if
2648 * we are opening the checkpointed state read-only, we have
2649 * not modified the labels. Therefore, we must ignore the
2650 * labels and continue using the spa_uberblock that was set
2651 * by spa_ld_checkpoint_rewind.
2653 * Note that it would be fine to ignore the labels when
2654 * rewinding (opening writeable) as well. However, if we
2655 * crash just after writing the labels, we will end up
2656 * searching the labels. Doing so in the common case means
2657 * that this code path gets exercised normally, rather than
2658 * just in the edge case.
2660 if (ub->ub_checkpoint_txg != 0 &&
2661 spa_importing_readonly_checkpoint(spa)) {
2662 spa_ld_select_uberblock_done(spa, ub);
2667 * Find the best uberblock.
2669 vdev_uberblock_load(rvd, ub, &label);
2672 * If we weren't able to find a single valid uberblock, return failure.
2674 if (ub->ub_txg == 0) {
2676 spa_load_failed(spa, "no valid uberblock found");
2677 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2680 spa_load_note(spa, "using uberblock with txg=%llu",
2681 (u_longlong_t)ub->ub_txg);
2684 * If the pool has an unsupported version we can't open it.
2686 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2688 spa_load_failed(spa, "version %llu is not supported",
2689 (u_longlong_t)ub->ub_version);
2690 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2693 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2697 * If we weren't able to find what's necessary for reading the
2698 * MOS in the label, return failure.
2700 if (label == NULL) {
2701 spa_load_failed(spa, "label config unavailable");
2702 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2706 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2709 spa_load_failed(spa, "invalid label: '%s' missing",
2710 ZPOOL_CONFIG_FEATURES_FOR_READ);
2711 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2716 * Update our in-core representation with the definitive values
2719 nvlist_free(spa->spa_label_features);
2720 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2726 * Look through entries in the label nvlist's features_for_read. If
2727 * there is a feature listed there which we don't understand then we
2728 * cannot open a pool.
2730 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2731 nvlist_t *unsup_feat;
2733 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2736 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2738 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2739 if (!zfeature_is_supported(nvpair_name(nvp))) {
2740 VERIFY(nvlist_add_string(unsup_feat,
2741 nvpair_name(nvp), "") == 0);
2745 if (!nvlist_empty(unsup_feat)) {
2746 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2747 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2748 nvlist_free(unsup_feat);
2749 spa_load_failed(spa, "some features are unsupported");
2750 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2754 nvlist_free(unsup_feat);
2757 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2758 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2759 spa_try_repair(spa, spa->spa_config);
2760 spa_config_exit(spa, SCL_ALL, FTAG);
2761 nvlist_free(spa->spa_config_splitting);
2762 spa->spa_config_splitting = NULL;
2766 * Initialize internal SPA structures.
2768 spa_ld_select_uberblock_done(spa, ub);
2774 spa_ld_open_rootbp(spa_t *spa)
2777 vdev_t *rvd = spa->spa_root_vdev;
2779 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2781 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2782 "[error=%d]", error);
2783 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2785 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2791 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2792 boolean_t reloading)
2794 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2795 nvlist_t *nv, *mos_config, *policy;
2796 int error = 0, copy_error;
2797 uint64_t healthy_tvds, healthy_tvds_mos;
2798 uint64_t mos_config_txg;
2800 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2802 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2805 * If we're assembling a pool from a split, the config provided is
2806 * already trusted so there is nothing to do.
2808 if (type == SPA_IMPORT_ASSEMBLE)
2811 healthy_tvds = spa_healthy_core_tvds(spa);
2813 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2815 spa_load_failed(spa, "unable to retrieve MOS config");
2816 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2820 * If we are doing an open, pool owner wasn't verified yet, thus do
2821 * the verification here.
2823 if (spa->spa_load_state == SPA_LOAD_OPEN) {
2824 error = spa_verify_host(spa, mos_config);
2826 nvlist_free(mos_config);
2831 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2833 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2836 * Build a new vdev tree from the trusted config
2838 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2841 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2842 * obtained by scanning /dev/dsk, then it will have the right vdev
2843 * paths. We update the trusted MOS config with this information.
2844 * We first try to copy the paths with vdev_copy_path_strict, which
2845 * succeeds only when both configs have exactly the same vdev tree.
2846 * If that fails, we fall back to a more flexible method that has a
2847 * best effort policy.
2849 copy_error = vdev_copy_path_strict(rvd, mrvd);
2850 if (copy_error != 0 || spa_load_print_vdev_tree) {
2851 spa_load_note(spa, "provided vdev tree:");
2852 vdev_dbgmsg_print_tree(rvd, 2);
2853 spa_load_note(spa, "MOS vdev tree:");
2854 vdev_dbgmsg_print_tree(mrvd, 2);
2856 if (copy_error != 0) {
2857 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2858 "back to vdev_copy_path_relaxed");
2859 vdev_copy_path_relaxed(rvd, mrvd);
2864 spa->spa_root_vdev = mrvd;
2866 spa_config_exit(spa, SCL_ALL, FTAG);
2869 * We will use spa_config if we decide to reload the spa or if spa_load
2870 * fails and we rewind. We must thus regenerate the config using the
2871 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
2872 * pass settings on how to load the pool and is not stored in the MOS.
2873 * We copy it over to our new, trusted config.
2875 mos_config_txg = fnvlist_lookup_uint64(mos_config,
2876 ZPOOL_CONFIG_POOL_TXG);
2877 nvlist_free(mos_config);
2878 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
2879 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
2881 fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
2882 spa_config_set(spa, mos_config);
2883 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2886 * Now that we got the config from the MOS, we should be more strict
2887 * in checking blkptrs and can make assumptions about the consistency
2888 * of the vdev tree. spa_trust_config must be set to true before opening
2889 * vdevs in order for them to be writeable.
2891 spa->spa_trust_config = B_TRUE;
2894 * Open and validate the new vdev tree
2896 error = spa_ld_open_vdevs(spa);
2900 error = spa_ld_validate_vdevs(spa);
2904 if (copy_error != 0 || spa_load_print_vdev_tree) {
2905 spa_load_note(spa, "final vdev tree:");
2906 vdev_dbgmsg_print_tree(rvd, 2);
2909 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2910 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2912 * Sanity check to make sure that we are indeed loading the
2913 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2914 * in the config provided and they happened to be the only ones
2915 * to have the latest uberblock, we could involuntarily perform
2916 * an extreme rewind.
2918 healthy_tvds_mos = spa_healthy_core_tvds(spa);
2919 if (healthy_tvds_mos - healthy_tvds >=
2920 SPA_SYNC_MIN_VDEVS) {
2921 spa_load_note(spa, "config provided misses too many "
2922 "top-level vdevs compared to MOS (%lld vs %lld). ",
2923 (u_longlong_t)healthy_tvds,
2924 (u_longlong_t)healthy_tvds_mos);
2925 spa_load_note(spa, "vdev tree:");
2926 vdev_dbgmsg_print_tree(rvd, 2);
2928 spa_load_failed(spa, "config was already "
2929 "provided from MOS. Aborting.");
2930 return (spa_vdev_err(rvd,
2931 VDEV_AUX_CORRUPT_DATA, EIO));
2933 spa_load_note(spa, "spa must be reloaded using MOS "
2935 return (SET_ERROR(EAGAIN));
2939 error = spa_check_for_missing_logs(spa);
2941 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2943 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2944 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2945 "guid sum (%llu != %llu)",
2946 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2947 (u_longlong_t)rvd->vdev_guid_sum);
2948 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2956 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2959 vdev_t *rvd = spa->spa_root_vdev;
2962 * Everything that we read before spa_remove_init() must be stored
2963 * on concreted vdevs. Therefore we do this as early as possible.
2965 error = spa_remove_init(spa);
2967 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2969 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2973 * Retrieve information needed to condense indirect vdev mappings.
2975 error = spa_condense_init(spa);
2977 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2979 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2986 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2989 vdev_t *rvd = spa->spa_root_vdev;
2991 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2992 boolean_t missing_feat_read = B_FALSE;
2993 nvlist_t *unsup_feat, *enabled_feat;
2995 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2996 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2997 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3000 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
3001 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
3002 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3005 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
3006 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
3007 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3010 enabled_feat = fnvlist_alloc();
3011 unsup_feat = fnvlist_alloc();
3013 if (!spa_features_check(spa, B_FALSE,
3014 unsup_feat, enabled_feat))
3015 missing_feat_read = B_TRUE;
3017 if (spa_writeable(spa) ||
3018 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3019 if (!spa_features_check(spa, B_TRUE,
3020 unsup_feat, enabled_feat)) {
3021 *missing_feat_writep = B_TRUE;
3025 fnvlist_add_nvlist(spa->spa_load_info,
3026 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3028 if (!nvlist_empty(unsup_feat)) {
3029 fnvlist_add_nvlist(spa->spa_load_info,
3030 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3033 fnvlist_free(enabled_feat);
3034 fnvlist_free(unsup_feat);
3036 if (!missing_feat_read) {
3037 fnvlist_add_boolean(spa->spa_load_info,
3038 ZPOOL_CONFIG_CAN_RDONLY);
3042 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3043 * twofold: to determine whether the pool is available for
3044 * import in read-write mode and (if it is not) whether the
3045 * pool is available for import in read-only mode. If the pool
3046 * is available for import in read-write mode, it is displayed
3047 * as available in userland; if it is not available for import
3048 * in read-only mode, it is displayed as unavailable in
3049 * userland. If the pool is available for import in read-only
3050 * mode but not read-write mode, it is displayed as unavailable
3051 * in userland with a special note that the pool is actually
3052 * available for open in read-only mode.
3054 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3055 * missing a feature for write, we must first determine whether
3056 * the pool can be opened read-only before returning to
3057 * userland in order to know whether to display the
3058 * abovementioned note.
3060 if (missing_feat_read || (*missing_feat_writep &&
3061 spa_writeable(spa))) {
3062 spa_load_failed(spa, "pool uses unsupported features");
3063 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3068 * Load refcounts for ZFS features from disk into an in-memory
3069 * cache during SPA initialization.
3071 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3074 error = feature_get_refcount_from_disk(spa,
3075 &spa_feature_table[i], &refcount);
3077 spa->spa_feat_refcount_cache[i] = refcount;
3078 } else if (error == ENOTSUP) {
3079 spa->spa_feat_refcount_cache[i] =
3080 SPA_FEATURE_DISABLED;
3082 spa_load_failed(spa, "error getting refcount "
3083 "for feature %s [error=%d]",
3084 spa_feature_table[i].fi_guid, error);
3085 return (spa_vdev_err(rvd,
3086 VDEV_AUX_CORRUPT_DATA, EIO));
3091 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3092 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3093 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3094 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3101 spa_ld_load_special_directories(spa_t *spa)
3104 vdev_t *rvd = spa->spa_root_vdev;
3106 spa->spa_is_initializing = B_TRUE;
3107 error = dsl_pool_open(spa->spa_dsl_pool);
3108 spa->spa_is_initializing = B_FALSE;
3110 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3111 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3118 spa_ld_get_props(spa_t *spa)
3122 vdev_t *rvd = spa->spa_root_vdev;
3124 /* Grab the secret checksum salt from the MOS. */
3125 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3126 DMU_POOL_CHECKSUM_SALT, 1,
3127 sizeof (spa->spa_cksum_salt.zcs_bytes),
3128 spa->spa_cksum_salt.zcs_bytes);
3129 if (error == ENOENT) {
3130 /* Generate a new salt for subsequent use */
3131 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3132 sizeof (spa->spa_cksum_salt.zcs_bytes));
3133 } else if (error != 0) {
3134 spa_load_failed(spa, "unable to retrieve checksum salt from "
3135 "MOS [error=%d]", error);
3136 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3139 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3140 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3141 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3143 spa_load_failed(spa, "error opening deferred-frees bpobj "
3144 "[error=%d]", error);
3145 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3149 * Load the bit that tells us to use the new accounting function
3150 * (raid-z deflation). If we have an older pool, this will not
3153 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3154 if (error != 0 && error != ENOENT)
3155 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3157 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3158 &spa->spa_creation_version, B_FALSE);
3159 if (error != 0 && error != ENOENT)
3160 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3163 * Load the persistent error log. If we have an older pool, this will
3166 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3168 if (error != 0 && error != ENOENT)
3169 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3171 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3172 &spa->spa_errlog_scrub, B_FALSE);
3173 if (error != 0 && error != ENOENT)
3174 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3177 * Load the history object. If we have an older pool, this
3178 * will not be present.
3180 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3181 if (error != 0 && error != ENOENT)
3182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3185 * Load the per-vdev ZAP map. If we have an older pool, this will not
3186 * be present; in this case, defer its creation to a later time to
3187 * avoid dirtying the MOS this early / out of sync context. See
3188 * spa_sync_config_object.
3191 /* The sentinel is only available in the MOS config. */
3192 nvlist_t *mos_config;
3193 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3194 spa_load_failed(spa, "unable to retrieve MOS config");
3195 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3198 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3199 &spa->spa_all_vdev_zaps, B_FALSE);
3201 if (error == ENOENT) {
3202 VERIFY(!nvlist_exists(mos_config,
3203 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3204 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3205 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3206 } else if (error != 0) {
3207 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3208 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3210 * An older version of ZFS overwrote the sentinel value, so
3211 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3212 * destruction to later; see spa_sync_config_object.
3214 spa->spa_avz_action = AVZ_ACTION_DESTROY;
3216 * We're assuming that no vdevs have had their ZAPs created
3217 * before this. Better be sure of it.
3219 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3221 nvlist_free(mos_config);
3223 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3225 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3227 if (error && error != ENOENT)
3228 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3231 uint64_t autoreplace;
3233 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3234 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3235 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3236 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3237 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3238 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3239 &spa->spa_dedup_ditto);
3241 spa->spa_autoreplace = (autoreplace != 0);
3245 * If we are importing a pool with missing top-level vdevs,
3246 * we enforce that the pool doesn't panic or get suspended on
3247 * error since the likelihood of missing data is extremely high.
3249 if (spa->spa_missing_tvds > 0 &&
3250 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3251 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3252 spa_load_note(spa, "forcing failmode to 'continue' "
3253 "as some top level vdevs are missing");
3254 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3261 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3264 vdev_t *rvd = spa->spa_root_vdev;
3267 * If we're assembling the pool from the split-off vdevs of
3268 * an existing pool, we don't want to attach the spares & cache
3273 * Load any hot spares for this pool.
3275 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3277 if (error != 0 && error != ENOENT)
3278 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3279 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3280 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3281 if (load_nvlist(spa, spa->spa_spares.sav_object,
3282 &spa->spa_spares.sav_config) != 0) {
3283 spa_load_failed(spa, "error loading spares nvlist");
3284 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3287 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3288 spa_load_spares(spa);
3289 spa_config_exit(spa, SCL_ALL, FTAG);
3290 } else if (error == 0) {
3291 spa->spa_spares.sav_sync = B_TRUE;
3295 * Load any level 2 ARC devices for this pool.
3297 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3298 &spa->spa_l2cache.sav_object, B_FALSE);
3299 if (error != 0 && error != ENOENT)
3300 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3301 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3302 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3303 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3304 &spa->spa_l2cache.sav_config) != 0) {
3305 spa_load_failed(spa, "error loading l2cache nvlist");
3306 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3309 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3310 spa_load_l2cache(spa);
3311 spa_config_exit(spa, SCL_ALL, FTAG);
3312 } else if (error == 0) {
3313 spa->spa_l2cache.sav_sync = B_TRUE;
3320 spa_ld_load_vdev_metadata(spa_t *spa)
3323 vdev_t *rvd = spa->spa_root_vdev;
3326 * If the 'autoreplace' property is set, then post a resource notifying
3327 * the ZFS DE that it should not issue any faults for unopenable
3328 * devices. We also iterate over the vdevs, and post a sysevent for any
3329 * unopenable vdevs so that the normal autoreplace handler can take
3332 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3333 spa_check_removed(spa->spa_root_vdev);
3335 * For the import case, this is done in spa_import(), because
3336 * at this point we're using the spare definitions from
3337 * the MOS config, not necessarily from the userland config.
3339 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3340 spa_aux_check_removed(&spa->spa_spares);
3341 spa_aux_check_removed(&spa->spa_l2cache);
3346 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3348 error = vdev_load(rvd);
3350 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3351 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3355 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3357 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3358 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3359 spa_config_exit(spa, SCL_ALL, FTAG);
3365 spa_ld_load_dedup_tables(spa_t *spa)
3368 vdev_t *rvd = spa->spa_root_vdev;
3370 error = ddt_load(spa);
3372 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3373 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3380 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3382 vdev_t *rvd = spa->spa_root_vdev;
3384 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3385 boolean_t missing = spa_check_logs(spa);
3387 if (spa->spa_missing_tvds != 0) {
3388 spa_load_note(spa, "spa_check_logs failed "
3389 "so dropping the logs");
3391 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3392 spa_load_failed(spa, "spa_check_logs failed");
3393 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3403 spa_ld_verify_pool_data(spa_t *spa)
3406 vdev_t *rvd = spa->spa_root_vdev;
3409 * We've successfully opened the pool, verify that we're ready
3410 * to start pushing transactions.
