#+title: Software Repositories #+date: 2026-05-25 Mon #+author: W. Kosior #+email: wkosior@agh.edu.pl * Do you remember… …the days of searching online for an installer for some piece of software? * Common Types of Software Distribution Channels - OS distro repository - e.g., repositories of Debian, RedHat, BSD Ports - special cases: MSYS2, MacPorts - third-party upstream → packages - LTS / Bleeding Edge - main threats: - vulnerable packages not fixed in time - human choosing another, less secure channel :( * Common Types of Software Distribution Channels, Cont. - OS distro repository - specialized software repository - aka programming language-specific software repository - some being cross-language - e.g., PyPI, Maven, npm Registry - programming libraries - main threats: - deliberate malware uploads - dependency constraints blocking security updates * Common Types of Software Distribution Channels, Cont… - OS distro repository - specialized software repository - software store - e.g., Microsoft Store, Google Play, F-Droid - applications - rarely the builders (except F-Droid) - main threats: - bundled dependencies not updated - deliberate malware uploads * Common Types of Software Distribution Channels, Cont… - OS distro repository - specialized software repository - software store - container image repository - e.g., DockerHub - special case: FlatHub (a software store at the same time) - main threats: - bundled dependencies not updates - deliberate malware uploads * *Typical* Characteristics of Software Distribution Channels | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | * *Typical* Characteristics of Software Distribution Channels, Cont. | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | * *Typical* Characteristics of Software Distribution Channels, Cont… | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | | source- or binary-based | varies | varies | binary | * *Typical* Characteristics of Software Distribution Channels, Cont… | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | | source- or binary-based | varies | varies | binary | | anyone can publish | no | yes | yes | | author = uploader | rarely | usually | usually | * *Typical* Characteristics of Software Distribution Channels, Cont… | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | | source- or binary-based | varies | varies | binary | | anyone can publish | no | yes | yes | | author = uploader | rarely | usually | usually | | packages signed by | repo | repo | varies | * *Typical* Characteristics of Software Distribution Channels, Cont… | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | | source- or binary-based | varies | varies | binary | | anyone can publish | no | yes | yes | | author = uploader | rarely | usually | usually | | packages signed by | repo | repo | varies | | pay to publish | no | no | sometimes | | pay to download | no | no | sometimes | * *Typical* Characteristics of Software Distribution Channels, Cont… | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | | source- or binary-based | varies | varies | binary | | anyone can publish | no | yes | yes | | author = uploader | rarely | usually | usually | | packages signed by | repo | repo | varies | | pay to publish | no | no | sometimes | | pay to download | no | no | sometimes | | security by patching (lts) | varies | sporadically | no | | security by updating (be) | varies | yes | yes | * *Typical* Characteristics of Software Distribution Channels | | OS distro repo | specialized repo | "store" | |----------------------------+​----------------+​--------------------+--------------------| | deps are separate packages | yes | yes | sometimes | | package counts | <= several 10k | <= several million | <= several million | | source- or binary-based | varies | varies | binary | | anyone can publish | no | yes | yes | | author = uploader | rarely | usually | usually | | packages signed by | repo | repo | varies | | pay to publish | no | no | sometimes | | pay to download | no | no | sometimes | | security by patching (lts) | varies | sporadically | no | | security by updating (be) | varies | yes | yes | | targets | users&devs | mostly devs | mostly users | * LTS vs Bleeding Edge - always get newest versions of software - "rolling release", aka "bleeding edge" - Arch, Fedora Rawhide, OpenSUSE Tumbleween, Debian Unstable, etc. - vulnerabilities → update to newer version * LTS vs Bleeding Edge, Cont. - always get newest versions of software - "rolling release", aka "bleeding edge" - Arch, Fedora Rawhide, OpenSUSE Tumbleween, Debian Unstable, etc. - vulnerabilities → update to newer version - what about stability? - snapshot current software versions - use snapshotted version for the next X years - exceptions - "stable release", aka "LTS" - Debian, openSUSE Leap, Fedora - vulnerabilities → patch the old version - or use LTS updates when offered by upstream * LTS in Debian - Debian Unstable — bleeding edge versions - Debian Testing - snapshot of Debian Unstable - kept for a few months to stabilize - Debian Stable - made from Debian Testing - oldstable - made from Stable once a new Stable kicks in - still receives security fixes for some time * Patch or Update? - updating - less work - the only viable option for, e.g., Mozilla browsers - patching - number of vulnerabilities *only decreases* :) * XZ Backdoor - Backdoor targeting SSH daemons - Debian, Ubuntu, Fedora, et al - through XZ compression library - backdoored XZ source release tarball - built by distros - detected in Debian Unstable in 2024 - before getting to Stable - already included in, e.g., openSUSE Tumbleweed - https://i.kym-cdn.com/photos/images/original/002/785/376/4db.png * From Source to Installed software - security guidance (OWASP et al) → SBoM - far from enough! - just what we *think* is in our system - https://slsa.dev/spec/v1.2/threats-overview - decent categorization ;) - distro software → securable (with some shortcomings) - specialized repo software → hard to secure deps - container images (DockerHub) → hard to secure deps * The Bazaar Model - "The Cathedral and the Bazaar" - essay by Eric S. Raymond - 1997 - two free/libre software development models - "given enough eyeballs, all bugs are shallow" - blobs in source :( - distro with guidelines → - → higher hygiene → - → smaller attack surface * Who Builds It? - author? - typical npm package → hundreds dependencies - (including recursive and dev deps) - hundreds of developer PCs - how hard is it to compromise at least one? - repo? - required distro resources - rebuildability → required higher packaging hygiene - a good thing, actually - "trusted" third party? - provenance attestation - Github Actions - GitLab CI/CD - lock-in :( - end machine (source-based distribution)? * Builds by Repo — nuances - Debian — binary uploads used sporadically (e.g., backports) - F-Droid — developer signature or repo signature? * Build Machine — SPOF - SolarWinds (Sunburst) backdoor - reproducible builds (future lecture) * Hidden Dependencies - build container security - what GitHub Actions' Ubuntu image had inside at build time? - downloaded deps & other files - we downloaded a web browser as part of automated tests - was it not compromised? - remedies - verification with a hash? - mandate network-less builds? * VCS vs Release Tarballs - tarballs: traditional form - scripts pre-generated by maintainer (Autotools) - typically PGP-signed - modified tarballs - e.g., Debian FSDG compliance - version control system - commits can be PGP-signed - does not include =configure= from Autoconf, etc. - may not be rebuildable - signed commits/tags possible - XZ backdoor - fewer eyeballs look into tarballs - *please build from VCS* 🥺 * Repository → Machine - signed package lists - offline search - not archived in, e.g., Debian - downgrade attack possibility? - online querying - specialized (aka language-specific) repos - repo public key used - distributon major version update? - subsequent LTS public keys distributed in old * Sample Package Formats — Debian (=control= file) #+begin_src Source: python-immutables Maintainer: John Smith Section: devel Priority: optional Standards-Version: 4.3.0 Build-Depends: debhelper (>= 11), dh-python, python3-all-dev (>= 3.6), python3-setuptools (>= 45), # also add test dependencies python3-pytest, python3-pytest-flake8, python3-mypy Package: python3-immutables Architecture: all Depends: ${python3:Depends}, ${misc:Depends} X-Python3-Version: >= 3.6 Description: an immutable mapping type for Python This library provides an immutable alternative to Python's native 'dict'. The data structure used if Hash Array Mapped Trie (HAMT) which is also utilized by some functional languages. #+end_src * Sample Package Formats — Debian (=rules= file) #+begin_src #! /usr/bin/make -f # Immutables debian/rules file #export DH_VERBOSE = 1 export PYBUILD_NAME = immutables export PYBUILD_TEST_PYTEST = 1 %: MYPYPATH=$$(pwd) dh $@ --with python3 --buildsystem=pybuild #+end_src * Extra: XZ Backdoor Presentation - https://git.koszko.org/AGH-xz-backdoor-presentation/