Re: [Qemu-block] [PATCH] [RFC] aio: properly bubble up errors from initialization

[Nishanth Aravamudan]

Hi Kevin,

On 15.06.2018 [10:41:26 +0200], Kevin Wolf wrote:
> Am 15.06.2018 um 01:21 hat Nishanth Aravamudan geschrieben:
> > laio_init() can fail for a couple of reasons, which will lead to a NULL
> > pointer dereference in laio_attach_aio_context().
> > 
> > To solve this, add a aio_linux_aio_setup() path which is called where
> > aio_get_linux_aio() is called currently, but can propogate errors up.
> > 
> > virtio-block and virtio-scsi call this new function before calling
> > blk_io_plug() (which eventually calls aio_get_linux_aio). This is
> > necessary because plug/unplug currently assume they do not fail.
> > 
> > It is trivial to make qemu segfault in my testing. Set
> > /proc/sys/fs/aio-max-nr to 0 and start a guest with
> > aio=native,cache=directsync. With this patch, the guest successfully
> > starts (but obviously isn't using native AIO). Setting aio-max-nr back
> > up to a reasonable value, AIO contexts are consumed normally.
> > 
> > Signed-off-by: Nishanth Aravamudan <address@hidden>
> 
> This is not a reasonable fix for several reasons:
> 
> * You frame this as a problem of blk_io_plug(), but that's not what it
>   is. It is a problem of delayed initialisation of Linux AIO that may
>   in the future affect other operations as well.
> 
> * This approch would need a fix in every device that uses a problematic
>   operation. You came across virtio + blk_io_plug(), but that are
>   probably not the only cases in the long run, which would make the code
>   spread much wider than it should.
> 
> * There is only a single block driver that actually implements the new
>   callback. This is a sign that this is not a generally useful callback.
> 
> Instead, the fix should be done locally in the file-posix driver, and
> the virtio devices shouldn't be touched at all. I think it would be good
> enough to call laio_init() when attaching to a new AioContext and to
> switch to the thread pool if it fails, like you already do. Maybe an
> error_report() would be appropriate to log the fact that we're not using
> the requested AIO mode.

Thank you for the constructive feedback! I will work on a v2 ASAP.

-Nish