mirror_zfs/module/zfs/zfs_vfsops.c

/*
 * 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.
 */

/* Portions Copyright 2010 Robert Milkowski */

#include <sys/types.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/sysmacros.h>
#include <sys/kmem.h>
#include <sys/pathname.h>
#include <sys/vnode.h>
#include <sys/vfs.h>
#include <sys/vfs_opreg.h>
#include <sys/mntent.h>
#include <sys/mount.h>
#include <sys/cmn_err.h>
#include "fs/fs_subr.h"
#include <sys/zfs_znode.h>
#include <sys/zfs_vnops.h>
#include <sys/zfs_dir.h>
#include <sys/zil.h>
#include <sys/fs/zfs.h>
#include <sys/dmu.h>
#include <sys/dsl_prop.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_deleg.h>
#include <sys/spa.h>
#include <sys/zap.h>
#include <sys/sa.h>
#include <sys/varargs.h>
#include <sys/policy.h>
#include <sys/atomic.h>
#include <sys/mkdev.h>
#include <sys/modctl.h>
#include <sys/refstr.h>
#include <sys/zfs_ioctl.h>
#include <sys/zfs_fuid.h>
#include <sys/bootconf.h>
#include <sys/sunddi.h>
#include <sys/dnlc.h>
#include <sys/dmu_objset.h>
#include <sys/spa_boot.h>
#include <sys/sa.h>
#include <sys/zpl.h>
#include "zfs_comutil.h"


/*ARGSUSED*/
int
zfs_sync(zfs_sb_t *zsb, short flag, cred_t *cr)
{
	/*
	 * Data integrity is job one.  We don't want a compromised kernel
	 * writing to the storage pool, so we never sync during panic.
	 */
	if (unlikely(oops_in_progress))
		return (0);

	if (zsb != NULL) {
		/*
		 * Sync a specific filesystem.
		 */
		dsl_pool_t *dp;

		ZFS_ENTER(zsb);
		dp = dmu_objset_pool(zsb->z_os);

#ifdef HAVE_SHUTDOWN
		/*
		 * If the system is shutting down, then skip any
		 * filesystems which may exist on a suspended pool.
		 *
		 * XXX: This can be implemented using the Linux reboot
		 *      notifiers: {un}register_reboot_notifier().
		 */
		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
			ZFS_EXIT(zsb);
			return (0);
		}
#endif /* HAVE_SHUTDOWN */

		if (zsb->z_log != NULL)
			zil_commit(zsb->z_log, 0);

		ZFS_EXIT(zsb);
	} else {
		/*
		 * Sync all ZFS filesystems.  This is what happens when you
		 * run sync(1M).  Unlike other filesystems, ZFS honors the
		 * request by waiting for all pools to commit all dirty data.
		 */
		spa_sync_allpools();
	}

	return (0);
}
EXPORT_SYMBOL(zfs_sync);

static void
atime_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	struct super_block *sb = zsb->z_sb;
	struct vfsmount *vfs = zsb->z_vfs;

	if (newval == TRUE) {
		vfs->mnt_flags &= ~MNT_NOATIME;
		sb->s_flags &= ~MS_NOATIME;
		zsb->z_atime = TRUE;
	} else {
		vfs->mnt_flags |= MNT_NOATIME;
		sb->s_flags |= MS_NOATIME;
		zsb->z_atime = FALSE;
	}
}

static void
xattr_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;

	if (newval == TRUE) {
		zsb->z_flags |= ZSB_XATTR_USER;
	} else {
		zsb->z_flags &= ~ZSB_XATTR_USER;
	}
}

static void
blksz_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;

	if (newval < SPA_MINBLOCKSIZE ||
	    newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
		newval = SPA_MAXBLOCKSIZE;

	zsb->z_max_blksz = newval;
}

static void
readonly_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	struct super_block *sb = zsb->z_sb;
	struct vfsmount *vfs = zsb->z_vfs;

	if (newval) {
		vfs->mnt_flags |= MNT_READONLY;
		sb->s_flags |= MS_RDONLY;
	} else {
		vfs->mnt_flags &= ~MNT_READONLY;
		sb->s_flags &= ~MS_RDONLY;
	}
}

static void
devices_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	struct super_block *sb = zsb->z_sb;
	struct vfsmount *vfs = zsb->z_vfs;

	if (newval == FALSE) {
		vfs->mnt_flags |= MNT_NODEV;
		sb->s_flags |= MS_NODEV;
	} else {
		vfs->mnt_flags &= ~MNT_NODEV;
		sb->s_flags &= ~MS_NODEV;
	}
}

static void
setuid_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	struct super_block *sb = zsb->z_sb;
	struct vfsmount *vfs = zsb->z_vfs;

	if (newval == FALSE) {
		vfs->mnt_flags |= MNT_NOSUID;
		sb->s_flags |= MS_NOSUID;
	} else {
		vfs->mnt_flags &= ~MNT_NOSUID;
		sb->s_flags &= ~MS_NOSUID;
	}
}

static void
exec_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	struct super_block *sb = zsb->z_sb;
	struct vfsmount *vfs = zsb->z_vfs;

	if (newval == FALSE) {
		vfs->mnt_flags |= MNT_NOEXEC;
		sb->s_flags |= MS_NOEXEC;
	} else {
		vfs->mnt_flags &= ~MNT_NOEXEC;
		sb->s_flags &= ~MS_NOEXEC;
	}
}

