[PATCH v8 0/2] allow KASAN to detect UAF in SLAB_TYPESAFE_BY_RCU slabs

[PATCH v8 0/2] allow KASAN to detect UAF in SLAB_TYPESAFE_BY_RCU slabs

[Jann Horn]

Hi!

The purpose of the series is to allow KASAN to detect use-after-free
access in SLAB_TYPESAFE_BY_RCU slab caches, by essentially making them
behave as if the cache was not SLAB_TYPESAFE_BY_RCU but instead every
kfree() in the cache was a kfree_rcu().
This is gated behind a config flag that is supposed to only be enabled
in fuzzing/testing builds where the performance impact doesn't matter.

Output of the new kunit testcase I added to the KASAN test suite:
==================================================================
BUG: KASAN: slab-use-after-free in kmem_cache_rcu_uaf+0x3ae/0x4d0
Read of size 1 at addr ffff888106224000 by task kunit_try_catch/224

CPU: 7 PID: 224 Comm: kunit_try_catch Tainted: G    B            N 6.10.0-00003-g065427d4b87f #430
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
Call Trace:
 <TASK>
 dump_stack_lvl+0x53/0x70
 print_report+0xce/0x670
[...]
 kasan_report+0xa5/0xe0
[...]
 kmem_cache_rcu_uaf+0x3ae/0x4d0
[...]
 kunit_try_run_case+0x1b3/0x490
[...]
 kunit_generic_run_threadfn_adapter+0x80/0xe0
 kthread+0x2a5/0x370
[...]
 ret_from_fork+0x34/0x70
[...]
 ret_from_fork_asm+0x1a/0x30
 </TASK>

Allocated by task 224:
 kasan_save_stack+0x33/0x60
 kasan_save_track+0x14/0x30
 __kasan_slab_alloc+0x6e/0x70
 kmem_cache_alloc_noprof+0xef/0x2b0
 kmem_cache_rcu_uaf+0x10d/0x4d0
 kunit_try_run_case+0x1b3/0x490
 kunit_generic_run_threadfn_adapter+0x80/0xe0
 kthread+0x2a5/0x370
 ret_from_fork+0x34/0x70
 ret_from_fork_asm+0x1a/0x30

Freed by task 0:
 kasan_save_stack+0x33/0x60
 kasan_save_track+0x14/0x30
 kasan_save_free_info+0x3b/0x60
 __kasan_slab_free+0x57/0x80
 slab_free_after_rcu_debug+0xe3/0x220
 rcu_core+0x676/0x15b0
 handle_softirqs+0x22f/0x690
 irq_exit_rcu+0x84/0xb0
 sysvec_apic_timer_interrupt+0x6a/0x80
 asm_sysvec_apic_timer_interrupt+0x1a/0x20

Last potentially related work creation:
 kasan_save_stack+0x33/0x60
 __kasan_record_aux_stack+0x8e/0xa0
 kmem_cache_free+0x10c/0x420
 kmem_cache_rcu_uaf+0x16e/0x4d0
 kunit_try_run_case+0x1b3/0x490
 kunit_generic_run_threadfn_adapter+0x80/0xe0
 kthread+0x2a5/0x370
 ret_from_fork+0x34/0x70
 ret_from_fork_asm+0x1a/0x30

The buggy address belongs to the object at ffff888106224000
 which belongs to the cache test_cache of size 200
The buggy address is located 0 bytes inside of
 freed 200-byte region [ffff888106224000, ffff8881062240c8)

The buggy address belongs to the physical page:
page: refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x106224
head: order:1 mapcount:0 entire_mapcount:0 nr_pages_mapped:0 pincount:0
flags: 0x200000000000040(head|node=0|zone=2)
page_type: 0xffffefff(slab)
raw: 0200000000000040 ffff88810621c140 dead000000000122 0000000000000000
raw: 0000000000000000 00000000801f001f 00000001ffffefff 0000000000000000
head: 0200000000000040 ffff88810621c140 dead000000000122 0000000000000000
head: 0000000000000000 00000000801f001f 00000001ffffefff 0000000000000000
head: 0200000000000001 ffffea0004188901 ffffffffffffffff 0000000000000000
head: 0000000000000002 0000000000000000 00000000ffffffff 0000000000000000
page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff888106223f00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 ffff888106223f80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>ffff888106224000: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                   ^
 ffff888106224080: fb fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc
 ffff888106224100: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
==================================================================
    ok 38 kmem_cache_rcu_uaf

Signed-off-by: Jann Horn <jannh@google.com>
---
Changes in v8:
- in patch 2/2:
  - move rcu_barrier() out of locked region (vbabka)
  - rearrange code in slab_free_after_rcu_debug (vbabka)
- Link to v7: https://lore.kernel.org/r/20240808-kasan-tsbrcu-v7-0-0d0590c54ae6@google.com

Changes in v7:
- in patch 2/2:
  - clarify kconfig comment (Marco)
  - fix memory leak (vbabka and dsterba)
  - move rcu_barrier() call up into kmem_cache_destroy() to hopefully
    make the merge conflict with vbabka's
    https://lore.kernel.org/all/20240807-b4-slab-kfree_rcu-destroy-v2-1-ea79102f428c@suse.cz/
    easier to deal with
- Link to v6: https://lore.kernel.org/r/20240802-kasan-tsbrcu-v6-0-60d86ea78416@google.com

Changes in v6:
- in patch 1/2:
  - fix commit message (Andrey)
  - change comments (Andrey)
  - fix mempool handling of kfence objects (Andrey)
- in patch 2/2:
  - fix is_kfence_address argument (syzbot and Marco)
  - refactor slab_free_hook() to create "still_accessible" variable
  - change kasan_slab_free() hook argument to "still_accessible"
  - add documentation to kasan_slab_free() hook
- Link to v5: https://lore.kernel.org/r/20240730-kasan-tsbrcu-v5-0-48d3cbdfccc5@google.com

Changes in v5:
- rebase to latest origin/master (akpm), no other changes from v4
- Link to v4: https://lore.kernel.org/r/20240729-kasan-tsbrcu-v4-0-57ec85ef80c6@google.com

Changes in v4:
- note I kept vbabka's ack for the SLUB changes in patch 1/2 since the
  SLUB part didn't change, even though I refactored a bunch of the
  KASAN parts
- in patch 1/2 (major rework):
  - fix commit message (Andrey)
  - add doc comments in header (Andrey)
  - remove "ip" argument from __kasan_slab_free()
  - rework the whole check_slab_free() thing and move code around (Andrey)
- in patch 2/2:
  - kconfig description and dependency changes (Andrey)
  - remove useless linebreak (Andrey)
  - fix comment style (Andrey)
  - fix do_slab_free() invocation (kernel test robot)
- Link to v3: https://lore.kernel.org/r/20240725-kasan-tsbrcu-v3-0-51c92f8f1101@google.com

Changes in v3:
- in patch 1/2, integrate akpm's fix for !CONFIG_KASAN build failure
- in patch 2/2, as suggested by vbabka, use dynamically allocated
  rcu_head to avoid having to add slab metadata
- in patch 2/2, add a warning in the kconfig help text that objects can
  be recycled immediately under memory pressure
- Link to v2: https://lore.kernel.org/r/20240724-kasan-tsbrcu-v2-0-45f898064468@google.com

Changes in v2:
Patch 1/2 is new; it's some necessary prep work for the main patch to
work, though the KASAN integration maybe is a bit ugly.
Patch 2/2 is a rebased version of the old patch, with some changes to
how the config is wired up, with poison/unpoison logic added as
suggested by dvyukov@ back then, with cache destruction fixed using
rcu_barrier() as pointed out by dvyukov@ and the test robot, and a test
added as suggested by elver@.

---
Jann Horn (2):
      kasan: catch invalid free before SLUB reinitializes the object
      slub: Introduce CONFIG_SLUB_RCU_DEBUG

 include/linux/kasan.h | 63 ++++++++++++++++++++++++++++++++++---
 mm/Kconfig.debug      | 32 +++++++++++++++++++
 mm/kasan/common.c     | 62 ++++++++++++++++++++++---------------
 mm/kasan/kasan_test.c | 46 +++++++++++++++++++++++++++
 mm/slab_common.c      | 16 ++++++++++
 mm/slub.c             | 86 ++++++++++++++++++++++++++++++++++++++++++++++-----
 6 files changed, 267 insertions(+), 38 deletions(-)
---
base-commit: 94ede2a3e9135764736221c080ac7c0ad993dc2d
change-id: 20240723-kasan-tsbrcu-b715a901f776
-- 
Jann Horn <jannh@google.com>

[PATCH v8 1/2] kasan: catch invalid free before SLUB reinitializes the object

[Jann Horn]

Currently, when KASAN is combined with init-on-free behavior, the
initialization happens before KASAN's "invalid free" checks.

More importantly, a subsequent commit will want to RCU-delay the actual
SLUB freeing of an object, and we'd like KASAN to still validate
synchronously that freeing the object is permitted. (Otherwise this
change will make the existing testcase kmem_cache_invalid_free fail.)

So add a new KASAN hook that allows KASAN to pre-validate a
kmem_cache_free() operation before SLUB actually starts modifying the
object or its metadata.

Inside KASAN, this:

 - moves checks from poison_slab_object() into check_slab_allocation()
 - moves kasan_arch_is_ready() up into callers of poison_slab_object()
 - removes "ip" argument of poison_slab_object() and __kasan_slab_free()
   (since those functions no longer do any reporting)

Acked-by: Vlastimil Babka <vbabka@suse.cz> #slub
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Signed-off-by: Jann Horn <jannh@google.com>
---
 include/linux/kasan.h | 54 ++++++++++++++++++++++++++++++++++++++++++---
 mm/kasan/common.c     | 61 ++++++++++++++++++++++++++++++---------------------
 mm/slub.c             |  7 ++++++
 3 files changed, 94 insertions(+), 28 deletions(-)

diff --git a/include/linux/kasan.h b/include/linux/kasan.h
index 70d6a8f6e25d..1570c7191176 100644
--- a/include/linux/kasan.h
+++ b/include/linux/kasan.h
@@ -172,19 +172,61 @@ static __always_inline void * __must_check kasan_init_slab_obj(
 {
 	if (kasan_enabled())
 		return __kasan_init_slab_obj(cache, object);
 	return (void *)object;
 }
 
-bool __kasan_slab_free(struct kmem_cache *s, void *object,
-			unsigned long ip, bool init);
+bool __kasan_slab_pre_free(struct kmem_cache *s, void *object,
+			unsigned long ip);
+/**
+ * kasan_slab_pre_free - Check whether freeing a slab object is safe.
+ * @object: Object to be freed.
+ *
+ * This function checks whether freeing the given object is safe. It may
+ * check for double-free and invalid-free bugs and report them.
+ *
+ * This function is intended only for use by the slab allocator.
+ *
+ * @Return true if freeing the object is unsafe; false otherwise.
+ */
+static __always_inline bool kasan_slab_pre_free(struct kmem_cache *s,
+						void *object)
+{
+	if (kasan_enabled())
+		return __kasan_slab_pre_free(s, object, _RET_IP_);
+	return false;
+}
+
+bool __kasan_slab_free(struct kmem_cache *s, void *object, bool init);
+/**
+ * kasan_slab_free - Poison, initialize, and quarantine a slab object.
+ * @object: Object to be freed.
+ * @init: Whether to initialize the object.
+ *
+ * This function informs that a slab object has been freed and is not
+ * supposed to be accessed anymore, except for objects in
+ * SLAB_TYPESAFE_BY_RCU caches.
+ *
+ * For KASAN modes that have integrated memory initialization
+ * (kasan_has_integrated_init() == true), this function also initializes
+ * the object's memory. For other modes, the @init argument is ignored.
+ *
+ * This function might also take ownership of the object to quarantine it.
+ * When this happens, KASAN will defer freeing the object to a later
+ * stage and handle it internally until then. The return value indicates
+ * whether KASAN took ownership of the object.
+ *
+ * This function is intended only for use by the slab allocator.
+ *
+ * @Return true if KASAN took ownership of the object; false otherwise.
+ */
 static __always_inline bool kasan_slab_free(struct kmem_cache *s,
 						void *object, bool init)
 {
 	if (kasan_enabled())
-		return __kasan_slab_free(s, object, _RET_IP_, init);
+		return __kasan_slab_free(s, object, init);
 	return false;
 }
 
