Newshot

Debian APT to drop legacy features, citing advances in AI security research

Original text(via HN AI)

EOL notice for various APT repository configurations

  • Subject: EOL notice for various APT repository configurations
  • Date: Thu, 19 Nov 2026 14:55:53 +0100
  • Message-id: <[🔎] av7_w0YhAlMjzbjL@jak-t14-g3>

        Hey,

the recent vulnerability in APT's downloading code and the
ongoing advancements in AI have put a sharp eye on the
incredible complexity of APT's downloading code.

Much of this complexity exists in an effort to maintain support
for legacy repository formats that are not actually being used
or can be trivially switched away from.

With the advancements in AI security research, the risk to
users from that complexity now significantly outweighs the
benefits of retaining backward compatibility.

As such, I will be removing support for various features in unstable
starting next week (the cut-off for changes landing next week is
Tuesday, November 24).

While I hope to continue to provide security support for these
features in stable releases, we may not be able to. Please
ensure you migrate all remaining repositories ASAP. Security
updates in 2027 onward may remove features without prior
warning.

Features being removed
----------------------
* Repositories with detached Release.gpg signatures

    Please add an InRelease file if you still have not, otherwise
    they will only work as untrusted repositories.

* Repositories with only weak hashes (MD5, SHA1)

    These could be enabled using the options:

    "Acquire::AllowWeakRepositories" / "allow-weak"

    But they create some complexity in keeping track of which files
    are allowed to be weak, creating the potential for confusion.

    Instead, I plan to extend the hashes framework to allow marking
    a hash as trusted using APT::Hashes::<name>::Trusted, to provide
    limited backward compatibility support for inspecting historical
    repositories.

* Repositories without hashes / Release file

    We previously supported Release files that do not have hashes
    (as well as flat repositories without a Release file).

    This requires probing all possible file paths in the repository
    without knowing the size or hashes of those files,
    exposing us to infinite responses amongst other issues.

This leaves repositories in two varieties:

1. Signed, with an InRelease file
2. Unsigned, with a Release file

with all further requests having trusted hashes and sizes available.

This should allow us to significantly further streamline the downloading
code going forward.

Removals under evaluation
-------------------------
* "Insecure" repositories

    The options

        "Acquire::AllowInsecureRepositories" / "allow-insecure"
        "Acquire::AllowDowngradeToInsecureRepositories" / "allow-downgrade-to-insecure"

    The feature seems small but we need to ask whether it makes sense
    to retain this logic vs only the "trusted" flag. Considering all
    repositories trusted either by real trust or force is conceptually
    easier.

* Fallbacks

    In a bunch of places we can conditionally fallback to a different
    file if one route fails, such as

    - Packages.xz to Packages.gz if Packages.xz is missing or wrong
    - Pdiff -> Packages files if the diffs are not complete
    - Translations

    Like repositories without hashes, this introduces complex control
    flow and state management which could contain plenty of bugs
    (notably their classes transition through various states; it's
    pretty bad).

* Other features

    Other non-essential features such as storing index files compressed
    may be removed without prior notice if they do not affect functionality.

    We're not aware of any such features having a direct security impact
    at this point; however, compressed index files make it harder for us
    to verify the final result of the sandbox before committing it.

As we gain a better understanding of the control flow we may be able
to reintroduce some features if we understand they are safe.

Note on the sandbox
-------------------
APT's sandbox is incomplete, files are being verified inside of it,
attempts to add a final verification steps so far have failed as we
do not have hashes for all files and things like OpenPGP signatures
are hard to verify a second time as root.


--
debian developer - deb.li/jak | jak-linux.org - free software dev
ubuntu core developer                              i speak de, en
      

Attachment: signature.asc Description: PGP signature

  • Julian Andres Klode (on-list)
  • Julian Andres Klode (off-list)
  • Prev by Date: Perl 5.44 transition under way
  • Previous by thread: Perl 5.44 transition under way
  • Index(es): Date Thread

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by HN AI
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