Re: [PATCH v2] x86: don't assume all fb devices are PCI devices

[Bjorn Helgaas]

On Mon, Mar 14, 2016 at 03:42:57PM +0100, Vitaly Kuznetsov wrote:
> When booting Hyper-V Generation 2 guests KASAN reports the following
> out-of-bounds access:
> 
> BUG: KASAN: slab-out-of-bounds in fb_is_primary_device+0x58/0x70 at addr
>   ffff880079cf0eb0
> Read of size 8 by task swapper/0/1
> ...
>  [<ffffffff81581308>] dump_stack+0x63/0x8b
>  [<ffffffff812e1f99>] print_trailer+0xf9/0x150
>  [<ffffffff812e7344>] object_err+0x34/0x40
>  [<ffffffff812e9630>] kasan_report_error+0x230/0x550
>  [<ffffffff812e9ee8>] kasan_report+0x58/0x60
>  [<ffffffff812e4500>] ? ___slab_alloc+0x80/0x490
>  [<ffffffff81878a28>] ? fb_is_primary_device+0x58/0x70
>  [<ffffffff812e87cd>] __asan_load8+0x5d/0x70
>  [<ffffffff81878a28>] fb_is_primary_device+0x58/0x70
>  [<ffffffff8162357a>] register_framebuffer+0xda/0x5b0
>  [<ffffffff816234a0>] ? remove_conflicting_framebuffers+0x50/0x50
> ...
> 
> The issue is caused by the to_pci_dev() call with no check that the given
> info->device is in fact a pci device and some fb devices (Hyper-V FB, EFI
> FB,...) are not.
> 
> Signed-off-by: Vitaly Kuznetsov <vkuznets@xxxxxxxxxx>

Acked-by: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>

> ---
> Changes since v1: use dev_is_pci() instead of full scan [Bjorn Helgaas].
> ---
>  arch/x86/video/fbdev.c | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/x86/video/fbdev.c b/arch/x86/video/fbdev.c
> index d5644bb..9f2fe63 100644
> --- a/arch/x86/video/fbdev.c
> +++ b/arch/x86/video/fbdev.c
> @@ -18,12 +18,11 @@ int fb_is_primary_device(struct fb_info *info)
>  	struct pci_dev *default_device = vga_default_device();
>  	struct resource *res = NULL;

The initializations of "pci_dev" and "res" to NULL are now
unnecessary.

>  
> -	if (device)
> -		pci_dev = to_pci_dev(device);
> -
> -	if (!pci_dev)
> +	if (!device || !dev_is_pci(device))
>  		return 0;
>  
> +	pci_dev = to_pci_dev(device);
> +
>  	if (default_device) {
>  		if (pci_dev == default_device)
>  			return 1;

The test for non-NULL "res" below is superfluous because we know
"pci_dev" is non-NULL:

  res = &pci_dev->resource[PCI_ROM_RESOURCE];
  if (res && res->flags & IORESOURCE_ROM_SHADOW)

I think your patch is functionally correct, so it's up to you and the
x86 guys whether you want to do these additional cleanups.

Bjorn