[PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[Gustavo A. R. Silva]

`struct urb` is a flexible structure, which means that it contains a
flexible-array member at the bottom. This could potentially lead to an
overwrite of the object `wq` at run-time with the contents of `urb`.

Fix this by placing object `urb` at the end of `struct smsusb_urb_t`.

Fixes: dd47fbd40e6e ("[media] smsusb: don't sleep while atomic")
Cc: stable@vger.kernel.org
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
---
 drivers/media/usb/siano/smsusb.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/media/usb/siano/smsusb.c b/drivers/media/usb/siano/smsusb.c
index 9d9e14c858e6..2c048f8e8371 100644
--- a/drivers/media/usb/siano/smsusb.c
+++ b/drivers/media/usb/siano/smsusb.c
@@ -40,10 +40,10 @@ struct smsusb_urb_t {
 	struct smscore_buffer_t *cb;
 	struct smsusb_device_t *dev;
 
-	struct urb urb;
-
 	/* For the bottom half */
 	struct work_struct wq;
+
+	struct urb urb;
 };
 
 struct smsusb_device_t {
-- 
2.34.1

Re: [PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[Jann Horn]

On Fri, Sep 29, 2023 at 5:42 PM Gustavo A. R. Silva
<gustavoars@kernel.org> wrote:
> `struct urb` is a flexible structure, which means that it contains a
> flexible-array member at the bottom. This could potentially lead to an
> overwrite of the object `wq` at run-time with the contents of `urb`.
>
> Fix this by placing object `urb` at the end of `struct smsusb_urb_t`.

Does this really change the situation? "struct smsusb_device_t"
contains an array of "struct smsusb_urb_t", so it seems to be like
you're just shifting the "VLA inside a non-final member of a struct"
thing around so that there is one more layer of abstraction in
between.

Comments on "struct urb" say:

 * Isochronous URBs have a different data transfer model, in part because
 * the quality of service is only "best effort".  Callers provide specially
 * allocated URBs, with number_of_packets worth of iso_frame_desc structures
 * at the end.

and:

/* (in) ISO ONLY */

And it looks like smsusb only uses that URB as a bulk URB, so the flex
array is unused and we can't have an overflow here?

If this is intended to make it possible to enable some kinda compiler
warning, it might be worth talking to the USB folks to figure out the
right approach here.

> Fixes: dd47fbd40e6e ("[media] smsusb: don't sleep while atomic")
> Cc: stable@vger.kernel.org
> Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
> ---
>  drivers/media/usb/siano/smsusb.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/media/usb/siano/smsusb.c b/drivers/media/usb/siano/smsusb.c
> index 9d9e14c858e6..2c048f8e8371 100644
> --- a/drivers/media/usb/siano/smsusb.c
> +++ b/drivers/media/usb/siano/smsusb.c
> @@ -40,10 +40,10 @@ struct smsusb_urb_t {
>         struct smscore_buffer_t *cb;
>         struct smsusb_device_t *dev;
>
> -       struct urb urb;
> -
>         /* For the bottom half */
>         struct work_struct wq;
> +
> +       struct urb urb;
>  };
>
>  struct smsusb_device_t {
> --
> 2.34.1
>
>

Re: [PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[Greg Kroah-Hartman]

On Fri, Sep 29, 2023 at 06:20:10PM +0200, Jann Horn wrote:
> On Fri, Sep 29, 2023 at 5:42 PM Gustavo A. R. Silva
> <gustavoars@kernel.org> wrote:
> > `struct urb` is a flexible structure, which means that it contains a
> > flexible-array member at the bottom. This could potentially lead to an
> > overwrite of the object `wq` at run-time with the contents of `urb`.
> >
> > Fix this by placing object `urb` at the end of `struct smsusb_urb_t`.
> 
> Does this really change the situation? "struct smsusb_device_t"
> contains an array of "struct smsusb_urb_t", so it seems to be like
> you're just shifting the "VLA inside a non-final member of a struct"
> thing around so that there is one more layer of abstraction in
> between.
> 
> Comments on "struct urb" say:
> 
>  * Isochronous URBs have a different data transfer model, in part because
>  * the quality of service is only "best effort".  Callers provide specially
>  * allocated URBs, with number_of_packets worth of iso_frame_desc structures
>  * at the end.
> 
> and:
> 
> /* (in) ISO ONLY */
> 
> And it looks like smsusb only uses that URB as a bulk URB, so the flex
> array is unused and we can't have an overflow here?
> 
> If this is intended to make it possible to enable some kinda compiler
> warning, it might be worth talking to the USB folks to figure out the
> right approach here.
> 
> > Fixes: dd47fbd40e6e ("[media] smsusb: don't sleep while atomic")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
> > ---
> >  drivers/media/usb/siano/smsusb.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/media/usb/siano/smsusb.c b/drivers/media/usb/siano/smsusb.c
> > index 9d9e14c858e6..2c048f8e8371 100644
> > --- a/drivers/media/usb/siano/smsusb.c
> > +++ b/drivers/media/usb/siano/smsusb.c
> > @@ -40,10 +40,10 @@ struct smsusb_urb_t {
> >         struct smscore_buffer_t *cb;
> >         struct smsusb_device_t *dev;
> >
> > -       struct urb urb;
> > -
> >         /* For the bottom half */
> >         struct work_struct wq;
> > +
> > +       struct urb urb;
> >  };

