Re: [PULL 6/7] usb/xhci: sanity check packet size (CVE-2021-3527)

[Remy Noel]

Hello

On Tue, May 04, 2021 at 10:53:16AM +0200, Gerd Hoffmann wrote:
> Make sure the usb packet size is within the
> bounds of the endpoint configuration.
>
> Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
> Message-Id: <20210503132915.2335822-5-kraxel@redhat.com>
> ---
> hw/usb/hcd-xhci.c | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/hw/usb/hcd-xhci.c b/hw/usb/hcd-xhci.c
> index 46212b1e695a..7acfb8137bc9 100644
> --- a/hw/usb/hcd-xhci.c
> +++ b/hw/usb/hcd-xhci.c
> @@ -1568,6 +1568,11 @@ static int xhci_setup_packet(XHCITransfer *xfer)
>         qemu_sglist_destroy(&xfer->sgl);
>         return -1;
>     }
> +    if (xfer->packet.iov.size > ep->max_packet_size) {
> +        usb_packet_unmap(&xfer->packet, &xfer->sgl);
> +        qemu_sglist_destroy(&xfer->sgl);
> +        return -1;
> +    }
>     DPRINTF("xhci: setup packet pid 0x%x addr %d ep %d\n",
>             xfer->packet.pid, ep->dev->addr, ep->nr);
>     return 0;
> --
So im my user's case (using a usb-Display-port adapter) i managed to trigger
this error.
According to him. the his displays works without this patch (but i can see 
him sending usbredir packets of up to 9MB) but breaks without (which is 
expected.)
I have trouble understanding whether this can be considered an illegitimate 
xhci use case (i'm very unfamiliar with it), but i don't see any burst 
working if we drop any buffer chain bigger than a single endpoint packet 
size.

I'm strill struggling to emulate SuperSpeed Bulk transfers from the vm in 
order to have a reproducer though.

By the way, the call to qemu_sglist_destroy seems unnecessary to me since 
usb_packet_unmap calls it already.


Cheers

Remy