Post #1882960

2026-04-30 03:33 UTC

The security guarantees we build our application and infrastructure upon—user isolation, filesystem permissions, process separation—are rendered moot when the kernel itself is compromised. The microkernel, Language-Based Security and capability folks have been screaming about this for the last three (more like 5 or 6 actually) decades.

Replies (2)

@lispi314@udongein.xyz 2026-04-30 03:46

Why did expert eyes, along with a plethora of automated analysis tools, miss this critical flaw for so long? Because so much of everything involved in that bug being possible is wrong that it just doesn't stand out to scrutiny. But generally, in-place modification should be avoided unless it's actually necessary. It's way too error prone to reason about without tools for help modeling it (the typesystem can be such a tool, in a number of languages). This scenario prompts a radical re-evaluation of current code review processes, security tool efficacy, and perhaps even the paradigms we use for kernel development. Yes, that thing people have been saying for around 50 years by now. To call UNIX "not up to the current state of the art" is to make a very generous understatement.

Open ##1882961
@kirtai@tech.lgbt 2026-04-30 16:20

@lispi314@udongein.xyz I can't help but remember that a very big part of the disagreement between the monolithic and microkernel people was "but muh performance" on the monolithic side.

Open ##2429460