forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2024-02-20 102a0743326a03cd1a1202ceda21e175b7d3575c 
 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);
....@@ -301,8 +315,15 @@
301315 
302316 	trace_f2fs_writepage(page, META);
303317 
304
-	if (unlikely(f2fs_cp_error(sbi)))
318
+	if (unlikely(f2fs_cp_error(sbi))) {
319
+		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
320
+			ClearPageUptodate(page);
321
+			dec_page_count(sbi, F2FS_DIRTY_META);
322
+			unlock_page(page);
323
+			return 0;
324
+		}
305325 		goto redirty_out;
326
+	}
306327 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
307328 		goto redirty_out;
308329 	if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
....@@ -348,13 +369,13 @@
348369 		goto skip_write;
349370 
350371 	/* if locked failed, cp will flush dirty pages instead */
351
-	if (!mutex_trylock(&sbi->cp_mutex))
372
+	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
352373 		goto skip_write;
353374 
354375 	trace_f2fs_writepages(mapping->host, wbc, META);
355376 	diff = nr_pages_to_write(sbi, META, wbc);
356377 	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
357
-	mutex_unlock(&sbi->cp_mutex);
378
+	f2fs_up_write(&sbi->cp_global_sem);
358379 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
359380 	return 0;
360381 
....@@ -442,8 +463,7 @@
442463 	if (!PageDirty(page)) {
443464 		__set_page_dirty_nobuffers(page);
444465 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
445
-		f2fs_set_page_private(page, 0);
446
-		f2fs_trace_pid(page);
466
+		set_page_private_reference(page);
447467 		return 1;
448468 	}
449469 	return 0;
....@@ -525,7 +545,7 @@
525545 	__remove_ino_entry(sbi, ino, type);
526546 }
527547 
528
-/* mode should be APPEND_INO or UPDATE_INO */
548
+/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
529549 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
530550 {
531551 	struct inode_management *im = &sbi->im[mode];
....@@ -649,7 +669,7 @@
649669 	/* truncate all the data during iput */
650670 	iput(inode);
651671 
652
-	err = f2fs_get_node_info(sbi, ino, &ni);
672
+	err = f2fs_get_node_info(sbi, ino, &ni, false);
653673 	if (err)
654674 		goto err_out;
655675 
....@@ -718,6 +738,7 @@
718738 		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
719739 		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
720740 			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
741
+
721742 			err = recover_orphan_inode(sbi, ino);
722743 			if (err) {
723744 				f2fs_put_page(page, 1);
....@@ -851,6 +872,7 @@
851872 	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
852873 	struct f2fs_checkpoint *cp_block = NULL;
853874 	unsigned long long cur_version = 0, pre_version = 0;
875
+	unsigned int cp_blocks;
854876 	int err;
855877 
856878 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
....@@ -858,15 +880,16 @@
858880 	if (err)
859881 		return NULL;
860882 
861
-	if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
862
-					sbi->blocks_per_seg) {
883
+	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
884
+
885
+	if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
863886 		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
864887 			  le32_to_cpu(cp_block->cp_pack_total_block_count));
865888 		goto invalid_cp;
866889 	}
867890 	pre_version = *version;
868891 
869
-	cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
892
+	cp_addr += cp_blocks - 1;
870893 	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
