Re: [PATCH v4 00/10] fscache: Replace and remove old I/O API

[David Howells]

Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> But:
> 
> > However, if you would rather I just removed all of fscache and (most of[*])
> > cachefiles, that I can do.
> 
> I assume and think that if you just do that part first, then the
> "convert to netfslib" of afs and ceph at that later stage will mean
> that the fallback code will never be needed?

The netfslib coversions for afs and ceph are already in your tree and I have a
patch here to do that for 9p (if you're willing to take that in the upcoming
merge window?).

The issue is cifs[*] and nfs.  I could leave caching in those disabled,
pending approved patches for those filesystems.  This would mean that I
wouldn't need the fallback code.

An alternative is that I could move the "fallback" code into fs/nfs/fscache.c
and fs/cifs/fscache.c if that would be easier and merge it into the functions
there.  The problem will come when the cache wants to do I/O in larger units
than page size to suit its own block size[**].

David

[*] As it happens, it turns out that cifs seems to have a bug in it that
causes the entire cache for a superblock to be discarded each time that
superblock is mounted.

[**] At some point the cache *has* to start keeping track of what data it is
holding rather than relying on bmap/SEEK_DATA/SEEK_HOLE to get round the
extent-bridging problem.  I'm trying to take a leaf out of the book of other
caching filesystems and use larger block sizes (e.g. 256K) to reduce the
overhead of cache metadata.