1 files changed, 397 insertions, 187 deletions

diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index fedbf67a0842..be9fd616736b 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -16,7 +16,6 @@
 #include <linux/kthread.h>
 #include <linux/delay.h>
 #include <linux/freezer.h>
-#include <linux/blkdev.h>
 
 #include "f2fs.h"
 #include "node.h"
@@ -29,17 +28,23 @@ static int gc_thread_func(void *data)
 	struct f2fs_sb_info *sbi = data;
 	struct f2fs_gc_kthread *gc_th = sbi->gc_thread;
 	wait_queue_head_t *wq = &sbi->gc_thread->gc_wait_queue_head;
-	long wait_ms;
+	unsigned int wait_ms;
 
 	wait_ms = gc_th->min_sleep_time;
 
+	set_freezable();
 	do {
+		wait_event_interruptible_timeout(*wq,
+				kthread_should_stop() || freezing(current) ||
+				gc_th->gc_wake,
+				msecs_to_jiffies(wait_ms));
+
+		/* give it a try one time */
+		if (gc_th->gc_wake)
+			gc_th->gc_wake = 0;
+
 		if (try_to_freeze())
 			continue;
-		else
-			wait_event_interruptible_timeout(*wq,
-						kthread_should_stop(),
-						msecs_to_jiffies(wait_ms));
 		if (kthread_should_stop())
 			break;
 
@@ -48,6 +53,16 @@ static int gc_thread_func(void *data)
 			continue;
 		}
 
+#ifdef CONFIG_F2FS_FAULT_INJECTION
+		if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
+			f2fs_show_injection_info(FAULT_CHECKPOINT);
+			f2fs_stop_checkpoint(sbi, false);
+		}
+#endif
+
+		if (!sb_start_write_trylock(sbi->sb))
+			continue;
+
 		/*
 		 * [GC triggering condition]
 		 * 0. GC is not conducted currently.
@@ -62,23 +77,28 @@ static int gc_thread_func(void *data)
 		 * So, I'd like to wait some time to collect dirty segments.
 		 */
 		if (!mutex_trylock(&sbi->gc_mutex))
-			continue;
+			goto next;
+
+		if (gc_th->gc_urgent) {
+			wait_ms = gc_th->urgent_sleep_time;
+			goto do_gc;
+		}
 
 		if (!is_idle(sbi)) {
 			increase_sleep_time(gc_th, &wait_ms);
 			mutex_unlock(&sbi->gc_mutex);
-			continue;
+			goto next;
 		}
 
 		if (has_enough_invalid_blocks(sbi))
 			decrease_sleep_time(gc_th, &wait_ms);
 		else
 			increase_sleep_time(gc_th, &wait_ms);
-
+do_gc:
 		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)))
+		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,
@@ -86,6 +106,8 @@ static int gc_thread_func(void *data)
 
 		/* balancing f2fs's metadata periodically */
 		f2fs_balance_fs_bg(sbi);
+next:
+		sb_end_write(sbi->sb);
 
 	} while (!kthread_should_stop());
 	return 0;
@@ -97,17 +119,20 @@ int start_gc_thread(struct f2fs_sb_info *sbi)
 	dev_t dev = sbi->sb->s_bdev->bd_dev;
 	int err = 0;
 
-	gc_th = kmalloc(sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
+	gc_th = f2fs_kmalloc(sbi, sizeof(struct f2fs_gc_kthread), GFP_KERNEL);
 	if (!gc_th) {
 		err = -ENOMEM;
 		goto out;
 	}
 
+	gc_th->urgent_sleep_time = DEF_GC_THREAD_URGENT_SLEEP_TIME;
 	gc_th->min_sleep_time = DEF_GC_THREAD_MIN_SLEEP_TIME;
 	gc_th->max_sleep_time = DEF_GC_THREAD_MAX_SLEEP_TIME;
 	gc_th->no_gc_sleep_time = DEF_GC_THREAD_NOGC_SLEEP_TIME;
 
 	gc_th->gc_idle = 0;
+	gc_th->gc_urgent = 0;
+	gc_th->gc_wake= 0;
 
 	sbi->gc_thread = gc_th;
 	init_waitqueue_head(&sbi->gc_thread->gc_wait_queue_head);
@@ -162,10 +187,15 @@ static void select_policy(struct f2fs_sb_info *sbi, int gc_type,
 		p->ofs_unit = sbi->segs_per_sec;
 	}
 
-	if (p->max_search > sbi->max_victim_search)
+	/* we need to check every dirty segments in the FG_GC case */
+	if (gc_type != FG_GC && p->max_search > sbi->max_victim_search)
 		p->max_search = sbi->max_victim_search;
 
