1 files changed, 85 insertions, 2 deletions

diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index a3e57f554f1b..901c1c938276 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -571,6 +571,89 @@ static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 	spin_unlock(&sbi->s_md_lock);
 }
 
+/*
+ * This writepage callback for write_cache_pages()
+ * takes care of a few cases after page cleaning.
+ *
+ * write_cache_pages() already checks for dirty pages
+ * and calls clear_page_dirty_for_io(), which we want,
+ * to write protect the pages.
+ *
+ * However, we may have to redirty a page (see below.)
+ */
+static int ext4_journalled_writepage_callback(struct page *page,
+					      struct writeback_control *wbc,
+					      void *data)
+{
+	transaction_t *transaction = (transaction_t *) data;
+	struct buffer_head *bh, *head;
+	struct journal_head *jh;
+
+	bh = head = page_buffers(page);
+	do {
+		/*
+		 * We have to redirty a page in these cases:
+		 * 1) If buffer is dirty, it means the page was dirty because it
+		 * contains a buffer that needs checkpointing. So the dirty bit
+		 * needs to be preserved so that checkpointing writes the buffer
+		 * properly.
+		 * 2) If buffer is not part of the committing transaction
+		 * (we may have just accidentally come across this buffer because
+		 * inode range tracking is not exact) or if the currently running
+		 * transaction already contains this buffer as well, dirty bit
+		 * needs to be preserved so that the buffer gets writeprotected
+		 * properly on running transaction's commit.
+		 */
+		jh = bh2jh(bh);
+		if (buffer_dirty(bh) ||
+		    (jh && (jh->b_transaction != transaction ||
+			    jh->b_next_transaction))) {
+			redirty_page_for_writepage(wbc, page);
+			goto out;
+		}
+	} while ((bh = bh->b_this_page) != head);
+
+out:
+	return AOP_WRITEPAGE_ACTIVATE;
+}
+
+static int ext4_journalled_submit_inode_data_buffers(struct jbd2_inode *jinode)
+{
+	struct address_space *mapping = jinode->i_vfs_inode->i_mapping;
+	struct writeback_control wbc = {
+		.sync_mode =  WB_SYNC_ALL,
+		.nr_to_write = LONG_MAX,
+		.range_start = jinode->i_dirty_start,
+		.range_end = jinode->i_dirty_end,
+        };
+
+	return write_cache_pages(mapping, &wbc,
+				 ext4_journalled_writepage_callback,
+				 jinode->i_transaction);
+}
+
+static int ext4_journal_submit_inode_data_buffers(struct jbd2_inode *jinode)
+{
+	int ret;
+
+	if (ext4_should_journal_data(jinode->i_vfs_inode))
+		ret = ext4_journalled_submit_inode_data_buffers(jinode);
+	else
+		ret = jbd2_journal_submit_inode_data_buffers(jinode);
+
+	return ret;
+}
+
+static int ext4_journal_finish_inode_data_buffers(struct jbd2_inode *jinode)
+{
+	int ret = 0;
+
+	if (!ext4_should_journal_data(jinode->i_vfs_inode))
+		ret = jbd2_journal_finish_inode_data_buffers(jinode);
+
+	return ret;
+}
+
 static bool system_going_down(void)
 {
 	return system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF
@@ -4753,9 +4836,9 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 
 	sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
 	sbi->s_journal->j_submit_inode_data_buffers =
-		jbd2_journal_submit_inode_data_buffers;
+		ext4_journal_submit_inode_data_buffers;
 	sbi->s_journal->j_finish_inode_data_buffers =
-		jbd2_journal_finish_inode_data_buffers;
+		ext4_journal_finish_inode_data_buffers;
 
 no_journal:
 	if (!test_opt(sb, NO_MBCACHE)) {