Re: TAKE 964002 - Don't wait for pending I/Os when purging blocks beyond eof.

[David Chinner <dgc@sgi.com>]

On Tue, Dec 11, 2007 at 08:11:17AM +0000, Christoph Hellwig wrote:
> On Mon, Dec 10, 2007 at 04:59:55PM +1100, Lachlan McIlroy wrote:
> > Don't wait for pending I/Os when purging blocks beyond eof.
> > 
> > On last close of a file we purge blocks beyond eof.  The same
> > code is used when we truncate the file size down.  In this case
> > we need to wait for any pending I/Os for dirty pages beyond the
> > new eof.  For the last close case we are not changing the file
> > size and therefore do not need to wait for any I/Os to complete.
> > This fixes a performance bottleneck where writes into the page
> > cache and cache flushes can become mutually exclusive.
> 
> I think I shortened from of this should be in the comment above
> the conditional vn_iowait intead of the current
> 
> 	/* wait for the completion of any pending DIOs */
> 
> which is wrong given that we don't wait for all pending direct I/O
> requests..  (and vn_iowait doesn't wait for direct I/O anyway)

vn_iowait() does wait for direct I/O. That was it's entire purpose - to be
able to prevent truncate vs direct I/O write races by tracking direct I/Os.
We increment ip->i_iocount in xfs_alloc_ioend() which is called from both the
buffered write and direct I/O write path, so vn_iowait() does wait for both
buffered and direct writes to complete.

FWIW, the freeze code makes use of this functionality (SYNC_IOWAIT) to ensure
all pending data writes are complete complete before the freeze is completed....

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group