871894 					&cp_page_2, version);
872895 	if (err)
....@@ -1016,8 +1039,7 @@
10161039 	inode_inc_dirty_pages(inode);
10171040 	spin_unlock(&sbi->inode_lock[type]);
10181041 
1019
-	f2fs_set_page_private(page, 0);
1020
-	f2fs_trace_pid(page);
1042
+	set_page_private_reference(page);
10211043 }
10221044 
10231045 void f2fs_remove_dirty_inode(struct inode *inode)
....@@ -1037,7 +1059,8 @@
10371059 	spin_unlock(&sbi->inode_lock[type]);
10381060 }
10391061 
1040
-int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1062
+int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1063
+						bool from_cp)
10411064 {
10421065 	struct list_head *head;
10431066 	struct inode *inode;
....@@ -1072,11 +1095,15 @@
10721095 	if (inode) {
10731096 		unsigned long cur_ino = inode->i_ino;
10741097 
1075
-		F2FS_I(inode)->cp_task = current;
1098
+		if (from_cp)
1099
+			F2FS_I(inode)->cp_task = current;
1100
+		F2FS_I(inode)->wb_task = current;
10761101 
10771102 		filemap_fdatawrite(inode->i_mapping);
10781103 
1079
-		F2FS_I(inode)->cp_task = NULL;
1104
+		F2FS_I(inode)->wb_task = NULL;
1105
+		if (from_cp)
1106
+			F2FS_I(inode)->cp_task = NULL;
10801107 
10811108 		iput(inode);
10821109 		/* We need to give cpu to another writers. */
....@@ -1147,7 +1174,8 @@
11471174 	if (!is_journalled_quota(sbi))
11481175 		return false;
11491176 
1150
-	down_write(&sbi->quota_sem);
1177
+	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1178
+		return true;
11511179 	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
11521180 		ret = false;
11531181 	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
....@@ -1158,7 +1186,7 @@
11581186 	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
11591187 		ret = true;
11601188 	}
1161
-	up_write(&sbi->quota_sem);
1189
+	f2fs_up_write(&sbi->quota_sem);
11621190 	return ret;
11631191 }
11641192 
....@@ -1204,7 +1232,7 @@
12041232 	/* write all the dirty dentry pages */
12051233 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
12061234 		f2fs_unlock_all(sbi);
1207
-		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1235
+		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
12081236 		if (err)
12091237 			return err;
12101238 		cond_resched();
....@@ -1215,10 +1243,10 @@
12151243 	 * POR: we should ensure that there are no dirty node pages
12161244 	 * until finishing nat/sit flush. inode->i_blocks can be updated.
12171245 	 */
1218
-	down_write(&sbi->node_change);
1246
+	f2fs_down_write(&sbi->node_change);
12191247 
12201248 	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1221
-		up_write(&sbi->node_change);
1249
+		f2fs_up_write(&sbi->node_change);
12221250 		f2fs_unlock_all(sbi);
12231251 		err = f2fs_sync_inode_meta(sbi);
12241252 		if (err)
....@@ -1228,15 +1256,15 @@
12281256 	}
12291257 
12301258 retry_flush_nodes:
1231
-	down_write(&sbi->node_write);
1259
+	f2fs_down_write(&sbi->node_write);
12321260 
12331261 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1234
-		up_write(&sbi->node_write);
1262
+		f2fs_up_write(&sbi->node_write);
12351263 		atomic_inc(&sbi->wb_sync_req[NODE]);
12361264 		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
12371265 		atomic_dec(&sbi->wb_sync_req[NODE]);
12381266 		if (err) {
1239
-			up_write(&sbi->node_change);
1267
+			f2fs_up_write(&sbi->node_change);
12401268 			f2fs_unlock_all(sbi);
12411269 			return err;
12421270 		}
....@@ -1249,13 +1277,13 @@
12491277 	 * dirty node blocks and some checkpoint values by block allocation.
12501278 	 */
12511279 	__prepare_cp_block(sbi);
1252
-	up_write(&sbi->node_change);
1280
+	f2fs_up_write(&sbi->node_change);
12531281 	return err;
12541282 }
12551283 
12561284 static void unblock_operations(struct f2fs_sb_info *sbi)
12571285 {
1258
-	up_write(&sbi->node_write);
1286
+	f2fs_up_write(&sbi->node_write);
12591287 	f2fs_unlock_all(sbi);
12601288 }
12611289 
....@@ -1264,17 +1292,20 @@
12641292 	DEFINE_WAIT(wait);
12651293 
12661294 	for (;;) {
1267
-		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1268
-
12691295 		if (!get_pages(sbi, type))
12701296 			break;
12711297 
1272
-		if (unlikely(f2fs_cp_error(sbi)))
1298
+		if (unlikely(f2fs_cp_error(sbi) &&
1299
+			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
12731300 			break;
12741301 
12751302 		if (type == F2FS_DIRTY_META)
12761303 			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
12771304 							FS_CP_META_IO);
1305
+		else if (type == F2FS_WB_CP_DATA)
1306
+			f2fs_submit_merged_write(sbi, DATA);
1307
+
1308
+		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
12781309 		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
12791310 	}
12801311 	finish_wait(&sbi->cp_wait, &wait);
....@@ -1383,6 +1414,26 @@
13831414 	f2fs_submit_merged_write(sbi, META_FLUSH);
13841415 }
13851416 
1417
+static inline u64 get_sectors_written(struct block_device *bdev)
1418
+{
1419
+	return (u64)part_stat_read(bdev->bd_part, sectors[STAT_WRITE]);
1420
+}
1421
+
1422
+u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1423
+{
1424
+	if (f2fs_is_multi_device(sbi)) {
1425
+		u64 sectors = 0;
1426
+		int i;
1427
+
1428
+		for (i = 0; i < sbi->s_ndevs; i++)
1429
+			sectors += get_sectors_written(FDEV(i).bdev);
1430
+
1431
+		return sectors;
1432
+	}
1433
+
1434
+	return get_sectors_written(sbi->sb->s_bdev);
1435
+}
1436
+
13861437 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
13871438 {
13881439 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
....@@ -1393,7 +1444,6 @@
13931444 	__u32 crc32 = 0;
13941445 	int i;
13951446 	int cp_payload_blks = __cp_payload(sbi);
1396
-	struct super_block *sb = sbi->sb;
13971447 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
13981448 	u64 kbytes_written;
13991449 	int err;
....@@ -1421,7 +1471,7 @@
14211471 				curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
14221472 	}
14231473 
1424
-	/* 2 cp  + n data seg summary + orphan inode blocks */
1474
+	/* 2 cp + n data seg summary + orphan inode blocks */
14251475 	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
14261476 	spin_lock_irqsave(&sbi->cp_lock, flags);
14271477 	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
....@@ -1435,7 +1485,7 @@
14351485 			orphan_blocks);
14361486 
14371487 	if (__remain_node_summaries(cpc->reason))
1438
-		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1488
+		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
14391489 				cp_payload_blks + data_sum_blocks +
14401490 				orphan_blocks + NR_CURSEG_NODE_TYPE);
14411491 	else
....@@ -1488,9 +1538,8 @@
14881538 
14891539 	/* Record write statistics in the hot node summary */
14901540 	kbytes_written = sbi->kbytes_written;
1491
-	if (sb->s_bdev->bd_part)
1492
-		kbytes_written += BD_PART_WRITTEN(sbi);
1493
-
1541
+	kbytes_written += (f2fs_get_sectors_written(sbi) -
1542
+				sbi->sectors_written_start) >> 1;
14941543 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
14951544 
14961545 	if (__remain_node_summaries(cpc->reason)) {
....@@ -1521,9 +1570,10 @@
15211570 
15221571 	/*
15231572 	 * invalidate intermediate page cache borrowed from meta inode which are
1524
-	 * used for migration of encrypted or verity inode's blocks.
1573
+	 * used for migration of encrypted, verity or compressed inode's blocks.
