On Wed, Apr 1, 2020 at 11:06 AM Daniel Vetter [off-list ref] wrote:

quoted

quoted

quoted

On Tue, Mar 31, 2020 at 2:18 PM Bartlomiej Zolnierkiewicz
[off-list ref] wrote:

quoted

On 3/31/20 12:18 PM, Dmitry Vyukov wrote:

quoted

On Tue, Mar 31, 2020 at 11:57 AM Peter Zijlstra [off-list ref] wrote:

quoted

On Mon, Mar 30, 2020 at 10:01:12PM -0700, syzbot wrote:

quoted

Hello,

syzbot found the following crash on:

HEAD commit:    9420e8ad Merge tag 'for-linus' of git://git.kernel.org/pub..
git tree:       upstream
console output: https://protect2.fireeye.com/url?k56a78d-5a9a6c49-07572cc2-0cc47a314e9a-e4dc8b657d340686&u=https://syzkaller.appspot.com/x/log.txt?x06ed4be00000
kernel config:  https://protect2.fireeye.com/url?kC211072-1eeddbb6-43209b3d-0cc47a314e9a-3bd45a19932c37c8&u=https://syzkaller.appspot.com/x/.config?x'392dd2975fd692
dashboard link: https://protect2.fireeye.com/url?k¿7a6153-e2b6aa97-bf7bea1c-0cc47a314e9a-c64073ee605efb7b&u=https://syzkaller.appspot.com/bug?extidè4d7ebd1361da13c356
compiler:       gcc (GCC) 9.0.0 20181231 (experimental)

Unfortunately, I don't have any reproducer for this crash yet.

IMPORTANT: if you fix the bug, please add the following tag to the commit:
Reported-by: syzbot+e84d7ebd1361da13c356@syzkaller.appspotmail.com

INFO: trying to register non-static key.
the code is fine but needs lockdep annotation.
turning off the locking correctness validator.
CPU: 1 PID: 1014 Comm: syz-executor.0 Not tainted 5.6.0-rc7-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 <IRQ>
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x188/0x20d lib/dump_stack.c:118
 assign_lock_key kernel/locking/lockdep.c:880 [inline]
 register_lock_class+0x14c4/0x1540 kernel/locking/lockdep.c:1189
 __lock_acquire+0xfc/0x3ca0 kernel/locking/lockdep.c:3836
 lock_acquire+0x197/0x420 kernel/locking/lockdep.c:4484
 __raw_spin_lock_irqsave include/linux/spinlock_api_smp.h:110 [inline]
 _raw_spin_lock_irqsave+0x8c/0xbf kernel/locking/spinlock.c:159
 try_to_wake_up+0x9f/0x17c0 kernel/sched/core.c:2547

That's p->pi_lock, which gets initialized in rt_mutex_init_task() in
copy_process(). This should be impossible. Very odd.

The stack mentions fbdev, which is a red flag at the moment. There are
a dozen of bad bugs in fbdev and around. Just few days ago Andy
pointed to another "impossible" crash "general protection fault in
do_syscall_64" which is related to dri:
https://protect2.fireeye.com/url?k

b8ad06-517466c2-0cb92649-0cc47a314e9a-a20c11191483c65b&u=https://syzkaller.appspot.com/bug?idc7b2602b1ff40f0d34f38baa4ba1640727c3d9

quoted

quoted

quoted

quoted

quoted

https://protect2.fireeye.com/url?ka4292e3-3c8e5927-614319ac-0cc47a314e9a-aeda6d72c01a7b0e&u=https://groups.google.com/forum/#!msg/syzkaller-bugs/ePqhfYx0-8M/Q_Urt97iAAAJ

There are probably more random manifestations of these bugs already,
and I guess we will be getting more.

+fbdev maintainers

Thank you for the report.

fbdev is in the maintenance mode and no new features or drivers are
being added so syzbot reports are not for a new bugs (regressions) and
are not a priority (at least to me).

Yup same here, I've seen a pile of syzbot reports for fbdev (and also
vt, or combinations of them since fbdev is linked to vt through fbcon)
fly by. But I really don't have to deal with these, my recommendation
to anyone who cares about security are:
- Don't enable vt
- Don't enable fbdev

1. How do we deliver this message to relevant people?

Because:

$ grep FBDEV syzkaller/dashboard/config/upstream-kasan.config
CONFIG_DRM_FBDEV_EMULATION=y
CONFIG_DRM_FBDEV_OVERALLOC0
# CONFIG_DRM_FBDEV_LEAK_PHYS_SMEM is not set
CONFIG_XEN_FBDEV_FRONTEND=y

and my current work machine:

$ grep FBDEV /boot/config-5.2.17-1-amd64
CONFIG_DRM_FBDEV_EMULATION=y
CONFIG_DRM_FBDEV_OVERALLOC0
# CONFIG_DRM_FBDEV_LEAK_PHYS_SMEM is not set
CONFIG_XEN_FBDEV_FRONTEND=y

Yeah I know it's been like this since forever. In theory you could
build a fbdev/fbcon less distro since years (the last bit for a proof
of concept was kmscon/systemd-consoled), but the amount of investment
into classic linux desktop is so little that it's impossible to get
this funded. CrOS fixed this a while ago iirc though.

I think to fix the syzbot issues all we'd need is a competent intern
for a few months, that should take care of the worst stuff. Obviously
wont include getting a test suite going, nor fixing any of the
fundamental issues. But duct-taping over all the bugs should be
possible (it's what we've been doing for well over 10 years by now in
fbdev/fbocn/vt code anyway). I'd be willing to help mentoring, but
that's about all I can do.

Adding Matthew Garret, I have discussed with him in the past finding
some funding for linux desktop stuff like this.

I will keep this in mind. We _may_ get some interns this year who
_may_ be interested in fixing Linux kernel bugs (but otherwise
extending syzkaller descriptions).

FTR, there is also some follow up on twitter re extending
https://github.com/a13xp0p0v/kconfig-hardened-check to capture such
recommendations:
https://twitter.com/dvyukov/status/1245969522869309441
https://github.com/a13xp0p0v/kconfig-hardened-check/issues/38

quoted

quoted

2. What do we do with fbdev testing on syzbot? Is there a way to
disable all of the unsupported stuff? But if we disable it, we don't
find any regressions as well. And in the end that's what is in the
mainline kernel and is still enabled in distros (at least in the 2
real configs I can grep now).

This would be bad I agree, but it's not any worse than the state of
things the past 10 years. That's roughly for as long as fbdev has been
in maintainance only mode, meaning "we'll apply patches if they come".
Without Bart volunteering, we wouldn't even have that much really.

Oh wrt disabling fbdev: Make sure CONFIG_FB isn't set. Unfortunately a
pile of other things select that, for convenience (like
CONFIG_DRM_KMS_FB_HELPER).

That should get rid of all the problematic fbdev code.

From what I've seen in some of the syzbot mails flying by we also have
issues in vt and console code blowing up (not just on fbcon/fbdev, but
also e.g. on vgacon). That stuff you'll still hit. But maybe you can
trick Greg KH into fixing the vt/console.c issues, he just claimed
that :-P
-Daniel

quoted

quoted

quoted

All that code has been developed long ago, in a much more innocent
time. If someone wants to fix this you'd not just need to fix all the
syzbot stuff, but also ramp up a full testsuite for all the ioctl, and
all the corner-cases. Plus also fix some of the horrendous locking in
there, probably.

Multi-year effort, easily.

Regressions I'll obviously try to handle, but none of these are. It's
just syzbot has become smarter at hitting bugs in fbdev and vt
subsystems (or maybe the hw the virtual machines emulate has become
more varied, some of the reports are for fun stuff like vgacon ...).

Cheers, Daniel

quoted

I have only resources to review/merge pending fbdev patches from time
to time so any help in fixing these syzbot reports is welcomed (there
have been a few fbdev related syzbot reports recently).

Also please note that fbdev is maintained through drm-misc tree so
patches can also be handled by other drm-misc maintainers in case I'm
not available / busy with other things.

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung R&D Institute Poland
Samsung Electronics

quoted

quoted

quoted

 wake_up_worker kernel/workqueue.c:836 [inline]
 insert_work+0x2ad/0x3a0 kernel/workqueue.c:1337
 __queue_work+0x50d/0x1280 kernel/workqueue.c:1488
 call_timer_fn+0x195/0x760 kernel/time/timer.c:1404
 expire_timers kernel/time/timer.c:1444 [inline]
 __run_timers kernel/time/timer.c:1773 [inline]
 __run_timers kernel/time/timer.c:1740 [inline]
 run_timer_softirq+0x412/0x1600 kernel/time/timer.c:1786
 __do_softirq+0x26c/0x99d kernel/softirq.c:292
 invoke_softirq kernel/softirq.c:373 [inline]
 irq_exit+0x192/0x1d0 kernel/softirq.c:413
 exiting_irq arch/x86/include/asm/apic.h:546 [inline]
 smp_apic_timer_interrupt+0x19e/0x600 arch/x86/kernel/apic/apic.c:1146
 apic_timer_interrupt+0xf/0x20 arch/x86/entry/entry_64.S:829
 </IRQ>

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel

--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch

--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch