On Tue, 15 Mar 2022 at 18:08, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, Mar 15, 2022 at 2:36 AM syzbot
> <syzbot+3f1ca6a6fec34d601788@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > syzbot suspects this issue was fixed by commit
> > 09688c0166e7 ("Linux 5.17-rc8")
>
> No, I'm afraid that means that the bisection is broken:
>
> > bisection log: https://syzkaller.appspot.com/x/bisect.txt?x=140283ad700000
>
> and yeah, looking at that log it looks like every single run has
>
> testing commit [...]
> run #0: crashed: KASAN: use-after-free Read in ath9k_hif_usb_rx_cb
> ...
> # git bisect good [...]
>
> and you never saw a "bad" commit that didn't have the issue, so the
> top-of-tree gets marked "good" (and I suspect you intentionally mark
> the broken case "good" in order to find where it got fixed, so you're
> using "git bisect" in a reverse way).
>
> I didn't look closer, but it does seem to not reproduce very reliably,
> maybe that is what confused the bot originally.
Hi Linus,
Thanks for taking a look. Yes, it's a "reverse" bisection that tries
to find the fix.
And your conclusion re flakiness looks right, there were few runs with
only 1/20 crashes.
But the bug looks to be fixed by something anyway. git log on the file
pretty clearly points to:
#syz fix: ath9k: Fix out-of-bound memcpy in ath9k_hif_usb_rx_stream
