Looking inside device files

[Tony Abernethy]

Andre Majorel wrote:
> 
> On 2007-08-21 05:55 -0400, Matt McCutchen wrote:
> > On 8/20/07, Andre Majorel <aym-cnysr at teaser.fr> wrote:
> > > Is there a way to make rsync look inside device files ? 
> The goal is 
> > > to copy the contents of a block device to a regular file 
> > > incrementally.
> > >
> > > Short of that, even just getting it to dump the block 
> checksums of 
> > > the content a block device would help (if I know which blocks are 
> > > out of sync, I can fix them with dd).
> > 
> > Are the block device and the target regular file on the 
> same machine?
> 
> They could be. Or not. Doesn't matter.
> 
> > Your remark about dd suggests that they are.
> 
> Not necessarily.
> 
>   dd if=myfile skip=100 | ssh user at host "dd of=myfile 
> seek=100 conv=notrunc"
> 
> > If so, why do you want an incremental copy?  It does at 
> least as much 
> > disk I/O as an ordinary copy (since it reads the entire 
> basis file), 
> > and the reduction in data passed from the sender process to the 
> > receiver process is irrelevant.
> 
> Sometimes, you need to copy a whole block device and you 
> can't just cat it to wherever you want, go out to lunch and 
> expect it to work.
> 
> The destination could be unreliable. The network could be 
> unreliable. The source could trigger DMA timeouts and lock up 
> the bus, forcing you to reboot. Many things can and do go wrong.
> 
> In those situations, you need to be able to verify that the 
> copy is identical to the original and where's it's not, fix 
> it (NOT start over, because if it failed the first time, it 
> may fail the second time).
> 
> rsync already has most of the mechanisms to do that easily 
> and efficiently. It would be even better if it could complete the copy
> *first* and check the beginning *second* but that's probably 
> more involved.
> 
> > Anyway, the attached patch for the current CVS rsync adds an option 
> > --copy-devices that makes rsync copy the contents of source device 
> > files as if they were regular files.  I tested the option on my 
> > computer to copy a block device to a regular file locally, and it 
> > seemed to work.  Wayne, you might like to include this patch in the 
> > patches/ dir of the source distribution.
> 
> Thanks for the patch. I hope something like that goes in. I 
> understand rsync was not meant as a block-device-level copy 
> tool, but since it already has most of what it takes, it's tempting.
> 

Could be extremely useful (in some cases)
but beware of rsync of /dev/zero and friends

One very useful stunt is an
rsync that takes a reasonable amount of time
followed immediately by an
rsync that takes a very short amount of time

??an offsite full-disk-backup of a moving target almost feasible??