I am working on a project that will be like a whole environment
bootstrapped from the smallest possible beginning.
I thought a hex-to-binary converter like hex0 was too much like
machine code, and I wanted everything to be source code, so I
chose to instead start with a super simple interpreted language
that is described in english and the user is supposed to write
an interpreter for it themselves in any way they want. This
way I connect the human to the computer using the language of
the human which I thought felt better and easier to understand.
This super simple interpreted language then builds a super
simple RISC-V assembler. Here I came up with a nice technique
where the functions in the assembler are called through
pointers, which makes them replacable while the asesmbler is
running. So as soon as the assembler has compiled a better
version of one of its own functions, that function is then
replaced with the better version while the asesmbler is running
and it continues to compile the rest of the code which can then
use the new features. So it evolves without having to restart.
I still have a small fee available from NLNet for the implementation of a RISC-V back-end for a compiler (more a transpiler) from a small stack language to assembly. I have already started the effort, but at this moment not the time to work on it. I am available for support. Back-ends to x86, x86_64 and arm64 are available and working. This is part of the MES-replacement project [1]. Let me know if you are interested (see email at the bottom of the website mentioned in my profile here or on GitHub.)
> The hex0 program provides a way to turn a string of hexadecimal text into a binary with those bytes.
I'll use this to ask for an explanation about the fundamental idea: How is hex text better than a binary blob? Both need detailed knowledge to understand; both need a tool to display (reading ASCII or binary), either of which could be compromised in a "trusting trust" sense.
It seems to me that the actually important aspects of the initial "seed" are its size (larger is harder to verify) and that the language used (whether it is Lisp, binary machine code or whatever) has rigorously defined semantics.
(Orthogonal to that, you'd want to store everything, seed and the rest, on a medium that cannot hide anything from you, and use that as the source. But that applies to all approaches.)
For one plain text is a bit more convenient to view, edit, and talk about on line.
And as you move up towards more complex capabilities, you can start adding simple features like comments and basic text macros to give a name to your common opcodes
The bootstrap aims to use a progressively more capable language at the same time that the size of the next object to review increases. Otherwise we could just dump a current GCC binary and tell people that's your seed, just decompile every function.
Opcode mnemonics is something that a tool can show when you inspect a binary. You will need to go up the ladder a lot before the source code is more expressive than a disassembly. Comments are a good point though, but since they do no affect the semantics, they could be provided as a separate file that does not have to be reproducible (since compromised comments are at worst misleading or confusing).
The GCC example is valid, but can also be explained by its size.
I like the idea of bootstrapping in a Lisp-like language since it is extremely expressive for its simplicity. But then you need an interpreter for that, which must be trusted. I've been nerd-sniped by the whole thing and I'm thinking about an extremely simple virtual machine for the seed, with the actual seed code being a binary for that VM that implements a Lisp. Then an audit of the seed becomes disassembling the VM code for the Lisp interpreter, which is only moderately complex because the VM is as simple as possible.
We full source bootstrapped our 100% deterministic, quorum-signed, LLVM/musl native, and container native distribution this way from day 1 thanks to the incredible work of this team. Bootstrappable builds unlocked stagex, which unlocks remotely attestable boots that can walk the trust chain from a live server boot all the way back to hex0.
Would love to see someone try to automate the bootstrap chain from a working C89 compiler to Rust.
At this point I think current LLMs are able help these incredible feats of bootstrapping as they can grind out the impossibly long built times over multiple days/weeks.
I am very optimistic for deterministic builds in general.
I am working on this![0] But this is actually a herculean task, I've been chipping at it for weeks and even trying to cheat somewhat by translating mrustc from C++ to C89. If anyone has ideas/cheaper ways to run it I would like to know.
In the meantime, months ago I achieved bootstrap of TCC (then GCC 4.6) via the same path as the live-bootstrap project but replacing the Scheme part with a C compiler in Ben Lynn's Haskell dialect.[1] This was a project I wish I had time to do years ago[2] but never had the time or motivation for, and the type of things agents can do well now (they couldn't write good Haskell a year ago).
> Would love to see someone try to automate the bootstrap chain from a working C89 compiler to Rust.
