Syft incorrectly identifying jruby jar files

[joshbressers]

For the purposes of this report, I used this container
docker.elastic.co/logstash/logstash:8.13.4

We are seeing java findings show up that are part of jruby packages. These findings are very broken as they're not really jar files, they are part of the jruby package.

For example

bress@anchore ➜ syft convert /tmp/logstash.json | grep nokogiri
[0000]  WARN convert is an experimental feature, run `syft convert -h` for help
nokogiri                                                                                   java-archive
nokogiri                                        1.16.4                                     gem

That turns up the nokogiri gem, and a nokogiri java-archive which isn't real.

The nokogiri gemspec can be found here

bress@anchore ➜  docker run --rm  docker.elastic.co/logstash/logstash:8.13.4 cat /usr/share/logstash/vendor/bundle/jruby/3.1.0/specifications/nokogiri-1.16.4-java.gemspec | head
# -*- encoding: utf-8 -*-
# stub: nokogiri 1.16.4 java liblib/nokogiri/jruby

Gem::Specification.new do |s|
  s.name = "nokogiri".freeze
  s.version = "1.16.4"
  s.platform = "java".freeze

  s.required_rubygems_version = Gem::Requirement.new(">= 0".freeze) if s.respond_to? :required_rubygems_version=
  s.metadata = { "bug_tracker_uri" => "https://github.com/sparklemotion/nokogiri/issues", "changelog_uri" => "https://nokogiri.org/CHANGELOG.html", "documentation_uri" => "https://nokogiri.org/rdoc/index.html", "homepage_uri" => "https://nokogiri.org", "rubygems_mfa_required" => "true", "source_code_uri" => "https://github.com/sparklemotion/nokogiri" } if s.respond_to? :metadata=

The nokogiri.jar is located at
/usr/share/logstash/vendor/bundle/jruby/3.1.0/gems/nokogiri-1.16.4-java/lib/nokogiri/nokogiri.jar

Here are the details that are turned up for that jar file

    {
      "id": "f64e536bd221a6de",
      "name": "nokogiri",
      "version": "",
      "type": "java-archive",
      "foundBy": "java-archive-cataloger",
      "locations": [
        {
          "path": "/usr/share/logstash/vendor/bundle/jruby/3.1.0/gems/nokogiri-1.16.4-java/lib/nokogiri/nokogiri.jar",
          "layerID": "sha256:dd49a5e5f509de1b29ebdeaa3e11ea6251e6c741157c2bc0c3a9228827bc80c6",
          "accessPath": "/usr/share/logstash/vendor/bundle/jruby/3.1.0/gems/nokogiri-1.16.4-java/lib/nokogiri/nokogiri.jar",
          "annotations": {
            "evidence": "primary"
          }
        }
      ],
      "licenses": [],
      "language": "java",
      "cpes": [
        {
          "cpe": "cpe:2.3:a:nokogiri:nokogiri:*:*:*:*:*:*:*:*",
          "source": "syft-generated"
        }
      ],
      "purl": "pkg:maven/nokogiri/nokogiri",
      "metadataType": "java-archive",
      "metadata": {
        "virtualPath": "/usr/share/logstash/vendor/bundle/jruby/3.1.0/gems/nokogiri-1.16.4-java/lib/nokogiri/nokogiri.jar",
        "manifest": {
          "main": [
            {
              "key": "Manifest-Version",
              "value": "1.0"
            },
            {
              "key": "Created-By",
              "value": "11.0.20.1 (Ubuntu)"
            }
          ]
        },
        "digest": [
          {
            "algorithm": "sha1",
            "value": "bcc6ebb7ac131f150bdfa54cc66b8fd406695767"
          }
        ]
      }
    },

I wouldn't expect such a jar to show up in the results

[luhring]

Hey @joshbressers, I'm curious to learn more... I'm seeing an issue with scans of jruby as well, and I'm curious if my issue is related to this one. In my case I'm seeing inconsistent results for CPEs and license data.

You mention a couple times that the JAR files "aren't real", could you say more? What do you mean by that exactly? While these files functions as gems, I had been thinking that they are simultaneously valid Java archives (in Syft terms).

[willmurphyscode]

@joshbressers how should we handle the jruby version of native extensions? nokogiri as a regular Ruby gem needs a bunch of native code built to work. I'm assuming that on jruby, this native code is written in Java not C. Maybe it's correct to simply omit the jar? Maybe we should track the jar as a component of the nokogiri gem in this case? I've added needs-discussion so we can talk about it.