diff mbox

[v2] drm/dumb-buffers: Integer overflow in drm_mode_create_ioctl()

Message ID 20180516140026.GA19340@mwanda (mailing list archive)
State New, archived
Headers show

Commit Message

Dan Carpenter May 16, 2018, 2 p.m. UTC
There is a comment here which says that DIV_ROUND_UP() and that's where
the problem comes from.  Say you pick:

	args->bpp = UINT_MAX - 7;
	args->width = 4;
	args->height = 1;

The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
because of how we picked args->width that means cpp < UINT_MAX / 4.

I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
removed the check for !cpp because it's not possible after this change.
I also changed all the 0xffffffffU references to U32_MAX.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
---
v2:  additional cleanups

Comments

Chris Wilson May 16, 2018, 2:26 p.m. UTC | #1
Quoting Dan Carpenter (2018-05-16 15:00:26)
> There is a comment here which says that DIV_ROUND_UP() and that's where
> the problem comes from.  Say you pick:
> 
>         args->bpp = UINT_MAX - 7;
>         args->width = 4;
>         args->height = 1;
> 
> The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
> because of how we picked args->width that means cpp < UINT_MAX / 4.
> 
> I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
> removed the check for !cpp because it's not possible after this change.
> I also changed all the 0xffffffffU references to U32_MAX.
> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>

I agree with Daniel that the !cpp check after DIV_ROUND_UP was
sufficient to guard the current code, but switching to a more idiomatic
style of overflow checking has its benefits too. Plus I like having
U32_MAX to compare the type ranges against.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
-Chris
Dan Carpenter May 16, 2018, 2:52 p.m. UTC | #2
On Wed, May 16, 2018 at 03:26:07PM +0100, Chris Wilson wrote:
> Quoting Dan Carpenter (2018-05-16 15:00:26)
> > There is a comment here which says that DIV_ROUND_UP() and that's where
> > the problem comes from.  Say you pick:
> > 
> >         args->bpp = UINT_MAX - 7;
> >         args->width = 4;
> >         args->height = 1;
> > 
> > The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
> > because of how we picked args->width that means cpp < UINT_MAX / 4.
> > 
> > I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
> > removed the check for !cpp because it's not possible after this change.
> > I also changed all the 0xffffffffU references to U32_MAX.
> > 
> > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> 
> I agree with Daniel that the !cpp check after DIV_ROUND_UP was
> sufficient to guard the current code, but switching to a more idiomatic
> style of overflow checking has its benefits too. Plus I like having
> U32_MAX to compare the type ranges against.
> 

I'm not totally sure what it means to say that the integer overflow is
sufficient.  The overflow check is definitely buggy but if you mean that
it isn't exploitable then you're probably right.  Anyway, let's say you
use use the values I provided in my changelog.  Then I believe we can
reach vgem_gem_dumb_create():

static int vgem_gem_dumb_create(struct drm_file *file, struct drm_device *dev,
   205                                  struct drm_mode_create_dumb *args)
   206  {
   207          struct drm_gem_object *gem_object;
   208          u64 pitch, size;
   209  
   210          pitch = args->width * DIV_ROUND_UP(args->bpp, 8);
                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^
This is now:	pitch = 4           * (U32_MAX / 8);

I would get a static checker warning here because of the integer
overflow in DIV_ROUND_UP().  I'll push that check soon.

   211          size = args->height * pitch;

This is now:	size = 1 * U32_MAX / 2;

   212          if (size == 0)
   213                  return -EINVAL;
   214  
   215          gem_object = vgem_gem_create(dev, file, &args->handle, size);

Then this fails because we can't allocate U32_MAX / 2 bytes.  There
are some other potential numbers we could try instead of the ones I
gave.  On 32bit arches  __vgem_gem_create() has integer overflows if
size is over U32_MAX - PAGE_SIZE so it gets a bit complicated...

   216          if (IS_ERR(gem_object))
   217                  return PTR_ERR(gem_object);
   218  
   219          args->size = gem_object->size;
   220          args->pitch = pitch;
   221  
   222          DRM_DEBUG_DRIVER("Created object of size %lld\n", size);
   223  
   224          return 0;
   225  }

