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,
617 uint_t value, uint_t flags)
619 boolean_t batch = B_FALSE;
623 return (NULL); /* no taskq needed */
626 ASSERT3U(value, >=, 1);
627 value = MAX(value, 1);
632 flags |= TASKQ_THREADS_CPU_PCT;
633 value = zio_taskq_batch_pct;
636 case zti_mode_online_percent:
637 flags |= TASKQ_THREADS_CPU_PCT;
641 panic("unrecognized mode for %s taskq (%u:%u) in "
647 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
649 flags |= TASKQ_DC_BATCH;
651 return (taskq_create_sysdc(name, value, 50, INT_MAX,
652 spa->spa_proc, zio_taskq_basedc, flags));
654 return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
655 spa->spa_proc, flags));
659 spa_create_zio_taskqs(spa_t *spa)
663 for (t = 0; t < ZIO_TYPES; t++) {
664 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
665 const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
666 enum zti_modes mode = ztip->zti_mode;
667 uint_t value = ztip->zti_value;
668 uint_t flags = TASKQ_PREPOPULATE;
671 if (t == ZIO_TYPE_WRITE)
672 flags |= TASKQ_NORECLAIM;
674 (void) snprintf(name, sizeof (name),
675 "%s_%s", zio_type_name[t], zio_taskq_types[q]);
677 spa->spa_zio_taskq[t][q] =
678 spa_taskq_create(spa, name, mode, value, flags);
683 #if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
685 spa_thread(void *arg)
690 user_t *pu = PTOU(curproc);
692 CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
695 ASSERT(curproc != &p0);
696 (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
697 "zpool-%s", spa->spa_name);
698 (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
700 /* bind this thread to the requested psrset */
701 if (zio_taskq_psrset_bind != PS_NONE) {
703 mutex_enter(&cpu_lock);
704 mutex_enter(&pidlock);
705 mutex_enter(&curproc->p_lock);
707 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
708 0, NULL, NULL) == 0) {
709 curthread->t_bind_pset = zio_taskq_psrset_bind;
712 "Couldn't bind process for zfs pool \"%s\" to "
713 "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
716 mutex_exit(&curproc->p_lock);
717 mutex_exit(&pidlock);
718 mutex_exit(&cpu_lock);
722 if (zio_taskq_sysdc) {
723 sysdc_thread_enter(curthread, 100, 0);
726 spa->spa_proc = curproc;
727 spa->spa_did = curthread->t_did;
729 spa_create_zio_taskqs(spa);
731 mutex_enter(&spa->spa_proc_lock);
732 ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
734 spa->spa_proc_state = SPA_PROC_ACTIVE;
735 cv_broadcast(&spa->spa_proc_cv);
737 CALLB_CPR_SAFE_BEGIN(&cprinfo);
738 while (spa->spa_proc_state == SPA_PROC_ACTIVE)
739 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
740 CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
742 ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
743 spa->spa_proc_state = SPA_PROC_GONE;
745 cv_broadcast(&spa->spa_proc_cv);
746 CALLB_CPR_EXIT(&cprinfo); /* drops spa_proc_lock */
748 mutex_enter(&curproc->p_lock);
754 * Activate an uninitialized pool.
757 spa_activate(spa_t *spa, int mode)
759 ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
761 spa->spa_state = POOL_STATE_ACTIVE;
762 spa->spa_mode = mode;
764 spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
765 spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
767 /* Try to create a covering process */
768 mutex_enter(&spa->spa_proc_lock);
769 ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
770 ASSERT(spa->spa_proc == &p0);
773 #ifdef HAVE_SPA_THREAD
774 /* Only create a process if we're going to be around a while. */
775 if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
776 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
778 spa->spa_proc_state = SPA_PROC_CREATED;
779 while (spa->spa_proc_state == SPA_PROC_CREATED) {
780 cv_wait(&spa->spa_proc_cv,
781 &spa->spa_proc_lock);
783 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
784 ASSERT(spa->spa_proc != &p0);
785 ASSERT(spa->spa_did != 0);
789 "Couldn't create process for zfs pool \"%s\"\n",
794 #endif /* HAVE_SPA_THREAD */
795 mutex_exit(&spa->spa_proc_lock);
797 /* If we didn't create a process, we need to create our taskqs. */
798 if (spa->spa_proc == &p0) {
799 spa_create_zio_taskqs(spa);
802 list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
803 offsetof(vdev_t, vdev_config_dirty_node));
804 list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
805 offsetof(vdev_t, vdev_state_dirty_node));
807 txg_list_create(&spa->spa_vdev_txg_list,
808 offsetof(struct vdev, vdev_txg_node));
810 avl_create(&spa->spa_errlist_scrub,
811 spa_error_entry_compare, sizeof (spa_error_entry_t),
812 offsetof(spa_error_entry_t, se_avl));
813 avl_create(&spa->spa_errlist_last,
814 spa_error_entry_compare, sizeof (spa_error_entry_t),
815 offsetof(spa_error_entry_t, se_avl));
819 * Opposite of spa_activate().
822 spa_deactivate(spa_t *spa)
826 ASSERT(spa->spa_sync_on == B_FALSE);
827 ASSERT(spa->spa_dsl_pool == NULL);
828 ASSERT(spa->spa_root_vdev == NULL);
829 ASSERT(spa->spa_async_zio_root == NULL);
830 ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
832 txg_list_destroy(&spa->spa_vdev_txg_list);
834 list_destroy(&spa->spa_config_dirty_list);
835 list_destroy(&spa->spa_state_dirty_list);
837 for (t = 0; t < ZIO_TYPES; t++) {
838 for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
839 if (spa->spa_zio_taskq[t][q] != NULL)
840 taskq_destroy(spa->spa_zio_taskq[t][q]);
841 spa->spa_zio_taskq[t][q] = NULL;
845 metaslab_class_destroy(spa->spa_normal_class);
846 spa->spa_normal_class = NULL;
848 metaslab_class_destroy(spa->spa_log_class);
849 spa->spa_log_class = NULL;
852 * If this was part of an import or the open otherwise failed, we may
853 * still have errors left in the queues. Empty them just in case.
855 spa_errlog_drain(spa);
857 avl_destroy(&spa->spa_errlist_scrub);
858 avl_destroy(&spa->spa_errlist_last);
860 spa->spa_state = POOL_STATE_UNINITIALIZED;
862 mutex_enter(&spa->spa_proc_lock);
863 if (spa->spa_proc_state != SPA_PROC_NONE) {
864 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
865 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
866 cv_broadcast(&spa->spa_proc_cv);
867 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
868 ASSERT(spa->spa_proc != &p0);
869 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
871 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
872 spa->spa_proc_state = SPA_PROC_NONE;
874 ASSERT(spa->spa_proc == &p0);
875 mutex_exit(&spa->spa_proc_lock);
878 * We want to make sure spa_thread() has actually exited the ZFS
879 * module, so that the module can't be unloaded out from underneath
882 if (spa->spa_did != 0) {
883 thread_join(spa->spa_did);
889 * Verify a pool configuration, and construct the vdev tree appropriately. This
890 * will create all the necessary vdevs in the appropriate layout, with each vdev
891 * in the CLOSED state. This will prep the pool before open/creation/import.
892 * All vdev validation is done by the vdev_alloc() routine.
895 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
896 uint_t id, int atype)
903 if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
906 if ((*vdp)->vdev_ops->vdev_op_leaf)
909 error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
921 for (c = 0; c < children; c++) {
923 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
931 ASSERT(*vdp != NULL);
937 * Opposite of spa_load().
940 spa_unload(spa_t *spa)
944 ASSERT(MUTEX_HELD(&spa_namespace_lock));
949 spa_async_suspend(spa);
954 if (spa->spa_sync_on) {
955 txg_sync_stop(spa->spa_dsl_pool);
956 spa->spa_sync_on = B_FALSE;
960 * Wait for any outstanding async I/O to complete.
962 if (spa->spa_async_zio_root != NULL) {
963 (void) zio_wait(spa->spa_async_zio_root);
964 spa->spa_async_zio_root = NULL;
967 bpobj_close(&spa->spa_deferred_bpobj);
970 * Close the dsl pool.
972 if (spa->spa_dsl_pool) {
973 dsl_pool_close(spa->spa_dsl_pool);
974 spa->spa_dsl_pool = NULL;
975 spa->spa_meta_objset = NULL;
980 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
983 * Drop and purge level 2 cache
985 spa_l2cache_drop(spa);
990 if (spa->spa_root_vdev)
991 vdev_free(spa->spa_root_vdev);
992 ASSERT(spa->spa_root_vdev == NULL);
994 for (i = 0; i < spa->spa_spares.sav_count; i++)
995 vdev_free(spa->spa_spares.sav_vdevs[i]);
996 if (spa->spa_spares.sav_vdevs) {
997 kmem_free(spa->spa_spares.sav_vdevs,
998 spa->spa_spares.sav_count * sizeof (void *));
999 spa->spa_spares.sav_vdevs = NULL;
1001 if (spa->spa_spares.sav_config) {
1002 nvlist_free(spa->spa_spares.sav_config);
1003 spa->spa_spares.sav_config = NULL;
1005 spa->spa_spares.sav_count = 0;
1007 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
1008 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1009 if (spa->spa_l2cache.sav_vdevs) {
1010 kmem_free(spa->spa_l2cache.sav_vdevs,
1011 spa->spa_l2cache.sav_count * sizeof (void *));
1012 spa->spa_l2cache.sav_vdevs = NULL;
1014 if (spa->spa_l2cache.sav_config) {
1015 nvlist_free(spa->spa_l2cache.sav_config);
1016 spa->spa_l2cache.sav_config = NULL;
1018 spa->spa_l2cache.sav_count = 0;
1020 spa->spa_async_suspended = 0;
1022 spa_config_exit(spa, SCL_ALL, FTAG);
1026 * Load (or re-load) the current list of vdevs describing the active spares for
1027 * this pool. When this is called, we have some form of basic information in
1028 * 'spa_spares.sav_config'. We parse this into vdevs, try to open them, and
1029 * then re-generate a more complete list including status information.
1032 spa_load_spares(spa_t *spa)
1039 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1042 * First, close and free any existing spare vdevs.
1044 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1045 vd = spa->spa_spares.sav_vdevs[i];
1047 /* Undo the call to spa_activate() below */
1048 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1049 B_FALSE)) != NULL && tvd->vdev_isspare)
1050 spa_spare_remove(tvd);
1055 if (spa->spa_spares.sav_vdevs)
1056 kmem_free(spa->spa_spares.sav_vdevs,
1057 spa->spa_spares.sav_count * sizeof (void *));
1059 if (spa->spa_spares.sav_config == NULL)
1062 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1063 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1065 spa->spa_spares.sav_count = (int)nspares;
1066 spa->spa_spares.sav_vdevs = NULL;
1072 * Construct the array of vdevs, opening them to get status in the
1073 * process. For each spare, there is potentially two different vdev_t
1074 * structures associated with it: one in the list of spares (used only
1075 * for basic validation purposes) and one in the active vdev
1076 * configuration (if it's spared in). During this phase we open and
1077 * validate each vdev on the spare list. If the vdev also exists in the
1078 * active configuration, then we also mark this vdev as an active spare.
1080 spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1082 for (i = 0; i < spa->spa_spares.sav_count; i++) {
1083 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1084 VDEV_ALLOC_SPARE) == 0);
1087 spa->spa_spares.sav_vdevs[i] = vd;
1089 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1090 B_FALSE)) != NULL) {
1091 if (!tvd->vdev_isspare)
1095 * We only mark the spare active if we were successfully
1096 * able to load the vdev. Otherwise, importing a pool
1097 * with a bad active spare would result in strange
1098 * behavior, because multiple pool would think the spare
1099 * is actively in use.
1101 * There is a vulnerability here to an equally bizarre
1102 * circumstance, where a dead active spare is later
1103 * brought back to life (onlined or otherwise). Given
1104 * the rarity of this scenario, and the extra complexity
1105 * it adds, we ignore the possibility.
1107 if (!vdev_is_dead(tvd))
1108 spa_spare_activate(tvd);
1112 vd->vdev_aux = &spa->spa_spares;
1114 if (vdev_open(vd) != 0)
1117 if (vdev_validate_aux(vd) == 0)
1122 * Recompute the stashed list of spares, with status information
1125 VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1126 DATA_TYPE_NVLIST_ARRAY) == 0);
1128 spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1130 for (i = 0; i < spa->spa_spares.sav_count; i++)
1131 spares[i] = vdev_config_generate(spa,
1132 spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1133 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1134 ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1135 for (i = 0; i < spa->spa_spares.sav_count; i++)
1136 nvlist_free(spares[i]);
1137 kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1141 * Load (or re-load) the current list of vdevs describing the active l2cache for
1142 * this pool. When this is called, we have some form of basic information in
1143 * 'spa_l2cache.sav_config'. We parse this into vdevs, try to open them, and
1144 * then re-generate a more complete list including status information.
1145 * Devices which are already active have their details maintained, and are
1149 spa_load_l2cache(spa_t *spa)
1153 int i, j, oldnvdevs;
1155 vdev_t *vd, **oldvdevs, **newvdevs = NULL;
1156 spa_aux_vdev_t *sav = &spa->spa_l2cache;
1158 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1160 if (sav->sav_config != NULL) {
1161 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1162 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1163 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1168 oldvdevs = sav->sav_vdevs;
1169 oldnvdevs = sav->sav_count;
1170 sav->sav_vdevs = NULL;
1174 * Process new nvlist of vdevs.
1176 for (i = 0; i < nl2cache; i++) {
1177 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1181 for (j = 0; j < oldnvdevs; j++) {
1183 if (vd != NULL && guid == vd->vdev_guid) {
1185 * Retain previous vdev for add/remove ops.
1193 if (newvdevs[i] == NULL) {
1197 VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1198 VDEV_ALLOC_L2CACHE) == 0);
1203 * Commit this vdev as an l2cache device,
1204 * even if it fails to open.
1206 spa_l2cache_add(vd);
1211 spa_l2cache_activate(vd);
1213 if (vdev_open(vd) != 0)
1216 (void) vdev_validate_aux(vd);
1218 if (!vdev_is_dead(vd))
1219 l2arc_add_vdev(spa, vd);
1224 * Purge vdevs that were dropped
1226 for (i = 0; i < oldnvdevs; i++) {
1231 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1232 pool != 0ULL && l2arc_vdev_present(vd))
1233 l2arc_remove_vdev(vd);
1234 (void) vdev_close(vd);
1235 spa_l2cache_remove(vd);
1240 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1242 if (sav->sav_config == NULL)
1245 sav->sav_vdevs = newvdevs;
1246 sav->sav_count = (int)nl2cache;
1249 * Recompute the stashed list of l2cache devices, with status
1250 * information this time.
1252 VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1253 DATA_TYPE_NVLIST_ARRAY) == 0);
1255 l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1256 for (i = 0; i < sav->sav_count; i++)
1257 l2cache[i] = vdev_config_generate(spa,
1258 sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1259 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1260 ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1262 for (i = 0; i < sav->sav_count; i++)
1263 nvlist_free(l2cache[i]);
1265 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1269 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1272 char *packed = NULL;
1277 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1278 nvsize = *(uint64_t *)db->db_data;
1279 dmu_buf_rele(db, FTAG);
1281 packed = kmem_alloc(nvsize, KM_SLEEP | KM_NODEBUG);
1282 error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1285 error = nvlist_unpack(packed, nvsize, value, 0);
1286 kmem_free(packed, nvsize);
1292 * Checks to see if the given vdev could not be opened, in which case we post a
1293 * sysevent to notify the autoreplace code that the device has been removed.
1296 spa_check_removed(vdev_t *vd)
1300 for (c = 0; c < vd->vdev_children; c++)
1301 spa_check_removed(vd->vdev_child[c]);
1303 if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1304 zfs_ereport_post(FM_EREPORT_RESOURCE_AUTOREPLACE,
1305 vd->vdev_spa, vd, NULL, 0, 0);
1306 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_CHECK);
1311 * Validate the current config against the MOS config
1314 spa_config_valid(spa_t *spa, nvlist_t *config)
1316 vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1320 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1322 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1323 VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1325 ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1328 * If we're doing a normal import, then build up any additional
1329 * diagnostic information about missing devices in this config.
1330 * We'll pass this up to the user for further processing.
1332 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1333 nvlist_t **child, *nv;
1336 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1338 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1340 for (c = 0; c < rvd->vdev_children; c++) {
1341 vdev_t *tvd = rvd->vdev_child[c];
1342 vdev_t *mtvd = mrvd->vdev_child[c];
1344 if (tvd->vdev_ops == &vdev_missing_ops &&
1345 mtvd->vdev_ops != &vdev_missing_ops &&
1347 child[idx++] = vdev_config_generate(spa, mtvd,
1352 VERIFY(nvlist_add_nvlist_array(nv,
1353 ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1354 VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1355 ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1357 for (i = 0; i < idx; i++)
1358 nvlist_free(child[i]);
1361 kmem_free(child, rvd->vdev_children * sizeof (char **));
1365 * Compare the root vdev tree with the information we have
1366 * from the MOS config (mrvd). Check each top-level vdev
1367 * with the corresponding MOS config top-level (mtvd).
1369 for (c = 0; c < rvd->vdev_children; c++) {
1370 vdev_t *tvd = rvd->vdev_child[c];
1371 vdev_t *mtvd = mrvd->vdev_child[c];
1374 * Resolve any "missing" vdevs in the current configuration.
1375 * If we find that the MOS config has more accurate information
1376 * about the top-level vdev then use that vdev instead.
1378 if (tvd->vdev_ops == &vdev_missing_ops &&
1379 mtvd->vdev_ops != &vdev_missing_ops) {
1381 if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1385 * Device specific actions.
1387 if (mtvd->vdev_islog) {
1388 spa_set_log_state(spa, SPA_LOG_CLEAR);
1391 * XXX - once we have 'readonly' pool
1392 * support we should be able to handle
1393 * missing data devices by transitioning
1394 * the pool to readonly.
1400 * Swap the missing vdev with the data we were
1401 * able to obtain from the MOS config.
1403 vdev_remove_child(rvd, tvd);
1404 vdev_remove_child(mrvd, mtvd);
1406 vdev_add_child(rvd, mtvd);
1407 vdev_add_child(mrvd, tvd);
1409 spa_config_exit(spa, SCL_ALL, FTAG);
1411 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1414 } else if (mtvd->vdev_islog) {
1416 * Load the slog device's state from the MOS config
1417 * since it's possible that the label does not
1418 * contain the most up-to-date information.
1420 vdev_load_log_state(tvd, mtvd);
1425 spa_config_exit(spa, SCL_ALL, FTAG);
1428 * Ensure we were able to validate the config.
1430 return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1434 * Check for missing log devices
1437 spa_check_logs(spa_t *spa)
1439 switch (spa->spa_log_state) {
1442 case SPA_LOG_MISSING:
1443 /* need to recheck in case slog has been restored */
1444 case SPA_LOG_UNKNOWN:
1445 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1446 DS_FIND_CHILDREN)) {
1447 spa_set_log_state(spa, SPA_LOG_MISSING);
1456 spa_passivate_log(spa_t *spa)
1458 vdev_t *rvd = spa->spa_root_vdev;
1459 boolean_t slog_found = B_FALSE;
1462 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1464 if (!spa_has_slogs(spa))
1467 for (c = 0; c < rvd->vdev_children; c++) {
1468 vdev_t *tvd = rvd->vdev_child[c];
1469 metaslab_group_t *mg = tvd->vdev_mg;
1471 if (tvd->vdev_islog) {
1472 metaslab_group_passivate(mg);
1473 slog_found = B_TRUE;
1477 return (slog_found);
1481 spa_activate_log(spa_t *spa)
1483 vdev_t *rvd = spa->spa_root_vdev;
1486 ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1488 for (c = 0; c < rvd->vdev_children; c++) {
1489 vdev_t *tvd = rvd->vdev_child[c];
1490 metaslab_group_t *mg = tvd->vdev_mg;
1492 if (tvd->vdev_islog)
1493 metaslab_group_activate(mg);
1498 spa_offline_log(spa_t *spa)
1502 if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1503 NULL, DS_FIND_CHILDREN)) == 0) {
1506 * We successfully offlined the log device, sync out the
1507 * current txg so that the "stubby" block can be removed
1510 txg_wait_synced(spa->spa_dsl_pool, 0);
1516 spa_aux_check_removed(spa_aux_vdev_t *sav)
1520 for (i = 0; i < sav->sav_count; i++)
1521 spa_check_removed(sav->sav_vdevs[i]);
1525 spa_claim_notify(zio_t *zio)
1527 spa_t *spa = zio->io_spa;
1532 mutex_enter(&spa->spa_props_lock); /* any mutex will do */
1533 if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1534 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1535 mutex_exit(&spa->spa_props_lock);
1538 typedef struct spa_load_error {
1539 uint64_t sle_meta_count;
1540 uint64_t sle_data_count;
1544 spa_load_verify_done(zio_t *zio)
1546 blkptr_t *bp = zio->io_bp;
1547 spa_load_error_t *sle = zio->io_private;
1548 dmu_object_type_t type = BP_GET_TYPE(bp);
1549 int error = zio->io_error;
1552 if ((BP_GET_LEVEL(bp) != 0 || dmu_ot[type].ot_metadata) &&
1553 type != DMU_OT_INTENT_LOG)
1554 atomic_add_64(&sle->sle_meta_count, 1);
1556 atomic_add_64(&sle->sle_data_count, 1);
1558 zio_data_buf_free(zio->io_data, zio->io_size);
1563 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1564 arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1568 size_t size = BP_GET_PSIZE(bp);
1569 void *data = zio_data_buf_alloc(size);
1571 zio_nowait(zio_read(rio, spa, bp, data, size,
1572 spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1573 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1574 ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1580 spa_load_verify(spa_t *spa)
1583 spa_load_error_t sle = { 0 };
1584 zpool_rewind_policy_t policy;
1585 boolean_t verify_ok = B_FALSE;
1588 zpool_get_rewind_policy(spa->spa_config, &policy);
1590 if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1593 rio = zio_root(spa, NULL, &sle,
1594 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1596 error = traverse_pool(spa, spa->spa_verify_min_txg,
1597 TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1599 (void) zio_wait(rio);
1601 spa->spa_load_meta_errors = sle.sle_meta_count;
1602 spa->spa_load_data_errors = sle.sle_data_count;
1604 if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1605 sle.sle_data_count <= policy.zrp_maxdata) {
1609 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1610 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1612 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1613 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1614 ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1615 VERIFY(nvlist_add_int64(spa->spa_load_info,
1616 ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1617 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1618 ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1620 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1624 if (error != ENXIO && error != EIO)
1629 return (verify_ok ? 0 : EIO);
1633 * Find a value in the pool props object.
1636 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1638 (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1639 zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1643 * Find a value in the pool directory object.
1646 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1648 return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1649 name, sizeof (uint64_t), 1, val));
1653 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1655 vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1660 * Fix up config after a partly-completed split. This is done with the
1661 * ZPOOL_CONFIG_SPLIT nvlist. Both the splitting pool and the split-off
1662 * pool have that entry in their config, but only the splitting one contains
1663 * a list of all the guids of the vdevs that are being split off.
1665 * This function determines what to do with that list: either rejoin
1666 * all the disks to the pool, or complete the splitting process. To attempt
1667 * the rejoin, each disk that is offlined is marked online again, and
1668 * we do a reopen() call. If the vdev label for every disk that was
1669 * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1670 * then we call vdev_split() on each disk, and complete the split.
1672 * Otherwise we leave the config alone, with all the vdevs in place in
1673 * the original pool.
1676 spa_try_repair(spa_t *spa, nvlist_t *config)
1683 boolean_t attempt_reopen;
1685 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1688 /* check that the config is complete */
1689 if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1690 &glist, &gcount) != 0)
1693 vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1695 /* attempt to online all the vdevs & validate */
1696 attempt_reopen = B_TRUE;
1697 for (i = 0; i < gcount; i++) {
1698 if (glist[i] == 0) /* vdev is hole */
1701 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1702 if (vd[i] == NULL) {
1704 * Don't bother attempting to reopen the disks;
1705 * just do the split.
1707 attempt_reopen = B_FALSE;
1709 /* attempt to re-online it */
1710 vd[i]->vdev_offline = B_FALSE;
1714 if (attempt_reopen) {
1715 vdev_reopen(spa->spa_root_vdev);
1717 /* check each device to see what state it's in */
1718 for (extracted = 0, i = 0; i < gcount; i++) {
1719 if (vd[i] != NULL &&
1720 vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1727 * If every disk has been moved to the new pool, or if we never
1728 * even attempted to look at them, then we split them off for
1731 if (!attempt_reopen || gcount == extracted) {
1732 for (i = 0; i < gcount; i++)
1735 vdev_reopen(spa->spa_root_vdev);
1738 kmem_free(vd, gcount * sizeof (vdev_t *));
1742 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1743 boolean_t mosconfig)
1745 nvlist_t *config = spa->spa_config;
1746 char *ereport = FM_EREPORT_ZFS_POOL;
1751 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1755 * Versioning wasn't explicitly added to the label until later, so if
1756 * it's not present treat it as the initial version.
1758 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1759 &spa->spa_ubsync.ub_version) != 0)
1760 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1762 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1763 &spa->spa_config_txg);
1765 if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1766 spa_guid_exists(pool_guid, 0)) {
1769 spa->spa_load_guid = pool_guid;
1771 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1773 VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1777 gethrestime(&spa->spa_loaded_ts);
1778 error = spa_load_impl(spa, pool_guid, config, state, type,
1779 mosconfig, &ereport);
1782 spa->spa_minref = refcount_count(&spa->spa_refcount);
1784 if (error != EEXIST) {
1785 spa->spa_loaded_ts.tv_sec = 0;
1786 spa->spa_loaded_ts.tv_nsec = 0;
1788 if (error != EBADF) {
1789 zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1792 spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1799 * Load an existing storage pool, using the pool's builtin spa_config as a
1800 * source of configuration information.
1802 __attribute__((always_inline))
1804 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1805 spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1809 nvlist_t *nvroot = NULL;
1811 uberblock_t *ub = &spa->spa_uberblock;
1812 uint64_t children, config_cache_txg = spa->spa_config_txg;
1813 int orig_mode = spa->spa_mode;
1818 * If this is an untrusted config, access the pool in read-only mode.
1819 * This prevents things like resilvering recently removed devices.
1822 spa->spa_mode = FREAD;
1824 ASSERT(MUTEX_HELD(&spa_namespace_lock));
1826 spa->spa_load_state = state;
1828 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1831 parse = (type == SPA_IMPORT_EXISTING ?
1832 VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1835 * Create "The Godfather" zio to hold all async IOs
1837 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1838 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1841 * Parse the configuration into a vdev tree. We explicitly set the
1842 * value that will be returned by spa_version() since parsing the
1843 * configuration requires knowing the version number.
1845 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1846 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1847 spa_config_exit(spa, SCL_ALL, FTAG);
1852 ASSERT(spa->spa_root_vdev == rvd);
1854 if (type != SPA_IMPORT_ASSEMBLE) {
1855 ASSERT(spa_guid(spa) == pool_guid);
1859 * Try to open all vdevs, loading each label in the process.
1861 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1862 error = vdev_open(rvd);
1863 spa_config_exit(spa, SCL_ALL, FTAG);
1868 * We need to validate the vdev labels against the configuration that
1869 * we have in hand, which is dependent on the setting of mosconfig. If
1870 * mosconfig is true then we're validating the vdev labels based on
1871 * that config. Otherwise, we're validating against the cached config
1872 * (zpool.cache) that was read when we loaded the zfs module, and then
1873 * later we will recursively call spa_load() and validate against
1876 * If we're assembling a new pool that's been split off from an
1877 * existing pool, the labels haven't yet been updated so we skip
1878 * validation for now.
1880 if (type != SPA_IMPORT_ASSEMBLE) {
1881 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1882 error = vdev_validate(rvd);
1883 spa_config_exit(spa, SCL_ALL, FTAG);
1888 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
1893 * Find the best uberblock.
1895 vdev_uberblock_load(NULL, rvd, ub);
1898 * If we weren't able to find a single valid uberblock, return failure.
1900 if (ub->ub_txg == 0)
1901 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
1904 * If the pool is newer than the code, we can't open it.
1906 if (ub->ub_version > SPA_VERSION)
1907 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
1910 * If the vdev guid sum doesn't match the uberblock, we have an
1911 * incomplete configuration. We first check to see if the pool
1912 * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
1913 * If it is, defer the vdev_guid_sum check till later so we
1914 * can handle missing vdevs.
1916 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
1917 &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
1918 rvd->vdev_guid_sum != ub->ub_guid_sum)
1919 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
1921 if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
1922 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1923 spa_try_repair(spa, config);
1924 spa_config_exit(spa, SCL_ALL, FTAG);
1925 nvlist_free(spa->spa_config_splitting);
1926 spa->spa_config_splitting = NULL;
1930 * Initialize internal SPA structures.
1932 spa->spa_state = POOL_STATE_ACTIVE;
1933 spa->spa_ubsync = spa->spa_uberblock;
1934 spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
1935 TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
1936 spa->spa_first_txg = spa->spa_last_ubsync_txg ?
1937 spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
1938 spa->spa_claim_max_txg = spa->spa_first_txg;
1939 spa->spa_prev_software_version = ub->ub_software_version;
1941 error = dsl_pool_open(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
1943 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1944 spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
1946 if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
1947 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1951 nvlist_t *policy = NULL, *nvconfig;
1953 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
1954 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
1956 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
1957 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
1959 unsigned long myhostid = 0;
1961 VERIFY(nvlist_lookup_string(nvconfig,
1962 ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
1965 myhostid = zone_get_hostid(NULL);
1968 * We're emulating the system's hostid in userland, so
1969 * we can't use zone_get_hostid().