 void __kasan_kfree_large(void *ptr, unsigned long ip);
 static __always_inline void kasan_kfree_large(void *ptr)
 {
@@ -368,12 +410,18 @@ static inline void kasan_poison_new_object(struct kmem_cache *cache,
 					void *object) {}
 static inline void *kasan_init_slab_obj(struct kmem_cache *cache,
 				const void *object)
 {
 	return (void *)object;
 }
+
+static inline bool kasan_slab_pre_free(struct kmem_cache *s, void *object)
+{
+	return false;
+}
+
 static inline bool kasan_slab_free(struct kmem_cache *s, void *object, bool init)
 {
 	return false;
 }
 static inline void kasan_kfree_large(void *ptr) {}
 static inline void *kasan_slab_alloc(struct kmem_cache *s, void *object,
diff --git a/mm/kasan/common.c b/mm/kasan/common.c
index 85e7c6b4575c..f26bbc087b3b 100644
--- a/mm/kasan/common.c
+++ b/mm/kasan/common.c
@@ -205,59 +205,65 @@ void * __must_check __kasan_init_slab_obj(struct kmem_cache *cache,
 	/* Tag is ignored in set_tag() without CONFIG_KASAN_SW/HW_TAGS */
 	object = set_tag(object, assign_tag(cache, object, true));
 
 	return (void *)object;
 }
 
-static inline bool poison_slab_object(struct kmem_cache *cache, void *object,
-				      unsigned long ip, bool init)
+/* Returns true when freeing the object is not safe. */
+static bool check_slab_allocation(struct kmem_cache *cache, void *object,
+				  unsigned long ip)
 {
-	void *tagged_object;
-
-	if (!kasan_arch_is_ready())
-		return false;
+	void *tagged_object = object;
 
-	tagged_object = object;
 	object = kasan_reset_tag(object);
 
 	if (unlikely(nearest_obj(cache, virt_to_slab(object), object) != object)) {
 		kasan_report_invalid_free(tagged_object, ip, KASAN_REPORT_INVALID_FREE);
 		return true;
 	}
 
-	/* RCU slabs could be legally used after free within the RCU period. */
-	if (unlikely(cache->flags & SLAB_TYPESAFE_BY_RCU))
-		return false;
-
 	if (!kasan_byte_accessible(tagged_object)) {
 		kasan_report_invalid_free(tagged_object, ip, KASAN_REPORT_DOUBLE_FREE);
 		return true;
 	}
 
+	return false;
+}
+
+static inline void poison_slab_object(struct kmem_cache *cache, void *object,
+				      bool init)
+{
+	void *tagged_object = object;
+
+	object = kasan_reset_tag(object);
+
+	/* RCU slabs could be legally used after free within the RCU period. */
+	if (unlikely(cache->flags & SLAB_TYPESAFE_BY_RCU))
+		return;
+
 	kasan_poison(object, round_up(cache->object_size, KASAN_GRANULE_SIZE),
 			KASAN_SLAB_FREE, init);
 
 	if (kasan_stack_collection_enabled())
 		kasan_save_free_info(cache, tagged_object);
+}
 
-	return false;
+bool __kasan_slab_pre_free(struct kmem_cache *cache, void *object,
+				unsigned long ip)
+{
+	if (!kasan_arch_is_ready() || is_kfence_address(object))
+		return false;
+	return check_slab_allocation(cache, object, ip);
 }
 
-bool __kasan_slab_free(struct kmem_cache *cache, void *object,
-				unsigned long ip, bool init)
+bool __kasan_slab_free(struct kmem_cache *cache, void *object, bool init)
 {
-	if (is_kfence_address(object))
+	if (!kasan_arch_is_ready() || is_kfence_address(object))
 		return false;
 
-	/*
-	 * If the object is buggy, do not let slab put the object onto the
-	 * freelist. The object will thus never be allocated again and its
-	 * metadata will never get released.
-	 */
-	if (poison_slab_object(cache, object, ip, init))
-		return true;
+	poison_slab_object(cache, object, init);
 
 	/*
 	 * If the object is put into quarantine, do not let slab put the object
 	 * onto the freelist for now. The object's metadata is kept until the
 	 * object gets evicted from quarantine.
 	 */
@@ -501,17 +507,22 @@ bool __kasan_mempool_poison_object(void *ptr, unsigned long ip)
 		if (check_page_allocation(ptr, ip))
 			return false;
 		kasan_poison(ptr, folio_size(folio), KASAN_PAGE_FREE, false);
 		return true;
 	}
 
-	if (is_kfence_address(ptr))
-		return false;
+	if (is_kfence_address(ptr) || !kasan_arch_is_ready())
+		return true;
 
 	slab = folio_slab(folio);
-	return !poison_slab_object(slab->slab_cache, ptr, ip, false);
+
+	if (check_slab_allocation(slab->slab_cache, ptr, ip))
+		return false;
+
+	poison_slab_object(slab->slab_cache, ptr, false);
+	return true;
 }
 
 void __kasan_mempool_unpoison_object(void *ptr, size_t size, unsigned long ip)
 {
 	struct slab *slab;
 	gfp_t flags = 0; /* Might be executing under a lock. */
diff --git a/mm/slub.c b/mm/slub.c
index 3520acaf9afa..0c98b6a2124f 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2223,12 +2223,19 @@ bool slab_free_hook(struct kmem_cache *s, void *x, bool init)
 		__kcsan_check_access(x, s->object_size,
 				     KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
 
 	if (kfence_free(x))
 		return false;
 
