Re: [PATCH 1/1] hw/ide/core: terminate in-flight DMA on IDE bus reset

[Fiona Ebner]

Am 26.09.23 um 16:45 schrieb John Snow:
> 
> 
> On Tue, Sep 26, 2023, 3:11 AM Fiona Ebner <f.ebner@proxmox.com
> <mailto:f.ebner@proxmox.com>> wrote:
> 
>     Am 25.09.23 um 21:53 schrieb John Snow:
>     > On Thu, Sep 21, 2023 at 12:07 PM Simon Rowe
>     <simon.rowe@nutanix.com <mailto:simon.rowe@nutanix.com>> wrote:
>     >>
>     >> When an IDE controller is reset, its internal state is being cleared
>     >> before any outstanding I/O is cancelled. If a response to DMA is
>     >> received in this window, the aio callback will incorrectly continue
>     >> with the next part of the transfer (now using sector 0 from
>     >> the cleared controller state).
>     >
>     > Eugh, yikes. It feels like we should fix the cancellation ...
>     Please note that there already is a patch for that on the list:
>     https://lists.nongnu.org/archive/html/qemu-devel/2023-09/msg01011.html 
> <https://lists.nongnu.org/archive/html/qemu-devel/2023-09/msg01011.html>
> 
>     Best Regards,
>     Fiona
> 
> 
> Gotcha, thanks for the pointer. I wonder if that's sufficient to fix the
> CVE here? I don't have the reproducer in my hands (that I know of ...
> it's genuinely possible I missed it, apologies)
> 

AFAICT, yes, because the DMA callback is invoked before resetting the
state now. But not 100% sure if it can't be triggered in some other way,
maybe Simon knows more? I don't have a reproducer for the CVE either,
but the second patch after the one linked above adds a qtest for the
reset scenario.

Best Regards,
Fiona