Close

POSIX is a curse

A project log for NPM - New Programming Model

NPM defines a programming model for new languages and "application processors", that emphasises safety, security and modularity.

Yann Guidon / YGDESYann Guidon / YGDES 05/30/2020 at 03:140 Comments

I'm sorry, dear Mr Stallman, but POSIX is defective by design and even though it was an amazing project 30+ years ago, it is now biting us E V E R Y - S I N G L E - D A Y. What was the very best in the mini- and micro-computers era is now totally unadapted and no hardware tweak or supervisory tool could ever solve the fundamental problems. And all the new language (even the "safe" ones) perpetuate the misconceptions that you could make an unsafe system safe by using a safe language. You should already know that the weakest link breaks first, right ?

POSIX was never meant to be safe by design. It is inherently tied to the standard libraries, which are inherently designed for and by the C language, which is well known to not be safe...

This co-evolution creates a macabre dance and there is no solution. The whole thing is rotten to the core but held in place because everybody uses it !


Of course the "new programming model" requires an adapted platform to run smoothly. However this is not a hard requirement and the system can be emulated at some level or another, to help bootstrap the transition.

Let's simply start with the most common source of vulnerabilities : buffer overflows. These should NEVER happen. Yet they do. All the time.

The i286 introduced the BOUND opcode in 1982... Where is it used ???

Oh yes, segmented and paginated memory is the solution...

But wait, flat space is sexier and segment registers are PITA.

Let's make fast SW and throw all the compiler warnings through the window...

IF you use a language that implements boundary checks. Some are more stringent than others... but speed always ends up being the excuse for indulging in the safety-sin again and again, and processors are tired and tired of checking the same pointers for nothing. Because it can never occur right ?

iAPX432 wanted to solve this with a micro-managed system with complex properties and it was slow. I understand why... but it would have been safer, right ?


But wait !

We're in 2020 and in hindsight (hahaha) 70% of all security bugs are memory safety issues and half of them "are use-after-free vulnerabilities, a type of security issue that arises from incorrect management of memory pointers"

So you use a block, free it and ... it can still be read ? Why would that be possible ?

Oh wait : we are bound to use the standard libs, even if our programming language has its own system.


"There will always be bugs" is an old refrain, and it is not false.

It is however a lousy pretext to accept preventable bugs and incapable platforms.

I see a kind of bugs that is not completely dependent on the tools and the programming languages : check the goddamned access rights. Check which process requests what service and in which context.

This is not the perfect solution but any system call should be easily traced by the callee to the root of the process tree. The whole history and tree of who calls what and with which access rights must be visible by the whole computer. So it's another hardware-assisted feature that must be etched in silicon (or whatever).


I am not blaming anyone here, and I know extensive efforts are being made, for example with Rust. However this is not a complete solution, as long as the whole computer system co-evolves around old premises dating from the 60s, when things were so simple and laid back.

Discussions