[pnfs] [PATCH 02/25] pnfs: fix pnfs4_*_done routines

[Benny Halevy]

On Feb. 15, 2008, 19:00 +0200, Fredric Isaman <iisaman at citi.umich.edu> wrote:
> 
> On Fri, 15 Feb 2008, Benny Halevy wrote:
> 
>> On Feb. 14, 2008, 20:21 +0200, Fred Isaman <iisaman at citi.umich.edu> wrote:
>>> The pnfs4_*_done routines are ignoring the possibility that the code
>>> fell back to nfs.  This patch chops out layoutcommit setup that is done
>>> there (incorrectly in the fallback case) which is already done by
>>> pnfs_writepages anyway.
>> Well, in the async case it is done in pnfs_writeback_done, but
>> only for pnfs_get_type(data->inode) != LAYOUT_NFSV4_FILES.
>> IIRC, last time I looked into this, I came to the conclusion
>> that the call to pnfs_update_last_write in pnfs4_write_done
>> is for the file layout rpc (via MDS) path.
>>
> 
> Actually, I was alooking at the call near the end of pnfs_writepages.
> I hadn't noticed it was also (conditionally) called during writeback_done.
> 
> 
>> Have you tested this patchset with over the file layout and made
>> sure layoutcommit is generated correctly after write?
>>
> 
> No, since patch 01 was guaranteed to break file layout.

OK, so please do :)
(you break it, you get to test it ;-)
I'll help test panlayout of course as you guys don't have a Panasas
system in house.