forked from ~ljy/RK356X_SDK_RELEASE
blame | history | patch | commit | commitdiff
hc
2023-12-09 95099d4622f8cb224d94e314c7a8e0df60b13f87 
 kernel/fs/f2fs/checkpoint.c
....@@ -13,22 +13,28 @@
1313 #include <linux/f2fs_fs.h>
1414 #include <linux/pagevec.h>
1515 #include <linux/swap.h>
16
+#include <linux/kthread.h>
1617 
1718 #include "f2fs.h"
1819 #include "node.h"
1920 #include "segment.h"
20
-#include "trace.h"
2121 #include <trace/events/f2fs.h>
22
+
23
+#define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
2224 
2325 static struct kmem_cache *ino_entry_slab;
2426 struct kmem_cache *f2fs_inode_entry_slab;
2527 
26
-void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
28
+void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io,
29
+						unsigned char reason)
2730 {
2831 	f2fs_build_fault_attr(sbi, 0, 0);
2932 	set_ckpt_flags(sbi, CP_ERROR_FLAG);
30
-	if (!end_io)
33
+	if (!end_io) {
3134 		f2fs_flush_merged_writes(sbi);
35
+
36
+		f2fs_handle_stop(sbi, reason);
37
+	}
3238 }
3339 
3440 /*
....@@ -37,7 +43,7 @@
3743 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
3844 {
3945 	struct address_space *mapping = META_MAPPING(sbi);
40
-	struct page *page = NULL;
46
+	struct page *page;
4147 repeat:
4248 	page = f2fs_grab_cache_page(mapping, index, false);
4349 	if (!page) {
....@@ -107,7 +113,7 @@
107113 	return __get_meta_page(sbi, index, true);
108114 }
109115 
110
-struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
116
+struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
111117 {
112118 	struct page *page;
113119 	int count = 0;
....@@ -118,7 +124,7 @@
118124 		if (PTR_ERR(page) == -EIO &&
119125 				++count <= DEFAULT_RETRY_IO_COUNT)
120126 			goto retry;
121
-		f2fs_stop_checkpoint(sbi, false);
127
+		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
122128 	}
123129 	return page;
124130 }
....@@ -136,7 +142,7 @@
136142 	unsigned int segno, offset;
137143 	bool exist;
138144 
139
-	if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
145
+	if (type == DATA_GENERIC)
140146 		return true;
141147 
142148 	segno = GET_SEGNO(sbi, blkaddr);
....@@ -144,11 +150,18 @@
144150 	se = get_seg_entry(sbi, segno);
145151 
146152 	exist = f2fs_test_bit(offset, se->cur_valid_map);
153
+	if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
154
+		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
155
+			 blkaddr, exist);
156
+		set_sbi_flag(sbi, SBI_NEED_FSCK);
157
+		return exist;
158
+	}
159
+
147160 	if (!exist && type == DATA_GENERIC_ENHANCE) {
148161 		f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
149162 			 blkaddr, exist);
150163 		set_sbi_flag(sbi, SBI_NEED_FSCK);
151
-		WARN_ON(1);
164
+		dump_stack();
152165 	}
153166 	return exist;
154167 }
....@@ -181,12 +194,13 @@
181194 	case DATA_GENERIC:
182195 	case DATA_GENERIC_ENHANCE:
183196 	case DATA_GENERIC_ENHANCE_READ:
197
+	case DATA_GENERIC_ENHANCE_UPDATE:
184198 		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
185199 				blkaddr < MAIN_BLKADDR(sbi))) {
186200 			f2fs_warn(sbi, "access invalid blkaddr:%u",
187201 				  blkaddr);
188202 			set_sbi_flag(sbi, SBI_NEED_FSCK);
189
-			WARN_ON(1);
203
+			dump_stack();
190204 			return false;
191205 		} else {
192206 			return __is_bitmap_valid(sbi, blkaddr, type);
....@@ -348,13 +362,13 @@
348362 		goto skip_write;
349363 
350364 	/* if locked failed, cp will flush dirty pages instead */
351
-	if (!mutex_trylock(&sbi->cp_mutex))
365
+	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
352366 		goto skip_write;
353367 
354368 	trace_f2fs_writepages(mapping->host, wbc, META);
355369 	diff = nr_pages_to_write(sbi, META, wbc);
356370 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
357
-	mutex_unlock(&sbi->cp_mutex);
371
+	f2fs_up_write(&sbi->cp_global_sem);
358372 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
359373 	return 0;
360374 
....@@ -442,8 +456,7 @@
442456 	if (!PageDirty(page)) {
443457 		__set_page_dirty_nobuffers(page);
444458 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
445
-		f2fs_set_page_private(page, 0);
446
-		f2fs_trace_pid(page);
459
+		set_page_private_reference(page);
447460 		return 1;
448461 	}
449462 	return 0;
....@@ -525,7 +538,7 @@
525538 	__remove_ino_entry(sbi, ino, type);
526539 }
527540 
528
-/* mode should be APPEND_INO or UPDATE_INO */
541
+/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
529542 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
530543 {
531544 	struct inode_management *im = &sbi->im[mode];
....@@ -649,7 +662,7 @@
649662 	/* truncate all the data during iput */
650663 	iput(inode);
651664 
652
-	err = f2fs_get_node_info(sbi, ino, &ni);
665
+	err = f2fs_get_node_info(sbi, ino, &ni, false);
653666 	if (err)
654667 		goto err_out;
