[OpenAFS] Recovering filenames from namei fileserver's /vicepa tree

Jeffrey Hutzelman jhutz@cmu.edu
Mon, 27 Jun 2005 19:02:49 -0400


On Sunday, June 26, 2005 08:53:56 PM -0400 Derrick J Brashear 
<shadow@dementia.org> wrote:

> On Sun, 26 Jun 2005, Richard Godbee wrote:
>
>> outside computer company the department hired to fix/maintain the
>> servers had  overwritten the first hard drive of the three disk software
>> RAID 5 set with  another Linux installation.  Just to make sure I
>> couldn't bring the RAID  array up by marking the first drive as failed,
>> they also fdisk'd the  remaining two drives.  Oh, did I mention there
>> are no backups?
>>
>> Here is where I stand now: I managed to recover november's RAID array,
>> and I  have copied both machines' /vicepa trees to my PowerBook.  At
>> this point,  we're uninterested in reconstructing their AFS cell.
>> They're bringing in a
>
> []
>
>> ** What is the fastest/least-painful way for me to take these two
>> /vicepa  trees and get meaningful pathnames/filenames attached to the
>> files? **
>
> Set up a fileserver and read the files out:
>> - Can I bring up a fresh Linux box, install the OpenAFS server+client,
>> point  that OpenAFS server at one of the /vicepa trees, and copy out the
>> files with  their names intact?
>
> Note that you'll want to be very careful to preserve file ownership and
> modes, otherwise the salvager may "help" you by deleting your data.

File ownership, modes, and the exact filename, including case.
If you have copied the /vicepa trees to an HFS+ filesystem on your 
powerbook, you may already have lost the correct filename case, 
particularly if there happened to be two files whose name differed only in 
case (note we are talking about the names of the fileserver's data files, 
not the users' filenames).

Also note that because of this issue, the encodings used on MacOS X 
fileservers are not the same as those used on other platforms.  So, you 
cannot feed a /vicepa tree from a Linux box to a fileserver running on a 
MacOS box, and expect it to work.

-- Jeffrey T. Hutzelman (N3NHS) <jhutz+@cmu.edu>
   Sr. Research Systems Programmer
   School of Computer Science - Research Computing Facility
   Carnegie Mellon University - Pittsburgh, PA