Message-ID: <4FD5CBF6.6030800@panasas.com>
Date: Mon, 11 Jun 2012 13:44:06 +0300
From: Boaz Harrosh <bharrosh@panasas.com>
MIME-Version: 1.0
To: Benny Halevy <bhalevy@tonian.com>, "Welch, Brent" <welch@panasas.com>,
        "Bhamare, Sachin" <sbhamare@panasas.com>
CC: "Adamson, Andy" <William.Adamson@netapp.com>,
        Andy Adamson <androsadamson@gmail.com>,
        "Myklebust, Trond" <Trond.Myklebust@netapp.com>,
        "<linux-nfs@vger.kernel.org>" <linux-nfs@vger.kernel.org>,
        "Huang, James" <jhuang@panasas.com>
Subject: Re: [PATCH 2/3] NFSv4.1 mark layout when already returned
References: <1338571178-2096-1-git-send-email-andros@netapp.com> <1338571178-2096-2-git-send-email-andros@netapp.com> <4FCA98E7.2030006@panasas.com> <1C92D18B-1977-4A12-A4DA-84DAC4B3E81E@netapp.com> <4FCE1DC1.6050100@panasas.com> <CAHVgHyXVCUdVFMO1WcRofOCvgDs7Reqsrs39qrZR7UnXWzG8RQ@mail.gmail.com> <4FD5C0EA.806@tonian.com>
In-Reply-To: <4FD5C0EA.806@tonian.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-nfs-owner@vger.kernel.org

On 06/11/2012 12:56 PM, Benny Halevy wrote:

> 
> It will for recovering files striped with RAID as the LAYOUTRETURN
> provides the server with a reliable "commit point" where it knows
> exactly what was written successfully to each stripe and can make
> the most efficient decision about recovering it (if needed).
> 
> Allowing I/O to the stripe post LAYOUTRETURN may result in data
> corruption due to parity inconsistency.
> 


The fact of the matter is that we - object guys - Which care about
LAYOUTRETURN the most, have neglected to notice and fix the current
situation which is even worse than After Andy's patch. Which will
send multiple LAYOUTRETURNs on the same layout, and might still
continue IO on returned layouts.

Thanks Andy for testing and highlighting the current problem,
I will prioritized this development internally at Panasas and
will work on a solution ASAP. I hope one that will satisfy
all.

Basically my plan is to postpone the LAYOUTRETURN send to
when the segment is last referenced, if a flag was set to
do so, and make sure reference counting is done proper, so
after flag set the last IO_done will send the LAYOUTRETURN.

If that is OK with all I hope?

> Benny
> 


Thanks
Boaz