Re: [pnfs] [PATCH 0/2] fix nfsd stateid encoding

From: Benny Halevy <bhalevy@panasas.com>
To: "J. Bruce Fields" <bfields@fieldses.org>
Cc: NFS list <linux-nfs@vger.kernel.org>,
	pNFS Mailing List <pnfs@linux-nfs.org>
Subject: Re: [pnfs] [PATCH 0/2] fix nfsd stateid encoding
Date: Mon, 11 Aug 2008 20:34:41 +0300	[thread overview]
Message-ID: <48A07831.2050602@panasas.com> (raw)
In-Reply-To: <20080811162747.GA772@fieldses.org>

On Aug. 11, 2008, 19:27 +0300, "J. Bruce Fields" <bfields@fieldses.org> wrote:
> On Mon, Aug 11, 2008 at 07:11:40PM +0300, Benny Halevy wrote:
>> On Aug. 11, 2008, 18:58 +0300, "J. Bruce Fields" <bfields@fieldses.org> wrote:
>>> On Mon, Aug 11, 2008 at 05:09:36PM +0300, Benny Halevy wrote:
>>>> Bruce, in a couple locations the nfsd needs to encode the stateid.seqid
>>>> as a uint32_t rather than as opaque.
>>> Agreed, thanks.
>>>
>>> Though I have a hard time figuring out whether this has any impact in
>>> practice.  Presumably the only change on the wire is that we'll get the
>>> endianness of the stateid4.seqid right?  But that field is mostly opaque
>>> to the client anyway; 3530 says
>>>
>>> 	The server is required to increment the seqid field
>>> 	monotonically at each transition of the stateid.  This is
>>> 	important since the client will inspect the seqid in OPEN
>>> 	stateids to determine the order of OPEN processing done by the
>>> 	server.
>>>
>>> but doesn't say why this is important.  I'm sure this has been brought
>>> up on the ietf list before, but can't recall whether someone came up
>>> with a justification for the importance of this.
>>>
>>> Anyway, so I figure these should be queued up for the next (2.6.28)
>>> merge window.  Thanks!
>> Actually, I think this breaks delegreturn.
>> Since we decode the stateid.si_generation correctly, it will get swabbed
>> in delegreturn on little-endian servers.  This will cause
>> nfs4_preprocess_stateid_op/check_stateid_generation as called by
>> nfsd4_delegreturn to fail. And eventually, unhash_delegation
>> wouldn't be called.
> 
> Yipes.  If delegations have always been broken and we haven't noticed,
> then there's a more serious problem here--we should at least look into
> why pynfs isn't catching that.
> 
> Oh, I see: si_generation is always zero for delegation stateid's!

Phew :-)

> 
>> Hence, I think these patches are appropriate for 2.6.27 and
>> even to older stable releases.
> 
> So unless I've missed something, I think this still looks more like a
> style/consistency question?

Yup.

Benny

> 
> --b.
> 
>> Benny
>>
>>> --b.
>>>
>>>> Patch #1 fixes that for cb_recall.
>>>> Patch #2 fixes the deleg stateid returned by open.
>>>>
>>>> The patches should apply to linux-2.6/master 
>>>> commit 796aadeb1b2db9b5d463946766c5bbfd7717158c
>>>>
>>>> Benny
>>> _______________________________________________
>>> pNFS mailing list
>>> pNFS@linux-nfs.org
>>> http://linux-nfs.org/cgi-bin/mailman/listinfo/pnfs