int
zfs_set_version(zfs_sb_t *zsb, uint64_t newvers)
{
	int error;
	objset_t *os = zsb->z_os;
	dmu_tx_t *tx;

	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
		return (SET_ERROR(EINVAL));

	if (newvers < zsb->z_version)
		return (SET_ERROR(EINVAL));

	if (zfs_spa_version_map(newvers) >
	    spa_version(dmu_objset_spa(zsb->z_os)))
		return (SET_ERROR(ENOTSUP));

	tx = dmu_tx_create(os);
	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
	if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) {
		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
		    ZFS_SA_ATTRS);
		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
	}
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		dmu_tx_abort(tx);
		return (error);
	}

	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
	    8, 1, &newvers, tx);

	if (error) {
		dmu_tx_commit(tx);
		return (error);
	}

	if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) {
		uint64_t sa_obj;

		ASSERT3U(spa_version(dmu_objset_spa(zsb->z_os)), >=,
		    SPA_VERSION_SA);
		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
		    DMU_OT_NONE, 0, tx);

		error = zap_add(os, MASTER_NODE_OBJ,
		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
		ASSERT0(error);

		VERIFY(0 == sa_set_sa_object(os, sa_obj));
		sa_register_update_callback(os, zfs_sa_upgrade);
	}

void
bpobj_decr_empty(objset_t *os, dmu_tx_t *tx)
{
	dsl_pool_t *dp = dmu_objset_pool(os);

	spa_feature_decr(dmu_objset_spa(os), SPA_FEATURE_EMPTY_BPOBJ, tx);
	if (!spa_feature_is_active(dmu_objset_spa(os),
	    SPA_FEATURE_EMPTY_BPOBJ)) {
		VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
		    DMU_POOL_DIRECTORY_OBJECT,
		    DMU_POOL_EMPTY_BPOBJ, tx));
		VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
		dp->dp_empty_bpobj = 0;
	}

/*
 * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
 */
uint64_t
bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx)
{
	zfeature_info_t *empty_bpobj_feat =
	    &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ];
	spa_t *spa = dmu_objset_spa(os);
	dsl_pool_t *dp = dmu_objset_pool(os);

	if (spa_feature_is_enabled(spa, empty_bpobj_feat)) {
		if (!spa_feature_is_active(spa, empty_bpobj_feat)) {
			ASSERT0(dp->dp_empty_bpobj);
			dp->dp_empty_bpobj =
			    bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx);
			VERIFY(zap_add(os,
			    DMU_POOL_DIRECTORY_OBJECT,
			    DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
			    &dp->dp_empty_bpobj, tx) == 0);
		}
		spa_feature_incr(spa, empty_bpobj_feat, tx);
		ASSERT(dp->dp_empty_bpobj != 0);
		return (dp->dp_empty_bpobj);
	} else {
		return (bpobj_alloc(os, blocksize, tx));
	}
}

static int osd_attr_get(const struct lu_env *env,
			struct dt_object *dt,
			struct lu_attr *attr)
{
	struct osd_object	*obj = osd_dt_obj(dt);
	uint64_t		 blocks;
	uint32_t		 blksize;

	LASSERT(dt_object_exists(dt));
	LASSERT(osd_invariant(obj));
	LASSERT(obj->oo_db);

	read_lock(&obj->oo_attr_lock);
	*attr = obj->oo_attr;
	read_unlock(&obj->oo_attr_lock);

	/* with ZFS_DEBUG zrl_add_debug() called by DB_DNODE_ENTER()
	 * from within sa_object_size() can block on a mutex, so
	 * we can't call sa_object_size() holding rwlock */
	sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
	/* we do not control size of indices, so always calculate
	 * it from number of blocks reported by DMU */
	if (S_ISDIR(attr->la_mode))
		attr->la_size = 512 * blocks;
	/* Block size may be not set; suggest maximal I/O transfers. */
	if (blksize == 0)
		blksize = osd_spa_maxblocksize(
			dmu_objset_spa(osd_obj2dev(obj)->od_os));

	attr->la_blksize = blksize;
	attr->la_blocks = blocks;
	attr->la_valid |= LA_BLOCKS | LA_BLKSIZE;

	return 0;
}

/*
 * Return (in *dsizep) the amount of space on the deadlist which is:
 * mintxg < blk_birth <= maxtxg
 */
int
bplist_space_birthrange(bplist_t *bpl, uint64_t mintxg, uint64_t maxtxg,
    uint64_t *dsizep)
{
	uint64_t size = 0;
	uint64_t itor = 0;
	blkptr_t bp;
	int err;