+	/*
+	 * Give KASAN a chance to notice an invalid free operation before we
+	 * modify the object.
+	 */
+	if (kasan_slab_pre_free(s, x))
+		return false;
+
 	/*
 	 * As memory initialization might be integrated into KASAN,
 	 * kasan_slab_free and initialization memset's must be
 	 * kept together to avoid discrepancies in behavior.
 	 *
 	 * The initialization memset's clear the object and the metadata,

-- 
2.46.0.76.ge559c4bf1a-goog

[PATCH v8 2/2] slub: Introduce CONFIG_SLUB_RCU_DEBUG

[Jann Horn]

Currently, KASAN is unable to catch use-after-free in SLAB_TYPESAFE_BY_RCU
slabs because use-after-free is allowed within the RCU grace period by
design.

Add a SLUB debugging feature which RCU-delays every individual
kmem_cache_free() before either actually freeing the object or handing it
off to KASAN, and change KASAN to poison freed objects as normal when this
option is enabled.

For now I've configured Kconfig.debug to default-enable this feature in the
KASAN GENERIC and SW_TAGS modes; I'm not enabling it by default in HW_TAGS
mode because I'm not sure if it might have unwanted performance degradation
effects there.

Note that this is mostly useful with KASAN in the quarantine-based GENERIC
mode; SLAB_TYPESAFE_BY_RCU slabs are basically always also slabs with a
->ctor, and KASAN's assign_tag() currently has to assign fixed tags for
those, reducing the effectiveness of SW_TAGS/HW_TAGS mode.
(A possible future extension of this work would be to also let SLUB call
the ->ctor() on every allocation instead of only when the slab page is
allocated; then tag-based modes would be able to assign new tags on every
reallocation.)

Tested-by: syzbot+263726e59eab6b442723@syzkaller.appspotmail.com
Reviewed-by: Andrey Konovalov <andreyknvl@gmail.com>
Acked-by: Marco Elver <elver@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz> (slab)
Signed-off-by: Jann Horn <jannh@google.com>
---
 include/linux/kasan.h | 17 +++++++----
 mm/Kconfig.debug      | 32 +++++++++++++++++++++
 mm/kasan/common.c     | 11 +++----
 mm/kasan/kasan_test.c | 46 ++++++++++++++++++++++++++++++
 mm/slab_common.c      | 16 +++++++++++
 mm/slub.c             | 79 +++++++++++++++++++++++++++++++++++++++++++++------
 6 files changed, 182 insertions(+), 19 deletions(-)

diff --git a/include/linux/kasan.h b/include/linux/kasan.h
index 1570c7191176..00a3bf7c0d8f 100644
--- a/include/linux/kasan.h
+++ b/include/linux/kasan.h
@@ -193,40 +193,44 @@ static __always_inline bool kasan_slab_pre_free(struct kmem_cache *s,
 {
 	if (kasan_enabled())
 		return __kasan_slab_pre_free(s, object, _RET_IP_);
 	return false;
 }
 
-bool __kasan_slab_free(struct kmem_cache *s, void *object, bool init);
+bool __kasan_slab_free(struct kmem_cache *s, void *object, bool init,
+		       bool still_accessible);
 /**
  * kasan_slab_free - Poison, initialize, and quarantine a slab object.
  * @object: Object to be freed.
  * @init: Whether to initialize the object.
+ * @still_accessible: Whether the object contents are still accessible.
  *
  * This function informs that a slab object has been freed and is not
- * supposed to be accessed anymore, except for objects in
- * SLAB_TYPESAFE_BY_RCU caches.
+ * supposed to be accessed anymore, except when @still_accessible is set
+ * (indicating that the object is in a SLAB_TYPESAFE_BY_RCU cache and an RCU
+ * grace period might not have passed yet).
  *
  * For KASAN modes that have integrated memory initialization
  * (kasan_has_integrated_init() == true), this function also initializes
  * the object's memory. For other modes, the @init argument is ignored.
  *
  * This function might also take ownership of the object to quarantine it.
  * When this happens, KASAN will defer freeing the object to a later
  * stage and handle it internally until then. The return value indicates
  * whether KASAN took ownership of the object.
  *
  * This function is intended only for use by the slab allocator.
  *
  * @Return true if KASAN took ownership of the object; false otherwise.
  */
 static __always_inline bool kasan_slab_free(struct kmem_cache *s,
-						void *object, bool init)
+						void *object, bool init,
+						bool still_accessible)
 {
 	if (kasan_enabled())
-		return __kasan_slab_free(s, object, init);
+		return __kasan_slab_free(s, object, init, still_accessible);
 	return false;
 }
 
 void __kasan_kfree_large(void *ptr, unsigned long ip);
 static __always_inline void kasan_kfree_large(void *ptr)
 {
@@ -416,13 +420,14 @@ static inline void *kasan_init_slab_obj(struct kmem_cache *cache,
 
 static inline bool kasan_slab_pre_free(struct kmem_cache *s, void *object)
 {
 	return false;
 }
 
-static inline bool kasan_slab_free(struct kmem_cache *s, void *object, bool init)
+static inline bool kasan_slab_free(struct kmem_cache *s, void *object,
+				   bool init, bool still_accessible)
 {
 	return false;
 }
 static inline void kasan_kfree_large(void *ptr) {}
 static inline void *kasan_slab_alloc(struct kmem_cache *s, void *object,
 				   gfp_t flags, bool init)
diff --git a/mm/Kconfig.debug b/mm/Kconfig.debug
index afc72fde0f03..41a58536531d 100644
--- a/mm/Kconfig.debug
+++ b/mm/Kconfig.debug
@@ -67,12 +67,44 @@ config SLUB_DEBUG_ON
 	  equivalent to specifying the "slab_debug" parameter on boot.
 	  There is no support for more fine grained debug control like
 	  possible with slab_debug=xxx. SLUB debugging may be switched
 	  off in a kernel built with CONFIG_SLUB_DEBUG_ON by specifying
 	  "slab_debug=-".
 