15251574 	 */
1526
-	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi))
1575
+	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1576
+		f2fs_sb_has_compression(sbi))
15271577 		invalidate_mapping_pages(META_MAPPING(sbi),
15281578 				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
15291579 
....@@ -1569,7 +1619,7 @@
15691619 		f2fs_warn(sbi, "Start checkpoint disabled!");
15701620 	}
15711621 	if (cpc->reason != CP_RESIZE)
1572
-		mutex_lock(&sbi->cp_mutex);
1622
+		f2fs_down_write(&sbi->cp_global_sem);
15731623 
15741624 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
15751625 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
....@@ -1597,7 +1647,7 @@
15971647 			goto out;
15981648 		}
15991649 
1600
-		if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1650
+		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
16011651 				SIT_I(sbi)->dirty_sentries == 0 &&
16021652 				prefree_segments(sbi) == 0) {
16031653 			f2fs_flush_sit_entries(sbi, cpc);
....@@ -1622,11 +1672,16 @@
16221672 
16231673 	f2fs_flush_sit_entries(sbi, cpc);
16241674 
1675
+	/* save inmem log status */
1676
+	f2fs_save_inmem_curseg(sbi);
1677
+
16251678 	err = do_checkpoint(sbi, cpc);
16261679 	if (err)
16271680 		f2fs_release_discard_addrs(sbi);
16281681 	else
16291682 		f2fs_clear_prefree_segments(sbi, cpc);
1683
+
1684
+	f2fs_restore_inmem_curseg(sbi);
16301685 stop:
16311686 	unblock_operations(sbi);
16321687 	stat_inc_cp_count(sbi->stat_info);
....@@ -1639,7 +1694,7 @@
16391694 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
16401695 out:
16411696 	if (cpc->reason != CP_RESIZE)
1642
-		mutex_unlock(&sbi->cp_mutex);
1697
+		f2fs_up_write(&sbi->cp_global_sem);
16431698 	return err;
16441699 }
16451700 
....@@ -1657,7 +1712,7 @@
16571712 	}
16581713 
16591714 	sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1660
-			NR_CURSEG_TYPE - __cp_payload(sbi)) *
1715
+			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
16611716 				F2FS_ORPHANS_PER_BLOCK;
16621717 }
16631718 
....@@ -1681,3 +1736,190 @@
16811736 	kmem_cache_destroy(ino_entry_slab);
16821737 	kmem_cache_destroy(f2fs_inode_entry_slab);
16831738 }
1739
+
1740
+static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1741
+{
1742
+	struct cp_control cpc = { .reason = CP_SYNC, };
1743
+	int err;
1744
+
1745
+	f2fs_down_write(&sbi->gc_lock);
1746
+	err = f2fs_write_checkpoint(sbi, &cpc);
1747
+	f2fs_up_write(&sbi->gc_lock);
1748
+
1749
+	return err;
1750
+}
1751
+
1752
+static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1753
+{
1754
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1755
+	struct ckpt_req *req, *next;
1756
+	struct llist_node *dispatch_list;
1757
+	u64 sum_diff = 0, diff, count = 0;
1758
+	int ret;
1759
+
1760
+	dispatch_list = llist_del_all(&cprc->issue_list);
1761
+	if (!dispatch_list)
1762
+		return;
1763
+	dispatch_list = llist_reverse_order(dispatch_list);
1764
+
1765
+	ret = __write_checkpoint_sync(sbi);
1766
+	atomic_inc(&cprc->issued_ckpt);
1767
+
1768
+	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1769
+		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1770
+		req->ret = ret;
1771
+		complete(&req->wait);
1772
+
1773
+		sum_diff += diff;
1774
+		count++;
1775
+	}
1776
+	atomic_sub(count, &cprc->queued_ckpt);
1777
+	atomic_add(count, &cprc->total_ckpt);
1778
+
1779
+	spin_lock(&cprc->stat_lock);
1780
+	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1781
+	if (cprc->peak_time < cprc->cur_time)
1782
+		cprc->peak_time = cprc->cur_time;
1783
+	spin_unlock(&cprc->stat_lock);
1784
+}
1785
+
1786
+static int issue_checkpoint_thread(void *data)
1787
+{
1788
+	struct f2fs_sb_info *sbi = data;
1789
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1790
+	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1791
+repeat:
1792
+	if (kthread_should_stop())
1793
+		return 0;
1794
+
1795
+	if (!llist_empty(&cprc->issue_list))
