On Tue, Jan 23, 2007 at 11:53:35AM -0800, Stephen Hahn wrote:
> * Nicolas Williams <Nicolas.Williams at Sun.COM> [2007-01-23 11:45]:
> > That's what I would think. It would be dangerous to provide a utility
> > like shred if it doesn't know to error out for files in ZFS.
> >
> > So it should least be modified to be ZFS aware. Since ZFS might not be
> > the only COW filesystem (think BSD 4.4 LFS) it may be a good idea to
> > come up with a generic interface for querying whether writes to a file
> > overwrite the same disk blocks or not.
>
> Well, this seems like an entree to make my economic point. We cannot
> staff each OSS integration effort such that it could modify a
> significant subset of commands (or the system) to meet policies and/or
> advised changes on initial integration; the freeware team is at best
> able to choose to ship or omit components. (This choice means that,
> in some cases, dependencies will not be satisfied that might have been
> satisfied on other systems.) If an incoming set of bug
> reports/customer calls implies that additional investment is
> justified, then that team will have to come back to declare their new
> proposed interfaces.
Then, IMO the fact that GNU shred(1) does not work on ZFS files should
result in:
- omission of shred from this case
- filing bugs against GNU shred to recognize ZFS and other COW
filesystems
- when GNU shred is so improved it could be added to Solaris then
- an RFE for an interface to ask the system whether writes to a file
overwrite its blocks, or to request that writes to a file overwrite
its blocks (the latter seems more robust, and ZFS would always reject
it)
Nico
--
