Received-SPF: pass (google.com: domain of linux-wireless-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <000000000000cb517b05a32c917b@google.com> <ed2b00dfda5b6ce46a2c2a33093ee56f77af6a8f.camel@sipsolutions.net>
In-Reply-To: <ed2b00dfda5b6ce46a2c2a33093ee56f77af6a8f.camel@sipsolutions.net>
From:   Dmitry Vyukov <dvyukov@google.com>
Date:   Wed, 15 Apr 2020 17:16:28 +0200
Message-ID: <CACT4Y+YtT4_An1wtzNWe3_=kMAF3Yhj+pr=GM5ZYOJ9TN3ryXA@mail.gmail.com>
Subject: Re: WARNING in hwsim_new_radio_nl
To:     Johannes Berg <johannes@sipsolutions.net>,
        syzkaller <syzkaller@googlegroups.com>
Cc:     syzbot <syzbot+a4aee3f42d7584d76761@syzkaller.appspotmail.com>,
        David Miller <davem@davemloft.net>,
        Kalle Valo <kvalo@codeaurora.org>,
        LKML <linux-kernel@vger.kernel.org>,
        linux-wireless@vger.kernel.org, mathew.j.martineau@linux.intel.com,
        matthieu.baerts@tessares.net, netdev <netdev@vger.kernel.org>,
        Paolo Abeni <pabeni@redhat.com>,
        syzkaller-bugs <syzkaller-bugs@googlegroups.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-wireless-owner@vger.kernel.org
Precedence: bulk

On Tue, Apr 14, 2020 at 12:41 PM Johannes Berg
<johannes@sipsolutions.net> wrote:
>
> Hi syzbot keepers,
>
> On Mon, 2020-04-13 at 07:05 -0700, syzbot wrote:
> > syzbot has bisected this bug to:
> >
> > commit 01cacb00b35cb62b139f07d5f84bcf0eeda8eff6
> > Author: Paolo Abeni <pabeni@redhat.com>
> > Date:   Fri Mar 27 21:48:51 2020 +0000
> >
> >     mptcp: add netlink-based PM
> >
> > bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=10225bb3e00000
>
> This is, fairly obviously, incorrect. Same with the bisection for
> 6693adf1698864d21734, which is really the same underlying problem as
> this one (though at a different code site).
>
> However, it stands out that this was bisected to a commit that adds a
> new generic netlink family in both cases.
>
> This makes sense - the reproducer identifies the family by *number*, but
> that number isn't stable, generic netlink families should be identified
> by *name*.
>
> Perhaps somehow syzbot could be taught that, so that the bisection is
> stable across kernels with different generic netlink families
> registered?
>
> Alternatively, we _could_ add some kind of stable ID mode, but I'm not
> sure we really want to ... since that would mean people start hardcoding
> IDs?

+syzkaller mailing list

Hi Johannes,

syzkaller has a pseudo-syscall to map string genetlink family ID to
int ID. If that syscall would have been used, then I assume it should
have worked. However in this case, it managed to trigger the bug with
a plain opaque blob with no knowledge about the blob contents
whatsoever. I don't see any realistic way to preserve family ID in
this case.