Btrfs: Maintain a list of inodes that are delalloc and a way to wait on them

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

fs/btrfs/btrfs_inode.h

@@ -42,6 +42,8 @@ struct btrfs_inode {
 	/* for keeping track of orphaned inodes */
 	struct list_head i_orphan;
 
+	struct list_head delalloc_inodes;
+
 	/*
 	 * transid of the trans_handle that last modified this inode
 	 */

fs/btrfs/ctree.h

@@ -551,6 +551,7 @@ struct btrfs_fs_info {
 	 */
 	spinlock_t ordered_extent_lock;
 	struct list_head ordered_extents;
+	struct list_head delalloc_inodes;
 
 	/*
 	 * there is a pool of worker threads for checksumming during writes
@@ -637,6 +638,7 @@ struct btrfs_root {
 	struct kobject root_kobj;
 	struct completion kobj_unregister;
 	struct mutex objectid_mutex;
+
 	u64 objectid;
 	u64 last_trans;
 
@@ -1651,6 +1653,8 @@ int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
 #define PageChecked PageFsMisc
 #endif
 
+int btrfs_start_delalloc_inodes(struct btrfs_root *root);
+int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end);
 int btrfs_writepages(struct address_space *mapping,
 		     struct writeback_control *wbc);
 int btrfs_create_subvol_root(struct btrfs_root *new_root,

fs/btrfs/disk-io.c

@@ -1234,6 +1234,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	INIT_LIST_HEAD(&fs_info->trans_list);
 	INIT_LIST_HEAD(&fs_info->dead_roots);
 	INIT_LIST_HEAD(&fs_info->hashers);
+	INIT_LIST_HEAD(&fs_info->delalloc_inodes);
 	spin_lock_init(&fs_info->hash_lock);
 	spin_lock_init(&fs_info->delalloc_lock);
 	spin_lock_init(&fs_info->new_trans_lock);

fs/btrfs/extent-tree.c

@@ -1230,7 +1230,6 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 		found->total_bytes += total_bytes;
 		found->bytes_used += bytes_used;
 		found->full = 0;
-		WARN_ON(found->total_bytes < found->bytes_used);
 		*space_info = found;
 		return 0;
 	}
@@ -2841,8 +2840,7 @@ static int noinline relocate_inode_pages(struct inode *inode, u64 start,
 		 */
 		clear_page_dirty_for_io(page);
 
-		set_extent_delalloc(io_tree, page_start,
-				    page_end, GFP_NOFS);
+		btrfs_set_extent_delalloc(inode, page_start, page_end);
 		set_page_dirty(page);
 
 		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
@@ -3319,6 +3317,13 @@ int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
 	key.type = 0;
 	cur_byte = key.objectid;
 
+	mutex_unlock(&root->fs_info->alloc_mutex);
+
+	btrfs_start_delalloc_inodes(root);
+	btrfs_wait_ordered_extents(tree_root);
+
+	mutex_lock(&root->fs_info->alloc_mutex);
+
 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 	if (ret < 0)
 		goto out;
@@ -3401,6 +3406,7 @@ int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 shrink_start)
 
 		btrfs_clean_old_snapshots(tree_root);
 
+		btrfs_start_delalloc_inodes(root);
 		btrfs_wait_ordered_extents(tree_root);
 
 		trans = btrfs_start_transaction(tree_root, 1);

fs/btrfs/file.c

@@ -312,8 +312,7 @@ static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 		 * to reset the delalloc bit on things that already have
 		 * extents reserved.
 		 */
-		set_extent_delalloc(io_tree, start_pos,
-				    end_of_last_block, GFP_NOFS);
+		btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 		for (i = 0; i < num_pages; i++) {
 			struct page *p = pages[i];
 			SetPageUptodate(p);

fs/btrfs/inode.c

@@ -303,6 +303,10 @@ int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 		spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 		BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 		root->fs_info->delalloc_bytes += end - start + 1;
+		if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
+			list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
+				      &root->fs_info->delalloc_inodes);
+		}
 		spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 	}
 	return 0;
@@ -325,6 +329,10 @@ int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 			root->fs_info->delalloc_bytes -= end - start + 1;
 			BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 		}
+		if (BTRFS_I(inode)->delalloc_bytes == 0 &&
+		    !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
+			list_del_init(&BTRFS_I(inode)->delalloc_inodes);
+		}
 		spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 	}
 	return 0;
@@ -408,6 +416,12 @@ static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
+{
+	return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
+				   GFP_NOFS);
+}
+
 struct btrfs_writepage_fixup {
 	struct page *page;
 	struct btrfs_work work;
@@ -453,8 +467,7 @@ void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 		goto again;
 	}
 
-	set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start, page_end,
-			    GFP_NOFS);
+	btrfs_set_extent_delalloc(inode, page_start, page_end);
 	ClearPageChecked(page);
 out:
 	unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
@@ -1530,8 +1543,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 		goto again;
 	}
 
-	set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
-			    page_end, GFP_NOFS);
+	btrfs_set_extent_delalloc(inode, page_start, page_end);
 	ret = 0;
 	if (offset != PAGE_CACHE_SIZE) {
 		kaddr = kmap(page);
@@ -1766,6 +1778,7 @@ static int btrfs_init_locked_inode(struct inode *inode, void *p)
 			     inode->i_mapping, GFP_NOFS);
 	extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 			     inode->i_mapping, GFP_NOFS);
+	INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
 	btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
 	mutex_init(&BTRFS_I(inode)->csum_mutex);
 	mutex_init(&BTRFS_I(inode)->extent_mutex);
@@ -2158,6 +2171,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 	extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 			     inode->i_mapping, GFP_NOFS);
 	btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
+	INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
 	mutex_init(&BTRFS_I(inode)->csum_mutex);
 	mutex_init(&BTRFS_I(inode)->extent_mutex);
 	BTRFS_I(inode)->delalloc_bytes = 0;
@@ -2400,6 +2414,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 				     inode->i_mapping, GFP_NOFS);
 		extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 				     inode->i_mapping, GFP_NOFS);
+		INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
 		mutex_init(&BTRFS_I(inode)->csum_mutex);
 		mutex_init(&BTRFS_I(inode)->extent_mutex);
 		BTRFS_I(inode)->delalloc_bytes = 0;
@@ -3049,8 +3064,7 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 		goto again;
 	}
 
-	set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
-			    page_end, GFP_NOFS);
+	btrfs_set_extent_delalloc(inode, page_start, page_end);
 	ret = 0;
 
 	/* page is wholly or partially inside EOF */
@@ -3373,6 +3387,26 @@ static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
 	return ret;
 }
 
+int btrfs_start_delalloc_inodes(struct btrfs_root *root)
+{
+	struct list_head *head = &root->fs_info->delalloc_inodes;
+	struct btrfs_inode *binode;
+	unsigned long flags;
+
+	spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
+	while(!list_empty(head)) {
+		binode = list_entry(head->next, struct btrfs_inode,
+				    delalloc_inodes);
+		atomic_inc(&binode->vfs_inode.i_count);
+		spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
+		filemap_write_and_wait(binode->vfs_inode.i_mapping);
+		iput(&binode->vfs_inode);
+		spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
+	}
+	spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
+	return 0;
+}
+
 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 			 const char *symname)
 {
@@ -3436,6 +3470,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 				     inode->i_mapping, GFP_NOFS);
 		extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 				     inode->i_mapping, GFP_NOFS);
+		INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
 		mutex_init(&BTRFS_I(inode)->csum_mutex);
 		mutex_init(&BTRFS_I(inode)->extent_mutex);
 		BTRFS_I(inode)->delalloc_bytes = 0;

fs/btrfs/ioctl.c

@@ -274,8 +274,7 @@ int btrfs_defrag_file(struct file *file)
 		 */
 		clear_page_dirty_for_io(page);
 
-		set_extent_delalloc(io_tree, page_start,
-				    page_end, GFP_NOFS);
+		btrfs_set_extent_delalloc(inode, page_start, page_end);
 
 		unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 		set_page_dirty(page);
@@ -784,6 +783,7 @@ long btrfs_ioctl(struct file *file, unsigned int
 	case BTRFS_IOC_TRANS_END:
 		return btrfs_ioctl_trans_end(file);
 	case BTRFS_IOC_SYNC:
+		btrfs_start_delalloc_inodes(root);
 		btrfs_sync_fs(file->f_dentry->d_sb, 1);
 		return 0;
 	}