3412 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3413 error = spa_load_verify(spa);
3415 spa_load_failed(spa, "spa_load_verify failed "
3416 "[error=%d]", error);
3417 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3426 spa_ld_claim_log_blocks(spa_t *spa)
3429 dsl_pool_t *dp = spa_get_dsl(spa);
3432 * Claim log blocks that haven't been committed yet.
3433 * This must all happen in a single txg.
3434 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3435 * invoked from zil_claim_log_block()'s i/o done callback.
3436 * Price of rollback is that we abandon the log.
3438 spa->spa_claiming = B_TRUE;
3440 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3441 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3442 zil_claim, tx, DS_FIND_CHILDREN);
3445 spa->spa_claiming = B_FALSE;
3447 spa_set_log_state(spa, SPA_LOG_GOOD);
3451 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3452 boolean_t update_config_cache)
3454 vdev_t *rvd = spa->spa_root_vdev;
3455 int need_update = B_FALSE;
3458 * If the config cache is stale, or we have uninitialized
3459 * metaslabs (see spa_vdev_add()), then update the config.
3461 * If this is a verbatim import, trust the current
3462 * in-core spa_config and update the disk labels.
3464 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3465 spa->spa_load_state == SPA_LOAD_IMPORT ||
3466 spa->spa_load_state == SPA_LOAD_RECOVER ||
3467 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3468 need_update = B_TRUE;
3470 for (int c = 0; c < rvd->vdev_children; c++)
3471 if (rvd->vdev_child[c]->vdev_ms_array == 0)
3472 need_update = B_TRUE;
3475 * Update the config cache asychronously in case we're the
3476 * root pool, in which case the config cache isn't writable yet.
3479 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3483 spa_ld_prepare_for_reload(spa_t *spa)
3485 int mode = spa->spa_mode;
3486 int async_suspended = spa->spa_async_suspended;
3489 spa_deactivate(spa);
3490 spa_activate(spa, mode);
3493 * We save the value of spa_async_suspended as it gets reset to 0 by
3494 * spa_unload(). We want to restore it back to the original value before
3495 * returning as we might be calling spa_async_resume() later.
3497 spa->spa_async_suspended = async_suspended;
3501 spa_ld_read_checkpoint_txg(spa_t *spa)
3503 uberblock_t checkpoint;
3506 ASSERT0(spa->spa_checkpoint_txg);
3507 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3509 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3510 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3511 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3513 if (error == ENOENT)
3519 ASSERT3U(checkpoint.ub_txg, !=, 0);
3520 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3521 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3522 spa->spa_checkpoint_txg = checkpoint.ub_txg;
3523 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3529 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3533 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3534 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3537 * Never trust the config that is provided unless we are assembling
3538 * a pool following a split.
3539 * This means don't trust blkptrs and the vdev tree in general. This
3540 * also effectively puts the spa in read-only mode since
3541 * spa_writeable() checks for spa_trust_config to be true.
3542 * We will later load a trusted config from the MOS.
3544 if (type != SPA_IMPORT_ASSEMBLE)
3545 spa->spa_trust_config = B_FALSE;
3548 * Parse the config provided to create a vdev tree.
3550 error = spa_ld_parse_config(spa, type);
3555 * Now that we have the vdev tree, try to open each vdev. This involves
3556 * opening the underlying physical device, retrieving its geometry and
3557 * probing the vdev with a dummy I/O. The state of each vdev will be set
3558 * based on the success of those operations. After this we'll be ready
3559 * to read from the vdevs.
3561 error = spa_ld_open_vdevs(spa);
3566 * Read the label of each vdev and make sure that the GUIDs stored
3567 * there match the GUIDs in the config provided.
3568 * If we're assembling a new pool that's been split off from an
3569 * existing pool, the labels haven't yet been updated so we skip
3570 * validation for now.
3572 if (type != SPA_IMPORT_ASSEMBLE) {
3573 error = spa_ld_validate_vdevs(spa);
3579 * Read all vdev labels to find the best uberblock (i.e. latest,
3580 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3581 * get the list of features required to read blkptrs in the MOS from
3582 * the vdev label with the best uberblock and verify that our version
3583 * of zfs supports them all.
3585 error = spa_ld_select_uberblock(spa, type);
3590 * Pass that uberblock to the dsl_pool layer which will open the root
3591 * blkptr. This blkptr points to the latest version of the MOS and will
3592 * allow us to read its contents.
3594 error = spa_ld_open_rootbp(spa);
3602 spa_ld_checkpoint_rewind(spa_t *spa)
3604 uberblock_t checkpoint;
3607 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3608 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3610 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3611 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3612 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3615 spa_load_failed(spa, "unable to retrieve checkpointed "
3616 "uberblock from the MOS config [error=%d]", error);
3618 if (error == ENOENT)
3619 error = ZFS_ERR_NO_CHECKPOINT;
3624 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3625 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3628 * We need to update the txg and timestamp of the checkpointed
3629 * uberblock to be higher than the latest one. This ensures that
3630 * the checkpointed uberblock is selected if we were to close and
3631 * reopen the pool right after we've written it in the vdev labels.
3632 * (also see block comment in vdev_uberblock_compare)
3634 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3635 checkpoint.ub_timestamp = gethrestime_sec();
3638 * Set current uberblock to be the checkpointed uberblock.
3640 spa->spa_uberblock = checkpoint;
3643 * If we are doing a normal rewind, then the pool is open for
3644 * writing and we sync the "updated" checkpointed uberblock to
3645 * disk. Once this is done, we've basically rewound the whole
3646 * pool and there is no way back.
3648 * There are cases when we don't want to attempt and sync the
3649 * checkpointed uberblock to disk because we are opening a
3650 * pool as read-only. Specifically, verifying the checkpointed
3651 * state with zdb, and importing the checkpointed state to get
3652 * a "preview" of its content.
3654 if (spa_writeable(spa)) {
3655 vdev_t *rvd = spa->spa_root_vdev;
3657 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3658 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3660 int children = rvd->vdev_children;
3661 int c0 = spa_get_random(children);
3663 for (int c = 0; c < children; c++) {
3664 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3666 /* Stop when revisiting the first vdev */
3667 if (c > 0 && svd[0] == vd)
3670 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3671 !vdev_is_concrete(vd))
3674 svd[svdcount++] = vd;
3675 if (svdcount == SPA_SYNC_MIN_VDEVS)
3678 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3680 spa->spa_last_synced_guid = rvd->vdev_guid;
3681 spa_config_exit(spa, SCL_ALL, FTAG);
3684 spa_load_failed(spa, "failed to write checkpointed "
3685 "uberblock to the vdev labels [error=%d]", error);
3694 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3695 boolean_t *update_config_cache)
3700 * Parse the config for pool, open and validate vdevs,
3701 * select an uberblock, and use that uberblock to open
3704 error = spa_ld_mos_init(spa, type);
3709 * Retrieve the trusted config stored in the MOS and use it to create
3710 * a new, exact version of the vdev tree, then reopen all vdevs.
3712 error = spa_ld_trusted_config(spa, type, B_FALSE);
3713 if (error == EAGAIN) {
3714 if (update_config_cache != NULL)
3715 *update_config_cache = B_TRUE;
3718 * Redo the loading process with the trusted config if it is
3719 * too different from the untrusted config.
3721 spa_ld_prepare_for_reload(spa);
3722 spa_load_note(spa, "RELOADING");
3723 error = spa_ld_mos_init(spa, type);
3727 error = spa_ld_trusted_config(spa, type, B_TRUE);
3731 } else if (error != 0) {
3739 * Load an existing storage pool, using the config provided. This config
3740 * describes which vdevs are part of the pool and is later validated against
3741 * partial configs present in each vdev's label and an entire copy of the
3742 * config stored in the MOS.
3745 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3748 boolean_t missing_feat_write = B_FALSE;
3749 boolean_t checkpoint_rewind =
3750 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3751 boolean_t update_config_cache = B_FALSE;
3753 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3754 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3756 spa_load_note(spa, "LOADING");
3758 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3763 * If we are rewinding to the checkpoint then we need to repeat
3764 * everything we've done so far in this function but this time
3765 * selecting the checkpointed uberblock and using that to open
3768 if (checkpoint_rewind) {
3770 * If we are rewinding to the checkpoint update config cache
3773 update_config_cache = B_TRUE;
3776 * Extract the checkpointed uberblock from the current MOS
3777 * and use this as the pool's uberblock from now on. If the
3778 * pool is imported as writeable we also write the checkpoint
3779 * uberblock to the labels, making the rewind permanent.
3781 error = spa_ld_checkpoint_rewind(spa);
3786 * Redo the loading process process again with the
3787 * checkpointed uberblock.
3789 spa_ld_prepare_for_reload(spa);
3790 spa_load_note(spa, "LOADING checkpointed uberblock");
3791 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3797 * Retrieve the checkpoint txg if the pool has a checkpoint.
3799 error = spa_ld_read_checkpoint_txg(spa);
3804 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3805 * from the pool and their contents were re-mapped to other vdevs. Note
3806 * that everything that we read before this step must have been
3807 * rewritten on concrete vdevs after the last device removal was
3808 * initiated. Otherwise we could be reading from indirect vdevs before
3809 * we have loaded their mappings.
3811 error = spa_ld_open_indirect_vdev_metadata(spa);
3816 * Retrieve the full list of active features from the MOS and check if
3817 * they are all supported.
3819 error = spa_ld_check_features(spa, &missing_feat_write);
3824 * Load several special directories from the MOS needed by the dsl_pool
3827 error = spa_ld_load_special_directories(spa);
3832 * Retrieve pool properties from the MOS.
3834 error = spa_ld_get_props(spa);
3839 * Retrieve the list of auxiliary devices - cache devices and spares -
3842 error = spa_ld_open_aux_vdevs(spa, type);
3847 * Load the metadata for all vdevs. Also check if unopenable devices
3848 * should be autoreplaced.
3850 error = spa_ld_load_vdev_metadata(spa);
3854 error = spa_ld_load_dedup_tables(spa);
3859 * Verify the logs now to make sure we don't have any unexpected errors
3860 * when we claim log blocks later.
3862 error = spa_ld_verify_logs(spa, type, ereport);
3866 if (missing_feat_write) {
3867 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3870 * At this point, we know that we can open the pool in
3871 * read-only mode but not read-write mode. We now have enough
3872 * information and can return to userland.
3874 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3879 * Traverse the last txgs to make sure the pool was left off in a safe
3880 * state. When performing an extreme rewind, we verify the whole pool,
3881 * which can take a very long time.
3883 error = spa_ld_verify_pool_data(spa);
3888 * Calculate the deflated space for the pool. This must be done before
3889 * we write anything to the pool because we'd need to update the space
3890 * accounting using the deflated sizes.
3892 spa_update_dspace(spa);
3895 * We have now retrieved all the information we needed to open the
3896 * pool. If we are importing the pool in read-write mode, a few
3897 * additional steps must be performed to finish the import.
3899 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3900 spa->spa_load_max_txg == UINT64_MAX)) {
3901 uint64_t config_cache_txg = spa->spa_config_txg;
3903 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3906 * In case of a checkpoint rewind, log the original txg
3907 * of the checkpointed uberblock.
3909 if (checkpoint_rewind) {
3910 spa_history_log_internal(spa, "checkpoint rewind",
3911 NULL, "rewound state to txg=%llu",
3912 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3916 * Traverse the ZIL and claim all blocks.
3918 spa_ld_claim_log_blocks(spa);
3921 * Kick-off the syncing thread.
3923 spa->spa_sync_on = B_TRUE;
3924 txg_sync_start(spa->spa_dsl_pool);
3927 * Wait for all claims to sync. We sync up to the highest
3928 * claimed log block birth time so that claimed log blocks
3929 * don't appear to be from the future. spa_claim_max_txg
3930 * will have been set for us by ZIL traversal operations
3933 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3936 * Check if we need to request an update of the config. On the
3937 * next sync, we would update the config stored in vdev labels
3938 * and the cachefile (by default /etc/zfs/zpool.cache).
3940 spa_ld_check_for_config_update(spa, config_cache_txg,
3941 update_config_cache);
3944 * Check all DTLs to see if anything needs resilvering.
3946 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3947 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3948 spa_async_request(spa, SPA_ASYNC_RESILVER);
3951 * Log the fact that we booted up (so that we can detect if
3952 * we rebooted in the middle of an operation).
3954 spa_history_log_version(spa, "open");
3957 * Delete any inconsistent datasets.
3959 (void) dmu_objset_find(spa_name(spa),
3960 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3963 * Clean up any stale temporary dataset userrefs.
3965 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3967 spa_restart_removal(spa);
3969 spa_spawn_aux_threads(spa);
3971 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3972 vdev_initialize_restart(spa->spa_root_vdev);
3973 spa_config_exit(spa, SCL_CONFIG, FTAG);
3976 spa_load_note(spa, "LOADED");
3982 spa_load_retry(spa_t *spa, spa_load_state_t state)
3984 int mode = spa->spa_mode;
3987 spa_deactivate(spa);
3989 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3991 spa_activate(spa, mode);
3992 spa_async_suspend(spa);
3994 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3995 (u_longlong_t)spa->spa_load_max_txg);
3997 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
4001 * If spa_load() fails this function will try loading prior txg's. If
4002 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
4003 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
4004 * function will not rewind the pool and will return the same error as
4008 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
4011 nvlist_t *loadinfo = NULL;
4012 nvlist_t *config = NULL;
4013 int load_error, rewind_error;
4014 uint64_t safe_rewind_txg;
4017 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4018 spa->spa_load_max_txg = spa->spa_load_txg;
4019 spa_set_log_state(spa, SPA_LOG_CLEAR);
4021 spa->spa_load_max_txg = max_request;
4022 if (max_request != UINT64_MAX)
4023 spa->spa_extreme_rewind = B_TRUE;
4026 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4027 if (load_error == 0)
4029 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4031 * When attempting checkpoint-rewind on a pool with no
4032 * checkpoint, we should not attempt to load uberblocks
4033 * from previous txgs when spa_load fails.
4035 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4036 return (load_error);
4039 if (spa->spa_root_vdev != NULL)
4040 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4042 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4043 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4045 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4046 nvlist_free(config);
4047 return (load_error);
4050 if (state == SPA_LOAD_RECOVER) {
4051 /* Price of rolling back is discarding txgs, including log */
4052 spa_set_log_state(spa, SPA_LOG_CLEAR);
4055 * If we aren't rolling back save the load info from our first
4056 * import attempt so that we can restore it after attempting
4059 loadinfo = spa->spa_load_info;
4060 spa->spa_load_info = fnvlist_alloc();
4063 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4064 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4065 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4066 TXG_INITIAL : safe_rewind_txg;
4069 * Continue as long as we're finding errors, we're still within
4070 * the acceptable rewind range, and we're still finding uberblocks
4072 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4073 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4074 if (spa->spa_load_max_txg < safe_rewind_txg)
4075 spa->spa_extreme_rewind = B_TRUE;
4076 rewind_error = spa_load_retry(spa, state);
4079 spa->spa_extreme_rewind = B_FALSE;
4080 spa->spa_load_max_txg = UINT64_MAX;
4082 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4083 spa_config_set(spa, config);
4085 nvlist_free(config);
4087 if (state == SPA_LOAD_RECOVER) {
4088 ASSERT3P(loadinfo, ==, NULL);
4089 return (rewind_error);
4091 /* Store the rewind info as part of the initial load info */
4092 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4093 spa->spa_load_info);
4095 /* Restore the initial load info */
4096 fnvlist_free(spa->spa_load_info);
4097 spa->spa_load_info = loadinfo;
4099 return (load_error);
4106 * The import case is identical to an open except that the configuration is sent
4107 * down from userland, instead of grabbed from the configuration cache. For the
4108 * case of an open, the pool configuration will exist in the
4109 * POOL_STATE_UNINITIALIZED state.
4111 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4112 * the same time open the pool, without having to keep around the spa_t in some
4116 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4120 spa_load_state_t state = SPA_LOAD_OPEN;
4122 int locked = B_FALSE;
4123 int firstopen = B_FALSE;
4128 * As disgusting as this is, we need to support recursive calls to this
4129 * function because dsl_dir_open() is called during spa_load(), and ends
4130 * up calling spa_open() again. The real fix is to figure out how to
4131 * avoid dsl_dir_open() calling this in the first place.
4133 if (mutex_owner(&spa_namespace_lock) != curthread) {
4134 mutex_enter(&spa_namespace_lock);
4138 if ((spa = spa_lookup(pool)) == NULL) {
4140 mutex_exit(&spa_namespace_lock);
4141 return (SET_ERROR(ENOENT));
4144 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
4145 zpool_load_policy_t policy;
4149 zpool_get_load_policy(nvpolicy ? nvpolicy : spa->spa_config,
4151 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
4152 state = SPA_LOAD_RECOVER;
4154 spa_activate(spa, spa_mode_global);
4156 if (state != SPA_LOAD_RECOVER)
4157 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4158 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4160 zfs_dbgmsg("spa_open_common: opening %s", pool);
4161 error = spa_load_best(spa, state, policy.zlp_txg,
4164 if (error == EBADF) {
4166 * If vdev_validate() returns failure (indicated by
4167 * EBADF), it indicates that one of the vdevs indicates
4168 * that the pool has been exported or destroyed. If
4169 * this is the case, the config cache is out of sync and
4170 * we should remove the pool from the namespace.
