Re: [PATCH -V14 0/11] Generic name to handle and open by handle syscalls

[Neil Brown <neilb@suse.de>]

On Wed, 7 Jul 2010 12:11:50 +1000
Dave Chinner <david@fromorbit.com> wrote:

> On Wed, Jul 07, 2010 at 09:36:29AM +1000, Neil Brown wrote:
> > On Wed, 7 Jul 2010 09:23:51 +1000
> > Dave Chinner <david@fromorbit.com> wrote:
> > 
> > > > I can add a new syscall that returns 
> > > > 
> > > > struct fs_uuid {
> > > >        u8 fs_uuid[16];  
> > > > };
> > > > 
> > > > long sys_get_fs_uuid(int dfd, char *name, struct fs_uuid *fsid, int flag);  
> > > 
> > > libblkid already provides the UUID to userspace applications, doesn't it?
> > 
> > Yes and no.
> > 
> > libblkid provides the uuid of the thing that uses a block device.  That
> > doesn't directly map to "UUID of a filesystem".
> 
> True.
> 
> > There are two types of filesystem that I can think of for which libblkid
> > cannot give a uuid.
> >   - network filesystems (or virtual filesystems, or fuse )
> 
> How would you guarantee persistent uniqueness for such filesystems?

Persistent shouldn't be too hard in many cases.
What uniqueness guarantees do we have anyway?  Mostly stochastic I expect.


> 
> >   - filesystems which share a block device, such as btrfs.
> >     btrfs can have 'subvols' - multiple "filesystems" within
> >     the one (set of) block device(s).  libblkid cannot be asked about these
> >     different subvols.
> > 
> > libblkid is useful, but not a real solution.
> 
> So libblkid doesn't cover everything, but I think my question is
> still valid - if we want per-filesystem UUIDs, why a syscall and not
> just publishing it somewhere where we already publish per-mount
> information?  e.g. in /proc/mounts?

The trouble with /proc/mounts is that it is somewhat clumsy to parse
(remember to handle \0ctal escapes) and doesn't include major/minor number
which is the primary key for identifying filesystems in Linux
(see /sys/class/bdi/MAJOR:MINOR which is e.g. the best place to configure
read-ahead for a filesystem).

So /proc/mounts could work (and would probably be better than a new syscall)
but I would really rather see something sane in /sys for
inspecting/configuring filesystems (rather than each filesystem doing their
own independent thing in /sys/fs).

NeilBrown