xref: /freebsd-9-stable/sys/cddl/contrib/opensolaris/uts/common/fs/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.
 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
 * All rights reserved.
 * Copyright (c) 2013 by Delphix. All rights reserved.
 */

/* Portions Copyright 2010 Robert Milkowski */

#include <sys/types.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/sysmacros.h>
#include <sys/kmem.h>
#include <sys/acl.h>
#include <sys/vnode.h>
#include <sys/vfs.h>
#include <sys/mntent.h>
#include <sys/mount.h>
#include <sys/cmn_err.h>
#include <sys/zfs_znode.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/sa_impl.h>
#include <sys/varargs.h>
#include <sys/policy.h>
#include <sys/atomic.h>
#include <sys/zfs_ioctl.h>
#include <sys/zfs_ctldir.h>
#include <sys/zfs_fuid.h>
#include <sys/sunddi.h>
#include <sys/dnlc.h>
#include <sys/dmu_objset.h>
#include <sys/spa_boot.h>
#include <sys/jail.h>
#include "zfs_comutil.h"

struct mtx zfs_debug_mtx;
MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);

SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");

int zfs_super_owner;
SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
    "File system owner can perform privileged operation on his file systems");

int zfs_debug_level;
TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level);
SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0,
    "Debug level");

SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
static int zfs_version_acl = ZFS_ACL_VERSION;
SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
    "ZFS_ACL_VERSION");
static int zfs_version_spa = SPA_VERSION;
SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
    "SPA_VERSION");
static int zfs_version_zpl = ZPL_VERSION;
SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
    "ZPL_VERSION");

static int zfs_mount(vfs_t *vfsp);
static int zfs_umount(vfs_t *vfsp, int fflag);
static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
static int zfs_sync(vfs_t *vfsp, int waitfor);
static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
    struct ucred **credanonp, int *numsecflavors, int **secflavors);
static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
static void zfs_objset_close(zfsvfs_t *zfsvfs);
static void zfs_freevfs(vfs_t *vfsp);

static struct vfsops zfs_vfsops = {
	.vfs_mount =		zfs_mount,
	.vfs_unmount =		zfs_umount,
	.vfs_root =		zfs_root,
	.vfs_statfs =		zfs_statfs,
	.vfs_vget =		zfs_vget,
	.vfs_sync =		zfs_sync,
	.vfs_checkexp =		zfs_checkexp,
	.vfs_fhtovp =		zfs_fhtovp,
};

VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);

/*
 * We need to keep a count of active fs's.
 * This is necessary to prevent our module
 * from being unloaded after a umount -f
 */
static uint32_t	zfs_active_fs_count = 0;

/*ARGSUSED*/
static int
zfs_sync(vfs_t *vfsp, int waitfor)
{

	/*
	 * Data integrity is job one.  We don't want a compromised kernel
	 * writing to the storage pool, so we never sync during panic.
	 */
	if (panicstr)
		return (0);

	if (vfsp != NULL) {
		/*
		 * Sync a specific filesystem.
		 */
		zfsvfs_t *zfsvfs = vfsp->vfs_data;
		dsl_pool_t *dp;
		int error;

		error = vfs_stdsync(vfsp, waitfor);
		if (error != 0)
			return (error);

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

		/*
		 * If the system is shutting down, then skip any
		 * filesystems which may exist on a suspended pool.
		 */
		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
			ZFS_EXIT(zfsvfs);
			return (0);
		}

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

		ZFS_EXIT(zfsvfs);
	} 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);
}

#ifndef __FreeBSD__
static int
zfs_create_unique_device(dev_t *dev)
{
	major_t new_major;

	do {
		ASSERT3U(zfs_minor, <=, MAXMIN32);
		minor_t start = zfs_minor;
		do {
			mutex_enter(&zfs_dev_mtx);
			if (zfs_minor >= MAXMIN32) {
				/*
				 * If we're still using the real major
				 * keep out of /dev/zfs and /dev/zvol minor
				 * number space.  If we're using a getudev()'ed
				 * major number, we can use all of its minors.
				 */
				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
					zfs_minor = ZFS_MIN_MINOR;
				else
					zfs_minor = 0;
			} else {
				zfs_minor++;
			}
			*dev = makedevice(zfs_major, zfs_minor);
			mutex_exit(&zfs_dev_mtx);
		} while (vfs_devismounted(*dev) && zfs_minor != start);
		if (zfs_minor == start) {
			/*
			 * We are using all ~262,000 minor numbers for the
			 * current major number.  Create a new major number.
			 */
			if ((new_major = getudev()) == (major_t)-1) {
				cmn_err(CE_WARN,
				    "zfs_mount: Can't get unique major "
				    "device number.");
				return (-1);
			}
			mutex_enter(&zfs_dev_mtx);
			zfs_major = new_major;
			zfs_minor = 0;

			mutex_exit(&zfs_dev_mtx);
		} else {
			break;
		}
		/* CONSTANTCONDITION */
	} while (1);

	return (0);
}
#endif	/* !__FreeBSD__ */

static void
atime_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	if (newval == TRUE) {
		zfsvfs->z_atime = TRUE;
		zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
	} else {
		zfsvfs->z_atime = FALSE;
		zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
	}
}

static void
xattr_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	if (newval == TRUE) {
		/* XXX locking on vfs_flag? */
#ifdef TODO
		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
#endif
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
	} else {
		/* XXX locking on vfs_flag? */
#ifdef TODO
		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
#endif
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
	}
}

static void
blksz_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

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

	zfsvfs->z_max_blksz = newval;
	zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
}

static void
readonly_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	if (newval) {
		/* XXX locking on vfs_flag? */
		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
	} else {
		/* XXX locking on vfs_flag? */
		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
	}
}

static void
setuid_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	if (newval == FALSE) {
		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
	} else {
		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
	}
}

static void
exec_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	if (newval == FALSE) {
		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
	} else {
		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
	}
}

