Using Vital Product Data For Persistent Device Naming
Abstract
We discuss an approach to Linux device naming that is similar to
the current udev/scsi_id combination. Our
approach uses a udev callout, but is much more general
than scsi_id. The first time a device is seen, a
name is allocated from a sequence in the traditional way.
However, this name is remembered by associating it with Vital
Product Data (VPD) elements such as manufacturer name, model
number and serial number (MF, TM, SN), or a world-wide identifier
(WWID), in a database. The approach works for any devices that
can be uniquely identified, rather than just SCSI devices.
Another advantage of this approach is that new devices do not need
to have their names configured before use, so the administrative
impact of adding a large number of new devices is quite low. In
general, we use (run-parts-style) ordered directories of files to
implement naming policy - this approach is known to be useful
because system administrators can modify individual files and
configuration information can be easily added automatically when
installing software packages. Our approach also caters for an
arbitrary number of aliases (or symlinks) for each device. One of
our naming policies is interesting in that it encodes only the
meta-policy - the desired naming policy is built into the database
structure, which can be easily changed by renaming or creating new
directories.
Note
No paper was written for this talk, just the slides that appear
below.
Download
|