Re: Persistent device names
On Wed, May 17, 2006 at 01:00:45AM +0200, Frans Pop wrote:
To discuss this subject, I've still wanted to set up a discussion with at
least Marco, you, Joey and Colin, but for different reasons have not
gotten around to that.
As I'll be on holiday for a few weeks after Debconf, let me at least list
the issues that I've had in mind and that should be considered when
implementing changes for persistent device naming.
Please add to the list and comment.
- We should like to keep support for devfs and regular device names in
the installer; loosing that would mean no support for 2.2/2.4 installs
(although those are likely to be dropped anyway) and no support for
installs of Sarge and I think may be a problem for debian-edu.
However, if it is unavoidable, we could drop this requirement.
But do 2.4 and 2.2 not use traditional device names after booting into
the installed system?
- Are there likely to be architecture specific issues with persistent
device naming?
- Or with filesystems other than ext2/ext3?
- Persistent device naming is mostly needed to get rid of errors on
reboot. Most common case is where a system has multiple disk
controllers and the order in which their drivers are loaded is
different after the reboot than in d-i leading to a different order
in device names.
Yes, and it would also help with upcoming issues. For instance, the 2.6
kernels seem to move towards supporting ordinary (PATA) IDE via libata,
meaning that hdaX devices would also become sdaX in the future. With
persistent names this wouldn't be a problem.
In addition, persistent names could be good for complex cases like
root-on-lvm-on-crypto-on-md-on-disks.
I've been considering whether it would be a good idea to introduce a new
parameter (for initramfs-tools/yaird) specifying a comma-separated list
of all devices needed to bring up the root dev. Something along the lines of
rootdeps=/dev/sda1,/dev/sda2,/dev/mapper/mainvg-rootlv.
Then the initramfs/initrd could do its best to bring up all those
devices before trying to mount root via the persistent device. This
would help with the complex scenarios...
It wouldn't help with grub/lilo/etc though.
Re,
David
Reply to: