Lustre devel - Mar 2007 - [Bug 11710] use PG_writeback in lustre

Please don''t reply to lustre-devel. Instead, comment in Bugzilla by
using the following link:
https://bugzilla.lustre.org/show_bug.cgi?id=11710



(In reply to comment #20)
> > Please explain, why you add llap_writeback member and don`t use 
> > PageWriteback()instead of. How i see this duplicate PG_writeback 
> > bit, but i don`t see case when his value is different to 
> > PageWriteback().
> 
> First off thanks for the quick review Alexey.  This was just a first
> cut at a new patch.
> 
> Your right llap_writeback is always set the same as PageWriteback().
> The only reason I did this initially was I was not 100% sure that
> the PG_writeback bit was never modified by the VFS.  By introducing
> llap_writeback in lustre I could just easily detect if they got out
> of sync and assert().  That all said I''ve pretty much convinced
myself
> that''s impossible so I have no objection to dropping that and
trusting
> the PG_writeback bit.
> 
Quick review contens in fs/ and mm/ show only two places when PG_writeback can
be cleared by kernel. One - in swap io, one in generic writepage.
seems use PageWriteback(page) should be safe.

> > Also you can simplify code in ll_ap_make_ready - if we double call
> > set_page_writeback for page - it`s not produce error. Same for 
> > ''ll_writepage''.
> 
> Yup, I think I can do something to simply this all.
> 
> On a related note, do you have any ideas on how to best handle clearing
> the error code for the file in the presense of recoverable errors. 
I''d
> really like fsync() to return success if it was able to eventually write 
> out all the pages.  Perhaps instead of a error flags being maintained
> a counter could be used instead...  I''m still not sure what the
right
> thing to do here is.
>       
I will be work on this.