Btrfs: Fix the defragmention code and the block relocation code for data=ordered

Before setting an extent to delalloc, the code needs to wait for
pending ordered extents.

Also, the relocation code needs to wait for ordered IO before scanning
the block group again.  This is because the extents are not removed
until the IO for the new extents is finished

Signed-off-by: Chris Mason <chris.mason@oracle.com>

6 files changed, 113 insertions, 20 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 8ecac2e..6675e91 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -539,6 +539,13 @@ struct btrfs_fs_info {
 	atomic_t nr_async_submits;
 
 	/*
+	 * this is used by the balancing code to wait for all the pending
+	 * ordered extents
+	 */
+	spinlock_t ordered_extent_lock;
+	struct list_head ordered_extents;
+
+	/*
 	 * there is a pool of worker threads for checksumming during writes
 	 * and a pool for checksumming after reads.  This is because readers
 	 * can run with FS locks held, and the writers may be waiting for
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 7ce3f83..ec01062 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1252,6 +1252,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	fs_info->btree_inode->i_nlink = 1;
 	fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
 
+	INIT_LIST_HEAD(&fs_info->ordered_extents);
+	spin_lock_init(&fs_info->ordered_extent_lock);
+
 	sb->s_blocksize = 4096;
 	sb->s_blocksize_bits = blksize_bits(4096);
 
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index febc629..f92b297 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2640,6 +2640,7 @@ static int noinline relocate_inode_pages(struct inode *inode, u64 start,
 	struct file_ra_state *ra;
 	unsigned long total_read = 0;
 	unsigned long ra_pages;
+	struct btrfs_ordered_extent *ordered;
 	struct btrfs_trans_handle *trans;
 
 	ra = kzalloc(sizeof(*ra), GFP_NOFS);
@@ -2658,9 +2659,9 @@ static int noinline relocate_inode_pages(struct inode *inode, u64 start,
 				       calc_ra(i, last_index, ra_pages));
 		}
 		total_read++;
-		if (((u64)i << PAGE_CACHE_SHIFT) > inode->i_size)
+again:
+		if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
 			goto truncate_racing;
-
 		page = grab_cache_page(inode->i_mapping, i);
 		if (!page) {
 			goto out_unlock;
@@ -2674,18 +2675,24 @@ static int noinline relocate_inode_pages(struct inode *inode, u64 start,
 				goto out_unlock;
 			}
 		}
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-		ClearPageDirty(page);
-#else
-		cancel_dirty_page(page, PAGE_CACHE_SIZE);
-#endif
 		wait_on_page_writeback(page);
-		set_page_extent_mapped(page);
+
 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 		page_end = page_start + PAGE_CACHE_SIZE - 1;
-
 		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 
+		ordered = btrfs_lookup_ordered_extent(inode, page_start);
+		if (ordered) {
+			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			goto again;
+		}
+		set_page_extent_mapped(page);
+
+
 		set_extent_delalloc(io_tree, page_start,
 				    page_end, GFP_NOFS);
 		set_page_dirty(page);
@@ -2694,10 +2701,18 @@ static int noinline relocate_inode_pages(struct inode *inode, u64 start,
 		unlock_page(page);
 		page_cache_release(page);
 	}
-	balance_dirty_pages_ratelimited_nr(inode->i_mapping,
-					   total_read);
 
 out_unlock:
+	/* we have to start the IO in order to get the ordered extents
+	 * instantiated.  This allows the relocation to code to wait
+	 * for all the ordered extents to hit the disk.
+	 *
+	 * Otherwise, it would constantly loop over the same extents
+	 * because the old ones don't get deleted  until the IO is
+	 * started
+	 */
+	btrfs_fdatawrite_range(inode->i_mapping, start, start + len - 1,
+			       WB_SYNC_NONE);
 	kfree(ra);
 	trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
 	if (trans) {
@@ -3238,6 +3253,8 @@ next:
 
 		btrfs_clean_old_snapshots(tree_root);
 
+		btrfs_wait_ordered_extents(tree_root);
+
 		trans = btrfs_start_transaction(tree_root, 1);
 		btrfs_commit_transaction(trans, tree_root);
 		mutex_lock(&root->fs_info->alloc_mutex);
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 83f17a5..a61f2e7 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -213,6 +213,7 @@ int btrfs_defrag_file(struct file *file)
 	struct inode *inode = fdentry(file)->d_inode;
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_ordered_extent *ordered;
 	struct page *page;
 	unsigned long last_index;
 	unsigned long ra_pages = root->fs_info->bdi.ra_pages;
@@ -234,6 +235,7 @@ int btrfs_defrag_file(struct file *file)
 				       min(last_index, i + ra_pages - 1));
 		}
 		total_read++;
+again:
 		page = grab_cache_page(inode->i_mapping, i);
 		if (!page)
 			goto out_unlock;
@@ -247,18 +249,23 @@ int btrfs_defrag_file(struct file *file)
 			}
 		}
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
-		ClearPageDirty(page);
-#else
-		cancel_dirty_page(page, PAGE_CACHE_SIZE);
-#endif
 		wait_on_page_writeback(page);
-		set_page_extent_mapped(page);
 
