On 08/03/2017 08:42 PM, Jeff Layton wrote:
On Fri, 2017-05-19 at 11:55 +0530, Soumya Koduri wrote:

On 05/18/2017 08:04 PM, Yun-Chih Chen wrote:
Hi, friends:

I'm studying the affect of NFSv4 on Linux.  As far as I know, NFS
implemented in Linux kernel supports read delegation but not write
delegation.  I wonder if Ganesha implements write delegation?  If yes,
can it take effect if accessed by Linux NFS client?

I use the following config file to observe delegation in Ganesha:

         Export_Id = 77;
         Path = /e;
         Pseudo = /e;
         Access_Type = RW;
         FSAL {
                 Name = XFS;

         Delegations = readwrite;

         Delegations = true;

However, I did not see any trace of delegation under various workloads
(example: repeated read [1], repeated write [2]) using tools like
nfstrace or tcpdump.  When running in debug mode, I always got the
following in the log file:

ganesha.nfsd-8999[main] display_fsinfo :FSAL :DEBUG :  delegations = 0

indicating that delegation was not on.
Is there something wrong with my config file?  Or any other clue
regarding NFS delegation?  Thanks ( ;

AFAIK except for FSAL_GPFS, other FSALs do not have support for
delegations yet. Also after switching to new extended APIs
(>=nfs-ganesha-2.4), delegations are disabled. It needs some additional
work. We are planning to address it as part of adding this support for
FSAL_GLUSTER (hopefully in 2.6 release). WRT XFS, I am not sure if
anyone is actively looking at it.


Hi Soumya,

I just started looking at delegation support in ganesha (mostly with an
eye toward plumbing in delegation support for Ceph).

I think we probably need to rework the whole delegation interface (maybe
even give it a dedicated FSAL op), and was wondering if you had started
any work along those lines.

Hi Jeff,

yes. I started it but have done only POC to mainly test the glusterfs lease support (which is experimental feature) at first. I could manage to get leases granted from and returned to the backend.

The only changes I had done so far in the core ganesha layer is to uncomment earlier delegation code path to get the request all the way till FSAL [1]. I now intent to fix the SAL layer and happy to collaborate with you. Agree that instead of clubbing with lock, a dedicated fop shall be better.


[1] https://review.gerrithub.io/372998


Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
Nfs-ganesha-devel mailing list

Reply via email to