Btrfs: Throttle file_write when data=ordered is flushing the inode

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

6 files changed, 28 insertions, 2 deletions

diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index fe6ef8e34166..5ba83894c8b7 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -31,6 +31,7 @@ struct btrfs_inode {
 	struct extent_io_tree io_tree;
 	struct extent_io_tree io_failure_tree;
 	struct inode vfs_inode;
+	atomic_t ordered_writeback;
 
 	u64 ordered_trans;
 	/*
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 5d537f26dc83..8effdf4f5d6f 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -977,6 +977,7 @@ out_nolock:
 		     (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 	}
 	current->backing_dev_info = NULL;
+	btrfs_ordered_throttle(root, inode);
 	return num_written ? num_written : err;
 }
 
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index d58b4d5a1fef..b31f52d4f2ca 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -1419,6 +1419,7 @@ 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);
+	atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 	return 0;
 }
 
@@ -1728,6 +1729,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 			     inode->i_mapping, GFP_NOFS);
 	extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 			     inode->i_mapping, GFP_NOFS);
+	atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 	BTRFS_I(inode)->delalloc_bytes = 0;
 	BTRFS_I(inode)->root = root;
 
@@ -1956,6 +1958,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
 		extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 				     inode->i_mapping, GFP_NOFS);
 		BTRFS_I(inode)->delalloc_bytes = 0;
+		atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 	}
 	dir->i_sb->s_dirt = 1;
@@ -3292,6 +3295,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
 		extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
 				     inode->i_mapping, GFP_NOFS);
 		BTRFS_I(inode)->delalloc_bytes = 0;
+		atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 	}
 	dir->i_sb->s_dirt = 1;
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 3ee51e10c187..b474902c90e2 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -269,3 +269,16 @@ int btrfs_del_ordered_inode(struct inode *inode)
 	return ret;
 }
 
+int btrfs_ordered_throttle(struct btrfs_root *root, struct inode *inode)
+{
+	struct btrfs_transaction *cur = root->fs_info->running_transaction;
+	while(cur == root->fs_info->running_transaction &&
+	      atomic_read(&BTRFS_I(inode)->ordered_writeback)) {
+#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
+		congestion_wait(WRITE, HZ/20);
+#else
+		blk_congestion_wait(WRITE, HZ/20);
+#endif
+	}
+	return 0;
+}
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index f25c6771ec64..29047e0abaab 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -39,4 +39,5 @@ int btrfs_find_first_ordered_inode(struct btrfs_ordered_inode_tree *tree,
 				       u64 *root_objectid, u64 *objectid,
 				       struct inode **inode);
 int btrfs_del_ordered_inode(struct inode *inode);
+int btrfs_ordered_throttle(struct btrfs_root *root, struct inode *inode);
 #endif
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 1ed179c020af..c85cb48d95ee 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -512,8 +512,11 @@ int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
 		mutex_unlock(&root->fs_info->trans_mutex);
 		mutex_unlock(&root->fs_info->fs_mutex);
 
-		if (S_ISREG(inode->i_mode))
+		if (S_ISREG(inode->i_mode)) {
+			atomic_inc(&BTRFS_I(inode)->ordered_writeback);
 			filemap_fdatawrite(inode->i_mapping);
+			atomic_dec(&BTRFS_I(inode)->ordered_writeback);
+		}
 		iput(inode);
 
 		mutex_lock(&root->fs_info->fs_mutex);
@@ -530,8 +533,11 @@ int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
 		mutex_unlock(&root->fs_info->trans_mutex);
 		mutex_unlock(&root->fs_info->fs_mutex);
 
-		if (S_ISREG(inode->i_mode))
+		if (S_ISREG(inode->i_mode)) {
+			atomic_inc(&BTRFS_I(inode)->ordered_writeback);
 			filemap_write_and_wait(inode->i_mapping);
+			atomic_dec(&BTRFS_I(inode)->ordered_writeback);
+		}
 		atomic_dec(&inode->i_count);
 		iput(inode);