v3:
- fix commit messages to properly reflect the direction of the overwrite
- justify the less-than-word-size patch better
- add Acks
- move some Fixes up into the commit log as just references
v2: https://lore.kernel.org/lkml/[email protected]
v1: https://lore.kernel.org/lkml/[email protected]
This fixes redzoning vs the freelist pointer (both for middle-position
and very small caches). Both are "theoretical" fixes, in that I see no
evidence of such small-sized caches actually be used in the kernel, but
that's no reason to let the bugs continue to exist. :)
Note on patch 2: Christopher NAKed it, but I actually think this is a
reasonable thing to add -- the "too small" check is only made when built
with CONFIG_DEBUG_VM, so it *is* actually possible for someone to trip
over this directly, even if it would never make it into a released
kernel. I see no reason to just leave this foot-gun in place, though, so
we might as well just fix it too. (Which seems to be what Longman was
similarly supporting, IIUC.)
Anyway, if patch 2 stays NAKed, that's fine. It's entirely separable,
and the other 2 can land. :)
Thanks!
-Kees
Kees Cook (3):
mm/slub: Clarify verification reporting
mm/slub: Fix redzoning for small allocations
mm/slub: Actually fix freelist pointer vs redzoning
Documentation/vm/slub.rst | 10 +++++-----
mm/slub.c | 36 +++++++++++++++---------------------
2 files changed, 20 insertions(+), 26 deletions(-)
--
2.25.1
It turns out that SLUB redzoning ("slub_debug=Z") checks from
s->object_size rather than from s->inuse (which is normally bumped
to make room for the freelist pointer), so a cache created with an
object size less than 24 would have the freelist pointer written beyond
s->object_size, causing the redzone to be corrupted by the freelist
pointer. This was very visible with "slub_debug=ZF":
BUG test (Tainted: G B ): Right Redzone overwritten
-----------------------------------------------------------------------------
INFO: 0xffff957ead1c05de-0xffff957ead1c05df @offset=1502. First byte 0x1a instead of 0xbb
INFO: Slab 0xffffef3950b47000 objects=170 used=170 fp=0x0000000000000000 flags=0x8000000000000200
INFO: Object 0xffff957ead1c05d8 @offset=1496 fp=0xffff957ead1c0620
Redzone (____ptrval____): bb bb bb bb bb bb bb bb ........
Object (____ptrval____): 00 00 00 00 00 f6 f4 a5 ........
Redzone (____ptrval____): 40 1d e8 1a aa @....
Padding (____ptrval____): 00 00 00 00 00 00 00 00 ........
Adjust the offset to stay within s->object_size.
(Note that no caches in this size range are known to exist in the kernel
currently.)
Reported-by: Marco Elver <[email protected]>
Link: https://lore.kernel.org/linux-mm/[email protected]/
Fixes: 89b83f282d8b (slub: avoid redzone when choosing freepointer location)
Cc: [email protected]
Tested-by: Marco Elver <[email protected]>
Link: https://lore.kernel.org/lkml/CANpmjNOwZ5VpKQn+SYWovTkFB4VsT-RPwyENBmaK0dLcpqStkA@mail.gmail.com
Signed-off-by: Kees Cook <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Link: https://lore.kernel.org/lkml/[email protected]/
---
mm/slub.c | 14 +++-----------
1 file changed, 3 insertions(+), 11 deletions(-)
diff --git a/mm/slub.c b/mm/slub.c
index 752fad36522c..6f115e56c5d0 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3637,7 +3637,6 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
{
slab_flags_t flags = s->flags;
unsigned int size = s->object_size;
- unsigned int freepointer_area;
unsigned int order;
/*
@@ -3646,13 +3645,6 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
* the possible location of the free pointer.
*/
size = ALIGN(size, sizeof(void *));
- /*
- * This is the area of the object where a freepointer can be
- * safely written. If redzoning adds more to the inuse size, we
- * can't use that portion for writing the freepointer, so
- * s->offset must be limited within this for the general case.
- */
- freepointer_area = size;
#ifdef CONFIG_SLUB_DEBUG
/*
@@ -3678,7 +3670,7 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
/*
* With that we have determined the number of bytes in actual use
- * by the object. This is the potential offset to the free pointer.
+ * by the object and redzoning.
