Re: [RFC PATCH] Generic name to handle and open by handle syscalls

[DENIEL Philippe <philippe.deniel@cea.fr>]

Hi,

I just found this message in this mailing list archive. This follows a
search I did after having a look at this thread on LWN.net
(http://lwn.net/Articles/375888/).

> The below set of patches implement open by handle support using 
exportfs > operations. This allows user space application to map a file
name to file > handle and later open the file using handle. This should
be usable > for userspace NFS and 9P server. XFS already support this
with the ioctls > XFS_IOC_PATH_TO_HANDLE and XFS_IOC_OPEN_BY_HANDLE.

After reading the mailing-list archive, it sounds like many people do
not see the benefit in having a way to address files from user space via
their fh and not only through the use of their name. I am leading a
project (see http://nfs-ganesha.sourceforge.net) that implements NFS
functionality from User Space. My server has various backends modules
that make it possible to address different kind of namespaces. When we
did the one dedicated to POSIX, we were strongly limited by the POSIX
functions that reference the files only by their names. We implemented
kind of "name <-> fh" tracking feature via the use of a db, but it is a
bit heavy and not performant. Having a way to translate name to fh, or
open file by fh would be great : it would make it possible to get rid of
all this heavy plumber we did with the db. In fact, half of the work is
already done in my project : we use LUSTRE a lot on our site, and we had
a need for exporting LUSTRE via NFS. There is a user space LUSTRE-API
was provide this "name to fid" and "open by fid" feature. We used it to
build a LUSTRE dedicated backend, and this fit perfectly our needs. So,
if patches are made to perform name-to-handle and open-by-handle
operation, this would be great. Of course, this is my personal point of
view, but personally I have a clear need of this feature.

Regards

Philippe