f2fs: handle EIO not to break fs consistency

There are two rules when EIO is occurred.
1. don't write any checkpoint data to preserve the previous checkpoint
2. don't lose the cached dentry/node/meta pages

So, at first, this patch adds set_page_dirty in f2fs_write_end_io's failure.
Then, writing checkpoint/dentry/node blocks is not allowed.

Note that, for the data pages, we can't just throw away by redirtying them.
Otherwise, kworker can fall into infinite loop to flush them.
(Ref. xfstests/019)
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

fs/f2fs/checkpoint.c

@@ -160,14 +160,11 @@ static int f2fs_write_meta_page(struct page *page,
 		goto redirty_out;
 	if (wbc->for_reclaim)
 		goto redirty_out;
-
-	/* Should not write any meta pages, if any IO error was occurred */
 	if (unlikely(f2fs_cp_error(sbi)))
-		goto no_write;
+		goto redirty_out;
 
 	f2fs_wait_on_page_writeback(page, META);
 	write_meta_page(sbi, page);
-no_write:
 	dec_page_count(sbi, F2FS_DIRTY_META);
 	unlock_page(page);
 	return 0;
@@ -737,7 +734,7 @@ void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
 /*
  * Freeze all the FS-operations for checkpoint.
  */
-static void block_operations(struct f2fs_sb_info *sbi)
+static int block_operations(struct f2fs_sb_info *sbi)
 {
 	struct writeback_control wbc = {
 		.sync_mode = WB_SYNC_ALL,
@@ -745,6 +742,7 @@ static void block_operations(struct f2fs_sb_info *sbi)
 		.for_reclaim = 0,
 	};
 	struct blk_plug plug;
+	int err = 0;
 
 	blk_start_plug(&plug);
 
@@ -754,6 +752,10 @@ static void block_operations(struct f2fs_sb_info *sbi)
 	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
 		f2fs_unlock_all(sbi);
 		sync_dirty_dir_inodes(sbi);
+		if (unlikely(f2fs_cp_error(sbi))) {
+			err = -EIO;
+			goto out;
+		}
 		goto retry_flush_dents;
 	}
 
@@ -767,9 +769,16 @@ static void block_operations(struct f2fs_sb_info *sbi)
 	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
 		up_write(&sbi->node_write);
 		sync_node_pages(sbi, 0, &wbc);
+		if (unlikely(f2fs_cp_error(sbi))) {
+			f2fs_unlock_all(sbi);
+			err = -EIO;
+			goto out;
+		}
 		goto retry_flush_nodes;
 	}
+out:
 	blk_finish_plug(&plug);
+	return err;
 }
 
 static void unblock_operations(struct f2fs_sb_info *sbi)
@@ -813,8 +822,11 @@ static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 	discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
 
 	/* Flush all the NAT/SIT pages */
-	while (get_pages(sbi, F2FS_DIRTY_META))
+	while (get_pages(sbi, F2FS_DIRTY_META)) {
 		sync_meta_pages(sbi, META, LONG_MAX);
+		if (unlikely(f2fs_cp_error(sbi)))
+			return;
+	}
 
 	next_free_nid(sbi, &last_nid);
 
@@ -924,6 +936,9 @@ static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 	/* wait for previous submitted node/meta pages writeback */
 	wait_on_all_pages_writeback(sbi);
 
+	if (unlikely(f2fs_cp_error(sbi)))
+		return;
+
 	filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
 	filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
 
@@ -934,11 +949,13 @@ static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 	/* Here, we only have one bio having CP pack */
 	sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
 
-	if (!f2fs_cp_error(sbi)) {
-		clear_prefree_segments(sbi);
-		release_dirty_inode(sbi);
-		F2FS_RESET_SB_DIRT(sbi);
-	}
+	release_dirty_inode(sbi);
+
+	if (unlikely(f2fs_cp_error(sbi)))
+		return;
+
+	clear_prefree_segments(sbi);
+	F2FS_RESET_SB_DIRT(sbi);
 }
 
 /*
@@ -955,8 +972,10 @@ void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
 
 	if (!sbi->s_dirty)
 		goto out;
-
-	block_operations(sbi);
+	if (unlikely(f2fs_cp_error(sbi)))
+		goto out;
+	if (block_operations(sbi))
+		goto out;
 
 	trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
 

fs/f2fs/data.c

@@ -53,7 +53,7 @@ static void f2fs_write_end_io(struct bio *bio, int err)
 		struct page *page = bvec->bv_page;
 
 		if (unlikely(err)) {
-			SetPageError(page);
+			set_page_dirty(page);
 			set_bit(AS_EIO, &page->mapping->flags);
 			f2fs_stop_checkpoint(sbi);
 		}
@@ -836,10 +836,19 @@ static int f2fs_write_data_page(struct page *page,
 
 	/* Dentry blocks are controlled by checkpoint */
 	if (S_ISDIR(inode->i_mode)) {
+		if (unlikely(f2fs_cp_error(sbi)))
+			goto redirty_out;
 		err = do_write_data_page(page, &fio);
 		goto done;
 	}
 
+	/* we should bypass data pages to proceed the kworkder jobs */
+	if (unlikely(f2fs_cp_error(sbi))) {
+		SetPageError(page);
+		unlock_page(page);
+		return 0;
+	}
+
 	if (!wbc->for_reclaim)
 		need_balance_fs = true;
 	else if (has_not_enough_free_secs(sbi, 0))

fs/f2fs/node.c

@@ -1215,6 +1215,8 @@ static int f2fs_write_node_page(struct page *page,
 
 	if (unlikely(sbi->por_doing))
 		goto redirty_out;
+	if (unlikely(f2fs_cp_error(sbi)))
+		goto redirty_out;
 
 	f2fs_wait_on_page_writeback(page, NODE);
 

fs/f2fs/super.c

@@ -435,7 +435,10 @@ static void f2fs_put_super(struct super_block *sb)
 	if (sbi->s_dirty)
 		write_checkpoint(sbi, true);
 
-	/* normally superblock is clean, so we need to release this */
+	/*
+	 * normally superblock is clean, so we need to release this.
+	 * In addition, EIO will skip do checkpoint, we need this as well.
+	 */
 	release_dirty_inode(sbi);
 
 	iput(sbi->node_inode);