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.
27 * This file contains all the routines used when modifying on-disk SPA state.
28 * This includes opening, importing, destroying, exporting a pool, and syncing a
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/spa_impl.h>
36 #include <sys/zio_checksum.h>
38 #include <sys/dmu_tx.h>
42 #include <sys/vdev_impl.h>
43 #include <sys/vdev_disk.h>
44 #include <sys/metaslab.h>
45 #include <sys/metaslab_impl.h>
46 #include <sys/uberblock_impl.h>
49 #include <sys/dmu_traverse.h>
50 #include <sys/dmu_objset.h>
51 #include <sys/unique.h>
52 #include <sys/dsl_pool.h>
53 #include <sys/dsl_dataset.h>
54 #include <sys/dsl_dir.h>
55 #include <sys/dsl_prop.h>
56 #include <sys/dsl_synctask.h>
57 #include <sys/fs/zfs.h>
59 #include <sys/callb.h>
60 #include <sys/systeminfo.h>
61 #include <sys/spa_boot.h>
62 #include <sys/zfs_ioctl.h>
63 #include <sys/dsl_scan.h>
66 #include <sys/bootprops.h>
67 #include <sys/callb.h>
68 #include <sys/cpupart.h>
70 #include <sys/sysdc.h>
75 #include "zfs_comutil.h"
77 typedef enum zti_modes {
78 zti_mode_fixed, /* value is # of threads (min 1) */
79 zti_mode_online_percent, /* value is % of online CPUs */
80 zti_mode_batch, /* cpu-intensive; value is ignored */
81 zti_mode_null, /* don't create a taskq */
85 #define ZTI_FIX(n) { zti_mode_fixed, (n) }
86 #define ZTI_PCT(n) { zti_mode_online_percent, (n) }
87 #define ZTI_BATCH { zti_mode_batch, 0 }
88 #define ZTI_NULL { zti_mode_null, 0 }
90 #define ZTI_ONE ZTI_FIX(1)
92 typedef struct zio_taskq_info {
93 enum zti_modes zti_mode;
97 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
98 "iss", "iss_h", "int", "int_h"
102 * Define the taskq threads for the following I/O types:
103 * NULL, READ, WRITE, FREE, CLAIM, and IOCTL
105 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
106 /* ISSUE ISSUE_HIGH INTR INTR_HIGH */
107 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
108 { ZTI_FIX(8), ZTI_NULL, ZTI_BATCH, ZTI_NULL },
109 { ZTI_BATCH, ZTI_FIX(5), ZTI_FIX(16), ZTI_FIX(5) },
110 { ZTI_FIX(100), ZTI_NULL, ZTI_ONE, ZTI_NULL },
111 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
112 { ZTI_ONE, ZTI_NULL, ZTI_ONE, ZTI_NULL },
115 static dsl_syncfunc_t spa_sync_props;
116 static boolean_t spa_has_active_shared_spare(spa_t *spa);
117 static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
118 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
120 static void spa_vdev_resilver_done(spa_t *spa);
122 uint_t zio_taskq_batch_pct = 100; /* 1 thread per cpu in pset */
123 id_t zio_taskq_psrset_bind = PS_NONE;
124 boolean_t zio_taskq_sysdc = B_TRUE; /* use SDC scheduling class */
125 uint_t zio_taskq_basedc = 80; /* base duty cycle */
127 boolean_t spa_create_process = B_TRUE; /* no process ==> no sysdc */
130 * This (illegal) pool name is used when temporarily importing a spa_t in order
131 * to get the vdev stats associated with the imported devices.
133 #define TRYIMPORT_NAME "$import"
136 * ==========================================================================
137 * SPA properties routines
138 * ==========================================================================
142 * Add a (source=src, propname=propval) list to an nvlist.
145 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
146 uint64_t intval, zprop_source_t src)
148 const char *propname = zpool_prop_to_name(prop);
151 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
152 VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
155 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
157 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
159 VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
160 nvlist_free(propval);
164 * Get property values from the spa configuration.
167 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
171 uint64_t cap, version;
172 zprop_source_t src = ZPROP_SRC_NONE;
173 spa_config_dirent_t *dp;
175 ASSERT(MUTEX_HELD(&spa->spa_props_lock));
177 if (spa->spa_root_vdev != NULL) {
178 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
179 size = metaslab_class_get_space(spa_normal_class(spa));
180 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
181 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
182 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
183 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
185 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
186 (spa_mode(spa) == FREAD), src);
188 cap = (size == 0) ? 0 : (alloc * 100 / size);
189 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
191 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
192 ddt_get_pool_dedup_ratio(spa), src);
194 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
195 spa->spa_root_vdev->vdev_state, src);
197 version = spa_version(spa);
198 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
199 src = ZPROP_SRC_DEFAULT;
201 src = ZPROP_SRC_LOCAL;
202 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
205 spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
207 if (spa->spa_root != NULL)
208 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
211 if ((dp = list_head(&spa->spa_config_list)) != NULL) {
212 if (dp->scd_path == NULL) {
213 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
214 "none", 0, ZPROP_SRC_LOCAL);
215 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
216 spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
217 dp->scd_path, 0, ZPROP_SRC_LOCAL);
223 * Get zpool property values.
226 spa_prop_get(spa_t *spa, nvlist_t **nvp)
228 objset_t *mos = spa->spa_meta_objset;
233 err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
237 mutex_enter(&spa->spa_props_lock);
240 * Get properties from the spa config.
242 spa_prop_get_config(spa, nvp);
244 /* If no pool property object, no more prop to get. */
245 if (mos == NULL || spa->spa_pool_props_object == 0) {
246 mutex_exit(&spa->spa_props_lock);
251 * Get properties from the MOS pool property object.
253 for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
254 (err = zap_cursor_retrieve(&zc, &za)) == 0;
255 zap_cursor_advance(&zc)) {
258 zprop_source_t src = ZPROP_SRC_DEFAULT;
261 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
264 switch (za.za_integer_length) {
266 /* integer property */
267 if (za.za_first_integer !=
268 zpool_prop_default_numeric(prop))
269 src = ZPROP_SRC_LOCAL;
271 if (prop == ZPOOL_PROP_BOOTFS) {
273 dsl_dataset_t *ds = NULL;
275 dp = spa_get_dsl(spa);
276 rw_enter(&dp->dp_config_rwlock, RW_READER);
277 if ((err = dsl_dataset_hold_obj(dp,
278 za.za_first_integer, FTAG, &ds))) {
279 rw_exit(&dp->dp_config_rwlock);
284 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
286 dsl_dataset_name(ds, strval);
287 dsl_dataset_rele(ds, FTAG);
288 rw_exit(&dp->dp_config_rwlock);
291 intval = za.za_first_integer;
294 spa_prop_add_list(*nvp, prop, strval, intval, src);
298 MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
303 /* string property */
304 strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
305 err = zap_lookup(mos, spa->spa_pool_props_object,
306 za.za_name, 1, za.za_num_integers, strval);
308 kmem_free(strval, za.za_num_integers);
311 spa_prop_add_list(*nvp, prop, strval, 0, src);
312 kmem_free(strval, za.za_num_integers);
319 zap_cursor_fini(&zc);
320 mutex_exit(&spa->spa_props_lock);
322 if (err && err != ENOENT) {
332 * Validate the given pool properties nvlist and modify the list
333 * for the property values to be set.
336 spa_prop_validate(spa_t *spa, nvlist_t *props)
339 int error = 0, reset_bootfs = 0;
343 while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
345 char *propname, *strval;
350 propname = nvpair_name(elem);
352 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
356 case ZPOOL_PROP_VERSION:
357 error = nvpair_value_uint64(elem, &intval);
359 (intval < spa_version(spa) || intval > SPA_VERSION))
363 case ZPOOL_PROP_DELEGATION:
364 case ZPOOL_PROP_AUTOREPLACE:
365 case ZPOOL_PROP_LISTSNAPS:
366 case ZPOOL_PROP_AUTOEXPAND:
367 error = nvpair_value_uint64(elem, &intval);
368 if (!error && intval > 1)
372 case ZPOOL_PROP_BOOTFS:
374 * If the pool version is less than SPA_VERSION_BOOTFS,
375 * or the pool is still being created (version == 0),
376 * the bootfs property cannot be set.
378 if (spa_version(spa) < SPA_VERSION_BOOTFS) {
384 * Make sure the vdev config is bootable
386 if (!vdev_is_bootable(spa->spa_root_vdev)) {
393 error = nvpair_value_string(elem, &strval);
398 if (strval == NULL || strval[0] == '\0') {
399 objnum = zpool_prop_default_numeric(
404 if ((error = dmu_objset_hold(strval,FTAG,&os)))
407 /* Must be ZPL and not gzip compressed. */
409 if (dmu_objset_type(os) != DMU_OST_ZFS) {
411 } else if ((error = dsl_prop_get_integer(strval,
412 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
413 &compress, NULL)) == 0 &&
414 !BOOTFS_COMPRESS_VALID(compress)) {
417 objnum = dmu_objset_id(os);
419 dmu_objset_rele(os, FTAG);
423 case ZPOOL_PROP_FAILUREMODE:
424 error = nvpair_value_uint64(elem, &intval);
425 if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
426 intval > ZIO_FAILURE_MODE_PANIC))
430 * This is a special case which only occurs when
431 * the pool has completely failed. This allows
432 * the user to change the in-core failmode property
433 * without syncing it out to disk (I/Os might
434 * currently be blocked). We do this by returning
435 * EIO to the caller (spa_prop_set) to trick it
436 * into thinking we encountered a property validation
439 if (!error && spa_suspended(spa)) {
440 spa->spa_failmode = intval;
445 case ZPOOL_PROP_CACHEFILE:
446 if ((error = nvpair_value_string(elem, &strval)) != 0)
449 if (strval[0] == '\0')
452 if (strcmp(strval, "none") == 0)
455 if (strval[0] != '/') {
460 slash = strrchr(strval, '/');
461 ASSERT(slash != NULL);
463 if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
464 strcmp(slash, "/..") == 0)
468 case ZPOOL_PROP_DEDUPDITTO:
469 if (spa_version(spa) < SPA_VERSION_DEDUP)
472 error = nvpair_value_uint64(elem, &intval);
474 intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
486 if (!error && reset_bootfs) {
487 error = nvlist_remove(props,
488 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
491 error = nvlist_add_uint64(props,
492 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
500 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
503 spa_config_dirent_t *dp;
505 if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
509 dp = kmem_alloc(sizeof (spa_config_dirent_t),
512 if (cachefile[0] == '\0')
513 dp->scd_path = spa_strdup(spa_config_path);
514 else if (strcmp(cachefile, "none") == 0)
517 dp->scd_path = spa_strdup(cachefile);
519 list_insert_head(&spa->spa_config_list, dp);
521 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
525 spa_prop_set(spa_t *spa, nvlist_t *nvp)
529 boolean_t need_sync = B_FALSE;
532 if ((error = spa_prop_validate(spa, nvp)) != 0)
536 while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
537 if ((prop = zpool_name_to_prop(
538 nvpair_name(elem))) == ZPROP_INVAL)
541 if (prop == ZPOOL_PROP_CACHEFILE ||
542 prop == ZPOOL_PROP_ALTROOT ||
543 prop == ZPOOL_PROP_READONLY)
551 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
558 * If the bootfs property value is dsobj, clear it.
561 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
563 if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
564 VERIFY(zap_remove(spa->spa_meta_objset,
565 spa->spa_pool_props_object,
566 zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
572 * ==========================================================================
573 * SPA state manipulation (open/create/destroy/import/export)
574 * ==========================================================================
578 spa_error_entry_compare(const void *a, const void *b)
580 spa_error_entry_t *sa = (spa_error_entry_t *)a;
581 spa_error_entry_t *sb = (spa_error_entry_t *)b;
584 ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
585 sizeof (zbookmark_t));
596 * Utility function which retrieves copies of the current logs and
597 * re-initializes them in the process.
600 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
602 ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
604 bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
605 bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
607 avl_create(&spa->spa_errlist_scrub,
608 spa_error_entry_compare, sizeof (spa_error_entry_t),
609 offsetof(spa_error_entry_t, se_avl));
610 avl_create(&spa->spa_errlist_last,
611 spa_error_entry_compare, sizeof (spa_error_entry_t),
612 offsetof(spa_error_entry_t, se_avl));
616 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
619 uint_t flags = TASKQ_PREPOPULATE;
620 boolean_t batch = B_FALSE;
624 return (NULL); /* no taskq needed */
627 ASSERT3U(value, >=, 1);
628 value = MAX(value, 1);
633 flags |= TASKQ_THREADS_CPU_PCT;
634 value = zio_taskq_batch_pct;
637 case zti_mode_online_percent:
638 flags |= TASKQ_THREADS_CPU_PCT;
642 panic("unrecognized mode for %s taskq (%u:%u) in "
648 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
650 flags |= TASKQ_DC_BATCH;
652 return (taskq_create_sysdc(name, value, 50, INT_MAX,
653 spa->spa_proc, zio_taskq_basedc, flags));
655 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
656 spa->spa_proc, flags));
660 spa_create_zio_taskqs(spa_t *spa)
664 for (t = 0; t < ZIO_TYPES; t++) {
665 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
666 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
667 enum zti_modes mode = ztip->zti_mode;
668 uint_t value = ztip->zti_value;
671 (void) snprintf(name, sizeof (name),
672 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
674 spa->spa_zio_taskq[t][q] =
675 spa_taskq_create(spa, name, mode, value);
680 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
682 spa_thread(void *arg)
687 user_t *pu = PTOU(curproc);
689 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
692 ASSERT(curproc != &p0);
693 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
694 "zpool-%s", spa->spa_name);
695 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
697 /* bind this thread to the requested psrset */
698 if (zio_taskq_psrset_bind != PS_NONE) {
700 mutex_enter(&cpu_lock);
701 mutex_enter(&pidlock);
702 mutex_enter(&curproc->p_lock);
704 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
705 0, NULL, NULL) == 0) {
706 curthread->t_bind_pset = zio_taskq_psrset_bind;
709 "Couldn't bind process for zfs pool \"%s\" to "
710 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
713 mutex_exit(&curproc->p_lock);
714 mutex_exit(&pidlock);
715 mutex_exit(&cpu_lock);
719 if (zio_taskq_sysdc) {
720 sysdc_thread_enter(curthread, 100, 0);
723 spa->spa_proc = curproc;
724 spa->spa_did = curthread->t_did;
726 spa_create_zio_taskqs(spa);
728 mutex_enter(&spa->spa_proc_lock);
729 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
731 spa->spa_proc_state = SPA_PROC_ACTIVE;
732 cv_broadcast(&spa->spa_proc_cv);
734 CALLB_CPR_SAFE_BEGIN(&cprinfo);
735 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
736 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
737 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
739 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
740 spa->spa_proc_state = SPA_PROC_GONE;
742 cv_broadcast(&spa->spa_proc_cv);
743 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
745 mutex_enter(&curproc->p_lock);
751 * Activate an uninitialized pool.
754 spa_activate(spa_t *spa, int mode)
756 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
758 spa->spa_state = POOL_STATE_ACTIVE;
759 spa->spa_mode = mode;
761 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
762 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
764 /* Try to create a covering process */
765 mutex_enter(&spa->spa_proc_lock);
766 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
767 ASSERT(spa->spa_proc == &p0);
770 #ifdef HAVE_SPA_THREAD
771 /* Only create a process if we're going to be around a while. */
772 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
773 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
775 spa->spa_proc_state = SPA_PROC_CREATED;
776 while (spa->spa_proc_state == SPA_PROC_CREATED) {
777 cv_wait(&spa->spa_proc_cv,
778 &spa->spa_proc_lock);
780 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
781 ASSERT(spa->spa_proc != &p0);
782 ASSERT(spa->spa_did != 0);
786 "Couldn't create process for zfs pool \"%s\"\n",
791 #endif /* HAVE_SPA_THREAD */
792 mutex_exit(&spa->spa_proc_lock);
794 /* If we didn't create a process, we need to create our taskqs. */
795 if (spa->spa_proc == &p0) {
796 spa_create_zio_taskqs(spa);
799 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
800 offsetof(vdev_t, vdev_config_dirty_node));
801 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
802 offsetof(vdev_t, vdev_state_dirty_node));
804 txg_list_create(&spa->spa_vdev_txg_list,
805 offsetof(struct vdev, vdev_txg_node));
807 avl_create(&spa->spa_errlist_scrub,
808 spa_error_entry_compare, sizeof (spa_error_entry_t),
809 offsetof(spa_error_entry_t, se_avl));
810 avl_create(&spa->spa_errlist_last,
811 spa_error_entry_compare, sizeof (spa_error_entry_t),
812 offsetof(spa_error_entry_t, se_avl));
816 * Opposite of spa_activate().
819 spa_deactivate(spa_t *spa)
823 ASSERT(spa->spa_sync_on == B_FALSE);
824 ASSERT(spa->spa_dsl_pool == NULL);
825 ASSERT(spa->spa_root_vdev == NULL);
826 ASSERT(spa->spa_async_zio_root == NULL);
827 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
829 txg_list_destroy(&spa->spa_vdev_txg_list);
831 list_destroy(&spa->spa_config_dirty_list);
832 list_destroy(&spa->spa_state_dirty_list);
834 for (t = 0; t < ZIO_TYPES; t++) {
835 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
836 if (spa->spa_zio_taskq[t][q] != NULL)
837 taskq_destroy(spa->spa_zio_taskq[t][q]);
838 spa->spa_zio_taskq[t][q] = NULL;
842 metaslab_class_destroy(spa->spa_normal_class);
843 spa->spa_normal_class = NULL;
845 metaslab_class_destroy(spa->spa_log_class);
846 spa->spa_log_class = NULL;
849 * If this was part of an import or the open otherwise failed, we may
850 * still have errors left in the queues. Empty them just in case.
852 spa_errlog_drain(spa);
854 avl_destroy(&spa->spa_errlist_scrub);
855 avl_destroy(&spa->spa_errlist_last);
857 spa->spa_state = POOL_STATE_UNINITIALIZED;
859 mutex_enter(&spa->spa_proc_lock);
860 if (spa->spa_proc_state != SPA_PROC_NONE) {
861 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
862 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
863 cv_broadcast(&spa->spa_proc_cv);
864 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
865 ASSERT(spa->spa_proc != &p0);
866 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
868 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
869 spa->spa_proc_state = SPA_PROC_NONE;
871 ASSERT(spa->spa_proc == &p0);
872 mutex_exit(&spa->spa_proc_lock);
875 * We want to make sure spa_thread() has actually exited the ZFS
876 * module, so that the module can't be unloaded out from underneath
879 if (spa->spa_did != 0) {
880 thread_join(spa->spa_did);
886 * Verify a pool configuration, and construct the vdev tree appropriately. This
887 * will create all the necessary vdevs in the appropriate layout, with each vdev
888 * in the CLOSED state. This will prep the pool before open/creation/import.
889 * All vdev validation is done by the vdev_alloc() routine.
892 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
893 uint_t id, int atype)
900 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
903 if ((*vdp)->vdev_ops->vdev_op_leaf)
906 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
918 for (c = 0; c < children; c++) {
920 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
928 ASSERT(*vdp != NULL);
934 * Opposite of spa_load().
937 spa_unload(spa_t *spa)
941 ASSERT(MUTEX_HELD(&spa_namespace_lock));
946 spa_async_suspend(spa);
951 if (spa->spa_sync_on) {
952 txg_sync_stop(spa->spa_dsl_pool);
953 spa->spa_sync_on = B_FALSE;
957 * Wait for any outstanding async I/O to complete.
959 if (spa->spa_async_zio_root != NULL) {
960 (void) zio_wait(spa->spa_async_zio_root);
961 spa->spa_async_zio_root = NULL;
964 bpobj_close(&spa->spa_deferred_bpobj);
967 * Close the dsl pool.
969 if (spa->spa_dsl_pool) {
970 dsl_pool_close(spa->spa_dsl_pool);
971 spa->spa_dsl_pool = NULL;
972 spa->spa_meta_objset = NULL;
977 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
980 * Drop and purge level 2 cache
982 spa_l2cache_drop(spa);
987 if (spa->spa_root_vdev)
988 vdev_free(spa->spa_root_vdev);
989 ASSERT(spa->spa_root_vdev == NULL);
991 for (i = 0; i < spa->spa_spares.sav_count; i++)
992 vdev_free(spa->spa_spares.sav_vdevs[i]);
993 if (spa->spa_spares.sav_vdevs) {
994 kmem_free(spa->spa_spares.sav_vdevs,
995 spa->spa_spares.sav_count * sizeof (void *));
996 spa->spa_spares.sav_vdevs = NULL;
998 if (spa->spa_spares.sav_config) {
999 nvlist_free(spa->spa_spares.sav_config);
1000 spa->spa_spares.sav_config = NULL;
1002 spa->spa_spares.sav_count = 0;
1004 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
1005 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1006 if (spa->spa_l2cache.sav_vdevs) {
1007 kmem_free(spa->spa_l2cache.sav_vdevs,
1008 spa->spa_l2cache.sav_count * sizeof (void *));
1009 spa->spa_l2cache.sav_vdevs = NULL;
1011 if (spa->spa_l2cache.sav_config) {
1012 nvlist_free(spa->spa_l2cache.sav_config);
1013 spa->spa_l2cache.sav_config = NULL;
1015 spa->spa_l2cache.sav_count = 0;
1017 spa->spa_async_suspended = 0;
1019 spa_config_exit(spa, SCL_ALL, FTAG);
1023 * Load (or re-load) the current list of vdevs describing the active spares for
1024 * this pool. When this is called, we have some form of basic information in
1025 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1026 * then re-generate a more complete list including status information.
1029 spa_load_spares(spa_t *spa)
1036 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1039 * First, close and free any existing spare vdevs.
1041 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1042 vd = spa->spa_spares.sav_vdevs[i];
1044 /* Undo the call to spa_activate() below */
1045 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1046 B_FALSE)) != NULL && tvd->vdev_isspare)
1047 spa_spare_remove(tvd);
1052 if (spa->spa_spares.sav_vdevs)
1053 kmem_free(spa->spa_spares.sav_vdevs,
1054 spa->spa_spares.sav_count * sizeof (void *));
1056 if (spa->spa_spares.sav_config == NULL)
1059 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1060 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1062 spa->spa_spares.sav_count = (int)nspares;
1063 spa->spa_spares.sav_vdevs = NULL;
1069 * Construct the array of vdevs, opening them to get status in the
1070 * process. For each spare, there is potentially two different vdev_t
1071 * structures associated with it: one in the list of spares (used only
1072 * for basic validation purposes) and one in the active vdev
1073 * configuration (if it's spared in). During this phase we open and
1074 * validate each vdev on the spare list. If the vdev also exists in the
1075 * active configuration, then we also mark this vdev as an active spare.
1077 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1079 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1080 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1081 VDEV_ALLOC_SPARE) == 0);
1084 spa->spa_spares.sav_vdevs[i] = vd;
1086 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1087 B_FALSE)) != NULL) {
1088 if (!tvd->vdev_isspare)
1092 * We only mark the spare active if we were successfully
1093 * able to load the vdev. Otherwise, importing a pool
1094 * with a bad active spare would result in strange
1095 * behavior, because multiple pool would think the spare
1096 * is actively in use.
1098 * There is a vulnerability here to an equally bizarre
1099 * circumstance, where a dead active spare is later
1100 * brought back to life (onlined or otherwise). Given
1101 * the rarity of this scenario, and the extra complexity
1102 * it adds, we ignore the possibility.
1104 if (!vdev_is_dead(tvd))
1105 spa_spare_activate(tvd);
1109 vd->vdev_aux = &spa->spa_spares;
1111 if (vdev_open(vd) != 0)
1114 if (vdev_validate_aux(vd) == 0)
1119 * Recompute the stashed list of spares, with status information
1122 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1123 DATA_TYPE_NVLIST_ARRAY) == 0);
1125 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1127 for (i = 0; i < spa->spa_spares.sav_count; i++)
1128 spares[i] = vdev_config_generate(spa,
1129 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1130 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1131 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1132 for (i = 0; i < spa->spa_spares.sav_count; i++)
1133 nvlist_free(spares[i]);
1134 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1138 * Load (or re-load) the current list of vdevs describing the active l2cache for
1139 * this pool. When this is called, we have some form of basic information in
1140 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1141 * then re-generate a more complete list including status information.
1142 * Devices which are already active have their details maintained, and are
1146 spa_load_l2cache(spa_t *spa)
1150 int i, j, oldnvdevs;
1152 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1153 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1155 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1157 if (sav->sav_config != NULL) {
1158 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1159 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1160 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1165 oldvdevs = sav->sav_vdevs;
1166 oldnvdevs = sav->sav_count;
1167 sav->sav_vdevs = NULL;
1171 * Process new nvlist of vdevs.
1173 for (i = 0; i < nl2cache; i++) {
1174 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1178 for (j = 0; j < oldnvdevs; j++) {
1180 if (vd != NULL && guid == vd->vdev_guid) {
1182 * Retain previous vdev for add/remove ops.
1190 if (newvdevs[i] == NULL) {
1194 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1195 VDEV_ALLOC_L2CACHE) == 0);
1200 * Commit this vdev as an l2cache device,
1201 * even if it fails to open.
1203 spa_l2cache_add(vd);
1208 spa_l2cache_activate(vd);
1210 if (vdev_open(vd) != 0)
1213 (void) vdev_validate_aux(vd);
1215 if (!vdev_is_dead(vd))
1216 l2arc_add_vdev(spa, vd);
1221 * Purge vdevs that were dropped
1223 for (i = 0; i < oldnvdevs; i++) {
1228 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1229 pool != 0ULL && l2arc_vdev_present(vd))
1230 l2arc_remove_vdev(vd);
1231 (void) vdev_close(vd);
1232 spa_l2cache_remove(vd);
1237 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1239 if (sav->sav_config == NULL)
1242 sav->sav_vdevs = newvdevs;
1243 sav->sav_count = (int)nl2cache;
1246 * Recompute the stashed list of l2cache devices, with status
1247 * information this time.
1249 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1250 DATA_TYPE_NVLIST_ARRAY) == 0);
1252 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1253 for (i = 0; i < sav->sav_count; i++)
1254 l2cache[i] = vdev_config_generate(spa,
1255 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1256 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1257 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1259 for (i = 0; i < sav->sav_count; i++)
1260 nvlist_free(l2cache[i]);
1262 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1266 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1269 char *packed = NULL;
1274 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1275 nvsize = *(uint64_t *)db->db_data;
1276 dmu_buf_rele(db, FTAG);
1278 packed = kmem_alloc(nvsize, KM_SLEEP | KM_NODEBUG);
1279 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1282 error = nvlist_unpack(packed, nvsize, value, 0);
1283 kmem_free(packed, nvsize);
1289 * Checks to see if the given vdev could not be opened, in which case we post a
1290 * sysevent to notify the autoreplace code that the device has been removed.
1293 spa_check_removed(vdev_t *vd)
1297 for (c = 0; c < vd->vdev_children; c++)
1298 spa_check_removed(vd->vdev_child[c]);
1300 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1301 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1302 vd->vdev_spa, vd, NULL, 0, 0);
1303 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1308 * Validate the current config against the MOS config
1311 spa_config_valid(spa_t *spa, nvlist_t *config)
1313 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1317 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1319 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1320 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1322 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1325 * If we're doing a normal import, then build up any additional
1326 * diagnostic information about missing devices in this config.
1327 * We'll pass this up to the user for further processing.
1329 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1330 nvlist_t **child, *nv;
1333 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1335 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1337 for (c = 0; c < rvd->vdev_children; c++) {
1338 vdev_t *tvd = rvd->vdev_child[c];
1339 vdev_t *mtvd = mrvd->vdev_child[c];
1341 if (tvd->vdev_ops == &vdev_missing_ops &&
1342 mtvd->vdev_ops != &vdev_missing_ops &&
1344 child[idx++] = vdev_config_generate(spa, mtvd,
1349 VERIFY(nvlist_add_nvlist_array(nv,
1350 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1351 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1352 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1354 for (i = 0; i < idx; i++)
1355 nvlist_free(child[i]);
1358 kmem_free(child, rvd->vdev_children * sizeof (char **));
1362 * Compare the root vdev tree with the information we have
1363 * from the MOS config (mrvd). Check each top-level vdev
1364 * with the corresponding MOS config top-level (mtvd).
1366 for (c = 0; c < rvd->vdev_children; c++) {
1367 vdev_t *tvd = rvd->vdev_child[c];
1368 vdev_t *mtvd = mrvd->vdev_child[c];
1371 * Resolve any "missing" vdevs in the current configuration.
1372 * If we find that the MOS config has more accurate information
1373 * about the top-level vdev then use that vdev instead.
1375 if (tvd->vdev_ops == &vdev_missing_ops &&
1376 mtvd->vdev_ops != &vdev_missing_ops) {
1378 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1382 * Device specific actions.
1384 if (mtvd->vdev_islog) {
1385 spa_set_log_state(spa, SPA_LOG_CLEAR);
1388 * XXX - once we have 'readonly' pool
1389 * support we should be able to handle
1390 * missing data devices by transitioning
1391 * the pool to readonly.
1397 * Swap the missing vdev with the data we were
1398 * able to obtain from the MOS config.
1400 vdev_remove_child(rvd, tvd);
1401 vdev_remove_child(mrvd, mtvd);
1403 vdev_add_child(rvd, mtvd);
1404 vdev_add_child(mrvd, tvd);
1406 spa_config_exit(spa, SCL_ALL, FTAG);
1408 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1411 } else if (mtvd->vdev_islog) {
1413 * Load the slog device's state from the MOS config
1414 * since it's possible that the label does not
1415 * contain the most up-to-date information.
1417 vdev_load_log_state(tvd, mtvd);
1422 spa_config_exit(spa, SCL_ALL, FTAG);
1425 * Ensure we were able to validate the config.
