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 2017 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/metaslab.h>
59 #include <sys/metaslab_impl.h>
60 #include <sys/uberblock_impl.h>
63 #include <sys/bpobj.h>
64 #include <sys/dmu_traverse.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/unique.h>
67 #include <sys/dsl_pool.h>
68 #include <sys/dsl_dataset.h>
69 #include <sys/dsl_dir.h>
70 #include <sys/dsl_prop.h>
71 #include <sys/dsl_synctask.h>
72 #include <sys/fs/zfs.h>
74 #include <sys/callb.h>
75 #include <sys/spa_boot.h>
76 #include <sys/zfs_ioctl.h>
77 #include <sys/dsl_scan.h>
78 #include <sys/dmu_send.h>
79 #include <sys/dsl_destroy.h>
80 #include <sys/dsl_userhold.h>
81 #include <sys/zfeature.h>
83 #include <sys/trim_map.h>
87 #include <sys/callb.h>
88 #include <sys/cpupart.h>
93 #include "zfs_comutil.h"
95 /* Check hostid on import? */
96 static int check_hostid = 1;
99 * The interval, in seconds, at which failed configuration cache file writes
102 int zfs_ccw_retry_interval = 300;
104 SYSCTL_DECL(_vfs_zfs);
105 SYSCTL_INT(_vfs_zfs, OID_AUTO, check_hostid, CTLFLAG_RWTUN, &check_hostid, 0,
106 "Check hostid on import?");
107 TUNABLE_INT("vfs.zfs.ccw_retry_interval", &zfs_ccw_retry_interval);
108 SYSCTL_INT(_vfs_zfs, OID_AUTO, ccw_retry_interval, CTLFLAG_RW,
109 &zfs_ccw_retry_interval, 0,
110 "Configuration cache file write, retry after failure, interval (seconds)");
112 typedef enum zti_modes {
113 ZTI_MODE_FIXED, /* value is # of threads (min 1) */
114 ZTI_MODE_BATCH, /* cpu-intensive; value is ignored */
115 ZTI_MODE_NULL, /* don't create a taskq */
119 #define ZTI_P(n, q) { ZTI_MODE_FIXED, (n), (q) }
120 #define ZTI_BATCH { ZTI_MODE_BATCH, 0, 1 }
121 #define ZTI_NULL { ZTI_MODE_NULL, 0, 0 }
123 #define ZTI_N(n) ZTI_P(n, 1)
124 #define ZTI_ONE ZTI_N(1)
126 typedef struct zio_taskq_info {
127 zti_modes_t zti_mode;
132 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
133 "issue", "issue_high", "intr", "intr_high"
137 * This table defines the taskq settings for each ZFS I/O type. When
138 * initializing a pool, we use this table to create an appropriately sized
139 * taskq. Some operations are low volume and therefore have a small, static
140 * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
141 * macros. Other operations process a large amount of data; the ZTI_BATCH
142 * macro causes us to create a taskq oriented for throughput. Some operations
143 * are so high frequency and short-lived that the taskq itself can become a a
144 * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
145 * additional degree of parallelism specified by the number of threads per-
146 * taskq and the number of taskqs; when dispatching an event in this case, the
147 * particular taskq is chosen at random.
149 * The different taskq priorities are to handle the different contexts (issue
150 * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
151 * need to be handled with minimum delay.
153 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
154 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
155 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* NULL */
156 { ZTI_N(8), ZTI_NULL, ZTI_P(12, 8), ZTI_NULL }, /* READ */
157 { ZTI_BATCH, ZTI_N(5), ZTI_N(8), ZTI_N(5) }, /* WRITE */
158 { ZTI_P(12, 8), ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* FREE */
159 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* CLAIM */
160 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL }, /* IOCTL */
163 static void spa_sync_version(void *arg, dmu_tx_t *tx);
164 static void spa_sync_props(void *arg, dmu_tx_t *tx);
165 static boolean_t spa_has_active_shared_spare(spa_t *spa);
166 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
167 static void spa_vdev_resilver_done(spa_t *spa);
169 uint_t zio_taskq_batch_pct = 75; /* 1 thread per cpu in pset */
171 id_t zio_taskq_psrset_bind = PS_NONE;
174 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
175 uint_t zio_taskq_basedc = 80; /* base duty cycle */
178 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
179 extern int zfs_sync_pass_deferred_free;
182 * Report any spa_load_verify errors found, but do not fail spa_load.
183 * This is used by zdb to analyze non-idle pools.
185 boolean_t spa_load_verify_dryrun = B_FALSE;
188 * This (illegal) pool name is used when temporarily importing a spa_t in order
189 * to get the vdev stats associated with the imported devices.
191 #define TRYIMPORT_NAME "$import"
194 * For debugging purposes: print out vdev tree during pool import.
196 int spa_load_print_vdev_tree = B_FALSE;
199 * A non-zero value for zfs_max_missing_tvds means that we allow importing
200 * pools with missing top-level vdevs. This is strictly intended for advanced
201 * pool recovery cases since missing data is almost inevitable. Pools with
202 * missing devices can only be imported read-only for safety reasons, and their
203 * fail-mode will be automatically set to "continue".
205 * With 1 missing vdev we should be able to import the pool and mount all
206 * datasets. User data that was not modified after the missing device has been
207 * added should be recoverable. This means that snapshots created prior to the
208 * addition of that device should be completely intact.
210 * With 2 missing vdevs, some datasets may fail to mount since there are
211 * dataset statistics that are stored as regular metadata. Some data might be
212 * recoverable if those vdevs were added recently.
214 * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
215 * may be missing entirely. Chances of data recovery are very low. Note that
216 * there are also risks of performing an inadvertent rewind as we might be
217 * missing all the vdevs with the latest uberblocks.
219 uint64_t zfs_max_missing_tvds = 0;
222 * The parameters below are similar to zfs_max_missing_tvds but are only
223 * intended for a preliminary open of the pool with an untrusted config which
224 * might be incomplete or out-dated.
226 * We are more tolerant for pools opened from a cachefile since we could have
227 * an out-dated cachefile where a device removal was not registered.
228 * We could have set the limit arbitrarily high but in the case where devices
229 * are really missing we would want to return the proper error codes; we chose
230 * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
231 * and we get a chance to retrieve the trusted config.
233 uint64_t zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
236 * In the case where config was assembled by scanning device paths (/dev/dsks
237 * by default) we are less tolerant since all the existing devices should have
238 * been detected and we want spa_load to return the right error codes.
240 uint64_t zfs_max_missing_tvds_scan = 0;
243 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_print_vdev_tree, CTLFLAG_RWTUN,
244 &spa_load_print_vdev_tree, 0,
245 "print out vdev tree during pool import");
246 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds, CTLFLAG_RWTUN,
247 &zfs_max_missing_tvds, 0,
248 "allow importing pools with missing top-level vdevs");
249 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_cachefile, CTLFLAG_RWTUN,
250 &zfs_max_missing_tvds_cachefile, 0,
251 "allow importing pools with missing top-level vdevs in cache file");
252 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, max_missing_tvds_scan, CTLFLAG_RWTUN,
253 &zfs_max_missing_tvds_scan, 0,
254 "allow importing pools with missing top-level vdevs during scan");
257 * Debugging aid that pauses spa_sync() towards the end.
259 boolean_t zfs_pause_spa_sync = B_FALSE;
262 * ==========================================================================
263 * SPA properties routines
264 * ==========================================================================
268 * Add a (source=src, propname=propval) list to an nvlist.
271 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
272 uint64_t intval, zprop_source_t src)
274 const char *propname = zpool_prop_to_name(prop);
277 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
278 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
281 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
283 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
285 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
286 nvlist_free(propval);
290 * Get property values from the spa configuration.
293 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
295 vdev_t *rvd = spa->spa_root_vdev;
296 dsl_pool_t *pool = spa->spa_dsl_pool;
297 uint64_t size, alloc, cap, version;
298 zprop_source_t src = ZPROP_SRC_NONE;
299 spa_config_dirent_t *dp;
300 metaslab_class_t *mc = spa_normal_class(spa);
302 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
305 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
306 size = metaslab_class_get_space(spa_normal_class(spa));
307 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
308 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
309 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
310 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
312 spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
313 spa->spa_checkpoint_info.sci_dspace, src);
315 spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
316 metaslab_class_fragmentation(mc), src);
317 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
318 metaslab_class_expandable_space(mc), src);
319 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
320 (spa_mode(spa) == FREAD), src);
322 cap = (size == 0) ? 0 : (alloc * 100 / size);
323 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
325 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
326 ddt_get_pool_dedup_ratio(spa), src);
328 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
329 rvd->vdev_state, src);
331 version = spa_version(spa);
332 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
333 src = ZPROP_SRC_DEFAULT;
335 src = ZPROP_SRC_LOCAL;
336 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
341 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
342 * when opening pools before this version freedir will be NULL.
344 if (pool->dp_free_dir != NULL) {
345 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
346 dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
349 spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
353 if (pool->dp_leak_dir != NULL) {
354 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
355 dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
358 spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
363 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
365 if (spa->spa_comment != NULL) {
366 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
370 if (spa->spa_root != NULL)
371 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
374 if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
375 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
376 MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
378 spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
379 SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
382 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
383 if (dp->scd_path == NULL) {
384 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
385 "none", 0, ZPROP_SRC_LOCAL);
386 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
387 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
388 dp->scd_path, 0, ZPROP_SRC_LOCAL);
394 * Get zpool property values.
397 spa_prop_get(spa_t *spa, nvlist_t **nvp)
399 objset_t *mos = spa->spa_meta_objset;
404 VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
406 mutex_enter(&spa->spa_props_lock);
409 * Get properties from the spa config.
411 spa_prop_get_config(spa, nvp);
413 /* If no pool property object, no more prop to get. */
414 if (mos == NULL || spa->spa_pool_props_object == 0) {
415 mutex_exit(&spa->spa_props_lock);
420 * Get properties from the MOS pool property object.
422 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
423 (err = zap_cursor_retrieve(&zc, &za)) == 0;
424 zap_cursor_advance(&zc)) {
427 zprop_source_t src = ZPROP_SRC_DEFAULT;
430 if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
433 switch (za.za_integer_length) {
435 /* integer property */
436 if (za.za_first_integer !=
437 zpool_prop_default_numeric(prop))
438 src = ZPROP_SRC_LOCAL;
440 if (prop == ZPOOL_PROP_BOOTFS) {
442 dsl_dataset_t *ds = NULL;
444 dp = spa_get_dsl(spa);
445 dsl_pool_config_enter(dp, FTAG);
446 if (err = dsl_dataset_hold_obj(dp,
447 za.za_first_integer, FTAG, &ds)) {
448 dsl_pool_config_exit(dp, FTAG);
452 strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
454 dsl_dataset_name(ds, strval);
455 dsl_dataset_rele(ds, FTAG);
456 dsl_pool_config_exit(dp, FTAG);
459 intval = za.za_first_integer;
462 spa_prop_add_list(*nvp, prop, strval, intval, src);
465 kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
470 /* string property */
471 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
472 err = zap_lookup(mos, spa->spa_pool_props_object,
473 za.za_name, 1, za.za_num_integers, strval);
475 kmem_free(strval, za.za_num_integers);
478 spa_prop_add_list(*nvp, prop, strval, 0, src);
479 kmem_free(strval, za.za_num_integers);
486 zap_cursor_fini(&zc);
487 mutex_exit(&spa->spa_props_lock);
489 if (err && err != ENOENT) {
499 * Validate the given pool properties nvlist and modify the list
500 * for the property values to be set.
503 spa_prop_validate(spa_t *spa, nvlist_t *props)
506 int error = 0, reset_bootfs = 0;
508 boolean_t has_feature = B_FALSE;
511 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
513 char *strval, *slash, *check, *fname;
514 const char *propname = nvpair_name(elem);
515 zpool_prop_t prop = zpool_name_to_prop(propname);
518 case ZPOOL_PROP_INVAL:
519 if (!zpool_prop_feature(propname)) {
520 error = SET_ERROR(EINVAL);
525 * Sanitize the input.
527 if (nvpair_type(elem) != DATA_TYPE_UINT64) {
528 error = SET_ERROR(EINVAL);
532 if (nvpair_value_uint64(elem, &intval) != 0) {
533 error = SET_ERROR(EINVAL);
538 error = SET_ERROR(EINVAL);
542 fname = strchr(propname, '@') + 1;
543 if (zfeature_lookup_name(fname, NULL) != 0) {
544 error = SET_ERROR(EINVAL);
548 has_feature = B_TRUE;
551 case ZPOOL_PROP_VERSION:
552 error = nvpair_value_uint64(elem, &intval);
554 (intval < spa_version(spa) ||
555 intval > SPA_VERSION_BEFORE_FEATURES ||
557 error = SET_ERROR(EINVAL);
560 case ZPOOL_PROP_DELEGATION:
561 case ZPOOL_PROP_AUTOREPLACE:
562 case ZPOOL_PROP_LISTSNAPS:
563 case ZPOOL_PROP_AUTOEXPAND:
564 error = nvpair_value_uint64(elem, &intval);
565 if (!error && intval > 1)
566 error = SET_ERROR(EINVAL);
569 case ZPOOL_PROP_BOOTFS:
571 * If the pool version is less than SPA_VERSION_BOOTFS,
572 * or the pool is still being created (version == 0),
573 * the bootfs property cannot be set.
575 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
576 error = SET_ERROR(ENOTSUP);
581 * Make sure the vdev config is bootable
583 if (!vdev_is_bootable(spa->spa_root_vdev)) {
584 error = SET_ERROR(ENOTSUP);
590 error = nvpair_value_string(elem, &strval);
596 if (strval == NULL || strval[0] == '\0') {
597 objnum = zpool_prop_default_numeric(
602 if (error = dmu_objset_hold(strval, FTAG, &os))
606 * Must be ZPL, and its property settings
607 * must be supported by GRUB (compression
608 * is not gzip, and large blocks are not used).
611 if (dmu_objset_type(os) != DMU_OST_ZFS) {
612 error = SET_ERROR(ENOTSUP);
614 dsl_prop_get_int_ds(dmu_objset_ds(os),
615 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
617 !BOOTFS_COMPRESS_VALID(propval)) {
618 error = SET_ERROR(ENOTSUP);
620 objnum = dmu_objset_id(os);
622 dmu_objset_rele(os, FTAG);
626 case ZPOOL_PROP_FAILUREMODE:
627 error = nvpair_value_uint64(elem, &intval);
628 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
629 intval > ZIO_FAILURE_MODE_PANIC))
630 error = SET_ERROR(EINVAL);
633 * This is a special case which only occurs when
634 * the pool has completely failed. This allows
635 * the user to change the in-core failmode property
636 * without syncing it out to disk (I/Os might
637 * currently be blocked). We do this by returning
638 * EIO to the caller (spa_prop_set) to trick it
639 * into thinking we encountered a property validation
642 if (!error && spa_suspended(spa)) {
643 spa->spa_failmode = intval;
644 error = SET_ERROR(EIO);
648 case ZPOOL_PROP_CACHEFILE:
649 if ((error = nvpair_value_string(elem, &strval)) != 0)
652 if (strval[0] == '\0')
655 if (strcmp(strval, "none") == 0)
658 if (strval[0] != '/') {
659 error = SET_ERROR(EINVAL);
663 slash = strrchr(strval, '/');
664 ASSERT(slash != NULL);
666 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
667 strcmp(slash, "/..") == 0)
668 error = SET_ERROR(EINVAL);
671 case ZPOOL_PROP_COMMENT:
672 if ((error = nvpair_value_string(elem, &strval)) != 0)
674 for (check = strval; *check != '\0'; check++) {
676 * The kernel doesn't have an easy isprint()
677 * check. For this kernel check, we merely
678 * check ASCII apart from DEL. Fix this if
679 * there is an easy-to-use kernel isprint().
681 if (*check >= 0x7f) {
682 error = SET_ERROR(EINVAL);
686 if (strlen(strval) > ZPROP_MAX_COMMENT)
690 case ZPOOL_PROP_DEDUPDITTO:
691 if (spa_version(spa) < SPA_VERSION_DEDUP)
692 error = SET_ERROR(ENOTSUP);
694 error = nvpair_value_uint64(elem, &intval);
696 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
697 error = SET_ERROR(EINVAL);
705 if (!error && reset_bootfs) {
706 error = nvlist_remove(props,
707 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
710 error = nvlist_add_uint64(props,
711 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
719 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
722 spa_config_dirent_t *dp;
724 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
728 dp = kmem_alloc(sizeof (spa_config_dirent_t),
731 if (cachefile[0] == '\0')
732 dp->scd_path = spa_strdup(spa_config_path);
733 else if (strcmp(cachefile, "none") == 0)
736 dp->scd_path = spa_strdup(cachefile);
738 list_insert_head(&spa->spa_config_list, dp);
740 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
744 spa_prop_set(spa_t *spa, nvlist_t *nvp)
747 nvpair_t *elem = NULL;
748 boolean_t need_sync = B_FALSE;
750 if ((error = spa_prop_validate(spa, nvp)) != 0)
753 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
754 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
756 if (prop == ZPOOL_PROP_CACHEFILE ||
757 prop == ZPOOL_PROP_ALTROOT ||
758 prop == ZPOOL_PROP_READONLY)
761 if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
764 if (prop == ZPOOL_PROP_VERSION) {
765 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
767 ASSERT(zpool_prop_feature(nvpair_name(elem)));
768 ver = SPA_VERSION_FEATURES;
772 /* Save time if the version is already set. */
773 if (ver == spa_version(spa))
777 * In addition to the pool directory object, we might
778 * create the pool properties object, the features for
779 * read object, the features for write object, or the
780 * feature descriptions object.
782 error = dsl_sync_task(spa->spa_name, NULL,
783 spa_sync_version, &ver,
784 6, ZFS_SPACE_CHECK_RESERVED);
795 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
796 nvp, 6, ZFS_SPACE_CHECK_RESERVED));
803 * If the bootfs property value is dsobj, clear it.
806 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
808 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
809 VERIFY(zap_remove(spa->spa_meta_objset,
810 spa->spa_pool_props_object,
811 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
818 spa_change_guid_check(void *arg, dmu_tx_t *tx)
820 uint64_t *newguid = arg;
821 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
822 vdev_t *rvd = spa->spa_root_vdev;
825 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
826 int error = (spa_has_checkpoint(spa)) ?
827 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
828 return (SET_ERROR(error));
831 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
832 vdev_state = rvd->vdev_state;
833 spa_config_exit(spa, SCL_STATE, FTAG);
835 if (vdev_state != VDEV_STATE_HEALTHY)
836 return (SET_ERROR(ENXIO));
838 ASSERT3U(spa_guid(spa), !=, *newguid);
844 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
846 uint64_t *newguid = arg;
847 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
849 vdev_t *rvd = spa->spa_root_vdev;
851 oldguid = spa_guid(spa);
853 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
854 rvd->vdev_guid = *newguid;
855 rvd->vdev_guid_sum += (*newguid - oldguid);
856 vdev_config_dirty(rvd);
857 spa_config_exit(spa, SCL_STATE, FTAG);
859 spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
864 * Change the GUID for the pool. This is done so that we can later
865 * re-import a pool built from a clone of our own vdevs. We will modify
866 * the root vdev's guid, our own pool guid, and then mark all of our
867 * vdevs dirty. Note that we must make sure that all our vdevs are
868 * online when we do this, or else any vdevs that weren't present
869 * would be orphaned from our pool. We are also going to issue a
870 * sysevent to update any watchers.
873 spa_change_guid(spa_t *spa)
878 mutex_enter(&spa->spa_vdev_top_lock);
879 mutex_enter(&spa_namespace_lock);
880 guid = spa_generate_guid(NULL);
882 error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
883 spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
886 spa_write_cachefile(spa, B_FALSE, B_TRUE);
887 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
890 mutex_exit(&spa_namespace_lock);
891 mutex_exit(&spa->spa_vdev_top_lock);
897 * ==========================================================================
898 * SPA state manipulation (open/create/destroy/import/export)
899 * ==========================================================================
903 spa_error_entry_compare(const void *a, const void *b)
905 spa_error_entry_t *sa = (spa_error_entry_t *)a;
906 spa_error_entry_t *sb = (spa_error_entry_t *)b;
909 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
910 sizeof (zbookmark_phys_t));
921 * Utility function which retrieves copies of the current logs and
922 * re-initializes them in the process.
925 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
927 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
929 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
930 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
932 avl_create(&spa->spa_errlist_scrub,
933 spa_error_entry_compare, sizeof (spa_error_entry_t),
934 offsetof(spa_error_entry_t, se_avl));
935 avl_create(&spa->spa_errlist_last,
936 spa_error_entry_compare, sizeof (spa_error_entry_t),
937 offsetof(spa_error_entry_t, se_avl));
941 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
943 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
944 enum zti_modes mode = ztip->zti_mode;
945 uint_t value = ztip->zti_value;
946 uint_t count = ztip->zti_count;
947 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
950 boolean_t batch = B_FALSE;
952 if (mode == ZTI_MODE_NULL) {
954 tqs->stqs_taskq = NULL;
958 ASSERT3U(count, >, 0);
960 tqs->stqs_count = count;
961 tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
965 ASSERT3U(value, >=, 1);
966 value = MAX(value, 1);
971 flags |= TASKQ_THREADS_CPU_PCT;
972 value = zio_taskq_batch_pct;
976 panic("unrecognized mode for %s_%s taskq (%u:%u) in "
978 zio_type_name[t], zio_taskq_types[q], mode, value);
982 for (uint_t i = 0; i < count; i++) {
986 (void) snprintf(name, sizeof (name), "%s_%s_%u",
987 zio_type_name[t], zio_taskq_types[q], i);
989 (void) snprintf(name, sizeof (name), "%s_%s",
990 zio_type_name[t], zio_taskq_types[q]);
994 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
996 flags |= TASKQ_DC_BATCH;
998 tq = taskq_create_sysdc(name, value, 50, INT_MAX,
999 spa->spa_proc, zio_taskq_basedc, flags);
1002 pri_t pri = maxclsyspri;
1004 * The write issue taskq can be extremely CPU
1005 * intensive. Run it at slightly lower priority
1006 * than the other taskqs.
1008 * - numerically higher priorities are lower priorities;
1009 * - if priorities divided by four (RQ_PPQ) are equal
1010 * then a difference between them is insignificant.
1012 if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1019 tq = taskq_create_proc(name, value, pri, 50,
1020 INT_MAX, spa->spa_proc, flags);
1025 tqs->stqs_taskq[i] = tq;
1030 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1032 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1034 if (tqs->stqs_taskq == NULL) {
1035 ASSERT0(tqs->stqs_count);
1039 for (uint_t i = 0; i < tqs->stqs_count; i++) {
1040 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1041 taskq_destroy(tqs->stqs_taskq[i]);
1044 kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1045 tqs->stqs_taskq = NULL;
1049 * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1050 * Note that a type may have multiple discrete taskqs to avoid lock contention
1051 * on the taskq itself. In that case we choose which taskq at random by using
1052 * the low bits of gethrtime().
1055 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1056 task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1058 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1061 ASSERT3P(tqs->stqs_taskq, !=, NULL);
1062 ASSERT3U(tqs->stqs_count, !=, 0);
1064 if (tqs->stqs_count == 1) {
1065 tq = tqs->stqs_taskq[0];
1068 tq = tqs->stqs_taskq[cpu_ticks() % tqs->stqs_count];
1070 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1074 taskq_dispatch_ent(tq, func, arg, flags, ent);
1078 spa_create_zio_taskqs(spa_t *spa)
1080 for (int t = 0; t < ZIO_TYPES; t++) {
1081 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1082 spa_taskqs_init(spa, t, q);
1090 spa_thread(void *arg)
1092 callb_cpr_t cprinfo;
1095 user_t *pu = PTOU(curproc);
1097 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1100 ASSERT(curproc != &p0);
1101 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1102 "zpool-%s", spa->spa_name);
1103 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1106 /* bind this thread to the requested psrset */
1107 if (zio_taskq_psrset_bind != PS_NONE) {
1109 mutex_enter(&cpu_lock);
1110 mutex_enter(&pidlock);
1111 mutex_enter(&curproc->p_lock);
1113 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1114 0, NULL, NULL) == 0) {
1115 curthread->t_bind_pset = zio_taskq_psrset_bind;
1118 "Couldn't bind process for zfs pool \"%s\" to "
1119 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1122 mutex_exit(&curproc->p_lock);
1123 mutex_exit(&pidlock);
1124 mutex_exit(&cpu_lock);
1130 if (zio_taskq_sysdc) {
1131 sysdc_thread_enter(curthread, 100, 0);
1135 spa->spa_proc = curproc;
1136 spa->spa_did = curthread->t_did;
1138 spa_create_zio_taskqs(spa);
1140 mutex_enter(&spa->spa_proc_lock);
1141 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1143 spa->spa_proc_state = SPA_PROC_ACTIVE;
1144 cv_broadcast(&spa->spa_proc_cv);
1146 CALLB_CPR_SAFE_BEGIN(&cprinfo);
1147 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1148 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1149 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1151 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1152 spa->spa_proc_state = SPA_PROC_GONE;
1153 spa->spa_proc = &p0;
1154 cv_broadcast(&spa->spa_proc_cv);
1155 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
1157 mutex_enter(&curproc->p_lock);
1160 #endif /* SPA_PROCESS */
1164 * Activate an uninitialized pool.
