A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 00:15 UTC
[Docs] Expand --allowed-media-domains security guidance with threat details (#36506)
Signed-off-by: Russell Bryant <rbryant@redhat.com>
Russell Bryant committed
Mar 9, 2026 at 13:43 UTC
d460a18fc656f7fb217b977d4c2ee1003af2a5b6
1 file changed
+16
-4
docs/usage/security.md
+16
-4
@@ -41,20 +41,20 @@ Key points from the PyTorch security guide:
41
- Messages are sent unencrypted
42
- Connections are accepted from anywhere without checks
43
44
-### Security Recommendations
44
+## Security Recommendations
45
46
-#### 1. **Network Isolation:**
46
+### 1. **Network Isolation:**
47
48
- Deploy vLLM nodes on a dedicated, isolated network
49
- Use network segmentation to prevent unauthorized access
50
- Implement appropriate firewall rules
51
52
-#### 2. **Configuration Best Practices:**
52
+### 2. **Configuration Best Practices:**
53
54
- Always set `VLLM_HOST_IP` to a specific IP address rather than using defaults
55
- Configure firewalls to only allow necessary ports between nodes
56
57
-#### 3. **Access Control:**
57
+### 3. **Access Control:**
58
59
- Restrict physical and network access to the deployment environment
60
- Implement proper authentication and authorization for management interfaces
@@ -66,6 +66,18 @@ Restrict domains that vLLM can access for media URLs by setting
66
`--allowed-media-domains` to prevent Server-Side Request Forgery (SSRF) attacks.
67
(e.g. `--allowed-media-domains upload.wikimedia.org github.com www.bogotobogo.com`)
68
69
+Without domain restrictions, a malicious user could supply URLs that:
70
+
71
+- **Target internal services**: Access internal network endpoints, cloud metadata
72
+ services (e.g. `169.254.169.254`), or other services not intended to be
73
+ publicly reachable (SSRF).
74
+- **Consume excessive resources**: Point to extremely large files or slow
75
+ endpoints, causing the server to download unbounded amounts of data and
76
+ exhausting memory, disk, or network bandwidth.
77
+
78
+By explicitly allowlisting only the domains you expect media to come from, you
79
+significantly reduce the attack surface for these types of abuse.
80
+
81
Also, consider setting `VLLM_MEDIA_URL_ALLOW_REDIRECTS=0` to prevent HTTP
82
redirects from being followed to bypass domain restrictions.
83