1427 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1431 * Check for missing log devices
1434 spa_check_logs(spa_t *spa)
1436 switch (spa->spa_log_state) {
1439 case SPA_LOG_MISSING:
1440 /* need to recheck in case slog has been restored */
1441 case SPA_LOG_UNKNOWN:
1442 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1443 DS_FIND_CHILDREN)) {
1444 spa_set_log_state(spa, SPA_LOG_MISSING);
1453 spa_passivate_log(spa_t *spa)
1455 vdev_t *rvd = spa->spa_root_vdev;
1456 boolean_t slog_found = B_FALSE;
1459 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1461 if (!spa_has_slogs(spa))
1464 for (c = 0; c < rvd->vdev_children; c++) {
1465 vdev_t *tvd = rvd->vdev_child[c];
1466 metaslab_group_t *mg = tvd->vdev_mg;
1468 if (tvd->vdev_islog) {
1469 metaslab_group_passivate(mg);
1470 slog_found = B_TRUE;
1474 return (slog_found);
1478 spa_activate_log(spa_t *spa)
1480 vdev_t *rvd = spa->spa_root_vdev;
1483 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1485 for (c = 0; c < rvd->vdev_children; c++) {
1486 vdev_t *tvd = rvd->vdev_child[c];
1487 metaslab_group_t *mg = tvd->vdev_mg;
1489 if (tvd->vdev_islog)
1490 metaslab_group_activate(mg);
1495 spa_offline_log(spa_t *spa)
1499 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1500 NULL, DS_FIND_CHILDREN)) == 0) {
1503 * We successfully offlined the log device, sync out the
1504 * current txg so that the "stubby" block can be removed
1507 txg_wait_synced(spa->spa_dsl_pool, 0);
1513 spa_aux_check_removed(spa_aux_vdev_t *sav)
1517 for (i = 0; i < sav->sav_count; i++)
1518 spa_check_removed(sav->sav_vdevs[i]);
1522 spa_claim_notify(zio_t *zio)
1524 spa_t *spa = zio->io_spa;
1529 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1530 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1531 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1532 mutex_exit(&spa->spa_props_lock);
1535 typedef struct spa_load_error {
1536 uint64_t sle_meta_count;
1537 uint64_t sle_data_count;
1541 spa_load_verify_done(zio_t *zio)
1543 blkptr_t *bp = zio->io_bp;
1544 spa_load_error_t *sle = zio->io_private;
1545 dmu_object_type_t type = BP_GET_TYPE(bp);
1546 int error = zio->io_error;
1549 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1550 type != DMU_OT_INTENT_LOG)
1551 atomic_add_64(&sle->sle_meta_count, 1);
1553 atomic_add_64(&sle->sle_data_count, 1);
1555 zio_data_buf_free(zio->io_data, zio->io_size);
1560 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1561 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1565 size_t size = BP_GET_PSIZE(bp);
1566 void *data = zio_data_buf_alloc(size);
1568 zio_nowait(zio_read(rio, spa, bp, data, size,
1569 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1570 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1571 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1577 spa_load_verify(spa_t *spa)
1580 spa_load_error_t sle = { 0 };
1581 zpool_rewind_policy_t policy;
1582 boolean_t verify_ok = B_FALSE;
1585 zpool_get_rewind_policy(spa->spa_config, &policy);
1587 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1590 rio = zio_root(spa, NULL, &sle,
1591 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1593 error = traverse_pool(spa, spa->spa_verify_min_txg,
1594 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1596 (void) zio_wait(rio);
1598 spa->spa_load_meta_errors = sle.sle_meta_count;
1599 spa->spa_load_data_errors = sle.sle_data_count;
1601 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1602 sle.sle_data_count <= policy.zrp_maxdata) {
1606 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1607 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1609 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1610 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1611 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1612 VERIFY(nvlist_add_int64(spa->spa_load_info,
1613 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1614 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1615 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1617 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1621 if (error != ENXIO && error != EIO)
1626 return (verify_ok ? 0 : EIO);
1630 * Find a value in the pool props object.
1633 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1635 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1636 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1640 * Find a value in the pool directory object.
1643 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1645 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1646 name, sizeof (uint64_t), 1, val));
1650 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1652 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1657 * Fix up config after a partly-completed split. This is done with the
1658 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1659 * pool have that entry in their config, but only the splitting one contains
1660 * a list of all the guids of the vdevs that are being split off.
1662 * This function determines what to do with that list: either rejoin
1663 * all the disks to the pool, or complete the splitting process. To attempt
1664 * the rejoin, each disk that is offlined is marked online again, and
1665 * we do a reopen() call. If the vdev label for every disk that was
1666 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1667 * then we call vdev_split() on each disk, and complete the split.
1669 * Otherwise we leave the config alone, with all the vdevs in place in
1670 * the original pool.
1673 spa_try_repair(spa_t *spa, nvlist_t *config)
1680 boolean_t attempt_reopen;
1682 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1685 /* check that the config is complete */
1686 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1687 &glist, &gcount) != 0)
1690 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1692 /* attempt to online all the vdevs & validate */
1693 attempt_reopen = B_TRUE;
1694 for (i = 0; i < gcount; i++) {
1695 if (glist[i] == 0) /* vdev is hole */
1698 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1699 if (vd[i] == NULL) {
1701 * Don't bother attempting to reopen the disks;
1702 * just do the split.
1704 attempt_reopen = B_FALSE;
1706 /* attempt to re-online it */
1707 vd[i]->vdev_offline = B_FALSE;
1711 if (attempt_reopen) {
1712 vdev_reopen(spa->spa_root_vdev);
1714 /* check each device to see what state it's in */
1715 for (extracted = 0, i = 0; i < gcount; i++) {
1716 if (vd[i] != NULL &&
1717 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1724 * If every disk has been moved to the new pool, or if we never
1725 * even attempted to look at them, then we split them off for
1728 if (!attempt_reopen || gcount == extracted) {
1729 for (i = 0; i < gcount; i++)
1732 vdev_reopen(spa->spa_root_vdev);
1735 kmem_free(vd, gcount * sizeof (vdev_t *));
1739 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1740 boolean_t mosconfig)
1742 nvlist_t *config = spa->spa_config;
1743 char *ereport = FM_EREPORT_ZFS_POOL;
1748 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1752 * Versioning wasn't explicitly added to the label until later, so if
1753 * it's not present treat it as the initial version.
1755 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1756 &spa->spa_ubsync.ub_version) != 0)
1757 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1759 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1760 &spa->spa_config_txg);
1762 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1763 spa_guid_exists(pool_guid, 0)) {
1766 spa->spa_load_guid = pool_guid;
1768 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1770 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1774 gethrestime(&spa->spa_loaded_ts);
1775 error = spa_load_impl(spa, pool_guid, config, state, type,
1776 mosconfig, &ereport);
1779 spa->spa_minref = refcount_count(&spa->spa_refcount);
1781 if (error != EEXIST) {
1782 spa->spa_loaded_ts.tv_sec = 0;
1783 spa->spa_loaded_ts.tv_nsec = 0;
1785 if (error != EBADF) {
1786 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1789 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1796 * Load an existing storage pool, using the pool's builtin spa_config as a
1797 * source of configuration information.
1799 __attribute__((always_inline))
1801 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1802 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1806 nvlist_t *nvroot = NULL;
1808 uberblock_t *ub = &spa->spa_uberblock;
1809 uint64_t children, config_cache_txg = spa->spa_config_txg;
1810 int orig_mode = spa->spa_mode;
1815 * If this is an untrusted config, access the pool in read-only mode.
1816 * This prevents things like resilvering recently removed devices.
1819 spa->spa_mode = FREAD;
1821 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1823 spa->spa_load_state = state;
1825 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1828 parse = (type == SPA_IMPORT_EXISTING ?
1829 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1832 * Create "The Godfather" zio to hold all async IOs
1834 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1835 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1838 * Parse the configuration into a vdev tree. We explicitly set the
1839 * value that will be returned by spa_version() since parsing the
1840 * configuration requires knowing the version number.
1842 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1843 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1844 spa_config_exit(spa, SCL_ALL, FTAG);
1849 ASSERT(spa->spa_root_vdev == rvd);
1851 if (type != SPA_IMPORT_ASSEMBLE) {
1852 ASSERT(spa_guid(spa) == pool_guid);
1856 * Try to open all vdevs, loading each label in the process.
1858 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1859 error = vdev_open(rvd);
1860 spa_config_exit(spa, SCL_ALL, FTAG);
1865 * We need to validate the vdev labels against the configuration that
1866 * we have in hand, which is dependent on the setting of mosconfig. If
1867 * mosconfig is true then we're validating the vdev labels based on
1868 * that config. Otherwise, we're validating against the cached config
1869 * (zpool.cache) that was read when we loaded the zfs module, and then
1870 * later we will recursively call spa_load() and validate against
1873 * If we're assembling a new pool that's been split off from an
1874 * existing pool, the labels haven't yet been updated so we skip
1875 * validation for now.
1877 if (type != SPA_IMPORT_ASSEMBLE) {
1878 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1879 error = vdev_validate(rvd);
1880 spa_config_exit(spa, SCL_ALL, FTAG);
1885 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1890 * Find the best uberblock.
1892 vdev_uberblock_load(NULL, rvd, ub);
1895 * If we weren't able to find a single valid uberblock, return failure.
1897 if (ub->ub_txg == 0)
1898 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1901 * If the pool is newer than the code, we can't open it.
1903 if (ub->ub_version > SPA_VERSION)
1904 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1907 * If the vdev guid sum doesn't match the uberblock, we have an
1908 * incomplete configuration. We first check to see if the pool
1909 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1910 * If it is, defer the vdev_guid_sum check till later so we
1911 * can handle missing vdevs.
1913 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1914 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1915 rvd->vdev_guid_sum != ub->ub_guid_sum)
1916 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1918 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1919 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1920 spa_try_repair(spa, config);
1921 spa_config_exit(spa, SCL_ALL, FTAG);
1922 nvlist_free(spa->spa_config_splitting);
1923 spa->spa_config_splitting = NULL;
1927 * Initialize internal SPA structures.
1929 spa->spa_state = POOL_STATE_ACTIVE;
1930 spa->spa_ubsync = spa->spa_uberblock;
1931 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1932 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1933 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1934 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1935 spa->spa_claim_max_txg = spa->spa_first_txg;
1936 spa->spa_prev_software_version = ub->ub_software_version;
1938 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1940 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1941 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1943 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1944 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1948 nvlist_t *policy = NULL, *nvconfig;
1950 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1951 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1953 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1954 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1956 unsigned long myhostid = 0;
1958 VERIFY(nvlist_lookup_string(nvconfig,
1959 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1962 myhostid = zone_get_hostid(NULL);
1965 * We're emulating the system's hostid in userland, so
1966 * we can't use zone_get_hostid().
1968 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1969 #endif /* _KERNEL */
1970 if (hostid != 0 && myhostid != 0 &&
1971 hostid != myhostid) {
1972 nvlist_free(nvconfig);
1973 cmn_err(CE_WARN, "pool '%s' could not be "
1974 "loaded as it was last accessed by "
1975 "another system (host: %s hostid: 0x%lx). "
1976 "See: http://www.sun.com/msg/ZFS-8000-EY",
1977 spa_name(spa), hostname,
1978 (unsigned long)hostid);
1982 if (nvlist_lookup_nvlist(spa->spa_config,
1983 ZPOOL_REWIND_POLICY, &policy) == 0)
1984 VERIFY(nvlist_add_nvlist(nvconfig,
1985 ZPOOL_REWIND_POLICY, policy) == 0);
1987 spa_config_set(spa, nvconfig);
1989 spa_deactivate(spa);
1990 spa_activate(spa, orig_mode);
1992 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1995 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1996 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1997 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
1999 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2002 * Load the bit that tells us to use the new accounting function
2003 * (raid-z deflation). If we have an older pool, this will not
2006 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2007 if (error != 0 && error != ENOENT)
2008 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2010 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2011 &spa->spa_creation_version);
2012 if (error != 0 && error != ENOENT)
2013 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2016 * Load the persistent error log. If we have an older pool, this will
2019 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2020 if (error != 0 && error != ENOENT)
2021 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2023 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2024 &spa->spa_errlog_scrub);
2025 if (error != 0 && error != ENOENT)
2026 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2029 * Load the history object. If we have an older pool, this
2030 * will not be present.
2032 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2033 if (error != 0 && error != ENOENT)
2034 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2037 * If we're assembling the pool from the split-off vdevs of
2038 * an existing pool, we don't want to attach the spares & cache
2043 * Load any hot spares for this pool.
2045 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2046 if (error != 0 && error != ENOENT)
2047 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2048 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2049 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2050 if (load_nvlist(spa, spa->spa_spares.sav_object,
2051 &spa->spa_spares.sav_config) != 0)
2052 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2054 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2055 spa_load_spares(spa);
2056 spa_config_exit(spa, SCL_ALL, FTAG);
2057 } else if (error == 0) {
2058 spa->spa_spares.sav_sync = B_TRUE;
2062 * Load any level 2 ARC devices for this pool.
2064 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2065 &spa->spa_l2cache.sav_object);
2066 if (error != 0 && error != ENOENT)
2067 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2068 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2069 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2070 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2071 &spa->spa_l2cache.sav_config) != 0)
2072 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2074 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2075 spa_load_l2cache(spa);
2076 spa_config_exit(spa, SCL_ALL, FTAG);
2077 } else if (error == 0) {
2078 spa->spa_l2cache.sav_sync = B_TRUE;
2081 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2083 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2084 if (error && error != ENOENT)
2085 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2088 uint64_t autoreplace;
2090 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2091 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2092 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2093 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2094 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2095 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2096 &spa->spa_dedup_ditto);
2098 spa->spa_autoreplace = (autoreplace != 0);
2102 * If the 'autoreplace' property is set, then post a resource notifying
2103 * the ZFS DE that it should not issue any faults for unopenable
2104 * devices. We also iterate over the vdevs, and post a sysevent for any
2105 * unopenable vdevs so that the normal autoreplace handler can take
2108 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2109 spa_check_removed(spa->spa_root_vdev);
2111 * For the import case, this is done in spa_import(), because
2112 * at this point we're using the spare definitions from
2113 * the MOS config, not necessarily from the userland config.
2115 if (state != SPA_LOAD_IMPORT) {
2116 spa_aux_check_removed(&spa->spa_spares);
2117 spa_aux_check_removed(&spa->spa_l2cache);
2122 * Load the vdev state for all toplevel vdevs.
2127 * Propagate the leaf DTLs we just loaded all the way up the tree.
2129 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2130 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2131 spa_config_exit(spa, SCL_ALL, FTAG);
2134 * Load the DDTs (dedup tables).
2136 error = ddt_load(spa);
2138 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2140 spa_update_dspace(spa);
2143 * Validate the config, using the MOS config to fill in any
2144 * information which might be missing. If we fail to validate
2145 * the config then declare the pool unfit for use. If we're
2146 * assembling a pool from a split, the log is not transferred
2149 if (type != SPA_IMPORT_ASSEMBLE) {
2152 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2153 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2155 if (!spa_config_valid(spa, nvconfig)) {
2156 nvlist_free(nvconfig);
2157 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2160 nvlist_free(nvconfig);
2163 * Now that we've validate the config, check the state of the
2164 * root vdev. If it can't be opened, it indicates one or
2165 * more toplevel vdevs are faulted.
2167 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2170 if (spa_check_logs(spa)) {
2171 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2172 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2177 * We've successfully opened the pool, verify that we're ready
2178 * to start pushing transactions.
2180 if (state != SPA_LOAD_TRYIMPORT) {
2181 if ((error = spa_load_verify(spa)))
2182 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2186 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2187 spa->spa_load_max_txg == UINT64_MAX)) {
2189 int need_update = B_FALSE;
2192 ASSERT(state != SPA_LOAD_TRYIMPORT);
2195 * Claim log blocks that haven't been committed yet.
2196 * This must all happen in a single txg.
2197 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2198 * invoked from zil_claim_log_block()'s i/o done callback.
2199 * Price of rollback is that we abandon the log.
2201 spa->spa_claiming = B_TRUE;
2203 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2204 spa_first_txg(spa));
2205 (void) dmu_objset_find(spa_name(spa),
2206 zil_claim, tx, DS_FIND_CHILDREN);
2209 spa->spa_claiming = B_FALSE;
2211 spa_set_log_state(spa, SPA_LOG_GOOD);
2212 spa->spa_sync_on = B_TRUE;
2213 txg_sync_start(spa->spa_dsl_pool);
2216 * Wait for all claims to sync. We sync up to the highest
2217 * claimed log block birth time so that claimed log blocks
2218 * don't appear to be from the future. spa_claim_max_txg
2219 * will have been set for us by either zil_check_log_chain()
2220 * (invoked from spa_check_logs()) or zil_claim() above.
2222 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2225 * If the config cache is stale, or we have uninitialized
2226 * metaslabs (see spa_vdev_add()), then update the config.
2228 * If this is a verbatim import, trust the current
2229 * in-core spa_config and update the disk labels.
2231 if (config_cache_txg != spa->spa_config_txg ||
2232 state == SPA_LOAD_IMPORT ||
2233 state == SPA_LOAD_RECOVER ||
2234 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2235 need_update = B_TRUE;
2237 for (c = 0; c < rvd->vdev_children; c++)
2238 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2239 need_update = B_TRUE;
2242 * Update the config cache asychronously in case we're the
2243 * root pool, in which case the config cache isn't writable yet.
2246 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2249 * Check all DTLs to see if anything needs resilvering.
2251 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2252 vdev_resilver_needed(rvd, NULL, NULL))
2253 spa_async_request(spa, SPA_ASYNC_RESILVER);
2256 * Delete any inconsistent datasets.
