[pnfs] [PATCH 4/5] Shut down the v4 style callback server for v4.1

[Benny Halevy]

Iyer, Rahul wrote:
>  
> 
>> -----Original Message-----
>> From: Benny Halevy [mailto:bhalevy at panasas.com] 
>> Sent: Tuesday, April 17, 2007 6:35 AM
>> To: Iyer, Rahul
>> Cc: pnfs at linux-nfs.org
>> Subject: Re: [pnfs] [PATCH 4/5] Shut down the v4 style 
>> callback server for v4.1
>>
>> iyer at netapp.com wrote:
>>> From: Rahul Iyer <iyer at netapp.com>
>>>
>>> This patch shuts down the v4.1 server in case the mount is a v4.1 
>>> mount. If it's a v4 mount, nothing changes.
>>>
>>> Signed-off-by: Rahul Iyer <iyer at netapp.com>
>>> ---
>>>  fs/nfs/super.c |   11 ++++++++++-
>>>  1 files changed, 10 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/fs/nfs/super.c b/fs/nfs/super.c index 89eb564..3411ea1 
>>> 100644
>>> --- a/fs/nfs/super.c
>>> +++ b/fs/nfs/super.c
>>> @@ -1270,8 +1270,17 @@ static int nfs4_fill_super(struct 
>> super_block *sb, struct nfs4_mount_data *data,
>>>                                  printk(KERN_EMERG 
>> "Couldn't mount using minorversion %d\n", i);
>>>                                  
>> rpc_shutdown_client(server->client);
>>>                          }
>>> -			else
>>> +			else {
>>> +				/* We successfully managed to 
>> set up a session.
>>> +				 * Since the v4.1 callbacks are 
>> different from
>>> +				 * that of v4, shutdown the v4 
>> callback server
>>> +				 * started up in nfs4_alloc_client
>>> +				 */
>>> +				if (i > 0)
>>> +					nfs_callback_down();
>> I really don't like this, and the comment that's needed to explain it.
>> Rather than tearing down the nfs callback path, it shouldn't 
>> have been established for i > 0 in the first place.
>> A flag asking to create a callback path should be passed down 
>> to nfs4_{create,get,alloc}_client.
> 
> Yeah! I noticed that. Actually, I found it rather odd that
> nfs_callback_up was being called in the alloc routine. This was the path
> of least resistance. I could do it the other way as well if this seems
> to hackish.
> Regards
> Rahul
> 
Yes, this is terribly hackish IMO.  I'd rather add a flag to
nfs4_create_client() as I suggested above, and either pass it
down the calling stack, or alternatively, carve out the call
to nfs_callback_up() from nfs4_alloc_client and move it up to
nfs4_create_client().

I think this means only that mounts of DSs for the files layout
type will not have a callback path.  Do they need it? (e.g. they
could have data delegations on the DS files, do we handle that
correctly?)

Benny