Quick links:
Currently at:
If you have any comments or suggestions just let me know:
XFS is a journalling filesystem developed by SGI and used in SGI's IRIX operating system. It is now also available under GPL for linux. It is extremely scalable, using btrees extensively to support large and/or sparse files, and extremely large directories. The journalling capability means no more waiting for fsck's or worrying about meta-data corruption.
Just point your web-browser to:
The best way to do this is to checkout the SGI XFS kernel from their CVS tree. How to do this is described at
After that you have two subtrees of importance: linux and cmd. The
firsts one - linux - is a normal linux kernel source tree containing
the XFS code. It is updated to the latest available linux kernel
from time to time but it may be a bit (not much) behind. Just build
your kernel the way you are used to do it and don't forget to enable
XFS and pagebuf under filesystems.
The other tree - cmd - contains all the tools you need
- most important: mkfs.xfs and xfs_repair - just go
to the xfs subdirectory in cmd and type make (maybe followed by make
install to install the binaries). The cmd tree has been restructured
a bit at the beginning of 2001: now you have to go into the various
subdirs (xfsprogs, attr, acl, xfsdump, dmapi) and run the make or the
below described Makepkgs script there. One thing which is also important
to note is that you need to have the e2fsprogs-devel package installed in
order to build the cmd tools - because they require the uuid stuff in
them. You may also build src and binary rpms by running:
./Makepkgs verbose
in the cmd/xfs or cmd/[xfsprogs, attr, ...] directory. The tools also have man-pages which you may consult for interesting options.
There exists also another way to get an XFS ready kernel - you may get kernel patches relative to a official kernel from:
and apply then to the kernel sources the patch is for. This is a good way for all the people who don't want to use CVS or do not have the bandwidth to checkout the whole kernel tree.
A third way to get a XFS ready system and kernel is to use the prepared rpm's and experimental redhat install ISO images from SGI you may find also in the download area.
There you will also find a little HOWTO about the nessecary steps.
Just reboot the new build kernel and create a filesytem on an empty partition
mkfs -t xfs /dev/foo
where /dev/foo is the the partition you want to use (you may have to use the -f option of mkfs -t xfs (which calls mkfs.xfs from the cmd directory which you must have installed) if this partition already contains an old filesystem which you want to overwrite). Now you can mount the filesystem using:
mount -t xfs /dev/foo /somewhere
It is stable and being used in production systems.
The current XFS tree seems to work just fine on ppc now (aside from some trivial compile fixes). It also runs well and is getting sporadically tested on the alpha, sparc64 and ia64. But on all of those platforms it is not as well tested as on i386, but so far there are no major problems on those platforms known. All in all it looks like XFS will be running across a lot of platforms fine soon (with all the platforms above we have 32/64bit and little/big-endian architectures supported. If you run it on a platform not mentioned here please let me know so that i can add it. Also an important note is that XFS is inherently platform independent in the on disk layout - so it should be possible to move a XFS disk from one linux platform to another out of the box.
If you are interested in this you may have a look at
Especially the ppc subdir there contains mostly quite up to date merged XFS and ppc trees which should work on ppc machines quite fine.
User quotas are supported, group/project quotas are work-in-progress.
To use quotas with XFS, you need to enable linux quota support, and
XFS quota support when you configure your kernel. You also need to specify
quota support when you mount.
The linux quotatools now include XFS quota controls. You can get the quota
utilities at their sourceforge website
http://sourceforge.net/projects/linuxquota/.
There are problems with the 1.0 release quota support. Upgrade to a later kernel version
to make it work. The next release (1.0.1) should have this fixed.
xfsdump and xfsrestore are now in the CVS tree. The tape format is
the same as for xfsdump and xfsrestore on IRIX and dump tapes should
be interchangable between systems.
The tape format is not the same as the classic Unix dump but should
work fine with tools like Amanda. Dumps produced with standard
other dump programs should be able to be restored onto an XFS filesystem
using the coresponding restore program.
This depens on where you install LILO.
For MBR installation: Yes.
