fbcon: Check font dimension limits

Commit Message

Samuel Thibault Jan. 26, 2023, 12:49 a.m. UTC

blit_x and blit_y are uint32_t, so fbcon currently cannot support fonts
larger than 32x32.

The 32x32 case also needs shifting an unsigned int, to properly set bit
31, otherwise we get "UBSAN: shift-out-of-bounds in fbcon_set_font",
as reported on

http://lore.kernel.org/all/[e-mail]
Kernel Branch: 6.2.0-rc5-next-20230124
Kernel config: https://drive.google.com/file/d/1F-LszDAizEEH0ZX0HcSR06v5q8FPl2Uv/view?usp=sharing
Reproducer: https://drive.google.com/file/d/1mP1jcLBY7vWCNM60OMf-ogw-urQRjNrm/view?usp=sharing

Reported-by: Sanan Hasanov <[e-mail]>
Signed-off-by: Samuel Thibault <[e-mail]>

Comments

Samuel Thibault Jan. 26, 2023, 12:51 a.m. UTC | #1

(FYI, the UB was already a problem before my big-font patch submission,
the checks I add here are already making sense in previous versions
anyway)

Samuel Thibault, le jeu. 26 janv. 2023 01:49:12 +0100, a ecrit:
> blit_x and blit_y are uint32_t, so fbcon currently cannot support fonts
> larger than 32x32.
> 
> The 32x32 case also needs shifting an unsigned int, to properly set bit
> 31, otherwise we get "UBSAN: shift-out-of-bounds in fbcon_set_font",
> as reported on
> 
> http://lore.kernel.org/all/[e-mail]
> Kernel Branch: 6.2.0-rc5-next-20230124
> Kernel config: https://drive.google.com/file/d/1F-LszDAizEEH0ZX0HcSR06v5q8FPl2Uv/view?usp=sharing
> Reproducer: https://drive.google.com/file/d/1mP1jcLBY7vWCNM60OMf-ogw-urQRjNrm/view?usp=sharing
> 
> Reported-by: Sanan Hasanov <[e-mail]>
> Signed-off-by: Samuel Thibault <[e-mail]>
> 
> Index: linux-6.0/drivers/video/fbdev/core/fbcon.c
> ===================================================================
> --- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
> +++ linux-6.0/drivers/video/fbdev/core/fbcon.c
> @@ -2489,9 +2489,12 @@ static int fbcon_set_font(struct vc_data
>  	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
>  		return -EINVAL;
>  
> +	if (font->width > 32 || font->height > 32)
> +		return -EINVAL;
> +
>  	/* Make sure drawing engine can handle the font */
> -	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
> -	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
> +	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
> +	    !(info->pixmap.blit_y & (1U << (font->height - 1))))
>  		return -EINVAL;
>  
>  	/* Make sure driver can handle the font length */
> 
>

Greg KH Jan. 26, 2023, 7:43 a.m. UTC | #2

On Thu, Jan 26, 2023 at 01:49:12AM +0100, Samuel Thibault wrote:
> blit_x and blit_y are uint32_t, so fbcon currently cannot support fonts
> larger than 32x32.

"u32" you mean, right?

> The 32x32 case also needs shifting an unsigned int, to properly set bit
> 31, otherwise we get "UBSAN: shift-out-of-bounds in fbcon_set_font",
> as reported on
> 
> http://lore.kernel.org/all/[e-mail]

Odd blank line?


> Kernel Branch: 6.2.0-rc5-next-20230124
> Kernel config: https://drive.google.com/file/d/1F-LszDAizEEH0ZX0HcSR06v5q8FPl2Uv/view?usp=sharing
> Reproducer: https://drive.google.com/file/d/1mP1jcLBY7vWCNM60OMf-ogw-urQRjNrm/view?usp=sharing

What are all of these lines for?

> 
> Reported-by: Sanan Hasanov <[e-mail]>
> Signed-off-by: Samuel Thibault <[e-mail]>

What commit id does this fix?  Should it go to stable kernels?

