Re: [PATCH] NFS: filelayout should use nfs_generic_pg_test

[Boaz Harrosh <bharrosh@panasas.com>]

On 06/01/2011 03:14 PM, Trond Myklebust wrote:
> On Wed, 2011-06-01 at 08:47 +0300, Boaz Harrosh wrote: 
>> On 06/01/2011 06:18 AM, Weston Andros Adamson wrote:
>>
>> pnfs_generic_pg_test is the one that gets the layout.
>>
>> What you've done is revert to MDS IO
> 
> The "files" layout type always gets the layout in the pg_doio() method
> instead of the pg_test().
> 

Well I don't see where? I fought this all day, when trying to make the
new code run with objlayout, which was missing the implementation of pg_test().
And never got a pnfs-IO.

I've searched the full tree for calls to pnfs_update_layout() the only
one I can see are in:
nfs_pagein_multi() - which means within a single page, right?
nfs_pagein_one()   - But is protected with list_is_singular() so only in the
                     single page case
nfs_flush_multi()  - Same as nfs_pagein_multi
nfs_flush_one()    - Also here protected with list_is_singular()

and the all mighty
pnfs_generic_pg_test()

I cannot see where the filelayout is different then other layouts
in that respect. Sorry to be slow, I would like to understand?

And also be careful with nfs_generic_pg_test() it inspects
desc->bsize which is negotiated with MDS, it's very small.

> Cheers
>   Trond
> 

Thanks
Boaz