-	p->offset = sbi->last_victim[p->gc_mode];
+	/* let's select beginning hot/small space first */
+	if (type == CURSEG_HOT_DATA || IS_NODESEG(type))
+		p->offset = 0;
+	else
+		p->offset = SIT_I(sbi)->last_victim[p->gc_mode];
 }
 
 static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
@@ -173,9 +203,9 @@ static unsigned int get_max_cost(struct f2fs_sb_info *sbi,
 {
 	/* SSR allocates in a segment unit */
 	if (p->alloc_mode == SSR)
-		return 1 << sbi->log_blocks_per_seg;
+		return sbi->blocks_per_seg;
 	if (p->gc_mode == GC_GREEDY)
-		return (1 << sbi->log_blocks_per_seg) * p->ofs_unit;
+		return 2 * sbi->blocks_per_seg * p->ofs_unit;
 	else if (p->gc_mode == GC_CB)
 		return UINT_MAX;
 	else /* No other gc_mode */
@@ -195,8 +225,12 @@ static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 	for_each_set_bit(secno, dirty_i->victim_secmap, MAIN_SECS(sbi)) {
 		if (sec_usage_check(sbi, secno))
 			continue;
+
+		if (no_fggc_candidate(sbi, secno))
+			continue;
+
 		clear_bit(secno, dirty_i->victim_secmap);
-		return secno * sbi->segs_per_sec;
+		return GET_SEG_FROM_SEC(sbi, secno);
 	}
 	return NULL_SEGNO;
 }
@@ -204,8 +238,8 @@ static unsigned int check_bg_victims(struct f2fs_sb_info *sbi)
 static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 {
 	struct sit_info *sit_i = SIT_I(sbi);
-	unsigned int secno = GET_SECNO(sbi, segno);
-	unsigned int start = secno * sbi->segs_per_sec;
+	unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
+	unsigned int start = GET_SEG_FROM_SEC(sbi, secno);
 	unsigned long long mtime = 0;
 	unsigned int vblocks;
 	unsigned char age = 0;
@@ -214,7 +248,7 @@ static unsigned int get_cb_cost(struct f2fs_sb_info *sbi, unsigned int segno)
 
 	for (i = 0; i < sbi->segs_per_sec; i++)
 		mtime += get_seg_entry(sbi, start + i)->mtime;
-	vblocks = get_valid_blocks(sbi, segno, sbi->segs_per_sec);
+	vblocks = get_valid_blocks(sbi, segno, true);
 
 	mtime = div_u64(mtime, sbi->segs_per_sec);
 	vblocks = div_u64(vblocks, sbi->segs_per_sec);
@@ -241,11 +275,23 @@ static inline unsigned int get_gc_cost(struct f2fs_sb_info *sbi,
 
 	/* alloc_mode == LFS */
 	if (p->gc_mode == GC_GREEDY)
-		return get_valid_blocks(sbi, segno, sbi->segs_per_sec);
+		return get_valid_blocks(sbi, segno, true);
 	else
 		return get_cb_cost(sbi, segno);
 }
 
+static unsigned int count_bits(const unsigned long *addr,
+				unsigned int offset, unsigned int len)
+{
+	unsigned int end = offset + len, sum = 0;
+
+	while (offset < end) {
+		if (test_bit(offset++, addr))
+			++sum;
+	}
+	return sum;
+}
+
 /*
  * This function is called from two paths.
  * One is garbage collection and the other is SSR segment selection.
@@ -258,10 +304,11 @@ 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, max_cost;
+	unsigned int secno, last_victim;
 	unsigned int last_segment = MAIN_SEGS(sbi);
-	int nsearched = 0;
+	unsigned int nsearched = 0;
 
 	mutex_lock(&dirty_i->seglist_lock);
 
@@ -269,11 +316,20 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
 	select_policy(sbi, gc_type, type, &p);
 
 	p.min_segno = NULL_SEGNO;
-	p.min_cost = max_cost = get_max_cost(sbi, &p);
+	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 = 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)
@@ -286,9 +342,10 @@ 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;
 			}
@@ -296,34 +353,45 @@ static int get_victim_by_default(struct f2fs_sb_info *sbi,
 		}
 
 		p.offset = segno + p.ofs_unit;
-		if (p.ofs_unit > 1)
+		if (p.ofs_unit > 1) {
 			p.offset -= segno % p.ofs_unit;
+			nsearched += count_bits(p.dirty_segmap,
+						p.offset - p.ofs_unit,
+						p.ofs_unit);
+		} else {
+			nsearched++;
+		}
 
