ext4: move aio completion after unwritten extent conversion
jiayingz@google.com (Jiaying Zhang) [Tue, 27 Jul 2010 15:56:06 +0000 (11:56 -0400)]
This patch is to be applied upon Christoph's "direct-io: move aio_complete
into ->end_io" patch. It adds iocb and result fields to struct ext4_io_end_t,
so that we can call aio_complete from  ext4_end_io_nolock() after the extent
conversion has finished.

I have verified with Christoph's aio-dio test that used to fail after a few
runs on an original kernel but now succeeds on the patched kernel.

See http://thread.gmane.org/gmane.comp.file-systems.ext4/19659 for details.

Signed-off-by: Jiaying Zhang <jiayingz@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

fs/ext4/ext4.h
fs/ext4/inode.c

index 4c7d472..fbb3947 100644 (file)
@@ -170,13 +170,15 @@ struct mpage_da_data {
 };
 #define        EXT4_IO_UNWRITTEN       0x1
 typedef struct ext4_io_end {
-       struct list_head        list;           /* per-file finished AIO list */
+       struct list_head        list;           /* per-file finished IO list */
        struct inode            *inode;         /* file being written to */
        unsigned int            flag;           /* unwritten or not */
        struct page             *page;          /* page struct for buffer write */
        loff_t                  offset;         /* offset in the file */
        ssize_t                 size;           /* size of the extent */
        struct work_struct      work;           /* data work queue */
+       struct kiocb            *iocb;          /* iocb struct for AIO */
+       int                     result;         /* error value for AIO */
 } ext4_io_end_t;
 
 /*
index 609159e..46d2079 100644 (file)
@@ -3668,6 +3668,8 @@ static int ext4_end_io_nolock(ext4_io_end_t *io)
                return ret;
        }
 
+       if (io->iocb)
+               aio_complete(io->iocb, io->result, 0);
        /* clear the DIO AIO unwritten flag */
        io->flag = 0;
        return ret;
@@ -3767,6 +3769,8 @@ static ext4_io_end_t *ext4_init_io_end (struct inode *inode, gfp_t flags)
                io->offset = 0;
                io->size = 0;
                io->page = NULL;
+               io->iocb = NULL;
+               io->result = 0;
                INIT_WORK(&io->work, ext4_end_io_work);
                INIT_LIST_HEAD(&io->list);
        }
@@ -3796,12 +3800,18 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
        if (io_end->flag != EXT4_IO_UNWRITTEN){
                ext4_free_io_end(io_end);
                iocb->private = NULL;
-               goto out;
+out:
+               if (is_async)
+                       aio_complete(iocb, ret, 0);
+               return;
        }
 
        io_end->offset = offset;
        io_end->size = size;
-       io_end->flag = EXT4_IO_UNWRITTEN;
+       if (is_async) {
+               io_end->iocb = iocb;
+               io_end->result = ret;
+       }
        wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
 
        /* queue the work to convert unwritten extents to written */
@@ -3813,9 +3823,6 @@ static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
        list_add_tail(&io_end->list, &ei->i_completed_io_list);
        spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
        iocb->private = NULL;
-out:
-       if (is_async)
-               aio_complete(iocb, ret, 0);
 }
 
 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)