[plug] Long-winded dev names

[Daniel Pittman]

Bernd Felsche <berfel at innovative.iinet.net.au> writes:
> Jim Householder <nofixed at westnet.com.au> wrote:
>
>>The following line is extracted from Rob Dunne's  "file not found at
>>start of boot sequence" message
>
>>root=UUID=1b672a8f-71a1-4d38-86a4-c4d43fb859f9 ro single
>
>>It appears that he is using SUSE or a derivative.
>
>>Can anyone tell me what the advantage or benefit is obtained by using
>>such a long-winded device name?
>
> It's a unique device ID.

It is a unique *filesystem* ID.  That distinction matters because you
can clone a filesystem and get duplicate ID values, while cloning a
physical device is harder. ;)


Moreover, the reason it matters: more and more modern hardware is
connected to one bus or another that *only* support discovery, not
simple enumeration.  The most common example, USB: it is absolutely
impossible, given the design of the bus, to provide stable device
enumeration.

(Even good old PCI is no longer assured stable: various netbook hardware
 uses PCIe hotplug support to add and remove devices dynamically from
 the PCI bus in the machine.)


So, the net effect is that the kernel cannot assure you that sda will
ever be the same device over two boots — today it *mostly* works, but it
will get less and less reliable every year.

Using a stable identifier for what you care about (the root filesystem)
rather than something unstable (whatever random SCSI node was assigned
the disk this boot) improves life.


Finally, why the UUID and not, say, the filesystem volume label?
Because the UUID is designed to be unique, while the label is commonly
duplicated; every RHEL4 box in the universe, more or less, has a
partition labeled '/' on disk.

> In theory, it means that you can shuffle the discs around willy-nilly
> and everything will still mount properly.  In practice, you sometimes
> get the above; especially across kernel versions and drivers.

I am yet to see any distribution released in the last few years fail to
mount the root partition correctly using a UUID or LABEL mount.

Presumably, though, you have had it fail; can you let me know where,
please — I like to know which distributions or circumstances might lead
to my discovering this the hard way. ;)


> Myself, I prefer volume names. Volume names can also be copied when
> you replace dodgy HDD.

If you clone the filesystem rather than the content you get the same
effect from a filesystem UUID — but, generally, LVM volume names are a
nice choice, too.

> Recovery from the UUID is to boot into rescue mode from CD/DVD,
> figure out which is the root partition (e.g. /dev/sda5), and set
> that at boot-time.

A much better approach is to boot from rescue mode, then run:

    ls -l /dev/disk/by-uuid/

Identify which is your root partition, and then update the UUID.
(Not that this was the problem that the OP had, but whatever. :)

Regards,
        Daniel