2258 (void) dmu_objset_find(spa_name(spa),
2259 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2262 * Clean up any stale temporary dataset userrefs.
2264 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2271 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2273 int mode = spa->spa_mode;
2276 spa_deactivate(spa);
2278 spa->spa_load_max_txg--;
2280 spa_activate(spa, mode);
2281 spa_async_suspend(spa);
2283 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2287 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2288 uint64_t max_request, int rewind_flags)
2290 nvlist_t *config = NULL;
2291 int load_error, rewind_error;
2292 uint64_t safe_rewind_txg;
2295 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2296 spa->spa_load_max_txg = spa->spa_load_txg;
2297 spa_set_log_state(spa, SPA_LOG_CLEAR);
2299 spa->spa_load_max_txg = max_request;
2302 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2304 if (load_error == 0)
2307 if (spa->spa_root_vdev != NULL)
2308 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2310 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2311 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2313 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2314 nvlist_free(config);
2315 return (load_error);
2318 /* Price of rolling back is discarding txgs, including log */
2319 if (state == SPA_LOAD_RECOVER)
2320 spa_set_log_state(spa, SPA_LOG_CLEAR);
2322 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2323 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2324 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2325 TXG_INITIAL : safe_rewind_txg;
2328 * Continue as long as we're finding errors, we're still within
2329 * the acceptable rewind range, and we're still finding uberblocks
2331 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2332 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2333 if (spa->spa_load_max_txg < safe_rewind_txg)
2334 spa->spa_extreme_rewind = B_TRUE;
2335 rewind_error = spa_load_retry(spa, state, mosconfig);
2338 spa->spa_extreme_rewind = B_FALSE;
2339 spa->spa_load_max_txg = UINT64_MAX;
2341 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2342 spa_config_set(spa, config);
2344 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2350 * The import case is identical to an open except that the configuration is sent
2351 * down from userland, instead of grabbed from the configuration cache. For the
2352 * case of an open, the pool configuration will exist in the
2353 * POOL_STATE_UNINITIALIZED state.
2355 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2356 * the same time open the pool, without having to keep around the spa_t in some
2360 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2364 spa_load_state_t state = SPA_LOAD_OPEN;
2366 int locked = B_FALSE;
2371 * As disgusting as this is, we need to support recursive calls to this
2372 * function because dsl_dir_open() is called during spa_load(), and ends
2373 * up calling spa_open() again. The real fix is to figure out how to
2374 * avoid dsl_dir_open() calling this in the first place.
2376 if (mutex_owner(&spa_namespace_lock) != curthread) {
2377 mutex_enter(&spa_namespace_lock);
2381 if ((spa = spa_lookup(pool)) == NULL) {
2383 mutex_exit(&spa_namespace_lock);
2387 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2388 zpool_rewind_policy_t policy;
2390 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2392 if (policy.zrp_request & ZPOOL_DO_REWIND)
2393 state = SPA_LOAD_RECOVER;
2395 spa_activate(spa, spa_mode_global);
2397 if (state != SPA_LOAD_RECOVER)
2398 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2400 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2401 policy.zrp_request);
2403 if (error == EBADF) {
2405 * If vdev_validate() returns failure (indicated by
2406 * EBADF), it indicates that one of the vdevs indicates
2407 * that the pool has been exported or destroyed. If
2408 * this is the case, the config cache is out of sync and
2409 * we should remove the pool from the namespace.
2412 spa_deactivate(spa);
2413 spa_config_sync(spa, B_TRUE, B_TRUE);
2416 mutex_exit(&spa_namespace_lock);
2422 * We can't open the pool, but we still have useful
2423 * information: the state of each vdev after the
2424 * attempted vdev_open(). Return this to the user.
2426 if (config != NULL && spa->spa_config) {
2427 VERIFY(nvlist_dup(spa->spa_config, config,
2429 VERIFY(nvlist_add_nvlist(*config,
2430 ZPOOL_CONFIG_LOAD_INFO,
2431 spa->spa_load_info) == 0);
2434 spa_deactivate(spa);
2435 spa->spa_last_open_failed = error;
2437 mutex_exit(&spa_namespace_lock);
2443 spa_open_ref(spa, tag);
2446 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2449 * If we've recovered the pool, pass back any information we
2450 * gathered while doing the load.
2452 if (state == SPA_LOAD_RECOVER) {
2453 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2454 spa->spa_load_info) == 0);
2458 spa->spa_last_open_failed = 0;
2459 spa->spa_last_ubsync_txg = 0;
2460 spa->spa_load_txg = 0;
2461 mutex_exit(&spa_namespace_lock);
2470 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2473 return (spa_open_common(name, spapp, tag, policy, config));
2477 spa_open(const char *name, spa_t **spapp, void *tag)
2479 return (spa_open_common(name, spapp, tag, NULL, NULL));
2483 * Lookup the given spa_t, incrementing the inject count in the process,
2484 * preventing it from being exported or destroyed.
2487 spa_inject_addref(char *name)
2491 mutex_enter(&spa_namespace_lock);
2492 if ((spa = spa_lookup(name)) == NULL) {
2493 mutex_exit(&spa_namespace_lock);
2496 spa->spa_inject_ref++;
2497 mutex_exit(&spa_namespace_lock);
2503 spa_inject_delref(spa_t *spa)
2505 mutex_enter(&spa_namespace_lock);
2506 spa->spa_inject_ref--;
2507 mutex_exit(&spa_namespace_lock);
2511 * Add spares device information to the nvlist.
2514 spa_add_spares(spa_t *spa, nvlist_t *config)
2524 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2526 if (spa->spa_spares.sav_count == 0)
2529 VERIFY(nvlist_lookup_nvlist(config,
2530 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2531 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2532 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2534 VERIFY(nvlist_add_nvlist_array(nvroot,
2535 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2536 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2537 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2540 * Go through and find any spares which have since been
2541 * repurposed as an active spare. If this is the case, update
2542 * their status appropriately.
2544 for (i = 0; i < nspares; i++) {
2545 VERIFY(nvlist_lookup_uint64(spares[i],
2546 ZPOOL_CONFIG_GUID, &guid) == 0);
2547 if (spa_spare_exists(guid, &pool, NULL) &&
2549 VERIFY(nvlist_lookup_uint64_array(
2550 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2551 (uint64_t **)&vs, &vsc) == 0);
2552 vs->vs_state = VDEV_STATE_CANT_OPEN;
2553 vs->vs_aux = VDEV_AUX_SPARED;
2560 * Add l2cache device information to the nvlist, including vdev stats.
2563 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2566 uint_t i, j, nl2cache;
2573 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2575 if (spa->spa_l2cache.sav_count == 0)
2578 VERIFY(nvlist_lookup_nvlist(config,
2579 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2580 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2581 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2582 if (nl2cache != 0) {
2583 VERIFY(nvlist_add_nvlist_array(nvroot,
2584 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2585 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2586 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2589 * Update level 2 cache device stats.
2592 for (i = 0; i < nl2cache; i++) {
2593 VERIFY(nvlist_lookup_uint64(l2cache[i],
2594 ZPOOL_CONFIG_GUID, &guid) == 0);
2597 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2599 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2600 vd = spa->spa_l2cache.sav_vdevs[j];
2606 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2607 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2609 vdev_get_stats(vd, vs);
2615 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2621 error = spa_open_common(name, &spa, FTAG, NULL, config);
2625 * This still leaves a window of inconsistency where the spares
2626 * or l2cache devices could change and the config would be
2627 * self-inconsistent.
2629 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2631 if (*config != NULL) {
2632 uint64_t loadtimes[2];
2634 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2635 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2636 VERIFY(nvlist_add_uint64_array(*config,
2637 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2639 VERIFY(nvlist_add_uint64(*config,
2640 ZPOOL_CONFIG_ERRCOUNT,
2641 spa_get_errlog_size(spa)) == 0);
2643 if (spa_suspended(spa))
2644 VERIFY(nvlist_add_uint64(*config,
2645 ZPOOL_CONFIG_SUSPENDED,
2646 spa->spa_failmode) == 0);
2648 spa_add_spares(spa, *config);
2649 spa_add_l2cache(spa, *config);
2654 * We want to get the alternate root even for faulted pools, so we cheat
2655 * and call spa_lookup() directly.
2659 mutex_enter(&spa_namespace_lock);
2660 spa = spa_lookup(name);
2662 spa_altroot(spa, altroot, buflen);
2666 mutex_exit(&spa_namespace_lock);
2668 spa_altroot(spa, altroot, buflen);
2673 spa_config_exit(spa, SCL_CONFIG, FTAG);
2674 spa_close(spa, FTAG);
2681 * Validate that the auxiliary device array is well formed. We must have an
2682 * array of nvlists, each which describes a valid leaf vdev. If this is an
2683 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2684 * specified, as long as they are well-formed.
2687 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2688 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2689 vdev_labeltype_t label)
2696 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2699 * It's acceptable to have no devs specified.
2701 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2708 * Make sure the pool is formatted with a version that supports this
2711 if (spa_version(spa) < version)
2715 * Set the pending device list so we correctly handle device in-use
2718 sav->sav_pending = dev;
2719 sav->sav_npending = ndev;
2721 for (i = 0; i < ndev; i++) {
2722 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2726 if (!vd->vdev_ops->vdev_op_leaf) {
2733 * The L2ARC currently only supports disk devices in
2734 * kernel context. For user-level testing, we allow it.
2737 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2738 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2745 if ((error = vdev_open(vd)) == 0 &&
2746 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2747 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2748 vd->vdev_guid) == 0);
2754 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2761 sav->sav_pending = NULL;
2762 sav->sav_npending = 0;
2767 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2771 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2773 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2774 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2775 VDEV_LABEL_SPARE)) != 0) {
2779 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2780 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2781 VDEV_LABEL_L2CACHE));
2785 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2790 if (sav->sav_config != NULL) {
2796 * Generate new dev list by concatentating with the
2799 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2800 &olddevs, &oldndevs) == 0);
2802 newdevs = kmem_alloc(sizeof (void *) *
2803 (ndevs + oldndevs), KM_SLEEP);
2804 for (i = 0; i < oldndevs; i++)
2805 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2807 for (i = 0; i < ndevs; i++)
2808 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2811 VERIFY(nvlist_remove(sav->sav_config, config,
2812 DATA_TYPE_NVLIST_ARRAY) == 0);
2814 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2815 config, newdevs, ndevs + oldndevs) == 0);
2816 for (i = 0; i < oldndevs + ndevs; i++)
2817 nvlist_free(newdevs[i]);
2818 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2821 * Generate a new dev list.
2823 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2825 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2831 * Stop and drop level 2 ARC devices
2834 spa_l2cache_drop(spa_t *spa)
2838 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2840 for (i = 0; i < sav->sav_count; i++) {
2843 vd = sav->sav_vdevs[i];
2846 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2847 pool != 0ULL && l2arc_vdev_present(vd))
2848 l2arc_remove_vdev(vd);
2849 if (vd->vdev_isl2cache)
2850 spa_l2cache_remove(vd);
2851 vdev_clear_stats(vd);
2852 (void) vdev_close(vd);
2860 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2861 const char *history_str, nvlist_t *zplprops)
2864 char *altroot = NULL;
2869 uint64_t txg = TXG_INITIAL;
2870 nvlist_t **spares, **l2cache;
2871 uint_t nspares, nl2cache;
2872 uint64_t version, obj;
2876 * If this pool already exists, return failure.
2878 mutex_enter(&spa_namespace_lock);
2879 if (spa_lookup(pool) != NULL) {
2880 mutex_exit(&spa_namespace_lock);
2885 * Allocate a new spa_t structure.
2887 (void) nvlist_lookup_string(props,
2888 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2889 spa = spa_add(pool, NULL, altroot);
2890 spa_activate(spa, spa_mode_global);
2892 if (props && (error = spa_prop_validate(spa, props))) {
2893 spa_deactivate(spa);
2895 mutex_exit(&spa_namespace_lock);
2899 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2901 version = SPA_VERSION;
2902 ASSERT(version <= SPA_VERSION);
2904 spa->spa_first_txg = txg;
2905 spa->spa_uberblock.ub_txg = txg - 1;
2906 spa->spa_uberblock.ub_version = version;
2907 spa->spa_ubsync = spa->spa_uberblock;
2910 * Create "The Godfather" zio to hold all async IOs
2912 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2913 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2916 * Create the root vdev.
2918 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2920 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2922 ASSERT(error != 0 || rvd != NULL);
2923 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2925 if (error == 0 && !zfs_allocatable_devs(nvroot))
2929 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2930 (error = spa_validate_aux(spa, nvroot, txg,
2931 VDEV_ALLOC_ADD)) == 0) {
2932 for (c = 0; c < rvd->vdev_children; c++) {
2933 vdev_metaslab_set_size(rvd->vdev_child[c]);
2934 vdev_expand(rvd->vdev_child[c], txg);
2938 spa_config_exit(spa, SCL_ALL, FTAG);
2942 spa_deactivate(spa);
2944 mutex_exit(&spa_namespace_lock);
2949 * Get the list of spares, if specified.
2951 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2952 &spares, &nspares) == 0) {
2953 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2955 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2956 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2957 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2958 spa_load_spares(spa);
2959 spa_config_exit(spa, SCL_ALL, FTAG);
2960 spa->spa_spares.sav_sync = B_TRUE;
2964 * Get the list of level 2 cache devices, if specified.
2966 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2967 &l2cache, &nl2cache) == 0) {
2968 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2969 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2970 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2971 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2972 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2973 spa_load_l2cache(spa);
2974 spa_config_exit(spa, SCL_ALL, FTAG);
2975 spa->spa_l2cache.sav_sync = B_TRUE;
2978 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2979 spa->spa_meta_objset = dp->dp_meta_objset;
2982 * Create DDTs (dedup tables).
2986 spa_update_dspace(spa);
2988 tx = dmu_tx_create_assigned(dp, txg);
2991 * Create the pool config object.
2993 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2994 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2995 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
2997 if (zap_add(spa->spa_meta_objset,
2998 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
2999 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3000 cmn_err(CE_PANIC, "failed to add pool config");
3003 if (zap_add(spa->spa_meta_objset,
3004 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3005 sizeof (uint64_t), 1, &version, tx) != 0) {
3006 cmn_err(CE_PANIC, "failed to add pool version");
3009 /* Newly created pools with the right version are always deflated. */
3010 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3011 spa->spa_deflate = TRUE;
3012 if (zap_add(spa->spa_meta_objset,
3013 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3014 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3015 cmn_err(CE_PANIC, "failed to add deflate");
3020 * Create the deferred-free bpobj. Turn off compression
3021 * because sync-to-convergence takes longer if the blocksize
3024 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3025 dmu_object_set_compress(spa->spa_meta_objset, obj,
3026 ZIO_COMPRESS_OFF, tx);
3027 if (zap_add(spa->spa_meta_objset,
3028 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3029 sizeof (uint64_t), 1, &obj, tx) != 0) {
3030 cmn_err(CE_PANIC, "failed to add bpobj");
3032 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3033 spa->spa_meta_objset, obj));
3036 * Create the pool's history object.
3038 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3039 spa_history_create_obj(spa, tx);
3042 * Set pool properties.
3044 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3045 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3046 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3047 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3049 if (props != NULL) {
3050 spa_configfile_set(spa, props, B_FALSE);
3051 spa_sync_props(spa, props, tx);
3056 spa->spa_sync_on = B_TRUE;
3057 txg_sync_start(spa->spa_dsl_pool);
3060 * We explicitly wait for the first transaction to complete so that our
3061 * bean counters are appropriately updated.
3063 txg_wait_synced(spa->spa_dsl_pool, txg);
3065 spa_config_sync(spa, B_FALSE, B_TRUE);
3067 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3068 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3069 spa_history_log_version(spa, LOG_POOL_CREATE);
3071 spa->spa_minref = refcount_count(&spa->spa_refcount);
3073 mutex_exit(&spa_namespace_lock);
3080 * Get the root pool information from the root disk, then import the root pool
3081 * during the system boot up time.
