A common pattern for device reset is currently:
vdev->config->reset(vdev);
.. cleanup ..
reset prevents new interrupts from arriving and waits for interrupt
handlers to finish.
However if - as is common - the handler queues a work request which is
flushed during the cleanup stage, we have code adding buffers / trying
to get buffers while device is reset. Not good.
This was reproduced by running
modprobe virtio_console
modprobe -r virtio_console
in a loop.
Fixing this comprehensively needs some thought, and new APIs.
Let's at least handle the specific case of virtio_console
removal that was reported in the field.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1786239
Signed-off-by: Michael S. Tsirkin <[email protected]>
---
drivers/char/virtio_console.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
index 7eaf303a7a86..c852ce0b4d56 100644
--- a/drivers/char/virtio_console.c
+++ b/drivers/char/virtio_console.c
@@ -1956,6 +1956,12 @@ static void virtcons_remove(struct virtio_device *vdev)
list_del(&portdev->list);
spin_unlock_irq(&pdrvdata_lock);
+ /* Device is going away, exit any polling for buffers */
+ virtio_break_device(vdev);
+ if (use_multiport(portdev))
+ flush_work(&portdev->control_work);
+ else
+ flush_work(&portdev->config_work);
/* Disable interrupts for vqs */
vdev->config->reset(vdev);
/* Finish up work that's lined up */
--
MST
On Tue, Oct 05, 2021 at 03:04:07AM -0400, Michael S. Tsirkin wrote:
> A common pattern for device reset is currently:
> vdev->config->reset(vdev);
> .. cleanup ..
>
> reset prevents new interrupts from arriving and waits for interrupt
> handlers to finish.
>
> However if - as is common - the handler queues a work request which is
> flushed during the cleanup stage, we have code adding buffers / trying
> to get buffers while device is reset. Not good.
>
> This was reproduced by running
> modprobe virtio_console
> modprobe -r virtio_console
> in a loop.
That's a pathological case that is not "in the field" except by people
who want to abuse the system as root. And they can do much worse things
than that.
> Fixing this comprehensively needs some thought, and new APIs.
> Let's at least handle the specific case of virtio_console
> removal that was reported in the field.
Let's fix this correctly, don't just hack it up now.
> Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1786239
> Signed-off-by: Michael S. Tsirkin <[email protected]>
> ---
> drivers/char/virtio_console.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
> index 7eaf303a7a86..c852ce0b4d56 100644
> --- a/drivers/char/virtio_console.c
> +++ b/drivers/char/virtio_console.c
> @@ -1956,6 +1956,12 @@ static void virtcons_remove(struct virtio_device *vdev)
> list_del(&portdev->list);
> spin_unlock_irq(&pdrvdata_lock);
>
> + /* Device is going away, exit any polling for buffers */
> + virtio_break_device(vdev);
> + if (use_multiport(portdev))
> + flush_work(&portdev->control_work);
> + else
> + flush_work(&portdev->config_work);
> /* Disable interrupts for vqs */
newline before comment?
thanks,
greg k-h
On Tue, Oct 05, 2021 at 03:33:42PM +0200, Greg Kroah-Hartman wrote:
> On Tue, Oct 05, 2021 at 03:04:07AM -0400, Michael S. Tsirkin wrote:
> > A common pattern for device reset is currently:
> > vdev->config->reset(vdev);
> > .. cleanup ..
> >
> > reset prevents new interrupts from arriving and waits for interrupt
> > handlers to finish.
> >
> > However if - as is common - the handler queues a work request which is
> > flushed during the cleanup stage, we have code adding buffers / trying
> > to get buffers while device is reset. Not good.
> >
> > This was reproduced by running
> > modprobe virtio_console
> > modprobe -r virtio_console
> > in a loop.
>
> That's a pathological case that is not "in the field" except by people
> who want to abuse the system as root. And they can do much worse things
> than that.
>
> > Fixing this comprehensively needs some thought, and new APIs.
> > Let's at least handle the specific case of virtio_console
> > removal that was reported in the field.
>
> Let's fix this correctly, don't just hack it up now.
Well I poked at it some more, and things are not as bad
as I thought. It's mostly just console and possibly virtio-mem.
Well and virtio-bt has a completely borken cleanup that
does not even bother to reset the device, but that's
a separate issue, discussing it with the maintainer.
So I wrote some patches to document the requirements better, added a
wrapper for reset and generally cleaned the API up a bit, and added a
patch for mem, but generally I no longer think we need a major API
change.
> > Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1786239
> > Signed-off-by: Michael S. Tsirkin <[email protected]>
> > ---
> > drivers/char/virtio_console.c | 6 ++++++
> > 1 file changed, 6 insertions(+)
> >
> > diff --git a/drivers/char/virtio_console.c b/drivers/char/virtio_console.c
> > index 7eaf303a7a86..c852ce0b4d56 100644
> > --- a/drivers/char/virtio_console.c
> > +++ b/drivers/char/virtio_console.c
> > @@ -1956,6 +1956,12 @@ static void virtcons_remove(struct virtio_device *vdev)
> > list_del(&portdev->list);
> > spin_unlock_irq(&pdrvdata_lock);
> >
> > + /* Device is going away, exit any polling for buffers */
> > + virtio_break_device(vdev);
> > + if (use_multiport(portdev))
> > + flush_work(&portdev->control_work);
> > + else
> > + flush_work(&portdev->config_work);
> > /* Disable interrupts for vqs */
>
> newline before comment?
sure
> thanks,
>
> greg k-h