Merge pull request #218 from flatrun/fix/peer-deployment-management

feat: Add scoped peer deployment management

Author: nfebe

CHANGELOG.md

@@ -10,6 +10,8 @@ Sixth beta of the Albacore release, making connected servers manageable as one F
 - Docker Swarm and k3s orchestration adapters with nginx and Traefik routing adapters
 - Grouped incidents and configurable notification targets and delivery rules
 - HTTP, TCP, and container command health checks for web services and databases
+- Deployment management grants for administering allowed deployments through a peer
+- Service-specific health checks for mixed web and database deployments
 
 ### Fixed
 - Existing Fleet peers gain default access policies during startup repair without reconnecting
@@ -21,6 +23,11 @@ Sixth beta of the Albacore release, making connected servers manageable as one F
 - Settings, notifications, and API keys require explicit access for non-admin roles
 - Repeated metric alerts share one incident until every affected series recovers
 - Email headers keep the white logo visible in clients that ignore inline CSS
+- Peer requests preserve uploads, downloads, query parameters, and response types
+- Peer deployment actions use deployment permissions instead of Fleet configuration access
+- Fleet access intersects module permissions, server-qualified user grants, and peer policy
+- Operators no longer receive host shell or process-control access by default
+- Certificate lists and actions are restricted to assigned deployments for non-admin users
 
 ## [0.4.0-beta.4] - 2026-08-21
 

internal/api/ai_handlers.go

@@ -113,8 +113,10 @@ func (s *Server) platformSection(deploymentName string) ai.Section {
 			} else {
 				fmt.Fprintf(&b, "This deployment is not exposed through the reverse proxy\n")
 			}
-			if healthCheckConfigured(meta.HealthCheck) {
-				fmt.Fprintf(&b, "Configured health check type: %s\n", healthCheckType(meta.HealthCheck))
+			if checks := meta.EffectiveHealthChecks(); len(checks) > 0 {
+				for _, check := range checks {
+					fmt.Fprintf(&b, "Configured health check for %s: %s\n", check.Service, healthCheckType(check))
+				}
 			}
 			if len(meta.Databases) > 0 {
 				aliases := make([]string, 0, len(meta.Databases))

internal/api/authz.go

@@ -35,6 +35,29 @@ func (s *Server) requireDeploymentAccess(c *gin.Context, deploymentName, level s
 	return true
 }
 
+func restrictClusterServiceResources(c *gin.Context) {
+	actor := auth.GetActorFromContext(c)
+	if actor == nil || actor.User == nil || actor.User.Role != auth.RoleService || actor.User.Username != "__flatrun_cluster" {
+		c.Next()
+		return
+	}
+
+	path := c.Request.URL.Path
+	if strings.HasPrefix(path, "/api/deployments/") || strings.HasPrefix(path, "/api/containers/") ||
+		strings.HasPrefix(path, "/api/proxy/") {
+		c.Next()
+		return
+	}
+	for _, prefix := range []string{"/api/backups", "/api/certificates", "/api/credentials", "/api/images", "/api/security"} {
+		if strings.HasPrefix(path, prefix) {
+			c.JSON(http.StatusForbidden, gin.H{"error": "Fleet credentials require a deployment-scoped endpoint"})
+			c.Abort()
+			return
+		}
+	}
+	c.Next()
+}
+
 func (s *Server) requireContainerAccess(c *gin.Context, containerID, level string) bool {
 	if containerID == "" {
 		c.JSON(http.StatusBadRequest, gin.H{"error": "Container ID required"})

internal/api/authz_test.go

@@ -39,6 +39,37 @@ func actorMiddleware(actor *auth.ActorContext) gin.HandlerFunc {
 	}
 }
 
+func TestClusterServiceCredentialsRejectUnscopedSensitiveResources(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	actor := &auth.ActorContext{
+		Type: "api_key",
+		Role: auth.RoleService,
+		User: &auth.User{Role: auth.RoleService, Username: "__flatrun_cluster"},
+	}
+
+	for _, path := range []string{"/api/backups/other", "/api/credentials", "/api/security/events"} {
+		router := gin.New()
+		router.Use(actorMiddleware(actor), restrictClusterServiceResources)
+		router.GET(path, func(c *gin.Context) { c.Status(http.StatusNoContent) })
+		request := httptest.NewRequest(http.MethodGet, path, nil)
+		response := httptest.NewRecorder()
+		router.ServeHTTP(response, request)
+		if response.Code != http.StatusForbidden {
+			t.Fatalf("%s status = %d", path, response.Code)
+		}
+	}
+
+	router := gin.New()
+	router.Use(actorMiddleware(actor), restrictClusterServiceResources)
+	router.GET("/api/deployments/:name/security", func(c *gin.Context) { c.Status(http.StatusNoContent) })
+	request := httptest.NewRequest(http.MethodGet, "/api/deployments/app/security", nil)
+	response := httptest.NewRecorder()
+	router.ServeHTTP(response, request)
+	if response.Code != http.StatusNoContent {
+		t.Fatalf("deployment-scoped status = %d", response.Code)
+	}
+}
+
 func TestListVirtualHostsFiltersByDeploymentAccess(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 

