Re: [f2fs-dev] [PATCH] f2fs: prevent meta updates while checkpoint is in progress

[Jaegeuk Kim <jaegeuk@kernel.org>]

On 04/14, Jaegeuk Kim wrote:
> On 04/13, Jaegeuk Kim wrote:
> > On 04/03, Jaegeuk Kim wrote:
> > > On 04/03, Jaegeuk Kim wrote:
> > > > On 04/01, Sahitya Tummala wrote:
> > > > > Hi Jaegeuk,
> > > > > 
> > > > > Got it.
> > > > > The diff below looks good to me.
> > > > > Would you like me to test it and put a patch for this?
> > > > 
> > > > Sahitya, Chao,
> > > > 
> > > > Could you please take a look at this patch and test intensively?
> > > > 
> > > > Thanks,

v5:
 - add signal handler

Sahitya raised an issue:
- prevent meta updates while checkpoint is in progress

allocate_segment_for_resize() can cause metapage updates if
it requires to change the current node/data segments for resizing.
Stop these meta updates when there is a checkpoint already
in progress to prevent inconsistent CP data.

Signed-off-by: Sahitya Tummala <stummala@codeaurora.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 fs/f2fs/checkpoint.c        |   6 +-
 fs/f2fs/f2fs.h              |   2 +-
 fs/f2fs/file.c              |   5 +-
 fs/f2fs/gc.c                | 114 ++++++++++++++++++++----------------
 fs/f2fs/super.c             |   1 -
 include/trace/events/f2fs.h |   4 +-
 6 files changed, 74 insertions(+), 58 deletions(-)

diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
index 6be357c8e0020..dcb3a15574c99 100644
--- a/fs/f2fs/checkpoint.c
+++ b/fs/f2fs/checkpoint.c
@@ -1554,7 +1554,8 @@ int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 			return 0;
 		f2fs_warn(sbi, "Start checkpoint disabled!");
 	}
-	mutex_lock(&sbi->cp_mutex);
+	if (cpc->reason != CP_RESIZE)
+		mutex_lock(&sbi->cp_mutex);
 
 	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
 		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
@@ -1623,7 +1624,8 @@ int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 	f2fs_update_time(sbi, CP_TIME);
 	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
 out:
-	mutex_unlock(&sbi->cp_mutex);
+	if (cpc->reason != CP_RESIZE)
+		mutex_unlock(&sbi->cp_mutex);
 	return err;
 }
 
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index dd9c1de7c59d2..9dd1fc957b943 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -194,6 +194,7 @@ enum {
 #define	CP_DISCARD	0x00000010
 #define CP_TRIMMED	0x00000020
 #define CP_PAUSE	0x00000040
+#define CP_RESIZE 	0x00000080
 
 #define MAX_DISCARD_BLOCKS(sbi)		BLKS_PER_SEC(sbi)
 #define DEF_MAX_DISCARD_REQUEST		8	/* issue 8 discards per round */
@@ -1423,7 +1424,6 @@ struct f2fs_sb_info {
 	unsigned int segs_per_sec;		/* segments per section */
 	unsigned int secs_per_zone;		/* sections per zone */
 	unsigned int total_sections;		/* total section count */
-	struct mutex resize_mutex;		/* for resize exclusion */
 	unsigned int total_node_count;		/* total node block count */
 	unsigned int total_valid_node_count;	/* valid node block count */
 	loff_t max_file_blocks;			/* max block index of file */
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index dc470358f25eb..212c5996d3807 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -3306,7 +3306,6 @@ static int f2fs_ioc_resize_fs(struct file *filp, unsigned long arg)
 {
 	struct f2fs_sb_info *sbi = F2FS_I_SB(file_inode(filp));
 	__u64 block_count;
-	int ret;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
@@ -3318,9 +3317,7 @@ static int f2fs_ioc_resize_fs(struct file *filp, unsigned long arg)
 			   sizeof(block_count)))
 		return -EFAULT;
 
-	ret = f2fs_resize_fs(sbi, block_count);
-
-	return ret;
+	return f2fs_resize_fs(sbi, block_count);
 }
 
 static int f2fs_ioc_enable_verity(struct file *filp, unsigned long arg)
diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index 26248c8936db0..c979dd1add5de 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -1399,12 +1399,29 @@ void f2fs_build_gc_manager(struct f2fs_sb_info *sbi)
 				GET_SEGNO(sbi, FDEV(0).end_blk) + 1;
 }
 
