Re: syzbot rcu/debugobjects warning

[Thomas Gleixner]

On Thu, 22 Mar 2018, Joel Fernandes wrote:

> Hi Paul, Thomas,
> 
> I received a crash report from syzbot on the android 4.9 kernel and I
> am looking into it, it seems the debugobjects subsystem is warning
> that a certain RCU structure is not allocated on the stack, but is
> annotated to be.
> 
> ------------[ cut here ]------------
> WARNING: CPU: 1 PID: 0 at lib/debugobjects.c:300
> debug_object_is_on_stack lib/debugobjects.c:300 [inline]
> WARNING: CPU: 1 PID: 0 at lib/debugobjects.c:300
> __debug_object_init+0x526/0xc40 lib/debugobjects.c:326
> [...]
>  <IRQ> [  150.631700]  [<ffffffff81d96069>] dump_stack+0xc1/0x128
> lib/dump_stack.c:51
>  [<ffffffff8142fbd1>] panic+0x1bc/0x3a8 kernel/panic.c:179
>  [<ffffffff81131894>] __warn+0x1c4/0x1e0 kernel/panic.c:542
>  [<ffffffff81131afc>] warn_slowpath_null+0x2c/0x40 kernel/panic.c:585
>  [<ffffffff81dfefb6>] debug_object_is_on_stack lib/debugobjects.c:300 [inline]
>  [<ffffffff81dfefb6>] __debug_object_init+0x526/0xc40 lib/debugobjects.c:326
>  [<ffffffff81dff709>] debug_object_init_on_stack+0x19/0x20
> lib/debugobjects.c:378
>  [<ffffffff81287a93>] init_rcu_head_on_stack kernel/rcu/update.c:403 [inline]
>  [<ffffffff81287a93>] __wait_rcu_gp+0x93/0x1b0 kernel/rcu/update.c:358
>  [<ffffffff81290251>] synchronize_rcu.part.65+0x101/0x110
> kernel/rcu/tree_plugin.h:678
>  [<ffffffff81290287>] synchronize_rcu+0x27/0x90 kernel/rcu/tree_plugin.h:679
>  [<ffffffff83588b35>] __l2tp_session_unhash+0x3d5/0x550
> net/l2tp/l2tp_core.c:1792
> 
> The full report is here:
> https://syzkaller.appspot.com/bug?extid=e6a19b585ab2dba3eee8

This is beyond useless. That brings me to a google 'Sign in' page. Please
use accessible storage. That information is hardly secrit.

> It seems as per the code that the structure is on the stack so its
> weird why debugobjects thinks its not.
> The object in question is allocated on the stack by the __wait_rcu_gp
> macro when its called from synchronize_rcu:
> 
> #define _wait_rcu_gp(checktiny, ...) \
> do {                                                                    \
>         call_rcu_func_t __crcu_array[] = { __VA_ARGS__ };               \
>         struct rcu_synchronize __rs_array[ARRAY_SIZE(__crcu_array)];    \
>         __wait_rcu_gp(checktiny, ARRAY_SIZE(__crcu_array),              \
>                         __crcu_array, __rs_array);                      \
> } while (0)
> 
> 
> Any debug ideas or thoughts about it?

I assume it emitted:
     pr_warn("object is not on stack, but annotated\n");

before dumping the WARN_ON(). Right? If so, then you might have run into a
stack corruption. But hard to tell. Please add something like this:

     pr_warn("object %p is not on stack %p, but annotated\n", obj,
     	     task_stack_page(current));
Thanks,

	tglx