mm/kmemleak: annotate a data race in checksum

PATCH In review Qian Cai 2 participants 3 messages Mar 17, 2020

Subsystems (2)

Memory management Kmemleak

lore

Share & tools

lore

Qian Cai

author

2020-03-17 13:22 UTC #0 lore raw

PATCH 1/1 mm/kmemleak: annotate a data race in checksum

Even if KCSAN is disabled for kmemleak, update_checksum() could still call crc32() (which is outside of kmemleak.c) to dereference object->pointer. Thus, the value of object->pointer could be accessed concurrently as noticed by KCSAN,

 BUG: KCSAN: data-race in crc32_le_base / do_raw_spin_lock

write to 0xffffb0ea683a7d50 of 4 bytes by task 23575 on cpu 12:

  do_raw_spin_lock+0x114/0x200
  debug_spin_lock_after at kernel/locking/spinlock_debug.c:91
  (inlined by) do_raw_spin_lock at kernel/locking/spinlock_debug.c:115
  _raw_spin_lock+0x40/0x50
  __handle_mm_fault+0xa9e/0xd00
  handle_mm_fault+0xfc/0x2f0
  do_page_fault+0x263/0x6f9
  page_fault+0x34/0x40

read to 0xffffb0ea683a7d50 of 4 bytes by task 839 on cpu 60:

  crc32_le_base+0x67/0x350
  crc32_le_base+0x67/0x350:
  crc32_body at lib/crc32.c:106
  (inlined by) crc32_le_generic at lib/crc32.c:179
  (inlined by) crc32_le at lib/crc32.c:197
  kmemleak_scan+0x528/0xd90
  update_checksum at mm/kmemleak.c:1172
  (inlined by) kmemleak_scan at mm/kmemleak.c:1497
  kmemleak_scan_thread+0xcc/0xfa
  kthread+0x1e0/0x200
  ret_from_fork+0x27/0x50

If a shattered value was returned due to a data race, it will be corrected in the next scan. Thus, annotate it as an intentional data race using the data_race() macro.

Signed-off-by Qian Cai <[email protected]>

---
 mm/kmemleak.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

mm/kmemleak.c +6 -1

index e362dc3d2028..d3327756c3a4 100644
@@ -1169,7 +1169,12 @@ static bool update_checksum(struct kmemleak_object *object)
 	u32 old_csum = object->checksum;
 
 	kasan_disable_current();
-	object->checksum = crc32(0, (void *)object->pointer, object->size);
+	/*
+	 * crc32() will dereference object->pointer. If an unstable value was
+	 * returned due to a data race, it will be corrected in the next scan.
+	 */
+	object->checksum = data_race(crc32(0, (void *)object->pointer,
+					   object->size));
 	kasan_enable_current();
 
 	return object->checksum != old_csum;

Signature · 1 line

2.21.0 (Apple Git-122.2)

Marco Elver

2020-03-17 13:31 UTC #1 lore raw

Quoted 45 lines · Qian Cai

Even if KCSAN is disabled for kmemleak, update_checksum() could still call crc32() (which is outside of kmemleak.c) to dereference object->pointer. Thus, the value of object->pointer could be accessed concurrently as noticed by KCSAN,
 BUG: KCSAN: data-race in crc32_le_base / do_raw_spin_lock
write to 0xffffb0ea683a7d50 of 4 bytes by task 23575 on cpu 12:
  do_raw_spin_lock+0x114/0x200
  debug_spin_lock_after at kernel/locking/spinlock_debug.c:91
  (inlined by) do_raw_spin_lock at kernel/locking/spinlock_debug.c:115
  _raw_spin_lock+0x40/0x50
  __handle_mm_fault+0xa9e/0xd00
  handle_mm_fault+0xfc/0x2f0
  do_page_fault+0x263/0x6f9
  page_fault+0x34/0x40
read to 0xffffb0ea683a7d50 of 4 bytes by task 839 on cpu 60:
  crc32_le_base+0x67/0x350
  crc32_le_base+0x67/0x350:
  crc32_body at lib/crc32.c:106
  (inlined by) crc32_le_generic at lib/crc32.c:179
  (inlined by) crc32_le at lib/crc32.c:197
  kmemleak_scan+0x528/0xd90
  update_checksum at mm/kmemleak.c:1172
  (inlined by) kmemleak_scan at mm/kmemleak.c:1497
  kmemleak_scan_thread+0xcc/0xfa
  kthread+0x1e0/0x200
  ret_from_fork+0x27/0x50
If a shattered value was returned due to a data race, it will be corrected in the next scan. Thus, annotate it as an intentional data race using the data_race() macro.
Signed-off-by Qian Cai <[email protected]>
---
 mm/kmemleak.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
mm/kmemleak.c +0 -0
index e362dc3d2028..d3327756c3a4 100644