/*
 * The nbmand mount option can be changed at mount time.
 * We can't allow it to be toggled on live file systems or incorrect
 * behavior may be seen from cifs clients
 *
 * This property isn't registered via dsl_prop_register(), but this callback
 * will be called when a file system is first mounted
 */
static void
nbmand_changed_cb(void *arg, uint64_t newval)
{
	zfs_sb_t *zsb = arg;
	struct super_block *sb = zsb->z_sb;

	if (newval == TRUE) {
		sb->s_flags |= MS_MANDLOCK;
	} else {
		sb->s_flags &= ~MS_MANDLOCK;
	}
}

static void
snapdir_changed_cb(void *arg, uint64_t newval)
{
	((zfs_sb_t *)arg)->z_show_ctldir = newval;
}

static void
vscan_changed_cb(void *arg, uint64_t newval)
{
	((zfs_sb_t *)arg)->z_vscan = newval;
}

static void
acl_inherit_changed_cb(void *arg, uint64_t newval)
{
	((zfs_sb_t *)arg)->z_acl_inherit = newval;
}

int
zfs_register_callbacks(zfs_sb_t *zsb)
{
	struct vfsmount *vfsp = zsb->z_vfs;
	struct dsl_dataset *ds = NULL;
	objset_t *os = zsb->z_os;
	uint64_t nbmand;
	boolean_t readonly = B_FALSE;
	boolean_t setuid = B_TRUE;
	boolean_t exec = B_TRUE;
	boolean_t devices = B_TRUE;
	boolean_t xattr = B_TRUE;
	boolean_t atime = B_TRUE;
	char osname[MAXNAMELEN];
	int error = 0;

	/*
	 * While Linux allows multiple vfs mounts per super block we have
	 * limited it artificially to one in zfs_fill_super.  Thus it is
	 * safe for us to modify the vfs mount fails through the callbacks.
	 */
	if ((vfsp->mnt_flags & MNT_READONLY) ||
	    !spa_writeable(dmu_objset_spa(os)))
		readonly = B_TRUE;

	if (vfsp->mnt_flags & MNT_NOSUID) {
		devices = B_FALSE;
		setuid = B_FALSE;
	} else {
		if (vfsp->mnt_flags & MNT_NODEV)
			devices = B_FALSE;
	}

	if (vfsp->mnt_flags & MNT_NOEXEC)
		exec = B_FALSE;

	if (vfsp->mnt_flags & MNT_NOATIME)
		atime = B_FALSE;

	/*
	 * nbmand is a special property which may only be changed at
	 * mount time.  Unfortunately, Linux does not have a VFS mount
	 * flag instead this is a super block flag.  So setting this
	 * option at mount time will have to wait until we can parse
	 * the mount option string.  For now we rely on the nbmand
	 * value stored with the object set.  Additional mount option
	 * string to be handled:
	 *
	 *   case: sensitive|insensitive|mixed
	 *   zerocopy: on|off
	 */

	dmu_objset_name(os, osname);
	if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand, NULL)))
		return (error);

	/*
	 * 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);
	error = dsl_prop_register(ds,
	    "atime", atime_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "xattr", xattr_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "recordsize", blksz_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "readonly", readonly_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "devices", devices_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "setuid", setuid_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "exec", exec_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "snapdir", snapdir_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "aclinherit", acl_inherit_changed_cb, zsb);
	error = error ? error : dsl_prop_register(ds,
	    "vscan", vscan_changed_cb, zsb);
	if (error)
		goto unregister;

	/*
	 * Invoke our callbacks to set required flags.
	 */
	readonly_changed_cb(zsb, readonly);
	setuid_changed_cb(zsb, setuid);
	exec_changed_cb(zsb, exec);
	devices_changed_cb(zsb, devices);
	xattr_changed_cb(zsb, xattr);
	atime_changed_cb(zsb, atime);
	nbmand_changed_cb(zsb, 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, "atime", atime_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zsb);
	(void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
	    zsb);
	(void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zsb);

	return (error);
}
EXPORT_SYMBOL(zfs_register_callbacks);

static int
zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
    uint64_t *userp, uint64_t *groupp)
{
	znode_phys_t *znp = data;
	int error = 0;

	/*
	 * Is it a valid type of object to track?
	 */
	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
		return (ENOENT);

	/*
	 * If we have a NULL data pointer
	 * then assume the id's aren't changing and
	 * return EEXIST to the dmu to let it know to
	 * use the same ids
	 */
	if (data == NULL)
		return (EEXIST);

	if (bonustype == DMU_OT_ZNODE) {
		*userp = znp->zp_uid;
		*groupp = znp->zp_gid;
	} else {
		int hdrsize;

