f2fs: add ioctl to flush data from faster device to cold area

This patch adds an ioctl to flush data in faster device to cold area. User can
give device number and number of segments to move. It doesn't move it if there
is only one device.

The parameter looks like:

struct f2fs_flush_device {
	u32 dev_num;		/* device number to flush */
	u32 segments;		/* # of segments to flush */
};

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

fs/f2fs/f2fs.h

@@ -300,6 +300,8 @@ static inline bool __has_cursum_space(struct f2fs_journal *journal,
 #define F2FS_IOC_DEFRAGMENT		_IO(F2FS_IOCTL_MAGIC, 8)
 #define F2FS_IOC_MOVE_RANGE		_IOWR(F2FS_IOCTL_MAGIC, 9,	\
 						struct f2fs_move_range)
+#define F2FS_IOC_FLUSH_DEVICE		_IOW(F2FS_IOCTL_MAGIC, 10,	\
+						struct f2fs_flush_device)
 
 #define F2FS_IOC_SET_ENCRYPTION_POLICY	FS_IOC_SET_ENCRYPTION_POLICY
 #define F2FS_IOC_GET_ENCRYPTION_POLICY	FS_IOC_GET_ENCRYPTION_POLICY
@@ -336,6 +338,11 @@ struct f2fs_move_range {
 	u64 len;		/* size to move */
 };
 
+struct f2fs_flush_device {
+	u32 dev_num;		/* device number to flush */
+	u32 segments;		/* # of segments to flush */
+};
+
 /*
  * For INODE and NODE manager
  */
@@ -980,7 +987,6 @@ struct f2fs_sb_info {
 	int bg_gc;				/* background gc calls */
 	unsigned int ndirty_inode[NR_INODE_TYPE];	/* # of dirty inodes */
 #endif
-	unsigned int last_victim[2];		/* last victim segment # */
 	spinlock_t stat_lock;			/* lock for stat operations */
 
 	/* For sysfs suppport */
@@ -2362,7 +2368,8 @@ int f2fs_migrate_page(struct address_space *mapping, struct page *newpage,
 int start_gc_thread(struct f2fs_sb_info *sbi);
 void stop_gc_thread(struct f2fs_sb_info *sbi);
 block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode);
-int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background);
+int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background,
+			unsigned int segno);
 void build_gc_manager(struct f2fs_sb_info *sbi);
 
 /*

fs/f2fs/file.c

@@ -1855,7 +1855,7 @@ static int f2fs_ioc_gc(struct file *filp, unsigned long arg)
 		mutex_lock(&sbi->gc_mutex);
 	}
 
-	ret = f2fs_gc(sbi, sync, true);
+	ret = f2fs_gc(sbi, sync, true, NULL_SEGNO);
 out:
 	mnt_drop_write_file(filp);
 	return ret;
@@ -2211,6 +2211,69 @@ err_out:
 	return err;
 }
 