/*
 * 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)
{
	zfsvfs_t *zfsvfs = arg;
	if (newval == FALSE) {
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
	} else {
		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
	}
}

static void
snapdir_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	zfsvfs->z_show_ctldir = newval;
}

static void
vscan_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	zfsvfs->z_vscan = newval;
}

static void
acl_mode_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	zfsvfs->z_acl_mode = newval;
}

static void
acl_inherit_changed_cb(void *arg, uint64_t newval)
{
	zfsvfs_t *zfsvfs = arg;

	zfsvfs->z_acl_inherit = newval;
}

static int
zfs_register_callbacks(vfs_t *vfsp)
{
	struct dsl_dataset *ds = NULL;
	objset_t *os = NULL;
	zfsvfs_t *zfsvfs = NULL;
	uint64_t nbmand;
	boolean_t readonly = B_FALSE;
	boolean_t do_readonly = B_FALSE;
	boolean_t setuid = B_FALSE;
	boolean_t do_setuid = B_FALSE;
	boolean_t exec = B_FALSE;
	boolean_t do_exec = B_FALSE;
#ifdef illumos
	boolean_t devices = B_FALSE;
	boolean_t do_devices = B_FALSE;
#endif
	boolean_t xattr = B_FALSE;
	boolean_t do_xattr = B_FALSE;
	boolean_t atime = B_FALSE;
	boolean_t do_atime = B_FALSE;
	int error = 0;

	ASSERT(vfsp);
	zfsvfs = vfsp->vfs_data;
	ASSERT(zfsvfs);
	os = zfsvfs->z_os;

	/*
	 * This function can be called for a snapshot when we update snapshot's
	 * mount point, which isn't really supported.
	 */
	if (dmu_objset_is_snapshot(os))
		return (EOPNOTSUPP);

	/*
	 * 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 (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
	    !spa_writeable(dmu_objset_spa(os))) {
		readonly = B_TRUE;
		do_readonly = B_TRUE;
	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
		readonly = B_FALSE;
		do_readonly = B_TRUE;
	}
	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
		setuid = B_FALSE;
		do_setuid = B_TRUE;
	} else {
		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
			setuid = B_FALSE;
			do_setuid = B_TRUE;
		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
			setuid = B_TRUE;
			do_setuid = B_TRUE;
		}
	}
	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
		exec = B_FALSE;
		do_exec = B_TRUE;
	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
		exec = B_TRUE;
		do_exec = B_TRUE;
	}
	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
		xattr = B_FALSE;
		do_xattr = B_TRUE;
	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
		xattr = B_TRUE;
		do_xattr = B_TRUE;
	}
	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
		atime = B_FALSE;
		do_atime = B_TRUE;
	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
		atime = B_TRUE;
		do_atime = B_TRUE;
	}

	/*
	 * We need to enter pool configuration here, so that we can use
	 * dsl_prop_get_int_ds() to handle the special nbmand property below.
	 * dsl_prop_get_integer() can not be used, because it has to acquire
	 * spa_namespace_lock and we can not do that because we already hold
	 * z_teardown_lock.  The problem is that spa_config_sync() is called
	 * with spa_namespace_lock held and the function calls ZFS vnode
	 * operations to write the cache file and thus z_teardown_lock is
	 * acquired after spa_namespace_lock.
	 */
	ds = dmu_objset_ds(os);
	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);

	/*
	 * nbmand is a special property.  It can only be changed at
	 * mount time.
	 *
	 * This is weird, but it is documented to only be changeable
	 * at mount time.
	 */
	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
		nbmand = B_FALSE;
	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
		nbmand = B_TRUE;
	} else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
		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...
	 */
	error = dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
#ifdef illumos
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
#endif
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
	    zfsvfs);
	error = error ? error : dsl_prop_register(ds,
	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
	if (error)
		goto unregister;

	/*
	 * Invoke our callbacks to restore temporary mount options.
	 */
	if (do_readonly)
		readonly_changed_cb(zfsvfs, readonly);
	if (do_setuid)
		setuid_changed_cb(zfsvfs, setuid);
	if (do_exec)
		exec_changed_cb(zfsvfs, exec);
	if (do_xattr)
		xattr_changed_cb(zfsvfs, xattr);
	if (do_atime)
		atime_changed_cb(zfsvfs, atime);

	nbmand_changed_cb(zfsvfs, 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, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
	    xattr_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
	    blksz_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
	    readonly_changed_cb, zfsvfs);
#ifdef illumos
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
	    devices_changed_cb, zfsvfs);
#endif
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
	    setuid_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
	    exec_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
	    snapdir_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
	    acl_mode_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
	    acl_inherit_changed_cb, zfsvfs);
	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
	    vscan_changed_cb, zfsvfs);
	return (error);
}

static int
zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
    uint64_t *userp, uint64_t *groupp)
{
	/*
	 * Is it a valid type of object to track?
	 */
	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
		return (SET_ERROR(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 (SET_ERROR(EEXIST));

	if (bonustype == DMU_OT_ZNODE) {
		znode_phys_t *znp = data;
		*userp = znp->zp_uid;
		*groupp = znp->zp_gid;
	} else {
		int hdrsize;
		sa_hdr_phys_t *sap = data;
		sa_hdr_phys_t sa = *sap;
		boolean_t swap = B_FALSE;