For root partitions: No, because the XFS superblock goes where LILO
would be installed. This is to maintain compatibility with the Irix
on-disk format. This will not be changed. Putting the Superblock on
the swap partition is reported to work.
No - so far there is no direct support in GRUB to boot from an XFS
filesystem but you may still boot an mostly XFS system with a small
ext2 /boot partition with all the GRUB stuff on it using GRUB.
There is work in progress for making GRUB XFS aware.
Yes XFS should run fine on top of LVM. If you plan to do so please keep
in mind that the current XFS cvs tree (and also the XFS 1.0 previews)
already contains the lvm beta6
code and some tweaks for XFS - currently it is recommended to stay
with this version instead of the current lvm version (beta7 at the
moment) because there are some logistical problems around that
version. Martin K. Petersen is keeping an eye on the lvm in the
XFS tree as soon as there is a recommended new version of it i think.
Keep one thing in mind with XFS on lvm: mounting snapshots of lvs
containing XFS filesystem does not work due to the journaling
nature of the filesystem (but a fix for this is on the TODO list).
Yes. If you are using a 2.4.2 based XFS kernel you need to apply Jens
Axboes loop-xfs-7c fix for it to work (the fix is for a problem in
2.4.2 and has nothing really to do with XFS). You may get this patch from
Yes. A document describing how to convert your system to XFS is currently
being worked on. The current state of this document you may find at
You might also have a look at my little stand-alone
environment for converting and maintaining a root XFS filesystem at
Linux native filesystem (83).
If you get something like:
you either don't have XFS compiled into the kernel (or you forgot
to load the modules) or you did not use the "-t xfs" option on mount
or the "xfs" option in /etc/fstab.
Yes - the current XFS tree and the latest released previews contain
fixes which should allow you to run XFS on top of any md RAID level.
Yes - hardware RAID is like a normal disk to XFS (like to any other
filesystem) and this should not be a problem.
Yes. To get good performance make shure to use an XFS tree from after
mid-march 2001. There were some important fixes for useable NFS-performance.
So far there are no more known problems with XFS and NFS since then.
An XFS filesystem may be enlarged within a partition using
xfs_growfs. You need to have free space after this partition to do so.
Remove partition recreate it larger with exact same starting point.
run xfs_growfs to make the partition larger. This operation is dangerous to your data
Back up your filesystem before using this tool.
Yes. The on-disk format of XFS is the same on IRIX and Linux. Obviously,
you should back-up your data before trying to move it between systems.
Filesystems must be "clean" when moved (ie unmounted correctly). If
plan to use IRIX disks on linux keep the followng things in mind: the
kernel needs to have SGI partition support enabled (to find in the
File systems -> Partition Types submenu of a "make menuconfig"), there
is no XLV support in linux, so you won't be able to read IRIX disks
using the XLV volume manager, also not all blocksizes available on
IRIX are available on linux for now (only the pagesize of the
architecture: 4k for i386, ppc, ... 8k for alpha, sparc, ... is
possible for now).
The only real caveat here is that at the moment XFS on linux only supports
filesystems where the disk block size equals the page size. Support
for other block sizes available on IRIX is work-in-progress.
Recompile your kernel with gcc 2.91.66 instead of gcc 2.95 this should
fix your problem. On RedHat 7.x systems you get egcs via the kgcc(7.0)
or compat-egcs(7.1) package. See the
comments in the top level Makefile of the kernel for more information
about this.
Yes. So far there were some problems reported with kernels built with
gcc 2.95 which were solved by compiling it with egcs 2.91.66. So for now please
use version gcc 2.91.66 (aka egcs 1.1.2) to build your XFS kernel. If you
are using a debian or SuSE based system this means that you may have to
find and install this egcs version. Please note that the problems with
gcc 2.95 seem to be restriced to the i386 platform - on the ppc it works
just fine with 2.95 for instance. On the other hand the gcc 2.95.3 (20010125) from
debian unstable seems to work.