1167 spa_activate(spa_t *spa, int mode)
1169 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1171 spa->spa_state = POOL_STATE_ACTIVE;
1172 spa->spa_mode = mode;
1174 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1175 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1177 /* Try to create a covering process */
1178 mutex_enter(&spa->spa_proc_lock);
1179 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1180 ASSERT(spa->spa_proc == &p0);
1184 /* Only create a process if we're going to be around a while. */
1185 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1186 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1188 spa->spa_proc_state = SPA_PROC_CREATED;
1189 while (spa->spa_proc_state == SPA_PROC_CREATED) {
1190 cv_wait(&spa->spa_proc_cv,
1191 &spa->spa_proc_lock);
1193 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1194 ASSERT(spa->spa_proc != &p0);
1195 ASSERT(spa->spa_did != 0);
1199 "Couldn't create process for zfs pool \"%s\"\n",
1204 #endif /* SPA_PROCESS */
1205 mutex_exit(&spa->spa_proc_lock);
1207 /* If we didn't create a process, we need to create our taskqs. */
1208 ASSERT(spa->spa_proc == &p0);
1209 if (spa->spa_proc == &p0) {
1210 spa_create_zio_taskqs(spa);
1214 * Start TRIM thread.
1216 trim_thread_create(spa);
1218 for (size_t i = 0; i < TXG_SIZE; i++)
1219 spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL, 0);
1221 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1222 offsetof(vdev_t, vdev_config_dirty_node));
1223 list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1224 offsetof(objset_t, os_evicting_node));
1225 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1226 offsetof(vdev_t, vdev_state_dirty_node));
1228 txg_list_create(&spa->spa_vdev_txg_list, spa,
1229 offsetof(struct vdev, vdev_txg_node));
1231 avl_create(&spa->spa_errlist_scrub,
1232 spa_error_entry_compare, sizeof (spa_error_entry_t),
1233 offsetof(spa_error_entry_t, se_avl));
1234 avl_create(&spa->spa_errlist_last,
1235 spa_error_entry_compare, sizeof (spa_error_entry_t),
1236 offsetof(spa_error_entry_t, se_avl));
1240 * Opposite of spa_activate().
1243 spa_deactivate(spa_t *spa)
1245 ASSERT(spa->spa_sync_on == B_FALSE);
1246 ASSERT(spa->spa_dsl_pool == NULL);
1247 ASSERT(spa->spa_root_vdev == NULL);
1248 ASSERT(spa->spa_async_zio_root == NULL);
1249 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1252 * Stop TRIM thread in case spa_unload() wasn't called directly
1253 * before spa_deactivate().
1255 trim_thread_destroy(spa);
1257 spa_evicting_os_wait(spa);
1259 txg_list_destroy(&spa->spa_vdev_txg_list);
1261 list_destroy(&spa->spa_config_dirty_list);
1262 list_destroy(&spa->spa_evicting_os_list);
1263 list_destroy(&spa->spa_state_dirty_list);
1265 for (int t = 0; t < ZIO_TYPES; t++) {
1266 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1267 spa_taskqs_fini(spa, t, q);
1271 for (size_t i = 0; i < TXG_SIZE; i++) {
1272 ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1273 VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1274 spa->spa_txg_zio[i] = NULL;
1277 metaslab_class_destroy(spa->spa_normal_class);
1278 spa->spa_normal_class = NULL;
1280 metaslab_class_destroy(spa->spa_log_class);
1281 spa->spa_log_class = NULL;
1284 * If this was part of an import or the open otherwise failed, we may
1285 * still have errors left in the queues. Empty them just in case.
1287 spa_errlog_drain(spa);
1289 avl_destroy(&spa->spa_errlist_scrub);
1290 avl_destroy(&spa->spa_errlist_last);
1292 spa->spa_state = POOL_STATE_UNINITIALIZED;
1294 mutex_enter(&spa->spa_proc_lock);
1295 if (spa->spa_proc_state != SPA_PROC_NONE) {
1296 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1297 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1298 cv_broadcast(&spa->spa_proc_cv);
1299 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1300 ASSERT(spa->spa_proc != &p0);
1301 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1303 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1304 spa->spa_proc_state = SPA_PROC_NONE;
1306 ASSERT(spa->spa_proc == &p0);
1307 mutex_exit(&spa->spa_proc_lock);
1311 * We want to make sure spa_thread() has actually exited the ZFS
1312 * module, so that the module can't be unloaded out from underneath
1315 if (spa->spa_did != 0) {
1316 thread_join(spa->spa_did);
1319 #endif /* SPA_PROCESS */
1323 * Verify a pool configuration, and construct the vdev tree appropriately. This
1324 * will create all the necessary vdevs in the appropriate layout, with each vdev
1325 * in the CLOSED state. This will prep the pool before open/creation/import.
1326 * All vdev validation is done by the vdev_alloc() routine.
1329 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1330 uint_t id, int atype)
1336 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1339 if ((*vdp)->vdev_ops->vdev_op_leaf)
1342 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1345 if (error == ENOENT)
1351 return (SET_ERROR(EINVAL));
1354 for (int c = 0; c < children; c++) {
1356 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1364 ASSERT(*vdp != NULL);
1370 * Opposite of spa_load().
1373 spa_unload(spa_t *spa)
1377 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1379 spa_load_note(spa, "UNLOADING");
1384 trim_thread_destroy(spa);
1389 spa_async_suspend(spa);
1394 if (spa->spa_sync_on) {
1395 txg_sync_stop(spa->spa_dsl_pool);
1396 spa->spa_sync_on = B_FALSE;
1400 * Even though vdev_free() also calls vdev_metaslab_fini, we need
1401 * to call it earlier, before we wait for async i/o to complete.
1402 * This ensures that there is no async metaslab prefetching, by
1403 * calling taskq_wait(mg_taskq).
1405 if (spa->spa_root_vdev != NULL) {
1406 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1407 for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++)
1408 vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
1409 spa_config_exit(spa, SCL_ALL, FTAG);
1413 * Wait for any outstanding async I/O to complete.
1415 if (spa->spa_async_zio_root != NULL) {
1416 for (int i = 0; i < max_ncpus; i++)
1417 (void) zio_wait(spa->spa_async_zio_root[i]);
1418 kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1419 spa->spa_async_zio_root = NULL;
1422 if (spa->spa_vdev_removal != NULL) {
1423 spa_vdev_removal_destroy(spa->spa_vdev_removal);
1424 spa->spa_vdev_removal = NULL;
1427 if (spa->spa_condense_zthr != NULL) {
1428 ASSERT(!zthr_isrunning(spa->spa_condense_zthr));
1429 zthr_destroy(spa->spa_condense_zthr);
1430 spa->spa_condense_zthr = NULL;
1433 if (spa->spa_checkpoint_discard_zthr != NULL) {
1434 ASSERT(!zthr_isrunning(spa->spa_checkpoint_discard_zthr));
1435 zthr_destroy(spa->spa_checkpoint_discard_zthr);
1436 spa->spa_checkpoint_discard_zthr = NULL;
1439 spa_condense_fini(spa);
1441 bpobj_close(&spa->spa_deferred_bpobj);
1443 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1448 if (spa->spa_root_vdev)
1449 vdev_free(spa->spa_root_vdev);
1450 ASSERT(spa->spa_root_vdev == NULL);
1453 * Close the dsl pool.
1455 if (spa->spa_dsl_pool) {
1456 dsl_pool_close(spa->spa_dsl_pool);
1457 spa->spa_dsl_pool = NULL;
1458 spa->spa_meta_objset = NULL;
1464 * Drop and purge level 2 cache
1466 spa_l2cache_drop(spa);
1468 for (i = 0; i < spa->spa_spares.sav_count; i++)
1469 vdev_free(spa->spa_spares.sav_vdevs[i]);
1470 if (spa->spa_spares.sav_vdevs) {
1471 kmem_free(spa->spa_spares.sav_vdevs,
1472 spa->spa_spares.sav_count * sizeof (void *));
1473 spa->spa_spares.sav_vdevs = NULL;
1475 if (spa->spa_spares.sav_config) {
1476 nvlist_free(spa->spa_spares.sav_config);
1477 spa->spa_spares.sav_config = NULL;
1479 spa->spa_spares.sav_count = 0;
1481 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1482 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1483 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1485 if (spa->spa_l2cache.sav_vdevs) {
1486 kmem_free(spa->spa_l2cache.sav_vdevs,
1487 spa->spa_l2cache.sav_count * sizeof (void *));
1488 spa->spa_l2cache.sav_vdevs = NULL;
1490 if (spa->spa_l2cache.sav_config) {
1491 nvlist_free(spa->spa_l2cache.sav_config);
1492 spa->spa_l2cache.sav_config = NULL;
1494 spa->spa_l2cache.sav_count = 0;
1496 spa->spa_async_suspended = 0;
1498 spa->spa_indirect_vdevs_loaded = B_FALSE;
1500 if (spa->spa_comment != NULL) {
1501 spa_strfree(spa->spa_comment);
1502 spa->spa_comment = NULL;
1505 spa_config_exit(spa, SCL_ALL, FTAG);
1509 * Load (or re-load) the current list of vdevs describing the active spares for
1510 * this pool. When this is called, we have some form of basic information in
1511 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1512 * then re-generate a more complete list including status information.
1515 spa_load_spares(spa_t *spa)
1524 * zdb opens both the current state of the pool and the
1525 * checkpointed state (if present), with a different spa_t.
1527 * As spare vdevs are shared among open pools, we skip loading
1528 * them when we load the checkpointed state of the pool.
1530 if (!spa_writeable(spa))
1534 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1537 * First, close and free any existing spare vdevs.
1539 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1540 vd = spa->spa_spares.sav_vdevs[i];
1542 /* Undo the call to spa_activate() below */
1543 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1544 B_FALSE)) != NULL && tvd->vdev_isspare)
1545 spa_spare_remove(tvd);
1550 if (spa->spa_spares.sav_vdevs)
1551 kmem_free(spa->spa_spares.sav_vdevs,
1552 spa->spa_spares.sav_count * sizeof (void *));
1554 if (spa->spa_spares.sav_config == NULL)
1557 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1558 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1560 spa->spa_spares.sav_count = (int)nspares;
1561 spa->spa_spares.sav_vdevs = NULL;
1567 * Construct the array of vdevs, opening them to get status in the
1568 * process. For each spare, there is potentially two different vdev_t
1569 * structures associated with it: one in the list of spares (used only
1570 * for basic validation purposes) and one in the active vdev
1571 * configuration (if it's spared in). During this phase we open and
1572 * validate each vdev on the spare list. If the vdev also exists in the
1573 * active configuration, then we also mark this vdev as an active spare.
1575 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1577 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1578 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1579 VDEV_ALLOC_SPARE) == 0);
1582 spa->spa_spares.sav_vdevs[i] = vd;
1584 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1585 B_FALSE)) != NULL) {
1586 if (!tvd->vdev_isspare)
1590 * We only mark the spare active if we were successfully
1591 * able to load the vdev. Otherwise, importing a pool
1592 * with a bad active spare would result in strange
1593 * behavior, because multiple pool would think the spare
1594 * is actively in use.
1596 * There is a vulnerability here to an equally bizarre
1597 * circumstance, where a dead active spare is later
1598 * brought back to life (onlined or otherwise). Given
1599 * the rarity of this scenario, and the extra complexity
1600 * it adds, we ignore the possibility.
1602 if (!vdev_is_dead(tvd))
1603 spa_spare_activate(tvd);
1607 vd->vdev_aux = &spa->spa_spares;
1609 if (vdev_open(vd) != 0)
1612 if (vdev_validate_aux(vd) == 0)
1617 * Recompute the stashed list of spares, with status information
1620 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1621 DATA_TYPE_NVLIST_ARRAY) == 0);
1623 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1625 for (i = 0; i < spa->spa_spares.sav_count; i++)
1626 spares[i] = vdev_config_generate(spa,
1627 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1628 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1629 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1630 for (i = 0; i < spa->spa_spares.sav_count; i++)
1631 nvlist_free(spares[i]);
1632 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1636 * Load (or re-load) the current list of vdevs describing the active l2cache for
1637 * this pool. When this is called, we have some form of basic information in
1638 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1639 * then re-generate a more complete list including status information.
1640 * Devices which are already active have their details maintained, and are
1644 spa_load_l2cache(spa_t *spa)
1648 int i, j, oldnvdevs;
1650 vdev_t *vd, **oldvdevs, **newvdevs;
1651 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1655 * zdb opens both the current state of the pool and the
1656 * checkpointed state (if present), with a different spa_t.
1658 * As L2 caches are part of the ARC which is shared among open
1659 * pools, we skip loading them when we load the checkpointed
1660 * state of the pool.
1662 if (!spa_writeable(spa))
1666 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1668 if (sav->sav_config != NULL) {
1669 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1670 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1671 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1677 oldvdevs = sav->sav_vdevs;
1678 oldnvdevs = sav->sav_count;
1679 sav->sav_vdevs = NULL;
1683 * Process new nvlist of vdevs.
1685 for (i = 0; i < nl2cache; i++) {
1686 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1690 for (j = 0; j < oldnvdevs; j++) {
1692 if (vd != NULL && guid == vd->vdev_guid) {
1694 * Retain previous vdev for add/remove ops.
1702 if (newvdevs[i] == NULL) {
1706 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1707 VDEV_ALLOC_L2CACHE) == 0);
1712 * Commit this vdev as an l2cache device,
1713 * even if it fails to open.
1715 spa_l2cache_add(vd);
1720 spa_l2cache_activate(vd);
1722 if (vdev_open(vd) != 0)
1725 (void) vdev_validate_aux(vd);
1727 if (!vdev_is_dead(vd))
1728 l2arc_add_vdev(spa, vd);
1733 * Purge vdevs that were dropped
1735 for (i = 0; i < oldnvdevs; i++) {
1740 ASSERT(vd->vdev_isl2cache);
1742 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1743 pool != 0ULL && l2arc_vdev_present(vd))
1744 l2arc_remove_vdev(vd);
1745 vdev_clear_stats(vd);
1751 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1753 if (sav->sav_config == NULL)
1756 sav->sav_vdevs = newvdevs;
1757 sav->sav_count = (int)nl2cache;
1760 * Recompute the stashed list of l2cache devices, with status
1761 * information this time.
1763 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1764 DATA_TYPE_NVLIST_ARRAY) == 0);
1766 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1767 for (i = 0; i < sav->sav_count; i++)
1768 l2cache[i] = vdev_config_generate(spa,
1769 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1770 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1771 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1773 for (i = 0; i < sav->sav_count; i++)
1774 nvlist_free(l2cache[i]);
1776 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1780 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1783 char *packed = NULL;
1788 error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1792 nvsize = *(uint64_t *)db->db_data;
1793 dmu_buf_rele(db, FTAG);
1795 packed = kmem_alloc(nvsize, KM_SLEEP);
1796 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1799 error = nvlist_unpack(packed, nvsize, value, 0);
1800 kmem_free(packed, nvsize);
1806 * Concrete top-level vdevs that are not missing and are not logs. At every
1807 * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1810 spa_healthy_core_tvds(spa_t *spa)
1812 vdev_t *rvd = spa->spa_root_vdev;
1815 for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1816 vdev_t *vd = rvd->vdev_child[i];
1819 if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1827 * Checks to see if the given vdev could not be opened, in which case we post a
1828 * sysevent to notify the autoreplace code that the device has been removed.
1831 spa_check_removed(vdev_t *vd)
1833 for (uint64_t c = 0; c < vd->vdev_children; c++)
1834 spa_check_removed(vd->vdev_child[c]);
1836 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1837 vdev_is_concrete(vd)) {
1838 zfs_post_autoreplace(vd->vdev_spa, vd);
1839 spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1844 spa_check_for_missing_logs(spa_t *spa)
1846 vdev_t *rvd = spa->spa_root_vdev;
1849 * If we're doing a normal import, then build up any additional
1850 * diagnostic information about missing log devices.
1851 * We'll pass this up to the user for further processing.
1853 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1854 nvlist_t **child, *nv;
1857 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1859 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1861 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1862 vdev_t *tvd = rvd->vdev_child[c];
1865 * We consider a device as missing only if it failed
1866 * to open (i.e. offline or faulted is not considered
1869 if (tvd->vdev_islog &&
1870 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1871 child[idx++] = vdev_config_generate(spa, tvd,
1872 B_FALSE, VDEV_CONFIG_MISSING);
1877 fnvlist_add_nvlist_array(nv,
1878 ZPOOL_CONFIG_CHILDREN, child, idx);
1879 fnvlist_add_nvlist(spa->spa_load_info,
1880 ZPOOL_CONFIG_MISSING_DEVICES, nv);
1882 for (uint64_t i = 0; i < idx; i++)
1883 nvlist_free(child[i]);
1886 kmem_free(child, rvd->vdev_children * sizeof (char **));
1889 spa_load_failed(spa, "some log devices are missing");
1890 vdev_dbgmsg_print_tree(rvd, 2);
1891 return (SET_ERROR(ENXIO));
1894 for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1895 vdev_t *tvd = rvd->vdev_child[c];
1897 if (tvd->vdev_islog &&
1898 tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1899 spa_set_log_state(spa, SPA_LOG_CLEAR);
1900 spa_load_note(spa, "some log devices are "
1901 "missing, ZIL is dropped.");
1902 vdev_dbgmsg_print_tree(rvd, 2);
1912 * Check for missing log devices
1915 spa_check_logs(spa_t *spa)
1917 boolean_t rv = B_FALSE;
1918 dsl_pool_t *dp = spa_get_dsl(spa);
1920 switch (spa->spa_log_state) {
1921 case SPA_LOG_MISSING:
1922 /* need to recheck in case slog has been restored */
1923 case SPA_LOG_UNKNOWN:
1924 rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1925 zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
1927 spa_set_log_state(spa, SPA_LOG_MISSING);
1934 spa_passivate_log(spa_t *spa)
1936 vdev_t *rvd = spa->spa_root_vdev;
1937 boolean_t slog_found = B_FALSE;
1939 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1941 if (!spa_has_slogs(spa))
1944 for (int c = 0; c < rvd->vdev_children; c++) {
1945 vdev_t *tvd = rvd->vdev_child[c];
1946 metaslab_group_t *mg = tvd->vdev_mg;
1948 if (tvd->vdev_islog) {
1949 metaslab_group_passivate(mg);
1950 slog_found = B_TRUE;
1954 return (slog_found);
1958 spa_activate_log(spa_t *spa)
1960 vdev_t *rvd = spa->spa_root_vdev;
1962 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1964 for (int c = 0; c < rvd->vdev_children; c++) {
1965 vdev_t *tvd = rvd->vdev_child[c];
1966 metaslab_group_t *mg = tvd->vdev_mg;
1968 if (tvd->vdev_islog)
1969 metaslab_group_activate(mg);
1974 spa_reset_logs(spa_t *spa)
1978 error = dmu_objset_find(spa_name(spa), zil_reset,
1979 NULL, DS_FIND_CHILDREN);
1982 * We successfully offlined the log device, sync out the
1983 * current txg so that the "stubby" block can be removed
1986 txg_wait_synced(spa->spa_dsl_pool, 0);
1992 spa_aux_check_removed(spa_aux_vdev_t *sav)
1996 for (i = 0; i < sav->sav_count; i++)
1997 spa_check_removed(sav->sav_vdevs[i]);
2001 spa_claim_notify(zio_t *zio)
2003 spa_t *spa = zio->io_spa;
2008 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
2009 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2010 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2011 mutex_exit(&spa->spa_props_lock);
2014 typedef struct spa_load_error {
2015 uint64_t sle_meta_count;
2016 uint64_t sle_data_count;
2020 spa_load_verify_done(zio_t *zio)
2022 blkptr_t *bp = zio->io_bp;
2023 spa_load_error_t *sle = zio->io_private;
2024 dmu_object_type_t type = BP_GET_TYPE(bp);
2025 int error = zio->io_error;
2026 spa_t *spa = zio->io_spa;
2028 abd_free(zio->io_abd);
2030 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2031 type != DMU_OT_INTENT_LOG)
2032 atomic_inc_64(&sle->sle_meta_count);
2034 atomic_inc_64(&sle->sle_data_count);
2037 mutex_enter(&spa->spa_scrub_lock);
2038 spa->spa_scrub_inflight--;
2039 cv_broadcast(&spa->spa_scrub_io_cv);
2040 mutex_exit(&spa->spa_scrub_lock);
2044 * Maximum number of concurrent scrub i/os to create while verifying
2045 * a pool while importing it.
2047 int spa_load_verify_maxinflight = 10000;
2048 boolean_t spa_load_verify_metadata = B_TRUE;
2049 boolean_t spa_load_verify_data = B_TRUE;
2051 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_maxinflight, CTLFLAG_RWTUN,
2052 &spa_load_verify_maxinflight, 0,
2053 "Maximum number of concurrent scrub I/Os to create while verifying a "
2054 "pool while importing it");
2056 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_metadata, CTLFLAG_RWTUN,
2057 &spa_load_verify_metadata, 0,
2058 "Check metadata on import?");
2060 SYSCTL_INT(_vfs_zfs, OID_AUTO, spa_load_verify_data, CTLFLAG_RWTUN,
2061 &spa_load_verify_data, 0,
2062 "Check user data on import?");
2066 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2067 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2069 if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2072 * Note: normally this routine will not be called if
2073 * spa_load_verify_metadata is not set. However, it may be useful
2074 * to manually set the flag after the traversal has begun.
2076 if (!spa_load_verify_metadata)
2078 if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2082 size_t size = BP_GET_PSIZE(bp);
2084 mutex_enter(&spa->spa_scrub_lock);
2085 while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
2086 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2087 spa->spa_scrub_inflight++;
2088 mutex_exit(&spa->spa_scrub_lock);
2090 zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2091 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2092 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2093 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2099 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2101 if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2102 return (SET_ERROR(ENAMETOOLONG));
2108 spa_load_verify(spa_t *spa)
2111 spa_load_error_t sle = { 0 };
2112 zpool_rewind_policy_t policy;
2113 boolean_t verify_ok = B_FALSE;
2116 zpool_get_rewind_policy(spa->spa_config, &policy);
2118 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
2121 dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2122 error = dmu_objset_find_dp(spa->spa_dsl_pool,
2123 spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2125 dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2129 rio = zio_root(spa, NULL, &sle,
2130 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2132 if (spa_load_verify_metadata) {
2133 if (spa->spa_extreme_rewind) {
2134 spa_load_note(spa, "performing a complete scan of the "
2135 "pool since extreme rewind is on. This may take "
2136 "a very long time.\n (spa_load_verify_data=%u, "
2137 "spa_load_verify_metadata=%u)",
2138 spa_load_verify_data, spa_load_verify_metadata);
2140 error = traverse_pool(spa, spa->spa_verify_min_txg,
2141 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
2142 spa_load_verify_cb, rio);
2145 (void) zio_wait(rio);
2147 spa->spa_load_meta_errors = sle.sle_meta_count;
2148 spa->spa_load_data_errors = sle.sle_data_count;
2150 if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2151 spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2152 "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2153 (u_longlong_t)sle.sle_data_count);
2156 if (spa_load_verify_dryrun ||
2157 (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
2158 sle.sle_data_count <= policy.zrp_maxdata)) {
2162 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2163 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2165 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2166 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2167 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2168 VERIFY(nvlist_add_int64(spa->spa_load_info,
2169 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2170 VERIFY(nvlist_add_uint64(spa->spa_load_info,
2171 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2173 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2176 if (spa_load_verify_dryrun)
2180 if (error != ENXIO && error != EIO)
2181 error = SET_ERROR(EIO);
2185 return (verify_ok ? 0 : EIO);
2189 * Find a value in the pool props object.
2192 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2194 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2195 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2199 * Find a value in the pool directory object.
2202 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2204 int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2205 name, sizeof (uint64_t), 1, val);
2207 if (error != 0 && (error != ENOENT || log_enoent)) {
2208 spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2209 "[error=%d]", name, error);
2216 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2218 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2219 return (SET_ERROR(err));
2223 spa_spawn_aux_threads(spa_t *spa)
2225 ASSERT(spa_writeable(spa));
2227 ASSERT(MUTEX_HELD(&spa_namespace_lock));
2229 spa_start_indirect_condensing_thread(spa);
2231 ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2232 spa->spa_checkpoint_discard_zthr =
2233 zthr_create(spa_checkpoint_discard_thread_check,
2234 spa_checkpoint_discard_thread, spa);
2238 * Fix up config after a partly-completed split. This is done with the
2239 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
2240 * pool have that entry in their config, but only the splitting one contains
2241 * a list of all the guids of the vdevs that are being split off.
