[PATCH 5/7] f2fs: set fsync mark only for the last dnode

[Jaegeuk Kim]

In order to give atomic writes, we should consider power failure during
sync_node_pages in fsync.
So, this patch marks fsync flag only in the last dnode block.

Signed-off-by: Jaegeuk Kim <jaegeuk@xxxxxxxxxx>
---
 fs/f2fs/f2fs.h     |   4 +-
 fs/f2fs/file.c     |  14 +++++--
 fs/f2fs/node.c     | 106 +++++++++++++++++++++++++++++++++++++++++++++++++----
 fs/f2fs/recovery.c |   9 ++---
 4 files changed, 113 insertions(+), 20 deletions(-)

diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 269abe5..ca828b0 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -159,7 +159,6 @@ struct fsync_inode_entry {
 	struct inode *inode;	/* vfs inode pointer */
 	block_t blkaddr;	/* block address locating the last fsync */
 	block_t last_dentry;	/* block address locating the last dentry */
-	block_t last_inode;	/* block address locating the last inode */
 };
 
 #define nats_in_cursum(jnl)		(le16_to_cpu(jnl->n_nats))
@@ -1784,7 +1783,8 @@ void ra_node_page(struct f2fs_sb_info *, nid_t);
 struct page *get_node_page(struct f2fs_sb_info *, pgoff_t);
 struct page *get_node_page_ra(struct page *, int);
 void sync_inode_page(struct dnode_of_data *);
-int fsync_node_pages(struct f2fs_sb_info *, nid_t, struct writeback_control *);
+int fsync_node_pages(struct f2fs_sb_info *, nid_t, struct writeback_control *,
+								bool);
 int sync_node_pages(struct f2fs_sb_info *, struct writeback_control *);
 bool alloc_nid(struct f2fs_sb_info *, nid_t *);
 void alloc_nid_done(struct f2fs_sb_info *, nid_t);
diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
index 60fd64c..dc47d5c 100644
--- a/fs/f2fs/file.c
+++ b/fs/f2fs/file.c
@@ -182,7 +182,8 @@ static void try_to_fix_pino(struct inode *inode)
 	}
 }
 
-int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
+static int f2fs_do_sync_file(struct file *file, loff_t start, loff_t end,
+						int datasync, bool atomic)
 {
 	struct inode *inode = file->f_mapping->host;
 	struct f2fs_inode_info *fi = F2FS_I(inode);
@@ -256,7 +257,7 @@ go_write:
 		goto out;
 	}
 sync_nodes:
-	ret = fsync_node_pages(sbi, ino, &wbc);
+	ret = fsync_node_pages(sbi, ino, &wbc, atomic);
 	if (ret)
 		goto out;
 
@@ -290,6 +291,11 @@ out:
 	return ret;
 }
 
+int f2fs_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
+{
+	return f2fs_do_sync_file(file, start, end, datasync, false);
+}
+
 static pgoff_t __get_first_dirty_index(struct address_space *mapping,
 						pgoff_t pgofs, int whence)
 {
@@ -1407,7 +1413,7 @@ static int f2fs_ioc_commit_atomic_write(struct file *filp)
 		}
 	}
 
-	ret = f2fs_sync_file(filp, 0, LLONG_MAX, 0);
+	ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
 err_out:
 	mnt_drop_write_file(filp);
 	return ret;
@@ -1465,7 +1471,7 @@ static int f2fs_ioc_abort_volatile_write(struct file *filp)
 		drop_inmem_pages(inode);
 	if (f2fs_is_volatile_file(inode)) {
 		clear_inode_flag(F2FS_I(inode), FI_VOLATILE_FILE);
-		ret = f2fs_sync_file(filp, 0, LLONG_MAX, 0);
+		ret = f2fs_do_sync_file(filp, 0, LLONG_MAX, 0, true);
 	}
 
 	mnt_drop_write_file(filp);
diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index 8a1e211..de070a5 100644
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -1222,13 +1222,81 @@ iput_out:
 	iput(inode);
 }
 
