Re: [PATCH v2] kasan: remove kasan_find_vm_area() to prevent possible deadlock

[Byungchul Park <byungchul@sk.com>]

On Thu, Jul 10, 2025 at 02:43:15PM +0200, Andrey Konovalov wrote:
> On Thu, Jul 3, 2025 at 8:10 PM Yeoreum Yun <yeoreum.yun@arm.com> wrote:
> >
> > find_vm_area() couldn't be called in atomic_context.
> > If find_vm_area() is called to reports vm area information,
> > kasan can trigger deadlock like:
> >
> > CPU0                                CPU1
> > vmalloc();
> >  alloc_vmap_area();
> >   spin_lock(&vn->busy.lock)
> >                                     spin_lock_bh(&some_lock);
> >    <interrupt occurs>
> >    <in softirq>
> >    spin_lock(&some_lock);
> >                                     <access invalid address>
> >                                     kasan_report();
> >                                      print_report();
> >                                       print_address_description();
> >                                        kasan_find_vm_area();
> >                                         find_vm_area();
> >                                          spin_lock(&vn->busy.lock) // deadlock!
> >
> > To prevent possible deadlock while kasan reports, remove kasan_find_vm_area().
> >
> > Fixes: c056a364e954 ("kasan: print virtual mapping info in reports")
> > Reported-by: Yunseong Kim <ysk@kzalloc.com>
> > Signed-off-by: Yeoreum Yun <yeoreum.yun@arm.com>
> 
> As a fix:
> 
> Acked-by: Andrey Konovalov <andreyknvl@gmail.com>
> 
> But it would be great to figure out a way to eventually restore this
> functionality; I'll file a bug for this once this patch lands. The
> virtual mapping info helps with real issues: e.g. just recently it
> helped me to quickly see the issue that caused a false-positive report

I checked the critical section by &vn->busy.lock in find_vm_area().  The
time complextity looks O(log N).  I don't think an irq disabled section
of O(log N) is harmful.  I still think using
spin_lock_irqsave(&vn->busy.lock) can resolve this issue with no worry
of significant irq delay.  Am I missing something?

If it's unacceptable for some reasons, why don't we introduce kind of
try_find_vm_area() using trylock so as to go ahead only if there's no
lock contention?

	Byungchul

> [1].
> 
> [1] https://lore.kernel.org/all/CA+fCnZfzHOFjVo43UZK8H6h3j=OHjfF13oFJvT0P-SM84Oc4qQ@mail.gmail.com/