Re: [PATCH] ext4: use directio end_io error status to finish unwritten aio dio correctly

[Christoph Hellwig <hch@infradead.org>]

Might help to tell that this is on top of a direct-io.c patch from the
XFS tree.

I don't think clearing any flags is the right thing - now that we
always call ->end_io the code dealing with it in ext4_ext_direct_IO
can simply be moved to the ->end_io handler.

Something like the untested patch below:

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 9db04dd..b741c79 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3166,23 +3166,25 @@ static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
 {
         ext4_io_end_t *io_end = iocb->private;
 
-	if (size <= 0)
-		return 0;
-
 	/* if not async direct IO just return */
 	if (!io_end)
 		return 0;
 
+	if (size <= 0) {
+		WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
+		goto out;
+	}
+
 	ext_debug("ext4_end_io_dio(): io_end 0x%p "
 		  "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
  		  iocb->private, io_end->inode->i_ino, iocb, offset,
 		  size);
 
-	iocb->private = NULL;
 	io_end->offset = offset;
 	io_end->size = size;
+out:
 	ext4_put_io_end(io_end);