]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/commitdiff
UBUNTU: SAUCE: (noup) remove completely the zfs code
authorKleber Sacilotto de Souza <kleber.souza@canonical.com>
Tue, 13 Aug 2019 14:51:31 +0000 (16:51 +0200)
committerKleber Sacilotto de Souza <kleber.souza@canonical.com>
Wed, 14 Aug 2019 09:18:49 +0000 (11:18 +0200)
BugLink: https://bugs.launchpad.net/bugs/1840051
Commit 1d1ecf03f066 "UBUNTU: [Packaging] dkms -- drop zfs/spl source
code from kernel" didn't remove completely the zfs source code from
Bionic tree.

Signed-off-by: Kleber Sacilotto de Souza <kleber.souza@canonical.com>
Acked-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Tyler Hicks <tyhicks@canonical.com>
Signed-off-by: Kleber Sacilotto de Souza <kleber.souza@canonical.com>
zfs/META [deleted file]
zfs/include/sys/spa_impl.h [deleted file]
zfs/module/zfs/dmu_traverse.c [deleted file]
zfs/module/zfs/spa.c [deleted file]

diff --git a/zfs/META b/zfs/META
deleted file mode 100644 (file)
index 0823859..0000000
--- a/zfs/META
+++ /dev/null
@@ -1,8 +0,0 @@
-Meta:         1
-Name:         zfs
-Branch:       1.0
-Version:      0.7.5
-Release:      1ubuntu16.6
-Release-Tags: relext
-License:      CDDL
-Author:       OpenZFS on Linux
diff --git a/zfs/include/sys/spa_impl.h b/zfs/include/sys/spa_impl.h
deleted file mode 100644 (file)
index 34b1199..0000000
+++ /dev/null
@@ -1,306 +0,0 @@
-/*
- * CDDL HEADER START
- *
- * The contents of this file are subject to the terms of the
- * Common Development and Distribution License (the "License").
- * You may not use this file except in compliance with the License.
- *
- * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
- * or http://www.opensolaris.org/os/licensing.
- * See the License for the specific language governing permissions
- * and limitations under the License.
- *
- * When distributing Covered Code, include this CDDL HEADER in each
- * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
- * If applicable, add the following below this CDDL HEADER, with the
- * fields enclosed by brackets "[]" replaced with your own identifying
- * information: Portions Copyright [yyyy] [name of copyright owner]
- *
- * CDDL HEADER END
- */
-/*
- * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
- * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
- * Copyright 2013 Saso Kiselkov. All rights reserved.
- * Copyright (c) 2016 Actifio, Inc. All rights reserved.
- * Copyright (c) 2017 Datto Inc.
- */
-
-#ifndef _SYS_SPA_IMPL_H
-#define        _SYS_SPA_IMPL_H
-
-#include <sys/spa.h>
-#include <sys/vdev.h>
-#include <sys/metaslab.h>
-#include <sys/dmu.h>
-#include <sys/dsl_pool.h>
-#include <sys/uberblock_impl.h>
-#include <sys/zfs_context.h>
-#include <sys/avl.h>
-#include <sys/refcount.h>
-#include <sys/bplist.h>
-#include <sys/bpobj.h>
-#include <sys/zfeature.h>
-#include <zfeature_common.h>
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-typedef struct spa_error_entry {
-       zbookmark_phys_t        se_bookmark;
-       char                    *se_name;
-       avl_node_t              se_avl;
-} spa_error_entry_t;
-
-typedef struct spa_history_phys {
-       uint64_t sh_pool_create_len;    /* ending offset of zpool create */
-       uint64_t sh_phys_max_off;       /* physical EOF */
-       uint64_t sh_bof;                /* logical BOF */
-       uint64_t sh_eof;                /* logical EOF */
-       uint64_t sh_records_lost;       /* num of records overwritten */
-} spa_history_phys_t;
-
-struct spa_aux_vdev {
-       uint64_t        sav_object;             /* MOS object for device list */
-       nvlist_t        *sav_config;            /* cached device config */
-       vdev_t          **sav_vdevs;            /* devices */
-       int             sav_count;              /* number devices */
-       boolean_t       sav_sync;               /* sync the device list */
-       nvlist_t        **sav_pending;          /* pending device additions */
-       uint_t          sav_npending;           /* # pending devices */
-};
-
-typedef struct spa_config_lock {
-       kmutex_t        scl_lock;
-       kthread_t       *scl_writer;
-       int             scl_write_wanted;
-       kcondvar_t      scl_cv;
-       refcount_t      scl_count;
-} spa_config_lock_t;
-
-typedef struct spa_config_dirent {
-       list_node_t     scd_link;
-       char            *scd_path;
-} spa_config_dirent_t;
-
-typedef enum zio_taskq_type {
-       ZIO_TASKQ_ISSUE = 0,
-       ZIO_TASKQ_ISSUE_HIGH,
-       ZIO_TASKQ_INTERRUPT,
-       ZIO_TASKQ_INTERRUPT_HIGH,
-       ZIO_TASKQ_TYPES
-} zio_taskq_type_t;
-
-/*
- * State machine for the zpool-poolname process.  The states transitions
- * are done as follows:
- *
- *     From               To                   Routine
- *     PROC_NONE       -> PROC_CREATED         spa_activate()
- *     PROC_CREATED    -> PROC_ACTIVE          spa_thread()
- *     PROC_ACTIVE     -> PROC_DEACTIVATE      spa_deactivate()
- *     PROC_DEACTIVATE -> PROC_GONE            spa_thread()
- *     PROC_GONE       -> PROC_NONE            spa_deactivate()
- */
-typedef enum spa_proc_state {
-       SPA_PROC_NONE,          /* spa_proc = &p0, no process created */
-       SPA_PROC_CREATED,       /* spa_activate() has proc, is waiting */
-       SPA_PROC_ACTIVE,        /* taskqs created, spa_proc set */
-       SPA_PROC_DEACTIVATE,    /* spa_deactivate() requests process exit */
-       SPA_PROC_GONE           /* spa_thread() is exiting, spa_proc = &p0 */
-} spa_proc_state_t;
-
-typedef struct spa_taskqs {
-       uint_t stqs_count;
-       taskq_t **stqs_taskq;
-} spa_taskqs_t;
-
-typedef enum spa_all_vdev_zap_action {
-       AVZ_ACTION_NONE = 0,
-       AVZ_ACTION_DESTROY,     /* Destroy all per-vdev ZAPs and the AVZ. */
-       AVZ_ACTION_REBUILD,     /* Populate the new AVZ, see spa_avz_rebuild */
-       AVZ_ACTION_INITIALIZE
-} spa_avz_action_t;
-
-struct spa {
-       /*
-        * Fields protected by spa_namespace_lock.
-        */
-       char            spa_name[ZFS_MAX_DATASET_NAME_LEN];     /* pool name */
-       char            *spa_comment;           /* comment */
-       avl_node_t      spa_avl;                /* node in spa_namespace_avl */
-       nvlist_t        *spa_config;            /* last synced config */
-       nvlist_t        *spa_config_syncing;    /* currently syncing config */
-       nvlist_t        *spa_config_splitting;  /* config for splitting */
-       nvlist_t        *spa_load_info;         /* info and errors from load */
-       uint64_t        spa_config_txg;         /* txg of last config change */
-       int             spa_sync_pass;          /* iterate-to-convergence */
-       pool_state_t    spa_state;              /* pool state */
-       int             spa_inject_ref;         /* injection references */
-       uint8_t         spa_sync_on;            /* sync threads are running */
-       spa_load_state_t spa_load_state;        /* current load operation */
-       uint64_t        spa_import_flags;       /* import specific flags */
-       spa_taskqs_t    spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES];
-       dsl_pool_t      *spa_dsl_pool;
-       boolean_t       spa_is_initializing;    /* true while opening pool */
-       metaslab_class_t *spa_normal_class;     /* normal data class */
-       metaslab_class_t *spa_log_class;        /* intent log data class */
-       uint64_t        spa_first_txg;          /* first txg after spa_open() */
-       uint64_t        spa_final_txg;          /* txg of export/destroy */
-       uint64_t        spa_freeze_txg;         /* freeze pool at this txg */
-       uint64_t        spa_load_max_txg;       /* best initial ub_txg */
-       uint64_t        spa_claim_max_txg;      /* highest claimed birth txg */
-       timespec_t      spa_loaded_ts;          /* 1st successful open time */
-       objset_t        *spa_meta_objset;       /* copy of dp->dp_meta_objset */
-       kmutex_t        spa_evicting_os_lock;   /* Evicting objset list lock */
-       list_t          spa_evicting_os_list;   /* Objsets being evicted. */
-       kcondvar_t      spa_evicting_os_cv;     /* Objset Eviction Completion */
-       txg_list_t      spa_vdev_txg_list;      /* per-txg dirty vdev list */
-       vdev_t          *spa_root_vdev;         /* top-level vdev container */
-       int             spa_min_ashift;         /* of vdevs in normal class */
-       int             spa_max_ashift;         /* of vdevs in normal class */
-       uint64_t        spa_config_guid;        /* config pool guid */
-       uint64_t        spa_load_guid;          /* spa_load initialized guid */
-       uint64_t        spa_last_synced_guid;   /* last synced guid */
-       list_t          spa_config_dirty_list;  /* vdevs with dirty config */
-       list_t          spa_state_dirty_list;   /* vdevs with dirty state */
-       kmutex_t        spa_alloc_lock;
-       avl_tree_t      spa_alloc_tree;
-       spa_aux_vdev_t  spa_spares;             /* hot spares */
-       spa_aux_vdev_t  spa_l2cache;            /* L2ARC cache devices */
-       nvlist_t        *spa_label_features;    /* Features for reading MOS */
-       uint64_t        spa_config_object;      /* MOS object for pool config */
-       uint64_t        spa_config_generation;  /* config generation number */
-       uint64_t        spa_syncing_txg;        /* txg currently syncing */
-       bpobj_t         spa_deferred_bpobj;     /* deferred-free bplist */
-       bplist_t        spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */
-       zio_cksum_salt_t spa_cksum_salt;        /* secret salt for cksum */
-       /* checksum context templates */
-       kmutex_t        spa_cksum_tmpls_lock;
-       void            *spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS];
-       uberblock_t     spa_ubsync;             /* last synced uberblock */
-       uberblock_t     spa_uberblock;          /* current uberblock */
-       boolean_t       spa_extreme_rewind;     /* rewind past deferred frees */
-       uint64_t        spa_last_io;            /* lbolt of last non-scan I/O */
-       kmutex_t        spa_scrub_lock;         /* resilver/scrub lock */
-       uint64_t        spa_scrub_inflight;     /* in-flight scrub I/Os */
-       kcondvar_t      spa_scrub_io_cv;        /* scrub I/O completion */
-       uint8_t         spa_scrub_active;       /* active or suspended? */
-       uint8_t         spa_scrub_type;         /* type of scrub we're doing */
-       uint8_t         spa_scrub_finished;     /* indicator to rotate logs */
-       uint8_t         spa_scrub_started;      /* started since last boot */
-       uint8_t         spa_scrub_reopen;       /* scrub doing vdev_reopen */
-       uint64_t        spa_scan_pass_start;    /* start time per pass/reboot */
-       uint64_t        spa_scan_pass_scrub_pause; /* scrub pause time */
-       uint64_t        spa_scan_pass_scrub_spent_paused; /* total paused */
-       uint64_t        spa_scan_pass_exam;     /* examined bytes per pass */
-       kmutex_t        spa_async_lock;         /* protect async state */
-       kthread_t       *spa_async_thread;      /* thread doing async task */
-       int             spa_async_suspended;    /* async tasks suspended */
-       kcondvar_t      spa_async_cv;           /* wait for thread_exit() */
-       uint16_t        spa_async_tasks;        /* async task mask */
-       char            *spa_root;              /* alternate root directory */
-       uint64_t        spa_ena;                /* spa-wide ereport ENA */
-       int             spa_last_open_failed;   /* error if last open failed */
-       uint64_t        spa_last_ubsync_txg;    /* "best" uberblock txg */
-       uint64_t        spa_last_ubsync_txg_ts; /* timestamp from that ub */
-       uint64_t        spa_load_txg;           /* ub txg that loaded */
-       uint64_t        spa_load_txg_ts;        /* timestamp from that ub */
-       uint64_t        spa_load_meta_errors;   /* verify metadata err count */
-       uint64_t        spa_load_data_errors;   /* verify data err count */
-       uint64_t        spa_verify_min_txg;     /* start txg of verify scrub */
-       kmutex_t        spa_errlog_lock;        /* error log lock */
-       uint64_t        spa_errlog_last;        /* last error log object */
-       uint64_t        spa_errlog_scrub;       /* scrub error log object */
-       kmutex_t        spa_errlist_lock;       /* error list/ereport lock */
-       avl_tree_t      spa_errlist_last;       /* last error list */
-       avl_tree_t      spa_errlist_scrub;      /* scrub error list */
-       uint64_t        spa_deflate;            /* should we deflate? */
-       uint64_t        spa_history;            /* history object */
-       kmutex_t        spa_history_lock;       /* history lock */
-       vdev_t          *spa_pending_vdev;      /* pending vdev additions */
-       kmutex_t        spa_props_lock;         /* property lock */
-       uint64_t        spa_pool_props_object;  /* object for properties */
-       uint64_t        spa_bootfs;             /* default boot filesystem */
-       uint64_t        spa_failmode;           /* failure mode for the pool */
-       uint64_t        spa_delegation;         /* delegation on/off */
-       list_t          spa_config_list;        /* previous cache file(s) */
-       /* per-CPU array of root of async I/O: */
-       zio_t           **spa_async_zio_root;
-       zio_t           *spa_suspend_zio_root;  /* root of all suspended I/O */
-       kmutex_t        spa_suspend_lock;       /* protects suspend_zio_root */
-       kcondvar_t      spa_suspend_cv;         /* notification of resume */
-       uint8_t         spa_suspended;          /* pool is suspended */
-       uint8_t         spa_claiming;           /* pool is doing zil_claim() */
-       boolean_t       spa_debug;              /* debug enabled? */
-       boolean_t       spa_is_root;            /* pool is root */
-       int             spa_minref;             /* num refs when first opened */
-       int             spa_mode;               /* FREAD | FWRITE */
-       spa_log_state_t spa_log_state;          /* log state */
-       uint64_t        spa_autoexpand;         /* lun expansion on/off */
-       ddt_t           *spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */
-       uint64_t        spa_ddt_stat_object;    /* DDT statistics */
-       uint64_t        spa_dedup_dspace;       /* Cache get_dedup_dspace() */
-       uint64_t        spa_dedup_ditto;        /* dedup ditto threshold */
-       uint64_t        spa_dedup_checksum;     /* default dedup checksum */
-       uint64_t        spa_dspace;             /* dspace in normal class */
-       kmutex_t        spa_vdev_top_lock;      /* dueling offline/remove */
-       kmutex_t        spa_proc_lock;          /* protects spa_proc* */
-       kcondvar_t      spa_proc_cv;            /* spa_proc_state transitions */
-       spa_proc_state_t spa_proc_state;        /* see definition */
-       proc_t          *spa_proc;              /* "zpool-poolname" process */
-       uint64_t        spa_did;                /* if procp != p0, did of t1 */
-       boolean_t       spa_autoreplace;        /* autoreplace set in open */
-       int             spa_vdev_locks;         /* locks grabbed */
-       uint64_t        spa_creation_version;   /* version at pool creation */
-       uint64_t        spa_prev_software_version; /* See ub_software_version */
-       uint64_t        spa_feat_for_write_obj; /* required to write to pool */
-       uint64_t        spa_feat_for_read_obj;  /* required to read from pool */
-       uint64_t        spa_feat_desc_obj;      /* Feature descriptions */
-       uint64_t        spa_feat_enabled_txg_obj; /* Feature enabled txg */
-       kmutex_t        spa_feat_stats_lock;    /* protects spa_feat_stats */
-       nvlist_t        *spa_feat_stats;        /* Cache of enabled features */
-       /* cache feature refcounts */
-       uint64_t        spa_feat_refcount_cache[SPA_FEATURES];
-       taskqid_t       spa_deadman_tqid;       /* Task id */
-       uint64_t        spa_deadman_calls;      /* number of deadman calls */
-       hrtime_t        spa_sync_starttime;     /* starting time of spa_sync */
-       uint64_t        spa_deadman_synctime;   /* deadman expiration timer */
-       uint64_t        spa_all_vdev_zaps;      /* ZAP of per-vd ZAP obj #s */
-       spa_avz_action_t        spa_avz_action; /* destroy/rebuild AVZ? */
-       uint64_t        spa_errata;             /* errata issues detected */
-       spa_stats_t     spa_stats;              /* assorted spa statistics */
-       hrtime_t        spa_ccw_fail_time;      /* Conf cache write fail time */
-       taskq_t         *spa_zvol_taskq;        /* Taskq for minor management */
-       taskq_t         *spa_prefetch_taskq;    /* Taskq for prefetch threads */
-       uint64_t        spa_multihost;          /* multihost aware (mmp) */
-       mmp_thread_t    spa_mmp;                /* multihost mmp thread */
-
-       /*
-        * spa_refcount & spa_config_lock must be the last elements
-        * because refcount_t changes size based on compilation options.
-        * In order for the MDB module to function correctly, the other
-        * fields must remain in the same location.
-        */
-       spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */
-       refcount_t      spa_refcount;           /* number of opens */
-
-       taskq_t         *spa_upgrade_taskq;     /* taskq for upgrade jobs */
-};
-
-extern char *spa_config_path;
-
-extern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
-    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent);
-extern void spa_taskq_dispatch_sync(spa_t *, zio_type_t t, zio_taskq_type_t q,
-    task_func_t *func, void *arg, uint_t flags);
-
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* _SYS_SPA_IMPL_H */
diff --git a/zfs/module/zfs/dmu_traverse.c b/zfs/module/zfs/dmu_traverse.c
deleted file mode 100644 (file)
index c9171ee..0000000
+++ /dev/null
@@ -1,752 +0,0 @@
-/*
- * CDDL HEADER START
- *
- * The contents of this file are subject to the terms of the
- * Common Development and Distribution License (the "License").
- * You may not use this file except in compliance with the License.
- *
- * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
- * or http://www.opensolaris.org/os/licensing.
- * See the License for the specific language governing permissions
- * and limitations under the License.
- *
- * When distributing Covered Code, include this CDDL HEADER in each
- * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
- * If applicable, add the following below this CDDL HEADER, with the
- * fields enclosed by brackets "[]" replaced with your own identifying
- * information: Portions Copyright [yyyy] [name of copyright owner]
- *
- * CDDL HEADER END
- */
-/*
- * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
- */
-
-#include <sys/zfs_context.h>
-#include <sys/dmu_objset.h>
-#include <sys/dmu_traverse.h>
-#include <sys/dsl_dataset.h>
-#include <sys/dsl_dir.h>
-#include <sys/dsl_pool.h>
-#include <sys/dnode.h>
-#include <sys/spa.h>
-#include <sys/spa_impl.h>
-#include <sys/zio.h>
-#include <sys/dmu_impl.h>
-#include <sys/sa.h>
-#include <sys/sa_impl.h>
-#include <sys/callb.h>
-#include <sys/zfeature.h>
-
-int32_t zfs_pd_bytes_max = 50 * 1024 * 1024;   /* 50MB */
-int32_t send_holes_without_birth_time = 1;
-
-typedef struct prefetch_data {
-       kmutex_t pd_mtx;
-       kcondvar_t pd_cv;
-       int32_t pd_bytes_fetched;
-       int pd_flags;
-       boolean_t pd_cancel;
-       boolean_t pd_exited;
-       zbookmark_phys_t pd_resume;
-} prefetch_data_t;
-
-typedef struct traverse_data {
-       spa_t *td_spa;
-       uint64_t td_objset;
-       blkptr_t *td_rootbp;
-       uint64_t td_min_txg;
-       zbookmark_phys_t *td_resume;
-       int td_flags;
-       prefetch_data_t *td_pfd;
-       boolean_t td_paused;
-       uint64_t td_hole_birth_enabled_txg;
-       blkptr_cb_t *td_func;
-       void *td_arg;
-       boolean_t td_realloc_possible;
-} traverse_data_t;
-
-static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
-    uint64_t objset, uint64_t object);
-static void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *,
-    uint64_t objset, uint64_t object);
-
-static int
-traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
-{
-       traverse_data_t *td = arg;
-       zbookmark_phys_t zb;
-
-       if (BP_IS_HOLE(bp))
-               return (0);
-
-       if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
-               return (0);
-
-       SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
-           bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
-
-       (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
-
-       return (0);
-}
-
-static int
-traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
-{
-       traverse_data_t *td = arg;
-
-       if (lrc->lrc_txtype == TX_WRITE) {
-               lr_write_t *lr = (lr_write_t *)lrc;
-               blkptr_t *bp = &lr->lr_blkptr;
-               zbookmark_phys_t zb;
-
-               if (BP_IS_HOLE(bp))
-                       return (0);
-
-               if (claim_txg == 0 || bp->blk_birth < claim_txg)
-                       return (0);
-
-               SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
-                   ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
-
-               (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
-                   td->td_arg);
-       }
-       return (0);
-}
-
-static void
-traverse_zil(traverse_data_t *td, zil_header_t *zh)
-{
-       uint64_t claim_txg = zh->zh_claim_txg;
-       zilog_t *zilog;
-
-       /*
-        * We only want to visit blocks that have been claimed but not yet
-        * replayed; plus, in read-only mode, blocks that are already stable.
-        */
-       if (claim_txg == 0 && spa_writeable(td->td_spa))
-               return;
-
-       zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
-
-       (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
-           claim_txg);
-
-       zil_free(zilog);
-}
-
-typedef enum resume_skip {
-       RESUME_SKIP_ALL,
-       RESUME_SKIP_NONE,
-       RESUME_SKIP_CHILDREN
-} resume_skip_t;
-
-/*
- * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
- * the block indicated by zb does not need to be visited at all. Returns
- * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
- * resume point. This indicates that this block should be visited but not its
- * children (since they must have been visited in a previous traversal).
- * Otherwise returns RESUME_SKIP_NONE.
- */
-static resume_skip_t
-resume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
-    const zbookmark_phys_t *zb)
-{
-       if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
-               /*
-                * If we already visited this bp & everything below,
-                * don't bother doing it again.
-                */
-               if (zbookmark_subtree_completed(dnp, zb, td->td_resume))
-                       return (RESUME_SKIP_ALL);
-
-               /*
-                * If we found the block we're trying to resume from, zero
-                * the bookmark out to indicate that we have resumed.
-                */
-               if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
-                       bzero(td->td_resume, sizeof (*zb));
-                       if (td->td_flags & TRAVERSE_POST)
-                               return (RESUME_SKIP_CHILDREN);
-               }
-       }
-       return (RESUME_SKIP_NONE);
-}
-
-static void
-traverse_prefetch_metadata(traverse_data_t *td,
-    const blkptr_t *bp, const zbookmark_phys_t *zb)
-{
-       arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
-
-       if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
-               return;
-       /*
-        * If we are in the process of resuming, don't prefetch, because
-        * some children will not be needed (and in fact may have already
-        * been freed).
-        */
-       if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
-               return;
-       if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
-               return;
-       if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
-               return;
-
-       (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
-           ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-}
-
-static boolean_t
-prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp)
-{
-       ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA);
-       if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
-           BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
-               return (B_FALSE);
-       return (B_TRUE);
-}
-
-static int
-traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
-    const blkptr_t *bp, const zbookmark_phys_t *zb)
-{
-       int err = 0;
-       arc_buf_t *buf = NULL;
-       prefetch_data_t *pd = td->td_pfd;
-
-       switch (resume_skip_check(td, dnp, zb)) {
-       case RESUME_SKIP_ALL:
-               return (0);
-       case RESUME_SKIP_CHILDREN:
-               goto post;
-       case RESUME_SKIP_NONE:
-               break;
-       default:
-               ASSERT(0);
-       }
-
-       if (bp->blk_birth == 0) {
-               /*
-                * Since this block has a birth time of 0 it must be one of
-                * two things: a hole created before the
-                * SPA_FEATURE_HOLE_BIRTH feature was enabled, or a hole
-                * which has always been a hole in an object.
-                *
-                * If a file is written sparsely, then the unwritten parts of
-                * the file were "always holes" -- that is, they have been
-                * holes since this object was allocated.  However, we (and
-                * our callers) can not necessarily tell when an object was
-                * allocated.  Therefore, if it's possible that this object
-                * was freed and then its object number reused, we need to
-                * visit all the holes with birth==0.
-                *
-                * If it isn't possible that the object number was reused,
-                * then if SPA_FEATURE_HOLE_BIRTH was enabled before we wrote
-                * all the blocks we will visit as part of this traversal,
-                * then this hole must have always existed, so we can skip
-                * it.  We visit blocks born after (exclusive) td_min_txg.
-                *
-                * Note that the meta-dnode cannot be reallocated.
-                */
-               if (!send_holes_without_birth_time &&
-                   (!td->td_realloc_possible ||
-                   zb->zb_object == DMU_META_DNODE_OBJECT) &&
-                   td->td_hole_birth_enabled_txg <= td->td_min_txg)
-                       return (0);
-       } else if (bp->blk_birth <= td->td_min_txg) {
-               return (0);
-       }
-
-       if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
-               uint64_t size = BP_GET_LSIZE(bp);
-               mutex_enter(&pd->pd_mtx);
-               ASSERT(pd->pd_bytes_fetched >= 0);
-               while (pd->pd_bytes_fetched < size && !pd->pd_exited)
-                       cv_wait_sig(&pd->pd_cv, &pd->pd_mtx);
-               pd->pd_bytes_fetched -= size;
-               cv_broadcast(&pd->pd_cv);
-               mutex_exit(&pd->pd_mtx);
-       }
-
-       if (BP_IS_HOLE(bp)) {
-               err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
-               if (err != 0)
-                       goto post;
-               return (0);
-       }
-
-       if (td->td_flags & TRAVERSE_PRE) {
-               err = td->td_func(td->td_spa, NULL, bp, zb, dnp,
-                   td->td_arg);
-               if (err == TRAVERSE_VISIT_NO_CHILDREN)
-                       return (0);
-               if (err != 0)
-                       goto post;
-       }
-
-       if (BP_GET_LEVEL(bp) > 0) {
-               uint32_t flags = ARC_FLAG_WAIT;
-               int32_t i;
-               int32_t epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
-               zbookmark_phys_t *czb;
-
-               err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
-                   ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-               if (err != 0)
-                       goto post;
-
-               czb = kmem_alloc(sizeof (zbookmark_phys_t), KM_SLEEP);
-
-               for (i = 0; i < epb; i++) {
-                       SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object,
-                           zb->zb_level - 1,
-                           zb->zb_blkid * epb + i);
-                       traverse_prefetch_metadata(td,
-                           &((blkptr_t *)buf->b_data)[i], czb);
-               }
-
-               /* recursively visitbp() blocks below this */
-               for (i = 0; i < epb; i++) {
-                       SET_BOOKMARK(czb, zb->zb_objset, zb->zb_object,
-                           zb->zb_level - 1,
-                           zb->zb_blkid * epb + i);
-                       err = traverse_visitbp(td, dnp,
-                           &((blkptr_t *)buf->b_data)[i], czb);
-                       if (err != 0)
-                               break;
-               }
-
-               kmem_free(czb, sizeof (zbookmark_phys_t));
-
-       } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
-               uint32_t flags = ARC_FLAG_WAIT;
-               int32_t i;
-               int32_t epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
-               dnode_phys_t *child_dnp;
-
-               err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
-                   ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-               if (err != 0)
-                       goto post;
-               child_dnp = buf->b_data;
-
-               for (i = 0; i < epb; i += child_dnp[i].dn_extra_slots + 1) {
-                       prefetch_dnode_metadata(td, &child_dnp[i],
-                           zb->zb_objset, zb->zb_blkid * epb + i);
-               }
-
-               /* recursively visitbp() blocks below this */
-               for (i = 0; i < epb; i += child_dnp[i].dn_extra_slots + 1) {
-                       err = traverse_dnode(td, &child_dnp[i],
-                           zb->zb_objset, zb->zb_blkid * epb + i);
-                       if (err != 0)
-                               break;
-               }
-       } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
-               arc_flags_t flags = ARC_FLAG_WAIT;
-               objset_phys_t *osp;
-
-               err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
-                   ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
-               if (err != 0)
-                       goto post;
-
-               osp = buf->b_data;
-               prefetch_dnode_metadata(td, &osp->os_meta_dnode, zb->zb_objset,
-                   DMU_META_DNODE_OBJECT);
-               /*
-                * See the block comment above for the goal of this variable.
-                * If the maxblkid of the meta-dnode is 0, then we know that
-                * we've never had more than DNODES_PER_BLOCK objects in the
-                * dataset, which means we can't have reused any object ids.
-                */
-               if (osp->os_meta_dnode.dn_maxblkid == 0)
-                       td->td_realloc_possible = B_FALSE;
-
-               if (arc_buf_size(buf) >= sizeof (objset_phys_t)) {
-                       prefetch_dnode_metadata(td, &osp->os_groupused_dnode,
-                           zb->zb_objset, DMU_GROUPUSED_OBJECT);
-                       prefetch_dnode_metadata(td, &osp->os_userused_dnode,
-                           zb->zb_objset, DMU_USERUSED_OBJECT);
-               }
-
-               err = traverse_dnode(td, &osp->os_meta_dnode, zb->zb_objset,
-                   DMU_META_DNODE_OBJECT);
-               if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
-                       err = traverse_dnode(td, &osp->os_groupused_dnode,
-                           zb->zb_objset, DMU_GROUPUSED_OBJECT);
-               }
-               if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
-                       err = traverse_dnode(td, &osp->os_userused_dnode,
-                           zb->zb_objset, DMU_USERUSED_OBJECT);
-               }
-       }
-
-       if (buf)
-               arc_buf_destroy(buf, &buf);
-
-post:
-       if (err == 0 && (td->td_flags & TRAVERSE_POST))
-               err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
-
-       if ((td->td_flags & TRAVERSE_HARD) && (err == EIO || err == ECKSUM)) {
-               /*
-                * Ignore this disk error as requested by the HARD flag,
-                * and continue traversal.
-                */
-               err = 0;
-       }
-
-       /*
-        * If we are stopping here, set td_resume.
-        */
-       if (td->td_resume != NULL && err != 0 && !td->td_paused) {
-               td->td_resume->zb_objset = zb->zb_objset;
-               td->td_resume->zb_object = zb->zb_object;
-               td->td_resume->zb_level = 0;
-               /*
-                * If we have stopped on an indirect block (e.g. due to
-                * i/o error), we have not visited anything below it.
-                * Set the bookmark to the first level-0 block that we need
-                * to visit.  This way, the resuming code does not need to
-                * deal with resuming from indirect blocks.
-                *
-                * Note, if zb_level <= 0, dnp may be NULL, so we don't want
-                * to dereference it.
