[pnfs] [PATCH 02/28] pnfs: fix pnfs_{read|write}pages

[Fredric Isaman]

On Wed, 12 Mar 2008, Benny Halevy wrote:

> On Mar. 12, 2008, 18:23 +0200, Fredric Isaman <iisaman at citi.umich.edu> wrote:
>> On Wed, 12 Mar 2008, Benny Halevy wrote:
>>
>>> On Mar. 11, 2008, 21:31 +0200, Fred Isaman <iisaman at citi.umich.edu> wrote:
>>>> THIS WILL BREAK OTHER LAYOUT DRIVERS that rely on old behavior.
>>> Fred, do you have the cycles to fix the files and objects LDs
>>> for the new behavior so we can merge this patch onto the pnfs branch?
>>>
>>
>> Yes, I'll work on that.
>>
>>  	Fred
>
> Thanks!
>
>


OK, I've been glancing through the code for the other layout drivers.
Basically, in the functions ld_io_ops->{read|write}_pagelist, the 
filelayout driver only returns 0, while the panlayout only returns 0 or a 
negative error.  So why do we have the checks for positive return 
value at all?

In addition, to clarify the intent of some of the panobject code:

Does sam_read/sam_write return PAN_ERR_IN_PROGRESS iff called with a 
completion callback?

Why the check against PNFS_ISSYNC in panlayout_*_done? Don't you still 
need to deallocate the memory used by the request?

On that note, it seems that PNFS_ISSYNC is never currently set/used at 
all.  Is this true?

 	Fred