-		secno = GET_SECNO(sbi, segno);
+		secno = GET_SEC_FROM_SEG(sbi, segno);
 
 		if (sec_usage_check(sbi, secno))
-			continue;
+			goto next;
 		if (gc_type == BG_GC && test_bit(secno, dirty_i->victim_secmap))
-			continue;
+			goto next;
+		if (gc_type == FG_GC && p.alloc_mode == LFS &&
+					no_fggc_candidate(sbi, secno))
+			goto next;
 
 		cost = get_gc_cost(sbi, segno, &p);
 
 		if (p.min_cost > cost) {
 			p.min_segno = segno;
 			p.min_cost = cost;
-		} else if (unlikely(cost == max_cost)) {
-			continue;
 		}
-
-		if (nsearched++ >= p.max_search) {
-			sbi->last_victim[p.gc_mode] = segno;
+next:
+		if (nsearched >= p.max_search) {
+			if (!sm->last_victim[p.gc_mode] && segno <= last_victim)
+				sm->last_victim[p.gc_mode] = last_victim + 1;
+			else
+				sm->last_victim[p.gc_mode] = segno + 1;
+			sm->last_victim[p.gc_mode] %= MAIN_SEGS(sbi);
 			break;
 		}
 	}
 	if (p.min_segno != NULL_SEGNO) {
 got_it:
 		if (p.alloc_mode == LFS) {
-			secno = GET_SECNO(sbi, p.min_segno);
+			secno = GET_SEC_FROM_SEG(sbi, p.min_segno);
 			if (gc_type == FG_GC)
 				sbi->cur_victim_sec = secno;
 			else
@@ -388,10 +456,10 @@ static int check_valid_map(struct f2fs_sb_info *sbi,
 	struct seg_entry *sentry;
 	int ret;
 
-	mutex_lock(&sit_i->sentry_lock);
+	down_read(&sit_i->sentry_lock);
 	sentry = get_seg_entry(sbi, segno);
 	ret = f2fs_test_bit(offset, sentry->cur_valid_map);
-	mutex_unlock(&sit_i->sentry_lock);
+	up_read(&sit_i->sentry_lock);
 	return ret;
 }
 
@@ -400,13 +468,13 @@ 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;
 	struct f2fs_summary *entry;
 	block_t start_addr;
 	int off;
+	int phase = 0;
 
 	start_addr = START_BLOCK(sbi, segno);
 
@@ -419,16 +487,24 @@ next_step:
 		struct node_info ni;
 
 		/* stop BG_GC if there is not enough free sections. */
-		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
-			return 0;
+		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
+			return;
 
 		if (check_valid_map(sbi, segno, off) == 0)
 			continue;
 
-		if (initial) {
+		if (phase == 0) {
+			ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
+							META_NAT, true);
+			continue;
+		}
+
+		if (phase == 1) {
 			ra_node_page(sbi, nid);
 			continue;
 		}
+
+		/* phase == 2 */
 		node_page = get_node_page(sbi, nid);
 		if (IS_ERR(node_page))
 			continue;
@@ -445,36 +521,12 @@ next_step:
 			continue;
 		}
 
-		/* set page dirty and write it */
-		if (gc_type == FG_GC) {
-			f2fs_wait_on_page_writeback(node_page, NODE);
-			set_page_dirty(node_page);
-		} else {
-			if (!PageWriteback(node_page))
-				set_page_dirty(node_page);
-		}
-		f2fs_put_page(node_page, 1);
+		move_node_page(node_page, gc_type);
 		stat_inc_node_blk_count(sbi, 1, gc_type);
 	}
 
-	if (initial) {
-		initial = false;
+	if (++phase < 3)
 		goto next_step;
-	}
-
-	if (gc_type == FG_GC) {
-		struct writeback_control wbc = {
-			.sync_mode = WB_SYNC_ALL,
-			.nr_to_write = LONG_MAX,
-			.for_reclaim = 0,
-		};
-		sync_node_pages(sbi, 0, &wbc);
-
-		/* return 1 only if FG_GC succefully reclaimed one */
-		if (get_valid_blocks(sbi, segno, 1) == 0)
-			return 1;
-	}
-	return 0;
 }
 
 /*
@@ -484,7 +536,7 @@ next_step:
  * as indirect or double indirect node blocks, are given, it must be a caller's
  * bug.
  */
-block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
+block_t start_bidx_of_node(unsigned int node_ofs, struct inode *inode)
 {
 	unsigned int indirect_blks = 2 * NIDS_PER_BLOCK + 4;
 	unsigned int bidx;
@@ -501,7 +553,7 @@ block_t start_bidx_of_node(unsigned int node_ofs, struct f2fs_inode_info *fi)
 		int dec = (node_ofs - indirect_blks - 3) / (NIDS_PER_BLOCK + 1);
 		bidx = node_ofs - 5 - dec;
 	}
-	return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(fi);
+	return bidx * ADDRS_PER_BLOCK + ADDRS_PER_INODE(inode);
 }
 
 static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
@@ -522,12 +574,14 @@ static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 	get_node_info(sbi, nid, dni);
 
 	if (sum->version != dni->version) {
-		f2fs_put_page(node_page, 1);
-		return false;
+		f2fs_msg(sbi->sb, KERN_WARNING,
+				"%s: valid data with mismatched node version.",
+				__func__);
+		set_sbi_flag(sbi, SBI_NEED_FSCK);
 	}
 
 	*nofs = ofs_of_node(node_page);
-	source_blkaddr = datablock_addr(node_page, ofs_in_node);
+	source_blkaddr = datablock_addr(NULL, node_page, ofs_in_node);
 	f2fs_put_page(node_page, 1);
 
 	if (source_blkaddr != blkaddr)
@@ -535,18 +589,28 @@ static bool is_alive(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
 	return true;
 }
 
-static void move_encrypted_block(struct inode *inode, block_t bidx)
+/*
+ * Move data block via META_MAPPING while keeping locked data page.
+ * This can be used to move blocks, aka LBAs, directly on disk.
+ */
+static void move_data_block(struct inode *inode, block_t bidx,
+					unsigned int segno, int off)
 {
 	struct f2fs_io_info fio = {
 		.sbi = F2FS_I_SB(inode),
+		.ino = inode->i_ino,
 		.type = DATA,
-		.rw = READ_SYNC,
+		.temp = COLD,
+		.op = REQ_OP_READ,
+		.op_flags = REQ_SYNC,
 		.encrypted_page = NULL,
+		.in_list = false,
 	};
 	struct dnode_of_data dn;
 	struct f2fs_summary sum;
 	struct node_info ni;
 	struct page *page;
+	block_t newaddr;
 	int err;
 
 	/* do not read out */
@@ -554,6 +618,12 @@ static void move_encrypted_block(struct inode *inode, block_t bidx)
 	if (!page)
 		return;
 
+	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
+		goto out;
+
+	if (f2fs_is_atomic_file(inode))
+		goto out;
+
 	set_new_dnode(&dn, inode, NULL, NULL, 0);
 	err = get_dnode_of_data(&dn, bidx, LOOKUP_NODE);
 	if (err)
@@ -568,21 +638,24 @@ static void move_encrypted_block(struct inode *inode, block_t bidx)
 	 * don't cache encrypted data into meta inode until previous dirty
 	 * data were writebacked to avoid racing between GC and flush.
 	 */
-	f2fs_wait_on_page_writeback(page, DATA);
+	f2fs_wait_on_page_writeback(page, DATA, true);
 
 	get_node_info(fio.sbi, dn.nid, &ni);
 	set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 
 	/* read page */
 	fio.page = page;
-	fio.blk_addr = dn.data_blkaddr;
+	fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr;
 
-	fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi),
-					fio.blk_addr,
-					FGP_LOCK|FGP_CREAT,
-					GFP_NOFS);
-	if (!fio.encrypted_page)
-		goto put_out;
+	allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr,
+					&sum, CURSEG_COLD_DATA, NULL, false);
+
+	fio.encrypted_page = f2fs_pagecache_get_page(META_MAPPING(fio.sbi),
+				newaddr, FGP_LOCK | FGP_CREAT, GFP_NOFS);
+	if (!fio.encrypted_page) {
+		err = -ENOMEM;
+		goto recover_block;
+	}
 
 	err = f2fs_submit_page_bio(&fio);
 	if (err)
@@ -591,40 +664,50 @@ static void move_encrypted_block(struct inode *inode, block_t bidx)
 	/* write page */
 	lock_page(fio.encrypted_page);
 
-	if (unlikely(!PageUptodate(fio.encrypted_page)))
+	if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi))) {
+		err = -EIO;
 		goto put_page_out;
-	if (unlikely(fio.encrypted_page->mapping != META_MAPPING(fio.sbi)))
+	}
+	if (unlikely(!PageUptodate(fio.encrypted_page))) {
+		err = -EIO;
 		goto put_page_out;
+	}
 
 	set_page_dirty(fio.encrypted_page);
-	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA);
+	f2fs_wait_on_page_writeback(fio.encrypted_page, DATA, true);
 	if (clear_page_dirty_for_io(fio.encrypted_page))
 		dec_page_count(fio.sbi, F2FS_DIRTY_META);
 
 	set_page_writeback(fio.encrypted_page);
 
 	/* allocate block address */
-	f2fs_wait_on_page_writeback(dn.node_page, NODE);
-	allocate_data_block(fio.sbi, NULL, fio.blk_addr,
-					&fio.blk_addr, &sum, CURSEG_COLD_DATA);
-	fio.rw = WRITE_SYNC;
-	f2fs_submit_page_mbio(&fio);
-
-	dn.data_blkaddr = fio.blk_addr;
-	set_data_blkaddr(&dn);
-	f2fs_update_extent_cache(&dn);
-	set_inode_flag(F2FS_I(inode), FI_APPEND_WRITE);
+	f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
+
+	fio.op = REQ_OP_WRITE;
+	fio.op_flags = REQ_SYNC | REQ_NOIDLE;
+	fio.new_blkaddr = newaddr;
+	f2fs_submit_page_write(&fio);
+
+	f2fs_update_iostat(fio.sbi, FS_GC_DATA_IO, F2FS_BLKSIZE);
+
+	f2fs_update_data_blkaddr(&dn, newaddr);
+	set_inode_flag(inode, FI_APPEND_WRITE);
 	if (page->index == 0)
-		set_inode_flag(F2FS_I(inode), FI_FIRST_BLOCK_WRITTEN);
+		set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 put_page_out:
 	f2fs_put_page(fio.encrypted_page, 1);
+recover_block:
+	if (err)
+		__f2fs_replace_block(fio.sbi, &sum, newaddr, fio.old_blkaddr,
+								true, true);
 put_out:
 	f2fs_put_dnode(&dn);
 out:
 	f2fs_put_page(page, 1);
 }
 
-static void move_data_page(struct inode *inode, block_t bidx, int gc_type)
+static void move_data_page(struct inode *inode, block_t bidx, int gc_type,
+							unsigned int segno, int off)
 {
 	struct page *page;
 
@@ -632,6 +715,12 @@ static void move_data_page(struct inode *inode, block_t bidx, int gc_type)
 	if (IS_ERR(page))
 		return;
 
+	if (!check_valid_map(F2FS_I_SB(inode), segno, off))
+		goto out;
+
+	if (f2fs_is_atomic_file(inode))
+		goto out;
+
 	if (gc_type == BG_GC) {
 		if (PageWriteback(page))
 			goto out;
@@ -640,18 +729,35 @@ static void move_data_page(struct inode *inode, block_t bidx, int gc_type)
 	} else {
 		struct f2fs_io_info fio = {
 			.sbi = F2FS_I_SB(inode),
+			.ino = inode->i_ino,
 			.type = DATA,
-			.rw = WRITE_SYNC,
+			.temp = COLD,
+			.op = REQ_OP_WRITE,
+			.op_flags = REQ_SYNC,
+			.old_blkaddr = NULL_ADDR,
 			.page = page,
 			.encrypted_page = NULL,
+			.need_lock = LOCK_REQ,
+			.io_type = FS_GC_DATA_IO,
 		};
+		bool is_dirty = PageDirty(page);
+		int err;
+
+retry:
 		set_page_dirty(page);
-		f2fs_wait_on_page_writeback(page, DATA);
-		if (clear_page_dirty_for_io(page))
+		f2fs_wait_on_page_writeback(page, DATA, true);
+		if (clear_page_dirty_for_io(page)) {
 			inode_dec_dirty_pages(inode);
+			remove_dirty_inode(inode);
+		}
+
 		set_cold_data(page);
-		do_write_data_page(&fio);
-		clear_cold_data(page);
+
+		err = do_write_data_page(&fio);
+		if (err == -ENOMEM && is_dirty) {
+			congestion_wait(BLK_RW_ASYNC, HZ/50);
+			goto retry;
+		}
 	}
 out:
 	f2fs_put_page(page, 1);
@@ -664,7 +770,7 @@ out:
  * 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 gc_inode_list *gc_list, unsigned int segno, int gc_type)
 {
 	struct super_block *sb = sbi->sb;
@@ -684,16 +790,23 @@ next_step:
 		struct node_info dni; /* dnode info for the data */
 		unsigned int ofs_in_node, nofs;
 		block_t start_bidx;
+		nid_t nid = le32_to_cpu(entry->nid);
 
 		/* stop BG_GC if there is not enough free sections. */
-		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0))
-			return 0;
+		if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0))
+			return;
 
 		if (check_valid_map(sbi, segno, off) == 0)
 			continue;
 
 		if (phase == 0) {
-			ra_node_page(sbi, le32_to_cpu(entry->nid));
+			ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), 1,
+							META_NAT, true);
+			continue;
+		}
+
+		if (phase == 1) {
+			ra_node_page(sbi, nid);
 			continue;
 		}
 
@@ -701,28 +814,35 @@ next_step:
 		if (!is_alive(sbi, entry, &dni, start_addr + off, &nofs))
 			continue;
 
-		if (phase == 1) {
+		if (phase == 2) {
 			ra_node_page(sbi, dni.ino);
 			continue;
 		}
 
 		ofs_in_node = le16_to_cpu(entry->ofs_in_node);
 
-		if (phase == 2) {
+		if (phase == 3) {
 			inode = f2fs_iget(sb, dni.ino);
 			if (IS_ERR(inode) || is_bad_inode(inode))
 				continue;
 
 			/* if encrypted inode, let's go phase 3 */
-			if (f2fs_encrypted_inode(inode) &&
-						S_ISREG(inode->i_mode)) {
+			if (f2fs_encrypted_file(inode)) {
 				add_gc_inode(gc_list, inode);
 				continue;
 			}
 
-			start_bidx = start_bidx_of_node(nofs, F2FS_I(inode));
+			if (!down_write_trylock(
+				&F2FS_I(inode)->dio_rwsem[WRITE])) {
+				iput(inode);
+				continue;
+			}
+
+			start_bidx = start_bidx_of_node(nofs, inode);
 			data_page = get_read_data_page(inode,
-					start_bidx + ofs_in_node, READA, true);
+					start_bidx + ofs_in_node, REQ_RAHEAD,
+					true);
+			up_write(&F2FS_I(inode)->dio_rwsem[WRITE]);
 			if (IS_ERR(data_page)) {
 				iput(inode);
 				continue;
@@ -733,30 +853,45 @@ next_step:
 			continue;
 		}
 
-		/* phase 3 */
+		/* phase 4 */
 		inode = find_gc_inode(gc_list, dni.ino);
 		if (inode) {
-			start_bidx = start_bidx_of_node(nofs, F2FS_I(inode))
+			struct f2fs_inode_info *fi = F2FS_I(inode);
+			bool locked = false;
+
+			if (S_ISREG(inode->i_mode)) {
+				if (!down_write_trylock(&fi->dio_rwsem[READ]))
+					continue;
+				if (!down_write_trylock(
+						&fi->dio_rwsem[WRITE])) {
+					up_write(&fi->dio_rwsem[READ]);
+					continue;
+				}
+				locked = true;
+
+				/* wait for all inflight aio data */
+				inode_dio_wait(inode);
+			}
+
+			start_bidx = start_bidx_of_node(nofs, inode)
 								+ ofs_in_node;
-			if (f2fs_encrypted_inode(inode) && S_ISREG(inode->i_mode))
-				move_encrypted_block(inode, start_bidx);
+			if (f2fs_encrypted_file(inode))
+				move_data_block(inode, start_bidx, segno, off);
 			else
-				move_data_page(inode, start_bidx, gc_type);
+				move_data_page(inode, start_bidx, gc_type,
+								segno, off);
+
+			if (locked) {
+				up_write(&fi->dio_rwsem[WRITE]);
+				up_write(&fi->dio_rwsem[READ]);
+			}
+
 			stat_inc_data_blk_count(sbi, 1, gc_type);
 		}
 	}
 
-	if (++phase < 4)
+	if (++phase < 5)
 		goto next_step;
-
-	if (gc_type == FG_GC) {
-		f2fs_submit_merged_bio(sbi, DATA, WRITE);
-
-		/* return 1 only if FG_GC succefully reclaimed one */
-		if (get_valid_blocks(sbi, segno, 1) == 0)
-			return 1;
-	}
-	return 0;
 }
 
 static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
@@ -765,115 +900,175 @@ static int __get_victim(struct f2fs_sb_info *sbi, unsigned int *victim,
 	struct sit_info *sit_i = SIT_I(sbi);
 	int ret;
 
-	mutex_lock(&sit_i->sentry_lock);
+	down_write(&sit_i->sentry_lock);
 	ret = DIRTY_I(sbi)->v_ops->get_victim(sbi, victim, gc_type,
 					      NO_CHECK_TYPE, LFS);
-	mutex_unlock(&sit_i->sentry_lock);
+	up_write(&sit_i->sentry_lock);
 	return ret;
 }
 
-static int do_garbage_collect(struct f2fs_sb_info *sbi, unsigned int segno,
+static int do_garbage_collect(struct f2fs_sb_info *sbi,
+				unsigned int start_segno,
 				struct gc_inode_list *gc_list, int gc_type)
 {
 	struct page *sum_page;
 	struct f2fs_summary_block *sum;
 	struct blk_plug plug;
-	int nfree = 0;
+	unsigned int segno = start_segno;
+	unsigned int end_segno = start_segno + sbi->segs_per_sec;
+	int seg_freed = 0;
+	unsigned char type = IS_DATASEG(get_seg_entry(sbi, segno)->type) ?
+						SUM_TYPE_DATA : SUM_TYPE_NODE;
+
+	/* readahead multi ssa blocks those have contiguous address */
+	if (sbi->segs_per_sec > 1)
+		ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno),
+					sbi->segs_per_sec, META_SSA, true);
 
-	/* read segment summary of victim */
-	sum_page = get_sum_page(sbi, segno);
+	/* reference all summary page */
+	while (segno < end_segno) {
+		sum_page = get_sum_page(sbi, segno++);
+		unlock_page(sum_page);
+	}
 
 	blk_start_plug(&plug);
 
-	sum = page_address(sum_page);
+	for (segno = start_segno; segno < end_segno; segno++) {
 
-	/*
-	 * this is to avoid deadlock:
-	 * - lock_page(sum_page)         - f2fs_replace_block
-	 *  - check_valid_map()            - mutex_lock(sentry_lock)
-	 *   - mutex_lock(sentry_lock)     - change_curseg()
-	 *                                  - lock_page(sum_page)
-	 */
-	unlock_page(sum_page);
-
-	switch (GET_SUM_TYPE((&sum->footer))) {
-	case SUM_TYPE_NODE:
-		nfree = gc_node_segment(sbi, sum->entries, segno, gc_type);
-		break;
-	case SUM_TYPE_DATA:
-		nfree = gc_data_segment(sbi, sum->entries, gc_list,
-							segno, gc_type);
-		break;
+		/* find segment summary of victim */
+		sum_page = find_get_page(META_MAPPING(sbi),
+					GET_SUM_BLOCK(sbi, segno));
+		f2fs_put_page(sum_page, 0);
+
+		if (get_valid_blocks(sbi, segno, false) == 0 ||
+				!PageUptodate(sum_page) ||
+				unlikely(f2fs_cp_error(sbi)))
+			goto next;
+
+		sum = page_address(sum_page);
+		f2fs_bug_on(sbi, type != GET_SUM_TYPE((&sum->footer)));
+
+		/*
+		 * this is to avoid deadlock:
+		 * - lock_page(sum_page)         - f2fs_replace_block
+		 *  - check_valid_map()            - down_write(sentry_lock)
+		 *   - down_read(sentry_lock)     - change_curseg()
+		 *                                  - lock_page(sum_page)
+		 */
+		if (type == SUM_TYPE_NODE)
+			gc_node_segment(sbi, sum->entries, segno, gc_type);
+		else
+			gc_data_segment(sbi, sum->entries, gc_list, segno,
+								gc_type);
+
+		stat_inc_seg_count(sbi, type, gc_type);
+
+		if (gc_type == FG_GC &&
+				get_valid_blocks(sbi, segno, false) == 0)
+			seg_freed++;
+next:
+		f2fs_put_page(sum_page, 0);
 	}
+
+	if (gc_type == FG_GC)
+		f2fs_submit_merged_write(sbi,
+				(type == SUM_TYPE_NODE) ? NODE : DATA);
+
 	blk_finish_plug(&plug);
 
-	stat_inc_seg_count(sbi, GET_SUM_TYPE((&sum->footer)), gc_type);
 	stat_inc_call_count(sbi->stat_info);
 
-	f2fs_put_page(sum_page, 0);
-	return nfree;
+	return seg_freed;
 }
 
-int f2fs_gc(struct f2fs_sb_info *sbi, bool sync)
+int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
+			bool background, unsigned int segno)
 {
-	unsigned int segno, i;
 	int gc_type = sync ? FG_GC : BG_GC;
-	int sec_freed = 0;
-	int ret = -EINVAL;
+	int sec_freed = 0, seg_freed = 0, total_freed = 0;
+	int ret = 0;
 	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),
 	};
 