		ASSERT(bonustype == DMU_OT_SA);
		hdrsize = sa_hdrsize(data);

		if (hdrsize != 0) {
			*userp = *((uint64_t *)((uintptr_t)data + hdrsize +
			    SA_UID_OFFSET));
			*groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
			    SA_GID_OFFSET));
		} else {
			/*
			 * This should only happen for newly created
			 * files that haven't had the znode data filled
			 * in yet.
			 */
			*userp = 0;
			*groupp = 0;
		}
	}
	return (error);
}

static void
fuidstr_to_sid(zfs_sb_t *zsb, const char *fuidstr,
    char *domainbuf, int buflen, uid_t *ridp)
{
	uint64_t fuid;
	const char *domain;

	fuid = strtonum(fuidstr, NULL);

	domain = zfs_fuid_find_by_idx(zsb, FUID_INDEX(fuid));
	if (domain)
		(void) strlcpy(domainbuf, domain, buflen);
	else
		domainbuf[0] = '\0';
	*ridp = FUID_RID(fuid);
}

static uint64_t
zfs_userquota_prop_to_obj(zfs_sb_t *zsb, zfs_userquota_prop_t type)
{
	switch (type) {
	case ZFS_PROP_USERUSED:
		return (DMU_USERUSED_OBJECT);
	case ZFS_PROP_GROUPUSED:
		return (DMU_GROUPUSED_OBJECT);
	case ZFS_PROP_USERQUOTA:
		return (zsb->z_userquota_obj);
	case ZFS_PROP_GROUPQUOTA:
		return (zsb->z_groupquota_obj);
	default:
		return (ENOTSUP);
	}
	return (0);
}

int
zfs_userspace_many(zfs_sb_t *zsb, zfs_userquota_prop_t type,
    uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
{
	int error;
	zap_cursor_t zc;
	zap_attribute_t za;
	zfs_useracct_t *buf = vbuf;
	uint64_t obj;

	if (!dmu_objset_userspace_present(zsb->z_os))
		return (ENOTSUP);

	obj = zfs_userquota_prop_to_obj(zsb, type);
	if (obj == 0) {
		*bufsizep = 0;
		return (0);
	}

	for (zap_cursor_init_serialized(&zc, zsb->z_os, obj, *cookiep);
	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
	    zap_cursor_advance(&zc)) {
		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
		    *bufsizep)
			break;

		fuidstr_to_sid(zsb, za.za_name,
		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);

		buf->zu_space = za.za_first_integer;
		buf++;
	}
	if (error == ENOENT)
		error = 0;

	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
	*cookiep = zap_cursor_serialize(&zc);
	zap_cursor_fini(&zc);
	return (error);
}
EXPORT_SYMBOL(zfs_userspace_many);

/*
 * buf must be big enough (eg, 32 bytes)
 */
static int
id_to_fuidstr(zfs_sb_t *zsb, const char *domain, uid_t rid,
    char *buf, boolean_t addok)
{
	uint64_t fuid;
	int domainid = 0;

	if (domain && domain[0]) {
		domainid = zfs_fuid_find_by_domain(zsb, domain, NULL, addok);
		if (domainid == -1)
			return (ENOENT);
	}
	fuid = FUID_ENCODE(domainid, rid);
	(void) sprintf(buf, "%llx", (longlong_t)fuid);
	return (0);
}

int
zfs_userspace_one(zfs_sb_t *zsb, zfs_userquota_prop_t type,
    const char *domain, uint64_t rid, uint64_t *valp)
{
	char buf[32];
	int err;
	uint64_t obj;

	*valp = 0;

	if (!dmu_objset_userspace_present(zsb->z_os))
		return (ENOTSUP);

	obj = zfs_userquota_prop_to_obj(zsb, type);
	if (obj == 0)
		return (0);

	err = id_to_fuidstr(zsb, domain, rid, buf, B_FALSE);
	if (err)
		return (err);

	err = zap_lookup(zsb->z_os, obj, buf, 8, 1, valp);
	if (err == ENOENT)
		err = 0;
	return (err);
}
EXPORT_SYMBOL(zfs_userspace_one);

int
zfs_set_userquota(zfs_sb_t *zsb, zfs_userquota_prop_t type,
    const char *domain, uint64_t rid, uint64_t quota)
{
	char buf[32];
	int err;
	dmu_tx_t *tx;
	uint64_t *objp;
	boolean_t fuid_dirtied;

	if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
		return (EINVAL);

	if (zsb->z_version < ZPL_VERSION_USERSPACE)
		return (ENOTSUP);

	objp = (type == ZFS_PROP_USERQUOTA) ? &zsb->z_userquota_obj :
	    &zsb->z_groupquota_obj;

	err = id_to_fuidstr(zsb, domain, rid, buf, B_TRUE);
	if (err)
		return (err);
	fuid_dirtied = zsb->z_fuid_dirty;

