f2fs: clarify and enhance the f2fs_gc flow

This patch makes clearer the ambiguous f2fs_gc flow as follows.

1. Remove intermediate checkpoint condition during f2fs_gc
 (i.e., should_do_checkpoint() and GC_BLOCKED)

2. Remove unnecessary return values of f2fs_gc because of #1.
 (i.e., GC_NODE, GC_OK, etc)

3. Simplify write_checkpoint() because of #2.

4. Clarify the main f2fs_gc flow.
 o monitor how many freed sections during one iteration of do_garbage_collect().
 o do GC more without checkpoints if we can't get enough free sections.
 o do checkpoint once we've got enough free sections through forground GCs.

5. Adopt thread-logging (Slack-Space-Recycle) scheme more aggressively on data
  log types. See. get_ssr_segement()
Signed-off-by: NJaegeuk Kim <jaegeuk.kim@samsung.com>

fs/f2fs/checkpoint.c

@@ -539,7 +539,7 @@ void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 /*
  * Freeze all the FS-operations for checkpoint.
  */
-void block_operations(struct f2fs_sb_info *sbi)
+static void block_operations(struct f2fs_sb_info *sbi)
 {
 	int t;
 	struct writeback_control wbc = {
@@ -722,15 +722,13 @@ static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 /*
  * We guarantee that this checkpoint procedure should not fail.
  */
-void write_checkpoint(struct f2fs_sb_info *sbi, bool blocked, bool is_umount)
+void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 {
 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
 	unsigned long long ckpt_ver;
 
-	if (!blocked) {
-		mutex_lock(&sbi->cp_mutex);
-		block_operations(sbi);
-	}
+	mutex_lock(&sbi->cp_mutex);
+	block_operations(sbi);
 
 	f2fs_submit_bio(sbi, DATA, true);
 	f2fs_submit_bio(sbi, NODE, true);

fs/f2fs/f2fs.h

@@ -969,8 +969,7 @@ int get_valid_checkpoint(struct f2fs_sb_info *);
 void set_dirty_dir_page(struct inode *, struct page *);
 void remove_dirty_dir_inode(struct inode *);
 void sync_dirty_dir_inodes(struct f2fs_sb_info *);
-void block_operations(struct f2fs_sb_info *);
-void write_checkpoint(struct f2fs_sb_info *, bool, bool);
+void write_checkpoint(struct f2fs_sb_info *, bool);
 void init_orphan_info(struct f2fs_sb_info *);
 int __init create_checkpoint_caches(void);
 void destroy_checkpoint_caches(void);

fs/f2fs/gc.c

@@ -78,7 +78,8 @@ static int gc_thread_func(void *data)
 
 		sbi->bg_gc++;
 
-		if (f2fs_gc(sbi) == GC_NONE)
+		/* if return value is not zero, no victim was selected */
+		if (f2fs_gc(sbi))
 			wait_ms = GC_THREAD_NOGC_SLEEP_TIME;
 		else if (wait_ms == GC_THREAD_NOGC_SLEEP_TIME)
 			wait_ms = GC_THREAD_MAX_SLEEP_TIME;
@@ -360,7 +361,7 @@ static int check_valid_map(struct f2fs_sb_info *sbi,
 	sentry = get_seg_entry(sbi, segno);
 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
 	mutex_unlock(&sit_i->sentry_lock);
-	return ret ? GC_OK : GC_NEXT;
+	return ret;
 }
 
 /*
@@ -368,7 +369,7 @@ static int check_valid_map(struct f2fs_sb_info *sbi,
  * On validity, copy that node with cold status, otherwise (invalid node)
  * ignore that.
  */
-static int gc_node_segment(struct f2fs_sb_info *sbi,
+static void gc_node_segment(struct f2fs_sb_info *sbi,
 		struct f2fs_summary *sum, unsigned int segno, int gc_type)
 {
 	bool initial = true;
@@ -380,21 +381,12 @@ static int gc_node_segment(struct f2fs_sb_info *sbi,
 	for (off = 0; off < sbi->blocks_per_seg; off++, entry++) {
 		nid_t nid = le32_to_cpu(entry->nid);
 		struct page *node_page;
-		int err;
 