4173 spa_deactivate(spa);
4174 spa_write_cachefile(spa, B_TRUE, B_TRUE);
4177 mutex_exit(&spa_namespace_lock);
4178 return (SET_ERROR(ENOENT));
4183 * We can't open the pool, but we still have useful
4184 * information: the state of each vdev after the
4185 * attempted vdev_open(). Return this to the user.
4187 if (config != NULL && spa->spa_config) {
4188 VERIFY(nvlist_dup(spa->spa_config, config,
4190 VERIFY(nvlist_add_nvlist(*config,
4191 ZPOOL_CONFIG_LOAD_INFO,
4192 spa->spa_load_info) == 0);
4195 spa_deactivate(spa);
4196 spa->spa_last_open_failed = error;
4198 mutex_exit(&spa_namespace_lock);
4204 spa_open_ref(spa, tag);
4207 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4210 * If we've recovered the pool, pass back any information we
4211 * gathered while doing the load.
4213 if (state == SPA_LOAD_RECOVER) {
4214 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4215 spa->spa_load_info) == 0);
4219 spa->spa_last_open_failed = 0;
4220 spa->spa_last_ubsync_txg = 0;
4221 spa->spa_load_txg = 0;
4222 mutex_exit(&spa_namespace_lock);
4226 zvol_create_minors(spa->spa_name);
4237 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4240 return (spa_open_common(name, spapp, tag, policy, config));
4244 spa_open(const char *name, spa_t **spapp, void *tag)
4246 return (spa_open_common(name, spapp, tag, NULL, NULL));
4250 * Lookup the given spa_t, incrementing the inject count in the process,
4251 * preventing it from being exported or destroyed.
4254 spa_inject_addref(char *name)
4258 mutex_enter(&spa_namespace_lock);
4259 if ((spa = spa_lookup(name)) == NULL) {
4260 mutex_exit(&spa_namespace_lock);
4263 spa->spa_inject_ref++;
4264 mutex_exit(&spa_namespace_lock);
4270 spa_inject_delref(spa_t *spa)
4272 mutex_enter(&spa_namespace_lock);
4273 spa->spa_inject_ref--;
4274 mutex_exit(&spa_namespace_lock);
4278 * Add spares device information to the nvlist.
4281 spa_add_spares(spa_t *spa, nvlist_t *config)
4291 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4293 if (spa->spa_spares.sav_count == 0)
4296 VERIFY(nvlist_lookup_nvlist(config,
4297 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4298 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4299 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4301 VERIFY(nvlist_add_nvlist_array(nvroot,
4302 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4303 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4304 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4307 * Go through and find any spares which have since been
4308 * repurposed as an active spare. If this is the case, update
4309 * their status appropriately.
4311 for (i = 0; i < nspares; i++) {
4312 VERIFY(nvlist_lookup_uint64(spares[i],
4313 ZPOOL_CONFIG_GUID, &guid) == 0);
4314 if (spa_spare_exists(guid, &pool, NULL) &&
4316 VERIFY(nvlist_lookup_uint64_array(
4317 spares[i], ZPOOL_CONFIG_VDEV_STATS,
4318 (uint64_t **)&vs, &vsc) == 0);
4319 vs->vs_state = VDEV_STATE_CANT_OPEN;
4320 vs->vs_aux = VDEV_AUX_SPARED;
4327 * Add l2cache device information to the nvlist, including vdev stats.
4330 spa_add_l2cache(spa_t *spa, nvlist_t *config)
4333 uint_t i, j, nl2cache;
4340 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4342 if (spa->spa_l2cache.sav_count == 0)
4345 VERIFY(nvlist_lookup_nvlist(config,
4346 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4347 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4348 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4349 if (nl2cache != 0) {
4350 VERIFY(nvlist_add_nvlist_array(nvroot,
4351 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4352 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4353 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4356 * Update level 2 cache device stats.
4359 for (i = 0; i < nl2cache; i++) {
4360 VERIFY(nvlist_lookup_uint64(l2cache[i],
4361 ZPOOL_CONFIG_GUID, &guid) == 0);
4364 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4366 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4367 vd = spa->spa_l2cache.sav_vdevs[j];
4373 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4374 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4376 vdev_get_stats(vd, vs);
4382 spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
4387 /* We may be unable to read features if pool is suspended. */
4388 if (spa_suspended(spa))
4391 if (spa->spa_feat_for_read_obj != 0) {
4392 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4393 spa->spa_feat_for_read_obj);
4394 zap_cursor_retrieve(&zc, &za) == 0;
4395 zap_cursor_advance(&zc)) {
4396 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4397 za.za_num_integers == 1);
4398 VERIFY0(nvlist_add_uint64(features, za.za_name,
4399 za.za_first_integer));
4401 zap_cursor_fini(&zc);
4404 if (spa->spa_feat_for_write_obj != 0) {
4405 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4406 spa->spa_feat_for_write_obj);
4407 zap_cursor_retrieve(&zc, &za) == 0;
4408 zap_cursor_advance(&zc)) {
4409 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4410 za.za_num_integers == 1);
4411 VERIFY0(nvlist_add_uint64(features, za.za_name,
4412 za.za_first_integer));
4414 zap_cursor_fini(&zc);
4419 spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
4423 for (i = 0; i < SPA_FEATURES; i++) {
4424 zfeature_info_t feature = spa_feature_table[i];
4427 if (feature_get_refcount(spa, &feature, &refcount) != 0)
4430 VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
4435 * Store a list of pool features and their reference counts in the
4438 * The first time this is called on a spa, allocate a new nvlist, fetch
4439 * the pool features and reference counts from disk, then save the list
4440 * in the spa. In subsequent calls on the same spa use the saved nvlist
4441 * and refresh its values from the cached reference counts. This
4442 * ensures we don't block here on I/O on a suspended pool so 'zpool
4443 * clear' can resume the pool.
4446 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4450 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4452 mutex_enter(&spa->spa_feat_stats_lock);
4453 features = spa->spa_feat_stats;
4455 if (features != NULL) {
4456 spa_feature_stats_from_cache(spa, features);
4458 VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
4459 spa->spa_feat_stats = features;
4460 spa_feature_stats_from_disk(spa, features);
4463 VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4466 mutex_exit(&spa->spa_feat_stats_lock);
4470 spa_get_stats(const char *name, nvlist_t **config,
4471 char *altroot, size_t buflen)
4477 error = spa_open_common(name, &spa, FTAG, NULL, config);
4481 * This still leaves a window of inconsistency where the spares
4482 * or l2cache devices could change and the config would be
4483 * self-inconsistent.
4485 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4487 if (*config != NULL) {
4488 uint64_t loadtimes[2];
4490 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4491 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4492 VERIFY(nvlist_add_uint64_array(*config,
4493 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4495 VERIFY(nvlist_add_uint64(*config,
4496 ZPOOL_CONFIG_ERRCOUNT,
4497 spa_get_errlog_size(spa)) == 0);
4499 if (spa_suspended(spa))
4500 VERIFY(nvlist_add_uint64(*config,
4501 ZPOOL_CONFIG_SUSPENDED,
4502 spa->spa_failmode) == 0);
4504 spa_add_spares(spa, *config);
4505 spa_add_l2cache(spa, *config);
4506 spa_add_feature_stats(spa, *config);
4511 * We want to get the alternate root even for faulted pools, so we cheat
4512 * and call spa_lookup() directly.
4516 mutex_enter(&spa_namespace_lock);
4517 spa = spa_lookup(name);
4519 spa_altroot(spa, altroot, buflen);
4523 mutex_exit(&spa_namespace_lock);
4525 spa_altroot(spa, altroot, buflen);
4530 spa_config_exit(spa, SCL_CONFIG, FTAG);
4531 spa_close(spa, FTAG);
4538 * Validate that the auxiliary device array is well formed. We must have an
4539 * array of nvlists, each which describes a valid leaf vdev. If this is an
4540 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4541 * specified, as long as they are well-formed.
4544 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4545 spa_aux_vdev_t *sav, const char *config, uint64_t version,
4546 vdev_labeltype_t label)
4553 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4556 * It's acceptable to have no devs specified.
4558 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4562 return (SET_ERROR(EINVAL));
4565 * Make sure the pool is formatted with a version that supports this
4568 if (spa_version(spa) < version)
4569 return (SET_ERROR(ENOTSUP));
4572 * Set the pending device list so we correctly handle device in-use
4575 sav->sav_pending = dev;
4576 sav->sav_npending = ndev;
4578 for (i = 0; i < ndev; i++) {
4579 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4583 if (!vd->vdev_ops->vdev_op_leaf) {
4585 error = SET_ERROR(EINVAL);
4590 * The L2ARC currently only supports disk devices in
4591 * kernel context. For user-level testing, we allow it.
4594 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4595 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4596 error = SET_ERROR(ENOTBLK);
4603 if ((error = vdev_open(vd)) == 0 &&
4604 (error = vdev_label_init(vd, crtxg, label)) == 0) {
4605 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4606 vd->vdev_guid) == 0);
4612 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4619 sav->sav_pending = NULL;
4620 sav->sav_npending = 0;
4625 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4629 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4631 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4632 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4633 VDEV_LABEL_SPARE)) != 0) {
4637 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4638 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4639 VDEV_LABEL_L2CACHE));
4643 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4648 if (sav->sav_config != NULL) {
4654 * Generate new dev list by concatentating with the
4657 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4658 &olddevs, &oldndevs) == 0);
4660 newdevs = kmem_alloc(sizeof (void *) *
4661 (ndevs + oldndevs), KM_SLEEP);
4662 for (i = 0; i < oldndevs; i++)
4663 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4665 for (i = 0; i < ndevs; i++)
4666 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4669 VERIFY(nvlist_remove(sav->sav_config, config,
4670 DATA_TYPE_NVLIST_ARRAY) == 0);
4672 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4673 config, newdevs, ndevs + oldndevs) == 0);
4674 for (i = 0; i < oldndevs + ndevs; i++)
4675 nvlist_free(newdevs[i]);
4676 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4679 * Generate a new dev list.
4681 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4683 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4689 * Stop and drop level 2 ARC devices
4692 spa_l2cache_drop(spa_t *spa)
4696 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4698 for (i = 0; i < sav->sav_count; i++) {
4701 vd = sav->sav_vdevs[i];
4704 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4705 pool != 0ULL && l2arc_vdev_present(vd))
4706 l2arc_remove_vdev(vd);
4714 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4718 char *altroot = NULL;
4723 uint64_t txg = TXG_INITIAL;
4724 nvlist_t **spares, **l2cache;
4725 uint_t nspares, nl2cache;
4726 uint64_t version, obj;
4727 boolean_t has_features;
4731 if (nvlist_lookup_string(props,
4732 zpool_prop_to_name(ZPOOL_PROP_TNAME), &poolname) != 0)
4733 poolname = (char *)pool;
4736 * If this pool already exists, return failure.
4738 mutex_enter(&spa_namespace_lock);
4739 if (spa_lookup(poolname) != NULL) {
4740 mutex_exit(&spa_namespace_lock);
4741 return (SET_ERROR(EEXIST));
4745 * Allocate a new spa_t structure.
4747 nvl = fnvlist_alloc();
4748 fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
4749 (void) nvlist_lookup_string(props,
4750 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4751 spa = spa_add(poolname, nvl, altroot);
4753 spa_activate(spa, spa_mode_global);
4755 if (props && (error = spa_prop_validate(spa, props))) {
4756 spa_deactivate(spa);
4758 mutex_exit(&spa_namespace_lock);
4763 * Temporary pool names should never be written to disk.
4765 if (poolname != pool)
4766 spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
4768 has_features = B_FALSE;
4769 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4770 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4771 if (zpool_prop_feature(nvpair_name(elem)))
4772 has_features = B_TRUE;
4775 if (has_features || nvlist_lookup_uint64(props,
4776 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4777 version = SPA_VERSION;
4779 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4781 spa->spa_first_txg = txg;
4782 spa->spa_uberblock.ub_txg = txg - 1;
4783 spa->spa_uberblock.ub_version = version;
4784 spa->spa_ubsync = spa->spa_uberblock;
4785 spa->spa_load_state = SPA_LOAD_CREATE;
4786 spa->spa_removing_phys.sr_state = DSS_NONE;
4787 spa->spa_removing_phys.sr_removing_vdev = -1;
4788 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4789 spa->spa_indirect_vdevs_loaded = B_TRUE;
4792 * Create "The Godfather" zio to hold all async IOs
4794 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4796 for (int i = 0; i < max_ncpus; i++) {
4797 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4798 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4799 ZIO_FLAG_GODFATHER);
4803 * Create the root vdev.
4805 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4807 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4809 ASSERT(error != 0 || rvd != NULL);
4810 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4812 if (error == 0 && !zfs_allocatable_devs(nvroot))
4813 error = SET_ERROR(EINVAL);
4816 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4817 (error = spa_validate_aux(spa, nvroot, txg,
4818 VDEV_ALLOC_ADD)) == 0) {
4819 for (int c = 0; c < rvd->vdev_children; c++) {
4820 vdev_ashift_optimize(rvd->vdev_child[c]);
4821 vdev_metaslab_set_size(rvd->vdev_child[c]);
4822 vdev_expand(rvd->vdev_child[c], txg);
4826 spa_config_exit(spa, SCL_ALL, FTAG);
4830 spa_deactivate(spa);
4832 mutex_exit(&spa_namespace_lock);
4837 * Get the list of spares, if specified.
4839 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4840 &spares, &nspares) == 0) {
4841 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4843 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4844 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4845 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4846 spa_load_spares(spa);
4847 spa_config_exit(spa, SCL_ALL, FTAG);
4848 spa->spa_spares.sav_sync = B_TRUE;
4852 * Get the list of level 2 cache devices, if specified.
4854 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4855 &l2cache, &nl2cache) == 0) {
4856 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4857 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4858 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4859 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4860 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4861 spa_load_l2cache(spa);
4862 spa_config_exit(spa, SCL_ALL, FTAG);
4863 spa->spa_l2cache.sav_sync = B_TRUE;
4866 spa->spa_is_initializing = B_TRUE;
4867 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4868 spa->spa_meta_objset = dp->dp_meta_objset;
4869 spa->spa_is_initializing = B_FALSE;
4872 * Create DDTs (dedup tables).
4876 spa_update_dspace(spa);
4878 tx = dmu_tx_create_assigned(dp, txg);
4881 * Create the pool config object.
4883 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4884 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4885 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4887 if (zap_add(spa->spa_meta_objset,
4888 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4889 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4890 cmn_err(CE_PANIC, "failed to add pool config");
4893 if (spa_version(spa) >= SPA_VERSION_FEATURES)
4894 spa_feature_create_zap_objects(spa, tx);
4896 if (zap_add(spa->spa_meta_objset,
4897 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4898 sizeof (uint64_t), 1, &version, tx) != 0) {
4899 cmn_err(CE_PANIC, "failed to add pool version");
4902 /* Newly created pools with the right version are always deflated. */
4903 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4904 spa->spa_deflate = TRUE;
4905 if (zap_add(spa->spa_meta_objset,
4906 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4907 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4908 cmn_err(CE_PANIC, "failed to add deflate");
4913 * Create the deferred-free bpobj. Turn off compression
4914 * because sync-to-convergence takes longer if the blocksize
4917 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4918 dmu_object_set_compress(spa->spa_meta_objset, obj,
4919 ZIO_COMPRESS_OFF, tx);
4920 if (zap_add(spa->spa_meta_objset,
4921 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4922 sizeof (uint64_t), 1, &obj, tx) != 0) {
4923 cmn_err(CE_PANIC, "failed to add bpobj");
4925 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4926 spa->spa_meta_objset, obj));
4929 * Create the pool's history object.
4931 if (version >= SPA_VERSION_ZPOOL_HISTORY)
4932 spa_history_create_obj(spa, tx);
4935 * Generate some random noise for salted checksums to operate on.
4937 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4938 sizeof (spa->spa_cksum_salt.zcs_bytes));
4941 * Set pool properties.
4943 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4944 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4945 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4946 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4948 if (props != NULL) {
4949 spa_configfile_set(spa, props, B_FALSE);
4950 spa_sync_props(props, tx);
4955 spa->spa_sync_on = B_TRUE;
4956 txg_sync_start(spa->spa_dsl_pool);
4959 * We explicitly wait for the first transaction to complete so that our
4960 * bean counters are appropriately updated.
4962 txg_wait_synced(spa->spa_dsl_pool, txg);
4964 spa_spawn_aux_threads(spa);
4966 spa_write_cachefile(spa, B_FALSE, B_TRUE);
4967 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4969 spa_history_log_version(spa, "create");
4972 * Don't count references from objsets that are already closed
4973 * and are making their way through the eviction process.
4975 spa_evicting_os_wait(spa);
4976 spa->spa_minref = refcount_count(&spa->spa_refcount);
4977 spa->spa_load_state = SPA_LOAD_NONE;
4979 mutex_exit(&spa_namespace_lock);
4987 * Get the root pool information from the root disk, then import the root pool
4988 * during the system boot up time.