1971 (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
1972 #endif /* _KERNEL */
1973 if (hostid != 0 && myhostid != 0 &&
1974 hostid != myhostid) {
1975 nvlist_free(nvconfig);
1976 cmn_err(CE_WARN, "pool '%s' could not be "
1977 "loaded as it was last accessed by "
1978 "another system (host: %s hostid: 0x%lx). "
1979 "See: http://www.sun.com/msg/ZFS-8000-EY",
1980 spa_name(spa), hostname,
1981 (unsigned long)hostid);
1985 if (nvlist_lookup_nvlist(spa->spa_config,
1986 ZPOOL_REWIND_POLICY, &policy) == 0)
1987 VERIFY(nvlist_add_nvlist(nvconfig,
1988 ZPOOL_REWIND_POLICY, policy) == 0);
1990 spa_config_set(spa, nvconfig);
1992 spa_deactivate(spa);
1993 spa_activate(spa, orig_mode);
1995 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
1998 if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
1999 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2000 error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2002 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2005 * Load the bit that tells us to use the new accounting function
2006 * (raid-z deflation). If we have an older pool, this will not
2009 error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2010 if (error != 0 && error != ENOENT)
2011 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2013 error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2014 &spa->spa_creation_version);
2015 if (error != 0 && error != ENOENT)
2016 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2019 * Load the persistent error log. If we have an older pool, this will
2022 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2023 if (error != 0 && error != ENOENT)
2024 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2026 error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2027 &spa->spa_errlog_scrub);
2028 if (error != 0 && error != ENOENT)
2029 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2032 * Load the history object. If we have an older pool, this
2033 * will not be present.
2035 error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2036 if (error != 0 && error != ENOENT)
2037 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2040 * If we're assembling the pool from the split-off vdevs of
2041 * an existing pool, we don't want to attach the spares & cache
2046 * Load any hot spares for this pool.
2048 error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2049 if (error != 0 && error != ENOENT)
2050 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2051 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2052 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2053 if (load_nvlist(spa, spa->spa_spares.sav_object,
2054 &spa->spa_spares.sav_config) != 0)
2055 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2057 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2058 spa_load_spares(spa);
2059 spa_config_exit(spa, SCL_ALL, FTAG);
2060 } else if (error == 0) {
2061 spa->spa_spares.sav_sync = B_TRUE;
2065 * Load any level 2 ARC devices for this pool.
2067 error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2068 &spa->spa_l2cache.sav_object);
2069 if (error != 0 && error != ENOENT)
2070 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2071 if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2072 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2073 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2074 &spa->spa_l2cache.sav_config) != 0)
2075 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2077 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2078 spa_load_l2cache(spa);
2079 spa_config_exit(spa, SCL_ALL, FTAG);
2080 } else if (error == 0) {
2081 spa->spa_l2cache.sav_sync = B_TRUE;
2084 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2086 error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2087 if (error && error != ENOENT)
2088 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2091 uint64_t autoreplace;
2093 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2094 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2095 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2096 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2097 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2098 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2099 &spa->spa_dedup_ditto);
2101 spa->spa_autoreplace = (autoreplace != 0);
2105 * If the 'autoreplace' property is set, then post a resource notifying
2106 * the ZFS DE that it should not issue any faults for unopenable
2107 * devices. We also iterate over the vdevs, and post a sysevent for any
2108 * unopenable vdevs so that the normal autoreplace handler can take
2111 if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2112 spa_check_removed(spa->spa_root_vdev);
2114 * For the import case, this is done in spa_import(), because
2115 * at this point we're using the spare definitions from
2116 * the MOS config, not necessarily from the userland config.
2118 if (state != SPA_LOAD_IMPORT) {
2119 spa_aux_check_removed(&spa->spa_spares);
2120 spa_aux_check_removed(&spa->spa_l2cache);
2125 * Load the vdev state for all toplevel vdevs.
2130 * Propagate the leaf DTLs we just loaded all the way up the tree.
2132 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2133 vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2134 spa_config_exit(spa, SCL_ALL, FTAG);
2137 * Load the DDTs (dedup tables).
2139 error = ddt_load(spa);
2141 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2143 spa_update_dspace(spa);
2146 * Validate the config, using the MOS config to fill in any
2147 * information which might be missing. If we fail to validate
2148 * the config then declare the pool unfit for use. If we're
2149 * assembling a pool from a split, the log is not transferred
2152 if (type != SPA_IMPORT_ASSEMBLE) {
2155 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2156 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2158 if (!spa_config_valid(spa, nvconfig)) {
2159 nvlist_free(nvconfig);
2160 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2163 nvlist_free(nvconfig);
2166 * Now that we've validate the config, check the state of the
2167 * root vdev. If it can't be opened, it indicates one or
2168 * more toplevel vdevs are faulted.
2170 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2173 if (spa_check_logs(spa)) {
2174 *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2175 return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2180 * We've successfully opened the pool, verify that we're ready
2181 * to start pushing transactions.
2183 if (state != SPA_LOAD_TRYIMPORT) {
2184 if ((error = spa_load_verify(spa)))
2185 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2189 if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2190 spa->spa_load_max_txg == UINT64_MAX)) {
2192 int need_update = B_FALSE;
2195 ASSERT(state != SPA_LOAD_TRYIMPORT);
2198 * Claim log blocks that haven't been committed yet.
2199 * This must all happen in a single txg.
2200 * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2201 * invoked from zil_claim_log_block()'s i/o done callback.
2202 * Price of rollback is that we abandon the log.
2204 spa->spa_claiming = B_TRUE;
2206 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2207 spa_first_txg(spa));
2208 (void) dmu_objset_find(spa_name(spa),
2209 zil_claim, tx, DS_FIND_CHILDREN);
2212 spa->spa_claiming = B_FALSE;
2214 spa_set_log_state(spa, SPA_LOG_GOOD);
2215 spa->spa_sync_on = B_TRUE;
2216 txg_sync_start(spa->spa_dsl_pool);
2219 * Wait for all claims to sync. We sync up to the highest
2220 * claimed log block birth time so that claimed log blocks
2221 * don't appear to be from the future. spa_claim_max_txg
2222 * will have been set for us by either zil_check_log_chain()
2223 * (invoked from spa_check_logs()) or zil_claim() above.
2225 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2228 * If the config cache is stale, or we have uninitialized
2229 * metaslabs (see spa_vdev_add()), then update the config.
2231 * If this is a verbatim import, trust the current
2232 * in-core spa_config and update the disk labels.
2234 if (config_cache_txg != spa->spa_config_txg ||
2235 state == SPA_LOAD_IMPORT ||
2236 state == SPA_LOAD_RECOVER ||
2237 (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2238 need_update = B_TRUE;
2240 for (c = 0; c < rvd->vdev_children; c++)
2241 if (rvd->vdev_child[c]->vdev_ms_array == 0)
2242 need_update = B_TRUE;
2245 * Update the config cache asychronously in case we're the
2246 * root pool, in which case the config cache isn't writable yet.
2249 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2252 * Check all DTLs to see if anything needs resilvering.
2254 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2255 vdev_resilver_needed(rvd, NULL, NULL))
2256 spa_async_request(spa, SPA_ASYNC_RESILVER);
2259 * Delete any inconsistent datasets.
2261 (void) dmu_objset_find(spa_name(spa),
2262 dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2265 * Clean up any stale temporary dataset userrefs.
2267 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2274 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2276 int mode = spa->spa_mode;
2279 spa_deactivate(spa);
2281 spa->spa_load_max_txg--;
2283 spa_activate(spa, mode);
2284 spa_async_suspend(spa);
2286 return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2290 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2291 uint64_t max_request, int rewind_flags)
2293 nvlist_t *config = NULL;
2294 int load_error, rewind_error;
2295 uint64_t safe_rewind_txg;
2298 if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2299 spa->spa_load_max_txg = spa->spa_load_txg;
2300 spa_set_log_state(spa, SPA_LOG_CLEAR);
2302 spa->spa_load_max_txg = max_request;
2305 load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2307 if (load_error == 0)
2310 if (spa->spa_root_vdev != NULL)
2311 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2313 spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2314 spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2316 if (rewind_flags & ZPOOL_NEVER_REWIND) {
2317 nvlist_free(config);
2318 return (load_error);
2321 /* Price of rolling back is discarding txgs, including log */
2322 if (state == SPA_LOAD_RECOVER)
2323 spa_set_log_state(spa, SPA_LOG_CLEAR);
2325 spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2326 safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2327 min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2328 TXG_INITIAL : safe_rewind_txg;
2331 * Continue as long as we're finding errors, we're still within
2332 * the acceptable rewind range, and we're still finding uberblocks
2334 while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2335 spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2336 if (spa->spa_load_max_txg < safe_rewind_txg)
2337 spa->spa_extreme_rewind = B_TRUE;
2338 rewind_error = spa_load_retry(spa, state, mosconfig);
2341 spa->spa_extreme_rewind = B_FALSE;
2342 spa->spa_load_max_txg = UINT64_MAX;
2344 if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2345 spa_config_set(spa, config);
2347 return (state == SPA_LOAD_RECOVER ? rewind_error : load_error);
2353 * The import case is identical to an open except that the configuration is sent
2354 * down from userland, instead of grabbed from the configuration cache. For the
2355 * case of an open, the pool configuration will exist in the
2356 * POOL_STATE_UNINITIALIZED state.
2358 * The stats information (gen/count/ustats) is used to gather vdev statistics at
2359 * the same time open the pool, without having to keep around the spa_t in some
2363 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2367 spa_load_state_t state = SPA_LOAD_OPEN;
2369 int locked = B_FALSE;
2374 * As disgusting as this is, we need to support recursive calls to this
2375 * function because dsl_dir_open() is called during spa_load(), and ends
2376 * up calling spa_open() again. The real fix is to figure out how to
2377 * avoid dsl_dir_open() calling this in the first place.
2379 if (mutex_owner(&spa_namespace_lock) != curthread) {
2380 mutex_enter(&spa_namespace_lock);
2384 if ((spa = spa_lookup(pool)) == NULL) {
2386 mutex_exit(&spa_namespace_lock);
2390 if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2391 zpool_rewind_policy_t policy;
2393 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2395 if (policy.zrp_request & ZPOOL_DO_REWIND)
2396 state = SPA_LOAD_RECOVER;
2398 spa_activate(spa, spa_mode_global);
2400 if (state != SPA_LOAD_RECOVER)
2401 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2403 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2404 policy.zrp_request);
2406 if (error == EBADF) {
2408 * If vdev_validate() returns failure (indicated by
2409 * EBADF), it indicates that one of the vdevs indicates
2410 * that the pool has been exported or destroyed. If
2411 * this is the case, the config cache is out of sync and
2412 * we should remove the pool from the namespace.
2415 spa_deactivate(spa);
2416 spa_config_sync(spa, B_TRUE, B_TRUE);
2419 mutex_exit(&spa_namespace_lock);
2425 * We can't open the pool, but we still have useful
2426 * information: the state of each vdev after the
2427 * attempted vdev_open(). Return this to the user.
2429 if (config != NULL && spa->spa_config) {
2430 VERIFY(nvlist_dup(spa->spa_config, config,
2432 VERIFY(nvlist_add_nvlist(*config,
2433 ZPOOL_CONFIG_LOAD_INFO,
2434 spa->spa_load_info) == 0);
2437 spa_deactivate(spa);
2438 spa->spa_last_open_failed = error;
2440 mutex_exit(&spa_namespace_lock);
2446 spa_open_ref(spa, tag);
2449 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2452 * If we've recovered the pool, pass back any information we
2453 * gathered while doing the load.
2455 if (state == SPA_LOAD_RECOVER) {
2456 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2457 spa->spa_load_info) == 0);
2461 spa->spa_last_open_failed = 0;
2462 spa->spa_last_ubsync_txg = 0;
2463 spa->spa_load_txg = 0;
2464 mutex_exit(&spa_namespace_lock);
2473 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2476 return (spa_open_common(name, spapp, tag, policy, config));
2480 spa_open(const char *name, spa_t **spapp, void *tag)
2482 return (spa_open_common(name, spapp, tag, NULL, NULL));
2486 * Lookup the given spa_t, incrementing the inject count in the process,
2487 * preventing it from being exported or destroyed.
2490 spa_inject_addref(char *name)
2494 mutex_enter(&spa_namespace_lock);
2495 if ((spa = spa_lookup(name)) == NULL) {
2496 mutex_exit(&spa_namespace_lock);
2499 spa->spa_inject_ref++;
2500 mutex_exit(&spa_namespace_lock);
2506 spa_inject_delref(spa_t *spa)
2508 mutex_enter(&spa_namespace_lock);
2509 spa->spa_inject_ref--;
2510 mutex_exit(&spa_namespace_lock);
2514 * Add spares device information to the nvlist.
2517 spa_add_spares(spa_t *spa, nvlist_t *config)
2527 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2529 if (spa->spa_spares.sav_count == 0)
2532 VERIFY(nvlist_lookup_nvlist(config,
2533 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2534 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2535 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2537 VERIFY(nvlist_add_nvlist_array(nvroot,
2538 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2539 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2540 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2543 * Go through and find any spares which have since been
2544 * repurposed as an active spare. If this is the case, update
2545 * their status appropriately.
2547 for (i = 0; i < nspares; i++) {
2548 VERIFY(nvlist_lookup_uint64(spares[i],
2549 ZPOOL_CONFIG_GUID, &guid) == 0);
2550 if (spa_spare_exists(guid, &pool, NULL) &&
2552 VERIFY(nvlist_lookup_uint64_array(
2553 spares[i], ZPOOL_CONFIG_VDEV_STATS,
2554 (uint64_t **)&vs, &vsc) == 0);
2555 vs->vs_state = VDEV_STATE_CANT_OPEN;
2556 vs->vs_aux = VDEV_AUX_SPARED;
2563 * Add l2cache device information to the nvlist, including vdev stats.
2566 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2569 uint_t i, j, nl2cache;
2576 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2578 if (spa->spa_l2cache.sav_count == 0)
2581 VERIFY(nvlist_lookup_nvlist(config,
2582 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2583 VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2584 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2585 if (nl2cache != 0) {
2586 VERIFY(nvlist_add_nvlist_array(nvroot,
2587 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2588 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2589 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2592 * Update level 2 cache device stats.
2595 for (i = 0; i < nl2cache; i++) {
2596 VERIFY(nvlist_lookup_uint64(l2cache[i],
2597 ZPOOL_CONFIG_GUID, &guid) == 0);
2600 for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2602 spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2603 vd = spa->spa_l2cache.sav_vdevs[j];
2609 VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2610 ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2612 vdev_get_stats(vd, vs);
2618 spa_get_stats(const char *name, nvlist_t **config, char *altroot, size_t buflen)
2624 error = spa_open_common(name, &spa, FTAG, NULL, config);
2628 * This still leaves a window of inconsistency where the spares
2629 * or l2cache devices could change and the config would be
2630 * self-inconsistent.
2632 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2634 if (*config != NULL) {
2635 uint64_t loadtimes[2];
2637 loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2638 loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
2639 VERIFY(nvlist_add_uint64_array(*config,
2640 ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
2642 VERIFY(nvlist_add_uint64(*config,
2643 ZPOOL_CONFIG_ERRCOUNT,
2644 spa_get_errlog_size(spa)) == 0);
2646 if (spa_suspended(spa))
2647 VERIFY(nvlist_add_uint64(*config,
2648 ZPOOL_CONFIG_SUSPENDED,
2649 spa->spa_failmode) == 0);
2651 spa_add_spares(spa, *config);
2652 spa_add_l2cache(spa, *config);
2657 * We want to get the alternate root even for faulted pools, so we cheat
2658 * and call spa_lookup() directly.