	/*
	 * As an optimization, if they want the whole txg range, just
	 * get bpl_bytes rather than iterating over the bps.
	 */
	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX) {
		mutex_enter(&bpl->bpl_lock);
		err = bplist_hold(bpl);
		if (err == 0)
			*dsizep = bpl->bpl_phys->bpl_bytes;
		mutex_exit(&bpl->bpl_lock);
		return (err);
	}

	while ((err = bplist_iterate(bpl, &itor, &bp)) == 0) {
		if (bp.blk_birth > mintxg && bp.blk_birth <= maxtxg) {
			size += bp_get_dsize(dmu_objset_spa(bpl->bpl_mos), &bp);
		}
	}
	if (err == ENOENT)
		err = 0;
	*dsizep = size;
	return (err);
}

zilog_t *
zil_alloc(objset_t *os, zil_header_t *zh_phys)
{
	zilog_t *zilog;

	zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP);

	zilog->zl_header = zh_phys;
	zilog->zl_os = os;
	zilog->zl_spa = dmu_objset_spa(os);
	zilog->zl_dmu_pool = dmu_objset_pool(os);
	zilog->zl_destroy_txg = TXG_INITIAL - 1;

	mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL);

	list_create(&zilog->zl_itx_list, sizeof (itx_t),
	    offsetof(itx_t, itx_node));

	list_create(&zilog->zl_lwb_list, sizeof (lwb_t),
	    offsetof(lwb_t, lwb_node));

	mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL);

	avl_create(&zilog->zl_vdev_tree, zil_vdev_compare,
	    sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node));

	cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL);
	cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL);

	return (zilog);
}

static void
blksz_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zsb->z_os)));
	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
	ASSERT(ISP2(newval));

	zsb->z_max_blksz = newval;
}

uint64_t
bplist_create(objset_t *mos, int blocksize, dmu_tx_t *tx)
{
	int size;

	size = spa_version(dmu_objset_spa(mos)) < SPA_VERSION_BPLIST_ACCOUNT ?
	    BPLIST_SIZE_V0 : sizeof (bplist_phys_t);

	return (dmu_object_alloc(mos, DMU_OT_BPLIST, blocksize,
	    DMU_OT_BPLIST_HDR, size, tx));
}

int
zfs_set_version(const char *name, uint64_t newvers)
{
	int error;
	objset_t *os;
	dmu_tx_t *tx;
	uint64_t curvers;

	/*
	 * XXX for now, require that the filesystem be unmounted.  Would
	 * be nice to find the zfsvfs_t and just update that if
	 * possible.
	 */

	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
		return (EINVAL);

	error = dmu_objset_open(name, DMU_OST_ZFS, DS_MODE_PRIMARY, &os);
	if (error)
		return (error);

	error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
	    8, 1, &curvers);
	if (error)
		goto out;
	if (newvers < curvers) {
		error = EINVAL;
		goto out;
	}

	tx = dmu_tx_create(os);
	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 0, ZPL_VERSION_STR);
	error = dmu_tx_assign(tx, TXG_WAIT);
	if (error) {
		dmu_tx_abort(tx);
		goto out;
	}
	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1,
	    &newvers, tx);

	spa_history_internal_log(LOG_DS_UPGRADE,
	    dmu_objset_spa(os), tx, CRED(),
	    "oldver=%llu newver=%llu dataset = %llu", curvers, newvers,
	    dmu_objset_id(os));
	dmu_tx_commit(tx);

out:
	dmu_objset_close(os);
	return (error);
}

static int
zvol_first_open(zvol_state_t *zv)
{
	objset_t *os;
	uint64_t volsize;
	int error;
	uint64_t ro;

	/* lie and say we're read-only */
	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os);
	if (error)
		return (SET_ERROR(-error));

	zv->zv_objset = os;

	error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
	if (error)
		goto out_owned;

	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
	if (error)
		goto out_owned;

