Re: WARNING: bad unlock balance in xfs_iunlock

From: Dmitry Vyukov
Date: Wed May 09 2018 - 04:43:40 EST


On Wed, May 9, 2018 at 4:48 AM, Eric Sandeen <sandeen@xxxxxxxxxxx> wrote:
>
>
> On 5/8/18 2:52 AM, Dmitry Vyukov wrote:
>>> Or put another way, how did you arrive at the fs image values in the reproducer,
>>> i.e.:
>> Currently they are completely random, nobody taught syzkaller about AGFs, etc.
>
> So you just combine a few megabytes of purely random bits out of thin air until
> you get something that approximates an xfs filesystem? Google must have more
> computing power than I was aware of.


syzbot uses very few cores for fuzzing of all of the hundreds of
kernel subsystems. But syzkaller (the underlying fuzzer) uses
coverage-guidance and this makes fuzzing exponentially more efficient
than blind techniques. Coverage-guidance is also combined with
grammar-based generation techniques, which gives additional synergy
(though there are no grammar descriptions for filesystem formats at
the moment).

Does "xfstests fuzzing infrastructure" use coverage-guidance? If not,
it would be useful to add. Among some solutions there are LibFuzzer
(https://llvm.org/docs/LibFuzzer.html), AFL
(http://lcamtuf.coredump.cx/afl/), kernel-fuzzing
(https://github.com/oracle/kernel-fuzzing). I don't know how xfstests
fuzzing works, so I can't say what would be more suitable there.



>>> oid loop()
>>> {
>>> memcpy((void*)0x20000000, "xfs", 4);
>>> memcpy((void*)0x20000100, "./file0", 8);
>>> *(uint64_t*)0x20000200 = 0x20010000;
>>> memcpy((void*)0x20010000,
>>> "\x58\x46\x53\x42\x00\x00\x10\x00\x00\x00\x00\x00\x00\x00\x10\x00\x00"
>>> "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x9f\x98"
>>> "\x99\xff\xcb\xa1\x4e\xe6\xad\x52\x08\x20\x67\x09\xed\x75\x00\x00\x00"
>>> "\x00\x00\x00\x00\x04\x00\x00\x00\x00\x00\x00\x35\xe0\x00\x00\x00\x00"
>>> "\x00\x00\x35\xe1\x00\x00\x00\x00\x00\x00\x35\xe2\x00\x00\x00\x01\x00"
>>> "\x00\x10\x00\x00\x00\x00\x01\x00\x00\x00\x00\x00\x00\x03\x55\xb4\xa4"
>>> "\x02\x00\x01\x00\x00\x10\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"
>>> "\x00\x0c\x09\x08\x04\x0c\x00\x00\x19\x00\x00\x00\x00\x00\x00\x00\x40"
>>> "\x00\x00\x00\x00\x00\x00\x00\x3d\x00\x00\x00\x00\x00\x00\x0c\xa3\x00"
>>> "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00"
>>> "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x02\x00\x00\x00"
>>> "\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x01\x00\x00\x02\x02",
>>> 204);
>>>
>>> ...
>>>
>>> The in-memory xfs filesystem it constructs is damaged, is that an intentional
>>> part of the fuzzing during the test?
>> Yes, invalid inputs is part of testing.
>