Btrfs: make delalloc inodes be flushed by multi-task

This patch introduce a new worker pool named "flush_workers", and if we
want to force all the inode with pending delalloc to the disks, we can
queue those inodes into the work queue of the worker pool, in this way,
those inodes will be flushed by multi-task.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@fusionio.com>

5 files changed, 103 insertions, 8 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 8fd9fe4282f5..cad16566da37 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -1333,6 +1333,7 @@ struct btrfs_fs_info {
 	struct btrfs_workers generic_worker;
 	struct btrfs_workers workers;
 	struct btrfs_workers delalloc_workers;
+	struct btrfs_workers flush_workers;
 	struct btrfs_workers endio_workers;
 	struct btrfs_workers endio_meta_workers;
 	struct btrfs_workers endio_meta_write_workers;
@@ -3277,6 +3278,19 @@ int btrfs_csum_truncate(struct btrfs_trans_handle *trans,
 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 			     struct list_head *list, int search_commit);
 /* inode.c */
+struct btrfs_delalloc_work {
+	struct inode *inode;
+	int wait;
+	int delay_iput;
+	struct completion completion;
+	struct list_head list;
+	struct btrfs_work work;
+};
+
+struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
+						    int wait, int delay_iput);
+void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work);
+
 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
 					   size_t pg_offset, u64 start, u64 len,
 					   int create);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 7cda51995c1e..bd70c2852ba0 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -2279,6 +2279,10 @@ int open_ctree(struct super_block *sb,
 			   fs_info->thread_pool_size,
 			   &fs_info->generic_worker);
 
+	btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
+			   fs_info->thread_pool_size,
+			   &fs_info->generic_worker);
+
 	btrfs_init_workers(&fs_info->submit_workers, "submit",
 			   min_t(u64, fs_devices->num_devices,
 			   fs_info->thread_pool_size),
@@ -2350,6 +2354,7 @@ int open_ctree(struct super_block *sb,
 	ret |= btrfs_start_workers(&fs_info->delayed_workers);
 	ret |= btrfs_start_workers(&fs_info->caching_workers);
 	ret |= btrfs_start_workers(&fs_info->readahead_workers);
+	ret |= btrfs_start_workers(&fs_info->flush_workers);
 	if (ret) {
 		err = -ENOMEM;
 		goto fail_sb_buffer;
@@ -2667,6 +2672,7 @@ fail_sb_buffer:
 	btrfs_stop_workers(&fs_info->submit_workers);
 	btrfs_stop_workers(&fs_info->delayed_workers);
 	btrfs_stop_workers(&fs_info->caching_workers);
+	btrfs_stop_workers(&fs_info->flush_workers);
 fail_alloc:
 fail_iput:
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
@@ -3339,6 +3345,7 @@ int close_ctree(struct btrfs_root *root)
 	btrfs_stop_workers(&fs_info->delayed_workers);
 	btrfs_stop_workers(&fs_info->caching_workers);
 	btrfs_stop_workers(&fs_info->readahead_workers);
+	btrfs_stop_workers(&fs_info->flush_workers);
 
 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
 	if (btrfs_test_opt(root, CHECK_INTEGRITY))
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index db3dd4ed057f..dce9e218b845 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -71,6 +71,7 @@ static const struct file_operations btrfs_dir_file_operations;
 static struct extent_io_ops btrfs_extent_io_ops;
 
 static struct kmem_cache *btrfs_inode_cachep;
+static struct kmem_cache *btrfs_delalloc_work_cachep;
 struct kmem_cache *btrfs_trans_handle_cachep;
 struct kmem_cache *btrfs_transaction_cachep;
 struct kmem_cache *btrfs_path_cachep;
@@ -7204,6 +7205,8 @@ void btrfs_destroy_cachep(void)
 		kmem_cache_destroy(btrfs_path_cachep);
 	if (btrfs_free_space_cachep)
 		kmem_cache_destroy(btrfs_free_space_cachep);
+	if (btrfs_delalloc_work_cachep)
+		kmem_cache_destroy(btrfs_delalloc_work_cachep);
 }
 
 int btrfs_init_cachep(void)
@@ -7238,6 +7241,13 @@ int btrfs_init_cachep(void)
 	if (!btrfs_free_space_cachep)
 		goto fail;
 
