aikido-zen 0.1.0.alpha4

data/lib/aikido/zen/scanners/ssrf_scanner.rb

@@ -0,0 +1,251 @@
+# frozen_string_literal: true
+
+require_relative "ssrf/private_ip_checker"
+require_relative "ssrf/dns_lookups"
+
+module Aikido::Zen
+  module Scanners
+    class SSRFScanner
+      # Checks if an outbound HTTP request is to a hostname supplied from user
+      # input that resolves to a "dangerous" address. This is called from two
+      # different places:
+      #
+      # * HTTP library sinks, before we make a request. In these cases we can
+      #   detect very obvious attempts such as a request that attempts to access
+      #   localhost or an internal IP.
+      # * DNS lookup sinks, after we resolve a hostname. For HTTP requests that
+      #   are not obviously an attack, we let the DNS resolution happen, and
+      #   then check again, now knowing if the domain name provided actually
+      #   resolves to an internal IP or not.
+      #
+      # NOTE: Because not all DNS resolutions might be happening in the context
+      # of a protected HTTP request, the +request+ argument below *might* be nil
+      # and we can then skip this scan.
+      #
+      # @param request [Aikido::Zen::Scanners::SSRFScanner::Request, nil]
+      #   the ongoing outbound HTTP request.
+      # @param context [Aikido::Zen::Context]
+      # @param sink [Aikido::Zen::Sink] the Sink that is running the scan.
+      # @param operation [Symbol, String] name of the method being scanned.
+      #   Expects +sink.operation+ being set to get the full module/name combo.
+      #
+      # @return [Aikido::Zen::Attacks::SSRFAttack, nil] an Attack if any user
+      #   input is detected to be attempting SSRF, or +nil+ if not.
+      def self.call(request:, sink:, context:, operation:, **)
+        return if context.nil?
+        return if request.nil? # See NOTE above.
+
+        context["ssrf.redirects"] ||= RedirectChains.new
+
+        context.payloads.each do |payload|
+          scanner = new(request.uri, payload.value, context["ssrf.redirects"])
+          next unless scanner.attack?
+
+          attack = Attacks::SSRFAttack.new(
+            sink: sink,
+            request: request,
+            input: payload,
+            context: context,
+            operation: "#{sink.operation}.#{operation}"
+          )
+
+          return attack
+        end
+
+        nil
+      end
+
+      # Track the origin of a redirection so we know if an attacker is using
+      # redirect chains to mask their use of a (seemingly) safe domain.
+      #
+      # @param request [Aikido::Zen::Scanners::SSRFScanner::Request]
+      # @param response [Aikido::Zen::Scanners::SSRFScanner::Response]
+      # @param context [Aikido::Zen::Context]
+      #
+      # @return [void]
+      def self.track_redirects(request:, response:, context: Aikido::Zen.current_context)
+        return unless response.redirect?
+
+        context["ssrf.redirects"] ||= RedirectChains.new
+        context["ssrf.redirects"].add(
+          source: request.uri,
+          destination: response.redirect_to
+        )
+      end
+
+      # @api private
+      def initialize(request_uri, input, redirects)
+        @request_uri = request_uri
+        @input = input
+        @redirects = redirects
+      end
+
+      # @api private
+      def attack?
+        return false if @input.nil? || @input.to_s.empty?
+
+        # If the request is not aimed at an internal IP, we can ignore it. (It
+        # might still be an SSRF if defined strictly, but it's unlikely to be
+        # exfiltrating data from the app's servers, and the risk for false
+        # positives is too high.)
+        return false unless private_ip?(@request_uri.hostname)
+
+        origins_for_request
+          .product(uris_from_input)
+          .any? { |(conn_uri, candidate)| match?(conn_uri, candidate) }
+      end
+
+      # @!visibility private
+      def self.private_ip_checker
+        @private_ip_checker ||= SSRF::PrivateIPChecker.new
+      end
+
+      private
+
+      def match?(conn_uri, input_uri)
+        return false if conn_uri.hostname.nil? || conn_uri.hostname.empty?
+        return false if input_uri.hostname.nil? || input_uri.hostname.empty?
+
+        # The URI library will automatically set the port to the default port
+        # for the current scheme if not provided, which means we can't just
+        # check if the port is present, as it always will be.
+        is_port_relevant = input_uri.port != input_uri.default_port
+        return false if is_port_relevant && input_uri.port != conn_uri.port
+
+        conn_uri.hostname == input_uri.hostname
+      end
+
+      def private_ip?(hostname)
+        self.class.private_ip_checker.private?(hostname)
+      end
+
+      def origins_for_request
+        [@request_uri, @redirects.origin(@request_uri)].compact
+      end
+
+      # Maps the current user input into a Set of URIs we can check against:
+      #
+      # * The input itself, if it already looks like a URI.
+      # * The input prefixed with http://
+      # * The input prefixed with https://
+      #
+      # @return [Set<URI>]
+      def uris_from_input
+        input = @input.to_s
+
+        # If you build a URI manually and set the hostname to an IPv6 string,
+        # the URI library will be helpful to wrap it in brackets so it's a
+        # valid hostname. We should do the same for the input.
+        input = format("[%s]", input) if unescaped_ipv6?(input)
+
+        [input, "http://#{input}", "https://#{input}"]
+          .map { |candidate| as_uri(candidate) }
+          .compact
+          .uniq
+      end
+
+      def as_uri(string)
+        URI(string)
+      rescue URI::InvalidURIError
+        nil
+      end
+
+      # Check if the input is an IPv6 that is not surrounded by square brackets.
+      def unescaped_ipv6?(input)
+        (
+          IPAddr::RE_IPV6ADDRLIKE_FULL.match?(input) ||
+          IPAddr::RE_IPV6ADDRLIKE_COMPRESSED.match?(input)
+        ) && !(input.start_with?("[") && input.end_with?("]"))
+      end
+
+      # @api private
+      module Headers
+        # @param headers [Hash<String, Object>]
+        # @param header_normalizer [Proc{Object => String}]
+        def initialize(headers:, header_normalizer: :to_s.to_proc)
+          @headers = headers.to_h
+          @header_normalizer = header_normalizer
+          @normalized_headers = false
+        end
+
+        # @return [Hash<String, String>]
+        def headers
+          return @headers if @normalized_headers
+
+          @headers
+            .transform_keys!(&:downcase)
+            .transform_values!(&@header_normalizer)
+            .tap { @normalized_headers = true }
+        end
+      end
+
+      # @api private
+      class Request
+        include Headers
+
+        attr_reader :verb
+        attr_reader :uri
+
+        def initialize(verb:, uri:, **header_options)
+          super(**header_options)
+          @verb = verb.to_s.upcase
+          @uri = URI(uri)
+        end
+
+        def to_s
+          [@verb, @uri.to_s].join(" ").strip
+        end
+      end
+
+      # @api private
+      class Response
+        include Headers
+
+        attr_reader :status
+
+        def initialize(status:, **header_options)
+          super(**header_options)
+          @status = status.to_s
+        end
+
+        def redirect?
+          @status.start_with?("3") && headers["location"]
+        end
+
+        def redirect_to
+          URI(headers["location"]) if redirect?
+        rescue URI::BadURIError
+          nil
+        end
+      end
+
+      # @api private
+      class RedirectChains
+        def initialize
+          @redirects = {}
+        end
+
+        def add(source:, destination:)
+          @redirects[destination] = source
+          self
+        end
+
+        # Recursively looks for the original URI that triggered the current
+        # chain. If given a URI that was not the result of a redirect chain, it
+        # returns +nil+
+        #
+        # @param uri [URI]
+        # @return [URI, nil]
+        def origin(uri)
+          source = @redirects[uri]
+
+          if @redirects[source]
+            origin(source)
+          else
+            source
+          end
+        end
+      end
+    end
+  end
+end

