I've been taking a look at the submit_bio_hook et al and I saw some opportunities for improvement. First I begin by simplifying the definition of submit_bio_hook callback and making one of its arguments explicit.
Patch 3 removes a redundant extent_io_tree argument that was passed through several layers of functions and ultimately made it somewhat cumbersome to reason what extent_io_tree is being used. Patch 4 and 5 hardcode a 0 being passed for the bio_offset parameter to btrfs_submit_bio_start/btree_submit_bio_start since it only matters for DIO. Patch 6 finally does away with the bio_offset parameter of submit_bio_hook. Here is the output of bloat-o-meter for posterity: add/remove: 0/0 grow/shrink: 4/7 up/down: 48/-319 (-271) Function old new delta read_extent_buffer_pages 784 822 +38 btrfs_add_ordered_sum 100 104 +4 btrfs_submit_bio_hook 364 367 +3 btree_submit_bio_hook 175 178 +3 end_bio_extent_readpage 2119 2116 -3 btrfs_reloc_clone_csums 272 262 -10 btrfs_csum_one_bio 1358 1348 -10 btree_read_extent_buffer_pages 282 272 -10 readahead_tree_block 73 55 -18 reada_tree_block_flagged 194 165 -29 submit_one_bio 313 74 -239 Total: Before=1072719, After=1072448, chg -0.03% Nikolay Borisov (6): btrfs: Define submit_bio_hook's type directly btrfs: Change submit_bio_hook to taking an inode directly btrfs: Remove 'tree' argument from read_extent_buffer_pages btrfs: Pass 0 for bio_offset to btrfs_wq_submit_bio btrfs: Always pass 0 bio_offset for btree_submit_bio_start btrfs: Remove bio_offset argument from submit_bio_hook fs/btrfs/disk-io.c | 22 +++++++--------------- fs/btrfs/extent_io.c | 10 ++++------ fs/btrfs/extent_io.h | 9 +++------ fs/btrfs/inode.c | 11 +++++------ 4 files changed, 19 insertions(+), 33 deletions(-) -- 2.17.1
