Message-ID: <4FBE3C4D.7050507@panasas.com>
Date: Thu, 24 May 2012 16:49:01 +0300
From: Boaz Harrosh <bharrosh@panasas.com>
MIME-Version: 1.0
To: Benny Halevy <bhalevy@tonian.com>
CC: NFS list <linux-nfs@vger.kernel.org>
Subject: Re: Does our Kernel PNFSD-Server supports recurring layout_get with
 open_state_id
References: <4FBD7DE8.9090002@panasas.com> <4FBE2C8C.9090403@tonian.com>
In-Reply-To: <4FBE2C8C.9090403@tonian.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-nfs-owner@vger.kernel.org

On 05/24/2012 03:41 PM, Benny Halevy wrote:

> On 2012-05-24 03:16, Boaz Harrosh wrote:
>>
>> Benny hi
>>
>> If I remember/understand correctly, there is a mode in the RFC errata
>> about the layout forgetful-model and a client sending a layout_get with
>> an open_state_id after he already had previous state (layouts) on the file.
>>
>> As I understood this is an indication to the server that client has
>> "forgotten" all it's layouts on a file, and Server can assume their
>> return.
>>
>> Is my understanding correct?
> 
> Yes
> 
>>
>> If Yes:
>> 	Did we implement the internal return of all layouts, if above
>> 	open_state_id is encountered?
>> 	I thought we did but I can't find this code.
> 
> True.  This is not implemented yet.
> 
>>
>> Currently, I always set ROC so there is no leak. But theoretically
>> ROC does not have to be set. I'm doing some heavy lifting of
>> layout_return, and I want to make sure I have not missed a spot.
>>
>> If I'm correct that it is needed, and it's missing:
>> My suggestion for now is that we always set ROC, disregarding FS so not to
>> leak layouts and therefor inode-refs, until such time that we implement it.
> 
> According to the new errata the server will have to simulate layout returns in the ROC
> case on last CLOSE if the (forgetful) client did not explicitly return the layout.
> This is not implemented either :-(
> 


Yes Benny it is implemented. I implemented it and you fix a bug I had. See:
	nfs4pnfsd.c::pnfsd_roc(...) 
and it's call site. It is called on last call and all is working well (except some locking
bugs I'll send a fix to later). Otherwise both exofs and DF would have crapped out for sure.

With ROC set all work well. This is why I say we should set it for now to make sure we do
not have the above bug.

Thanks
Boaz

> Benny
> 
>>
>> Thanks
>> Boaz
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html