[Parted-maintainers] Bug#276721: tries to open non-existing device files and hangs

[Sven Luther]

On Sat, Oct 16, 2004 at 05:01:09PM +0200, Eduard Bloch wrote:
> #include <hallo.h>
> * Sven Luther [Sat, Oct 16 2004, 03:31:24PM]:
> 
> > > > So, if you are trying to format /dev/hda1 or resize it or whatever, so why in
> > > > hell are you trying to setup the devfs disk devices ? I also think this is
> > > 
> > > ??
> > > 
> > > I did not try to setup any new devfs related thing. I was just *using*
> > > devfs, all the time.
> > 
> > You should not use evil devfs at all. :)
> > 
> > Seriously, it is going to be removed from the 2.6 kernels very soon, and you
> > should maybe try using udev instead.
> 
> Seriosly, I could get gray hairs if I would read the changelog of every
> version on LKML. This thing could be the future, but currently I would
> not use for any production system.

Well, i also don't follow LKML, but from second hand report it seems this is
the strong concensus.

> > What confused me is you speaking of /dev/hda1. Are you sure you have the devfs
> > stuff ? What kernels are you using anyway ?
> 
> Yes. And I have devfsd to not remember all the long number. So yes, I
> have both, devfs and /dev/hda ;)

Ok. This may be what is confusing parted maybe, can you try again without
having devfsd running ? 

> > > > related to some problem with the resize design that was spoken of in the
> > > > parted mailing list. I will follow up on this. Basically, if i understood
> > > > well, when resizing, libparted is syncing all the partitions, and then
> > > > reopening it, or something such.
> > > 
> > > Something like that. Maybe it tried to open the "new" partition, which
> > > did not exist because the kernel was not able to reread the part. table,
> > > and then it started to wreak random havoc. Just a theory.
> > 
> > Did you get any message about rereading the partition table ? 
> 
> Only while exiting parted. Not when the problem occured.

Mmm.

> > > I could, but not yet. And if my assumption above is correct, it will
> > > work because on the installer, nothing is mounted so the kernel will
> > > reread the partition table cleanly and create the new device file.
> > 
> > Err, are you trying to modify a partition while it is mounted ? This is _NOT_
> > supposed to happen. But please give d-i it a try anyway, it is best to have
> 
> Heh? When I write nothing, I mean nothing. The counterpart is is
> "something" and this something is the second partition where Linux was
> running. I did write that before, damn. So the modified partition was
> not mounted, or do you think I am going nuts?

Ok, just making sure.

> > fact and not assumptions when tracking stuff like that. I cannot really help
> > you all that much myself, since i don't use devfs, have no 2.4 kernel anymore,
> > and anyway don't have any disk with a MBR or a fat partition on it anyway.
> 
> I seriously think about going back to statical /dev. Kernel developers
> wanna devfs die for no good reason (maybe personal issues with its
> author) and udev is a cludge IMHO.

Well, my understanding is that devfs was never really adopted from the start,
so it doesn't surprise me it is going away, and the devfs names are ugly and
such. I don't know the internals of udev though, and only use it for the
auto-mounting of hoptplug devices in gnome myself.

Friendly,

Sven Luther