Hi Jan,
> Hum, these games with several callbacks, lists, etc. look awkward and
> unnecessary. It think they mostly come from the fact that we call separate
> freeing callback for each extent to free which doesn't fit the needs of
> async discard well.
> So instead of adding post_cb_list and several callback functions, it would
> seem easier to have just one callback structure instead of one for every
> extent. Then the structure would contain a list of extents that need to be
> freed freed. So something like:
> struct ext4_free_data {
> struct ext4_journal_cb_entry efd_jce;
> struct list_head efd_extents;
> }
> struct ext4_freed_extent {
> struct list_head efe_list;
> struct rb_node efe_node;
> ext4_group_t efe_group;
> ext4_grpblk_t efe_start_cluster;
> ext4_grpblk_t efe_count;
> tid_t efe_tid;
> }
> When commit happens, we can just walk the efd_extents list while efe_tid is
> equal tid of the transaction for which the callback was called and submit all
> discard requests. You can use bio chaining implemented in
> __blkdev_issue_discard() which XFS already uses and so the result of all
> the discards you submit will be just one bio. Then you walk the list of
> extents again and free them in the buddy bitmaps. And finally, you wait for
> the bio to complete. All will be then happening in one function and it will
> be much easier to understand.
It's right. the patch didn't look neat because of a few callbacks and the
post callback list. I will modify the patch as your suggestion. It will
look better.
Thank you very much. :-)