	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
	if (error)
		goto out_owned;

	set_capacity(zv->zv_disk, volsize >> 9);
	zv->zv_volsize = volsize;
	zv->zv_zilog = zil_open(os, zvol_get_data);

	if (ro || dmu_objset_is_snapshot(os) ||
	    !spa_writeable(dmu_objset_spa(os))) {
		set_disk_ro(zv->zv_disk, 1);
		zv->zv_flags |= ZVOL_RDONLY;
	} else {
		set_disk_ro(zv->zv_disk, 0);
		zv->zv_flags &= ~ZVOL_RDONLY;
	}

out_owned:
	if (error) {
		dmu_objset_disown(os, zvol_tag);
		zv->zv_objset = NULL;
	}

	return (SET_ERROR(-error));
}

int
bplist_enqueue(bplist_t *bpl, const blkptr_t *bp, dmu_tx_t *tx)
{
	uint64_t blk, off;
	blkptr_t *bparray;
	int err;

	ASSERT(!BP_IS_HOLE(bp));
	mutex_enter(&bpl->bpl_lock);
	err = bplist_hold(bpl);
	if (err)
		return (err);

	blk = bpl->bpl_phys->bpl_entries >> bpl->bpl_bpshift;
	off = P2PHASE(bpl->bpl_phys->bpl_entries, 1ULL << bpl->bpl_bpshift);

	err = bplist_cache(bpl, blk);
	if (err) {
		mutex_exit(&bpl->bpl_lock);
		return (err);
	}

	dmu_buf_will_dirty(bpl->bpl_cached_dbuf, tx);
	bparray = bpl->bpl_cached_dbuf->db_data;
	bparray[off] = *bp;

	/* We never need the fill count. */
	bparray[off].blk_fill = 0;

	/* The bplist will compress better if we can leave off the checksum */
	if (!BP_GET_DEDUP(&bparray[off]))
		bzero(&bparray[off].blk_cksum, sizeof (bparray[off].blk_cksum));

	dmu_buf_will_dirty(bpl->bpl_dbuf, tx);
	bpl->bpl_phys->bpl_entries++;
	bpl->bpl_phys->bpl_bytes +=
	    bp_get_dsize_sync(dmu_objset_spa(bpl->bpl_mos), bp);
	if (bpl->bpl_havecomp) {
		bpl->bpl_phys->bpl_comp += BP_GET_PSIZE(bp);
		bpl->bpl_phys->bpl_uncomp += BP_GET_UCSIZE(bp);
	}
	mutex_exit(&bpl->bpl_lock);

	return (0);
}

static uint64_t
zpios_dmu_object_create(run_args_t *run_args, objset_t *os)
{
	struct dmu_tx *tx;
	uint64_t obj = 0ULL;
	uint64_t blksize = run_args->block_size;
	int rc;

	if (blksize < SPA_MINBLOCKSIZE ||
	    blksize > spa_maxblocksize(dmu_objset_spa(os)) ||
	    !ISP2(blksize)) {
		zpios_print(run_args->file,
		    "invalid block size for pool: %d\n", (int)blksize);
		return (obj);
	}

	tx = dmu_tx_create(os);
	dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, OBJ_SIZE);
	rc = dmu_tx_assign(tx, TXG_WAIT);
	if (rc) {
		zpios_print(run_args->file,
		    "dmu_tx_assign() failed: %d\n", rc);
		dmu_tx_abort(tx);
		return (obj);
	}

	obj = dmu_object_alloc(os, DMU_OT_UINT64_OTHER, 0, DMU_OT_NONE, 0, tx);
	rc = dmu_object_set_blocksize(os, obj, blksize, 0, tx);
	if (rc) {
		zpios_print(run_args->file,
		    "dmu_object_set_blocksize to %d failed: %d\n",
		    (int)blksize, rc);
		dmu_tx_abort(tx);
		return (obj);
	}

	dmu_tx_commit(tx);

	return (obj);
}

static int
zvol_first_open(zvol_state_t *zv)
{
	objset_t *os;
	uint64_t volsize;
	int locked = 0;
	int error;
	uint64_t ro;

