[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-block] [PATCH] block: check BlockDriverState object before der
|
From: |
Kevin Wolf |
|
Subject: |
Re: [Qemu-block] [PATCH] block: check BlockDriverState object before dereference |
|
Date: |
Tue, 1 Aug 2017 15:40:26 +0200 |
|
User-agent: |
Mutt/1.8.3 (2017-05-23) |
Am 01.08.2017 um 10:35 hat Paolo Bonzini geschrieben:
> On 01/08/2017 02:14, John Snow wrote:
> > I may need some nudging towards understanding what the right solution
> > here is, though. Should the blk_aio_flush assume that there always is a
> > root BDS? should it not assume that?
>
> I think blk_aio_flush is not special. If there is no root BDS, either
> you return -ENOMEDIUM, or you crash. But all functions should be doing
> the same.
The intended semantics is that they return -ENOMEDIUM (or fail at
least). This is how things have always worked, and that it crashes now
because of the bdrv_inc_in_flight() was not an intentional change, but
simply a bug in the patch.
> The former makes sense, but right now blk_prwv for one are crashing if
> there is no root BDS so the minimum patch would fix the caller rather
> than blk_aio_flush.
The synchronous versions don't crash, and bdrv_aio_prwv() would fix all
cases if bdrv_inc_in_flight() were moved inside the coroutine; probably
right before blk_aio_complete(). This would be more consistent with how
the synchronous versions work, too, increasing the in-flight count only
by 1 rather than 2 for an AIO request.
Kevin