-                */
-               td->td_resume->zb_blkid = zb->zb_blkid;
-               if (zb->zb_level > 0) {
-                       td->td_resume->zb_blkid <<= zb->zb_level *
-                           (dnp->dn_indblkshift - SPA_BLKPTRSHIFT);
-               }
-               td->td_paused = B_TRUE;
-       }
-
-       return (err);
-}
-
-static void
-prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
-    uint64_t objset, uint64_t object)
-{
-       int j;
-       zbookmark_phys_t czb;
-
-       for (j = 0; j < dnp->dn_nblkptr; j++) {
-               SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
-               traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
-       }
-
-       if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
-               SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
-               traverse_prefetch_metadata(td, DN_SPILL_BLKPTR(dnp), &czb);
-       }
-}
-
-static int
-traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
-    uint64_t objset, uint64_t object)
-{
-       int j, err = 0;
-       zbookmark_phys_t czb;
-
-       if (object != DMU_META_DNODE_OBJECT && td->td_resume != NULL &&
-           object < td->td_resume->zb_object)
-               return (0);
-
-       if (td->td_flags & TRAVERSE_PRE) {
-               SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
-                   ZB_DNODE_BLKID);
-               err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
-                   td->td_arg);
-               if (err == TRAVERSE_VISIT_NO_CHILDREN)
-                       return (0);
-               if (err != 0)
-                       return (err);
-       }
-
-       for (j = 0; j < dnp->dn_nblkptr; j++) {
-               SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
-               err = traverse_visitbp(td, dnp, &dnp->dn_blkptr[j], &czb);
-               if (err != 0)
-                       break;
-       }
-
-       if (err == 0 && (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR)) {
-               SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
-               err = traverse_visitbp(td, dnp, DN_SPILL_BLKPTR(dnp), &czb);
-       }
-
-       if (err == 0 && (td->td_flags & TRAVERSE_POST)) {
-               SET_BOOKMARK(&czb, objset, object, ZB_DNODE_LEVEL,
-                   ZB_DNODE_BLKID);
-               err = td->td_func(td->td_spa, NULL, NULL, &czb, dnp,
-                   td->td_arg);
-               if (err == TRAVERSE_VISIT_NO_CHILDREN)
-                       return (0);
-               if (err != 0)
-                       return (err);
-       }
-       return (err);
-}
-
-/* ARGSUSED */
-static int
-traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
-    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
-{
-       prefetch_data_t *pfd = arg;
-       arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
-
-       ASSERT(pfd->pd_bytes_fetched >= 0);
-       if (bp == NULL)
-               return (0);
-       if (pfd->pd_cancel)
-               return (SET_ERROR(EINTR));
-
-       if (!prefetch_needed(pfd, bp))
-               return (0);
-
-       mutex_enter(&pfd->pd_mtx);
-       while (!pfd->pd_cancel && pfd->pd_bytes_fetched >= zfs_pd_bytes_max)
-               cv_wait_sig(&pfd->pd_cv, &pfd->pd_mtx);
-       pfd->pd_bytes_fetched += BP_GET_LSIZE(bp);
-       cv_broadcast(&pfd->pd_cv);
-       mutex_exit(&pfd->pd_mtx);
-
-       (void) arc_read(NULL, spa, bp, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
-           ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, &aflags, zb);
-
-       return (0);
-}
-
-static void
-traverse_prefetch_thread(void *arg)
-{
-       traverse_data_t *td_main = arg;
-       traverse_data_t td = *td_main;
-       zbookmark_phys_t czb;
-       fstrans_cookie_t cookie = spl_fstrans_mark();
-
-       td.td_func = traverse_prefetcher;
-       td.td_arg = td_main->td_pfd;
-       td.td_pfd = NULL;
-       td.td_resume = &td_main->td_pfd->pd_resume;
-
-       SET_BOOKMARK(&czb, td.td_objset,
-           ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
-       (void) traverse_visitbp(&td, NULL, td.td_rootbp, &czb);
-
-       mutex_enter(&td_main->td_pfd->pd_mtx);
-       td_main->td_pfd->pd_exited = B_TRUE;
-       cv_broadcast(&td_main->td_pfd->pd_cv);
-       mutex_exit(&td_main->td_pfd->pd_mtx);
-       spl_fstrans_unmark(cookie);
-}
-
-/*
- * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
- * in syncing context).
- */
-static int
-traverse_impl(spa_t *spa, dsl_dataset_t *ds, uint64_t objset, blkptr_t *rootbp,
-    uint64_t txg_start, zbookmark_phys_t *resume, int flags,
-    blkptr_cb_t func, void *arg)
-{
-       traverse_data_t *td;
-       prefetch_data_t *pd;
-       zbookmark_phys_t *czb;
-       int err;
-
-       ASSERT(ds == NULL || objset == ds->ds_object);
-       ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));
-
-       td = kmem_alloc(sizeof (traverse_data_t), KM_SLEEP);
-       pd = kmem_zalloc(sizeof (prefetch_data_t), KM_SLEEP);
-       czb = kmem_alloc(sizeof (zbookmark_phys_t), KM_SLEEP);
-
-       td->td_spa = spa;
-       td->td_objset = objset;
-       td->td_rootbp = rootbp;
-       td->td_min_txg = txg_start;
-       td->td_resume = resume;
-       td->td_func = func;
-       td->td_arg = arg;
-       td->td_pfd = pd;
-       td->td_flags = flags;
-       td->td_paused = B_FALSE;
-       td->td_realloc_possible = (txg_start == 0 ? B_FALSE : B_TRUE);
-
-       if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
-               VERIFY(spa_feature_enabled_txg(spa,
-                   SPA_FEATURE_HOLE_BIRTH, &td->td_hole_birth_enabled_txg));
-       } else {
-               td->td_hole_birth_enabled_txg = UINT64_MAX;
-       }
-
-       pd->pd_flags = flags;
-       if (resume != NULL)
-               pd->pd_resume = *resume;
-       mutex_init(&pd->pd_mtx, NULL, MUTEX_DEFAULT, NULL);
-       cv_init(&pd->pd_cv, NULL, CV_DEFAULT, NULL);
-
-       SET_BOOKMARK(czb, td->td_objset,
-           ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
-
-       /* See comment on ZIL traversal in dsl_scan_visitds. */
-       if (ds != NULL && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) {
-               uint32_t flags = ARC_FLAG_WAIT;
-               objset_phys_t *osp;
-               arc_buf_t *buf;
-
-               err = arc_read(NULL, td->td_spa, rootbp,
-                   arc_getbuf_func, &buf,
-                   ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, czb);
-               if (err != 0)
-                       return (err);
-
-               if (err != 0) {
-                       /*
-                        * If both TRAVERSE_HARD and TRAVERSE_PRE are set,
-                        * continue to visitbp so that td_func can be called
-                        * in pre stage, and err will reset to zero.
-                        */
-                       if (!(td->td_flags & TRAVERSE_HARD) ||
-                           !(td->td_flags & TRAVERSE_PRE))
-                               return (err);
-               } else {
-                       osp = buf->b_data;
-                       traverse_zil(td, &osp->os_zil_header);
-                       arc_buf_destroy(buf, &buf);
-               }
-       }
-
-       if (!(flags & TRAVERSE_PREFETCH_DATA) ||
-           taskq_dispatch(spa->spa_prefetch_taskq, traverse_prefetch_thread,
-           td, TQ_NOQUEUE) == TASKQID_INVALID)
-               pd->pd_exited = B_TRUE;
-
-       err = traverse_visitbp(td, NULL, rootbp, czb);
-
-       mutex_enter(&pd->pd_mtx);
-       pd->pd_cancel = B_TRUE;
-       cv_broadcast(&pd->pd_cv);
-       while (!pd->pd_exited)
-               cv_wait_sig(&pd->pd_cv, &pd->pd_mtx);
-       mutex_exit(&pd->pd_mtx);
-
-       mutex_destroy(&pd->pd_mtx);
-       cv_destroy(&pd->pd_cv);
-
-       kmem_free(czb, sizeof (zbookmark_phys_t));
-       kmem_free(pd, sizeof (struct prefetch_data));
-       kmem_free(td, sizeof (struct traverse_data));
-
-       return (err);
-}
-
-/*
- * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
- * in syncing context).
- */
-int
-traverse_dataset_resume(dsl_dataset_t *ds, uint64_t txg_start,
-    zbookmark_phys_t *resume,
-    int flags, blkptr_cb_t func, void *arg)
-{
-       return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds, ds->ds_object,
-           &dsl_dataset_phys(ds)->ds_bp, txg_start, resume, flags, func, arg));
-}
-
-int
-traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start,
-    int flags, blkptr_cb_t func, void *arg)
-{
-       return (traverse_dataset_resume(ds, txg_start, NULL, flags, func, arg));
-}
-
-int
-traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
-    uint64_t txg_start, zbookmark_phys_t *resume, int flags,
-    blkptr_cb_t func, void *arg)
-{
-       return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
-           blkptr, txg_start, resume, flags, func, arg));
-}
-
-/*
- * NB: pool must not be changing on-disk (eg, from zdb or sync context).
- */
-int
-traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
-    blkptr_cb_t func, void *arg)
-{
-       int err;
-       uint64_t obj;
-       dsl_pool_t *dp = spa_get_dsl(spa);
-       objset_t *mos = dp->dp_meta_objset;
-       boolean_t hard = (flags & TRAVERSE_HARD);
-
-       /* visit the MOS */
-       err = traverse_impl(spa, NULL, 0, spa_get_rootblkptr(spa),
-           txg_start, NULL, flags, func, arg);
-       if (err != 0)
-               return (err);
-
-       /* visit each dataset */
-       for (obj = 1; err == 0;
-           err = dmu_object_next(mos, &obj, B_FALSE, txg_start)) {
-               dmu_object_info_t doi;
-
-               err = dmu_object_info(mos, obj, &doi);
-               if (err != 0) {
-                       if (hard)
-                               continue;
-                       break;
-               }
-
-               if (doi.doi_bonus_type == DMU_OT_DSL_DATASET) {
-                       dsl_dataset_t *ds;
-                       uint64_t txg = txg_start;
-
-                       dsl_pool_config_enter(dp, FTAG);
-                       err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
-                       dsl_pool_config_exit(dp, FTAG);
-                       if (err != 0) {
-                               if (hard)
-                                       continue;
-                               break;
-                       }
-                       if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg)
-                               txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
-                       err = traverse_dataset(ds, txg, flags, func, arg);
-                       dsl_dataset_rele(ds, FTAG);
-                       if (err != 0)
-                               break;
-               }
-       }
-       if (err == ESRCH)
-               err = 0;
-       return (err);
-}
-
-#if defined(_KERNEL) && defined(HAVE_SPL)
-EXPORT_SYMBOL(traverse_dataset);
-EXPORT_SYMBOL(traverse_pool);
-
-module_param(zfs_pd_bytes_max, int, 0644);
-MODULE_PARM_DESC(zfs_pd_bytes_max, "Max number of bytes to prefetch");
-
-module_param_named(ignore_hole_birth, send_holes_without_birth_time, int, 0644);
-MODULE_PARM_DESC(ignore_hole_birth, "Alias for send_holes_without_birth_time");
-
-module_param_named(send_holes_without_birth_time,
-       send_holes_without_birth_time, int, 0644);
-MODULE_PARM_DESC(send_holes_without_birth_time,
-       "Ignore hole_birth txg for zfs send");
-#endif
diff --git a/zfs/module/zfs/spa.c b/zfs/module/zfs/spa.c
deleted file mode 100644 (file)
index b777afc..0000000
+++ /dev/null
@@ -1,7321 +0,0 @@
-/*
- * CDDL HEADER START
- *
- * The contents of this file are subject to the terms of the
- * Common Development and Distribution License (the "License").
- * You may not use this file except in compliance with the License.
- *
- * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
- * or http://www.opensolaris.org/os/licensing.
- * See the License for the specific language governing permissions
- * and limitations under the License.
- *
- * When distributing Covered Code, include this CDDL HEADER in each
- * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
- * If applicable, add the following below this CDDL HEADER, with the
- * fields enclosed by brackets "[]" replaced with your own identifying
- * information: Portions Copyright [yyyy] [name of copyright owner]
- *
- * CDDL HEADER END
- */
-
-/*
- * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
- * Copyright (c) 2015, Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2013, 2014, Nexenta Systems, Inc.  All rights reserved.
- * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
- * Copyright 2013 Saso Kiselkov. All rights reserved.
- * Copyright (c) 2014 Integros [integros.com]
- * Copyright 2016 Toomas Soome <tsoome@me.com>
- * Copyright (c) 2016 Actifio, Inc. All rights reserved.
- * Copyright (c) 2017 Datto Inc.
- * Copyright 2017 Joyent, Inc.
- */
-
-/*
- * SPA: Storage Pool Allocator
- *
- * This file contains all the routines used when modifying on-disk SPA state.
- * This includes opening, importing, destroying, exporting a pool, and syncing a
- * pool.
- */
-
-#include <sys/zfs_context.h>
-#include <sys/fm/fs/zfs.h>
-#include <sys/spa_impl.h>
-#include <sys/zio.h>
-#include <sys/zio_checksum.h>
-#include <sys/dmu.h>
-#include <sys/dmu_tx.h>
-#include <sys/zap.h>
-#include <sys/zil.h>
-#include <sys/ddt.h>
-#include <sys/vdev_impl.h>
-#include <sys/vdev_disk.h>
-#include <sys/metaslab.h>
-#include <sys/metaslab_impl.h>
-#include <sys/mmp.h>
-#include <sys/uberblock_impl.h>
-#include <sys/txg.h>
-#include <sys/avl.h>
-#include <sys/dmu_traverse.h>
-#include <sys/dmu_objset.h>
-#include <sys/unique.h>
-#include <sys/dsl_pool.h>
-#include <sys/dsl_dataset.h>
-#include <sys/dsl_dir.h>
-#include <sys/dsl_prop.h>
-#include <sys/dsl_synctask.h>
-#include <sys/fs/zfs.h>
-#include <sys/arc.h>
-#include <sys/callb.h>
-#include <sys/systeminfo.h>
-#include <sys/spa_boot.h>
-#include <sys/zfs_ioctl.h>
-#include <sys/dsl_scan.h>
-#include <sys/zfeature.h>
-#include <sys/dsl_destroy.h>
-#include <sys/zvol.h>
-
-#ifdef _KERNEL
-#include <sys/fm/protocol.h>
-#include <sys/fm/util.h>
-#include <sys/bootprops.h>
-#include <sys/callb.h>
-#include <sys/cpupart.h>
-#include <sys/pool.h>
-#include <sys/sysdc.h>
-#include <sys/zone.h>
-#endif /* _KERNEL */
-
-#include "zfs_prop.h"
-#include "zfs_comutil.h"
-
-/*
- * The interval, in seconds, at which failed configuration cache file writes
- * should be retried.
- */
-static int zfs_ccw_retry_interval = 300;
-
-typedef enum zti_modes {
-       ZTI_MODE_FIXED,                 /* value is # of threads (min 1) */
-       ZTI_MODE_BATCH,                 /* cpu-intensive; value is ignored */
-       ZTI_MODE_NULL,                  /* don't create a taskq */
-       ZTI_NMODES
-} zti_modes_t;
-
-#define        ZTI_P(n, q)     { ZTI_MODE_FIXED, (n), (q) }
-#define        ZTI_PCT(n)      { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
-#define        ZTI_BATCH       { ZTI_MODE_BATCH, 0, 1 }
-#define        ZTI_NULL        { ZTI_MODE_NULL, 0, 0 }
-
-#define        ZTI_N(n)        ZTI_P(n, 1)
-#define        ZTI_ONE         ZTI_N(1)
-
-typedef struct zio_taskq_info {
-       zti_modes_t zti_mode;
-       uint_t zti_value;
-       uint_t zti_count;
-} zio_taskq_info_t;
-
-static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
-       "iss", "iss_h", "int", "int_h"
-};
-
-/*
- * This table defines the taskq settings for each ZFS I/O type. When
- * initializing a pool, we use this table to create an appropriately sized
- * taskq. Some operations are low volume and therefore have a small, static
- * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
- * macros. Other operations process a large amount of data; the ZTI_BATCH
- * macro causes us to create a taskq oriented for throughput. Some operations
- * are so high frequency and short-lived that the taskq itself can become a a
- * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
- * additional degree of parallelism specified by the number of threads per-
- * taskq and the number of taskqs; when dispatching an event in this case, the
- * particular taskq is chosen at random.
- *
- * The different taskq priorities are to handle the different contexts (issue
- * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
- * need to be handled with minimum delay.
- */
-const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
-       /* ISSUE        ISSUE_HIGH      INTR            INTR_HIGH */
-       { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* NULL */
-       { ZTI_N(8),     ZTI_NULL,       ZTI_P(12, 8),   ZTI_NULL }, /* READ */
-       { ZTI_BATCH,    ZTI_N(5),       ZTI_P(12, 8),   ZTI_N(5) }, /* WRITE */
-       { ZTI_P(12, 8), ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* FREE */
-       { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* CLAIM */
-       { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* IOCTL */
-};
-
-static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl,
-    const char *name);
-static void spa_event_post(sysevent_t *ev);
-static void spa_sync_version(void *arg, dmu_tx_t *tx);
-static void spa_sync_props(void *arg, dmu_tx_t *tx);
-static boolean_t spa_has_active_shared_spare(spa_t *spa);
-static inline int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
-    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
-    char **ereport);
-static void spa_vdev_resilver_done(spa_t *spa);
-
-uint_t         zio_taskq_batch_pct = 75;       /* 1 thread per cpu in pset */
-id_t           zio_taskq_psrset_bind = PS_NONE;
-boolean_t      zio_taskq_sysdc = B_TRUE;       /* use SDC scheduling class */
-uint_t         zio_taskq_basedc = 80;          /* base duty cycle */
-
-boolean_t      spa_create_process = B_TRUE;    /* no process ==> no sysdc */
-
-/*
- * This (illegal) pool name is used when temporarily importing a spa_t in order
- * to get the vdev stats associated with the imported devices.
- */
-#define        TRYIMPORT_NAME  "$import"
-
-/*
- * ==========================================================================
- * SPA properties routines
- * ==========================================================================
- */
-
-/*
- * Add a (source=src, propname=propval) list to an nvlist.
- */
-static void
-spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
-    uint64_t intval, zprop_source_t src)
-{
-       const char *propname = zpool_prop_to_name(prop);
-       nvlist_t *propval;
-
-       VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
-       VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
-
-       if (strval != NULL)
-               VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
-       else
-               VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
-
-       VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
-       nvlist_free(propval);
-}
-
-/*
- * Get property values from the spa configuration.
- */
-static void
-spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
-{
-       vdev_t *rvd = spa->spa_root_vdev;
-       dsl_pool_t *pool = spa->spa_dsl_pool;
-       uint64_t size, alloc, cap, version;
-       const zprop_source_t src = ZPROP_SRC_NONE;
-       spa_config_dirent_t *dp;
-       metaslab_class_t *mc = spa_normal_class(spa);
-
-       ASSERT(MUTEX_HELD(&spa->spa_props_lock));
-
-       if (rvd != NULL) {
-               alloc = metaslab_class_get_alloc(spa_normal_class(spa));
-               size = metaslab_class_get_space(spa_normal_class(spa));
-               spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
-               spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
-               spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
-               spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
-                   size - alloc, src);
-
-               spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
-                   metaslab_class_fragmentation(mc), src);
-               spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
-                   metaslab_class_expandable_space(mc), src);
-               spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
-                   (spa_mode(spa) == FREAD), src);
-
-               cap = (size == 0) ? 0 : (alloc * 100 / size);
-               spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
-
-               spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
-                   ddt_get_pool_dedup_ratio(spa), src);
-
-               spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
-                   rvd->vdev_state, src);
-
-               version = spa_version(spa);
-               if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION)) {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
-                           version, ZPROP_SRC_DEFAULT);
-               } else {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL,
-                           version, ZPROP_SRC_LOCAL);
-               }
-       }
-
-       if (pool != NULL) {
-               /*
-                * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
-                * when opening pools before this version freedir will be NULL.
-                */
-               if (pool->dp_free_dir != NULL) {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
-                           dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
-                           src);
-               } else {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
-                           NULL, 0, src);
-               }
-
-               if (pool->dp_leak_dir != NULL) {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
-                           dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
-                           src);
-               } else {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
-                           NULL, 0, src);
-               }
-       }
-
-       spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
-
-       if (spa->spa_comment != NULL) {
-               spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
-                   0, ZPROP_SRC_LOCAL);
-       }
-
-       if (spa->spa_root != NULL)
-               spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
-                   0, ZPROP_SRC_LOCAL);
-
-       if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
-               spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
-                   MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
-       } else {
-               spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
-                   SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
-       }
-
-       if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
-               spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
-                   DNODE_MAX_SIZE, ZPROP_SRC_NONE);
-       } else {
-               spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
-                   DNODE_MIN_SIZE, ZPROP_SRC_NONE);
-       }
-
-       if ((dp = list_head(&spa->spa_config_list)) != NULL) {
-               if (dp->scd_path == NULL) {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
-                           "none", 0, ZPROP_SRC_LOCAL);
-               } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
-                       spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
-                           dp->scd_path, 0, ZPROP_SRC_LOCAL);
-               }
-       }
-}
-
-/*
- * Get zpool property values.
- */
-int
-spa_prop_get(spa_t *spa, nvlist_t **nvp)
-{
-       objset_t *mos = spa->spa_meta_objset;
-       zap_cursor_t zc;
-       zap_attribute_t za;
-       int err;
-
-       err = nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP);
-       if (err)
-               return (err);
-
-       mutex_enter(&spa->spa_props_lock);
-
-       /*
-        * Get properties from the spa config.
-        */
-       spa_prop_get_config(spa, nvp);
-
-       /* If no pool property object, no more prop to get. */
-       if (mos == NULL || spa->spa_pool_props_object == 0) {
-               mutex_exit(&spa->spa_props_lock);
-               goto out;
-       }
-
-       /*
-        * Get properties from the MOS pool property object.
-        */
-       for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
-           (err = zap_cursor_retrieve(&zc, &za)) == 0;
-           zap_cursor_advance(&zc)) {
-               uint64_t intval = 0;
-               char *strval = NULL;
-               zprop_source_t src = ZPROP_SRC_DEFAULT;
-               zpool_prop_t prop;
-
-               if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
-                       continue;
-
-               switch (za.za_integer_length) {
-               case 8:
-                       /* integer property */
-                       if (za.za_first_integer !=
-                           zpool_prop_default_numeric(prop))
-                               src = ZPROP_SRC_LOCAL;
-
-                       if (prop == ZPOOL_PROP_BOOTFS) {
-                               dsl_pool_t *dp;
-                               dsl_dataset_t *ds = NULL;
-
-                               dp = spa_get_dsl(spa);
-                               dsl_pool_config_enter(dp, FTAG);
-                               if ((err = dsl_dataset_hold_obj(dp,
-                                   za.za_first_integer, FTAG, &ds))) {
-                                       dsl_pool_config_exit(dp, FTAG);
-                                       break;
-                               }
-
-                               strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
-                                   KM_SLEEP);
-                               dsl_dataset_name(ds, strval);
-                               dsl_dataset_rele(ds, FTAG);
-                               dsl_pool_config_exit(dp, FTAG);
-                       } else {
-                               strval = NULL;
-                               intval = za.za_first_integer;
-                       }
-
-                       spa_prop_add_list(*nvp, prop, strval, intval, src);
-
-                       if (strval != NULL)
-                               kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
-
-                       break;
-
-               case 1:
-                       /* string property */
-                       strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
-                       err = zap_lookup(mos, spa->spa_pool_props_object,
-                           za.za_name, 1, za.za_num_integers, strval);
-                       if (err) {
-                               kmem_free(strval, za.za_num_integers);
-                               break;
-                       }
-                       spa_prop_add_list(*nvp, prop, strval, 0, src);
-                       kmem_free(strval, za.za_num_integers);
-                       break;
-
-               default:
-                       break;
-               }
-       }
-       zap_cursor_fini(&zc);
-       mutex_exit(&spa->spa_props_lock);
-out:
-       if (err && err != ENOENT) {
-               nvlist_free(*nvp);
-               *nvp = NULL;
-               return (err);
-       }
-
-       return (0);
-}
-
-/*
- * Validate the given pool properties nvlist and modify the list
- * for the property values to be set.
- */
-static int
-spa_prop_validate(spa_t *spa, nvlist_t *props)
-{
-       nvpair_t *elem;
-       int error = 0, reset_bootfs = 0;
-       uint64_t objnum = 0;
-       boolean_t has_feature = B_FALSE;
-
-       elem = NULL;
-       while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
-               uint64_t intval;
-               char *strval, *slash, *check, *fname;
-               const char *propname = nvpair_name(elem);
-               zpool_prop_t prop = zpool_name_to_prop(propname);
-
-               switch ((int)prop) {
-               case ZPROP_INVAL:
-                       if (!zpool_prop_feature(propname)) {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-
-                       /*
-                        * Sanitize the input.
-                        */
-                       if (nvpair_type(elem) != DATA_TYPE_UINT64) {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-
-                       if (nvpair_value_uint64(elem, &intval) != 0) {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-
-                       if (intval != 0) {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-
-                       fname = strchr(propname, '@') + 1;
-                       if (zfeature_lookup_name(fname, NULL) != 0) {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-
-                       has_feature = B_TRUE;
-                       break;
-
-               case ZPOOL_PROP_VERSION:
-                       error = nvpair_value_uint64(elem, &intval);
-                       if (!error &&
-                           (intval < spa_version(spa) ||
-                           intval > SPA_VERSION_BEFORE_FEATURES ||
-                           has_feature))
-                               error = SET_ERROR(EINVAL);
-                       break;
-
-               case ZPOOL_PROP_DELEGATION:
-               case ZPOOL_PROP_AUTOREPLACE:
-               case ZPOOL_PROP_LISTSNAPS:
-               case ZPOOL_PROP_AUTOEXPAND:
-                       error = nvpair_value_uint64(elem, &intval);
-                       if (!error && intval > 1)
-                               error = SET_ERROR(EINVAL);
-                       break;
-
-               case ZPOOL_PROP_MULTIHOST:
-                       error = nvpair_value_uint64(elem, &intval);
-                       if (!error && intval > 1)
-                               error = SET_ERROR(EINVAL);
-
-                       if (!error && !spa_get_hostid())
-                               error = SET_ERROR(ENOTSUP);
-
-                       break;
-
-               case ZPOOL_PROP_BOOTFS:
-                       /*
-                        * If the pool version is less than SPA_VERSION_BOOTFS,
-                        * or the pool is still being created (version == 0),
-                        * the bootfs property cannot be set.
-                        */
-                       if (spa_version(spa) < SPA_VERSION_BOOTFS) {
-                               error = SET_ERROR(ENOTSUP);
-                               break;
-                       }
-
-                       /*
-                        * Make sure the vdev config is bootable
-                        */
-                       if (!vdev_is_bootable(spa->spa_root_vdev)) {
-                               error = SET_ERROR(ENOTSUP);
-                               break;
-                       }
-
-                       reset_bootfs = 1;
-
-                       error = nvpair_value_string(elem, &strval);
-
-                       if (!error) {
-                               objset_t *os;
-                               uint64_t propval;
-
-                               if (strval == NULL || strval[0] == '\0') {
-                                       objnum = zpool_prop_default_numeric(
-                                           ZPOOL_PROP_BOOTFS);
-                                       break;
-                               }
-
-                               error = dmu_objset_hold(strval, FTAG, &os);
-                               if (error)
-                                       break;
-
-                               /*
-                                * Must be ZPL, and its property settings
-                                * must be supported by GRUB (compression
-                                * is not gzip, and large blocks or large
-                                * dnodes are not used).
-                                */
-
-                               if (dmu_objset_type(os) != DMU_OST_ZFS) {
-                                       error = SET_ERROR(ENOTSUP);
-                               } else if ((error =
-                                   dsl_prop_get_int_ds(dmu_objset_ds(os),
-                                   zfs_prop_to_name(ZFS_PROP_COMPRESSION),
-                                   &propval)) == 0 &&
-                                   !BOOTFS_COMPRESS_VALID(propval)) {
-                                       error = SET_ERROR(ENOTSUP);
-                               } else if ((error =
-                                   dsl_prop_get_int_ds(dmu_objset_ds(os),
-                                   zfs_prop_to_name(ZFS_PROP_DNODESIZE),
-                                   &propval)) == 0 &&
-                                   propval != ZFS_DNSIZE_LEGACY) {
-                                       error = SET_ERROR(ENOTSUP);
-                               } else {
-                                       objnum = dmu_objset_id(os);
-                               }
-                               dmu_objset_rele(os, FTAG);
-                       }
-                       break;
-
-               case ZPOOL_PROP_FAILUREMODE:
-                       error = nvpair_value_uint64(elem, &intval);
-                       if (!error && intval > ZIO_FAILURE_MODE_PANIC)
-                               error = SET_ERROR(EINVAL);
-
-                       /*
-                        * This is a special case which only occurs when
-                        * the pool has completely failed. This allows
-                        * the user to change the in-core failmode property
-                        * without syncing it out to disk (I/Os might
-                        * currently be blocked). We do this by returning
-                        * EIO to the caller (spa_prop_set) to trick it
-                        * into thinking we encountered a property validation
-                        * error.
-                        */
-                       if (!error && spa_suspended(spa)) {
-                               spa->spa_failmode = intval;
-                               error = SET_ERROR(EIO);
-                       }
-                       break;
-
-               case ZPOOL_PROP_CACHEFILE:
-                       if ((error = nvpair_value_string(elem, &strval)) != 0)
-                               break;
-
-                       if (strval[0] == '\0')
-                               break;
-
-                       if (strcmp(strval, "none") == 0)
-                               break;
-
-                       if (strval[0] != '/') {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-
-                       slash = strrchr(strval, '/');
-                       ASSERT(slash != NULL);
-
-                       if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
-                           strcmp(slash, "/..") == 0)
-                               error = SET_ERROR(EINVAL);
-                       break;
-
-               case ZPOOL_PROP_COMMENT:
-                       if ((error = nvpair_value_string(elem, &strval)) != 0)
-                               break;
-                       for (check = strval; *check != '\0'; check++) {
-                               if (!isprint(*check)) {
-                                       error = SET_ERROR(EINVAL);
-                                       break;
-                               }
-                       }
-                       if (strlen(strval) > ZPROP_MAX_COMMENT)
-                               error = SET_ERROR(E2BIG);
-                       break;
-
-               case ZPOOL_PROP_DEDUPDITTO:
-                       if (spa_version(spa) < SPA_VERSION_DEDUP)
-                               error = SET_ERROR(ENOTSUP);
-                       else
-                               error = nvpair_value_uint64(elem, &intval);
-                       if (error == 0 &&
-                           intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
-                               error = SET_ERROR(EINVAL);
-                       break;
-
-               default:
-                       break;
-               }
-
-               if (error)
-                       break;
-       }
-
-       if (!error && reset_bootfs) {
-               error = nvlist_remove(props,
-                   zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
-
-               if (!error) {
-                       error = nvlist_add_uint64(props,
-                           zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
-               }
-       }
-
-       return (error);
-}
-
-void
-spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
-{
-       char *cachefile;
-       spa_config_dirent_t *dp;
-
-       if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
-           &cachefile) != 0)
-               return;
-
-       dp = kmem_alloc(sizeof (spa_config_dirent_t),
-           KM_SLEEP);
-
-       if (cachefile[0] == '\0')
-               dp->scd_path = spa_strdup(spa_config_path);
-       else if (strcmp(cachefile, "none") == 0)
-               dp->scd_path = NULL;
-       else
-               dp->scd_path = spa_strdup(cachefile);
-
-       list_insert_head(&spa->spa_config_list, dp);
-       if (need_sync)
-               spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
-}
-
-int
-spa_prop_set(spa_t *spa, nvlist_t *nvp)
-{
-       int error;
-       nvpair_t *elem = NULL;
-       boolean_t need_sync = B_FALSE;
-
-       if ((error = spa_prop_validate(spa, nvp)) != 0)
-               return (error);
-
-       while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
-               zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
-
-               if (prop == ZPOOL_PROP_CACHEFILE ||
-                   prop == ZPOOL_PROP_ALTROOT ||
-                   prop == ZPOOL_PROP_READONLY)
-                       continue;
-
-               if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
-                       uint64_t ver;
-
-                       if (prop == ZPOOL_PROP_VERSION) {
-                               VERIFY(nvpair_value_uint64(elem, &ver) == 0);
-                       } else {
-                               ASSERT(zpool_prop_feature(nvpair_name(elem)));
-                               ver = SPA_VERSION_FEATURES;
-                               need_sync = B_TRUE;
-                       }
-
-                       /* Save time if the version is already set. */
-                       if (ver == spa_version(spa))
-                               continue;
-
-                       /*
-                        * In addition to the pool directory object, we might
-                        * create the pool properties object, the features for
-                        * read object, the features for write object, or the
-                        * feature descriptions object.
-                        */
-                       error = dsl_sync_task(spa->spa_name, NULL,
-                           spa_sync_version, &ver,
-                           6, ZFS_SPACE_CHECK_RESERVED);
-                       if (error)
-                               return (error);
-                       continue;
-               }
-
-               need_sync = B_TRUE;
-               break;
-       }
-
-       if (need_sync) {
-               return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
-                   nvp, 6, ZFS_SPACE_CHECK_RESERVED));
-       }
-
-       return (0);
-}
-
-/*
- * If the bootfs property value is dsobj, clear it.
- */
-void
-spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
-{
-       if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
-               VERIFY(zap_remove(spa->spa_meta_objset,
-                   spa->spa_pool_props_object,
-                   zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
-               spa->spa_bootfs = 0;
-       }
-}
-
-/*ARGSUSED*/
-static int
-spa_change_guid_check(void *arg, dmu_tx_t *tx)
-{
-       spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-       vdev_t *rvd = spa->spa_root_vdev;
-       uint64_t vdev_state;
-       ASSERTV(uint64_t *newguid = arg);
-
-       spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
-       vdev_state = rvd->vdev_state;
-       spa_config_exit(spa, SCL_STATE, FTAG);
-
-       if (vdev_state != VDEV_STATE_HEALTHY)
-               return (SET_ERROR(ENXIO));
-
-       ASSERT3U(spa_guid(spa), !=, *newguid);
-
-       return (0);
-}
-
-static void
-spa_change_guid_sync(void *arg, dmu_tx_t *tx)
-{
-       uint64_t *newguid = arg;
-       spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-       uint64_t oldguid;
-       vdev_t *rvd = spa->spa_root_vdev;
-
-       oldguid = spa_guid(spa);
-
-       spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
-       rvd->vdev_guid = *newguid;
-       rvd->vdev_guid_sum += (*newguid - oldguid);
-       vdev_config_dirty(rvd);
-       spa_config_exit(spa, SCL_STATE, FTAG);
-
-       spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
-           oldguid, *newguid);
-}
-
-/*
- * Change the GUID for the pool.  This is done so that we can later
- * re-import a pool built from a clone of our own vdevs.  We will modify
- * the root vdev's guid, our own pool guid, and then mark all of our
- * vdevs dirty.  Note that we must make sure that all our vdevs are
- * online when we do this, or else any vdevs that weren't present
- * would be orphaned from our pool.  We are also going to issue a
- * sysevent to update any watchers.
- */
-int
-spa_change_guid(spa_t *spa)
-{
-       int error;
-       uint64_t guid;
-
-       mutex_enter(&spa->spa_vdev_top_lock);
-       mutex_enter(&spa_namespace_lock);
-       guid = spa_generate_guid(NULL);
-
-       error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
-           spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
-
-       if (error == 0) {
-               spa_config_sync(spa, B_FALSE, B_TRUE);
-               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
-       }
-
-       mutex_exit(&spa_namespace_lock);
-       mutex_exit(&spa->spa_vdev_top_lock);
-
-       return (error);
-}
-
-/*
- * ==========================================================================
- * SPA state manipulation (open/create/destroy/import/export)
- * ==========================================================================
- */
-
-static int
-spa_error_entry_compare(const void *a, const void *b)
-{
-       const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
-       const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
-       int ret;
-
-       ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
-           sizeof (zbookmark_phys_t));
-
-       return (AVL_ISIGN(ret));
-}
-
-/*
- * Utility function which retrieves copies of the current logs and
- * re-initializes them in the process.
- */
-void
-spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
-{
-       ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
-
-       bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
-       bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
-
-       avl_create(&spa->spa_errlist_scrub,
-           spa_error_entry_compare, sizeof (spa_error_entry_t),
-           offsetof(spa_error_entry_t, se_avl));
-       avl_create(&spa->spa_errlist_last,
-           spa_error_entry_compare, sizeof (spa_error_entry_t),
-           offsetof(spa_error_entry_t, se_avl));
-}
-
-static void
-spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
-{
-       const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
-       enum zti_modes mode = ztip->zti_mode;
-       uint_t value = ztip->zti_value;
-       uint_t count = ztip->zti_count;
-       spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
-       char name[32];
-       uint_t i, flags = 0;
-       boolean_t batch = B_FALSE;
-
-       if (mode == ZTI_MODE_NULL) {
-               tqs->stqs_count = 0;
-               tqs->stqs_taskq = NULL;
-               return;
-       }
-
-       ASSERT3U(count, >, 0);
-
-       tqs->stqs_count = count;
-       tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
-
-       switch (mode) {
-       case ZTI_MODE_FIXED:
-               ASSERT3U(value, >=, 1);
-               value = MAX(value, 1);
-               flags |= TASKQ_DYNAMIC;
-               break;
-
-       case ZTI_MODE_BATCH:
-               batch = B_TRUE;
-               flags |= TASKQ_THREADS_CPU_PCT;
-               value = MIN(zio_taskq_batch_pct, 100);
-               break;
-
-       default:
-               panic("unrecognized mode for %s_%s taskq (%u:%u) in "
-                   "spa_activate()",
-                   zio_type_name[t], zio_taskq_types[q], mode, value);
-               break;
-       }
-
-       for (i = 0; i < count; i++) {
-               taskq_t *tq;
-
-               if (count > 1) {
-                       (void) snprintf(name, sizeof (name), "%s_%s_%u",
-                           zio_type_name[t], zio_taskq_types[q], i);
-               } else {
-                       (void) snprintf(name, sizeof (name), "%s_%s",
-                           zio_type_name[t], zio_taskq_types[q]);
-               }
-
-               if (zio_taskq_sysdc && spa->spa_proc != &p0) {
-                       if (batch)
-                               flags |= TASKQ_DC_BATCH;
-
-                       tq = taskq_create_sysdc(name, value, 50, INT_MAX,
-                           spa->spa_proc, zio_taskq_basedc, flags);
-               } else {
-                       pri_t pri = maxclsyspri;
-                       /*
-                        * The write issue taskq can be extremely CPU
-                        * intensive.  Run it at slightly less important
-                        * priority than the other taskqs.  Under Linux this
-                        * means incrementing the priority value on platforms
-                        * like illumos it should be decremented.
-                        */
-                       if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
-                               pri++;
-
-                       tq = taskq_create_proc(name, value, pri, 50,
-                           INT_MAX, spa->spa_proc, flags);
-               }
-
-               tqs->stqs_taskq[i] = tq;
-       }
-}
-
-static void
-spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
-{
-       spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
-       uint_t i;
-
-       if (tqs->stqs_taskq == NULL) {
-               ASSERT3U(tqs->stqs_count, ==, 0);
-               return;
-       }
-
-       for (i = 0; i < tqs->stqs_count; i++) {
-               ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
-               taskq_destroy(tqs->stqs_taskq[i]);
-       }
-
-       kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
-       tqs->stqs_taskq = NULL;
-}
-
-/*
- * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
- * Note that a type may have multiple discrete taskqs to avoid lock contention
- * on the taskq itself. In that case we choose which taskq at random by using
- * the low bits of gethrtime().
- */
-void
-spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
-    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
-{
-       spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
-       taskq_t *tq;
-
-       ASSERT3P(tqs->stqs_taskq, !=, NULL);
-       ASSERT3U(tqs->stqs_count, !=, 0);
-
-       if (tqs->stqs_count == 1) {
-               tq = tqs->stqs_taskq[0];
-       } else {
-               tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
-       }
-
-       taskq_dispatch_ent(tq, func, arg, flags, ent);
-}
-
-/*
- * Same as spa_taskq_dispatch_ent() but block on the task until completion.
- */
-void
-spa_taskq_dispatch_sync(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
-    task_func_t *func, void *arg, uint_t flags)
-{
-       spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
-       taskq_t *tq;
-       taskqid_t id;
-
-       ASSERT3P(tqs->stqs_taskq, !=, NULL);
-       ASSERT3U(tqs->stqs_count, !=, 0);
-
-       if (tqs->stqs_count == 1) {
-               tq = tqs->stqs_taskq[0];
-       } else {
-               tq = tqs->stqs_taskq[((uint64_t)gethrtime()) % tqs->stqs_count];
-       }
-
-       id = taskq_dispatch(tq, func, arg, flags);
-       if (id)
-               taskq_wait_id(tq, id);
-}
-
-static void
-spa_create_zio_taskqs(spa_t *spa)
-{
-       int t, q;
-
-       for (t = 0; t < ZIO_TYPES; t++) {
-               for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
-                       spa_taskqs_init(spa, t, q);
-               }
-       }
-}
-
-#if defined(_KERNEL) && defined(HAVE_SPA_THREAD)
-static void
-spa_thread(void *arg)
-{
-       callb_cpr_t cprinfo;
-
-       spa_t *spa = arg;
-       user_t *pu = PTOU(curproc);
-
-       CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
-           spa->spa_name);
-
-       ASSERT(curproc != &p0);
-       (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
-           "zpool-%s", spa->spa_name);
-       (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
-
-       /* bind this thread to the requested psrset */
-       if (zio_taskq_psrset_bind != PS_NONE) {
-               pool_lock();
-               mutex_enter(&cpu_lock);
-               mutex_enter(&pidlock);
-               mutex_enter(&curproc->p_lock);
-
-               if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
-                   0, NULL, NULL) == 0)  {
-                       curthread->t_bind_pset = zio_taskq_psrset_bind;
-               } else {
-                       cmn_err(CE_WARN,
-                           "Couldn't bind process for zfs pool \"%s\" to "
-                           "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
-               }
-
-               mutex_exit(&curproc->p_lock);
-               mutex_exit(&pidlock);
-               mutex_exit(&cpu_lock);
-               pool_unlock();
-       }
-
-       if (zio_taskq_sysdc) {
-               sysdc_thread_enter(curthread, 100, 0);
-       }
-
-       spa->spa_proc = curproc;
-       spa->spa_did = curthread->t_did;
-
-       spa_create_zio_taskqs(spa);
-
-       mutex_enter(&spa->spa_proc_lock);
-       ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
-
-       spa->spa_proc_state = SPA_PROC_ACTIVE;
-       cv_broadcast(&spa->spa_proc_cv);
-
-       CALLB_CPR_SAFE_BEGIN(&cprinfo);
-       while (spa->spa_proc_state == SPA_PROC_ACTIVE)
-               cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
-       CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
-
-       ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
-       spa->spa_proc_state = SPA_PROC_GONE;
-       spa->spa_proc = &p0;
-       cv_broadcast(&spa->spa_proc_cv);
-       CALLB_CPR_EXIT(&cprinfo);       /* drops spa_proc_lock */
-
-       mutex_enter(&curproc->p_lock);
-       lwp_exit();
-}
-#endif
-
-/*
- * Activate an uninitialized pool.
- */
-static void
-spa_activate(spa_t *spa, int mode)
-{
-       ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
-
-       spa->spa_state = POOL_STATE_ACTIVE;
-       spa->spa_mode = mode;
-
-       spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
-       spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
-
-       /* Try to create a covering process */
-       mutex_enter(&spa->spa_proc_lock);
-       ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
-       ASSERT(spa->spa_proc == &p0);
-       spa->spa_did = 0;
-
-#ifdef HAVE_SPA_THREAD
-       /* Only create a process if we're going to be around a while. */
-       if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
-               if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
-                   NULL, 0) == 0) {
-                       spa->spa_proc_state = SPA_PROC_CREATED;
-                       while (spa->spa_proc_state == SPA_PROC_CREATED) {
-                               cv_wait(&spa->spa_proc_cv,
-                                   &spa->spa_proc_lock);
-                       }
-                       ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
-                       ASSERT(spa->spa_proc != &p0);
-                       ASSERT(spa->spa_did != 0);
-               } else {
-#ifdef _KERNEL
-                       cmn_err(CE_WARN,
-                           "Couldn't create process for zfs pool \"%s\"\n",
-                           spa->spa_name);
-#endif
-               }
-       }
-#endif /* HAVE_SPA_THREAD */
-       mutex_exit(&spa->spa_proc_lock);
-
-       /* If we didn't create a process, we need to create our taskqs. */
-       if (spa->spa_proc == &p0) {
-               spa_create_zio_taskqs(spa);
-       }
-
-       list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
-           offsetof(vdev_t, vdev_config_dirty_node));
-       list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
-           offsetof(objset_t, os_evicting_node));
-       list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
-           offsetof(vdev_t, vdev_state_dirty_node));
-
-       txg_list_create(&spa->spa_vdev_txg_list, spa,
-           offsetof(struct vdev, vdev_txg_node));
-
-       avl_create(&spa->spa_errlist_scrub,
-           spa_error_entry_compare, sizeof (spa_error_entry_t),
-           offsetof(spa_error_entry_t, se_avl));
-       avl_create(&spa->spa_errlist_last,
-           spa_error_entry_compare, sizeof (spa_error_entry_t),
-           offsetof(spa_error_entry_t, se_avl));
-
-       /*
-        * This taskq is used to perform zvol-minor-related tasks
-        * asynchronously. This has several advantages, including easy
-        * resolution of various deadlocks (zfsonlinux bug #3681).
-        *
-        * The taskq must be single threaded to ensure tasks are always
-        * processed in the order in which they were dispatched.
-        *
-        * A taskq per pool allows one to keep the pools independent.
-        * This way if one pool is suspended, it will not impact another.
-        *
-        * The preferred location to dispatch a zvol minor task is a sync
-        * task. In this context, there is easy access to the spa_t and minimal
-        * error handling is required because the sync task must succeed.
-        */
-       spa->spa_zvol_taskq = taskq_create("z_zvol", 1, defclsyspri,
-           1, INT_MAX, 0);
-
-       /*
-        * Taskq dedicated to prefetcher threads: this is used to prevent the
-        * pool traverse code from monopolizing the global (and limited)
-        * system_taskq by inappropriately scheduling long running tasks on it.
-        */
-       spa->spa_prefetch_taskq = taskq_create("z_prefetch", boot_ncpus,
-           defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
-
-       /*
-        * The taskq to upgrade datasets in this pool. Currently used by
-        * feature SPA_FEATURE_USEROBJ_ACCOUNTING.
-        */
-       spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus,
-           defclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
-}
-
-/*
- * Opposite of spa_activate().
- */
-static void
-spa_deactivate(spa_t *spa)
-{
-       int t, q;
-
-       ASSERT(spa->spa_sync_on == B_FALSE);
-       ASSERT(spa->spa_dsl_pool == NULL);
-       ASSERT(spa->spa_root_vdev == NULL);
-       ASSERT(spa->spa_async_zio_root == NULL);
-       ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
-
-       spa_evicting_os_wait(spa);
-
-       if (spa->spa_zvol_taskq) {
-               taskq_destroy(spa->spa_zvol_taskq);
-               spa->spa_zvol_taskq = NULL;
-       }
-
-       if (spa->spa_prefetch_taskq) {
-               taskq_destroy(spa->spa_prefetch_taskq);
-               spa->spa_prefetch_taskq = NULL;
-       }
-
-       if (spa->spa_upgrade_taskq) {
-               taskq_destroy(spa->spa_upgrade_taskq);
-               spa->spa_upgrade_taskq = NULL;
-       }
-
-       txg_list_destroy(&spa->spa_vdev_txg_list);
-
-       list_destroy(&spa->spa_config_dirty_list);
-       list_destroy(&spa->spa_evicting_os_list);
-       list_destroy(&spa->spa_state_dirty_list);
-
-       taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
-
-       for (t = 0; t < ZIO_TYPES; t++) {
-               for (q = 0; q < ZIO_TASKQ_TYPES; q++) {
-                       spa_taskqs_fini(spa, t, q);
-               }
-       }
-
-       metaslab_class_destroy(spa->spa_normal_class);
-       spa->spa_normal_class = NULL;
-
-       metaslab_class_destroy(spa->spa_log_class);
-       spa->spa_log_class = NULL;
-
-       /*
-        * If this was part of an import or the open otherwise failed, we may
-        * still have errors left in the queues.  Empty them just in case.
-        */
-       spa_errlog_drain(spa);
-
-       avl_destroy(&spa->spa_errlist_scrub);
-       avl_destroy(&spa->spa_errlist_last);
-
-       spa->spa_state = POOL_STATE_UNINITIALIZED;
-
-       mutex_enter(&spa->spa_proc_lock);
-       if (spa->spa_proc_state != SPA_PROC_NONE) {
-               ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
-               spa->spa_proc_state = SPA_PROC_DEACTIVATE;
-               cv_broadcast(&spa->spa_proc_cv);
-               while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
-                       ASSERT(spa->spa_proc != &p0);
-                       cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
-               }
-               ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
-               spa->spa_proc_state = SPA_PROC_NONE;
-       }
-       ASSERT(spa->spa_proc == &p0);
-       mutex_exit(&spa->spa_proc_lock);
-
-       /*
-        * We want to make sure spa_thread() has actually exited the ZFS
-        * module, so that the module can't be unloaded out from underneath
-        * it.
-        */
-       if (spa->spa_did != 0) {
-               thread_join(spa->spa_did);
-               spa->spa_did = 0;
-       }
-}
-
-/*
- * Verify a pool configuration, and construct the vdev tree appropriately.  This
- * will create all the necessary vdevs in the appropriate layout, with each vdev
- * in the CLOSED state.  This will prep the pool before open/creation/import.
- * All vdev validation is done by the vdev_alloc() routine.
- */
-static int
-spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
-    uint_t id, int atype)
-{
-       nvlist_t **child;
-       uint_t children;
-       int error;
-       int c;
-
-       if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
-               return (error);
-
-       if ((*vdp)->vdev_ops->vdev_op_leaf)
-               return (0);
-
-       error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
-           &child, &children);
-
-       if (error == ENOENT)
-               return (0);
-
-       if (error) {
-               vdev_free(*vdp);
-               *vdp = NULL;
-               return (SET_ERROR(EINVAL));
-       }
-
-       for (c = 0; c < children; c++) {
-               vdev_t *vd;
-               if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
-                   atype)) != 0) {
-                       vdev_free(*vdp);
-                       *vdp = NULL;
-                       return (error);
-               }
-       }
-
-       ASSERT(*vdp != NULL);
-
-       return (0);
-}
-
-/*
- * Opposite of spa_load().
- */
-static void
-spa_unload(spa_t *spa)
-{
-       int i, c;
-
-       ASSERT(MUTEX_HELD(&spa_namespace_lock));
-
-       /*
-        * Stop async tasks.
-        */
-       spa_async_suspend(spa);
-
-       /*
-        * Stop syncing.
-        */
-       if (spa->spa_sync_on) {
-               txg_sync_stop(spa->spa_dsl_pool);
-               spa->spa_sync_on = B_FALSE;
-       }
-
-       /*
-        * Even though vdev_free() also calls vdev_metaslab_fini, we need
-        * to call it earlier, before we wait for async i/o to complete.
-        * This ensures that there is no async metaslab prefetching, by
-        * calling taskq_wait(mg_taskq).
-        */
-       if (spa->spa_root_vdev != NULL) {
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               for (c = 0; c < spa->spa_root_vdev->vdev_children; c++)
-                       vdev_metaslab_fini(spa->spa_root_vdev->vdev_child[c]);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-       }
-
-       if (spa->spa_mmp.mmp_thread)
-               mmp_thread_stop(spa);
-
-       /*
-        * Wait for any outstanding async I/O to complete.
-        */
-       if (spa->spa_async_zio_root != NULL) {
-               for (i = 0; i < max_ncpus; i++)
-                       (void) zio_wait(spa->spa_async_zio_root[i]);
-               kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
-               spa->spa_async_zio_root = NULL;
-       }
-
-       bpobj_close(&spa->spa_deferred_bpobj);
-
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-
-       /*
-        * Close all vdevs.
-        */
-       if (spa->spa_root_vdev)
-               vdev_free(spa->spa_root_vdev);
-       ASSERT(spa->spa_root_vdev == NULL);
-
-       /*
-        * Close the dsl pool.
-        */
-       if (spa->spa_dsl_pool) {
-               dsl_pool_close(spa->spa_dsl_pool);
-               spa->spa_dsl_pool = NULL;
-               spa->spa_meta_objset = NULL;
-       }
-
-       ddt_unload(spa);
-
-       /*
-        * Drop and purge level 2 cache
-        */
-       spa_l2cache_drop(spa);
-
-       for (i = 0; i < spa->spa_spares.sav_count; i++)
-               vdev_free(spa->spa_spares.sav_vdevs[i]);
-       if (spa->spa_spares.sav_vdevs) {
-               kmem_free(spa->spa_spares.sav_vdevs,
-                   spa->spa_spares.sav_count * sizeof (void *));
-               spa->spa_spares.sav_vdevs = NULL;
-       }
-       if (spa->spa_spares.sav_config) {
-               nvlist_free(spa->spa_spares.sav_config);
-               spa->spa_spares.sav_config = NULL;
-       }
-       spa->spa_spares.sav_count = 0;
-
-       for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
-               vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
-               vdev_free(spa->spa_l2cache.sav_vdevs[i]);
-       }
-       if (spa->spa_l2cache.sav_vdevs) {
-               kmem_free(spa->spa_l2cache.sav_vdevs,
-                   spa->spa_l2cache.sav_count * sizeof (void *));
-               spa->spa_l2cache.sav_vdevs = NULL;
-       }
-       if (spa->spa_l2cache.sav_config) {
-               nvlist_free(spa->spa_l2cache.sav_config);
-               spa->spa_l2cache.sav_config = NULL;
-       }
-       spa->spa_l2cache.sav_count = 0;
-
-       spa->spa_async_suspended = 0;
-
-       if (spa->spa_comment != NULL) {
-               spa_strfree(spa->spa_comment);
-               spa->spa_comment = NULL;
-       }
-
-       spa_config_exit(spa, SCL_ALL, FTAG);
-}
-
-/*
- * Load (or re-load) the current list of vdevs describing the active spares for
- * this pool.  When this is called, we have some form of basic information in
- * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
- * then re-generate a more complete list including status information.
- */
-static void
-spa_load_spares(spa_t *spa)
-{
-       nvlist_t **spares;
-       uint_t nspares;
-       int i;
-       vdev_t *vd, *tvd;
-
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
-
-       /*
-        * First, close and free any existing spare vdevs.
-        */
-       for (i = 0; i < spa->spa_spares.sav_count; i++) {
-               vd = spa->spa_spares.sav_vdevs[i];
-
-               /* Undo the call to spa_activate() below */
-               if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
-                   B_FALSE)) != NULL && tvd->vdev_isspare)
-                       spa_spare_remove(tvd);
-               vdev_close(vd);
-               vdev_free(vd);
-       }
-
-       if (spa->spa_spares.sav_vdevs)
-               kmem_free(spa->spa_spares.sav_vdevs,
-                   spa->spa_spares.sav_count * sizeof (void *));
-
-       if (spa->spa_spares.sav_config == NULL)
-               nspares = 0;
-       else
-               VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
-                   ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
-
-       spa->spa_spares.sav_count = (int)nspares;
-       spa->spa_spares.sav_vdevs = NULL;
-
-       if (nspares == 0)
-               return;
-
-       /*
-        * Construct the array of vdevs, opening them to get status in the
-        * process.   For each spare, there is potentially two different vdev_t
-        * structures associated with it: one in the list of spares (used only
-        * for basic validation purposes) and one in the active vdev
-        * configuration (if it's spared in).  During this phase we open and
-        * validate each vdev on the spare list.  If the vdev also exists in the
-        * active configuration, then we also mark this vdev as an active spare.
-        */
-       spa->spa_spares.sav_vdevs = kmem_zalloc(nspares * sizeof (void *),
-           KM_SLEEP);
-       for (i = 0; i < spa->spa_spares.sav_count; i++) {
-               VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
-                   VDEV_ALLOC_SPARE) == 0);
-               ASSERT(vd != NULL);
-
-               spa->spa_spares.sav_vdevs[i] = vd;
-
-               if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
-                   B_FALSE)) != NULL) {
-                       if (!tvd->vdev_isspare)
-                               spa_spare_add(tvd);
-
-                       /*
-                        * We only mark the spare active if we were successfully
-                        * able to load the vdev.  Otherwise, importing a pool
-                        * with a bad active spare would result in strange
-                        * behavior, because multiple pool would think the spare
-                        * is actively in use.
-                        *
-                        * There is a vulnerability here to an equally bizarre
-                        * circumstance, where a dead active spare is later
-                        * brought back to life (onlined or otherwise).  Given
-                        * the rarity of this scenario, and the extra complexity
-                        * it adds, we ignore the possibility.
-                        */
-                       if (!vdev_is_dead(tvd))
-                               spa_spare_activate(tvd);
-               }
-
-               vd->vdev_top = vd;
-               vd->vdev_aux = &spa->spa_spares;
-
-               if (vdev_open(vd) != 0)
-                       continue;
-
-               if (vdev_validate_aux(vd) == 0)
-                       spa_spare_add(vd);
-       }
-
-       /*
-        * Recompute the stashed list of spares, with status information
-        * this time.
-        */
-       VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
-           DATA_TYPE_NVLIST_ARRAY) == 0);
-
-       spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
-           KM_SLEEP);
-       for (i = 0; i < spa->spa_spares.sav_count; i++)
-               spares[i] = vdev_config_generate(spa,
-                   spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
-       VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
-           ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
-       for (i = 0; i < spa->spa_spares.sav_count; i++)
-               nvlist_free(spares[i]);
-       kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
-}
-
-/*
- * Load (or re-load) the current list of vdevs describing the active l2cache for
- * this pool.  When this is called, we have some form of basic information in
- * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
- * then re-generate a more complete list including status information.
- * Devices which are already active have their details maintained, and are
- * not re-opened.
- */
-static void
-spa_load_l2cache(spa_t *spa)
-{
-       nvlist_t **l2cache = NULL;
-       uint_t nl2cache;
-       int i, j, oldnvdevs;
-       uint64_t guid;
-       vdev_t *vd, **oldvdevs, **newvdevs;
-       spa_aux_vdev_t *sav = &spa->spa_l2cache;
-
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
-
-       oldvdevs = sav->sav_vdevs;
-       oldnvdevs = sav->sav_count;
-       sav->sav_vdevs = NULL;
-       sav->sav_count = 0;
-
-       if (sav->sav_config == NULL) {
-               nl2cache = 0;
-               newvdevs = NULL;
-               goto out;
-       }
-
-       VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
-           ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
-       newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
-
-       /*
-        * Process new nvlist of vdevs.
-        */
-       for (i = 0; i < nl2cache; i++) {
-               VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
-                   &guid) == 0);
-
-               newvdevs[i] = NULL;
-               for (j = 0; j < oldnvdevs; j++) {
-                       vd = oldvdevs[j];
-                       if (vd != NULL && guid == vd->vdev_guid) {
-                               /*
-                                * Retain previous vdev for add/remove ops.
-                                */
-                               newvdevs[i] = vd;
-                               oldvdevs[j] = NULL;
-                               break;
-                       }
-               }
-
-               if (newvdevs[i] == NULL) {
-                       /*
-                        * Create new vdev
-                        */
-                       VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
-                           VDEV_ALLOC_L2CACHE) == 0);
-                       ASSERT(vd != NULL);
-                       newvdevs[i] = vd;
-
-                       /*
-                        * Commit this vdev as an l2cache device,
-                        * even if it fails to open.
-                        */
-                       spa_l2cache_add(vd);
-
-                       vd->vdev_top = vd;
-                       vd->vdev_aux = sav;
-
-                       spa_l2cache_activate(vd);
-
-                       if (vdev_open(vd) != 0)
-                               continue;
-
-                       (void) vdev_validate_aux(vd);
-
-                       if (!vdev_is_dead(vd))
-                               l2arc_add_vdev(spa, vd);
-               }
-       }
-
-       sav->sav_vdevs = newvdevs;
-       sav->sav_count = (int)nl2cache;
-
-       /*
-        * Recompute the stashed list of l2cache devices, with status
-        * information this time.
-        */
-       VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
-           DATA_TYPE_NVLIST_ARRAY) == 0);
-
-       if (sav->sav_count > 0)
-               l2cache = kmem_alloc(sav->sav_count * sizeof (void *),
-                   KM_SLEEP);
-       for (i = 0; i < sav->sav_count; i++)
-               l2cache[i] = vdev_config_generate(spa,
-                   sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
-       VERIFY(nvlist_add_nvlist_array(sav->sav_config,
-           ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
-
-out:
-       /*
-        * Purge vdevs that were dropped
-        */
-       for (i = 0; i < oldnvdevs; i++) {
-               uint64_t pool;
-
-               vd = oldvdevs[i];
-               if (vd != NULL) {
-                       ASSERT(vd->vdev_isl2cache);
-
-                       if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
-                           pool != 0ULL && l2arc_vdev_present(vd))
-                               l2arc_remove_vdev(vd);
-                       vdev_clear_stats(vd);
-                       vdev_free(vd);
-               }
-       }
-
-       if (oldvdevs)
-               kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
-
-       for (i = 0; i < sav->sav_count; i++)
-               nvlist_free(l2cache[i]);
-       if (sav->sav_count)
-               kmem_free(l2cache, sav->sav_count * sizeof (void *));
-}
-
-static int
-load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
-{
-       dmu_buf_t *db;
-       char *packed = NULL;
-       size_t nvsize = 0;
-       int error;
-       *value = NULL;
-
-       error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
-       if (error)
-               return (error);
-
-       nvsize = *(uint64_t *)db->db_data;
-       dmu_buf_rele(db, FTAG);
-
-       packed = vmem_alloc(nvsize, KM_SLEEP);
-       error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
-           DMU_READ_PREFETCH);
-       if (error == 0)
-               error = nvlist_unpack(packed, nvsize, value, 0);
-       vmem_free(packed, nvsize);
-
-       return (error);
-}
-
-/*
- * Checks to see if the given vdev could not be opened, in which case we post a
- * sysevent to notify the autoreplace code that the device has been removed.
- */
-static void
-spa_check_removed(vdev_t *vd)
-{
-       int c;
-
-       for (c = 0; c < vd->vdev_children; c++)
-               spa_check_removed(vd->vdev_child[c]);
-
-       if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
-           !vd->vdev_ishole) {
-               zfs_post_autoreplace(vd->vdev_spa, vd);
-               spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
-       }
-}
-
-static void
-spa_config_valid_zaps(vdev_t *vd, vdev_t *mvd)
-{
-       uint64_t i;
-
-       ASSERT3U(vd->vdev_children, ==, mvd->vdev_children);
-
-       vd->vdev_top_zap = mvd->vdev_top_zap;
-       vd->vdev_leaf_zap = mvd->vdev_leaf_zap;
-
-       for (i = 0; i < vd->vdev_children; i++) {
-               spa_config_valid_zaps(vd->vdev_child[i], mvd->vdev_child[i]);
-       }
-}
-
-/*
- * Validate the current config against the MOS config
- */
-static boolean_t
-spa_config_valid(spa_t *spa, nvlist_t *config)
-{
-       vdev_t *mrvd, *rvd = spa->spa_root_vdev;
-       nvlist_t *nv;
-       int c, i;
-
-       VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
-
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-       VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
-
-       ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
-
-       /*
-        * If we're doing a normal import, then build up any additional
-        * diagnostic information about missing devices in this config.
-        * We'll pass this up to the user for further processing.
-        */
-       if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
-               nvlist_t **child, *nv;
-               uint64_t idx = 0;
-
-               child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
-                   KM_SLEEP);
-               VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
-
-               for (c = 0; c < rvd->vdev_children; c++) {
-                       vdev_t *tvd = rvd->vdev_child[c];
-                       vdev_t *mtvd  = mrvd->vdev_child[c];
-
-                       if (tvd->vdev_ops == &vdev_missing_ops &&
-                           mtvd->vdev_ops != &vdev_missing_ops &&
-                           mtvd->vdev_islog)
-                               child[idx++] = vdev_config_generate(spa, mtvd,
-                                   B_FALSE, 0);
-               }
-
-               if (idx) {
-                       VERIFY(nvlist_add_nvlist_array(nv,
-                           ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
-                       VERIFY(nvlist_add_nvlist(spa->spa_load_info,
-                           ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
-
-                       for (i = 0; i < idx; i++)
-                               nvlist_free(child[i]);
-               }
-               nvlist_free(nv);
-               kmem_free(child, rvd->vdev_children * sizeof (char **));
-       }
-
-       /*
-        * Compare the root vdev tree with the information we have
-        * from the MOS config (mrvd). Check each top-level vdev
-        * with the corresponding MOS config top-level (mtvd).
-        */
-       for (c = 0; c < rvd->vdev_children; c++) {
-               vdev_t *tvd = rvd->vdev_child[c];
-               vdev_t *mtvd  = mrvd->vdev_child[c];
-
-               /*
-                * Resolve any "missing" vdevs in the current configuration.
-                * If we find that the MOS config has more accurate information
-                * about the top-level vdev then use that vdev instead.
-                */
-               if (tvd->vdev_ops == &vdev_missing_ops &&
-                   mtvd->vdev_ops != &vdev_missing_ops) {
-
-                       if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
-                               continue;
-
-                       /*
-                        * Device specific actions.