	/*
	 * In all other cases the spa_namespace_lock is taken before the
	 * bdev->bd_mutex lock.  But in this case the Linux __blkdev_get()
	 * function calls fops->open() with the bdev->bd_mutex lock held.
	 *
	 * To avoid a potential lock inversion deadlock we preemptively
	 * try to take the spa_namespace_lock().  Normally it will not
	 * be contended and this is safe because spa_open_common() handles
	 * the case where the caller already holds the spa_namespace_lock.
	 *
	 * When it is contended we risk a lock inversion if we were to
	 * block waiting for the lock.  Luckily, the __blkdev_get()
	 * function allows us to return -ERESTARTSYS which will result in
	 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
	 * called again.  This process can be repeated safely until both
	 * locks are acquired.
	 */
	if (!mutex_owned(&spa_namespace_lock)) {
		locked = mutex_tryenter(&spa_namespace_lock);
		if (!locked)
			return (-ERESTARTSYS);
	}

	/* lie and say we're read-only */
	error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os);
	if (error)
		goto out_mutex;

	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
	if (error) {
		dmu_objset_disown(os, zvol_tag);
		goto out_mutex;
	}

	zv->zv_objset = os;
	error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
	if (error) {
		dmu_objset_disown(os, zvol_tag);
		goto out_mutex;
	}

	set_capacity(zv->zv_disk, volsize >> 9);
	zv->zv_volsize = volsize;
	zv->zv_zilog = zil_open(os, zvol_get_data);

	VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL) == 0);
	if (ro || dmu_objset_is_snapshot(os) ||
	    !spa_writeable(dmu_objset_spa(os))) {
		set_disk_ro(zv->zv_disk, 1);
		zv->zv_flags |= ZVOL_RDONLY;
	} else {
		set_disk_ro(zv->zv_disk, 0);
		zv->zv_flags &= ~ZVOL_RDONLY;
	}

out_mutex:
	if (locked)
		mutex_exit(&spa_namespace_lock);

	return (-error);
}

/*
 * Create a block device minor node and setup the linkage between it
 * and the specified volume.  Once this function returns the block
 * device is live and ready for use.
 */
static int
zvol_create_minor_impl(const char *name)
{
	zvol_state_t *zv;
	objset_t *os;
	dmu_object_info_t *doi;
	uint64_t volsize;
	uint64_t len;
	unsigned minor = 0;
	int error = 0;

	mutex_enter(&zvol_state_lock);

	zv = zvol_find_by_name(name);
	if (zv) {
		error = SET_ERROR(EEXIST);
		goto out;
	}

	doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);

	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os);
	if (error)
		goto out_doi;

	error = dmu_object_info(os, ZVOL_OBJ, doi);
	if (error)
		goto out_dmu_objset_disown;

	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
	if (error)
		goto out_dmu_objset_disown;

	error = zvol_find_minor(&minor);
	if (error)
		goto out_dmu_objset_disown;

	zv = zvol_alloc(MKDEV(zvol_major, minor), name);
	if (zv == NULL) {
		error = SET_ERROR(EAGAIN);
		goto out_dmu_objset_disown;
	}

	if (dmu_objset_is_snapshot(os))
		zv->zv_flags |= ZVOL_RDONLY;

	zv->zv_volblocksize = doi->doi_data_block_size;
	zv->zv_volsize = volsize;
	zv->zv_objset = os;

	set_capacity(zv->zv_disk, zv->zv_volsize >> 9);

	blk_queue_max_hw_sectors(zv->zv_queue, (DMU_MAX_ACCESS / 4) >> 9);
	blk_queue_max_segments(zv->zv_queue, UINT16_MAX);
	blk_queue_max_segment_size(zv->zv_queue, UINT_MAX);
	blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize);
	blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize);
	blk_queue_max_discard_sectors(zv->zv_queue,
	    (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9);
	blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize);
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue);
#ifdef QUEUE_FLAG_NONROT
	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue);
#endif
#ifdef QUEUE_FLAG_ADD_RANDOM
	queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zv->zv_queue);
#endif

	if (spa_writeable(dmu_objset_spa(os))) {
		if (zil_replay_disable)
			zil_destroy(dmu_objset_zil(os), B_FALSE);
		else
			zil_replay(os, zv, zvol_replay_vector);
	}

	/*
	 * When udev detects the addition of the device it will immediately
	 * invoke blkid(8) to determine the type of content on the device.
	 * Prefetching the blocks commonly scanned by blkid(8) will speed
	 * up this process.
	 */
	len = MIN(MAX(zvol_prefetch_bytes, 0), SPA_MAXBLOCKSIZE);
	if (len > 0) {
		dmu_prefetch(os, ZVOL_OBJ, 0, 0, len, ZIO_PRIORITY_SYNC_READ);
		dmu_prefetch(os, ZVOL_OBJ, 0, volsize - len, len,
			ZIO_PRIORITY_SYNC_READ);
	}

	zv->zv_objset = NULL;
out_dmu_objset_disown:
	dmu_objset_disown(os, zvol_tag);
out_doi:
	kmem_free(doi, sizeof (dmu_object_info_t));
out:

//.........这里部分代码省略.........

int
zfs_register_callbacks(zfs_sb_t *zsb)
{
	struct dsl_dataset *ds = NULL;
	objset_t *os = zsb->z_os;
	zfs_mntopts_t *zmo = zsb->z_mntopts;
	int error = 0;

	ASSERT(zsb);
	ASSERT(zmo);

	/*
	 * The act of registering our callbacks will destroy any mount
	 * options we may have.  In order to enable temporary overrides
	 * of mount options, we stash away the current values and
	 * restore them after we register the callbacks.
	 */
	if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os))) {
		zmo->z_do_readonly = B_TRUE;
		zmo->z_readonly = B_TRUE;
	}

	/*
	 * Register property callbacks.
	 *
	 * It would probably be fine to just check for i/o error from
	 * the first prop_register(), but I guess I like to go
	 * overboard...
	 */
	ds = dmu_objset_ds(os);
	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
	error = dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zsb);
	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
	if (error)
		goto unregister;

	/*
	 * Invoke our callbacks to restore temporary mount options.
	 */
	if (zmo->z_do_readonly)
		readonly_changed_cb(zsb, zmo->z_readonly);
	if (zmo->z_do_setuid)
		setuid_changed_cb(zsb, zmo->z_setuid);
	if (zmo->z_do_exec)
		exec_changed_cb(zsb, zmo->z_exec);
	if (zmo->z_do_devices)
		devices_changed_cb(zsb, zmo->z_devices);
	if (zmo->z_do_xattr)
		xattr_changed_cb(zsb, zmo->z_xattr);
	if (zmo->z_do_atime)
		atime_changed_cb(zsb, zmo->z_atime);
	if (zmo->z_do_relatime)
		relatime_changed_cb(zsb, zmo->z_relatime);
	if (zmo->z_do_nbmand)
		nbmand_changed_cb(zsb, zmo->z_nbmand);

	return (0);

unregister:
	/*
	 * We may attempt to unregister some callbacks that are not
	 * registered, but this is OK; it will simply return ENOMSG,
	 * which we will ignore.
	 */
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
	    atime_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RELATIME),
	    relatime_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
	    xattr_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
	    blksz_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
	    readonly_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
//.........这里部分代码省略.........

int
zfs_sb_setup(zfs_sb_t *zsb, boolean_t mounting)
{
	int error;

	error = zfs_register_callbacks(zsb);
	if (error)
		return (error);

	/*
	 * Set the objset user_ptr to track its zsb.
	 */
	mutex_enter(&zsb->z_os->os_user_ptr_lock);
	dmu_objset_set_user(zsb->z_os, zsb);
	mutex_exit(&zsb->z_os->os_user_ptr_lock);

	zsb->z_log = zil_open(zsb->z_os, zfs_get_data);

	/*
	 * If we are not mounting (ie: online recv), then we don't
	 * have to worry about replaying the log as we blocked all
	 * operations out since we closed the ZIL.
	 */
	if (mounting) {
		boolean_t readonly;

		/*
		 * During replay we remove the read only flag to
		 * allow replays to succeed.
		 */
		readonly = zfs_is_readonly(zsb);
		if (readonly != 0)
			readonly_changed_cb(zsb, B_FALSE);
		else
			zfs_unlinked_drain(zsb);