+config SLUB_RCU_DEBUG
+	bool "Enable UAF detection in TYPESAFE_BY_RCU caches (for KASAN)"
+	depends on SLUB_DEBUG
+	# SLUB_RCU_DEBUG should build fine without KASAN, but is currently useless
+	# without KASAN, so mark it as a dependency of KASAN for now.
+	depends on KASAN
+	default KASAN_GENERIC || KASAN_SW_TAGS
+	help
+	  Make SLAB_TYPESAFE_BY_RCU caches behave approximately as if the cache
+	  was not marked as SLAB_TYPESAFE_BY_RCU and every caller used
+	  kfree_rcu() instead.
+
+	  This is intended for use in combination with KASAN, to enable KASAN to
+	  detect use-after-free accesses in such caches.
+	  (KFENCE is able to do that independent of this flag.)
+
+	  This might degrade performance.
+	  Unfortunately this also prevents a very specific bug pattern from
+	  triggering (insufficient checks against an object being recycled
+	  within the RCU grace period); so this option can be turned off even on
+	  KASAN builds, in case you want to test for such a bug.
+
+	  If you're using this for testing bugs / fuzzing and care about
+	  catching all the bugs WAY more than performance, you might want to
+	  also turn on CONFIG_RCU_STRICT_GRACE_PERIOD.
+
+	  WARNING:
+	  This is designed as a debugging feature, not a security feature.
+	  Objects are sometimes recycled without RCU delay under memory pressure.
+
+	  If unsure, say N.
+
 config PAGE_OWNER
 	bool "Track page owner"
 	depends on DEBUG_KERNEL && STACKTRACE_SUPPORT
 	select DEBUG_FS
 	select STACKTRACE
 	select STACKDEPOT
diff --git a/mm/kasan/common.c b/mm/kasan/common.c
index f26bbc087b3b..ed4873e18c75 100644
--- a/mm/kasan/common.c
+++ b/mm/kasan/common.c
@@ -227,43 +227,44 @@ static bool check_slab_allocation(struct kmem_cache *cache, void *object,
 	}
 
 	return false;
 }
 
 static inline void poison_slab_object(struct kmem_cache *cache, void *object,
-				      bool init)
+				      bool init, bool still_accessible)
 {
 	void *tagged_object = object;
 
 	object = kasan_reset_tag(object);
 
 	/* RCU slabs could be legally used after free within the RCU period. */
-	if (unlikely(cache->flags & SLAB_TYPESAFE_BY_RCU))
+	if (unlikely(still_accessible))
 		return;
 
 	kasan_poison(object, round_up(cache->object_size, KASAN_GRANULE_SIZE),
 			KASAN_SLAB_FREE, init);
 
 	if (kasan_stack_collection_enabled())
 		kasan_save_free_info(cache, tagged_object);
 }
 
 bool __kasan_slab_pre_free(struct kmem_cache *cache, void *object,
 				unsigned long ip)
 {
 	if (!kasan_arch_is_ready() || is_kfence_address(object))
 		return false;
 	return check_slab_allocation(cache, object, ip);
 }
 
-bool __kasan_slab_free(struct kmem_cache *cache, void *object, bool init)
+bool __kasan_slab_free(struct kmem_cache *cache, void *object, bool init,
+		       bool still_accessible)
 {
 	if (!kasan_arch_is_ready() || is_kfence_address(object))
 		return false;
 
-	poison_slab_object(cache, object, init);
+	poison_slab_object(cache, object, init, still_accessible);
 
 	/*
 	 * If the object is put into quarantine, do not let slab put the object
 	 * onto the freelist for now. The object's metadata is kept until the
 	 * object gets evicted from quarantine.
 	 */
@@ -515,13 +516,13 @@ bool __kasan_mempool_poison_object(void *ptr, unsigned long ip)
 
 	slab = folio_slab(folio);
 
 	if (check_slab_allocation(slab->slab_cache, ptr, ip))
 		return false;
 
-	poison_slab_object(slab->slab_cache, ptr, false);
+	poison_slab_object(slab->slab_cache, ptr, false, false);
 	return true;
 }
 
 void __kasan_mempool_unpoison_object(void *ptr, size_t size, unsigned long ip)
 {
 	struct slab *slab;
diff --git a/mm/kasan/kasan_test.c b/mm/kasan/kasan_test.c
index 7b32be2a3cf0..567d33b493e2 100644
--- a/mm/kasan/kasan_test.c
+++ b/mm/kasan/kasan_test.c
@@ -993,12 +993,57 @@ static void kmem_cache_invalid_free(struct kunit *test)
 	 */
 	kmem_cache_free(cache, p);
 
 	kmem_cache_destroy(cache);
 }
 
+static void kmem_cache_rcu_uaf(struct kunit *test)
+{
+	char *p;
+	size_t size = 200;
+	struct kmem_cache *cache;
+
+	KASAN_TEST_NEEDS_CONFIG_ON(test, CONFIG_SLUB_RCU_DEBUG);
+
+	cache = kmem_cache_create("test_cache", size, 0, SLAB_TYPESAFE_BY_RCU,
+				  NULL);
+	KUNIT_ASSERT_NOT_ERR_OR_NULL(test, cache);
+
+	p = kmem_cache_alloc(cache, GFP_KERNEL);
+	if (!p) {
+		kunit_err(test, "Allocation failed: %s\n", __func__);
+		kmem_cache_destroy(cache);
+		return;
+	}
+	*p = 1;
+
+	rcu_read_lock();
+
+	/* Free the object - this will internally schedule an RCU callback. */
+	kmem_cache_free(cache, p);
+
+	/*
+	 * We should still be allowed to access the object at this point because
+	 * the cache is SLAB_TYPESAFE_BY_RCU and we've been in an RCU read-side
+	 * critical section since before the kmem_cache_free().
+	 */
+	READ_ONCE(*p);
+
+	rcu_read_unlock();
+
+	/*
+	 * Wait for the RCU callback to execute; after this, the object should
+	 * have actually been freed from KASAN's perspective.
+	 */
+	rcu_barrier();
+
+	KUNIT_EXPECT_KASAN_FAIL(test, READ_ONCE(*p));
+
+	kmem_cache_destroy(cache);
+}
+
 static void empty_cache_ctor(void *object) { }
 
 static void kmem_cache_double_destroy(struct kunit *test)
 {
 	struct kmem_cache *cache;
 