The RedHat included gcc 2.96 compiler still has issues ranging from oopses
to hangs and fs corruption. If you have spare time to figure out what is going wrong
you may send reports to the list. If you are building kernels for your production
environment USE kgcc if your data is worth something to you.
This is the only compiler which has been thoroughly tested and the original XFS port
was made using this compiler. Later kernels then 2.4.6-prex may work for you but it
is untested teritory.
If you are using gcc 3.0 and it gives problems or does not compile, drop a note on
the list with the oops and ksymoops output. We will be working on getting XFS fully
functional with gcc 3.0.
If you are looking for egcs(gcc 2.91.66) compilers for your system you can find it
in the following places. Under RedHat 7.0 systems it is located in the kgcc
package on 7.0 and for 7.1 it is in the compat-egcs package.
On Linux-Mandrake systems it's located in the egcs package.
Some of them are other interesting tools like: db - xfs_db is a xfs
filesystem debugger (working) , copy - xfs_copy is tool for
effectively copying one filesystem to another device (not yet ported),
fsr - xfs_fsr is a defragmenter for xfs (working),
repair - xfs_repair is the consistency checker for an xfs
filesystem (working). As already mentioned earlier: the cmd structure
has changed a bit at the beginning of 2001 - now it all looks a bit
clearer i think (modeled a bit after the ext2 tools structure). The
only two other subdirs are: quota - a quota-tools tree with the XFS
additions in it and xfstests - the SGI XFS stress test suite.
This is because there is a syscall conflict which results in this
problem. Please use the latest XFS code from the cvs tree which has
this fixed. Eventually the beta will be rerolled for that.
Don't run those commands on a mounted filesystem - in this case it is
normal that you see inconsistencies. Never run xfs_repair on a mounted
filesystem without the "-n" switch at all. If you ran the commands on
an unmounted filesystem then there seems to be something wrong with
the filesystem.
First dbench: this is a very metadata intensive benchmark which might
be a little limited by the default log size. You will getter better
results by creating the filesystems with
Q: Do quotas work on XFS?
Q: Is there any dump/restore for XFS?
Q: Does LILO work with XFS?
Q: Does GRUB work with XFS?
Q: Can I run XFS on top of LVM?
Q: Can I use XFS on loopback block devices?
Q: Can XFS be used for a root filesystem?
Q: What partition type should I use for XFS?
Q: Mounting the XFS filesystem does not work - what is wrong?
mount: /dev/hda5 has wrong major or minor number
Q: Does XFS run with the linux software RAID md driver?
Q: Does XFS run on hardware RAID?
Q: Can I run NFS on top of an XFS partition?
Q: Is there a way to make a XFS filesystem larger or smaller?
You can NOT make a XFS partition smaller. The only way to do so would
be a complete dump, format and restore.
Using XFS filesystems on top of LVM makes this a lot easier.
Q: Will I be able to use my old IRIX XFS disks on linux?
Q: Processes are hanging when acting on an XFS filesystem (for
instance in sv_wait) - what is this?
Q: Are there any known issues about gcc for compiling the XFS kernel tree?
Q: What are all those other things in the cmd/xfs directory?
Q: Why my RedHat 7.0 system does not boot with the XFS beta kernel?
Q: My filesystem is ok - but xfs_check or xfs_repair shows errors - whats wrong here?
Q: I am getting very bad dbench (etc.) results with XFS - whats wrong here?
mkfs -t xfs -l size=32768b -f /dev/device
This creates a bigger logspace. Also it is a good idea to mount meta-data intensive filesystems with
mount -t xfs -o logbufs=8,logbsize=32768 /dev/device /mountpoint
Note that in some earlier versions you may only use 4 logbufs as a maximum. Using more logbufs can fail if your system has not enough ram. Using 8 logbufs on a machine with 128MB ram wil probably fail. Also since mid-march there are some changes in the tree which improve the overall dbench performance a bit. Have a look for the xfs.txt file in the Documentation/filesystems subdirectory of your kernel sources for those and other XFS mount options.
In general you should get about the same or better performance values with XFS in various benchmarks. One thing XFS is usually bad at is removing large amounts of files (rm -rf or bonnie++).
