microsoft / microsoft/semantic-kernel

Python: server_url_validator allows Azure WireServer (168.63.129.16) and IPv6-embedded IPv4 forms (NAT64/6to4/Teredo)

Open
#14,240 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
C#
Stars
28.6k
Forks
4.8k
Avg merge
14h 13m
Merged PRs (30d)
18

Description

Describe the bug

server_url_validator is the default SSRF guard for OpenAPI plugin calls (allow_private_network_access defaults to False), and it classifies IPv4 by explicit octet ranges and IPv6 by a short list of properties. Two categories fall outside both:

  • 168.63.129.16 — Azure's WireServer / platform channel, reachable from every Azure VM. It's a publicly routable address, so none of the _try_classify_ipv4 branches match it.
  • IPv6 addresses that carry an IPv4 target inside them. _try_classify_ipv6 unwraps ::ffff: via ipv4_mapped, but NAT64, 6to4 and Teredo encodings aren't decoded.

The URL an OpenAPI plugin ends up calling is influenced by the model (server variables, path parameters), which is what makes this the SSRF surface the module exists for.

To Reproduce
import asyncio
from semantic_kernel.connectors.openapi_plugin.server_url_validator import validate_server_url

asyncio.run(validate_server_url("https://168.63.129.16/machine/"))            # returns
asyncio.run(validate_server_url("https://[64:ff9b::169.254.169.254]/latest/meta-data/"))  # returns

Full sweep on main @ 7d885dd:

address result category reported
169.254.169.254 (AWS/GCP/Azure/OCI/DO IMDS) blocked link-local
169.254.170.2 (AWS ECS task creds) blocked link-local
169.254.170.23 (AWS EKS Pod Identity) blocked link-local
169.254.42.42 (Scaleway) blocked link-local
100.100.100.200 (Alibaba) blocked carrier-grade NAT
192.0.0.192 (Oracle Cloud Classic) blocked reserved
168.63.129.16 (Azure WireServer) allowed
IPv6 form result
::ffff:169.254.169.254 (IPv4-mapped) blocked
::1, fd00:ec2::254 blocked
64:ff9b::169.254.169.254 (NAT64 well-known, RFC 6052) allowed
2002:a9fe:a9fe:: (6to4, RFC 3056) allowed
2001:0:4136:e378:8000:63bf:3fff:fdd2 (Teredo, RFC 4380) allowed
Expected behavior

168.63.129.16 blocked regardless of the private-range classification, and IPv6 forms carrying an embedded IPv4 decoded to that IPv4 before classification.

The Azure one can't be caught by range logic — it needs an explicit metadata denylist, checked before allow_private_network_access returns early, since "let me reach my own network" and "let me reach the host agent's credential endpoint" aren't the same request.

NAT64 needs the 64:ff9b::/96 and 64:ff9b:1::/48 prefixes decoded per RFC 6052 §2.2; 6to4 takes the IPv4 from bits 16–47; Teredo's is the low 32 bits XOR'd with all-ones.

Platform
  • OS: Linux
  • Python: 3.10.12
  • Semantic Kernel: main @ 7d885dd (python)
Additional context

Reachability varies by deployment — NAT64 needs a gateway on the path, and 6to4/Teredo relays are largely retired, so those three are defence-in-depth rather than universally exploitable. 168.63.129.16 is different: it is directly reachable on any Azure VM with no special network setup, which is where this repo's users are most likely to be running.

The IPv4-mapped unwrapping in try_categorize_non_public_address is already the right shape for the IPv6 side; the other encodings would slot in next to it.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start in semantic_kernel.connectors.openapi_plugin.server_url_validator at validate_server_url, then inspect try_categorize_non_public_address and the IPv4/IPv6 classification helpers. Use the reproduction addresses and expected behavior in the issue as checks; done means Azure WireServer is blocked and NAT64, 6to4, and Teredo forms are decoded before classification.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
api, security
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
55/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.