Btrfs: remove transaction from send

Lets try this again.  We can deadlock the box if we send on a box and try to
write onto the same fs with the app that is trying to listen to the send pipe.
This is because the writer could get stuck waiting for a transaction commit
which is being blocked by the send.  So fix this by making sure looking at the
commit roots is always going to be consistent.  We do this by keeping track of
which roots need to have their commit roots swapped during commit, and then
taking the commit_root_sem and swapping them all at once.  Then make sure we
take a read lock on the commit_root_sem in cases where we search the commit root
to make sure we're always looking at a consistent view of the commit roots.
Previously we had problems with this because we would swap a fs tree commit root
and then swap the extent tree commit root independently which would cause the
backref walking code to screw up sometimes.  With this patch we no longer
deadlock and pass all the weird send/receive corner cases.  Thanks,

Reportedy-by: Hugo Mills <hugo@carfax.org.uk>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Chris Mason <clm@fb.com>

1 files changed, 2 insertions, 55 deletions

diff --git a/fs/btrfs/send.c b/fs/btrfs/send.c
index d00534b7838..6b5f1365931 100644
--- a/fs/btrfs/send.c
+++ b/fs/btrfs/send.c
@@ -1268,8 +1268,10 @@ static int find_extent_clone(struct send_ctx *sctx,
 	}
 	logical = disk_byte + btrfs_file_extent_offset(eb, fi);
 
+	down_read(&sctx->send_root->fs_info->commit_root_sem);
 	ret = extent_from_logical(sctx->send_root->fs_info, disk_byte, tmp_path,
 				  &found_key, &flags);
+	up_read(&sctx->send_root->fs_info->commit_root_sem);
 	btrfs_release_path(tmp_path);
 
 	if (ret < 0)
@@ -5367,57 +5369,21 @@ out:
 static int full_send_tree(struct send_ctx *sctx)
 {
 	int ret;
-	struct btrfs_trans_handle *trans = NULL;
 	struct btrfs_root *send_root = sctx->send_root;
 	struct btrfs_key key;
 	struct btrfs_key found_key;
 	struct btrfs_path *path;
 	struct extent_buffer *eb;
 	int slot;
-	u64 start_ctransid;
-	u64 ctransid;
 
 	path = alloc_path_for_send();
 	if (!path)
 		return -ENOMEM;
 
-	spin_lock(&send_root->root_item_lock);
-	start_ctransid = btrfs_root_ctransid(&send_root->root_item);
-	spin_unlock(&send_root->root_item_lock);
-
 	key.objectid = BTRFS_FIRST_FREE_OBJECTID;
 	key.type = BTRFS_INODE_ITEM_KEY;
 	key.offset = 0;
 
-join_trans:
-	/*
-	 * We need to make sure the transaction does not get committed
-	 * while we do anything on commit roots. Join a transaction to prevent
-	 * this.
-	 */
-	trans = btrfs_join_transaction(send_root);
-	if (IS_ERR(trans)) {
-		ret = PTR_ERR(trans);
-		trans = NULL;
-		goto out;
-	}
-
-	/*
-	 * Make sure the tree has not changed after re-joining. We detect this
-	 * by comparing start_ctransid and ctransid. They should always match.
-	 */
-	spin_lock(&send_root->root_item_lock);
-	ctransid = btrfs_root_ctransid(&send_root->root_item);
-	spin_unlock(&send_root->root_item_lock);
-
-	if (ctransid != start_ctransid) {
-		WARN(1, KERN_WARNING "BTRFS: the root that you're trying to "
-				     "send was modified in between. This is "
-				     "probably a bug.\n");
-		ret = -EIO;
-		goto out;
-	}
-
 	ret = btrfs_search_slot_for_read(send_root, &key, path, 1, 0);
 	if (ret < 0)
 		goto out;
@@ -5425,19 +5391,6 @@ join_trans:
 		goto out_finish;
 
 	while (1) {
-		/*
-		 * When someone want to commit while we iterate, end the
-		 * joined transaction and rejoin.
-		 */
-		if (btrfs_should_end_transaction(trans, send_root)) {
-			ret = btrfs_end_transaction(trans, send_root);
-			trans = NULL;
-			if (ret < 0)
-				goto out;
-			btrfs_release_path(path);
-			goto join_trans;
-		}
-
 		eb = path->nodes[0];
 		slot = path->slots[0];
 		btrfs_item_key_to_cpu(eb, &found_key, slot);
@@ -5465,12 +5418,6 @@ out_finish:
 
 out:
 	btrfs_free_path(path);
-	if (trans) {
-		if (!ret)
-			ret = btrfs_end_transaction(trans, send_root);
-		else
-			btrfs_end_transaction(trans, send_root);
-	}
 	return ret;
 }