This is fixed in the current code (mid-march). But also before this is normal and harmless - it only takes a bit of time (on boot and halt with SuSE startup scripts - only on halt with RedHat based scripts).
It produces the following error.
SCSI subsystem driver Revision: 1.00
PCI: Found IRQ 11 for device 00:0c.0
scsi0: Adaptec AIC7XXX EISA/VLB/PCI SCSI HBA DRIVER, Rev 6.1.13
aic7890/91: Ultra2 Wide Channel A, SCSI Id=7, 32/255 SCBs
ahc_intr: AWAITING_MSG for an SCB that does not have a waiting message
SCSIID = 7, target_mask = 1
Kernel panic: SCB = 3, SCB Control = 40, MSG_OUT = 80 SCB flags = 6000
In interrupt handler - not syncing
The Adaptec driver in 2.4.5 and later needs to have the following
selected in order to work.
[ ] Build Adapter Firmware with Kernel Build (NEW)
It spits out the following error during compile
gcc -I/usr/include -ldb1 aicasm_gram.c aicasm_scan.c aicasm.c aicasm_symbol.c -o aicasm aicasm_symbol.c:39: db1/db.h: No such file or directory make[5]: *** [aicasm] Error 1The Adaptec driver in newer 2.4.2 kernels and later need to have the db headers. These can be found in the db-devel packages.
Or it produces the error that it cannot find two header files (.h) in the build directory. In this case you should activate the following option.
[ ] Build Adapter Firmware with Kernel Build (NEW)Make sure to run make mrproper after selecting this option.
Before trying to compile the cvs tree do a
make mrproperIf you also copied your own .config into the tree make sure to run
make oldconfig
Yes, XFS supports files larger then 2GB. The large file support (LFS) is
largely dependent on the C library of your computer. Glibc 2.2 and higher has
full LFS support. If your C lib does not support it you will get errors that
the valued is too large for the defined data type.
Userland software needs to be compiled against the LFS compliant C lib in
order to work. You will be able to create 10GB files on non LFS systems but the
tools will not be able to stat them.
Distributions based on Glibc 2.2.x and higher will function normally. Note that
some userspace programs like tcsh do not correctly behave even if they are compiled
against glibc 2.2.x
You may need to contact your vendor/developer if this is the case.
Here is a snippet of email conversation with Steve Lord on the topic of the maximum filesize of XFS under linux.
I would challenge any filesystem running on Linux on an ia32, and using the page cache to get past the practical limit of 16 Tbytes using buffered I/O. At this point you run out of space to address pages in the cache since the core kernel code uses a 32 bit number as the index number of a page in the cache. As for XFS itself, this is a constant definition from the code: #define XFS_MAX_FILE_OFFSET ((long long)((1ULL<<63)-1ULL)) So 2^63 bytes is theoretically possible. All of this is ignoring the current limitation of 2 Tbytes of address space for block devices (including logical volumes). The only way to get a file bigger than this of course is to have large holes in it. And to get past 16 Tbytes you have to used direct I/O.Which would would mean a theoretical 8388608TB file size. Large enough?
At least one user has experienced problems when resizing his lvm volume multiple times. The workaround is to run xfs_repair after each xfs_growfs. They are not able to replicate this at SGI on the moment.
The 2.4.5 tree from beginning of june had a problem that permissions over NFS were reset. This is fixed in the development tree.
No, there is no relation. Newer utilities only have fixes and checks the previous versions might not have. These are the same utilities that have been used under Irix for years so they have been well developed.
XFS is a huge amount of kernel code which means that your kernel is probably
to big to fit on a boot floppy together with a inital ramdisk and scsi drivers.
There is patch available for mkbootdisk to properly format a floppy with a larger size.
Your mileage may very on booting overformatted floppy's of 1.68 or 1.72MB.
The patch can be found at
http://iserv.nl/files/xfs/mkbootdisk.large.patch Use the /dev/fd0u1680 for making
a 1.68MB floppy.
