[OpenAFS-devel] Proposed change for handling backup volumes

[Kris Van Hees]

On Mon, Jul 07, 2003 at 05:09:31PM -0500, Nathan Neulinger wrote:
> I was thinking something along the lines of "if you traversed through a
> force-BK mountpt, always go BK until you hit a volume without a BK
> volume. The idea would be that you could make a single force-bk mount
> point of root.cell.
> 
> Not sure it's even possible to do as I described though, since you
> probably only have the context of the current volume the mount is being
> stat'd from. An approach might be to dynamically flag the volume entry
> when you traverse into it as 'force-bk' depending on how you got into
> it, and then check that flag when expanding mount points within that
> volume.

You indeed only have a context of the volume where the mountpoint is located
in.  Passing on a flag through the volume structs as they are being initialized
thorugh EvalMountPoint() seems a possible risk in that I believe volume structs
are shared in the cache manager, i.e. if two mount points resolve to the same
volume, they are both resolving to the same struct, and so setting a flag in
the struct would have rather unexpected side effects.
> 
> I just hate to dramatically change the semantics of mounts like you
> describe by default since it makes it impossible to use a backup volume
> in a way that it had previously been used. The idea does seem very
> useful though.

Well, it is only a change in the default behaviour of backup volumes when you
enable this feature (at afsd start, or at runtime).  So it is a deliberate
choice to have this behaviour over the default one.  The default behaviour of
backup volumes will remain the same, unless the feature is enabled.

	Kris