btrfs: catch cow on deleting snapshots

When debugging some weird extent reference bug I suspected that we were
changing a snapshot while we were deleting it, which could explain my
bug.  This was indeed what was happening, and this patch helped me
verify my theory.  It is never correct to modify the snapshot once it's
being deleted, so mark the root when we are deleting it and make sure we
complain about it when it happens.
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/ctree.c

@@ -1440,6 +1440,10 @@ noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
 	u64 search_start;
 	int ret;
 
+	if (test_bit(BTRFS_ROOT_DELETING, &root->state))
+		btrfs_err(fs_info,
+			"COW'ing blocks on a fs root that's being dropped");
+
 	if (trans->transaction != fs_info->running_transaction)
 		WARN(1, KERN_CRIT "trans %llu running %llu\n",
 		       trans->transid,

fs/btrfs/ctree.h

@@ -1197,6 +1197,7 @@ enum {
 	BTRFS_ROOT_FORCE_COW,
 	BTRFS_ROOT_MULTI_LOG_TASKS,
 	BTRFS_ROOT_DIRTY,
+	BTRFS_ROOT_DELETING,
 };
 
 /*

fs/btrfs/extent-tree.c

@@ -9275,6 +9275,15 @@ int btrfs_drop_snapshot(struct btrfs_root *root,
 	if (block_rsv)
 		trans->block_rsv = block_rsv;
 
+	/*
+	 * This will help us catch people modifying the fs tree while we're
+	 * dropping it.  It is unsafe to mess with the fs tree while it's being
+	 * dropped as we unlock the root node and parent nodes as we walk down
+	 * the tree, assuming nothing will change.  If something does change
+	 * then we'll have stale information and drop references to blocks we've
+	 * already dropped.
+	 */
+	set_bit(BTRFS_ROOT_DELETING, &root->state);
 	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
 		level = btrfs_header_level(root->node);
 		path->nodes[level] = btrfs_lock_root_node(root);