Skip to content

Latest commit

 

History

History
129 lines (91 loc) · 5.5 KB

summary.md

File metadata and controls

129 lines (91 loc) · 5.5 KB

Survey Summary and Takeaways

This document summarizes the survey results and gives some insight into some takeaways from each category of the survey.

In total, we had package manager maintainers (and/or reputable sources) from 11 different package manager ecosystems responded to the survey. Including:

  • PyPI (Python)
  • Gradle (Java)
  • Basalt (Bash)
  • Opam (OCaml)
  • RubyGems (Ruby)
  • npm (javascript)
  • Pkg.go.dev (Go)
  • Cargo (Rust)
  • Pub (Dart)
  • Apache Maven (Java)
  • Maven Central (Java)

Integrity

Most support is still around developer owned keys for signing (with PGP being the most supported signature type), with some a couple ecosystems which do not support signing yet - but there is interest/plans.

There are no package managers that support keyless signing yet.

Java, OCaml, Go, Ruby and Rust support reproducible builds with Maven Central making it mandatory.

image

Typos & dependency confusion

45.5% of package managers do not require and do not plan on having DNS verification proof when reserving/exercising a domain/namespace or domain name attribute of a package. This could lead to impersonation of organizations/entities. A method to reduce the cost of protecting against this threat or education of the seriousness of this threat is needed.

General typos and dependency confusion is observed and handled in a myriad of different ways across ecosystems. It would be helpful to provide guidance on a list of methods and evaluation of their effectiveness.

image

Handling compromise and malice

Most ecosystems don't see the responsibility of finding malware in the packages as their responsibility. However, they provide the ability to flag and handle malware in releases when it happens.

Rust, Ruby and Gradle have no plans on ensuring that install and build of code doesn't allow execution of arbitrary code on the host in an unprotected way. This is an area that requires further investigation into why that is not a priority and look into a way to be able to have isolation and provenance of builds without this property.

image

Protecting the ecosystem

Most ecosystems are not currently performing any security testing against code/infrastructure. However, all of them are interested (7) or have it planned (2). These ecosystems could be funded for security testing.

All ecosystems have a process in place for vulnerability disclosure or are interested in doing so.

Only 1 ecosystem (Go) has additional supply chain security guarantees for the binaries/tooling that it produces. This is an area for improvement overall.

image

Authentication and credential management

Generally there is interest in protecting token security, which is positive feedback. There are 2 ecosystems that currently have no plans on requiring MFA. Further investigation on why would be helpful.

Policy

With policy, there is a mix of implementations that help provide ability to create policy on keys/hashes as well as checking for known vulnerabilities. A significant number of ecosystems (4-5) currently do not but are interested.

For policies that take into account 3rd party audits and attestations, only 1 ecosystem provided functionality. The rest are mostly interested (8) and planned (1). This could be somewhere that a supply chain knowledge graph project and public instance can be leveraged to lower the cost to do so.

image

SBOMs

7 of 11 ecosystems do not currently have support for generating SBOM but are interested (5) or planned (2). Additional tooling support is required and could be a place where the OpenSSF SBOMs everywhere workstream 9 can be helpful.

Most SBOM formats are around SPDX and CycloneDX, with the exception of Go that provides an intermediary format. This is a positive indicator of convergence around the two SBOM standards.

image

Identify / produce good dependencies

In general most ecosystems allow pinned dependencies that are immutably verifiable. However, there are 3 ecosystems that do not have this yet. Further investigation on why would be helpful.

Only 27.3% of ecosystems have fuzzing services and utilities available, 45.4% have no plans on implementing and the last 27.3% are interested in exploring. This could be an avenue for education of the value of fuzzing within the different ecosystems.

There is a surprising lack of SAST/DAST services available for ecosystems to detect bad code patterns. Currently, 5 ecosystems of the 11 are without the capabilities. Funding of tooling could be helpful here.

There is a high demand (8) for services and tools for developers to help keep their code and packages in good shape. E.g. like dependabot for the package ecosystem. This would be a gap that could be filled by additional tooling.

Across the board, there are no ways to score dependencies based on their hygiene or security posture. Further education on scorecards, or evaluation of how scorecards can be made more useful to package ecosystems will be useful.