+static int f2fs_ioc_flush_device(struct file *filp, unsigned long arg)
+{
+	struct inode *inode = file_inode(filp);
+	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+	struct sit_info *sm = SIT_I(sbi);
+	unsigned int start_segno = 0, end_segno = 0;
+	unsigned int dev_start_segno = 0, dev_end_segno = 0;
+	struct f2fs_flush_device range;
+	int ret;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (f2fs_readonly(sbi->sb))
+		return -EROFS;
+
+	if (copy_from_user(&range, (struct f2fs_flush_device __user *)arg,
+							sizeof(range)))
+		return -EFAULT;
+
+	if (sbi->s_ndevs <= 1 || sbi->s_ndevs - 1 <= range.dev_num ||
+			sbi->segs_per_sec != 1) {
+		f2fs_msg(sbi->sb, KERN_WARNING,
+			"Can't flush %u in %d for segs_per_sec %u != 1\n",
+				range.dev_num, sbi->s_ndevs,
+				sbi->segs_per_sec);
+		return -EINVAL;
+	}
+
+	ret = mnt_want_write_file(filp);
+	if (ret)
+		return ret;
+
+	if (range.dev_num != 0)
+		dev_start_segno = GET_SEGNO(sbi, FDEV(range.dev_num).start_blk);
+	dev_end_segno = GET_SEGNO(sbi, FDEV(range.dev_num).end_blk);
+
+	start_segno = sm->last_victim[FLUSH_DEVICE];
+	if (start_segno < dev_start_segno || start_segno >= dev_end_segno)
+		start_segno = dev_start_segno;
+	end_segno = min(start_segno + range.segments, dev_end_segno);
+
+	while (start_segno < end_segno) {
+		if (!mutex_trylock(&sbi->gc_mutex)) {
+			ret = -EBUSY;
+			goto out;
+		}
+		sm->last_victim[GC_CB] = end_segno + 1;
+		sm->last_victim[GC_GREEDY] = end_segno + 1;
+		sm->last_victim[ALLOC_NEXT] = end_segno + 1;
+		ret = f2fs_gc(sbi, true, true, start_segno);
+		if (ret == -EAGAIN)
+			ret = 0;
+		else if (ret < 0)
+			break;
+		start_segno++;
+	}
+out:
+	mnt_drop_write_file(filp);
+	return ret;
+}
+
+
 long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
 	switch (cmd) {
@@ -2248,6 +2311,8 @@ long f2fs_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 		return f2fs_ioc_defragment(filp, arg);
 	case F2FS_IOC_MOVE_RANGE:
 		return f2fs_ioc_move_range(filp, arg);
+	case F2FS_IOC_FLUSH_DEVICE:
+		return f2fs_ioc_flush_device(filp, arg);
 	default:
 		return -ENOTTY;
 	}
@@ -2315,8 +2380,8 @@ long f2fs_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 	case F2FS_IOC_GARBAGE_COLLECT:
 	case F2FS_IOC_WRITE_CHECKPOINT:
 	case F2FS_IOC_DEFRAGMENT:
-		break;
 	case F2FS_IOC_MOVE_RANGE:
+	case F2FS_IOC_FLUSH_DEVICE:
 		break;
 	default:
 		return -ENOIOCTLCMD;

fs/f2fs/gc.c

@@ -84,7 +84,7 @@ static int gc_thread_func(void *data)
 		stat_inc_bggc_count(sbi);
 
 		/* if return value is not zero, no victim was selected */
-		if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true))
+		if (f2fs_gc(sbi, test_opt(sbi, FORCE_FG_GC), true, NULL_SEGNO))
 			wait_ms = gc_th->no_gc_sleep_time;
 
 		trace_f2fs_background_gc(sbi->sb, wait_ms,
@@ -176,7 +176,7 @@ static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 	if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
 		p->offset = 0;
 	else
-		p->offset = sbi->last_victim[p->gc_mode];
+		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 }
 
 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
@@ -295,6 +295,7 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
 		unsigned int *result, int gc_type, int type, char alloc_mode)
 {
 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
+	struct sit_info *sm = SIT_I(sbi);
 	struct victim_sel_policy p;
 	unsigned int secno, last_victim;
 	unsigned int last_segment = MAIN_SEGS(sbi);
@@ -308,10 +309,18 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
 	p.min_segno = NULL_SEGNO;
 	p.min_cost = get_max_cost(sbi, &p);
 
+	if (*result != NULL_SEGNO) {
+		if (IS_DATASEG(get_seg_entry(sbi, *result)->type) &&
+			get_valid_blocks(sbi, *result, false) &&
+			!sec_usage_check(sbi, GET_SEC_FROM_SEG(sbi, *result)))
+			p.min_segno = *result;
+		goto out;
+	}
+
 	if (p.max_search == 0)
 		goto out;
 
-	last_victim = sbi->last_victim[p.gc_mode];
+	last_victim = sm->last_victim[p.gc_mode];
 	if (p.alloc_mode == LFS && gc_type == FG_GC) {
 		p.min_segno = check_bg_victims(sbi);
 		if (p.min_segno != NULL_SEGNO)
@@ -324,9 +333,10 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
 
 		segno = find_next_bit(p.dirty_segmap, last_segment, p.offset);
 		if (segno >= last_segment) {
-			if (sbi->last_victim[p.gc_mode]) {
-				last_segment = sbi->last_victim[p.gc_mode];
-				sbi->last_victim[p.gc_mode] = 0;
+			if (sm->last_victim[p.gc_mode]) {
+				last_segment =
+					sm->last_victim[p.gc_mode];
+				sm->last_victim[p.gc_mode] = 0;
 				p.offset = 0;
 				continue;
 			}
@@ -361,11 +371,11 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
 		}
 next:
 		if (nsearched >= p.max_search) {
-			if (!sbi->last_victim[p.gc_mode] && segno <= last_victim)
-				sbi->last_victim[p.gc_mode] = last_victim + 1;
+			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
+				sm->last_victim[p.gc_mode] = last_victim + 1;
 			else
-				sbi->last_victim[p.gc_mode] = segno + 1;
-			sbi->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
+				sm->last_victim[p.gc_mode] = segno + 1;
+			sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
 			break;
 		}
 	}
@@ -912,7 +922,6 @@ static int do_garbage_collect(struct f2fs_sb_info *sbi,
 		 *   - mutex_lock(sentry_lock)     - change_curseg()
 		 *                                  - lock_page(sum_page)
 		 */
-
 		if (type == SUM_TYPE_NODE)
 			gc_node_segment(sbi, sum->entries, segno, gc_type);
 		else
@@ -939,13 +948,14 @@ next:
 	return sec_freed;
 }
 
-int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background)
+int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
+			bool background, unsigned int segno)
 {
-	unsigned int segno;
 	int gc_type = sync ? FG_GC : BG_GC;
 	int sec_freed = 0;
 	int ret = -EINVAL;
 	struct cp_control cpc;
+	unsigned int init_segno = segno;
 	struct gc_inode_list gc_list = {
 		.ilist = LIST_HEAD_INIT(gc_list.ilist),
 		.iroot = RADIX_TREE_INIT(GFP_NOFS),
@@ -990,13 +1000,17 @@ gc_more:
 		sbi->cur_victim_sec = NULL_SEGNO;
 
 	if (!sync) {
-		if (has_not_enough_free_secs(sbi, sec_freed, 0))
+		if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
+			segno = NULL_SEGNO;
 			goto gc_more;
+		}
 
 		if (gc_type == FG_GC)
 			ret = write_checkpoint(sbi, &cpc);
 	}
 stop:
+	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
+	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
 	mutex_unlock(&sbi->gc_mutex);
 
 	put_gc_inode(&gc_list);

fs/f2fs/segment.c

@@ -401,7 +401,7 @@ void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 	 */
 	if (has_not_enough_free_secs(sbi, 0, 0)) {
 		mutex_lock(&sbi->gc_mutex);
-		f2fs_gc(sbi, false, false);
+		f2fs_gc(sbi, false, false, NULL_SEGNO);
 	}
 }
 
@@ -1755,6 +1755,8 @@ static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
 	if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
 		return 0;
 
+	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
+		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
 	return CURSEG_I(sbi, type)->segno;
 }
 
@@ -1852,12 +1854,15 @@ 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;
+	unsigned segno = NULL_SEGNO;
 	int i, cnt;
 	bool reversed = false;
 
 	/* need_SSR() already forces to do this */
-	if (v_ops->get_victim(sbi, &(curseg)->next_segno, BG_GC, type, SSR))
+	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
+		curseg->next_segno = segno;
 		return 1;
+	}
 
 	/* For node segments, let's do SSR more intensively */
 	if (IS_NODESEG(type)) {
@@ -1881,9 +1886,10 @@ static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
 	for (; cnt-- > 0; reversed ? i-- : i++) {
 		if (i == type)
 			continue;
-		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
-						BG_GC, i, SSR))
+		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
+			curseg->next_segno = segno;
 			return 1;
+		}
 	}
 	return 0;
 }

fs/f2fs/segment.h

@@ -138,7 +138,10 @@ enum {
  */
 enum {
 	GC_CB = 0,
-	GC_GREEDY
+	GC_GREEDY,
+	ALLOC_NEXT,
+	FLUSH_DEVICE,
+	MAX_GC_POLICY,
 };
 
 /*
@@ -233,6 +236,8 @@ struct sit_info {
 	unsigned long long mounted_time;	/* mount time */
 	unsigned long long min_mtime;		/* min. modification time */
 	unsigned long long max_mtime;		/* max. modification time */
+
+	unsigned int last_victim[MAX_GC_POLICY]; /* last victim segment # */
 };
 
 struct free_segmap_info {