linux-kernel - Re: [PATCH] drm/radeon: fix a missing-check bug

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <90c22763-fdba-0415-c570-ef8e85bf5889@amd.com>
Date:   Fri, 19 Oct 2018 07:02:16 +0000
From:   "Koenig, Christian" <Christian.Koenig@....com>
To:     Wenwen Wang <wang6495@....edu>
CC:     Kangjie Lu <kjlu@....edu>,
        "Deucher, Alexander" <Alexander.Deucher@....com>,
        "Zhou, David(ChunMing)" <David1.Zhou@....com>,
        David Airlie <airlied@...ux.ie>,
        "open list:RADEON and AMDGPU DRM DRIVERS" 
        <amd-gfx@...ts.freedesktop.org>,
        "open list:DRM DRIVERS" <dri-devel@...ts.freedesktop.org>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] drm/radeon: fix a missing-check bug

Am 18.10.18 um 19:13 schrieb Wenwen Wang:
> In radeon_read_bios(), the bios rom is firstly mapped to the IO memory
> region 'bios' through pci_map_rom(). Then the first two bytes of 'bios' are
> copied to 'val1' and 'val2' respectively through readb(). After that,
> 'val1' and 'val2' are checked to see whether they have expected values,
> i.e., 0x55 and 0xaa, respectively. If yes, the whole data in 'bios' is then
> copied to 'rdev->bios' through memcpy_fromio(). Obviously, the first two
> bytes in 'bios' are copied twice. More importantly, no check is enforced on
> the first two bytes of 'rdev->bios' after memcpy_fromio(). Given that the
> IO memory region can also be accessed by the device, it is possible that a
> malicious device can race to modify these two bytes between the two copies
> and thus after memcpy_fromio(), the first two bytes in 'rdev->bios' can
> have unexpected values.  This can cause undefined behavior of the kernel
> and introduce potential security risk, if the device can be controlled by
> attackers.
>
> This patch rewrites the first two bytes of 'rdev->bios' after
> memcpy_fromio() with expected values. Through this way, the above issue can
> be avoided.

Well NAK, that doesn't make any sense to me.

First of all we don't map VRAM, but rather the ROM which is a read only 
flash. Writing to the flash is not supported through the BAR as far as I 
know.

Then we check the first two bytes to make sure that the ROM is correctly 
mapped and not to prevent any malicious attacks.

Regards,
Christian.

>
> Signed-off-by: Wenwen Wang <wang6495@....edu>
> ---
>   drivers/gpu/drm/radeon/radeon_bios.c | 2 ++
>   1 file changed, 2 insertions(+)
>
> diff --git a/drivers/gpu/drm/radeon/radeon_bios.c b/drivers/gpu/drm/radeon/radeon_bios.c
> index 04c0ed4..f336719 100644
> --- a/drivers/gpu/drm/radeon/radeon_bios.c
> +++ b/drivers/gpu/drm/radeon/radeon_bios.c
> @@ -98,6 +98,8 @@ static bool radeon_read_bios(struct radeon_device *rdev)
>   		return false;
>   	}
>   	memcpy_fromio(rdev->bios, bios, size);
> +	rdev->bios[0] = val1;
> +	rdev->bios[1] = val2;
>   	pci_unmap_rom(rdev->pdev, bios);
>   	return true;
>   }