3083 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3086 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3089 nvlist_t *nvtop, *nvroot;
3092 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3096 * Add this top-level vdev to the child array.
3098 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3100 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3102 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3105 * Put this pool's top-level vdevs into a root vdev.
3107 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3108 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3109 VDEV_TYPE_ROOT) == 0);
3110 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3111 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3112 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3116 * Replace the existing vdev_tree with the new root vdev in
3117 * this pool's configuration (remove the old, add the new).
3119 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3120 nvlist_free(nvroot);
3125 * Walk the vdev tree and see if we can find a device with "better"
3126 * configuration. A configuration is "better" if the label on that
3127 * device has a more recent txg.
3130 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3134 for (c = 0; c < vd->vdev_children; c++)
3135 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3137 if (vd->vdev_ops->vdev_op_leaf) {
3141 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3145 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3149 * Do we have a better boot device?
3151 if (label_txg > *txg) {
3160 * Import a root pool.
3162 * For x86. devpath_list will consist of devid and/or physpath name of
3163 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3164 * The GRUB "findroot" command will return the vdev we should boot.
3166 * For Sparc, devpath_list consists the physpath name of the booting device
3167 * no matter the rootpool is a single device pool or a mirrored pool.
3169 * "/pci@1f,0/ide@d/disk@0,0:a"
3172 spa_import_rootpool(char *devpath, char *devid)
3175 vdev_t *rvd, *bvd, *avd = NULL;
3176 nvlist_t *config, *nvtop;
3182 * Read the label from the boot device and generate a configuration.
3184 config = spa_generate_rootconf(devpath, devid, &guid);
3185 #if defined(_OBP) && defined(_KERNEL)
3186 if (config == NULL) {
3187 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3189 get_iscsi_bootpath_phy(devpath);
3190 config = spa_generate_rootconf(devpath, devid, &guid);
3194 if (config == NULL) {
3195 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3200 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3202 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3204 mutex_enter(&spa_namespace_lock);
3205 if ((spa = spa_lookup(pname)) != NULL) {
3207 * Remove the existing root pool from the namespace so that we
3208 * can replace it with the correct config we just read in.
3213 spa = spa_add(pname, config, NULL);
3214 spa->spa_is_root = B_TRUE;
3215 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3218 * Build up a vdev tree based on the boot device's label config.
3220 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3222 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3223 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3224 VDEV_ALLOC_ROOTPOOL);
3225 spa_config_exit(spa, SCL_ALL, FTAG);
3227 mutex_exit(&spa_namespace_lock);
3228 nvlist_free(config);
3229 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3235 * Get the boot vdev.
3237 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3238 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3239 (u_longlong_t)guid);
3245 * Determine if there is a better boot device.
3248 spa_alt_rootvdev(rvd, &avd, &txg);
3250 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3251 "try booting from '%s'", avd->vdev_path);
3257 * If the boot device is part of a spare vdev then ensure that
3258 * we're booting off the active spare.
3260 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3261 !bvd->vdev_isspare) {
3262 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3263 "try booting from '%s'",
3265 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3271 spa_history_log_version(spa, LOG_POOL_IMPORT);
3273 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3275 spa_config_exit(spa, SCL_ALL, FTAG);
3276 mutex_exit(&spa_namespace_lock);
3278 nvlist_free(config);
3285 * Import a non-root pool into the system.
3288 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3291 char *altroot = NULL;
3292 spa_load_state_t state = SPA_LOAD_IMPORT;
3293 zpool_rewind_policy_t policy;
3294 uint64_t mode = spa_mode_global;
3295 uint64_t readonly = B_FALSE;
3298 nvlist_t **spares, **l2cache;
3299 uint_t nspares, nl2cache;
3302 * If a pool with this name exists, return failure.
3304 mutex_enter(&spa_namespace_lock);
3305 if (spa_lookup(pool) != NULL) {
3306 mutex_exit(&spa_namespace_lock);
3311 * Create and initialize the spa structure.
3313 (void) nvlist_lookup_string(props,
3314 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3315 (void) nvlist_lookup_uint64(props,
3316 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3319 spa = spa_add(pool, config, altroot);
3320 spa->spa_import_flags = flags;
3323 * Verbatim import - Take a pool and insert it into the namespace
3324 * as if it had been loaded at boot.
3326 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3328 spa_configfile_set(spa, props, B_FALSE);
3330 spa_config_sync(spa, B_FALSE, B_TRUE);
3332 mutex_exit(&spa_namespace_lock);
3333 spa_history_log_version(spa, LOG_POOL_IMPORT);
3338 spa_activate(spa, mode);
3341 * Don't start async tasks until we know everything is healthy.
3343 spa_async_suspend(spa);
3345 zpool_get_rewind_policy(config, &policy);
3346 if (policy.zrp_request & ZPOOL_DO_REWIND)
3347 state = SPA_LOAD_RECOVER;
3350 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3351 * because the user-supplied config is actually the one to trust when
3354 if (state != SPA_LOAD_RECOVER)
3355 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3357 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3358 policy.zrp_request);
3361 * Propagate anything learned while loading the pool and pass it
3362 * back to caller (i.e. rewind info, missing devices, etc).
3364 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3365 spa->spa_load_info) == 0);
3367 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3369 * Toss any existing sparelist, as it doesn't have any validity
3370 * anymore, and conflicts with spa_has_spare().
3372 if (spa->spa_spares.sav_config) {
3373 nvlist_free(spa->spa_spares.sav_config);
3374 spa->spa_spares.sav_config = NULL;
3375 spa_load_spares(spa);
3377 if (spa->spa_l2cache.sav_config) {
3378 nvlist_free(spa->spa_l2cache.sav_config);
3379 spa->spa_l2cache.sav_config = NULL;
3380 spa_load_l2cache(spa);
3383 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3386 error = spa_validate_aux(spa, nvroot, -1ULL,
3389 error = spa_validate_aux(spa, nvroot, -1ULL,
3390 VDEV_ALLOC_L2CACHE);
3391 spa_config_exit(spa, SCL_ALL, FTAG);
3394 spa_configfile_set(spa, props, B_FALSE);
3396 if (error != 0 || (props && spa_writeable(spa) &&
3397 (error = spa_prop_set(spa, props)))) {
3399 spa_deactivate(spa);
3401 mutex_exit(&spa_namespace_lock);
3405 spa_async_resume(spa);
3408 * Override any spares and level 2 cache devices as specified by
3409 * the user, as these may have correct device names/devids, etc.
3411 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3412 &spares, &nspares) == 0) {
3413 if (spa->spa_spares.sav_config)
3414 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3415 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3417 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3418 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3419 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3420 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3421 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3422 spa_load_spares(spa);
3423 spa_config_exit(spa, SCL_ALL, FTAG);
3424 spa->spa_spares.sav_sync = B_TRUE;
3426 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3427 &l2cache, &nl2cache) == 0) {
3428 if (spa->spa_l2cache.sav_config)
3429 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3430 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3432 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3433 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3434 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3435 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3436 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3437 spa_load_l2cache(spa);
3438 spa_config_exit(spa, SCL_ALL, FTAG);
3439 spa->spa_l2cache.sav_sync = B_TRUE;
3443 * Check for any removed devices.
3445 if (spa->spa_autoreplace) {
3446 spa_aux_check_removed(&spa->spa_spares);
3447 spa_aux_check_removed(&spa->spa_l2cache);
3450 if (spa_writeable(spa)) {
3452 * Update the config cache to include the newly-imported pool.
3454 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3458 * It's possible that the pool was expanded while it was exported.
3459 * We kick off an async task to handle this for us.
3461 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3463 mutex_exit(&spa_namespace_lock);
3464 spa_history_log_version(spa, LOG_POOL_IMPORT);
3470 spa_tryimport(nvlist_t *tryconfig)
3472 nvlist_t *config = NULL;
3478 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3481 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3485 * Create and initialize the spa structure.
3487 mutex_enter(&spa_namespace_lock);
3488 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3489 spa_activate(spa, FREAD);
3492 * Pass off the heavy lifting to spa_load().
3493 * Pass TRUE for mosconfig because the user-supplied config
3494 * is actually the one to trust when doing an import.
3496 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3499 * If 'tryconfig' was at least parsable, return the current config.
3501 if (spa->spa_root_vdev != NULL) {
3502 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3503 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3505 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3507 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3508 spa->spa_uberblock.ub_timestamp) == 0);
3511 * If the bootfs property exists on this pool then we
3512 * copy it out so that external consumers can tell which
3513 * pools are bootable.
3515 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3516 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3519 * We have to play games with the name since the
3520 * pool was opened as TRYIMPORT_NAME.
3522 if (dsl_dsobj_to_dsname(spa_name(spa),
3523 spa->spa_bootfs, tmpname) == 0) {
3525 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3527 cp = strchr(tmpname, '/');
3529 (void) strlcpy(dsname, tmpname,
3532 (void) snprintf(dsname, MAXPATHLEN,
3533 "%s/%s", poolname, ++cp);
3535 VERIFY(nvlist_add_string(config,
3536 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3537 kmem_free(dsname, MAXPATHLEN);
3539 kmem_free(tmpname, MAXPATHLEN);
3543 * Add the list of hot spares and level 2 cache devices.
3545 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3546 spa_add_spares(spa, config);
3547 spa_add_l2cache(spa, config);
3548 spa_config_exit(spa, SCL_CONFIG, FTAG);
3552 spa_deactivate(spa);
3554 mutex_exit(&spa_namespace_lock);
3560 * Pool export/destroy
3562 * The act of destroying or exporting a pool is very simple. We make sure there
3563 * is no more pending I/O and any references to the pool are gone. Then, we
3564 * update the pool state and sync all the labels to disk, removing the
3565 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3566 * we don't sync the labels or remove the configuration cache.
3569 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3570 boolean_t force, boolean_t hardforce)
3577 if (!(spa_mode_global & FWRITE))
3580 mutex_enter(&spa_namespace_lock);
3581 if ((spa = spa_lookup(pool)) == NULL) {
3582 mutex_exit(&spa_namespace_lock);
3587 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3588 * reacquire the namespace lock, and see if we can export.
3590 spa_open_ref(spa, FTAG);
3591 mutex_exit(&spa_namespace_lock);
3592 spa_async_suspend(spa);
3593 mutex_enter(&spa_namespace_lock);
3594 spa_close(spa, FTAG);
3597 * The pool will be in core if it's openable,
3598 * in which case we can modify its state.
3600 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3602 * Objsets may be open only because they're dirty, so we
3603 * have to force it to sync before checking spa_refcnt.
3605 txg_wait_synced(spa->spa_dsl_pool, 0);
3608 * A pool cannot be exported or destroyed if there are active
3609 * references. If we are resetting a pool, allow references by
3610 * fault injection handlers.
3612 if (!spa_refcount_zero(spa) ||
3613 (spa->spa_inject_ref != 0 &&
3614 new_state != POOL_STATE_UNINITIALIZED)) {
3615 spa_async_resume(spa);
3616 mutex_exit(&spa_namespace_lock);
3621 * A pool cannot be exported if it has an active shared spare.
3622 * This is to prevent other pools stealing the active spare
3623 * from an exported pool. At user's own will, such pool can
3624 * be forcedly exported.
3626 if (!force && new_state == POOL_STATE_EXPORTED &&
3627 spa_has_active_shared_spare(spa)) {
3628 spa_async_resume(spa);
3629 mutex_exit(&spa_namespace_lock);
3634 * We want this to be reflected on every label,
3635 * so mark them all dirty. spa_unload() will do the
3636 * final sync that pushes these changes out.
3638 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3639 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3640 spa->spa_state = new_state;
3641 spa->spa_final_txg = spa_last_synced_txg(spa) +
3643 vdev_config_dirty(spa->spa_root_vdev);
3644 spa_config_exit(spa, SCL_ALL, FTAG);
3648 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
3650 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3652 spa_deactivate(spa);
3655 if (oldconfig && spa->spa_config)
3656 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3658 if (new_state != POOL_STATE_UNINITIALIZED) {
3660 spa_config_sync(spa, B_TRUE, B_TRUE);
3663 mutex_exit(&spa_namespace_lock);
3669 * Destroy a storage pool.
3672 spa_destroy(char *pool)
3674 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3679 * Export a storage pool.
3682 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3683 boolean_t hardforce)
3685 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3690 * Similar to spa_export(), this unloads the spa_t without actually removing it
3691 * from the namespace in any way.
3694 spa_reset(char *pool)
3696 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3701 * ==========================================================================
3702 * Device manipulation
3703 * ==========================================================================
3707 * Add a device to a storage pool.
3710 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3714 vdev_t *rvd = spa->spa_root_vdev;
3716 nvlist_t **spares, **l2cache;
3717 uint_t nspares, nl2cache;
3720 ASSERT(spa_writeable(spa));
3722 txg = spa_vdev_enter(spa);
3724 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3725 VDEV_ALLOC_ADD)) != 0)
3726 return (spa_vdev_exit(spa, NULL, txg, error));
3728 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3730 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3734 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3738 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3739 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3741 if (vd->vdev_children != 0 &&
3742 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3743 return (spa_vdev_exit(spa, vd, txg, error));
3746 * We must validate the spares and l2cache devices after checking the
3747 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3749 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3750 return (spa_vdev_exit(spa, vd, txg, error));
3753 * Transfer each new top-level vdev from vd to rvd.
3755 for (c = 0; c < vd->vdev_children; c++) {
3758 * Set the vdev id to the first hole, if one exists.
3760 for (id = 0; id < rvd->vdev_children; id++) {
3761 if (rvd->vdev_child[id]->vdev_ishole) {
3762 vdev_free(rvd->vdev_child[id]);
3766 tvd = vd->vdev_child[c];
3767 vdev_remove_child(vd, tvd);
3769 vdev_add_child(rvd, tvd);
3770 vdev_config_dirty(tvd);
3774 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3775 ZPOOL_CONFIG_SPARES);
3776 spa_load_spares(spa);
3777 spa->spa_spares.sav_sync = B_TRUE;
3780 if (nl2cache != 0) {
3781 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3782 ZPOOL_CONFIG_L2CACHE);
3783 spa_load_l2cache(spa);
3784 spa->spa_l2cache.sav_sync = B_TRUE;
3788 * We have to be careful when adding new vdevs to an existing pool.
3789 * If other threads start allocating from these vdevs before we
3790 * sync the config cache, and we lose power, then upon reboot we may
3791 * fail to open the pool because there are DVAs that the config cache
3792 * can't translate. Therefore, we first add the vdevs without
3793 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3794 * and then let spa_config_update() initialize the new metaslabs.
3796 * spa_load() checks for added-but-not-initialized vdevs, so that
3797 * if we lose power at any point in this sequence, the remaining
3798 * steps will be completed the next time we load the pool.
3800 (void) spa_vdev_exit(spa, vd, txg, 0);
3802 mutex_enter(&spa_namespace_lock);
3803 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3804 mutex_exit(&spa_namespace_lock);
3810 * Attach a device to a mirror. The arguments are the path to any device
3811 * in the mirror, and the nvroot for the new device. If the path specifies
3812 * a device that is not mirrored, we automatically insert the mirror vdev.
3814 * If 'replacing' is specified, the new device is intended to replace the
3815 * existing device; in this case the two devices are made into their own
3816 * mirror using the 'replacing' vdev, which is functionally identical to
3817 * the mirror vdev (it actually reuses all the same ops) but has a few
3818 * extra rules: you can't attach to it after it's been created, and upon
3819 * completion of resilvering, the first disk (the one being replaced)
3820 * is automatically detached.
