MIME-Version: 1.0
In-Reply-To: <53FCBE9C.40906@plexistor.com>
References: <53FCA183.8000605@plexistor.com>
	<1435166875.78.1409066662291.JavaMail.root@thunderbeast.private.linuxbox.com>
	<CAHQdGtT-B7NTP8T7ETgTJD1+84k_KW_5aOfrYjLpA04hoR3YsA@mail.gmail.com>
	<53FCBC5A.4060304@plexistor.com>
	<CAHQdGtSghgbrPr=X8O2YLq=vXbKY1y5OuOMisf6vRoj8WeF0YA@mail.gmail.com>
	<53FCBE9C.40906@plexistor.com>
Date: Tue, 26 Aug 2014 13:54:22 -0400
Message-ID: <CAHQdGtTgcGiWbcZyAsaOzdB=RpdUx17vHcw1yxg7vS56BfwY+A@mail.gmail.com>
Subject: Re: [PATCH] pnfs: Kick a pnfs_layoutcommit_inode on recall
From: Trond Myklebust <trond.myklebust@primarydata.com>
To: Boaz Harrosh <boaz@plexistor.com>
Cc: "Matt W. Benjamin" <matt@linuxbox.com>, Christoph Hellwig <hch@lst.de>,
        Linux NFS Mailing List <linux-nfs@vger.kernel.org>,
        "Adam C. Emerson" <aemerson@linuxbox.com>
Content-Type: text/plain; charset=UTF-8
Sender: linux-nfs-owner@vger.kernel.org

On Tue, Aug 26, 2014 at 1:06 PM, Boaz Harrosh <boaz@plexistor.com> wrote:
> On 08/26/2014 07:59 PM, Trond Myklebust wrote:
>> On Tue, Aug 26, 2014 at 12:56 PM, Boaz Harrosh <boaz@plexistor.com> wrote:
>>> On 08/26/2014 06:36 PM, Trond Myklebust wrote:
>>>> On Tue, Aug 26, 2014 at 11:24 AM, Matt W. Benjamin <matt@linuxbox.com> wrote:
>>>>> IIUC, the problem is the forechannel slot count, since the call you want to make synchronously is on the forechannel?
>>>
>>>
>>> Matt no top post on a Linux mailing list ;-)
>>>
>>>> Yep. layoutcommit will be sent on the fore channel, which is why it
>>>> can deadlock with the initial layoutget (or whatever operation that
>>>> triggered the layout recall).
>>>
>>> Trond you said below:
>>>> The above can deadlock if there are no session slots available to send
>>>> the layoutcommit, in which case the recall won't complete, and the
>>>> layoutget won't get a reply (which would free up the slot).
>>>
>>> Why would the layoutget not-get-a-reply ?
>>> This is how it goes with Both ganesha server and knfsd last I tested.
>>>
>>> [1]
>>> The LAYOUT_GET cause LAYOUT_RECALL case: (including the lo_commit)
>>>
>>> client          Server                                     comments
>>> ~~~~~~          ~~~~~~                                     ~~~~~~~~
>>> LAYOUT_GET ==>
>>>            <==   LAYOUT_GET_REPLAY(ERR_RECALL_CONFLICT)
>>>                                                 <--------- fore-channel is free
>>>            <==   RECALL
>>> LAYOUT_COMMIT ==>
>>>             <==  LAYOUT_COMMIT_REPLAY
>>>                                                 <--------- fore-channel is free
>>
>> Beep! No free slots, so this hangs.
>>
>
> Beep! does not do a very good of a job to explain. Sorry
>
> What do you mean? which slot? which channel? Just above your text it says
> "fore-channel is free" so are you saying it is not free? why not.
> Please use more then one line of text to explain. It might be clear to
> you but not to me.

The deadlock occurs _if_ the above layout commit  is unable to get a
slot. You can't guarantee that it will, because the slot table is a
finite resource and it can be exhausted if you allow fore channel
calls to trigger synchronous recalls on the back channel that again
trigger synchronous calls on the fore channel. You're basically saying
that the client needs to guarantee that it can allocate 2 slots before
it is allowed to send a layoutget just in case the server needs to
recall a layout.

If, OTOH, the layoutcommit is asynchronous, then there is no
serialisation and the back channel thread can happily reply to the
layout recall even if there are no free slots in the fore channel.

>>> RECALL_REPLY(NO_MATCHING) =>
>>>                                                 <--------- back-channel is free
>
> Thanks
> Boaz
>


-- 
Trond Myklebust

Linux NFS client maintainer, PrimaryData

trond.myklebust@primarydata.com