-static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
-							unsigned int end)
+static int free_segment_range(struct f2fs_sb_info *sbi,
+				unsigned int secs, bool gc_only)
 {
-	int type;
-	unsigned int segno, next_inuse;
+	unsigned int segno, next_inuse, start, end;
+	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
+	int gc_mode, gc_type;
 	int err = 0;
+	int type;
+
+	/* Force block allocation for GC */
+	MAIN_SECS(sbi) -= secs;
+	start = MAIN_SECS(sbi) * sbi->segs_per_sec;
+	end = MAIN_SEGS(sbi) - 1;
+
+	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
+	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
+		if (SIT_I(sbi)->last_victim[gc_mode] >= start)
+			SIT_I(sbi)->last_victim[gc_mode] = 0;
+
+	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
+		if (sbi->next_victim_seg[gc_type] >= start)
+			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
+	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
 
 	/* Move out cursegs from the target range */
 	for (type = CURSEG_HOT_DATA; type < NR_CURSEG_TYPE; type++)
@@ -1417,18 +1434,22 @@ static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
 			.iroot = RADIX_TREE_INIT(gc_list.iroot, GFP_NOFS),
 		};
 
-		down_write(&sbi->gc_lock);
 		do_garbage_collect(sbi, segno, &gc_list, FG_GC);
-		up_write(&sbi->gc_lock);
 		put_gc_inode(&gc_list);
 
-		if (get_valid_blocks(sbi, segno, true))
-			return -EAGAIN;
+		if (!gc_only && get_valid_blocks(sbi, segno, true)) {
+			err = -EAGAIN;
+			goto out;
+		}
+		if (fatal_signal_pending(current))
+			return -ERESTARTSYS;
 	}
+	if (gc_only)
+		goto out;
 
-	err = f2fs_sync_fs(sbi->sb, 1);
+	err = f2fs_write_checkpoint(sbi, &cpc);
 	if (err)
-		return err;
+		goto out;
 
 	next_inuse = find_next_inuse(FREE_I(sbi), end + 1, start);
 	if (next_inuse <= end) {
@@ -1436,6 +1457,8 @@ static int free_segment_range(struct f2fs_sb_info *sbi, unsigned int start,
 			 next_inuse);
 		f2fs_bug_on(sbi, 1);
 	}
+out:
+	MAIN_SECS(sbi) -= secs;
 	return err;
 }
 
@@ -1481,6 +1504,7 @@ static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
 
 	SM_I(sbi)->segment_count = (int)SM_I(sbi)->segment_count + segs;
 	MAIN_SEGS(sbi) = (int)MAIN_SEGS(sbi) + segs;
+	MAIN_SECS(sbi) += secs;
 	FREE_I(sbi)->free_sections = (int)FREE_I(sbi)->free_sections + secs;
 	FREE_I(sbi)->free_segments = (int)FREE_I(sbi)->free_segments + segs;
 	F2FS_CKPT(sbi)->user_block_count = cpu_to_le64(user_block_count + blks);
@@ -1502,8 +1526,8 @@ static void update_fs_metadata(struct f2fs_sb_info *sbi, int secs)
 int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
 {
 	__u64 old_block_count, shrunk_blocks;
+	struct cp_control cpc = { CP_RESIZE, 0, 0, 0 };
 	unsigned int secs;
-	int gc_mode, gc_type;
 	int err = 0;
 	__u32 rem;
 
@@ -1538,10 +1562,27 @@ int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
 		return -EINVAL;
 	}
 
-	freeze_bdev(sbi->sb->s_bdev);
-
 	shrunk_blocks = old_block_count - block_count;
 	secs = div_u64(shrunk_blocks, BLKS_PER_SEC(sbi));
+
+	/* stop other GC */
+	if (!down_write_trylock(&sbi->gc_lock))
+		return -EAGAIN;
+
+	/* stop CP to protect MAIN_SEC in free_segment_range */
+	f2fs_lock_op(sbi);
+	err = free_segment_range(sbi, secs, true);
+	f2fs_unlock_op(sbi);
+	up_write(&sbi->gc_lock);
+	if (err)
+		return err;
+
+	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
+
+	freeze_super(sbi->sb);
+	down_write(&sbi->gc_lock);
+	mutex_lock(&sbi->cp_mutex);
+
 	spin_lock(&sbi->stat_lock);
 	if (shrunk_blocks + valid_user_blocks(sbi) +
 		sbi->current_reserved_blocks + sbi->unusable_block_count +
@@ -1550,69 +1591,44 @@ int f2fs_resize_fs(struct f2fs_sb_info *sbi, __u64 block_count)
 	else
 		sbi->user_block_count -= shrunk_blocks;
 	spin_unlock(&sbi->stat_lock);
-	if (err) {
-		thaw_bdev(sbi->sb->s_bdev, sbi->sb);
-		return err;
-	}
-
-	mutex_lock(&sbi->resize_mutex);
-	set_sbi_flag(sbi, SBI_IS_RESIZEFS);
-
-	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
-
-	MAIN_SECS(sbi) -= secs;
-
-	for (gc_mode = 0; gc_mode < MAX_GC_POLICY; gc_mode++)
-		if (SIT_I(sbi)->last_victim[gc_mode] >=
-					MAIN_SECS(sbi) * sbi->segs_per_sec)
-			SIT_I(sbi)->last_victim[gc_mode] = 0;
-
-	for (gc_type = BG_GC; gc_type <= FG_GC; gc_type++)
-		if (sbi->next_victim_seg[gc_type] >=
-					MAIN_SECS(sbi) * sbi->segs_per_sec)
-			sbi->next_victim_seg[gc_type] = NULL_SEGNO;
-
-	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
+	if (err)
+		goto out_err;
 
