Re: GFS2 and DLM

[Ingo Molnar <mingo@elte.hu>]

* Christoph Hellwig <hch@infradead.org> wrote:

> > The part where you needed file_read_actor looks like pretty much a stright
> > cut and paste from __generic_file_aio_read, which indicates that you might
> > be exporting at the wrong level.
> 
> A definitive NACK to this export.  All other filesystems manage to use 
> the generic file read code so GFS should do so aswell.  And there's a 
> technical reason for not exporting aswell as the generic file read 
> interface is far too complicated already.

GFS is different here mostly due to locking, because one of its strong 
features is an implementation of pretty strict POSIX semantics in a 
clustered environment, something that no other Linux FS (that is 
available in source code) has done so far. (OCFS2 does not do it as 
strictly - it has a very specific application in mind)

so i'd reformulate your request as a request to extend the VFS to unify 
clustering filesystems - which is a nice cleanup goal but not a merge 
showstopper to me.

> > Not sure about the tty_ export. Would it be better to make a generic 
> > printfish interface on top of it and also replace the interesting 
> > dquot.c gymnastics? (I don't know)
> 
> In fact neither gfs nor dquot should use it at all.  The xfs quota 
> code is fine without this nonsense.

yeah, the tty_ export is unnecessary and should be fixed. But this seems 
quite easy to do.

	Ingo