Re: kernel BUG at drivers/android/binder_alloc.c:LINE!

[Dmitry Vyukov]

On Wed, Jan 31, 2018 at 9:08 AM, Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
> On Tue, Jan 30, 2018 at 11:59:47PM -0800, Eric Biggers wrote:
>> On Fri, Dec 01, 2017 at 04:22:00PM -0800, syzbot wrote:
>> > syzkaller has found reproducer for the following crash on
>> > 3c1c4ddffb58b9e10b3365764fe59546130b3f32
>> > git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/master
>> > compiler: gcc (GCC) 7.1.1 20170620
>> > .config is attached
>> > Raw console output is attached.
>> > C reproducer is attached
>> > syzkaller reproducer is attached. See https://goo.gl/kgGztJ
>> > for information about syzkaller reproducers
>> >
>> >
>> > binder: 3087:3087 ERROR: BC_REGISTER_LOOPER called without request
>> > binder: 3087:3087 ioctl c0306201 2000dfd0 returned -14
>> > ------------[ cut here ]------------
>> > kernel BUG at drivers/android/binder_alloc.c:750!
>> > invalid opcode: 0000 [#1] SMP KASAN
>> > Dumping ftrace buffer:
>> >    (ftrace buffer empty)
>> > Modules linked in:
>> > CPU: 0 PID: 1404 Comm: kworker/0:2 Not tainted 4.15.0-rc1+ #203
>> > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
>> > Google 01/01/2011
>> > Workqueue: events binder_deferred_func
>> > task: 0000000009b739b6 task.stack: 00000000c1d4442c
>> > RIP: 0010:binder_alloc_deferred_release+0x146/0xa40
>> > drivers/android/binder_alloc.c:750
>> > RSP: 0018:ffff8801d2b16fd8 EFLAGS: 00010293
>> > RAX: ffff8801d2b08080 RBX: ffff8801d7829300 RCX: ffffffff8403b856
>> > RDX: 0000000000000000 RSI: 0000000000000004 RDI: ffff8801d7829330
>> > RBP: ffff8801d2b17138 R08: ffffffff8403b7d9 R09: 1ffffffff0e53001
>> > R10: ffff8801d2b16fc8 R11: ffffffff87489d60 R12: 0000000000000000
>> > R13: dffffc0000000000 R14: ffff8801d2b17110 R15: ffff8801d7829310
>> > FS:  0000000000000000(0000) GS:ffff8801db400000(0000) knlGS:0000000000000000
>> > CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> > CR2: 00000000006d00a8 CR3: 0000000005e25000 CR4: 00000000001406f0
>> > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
>> > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
>> > Call Trace:
>> >  binder_free_proc drivers/android/binder.c:4200 [inline]
>> >  binder_proc_dec_tmpref+0x2f3/0x420 drivers/android/binder.c:1833
>> >  binder_deferred_release drivers/android/binder.c:4858 [inline]
>> >  binder_deferred_func+0xe22/0x12f0 drivers/android/binder.c:4893
>> >  process_one_work+0xbfd/0x1be0 kernel/workqueue.c:2112
>> >  worker_thread+0x223/0x1990 kernel/workqueue.c:2246
>> >  kthread+0x37a/0x440 kernel/kthread.c:238
>> >  ret_from_fork+0x24/0x30 arch/x86/entry/entry_64.S:441
>> > Code: e8 00 40 6c fd 49 8d 7f 20 49 8d 5f f0 48 89 fa 48 c1 ea 03 42 80 3c
>> > 2a 00 0f 85 84 07 00 00 49 83 7f 20 00 74 a9 e8 da 3f 6c fd <0f> 0b 48 8b 9d
>> > e8 fe ff ff 44 89 a5 bc fe ff ff e8 c5 3f 6c fd
>> > RIP: binder_alloc_deferred_release+0x146/0xa40
>> > drivers/android/binder_alloc.c:750 RSP: ffff8801d2b16fd8
>> > ---[ end trace 616e085d0dbf3c21 ]---
>> > Kernel panic - not syncing: Fatal exception
>> > Dumping ftrace buffer:
>> >    (ftrace buffer empty)
>> > Kernel Offset: disabled
>> > Rebooting in 86400 seconds..
>>
>> This crash stopped occurring in v4.15-rc3, and I verified that it was commit
>> fb2c445277e which fixed it (thanks Martijn!).  So, telling syzbot so that it can
>> start reporting similar crashes again:
>>
>> #syz fix: ANDROID: binder: fix transaction leak.
>
> Could you change the syz fix format to include the hash in case we want
> to look it up?

Hi Dan,

There is an unpleasant issue related to this. Some email clients
(notably gmail) reflow text emails to 72/80 columns and this
transformation is irreversible. So if we add hash, this increases
chances of commit title (what is actually used by syzbot) being
corrupted.
The Reported-by tags allow precise matching of bugs to reports and
back (and that's what I think now is the preferred way to doing the
association as it does not require somebody to go back to the report
email and attach the tag).
Or I can suggest to provide commit hash on a separate line.
The reflow thing is nasty, but hit us periodically. The problem that
should not be there at all, but I don't see a good solution for it.
Obviously requiring developers to follow some formal escaping rules
for titles is not going to work either.