Re: [PATCH] O_DIRECT for ext3 (2.4.21)

[Andrew Morton]

Andrew Morton <[e-mail]> wrote:
>
> I think the check should be implemented in (the new) ext3_open().  Because
>  checking the return from open() is the way in which a good application would
>  determine whether the underlying fs supports O_DIRECT.
> 
>  Unfortunately O_DIRECT can also be set with fcntl(F_SETFL), and we seem to
>  have forgotten to provide a way for the fs to be told about fcntl.

It works out OK in 2.5, and we should do it this way in 2.4 too:

- dentry_open() checks for inode->i_mapping->a_ops->direct_IO

- setfl() checks for inode->i_mapping->a_ops->direct_IO

- the a_ops for data-journalled inodes have a null ->direct_IO.