2243 * This function determines what to do with that list: either rejoin
2244 * all the disks to the pool, or complete the splitting process. To attempt
2245 * the rejoin, each disk that is offlined is marked online again, and
2246 * we do a reopen() call. If the vdev label for every disk that was
2247 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2248 * then we call vdev_split() on each disk, and complete the split.
2250 * Otherwise we leave the config alone, with all the vdevs in place in
2251 * the original pool.
2254 spa_try_repair(spa_t *spa, nvlist_t *config)
2261 boolean_t attempt_reopen;
2263 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2266 /* check that the config is complete */
2267 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2268 &glist, &gcount) != 0)
2271 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2273 /* attempt to online all the vdevs & validate */
2274 attempt_reopen = B_TRUE;
2275 for (i = 0; i < gcount; i++) {
2276 if (glist[i] == 0) /* vdev is hole */
2279 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2280 if (vd[i] == NULL) {
2282 * Don't bother attempting to reopen the disks;
2283 * just do the split.
2285 attempt_reopen = B_FALSE;
2287 /* attempt to re-online it */
2288 vd[i]->vdev_offline = B_FALSE;
2292 if (attempt_reopen) {
2293 vdev_reopen(spa->spa_root_vdev);
2295 /* check each device to see what state it's in */
2296 for (extracted = 0, i = 0; i < gcount; i++) {
2297 if (vd[i] != NULL &&
2298 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2305 * If every disk has been moved to the new pool, or if we never
2306 * even attempted to look at them, then we split them off for
2309 if (!attempt_reopen || gcount == extracted) {
2310 for (i = 0; i < gcount; i++)
2313 vdev_reopen(spa->spa_root_vdev);
2316 kmem_free(vd, gcount * sizeof (vdev_t *));
2320 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2322 char *ereport = FM_EREPORT_ZFS_POOL;
2325 spa->spa_load_state = state;
2327 gethrestime(&spa->spa_loaded_ts);
2328 error = spa_load_impl(spa, type, &ereport);
2331 * Don't count references from objsets that are already closed
2332 * and are making their way through the eviction process.
2334 spa_evicting_os_wait(spa);
2335 spa->spa_minref = refcount_count(&spa->spa_refcount);
2337 if (error != EEXIST) {
2338 spa->spa_loaded_ts.tv_sec = 0;
2339 spa->spa_loaded_ts.tv_nsec = 0;
2341 if (error != EBADF) {
2342 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2345 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2352 * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2353 * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2354 * spa's per-vdev ZAP list.
2357 vdev_count_verify_zaps(vdev_t *vd)
2359 spa_t *spa = vd->vdev_spa;
2361 if (vd->vdev_top_zap != 0) {
2363 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2364 spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2366 if (vd->vdev_leaf_zap != 0) {
2368 ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2369 spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2372 for (uint64_t i = 0; i < vd->vdev_children; i++) {
2373 total += vdev_count_verify_zaps(vd->vdev_child[i]);
2380 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2384 uint64_t myhostid = 0;
2386 if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2387 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2388 hostname = fnvlist_lookup_string(mos_config,
2389 ZPOOL_CONFIG_HOSTNAME);
2391 myhostid = zone_get_hostid(NULL);
2393 if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2394 cmn_err(CE_WARN, "pool '%s' could not be "
2395 "loaded as it was last accessed by "
2396 "another system (host: %s hostid: 0x%llx). "
2397 "See: http://illumos.org/msg/ZFS-8000-EY",
2398 spa_name(spa), hostname, (u_longlong_t)hostid);
2399 spa_load_failed(spa, "hostid verification failed: pool "
2400 "last accessed by host: %s (hostid: 0x%llx)",
2401 hostname, (u_longlong_t)hostid);
2402 return (SET_ERROR(EBADF));
2410 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2413 nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2420 * Versioning wasn't explicitly added to the label until later, so if
2421 * it's not present treat it as the initial version.
2423 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2424 &spa->spa_ubsync.ub_version) != 0)
2425 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2427 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2428 spa_load_failed(spa, "invalid config provided: '%s' missing",
2429 ZPOOL_CONFIG_POOL_GUID);
2430 return (SET_ERROR(EINVAL));
2434 * If we are doing an import, ensure that the pool is not already
2435 * imported by checking if its pool guid already exists in the
2438 * The only case that we allow an already imported pool to be
2439 * imported again, is when the pool is checkpointed and we want to
2440 * look at its checkpointed state from userland tools like zdb.
2443 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2444 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2445 spa_guid_exists(pool_guid, 0)) {
2447 if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2448 spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2449 spa_guid_exists(pool_guid, 0) &&
2450 !spa_importing_readonly_checkpoint(spa)) {
2452 spa_load_failed(spa, "a pool with guid %llu is already open",
2453 (u_longlong_t)pool_guid);
2454 return (SET_ERROR(EEXIST));
2457 spa->spa_config_guid = pool_guid;
2459 nvlist_free(spa->spa_load_info);
2460 spa->spa_load_info = fnvlist_alloc();
2462 ASSERT(spa->spa_comment == NULL);
2463 if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2464 spa->spa_comment = spa_strdup(comment);
2466 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2467 &spa->spa_config_txg);
2469 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2470 spa->spa_config_splitting = fnvlist_dup(nvl);
2472 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2473 spa_load_failed(spa, "invalid config provided: '%s' missing",
2474 ZPOOL_CONFIG_VDEV_TREE);
2475 return (SET_ERROR(EINVAL));
2479 * Create "The Godfather" zio to hold all async IOs
2481 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2483 for (int i = 0; i < max_ncpus; i++) {
2484 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2485 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2486 ZIO_FLAG_GODFATHER);
2490 * Parse the configuration into a vdev tree. We explicitly set the
2491 * value that will be returned by spa_version() since parsing the
2492 * configuration requires knowing the version number.
2494 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2495 parse = (type == SPA_IMPORT_EXISTING ?
2496 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2497 error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2498 spa_config_exit(spa, SCL_ALL, FTAG);
2501 spa_load_failed(spa, "unable to parse config [error=%d]",
2506 ASSERT(spa->spa_root_vdev == rvd);
2507 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2508 ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2510 if (type != SPA_IMPORT_ASSEMBLE) {
2511 ASSERT(spa_guid(spa) == pool_guid);
2518 * Recursively open all vdevs in the vdev tree. This function is called twice:
2519 * first with the untrusted config, then with the trusted config.
2522 spa_ld_open_vdevs(spa_t *spa)
2527 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2528 * missing/unopenable for the root vdev to be still considered openable.
2530 if (spa->spa_trust_config) {
2531 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2532 } else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2533 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2534 } else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2535 spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2537 spa->spa_missing_tvds_allowed = 0;
2540 spa->spa_missing_tvds_allowed =
2541 MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2543 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2544 error = vdev_open(spa->spa_root_vdev);
2545 spa_config_exit(spa, SCL_ALL, FTAG);
2547 if (spa->spa_missing_tvds != 0) {
2548 spa_load_note(spa, "vdev tree has %lld missing top-level "
2549 "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2550 if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2552 * Although theoretically we could allow users to open
2553 * incomplete pools in RW mode, we'd need to add a lot
2554 * of extra logic (e.g. adjust pool space to account
2555 * for missing vdevs).
2556 * This limitation also prevents users from accidentally
2557 * opening the pool in RW mode during data recovery and
2558 * damaging it further.
2560 spa_load_note(spa, "pools with missing top-level "
2561 "vdevs can only be opened in read-only mode.");
2562 error = SET_ERROR(ENXIO);
2564 spa_load_note(spa, "current settings allow for maximum "
2565 "%lld missing top-level vdevs at this stage.",
2566 (u_longlong_t)spa->spa_missing_tvds_allowed);
2570 spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2573 if (spa->spa_missing_tvds != 0 || error != 0)
2574 vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2580 * We need to validate the vdev labels against the configuration that
2581 * we have in hand. This function is called twice: first with an untrusted
2582 * config, then with a trusted config. The validation is more strict when the
2583 * config is trusted.
2586 spa_ld_validate_vdevs(spa_t *spa)
2589 vdev_t *rvd = spa->spa_root_vdev;
2591 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2592 error = vdev_validate(rvd);
2593 spa_config_exit(spa, SCL_ALL, FTAG);
2596 spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2600 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2601 spa_load_failed(spa, "cannot open vdev tree after invalidating "
2603 vdev_dbgmsg_print_tree(rvd, 2);
2604 return (SET_ERROR(ENXIO));
2611 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2613 spa->spa_state = POOL_STATE_ACTIVE;
2614 spa->spa_ubsync = spa->spa_uberblock;
2615 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2616 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2617 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2618 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2619 spa->spa_claim_max_txg = spa->spa_first_txg;
2620 spa->spa_prev_software_version = ub->ub_software_version;
2624 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2626 vdev_t *rvd = spa->spa_root_vdev;
2628 uberblock_t *ub = &spa->spa_uberblock;
2631 * If we are opening the checkpointed state of the pool by
2632 * rewinding to it, at this point we will have written the
2633 * checkpointed uberblock to the vdev labels, so searching
2634 * the labels will find the right uberblock. However, if
2635 * we are opening the checkpointed state read-only, we have
2636 * not modified the labels. Therefore, we must ignore the
2637 * labels and continue using the spa_uberblock that was set
2638 * by spa_ld_checkpoint_rewind.
2640 * Note that it would be fine to ignore the labels when
2641 * rewinding (opening writeable) as well. However, if we
2642 * crash just after writing the labels, we will end up
2643 * searching the labels. Doing so in the common case means
2644 * that this code path gets exercised normally, rather than
2645 * just in the edge case.
2647 if (ub->ub_checkpoint_txg != 0 &&
2648 spa_importing_readonly_checkpoint(spa)) {
2649 spa_ld_select_uberblock_done(spa, ub);
2654 * Find the best uberblock.
2656 vdev_uberblock_load(rvd, ub, &label);
2659 * If we weren't able to find a single valid uberblock, return failure.
2661 if (ub->ub_txg == 0) {
2663 spa_load_failed(spa, "no valid uberblock found");
2664 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2667 spa_load_note(spa, "using uberblock with txg=%llu",
2668 (u_longlong_t)ub->ub_txg);
2671 * If the pool has an unsupported version we can't open it.
2673 if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2675 spa_load_failed(spa, "version %llu is not supported",
2676 (u_longlong_t)ub->ub_version);
2677 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2680 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2684 * If we weren't able to find what's necessary for reading the
2685 * MOS in the label, return failure.
2687 if (label == NULL) {
2688 spa_load_failed(spa, "label config unavailable");
2689 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2693 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
2696 spa_load_failed(spa, "invalid label: '%s' missing",
2697 ZPOOL_CONFIG_FEATURES_FOR_READ);
2698 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2703 * Update our in-core representation with the definitive values
2706 nvlist_free(spa->spa_label_features);
2707 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2713 * Look through entries in the label nvlist's features_for_read. If
2714 * there is a feature listed there which we don't understand then we
2715 * cannot open a pool.
2717 if (ub->ub_version >= SPA_VERSION_FEATURES) {
2718 nvlist_t *unsup_feat;
2720 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2723 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2725 nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2726 if (!zfeature_is_supported(nvpair_name(nvp))) {
2727 VERIFY(nvlist_add_string(unsup_feat,
2728 nvpair_name(nvp), "") == 0);
2732 if (!nvlist_empty(unsup_feat)) {
2733 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2734 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2735 nvlist_free(unsup_feat);
2736 spa_load_failed(spa, "some features are unsupported");
2737 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2741 nvlist_free(unsup_feat);
2744 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2745 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2746 spa_try_repair(spa, spa->spa_config);
2747 spa_config_exit(spa, SCL_ALL, FTAG);
2748 nvlist_free(spa->spa_config_splitting);
2749 spa->spa_config_splitting = NULL;
2753 * Initialize internal SPA structures.
2755 spa_ld_select_uberblock_done(spa, ub);
2761 spa_ld_open_rootbp(spa_t *spa)
2764 vdev_t *rvd = spa->spa_root_vdev;
2766 error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2768 spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
2769 "[error=%d]", error);
2770 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2772 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2778 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
2779 boolean_t reloading)
2781 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
2782 nvlist_t *nv, *mos_config, *policy;
2783 int error = 0, copy_error;
2784 uint64_t healthy_tvds, healthy_tvds_mos;
2785 uint64_t mos_config_txg;
2787 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
2789 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2792 * If we're assembling a pool from a split, the config provided is
2793 * already trusted so there is nothing to do.
2795 if (type == SPA_IMPORT_ASSEMBLE)
2798 healthy_tvds = spa_healthy_core_tvds(spa);
2800 if (load_nvlist(spa, spa->spa_config_object, &mos_config)
2802 spa_load_failed(spa, "unable to retrieve MOS config");
2803 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2807 * If we are doing an open, pool owner wasn't verified yet, thus do
2808 * the verification here.
2810 if (spa->spa_load_state == SPA_LOAD_OPEN) {
2811 error = spa_verify_host(spa, mos_config);
2813 nvlist_free(mos_config);
2818 nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
2820 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2823 * Build a new vdev tree from the trusted config
2825 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
2828 * Vdev paths in the MOS may be obsolete. If the untrusted config was
2829 * obtained by scanning /dev/dsk, then it will have the right vdev
2830 * paths. We update the trusted MOS config with this information.
2831 * We first try to copy the paths with vdev_copy_path_strict, which
2832 * succeeds only when both configs have exactly the same vdev tree.
2833 * If that fails, we fall back to a more flexible method that has a
2834 * best effort policy.
2836 copy_error = vdev_copy_path_strict(rvd, mrvd);
2837 if (copy_error != 0 || spa_load_print_vdev_tree) {
2838 spa_load_note(spa, "provided vdev tree:");
2839 vdev_dbgmsg_print_tree(rvd, 2);
2840 spa_load_note(spa, "MOS vdev tree:");
2841 vdev_dbgmsg_print_tree(mrvd, 2);
2843 if (copy_error != 0) {
2844 spa_load_note(spa, "vdev_copy_path_strict failed, falling "
2845 "back to vdev_copy_path_relaxed");
2846 vdev_copy_path_relaxed(rvd, mrvd);
2851 spa->spa_root_vdev = mrvd;
2853 spa_config_exit(spa, SCL_ALL, FTAG);
2856 * We will use spa_config if we decide to reload the spa or if spa_load
2857 * fails and we rewind. We must thus regenerate the config using the
2858 * MOS information with the updated paths. Rewind policy is an import
2859 * setting and is not in the MOS. We copy it over to our new, trusted
2862 mos_config_txg = fnvlist_lookup_uint64(mos_config,
2863 ZPOOL_CONFIG_POOL_TXG);
2864 nvlist_free(mos_config);
2865 mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
2866 if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_REWIND_POLICY,
2868 fnvlist_add_nvlist(mos_config, ZPOOL_REWIND_POLICY, policy);
2869 spa_config_set(spa, mos_config);
2870 spa->spa_config_source = SPA_CONFIG_SRC_MOS;
2873 * Now that we got the config from the MOS, we should be more strict
2874 * in checking blkptrs and can make assumptions about the consistency
2875 * of the vdev tree. spa_trust_config must be set to true before opening
2876 * vdevs in order for them to be writeable.
2878 spa->spa_trust_config = B_TRUE;
2881 * Open and validate the new vdev tree
2883 error = spa_ld_open_vdevs(spa);
2887 error = spa_ld_validate_vdevs(spa);
2891 if (copy_error != 0 || spa_load_print_vdev_tree) {
2892 spa_load_note(spa, "final vdev tree:");
2893 vdev_dbgmsg_print_tree(rvd, 2);
2896 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT &&
2897 !spa->spa_extreme_rewind && zfs_max_missing_tvds == 0) {
2899 * Sanity check to make sure that we are indeed loading the
2900 * latest uberblock. If we missed SPA_SYNC_MIN_VDEVS tvds
2901 * in the config provided and they happened to be the only ones
2902 * to have the latest uberblock, we could involuntarily perform
2903 * an extreme rewind.
2905 healthy_tvds_mos = spa_healthy_core_tvds(spa);
2906 if (healthy_tvds_mos - healthy_tvds >=
2907 SPA_SYNC_MIN_VDEVS) {
2908 spa_load_note(spa, "config provided misses too many "
2909 "top-level vdevs compared to MOS (%lld vs %lld). ",
2910 (u_longlong_t)healthy_tvds,
2911 (u_longlong_t)healthy_tvds_mos);
2912 spa_load_note(spa, "vdev tree:");
2913 vdev_dbgmsg_print_tree(rvd, 2);
2915 spa_load_failed(spa, "config was already "
2916 "provided from MOS. Aborting.");
2917 return (spa_vdev_err(rvd,
2918 VDEV_AUX_CORRUPT_DATA, EIO));
2920 spa_load_note(spa, "spa must be reloaded using MOS "
2922 return (SET_ERROR(EAGAIN));
2926 error = spa_check_for_missing_logs(spa);
2928 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2930 if (rvd->vdev_guid_sum != spa->spa_uberblock.ub_guid_sum) {
2931 spa_load_failed(spa, "uberblock guid sum doesn't match MOS "
2932 "guid sum (%llu != %llu)",
2933 (u_longlong_t)spa->spa_uberblock.ub_guid_sum,
2934 (u_longlong_t)rvd->vdev_guid_sum);
2935 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2943 spa_ld_open_indirect_vdev_metadata(spa_t *spa)
2946 vdev_t *rvd = spa->spa_root_vdev;
2949 * Everything that we read before spa_remove_init() must be stored
2950 * on concreted vdevs. Therefore we do this as early as possible.
2952 error = spa_remove_init(spa);
2954 spa_load_failed(spa, "spa_remove_init failed [error=%d]",
2956 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2960 * Retrieve information needed to condense indirect vdev mappings.
2962 error = spa_condense_init(spa);
2964 spa_load_failed(spa, "spa_condense_init failed [error=%d]",
2966 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
2973 spa_ld_check_features(spa_t *spa, boolean_t *missing_feat_writep)
2976 vdev_t *rvd = spa->spa_root_vdev;
2978 if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2979 boolean_t missing_feat_read = B_FALSE;
2980 nvlist_t *unsup_feat, *enabled_feat;
2982 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2983 &spa->spa_feat_for_read_obj, B_TRUE) != 0) {
2984 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2987 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2988 &spa->spa_feat_for_write_obj, B_TRUE) != 0) {
2989 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2992 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2993 &spa->spa_feat_desc_obj, B_TRUE) != 0) {
2994 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2997 enabled_feat = fnvlist_alloc();
2998 unsup_feat = fnvlist_alloc();
3000 if (!spa_features_check(spa, B_FALSE,
3001 unsup_feat, enabled_feat))
3002 missing_feat_read = B_TRUE;
3004 if (spa_writeable(spa) ||
3005 spa->spa_load_state == SPA_LOAD_TRYIMPORT) {
3006 if (!spa_features_check(spa, B_TRUE,
3007 unsup_feat, enabled_feat)) {
3008 *missing_feat_writep = B_TRUE;
3012 fnvlist_add_nvlist(spa->spa_load_info,
3013 ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
3015 if (!nvlist_empty(unsup_feat)) {
3016 fnvlist_add_nvlist(spa->spa_load_info,
3017 ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
3020 fnvlist_free(enabled_feat);
3021 fnvlist_free(unsup_feat);
3023 if (!missing_feat_read) {
3024 fnvlist_add_boolean(spa->spa_load_info,
3025 ZPOOL_CONFIG_CAN_RDONLY);
3029 * If the state is SPA_LOAD_TRYIMPORT, our objective is
3030 * twofold: to determine whether the pool is available for
3031 * import in read-write mode and (if it is not) whether the
3032 * pool is available for import in read-only mode. If the pool
3033 * is available for import in read-write mode, it is displayed
3034 * as available in userland; if it is not available for import
3035 * in read-only mode, it is displayed as unavailable in
3036 * userland. If the pool is available for import in read-only
3037 * mode but not read-write mode, it is displayed as unavailable
3038 * in userland with a special note that the pool is actually
3039 * available for open in read-only mode.
3041 * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
3042 * missing a feature for write, we must first determine whether
3043 * the pool can be opened read-only before returning to
3044 * userland in order to know whether to display the
3045 * abovementioned note.
3047 if (missing_feat_read || (*missing_feat_writep &&
3048 spa_writeable(spa))) {
3049 spa_load_failed(spa, "pool uses unsupported features");
3050 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3055 * Load refcounts for ZFS features from disk into an in-memory
3056 * cache during SPA initialization.
3058 for (spa_feature_t i = 0; i < SPA_FEATURES; i++) {
3061 error = feature_get_refcount_from_disk(spa,
3062 &spa_feature_table[i], &refcount);
3064 spa->spa_feat_refcount_cache[i] = refcount;
3065 } else if (error == ENOTSUP) {
3066 spa->spa_feat_refcount_cache[i] =
3067 SPA_FEATURE_DISABLED;
3069 spa_load_failed(spa, "error getting refcount "
3070 "for feature %s [error=%d]",
3071 spa_feature_table[i].fi_guid, error);
3072 return (spa_vdev_err(rvd,
3073 VDEV_AUX_CORRUPT_DATA, EIO));
3078 if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
3079 if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
3080 &spa->spa_feat_enabled_txg_obj, B_TRUE) != 0)
3081 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3088 spa_ld_load_special_directories(spa_t *spa)
3091 vdev_t *rvd = spa->spa_root_vdev;
3093 spa->spa_is_initializing = B_TRUE;
3094 error = dsl_pool_open(spa->spa_dsl_pool);
3095 spa->spa_is_initializing = B_FALSE;
3097 spa_load_failed(spa, "dsl_pool_open failed [error=%d]", error);
3098 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3105 spa_ld_get_props(spa_t *spa)
3109 vdev_t *rvd = spa->spa_root_vdev;
3111 /* Grab the secret checksum salt from the MOS. */
3112 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3113 DMU_POOL_CHECKSUM_SALT, 1,
3114 sizeof (spa->spa_cksum_salt.zcs_bytes),
3115 spa->spa_cksum_salt.zcs_bytes);
3116 if (error == ENOENT) {
3117 /* Generate a new salt for subsequent use */
3118 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
3119 sizeof (spa->spa_cksum_salt.zcs_bytes));
3120 } else if (error != 0) {
3121 spa_load_failed(spa, "unable to retrieve checksum salt from "
3122 "MOS [error=%d]", error);
3123 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3126 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj, B_TRUE) != 0)
3127 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3128 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
3130 spa_load_failed(spa, "error opening deferred-frees bpobj "
3131 "[error=%d]", error);
3132 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3136 * Load the bit that tells us to use the new accounting function
3137 * (raid-z deflation). If we have an older pool, this will not
3140 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate, B_FALSE);
3141 if (error != 0 && error != ENOENT)
3142 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3144 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
3145 &spa->spa_creation_version, B_FALSE);
3146 if (error != 0 && error != ENOENT)
3147 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3150 * Load the persistent error log. If we have an older pool, this will
3153 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last,
3155 if (error != 0 && error != ENOENT)
3156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3158 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
3159 &spa->spa_errlog_scrub, B_FALSE);
3160 if (error != 0 && error != ENOENT)
3161 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3164 * Load the history object. If we have an older pool, this
3165 * will not be present.
3167 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history, B_FALSE);
3168 if (error != 0 && error != ENOENT)
3169 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3172 * Load the per-vdev ZAP map. If we have an older pool, this will not
3173 * be present; in this case, defer its creation to a later time to
3174 * avoid dirtying the MOS this early / out of sync context. See
3175 * spa_sync_config_object.
3178 /* The sentinel is only available in the MOS config. */
3179 nvlist_t *mos_config;
3180 if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0) {
3181 spa_load_failed(spa, "unable to retrieve MOS config");
3182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3185 error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
3186 &spa->spa_all_vdev_zaps, B_FALSE);
3188 if (error == ENOENT) {
3189 VERIFY(!nvlist_exists(mos_config,
3190 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
3191 spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
3192 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3193 } else if (error != 0) {
3194 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3195 } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
3197 * An older version of ZFS overwrote the sentinel value, so
3198 * we have orphaned per-vdev ZAPs in the MOS. Defer their
3199 * destruction to later; see spa_sync_config_object.
3201 spa->spa_avz_action = AVZ_ACTION_DESTROY;
3203 * We're assuming that no vdevs have had their ZAPs created
3204 * before this. Better be sure of it.
3206 ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
3208 nvlist_free(mos_config);
3210 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3212 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object,
3214 if (error && error != ENOENT)
3215 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3218 uint64_t autoreplace;
3220 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
3221 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
3222 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
3223 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
3224 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
3225 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
3226 &spa->spa_dedup_ditto);
3228 spa->spa_autoreplace = (autoreplace != 0);
3232 * If we are importing a pool with missing top-level vdevs,
3233 * we enforce that the pool doesn't panic or get suspended on
3234 * error since the likelihood of missing data is extremely high.
