Skip to contentSkip to conversation

f2fs: write checkpoint during FG_GC

PATCH In review Jaegeuk Kim 2 participants 6 messages May 3, 2022 – May 5, 2022 F2FS file system

lore
Share & tools
lore

Jaegeuk Kim

author maintainer
#0 lore raw

PATCH 1/2 f2fs: write checkpoint during FG_GC

From: Byungki Lee <[email protected]>

If there's not enough free sections each of which consistis of large segments, we can hit no free section for upcoming section allocation. Let's reclaim some prefree segments by writing checkpoints.

---
 fs/f2fs/gc.c | 38 +++++++++++++++++++++++---------------
 1 file changed, 23 insertions(+), 15 deletions(-)
fs/f2fs/gc.c +23 -15
index eddd2e81a2a2..4d47723523c3 100644
@@ -1841,23 +1841,31 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
 	if (sync)
 		goto stop;
 
-	if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
-		if (skipped_round <= MAX_SKIP_GC_COUNT ||
-					skipped_round * 2 < round) {
-			segno = NULL_SEGNO;
-			goto gc_more;
-		}
+	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
+		goto stop;
 
-		if (first_skipped < last_skipped &&
-				(last_skipped - first_skipped) >
-						sbi->skipped_gc_rwsem) {
-			f2fs_drop_inmem_pages_all(sbi, true);
-			segno = NULL_SEGNO;
-			goto gc_more;
-		}
-		if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
+	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
+
+		/* Write checkpoint to reclaim prefree segments */
+		if (free_sections(sbi) < NR_CURSEG_PERSIST_TYPE &&
+				prefree_segments(sbi) &&
+				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
 			ret = f2fs_write_checkpoint(sbi, &cpc);
-	}
+			if (ret)
+				goto stop;
+		}
+		segno = NULL_SEGNO;
+		goto gc_more;
+	}
+	if (first_skipped < last_skipped &&
+			(last_skipped - first_skipped) >
+					sbi->skipped_gc_rwsem) {
+		f2fs_drop_inmem_pages_all(sbi, true);
+		segno = NULL_SEGNO;
+		goto gc_more;
+	}
+	if (gc_type == FG_GC && !is_sbi_flag_set(sbi, SBI_CP_DISABLED))
+		ret = f2fs_write_checkpoint(sbi, &cpc);
 stop:
 	SIT_I(sbi)->last_victim[ALLOC_NEXT] = 0;
 	SIT_I(sbi)->last_victim[FLUSH_DEVICE] = init_segno;
Signature · 8 lines
2.36.0.464.gb9c8b46e94-goog



_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
author maintainer
#1 lore raw

PATCH 2/2 f2fs: avoid to select pinned section during checkpoint=disable

The f2fs_gc uses a bitmap to indicate pinned sections, but when disabling chckpoint, we call f2fs_gc() with NULL_SEGNO which selects the same dirty segment as a victim all the time, resulting in checkpoint=disable failure. Let's pick another one, if we fail to collect it.

---
 fs/f2fs/gc.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
fs/f2fs/gc.c +1 -4
index 4d47723523c3..5ba8104e138b 100644
@@ -1838,10 +1838,7 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
 	if (gc_type == FG_GC)
 		sbi->cur_victim_sec = NULL_SEGNO;
 
-	if (sync)
-		goto stop;
-
-	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
+	if (!has_not_enough_free_secs(sbi, sec_freed, 0) && seg_freed)
 		goto stop;
 
 	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
Signature · 1 line
2.36.0.464.gb9c8b46e94-goog
maintainer
#2 lore raw

Re: [f2fs-dev] [PATCH 2/2] f2fs: avoid to select pinned section during checkpoint=disable

Quoted 15 lines · Jaegeuk Kim

The f2fs_gc uses a bitmap to indicate pinned sections, but when disabling chckpoint, we call f2fs_gc() with NULL_SEGNO which selects the same dirty segment as a victim all the time, resulting in checkpoint=disable failure. Let's pick another one, if we fail to collect it.

---
  fs/f2fs/gc.c | 5 +----
  1 file changed, 1 insertion(+), 4 deletions(-)
fs/f2fs/gc.c +0 -0
index 4d47723523c3..5ba8104e138b 100644

fs/f2fs/gc.c int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, · line 1842

Quoted 5 lines
fs/f2fs/gc.c +0 -2
@@ -1838,10 +1838,7 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
  	if (gc_type == FG_GC)
  		sbi->cur_victim_sec = NULL_SEGNO;
 
-	if (sync)
-		goto stop;

Semantics of 'sync' parameter should be: migrate one section at least before exit?

So, should it be?

if (seg_freed && (sync || !has_not_enough_free_secs(sbi, sec_freed, 0)))

	goto stop;

Thanks,

Quoted 6 lines
fs/f2fs/gc.c +1 -2
-
-	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
+	if (!has_not_enough_free_secs(sbi, sec_freed, 0) && seg_freed)
  		goto stop;
 
  	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
author maintainer
#3 lore raw

Re: [PATCH 2/2 v2] f2fs: avoid to select pinned section during checkpoint=disable

In reply to Jaegeuk Kim

The f2fs_gc uses a bitmap to indicate pinned sections, but when disabling chckpoint, we call f2fs_gc() with NULL_SEGNO which selects the same dirty segment as a victim all the time, resulting in checkpoint=disable failure. Let's pick another one, if we fail to collect it.

---

Change log from v1:
- keep sync condition to stop GC

 fs/f2fs/gc.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
fs/f2fs/gc.c +1 -4
index 4d47723523c3..441190ab8593 100644
@@ -1838,10 +1838,7 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
 	if (gc_type == FG_GC)
 		sbi->cur_victim_sec = NULL_SEGNO;
 
-	if (sync)
-		goto stop;
-
-	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
+	if ((!has_not_enough_free_secs(sbi, sec_freed, 0) || sync) && seg_freed)
 		goto stop;
 
 	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
Signature · 1 line
2.36.0.464.gb9c8b46e94-goog
maintainer
#4 lore raw

Re: [f2fs-dev] [PATCH 2/2 v2] f2fs: avoid to select pinned section during checkpoint=disable

Quoted 19 lines · Jaegeuk Kim

The f2fs_gc uses a bitmap to indicate pinned sections, but when disabling chckpoint, we call f2fs_gc() with NULL_SEGNO which selects the same dirty segment as a victim all the time, resulting in checkpoint=disable failure. Let's pick another one, if we fail to collect it.

---

Change log from v1:
- keep sync condition to stop GC

  fs/f2fs/gc.c | 5 +----
  1 file changed, 1 insertion(+), 4 deletions(-)
fs/f2fs/gc.c +0 -0
index 4d47723523c3..441190ab8593 100644

fs/f2fs/gc.c int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, · line 1841

Quoted 8 lines
fs/f2fs/gc.c +1 -4
@@ -1838,10 +1838,7 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
  	if (gc_type == FG_GC)
  		sbi->cur_victim_sec = NULL_SEGNO;
 
-	if (sync)
-		goto stop;
-
-	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
+	if ((!has_not_enough_free_secs(sbi, sec_freed, 0) || sync) && seg_freed)

Wait, it doesn't make sense that it needs to migrate one section successfully when there is enough free sections, e.g. for background GC case.

I doubt above implementation may force bggc to migrate all dirty sections? due to seg_freed will always be zero, because it increases only if gc_type is FG_GC?

		if (gc_type == FG_GC &&
				get_valid_blocks(sbi, segno, false) == 0)
			seg_freed++;

And I didn't see there is one or combinations of parameters can indicate
meaning: "it needs to migrate one section at least before exit", so how about introducing another parameter for this requirement?

From: Jaegeuk Kim <[email protected]>
Date: Sat, 30 Apr 2022 03:27:20 -0700
Subject: [PATCH] f2fs: avoid to select pinned section during
checkpoint=disable

The f2fs_gc uses a bitmap to indicate pinned sections, but when disabling chckpoint, we call f2fs_gc() with NULL_SEGNO which selects the same dirty segment as a victim all the time, resulting in checkpoint=disable failure. Let's pick another one, if we fail to collect it.

---
 fs/f2fs/f2fs.h    |  2 +-
 fs/f2fs/file.c    |  8 ++++----
 fs/f2fs/gc.c      | 17 ++++++++++++-----
 fs/f2fs/segment.c |  2 +-
 fs/f2fs/super.c   |  2 +-
 5 files changed, 19 insertions(+), 12 deletions(-)
fs/f2fs/f2fs.h +1 -1
index 56adc3b68e14..0f8c426aed50 100644
@@ -3787,7 +3787,7 @@ int f2fs_start_gc_thread(struct f2fs_sb_info *sbi);
 void f2fs_stop_gc_thread(struct f2fs_sb_info *sbi);
 block_t f2fs_start_bidx_of_node(unsigned int node_ofs, struct inode *inode);
 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, bool background, bool force,
-			unsigned int segno);
+			unsigned int segno, unsigned int nr_section);
 void f2fs_build_gc_manager(struct f2fs_sb_info *sbi);
 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count);
 int __init f2fs_create_garbage_collection_cache(void);
fs/f2fs/file.c +4 -4
index e4cf8b7b23aa..07d2fa6c774b 100644
@@ -1684,7 +1684,7 @@ static int expand_inode_data(struct inode *inode, loff_t offset,
 		if (has_not_enough_free_secs(sbi, 0,
 			GET_SEC_FROM_SEG(sbi, overprovision_segments(sbi)))) {
 			f2fs_down_write(&sbi->gc_lock);
-			err = f2fs_gc(sbi, true, false, false, NULL_SEGNO);
+			err = f2fs_gc(sbi, true, false, false, NULL_SEGNO, 0);
 			if (err && err != -ENODATA && err != -EAGAIN)
 				goto out_err;
 		}
@@ -2472,7 +2472,7 @@ static int f2fs_ioc_gc(struct file *filp, unsigned long arg)
 		f2fs_down_write(&sbi->gc_lock);
 	}
 
