コード例 #1
0
ファイル: dsl_userhold.c プロジェクト: AB17/zfs
/*
 * Called at spa_load time to release a stale temporary user hold.
 * Also called by the onexit code.
 */
void
dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, const char *htag)
{
	dsl_dataset_user_release_tmp_arg_t ddurta;

#ifdef _KERNEL
	dsl_dataset_t *ds;
	int error;

	/* Make sure it is not mounted. */
	dsl_pool_config_enter(dp, FTAG);
	error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
	if (error == 0) {
		char name[MAXNAMELEN];
		dsl_dataset_name(ds, name);
		dsl_dataset_rele(ds, FTAG);
		dsl_pool_config_exit(dp, FTAG);
		zfs_unmount_snap(name);
	} else {
		dsl_pool_config_exit(dp, FTAG);
	}
#endif

	ddurta.ddurta_dsobj = dsobj;
	ddurta.ddurta_holds = fnvlist_alloc();
	fnvlist_add_boolean(ddurta.ddurta_holds, htag);

	(void) dsl_sync_task(spa_name(dp->dp_spa),
	    dsl_dataset_user_release_tmp_check,
	    dsl_dataset_user_release_tmp_sync, &ddurta, 1);
	fnvlist_free(ddurta.ddurta_holds);
}
コード例 #2
0
ファイル: dsl_synctask.c プロジェクト: 64116278/zfs
/*
 * Called from open context to perform a callback in syncing context.  Waits
 * for the operation to complete.
 *
 * The checkfunc will be called from open context as a preliminary check
 * which can quickly fail.  If it succeeds, it will be called again from
 * syncing context.  The checkfunc should generally be designed to work
 * properly in either context, but if necessary it can check
 * dmu_tx_is_syncing(tx).
 *
 * The synctask infrastructure enforces proper locking strategy with respect
 * to the dp_config_rwlock -- the lock will always be held when the callbacks
 * are called.  It will be held for read during the open-context (preliminary)
 * call to the checkfunc, and then held for write from syncing context during
 * the calls to the check and sync funcs.
 *
 * A dataset or pool name can be passed as the first argument.  Typically,
 * the check func will hold, check the return value of the hold, and then
 * release the dataset.  The sync func will VERIFYO(hold()) the dataset.
 * This is safe because no changes can be made between the check and sync funcs,
 * and the sync func will only be called if the check func successfully opened
 * the dataset.
 */
int
dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc,
    dsl_syncfunc_t *syncfunc, void *arg,
    int blocks_modified, zfs_space_check_t space_check)
{
	spa_t *spa;
	dmu_tx_t *tx;
	int err;
	dsl_sync_task_t dst = { { { NULL } } };
	dsl_pool_t *dp;

	err = spa_open(pool, &spa, FTAG);
	if (err != 0)
		return (err);
	dp = spa_get_dsl(spa);

top:
	tx = dmu_tx_create_dd(dp->dp_mos_dir);
	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));

	dst.dst_pool = dp;
	dst.dst_txg = dmu_tx_get_txg(tx);
	dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT;
	dst.dst_space_check = space_check;
	dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc;
	dst.dst_syncfunc = syncfunc;
	dst.dst_arg = arg;
	dst.dst_error = 0;
	dst.dst_nowaiter = B_FALSE;

	dsl_pool_config_enter(dp, FTAG);
	err = dst.dst_checkfunc(arg, tx);
	dsl_pool_config_exit(dp, FTAG);

	if (err != 0) {
		dmu_tx_commit(tx);
		spa_close(spa, FTAG);
		return (err);
	}

	VERIFY(txg_list_add_tail(&dp->dp_sync_tasks, &dst, dst.dst_txg));

	dmu_tx_commit(tx);

	txg_wait_synced(dp, dst.dst_txg);

	if (dst.dst_error == EAGAIN) {
		txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE);
		goto top;
	}

	spa_close(spa, FTAG);
	return (dst.dst_error);
}
コード例 #3
0
/* ARGSUSED */
static int
zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
    offset_t *offp, offset_t *nextp, void *data, int flags)
{
	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
	char snapname[ZFS_MAX_DATASET_NAME_LEN];
	uint64_t id, cookie;
	boolean_t case_conflict;
	int error;