3236 if (spa->spa_missing_tvds > 0 &&
3237 spa->spa_failmode != ZIO_FAILURE_MODE_CONTINUE &&
3238 spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3239 spa_load_note(spa, "forcing failmode to 'continue' "
3240 "as some top level vdevs are missing");
3241 spa->spa_failmode = ZIO_FAILURE_MODE_CONTINUE;
3248 spa_ld_open_aux_vdevs(spa_t *spa, spa_import_type_t type)
3251 vdev_t *rvd = spa->spa_root_vdev;
3254 * If we're assembling the pool from the split-off vdevs of
3255 * an existing pool, we don't want to attach the spares & cache
3260 * Load any hot spares for this pool.
3262 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object,
3264 if (error != 0 && error != ENOENT)
3265 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3266 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3267 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
3268 if (load_nvlist(spa, spa->spa_spares.sav_object,
3269 &spa->spa_spares.sav_config) != 0) {
3270 spa_load_failed(spa, "error loading spares nvlist");
3271 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3274 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3275 spa_load_spares(spa);
3276 spa_config_exit(spa, SCL_ALL, FTAG);
3277 } else if (error == 0) {
3278 spa->spa_spares.sav_sync = B_TRUE;
3282 * Load any level 2 ARC devices for this pool.
3284 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
3285 &spa->spa_l2cache.sav_object, B_FALSE);
3286 if (error != 0 && error != ENOENT)
3287 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3288 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
3289 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
3290 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
3291 &spa->spa_l2cache.sav_config) != 0) {
3292 spa_load_failed(spa, "error loading l2cache nvlist");
3293 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3297 spa_load_l2cache(spa);
3298 spa_config_exit(spa, SCL_ALL, FTAG);
3299 } else if (error == 0) {
3300 spa->spa_l2cache.sav_sync = B_TRUE;
3307 spa_ld_load_vdev_metadata(spa_t *spa)
3310 vdev_t *rvd = spa->spa_root_vdev;
3313 * If the 'autoreplace' property is set, then post a resource notifying
3314 * the ZFS DE that it should not issue any faults for unopenable
3315 * devices. We also iterate over the vdevs, and post a sysevent for any
3316 * unopenable vdevs so that the normal autoreplace handler can take
3319 if (spa->spa_autoreplace && spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3320 spa_check_removed(spa->spa_root_vdev);
3322 * For the import case, this is done in spa_import(), because
3323 * at this point we're using the spare definitions from
3324 * the MOS config, not necessarily from the userland config.
3326 if (spa->spa_load_state != SPA_LOAD_IMPORT) {
3327 spa_aux_check_removed(&spa->spa_spares);
3328 spa_aux_check_removed(&spa->spa_l2cache);
3333 * Load the vdev metadata such as metaslabs, DTLs, spacemap object, etc.
3335 error = vdev_load(rvd);
3337 spa_load_failed(spa, "vdev_load failed [error=%d]", error);
3338 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, error));
3342 * Propagate the leaf DTLs we just loaded all the way up the vdev tree.
3344 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3345 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
3346 spa_config_exit(spa, SCL_ALL, FTAG);
3352 spa_ld_load_dedup_tables(spa_t *spa)
3355 vdev_t *rvd = spa->spa_root_vdev;
3357 error = ddt_load(spa);
3359 spa_load_failed(spa, "ddt_load failed [error=%d]", error);
3360 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3367 spa_ld_verify_logs(spa_t *spa, spa_import_type_t type, char **ereport)
3369 vdev_t *rvd = spa->spa_root_vdev;
3371 if (type != SPA_IMPORT_ASSEMBLE && spa_writeable(spa)) {
3372 boolean_t missing = spa_check_logs(spa);
3374 if (spa->spa_missing_tvds != 0) {
3375 spa_load_note(spa, "spa_check_logs failed "
3376 "so dropping the logs");
3378 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
3379 spa_load_failed(spa, "spa_check_logs failed");
3380 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG,
3390 spa_ld_verify_pool_data(spa_t *spa)
3393 vdev_t *rvd = spa->spa_root_vdev;
3396 * We've successfully opened the pool, verify that we're ready
3397 * to start pushing transactions.
3399 if (spa->spa_load_state != SPA_LOAD_TRYIMPORT) {
3400 error = spa_load_verify(spa);
3402 spa_load_failed(spa, "spa_load_verify failed "
3403 "[error=%d]", error);
3404 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3413 spa_ld_claim_log_blocks(spa_t *spa)
3416 dsl_pool_t *dp = spa_get_dsl(spa);
3419 * Claim log blocks that haven't been committed yet.
3420 * This must all happen in a single txg.
3421 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
3422 * invoked from zil_claim_log_block()'s i/o done callback.
3423 * Price of rollback is that we abandon the log.
3425 spa->spa_claiming = B_TRUE;
3427 tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
3428 (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3429 zil_claim, tx, DS_FIND_CHILDREN);
3432 spa->spa_claiming = B_FALSE;
3434 spa_set_log_state(spa, SPA_LOG_GOOD);
3438 spa_ld_check_for_config_update(spa_t *spa, uint64_t config_cache_txg,
3439 boolean_t update_config_cache)
3441 vdev_t *rvd = spa->spa_root_vdev;
3442 int need_update = B_FALSE;
3445 * If the config cache is stale, or we have uninitialized
3446 * metaslabs (see spa_vdev_add()), then update the config.
3448 * If this is a verbatim import, trust the current
3449 * in-core spa_config and update the disk labels.
3451 if (update_config_cache || config_cache_txg != spa->spa_config_txg ||
3452 spa->spa_load_state == SPA_LOAD_IMPORT ||
3453 spa->spa_load_state == SPA_LOAD_RECOVER ||
3454 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
3455 need_update = B_TRUE;
3457 for (int c = 0; c < rvd->vdev_children; c++)
3458 if (rvd->vdev_child[c]->vdev_ms_array == 0)
3459 need_update = B_TRUE;
3462 * Update the config cache asychronously in case we're the
3463 * root pool, in which case the config cache isn't writable yet.
3466 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
3470 spa_ld_prepare_for_reload(spa_t *spa)
3472 int mode = spa->spa_mode;
3473 int async_suspended = spa->spa_async_suspended;
3476 spa_deactivate(spa);
3477 spa_activate(spa, mode);
3480 * We save the value of spa_async_suspended as it gets reset to 0 by
3481 * spa_unload(). We want to restore it back to the original value before
3482 * returning as we might be calling spa_async_resume() later.
3484 spa->spa_async_suspended = async_suspended;
3488 spa_ld_read_checkpoint_txg(spa_t *spa)
3490 uberblock_t checkpoint;
3493 ASSERT0(spa->spa_checkpoint_txg);
3494 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3496 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3497 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3498 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3500 if (error == ENOENT)
3506 ASSERT3U(checkpoint.ub_txg, !=, 0);
3507 ASSERT3U(checkpoint.ub_checkpoint_txg, !=, 0);
3508 ASSERT3U(checkpoint.ub_timestamp, !=, 0);
3509 spa->spa_checkpoint_txg = checkpoint.ub_txg;
3510 spa->spa_checkpoint_info.sci_timestamp = checkpoint.ub_timestamp;
3516 spa_ld_mos_init(spa_t *spa, spa_import_type_t type)
3520 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3521 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3524 * Never trust the config that is provided unless we are assembling
3525 * a pool following a split.
3526 * This means don't trust blkptrs and the vdev tree in general. This
3527 * also effectively puts the spa in read-only mode since
3528 * spa_writeable() checks for spa_trust_config to be true.
3529 * We will later load a trusted config from the MOS.
3531 if (type != SPA_IMPORT_ASSEMBLE)
3532 spa->spa_trust_config = B_FALSE;
3535 * Parse the config provided to create a vdev tree.
3537 error = spa_ld_parse_config(spa, type);
3542 * Now that we have the vdev tree, try to open each vdev. This involves
3543 * opening the underlying physical device, retrieving its geometry and
3544 * probing the vdev with a dummy I/O. The state of each vdev will be set
3545 * based on the success of those operations. After this we'll be ready
3546 * to read from the vdevs.
3548 error = spa_ld_open_vdevs(spa);
3553 * Read the label of each vdev and make sure that the GUIDs stored
3554 * there match the GUIDs in the config provided.
3555 * If we're assembling a new pool that's been split off from an
3556 * existing pool, the labels haven't yet been updated so we skip
3557 * validation for now.
3559 if (type != SPA_IMPORT_ASSEMBLE) {
3560 error = spa_ld_validate_vdevs(spa);
3566 * Read all vdev labels to find the best uberblock (i.e. latest,
3567 * unless spa_load_max_txg is set) and store it in spa_uberblock. We
3568 * get the list of features required to read blkptrs in the MOS from
3569 * the vdev label with the best uberblock and verify that our version
3570 * of zfs supports them all.
3572 error = spa_ld_select_uberblock(spa, type);
3577 * Pass that uberblock to the dsl_pool layer which will open the root
3578 * blkptr. This blkptr points to the latest version of the MOS and will
3579 * allow us to read its contents.
3581 error = spa_ld_open_rootbp(spa);
3589 spa_ld_checkpoint_rewind(spa_t *spa)
3591 uberblock_t checkpoint;
3594 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3595 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3597 error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3598 DMU_POOL_ZPOOL_CHECKPOINT, sizeof (uint64_t),
3599 sizeof (uberblock_t) / sizeof (uint64_t), &checkpoint);
3602 spa_load_failed(spa, "unable to retrieve checkpointed "
3603 "uberblock from the MOS config [error=%d]", error);
3605 if (error == ENOENT)
3606 error = ZFS_ERR_NO_CHECKPOINT;
3611 ASSERT3U(checkpoint.ub_txg, <, spa->spa_uberblock.ub_txg);
3612 ASSERT3U(checkpoint.ub_txg, ==, checkpoint.ub_checkpoint_txg);
3615 * We need to update the txg and timestamp of the checkpointed
3616 * uberblock to be higher than the latest one. This ensures that
3617 * the checkpointed uberblock is selected if we were to close and
3618 * reopen the pool right after we've written it in the vdev labels.
3619 * (also see block comment in vdev_uberblock_compare)
3621 checkpoint.ub_txg = spa->spa_uberblock.ub_txg + 1;
3622 checkpoint.ub_timestamp = gethrestime_sec();
3625 * Set current uberblock to be the checkpointed uberblock.
3627 spa->spa_uberblock = checkpoint;
3630 * If we are doing a normal rewind, then the pool is open for
3631 * writing and we sync the "updated" checkpointed uberblock to
3632 * disk. Once this is done, we've basically rewound the whole
3633 * pool and there is no way back.
3635 * There are cases when we don't want to attempt and sync the
3636 * checkpointed uberblock to disk because we are opening a
3637 * pool as read-only. Specifically, verifying the checkpointed
3638 * state with zdb, and importing the checkpointed state to get
3639 * a "preview" of its content.
3641 if (spa_writeable(spa)) {
3642 vdev_t *rvd = spa->spa_root_vdev;
3644 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3645 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
3647 int children = rvd->vdev_children;
3648 int c0 = spa_get_random(children);
3650 for (int c = 0; c < children; c++) {
3651 vdev_t *vd = rvd->vdev_child[(c0 + c) % children];
3653 /* Stop when revisiting the first vdev */
3654 if (c > 0 && svd[0] == vd)
3657 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
3658 !vdev_is_concrete(vd))
3661 svd[svdcount++] = vd;
3662 if (svdcount == SPA_SYNC_MIN_VDEVS)
3665 error = vdev_config_sync(svd, svdcount, spa->spa_first_txg);
3667 spa->spa_last_synced_guid = rvd->vdev_guid;
3668 spa_config_exit(spa, SCL_ALL, FTAG);
3671 spa_load_failed(spa, "failed to write checkpointed "
3672 "uberblock to the vdev labels [error=%d]", error);
3681 spa_ld_mos_with_trusted_config(spa_t *spa, spa_import_type_t type,
3682 boolean_t *update_config_cache)
3687 * Parse the config for pool, open and validate vdevs,
3688 * select an uberblock, and use that uberblock to open
3691 error = spa_ld_mos_init(spa, type);
3696 * Retrieve the trusted config stored in the MOS and use it to create
3697 * a new, exact version of the vdev tree, then reopen all vdevs.
3699 error = spa_ld_trusted_config(spa, type, B_FALSE);
3700 if (error == EAGAIN) {
3701 if (update_config_cache != NULL)
3702 *update_config_cache = B_TRUE;
3705 * Redo the loading process with the trusted config if it is
3706 * too different from the untrusted config.
3708 spa_ld_prepare_for_reload(spa);
3709 spa_load_note(spa, "RELOADING");
3710 error = spa_ld_mos_init(spa, type);
3714 error = spa_ld_trusted_config(spa, type, B_TRUE);
3718 } else if (error != 0) {
3726 * Load an existing storage pool, using the config provided. This config
3727 * describes which vdevs are part of the pool and is later validated against
3728 * partial configs present in each vdev's label and an entire copy of the
3729 * config stored in the MOS.
3732 spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport)
3735 boolean_t missing_feat_write = B_FALSE;
3736 boolean_t checkpoint_rewind =
3737 (spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
3738 boolean_t update_config_cache = B_FALSE;
3740 ASSERT(MUTEX_HELD(&spa_namespace_lock));
3741 ASSERT(spa->spa_config_source != SPA_CONFIG_SRC_NONE);
3743 spa_load_note(spa, "LOADING");
3745 error = spa_ld_mos_with_trusted_config(spa, type, &update_config_cache);
3750 * If we are rewinding to the checkpoint then we need to repeat
3751 * everything we've done so far in this function but this time
3752 * selecting the checkpointed uberblock and using that to open
3755 if (checkpoint_rewind) {
3757 * If we are rewinding to the checkpoint update config cache
3760 update_config_cache = B_TRUE;
3763 * Extract the checkpointed uberblock from the current MOS
3764 * and use this as the pool's uberblock from now on. If the
3765 * pool is imported as writeable we also write the checkpoint
3766 * uberblock to the labels, making the rewind permanent.
3768 error = spa_ld_checkpoint_rewind(spa);
3773 * Redo the loading process process again with the
3774 * checkpointed uberblock.
3776 spa_ld_prepare_for_reload(spa);
3777 spa_load_note(spa, "LOADING checkpointed uberblock");
3778 error = spa_ld_mos_with_trusted_config(spa, type, NULL);
3784 * Retrieve the checkpoint txg if the pool has a checkpoint.
3786 error = spa_ld_read_checkpoint_txg(spa);
3791 * Retrieve the mapping of indirect vdevs. Those vdevs were removed
3792 * from the pool and their contents were re-mapped to other vdevs. Note
3793 * that everything that we read before this step must have been
3794 * rewritten on concrete vdevs after the last device removal was
3795 * initiated. Otherwise we could be reading from indirect vdevs before
3796 * we have loaded their mappings.
3798 error = spa_ld_open_indirect_vdev_metadata(spa);
3803 * Retrieve the full list of active features from the MOS and check if
3804 * they are all supported.
3806 error = spa_ld_check_features(spa, &missing_feat_write);
3811 * Load several special directories from the MOS needed by the dsl_pool
3814 error = spa_ld_load_special_directories(spa);
3819 * Retrieve pool properties from the MOS.
3821 error = spa_ld_get_props(spa);
3826 * Retrieve the list of auxiliary devices - cache devices and spares -
3829 error = spa_ld_open_aux_vdevs(spa, type);
3834 * Load the metadata for all vdevs. Also check if unopenable devices
3835 * should be autoreplaced.
3837 error = spa_ld_load_vdev_metadata(spa);
3841 error = spa_ld_load_dedup_tables(spa);
3846 * Verify the logs now to make sure we don't have any unexpected errors
3847 * when we claim log blocks later.
3849 error = spa_ld_verify_logs(spa, type, ereport);
3853 if (missing_feat_write) {
3854 ASSERT(spa->spa_load_state == SPA_LOAD_TRYIMPORT);
3857 * At this point, we know that we can open the pool in
3858 * read-only mode but not read-write mode. We now have enough
3859 * information and can return to userland.
3861 return (spa_vdev_err(spa->spa_root_vdev, VDEV_AUX_UNSUP_FEAT,
3866 * Traverse the last txgs to make sure the pool was left off in a safe
3867 * state. When performing an extreme rewind, we verify the whole pool,
3868 * which can take a very long time.
3870 error = spa_ld_verify_pool_data(spa);
3875 * Calculate the deflated space for the pool. This must be done before
3876 * we write anything to the pool because we'd need to update the space
3877 * accounting using the deflated sizes.
3879 spa_update_dspace(spa);
3882 * We have now retrieved all the information we needed to open the
3883 * pool. If we are importing the pool in read-write mode, a few
3884 * additional steps must be performed to finish the import.
3886 if (spa_writeable(spa) && (spa->spa_load_state == SPA_LOAD_RECOVER ||
3887 spa->spa_load_max_txg == UINT64_MAX)) {
3888 uint64_t config_cache_txg = spa->spa_config_txg;
3890 ASSERT(spa->spa_load_state != SPA_LOAD_TRYIMPORT);
3893 * In case of a checkpoint rewind, log the original txg
3894 * of the checkpointed uberblock.
3896 if (checkpoint_rewind) {
3897 spa_history_log_internal(spa, "checkpoint rewind",
3898 NULL, "rewound state to txg=%llu",
3899 (u_longlong_t)spa->spa_uberblock.ub_checkpoint_txg);
3903 * Traverse the ZIL and claim all blocks.
3905 spa_ld_claim_log_blocks(spa);
3908 * Kick-off the syncing thread.
3910 spa->spa_sync_on = B_TRUE;
3911 txg_sync_start(spa->spa_dsl_pool);
3914 * Wait for all claims to sync. We sync up to the highest
3915 * claimed log block birth time so that claimed log blocks
3916 * don't appear to be from the future. spa_claim_max_txg
3917 * will have been set for us by ZIL traversal operations
3920 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
3923 * Check if we need to request an update of the config. On the
3924 * next sync, we would update the config stored in vdev labels
3925 * and the cachefile (by default /etc/zfs/zpool.cache).
3927 spa_ld_check_for_config_update(spa, config_cache_txg,
3928 update_config_cache);
3931 * Check all DTLs to see if anything needs resilvering.
3933 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
3934 vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
3935 spa_async_request(spa, SPA_ASYNC_RESILVER);
3938 * Log the fact that we booted up (so that we can detect if
3939 * we rebooted in the middle of an operation).
3941 spa_history_log_version(spa, "open");
3944 * Delete any inconsistent datasets.
3946 (void) dmu_objset_find(spa_name(spa),
3947 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
3950 * Clean up any stale temporary dataset userrefs.
3952 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
3954 spa_restart_removal(spa);
3956 spa_spawn_aux_threads(spa);
3959 spa_load_note(spa, "LOADED");
3965 spa_load_retry(spa_t *spa, spa_load_state_t state)
3967 int mode = spa->spa_mode;
3970 spa_deactivate(spa);
3972 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
3974 spa_activate(spa, mode);
3975 spa_async_suspend(spa);
3977 spa_load_note(spa, "spa_load_retry: rewind, max txg: %llu",
3978 (u_longlong_t)spa->spa_load_max_txg);
3980 return (spa_load(spa, state, SPA_IMPORT_EXISTING));
3984 * If spa_load() fails this function will try loading prior txg's. If
3985 * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
3986 * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
3987 * function will not rewind the pool and will return the same error as
3991 spa_load_best(spa_t *spa, spa_load_state_t state, uint64_t max_request,
3994 nvlist_t *loadinfo = NULL;
3995 nvlist_t *config = NULL;
3996 int load_error, rewind_error;
3997 uint64_t safe_rewind_txg;
4000 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
4001 spa->spa_load_max_txg = spa->spa_load_txg;
4002 spa_set_log_state(spa, SPA_LOG_CLEAR);
4004 spa->spa_load_max_txg = max_request;
4005 if (max_request != UINT64_MAX)
4006 spa->spa_extreme_rewind = B_TRUE;
4009 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING);
4010 if (load_error == 0)
4012 if (load_error == ZFS_ERR_NO_CHECKPOINT) {
4014 * When attempting checkpoint-rewind on a pool with no
4015 * checkpoint, we should not attempt to load uberblocks
4016 * from previous txgs when spa_load fails.
4018 ASSERT(spa->spa_import_flags & ZFS_IMPORT_CHECKPOINT);
4019 return (load_error);
4022 if (spa->spa_root_vdev != NULL)
4023 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4025 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
4026 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
4028 if (rewind_flags & ZPOOL_NEVER_REWIND) {
4029 nvlist_free(config);
4030 return (load_error);
4033 if (state == SPA_LOAD_RECOVER) {
4034 /* Price of rolling back is discarding txgs, including log */
4035 spa_set_log_state(spa, SPA_LOG_CLEAR);
4038 * If we aren't rolling back save the load info from our first
4039 * import attempt so that we can restore it after attempting
4042 loadinfo = spa->spa_load_info;
4043 spa->spa_load_info = fnvlist_alloc();
4046 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
4047 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
4048 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
4049 TXG_INITIAL : safe_rewind_txg;
4052 * Continue as long as we're finding errors, we're still within
4053 * the acceptable rewind range, and we're still finding uberblocks
4055 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
4056 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
4057 if (spa->spa_load_max_txg < safe_rewind_txg)
4058 spa->spa_extreme_rewind = B_TRUE;
4059 rewind_error = spa_load_retry(spa, state);
4062 spa->spa_extreme_rewind = B_FALSE;
4063 spa->spa_load_max_txg = UINT64_MAX;
4065 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
4066 spa_config_set(spa, config);
4068 nvlist_free(config);
4070 if (state == SPA_LOAD_RECOVER) {
4071 ASSERT3P(loadinfo, ==, NULL);
4072 return (rewind_error);
4074 /* Store the rewind info as part of the initial load info */
4075 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
4076 spa->spa_load_info);
4078 /* Restore the initial load info */
4079 fnvlist_free(spa->spa_load_info);
4080 spa->spa_load_info = loadinfo;
4082 return (load_error);
4089 * The import case is identical to an open except that the configuration is sent
4090 * down from userland, instead of grabbed from the configuration cache. For the
4091 * case of an open, the pool configuration will exist in the
4092 * POOL_STATE_UNINITIALIZED state.
4094 * The stats information (gen/count/ustats) is used to gather vdev statistics at
4095 * the same time open the pool, without having to keep around the spa_t in some
4099 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
4103 spa_load_state_t state = SPA_LOAD_OPEN;
4105 int locked = B_FALSE;
4106 int firstopen = B_FALSE;
4111 * As disgusting as this is, we need to support recursive calls to this
4112 * function because dsl_dir_open() is called during spa_load(), and ends
4113 * up calling spa_open() again. The real fix is to figure out how to
4114 * avoid dsl_dir_open() calling this in the first place.
4116 if (mutex_owner(&spa_namespace_lock) != curthread) {
4117 mutex_enter(&spa_namespace_lock);
4121 if ((spa = spa_lookup(pool)) == NULL) {
4123 mutex_exit(&spa_namespace_lock);
4124 return (SET_ERROR(ENOENT));
4127 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
4128 zpool_rewind_policy_t policy;
4132 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
4134 if (policy.zrp_request & ZPOOL_DO_REWIND)
4135 state = SPA_LOAD_RECOVER;
4137 spa_activate(spa, spa_mode_global);
4139 if (state != SPA_LOAD_RECOVER)
4140 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
4141 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
4143 zfs_dbgmsg("spa_open_common: opening %s", pool);
4144 error = spa_load_best(spa, state, policy.zrp_txg,
4145 policy.zrp_request);
4147 if (error == EBADF) {
4149 * If vdev_validate() returns failure (indicated by
4150 * EBADF), it indicates that one of the vdevs indicates
4151 * that the pool has been exported or destroyed. If
4152 * this is the case, the config cache is out of sync and
4153 * we should remove the pool from the namespace.
4156 spa_deactivate(spa);
4157 spa_write_cachefile(spa, B_TRUE, B_TRUE);
4160 mutex_exit(&spa_namespace_lock);
4161 return (SET_ERROR(ENOENT));
4166 * We can't open the pool, but we still have useful
4167 * information: the state of each vdev after the
4168 * attempted vdev_open(). Return this to the user.
4170 if (config != NULL && spa->spa_config) {
4171 VERIFY(nvlist_dup(spa->spa_config, config,
4173 VERIFY(nvlist_add_nvlist(*config,
4174 ZPOOL_CONFIG_LOAD_INFO,
4175 spa->spa_load_info) == 0);
4178 spa_deactivate(spa);
4179 spa->spa_last_open_failed = error;
4181 mutex_exit(&spa_namespace_lock);
4187 spa_open_ref(spa, tag);
4190 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4193 * If we've recovered the pool, pass back any information we
4194 * gathered while doing the load.
4196 if (state == SPA_LOAD_RECOVER) {
4197 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
4198 spa->spa_load_info) == 0);
4202 spa->spa_last_open_failed = 0;
4203 spa->spa_last_ubsync_txg = 0;
4204 spa->spa_load_txg = 0;
4205 mutex_exit(&spa_namespace_lock);
4209 zvol_create_minors(spa->spa_name);
4220 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
4223 return (spa_open_common(name, spapp, tag, policy, config));
4227 spa_open(const char *name, spa_t **spapp, void *tag)
4229 return (spa_open_common(name, spapp, tag, NULL, NULL));
4233 * Lookup the given spa_t, incrementing the inject count in the process,
4234 * preventing it from being exported or destroyed.