3823 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3825 uint64_t txg, dtl_max_txg;
3826 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3827 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3829 char *oldvdpath, *newvdpath;
3833 ASSERT(spa_writeable(spa));
3835 txg = spa_vdev_enter(spa);
3837 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3840 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3842 if (!oldvd->vdev_ops->vdev_op_leaf)
3843 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3845 pvd = oldvd->vdev_parent;
3847 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3848 VDEV_ALLOC_ADD)) != 0)
3849 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3851 if (newrootvd->vdev_children != 1)
3852 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3854 newvd = newrootvd->vdev_child[0];
3856 if (!newvd->vdev_ops->vdev_op_leaf)
3857 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3859 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3860 return (spa_vdev_exit(spa, newrootvd, txg, error));
3863 * Spares can't replace logs
3865 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3866 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3870 * For attach, the only allowable parent is a mirror or the root
3873 if (pvd->vdev_ops != &vdev_mirror_ops &&
3874 pvd->vdev_ops != &vdev_root_ops)
3875 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3877 pvops = &vdev_mirror_ops;
3880 * Active hot spares can only be replaced by inactive hot
3883 if (pvd->vdev_ops == &vdev_spare_ops &&
3884 oldvd->vdev_isspare &&
3885 !spa_has_spare(spa, newvd->vdev_guid))
3886 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3889 * If the source is a hot spare, and the parent isn't already a
3890 * spare, then we want to create a new hot spare. Otherwise, we
3891 * want to create a replacing vdev. The user is not allowed to
3892 * attach to a spared vdev child unless the 'isspare' state is
3893 * the same (spare replaces spare, non-spare replaces
3896 if (pvd->vdev_ops == &vdev_replacing_ops &&
3897 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3898 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3899 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3900 newvd->vdev_isspare != oldvd->vdev_isspare) {
3901 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3904 if (newvd->vdev_isspare)
3905 pvops = &vdev_spare_ops;
3907 pvops = &vdev_replacing_ops;
3911 * Make sure the new device is big enough.
3913 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3914 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3917 * The new device cannot have a higher alignment requirement
3918 * than the top-level vdev.
3920 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3921 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3924 * If this is an in-place replacement, update oldvd's path and devid
3925 * to make it distinguishable from newvd, and unopenable from now on.
3927 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3928 spa_strfree(oldvd->vdev_path);
3929 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3931 (void) sprintf(oldvd->vdev_path, "%s/%s",
3932 newvd->vdev_path, "old");
3933 if (oldvd->vdev_devid != NULL) {
3934 spa_strfree(oldvd->vdev_devid);
3935 oldvd->vdev_devid = NULL;
3939 /* mark the device being resilvered */
3940 newvd->vdev_resilvering = B_TRUE;
3943 * If the parent is not a mirror, or if we're replacing, insert the new
3944 * mirror/replacing/spare vdev above oldvd.
3946 if (pvd->vdev_ops != pvops)
3947 pvd = vdev_add_parent(oldvd, pvops);
3949 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3950 ASSERT(pvd->vdev_ops == pvops);
3951 ASSERT(oldvd->vdev_parent == pvd);
3954 * Extract the new device from its root and add it to pvd.
3956 vdev_remove_child(newrootvd, newvd);
3957 newvd->vdev_id = pvd->vdev_children;
3958 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3959 vdev_add_child(pvd, newvd);
3961 tvd = newvd->vdev_top;
3962 ASSERT(pvd->vdev_top == tvd);
3963 ASSERT(tvd->vdev_parent == rvd);
3965 vdev_config_dirty(tvd);
3968 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3969 * for any dmu_sync-ed blocks. It will propagate upward when
3970 * spa_vdev_exit() calls vdev_dtl_reassess().
3972 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3974 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3975 dtl_max_txg - TXG_INITIAL);
3977 if (newvd->vdev_isspare) {
3978 spa_spare_activate(newvd);
3979 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
3982 oldvdpath = spa_strdup(oldvd->vdev_path);
3983 newvdpath = spa_strdup(newvd->vdev_path);
3984 newvd_isspare = newvd->vdev_isspare;
3987 * Mark newvd's DTL dirty in this txg.
3989 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3992 * Restart the resilver
3994 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
3999 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4001 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4002 "%s vdev=%s %s vdev=%s",
4003 replacing && newvd_isspare ? "spare in" :
4004 replacing ? "replace" : "attach", newvdpath,
4005 replacing ? "for" : "to", oldvdpath);
4007 spa_strfree(oldvdpath);
4008 spa_strfree(newvdpath);
4010 if (spa->spa_bootfs)
4011 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4017 * Detach a device from a mirror or replacing vdev.
4018 * If 'replace_done' is specified, only detach if the parent
4019 * is a replacing vdev.
4022 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4026 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4027 vdev_t *vd, *pvd, *cvd, *tvd;
4028 boolean_t unspare = B_FALSE;
4029 uint64_t unspare_guid = 0;
4033 ASSERT(spa_writeable(spa));
4035 txg = spa_vdev_enter(spa);
4037 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4040 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4042 if (!vd->vdev_ops->vdev_op_leaf)
4043 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4045 pvd = vd->vdev_parent;
4048 * If the parent/child relationship is not as expected, don't do it.
4049 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4050 * vdev that's replacing B with C. The user's intent in replacing
4051 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4052 * the replace by detaching C, the expected behavior is to end up
4053 * M(A,B). But suppose that right after deciding to detach C,
4054 * the replacement of B completes. We would have M(A,C), and then
4055 * ask to detach C, which would leave us with just A -- not what
4056 * the user wanted. To prevent this, we make sure that the
4057 * parent/child relationship hasn't changed -- in this example,
4058 * that C's parent is still the replacing vdev R.
4060 if (pvd->vdev_guid != pguid && pguid != 0)
4061 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4064 * Only 'replacing' or 'spare' vdevs can be replaced.
4066 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4067 pvd->vdev_ops != &vdev_spare_ops)
4068 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4070 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4071 spa_version(spa) >= SPA_VERSION_SPARES);
4074 * Only mirror, replacing, and spare vdevs support detach.
4076 if (pvd->vdev_ops != &vdev_replacing_ops &&
4077 pvd->vdev_ops != &vdev_mirror_ops &&
4078 pvd->vdev_ops != &vdev_spare_ops)
4079 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4082 * If this device has the only valid copy of some data,
4083 * we cannot safely detach it.
4085 if (vdev_dtl_required(vd))
4086 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4088 ASSERT(pvd->vdev_children >= 2);
4091 * If we are detaching the second disk from a replacing vdev, then
4092 * check to see if we changed the original vdev's path to have "/old"
4093 * at the end in spa_vdev_attach(). If so, undo that change now.
4095 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4096 vd->vdev_path != NULL) {
4097 size_t len = strlen(vd->vdev_path);
4099 for (c = 0; c < pvd->vdev_children; c++) {
4100 cvd = pvd->vdev_child[c];
4102 if (cvd == vd || cvd->vdev_path == NULL)
4105 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4106 strcmp(cvd->vdev_path + len, "/old") == 0) {
4107 spa_strfree(cvd->vdev_path);
4108 cvd->vdev_path = spa_strdup(vd->vdev_path);
4115 * If we are detaching the original disk from a spare, then it implies
4116 * that the spare should become a real disk, and be removed from the
4117 * active spare list for the pool.
4119 if (pvd->vdev_ops == &vdev_spare_ops &&
4121 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4125 * Erase the disk labels so the disk can be used for other things.
4126 * This must be done after all other error cases are handled,
4127 * but before we disembowel vd (so we can still do I/O to it).
4128 * But if we can't do it, don't treat the error as fatal --
4129 * it may be that the unwritability of the disk is the reason
4130 * it's being detached!
4132 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4135 * Remove vd from its parent and compact the parent's children.
4137 vdev_remove_child(pvd, vd);
4138 vdev_compact_children(pvd);
4141 * Remember one of the remaining children so we can get tvd below.
4143 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4146 * If we need to remove the remaining child from the list of hot spares,
4147 * do it now, marking the vdev as no longer a spare in the process.
4148 * We must do this before vdev_remove_parent(), because that can
4149 * change the GUID if it creates a new toplevel GUID. For a similar
4150 * reason, we must remove the spare now, in the same txg as the detach;
4151 * otherwise someone could attach a new sibling, change the GUID, and
4152 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4155 ASSERT(cvd->vdev_isspare);
4156 spa_spare_remove(cvd);
4157 unspare_guid = cvd->vdev_guid;
4158 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4159 cvd->vdev_unspare = B_TRUE;
4163 * If the parent mirror/replacing vdev only has one child,
4164 * the parent is no longer needed. Remove it from the tree.
4166 if (pvd->vdev_children == 1) {
4167 if (pvd->vdev_ops == &vdev_spare_ops)
4168 cvd->vdev_unspare = B_FALSE;
4169 vdev_remove_parent(cvd);
4170 cvd->vdev_resilvering = B_FALSE;
4175 * We don't set tvd until now because the parent we just removed
4176 * may have been the previous top-level vdev.
4178 tvd = cvd->vdev_top;
4179 ASSERT(tvd->vdev_parent == rvd);
4182 * Reevaluate the parent vdev state.
4184 vdev_propagate_state(cvd);
4187 * If the 'autoexpand' property is set on the pool then automatically
4188 * try to expand the size of the pool. For example if the device we
4189 * just detached was smaller than the others, it may be possible to
4190 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4191 * first so that we can obtain the updated sizes of the leaf vdevs.
4193 if (spa->spa_autoexpand) {
4195 vdev_expand(tvd, txg);
4198 vdev_config_dirty(tvd);
4201 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4202 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4203 * But first make sure we're not on any *other* txg's DTL list, to
4204 * prevent vd from being accessed after it's freed.
4206 vdpath = spa_strdup(vd->vdev_path);
4207 for (t = 0; t < TXG_SIZE; t++)
4208 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4209 vd->vdev_detached = B_TRUE;
4210 vdev_dirty(tvd, VDD_DTL, vd, txg);
4212 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4214 /* hang on to the spa before we release the lock */
4215 spa_open_ref(spa, FTAG);
4217 error = spa_vdev_exit(spa, vd, txg, 0);
4219 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4221 spa_strfree(vdpath);
4224 * If this was the removal of the original device in a hot spare vdev,
4225 * then we want to go through and remove the device from the hot spare
4226 * list of every other pool.
4229 spa_t *altspa = NULL;
4231 mutex_enter(&spa_namespace_lock);
4232 while ((altspa = spa_next(altspa)) != NULL) {
4233 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4237 spa_open_ref(altspa, FTAG);
4238 mutex_exit(&spa_namespace_lock);
4239 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4240 mutex_enter(&spa_namespace_lock);
4241 spa_close(altspa, FTAG);
4243 mutex_exit(&spa_namespace_lock);
4245 /* search the rest of the vdevs for spares to remove */
4246 spa_vdev_resilver_done(spa);
4249 /* all done with the spa; OK to release */
4250 mutex_enter(&spa_namespace_lock);
4251 spa_close(spa, FTAG);
4252 mutex_exit(&spa_namespace_lock);
4258 * Split a set of devices from their mirrors, and create a new pool from them.
4261 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4262 nvlist_t *props, boolean_t exp)
4265 uint64_t txg, *glist;
4267 uint_t c, children, lastlog;
4268 nvlist_t **child, *nvl, *tmp;
4270 char *altroot = NULL;
4271 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4272 boolean_t activate_slog;
4274 ASSERT(spa_writeable(spa));
4276 txg = spa_vdev_enter(spa);
4278 /* clear the log and flush everything up to now */
4279 activate_slog = spa_passivate_log(spa);
4280 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4281 error = spa_offline_log(spa);
4282 txg = spa_vdev_config_enter(spa);
4285 spa_activate_log(spa);
4288 return (spa_vdev_exit(spa, NULL, txg, error));
4290 /* check new spa name before going any further */
4291 if (spa_lookup(newname) != NULL)
4292 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4295 * scan through all the children to ensure they're all mirrors
4297 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4298 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4300 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4302 /* first, check to ensure we've got the right child count */
4303 rvd = spa->spa_root_vdev;
4305 for (c = 0; c < rvd->vdev_children; c++) {
4306 vdev_t *vd = rvd->vdev_child[c];
4308 /* don't count the holes & logs as children */
4309 if (vd->vdev_islog || vd->vdev_ishole) {
4317 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4318 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4320 /* next, ensure no spare or cache devices are part of the split */
4321 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4322 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4323 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4325 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4326 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4328 /* then, loop over each vdev and validate it */
4329 for (c = 0; c < children; c++) {
4330 uint64_t is_hole = 0;
4332 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4336 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4337 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4345 /* which disk is going to be split? */
4346 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4352 /* look it up in the spa */
4353 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4354 if (vml[c] == NULL) {
4359 /* make sure there's nothing stopping the split */
4360 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4361 vml[c]->vdev_islog ||
4362 vml[c]->vdev_ishole ||
4363 vml[c]->vdev_isspare ||
4364 vml[c]->vdev_isl2cache ||
4365 !vdev_writeable(vml[c]) ||
4366 vml[c]->vdev_children != 0 ||
4367 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4368 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4373 if (vdev_dtl_required(vml[c])) {
4378 /* we need certain info from the top level */
4379 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4380 vml[c]->vdev_top->vdev_ms_array) == 0);
4381 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4382 vml[c]->vdev_top->vdev_ms_shift) == 0);
4383 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4384 vml[c]->vdev_top->vdev_asize) == 0);
4385 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4386 vml[c]->vdev_top->vdev_ashift) == 0);
4390 kmem_free(vml, children * sizeof (vdev_t *));
4391 kmem_free(glist, children * sizeof (uint64_t));
4392 return (spa_vdev_exit(spa, NULL, txg, error));
4395 /* stop writers from using the disks */
4396 for (c = 0; c < children; c++) {
4398 vml[c]->vdev_offline = B_TRUE;
4400 vdev_reopen(spa->spa_root_vdev);
4403 * Temporarily record the splitting vdevs in the spa config. This
4404 * will disappear once the config is regenerated.
4406 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4407 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4408 glist, children) == 0);
4409 kmem_free(glist, children * sizeof (uint64_t));
4411 mutex_enter(&spa->spa_props_lock);
4412 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4414 mutex_exit(&spa->spa_props_lock);
4415 spa->spa_config_splitting = nvl;
4416 vdev_config_dirty(spa->spa_root_vdev);
4418 /* configure and create the new pool */
4419 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4420 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4421 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4422 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4423 spa_version(spa)) == 0);
4424 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4425 spa->spa_config_txg) == 0);
4426 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4427 spa_generate_guid(NULL)) == 0);
4428 (void) nvlist_lookup_string(props,
4429 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4431 /* add the new pool to the namespace */
4432 newspa = spa_add(newname, config, altroot);
4433 newspa->spa_config_txg = spa->spa_config_txg;
4434 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4436 /* release the spa config lock, retaining the namespace lock */
4437 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4439 if (zio_injection_enabled)
4440 zio_handle_panic_injection(spa, FTAG, 1);
4442 spa_activate(newspa, spa_mode_global);
4443 spa_async_suspend(newspa);
4445 /* create the new pool from the disks of the original pool */
4446 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4450 /* if that worked, generate a real config for the new pool */
4451 if (newspa->spa_root_vdev != NULL) {
4452 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4453 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4454 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4455 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4456 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4461 if (props != NULL) {
4462 spa_configfile_set(newspa, props, B_FALSE);
4463 error = spa_prop_set(newspa, props);
4468 /* flush everything */
4469 txg = spa_vdev_config_enter(newspa);
4470 vdev_config_dirty(newspa->spa_root_vdev);
4471 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4473 if (zio_injection_enabled)
4474 zio_handle_panic_injection(spa, FTAG, 2);
4476 spa_async_resume(newspa);
4478 /* finally, update the original pool's config */
4479 txg = spa_vdev_config_enter(spa);
4480 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4481 error = dmu_tx_assign(tx, TXG_WAIT);
4484 for (c = 0; c < children; c++) {
4485 if (vml[c] != NULL) {
4488 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4494 vdev_config_dirty(spa->spa_root_vdev);
4495 spa->spa_config_splitting = NULL;
4499 (void) spa_vdev_exit(spa, NULL, txg, 0);
4501 if (zio_injection_enabled)
4502 zio_handle_panic_injection(spa, FTAG, 3);
4504 /* split is complete; log a history record */
4505 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4506 "split new pool %s from pool %s", newname, spa_name(spa));
4508 kmem_free(vml, children * sizeof (vdev_t *));
4510 /* if we're not going to mount the filesystems in userland, export */
4512 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4519 spa_deactivate(newspa);
4522 txg = spa_vdev_config_enter(spa);
4524 /* re-online all offlined disks */
4525 for (c = 0; c < children; c++) {
4527 vml[c]->vdev_offline = B_FALSE;
4529 vdev_reopen(spa->spa_root_vdev);
4531 nvlist_free(spa->spa_config_splitting);
4532 spa->spa_config_splitting = NULL;
4533 (void) spa_vdev_exit(spa, NULL, txg, error);
4535 kmem_free(vml, children * sizeof (vdev_t *));
4540 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4544 for (i = 0; i < count; i++) {
4547 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4550 if (guid == target_guid)
4558 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4559 nvlist_t *dev_to_remove)
4561 nvlist_t **newdev = NULL;
4565 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4567 for (i = 0, j = 0; i < count; i++) {
4568 if (dev[i] == dev_to_remove)
4570 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4573 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4574 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4576 for (i = 0; i < count - 1; i++)
4577 nvlist_free(newdev[i]);
4580 kmem_free(newdev, (count - 1) * sizeof (void *));
4584 * Evacuate the device.
