On Thu, Jan 11, 2024 at 4:00 PM Tetsuo Handa
<[email protected]> wrote:
>
> Commit a414d4286f34 ("kasan: handle concurrent kasan_record_aux_stack calls")
> calls raw_spin_lock_init(&alloc_meta->aux_lock) after __memset() in
> kasan_init_object_meta(), but does not call raw_spin_lock_init() after __memset()
> in release_alloc_meta(), resulting in lock map information being zeroed out?
>
> We should not zero out the whole sizeof(struct kasan_alloc_meta) bytes from
> release_alloc_meta() in order not to undo raw_spin_lock_init() from
> kasan_init_object_meta() ?
Yes, already sent a fix:
https://lore.kernel.org/linux-mm/[email protected]/
Thank you!