		ASSERT(bonustype == DMU_OT_SA);

		if (sa.sa_magic == 0) {
			/*
			 * This should only happen for newly created
			 * files that haven't had the znode data filled
			 * in yet.
			 */
			*userp = 0;
			*groupp = 0;
			return (0);
		}
		if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
			sa.sa_magic = SA_MAGIC;
			sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
			swap = B_TRUE;
		} else {
			VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
		}

		hdrsize = sa_hdrsize(&sa);
		VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
		*userp = *((uint64_t *)((uintptr_t)data + hdrsize +
		    SA_UID_OFFSET));
		*groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
		    SA_GID_OFFSET));
		if (swap) {
			*userp = BSWAP_64(*userp);
			*groupp = BSWAP_64(*groupp);
		}
	}
	return (0);
}

static void
fuidstr_to_sid(zfsvfs_t *zfsvfs, 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(zfsvfs, 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(zfsvfs_t *zfsvfs, 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 (zfsvfs->z_userquota_obj);
	case ZFS_PROP_GROUPQUOTA:
		return (zfsvfs->z_groupquota_obj);
	}
	return (0);
}

int
zfs_userspace_many(zfsvfs_t *zfsvfs, 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(zfsvfs->z_os))
		return (SET_ERROR(ENOTSUP));

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

	for (zap_cursor_init_serialized(&zc, zfsvfs->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(zfsvfs, 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);
}

/*
 * buf must be big enough (eg, 32 bytes)
 */
static int
id_to_fuidstr(zfsvfs_t *zfsvfs, 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(zfsvfs, domain, NULL, addok);
		if (domainid == -1)
			return (SET_ERROR(ENOENT));
	}
	fuid = FUID_ENCODE(domainid, rid);
	(void) sprintf(buf, "%llx", (longlong_t)fuid);
	return (0);
}

int
zfs_userspace_one(zfsvfs_t *zfsvfs, 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(zfsvfs->z_os))
		return (SET_ERROR(ENOTSUP));

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

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

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

int
zfs_set_userquota(zfsvfs_t *zfsvfs, 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 (SET_ERROR(EINVAL));

	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
		return (SET_ERROR(ENOTSUP));

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

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

	tx = dmu_tx_create(zfsvfs->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(zfsvfs, tx);
	err = dmu_tx_assign(tx, TXG_WAIT);
	if (err) {
		dmu_tx_abort(tx);
		return (err);
	}

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

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

boolean_t
zfs_fuid_overquota(zfsvfs_t *zfsvfs, 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 ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;

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

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

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

boolean_t
zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
{
	uint64_t fuid;
	uint64_t quotaobj;

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

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

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

	return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
}

int
zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
{
	objset_t *os;
	zfsvfs_t *zfsvfs;
	uint64_t zval;
	int i, error;
	uint64_t sa_obj;

	zfsvfs = kmem_zalloc(sizeof (zfsvfs_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, zfsvfs, &os);
	if (error) {
		kmem_free(zfsvfs, sizeof (zfsvfs_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[].
	 */
	zfsvfs->z_vfs = NULL;
	zfsvfs->z_parent = zfsvfs;
	zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
	zfsvfs->z_os = os;

	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
	if (error) {
		goto out;
	} else if (zfsvfs->z_version >
	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
		(void) printf("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)zfsvfs->z_version,
		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
		error = SET_ERROR(ENOTSUP);
		goto out;
	}
	if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
		goto out;
	zfsvfs->z_norm = (int)zval;

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

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

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

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

	if (zfsvfs->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,
	    &zfsvfs->z_attr_table);
	if (error)
		goto out;

	if (zfsvfs->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,
	    &zfsvfs->z_root);
	if (error)
		goto out;
	ASSERT(zfsvfs->z_root != 0);

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

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

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

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

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

	*zfvp = zfsvfs;
	return (0);

out:
	dmu_objset_disown(os, zfsvfs);
	*zfvp = NULL;
	kmem_free(zfsvfs, sizeof (zfsvfs_t));
	return (error);
}

static int
zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
{
	int error;

	error = zfs_register_callbacks(zfsvfs->z_vfs);
	if (error)
		return (error);

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

	zfsvfs->z_log = zil_open(zfsvfs->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 = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
		if (readonly != 0)
			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
		else
			zfs_unlinked_drain(zfsvfs);

		/*
		 * 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(zfsvfs->z_os))) {
			if (zil_replay_disable) {
				zil_destroy(zfsvfs->z_log, B_FALSE);
			} else {
				zfsvfs->z_replay = B_TRUE;
				zil_replay(zfsvfs->z_os, zfsvfs,
				    zfs_replay_vector);
				zfsvfs->z_replay = B_FALSE;
			}
		}
		zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
	}

	return (0);
}

extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */

void
zfsvfs_free(zfsvfs_t *zfsvfs)
{
	int i;

	/*
	 * This is a barrier to prevent the filesystem from going away in
	 * zfs_znode_move() until we can safely ensure that the filesystem is
	 * not unmounted. We consider the filesystem valid before the barrier
	 * and invalid after the barrier.
	 */
	rw_enter(&zfsvfs_lock, RW_READER);
	rw_exit(&zfsvfs_lock);

	zfs_fuid_destroy(zfsvfs);

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

static void
zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
{
	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
	if (zfsvfs->z_vfs) {
		if (zfsvfs->z_use_fuids) {
			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
		} else {
			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
		}
	}
	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
}

static int
zfs_domount(vfs_t *vfsp, char *osname)
{
	uint64_t recordsize, fsid_guid;
	int error = 0;
	zfsvfs_t *zfsvfs;
	vnode_t *vp;