4237 spa_inject_addref(char *name)
4241 mutex_enter(&spa_namespace_lock);
4242 if ((spa = spa_lookup(name)) == NULL) {
4243 mutex_exit(&spa_namespace_lock);
4246 spa->spa_inject_ref++;
4247 mutex_exit(&spa_namespace_lock);
4253 spa_inject_delref(spa_t *spa)
4255 mutex_enter(&spa_namespace_lock);
4256 spa->spa_inject_ref--;
4257 mutex_exit(&spa_namespace_lock);
4261 * Add spares device information to the nvlist.
4264 spa_add_spares(spa_t *spa, nvlist_t *config)
4274 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4276 if (spa->spa_spares.sav_count == 0)
4279 VERIFY(nvlist_lookup_nvlist(config,
4280 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4281 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4282 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4284 VERIFY(nvlist_add_nvlist_array(nvroot,
4285 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4286 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4287 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
4290 * Go through and find any spares which have since been
4291 * repurposed as an active spare. If this is the case, update
4292 * their status appropriately.
4294 for (i = 0; i < nspares; i++) {
4295 VERIFY(nvlist_lookup_uint64(spares[i],
4296 ZPOOL_CONFIG_GUID, &guid) == 0);
4297 if (spa_spare_exists(guid, &pool, NULL) &&
4299 VERIFY(nvlist_lookup_uint64_array(
4300 spares[i], ZPOOL_CONFIG_VDEV_STATS,
4301 (uint64_t **)&vs, &vsc) == 0);
4302 vs->vs_state = VDEV_STATE_CANT_OPEN;
4303 vs->vs_aux = VDEV_AUX_SPARED;
4310 * Add l2cache device information to the nvlist, including vdev stats.
4313 spa_add_l2cache(spa_t *spa, nvlist_t *config)
4316 uint_t i, j, nl2cache;
4323 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4325 if (spa->spa_l2cache.sav_count == 0)
4328 VERIFY(nvlist_lookup_nvlist(config,
4329 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4330 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4331 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4332 if (nl2cache != 0) {
4333 VERIFY(nvlist_add_nvlist_array(nvroot,
4334 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4335 VERIFY(nvlist_lookup_nvlist_array(nvroot,
4336 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
4339 * Update level 2 cache device stats.
4342 for (i = 0; i < nl2cache; i++) {
4343 VERIFY(nvlist_lookup_uint64(l2cache[i],
4344 ZPOOL_CONFIG_GUID, &guid) == 0);
4347 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
4349 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
4350 vd = spa->spa_l2cache.sav_vdevs[j];
4356 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
4357 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
4359 vdev_get_stats(vd, vs);
4365 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
4371 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
4372 VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4374 /* We may be unable to read features if pool is suspended. */
4375 if (spa_suspended(spa))
4378 if (spa->spa_feat_for_read_obj != 0) {
4379 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4380 spa->spa_feat_for_read_obj);
4381 zap_cursor_retrieve(&zc, &za) == 0;
4382 zap_cursor_advance(&zc)) {
4383 ASSERT(za.za_integer_length == sizeof (uint64_t) &&
4384 za.za_num_integers == 1);
4385 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4386 za.za_first_integer));
4388 zap_cursor_fini(&zc);
4391 if (spa->spa_feat_for_write_obj != 0) {
4392 for (zap_cursor_init(&zc, spa->spa_meta_objset,
4393 spa->spa_feat_for_write_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 VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
4399 za.za_first_integer));
4401 zap_cursor_fini(&zc);
4405 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
4407 nvlist_free(features);
4411 spa_get_stats(const char *name, nvlist_t **config,
4412 char *altroot, size_t buflen)
4418 error = spa_open_common(name, &spa, FTAG, NULL, config);
4422 * This still leaves a window of inconsistency where the spares
4423 * or l2cache devices could change and the config would be
4424 * self-inconsistent.
4426 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4428 if (*config != NULL) {
4429 uint64_t loadtimes[2];
4431 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
4432 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
4433 VERIFY(nvlist_add_uint64_array(*config,
4434 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
4436 VERIFY(nvlist_add_uint64(*config,
4437 ZPOOL_CONFIG_ERRCOUNT,
4438 spa_get_errlog_size(spa)) == 0);
4440 if (spa_suspended(spa))
4441 VERIFY(nvlist_add_uint64(*config,
4442 ZPOOL_CONFIG_SUSPENDED,
4443 spa->spa_failmode) == 0);
4445 spa_add_spares(spa, *config);
4446 spa_add_l2cache(spa, *config);
4447 spa_add_feature_stats(spa, *config);
4452 * We want to get the alternate root even for faulted pools, so we cheat
4453 * and call spa_lookup() directly.
4457 mutex_enter(&spa_namespace_lock);
4458 spa = spa_lookup(name);
4460 spa_altroot(spa, altroot, buflen);
4464 mutex_exit(&spa_namespace_lock);
4466 spa_altroot(spa, altroot, buflen);
4471 spa_config_exit(spa, SCL_CONFIG, FTAG);
4472 spa_close(spa, FTAG);
4479 * Validate that the auxiliary device array is well formed. We must have an
4480 * array of nvlists, each which describes a valid leaf vdev. If this is an
4481 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
4482 * specified, as long as they are well-formed.
4485 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
4486 spa_aux_vdev_t *sav, const char *config, uint64_t version,
4487 vdev_labeltype_t label)
4494 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4497 * It's acceptable to have no devs specified.
4499 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
4503 return (SET_ERROR(EINVAL));
4506 * Make sure the pool is formatted with a version that supports this
4509 if (spa_version(spa) < version)
4510 return (SET_ERROR(ENOTSUP));
4513 * Set the pending device list so we correctly handle device in-use
4516 sav->sav_pending = dev;
4517 sav->sav_npending = ndev;
4519 for (i = 0; i < ndev; i++) {
4520 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
4524 if (!vd->vdev_ops->vdev_op_leaf) {
4526 error = SET_ERROR(EINVAL);
4531 * The L2ARC currently only supports disk devices in
4532 * kernel context. For user-level testing, we allow it.
4535 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
4536 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
4537 error = SET_ERROR(ENOTBLK);
4544 if ((error = vdev_open(vd)) == 0 &&
4545 (error = vdev_label_init(vd, crtxg, label)) == 0) {
4546 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
4547 vd->vdev_guid) == 0);
4553 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
4560 sav->sav_pending = NULL;
4561 sav->sav_npending = 0;
4566 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
4570 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4572 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4573 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
4574 VDEV_LABEL_SPARE)) != 0) {
4578 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
4579 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
4580 VDEV_LABEL_L2CACHE));
4584 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
4589 if (sav->sav_config != NULL) {
4595 * Generate new dev list by concatentating with the
4598 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
4599 &olddevs, &oldndevs) == 0);
4601 newdevs = kmem_alloc(sizeof (void *) *
4602 (ndevs + oldndevs), KM_SLEEP);
4603 for (i = 0; i < oldndevs; i++)
4604 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
4606 for (i = 0; i < ndevs; i++)
4607 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
4610 VERIFY(nvlist_remove(sav->sav_config, config,
4611 DATA_TYPE_NVLIST_ARRAY) == 0);
4613 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
4614 config, newdevs, ndevs + oldndevs) == 0);
4615 for (i = 0; i < oldndevs + ndevs; i++)
4616 nvlist_free(newdevs[i]);
4617 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
4620 * Generate a new dev list.
4622 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
4624 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
4630 * Stop and drop level 2 ARC devices
4633 spa_l2cache_drop(spa_t *spa)
4637 spa_aux_vdev_t *sav = &spa->spa_l2cache;
4639 for (i = 0; i < sav->sav_count; i++) {
4642 vd = sav->sav_vdevs[i];
4645 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
4646 pool != 0ULL && l2arc_vdev_present(vd))
4647 l2arc_remove_vdev(vd);
4655 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
4659 char *altroot = NULL;
4664 uint64_t txg = TXG_INITIAL;
4665 nvlist_t **spares, **l2cache;
4666 uint_t nspares, nl2cache;
4667 uint64_t version, obj;
4668 boolean_t has_features;
4671 * If this pool already exists, return failure.
4673 mutex_enter(&spa_namespace_lock);
4674 if (spa_lookup(pool) != NULL) {
4675 mutex_exit(&spa_namespace_lock);
4676 return (SET_ERROR(EEXIST));
4680 * Allocate a new spa_t structure.
4682 (void) nvlist_lookup_string(props,
4683 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4684 spa = spa_add(pool, NULL, altroot);
4685 spa_activate(spa, spa_mode_global);
4687 if (props && (error = spa_prop_validate(spa, props))) {
4688 spa_deactivate(spa);
4690 mutex_exit(&spa_namespace_lock);
4694 has_features = B_FALSE;
4695 for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
4696 elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
4697 if (zpool_prop_feature(nvpair_name(elem)))
4698 has_features = B_TRUE;
4701 if (has_features || nvlist_lookup_uint64(props,
4702 zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
4703 version = SPA_VERSION;
4705 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
4707 spa->spa_first_txg = txg;
4708 spa->spa_uberblock.ub_txg = txg - 1;
4709 spa->spa_uberblock.ub_version = version;
4710 spa->spa_ubsync = spa->spa_uberblock;
4711 spa->spa_load_state = SPA_LOAD_CREATE;
4712 spa->spa_removing_phys.sr_state = DSS_NONE;
4713 spa->spa_removing_phys.sr_removing_vdev = -1;
4714 spa->spa_removing_phys.sr_prev_indirect_vdev = -1;
4717 * Create "The Godfather" zio to hold all async IOs
4719 spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
4721 for (int i = 0; i < max_ncpus; i++) {
4722 spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
4723 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
4724 ZIO_FLAG_GODFATHER);
4728 * Create the root vdev.
4730 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4732 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
4734 ASSERT(error != 0 || rvd != NULL);
4735 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
4737 if (error == 0 && !zfs_allocatable_devs(nvroot))
4738 error = SET_ERROR(EINVAL);
4741 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
4742 (error = spa_validate_aux(spa, nvroot, txg,
4743 VDEV_ALLOC_ADD)) == 0) {
4744 for (int c = 0; c < rvd->vdev_children; c++) {
4745 vdev_ashift_optimize(rvd->vdev_child[c]);
4746 vdev_metaslab_set_size(rvd->vdev_child[c]);
4747 vdev_expand(rvd->vdev_child[c], txg);
4751 spa_config_exit(spa, SCL_ALL, FTAG);
4755 spa_deactivate(spa);
4757 mutex_exit(&spa_namespace_lock);
4762 * Get the list of spares, if specified.
4764 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
4765 &spares, &nspares) == 0) {
4766 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
4768 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
4769 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
4770 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4771 spa_load_spares(spa);
4772 spa_config_exit(spa, SCL_ALL, FTAG);
4773 spa->spa_spares.sav_sync = B_TRUE;
4777 * Get the list of level 2 cache devices, if specified.
4779 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
4780 &l2cache, &nl2cache) == 0) {
4781 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
4782 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4783 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
4784 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
4785 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4786 spa_load_l2cache(spa);
4787 spa_config_exit(spa, SCL_ALL, FTAG);
4788 spa->spa_l2cache.sav_sync = B_TRUE;
4791 spa->spa_is_initializing = B_TRUE;
4792 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
4793 spa->spa_meta_objset = dp->dp_meta_objset;
4794 spa->spa_is_initializing = B_FALSE;
4797 * Create DDTs (dedup tables).
4801 spa_update_dspace(spa);
4803 tx = dmu_tx_create_assigned(dp, txg);
4806 * Create the pool config object.
4808 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
4809 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
4810 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
4812 if (zap_add(spa->spa_meta_objset,
4813 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
4814 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
4815 cmn_err(CE_PANIC, "failed to add pool config");
4818 if (spa_version(spa) >= SPA_VERSION_FEATURES)
4819 spa_feature_create_zap_objects(spa, tx);
4821 if (zap_add(spa->spa_meta_objset,
4822 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
4823 sizeof (uint64_t), 1, &version, tx) != 0) {
4824 cmn_err(CE_PANIC, "failed to add pool version");
4827 /* Newly created pools with the right version are always deflated. */
4828 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
4829 spa->spa_deflate = TRUE;
4830 if (zap_add(spa->spa_meta_objset,
4831 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
4832 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
4833 cmn_err(CE_PANIC, "failed to add deflate");
4838 * Create the deferred-free bpobj. Turn off compression
4839 * because sync-to-convergence takes longer if the blocksize
4842 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
4843 dmu_object_set_compress(spa->spa_meta_objset, obj,
4844 ZIO_COMPRESS_OFF, tx);
4845 if (zap_add(spa->spa_meta_objset,
4846 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
4847 sizeof (uint64_t), 1, &obj, tx) != 0) {
4848 cmn_err(CE_PANIC, "failed to add bpobj");
4850 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
4851 spa->spa_meta_objset, obj));
4854 * Create the pool's history object.
4856 if (version >= SPA_VERSION_ZPOOL_HISTORY)
4857 spa_history_create_obj(spa, tx);
4860 * Generate some random noise for salted checksums to operate on.
4862 (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
4863 sizeof (spa->spa_cksum_salt.zcs_bytes));
4866 * Set pool properties.
4868 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
4869 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
4870 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
4871 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
4873 if (props != NULL) {
4874 spa_configfile_set(spa, props, B_FALSE);
4875 spa_sync_props(props, tx);
4880 spa->spa_sync_on = B_TRUE;
4881 txg_sync_start(spa->spa_dsl_pool);
4884 * We explicitly wait for the first transaction to complete so that our
4885 * bean counters are appropriately updated.
4887 txg_wait_synced(spa->spa_dsl_pool, txg);
4889 spa_spawn_aux_threads(spa);
4891 spa_write_cachefile(spa, B_FALSE, B_TRUE);
4892 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
4894 spa_history_log_version(spa, "create");
4897 * Don't count references from objsets that are already closed
4898 * and are making their way through the eviction process.
4900 spa_evicting_os_wait(spa);
4901 spa->spa_minref = refcount_count(&spa->spa_refcount);
4902 spa->spa_load_state = SPA_LOAD_NONE;
4904 mutex_exit(&spa_namespace_lock);
4912 * Get the root pool information from the root disk, then import the root pool
4913 * during the system boot up time.
4915 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
4918 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
4921 nvlist_t *nvtop, *nvroot;
4924 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
4928 * Add this top-level vdev to the child array.
4930 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4932 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4934 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
4937 * Put this pool's top-level vdevs into a root vdev.
4939 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4940 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
4941 VDEV_TYPE_ROOT) == 0);
4942 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
4943 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
4944 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4948 * Replace the existing vdev_tree with the new root vdev in
4949 * this pool's configuration (remove the old, add the new).
4951 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
4952 nvlist_free(nvroot);
4957 * Walk the vdev tree and see if we can find a device with "better"
4958 * configuration. A configuration is "better" if the label on that
4959 * device has a more recent txg.
4962 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
4964 for (int c = 0; c < vd->vdev_children; c++)
4965 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
4967 if (vd->vdev_ops->vdev_op_leaf) {
4971 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
4975 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
4979 * Do we have a better boot device?
4981 if (label_txg > *txg) {
4990 * Import a root pool.
4992 * For x86. devpath_list will consist of devid and/or physpath name of
4993 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
4994 * The GRUB "findroot" command will return the vdev we should boot.
4996 * For Sparc, devpath_list consists the physpath name of the booting device
4997 * no matter the rootpool is a single device pool or a mirrored pool.
4999 * "/pci@1f,0/ide@d/disk@0,0:a"
5002 spa_import_rootpool(char *devpath, char *devid)
5005 vdev_t *rvd, *bvd, *avd = NULL;
5006 nvlist_t *config, *nvtop;
5012 * Read the label from the boot device and generate a configuration.
5014 config = spa_generate_rootconf(devpath, devid, &guid);
5015 #if defined(_OBP) && defined(_KERNEL)
5016 if (config == NULL) {
5017 if (strstr(devpath, "/iscsi/ssd") != NULL) {
5019 get_iscsi_bootpath_phy(devpath);
5020 config = spa_generate_rootconf(devpath, devid, &guid);
5024 if (config == NULL) {
5025 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
5027 return (SET_ERROR(EIO));
5030 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5032 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
5034 mutex_enter(&spa_namespace_lock);
5035 if ((spa = spa_lookup(pname)) != NULL) {
5037 * Remove the existing root pool from the namespace so that we
5038 * can replace it with the correct config we just read in.
5043 spa = spa_add(pname, config, NULL);
5044 spa->spa_is_root = B_TRUE;
5045 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5046 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5047 &spa->spa_ubsync.ub_version) != 0)
5048 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5051 * Build up a vdev tree based on the boot device's label config.
5053 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5055 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5056 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5057 VDEV_ALLOC_ROOTPOOL);
5058 spa_config_exit(spa, SCL_ALL, FTAG);
5060 mutex_exit(&spa_namespace_lock);
5061 nvlist_free(config);
5062 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5068 * Get the boot vdev.
5070 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
5071 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
5072 (u_longlong_t)guid);
5073 error = SET_ERROR(ENOENT);
5078 * Determine if there is a better boot device.
5081 spa_alt_rootvdev(rvd, &avd, &txg);
5083 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
5084 "try booting from '%s'", avd->vdev_path);
5085 error = SET_ERROR(EINVAL);
5090 * If the boot device is part of a spare vdev then ensure that
5091 * we're booting off the active spare.
5093 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
5094 !bvd->vdev_isspare) {
5095 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
5096 "try booting from '%s'",
5098 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
5099 error = SET_ERROR(EINVAL);
5105 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5107 spa_config_exit(spa, SCL_ALL, FTAG);
5108 mutex_exit(&spa_namespace_lock);
5110 nvlist_free(config);
5114 #else /* !illumos */
5116 extern int vdev_geom_read_pool_label(const char *name, nvlist_t ***configs,
5120 spa_generate_rootconf(const char *name)
5122 nvlist_t **configs, **tops;
5124 nvlist_t *best_cfg, *nvtop, *nvroot;
5133 if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
5136 ASSERT3U(count, !=, 0);
5138 for (i = 0; i < count; i++) {
5141 VERIFY(nvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG,
5143 if (txg > best_txg) {
5145 best_cfg = configs[i];
5150 nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
5152 nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
5155 tops = kmem_zalloc(nchildren * sizeof(void *), KM_SLEEP);
5156 for (i = 0; i < nchildren; i++) {
5159 if (configs[i] == NULL)
5161 VERIFY(nvlist_lookup_nvlist(configs[i], ZPOOL_CONFIG_VDEV_TREE,
5163 nvlist_dup(nvtop, &tops[i], KM_SLEEP);
5165 for (i = 0; holes != NULL && i < nholes; i++) {
5168 if (tops[holes[i]] != NULL)
5170 nvlist_alloc(&tops[holes[i]], NV_UNIQUE_NAME, KM_SLEEP);
5171 VERIFY(nvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
5172 VDEV_TYPE_HOLE) == 0);
5173 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID,
5175 VERIFY(nvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID,
5178 for (i = 0; i < nchildren; i++) {
5179 if (tops[i] != NULL)
5181 nvlist_alloc(&tops[i], NV_UNIQUE_NAME, KM_SLEEP);
5182 VERIFY(nvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
5183 VDEV_TYPE_MISSING) == 0);
5184 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID,
5186 VERIFY(nvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID,
5191 * Create pool config based on the best vdev config.
5193 nvlist_dup(best_cfg, &config, KM_SLEEP);
5196 * Put this pool's top-level vdevs into a root vdev.
5198 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
5200 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5201 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
5202 VDEV_TYPE_ROOT) == 0);
5203 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
5204 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
5205 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
5206 tops, nchildren) == 0);
5209 * Replace the existing vdev_tree with the new root vdev in
5210 * this pool's configuration (remove the old, add the new).
5212 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
5215 * Drop vdev config elements that should not be present at pool level.
5217 nvlist_remove(config, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64);
5218 nvlist_remove(config, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64);
5220 for (i = 0; i < count; i++)
5221 nvlist_free(configs[i]);
5222 kmem_free(configs, count * sizeof(void *));
5223 for (i = 0; i < nchildren; i++)
5224 nvlist_free(tops[i]);
5225 kmem_free(tops, nchildren * sizeof(void *));
5226 nvlist_free(nvroot);
5231 spa_import_rootpool(const char *name)
5234 vdev_t *rvd, *bvd, *avd = NULL;
5235 nvlist_t *config, *nvtop;
5241 * Read the label from the boot device and generate a configuration.
5243 config = spa_generate_rootconf(name);
5245 mutex_enter(&spa_namespace_lock);
5246 if (config != NULL) {
5247 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
5248 &pname) == 0 && strcmp(name, pname) == 0);
5249 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg)
5252 if ((spa = spa_lookup(pname)) != NULL) {
5254 * The pool could already be imported,
5255 * e.g., after reboot -r.
5257 if (spa->spa_state == POOL_STATE_ACTIVE) {
5258 mutex_exit(&spa_namespace_lock);
5259 nvlist_free(config);
5264 * Remove the existing root pool from the namespace so
5265 * that we can replace it with the correct config
5270 spa = spa_add(pname, config, NULL);
5273 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
5274 * via spa_version().
5276 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
5277 &spa->spa_ubsync.ub_version) != 0)
5278 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
5279 } else if ((spa = spa_lookup(name)) == NULL) {
5280 mutex_exit(&spa_namespace_lock);
5281 nvlist_free(config);
5282 cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
5286 VERIFY(nvlist_dup(spa->spa_config, &config, KM_SLEEP) == 0);
5288 spa->spa_is_root = B_TRUE;
5289 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
5292 * Build up a vdev tree based on the boot device's label config.
5294 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5296 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5297 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
5298 VDEV_ALLOC_ROOTPOOL);
5299 spa_config_exit(spa, SCL_ALL, FTAG);
5301 mutex_exit(&spa_namespace_lock);
5302 nvlist_free(config);
5303 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
5308 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5310 spa_config_exit(spa, SCL_ALL, FTAG);
5311 mutex_exit(&spa_namespace_lock);
5313 nvlist_free(config);
5317 #endif /* illumos */
5318 #endif /* _KERNEL */
5321 * Import a non-root pool into the system.
5324 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
5327 char *altroot = NULL;
5328 spa_load_state_t state = SPA_LOAD_IMPORT;
5329 zpool_rewind_policy_t policy;
5330 uint64_t mode = spa_mode_global;
5331 uint64_t readonly = B_FALSE;
5334 nvlist_t **spares, **l2cache;
5335 uint_t nspares, nl2cache;
5338 * If a pool with this name exists, return failure.
5340 mutex_enter(&spa_namespace_lock);
5341 if (spa_lookup(pool) != NULL) {
5342 mutex_exit(&spa_namespace_lock);
5343 return (SET_ERROR(EEXIST));
5347 * Create and initialize the spa structure.
5349 (void) nvlist_lookup_string(props,
5350 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
5351 (void) nvlist_lookup_uint64(props,
5352 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
5355 spa = spa_add(pool, config, altroot);
5356 spa->spa_import_flags = flags;
5359 * Verbatim import - Take a pool and insert it into the namespace
5360 * as if it had been loaded at boot.
5362 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
5364 spa_configfile_set(spa, props, B_FALSE);
5366 spa_write_cachefile(spa, B_FALSE, B_TRUE);
5367 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5368 zfs_dbgmsg("spa_import: verbatim import of %s", pool);
5369 mutex_exit(&spa_namespace_lock);
5373 spa_activate(spa, mode);
5376 * Don't start async tasks until we know everything is healthy.
5378 spa_async_suspend(spa);
5380 zpool_get_rewind_policy(config, &policy);
5381 if (policy.zrp_request & ZPOOL_DO_REWIND)
5382 state = SPA_LOAD_RECOVER;
5384 spa->spa_config_source = SPA_CONFIG_SRC_TRYIMPORT;
5386 if (state != SPA_LOAD_RECOVER) {
5387 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
5388 zfs_dbgmsg("spa_import: importing %s", pool);
5390 zfs_dbgmsg("spa_import: importing %s, max_txg=%lld "
5391 "(RECOVERY MODE)", pool, (longlong_t)policy.zrp_txg);
5393 error = spa_load_best(spa, state, policy.zrp_txg, policy.zrp_request);
5396 * Propagate anything learned while loading the pool and pass it
5397 * back to caller (i.e. rewind info, missing devices, etc).
5399 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5400 spa->spa_load_info) == 0);
5402 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5404 * Toss any existing sparelist, as it doesn't have any validity
5405 * anymore, and conflicts with spa_has_spare().
