The functions txEnd() and txExit() can be concurrently executed in the
following call contexts:

Thread1: 
  jfs_lazycommit()
    txLazyCommit()
      txEnd()
Thread2:
  exit_jfs_fs()
    txExit()

In txEnd():
  struct tblock *tblk = tid_to_tblock(tid);
  // #define tid_to_tblock(tid) (&TxBlock[tid])

In txExit():
  vfree(TxBlock);
  TxBlock = NULL;

Thus, data races can occur for the global variable TxBlock.
And these data races can cause use-after-free bugs and null-pointer
dereferences.

To fix these data races, txExit() is called after the filesystem stops 
the threads that run jfs_lazycommit().

These data races are found by our concurrency fuzzer.

Signed-off-by: Jia-Ju Bai <baijiaju1...@gmail.com>
---
 fs/jfs/super.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/jfs/super.c b/fs/jfs/super.c
index b2dc4d1f9dcc..8c80397df336 100644
--- a/fs/jfs/super.c
+++ b/fs/jfs/super.c
@@ -1027,13 +1027,13 @@ static void __exit exit_jfs_fs(void)
 
        jfs_info("exit_jfs_fs called");
 
-       txExit();
        metapage_exit();
 
        kthread_stop(jfsIOthread);
        for (i = 0; i < commit_threads; i++)
                kthread_stop(jfsCommitThread[i]);
        kthread_stop(jfsSyncThread);
+       txExit();
 #ifdef PROC_FS_JFS
        jfs_proc_clean();
 #endif
-- 
2.17.1

Reply via email to