-                        */
-                       if (mtvd->vdev_islog) {
-                               spa_set_log_state(spa, SPA_LOG_CLEAR);
-                       } else {
-                               /*
-                                * XXX - once we have 'readonly' pool
-                                * support we should be able to handle
-                                * missing data devices by transitioning
-                                * the pool to readonly.
-                                */
-                               continue;
-                       }
-
-                       /*
-                        * Swap the missing vdev with the data we were
-                        * able to obtain from the MOS config.
-                        */
-                       vdev_remove_child(rvd, tvd);
-                       vdev_remove_child(mrvd, mtvd);
-
-                       vdev_add_child(rvd, mtvd);
-                       vdev_add_child(mrvd, tvd);
-
-                       spa_config_exit(spa, SCL_ALL, FTAG);
-                       vdev_load(mtvd);
-                       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-
-                       vdev_reopen(rvd);
-               } else {
-                       if (mtvd->vdev_islog) {
-                               /*
-                                * Load the slog device's state from the MOS
-                                * config since it's possible that the label
-                                * does not contain the most up-to-date
-                                * information.
-                                */
-                               vdev_load_log_state(tvd, mtvd);
-                               vdev_reopen(tvd);
-                       }
-
-                       /*
-                        * Per-vdev ZAP info is stored exclusively in the MOS.
-                        */
-                       spa_config_valid_zaps(tvd, mtvd);
-               }
-       }
-
-       vdev_free(mrvd);
-       spa_config_exit(spa, SCL_ALL, FTAG);
-
-       /*
-        * Ensure we were able to validate the config.
-        */
-       return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
-}
-
-/*
- * Check for missing log devices
- */
-static boolean_t
-spa_check_logs(spa_t *spa)
-{
-       boolean_t rv = B_FALSE;
-       dsl_pool_t *dp = spa_get_dsl(spa);
-
-       switch (spa->spa_log_state) {
-       default:
-               break;
-       case SPA_LOG_MISSING:
-               /* need to recheck in case slog has been restored */
-       case SPA_LOG_UNKNOWN:
-               rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
-                   zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
-               if (rv)
-                       spa_set_log_state(spa, SPA_LOG_MISSING);
-               break;
-       }
-       return (rv);
-}
-
-static boolean_t
-spa_passivate_log(spa_t *spa)
-{
-       vdev_t *rvd = spa->spa_root_vdev;
-       boolean_t slog_found = B_FALSE;
-       int c;
-
-       ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
-
-       if (!spa_has_slogs(spa))
-               return (B_FALSE);
-
-       for (c = 0; c < rvd->vdev_children; c++) {
-               vdev_t *tvd = rvd->vdev_child[c];
-               metaslab_group_t *mg = tvd->vdev_mg;
-
-               if (tvd->vdev_islog) {
-                       metaslab_group_passivate(mg);
-                       slog_found = B_TRUE;
-               }
-       }
-
-       return (slog_found);
-}
-
-static void
-spa_activate_log(spa_t *spa)
-{
-       vdev_t *rvd = spa->spa_root_vdev;
-       int c;
-
-       ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
-
-       for (c = 0; c < rvd->vdev_children; c++) {
-               vdev_t *tvd = rvd->vdev_child[c];
-               metaslab_group_t *mg = tvd->vdev_mg;
-
-               if (tvd->vdev_islog)
-                       metaslab_group_activate(mg);
-       }
-}
-
-int
-spa_offline_log(spa_t *spa)
-{
-       int error;
-
-       error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
-           NULL, DS_FIND_CHILDREN);
-       if (error == 0) {
-               /*
-                * We successfully offlined the log device, sync out the
-                * current txg so that the "stubby" block can be removed
-                * by zil_sync().
-                */
-               txg_wait_synced(spa->spa_dsl_pool, 0);
-       }
-       return (error);
-}
-
-static void
-spa_aux_check_removed(spa_aux_vdev_t *sav)
-{
-       int i;
-
-       for (i = 0; i < sav->sav_count; i++)
-               spa_check_removed(sav->sav_vdevs[i]);
-}
-
-void
-spa_claim_notify(zio_t *zio)
-{
-       spa_t *spa = zio->io_spa;
-
-       if (zio->io_error)
-               return;
-
-       mutex_enter(&spa->spa_props_lock);      /* any mutex will do */
-       if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
-               spa->spa_claim_max_txg = zio->io_bp->blk_birth;
-       mutex_exit(&spa->spa_props_lock);
-}
-
-typedef struct spa_load_error {
-       uint64_t        sle_meta_count;
-       uint64_t        sle_data_count;
-} spa_load_error_t;
-
-static void
-spa_load_verify_done(zio_t *zio)
-{
-       blkptr_t *bp = zio->io_bp;
-       spa_load_error_t *sle = zio->io_private;
-       dmu_object_type_t type = BP_GET_TYPE(bp);
-       int error = zio->io_error;
-       spa_t *spa = zio->io_spa;
-
-       abd_free(zio->io_abd);
-       if (error) {
-               if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
-                   type != DMU_OT_INTENT_LOG)
-                       atomic_inc_64(&sle->sle_meta_count);
-               else
-                       atomic_inc_64(&sle->sle_data_count);
-       }
-
-       mutex_enter(&spa->spa_scrub_lock);
-       spa->spa_scrub_inflight--;
-       cv_broadcast(&spa->spa_scrub_io_cv);
-       mutex_exit(&spa->spa_scrub_lock);
-}
-
-/*
- * Maximum number of concurrent scrub i/os to create while verifying
- * a pool while importing it.
- */
-int spa_load_verify_maxinflight = 10000;
-int spa_load_verify_metadata = B_TRUE;
-int spa_load_verify_data = B_TRUE;
-
-/*ARGSUSED*/
-static int
-spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
-    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
-{
-       zio_t *rio;
-       size_t size;
-
-       if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
-               return (0);
-       /*
-        * Note: normally this routine will not be called if
-        * spa_load_verify_metadata is not set.  However, it may be useful
-        * to manually set the flag after the traversal has begun.
-        */
-       if (!spa_load_verify_metadata)
-               return (0);
-       if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
-               return (0);
-
-       rio = arg;
-       size = BP_GET_PSIZE(bp);
-
-       mutex_enter(&spa->spa_scrub_lock);
-       while (spa->spa_scrub_inflight >= spa_load_verify_maxinflight)
-               cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
-       spa->spa_scrub_inflight++;
-       mutex_exit(&spa->spa_scrub_lock);
-
-       zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
-           spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
-           ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
-           ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
-       return (0);
-}
-
-/* ARGSUSED */
-int
-verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
-{
-       if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
-               return (SET_ERROR(ENAMETOOLONG));
-
-       return (0);
-}
-
-static int
-spa_load_verify(spa_t *spa)
-{
-       zio_t *rio;
-       spa_load_error_t sle = { 0 };
-       zpool_rewind_policy_t policy;
-       boolean_t verify_ok = B_FALSE;
-       int error = 0;
-
-       zpool_get_rewind_policy(spa->spa_config, &policy);
-
-       if (policy.zrp_request & ZPOOL_NEVER_REWIND)
-               return (0);
-
-       dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
-       error = dmu_objset_find_dp(spa->spa_dsl_pool,
-           spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
-           DS_FIND_CHILDREN);
-       dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
-       if (error != 0)
-               return (error);
-
-       rio = zio_root(spa, NULL, &sle,
-           ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
-
-       if (spa_load_verify_metadata) {
-               error = traverse_pool(spa, spa->spa_verify_min_txg,
-                   TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
-                   spa_load_verify_cb, rio);
-       }
-
-       (void) zio_wait(rio);
-
-       spa->spa_load_meta_errors = sle.sle_meta_count;
-       spa->spa_load_data_errors = sle.sle_data_count;
-
-       if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
-           sle.sle_data_count <= policy.zrp_maxdata) {
-               int64_t loss = 0;
-
-               verify_ok = B_TRUE;
-               spa->spa_load_txg = spa->spa_uberblock.ub_txg;
-               spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
-
-               loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
-               VERIFY(nvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
-               VERIFY(nvlist_add_int64(spa->spa_load_info,
-                   ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
-               VERIFY(nvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
-       } else {
-               spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
-       }
-
-       if (error) {
-               if (error != ENXIO && error != EIO)
-                       error = SET_ERROR(EIO);
-               return (error);
-       }
-
-       return (verify_ok ? 0 : EIO);
-}
-
-/*
- * Find a value in the pool props object.
- */
-static void
-spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
-{
-       (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
-           zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
-}
-
-/*
- * Find a value in the pool directory object.
- */
-static int
-spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
-{
-       return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
-           name, sizeof (uint64_t), 1, val));
-}
-
-static int
-spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
-{
-       vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
-       return (err);
-}
-
-/*
- * Fix up config after a partly-completed split.  This is done with the
- * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
- * pool have that entry in their config, but only the splitting one contains
- * a list of all the guids of the vdevs that are being split off.
- *
- * This function determines what to do with that list: either rejoin
- * all the disks to the pool, or complete the splitting process.  To attempt
- * the rejoin, each disk that is offlined is marked online again, and
- * we do a reopen() call.  If the vdev label for every disk that was
- * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
- * then we call vdev_split() on each disk, and complete the split.
- *
- * Otherwise we leave the config alone, with all the vdevs in place in
- * the original pool.
- */
-static void
-spa_try_repair(spa_t *spa, nvlist_t *config)
-{
-       uint_t extracted;
-       uint64_t *glist;
-       uint_t i, gcount;
-       nvlist_t *nvl;
-       vdev_t **vd;
-       boolean_t attempt_reopen;
-
-       if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
-               return;
-
-       /* check that the config is complete */
-       if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
-           &glist, &gcount) != 0)
-               return;
-
-       vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
-
-       /* attempt to online all the vdevs & validate */
-       attempt_reopen = B_TRUE;
-       for (i = 0; i < gcount; i++) {
-               if (glist[i] == 0)      /* vdev is hole */
-                       continue;
-
-               vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
-               if (vd[i] == NULL) {
-                       /*
-                        * Don't bother attempting to reopen the disks;
-                        * just do the split.
-                        */
-                       attempt_reopen = B_FALSE;
-               } else {
-                       /* attempt to re-online it */
-                       vd[i]->vdev_offline = B_FALSE;
-               }
-       }
-
-       if (attempt_reopen) {
-               vdev_reopen(spa->spa_root_vdev);
-
-               /* check each device to see what state it's in */
-               for (extracted = 0, i = 0; i < gcount; i++) {
-                       if (vd[i] != NULL &&
-                           vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
-                               break;
-                       ++extracted;
-               }
-       }
-
-       /*
-        * If every disk has been moved to the new pool, or if we never
-        * even attempted to look at them, then we split them off for
-        * good.
-        */
-       if (!attempt_reopen || gcount == extracted) {
-               for (i = 0; i < gcount; i++)
-                       if (vd[i] != NULL)
-                               vdev_split(vd[i]);
-               vdev_reopen(spa->spa_root_vdev);
-       }
-
-       kmem_free(vd, gcount * sizeof (vdev_t *));
-}
-
-static int
-spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
-    boolean_t mosconfig)
-{
-       nvlist_t *config = spa->spa_config;
-       char *ereport = FM_EREPORT_ZFS_POOL;
-       char *comment;
-       int error;
-       uint64_t pool_guid;
-       nvlist_t *nvl;
-
-       if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
-               return (SET_ERROR(EINVAL));
-
-       ASSERT(spa->spa_comment == NULL);
-       if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
-               spa->spa_comment = spa_strdup(comment);
-
-       /*
-        * Versioning wasn't explicitly added to the label until later, so if
-        * it's not present treat it as the initial version.
-        */
-       if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
-           &spa->spa_ubsync.ub_version) != 0)
-               spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
-
-       (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
-           &spa->spa_config_txg);
-
-       if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
-           spa_guid_exists(pool_guid, 0)) {
-               error = SET_ERROR(EEXIST);
-       } else {
-               spa->spa_config_guid = pool_guid;
-
-               if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
-                   &nvl) == 0) {
-                       VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
-                           KM_SLEEP) == 0);
-               }
-
-               nvlist_free(spa->spa_load_info);
-               spa->spa_load_info = fnvlist_alloc();
-
-               gethrestime(&spa->spa_loaded_ts);
-               error = spa_load_impl(spa, pool_guid, config, state, type,
-                   mosconfig, &ereport);
-       }
-
-       /*
-        * Don't count references from objsets that are already closed
-        * and are making their way through the eviction process.
-        */
-       spa_evicting_os_wait(spa);
-       spa->spa_minref = refcount_count(&spa->spa_refcount);
-       if (error) {
-               if (error != EEXIST) {
-                       spa->spa_loaded_ts.tv_sec = 0;
-                       spa->spa_loaded_ts.tv_nsec = 0;
-               }
-               if (error != EBADF) {
-                       zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
-               }
-       }
-       spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
-       spa->spa_ena = 0;
-
-       return (error);
-}
-
-#ifdef ZFS_DEBUG
-/*
- * Count the number of per-vdev ZAPs associated with all of the vdevs in the
- * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
- * spa's per-vdev ZAP list.
- */
-static uint64_t
-vdev_count_verify_zaps(vdev_t *vd)
-{
-       spa_t *spa = vd->vdev_spa;
-       uint64_t total = 0;
-       uint64_t i;
-
-       if (vd->vdev_top_zap != 0) {
-               total++;
-               ASSERT0(zap_lookup_int(spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps, vd->vdev_top_zap));
-       }
-       if (vd->vdev_leaf_zap != 0) {
-               total++;
-               ASSERT0(zap_lookup_int(spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
-       }
-
-       for (i = 0; i < vd->vdev_children; i++) {
-               total += vdev_count_verify_zaps(vd->vdev_child[i]);
-       }
-
-       return (total);
-}
-#endif
-
-/*
- * Determine whether the activity check is required.
- */
-static boolean_t
-spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
-    nvlist_t *config)
-{
-       uint64_t state = 0;
-       uint64_t hostid = 0;
-       uint64_t tryconfig_txg = 0;
-       uint64_t tryconfig_timestamp = 0;
-       nvlist_t *nvinfo;
-
-       if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
-               nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
-               (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
-                   &tryconfig_txg);
-               (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
-                   &tryconfig_timestamp);
-       }
-
-       (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
-
-       /*
-        * Disable the MMP activity check - This is used by zdb which
-        * is intended to be used on potentially active pools.
-        */
-       if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
-               return (B_FALSE);
-
-       /*
-        * Skip the activity check when the MMP feature is disabled.
-        */
-       if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
-               return (B_FALSE);
-       /*
-        * If the tryconfig_* values are nonzero, they are the results of an
-        * earlier tryimport.  If they match the uberblock we just found, then
-        * the pool has not changed and we return false so we do not test a
-        * second time.
-        */
-       if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
-           tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp)
-               return (B_FALSE);
-
-       /*
-        * Allow the activity check to be skipped when importing the pool
-        * on the same host which last imported it.  Since the hostid from
-        * configuration may be stale use the one read from the label.
-        */
-       if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
-               hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
-
-       if (hostid == spa_get_hostid())
-               return (B_FALSE);
-
-       /*
-        * Skip the activity test when the pool was cleanly exported.
-        */
-       if (state != POOL_STATE_ACTIVE)
-               return (B_FALSE);
-
-       return (B_TRUE);
-}
-
-/*
- * Perform the import activity check.  If the user canceled the import or
- * we detected activity then fail.
- */
-static int
-spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
-{
-       uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
-       uint64_t txg = ub->ub_txg;
-       uint64_t timestamp = ub->ub_timestamp;
-       uint64_t import_delay = NANOSEC;
-       hrtime_t import_expire;
-       nvlist_t *mmp_label = NULL;
-       vdev_t *rvd = spa->spa_root_vdev;
-       kcondvar_t cv;
-       kmutex_t mtx;
-       int error = 0;
-
-       cv_init(&cv, NULL, CV_DEFAULT, NULL);
-       mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
-       mutex_enter(&mtx);
-
-       /*
-        * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
-        * during the earlier tryimport.  If the txg recorded there is 0 then
-        * the pool is known to be active on another host.
-        *
-        * Otherwise, the pool might be in use on another node.  Check for
-        * changes in the uberblocks on disk if necessary.
-        */
-       if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
-               nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
-                   ZPOOL_CONFIG_LOAD_INFO);
-
-               if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
-                   fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
-                       vdev_uberblock_load(rvd, ub, &mmp_label);
-                       error = SET_ERROR(EREMOTEIO);
-                       goto out;
-               }
-       }
-
-       /*
-        * Preferentially use the zfs_multihost_interval from the node which
-        * last imported the pool.  This value is stored in an MMP uberblock as.
-        *
-        * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
-        */
-       if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay)
-               import_delay = MAX(import_delay, import_intervals *
-                   ub->ub_mmp_delay * MAX(vdev_count_leaves(spa), 1));
-
-       /* Apply a floor using the local default values. */
-       import_delay = MAX(import_delay, import_intervals *
-           MSEC2NSEC(MAX(zfs_multihost_interval, MMP_MIN_INTERVAL)));
-
-       /* Add a small random factor in case of simultaneous imports (0-25%) */
-       import_expire = gethrtime() + import_delay +
-           (import_delay * spa_get_random(250) / 1000);
-
-       while (gethrtime() < import_expire) {
-               vdev_uberblock_load(rvd, ub, &mmp_label);
-
-               if (txg != ub->ub_txg || timestamp != ub->ub_timestamp) {
-                       error = SET_ERROR(EREMOTEIO);
-                       break;
-               }
-
-               if (mmp_label) {
-                       nvlist_free(mmp_label);
-                       mmp_label = NULL;
-               }
-
-               error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
-               if (error != -1) {
-                       error = SET_ERROR(EINTR);
-                       break;
-               }
-               error = 0;
-       }
-
-out:
-       mutex_exit(&mtx);
-       mutex_destroy(&mtx);
-       cv_destroy(&cv);
-
-       /*
-        * If the pool is determined to be active store the status in the
-        * spa->spa_load_info nvlist.  If the remote hostname or hostid are
-        * available from configuration read from disk store them as well.
-        * This allows 'zpool import' to generate a more useful message.
-        *
-        * ZPOOL_CONFIG_MMP_STATE    - observed pool status (mandatory)
-        * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
-        * ZPOOL_CONFIG_MMP_HOSTID   - hostid from the active pool
-        */
-       if (error == EREMOTEIO) {
-               char *hostname = "<unknown>";
-               uint64_t hostid = 0;
-
-               if (mmp_label) {
-                       if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
-                               hostname = fnvlist_lookup_string(mmp_label,
-                                   ZPOOL_CONFIG_HOSTNAME);
-                               fnvlist_add_string(spa->spa_load_info,
-                                   ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
-                       }
-
-                       if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
-                               hostid = fnvlist_lookup_uint64(mmp_label,
-                                   ZPOOL_CONFIG_HOSTID);
-                               fnvlist_add_uint64(spa->spa_load_info,
-                                   ZPOOL_CONFIG_MMP_HOSTID, hostid);
-                       }
-               }
-
-               fnvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
-               fnvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_MMP_TXG, 0);
-
-               error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
-       }
-
-       if (mmp_label)
-               nvlist_free(mmp_label);
-
-       return (error);
-}
-
-/*
- * Load an existing storage pool, using the pool's builtin spa_config as a
- * source of configuration information.
- */
-__attribute__((always_inline))
-static inline int
-spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
-    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
-    char **ereport)
-{
-       int error = 0;
-       nvlist_t *nvroot = NULL;
-       nvlist_t *label;
-       vdev_t *rvd;
-       uberblock_t *ub = &spa->spa_uberblock;
-       uint64_t children, config_cache_txg = spa->spa_config_txg;
-       int orig_mode = spa->spa_mode;
-       int parse, i;
-       uint64_t obj;
-       boolean_t missing_feat_write = B_FALSE;
-       boolean_t activity_check = B_FALSE;
-       nvlist_t *mos_config;
-
-       /*
-        * If this is an untrusted config, access the pool in read-only mode.
-        * This prevents things like resilvering recently removed devices.
-        */
-       if (!mosconfig)
-               spa->spa_mode = FREAD;
-
-       ASSERT(MUTEX_HELD(&spa_namespace_lock));
-
-       spa->spa_load_state = state;
-
-       if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
-               return (SET_ERROR(EINVAL));
-
-       parse = (type == SPA_IMPORT_EXISTING ?
-           VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
-
-       /*
-        * Create "The Godfather" zio to hold all async IOs
-        */
-       spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
-           KM_SLEEP);
-       for (i = 0; i < max_ncpus; i++) {
-               spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
-                   ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
-                   ZIO_FLAG_GODFATHER);
-       }
-
-       /*
-        * Parse the configuration into a vdev tree.  We explicitly set the
-        * value that will be returned by spa_version() since parsing the
-        * configuration requires knowing the version number.
-        */
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-       error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
-       spa_config_exit(spa, SCL_ALL, FTAG);
-
-       if (error != 0)
-               return (error);
-
-       ASSERT(spa->spa_root_vdev == rvd);
-       ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
-       ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
-
-       if (type != SPA_IMPORT_ASSEMBLE) {
-               ASSERT(spa_guid(spa) == pool_guid);
-       }
-
-       /*
-        * Try to open all vdevs, loading each label in the process.
-        */
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-       error = vdev_open(rvd);
-       spa_config_exit(spa, SCL_ALL, FTAG);
-       if (error != 0)
-               return (error);
-
-       /*
-        * We need to validate the vdev labels against the configuration that
-        * we have in hand, which is dependent on the setting of mosconfig. If
-        * mosconfig is true then we're validating the vdev labels based on
-        * that config.  Otherwise, we're validating against the cached config
-        * (zpool.cache) that was read when we loaded the zfs module, and then
-        * later we will recursively call spa_load() and validate against
-        * the vdev config.
-        *
-        * If we're assembling a new pool that's been split off from an
-        * existing pool, the labels haven't yet been updated so we skip
-        * validation for now.
-        */
-       if (type != SPA_IMPORT_ASSEMBLE) {
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               error = vdev_validate(rvd, mosconfig);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-
-               if (error != 0)
-                       return (error);
-
-               if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
-                       return (SET_ERROR(ENXIO));
-       }
-
-       /*
-        * Find the best uberblock.
-        */
-       vdev_uberblock_load(rvd, ub, &label);
-
-       /*
-        * If we weren't able to find a single valid uberblock, return failure.
-        */
-       if (ub->ub_txg == 0) {
-               nvlist_free(label);
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
-       }
-
-       /*
-        * For pools which have the multihost property on determine if the
-        * pool is truly inactive and can be safely imported.  Prevent
-        * hosts which don't have a hostid set from importing the pool.
-        */
-       activity_check = spa_activity_check_required(spa, ub, label, config);
-       if (activity_check) {
-               if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
-                   spa_get_hostid() == 0) {
-                       nvlist_free(label);
-                       fnvlist_add_uint64(spa->spa_load_info,
-                           ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
-                       return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
-               }
-
-               error = spa_activity_check(spa, ub, config);
-               if (error) {
-                       nvlist_free(label);
-                       return (error);
-               }
-
-               fnvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
-               fnvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
-       }
-
-       /*
-        * If the pool has an unsupported version we can't open it.
-        */
-       if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
-               nvlist_free(label);
-               return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
-       }
-
-       if (ub->ub_version >= SPA_VERSION_FEATURES) {
-               nvlist_t *features;
-
-               /*
-                * If we weren't able to find what's necessary for reading the
-                * MOS in the label, return failure.
-                */
-               if (label == NULL || nvlist_lookup_nvlist(label,
-                   ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
-                       nvlist_free(label);
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
-                           ENXIO));
-               }
-
-               /*
-                * Update our in-core representation with the definitive values
-                * from the label.
-                */
-               nvlist_free(spa->spa_label_features);
-               VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
-       }
-
-       nvlist_free(label);
-
-       /*
-        * Look through entries in the label nvlist's features_for_read. If
-        * there is a feature listed there which we don't understand then we
-        * cannot open a pool.
-        */
-       if (ub->ub_version >= SPA_VERSION_FEATURES) {
-               nvlist_t *unsup_feat;
-               nvpair_t *nvp;
-
-               VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
-                   0);
-
-               for (nvp = nvlist_next_nvpair(spa->spa_label_features, NULL);
-                   nvp != NULL;
-                   nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
-                       if (!zfeature_is_supported(nvpair_name(nvp))) {
-                               VERIFY(nvlist_add_string(unsup_feat,
-                                   nvpair_name(nvp), "") == 0);
-                       }
-               }
-
-               if (!nvlist_empty(unsup_feat)) {
-                       VERIFY(nvlist_add_nvlist(spa->spa_load_info,
-                           ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
-                       nvlist_free(unsup_feat);
-                       return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
-                           ENOTSUP));
-               }
-
-               nvlist_free(unsup_feat);
-       }
-
-       /*
-        * If the vdev guid sum doesn't match the uberblock, we have an
-        * incomplete configuration.  We first check to see if the pool
-        * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
-        * If it is, defer the vdev_guid_sum check till later so we
-        * can handle missing vdevs.
-        */
-       if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
-           &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
-           rvd->vdev_guid_sum != ub->ub_guid_sum)
-               return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
-
-       if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_try_repair(spa, config);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-               nvlist_free(spa->spa_config_splitting);
-               spa->spa_config_splitting = NULL;
-       }
-
-       /*
-        * Initialize internal SPA structures.
-        */
-       spa->spa_state = POOL_STATE_ACTIVE;
-       spa->spa_ubsync = spa->spa_uberblock;
-       spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
-           TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
-       spa->spa_first_txg = spa->spa_last_ubsync_txg ?
-           spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
-       spa->spa_claim_max_txg = spa->spa_first_txg;
-       spa->spa_prev_software_version = ub->ub_software_version;
-
-       error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
-       if (error)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
-
-       if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       if (spa_version(spa) >= SPA_VERSION_FEATURES) {
-               boolean_t missing_feat_read = B_FALSE;
-               nvlist_t *unsup_feat, *enabled_feat;
-               spa_feature_t i;
-
-               if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
-                   &spa->spa_feat_for_read_obj) != 0) {
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-               }
-
-               if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
-                   &spa->spa_feat_for_write_obj) != 0) {
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-               }
-
-               if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
-                   &spa->spa_feat_desc_obj) != 0) {
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-               }
-
-               enabled_feat = fnvlist_alloc();
-               unsup_feat = fnvlist_alloc();
-
-               if (!spa_features_check(spa, B_FALSE,
-                   unsup_feat, enabled_feat))
-                       missing_feat_read = B_TRUE;
-
-               if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
-                       if (!spa_features_check(spa, B_TRUE,
-                           unsup_feat, enabled_feat)) {
-                               missing_feat_write = B_TRUE;
-                       }
-               }
-
-               fnvlist_add_nvlist(spa->spa_load_info,
-                   ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
-
-               if (!nvlist_empty(unsup_feat)) {
-                       fnvlist_add_nvlist(spa->spa_load_info,
-                           ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
-               }
-
-               fnvlist_free(enabled_feat);
-               fnvlist_free(unsup_feat);
-
-               if (!missing_feat_read) {
-                       fnvlist_add_boolean(spa->spa_load_info,
-                           ZPOOL_CONFIG_CAN_RDONLY);
-               }
-
-               /*
-                * If the state is SPA_LOAD_TRYIMPORT, our objective is
-                * twofold: to determine whether the pool is available for
-                * import in read-write mode and (if it is not) whether the
-                * pool is available for import in read-only mode. If the pool
-                * is available for import in read-write mode, it is displayed
-                * as available in userland; if it is not available for import
-                * in read-only mode, it is displayed as unavailable in
-                * userland. If the pool is available for import in read-only
-                * mode but not read-write mode, it is displayed as unavailable
-                * in userland with a special note that the pool is actually
-                * available for open in read-only mode.
-                *
-                * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
-                * missing a feature for write, we must first determine whether
-                * the pool can be opened read-only before returning to
-                * userland in order to know whether to display the
-                * abovementioned note.
-                */
-               if (missing_feat_read || (missing_feat_write &&
-                   spa_writeable(spa))) {
-                       return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
-                           ENOTSUP));
-               }
-
-               /*
-                * Load refcounts for ZFS features from disk into an in-memory
-                * cache during SPA initialization.
-                */
-               for (i = 0; i < SPA_FEATURES; i++) {
-                       uint64_t refcount;
-
-                       error = feature_get_refcount_from_disk(spa,
-                           &spa_feature_table[i], &refcount);
-                       if (error == 0) {
-                               spa->spa_feat_refcount_cache[i] = refcount;
-                       } else if (error == ENOTSUP) {
-                               spa->spa_feat_refcount_cache[i] =
-                                   SPA_FEATURE_DISABLED;
-                       } else {
-                               return (spa_vdev_err(rvd,
-                                   VDEV_AUX_CORRUPT_DATA, EIO));
-                       }
-               }
-       }
-
-       if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
-               if (spa_dir_prop(spa, DMU_POOL_FEATURE_ENABLED_TXG,
-                   &spa->spa_feat_enabled_txg_obj) != 0)
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       }
-
-       spa->spa_is_initializing = B_TRUE;
-       error = dsl_pool_open(spa->spa_dsl_pool);
-       spa->spa_is_initializing = B_FALSE;
-       if (error != 0)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       if (!mosconfig) {
-               uint64_t hostid;
-               nvlist_t *policy = NULL, *nvconfig;
-
-               if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-               if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
-                   ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
-                       char *hostname;
-                       unsigned long myhostid = 0;
-
-                       VERIFY(nvlist_lookup_string(nvconfig,
-                           ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
-
-                       myhostid = spa_get_hostid();
-                       if (hostid && myhostid && hostid != myhostid) {
-                               nvlist_free(nvconfig);
-                               return (SET_ERROR(EBADF));
-                       }
-               }
-               if (nvlist_lookup_nvlist(spa->spa_config,
-                   ZPOOL_REWIND_POLICY, &policy) == 0)
-                       VERIFY(nvlist_add_nvlist(nvconfig,
-                           ZPOOL_REWIND_POLICY, policy) == 0);
-
-               spa_config_set(spa, nvconfig);
-               spa_unload(spa);
-               spa_deactivate(spa);
-               spa_activate(spa, orig_mode);
-
-               return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
-       }
-
-       /* Grab the checksum salt from the MOS. */
-       error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
-           DMU_POOL_CHECKSUM_SALT, 1,
-           sizeof (spa->spa_cksum_salt.zcs_bytes),
-           spa->spa_cksum_salt.zcs_bytes);
-       if (error == ENOENT) {
-               /* Generate a new salt for subsequent use */
-               (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
-                   sizeof (spa->spa_cksum_salt.zcs_bytes));
-       } else if (error != 0) {
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       }
-
-       if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
-       if (error != 0)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       /*
-        * Load the bit that tells us to use the new accounting function
-        * (raid-z deflation).  If we have an older pool, this will not
-        * be present.
-        */
-       error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
-           &spa->spa_creation_version);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       /*
-        * Load the persistent error log.  If we have an older pool, this will
-        * not be present.
-        */
-       error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
-           &spa->spa_errlog_scrub);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       /*
-        * Load the history object.  If we have an older pool, this
-        * will not be present.
-        */
-       error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       /*
-        * Load the per-vdev ZAP map. If we have an older pool, this will not
-        * be present; in this case, defer its creation to a later time to
-        * avoid dirtying the MOS this early / out of sync context. See
-        * spa_sync_config_object.
-        */
-
-       /* The sentinel is only available in the MOS config. */
-       if (load_nvlist(spa, spa->spa_config_object, &mos_config) != 0)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       error = spa_dir_prop(spa, DMU_POOL_VDEV_ZAP_MAP,
-           &spa->spa_all_vdev_zaps);
-
-       if (error == ENOENT) {
-               VERIFY(!nvlist_exists(mos_config,
-                   ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
-               spa->spa_avz_action = AVZ_ACTION_INITIALIZE;
-               ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
-       } else if (error != 0) {
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       } else if (!nvlist_exists(mos_config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS)) {
-               /*
-                * An older version of ZFS overwrote the sentinel value, so
-                * we have orphaned per-vdev ZAPs in the MOS. Defer their
-                * destruction to later; see spa_sync_config_object.
-                */
-               spa->spa_avz_action = AVZ_ACTION_DESTROY;
-               /*
-                * We're assuming that no vdevs have had their ZAPs created
-                * before this. Better be sure of it.
