First, thanks a lot for the responses, especially to Stefan for the
various options. I have further questions on these options to fully
understand them.

quoted text
>> Is there a reason why you can't use filesystem labels or UUIDs?
>
> Indeed.  Shawn, I see these options:
>
>  1. Mount filesystems by UUID or by label.
>    Doesn't work with some filesystems.

The bootloader I'm using is LILO. So I have to specify the boot device
and the root device in lilo.conf. For example, boot=/dev/sda,
root=/dev/sda1. I'm not sure how mounting filesystem by UUID or by
label can help. A little bit elaboration will be more helpful.

quoted text
>  2. Refer to devices by device ID or by device path.
>    Requires a minimal udev environment which gives you
>    /dev/disk/by-id or /dev/disk/by-path, or something
>    similar to udev.  People had things like this already
>    working under Linux 2.4, using programs like devlabel
>    IIRC.  Requires you to set up an initrd with respective
>    scripts in it.
>    By-path only works if you never alter the USB topology
>    and PCI topology.  I.e. prefer by-id.

Every USB thumb drive has a unique ID. So will I have to have a
different root filesystem image if it's install in a different usb
drive if the by-id way is adopted?

quoted text
>  3. Influence the order of disk probing so that your USB
>    disk is the first one.  Works only if there is at most
>    one disk on the USB bus and requires you to wait with
>    insertion or probing of the SATA controller driver (PCI
>    driver) until after the USB disk was probed by sd_mod.
>
>    There may or may not be very simple ways to do this.
>    Simplest would be to configure the SATA controller driver
>    as module and load that module in an rc script after the
>    root filesystem was mounted.  That would actually be much
>    simpler than option 2, so maybe you want to try this
>    before 2.

I think I understand this option. Of course this requires an initrd or
initramfs, which I need to build.

quoted text
>  4. Last resort:  Hack sd_mod to defer any probes until after
>    the first probe of an USB attached disk.  Requires some
>    experience in kernel driver programming.

As you said, this is the last resort, which I also try to avoid. ;-)

-Shawn.
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/