mm/kmemleak.c static bool update_checksum(struct kmemleak_object *object) · line 1171

Quoted 3 lines

mm/kmemleak.c +0 -0

@@ -1169,7 +1169,12 @@ static bool update_checksum(struct kmemleak_object *object)
        u32 old_csum = object->checksum;
 
        kasan_disable_current();

Suggested:
+ kcsan_disable_current();

mm/kmemleak.c static bool update_checksum(struct kmemleak_object *object) · line 1177

Quoted 7 lines

mm/kmemleak.c +6 -1

-       object->checksum = crc32(0, (void *)object->pointer, object->size);
+       /*
+        * crc32() will dereference object->pointer. If an unstable value was
+        * returned due to a data race, it will be corrected in the next scan.
+        */
+       object->checksum = data_race(crc32(0, (void *)object->pointer,
+                                          object->size));

This will work with the default config, because for word-sized-aligned writes no marking is enforced. But this will still cause a data race if the write is e.g. due to a memcpy.

There are already markers for KASAN around, so the most reliable thing is to just disable KCSAN in this region.

mm/kmemleak.c static bool update_checksum(struct kmemleak_object *object) · line 1178

Quoted 1 line

mm/kmemleak.c +0 -0

        kasan_enable_current();

Suggested:
+ kcsan_enable_current();

Thanks,
-- Marco

Quoted 4 lines

mm/kmemleak.c +0 -1
        return object->checksum != old_csum;
--
2.21.0 (Apple Git-122.2)

Qian Cai

author

2020-03-17 13:42 UTC #2 lore raw

Quoted 65 lines · Marco Elver

Quoted 45 lines · Qian Cai
Even if KCSAN is disabled for kmemleak, update_checksum() could still call crc32() (which is outside of kmemleak.c) to dereference object->pointer. Thus, the value of object->pointer could be accessed concurrently as noticed by KCSAN,
BUG: KCSAN: data-race in crc32_le_base / do_raw_spin_lock
write to 0xffffb0ea683a7d50 of 4 bytes by task 23575 on cpu 12:
 do_raw_spin_lock+0x114/0x200
 debug_spin_lock_after at kernel/locking/spinlock_debug.c:91
 (inlined by) do_raw_spin_lock at kernel/locking/spinlock_debug.c:115
 _raw_spin_lock+0x40/0x50
 __handle_mm_fault+0xa9e/0xd00
 handle_mm_fault+0xfc/0x2f0
 do_page_fault+0x263/0x6f9
 page_fault+0x34/0x40
read to 0xffffb0ea683a7d50 of 4 bytes by task 839 on cpu 60:
 crc32_le_base+0x67/0x350
 crc32_le_base+0x67/0x350:
 crc32_body at lib/crc32.c:106
 (inlined by) crc32_le_generic at lib/crc32.c:179
 (inlined by) crc32_le at lib/crc32.c:197
 kmemleak_scan+0x528/0xd90
 update_checksum at mm/kmemleak.c:1172
 (inlined by) kmemleak_scan at mm/kmemleak.c:1497
 kmemleak_scan_thread+0xcc/0xfa
 kthread+0x1e0/0x200
 ret_from_fork+0x27/0x50
If a shattered value was returned due to a data race, it will be corrected in the next scan. Thus, annotate it as an intentional data race using the data_race() macro.
Signed-off-by Qian Cai <[email protected]>
---
mm/kmemleak.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
mm/kmemleak.c +0 -0
index e362dc3d2028..d3327756c3a4 100644
mm/kmemleak.c static bool update_checksum(struct kmemleak_object *object) · line 1171
Quoted 3 lines
mm/kmemleak.c +0 -0
@@ -1169,7 +1169,12 @@ static bool update_checksum(struct kmemleak_object *object)
       u32 old_csum = object->checksum;
 