-                */
-               ASSERT0(vdev_count_verify_zaps(spa->spa_root_vdev));
-       }
-       nvlist_free(mos_config);
-
-       /*
-        * If we're assembling the pool from the split-off vdevs of
-        * an existing pool, we don't want to attach the spares & cache
-        * devices.
-        */
-
-       /*
-        * Load any hot spares for this pool.
-        */
-       error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
-               ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
-               if (load_nvlist(spa, spa->spa_spares.sav_object,
-                   &spa->spa_spares.sav_config) != 0)
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_load_spares(spa);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-       } else if (error == 0) {
-               spa->spa_spares.sav_sync = B_TRUE;
-       }
-
-       /*
-        * Load any level 2 ARC devices for this pool.
-        */
-       error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
-           &spa->spa_l2cache.sav_object);
-       if (error != 0 && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-       if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
-               ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
-               if (load_nvlist(spa, spa->spa_l2cache.sav_object,
-                   &spa->spa_l2cache.sav_config) != 0)
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_load_l2cache(spa);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-       } else if (error == 0) {
-               spa->spa_l2cache.sav_sync = B_TRUE;
-       }
-
-       spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
-
-       error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
-       if (error && error != ENOENT)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       if (error == 0) {
-               uint64_t autoreplace = 0;
-
-               spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
-               spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
-               spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
-               spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
-               spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
-               spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
-               spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
-                   &spa->spa_dedup_ditto);
-
-               spa->spa_autoreplace = (autoreplace != 0);
-       }
-
-       /*
-        * If the 'multihost' property is set, then never allow a pool to
-        * be imported when the system hostid is zero.  The exception to
-        * this rule is zdb which is always allowed to access pools.
-        */
-       if (spa_multihost(spa) && spa_get_hostid() == 0 &&
-           (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
-               fnvlist_add_uint64(spa->spa_load_info,
-                   ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
-               return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
-       }
-
-       /*
-        * If the 'autoreplace' property is set, then post a resource notifying
-        * the ZFS DE that it should not issue any faults for unopenable
-        * devices.  We also iterate over the vdevs, and post a sysevent for any
-        * unopenable vdevs so that the normal autoreplace handler can take
-        * over.
-        */
-       if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
-               spa_check_removed(spa->spa_root_vdev);
-               /*
-                * For the import case, this is done in spa_import(), because
-                * at this point we're using the spare definitions from
-                * the MOS config, not necessarily from the userland config.
-                */
-               if (state != SPA_LOAD_IMPORT) {
-                       spa_aux_check_removed(&spa->spa_spares);
-                       spa_aux_check_removed(&spa->spa_l2cache);
-               }
-       }
-
-       /*
-        * Load the vdev state for all toplevel vdevs.
-        */
-       vdev_load(rvd);
-
-       /*
-        * Propagate the leaf DTLs we just loaded all the way up the tree.
-        */
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-       vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
-       spa_config_exit(spa, SCL_ALL, FTAG);
-
-       /*
-        * Load the DDTs (dedup tables).
-        */
-       error = ddt_load(spa);
-       if (error != 0)
-               return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-       spa_update_dspace(spa);
-
-       /*
-        * Validate the config, using the MOS config to fill in any
-        * information which might be missing.  If we fail to validate
-        * the config then declare the pool unfit for use. If we're
-        * assembling a pool from a split, the log is not transferred
-        * over.
-        */
-       if (type != SPA_IMPORT_ASSEMBLE) {
-               nvlist_t *nvconfig;
-
-               if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
-
-               if (!spa_config_valid(spa, nvconfig)) {
-                       nvlist_free(nvconfig);
-                       return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
-                           ENXIO));
-               }
-               nvlist_free(nvconfig);
-
-               /*
-                * Now that we've validated the config, check the state of the
-                * root vdev.  If it can't be opened, it indicates one or
-                * more toplevel vdevs are faulted.
-                */
-               if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
-                       return (SET_ERROR(ENXIO));
-
-               if (spa_writeable(spa) && spa_check_logs(spa)) {
-                       *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
-                       return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
-               }
-       }
-
-       if (missing_feat_write) {
-               ASSERT(state == SPA_LOAD_TRYIMPORT);
-
-               /*
-                * At this point, we know that we can open the pool in
-                * read-only mode but not read-write mode. We now have enough
-                * information and can return to userland.
-                */
-               return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
-       }
-
-       /*
-        * We've successfully opened the pool, verify that we're ready
-        * to start pushing transactions.
-        */
-       if (state != SPA_LOAD_TRYIMPORT) {
-               if ((error = spa_load_verify(spa)))
-                       return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
-                           error));
-       }
-
-       if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
-           spa->spa_load_max_txg == UINT64_MAX)) {
-               dmu_tx_t *tx;
-               int need_update = B_FALSE;
-               dsl_pool_t *dp = spa_get_dsl(spa);
-               int c;
-
-               ASSERT(state != SPA_LOAD_TRYIMPORT);
-
-               /*
-                * Claim log blocks that haven't been committed yet.
-                * This must all happen in a single txg.
-                * Note: spa_claim_max_txg is updated by spa_claim_notify(),
-                * invoked from zil_claim_log_block()'s i/o done callback.
-                * Price of rollback is that we abandon the log.
-                */
-               spa->spa_claiming = B_TRUE;
-
-               tx = dmu_tx_create_assigned(dp, spa_first_txg(spa));
-               (void) dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
-                   zil_claim, tx, DS_FIND_CHILDREN);
-               dmu_tx_commit(tx);
-
-               spa->spa_claiming = B_FALSE;
-
-               spa_set_log_state(spa, SPA_LOG_GOOD);
-               spa->spa_sync_on = B_TRUE;
-               txg_sync_start(spa->spa_dsl_pool);
-               mmp_thread_start(spa);
-
-               /*
-                * Wait for all claims to sync.  We sync up to the highest
-                * claimed log block birth time so that claimed log blocks
-                * don't appear to be from the future.  spa_claim_max_txg
-                * will have been set for us by either zil_check_log_chain()
-                * (invoked from spa_check_logs()) or zil_claim() above.
-                */
-               txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
-
-               /*
-                * If the config cache is stale, or we have uninitialized
-                * metaslabs (see spa_vdev_add()), then update the config.
-                *
-                * If this is a verbatim import, trust the current
-                * in-core spa_config and update the disk labels.
-                */
-               if (config_cache_txg != spa->spa_config_txg ||
-                   state == SPA_LOAD_IMPORT ||
-                   state == SPA_LOAD_RECOVER ||
-                   (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
-                       need_update = B_TRUE;
-
-               for (c = 0; c < rvd->vdev_children; c++)
-                       if (rvd->vdev_child[c]->vdev_ms_array == 0)
-                               need_update = B_TRUE;
-
-               /*
-                * Update the config cache asychronously in case we're the
-                * root pool, in which case the config cache isn't writable yet.
-                */
-               if (need_update)
-                       spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
-
-               /*
-                * Check all DTLs to see if anything needs resilvering.
-                */
-               if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
-                   vdev_resilver_needed(rvd, NULL, NULL))
-                       spa_async_request(spa, SPA_ASYNC_RESILVER);
-
-               /*
-                * Log the fact that we booted up (so that we can detect if
-                * we rebooted in the middle of an operation).
-                */
-               spa_history_log_version(spa, "open", NULL);
-
-               /*
-                * Delete any inconsistent datasets.
-                */
-               (void) dmu_objset_find(spa_name(spa),
-                   dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
-
-               /*
-                * Clean up any stale temporary dataset userrefs.
-                */
-               dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
-       }
-
-       return (0);
-}
-
-static int
-spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
-{
-       int mode = spa->spa_mode;
-
-       spa_unload(spa);
-       spa_deactivate(spa);
-
-       spa->spa_load_max_txg = spa->spa_uberblock.ub_txg - 1;
-
-       spa_activate(spa, mode);
-       spa_async_suspend(spa);
-
-       return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
-}
-
-/*
- * If spa_load() fails this function will try loading prior txg's. If
- * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
- * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
- * function will not rewind the pool and will return the same error as
- * spa_load().
- */
-static int
-spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
-    uint64_t max_request, int rewind_flags)
-{
-       nvlist_t *loadinfo = NULL;
-       nvlist_t *config = NULL;
-       int load_error, rewind_error;
-       uint64_t safe_rewind_txg;
-       uint64_t min_txg;
-
-       if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
-               spa->spa_load_max_txg = spa->spa_load_txg;
-               spa_set_log_state(spa, SPA_LOG_CLEAR);
-       } else {
-               spa->spa_load_max_txg = max_request;
-               if (max_request != UINT64_MAX)
-                       spa->spa_extreme_rewind = B_TRUE;
-       }
-
-       load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
-           mosconfig);
-       if (load_error == 0)
-               return (0);
-
-       if (spa->spa_root_vdev != NULL)
-               config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
-
-       spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
-       spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
-
-       if (rewind_flags & ZPOOL_NEVER_REWIND) {
-               nvlist_free(config);
-               return (load_error);
-       }
-
-       if (state == SPA_LOAD_RECOVER) {
-               /* Price of rolling back is discarding txgs, including log */
-               spa_set_log_state(spa, SPA_LOG_CLEAR);
-       } else {
-               /*
-                * If we aren't rolling back save the load info from our first
-                * import attempt so that we can restore it after attempting
-                * to rewind.
-                */
-               loadinfo = spa->spa_load_info;
-               spa->spa_load_info = fnvlist_alloc();
-       }
-
-       spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
-       safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
-       min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
-           TXG_INITIAL : safe_rewind_txg;
-
-       /*
-        * Continue as long as we're finding errors, we're still within
-        * the acceptable rewind range, and we're still finding uberblocks
-        */
-       while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
-           spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
-               if (spa->spa_load_max_txg < safe_rewind_txg)
-                       spa->spa_extreme_rewind = B_TRUE;
-               rewind_error = spa_load_retry(spa, state, mosconfig);
-       }
-
-       spa->spa_extreme_rewind = B_FALSE;
-       spa->spa_load_max_txg = UINT64_MAX;
-
-       if (config && (rewind_error || state != SPA_LOAD_RECOVER))
-               spa_config_set(spa, config);
-       else
-               nvlist_free(config);
-
-       if (state == SPA_LOAD_RECOVER) {
-               ASSERT3P(loadinfo, ==, NULL);
-               return (rewind_error);
-       } else {
-               /* Store the rewind info as part of the initial load info */
-               fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
-                   spa->spa_load_info);
-
-               /* Restore the initial load info */
-               fnvlist_free(spa->spa_load_info);
-               spa->spa_load_info = loadinfo;
-
-               return (load_error);
-       }
-}
-
-/*
- * Pool Open/Import
- *
- * The import case is identical to an open except that the configuration is sent
- * down from userland, instead of grabbed from the configuration cache.  For the
- * case of an open, the pool configuration will exist in the
- * POOL_STATE_UNINITIALIZED state.
- *
- * The stats information (gen/count/ustats) is used to gather vdev statistics at
- * the same time open the pool, without having to keep around the spa_t in some
- * ambiguous state.
- */
-static int
-spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
-    nvlist_t **config)
-{
-       spa_t *spa;
-       spa_load_state_t state = SPA_LOAD_OPEN;
-       int error;
-       int locked = B_FALSE;
-       int firstopen = B_FALSE;
-
-       *spapp = NULL;
-
-       /*
-        * As disgusting as this is, we need to support recursive calls to this
-        * function because dsl_dir_open() is called during spa_load(), and ends
-        * up calling spa_open() again.  The real fix is to figure out how to
-        * avoid dsl_dir_open() calling this in the first place.
-        */
-       if (mutex_owner(&spa_namespace_lock) != curthread) {
-               mutex_enter(&spa_namespace_lock);
-               locked = B_TRUE;
-       }
-
-       if ((spa = spa_lookup(pool)) == NULL) {
-               if (locked)
-                       mutex_exit(&spa_namespace_lock);
-               return (SET_ERROR(ENOENT));
-       }
-
-       if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
-               zpool_rewind_policy_t policy;
-
-               firstopen = B_TRUE;
-
-               zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
-                   &policy);
-               if (policy.zrp_request & ZPOOL_DO_REWIND)
-                       state = SPA_LOAD_RECOVER;
-
-               spa_activate(spa, spa_mode_global);
-
-               if (state != SPA_LOAD_RECOVER)
-                       spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
-
-               error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
-                   policy.zrp_request);
-
-               if (error == EBADF) {
-                       /*
-                        * If vdev_validate() returns failure (indicated by
-                        * EBADF), it indicates that one of the vdevs indicates
-                        * that the pool has been exported or destroyed.  If
-                        * this is the case, the config cache is out of sync and
-                        * we should remove the pool from the namespace.
-                        */
-                       spa_unload(spa);
-                       spa_deactivate(spa);
-                       spa_config_sync(spa, B_TRUE, B_TRUE);
-                       spa_remove(spa);
-                       if (locked)
-                               mutex_exit(&spa_namespace_lock);
-                       return (SET_ERROR(ENOENT));
-               }
-
-               if (error) {
-                       /*
-                        * We can't open the pool, but we still have useful
-                        * information: the state of each vdev after the
-                        * attempted vdev_open().  Return this to the user.
-                        */
-                       if (config != NULL && spa->spa_config) {
-                               VERIFY(nvlist_dup(spa->spa_config, config,
-                                   KM_SLEEP) == 0);
-                               VERIFY(nvlist_add_nvlist(*config,
-                                   ZPOOL_CONFIG_LOAD_INFO,
-                                   spa->spa_load_info) == 0);
-                       }
-                       spa_unload(spa);
-                       spa_deactivate(spa);
-                       spa->spa_last_open_failed = error;
-                       if (locked)
-                               mutex_exit(&spa_namespace_lock);
-                       *spapp = NULL;
-                       return (error);
-               }
-       }
-
-       spa_open_ref(spa, tag);
-
-       if (config != NULL)
-               *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
-
-       /*
-        * If we've recovered the pool, pass back any information we
-        * gathered while doing the load.
-        */
-       if (state == SPA_LOAD_RECOVER) {
-               VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
-                   spa->spa_load_info) == 0);
-       }
-
-       if (locked) {
-               spa->spa_last_open_failed = 0;
-               spa->spa_last_ubsync_txg = 0;
-               spa->spa_load_txg = 0;
-               mutex_exit(&spa_namespace_lock);
-       }
-
-       if (firstopen)
-               zvol_create_minors(spa, spa_name(spa), B_TRUE);
-
-       *spapp = spa;
-
-       return (0);
-}
-
-int
-spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
-    nvlist_t **config)
-{
-       return (spa_open_common(name, spapp, tag, policy, config));
-}
-
-int
-spa_open(const char *name, spa_t **spapp, void *tag)
-{
-       return (spa_open_common(name, spapp, tag, NULL, NULL));
-}
-
-/*
- * Lookup the given spa_t, incrementing the inject count in the process,
- * preventing it from being exported or destroyed.
- */
-spa_t *
-spa_inject_addref(char *name)
-{
-       spa_t *spa;
-
-       mutex_enter(&spa_namespace_lock);
-       if ((spa = spa_lookup(name)) == NULL) {
-               mutex_exit(&spa_namespace_lock);
-               return (NULL);
-       }
-       spa->spa_inject_ref++;
-       mutex_exit(&spa_namespace_lock);
-
-       return (spa);
-}
-
-void
-spa_inject_delref(spa_t *spa)
-{
-       mutex_enter(&spa_namespace_lock);
-       spa->spa_inject_ref--;
-       mutex_exit(&spa_namespace_lock);
-}
-
-/*
- * Add spares device information to the nvlist.
- */
-static void
-spa_add_spares(spa_t *spa, nvlist_t *config)
-{
-       nvlist_t **spares;
-       uint_t i, nspares;
-       nvlist_t *nvroot;
-       uint64_t guid;
-       vdev_stat_t *vs;
-       uint_t vsc;
-       uint64_t pool;
-
-       ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
-
-       if (spa->spa_spares.sav_count == 0)
-               return;
-
-       VERIFY(nvlist_lookup_nvlist(config,
-           ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
-       VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
-           ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
-       if (nspares != 0) {
-               VERIFY(nvlist_add_nvlist_array(nvroot,
-                   ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
-               VERIFY(nvlist_lookup_nvlist_array(nvroot,
-                   ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
-
-               /*
-                * Go through and find any spares which have since been
-                * repurposed as an active spare.  If this is the case, update
-                * their status appropriately.
-                */
-               for (i = 0; i < nspares; i++) {
-                       VERIFY(nvlist_lookup_uint64(spares[i],
-                           ZPOOL_CONFIG_GUID, &guid) == 0);
-                       if (spa_spare_exists(guid, &pool, NULL) &&
-                           pool != 0ULL) {
-                               VERIFY(nvlist_lookup_uint64_array(
-                                   spares[i], ZPOOL_CONFIG_VDEV_STATS,
-                                   (uint64_t **)&vs, &vsc) == 0);
-                               vs->vs_state = VDEV_STATE_CANT_OPEN;
-                               vs->vs_aux = VDEV_AUX_SPARED;
-                       }
-               }
-       }
-}
-
-/*
- * Add l2cache device information to the nvlist, including vdev stats.
- */
-static void
-spa_add_l2cache(spa_t *spa, nvlist_t *config)
-{
-       nvlist_t **l2cache;
-       uint_t i, j, nl2cache;
-       nvlist_t *nvroot;
-       uint64_t guid;
-       vdev_t *vd;
-       vdev_stat_t *vs;
-       uint_t vsc;
-
-       ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
-
-       if (spa->spa_l2cache.sav_count == 0)
-               return;
-
-       VERIFY(nvlist_lookup_nvlist(config,
-           ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
-       VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
-           ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
-       if (nl2cache != 0) {
-               VERIFY(nvlist_add_nvlist_array(nvroot,
-                   ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
-               VERIFY(nvlist_lookup_nvlist_array(nvroot,
-                   ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
-
-               /*
-                * Update level 2 cache device stats.
-                */
-
-               for (i = 0; i < nl2cache; i++) {
-                       VERIFY(nvlist_lookup_uint64(l2cache[i],
-                           ZPOOL_CONFIG_GUID, &guid) == 0);
-
-                       vd = NULL;
-                       for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
-                               if (guid ==
-                                   spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
-                                       vd = spa->spa_l2cache.sav_vdevs[j];
-                                       break;
-                               }
-                       }
-                       ASSERT(vd != NULL);
-
-                       VERIFY(nvlist_lookup_uint64_array(l2cache[i],
-                           ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
-                           == 0);
-                       vdev_get_stats(vd, vs);
-                       vdev_config_generate_stats(vd, l2cache[i]);
-
-               }
-       }
-}
-
-static void
-spa_feature_stats_from_disk(spa_t *spa, nvlist_t *features)
-{
-       zap_cursor_t zc;
-       zap_attribute_t za;
-
-       if (spa->spa_feat_for_read_obj != 0) {
-               for (zap_cursor_init(&zc, spa->spa_meta_objset,
-                   spa->spa_feat_for_read_obj);
-                   zap_cursor_retrieve(&zc, &za) == 0;
-                   zap_cursor_advance(&zc)) {
-                       ASSERT(za.za_integer_length == sizeof (uint64_t) &&
-                           za.za_num_integers == 1);
-                       VERIFY0(nvlist_add_uint64(features, za.za_name,
-                           za.za_first_integer));
-               }
-               zap_cursor_fini(&zc);
-       }
-
-       if (spa->spa_feat_for_write_obj != 0) {
-               for (zap_cursor_init(&zc, spa->spa_meta_objset,
-                   spa->spa_feat_for_write_obj);
-                   zap_cursor_retrieve(&zc, &za) == 0;
-                   zap_cursor_advance(&zc)) {
-                       ASSERT(za.za_integer_length == sizeof (uint64_t) &&
-                           za.za_num_integers == 1);
-                       VERIFY0(nvlist_add_uint64(features, za.za_name,
-                           za.za_first_integer));
-               }
-               zap_cursor_fini(&zc);
-       }
-}
-
-static void
-spa_feature_stats_from_cache(spa_t *spa, nvlist_t *features)
-{
-       int i;
-
-       for (i = 0; i < SPA_FEATURES; i++) {
-               zfeature_info_t feature = spa_feature_table[i];
-               uint64_t refcount;
-
-               if (feature_get_refcount(spa, &feature, &refcount) != 0)
-                       continue;
-
-               VERIFY0(nvlist_add_uint64(features, feature.fi_guid, refcount));
-       }
-}
-
-/*
- * Store a list of pool features and their reference counts in the
- * config.
- *
- * The first time this is called on a spa, allocate a new nvlist, fetch
- * the pool features and reference counts from disk, then save the list
- * in the spa. In subsequent calls on the same spa use the saved nvlist
- * and refresh its values from the cached reference counts.  This
- * ensures we don't block here on I/O on a suspended pool so 'zpool
- * clear' can resume the pool.
- */
-static void
-spa_add_feature_stats(spa_t *spa, nvlist_t *config)
-{
-       nvlist_t *features;
-
-       ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
-
-       mutex_enter(&spa->spa_feat_stats_lock);
-       features = spa->spa_feat_stats;
-
-       if (features != NULL) {
-               spa_feature_stats_from_cache(spa, features);
-       } else {
-               VERIFY0(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP));
-               spa->spa_feat_stats = features;
-               spa_feature_stats_from_disk(spa, features);
-       }
-
-       VERIFY0(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
-           features));
-
-       mutex_exit(&spa->spa_feat_stats_lock);
-}
-
-int
-spa_get_stats(const char *name, nvlist_t **config,
-    char *altroot, size_t buflen)
-{
-       int error;
-       spa_t *spa;
-
-       *config = NULL;
-       error = spa_open_common(name, &spa, FTAG, NULL, config);
-
-       if (spa != NULL) {
-               /*
-                * This still leaves a window of inconsistency where the spares
-                * or l2cache devices could change and the config would be
-                * self-inconsistent.
-                */
-               spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
-
-               if (*config != NULL) {
-                       uint64_t loadtimes[2];
-
-                       loadtimes[0] = spa->spa_loaded_ts.tv_sec;
-                       loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
-                       VERIFY(nvlist_add_uint64_array(*config,
-                           ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
-
-                       VERIFY(nvlist_add_uint64(*config,
-                           ZPOOL_CONFIG_ERRCOUNT,
-                           spa_get_errlog_size(spa)) == 0);
-
-                       if (spa_suspended(spa))
-                               VERIFY(nvlist_add_uint64(*config,
-                                   ZPOOL_CONFIG_SUSPENDED,
-                                   spa->spa_failmode) == 0);
-
-                       spa_add_spares(spa, *config);
-                       spa_add_l2cache(spa, *config);
-                       spa_add_feature_stats(spa, *config);
-               }
-       }
-
-       /*
-        * We want to get the alternate root even for faulted pools, so we cheat
-        * and call spa_lookup() directly.
-        */
-       if (altroot) {
-               if (spa == NULL) {
-                       mutex_enter(&spa_namespace_lock);
-                       spa = spa_lookup(name);
-                       if (spa)
-                               spa_altroot(spa, altroot, buflen);
-                       else
-                               altroot[0] = '\0';
-                       spa = NULL;
-                       mutex_exit(&spa_namespace_lock);
-               } else {
-                       spa_altroot(spa, altroot, buflen);
-               }
-       }
-
-       if (spa != NULL) {
-               spa_config_exit(spa, SCL_CONFIG, FTAG);
-               spa_close(spa, FTAG);
-       }
-
-       return (error);
-}
-
-/*
- * Validate that the auxiliary device array is well formed.  We must have an
- * array of nvlists, each which describes a valid leaf vdev.  If this is an
- * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
- * specified, as long as they are well-formed.
- */
-static int
-spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
-    spa_aux_vdev_t *sav, const char *config, uint64_t version,
-    vdev_labeltype_t label)
-{
-       nvlist_t **dev;
-       uint_t i, ndev;
-       vdev_t *vd;
-       int error;
-
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
-
-       /*
-        * It's acceptable to have no devs specified.
-        */
-       if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
-               return (0);
-
-       if (ndev == 0)
-               return (SET_ERROR(EINVAL));
-
-       /*
-        * Make sure the pool is formatted with a version that supports this
-        * device type.
-        */
-       if (spa_version(spa) < version)
-               return (SET_ERROR(ENOTSUP));
-
-       /*
-        * Set the pending device list so we correctly handle device in-use
-        * checking.
-        */
-       sav->sav_pending = dev;
-       sav->sav_npending = ndev;
-
-       for (i = 0; i < ndev; i++) {
-               if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
-                   mode)) != 0)
-                       goto out;
-
-               if (!vd->vdev_ops->vdev_op_leaf) {
-                       vdev_free(vd);
-                       error = SET_ERROR(EINVAL);
-                       goto out;
-               }
-
-               vd->vdev_top = vd;
-
-               if ((error = vdev_open(vd)) == 0 &&
-                   (error = vdev_label_init(vd, crtxg, label)) == 0) {
-                       VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
-                           vd->vdev_guid) == 0);
-               }
-
-               vdev_free(vd);
-
-               if (error &&
-                   (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
-                       goto out;
-               else
-                       error = 0;
-       }
-
-out:
-       sav->sav_pending = NULL;
-       sav->sav_npending = 0;
-       return (error);
-}
-
-static int
-spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
-{
-       int error;
-
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
-
-       if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
-           &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
-           VDEV_LABEL_SPARE)) != 0) {
-               return (error);
-       }
-
-       return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
-           &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
-           VDEV_LABEL_L2CACHE));
-}
-
-static void
-spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
-    const char *config)
-{
-       int i;
-
-       if (sav->sav_config != NULL) {
-               nvlist_t **olddevs;
-               uint_t oldndevs;
-               nvlist_t **newdevs;
-
-               /*
-                * Generate new dev list by concatenating with the
-                * current dev list.
-                */
-               VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
-                   &olddevs, &oldndevs) == 0);
-
-               newdevs = kmem_alloc(sizeof (void *) *
-                   (ndevs + oldndevs), KM_SLEEP);
-               for (i = 0; i < oldndevs; i++)
-                       VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
-                           KM_SLEEP) == 0);
-               for (i = 0; i < ndevs; i++)
-                       VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
-                           KM_SLEEP) == 0);
-
-               VERIFY(nvlist_remove(sav->sav_config, config,
-                   DATA_TYPE_NVLIST_ARRAY) == 0);
-
-               VERIFY(nvlist_add_nvlist_array(sav->sav_config,
-                   config, newdevs, ndevs + oldndevs) == 0);
-               for (i = 0; i < oldndevs + ndevs; i++)
-                       nvlist_free(newdevs[i]);
-               kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
-       } else {
-               /*
-                * Generate a new dev list.
-                */
-               VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
-                   KM_SLEEP) == 0);
-               VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
-                   devs, ndevs) == 0);
-       }
-}
-
-/*
- * Stop and drop level 2 ARC devices
- */
-void
-spa_l2cache_drop(spa_t *spa)
-{
-       vdev_t *vd;
-       int i;
-       spa_aux_vdev_t *sav = &spa->spa_l2cache;
-
-       for (i = 0; i < sav->sav_count; i++) {
-               uint64_t pool;
-
-               vd = sav->sav_vdevs[i];
-               ASSERT(vd != NULL);
-
-               if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
-                   pool != 0ULL && l2arc_vdev_present(vd))
-                       l2arc_remove_vdev(vd);
-       }
-}
-
-/*
- * Pool Creation
- */
-int
-spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
-    nvlist_t *zplprops)
-{
-       spa_t *spa;
-       char *altroot = NULL;
-       vdev_t *rvd;
-       dsl_pool_t *dp;
-       dmu_tx_t *tx;
-       int error = 0;
-       uint64_t txg = TXG_INITIAL;
-       nvlist_t **spares, **l2cache;
-       uint_t nspares, nl2cache;
-       uint64_t version, obj;
-       boolean_t has_features;
-       nvpair_t *elem;
-       int c, i;
-       char *poolname;
-       nvlist_t *nvl;
-
-       if (nvlist_lookup_string(props, "tname", &poolname) != 0)
-               poolname = (char *)pool;
-
-       /*
-        * If this pool already exists, return failure.
-        */
-       mutex_enter(&spa_namespace_lock);
-       if (spa_lookup(poolname) != NULL) {
-               mutex_exit(&spa_namespace_lock);
-               return (SET_ERROR(EEXIST));
-       }
-
-       /*
-        * Allocate a new spa_t structure.
-        */
-       nvl = fnvlist_alloc();
-       fnvlist_add_string(nvl, ZPOOL_CONFIG_POOL_NAME, pool);
-       (void) nvlist_lookup_string(props,
-           zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
-       spa = spa_add(poolname, nvl, altroot);
-       fnvlist_free(nvl);
-       spa_activate(spa, spa_mode_global);
-
-       if (props && (error = spa_prop_validate(spa, props))) {
-               spa_deactivate(spa);
-               spa_remove(spa);
-               mutex_exit(&spa_namespace_lock);
-               return (error);
-       }
-
-       /*
-        * Temporary pool names should never be written to disk.
-        */
-       if (poolname != pool)
-               spa->spa_import_flags |= ZFS_IMPORT_TEMP_NAME;
-
-       has_features = B_FALSE;
-       for (elem = nvlist_next_nvpair(props, NULL);
-           elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
-               if (zpool_prop_feature(nvpair_name(elem)))
-                       has_features = B_TRUE;
-       }
-
-       if (has_features || nvlist_lookup_uint64(props,
-           zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
-               version = SPA_VERSION;
-       }
-       ASSERT(SPA_VERSION_IS_SUPPORTED(version));
-
-       spa->spa_first_txg = txg;
-       spa->spa_uberblock.ub_txg = txg - 1;
-       spa->spa_uberblock.ub_version = version;
-       spa->spa_ubsync = spa->spa_uberblock;
-       spa->spa_load_state = SPA_LOAD_CREATE;
-
-       /*
-        * Create "The Godfather" zio to hold all async IOs
-        */
-       spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
-           KM_SLEEP);
-       for (i = 0; i < max_ncpus; i++) {
-               spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
-                   ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
-                   ZIO_FLAG_GODFATHER);
-       }
-
-       /*
-        * Create the root vdev.
-        */
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-
-       error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
-
-       ASSERT(error != 0 || rvd != NULL);
-       ASSERT(error != 0 || spa->spa_root_vdev == rvd);
-
-       if (error == 0 && !zfs_allocatable_devs(nvroot))
-               error = SET_ERROR(EINVAL);
-
-       if (error == 0 &&
-           (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
-           (error = spa_validate_aux(spa, nvroot, txg,
-           VDEV_ALLOC_ADD)) == 0) {
-               for (c = 0; c < rvd->vdev_children; c++) {
-                       vdev_metaslab_set_size(rvd->vdev_child[c]);
-                       vdev_expand(rvd->vdev_child[c], txg);
-               }
-       }
-
-       spa_config_exit(spa, SCL_ALL, FTAG);
-
-       if (error != 0) {
-               spa_unload(spa);
-               spa_deactivate(spa);
-               spa_remove(spa);
-               mutex_exit(&spa_namespace_lock);
-               return (error);
-       }
-
-       /*
-        * Get the list of spares, if specified.
-        */
-       if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
-           &spares, &nspares) == 0) {
-               VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
-                   KM_SLEEP) == 0);
-               VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
-                   ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_load_spares(spa);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-               spa->spa_spares.sav_sync = B_TRUE;
-       }
-
-       /*
-        * Get the list of level 2 cache devices, if specified.
-        */
-       if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
-           &l2cache, &nl2cache) == 0) {
-               VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
-                   NV_UNIQUE_NAME, KM_SLEEP) == 0);
-               VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
-                   ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_load_l2cache(spa);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-               spa->spa_l2cache.sav_sync = B_TRUE;
-       }
-
-       spa->spa_is_initializing = B_TRUE;
-       spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
-       spa->spa_meta_objset = dp->dp_meta_objset;
-       spa->spa_is_initializing = B_FALSE;
-
-       /*
-        * Create DDTs (dedup tables).
-        */
-       ddt_create(spa);
-
-       spa_update_dspace(spa);
-
-       tx = dmu_tx_create_assigned(dp, txg);
-
-       /*
-        * Create the pool's history object.
-        */
-       if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history)
-               spa_history_create_obj(spa, tx);
-
-       spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
-       spa_history_log_version(spa, "create", tx);
-
-       /*
-        * Create the pool config object.
