Re: WARNING: kobject bug in sysfs_warn_dup

From: Dmitry Vyukov
Date: Wed Apr 11 2018 - 11:28:28 EST


On Wed, Apr 11, 2018 at 5:04 PM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
> On Thu, Apr 5, 2018 at 4:23 PM, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
>>> >> On 05/04/18 09:52, Dmitry Vyukov wrote:
>>> >> > On Thu, Apr 5, 2018 at 10:36 AM, Steven Whitehouse <swhiteho@xxxxxxxxxx> wrote:
>>> >> > > Hi,
>>> >> > >
>>> >> > >
>>> >> > >
>>> >> > > On 05/04/18 09:19, Dmitry Vyukov wrote:
>>> >> > > > On Thu, Apr 5, 2018 at 8:34 AM, Greg KH <gregkh@xxxxxxxxxxxxxxxxxxx>
>>> >> > > > wrote:
>>> >> > > > > On Wed, Apr 04, 2018 at 07:02:01PM -0700, syzbot wrote:
>>> >> > > > > > Hello,
>>> >> > > > > >
>>> >> > > > > > syzbot hit the following crash on upstream commit
>>> >> > > > > > 3e968c9f1401088abc9a19ae6ff571644d37a355 (Wed Apr 4 21:19:24 2018 +0000)
>>> >> > > > > > Merge tag 'ext4_for_linus' of
>>> >> > > > > > git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
>>> >> > > > > > syzbot dashboard link:
>>> >> > > > > > https://syzkaller.appspot.com/bug?extid=ff87a28e665c163aa7f5
>>> >> > > > > >
>>> >> > > > > > C reproducer:
>>> >> > > > > > https://syzkaller.appspot.com/x/repro.c?id=5104666266304512
>>> >> > > > > > syzkaller reproducer:
>>> >> > > > > > https://syzkaller.appspot.com/x/repro.syz?id=5683447737614336
>>> >> > > > > > Raw console output:
>>> >> > > > > > https://syzkaller.appspot.com/x/log.txt?id=5104818200772608
>>> >> > > > > > Kernel config:
>>> >> > > > > > https://syzkaller.appspot.com/x/.config?id=9118669095563550941
>>> >> > > > > > compiler: gcc (GCC) 7.1.1 20170620
>>> >> > > > > >
>>> >> > > > > > IMPORTANT: if you fix the bug, please add the following tag to the
>>> >> > > > > > commit:
>>> >> > > > > > Reported-by: syzbot+ff87a28e665c163aa7f5@xxxxxxxxxxxxxxxxxxxxxxxxx
>>> >> > > > > > It will help syzbot understand when the bug is fixed. See footer for
>>> >> > > > > > details.
>>> >> > > > > > If you forward the report, please keep this part and the footer.
>>> >> > > > > >
>>> >> > > > > > R10: 0000000000000000 R11: 0000000000000286 R12: 0000000000000003
>>> >> > > > > > R13: 0000000000000004 R14: 0000000000000000 R15: 0000000000000000
>>> >> > > > > > ------------[ cut here ]------------
>>> >> > > > > > kobject_add_internal failed for nodev( with -EEXIST, don't try to
>>> >> > > > > > register
>>> >> > > > > > things with the same name in the same directory.
>>> >> > > > > > sysfs: cannot create duplicate filename '/fs/gfs2/nodev('
>>> >> > > > > > WARNING: CPU: 1 PID: 4473 at lib/kobject.c:238
>>> >> > > > > > kobject_add_internal+0x8d4/0xbc0 lib/kobject.c:235
>>> >> > > > > > CPU: 0 PID: 4474 Comm: syzkaller533472 Not tainted 4.16.0+ #15
>>> >> > > > > > Kernel panic - not syncing: panic_on_warn set ...
>>> >> > > > > >
>>> >> > > > > > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
>>> >> > > > > > Google 01/01/2011
>>> >> > > > > > Call Trace:
>>> >> > > > > > __dump_stack lib/dump_stack.c:17 [inline]
>>> >> > > > > > dump_stack+0x1a7/0x27d lib/dump_stack.c:53
>>> >> > > > > > sysfs_warn_dup+0x83/0xa0 fs/sysfs/dir.c:30
>>> >> > > > > > sysfs_create_dir_ns+0x178/0x1d0 fs/sysfs/dir.c:58
>>> >> > > > > > create_dir lib/kobject.c:69 [inline]
>>> >> > > > > > kobject_add_internal+0x335/0xbc0 lib/kobject.c:227
>>> >> > > > > > kobject_add_varg lib/kobject.c:364 [inline]
>>> >> > > > > > kobject_init_and_add+0xf9/0x150 lib/kobject.c:436
>>> >> > > > > > gfs2_sys_fs_add+0x1ff/0x580 fs/gfs2/sys.c:652
>>> >> > > > > > fill_super+0x86f/0x1d70 fs/gfs2/ops_fstype.c:1118
>>> >> > > > > > gfs2_mount+0x587/0x6e0 fs/gfs2/ops_fstype.c:1321
>>> >> > > > > gfs2 bug, not a sysfs bug, we are correctly warning about an incorrect
>>> >> > > > > usage of the api.
>>> >> > > > Then +gfs2 maintainers.
>>> >> > > >
>>> >> > > > > Now if we should turn this into a non-WARN message, that's a different