655668 
....@@ -718,6 +731,7 @@
718731 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
719732 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
720733 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
734
+
721735 			err = recover_orphan_inode(sbi, ino);
722736 			if (err) {
723737 				f2fs_put_page(page, 1);
....@@ -851,6 +865,7 @@
851865 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
852866 	struct f2fs_checkpoint *cp_block = NULL;
853867 	unsigned long long cur_version = 0, pre_version = 0;
868
+	unsigned int cp_blocks;
854869 	int err;
855870 
856871 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
....@@ -858,15 +873,16 @@
858873 	if (err)
859874 		return NULL;
860875 
861
-	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
862
-					sbi->blocks_per_seg) {
876
+	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
877
+
878
+	if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
863879 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
864880 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
865881 		goto invalid_cp;
866882 	}
867883 	pre_version = *version;
868884 
869
-	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
885
+	cp_addr += cp_blocks - 1;
870886 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
871887 					&cp_page_2, version);
872888 	if (err)
....@@ -1016,8 +1032,7 @@
10161032 	inode_inc_dirty_pages(inode);
10171033 	spin_unlock(&sbi->inode_lock[type]);
10181034 
1019
-	f2fs_set_page_private(page, 0);
1020
-	f2fs_trace_pid(page);
1035
+	set_page_private_reference(page);
10211036 }
10221037 
10231038 void f2fs_remove_dirty_inode(struct inode *inode)
....@@ -1037,7 +1052,8 @@
10371052 	spin_unlock(&sbi->inode_lock[type]);
10381053 }
10391054 
1040
-int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1055
+int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1056
+						bool from_cp)
10411057 {
10421058 	struct list_head *head;
10431059 	struct inode *inode;
....@@ -1072,11 +1088,15 @@
10721088 	if (inode) {
10731089 		unsigned long cur_ino = inode->i_ino;
10741090 
1075
-		F2FS_I(inode)->cp_task = current;
1091
+		if (from_cp)
1092
+			F2FS_I(inode)->cp_task = current;
1093
+		F2FS_I(inode)->wb_task = current;
10761094 
10771095 		filemap_fdatawrite(inode->i_mapping);
10781096 
1079
-		F2FS_I(inode)->cp_task = NULL;
1097
+		F2FS_I(inode)->wb_task = NULL;
1098
+		if (from_cp)
1099
+			F2FS_I(inode)->cp_task = NULL;
10801100 
10811101 		iput(inode);
10821102 		/* We need to give cpu to another writers. */
....@@ -1147,7 +1167,8 @@
11471167 	if (!is_journalled_quota(sbi))
11481168 		return false;
11491169 
1150
-	down_write(&sbi->quota_sem);
1170
+	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1171
+		return true;
11511172 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
11521173 		ret = false;
11531174 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
....@@ -1158,7 +1179,7 @@
11581179 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
11591180 		ret = true;
11601181 	}
1161
-	up_write(&sbi->quota_sem);
1182
+	f2fs_up_write(&sbi->quota_sem);
11621183 	return ret;
11631184 }
11641185 
....@@ -1204,7 +1225,7 @@
12041225 	/* write all the dirty dentry pages */
12051226 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
12061227 		f2fs_unlock_all(sbi);
1207
-		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1228
+		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
12081229 		if (err)
12091230 			return err;
12101231 		cond_resched();
....@@ -1215,10 +1236,10 @@
12151236 	 * POR: we should ensure that there are no dirty node pages
12161237 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
12171238 	 */
1218
-	down_write(&sbi->node_change);
1239
+	f2fs_down_write(&sbi->node_change);
12191240 
12201241 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1221
-		up_write(&sbi->node_change);
1242
+		f2fs_up_write(&sbi->node_change);
12221243 		f2fs_unlock_all(sbi);
12231244 		err = f2fs_sync_inode_meta(sbi);
12241245 		if (err)
....@@ -1228,15 +1249,15 @@
12281249 	}
12291250 
12301251 retry_flush_nodes:
1231
-	down_write(&sbi->node_write);
1252
+	f2fs_down_write(&sbi->node_write);
12321253 
12331254 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1234
-		up_write(&sbi->node_write);
1255
+		f2fs_up_write(&sbi->node_write);
12351256 		atomic_inc(&sbi->wb_sync_req[NODE]);
12361257 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
12371258 		atomic_dec(&sbi->wb_sync_req[NODE]);
12381259 		if (err) {
1239
-			up_write(&sbi->node_change);
1260
+			f2fs_up_write(&sbi->node_change);
12401261 			f2fs_unlock_all(sbi);
12411262 			return err;
12421263 		}
....@@ -1249,13 +1270,13 @@
12491270 	 * dirty node blocks and some checkpoint values by block allocation.
12501271 	 */
12511272 	__prepare_cp_block(sbi);
1252
-	up_write(&sbi->node_change);
1273
+	f2fs_up_write(&sbi->node_change);
12531274 	return err;
12541275 }
12551276 
12561277 static void unblock_operations(struct f2fs_sb_info *sbi)
12571278 {
1258
-	up_write(&sbi->node_write);
1279
+	f2fs_up_write(&sbi->node_write);
12591280 	f2fs_unlock_all(sbi);
12601281 }
12611282 
....@@ -1264,8 +1285,6 @@
12641285 	DEFINE_WAIT(wait);
12651286 
12661287 	for (;;) {
1267
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1268
-
12691288 		if (!get_pages(sbi, type))
12701289 			break;
12711290 
....@@ -1275,6 +1294,10 @@
12751294 		if (type == F2FS_DIRTY_META)
12761295 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
12771296 							FS_CP_META_IO);
1297
+		else if (type == F2FS_WB_CP_DATA)
1298
+			f2fs_submit_merged_write(sbi, DATA);
1299
+
1300
+		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
12781301 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
12791302 	}
12801303 	finish_wait(&sbi->cp_wait, &wait);
....@@ -1383,6 +1406,26 @@
13831406 	f2fs_submit_merged_write(sbi, META_FLUSH);
13841407 }
13851408 
1409
+static inline u64 get_sectors_written(struct block_device *bdev)
1410
+{
1411
+	return (u64)part_stat_read(bdev->bd_part, sectors[STAT_WRITE]);
1412
+}
1413
+
1414
+u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1415
+{
1416
+	if (f2fs_is_multi_device(sbi)) {
1417
+		u64 sectors = 0;
1418
+		int i;
1419
+
1420
+		for (i = 0; i < sbi->s_ndevs; i++)
1421
+			sectors += get_sectors_written(FDEV(i).bdev);
1422
+
1423
+		return sectors;
1424
+	}
1425
+
1426
+	return get_sectors_written(sbi->sb->s_bdev);
1427
+}
1428
+
13861429 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
13871430 {
13881431 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
....@@ -1393,7 +1436,6 @@
13931436 	__u32 crc32 = 0;
13941437 	int i;
13951438 	int cp_payload_blks = __cp_payload(sbi);
1396
-	struct super_block *sb = sbi->sb;
13971439 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
13981440 	u64 kbytes_written;
13991441 	int err;
....@@ -1421,7 +1463,7 @@
14211463 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
14221464 	}
14231465 
1424
-	/* 2 cp  + n data seg summary + orphan inode blocks */
1466
+	/* 2 cp + n data seg summary + orphan inode blocks */
14251467 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
14261468 	spin_lock_irqsave(&sbi->cp_lock, flags);
14271469 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
....@@ -1435,7 +1477,7 @@
14351477 			orphan_blocks);
14361478 
14371479 	if (__remain_node_summaries(cpc->reason))
1438
-		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1480
+		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
14391481 				cp_payload_blks + data_sum_blocks +
14401482 				orphan_blocks + NR_CURSEG_NODE_TYPE);
14411483 	else
....@@ -1488,9 +1530,8 @@
14881530 
14891531 	/* Record write statistics in the hot node summary */
14901532 	kbytes_written = sbi->kbytes_written;
1491
-	if (sb->s_bdev->bd_part)
1492
-		kbytes_written += BD_PART_WRITTEN(sbi);
1493
-
1533
+	kbytes_written += (f2fs_get_sectors_written(sbi) -
1534
+				sbi->sectors_written_start) >> 1;
14941535 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
14951536 
14961537 	if (__remain_node_summaries(cpc->reason)) {
....@@ -1521,9 +1562,10 @@
15211562 
15221563 	/*
15231564 	 * invalidate intermediate page cache borrowed from meta inode which are
1524
-	 * used for migration of encrypted or verity inode's blocks.