4990 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4993 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4996 nvlist_t *nvtop, *nvroot;
4999 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
5003 * Add this top-level vdev to the child array.
5005 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5007 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5009 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
5012 * Put this pool's top-level vdevs into a root vdev.
5014 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5015 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5016 VDEV_TYPE_ROOT) == 0);
5017 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5018 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5019 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5023 * Replace the existing vdev_tree with the new root vdev in
5024 * this pool's configuration (remove the old, add the new).
5026 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5027 nvlist_free(nvroot);
5032 * Walk the vdev tree and see if we can find a device with "better"
5033 * configuration. A configuration is "better" if the label on that
5034 * device has a more recent txg.
5037 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
5039 for (int c = 0; c < vd->vdev_children; c++)
5040 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
5042 if (vd->vdev_ops->vdev_op_leaf) {
5046 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
5050 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
5054 * Do we have a better boot device?
5056 if (label_txg > *txg) {
5065 * Import a root pool.
5067 * For x86. devpath_list will consist of devid and/or physpath name of
5068 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
5069 * The GRUB "findroot" command will return the vdev we should boot.
5071 * For Sparc, devpath_list consists the physpath name of the booting device
5072 * no matter the rootpool is a single device pool or a mirrored pool.
5074 * "/pci@1f,0/ide@d/disk@0,0:a"
5077 spa_import_rootpool(char *devpath, char *devid)
5080 vdev_t *rvd, *bvd, *avd = NULL;
5081 nvlist_t *config, *nvtop;
5087 * Read the label from the boot device and generate a configuration.
5089 config = spa_generate_rootconf(devpath, devid, &guid);
5090 #if defined(_OBP) && defined(_KERNEL)
5091 if (config == NULL) {
5092 if (strstr(devpath, "/iscsi/ssd") != NULL) {
5094 get_iscsi_bootpath_phy(devpath);
5095 config = spa_generate_rootconf(devpath, devid, &guid);
5099 if (config == NULL) {
5100 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
5102 return (SET_ERROR(EIO));
5105 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5107 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
5109 mutex_enter(&spa_namespace_lock);
5110 if ((spa = spa_lookup(pname)) != NULL) {
5112 * Remove the existing root pool from the namespace so that we
5113 * can replace it with the correct config we just read in.
5118 spa = spa_add(pname, config, NULL);
5119 spa->spa_is_root = B_TRUE;
5120 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5121 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5122 &spa->spa_ubsync.ub_version) != 0)
5123 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5126 * Build up a vdev tree based on the boot device's label config.
5128 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5130 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5131 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5132 VDEV_ALLOC_ROOTPOOL);
5133 spa_config_exit(spa, SCL_ALL, FTAG);
5135 mutex_exit(&spa_namespace_lock);
5136 nvlist_free(config);
5137 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5143 * Get the boot vdev.
5145 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5146 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5147 (u_longlong_t)guid);
5148 error = SET_ERROR(ENOENT);
5153 * Determine if there is a better boot device.
5156 spa_alt_rootvdev(rvd, &avd, &txg);
5158 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5159 "try booting from '%s'", avd->vdev_path);
5160 error = SET_ERROR(EINVAL);
5165 * If the boot device is part of a spare vdev then ensure that
5166 * we're booting off the active spare.
5168 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5169 !bvd->vdev_isspare) {
5170 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5171 "try booting from '%s'",
5173 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5174 error = SET_ERROR(EINVAL);
5180 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5182 spa_config_exit(spa, SCL_ALL, FTAG);
5183 mutex_exit(&spa_namespace_lock);
5185 nvlist_free(config);
5189 #else /* !illumos */
5191 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
5195 spa_generate_rootconf(const char *name)
5197 nvlist_t **configs, **tops;
5199 nvlist_t *best_cfg, *nvtop, *nvroot;
5208 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
5211 ASSERT3U(count, !=, 0);
5213 for (i = 0; i < count; i++) {
5216 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
5218 if (txg > best_txg) {
5220 best_cfg = configs[i];
5225 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
5227 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
5230 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
5231 for (i = 0; i < nchildren; i++) {
5234 if (configs[i] == NULL)
5236 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
5238 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
5240 for (i = 0; holes != NULL && i < nholes; i++) {
5243 if (tops[holes[i]] != NULL)
5245 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
5246 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
5247 VDEV_TYPE_HOLE) == 0);
5248 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
5250 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
5253 for (i = 0; i < nchildren; i++) {
5254 if (tops[i] != NULL)
5256 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
5257 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
5258 VDEV_TYPE_MISSING) == 0);
5259 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
5261 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
5266 * Create pool config based on the best vdev config.
5268 nvlist_dup(best_cfg, &config, KM_SLEEP);
5271 * Put this pool's top-level vdevs into a root vdev.
5273 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5275 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5276 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5277 VDEV_TYPE_ROOT) == 0);
5278 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5279 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5280 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5281 tops, nchildren) == 0);
5284 * Replace the existing vdev_tree with the new root vdev in
5285 * this pool's configuration (remove the old, add the new).
5287 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5290 * Drop vdev config elements that should not be present at pool level.
5292 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
5293 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
5295 for (i = 0; i < count; i++)
5296 nvlist_free(configs[i]);
5297 kmem_free(configs, count * sizeof(void *));
5298 for (i = 0; i < nchildren; i++)
5299 nvlist_free(tops[i]);
5300 kmem_free(tops, nchildren * sizeof(void *));
5301 nvlist_free(nvroot);
5306 spa_import_rootpool(const char *name)
5309 vdev_t *rvd, *bvd, *avd = NULL;
5310 nvlist_t *config, *nvtop;
5316 * Read the label from the boot device and generate a configuration.
5318 config = spa_generate_rootconf(name);
5320 mutex_enter(&spa_namespace_lock);
5321 if (config != NULL) {
5322 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5323 &pname) == 0 && strcmp(name, pname) == 0);
5324 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
5327 if ((spa = spa_lookup(pname)) != NULL) {
5329 * The pool could already be imported,
5330 * e.g., after reboot -r.
5332 if (spa->spa_state == POOL_STATE_ACTIVE) {
5333 mutex_exit(&spa_namespace_lock);
5334 nvlist_free(config);
5339 * Remove the existing root pool from the namespace so
5340 * that we can replace it with the correct config
5345 spa = spa_add(pname, config, NULL);
5348 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
5349 * via spa_version().
5351 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5352 &spa->spa_ubsync.ub_version) != 0)
5353 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5354 } else if ((spa = spa_lookup(name)) == NULL) {
5355 mutex_exit(&spa_namespace_lock);
5356 nvlist_free(config);
5357 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
5361 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
5363 spa->spa_is_root = B_TRUE;
5364 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5367 * Build up a vdev tree based on the boot device's label config.
5369 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5371 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5372 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5373 VDEV_ALLOC_ROOTPOOL);
5374 spa_config_exit(spa, SCL_ALL, FTAG);
5376 mutex_exit(&spa_namespace_lock);
5377 nvlist_free(config);
5378 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5383 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5385 spa_config_exit(spa, SCL_ALL, FTAG);
5386 mutex_exit(&spa_namespace_lock);
5388 nvlist_free(config);
5392 #endif /* illumos */
5393 #endif /* _KERNEL */
5396 * Import a non-root pool into the system.
5399 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5402 char *altroot = NULL;
5403 spa_load_state_t state = SPA_LOAD_IMPORT;
5404 zpool_load_policy_t policy;
5405 uint64_t mode = spa_mode_global;
5406 uint64_t readonly = B_FALSE;
5409 nvlist_t **spares, **l2cache;
5410 uint_t nspares, nl2cache;
5413 * If a pool with this name exists, return failure.
5415 mutex_enter(&spa_namespace_lock);
5416 if (spa_lookup(pool) != NULL) {
5417 mutex_exit(&spa_namespace_lock);
5418 return (SET_ERROR(EEXIST));
5422 * Create and initialize the spa structure.
5424 (void) nvlist_lookup_string(props,
5425 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5426 (void) nvlist_lookup_uint64(props,
5427 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5430 spa = spa_add(pool, config, altroot);
5431 spa->spa_import_flags = flags;
5434 * Verbatim import - Take a pool and insert it into the namespace
5435 * as if it had been loaded at boot.
5437 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5439 spa_configfile_set(spa, props, B_FALSE);
5441 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5442 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5443 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5444 mutex_exit(&spa_namespace_lock);
5448 spa_activate(spa, mode);
5451 * Don't start async tasks until we know everything is healthy.
5453 spa_async_suspend(spa);
5455 zpool_get_load_policy(config, &policy);
5456 if (policy.zlp_rewind & ZPOOL_DO_REWIND)
5457 state = SPA_LOAD_RECOVER;
5459 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5461 if (state != SPA_LOAD_RECOVER) {
5462 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5463 zfs_dbgmsg("spa_import: importing %s", pool);
5465 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5466 "(RECOVERY MODE)", pool, (longlong_t)policy.zlp_txg);
5468 error = spa_load_best(spa, state, policy.zlp_txg, policy.zlp_rewind);
5471 * Propagate anything learned while loading the pool and pass it
5472 * back to caller (i.e. rewind info, missing devices, etc).
5474 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5475 spa->spa_load_info) == 0);
5477 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5479 * Toss any existing sparelist, as it doesn't have any validity
5480 * anymore, and conflicts with spa_has_spare().
5482 if (spa->spa_spares.sav_config) {
5483 nvlist_free(spa->spa_spares.sav_config);
5484 spa->spa_spares.sav_config = NULL;
5485 spa_load_spares(spa);
5487 if (spa->spa_l2cache.sav_config) {
5488 nvlist_free(spa->spa_l2cache.sav_config);
5489 spa->spa_l2cache.sav_config = NULL;
5490 spa_load_l2cache(spa);
5493 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5496 error = spa_validate_aux(spa, nvroot, -1ULL,
5499 error = spa_validate_aux(spa, nvroot, -1ULL,
5500 VDEV_ALLOC_L2CACHE);
5501 spa_config_exit(spa, SCL_ALL, FTAG);
5504 spa_configfile_set(spa, props, B_FALSE);
5506 if (error != 0 || (props && spa_writeable(spa) &&
5507 (error = spa_prop_set(spa, props)))) {
5509 spa_deactivate(spa);
5511 mutex_exit(&spa_namespace_lock);
5515 spa_async_resume(spa);
5518 * Override any spares and level 2 cache devices as specified by
5519 * the user, as these may have correct device names/devids, etc.
5521 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5522 &spares, &nspares) == 0) {
5523 if (spa->spa_spares.sav_config)
5524 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5525 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5527 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5528 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5529 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5530 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5531 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5532 spa_load_spares(spa);
5533 spa_config_exit(spa, SCL_ALL, FTAG);
5534 spa->spa_spares.sav_sync = B_TRUE;
5536 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5537 &l2cache, &nl2cache) == 0) {
5538 if (spa->spa_l2cache.sav_config)
5539 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5540 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5542 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5543 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5544 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5545 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5546 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5547 spa_load_l2cache(spa);
5548 spa_config_exit(spa, SCL_ALL, FTAG);
5549 spa->spa_l2cache.sav_sync = B_TRUE;
5553 * Check for any removed devices.
5555 if (spa->spa_autoreplace) {
5556 spa_aux_check_removed(&spa->spa_spares);
5557 spa_aux_check_removed(&spa->spa_l2cache);
5560 if (spa_writeable(spa)) {
5562 * Update the config cache to include the newly-imported pool.
5564 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5568 * It's possible that the pool was expanded while it was exported.
5569 * We kick off an async task to handle this for us.
5571 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5573 spa_history_log_version(spa, "import");
5575 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5577 mutex_exit(&spa_namespace_lock);
5581 zvol_create_minors(pool);
5588 spa_tryimport(nvlist_t *tryconfig)
5590 nvlist_t *config = NULL;
5591 char *poolname, *cachefile;
5595 zpool_load_policy_t policy;
5597 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5600 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5604 * Create and initialize the spa structure.
5606 mutex_enter(&spa_namespace_lock);
5607 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5608 spa_activate(spa, FREAD);
5611 * Rewind pool if a max txg was provided.
5613 zpool_get_load_policy(spa->spa_config, &policy);
5614 if (policy.zlp_txg != UINT64_MAX) {
5615 spa->spa_load_max_txg = policy.zlp_txg;
5616 spa->spa_extreme_rewind = B_TRUE;
5617 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5618 poolname, (longlong_t)policy.zlp_txg);
5620 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5623 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5625 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5626 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5628 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5631 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5634 * If 'tryconfig' was at least parsable, return the current config.
5636 if (spa->spa_root_vdev != NULL) {
5637 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5638 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5640 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5642 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5643 spa->spa_uberblock.ub_timestamp) == 0);
5644 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5645 spa->spa_load_info) == 0);
5648 * If the bootfs property exists on this pool then we
5649 * copy it out so that external consumers can tell which
5650 * pools are bootable.
5652 if ((!error || error == EEXIST) && spa->spa_bootfs) {
5653 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5656 * We have to play games with the name since the
5657 * pool was opened as TRYIMPORT_NAME.
5659 if (dsl_dsobj_to_dsname(spa_name(spa),
5660 spa->spa_bootfs, tmpname) == 0) {
5662 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5664 cp = strchr(tmpname, '/');
5666 (void) strlcpy(dsname, tmpname,
5669 (void) snprintf(dsname, MAXPATHLEN,
5670 "%s/%s", poolname, ++cp);
5672 VERIFY(nvlist_add_string(config,
5673 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5674 kmem_free(dsname, MAXPATHLEN);
5676 kmem_free(tmpname, MAXPATHLEN);
5680 * Add the list of hot spares and level 2 cache devices.
5682 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5683 spa_add_spares(spa, config);
5684 spa_add_l2cache(spa, config);
5685 spa_config_exit(spa, SCL_CONFIG, FTAG);
5689 spa_deactivate(spa);
5691 mutex_exit(&spa_namespace_lock);
5697 * Pool export/destroy
5699 * The act of destroying or exporting a pool is very simple. We make sure there
5700 * is no more pending I/O and any references to the pool are gone. Then, we
5701 * update the pool state and sync all the labels to disk, removing the
5702 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5703 * we don't sync the labels or remove the configuration cache.
5706 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5707 boolean_t force, boolean_t hardforce)
5714 if (!(spa_mode_global & FWRITE))
5715 return (SET_ERROR(EROFS));
5717 mutex_enter(&spa_namespace_lock);
5718 if ((spa = spa_lookup(pool)) == NULL) {
5719 mutex_exit(&spa_namespace_lock);
5720 return (SET_ERROR(ENOENT));
5724 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5725 * reacquire the namespace lock, and see if we can export.
5727 spa_open_ref(spa, FTAG);
5728 mutex_exit(&spa_namespace_lock);
5729 spa_async_suspend(spa);
5730 mutex_enter(&spa_namespace_lock);
5731 spa_close(spa, FTAG);
5734 * The pool will be in core if it's openable,
5735 * in which case we can modify its state.
5737 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5740 * Objsets may be open only because they're dirty, so we
5741 * have to force it to sync before checking spa_refcnt.
5743 txg_wait_synced(spa->spa_dsl_pool, 0);
5744 spa_evicting_os_wait(spa);
5747 * A pool cannot be exported or destroyed if there are active
5748 * references. If we are resetting a pool, allow references by
5749 * fault injection handlers.
5751 if (!spa_refcount_zero(spa) ||
5752 (spa->spa_inject_ref != 0 &&
5753 new_state != POOL_STATE_UNINITIALIZED)) {
5754 spa_async_resume(spa);
5755 mutex_exit(&spa_namespace_lock);
5756 return (SET_ERROR(EBUSY));
5760 * A pool cannot be exported if it has an active shared spare.
5761 * This is to prevent other pools stealing the active spare
5762 * from an exported pool. At user's own will, such pool can
5763 * be forcedly exported.
5765 if (!force && new_state == POOL_STATE_EXPORTED &&
5766 spa_has_active_shared_spare(spa)) {
5767 spa_async_resume(spa);
5768 mutex_exit(&spa_namespace_lock);
5769 return (SET_ERROR(EXDEV));
5773 * We're about to export or destroy this pool. Make sure
5774 * we stop all initializtion activity here before we
5775 * set the spa_final_txg. This will ensure that all
5776 * dirty data resulting from the initialization is
5777 * committed to disk before we unload the pool.
5779 if (spa->spa_root_vdev != NULL) {
5780 vdev_initialize_stop_all(spa->spa_root_vdev,
5781 VDEV_INITIALIZE_ACTIVE);
5785 * We want this to be reflected on every label,
5786 * so mark them all dirty. spa_unload() will do the
5787 * final sync that pushes these changes out.
5789 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5790 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5791 spa->spa_state = new_state;
5792 spa->spa_final_txg = spa_last_synced_txg(spa) +
5794 vdev_config_dirty(spa->spa_root_vdev);
5795 spa_config_exit(spa, SCL_ALL, FTAG);
5799 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5801 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5803 spa_deactivate(spa);
5806 if (oldconfig && spa->spa_config)
5807 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5809 if (new_state != POOL_STATE_UNINITIALIZED) {
5811 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5814 mutex_exit(&spa_namespace_lock);
5820 * Destroy a storage pool.