@@ -1934,12 +1979,13 @@ static struct kunit_case kasan_kunit_test_cases[] = {
 	KUNIT_CASE(workqueue_uaf),
 	KUNIT_CASE(kfree_via_page),
 	KUNIT_CASE(kfree_via_phys),
 	KUNIT_CASE(kmem_cache_oob),
 	KUNIT_CASE(kmem_cache_double_free),
 	KUNIT_CASE(kmem_cache_invalid_free),
+	KUNIT_CASE(kmem_cache_rcu_uaf),
 	KUNIT_CASE(kmem_cache_double_destroy),
 	KUNIT_CASE(kmem_cache_accounted),
 	KUNIT_CASE(kmem_cache_bulk),
 	KUNIT_CASE(mempool_kmalloc_oob_right),
 	KUNIT_CASE(mempool_kmalloc_large_oob_right),
 	KUNIT_CASE(mempool_slab_oob_right),
diff --git a/mm/slab_common.c b/mm/slab_common.c
index 40b582a014b8..9025e85c6750 100644
--- a/mm/slab_common.c
+++ b/mm/slab_common.c
@@ -573,12 +573,28 @@ void kmem_cache_destroy(struct kmem_cache *s)
 	int err = -EBUSY;
 	bool rcu_set;
 
 	if (unlikely(!s) || !kasan_check_byte(s))
 		return;
 
+	if (IS_ENABLED(CONFIG_SLUB_RCU_DEBUG) &&
+	    (s->flags & SLAB_TYPESAFE_BY_RCU)) {
+		/*
+		 * Under CONFIG_SLUB_RCU_DEBUG, when objects in a
+		 * SLAB_TYPESAFE_BY_RCU slab are freed, SLUB will internally
+		 * defer their freeing with call_rcu().
+		 * Wait for such call_rcu() invocations here before actually
+		 * destroying the cache.
+		 *
+		 * It doesn't matter that we haven't looked at the slab refcount
+		 * yet - slabs with SLAB_TYPESAFE_BY_RCU can't be merged, so
+		 * the refcount should be 1 here.
+		 */
+		rcu_barrier();
+	}
+
 	cpus_read_lock();
 	mutex_lock(&slab_mutex);
 
 	rcu_set = s->flags & SLAB_TYPESAFE_BY_RCU;
 
 	s->refcount--;
diff --git a/mm/slub.c b/mm/slub.c
index 0c98b6a2124f..86ab9477a1ae 100644
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -2197,45 +2197,81 @@ static inline bool memcg_slab_post_alloc_hook(struct kmem_cache *s,
 static inline void memcg_slab_free_hook(struct kmem_cache *s, struct slab *slab,
 					void **p, int objects)
 {
 }
 #endif /* CONFIG_MEMCG */
 
+#ifdef CONFIG_SLUB_RCU_DEBUG
+static void slab_free_after_rcu_debug(struct rcu_head *rcu_head);
+
+struct rcu_delayed_free {
+	struct rcu_head head;
+	void *object;
+};
+#endif
+
 /*
  * Hooks for other subsystems that check memory allocations. In a typical
  * production configuration these hooks all should produce no code at all.
  *
  * Returns true if freeing of the object can proceed, false if its reuse
- * was delayed by KASAN quarantine, or it was returned to KFENCE.
+ * was delayed by CONFIG_SLUB_RCU_DEBUG or KASAN quarantine, or it was returned
+ * to KFENCE.
  */
 static __always_inline
-bool slab_free_hook(struct kmem_cache *s, void *x, bool init)
+bool slab_free_hook(struct kmem_cache *s, void *x, bool init,
+		    bool after_rcu_delay)
 {
+	/* Are the object contents still accessible? */
+	bool still_accessible = (s->flags & SLAB_TYPESAFE_BY_RCU) && !after_rcu_delay;
+
 	kmemleak_free_recursive(x, s->flags);
 	kmsan_slab_free(s, x);
 
 	debug_check_no_locks_freed(x, s->object_size);
 
 	if (!(s->flags & SLAB_DEBUG_OBJECTS))
 		debug_check_no_obj_freed(x, s->object_size);
 
 	/* Use KCSAN to help debug racy use-after-free. */
-	if (!(s->flags & SLAB_TYPESAFE_BY_RCU))
+	if (!still_accessible)
 		__kcsan_check_access(x, s->object_size,
 				     KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT);
 
 	if (kfence_free(x))
 		return false;
 
 	/*
 	 * Give KASAN a chance to notice an invalid free operation before we
 	 * modify the object.
 	 */
 	if (kasan_slab_pre_free(s, x))
 		return false;
 
+#ifdef CONFIG_SLUB_RCU_DEBUG
+	if (still_accessible) {
+		struct rcu_delayed_free *delayed_free;
+
+		delayed_free = kmalloc(sizeof(*delayed_free), GFP_NOWAIT);
+		if (delayed_free) {
+			/*
+			 * Let KASAN track our call stack as a "related work
+			 * creation", just like if the object had been freed
+			 * normally via kfree_rcu().
+			 * We have to do this manually because the rcu_head is
+			 * not located inside the object.
+			 */
+			kasan_record_aux_stack_noalloc(x);
+
+			delayed_free->object = x;
+			call_rcu(&delayed_free->head, slab_free_after_rcu_debug);
+			return false;
+		}
+	}
+#endif /* CONFIG_SLUB_RCU_DEBUG */
+
 	/*
 	 * As memory initialization might be integrated into KASAN,
 	 * kasan_slab_free and initialization memset's must be
 	 * kept together to avoid discrepancies in behavior.
 	 *
 	 * The initialization memset's clear the object and the metadata,
@@ -2253,42 +2289,42 @@ bool slab_free_hook(struct kmem_cache *s, void *x, bool init)
 			memset(kasan_reset_tag(x), 0, s->object_size);
 		rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0;
 		memset((char *)kasan_reset_tag(x) + inuse, 0,
 		       s->size - inuse - rsize);
 	}
 	/* KASAN might put x into memory quarantine, delaying its reuse. */
-	return !kasan_slab_free(s, x, init);
+	return !kasan_slab_free(s, x, init, still_accessible);
 }
 
 static __fastpath_inline
 bool slab_free_freelist_hook(struct kmem_cache *s, void **head, void **tail,
 			     int *cnt)
 {
 
 	void *object;
 	void *next = *head;
 	void *old_tail = *tail;
 	bool init;
 
 	if (is_kfence_address(next)) {
-		slab_free_hook(s, next, false);
+		slab_free_hook(s, next, false, false);
 		return false;
 	}
 
 	/* Head and tail of the reconstructed freelist */
 	*head = NULL;
 	*tail = NULL;
 
 	init = slab_want_init_on_free(s);
 
 	do {
 		object = next;
 		next = get_freepointer(s, object);
 
 		/* If object's reuse doesn't have to be delayed */
-		if (likely(slab_free_hook(s, object, init))) {
+		if (likely(slab_free_hook(s, object, init, false))) {
 			/* Move object to the new freelist */
 			set_freepointer(s, object, *head);
 			*head = object;
 			if (!*tail)
 				*tail = object;
 		} else {
@@ -4474,40 +4510,67 @@ static __fastpath_inline
 void slab_free(struct kmem_cache *s, struct slab *slab, void *object,
 	       unsigned long addr)
 {
 	memcg_slab_free_hook(s, slab, &object, 1);
 	alloc_tagging_slab_free_hook(s, slab, &object, 1);
 
-	if (likely(slab_free_hook(s, object, slab_want_init_on_free(s))))
+	if (likely(slab_free_hook(s, object, slab_want_init_on_free(s), false)))
 		do_slab_free(s, slab, object, object, 1, addr);
 }
 
 #ifdef CONFIG_MEMCG
 /* Do not inline the rare memcg charging failed path into the allocation path */
 static noinline
 void memcg_alloc_abort_single(struct kmem_cache *s, void *object)
 {
-	if (likely(slab_free_hook(s, object, slab_want_init_on_free(s))))
+	if (likely(slab_free_hook(s, object, slab_want_init_on_free(s), false)))
 		do_slab_free(s, virt_to_slab(object), object, object, 1, _RET_IP_);
 }
 #endif
 
 static __fastpath_inline
 void slab_free_bulk(struct kmem_cache *s, struct slab *slab, void *head,
 		    void *tail, void **p, int cnt, unsigned long addr)
 {
 	memcg_slab_free_hook(s, slab, p, cnt);
 	alloc_tagging_slab_free_hook(s, slab, p, cnt);
 	/*
 	 * With KASAN enabled slab_free_freelist_hook modifies the freelist
 	 * to remove objects, whose reuse must be delayed.
 	 */
 	if (likely(slab_free_freelist_hook(s, &head, &tail, &cnt)))
 		do_slab_free(s, slab, head, tail, cnt, addr);
 }
 
+#ifdef CONFIG_SLUB_RCU_DEBUG
+static void slab_free_after_rcu_debug(struct rcu_head *rcu_head)
+{
+	struct rcu_delayed_free *delayed_free =
+			container_of(rcu_head, struct rcu_delayed_free, head);
+	void *object = delayed_free->object;
+	struct slab *slab = virt_to_slab(object);
+	struct kmem_cache *s;
+
+	kfree(delayed_free);
+
+	if (WARN_ON(is_kfence_address(object)))
+		return;
+
+	/* find the object and the cache again */
+	if (WARN_ON(!slab))
+		return;
+	s = slab->slab_cache;
+	if (WARN_ON(!(s->flags & SLAB_TYPESAFE_BY_RCU)))
+		return;
+
+	/* resume freeing */
+	if (slab_free_hook(s, object, slab_want_init_on_free(s), true))
+		do_slab_free(s, slab, object, object, 1, _THIS_IP_);
+}
+#endif /* CONFIG_SLUB_RCU_DEBUG */
+
 #ifdef CONFIG_KASAN_GENERIC
 void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
 {
 	do_slab_free(cache, virt_to_slab(x), x, x, 1, addr);
 }
 #endif

-- 
2.46.0.76.ge559c4bf1a-goog

Re: [PATCH v8 0/2] allow KASAN to detect UAF in SLAB_TYPESAFE_BY_RCU slabs

[David Sterba]

On Fri, Aug 09, 2024 at 05:36:54PM +0200, Jann Horn wrote:
> Changes in v8:
> - in patch 2/2:
>   - move rcu_barrier() out of locked region (vbabka)
>   - rearrange code in slab_free_after_rcu_debug (vbabka)
> - Link to v7: https://lore.kernel.org/r/20240808-kasan-tsbrcu-v7-0-0d0590c54ae6@google.com
> 
> Changes in v7:
> - in patch 2/2:
>   - clarify kconfig comment (Marco)
>   - fix memory leak (vbabka and dsterba)

FWIW, I've retested v7 and got no OOM (caused by the leak), on KASAN,
KFENCE and other debugging options enabled.

