Re: [PATCH] ext2fs and large stores (> 1.5G)

[Ognyan Kulev]

Neal H. Walfield wrote:
> > This is attractive idea but the thing that stopped me to use it was
> > that pages can be flushed in _pokel_exec without actual call to
> > disk_pager_write_page.
>
> How?  If a page is flushed and it is dirty, it *will* be returned via
> disk_pager_write_page.

If _pokel_exec don't call pager_sync_some, this flushing of the page is 
delayed for unpredictable moment in the future, and during this period 
the block will be locked for reusing.

> > Thinking again about it, disk_cache_clear can
> > be called in _pokel_sync when !sync.  I'll try this and if it works
> > fine I'll upload new snapshot tomorrow.
>
> I still believe this appoarch is wrong.  Please convince me otherwise.

When I started writing the patch the only thing I wanted was something 
that works.  I'm very glad that we now discuss how to do it the right way.

Leaving the housekeeping (flushing etc) to microkernel is natural way in 
Hurd environment, but as I said in another mail, I need a reliable way 
to be notified when the microkernel evicts page so that it can be reused.

Regards
--
Ognyan Kulev <ogi@fmi.uni-sofia.bg>, "\"Programmer\""