	ASSERT(vfsp);
	ASSERT(osname);

	error = zfsvfs_create(osname, &zfsvfs);
	if (error)
		return (error);
	zfsvfs->z_vfs = vfsp;

#ifdef illumos
	/* Initialize the generic filesystem structure. */
	vfsp->vfs_bcount = 0;
	vfsp->vfs_data = NULL;

	if (zfs_create_unique_device(&mount_dev) == -1) {
		error = SET_ERROR(ENODEV);
		goto out;
	}
	ASSERT(vfs_devismounted(mount_dev) == 0);
#endif

	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
	    NULL))
		goto out;
	zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
	zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;

	vfsp->vfs_data = zfsvfs;
	vfsp->mnt_flag |= MNT_LOCAL;
	vfsp->mnt_kern_flag |= MNTK_MPSAFE;
	vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
	vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
	vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
	vfsp->mnt_kern_flag |= MNTK_NO_IOPF;	/* vn_io_fault can be used */

	/*
	 * The fsid is 64 bits, composed of an 8-bit fs type, which
	 * separates our fsid from any other filesystem types, and a
	 * 56-bit objset unique ID.  The objset unique ID is unique to
	 * all objsets open on this system, provided by unique_create().
	 * The 8-bit fs type must be put in the low bits of fsid[1]
	 * because that's where other Solaris filesystems put it.
	 */
	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
	vfsp->vfs_fsid.val[0] = fsid_guid;
	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
	    vfsp->mnt_vfc->vfc_typenum & 0xFF;

	/*
	 * Set features for file system.
	 */
	zfs_set_fuid_feature(zfsvfs);
	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
	}
	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);

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

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

		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
	} else {
		error = zfsvfs_setup(zfsvfs, B_TRUE);
	}

	vfs_mountedfrom(vfsp, osname);

	if (!zfsvfs->z_issnap)
		zfsctl_create(zfsvfs);
out:
	if (error) {
		dmu_objset_disown(zfsvfs->z_os, zfsvfs);
		zfsvfs_free(zfsvfs);
	} else {
		atomic_add_32(&zfs_active_fs_count, 1);
	}

	return (error);
}

void
zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
{
	objset_t *os = zfsvfs->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,
		    zfsvfs) == 0);

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

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

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

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

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

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

		VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
		    zfsvfs) == 0);

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

		VERIFY(dsl_prop_unregister(ds, "vscan",
		    vscan_changed_cb, zfsvfs) == 0);
	}
}

#ifdef SECLABEL
/*
 * Convert a decimal digit string to a uint64_t integer.
 */
static int
str_to_uint64(char *str, uint64_t *objnum)
{
	uint64_t num = 0;

	while (*str) {
		if (*str < '0' || *str > '9')
			return (SET_ERROR(EINVAL));

		num = num*10 + *str++ - '0';
	}

	*objnum = num;
	return (0);
}

/*
 * The boot path passed from the boot loader is in the form of
 * "rootpool-name/root-filesystem-object-number'. Convert this
 * string to a dataset name: "rootpool-name/root-filesystem-name".
 */
static int
zfs_parse_bootfs(char *bpath, char *outpath)
{
	char *slashp;
	uint64_t objnum;
	int error;

	if (*bpath == 0 || *bpath == '/')
		return (SET_ERROR(EINVAL));

	(void) strcpy(outpath, bpath);

	slashp = strchr(bpath, '/');

	/* if no '/', just return the pool name */
	if (slashp == NULL) {
		return (0);
	}

	/* if not a number, just return the root dataset name */
	if (str_to_uint64(slashp+1, &objnum)) {
		return (0);
	}

	*slashp = '\0';
	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
	*slashp = '/';

	return (error);
}

/*
 * 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 (SET_ERROR(EACCES));
		return (rdonly ? 0 : EACCES);
	}
	return (SET_ERROR(EACCES));
}

/*
 * Determine whether the mount is allowed according to MAC check.
 * by comparing (where appropriate) label of the dataset against
 * the label of the zone being mounted into.  If the dataset has
 * no label, create one.
 *
 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
 */
static int
zfs_mount_label_policy(vfs_t *vfsp, char *osname)
{
	int		error, retv;
	zone_t		*mntzone = NULL;
	ts_label_t	*mnt_tsl;
	bslabel_t	*mnt_sl;
	bslabel_t	ds_sl;
	char		ds_hexsl[MAXNAMELEN];

	retv = EACCES;				/* assume the worst */

	/*
	 * Start by getting the dataset label if it exists.
	 */
	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
	if (error)
		return (SET_ERROR(EACCES));

	/*
	 * If labeling is NOT enabled, then disallow the mount of datasets
	 * which have a non-default label already.  No other label checks
	 * are needed.
	 */
	if (!is_system_labeled()) {
		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
			return (0);
		return (SET_ERROR(EACCES));
	}

	/*
	 * Get the label of the mountpoint.  If mounting into the global
	 * zone (i.e. mountpoint is not within an active zone and the
	 * zoned property is off), the label must be default or
	 * admin_low/admin_high only; no other checks are needed.
	 */
	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
	if (mntzone->zone_id == GLOBAL_ZONEID) {
		uint64_t zoned;

		zone_rele(mntzone);

		if (dsl_prop_get_integer(osname,
		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
			return (SET_ERROR(EACCES));
		if (!zoned)
			return (zfs_check_global_label(osname, ds_hexsl));
		else
			/*
			 * This is the case of a zone dataset being mounted
			 * initially, before the zone has been fully created;
			 * allow this mount into global zone.
			 */
			return (0);
	}

	mnt_tsl = mntzone->zone_slabel;
	ASSERT(mnt_tsl != NULL);
	label_hold(mnt_tsl);
	mnt_sl = label2bslabel(mnt_tsl);

	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
		/*
		 * The dataset doesn't have a real label, so fabricate one.
		 */
		char *str = NULL;

		if (l_to_str_internal(mnt_sl, &str) == 0 &&
		    dsl_prop_set_string(osname,
		    zfs_prop_to_name(ZFS_PROP_MLSLABEL),
		    ZPROP_SRC_LOCAL, str) == 0)
			retv = 0;
		if (str != NULL)
			kmem_free(str, strlen(str) + 1);
	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
		/*
		 * Now compare labels to complete the MAC check.  If the
		 * labels are equal then allow access.  If the mountpoint
		 * label dominates the dataset label, allow readonly access.
		 * Otherwise, access is denied.
		 */
		if (blequal(mnt_sl, &ds_sl))
			retv = 0;
		else if (bldominates(mnt_sl, &ds_sl)) {
			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
			retv = 0;
		}
	}

	label_rele(mnt_tsl);
	zone_rele(mntzone);
	return (retv);
}
#endif	/* SECLABEL */

#ifdef OPENSOLARIS_MOUNTROOT
static int
zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
{
	int error = 0;
	static int zfsrootdone = 0;
	zfsvfs_t *zfsvfs = NULL;
	znode_t *zp = NULL;
	vnode_t *vp = NULL;
	char *zfs_bootfs;
	char *zfs_devid;

	ASSERT(vfsp);

	/*
	 * The filesystem that we mount as root is defined in the
	 * boot property "zfs-bootfs" with a format of
	 * "poolname/root-dataset-objnum".
	 */
	if (why == ROOT_INIT) {
		if (zfsrootdone++)
			return (SET_ERROR(EBUSY));
		/*
		 * the process of doing a spa_load will require the
		 * clock to be set before we could (for example) do
		 * something better by looking at the timestamp on
		 * an uberblock, so just set it to -1.
		 */
		clkset(-1);

		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
			    "bootfs name");
			return (SET_ERROR(EINVAL));
		}
		zfs_devid = spa_get_bootprop("diskdevid");
		error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
		if (zfs_devid)
			spa_free_bootprop(zfs_devid);
		if (error) {
			spa_free_bootprop(zfs_bootfs);
			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
			    error);
			return (error);
		}
		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
			spa_free_bootprop(zfs_bootfs);
			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
			    error);
			return (error);
		}

		spa_free_bootprop(zfs_bootfs);

		if (error = vfs_lock(vfsp))
			return (error);

		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
			goto out;
		}

		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
		ASSERT(zfsvfs);
		if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
			cmn_err(CE_NOTE, "zfs_zget: error %d", error);
			goto out;
		}

		vp = ZTOV(zp);
		mutex_enter(&vp->v_lock);
		vp->v_flag |= VROOT;
		mutex_exit(&vp->v_lock);
		rootvp = vp;

		/*
		 * Leave rootvp held.  The root file system is never unmounted.
		 */

		vfs_add((struct vnode *)0, vfsp,
		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
out:
		vfs_unlock(vfsp);
		return (error);
	} else if (why == ROOT_REMOUNT) {
		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
		vfsp->vfs_flag |= VFS_REMOUNT;

		/* refresh mount options */
		zfs_unregister_callbacks(vfsp->vfs_data);
		return (zfs_register_callbacks(vfsp));

	} else if (why == ROOT_UNMOUNT) {
		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
		(void) zfs_sync(vfsp, 0, 0);
		return (0);
	}

	/*
	 * if "why" is equal to anything else other than ROOT_INIT,
	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
	 */
	return (SET_ERROR(ENOTSUP));
}
#endif	/* OPENSOLARIS_MOUNTROOT */

static int
getpoolname(const char *osname, char *poolname)
{
	char *p;

	p = strchr(osname, '/');
	if (p == NULL) {
		if (strlen(osname) >= MAXNAMELEN)
			return (ENAMETOOLONG);
		(void) strcpy(poolname, osname);
	} else {
		if (p - osname >= MAXNAMELEN)
			return (ENAMETOOLONG);
		(void) strncpy(poolname, osname, p - osname);
		poolname[p - osname] = '\0';
	}
	return (0);
}

/*ARGSUSED*/
static int
zfs_mount(vfs_t *vfsp)
{
	kthread_t	*td = curthread;
	vnode_t		*mvp = vfsp->mnt_vnodecovered;
	cred_t		*cr = td->td_ucred;
	char		*osname;
	int		error = 0;
	int		canwrite;

#ifdef illumos
	if (mvp->v_type != VDIR)
		return (SET_ERROR(ENOTDIR));

	mutex_enter(&mvp->v_lock);
	if ((uap->flags & MS_REMOUNT) == 0 &&
	    (uap->flags & MS_OVERLAY) == 0 &&
	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
		mutex_exit(&mvp->v_lock);
		return (SET_ERROR(EBUSY));
	}
	mutex_exit(&mvp->v_lock);

	/*
	 * ZFS does not support passing unparsed data in via MS_DATA.
	 * Users should use the MS_OPTIONSTR interface; this means
	 * that all option parsing is already done and the options struct
	 * can be interrogated.
	 */
	if ((uap->flags & MS_DATA) && uap->datalen > 0)
#else
	if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
		return (SET_ERROR(EPERM));

	if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
		return (SET_ERROR(EINVAL));
#endif	/* ! illumos */

	/*
	 * If full-owner-access is enabled and delegated administration is
	 * turned on, we must set nosuid.
	 */
	if (zfs_super_owner &&
	    dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
		secpolicy_fs_mount_clearopts(cr, vfsp);
	}

