f2fs: avoid using __GFP_NOFAIL

From: Chao Yu <[email protected]>

This patch slightly changes need_do_checkpoint to return the detail info that indicates why we need do checkpoint, then caller could print it with trace message.

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/file.c              | 34 ++++++++++++++++++----------------
 include/trace/events/f2fs.h | 12 ++++++------
 2 files changed, 24 insertions(+), 22 deletions(-)

fs/f2fs/file.c +8 -7

index cee0f36ba9a1..d9d8b1c372f8 100644
@@ -141,27 +141,29 @@ static int get_parent_ino(struct inode *inode, nid_t *pino)
 	return 1;
 }
 
-static inline bool need_do_checkpoint(struct inode *inode)
+static inline int need_do_checkpoint(struct inode *inode)
 {
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
-	bool need_cp = false;
+	int cp_reason = 0;
 
-	if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1)
-		need_cp = true;
+	if (!S_ISREG(inode->i_mode))
+		cp_reason = 1;
+	else if (inode->i_nlink != 1)
+		cp_reason = 2;
 	else if (is_sbi_flag_set(sbi, SBI_NEED_CP))
-		need_cp = true;
+		cp_reason = 3;
 	else if (file_wrong_pino(inode))
-		need_cp = true;
+		cp_reason = 4;
 	else if (!space_for_roll_forward(sbi))
-		need_cp = true;

fs/f2fs/file.c +6 -6

-	return need_cp;
+	return cp_reason;
 }
 
 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino)

@@ -196,7 +198,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	nid_t ino = inode->i_ino;
 	int ret = 0;
-	bool need_cp = false;
+	bool cp_reason = 0;
 	struct writeback_control wbc = {
 		.sync_mode = WB_SYNC_ALL,
 		.nr_to_write = LONG_MAX,

@@ -215,7 +217,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	clear_inode_flag(inode, FI_NEED_IPU);
 
 	if (ret) {
-		trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
+		trace_f2fs_sync_file_exit(inode, cp_reason, datasync, ret);
 		return ret;
 	}
 

@@ -246,10 +248,10 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	 * sudden-power-off.
 	 */
 	down_read(&F2FS_I(inode)->i_sem);
-	need_cp = need_do_checkpoint(inode);
+	cp_reason = need_do_checkpoint(inode);
 	up_read(&F2FS_I(inode)->i_sem);
 
-	if (need_cp) {
+	if (cp_reason) {
 		/* all the dirty node pages should be flushed for POR */
 		ret = f2fs_sync_fs(inode->i_sb, 1);
 

@@ -306,7 +308,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	}
 	f2fs_update_time(sbi, REQ_TIME);
 out:
-	trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
+	trace_f2fs_sync_file_exit(inode, cp_reason, datasync, ret);
 	f2fs_trace_ios(NULL, 1);
 	return ret;
 }

include/trace/events/f2fs.h +6 -6

index 009745564214..f1a317e6db79 100644
@@ -210,14 +210,14 @@ DEFINE_EVENT(f2fs__inode, f2fs_sync_file_enter,
 
 TRACE_EVENT(f2fs_sync_file_exit,
 
-	TP_PROTO(struct inode *inode, int need_cp, int datasync, int ret),
+	TP_PROTO(struct inode *inode, int cp_reason, int datasync, int ret),
 
-	TP_ARGS(inode, need_cp, datasync, ret),
+	TP_ARGS(inode, cp_reason, datasync, ret),
 
 	TP_STRUCT__entry(
 		__field(dev_t,	dev)
 		__field(ino_t,	ino)
-		__field(int,	need_cp)
+		__field(int,	cp_reason)
 		__field(int,	datasync)
 		__field(int,	ret)
 	),

@@ -225,15 +225,15 @@ TRACE_EVENT(f2fs_sync_file_exit,
 	TP_fast_assign(
 		__entry->dev		= inode->i_sb->s_dev;
 		__entry->ino		= inode->i_ino;
-		__entry->need_cp	= need_cp;
+		__entry->cp_reason	= cp_reason;
 		__entry->datasync	= datasync;
 		__entry->ret		= ret;
 	),
 
-	TP_printk("dev = (%d,%d), ino = %lu, checkpoint is %s, "
+	TP_printk("dev = (%d,%d), ino = %lu, cp_reason:%d, "
 		"datasync = %d, ret = %d",
 		show_dev_ino(__entry),
-		__entry->need_cp ? "needed" : "not needed",
+		__entry->cp_reason,
 		__entry->datasync,
 		__entry->ret)
 );

Signature · 1 line

2.14.1.145.gb3622a4ee

Quoted 46 lines · Chao Yu

From: Chao Yu <[email protected]>

This patch slightly changes need_do_checkpoint to return the detail info that indicates why we need do checkpoint, then caller could print it with trace message.

