On 11th August, I wrote:
> I am at a loss to understand how to use vdump & vrestore to restore an ADVfs
> fileset to the state it was in at the time of an incremental backup. As far as
> I can make out, restoring from an incremental vdump merely brings back all the
> files that are new or changed since the base dump. It doesn't redo any delete
> or rename operations that were done since the base dump.
Unfortunately, nobody was able to enlighten me. I received just one reply,
which was from somebody who appeared to have missed the point.
I therefore conclude that I am correct, and that the vdump/vrestore system
simply does not work properly.
In case anybody has not realised the implications of this, let me give a
specific example of how it can all go wrong:
1. A user creates a file, "prog.c" say.
2. Sysadmin does level 0 vdump.
3. The user does
mv prog.c prog.backup
and then creates a new prog.c
4. Sysadmin does a level 1 vdump.
5. The disc fails.
6. Sysadmin vrestores from the level 0, then the level 1.
7. The user finds that he has TOTALLY LOST the file "prog.backup", and may
justifiably feel aggrieved.
Now the "lost" file "prog.backup" does in fact exist on the level 0 dump as
"prog.c", but the sysadmin has no way of knowing that. Moreover, the user is
unlikely to be able to remember exactly when the renamings were done,
especially as he has no idea when the sysadmin takes the backups.
There are far worse scenarios if people rename directories, or change
directories into regular files or vice versa. Multiply all of this by
hundreds of user and thousands of files and it should be clear that with vdump
and vrestore as they stand, the sysadmin doesn't stand much of a chance.
The really annoying thing is that I am fairly convinced that vdump does in
fact dump enough information to be able to restore the state properly. It's
just that vrestore lacks the mechanism to do it. If the format of the vdump
saveset were documented, I could probably write something that would enable me
to put things back together from the dumps. But I cannot find any such
documentation, and even the SOURCE distribution does not contain these
programs.
The only solution I can think of is to maintain a separate map of the
directory structure (i.e. name->inumber mappings) taken at the same time as
each dump. This would enable reconstruction of the filesytem from the dumps.
But it would be an awful lot of work, and a real pain to have to maintain a
map which merely duplicates information held in a secret format in the saveset.
Martyn Johnson maj_at_cl.cam.ac.uk
University of Cambridge Computer Lab
Cambridge UK
Received on Wed Aug 30 1995 - 16:39:30 NZST