re: media: dvb-usb-v2/gl861: ensure USB message buffers DMA'able

re: media: dvb-usb-v2/gl861: ensure USB message buffers DMA'able

[Colin Ian King]

Hi,

While running static analysis on linux-next with CoverityScan an issue
was detected with recent commit:

commit 86f65c218123c4e36fd855fbbc38147ffaf29974
Author: Akihiro Tsukada <tskd08@gmail.com>
Date:   Sun Apr 8 13:21:38 2018 -0400

    media: dvb-usb-v2/gl861: ensure USB message buffers DMA'able

The report is as follows:

46        }

assign_zero: Assigning: buf = NULL.

 47        buf = NULL;

Condition rlen > 0, taking false branch.

 48        if (rlen > 0) {
 49                buf = kmalloc(rlen, GFP_KERNEL);
 50                if (!buf)
 51                        return -ENOMEM;
 52        }

 53        usleep_range(1000, 2000); /* avoid I2C errors */
 54
   CID 1470241 (#1 of 1): Explicit null dereferenced (FORWARD_NULL).
var_deref_model: Passing null pointer buf to usb_control_msg, which
dereferences it.

 55        ret = usb_control_msg(d->udev, usb_rcvctrlpipe(d->udev, 0),
req, type,
 56                              value, index, buf, rlen, 2000);


The assignment of buf = NULL means a null buffer is passed down the usb
control message stack until it eventually gets dereferenced. This only
occurs when rlen <= 0.   I was unsure how to fix this for the case when
rlen <= 0, so I am flagging this up as an issue that needs fixing.

Regards,

Colin

Re: media: dvb-usb-v2/gl861: ensure USB message buffers DMA'able

[Akihiro TSUKADA]

Hi,
thanks for the report.

>  47        buf = NULL;
> 
> Condition rlen > 0, taking false branch.
> 
>  48        if (rlen > 0) {
>  49                buf = kmalloc(rlen, GFP_KERNEL);
>  50                if (!buf)
>  51                        return -ENOMEM;
>  52        }
> 
>  53        usleep_range(1000, 2000); /* avoid I2C errors */
>  54
>    CID 1470241 (#1 of 1): Explicit null dereferenced (FORWARD_NULL).
> var_deref_model: Passing null pointer buf to usb_control_msg, which
> dereferences it.
> 
>  55        ret = usb_control_msg(d->udev, usb_rcvctrlpipe(d->udev, 0),
> req, type,
>  56                              value, index, buf, rlen, 2000);
> 
> 
> The assignment of buf = NULL means a null buffer is passed down the usb
> control message stack until it eventually gets dereferenced. This only
> occurs when rlen <= 0.   I was unsure how to fix this for the case when
> rlen <= 0, so I am flagging this up as an issue that needs fixing.
> 

Since rlen is an u16, null pointer is passed only when rlen == 0,
so I think it is not a problem,
but I am OK to add a guard in order to make scan result clean.

regards,
Akihiro

Re: media: dvb-usb-v2/gl861: ensure USB message buffers DMA'able

[Mauro Carvalho Chehab]

Em Tue, 3 Jul 2018 21:07:07 +0900
Akihiro TSUKADA <tskd08@gmail.com> escreveu:

> Hi,
> thanks for the report.
> 
> >  47        buf = NULL;
> > 
> > Condition rlen > 0, taking false branch.
> > 
> >  48        if (rlen > 0) {
> >  49                buf = kmalloc(rlen, GFP_KERNEL);
> >  50                if (!buf)
> >  51                        return -ENOMEM;
> >  52        }
> > 
> >  53        usleep_range(1000, 2000); /* avoid I2C errors */
> >  54
> >    CID 1470241 (#1 of 1): Explicit null dereferenced (FORWARD_NULL).
> > var_deref_model: Passing null pointer buf to usb_control_msg, which
> > dereferences it.
> > 
> >  55        ret = usb_control_msg(d->udev, usb_rcvctrlpipe(d->udev, 0),
> > req, type,
> >  56                              value, index, buf, rlen, 2000);
> > 
> > 
> > The assignment of buf = NULL means a null buffer is passed down the usb
> > control message stack until it eventually gets dereferenced. This only
> > occurs when rlen <= 0.   I was unsure how to fix this for the case when
> > rlen <= 0, so I am flagging this up as an issue that needs fixing.
> >   
> 
> Since rlen is an u16, null pointer is passed only when rlen == 0,
> so I think it is not a problem,
> but I am OK to add a guard in order to make scan result clean.

There was another patch proposed to fix this issue with does the
right thing when rlen == 0. I rebased it on the top of the current
tree:
	https://git.linuxtv.org/media_tree.git/commit/?id=0b666e1c8120c0b17a8a68aaed58e22011f06ab3

That should cover both cases.

Thanks,
Mauro

Re: media: dvb-usb-v2/gl861: ensure USB message buffers DMA'able

[Akihiro TSUKADA]

Hi,

On 2018年07月25日 22:57, Mauro Carvalho Chehab wrote:
...
> There was another patch proposed to fix this issue with does the
> right thing when rlen == 0. I rebased it on the top of the current
> tree:
> 	https://git.linuxtv.org/media_tree.git/commit/?id=0b666e1c8120c0b17a8a68aaed58e22011f06ab3
> 
> That should cover both cases.

When wlen is checked to be <= 2 and
wbuf[0],wbuf[1] are already used in 'index','value',
why this patch copys wbuf and passes it again in usb_control_msg when wo==1 ?
(just to silence static analiyzers?)

Furthermore, 
I am afraid that a static analyzer might warn on a possible
buffer overrun in usb_control_msg in the case of rbuf == NULL && rlen > 2,
since buf's length is passed as rlen but is actually wlen.

regards,
Akihiro