f2fs: fix race in between GC and atomic open
authorChao Yu <yuchao0@huawei.com>
Tue, 17 Apr 2018 09:51:28 +0000 (17:51 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Thu, 31 May 2018 18:31:46 +0000 (11:31 -0700)
Thread GC thread
- f2fs_ioc_start_atomic_write
 - get_dirty_pages
 - filemap_write_and_wait_range
- f2fs_gc
 - do_garbage_collect
  - gc_data_segment
   - move_data_page
    - f2fs_is_atomic_file
    - set_page_dirty
 - set_inode_flag(, FI_ATOMIC_FILE)

Dirty data page can still be generated by GC in race condition as
above call stack.

This patch adds fi->dio_rwsem[WRITE] in f2fs_ioc_start_atomic_write
to avoid such race.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/file.c

index f639b5fcd66f3a73e54547b6c76a5618cb22e96c..18be189075297ec66d6319c24021442e114fcaa4 100644 (file)
@@ -1677,6 +1677,8 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
 
        inode_lock(inode);
 
+       down_write(&F2FS_I(inode)->dio_rwsem[WRITE]);
+
        if (f2fs_is_atomic_file(inode))
                goto out;
 
@@ -1706,6 +1708,7 @@ inc_stat:
        stat_inc_atomic_write(inode);
        stat_update_max_atomic_write(inode);
 out:
+       up_write(&F2FS_I(inode)->dio_rwsem[WRITE]);
        inode_unlock(inode);
        mnt_drop_write_file(filp);
        return ret;