internal/api/backup_handlers.go

@@ -70,6 +70,36 @@ func (s *Server) getBackup(c *gin.Context) {
 	c.JSON(http.StatusOK, gin.H{"backup": b})
 }
 
+func (s *Server) requireBackupDeployment(c *gin.Context) {
+	if s.backupManager == nil {
+		c.JSON(http.StatusServiceUnavailable, gin.H{"error": "Backup manager not enabled"})
+		c.Abort()
+		return
+	}
+	b, err := s.backupManager.GetBackup(c.Param("id"))
+	if err != nil || b.DeploymentName != c.Param("name") {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Backup not found"})
+		c.Abort()
+		return
+	}
+	c.Next()
+}
+
+func (s *Server) requireBackupJobDeployment(c *gin.Context) {
+	if s.backupManager == nil {
+		c.JSON(http.StatusServiceUnavailable, gin.H{"error": "Backup manager not enabled"})
+		c.Abort()
+		return
+	}
+	job := s.backupManager.GetJob(c.Param("id"))
+	if job == nil || job.DeploymentName != c.Param("name") {
+		c.JSON(http.StatusNotFound, gin.H{"error": "Job not found"})
+		c.Abort()
+		return
+	}
+	c.Next()
+}
+
 func (s *Server) createBackup(c *gin.Context) {
 	if s.backupManager == nil {
 		c.JSON(http.StatusServiceUnavailable, gin.H{"error": "Backup manager not enabled"})

internal/api/cert_renewal_test.go

@@ -16,6 +16,7 @@ import (
 	"testing"
 	"time"
 
+	"github.com/flatrun/agent/internal/auth"
 	"github.com/flatrun/agent/internal/docker"
 	"github.com/flatrun/agent/internal/nginx"
 	"github.com/flatrun/agent/internal/proxy"
@@ -173,6 +174,31 @@ func TestListCertificates_AnnotatesDeploymentID(t *testing.T) {
 	}
 }
 
+func TestListCertificates_FiltersByDeploymentAccess(t *testing.T) {
+	server, deploymentsPath, certsPath := setupRenewalTestServer(t)
+	writeSelfSignedCert(t, certsPath, "mine.example.com")
+	writeSelfSignedCert(t, certsPath, "other.example.com")
+	writeSelfSignedCert(t, certsPath, "orphan.example.com")
+	writeDeploymentWithDomains(t, deploymentsPath, "mine", []models.DomainConfig{{Domain: "mine.example.com"}})
+	writeDeploymentWithDomains(t, deploymentsPath, "other", []models.DomainConfig{{Domain: "other.example.com"}})
+
+	router := gin.New()
+	router.Use(actorMiddleware(testActor(auth.RoleOperator, map[string]string{"mine": auth.AccessLevelRead})))
+	router.GET("/certificates", server.listCertificates)
+	response := httptest.NewRecorder()
+	router.ServeHTTP(response, httptest.NewRequest(http.MethodGet, "/certificates", nil))
+
+	var payload struct {
+		Certificates []models.Certificate `json:"certificates"`
+	}
+	if response.Code != http.StatusOK || json.Unmarshal(response.Body.Bytes(), &payload) != nil {
+		t.Fatalf("status = %d, body = %s", response.Code, response.Body.String())
+	}
+	if len(payload.Certificates) != 1 || payload.Certificates[0].Domain != "mine.example.com" {
+		t.Fatalf("unexpected certificates: %+v", payload.Certificates)
+	}
+}
+
 func TestRenewDeploymentCertificates_CollectsAllDomainsAndAliases(t *testing.T) {
 	server, deploymentsPath, certsPath := setupRenewalTestServer(t)
 

internal/api/cluster_handlers.go

@@ -23,6 +23,7 @@ import (
 	"github.com/flatrun/agent/internal/routing"
 	"github.com/flatrun/agent/internal/system"
 	"github.com/flatrun/agent/pkg/config"
+	"github.com/flatrun/agent/pkg/models"
 	"github.com/flatrun/agent/pkg/version"
 	"github.com/gin-gonic/gin"
 )
@@ -709,6 +710,19 @@ func clusterPolicyAccess(policy cluster.PeerPolicy) ([]string, auth.DeploymentAc
 			permissions[auth.PermContainersRead.String()] = true
 			permissions[auth.PermContainersWrite.String()] = true
 			unrestrictedDeployments = mergeClusterDeploymentAccess(deployments, grant.Deployments, auth.AccessLevelWrite, unrestrictedDeployments)
+		case cluster.CapabilityDeploymentsManage:
+			for _, permission := range []auth.Permission{
+				auth.PermDeploymentsRead, auth.PermDeploymentsWrite, auth.PermDeploymentsDelete,
+				auth.PermContainersRead, auth.PermContainersWrite, auth.PermContainersDelete,
+				auth.PermCertificatesRead, auth.PermCertificatesWrite, auth.PermCertificatesDelete,
+				auth.PermSecurityRead, auth.PermSecurityWrite, auth.PermImagesRead,
+				auth.PermImagesWrite, auth.PermImagesDelete, auth.PermBackupsRead,
+				auth.PermBackupsWrite, auth.PermBackupsDelete,
+				auth.PermSchedulerRead, auth.PermSchedulerWrite, auth.PermSchedulerDelete,
+			} {
+				permissions[permission.String()] = true
+			}
+			unrestrictedDeployments = mergeClusterDeploymentAccess(deployments, grant.Deployments, auth.AccessLevelAdmin, unrestrictedDeployments)
 		case cluster.CapabilityCapacityRead:
 			permissions[auth.PermSystemRead.String()] = true
 		case cluster.CapabilityCapacityOffer:
@@ -734,13 +748,26 @@ func mergeClusterDeploymentAccess(access auth.DeploymentAccess, names []string,
 		return true
 	}
 	for _, name := range names {
-		if current, ok := access[name]; !ok || current == auth.AccessLevelRead && level == auth.AccessLevelWrite {
+		if current, ok := access[name]; !ok || clusterAccessLevelRank(level) > clusterAccessLevelRank(current) {
 			access[name] = level
 		}
 	}
 	return false
 }
 
+func clusterAccessLevelRank(level string) int {
+	switch level {
+	case auth.AccessLevelRead:
+		return 1
+	case auth.AccessLevelWrite:
+		return 2
+	case auth.AccessLevelAdmin:
+		return 3
+	default:
+		return 0
+	}
+}
+
 func (s *Server) applyClusterPeerPolicy(policy cluster.PeerPolicy) error {
 	if s.authManager == nil {
 		return fmt.Errorf("Authentication manager is not available")
@@ -833,6 +860,9 @@ func (s *Server) deleteClusterAPIKey(peerName string) error {
 }
 
 func (s *Server) clusterProxy(c *gin.Context) {
+	if !authorizePeerProxy(c) {
+		return
+	}
 	mgr := s.getClusterManager()
 	if mgr == nil {
 		c.JSON(http.StatusBadRequest, gin.H{"error": "Cluster is not enabled"})
@@ -853,18 +883,143 @@ func (s *Server) clusterProxy(c *gin.Context) {
 		body = c.Request.Body
 	}
 
-	data, status, headers, err := client.Forward(c.Request.Context(), c.Request.Method, "/api"+path, body)
+	forwardPath := "/api" + path
+	if c.Request.URL.RawQuery != "" {
+		forwardPath += "?" + c.Request.URL.RawQuery
+	}
+	resp, err := client.DoWithHeaders(c.Request.Context(), c.Request.Method, forwardPath, c.Request.Header, body)
 	if err != nil {
 		c.JSON(http.StatusBadGateway, gin.H{"error": fmt.Sprintf("Failed to proxy request: %v", err)})
 		return
 	}
 
