Whitelist classes and symbols that are in Gem spec YAML

This patch adds a method for loading YAML specs from a gem and
whitelists classes and symbols that are allowed in the spec.  Then it
changes calls to YAML.load to call the whitelisted "safe" loader
instead.

[CVE-2017-0903]

Author: tenderlove

Manifest.txt

@@ -387,6 +387,7 @@ lib/rubygems/resolver/specification.rb
 lib/rubygems/resolver/stats.rb
 lib/rubygems/resolver/vendor_set.rb
 lib/rubygems/resolver/vendor_specification.rb
+lib/rubygems/safe_yaml.rb
 lib/rubygems/security.rb
 lib/rubygems/security/policies.rb
 lib/rubygems/security/policy.rb

lib/rubygems.rb

@@ -675,7 +675,7 @@ def self.load_yaml
 
     unless test_syck
       begin
-        gem 'psych', '>= 1.2.1'
+        gem 'psych', '>= 2.0.0'
       rescue Gem::LoadError
         # It's OK if the user does not have the psych gem installed.  We will
         # attempt to require the stdlib version
@@ -699,6 +699,7 @@ def self.load_yaml
     end
 
     require 'yaml'
+    require 'rubygems/safe_yaml'
 
     # If we're supposed to be using syck, then we may have to force
     # activate it via the YAML::ENGINE API.

lib/rubygems/config_file.rb

@@ -345,7 +345,7 @@ def load_file(filename)
     return {} unless filename and File.exist? filename
 
     begin
-      content = YAML.load(File.read(filename))
+      content = Gem::SafeYAML.load(File.read(filename))
       unless content.kind_of? Hash
         warn "Failed to load #{filename} because it doesn't contain valid YAML hash"
         return {}

lib/rubygems/package.rb

@@ -468,7 +468,7 @@ def read_checksums gem
 
     @checksums = gem.seek 'checksums.yaml.gz' do |entry|
       Zlib::GzipReader.wrap entry do |gz_io|
-        YAML.load gz_io.read
+        Gem::SafeYAML.safe_load gz_io.read
       end
     end
   end

lib/rubygems/package/old.rb

@@ -101,7 +101,7 @@ def file_list io # :nodoc:
       header << line
     end
 
-    YAML.load header
+    Gem::SafeYAML.safe_load header
   end
 
   ##

lib/rubygems/safe_yaml.rb

@@ -0,0 +1,48 @@
+module Gem
+
+  ###
+  # This module is used for safely loading YAML specs from a gem.  The
+  # `safe_load` method defined on this module is specifically designed for
+  # loading Gem specifications.  For loading other YAML safely, please see
+  # Psych.safe_load
+
+  module SafeYAML
+    WHITELISTED_CLASSES = %w(
+      Symbol
+      Time
+      Date
+      Gem::Dependency
+      Gem::Platform
+      Gem::Requirement
+      Gem::Specification
+      Gem::Version
+      Gem::Version::Requirement
+      YAML::Syck::DefaultKey
+      Syck::DefaultKey
+    )
+
+    WHITELISTED_SYMBOLS = %w(
+      development
+      runtime
+    )
+
+    if ::YAML.respond_to? :safe_load
+      def self.safe_load input
+        ::YAML.safe_load(input, WHITELISTED_CLASSES, WHITELISTED_SYMBOLS, true)
+      end
+
+      def self.load input
+        ::YAML.safe_load(input, [::Symbol])
+      end
+    else
+      warn "YAML safe loading is not available. Please upgrade psych to a version that supports safe loading (>= 2.0)."
+      def self.safe_load input, *args
+        ::YAML.load input
+      end
+
+      def self.load input
+        ::YAML.load input
+      end
+    end
+  end
+end

lib/rubygems/specification.rb

@@ -1101,7 +1101,7 @@ def self.from_yaml(input)
     Gem.load_yaml
 
     input = normalize_yaml_input input
-    spec = YAML.load input
+    spec = Gem::SafeYAML.safe_load input
 
     if spec && spec.class == FalseClass then
       raise Gem::EndOfYAMLException