*/
s->inuse = size;
@@ -3701,13 +3693,13 @@ static int calculate_sizes(struct kmem_cache *s, int forced_order)
*/
s->offset = size;
size += sizeof(void *);
- } else if (freepointer_area > sizeof(void *)) {
+ } else {
/*
* Store freelist pointer near middle of object to keep
* it away from the edges of the object to avoid small
* sized over/underflows from neighboring allocations.
*/
- s->offset = ALIGN(freepointer_area / 2, sizeof(void *));
+ s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *));
}
#ifdef CONFIG_SLUB_DEBUG
--
2.25.1
On 10/15/20 10:23 AM, Christopher Lameter wrote:
> On Wed, 14 Oct 2020, Kees Cook wrote:
>
>> Note on patch 2: Christopher NAKed it, but I actually think this is a
>> reasonable thing to add -- the "too small" check is only made when built
>> with CONFIG_DEBUG_VM, so it *is* actually possible for someone to trip
>> over this directly, even if it would never make it into a released
>> kernel. I see no reason to just leave this foot-gun in place, though, so
>> we might as well just fix it too. (Which seems to be what Longman was
>> similarly supporting, IIUC.)
>
> Well then remove the duplication of checks. The NAK was there because it
> seems that you were not aware of the existing checks.
>
>> Anyway, if patch 2 stays NAKed, that's fine. It's entirely separable,
>> and the other 2 can land. :)
>
> Just deal with the old checks too and it will be fine.
Yeah, the existing check is under CONFIG_DEBUG_VM, which means it's not active
on some configurations. Creating a cache is not exactly fast path operation, so
I would remove this guard.
As for the minimum size check, I would probably remove it (but watch out if
SLAB/SLOB can handle it). It's not effective to use slab cache for 4-byte
objects, but why make it an error.
On Wed, 14 Oct 2020, Kees Cook wrote:
> Note on patch 2: Christopher NAKed it, but I actually think this is a
> reasonable thing to add -- the "too small" check is only made when built
> with CONFIG_DEBUG_VM, so it *is* actually possible for someone to trip
> over this directly, even if it would never make it into a released
> kernel. I see no reason to just leave this foot-gun in place, though, so
> we might as well just fix it too. (Which seems to be what Longman was
> similarly supporting, IIUC.)
Well then remove the duplication of checks. The NAK was there because it
seems that you were not aware of the existing checks.
> Anyway, if patch 2 stays NAKed, that's fine. It's entirely separable,
> and the other 2 can land. :)
Just deal with the old checks too and it will be fine.
On Thu, Oct 15, 2020 at 11:44:15AM +0200, Vlastimil Babka wrote:
> On 10/15/20 10:23 AM, Christopher Lameter wrote:
> > On Wed, 14 Oct 2020, Kees Cook wrote:
> >
> > > Note on patch 2: Christopher NAKed it, but I actually think this is a
> > > reasonable thing to add -- the "too small" check is only made when built
> > > with CONFIG_DEBUG_VM, so it *is* actually possible for someone to trip
> > > over this directly, even if it would never make it into a released
> > > kernel. I see no reason to just leave this foot-gun in place, though, so
> > > we might as well just fix it too. (Which seems to be what Longman was
> > > similarly supporting, IIUC.)
> >
> > Well then remove the duplication of checks. The NAK was there because it
> > seems that you were not aware of the existing checks.
> >
> > > Anyway, if patch 2 stays NAKed, that's fine. It's entirely separable,
> > > and the other 2 can land. :)
> >
> > Just deal with the old checks too and it will be fine.
>
> Yeah, the existing check is under CONFIG_DEBUG_VM, which means it's not
> active on some configurations. Creating a cache is not exactly fast path
> operation, so I would remove this guard.
> As for the minimum size check, I would probably remove it (but watch out if
> SLAB/SLOB can handle it). It's not effective to use slab cache for 4-byte
> objects, but why make it an error.
Err, why did the check exist to begin with? If the check isn't wanted,
that's one thing, but I was just trying to fix what I saw in the redzone
handling. What is preferred here?
1) drop patch 2
2) keep patch 2, but also:
a) validate slab/slob can handle < word-sized allocations
b) remove check in kmem_cache_sanity_check
option 2a seems like it could be fragile if I miss something. I think
I'd rather just take option 1.
--
Kees Cook