-	err = free_segment_range(sbi, MAIN_SECS(sbi) * sbi->segs_per_sec,
-			MAIN_SEGS(sbi) - 1);
+	err = free_segment_range(sbi, secs, false);
 	if (err)
-		goto out;
+		goto recover_out;
 
 	update_sb_metadata(sbi, -secs);
 
 	err = f2fs_commit_super(sbi, false);
 	if (err) {
 		update_sb_metadata(sbi, secs);
-		goto out;
+		goto recover_out;
 	}
 
-	mutex_lock(&sbi->cp_mutex);
 	update_fs_metadata(sbi, -secs);
 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
 	set_sbi_flag(sbi, SBI_IS_DIRTY);
-	mutex_unlock(&sbi->cp_mutex);
 
-	err = f2fs_sync_fs(sbi->sb, 1);
+	err = f2fs_write_checkpoint(sbi, &cpc);
 	if (err) {
-		mutex_lock(&sbi->cp_mutex);
 		update_fs_metadata(sbi, secs);
-		mutex_unlock(&sbi->cp_mutex);
 		update_sb_metadata(sbi, secs);
 		f2fs_commit_super(sbi, false);
 	}
-out:
+recover_out:
 	if (err) {
 		set_sbi_flag(sbi, SBI_NEED_FSCK);
 		f2fs_err(sbi, "resize_fs failed, should run fsck to repair!");
 
-		MAIN_SECS(sbi) += secs;
 		spin_lock(&sbi->stat_lock);
 		sbi->user_block_count += shrunk_blocks;
 		spin_unlock(&sbi->stat_lock);
 	}
+out_err:
+	mutex_unlock(&sbi->cp_mutex);
+	up_write(&sbi->gc_lock);
+	thaw_super(sbi->sb);
 	clear_sbi_flag(sbi, SBI_IS_RESIZEFS);
-	mutex_unlock(&sbi->resize_mutex);
-	thaw_bdev(sbi->sb->s_bdev, sbi->sb);
 	return err;
 }
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index e3a323ff04c34..ad3b66c3dbe0e 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -3420,7 +3420,6 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
 	init_rwsem(&sbi->gc_lock);
 	mutex_init(&sbi->writepages);
 	mutex_init(&sbi->cp_mutex);
-	mutex_init(&sbi->resize_mutex);
 	init_rwsem(&sbi->node_write);
 	init_rwsem(&sbi->node_change);
 
diff --git a/include/trace/events/f2fs.h b/include/trace/events/f2fs.h
index 3577fa67690af..421a661bfd5aa 100644
--- a/include/trace/events/f2fs.h
+++ b/include/trace/events/f2fs.h
@@ -50,6 +50,7 @@ TRACE_DEFINE_ENUM(CP_RECOVERY);
 TRACE_DEFINE_ENUM(CP_DISCARD);
 TRACE_DEFINE_ENUM(CP_TRIMMED);
 TRACE_DEFINE_ENUM(CP_PAUSE);
+TRACE_DEFINE_ENUM(CP_RESIZE);
 
 #define show_block_type(type)						\
 	__print_symbolic(type,						\
@@ -126,7 +127,8 @@ TRACE_DEFINE_ENUM(CP_PAUSE);
 		{ CP_RECOVERY,	"Recovery" },				\
 		{ CP_DISCARD,	"Discard" },				\
 		{ CP_PAUSE,	"Pause" },				\
-		{ CP_TRIMMED,	"Trimmed" })
+		{ CP_TRIMMED,	"Trimmed" },				\
+		{ CP_RESIZE,	"Resize" })
 
 #define show_fsync_cpreason(type)					\
 	__print_symbolic(type,						\
-- 
2.26.1.301.g55bc3eb7cb9-goog