-		/*
-		 * It makes sure that free segments are able to write
-		 * all the dirty node pages before CP after this CP.
-		 * So let's check the space of dirty node pages.
-		 */
-		if (should_do_checkpoint(sbi)) {
-			mutex_lock(&sbi->cp_mutex);
-			block_operations(sbi);
-			return GC_BLOCKED;
-		}
+		/* stop BG_GC if there is not enough free sections. */
+		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
+			return;
 
-		err = check_valid_map(sbi, segno, off);
-		if (err == GC_NEXT)
+		if (check_valid_map(sbi, segno, off) == 0)
 			continue;
 
 		if (initial) {
@@ -424,7 +416,6 @@ static int gc_node_segment(struct f2fs_sb_info *sbi,
 		};
 		sync_node_pages(sbi, 0, &wbc);
 	}
-	return GC_DONE;
 }
 
 /*
@@ -467,13 +458,13 @@ static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 
 	node_page = get_node_page(sbi, nid);
 	if (IS_ERR(node_page))
-		return GC_NEXT;
+		return 0;
 
 	get_node_info(sbi, nid, dni);
 
 	if (sum->version != dni->version) {
 		f2fs_put_page(node_page, 1);
-		return GC_NEXT;
+		return 0;
 	}
 
 	*nofs = ofs_of_node(node_page);
@@ -481,8 +472,8 @@ static int check_dnode(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 	f2fs_put_page(node_page, 1);
 
 	if (source_blkaddr != blkaddr)
-		return GC_NEXT;
-	return GC_OK;
+		return 0;
+	return 1;
 }
 
 static void move_data_page(struct inode *inode, struct page *page, int gc_type)
@@ -523,13 +514,13 @@ static void move_data_page(struct inode *inode, struct page *page, int gc_type)
  * If the parent node is not valid or the data block address is different,
  * the victim data block is ignored.
  */
-static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
+static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 		struct list_head *ilist, unsigned int segno, int gc_type)
 {
 	struct super_block *sb = sbi->sb;
 	struct f2fs_summary *entry;
 	block_t start_addr;
-	int err, off;
+	int off;
 	int phase = 0;
 
 	start_addr = START_BLOCK(sbi, segno);
@@ -543,20 +534,11 @@ static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 		unsigned int ofs_in_node, nofs;
 		block_t start_bidx;
 
-		/*
-		 * It makes sure that free segments are able to write
-		 * all the dirty node pages before CP after this CP.
-		 * So let's check the space of dirty node pages.
-		 */
-		if (should_do_checkpoint(sbi)) {
-			mutex_lock(&sbi->cp_mutex);
-			block_operations(sbi);
-			err = GC_BLOCKED;
-			goto stop;
-		}
+		/* stop BG_GC if there is not enough free sections. */
+		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
+			return;
 
-		err = check_valid_map(sbi, segno, off);
-		if (err == GC_NEXT)
+		if (check_valid_map(sbi, segno, off) == 0)
 			continue;
 
 		if (phase == 0) {
@@ -565,8 +547,7 @@ static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 		}
 
 		/* Get an inode by ino with checking validity */
-		err = check_dnode(sbi, entry, &dni, start_addr + off, &nofs);
-		if (err == GC_NEXT)
+		if (check_dnode(sbi, entry, &dni, start_addr + off, &nofs) == 0)
 			continue;
 
 		if (phase == 1) {
@@ -606,11 +587,9 @@ static int gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 	}
 	if (++phase < 4)
 		goto next_step;
-	err = GC_DONE;
-stop:
+
 	if (gc_type == FG_GC)
 		f2fs_submit_bio(sbi, DATA, true);
-	return err;
 }
 
 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
@@ -624,17 +603,16 @@ static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 	return ret;
 }
 