Another method is building XFS as module but this is probably more hassle then trying to
boot a superformatted floppy.
Kelly Eicher has made a boot floppy set available on his homepage http://www.astro.umn.edu/~carde These are very helpful and easy to use in migrating or repairing a system.
You can backup a XFS filesystem with utilities like xfsdump and standard tar for standard files. If you want to backup acls you will need to use xfsdump. This is the only tool at the moment that supports backing up of acls. Support for XFS and acls is underway at several commercial backup tools. xfsdump can be made to work with amanda.
Good question, there still is a decent amount of work to be done to make the
xfs patch less intrusive on the standard Linus kernel. There are still some points
that need to be resolved before it can be sent to Linus. Maybe if we
are lucky it might even be included at some point in the 2.4 series.
We are sending patches to Linus but nothing is coming back yet ;)
XFS also interacts with the VM very closely. It will probably be included in
the 2.5 tree. In the mean time there will be patches made available
for patching a Linus kernel with XFS.
The easiest way to get the "latest and greatest" is by checking the out the xfs
development tree by cvs.
XFS supports locking but it is not known to VMware. It has been reported to VMware but the status is unknown at this point. If you follow the instructions that VMware gives when starting up you should be fine.
Take a look at http://acl.bestbits.at/pre and get the posix 1003.1e draft.
There are some papers available on the SGI XFS project page (mentioned above) and I have set up an LXR indexed version of the SGI XFS CVS tree at:
which might help to find an easy and good way through the sources. I plan to keep this tree automatically updated to the current SGI XFS CVS version on a daily basis within the next days. If anyone has pointers to other XFS related docs - just send me a mail (address - see above).
So for those of you with money to spare:
http://www1.fatbrain.com/asp/bookinfo/bookinfo.asp?theisbn=1400512638&vm=
And for those of you who are cheapskates:
http://techpubs.sgi.com:80/library/tpl/cgi-bin/download.cgi?coll=0530&db=bks&pth=/SGI_Admin/XFS_AG
I think this is externally accessible, and you can get the pdf form of the
book from here for the price of a 500K download.
Zoltan Kraus
I have uploaded new xfs floppies under the BETA directory on my site, I have not tested them, but they include the
preliminary code for a menu driven XFS install written by damasta. Someone give them a spin, I'd like to see if they
work properly....
You can find the floppies at the following site.
http://markybobdeb.sourceforge.net/zoltan/disks/xfs/BETA
Yes, the nVIDIA drivers work fine on XFS systems. Be sure to use the
1.0-1251 release or later of the nVIDIA linux drivers.
These are known to work starting from 2.4.0-test5 and later.
The FAQ maintainer (me) uses these nVIDIA drivers on his own XFS system so
you will get notice which version does not work.
The drivers can be found at http://www.nvidia.com/
If you are using the 1.0 release of XFS I suggest disabling devfs. Devfs can be disabled by editing your lilo.conf and inserting the following line
append="devfs=nomount"This will prevent devfs from interfering. Other people suggest iserting the following magic in /etc/rc.d/rc.local (on RedHat systems).
major=195
for i in 0 1 2 3; do
devfile="/dev/nvidia$i"
rm -f $devfile
if ! mknod $devfile c $major $i || ! chmod 0666 $devfile; then
echo "Couldn't create device \"$devfile\"."
exit 1
fi
done
devfile=/dev/nvidiactl
rm -f $devfile
mknod $devfile c $major 255
chmod 0666 $devfile
This will create the nvidia devices with each boot.
There are multiple slackware boot disks available. The first is
http://village.flashnet.it/users/fn048069/linux-xfs.html. The page will explain what to do and what the disks
contain. These author of these disks can be contacted at the following address daedalus@freemail.it
The second is located at http://slackjfs.setcom.bg/ which has slackware
install disks for most available Journaling filesystems including XFS, ReiserFS and JFS.