regards,
dan carpenter
Chris Wilson May 16, 2018, 2:56 p.m. UTC | #3
Quoting Dan Carpenter (2018-05-16 15:52:57)
> On Wed, May 16, 2018 at 03:26:07PM +0100, Chris Wilson wrote:
> > Quoting Dan Carpenter (2018-05-16 15:00:26)
> > > There is a comment here which says that DIV_ROUND_UP() and that's where
> > > the problem comes from.  Say you pick:
> > > 
> > >         args->bpp = UINT_MAX - 7;
> > >         args->width = 4;
> > >         args->height = 1;
> > > 
> > > The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
> > > because of how we picked args->width that means cpp < UINT_MAX / 4.
> > > 
> > > I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
> > > removed the check for !cpp because it's not possible after this change.
> > > I also changed all the 0xffffffffU references to U32_MAX.
> > > 
> > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> > 
> > I agree with Daniel that the !cpp check after DIV_ROUND_UP was
> > sufficient to guard the current code, but switching to a more idiomatic
> > style of overflow checking has its benefits too. Plus I like having
> > U32_MAX to compare the type ranges against.
> > 
> 
> I'm not totally sure what it means to say that the integer overflow is
> sufficient.  The overflow check is definitely buggy but if you mean that
> it isn't exploitable then you're probably right.  Anyway, let's say you
> use use the values I provided in my changelog.  Then I believe we can
> reach vgem_gem_dumb_create():

But we are talking about

	cpp = (U32_MAX + 8) / 8

which is 0. So the !cpp does catch the overflow.

Or am I completely off in unsigned wraparound?
-Chris
Dan Carpenter May 16, 2018, 3:15 p.m. UTC | #4
On Wed, May 16, 2018 at 03:56:55PM +0100, Chris Wilson wrote:
> Quoting Dan Carpenter (2018-05-16 15:52:57)
> > On Wed, May 16, 2018 at 03:26:07PM +0100, Chris Wilson wrote:
> > > Quoting Dan Carpenter (2018-05-16 15:00:26)
> > > > There is a comment here which says that DIV_ROUND_UP() and that's where
> > > > the problem comes from.  Say you pick:
> > > > 
> > > >         args->bpp = UINT_MAX - 7;
> > > >         args->width = 4;
> > > >         args->height = 1;
> > > > 
> > > > The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
> > > > because of how we picked args->width that means cpp < UINT_MAX / 4.
> > > > 
> > > > I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
> > > > removed the check for !cpp because it's not possible after this change.
> > > > I also changed all the 0xffffffffU references to U32_MAX.
> > > > 
> > > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> > > 
> > > I agree with Daniel that the !cpp check after DIV_ROUND_UP was
> > > sufficient to guard the current code, but switching to a more idiomatic
> > > style of overflow checking has its benefits too. Plus I like having
> > > U32_MAX to compare the type ranges against.
> > > 
> > 
> > I'm not totally sure what it means to say that the integer overflow is
> > sufficient.  The overflow check is definitely buggy but if you mean that
> > it isn't exploitable then you're probably right.  Anyway, let's say you
> > use use the values I provided in my changelog.  Then I believe we can
> > reach vgem_gem_dumb_create():
> 
> But we are talking about
> 
> 	cpp = (U32_MAX + 8) / 8
> 

No, no.  That's not the problem.  The problem is the - 1 subtraction.  I
explained poorly.  The issue is when "bpp" is U32_MAX - 7.  The define
looks like:

#define __KERNEL_DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))

So "n" is (U32_MAX - 7) and "d" is 8.  "(n) + (d)" wraps to zero.  We do
the "- 1" subtraction so that gives us U32_MAX.  And finally we divide
by "d" to get "U32_MAX / 8".

regards,
dan carpenter
Chris Wilson May 16, 2018, 3:25 p.m. UTC | #5
Quoting Dan Carpenter (2018-05-16 16:15:54)
> On Wed, May 16, 2018 at 03:56:55PM +0100, Chris Wilson wrote:
> > Quoting Dan Carpenter (2018-05-16 15:52:57)
> > > On Wed, May 16, 2018 at 03:26:07PM +0100, Chris Wilson wrote:
> > > > Quoting Dan Carpenter (2018-05-16 15:00:26)
> > > > > There is a comment here which says that DIV_ROUND_UP() and that's where
> > > > > the problem comes from.  Say you pick:
> > > > > 
> > > > >         args->bpp = UINT_MAX - 7;
> > > > >         args->width = 4;
> > > > >         args->height = 1;
> > > > > 
> > > > > The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
> > > > > because of how we picked args->width that means cpp < UINT_MAX / 4.
> > > > > 
> > > > > I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
> > > > > removed the check for !cpp because it's not possible after this change.
> > > > > I also changed all the 0xffffffffU references to U32_MAX.
> > > > > 
> > > > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> > > > 
> > > > I agree with Daniel that the !cpp check after DIV_ROUND_UP was
> > > > sufficient to guard the current code, but switching to a more idiomatic
> > > > style of overflow checking has its benefits too. Plus I like having
> > > > U32_MAX to compare the type ranges against.
> > > > 
> > > 
> > > I'm not totally sure what it means to say that the integer overflow is
> > > sufficient.  The overflow check is definitely buggy but if you mean that
> > > it isn't exploitable then you're probably right.  Anyway, let's say you
> > > use use the values I provided in my changelog.  Then I believe we can
> > > reach vgem_gem_dumb_create():
> > 
> > But we are talking about
> > 
> >       cpp = (U32_MAX + 8) / 8
> > 
> 
> No, no.  That's not the problem.  The problem is the - 1 subtraction.  I
> explained poorly.  The issue is when "bpp" is U32_MAX - 7.  The define
> looks like:
> 
> #define __KERNEL_DIV_ROUND_UP(n, d) (((n) + (d) - 1) / (d))
> 
> So "n" is (U32_MAX - 7) and "d" is 8.  "(n) + (d)" wraps to zero.  We do
> the "- 1" subtraction so that gives us U32_MAX.  And finally we divide
> by "d" to get "U32_MAX / 8".