5823 spa_destroy(char *pool)
5825 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5830 * Export a storage pool.
5833 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5834 boolean_t hardforce)
5836 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5841 * Similar to spa_export(), this unloads the spa_t without actually removing it
5842 * from the namespace in any way.
5845 spa_reset(char *pool)
5847 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5852 * ==========================================================================
5853 * Device manipulation
5854 * ==========================================================================
5858 * Add a device to a storage pool.
5861 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5865 vdev_t *rvd = spa->spa_root_vdev;
5867 nvlist_t **spares, **l2cache;
5868 uint_t nspares, nl2cache;
5870 ASSERT(spa_writeable(spa));
5872 txg = spa_vdev_enter(spa);
5874 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5875 VDEV_ALLOC_ADD)) != 0)
5876 return (spa_vdev_exit(spa, NULL, txg, error));
5878 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
5880 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5884 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5888 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5889 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5891 if (vd->vdev_children != 0 &&
5892 (error = vdev_create(vd, txg, B_FALSE)) != 0)
5893 return (spa_vdev_exit(spa, vd, txg, error));
5896 * We must validate the spares and l2cache devices after checking the
5897 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5899 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5900 return (spa_vdev_exit(spa, vd, txg, error));
5903 * If we are in the middle of a device removal, we can only add
5904 * devices which match the existing devices in the pool.
5905 * If we are in the middle of a removal, or have some indirect
5906 * vdevs, we can not add raidz toplevels.
5908 if (spa->spa_vdev_removal != NULL ||
5909 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5910 for (int c = 0; c < vd->vdev_children; c++) {
5911 tvd = vd->vdev_child[c];
5912 if (spa->spa_vdev_removal != NULL &&
5913 tvd->vdev_ashift != spa->spa_max_ashift) {
5914 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5916 /* Fail if top level vdev is raidz */
5917 if (tvd->vdev_ops == &vdev_raidz_ops) {
5918 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5921 * Need the top level mirror to be
5922 * a mirror of leaf vdevs only
5924 if (tvd->vdev_ops == &vdev_mirror_ops) {
5925 for (uint64_t cid = 0;
5926 cid < tvd->vdev_children; cid++) {
5927 vdev_t *cvd = tvd->vdev_child[cid];
5928 if (!cvd->vdev_ops->vdev_op_leaf) {
5929 return (spa_vdev_exit(spa, vd,
5937 for (int c = 0; c < vd->vdev_children; c++) {
5940 * Set the vdev id to the first hole, if one exists.
5942 for (id = 0; id < rvd->vdev_children; id++) {
5943 if (rvd->vdev_child[id]->vdev_ishole) {
5944 vdev_free(rvd->vdev_child[id]);
5948 tvd = vd->vdev_child[c];
5949 vdev_remove_child(vd, tvd);
5951 vdev_add_child(rvd, tvd);
5952 vdev_config_dirty(tvd);
5956 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5957 ZPOOL_CONFIG_SPARES);
5958 spa_load_spares(spa);
5959 spa->spa_spares.sav_sync = B_TRUE;
5962 if (nl2cache != 0) {
5963 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5964 ZPOOL_CONFIG_L2CACHE);
5965 spa_load_l2cache(spa);
5966 spa->spa_l2cache.sav_sync = B_TRUE;
5970 * We have to be careful when adding new vdevs to an existing pool.
5971 * If other threads start allocating from these vdevs before we
5972 * sync the config cache, and we lose power, then upon reboot we may
5973 * fail to open the pool because there are DVAs that the config cache
5974 * can't translate. Therefore, we first add the vdevs without
5975 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5976 * and then let spa_config_update() initialize the new metaslabs.
5978 * spa_load() checks for added-but-not-initialized vdevs, so that
5979 * if we lose power at any point in this sequence, the remaining
5980 * steps will be completed the next time we load the pool.
5982 (void) spa_vdev_exit(spa, vd, txg, 0);
5984 mutex_enter(&spa_namespace_lock);
5985 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5986 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5987 mutex_exit(&spa_namespace_lock);
5993 * Attach a device to a mirror. The arguments are the path to any device
5994 * in the mirror, and the nvroot for the new device. If the path specifies
5995 * a device that is not mirrored, we automatically insert the mirror vdev.
5997 * If 'replacing' is specified, the new device is intended to replace the
5998 * existing device; in this case the two devices are made into their own
5999 * mirror using the 'replacing' vdev, which is functionally identical to
6000 * the mirror vdev (it actually reuses all the same ops) but has a few
6001 * extra rules: you can't attach to it after it's been created, and upon
6002 * completion of resilvering, the first disk (the one being replaced)
6003 * is automatically detached.
6006 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
6008 uint64_t txg, dtl_max_txg;
6009 vdev_t *rvd = spa->spa_root_vdev;
6010 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
6012 char *oldvdpath, *newvdpath;
6016 ASSERT(spa_writeable(spa));
6018 txg = spa_vdev_enter(spa);
6020 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
6022 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6023 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6024 error = (spa_has_checkpoint(spa)) ?
6025 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6026 return (spa_vdev_exit(spa, NULL, txg, error));
6029 if (spa->spa_vdev_removal != NULL)
6030 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6033 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6035 if (!oldvd->vdev_ops->vdev_op_leaf)
6036 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6038 pvd = oldvd->vdev_parent;
6040 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
6041 VDEV_ALLOC_ATTACH)) != 0)
6042 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6044 if (newrootvd->vdev_children != 1)
6045 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6047 newvd = newrootvd->vdev_child[0];
6049 if (!newvd->vdev_ops->vdev_op_leaf)
6050 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
6052 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
6053 return (spa_vdev_exit(spa, newrootvd, txg, error));
6056 * Spares can't replace logs
6058 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
6059 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6063 * For attach, the only allowable parent is a mirror or the root
6066 if (pvd->vdev_ops != &vdev_mirror_ops &&
6067 pvd->vdev_ops != &vdev_root_ops)
6068 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6070 pvops = &vdev_mirror_ops;
6073 * Active hot spares can only be replaced by inactive hot
6076 if (pvd->vdev_ops == &vdev_spare_ops &&
6077 oldvd->vdev_isspare &&
6078 !spa_has_spare(spa, newvd->vdev_guid))
6079 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6082 * If the source is a hot spare, and the parent isn't already a
6083 * spare, then we want to create a new hot spare. Otherwise, we
6084 * want to create a replacing vdev. The user is not allowed to
6085 * attach to a spared vdev child unless the 'isspare' state is
6086 * the same (spare replaces spare, non-spare replaces
6089 if (pvd->vdev_ops == &vdev_replacing_ops &&
6090 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6091 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6092 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6093 newvd->vdev_isspare != oldvd->vdev_isspare) {
6094 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6097 if (newvd->vdev_isspare)
6098 pvops = &vdev_spare_ops;
6100 pvops = &vdev_replacing_ops;
6104 * Make sure the new device is big enough.
6106 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6107 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6110 * The new device cannot have a higher alignment requirement
6111 * than the top-level vdev.
6113 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6114 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
6117 * If this is an in-place replacement, update oldvd's path and devid
6118 * to make it distinguishable from newvd, and unopenable from now on.
6120 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6121 spa_strfree(oldvd->vdev_path);
6122 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6124 (void) sprintf(oldvd->vdev_path, "%s/%s",
6125 newvd->vdev_path, "old");
6126 if (oldvd->vdev_devid != NULL) {
6127 spa_strfree(oldvd->vdev_devid);
6128 oldvd->vdev_devid = NULL;
6132 /* mark the device being resilvered */
6133 newvd->vdev_resilver_txg = txg;
6136 * If the parent is not a mirror, or if we're replacing, insert the new
6137 * mirror/replacing/spare vdev above oldvd.
6139 if (pvd->vdev_ops != pvops)
6140 pvd = vdev_add_parent(oldvd, pvops);
6142 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6143 ASSERT(pvd->vdev_ops == pvops);
6144 ASSERT(oldvd->vdev_parent == pvd);
6147 * Extract the new device from its root and add it to pvd.
6149 vdev_remove_child(newrootvd, newvd);
6150 newvd->vdev_id = pvd->vdev_children;
6151 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6152 vdev_add_child(pvd, newvd);
6154 tvd = newvd->vdev_top;
6155 ASSERT(pvd->vdev_top == tvd);
6156 ASSERT(tvd->vdev_parent == rvd);
6158 vdev_config_dirty(tvd);
6161 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6162 * for any dmu_sync-ed blocks. It will propagate upward when
6163 * spa_vdev_exit() calls vdev_dtl_reassess().
6165 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6167 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
6168 dtl_max_txg - TXG_INITIAL);
6170 if (newvd->vdev_isspare) {
6171 spa_spare_activate(newvd);
6172 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6175 oldvdpath = spa_strdup(oldvd->vdev_path);
6176 newvdpath = spa_strdup(newvd->vdev_path);
6177 newvd_isspare = newvd->vdev_isspare;
6180 * Mark newvd's DTL dirty in this txg.
6182 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6185 * Schedule the resilver to restart in the future. We do this to
6186 * ensure that dmu_sync-ed blocks have been stitched into the
6187 * respective datasets.
6189 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
6191 if (spa->spa_bootfs)
6192 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6194 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6199 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6201 spa_history_log_internal(spa, "vdev attach", NULL,
6202 "%s vdev=%s %s vdev=%s",
6203 replacing && newvd_isspare ? "spare in" :
6204 replacing ? "replace" : "attach", newvdpath,
6205 replacing ? "for" : "to", oldvdpath);
6207 spa_strfree(oldvdpath);
6208 spa_strfree(newvdpath);
6214 * Detach a device from a mirror or replacing vdev.
6216 * If 'replace_done' is specified, only detach if the parent
6217 * is a replacing vdev.
6220 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6224 vdev_t *rvd = spa->spa_root_vdev;
6225 vdev_t *vd, *pvd, *cvd, *tvd;
6226 boolean_t unspare = B_FALSE;
6227 uint64_t unspare_guid = 0;
6230 ASSERT(spa_writeable(spa));
6232 txg = spa_vdev_enter(spa);
6234 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6237 * Besides being called directly from the userland through the
6238 * ioctl interface, spa_vdev_detach() can be potentially called
6239 * at the end of spa_vdev_resilver_done().
6241 * In the regular case, when we have a checkpoint this shouldn't
6242 * happen as we never empty the DTLs of a vdev during the scrub
6243 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6244 * should never get here when we have a checkpoint.
6246 * That said, even in a case when we checkpoint the pool exactly
6247 * as spa_vdev_resilver_done() calls this function everything
6248 * should be fine as the resilver will return right away.
6250 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6251 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6252 error = (spa_has_checkpoint(spa)) ?
6253 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6254 return (spa_vdev_exit(spa, NULL, txg, error));
6258 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6260 if (!vd->vdev_ops->vdev_op_leaf)
6261 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6263 pvd = vd->vdev_parent;
6266 * If the parent/child relationship is not as expected, don't do it.
6267 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6268 * vdev that's replacing B with C. The user's intent in replacing
6269 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6270 * the replace by detaching C, the expected behavior is to end up
6271 * M(A,B). But suppose that right after deciding to detach C,
6272 * the replacement of B completes. We would have M(A,C), and then
6273 * ask to detach C, which would leave us with just A -- not what
6274 * the user wanted. To prevent this, we make sure that the
6275 * parent/child relationship hasn't changed -- in this example,
6276 * that C's parent is still the replacing vdev R.
6278 if (pvd->vdev_guid != pguid && pguid != 0)
6279 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6282 * Only 'replacing' or 'spare' vdevs can be replaced.
6284 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6285 pvd->vdev_ops != &vdev_spare_ops)
6286 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6288 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6289 spa_version(spa) >= SPA_VERSION_SPARES);
6292 * Only mirror, replacing, and spare vdevs support detach.
6294 if (pvd->vdev_ops != &vdev_replacing_ops &&
6295 pvd->vdev_ops != &vdev_mirror_ops &&
6296 pvd->vdev_ops != &vdev_spare_ops)
6297 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6300 * If this device has the only valid copy of some data,
6301 * we cannot safely detach it.
6303 if (vdev_dtl_required(vd))
6304 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6306 ASSERT(pvd->vdev_children >= 2);
6309 * If we are detaching the second disk from a replacing vdev, then
6310 * check to see if we changed the original vdev's path to have "/old"
6311 * at the end in spa_vdev_attach(). If so, undo that change now.
6313 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6314 vd->vdev_path != NULL) {
6315 size_t len = strlen(vd->vdev_path);
6317 for (int c = 0; c < pvd->vdev_children; c++) {
6318 cvd = pvd->vdev_child[c];
6320 if (cvd == vd || cvd->vdev_path == NULL)
6323 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6324 strcmp(cvd->vdev_path + len, "/old") == 0) {
6325 spa_strfree(cvd->vdev_path);
6326 cvd->vdev_path = spa_strdup(vd->vdev_path);
6333 * If we are detaching the original disk from a spare, then it implies
6334 * that the spare should become a real disk, and be removed from the
6335 * active spare list for the pool.
6337 if (pvd->vdev_ops == &vdev_spare_ops &&
6339 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6343 * Erase the disk labels so the disk can be used for other things.
6344 * This must be done after all other error cases are handled,
6345 * but before we disembowel vd (so we can still do I/O to it).
6346 * But if we can't do it, don't treat the error as fatal --
6347 * it may be that the unwritability of the disk is the reason
6348 * it's being detached!
6350 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6353 * Remove vd from its parent and compact the parent's children.
6355 vdev_remove_child(pvd, vd);
6356 vdev_compact_children(pvd);
6359 * Remember one of the remaining children so we can get tvd below.
6361 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6364 * If we need to remove the remaining child from the list of hot spares,
6365 * do it now, marking the vdev as no longer a spare in the process.
6366 * We must do this before vdev_remove_parent(), because that can
6367 * change the GUID if it creates a new toplevel GUID. For a similar
6368 * reason, we must remove the spare now, in the same txg as the detach;
6369 * otherwise someone could attach a new sibling, change the GUID, and
6370 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6373 ASSERT(cvd->vdev_isspare);
6374 spa_spare_remove(cvd);
6375 unspare_guid = cvd->vdev_guid;
6376 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6377 cvd->vdev_unspare = B_TRUE;
6381 * If the parent mirror/replacing vdev only has one child,
6382 * the parent is no longer needed. Remove it from the tree.
6384 if (pvd->vdev_children == 1) {
6385 if (pvd->vdev_ops == &vdev_spare_ops)
6386 cvd->vdev_unspare = B_FALSE;
6387 vdev_remove_parent(cvd);
6392 * We don't set tvd until now because the parent we just removed
6393 * may have been the previous top-level vdev.
6395 tvd = cvd->vdev_top;
6396 ASSERT(tvd->vdev_parent == rvd);
6399 * Reevaluate the parent vdev state.
6401 vdev_propagate_state(cvd);
6404 * If the 'autoexpand' property is set on the pool then automatically
6405 * try to expand the size of the pool. For example if the device we
6406 * just detached was smaller than the others, it may be possible to
6407 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6408 * first so that we can obtain the updated sizes of the leaf vdevs.
6410 if (spa->spa_autoexpand) {
6412 vdev_expand(tvd, txg);
6415 vdev_config_dirty(tvd);
6418 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6419 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6420 * But first make sure we're not on any *other* txg's DTL list, to
6421 * prevent vd from being accessed after it's freed.
6423 vdpath = spa_strdup(vd->vdev_path);
6424 for (int t = 0; t < TXG_SIZE; t++)
6425 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6426 vd->vdev_detached = B_TRUE;
6427 vdev_dirty(tvd, VDD_DTL, vd, txg);
6429 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6431 /* hang on to the spa before we release the lock */
6432 spa_open_ref(spa, FTAG);
6434 error = spa_vdev_exit(spa, vd, txg, 0);
6436 spa_history_log_internal(spa, "detach", NULL,
6438 spa_strfree(vdpath);
6441 * If this was the removal of the original device in a hot spare vdev,
6442 * then we want to go through and remove the device from the hot spare
6443 * list of every other pool.
6446 spa_t *altspa = NULL;
6448 mutex_enter(&spa_namespace_lock);
6449 while ((altspa = spa_next(altspa)) != NULL) {
6450 if (altspa->spa_state != POOL_STATE_ACTIVE ||
6454 spa_open_ref(altspa, FTAG);
6455 mutex_exit(&spa_namespace_lock);
6456 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6457 mutex_enter(&spa_namespace_lock);
6458 spa_close(altspa, FTAG);
6460 mutex_exit(&spa_namespace_lock);
6462 /* search the rest of the vdevs for spares to remove */
6463 spa_vdev_resilver_done(spa);
6466 /* all done with the spa; OK to release */
6467 mutex_enter(&spa_namespace_lock);
6468 spa_close(spa, FTAG);
6469 mutex_exit(&spa_namespace_lock);
6475 spa_vdev_initialize(spa_t *spa, uint64_t guid, uint64_t cmd_type)
6478 * We hold the namespace lock through the whole function
6479 * to prevent any changes to the pool while we're starting or
6480 * stopping initialization. The config and state locks are held so that
6481 * we can properly assess the vdev state before we commit to
6482 * the initializing operation.