-static int do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
+static void do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
 				struct list_head *ilist, int gc_type)
 {
 	struct page *sum_page;
 	struct f2fs_summary_block *sum;
-	int ret = GC_DONE;
 
 	/* read segment summary of victim */
 	sum_page = get_sum_page(sbi, segno);
 	if (IS_ERR(sum_page))
-		return GC_ERROR;
+		return;
 
 	/*
 	 * CP needs to lock sum_page. In this time, we don't need
@@ -646,17 +624,16 @@ static int do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
 
 	switch (GET_SUM_TYPE((&sum->footer))) {
 	case SUM_TYPE_NODE:
-		ret = gc_node_segment(sbi, sum->entries, segno, gc_type);
+		gc_node_segment(sbi, sum->entries, segno, gc_type);
 		break;
 	case SUM_TYPE_DATA:
-		ret = gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
+		gc_data_segment(sbi, sum->entries, ilist, segno, gc_type);
 		break;
 	}
 	stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)));
 	stat_inc_call_count(sbi->stat_info);
 
 	f2fs_put_page(sum_page, 0);
-	return ret;
 }
 
 int f2fs_gc(struct f2fs_sb_info *sbi)
@@ -664,40 +641,38 @@ int f2fs_gc(struct f2fs_sb_info *sbi)
 	struct list_head ilist;
 	unsigned int segno, i;
 	int gc_type = BG_GC;
-	int gc_status = GC_NONE;
+	int nfree = 0;
+	int ret = -1;
 
 	INIT_LIST_HEAD(&ilist);
 gc_more:
 	if (!(sbi->sb->s_flags & MS_ACTIVE))
 		goto stop;
 
-	if (gc_type == BG_GC && has_not_enough_free_secs(sbi))
+	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, nfree))
 		gc_type = FG_GC;
 
 	if (!__get_victim(sbi, &segno, gc_type, NO_CHECK_TYPE))
 		goto stop;
+	ret = 0;
 
-	for (i = 0; i < sbi->segs_per_sec; i++) {
-		/*
-		 * do_garbage_collect will give us three gc_status:
-		 * GC_ERROR, GC_DONE, and GC_BLOCKED.
-		 * If GC is finished uncleanly, we have to return
-		 * the victim to dirty segment list.
-		 */
-		gc_status = do_garbage_collect(sbi, segno + i, &ilist, gc_type);
-		if (gc_status != GC_DONE)
-			break;
-	}
-	if (has_not_enough_free_secs(sbi)) {
-		write_checkpoint(sbi, (gc_status == GC_BLOCKED), false);
-		if (has_not_enough_free_secs(sbi))
-			goto gc_more;
-	}
+	for (i = 0; i < sbi->segs_per_sec; i++)
+		do_garbage_collect(sbi, segno + i, &ilist, gc_type);
+
+	if (gc_type == FG_GC &&
+			get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
+		nfree++;
+
+	if (has_not_enough_free_secs(sbi, nfree))
+		goto gc_more;
+
+	if (gc_type == FG_GC)
+		write_checkpoint(sbi, false);
 stop:
 	mutex_unlock(&sbi->gc_mutex);
 
 	put_gc_inode(&ilist);
-	return gc_status;
+	return ret;
 }
 
 void build_gc_manager(struct f2fs_sb_info *sbi)

fs/f2fs/gc.h

@@ -22,15 +22,6 @@
 /* Search max. number of dirty segments to select a victim segment */
 #define MAX_VICTIM_SEARCH	20
 
-enum {
-	GC_NONE = 0,
-	GC_ERROR,
-	GC_OK,
-	GC_NEXT,
-	GC_BLOCKED,
-	GC_DONE,
-};
-
 struct f2fs_gc_kthread {
 	struct task_struct *f2fs_gc_task;
 	wait_queue_head_t gc_wait_queue_head;
@@ -103,10 +94,3 @@ static inline int is_idle(struct f2fs_sb_info *sbi)
 	struct request_list *rl = &q->root_rl;
 	return !(rl->count[BLK_RW_SYNC]) && !(rl->count[BLK_RW_ASYNC]);
 }
-
-static inline bool should_do_checkpoint(struct f2fs_sb_info *sbi)
-{
-	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
-	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
-	return free_sections(sbi) <= (node_secs + 2 * dent_secs + 2);
-}

fs/f2fs/node.c

@@ -1135,7 +1135,7 @@ static int f2fs_write_node_pages(struct address_space *mapping,
 
