exfat: do not sync parent dir if just update timestamp

When sync or dir_sync is enabled, there is no need to sync the
parent directory's inode if only for updating its timestamp.

1. If an unexpected power failure occurs, the timestamp of the
   parent directory is not updated to the storage, which has no
   impact on the user.

2. The number of writes will be greatly reduced, which can not
   only improve performance, but also prolong device life.

Signed-off-by: Yuezhang Mo <[email protected]>
Reviewed-by: Andy Wu <[email protected]>
Reviewed-by: Aoyama Wataru <[email protected]>
Signed-off-by: Namjae Jeon <[email protected]>

namei.c

@@ -553,6 +553,7 @@ static int exfat_create(struct user_namespace *mnt_userns, struct inode *dir,
 	struct exfat_dir_entry info;
 	loff_t i_pos;
 	int err;
+	loff_t size = i_size_read(dir);
 
 	mutex_lock(&EXFAT_SB(sb)->s_lock);
 	exfat_set_volume_dirty(sb);
@@ -571,7 +572,7 @@ static int exfat_create(struct user_namespace *mnt_userns, struct inode *dir,
 #else
 	dir->i_ctime = dir->i_mtime = current_time(dir);
 #endif
-	if (IS_DIRSYNC(dir))
+	if (IS_DIRSYNC(dir) && size != i_size_read(dir))
 		exfat_sync_inode(dir);
 	else
 		mark_inode_dirty(dir);
@@ -835,10 +836,7 @@ static int exfat_unlink(struct inode *dir, struct dentry *dentry)
 #endif
 	exfat_truncate_atime(&dir->i_atime);
 #endif
-	if (IS_DIRSYNC(dir))
-		exfat_sync_inode(dir);
-	else
-		mark_inode_dirty(dir);
+	mark_inode_dirty(dir);
 
 	clear_nlink(inode);
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(6, 7, 0)
@@ -873,6 +871,7 @@ static int exfat_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 	struct exfat_chain cdir;
 	loff_t i_pos;
 	int err;
+	loff_t size = i_size_read(dir);
 
 	mutex_lock(&EXFAT_SB(sb)->s_lock);
 	exfat_set_volume_dirty(sb);
@@ -892,7 +891,7 @@ static int exfat_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 #else
 	dir->i_ctime = dir->i_mtime = current_time(dir);
 #endif
-	if (IS_DIRSYNC(dir))
+	if (IS_DIRSYNC(dir) && size != i_size_read(dir))
 		exfat_sync_inode(dir);
 	else
 		mark_inode_dirty(dir);
@@ -1331,6 +1330,7 @@ static int exfat_rename(struct user_namespace *mnt_userns,
 	struct super_block *sb = old_dir->i_sb;
 	loff_t i_pos;
 	int err;
+	loff_t size = i_size_read(new_dir);
 
 	/*
 	 * The VFS already checks for existence, so for local filesystems
@@ -1361,7 +1361,7 @@ static int exfat_rename(struct user_namespace *mnt_userns,
 #else
 	exfat_truncate_atime(&new_dir->i_atime);
 #endif
-	if (IS_DIRSYNC(new_dir))
+	if (IS_DIRSYNC(new_dir) && size != i_size_read(new_dir))
 		exfat_sync_inode(new_dir);
 	else
 		mark_inode_dirty(new_dir);
@@ -1385,10 +1385,7 @@ static int exfat_rename(struct user_namespace *mnt_userns,
 #if LINUX_VERSION_CODE <= KERNEL_VERSION(6, 6, 0)
 	old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
 #endif
-	if (IS_DIRSYNC(old_dir))
-		exfat_sync_inode(old_dir);
-	else
-		mark_inode_dirty(old_dir);
+	mark_inode_dirty(old_dir);
 
 	if (new_inode) {
 		exfat_unhash_inode(new_inode);