2662 mutex_enter(&spa_namespace_lock);
2663 spa = spa_lookup(name);
2665 spa_altroot(spa, altroot, buflen);
2669 mutex_exit(&spa_namespace_lock);
2671 spa_altroot(spa, altroot, buflen);
2676 spa_config_exit(spa, SCL_CONFIG, FTAG);
2677 spa_close(spa, FTAG);
2684 * Validate that the auxiliary device array is well formed. We must have an
2685 * array of nvlists, each which describes a valid leaf vdev. If this is an
2686 * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
2687 * specified, as long as they are well-formed.
2690 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
2691 spa_aux_vdev_t *sav, const char *config, uint64_t version,
2692 vdev_labeltype_t label)
2699 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2702 * It's acceptable to have no devs specified.
2704 if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
2711 * Make sure the pool is formatted with a version that supports this
2714 if (spa_version(spa) < version)
2718 * Set the pending device list so we correctly handle device in-use
2721 sav->sav_pending = dev;
2722 sav->sav_npending = ndev;
2724 for (i = 0; i < ndev; i++) {
2725 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
2729 if (!vd->vdev_ops->vdev_op_leaf) {
2736 * The L2ARC currently only supports disk devices in
2737 * kernel context. For user-level testing, we allow it.
2740 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
2741 strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
2748 if ((error = vdev_open(vd)) == 0 &&
2749 (error = vdev_label_init(vd, crtxg, label)) == 0) {
2750 VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
2751 vd->vdev_guid) == 0);
2757 (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
2764 sav->sav_pending = NULL;
2765 sav->sav_npending = 0;
2770 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
2774 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
2776 if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2777 &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
2778 VDEV_LABEL_SPARE)) != 0) {
2782 return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
2783 &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
2784 VDEV_LABEL_L2CACHE));
2788 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
2793 if (sav->sav_config != NULL) {
2799 * Generate new dev list by concatentating with the
2802 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
2803 &olddevs, &oldndevs) == 0);
2805 newdevs = kmem_alloc(sizeof (void *) *
2806 (ndevs + oldndevs), KM_SLEEP);
2807 for (i = 0; i < oldndevs; i++)
2808 VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
2810 for (i = 0; i < ndevs; i++)
2811 VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
2814 VERIFY(nvlist_remove(sav->sav_config, config,
2815 DATA_TYPE_NVLIST_ARRAY) == 0);
2817 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
2818 config, newdevs, ndevs + oldndevs) == 0);
2819 for (i = 0; i < oldndevs + ndevs; i++)
2820 nvlist_free(newdevs[i]);
2821 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
2824 * Generate a new dev list.
2826 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
2828 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
2834 * Stop and drop level 2 ARC devices
2837 spa_l2cache_drop(spa_t *spa)
2841 spa_aux_vdev_t *sav = &spa->spa_l2cache;
2843 for (i = 0; i < sav->sav_count; i++) {
2846 vd = sav->sav_vdevs[i];
2849 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
2850 pool != 0ULL && l2arc_vdev_present(vd))
2851 l2arc_remove_vdev(vd);
2852 if (vd->vdev_isl2cache)
2853 spa_l2cache_remove(vd);
2854 vdev_clear_stats(vd);
2855 (void) vdev_close(vd);
2863 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
2864 const char *history_str, nvlist_t *zplprops)
2867 char *altroot = NULL;
2872 uint64_t txg = TXG_INITIAL;
2873 nvlist_t **spares, **l2cache;
2874 uint_t nspares, nl2cache;
2875 uint64_t version, obj;
2879 * If this pool already exists, return failure.
2881 mutex_enter(&spa_namespace_lock);
2882 if (spa_lookup(pool) != NULL) {
2883 mutex_exit(&spa_namespace_lock);
2888 * Allocate a new spa_t structure.
2890 (void) nvlist_lookup_string(props,
2891 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
2892 spa = spa_add(pool, NULL, altroot);
2893 spa_activate(spa, spa_mode_global);
2895 if (props && (error = spa_prop_validate(spa, props))) {
2896 spa_deactivate(spa);
2898 mutex_exit(&spa_namespace_lock);
2902 if (nvlist_lookup_uint64(props, zpool_prop_to_name(ZPOOL_PROP_VERSION),
2904 version = SPA_VERSION;
2905 ASSERT(version <= SPA_VERSION);
2907 spa->spa_first_txg = txg;
2908 spa->spa_uberblock.ub_txg = txg - 1;
2909 spa->spa_uberblock.ub_version = version;
2910 spa->spa_ubsync = spa->spa_uberblock;
2913 * Create "The Godfather" zio to hold all async IOs
2915 spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2916 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2919 * Create the root vdev.
2921 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2923 error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
2925 ASSERT(error != 0 || rvd != NULL);
2926 ASSERT(error != 0 || spa->spa_root_vdev == rvd);
2928 if (error == 0 && !zfs_allocatable_devs(nvroot))
2932 (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
2933 (error = spa_validate_aux(spa, nvroot, txg,
2934 VDEV_ALLOC_ADD)) == 0) {
2935 for (c = 0; c < rvd->vdev_children; c++) {
2936 vdev_metaslab_set_size(rvd->vdev_child[c]);
2937 vdev_expand(rvd->vdev_child[c], txg);
2941 spa_config_exit(spa, SCL_ALL, FTAG);
2945 spa_deactivate(spa);
2947 mutex_exit(&spa_namespace_lock);
2952 * Get the list of spares, if specified.
2954 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
2955 &spares, &nspares) == 0) {
2956 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
2958 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
2959 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2960 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2961 spa_load_spares(spa);
2962 spa_config_exit(spa, SCL_ALL, FTAG);
2963 spa->spa_spares.sav_sync = B_TRUE;
2967 * Get the list of level 2 cache devices, if specified.
2969 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
2970 &l2cache, &nl2cache) == 0) {
2971 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
2972 NV_UNIQUE_NAME, KM_SLEEP) == 0);
2973 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
2974 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2975 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2976 spa_load_l2cache(spa);
2977 spa_config_exit(spa, SCL_ALL, FTAG);
2978 spa->spa_l2cache.sav_sync = B_TRUE;
2981 spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
2982 spa->spa_meta_objset = dp->dp_meta_objset;
2985 * Create DDTs (dedup tables).
2989 spa_update_dspace(spa);
2991 tx = dmu_tx_create_assigned(dp, txg);
2994 * Create the pool config object.
2996 spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
2997 DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
2998 DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3000 if (zap_add(spa->spa_meta_objset,
3001 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3002 sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3003 cmn_err(CE_PANIC, "failed to add pool config");
3006 if (zap_add(spa->spa_meta_objset,
3007 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3008 sizeof (uint64_t), 1, &version, tx) != 0) {
3009 cmn_err(CE_PANIC, "failed to add pool version");
3012 /* Newly created pools with the right version are always deflated. */
3013 if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3014 spa->spa_deflate = TRUE;
3015 if (zap_add(spa->spa_meta_objset,
3016 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3017 sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3018 cmn_err(CE_PANIC, "failed to add deflate");
3023 * Create the deferred-free bpobj. Turn off compression
3024 * because sync-to-convergence takes longer if the blocksize
3027 obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3028 dmu_object_set_compress(spa->spa_meta_objset, obj,
3029 ZIO_COMPRESS_OFF, tx);
3030 if (zap_add(spa->spa_meta_objset,
3031 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3032 sizeof (uint64_t), 1, &obj, tx) != 0) {
3033 cmn_err(CE_PANIC, "failed to add bpobj");
3035 VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3036 spa->spa_meta_objset, obj));
3039 * Create the pool's history object.
3041 if (version >= SPA_VERSION_ZPOOL_HISTORY)
3042 spa_history_create_obj(spa, tx);
3045 * Set pool properties.
3047 spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3048 spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3049 spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3050 spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3052 if (props != NULL) {
3053 spa_configfile_set(spa, props, B_FALSE);
3054 spa_sync_props(spa, props, tx);
3059 spa->spa_sync_on = B_TRUE;
3060 txg_sync_start(spa->spa_dsl_pool);
3063 * We explicitly wait for the first transaction to complete so that our
3064 * bean counters are appropriately updated.
3066 txg_wait_synced(spa->spa_dsl_pool, txg);
3068 spa_config_sync(spa, B_FALSE, B_TRUE);
3070 if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3071 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3072 spa_history_log_version(spa, LOG_POOL_CREATE);
3074 spa->spa_minref = refcount_count(&spa->spa_refcount);
3076 mutex_exit(&spa_namespace_lock);
3083 * Get the root pool information from the root disk, then import the root pool
3084 * during the system boot up time.
3086 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3089 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3092 nvlist_t *nvtop, *nvroot;
3095 if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3099 * Add this top-level vdev to the child array.
3101 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3103 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3105 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3108 * Put this pool's top-level vdevs into a root vdev.
3110 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3111 VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3112 VDEV_TYPE_ROOT) == 0);
3113 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3114 VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3115 VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3119 * Replace the existing vdev_tree with the new root vdev in
3120 * this pool's configuration (remove the old, add the new).
3122 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3123 nvlist_free(nvroot);
3128 * Walk the vdev tree and see if we can find a device with "better"
3129 * configuration. A configuration is "better" if the label on that
3130 * device has a more recent txg.
3133 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3137 for (c = 0; c < vd->vdev_children; c++)
3138 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3140 if (vd->vdev_ops->vdev_op_leaf) {
3144 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3148 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3152 * Do we have a better boot device?
3154 if (label_txg > *txg) {
3163 * Import a root pool.
3165 * For x86. devpath_list will consist of devid and/or physpath name of
3166 * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3167 * The GRUB "findroot" command will return the vdev we should boot.
3169 * For Sparc, devpath_list consists the physpath name of the booting device
3170 * no matter the rootpool is a single device pool or a mirrored pool.
3172 * "/pci@1f,0/ide@d/disk@0,0:a"
3175 spa_import_rootpool(char *devpath, char *devid)
3178 vdev_t *rvd, *bvd, *avd = NULL;
3179 nvlist_t *config, *nvtop;
3185 * Read the label from the boot device and generate a configuration.
3187 config = spa_generate_rootconf(devpath, devid, &guid);
3188 #if defined(_OBP) && defined(_KERNEL)
3189 if (config == NULL) {
3190 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3192 get_iscsi_bootpath_phy(devpath);
3193 config = spa_generate_rootconf(devpath, devid, &guid);
3197 if (config == NULL) {
3198 cmn_err(CE_NOTE, "Can not read the pool label from '%s'",
3203 VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3205 VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3207 mutex_enter(&spa_namespace_lock);
3208 if ((spa = spa_lookup(pname)) != NULL) {
3210 * Remove the existing root pool from the namespace so that we
3211 * can replace it with the correct config we just read in.
3216 spa = spa_add(pname, config, NULL);
3217 spa->spa_is_root = B_TRUE;
3218 spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3221 * Build up a vdev tree based on the boot device's label config.
3223 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3225 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3226 error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3227 VDEV_ALLOC_ROOTPOOL);
3228 spa_config_exit(spa, SCL_ALL, FTAG);
3230 mutex_exit(&spa_namespace_lock);
3231 nvlist_free(config);
3232 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3238 * Get the boot vdev.
3240 if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3241 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3242 (u_longlong_t)guid);
3248 * Determine if there is a better boot device.
3251 spa_alt_rootvdev(rvd, &avd, &txg);
3253 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3254 "try booting from '%s'", avd->vdev_path);
3260 * If the boot device is part of a spare vdev then ensure that
3261 * we're booting off the active spare.
3263 if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3264 !bvd->vdev_isspare) {
3265 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3266 "try booting from '%s'",
3268 vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3274 spa_history_log_version(spa, LOG_POOL_IMPORT);
3276 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3278 spa_config_exit(spa, SCL_ALL, FTAG);
3279 mutex_exit(&spa_namespace_lock);
3281 nvlist_free(config);
3288 * Import a non-root pool into the system.
3291 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3294 char *altroot = NULL;
3295 spa_load_state_t state = SPA_LOAD_IMPORT;
3296 zpool_rewind_policy_t policy;
3297 uint64_t mode = spa_mode_global;
3298 uint64_t readonly = B_FALSE;
3301 nvlist_t **spares, **l2cache;
3302 uint_t nspares, nl2cache;
3305 * If a pool with this name exists, return failure.
3307 mutex_enter(&spa_namespace_lock);
3308 if (spa_lookup(pool) != NULL) {
3309 mutex_exit(&spa_namespace_lock);
3314 * Create and initialize the spa structure.
3316 (void) nvlist_lookup_string(props,
3317 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3318 (void) nvlist_lookup_uint64(props,
3319 zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3322 spa = spa_add(pool, config, altroot);
3323 spa->spa_import_flags = flags;
3326 * Verbatim import - Take a pool and insert it into the namespace
3327 * as if it had been loaded at boot.
3329 if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3331 spa_configfile_set(spa, props, B_FALSE);
3333 spa_config_sync(spa, B_FALSE, B_TRUE);
3335 mutex_exit(&spa_namespace_lock);
3336 spa_history_log_version(spa, LOG_POOL_IMPORT);
3341 spa_activate(spa, mode);
3344 * Don't start async tasks until we know everything is healthy.
3346 spa_async_suspend(spa);
3348 zpool_get_rewind_policy(config, &policy);
3349 if (policy.zrp_request & ZPOOL_DO_REWIND)
3350 state = SPA_LOAD_RECOVER;
3353 * Pass off the heavy lifting to spa_load(). Pass TRUE for mosconfig
3354 * because the user-supplied config is actually the one to trust when
3357 if (state != SPA_LOAD_RECOVER)
3358 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3360 error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3361 policy.zrp_request);
3364 * Propagate anything learned while loading the pool and pass it
3365 * back to caller (i.e. rewind info, missing devices, etc).
3367 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3368 spa->spa_load_info) == 0);
3370 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3372 * Toss any existing sparelist, as it doesn't have any validity
3373 * anymore, and conflicts with spa_has_spare().
3375 if (spa->spa_spares.sav_config) {
3376 nvlist_free(spa->spa_spares.sav_config);
3377 spa->spa_spares.sav_config = NULL;
3378 spa_load_spares(spa);
3380 if (spa->spa_l2cache.sav_config) {
3381 nvlist_free(spa->spa_l2cache.sav_config);
3382 spa->spa_l2cache.sav_config = NULL;
3383 spa_load_l2cache(spa);
3386 VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3389 error = spa_validate_aux(spa, nvroot, -1ULL,
3392 error = spa_validate_aux(spa, nvroot, -1ULL,
3393 VDEV_ALLOC_L2CACHE);
3394 spa_config_exit(spa, SCL_ALL, FTAG);
3397 spa_configfile_set(spa, props, B_FALSE);
3399 if (error != 0 || (props && spa_writeable(spa) &&
3400 (error = spa_prop_set(spa, props)))) {
3402 spa_deactivate(spa);
3404 mutex_exit(&spa_namespace_lock);
3408 spa_async_resume(spa);
3411 * Override any spares and level 2 cache devices as specified by
3412 * the user, as these may have correct device names/devids, etc.