-	ret = f2fs_gc(sbi, sync, true, false, NULL_SEGNO);
+	ret = f2fs_gc(sbi, sync, true, false, NULL_SEGNO, 0);
 out:
 	mnt_drop_write_file(filp);
 	return ret;
@@ -2509,7 +2509,7 @@ static int __f2fs_ioc_gc_range(struct file *filp, struct f2fs_gc_range *range)
 	}
 
 	ret = f2fs_gc(sbi, range->sync, true, false,
-				GET_SEGNO(sbi, range->start));
+				GET_SEGNO(sbi, range->start), 0);
 	if (ret) {
 		if (ret == -EBUSY)
 			ret = -EAGAIN;
@@ -2966,7 +2966,7 @@ static int f2fs_ioc_flush_device(struct file *filp, unsigned long arg)
 		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, true, start_segno);
+		ret = f2fs_gc(sbi, true, true, true, start_segno, 0);
 		if (ret == -EAGAIN)
 			ret = 0;
 		else if (ret < 0)
fs/f2fs/gc.c +12 -5
index 4d47723523c3..2364ffe6bbbf 100644
@@ -142,7 +142,7 @@ static int gc_thread_func(void *data)
 			sync_mode = false;
 
 		/* if return value is not zero, no victim was selected */
-		if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO))
+		if (f2fs_gc(sbi, sync_mode, !foreground, false, NULL_SEGNO, 0))
 			wait_ms = gc_th->no_gc_sleep_time;
 
 		if (foreground)
@@ -1751,7 +1751,8 @@ static int do_garbage_collect(struct f2fs_sb_info *sbi,
 }
 
 int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
-			bool background, bool force, unsigned int segno)
+			bool background, bool force, unsigned int segno,
+			unsigned int nr_section)
 {
 	int gc_type = sync ? FG_GC : BG_GC;
 	int sec_freed = 0, seg_freed = 0, total_freed = 0;
@@ -1838,11 +1839,17 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
 	if (gc_type == FG_GC)
 		sbi->cur_victim_sec = NULL_SEGNO;
 
-	if (sync)
-		goto stop;
+	
 
-	if (!has_not_enough_free_secs(sbi, sec_freed, 0))
+	if (sync || !has_not_enough_free_secs(sbi, sec_freed, 0)) {
+		/*
+		 * f2fs_disable_checkpoint() needs to migrate enough section
+		 * before f2fs_gc() exits.
+		 */
+		if (total_freed < nr_section)
+			goto gc_more;
 		goto stop;
+	}
 
 	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {
 
fs/f2fs/segment.c +1 -1
index 8c17fed8987e..3a3e2cec2ac4 100644
@@ -524,7 +524,7 @@ void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
 			finish_wait(&sbi->gc_thread->fggc_wq, &wait);
 		} else {
 			f2fs_down_write(&sbi->gc_lock);
-			f2fs_gc(sbi, false, false, false, NULL_SEGNO);
+			f2fs_gc(sbi, false, false, false, NULL_SEGNO, 0);
 		}
 	}
 }
fs/f2fs/super.c +1 -1
index d06a577a1208..2f0ad9ffbf17 100644
@@ -2076,7 +2076,7 @@ static int f2fs_disable_checkpoint(struct f2fs_sb_info *sbi)
 
 	while (!f2fs_time_over(sbi, DISABLE_TIME)) {
 		f2fs_down_write(&sbi->gc_lock);
-		err = f2fs_gc(sbi, true, false, false, NULL_SEGNO);
+		err = f2fs_gc(sbi, true, false, false, NULL_SEGNO, 1);
 		if (err == -ENODATA) {
 			err = 0;
 			break;
Signature · 1 line
2.32.0

fs/f2fs/gc.c int f2fs_gc(struct f2fs_sb_info *sbi, bool sync, · line 1844

Quoted 3 lines
fs/f2fs/gc.c +0 -0
  		goto stop;
 
  	if (skipped_round <= MAX_SKIP_GC_COUNT || skipped_round * 2 < round) {

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

maintainer Reviewed-by on v1 1/2
#5 lore raw
In reply to Jaegeuk Kim
Quoted 8 lines · Jaegeuk Kim

From: Byungki Lee <[email protected]>

If there's not enough free sections each of which consistis of large segments, we can hit no free section for upcoming section allocation. Let's reclaim some prefree segments by writing checkpoints.

Thanks,