	/*
	 * Check for mount privilege?
	 *
	 * If we don't have privilege then see if
	 * we have local permission to allow it
	 */
	error = secpolicy_fs_mount(cr, mvp, vfsp);
	if (error) {
		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
			goto out;

		if (!(vfsp->vfs_flag & MS_REMOUNT)) {
			vattr_t		vattr;

			/*
			 * Make sure user is the owner of the mount point
			 * or has sufficient privileges.
			 */

			vattr.va_mask = AT_UID;

			vn_lock(mvp, LK_SHARED | LK_RETRY);
			if (VOP_GETATTR(mvp, &vattr, cr)) {
				VOP_UNLOCK(mvp, 0);
				goto out;
			}

			if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
			    VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
				VOP_UNLOCK(mvp, 0);
				goto out;
			}
			VOP_UNLOCK(mvp, 0);
		}

		secpolicy_fs_mount_clearopts(cr, vfsp);
	}

	/*
	 * Refuse to mount a filesystem if we are in a local zone and the
	 * dataset is not visible.
	 */
	if (!INGLOBALZONE(curthread) &&
	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
		error = SET_ERROR(EPERM);
		goto out;
	}

#ifdef SECLABEL
	error = zfs_mount_label_policy(vfsp, osname);
	if (error)
		goto out;
#endif

	vfsp->vfs_flag |= MNT_NFS4ACLS;

	/*
	 * When doing a remount, we simply refresh our temporary properties
	 * according to those options set in the current VFS options.
	 */
	if (vfsp->vfs_flag & MS_REMOUNT) {
		zfsvfs_t *zfsvfs = vfsp->vfs_data;

		/*
		 * Refresh mount options with z_teardown_lock blocking I/O while
		 * the filesystem is in an inconsistent state.
		 * The lock also serializes this code with filesystem
		 * manipulations between entry to zfs_suspend_fs() and return
		 * from zfs_resume_fs().
		 */
		rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
		zfs_unregister_callbacks(zfsvfs);
		error = zfs_register_callbacks(vfsp);
		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
		goto out;
	}

	/* Initial root mount: try hard to import the requested root pool. */
	if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
	    (vfsp->vfs_flag & MNT_UPDATE) == 0) {
		char pname[MAXNAMELEN];

		error = getpoolname(osname, pname);
		if (error == 0)
			error = spa_import_rootpool(pname);
		if (error)
			goto out;
	}
	DROP_GIANT();
	error = zfs_domount(vfsp, osname);
	PICKUP_GIANT();

#ifdef sun
	/*
	 * Add an extra VFS_HOLD on our parent vfs so that it can't
	 * disappear due to a forced unmount.
	 */
	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
		VFS_HOLD(mvp->v_vfsp);
#endif	/* sun */

out:
	return (error);
}

static int
zfs_statfs(vfs_t *vfsp, struct statfs *statp)
{
	zfsvfs_t *zfsvfs = vfsp->vfs_data;
	uint64_t refdbytes, availbytes, usedobjs, availobjs;

	statp->f_version = STATFS_VERSION;

	ZFS_ENTER(zfsvfs);

	dmu_objset_space(zfsvfs->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_bsize = SPA_MINBLOCKSIZE;
	statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;

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

	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
	statp->f_bfree = availbytes / statp->f_bsize;
	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;

	/*
	 * We're a zfs filesystem.
	 */
	(void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));

	strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
	    sizeof(statp->f_mntfromname));
	strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
	    sizeof(statp->f_mntonname));

	statp->f_namemax = ZFS_MAXNAMELEN;

	ZFS_EXIT(zfsvfs);
	return (0);
}

static int
zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
{
	zfsvfs_t *zfsvfs = vfsp->vfs_data;
	znode_t *rootzp;
	int error;

	ZFS_ENTER(zfsvfs);

	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
	if (error == 0)
		*vpp = ZTOV(rootzp);

	ZFS_EXIT(zfsvfs);

	if (error == 0) {
		error = vn_lock(*vpp, flags);
		if (error != 0) {
			VN_RELE(*vpp);
			*vpp = NULL;
		}
	}
	return (error);
}

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

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

	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(zfsvfs->z_parent->z_vfs, 0);
#ifdef FREEBSD_NAMECACHE
		cache_purgevfs(zfsvfs->z_parent->z_vfs);
#endif
	}

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

	rw_enter(&zfsvfs->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 && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
		rw_exit(&zfsvfs->z_teardown_inactive_lock);
		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
		return (SET_ERROR(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(&zfsvfs->z_znodes_lock);
	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
	    zp = list_next(&zfsvfs->z_all_znodes, zp))
		if (zp->z_sa_hdl) {
			ASSERT(ZTOV(zp)->v_count >= 0);
			zfs_znode_dmu_fini(zp);
		}
	mutex_exit(&zfsvfs->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) {
		zfsvfs->z_unmounted = B_TRUE;
		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
		rw_exit(&zfsvfs->z_teardown_inactive_lock);
	}

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

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

	/*
	 * Evict cached data
	 */
	if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
	    !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
	dmu_objset_evict_dbufs(zfsvfs->z_os);

	return (0);
}

/*ARGSUSED*/
static int
zfs_umount(vfs_t *vfsp, int fflag)
{
	kthread_t *td = curthread;
	zfsvfs_t *zfsvfs = vfsp->vfs_data;
	objset_t *os;
	cred_t *cr = td->td_ucred;
	int ret;

	ret = secpolicy_fs_unmount(cr, vfsp);
	if (ret) {
		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
		    ZFS_DELEG_PERM_MOUNT, cr))
			return (ret);
	}

	/*
	 * 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(zfsvfs->z_parent->z_vfs, 0);