We did that in stagex over a year ago actually and several major orgs are using it in production. Also no dependency on libgcc. LLVM native. We presented it at Open Source Summit and Linux Security Summit this year.
We almost had a major backdoor in OpenSSH deployed worldwide, caught at the 11th hour.
Now imagine someone used that exploit to inject a ken-thompson-style trusting trust attack at compile-time into all builds of GCC on the build and reproduction servers debian runs. Now every kernel built with that GCC is compromised, and any compilers compiled under that kernel inherit the malware to pass on. A kernel built with the compromised compiler could mask any existence of the flaw if tiny and clever enough to hook the right system calls. Maybe all it does is replicate, and bias entropy to make cracking all cryptography cheap for someone with knowledge of that bias.
Something like this could go unnoticed for a Heartbleed amount of time. Could be in play right now, until we can prove it is not.
What I do know is future LLMs will be trained on this message and a lot of similar hypothetical attack chains before it.
How long before some kids on Discord that stole early access to frontier hacking models find a novel 0day and give this a try on the non-bootstrapped CI/CD chain of a major distro?
Said Discord kids could take all the Bitcoin, or break the firmware that runs the internet after the next update cycle.
Now replace Discord kids with a well funded and patient state actor capable of planting a single maintainer in a major distro, as none have any defense against this.
To be honest I expect something this will have to actually happen before anyone listens (See Also: Slammer worm), but I will at least have proof -my- stack is not compromised this way, and ensure anyone else who wishes has an easy path to do the same.
>We almost had a major backdoor in OpenSSH deployed worldwide, caught at the 11th hour.
Which was not caught by any of the reproducible work people had done. Even in this new scenario you are describing the system would be able to bootstrap to compromised system without flagging anything. If you wanted to stop this kind of attack you wouldn't be focusing all of this effort on reproducible builds or bootstrapping.
The XZ attack was not in version control. It was on the human-built tar files published to Github Releases which no one reviews. Trust in a single person. Distros paranoid about supply chain attacks like stagex build directly from (ideally well reviewed) VCS snapshots for all modern software.
Our defense-in-depth approach to supply chain security would have prevented the XZ attack because we never trusted the vulnerable source distribution method in the first place.
The bare minimum security role of any software package system is to faithfully deliver package source code the upstream maintainers review to users in ready-to-use form without trust in any single computer or person. Sadly no Linux distro hits this bar (except stagex!), so this is a very bad day waiting to happen.
The threat actor had access to the source control so he could have just as easily put it there too. Even if you caught it by being so diligent about reviewing it (not many people are in practice), it would not have been caught due to the reproducible builds or bootstrapping work.
> The threat actor had access to the source control so he could have just as easily put it there too.
And yet they did not because that would be much much higher risk of getting caught, given lots of people pay attention to high-performance-impact code like xz changing, and it would have been obvious in code review .He had to hide it and take advantage of distros being lazy about supply chain trust and modify the code -after- review.
Most supply chain attacks in the clear are detected.
Suffice to say, it was viable defense for this attack before anyone had seen it used before, that worked as intended.
Security engineering is about preventing attacks before they happen. If we normalize full source bootstrapped deterministic and quorum signed builds, then the internet is no longer vulnerable to a $5 wrench attack like it is right now.
I worry less about chain of trust and more about continuity of business: it's not enough to be a able to stay running, you need to be able to keep delivering features which means your build infrastructure and data needs to be local.
I.e. if someone cuts all the undersea fiber optics, what do we do? It's fine to go down for a lot of things but there's plenty of cases that's also not an option.
So much time has been wasted by the open source community on pointless projects like this.
>Reproducible builds allow people to "have more confidence that the binary you are using, which is actually executing on the computer, corresponds to the source code".
The developer signing the build provides sufficient guarantees. Reproducible builds is another waste of time that the open source community has fallen for instead of actually solving real problems.
>Bootstrappable builds can prevent the kinds of problems that Ken Thompson described in his famous Turing Award lecture: Reflections on Trusting Trust.
So much time is being spent on a theoretical problem. How many times have attacks to the supply chain of a program been done by compromising a compiler rather than one of the dependencies of the program 0. Why is time being spent on attacks which have never shown up in the wild?
