Snapshot directories should be separate mountpoints or snapshotted files should have unique inode numbers

In Linux, files in Spectrum Scale snapshots have the same device id
(st_dev) and inode number (st_ino) as their live counterparts. They
have the same device because all snapshots in a filesystem are under the
same mountpoint as the live version of the filesystem.

This breaks the following guarantee in Linux (and POSIX if one wants to
be picky): 'The st_ino and st_dev fields taken together uniquely
identify the file within the system.'

When using snapshots in Spectrum Scale, two different files (a
snapshotted file and a corresponding live file) can have the same device
id and inode number, and this in turn can break applications, as
demonstrated in the use case examples of this RFE.

How does other snapshotted filesystems handle this? For the ones that I
have looked at, the guarantee is kept by letting each snapshot be a
separate mountpoint with a unique device id for each snapshot. I guess
an alternative solution would be to change the inode number for a
snapshotted file when the live file is modified. At least the inode
number returned by stat(2) -- I assume Spectrum Scale would need to keep
the original inode number internally to track snapshotted files
relations to the live files.

For this RFE, we would be happy if either

- each snapshot in the global snapdir is a separate mountpoint with a
unique device id, or
- stat(2) for a snapshotted file returns a unique inode number if the
live file has been modified.

To limit the scope of this RFE: we are not interested in how
'mmsnapdir -a' or fileset snapshots are handled, so this RFE covers the
global snapdir only.