Btrfs: Add a per-inode csum mutex to avoid races creating csum items

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

6 files changed, 21 insertions, 6 deletions

diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index 5ba83894c8b7..40b4e0c9cd09 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -30,6 +30,7 @@ struct btrfs_inode {
 	struct extent_map_tree extent_tree;
 	struct extent_io_tree io_tree;
 	struct extent_io_tree io_failure_tree;
+	struct mutex csum_mutex;
 	struct inode vfs_inode;
 	atomic_t ordered_writeback;
 
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index ece221cba90c..8037792f8789 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -267,13 +267,13 @@ static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
 
 	/* FIXME...EIEIO, ENOSPC and more */
 	/* insert any holes we need to create */
-	if (isize < end_pos) {
+	if (isize < start_pos) {
 		u64 last_pos_in_file;
 		u64 hole_size;
 		u64 mask = root->sectorsize - 1;
 		last_pos_in_file = (isize + mask) & ~mask;
-		hole_size = (end_pos - last_pos_in_file + mask) & ~mask;
-		if (last_pos_in_file < end_pos) {
+		hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
+		if (last_pos_in_file < start_pos) {
 			err = btrfs_drop_extents(trans, root, inode,
 						 last_pos_in_file,
 						 last_pos_in_file + hole_size,
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index bbba3350d023..d39433dfb2c7 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -351,7 +351,9 @@ int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 	trans = btrfs_start_transaction(root, 1);
 
 	btrfs_set_trans_block_group(trans, inode);
+	mutex_lock(&BTRFS_I(inode)->csum_mutex);
 	btrfs_csum_file_blocks(trans, root, inode, bio, sums);
+	mutex_unlock(&BTRFS_I(inode)->csum_mutex);
 
 	ret = btrfs_end_transaction(trans, root);
 	BUG_ON(ret);
@@ -1400,6 +1402,7 @@ 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);
+	mutex_init(&BTRFS_I(inode)->csum_mutex);
 	atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 	return 0;
 }
@@ -1701,6 +1704,7 @@ 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);
+	mutex_init(&BTRFS_I(inode)->csum_mutex);
 	atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 	BTRFS_I(inode)->delalloc_bytes = 0;
 	BTRFS_I(inode)->root = root;
@@ -1924,6 +1928,7 @@ 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);
+		mutex_init(&BTRFS_I(inode)->csum_mutex);
 		BTRFS_I(inode)->delalloc_bytes = 0;
 		atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
@@ -2862,6 +2867,7 @@ 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);
+		mutex_init(&BTRFS_I(inode)->csum_mutex);
 		BTRFS_I(inode)->delalloc_bytes = 0;
 		atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
 		BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 5e4c0d95ce43..254da8225664 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -152,12 +152,13 @@ int btrfs_add_ordered_inode(struct inode *inode)
 			   inode->i_ino, &entry->rb_node);
 
 	BTRFS_I(inode)->ordered_trans = transid;
+	if (!node)
+		igrab(inode);
 
 	write_unlock(&tree->lock);
+
 	if (node)
 		kfree(entry);
-	else
-		igrab(inode);
 	return 0;
 }
 
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 0c53ff775b92..8e909cb97c6d 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -477,7 +477,7 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 			if (err)
 				ret = err;
 			nr = trans->blocks_used;
-			ret = btrfs_end_transaction(trans, tree_root);
+			ret = btrfs_end_transaction_throttle(trans, tree_root);
 			BUG_ON(ret);
 
 			mutex_unlock(&root->fs_info->drop_mutex);
diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index 1677e4edaf6f..b17693f61fbc 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -38,8 +38,15 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 	u64 last_ret = 0;
 
 	if (root->fs_info->extent_root == root) {
+		/*
+		 * there's recursion here right now in the tree locking,
+		 * we can't defrag the extent root without deadlock
+		 */
+		goto out;
+#if 0
 		mutex_lock(&root->fs_info->alloc_mutex);
 		is_extent = 1;
+#endif
 	}
 
 	if (root->ref_cows == 0 && !is_extent)