4587 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4592 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4593 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4594 ASSERT(vd == vd->vdev_top);
4597 * Evacuate the device. We don't hold the config lock as writer
4598 * since we need to do I/O but we do keep the
4599 * spa_namespace_lock held. Once this completes the device
4600 * should no longer have any blocks allocated on it.
4602 if (vd->vdev_islog) {
4603 if (vd->vdev_stat.vs_alloc != 0)
4604 error = spa_offline_log(spa);
4613 * The evacuation succeeded. Remove any remaining MOS metadata
4614 * associated with this vdev, and wait for these changes to sync.
4616 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4617 txg = spa_vdev_config_enter(spa);
4618 vd->vdev_removing = B_TRUE;
4619 vdev_dirty(vd, 0, NULL, txg);
4620 vdev_config_dirty(vd);
4621 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4627 * Complete the removal by cleaning up the namespace.
4630 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4632 vdev_t *rvd = spa->spa_root_vdev;
4633 uint64_t id = vd->vdev_id;
4634 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4636 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4637 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4638 ASSERT(vd == vd->vdev_top);
4641 * Only remove any devices which are empty.
4643 if (vd->vdev_stat.vs_alloc != 0)
4646 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4648 if (list_link_active(&vd->vdev_state_dirty_node))
4649 vdev_state_clean(vd);
4650 if (list_link_active(&vd->vdev_config_dirty_node))
4651 vdev_config_clean(vd);
4656 vdev_compact_children(rvd);
4658 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4659 vdev_add_child(rvd, vd);
4661 vdev_config_dirty(rvd);
4664 * Reassess the health of our root vdev.
4670 * Remove a device from the pool -
4672 * Removing a device from the vdev namespace requires several steps
4673 * and can take a significant amount of time. As a result we use
4674 * the spa_vdev_config_[enter/exit] functions which allow us to
4675 * grab and release the spa_config_lock while still holding the namespace
4676 * lock. During each step the configuration is synced out.
4680 * Remove a device from the pool. Currently, this supports removing only hot
4681 * spares, slogs, and level 2 ARC devices.
4684 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4687 metaslab_group_t *mg;
4688 nvlist_t **spares, **l2cache, *nv;
4690 uint_t nspares, nl2cache;
4692 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4694 ASSERT(spa_writeable(spa));
4697 txg = spa_vdev_enter(spa);
4699 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4701 if (spa->spa_spares.sav_vdevs != NULL &&
4702 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4703 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4704 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4706 * Only remove the hot spare if it's not currently in use
4709 if (vd == NULL || unspare) {
4710 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4711 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4712 spa_load_spares(spa);
4713 spa->spa_spares.sav_sync = B_TRUE;
4717 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4718 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4719 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4720 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4722 * Cache devices can always be removed.
4724 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4725 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4726 spa_load_l2cache(spa);
4727 spa->spa_l2cache.sav_sync = B_TRUE;
4728 } else if (vd != NULL && vd->vdev_islog) {
4730 ASSERT(vd == vd->vdev_top);
4733 * XXX - Once we have bp-rewrite this should
4734 * become the common case.
4740 * Stop allocating from this vdev.
4742 metaslab_group_passivate(mg);
4745 * Wait for the youngest allocations and frees to sync,
4746 * and then wait for the deferral of those frees to finish.
4748 spa_vdev_config_exit(spa, NULL,
4749 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4752 * Attempt to evacuate the vdev.
4754 error = spa_vdev_remove_evacuate(spa, vd);
4756 txg = spa_vdev_config_enter(spa);
4759 * If we couldn't evacuate the vdev, unwind.
4762 metaslab_group_activate(mg);
4763 return (spa_vdev_exit(spa, NULL, txg, error));
4767 * Clean up the vdev namespace.
4769 spa_vdev_remove_from_namespace(spa, vd);
4771 } else if (vd != NULL) {
4773 * Normal vdevs cannot be removed (yet).
4778 * There is no vdev of any kind with the specified guid.
4784 return (spa_vdev_exit(spa, NULL, txg, error));
4790 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4791 * current spared, so we can detach it.
4794 spa_vdev_resilver_done_hunt(vdev_t *vd)
4796 vdev_t *newvd, *oldvd;
4799 for (c = 0; c < vd->vdev_children; c++) {
4800 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4806 * Check for a completed replacement. We always consider the first
4807 * vdev in the list to be the oldest vdev, and the last one to be
4808 * the newest (see spa_vdev_attach() for how that works). In
4809 * the case where the newest vdev is faulted, we will not automatically
4810 * remove it after a resilver completes. This is OK as it will require
4811 * user intervention to determine which disk the admin wishes to keep.
4813 if (vd->vdev_ops == &vdev_replacing_ops) {
4814 ASSERT(vd->vdev_children > 1);
4816 newvd = vd->vdev_child[vd->vdev_children - 1];
4817 oldvd = vd->vdev_child[0];
4819 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4820 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4821 !vdev_dtl_required(oldvd))
4826 * Check for a completed resilver with the 'unspare' flag set.
4828 if (vd->vdev_ops == &vdev_spare_ops) {
4829 vdev_t *first = vd->vdev_child[0];
4830 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4832 if (last->vdev_unspare) {
4835 } else if (first->vdev_unspare) {
4842 if (oldvd != NULL &&
4843 vdev_dtl_empty(newvd, DTL_MISSING) &&
4844 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4845 !vdev_dtl_required(oldvd))
4849 * If there are more than two spares attached to a disk,
4850 * and those spares are not required, then we want to
4851 * attempt to free them up now so that they can be used
4852 * by other pools. Once we're back down to a single
4853 * disk+spare, we stop removing them.
4855 if (vd->vdev_children > 2) {
4856 newvd = vd->vdev_child[1];
4858 if (newvd->vdev_isspare && last->vdev_isspare &&
4859 vdev_dtl_empty(last, DTL_MISSING) &&
4860 vdev_dtl_empty(last, DTL_OUTAGE) &&
4861 !vdev_dtl_required(newvd))
4870 spa_vdev_resilver_done(spa_t *spa)
4872 vdev_t *vd, *pvd, *ppvd;
4873 uint64_t guid, sguid, pguid, ppguid;
4875 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4877 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4878 pvd = vd->vdev_parent;
4879 ppvd = pvd->vdev_parent;
4880 guid = vd->vdev_guid;
4881 pguid = pvd->vdev_guid;
4882 ppguid = ppvd->vdev_guid;
4885 * If we have just finished replacing a hot spared device, then
4886 * we need to detach the parent's first child (the original hot
4889 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4890 ppvd->vdev_children == 2) {
4891 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4892 sguid = ppvd->vdev_child[1]->vdev_guid;
4894 spa_config_exit(spa, SCL_ALL, FTAG);
4895 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4897 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4899 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4902 spa_config_exit(spa, SCL_ALL, FTAG);
4906 * Update the stored path or FRU for this vdev.
4909 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4913 boolean_t sync = B_FALSE;
4915 ASSERT(spa_writeable(spa));
4917 spa_vdev_state_enter(spa, SCL_ALL);
4919 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4920 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4922 if (!vd->vdev_ops->vdev_op_leaf)
4923 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4926 if (strcmp(value, vd->vdev_path) != 0) {
4927 spa_strfree(vd->vdev_path);
4928 vd->vdev_path = spa_strdup(value);
4932 if (vd->vdev_fru == NULL) {
4933 vd->vdev_fru = spa_strdup(value);
4935 } else if (strcmp(value, vd->vdev_fru) != 0) {
4936 spa_strfree(vd->vdev_fru);
4937 vd->vdev_fru = spa_strdup(value);
4942 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4946 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4948 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4952 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4954 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4958 * ==========================================================================
4960 * ==========================================================================
4964 spa_scan_stop(spa_t *spa)
4966 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4967 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4969 return (dsl_scan_cancel(spa->spa_dsl_pool));
4973 spa_scan(spa_t *spa, pool_scan_func_t func)
4975 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4977 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4981 * If a resilver was requested, but there is no DTL on a
4982 * writeable leaf device, we have nothing to do.
4984 if (func == POOL_SCAN_RESILVER &&
4985 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4986 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4990 return (dsl_scan(spa->spa_dsl_pool, func));
4994 * ==========================================================================
4995 * SPA async task processing
4996 * ==========================================================================
5000 spa_async_remove(spa_t *spa, vdev_t *vd)
5004 if (vd->vdev_remove_wanted) {
5005 vd->vdev_remove_wanted = B_FALSE;
5006 vd->vdev_delayed_close = B_FALSE;
5007 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5010 * We want to clear the stats, but we don't want to do a full
5011 * vdev_clear() as that will cause us to throw away
5012 * degraded/faulted state as well as attempt to reopen the
5013 * device, all of which is a waste.
5015 vd->vdev_stat.vs_read_errors = 0;
5016 vd->vdev_stat.vs_write_errors = 0;
5017 vd->vdev_stat.vs_checksum_errors = 0;
5019 vdev_state_dirty(vd->vdev_top);
5022 for (c = 0; c < vd->vdev_children; c++)
5023 spa_async_remove(spa, vd->vdev_child[c]);
5027 spa_async_probe(spa_t *spa, vdev_t *vd)
5031 if (vd->vdev_probe_wanted) {
5032 vd->vdev_probe_wanted = B_FALSE;
5033 vdev_reopen(vd); /* vdev_open() does the actual probe */
5036 for (c = 0; c < vd->vdev_children; c++)
5037 spa_async_probe(spa, vd->vdev_child[c]);
5041 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5045 if (!spa->spa_autoexpand)
5048 for (c = 0; c < vd->vdev_children; c++) {
5049 vdev_t *cvd = vd->vdev_child[c];
5050 spa_async_autoexpand(spa, cvd);
5053 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5056 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5060 spa_async_thread(spa_t *spa)
5064 ASSERT(spa->spa_sync_on);
5066 mutex_enter(&spa->spa_async_lock);
5067 tasks = spa->spa_async_tasks;
5068 spa->spa_async_tasks = 0;
5069 mutex_exit(&spa->spa_async_lock);
5072 * See if the config needs to be updated.
5074 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5075 uint64_t old_space, new_space;
5077 mutex_enter(&spa_namespace_lock);
5078 old_space = metaslab_class_get_space(spa_normal_class(spa));
5079 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5080 new_space = metaslab_class_get_space(spa_normal_class(spa));
5081 mutex_exit(&spa_namespace_lock);
5084 * If the pool grew as a result of the config update,
5085 * then log an internal history event.
5087 if (new_space != old_space) {
5088 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5090 "pool '%s' size: %llu(+%llu)",
5091 spa_name(spa), new_space, new_space - old_space);
5096 * See if any devices need to be marked REMOVED.
5098 if (tasks & SPA_ASYNC_REMOVE) {
5099 spa_vdev_state_enter(spa, SCL_NONE);
5100 spa_async_remove(spa, spa->spa_root_vdev);
5101 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5102 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5103 for (i = 0; i < spa->spa_spares.sav_count; i++)
5104 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5105 (void) spa_vdev_state_exit(spa, NULL, 0);
5108 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5109 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5110 spa_async_autoexpand(spa, spa->spa_root_vdev);
5111 spa_config_exit(spa, SCL_CONFIG, FTAG);
5115 * See if any devices need to be probed.
5117 if (tasks & SPA_ASYNC_PROBE) {
5118 spa_vdev_state_enter(spa, SCL_NONE);
5119 spa_async_probe(spa, spa->spa_root_vdev);
5120 (void) spa_vdev_state_exit(spa, NULL, 0);
5124 * If any devices are done replacing, detach them.
5126 if (tasks & SPA_ASYNC_RESILVER_DONE)
5127 spa_vdev_resilver_done(spa);
5130 * Kick off a resilver.
5132 if (tasks & SPA_ASYNC_RESILVER)
5133 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5136 * Let the world know that we're done.
5138 mutex_enter(&spa->spa_async_lock);
5139 spa->spa_async_thread = NULL;
5140 cv_broadcast(&spa->spa_async_cv);
5141 mutex_exit(&spa->spa_async_lock);
5146 spa_async_suspend(spa_t *spa)
5148 mutex_enter(&spa->spa_async_lock);
5149 spa->spa_async_suspended++;
5150 while (spa->spa_async_thread != NULL)
5151 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5152 mutex_exit(&spa->spa_async_lock);
5156 spa_async_resume(spa_t *spa)
5158 mutex_enter(&spa->spa_async_lock);
5159 ASSERT(spa->spa_async_suspended != 0);
5160 spa->spa_async_suspended--;
5161 mutex_exit(&spa->spa_async_lock);
5165 spa_async_dispatch(spa_t *spa)
5167 mutex_enter(&spa->spa_async_lock);
5168 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5169 spa->spa_async_thread == NULL &&
5170 rootdir != NULL && !vn_is_readonly(rootdir))
5171 spa->spa_async_thread = thread_create(NULL, 0,
5172 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5173 mutex_exit(&spa->spa_async_lock);
5177 spa_async_request(spa_t *spa, int task)
5179 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5180 mutex_enter(&spa->spa_async_lock);
5181 spa->spa_async_tasks |= task;
5182 mutex_exit(&spa->spa_async_lock);
5186 * ==========================================================================
5187 * SPA syncing routines
5188 * ==========================================================================
5192 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5195 bpobj_enqueue(bpo, bp, tx);
5200 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5204 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5210 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5212 char *packed = NULL;
5217 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5220 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5221 * information. This avoids the dbuf_will_dirty() path and
5222 * saves us a pre-read to get data we don't actually care about.
5224 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5225 packed = vmem_alloc(bufsize, KM_SLEEP);
5227 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5229 bzero(packed + nvsize, bufsize - nvsize);
5231 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5233 vmem_free(packed, bufsize);
5235 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5236 dmu_buf_will_dirty(db, tx);
5237 *(uint64_t *)db->db_data = nvsize;
5238 dmu_buf_rele(db, FTAG);
5242 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5243 const char *config, const char *entry)
5253 * Update the MOS nvlist describing the list of available devices.
5254 * spa_validate_aux() will have already made sure this nvlist is
5255 * valid and the vdevs are labeled appropriately.
5257 if (sav->sav_object == 0) {
5258 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5259 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5260 sizeof (uint64_t), tx);
5261 VERIFY(zap_update(spa->spa_meta_objset,
5262 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5263 &sav->sav_object, tx) == 0);
5266 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5267 if (sav->sav_count == 0) {
5268 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5270 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5271 for (i = 0; i < sav->sav_count; i++)
5272 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5273 B_FALSE, VDEV_CONFIG_L2CACHE);
5274 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5275 sav->sav_count) == 0);
5276 for (i = 0; i < sav->sav_count; i++)
5277 nvlist_free(list[i]);
5278 kmem_free(list, sav->sav_count * sizeof (void *));
5281 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5282 nvlist_free(nvroot);
5284 sav->sav_sync = B_FALSE;
5288 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5292 if (list_is_empty(&spa->spa_config_dirty_list))
5295 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5297 config = spa_config_generate(spa, spa->spa_root_vdev,
5298 dmu_tx_get_txg(tx), B_FALSE);
5300 spa_config_exit(spa, SCL_STATE, FTAG);
5302 if (spa->spa_config_syncing)
5303 nvlist_free(spa->spa_config_syncing);
5304 spa->spa_config_syncing = config;
5306 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5310 * Set zpool properties.
