[PATCH 3/3 v4] vfs_glusterfs: Samba VFS module for glusterfs

[Jeremy Allison]

On Fri, May 24, 2013 at 04:17:48PM -0700, Anand Avati wrote:
> On 5/24/13 5:18 AM, Simo wrote:
> >
> >By using an array you are forced to create this cleanup function,
> >however if you use a doubly linked list you could simply free the fs and
> >use a talloc destructor to perform the clean up. Have you though about
> >doing that ?
> 
> I did not look around enough in the samba code to find a list
> template and did not want to write list primitives from scratch. Is
> there a specific list template in samba which you recommend?

Check out all uses of DLIST_XXX() functions (found in
lib/util/dlinklist.h.

> fallocate() support is currently under review in glusterfs (likely
> to be merged soon). This vfs_slow_fallocate() behavior is not a new
> issue as this has been how it worked with the FUSE mount as well.
> Once fallocate() support is merged, I will send a new patch to use
> the API call.

Having a native gluster fallocate will help a *LOT*.

> Does this flock() strictly have to be in a separate lock space from
> fcntl()? In Linux they are separete, but how does samba deal with
> other OSes where they compete in the same lockspace? In gluster we
> have a single lockspace. Will errno = ENOSYS make samba autodetect
> the lack of 'kernel share modes' support?

Yes.

> Gluster's ACL is byte-compatible with Linux's POSIX ACL format. The
> use of acl_copy_int() here is possibly dangerous on non-Linux
> servers (though in reality most of the OS's POSIX ACL format is byte
> compatible?). In that sense, maybe it makes sense to have a Gluster
> format to smb_acl_t converter implemented natively in this module
> (though it might look similar to smb_acl_to_internal).

Yes.

Jeremy.