D'oh. Thanks for spelling it out.
-Chris
Dan Carpenter May 16, 2018, 3:41 p.m. UTC | #6
Btw, I've looked at this some more and I'm 99% sure there is no way to
exploit it.  The "if (PAGE_ALIGN(size) == 0)" prevents the integer
overflow in __vgem_gem_create() that I was worried about.

regards,
dan carpenter
Daniel Vetter May 16, 2018, 3:57 p.m. UTC | #7
On Wed, May 16, 2018 at 05:00:26PM +0300, Dan Carpenter wrote:
> There is a comment here which says that DIV_ROUND_UP() and that's where
> the problem comes from.  Say you pick:
> 
> 	args->bpp = UINT_MAX - 7;
> 	args->width = 4;
> 	args->height = 1;
> 
> The integer overflow in DIV_ROUND_UP() means "cpp" is UINT_MAX / 8 and
> because of how we picked args->width that means cpp < UINT_MAX / 4.
> 
> I've fixed it by preventing the integer overflow in DIV_ROUND_UP().  I
> removed the check for !cpp because it's not possible after this change.
> I also changed all the 0xffffffffU references to U32_MAX.
> 
> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> ---
> v2:  additional cleanups

Thanks a lot for respinning, applied to drm-misc-fixes.
-Daniel

> 
> diff --git a/drivers/gpu/drm/drm_dumb_buffers.c b/drivers/gpu/drm/drm_dumb_buffers.c
> index 39ac15ce4702..9e2ae02f31e0 100644
> --- a/drivers/gpu/drm/drm_dumb_buffers.c
> +++ b/drivers/gpu/drm/drm_dumb_buffers.c
> @@ -65,12 +65,13 @@ int drm_mode_create_dumb_ioctl(struct drm_device *dev,
>  		return -EINVAL;
>  
>  	/* overflow checks for 32bit size calculations */
> -	/* NOTE: DIV_ROUND_UP() can overflow */
> +	if (args->bpp > U32_MAX - 8)
> +		return -EINVAL;
>  	cpp = DIV_ROUND_UP(args->bpp, 8);
> -	if (!cpp || cpp > 0xffffffffU / args->width)
> +	if (cpp > U32_MAX / args->width)
>  		return -EINVAL;
>  	stride = cpp * args->width;
> -	if (args->height > 0xffffffffU / stride)
> +	if (args->height > U32_MAX / stride)
>  		return -EINVAL;
>  
>  	/* test for wrap-around */
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel
diff mbox

Patch

diff --git a/drivers/gpu/drm/drm_dumb_buffers.c b/drivers/gpu/drm/drm_dumb_buffers.c
index 39ac15ce4702..9e2ae02f31e0 100644
--- a/drivers/gpu/drm/drm_dumb_buffers.c
+++ b/drivers/gpu/drm/drm_dumb_buffers.c
@@ -65,12 +65,13 @@  int drm_mode_create_dumb_ioctl(struct drm_device *dev,
 		return -EINVAL;
 
 	/* overflow checks for 32bit size calculations */
-	/* NOTE: DIV_ROUND_UP() can overflow */
+	if (args->bpp > U32_MAX - 8)
+		return -EINVAL;
 	cpp = DIV_ROUND_UP(args->bpp, 8);
-	if (!cpp || cpp > 0xffffffffU / args->width)
+	if (cpp > U32_MAX / args->width)
 		return -EINVAL;
 	stride = cpp * args->width;
-	if (args->height > 0xffffffffU / stride)
+	if (args->height > U32_MAX / stride)
 		return -EINVAL;
 
 	/* test for wrap-around */