Re: Signing causes to loose hardlinks.

[Jeff Johnson <jbj redhat com>]

On Mon, Aug 25, 2003 at 01:53:00PM +0200, Dag Wieers wrote:
> On Mon, 25 Aug 2003, Jeff Johnson wrote:
> 
> > Dag Wieers wrote:
> > 
> > >Hi,
> > >
> > >I guess this may have already been covered by someone else, but if I 
> > >resign a package that is hardlinked, rpm causes the package to have 
> > >another inode, loosing all the hardlinks in return.
> > >
> > >Could this be fixed in a future version, or is this done on purpose ?
> > 
> > Hmmm, I expect just the opposite, as rpm goes to lengths to reopen the 
> > original
> > package with O_TRUNC to preserve the inode in the face of hardlinks.
> 
> Feel free to try it out. I did that on a 2.5GB repository and quickly 
> found out that I didn't have twice the diskspace available ;)
> 

Hmmm, you're right, hardlinks not preserved. Used to work, dunno when
change occured. I have had request to *not* honor hardlinks because
Red Hat now uses symlinks rather than hardlinks, that was maybe 10 months
ago, presumably the behavior was different then.

Again, I don't know whether honoring hardlinks or not is the correct
behavior. Rewriting in place with O_TRUNC relies on file system support,
and a robust solution might very well involve teaching rpm about the
expected behavior of various file systems. Ick.

> And the file it was doing when the disk was full, had a leftover 
> *.rpm.TYURS kind of file that was truncated.

Hmmm, what file system? Adding a suffix like that was not done by
rpm code afaik.

> 
> PS I also came across RH73 problems when resigning about 2000 packages 
> where it caused RPM to stop suddenly because of an error. It didn't happen 
> on a RH9 (but I didn't use 2000 packages at once there).

Thanks for the report.

73 de Jeff

-- 
Jeff Johnson	ARS N3NPQ
jbj@redhat.com (jbj@jbj.org)
Chapel Hill, NC