Re: [REGRESSION] bpf verifier slowdown due to vrealloc() change since 6.15-rc6

[Kees Cook <kees@kernel.org>]

On Thu, May 15, 2025 at 08:55:47AM -0700, Andrii Nakryiko wrote:
> On Thu, May 15, 2025 at 8:53 AM Kees Cook <kees@kernel.org> wrote:
> >
> > On Thu, May 15, 2025 at 08:47:47AM -0700, Kees Cook wrote:
> > > On Thu, May 15, 2025 at 09:12:25PM +0800, Shung-Hsi Yu wrote:
> > > > Bisect was done by Pawan and got to commit a0309faf1cb0 "mm: vmalloc:
> > > > support more granular vrealloc() sizing"[2]. To further zoom in the
> > >
> > > Can you try this patch? It's a clear bug fix, but if it doesn't improve
> > > things, I have another idea to rearrange the memset.
> >
> > Here's the patch (on top of the prior one) that relocates the memset:
> >
> >
> > From 0bc71b78603500705aca77f82de8ed1fc595c4c3 Mon Sep 17 00:00:00 2001
> > From: Kees Cook <kees@kernel.org>
> > Date: Thu, 15 May 2025 08:48:24 -0700
> > Subject: [PATCH] mm: vmalloc: Only zero-init on vrealloc shrink
> >
> > The common case is to grow reallocations, and since init_on_alloc will
> > have already zeroed the whole allocation, we only need to zero when
> > shrinking the allocation.
> >
> > Fixes: a0309faf1cb0 ("mm: vmalloc: support more granular vrealloc() sizing")
> > Signed-off-by: Kees Cook <kees@kernel.org>
> > ---
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Cc: Uladzislau Rezki <urezki@gmail.com>
> > Cc: <linux-mm@kvack.org>
> > ---
> >  mm/vmalloc.c | 12 +++++++-----
> >  1 file changed, 7 insertions(+), 5 deletions(-)
> >
> > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > index 74bd00fd734d..83bedb1559ac 100644
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -4093,8 +4093,8 @@ void *vrealloc_noprof(const void *p, size_t size, gfp_t flags)
> >          * would be a good heuristic for when to shrink the vm_area?
> >          */
> >         if (size <= old_size) {
> > -               /* Zero out "freed" memory. */
> > -               if (want_init_on_free())
> > +               /* Zero out "freed" memory, potentially for future realloc. */
> > +               if (want_init_on_free() || want_init_on_alloc(flags))
> >                         memset((void *)p + size, 0, old_size - size);
> >                 vm->requested_size = size;
> >                 kasan_poison_vmalloc(p + size, old_size - size);
> > @@ -4107,9 +4107,11 @@ void *vrealloc_noprof(const void *p, size_t size, gfp_t flags)
> >         if (size <= alloced_size) {
> >                 kasan_unpoison_vmalloc(p + old_size, size - old_size,
> >                                        KASAN_VMALLOC_PROT_NORMAL);
> > -               /* Zero out "alloced" memory. */
> > -               if (want_init_on_alloc(flags))
> > -                       memset((void *)p + old_size, 0, size - old_size);
> > +               /*
> > +                * No need to zero memory here, as unused memory will have
> > +                * already been zeroed at initial allocation time or during
> > +                * realloc shrink time.
> > +                */
> >                 vm->requested_size = size;
> 
> This vm->requested_size change you are adding should also fix the
> kasan issue reported by syzbot ([0]).
> 
>   [0] https://lore.kernel.org/bpf/68213ddf.050a0220.f2294.0045.GAE@google.com/

Yes, this looks very much like the kasan oops that motivated the initial
patch:

https://lore.kernel.org/all/20250408192503.6149a816@outsider.home/

-- 
Kees Cook