 		page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 		page_end = page_start + PAGE_CACHE_SIZE - 1;
-
 		lock_extent(io_tree, page_start, page_end, GFP_NOFS);
+
+		ordered = btrfs_lookup_ordered_extent(inode, page_start);
+		if (ordered) {
+			unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+			unlock_page(page);
+			page_cache_release(page);
+			btrfs_start_ordered_extent(inode, ordered, 1);
+			btrfs_put_ordered_extent(ordered);
+			goto again;
+		}
+		set_page_extent_mapped(page);
+
 		set_extent_delalloc(io_tree, page_start,
 				    page_end, GFP_NOFS);
 
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index e42fd23..676e4bd 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -167,20 +167,28 @@ int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset,
 	entry->file_offset = file_offset;
 	entry->start = start;
 	entry->len = len;
+	entry->inode = inode;
+
 	/* one ref for the tree */
 	atomic_set(&entry->refs, 1);
 	init_waitqueue_head(&entry->wait);
 	INIT_LIST_HEAD(&entry->list);
+	INIT_LIST_HEAD(&entry->root_extent_list);
 
 	node = tree_insert(&tree->tree, file_offset,
 			   &entry->rb_node);
 	if (node) {
-		entry = rb_entry(node, struct btrfs_ordered_extent, rb_node);
-		atomic_inc(&entry->refs);
+		printk("warning dup entry from add_ordered_extent\n");
+		BUG();
 	}
 	set_extent_ordered(&BTRFS_I(inode)->io_tree, file_offset,
 			   entry_end(entry) - 1, GFP_NOFS);
 
+	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+	list_add_tail(&entry->root_extent_list,
+		      &BTRFS_I(inode)->root->fs_info->ordered_extents);
+	spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+
 	mutex_unlock(&tree->mutex);
 	BUG_ON(node);
 	return 0;
@@ -285,11 +293,55 @@ int btrfs_remove_ordered_extent(struct inode *inode,
 	rb_erase(node, &tree->tree);
 	tree->last = NULL;
 	set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags);
+
+	spin_lock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+	list_del_init(&entry->root_extent_list);
+	spin_unlock(&BTRFS_I(inode)->root->fs_info->ordered_extent_lock);
+
 	mutex_unlock(&tree->mutex);
 	wake_up(&entry->wait);
 	return 0;
 }
 
+int btrfs_wait_ordered_extents(struct btrfs_root *root)
+{
+	struct list_head splice;
+	struct list_head *cur;
+	struct btrfs_ordered_extent *ordered;
+	struct inode *inode;
+
+	INIT_LIST_HEAD(&splice);
+
+	spin_lock(&root->fs_info->ordered_extent_lock);
+	list_splice_init(&root->fs_info->ordered_extents, &splice);
+	while(!list_empty(&splice)) {
+		cur = splice.next;
+		ordered = list_entry(cur, struct btrfs_ordered_extent,
+				     root_extent_list);
+		list_del_init(&ordered->root_extent_list);
+		atomic_inc(&ordered->refs);
+		inode = ordered->inode;
+
+		/*
+		 * the inode can't go away until all the pages are gone
+		 * and the pages won't go away while there is still
+		 * an ordered extent and the ordered extent won't go
+		 * away until it is off this list.  So, we can safely
+		 * increment i_count here and call iput later
+		 */
+		atomic_inc(&inode->i_count);
+		spin_unlock(&root->fs_info->ordered_extent_lock);
+
+		btrfs_start_ordered_extent(inode, ordered, 1);
+		btrfs_put_ordered_extent(ordered);
+		iput(inode);
+
+		spin_lock(&root->fs_info->ordered_extent_lock);
+	}
+	spin_unlock(&root->fs_info->ordered_extent_lock);
+	return 0;
+}
+
 /*
  * Used to start IO or wait for a given ordered extent to finish.
  *
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index 199cb0b..5efe6b6 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -80,6 +80,9 @@ struct btrfs_ordered_extent {
 	/* reference count */
 	atomic_t refs;
 
+	/* the inode we belong to */
+	struct inode *inode;
+
 	/* list of checksums for insertion when the extent io is done */
 	struct list_head list;
 
@@ -88,6 +91,9 @@ struct btrfs_ordered_extent {
 
 	/* our friendly rbtree entry */
 	struct rb_node rb_node;
+
+	/* a per root list of all the pending ordered extents */
+	struct list_head root_extent_list;
 };
 
 
@@ -137,4 +143,5 @@ int btrfs_wait_on_page_writeback_range(struct address_space *mapping,
 				       pgoff_t start, pgoff_t end);
 int btrfs_fdatawrite_range(struct address_space *mapping, loff_t start,
 			   loff_t end, int sync_mode);
+int btrfs_wait_ordered_extents(struct btrfs_root *root);
 #endif