6484 mutex_enter(&spa_namespace_lock);
6485 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6487 /* Look up vdev and ensure it's a leaf. */
6488 vdev_t *vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6489 if (vd == NULL || vd->vdev_detached) {
6490 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6491 mutex_exit(&spa_namespace_lock);
6492 return (SET_ERROR(ENODEV));
6493 } else if (!vd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(vd)) {
6494 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6495 mutex_exit(&spa_namespace_lock);
6496 return (SET_ERROR(EINVAL));
6497 } else if (!vdev_writeable(vd)) {
6498 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6499 mutex_exit(&spa_namespace_lock);
6500 return (SET_ERROR(EROFS));
6502 mutex_enter(&vd->vdev_initialize_lock);
6503 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6506 * When we activate an initialize action we check to see
6507 * if the vdev_initialize_thread is NULL. We do this instead
6508 * of using the vdev_initialize_state since there might be
6509 * a previous initialization process which has completed but
6510 * the thread is not exited.
6512 if (cmd_type == POOL_INITIALIZE_DO &&
6513 (vd->vdev_initialize_thread != NULL ||
6514 vd->vdev_top->vdev_removing)) {
6515 mutex_exit(&vd->vdev_initialize_lock);
6516 mutex_exit(&spa_namespace_lock);
6517 return (SET_ERROR(EBUSY));
6518 } else if (cmd_type == POOL_INITIALIZE_CANCEL &&
6519 (vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE &&
6520 vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED)) {
6521 mutex_exit(&vd->vdev_initialize_lock);
6522 mutex_exit(&spa_namespace_lock);
6523 return (SET_ERROR(ESRCH));
6524 } else if (cmd_type == POOL_INITIALIZE_SUSPEND &&
6525 vd->vdev_initialize_state != VDEV_INITIALIZE_ACTIVE) {
6526 mutex_exit(&vd->vdev_initialize_lock);
6527 mutex_exit(&spa_namespace_lock);
6528 return (SET_ERROR(ESRCH));
6532 case POOL_INITIALIZE_DO:
6533 vdev_initialize(vd);
6535 case POOL_INITIALIZE_CANCEL:
6536 vdev_initialize_stop(vd, VDEV_INITIALIZE_CANCELED);
6538 case POOL_INITIALIZE_SUSPEND:
6539 vdev_initialize_stop(vd, VDEV_INITIALIZE_SUSPENDED);
6542 panic("invalid cmd_type %llu", (unsigned long long)cmd_type);
6544 mutex_exit(&vd->vdev_initialize_lock);
6546 /* Sync out the initializing state */
6547 txg_wait_synced(spa->spa_dsl_pool, 0);
6548 mutex_exit(&spa_namespace_lock);
6555 * Split a set of devices from their mirrors, and create a new pool from them.
6558 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6559 nvlist_t *props, boolean_t exp)
6562 uint64_t txg, *glist;
6564 uint_t c, children, lastlog;
6565 nvlist_t **child, *nvl, *tmp;
6567 char *altroot = NULL;
6568 vdev_t *rvd, **vml = NULL; /* vdev modify list */
6569 boolean_t activate_slog;
6571 ASSERT(spa_writeable(spa));
6573 txg = spa_vdev_enter(spa);
6575 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6576 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6577 error = (spa_has_checkpoint(spa)) ?
6578 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6579 return (spa_vdev_exit(spa, NULL, txg, error));
6582 /* clear the log and flush everything up to now */
6583 activate_slog = spa_passivate_log(spa);
6584 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6585 error = spa_reset_logs(spa);
6586 txg = spa_vdev_config_enter(spa);
6589 spa_activate_log(spa);
6592 return (spa_vdev_exit(spa, NULL, txg, error));
6594 /* check new spa name before going any further */
6595 if (spa_lookup(newname) != NULL)
6596 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
6599 * scan through all the children to ensure they're all mirrors
6601 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
6602 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
6604 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6606 /* first, check to ensure we've got the right child count */
6607 rvd = spa->spa_root_vdev;
6609 for (c = 0; c < rvd->vdev_children; c++) {
6610 vdev_t *vd = rvd->vdev_child[c];
6612 /* don't count the holes & logs as children */
6613 if (vd->vdev_islog || !vdev_is_concrete(vd)) {
6621 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
6622 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6624 /* next, ensure no spare or cache devices are part of the split */
6625 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
6626 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
6627 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6629 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
6630 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
6632 /* then, loop over each vdev and validate it */
6633 for (c = 0; c < children; c++) {
6634 uint64_t is_hole = 0;
6636 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
6640 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
6641 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
6644 error = SET_ERROR(EINVAL);
6649 /* which disk is going to be split? */
6650 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
6652 error = SET_ERROR(EINVAL);
6656 /* look it up in the spa */
6657 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
6658 if (vml[c] == NULL) {
6659 error = SET_ERROR(ENODEV);
6663 /* make sure there's nothing stopping the split */
6664 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
6665 vml[c]->vdev_islog ||
6666 !vdev_is_concrete(vml[c]) ||
6667 vml[c]->vdev_isspare ||
6668 vml[c]->vdev_isl2cache ||
6669 !vdev_writeable(vml[c]) ||
6670 vml[c]->vdev_children != 0 ||
6671 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
6672 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
6673 error = SET_ERROR(EINVAL);
6677 if (vdev_dtl_required(vml[c])) {
6678 error = SET_ERROR(EBUSY);
6682 /* we need certain info from the top level */
6683 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
6684 vml[c]->vdev_top->vdev_ms_array) == 0);
6685 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
6686 vml[c]->vdev_top->vdev_ms_shift) == 0);
6687 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
6688 vml[c]->vdev_top->vdev_asize) == 0);
6689 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
6690 vml[c]->vdev_top->vdev_ashift) == 0);
6692 /* transfer per-vdev ZAPs */
6693 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
6694 VERIFY0(nvlist_add_uint64(child[c],
6695 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
6697 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
6698 VERIFY0(nvlist_add_uint64(child[c],
6699 ZPOOL_CONFIG_VDEV_TOP_ZAP,
6700 vml[c]->vdev_parent->vdev_top_zap));
6704 kmem_free(vml, children * sizeof (vdev_t *));
6705 kmem_free(glist, children * sizeof (uint64_t));
6706 return (spa_vdev_exit(spa, NULL, txg, error));
6709 /* stop writers from using the disks */
6710 for (c = 0; c < children; c++) {
6712 vml[c]->vdev_offline = B_TRUE;
6714 vdev_reopen(spa->spa_root_vdev);
6717 * Temporarily record the splitting vdevs in the spa config. This
6718 * will disappear once the config is regenerated.
6720 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6721 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
6722 glist, children) == 0);
6723 kmem_free(glist, children * sizeof (uint64_t));
6725 mutex_enter(&spa->spa_props_lock);
6726 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
6728 mutex_exit(&spa->spa_props_lock);
6729 spa->spa_config_splitting = nvl;
6730 vdev_config_dirty(spa->spa_root_vdev);
6732 /* configure and create the new pool */
6733 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
6734 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6735 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
6736 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6737 spa_version(spa)) == 0);
6738 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
6739 spa->spa_config_txg) == 0);
6740 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
6741 spa_generate_guid(NULL)) == 0);
6742 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
6743 (void) nvlist_lookup_string(props,
6744 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6746 /* add the new pool to the namespace */
6747 newspa = spa_add(newname, config, altroot);
6748 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
6749 newspa->spa_config_txg = spa->spa_config_txg;
6750 spa_set_log_state(newspa, SPA_LOG_CLEAR);
6752 /* release the spa config lock, retaining the namespace lock */
6753 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6755 if (zio_injection_enabled)
6756 zio_handle_panic_injection(spa, FTAG, 1);
6758 spa_activate(newspa, spa_mode_global);
6759 spa_async_suspend(newspa);
6761 for (c = 0; c < children; c++) {
6762 if (vml[c] != NULL) {
6764 * Temporarily stop the initializing activity. We set
6765 * the state to ACTIVE so that we know to resume
6766 * the initializing once the split has completed.
6768 mutex_enter(&vml[c]->vdev_initialize_lock);
6769 vdev_initialize_stop(vml[c], VDEV_INITIALIZE_ACTIVE);
6770 mutex_exit(&vml[c]->vdev_initialize_lock);
6775 /* mark that we are creating new spa by splitting */
6776 newspa->spa_splitting_newspa = B_TRUE;
6778 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
6780 /* create the new pool from the disks of the original pool */
6781 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
6783 newspa->spa_splitting_newspa = B_FALSE;
6788 /* if that worked, generate a real config for the new pool */
6789 if (newspa->spa_root_vdev != NULL) {
6790 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
6791 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6792 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
6793 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
6794 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
6799 if (props != NULL) {
6800 spa_configfile_set(newspa, props, B_FALSE);
6801 error = spa_prop_set(newspa, props);
6806 /* flush everything */
6807 txg = spa_vdev_config_enter(newspa);
6808 vdev_config_dirty(newspa->spa_root_vdev);
6809 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
6811 if (zio_injection_enabled)
6812 zio_handle_panic_injection(spa, FTAG, 2);
6814 spa_async_resume(newspa);
6816 /* finally, update the original pool's config */
6817 txg = spa_vdev_config_enter(spa);
6818 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
6819 error = dmu_tx_assign(tx, TXG_WAIT);
6822 for (c = 0; c < children; c++) {
6823 if (vml[c] != NULL) {
6826 spa_history_log_internal(spa, "detach", tx,
6827 "vdev=%s", vml[c]->vdev_path);
6832 spa->spa_avz_action = AVZ_ACTION_REBUILD;
6833 vdev_config_dirty(spa->spa_root_vdev);
6834 spa->spa_config_splitting = NULL;
6838 (void) spa_vdev_exit(spa, NULL, txg, 0);
6840 if (zio_injection_enabled)
6841 zio_handle_panic_injection(spa, FTAG, 3);
6843 /* split is complete; log a history record */
6844 spa_history_log_internal(newspa, "split", NULL,
6845 "from pool %s", spa_name(spa));
6847 kmem_free(vml, children * sizeof (vdev_t *));
6849 /* if we're not going to mount the filesystems in userland, export */
6851 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
6858 spa_deactivate(newspa);
6861 txg = spa_vdev_config_enter(spa);
6863 /* re-online all offlined disks */
6864 for (c = 0; c < children; c++) {
6866 vml[c]->vdev_offline = B_FALSE;
6869 /* restart initializing disks as necessary */
6870 spa_async_request(spa, SPA_ASYNC_INITIALIZE_RESTART);
6872 vdev_reopen(spa->spa_root_vdev);
6874 nvlist_free(spa->spa_config_splitting);
6875 spa->spa_config_splitting = NULL;
6876 (void) spa_vdev_exit(spa, NULL, txg, error);
6878 kmem_free(vml, children * sizeof (vdev_t *));
6883 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6884 * currently spared, so we can detach it.
6887 spa_vdev_resilver_done_hunt(vdev_t *vd)
6889 vdev_t *newvd, *oldvd;
6891 for (int c = 0; c < vd->vdev_children; c++) {
6892 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
6898 * Check for a completed replacement. We always consider the first
6899 * vdev in the list to be the oldest vdev, and the last one to be
6900 * the newest (see spa_vdev_attach() for how that works). In
6901 * the case where the newest vdev is faulted, we will not automatically
6902 * remove it after a resilver completes. This is OK as it will require
6903 * user intervention to determine which disk the admin wishes to keep.
6905 if (vd->vdev_ops == &vdev_replacing_ops) {
6906 ASSERT(vd->vdev_children > 1);
6908 newvd = vd->vdev_child[vd->vdev_children - 1];
6909 oldvd = vd->vdev_child[0];
6911 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
6912 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6913 !vdev_dtl_required(oldvd))
6918 * Check for a completed resilver with the 'unspare' flag set.
6919 * Also potentially update faulted state.
6921 if (vd->vdev_ops == &vdev_spare_ops) {
6922 vdev_t *first = vd->vdev_child[0];
6923 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
6925 if (last->vdev_unspare) {
6928 } else if (first->vdev_unspare) {
6935 if (oldvd != NULL &&
6936 vdev_dtl_empty(newvd, DTL_MISSING) &&
6937 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6938 !vdev_dtl_required(oldvd))
6941 vdev_propagate_state(vd);
6944 * If there are more than two spares attached to a disk,
6945 * and those spares are not required, then we want to
6946 * attempt to free them up now so that they can be used
6947 * by other pools. Once we're back down to a single
6948 * disk+spare, we stop removing them.
6950 if (vd->vdev_children > 2) {
6951 newvd = vd->vdev_child[1];
6953 if (newvd->vdev_isspare && last->vdev_isspare &&
6954 vdev_dtl_empty(last, DTL_MISSING) &&
6955 vdev_dtl_empty(last, DTL_OUTAGE) &&
6956 !vdev_dtl_required(newvd))
6965 spa_vdev_resilver_done(spa_t *spa)
6967 vdev_t *vd, *pvd, *ppvd;
6968 uint64_t guid, sguid, pguid, ppguid;
6970 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6972 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
6973 pvd = vd->vdev_parent;
6974 ppvd = pvd->vdev_parent;
6975 guid = vd->vdev_guid;
6976 pguid = pvd->vdev_guid;
6977 ppguid = ppvd->vdev_guid;
6980 * If we have just finished replacing a hot spared device, then
6981 * we need to detach the parent's first child (the original hot
6984 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
6985 ppvd->vdev_children == 2) {
6986 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
6987 sguid = ppvd->vdev_child[1]->vdev_guid;
6989 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
6991 spa_config_exit(spa, SCL_ALL, FTAG);
6992 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
6994 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
6996 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6999 spa_config_exit(spa, SCL_ALL, FTAG);
7003 * Update the stored path or FRU for this vdev.
7006 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
7010 boolean_t sync = B_FALSE;
7012 ASSERT(spa_writeable(spa));
7014 spa_vdev_state_enter(spa, SCL_ALL);
7016 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
7017 return (spa_vdev_state_exit(spa, NULL, ENOENT));
7019 if (!vd->vdev_ops->vdev_op_leaf)
7020 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
7023 if (strcmp(value, vd->vdev_path) != 0) {
7024 spa_strfree(vd->vdev_path);
7025 vd->vdev_path = spa_strdup(value);
7029 if (vd->vdev_fru == NULL) {
7030 vd->vdev_fru = spa_strdup(value);
7032 } else if (strcmp(value, vd->vdev_fru) != 0) {
7033 spa_strfree(vd->vdev_fru);
7034 vd->vdev_fru = spa_strdup(value);
7039 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
7043 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
7045 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
7049 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
7051 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
7055 * ==========================================================================
7057 * ==========================================================================
7060 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
7062 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7064 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7065 return (SET_ERROR(EBUSY));
7067 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
7071 spa_scan_stop(spa_t *spa)
7073 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7074 if (dsl_scan_resilvering(spa->spa_dsl_pool))
7075 return (SET_ERROR(EBUSY));
7076 return (dsl_scan_cancel(spa->spa_dsl_pool));
7080 spa_scan(spa_t *spa, pool_scan_func_t func)
7082 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
7084 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
7085 return (SET_ERROR(ENOTSUP));
7088 * If a resilver was requested, but there is no DTL on a
7089 * writeable leaf device, we have nothing to do.
7091 if (func == POOL_SCAN_RESILVER &&
7092 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
7093 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
7097 return (dsl_scan(spa->spa_dsl_pool, func));
7101 * ==========================================================================
7102 * SPA async task processing
7103 * ==========================================================================
7107 spa_async_remove(spa_t *spa, vdev_t *vd)
7109 if (vd->vdev_remove_wanted) {
7110 vd->vdev_remove_wanted = B_FALSE;
7111 vd->vdev_delayed_close = B_FALSE;
7112 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
7115 * We want to clear the stats, but we don't want to do a full
7116 * vdev_clear() as that will cause us to throw away
7117 * degraded/faulted state as well as attempt to reopen the
7118 * device, all of which is a waste.
7120 vd->vdev_stat.vs_read_errors = 0;
7121 vd->vdev_stat.vs_write_errors = 0;
7122 vd->vdev_stat.vs_checksum_errors = 0;
7124 vdev_state_dirty(vd->vdev_top);
7125 /* Tell userspace that the vdev is gone. */
7126 zfs_post_remove(spa, vd);
7129 for (int c = 0; c < vd->vdev_children; c++)
7130 spa_async_remove(spa, vd->vdev_child[c]);
7134 spa_async_probe(spa_t *spa, vdev_t *vd)
7136 if (vd->vdev_probe_wanted) {
7137 vd->vdev_probe_wanted = B_FALSE;
7138 vdev_reopen(vd); /* vdev_open() does the actual probe */
7141 for (int c = 0; c < vd->vdev_children; c++)
7142 spa_async_probe(spa, vd->vdev_child[c]);
7146 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
7152 if (!spa->spa_autoexpand)
7155 for (int c = 0; c < vd->vdev_children; c++) {
7156 vdev_t *cvd = vd->vdev_child[c];
7157 spa_async_autoexpand(spa, cvd);
7160 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
7163 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
7164 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
7166 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7167 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
7169 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
7170 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
7173 kmem_free(physpath, MAXPATHLEN);
7177 spa_async_thread(void *arg)
7179 spa_t *spa = (spa_t *)arg;
7182 ASSERT(spa->spa_sync_on);
7184 mutex_enter(&spa->spa_async_lock);
7185 tasks = spa->spa_async_tasks;
7186 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
7187 mutex_exit(&spa->spa_async_lock);
7190 * See if the config needs to be updated.