5313 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5316 objset_t *mos = spa->spa_meta_objset;
5317 nvlist_t *nvp = arg2;
5322 const char *propname;
5323 zprop_type_t proptype;
5325 mutex_enter(&spa->spa_props_lock);
5328 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5329 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5330 case ZPOOL_PROP_VERSION:
5332 * Only set version for non-zpool-creation cases
5333 * (set/import). spa_create() needs special care
5334 * for version setting.
5336 if (tx->tx_txg != TXG_INITIAL) {
5337 VERIFY(nvpair_value_uint64(elem,
5339 ASSERT(intval <= SPA_VERSION);
5340 ASSERT(intval >= spa_version(spa));
5341 spa->spa_uberblock.ub_version = intval;
5342 vdev_config_dirty(spa->spa_root_vdev);
5346 case ZPOOL_PROP_ALTROOT:
5348 * 'altroot' is a non-persistent property. It should
5349 * have been set temporarily at creation or import time.
5351 ASSERT(spa->spa_root != NULL);
5354 case ZPOOL_PROP_READONLY:
5355 case ZPOOL_PROP_CACHEFILE:
5357 * 'readonly' and 'cachefile' are also non-persisitent
5363 * Set pool property values in the poolprops mos object.
5365 if (spa->spa_pool_props_object == 0) {
5366 VERIFY((spa->spa_pool_props_object =
5367 zap_create(mos, DMU_OT_POOL_PROPS,
5368 DMU_OT_NONE, 0, tx)) > 0);
5370 VERIFY(zap_update(mos,
5371 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5372 8, 1, &spa->spa_pool_props_object, tx)
5376 /* normalize the property name */
5377 propname = zpool_prop_to_name(prop);
5378 proptype = zpool_prop_get_type(prop);
5380 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5381 ASSERT(proptype == PROP_TYPE_STRING);
5382 VERIFY(nvpair_value_string(elem, &strval) == 0);
5383 VERIFY(zap_update(mos,
5384 spa->spa_pool_props_object, propname,
5385 1, strlen(strval) + 1, strval, tx) == 0);
5387 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5388 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5390 if (proptype == PROP_TYPE_INDEX) {
5392 VERIFY(zpool_prop_index_to_string(
5393 prop, intval, &unused) == 0);
5395 VERIFY(zap_update(mos,
5396 spa->spa_pool_props_object, propname,
5397 8, 1, &intval, tx) == 0);
5399 ASSERT(0); /* not allowed */
5403 case ZPOOL_PROP_DELEGATION:
5404 spa->spa_delegation = intval;
5406 case ZPOOL_PROP_BOOTFS:
5407 spa->spa_bootfs = intval;
5409 case ZPOOL_PROP_FAILUREMODE:
5410 spa->spa_failmode = intval;
5412 case ZPOOL_PROP_AUTOEXPAND:
5413 spa->spa_autoexpand = intval;
5414 if (tx->tx_txg != TXG_INITIAL)
5415 spa_async_request(spa,
5416 SPA_ASYNC_AUTOEXPAND);
5418 case ZPOOL_PROP_DEDUPDITTO:
5419 spa->spa_dedup_ditto = intval;
5426 /* log internal history if this is not a zpool create */
5427 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5428 tx->tx_txg != TXG_INITIAL) {
5429 spa_history_log_internal(LOG_POOL_PROPSET,
5430 spa, tx, "%s %lld %s",
5431 nvpair_name(elem), intval, spa_name(spa));
5435 mutex_exit(&spa->spa_props_lock);
5439 * Perform one-time upgrade on-disk changes. spa_version() does not
5440 * reflect the new version this txg, so there must be no changes this
5441 * txg to anything that the upgrade code depends on after it executes.
5442 * Therefore this must be called after dsl_pool_sync() does the sync
5446 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5448 dsl_pool_t *dp = spa->spa_dsl_pool;
5450 ASSERT(spa->spa_sync_pass == 1);
5452 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5453 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5454 dsl_pool_create_origin(dp, tx);
5456 /* Keeping the origin open increases spa_minref */
5457 spa->spa_minref += 3;
5460 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5461 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5462 dsl_pool_upgrade_clones(dp, tx);
5465 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5466 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5467 dsl_pool_upgrade_dir_clones(dp, tx);
5469 /* Keeping the freedir open increases spa_minref */
5470 spa->spa_minref += 3;
5475 * Sync the specified transaction group. New blocks may be dirtied as
5476 * part of the process, so we iterate until it converges.
5479 spa_sync(spa_t *spa, uint64_t txg)
5481 dsl_pool_t *dp = spa->spa_dsl_pool;
5482 objset_t *mos = spa->spa_meta_objset;
5483 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5484 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5485 vdev_t *rvd = spa->spa_root_vdev;
5491 VERIFY(spa_writeable(spa));
5494 * Lock out configuration changes.
5496 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5498 spa->spa_syncing_txg = txg;
5499 spa->spa_sync_pass = 0;
5502 * If there are any pending vdev state changes, convert them
5503 * into config changes that go out with this transaction group.
5505 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5506 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5508 * We need the write lock here because, for aux vdevs,
5509 * calling vdev_config_dirty() modifies sav_config.
5510 * This is ugly and will become unnecessary when we
5511 * eliminate the aux vdev wart by integrating all vdevs
5512 * into the root vdev tree.
5514 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5515 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5516 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5517 vdev_state_clean(vd);
5518 vdev_config_dirty(vd);
5520 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5521 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5523 spa_config_exit(spa, SCL_STATE, FTAG);
5525 tx = dmu_tx_create_assigned(dp, txg);
5528 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5529 * set spa_deflate if we have no raid-z vdevs.
5531 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5532 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5535 for (i = 0; i < rvd->vdev_children; i++) {
5536 vd = rvd->vdev_child[i];
5537 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5540 if (i == rvd->vdev_children) {
5541 spa->spa_deflate = TRUE;
5542 VERIFY(0 == zap_add(spa->spa_meta_objset,
5543 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5544 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5549 * If anything has changed in this txg, or if someone is waiting
5550 * for this txg to sync (eg, spa_vdev_remove()), push the
5551 * deferred frees from the previous txg. If not, leave them
5552 * alone so that we don't generate work on an otherwise idle
5555 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5556 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5557 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5558 ((dsl_scan_active(dp->dp_scan) ||
5559 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5560 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5561 VERIFY3U(bpobj_iterate(defer_bpo,
5562 spa_free_sync_cb, zio, tx), ==, 0);
5563 VERIFY3U(zio_wait(zio), ==, 0);
5567 * Iterate to convergence.
5570 int pass = ++spa->spa_sync_pass;
5572 spa_sync_config_object(spa, tx);
5573 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5574 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5575 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5576 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5577 spa_errlog_sync(spa, txg);
5578 dsl_pool_sync(dp, txg);
5580 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5581 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5582 bplist_iterate(free_bpl, spa_free_sync_cb,
5584 VERIFY(zio_wait(zio) == 0);
5586 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5591 dsl_scan_sync(dp, tx);
5593 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5597 spa_sync_upgrades(spa, tx);
5599 } while (dmu_objset_is_dirty(mos, txg));
5602 * Rewrite the vdev configuration (which includes the uberblock)
5603 * to commit the transaction group.
5605 * If there are no dirty vdevs, we sync the uberblock to a few
5606 * random top-level vdevs that are known to be visible in the
5607 * config cache (see spa_vdev_add() for a complete description).
5608 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5612 * We hold SCL_STATE to prevent vdev open/close/etc.
5613 * while we're attempting to write the vdev labels.
5615 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5617 if (list_is_empty(&spa->spa_config_dirty_list)) {
5618 vdev_t *svd[SPA_DVAS_PER_BP];
5620 int children = rvd->vdev_children;
5621 int c0 = spa_get_random(children);
5623 for (c = 0; c < children; c++) {
5624 vd = rvd->vdev_child[(c0 + c) % children];
5625 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5627 svd[svdcount++] = vd;
5628 if (svdcount == SPA_DVAS_PER_BP)
5631 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5633 error = vdev_config_sync(svd, svdcount, txg,
5636 error = vdev_config_sync(rvd->vdev_child,
5637 rvd->vdev_children, txg, B_FALSE);
5639 error = vdev_config_sync(rvd->vdev_child,
5640 rvd->vdev_children, txg, B_TRUE);
5643 spa_config_exit(spa, SCL_STATE, FTAG);
5647 zio_suspend(spa, NULL);
5648 zio_resume_wait(spa);
5653 * Clear the dirty config list.
5655 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5656 vdev_config_clean(vd);
5659 * Now that the new config has synced transactionally,
5660 * let it become visible to the config cache.
5662 if (spa->spa_config_syncing != NULL) {
5663 spa_config_set(spa, spa->spa_config_syncing);
5664 spa->spa_config_txg = txg;
5665 spa->spa_config_syncing = NULL;
5668 spa->spa_ubsync = spa->spa_uberblock;
5670 dsl_pool_sync_done(dp, txg);
5673 * Update usable space statistics.
5675 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5676 vdev_sync_done(vd, txg);
5678 spa_update_dspace(spa);
5681 * It had better be the case that we didn't dirty anything
5682 * since vdev_config_sync().
5684 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5685 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5686 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5688 spa->spa_sync_pass = 0;
5690 spa_config_exit(spa, SCL_CONFIG, FTAG);
5692 spa_handle_ignored_writes(spa);
5695 * If any async tasks have been requested, kick them off.
5697 spa_async_dispatch(spa);
5701 * Sync all pools. We don't want to hold the namespace lock across these
5702 * operations, so we take a reference on the spa_t and drop the lock during the
5706 spa_sync_allpools(void)
5709 mutex_enter(&spa_namespace_lock);
5710 while ((spa = spa_next(spa)) != NULL) {
5711 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5712 !spa_writeable(spa) || spa_suspended(spa))
5714 spa_open_ref(spa, FTAG);
5715 mutex_exit(&spa_namespace_lock);
5716 txg_wait_synced(spa_get_dsl(spa), 0);
5717 mutex_enter(&spa_namespace_lock);
5718 spa_close(spa, FTAG);
5720 mutex_exit(&spa_namespace_lock);
5724 * ==========================================================================
5725 * Miscellaneous routines
5726 * ==========================================================================
5730 * Remove all pools in the system.
5738 * Remove all cached state. All pools should be closed now,
5739 * so every spa in the AVL tree should be unreferenced.
5741 mutex_enter(&spa_namespace_lock);
5742 while ((spa = spa_next(NULL)) != NULL) {
5744 * Stop async tasks. The async thread may need to detach
5745 * a device that's been replaced, which requires grabbing
5746 * spa_namespace_lock, so we must drop it here.
5748 spa_open_ref(spa, FTAG);
5749 mutex_exit(&spa_namespace_lock);
5750 spa_async_suspend(spa);
5751 mutex_enter(&spa_namespace_lock);
5752 spa_close(spa, FTAG);
5754 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5756 spa_deactivate(spa);
5760 mutex_exit(&spa_namespace_lock);
5764 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5769 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5773 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5774 vd = spa->spa_l2cache.sav_vdevs[i];
5775 if (vd->vdev_guid == guid)
5779 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5780 vd = spa->spa_spares.sav_vdevs[i];
5781 if (vd->vdev_guid == guid)
5790 spa_upgrade(spa_t *spa, uint64_t version)
5792 ASSERT(spa_writeable(spa));
5794 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5797 * This should only be called for a non-faulted pool, and since a
5798 * future version would result in an unopenable pool, this shouldn't be
5801 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5802 ASSERT(version >= spa->spa_uberblock.ub_version);
5804 spa->spa_uberblock.ub_version = version;
5805 vdev_config_dirty(spa->spa_root_vdev);
5807 spa_config_exit(spa, SCL_ALL, FTAG);
5809 txg_wait_synced(spa_get_dsl(spa), 0);
5813 spa_has_spare(spa_t *spa, uint64_t guid)
5817 spa_aux_vdev_t *sav = &spa->spa_spares;
5819 for (i = 0; i < sav->sav_count; i++)
5820 if (sav->sav_vdevs[i]->vdev_guid == guid)
5823 for (i = 0; i < sav->sav_npending; i++) {
5824 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5825 &spareguid) == 0 && spareguid == guid)
5833 * Check if a pool has an active shared spare device.
5834 * Note: reference count of an active spare is 2, as a spare and as a replace
5837 spa_has_active_shared_spare(spa_t *spa)
5841 spa_aux_vdev_t *sav = &spa->spa_spares;
5843 for (i = 0; i < sav->sav_count; i++) {
5844 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5845 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5854 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5855 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5856 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5857 * or zdb as real changes.
5860 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5863 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
5867 #if defined(_KERNEL) && defined(HAVE_SPL)
5868 /* state manipulation functions */
5869 EXPORT_SYMBOL(spa_open);
5870 EXPORT_SYMBOL(spa_open_rewind);
5871 EXPORT_SYMBOL(spa_get_stats);
5872 EXPORT_SYMBOL(spa_create);
5873 EXPORT_SYMBOL(spa_import_rootpool);
5874 EXPORT_SYMBOL(spa_import);
5875 EXPORT_SYMBOL(spa_tryimport);
5876 EXPORT_SYMBOL(spa_destroy);
5877 EXPORT_SYMBOL(spa_export);
5878 EXPORT_SYMBOL(spa_reset);
5879 EXPORT_SYMBOL(spa_async_request);
5880 EXPORT_SYMBOL(spa_async_suspend);
5881 EXPORT_SYMBOL(spa_async_resume);
5882 EXPORT_SYMBOL(spa_inject_addref);
5883 EXPORT_SYMBOL(spa_inject_delref);
5884 EXPORT_SYMBOL(spa_scan_stat_init);
5885 EXPORT_SYMBOL(spa_scan_get_stats);
5887 /* device maniion */
5888 EXPORT_SYMBOL(spa_vdev_add);
5889 EXPORT_SYMBOL(spa_vdev_attach);
5890 EXPORT_SYMBOL(spa_vdev_detach);
5891 EXPORT_SYMBOL(spa_vdev_remove);
5892 EXPORT_SYMBOL(spa_vdev_setpath);
5893 EXPORT_SYMBOL(spa_vdev_setfru);
5894 EXPORT_SYMBOL(spa_vdev_split_mirror);
5896 /* spare statech is global across all pools) */
5897 EXPORT_SYMBOL(spa_spare_add);
5898 EXPORT_SYMBOL(spa_spare_remove);
5899 EXPORT_SYMBOL(spa_spare_exists);
5900 EXPORT_SYMBOL(spa_spare_activate);
5902 /* L2ARC statech is global across all pools) */
5903 EXPORT_SYMBOL(spa_l2cache_add);
5904 EXPORT_SYMBOL(spa_l2cache_remove);
5905 EXPORT_SYMBOL(spa_l2cache_exists);
5906 EXPORT_SYMBOL(spa_l2cache_activate);
5907 EXPORT_SYMBOL(spa_l2cache_drop);
5910 EXPORT_SYMBOL(spa_scan);
5911 EXPORT_SYMBOL(spa_scan_stop);
5914 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
5915 EXPORT_SYMBOL(spa_sync_allpools);
5918 EXPORT_SYMBOL(spa_prop_set);
5919 EXPORT_SYMBOL(spa_prop_get);
5920 EXPORT_SYMBOL(spa_prop_clear_bootfs);
5922 /* asynchronous event notification */
5923 EXPORT_SYMBOL(spa_event_notify);