> 
> Index: linux-6.0/drivers/video/fbdev/core/fbcon.c
> ===================================================================
> --- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
> +++ linux-6.0/drivers/video/fbdev/core/fbcon.c
> @@ -2489,9 +2489,12 @@ static int fbcon_set_font(struct vc_data
>  	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
>  		return -EINVAL;
>  
> +	if (font->width > 32 || font->height > 32)
> +		return -EINVAL;
> +
>  	/* Make sure drawing engine can handle the font */
> -	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
> -	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
> +	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
> +	    !(info->pixmap.blit_y & (1U << (font->height - 1))))

Are you sure this is still needed with the above check added?  If so,
why?  What is the difference in the compiled code?

thanks,

greg k-h

Jiri Slaby Jan. 26, 2023, 9:02 a.m. UTC | #3

On 26. 01. 23, 1:49, Samuel Thibault wrote:
> blit_x and blit_y are uint32_t, so fbcon currently cannot support fonts
> larger than 32x32.
> 
> The 32x32 case also needs shifting an unsigned int, to properly set bit
> 31, otherwise we get "UBSAN: shift-out-of-bounds in fbcon_set_font",
> as reported on
> 
> http://lore.kernel.org/all/[e-mail]
> Kernel Branch: 6.2.0-rc5-next-20230124
> Kernel config: https://drive.google.com/file/d/1F-LszDAizEEH0ZX0HcSR06v5q8FPl2Uv/view?usp=sharing
> Reproducer: https://drive.google.com/file/d/1mP1jcLBY7vWCNM60OMf-ogw-urQRjNrm/view?usp=sharing
> 
> Reported-by: Sanan Hasanov <[e-mail]>
> Signed-off-by: Samuel Thibault <[e-mail]>
> 
> Index: linux-6.0/drivers/video/fbdev/core/fbcon.c
> ===================================================================
> --- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
> +++ linux-6.0/drivers/video/fbdev/core/fbcon.c
> @@ -2489,9 +2489,12 @@ static int fbcon_set_font(struct vc_data
>   	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
>   		return -EINVAL;
>   
> +	if (font->width > 32 || font->height > 32)
> +		return -EINVAL;
> +
>   	/* Make sure drawing engine can handle the font */
> -	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
> -	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
> +	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
> +	    !(info->pixmap.blit_y & (1U << (font->height - 1))))

So use BIT() properly then? That should be used in all these shifts 
anyway. Exactly to avoid UB.

thanks,

Jiri Slaby Jan. 26, 2023, 9:08 a.m. UTC | #4

On 26. 01. 23, 8:43, Greg KH wrote:
>> --- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
>> +++ linux-6.0/drivers/video/fbdev/core/fbcon.c
>> @@ -2489,9 +2489,12 @@ static int fbcon_set_font(struct vc_data
>>   	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
>>   		return -EINVAL;
>>   
>> +	if (font->width > 32 || font->height > 32)
>> +		return -EINVAL;
>> +
>>   	/* Make sure drawing engine can handle the font */
>> -	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
>> -	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
>> +	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
>> +	    !(info->pixmap.blit_y & (1U << (font->height - 1))))
> 
> Are you sure this is still needed with the above check added?  If so,
> why?  What is the difference in the compiled code?

For font->{width,height} == 32, definitely. IMO, 1 << 31 is undefined as 
1 << 31 cannot be represented by an (signed) int.

Samuel Thibault Jan. 29, 2023, 3:04 p.m. UTC | #5

Jiri Slaby, le jeu. 26 janv. 2023 10:02:55 +0100, a ecrit:
> On 26. 01. 23, 1:49, Samuel Thibault wrote:
> > Index: linux-6.0/drivers/video/fbdev/core/fbcon.c
> > ===================================================================
> > --- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
> > +++ linux-6.0/drivers/video/fbdev/core/fbcon.c
> > @@ -2489,9 +2489,12 @@ static int fbcon_set_font(struct vc_data
> >   	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
> >   		return -EINVAL;
> > +	if (font->width > 32 || font->height > 32)
> > +		return -EINVAL;
> > +
> >   	/* Make sure drawing engine can handle the font */
> > -	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
> > -	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
> > +	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
> > +	    !(info->pixmap.blit_y & (1U << (font->height - 1))))
> 
> So use BIT() properly then? That should be used in all these shifts anyway.
> Exactly to avoid UB.

