fs: always maintain i_dio_count

Maintain i_dio_count for all filesystems, not just those using DIO_LOCKING.
This these filesystems to also protect truncate against direct I/O requests
by using common code.  Right now the only non-DIO_LOCKING filesystem that
appears to do so is XFS, which uses an opencoded variant of the i_dio_count
scheme.

Behaviour doesn't change for filesystems never calling inode_dio_wait.
For ext4 behaviour changes when using the dioread_nonlock option, which
previously was missing any protection between truncate and direct I/O reads.
For ocfs2 that handcrafted i_dio_count manipulations are replaced with
the common code now enable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fs/direct-io.c

@@ -297,7 +297,6 @@ static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is
 		aio_complete(dio->iocb, ret, 0);
 	}
 
-	if (dio->flags & DIO_LOCKING)
 	inode_dio_done(dio->inode);
 	return ret;
 }
@@ -1185,14 +1184,16 @@ direct_io_worker(int rw, struct kiocb *iocb, struct inode *inode,
  *    For writes this function is called under i_mutex and returns with
  *    i_mutex held, for reads, i_mutex is not held on entry, but it is
  *    taken and dropped again before returning.
- *    The i_dio_count counter keeps track of the number of outstanding
- *    direct I/O requests, and truncate waits for it to reach zero.
- *    New references to i_dio_count must only be grabbed with i_mutex
- *    held.
- *
  *  - if the flags value does NOT contain DIO_LOCKING we don't use any
  *    internal locking but rather rely on the filesystem to synchronize
  *    direct I/O reads/writes versus each other and truncate.
+ *
+ * To help with locking against truncate we incremented the i_dio_count
+ * counter before starting direct I/O, and decrement it once we are done.
+ * Truncate can wait for it to reach zero to provide exclusion.  It is
+ * expected that filesystem provide exclusion between new direct I/O
+ * and truncates.  For DIO_LOCKING filesystems this is done by i_mutex,
+ * but other filesystems need to take care of this on their own.
  */
 ssize_t
 __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
@@ -1270,12 +1271,12 @@ __blockdev_direct_IO(int rw, struct kiocb *iocb, struct inode *inode,
 				goto out;
 			}
 		}
+	}
 
 	/*
 	 * Will be decremented at I/O completion time.
 	 */
 	atomic_inc(&inode->i_dio_count);
-	}
 
 	/*
 	 * For file extending writes updating i_size before data

fs/ocfs2/aops.c

@@ -567,10 +567,8 @@ static void ocfs2_dio_end_io(struct kiocb *iocb,
 	/* this io's submitter should not have unlocked this before we could */
 	BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
 
-	if (ocfs2_iocb_is_sem_locked(iocb)) {
-		inode_dio_done(inode);
+	if (ocfs2_iocb_is_sem_locked(iocb))
 		ocfs2_iocb_clear_sem_locked(iocb);
-	}
 
 	ocfs2_iocb_clear_rw_locked(iocb);
 

fs/ocfs2/file.c

@@ -2240,7 +2240,6 @@ static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
 relock:
 	/* to match setattr's i_mutex -> rw_lock ordering */
 	if (direct_io) {
-		atomic_inc(&inode->i_dio_count);
 		have_alloc_sem = 1;
 		/* communicate with ocfs2_dio_end_io */
 		ocfs2_iocb_set_sem_locked(iocb);
@@ -2292,7 +2291,6 @@ relock:
 	 */
 	if (direct_io && !can_do_direct) {
 		ocfs2_rw_unlock(inode, rw_level);
-		inode_dio_done(inode);
 
 		have_alloc_sem = 0;
 		rw_level = -1;
@@ -2379,10 +2377,8 @@ out:
 		ocfs2_rw_unlock(inode, rw_level);
 
 out_sems:
-	if (have_alloc_sem) {
-		inode_dio_done(inode);
+	if (have_alloc_sem)
 		ocfs2_iocb_clear_sem_locked(iocb);
-	}
 
 	mutex_unlock(&inode->i_mutex);
 
@@ -2533,7 +2529,6 @@ static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
 	 */
 	if (filp->f_flags & O_DIRECT) {
 		have_alloc_sem = 1;
-		atomic_inc(&inode->i_dio_count);
 		ocfs2_iocb_set_sem_locked(iocb);
 
 		ret = ocfs2_rw_lock(inode, 0);
@@ -2575,10 +2570,9 @@ static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
 	}
 
 bail:
-	if (have_alloc_sem) {
-		inode_dio_done(inode);
+	if (have_alloc_sem)
 		ocfs2_iocb_clear_sem_locked(iocb);
-	}
+
 	if (rw_level != -1)
 		ocfs2_rw_unlock(inode, rw_level);