7192 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7193 uint64_t old_space, new_space;
7195 mutex_enter(&spa_namespace_lock);
7196 old_space = metaslab_class_get_space(spa_normal_class(spa));
7197 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7198 new_space = metaslab_class_get_space(spa_normal_class(spa));
7199 mutex_exit(&spa_namespace_lock);
7202 * If the pool grew as a result of the config update,
7203 * then log an internal history event.
7205 if (new_space != old_space) {
7206 spa_history_log_internal(spa, "vdev online", NULL,
7207 "pool '%s' size: %llu(+%llu)",
7208 spa_name(spa), new_space, new_space - old_space);
7212 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7213 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7214 spa_async_autoexpand(spa, spa->spa_root_vdev);
7215 spa_config_exit(spa, SCL_CONFIG, FTAG);
7219 * See if any devices need to be probed.
7221 if (tasks & SPA_ASYNC_PROBE) {
7222 spa_vdev_state_enter(spa, SCL_NONE);
7223 spa_async_probe(spa, spa->spa_root_vdev);
7224 (void) spa_vdev_state_exit(spa, NULL, 0);
7228 * If any devices are done replacing, detach them.
7230 if (tasks & SPA_ASYNC_RESILVER_DONE)
7231 spa_vdev_resilver_done(spa);
7234 * Kick off a resilver.
7236 if (tasks & SPA_ASYNC_RESILVER)
7237 dsl_resilver_restart(spa->spa_dsl_pool, 0);
7239 if (tasks & SPA_ASYNC_INITIALIZE_RESTART) {
7240 mutex_enter(&spa_namespace_lock);
7241 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7242 vdev_initialize_restart(spa->spa_root_vdev);
7243 spa_config_exit(spa, SCL_CONFIG, FTAG);
7244 mutex_exit(&spa_namespace_lock);
7248 * Let the world know that we're done.
7250 mutex_enter(&spa->spa_async_lock);
7251 spa->spa_async_thread = NULL;
7252 cv_broadcast(&spa->spa_async_cv);
7253 mutex_exit(&spa->spa_async_lock);
7258 spa_async_thread_vd(void *arg)
7263 mutex_enter(&spa->spa_async_lock);
7264 tasks = spa->spa_async_tasks;
7266 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
7267 mutex_exit(&spa->spa_async_lock);
7270 * See if any devices need to be marked REMOVED.
7272 if (tasks & SPA_ASYNC_REMOVE) {
7273 spa_vdev_state_enter(spa, SCL_NONE);
7274 spa_async_remove(spa, spa->spa_root_vdev);
7275 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7276 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7277 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7278 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7279 (void) spa_vdev_state_exit(spa, NULL, 0);
7283 * Let the world know that we're done.
7285 mutex_enter(&spa->spa_async_lock);
7286 tasks = spa->spa_async_tasks;
7287 if ((tasks & SPA_ASYNC_REMOVE) != 0)
7289 spa->spa_async_thread_vd = NULL;
7290 cv_broadcast(&spa->spa_async_cv);
7291 mutex_exit(&spa->spa_async_lock);
7296 spa_async_suspend(spa_t *spa)
7298 mutex_enter(&spa->spa_async_lock);
7299 spa->spa_async_suspended++;
7300 while (spa->spa_async_thread != NULL ||
7301 spa->spa_async_thread_vd != NULL)
7302 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
7303 mutex_exit(&spa->spa_async_lock);
7305 spa_vdev_remove_suspend(spa);
7307 zthr_t *condense_thread = spa->spa_condense_zthr;
7308 if (condense_thread != NULL && zthr_isrunning(condense_thread))
7309 VERIFY0(zthr_cancel(condense_thread));
7311 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7312 if (discard_thread != NULL && zthr_isrunning(discard_thread))
7313 VERIFY0(zthr_cancel(discard_thread));
7317 spa_async_resume(spa_t *spa)
7319 mutex_enter(&spa->spa_async_lock);
7320 ASSERT(spa->spa_async_suspended != 0);
7321 spa->spa_async_suspended--;
7322 mutex_exit(&spa->spa_async_lock);
7323 spa_restart_removal(spa);
7325 zthr_t *condense_thread = spa->spa_condense_zthr;
7326 if (condense_thread != NULL && !zthr_isrunning(condense_thread))
7327 zthr_resume(condense_thread);
7329 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7330 if (discard_thread != NULL && !zthr_isrunning(discard_thread))
7331 zthr_resume(discard_thread);
7335 spa_async_tasks_pending(spa_t *spa)
7337 uint_t non_config_tasks;
7339 boolean_t config_task_suspended;
7341 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
7343 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
7344 if (spa->spa_ccw_fail_time == 0) {
7345 config_task_suspended = B_FALSE;
7347 config_task_suspended =
7348 (gethrtime() - spa->spa_ccw_fail_time) <
7349 (zfs_ccw_retry_interval * NANOSEC);
7352 return (non_config_tasks || (config_task && !config_task_suspended));
7356 spa_async_dispatch(spa_t *spa)
7358 mutex_enter(&spa->spa_async_lock);
7359 if (spa_async_tasks_pending(spa) &&
7360 !spa->spa_async_suspended &&
7361 spa->spa_async_thread == NULL &&
7363 spa->spa_async_thread = thread_create(NULL, 0,
7364 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
7365 mutex_exit(&spa->spa_async_lock);
7369 spa_async_dispatch_vd(spa_t *spa)
7371 mutex_enter(&spa->spa_async_lock);
7372 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
7373 !spa->spa_async_suspended &&
7374 spa->spa_async_thread_vd == NULL &&
7376 spa->spa_async_thread_vd = thread_create(NULL, 0,
7377 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
7378 mutex_exit(&spa->spa_async_lock);
7382 spa_async_request(spa_t *spa, int task)
7384 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
7385 mutex_enter(&spa->spa_async_lock);
7386 spa->spa_async_tasks |= task;
7387 mutex_exit(&spa->spa_async_lock);
7388 spa_async_dispatch_vd(spa);
7392 * ==========================================================================
7393 * SPA syncing routines
7394 * ==========================================================================
7398 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7401 bpobj_enqueue(bpo, bp, tx);
7406 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7410 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
7411 BP_GET_PSIZE(bp), zio->io_flags));
7416 * Note: this simple function is not inlined to make it easier to dtrace the
7417 * amount of time spent syncing frees.
7420 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
7422 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7423 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
7424 VERIFY(zio_wait(zio) == 0);
7428 * Note: this simple function is not inlined to make it easier to dtrace the
7429 * amount of time spent syncing deferred frees.
7432 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
7434 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7435 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
7436 spa_free_sync_cb, zio, tx), ==, 0);
7437 VERIFY0(zio_wait(zio));
7442 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
7444 char *packed = NULL;
7449 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
7452 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
7453 * information. This avoids the dmu_buf_will_dirty() path and
7454 * saves us a pre-read to get data we don't actually care about.
7456 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
7457 packed = kmem_alloc(bufsize, KM_SLEEP);
7459 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
7461 bzero(packed + nvsize, bufsize - nvsize);
7463 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
7465 kmem_free(packed, bufsize);
7467 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
7468 dmu_buf_will_dirty(db, tx);
7469 *(uint64_t *)db->db_data = nvsize;
7470 dmu_buf_rele(db, FTAG);
7474 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
7475 const char *config, const char *entry)
7485 * Update the MOS nvlist describing the list of available devices.
7486 * spa_validate_aux() will have already made sure this nvlist is
7487 * valid and the vdevs are labeled appropriately.
7489 if (sav->sav_object == 0) {
7490 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
7491 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
7492 sizeof (uint64_t), tx);
7493 VERIFY(zap_update(spa->spa_meta_objset,
7494 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
7495 &sav->sav_object, tx) == 0);
7498 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7499 if (sav->sav_count == 0) {
7500 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
7502 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
7503 for (i = 0; i < sav->sav_count; i++)
7504 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
7505 B_FALSE, VDEV_CONFIG_L2CACHE);
7506 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
7507 sav->sav_count) == 0);
7508 for (i = 0; i < sav->sav_count; i++)
7509 nvlist_free(list[i]);
7510 kmem_free(list, sav->sav_count * sizeof (void *));
7513 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
7514 nvlist_free(nvroot);
7516 sav->sav_sync = B_FALSE;
7520 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
7521 * The all-vdev ZAP must be empty.
7524 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
7526 spa_t *spa = vd->vdev_spa;
7527 if (vd->vdev_top_zap != 0) {
7528 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7529 vd->vdev_top_zap, tx));
7531 if (vd->vdev_leaf_zap != 0) {
7532 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7533 vd->vdev_leaf_zap, tx));
7535 for (uint64_t i = 0; i < vd->vdev_children; i++) {
7536 spa_avz_build(vd->vdev_child[i], avz, tx);
7541 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
7546 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
7547 * its config may not be dirty but we still need to build per-vdev ZAPs.
7548 * Similarly, if the pool is being assembled (e.g. after a split), we
7549 * need to rebuild the AVZ although the config may not be dirty.
7551 if (list_is_empty(&spa->spa_config_dirty_list) &&
7552 spa->spa_avz_action == AVZ_ACTION_NONE)
7555 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7557 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
7558 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
7559 spa->spa_all_vdev_zaps != 0);
7561 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
7562 /* Make and build the new AVZ */
7563 uint64_t new_avz = zap_create(spa->spa_meta_objset,
7564 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
7565 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
7567 /* Diff old AVZ with new one */
7571 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7572 spa->spa_all_vdev_zaps);
7573 zap_cursor_retrieve(&zc, &za) == 0;
7574 zap_cursor_advance(&zc)) {
7575 uint64_t vdzap = za.za_first_integer;
7576 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
7579 * ZAP is listed in old AVZ but not in new one;
7582 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
7587 zap_cursor_fini(&zc);
7589 /* Destroy the old AVZ */
7590 VERIFY0(zap_destroy(spa->spa_meta_objset,
7591 spa->spa_all_vdev_zaps, tx));
7593 /* Replace the old AVZ in the dir obj with the new one */
7594 VERIFY0(zap_update(spa->spa_meta_objset,
7595 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
7596 sizeof (new_avz), 1, &new_avz, tx));
7598 spa->spa_all_vdev_zaps = new_avz;
7599 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
7603 /* Walk through the AVZ and destroy all listed ZAPs */
7604 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7605 spa->spa_all_vdev_zaps);
7606 zap_cursor_retrieve(&zc, &za) == 0;
7607 zap_cursor_advance(&zc)) {
7608 uint64_t zap = za.za_first_integer;
7609 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
7612 zap_cursor_fini(&zc);
7614 /* Destroy and unlink the AVZ itself */
7615 VERIFY0(zap_destroy(spa->spa_meta_objset,
7616 spa->spa_all_vdev_zaps, tx));
7617 VERIFY0(zap_remove(spa->spa_meta_objset,
7618 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
7619 spa->spa_all_vdev_zaps = 0;
7622 if (spa->spa_all_vdev_zaps == 0) {
7623 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
7624 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
7625 DMU_POOL_VDEV_ZAP_MAP, tx);
7627 spa->spa_avz_action = AVZ_ACTION_NONE;
7629 /* Create ZAPs for vdevs that don't have them. */
7630 vdev_construct_zaps(spa->spa_root_vdev, tx);
7632 config = spa_config_generate(spa, spa->spa_root_vdev,
7633 dmu_tx_get_txg(tx), B_FALSE);
7636 * If we're upgrading the spa version then make sure that
7637 * the config object gets updated with the correct version.
7639 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
7640 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7641 spa->spa_uberblock.ub_version);
7643 spa_config_exit(spa, SCL_STATE, FTAG);
7645 nvlist_free(spa->spa_config_syncing);
7646 spa->spa_config_syncing = config;
7648 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
7652 spa_sync_version(void *arg, dmu_tx_t *tx)
7654 uint64_t *versionp = arg;
7655 uint64_t version = *versionp;
7656 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7659 * Setting the version is special cased when first creating the pool.
7661 ASSERT(tx->tx_txg != TXG_INITIAL);
7663 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
7664 ASSERT(version >= spa_version(spa));
7666 spa->spa_uberblock.ub_version = version;
7667 vdev_config_dirty(spa->spa_root_vdev);
7668 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
7672 * Set zpool properties.
7675 spa_sync_props(void *arg, dmu_tx_t *tx)
7677 nvlist_t *nvp = arg;
7678 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7679 objset_t *mos = spa->spa_meta_objset;
7680 nvpair_t *elem = NULL;
7682 mutex_enter(&spa->spa_props_lock);
7684 while ((elem = nvlist_next_nvpair(nvp, elem))) {
7686 char *strval, *fname;
7688 const char *propname;
7689 zprop_type_t proptype;
7692 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
7693 case ZPOOL_PROP_INVAL:
7695 * We checked this earlier in spa_prop_validate().
7697 ASSERT(zpool_prop_feature(nvpair_name(elem)));
7699 fname = strchr(nvpair_name(elem), '@') + 1;
7700 VERIFY0(zfeature_lookup_name(fname, &fid));
7702 spa_feature_enable(spa, fid, tx);
7703 spa_history_log_internal(spa, "set", tx,
7704 "%s=enabled", nvpair_name(elem));
7707 case ZPOOL_PROP_VERSION:
7708 intval = fnvpair_value_uint64(elem);
7710 * The version is synced seperatly before other
7711 * properties and should be correct by now.
7713 ASSERT3U(spa_version(spa), >=, intval);
7716 case ZPOOL_PROP_ALTROOT:
7718 * 'altroot' is a non-persistent property. It should
7719 * have been set temporarily at creation or import time.
7721 ASSERT(spa->spa_root != NULL);
7724 case ZPOOL_PROP_READONLY:
7725 case ZPOOL_PROP_CACHEFILE:
7727 * 'readonly' and 'cachefile' are also non-persisitent
7731 case ZPOOL_PROP_COMMENT:
7732 strval = fnvpair_value_string(elem);
7733 if (spa->spa_comment != NULL)
7734 spa_strfree(spa->spa_comment);
7735 spa->spa_comment = spa_strdup(strval);
7737 * We need to dirty the configuration on all the vdevs
7738 * so that their labels get updated. It's unnecessary
7739 * to do this for pool creation since the vdev's
7740 * configuratoin has already been dirtied.
7742 if (tx->tx_txg != TXG_INITIAL)
7743 vdev_config_dirty(spa->spa_root_vdev);
7744 spa_history_log_internal(spa, "set", tx,
7745 "%s=%s", nvpair_name(elem), strval);
7749 * Set pool property values in the poolprops mos object.
7751 if (spa->spa_pool_props_object == 0) {
7752 spa->spa_pool_props_object =
7753 zap_create_link(mos, DMU_OT_POOL_PROPS,
7754 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
7758 /* normalize the property name */
7759 propname = zpool_prop_to_name(prop);
7760 proptype = zpool_prop_get_type(prop);
7762 if (nvpair_type(elem) == DATA_TYPE_STRING) {
7763 ASSERT(proptype == PROP_TYPE_STRING);
7764 strval = fnvpair_value_string(elem);
7765 VERIFY0(zap_update(mos,
7766 spa->spa_pool_props_object, propname,
7767 1, strlen(strval) + 1, strval, tx));
7768 spa_history_log_internal(spa, "set", tx,
7769 "%s=%s", nvpair_name(elem), strval);
7770 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
7771 intval = fnvpair_value_uint64(elem);
7773 if (proptype == PROP_TYPE_INDEX) {
7775 VERIFY0(zpool_prop_index_to_string(
7776 prop, intval, &unused));
7778 VERIFY0(zap_update(mos,
7779 spa->spa_pool_props_object, propname,
7780 8, 1, &intval, tx));
7781 spa_history_log_internal(spa, "set", tx,
7782 "%s=%lld", nvpair_name(elem), intval);
7784 ASSERT(0); /* not allowed */
7788 case ZPOOL_PROP_DELEGATION:
7789 spa->spa_delegation = intval;
7791 case ZPOOL_PROP_BOOTFS:
7792 spa->spa_bootfs = intval;
7794 case ZPOOL_PROP_FAILUREMODE:
7795 spa->spa_failmode = intval;
7797 case ZPOOL_PROP_AUTOEXPAND:
7798 spa->spa_autoexpand = intval;
7799 if (tx->tx_txg != TXG_INITIAL)
7800 spa_async_request(spa,
7801 SPA_ASYNC_AUTOEXPAND);
7803 case ZPOOL_PROP_DEDUPDITTO:
7804 spa->spa_dedup_ditto = intval;
7813 mutex_exit(&spa->spa_props_lock);
7817 * Perform one-time upgrade on-disk changes. spa_version() does not
7818 * reflect the new version this txg, so there must be no changes this
7819 * txg to anything that the upgrade code depends on after it executes.