1796
+		__checkpoint_and_complete_reqs(sbi);
1797
+
1798
+	wait_event_interruptible(*q,
1799
+		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1800
+	goto repeat;
1801
+}
1802
+
1803
+static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1804
+		struct ckpt_req *wait_req)
1805
+{
1806
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1807
+
1808
+	if (!llist_empty(&cprc->issue_list)) {
1809
+		__checkpoint_and_complete_reqs(sbi);
1810
+	} else {
1811
+		/* already dispatched by issue_checkpoint_thread */
1812
+		if (wait_req)
1813
+			wait_for_completion(&wait_req->wait);
1814
+	}
1815
+}
1816
+
1817
+static void init_ckpt_req(struct ckpt_req *req)
1818
+{
1819
+	memset(req, 0, sizeof(struct ckpt_req));
1820
+
1821
+	init_completion(&req->wait);
1822
+	req->queue_time = ktime_get();
1823
+}
1824
+
1825
+int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1826
+{
1827
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1828
+	struct ckpt_req req;
1829
+	struct cp_control cpc;
1830
+
1831
+	cpc.reason = __get_cp_reason(sbi);
1832
+	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1833
+		int ret;
1834
+
1835
+		f2fs_down_write(&sbi->gc_lock);
1836
+		ret = f2fs_write_checkpoint(sbi, &cpc);
1837
+		f2fs_up_write(&sbi->gc_lock);
1838
+
1839
+		return ret;
1840
+	}
1841
+
1842
+	if (!cprc->f2fs_issue_ckpt)
1843
+		return __write_checkpoint_sync(sbi);
1844
+
1845
+	init_ckpt_req(&req);
1846
+
1847
+	llist_add(&req.llnode, &cprc->issue_list);
1848
+	atomic_inc(&cprc->queued_ckpt);
1849
+
1850
+	/*
1851
+	 * update issue_list before we wake up issue_checkpoint thread,
1852
+	 * this smp_mb() pairs with another barrier in ___wait_event(),
1853
+	 * see more details in comments of waitqueue_active().
1854
+	 */
1855
+	smp_mb();
1856
+
1857
+	if (waitqueue_active(&cprc->ckpt_wait_queue))
1858
+		wake_up(&cprc->ckpt_wait_queue);
1859
+
1860
+	if (cprc->f2fs_issue_ckpt)
1861
+		wait_for_completion(&req.wait);
1862
+	else
1863
+		flush_remained_ckpt_reqs(sbi, &req);
1864
+
1865
+	return req.ret;
1866
+}
1867
+
1868
+int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1869
+{
1870
+	dev_t dev = sbi->sb->s_bdev->bd_dev;
1871
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1872
+
1873
+	if (cprc->f2fs_issue_ckpt)
1874
+		return 0;
1875
+
1876
+	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1877
+			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1878
+	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1879
+		cprc->f2fs_issue_ckpt = NULL;
1880
+		return -ENOMEM;
1881
+	}
1882
+
1883
+	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1884
+
1885
+	return 0;
1886
+}
1887
+
1888
+void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1889
+{
1890
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1891
+	struct task_struct *ckpt_task;
1892
+
1893
+	if (!cprc->f2fs_issue_ckpt)
1894
+		return;
1895
+
1896
+	ckpt_task = cprc->f2fs_issue_ckpt;
1897
+	cprc->f2fs_issue_ckpt = NULL;
1898
+	kthread_stop(ckpt_task);
1899
+
1900
+	f2fs_flush_ckpt_thread(sbi);
1901
+}
1902
+
1903
+void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1904
+{
1905
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1906
+
1907
+	flush_remained_ckpt_reqs(sbi, NULL);
1908
+
1909
+	/* Let's wait for the previous dispatched checkpoint. */
1910
+	while (atomic_read(&cprc->queued_ckpt))
1911
+		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1912
+}
1913
+
1914
+void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1915
+{
1916
+	struct ckpt_req_control *cprc = &sbi->cprc_info;
1917
+
1918
+	atomic_set(&cprc->issued_ckpt, 0);
1919
+	atomic_set(&cprc->total_ckpt, 0);
1920
+	atomic_set(&cprc->queued_ckpt, 0);
1921
+	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1922
+	init_waitqueue_head(&cprc->ckpt_wait_queue);
1923
+	init_llist_head(&cprc->issue_list);
1924
+	spin_lock_init(&cprc->stat_lock);
1925
+}