3414 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3415 &spares, &nspares) == 0) {
3416 if (spa->spa_spares.sav_config)
3417 VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3418 ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3420 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3421 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3422 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3423 ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3424 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3425 spa_load_spares(spa);
3426 spa_config_exit(spa, SCL_ALL, FTAG);
3427 spa->spa_spares.sav_sync = B_TRUE;
3429 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3430 &l2cache, &nl2cache) == 0) {
3431 if (spa->spa_l2cache.sav_config)
3432 VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3433 ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3435 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3436 NV_UNIQUE_NAME, KM_SLEEP) == 0);
3437 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3438 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3439 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3440 spa_load_l2cache(spa);
3441 spa_config_exit(spa, SCL_ALL, FTAG);
3442 spa->spa_l2cache.sav_sync = B_TRUE;
3446 * Check for any removed devices.
3448 if (spa->spa_autoreplace) {
3449 spa_aux_check_removed(&spa->spa_spares);
3450 spa_aux_check_removed(&spa->spa_l2cache);
3453 if (spa_writeable(spa)) {
3455 * Update the config cache to include the newly-imported pool.
3457 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3461 * It's possible that the pool was expanded while it was exported.
3462 * We kick off an async task to handle this for us.
3464 spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3466 mutex_exit(&spa_namespace_lock);
3467 spa_history_log_version(spa, LOG_POOL_IMPORT);
3473 spa_tryimport(nvlist_t *tryconfig)
3475 nvlist_t *config = NULL;
3481 if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3484 if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3488 * Create and initialize the spa structure.
3490 mutex_enter(&spa_namespace_lock);
3491 spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3492 spa_activate(spa, FREAD);
3495 * Pass off the heavy lifting to spa_load().
3496 * Pass TRUE for mosconfig because the user-supplied config
3497 * is actually the one to trust when doing an import.
3499 error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3502 * If 'tryconfig' was at least parsable, return the current config.
3504 if (spa->spa_root_vdev != NULL) {
3505 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3506 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3508 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3510 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3511 spa->spa_uberblock.ub_timestamp) == 0);
3514 * If the bootfs property exists on this pool then we
3515 * copy it out so that external consumers can tell which
3516 * pools are bootable.
3518 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3519 char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3522 * We have to play games with the name since the
3523 * pool was opened as TRYIMPORT_NAME.
3525 if (dsl_dsobj_to_dsname(spa_name(spa),
3526 spa->spa_bootfs, tmpname) == 0) {
3528 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3530 cp = strchr(tmpname, '/');
3532 (void) strlcpy(dsname, tmpname,
3535 (void) snprintf(dsname, MAXPATHLEN,
3536 "%s/%s", poolname, ++cp);
3538 VERIFY(nvlist_add_string(config,
3539 ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3540 kmem_free(dsname, MAXPATHLEN);
3542 kmem_free(tmpname, MAXPATHLEN);
3546 * Add the list of hot spares and level 2 cache devices.
3548 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3549 spa_add_spares(spa, config);
3550 spa_add_l2cache(spa, config);
3551 spa_config_exit(spa, SCL_CONFIG, FTAG);
3555 spa_deactivate(spa);
3557 mutex_exit(&spa_namespace_lock);
3563 * Pool export/destroy
3565 * The act of destroying or exporting a pool is very simple. We make sure there
3566 * is no more pending I/O and any references to the pool are gone. Then, we
3567 * update the pool state and sync all the labels to disk, removing the
3568 * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3569 * we don't sync the labels or remove the configuration cache.
3572 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3573 boolean_t force, boolean_t hardforce)
3580 if (!(spa_mode_global & FWRITE))
3583 mutex_enter(&spa_namespace_lock);
3584 if ((spa = spa_lookup(pool)) == NULL) {
3585 mutex_exit(&spa_namespace_lock);
3590 * Put a hold on the pool, drop the namespace lock, stop async tasks,
3591 * reacquire the namespace lock, and see if we can export.
3593 spa_open_ref(spa, FTAG);
3594 mutex_exit(&spa_namespace_lock);
3595 spa_async_suspend(spa);
3596 mutex_enter(&spa_namespace_lock);
3597 spa_close(spa, FTAG);
3600 * The pool will be in core if it's openable,
3601 * in which case we can modify its state.
3603 if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3605 * Objsets may be open only because they're dirty, so we
3606 * have to force it to sync before checking spa_refcnt.
3608 txg_wait_synced(spa->spa_dsl_pool, 0);
3611 * A pool cannot be exported or destroyed if there are active
3612 * references. If we are resetting a pool, allow references by
3613 * fault injection handlers.
3615 if (!spa_refcount_zero(spa) ||
3616 (spa->spa_inject_ref != 0 &&
3617 new_state != POOL_STATE_UNINITIALIZED)) {
3618 spa_async_resume(spa);
3619 mutex_exit(&spa_namespace_lock);
3624 * A pool cannot be exported if it has an active shared spare.
3625 * This is to prevent other pools stealing the active spare
3626 * from an exported pool. At user's own will, such pool can
3627 * be forcedly exported.
3629 if (!force && new_state == POOL_STATE_EXPORTED &&
3630 spa_has_active_shared_spare(spa)) {
3631 spa_async_resume(spa);
3632 mutex_exit(&spa_namespace_lock);
3637 * We want this to be reflected on every label,
3638 * so mark them all dirty. spa_unload() will do the
3639 * final sync that pushes these changes out.
3641 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
3642 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3643 spa->spa_state = new_state;
3644 spa->spa_final_txg = spa_last_synced_txg(spa) +
3646 vdev_config_dirty(spa->spa_root_vdev);
3647 spa_config_exit(spa, SCL_ALL, FTAG);
3651 spa_event_notify(spa, NULL, FM_EREPORT_ZFS_POOL_DESTROY);
3653 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
3655 spa_deactivate(spa);
3658 if (oldconfig && spa->spa_config)
3659 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
3661 if (new_state != POOL_STATE_UNINITIALIZED) {
3663 spa_config_sync(spa, B_TRUE, B_TRUE);
3666 mutex_exit(&spa_namespace_lock);
3672 * Destroy a storage pool.
3675 spa_destroy(char *pool)
3677 return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
3682 * Export a storage pool.
3685 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
3686 boolean_t hardforce)
3688 return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
3693 * Similar to spa_export(), this unloads the spa_t without actually removing it
3694 * from the namespace in any way.
3697 spa_reset(char *pool)
3699 return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
3704 * ==========================================================================
3705 * Device manipulation
3706 * ==========================================================================
3710 * Add a device to a storage pool.
3713 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
3717 vdev_t *rvd = spa->spa_root_vdev;
3719 nvlist_t **spares, **l2cache;
3720 uint_t nspares, nl2cache;
3723 ASSERT(spa_writeable(spa));
3725 txg = spa_vdev_enter(spa);
3727 if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
3728 VDEV_ALLOC_ADD)) != 0)
3729 return (spa_vdev_exit(spa, NULL, txg, error));
3731 spa->spa_pending_vdev = vd; /* spa_vdev_exit() will clear this */
3733 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
3737 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
3741 if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
3742 return (spa_vdev_exit(spa, vd, txg, EINVAL));
3744 if (vd->vdev_children != 0 &&
3745 (error = vdev_create(vd, txg, B_FALSE)) != 0)
3746 return (spa_vdev_exit(spa, vd, txg, error));
3749 * We must validate the spares and l2cache devices after checking the
3750 * children. Otherwise, vdev_inuse() will blindly overwrite the spare.
3752 if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
3753 return (spa_vdev_exit(spa, vd, txg, error));
3756 * Transfer each new top-level vdev from vd to rvd.
3758 for (c = 0; c < vd->vdev_children; c++) {
3761 * Set the vdev id to the first hole, if one exists.
3763 for (id = 0; id < rvd->vdev_children; id++) {
3764 if (rvd->vdev_child[id]->vdev_ishole) {
3765 vdev_free(rvd->vdev_child[id]);
3769 tvd = vd->vdev_child[c];
3770 vdev_remove_child(vd, tvd);
3772 vdev_add_child(rvd, tvd);
3773 vdev_config_dirty(tvd);
3777 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
3778 ZPOOL_CONFIG_SPARES);
3779 spa_load_spares(spa);
3780 spa->spa_spares.sav_sync = B_TRUE;
3783 if (nl2cache != 0) {
3784 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
3785 ZPOOL_CONFIG_L2CACHE);
3786 spa_load_l2cache(spa);
3787 spa->spa_l2cache.sav_sync = B_TRUE;
3791 * We have to be careful when adding new vdevs to an existing pool.
3792 * If other threads start allocating from these vdevs before we
3793 * sync the config cache, and we lose power, then upon reboot we may
3794 * fail to open the pool because there are DVAs that the config cache
3795 * can't translate. Therefore, we first add the vdevs without
3796 * initializing metaslabs; sync the config cache (via spa_vdev_exit());
3797 * and then let spa_config_update() initialize the new metaslabs.
3799 * spa_load() checks for added-but-not-initialized vdevs, so that
3800 * if we lose power at any point in this sequence, the remaining
3801 * steps will be completed the next time we load the pool.
3803 (void) spa_vdev_exit(spa, vd, txg, 0);
3805 mutex_enter(&spa_namespace_lock);
3806 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3807 mutex_exit(&spa_namespace_lock);
3813 * Attach a device to a mirror. The arguments are the path to any device
3814 * in the mirror, and the nvroot for the new device. If the path specifies
3815 * a device that is not mirrored, we automatically insert the mirror vdev.
3817 * If 'replacing' is specified, the new device is intended to replace the
3818 * existing device; in this case the two devices are made into their own
3819 * mirror using the 'replacing' vdev, which is functionally identical to
3820 * the mirror vdev (it actually reuses all the same ops) but has a few
3821 * extra rules: you can't attach to it after it's been created, and upon
3822 * completion of resilvering, the first disk (the one being replaced)
3823 * is automatically detached.
3826 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
3828 uint64_t txg, dtl_max_txg;
3829 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
3830 vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
3832 char *oldvdpath, *newvdpath;
3836 ASSERT(spa_writeable(spa));
3838 txg = spa_vdev_enter(spa);
3840 oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
3843 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
3845 if (!oldvd->vdev_ops->vdev_op_leaf)
3846 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
3848 pvd = oldvd->vdev_parent;
3850 if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
3851 VDEV_ALLOC_ADD)) != 0)
3852 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
3854 if (newrootvd->vdev_children != 1)
3855 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3857 newvd = newrootvd->vdev_child[0];
3859 if (!newvd->vdev_ops->vdev_op_leaf)
3860 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
3862 if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
3863 return (spa_vdev_exit(spa, newrootvd, txg, error));
3866 * Spares can't replace logs
3868 if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
3869 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3873 * For attach, the only allowable parent is a mirror or the root
3876 if (pvd->vdev_ops != &vdev_mirror_ops &&
3877 pvd->vdev_ops != &vdev_root_ops)
3878 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3880 pvops = &vdev_mirror_ops;
3883 * Active hot spares can only be replaced by inactive hot
3886 if (pvd->vdev_ops == &vdev_spare_ops &&
3887 oldvd->vdev_isspare &&
3888 !spa_has_spare(spa, newvd->vdev_guid))
3889 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3892 * If the source is a hot spare, and the parent isn't already a
3893 * spare, then we want to create a new hot spare. Otherwise, we
3894 * want to create a replacing vdev. The user is not allowed to
3895 * attach to a spared vdev child unless the 'isspare' state is
3896 * the same (spare replaces spare, non-spare replaces
3899 if (pvd->vdev_ops == &vdev_replacing_ops &&
3900 spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
3901 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3902 } else if (pvd->vdev_ops == &vdev_spare_ops &&
3903 newvd->vdev_isspare != oldvd->vdev_isspare) {
3904 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
3907 if (newvd->vdev_isspare)
3908 pvops = &vdev_spare_ops;
3910 pvops = &vdev_replacing_ops;
3914 * Make sure the new device is big enough.
3916 if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
3917 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
3920 * The new device cannot have a higher alignment requirement
3921 * than the top-level vdev.
3923 if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
3924 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
3927 * If this is an in-place replacement, update oldvd's path and devid
3928 * to make it distinguishable from newvd, and unopenable from now on.
3930 if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
3931 spa_strfree(oldvd->vdev_path);
3932 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
3934 (void) sprintf(oldvd->vdev_path, "%s/%s",
3935 newvd->vdev_path, "old");
3936 if (oldvd->vdev_devid != NULL) {
3937 spa_strfree(oldvd->vdev_devid);
3938 oldvd->vdev_devid = NULL;
3942 /* mark the device being resilvered */
3943 newvd->vdev_resilvering = B_TRUE;
3946 * If the parent is not a mirror, or if we're replacing, insert the new
3947 * mirror/replacing/spare vdev above oldvd.
3949 if (pvd->vdev_ops != pvops)
3950 pvd = vdev_add_parent(oldvd, pvops);
3952 ASSERT(pvd->vdev_top->vdev_parent == rvd);
3953 ASSERT(pvd->vdev_ops == pvops);
3954 ASSERT(oldvd->vdev_parent == pvd);
3957 * Extract the new device from its root and add it to pvd.
3959 vdev_remove_child(newrootvd, newvd);
3960 newvd->vdev_id = pvd->vdev_children;
3961 newvd->vdev_crtxg = oldvd->vdev_crtxg;
3962 vdev_add_child(pvd, newvd);
3964 tvd = newvd->vdev_top;
3965 ASSERT(pvd->vdev_top == tvd);
3966 ASSERT(tvd->vdev_parent == rvd);
3968 vdev_config_dirty(tvd);
3971 * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
3972 * for any dmu_sync-ed blocks. It will propagate upward when
3973 * spa_vdev_exit() calls vdev_dtl_reassess().
3975 dtl_max_txg = txg + TXG_CONCURRENT_STATES;
3977 vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
3978 dtl_max_txg - TXG_INITIAL);
3980 if (newvd->vdev_isspare) {
3981 spa_spare_activate(newvd);
3982 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_DEVICE_SPARE);
3985 oldvdpath = spa_strdup(oldvd->vdev_path);
3986 newvdpath = spa_strdup(newvd->vdev_path);
3987 newvd_isspare = newvd->vdev_isspare;
3990 * Mark newvd's DTL dirty in this txg.
3992 vdev_dirty(tvd, VDD_DTL, newvd, txg);
3995 * Restart the resilver
3997 dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4002 (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4004 spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4005 "%s vdev=%s %s vdev=%s",
4006 replacing && newvd_isspare ? "spare in" :
4007 replacing ? "replace" : "attach", newvdpath,
4008 replacing ? "for" : "to", oldvdpath);
4010 spa_strfree(oldvdpath);
4011 spa_strfree(newvdpath);
4013 if (spa->spa_bootfs)
4014 spa_event_notify(spa, newvd, FM_EREPORT_ZFS_BOOTFS_VDEV_ATTACH);
4020 * Detach a device from a mirror or replacing vdev.
4021 * If 'replace_done' is specified, only detach if the parent
4022 * is a replacing vdev.
4025 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4029 ASSERTV(vdev_t *rvd = spa->spa_root_vdev;)
4030 vdev_t *vd, *pvd, *cvd, *tvd;
4031 boolean_t unspare = B_FALSE;
4032 uint64_t unspare_guid = 0;
4036 ASSERT(spa_writeable(spa));
4038 txg = spa_vdev_enter(spa);
4040 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4043 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4045 if (!vd->vdev_ops->vdev_op_leaf)
4046 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4048 pvd = vd->vdev_parent;
4051 * If the parent/child relationship is not as expected, don't do it.
4052 * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4053 * vdev that's replacing B with C. The user's intent in replacing
4054 * is to go from M(A,B) to M(A,C). If the user decides to cancel
4055 * the replace by detaching C, the expected behavior is to end up
4056 * M(A,B). But suppose that right after deciding to detach C,
4057 * the replacement of B completes. We would have M(A,C), and then
4058 * ask to detach C, which would leave us with just A -- not what
4059 * the user wanted. To prevent this, we make sure that the
4060 * parent/child relationship hasn't changed -- in this example,
4061 * that C's parent is still the replacing vdev R.
4063 if (pvd->vdev_guid != pguid && pguid != 0)
4064 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4067 * Only 'replacing' or 'spare' vdevs can be replaced.
4069 if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4070 pvd->vdev_ops != &vdev_spare_ops)
4071 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4073 ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4074 spa_version(spa) >= SPA_VERSION_SPARES);
4077 * Only mirror, replacing, and spare vdevs support detach.
4079 if (pvd->vdev_ops != &vdev_replacing_ops &&
4080 pvd->vdev_ops != &vdev_mirror_ops &&
4081 pvd->vdev_ops != &vdev_spare_ops)
4082 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4085 * If this device has the only valid copy of some data,
4086 * we cannot safely detach it.
4088 if (vdev_dtl_required(vd))
4089 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4091 ASSERT(pvd->vdev_children >= 2);
4094 * If we are detaching the second disk from a replacing vdev, then
4095 * check to see if we changed the original vdev's path to have "/old"
4096 * at the end in spa_vdev_attach(). If so, undo that change now.
4098 if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4099 vd->vdev_path != NULL) {
4100 size_t len = strlen(vd->vdev_path);
4102 for (c = 0; c < pvd->vdev_children; c++) {
4103 cvd = pvd->vdev_child[c];
4105 if (cvd == vd || cvd->vdev_path == NULL)
4108 if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4109 strcmp(cvd->vdev_path + len, "/old") == 0) {
4110 spa_strfree(cvd->vdev_path);
4111 cvd->vdev_path = spa_strdup(vd->vdev_path);
4118 * If we are detaching the original disk from a spare, then it implies
4119 * that the spare should become a real disk, and be removed from the
4120 * active spare list for the pool.
4122 if (pvd->vdev_ops == &vdev_spare_ops &&
4124 pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4128 * Erase the disk labels so the disk can be used for other things.
4129 * This must be done after all other error cases are handled,
4130 * but before we disembowel vd (so we can still do I/O to it).
4131 * But if we can't do it, don't treat the error as fatal --
4132 * it may be that the unwritability of the disk is the reason
4133 * it's being detached!
4135 error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4138 * Remove vd from its parent and compact the parent's children.
4140 vdev_remove_child(pvd, vd);
4141 vdev_compact_children(pvd);
4144 * Remember one of the remaining children so we can get tvd below.
4146 cvd = pvd->vdev_child[pvd->vdev_children - 1];
4149 * If we need to remove the remaining child from the list of hot spares,
4150 * do it now, marking the vdev as no longer a spare in the process.
4151 * We must do this before vdev_remove_parent(), because that can
4152 * change the GUID if it creates a new toplevel GUID. For a similar
4153 * reason, we must remove the spare now, in the same txg as the detach;
4154 * otherwise someone could attach a new sibling, change the GUID, and
4155 * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4158 ASSERT(cvd->vdev_isspare);
4159 spa_spare_remove(cvd);
4160 unspare_guid = cvd->vdev_guid;
4161 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4162 cvd->vdev_unspare = B_TRUE;
4166 * If the parent mirror/replacing vdev only has one child,
4167 * the parent is no longer needed. Remove it from the tree.
4169 if (pvd->vdev_children == 1) {
4170 if (pvd->vdev_ops == &vdev_spare_ops)
4171 cvd->vdev_unspare = B_FALSE;
4172 vdev_remove_parent(cvd);
4173 cvd->vdev_resilvering = B_FALSE;
4178 * We don't set tvd until now because the parent we just removed
4179 * may have been the previous top-level vdev.
4181 tvd = cvd->vdev_top;
4182 ASSERT(tvd->vdev_parent == rvd);
4185 * Reevaluate the parent vdev state.
4187 vdev_propagate_state(cvd);
4190 * If the 'autoexpand' property is set on the pool then automatically
4191 * try to expand the size of the pool. For example if the device we
4192 * just detached was smaller than the others, it may be possible to
4193 * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4194 * first so that we can obtain the updated sizes of the leaf vdevs.
4196 if (spa->spa_autoexpand) {
4198 vdev_expand(tvd, txg);
4201 vdev_config_dirty(tvd);
4204 * Mark vd's DTL as dirty in this txg. vdev_dtl_sync() will see that
4205 * vd->vdev_detached is set and free vd's DTL object in syncing context.
4206 * But first make sure we're not on any *other* txg's DTL list, to
4207 * prevent vd from being accessed after it's freed.
4209 vdpath = spa_strdup(vd->vdev_path);
4210 for (t = 0; t < TXG_SIZE; t++)
4211 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4212 vd->vdev_detached = B_TRUE;
4213 vdev_dirty(tvd, VDD_DTL, vd, txg);
4215 spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_REMOVE);
4217 /* hang on to the spa before we release the lock */
4218 spa_open_ref(spa, FTAG);
4220 error = spa_vdev_exit(spa, vd, txg, 0);
4222 spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4224 spa_strfree(vdpath);
4227 * If this was the removal of the original device in a hot spare vdev,
4228 * then we want to go through and remove the device from the hot spare
4229 * list of every other pool.
4232 spa_t *altspa = NULL;
4234 mutex_enter(&spa_namespace_lock);
4235 while ((altspa = spa_next(altspa)) != NULL) {
4236 if (altspa->spa_state != POOL_STATE_ACTIVE ||
4240 spa_open_ref(altspa, FTAG);
4241 mutex_exit(&spa_namespace_lock);
4242 (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4243 mutex_enter(&spa_namespace_lock);
4244 spa_close(altspa, FTAG);
4246 mutex_exit(&spa_namespace_lock);
4248 /* search the rest of the vdevs for spares to remove */
4249 spa_vdev_resilver_done(spa);
4252 /* all done with the spa; OK to release */
4253 mutex_enter(&spa_namespace_lock);
4254 spa_close(spa, FTAG);
4255 mutex_exit(&spa_namespace_lock);
4261 * Split a set of devices from their mirrors, and create a new pool from them.
4264 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4265 nvlist_t *props, boolean_t exp)
4268 uint64_t txg, *glist;
4270 uint_t c, children, lastlog;
4271 nvlist_t **child, *nvl, *tmp;
4273 char *altroot = NULL;
4274 vdev_t *rvd, **vml = NULL; /* vdev modify list */
4275 boolean_t activate_slog;
4277 ASSERT(spa_writeable(spa));
4279 txg = spa_vdev_enter(spa);
4281 /* clear the log and flush everything up to now */
4282 activate_slog = spa_passivate_log(spa);
4283 (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4284 error = spa_offline_log(spa);
4285 txg = spa_vdev_config_enter(spa);
4288 spa_activate_log(spa);
4291 return (spa_vdev_exit(spa, NULL, txg, error));
4293 /* check new spa name before going any further */
4294 if (spa_lookup(newname) != NULL)
4295 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4298 * scan through all the children to ensure they're all mirrors
4300 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4301 nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4303 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4305 /* first, check to ensure we've got the right child count */
4306 rvd = spa->spa_root_vdev;
4308 for (c = 0; c < rvd->vdev_children; c++) {
4309 vdev_t *vd = rvd->vdev_child[c];
4311 /* don't count the holes & logs as children */
4312 if (vd->vdev_islog || vd->vdev_ishole) {
4320 if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4321 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4323 /* next, ensure no spare or cache devices are part of the split */
4324 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4325 nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4326 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4328 vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4329 glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4331 /* then, loop over each vdev and validate it */
4332 for (c = 0; c < children; c++) {
4333 uint64_t is_hole = 0;
4335 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4339 if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4340 spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4348 /* which disk is going to be split? */
4349 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4355 /* look it up in the spa */
4356 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4357 if (vml[c] == NULL) {
4362 /* make sure there's nothing stopping the split */
4363 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4364 vml[c]->vdev_islog ||
4365 vml[c]->vdev_ishole ||
4366 vml[c]->vdev_isspare ||
4367 vml[c]->vdev_isl2cache ||
4368 !vdev_writeable(vml[c]) ||
4369 vml[c]->vdev_children != 0 ||
4370 vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4371 c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4376 if (vdev_dtl_required(vml[c])) {
4381 /* we need certain info from the top level */
4382 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4383 vml[c]->vdev_top->vdev_ms_array) == 0);
4384 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4385 vml[c]->vdev_top->vdev_ms_shift) == 0);
4386 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4387 vml[c]->vdev_top->vdev_asize) == 0);
4388 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4389 vml[c]->vdev_top->vdev_ashift) == 0);
4393 kmem_free(vml, children * sizeof (vdev_t *));
4394 kmem_free(glist, children * sizeof (uint64_t));
4395 return (spa_vdev_exit(spa, NULL, txg, error));
4398 /* stop writers from using the disks */
4399 for (c = 0; c < children; c++) {
4401 vml[c]->vdev_offline = B_TRUE;
4403 vdev_reopen(spa->spa_root_vdev);
4406 * Temporarily record the splitting vdevs in the spa config. This
4407 * will disappear once the config is regenerated.
4409 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4410 VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4411 glist, children) == 0);
4412 kmem_free(glist, children * sizeof (uint64_t));
4414 mutex_enter(&spa->spa_props_lock);
4415 VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4417 mutex_exit(&spa->spa_props_lock);
4418 spa->spa_config_splitting = nvl;
4419 vdev_config_dirty(spa->spa_root_vdev);
4421 /* configure and create the new pool */
4422 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4423 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4424 exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4425 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4426 spa_version(spa)) == 0);
4427 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4428 spa->spa_config_txg) == 0);
4429 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4430 spa_generate_guid(NULL)) == 0);
4431 (void) nvlist_lookup_string(props,
4432 zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4434 /* add the new pool to the namespace */
4435 newspa = spa_add(newname, config, altroot);
4436 newspa->spa_config_txg = spa->spa_config_txg;
4437 spa_set_log_state(newspa, SPA_LOG_CLEAR);
4439 /* release the spa config lock, retaining the namespace lock */
4440 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4442 if (zio_injection_enabled)
4443 zio_handle_panic_injection(spa, FTAG, 1);
4445 spa_activate(newspa, spa_mode_global);
4446 spa_async_suspend(newspa);
4448 /* create the new pool from the disks of the original pool */
4449 error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4453 /* if that worked, generate a real config for the new pool */
4454 if (newspa->spa_root_vdev != NULL) {
4455 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4456 NV_UNIQUE_NAME, KM_SLEEP) == 0);
4457 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4458 ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4459 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4464 if (props != NULL) {
4465 spa_configfile_set(newspa, props, B_FALSE);
4466 error = spa_prop_set(newspa, props);
4471 /* flush everything */
4472 txg = spa_vdev_config_enter(newspa);
4473 vdev_config_dirty(newspa->spa_root_vdev);
4474 (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4476 if (zio_injection_enabled)
4477 zio_handle_panic_injection(spa, FTAG, 2);
4479 spa_async_resume(newspa);
4481 /* finally, update the original pool's config */
4482 txg = spa_vdev_config_enter(spa);
4483 tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4484 error = dmu_tx_assign(tx, TXG_WAIT);
4487 for (c = 0; c < children; c++) {
4488 if (vml[c] != NULL) {
4491 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4497 vdev_config_dirty(spa->spa_root_vdev);
4498 spa->spa_config_splitting = NULL;
4502 (void) spa_vdev_exit(spa, NULL, txg, 0);
4504 if (zio_injection_enabled)
4505 zio_handle_panic_injection(spa, FTAG, 3);
4507 /* split is complete; log a history record */
4508 spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4509 "split new pool %s from pool %s", newname, spa_name(spa));
4511 kmem_free(vml, children * sizeof (vdev_t *));
4513 /* if we're not going to mount the filesystems in userland, export */
4515 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4522 spa_deactivate(newspa);
4525 txg = spa_vdev_config_enter(spa);
4527 /* re-online all offlined disks */
4528 for (c = 0; c < children; c++) {
4530 vml[c]->vdev_offline = B_FALSE;
4532 vdev_reopen(spa->spa_root_vdev);
4534 nvlist_free(spa->spa_config_splitting);
4535 spa->spa_config_splitting = NULL;
4536 (void) spa_vdev_exit(spa, NULL, txg, error);
4538 kmem_free(vml, children * sizeof (vdev_t *));
4543 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4547 for (i = 0; i < count; i++) {
4550 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4553 if (guid == target_guid)
4561 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4562 nvlist_t *dev_to_remove)
4564 nvlist_t **newdev = NULL;
4568 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4570 for (i = 0, j = 0; i < count; i++) {
4571 if (dev[i] == dev_to_remove)
4573 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4576 VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4577 VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4579 for (i = 0; i < count - 1; i++)
4580 nvlist_free(newdev[i]);
4583 kmem_free(newdev, (count - 1) * sizeof (void *));
4587 * Evacuate the device.
4590 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4595 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4596 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4597 ASSERT(vd == vd->vdev_top);
4600 * Evacuate the device. We don't hold the config lock as writer
4601 * since we need to do I/O but we do keep the
4602 * spa_namespace_lock held. Once this completes the device
4603 * should no longer have any blocks allocated on it.
4605 if (vd->vdev_islog) {
4606 if (vd->vdev_stat.vs_alloc != 0)
4607 error = spa_offline_log(spa);
4616 * The evacuation succeeded. Remove any remaining MOS metadata
4617 * associated with this vdev, and wait for these changes to sync.
4619 ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4620 txg = spa_vdev_config_enter(spa);
4621 vd->vdev_removing = B_TRUE;
4622 vdev_dirty(vd, 0, NULL, txg);
4623 vdev_config_dirty(vd);
4624 spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4630 * Complete the removal by cleaning up the namespace.
4633 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
4635 vdev_t *rvd = spa->spa_root_vdev;
4636 uint64_t id = vd->vdev_id;
4637 boolean_t last_vdev = (id == (rvd->vdev_children - 1));
4639 ASSERT(MUTEX_HELD(&spa_namespace_lock));
4640 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
4641 ASSERT(vd == vd->vdev_top);
4644 * Only remove any devices which are empty.
4646 if (vd->vdev_stat.vs_alloc != 0)
4649 (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4651 if (list_link_active(&vd->vdev_state_dirty_node))
4652 vdev_state_clean(vd);
4653 if (list_link_active(&vd->vdev_config_dirty_node))
4654 vdev_config_clean(vd);
4659 vdev_compact_children(rvd);
4661 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
4662 vdev_add_child(rvd, vd);
4664 vdev_config_dirty(rvd);
4667 * Reassess the health of our root vdev.
4673 * Remove a device from the pool -
4675 * Removing a device from the vdev namespace requires several steps
4676 * and can take a significant amount of time. As a result we use
4677 * the spa_vdev_config_[enter/exit] functions which allow us to
4678 * grab and release the spa_config_lock while still holding the namespace
4679 * lock. During each step the configuration is synced out.
4683 * Remove a device from the pool. Currently, this supports removing only hot
4684 * spares, slogs, and level 2 ARC devices.