1565
+	 * used for migration of encrypted, verity or compressed inode's blocks.
15251566 	 */
1526
-	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi))
1567
+	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1568
+		f2fs_sb_has_compression(sbi))
15271569 		invalidate_mapping_pages(META_MAPPING(sbi),
15281570 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
15291571 
....@@ -1569,7 +1611,7 @@
15691611 		f2fs_warn(sbi, "Start checkpoint disabled!");
15701612 	}
15711613 	if (cpc->reason != CP_RESIZE)
1572
-		mutex_lock(&sbi->cp_mutex);
1614
+		f2fs_down_write(&sbi->cp_global_sem);
15731615 
15741616 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
15751617 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
....@@ -1597,7 +1639,7 @@
15971639 			goto out;
15981640 		}
15991641 
1600
-		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1642
+		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
16011643 				SIT_I(sbi)->dirty_sentries == 0 &&
16021644 				prefree_segments(sbi) == 0) {
16031645 			f2fs_flush_sit_entries(sbi, cpc);
....@@ -1622,11 +1664,16 @@
16221664 
16231665 	f2fs_flush_sit_entries(sbi, cpc);
16241666 
1667
+	/* save inmem log status */
1668
+	f2fs_save_inmem_curseg(sbi);
1669
+
16251670 	err = do_checkpoint(sbi, cpc);
16261671 	if (err)
16271672 		f2fs_release_discard_addrs(sbi);
16281673 	else
16291674 		f2fs_clear_prefree_segments(sbi, cpc);
1675
+
1676
+	f2fs_restore_inmem_curseg(sbi);
16301677 stop:
16311678 	unblock_operations(sbi);
16321679 	stat_inc_cp_count(sbi->stat_info);
....@@ -1639,7 +1686,7 @@
16391686 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
16401687 out:
16411688 	if (cpc->reason != CP_RESIZE)
1642
-		mutex_unlock(&sbi->cp_mutex);
1689
+		f2fs_up_write(&sbi->cp_global_sem);
16431690 	return err;
16441691 }
16451692 
....@@ -1657,7 +1704,7 @@
16571704 	}
16581705 
16591706 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1660
-			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1707
+			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
16611708 				F2FS_ORPHANS_PER_BLOCK;
16621709 }
16631710 
....@@ -1681,3 +1728,190 @@
16811728 	kmem_cache_destroy(ino_entry_slab);
16821729 	kmem_cache_destroy(f2fs_inode_entry_slab);
16831730 }
1731
+
1732
+static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1733
+{
1734
+	struct cp_control cpc = { .reason = CP_SYNC, };
1735
+	int err;
1736
+
1737
+	f2fs_down_write(&sbi->gc_lock);
1738
+	err = f2fs_write_checkpoint(sbi, &cpc);
1739
+	f2fs_up_write(&sbi->gc_lock);
1740
+
1741
+	return err;
1742
+}
1743
+
1744
+static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1745
+{
1746
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1747
+	struct ckpt_req *req, *next;
1748
+	struct llist_node *dispatch_list;
1749
+	u64 sum_diff = 0, diff, count = 0;
1750
+	int ret;
1751
+
1752
+	dispatch_list = llist_del_all(&cprc->issue_list);
1753
+	if (!dispatch_list)
1754
+		return;
1755
+	dispatch_list = llist_reverse_order(dispatch_list);
1756
+
1757
+	ret = __write_checkpoint_sync(sbi);
1758
+	atomic_inc(&cprc->issued_ckpt);
1759
+
1760
+	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1761
+		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1762
+		req->ret = ret;
1763
+		complete(&req->wait);
1764
+
1765
+		sum_diff += diff;
1766
+		count++;
1767
+	}
1768
+	atomic_sub(count, &cprc->queued_ckpt);
1769
+	atomic_add(count, &cprc->total_ckpt);
1770
+
1771
+	spin_lock(&cprc->stat_lock);
1772
+	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1773
+	if (cprc->peak_time < cprc->cur_time)
1774
+		cprc->peak_time = cprc->cur_time;
1775
+	spin_unlock(&cprc->stat_lock);
1776
+}
1777
+
1778
+static int issue_checkpoint_thread(void *data)
1779
+{
1780
+	struct f2fs_sb_info *sbi = data;
1781
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1782
+	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1783
+repeat:
1784
+	if (kthread_should_stop())
1785
+		return 0;
