ext4: fix ext4_end_io_dio() racing against fsync()
authorTheodore Ts'o <tytso@mit.edu>
Mon, 12 Dec 2011 15:53:02 +0000 (10:53 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Mon, 12 Dec 2011 15:53:02 +0000 (10:53 -0500)
We need to make sure iocb->private is cleared *before* we put the
io_end structure on i_completed_io_list.  Otherwise fsync() could
potentially run on another CPU and free the iocb structure out from
under us.

Reported-by: Kent Overstreet <koverstreet@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
fs/ext4/inode.c

index 848f436df29f6ffec6e8487549764d98c63d5b6b..d2c86633811f997fec8ed00d7978e8e16f7f692a 100644 (file)
@@ -2781,10 +2781,11 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
                  iocb->private, io_end->inode->i_ino, iocb, offset,
                  size);
 
+       iocb->private = NULL;
+
        /* if not aio dio with unwritten extents, just free io and return */
        if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
                ext4_free_io_end(io_end);
-               iocb->private = NULL;
 out:
                if (is_async)
                        aio_complete(iocb, ret, 0);
@@ -2807,7 +2808,6 @@ out:
        spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
 
        /* queue the work to convert unwritten extents to written */
-       iocb->private = NULL;
        queue_work(wq, &io_end->work);
 
        /* XXX: probably should move into the real I/O completion handler */