	tx = dmu_tx_create(zsb->z_os);
	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
	if (*objp == 0) {
		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
		    zfs_userquota_prop_prefixes[type]);
	}
	if (fuid_dirtied)
		zfs_fuid_txhold(zsb, tx);
	err = dmu_tx_assign(tx, TXG_WAIT);
	if (err) {
		dmu_tx_abort(tx);
		return (err);
	}

	mutex_enter(&zsb->z_lock);
	if (*objp == 0) {
		*objp = zap_create(zsb->z_os, DMU_OT_USERGROUP_QUOTA,
		    DMU_OT_NONE, 0, tx);
		VERIFY(0 == zap_add(zsb->z_os, MASTER_NODE_OBJ,
		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
	}
	mutex_exit(&zsb->z_lock);

	if (quota == 0) {
		err = zap_remove(zsb->z_os, *objp, buf, tx);
		if (err == ENOENT)
			err = 0;
	} else {
		err = zap_update(zsb->z_os, *objp, buf, 8, 1, &quota, tx);
	}
	ASSERT(err == 0);
	if (fuid_dirtied)
		zfs_fuid_sync(zsb, tx);
	dmu_tx_commit(tx);
	return (err);
}
EXPORT_SYMBOL(zfs_set_userquota);

boolean_t
zfs_fuid_overquota(zfs_sb_t *zsb, boolean_t isgroup, uint64_t fuid)
{
	char buf[32];
	uint64_t used, quota, usedobj, quotaobj;
	int err;

	usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
	quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj;

	if (quotaobj == 0 || zsb->z_replay)
		return (B_FALSE);

	(void) sprintf(buf, "%llx", (longlong_t)fuid);
	err = zap_lookup(zsb->z_os, quotaobj, buf, 8, 1, &quota);
	if (err != 0)
		return (B_FALSE);

	err = zap_lookup(zsb->z_os, usedobj, buf, 8, 1, &used);
	if (err != 0)
		return (B_FALSE);
	return (used >= quota);
}
EXPORT_SYMBOL(zfs_fuid_overquota);

boolean_t
zfs_owner_overquota(zfs_sb_t *zsb, znode_t *zp, boolean_t isgroup)
{
	uint64_t fuid;
	uint64_t quotaobj;

	quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj;

	fuid = isgroup ? zp->z_gid : zp->z_uid;

	if (quotaobj == 0 || zsb->z_replay)
		return (B_FALSE);

	return (zfs_fuid_overquota(zsb, isgroup, fuid));
}
EXPORT_SYMBOL(zfs_owner_overquota);

int
zfs_sb_create(const char *osname, zfs_sb_t **zsbp)
{
	objset_t *os;
	zfs_sb_t *zsb;
	uint64_t zval;
	int i, error;
	uint64_t sa_obj;

	zsb = kmem_zalloc(sizeof (zfs_sb_t), KM_SLEEP);

	/*
	 * We claim to always be readonly so we can open snapshots;
	 * other ZPL code will prevent us from writing to snapshots.
	 */
	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zsb, &os);
	if (error) {
		kmem_free(zsb, sizeof (zfs_sb_t));
		return (error);
	}

	/*
	 * Initialize the zfs-specific filesystem structure.
	 * Should probably make this a kmem cache, shuffle fields,
	 * and just bzero up to z_hold_mtx[].
	 */
	zsb->z_vfs = NULL;
	zsb->z_parent = zsb;
	zsb->z_max_blksz = SPA_MAXBLOCKSIZE;
	zsb->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
	zsb->z_os = os;

	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zsb->z_version);
	if (error) {
		goto out;
	} else if (zsb->z_version >
	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
		(void) printk("Can't mount a version %lld file system "
		    "on a version %lld pool\n. Pool must be upgraded to mount "
		    "this file system.", (u_longlong_t)zsb->z_version,
		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
		error = ENOTSUP;
		goto out;
	}
	if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
		goto out;
	zsb->z_norm = (int)zval;

	if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
		goto out;
	zsb->z_utf8 = (zval != 0);

	if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
		goto out;
	zsb->z_case = (uint_t)zval;

	/*
	 * Fold case on file systems that are always or sometimes case
	 * insensitive.
	 */
	if (zsb->z_case == ZFS_CASE_INSENSITIVE ||
	    zsb->z_case == ZFS_CASE_MIXED)
		zsb->z_norm |= U8_TEXTPREP_TOUPPER;

	zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os);
	zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os);

	if (zsb->z_use_sa) {
		/* should either have both of these objects or none */
		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
		    &sa_obj);
		if (error)
			return (error);
	} else {
		/*
		 * Pre SA versions file systems should never touch
		 * either the attribute registration or layout objects.
		 */
		sa_obj = 0;
	}

	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
	    &zsb->z_attr_table);
	if (error)
		goto out;

	if (zsb->z_version >= ZPL_VERSION_SA)
		sa_register_update_callback(os, zfs_sa_upgrade);

