Re: WARNING: ODEBUG bug in tcindex_destroy_work (3)

[Thomas Gleixner]

Cong Wang <xiyou.wangcong@xxxxxxxxx> writes:
> On Sat, Mar 21, 2020 at 3:19 AM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> > ------------[ cut here ]------------
>> > ODEBUG: free active (active state 0) object type: work_struct hint: tcindex_destroy_rexts_work+0x0/0x20 net/sched/cls_tcindex.c:143
>> ...
>> >  __debug_check_no_obj_freed lib/debugobjects.c:967 [inline]
>> >  debug_check_no_obj_freed+0x2e1/0x445 lib/debugobjects.c:998
>> >  kfree+0xf6/0x2b0 mm/slab.c:3756
>> >  tcindex_destroy_work+0x2e/0x70 net/sched/cls_tcindex.c:231
>>
>> So this is:
>>
>>         kfree(p->perfect);
>>
>> Looking at the place which queues that work:
>>
>> tcindex_destroy()
>>
>>    if (p->perfect) {
>>         if (tcf_exts_get_net(&r->exts))
>>             tcf_queue_work(&r-rwork, tcindex_destroy_rexts_work);
>>         else
>>             __tcindex_destroy_rexts(r)
>>    }
>>
>>    .....
>>
>>    tcf_queue_work(&p->rwork, tcindex_destroy_work);
>>
>> So obviously if tcindex_destroy_work() runs before
>> tcindex_destroy_rexts_work() then the above happens.
>
> We use an ordered workqueue for tc filters, so these two
> works are executed in the same order as they are queued.

The workqueue is ordered, but look how the work is queued on the work
queue:

tcf_queue_work()
  queue_rcu_work()
    call_rcu(&rwork->rcu, rcu_work_rcufn);

So after the grace period elapses rcu_work_rcufn() queues it in the
actual work queue.

Now tcindex_destroy() is invoked via tcf_proto_destroy() which can be
invoked from preemtible context. Now assume the following:

CPU0
  tcf_queue_work()
    tcf_queue_work(&r->rwork, tcindex_destroy_rexts_work);

-> Migration

CPU1
   tcf_queue_work(&p->rwork, tcindex_destroy_work);

So your RCU callbacks can be placed on different CPUs which obviously
has no ordering guarantee at all. See also:

  https://mirrors.edge.kernel.org/pub/linux/kernel/people/paulmck/Answers/RCU/RCUCBordering.html

Disabling preemption would "fix" it today, but that documentation
explicitely says that it is an implementation detail, but not
guaranteed by design.

Thanks,

        tglx