Re: [Qemu-devel] [PULL 1/3] usb-mtp: Add some NULL checks for issues pointed out by coverity

[Bandan Das <bsd@redhat.com>]

Peter Maydell <peter.maydell@linaro.org> writes:

> On 7 May 2018 at 10:44, Gerd Hoffmann <kraxel@redhat.com> wrote:
>> From: Bandan Das <bsd@redhat.com>
>>
>> CID 1390578: In usb_mtp_write_metadata, parent can never be NULL but
>> just in case, add an assert
>> CID 1390592: Check for o->format only if o !=NULL
>> CID 1390604: Check s->data_out != NULL in usb_mtp_handle_data
>
>> @@ -1838,7 +1838,7 @@ static void usb_mtp_handle_data(USBDevice *dev, USBPacket *p)
>>              p->status = USB_RET_STALL;
>>              return;
>>          }
>> -        if (s->data_out && !s->data_out->first) {
>> +        if ((s->data_out != NULL) && !s->data_out->first) {
>>              container_type = TYPE_DATA;
>>          } else {
>>              usb_packet_copy(p, &container, sizeof(container));
>> --
>
> I just noticed this, but this isn't actually changing the logic
> in this function: "s->data_out" and "s->data_out != NULL" are
> exactly the same test. So this won't change CID 1390604.

Right, indeed they are. I am unfamiliar with coverity and just
incorrectly assumed that somehow it's assuming the test for NULL
can return false.

> Looking back at my previous email on this, it looks like I
> managed to completely misinterpret the code and/or what
> coverity is talking about, which is probably the source of
> the confusion. Let me try again...
>
> In the function usb_mtp_handle_data() EP_DATA_OUT case, we have:
>  (1) a check for whether s->data_out is NULL, which implies
>      that it might be NULL sometimes
>  (2) some time later, a call to usb_mtp_get_data() which is
>      not protected by a test for whether s->data_out is NULL
>
> and if s->data_out is NULL at point (2) then usb_mtp_get_data()
> will crash.
>
> Obviously the code path that goes through "containe_type = TYPE_DATA"
> must have s->data_out being non-NULL. But in the else branch of
> that, can the container_type we get via usb_packet_copy() ever
> be TYPE_DATA ? (It's not clear to me whether that comes from
> a trusted or untrusted source.)
>
> If this is a "can't happen" situation we can mark it as a false
> positive in coverity.

The protocol ofcourse won't let this happen but the guest can't be
trusted. It can easily send a packet with TYPE_DATA without sending
OBJECT_INFO first that allocates memory for data_out. I will submit a
fix.

Thanks for clearing out the confusion.

Bandan

> thanks
> -- PMM