	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
	    &zsb->z_root);
	if (error)
		goto out;
	ASSERT(zsb->z_root != 0);

	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
	    &zsb->z_unlinkedobj);
	if (error)
		goto out;

	error = zap_lookup(os, MASTER_NODE_OBJ,
	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
	    8, 1, &zsb->z_userquota_obj);
	if (error && error != ENOENT)
		goto out;

	error = zap_lookup(os, MASTER_NODE_OBJ,
	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
	    8, 1, &zsb->z_groupquota_obj);
	if (error && error != ENOENT)
		goto out;

	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
	    &zsb->z_fuid_obj);
	if (error && error != ENOENT)
		goto out;

	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
	    &zsb->z_shares_dir);
	if (error && error != ENOENT)
		goto out;

	mutex_init(&zsb->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&zsb->z_lock, NULL, MUTEX_DEFAULT, NULL);
	list_create(&zsb->z_all_znodes, sizeof (znode_t),
	    offsetof(znode_t, z_link_node));
	rrw_init(&zsb->z_teardown_lock);
	rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
	rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL);
	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
		mutex_init(&zsb->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);

	*zsbp = zsb;
	return (0);

out:
	dmu_objset_disown(os, zsb);
	*zsbp = NULL;
	kmem_free(zsb, sizeof (zfs_sb_t));
	return (error);
}

static 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 = zsb->z_vfs->mnt_flags & MNT_READONLY;
		if (readonly != 0)
			zsb->z_vfs->mnt_flags &= ~MNT_READONLY;
		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;
			}
		}
		zsb->z_vfs->mnt_flags |= readonly; /* restore readonly bit */
	}

	return (0);
}

void
zfs_sb_free(zfs_sb_t *zsb)
{
	int i;

	zfs_fuid_destroy(zsb);

	mutex_destroy(&zsb->z_znodes_lock);
	mutex_destroy(&zsb->z_lock);
	list_destroy(&zsb->z_all_znodes);
	rrw_destroy(&zsb->z_teardown_lock);
	rw_destroy(&zsb->z_teardown_inactive_lock);
	rw_destroy(&zsb->z_fuid_lock);
	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
		mutex_destroy(&zsb->z_hold_mtx[i]);
	kmem_free(zsb, sizeof (zfs_sb_t));
}

static void
zfs_set_fuid_feature(zfs_sb_t *zsb)
{
	zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os);
	zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os);
}

void
zfs_unregister_callbacks(zfs_sb_t *zsb)
{
	objset_t *os = zsb->z_os;
	struct dsl_dataset *ds;

	/*
	 * Unregister properties.
	 */
	if (!dmu_objset_is_snapshot(os)) {
		ds = dmu_objset_ds(os);
		VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
		    zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "aclinherit",
		    acl_inherit_changed_cb, zsb) == 0);

		VERIFY(dsl_prop_unregister(ds, "vscan",
		    vscan_changed_cb, zsb) == 0);
	}
}
EXPORT_SYMBOL(zfs_unregister_callbacks);

#ifdef HAVE_MLSLABEL
/*
 * zfs_check_global_label:
 *	Check that the hex label string is appropriate for the dataset
 *	being mounted into the global_zone proper.
 *
 *	Return an error if the hex label string is not default or
 *	admin_low/admin_high.  For admin_low labels, the corresponding
 *	dataset must be readonly.
 */
int
zfs_check_global_label(const char *dsname, const char *hexsl)
{
	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
		return (0);
	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
		return (0);
	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
		/* must be readonly */
		uint64_t rdonly;

		if (dsl_prop_get_integer(dsname,
		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
			return (EACCES);
		return (rdonly ? 0 : EACCES);
	}
	return (EACCES);
}
#endif /* HAVE_MLSLABEL */

int
zfs_statvfs(struct dentry *dentry, struct kstatfs *statp)
{
	zfs_sb_t *zsb = dentry->d_sb->s_fs_info;
	uint64_t refdbytes, availbytes, usedobjs, availobjs;
	uint32_t bshift;

	ZFS_ENTER(zsb);

	dmu_objset_space(zsb->z_os,
	    &refdbytes, &availbytes, &usedobjs, &availobjs);

	/*
	 * The underlying storage pool actually uses multiple block sizes.
	 * We report the fragsize as the smallest block size we support,
	 * and we report our blocksize as the filesystem's maximum blocksize.
	 */
	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
	statp->f_bsize = zsb->z_max_blksz;
	bshift = fls(statp->f_bsize) - 1;

	/*
	 * The following report "total" blocks of various kinds in
	 * the file system, but reported in terms of f_bsize - the
	 * "preferred" size.
	 */

	statp->f_blocks = (refdbytes + availbytes) >> bshift;
	statp->f_bfree = availbytes >> bshift;
	statp->f_bavail = statp->f_bfree; /* no root reservation */