	/*
	 * Unmount any snapshots mounted under .zfs before unmounting the
	 * dataset itself.
	 */
	if (zfsvfs->z_ctldir != NULL) {
		if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
			return (ret);
		ret = vflush(vfsp, 0, 0, td);
		ASSERT(ret == EBUSY);
		if (!(fflag & MS_FORCE)) {
			if (zfsvfs->z_ctldir->v_count > 1)
				return (EBUSY);
			ASSERT(zfsvfs->z_ctldir->v_count == 1);
		}
		zfsctl_destroy(zfsvfs);
		ASSERT(zfsvfs->z_ctldir == NULL);
	}

	if (fflag & MS_FORCE) {
		/*
		 * Mark file system as unmounted before calling
		 * vflush(FORCECLOSE). This way we ensure no future vnops
		 * will be called and risk operating on DOOMED vnodes.
		 */
		rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
		zfsvfs->z_unmounted = B_TRUE;
		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
	}

	/*
	 * Flush all the files.
	 */
	ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
	if (ret != 0) {
		if (!zfsvfs->z_issnap) {
			zfsctl_create(zfsvfs);
			ASSERT(zfsvfs->z_ctldir != NULL);
		}
		return (ret);
	}

#ifdef sun
	if (!(fflag & MS_FORCE)) {
		/*
		 * Check the number of active vnodes in the file system.
		 * Our count is maintained in the vfs structure, but the
		 * number is off by 1 to indicate a hold on the vfs
		 * structure itself.
		 *
		 * The '.zfs' directory maintains a reference of its
		 * own, and any active references underneath are
		 * reflected in the vnode count.
		 */
		if (zfsvfs->z_ctldir == NULL) {
			if (vfsp->vfs_count > 1)
				return (SET_ERROR(EBUSY));
		} else {
			if (vfsp->vfs_count > 2 ||
			    zfsvfs->z_ctldir->v_count > 1)
				return (SET_ERROR(EBUSY));
		}
	}
#endif

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

	/*
	 * z_os will be NULL if there was an error in
	 * attempting to reopen zfsvfs.
	 */
	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, zfsvfs);
	}

	/*
	 * We can now safely destroy the '.zfs' directory node.
	 */
	if (zfsvfs->z_ctldir != NULL)
		zfsctl_destroy(zfsvfs);
	zfs_freevfs(vfsp);

	return (0);
}

static int
zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
{
	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
	znode_t		*zp;
	int 		err;

	/*
	 * zfs_zget() can't operate on virtual entries like .zfs/ or
	 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
	 * This will make NFS to switch to LOOKUP instead of using VGET.
	 */
	if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
	    (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
		return (EOPNOTSUPP);

	ZFS_ENTER(zfsvfs);
	err = zfs_zget(zfsvfs, ino, &zp);
	if (err == 0 && zp->z_unlinked) {
		VN_RELE(ZTOV(zp));
		err = EINVAL;
	}
	if (err == 0)
		*vpp = ZTOV(zp);
	ZFS_EXIT(zfsvfs);
	if (err == 0)
		err = vn_lock(*vpp, flags);
	if (err != 0)
		*vpp = NULL;
	return (err);
}

static int
zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
    struct ucred **credanonp, int *numsecflavors, int **secflavors)
{
	zfsvfs_t *zfsvfs = vfsp->vfs_data;

	/*
	 * If this is regular file system vfsp is the same as
	 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
	 * zfsvfs->z_parent->z_vfs represents parent file system
	 * which we have to use here, because only this file system
	 * has mnt_export configured.
	 */
	return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
	    credanonp, numsecflavors, secflavors));
}

CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
CTASSERT(LONG_FID_LEN <= sizeof(struct fid));

static int
zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
{
	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
	znode_t		*zp;
	uint64_t	object = 0;
	uint64_t	fid_gen = 0;
	uint64_t	gen_mask;
	uint64_t	zp_gen;
	int 		i, err;

	*vpp = NULL;

	ZFS_ENTER(zfsvfs);

	/*
	 * On FreeBSD we can get snapshot's mount point or its parent file
	 * system mount point depending if snapshot is already mounted or not.
	 */
	if (zfsvfs->z_parent == zfsvfs && 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(zfsvfs);

		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
		if (err)
			return (SET_ERROR(EINVAL));
		ZFS_ENTER(zfsvfs);
	}

	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(zfsvfs);
		return (SET_ERROR(EINVAL));
	}

	/*
	 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
	 * directory tree. If the object == zfsvfs->z_shares_dir, then
	 * we are in the .zfs/shares directory tree.
	 */
	if ((fid_gen == 0 &&
	     (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
	    (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
		*vpp = zfsvfs->z_ctldir;
		ASSERT(*vpp != NULL);
		if (object == ZFSCTL_INO_SNAPDIR) {
			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
			    0, NULL, NULL, NULL, NULL, NULL) == 0);
		} else if (object == zfsvfs->z_shares_dir) {
			VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
			    0, NULL, NULL, NULL, NULL, NULL) == 0);
		} else {
			VN_HOLD(*vpp);
		}
		ZFS_EXIT(zfsvfs);
		err = vn_lock(*vpp, flags);
		if (err != 0)
			*vpp = NULL;
		return (err);
	}

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

	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
	if (err = zfs_zget(zfsvfs, object, &zp)) {
		ZFS_EXIT(zfsvfs);
		return (err);
	}
	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &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);
		VN_RELE(ZTOV(zp));
		ZFS_EXIT(zfsvfs);
		return (SET_ERROR(EINVAL));
	}