Yeah, this is going to get messy.  Ideally, just dynamically create the
urb and change this to a "struct urb *urb;" instead.

thanks,

greg k-h

Re: [PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[Gustavo A. R. Silva]

On 9/30/23 09:01, Greg Kroah-Hartman wrote:
> On Fri, Sep 29, 2023 at 06:20:10PM +0200, Jann Horn wrote:
>> On Fri, Sep 29, 2023 at 5:42 PM Gustavo A. R. Silva
>> <gustavoars@kernel.org> wrote:
>>> `struct urb` is a flexible structure, which means that it contains a
>>> flexible-array member at the bottom. This could potentially lead to an
>>> overwrite of the object `wq` at run-time with the contents of `urb`.
>>>
>>> Fix this by placing object `urb` at the end of `struct smsusb_urb_t`.
>>
>> Does this really change the situation? "struct smsusb_device_t"
>> contains an array of "struct smsusb_urb_t", so it seems to be like

Yeah. I noticed that too.

Probably what Greg suggests (dynamically create the urb) can fix this, too.

I haven't taken a deep dive into this particular case. So, let me go and
figure something out.

>> you're just shifting the "VLA inside a non-final member of a struct"
>> thing around so that there is one more layer of abstraction in
>> between.
>>
>> Comments on "struct urb" say:
>>
>>   * Isochronous URBs have a different data transfer model, in part because
>>   * the quality of service is only "best effort".  Callers provide specially
>>   * allocated URBs, with number_of_packets worth of iso_frame_desc structures
>>   * at the end.
>>
>> and:
>>
>> /* (in) ISO ONLY */
>>
>> And it looks like smsusb only uses that URB as a bulk URB, so the flex
>> array is unused and we can't have an overflow here?
>>
>> If this is intended to make it possible to enable some kinda compiler
>> warning, it might be worth talking to the USB folks to figure out the
>> right approach here.
>>
>>> Fixes: dd47fbd40e6e ("[media] smsusb: don't sleep while atomic")
>>> Cc: stable@vger.kernel.org
>>> Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
>>> ---
>>>   drivers/media/usb/siano/smsusb.c | 4 ++--
>>>   1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/media/usb/siano/smsusb.c b/drivers/media/usb/siano/smsusb.c
>>> index 9d9e14c858e6..2c048f8e8371 100644
>>> --- a/drivers/media/usb/siano/smsusb.c
>>> +++ b/drivers/media/usb/siano/smsusb.c
>>> @@ -40,10 +40,10 @@ struct smsusb_urb_t {
>>>          struct smscore_buffer_t *cb;
>>>          struct smsusb_device_t *dev;
>>>
>>> -       struct urb urb;
>>> -
>>>          /* For the bottom half */
>>>          struct work_struct wq;
>>> +
>>> +       struct urb urb;
>>>   };
> 
> Yeah, this is going to get messy.  Ideally, just dynamically create the
> urb and change this to a "struct urb *urb;" instead.

Probably, yes.

Thanks
--
Gustavo

Re: [PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[Kees Cook]

On Fri, Sep 29, 2023 at 05:42:11PM +0200, Gustavo A. R. Silva wrote:
> `struct urb` is a flexible structure, which means that it contains a
> flexible-array member at the bottom. This could potentially lead to an
> overwrite of the object `wq` at run-time with the contents of `urb`.
> 
> Fix this by placing object `urb` at the end of `struct smsusb_urb_t`.
> 
> Fixes: dd47fbd40e6e ("[media] smsusb: don't sleep while atomic")
> Cc: stable@vger.kernel.org
> Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>

As Jann pointed out, it's unlikely there is a function bug here, but I
still think it's right to make sure this is robust and clears the way
for -Wflex-array-member-not-at-end.

Reviewed-by: Kees Cook <keescook@chromium.org>

-- 
Kees Cook

Re: [PATCH][next] media: usb: siano: Fix undefined behavior bug in struct smsusb_urb_t

[Jann Horn]

On Fri, Sep 29, 2023 at 7:29 PM Kees Cook <keescook@chromium.org> wrote:
> On Fri, Sep 29, 2023 at 05:42:11PM +0200, Gustavo A. R. Silva wrote:
> > `struct urb` is a flexible structure, which means that it contains a
> > flexible-array member at the bottom. This could potentially lead to an
> > overwrite of the object `wq` at run-time with the contents of `urb`.
> >
> > Fix this by placing object `urb` at the end of `struct smsusb_urb_t`.
> >
> > Fixes: dd47fbd40e6e ("[media] smsusb: don't sleep while atomic")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
>
> As Jann pointed out, it's unlikely there is a function bug here, but I
> still think it's right to make sure this is robust and clears the way
> for -Wflex-array-member-not-at-end.

But if this change makes the warning go away, that just means the
warning is implemented badly, right?

Like, before we had:

struct urb {
  ...
  struct usb_iso_packet_descriptor iso_frame_desc[];
};
struct smsusb_urb_t {
  ...
  struct urb urb; // not last element
  ...
};

whereas afterwards we have:

struct urb {
  ...
  struct usb_iso_packet_descriptor iso_frame_desc[];
};
struct smsusb_urb_t {
  ...
  struct urb urb;
};
struct smsusb_device_t {
  ...
  struct smsusb_urb_t surbs[MAX_URBS]; // array, and not last element
  ...
};

That's basically the same pattern! Except that the new version has one
more layer of indirection and there's an array involved.

And you can't address that by moving struct members around because of
the involvement of the array.