…= NULL

The logic in ll_create relies on *fhp being zeroed out on entry into the
function, but that's no longer being done since commit e08210d. Fix
this by reinstating that initialization.

Luckily, it looks like the two main callers (FUSE and nfs-ganesha)
already handle this safely, but the SyntheticClient does not. Move the
zeroing of the pointer into ll_create, and allow callers to pass in a
pointer to an uninitialized value. With this, we can remove the
initialization in the fuse client code as well.

Also, we can get rid of some unneeded NULL pointer checks if we don't
allow callers to pass in a NULL fhp pointer. As best I can tell, no one
does that currently and no comments over ceph_ll_create seem to claim
that this should be allowed.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

Holdover from the proposed kernel API that doesn't really help anything
here. If we ever want to extend this struct, we'll just append to the
end and add new want mask bits.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

This is just much more convenient for applications that want to use
the API. We're not talking to the kernel here, so there's no real
reason we need to use fixed length fields.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

Signed-off-by: Jeff Layton <jlayton@redhat.com>

This reverts commit 2115de0.

This is unnecessary now that we're using dev_t's in ceph_statx.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

We don't provide stx_version and stx_information is currently unused.

Signed-off-by: Jeff Layton <jlayton@redhat.com>

The proposed statx interface does this, and this would allow ganesha
to get away without requesting CEPH_STATX_MODE when all it wants is
enough to instantiate a filehandle.

Signed-off-by: Jeff Layton <jlayton@redhat.com>