+	trace_f2fs_gc_begin(sbi->sb, sync, background,
+				get_pages(sbi, F2FS_DIRTY_NODES),
+				get_pages(sbi, F2FS_DIRTY_DENTS),
+				get_pages(sbi, F2FS_DIRTY_IMETA),
+				free_sections(sbi),
+				free_segments(sbi),
+				reserved_segments(sbi),
+				prefree_segments(sbi));
+
 	cpc.reason = __get_cp_reason(sbi);
 gc_more:
-	segno = NULL_SEGNO;
-
-	if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE)))
+	if (unlikely(!(sbi->sb->s_flags & MS_ACTIVE))) {
+		ret = -EINVAL;
 		goto stop;
-	if (unlikely(f2fs_cp_error(sbi)))
-		goto stop;
-
-	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, sec_freed)) {
-		gc_type = FG_GC;
-		if (__get_victim(sbi, &segno, gc_type) || prefree_segments(sbi))
-			write_checkpoint(sbi, &cpc);
 	}
-
-	if (segno == NULL_SEGNO && !__get_victim(sbi, &segno, gc_type))
+	if (unlikely(f2fs_cp_error(sbi))) {
+		ret = -EIO;
 		goto stop;
-	ret = 0;
-
-	/* readahead multi ssa blocks those have contiguous address */
-	if (sbi->segs_per_sec > 1)
-		ra_meta_pages(sbi, GET_SUM_BLOCK(sbi, segno), sbi->segs_per_sec,
-							META_SSA, true);
+	}
 
-	for (i = 0; i < sbi->segs_per_sec; i++) {
+	if (gc_type == BG_GC && has_not_enough_free_secs(sbi, 0, 0)) {
 		/*
-		 * for FG_GC case, halt gcing left segments once failed one
-		 * of segments in selected section to avoid long latency.
+		 * For example, if there are many prefree_segments below given
+		 * threshold, we can make them free by checkpoint. Then, we
+		 * secure free segments which doesn't need fggc any more.
 		 */
-		if (!do_garbage_collect(sbi, segno + i, &gc_list, gc_type) &&
-				gc_type == FG_GC)
-			break;
+		if (prefree_segments(sbi)) {
+			ret = write_checkpoint(sbi, &cpc);
+			if (ret)
+				goto stop;
+		}
+		if (has_not_enough_free_secs(sbi, 0, 0))
+			gc_type = FG_GC;
+	}
+
+	/* f2fs_balance_fs doesn't need to do BG_GC in critical path. */
+	if (gc_type == BG_GC && !background) {
+		ret = -EINVAL;
+		goto stop;
+	}
+	if (!__get_victim(sbi, &segno, gc_type)) {
+		ret = -ENODATA;
+		goto stop;
 	}
 
-	if (i == sbi->segs_per_sec && gc_type == FG_GC)
+	seg_freed = do_garbage_collect(sbi, segno, &gc_list, gc_type);
+	if (gc_type == FG_GC && seg_freed == sbi->segs_per_sec)
 		sec_freed++;
+	total_freed += seg_freed;
 
 	if (gc_type == FG_GC)
 		sbi->cur_victim_sec = NULL_SEGNO;
 
 	if (!sync) {
-		if (has_not_enough_free_secs(sbi, sec_freed))
+		if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
+			segno = NULL_SEGNO;
 			goto gc_more;
+		}
 
 		if (gc_type == FG_GC)
-			write_checkpoint(sbi, &cpc);
+			ret = write_checkpoint(sbi, &cpc);
 	}
 stop:
+	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
+	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
+
+	trace_f2fs_gc_end(sbi->sb, ret, total_freed, sec_freed,
+				get_pages(sbi, F2FS_DIRTY_NODES),
+				get_pages(sbi, F2FS_DIRTY_DENTS),
+				get_pages(sbi, F2FS_DIRTY_IMETA),
+				free_sections(sbi),
+				free_segments(sbi),
+				reserved_segments(sbi),
+				prefree_segments(sbi));
+
 	mutex_unlock(&sbi->gc_mutex);
 
 	put_gc_inode(&gc_list);
@@ -885,5 +1080,20 @@ stop:
 
 void build_gc_manager(struct f2fs_sb_info *sbi)
 {
+	u64 main_count, resv_count, ovp_count;
+
 	DIRTY_I(sbi)->v_ops = &default_v_ops;
+
+	/* threshold of # of valid blocks in a section for victims of FG_GC */
+	main_count = SM_I(sbi)->main_segments << sbi->log_blocks_per_seg;
+	resv_count = SM_I(sbi)->reserved_segments << sbi->log_blocks_per_seg;
+	ovp_count = SM_I(sbi)->ovp_segments << sbi->log_blocks_per_seg;
+
+	sbi->fggc_threshold = div64_u64((main_count - ovp_count) *
+				BLKS_PER_SEC(sbi), (main_count - resv_count));
+
+	/* give warm/cold data area from slower device */
+	if (sbi->s_ndevs && sbi->segs_per_sec == 1)
+		SIT_I(sbi)->last_victim[ALLOC_NEXT] =
+				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
 }