ext4: protect i_disksize update by i_data_sem in direct write path

i_disksize update should be protected by i_data_sem, by either taking
the lock explicitly or by using ext4_update_i_disksize() helper. But the
i_disksize updates in ext4_direct_IO_write() are not protected at all,
which may be racing with i_disksize updates in writeback path in
delalloc buffer write path.

This is found by code inspection, and I didn't hit any i_disksize
corruption due to this bug. Thanks to Jan Kara for catching this bug and
suggesting the fix!

Reported-by: Jan Kara <jack@suse.cz>
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Eryu Guan <guaneryu@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@vger.kernel.org

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index c94780075b04f752b1cfdec9ccf6f631bd874116..bff44b4a0783f38d6819542314d9e4c70890aa15 100644 (file)
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3658,7 +3658,6 @@ static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
 {
        struct file *file = iocb->ki_filp;
        struct inode *inode = file->f_mapping->host;
-       struct ext4_inode_info *ei = EXT4_I(inode);
        ssize_t ret;
        loff_t offset = iocb->ki_pos;
        size_t count = iov_iter_count(iter);
@@ -3682,7 +3681,7 @@ static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
                        goto out;
                }
                orphan = 1;
-               ei->i_disksize = inode->i_size;
+               ext4_update_i_disksize(inode, inode->i_size);
                ext4_journal_stop(handle);
        }
 
@@ -3790,7 +3789,7 @@ static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
                if (ret > 0) {
                        loff_t end = offset + ret;
                        if (end > inode->i_size) {
-                               ei->i_disksize = end;
+                               ext4_update_i_disksize(inode, end);
                                i_size_write(inode, end);
                                /*
                                 * We're going to return a positive `ret'