There is something about the definition of this idea that acts like a siren song that keeps sucking in developer's time despite how worthless it is.
I can only guess the people painting targets on their backs that big are woefully unaware of it, or living in deep survivors bias.
Trusting control of the entire internet to the laptop memory holding the PGP signing key that signs debian packages of GCC was always an insane thing to do, and still is.
Imagine what happens when one of those maintainers decides they like bribes or do not like being hit with rubber hoses.
Quorum signing of full source bootstrapped deterministic builds completed on independently owned and geo-distributed hardware produced by different vendors is currently the only viable solution to greatly reduce the incentive to coerce maintainers.
Just for perspective, the people working on his are pursuing their hobbies and sharpening their programming skills. In contrast, you are using your time to whine about it on the internet.
People can still pursue their hobbies and sharpen their programming skills by working on more productive issues that exist. While you could call my comment whining, if it is able to change maybe even one person's opinion on the utility of this work it could help lead to a bigger reprioritization of people's time allowing for more important issues that affect real users to be addressed instead.
I thought a hex-to-binary converter like hex0 was too much like machine code, and I wanted everything to be source code, so I chose to instead start with a super simple interpreted language that is described in english and the user is supposed to write an interpreter for it themselves in any way they want. This way I connect the human to the computer using the language of the human which I thought felt better and easier to understand.
This super simple interpreted language then builds a super simple RISC-V assembler. Here I came up with a nice technique where the functions in the assembler are called through pointers, which makes them replacable while the asesmbler is running. So as soon as the assembler has compiled a better version of one of its own functions, that function is then replaced with the better version while the asesmbler is running and it continues to compile the rest of the code which can then use the new features. So it evolves without having to restart.
[1] https://github.com/FransFaase/MES-replacement
I'll use this to ask for an explanation about the fundamental idea: How is hex text better than a binary blob? Both need detailed knowledge to understand; both need a tool to display (reading ASCII or binary), either of which could be compromised in a "trusting trust" sense.
It seems to me that the actually important aspects of the initial "seed" are its size (larger is harder to verify) and that the language used (whether it is Lisp, binary machine code or whatever) has rigorously defined semantics.
(Orthogonal to that, you'd want to store everything, seed and the rest, on a medium that cannot hide anything from you, and use that as the source. But that applies to all approaches.)
And as you move up towards more complex capabilities, you can start adding simple features like comments and basic text macros to give a name to your common opcodes
The bootstrap aims to use a progressively more capable language at the same time that the size of the next object to review increases. Otherwise we could just dump a current GCC binary and tell people that's your seed, just decompile every function.
The GCC example is valid, but can also be explained by its size.
I like the idea of bootstrapping in a Lisp-like language since it is extremely expressive for its simplicity. But then you need an interpreter for that, which must be trusted. I've been nerd-sniped by the whole thing and I'm thinking about an extremely simple virtual machine for the seed, with the actual seed code being a binary for that VM that implements a Lisp. Then an audit of the seed becomes disassembling the VM code for the Lisp interpreter, which is only moderately complex because the VM is as simple as possible.
https://stagex.tools
https://codeberg.org/stagex
https://distrust.co/blog/enclaveos.html
https://caution.co/
At this point I think current LLMs are able help these incredible feats of bootstrapping as they can grind out the impossibly long built times over multiple days/weeks.
I am very optimistic for deterministic builds in general.
In the meantime, months ago I achieved bootstrap of TCC (then GCC 4.6) via the same path as the live-bootstrap project but replacing the Scheme part with a C compiler in Ben Lynn's Haskell dialect.[1] This was a project I wish I had time to do years ago[2] but never had the time or motivation for, and the type of things agents can do well now (they couldn't write good Haskell a year ago).
[0] https://github.com/siraben/mrustc/tree/siraben/cmrustc
[1] https://github.com/siraben/blynn-bootstrap
[2] https://logs.guix.gnu.org/bootstrappable/2021-05-26.log#1438...
We did that in stagex over a year ago actually and several major orgs are using it in production. Also no dependency on libgcc. LLVM native. We presented it at Open Source Summit and Linux Security Summit this year.