4687 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
4690 metaslab_group_t *mg;
4691 nvlist_t **spares, **l2cache, *nv;
4693 uint_t nspares, nl2cache;
4695 boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
4697 ASSERT(spa_writeable(spa));
4700 txg = spa_vdev_enter(spa);
4702 vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4704 if (spa->spa_spares.sav_vdevs != NULL &&
4705 nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
4706 ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
4707 (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
4709 * Only remove the hot spare if it's not currently in use
4712 if (vd == NULL || unspare) {
4713 spa_vdev_remove_aux(spa->spa_spares.sav_config,
4714 ZPOOL_CONFIG_SPARES, spares, nspares, nv);
4715 spa_load_spares(spa);
4716 spa->spa_spares.sav_sync = B_TRUE;
4720 } else if (spa->spa_l2cache.sav_vdevs != NULL &&
4721 nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
4722 ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
4723 (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
4725 * Cache devices can always be removed.
4727 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
4728 ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
4729 spa_load_l2cache(spa);
4730 spa->spa_l2cache.sav_sync = B_TRUE;
4731 } else if (vd != NULL && vd->vdev_islog) {
4733 ASSERT(vd == vd->vdev_top);
4736 * XXX - Once we have bp-rewrite this should
4737 * become the common case.
4743 * Stop allocating from this vdev.
4745 metaslab_group_passivate(mg);
4748 * Wait for the youngest allocations and frees to sync,
4749 * and then wait for the deferral of those frees to finish.
4751 spa_vdev_config_exit(spa, NULL,
4752 txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
4755 * Attempt to evacuate the vdev.
4757 error = spa_vdev_remove_evacuate(spa, vd);
4759 txg = spa_vdev_config_enter(spa);
4762 * If we couldn't evacuate the vdev, unwind.
4765 metaslab_group_activate(mg);
4766 return (spa_vdev_exit(spa, NULL, txg, error));
4770 * Clean up the vdev namespace.
4772 spa_vdev_remove_from_namespace(spa, vd);
4774 } else if (vd != NULL) {
4776 * Normal vdevs cannot be removed (yet).
4781 * There is no vdev of any kind with the specified guid.
4787 return (spa_vdev_exit(spa, NULL, txg, error));
4793 * Find any device that's done replacing, or a vdev marked 'unspare' that's
4794 * current spared, so we can detach it.
4797 spa_vdev_resilver_done_hunt(vdev_t *vd)
4799 vdev_t *newvd, *oldvd;
4802 for (c = 0; c < vd->vdev_children; c++) {
4803 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
4809 * Check for a completed replacement. We always consider the first
4810 * vdev in the list to be the oldest vdev, and the last one to be
4811 * the newest (see spa_vdev_attach() for how that works). In
4812 * the case where the newest vdev is faulted, we will not automatically
4813 * remove it after a resilver completes. This is OK as it will require
4814 * user intervention to determine which disk the admin wishes to keep.
4816 if (vd->vdev_ops == &vdev_replacing_ops) {
4817 ASSERT(vd->vdev_children > 1);
4819 newvd = vd->vdev_child[vd->vdev_children - 1];
4820 oldvd = vd->vdev_child[0];
4822 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
4823 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4824 !vdev_dtl_required(oldvd))
4829 * Check for a completed resilver with the 'unspare' flag set.
4831 if (vd->vdev_ops == &vdev_spare_ops) {
4832 vdev_t *first = vd->vdev_child[0];
4833 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
4835 if (last->vdev_unspare) {
4838 } else if (first->vdev_unspare) {
4845 if (oldvd != NULL &&
4846 vdev_dtl_empty(newvd, DTL_MISSING) &&
4847 vdev_dtl_empty(newvd, DTL_OUTAGE) &&
4848 !vdev_dtl_required(oldvd))
4852 * If there are more than two spares attached to a disk,
4853 * and those spares are not required, then we want to
4854 * attempt to free them up now so that they can be used
4855 * by other pools. Once we're back down to a single
4856 * disk+spare, we stop removing them.
4858 if (vd->vdev_children > 2) {
4859 newvd = vd->vdev_child[1];
4861 if (newvd->vdev_isspare && last->vdev_isspare &&
4862 vdev_dtl_empty(last, DTL_MISSING) &&
4863 vdev_dtl_empty(last, DTL_OUTAGE) &&
4864 !vdev_dtl_required(newvd))
4873 spa_vdev_resilver_done(spa_t *spa)
4875 vdev_t *vd, *pvd, *ppvd;
4876 uint64_t guid, sguid, pguid, ppguid;
4878 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4880 while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
4881 pvd = vd->vdev_parent;
4882 ppvd = pvd->vdev_parent;
4883 guid = vd->vdev_guid;
4884 pguid = pvd->vdev_guid;
4885 ppguid = ppvd->vdev_guid;
4888 * If we have just finished replacing a hot spared device, then
4889 * we need to detach the parent's first child (the original hot
4892 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
4893 ppvd->vdev_children == 2) {
4894 ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
4895 sguid = ppvd->vdev_child[1]->vdev_guid;
4897 spa_config_exit(spa, SCL_ALL, FTAG);
4898 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
4900 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
4902 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4905 spa_config_exit(spa, SCL_ALL, FTAG);
4909 * Update the stored path or FRU for this vdev.
4912 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
4916 boolean_t sync = B_FALSE;
4918 ASSERT(spa_writeable(spa));
4920 spa_vdev_state_enter(spa, SCL_ALL);
4922 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
4923 return (spa_vdev_state_exit(spa, NULL, ENOENT));
4925 if (!vd->vdev_ops->vdev_op_leaf)
4926 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
4929 if (strcmp(value, vd->vdev_path) != 0) {
4930 spa_strfree(vd->vdev_path);
4931 vd->vdev_path = spa_strdup(value);
4935 if (vd->vdev_fru == NULL) {
4936 vd->vdev_fru = spa_strdup(value);
4938 } else if (strcmp(value, vd->vdev_fru) != 0) {
4939 spa_strfree(vd->vdev_fru);
4940 vd->vdev_fru = spa_strdup(value);
4945 return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
4949 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
4951 return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
4955 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
4957 return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
4961 * ==========================================================================
4963 * ==========================================================================
4967 spa_scan_stop(spa_t *spa)
4969 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4970 if (dsl_scan_resilvering(spa->spa_dsl_pool))
4972 return (dsl_scan_cancel(spa->spa_dsl_pool));
4976 spa_scan(spa_t *spa, pool_scan_func_t func)
4978 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4980 if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
4984 * If a resilver was requested, but there is no DTL on a
4985 * writeable leaf device, we have nothing to do.
4987 if (func == POOL_SCAN_RESILVER &&
4988 !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
4989 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
4993 return (dsl_scan(spa->spa_dsl_pool, func));
4997 * ==========================================================================
4998 * SPA async task processing
4999 * ==========================================================================
5003 spa_async_remove(spa_t *spa, vdev_t *vd)
5007 if (vd->vdev_remove_wanted) {
5008 vd->vdev_remove_wanted = B_FALSE;
5009 vd->vdev_delayed_close = B_FALSE;
5010 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5013 * We want to clear the stats, but we don't want to do a full
5014 * vdev_clear() as that will cause us to throw away
5015 * degraded/faulted state as well as attempt to reopen the
5016 * device, all of which is a waste.
5018 vd->vdev_stat.vs_read_errors = 0;
5019 vd->vdev_stat.vs_write_errors = 0;
5020 vd->vdev_stat.vs_checksum_errors = 0;
5022 vdev_state_dirty(vd->vdev_top);
5025 for (c = 0; c < vd->vdev_children; c++)
5026 spa_async_remove(spa, vd->vdev_child[c]);
5030 spa_async_probe(spa_t *spa, vdev_t *vd)
5034 if (vd->vdev_probe_wanted) {
5035 vd->vdev_probe_wanted = B_FALSE;
5036 vdev_reopen(vd); /* vdev_open() does the actual probe */
5039 for (c = 0; c < vd->vdev_children; c++)
5040 spa_async_probe(spa, vd->vdev_child[c]);
5044 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5048 if (!spa->spa_autoexpand)
5051 for (c = 0; c < vd->vdev_children; c++) {
5052 vdev_t *cvd = vd->vdev_child[c];
5053 spa_async_autoexpand(spa, cvd);
5056 if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5059 spa_event_notify(vd->vdev_spa, vd, FM_EREPORT_ZFS_DEVICE_AUTOEXPAND);
5063 spa_async_thread(spa_t *spa)
5067 ASSERT(spa->spa_sync_on);
5069 mutex_enter(&spa->spa_async_lock);
5070 tasks = spa->spa_async_tasks;
5071 spa->spa_async_tasks = 0;
5072 mutex_exit(&spa->spa_async_lock);
5075 * See if the config needs to be updated.
5077 if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5078 uint64_t old_space, new_space;
5080 mutex_enter(&spa_namespace_lock);
5081 old_space = metaslab_class_get_space(spa_normal_class(spa));
5082 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5083 new_space = metaslab_class_get_space(spa_normal_class(spa));
5084 mutex_exit(&spa_namespace_lock);
5087 * If the pool grew as a result of the config update,
5088 * then log an internal history event.
5090 if (new_space != old_space) {
5091 spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5093 "pool '%s' size: %llu(+%llu)",
5094 spa_name(spa), new_space, new_space - old_space);
5099 * See if any devices need to be marked REMOVED.
5101 if (tasks & SPA_ASYNC_REMOVE) {
5102 spa_vdev_state_enter(spa, SCL_NONE);
5103 spa_async_remove(spa, spa->spa_root_vdev);
5104 for (i = 0; i < spa->spa_l2cache.sav_count; i++)
5105 spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5106 for (i = 0; i < spa->spa_spares.sav_count; i++)
5107 spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5108 (void) spa_vdev_state_exit(spa, NULL, 0);
5111 if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5112 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5113 spa_async_autoexpand(spa, spa->spa_root_vdev);
5114 spa_config_exit(spa, SCL_CONFIG, FTAG);
5118 * See if any devices need to be probed.
5120 if (tasks & SPA_ASYNC_PROBE) {
5121 spa_vdev_state_enter(spa, SCL_NONE);
5122 spa_async_probe(spa, spa->spa_root_vdev);
5123 (void) spa_vdev_state_exit(spa, NULL, 0);
5127 * If any devices are done replacing, detach them.
5129 if (tasks & SPA_ASYNC_RESILVER_DONE)
5130 spa_vdev_resilver_done(spa);
5133 * Kick off a resilver.
5135 if (tasks & SPA_ASYNC_RESILVER)
5136 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5139 * Let the world know that we're done.
5141 mutex_enter(&spa->spa_async_lock);
5142 spa->spa_async_thread = NULL;
5143 cv_broadcast(&spa->spa_async_cv);
5144 mutex_exit(&spa->spa_async_lock);
5149 spa_async_suspend(spa_t *spa)
5151 mutex_enter(&spa->spa_async_lock);
5152 spa->spa_async_suspended++;
5153 while (spa->spa_async_thread != NULL)
5154 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5155 mutex_exit(&spa->spa_async_lock);
5159 spa_async_resume(spa_t *spa)
5161 mutex_enter(&spa->spa_async_lock);
5162 ASSERT(spa->spa_async_suspended != 0);
5163 spa->spa_async_suspended--;
5164 mutex_exit(&spa->spa_async_lock);
5168 spa_async_dispatch(spa_t *spa)
5170 mutex_enter(&spa->spa_async_lock);
5171 if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5172 spa->spa_async_thread == NULL &&
5173 rootdir != NULL && !vn_is_readonly(rootdir))
5174 spa->spa_async_thread = thread_create(NULL, 0,
5175 spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5176 mutex_exit(&spa->spa_async_lock);
5180 spa_async_request(spa_t *spa, int task)
5182 zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5183 mutex_enter(&spa->spa_async_lock);
5184 spa->spa_async_tasks |= task;
5185 mutex_exit(&spa->spa_async_lock);
5189 * ==========================================================================
5190 * SPA syncing routines
5191 * ==========================================================================
5195 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5198 bpobj_enqueue(bpo, bp, tx);
5203 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5207 zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5213 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5215 char *packed = NULL;
5220 VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5223 * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5224 * information. This avoids the dbuf_will_dirty() path and
5225 * saves us a pre-read to get data we don't actually care about.
5227 bufsize = P2ROUNDUP(nvsize, SPA_CONFIG_BLOCKSIZE);
5228 packed = vmem_alloc(bufsize, KM_SLEEP);
5230 VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5232 bzero(packed + nvsize, bufsize - nvsize);
5234 dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5236 vmem_free(packed, bufsize);
5238 VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5239 dmu_buf_will_dirty(db, tx);
5240 *(uint64_t *)db->db_data = nvsize;
5241 dmu_buf_rele(db, FTAG);
5245 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5246 const char *config, const char *entry)
5256 * Update the MOS nvlist describing the list of available devices.
5257 * spa_validate_aux() will have already made sure this nvlist is
5258 * valid and the vdevs are labeled appropriately.
5260 if (sav->sav_object == 0) {
5261 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5262 DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5263 sizeof (uint64_t), tx);
5264 VERIFY(zap_update(spa->spa_meta_objset,
5265 DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5266 &sav->sav_object, tx) == 0);
5269 VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5270 if (sav->sav_count == 0) {
5271 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5273 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5274 for (i = 0; i < sav->sav_count; i++)
5275 list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5276 B_FALSE, VDEV_CONFIG_L2CACHE);
5277 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5278 sav->sav_count) == 0);
5279 for (i = 0; i < sav->sav_count; i++)
5280 nvlist_free(list[i]);
5281 kmem_free(list, sav->sav_count * sizeof (void *));
5284 spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5285 nvlist_free(nvroot);
5287 sav->sav_sync = B_FALSE;
5291 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5295 if (list_is_empty(&spa->spa_config_dirty_list))
5298 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5300 config = spa_config_generate(spa, spa->spa_root_vdev,
5301 dmu_tx_get_txg(tx), B_FALSE);
5303 spa_config_exit(spa, SCL_STATE, FTAG);
5305 if (spa->spa_config_syncing)
5306 nvlist_free(spa->spa_config_syncing);
5307 spa->spa_config_syncing = config;
5309 spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5313 * Set zpool properties.
5316 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5319 objset_t *mos = spa->spa_meta_objset;
5320 nvlist_t *nvp = arg2;
5325 const char *propname;
5326 zprop_type_t proptype;
5328 mutex_enter(&spa->spa_props_lock);
5331 while ((elem = nvlist_next_nvpair(nvp, elem))) {
5332 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5333 case ZPOOL_PROP_VERSION:
5335 * Only set version for non-zpool-creation cases
5336 * (set/import). spa_create() needs special care
5337 * for version setting.
5339 if (tx->tx_txg != TXG_INITIAL) {
5340 VERIFY(nvpair_value_uint64(elem,
5342 ASSERT(intval <= SPA_VERSION);
5343 ASSERT(intval >= spa_version(spa));
5344 spa->spa_uberblock.ub_version = intval;
5345 vdev_config_dirty(spa->spa_root_vdev);
5349 case ZPOOL_PROP_ALTROOT:
5351 * 'altroot' is a non-persistent property. It should
5352 * have been set temporarily at creation or import time.
