sysfs: Serialize updates to the vfs inode

The vfs depends upon filesystem methods to update the
vfs inode.   Sysfs adds to the normal number of places
where the vfs inode is updated by also updatng the
vfs inode in sysfs_refresh_inode.

Typically the inode mutex is used to serialize updates
to the vfs inode, but grabbing the inode mutex in
sysfs_permission and sysfs_getattr causes deadlocks,
because sometimes the vfs calls those operations with
the inode mutex held.  Therefore sysfs  can not use the
inode mutex to serial updates to the vfs inode.

The sysfs_mutex is acquired in all of the routines
where sysfs updates the vfs inode, and with a small
change we can consistently protext sysfs vfs inode
updates with the sysfs_mutex. To protect the sysfs
vfs inode updates with the sysfs_mutex simply requires
extending the scope of sysfs_mutex in sysfs_setattr
over inode_setattr, and over inode_change_ok (so we
have an unchanging inode when we perform the check).
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

fs/sysfs/inode.c

@@ -111,20 +111,20 @@ int sysfs_setattr(struct dentry *dentry, struct iattr *iattr)
 	if (!sd)
 		return -EINVAL;
 
+	mutex_lock(&sysfs_mutex);
 	error = inode_change_ok(inode, iattr);
 	if (error)
-		return error;
+		goto out;
 
 	iattr->ia_valid &= ~ATTR_SIZE; /* ignore size changes */
 
 	error = inode_setattr(inode, iattr);
 	if (error)
-		return error;
+		goto out;
 
-	mutex_lock(&sysfs_mutex);
 	error = sysfs_sd_setattr(sd, iattr);
+out:
 	mutex_unlock(&sysfs_mutex);
-
 	return error;
 }