5407 if (spa->spa_spares.sav_config) {
5408 nvlist_free(spa->spa_spares.sav_config);
5409 spa->spa_spares.sav_config = NULL;
5410 spa_load_spares(spa);
5412 if (spa->spa_l2cache.sav_config) {
5413 nvlist_free(spa->spa_l2cache.sav_config);
5414 spa->spa_l2cache.sav_config = NULL;
5415 spa_load_l2cache(spa);
5418 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
5421 error = spa_validate_aux(spa, nvroot, -1ULL,
5424 error = spa_validate_aux(spa, nvroot, -1ULL,
5425 VDEV_ALLOC_L2CACHE);
5426 spa_config_exit(spa, SCL_ALL, FTAG);
5429 spa_configfile_set(spa, props, B_FALSE);
5431 if (error != 0 || (props && spa_writeable(spa) &&
5432 (error = spa_prop_set(spa, props)))) {
5434 spa_deactivate(spa);
5436 mutex_exit(&spa_namespace_lock);
5440 spa_async_resume(spa);
5443 * Override any spares and level 2 cache devices as specified by
5444 * the user, as these may have correct device names/devids, etc.
5446 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
5447 &spares, &nspares) == 0) {
5448 if (spa->spa_spares.sav_config)
5449 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
5450 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
5452 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
5453 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5454 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
5455 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
5456 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5457 spa_load_spares(spa);
5458 spa_config_exit(spa, SCL_ALL, FTAG);
5459 spa->spa_spares.sav_sync = B_TRUE;
5461 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
5462 &l2cache, &nl2cache) == 0) {
5463 if (spa->spa_l2cache.sav_config)
5464 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
5465 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
5467 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
5468 NV_UNIQUE_NAME, KM_SLEEP) == 0);
5469 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
5470 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
5471 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5472 spa_load_l2cache(spa);
5473 spa_config_exit(spa, SCL_ALL, FTAG);
5474 spa->spa_l2cache.sav_sync = B_TRUE;
5478 * Check for any removed devices.
5480 if (spa->spa_autoreplace) {
5481 spa_aux_check_removed(&spa->spa_spares);
5482 spa_aux_check_removed(&spa->spa_l2cache);
5485 if (spa_writeable(spa)) {
5487 * Update the config cache to include the newly-imported pool.
5489 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5493 * It's possible that the pool was expanded while it was exported.
5494 * We kick off an async task to handle this for us.
5496 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
5498 spa_history_log_version(spa, "import");
5500 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
5502 mutex_exit(&spa_namespace_lock);
5506 zvol_create_minors(pool);
5513 spa_tryimport(nvlist_t *tryconfig)
5515 nvlist_t *config = NULL;
5516 char *poolname, *cachefile;
5520 zpool_rewind_policy_t policy;
5522 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
5525 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
5529 * Create and initialize the spa structure.
5531 mutex_enter(&spa_namespace_lock);
5532 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
5533 spa_activate(spa, FREAD);
5536 * Rewind pool if a max txg was provided. Note that even though we
5537 * retrieve the complete rewind policy, only the rewind txg is relevant
5540 zpool_get_rewind_policy(spa->spa_config, &policy);
5541 if (policy.zrp_txg != UINT64_MAX) {
5542 spa->spa_load_max_txg = policy.zrp_txg;
5543 spa->spa_extreme_rewind = B_TRUE;
5544 zfs_dbgmsg("spa_tryimport: importing %s, max_txg=%lld",
5545 poolname, (longlong_t)policy.zrp_txg);
5547 zfs_dbgmsg("spa_tryimport: importing %s", poolname);
5550 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_CACHEFILE, &cachefile)
5552 zfs_dbgmsg("spa_tryimport: using cachefile '%s'", cachefile);
5553 spa->spa_config_source = SPA_CONFIG_SRC_CACHEFILE;
5555 spa->spa_config_source = SPA_CONFIG_SRC_SCAN;
5558 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING);
5561 * If 'tryconfig' was at least parsable, return the current config.
5563 if (spa->spa_root_vdev != NULL) {
5564 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
5565 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
5567 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
5569 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
5570 spa->spa_uberblock.ub_timestamp) == 0);
5571 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
5572 spa->spa_load_info) == 0);
5575 * If the bootfs property exists on this pool then we
5576 * copy it out so that external consumers can tell which
5577 * pools are bootable.
5579 if ((!error || error == EEXIST) && spa->spa_bootfs) {
5580 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5583 * We have to play games with the name since the
5584 * pool was opened as TRYIMPORT_NAME.
5586 if (dsl_dsobj_to_dsname(spa_name(spa),
5587 spa->spa_bootfs, tmpname) == 0) {
5589 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5591 cp = strchr(tmpname, '/');
5593 (void) strlcpy(dsname, tmpname,
5596 (void) snprintf(dsname, MAXPATHLEN,
5597 "%s/%s", poolname, ++cp);
5599 VERIFY(nvlist_add_string(config,
5600 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
5601 kmem_free(dsname, MAXPATHLEN);
5603 kmem_free(tmpname, MAXPATHLEN);
5607 * Add the list of hot spares and level 2 cache devices.
5609 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5610 spa_add_spares(spa, config);
5611 spa_add_l2cache(spa, config);
5612 spa_config_exit(spa, SCL_CONFIG, FTAG);
5616 spa_deactivate(spa);
5618 mutex_exit(&spa_namespace_lock);
5624 * Pool export/destroy
5626 * The act of destroying or exporting a pool is very simple. We make sure there
5627 * is no more pending I/O and any references to the pool are gone. Then, we
5628 * update the pool state and sync all the labels to disk, removing the
5629 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
5630 * we don't sync the labels or remove the configuration cache.
5633 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
5634 boolean_t force, boolean_t hardforce)
5641 if (!(spa_mode_global & FWRITE))
5642 return (SET_ERROR(EROFS));
5644 mutex_enter(&spa_namespace_lock);
5645 if ((spa = spa_lookup(pool)) == NULL) {
5646 mutex_exit(&spa_namespace_lock);
5647 return (SET_ERROR(ENOENT));
5651 * Put a hold on the pool, drop the namespace lock, stop async tasks,
5652 * reacquire the namespace lock, and see if we can export.
5654 spa_open_ref(spa, FTAG);
5655 mutex_exit(&spa_namespace_lock);
5656 spa_async_suspend(spa);
5657 mutex_enter(&spa_namespace_lock);
5658 spa_close(spa, FTAG);
5661 * The pool will be in core if it's openable,
5662 * in which case we can modify its state.
5664 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
5666 * Objsets may be open only because they're dirty, so we
5667 * have to force it to sync before checking spa_refcnt.
5669 txg_wait_synced(spa->spa_dsl_pool, 0);
5670 spa_evicting_os_wait(spa);
5673 * A pool cannot be exported or destroyed if there are active
5674 * references. If we are resetting a pool, allow references by
5675 * fault injection handlers.
5677 if (!spa_refcount_zero(spa) ||
5678 (spa->spa_inject_ref != 0 &&
5679 new_state != POOL_STATE_UNINITIALIZED)) {
5680 spa_async_resume(spa);
5681 mutex_exit(&spa_namespace_lock);
5682 return (SET_ERROR(EBUSY));
5686 * A pool cannot be exported if it has an active shared spare.
5687 * This is to prevent other pools stealing the active spare
5688 * from an exported pool. At user's own will, such pool can
5689 * be forcedly exported.
5691 if (!force && new_state == POOL_STATE_EXPORTED &&
5692 spa_has_active_shared_spare(spa)) {
5693 spa_async_resume(spa);
5694 mutex_exit(&spa_namespace_lock);
5695 return (SET_ERROR(EXDEV));
5699 * We want this to be reflected on every label,
5700 * so mark them all dirty. spa_unload() will do the
5701 * final sync that pushes these changes out.
5703 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
5704 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5705 spa->spa_state = new_state;
5706 spa->spa_final_txg = spa_last_synced_txg(spa) +
5708 vdev_config_dirty(spa->spa_root_vdev);
5709 spa_config_exit(spa, SCL_ALL, FTAG);
5713 spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
5715 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5717 spa_deactivate(spa);
5720 if (oldconfig && spa->spa_config)
5721 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
5723 if (new_state != POOL_STATE_UNINITIALIZED) {
5725 spa_write_cachefile(spa, B_TRUE, B_TRUE);
5728 mutex_exit(&spa_namespace_lock);
5734 * Destroy a storage pool.
5737 spa_destroy(char *pool)
5739 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
5744 * Export a storage pool.
5747 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
5748 boolean_t hardforce)
5750 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
5755 * Similar to spa_export(), this unloads the spa_t without actually removing it
5756 * from the namespace in any way.
5759 spa_reset(char *pool)
5761 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
5766 * ==========================================================================
5767 * Device manipulation
5768 * ==========================================================================
5772 * Add a device to a storage pool.
5775 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
5779 vdev_t *rvd = spa->spa_root_vdev;
5781 nvlist_t **spares, **l2cache;
5782 uint_t nspares, nl2cache;
5784 ASSERT(spa_writeable(spa));
5786 txg = spa_vdev_enter(spa);
5788 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
5789 VDEV_ALLOC_ADD)) != 0)
5790 return (spa_vdev_exit(spa, NULL, txg, error));
5792 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
5794 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
5798 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
5802 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
5803 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5805 if (vd->vdev_children != 0 &&
5806 (error = vdev_create(vd, txg, B_FALSE)) != 0)
5807 return (spa_vdev_exit(spa, vd, txg, error));
5810 * We must validate the spares and l2cache devices after checking the
5811 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
5813 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
5814 return (spa_vdev_exit(spa, vd, txg, error));
5817 * If we are in the middle of a device removal, we can only add
5818 * devices which match the existing devices in the pool.
5819 * If we are in the middle of a removal, or have some indirect
5820 * vdevs, we can not add raidz toplevels.
5822 if (spa->spa_vdev_removal != NULL ||
5823 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5824 for (int c = 0; c < vd->vdev_children; c++) {
5825 tvd = vd->vdev_child[c];
5826 if (spa->spa_vdev_removal != NULL &&
5828 spa->spa_vdev_removal->svr_vdev->vdev_ashift) {
5829 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5831 /* Fail if top level vdev is raidz */
5832 if (tvd->vdev_ops == &vdev_raidz_ops) {
5833 return (spa_vdev_exit(spa, vd, txg, EINVAL));
5836 * Need the top level mirror to be
5837 * a mirror of leaf vdevs only
5839 if (tvd->vdev_ops == &vdev_mirror_ops) {
5840 for (uint64_t cid = 0;
5841 cid < tvd->vdev_children; cid++) {
5842 vdev_t *cvd = tvd->vdev_child[cid];
5843 if (!cvd->vdev_ops->vdev_op_leaf) {
5844 return (spa_vdev_exit(spa, vd,
5852 for (int c = 0; c < vd->vdev_children; c++) {
5855 * Set the vdev id to the first hole, if one exists.
5857 for (id = 0; id < rvd->vdev_children; id++) {
5858 if (rvd->vdev_child[id]->vdev_ishole) {
5859 vdev_free(rvd->vdev_child[id]);
5863 tvd = vd->vdev_child[c];
5864 vdev_remove_child(vd, tvd);
5866 vdev_add_child(rvd, tvd);
5867 vdev_config_dirty(tvd);
5871 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
5872 ZPOOL_CONFIG_SPARES);
5873 spa_load_spares(spa);
5874 spa->spa_spares.sav_sync = B_TRUE;
5877 if (nl2cache != 0) {
5878 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
5879 ZPOOL_CONFIG_L2CACHE);
5880 spa_load_l2cache(spa);
5881 spa->spa_l2cache.sav_sync = B_TRUE;
5885 * We have to be careful when adding new vdevs to an existing pool.
5886 * If other threads start allocating from these vdevs before we
5887 * sync the config cache, and we lose power, then upon reboot we may
5888 * fail to open the pool because there are DVAs that the config cache
5889 * can't translate. Therefore, we first add the vdevs without
5890 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
5891 * and then let spa_config_update() initialize the new metaslabs.
5893 * spa_load() checks for added-but-not-initialized vdevs, so that
5894 * if we lose power at any point in this sequence, the remaining
5895 * steps will be completed the next time we load the pool.
5897 (void) spa_vdev_exit(spa, vd, txg, 0);
5899 mutex_enter(&spa_namespace_lock);
5900 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5901 spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
5902 mutex_exit(&spa_namespace_lock);
5908 * Attach a device to a mirror. The arguments are the path to any device
5909 * in the mirror, and the nvroot for the new device. If the path specifies
5910 * a device that is not mirrored, we automatically insert the mirror vdev.
5912 * If 'replacing' is specified, the new device is intended to replace the
5913 * existing device; in this case the two devices are made into their own
5914 * mirror using the 'replacing' vdev, which is functionally identical to
5915 * the mirror vdev (it actually reuses all the same ops) but has a few
5916 * extra rules: you can't attach to it after it's been created, and upon
5917 * completion of resilvering, the first disk (the one being replaced)
5918 * is automatically detached.
5921 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
5923 uint64_t txg, dtl_max_txg;
5924 vdev_t *rvd = spa->spa_root_vdev;
5925 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
5927 char *oldvdpath, *newvdpath;
5931 ASSERT(spa_writeable(spa));
5933 txg = spa_vdev_enter(spa);
5935 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
5937 ASSERT(MUTEX_HELD(&spa_namespace_lock));
5938 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
5939 error = (spa_has_checkpoint(spa)) ?
5940 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
5941 return (spa_vdev_exit(spa, NULL, txg, error));
5944 if (spa->spa_vdev_removal != NULL ||
5945 spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
5946 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
5950 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
5952 if (!oldvd->vdev_ops->vdev_op_leaf)
5953 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
5955 pvd = oldvd->vdev_parent;
5957 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
5958 VDEV_ALLOC_ATTACH)) != 0)
5959 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
5961 if (newrootvd->vdev_children != 1)
5962 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5964 newvd = newrootvd->vdev_child[0];
5966 if (!newvd->vdev_ops->vdev_op_leaf)
5967 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
5969 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
5970 return (spa_vdev_exit(spa, newrootvd, txg, error));
5973 * Spares can't replace logs
5975 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
5976 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5980 * For attach, the only allowable parent is a mirror or the root
5983 if (pvd->vdev_ops != &vdev_mirror_ops &&
5984 pvd->vdev_ops != &vdev_root_ops)
5985 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5987 pvops = &vdev_mirror_ops;
5990 * Active hot spares can only be replaced by inactive hot
5993 if (pvd->vdev_ops == &vdev_spare_ops &&
5994 oldvd->vdev_isspare &&
5995 !spa_has_spare(spa, newvd->vdev_guid))
5996 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
5999 * If the source is a hot spare, and the parent isn't already a
6000 * spare, then we want to create a new hot spare. Otherwise, we
6001 * want to create a replacing vdev. The user is not allowed to
6002 * attach to a spared vdev child unless the 'isspare' state is
6003 * the same (spare replaces spare, non-spare replaces
6006 if (pvd->vdev_ops == &vdev_replacing_ops &&
6007 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
6008 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6009 } else if (pvd->vdev_ops == &vdev_spare_ops &&
6010 newvd->vdev_isspare != oldvd->vdev_isspare) {
6011 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
6014 if (newvd->vdev_isspare)
6015 pvops = &vdev_spare_ops;
6017 pvops = &vdev_replacing_ops;
6021 * Make sure the new device is big enough.
6023 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
6024 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
6027 * The new device cannot have a higher alignment requirement
6028 * than the top-level vdev.
6030 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
6031 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
6034 * If this is an in-place replacement, update oldvd's path and devid
6035 * to make it distinguishable from newvd, and unopenable from now on.
6037 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
6038 spa_strfree(oldvd->vdev_path);
6039 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
6041 (void) sprintf(oldvd->vdev_path, "%s/%s",
6042 newvd->vdev_path, "old");
6043 if (oldvd->vdev_devid != NULL) {
6044 spa_strfree(oldvd->vdev_devid);
6045 oldvd->vdev_devid = NULL;
6049 /* mark the device being resilvered */
6050 newvd->vdev_resilver_txg = txg;
6053 * If the parent is not a mirror, or if we're replacing, insert the new
6054 * mirror/replacing/spare vdev above oldvd.
6056 if (pvd->vdev_ops != pvops)
6057 pvd = vdev_add_parent(oldvd, pvops);
6059 ASSERT(pvd->vdev_top->vdev_parent == rvd);
6060 ASSERT(pvd->vdev_ops == pvops);
6061 ASSERT(oldvd->vdev_parent == pvd);
6064 * Extract the new device from its root and add it to pvd.
6066 vdev_remove_child(newrootvd, newvd);
6067 newvd->vdev_id = pvd->vdev_children;
6068 newvd->vdev_crtxg = oldvd->vdev_crtxg;
6069 vdev_add_child(pvd, newvd);
6071 tvd = newvd->vdev_top;
6072 ASSERT(pvd->vdev_top == tvd);
6073 ASSERT(tvd->vdev_parent == rvd);
6075 vdev_config_dirty(tvd);
6078 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
6079 * for any dmu_sync-ed blocks. It will propagate upward when
6080 * spa_vdev_exit() calls vdev_dtl_reassess().
6082 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
6084 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
6085 dtl_max_txg - TXG_INITIAL);
6087 if (newvd->vdev_isspare) {
6088 spa_spare_activate(newvd);
6089 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
6092 oldvdpath = spa_strdup(oldvd->vdev_path);
6093 newvdpath = spa_strdup(newvd->vdev_path);
6094 newvd_isspare = newvd->vdev_isspare;
6097 * Mark newvd's DTL dirty in this txg.
6099 vdev_dirty(tvd, VDD_DTL, newvd, txg);
6102 * Schedule the resilver to restart in the future. We do this to
6103 * ensure that dmu_sync-ed blocks have been stitched into the
6104 * respective datasets.
6106 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
6108 if (spa->spa_bootfs)
6109 spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
6111 spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
6116 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
6118 spa_history_log_internal(spa, "vdev attach", NULL,
6119 "%s vdev=%s %s vdev=%s",
6120 replacing && newvd_isspare ? "spare in" :
6121 replacing ? "replace" : "attach", newvdpath,
6122 replacing ? "for" : "to", oldvdpath);
6124 spa_strfree(oldvdpath);
6125 spa_strfree(newvdpath);
6131 * Detach a device from a mirror or replacing vdev.
6133 * If 'replace_done' is specified, only detach if the parent
6134 * is a replacing vdev.
6137 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
6141 vdev_t *rvd = spa->spa_root_vdev;
6142 vdev_t *vd, *pvd, *cvd, *tvd;
6143 boolean_t unspare = B_FALSE;
6144 uint64_t unspare_guid = 0;
6147 ASSERT(spa_writeable(spa));
6149 txg = spa_vdev_enter(spa);
6151 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
6154 * Besides being called directly from the userland through the
6155 * ioctl interface, spa_vdev_detach() can be potentially called
6156 * at the end of spa_vdev_resilver_done().
6158 * In the regular case, when we have a checkpoint this shouldn't
6159 * happen as we never empty the DTLs of a vdev during the scrub
6160 * [see comment in dsl_scan_done()]. Thus spa_vdev_resilvering_done()
6161 * should never get here when we have a checkpoint.
6163 * That said, even in a case when we checkpoint the pool exactly
6164 * as spa_vdev_resilver_done() calls this function everything
6165 * should be fine as the resilver will return right away.
6167 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6168 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6169 error = (spa_has_checkpoint(spa)) ?
6170 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6171 return (spa_vdev_exit(spa, NULL, txg, error));
6175 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
6177 if (!vd->vdev_ops->vdev_op_leaf)
6178 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6180 pvd = vd->vdev_parent;
6183 * If the parent/child relationship is not as expected, don't do it.
6184 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
6185 * vdev that's replacing B with C. The user's intent in replacing
6186 * is to go from M(A,B) to M(A,C). If the user decides to cancel
6187 * the replace by detaching C, the expected behavior is to end up
6188 * M(A,B). But suppose that right after deciding to detach C,
6189 * the replacement of B completes. We would have M(A,C), and then
6190 * ask to detach C, which would leave us with just A -- not what
6191 * the user wanted. To prevent this, we make sure that the
6192 * parent/child relationship hasn't changed -- in this example,
6193 * that C's parent is still the replacing vdev R.
6195 if (pvd->vdev_guid != pguid && pguid != 0)
6196 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6199 * Only 'replacing' or 'spare' vdevs can be replaced.
6201 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
6202 pvd->vdev_ops != &vdev_spare_ops)
6203 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6205 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
6206 spa_version(spa) >= SPA_VERSION_SPARES);
6209 * Only mirror, replacing, and spare vdevs support detach.
6211 if (pvd->vdev_ops != &vdev_replacing_ops &&
6212 pvd->vdev_ops != &vdev_mirror_ops &&
6213 pvd->vdev_ops != &vdev_spare_ops)
6214 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
6217 * If this device has the only valid copy of some data,
6218 * we cannot safely detach it.
6220 if (vdev_dtl_required(vd))
6221 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
6223 ASSERT(pvd->vdev_children >= 2);
6226 * If we are detaching the second disk from a replacing vdev, then
6227 * check to see if we changed the original vdev's path to have "/old"
6228 * at the end in spa_vdev_attach(). If so, undo that change now.
6230 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
6231 vd->vdev_path != NULL) {
6232 size_t len = strlen(vd->vdev_path);
6234 for (int c = 0; c < pvd->vdev_children; c++) {
6235 cvd = pvd->vdev_child[c];
6237 if (cvd == vd || cvd->vdev_path == NULL)
6240 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
6241 strcmp(cvd->vdev_path + len, "/old") == 0) {
6242 spa_strfree(cvd->vdev_path);
6243 cvd->vdev_path = spa_strdup(vd->vdev_path);
6250 * If we are detaching the original disk from a spare, then it implies
6251 * that the spare should become a real disk, and be removed from the
6252 * active spare list for the pool.
6254 if (pvd->vdev_ops == &vdev_spare_ops &&
6256 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
6260 * Erase the disk labels so the disk can be used for other things.
6261 * This must be done after all other error cases are handled,
6262 * but before we disembowel vd (so we can still do I/O to it).
6263 * But if we can't do it, don't treat the error as fatal --
6264 * it may be that the unwritability of the disk is the reason
6265 * it's being detached!
6267 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
6270 * Remove vd from its parent and compact the parent's children.
6272 vdev_remove_child(pvd, vd);
6273 vdev_compact_children(pvd);
6276 * Remember one of the remaining children so we can get tvd below.
6278 cvd = pvd->vdev_child[pvd->vdev_children - 1];
6281 * If we need to remove the remaining child from the list of hot spares,
6282 * do it now, marking the vdev as no longer a spare in the process.
6283 * We must do this before vdev_remove_parent(), because that can
6284 * change the GUID if it creates a new toplevel GUID. For a similar
6285 * reason, we must remove the spare now, in the same txg as the detach;
6286 * otherwise someone could attach a new sibling, change the GUID, and
6287 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
6290 ASSERT(cvd->vdev_isspare);
6291 spa_spare_remove(cvd);
6292 unspare_guid = cvd->vdev_guid;
6293 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
6294 cvd->vdev_unspare = B_TRUE;
6298 * If the parent mirror/replacing vdev only has one child,
6299 * the parent is no longer needed. Remove it from the tree.
6301 if (pvd->vdev_children == 1) {
6302 if (pvd->vdev_ops == &vdev_spare_ops)
6303 cvd->vdev_unspare = B_FALSE;
6304 vdev_remove_parent(cvd);
6309 * We don't set tvd until now because the parent we just removed
6310 * may have been the previous top-level vdev.
6312 tvd = cvd->vdev_top;
6313 ASSERT(tvd->vdev_parent == rvd);
6316 * Reevaluate the parent vdev state.
6318 vdev_propagate_state(cvd);
6321 * If the 'autoexpand' property is set on the pool then automatically
6322 * try to expand the size of the pool. For example if the device we
6323 * just detached was smaller than the others, it may be possible to
6324 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
6325 * first so that we can obtain the updated sizes of the leaf vdevs.
6327 if (spa->spa_autoexpand) {
6329 vdev_expand(tvd, txg);
6332 vdev_config_dirty(tvd);
6335 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
6336 * vd->vdev_detached is set and free vd's DTL object in syncing context.
6337 * But first make sure we're not on any *other* txg's DTL list, to
6338 * prevent vd from being accessed after it's freed.
6340 vdpath = spa_strdup(vd->vdev_path);
6341 for (int t = 0; t < TXG_SIZE; t++)
6342 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
6343 vd->vdev_detached = B_TRUE;
6344 vdev_dirty(tvd, VDD_DTL, vd, txg);
6346 spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
6348 /* hang on to the spa before we release the lock */
6349 spa_open_ref(spa, FTAG);
6351 error = spa_vdev_exit(spa, vd, txg, 0);
6353 spa_history_log_internal(spa, "detach", NULL,
6355 spa_strfree(vdpath);
6358 * If this was the removal of the original device in a hot spare vdev,
6359 * then we want to go through and remove the device from the hot spare
6360 * list of every other pool.
6363 spa_t *altspa = NULL;
6365 mutex_enter(&spa_namespace_lock);
6366 while ((altspa = spa_next(altspa)) != NULL) {
6367 if (altspa->spa_state != POOL_STATE_ACTIVE ||
6371 spa_open_ref(altspa, FTAG);
6372 mutex_exit(&spa_namespace_lock);
6373 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
6374 mutex_enter(&spa_namespace_lock);
6375 spa_close(altspa, FTAG);
6377 mutex_exit(&spa_namespace_lock);
6379 /* search the rest of the vdevs for spares to remove */
6380 spa_vdev_resilver_done(spa);
6383 /* all done with the spa; OK to release */
6384 mutex_enter(&spa_namespace_lock);
6385 spa_close(spa, FTAG);
6386 mutex_exit(&spa_namespace_lock);
6392 * Split a set of devices from their mirrors, and create a new pool from them.
6395 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
6396 nvlist_t *props, boolean_t exp)
6399 uint64_t txg, *glist;
6401 uint_t c, children, lastlog;
6402 nvlist_t **child, *nvl, *tmp;
6404 char *altroot = NULL;
6405 vdev_t *rvd, **vml = NULL; /* vdev modify list */
6406 boolean_t activate_slog;
6408 ASSERT(spa_writeable(spa));
6410 txg = spa_vdev_enter(spa);
6412 ASSERT(MUTEX_HELD(&spa_namespace_lock));
6413 if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
6414 error = (spa_has_checkpoint(spa)) ?
6415 ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
6416 return (spa_vdev_exit(spa, NULL, txg, error));
6419 /* clear the log and flush everything up to now */
6420 activate_slog = spa_passivate_log(spa);
6421 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6422 error = spa_reset_logs(spa);
6423 txg = spa_vdev_config_enter(spa);
6426 spa_activate_log(spa);
6429 return (spa_vdev_exit(spa, NULL, txg, error));
6431 /* check new spa name before going any further */
6432 if (spa_lookup(newname) != NULL)
6433 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
6436 * scan through all the children to ensure they're all mirrors
6438 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
6439 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
6441 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6443 /* first, check to ensure we've got the right child count */
6444 rvd = spa->spa_root_vdev;
6446 for (c = 0; c < rvd->vdev_children; c++) {
6447 vdev_t *vd = rvd->vdev_child[c];
6449 /* don't count the holes & logs as children */
6450 if (vd->vdev_islog || !vdev_is_concrete(vd)) {
6458 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
6459 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6461 /* next, ensure no spare or cache devices are part of the split */
6462 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
6463 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
6464 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
6466 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
6467 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
6469 /* then, loop over each vdev and validate it */
6470 for (c = 0; c < children; c++) {
6471 uint64_t is_hole = 0;
6473 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
6477 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
6478 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
6481 error = SET_ERROR(EINVAL);
6486 /* which disk is going to be split? */
6487 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
6489 error = SET_ERROR(EINVAL);
6493 /* look it up in the spa */
6494 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
6495 if (vml[c] == NULL) {
6496 error = SET_ERROR(ENODEV);
6500 /* make sure there's nothing stopping the split */
6501 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
6502 vml[c]->vdev_islog ||
6503 !vdev_is_concrete(vml[c]) ||
6504 vml[c]->vdev_isspare ||
6505 vml[c]->vdev_isl2cache ||
6506 !vdev_writeable(vml[c]) ||
6507 vml[c]->vdev_children != 0 ||
6508 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
6509 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
6510 error = SET_ERROR(EINVAL);
6514 if (vdev_dtl_required(vml[c])) {
6515 error = SET_ERROR(EBUSY);
6519 /* we need certain info from the top level */
6520 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
6521 vml[c]->vdev_top->vdev_ms_array) == 0);
6522 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
6523 vml[c]->vdev_top->vdev_ms_shift) == 0);
6524 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
6525 vml[c]->vdev_top->vdev_asize) == 0);
6526 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
6527 vml[c]->vdev_top->vdev_ashift) == 0);
6529 /* transfer per-vdev ZAPs */
6530 ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
6531 VERIFY0(nvlist_add_uint64(child[c],
6532 ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
6534 ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
6535 VERIFY0(nvlist_add_uint64(child[c],
6536 ZPOOL_CONFIG_VDEV_TOP_ZAP,
6537 vml[c]->vdev_parent->vdev_top_zap));
6541 kmem_free(vml, children * sizeof (vdev_t *));
6542 kmem_free(glist, children * sizeof (uint64_t));
6543 return (spa_vdev_exit(spa, NULL, txg, error));
6546 /* stop writers from using the disks */
6547 for (c = 0; c < children; c++) {
6549 vml[c]->vdev_offline = B_TRUE;
6551 vdev_reopen(spa->spa_root_vdev);
6554 * Temporarily record the splitting vdevs in the spa config. This
6555 * will disappear once the config is regenerated.
6557 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6558 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
6559 glist, children) == 0);
6560 kmem_free(glist, children * sizeof (uint64_t));
6562 mutex_enter(&spa->spa_props_lock);
6563 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
6565 mutex_exit(&spa->spa_props_lock);
6566 spa->spa_config_splitting = nvl;
6567 vdev_config_dirty(spa->spa_root_vdev);
6569 /* configure and create the new pool */
6570 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
6571 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
6572 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
6573 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
6574 spa_version(spa)) == 0);
6575 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
6576 spa->spa_config_txg) == 0);
6577 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
6578 spa_generate_guid(NULL)) == 0);
6579 VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
6580 (void) nvlist_lookup_string(props,
6581 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
6583 /* add the new pool to the namespace */
6584 newspa = spa_add(newname, config, altroot);
6585 newspa->spa_avz_action = AVZ_ACTION_REBUILD;
6586 newspa->spa_config_txg = spa->spa_config_txg;
6587 spa_set_log_state(newspa, SPA_LOG_CLEAR);
6589 /* release the spa config lock, retaining the namespace lock */
6590 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
6592 if (zio_injection_enabled)
6593 zio_handle_panic_injection(spa, FTAG, 1);
6595 spa_activate(newspa, spa_mode_global);
6596 spa_async_suspend(newspa);
6599 /* mark that we are creating new spa by splitting */
6600 newspa->spa_splitting_newspa = B_TRUE;
6602 newspa->spa_config_source = SPA_CONFIG_SRC_SPLIT;
6604 /* create the new pool from the disks of the original pool */
6605 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE);
6607 newspa->spa_splitting_newspa = B_FALSE;
6612 /* if that worked, generate a real config for the new pool */
6613 if (newspa->spa_root_vdev != NULL) {
6614 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
6615 NV_UNIQUE_NAME, KM_SLEEP) == 0);
6616 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
6617 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
6618 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
6623 if (props != NULL) {
6624 spa_configfile_set(newspa, props, B_FALSE);
6625 error = spa_prop_set(newspa, props);
6630 /* flush everything */
6631 txg = spa_vdev_config_enter(newspa);
6632 vdev_config_dirty(newspa->spa_root_vdev);
6633 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
6635 if (zio_injection_enabled)
6636 zio_handle_panic_injection(spa, FTAG, 2);
6638 spa_async_resume(newspa);
6640 /* finally, update the original pool's config */
6641 txg = spa_vdev_config_enter(spa);
6642 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
6643 error = dmu_tx_assign(tx, TXG_WAIT);
6646 for (c = 0; c < children; c++) {
6647 if (vml[c] != NULL) {
6650 spa_history_log_internal(spa, "detach", tx,
6651 "vdev=%s", vml[c]->vdev_path);
6656 spa->spa_avz_action = AVZ_ACTION_REBUILD;
6657 vdev_config_dirty(spa->spa_root_vdev);
6658 spa->spa_config_splitting = NULL;
6662 (void) spa_vdev_exit(spa, NULL, txg, 0);
6664 if (zio_injection_enabled)
6665 zio_handle_panic_injection(spa, FTAG, 3);
6667 /* split is complete; log a history record */
6668 spa_history_log_internal(newspa, "split", NULL,
6669 "from pool %s", spa_name(spa));
6671 kmem_free(vml, children * sizeof (vdev_t *));
6673 /* if we're not going to mount the filesystems in userland, export */
6675 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
6682 spa_deactivate(newspa);
6685 txg = spa_vdev_config_enter(spa);
6687 /* re-online all offlined disks */
6688 for (c = 0; c < children; c++) {
6690 vml[c]->vdev_offline = B_FALSE;
6692 vdev_reopen(spa->spa_root_vdev);
6694 nvlist_free(spa->spa_config_splitting);
6695 spa->spa_config_splitting = NULL;
6696 (void) spa_vdev_exit(spa, NULL, txg, error);
6698 kmem_free(vml, children * sizeof (vdev_t *));
6703 * Find any device that's done replacing, or a vdev marked 'unspare' that's
6704 * currently spared, so we can detach it.
6707 spa_vdev_resilver_done_hunt(vdev_t *vd)
6709 vdev_t *newvd, *oldvd;
6711 for (int c = 0; c < vd->vdev_children; c++) {
6712 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
6718 * Check for a completed replacement. We always consider the first
6719 * vdev in the list to be the oldest vdev, and the last one to be
6720 * the newest (see spa_vdev_attach() for how that works). In
6721 * the case where the newest vdev is faulted, we will not automatically
6722 * remove it after a resilver completes. This is OK as it will require
6723 * user intervention to determine which disk the admin wishes to keep.
6725 if (vd->vdev_ops == &vdev_replacing_ops) {
6726 ASSERT(vd->vdev_children > 1);
6728 newvd = vd->vdev_child[vd->vdev_children - 1];
6729 oldvd = vd->vdev_child[0];
6731 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
6732 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6733 !vdev_dtl_required(oldvd))
6738 * Check for a completed resilver with the 'unspare' flag set.
6740 if (vd->vdev_ops == &vdev_spare_ops) {
6741 vdev_t *first = vd->vdev_child[0];
6742 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
6744 if (last->vdev_unspare) {
6747 } else if (first->vdev_unspare) {
6754 if (oldvd != NULL &&
6755 vdev_dtl_empty(newvd, DTL_MISSING) &&
6756 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
6757 !vdev_dtl_required(oldvd))
6761 * If there are more than two spares attached to a disk,
6762 * and those spares are not required, then we want to
6763 * attempt to free them up now so that they can be used
6764 * by other pools. Once we're back down to a single
6765 * disk+spare, we stop removing them.
6767 if (vd->vdev_children > 2) {
6768 newvd = vd->vdev_child[1];
6770 if (newvd->vdev_isspare && last->vdev_isspare &&
6771 vdev_dtl_empty(last, DTL_MISSING) &&
6772 vdev_dtl_empty(last, DTL_OUTAGE) &&
6773 !vdev_dtl_required(newvd))
6782 spa_vdev_resilver_done(spa_t *spa)
6784 vdev_t *vd, *pvd, *ppvd;
6785 uint64_t guid, sguid, pguid, ppguid;
6787 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6789 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
6790 pvd = vd->vdev_parent;
6791 ppvd = pvd->vdev_parent;
6792 guid = vd->vdev_guid;
6793 pguid = pvd->vdev_guid;
6794 ppguid = ppvd->vdev_guid;
6797 * If we have just finished replacing a hot spared device, then
6798 * we need to detach the parent's first child (the original hot
6801 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
6802 ppvd->vdev_children == 2) {
6803 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
6804 sguid = ppvd->vdev_child[1]->vdev_guid;
6806 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
6808 spa_config_exit(spa, SCL_ALL, FTAG);
6809 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
6811 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
6813 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6816 spa_config_exit(spa, SCL_ALL, FTAG);
6820 * Update the stored path or FRU for this vdev.
6823 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
6827 boolean_t sync = B_FALSE;
6829 ASSERT(spa_writeable(spa));
6831 spa_vdev_state_enter(spa, SCL_ALL);
6833 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
6834 return (spa_vdev_state_exit(spa, NULL, ENOENT));
6836 if (!vd->vdev_ops->vdev_op_leaf)
6837 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
6840 if (strcmp(value, vd->vdev_path) != 0) {
6841 spa_strfree(vd->vdev_path);
6842 vd->vdev_path = spa_strdup(value);
6846 if (vd->vdev_fru == NULL) {
6847 vd->vdev_fru = spa_strdup(value);
6849 } else if (strcmp(value, vd->vdev_fru) != 0) {
6850 spa_strfree(vd->vdev_fru);
6851 vd->vdev_fru = spa_strdup(value);
6856 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
6860 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
6862 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
6866 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
6868 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
6872 * ==========================================================================
6874 * ==========================================================================
6877 spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
6879 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6881 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6882 return (SET_ERROR(EBUSY));
6884 return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
6888 spa_scan_stop(spa_t *spa)
6890 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6891 if (dsl_scan_resilvering(spa->spa_dsl_pool))
6892 return (SET_ERROR(EBUSY));
6893 return (dsl_scan_cancel(spa->spa_dsl_pool));
6897 spa_scan(spa_t *spa, pool_scan_func_t func)
6899 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
6901 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
6902 return (SET_ERROR(ENOTSUP));
6905 * If a resilver was requested, but there is no DTL on a
6906 * writeable leaf device, we have nothing to do.
6908 if (func == POOL_SCAN_RESILVER &&
6909 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
6910 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
6914 return (dsl_scan(spa->spa_dsl_pool, func));
6918 * ==========================================================================
6919 * SPA async task processing
6920 * ==========================================================================
6924 spa_async_remove(spa_t *spa, vdev_t *vd)
6926 if (vd->vdev_remove_wanted) {
6927 vd->vdev_remove_wanted = B_FALSE;
6928 vd->vdev_delayed_close = B_FALSE;
6929 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
6932 * We want to clear the stats, but we don't want to do a full
6933 * vdev_clear() as that will cause us to throw away
6934 * degraded/faulted state as well as attempt to reopen the
6935 * device, all of which is a waste.
6937 vd->vdev_stat.vs_read_errors = 0;
6938 vd->vdev_stat.vs_write_errors = 0;
6939 vd->vdev_stat.vs_checksum_errors = 0;
6941 vdev_state_dirty(vd->vdev_top);
6942 /* Tell userspace that the vdev is gone. */
6943 zfs_post_remove(spa, vd);
6946 for (int c = 0; c < vd->vdev_children; c++)
6947 spa_async_remove(spa, vd->vdev_child[c]);
6951 spa_async_probe(spa_t *spa, vdev_t *vd)
6953 if (vd->vdev_probe_wanted) {
6954 vd->vdev_probe_wanted = B_FALSE;
6955 vdev_reopen(vd); /* vdev_open() does the actual probe */
6958 for (int c = 0; c < vd->vdev_children; c++)
6959 spa_async_probe(spa, vd->vdev_child[c]);
6963 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
6969 if (!spa->spa_autoexpand)
6972 for (int c = 0; c < vd->vdev_children; c++) {
6973 vdev_t *cvd = vd->vdev_child[c];
6974 spa_async_autoexpand(spa, cvd);
6977 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
6980 physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
6981 (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
6983 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
6984 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
6986 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
6987 ESC_ZFS_VDEV_AUTOEXPAND, attr, &eid, DDI_SLEEP);
6990 kmem_free(physpath, MAXPATHLEN);
6994 spa_async_thread(void *arg)
6996 spa_t *spa = (spa_t *)arg;
6999 ASSERT(spa->spa_sync_on);
7001 mutex_enter(&spa->spa_async_lock);
7002 tasks = spa->spa_async_tasks;
7003 spa->spa_async_tasks &= SPA_ASYNC_REMOVE;
7004 mutex_exit(&spa->spa_async_lock);
7007 * See if the config needs to be updated.
7009 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
7010 uint64_t old_space, new_space;
7012 mutex_enter(&spa_namespace_lock);
7013 old_space = metaslab_class_get_space(spa_normal_class(spa));
7014 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
7015 new_space = metaslab_class_get_space(spa_normal_class(spa));
7016 mutex_exit(&spa_namespace_lock);
7019 * If the pool grew as a result of the config update,
7020 * then log an internal history event.
7022 if (new_space != old_space) {
7023 spa_history_log_internal(spa, "vdev online", NULL,
7024 "pool '%s' size: %llu(+%llu)",
7025 spa_name(spa), new_space, new_space - old_space);
7029 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
7030 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7031 spa_async_autoexpand(spa, spa->spa_root_vdev);
7032 spa_config_exit(spa, SCL_CONFIG, FTAG);
7036 * See if any devices need to be probed.
7038 if (tasks & SPA_ASYNC_PROBE) {
7039 spa_vdev_state_enter(spa, SCL_NONE);
7040 spa_async_probe(spa, spa->spa_root_vdev);
7041 (void) spa_vdev_state_exit(spa, NULL, 0);
7045 * If any devices are done replacing, detach them.
7047 if (tasks & SPA_ASYNC_RESILVER_DONE)
7048 spa_vdev_resilver_done(spa);
7051 * Kick off a resilver.
7053 if (tasks & SPA_ASYNC_RESILVER)
7054 dsl_resilver_restart(spa->spa_dsl_pool, 0);
7057 * Let the world know that we're done.
7059 mutex_enter(&spa->spa_async_lock);
7060 spa->spa_async_thread = NULL;
7061 cv_broadcast(&spa->spa_async_cv);
7062 mutex_exit(&spa->spa_async_lock);
7067 spa_async_thread_vd(void *arg)
7072 mutex_enter(&spa->spa_async_lock);
7073 tasks = spa->spa_async_tasks;
7075 spa->spa_async_tasks &= ~SPA_ASYNC_REMOVE;
7076 mutex_exit(&spa->spa_async_lock);
7079 * See if any devices need to be marked REMOVED.
7081 if (tasks & SPA_ASYNC_REMOVE) {
7082 spa_vdev_state_enter(spa, SCL_NONE);
7083 spa_async_remove(spa, spa->spa_root_vdev);
7084 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
7085 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
7086 for (int i = 0; i < spa->spa_spares.sav_count; i++)
7087 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
7088 (void) spa_vdev_state_exit(spa, NULL, 0);
7092 * Let the world know that we're done.
7094 mutex_enter(&spa->spa_async_lock);
7095 tasks = spa->spa_async_tasks;
7096 if ((tasks & SPA_ASYNC_REMOVE) != 0)
7098 spa->spa_async_thread_vd = NULL;
7099 cv_broadcast(&spa->spa_async_cv);
7100 mutex_exit(&spa->spa_async_lock);
7105 spa_async_suspend(spa_t *spa)
7107 mutex_enter(&spa->spa_async_lock);
7108 spa->spa_async_suspended++;
7109 while (spa->spa_async_thread != NULL ||
7110 spa->spa_async_thread_vd != NULL)
7111 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
7112 mutex_exit(&spa->spa_async_lock);
7114 spa_vdev_remove_suspend(spa);
7116 zthr_t *condense_thread = spa->spa_condense_zthr;
7117 if (condense_thread != NULL && zthr_isrunning(condense_thread))
7118 VERIFY0(zthr_cancel(condense_thread));
7120 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7121 if (discard_thread != NULL && zthr_isrunning(discard_thread))
7122 VERIFY0(zthr_cancel(discard_thread));
7126 spa_async_resume(spa_t *spa)
7128 mutex_enter(&spa->spa_async_lock);
7129 ASSERT(spa->spa_async_suspended != 0);
7130 spa->spa_async_suspended--;
7131 mutex_exit(&spa->spa_async_lock);
7132 spa_restart_removal(spa);
7134 zthr_t *condense_thread = spa->spa_condense_zthr;
7135 if (condense_thread != NULL && !zthr_isrunning(condense_thread))
7136 zthr_resume(condense_thread);
7138 zthr_t *discard_thread = spa->spa_checkpoint_discard_zthr;
7139 if (discard_thread != NULL && !zthr_isrunning(discard_thread))
7140 zthr_resume(discard_thread);
7144 spa_async_tasks_pending(spa_t *spa)
7146 uint_t non_config_tasks;
7148 boolean_t config_task_suspended;
7150 non_config_tasks = spa->spa_async_tasks & ~(SPA_ASYNC_CONFIG_UPDATE |
7152 config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
7153 if (spa->spa_ccw_fail_time == 0) {
7154 config_task_suspended = B_FALSE;
7156 config_task_suspended =
7157 (gethrtime() - spa->spa_ccw_fail_time) <
7158 (zfs_ccw_retry_interval * NANOSEC);
7161 return (non_config_tasks || (config_task && !config_task_suspended));
7165 spa_async_dispatch(spa_t *spa)
7167 mutex_enter(&spa->spa_async_lock);
7168 if (spa_async_tasks_pending(spa) &&
7169 !spa->spa_async_suspended &&
7170 spa->spa_async_thread == NULL &&
7172 spa->spa_async_thread = thread_create(NULL, 0,
7173 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
7174 mutex_exit(&spa->spa_async_lock);
7178 spa_async_dispatch_vd(spa_t *spa)
7180 mutex_enter(&spa->spa_async_lock);
7181 if ((spa->spa_async_tasks & SPA_ASYNC_REMOVE) != 0 &&
7182 !spa->spa_async_suspended &&
7183 spa->spa_async_thread_vd == NULL &&
7185 spa->spa_async_thread_vd = thread_create(NULL, 0,
7186 spa_async_thread_vd, spa, 0, &p0, TS_RUN, maxclsyspri);
7187 mutex_exit(&spa->spa_async_lock);
7191 spa_async_request(spa_t *spa, int task)
7193 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
7194 mutex_enter(&spa->spa_async_lock);
7195 spa->spa_async_tasks |= task;
7196 mutex_exit(&spa->spa_async_lock);
7197 spa_async_dispatch_vd(spa);
7201 * ==========================================================================
7202 * SPA syncing routines
7203 * ==========================================================================
7207 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7210 bpobj_enqueue(bpo, bp, tx);
7215 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
7219 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
7220 BP_GET_PSIZE(bp), zio->io_flags));
7225 * Note: this simple function is not inlined to make it easier to dtrace the
7226 * amount of time spent syncing frees.
7229 spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
7231 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7232 bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
7233 VERIFY(zio_wait(zio) == 0);
7237 * Note: this simple function is not inlined to make it easier to dtrace the
7238 * amount of time spent syncing deferred frees.
7241 spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
7243 zio_t *zio = zio_root(spa, NULL, NULL, 0);
7244 VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
7245 spa_free_sync_cb, zio, tx), ==, 0);
7246 VERIFY0(zio_wait(zio));
7251 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
7253 char *packed = NULL;
7258 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
7261 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
7262 * information. This avoids the dmu_buf_will_dirty() path and
7263 * saves us a pre-read to get data we don't actually care about.
7265 bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
7266 packed = kmem_alloc(bufsize, KM_SLEEP);
7268 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
7270 bzero(packed + nvsize, bufsize - nvsize);
7272 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
7274 kmem_free(packed, bufsize);
7276 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
7277 dmu_buf_will_dirty(db, tx);
7278 *(uint64_t *)db->db_data = nvsize;
7279 dmu_buf_rele(db, FTAG);
7283 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
7284 const char *config, const char *entry)
7294 * Update the MOS nvlist describing the list of available devices.
7295 * spa_validate_aux() will have already made sure this nvlist is
7296 * valid and the vdevs are labeled appropriately.
7298 if (sav->sav_object == 0) {
7299 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
7300 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
7301 sizeof (uint64_t), tx);
7302 VERIFY(zap_update(spa->spa_meta_objset,
7303 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
7304 &sav->sav_object, tx) == 0);
7307 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
7308 if (sav->sav_count == 0) {
7309 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
7311 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
7312 for (i = 0; i < sav->sav_count; i++)
7313 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
7314 B_FALSE, VDEV_CONFIG_L2CACHE);
7315 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
7316 sav->sav_count) == 0);
7317 for (i = 0; i < sav->sav_count; i++)
7318 nvlist_free(list[i]);
7319 kmem_free(list, sav->sav_count * sizeof (void *));
7322 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
7323 nvlist_free(nvroot);
7325 sav->sav_sync = B_FALSE;
7329 * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
7330 * The all-vdev ZAP must be empty.
7333 spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
7335 spa_t *spa = vd->vdev_spa;
7336 if (vd->vdev_top_zap != 0) {
7337 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7338 vd->vdev_top_zap, tx));
7340 if (vd->vdev_leaf_zap != 0) {
7341 VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
7342 vd->vdev_leaf_zap, tx));
7344 for (uint64_t i = 0; i < vd->vdev_children; i++) {
7345 spa_avz_build(vd->vdev_child[i], avz, tx);
7350 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
7355 * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
7356 * its config may not be dirty but we still need to build per-vdev ZAPs.
7357 * Similarly, if the pool is being assembled (e.g. after a split), we
7358 * need to rebuild the AVZ although the config may not be dirty.
7360 if (list_is_empty(&spa->spa_config_dirty_list) &&
7361 spa->spa_avz_action == AVZ_ACTION_NONE)
7364 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7366 ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
7367 spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
7368 spa->spa_all_vdev_zaps != 0);
7370 if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
7371 /* Make and build the new AVZ */
7372 uint64_t new_avz = zap_create(spa->spa_meta_objset,
7373 DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
7374 spa_avz_build(spa->spa_root_vdev, new_avz, tx);
7376 /* Diff old AVZ with new one */
7380 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7381 spa->spa_all_vdev_zaps);
7382 zap_cursor_retrieve(&zc, &za) == 0;
7383 zap_cursor_advance(&zc)) {
7384 uint64_t vdzap = za.za_first_integer;
7385 if (zap_lookup_int(spa->spa_meta_objset, new_avz,
7388 * ZAP is listed in old AVZ but not in new one;
7391 VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
7396 zap_cursor_fini(&zc);
7398 /* Destroy the old AVZ */
7399 VERIFY0(zap_destroy(spa->spa_meta_objset,
7400 spa->spa_all_vdev_zaps, tx));
7402 /* Replace the old AVZ in the dir obj with the new one */
7403 VERIFY0(zap_update(spa->spa_meta_objset,
7404 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
7405 sizeof (new_avz), 1, &new_avz, tx));
7407 spa->spa_all_vdev_zaps = new_avz;
7408 } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
7412 /* Walk through the AVZ and destroy all listed ZAPs */
7413 for (zap_cursor_init(&zc, spa->spa_meta_objset,
7414 spa->spa_all_vdev_zaps);
7415 zap_cursor_retrieve(&zc, &za) == 0;
7416 zap_cursor_advance(&zc)) {
7417 uint64_t zap = za.za_first_integer;
7418 VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
7421 zap_cursor_fini(&zc);
7423 /* Destroy and unlink the AVZ itself */
7424 VERIFY0(zap_destroy(spa->spa_meta_objset,
7425 spa->spa_all_vdev_zaps, tx));
7426 VERIFY0(zap_remove(spa->spa_meta_objset,
7427 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
7428 spa->spa_all_vdev_zaps = 0;
7431 if (spa->spa_all_vdev_zaps == 0) {
7432 spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
7433 DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
7434 DMU_POOL_VDEV_ZAP_MAP, tx);
7436 spa->spa_avz_action = AVZ_ACTION_NONE;
7438 /* Create ZAPs for vdevs that don't have them. */
7439 vdev_construct_zaps(spa->spa_root_vdev, tx);
7441 config = spa_config_generate(spa, spa->spa_root_vdev,
7442 dmu_tx_get_txg(tx), B_FALSE);
7445 * If we're upgrading the spa version then make sure that
7446 * the config object gets updated with the correct version.
7448 if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
7449 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
7450 spa->spa_uberblock.ub_version);
7452 spa_config_exit(spa, SCL_STATE, FTAG);
7454 nvlist_free(spa->spa_config_syncing);
7455 spa->spa_config_syncing = config;
7457 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
7461 spa_sync_version(void *arg, dmu_tx_t *tx)
7463 uint64_t *versionp = arg;
7464 uint64_t version = *versionp;
7465 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7468 * Setting the version is special cased when first creating the pool.
7470 ASSERT(tx->tx_txg != TXG_INITIAL);
7472 ASSERT(SPA_VERSION_IS_SUPPORTED(version));
7473 ASSERT(version >= spa_version(spa));
7475 spa->spa_uberblock.ub_version = version;
7476 vdev_config_dirty(spa->spa_root_vdev);
7477 spa_history_log_internal(spa, "set", tx, "version=%lld", version);
7481 * Set zpool properties.
7484 spa_sync_props(void *arg, dmu_tx_t *tx)
7486 nvlist_t *nvp = arg;
7487 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
7488 objset_t *mos = spa->spa_meta_objset;
7489 nvpair_t *elem = NULL;
7491 mutex_enter(&spa->spa_props_lock);
7493 while ((elem = nvlist_next_nvpair(nvp, elem))) {
7495 char *strval, *fname;
7497 const char *propname;
7498 zprop_type_t proptype;
7501 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
7502 case ZPOOL_PROP_INVAL:
7504 * We checked this earlier in spa_prop_validate().
7506 ASSERT(zpool_prop_feature(nvpair_name(elem)));
7508 fname = strchr(nvpair_name(elem), '@') + 1;
7509 VERIFY0(zfeature_lookup_name(fname, &fid));
7511 spa_feature_enable(spa, fid, tx);
7512 spa_history_log_internal(spa, "set", tx,
7513 "%s=enabled", nvpair_name(elem));
7516 case ZPOOL_PROP_VERSION:
7517 intval = fnvpair_value_uint64(elem);
7519 * The version is synced seperatly before other
7520 * properties and should be correct by now.
7522 ASSERT3U(spa_version(spa), >=, intval);
7525 case ZPOOL_PROP_ALTROOT:
7527 * 'altroot' is a non-persistent property. It should
7528 * have been set temporarily at creation or import time.
7530 ASSERT(spa->spa_root != NULL);
7533 case ZPOOL_PROP_READONLY:
7534 case ZPOOL_PROP_CACHEFILE:
7536 * 'readonly' and 'cachefile' are also non-persisitent
7540 case ZPOOL_PROP_COMMENT:
7541 strval = fnvpair_value_string(elem);
7542 if (spa->spa_comment != NULL)
7543 spa_strfree(spa->spa_comment);
7544 spa->spa_comment = spa_strdup(strval);
7546 * We need to dirty the configuration on all the vdevs
7547 * so that their labels get updated. It's unnecessary
7548 * to do this for pool creation since the vdev's
7549 * configuratoin has already been dirtied.
7551 if (tx->tx_txg != TXG_INITIAL)
7552 vdev_config_dirty(spa->spa_root_vdev);
7553 spa_history_log_internal(spa, "set", tx,
7554 "%s=%s", nvpair_name(elem), strval);
7558 * Set pool property values in the poolprops mos object.
7560 if (spa->spa_pool_props_object == 0) {
7561 spa->spa_pool_props_object =
7562 zap_create_link(mos, DMU_OT_POOL_PROPS,
7563 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
7567 /* normalize the property name */
7568 propname = zpool_prop_to_name(prop);
7569 proptype = zpool_prop_get_type(prop);
7571 if (nvpair_type(elem) == DATA_TYPE_STRING) {
7572 ASSERT(proptype == PROP_TYPE_STRING);
7573 strval = fnvpair_value_string(elem);
7574 VERIFY0(zap_update(mos,
7575 spa->spa_pool_props_object, propname,
7576 1, strlen(strval) + 1, strval, tx));
7577 spa_history_log_internal(spa, "set", tx,
7578 "%s=%s", nvpair_name(elem), strval);
7579 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
7580 intval = fnvpair_value_uint64(elem);
7582 if (proptype == PROP_TYPE_INDEX) {
7584 VERIFY0(zpool_prop_index_to_string(
7585 prop, intval, &unused));
7587 VERIFY0(zap_update(mos,
7588 spa->spa_pool_props_object, propname,
7589 8, 1, &intval, tx));
7590 spa_history_log_internal(spa, "set", tx,
7591 "%s=%lld", nvpair_name(elem), intval);
7593 ASSERT(0); /* not allowed */
7597 case ZPOOL_PROP_DELEGATION:
7598 spa->spa_delegation = intval;
7600 case ZPOOL_PROP_BOOTFS:
7601 spa->spa_bootfs = intval;
7603 case ZPOOL_PROP_FAILUREMODE:
7604 spa->spa_failmode = intval;
7606 case ZPOOL_PROP_AUTOEXPAND:
7607 spa->spa_autoexpand = intval;
7608 if (tx->tx_txg != TXG_INITIAL)
7609 spa_async_request(spa,
7610 SPA_ASYNC_AUTOEXPAND);
7612 case ZPOOL_PROP_DEDUPDITTO:
7613 spa->spa_dedup_ditto = intval;
7622 mutex_exit(&spa->spa_props_lock);
7626 * Perform one-time upgrade on-disk changes. spa_version() does not
7627 * reflect the new version this txg, so there must be no changes this
7628 * txg to anything that the upgrade code depends on after it executes.
7629 * Therefore this must be called after dsl_pool_sync() does the sync
7633 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
7635 dsl_pool_t *dp = spa->spa_dsl_pool;
7637 ASSERT(spa->spa_sync_pass == 1);
7639 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7641 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
7642 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
7643 dsl_pool_create_origin(dp, tx);
7645 /* Keeping the origin open increases spa_minref */
7646 spa->spa_minref += 3;
7649 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
7650 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
7651 dsl_pool_upgrade_clones(dp, tx);
7654 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
7655 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
7656 dsl_pool_upgrade_dir_clones(dp, tx);
7658 /* Keeping the freedir open increases spa_minref */
7659 spa->spa_minref += 3;
7662 if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
7663 spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7664 spa_feature_create_zap_objects(spa, tx);
7668 * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
7669 * when possibility to use lz4 compression for metadata was added
7670 * Old pools that have this feature enabled must be upgraded to have
7671 * this feature active
7673 if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
7674 boolean_t lz4_en = spa_feature_is_enabled(spa,
7675 SPA_FEATURE_LZ4_COMPRESS);
7676 boolean_t lz4_ac = spa_feature_is_active(spa,
7677 SPA_FEATURE_LZ4_COMPRESS);
7679 if (lz4_en && !lz4_ac)
7680 spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
7684 * If we haven't written the salt, do so now. Note that the
7685 * feature may not be activated yet, but that's fine since
7686 * the presence of this ZAP entry is backwards compatible.
7688 if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
7689 DMU_POOL_CHECKSUM_SALT) == ENOENT) {
7690 VERIFY0(zap_add(spa->spa_meta_objset,
7691 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
7692 sizeof (spa->spa_cksum_salt.zcs_bytes),
7693 spa->spa_cksum_salt.zcs_bytes, tx));
7696 rrw_exit(&dp->dp_config_rwlock, FTAG);
7700 vdev_indirect_state_sync_verify(vdev_t *vd)
7702 vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
7703 vdev_indirect_births_t *vib = vd->vdev_indirect_births;
7705 if (vd->vdev_ops == &vdev_indirect_ops) {
7706 ASSERT(vim != NULL);
7707 ASSERT(vib != NULL);
7710 if (vdev_obsolete_sm_object(vd) != 0) {
7711 ASSERT(vd->vdev_obsolete_sm != NULL);
7712 ASSERT(vd->vdev_removing ||
7713 vd->vdev_ops == &vdev_indirect_ops);
7714 ASSERT(vdev_indirect_mapping_num_entries(vim) > 0);
7715 ASSERT(vdev_indirect_mapping_bytes_mapped(vim) > 0);
7717 ASSERT3U(vdev_obsolete_sm_object(vd), ==,
7718 space_map_object(vd->vdev_obsolete_sm));
7719 ASSERT3U(vdev_indirect_mapping_bytes_mapped(vim), >=,
7720 space_map_allocated(vd->vdev_obsolete_sm));
7722 ASSERT(vd->vdev_obsolete_segments != NULL);
7725 * Since frees / remaps to an indirect vdev can only
7726 * happen in syncing context, the obsolete segments
7727 * tree must be empty when we start syncing.
7729 ASSERT0(range_tree_space(vd->vdev_obsolete_segments));
7733 * Sync the specified transaction group. New blocks may be dirtied as
7734 * part of the process, so we iterate until it converges.
7737 spa_sync(spa_t *spa, uint64_t txg)
7739 dsl_pool_t *dp = spa->spa_dsl_pool;
7740 objset_t *mos = spa->spa_meta_objset;
7741 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
7742 vdev_t *rvd = spa->spa_root_vdev;
7746 uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
7747 zfs_vdev_queue_depth_pct / 100;
7749 VERIFY(spa_writeable(spa));
7752 * Wait for i/os issued in open context that need to complete
7753 * before this txg syncs.
7755 VERIFY0(zio_wait(spa->spa_txg_zio[txg & TXG_MASK]));
7756 spa->spa_txg_zio[txg & TXG_MASK] = zio_root(spa, NULL, NULL, 0);
7759 * Lock out configuration changes.
7761 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
7763 spa->spa_syncing_txg = txg;
7764 spa->spa_sync_pass = 0;
7766 mutex_enter(&spa->spa_alloc_lock);
7767 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
7768 mutex_exit(&spa->spa_alloc_lock);
7771 * If there are any pending vdev state changes, convert them
7772 * into config changes that go out with this transaction group.
7774 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7775 while (list_head(&spa->spa_state_dirty_list) != NULL) {
7777 * We need the write lock here because, for aux vdevs,
7778 * calling vdev_config_dirty() modifies sav_config.
7779 * This is ugly and will become unnecessary when we
7780 * eliminate the aux vdev wart by integrating all vdevs
7781 * into the root vdev tree.
7783 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7784 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
7785 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
7786 vdev_state_clean(vd);
7787 vdev_config_dirty(vd);
7789 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
7790 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
7792 spa_config_exit(spa, SCL_STATE, FTAG);
7794 tx = dmu_tx_create_assigned(dp, txg);
7796 spa->spa_sync_starttime = gethrtime();
7798 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
7799 spa->spa_sync_starttime + spa->spa_deadman_synctime));
7800 #else /* !illumos */
7802 callout_schedule(&spa->spa_deadman_cycid,
7803 hz * spa->spa_deadman_synctime / NANOSEC);
7805 #endif /* illumos */
7808 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
7809 * set spa_deflate if we have no raid-z vdevs.
7811 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
7812 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
7815 for (i = 0; i < rvd->vdev_children; i++) {
7816 vd = rvd->vdev_child[i];
7817 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
7820 if (i == rvd->vdev_children) {
7821 spa->spa_deflate = TRUE;
7822 VERIFY(0 == zap_add(spa->spa_meta_objset,
7823 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
7824 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
7829 * Set the top-level vdev's max queue depth. Evaluate each
7830 * top-level's async write queue depth in case it changed.
7831 * The max queue depth will not change in the middle of syncing
7834 uint64_t queue_depth_total = 0;
7835 for (int c = 0; c < rvd->vdev_children; c++) {
7836 vdev_t *tvd = rvd->vdev_child[c];
7837 metaslab_group_t *mg = tvd->vdev_mg;
7839 if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
7840 !metaslab_group_initialized(mg))
7844 * It is safe to do a lock-free check here because only async
7845 * allocations look at mg_max_alloc_queue_depth, and async
7846 * allocations all happen from spa_sync().
7848 ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
7849 mg->mg_max_alloc_queue_depth = max_queue_depth;
7850 queue_depth_total += mg->mg_max_alloc_queue_depth;
7852 metaslab_class_t *mc = spa_normal_class(spa);
7853 ASSERT0(refcount_count(&mc->mc_alloc_slots));
7854 mc->mc_alloc_max_slots = queue_depth_total;
7855 mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
7857 ASSERT3U(mc->mc_alloc_max_slots, <=,
7858 max_queue_depth * rvd->vdev_children);
7860 for (int c = 0; c < rvd->vdev_children; c++) {
7861 vdev_t *vd = rvd->vdev_child[c];
7862 vdev_indirect_state_sync_verify(vd);
7864 if (vdev_indirect_should_condense(vd)) {
7865 spa_condense_indirect_start_sync(vd, tx);
7871 * Iterate to convergence.
7874 int pass = ++spa->spa_sync_pass;
7876 spa_sync_config_object(spa, tx);
7877 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
7878 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
7879 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
7880 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
7881 spa_errlog_sync(spa, txg);
7882 dsl_pool_sync(dp, txg);
7884 if (pass < zfs_sync_pass_deferred_free) {
7885 spa_sync_frees(spa, free_bpl, tx);
7888 * We can not defer frees in pass 1, because
7889 * we sync the deferred frees later in pass 1.
7891 ASSERT3U(pass, >, 1);
7892 bplist_iterate(free_bpl, bpobj_enqueue_cb,
7893 &spa->spa_deferred_bpobj, tx);
7897 dsl_scan_sync(dp, tx);
7899 if (spa->spa_vdev_removal != NULL)
7902 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
7907 spa_sync_upgrades(spa, tx);
7909 spa->spa_uberblock.ub_rootbp.blk_birth);
7911 * Note: We need to check if the MOS is dirty
7912 * because we could have marked the MOS dirty
7913 * without updating the uberblock (e.g. if we
7914 * have sync tasks but no dirty user data). We
7915 * need to check the uberblock's rootbp because
7916 * it is updated if we have synced out dirty
7917 * data (though in this case the MOS will most
7918 * likely also be dirty due to second order
7919 * effects, we don't want to rely on that here).
7921 if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
7922 !dmu_objset_is_dirty(mos, txg)) {
7924 * Nothing changed on the first pass,
7925 * therefore this TXG is a no-op. Avoid
7926 * syncing deferred frees, so that we
7927 * can keep this TXG as a no-op.
7929 ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
7931 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
7932 ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
7933 ASSERT(txg_list_empty(&dp->dp_early_sync_tasks,
7937 spa_sync_deferred_frees(spa, tx);
7940 } while (dmu_objset_is_dirty(mos, txg));
7942 if (!list_is_empty(&spa->spa_config_dirty_list)) {
7944 * Make sure that the number of ZAPs for all the vdevs matches
7945 * the number of ZAPs in the per-vdev ZAP list. This only gets
7946 * called if the config is dirty; otherwise there may be
7947 * outstanding AVZ operations that weren't completed in
7948 * spa_sync_config_object.
7950 uint64_t all_vdev_zap_entry_count;
7951 ASSERT0(zap_count(spa->spa_meta_objset,
7952 spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
7953 ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
7954 all_vdev_zap_entry_count);
7957 if (spa->spa_vdev_removal != NULL) {
7958 ASSERT0(spa->spa_vdev_removal->svr_bytes_done[txg & TXG_MASK]);
7962 * Rewrite the vdev configuration (which includes the uberblock)
7963 * to commit the transaction group.
7965 * If there are no dirty vdevs, we sync the uberblock to a few
7966 * random top-level vdevs that are known to be visible in the
7967 * config cache (see spa_vdev_add() for a complete description).
7968 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
7972 * We hold SCL_STATE to prevent vdev open/close/etc.
7973 * while we're attempting to write the vdev labels.
7975 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
7977 if (list_is_empty(&spa->spa_config_dirty_list)) {
7978 vdev_t *svd[SPA_SYNC_MIN_VDEVS] = { NULL };
7980 int children = rvd->vdev_children;
7981 int c0 = spa_get_random(children);
7983 for (int c = 0; c < children; c++) {
7984 vd = rvd->vdev_child[(c0 + c) % children];
7986 /* Stop when revisiting the first vdev */
7987 if (c > 0 && svd[0] == vd)
7990 if (vd->vdev_ms_array == 0 || vd->vdev_islog ||
7991 !vdev_is_concrete(vd))
7994 svd[svdcount++] = vd;
7995 if (svdcount == SPA_SYNC_MIN_VDEVS)
7998 error = vdev_config_sync(svd, svdcount, txg);
8000 error = vdev_config_sync(rvd->vdev_child,
8001 rvd->vdev_children, txg);
8005 spa->spa_last_synced_guid = rvd->vdev_guid;
8007 spa_config_exit(spa, SCL_STATE, FTAG);
8011 zio_suspend(spa, NULL);
8012 zio_resume_wait(spa);
8017 VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
8018 #else /* !illumos */
8020 callout_drain(&spa->spa_deadman_cycid);
8022 #endif /* illumos */
8025 * Clear the dirty config list.
8027 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
8028 vdev_config_clean(vd);
8031 * Now that the new config has synced transactionally,
8032 * let it become visible to the config cache.
8034 if (spa->spa_config_syncing != NULL) {
8035 spa_config_set(spa, spa->spa_config_syncing);
8036 spa->spa_config_txg = txg;
8037 spa->spa_config_syncing = NULL;
8040 dsl_pool_sync_done(dp, txg);
8042 mutex_enter(&spa->spa_alloc_lock);
8043 VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
8044 mutex_exit(&spa->spa_alloc_lock);
8047 * Update usable space statistics.
8049 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
8050 vdev_sync_done(vd, txg);
8052 spa_update_dspace(spa);
8055 * It had better be the case that we didn't dirty anything
8056 * since vdev_config_sync().
8058 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
8059 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
8060 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
8062 while (zfs_pause_spa_sync)
8065 spa->spa_sync_pass = 0;
8068 * Update the last synced uberblock here. We want to do this at
8069 * the end of spa_sync() so that consumers of spa_last_synced_txg()
8070 * will be guaranteed that all the processing associated with
8071 * that txg has been completed.
8073 spa->spa_ubsync = spa->spa_uberblock;
8074 spa_config_exit(spa, SCL_CONFIG, FTAG);
8076 spa_handle_ignored_writes(spa);
8079 * If any async tasks have been requested, kick them off.
8081 spa_async_dispatch(spa);
8082 spa_async_dispatch_vd(spa);
8086 * Sync all pools. We don't want to hold the namespace lock across these
8087 * operations, so we take a reference on the spa_t and drop the lock during the
8091 spa_sync_allpools(void)
8094 mutex_enter(&spa_namespace_lock);
8095 while ((spa = spa_next(spa)) != NULL) {
8096 if (spa_state(spa) != POOL_STATE_ACTIVE ||
8097 !spa_writeable(spa) || spa_suspended(spa))
8099 spa_open_ref(spa, FTAG);
8100 mutex_exit(&spa_namespace_lock);
8101 txg_wait_synced(spa_get_dsl(spa), 0);
8102 mutex_enter(&spa_namespace_lock);
8103 spa_close(spa, FTAG);
8105 mutex_exit(&spa_namespace_lock);
8109 * ==========================================================================
8110 * Miscellaneous routines
8111 * ==========================================================================
8115 * Remove all pools in the system.
8123 * Remove all cached state. All pools should be closed now,
8124 * so every spa in the AVL tree should be unreferenced.
8126 mutex_enter(&spa_namespace_lock);
8127 while ((spa = spa_next(NULL)) != NULL) {
8129 * Stop async tasks. The async thread may need to detach
8130 * a device that's been replaced, which requires grabbing
8131 * spa_namespace_lock, so we must drop it here.
8133 spa_open_ref(spa, FTAG);
8134 mutex_exit(&spa_namespace_lock);
8135 spa_async_suspend(spa);
8136 mutex_enter(&spa_namespace_lock);
8137 spa_close(spa, FTAG);
8139 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
8141 spa_deactivate(spa);
8145 mutex_exit(&spa_namespace_lock);
8149 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
8154 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
8158 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
8159 vd = spa->spa_l2cache.sav_vdevs[i];
8160 if (vd->vdev_guid == guid)
8164 for (i = 0; i < spa->spa_spares.sav_count; i++) {
8165 vd = spa->spa_spares.sav_vdevs[i];
8166 if (vd->vdev_guid == guid)
8175 spa_upgrade(spa_t *spa, uint64_t version)
8177 ASSERT(spa_writeable(spa));
8179 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
8182 * This should only be called for a non-faulted pool, and since a
8183 * future version would result in an unopenable pool, this shouldn't be
8186 ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
8187 ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
8189 spa->spa_uberblock.ub_version = version;
8190 vdev_config_dirty(spa->spa_root_vdev);
8192 spa_config_exit(spa, SCL_ALL, FTAG);
8194 txg_wait_synced(spa_get_dsl(spa), 0);
8198 spa_has_spare(spa_t *spa, uint64_t guid)
8202 spa_aux_vdev_t *sav = &spa->spa_spares;
8204 for (i = 0; i < sav->sav_count; i++)
8205 if (sav->sav_vdevs[i]->vdev_guid == guid)
8208 for (i = 0; i < sav->sav_npending; i++) {
8209 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
8210 &spareguid) == 0 && spareguid == guid)
8218 * Check if a pool has an active shared spare device.
8219 * Note: reference count of an active spare is 2, as a spare and as a replace
8222 spa_has_active_shared_spare(spa_t *spa)
8226 spa_aux_vdev_t *sav = &spa->spa_spares;
8228 for (i = 0; i < sav->sav_count; i++) {
8229 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
8230 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
8239 spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8241 sysevent_t *ev = NULL;
8243 sysevent_attr_list_t *attr = NULL;
8244 sysevent_value_t value;
8246 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
8250 value.value_type = SE_DATA_TYPE_STRING;
8251 value.value.sv_string = spa_name(spa);
8252 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
8255 value.value_type = SE_DATA_TYPE_UINT64;
8256 value.value.sv_uint64 = spa_guid(spa);
8257 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
8261 value.value_type = SE_DATA_TYPE_UINT64;
8262 value.value.sv_uint64 = vd->vdev_guid;
8263 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
8267 if (vd->vdev_path) {
8268 value.value_type = SE_DATA_TYPE_STRING;
8269 value.value.sv_string = vd->vdev_path;
8270 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
8271 &value, SE_SLEEP) != 0)
8276 if (hist_nvl != NULL) {
8277 fnvlist_merge((nvlist_t *)attr, hist_nvl);
8280 if (sysevent_attach_attributes(ev, attr) != 0)
8286 sysevent_free_attr(attr);
8293 spa_event_post(sysevent_t *ev)
8298 (void) log_sysevent(ev, SE_SLEEP, &eid);
8304 spa_event_discard(sysevent_t *ev)
8312 * Post a sysevent corresponding to the given event. The 'name' must be one of
8313 * the event definitions in sys/sysevent/eventdefs.h. The payload will be
8314 * filled in from the spa and (optionally) the vdev and history nvl. This
8315 * doesn't do anything in the userland libzpool, as we don't want consumers to
8316 * misinterpret ztest or zdb as real changes.
8319 spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
8321 spa_event_post(spa_event_create(spa, vd, hist_nvl, name));