Re: Adding of peter gutmann's paper concept to /bin/rm

[Marcus Watts <mdw_(_at_)_umich_(_dot_)_edu>]

Dirk-Willem van Gulik <dirkx_(_at_)_webweaving_(_dot_)_org> writes:
> On Wed, 26 Mar 2003, Ted Unangst wrote:
> 
> > On Wed, 26 Mar 2003, dreamwvr_(_at_)_dreamwvr_(_dot_)_com wrote:
> ...
> > perhaps).  read the epilogue to the paper.  "In fact performing the full
> > 35-pass overwrite is pointless for any drive... you never need to perform
> > all 35 passes."
> 
> But in order to deceide which passes can be skipped; rm(1) would need to
> be tought the hardware characteristics of each drive type. This
> information is essentially not available to the kernel.
> 
> So the alternative would be to have flags for PRML, RLL or MFM drive types
> and rely on the user to know his or her disks. Most, if not all, users
> would punt in this case and opt for the 'paranoid' flag to run all 35
> passes - just incase.
> 
> Dw
> 
> 

Well, that's a nice theory.  But I don't think they make MFM drives
anymore.  Modern drives generally provide a sort of "virtual" interface
to the actual physical drive, which may actually work considerably
different than the interface might suggest.  For instance, the number
of heads, cylinders, & sectors is very likely to be way different, and
the number of sectors per track may vary between the inside & outside
of the disk.

But there's a way worse problem than this.  Many modern drives at least
support some form of hardware bad blocking.  This could happen at any
time, and if it does, the "bad" block may contain recoverable data.
It's also quite common to support some sort of buffering - that means
the interface layer reads and writes don't necessarily translate
directly into hardware reads and writes.  It's perfectly possible that
the drive might perform some sort of write compression (deleting
"redundant" writes to the same spot on the disk), and it might also
leave data in the buffer that isn't written to disk.  That buffer may
even be stored in nvram.  Even ordinary cheap laptop disks may have all
of these characteristics, but most forms of hardware RAID, especially
the higher levels, are guaranteed to have these features in spades.

				-Marcus Watts