Re: [syzbot] KFENCE: use-after-free in kvm_fastop_exception

[Matthieu Baerts]

Hi Pavel,

On 18/08/2021 10:12, Pavel Skripkin wrote:
> On 8/18/21 11:02 AM, Matthieu Baerts wrote:
>> Hello,
>>
>> On 18/08/2021 00:21, syzbot wrote:
>>> syzbot has bisected this issue to:
>>>
>>> commit c4512c63b1193c73b3f09c598a6d0a7f88da1dd8
>>> Author: Matthieu Baerts <matthieu.baerts@xxxxxxxxxxxx>
>>> Date:   Fri Jun 25 21:25:22 2021 +0000
>>>
>>>     mptcp: fix 'masking a bool' warning
>>>
>>> bisection log: 
>>> https://syzkaller.appspot.com/x/bisect.txt?x=122b0655300000
>>> start commit:   b9011c7e671d Add linux-next specific files for 20210816
>>> git tree:       linux-next
>>> final oops:    
>>> https://syzkaller.appspot.com/x/report.txt?x=112b0655300000
>>> console output: https://syzkaller.appspot.com/x/log.txt?x=162b0655300000
>>> kernel config: 
>>> https://syzkaller.appspot.com/x/.config?x=a245d1aa4f055cc1
>>> dashboard link:
>>> https://syzkaller.appspot.com/bug?extid=7b938780d5deeaaf938f
>>> syz repro:     
>>> https://syzkaller.appspot.com/x/repro.syz?x=157a41ee300000
>>> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=14f78ff9300000
>>
>> I'm pretty sure the commit c4512c63b119 ("mptcp: fix 'masking a bool'
>> warning") doesn't introduce the reported bug. This minor fix is specific
>> to MPTCP which doesn't seem to be used here.
>>
>> I'm not sure how I can tell syzbot this is a false positive.
>>
> 
> 
> looks like it's fs/namei bug. Similar reports:
> 
> https://syzkaller.appspot.com/bug?id=517fa734b92b7db404c409b924cf5c997640e324
> 
> 
> https://syzkaller.appspot.com/bug?id=484483daf3652b40dae18531923aa9175d392a4d

Thank you for having checked!
Should we mark them as "#syz dup" if you think they have the same root
cause?

> It's not false positive. I've suggested the fix here:
> https://groups.google.com/g/syzkaller-bugs/c/HE3c2fP5nic/m/1Yk17GBeAwAJ
> I am waiting for author comments about the fix :)
> 
> But, yes, syzbot bisection is often wrong, so don't rely on it much :)

Yes sorry, I wanted to say the bisection picked a wrong commit :)

All good then if syzbot often blames the wrong modification :)

Cheers,
Matt
-- 
Tessares | Belgium | Hybrid Access Solutions
www.tessares.net