[PATCH] 2.6.10 - direct-io async short read bug

[Sébastien Dugué]

 Hi,

 When reading a file in async mode (using kernel AIO), and the file
size is lower than the requested size (short read),  the direct IO
layer reports an incorrect number of bytes read (transferred).

 That case is handled for the sync path in 'direct_io_worker' 
(fs/direct-io.c) where a check is made against the file size.

 This patch does the same thing for the async path. It applies to
a vanilla 2.6.10 kernel.

 Please CC: me any comments and/or suggestions.

 Sébastien.

------------------------------------------------------

  Sébastien Dugué                BULL/FREC:B1-247
  phone: (+33) 476 29 77 70      Bullcom: 229-7770

  mailto:sebastien.dugue@xxxxxxxx

  Linux POSIX AIO: http://www.bullopensource.org/posix
  
------------------------------------------------------
--- linux-2.6.10/fs/direct-io.c.orig	2005-03-03 12:41:05.541074051 +0100
+++ linux-2.6.10/fs/direct-io.c	2005-03-03 12:01:46.206210808 +0100
@@ -230,17 +230,30 @@ static void finished_one_bio(struct dio 
 	spin_lock_irqsave(&dio->bio_lock, flags);
 	if (dio->bio_count == 1) {
 		if (dio->is_async) {
+			ssize_t transferred;
+			ssize_t i_size;
+			loff_t offset;
+
 			/*
 			 * Last reference to the dio is going away.
 			 * Drop spinlock and complete the DIO.
 			 */
 			spin_unlock_irqrestore(&dio->bio_lock, flags);
-			dio_complete(dio, dio->block_in_file << dio->blkbits,
-					dio->result);
+
+			/* Check for short read case */
+			transferred = dio->result;
+			i_size = i_size_read (dio->inode);
+			offset = dio->iocb->ki_pos;
+
+			if ((dio->rw == READ) && ((offset + transferred) > i_size))
+				transferred = i_size - offset;
+
+			dio_complete(dio, offset, transferred);
+
 			/* Complete AIO later if falling back to buffered i/o */
 			if (dio->result == dio->size ||
 				((dio->rw == READ) && dio->result)) {
-				aio_complete(dio->iocb, dio->result, 0);
+				aio_complete(dio->iocb, transferred, 0);
 				kfree(dio);
 				return;
 			} else {