-        */
-       spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
-           DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
-           DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
-
-       if (zap_add(spa->spa_meta_objset,
-           DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
-           sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
-               cmn_err(CE_PANIC, "failed to add pool config");
-       }
-
-       if (spa_version(spa) >= SPA_VERSION_FEATURES)
-               spa_feature_create_zap_objects(spa, tx);
-
-       if (zap_add(spa->spa_meta_objset,
-           DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
-           sizeof (uint64_t), 1, &version, tx) != 0) {
-               cmn_err(CE_PANIC, "failed to add pool version");
-       }
-
-       /* Newly created pools with the right version are always deflated. */
-       if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
-               spa->spa_deflate = TRUE;
-               if (zap_add(spa->spa_meta_objset,
-                   DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
-                   sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
-                       cmn_err(CE_PANIC, "failed to add deflate");
-               }
-       }
-
-       /*
-        * Create the deferred-free bpobj.  Turn off compression
-        * because sync-to-convergence takes longer if the blocksize
-        * keeps changing.
-        */
-       obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
-       dmu_object_set_compress(spa->spa_meta_objset, obj,
-           ZIO_COMPRESS_OFF, tx);
-       if (zap_add(spa->spa_meta_objset,
-           DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
-           sizeof (uint64_t), 1, &obj, tx) != 0) {
-               cmn_err(CE_PANIC, "failed to add bpobj");
-       }
-       VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
-           spa->spa_meta_objset, obj));
-
-       /*
-        * Generate some random noise for salted checksums to operate on.
-        */
-       (void) random_get_pseudo_bytes(spa->spa_cksum_salt.zcs_bytes,
-           sizeof (spa->spa_cksum_salt.zcs_bytes));
-
-       /*
-        * Set pool properties.
-        */
-       spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
-       spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
-       spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
-       spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
-       spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
-
-       if (props != NULL) {
-               spa_configfile_set(spa, props, B_FALSE);
-               spa_sync_props(props, tx);
-       }
-
-       dmu_tx_commit(tx);
-
-       spa->spa_sync_on = B_TRUE;
-       txg_sync_start(spa->spa_dsl_pool);
-       mmp_thread_start(spa);
-
-       /*
-        * We explicitly wait for the first transaction to complete so that our
-        * bean counters are appropriately updated.
-        */
-       txg_wait_synced(spa->spa_dsl_pool, txg);
-
-       spa_config_sync(spa, B_FALSE, B_TRUE);
-
-       /*
-        * Don't count references from objsets that are already closed
-        * and are making their way through the eviction process.
-        */
-       spa_evicting_os_wait(spa);
-       spa->spa_minref = refcount_count(&spa->spa_refcount);
-       spa->spa_load_state = SPA_LOAD_NONE;
-
-       mutex_exit(&spa_namespace_lock);
-
-       return (0);
-}
-
-/*
- * Import a non-root pool into the system.
- */
-int
-spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
-{
-       spa_t *spa;
-       char *altroot = NULL;
-       spa_load_state_t state = SPA_LOAD_IMPORT;
-       zpool_rewind_policy_t policy;
-       uint64_t mode = spa_mode_global;
-       uint64_t readonly = B_FALSE;
-       int error;
-       nvlist_t *nvroot;
-       nvlist_t **spares, **l2cache;
-       uint_t nspares, nl2cache;
-
-       /*
-        * If a pool with this name exists, return failure.
-        */
-       mutex_enter(&spa_namespace_lock);
-       if (spa_lookup(pool) != NULL) {
-               mutex_exit(&spa_namespace_lock);
-               return (SET_ERROR(EEXIST));
-       }
-
-       /*
-        * Create and initialize the spa structure.
-        */
-       (void) nvlist_lookup_string(props,
-           zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
-       (void) nvlist_lookup_uint64(props,
-           zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
-       if (readonly)
-               mode = FREAD;
-       spa = spa_add(pool, config, altroot);
-       spa->spa_import_flags = flags;
-
-       /*
-        * Verbatim import - Take a pool and insert it into the namespace
-        * as if it had been loaded at boot.
-        */
-       if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
-               if (props != NULL)
-                       spa_configfile_set(spa, props, B_FALSE);
-
-               spa_config_sync(spa, B_FALSE, B_TRUE);
-               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
-
-               mutex_exit(&spa_namespace_lock);
-               return (0);
-       }
-
-       spa_activate(spa, mode);
-
-       /*
-        * Don't start async tasks until we know everything is healthy.
-        */
-       spa_async_suspend(spa);
-
-       zpool_get_rewind_policy(config, &policy);
-       if (policy.zrp_request & ZPOOL_DO_REWIND)
-               state = SPA_LOAD_RECOVER;
-
-       /*
-        * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
-        * because the user-supplied config is actually the one to trust when
-        * doing an import.
-        */
-       if (state != SPA_LOAD_RECOVER)
-               spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
-
-       error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
-           policy.zrp_request);
-
-       /*
-        * Propagate anything learned while loading the pool and pass it
-        * back to caller (i.e. rewind info, missing devices, etc).
-        */
-       VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
-           spa->spa_load_info) == 0);
-
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-       /*
-        * Toss any existing sparelist, as it doesn't have any validity
-        * anymore, and conflicts with spa_has_spare().
-        */
-       if (spa->spa_spares.sav_config) {
-               nvlist_free(spa->spa_spares.sav_config);
-               spa->spa_spares.sav_config = NULL;
-               spa_load_spares(spa);
-       }
-       if (spa->spa_l2cache.sav_config) {
-               nvlist_free(spa->spa_l2cache.sav_config);
-               spa->spa_l2cache.sav_config = NULL;
-               spa_load_l2cache(spa);
-       }
-
-       VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
-           &nvroot) == 0);
-       spa_config_exit(spa, SCL_ALL, FTAG);
-
-       if (props != NULL)
-               spa_configfile_set(spa, props, B_FALSE);
-
-       if (error != 0 || (props && spa_writeable(spa) &&
-           (error = spa_prop_set(spa, props)))) {
-               spa_unload(spa);
-               spa_deactivate(spa);
-               spa_remove(spa);
-               mutex_exit(&spa_namespace_lock);
-               return (error);
-       }
-
-       spa_async_resume(spa);
-
-       /*
-        * Override any spares and level 2 cache devices as specified by
-        * the user, as these may have correct device names/devids, etc.
-        */
-       if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
-           &spares, &nspares) == 0) {
-               if (spa->spa_spares.sav_config)
-                       VERIFY(nvlist_remove(spa->spa_spares.sav_config,
-                           ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
-               else
-                       VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
-                           NV_UNIQUE_NAME, KM_SLEEP) == 0);
-               VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
-                   ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_load_spares(spa);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-               spa->spa_spares.sav_sync = B_TRUE;
-       }
-       if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
-           &l2cache, &nl2cache) == 0) {
-               if (spa->spa_l2cache.sav_config)
-                       VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
-                           ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
-               else
-                       VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
-                           NV_UNIQUE_NAME, KM_SLEEP) == 0);
-               VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
-                   ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-               spa_load_l2cache(spa);
-               spa_config_exit(spa, SCL_ALL, FTAG);
-               spa->spa_l2cache.sav_sync = B_TRUE;
-       }
-
-       /*
-        * Check for any removed devices.
-        */
-       if (spa->spa_autoreplace) {
-               spa_aux_check_removed(&spa->spa_spares);
-               spa_aux_check_removed(&spa->spa_l2cache);
-       }
-
-       if (spa_writeable(spa)) {
-               /*
-                * Update the config cache to include the newly-imported pool.
-                */
-               spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
-       }
-
-       /*
-        * It's possible that the pool was expanded while it was exported.
-        * We kick off an async task to handle this for us.
-        */
-       spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
-
-       spa_history_log_version(spa, "import", NULL);
-
-       spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
-
-       zvol_create_minors(spa, pool, B_TRUE);
-
-       mutex_exit(&spa_namespace_lock);
-
-       return (0);
-}
-
-nvlist_t *
-spa_tryimport(nvlist_t *tryconfig)
-{
-       nvlist_t *config = NULL;
-       char *poolname;
-       spa_t *spa;
-       uint64_t state;
-       int error;
-
-       if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
-               return (NULL);
-
-       if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
-               return (NULL);
-
-       /*
-        * Create and initialize the spa structure.
-        */
-       mutex_enter(&spa_namespace_lock);
-       spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
-       spa_activate(spa, FREAD);
-
-       /*
-        * Pass off the heavy lifting to spa_load().
-        * Pass TRUE for mosconfig because the user-supplied config
-        * is actually the one to trust when doing an import.
-        */
-       error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
-
-       /*
-        * If 'tryconfig' was at least parsable, return the current config.
-        */
-       if (spa->spa_root_vdev != NULL) {
-               config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
-               VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
-                   poolname) == 0);
-               VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
-                   state) == 0);
-               VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
-                   spa->spa_uberblock.ub_timestamp) == 0);
-               VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
-                   spa->spa_load_info) == 0);
-               VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
-                   spa->spa_errata) == 0);
-
-               /*
-                * If the bootfs property exists on this pool then we
-                * copy it out so that external consumers can tell which
-                * pools are bootable.
-                */
-               if ((!error || error == EEXIST) && spa->spa_bootfs) {
-                       char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
-
-                       /*
-                        * We have to play games with the name since the
-                        * pool was opened as TRYIMPORT_NAME.
-                        */
-                       if (dsl_dsobj_to_dsname(spa_name(spa),
-                           spa->spa_bootfs, tmpname) == 0) {
-                               char *cp;
-                               char *dsname;
-
-                               dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
-
-                               cp = strchr(tmpname, '/');
-                               if (cp == NULL) {
-                                       (void) strlcpy(dsname, tmpname,
-                                           MAXPATHLEN);
-                               } else {
-                                       (void) snprintf(dsname, MAXPATHLEN,
-                                           "%s/%s", poolname, ++cp);
-                               }
-                               VERIFY(nvlist_add_string(config,
-                                   ZPOOL_CONFIG_BOOTFS, dsname) == 0);
-                               kmem_free(dsname, MAXPATHLEN);
-                       }
-                       kmem_free(tmpname, MAXPATHLEN);
-               }
-
-               /*
-                * Add the list of hot spares and level 2 cache devices.
-                */
-               spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
-               spa_add_spares(spa, config);
-               spa_add_l2cache(spa, config);
-               spa_config_exit(spa, SCL_CONFIG, FTAG);
-       }
-
-       spa_unload(spa);
-       spa_deactivate(spa);
-       spa_remove(spa);
-       mutex_exit(&spa_namespace_lock);
-
-       return (config);
-}
-
-/*
- * Pool export/destroy
- *
- * The act of destroying or exporting a pool is very simple.  We make sure there
- * is no more pending I/O and any references to the pool are gone.  Then, we
- * update the pool state and sync all the labels to disk, removing the
- * configuration from the cache afterwards. If the 'hardforce' flag is set, then
- * we don't sync the labels or remove the configuration cache.
- */
-static int
-spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
-    boolean_t force, boolean_t hardforce)
-{
-       spa_t *spa;
-
-       if (oldconfig)
-               *oldconfig = NULL;
-
-       if (!(spa_mode_global & FWRITE))
-               return (SET_ERROR(EROFS));
-
-       mutex_enter(&spa_namespace_lock);
-       if ((spa = spa_lookup(pool)) == NULL) {
-               mutex_exit(&spa_namespace_lock);
-               return (SET_ERROR(ENOENT));
-       }
-
-       /*
-        * Put a hold on the pool, drop the namespace lock, stop async tasks,
-        * reacquire the namespace lock, and see if we can export.
-        */
-       spa_open_ref(spa, FTAG);
-       mutex_exit(&spa_namespace_lock);
-       spa_async_suspend(spa);
-       if (spa->spa_zvol_taskq) {
-               zvol_remove_minors(spa, spa_name(spa), B_TRUE);
-               taskq_wait(spa->spa_zvol_taskq);
-       }
-       mutex_enter(&spa_namespace_lock);
-       spa_close(spa, FTAG);
-
-       if (spa->spa_state == POOL_STATE_UNINITIALIZED)
-               goto export_spa;
-       /*
-        * The pool will be in core if it's openable, in which case we can
-        * modify its state.  Objsets may be open only because they're dirty,
-        * so we have to force it to sync before checking spa_refcnt.
-        */
-       if (spa->spa_sync_on) {
-               txg_wait_synced(spa->spa_dsl_pool, 0);
-               spa_evicting_os_wait(spa);
-       }
-
-       /*
-        * A pool cannot be exported or destroyed if there are active
-        * references.  If we are resetting a pool, allow references by
-        * fault injection handlers.
-        */
-       if (!spa_refcount_zero(spa) ||
-           (spa->spa_inject_ref != 0 &&
-           new_state != POOL_STATE_UNINITIALIZED)) {
-               spa_async_resume(spa);
-               mutex_exit(&spa_namespace_lock);
-               return (SET_ERROR(EBUSY));
-       }
-
-       if (spa->spa_sync_on) {
-               /*
-                * A pool cannot be exported if it has an active shared spare.
-                * This is to prevent other pools stealing the active spare
-                * from an exported pool. At user's own will, such pool can
-                * be forcedly exported.
-                */
-               if (!force && new_state == POOL_STATE_EXPORTED &&
-                   spa_has_active_shared_spare(spa)) {
-                       spa_async_resume(spa);
-                       mutex_exit(&spa_namespace_lock);
-                       return (SET_ERROR(EXDEV));
-               }
-
-               /*
-                * We want this to be reflected on every label,
-                * so mark them all dirty.  spa_unload() will do the
-                * final sync that pushes these changes out.
-                */
-               if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
-                       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-                       spa->spa_state = new_state;
-                       spa->spa_final_txg = spa_last_synced_txg(spa) +
-                           TXG_DEFER_SIZE + 1;
-                       vdev_config_dirty(spa->spa_root_vdev);
-                       spa_config_exit(spa, SCL_ALL, FTAG);
-               }
-       }
-
-export_spa:
-       if (new_state == POOL_STATE_DESTROYED)
-               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
-       else if (new_state == POOL_STATE_EXPORTED)
-               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT);
-
-       if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
-               spa_unload(spa);
-               spa_deactivate(spa);
-       }
-
-       if (oldconfig && spa->spa_config)
-               VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
-
-       if (new_state != POOL_STATE_UNINITIALIZED) {
-               if (!hardforce)
-                       spa_config_sync(spa, B_TRUE, B_TRUE);
-               spa_remove(spa);
-       }
-       mutex_exit(&spa_namespace_lock);
-
-       return (0);
-}
-
-/*
- * Destroy a storage pool.
- */
-int
-spa_destroy(char *pool)
-{
-       return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
-           B_FALSE, B_FALSE));
-}
-
-/*
- * Export a storage pool.
- */
-int
-spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
-    boolean_t hardforce)
-{
-       return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
-           force, hardforce));
-}
-
-/*
- * Similar to spa_export(), this unloads the spa_t without actually removing it
- * from the namespace in any way.
- */
-int
-spa_reset(char *pool)
-{
-       return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
-           B_FALSE, B_FALSE));
-}
-
-/*
- * ==========================================================================
- * Device manipulation
- * ==========================================================================
- */
-
-/*
- * Add a device to a storage pool.
- */
-int
-spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
-{
-       uint64_t txg, id;
-       int error;
-       vdev_t *rvd = spa->spa_root_vdev;
-       vdev_t *vd, *tvd;
-       nvlist_t **spares, **l2cache;
-       uint_t nspares, nl2cache;
-       int c;
-
-       ASSERT(spa_writeable(spa));
-
-       txg = spa_vdev_enter(spa);
-
-       if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
-           VDEV_ALLOC_ADD)) != 0)
-               return (spa_vdev_exit(spa, NULL, txg, error));
-
-       spa->spa_pending_vdev = vd;     /* spa_vdev_exit() will clear this */
-
-       if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
-           &nspares) != 0)
-               nspares = 0;
-
-       if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
-           &nl2cache) != 0)
-               nl2cache = 0;
-
-       if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
-               return (spa_vdev_exit(spa, vd, txg, EINVAL));
-
-       if (vd->vdev_children != 0 &&
-           (error = vdev_create(vd, txg, B_FALSE)) != 0)
-               return (spa_vdev_exit(spa, vd, txg, error));
-
-       /*
-        * We must validate the spares and l2cache devices after checking the
-        * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
-        */
-       if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
-               return (spa_vdev_exit(spa, vd, txg, error));
-
-       /*
-        * Transfer each new top-level vdev from vd to rvd.
-        */
-       for (c = 0; c < vd->vdev_children; c++) {
-
-               /*
-                * Set the vdev id to the first hole, if one exists.
-                */
-               for (id = 0; id < rvd->vdev_children; id++) {
-                       if (rvd->vdev_child[id]->vdev_ishole) {
-                               vdev_free(rvd->vdev_child[id]);
-                               break;
-                       }
-               }
-               tvd = vd->vdev_child[c];
-               vdev_remove_child(vd, tvd);
-               tvd->vdev_id = id;
-               vdev_add_child(rvd, tvd);
-               vdev_config_dirty(tvd);
-       }
-
-       if (nspares != 0) {
-               spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
-                   ZPOOL_CONFIG_SPARES);
-               spa_load_spares(spa);
-               spa->spa_spares.sav_sync = B_TRUE;
-       }
-
-       if (nl2cache != 0) {
-               spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
-                   ZPOOL_CONFIG_L2CACHE);
-               spa_load_l2cache(spa);
-               spa->spa_l2cache.sav_sync = B_TRUE;
-       }
-
-       /*
-        * We have to be careful when adding new vdevs to an existing pool.
-        * If other threads start allocating from these vdevs before we
-        * sync the config cache, and we lose power, then upon reboot we may
-        * fail to open the pool because there are DVAs that the config cache
-        * can't translate.  Therefore, we first add the vdevs without
-        * initializing metaslabs; sync the config cache (via spa_vdev_exit());
-        * and then let spa_config_update() initialize the new metaslabs.
-        *
-        * spa_load() checks for added-but-not-initialized vdevs, so that
-        * if we lose power at any point in this sequence, the remaining
-        * steps will be completed the next time we load the pool.
-        */
-       (void) spa_vdev_exit(spa, vd, txg, 0);
-
-       mutex_enter(&spa_namespace_lock);
-       spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
-       spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
-       mutex_exit(&spa_namespace_lock);
-
-       return (0);
-}
-
-/*
- * Attach a device to a mirror.  The arguments are the path to any device
- * in the mirror, and the nvroot for the new device.  If the path specifies
- * a device that is not mirrored, we automatically insert the mirror vdev.
- *
- * If 'replacing' is specified, the new device is intended to replace the
- * existing device; in this case the two devices are made into their own
- * mirror using the 'replacing' vdev, which is functionally identical to
- * the mirror vdev (it actually reuses all the same ops) but has a few
- * extra rules: you can't attach to it after it's been created, and upon
- * completion of resilvering, the first disk (the one being replaced)
- * is automatically detached.
- */
-int
-spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
-{
-       uint64_t txg, dtl_max_txg;
-       vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
-       vdev_ops_t *pvops;
-       char *oldvdpath, *newvdpath;
-       int newvd_isspare;
-       int error;
-       ASSERTV(vdev_t *rvd = spa->spa_root_vdev);
-
-       ASSERT(spa_writeable(spa));
-
-       txg = spa_vdev_enter(spa);
-
-       oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
-
-       if (oldvd == NULL)
-               return (spa_vdev_exit(spa, NULL, txg, ENODEV));
-
-       if (!oldvd->vdev_ops->vdev_op_leaf)
-               return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
-
-       pvd = oldvd->vdev_parent;
-
-       if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
-           VDEV_ALLOC_ATTACH)) != 0)
-               return (spa_vdev_exit(spa, NULL, txg, EINVAL));
-
-       if (newrootvd->vdev_children != 1)
-               return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
-
-       newvd = newrootvd->vdev_child[0];
-
-       if (!newvd->vdev_ops->vdev_op_leaf)
-               return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
-
-       if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
-               return (spa_vdev_exit(spa, newrootvd, txg, error));
-
-       /*
-        * Spares can't replace logs
-        */
-       if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
-               return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
-
-       if (!replacing) {
-               /*
-                * For attach, the only allowable parent is a mirror or the root
-                * vdev.
-                */
-               if (pvd->vdev_ops != &vdev_mirror_ops &&
-                   pvd->vdev_ops != &vdev_root_ops)
-                       return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
-
-               pvops = &vdev_mirror_ops;
-       } else {
-               /*
-                * Active hot spares can only be replaced by inactive hot
-                * spares.
-                */
-               if (pvd->vdev_ops == &vdev_spare_ops &&
-                   oldvd->vdev_isspare &&
-                   !spa_has_spare(spa, newvd->vdev_guid))
-                       return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
-
-               /*
-                * If the source is a hot spare, and the parent isn't already a
-                * spare, then we want to create a new hot spare.  Otherwise, we
-                * want to create a replacing vdev.  The user is not allowed to
-                * attach to a spared vdev child unless the 'isspare' state is
-                * the same (spare replaces spare, non-spare replaces
-                * non-spare).
-                */
-               if (pvd->vdev_ops == &vdev_replacing_ops &&
-                   spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
-                       return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
-               } else if (pvd->vdev_ops == &vdev_spare_ops &&
-                   newvd->vdev_isspare != oldvd->vdev_isspare) {
-                       return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
-               }
-
-               if (newvd->vdev_isspare)
-                       pvops = &vdev_spare_ops;
-               else
-                       pvops = &vdev_replacing_ops;
-       }
-
-       /*
-        * Make sure the new device is big enough.
-        */
-       if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
-               return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
-
-       /*
-        * The new device cannot have a higher alignment requirement
-        * than the top-level vdev.
-        */
-       if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
-               return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
-
-       /*
-        * If this is an in-place replacement, update oldvd's path and devid
-        * to make it distinguishable from newvd, and unopenable from now on.
-        */
-       if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
-               spa_strfree(oldvd->vdev_path);
-               oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
-                   KM_SLEEP);
-               (void) sprintf(oldvd->vdev_path, "%s/%s",
-                   newvd->vdev_path, "old");
-               if (oldvd->vdev_devid != NULL) {
-                       spa_strfree(oldvd->vdev_devid);
-                       oldvd->vdev_devid = NULL;
-               }
-       }
-
-       /* mark the device being resilvered */
-       newvd->vdev_resilver_txg = txg;
-
-       /*
-        * If the parent is not a mirror, or if we're replacing, insert the new
-        * mirror/replacing/spare vdev above oldvd.
-        */
-       if (pvd->vdev_ops != pvops)
-               pvd = vdev_add_parent(oldvd, pvops);
-
-       ASSERT(pvd->vdev_top->vdev_parent == rvd);
-       ASSERT(pvd->vdev_ops == pvops);
-       ASSERT(oldvd->vdev_parent == pvd);
-
-       /*
-        * Extract the new device from its root and add it to pvd.
-        */
-       vdev_remove_child(newrootvd, newvd);
-       newvd->vdev_id = pvd->vdev_children;
-       newvd->vdev_crtxg = oldvd->vdev_crtxg;
-       vdev_add_child(pvd, newvd);
-
-       /*
-        * Reevaluate the parent vdev state.
-        */
-       vdev_propagate_state(pvd);
-
-       tvd = newvd->vdev_top;
-       ASSERT(pvd->vdev_top == tvd);
-       ASSERT(tvd->vdev_parent == rvd);
-
-       vdev_config_dirty(tvd);
-
-       /*
-        * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
-        * for any dmu_sync-ed blocks.  It will propagate upward when
-        * spa_vdev_exit() calls vdev_dtl_reassess().
-        */
-       dtl_max_txg = txg + TXG_CONCURRENT_STATES;
-
-       vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
-           dtl_max_txg - TXG_INITIAL);
-
-       if (newvd->vdev_isspare) {
-               spa_spare_activate(newvd);
-               spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
-       }
-
-       oldvdpath = spa_strdup(oldvd->vdev_path);
-       newvdpath = spa_strdup(newvd->vdev_path);
-       newvd_isspare = newvd->vdev_isspare;
-
-       /*
-        * Mark newvd's DTL dirty in this txg.
-        */
-       vdev_dirty(tvd, VDD_DTL, newvd, txg);
-
-       /*
-        * Schedule the resilver to restart in the future. We do this to
-        * ensure that dmu_sync-ed blocks have been stitched into the
-        * respective datasets.
-        */
-       dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
-
-       if (spa->spa_bootfs)
-               spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
-
-       spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
-
-       /*
-        * Commit the config
-        */
-       (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
-
-       spa_history_log_internal(spa, "vdev attach", NULL,
-           "%s vdev=%s %s vdev=%s",
-           replacing && newvd_isspare ? "spare in" :
-           replacing ? "replace" : "attach", newvdpath,
-           replacing ? "for" : "to", oldvdpath);
-
-       spa_strfree(oldvdpath);
-       spa_strfree(newvdpath);
-
-       return (0);
-}
-
-/*
- * Detach a device from a mirror or replacing vdev.
- *
- * If 'replace_done' is specified, only detach if the parent
- * is a replacing vdev.
- */
-int
-spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
-{
-       uint64_t txg;
-       int error;
-       vdev_t *vd, *pvd, *cvd, *tvd;
-       boolean_t unspare = B_FALSE;
-       uint64_t unspare_guid = 0;
-       char *vdpath;
-       int c, t;
-       ASSERTV(vdev_t *rvd = spa->spa_root_vdev);
-       ASSERT(spa_writeable(spa));
-
-       txg = spa_vdev_enter(spa);
-
-       vd = spa_lookup_by_guid(spa, guid, B_FALSE);
-
-       if (vd == NULL)
-               return (spa_vdev_exit(spa, NULL, txg, ENODEV));
-
-       if (!vd->vdev_ops->vdev_op_leaf)
-               return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
-
-       pvd = vd->vdev_parent;
-
-       /*
-        * If the parent/child relationship is not as expected, don't do it.
-        * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
-        * vdev that's replacing B with C.  The user's intent in replacing
-        * is to go from M(A,B) to M(A,C).  If the user decides to cancel
-        * the replace by detaching C, the expected behavior is to end up
-        * M(A,B).  But suppose that right after deciding to detach C,
-        * the replacement of B completes.  We would have M(A,C), and then
-        * ask to detach C, which would leave us with just A -- not what
-        * the user wanted.  To prevent this, we make sure that the
-        * parent/child relationship hasn't changed -- in this example,
-        * that C's parent is still the replacing vdev R.
-        */
-       if (pvd->vdev_guid != pguid && pguid != 0)
-               return (spa_vdev_exit(spa, NULL, txg, EBUSY));
-
-       /*
-        * Only 'replacing' or 'spare' vdevs can be replaced.
-        */
-       if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
-           pvd->vdev_ops != &vdev_spare_ops)
-               return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
-
-       ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
-           spa_version(spa) >= SPA_VERSION_SPARES);
-
-       /*
-        * Only mirror, replacing, and spare vdevs support detach.
-        */
-       if (pvd->vdev_ops != &vdev_replacing_ops &&
-           pvd->vdev_ops != &vdev_mirror_ops &&
-           pvd->vdev_ops != &vdev_spare_ops)
-               return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
-
-       /*
-        * If this device has the only valid copy of some data,
-        * we cannot safely detach it.
-        */
-       if (vdev_dtl_required(vd))
-               return (spa_vdev_exit(spa, NULL, txg, EBUSY));
-
-       ASSERT(pvd->vdev_children >= 2);
-
-       /*
-        * If we are detaching the second disk from a replacing vdev, then
-        * check to see if we changed the original vdev's path to have "/old"
-        * at the end in spa_vdev_attach().  If so, undo that change now.
-        */
-       if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
-           vd->vdev_path != NULL) {
-               size_t len = strlen(vd->vdev_path);
-
-               for (c = 0; c < pvd->vdev_children; c++) {
-                       cvd = pvd->vdev_child[c];
-
-                       if (cvd == vd || cvd->vdev_path == NULL)
-                               continue;
-
-                       if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
-                           strcmp(cvd->vdev_path + len, "/old") == 0) {
-                               spa_strfree(cvd->vdev_path);
-                               cvd->vdev_path = spa_strdup(vd->vdev_path);
-                               break;
-                       }
-               }
-       }
-
-       /*
-        * If we are detaching the original disk from a spare, then it implies
-        * that the spare should become a real disk, and be removed from the
-        * active spare list for the pool.
-        */
-       if (pvd->vdev_ops == &vdev_spare_ops &&
-           vd->vdev_id == 0 &&
-           pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
-               unspare = B_TRUE;
-
-       /*
-        * Erase the disk labels so the disk can be used for other things.
-        * This must be done after all other error cases are handled,
-        * but before we disembowel vd (so we can still do I/O to it).
-        * But if we can't do it, don't treat the error as fatal --
-        * it may be that the unwritability of the disk is the reason
-        * it's being detached!
-        */
-       error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
-
-       /*
-        * Remove vd from its parent and compact the parent's children.
-        */
-       vdev_remove_child(pvd, vd);
-       vdev_compact_children(pvd);
-
-       /*
-        * Remember one of the remaining children so we can get tvd below.
-        */
-       cvd = pvd->vdev_child[pvd->vdev_children - 1];
-
-       /*
-        * If we need to remove the remaining child from the list of hot spares,
-        * do it now, marking the vdev as no longer a spare in the process.
-        * We must do this before vdev_remove_parent(), because that can
-        * change the GUID if it creates a new toplevel GUID.  For a similar
-        * reason, we must remove the spare now, in the same txg as the detach;
-        * otherwise someone could attach a new sibling, change the GUID, and
-        * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
-        */
-       if (unspare) {
-               ASSERT(cvd->vdev_isspare);
-               spa_spare_remove(cvd);
-               unspare_guid = cvd->vdev_guid;
-               (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
-               cvd->vdev_unspare = B_TRUE;
-       }
-
-       /*
-        * If the parent mirror/replacing vdev only has one child,
-        * the parent is no longer needed.  Remove it from the tree.
-        */
-       if (pvd->vdev_children == 1) {
-               if (pvd->vdev_ops == &vdev_spare_ops)
-                       cvd->vdev_unspare = B_FALSE;
-               vdev_remove_parent(cvd);
-       }
-
-
-       /*
-        * We don't set tvd until now because the parent we just removed
-        * may have been the previous top-level vdev.
-        */
-       tvd = cvd->vdev_top;
-       ASSERT(tvd->vdev_parent == rvd);
-
-       /*
-        * Reevaluate the parent vdev state.
-        */
-       vdev_propagate_state(cvd);
-
-       /*
-        * If the 'autoexpand' property is set on the pool then automatically
-        * try to expand the size of the pool. For example if the device we
-        * just detached was smaller than the others, it may be possible to
-        * add metaslabs (i.e. grow the pool). We need to reopen the vdev
-        * first so that we can obtain the updated sizes of the leaf vdevs.
-        */
-       if (spa->spa_autoexpand) {
-               vdev_reopen(tvd);
-               vdev_expand(tvd, txg);
-       }
-
-       vdev_config_dirty(tvd);
-
-       /*
-        * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
-        * vd->vdev_detached is set and free vd's DTL object in syncing context.
-        * But first make sure we're not on any *other* txg's DTL list, to
-        * prevent vd from being accessed after it's freed.
-        */
-       vdpath = spa_strdup(vd->vdev_path ? vd->vdev_path : "none");
-       for (t = 0; t < TXG_SIZE; t++)
-               (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
-       vd->vdev_detached = B_TRUE;
-       vdev_dirty(tvd, VDD_DTL, vd, txg);
-
-       spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
-
-       /* hang on to the spa before we release the lock */
-       spa_open_ref(spa, FTAG);
-
-       error = spa_vdev_exit(spa, vd, txg, 0);
-
-       spa_history_log_internal(spa, "detach", NULL,
-           "vdev=%s", vdpath);
-       spa_strfree(vdpath);
-
-       /*
-        * If this was the removal of the original device in a hot spare vdev,
-        * then we want to go through and remove the device from the hot spare
-        * list of every other pool.