Right, I'll use that in next version.

Samuel

Samuel Thibault Jan. 29, 2023, 3:13 p.m. UTC | #6

Greg KH, le jeu. 26 janv. 2023 08:43:02 +0100, a ecrit:
> On Thu, Jan 26, 2023 at 01:49:12AM +0100, Samuel Thibault wrote:
> > blit_x and blit_y are uint32_t, so fbcon currently cannot support fonts
> > larger than 32x32.
> 
> "u32" you mean, right?

Right :)

> > The 32x32 case also needs shifting an unsigned int, to properly set bit
> > 31, otherwise we get "UBSAN: shift-out-of-bounds in fbcon_set_font",
> > as reported on
> > 
> > http://lore.kernel.org/all/[e-mail]
> 
> Odd blank line?

Not sure what you mean? I found it easier to read to put a blank line
between the text and the references.

> > Kernel Branch: 6.2.0-rc5-next-20230124
> > Kernel config: https://drive.google.com/file/d/1F-LszDAizEEH0ZX0HcSR06v5q8FPl2Uv/view?usp=sharing
> > Reproducer: https://drive.google.com/file/d/1mP1jcLBY7vWCNM60OMf-ogw-urQRjNrm/view?usp=sharing
> 
> What are all of these lines for?

They provide the references that were set in the original report

http://lore.kernel.org/all/[e-mail]

Arguably the "branch" and "config" are not that useful since the bug has
been there for more than a dozen years, but notably the "Reproducer" is
useful to provide a userland program that triggers the UB.

> > Reported-by: Sanan Hasanov <[e-mail]>
> > Signed-off-by: Samuel Thibault <[e-mail]>
> 
> What commit id does this fix?

I though it was there forever, but it seems the check was added in 2007,
so I'll add it in next version.

> Should it go to stable kernels?

Yes. I added stable in Cc of the mail but missed adding it in the text,
I will add it.

> > Index: linux-6.0/drivers/video/fbdev/core/fbcon.c
> > ===================================================================
> > --- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
> > +++ linux-6.0/drivers/video/fbdev/core/fbcon.c
> > @@ -2489,9 +2489,12 @@ static int fbcon_set_font(struct vc_data
> >  	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
> >  		return -EINVAL;
> >  
> > +	if (font->width > 32 || font->height > 32)
> > +		return -EINVAL;
> > +
> >  	/* Make sure drawing engine can handle the font */
> > -	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
> > -	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
> > +	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
> > +	    !(info->pixmap.blit_y & (1U << (font->height - 1))))
> 
> Are you sure this is still needed with the above check added?  If so,
> why?  What is the difference in the compiled code?

As mentioned by Jiri, yes in the 32 case it's needed otherwise it's UB.

Samuel

Patch

Index: linux-6.0/drivers/video/fbdev/core/fbcon.c
===================================================================
--- linux-6.0.orig/drivers/video/fbdev/core/fbcon.c
+++ linux-6.0/drivers/video/fbdev/core/fbcon.c
@@ -2489,9 +2489,12 @@  static int fbcon_set_font(struct vc_data
 	    h > FBCON_SWAP(info->var.rotate, info->var.yres, info->var.xres))
 		return -EINVAL;
 
+	if (font->width > 32 || font->height > 32)
+		return -EINVAL;
+
 	/* Make sure drawing engine can handle the font */
-	if (!(info->pixmap.blit_x & (1 << (font->width - 1))) ||
-	    !(info->pixmap.blit_y & (1 << (font->height - 1))))
+	if (!(info->pixmap.blit_x & (1U << (font->width - 1))) ||
+	    !(info->pixmap.blit_y & (1U << (font->height - 1))))
 		return -EINVAL;
 
 	/* Make sure driver can handle the font length */