		/*
		 * Parse and replay the intent log.
		 *
		 * Because of ziltest, this must be done after
		 * zfs_unlinked_drain().  (Further note: ziltest
		 * doesn't use readonly mounts, where
		 * zfs_unlinked_drain() isn't called.)  This is because
		 * ziltest causes spa_sync() to think it's committed,
		 * but actually it is not, so the intent log contains
		 * many txg's worth of changes.
		 *
		 * In particular, if object N is in the unlinked set in
		 * the last txg to actually sync, then it could be
		 * actually freed in a later txg and then reallocated
		 * in a yet later txg.  This would write a "create
		 * object N" record to the intent log.  Normally, this
		 * would be fine because the spa_sync() would have
		 * written out the fact that object N is free, before
		 * we could write the "create object N" intent log
		 * record.
		 *
		 * But when we are in ziltest mode, we advance the "open
		 * txg" without actually spa_sync()-ing the changes to
		 * disk.  So we would see that object N is still
		 * allocated and in the unlinked set, and there is an
		 * intent log record saying to allocate it.
		 */
		if (spa_writeable(dmu_objset_spa(zsb->z_os))) {
			if (zil_replay_disable) {
				zil_destroy(zsb->z_log, B_FALSE);
			} else {
				zsb->z_replay = B_TRUE;
				zil_replay(zsb->z_os, zsb,
				    zfs_replay_vector);
				zsb->z_replay = B_FALSE;
			}
		}

		/* restore readonly bit */
		if (readonly != 0)
			readonly_changed_cb(zsb, B_TRUE);
	}

	return (0);
}

static int
__zvol_create_minor(const char *name, boolean_t ignore_snapdev)
{
	zvol_state_t *zv;
	objset_t *os;
	dmu_object_info_t *doi;
	uint64_t volsize;
	unsigned minor = 0;
	int error = 0;

	ASSERT(MUTEX_HELD(&zvol_state_lock));

	zv = zvol_find_by_name(name);
	if (zv) {
		error = EEXIST;
		goto out;
	}

	if (ignore_snapdev == B_FALSE) {
		error = __zvol_snapdev_hidden(name);
		if (error)
			goto out;
	}

	doi = kmem_alloc(sizeof(dmu_object_info_t), KM_SLEEP);

	error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os);
	if (error)
		goto out_doi;

	error = dmu_object_info(os, ZVOL_OBJ, doi);
	if (error)
		goto out_dmu_objset_disown;

	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
	if (error)
		goto out_dmu_objset_disown;

	error = zvol_find_minor(&minor);
	if (error)
		goto out_dmu_objset_disown;

	zv = zvol_alloc(MKDEV(zvol_major, minor), name);
	if (zv == NULL) {
		error = EAGAIN;
		goto out_dmu_objset_disown;
	}

	if (dmu_objset_is_snapshot(os))
		zv->zv_flags |= ZVOL_RDONLY;

	zv->zv_volblocksize = doi->doi_data_block_size;
	zv->zv_volsize = volsize;
	zv->zv_objset = os;

	set_capacity(zv->zv_disk, zv->zv_volsize >> 9);

	blk_queue_max_hw_sectors(zv->zv_queue, UINT_MAX);
	blk_queue_max_segments(zv->zv_queue, UINT16_MAX);
	blk_queue_max_segment_size(zv->zv_queue, UINT_MAX);
	blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize);
	blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize);
#ifdef HAVE_BLK_QUEUE_DISCARD
	blk_queue_max_discard_sectors(zv->zv_queue,
	    (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9);
	blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize);
	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue);
#endif
#ifdef HAVE_BLK_QUEUE_NONROT
	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue);
#endif

	if (spa_writeable(dmu_objset_spa(os))) {
		if (zil_replay_disable)
			zil_destroy(dmu_objset_zil(os), B_FALSE);
		else
			zil_replay(os, zv, zvol_replay_vector);
	}

	zv->zv_objset = NULL;
out_dmu_objset_disown:
	dmu_objset_disown(os, zvol_tag);
out_doi:
	kmem_free(doi, sizeof(dmu_object_info_t));
out:

	if (error == 0) {
		zvol_insert(zv);
		add_disk(zv->zv_disk);
	}

	return (error);
}

static int
feature_do_action(objset_t *os, uint64_t read_obj, uint64_t write_obj,
                  uint64_t desc_obj, zfeature_info_t *feature, feature_action_t action,
                  dmu_tx_t *tx)
{
    int error;
    uint64_t refcount;
    uint64_t zapobj = feature->fi_can_readonly ? write_obj : read_obj;

    ASSERT(0 != zapobj);
    ASSERT(zfeature_is_valid_guid(feature->fi_guid));

    error = zap_lookup(os, zapobj, feature->fi_guid,
                       sizeof (uint64_t), 1, &refcount);

