Information Disclosure

The Lure of the Schema

A primary target for attackers is the GraphQL schema, which reveals the structure of an application's data. It's like having a map of the database, including all fields and types, and it is often exposed via the introspection feature which is enabled by default. This allows hackers to understand the data model, business logic and potential attack vectors. Tools like InQL can automate this process, extracting schema information and outputting it in formats that are useful for security testing. However, some GraphQL implementations allow the disabling of introspection, but there are ways around that.

inql -t http://localhost:5013/graphql --generate-tsv

Bypassing Disabled Introspection

Even if introspection is disabled, there are several techniques to gather schema information:

• Non-Production Environments: Development and staging environments may have less stringent security than production, and therefore introspection might be enabled. Subdomains like 'staging' or 'dev' are worth checking for GraphQL services with introspection enabled.

Black-Hat-GraphQL/resources/non-production-graphql-urls.txt at master · dolevf/Black-Hat-GraphQLGitHub

• The __type Meta-field:

The WAF contains rules tailored to GraphQL applications, one of which blocks attempts to introspect the GraphQL API via the __schema meta-field but doesn’t take into consideration other introspection meta-fields. The rule itself is defined in JSON in the following way:

This can be used as a "canary" to determine if introspection is disabled. By sending a query using __type and checking the response, an attacker can confirm whether the meta-field is available.

# A __type introspection canary query
{
    __type(name:"Query") {
    name
    }
}

---

• Field Suggestions: When a client misspells a field, the server may return a suggestion, which can be abused to discover the fields in the schema. Tools like Clairvoyance exploit this feature by sending queries based on a dictionary of common words to reconstruct the schema. The edit-distance algorithm will determine whether suggestions are provided.

query {
    pastes {
        owne
    }
}

-----------------
{"errors": [{"message": "Cannot query field \"owne\" on type \"Paste Object\". Did you mean \"owner\" or \"ownerId\"?", "locations": [
{"line": 24,"column": 3}

---

• Field Stuffing: Attackers insert lists of potential field names into queries to discover additional information. By observing what is returned, they can uncover sensitive fields that are not intended for public access.

Schema Naming ConventionsApollo GraphQL Docs

query {
    user {
        name
        username
        address
        birthday
        age
        password
        sin
        ssn
        apiKey
        token
        emailAddress
        status
    }
}

---

Type Stuffing in the __type Meta-field:

Type stuffing exploits weaknesses in GraphQL's introspection disabling by targeting the __type meta-field to uncover schema details.

• Technique:

  • By supplying potential type names in the __type(name:"TypeName") query, attackers can identify valid types and their fields.

  • For example, querying for PasteObject might reveal its fields, such as id, title, content, public, and more.

• Naming Convention: GraphQL type names typically follow UpperCamelCase, like PrivatePasteProperties. Testing these systematically can uncover existing types.

• Example Query:

  {
    __type(name: "PasteObject") {
      name
      fields {
        name
      }
    }
  }

  Response:

  {
    "data": {
      "__type": {
        "name": "PasteObject",
        "fields": [
          {"name": "id"},
          {"name": "title"},
          {"name": "content"},
          {"name": "public"},
          ...
        ]
      }
    }
  }

---

Automating Field Suggestion and Stuffing Using Clairvoyance

GraphQL Voyager

python3 -m clairvoyance http://localhost:5013/graphql -w ~/high-frequency-vocabulary/30k.txt -o clairvoyance-dvga-schema.json

The original DVGA schema

Tools like the Custom Word List Generator (CeWL), which comes preinstalled in Kali, can extract keywords from the application’s frontend HTML. Try using the following one-liner to profile and extract information from the DVGA interface:

cewl http://localhost:5013/

This command will return a list of words that you can use in a manual field-stuffing attack. Alternatively, merge it with your list of 30,000 words and use it with Clairvoyance. You can merge two text files by using a simple Bash command:

paste -d "\n" wordlist1.txt wordlist2.txt > merged_wordlist.txt

---

Error Messages: A Goldmine for Attackers

GraphQL’s tendency to return verbose error messages, while helpful for developers, can be exploited. These messages may reveal internal information such as:

• SQL statements used by the server to interact with the database.

• Database column names.

• User credentials.

• Inferring Information from Stack Traces:

query {
    pastes {
        titled
    }
}

---------------------------
{"errors": [{"message": "Cannot query field \"titled\" on type \"PasteObject\".Did you mean \"title\"?","extensions": {"exception": {"stack": [" File \"/Users/dvga-user/Desktop/Damn-Vulnerable-GraphQL-Application/venv/lib/python3.x/site-packages/gevent/baseserver.py\", line 34,in _handle_and_close_when_done\nreturn handle(*args_tuple)\n",--snip--" File \"/Users/dvga-user/Desktop/Damn-Vulnerable-GraphQL-Application/venv/lib/python3.x/site-packages/flask/app.py\", line 2464,in __call__\nreturn self.wsgi_app(environ, start_response)\n",--snip--],"debug": "Traceback (most recent call last):\n File \"/Users/dvga-user/Desktop/Damn-Vulnerable-GraphQL-Application/venv/lib/python3.x/site-packages/flask_sockets.py\", line 40, in __call__\n ..."path": \"/Users/dvga-user/Desktop/Damn-Vulnerable-GraphQL-Application/core/view_override.py"}}}]}

---

Leaking Data via GET Requests

Some GraphQL implementations allow queries to be sent using the HTTP GET method. This may expose sensitive data, as the data is included in the URL and can be stored in various locations, such as browser history, referrer headers, and proxies.

• Enabling Debugging: http://example.com/graphql?debug=1

Tools of the Trade

Several tools are mentioned to aid in information disclosure:

• InQL: Extracts schema information, assists in fuzzing and brute-forcing, and can be used to automate tasks from the command line.

• Clairvoyance: Uses field suggestions to reconstruct the schema when introspection is disabled.

• Burp Suite: Used to intercept traffic, capture queries, and observe application behavior.

• CeWL: Extracts keywords from the frontend HTML of an application, which can then be used in field-stuffing attacks.