Re: kernel-managed /dev, revisited

Miguel de Icaza (
Fri, 11 Jul 1997 19:12:26 -0500

> What if you've got filesystems that don't take labels?

Well, that will be an inconvenience for those who happen to not use
the one true file system in all of their disks.

But seriously, this feature will not make the life any different to
those users. In the other hand, it would allow a lot of Linux users
to move freely their volume-labeled file systems.

What you suggest is equivalent to not implementing a SPARC
feature in the SPARC port just because Intel users would not
take advantage of it.

> What if you've got _two_ usr filesystems (my case: a
> RedHat "working" system, and a "test-drive" Debian one)?

Well, you should be able to do e2label /dev/sda2 "redhat/usr"; e2label
/dev/sdb2 "debian/usr" and modify your setup files accordingly.
Linux installation programs would let you select a proper label at
install time.

> What if some of the partitions aren't filesystems but just raw
> devices (like a floppy you simply tar over), or before mkfs-ing it?

Well, what is wrong with those? You can still access your devices
trought the device files, I am just suggesting to add to the kernel
support for labels on the file systems. (or to a user-level daemon or
to the mount tool or to any combination of those).

> Or filesystems/devices that need to load a module to access?

This would work by a piece of code that knows about the layout of the
different file systems supported. You do not need to load the file
system code into the kernel to figure out what kind of file system you
are dealing with. For instance, some twisted hacker could probably
make an /etc/magic entry for recognizing file systems :-).