-	for k, v := range headers {
-		if k != "Content-Length" && k != "Transfer-Encoding" {
-			c.Header(k, v)
+	defer resp.Body.Close()
+	for k, values := range resp.Header {
+		if k != "Content-Length" && k != "Transfer-Encoding" && k != "Connection" {
+			for _, value := range values {
+				c.Writer.Header().Add(k, value)
+			}
+		}
+	}
+	if c.Request.Method == http.MethodGet && path == "/deployments" {
+		s.writeScopedPeerDeployments(c, name, resp)
+		return
+	}
+	c.Status(resp.StatusCode)
+	_, _ = io.Copy(c.Writer, resp.Body)
+}
+
+func (s *Server) writeScopedPeerDeployments(c *gin.Context, peer string, resp *http.Response) {
+	actor := auth.GetActorFromContext(c)
+	if actor == nil || actor.Role == auth.RoleAdmin || resp.StatusCode != http.StatusOK {
+		c.Status(resp.StatusCode)
+		_, _ = io.Copy(c.Writer, resp.Body)
+		return
+	}
+	var payload struct {
+		Deployments []models.Deployment `json:"deployments"`
+		Path        string              `json:"path,omitempty"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&payload); err != nil {
+		c.JSON(http.StatusBadGateway, gin.H{"error": "Peer returned an invalid deployment list"})
+		return
+	}
+	visible := payload.Deployments[:0]
+	for _, deployment := range payload.Deployments {
+		if actor.CanAccessPeerDeployment(peer, deployment.Name, auth.AccessLevelRead) {
+			visible = append(visible, deployment)
+		}
+	}
+	payload.Deployments = visible
+	c.JSON(http.StatusOK, payload)
+}
+
+func authorizePeerProxy(c *gin.Context) bool {
+	actor := auth.GetActorFromContext(c)
+	if actor == nil {
+		c.JSON(http.StatusUnauthorized, gin.H{"error": "Not authenticated"})
+		return false
+	}
+
+	path := c.Param("path")
+	peer := c.Param("name")
+	deployment := peerProxyDeployment(path)
+	if deployment == "" {
+		deployment = strings.TrimSpace(c.GetHeader("X-FlatRun-Deployment"))
+	}
+	if actor.Role != auth.RoleAdmin && path != "/deployments" && deployment == "" {
+		c.JSON(http.StatusForbidden, gin.H{"error": "A deployment scope is required"})
+		return false
+	}
+	requiredLevel := auth.AccessLevelRead
+	if c.Request.Method != http.MethodGet && c.Request.Method != http.MethodHead {
+		requiredLevel = auth.AccessLevelWrite
+	}
+	if c.Request.Method == http.MethodDelete {
+		requiredLevel = auth.AccessLevelAdmin
+	}
+	if deployment != "" && !actor.CanAccessPeerDeployment(peer, deployment, requiredLevel) {
+		c.JSON(http.StatusForbidden, gin.H{"error": "No access to this peer deployment"})
+		return false
+	}
+	permission := auth.PermDeploymentsRead
+	if c.Request.Method != http.MethodGet && c.Request.Method != http.MethodHead {
+		permission = auth.PermDeploymentsWrite
+	}
+	if c.Request.Method == http.MethodDelete {
+		permission = auth.PermDeploymentsDelete
+	}
+
+	switch {
+	case strings.HasPrefix(path, "/containers/"):
+		permission = auth.PermContainersRead
+		if c.Request.Method != http.MethodGet && c.Request.Method != http.MethodHead {
+			permission = auth.PermContainersWrite
+		}
+		if c.Request.Method == http.MethodDelete {
+			permission = auth.PermContainersDelete
+		}
+	case strings.HasPrefix(path, "/certificates"), strings.HasPrefix(path, "/proxy/"):
+		permission = auth.PermCertificatesRead
+		if c.Request.Method != http.MethodGet && c.Request.Method != http.MethodHead {
+			permission = auth.PermCertificatesWrite
+		}
+		if c.Request.Method == http.MethodDelete {
+			permission = auth.PermCertificatesDelete
+		}
+	case strings.Contains(path, "/security"):
+		permission = auth.PermSecurityRead
+		if c.Request.Method != http.MethodGet && c.Request.Method != http.MethodHead {
+			permission = auth.PermSecurityWrite
+		}
+	case strings.HasPrefix(path, "/backups"), strings.Contains(path, "/backups"):
+		permission = auth.PermBackupsRead
+		if c.Request.Method != http.MethodGet && c.Request.Method != http.MethodHead {
+			permission = auth.PermBackupsWrite
+		}
+		if c.Request.Method == http.MethodDelete {
+			permission = auth.PermBackupsDelete
+		}
+	case strings.HasPrefix(path, "/credentials"):
+		permission = auth.PermRegistriesRead
+	}
+
+	if !actor.HasPermission(permission) {
+		c.JSON(http.StatusForbidden, gin.H{"error": "Permission denied", "required": permission})
+		return false
+	}
+	return true
+}
+
+func peerProxyDeployment(path string) string {
+	parts := strings.Split(strings.Trim(path, "/"), "/")
+	if len(parts) >= 2 && parts[0] == "deployments" {
+		name, err := url.PathUnescape(parts[1])
+		if err == nil {
+			return name
 		}
 	}
-	c.Data(status, "application/json", data)
+	return ""
 }
 
 func (s *Server) clusterAggregateDeployments(c *gin.Context) {