+static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
+{
+	pgoff_t index, end;
+	struct pagevec pvec;
+	struct page *last_page = NULL;
+
+	pagevec_init(&pvec, 0);
+	index = 0;
+	end = ULONG_MAX;
+
+	while (index <= end) {
+		int i, nr_pages;
+		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
+				PAGECACHE_TAG_DIRTY,
+				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
+		if (nr_pages == 0)
+			break;
+
+		for (i = 0; i < nr_pages; i++) {
+			struct page *page = pvec.pages[i];
+
+			if (unlikely(f2fs_cp_error(sbi))) {
+				f2fs_put_page(last_page, 0);
+				pagevec_release(&pvec);
+				return ERR_PTR(-EIO);
+			}
+
+			if (!IS_DNODE(page) || !is_cold_node(page))
+				continue;
+			if (ino_of_node(page) != ino)
+				continue;
+
+			lock_page(page);
+
+			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
+continue_unlock:
+				unlock_page(page);
+				continue;
+			}
+			if (ino_of_node(page) != ino)
+				goto continue_unlock;
+
+			if (!PageDirty(page)) {
+				/* someone wrote it for us */
+				goto continue_unlock;
+			}
+
+			if (last_page)
+				f2fs_put_page(last_page, 0);
+
+			get_page(page);
+			last_page = page;
+			unlock_page(page);
+		}
+		pagevec_release(&pvec);
+		cond_resched();
+	}
+	return last_page;
+}
+
 int fsync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
-					struct writeback_control *wbc)
+			struct writeback_control *wbc, bool atomic)
 {
 	pgoff_t index, end;
 	struct pagevec pvec;
 	int ret = 0;
+	struct page *last_page = NULL;
+	bool marked = false;
 
+	if (atomic) {
+		last_page = last_fsync_dnode(sbi, ino);
+		if (IS_ERR_OR_NULL(last_page))
+			return PTR_ERR_OR_ZERO(last_page);
+	}
+retry:
 	pagevec_init(&pvec, 0);
 	index = 0;
 	end = ULONG_MAX;
@@ -1245,6 +1313,7 @@ int fsync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
 			struct page *page = pvec.pages[i];
 
 			if (unlikely(f2fs_cp_error(sbi))) {
+				f2fs_put_page(last_page, 0);
 				pagevec_release(&pvec);
 				return -EIO;
 			}
@@ -1264,33 +1333,54 @@ continue_unlock:
 			if (ino_of_node(page) != ino)
 				goto continue_unlock;
 
-			if (!PageDirty(page)) {
+			if (!PageDirty(page) && page != last_page) {
 				/* someone wrote it for us */
 				goto continue_unlock;
 			}
 
 			f2fs_wait_on_page_writeback(page, NODE, true);
 			BUG_ON(PageWriteback(page));
-			if (!clear_page_dirty_for_io(page))
-				goto continue_unlock;
 
-			set_fsync_mark(page, 1);
-			if (IS_INODE(page))
-				set_dentry_mark(page,
+			if (!atomic || page == last_page) {
+				set_fsync_mark(page, 1);
+				if (IS_INODE(page))
+					set_dentry_mark(page,
 						need_dentry_mark(sbi, ino));
+				/*  may be written by other thread */
+				if (!PageDirty(page))
+					set_page_dirty(page);
+			}
+
+			if (!clear_page_dirty_for_io(page))
+				goto continue_unlock;
 
 			ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
 			if (ret) {
 				unlock_page(page);
+				f2fs_put_page(last_page, 0);
+				break;
+			}
+			if (page == last_page) {
+				f2fs_put_page(page, 0);
+				marked = true;
 				break;
 			}
 		}
 		pagevec_release(&pvec);
 		cond_resched();
 
-		if (ret)
+		if (ret || marked)
 			break;
 	}
+	if (!ret && atomic && !marked) {
+		f2fs_msg(sbi->sb, KERN_DEBUG,
+			"Retry to write fsync mark: ino=%u, idx=%lx",
+					ino, last_page->index);
+		lock_page(last_page);
+		set_page_dirty(last_page);
+		unlock_page(last_page);
+		goto retry;
+	}
 	return ret ? -EIO: 0;
 }
 
diff --git a/fs/f2fs/recovery.c b/fs/f2fs/recovery.c
index 2c87c12..a646d3b 100644
--- a/fs/f2fs/recovery.c
+++ b/fs/f2fs/recovery.c
@@ -257,11 +257,8 @@ static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 		}
 		entry->blkaddr = blkaddr;
 
-		if (IS_INODE(page)) {
-			entry->last_inode = blkaddr;
-			if (is_dent_dnode(page))
-				entry->last_dentry = blkaddr;
-		}
+		if (IS_INODE(page) && is_dent_dnode(page))
+			entry->last_dentry = blkaddr;
 next:
 		/* check next segment */
 		blkaddr = next_blkaddr_of_node(page);
@@ -521,7 +518,7 @@ static int recover_data(struct f2fs_sb_info *sbi, struct list_head *head)
 		 * In this case, we can lose the latest inode(x).
 		 * So, call recover_inode for the inode update.
 		 */
-		if (entry->last_inode == blkaddr)
+		if (IS_INODE(page))
 			recover_inode(entry->inode, page);
 		if (entry->last_dentry == blkaddr) {
 			err = recover_dentry(entry->inode, page);
-- 
2.6.3