MIME-Version: 1.0
In-Reply-To: <20180104235004.GO23371@thunk.org>
References: <CACT4Y+ZNJRzPxxPQNR88vcPxD54mHY-evhMwR39iMsJh_NwaEA@mail.gmail.com>
 <20180104092327.GA26245@kroah.com> <CACT4Y+aN4TxTE4PYjL7Uos7hatWMonU73As5ddq8iU0cPb29rg@mail.gmail.com>
 <CACT4Y+Z_DV6sgQ7FodoxBj8MdVkcij3bS0P1pp9GOMDS4-f92A@mail.gmail.com> <20180104235004.GO23371@thunk.org>
From: Dmitry Vyukov <dvyukov@google.com>
Date: Mon, 8 Jan 2018 13:11:49 +0100
Message-ID: <CACT4Y+ZF_wGDG8-3J4MPMTeNaL0tTARjtktJPC3bJRDd4aODqQ@mail.gmail.com>
Subject: Re: LKML admins (syzbot emails are not delivered)
To: "Theodore Ts'o" <tytso@mit.edu>,
        Dmitry Vyukov <dvyukov@google.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        syzkaller <syzkaller@googlegroups.com>,
        David Miller <davem@davemloft.net>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org

On Fri, Jan 5, 2018 at 12:50 AM, Theodore Ts'o <tytso@mit.edu> wrote:
> It also seems to me, looking at other complaints on this thread, that
> there is the opportunity for the syzbot to do much more.  For example,
> you can see if it repro's on the last released mainline kernel (such
> as 4.14) and if so, have the syzbot automatically do a bisection
> search, so you can make sure the report goes to the best set of
> developers to fix it, a pointer to the guilty commit.

Hi Ted,

I've filed https://github.com/google/syzkaller/issues/501 for this as
this come up several times. There is bunch of problems, though:

 - unreliable reproducers
 - triggering of unrelated bugs (happens frequently)
 - flakes (bugs in GCE, crashes in tty on first ssh connection, etc)
 - bugs (races) that manifest in multiple different ways
 - bugs that will be attributed to tools improvements (e.g. KASAN,
fault injection improvements)
 - reproducers that need slight changes on different kernel revisions

Not sure what quality of bisection we can achieve. And kernel
developers tend to be negative to any kind of bot gaffe, so we can
lose both ways.