5354 ASSERT(spa->spa_root != NULL);
5357 case ZPOOL_PROP_READONLY:
5358 case ZPOOL_PROP_CACHEFILE:
5360 * 'readonly' and 'cachefile' are also non-persisitent
5366 * Set pool property values in the poolprops mos object.
5368 if (spa->spa_pool_props_object == 0) {
5369 VERIFY((spa->spa_pool_props_object =
5370 zap_create(mos, DMU_OT_POOL_PROPS,
5371 DMU_OT_NONE, 0, tx)) > 0);
5373 VERIFY(zap_update(mos,
5374 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5375 8, 1, &spa->spa_pool_props_object, tx)
5379 /* normalize the property name */
5380 propname = zpool_prop_to_name(prop);
5381 proptype = zpool_prop_get_type(prop);
5383 if (nvpair_type(elem) == DATA_TYPE_STRING) {
5384 ASSERT(proptype == PROP_TYPE_STRING);
5385 VERIFY(nvpair_value_string(elem, &strval) == 0);
5386 VERIFY(zap_update(mos,
5387 spa->spa_pool_props_object, propname,
5388 1, strlen(strval) + 1, strval, tx) == 0);
5390 } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5391 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5393 if (proptype == PROP_TYPE_INDEX) {
5395 VERIFY(zpool_prop_index_to_string(
5396 prop, intval, &unused) == 0);
5398 VERIFY(zap_update(mos,
5399 spa->spa_pool_props_object, propname,
5400 8, 1, &intval, tx) == 0);
5402 ASSERT(0); /* not allowed */
5406 case ZPOOL_PROP_DELEGATION:
5407 spa->spa_delegation = intval;
5409 case ZPOOL_PROP_BOOTFS:
5410 spa->spa_bootfs = intval;
5412 case ZPOOL_PROP_FAILUREMODE:
5413 spa->spa_failmode = intval;
5415 case ZPOOL_PROP_AUTOEXPAND:
5416 spa->spa_autoexpand = intval;
5417 if (tx->tx_txg != TXG_INITIAL)
5418 spa_async_request(spa,
5419 SPA_ASYNC_AUTOEXPAND);
5421 case ZPOOL_PROP_DEDUPDITTO:
5422 spa->spa_dedup_ditto = intval;
5429 /* log internal history if this is not a zpool create */
5430 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5431 tx->tx_txg != TXG_INITIAL) {
5432 spa_history_log_internal(LOG_POOL_PROPSET,
5433 spa, tx, "%s %lld %s",
5434 nvpair_name(elem), intval, spa_name(spa));
5438 mutex_exit(&spa->spa_props_lock);
5442 * Perform one-time upgrade on-disk changes. spa_version() does not
5443 * reflect the new version this txg, so there must be no changes this
5444 * txg to anything that the upgrade code depends on after it executes.
5445 * Therefore this must be called after dsl_pool_sync() does the sync
5449 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5451 dsl_pool_t *dp = spa->spa_dsl_pool;
5453 ASSERT(spa->spa_sync_pass == 1);
5455 if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5456 spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5457 dsl_pool_create_origin(dp, tx);
5459 /* Keeping the origin open increases spa_minref */
5460 spa->spa_minref += 3;
5463 if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5464 spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5465 dsl_pool_upgrade_clones(dp, tx);
5468 if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5469 spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5470 dsl_pool_upgrade_dir_clones(dp, tx);
5472 /* Keeping the freedir open increases spa_minref */
5473 spa->spa_minref += 3;
5478 * Sync the specified transaction group. New blocks may be dirtied as
5479 * part of the process, so we iterate until it converges.
5482 spa_sync(spa_t *spa, uint64_t txg)
5484 dsl_pool_t *dp = spa->spa_dsl_pool;
5485 objset_t *mos = spa->spa_meta_objset;
5486 bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5487 bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5488 vdev_t *rvd = spa->spa_root_vdev;
5494 VERIFY(spa_writeable(spa));
5497 * Lock out configuration changes.
5499 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5501 spa->spa_syncing_txg = txg;
5502 spa->spa_sync_pass = 0;
5505 * If there are any pending vdev state changes, convert them
5506 * into config changes that go out with this transaction group.
5508 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5509 while (list_head(&spa->spa_state_dirty_list) != NULL) {
5511 * We need the write lock here because, for aux vdevs,
5512 * calling vdev_config_dirty() modifies sav_config.
5513 * This is ugly and will become unnecessary when we
5514 * eliminate the aux vdev wart by integrating all vdevs
5515 * into the root vdev tree.
5517 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5518 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5519 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5520 vdev_state_clean(vd);
5521 vdev_config_dirty(vd);
5523 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5524 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5526 spa_config_exit(spa, SCL_STATE, FTAG);
5528 tx = dmu_tx_create_assigned(dp, txg);
5531 * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5532 * set spa_deflate if we have no raid-z vdevs.
5534 if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5535 spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5538 for (i = 0; i < rvd->vdev_children; i++) {
5539 vd = rvd->vdev_child[i];
5540 if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5543 if (i == rvd->vdev_children) {
5544 spa->spa_deflate = TRUE;
5545 VERIFY(0 == zap_add(spa->spa_meta_objset,
5546 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5547 sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5552 * If anything has changed in this txg, or if someone is waiting
5553 * for this txg to sync (eg, spa_vdev_remove()), push the
5554 * deferred frees from the previous txg. If not, leave them
5555 * alone so that we don't generate work on an otherwise idle
5558 if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5559 !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5560 !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5561 ((dsl_scan_active(dp->dp_scan) ||
5562 txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5563 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5564 VERIFY3U(bpobj_iterate(defer_bpo,
5565 spa_free_sync_cb, zio, tx), ==, 0);
5566 VERIFY3U(zio_wait(zio), ==, 0);
5570 * Iterate to convergence.
5573 int pass = ++spa->spa_sync_pass;
5575 spa_sync_config_object(spa, tx);
5576 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5577 ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5578 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5579 ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5580 spa_errlog_sync(spa, txg);
5581 dsl_pool_sync(dp, txg);
5583 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5584 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5585 bplist_iterate(free_bpl, spa_free_sync_cb,
5587 VERIFY(zio_wait(zio) == 0);
5589 bplist_iterate(free_bpl, bpobj_enqueue_cb,
5594 dsl_scan_sync(dp, tx);
5596 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
5600 spa_sync_upgrades(spa, tx);
5602 } while (dmu_objset_is_dirty(mos, txg));
5605 * Rewrite the vdev configuration (which includes the uberblock)
5606 * to commit the transaction group.
5608 * If there are no dirty vdevs, we sync the uberblock to a few
5609 * random top-level vdevs that are known to be visible in the
5610 * config cache (see spa_vdev_add() for a complete description).
5611 * If there *are* dirty vdevs, sync the uberblock to all vdevs.
5615 * We hold SCL_STATE to prevent vdev open/close/etc.
5616 * while we're attempting to write the vdev labels.
5618 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5620 if (list_is_empty(&spa->spa_config_dirty_list)) {
5621 vdev_t *svd[SPA_DVAS_PER_BP];
5623 int children = rvd->vdev_children;
5624 int c0 = spa_get_random(children);
5626 for (c = 0; c < children; c++) {
5627 vd = rvd->vdev_child[(c0 + c) % children];
5628 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
5630 svd[svdcount++] = vd;
5631 if (svdcount == SPA_DVAS_PER_BP)
5634 error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
5636 error = vdev_config_sync(svd, svdcount, txg,
5639 error = vdev_config_sync(rvd->vdev_child,
5640 rvd->vdev_children, txg, B_FALSE);
5642 error = vdev_config_sync(rvd->vdev_child,
5643 rvd->vdev_children, txg, B_TRUE);
5646 spa_config_exit(spa, SCL_STATE, FTAG);
5650 zio_suspend(spa, NULL);
5651 zio_resume_wait(spa);
5656 * Clear the dirty config list.
5658 while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
5659 vdev_config_clean(vd);
5662 * Now that the new config has synced transactionally,
5663 * let it become visible to the config cache.
5665 if (spa->spa_config_syncing != NULL) {
5666 spa_config_set(spa, spa->spa_config_syncing);
5667 spa->spa_config_txg = txg;
5668 spa->spa_config_syncing = NULL;
5671 spa->spa_ubsync = spa->spa_uberblock;
5673 dsl_pool_sync_done(dp, txg);
5676 * Update usable space statistics.
5678 while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
5679 vdev_sync_done(vd, txg);
5681 spa_update_dspace(spa);
5684 * It had better be the case that we didn't dirty anything
5685 * since vdev_config_sync().
5687 ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
5688 ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
5689 ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
5691 spa->spa_sync_pass = 0;
5693 spa_config_exit(spa, SCL_CONFIG, FTAG);
5695 spa_handle_ignored_writes(spa);
5698 * If any async tasks have been requested, kick them off.
5700 spa_async_dispatch(spa);
5704 * Sync all pools. We don't want to hold the namespace lock across these
5705 * operations, so we take a reference on the spa_t and drop the lock during the
5709 spa_sync_allpools(void)
5712 mutex_enter(&spa_namespace_lock);
5713 while ((spa = spa_next(spa)) != NULL) {
5714 if (spa_state(spa) != POOL_STATE_ACTIVE ||
5715 !spa_writeable(spa) || spa_suspended(spa))
5717 spa_open_ref(spa, FTAG);
5718 mutex_exit(&spa_namespace_lock);
5719 txg_wait_synced(spa_get_dsl(spa), 0);
5720 mutex_enter(&spa_namespace_lock);
5721 spa_close(spa, FTAG);
5723 mutex_exit(&spa_namespace_lock);
5727 * ==========================================================================
5728 * Miscellaneous routines
5729 * ==========================================================================
5733 * Remove all pools in the system.
5741 * Remove all cached state. All pools should be closed now,
5742 * so every spa in the AVL tree should be unreferenced.
5744 mutex_enter(&spa_namespace_lock);
5745 while ((spa = spa_next(NULL)) != NULL) {
5747 * Stop async tasks. The async thread may need to detach
5748 * a device that's been replaced, which requires grabbing
5749 * spa_namespace_lock, so we must drop it here.
5751 spa_open_ref(spa, FTAG);
5752 mutex_exit(&spa_namespace_lock);
5753 spa_async_suspend(spa);
5754 mutex_enter(&spa_namespace_lock);
5755 spa_close(spa, FTAG);
5757 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
5759 spa_deactivate(spa);
5763 mutex_exit(&spa_namespace_lock);
5767 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
5772 if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
5776 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
5777 vd = spa->spa_l2cache.sav_vdevs[i];
5778 if (vd->vdev_guid == guid)
5782 for (i = 0; i < spa->spa_spares.sav_count; i++) {
5783 vd = spa->spa_spares.sav_vdevs[i];
5784 if (vd->vdev_guid == guid)
5793 spa_upgrade(spa_t *spa, uint64_t version)
5795 ASSERT(spa_writeable(spa));
5797 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5800 * This should only be called for a non-faulted pool, and since a
5801 * future version would result in an unopenable pool, this shouldn't be
5804 ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
5805 ASSERT(version >= spa->spa_uberblock.ub_version);
5807 spa->spa_uberblock.ub_version = version;
5808 vdev_config_dirty(spa->spa_root_vdev);
5810 spa_config_exit(spa, SCL_ALL, FTAG);
5812 txg_wait_synced(spa_get_dsl(spa), 0);
5816 spa_has_spare(spa_t *spa, uint64_t guid)
5820 spa_aux_vdev_t *sav = &spa->spa_spares;
5822 for (i = 0; i < sav->sav_count; i++)
5823 if (sav->sav_vdevs[i]->vdev_guid == guid)
5826 for (i = 0; i < sav->sav_npending; i++) {
5827 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
5828 &spareguid) == 0 && spareguid == guid)
5836 * Check if a pool has an active shared spare device.
5837 * Note: reference count of an active spare is 2, as a spare and as a replace
5840 spa_has_active_shared_spare(spa_t *spa)
5844 spa_aux_vdev_t *sav = &spa->spa_spares;
5846 for (i = 0; i < sav->sav_count; i++) {
5847 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
5848 &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
5857 * Post a FM_EREPORT_ZFS_* event from sys/fm/fs/zfs.h. The payload will be
5858 * filled in from the spa and (optionally) the vdev. This doesn't do anything
5859 * in the userland libzpool, as we don't want consumers to misinterpret ztest
5860 * or zdb as real changes.
5863 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
5866 zfs_ereport_post(name, spa, vd, NULL, 0, 0);
5870 #if defined(_KERNEL) && defined(HAVE_SPL)
5871 /* state manipulation functions */
5872 EXPORT_SYMBOL(spa_open);
5873 EXPORT_SYMBOL(spa_open_rewind);
5874 EXPORT_SYMBOL(spa_get_stats);
5875 EXPORT_SYMBOL(spa_create);
5876 EXPORT_SYMBOL(spa_import_rootpool);
5877 EXPORT_SYMBOL(spa_import);
5878 EXPORT_SYMBOL(spa_tryimport);
5879 EXPORT_SYMBOL(spa_destroy);
5880 EXPORT_SYMBOL(spa_export);
5881 EXPORT_SYMBOL(spa_reset);
5882 EXPORT_SYMBOL(spa_async_request);
5883 EXPORT_SYMBOL(spa_async_suspend);
5884 EXPORT_SYMBOL(spa_async_resume);
5885 EXPORT_SYMBOL(spa_inject_addref);
5886 EXPORT_SYMBOL(spa_inject_delref);
5887 EXPORT_SYMBOL(spa_scan_stat_init);
5888 EXPORT_SYMBOL(spa_scan_get_stats);
5890 /* device maniion */
5891 EXPORT_SYMBOL(spa_vdev_add);
5892 EXPORT_SYMBOL(spa_vdev_attach);
5893 EXPORT_SYMBOL(spa_vdev_detach);
5894 EXPORT_SYMBOL(spa_vdev_remove);
5895 EXPORT_SYMBOL(spa_vdev_setpath);
5896 EXPORT_SYMBOL(spa_vdev_setfru);
5897 EXPORT_SYMBOL(spa_vdev_split_mirror);
5899 /* spare statech is global across all pools) */
5900 EXPORT_SYMBOL(spa_spare_add);
5901 EXPORT_SYMBOL(spa_spare_remove);
5902 EXPORT_SYMBOL(spa_spare_exists);
5903 EXPORT_SYMBOL(spa_spare_activate);
5905 /* L2ARC statech is global across all pools) */
5906 EXPORT_SYMBOL(spa_l2cache_add);
5907 EXPORT_SYMBOL(spa_l2cache_remove);
5908 EXPORT_SYMBOL(spa_l2cache_exists);
5909 EXPORT_SYMBOL(spa_l2cache_activate);
5910 EXPORT_SYMBOL(spa_l2cache_drop);
5913 EXPORT_SYMBOL(spa_scan);
5914 EXPORT_SYMBOL(spa_scan_stop);
5917 EXPORT_SYMBOL(spa_sync); /* only for DMU use */
5918 EXPORT_SYMBOL(spa_sync_allpools);
5921 EXPORT_SYMBOL(spa_prop_set);
5922 EXPORT_SYMBOL(spa_prop_get);
5923 EXPORT_SYMBOL(spa_prop_clear_bootfs);
5925 /* asynchronous event notification */
5926 EXPORT_SYMBOL(spa_event_notify);