[v4,16/35] mm: slub: Unpoison the memchr_inv() return value

Commit Message

Ilya Leoshkevich June 13, 2024, 3:34 p.m. UTC

Even though the KMSAN warnings generated by memchr_inv() are suppressed
by metadata_access_enable(), its return value may still be poisoned.

The reason is that the last iteration of memchr_inv() returns
`*start != value ? start : NULL`, where *start is poisoned. Because of
this, somewhat counterintuitively, the shadow value computed by
visitSelectInst() is equal to `(uintptr_t)start`.

The intention behind guarding memchr_inv() behind
metadata_access_enable() is to touch poisoned metadata without
triggering KMSAN, so unpoison its return value.

Acked-by: Vlastimil Babka <[e-mail]>
Signed-off-by: Ilya Leoshkevich <[e-mail]>
---
 mm/slub.c | 2 ++
 1 file changed, 2 insertions(+)

Comments

Alexander Potapenko June 18, 2024, 2:38 p.m. UTC | #1

On Thu, Jun 13, 2024 at 5:39 PM Ilya Leoshkevich <[e-mail]> wrote:
>
> Even though the KMSAN warnings generated by memchr_inv() are suppressed
> by metadata_access_enable(), its return value may still be poisoned.
>
> The reason is that the last iteration of memchr_inv() returns
> `*start != value ? start : NULL`, where *start is poisoned. Because of
> this, somewhat counterintuitively, the shadow value computed by
> visitSelectInst() is equal to `(uintptr_t)start`.
>
> The intention behind guarding memchr_inv() behind
> metadata_access_enable() is to touch poisoned metadata without
> triggering KMSAN, so unpoison its return value.

What do you think about applying __no_kmsan_checks to these functions instead?

Ilya Leoshkevich June 19, 2024, 11:46 a.m. UTC | #2

On Tue, 2024-06-18 at 16:38 +0200, Alexander Potapenko wrote:
> On Thu, Jun 13, 2024 at 5:39 PM Ilya Leoshkevich <[e-mail]>
> wrote:
> > 
> > Even though the KMSAN warnings generated by memchr_inv() are
> > suppressed
> > by metadata_access_enable(), its return value may still be
> > poisoned.
> > 
> > The reason is that the last iteration of memchr_inv() returns
> > `*start != value ? start : NULL`, where *start is poisoned. Because
> > of
> > this, somewhat counterintuitively, the shadow value computed by
> > visitSelectInst() is equal to `(uintptr_t)start`.
> > 
> > The intention behind guarding memchr_inv() behind
> > metadata_access_enable() is to touch poisoned metadata without
> > triggering KMSAN, so unpoison its return value.
> 
> What do you think about applying __no_kmsan_checks to these functions
> instead?

Ok, will do. The __no_kmsan_checks approach is already taken by
"mm: kfence: Disable KMSAN when checking the canary", so we might as
well be consistent in how we fix these issues.

Patch

diff --git a/mm/slub.c b/mm/slub.c
index a290f6c63e7b..b9101b2dc9aa 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -1185,6 +1185,7 @@  static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab,
 	metadata_access_enable();
 	fault = memchr_inv(kasan_reset_tag(start), value, bytes);
 	metadata_access_disable();
+	kmsan_unpoison_memory(&fault, sizeof(fault));
 	if (!fault)
 		return 1;
 
@@ -1291,6 +1292,7 @@  static void slab_pad_check(struct kmem_cache *s, struct slab *slab)
 	metadata_access_enable();
 	fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder);
 	metadata_access_disable();
+	kmsan_unpoison_memory(&fault, sizeof(fault));
 	if (!fault)
 		return;
 	while (end > fault && end[-1] == POISON_INUSE)