f2fs: flush inode metadata when checkpoint is doing

This patch registers all the inodes which have dirty metadata to sync when
checkpoint is doing.

Signed-off-by: Jaegeuk Kim <[e-mail]>

fs/f2fs/checkpoint.c

···
 	goto retry;
 }
 
+int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
+{
+	struct list_head *head = &sbi->inode_list[DIRTY_META];
+	struct inode *inode;
+	struct f2fs_inode_info *fi;
+	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
+
+	while (total--) {
+		if (unlikely(f2fs_cp_error(sbi)))
+			return -EIO;
+
+		spin_lock(&sbi->inode_lock[DIRTY_META]);
+		if (list_empty(head)) {
+			spin_unlock(&sbi->inode_lock[DIRTY_META]);
+			return 0;
+		}
+		fi = list_entry(head->next, struct f2fs_inode_info,
+							gdirty_list);
+		inode = igrab(&fi->vfs_inode);
+		spin_unlock(&sbi->inode_lock[DIRTY_META]);
+		if (inode) {
+			update_inode_page(inode);
+			iput(inode);
+		}
+	};
+	return 0;
+}
+
 /*
  * Freeze all the FS-operations for checkpoint.
  */
···
 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 		f2fs_unlock_all(sbi);
 		err = sync_dirty_inodes(sbi, DIR_INODE);
+		if (err)
+			goto out;
+		goto retry_flush_dents;
+	}
+
+	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
+		f2fs_unlock_all(sbi);
+		err = f2fs_sync_inode_meta(sbi);
 		if (err)
 			goto out;
 		goto retry_flush_dents;

fs/f2fs/debug.c

···
 	si->ndirty_data = get_pages(sbi, F2FS_DIRTY_DATA);
 	si->ndirty_dirs = sbi->ndirty_inode[DIR_INODE];
 	si->ndirty_files = sbi->ndirty_inode[FILE_INODE];
+	si->ndirty_all = sbi->ndirty_inode[DIRTY_META];
 	si->inmem_pages = get_pages(sbi, F2FS_INMEM_PAGES);
 	si->wb_bios = atomic_read(&sbi->nr_wb_bios);
 	si->total_count = (int)sbi->user_block_count / sbi->blocks_per_seg;
···
 			   si->inmem_pages, si->wb_bios);
 		seq_printf(s, "  - nodes: %4lld in %4d\n",
 			   si->ndirty_node, si->node_pages);
-		seq_printf(s, "  - dents: %4lld in dirs:%4d\n",
-			   si->ndirty_dent, si->ndirty_dirs);
+		seq_printf(s, "  - dents: %4lld in dirs:%4d (%4d)\n",
+			   si->ndirty_dent, si->ndirty_dirs, si->ndirty_all);
 		seq_printf(s, "  - datas: %4lld in files:%4d\n",
 			   si->ndirty_data, si->ndirty_files);
 		seq_printf(s, "  - meta: %4lld in %4d\n",

fs/f2fs/f2fs.h

···
 	nid_t i_xattr_nid;		/* node id that contains xattrs */
 	unsigned long long xattr_ver;	/* cp version of xattr modification */
 
-	struct list_head dirty_list;	/* linked in global dirty list */
+	struct list_head dirty_list;	/* dirty list for dirs and files */
+	struct list_head gdirty_list;	/* linked in global dirty list */
 	struct list_head inmem_pages;	/* inmemory pages managed by f2fs */
 	struct mutex inmem_lock;	/* lock for inmemory pages */
 	struct extent_tree *extent_tree;	/* cached extent_tree entry */
···
 	F2FS_DIRTY_NODES,
 	F2FS_DIRTY_META,
 	F2FS_INMEM_PAGES,
+	F2FS_DIRTY_IMETA,
 	NR_COUNT_TYPE,
 };
 
···
 enum inode_type {
 	DIR_INODE,			/* for dirty dir inode */
 	FILE_INODE,			/* for dirty regular/symlink inode */
+	DIRTY_META,			/* for all dirtied inode metadata */
 	NR_INODE_TYPE,
 };
 
···
 /*
  * super.c
  */
+void f2fs_inode_synced(struct inode *);
 int f2fs_commit_super(struct f2fs_sb_info *, bool);
 int f2fs_sync_fs(struct super_block *, int);
 extern __printf(3, 4)
···
 void remove_ino_entry(struct f2fs_sb_info *, nid_t, int type);
 void release_ino_entry(struct f2fs_sb_info *, bool);
 bool exist_written_data(struct f2fs_sb_info *, nid_t, int);
+int f2fs_sync_inode_meta(struct f2fs_sb_info *);
 int acquire_orphan_inode(struct f2fs_sb_info *);
 void release_orphan_inode(struct f2fs_sb_info *);
 void add_orphan_inode(struct f2fs_sb_info *, nid_t);
···
 	unsigned long long hit_total, total_ext;
 	int ext_tree, zombie_tree, ext_node;
 	s64 ndirty_node, ndirty_dent, ndirty_meta, ndirty_data, inmem_pages;
-	unsigned int ndirty_dirs, ndirty_files;
+	unsigned int ndirty_dirs, ndirty_files, ndirty_all;
 	int nats, dirty_nats, sits, dirty_sits, fnids;
 	int total_count, utilization;
 	int bg_gc, wb_bios;

fs/f2fs/inode.c

···
 
 	__set_inode_rdev(inode, ri);
 	set_cold_node(inode, node_page);
-	clear_inode_flag(inode, FI_DIRTY_INODE);
+	f2fs_inode_synced(inode);
 
