Commit 105d931d authored by Chris Mason's avatar Chris Mason
Browse files

Btrfs: switch back to wait_on_page_writeback to wait on metadata writes 



The extent based waiting was using more CPU, and other fixes have helped
with the unplug storm problems.

Signed-off-by: default avatarChris Mason <chris.mason@oracle.com>
parent 73e9f5be

fs/btrfs/transaction.c

+1 −28
Original line number Diff line number Diff line
@@ -332,7 +332,6 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 
	int werr = 0;

	struct page *page;

	struct inode *btree_inode = root->fs_info->btree_inode;

	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;

	u64 start = 0;

	u64 end;

	unsigned long index;
@@ -373,11 +372,6 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 
			page_cache_release(page);

		}

	}

	/*

	 * we unplug once and then use the wait_on_extent_bit for

	 * everything else

	 */

	blk_run_address_space(btree_inode->i_mapping);

	while(1) {

		ret = find_first_extent_bit(dirty_pages, 0, &start, &end,

					    EXTENT_DIRTY);
@@ -398,28 +392,7 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 
				if (err)

					werr = err;

			}

			if (PageWriteback(page)) {

				/*

				 * we don't wait on the page writeback bit

				 * because that triggers a lot of unplugs.

				 * The extent bits are much nicer to

				 * the disks, but come with a slightly

				 * higher latency because we aren't forcing

				 * unplugs.

				 */

				wait_on_extent_writeback(io_tree,

					 page_offset(page),

					 page_offset(page) +

					 PAGE_CACHE_SIZE - 1);

			}

			if (PageWriteback(page)) {

				/*

				 * the state bits get cleared before the

				 * page bits, lets add some extra

				 * paranoia here

				 */

			wait_on_page_writeback(page);

			}

			page_cache_release(page);

			cond_resched();

		}