extent_io_ops has a set of 8 optional hooks which are set only for data and freespace inodes. The majority of them actually deal with delallocs in one way or another. Inspecting the code it transpired that there is actually no need to have them as function pointers in a structure. Data/freespace inodes can easily be distinguished from the btree_inode (which is pending removal anyway) by inspecting extent_io_tree::private_data. This member is set by all data/freespace inodes. This series exploits this fact to remove the majority of them. Others, such as fill_delalloc, writepage_start_hook and writepage_end_io_hook are always called from the data writeout path and can be directly called without having to check whether the respective pointers are set.
This series has undergone multiple xfstest runs and no regressions were identified. Additionally all but run_delalloc_range functions are given more descriptive names, related to their actual intent. Nikolay Borisov (8): btrfs: Remove extent_io_ops::fill_delalloc btrfs: Remove extent_io_ops::writepage_start_hook btrfs: Remove extent_io_ops::writepage_end_io_hook btrfs: Remove extent_io_ops::check_extent_io_range callback btrfs: Remove extent_io_ops::set_bit_hook extent_io callback btrfs: Remove extent_io_ops::clear_bit_hook callback btrfs: Remove extent_io_ops::merge_extent_hook callback btrfs: Remove extent_io_ops::split_extent_hook callback fs/btrfs/compression.c | 11 ++-- fs/btrfs/ctree.h | 15 +++++ fs/btrfs/extent_io.c | 131 +++++++++++++++++---------------------- fs/btrfs/extent_io.h | 24 ------- fs/btrfs/inode.c | 86 +++++++++---------------- fs/btrfs/tests/extent-io-tests.c | 2 +- 6 files changed, 105 insertions(+), 164 deletions(-) -- 2.7.4
