Re: Hacking linux-utils for swap label

[Theodore Ts'o <tytso@mit.edu>]

On Thu, Apr 13, 2006 at 10:10:42AM -0700, Greg KH wrote:
> I don't see that requirement anywhere in the vol_id code.  It's a
> stand-alone binary that should work with just about any kernel version,
> as it just talks to the block device directly.  No udev requirement at
> all (it just happens to live in the udev source tree for now, I think
> that's changing as HAL and some other utilities are starting to rely on
> it.)

The blkid library/program does many things, beyond just simply
identifying the filesystem type, label, and uuid given a particular
device.  In particular, the blkid library uses a cache (so you don't
have to scan all of the attached block devices, which in the case of a
system with a truly vast number of san attached storage devices, could
take very long time), so that an application can find a block device
given a particular label or uuid as a search parameter.

That was what was originally requested, and that is something which
vol_id can not do unassisted (but which blkid can do, either from the
command-line or from the C library interface).  

Udev can provide this functionality, if you are running on a modern
distribution that uses udev, and it does so by scanning all block
devices at hotplug/coldplug time.  I'm not entirely convinced that is
the right approach if there are several hundred thousand potential
devices available via a SAN fabric, or via iSCSI --- and to be fair,
blkid probably won't be enough in that environment, either.  But it
provides a nice abstraction interface so that in the future, blkid
could optionally call out to some directory service to find the
location of the iSCSI device.  In contrast, the udev solution to the
find-the-device-given-a-uuid assumes that a symlink has been created,
which makes it more difficult to trap the request to look for a
particular UUID/label.

> Anyway, I don't have any objection to blkid, just trying to point out
> another solution, instead of having the original poster try to reinvent
> the wheel again :)

And I don't have an objection to udev, although I'm not convineced
it's proposed solution for finding block devices given a UUID or label
is the best one....

						- Ted