Re: [mm/slub] 555b8c8cb3: WARNING:at_lib/stackdepot.c:#stack_depot_fetch

[Hyeonggon Yoo <42.hyeyoo@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2422 bytes --]

On Mon, Apr 04, 2022 at 05:18:16PM +0200, Marco Elver wrote:
> On Mon, 4 Apr 2022 at 16:20, Vlastimil Babka <vbabka@suse.cz> wrote:
> >
> > On 4/4/22 10:10, Marco Elver wrote:
> > > On Mon, Apr 04, 2022 at 12:05PM +0900, Hyeonggon Yoo wrote:
> > > (Maybe CONFIG_KCSAN_STRICT=y is going to yield something? I still doubt
> > > it thought, this bug is related to corrupted stackdepot handle
> > > somewhere...)
> > >
> > >> I noticed that it is not reproduced when KASAN=y and KFENCE=n (reproduced 0 of 181).
> > >> and it was reproduced 56 of 196 when KASAN=n and KFENCE=y
> > >>
> > >> maybe this issue is related to kfence?
> >
> > Hmm kfence seems to be a good lead. If I understand kfence_guarded_alloc()
> > correctly, it tries to set up something that really looks like a normal slab
> > page? Especially the part with comment /* Set required slab fields. */
> > But it doesn't seem to cover the debugging parts that SLUB sets up with
> > alloc_debug_processing(). This includes alloc stack saving, thus, after
> > commit 555b8c8cb3, a stackdepot handle setting. It probably normally doesn't
> > matter as is_kfence_address() redirects processing of kfence-allocated
> > objects so we don't hit any slub code that expects the debugging parts to be
> > properly initialized.
> >
> > But here we are in mem_dump_obj() -> kmem_dump_obj() -> kmem_obj_info().
> > Because kmem_valid_obj() returned true, fooled by folio_test_slab()
> > returning true because of the /* Set required slab fields. */ code.
> > Yet the illusion is not perfect and we read garbage instead of a valid
> > stackdepot handle.
> >
> > IMHO we should e.g. add the appropriate is_kfence_address() test into
> > kmem_valid_obj(), to exclude kfence-allocated objects? Sounds much simpler
> > than trying to extend the illusion further to make kmem_dump_obj() work?
> > Instead kfence could add its own specific handler to mem_dump_obj() to print
> > its debugging data?
> 
> I think this explanation makes sense!  Indeed, KFENCE already records
> allocation stacks internally anyway, so it should be straightforward
> to convince it to just print that.
>

Thank you both! Yeah the explanation makes sense... thats why KASAN/KCSAN couldn't yield anything -- it was not overwritten.

I'm writing a fix and will test if the bug disappears.
This may take few days.

Thanks!
Hyeonggon

> Thanks,
> -- Marco