Improvement of OpenSSF Scorecard Score

[harshitasao]

Describe the feature

Is this a bug report or feature request?

• Feature Request

What should the feature do:
Hi, I'm Harshita. I’m working with CNCF and the Google Open Source Security Team for the GSoC 2024 term. We are collaborating to enhance security practices across various CNCF projects. The goal is to improve security for all CNCF projects by both using OpenSSF Scorecards and implementing its security improvements.

After the opening of the scorecard action addition PR, I'm here to increase the final score by going over each check. I've listed all of the checks where work needs to be done, in order of its criticality. I plan to submit each PR for each fix. Please let me know what you think and for which ones a PR is welcome that I will submit it ASAP.

Current Score: 7.3

Scorecard report: https://scorecard.dev/viewer/?uri=github.com/knative/serving

Here's a few checks we can work on to improve the project's security posture [ in order of its criticality ]:

• Token-Permissions: Score = 0

  • The issue here is that many workflows doesn’t have a top-level read-only permissions block like present in the other workflows. Scorecard is quite severe in this check: a single workflow without top-level permissions gets a 0/10 for the check.

• Vulnerabilities: Score = 7

  • After running the osv scanner locally, a significant amount of vulnerabilities in Go dependencies were found that need to be fixed.
  • May need to add an osv-scanner.toml to mark some of these as not impacting/ignored.
  • Open vulnerabilities are easily exploited by attackers and should be fixed as soon as possible.

• Signed-Releases: Score = 8

  • SLSA provenance needs to be added, which can be done using this, increasing the overall score to 10.

• Fuzzing: Score = 0

  • Integrating the project with OSS-Fuzz by following the instructions here. The most difficult one on the list, maintainers help, is highly appreciated. For example, helping in identifying the components where fuzz testing will be added.

• Pinned-Dependencies: Score = 3

  • Github actions and go commands are not pinned by hash, which is resulting in a low score. But pinning go commands would introduce the risk of running outdated versions.
  • Pinning dependencies to a specific hash rather than allowing mutable versions or ranges of versions improves supply chain security.
  • PR: fix: fixed the pinned dependencies issue #15475

/cc @joycebrum @diogoteles08 @pnacht @nate-double-u

[harshitasao]

cc @davidhadas

[davidhadas]

Welcome @harshitasao

As for Fuzzing - see https://knative.dev/blog/events/fuzzing-audit-2023/
See also https://knative.dev/blog/events/security-audit-2023/

cc @evankanderson

[ReToCode]

Hm, regarding #15475 I'm not sure I fully agree with:

Github actions and go commands are not pinned by hash, which is resulting in a low score. But pinning go commands would introduce the risk of running outdated versions.
Pinning dependencies to a specific hash rather than allowing mutable versions or ranges of versions improves supply chain security.

Currently, we get every patch of lets say v4 of an action. Sure we have do work to get to v5, but with having fixed hashes without a mechanism to patch those automatically, we would miss 4.1, 4.2 (just an example) and so on.

Opinions @dprotaso @skonto ?

[pnacht]

Hey @ReToCode. I see Knative Serving already uses Dependabot to bump Actions.

Dependabot also works with hash-pinned Actions, so it will simply send you PRs whenever an Action releases a new version. It'll even update the "version comment" next to each hash.

If you are concerned about receiving more such PRs (updating from minor versions you get "for free" by using @v4), you can modify your dependabot.yml config file to use "grouped updates", where you'll receive a single PR updating all Actions at once (example PR):

  - package-ecosystem: "github-actions"
    directory: "/"
    schedule:
      interval: "monthly"
    groups:
      github-actions:
        patterns:
          - "*"

[ReToCode]

@pnacht I see. I still don't really follow the initial argumentation. Hashes are not more secure nor more up to date than versions (assuming dependabot is used to bump both). Furthermore, preventing supply chain attacks would require us to review the actual changes between versions of actions, otherwise how can we make sure a new version/hash is ok?

@cardil WDYT about the grouped updates? I think that could help to prevent even more automated, noisy PRs to all repos.

[pnacht]

Hey @ReToCode. Indeed, the security of hash-pinning + Dependabot PRs comes from the "lag" between the new version being released and the PR being sent. This is actually why I suggest using monthly updates, not weekly: this will mean that a new version will exist in the wild for ~2 weeks (on average) before you receive the PR itself. During this period, the Actions' maintainers and other users might discover problems with the new version and release other fixes, such that Knative Service won't ever be exposed to the problematic release.

I also suggest that you enable Dependabot Security Updates if you haven't already. With this enabled, Dependabot will send "out of season" PRs to immediately update dependencies with known vulnerabilities.

Therefore, hash-pinning + monthly updates you get some "bake time" before being exposed to new (and untested) versions of dependencies, and "out of season" security updates help you quickly get updated to new versions in case one is eventually found to be dangerous.