ext4: fix race between unwritten extent conversion and truncate
authorJeff Moyer <jmoyer@redhat.com>
Mon, 20 Feb 2012 22:59:24 +0000 (17:59 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Mon, 20 Feb 2012 22:59:24 +0000 (17:59 -0500)
The following comment in ext4_end_io_dio caught my attention:

/* XXX: probably should move into the real I/O completion handler */
        inode_dio_done(inode);

The truncate code takes i_mutex, then calls inode_dio_wait.  Because the
ext4 code path above will end up dropping the mutex before it is
reacquired by the worker thread that does the extent conversion, it
seems to me that the truncate can happen out of order.  Jan Kara
mentioned that this might result in error messages in the system logs,
but that should be the extent of the "damage."

The fix is pretty straight-forward: don't call inode_dio_done until the
extent conversion is complete.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
fs/ext4/ext4.h
fs/ext4/inode.c
fs/ext4/page-io.c

index c2314dc..4076746 100644 (file)
@@ -184,6 +184,7 @@ struct mpage_da_data {
 #define        EXT4_IO_END_UNWRITTEN   0x0001
 #define EXT4_IO_END_ERROR      0x0002
 #define EXT4_IO_END_QUEUED     0x0004
+#define EXT4_IO_END_DIRECT     0x0008
 
 struct ext4_io_page {
        struct page     *p_page;
index a58812e..aafc626 100644 (file)
@@ -2796,9 +2796,6 @@ out:
 
        /* queue the work to convert unwritten extents to written */
        queue_work(wq, &io_end->work);
-
-       /* XXX: probably should move into the real I/O completion handler */
-       inode_dio_done(inode);
 }
 
 static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate)
@@ -2922,9 +2919,12 @@ static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
                iocb->private = NULL;
                EXT4_I(inode)->cur_aio_dio = NULL;
                if (!is_sync_kiocb(iocb)) {
-                       iocb->private = ext4_init_io_end(inode, GFP_NOFS);
-                       if (!iocb->private)
+                       ext4_io_end_t *io_end =
+                               ext4_init_io_end(inode, GFP_NOFS);
+                       if (!io_end)
                                return -ENOMEM;
+                       io_end->flag |= EXT4_IO_END_DIRECT;
+                       iocb->private = io_end;
                        /*
                         * we save the io structure for current async
                         * direct IO, so that later ext4_map_blocks()
index 4758518..9e1b8eb 100644 (file)
@@ -110,6 +110,8 @@ int ext4_end_io_nolock(ext4_io_end_t *io)
        if (io->iocb)
                aio_complete(io->iocb, io->result, 0);
 
+       if (io->flag & EXT4_IO_END_DIRECT)
+               inode_dio_done(inode);
        /* Wake up anyone waiting on unwritten extent conversion */
        if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten))
                wake_up_all(ext4_ioend_wq(io->inode));