	/*
	 * statvfs() should really be called statufs(), because it assumes
	 * static metadata.  ZFS doesn't preallocate files, so the best
	 * we can do is report the max that could possibly fit in f_files,
	 * and that minus the number actually used in f_ffree.
	 * For f_ffree, report the smaller of the number of object available
	 * and the number of blocks (each object will take at least a block).
	 */
	statp->f_ffree = MIN(availobjs, statp->f_bfree);
	statp->f_files = statp->f_ffree + usedobjs;
	statp->f_fsid.val[0] = 0; /* XXX: Map up some unique ID */
	statp->f_fsid.val[1] = 0;
	statp->f_type = ZFS_SUPER_MAGIC;
	statp->f_namelen = ZFS_MAXNAMELEN;

	/*
	 * We have all of 40 characters to stuff a string here.
	 * Is there anything useful we could/should provide?
	 */
	bzero(statp->f_spare, sizeof (statp->f_spare));

	ZFS_EXIT(zsb);
	return (0);
}
EXPORT_SYMBOL(zfs_statvfs);

int
zfs_root(zfs_sb_t *zsb, struct inode **ipp)
{
	znode_t *rootzp;
	int error;

	ZFS_ENTER(zsb);

	error = zfs_zget(zsb, zsb->z_root, &rootzp);
	if (error == 0)
		*ipp = ZTOI(rootzp);

	ZFS_EXIT(zsb);
	return (error);
}
EXPORT_SYMBOL(zfs_root);

/*
 * Teardown the zfs_sb_t::z_os.
 *
 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
 * and 'z_teardown_inactive_lock' held.
 */
int
zfsvfs_teardown(zfs_sb_t *zsb, boolean_t unmounting)
{
	znode_t	*zp;

	rrw_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG);

#ifdef HAVE_DNLC
	if (!unmounting) {
		/*
		 * We purge the parent filesystem's vfsp as the parent
		 * filesystem and all of its snapshots have their vnode's
		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
		 * 'z_parent' is self referential for non-snapshots.
		 */
		(void) dnlc_purge_vfsp(zsb->z_parent->z_vfs, 0);
	}
#endif /* HAVE_DNLC */

	/*
	 * Close the zil. NB: Can't close the zil while zfs_inactive
	 * threads are blocked as zil_close can call zfs_inactive.
	 */
	if (zsb->z_log) {
		zil_close(zsb->z_log);
		zsb->z_log = NULL;
	}

	rw_enter(&zsb->z_teardown_inactive_lock, RW_WRITER);

	/*
	 * If we are not unmounting (ie: online recv) and someone already
	 * unmounted this file system while we were doing the switcheroo,
	 * or a reopen of z_os failed then just bail out now.
	 */
	if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) {
		rw_exit(&zsb->z_teardown_inactive_lock);
		rrw_exit(&zsb->z_teardown_lock, FTAG);
		return (EIO);
	}

	/*
	 * At this point there are no vops active, and any new vops will
	 * fail with EIO since we have z_teardown_lock for writer (only
	 * relavent for forced unmount).
	 *
	 * Release all holds on dbufs.
	 */
	mutex_enter(&zsb->z_znodes_lock);
	for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
	    zp = list_next(&zsb->z_all_znodes, zp))
		if (zp->z_sa_hdl) {
			ASSERT(atomic_read(&ZTOI(zp)->i_count) > 0);
			zfs_znode_dmu_fini(zp);
		}
	mutex_exit(&zsb->z_znodes_lock);

	/*
	 * If we are unmounting, set the unmounted flag and let new vops
	 * unblock.  zfs_inactive will have the unmounted behavior, and all
	 * other vops will fail with EIO.
	 */
	if (unmounting) {
		zsb->z_unmounted = B_TRUE;
		rrw_exit(&zsb->z_teardown_lock, FTAG);
		rw_exit(&zsb->z_teardown_inactive_lock);
	}

	/*
	 * z_os will be NULL if there was an error in attempting to reopen
	 * zsb, so just return as the properties had already been
	 *
	 * unregistered and cached data had been evicted before.
	 */
	if (zsb->z_os == NULL)
		return (0);

	/*
	 * Unregister properties.
	 */
	zfs_unregister_callbacks(zsb);

	/*
	 * Evict cached data
	 */
	if (dmu_objset_is_dirty_anywhere(zsb->z_os))
		if (!(zsb->z_vfs->mnt_flags & MNT_READONLY))
			txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
	(void) dmu_objset_evict_dbufs(zsb->z_os);

	return (0);
}

int
zfs_domount(struct super_block *sb, void *data, int silent)
{
	zpl_mount_data_t *zmd = data;
	const char *osname = zmd->z_osname;
	zfs_sb_t *zsb;
	struct inode *root_inode;
	uint64_t recordsize;
	int error;