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/file.c              | 34 ++++++++++++++++++----------------
 include/trace/events/f2fs.h | 12 ++++++------
 2 files changed, 24 insertions(+), 22 deletions(-)

fs/f2fs/file.c +10 -8

index cee0f36ba9a1..d9d8b1c372f8 100644
@@ -141,27 +141,29 @@ static int get_parent_ino(struct inode *inode, nid_t *pino)
 	return 1;
 }
 
-static inline bool need_do_checkpoint(struct inode *inode)
+static inline int need_do_checkpoint(struct inode *inode)
 {
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
-	bool need_cp = false;
+	int cp_reason = 0;
 
-	if (!S_ISREG(inode->i_mode) || inode->i_nlink != 1)
-		need_cp = true;
+	if (!S_ISREG(inode->i_mode))
+		cp_reason = 1;
+	else if (inode->i_nlink != 1)
+		cp_reason = 2;
 	else if (is_sbi_flag_set(sbi, SBI_NEED_CP))
-		need_cp = true;
+		cp_reason = 3;
 	else if (file_wrong_pino(inode))
-		need_cp = true;
+		cp_reason = 4;
 	else if (!space_for_roll_forward(sbi))
-		need_cp = true;
+		cp_reason = 5;
 	else if (!is_checkpointed_node(sbi, F2FS_I(inode)->i_pino))
-		need_cp = true;
+		cp_reason = 6;
 	else if (test_opt(sbi, FASTBOOT))

fs/f2fs/file.c static int get_parent_ino(struct inode *inode, nid_t *pino) · line 164

Quoted 5 lines

fs/f2fs/file.c +2 -2

-		need_cp = true;
+		cp_reason = 7;
 	else if (sbi->active_logs == 2)
-		need_cp = true;
+		cp_reason = 8;

We need to make the numbers by enum so that we could interpret them in tracepoint below.

Thanks,

Quoted 89 lines

fs/f2fs/file.c +6 -6

-	return need_cp;
+	return cp_reason;
 }
 
 static bool need_inode_page_update(struct f2fs_sb_info *sbi, nid_t ino)

@@ -196,7 +198,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 	nid_t ino = inode->i_ino;
 	int ret = 0;
-	bool need_cp = false;
+	bool cp_reason = 0;
 	struct writeback_control wbc = {
 		.sync_mode = WB_SYNC_ALL,
 		.nr_to_write = LONG_MAX,

@@ -215,7 +217,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	clear_inode_flag(inode, FI_NEED_IPU);
 
 	if (ret) {
-		trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
+		trace_f2fs_sync_file_exit(inode, cp_reason, datasync, ret);
 		return ret;
 	}
 

@@ -246,10 +248,10 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	 * sudden-power-off.
 	 */
 	down_read(&F2FS_I(inode)->i_sem);
-	need_cp = need_do_checkpoint(inode);
+	cp_reason = need_do_checkpoint(inode);
 	up_read(&F2FS_I(inode)->i_sem);
 
-	if (need_cp) {
+	if (cp_reason) {
 		/* all the dirty node pages should be flushed for POR */
 		ret = f2fs_sync_fs(inode->i_sb, 1);
 

@@ -306,7 +308,7 @@ static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
 	}
 	f2fs_update_time(sbi, REQ_TIME);
 out:
-	trace_f2fs_sync_file_exit(inode, need_cp, datasync, ret);
+	trace_f2fs_sync_file_exit(inode, cp_reason, datasync, ret);
 	f2fs_trace_ios(NULL, 1);
 	return ret;
 }

include/trace/events/f2fs.h +6 -6

index 009745564214..f1a317e6db79 100644
@@ -210,14 +210,14 @@ DEFINE_EVENT(f2fs__inode, f2fs_sync_file_enter,
 
 TRACE_EVENT(f2fs_sync_file_exit,
 
-	TP_PROTO(struct inode *inode, int need_cp, int datasync, int ret),
+	TP_PROTO(struct inode *inode, int cp_reason, int datasync, int ret),
 
-	TP_ARGS(inode, need_cp, datasync, ret),
+	TP_ARGS(inode, cp_reason, datasync, ret),
 
 	TP_STRUCT__entry(
 		__field(dev_t,	dev)
 		__field(ino_t,	ino)
-		__field(int,	need_cp)
+		__field(int,	cp_reason)
 		__field(int,	datasync)
 		__field(int,	ret)
 	),

@@ -225,15 +225,15 @@ TRACE_EVENT(f2fs_sync_file_exit,
 	TP_fast_assign(
 		__entry->dev		= inode->i_sb->s_dev;
 		__entry->ino		= inode->i_ino;
-		__entry->need_cp	= need_cp;
+		__entry->cp_reason	= cp_reason;
 		__entry->datasync	= datasync;
 		__entry->ret		= ret;
 	),
 
-	TP_printk("dev = (%d,%d), ino = %lu, checkpoint is %s, "
+	TP_printk("dev = (%d,%d), ino = %lu, cp_reason:%d, "
 		"datasync = %d, ret = %d",
 		show_dev_ino(__entry),
-		__entry->need_cp ? "needed" : "not needed",
+		__entry->cp_reason,
 		__entry->datasync,
 		__entry->ret)
 );

Signature · 1 line

2.14.1.145.gb3622a4ee

.

