Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
In-Reply-To: <CAJ+HfNh9pRGcd9EO7BEfPPEdCmP5EDdu_rNgLR7r4oDrcLgvQQ@mail.gmail.com>
References: <00000000000092de58056e3d4b96@google.com> <10d6b170-b820-3077-8737-c9d06e98d0fb@I-love.SAKURA.ne.jp>
 <CAJ+HfNhnmvTHsE7cgi8pROjprsYvCZuzZDmYmpiUx4jqtHboUw@mail.gmail.com>
 <CACT4Y+bfZK6yucoDrXPPW3Tc40RYq61u0Zcb=CKtdao8E+v1cQ@mail.gmail.com>
 <13f6777a-2170-d0cc-1066-1b48a27ec981@i-love.sakura.ne.jp> <CAJ+HfNh9pRGcd9EO7BEfPPEdCmP5EDdu_rNgLR7r4oDrcLgvQQ@mail.gmail.com>
From:   Dmitry Vyukov <dvyukov@google.com>
Date:   Mon, 11 Jun 2018 07:49:41 +0200
Message-ID: <CACT4Y+ZYNjPTkDZxCranM9X1C3yB4GSy8qQd=U0NoeHiJtmV-w@mail.gmail.com>
Subject: Re: WARNING: kmalloc bug in xdp_umem_create
To:     =?UTF-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn.topel@gmail.com>
Cc:     Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
        syzbot+4abadc5d69117b346506@syzkaller.appspotmail.com,
        =?UTF-8?B?QmrDtnJuIFTDtnBlbA==?= <bjorn.topel@intel.com>,
        "Karlsson, Magnus" <magnus.karlsson@intel.com>,
        David Miller <davem@davemloft.net>,
        LKML <linux-kernel@vger.kernel.org>,
        Netdev <netdev@vger.kernel.org>,
        syzkaller-bugs <syzkaller-bugs@googlegroups.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, Jun 10, 2018 at 3:03 PM, Bj=C3=B6rn T=C3=B6pel <bjorn.topel@gmail.c=
om> wrote:
>> On 2018/06/10 20:52, Dmitry Vyukov wrote:
>> > On Sun, Jun 10, 2018 at 11:31 AM, Bj=C3=B6rn T=C3=B6pel <bjorn.topel@g=
mail.com> wrote:
>> >> Den s=C3=B6n 10 juni 2018 kl 04:53 skrev Tetsuo Handa
>> >> <penguin-kernel@i-love.sakura.ne.jp>:
>> >>>
>> >>> On 2018/06/10 7:47, syzbot wrote:
>> >>>> Hello,
>> >>>>
>> >>>> syzbot found the following crash on:
>> >>>>
>> >>>> HEAD commit:    7d3bf613e99a Merge tag 'libnvdimm-for-4.18' of git:=
//git.k..
>> >>>> git tree:       upstream
>> >>>> console output: https://syzkaller.appspot.com/x/log.txt?x=3D1073f68=
f800000
>> >>>> kernel config:  https://syzkaller.appspot.com/x/.config?x=3Df04d8d0=
a2afb789a
>> >>>> dashboard link: https://syzkaller.appspot.com/bug?extid=3D4abadc5d6=
9117b346506
>> >>>> compiler:       gcc (GCC) 8.0.1 20180413 (experimental)
>> >>>> syzkaller repro:https://syzkaller.appspot.com/x/repro.syz?x=3D13c97=
56f800000
>> >>>> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=3D16366f9=
f800000
>> >>>>
>> >>>> IMPORTANT: if you fix the bug, please add the following tag to the =
commit:
>> >>>> Reported-by: syzbot+4abadc5d69117b346506@syzkaller.appspotmail.com
>> >>>>
>> >>>> random: sshd: uninitialized urandom read (32 bytes read)
>> >>>> random: sshd: uninitialized urandom read (32 bytes read)
>> >>>> random: sshd: uninitialized urandom read (32 bytes read)
>> >>>> random: sshd: uninitialized urandom read (32 bytes read)
>> >>>> random: sshd: uninitialized urandom read (32 bytes read)
>> >>>> WARNING: CPU: 1 PID: 4537 at mm/slab_common.c:996 kmalloc_slab+0x56=
/0x70 mm/slab_common.c:996
>> >>>> Kernel panic - not syncing: panic_on_warn set ...
>> >>>
>> >>> syzbot gave up upon kmalloc(), but actually error handling path has
>> >>> NULL pointer dereference bug.
>> >>>
>> >>
>> >> Thanks Tetsuo! This crash has been fixed by Daniel Borkmann in commit
>> >> c09290c56376 ("bpf, xdp: fix crash in xdp_umem_unaccount_pages").
>> >
>> > Let's tell syzbot about this:
>> >
>> > #syz fix: bpf, xdp: fix crash in xdp_umem_unaccount_pages
>> >
>> >
>> Excuse me, but that patch fixes NULL pointer dereference which occurs af=
ter kmalloc()'s
>> "WARNING: CPU: 1 PID: 4537 at mm/slab_common.c:996 kmalloc_slab+0x56/0x7=
0 mm/slab_common.c:996"
>> message. That is, "Too large memory allocation" itself is not yet fixed.
>
> The code relies on that the sl{u,a,o}b layer says no, and the
> setsockopt bails out. The warning could be opted out using
> __GFP_NOWARN. Is there another preferred way? Two get_user_pages
> calls, where the first call would set pages to NULL just to fault the
> region? Walk the process' VMAs? Something else?

Hi Bj=C3=B6rn,

Yes, either __GFP_NOWARN for allocations with user-controllable size
or stricter custom limit (if we don't want current sla/u/ob
implementation details to be part of public kernel interface).