-        */
-       if (unspare) {
-               spa_t *altspa = NULL;
-
-               mutex_enter(&spa_namespace_lock);
-               while ((altspa = spa_next(altspa)) != NULL) {
-                       if (altspa->spa_state != POOL_STATE_ACTIVE ||
-                           altspa == spa)
-                               continue;
-
-                       spa_open_ref(altspa, FTAG);
-                       mutex_exit(&spa_namespace_lock);
-                       (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
-                       mutex_enter(&spa_namespace_lock);
-                       spa_close(altspa, FTAG);
-               }
-               mutex_exit(&spa_namespace_lock);
-
-               /* search the rest of the vdevs for spares to remove */
-               spa_vdev_resilver_done(spa);
-       }
-
-       /* all done with the spa; OK to release */
-       mutex_enter(&spa_namespace_lock);
-       spa_close(spa, FTAG);
-       mutex_exit(&spa_namespace_lock);
-
-       return (error);
-}
-
-/*
- * Split a set of devices from their mirrors, and create a new pool from them.
- */
-int
-spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
-    nvlist_t *props, boolean_t exp)
-{
-       int error = 0;
-       uint64_t txg, *glist;
-       spa_t *newspa;
-       uint_t c, children, lastlog;
-       nvlist_t **child, *nvl, *tmp;
-       dmu_tx_t *tx;
-       char *altroot = NULL;
-       vdev_t *rvd, **vml = NULL;                      /* vdev modify list */
-       boolean_t activate_slog;
-
-       ASSERT(spa_writeable(spa));
-
-       txg = spa_vdev_enter(spa);
-
-       /* clear the log and flush everything up to now */
-       activate_slog = spa_passivate_log(spa);
-       (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
-       error = spa_offline_log(spa);
-       txg = spa_vdev_config_enter(spa);
-
-       if (activate_slog)
-               spa_activate_log(spa);
-
-       if (error != 0)
-               return (spa_vdev_exit(spa, NULL, txg, error));
-
-       /* check new spa name before going any further */
-       if (spa_lookup(newname) != NULL)
-               return (spa_vdev_exit(spa, NULL, txg, EEXIST));
-
-       /*
-        * scan through all the children to ensure they're all mirrors
-        */
-       if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
-           nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
-           &children) != 0)
-               return (spa_vdev_exit(spa, NULL, txg, EINVAL));
-
-       /* first, check to ensure we've got the right child count */
-       rvd = spa->spa_root_vdev;
-       lastlog = 0;
-       for (c = 0; c < rvd->vdev_children; c++) {
-               vdev_t *vd = rvd->vdev_child[c];
-
-               /* don't count the holes & logs as children */
-               if (vd->vdev_islog || vd->vdev_ishole) {
-                       if (lastlog == 0)
-                               lastlog = c;
-                       continue;
-               }
-
-               lastlog = 0;
-       }
-       if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
-               return (spa_vdev_exit(spa, NULL, txg, EINVAL));
-
-       /* next, ensure no spare or cache devices are part of the split */
-       if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
-           nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
-               return (spa_vdev_exit(spa, NULL, txg, EINVAL));
-
-       vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
-       glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
-
-       /* then, loop over each vdev and validate it */
-       for (c = 0; c < children; c++) {
-               uint64_t is_hole = 0;
-
-               (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
-                   &is_hole);
-
-               if (is_hole != 0) {
-                       if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
-                           spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
-                               continue;
-                       } else {
-                               error = SET_ERROR(EINVAL);
-                               break;
-                       }
-               }
-
-               /* which disk is going to be split? */
-               if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
-                   &glist[c]) != 0) {
-                       error = SET_ERROR(EINVAL);
-                       break;
-               }
-
-               /* look it up in the spa */
-               vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
-               if (vml[c] == NULL) {
-                       error = SET_ERROR(ENODEV);
-                       break;
-               }
-
-               /* make sure there's nothing stopping the split */
-               if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
-                   vml[c]->vdev_islog ||
-                   vml[c]->vdev_ishole ||
-                   vml[c]->vdev_isspare ||
-                   vml[c]->vdev_isl2cache ||
-                   !vdev_writeable(vml[c]) ||
-                   vml[c]->vdev_children != 0 ||
-                   vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
-                   c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
-                       error = SET_ERROR(EINVAL);
-                       break;
-               }
-
-               if (vdev_dtl_required(vml[c])) {
-                       error = SET_ERROR(EBUSY);
-                       break;
-               }
-
-               /* we need certain info from the top level */
-               VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
-                   vml[c]->vdev_top->vdev_ms_array) == 0);
-               VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
-                   vml[c]->vdev_top->vdev_ms_shift) == 0);
-               VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
-                   vml[c]->vdev_top->vdev_asize) == 0);
-               VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
-                   vml[c]->vdev_top->vdev_ashift) == 0);
-
-               /* transfer per-vdev ZAPs */
-               ASSERT3U(vml[c]->vdev_leaf_zap, !=, 0);
-               VERIFY0(nvlist_add_uint64(child[c],
-                   ZPOOL_CONFIG_VDEV_LEAF_ZAP, vml[c]->vdev_leaf_zap));
-
-               ASSERT3U(vml[c]->vdev_top->vdev_top_zap, !=, 0);
-               VERIFY0(nvlist_add_uint64(child[c],
-                   ZPOOL_CONFIG_VDEV_TOP_ZAP,
-                   vml[c]->vdev_parent->vdev_top_zap));
-       }
-
-       if (error != 0) {
-               kmem_free(vml, children * sizeof (vdev_t *));
-               kmem_free(glist, children * sizeof (uint64_t));
-               return (spa_vdev_exit(spa, NULL, txg, error));
-       }
-
-       /* stop writers from using the disks */
-       for (c = 0; c < children; c++) {
-               if (vml[c] != NULL)
-                       vml[c]->vdev_offline = B_TRUE;
-       }
-       vdev_reopen(spa->spa_root_vdev);
-
-       /*
-        * Temporarily record the splitting vdevs in the spa config.  This
-        * will disappear once the config is regenerated.
-        */
-       VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
-       VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
-           glist, children) == 0);
-       kmem_free(glist, children * sizeof (uint64_t));
-
-       mutex_enter(&spa->spa_props_lock);
-       VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
-           nvl) == 0);
-       mutex_exit(&spa->spa_props_lock);
-       spa->spa_config_splitting = nvl;
-       vdev_config_dirty(spa->spa_root_vdev);
-
-       /* configure and create the new pool */
-       VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
-       VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
-           exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
-       VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
-           spa_version(spa)) == 0);
-       VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
-           spa->spa_config_txg) == 0);
-       VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
-           spa_generate_guid(NULL)) == 0);
-       VERIFY0(nvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS));
-       (void) nvlist_lookup_string(props,
-           zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
-
-       /* add the new pool to the namespace */
-       newspa = spa_add(newname, config, altroot);
-       newspa->spa_avz_action = AVZ_ACTION_REBUILD;
-       newspa->spa_config_txg = spa->spa_config_txg;
-       spa_set_log_state(newspa, SPA_LOG_CLEAR);
-
-       /* release the spa config lock, retaining the namespace lock */
-       spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
-
-       if (zio_injection_enabled)
-               zio_handle_panic_injection(spa, FTAG, 1);
-
-       spa_activate(newspa, spa_mode_global);
-       spa_async_suspend(newspa);
-
-       /* create the new pool from the disks of the original pool */
-       error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
-       if (error)
-               goto out;
-
-       /* if that worked, generate a real config for the new pool */
-       if (newspa->spa_root_vdev != NULL) {
-               VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
-                   NV_UNIQUE_NAME, KM_SLEEP) == 0);
-               VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
-                   ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
-               spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
-                   B_TRUE));
-       }
-
-       /* set the props */
-       if (props != NULL) {
-               spa_configfile_set(newspa, props, B_FALSE);
-               error = spa_prop_set(newspa, props);
-               if (error)
-                       goto out;
-       }
-
-       /* flush everything */
-       txg = spa_vdev_config_enter(newspa);
-       vdev_config_dirty(newspa->spa_root_vdev);
-       (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
-
-       if (zio_injection_enabled)
-               zio_handle_panic_injection(spa, FTAG, 2);
-
-       spa_async_resume(newspa);
-
-       /* finally, update the original pool's config */
-       txg = spa_vdev_config_enter(spa);
-       tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
-       error = dmu_tx_assign(tx, TXG_WAIT);
-       if (error != 0)
-               dmu_tx_abort(tx);
-       for (c = 0; c < children; c++) {
-               if (vml[c] != NULL) {
-                       vdev_split(vml[c]);
-                       if (error == 0)
-                               spa_history_log_internal(spa, "detach", tx,
-                                   "vdev=%s", vml[c]->vdev_path);
-
-                       vdev_free(vml[c]);
-               }
-       }
-       spa->spa_avz_action = AVZ_ACTION_REBUILD;
-       vdev_config_dirty(spa->spa_root_vdev);
-       spa->spa_config_splitting = NULL;
-       nvlist_free(nvl);
-       if (error == 0)
-               dmu_tx_commit(tx);
-       (void) spa_vdev_exit(spa, NULL, txg, 0);
-
-       if (zio_injection_enabled)
-               zio_handle_panic_injection(spa, FTAG, 3);
-
-       /* split is complete; log a history record */
-       spa_history_log_internal(newspa, "split", NULL,
-           "from pool %s", spa_name(spa));
-
-       kmem_free(vml, children * sizeof (vdev_t *));
-
-       /* if we're not going to mount the filesystems in userland, export */
-       if (exp)
-               error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
-                   B_FALSE, B_FALSE);
-
-       return (error);
-
-out:
-       spa_unload(newspa);
-       spa_deactivate(newspa);
-       spa_remove(newspa);
-
-       txg = spa_vdev_config_enter(spa);
-
-       /* re-online all offlined disks */
-       for (c = 0; c < children; c++) {
-               if (vml[c] != NULL)
-                       vml[c]->vdev_offline = B_FALSE;
-       }
-       vdev_reopen(spa->spa_root_vdev);
-
-       nvlist_free(spa->spa_config_splitting);
-       spa->spa_config_splitting = NULL;
-       (void) spa_vdev_exit(spa, NULL, txg, error);
-
-       kmem_free(vml, children * sizeof (vdev_t *));
-       return (error);
-}
-
-static nvlist_t *
-spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
-{
-       int i;
-
-       for (i = 0; i < count; i++) {
-               uint64_t guid;
-
-               VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
-                   &guid) == 0);
-
-               if (guid == target_guid)
-                       return (nvpp[i]);
-       }
-
-       return (NULL);
-}
-
-static void
-spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
-    nvlist_t *dev_to_remove)
-{
-       nvlist_t **newdev = NULL;
-       int i, j;
-
-       if (count > 1)
-               newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
-
-       for (i = 0, j = 0; i < count; i++) {
-               if (dev[i] == dev_to_remove)
-                       continue;
-               VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
-       }
-
-       VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
-       VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
-
-       for (i = 0; i < count - 1; i++)
-               nvlist_free(newdev[i]);
-
-       if (count > 1)
-               kmem_free(newdev, (count - 1) * sizeof (void *));
-}
-
-/*
- * Evacuate the device.
- */
-static int
-spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
-{
-       uint64_t txg;
-       int error = 0;
-
-       ASSERT(MUTEX_HELD(&spa_namespace_lock));
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
-       ASSERT(vd == vd->vdev_top);
-
-       /*
-        * Evacuate the device.  We don't hold the config lock as writer
-        * since we need to do I/O but we do keep the
-        * spa_namespace_lock held.  Once this completes the device
-        * should no longer have any blocks allocated on it.
-        */
-       if (vd->vdev_islog) {
-               if (vd->vdev_stat.vs_alloc != 0)
-                       error = spa_offline_log(spa);
-       } else {
-               error = SET_ERROR(ENOTSUP);
-       }
-
-       if (error)
-               return (error);
-
-       /*
-        * The evacuation succeeded.  Remove any remaining MOS metadata
-        * associated with this vdev, and wait for these changes to sync.
-        */
-       ASSERT0(vd->vdev_stat.vs_alloc);
-       txg = spa_vdev_config_enter(spa);
-       vd->vdev_removing = B_TRUE;
-       vdev_dirty_leaves(vd, VDD_DTL, txg);
-       vdev_config_dirty(vd);
-       spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
-
-       return (0);
-}
-
-/*
- * Complete the removal by cleaning up the namespace.
- */
-static void
-spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
-{
-       vdev_t *rvd = spa->spa_root_vdev;
-       uint64_t id = vd->vdev_id;
-       boolean_t last_vdev = (id == (rvd->vdev_children - 1));
-
-       ASSERT(MUTEX_HELD(&spa_namespace_lock));
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
-       ASSERT(vd == vd->vdev_top);
-
-       /*
-        * Only remove any devices which are empty.
-        */
-       if (vd->vdev_stat.vs_alloc != 0)
-               return;
-
-       (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
-
-       if (list_link_active(&vd->vdev_state_dirty_node))
-               vdev_state_clean(vd);
-       if (list_link_active(&vd->vdev_config_dirty_node))
-               vdev_config_clean(vd);
-
-       vdev_free(vd);
-
-       if (last_vdev) {
-               vdev_compact_children(rvd);
-       } else {
-               vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
-               vdev_add_child(rvd, vd);
-       }
-       vdev_config_dirty(rvd);
-
-       /*
-        * Reassess the health of our root vdev.
-        */
-       vdev_reopen(rvd);
-}
-
-/*
- * Remove a device from the pool -
- *
- * Removing a device from the vdev namespace requires several steps
- * and can take a significant amount of time.  As a result we use
- * the spa_vdev_config_[enter/exit] functions which allow us to
- * grab and release the spa_config_lock while still holding the namespace
- * lock.  During each step the configuration is synced out.
- *
- * Currently, this supports removing only hot spares, slogs, and level 2 ARC
- * devices.
- */
-int
-spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
-{
-       vdev_t *vd;
-       sysevent_t *ev = NULL;
-       metaslab_group_t *mg;
-       nvlist_t **spares, **l2cache, *nv;
-       uint64_t txg = 0;
-       uint_t nspares, nl2cache;
-       int error = 0;
-       boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
-
-       ASSERT(spa_writeable(spa));
-
-       if (!locked)
-               txg = spa_vdev_enter(spa);
-
-       vd = spa_lookup_by_guid(spa, guid, B_FALSE);
-
-       if (spa->spa_spares.sav_vdevs != NULL &&
-           nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
-           ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
-           (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
-               /*
-                * Only remove the hot spare if it's not currently in use
-                * in this pool.
-                */
-               if (vd == NULL || unspare) {
-                       if (vd == NULL)
-                               vd = spa_lookup_by_guid(spa, guid, B_TRUE);
-                       ev = spa_event_create(spa, vd, NULL,
-                           ESC_ZFS_VDEV_REMOVE_AUX);
-                       spa_vdev_remove_aux(spa->spa_spares.sav_config,
-                           ZPOOL_CONFIG_SPARES, spares, nspares, nv);
-                       spa_load_spares(spa);
-                       spa->spa_spares.sav_sync = B_TRUE;
-               } else {
-                       error = SET_ERROR(EBUSY);
-               }
-       } else if (spa->spa_l2cache.sav_vdevs != NULL &&
-           nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
-           ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
-           (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
-               /*
-                * Cache devices can always be removed.
-                */
-               vd = spa_lookup_by_guid(spa, guid, B_TRUE);
-               ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_AUX);
-               spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
-                   ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
-               spa_load_l2cache(spa);
-               spa->spa_l2cache.sav_sync = B_TRUE;
-       } else if (vd != NULL && vd->vdev_islog) {
-               ASSERT(!locked);
-               ASSERT(vd == vd->vdev_top);
-
-               mg = vd->vdev_mg;
-
-               /*
-                * Stop allocating from this vdev.
-                */
-               metaslab_group_passivate(mg);
-
-               /*
-                * Wait for the youngest allocations and frees to sync,
-                * and then wait for the deferral of those frees to finish.
-                */
-               spa_vdev_config_exit(spa, NULL,
-                   txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
-
-               /*
-                * Attempt to evacuate the vdev.
-                */
-               error = spa_vdev_remove_evacuate(spa, vd);
-
-               txg = spa_vdev_config_enter(spa);
-
-               /*
-                * If we couldn't evacuate the vdev, unwind.
-                */
-               if (error) {
-                       metaslab_group_activate(mg);
-                       return (spa_vdev_exit(spa, NULL, txg, error));
-               }
-
-               /*
-                * Clean up the vdev namespace.
-                */
-               ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_DEV);
-               spa_vdev_remove_from_namespace(spa, vd);
-
-       } else if (vd != NULL) {
-               /*
-                * Normal vdevs cannot be removed (yet).
-                */
-               error = SET_ERROR(ENOTSUP);
-       } else {
-               /*
-                * There is no vdev of any kind with the specified guid.
-                */
-               error = SET_ERROR(ENOENT);
-       }
-
-       if (!locked)
-               error = spa_vdev_exit(spa, NULL, txg, error);
-
-       if (ev)
-               spa_event_post(ev);
-
-       return (error);
-}
-
-/*
- * Find any device that's done replacing, or a vdev marked 'unspare' that's
- * currently spared, so we can detach it.
- */
-static vdev_t *
-spa_vdev_resilver_done_hunt(vdev_t *vd)
-{
-       vdev_t *newvd, *oldvd;
-       int c;
-
-       for (c = 0; c < vd->vdev_children; c++) {
-               oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
-               if (oldvd != NULL)
-                       return (oldvd);
-       }
-
-       /*
-        * Check for a completed replacement.  We always consider the first
-        * vdev in the list to be the oldest vdev, and the last one to be
-        * the newest (see spa_vdev_attach() for how that works).  In
-        * the case where the newest vdev is faulted, we will not automatically
-        * remove it after a resilver completes.  This is OK as it will require
-        * user intervention to determine which disk the admin wishes to keep.
-        */
-       if (vd->vdev_ops == &vdev_replacing_ops) {
-               ASSERT(vd->vdev_children > 1);
-
-               newvd = vd->vdev_child[vd->vdev_children - 1];
-               oldvd = vd->vdev_child[0];
-
-               if (vdev_dtl_empty(newvd, DTL_MISSING) &&
-                   vdev_dtl_empty(newvd, DTL_OUTAGE) &&
-                   !vdev_dtl_required(oldvd))
-                       return (oldvd);
-       }
-
-       /*
-        * Check for a completed resilver with the 'unspare' flag set.
-        */
-       if (vd->vdev_ops == &vdev_spare_ops) {
-               vdev_t *first = vd->vdev_child[0];
-               vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
-
-               if (last->vdev_unspare) {
-                       oldvd = first;
-                       newvd = last;
-               } else if (first->vdev_unspare) {
-                       oldvd = last;
-                       newvd = first;
-               } else {
-                       oldvd = NULL;
-               }
-
-               if (oldvd != NULL &&
-                   vdev_dtl_empty(newvd, DTL_MISSING) &&
-                   vdev_dtl_empty(newvd, DTL_OUTAGE) &&
-                   !vdev_dtl_required(oldvd))
-                       return (oldvd);
-
-               /*
-                * If there are more than two spares attached to a disk,
-                * and those spares are not required, then we want to
-                * attempt to free them up now so that they can be used
-                * by other pools.  Once we're back down to a single
-                * disk+spare, we stop removing them.
-                */
-               if (vd->vdev_children > 2) {
-                       newvd = vd->vdev_child[1];
-
-                       if (newvd->vdev_isspare && last->vdev_isspare &&
-                           vdev_dtl_empty(last, DTL_MISSING) &&
-                           vdev_dtl_empty(last, DTL_OUTAGE) &&
-                           !vdev_dtl_required(newvd))
-                               return (newvd);
-               }
-       }
-
-       return (NULL);
-}
-
-static void
-spa_vdev_resilver_done(spa_t *spa)
-{
-       vdev_t *vd, *pvd, *ppvd;
-       uint64_t guid, sguid, pguid, ppguid;
-
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-
-       while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
-               pvd = vd->vdev_parent;
-               ppvd = pvd->vdev_parent;
-               guid = vd->vdev_guid;
-               pguid = pvd->vdev_guid;
-               ppguid = ppvd->vdev_guid;
-               sguid = 0;
-               /*
-                * If we have just finished replacing a hot spared device, then
-                * we need to detach the parent's first child (the original hot
-                * spare) as well.
-                */
-               if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
-                   ppvd->vdev_children == 2) {
-                       ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
-                       sguid = ppvd->vdev_child[1]->vdev_guid;
-               }
-               ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
-
-               spa_config_exit(spa, SCL_ALL, FTAG);
-               if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
-                       return;
-               if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
-                       return;
-               spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-       }
-
-       spa_config_exit(spa, SCL_ALL, FTAG);
-}
-
-/*
- * Update the stored path or FRU for this vdev.
- */
-int
-spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
-    boolean_t ispath)
-{
-       vdev_t *vd;
-       boolean_t sync = B_FALSE;
-
-       ASSERT(spa_writeable(spa));
-
-       spa_vdev_state_enter(spa, SCL_ALL);
-
-       if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
-               return (spa_vdev_state_exit(spa, NULL, ENOENT));
-
-       if (!vd->vdev_ops->vdev_op_leaf)
-               return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
-
-       if (ispath) {
-               if (strcmp(value, vd->vdev_path) != 0) {
-                       spa_strfree(vd->vdev_path);
-                       vd->vdev_path = spa_strdup(value);
-                       sync = B_TRUE;
-               }
-       } else {
-               if (vd->vdev_fru == NULL) {
-                       vd->vdev_fru = spa_strdup(value);
-                       sync = B_TRUE;
-               } else if (strcmp(value, vd->vdev_fru) != 0) {
-                       spa_strfree(vd->vdev_fru);
-                       vd->vdev_fru = spa_strdup(value);
-                       sync = B_TRUE;
-               }
-       }
-
-       return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
-}
-
-int
-spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
-{
-       return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
-}
-
-int
-spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
-{
-       return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
-}
-
-/*
- * ==========================================================================
- * SPA Scanning
- * ==========================================================================
- */
-int
-spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
-{
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
-
-       if (dsl_scan_resilvering(spa->spa_dsl_pool))
-               return (SET_ERROR(EBUSY));
-
-       return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
-}
-
-int
-spa_scan_stop(spa_t *spa)
-{
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
-       if (dsl_scan_resilvering(spa->spa_dsl_pool))
-               return (SET_ERROR(EBUSY));
-       return (dsl_scan_cancel(spa->spa_dsl_pool));
-}
-
-int
-spa_scan(spa_t *spa, pool_scan_func_t func)
-{
-       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
-
-       if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
-               return (SET_ERROR(ENOTSUP));
-
-       /*
-        * If a resilver was requested, but there is no DTL on a
-        * writeable leaf device, we have nothing to do.
-        */
-       if (func == POOL_SCAN_RESILVER &&
-           !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
-               spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
-               return (0);
-       }
-
-       return (dsl_scan(spa->spa_dsl_pool, func));
-}
-
-/*
- * ==========================================================================
- * SPA async task processing
- * ==========================================================================
- */
-
-static void
-spa_async_remove(spa_t *spa, vdev_t *vd)
-{
-       int c;
-
-       if (vd->vdev_remove_wanted) {
-               vd->vdev_remove_wanted = B_FALSE;
-               vd->vdev_delayed_close = B_FALSE;
-               vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
-
-               /*
-                * We want to clear the stats, but we don't want to do a full
-                * vdev_clear() as that will cause us to throw away
-                * degraded/faulted state as well as attempt to reopen the
-                * device, all of which is a waste.
-                */
-               vd->vdev_stat.vs_read_errors = 0;
-               vd->vdev_stat.vs_write_errors = 0;
-               vd->vdev_stat.vs_checksum_errors = 0;
-
-               vdev_state_dirty(vd->vdev_top);
-       }
-
-       for (c = 0; c < vd->vdev_children; c++)
-               spa_async_remove(spa, vd->vdev_child[c]);
-}
-
-static void
-spa_async_probe(spa_t *spa, vdev_t *vd)
-{
-       int c;
-
-       if (vd->vdev_probe_wanted) {
-               vd->vdev_probe_wanted = B_FALSE;
-               vdev_reopen(vd);        /* vdev_open() does the actual probe */
-       }
-
-       for (c = 0; c < vd->vdev_children; c++)
-               spa_async_probe(spa, vd->vdev_child[c]);
-}
-
-static void
-spa_async_autoexpand(spa_t *spa, vdev_t *vd)
-{
-       int c;
-
-       if (!spa->spa_autoexpand)
-               return;
-
-       for (c = 0; c < vd->vdev_children; c++) {
-               vdev_t *cvd = vd->vdev_child[c];
-               spa_async_autoexpand(spa, cvd);
-       }
-
-       if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
-               return;
-
-       spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND);
-}
-
-static void
-spa_async_thread(spa_t *spa)
-{
-       int tasks, i;
-
-       ASSERT(spa->spa_sync_on);
-
-       mutex_enter(&spa->spa_async_lock);
-       tasks = spa->spa_async_tasks;
-       spa->spa_async_tasks = 0;
-       mutex_exit(&spa->spa_async_lock);
-
-       /*
-        * See if the config needs to be updated.
-        */
-       if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
-               uint64_t old_space, new_space;
-
-               mutex_enter(&spa_namespace_lock);
-               old_space = metaslab_class_get_space(spa_normal_class(spa));
-               spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
-               new_space = metaslab_class_get_space(spa_normal_class(spa));
-               mutex_exit(&spa_namespace_lock);
-
-               /*
-                * If the pool grew as a result of the config update,
-                * then log an internal history event.
-                */
-               if (new_space != old_space) {
-                       spa_history_log_internal(spa, "vdev online", NULL,
-                           "pool '%s' size: %llu(+%llu)",
-                           spa_name(spa), new_space, new_space - old_space);
-               }
-       }
-
-       /*
-        * See if any devices need to be marked REMOVED.
-        */
-       if (tasks & SPA_ASYNC_REMOVE) {
-               spa_vdev_state_enter(spa, SCL_NONE);
-               spa_async_remove(spa, spa->spa_root_vdev);
-               for (i = 0; i < spa->spa_l2cache.sav_count; i++)
-                       spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
-               for (i = 0; i < spa->spa_spares.sav_count; i++)
-                       spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
-               (void) spa_vdev_state_exit(spa, NULL, 0);
-       }
-
-       if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
-               spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
-               spa_async_autoexpand(spa, spa->spa_root_vdev);
-               spa_config_exit(spa, SCL_CONFIG, FTAG);
-       }
-
-       /*
-        * See if any devices need to be probed.
-        */
-       if (tasks & SPA_ASYNC_PROBE) {
-               spa_vdev_state_enter(spa, SCL_NONE);
-               spa_async_probe(spa, spa->spa_root_vdev);
-               (void) spa_vdev_state_exit(spa, NULL, 0);
-       }
-
-       /*
-        * If any devices are done replacing, detach them.
-        */
-       if (tasks & SPA_ASYNC_RESILVER_DONE)
-               spa_vdev_resilver_done(spa);
-
-       /*
-        * Kick off a resilver.
-        */
-       if (tasks & SPA_ASYNC_RESILVER)
-               dsl_resilver_restart(spa->spa_dsl_pool, 0);
-
-       /*
-        * Let the world know that we're done.
-        */
-       mutex_enter(&spa->spa_async_lock);
-       spa->spa_async_thread = NULL;
-       cv_broadcast(&spa->spa_async_cv);
-       mutex_exit(&spa->spa_async_lock);
-       thread_exit();
-}
-
-void
-spa_async_suspend(spa_t *spa)
-{
-       mutex_enter(&spa->spa_async_lock);
-       spa->spa_async_suspended++;
-       while (spa->spa_async_thread != NULL)
-               cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
-       mutex_exit(&spa->spa_async_lock);
-}
-
-void
-spa_async_resume(spa_t *spa)
-{
-       mutex_enter(&spa->spa_async_lock);
-       ASSERT(spa->spa_async_suspended != 0);
-       spa->spa_async_suspended--;
-       mutex_exit(&spa->spa_async_lock);
-}
-
-static boolean_t
-spa_async_tasks_pending(spa_t *spa)
-{
-       uint_t non_config_tasks;
-       uint_t config_task;
-       boolean_t config_task_suspended;
-
-       non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
-       config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
-       if (spa->spa_ccw_fail_time == 0) {
-               config_task_suspended = B_FALSE;
-       } else {
-               config_task_suspended =
-                   (gethrtime() - spa->spa_ccw_fail_time) <
-                   ((hrtime_t)zfs_ccw_retry_interval * NANOSEC);
-       }
-
-       return (non_config_tasks || (config_task && !config_task_suspended));
-}
-
-static void
-spa_async_dispatch(spa_t *spa)
-{
-       mutex_enter(&spa->spa_async_lock);
-       if (spa_async_tasks_pending(spa) &&
-           !spa->spa_async_suspended &&
-           spa->spa_async_thread == NULL &&
-           rootdir != NULL)
-               spa->spa_async_thread = thread_create(NULL, 0,
-                   spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
-       mutex_exit(&spa->spa_async_lock);
-}
-
-void
-spa_async_request(spa_t *spa, int task)
-{
-       zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
-       mutex_enter(&spa->spa_async_lock);
-       spa->spa_async_tasks |= task;
-       mutex_exit(&spa->spa_async_lock);
-}
-
-/*
- * ==========================================================================
- * SPA syncing routines
- * ==========================================================================
- */
-
-static int
-bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
-{
-       bpobj_t *bpo = arg;
-       bpobj_enqueue(bpo, bp, tx);
-       return (0);
-}
-
-static int
-spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
-{
-       zio_t *zio = arg;
-
-       zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
-           zio->io_flags));
-       return (0);
-}
-
-/*
- * Note: this simple function is not inlined to make it easier to dtrace the
- * amount of time spent syncing frees.
- */
-static void
-spa_sync_frees(spa_t *spa, bplist_t *bpl, dmu_tx_t *tx)
-{
-       zio_t *zio = zio_root(spa, NULL, NULL, 0);
-       bplist_iterate(bpl, spa_free_sync_cb, zio, tx);
-       VERIFY(zio_wait(zio) == 0);
-}
-
-/*
- * Note: this simple function is not inlined to make it easier to dtrace the
- * amount of time spent syncing deferred frees.
- */
-static void
-spa_sync_deferred_frees(spa_t *spa, dmu_tx_t *tx)
-{
-       zio_t *zio = zio_root(spa, NULL, NULL, 0);
-       VERIFY3U(bpobj_iterate(&spa->spa_deferred_bpobj,
-           spa_free_sync_cb, zio, tx), ==, 0);
-       VERIFY0(zio_wait(zio));
-}
-
-static void
-spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
-{
-       char *packed = NULL;
-       size_t bufsize;
-       size_t nvsize = 0;
-       dmu_buf_t *db;
-
-       VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
-
-       /*
-        * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
-        * information.  This avoids the dmu_buf_will_dirty() path and
-        * saves us a pre-read to get data we don't actually care about.
-        */
-       bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
-       packed = vmem_alloc(bufsize, KM_SLEEP);
-
-       VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
-           KM_SLEEP) == 0);
-       bzero(packed + nvsize, bufsize - nvsize);
-
-       dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
-
-       vmem_free(packed, bufsize);
-
-       VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
-       dmu_buf_will_dirty(db, tx);
-       *(uint64_t *)db->db_data = nvsize;
-       dmu_buf_rele(db, FTAG);
-}
-
-static void
-spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
-    const char *config, const char *entry)
-{
-       nvlist_t *nvroot;
-       nvlist_t **list;
-       int i;
-
-       if (!sav->sav_sync)
-               return;
-
-       /*
-        * Update the MOS nvlist describing the list of available devices.
-        * spa_validate_aux() will have already made sure this nvlist is
-        * valid and the vdevs are labeled appropriately.
-        */
-       if (sav->sav_object == 0) {
-               sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
-                   DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
-                   sizeof (uint64_t), tx);
-               VERIFY(zap_update(spa->spa_meta_objset,
-                   DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
-                   &sav->sav_object, tx) == 0);
-       }
-
-       VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
-       if (sav->sav_count == 0) {
-               VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
-       } else {
-               list = kmem_alloc(sav->sav_count*sizeof (void *), KM_SLEEP);
-               for (i = 0; i < sav->sav_count; i++)
-                       list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
-                           B_FALSE, VDEV_CONFIG_L2CACHE);
-               VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
-                   sav->sav_count) == 0);
-               for (i = 0; i < sav->sav_count; i++)
-                       nvlist_free(list[i]);
-               kmem_free(list, sav->sav_count * sizeof (void *));
-       }
-
-       spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
-       nvlist_free(nvroot);
-
-       sav->sav_sync = B_FALSE;
-}
-
-/*
- * Rebuild spa's all-vdev ZAP from the vdev ZAPs indicated in each vdev_t.