    /*
     * If we can't ascertain the status of the specified feature, an I/O
     * error occurred.
     */
    if (error != 0 && error != ENOENT)
        return (error);

    switch (action) {
    case FEATURE_ACTION_ENABLE:
        /*
         * If the feature is already enabled, ignore the request.
         */
        if (error == 0)
            return (0);
        refcount = 0;
        break;
    case FEATURE_ACTION_INCR:
        if (error == ENOENT)
            return (ENOTSUP);
        if (refcount == UINT64_MAX)
            return (EOVERFLOW);
        refcount++;
        break;
    case FEATURE_ACTION_DECR:
        if (error == ENOENT)
            return (ENOTSUP);
        if (refcount == 0)
            return (EOVERFLOW);
        refcount--;
        break;
    default:
        ASSERT(0);
        break;
    }

    if (action == FEATURE_ACTION_ENABLE) {
        int i;

        for (i = 0; feature->fi_depends[i] != NULL; i++) {
            zfeature_info_t *dep = feature->fi_depends[i];

            error = feature_do_action(os, read_obj, write_obj,
                                      desc_obj, dep, FEATURE_ACTION_ENABLE, tx);
            if (error != 0)
                return (error);
        }
    }

    error = zap_update(os, zapobj, feature->fi_guid,
                       sizeof (uint64_t), 1, &refcount, tx);
    if (error != 0)
        return (error);

    if (action == FEATURE_ACTION_ENABLE) {
        error = zap_update(os, desc_obj,
                           feature->fi_guid, 1, strlen(feature->fi_desc) + 1,
                           feature->fi_desc, tx);
        if (error != 0)
            return (error);
    }

    if (action == FEATURE_ACTION_INCR && refcount == 1 && feature->fi_mos) {
        spa_activate_mos_feature(dmu_objset_spa(os), feature->fi_guid);
    }

    if (action == FEATURE_ACTION_DECR && refcount == 0) {
        spa_deactivate_mos_feature(dmu_objset_spa(os),
                                   feature->fi_guid);
    }

    return (0);
}

/*
 * Check if user has requested permission.
 */
int
dsl_deleg_access(const char *ddname, const char *perm, cred_t *cr)
{
	dsl_dir_t *dd, *startdd;
	dsl_pool_t *dp;
	void *cookie;
	int	error;
	char	checkflag = ZFS_DELEG_LOCAL;
	const char *tail;
	objset_t *mos;
	avl_tree_t permsets;
	perm_set_t *setnode;

	/*
	 * Use tail so that zfs_ioctl() code doesn't have
	 * to always to to figure out parent name in order
	 * to do access check.  for example renaming a snapshot
	 */
	error = dsl_dir_open(ddname, FTAG, &startdd, &tail);
	if (error)
		return (error);

	if (tail && tail[0] != '@') {
		dsl_dir_close(startdd, FTAG);
		return (ENOENT);
	}
	dp = startdd->dd_pool;
	mos = dp->dp_meta_objset;

	if (dsl_delegation_on(mos) == B_FALSE) {
		dsl_dir_close(startdd, FTAG);
		return (ECANCELED);
	}

	if (spa_version(dmu_objset_spa(dp->dp_meta_objset)) <
	    SPA_VERSION_DELEGATED_PERMS) {
		dsl_dir_close(startdd, FTAG);
		return (EPERM);
	}

	avl_create(&permsets, perm_set_compare, sizeof (perm_set_t),
	    offsetof(perm_set_t, p_node));

	rw_enter(&dp->dp_config_rwlock, RW_READER);
	for (dd = startdd; dd != NULL; dd = dd->dd_parent,
	    checkflag = ZFS_DELEG_DESCENDENT) {
		uint64_t zapobj;
		boolean_t expanded;

		/*
		 * If not in global zone then make sure
		 * the zoned property is set
		 */
		if (!INGLOBALZONE(curproc)) {
			uint64_t zoned;

			if (dsl_prop_get_ds_locked(dd,
			    zfs_prop_to_name(ZFS_PROP_ZONED),
			    8, 1, &zoned, NULL) != 0)
				break;
			if (!zoned)
				break;
		}
		zapobj = dd->dd_phys->dd_deleg_zapobj;

		if (zapobj == 0)
			continue;

		dsl_load_user_sets(mos, zapobj, &permsets, checkflag, cr);
again:
		expanded = B_FALSE;
		for (setnode = avl_first(&permsets); setnode;
		    setnode = AVL_NEXT(&permsets, setnode)) {
			if (setnode->p_matched == B_TRUE)
				continue;