1786
+
1787
+	if (!llist_empty(&cprc->issue_list))
1788
+		__checkpoint_and_complete_reqs(sbi);
1789
+
1790
+	wait_event_interruptible(*q,
1791
+		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1792
+	goto repeat;
1793
+}
1794
+
1795
+static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1796
+		struct ckpt_req *wait_req)
1797
+{
1798
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1799
+
1800
+	if (!llist_empty(&cprc->issue_list)) {
1801
+		__checkpoint_and_complete_reqs(sbi);
1802
+	} else {
1803
+		/* already dispatched by issue_checkpoint_thread */
1804
+		if (wait_req)
1805
+			wait_for_completion(&wait_req->wait);
1806
+	}
1807
+}
1808
+
1809
+static void init_ckpt_req(struct ckpt_req *req)
1810
+{
1811
+	memset(req, 0, sizeof(struct ckpt_req));
1812
+
1813
+	init_completion(&req->wait);
1814
+	req->queue_time = ktime_get();
1815
+}
1816
+
1817
+int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1818
+{
1819
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1820
+	struct ckpt_req req;
1821
+	struct cp_control cpc;
1822
+
1823
+	cpc.reason = __get_cp_reason(sbi);
1824
+	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1825
+		int ret;
1826
+
1827
+		f2fs_down_write(&sbi->gc_lock);
1828
+		ret = f2fs_write_checkpoint(sbi, &cpc);
1829
+		f2fs_up_write(&sbi->gc_lock);
1830
+
1831
+		return ret;
1832
+	}
1833
+
1834
+	if (!cprc->f2fs_issue_ckpt)
1835
+		return __write_checkpoint_sync(sbi);
1836
+
1837
+	init_ckpt_req(&req);
1838
+
1839
+	llist_add(&req.llnode, &cprc->issue_list);
1840
+	atomic_inc(&cprc->queued_ckpt);
1841
+
1842
+	/*
1843
+	 * update issue_list before we wake up issue_checkpoint thread,
1844
+	 * this smp_mb() pairs with another barrier in ___wait_event(),
1845
+	 * see more details in comments of waitqueue_active().
1846
+	 */
1847
+	smp_mb();
1848
+
1849
+	if (waitqueue_active(&cprc->ckpt_wait_queue))
1850
+		wake_up(&cprc->ckpt_wait_queue);
1851
+
1852
+	if (cprc->f2fs_issue_ckpt)
1853
+		wait_for_completion(&req.wait);
1854
+	else
1855
+		flush_remained_ckpt_reqs(sbi, &req);
1856
+
1857
+	return req.ret;
1858
+}
1859
+
1860
+int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1861
+{
1862
+	dev_t dev = sbi->sb->s_bdev->bd_dev;
1863
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1864
+
1865
+	if (cprc->f2fs_issue_ckpt)
1866
+		return 0;
1867
+
1868
+	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1869
+			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1870
+	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1871
+		cprc->f2fs_issue_ckpt = NULL;
1872
+		return -ENOMEM;
1873
+	}
1874
+
1875
+	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1876
+
1877
+	return 0;
1878
+}
1879
+
1880
+void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1881
+{
1882
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1883
+	struct task_struct *ckpt_task;
1884
+
1885
+	if (!cprc->f2fs_issue_ckpt)
1886
+		return;
1887
+
1888
+	ckpt_task = cprc->f2fs_issue_ckpt;
1889
+	cprc->f2fs_issue_ckpt = NULL;
1890
+	kthread_stop(ckpt_task);
1891
+
1892
+	f2fs_flush_ckpt_thread(sbi);
1893
+}
1894
+
1895
+void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1896
+{
1897
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1898
+
1899
+	flush_remained_ckpt_reqs(sbi, NULL);
1900
+
1901
+	/* Let's wait for the previous dispatched checkpoint. */
1902
+	while (atomic_read(&cprc->queued_ckpt))
1903
+		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1904
+}
1905
+
1906
+void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1907
+{
1908
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1909
+
1910
+	atomic_set(&cprc->issued_ckpt, 0);
1911
+	atomic_set(&cprc->total_ckpt, 0);
1912
+	atomic_set(&cprc->queued_ckpt, 0);
1913
+	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1914
+	init_waitqueue_head(&cprc->ckpt_wait_queue);
1915
+	init_llist_head(&cprc->issue_list);
1916
+	spin_lock_init(&cprc->stat_lock);
1917
+}