Re: tar -c -G -N <date> query

[Tom Crane]

> Hello Tom.
Hi,

> 
> On Wed, Jul 16, 2003 at 07:00:47PM +0100, Tom Crane wrote:
> > One question: Can I fabricate a snapshotfile? The format appear to be
> 
> though I don't have experiences with it myself, I think I heard it's
> possible.
> 
> > time_t format timestamp
> > device_number directory-inode-number directory-spec
> > device_number directory-inode-number directory-spec
> > device_number directory-inode-number directory-spec
> > etc.
> 
> > If I were to invent dummy device_numbers (or obtain the device-->numbers
> > mapping??), read the directory inode numbers and directory names out of my
> > *full.tar.bz2 tarfiles, and write them to a file with the above format,
> > would this do? 
> 
> I think it should work.  You can try the method on a smaller test tree.
> 
> But it's not enough to invent device numbers.  These numbers are unique
> for each device listed when you run ``mount''.
> (They are probably calculated from major and minor device numbers, as
> listed by eg. ``ls -l /dev/hda1''.)

I did wonder about that... but it seemed too nasty to be true --  I mean 
to tie the tarfile to the H/W-particulars rather than the mount-point for 
example. I've not tried it but I wonder if this means that a tarfile
created with listed-incremental could not be extracted on to a different
IDE HD partition, or a device on a different bus altogether. Surely this
can't be true...? ;-| 

> 
> I'd suggest that you run tar with ``-g'' on a small portion of each
> tree mounted by ``mount'' to get the mapping.  Then you can fabricate
> the file.

Just checked it. It does work, the mapping is device_number = major_id*256 + 
minor_id
 
> (BTW: if the device is a nfs device, the device numbers are not significant
> and you can use any dummy number.)

Agreed.

> 
> Good luck,
>       Stepan Kasal

Cheers
Tom.

Ps. CCed: address@hidden