Security: prevent XXE exploits

config/initializers/sax_machine.rb

@@ -0,0 +1,21 @@
+# frozen_string_literal: true
+
+# sax-machine hardcodes `ctx.replace_entities = true` in its Nokogiri handler,
+# so a hostile feed can declare an external entity and have us resolve it while
+# parsing (XXE): `file://` reads local files and `http://` reaches private
+# addresses, both outside the SafeFetch guard that protects the fetch itself.
+#
+# Turning substitution off only affects entities that point at a SYSTEM
+# resource. Predefined (`&`), numeric (`é`) and internally declared
+# entities still resolve, so feeds that declare their own HTML entities in a
+# DOCTYPE keep working.
+module SAXMachine::DisableExternalEntities
+  def sax_parse(xml_input)
+    parser = Nokogiri::XML::SAX::Parser.new(self)
+    parser.parse(xml_input) { |ctx| ctx.replace_entities = false }
+  end
+end
+
+if SAXMachine.handler == :nokogiri
+  SAXMachine::SAXNokogiriHandler.prepend(SAXMachine::DisableExternalEntities)
+end

spec/utils/feedjira_spec.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+# Guards config/initializers/sax_machine.rb. These assert parser behaviour
+# rather than the patch itself, so they still fail if a sax-machine upgrade
+# stops the prepend from applying.
+RSpec.describe Feedjira do
+  def secret_file
+    Tempfile.new("xxe").tap do |file|
+      file.write("TOP-SECRET-CONTENTS")
+      file.close
+    end
+  end
+
+  def parse_feed(doctype, title)
+    described_class.parse(<<~XML)
+      <?xml version="1.0"?>
+      #{doctype}
+      <rss version="2.0"><channel>
+        <title>#{title}</title>
+        <link>http://example.com</link>
+      </channel></rss>
+    XML
+  end
+
+  it "does not resolve external entities pointing at local files" do
+    secret = secret_file
+    doctype = %(<!DOCTYPE rss [<!ENTITY xxe SYSTEM "file://#{secret.path}">]>)
+
+    expect(parse_feed(doctype, "&xxe;").title.to_s)
+      .not_to include("TOP-SECRET-CONTENTS")
+  end
+
+  it "still resolves predefined and numeric entities" do
+    expect(parse_feed("", "Tom &amp; Jerry caf&#233;").title)
+      .to eq("Tom & Jerry café")
+  end
+
+  it "still resolves entities declared inside the document" do
+    doctype = %(<!DOCTYPE rss [<!ENTITY nbsp "&#160;">]>)
+
+    expect(parse_feed(doctype, "Feed&nbsp;Title").title)
+      .to eq("Feed\u00A0Title")
+  end
+end