xds: document how authorization works

agent/xds/delta.go

@@ -138,7 +138,7 @@ func (s *Server) processDelta(stream ADSDeltaStream, reqCh <-chan *envoy_discove
 	}
 
 	checkStreamACLs := func(cfgSnap *proxycfg.ConfigSnapshot) error {
-		return s.checkStreamACLs(stream.Context(), cfgSnap)
+		return s.authorize(stream.Context(), cfgSnap)
 	}
 
 	for {

agent/xds/server.go

@@ -327,7 +327,7 @@ func (s *Server) process(stream ADSStream, reqCh <-chan *envoy_discovery_v3.Disc
 	}
 
 	checkStreamACLs := func(cfgSnap *proxycfg.ConfigSnapshot) error {
-		return s.checkStreamACLs(stream.Context(), cfgSnap)
+		return s.authorize(stream.Context(), cfgSnap)
 	}
 
 	for {
@@ -564,13 +564,22 @@ func NewGRPCServer(s *Server, tlsConfigurator *tlsutil.Configurator) *grpc.Serve
 	return srv
 }
 
-func (s *Server) checkStreamACLs(streamCtx context.Context, cfgSnap *proxycfg.ConfigSnapshot) error {
+// authorize the xDS request using the token stored in ctx. This authorization is
+// a bit different from most interfaces. Instead of explicitly authorizing or
+// filtering each piece of data in the response, the request is authorized
+// by checking the token has `service:write` for the service ID of the destination
+// service (for kind=ConnectProxy), or the gateway service (for other kinds).
+// This authorization strategy requires that agent/proxycfg only fetches data
+// using a token with the same permissions, and that it stores the data by
+// proxy ID. We assume that any data in the snapshot was already filtered,
+// which allows this authorization to be a shallow authorization check
+// for all the data in a ConfigSnapshot.
+func (s *Server) authorize(ctx context.Context, cfgSnap *proxycfg.ConfigSnapshot) error {
 	if cfgSnap == nil {
 		return status.Errorf(codes.Unauthenticated, "unauthenticated: no config snapshot")
 	}
 
-	authz, err := s.ResolveToken(tokenFromContext(streamCtx))
+	authz, err := s.ResolveToken(tokenFromContext(ctx))
-
 	if acl.IsErrNotFound(err) {
 		return status.Errorf(codes.Unauthenticated, "unauthenticated: %v", err)
 	} else if acl.IsErrPermissionDenied(err) {

contributing/service-mesh/README.md

@@ -1,12 +1,15 @@
 # Service Mesh (Connect)
 
 - call out: envoy/proxy is the data plane, Consul is the control plane
-- agent/xds - gRPC service that implements
+- [xDS Server] - a gRPC service that implements [xDS] and handles requests from an [envoy proxy].
-  [xDS](https://www.envoyproxy.io/docs/envoy/latest/api-docs/xds_protocol)
+- [agent/proxycfg]
-- [agent/proxycfg](https://github.com/hashicorp/consul/blob/master/agent/proxycfg/proxycfg.go)
 - CA Manager - certificate authority
 - command/connect/envoy - bootstrapping and running envoy
 - command/connect/proxy - built-in proxy that is dev-only and not supported 
   for production.
 - `connect/` - "Native" service mesh
 
+[xDS Server]: ./xds.md
+[xDS]: https://www.envoyproxy.io/docs/envoy/latest/api-docs/xds_protocol
+[envoy proxy]: https://www.consul.io/docs/connect/proxies/envoy
+[agent/proxycfg]: https://github.com/hashicorp/consul/blob/main/agent/proxycfg

contributing/service-mesh/xds.md

@@ -0,0 +1,25 @@
+# xDS Server
+
+The xDS Server is a gRPC service that implements [xDS] and handles requests from
+an [envoy proxy].
+
+[xDS]: https://www.envoyproxy.io/docs/envoy/latest/api-docs/xds_protocol
+[envoy proxy]: https://www.consul.io/docs/connect/proxies/envoy
+
+
+## Authorization
+
+Requests to the xDS server are authorized based on an assumption of how
+`proxycfg.ConfigSnapshot` are constructed. Most interfaces (HTTP, DNS, RPC)
+authorize requests by authorizing the data in the response, or by filtering
+out data that the requester is not authorized to view. The xDS server authorizes
+requests by looking at the proxy ID in the request and ensuring the ACL token has
+`service:write` access to either the destination service (for kind=ConnectProxy), or
+the gateway service (for other kinds).
+
+This authorization strategy requires that [agent/proxycfg] only fetches data using a
+token with the same permissions, and that it only stores data by proxy ID. We assume
+that any data in the snapshot was already filtered, which allows this authorization to
+only perform a shallow check against the proxy ID.
+
+[agent/proxycfg]: https://github.com/hashicorp/consul/blob/main/agent/proxycfg