[Tfug] Cloning systems

[Bexley Hall]

Hi,

I had to play swap-a-disk over the weekend (it's
a game similar to whack-a-mole... except you use
a BIGGER HAMMER!).  This is on a Slowaris box but 
I think the issues apply to all non-toy OS's.

My initial goal was to move a file hierarchy onto
a new drive having a different size and geometry.
I typically use tar to do this:

# #mount new filesystem hierarchy under /mnt
# tar cpf - <names> | (cd /mnt; tar xpf - )

where "tar" is "/usr/local/bin/gtar" in this case
(I never trust Solaris' tar(1) ).

Unfortunately, this messes the bed under Solaris
since hard links aren't supported accross file 
systems so you end up with /mnt/whatever linking
to /whatever (no goodness there).

A workaround is:

# tar cpf - <names> | chroot /mnt tar xpf -

But, this requires several things to exist in 
the chroot'ed file hierarchy just to implement
the command itself (e.g., "tar" as well as the
three or four .so's that it uses -- my gtar is
not statically linked).

While this may not be rocket science, it is, 
nevertheless, a royal PITA.  :<  Thus raising
the question:  is there an *easier* way of doing 
this?

Having built the new file hierarchy, I installed
boot blocks and verified the drive would boot
correctly and "run" as the original did.

Now, the second goal was to *clone* this drive
onto an identical mate (i.e., same size and
geometry).  My first attempt was:

# dd if=/dev/rdsk/c0t0d0s2 of=/dev/rdsk/c0t1d0s2
bs=SxHb

where:

S is number of sectors per track
H is number of heads per track

(note that the system has been in single user mode
for all of this and it is largely quiescent with
me as the sole user at the text console)

Note that the s2 slice is the "backup" slice which
should represent the entire disk (i.e., equivalent
to the "c" slice in a NetBSD/SPARC system)

My thinking was that this should copy the contents
of all of the files as well as all of the supporting
filesystem scaffolding.  E.g., I shouldn't even have
to dick with boot blocks on the "new" drive!

This *appeared* to work -- displaying the number
of records written as identical to the number of
*cylinders* in the geometry (i.e. "as expected").

Of course, it *didn't*! :<  So, where did I go
wrong (i.e. which assumptions are in error)?

Note that I have subsequently cloned the drive just 
by repeating the tar(1) procedure outlined above
so this question is more of an academic nature, now.
But, I'd like to know if my approach was inherently
flawed or if I just got bitten by gremlins in my
*application* of that approach (i.e., made a typo
somewhere).

Thx!
--1138


      ____________________________________________________________________________________
Never miss a thing.  Make Yahoo your home page. 
http://www.yahoo.com/r/hs