	/*
	 * Linux allows multiple vfs mounts per super block.  However, the
	 * zfs_sb_t only contains a pointer for a single vfs mount.  This
	 * back reference in the long term could be extended to a list of
	 * vfs mounts if a hook were added to the kernel to notify us when
	 * a vfsmount is destroyed.  Until then we must limit the number
	 * of mounts per super block to one.
	 */
	if (atomic_read(&sb->s_active) > 1)
		return (EBUSY);

	error = zfs_sb_create(osname, &zsb);
	if (error)
		return (error);

	if ((error = dsl_prop_get_integer(osname, "recordsize",
	    &recordsize, NULL)))
		goto out;

	zsb->z_sb = sb;
	zsb->z_vfs = zmd->z_vfs;
	sb->s_fs_info = zsb;
	sb->s_magic = ZFS_SUPER_MAGIC;
	sb->s_maxbytes = MAX_LFS_FILESIZE;
	sb->s_time_gran = 1;
	sb->s_blocksize = recordsize;
	sb->s_blocksize_bits = ilog2(recordsize);

	/* Set callback operations for the file system. */
	sb->s_op = &zpl_super_operations;
	sb->s_xattr = zpl_xattr_handlers;
#ifdef HAVE_EXPORTS
        sb->s_export_op = &zpl_export_operations;
#endif /* HAVE_EXPORTS */

	/* Set features for file system. */
	zfs_set_fuid_feature(zsb);

	if (dmu_objset_is_snapshot(zsb->z_os)) {
		uint64_t pval;

		atime_changed_cb(zsb, B_FALSE);
		readonly_changed_cb(zsb, B_TRUE);
		if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
			goto out;
		xattr_changed_cb(zsb, pval);
		zsb->z_issnap = B_TRUE;
		zsb->z_os->os_sync = ZFS_SYNC_DISABLED;

		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);
	} else {
		error = zfs_sb_setup(zsb, B_TRUE);
#ifdef HAVE_SNAPSHOT
		(void) zfs_snap_create(zsb);
#endif /* HAVE_SNAPSHOT */
	}

	/* Allocate a root inode for the filesystem. */
	error = zfs_root(zsb, &root_inode);
	if (error) {
		(void) zfs_umount(sb);
		goto out;
	}

	/* Allocate a root dentry for the filesystem */
	sb->s_root = d_alloc_root(root_inode);
	if (sb->s_root == NULL) {
		(void) zfs_umount(sb);
		error = ENOMEM;
		goto out;
	}
out:
	if (error) {
		dmu_objset_disown(zsb->z_os, zsb);
		zfs_sb_free(zsb);
	}

	return (error);
}
EXPORT_SYMBOL(zfs_domount);

/*ARGSUSED*/
int
zfs_umount(struct super_block *sb)
{
	zfs_sb_t *zsb = sb->s_fs_info;
	objset_t *os;

	VERIFY(zfsvfs_teardown(zsb, B_TRUE) == 0);
	os = zsb->z_os;

	/*
	 * z_os will be NULL if there was an error in
	 * attempting to reopen zsb.
	 */
	if (os != NULL) {
		/*
		 * Unset the objset user_ptr.
		 */
		mutex_enter(&os->os_user_ptr_lock);
		dmu_objset_set_user(os, NULL);
		mutex_exit(&os->os_user_ptr_lock);

		/*
		 * Finally release the objset
		 */
		dmu_objset_disown(os, zsb);
	}

	zfs_sb_free(zsb);
	return (0);
}
EXPORT_SYMBOL(zfs_umount);

int
zfs_vget(struct vfsmount *vfsp, struct inode **ipp, fid_t *fidp)
{
	zfs_sb_t	*zsb = VTOZSB(vfsp);
	znode_t		*zp;
	uint64_t	object = 0;
	uint64_t	fid_gen = 0;
	uint64_t	gen_mask;
	uint64_t	zp_gen;
	int		i, err;

	*ipp = NULL;

	ZFS_ENTER(zsb);

	if (fidp->fid_len == LONG_FID_LEN) {
		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
		uint64_t	objsetid = 0;
		uint64_t	setgen = 0;

		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);

		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);

		ZFS_EXIT(zsb);

#ifdef HAVE_SNAPSHOT
		err = zfsctl_lookup_objset(vfsp, objsetid, &zsb);
		if (err)
			return (EINVAL);
#endif /* HAVE_SNAPSHOT */
		ZFS_ENTER(zsb);
	}

	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
		zfid_short_t	*zfid = (zfid_short_t *)fidp;

		for (i = 0; i < sizeof (zfid->zf_object); i++)
			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);

		for (i = 0; i < sizeof (zfid->zf_gen); i++)
			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
	} else {
		ZFS_EXIT(zsb);
		return (EINVAL);
	}

#ifdef HAVE_SNAPSHOT
	/* A zero fid_gen means we are in the .zfs control directories */
	if (fid_gen == 0 &&
	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
		*ipp = zsb->z_ctldir;
		ASSERT(*ipp != NULL);
		if (object == ZFSCTL_INO_SNAPDIR) {
			VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp, NULL,
			    0, NULL, NULL, NULL, NULL, NULL) == 0);
		} else {
			igrab(*ipp);
		}
		ZFS_EXIT(zsb);
		return (0);
	}
#endif /* HAVE_SNAPSHOT */

	gen_mask = -1ULL >> (64 - 8 * i);

