You must take the following issues into account when you create and administer XFS filesystems with XFS for Linux pre-release 0.9.
Currently XFS is limited to filesystems smaller that 2 terabytes. This is due to limitations in the Linux block device I/O layers.
The XFS team is working with Linux developers to improve the Linux I/O layers. The improvements will include the support necessary to exceed 2 Tbyte filesystems.
Due to known problems with later versions of gcc, gcc version 2.91.66 (aka egcs-1.1.2) must be used when compiling XFS and the associated kernel. This is the default compiler on RedHat 6.2 systems. Other distributions may be using a later version. The easiest way to check what version is in use is by running gcc -v.
The version of glibc primarily used for development and testing has been 2.1.3. Earlier versions should be considered "status unknown".
Red Hat 7.0 ships with 2 compilers, "gcc" and "kgcc." kgcc is essentially an older gcc (egcs-2.91.66) which should be used to compile the kernel. To use this compiler, comment out the following line in the top level kernel makefile:
CC = $(CROSS_COMPILE)gcc -V egcs-2.91.66And uncomment the following line:
CC = $(CROSS_COMPILE)kgcc
A preliminary set of system calls to support extended attributes has been added. This interface will most likely change as a consensus on appropriate interfaces is formed in the Linux community.
The extended attribute system call is attrctl(2). Library calls implement the various extented attribute operations: attr_get(3), attr_getf(3), attr_list(3), attr_listf(3), attr_multi(3), attr_multif(3), attr_remove(3), attr_removef(3), attr_set(3), attr_setf(3). Extended attributes can be manipulated using the command attr(1). See the associated man pages for details.
The access control list system calls are acl_get(2) and acl_set(2). Library calls implement many of the main Posix 1003.1e proposed functions, such as: acl_copy_ext(3), acl_delete_def_file(3), acl_dup(3), acl_free(3), acl_from_text(3), acl_get_fd(3), acl_get_file(3), acl_set_fd(3), acl_set_file(3), acl_size(3), acl_valid(3). Access control lists can be manipulated using the command chacl(1). See the associated man pages for details.
The Linux version of dump and restore can be used to migrate data from a non-XFS volume to an XFS volume. Linux dump, however, cannot be used to dump an XFS volume as it uses ext2 library functions.
xfsdump and xfsrestore can be used to migrate an IRIX XFS volume to a Linux XFS volume and vice versa.
xfsdump and xfsrestore should be used for backup and restore of XFS volumes. xfsdump and xfsrestore currently support file extents but not extended attributes. DMAPI is supported in XFS Linux, but a DMAPI-managed filesystem cannot be moved between IRIX and Linux, by any means including dump/restore.
XFS supports two directory formats, referred to as version 1 and version 2. For Linux XFS, the version 2 directory format is strongly recommended. Due to the way glibc getdents interfaces with the kernel implementation, XFS version 1 directories can exhibit problems, usually missing directory entries. By default mkfs will create version 2 directory format filesystems.
For the current release of XFS, the filesystem block size is limited to the size of a memory page. On a x86 architecture that size is 4 Kbytes.
Note that files systems created on an IRIX/MIPS platform must have been created with a 4 Kbyte block size in order to be mounted on a ia32 Linux system. File systems not created with a 4k block will fail to mount with an error indicating the mis-match.
The unwritten extent feature is not yet operational and should be turned off when mkfs is executed, using the following argument to the mkfs command:
-d unwritten=0
Currently this is the default setting for mkfs.xfs on Linux.
Specifying O_SYNC on open will not have the desired effect. write system calls can return before data is on disk. fsync will force data out to disk. This will be fixed in a later beta.
It is scheduled for a later beta release.
User quotas are now functional. XFS uses journaling to provide a higher level guarantee of quota consistency than other filesystems, and as such requires a modified version of the standard quota utilities. These tools can currently only be obtained from the "cmd/quota" directory in the XFS CVS tree.
Group quotas are not yet supported.
It is possible to remount an XFS filesystem to change its permissions from read/write to read-only. This support is needed for XFS to function as a root filesystem.
Note: This operation will most likely result in a hung file system if performed on a busy partition.
2.4.0 has several bugs in the Multiple Disk driver which have been fixed in 2.4.1. Furthermore, a few problems have been identified in the interaction between XFS and MD which could cause corruption during resync after a disk failure. These will be fixed in the next release.
2.4.0 features an upgrade to the Linux Logical Volume Manager. However, there are several known issues with the included version. If you intend to use LVM, it is recommended you recompile your kernel with the latest patches from The LVM Development Team. You will also need the 0.9 userland utilities included in this patch to access your LVM partitions, as the new driver isn't backwards compatible with the 0.8 tools.
mkfs.xfs does not currently support querying the stripe size/stripe width from LVM 0.9 volumes. You'll have to specify these by hand using the sunit and swidth parameters when you create a new file system. This will be fixed in our next release.