	ZFS_ENTER(zfsvfs);

	cookie = *offp;
	dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
	error = dmu_snapshot_list_next(zfsvfs->z_os,
	    sizeof (snapname), snapname, &id, &cookie, &case_conflict);
	dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
	if (error) {
		ZFS_EXIT(zfsvfs);
		if (error == ENOENT) {
			*eofp = 1;
			return (0);
		}
		return (error);
	}

	if (flags & V_RDDIR_ENTFLAGS) {
		edirent_t *eodp = dp;

		(void) strcpy(eodp->ed_name, snapname);
		eodp->ed_ino = ZFSCTL_INO_SNAP(id);
		eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
	} else {
		struct dirent64 *odp = dp;

		(void) strcpy(odp->d_name, snapname);
		odp->d_ino = ZFSCTL_INO_SNAP(id);
	}
	*nextp = cookie;

	ZFS_EXIT(zfsvfs);

	return (0);
}
コード例 #4
0
ファイル: dmu_objset.c プロジェクト: ngkaho1234/freebsd
/*
 * When we are called, os MUST refer to an objset associated with a dataset
 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
 * == tag.  We will then release and reacquire ownership of the dataset while
 * holding the pool config_rwlock to avoid intervening namespace or ownership
 * changes may occur.
 *
 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
 * release the hold on its dataset and acquire a new one on the dataset of the
 * same name so that it can be partially torn down and reconstructed.
 */
void
dmu_objset_refresh_ownership(objset_t *os, void *tag)
{
	dsl_pool_t *dp;
	dsl_dataset_t *ds, *newds;
	char name[MAXNAMELEN];

	ds = os->os_dsl_dataset;
	VERIFY3P(ds, !=, NULL);
	VERIFY3P(ds->ds_owner, ==, tag);
	VERIFY(dsl_dataset_long_held(ds));

	dsl_dataset_name(ds, name);
	dp = dmu_objset_pool(os);
	dsl_pool_config_enter(dp, FTAG);
	dmu_objset_disown(os, tag);
	VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
	VERIFY3P(newds, ==, os->os_dsl_dataset);
	dsl_pool_config_exit(dp, FTAG);
}
コード例 #5
0
ファイル: zpl_ctldir.c プロジェクト: awein/zfs
static int
zpl_snapdir_iterate(struct file *filp, struct dir_context *ctx)
{
	zfs_sb_t *zsb = ITOZSB(filp->f_path.dentry->d_inode);
	fstrans_cookie_t cookie;
	char snapname[MAXNAMELEN];
	boolean_t case_conflict;
	uint64_t id, pos;
	int error = 0;

	ZFS_ENTER(zsb);
	cookie = spl_fstrans_mark();

	if (!dir_emit_dots(filp, ctx))
		goto out;

	pos = ctx->pos;
	while (error == 0) {
		dsl_pool_config_enter(dmu_objset_pool(zsb->z_os), FTAG);
		error = -dmu_snapshot_list_next(zsb->z_os, MAXNAMELEN,
		    snapname, &id, &pos, &case_conflict);
		dsl_pool_config_exit(dmu_objset_pool(zsb->z_os), FTAG);
		if (error)
			goto out;

		if (!dir_emit(ctx, snapname, strlen(snapname),
		    ZFSCTL_INO_SHARES - id, DT_DIR))
			goto out;

		ctx->pos = pos;
	}
out:
	spl_fstrans_unmark(cookie);
	ZFS_EXIT(zsb);

	if (error == -ENOENT)
		return (0);

	return (error);
}
コード例 #6
0
ファイル: dsl_userhold.c プロジェクト: apprisi/illumos-gate
/*
 * The full semantics of this function are described in the comment above
 * lzc_release().
 *
 * To summarize:
 * Releases holds specified in the nvl holds.
 *
 * holds is nvl of snapname -> { holdname, ... }
 * errlist will be filled in with snapname -> error
 *
 * If tmpdp is not NULL the names for holds should be the dsobj's of snapshots,
 * otherwise they should be the names of shapshots.
 *
 * As a release may cause snapshots to be destroyed this trys to ensure they
 * aren't mounted.
 *
 * The release of non-existent holds are skipped.
 *
 * At least one hold must have been released for the this function to succeed
 * and return 0.
 */
static int
dsl_dataset_user_release_impl(nvlist_t *holds, nvlist_t *errlist,
    dsl_pool_t *tmpdp)
{
	dsl_dataset_user_release_arg_t ddura;
	nvpair_t *pair;
	char *pool;
	int error;

	pair = nvlist_next_nvpair(holds, NULL);
	if (pair == NULL)
		return (0);