data/lib/aikido/zen/scanners/stored_ssrf_scanner.rb

@@ -0,0 +1,43 @@
+# frozen_string_literal: true
+
+module Aikido::Zen
+  module Scanners
+    # Inspects the result of DNS lookups, to determine if we're being the target
+    # of a stored SSRF targeting IMDS addresses (169.254.169.254).
+    class StoredSSRFScanner
+      def self.call(hostname:, addresses:, operation:, sink:, context:, **opts)
+        offending_address = new(hostname, addresses).attack?
+        return if offending_address.nil?
+
+        Attacks::StoredSSRFAttack.new(
+          hostname: hostname,
+          address: offending_address,
+          sink: sink,
+          context: context,
+          operation: "#{sink.operation}.#{operation}"
+        )
+      end
+
+      def initialize(hostname, addresses, config: Aikido::Zen.config)
+        @hostname = hostname
+        @addresses = addresses
+        @config = config
+      end
+
+      # @return [String, nil] either the offending address, or +nil+ if no
+      #   address is deemed dangerous.
+      def attack?
+        return false if @config.imds_allowed_hosts.include?(@hostname)
+
+        @addresses.find do |candidate|
+          DANGEROUS_ADDRESSES.any? { |address| address === candidate }
+        end
+      end
+
+      DANGEROUS_ADDRESSES = [
+        IPAddr.new("169.254.169.254"),
+        IPAddr.new("fd00:ec2::254")
+      ]
+    end
+  end
+end

data/lib/aikido/zen/scanners.rb

@@ -0,0 +1,5 @@
+# frozen_string_literal: true
+
+require_relative "scanners/sql_injection_scanner"
+require_relative "scanners/stored_ssrf_scanner"
+require_relative "scanners/ssrf_scanner"

data/lib/aikido/zen/sink.rb

@@ -0,0 +1,108 @@
+# frozen_string_literal: true
+
+require_relative "scan"
+
+module Aikido::Zen
+  module Sinks
+    # @api internal
+    # @return [Hash<String, Sink>]
+    def self.registry
+      @registry ||= {}
+    end
+
+    # Primary interface to set up a sink with a list of given scanners.
+    #
+    # @param name [String] name of the library being patched. (This must
+    #   match the name of the gem, or we won't report that gem as
+    #   supported.)
+    # @param scanners [Array<#call>] a list of objects that respond to
+    #   #call with a Hash and return an Attack or nil.
+    # @param opts [Hash<Symbol, Object>] any other options to pass to
+    #   the Sink initializer.
+    #
+    # @return [void]
+    # @raise [ArgumentError] if a Sink with this name has already been
+    #   registered.
+    def self.add(name, scanners:, **opts)
+      raise ArgumentError, "Sink #{name} already registered" if registry.key?(name.to_s)
+      registry[name.to_s] = Sink.new(name.to_s, scanners: scanners, **opts)
+    end
+  end
+
+  # Sinks serve as the proxies between a given library that we protect
+  # (such as a database adapter that we patch to prevent SQL injections)
+  # and the reporting agent.
+  #
+  # When a library is patched to track and potentially block attacks, we
+  # rely on a sink to run any scans required, and report any attacks to
+  # our agent.
+  #
+  # @see ./sinks/trilogy.rb for a reference implementation.
+  class Sink
+    # @return [String] name of the patched library (e.g. "mysql2").
+    attr_reader :name
+
+    # @return [Array<#call>] list of registered scanners for this sink.
+    attr_reader :scanners
+
+    # @return [String] descriptor of the module / method being scanned
+    #   for attacks. This is fed into Attacks when instantiated. In
+    #   certain cases, some scanners allow you to specialize this by
+    #   using an +operation+ param of their own.
+    attr_reader :operation
+
+    DEFAULT_REPORTER = ->(scan) { Aikido::Zen.track_scan(scan) }
+
+    def initialize(name, scanners:, operation: name, reporter: DEFAULT_REPORTER)
+      raise ArgumentError, "scanners cannot be empty" if scanners.empty?
+
+      @name = name
+      @operation = operation
+      @scanners = scanners
+      @reporter = reporter
+    end
+
+    # Run the given arguments through all the registered scanners, until
+    # one of them returns an Attack or all return +nil+, and report the
+    # findings back to the Sink's +reporter+ to track statistics and
+    # potentially handle the +Attack+, if anything.
+    #
+    # This checks if runtime protection has been turned off for the current
+    # route first, and if so skips the scanning altogether, returning nil.
+    #
+    # @param scan_params [Hash] data to pass to all registered scanners.
+    # @option scan_params [Aikido::Zen::Context, nil] :context
+    #   The current Context, including the HTTP request being inspected, or
+    #   +nil+ if we're scanning outside of an HTTP request.
+    #
+    # @return [Aikido::Zen::Scan, nil] the result of the scan, or +nil+ if the
+    #   scan was skipped due to protection being disabled for the current route.
+    # @raise [Aikido::UnderAttackError] if an attack is detected and
+    #   blocking_mode is enabled.
+    def scan(context: Aikido::Zen.current_context, **scan_params)
+      return if context&.protection_disabled?
+
+      scan = Scan.new(sink: self, context: context)
+
+      scan.perform do
+        result = nil
+
+        scanners.each do |scanner|
+          result = scanner.call(sink: self, context: context, **scan_params)
+          break result if result
+        rescue Aikido::Zen::InternalsError => error
+          Aikido::Zen.config.logger.warn(error.message)
+          scan.track_error(error, scanner)
+        rescue => error
+          scan.track_error(error, scanner)
+        end
+
+        result
+      end
+
+      @reporter.call(scan)
+
+      scan
+    end
+  end
+end

data/lib/aikido/zen/sinks/async_http.rb

@@ -0,0 +1,63 @@
+# frozen_string_literal: true
+
+require_relative "../sink"
+require_relative "../outbound_connection_monitor"
+
+module Aikido::Zen
+  module Sinks
+    module Async
+      module HTTP
+        SINK = Sinks.add("async-http", scanners: [
+          Aikido::Zen::Scanners::SSRFScanner,
+          Aikido::Zen::OutboundConnectionMonitor
+        ])
+
+        module Extensions
+          def call(request)
+            uri = URI(format("%<scheme>s://%<authority>s%<path>s", {
+              scheme: request.scheme || scheme,
+              authority: request.authority || authority,
+              path: request.path
+            }))
+
+            wrapped_request = Aikido::Zen::Scanners::SSRFScanner::Request.new(
+              verb: request.method,
+              uri: uri,
+              headers: request.headers.to_h,
+              header_normalizer: ->(value) { Array(value).join(", ") }
+            )
+
+            # Store the request information so the DNS sinks can pick it up.
+            if (context = Aikido::Zen.current_context)
+              prev_request = context["ssrf.request"]
+              context["ssrf.request"] = wrapped_request
+            end
+
+            SINK.scan(
+              connection: Aikido::Zen::OutboundConnection.from_uri(uri),
+              request: wrapped_request,
+              operation: "request"
+            )
+
+            response = super
+
+            Aikido::Zen::Scanners::SSRFScanner.track_redirects(
+              request: wrapped_request,
+              response: Aikido::Zen::Scanners::SSRFScanner::Response.new(
+                status: response.status,
+                headers: response.headers.to_h,
+                header_normalizer: ->(value) { Array(value).join(", ") }
+              )
+            )
+
+            response
+          ensure
+            context["ssrf.request"] = prev_request if context
+          end
+        end
+      end
+    end
+  end
+end
+
+::Async::HTTP::Client.prepend(Aikido::Zen::Sinks::Async::HTTP::Extensions)

data/lib/aikido/zen/sinks/curb.rb

@@ -0,0 +1,89 @@
+# frozen_string_literal: true
+
+require_relative "../sink"
+require_relative "../outbound_connection_monitor"
+
+module Aikido::Zen
+  module Sinks
+    module Curl
+      SINK = Sinks.add("curb", scanners: [
+        Aikido::Zen::Scanners::SSRFScanner,
+        Aikido::Zen::OutboundConnectionMonitor
+      ])
+
+      module Extensions
+        def self.wrap_request(curl, url: curl.url)
+          Aikido::Zen::Scanners::SSRFScanner::Request.new(
+            verb: nil, # Curb hides this by directly setting an option in C
+            uri: URI(url),
+            headers: curl.headers
+          )
+        end
+
+        def self.wrap_response(curl)
+          # Curb made an… interesting choice by not parsing the response headers
+          # and forcing users to do this manually if they need to look at them.
+          _, *headers = curl.header_str.split(/[\r\n]+/).map(&:strip)
+          headers = headers.flat_map { |str| str.scan(/\A(\S+): (.+)\z/) }.to_h
+
+          if curl.url != curl.last_effective_url
+            status = 302 # We can't know what the original status was, but we just need a 3XX
+            headers["Location"] = curl.last_effective_url
+          else
+            status = curl.status.to_i
+          end
+
+          Aikido::Zen::Scanners::SSRFScanner::Response.new(status: status, headers: headers)
+        end
+
+        def perform
+          wrapped_request = Extensions.wrap_request(self)
+
+          # Store the request information so the DNS sinks can pick it up.
+          if (context = Aikido::Zen.current_context)
+            prev_request = context["ssrf.request"]
+            context["ssrf.request"] = wrapped_request
+          end
+
+          SINK.scan(
+            connection: Aikido::Zen::OutboundConnection.from_uri(URI(url)),
+            request: wrapped_request,
+            operation: "request"
+          )
+
+          response = super
+
+          Aikido::Zen::Scanners::SSRFScanner.track_redirects(
+            request: wrapped_request,
+            response: Extensions.wrap_response(self)
+          )
+
+          # When libcurl has follow_location set, it will handle redirections
+          # internally, and expose the "last_effective_url" as the URI that was
+          # last requested in the redirect chain.
+          #
+          # In this case, we can't actually stop the request from happening, but
+          # we can scan again (now that we know another request happened), to
+          # stop the response from being exposed to the user. This downgrades
+          # the SSRF into a blind SSRF, which is better than doing nothing.
+          if url != last_effective_url
+            last_effective_request = Extensions.wrap_request(self, url: last_effective_url)
+            context["ssrf.request"] = last_effective_request if context
+
+            SINK.scan(
+              connection: Aikido::Zen::OutboundConnection.from_uri(URI(last_effective_url)),
+              request: last_effective_request,
+              operation: "request"
+            )
+          end
+
+          response
+        ensure
+          context["ssrf.request"] = prev_request if context
+        end
+      end
+    end
+  end
+end
+
+::Curl::Easy.prepend(Aikido::Zen::Sinks::Curl::Extensions)

data/lib/aikido/zen/sinks/em_http.rb

@@ -0,0 +1,71 @@
+# frozen_string_literal: true
+
+require_relative "../sink"
+require_relative "../outbound_connection_monitor"
+
+module Aikido::Zen
+  module Sinks
+    module EventMachine
+      module HttpRequest
+        SINK = Sinks.add("em-http-request", scanners: [
+          Aikido::Zen::Scanners::SSRFScanner,
+          Aikido::Zen::OutboundConnectionMonitor
+        ])
+
+        module Extensions
+          def send_request(*)
+            wrapped_request = Aikido::Zen::Scanners::SSRFScanner::Request.new(
+              verb: req.method.to_s,
+              uri: URI(req.uri),
+              headers: req.headers
+            )
+
+            # Store the request information so the DNS sinks can pick it up.
+            context = Aikido::Zen.current_context
+            context["ssrf.request"] = wrapped_request if context
+
+            SINK.scan(
+              connection: Aikido::Zen::OutboundConnection.new(
+                host: req.host,
+                port: req.port
+              ),
+              request: wrapped_request,
+              operation: "request"
+            )
+
+            super
+          end
+        end
+
+        class Middleware
+          def response(client)
+            # Store the request information so the DNS sinks can pick it up.
+            context = Aikido::Zen.current_context
+            context["ssrf.request"] = nil if context
+
+            Aikido::Zen::Scanners::SSRFScanner.track_redirects(
+              request: Aikido::Zen::Scanners::SSRFScanner::Request.new(
+                verb: client.req.method,
+                uri: URI(client.req.uri),
+                headers: client.req.headers
+              ),
+              response: Aikido::Zen::Scanners::SSRFScanner::Response.new(
+                status: client.response_header.status,
+                headers: client.response_header.to_h
+              )
+            )
+          end
+        end
+      end
+    end
+  end
+end
+
+::EventMachine::HttpRequest
+  .use(Aikido::Zen::Sinks::EventMachine::HttpRequest::Middleware)
+
+# NOTE: We can't use middleware to intercept requests as we want to ensure any
+# modifications to the request from user-supplied middleware are already applied
+# before we scan the request.
+::EventMachine::HttpClient
+  .prepend(Aikido::Zen::Sinks::EventMachine::HttpRequest::Extensions)