 	/* First check balancing cached NAT entries */
 	if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
-		write_checkpoint(sbi, false, false);
+		write_checkpoint(sbi, false);
 		return 0;
 	}
 

fs/f2fs/recovery.c

@@ -373,5 +373,5 @@ void recover_fsync_data(struct f2fs_sb_info *sbi)
 out:
 	destroy_fsync_dnodes(sbi, &inode_list);
 	kmem_cache_destroy(fsync_entry_slab);
-	write_checkpoint(sbi, false, false);
+	write_checkpoint(sbi, false);
 }

fs/f2fs/segment.c

@@ -29,7 +29,7 @@ void f2fs_balance_fs(struct f2fs_sb_info *sbi)
 	 * We should do GC or end up with checkpoint, if there are so many dirty
 	 * dir/node pages without enough free segments.
 	 */
-	if (has_not_enough_free_secs(sbi)) {
+	if (has_not_enough_free_secs(sbi, 0)) {
 		mutex_lock(&sbi->gc_mutex);
 		f2fs_gc(sbi);
 	}
@@ -308,7 +308,7 @@ static unsigned int check_prefree_segments(struct f2fs_sb_info *sbi,
 	 * If there is not enough reserved sections,
 	 * we should not reuse prefree segments.
 	 */
-	if (has_not_enough_free_secs(sbi))
+	if (has_not_enough_free_secs(sbi, 0))
 		return NULL_SEGNO;
 
 	/*
@@ -536,6 +536,23 @@ static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
 	}
 }
 
+static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
+{
+	struct curseg_info *curseg = CURSEG_I(sbi, type);
+	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
+
+	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
+		return v_ops->get_victim(sbi,
+				&(curseg)->next_segno, BG_GC, type, SSR);
+
+	/* For data segments, let's do SSR more intensively */
+	for (; type >= CURSEG_HOT_DATA; type--)
+		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
+						BG_GC, type, SSR))
+			return 1;
+	return 0;
+}
+
 /*
  * flush out current segment and replace it with new segment
  * This function should be returned with success, otherwise BUG

fs/f2fs/segment.h

@@ -450,21 +450,15 @@ static inline bool need_SSR(struct f2fs_sb_info *sbi)
 	return (free_sections(sbi) < overprovision_sections(sbi));
 }
 
-static inline int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
-{
-	struct curseg_info *curseg = CURSEG_I(sbi, type);
-	return DIRTY_I(sbi)->v_ops->get_victim(sbi,
-				&(curseg)->next_segno, BG_GC, type, SSR);
-}
-
-static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi)
+static inline bool has_not_enough_free_secs(struct f2fs_sb_info *sbi, int freed)
 {
 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
+
 	if (sbi->por_doing)
 		return false;
 
-	return (free_sections(sbi) <= (node_secs + 2 * dent_secs +
+	return ((free_sections(sbi) + freed) <= (node_secs + 2 * dent_secs +
 						reserved_sections(sbi)));
 }
 

fs/f2fs/super.c

@@ -112,7 +112,7 @@ static void f2fs_put_super(struct super_block *sb)
 	f2fs_destroy_stats(sbi);
 	stop_gc_thread(sbi);
 
-	write_checkpoint(sbi, false, true);
+	write_checkpoint(sbi, true);
 
 	iput(sbi->node_inode);
 	iput(sbi->meta_inode);
@@ -136,7 +136,7 @@ int f2fs_sync_fs(struct super_block *sb, int sync)
 		return 0;
 
 	if (sync)
-		write_checkpoint(sbi, false, false);
+		write_checkpoint(sbi, false);
 	else
 		f2fs_balance_fs(sbi);