mm/memcontrol: fix a data race in scan count
authorQian Cai <cai@lca.pw>
Sat, 15 Aug 2020 00:31:37 +0000 (17:31 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 15 Aug 2020 02:56:57 +0000 (19:56 -0700)
struct mem_cgroup_per_node mz.lru_zone_size[zone_idx][lru] could be
accessed concurrently as noticed by KCSAN,

 BUG: KCSAN: data-race in lruvec_lru_size / mem_cgroup_update_lru_size

 write to 0xffff9c804ca285f8 of 8 bytes by task 50951 on cpu 12:
  mem_cgroup_update_lru_size+0x11c/0x1d0
  mem_cgroup_update_lru_size at mm/memcontrol.c:1266
  isolate_lru_pages+0x6a9/0xf30
  shrink_active_list+0x123/0xcc0
  shrink_lruvec+0x8fd/0x1380
  shrink_node+0x317/0xd80
  do_try_to_free_pages+0x1f7/0xa10
  try_to_free_pages+0x26c/0x5e0
  __alloc_pages_slowpath+0x458/0x1290
  __alloc_pages_nodemask+0x3bb/0x450
  alloc_pages_vma+0x8a/0x2c0
  do_anonymous_page+0x170/0x700
  __handle_mm_fault+0xc9f/0xd00
  handle_mm_fault+0xfc/0x2f0
  do_page_fault+0x263/0x6f9
  page_fault+0x34/0x40

 read to 0xffff9c804ca285f8 of 8 bytes by task 50964 on cpu 95:
  lruvec_lru_size+0xbb/0x270
  mem_cgroup_get_zone_lru_size at include/linux/memcontrol.h:536
  (inlined by) lruvec_lru_size at mm/vmscan.c:326
  shrink_lruvec+0x1d0/0x1380
  shrink_node+0x317/0xd80
  do_try_to_free_pages+0x1f7/0xa10
  try_to_free_pages+0x26c/0x5e0
  __alloc_pages_slowpath+0x458/0x1290
  __alloc_pages_nodemask+0x3bb/0x450
  alloc_pages_current+0xa6/0x120
  alloc_slab_page+0x3b1/0x540
  allocate_slab+0x70/0x660
  new_slab+0x46/0x70
  ___slab_alloc+0x4ad/0x7d0
  __slab_alloc+0x43/0x70
  kmem_cache_alloc+0x2c3/0x420
  getname_flags+0x4c/0x230
  getname+0x22/0x30
  do_sys_openat2+0x205/0x3b0
  do_sys_open+0x9a/0xf0
  __x64_sys_openat+0x62/0x80
  do_syscall_64+0x91/0xb47
  entry_SYSCALL_64_after_hwframe+0x49/0xbe

 Reported by Kernel Concurrency Sanitizer on:
 CPU: 95 PID: 50964 Comm: cc1 Tainted: G        W  O L    5.5.0-next-20200204+ #6
 Hardware name: HPE ProLiant DL385 Gen10/ProLiant DL385 Gen10, BIOS A40 07/10/2019

The write is under lru_lock, but the read is done as lockless.  The scan
count is used to determine how aggressively the anon and file LRU lists
should be scanned.  Load tearing could generate an inefficient heuristic,
so fix it by adding READ_ONCE() for the read.

Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Link: http://lkml.kernel.org/r/20200206034945.2481-1-cai@lca.pw
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/memcontrol.h

index 385237e..d0b0361 100644 (file)
@@ -630,7 +630,7 @@ unsigned long mem_cgroup_get_zone_lru_size(struct lruvec *lruvec,
        struct mem_cgroup_per_node *mz;
 
        mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec);
-       return mz->lru_zone_size[zone_idx][lru];
+       return READ_ONCE(mz->lru_zone_size[zone_idx][lru]);
 }
 
 void mem_cgroup_handle_over_high(void);