Re: [arch-announce] The xz package has been backdoored
On Fri, 2024-03-29 at 18:55 +0000, Arch Linux: Recent news updates: David Runge wrote:
TL;DR: Upgrade your systems and container images **now**!
Thanks for sharing. Truly an astounding revelation. This is a very, very sophisticated tool-chain attack along the lines of Ken Thompson's famous compiler trust example [1] Arch has been a strong advocate for reproducible builds [2] which can be part of a defense strategy [2]. I note that our xz package is marked as good in this regard [3]. I wonder what more we can reasonably do in the near term. Since git gives us decent tools to check what changed etc., I would imagine that can provide a stronger base on which to check things than working with tarballs or tarballs alone. This may have gone largely un-noticed for so long as people are probably more likely to check the source than the tarball itself. In this case, it seems, it was a primary developer doing the naughty - but they chose to leave the git repo alone and only infect the tarball. Question: -------- Would it make sense, therefore, to switch builds, where possible, away from tar files and instead pull directly from git source (signed tags where possible as usual etc)? Of course a git repo can also carry infections - perhaps taht's a little less likely. Or is this not worth the trouble? Gene [1] https://wiki.c2.com/?TheKenThompsonHack [2] https://reproducible-builds.org/ https://wiki.archlinux.org/title/DeveloperWiki:ReproducibleBuilds https://bootstrappable.org/ [3] https://reproducible.archlinux.org/ -- Gene
On 3/30/24 12:34, Genes Lists wrote:
On Fri, 2024-03-29 at 18:55 +0000, Arch Linux: Recent news updates: David Runge wrote:
TL;DR: Upgrade your systems and container images **now**!
<snip> Question:
--------
Would it make sense, therefore, to switch builds, where possible, away from tar files and instead pull directly from git source (signed tags where possible as usual etc)? Of course a git repo can also carry infections - perhaps taht's a little less likely.
Or is this not worth the trouble?
I have public servers -- so was quite terrifying. However, the consensus was that Arch was never vulnerable given that the .m4 script is not used in the PKGBUILD and is limited to use in .deb or .rpm packaging. (that's to say the compromised test files are present, but not invoked to inject themselves into the library as part of the build) The lack of freak-out by Allan was the most comforting aspect. Long discussion, frustrating abundance of "opinions" and light on "concrete facts", but worth the read on just how Arch handles xz: https://gitlab.archlinux.org/archlinux/packaging/packages/xz/-/issues/2 -- David C. Rankin, J.D.,P.E.
participants (2)
-
David C. Rankin
-
Genes Lists