RE: Installing five identical systems...

[Chance Reschke <reschke astro washington edu>]

On Fri, 10 Dec 1999, Greg Lindahl wrote:

> > Using fsk to check the validity of the target isn't a reliable check
> > because the copy is done below the filesystem level.
> 
> The second clause of this sentence is true, but the first clause doesn't go
> with it. Since all the files you care about aren't changing, this technique
> does work, and I've used it about 1000 times with no failures. If you are
> really worried, boot into single user mode, remount the source partition
> read only, and then make the copy. But it's a waste of your time.

The second clause does, in fact, go with the first.  Under certain
circumstances, making a blockwise copy from from a rw mounted disk to
another might be "safe"; however, running fsk isn't necessarily
sufficient to identify copying problems.  

If dd has already copied the inode table for a given set of filesystem
blocks, then, before dd completes writing these blocks, they are
changed, fsck will be unable to identify the copying problem.  This
means that any file which is written to during the copy is vulnerable
to undetectable (with fsck) corruption.

This means what?  That fsck "isn't a reliable check" of blockwise
copies of rw mounted filesystems.  Anyone planing to use dd to
duplicate disks should be aware of this.

> > If you create a minimal system which allows you to boot from a diskette
> > and connect to a network, you can perform this type of disk duplication
> > over your LAN (dd if=/dev/sda | rsh somehost 'dd of=/dev/sda').
> 
> This is impossible on an alpha, because the binaries are too big. Intead I
> wrote a program which compiles down to just 30 kbytes which configs the
> interface and does the copy and uncompress.

When you are ready to release this I'm sure many people would find it
useful.

 - C