ksm: fix conflict between mmput and scan_get_next_rmap_item
authorZhou Chengming <zhouchengming1@huawei.com>
Thu, 12 May 2016 22:42:21 +0000 (15:42 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 12 May 2016 22:52:50 +0000 (15:52 -0700)
A concurrency issue about KSM in the function scan_get_next_rmap_item.

task A (ksmd): |task B (the mm's task):
|
mm = slot->mm; |
down_read(&mm->mmap_sem); |
|
... |
|
spin_lock(&ksm_mmlist_lock); |
|
ksm_scan.mm_slot go to the next slot; |
|
spin_unlock(&ksm_mmlist_lock); |
|mmput() ->
| ksm_exit():
|
|spin_lock(&ksm_mmlist_lock);
|if (mm_slot && ksm_scan.mm_slot != mm_slot) {
| if (!mm_slot->rmap_list) {
| easy_to_free = 1;
| ...
|
|if (easy_to_free) {
| mmdrop(mm);
| ...
|
|So this mm_struct may be freed in the mmput().
|
up_read(&mm->mmap_sem); |

As we can see above, the ksmd thread may access a mm_struct that already
been freed to the kmem_cache.  Suppose a fork will get this mm_struct from
the kmem_cache, the ksmd thread then call up_read(&mm->mmap_sem), will
cause mmap_sem.count to become -1.

As suggested by Andrea Arcangeli, unmerge_and_remove_all_rmap_items has
the same SMP race condition, so fix it too.  My prev fix in function
scan_get_next_rmap_item will introduce a different SMP race condition, so
just invert the up_read/spin_unlock order as Andrea Arcangeli said.

Link: http://lkml.kernel.org/r/1462708815-31301-1-git-send-email-zhouchengming1@huawei.com
Signed-off-by: Zhou Chengming <zhouchengming1@huawei.com>
Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Geliang Tang <geliangtang@163.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: Ding Tianhong <dingtianhong@huawei.com>
Cc: Li Bin <huawei.libin@huawei.com>
Cc: Zhen Lei <thunder.leizhen@huawei.com>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/ksm.c

index b99e828..4786b41 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -783,6 +783,7 @@ static int unmerge_and_remove_all_rmap_items(void)
                }
 
                remove_trailing_rmap_items(mm_slot, &mm_slot->rmap_list);
+               up_read(&mm->mmap_sem);
 
                spin_lock(&ksm_mmlist_lock);
                ksm_scan.mm_slot = list_entry(mm_slot->mm_list.next,
@@ -794,12 +795,9 @@ static int unmerge_and_remove_all_rmap_items(void)
 
                        free_mm_slot(mm_slot);
                        clear_bit(MMF_VM_MERGEABLE, &mm->flags);
-                       up_read(&mm->mmap_sem);
                        mmdrop(mm);
-               } else {
+               } else
                        spin_unlock(&ksm_mmlist_lock);
-                       up_read(&mm->mmap_sem);
-               }
        }
 
        /* Clean up stable nodes, but don't worry if some are still busy */
@@ -1663,8 +1661,15 @@ next_mm:
                up_read(&mm->mmap_sem);
                mmdrop(mm);
        } else {
-               spin_unlock(&ksm_mmlist_lock);
                up_read(&mm->mmap_sem);
+               /*
+                * up_read(&mm->mmap_sem) first because after
+                * spin_unlock(&ksm_mmlist_lock) run, the "mm" may
+                * already have been freed under us by __ksm_exit()
+                * because the "mm_slot" is still hashed and
+                * ksm_scan.mm_slot doesn't point to it anymore.
+                */
+               spin_unlock(&ksm_mmlist_lock);
        }
 
        /* Repeat until we've completed scanning the whole list */