	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
	if ((err = zfs_zget(zsb, object, &zp))) {
		ZFS_EXIT(zsb);
		return (err);
	}
	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb), &zp_gen,
	    sizeof (uint64_t));
	zp_gen = zp_gen & gen_mask;
	if (zp_gen == 0)
		zp_gen = 1;
	if (zp->z_unlinked || zp_gen != fid_gen) {
		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
		iput(ZTOI(zp));
		ZFS_EXIT(zsb);
		return (EINVAL);
	}

	*ipp = ZTOI(zp);
	if (*ipp)
		zfs_inode_update(ITOZ(*ipp));

	ZFS_EXIT(zsb);
	return (0);
}
EXPORT_SYMBOL(zfs_vget);

/*
 * Block out VOPs and close zfs_sb_t::z_os
 *
 * Note, if successful, then we return with the 'z_teardown_lock' and
 * 'z_teardown_inactive_lock' write held.
 */
int
zfs_suspend_fs(zfs_sb_t *zsb)
{
	int error;

	if ((error = zfsvfs_teardown(zsb, B_FALSE)) != 0)
		return (error);
	dmu_objset_disown(zsb->z_os, zsb);

	return (0);
}
EXPORT_SYMBOL(zfs_suspend_fs);

/*
 * Reopen zfs_sb_t::z_os and release VOPs.
 */
int
zfs_resume_fs(zfs_sb_t *zsb, const char *osname)
{
	int err, err2;

	ASSERT(RRW_WRITE_HELD(&zsb->z_teardown_lock));
	ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock));

	err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zsb, &zsb->z_os);
	if (err) {
		zsb->z_os = NULL;
	} else {
		znode_t *zp;
		uint64_t sa_obj = 0;

		err2 = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
		    ZFS_SA_ATTRS, 8, 1, &sa_obj);

		if ((err || err2) && zsb->z_version >= ZPL_VERSION_SA)
			goto bail;


		if ((err = sa_setup(zsb->z_os, sa_obj,
		    zfs_attr_table,  ZPL_END, &zsb->z_attr_table)) != 0)
			goto bail;

		VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);

		/*
		 * Attempt to re-establish all the active znodes with
		 * their dbufs.  If a zfs_rezget() fails, then we'll let
		 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
		 * when they try to use their znode.
		 */
		mutex_enter(&zsb->z_znodes_lock);
		for (zp = list_head(&zsb->z_all_znodes); zp;
		    zp = list_next(&zsb->z_all_znodes, zp)) {
			(void) zfs_rezget(zp);
		}
		mutex_exit(&zsb->z_znodes_lock);

	}

bail:
	/* release the VOPs */
	rw_exit(&zsb->z_teardown_inactive_lock);
	rrw_exit(&zsb->z_teardown_lock, FTAG);

	if (err) {
		/*
		 * Since we couldn't reopen zfs_sb_t::z_os, force
		 * unmount this file system.
		 */
		(void) zfs_umount(zsb->z_sb);
	}
	return (err);
}
EXPORT_SYMBOL(zfs_resume_fs);

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 (EINVAL);

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

	if (zfs_spa_version_map(newvers) >
	    spa_version(dmu_objset_spa(zsb->z_os)))
		return (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);
		ASSERT3U(error, ==, 0);

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

	spa_history_log_internal(LOG_DS_UPGRADE,
	    dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
	    zsb->z_version, newvers, dmu_objset_id(os));

	dmu_tx_commit(tx);

	zsb->z_version = newvers;

	if (zsb->z_version >= ZPL_VERSION_FUID)
		zfs_set_fuid_feature(zsb);

	return (0);
}
EXPORT_SYMBOL(zfs_set_version);

/*
 * Read a property stored within the master node.
 */
int
zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
{
	const char *pname;
	int error = ENOENT;

	/*
	 * Look up the file system's value for the property.  For the
	 * version property, we look up a slightly different string.
	 */
	if (prop == ZFS_PROP_VERSION)
		pname = ZPL_VERSION_STR;
	else
		pname = zfs_prop_to_name(prop);

	if (os != NULL)
		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);

	if (error == ENOENT) {
		/* No value set, use the default value */
		switch (prop) {
		case ZFS_PROP_VERSION:
			*value = ZPL_VERSION;
			break;
		case ZFS_PROP_NORMALIZE:
		case ZFS_PROP_UTF8ONLY:
			*value = 0;
			break;
		case ZFS_PROP_CASE:
			*value = ZFS_CASE_SENSITIVE;
			break;
		default:
			return (error);
		}
		error = 0;
	}
	return (error);
}

void
zfs_init(void)
{
	zfs_znode_init();
	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
	register_filesystem(&zpl_fs_type);
}

void
zfs_fini(void)
{
	unregister_filesystem(&zpl_fs_type);
	zfs_znode_fini();
}