ext4: wait for existing dio workers in ext4_alloc_file_blocks()

Currently existing dio workers can jump in and potentially increase
extent tree depth while we're allocating blocks in
ext4_alloc_file_blocks().  This may cause us to underestimate the
number of credits needed for the transaction because the extent tree
depth can change after our estimation.

Fix this by waiting for all the existing dio workers in the same way
as we do it in ext4_punch_hole.  We've seen errors caused by this in
xfstest generic/299, however it's really hard to reproduce.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>

diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c
index fc76bda..1ba8b4a 100644 (file)
--- a/fs/ext4/extents.c
+++ b/fs/ext4/extents.c
@@ -4678,6 +4678,10 @@ static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
        if (len <= EXT_UNWRITTEN_MAX_LEN)
                flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
 
+       /* Wait all existing dio workers, newcomers will block on i_mutex */
+       ext4_inode_block_unlocked_dio(inode);
+       inode_dio_wait(inode);
+
        /*
         * credits to insert 1 extent into extent tree
         */
@@ -4741,6 +4745,8 @@ retry:
                goto retry;
        }
 
+       ext4_inode_resume_unlocked_dio(inode);
+
        return ret > 0 ? ret2 : ret;
 }