Patches for getting a kernel with XFS can be found on the FTP server in ftp://oss.sgi.com/projects/xfs/download/patches. The most patches here are for the linus tree, patches for the -ac (Alan Cox) series are not available. Alan Cox can somtimes produce more then 3 kernels at a day which is a pace the SGI people can not keep up with. If you want to make unofficial patches available for the -ac series and think you can keep up with the pace drop us a note on the list.
The patches you can find here are provided for recent linus kernels to
either seed a cvs tree which makes it faster to make your own local CVS
tree or patch a linus tree in to a XFS capable kernel tree.
These patches are remade every once in a while to have a patch that includes fixes that are backported from the CVS.
Follow the following link to look what version is in the CVS tree. The top level Makefile for the current development CVS version.
There are a few mountoptions to influence a XFS filesystem described below.
At mount time, there are really three options which will make a difference o biosize (in the released tree the default is 16 or 64K, in the development tree the default is 12 or 4K). Making this larger may help some applications, it will hinder others. o osyncisdsync - indicates that O_SYNC is treated as O_DSYNC, which is the behavior ext2 gives you by default. Without this option, O_SYNC file I/O will sync more metadata for the file. o logbufs=4 or logbufs=8, this increases (from 2) the number of in memory log buffers. This means you can have more active transactions at once, and can still perform metadata changes while the log is being synced to disk. The flip side of this is that the amount of metadata changes which may be lost on crash is greater.
In the original 1.0 release the default biosize was 16 or 64K which resulted in a possibility to make a _really_ large rpm database when rebuilsing it. This has been fixed in versions after 2.4.4. This will be fixed in the 1.0.1 version.
Yes you can! Here is an example to fetch a release iso.
rsync -avu oss.sgi.com::xfsftp/Release-1.0.1/installer/RH7.1-SGI-XFS-1.0.1.iso .Note that using compression may give problems.
Yes, this should work without to much trouble. Someone already has patches for making these two work together.
I'm running such a setup for several months already. This only works, because Steve Best added a little tweak upon my request to get this going because XFS modifies some type declaration that JFS depends on. I'm maintaining patches with XFS plus JFS on my ftp server: ftp://ftp.uni-duisburg.de:/Linux/filesys/or see here http://oss.sgi.com/projects/xfs/mail_archive/0107/msg00025.html
For people that want a searchable archive you can find it at the following url. http://marc.theaimsgroup.com/?l=linux-xfs&r=1&w=2
mkfs.xfs: warning - cannot set blocksize on block device /dev/hde1:You are doing nothing wrong. XFS is using an extra ioctl to set the block size of the device, it is not implemented for this device. However, a recent review of the code seems to show that we do not need actually need the ioctl anymore. Since the filesystem was made anyway, this is a message you should be able to ignore. Try mounting the filesystem and see what happens.
Things to include are what version of XFS you are using, if this is
a CVS version of what date and version of the kernel. If you have
problems with userland packages please report the version of the
package you are using. This also aplies to what distribution you are running.
If you get fs corruption report in what files and which applications. If you
experience an oops, please run it through ksymoops so it is actually understandable
for the developers what is going wrong.
If you see strange errors in the log about things failing and see lockups be sure
to report what hardware you are on and in what configuration.
Be sure to read the How to Report Bugs Effectively document.
The error 990 stands for EFSCORRUPTED which usually means XFS has
detected a metadata problem on the disk and has shut it down.
There should be a console message when this happens. Suspect hardware problems
or serious software problems. People using HP servers have noticed this corruption with a AMI megaraid
based raid controller.
XFS will slow down doing allocations when it is really full, you are no where near full, 99.x% full is nearly full. Basically XFS chops the filesystem into allocation groups (1 to 4 Gbytes each), free space is managed independently in each of these. The slowdown happens when you have to scan through lots of allocation groups looking for space to extend a file. There is an in memory summary structure which tells you if it is worth even looking in an allocation group, so it is not a major slowdown - unless you have lots of parallel allocation calls going on at the time.
There is no undelete in XFS, in fact once you delete something, the chances are the space it used to occupy is the first thing reused. Undelete is really something you have to design in from the start.
<& xfsTemplate,bottom=>1 &>