Implementation of delegation

Implementation of delegation

[Yudong Gao]

Hi,

I am reading the NFS source code of the file open function in kernel
2.6.34, but I cannot find any implementation with delegation to allow
client to open a file locally without contacting the server. We are
tying to use delegation to do something interesting. So I am wondering
how much delegation support specified in the NFS 4.1 protocol is
available in the current Linux implementation?

Thanks a lot!

best,

Yudong

Re: Implementation of delegation

[Trond Myklebust]

On Mon, 2010-07-12 at 14:37 -0700, Yudong Gao wrote:
> Hi,
> 
> I am reading the NFS source code of the file open function in kernel
> 2.6.34, but I cannot find any implementation with delegation to allow
> client to open a file locally without contacting the server. We are
> tying to use delegation to do something interesting. So I am wondering
> how much delegation support specified in the NFS 4.1 protocol is
> available in the current Linux implementation?

What's wrong with nfs4_try_open_cached()? It even works with NFSv4.0...

Trond

Re: Implementation of delegation

[Yudong Gao]

Hi Trond,

Thanks for the prompt reply!

I find that in _nfs_do_open(), _nfs4_proc_open(), which do the RPC
call to the remote server, is always called before
nfs4_opendata_to_nfs4_state(), which will further invoke
nfs4_try_open_cached().

Does this means that whenever is file is opened, the client always
need to talk to the server first?

Thanks!

best,

Yudong

On Mon, Jul 12, 2010 at 2:48 PM, Trond Myklebust
<trond.myklebust@fys.uio.no> wrote:
> On Mon, 2010-07-12 at 14:37 -0700, Yudong Gao wrote:
>> Hi,
>>
>> I am reading the NFS source code of the file open function in kernel
>> 2.6.34, but I cannot find any implementation with delegation to allow
>> client to open a file locally without contacting the server. We are
>> tying to use delegation to do something interesting. So I am wondering
>> how much delegation support specified in the NFS 4.1 protocol is
>> available in the current Linux implementation?
>
> What's wrong with nfs4_try_open_cached()? It even works with NFSv4.0...
>
> Trond
>
>

Re: Implementation of delegation

[Trond Myklebust]

On Mon, 2010-07-12 at 15:00 -0700, Yudong Gao wrote:
> Hi Trond,
> 
> Thanks for the prompt reply!
> 
> I find that in _nfs_do_open(), _nfs4_proc_open(), which do the RPC
> call to the remote server, is always called before
> nfs4_opendata_to_nfs4_state(), which will further invoke
> nfs4_try_open_cached().
> 
> Does this means that whenever is file is opened, the client always
> need to talk to the server first?

The actual RPC call is skipped if the call to can_open_cached() succeeds
in nfs4_open_prepare(). In that case, we just grab the sequence, which
ensures that we remain serialised w.r.t. CLOSE and OPEN_DOWNGRADE calls.

Trond

Re: Implementation of delegation

[Yudong Gao]

I see! Thanks a lot for the pointer!

best,

Yudong

On Mon, Jul 12, 2010 at 3:07 PM, Trond Myklebust
<trond.myklebust@fys.uio.no> wrote:
> On Mon, 2010-07-12 at 15:00 -0700, Yudong Gao wrote:
>> Hi Trond,
>>
>> Thanks for the prompt reply!
>>
>> I find that in _nfs_do_open(), _nfs4_proc_open(), which do the RPC
>> call to the remote server, is always called before
>> nfs4_opendata_to_nfs4_state(), which will further invoke
>> nfs4_try_open_cached().
>>
>> Does this means that whenever is file is opened, the client always
>> need to talk to the server first?
>
> The actual RPC call is skipped if the call to can_open_cached() succeeds
> in nfs4_open_prepare(). In that case, we just grab the sequence, which
> ensures that we remain serialised w.r.t. CLOSE and OPEN_DOWNGRADE calls.
>
> Trond
>
>