       kasan_disable_current();
Suggested:
+ kcsan_disable_current();
mm/kmemleak.c static bool update_checksum(struct kmemleak_object *object) · line 1177
Quoted 7 lines
mm/kmemleak.c +6 -1
-       object->checksum = crc32(0, (void *)object->pointer, object->size);
+       /*
+        * crc32() will dereference object->pointer. If an unstable value was
+        * returned due to a data race, it will be corrected in the next scan.
+        */
+       object->checksum = data_race(crc32(0, (void *)object->pointer,
+                                          object->size));

Quoted 3 lines

This will work with the default config, because for word-sized-aligned writes no marking is enforced. But this will still cause a data race if the write is e.g. due to a memcpy.

I saw this spla atmt but just decided to reuse an old one to save some time.

Looks like that "head->func = func;” not aligned.

[77392.095571][  T839] BUG: KCSAN: data-race in call_rcu / crc32_le_base
[77392.102066][  T839] 
[77392.104297][  T839] write to 0xffff898ea73a8748 of 8 bytes by task 114682 on cpu 79:
[77392.112111][  T839]  call_rcu+0xe8/0x4b0
__call_rcu at kernel/rcu/tree.c:2701
(inlined by) call_rcu at kernel/rcu/tree.c:2777
[77392.116084][  T839]  __fput+0x23a/0x3d0
[77392.119970][  T839]  ____fput+0x1e/0x30
[77392.123852][  T839]  task_work_run+0xba/0x120
[77392.128257][  T839]  do_syscall_64+0x7d7/0xb05
[77392.132753][  T839]  entry_SYSCALL_64_after_hwframe+0x49/0xb3
[77392.138544][  T839] 
[77392.140760][  T839] INFO: lockdep is turned off.
[77392.145478][  T839] irq event stamp: 0
[77392.149270][  T839] hardirqs last  enabled at (0): [<0000000000000000>] 0x0
[77392.156307][  T839] hardirqs last disabled at (0): [<ffffffffb0ab4d42>] copy_process+0x1122/0x3240
[77392.165348][  T839] softirqs last  enabled at (0): [<ffffffffb0ab4d42>] copy_process+0x1122/0x3240
[77392.174384][  T839] softirqs last disabled at (0): [<0000000000000000>] 0x0
[77392.181405][  T839] 
[77392.183625][  T839] read to 0xffff898ea73a8748 of 4 bytes by task 839 on cpu 46:
[77392.191088][  T839]  crc32_le_base+0x67/0x350
[77392.195498][  T839]  kmemleak_scan+0x3ee/0x9f0
[77392.199992][  T839]  kmemleak_scan_thread+0x9f/0xc4
[77392.204921][  T839]  kthread+0x1cd/0x1f0
[77392.208894][  T839]  ret_from_fork+0x27/0x50

Quoted 3 lines

There are already markers for KASAN around, so the most reliable thing is to just disable KCSAN in this region.

OK, I’ll test that a bit first.

Quoted 12 lines

mm/kmemleak.c static bool update_checksum(struct kmemleak_object *object) · line 1178
Quoted 2 lines
mm/kmemleak.c +0 -0
       kasan_enable_current();
Suggested:
+ kcsan_enable_current();
Thanks,
-- Marco
Quoted 3 lines
mm/kmemleak.c +0 -1
       return object->checksum != old_csum;
--
2.21.0 (Apple Git-122.2)