Re: [PATCH v2] fbdev: Add bounds checking in bit_putcs to fix vmalloc-out-of-bounds

[Albin Babu Varghese <albinbabuvarghese20@gmail.com>]

On Sat, Oct 04, 2025 at 02:43:33AM +0200, Helge Deller wrote:
> On 10/3/25 09:32, Albin Babu Varghese wrote:
> > Add bounds checking to prevent writes past framebuffer boundaries when
> > rendering text near screen edges. Return early if the Y position is off-screen
> > and clip image height to screen boundary. Break from the rendering loop if the
> > X position is off-screen. When clipping image width to fit the screen, update
> > the character count to match the clipped width to prevent buffer size
> > mismatches.
> > 
> > Without the character count update, bit_putcs_aligned and bit_putcs_unaligned
> > receive mismatched parameters where the buffer is allocated for the clipped
> > width but cnt reflects the original larger count, causing out-of-bounds writes.
> > 
> > Reported-by: syzbot+48b0652a95834717f190@syzkaller.appspotmail.com
> > Closes: https://syzkaller.appspot.com/bug?extid=48b0652a95834717f190
> > Suggested-by: Helge Deller <deller@gmx.de>
> > Tested-by: syzbot+48b0652a95834717f190@syzkaller.appspotmail.com
> > Signed-off-by: Albin Babu Varghese <albinbabuvarghese20@gmail.com>
> > ---
> > Changes in v2:
> > - Partially render when height exceeding screen boundaries instead of skipping
> > - Update character count when width is clipped to prevent buffer mismatch
> > 
> > Link to v1:
> > https://lore.kernel.org/all/20250927075010.119671-1-albinbabuvarghese20@gmail.com/
> > ---
> >   drivers/video/fbdev/core/bitblit.c | 20 ++++++++++++++++++++
> >   1 file changed, 20 insertions(+)
> 
> applied.
> 
> Thanks!
> Helge
Thank you for merging the patch.

After the patch appeared in mainline, I observed that syzbot continues
to find the same bug through different execution paths. My fix addressed
bit_putcs, but the crashes now occur through bit_cursor and cw_putcs,
which bypass bit_putcs entirely and go directly to sys_imageblit():

Crash 1 (cursor path):
https://syzkaller.appspot.com/text?tag=CrashReport&x=11fe95e2580000
Call trace: hide_cursor → bit_cursor → soft_cursor → sys_imageblit

Crash 2 (rotation path):
https://syzkaller.appspot.com/text?tag=CrashReport&x=164f0b04580000
Call trace: fbcon_modechanged → cw_putcs → sys_imageblit

The original syzbot reproducer depended on character height going out of
bounds, so I focused on bit_putcs where character images are drawn. I
naively overlooked cursor drawing - apologies for that. That's why I
looked at the other crash reports after the merge, because it seemed odd
that it was still hitting the bug after the fix.

I believe adding the same clipping logic in sys_imageblit() would provide
comprehensive protection. Something like this:

void sys_imageblit(struct fb_info *p, const struct fb_image *image)
  {
 +       struct fb_image clipped;
 +       u32 width, height;
 +
         if (!(p->flags & FBINFO_VIRTFB))
                 fb_warn_once(p, "%s: framebuffer is not in virtual address space.\n", __func__);
 
 -       fb_imageblit(p, image);
 +       if (!image || image->dx >= p->var.xres || image->dy >= p->var.yres)
 +               return;
 +
 +       if (image->dx + image->width > p->var.xres || image->dy + image->height > p->var.yres) {
 +               clipped = *image;
 +
 +               height = clipped.height;
 +               if (clipped.dy + height > p->var.yres)
 +                       height = p->var.yres - clipped.dy;
 +
 +               clipped.height = height;
 +
 +               width = clipped.width;
 +               if (clipped.dx + width > p->var.xres)
 +                       width = p->var.xres - clipped.dx;
 +
 +               clipped.width = width;
 +
 +               if (clipped.width == 0 || clipped.height == 0)
 +                       return;
 +
 +               fb_imageblit(p, &clipped);
 +       } else {
 +               fb_imageblit(p, image);
 +       }
  }
 
I don't have any way to test this other than testing it with the existing
reproducers. I can prepare a patch if this approach sounds reasonable.

Thanks,
Albin