	/*
	 * The release may cause snapshots to be destroyed; make sure they
	 * are not mounted.
	 */
	if (tmpdp != NULL) {
		/* Temporary holds are specified by dsobj string. */
		ddura.ddura_holdfunc = dsl_dataset_hold_obj_string;
		pool = spa_name(tmpdp->dp_spa);
#ifdef _KERNEL
		for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
		    pair = nvlist_next_nvpair(holds, pair)) {
			dsl_dataset_t *ds;

			dsl_pool_config_enter(tmpdp, FTAG);
			error = dsl_dataset_hold_obj_string(tmpdp,
			    nvpair_name(pair), FTAG, &ds);
			if (error == 0) {
				char name[MAXNAMELEN];
				dsl_dataset_name(ds, name);
				dsl_pool_config_exit(tmpdp, FTAG);
				dsl_dataset_rele(ds, FTAG);
				(void) zfs_unmount_snap(name);
			} else {
				dsl_pool_config_exit(tmpdp, FTAG);
			}
		}
#endif
	} else {
		/* Non-temporary holds are specified by name. */
		ddura.ddura_holdfunc = dsl_dataset_hold;
		pool = nvpair_name(pair);
#ifdef _KERNEL
		for (pair = nvlist_next_nvpair(holds, NULL); pair != NULL;
		    pair = nvlist_next_nvpair(holds, pair)) {
			(void) zfs_unmount_snap(nvpair_name(pair));
		}
#endif
	}

	ddura.ddura_holds = holds;
	ddura.ddura_errlist = errlist;
	ddura.ddura_todelete = fnvlist_alloc();
	ddura.ddura_chkholds = fnvlist_alloc();

	error = dsl_sync_task(pool, dsl_dataset_user_release_check,
	    dsl_dataset_user_release_sync, &ddura, 0);
	fnvlist_free(ddura.ddura_todelete);
	fnvlist_free(ddura.ddura_chkholds);

	return (error);
}
コード例 #7
0
ファイル: zfs_vfsops.c プロジェクト: nwf/zfs
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),
	    devices_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
	    setuid_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
	    exec_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
	    snapdir_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLTYPE),
	    acltype_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
	    acl_inherit_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
	    vscan_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND),
	    nbmand_changed_cb, zsb);

	return (error);
}
コード例 #8
0
ファイル: zfs_vfsops.c プロジェクト: AB17/zfs
int
zfs_register_callbacks(zfs_sb_t *zsb)
{
	struct dsl_dataset *ds = NULL;
	objset_t *os = zsb->z_os;
	boolean_t do_readonly = B_FALSE;
	int error = 0;

	if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os)))
		do_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_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_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;

	if (do_readonly)
		readonly_changed_cb(zsb, B_TRUE);

	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_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),
	    devices_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
	    setuid_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
	    exec_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
	    snapdir_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
	    acl_inherit_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
	    vscan_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND),
	    nbmand_changed_cb, zsb);

	return (error);
}
コード例 #9
0
ファイル: dmu_objset.c プロジェクト: KingD512/freebsd
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
    objset_t **osp)
{
	objset_t *os;
	int i, err;

	ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));

	os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
	os->os_dsl_dataset = ds;
	os->os_spa = spa;
	os->os_rootbp = bp;
	if (!BP_IS_HOLE(os->os_rootbp)) {
		arc_flags_t aflags = ARC_FLAG_WAIT;
		zbookmark_phys_t zb;
		SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
		    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);

		if (DMU_OS_IS_L2CACHEABLE(os))
			aflags |= ARC_FLAG_L2CACHE;

		dprintf_bp(os->os_rootbp, "reading %s", "");
		err = arc_read(NULL, spa, os->os_rootbp,
		    arc_getbuf_func, &os->os_phys_buf,
		    ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
		if (err != 0) {
			kmem_free(os, sizeof (objset_t));
			/* convert checksum errors into IO errors */
			if (err == ECKSUM)
				err = SET_ERROR(EIO);
			return (err);
		}

		/* Increase the blocksize if we are permitted. */
		if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
		    arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
			arc_buf_t *buf = arc_alloc_buf(spa,
			    sizeof (objset_phys_t), &os->os_phys_buf,
			    ARC_BUFC_METADATA);
			bzero(buf->b_data, sizeof (objset_phys_t));
			bcopy(os->os_phys_buf->b_data, buf->b_data,
			    arc_buf_size(os->os_phys_buf));
			arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
			os->os_phys_buf = buf;
		}

		os->os_phys = os->os_phys_buf->b_data;
		os->os_flags = os->os_phys->os_flags;
	} else {
		int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
		    sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
		os->os_phys_buf = arc_alloc_buf(spa, size,
		    &os->os_phys_buf, ARC_BUFC_METADATA);
		os->os_phys = os->os_phys_buf->b_data;
		bzero(os->os_phys, size);
	}

	/*
	 * Note: the changed_cb will be called once before the register
	 * func returns, thus changing the checksum/compression from the
	 * default (fletcher2/off).  Snapshots don't need to know about
	 * checksum/compression/copies.
	 */
	if (ds != NULL) {
		boolean_t needlock = B_FALSE;

		/*
		 * Note: it's valid to open the objset if the dataset is
		 * long-held, in which case the pool_config lock will not
		 * be held.
		 */
		if (!dsl_pool_config_held(dmu_objset_pool(os))) {
			needlock = B_TRUE;
			dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
		}
		err = dsl_prop_register(ds,
		    zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
		    primary_cache_changed_cb, os);
		if (err == 0) {
			err = dsl_prop_register(ds,
			    zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
			    secondary_cache_changed_cb, os);
		}
		if (!ds->ds_is_snapshot) {
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_CHECKSUM),
				    checksum_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
				    compression_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_COPIES),
				    copies_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_DEDUP),
				    dedup_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_LOGBIAS),
				    logbias_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_SYNC),
				    sync_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(
				    ZFS_PROP_REDUNDANT_METADATA),
				    redundant_metadata_changed_cb, os);
			}
			if (err == 0) {
				err = dsl_prop_register(ds,
				    zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
				    recordsize_changed_cb, os);
			}
		}
		if (needlock)
			dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
		if (err != 0) {
			arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
			kmem_free(os, sizeof (objset_t));
			return (err);
		}
	} else {
		/* It's the meta-objset. */
		os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
		os->os_compress = ZIO_COMPRESS_ON;
		os->os_copies = spa_max_replication(spa);
		os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
		os->os_dedup_verify = B_FALSE;
		os->os_logbias = ZFS_LOGBIAS_LATENCY;
		os->os_sync = ZFS_SYNC_STANDARD;
		os->os_primary_cache = ZFS_CACHE_ALL;
		os->os_secondary_cache = ZFS_CACHE_ALL;
	}

	if (ds == NULL || !ds->ds_is_snapshot)
		os->os_zil_header = os->os_phys->os_zil_header;
	os->os_zil = zil_alloc(os, &os->os_zil_header);

	for (i = 0; i < TXG_SIZE; i++) {
		list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
		    offsetof(dnode_t, dn_dirty_link[i]));
		list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
		    offsetof(dnode_t, dn_dirty_link[i]));
	}
	list_create(&os->os_dnodes, sizeof (dnode_t),
	    offsetof(dnode_t, dn_link));
	list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
	    offsetof(dmu_buf_impl_t, db_link));

	mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);

	dnode_special_open(os, &os->os_phys->os_meta_dnode,
	    DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
	if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
		dnode_special_open(os, &os->os_phys->os_userused_dnode,
		    DMU_USERUSED_OBJECT, &os->os_userused_dnode);
		dnode_special_open(os, &os->os_phys->os_groupused_dnode,
		    DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
	}

	*osp = os;
	return (0);
}