From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from eggs.gnu.org ([208.118.235.92]:44960)
	by lists.gnu.org with esmtp (Exim 4.71)
	(envelope-from <stefanha@redhat.com>) id 1UaSSl-0007LE-Rv
	for qemu-devel@nongnu.org; Thu, 09 May 2013 11:09:02 -0400
Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71)
	(envelope-from <stefanha@redhat.com>) id 1UaSSi-0006eJ-U1
	for qemu-devel@nongnu.org; Thu, 09 May 2013 11:08:55 -0400
Received: from mx1.redhat.com ([209.132.183.28]:55736)
	by eggs.gnu.org with esmtp (Exim 4.71)
	(envelope-from <stefanha@redhat.com>) id 1UaSSi-0006dz-Ms
	for qemu-devel@nongnu.org; Thu, 09 May 2013 11:08:52 -0400
Date: Thu, 9 May 2013 17:07:56 +0200
From: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20130509150756.GA26728@stefanha-thinkpad.redhat.com>
References: <1365581513-3475-1-git-send-email-wdongxu@linux.vnet.ibm.com>
	<1365581513-3475-6-git-send-email-wdongxu@linux.vnet.ibm.com>
	<20130418100336.GC19587@stefanha-thinkpad.redhat.com>
	<CAGrFBsha0kQKTtEEwveOWFgMSs4D3Qhq7ifNNquuiVbDzjSd4Q@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAGrFBsha0kQKTtEEwveOWFgMSs4D3Qhq7ifNNquuiVbDzjSd4Q@mail.gmail.com>
Subject: Re: [Qemu-devel] [PATCH V18 5/6] add-cow file format core code.
List-Id: <qemu-devel.nongnu.org>
List-Unsubscribe: <https://lists.nongnu.org/mailman/options/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=unsubscribe>
List-Archive: <http://lists.nongnu.org/archive/html/qemu-devel>
List-Post: <mailto:qemu-devel@nongnu.org>
List-Help: <mailto:qemu-devel-request@nongnu.org?subject=help>
List-Subscribe: <https://lists.nongnu.org/mailman/listinfo/qemu-devel>,
	<mailto:qemu-devel-request@nongnu.org?subject=subscribe>
To: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
Cc: Kevin Wolf <kwolf@redhat.com>, Stefan Hajnoczi <stefanha@gmail.com>, qemu-devel <qemu-devel@nongnu.org>

On Thu, May 09, 2013 at 02:24:20PM +0800, Dong Xu Wang wrote:
> On Thu, Apr 18, 2013 at 6:03 PM, Stefan Hajnoczi <stefanha@gmail.com> wrote:
> > On Wed, Apr 10, 2013 at 04:11:52PM +0800, Dong Xu Wang wrote:
> >> +static coroutine_fn int add_cow_co_flush(BlockDriverState *bs)
> >> +{
> >> +    BDRVAddCowState *s = bs->opaque;
> >> +    int ret;
> >> +
> >> +    qemu_co_mutex_lock(&s->lock);
> >> +    if (s->bitmap_cache) {
> >> +        ret = block_cache_flush(bs, s->bitmap_cache);
> >> +        if (ret < 0) {
> >> +            return ret;
> >> +        }
> >> +    }
> >> +    ret = bdrv_flush(s->image_hd);
> >
> > This is the wrong way around.  We must flush image_hd first so that
> > valid data is on disk.  Then we can flush bitmap_cache to mark the
> > clusters allocated.
> >
> > Beyond explicit flushes you also need to make sure that image_hd is
> > flushed *before* bitmap_cache tables are written out (e.g. cache
> > eviction when the cache becomes full).  It seems this code is missing.
> >
> Hi  Stefan, how can I make sure image_hd "flush" operation completed before
> I write bitmap_cache tables? Is there any functions I can use?

You could extend the block cache like this:

block_cache_add_dependency_func(BlockCache *c,
                                int (*dependency_fn)(BlockCache *c, void *opaque),
				void *opaque);

Then add-cow would register a custom dependency function that flushes
image_hd.

A hackier way of achieving the same thing is to instantiate an
image_hd_cache and set a dependency from the bitmap_cache on the
image_hd_cache.  The image_hd_cache actually has no cache entries, but
it uses image_hd and will flush it.

Stefan