Re: [PATCH v8 2/2] slub: Introduce CONFIG_SLUB_RCU_DEBUG

[Shinichiro Kawasaki]

Hello Jann, let me ask a question about this patch. When I tested the
next-20240808 kernel which includes this patch, I observed that
slab_free_after_rcu_debug() reports many WARNs. Please find my question in line.

On Aug 09, 2024 / 17:36, Jann Horn wrote:
> Currently, KASAN is unable to catch use-after-free in SLAB_TYPESAFE_BY_RCU
> slabs because use-after-free is allowed within the RCU grace period by
> design.
> 
> Add a SLUB debugging feature which RCU-delays every individual
> kmem_cache_free() before either actually freeing the object or handing it
> off to KASAN, and change KASAN to poison freed objects as normal when this
> option is enabled.
> 
> For now I've configured Kconfig.debug to default-enable this feature in the
> KASAN GENERIC and SW_TAGS modes; I'm not enabling it by default in HW_TAGS
> mode because I'm not sure if it might have unwanted performance degradation
> effects there.
> 
> Note that this is mostly useful with KASAN in the quarantine-based GENERIC
> mode; SLAB_TYPESAFE_BY_RCU slabs are basically always also slabs with a
> ->ctor, and KASAN's assign_tag() currently has to assign fixed tags for
> those, reducing the effectiveness of SW_TAGS/HW_TAGS mode.
> (A possible future extension of this work would be to also let SLUB call
> the ->ctor() on every allocation instead of only when the slab page is
> allocated; then tag-based modes would be able to assign new tags on every
> reallocation.)

[...]

> diff --git a/mm/Kconfig.debug b/mm/Kconfig.debug
> index afc72fde0f03..41a58536531d 100644
> --- a/mm/Kconfig.debug
> +++ b/mm/Kconfig.debug
> @@ -67,12 +67,44 @@ config SLUB_DEBUG_ON
>  	  equivalent to specifying the "slab_debug" parameter on boot.
>  	  There is no support for more fine grained debug control like
>  	  possible with slab_debug=xxx. SLUB debugging may be switched
>  	  off in a kernel built with CONFIG_SLUB_DEBUG_ON by specifying
>  	  "slab_debug=-".
>  
> +config SLUB_RCU_DEBUG
> +	bool "Enable UAF detection in TYPESAFE_BY_RCU caches (for KASAN)"
> +	depends on SLUB_DEBUG
> +	# SLUB_RCU_DEBUG should build fine without KASAN, but is currently useless
> +	# without KASAN, so mark it as a dependency of KASAN for now.
> +	depends on KASAN
> +	default KASAN_GENERIC || KASAN_SW_TAGS

When I tested the next-20240808 kernel which includes this patch, I saw the
SLUB_RCU_DEBUG was enabled because I enable KASAN_GENERIC and KASAN_SW_TAGS
for my test target kernels. I also enable KFENCE.

[...]

> +#ifdef CONFIG_SLUB_RCU_DEBUG
> +static void slab_free_after_rcu_debug(struct rcu_head *rcu_head)
> +{
> +	struct rcu_delayed_free *delayed_free =
> +			container_of(rcu_head, struct rcu_delayed_free, head);
> +	void *object = delayed_free->object;
> +	struct slab *slab = virt_to_slab(object);
> +	struct kmem_cache *s;
> +
> +	kfree(delayed_free);
> +
> +	if (WARN_ON(is_kfence_address(object)))
> +		return;

With the kernel configs above, I see the many WARNs are reported here.
When SLUB_RCU_DEBUG is enabled, should I disable KFENCE?

Re: [PATCH v8 2/2] slub: Introduce CONFIG_SLUB_RCU_DEBUG

[Jann Horn]

Hi!

On Tue, Aug 13, 2024 at 11:03 AM Shinichiro Kawasaki
<shinichiro.kawasaki@wdc.com> wrote:
> Hello Jann, let me ask a question about this patch. When I tested the
> next-20240808 kernel which includes this patch, I observed that
> slab_free_after_rcu_debug() reports many WARNs. Please find my question in line.

Thanks for testing linux-next.

> On Aug 09, 2024 / 17:36, Jann Horn wrote:
[...]
> > +#ifdef CONFIG_SLUB_RCU_DEBUG
> > +static void slab_free_after_rcu_debug(struct rcu_head *rcu_head)
> > +{
> > +     struct rcu_delayed_free *delayed_free =
> > +                     container_of(rcu_head, struct rcu_delayed_free, head);
> > +     void *object = delayed_free->object;
> > +     struct slab *slab = virt_to_slab(object);
> > +     struct kmem_cache *s;
> > +
> > +     kfree(delayed_free);
> > +
> > +     if (WARN_ON(is_kfence_address(object)))
> > +             return;
>
> With the kernel configs above, I see the many WARNs are reported here.
> When SLUB_RCU_DEBUG is enabled, should I disable KFENCE?

These features are supposed to be compatible.

In the version you tested
(https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next-history.git/tree/mm/slub.c?h=next-20240808#n4550),
I made a mistake and wrote "if (WARN_ON(is_kfence_address(rcu_head)))"
instead of "if (WARN_ON(is_kfence_address(object)))". That issue was
fixed in v6 of the series after syzbot and the Intel test bot ran into
the same issue.

Re: [PATCH v8 2/2] slub: Introduce CONFIG_SLUB_RCU_DEBUG

[Shinichiro Kawasaki]

On Aug 13, 2024 / 16:35, Jann Horn wrote:
[...]
> In the version you tested
> (https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next-history.git/tree/mm/slub.c?h=next-20240808#n4550),
> I made a mistake and wrote "if (WARN_ON(is_kfence_address(rcu_head)))"
> instead of "if (WARN_ON(is_kfence_address(object)))". That issue was
> fixed in v6 of the series after syzbot and the Intel test bot ran into
> the same issue.

Ah, I overlooked that 'rcu_head' has changed to 'object'... I should have
checked the patch change history. Good to know that the WARN is already fixed.
Thank you for the clarification.