 	/* deleted inode */
 	if (inode->i_nlink == 0)
···
 		} else if (err != -ENOENT) {
 			f2fs_stop_checkpoint(sbi, false);
 		}
+		f2fs_inode_synced(inode);
 		return 0;
 	}
 	ret = update_inode(inode, node_page);
···
 		goto retry;
 	}
 
+	if (err)
+		update_inode_page(inode);
 	sb_end_intwrite(inode->i_sb);
 no_delete:
 	stat_dec_inline_xattr(inode);
···
 		!exist_written_data(sbi, inode->i_ino, ORPHAN_INO));
 out_clear:
 	fscrypt_put_encryption_info(inode, NULL);
+
+	f2fs_bug_on(sbi, is_inode_flag_set(inode, FI_DIRTY_INODE));
 	clear_inode(inode);
 }
 

fs/f2fs/node.c

···
 	if (dn->nid == dn->inode->i_ino) {
 		remove_orphan_inode(sbi, dn->nid);
 		dec_valid_inode_count(sbi);
+		f2fs_inode_synced(dn->inode);
 	} else {
 		sync_inode_page(dn);
 	}

fs/f2fs/segment.h

···
 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
 
+	node_secs += get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
+
 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 		return false;
 

fs/f2fs/super.c

···
 	fi->i_advise = 0;
 	init_rwsem(&fi->i_sem);
 	INIT_LIST_HEAD(&fi->dirty_list);
+	INIT_LIST_HEAD(&fi->gdirty_list);
 	INIT_LIST_HEAD(&fi->inmem_pages);
 	mutex_init(&fi->inmem_lock);
 
···
 
 static int f2fs_drop_inode(struct inode *inode)
 {
+	int ret;
+
 	/*
 	 * This is to avoid a deadlock condition like below.
 	 * writeback_single_inode(inode)
···
 	 *    - f2fs_gc -> iput -> evict
 	 *       - inode_wait_for_writeback(inode)
 	 */
-	if (!inode_unhashed(inode) && inode->i_state & I_SYNC) {
+	if ((!inode_unhashed(inode) && inode->i_state & I_SYNC)) {
 		if (!inode->i_nlink && !is_bad_inode(inode)) {
 			/* to avoid evict_inode call simultaneously */
 			atomic_inc(&inode->i_count);
···
 		}
 		return 0;
 	}
-	return generic_drop_inode(inode);
+
+	ret = generic_drop_inode(inode);
+	if (is_inode_flag_set(inode, FI_DIRTY_INODE)) {
+		if (ret)
+			inode->i_state |= I_WILL_FREE;
+		spin_unlock(&inode->i_lock);
+
+		update_inode_page(inode);
+
+		spin_lock(&inode->i_lock);
+		if (ret)
+			inode->i_state &= ~I_WILL_FREE;
+	}
+	return ret;
 }
 
 /*
···
  */
 static void f2fs_dirty_inode(struct inode *inode, int flags)
 {
+	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+
+	if (inode->i_ino == F2FS_NODE_INO(sbi) ||
+			inode->i_ino == F2FS_META_INO(sbi))
+		return;
+
+	spin_lock(&sbi->inode_lock[DIRTY_META]);
+	if (is_inode_flag_set(inode, FI_DIRTY_INODE)) {
+		spin_unlock(&sbi->inode_lock[DIRTY_META]);
+		return;
+	}
+
 	set_inode_flag(inode, FI_DIRTY_INODE);
+	list_add_tail(&F2FS_I(inode)->gdirty_list,
+				&sbi->inode_list[DIRTY_META]);
+	inc_page_count(sbi, F2FS_DIRTY_IMETA);
+	spin_unlock(&sbi->inode_lock[DIRTY_META]);
+	stat_inc_dirty_inode(sbi, DIRTY_META);
+}
+
+void f2fs_inode_synced(struct inode *inode)
+{
+	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
+
+	spin_lock(&sbi->inode_lock[DIRTY_META]);
+	if (!is_inode_flag_set(inode, FI_DIRTY_INODE)) {
+		spin_unlock(&sbi->inode_lock[DIRTY_META]);
+		return;
+	}
+	list_del_init(&F2FS_I(inode)->gdirty_list);
+	clear_inode_flag(inode, FI_DIRTY_INODE);
+	dec_page_count(sbi, F2FS_DIRTY_IMETA);
+	spin_unlock(&sbi->inode_lock[DIRTY_META]);
+	stat_dec_dirty_inode(F2FS_I_SB(inode), DIRTY_META);
 }
 
 static void f2fs_i_callback(struct rcu_head *head)
···
 	return 0;
 
 free_kobj:
+	f2fs_sync_inode_meta(sbi);
 	kobject_del(&sbi->s_kobj);
 	kobject_put(&sbi->s_kobj);
 	wait_for_completion(&sbi->s_kobj_unregister);