- * The all-vdev ZAP must be empty.
- */
-static void
-spa_avz_build(vdev_t *vd, uint64_t avz, dmu_tx_t *tx)
-{
-       spa_t *spa = vd->vdev_spa;
-       uint64_t i;
-
-       if (vd->vdev_top_zap != 0) {
-               VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
-                   vd->vdev_top_zap, tx));
-       }
-       if (vd->vdev_leaf_zap != 0) {
-               VERIFY0(zap_add_int(spa->spa_meta_objset, avz,
-                   vd->vdev_leaf_zap, tx));
-       }
-       for (i = 0; i < vd->vdev_children; i++) {
-               spa_avz_build(vd->vdev_child[i], avz, tx);
-       }
-}
-
-static void
-spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
-{
-       nvlist_t *config;
-
-       /*
-        * If the pool is being imported from a pre-per-vdev-ZAP version of ZFS,
-        * its config may not be dirty but we still need to build per-vdev ZAPs.
-        * Similarly, if the pool is being assembled (e.g. after a split), we
-        * need to rebuild the AVZ although the config may not be dirty.
-        */
-       if (list_is_empty(&spa->spa_config_dirty_list) &&
-           spa->spa_avz_action == AVZ_ACTION_NONE)
-               return;
-
-       spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
-
-       ASSERT(spa->spa_avz_action == AVZ_ACTION_NONE ||
-           spa->spa_avz_action == AVZ_ACTION_INITIALIZE ||
-           spa->spa_all_vdev_zaps != 0);
-
-       if (spa->spa_avz_action == AVZ_ACTION_REBUILD) {
-               zap_cursor_t zc;
-               zap_attribute_t za;
-
-               /* Make and build the new AVZ */
-               uint64_t new_avz = zap_create(spa->spa_meta_objset,
-                   DMU_OTN_ZAP_METADATA, DMU_OT_NONE, 0, tx);
-               spa_avz_build(spa->spa_root_vdev, new_avz, tx);
-
-               /* Diff old AVZ with new one */
-               for (zap_cursor_init(&zc, spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps);
-                   zap_cursor_retrieve(&zc, &za) == 0;
-                   zap_cursor_advance(&zc)) {
-                       uint64_t vdzap = za.za_first_integer;
-                       if (zap_lookup_int(spa->spa_meta_objset, new_avz,
-                           vdzap) == ENOENT) {
-                               /*
-                                * ZAP is listed in old AVZ but not in new one;
-                                * destroy it
-                                */
-                               VERIFY0(zap_destroy(spa->spa_meta_objset, vdzap,
-                                   tx));
-                       }
-               }
-
-               zap_cursor_fini(&zc);
-
-               /* Destroy the old AVZ */
-               VERIFY0(zap_destroy(spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps, tx));
-
-               /* Replace the old AVZ in the dir obj with the new one */
-               VERIFY0(zap_update(spa->spa_meta_objset,
-                   DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP,
-                   sizeof (new_avz), 1, &new_avz, tx));
-
-               spa->spa_all_vdev_zaps = new_avz;
-       } else if (spa->spa_avz_action == AVZ_ACTION_DESTROY) {
-               zap_cursor_t zc;
-               zap_attribute_t za;
-
-               /* Walk through the AVZ and destroy all listed ZAPs */
-               for (zap_cursor_init(&zc, spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps);
-                   zap_cursor_retrieve(&zc, &za) == 0;
-                   zap_cursor_advance(&zc)) {
-                       uint64_t zap = za.za_first_integer;
-                       VERIFY0(zap_destroy(spa->spa_meta_objset, zap, tx));
-               }
-
-               zap_cursor_fini(&zc);
-
-               /* Destroy and unlink the AVZ itself */
-               VERIFY0(zap_destroy(spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps, tx));
-               VERIFY0(zap_remove(spa->spa_meta_objset,
-                   DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_VDEV_ZAP_MAP, tx));
-               spa->spa_all_vdev_zaps = 0;
-       }
-
-       if (spa->spa_all_vdev_zaps == 0) {
-               spa->spa_all_vdev_zaps = zap_create_link(spa->spa_meta_objset,
-                   DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
-                   DMU_POOL_VDEV_ZAP_MAP, tx);
-       }
-       spa->spa_avz_action = AVZ_ACTION_NONE;
-
-       /* Create ZAPs for vdevs that don't have them. */
-       vdev_construct_zaps(spa->spa_root_vdev, tx);
-
-       config = spa_config_generate(spa, spa->spa_root_vdev,
-           dmu_tx_get_txg(tx), B_FALSE);
-
-       /*
-        * If we're upgrading the spa version then make sure that
-        * the config object gets updated with the correct version.
-        */
-       if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
-               fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
-                   spa->spa_uberblock.ub_version);
-
-       spa_config_exit(spa, SCL_STATE, FTAG);
-
-       nvlist_free(spa->spa_config_syncing);
-       spa->spa_config_syncing = config;
-
-       spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
-}
-
-static void
-spa_sync_version(void *arg, dmu_tx_t *tx)
-{
-       uint64_t *versionp = arg;
-       uint64_t version = *versionp;
-       spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-
-       /*
-        * Setting the version is special cased when first creating the pool.
-        */
-       ASSERT(tx->tx_txg != TXG_INITIAL);
-
-       ASSERT(SPA_VERSION_IS_SUPPORTED(version));
-       ASSERT(version >= spa_version(spa));
-
-       spa->spa_uberblock.ub_version = version;
-       vdev_config_dirty(spa->spa_root_vdev);
-       spa_history_log_internal(spa, "set", tx, "version=%lld", version);
-}
-
-/*
- * Set zpool properties.
- */
-static void
-spa_sync_props(void *arg, dmu_tx_t *tx)
-{
-       nvlist_t *nvp = arg;
-       spa_t *spa = dmu_tx_pool(tx)->dp_spa;
-       objset_t *mos = spa->spa_meta_objset;
-       nvpair_t *elem = NULL;
-
-       mutex_enter(&spa->spa_props_lock);
-
-       while ((elem = nvlist_next_nvpair(nvp, elem))) {
-               uint64_t intval;
-               char *strval, *fname;
-               zpool_prop_t prop;
-               const char *propname;
-               zprop_type_t proptype;
-               spa_feature_t fid;
-
-               prop = zpool_name_to_prop(nvpair_name(elem));
-               switch ((int)prop) {
-               case ZPROP_INVAL:
-                       /*
-                        * We checked this earlier in spa_prop_validate().
-                        */
-                       ASSERT(zpool_prop_feature(nvpair_name(elem)));
-
-                       fname = strchr(nvpair_name(elem), '@') + 1;
-                       VERIFY0(zfeature_lookup_name(fname, &fid));
-
-                       spa_feature_enable(spa, fid, tx);
-                       spa_history_log_internal(spa, "set", tx,
-                           "%s=enabled", nvpair_name(elem));
-                       break;
-
-               case ZPOOL_PROP_VERSION:
-                       intval = fnvpair_value_uint64(elem);
-                       /*
-                        * The version is synced separately before other
-                        * properties and should be correct by now.
-                        */
-                       ASSERT3U(spa_version(spa), >=, intval);
-                       break;
-
-               case ZPOOL_PROP_ALTROOT:
-                       /*
-                        * 'altroot' is a non-persistent property. It should
-                        * have been set temporarily at creation or import time.
-                        */
-                       ASSERT(spa->spa_root != NULL);
-                       break;
-
-               case ZPOOL_PROP_READONLY:
-               case ZPOOL_PROP_CACHEFILE:
-                       /*
-                        * 'readonly' and 'cachefile' are also non-persisitent
-                        * properties.
-                        */
-                       break;
-               case ZPOOL_PROP_COMMENT:
-                       strval = fnvpair_value_string(elem);
-                       if (spa->spa_comment != NULL)
-                               spa_strfree(spa->spa_comment);
-                       spa->spa_comment = spa_strdup(strval);
-                       /*
-                        * We need to dirty the configuration on all the vdevs
-                        * so that their labels get updated.  It's unnecessary
-                        * to do this for pool creation since the vdev's
-                        * configuration has already been dirtied.
-                        */
-                       if (tx->tx_txg != TXG_INITIAL)
-                               vdev_config_dirty(spa->spa_root_vdev);
-                       spa_history_log_internal(spa, "set", tx,
-                           "%s=%s", nvpair_name(elem), strval);
-                       break;
-               default:
-                       /*
-                        * Set pool property values in the poolprops mos object.
-                        */
-                       if (spa->spa_pool_props_object == 0) {
-                               spa->spa_pool_props_object =
-                                   zap_create_link(mos, DMU_OT_POOL_PROPS,
-                                   DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
-                                   tx);
-                       }
-
-                       /* normalize the property name */
-                       propname = zpool_prop_to_name(prop);
-                       proptype = zpool_prop_get_type(prop);
-
-                       if (nvpair_type(elem) == DATA_TYPE_STRING) {
-                               ASSERT(proptype == PROP_TYPE_STRING);
-                               strval = fnvpair_value_string(elem);
-                               VERIFY0(zap_update(mos,
-                                   spa->spa_pool_props_object, propname,
-                                   1, strlen(strval) + 1, strval, tx));
-                               spa_history_log_internal(spa, "set", tx,
-                                   "%s=%s", nvpair_name(elem), strval);
-                       } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
-                               intval = fnvpair_value_uint64(elem);
-
-                               if (proptype == PROP_TYPE_INDEX) {
-                                       const char *unused;
-                                       VERIFY0(zpool_prop_index_to_string(
-                                           prop, intval, &unused));
-                               }
-                               VERIFY0(zap_update(mos,
-                                   spa->spa_pool_props_object, propname,
-                                   8, 1, &intval, tx));
-                               spa_history_log_internal(spa, "set", tx,
-                                   "%s=%lld", nvpair_name(elem), intval);
-                       } else {
-                               ASSERT(0); /* not allowed */
-                       }
-
-                       switch (prop) {
-                       case ZPOOL_PROP_DELEGATION:
-                               spa->spa_delegation = intval;
-                               break;
-                       case ZPOOL_PROP_BOOTFS:
-                               spa->spa_bootfs = intval;
-                               break;
-                       case ZPOOL_PROP_FAILUREMODE:
-                               spa->spa_failmode = intval;
-                               break;
-                       case ZPOOL_PROP_AUTOEXPAND:
-                               spa->spa_autoexpand = intval;
-                               if (tx->tx_txg != TXG_INITIAL)
-                                       spa_async_request(spa,
-                                           SPA_ASYNC_AUTOEXPAND);
-                               break;
-                       case ZPOOL_PROP_MULTIHOST:
-                               spa->spa_multihost = intval;
-                               break;
-                       case ZPOOL_PROP_DEDUPDITTO:
-                               spa->spa_dedup_ditto = intval;
-                               break;
-                       default:
-                               break;
-                       }
-               }
-
-       }
-
-       mutex_exit(&spa->spa_props_lock);
-}
-
-/*
- * Perform one-time upgrade on-disk changes.  spa_version() does not
- * reflect the new version this txg, so there must be no changes this
- * txg to anything that the upgrade code depends on after it executes.
- * Therefore this must be called after dsl_pool_sync() does the sync
- * tasks.
- */
-static void
-spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
-{
-       dsl_pool_t *dp = spa->spa_dsl_pool;
-
-       ASSERT(spa->spa_sync_pass == 1);
-
-       rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
-
-       if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
-           spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
-               dsl_pool_create_origin(dp, tx);
-
-               /* Keeping the origin open increases spa_minref */
-               spa->spa_minref += 3;
-       }
-
-       if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
-           spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
-               dsl_pool_upgrade_clones(dp, tx);
-       }
-
-       if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
-           spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
-               dsl_pool_upgrade_dir_clones(dp, tx);
-
-               /* Keeping the freedir open increases spa_minref */
-               spa->spa_minref += 3;
-       }
-
-       if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
-           spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
-               spa_feature_create_zap_objects(spa, tx);
-       }
-
-       /*
-        * LZ4_COMPRESS feature's behaviour was changed to activate_on_enable
-        * when possibility to use lz4 compression for metadata was added
-        * Old pools that have this feature enabled must be upgraded to have
-        * this feature active
-        */
-       if (spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
-               boolean_t lz4_en = spa_feature_is_enabled(spa,
-                   SPA_FEATURE_LZ4_COMPRESS);
-               boolean_t lz4_ac = spa_feature_is_active(spa,
-                   SPA_FEATURE_LZ4_COMPRESS);
-
-               if (lz4_en && !lz4_ac)
-                       spa_feature_incr(spa, SPA_FEATURE_LZ4_COMPRESS, tx);
-       }
-
-       /*
-        * If we haven't written the salt, do so now.  Note that the
-        * feature may not be activated yet, but that's fine since
-        * the presence of this ZAP entry is backwards compatible.
-        */
-       if (zap_contains(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
-           DMU_POOL_CHECKSUM_SALT) == ENOENT) {
-               VERIFY0(zap_add(spa->spa_meta_objset,
-                   DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CHECKSUM_SALT, 1,
-                   sizeof (spa->spa_cksum_salt.zcs_bytes),
-                   spa->spa_cksum_salt.zcs_bytes, tx));
-       }
-
-       rrw_exit(&dp->dp_config_rwlock, FTAG);
-}
-
-/*
- * Sync the specified transaction group.  New blocks may be dirtied as
- * part of the process, so we iterate until it converges.
- */
-void
-spa_sync(spa_t *spa, uint64_t txg)
-{
-       dsl_pool_t *dp = spa->spa_dsl_pool;
-       objset_t *mos = spa->spa_meta_objset;
-       bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
-       metaslab_class_t *mc;
-       vdev_t *rvd = spa->spa_root_vdev;
-       vdev_t *vd;
-       dmu_tx_t *tx;
-       int error;
-       uint32_t max_queue_depth = zfs_vdev_async_write_max_active *
-           zfs_vdev_queue_depth_pct / 100;
-       uint64_t queue_depth_total;
-       int c;
-
-       VERIFY(spa_writeable(spa));
-
-       /*
-        * Lock out configuration changes.
-        */
-       spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
-
-       spa->spa_syncing_txg = txg;
-       spa->spa_sync_pass = 0;
-
-       mutex_enter(&spa->spa_alloc_lock);
-       VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
-       mutex_exit(&spa->spa_alloc_lock);
-
-       /*
-        * If there are any pending vdev state changes, convert them
-        * into config changes that go out with this transaction group.
-        */
-       spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
-       while (list_head(&spa->spa_state_dirty_list) != NULL) {
-               /*
-                * We need the write lock here because, for aux vdevs,
-                * calling vdev_config_dirty() modifies sav_config.
-                * This is ugly and will become unnecessary when we
-                * eliminate the aux vdev wart by integrating all vdevs
-                * into the root vdev tree.
-                */
-               spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
-               spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
-               while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
-                       vdev_state_clean(vd);
-                       vdev_config_dirty(vd);
-               }
-               spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
-               spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
-       }
-       spa_config_exit(spa, SCL_STATE, FTAG);
-
-       tx = dmu_tx_create_assigned(dp, txg);
-
-       spa->spa_sync_starttime = gethrtime();
-       taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
-       spa->spa_deadman_tqid = taskq_dispatch_delay(system_delay_taskq,
-           spa_deadman, spa, TQ_SLEEP, ddi_get_lbolt() +
-           NSEC_TO_TICK(spa->spa_deadman_synctime));
-
-       /*
-        * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
-        * set spa_deflate if we have no raid-z vdevs.
-        */
-       if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
-           spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
-               int i;
-
-               for (i = 0; i < rvd->vdev_children; i++) {
-                       vd = rvd->vdev_child[i];
-                       if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
-                               break;
-               }
-               if (i == rvd->vdev_children) {
-                       spa->spa_deflate = TRUE;
-                       VERIFY(0 == zap_add(spa->spa_meta_objset,
-                           DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
-                           sizeof (uint64_t), 1, &spa->spa_deflate, tx));
-               }
-       }
-
-       /*
-        * Set the top-level vdev's max queue depth. Evaluate each
-        * top-level's async write queue depth in case it changed.
-        * The max queue depth will not change in the middle of syncing
-        * out this txg.
-        */
-       queue_depth_total = 0;
-       for (c = 0; c < rvd->vdev_children; c++) {
-               vdev_t *tvd = rvd->vdev_child[c];
-               metaslab_group_t *mg = tvd->vdev_mg;
-
-               if (mg == NULL || mg->mg_class != spa_normal_class(spa) ||
-                   !metaslab_group_initialized(mg))
-                       continue;
-
-               /*
-                * It is safe to do a lock-free check here because only async
-                * allocations look at mg_max_alloc_queue_depth, and async
-                * allocations all happen from spa_sync().
-                */
-               ASSERT0(refcount_count(&mg->mg_alloc_queue_depth));
-               mg->mg_max_alloc_queue_depth = max_queue_depth;
-               queue_depth_total += mg->mg_max_alloc_queue_depth;
-       }
-       mc = spa_normal_class(spa);
-       ASSERT0(refcount_count(&mc->mc_alloc_slots));
-       mc->mc_alloc_max_slots = queue_depth_total;
-       mc->mc_alloc_throttle_enabled = zio_dva_throttle_enabled;
-
-       ASSERT3U(mc->mc_alloc_max_slots, <=,
-           max_queue_depth * rvd->vdev_children);
-
-       /*
-        * Iterate to convergence.
-        */
-       do {
-               int pass = ++spa->spa_sync_pass;
-
-               spa_sync_config_object(spa, tx);
-               spa_sync_aux_dev(spa, &spa->spa_spares, tx,
-                   ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
-               spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
-                   ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
-               spa_errlog_sync(spa, txg);
-               dsl_pool_sync(dp, txg);
-
-               if (pass < zfs_sync_pass_deferred_free) {
-                       spa_sync_frees(spa, free_bpl, tx);
-               } else {
-                       /*
-                        * We can not defer frees in pass 1, because
-                        * we sync the deferred frees later in pass 1.
-                        */
-                       ASSERT3U(pass, >, 1);
-                       bplist_iterate(free_bpl, bpobj_enqueue_cb,
-                           &spa->spa_deferred_bpobj, tx);
-               }
-
-               ddt_sync(spa, txg);
-               dsl_scan_sync(dp, tx);
-
-               while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, txg)))
-                       vdev_sync(vd, txg);
-
-               if (pass == 1) {
-                       spa_sync_upgrades(spa, tx);
-                       ASSERT3U(txg, >=,
-                           spa->spa_uberblock.ub_rootbp.blk_birth);
-                       /*
-                        * Note: We need to check if the MOS is dirty
-                        * because we could have marked the MOS dirty
-                        * without updating the uberblock (e.g. if we
-                        * have sync tasks but no dirty user data).  We
-                        * need to check the uberblock's rootbp because
-                        * it is updated if we have synced out dirty
-                        * data (though in this case the MOS will most
-                        * likely also be dirty due to second order
-                        * effects, we don't want to rely on that here).
-                        */
-                       if (spa->spa_uberblock.ub_rootbp.blk_birth < txg &&
-                           !dmu_objset_is_dirty(mos, txg)) {
-                               /*
-                                * Nothing changed on the first pass,
-                                * therefore this TXG is a no-op.  Avoid
-                                * syncing deferred frees, so that we
-                                * can keep this TXG as a no-op.
-                                */
-                               ASSERT(txg_list_empty(&dp->dp_dirty_datasets,
-                                   txg));
-                               ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
-                               ASSERT(txg_list_empty(&dp->dp_sync_tasks, txg));
-                               break;
-                       }
-                       spa_sync_deferred_frees(spa, tx);
-               }
-
-       } while (dmu_objset_is_dirty(mos, txg));
-
-#ifdef ZFS_DEBUG
-       if (!list_is_empty(&spa->spa_config_dirty_list)) {
-               /*
-                * Make sure that the number of ZAPs for all the vdevs matches
-                * the number of ZAPs in the per-vdev ZAP list. This only gets
-                * called if the config is dirty; otherwise there may be
-                * outstanding AVZ operations that weren't completed in
-                * spa_sync_config_object.
-                */
-               uint64_t all_vdev_zap_entry_count;
-               ASSERT0(zap_count(spa->spa_meta_objset,
-                   spa->spa_all_vdev_zaps, &all_vdev_zap_entry_count));
-               ASSERT3U(vdev_count_verify_zaps(spa->spa_root_vdev), ==,
-                   all_vdev_zap_entry_count);
-       }
-#endif
-
-       /*
-        * Rewrite the vdev configuration (which includes the uberblock)
-        * to commit the transaction group.
-        *
-        * If there are no dirty vdevs, we sync the uberblock to a few
-        * random top-level vdevs that are known to be visible in the
-        * config cache (see spa_vdev_add() for a complete description).
-        * If there *are* dirty vdevs, sync the uberblock to all vdevs.
-        */
-       for (;;) {
-               /*
-                * We hold SCL_STATE to prevent vdev open/close/etc.
-                * while we're attempting to write the vdev labels.
-                */
-               spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
-
-               if (list_is_empty(&spa->spa_config_dirty_list)) {
-                       vdev_t *svd[SPA_DVAS_PER_BP];
-                       int svdcount = 0;
-                       int children = rvd->vdev_children;
-                       int c0 = spa_get_random(children);
-
-                       for (c = 0; c < children; c++) {
-                               vd = rvd->vdev_child[(c0 + c) % children];
-                               if (vd->vdev_ms_array == 0 || vd->vdev_islog)
-                                       continue;
-                               svd[svdcount++] = vd;
-                               if (svdcount == SPA_DVAS_PER_BP)
-                                       break;
-                       }
-                       error = vdev_config_sync(svd, svdcount, txg);
-               } else {
-                       error = vdev_config_sync(rvd->vdev_child,
-                           rvd->vdev_children, txg);
-               }
-
-               if (error == 0)
-                       spa->spa_last_synced_guid = rvd->vdev_guid;
-
-               spa_config_exit(spa, SCL_STATE, FTAG);
-
-               if (error == 0)
-                       break;
-               zio_suspend(spa, NULL);
-               zio_resume_wait(spa);
-       }
-       dmu_tx_commit(tx);
-
-       taskq_cancel_id(system_delay_taskq, spa->spa_deadman_tqid);
-       spa->spa_deadman_tqid = 0;
-
-       /*
-        * Clear the dirty config list.
-        */
-       while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
-               vdev_config_clean(vd);
-
-       /*
-        * Now that the new config has synced transactionally,
-        * let it become visible to the config cache.
-        */
-       if (spa->spa_config_syncing != NULL) {
-               spa_config_set(spa, spa->spa_config_syncing);
-               spa->spa_config_txg = txg;
-               spa->spa_config_syncing = NULL;
-       }
-
-       dsl_pool_sync_done(dp, txg);
-
-       mutex_enter(&spa->spa_alloc_lock);
-       VERIFY0(avl_numnodes(&spa->spa_alloc_tree));
-       mutex_exit(&spa->spa_alloc_lock);
-
-       /*
-        * Update usable space statistics.
-        */
-       while ((vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg))))
-               vdev_sync_done(vd, txg);
-
-       spa_update_dspace(spa);
-
-       /*
-        * It had better be the case that we didn't dirty anything
-        * since vdev_config_sync().
-        */
-       ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
-       ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
-       ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
-
-       spa->spa_sync_pass = 0;
-
-       /*
-        * Update the last synced uberblock here. We want to do this at
-        * the end of spa_sync() so that consumers of spa_last_synced_txg()
-        * will be guaranteed that all the processing associated with
-        * that txg has been completed.
-        */
-       spa->spa_ubsync = spa->spa_uberblock;
-       spa_config_exit(spa, SCL_CONFIG, FTAG);
-
-       spa_handle_ignored_writes(spa);
-
-       /*
-        * If any async tasks have been requested, kick them off.
-        */
-       spa_async_dispatch(spa);
-}
-
-/*
- * Sync all pools.  We don't want to hold the namespace lock across these
- * operations, so we take a reference on the spa_t and drop the lock during the
- * sync.
- */
-void
-spa_sync_allpools(void)
-{
-       spa_t *spa = NULL;
-       mutex_enter(&spa_namespace_lock);
-       while ((spa = spa_next(spa)) != NULL) {
-               if (spa_state(spa) != POOL_STATE_ACTIVE ||
-                   !spa_writeable(spa) || spa_suspended(spa))
-                       continue;
-               spa_open_ref(spa, FTAG);
-               mutex_exit(&spa_namespace_lock);
-               txg_wait_synced(spa_get_dsl(spa), 0);
-               mutex_enter(&spa_namespace_lock);
-               spa_close(spa, FTAG);
-       }
-       mutex_exit(&spa_namespace_lock);
-}
-
-/*
- * ==========================================================================
- * Miscellaneous routines
- * ==========================================================================
- */
-
-/*
- * Remove all pools in the system.
- */
-void
-spa_evict_all(void)
-{
-       spa_t *spa;
-
-       /*
-        * Remove all cached state.  All pools should be closed now,
-        * so every spa in the AVL tree should be unreferenced.
-        */
-       mutex_enter(&spa_namespace_lock);
-       while ((spa = spa_next(NULL)) != NULL) {
-               /*
-                * Stop async tasks.  The async thread may need to detach
-                * a device that's been replaced, which requires grabbing
-                * spa_namespace_lock, so we must drop it here.
-                */
-               spa_open_ref(spa, FTAG);
-               mutex_exit(&spa_namespace_lock);
-               spa_async_suspend(spa);
-               mutex_enter(&spa_namespace_lock);
-               spa_close(spa, FTAG);
-
-               if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
-                       spa_unload(spa);
-                       spa_deactivate(spa);
-               }
-               spa_remove(spa);
-       }
-       mutex_exit(&spa_namespace_lock);
-}
-
-vdev_t *
-spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
-{
-       vdev_t *vd;
-       int i;
-
-       if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
-               return (vd);
-
-       if (aux) {
-               for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
-                       vd = spa->spa_l2cache.sav_vdevs[i];
-                       if (vd->vdev_guid == guid)
-                               return (vd);
-               }
-
-               for (i = 0; i < spa->spa_spares.sav_count; i++) {
-                       vd = spa->spa_spares.sav_vdevs[i];
-                       if (vd->vdev_guid == guid)
-                               return (vd);
-               }
-       }
-
-       return (NULL);
-}
-
-void
-spa_upgrade(spa_t *spa, uint64_t version)
-{
-       ASSERT(spa_writeable(spa));
-
-       spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
-
-       /*
-        * This should only be called for a non-faulted pool, and since a
-        * future version would result in an unopenable pool, this shouldn't be
-        * possible.
-        */
-       ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
-       ASSERT3U(version, >=, spa->spa_uberblock.ub_version);
-
-       spa->spa_uberblock.ub_version = version;
-       vdev_config_dirty(spa->spa_root_vdev);
-
-       spa_config_exit(spa, SCL_ALL, FTAG);
-
-       txg_wait_synced(spa_get_dsl(spa), 0);
-}
-
-boolean_t
-spa_has_spare(spa_t *spa, uint64_t guid)
-{
-       int i;
-       uint64_t spareguid;
-       spa_aux_vdev_t *sav = &spa->spa_spares;
-
-       for (i = 0; i < sav->sav_count; i++)
-               if (sav->sav_vdevs[i]->vdev_guid == guid)
-                       return (B_TRUE);
-
-       for (i = 0; i < sav->sav_npending; i++) {
-               if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
-                   &spareguid) == 0 && spareguid == guid)
-                       return (B_TRUE);
-       }
-
-       return (B_FALSE);
-}
-
-/*
- * Check if a pool has an active shared spare device.
- * Note: reference count of an active spare is 2, as a spare and as a replace
- */
-static boolean_t
-spa_has_active_shared_spare(spa_t *spa)
-{
-       int i, refcnt;
-       uint64_t pool;
-       spa_aux_vdev_t *sav = &spa->spa_spares;
-
-       for (i = 0; i < sav->sav_count; i++) {
-               if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
-                   &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
-                   refcnt > 2)
-                       return (B_TRUE);
-       }
-
-       return (B_FALSE);
-}
-
-static sysevent_t *
-spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
-{
-       sysevent_t *ev = NULL;
-#ifdef _KERNEL
-       nvlist_t *resource;
-
-       resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl);
-       if (resource) {
-               ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP);
-               ev->resource = resource;
-       }
-#endif
-       return (ev);
-}
-
-static void
-spa_event_post(sysevent_t *ev)
-{
-#ifdef _KERNEL
-       if (ev) {
-               zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb);
-               kmem_free(ev, sizeof (*ev));
-       }
-#endif
-}
-
-/*
- * Post a zevent corresponding to the given sysevent.   The 'name' must be one
- * of the event definitions in sys/sysevent/eventdefs.h.  The payload will be
- * filled in from the spa and (optionally) the vdev.  This doesn't do anything
- * in the userland libzpool, as we don't want consumers to misinterpret ztest
- * or zdb as real changes.
- */
-void
-spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
-{
-       spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
-}
-
-#if defined(_KERNEL) && defined(HAVE_SPL)
-/* state manipulation functions */
-EXPORT_SYMBOL(spa_open);
-EXPORT_SYMBOL(spa_open_rewind);
-EXPORT_SYMBOL(spa_get_stats);
-EXPORT_SYMBOL(spa_create);
-EXPORT_SYMBOL(spa_import);
-EXPORT_SYMBOL(spa_tryimport);
-EXPORT_SYMBOL(spa_destroy);
-EXPORT_SYMBOL(spa_export);
-EXPORT_SYMBOL(spa_reset);
-EXPORT_SYMBOL(spa_async_request);
-EXPORT_SYMBOL(spa_async_suspend);
-EXPORT_SYMBOL(spa_async_resume);
-EXPORT_SYMBOL(spa_inject_addref);
-EXPORT_SYMBOL(spa_inject_delref);
-EXPORT_SYMBOL(spa_scan_stat_init);
-EXPORT_SYMBOL(spa_scan_get_stats);
-
-/* device maniion */
-EXPORT_SYMBOL(spa_vdev_add);
-EXPORT_SYMBOL(spa_vdev_attach);
-EXPORT_SYMBOL(spa_vdev_detach);
-EXPORT_SYMBOL(spa_vdev_remove);
-EXPORT_SYMBOL(spa_vdev_setpath);
-EXPORT_SYMBOL(spa_vdev_setfru);
-EXPORT_SYMBOL(spa_vdev_split_mirror);
-
-/* spare statech is global across all pools) */
-EXPORT_SYMBOL(spa_spare_add);
-EXPORT_SYMBOL(spa_spare_remove);
-EXPORT_SYMBOL(spa_spare_exists);
-EXPORT_SYMBOL(spa_spare_activate);
-
-/* L2ARC statech is global across all pools) */
-EXPORT_SYMBOL(spa_l2cache_add);
-EXPORT_SYMBOL(spa_l2cache_remove);
-EXPORT_SYMBOL(spa_l2cache_exists);
-EXPORT_SYMBOL(spa_l2cache_activate);
-EXPORT_SYMBOL(spa_l2cache_drop);
-
-/* scanning */
-EXPORT_SYMBOL(spa_scan);
-EXPORT_SYMBOL(spa_scan_stop);
-
-/* spa syncing */
-EXPORT_SYMBOL(spa_sync); /* only for DMU use */
-EXPORT_SYMBOL(spa_sync_allpools);
-
-/* properties */
-EXPORT_SYMBOL(spa_prop_set);
-EXPORT_SYMBOL(spa_prop_get);
-EXPORT_SYMBOL(spa_prop_clear_bootfs);
-
-/* asynchronous event notification */
-EXPORT_SYMBOL(spa_event_notify);
-#endif
-
-#if defined(_KERNEL) && defined(HAVE_SPL)
-module_param(spa_load_verify_maxinflight, int, 0644);
-MODULE_PARM_DESC(spa_load_verify_maxinflight,
-       "Max concurrent traversal I/Os while verifying pool during import -X");
-
-module_param(spa_load_verify_metadata, int, 0644);
-MODULE_PARM_DESC(spa_load_verify_metadata,
-       "Set to traverse metadata on pool import");
-
-module_param(spa_load_verify_data, int, 0644);
-MODULE_PARM_DESC(spa_load_verify_data,
-       "Set to traverse data on pool import");
-
-/* CSTYLED */
-module_param(zio_taskq_batch_pct, uint, 0444);
-MODULE_PARM_DESC(zio_taskq_batch_pct,
-       "Percentage of CPUs to run an IO worker thread");
-
-#endif