Arch Linux

archboot does support that, official media i don't know.

This sounds ok to me. What do you guys think? We could just add the module to the mkinitcpio config

sounds good to me. but i'm not familiar enough with this to see any disadvantages.

AFAIK this should be already possible with 2009.08-beta1 images (during the change in boot-device detection).
We have had vfat always as module in initrd images, with new boot device detection we don't mount explicit a ext2 FS anymore, we use:
eval $(fstype < /dev/archiso 2>/dev/null)
if [ -n "${FSTYPE}" -a "${FSTYPE}" != "unknown" ]; then
if mount -r -t "${FSTYPE}" /dev/archiso /bootmnt >/dev/null 2>&1; then
...........

So @Robert: would you please test 2009.08-beta1 (http://build.archlinux.org/isos/) if it fits your request without we have to modify anything...

Hi. fstype doesn't detect vfat!

Used archlinux-2009.08-core-i686.img.
Inflated to 2gb vfat partition.
On boot, /dev/sda is internal HDD and /dev/sdb is USB HDD.
/dev/archiso is created, but not mounted. (To be able to enter the prompt mode, I used boot option: archisolabel=INVALID)
*************************
ramfs$ fstype < /dev/sda1
FSTYPE=ext3
FSSIZE=858.....
ramfs$ fstype < /dev/sdb1
FSTYPE=unknown
FSSIZE=0
ramfs$ mount -r -t vfat /dev/sdb1 /bootmnt
ramfs$ echo /bootmnt/isom*
/bootmnt/isomounts
*************************
so sdb1 can be mounted, but fstype not detecting it's type

removed myself from assignees. i don't care about vfat.

So you're doing something crazy and unsupported... what are you expecting to happen?

fstype actually does not detect fat32....

If you want to make the archiso boot on a fat32 device, you're going to have to hand edit the scripts or post patches. I have no interest in doing any work to support this.

Ok, modified. hope someone finds this useful:
***************
eval $(fstype < /dev/archiso 2>/dev/null)
if [ -z "${FSTYPE}" -o "${FSTYPE}" = "unknown" ]; then
echo "WARNING: /dev/archiso found, but the filesystem type is unknown. Assuming vfat"
FSTYPE=vfat
fi

if mount -r -t "${FSTYPE}" /dev/archiso /bootmnt >/dev/null 2>&1; then
if [ -e "/bootmnt/isomounts" ]; then
echo "SUCCESS: Mounted archiso volume successfully."
else
echo "ERROR: Mounting was successful, but the isomounts file does not exist."
exit 1
fi
else
echo "ERROR: Failed to mount /dev/archiso"
exit 1
fi
***************

vfat support actually is in the code now that I look at it.
http://projects.archlinux.org/?p=archiso.git;a=commitdiff;h=fd93e7c9bd614a026f1d661f359012a4c1e080d7

However, I think this problem results from your "archisolabel=INVALID" setting. What is /dev/archiso symlinked too? Is your usb device ACTUALLY labeled "INVALID" ?

Thanks for the link. Looks like it will be solved in next release.
archisolabel=INVALID was just to get interactive console to test fstype. With correct archisolabel /dev/archiso was created, not mounted and kernel panic as the result.

@Gerhard: Was this vfat support added before or after the last release?

after.

Is this bug still relevant? I posted a feature request for vfat and udf on the klibc mailing list, hoping that fstype will be able to recognize file systems a it better: http://www.zytor.com/pipermail/klibc/2009-November/002467.html
Are we going to wait or implement hacky solutions? I think we should switch to the mount way of recognizing fs types which makes use of blkid. Actually, why not just let mount do its thing? Is there any gain in making fstype probe for the fs type when mount is better at that?

@Svenstaro: The current archiso hook support detection of fstype, so works fine when copy the files "*.sqfs isomounts boot/*" to a VFAT partition and setup the grub on it. The only issue is the "archisolabel", it has more than 11 chars allowed for use in VFAT label (archiso_NNNNNNNN), so won't work without changing it. So the steps are setting a dosfslabel on the vfat media and edit boot/grub/menu.lst to match this custom label.

This doesn't quite address my question though. Why do we rely on fstype when mount uses blkid which detects more file systems? For example, fstype can't do udf which is quite annoying for me.

mmm maybe is better to open a separate feature request for it. This topic is centered on vfat support that is done now.

Done: http://bugs.archlinux.org/task/17237

The only thing remain is fix the label for the filesystem: "ARCHISO_NNNNNNN" is longer than 11 chars supported by VFAT.

Is there any objetion in changing this label to ARCH_XXXXXX ?

-export LABEL="ARCHISO_$(pwgen -n 8 1 | tr [a-z] [A-Z])"
+export LABEL="ARCH_$(pwgen -n 6 1 | tr [a-z] [A-Z])"

With this, users only need to edit the label in your target media without changing grub/menu.lst.

I wonder why we even use pwgen here. Why not make the label "ARCH_200912" or something. It fits perfectly and makes far more sense.

Sure!, a trivial patch for this is attached.

push -> close task :)