Dear David Sterba,
David Sterba wrote:
> On Fri, Oct 05, 2012 at 08:56:44PM +0900, ????????? wrote:
> > +struct node_footer {
> > + __le32 nid; /* node id */
> > + __le32 ino; /* inode nunmber */
> > + __le32 cold:1; /* cold mark */
> > + __le32 fsync:1; /* fsync mark */
> > + __le32 dentry:1; /* dentry mark */
> > + __le32 offset:29; /* offset in inode's node space */
>
> A bitfield for a on-disk structure? This will have endianity issues,
> (but I don't know if you intend to support big-endian). It's not enough
> to use cpu_to_le* as in
>
> fill_node_footer(...) {
>
> rn->footer.offset = cpu_to_le32(ofs);
>
> }
>
> because the bitfield inside the structure will be already defined
> reversed. The cpu_to_le macro will only convert value of 'ofs' but will
> place it to different bits than it would on a little-endian arch.
>
> There are macros to define bitfields in an endian-neutral way (or do it
> by #ifdefs though it also involves duplicating the item names), or you
> can alternatively use two structs fr disk-only and memory-only access,
> the disk one stores __le32 with value combined of all and the in-memory
> gets set up properly and will look like your current version of the
> structure.
>
> (More about not using bitfields
> http://yarchive.net/comp/linux/bitfields.html)
I appreciate your kind feedback, but we should've avoided using bitfields
for on-disk structure. As you suggested, we'll revise them. The latter
approach would be good.
>
>
> david
Chul
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