			/* See if this set directly grants this permission */
			error = dsl_check_access(mos, zapobj,
			    ZFS_DELEG_NAMED_SET, 0, setnode->p_setname, perm);
			if (error == 0)
				goto success;
			if (error == EPERM)
				setnode->p_matched = B_TRUE;

			/* See if this set includes other sets */
			error = dsl_load_sets(mos, zapobj,
			    ZFS_DELEG_NAMED_SET_SETS, 0,
			    setnode->p_setname, &permsets);
			if (error == 0)
				setnode->p_matched = expanded = B_TRUE;
		}
		/*
		 * If we expanded any sets, that will define more sets,
		 * which we need to check.
		 */
		if (expanded)
			goto again;
//.........这里部分代码省略.........

void
dataset_kstats_create(dataset_kstats_t *dk, objset_t *objset)
{
	/*
	 * There should not be anything wrong with having kstats for
	 * snapshots. Since we are not sure how useful they would be
	 * though nor how much their memory overhead would matter in
	 * a filesystem with many snapshots, we skip them for now.
	 */
	if (dmu_objset_is_snapshot(objset))
		return;

	/*
	 * At the time of this writing, KSTAT_STRLEN is 255 in Linux,
	 * and the spa_name can theoretically be up to 256 characters.
	 * In reality though the spa_name can be 240 characters max
	 * [see origin directory name check in pool_namecheck()]. Thus,
	 * the naming scheme for the module name below should not cause
	 * any truncations. In the event that a truncation does happen
	 * though, due to some future change, we silently skip creating
	 * the kstat and log the event.
	 */
	char kstat_module_name[KSTAT_STRLEN];
	int n = snprintf(kstat_module_name, sizeof (kstat_module_name),
	    "zfs/%s", spa_name(dmu_objset_spa(objset)));
	if (n < 0) {
		zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
		    " snprintf() for kstat module name returned %d",
		    (unsigned long long)dmu_objset_id(objset), n);
		return;
	} else if (n >= KSTAT_STRLEN) {
		zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
		    "kstat module name length (%d) exceeds limit (%d)",
		    (unsigned long long)dmu_objset_id(objset),
		    n, KSTAT_STRLEN);
		return;
	}

	char kstat_name[KSTAT_STRLEN];
	n = snprintf(kstat_name, sizeof (kstat_name), "objset-0x%llx",
	    (unsigned long long)dmu_objset_id(objset));
	if (n < 0) {
		zfs_dbgmsg("failed to create dataset kstat for objset %lld: "
		    " snprintf() for kstat name returned %d",
		    (unsigned long long)dmu_objset_id(objset), n);
		return;
	}
	ASSERT3U(n, <, KSTAT_STRLEN);

	kstat_t *kstat = kstat_create(kstat_module_name, 0, kstat_name,
	    "dataset", KSTAT_TYPE_NAMED,
	    sizeof (empty_dataset_kstats) / sizeof (kstat_named_t),
	    KSTAT_FLAG_VIRTUAL);
	if (kstat == NULL)
		return;

	dataset_kstat_values_t *dk_kstats =
	    kmem_alloc(sizeof (empty_dataset_kstats), KM_SLEEP);
	bcopy(&empty_dataset_kstats, dk_kstats,
	    sizeof (empty_dataset_kstats));

	char *ds_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
	dsl_dataset_name(objset->os_dsl_dataset, ds_name);
	KSTAT_NAMED_STR_PTR(&dk_kstats->dkv_ds_name) = ds_name;
	KSTAT_NAMED_STR_BUFLEN(&dk_kstats->dkv_ds_name) =
	    ZFS_MAX_DATASET_NAME_LEN;

	kstat->ks_data = dk_kstats;
	kstat->ks_update = dataset_kstats_update;
	kstat->ks_private = dk;

	kstat_install(kstat);
	dk->dk_kstats = kstat;

	aggsum_init(&dk->dk_aggsums.das_writes, 0);
	aggsum_init(&dk->dk_aggsums.das_nwritten, 0);
	aggsum_init(&dk->dk_aggsums.das_reads, 0);
	aggsum_init(&dk->dk_aggsums.das_nread, 0);
	aggsum_init(&dk->dk_aggsums.das_nunlinks, 0);
	aggsum_init(&dk->dk_aggsums.das_nunlinked, 0);
}