Re: KASAN: slab-out-of-bounds Read in fbcon_get_font

[Andrey Ryabinin]

On 12/4/19 9:33 AM, Dmitry Vyukov wrote:
> On Tue, Dec 3, 2019 at 11:37 PM Daniel Vetter <daniel.vetter@xxxxxxxx> wrote:
>>
>> On Tue, Dec 3, 2019 at 11:25 PM syzbot
>> <syzbot+4455ca3b3291de891abc@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
>>>
>>> Hello,
>>>
>>> syzbot found the following crash on:
>>>
>>> HEAD commit:    76bb8b05 Merge tag 'kbuild-v5.5' of git://git.kernel.org/p..
>>> git tree:       upstream
>>> console output: https://syzkaller.appspot.com/x/log.txt?x=10bfe282e00000
>>> kernel config:  https://syzkaller.appspot.com/x/.config?x=dd226651cb0f364b
>>> dashboard link: https://syzkaller.appspot.com/bug?extid=4455ca3b3291de891abc
>>> compiler:       gcc (GCC) 9.0.0 20181231 (experimental)
>>> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=11181edae00000
>>> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=105cbb7ae00000
>>>
>>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
>>> Reported-by: syzbot+4455ca3b3291de891abc@xxxxxxxxxxxxxxxxxxxxxxxxx
>>>
>>> ==================================================================
>>> BUG: KASAN: slab-out-of-bounds in memcpy include/linux/string.h:380 [inline]
>>> BUG: KASAN: slab-out-of-bounds in fbcon_get_font+0x2b2/0x5e0
>>> drivers/video/fbdev/core/fbcon.c:2465
>>> Read of size 16 at addr ffff888094b0aa10 by task syz-executor414/9999
>>
>> So fbcon allocates some memory, security/tomoyo goes around and frees
>> it, fbcon goes boom because the memory is gone. I'm kinda leaning
>> towards "not an fbcon bug". Adding relevant security folks and mailing
>> lists.
>>
>> But from a very quick look in tomoyo it loosk more like "machine on
>> fire, random corruption all over". No idea what's going on here.
> 
> Hi Daniel,
> 
> This is an out-of-bounds access, not use-after-free.
> I don't know why we print the free stack at all (maybe +Andrey knows),
> but that's what KASAN did from day one. I filed
> https://bugzilla.kernel.org/show_bug.cgi?id=198425 which I think is a
> good idea, I will add your confusion as a data point :)

Because we have that information (free stack) and it usually better to provide
all the information we have rather than hide it. You never known what information
might be needed to fix the bug.
Free memory might be reused and what we report as OOB might be an UAF and free stack
could be useful in such case.