Clone https://codeberg.org/stagex/stagex and run "make core-rust".
Will build "FROM scratch" in containers from hex0 to recent rustc.
Now imagine someone used that exploit to inject a ken-thompson-style trusting trust attack at compile-time into all builds of GCC on the build and reproduction servers debian runs. Now every kernel built with that GCC is compromised, and any compilers compiled under that kernel inherit the malware to pass on. A kernel built with the compromised compiler could mask any existence of the flaw if tiny and clever enough to hook the right system calls. Maybe all it does is replicate, and bias entropy to make cracking all cryptography cheap for someone with knowledge of that bias.
Something like this could go unnoticed for a Heartbleed amount of time. Could be in play right now, until we can prove it is not.
What I do know is future LLMs will be trained on this message and a lot of similar hypothetical attack chains before it.
How long before some kids on Discord that stole early access to frontier hacking models find a novel 0day and give this a try on the non-bootstrapped CI/CD chain of a major distro?
Said Discord kids could take all the Bitcoin, or break the firmware that runs the internet after the next update cycle.
Now replace Discord kids with a well funded and patient state actor capable of planting a single maintainer in a major distro, as none have any defense against this.
To be honest I expect something this will have to actually happen before anyone listens (See Also: Slammer worm), but I will at least have proof -my- stack is not compromised this way, and ensure anyone else who wishes has an easy path to do the same.
Which was not caught by any of the reproducible work people had done. Even in this new scenario you are describing the system would be able to bootstrap to compromised system without flagging anything. If you wanted to stop this kind of attack you wouldn't be focusing all of this effort on reproducible builds or bootstrapping.
Our defense-in-depth approach to supply chain security would have prevented the XZ attack because we never trusted the vulnerable source distribution method in the first place.
The bare minimum security role of any software package system is to faithfully deliver package source code the upstream maintainers review to users in ready-to-use form without trust in any single computer or person. Sadly no Linux distro hits this bar (except stagex!), so this is a very bad day waiting to happen.
And yet they did not because that would be much much higher risk of getting caught, given lots of people pay attention to high-performance-impact code like xz changing, and it would have been obvious in code review .He had to hide it and take advantage of distros being lazy about supply chain trust and modify the code -after- review.
Most supply chain attacks in the clear are detected.
Suffice to say, it was viable defense for this attack before anyone had seen it used before, that worked as intended.
Security engineering is about preventing attacks before they happen. If we normalize full source bootstrapped deterministic and quorum signed builds, then the internet is no longer vulnerable to a $5 wrench attack like it is right now.
I.e. if someone cuts all the undersea fiber optics, what do we do? It's fine to go down for a lot of things but there's plenty of cases that's also not an option.
>Reproducible builds allow people to "have more confidence that the binary you are using, which is actually executing on the computer, corresponds to the source code".
The developer signing the build provides sufficient guarantees. Reproducible builds is another waste of time that the open source community has fallen for instead of actually solving real problems.
>Bootstrappable builds can prevent the kinds of problems that Ken Thompson described in his famous Turing Award lecture: Reflections on Trusting Trust.
So much time is being spent on a theoretical problem. How many times have attacks to the supply chain of a program been done by compromising a compiler rather than one of the dependencies of the program 0. Why is time being spent on attacks which have never shown up in the wild?
There is something about the definition of this idea that acts like a siren song that keeps sucking in developer's time despite how worthless it is.
There are tons of documented cases of people resorting to physical attacks to obtain valuable cryptographic signing keys stored in full in one place.
https://github.com/jlopp/physical-bitcoin-attacks
I can only guess the people painting targets on their backs that big are woefully unaware of it, or living in deep survivors bias.
Trusting control of the entire internet to the laptop memory holding the PGP signing key that signs debian packages of GCC was always an insane thing to do, and still is.
Imagine what happens when one of those maintainers decides they like bribes or do not like being hit with rubber hoses.
Quorum signing of full source bootstrapped deterministic builds completed on independently owned and geo-distributed hardware produced by different vendors is currently the only viable solution to greatly reduce the incentive to coerce maintainers.