From: Chao Yu <[email protected]>

We will keep __add_ino_entry success all the time, for ENOMEM failure case, we have already handled it with an opened loop code, so we don't have to use redundant __GFP_NOFAIL in radix_tree_preload, remove it.

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/checkpoint.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

fs/f2fs/checkpoint.c +7 -3

index 98777c1ae70c..43ee9d97fd8f 100644
@@ -405,10 +405,11 @@ static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 {
 	struct inode_management *im = &sbi->im[type];
 	struct ino_entry *e, *tmp;
+	bool preloaded;
 
 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 retry:
-	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
+	preloaded = !radix_tree_preload(GFP_NOFS);
 
 	spin_lock(&im->ino_lock);
 	e = radix_tree_lookup(&im->ino_root, ino);

@@ -416,7 +417,8 @@ static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 		e = tmp;
 		if (radix_tree_insert(&im->ino_root, ino, e)) {
 			spin_unlock(&im->ino_lock);
-			radix_tree_preload_end();
+			if (preloaded)
+				radix_tree_preload_end();
 			goto retry;
 		}
 		memset(e, 0, sizeof(struct ino_entry));

@@ -431,7 +433,9 @@ static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
 
 	spin_unlock(&im->ino_lock);
-	radix_tree_preload_end();
+
+	if (preloaded)
+		radix_tree_preload_end();
 
 	if (e != tmp)
 		kmem_cache_free(ino_entry_slab, tmp);

Signature · 1 line

2.14.1.145.gb3622a4ee

Quoted 5 lines · Chao Yu

From: Chao Yu <[email protected]>

We will keep __add_ino_entry success all the time, for ENOMEM failure case, we have already handled it with an opened loop code, so we don't have to use redundant __GFP_NOFAIL in radix_tree_preload, remove it.

Why do you think an open coded allocation retry loop is better than
having the MM do all it can when the nofail is requested explicitly?> E.g. giving it an access to memory reserves to allow forward progress.

Quoted 46 lines

Signed-off-by Chao Yu <[email protected]>

---
 fs/f2fs/checkpoint.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

fs/f2fs/checkpoint.c +7 -3

index 98777c1ae70c..43ee9d97fd8f 100644
@@ -405,10 +405,11 @@ static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 {
 	struct inode_management *im = &sbi->im[type];
 	struct ino_entry *e, *tmp;
+	bool preloaded;
 
 	tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
 retry:
-	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
+	preloaded = !radix_tree_preload(GFP_NOFS);
 
 	spin_lock(&im->ino_lock);
 	e = radix_tree_lookup(&im->ino_root, ino);

@@ -416,7 +417,8 @@ static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 		e = tmp;
 		if (radix_tree_insert(&im->ino_root, ino, e)) {
 			spin_unlock(&im->ino_lock);
-			radix_tree_preload_end();
+			if (preloaded)
+				radix_tree_preload_end();
 			goto retry;
 		}
 		memset(e, 0, sizeof(struct ino_entry));

@@ -431,7 +433,9 @@ static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 		f2fs_set_bit(devidx, (char *)&e->dirty_device);
 
 	spin_unlock(&im->ino_lock);
-	radix_tree_preload_end();
+
+	if (preloaded)
+		radix_tree_preload_end();
 
 	if (e != tmp)
 		kmem_cache_free(ino_entry_slab, tmp);

Signature · 1 line

2.14.1.145.gb3622a4ee

BTW, I notice the comments of __GFP_NOFAIL, what does this mean?
* Using this flag for costly allocations is _highly_ discouraged.

On Tue 07-11-17 00:08:25, Chao Yu wrote:
[...]

Quoted 2 lines

BTW, I notice the comments of __GFP_NOFAIL, what does this mean?
* Using this flag for costly allocations is _highly_ discouraged.

This means that using __GFP_NOFAIL for high order allocations (especially those with order > PAGE_ALLOC_COSTLY_ORDER) are highly discouraged because we those are quite hard to get and looping inside the allocator basically for ever is not a wise thing to do. That being said replacing __GFP_NOFAIL by an open coded retry loop might make sense for those e.g. to check signals or other termination conditions.