f2fs: let sync node IO interrupt async one

Although mixed sync/async IOs can have continuous LBA, as they have
different IO priority, block IO scheduler will add them into different
queues and commit them separately, result in splited IOs which causes
wrose performance.

This patch gives high priority to synchronous IO of nodes, means that
once synchronous flow starts, it can interrupt asynchronous writeback
flow of system flusher, so more big IOs can be expected.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

7 files changed, 37 insertions, 9 deletions

diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index 60b4886f5bb6..e255e9b5538f 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -1090,7 +1090,9 @@ retry_flush_nodes:
 
 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 		up_write(&sbi->node_write);
+		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) {
 			up_write(&sbi->node_change);
 			f2fs_unlock_all(sbi);
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index 4b0db685e5d5..a166927355c8 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -1930,6 +1930,7 @@ static int f2fs_write_cache_pages(struct address_space *mapping,
 	int ret = 0;
 	int done = 0;
 	struct pagevec pvec;
+	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 	int nr_pages;
 	pgoff_t uninitialized_var(writeback_index);
 	pgoff_t index;
@@ -1984,7 +1985,7 @@ retry:
 			bool submitted = false;
 
 			/* give a priority to WB_SYNC threads */
-			if (atomic_read(&F2FS_M_SB(mapping)->wb_sync_req) &&
+			if (atomic_read(&sbi->wb_sync_req[DATA]) &&
 					wbc->sync_mode == WB_SYNC_NONE) {
 				done = 1;
 				break;
@@ -2104,8 +2105,8 @@ static int __f2fs_write_data_pages(struct address_space *mapping,
 
 	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
 	if (wbc->sync_mode == WB_SYNC_ALL)
-		atomic_inc(&sbi->wb_sync_req);
-	else if (atomic_read(&sbi->wb_sync_req))
+		atomic_inc(&sbi->wb_sync_req[DATA]);
+	else if (atomic_read(&sbi->wb_sync_req[DATA]))
 		goto skip_write;
 
 	blk_start_plug(&plug);
@@ -2113,7 +2114,7 @@ static int __f2fs_write_data_pages(struct address_space *mapping,
 	blk_finish_plug(&plug);
 
 	if (wbc->sync_mode == WB_SYNC_ALL)
-		atomic_dec(&sbi->wb_sync_req);
+		atomic_dec(&sbi->wb_sync_req[DATA]);
 	/*
 	 * if some pages were truncated, we cannot guarantee its mapping->host
 	 * to detect pending bios.
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index e91f7ff71dc6..6873b321c2c1 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -1265,7 +1265,7 @@ struct f2fs_sb_info {
 	struct percpu_counter alloc_valid_block_count;
 
 	/* writeback control */
-	atomic_t wb_sync_req;			/* count # of WB_SYNC threads */
+	atomic_t wb_sync_req[META];	/* count # of WB_SYNC threads */
 
 	/* valid inode count */
 	struct percpu_counter total_valid_inode_count;
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index de1c712777c9..8b0002f05451 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -277,7 +277,9 @@ go_write:
 		goto out;
 	}
 sync_nodes:
+	atomic_inc(&sbi->wb_sync_req[NODE]);
 	ret = f2fs_fsync_node_pages(sbi, inode, &wbc, atomic);
+	atomic_dec(&sbi->wb_sync_req[NODE]);
 	if (ret)
 		goto out;
 
diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index dcadc0691a3e..772ef64d2035 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -473,12 +473,16 @@ static void gc_node_segment(struct f2fs_sb_info *sbi,
 	block_t start_addr;
 	int off;
 	int phase = 0;
+	bool fggc = (gc_type == FG_GC);
 
 	start_addr = START_BLOCK(sbi, segno);
 
 next_step:
 	entry = sum;
 
+	if (fggc && phase == 2)
+		atomic_inc(&sbi->wb_sync_req[NODE]);
+
 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 		nid_t nid = le32_to_cpu(entry->nid);
 		struct page *node_page;
@@ -525,6 +529,9 @@ next_step:
 
 	if (++phase < 3)
 		goto next_step;
+
+	if (fggc)
+		atomic_dec(&sbi->wb_sync_req[NODE]);
 }
 
 /*
diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index 5264b079b93e..baa8ee1aca38 100644
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -1596,21 +1596,28 @@ int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
 	int step = 0;
 	int nwritten = 0;
 	int ret = 0;
-	int nr_pages;
+	int nr_pages, done = 0;
 
 	pagevec_init(&pvec, 0);
 
 next_step:
 	index = 0;
 
-	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
-				PAGECACHE_TAG_DIRTY))) {
+	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
+			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
 		int i;
 
 		for (i = 0; i < nr_pages; i++) {
 			struct page *page = pvec.pages[i];
 			bool submitted = false;
 
+			/* give a priority to WB_SYNC threads */
+			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
+					wbc->sync_mode == WB_SYNC_NONE) {
+				done = 1;
+				break;
+			}
+
 			/*
 			 * flushing sequence with step:
 			 * 0. indirect nodes
@@ -1741,6 +1748,11 @@ static int f2fs_write_node_pages(struct address_space *mapping,
 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
 		goto skip_write;
 
+	if (wbc->sync_mode == WB_SYNC_ALL)
+		atomic_inc(&sbi->wb_sync_req[NODE]);
+	else if (atomic_read(&sbi->wb_sync_req[NODE]))
+		goto skip_write;
+
 	trace_f2fs_writepages(mapping->host, wbc, NODE);
 
 	diff = nr_pages_to_write(sbi, NODE, wbc);
@@ -1748,6 +1760,9 @@ static int f2fs_write_node_pages(struct address_space *mapping,
 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
 	blk_finish_plug(&plug);
 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
+
+	if (wbc->sync_mode == WB_SYNC_ALL)
+		atomic_dec(&sbi->wb_sync_req[NODE]);
 	return 0;
 
 skip_write:
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index d306725d7399..0b803213ed64 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -2360,7 +2360,8 @@ static void init_sb_info(struct f2fs_sb_info *sbi)
 	for (i = 0; i < NR_COUNT_TYPE; i++)
 		atomic_set(&sbi->nr_pages[i], 0);
 
-	atomic_set(&sbi->wb_sync_req, 0);
+	for (i = 0; i < META; i++)
+		atomic_set(&sbi->wb_sync_req[i], 0);
 
 	INIT_LIST_HEAD(&sbi->s_list);
 	mutex_init(&sbi->umount_mutex);