resolve user identifier to stable ID

currently, the policy approach node to user matching
with a quite naive approach looking at the username
provided in the policy and matched it with the username
on the nodes. This worked ok as long as usernames were
unique and did not change.

As usernames are no longer guarenteed to be unique in
an OIDC environment we cant rely on this.

This changes the mechanism that matches the user string
(now user token) with nodes:

- first find all potential users by looking up:
  - database ID
  - provider ID (OIDC)
  - username/email

If more than one user is matching, then the query is
rejected, and zero matching nodes are returned.

When a single user is found, the node is matched against
the User database ID, which are also present on the actual
node.

This means that from this commit, users can use the following
to identify users in the policy:
- provider identity (iss + sub)
- username
- email
- database id

There are more changes coming to this, so it is not recommended
to start using any of these new abilities, with the exception
of email, which will not change since it includes an @.

Signed-off-by: Kristoffer Dalby <[email protected]>

Author: kradalby

hscontrol/app.go

@@ -1026,14 +1026,18 @@ func (h *Headscale) loadACLPolicy() error {
 		if err != nil {
 			return fmt.Errorf("loading nodes from database to validate policy: %w", err)
 		}
+		users, err := h.db.ListUsers()
+		if err != nil {
+			return fmt.Errorf("loading users from database to validate policy: %w", err)
+		}
 
-		_, err = pol.CompileFilterRules(nodes)
+		_, err = pol.CompileFilterRules(users, nodes)
 		if err != nil {
 			return fmt.Errorf("verifying policy rules: %w", err)
 		}
 
 		if len(nodes) > 0 {
-			_, err = pol.CompileSSHPolicy(nodes[0], nodes)
+			_, err = pol.CompileSSHPolicy(nodes[0], users, nodes)
 			if err != nil {
 				return fmt.Errorf("verifying SSH rules: %w", err)
 			}

hscontrol/db/node_test.go

@@ -255,10 +255,10 @@ func (s *Suite) TestGetACLFilteredPeers(c *check.C) {
 	c.Assert(err, check.IsNil)
 	c.Assert(len(testPeers), check.Equals, 9)
 
-	adminRules, _, err := policy.GenerateFilterAndSSHRulesForTests(aclPolicy, adminNode, adminPeers)
+	adminRules, _, err := policy.GenerateFilterAndSSHRulesForTests(aclPolicy, adminNode, adminPeers, []types.User{*stor[0].user, *stor[1].user})
 	c.Assert(err, check.IsNil)
 
-	testRules, _, err := policy.GenerateFilterAndSSHRulesForTests(aclPolicy, testNode, testPeers)
+	testRules, _, err := policy.GenerateFilterAndSSHRulesForTests(aclPolicy, testNode, testPeers, []types.User{*stor[0].user, *stor[1].user})
 	c.Assert(err, check.IsNil)
 
 	peersOfAdminNode := policy.FilterNodesByACL(adminNode, adminPeers, adminRules)

hscontrol/db/routes.go

@@ -648,8 +648,13 @@ func EnableAutoApprovedRoutes(
 			if approvedAlias == node.User.Username() {
 				approvedRoutes = append(approvedRoutes, advertisedRoute)
 			} else {
+				users, err := ListUsers(tx)
+				if err != nil {
+					return fmt.Errorf("looking up users to expand route alias: %w", err)
+				}
+
 				// TODO(kradalby): figure out how to get this to depend on less stuff
-				approvedIps, err := aclPolicy.ExpandAlias(types.Nodes{node}, approvedAlias)
+				approvedIps, err := aclPolicy.ExpandAlias(types.Nodes{node}, users, approvedAlias)
 				if err != nil {
 					return fmt.Errorf("expanding alias %q for autoApprovers: %w", approvedAlias, err)
 				}

hscontrol/grpcv1.go

@@ -737,14 +737,18 @@ func (api headscaleV1APIServer) SetPolicy(
 	if err != nil {
 		return nil, fmt.Errorf("loading nodes from database to validate policy: %w", err)
 	}
+	users, err := api.h.db.ListUsers()
+	if err != nil {
+		return nil, fmt.Errorf("loading users from database to validate policy: %w", err)
+	}
 
-	_, err = pol.CompileFilterRules(nodes)
+	_, err = pol.CompileFilterRules(users, nodes)
 	if err != nil {
 		return nil, fmt.Errorf("verifying policy rules: %w", err)
 	}
 
 	if len(nodes) > 0 {
-		_, err = pol.CompileSSHPolicy(nodes[0], nodes)
+		_, err = pol.CompileSSHPolicy(nodes[0], users, nodes)
 		if err != nil {
 			return nil, fmt.Errorf("verifying SSH rules: %w", err)
 		}

hscontrol/mapper/mapper.go

@@ -153,6 +153,7 @@ func addNextDNSMetadata(resolvers []*dnstype.Resolver, node *types.Node) {
 func (m *Mapper) fullMapResponse(
 	node *types.Node,
 	peers types.Nodes,
+	users []types.User,
 	pol *policy.ACLPolicy,
 	capVer tailcfg.CapabilityVersion,
 ) (*tailcfg.MapResponse, error) {
@@ -167,6 +168,7 @@ func (m *Mapper) fullMapResponse(
 		pol,
 		node,
 		capVer,
+		users,
 		peers,
 		peers,
 		m.cfg,
@@ -189,8 +191,12 @@ func (m *Mapper) FullMapResponse(
 	if err != nil {
 		return nil, err
 	}
+	users, err := m.db.ListUsers()
+	if err != nil {
+		return nil, err
+	}
 
-	resp, err := m.fullMapResponse(node, peers, pol, mapRequest.Version)
+	resp, err := m.fullMapResponse(node, peers, users, pol, mapRequest.Version)
 	if err != nil {
 		return nil, err
 	}
@@ -253,6 +259,11 @@ func (m *Mapper) PeerChangedResponse(
 		return nil, err
 	}
 
+	users, err := m.db.ListUsers()
+	if err != nil {
+		return nil, fmt.Errorf("listing users for map response: %w", err)
+	}
+
 	var removedIDs []tailcfg.NodeID
 	var changedIDs []types.NodeID
 	for nodeID, nodeChanged := range changed {
@@ -276,6 +287,7 @@ func (m *Mapper) PeerChangedResponse(
 		pol,
 		node,
 		mapRequest.Version,
+		users,
 		peers,
 		changedNodes,
 		m.cfg,
@@ -508,16 +520,17 @@ func appendPeerChanges(
 	pol *policy.ACLPolicy,
 	node *types.Node,
 	capVer tailcfg.CapabilityVersion,
+	users []types.User,
 	peers types.Nodes,
 	changed types.Nodes,
 	cfg *types.Config,
 ) error {
-	packetFilter, err := pol.CompileFilterRules(append(peers, node))
+	packetFilter, err := pol.CompileFilterRules(users, append(peers, node))
 	if err != nil {
 		return err
 	}
 
-	sshPolicy, err := pol.CompileSSHPolicy(node, peers)
+	sshPolicy, err := pol.CompileSSHPolicy(node, users, peers)
 	if err != nil {
 		return err
 	}

hscontrol/mapper/mapper_test.go

@@ -159,6 +159,9 @@ func Test_fullMapResponse(t *testing.T) {
 	lastSeen := time.Date(2009, time.November, 10, 23, 9, 0, 0, time.UTC)
 	expire := time.Date(2500, time.November, 11, 23, 0, 0, 0, time.UTC)
 
+	user1 := types.User{Model: gorm.Model{ID: 0}, Name: "mini"}
+	user2 := types.User{Model: gorm.Model{ID: 1}, Name: "peer2"}
+
 	mini := &types.Node{
 		ID: 0,
 		MachineKey: mustMK(
@@ -173,8 +176,8 @@ func Test_fullMapResponse(t *testing.T) {
 		IPv4:       iap("100.64.0.1"),
 		Hostname:   "mini",
 		GivenName:  "mini",
-		UserID:     0,
-		User:       types.User{Name: "mini"},
+		UserID:     user1.ID,
+		User:       user1,
 		ForcedTags: []string{},
 		AuthKey:    &types.PreAuthKey{},
 		LastSeen:   &lastSeen,
@@ -253,8 +256,8 @@ func Test_fullMapResponse(t *testing.T) {
 		IPv4:       iap("100.64.0.2"),
 		Hostname:   "peer1",
 		GivenName:  "peer1",
-		UserID:     0,
-		User:       types.User{Name: "mini"},
+		UserID:     user1.ID,
+		User:       user1,
 		ForcedTags: []string{},
 		LastSeen:   &lastSeen,
 		Expiry:     &expire,
@@ -308,8 +311,8 @@ func Test_fullMapResponse(t *testing.T) {
 		IPv4:       iap("100.64.0.3"),
 		Hostname:   "peer2",
 		GivenName:  "peer2",
-		UserID:     1,
-		User:       types.User{Name: "peer2"},
+		UserID:     user2.ID,
+		User:       user2,
 		ForcedTags: []string{},
 		LastSeen:   &lastSeen,
 		Expiry:     &expire,
@@ -468,6 +471,7 @@ func Test_fullMapResponse(t *testing.T) {
 			got, err := mappy.fullMapResponse(
 				tt.node,
 				tt.peers,
+				[]types.User{user1, user2},
 				tt.pol,
 				0,
 			)