7820 * Therefore this must be called after dsl_pool_sync() does the sync
7824 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
7826 dsl_pool_t *dp = spa->spa_dsl_pool;
7828 ASSERT(spa->spa_sync_pass == 1);
7830 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7832 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
7833 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
7834 dsl_pool_create_origin(dp, tx);
7836 /* Keeping the origin open increases spa_minref */
7837 spa->spa_minref += 3;
7840 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
7841 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
7842 dsl_pool_upgrade_clones(dp, tx);
7845 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
7846 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
7847 dsl_pool_upgrade_dir_clones(dp, tx);
7849 /* Keeping the freedir open increases spa_minref */
7850 spa->spa_minref += 3;
7853 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
7854 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7855 spa_feature_create_zap_objects(spa, tx);
7859 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7860 * when possibility to use lz4 compression for metadata was added
7861 * Old pools that have this feature enabled must be upgraded to have
7862 * this feature active
7864 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7865 boolean_t lz4_en = spa_feature_is_enabled(spa,
7866 SPA_FEATURE_LZ4_COMPRESS);
7867 boolean_t lz4_ac = spa_feature_is_active(spa,
7868 SPA_FEATURE_LZ4_COMPRESS);
7870 if (lz4_en && !lz4_ac)
7871 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
7875 * If we haven't written the salt, do so now. Note that the
7876 * feature may not be activated yet, but that's fine since
7877 * the presence of this ZAP entry is backwards compatible.
7879 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7880 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
7881 VERIFY0(zap_add(spa->spa_meta_objset,
7882 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
7883 sizeof (spa->spa_cksum_salt.zcs_bytes),
7884 spa->spa_cksum_salt.zcs_bytes, tx));
7887 rrw_exit(&dp->dp_config_rwlock, FTAG);
7891 vdev_indirect_state_sync_verify(vdev_t *vd)
7893 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7894 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
7896 if (vd->vdev_ops == &vdev_indirect_ops) {
7897 ASSERT(vim != NULL);
7898 ASSERT(vib != NULL);
7901 if (vdev_obsolete_sm_object(vd) != 0) {
7902 ASSERT(vd->vdev_obsolete_sm != NULL);
7903 ASSERT(vd->vdev_removing ||
7904 vd->vdev_ops == &vdev_indirect_ops);
7905 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
7906 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
7908 ASSERT3U(vdev_obsolete_sm_object(vd), ==,
7909 space_map_object(vd->vdev_obsolete_sm));
7910 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
7911 space_map_allocated(vd->vdev_obsolete_sm));
7913 ASSERT(vd->vdev_obsolete_segments != NULL);
7916 * Since frees / remaps to an indirect vdev can only
7917 * happen in syncing context, the obsolete segments
7918 * tree must be empty when we start syncing.
7920 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
7924 * Sync the specified transaction group. New blocks may be dirtied as
7925 * part of the process, so we iterate until it converges.
7928 spa_sync(spa_t *spa, uint64_t txg)
7930 dsl_pool_t *dp = spa->spa_dsl_pool;
7931 objset_t *mos = spa->spa_meta_objset;
7932 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
7933 vdev_t *rvd = spa->spa_root_vdev;
7937 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
7938 zfs_vdev_queue_depth_pct / 100;
7940 VERIFY(spa_writeable(spa));
7943 * Wait for i/os issued in open context that need to complete
7944 * before this txg syncs.
7946 (void) zio_wait(spa->spa_txg_zio[txg & TXG_MASK]);
7947 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL,
7951 * Lock out configuration changes.
7953 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7955 spa->spa_syncing_txg = txg;
7956 spa->spa_sync_pass = 0;
7958 for (int i = 0; i < spa->spa_alloc_count; i++) {
7959 mutex_enter(&spa->spa_alloc_locks[i]);
7960 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
7961 mutex_exit(&spa->spa_alloc_locks[i]);
7965 * If there are any pending vdev state changes, convert them
7966 * into config changes that go out with this transaction group.
7968 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7969 while (list_head(&spa->spa_state_dirty_list) != NULL) {
7971 * We need the write lock here because, for aux vdevs,
7972 * calling vdev_config_dirty() modifies sav_config.
7973 * This is ugly and will become unnecessary when we
7974 * eliminate the aux vdev wart by integrating all vdevs
7975 * into the root vdev tree.
7977 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7978 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
7979 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
7980 vdev_state_clean(vd);
7981 vdev_config_dirty(vd);
7983 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7984 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7986 spa_config_exit(spa, SCL_STATE, FTAG);
7988 tx = dmu_tx_create_assigned(dp, txg);
7990 spa->spa_sync_starttime = gethrtime();
7992 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
7993 spa->spa_sync_starttime + spa->spa_deadman_synctime));
7994 #else /* !illumos */
7996 callout_schedule(&spa->spa_deadman_cycid,
7997 hz * spa->spa_deadman_synctime / NANOSEC);
7999 #endif /* illumos */
8002 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
8003 * set spa_deflate if we have no raid-z vdevs.
8005 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
8006 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
8009 for (i = 0; i < rvd->vdev_children; i++) {
8010 vd = rvd->vdev_child[i];
8011 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
8014 if (i == rvd->vdev_children) {
8015 spa->spa_deflate = TRUE;
8016 VERIFY(0 == zap_add(spa->spa_meta_objset,
8017 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
8018 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
8023 * Set the top-level vdev's max queue depth. Evaluate each
8024 * top-level's async write queue depth in case it changed.
8025 * The max queue depth will not change in the middle of syncing
8028 uint64_t slots_per_allocator = 0;
8029 for (int c = 0; c < rvd->vdev_children; c++) {
8030 vdev_t *tvd = rvd->vdev_child[c];
8031 metaslab_group_t *mg = tvd->vdev_mg;
8033 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
8034 !metaslab_group_initialized(mg))
8038 * It is safe to do a lock-free check here because only async
8039 * allocations look at mg_max_alloc_queue_depth, and async
8040 * allocations all happen from spa_sync().
8042 for (int i = 0; i < spa->spa_alloc_count; i++)
8043 ASSERT0(refcount_count(&(mg->mg_alloc_queue_depth[i])));
8044 mg->mg_max_alloc_queue_depth = max_queue_depth;
8046 for (int i = 0; i < spa->spa_alloc_count; i++) {
8047 mg->mg_cur_max_alloc_queue_depth[i] =
8048 zfs_vdev_def_queue_depth;
8050 slots_per_allocator += zfs_vdev_def_queue_depth;
8052 metaslab_class_t *mc = spa_normal_class(spa);
8053 for (int i = 0; i < spa->spa_alloc_count; i++) {
8054 ASSERT0(refcount_count(&mc->mc_alloc_slots[i]));
8055 mc->mc_alloc_max_slots[i] = slots_per_allocator;
8057 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
8059 for (int c = 0; c < rvd->vdev_children; c++) {
8060 vdev_t *vd = rvd->vdev_child[c];
8061 vdev_indirect_state_sync_verify(vd);
8063 if (vdev_indirect_should_condense(vd)) {
8064 spa_condense_indirect_start_sync(vd, tx);
8070 * Iterate to convergence.
8073 int pass = ++spa->spa_sync_pass;
8075 spa_sync_config_object(spa, tx);
8076 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
8077 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
8078 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
8079 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
8080 spa_errlog_sync(spa, txg);
8081 dsl_pool_sync(dp, txg);
8083 if (pass < zfs_sync_pass_deferred_free) {
8084 spa_sync_frees(spa, free_bpl, tx);
8087 * We can not defer frees in pass 1, because
8088 * we sync the deferred frees later in pass 1.
8090 ASSERT3U(pass, >, 1);
8091 bplist_iterate(free_bpl, bpobj_enqueue_cb,
8092 &spa->spa_deferred_bpobj, tx);
8096 dsl_scan_sync(dp, tx);
8098 if (spa->spa_vdev_removal != NULL)
8101 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
8106 spa_sync_upgrades(spa, tx);
8108 spa->spa_uberblock.ub_rootbp.blk_birth);
8110 * Note: We need to check if the MOS is dirty
8111 * because we could have marked the MOS dirty
8112 * without updating the uberblock (e.g. if we
8113 * have sync tasks but no dirty user data). We
8114 * need to check the uberblock's rootbp because
8115 * it is updated if we have synced out dirty
8116 * data (though in this case the MOS will most
8117 * likely also be dirty due to second order
8118 * effects, we don't want to rely on that here).
8120 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
8121 !dmu_objset_is_dirty(mos, txg)) {
8123 * Nothing changed on the first pass,
8124 * therefore this TXG is a no-op. Avoid
8125 * syncing deferred frees, so that we
8126 * can keep this TXG as a no-op.
8128 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
8130 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8131 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
8132 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks,
8136 spa_sync_deferred_frees(spa, tx);
8139 } while (dmu_objset_is_dirty(mos, txg));
8141 if (!list_is_empty(&spa->spa_config_dirty_list)) {
8143 * Make sure that the number of ZAPs for all the vdevs matches
8144 * the number of ZAPs in the per-vdev ZAP list. This only gets
8145 * called if the config is dirty; otherwise there may be
8146 * outstanding AVZ operations that weren't completed in
8147 * spa_sync_config_object.
8149 uint64_t all_vdev_zap_entry_count;
8150 ASSERT0(zap_count(spa->spa_meta_objset,
8151 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
8152 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
8153 all_vdev_zap_entry_count);
8156 if (spa->spa_vdev_removal != NULL) {
8157 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
8161 * Rewrite the vdev configuration (which includes the uberblock)
8162 * to commit the transaction group.
8164 * If there are no dirty vdevs, we sync the uberblock to a few
8165 * random top-level vdevs that are known to be visible in the
8166 * config cache (see spa_vdev_add() for a complete description).
8167 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
8171 * We hold SCL_STATE to prevent vdev open/close/etc.
8172 * while we're attempting to write the vdev labels.
8174 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
8176 if (list_is_empty(&spa->spa_config_dirty_list)) {
8177 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
8179 int children = rvd->vdev_children;
8180 int c0 = spa_get_random(children);
8182 for (int c = 0; c < children; c++) {
8183 vd = rvd->vdev_child[(c0 + c) % children];
8185 /* Stop when revisiting the first vdev */
8186 if (c > 0 && svd[0] == vd)
8189 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
8190 !vdev_is_concrete(vd))
8193 svd[svdcount++] = vd;
8194 if (svdcount == SPA_SYNC_MIN_VDEVS)
8197 error = vdev_config_sync(svd, svdcount, txg);
8199 error = vdev_config_sync(rvd->vdev_child,
8200 rvd->vdev_children, txg);
8204 spa->spa_last_synced_guid = rvd->vdev_guid;
8206 spa_config_exit(spa, SCL_STATE, FTAG);
8210 zio_suspend(spa, NULL);
8211 zio_resume_wait(spa);
8216 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
8217 #else /* !illumos */
8219 callout_drain(&spa->spa_deadman_cycid);
8221 #endif /* illumos */
8224 * Clear the dirty config list.
8226 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
8227 vdev_config_clean(vd);
8230 * Now that the new config has synced transactionally,
8231 * let it become visible to the config cache.
8233 if (spa->spa_config_syncing != NULL) {
8234 spa_config_set(spa, spa->spa_config_syncing);
8235 spa->spa_config_txg = txg;
8236 spa->spa_config_syncing = NULL;
8239 dsl_pool_sync_done(dp, txg);
8241 for (int i = 0; i < spa->spa_alloc_count; i++) {
8242 mutex_enter(&spa->spa_alloc_locks[i]);
8243 VERIFY0(avl_numnodes(&spa->spa_alloc_trees[i]));
8244 mutex_exit(&spa->spa_alloc_locks[i]);
8248 * Update usable space statistics.
8250 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
8252 vdev_sync_done(vd, txg);
8254 spa_update_dspace(spa);
8257 * It had better be the case that we didn't dirty anything
8258 * since vdev_config_sync().
8260 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8261 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8262 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
8264 while (zfs_pause_spa_sync)
8267 spa->spa_sync_pass = 0;
8270 * Update the last synced uberblock here. We want to do this at
8271 * the end of spa_sync() so that consumers of spa_last_synced_txg()
8272 * will be guaranteed that all the processing associated with
8273 * that txg has been completed.
8275 spa->spa_ubsync = spa->spa_uberblock;
8276 spa_config_exit(spa, SCL_CONFIG, FTAG);
8278 spa_handle_ignored_writes(spa);
8281 * If any async tasks have been requested, kick them off.
8283 spa_async_dispatch(spa);
8284 spa_async_dispatch_vd(spa);
8288 * Sync all pools. We don't want to hold the namespace lock across these
8289 * operations, so we take a reference on the spa_t and drop the lock during the
8293 spa_sync_allpools(void)
8296 mutex_enter(&spa_namespace_lock);
8297 while ((spa = spa_next(spa)) != NULL) {
8298 if (spa_state(spa) != POOL_STATE_ACTIVE ||
8299 !spa_writeable(spa) || spa_suspended(spa))
8301 spa_open_ref(spa, FTAG);
8302 mutex_exit(&spa_namespace_lock);
8303 txg_wait_synced(spa_get_dsl(spa), 0);
8304 mutex_enter(&spa_namespace_lock);
8305 spa_close(spa, FTAG);
8307 mutex_exit(&spa_namespace_lock);
8311 * ==========================================================================
8312 * Miscellaneous routines
8313 * ==========================================================================
8317 * Remove all pools in the system.
8325 * Remove all cached state. All pools should be closed now,
8326 * so every spa in the AVL tree should be unreferenced.
8328 mutex_enter(&spa_namespace_lock);
8329 while ((spa = spa_next(NULL)) != NULL) {
8331 * Stop async tasks. The async thread may need to detach
8332 * a device that's been replaced, which requires grabbing
8333 * spa_namespace_lock, so we must drop it here.
8335 spa_open_ref(spa, FTAG);
8336 mutex_exit(&spa_namespace_lock);
8337 spa_async_suspend(spa);
8338 mutex_enter(&spa_namespace_lock);
8339 spa_close(spa, FTAG);
8341 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
8343 spa_deactivate(spa);
8347 mutex_exit(&spa_namespace_lock);
8351 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
8356 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
8360 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
8361 vd = spa->spa_l2cache.sav_vdevs[i];
8362 if (vd->vdev_guid == guid)
8366 for (i = 0; i < spa->spa_spares.sav_count; i++) {
8367 vd = spa->spa_spares.sav_vdevs[i];
8368 if (vd->vdev_guid == guid)
8377 spa_upgrade(spa_t *spa, uint64_t version)
8379 ASSERT(spa_writeable(spa));
8381 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
8384 * This should only be called for a non-faulted pool, and since a
8385 * future version would result in an unopenable pool, this shouldn't be
8388 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
8389 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
8391 spa->spa_uberblock.ub_version = version;
8392 vdev_config_dirty(spa->spa_root_vdev);
8394 spa_config_exit(spa, SCL_ALL, FTAG);
8396 txg_wait_synced(spa_get_dsl(spa), 0);
8400 spa_has_spare(spa_t *spa, uint64_t guid)
8404 spa_aux_vdev_t *sav = &spa->spa_spares;
8406 for (i = 0; i < sav->sav_count; i++)
8407 if (sav->sav_vdevs[i]->vdev_guid == guid)
8410 for (i = 0; i < sav->sav_npending; i++) {
8411 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
8412 &spareguid) == 0 && spareguid == guid)
8420 * Check if a pool has an active shared spare device.
8421 * Note: reference count of an active spare is 2, as a spare and as a replace
8424 spa_has_active_shared_spare(spa_t *spa)
8428 spa_aux_vdev_t *sav = &spa->spa_spares;
8430 for (i = 0; i < sav->sav_count; i++) {
8431 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
8432 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
8441 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8443 sysevent_t *ev = NULL;
8445 sysevent_attr_list_t *attr = NULL;
8446 sysevent_value_t value;
8448 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
8452 value.value_type = SE_DATA_TYPE_STRING;
8453 value.value.sv_string = spa_name(spa);
8454 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
8457 value.value_type = SE_DATA_TYPE_UINT64;
8458 value.value.sv_uint64 = spa_guid(spa);
8459 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
8463 value.value_type = SE_DATA_TYPE_UINT64;
8464 value.value.sv_uint64 = vd->vdev_guid;
8465 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
8469 if (vd->vdev_path) {
8470 value.value_type = SE_DATA_TYPE_STRING;
8471 value.value.sv_string = vd->vdev_path;
8472 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
8473 &value, SE_SLEEP) != 0)
8478 if (hist_nvl != NULL) {
8479 fnvlist_merge((nvlist_t *)attr, hist_nvl);
8482 if (sysevent_attach_attributes(ev, attr) != 0)
8488 sysevent_free_attr(attr);
8495 spa_event_post(sysevent_t *ev)
8500 (void) log_sysevent(ev, SE_SLEEP, &eid);
8506 spa_event_discard(sysevent_t *ev)
8514 * Post a sysevent corresponding to the given event. The 'name' must be one of
8515 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
8516 * filled in from the spa and (optionally) the vdev and history nvl. This
8517 * doesn't do anything in the userland libzpool, as we don't want consumers to
8518 * misinterpret ztest or zdb as real changes.
8521 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8523 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));