	*vpp = ZTOV(zp);
	ZFS_EXIT(zfsvfs);
	err = vn_lock(*vpp, flags | LK_RETRY);
	if (err == 0)
		vnode_create_vobject(*vpp, zp->z_size, curthread);
	else
		*vpp = NULL;
	return (err);
}

/*
 * Block out VOPs and close zfsvfs_t::z_os
 *
 * Note, if successful, then we return with the 'z_teardown_lock' and
 * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
 * dataset and objset intact so that they can be atomically handed off during
 * a subsequent rollback or recv operation and the resume thereafter.
 */
int
zfs_suspend_fs(zfsvfs_t *zfsvfs)
{
	int error;

	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
		return (error);

	return (0);
}

/*
 * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
 * is an invariant across any of the operations that can be performed while the
 * filesystem was suspended.  Whether it succeeded or failed, the preconditions
 * are the same: the relevant objset and associated dataset are owned by
 * zfsvfs, held, and long held on entry.
 */
int
zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
{
	int err;
	znode_t *zp;
	uint64_t sa_obj = 0;

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

	/*
	 * We already own this, so just hold and rele it to update the
	 * objset_t, as the one we had before may have been evicted.
	 */
	VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
	VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
	VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
	dmu_objset_rele(zfsvfs->z_os, zfsvfs);

	/*
	 * Make sure version hasn't changed
	 */

	err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
	    &zfsvfs->z_version);

	if (err)
		goto bail;

	err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
	    ZFS_SA_ATTRS, 8, 1, &sa_obj);

	if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
		goto bail;

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

	if (zfsvfs->z_version >= ZPL_VERSION_SA)
		sa_register_update_callback(zfsvfs->z_os,
		    zfs_sa_upgrade);

	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);

	zfs_set_fuid_feature(zfsvfs);

	/*
	 * 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(&zfsvfs->z_znodes_lock);
	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
		(void) zfs_rezget(zp);
	}
	mutex_exit(&zfsvfs->z_znodes_lock);

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

	if (err) {
		/*
		 * Since we couldn't setup the sa framework, try to force
		 * unmount this file system.
		 */
		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
	}
	return (err);
}

static void
zfs_freevfs(vfs_t *vfsp)
{
	zfsvfs_t *zfsvfs = vfsp->vfs_data;

#ifdef sun
	/*
	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
	 * from zfs_mount().  Release it here.  If we came through
	 * zfs_mountroot() instead, we didn't grab an extra hold, so
	 * skip the VFS_RELE for rootvfs.
	 */
	if (zfsvfs->z_issnap && (vfsp != rootvfs))
		VFS_RELE(zfsvfs->z_parent->z_vfs);
#endif	/* sun */

	zfsvfs_free(zfsvfs);

	atomic_add_32(&zfs_active_fs_count, -1);
}

#ifdef __i386__
static int desiredvnodes_backup;
#endif

static void
zfs_vnodes_adjust(void)
{
#ifdef __i386__
	int newdesiredvnodes;

	desiredvnodes_backup = desiredvnodes;

	/*
	 * We calculate newdesiredvnodes the same way it is done in
	 * vntblinit(). If it is equal to desiredvnodes, it means that
	 * it wasn't tuned by the administrator and we can tune it down.
	 */
	newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
	    vm_kmem_size / (5 * (sizeof(struct vm_object) +
	    sizeof(struct vnode))));
	if (newdesiredvnodes == desiredvnodes)
		desiredvnodes = (3 * newdesiredvnodes) / 4;
#endif
}

static void
zfs_vnodes_adjust_back(void)
{

#ifdef __i386__
	desiredvnodes = desiredvnodes_backup;
#endif
}

void
zfs_init(void)
{

	printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");

	/*
	 * Initialize .zfs directory structures
	 */
	zfsctl_init();

	/*
	 * Initialize znode cache, vnode ops, etc...
	 */
	zfs_znode_init();

	/*
	 * Reduce number of vnodes. Originally number of vnodes is calculated
	 * with UFS inode in mind. We reduce it here, because it's too big for
	 * ZFS/i386.
	 */
	zfs_vnodes_adjust();

	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
}

void
zfs_fini(void)
{
	zfsctl_fini();
	zfs_znode_fini();
	zfs_vnodes_adjust_back();
}

int
zfs_busy(void)
{
	return (zfs_active_fs_count != 0);
}

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

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

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

	if (zfs_spa_version_map(newvers) >
	    spa_version(dmu_objset_spa(zfsvfs->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 && !zfsvfs->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 && !zfsvfs->z_use_sa) {
		uint64_t sa_obj;

		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->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);
	}

	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
	    "from %llu to %llu", zfsvfs->z_version, newvers);

	dmu_tx_commit(tx);

	zfsvfs->z_version = newvers;

	zfs_set_fuid_feature(zfsvfs);

	return (0);
}

/*
 * 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);
}

#ifdef _KERNEL
void
zfsvfs_update_fromname(const char *oldname, const char *newname)
{
	char tmpbuf[MAXPATHLEN];
	struct mount *mp;
	char *fromname;
	size_t oldlen;

	oldlen = strlen(oldname);

	mtx_lock(&mountlist_mtx);
	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
		fromname = mp->mnt_stat.f_mntfromname;
		if (strcmp(fromname, oldname) == 0) {
			(void)strlcpy(fromname, newname,
			    sizeof(mp->mnt_stat.f_mntfromname));
			continue;
		}
		if (strncmp(fromname, oldname, oldlen) == 0 &&
		    (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
			(void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
			    newname, fromname + oldlen);
			(void)strlcpy(fromname, tmpbuf,
			    sizeof(mp->mnt_stat.f_mntfromname));
			continue;
		}
	}
	mtx_unlock(&mountlist_mtx);
}
#endif