+	btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
+			sizeof(struct btrfs_delalloc_work), 0,
+			SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
+			NULL);
+	if (!btrfs_delalloc_work_cachep)
+		goto fail;
+
 	return 0;
 fail:
 	btrfs_destroy_cachep();
@@ -7448,6 +7458,49 @@ out_notrans:
 	return ret;
 }
 
+static void btrfs_run_delalloc_work(struct btrfs_work *work)
+{
+	struct btrfs_delalloc_work *delalloc_work;
+
+	delalloc_work = container_of(work, struct btrfs_delalloc_work,
+				     work);
+	if (delalloc_work->wait)
+		btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
+	else
+		filemap_flush(delalloc_work->inode->i_mapping);
+
+	if (delalloc_work->delay_iput)
+		btrfs_add_delayed_iput(delalloc_work->inode);
+	else
+		iput(delalloc_work->inode);
+	complete(&delalloc_work->completion);
+}
+
+struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
+						    int wait, int delay_iput)
+{
+	struct btrfs_delalloc_work *work;
+
+	work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
+	if (!work)
+		return NULL;
+
+	init_completion(&work->completion);
+	INIT_LIST_HEAD(&work->list);
+	work->inode = inode;
+	work->wait = wait;
+	work->delay_iput = delay_iput;
+	work->work.func = btrfs_run_delalloc_work;
+
+	return work;
+}
+
+void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
+{
+	wait_for_completion(&work->completion);
+	kmem_cache_free(btrfs_delalloc_work_cachep, work);
+}
+
 /*
  * some fairly slow code that needs optimization. This walks the list
  * of all the inodes with pending delalloc and forces them to disk.
@@ -7457,10 +7510,15 @@ int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
 	struct list_head *head = &root->fs_info->delalloc_inodes;
 	struct btrfs_inode *binode;
 	struct inode *inode;
+	struct btrfs_delalloc_work *work, *next;
+	struct list_head works;
+	int ret = 0;
 
 	if (root->fs_info->sb->s_flags & MS_RDONLY)
 		return -EROFS;
 
+	INIT_LIST_HEAD(&works);
+
 	spin_lock(&root->fs_info->delalloc_lock);
 	while (!list_empty(head)) {
 		binode = list_entry(head->next, struct btrfs_inode,
@@ -7470,11 +7528,14 @@ int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
 			list_del_init(&binode->delalloc_inodes);
 		spin_unlock(&root->fs_info->delalloc_lock);
 		if (inode) {
-			filemap_flush(inode->i_mapping);
-			if (delay_iput)
-				btrfs_add_delayed_iput(inode);
-			else
-				iput(inode);
+			work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
+			if (!work) {
+				ret = -ENOMEM;
+				goto out;
+			}
+			list_add_tail(&work->list, &works);
+			btrfs_queue_worker(&root->fs_info->flush_workers,
+					   &work->work);
 		}
 		cond_resched();
 		spin_lock(&root->fs_info->delalloc_lock);
@@ -7493,7 +7554,12 @@ int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
 		    atomic_read(&root->fs_info->async_delalloc_pages) == 0));
 	}
 	atomic_dec(&root->fs_info->async_submit_draining);
-	return 0;
+out:
+	list_for_each_entry_safe(work, next, &works, list) {
+		list_del_init(&work->list);
+		btrfs_wait_and_free_delalloc_work(work);
+	}
+	return ret;
 }
 
 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index 242d6de4d8eb..270f24ffe1be 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -4061,7 +4061,11 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
 	       (unsigned long long)rc->block_group->key.objectid,
 	       (unsigned long long)rc->block_group->flags);
 
-	btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
+	ret = btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
+	if (ret < 0) {
+		err = ret;
+		goto out;
+	}
 	btrfs_wait_ordered_extents(fs_info->tree_root, 0);
 
 	while (1) {
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 4e1def4c06b1..9c466f9f8175 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -1497,7 +1497,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 		WARN_ON(cur_trans != trans->transaction);
 
 		if (flush_on_commit || snap_pending) {
-			btrfs_start_delalloc_inodes(root, 1);
+			ret = btrfs_start_delalloc_inodes(root, 1);
+			if (ret) {
+				btrfs_abort_transaction(trans, root, ret);
+				goto cleanup_transaction;
+			}
 			btrfs_wait_ordered_extents(root, 1);
 		}