>>> >> > > > > thing, I'll gladly take a patch for that.
>>> >> > > > If it's API usage bug in higher level code, then I think WARN is a
>>> >> > > > proper thing. We already had similar ones and they were fixed.
>>> >> > >
>>> >> > > I'm trying to figure out what the test is doing, but it is not very clear.
>>> >> > > At a guess I'd say that perhaps it is trying to mount multiple filesystems
>>> >> > > with the same label? If that is the case then it is not allowed, and it
>>> >> > > should be caught be the sysfs code and result in a refusal to mount, which
>>> >> > > is what I think I see here. Knowing which sysfs directory is involved would
>>> >> > > allow us to confirm, but I suspect that the test needs altering to give each
>>> >> > > gfs2 mount a different label at an initial guess,
>>> >> >
>>> >> > Hi Steve,
>>> >> >
>>> >> > But Greg claims that this is incorrect usage of sysfs API:
>>> >> >
>>> >> > > gfs2 bug, not a sysfs bug, we are correctly warning about an incorrect
>>> >> > > usage of the api.
>>> >> > I think this means that sysfs callers must not try to create the same
>>> >> > thing twice.
>>> >> >
>>> >> > Either way user-space code must not be able to triggers WARNINGs in
>>> >> > kernel. If it does than this is something to fix in kernel.
>>> >>
>>> >> I guess that this warning was added more recently as I've not seen it
>>> >> before.
>>> >
>>> > No, it has been there since at least the 3.13 kernel release (back in
>>> > 2013), which is where it got moved to a separate function, but the logic
>>> > has been around in the kernel tree for much longer than that as seen in
>>> > commit d1c1459e4594 ("sysfs: separate out dup filename warning into a
>>> > separate function")
>>> >
>>> >> My expectation is that it will return -EEXIST and not print a
>>> >> warning there. To avoid that we would have to create a new list of GFS2
>>> >> superblocks, and check the list for each mount I think. We could do that,
>>> >> but it seems a bit odd to duplicate code that is already there and working.
>>> >
>>> > Don't you have a list of the "names" of the things you are creating
>>> > somewhere? Or are you relying on sysfs to do your housekeeping for you?
>>> >
>>> > Also, why did this trigger a syzbot report? It's only a dump_stack()
>>> > reference, one showing that yes, this is something that should not be
>>> > done, but the kernel keeps on working afterward.
>>>
>>> There is a WARN(), not just dump_stack():
>>>
>>> WARN(1, "%s failed for %s with "
>>> "-EEXIST, don't try to register things with "
>>> "the same name in the same directory.\n",
>>> __func__, kobject_name(kobj));
>>
>> Ah, that's a kobject warning, not a sysfs one. Was looking too far down
>> the call chain. We can change this to be dump_stack() as well if you
>> think that would help out. Maybe remove it entirely as sysfs already
>> does dump the stack?
>
>
> kobject_add is called not only from sysfs, right?

Wait, it's the other way around.
I've already mailed a patch that does s/WARN/pr_err/. But can remove
it entirely, whichever you prefer.

> I was just going to send a patch that changes WARN to
> pr_err+dump_stack, but noticed that we have 4 active bugs on this
> WARNING:
>
> WARNING: kobject bug in gfs2_sys_fs_add
> https://syzkaller.appspot.com/bug?id=057673a56dab61b3a447989b67f10b205111c8f4
>
> WARNING: kobject bug in br_add_if
> https://syzkaller.appspot.com/bug?id=3e0339080acd6a2a350a900bc6533b03f5498490
>
> WARNING: kobject bug in netdev_queue_update_kobjects
> https://syzkaller.appspot.com/bug?id=86a8e2ab50527d5a5eb4fad2fc15df609f22d86a
>
> WARNING: kobject bug in device_add
> https://syzkaller.appspot.com/bug?id=57eba87aff7669512fb68e56a932b01805342d13
>
> We want to ignore all of them?
> As a side signal none of them got any attention, except Eric noting
> that, yes, it is noisy:
> https://groups.google.com/d/msg/syzkaller-bugs/XDTC7Iv4IKY/Ab_tgZ4HAQAJ
>
> So I guess I will still mail the patch.