[f2fs-dev] [PATCH RFC] f2fs: sync node page without cp_rwsem during block_operations()

[Chao Yu <chao@kernel.org>]

Let's avoid holding cp_rwsem write lock during f2fs_sync_node_pages()
in block_operations(), then cp_rwsem write lock's coverage in checkpoint
can be shrunk, then it expects there will be more opportunity that
foreground operator can grab cp_rwsem read lock.

Signed-off-by: Chao Yu <chao@kernel.org>
---
 fs/f2fs/checkpoint.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index f7cb4277de70..7596faa0be45 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -1190,6 +1190,8 @@ static int block_operations(struct f2fs_sb_info *sbi)
 		.for_reclaim = 0,
 	};
 	int err = 0, cnt = 0;
+	bool sync_lockless = true;
+	unsigned int retry_cnt = 3;
 
 	/*
 	 * Let's flush inline_data in dirty node pages.
@@ -1249,15 +1251,25 @@ static int block_operations(struct f2fs_sb_info *sbi)
 
 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 		f2fs_up_write(&sbi->node_write);
+		if (!retry_cnt--)
+			sync_lockless = false;
+		if (sync_lockless) {
+			f2fs_up_write(&sbi->node_change);
+			f2fs_unlock_all(sbi);
+		}
 		atomic_inc(&sbi->wb_sync_req[NODE]);
 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
 		atomic_dec(&sbi->wb_sync_req[NODE]);
 		if (err) {
-			f2fs_up_write(&sbi->node_change);
-			f2fs_unlock_all(sbi);
+			if (!sync_lockless) {
+				f2fs_up_write(&sbi->node_change);
+				f2fs_unlock_all(sbi);
+			}
 			return err;
 		}
 		cond_resched();
+		if (sync_lockless)
+			goto retry_flush_quotas;
 		goto retry_flush_nodes;
 	}
 
-- 
2.25.1



_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel