Matthew Garrett

https://nondeterministic.computer/@mjg59

Former biologist. Actual PhD in genetics. Security, OS security teaching at https://www.ischool.berkeley.edu. Blog: https://codon.org.uk/~mjg59/blog . He/him. Opinions do not reflect those of my employer.

Blog: https://codon.org.uk/~mjg59/blog
Signal: @mjg.59

Public Key

npub1jpfx7gvwn68l9rw8def53lkljwg0yv7fn99tqdzxyn68k29p8gmq9qvqhj

NIP-05 Address

mjg59_at_nondeterministic.computer@momostr.pink

Profile Code

nprofile1qqsfq5n0yx8farlj3hrku56glm0e8y8jx0yejj4sx3rzfarm9zsn5dspz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dswff2wh

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1jpfx7gvwn68l9rw8def53lkljwg0yv7fn99tqdzxyn68k29p8gmq9qvqhj

Show more details

Last Notes

2026-05-18 19:42:27 UTC

- reply

Slack, it turns out, is not a great mechanism for dealing with MFA logins

2026-05-18 19:41:36 UTC

Current fuckery is users being unable to authenticate to the VPN because it works by writing out an HTML page to the local filesystem and asking the browser to open that and Slack has registered itself as the default text/html handler

2026-05-13 06:35:17 UTC

Free as in Hoover free flights promotion

2026-05-07 15:56:46 UTC

The downside of full-on self-hosting is, of course, that your server will inevitably choose to have weird hardware issues while you're in another country

2026-04-27 08:12:54 UTC

Speaking of Unicode, I think many people would be surprised just how strong the pushback against it was even in the early to mid 2000s. UTF-8 adoption in Linux was a fight.

2026-04-27 07:54:17 UTC

- reply

Unicode is one of the most transformative technologies of the past 50 years

2026-04-27 07:53:13 UTC

- reply

It is absolutely wild that it is only in this century that it has become possible to reliably send a message that contains a mixture of latin, kanji, and hangul

2026-04-27 07:52:00 UTC

Revisiting the fact that DVD subtitles are bitmap images and sighing and also accepting that in the 90s the alternative was having DVD players speak all of the weird language encodings that existed before widespread Unicode adoption

2026-04-22 06:05:17 UTC

Garrett Hardin: I described the tragedy of the commons as a warning against a system that drives people to consume common resources for personal benefit
AI companies: Good news, we have implemented the tragedy of the commons as a service

2026-04-19 20:08:27 UTC

Palantir telling their foreign customers that their priority will always be to defend the USA is certainly a choice

2026-04-18 23:49:06 UTC

- reply

I should be clear that I write code by hand and enjoy the process, and also agree that the only way you're going to get high quality human developers is through doing that. But I also think that the world is probably better if more people are able to modify code to meet their needs, even if those people never turn into high quality human developers as a result.

2026-04-18 09:27:43 UTC

- reply

Clearly my most unpopular thread ever, so let me add a clarification: submitting LLM generated code you don't understand to an upstream project is absolute bullshit and you should never do that. Having an LLM turn an existing codebase into something that meets your local needs? Do it. The code may be awful, it may break stuff you don't care about, and that's what all my early patches to free software looked like. It's ok to solve your problem locally.

2026-04-18 09:05:07 UTC

- reply

If you're willing to accept that then what's the problem? Are we threatening to burn down Ikea stores now?

2026-04-18 08:29:36 UTC

- reply

Personally I'm not going to literally copy code from a codebase under an incompatible license because that is what the law says, but have I read proprietary code and learned the underlying creative aspect and then written new code that embodies it? Yes! Anyone claiming otherwise is lying!

2026-04-18 08:25:15 UTC

- reply

(Yes ok there are cases where code is beauty and embodies an idea that could make a grown man cry and:
(1) your code is not that code
(2) you would think nothing of copying the creative aspect of that code if you needed to don't fucking lie to me)

2026-04-18 08:23:06 UTC

- reply

https://nondeterministic.computer/system/media_attachments/files/116/424/735/049/733/841/original/45f5624c087043e5.png

2026-04-18 08:20:31 UTC

- reply

Look, coders, we are not writers. There's no way to turn "increment this variable" into life changing prose. The creativity exists outside the code. It always has done and it always will do. Let it go.

2026-04-18 08:17:49 UTC

- reply

When I write code I am turning a creative idea into a mechanical embodiment of that idea. I am not creating beauty. Every line of code I write is a copy of another line of code I've read somewhere before, lightly modified to meet my needs. My code is not intended to evoke emotion. It does not change people think about the world. The idea→code pipeline in my head is not obviously distinguishable from the prompt->code process in an LLM

2026-04-18 08:14:40 UTC

Free software people: A major goal of free software is for individuals to be able to cause software to behave in the way they want it to
LLMs: (enable that)
Free software people: Oh no not like that

2026-04-17 07:34:01 UTC

Fucksake I'm going to need to write some elaborate agent that snoops my packets and rewrites anything heading to the BBC through a VPN just so I don't get paywalled, but also this is a great reason to figure out what level of ebpf bullshit can be carried out

2026-04-17 07:32:21 UTC

- reply

(For absolute clarity there is no secret third party paying my legal bills, as I said in court under penalty of perjury I am in the unfortunate position of having to do so myself)

2026-04-17 05:09:24 UTC

RE: https://infosec.exchange/@0xabad1dea/116418277310717935
The secret third party paying my legal bills has finally been revealed
#note15xg…7amt

2026-04-02 09:10:17 UTC

- reply

Just think of your poor grandmother's love of exploring open buckets

2026-03-21 21:35:06 UTC

Blog post about my #bsidessf talk on using SSH certificates for git signing: https://codon.org.uk/~mjg59/blog/p/ssh-certificates-and-git-signing/

2026-03-20 06:01:02 UTC

Trying to convince my students that having all your security policy changes include a design doc describing the status quo, the desired outcome, why this change will achieve it, why alternatives were rejected, and then implementing it via some automation schema so it can't accidentally be reverted for no obvious reason is good actually

2026-03-16 08:51:13 UTC

- reply

What I'm taking away from this right now is that for the extremely narrow case of a problem that humans have documented well and written strong tests for, a robot can be competitive with a human expert. And then also today I got an LLM slop bug report that entirely misrepresented the code in question and had entirely fake conclusions. I'm still solidly team human.

2026-03-16 08:10:56 UTC

I am watching an absolutely fascinating exchange between a domain expert and someone driving Claude compete to produce the smallest functional implementation of a well-defined problem and wow the robot is being competitive, but this is very much a best case scenario of there being a very clear functional specification the output can be tested against

2026-03-12 06:07:22 UTC

Way back in the day a friend had to move an Apple Xserve across London to get it racked, and because massively rich South Africans are stingy with money he couldn't expense a cab so took it on the underground and someone looked at him, looked at the Xserve, looked back at him and said "That's a fucking big iPod"

2026-02-23 04:58:01 UTC

- reply

The nature of side channel attacks is that CPU-imposed barriers are no longer as strict as they should be. This means that hypervisor boundaries are porous - it's possible to extract material from other virtual machines. I understand the perspective that simply avoiding executing any untrusted software removes that risk, but I do not control the software running in other VMs on the same hardware. How do you solve that?

2026-02-21 11:06:47 UTC

- reply

Never going to stop being angry about the subset of the free software community that would rather pretend that non-free software embedded in hardware doesn't exist than allow that non-free software to be replaced with equivalently non-free software that fixes bugs

2026-02-21 11:05:12 UTC

18 years of the Linux fork that refuses to load the microcode that mitigates a range of CPU vulnerabilities

2026-02-13 10:19:40 UTC

Ring cameras using wifi are obviously vulnerable (as are all other wifi devices) to just jamming wifi channels so they can't communicate, but that's noisy and attracts attention. But they also don't appear to implement WPA3 or 802.11w and so you can also just spam them with deauth frames while being much less obvious. This is very easy and also, in the US at least, very illegal. You shouldn't do it.

2026-02-10 05:05:15 UTC

Enjoying the "Tech is more important than ideology" people removing rust on ideological grounds

2026-02-06 03:45:26 UTC

- reply

It doesn't matter whether C is good or not. It matters that if I write code in two languages that aren't C, and I want it to all be part of the same process, I need to care about C. C pervades all. You cannot escape it. C will outlive all of us. The language will die and the ABI will persist. The far future will involve students learning about C just to explain their present day. Our robot overlords will use null terminated strings. C will outlive fungi.

2026-02-06 03:34:17 UTC

https://faultlore.com/blah/c-isnt-a-language/ deserves a fucking record for managing to trigger people into being extremely upset while also demonstrating that they don't understand the actual point being made

2026-02-03 00:11:56 UTC

- reply

Ah! https://gitlab.freedesktop.org/hadess/iio-sensor-proxy/-/issues/216 ok I'll dig into this

2026-02-02 21:47:59 UTC

If you're finding that gdm is falling back to X11 and refuses to start Wayland sessions then uh make sure you haven't accidentally changed the owner of / because that results in some exceptionally weird breakage

2026-02-02 20:48:49 UTC

[@hadessuk](https://hachyderm.io/@hadessuk ) I have a tablet that exposes two accelerometers via iio which are supposed to be used to identify whether it's in tablet mode or not. Any idea what a generalisable solution to that would look like? There's various hacky device specific ones.

2026-01-26 04:51:30 UTC

- reply

Frankly: binaries are the thing that executes on your system and embody the truth of software behaviour, and with modern technology it's often *easier* to determine that truth through the binary than through the source code (throw the "login" app from Reflections on Trusting Trust into Ghidra and you'd learn the truth even if the source code wouldn't tell you that)

2026-01-26 04:49:45 UTC

- reply

But merely being free software isn't sufficient - software developed in a way that prevents arbitrary observers from witnessing design conversations may still be free software, but doesn't give us a strong reason to trust the developers. We all know how easy it is to hide dubious code in the open. The libxz backdoor was discovered by examining the binary and tracking that back to the source, not through source examination.

2026-01-26 04:46:56 UTC

The presumption that free software is sufficient or necessary to ensure all software you depend on is trustworthy is simultaneously naive and ignorant of what software is capable of. The only realistic way to develop trust in software is to trust the people who write it, and development processes associated with free software make that trust easier.

2026-01-21 08:36:14 UTC

"Why is my IPv6 not working" because I assigned that machine a static address associated with an he.net tunnel I'm no longer using because Google corp security had it tagged as being a Chinese VPN and disabled employee accounts if they joined my wifi from a company device
Sometimes I feel like I'm @npub1838…krhz except my fuckups involve less nuclear material and fewer geodesic domes

2026-01-20 20:32:49 UTC

https://www.openwall.com/lists/oss-security/2026/01/20/2 OH MY FUCKING GOD complete auth bypass in inetutils telnetd for over a decade and obviously nobody inside that decade should have been running a telnet daemon but wow

2026-01-15 09:56:32 UTC

- reply

Still going to argue that https://gitlab.com/mjg59/linux/-/commit/13cd6ec5e0e99124dd730156a4d921b20f192e2d would maybe be the most security per lines of code this decade

2026-01-15 09:54:04 UTC

Can someone please just write an ebpf script that blocks ../ and make it possible to automatically load that in systemd units and docker and any other security boundary

2026-01-13 07:12:11 UTC

- reply

Screen *locking* is actually pretty arcane magic and it's been very bound up in screen saving and as a result the complexity of both has become conflated

2026-01-13 07:00:13 UTC

- reply

So: In the past the infrastructure was simpler, the underlying hardware was simpler, the number of use cases you had to satisfy was smaller. And now everything is fundamentally more complicated and you're competing with platforms that have millions of absolutely normal computer users using them. But we've also got greater avenues of sharing knowledge, collaboration, better understanding of how to build abstractions. Was the golden age 30 years ago, or is it now?

2026-01-13 06:45:34 UTC

- reply

And yeah there's all sorts of cases where X exposes functionality that doesn't really exist in Wayland and I understand people wanting that but also I shared my desktop with some people today on a call and got a notification icon letting me know it was happening and I didn't have to rely on the client to behave correctly for that to be something I could rely on

2026-01-13 06:42:59 UTC

- reply

I'm pretty fascinated about why this is happening and my gut feeling is that people have simply written better abstraction layers on top of Wayland than were possible on top of X and now people can just write shit without needing to care anywhere near as much about the sharp edges that exist everywhere

2026-01-13 06:41:40 UTC

It is actually kind of wild that we're simultaneously in an era of people complaining that Wayland is destroying choice and also maybe the greatest number of high-quality desktop environments aimed at different use cases the free software world has ever had

2026-01-04 18:47:20 UTC

Hello everyone I moved to a new self-hosted blog thing and here is my first post on why "PC Compatible" is a social construct https://www.codon.org.uk/~mjg59/blog/p/what-is-a-pc-compatible/

2025-12-17 13:21:40 UTC

I had to spend a bunch of time researching the implementation of IRC ping timeouts for a lawsuit so now you get to know as well: https://mjg59.dreamwidth.org/73777.html

2025-12-15 03:04:32 UTC

iRobot apparently just declared bankruptcy, so if you have an internet connected one and want to retain control if the cloud platform vanishes, take a look at https://github.com/koalazak/dorita980#how-to-get-your-usernameblid-and-password and stash that information somewhere safe (and note that it changes if you ever factory reset the device, so try not to do that)

2025-12-10 07:25:12 UTC

If you wanted to hear me maybe talk about a thing at EMFCamp next year, what would be a thing you'd like to hear me talk about?

2025-12-07 00:14:21 UTC

Why does my soundbar have 184 systemd service files

2025-11-28 05:51:38 UTC

Poll for people from the UK and US only sorry I am being restrictive there is actually a reason

2025-11-20 12:12:19 UTC

In and around 2023, Roy and Rianne Schestowitz were subject to a horrific campaign of online harassment. Unfortunately they blamed me for it, and in turn wrote and published an astonishing array of articles making false accusations against me. Last year, I sued them in the high court in London. In turn, they countersued me for harassment. The case was heard last month and I'm pleased to say that the counterclaim was dismissed and I prevailed in my case. The court awarded me £70,000 in damages.

2025-09-24 22:25:29 UTC

I have been learning more about PDFs than I really wanted to for maybe the absolutely most funny reason possible - letting agency forgery: https://mjg59.dreamwidth.org/73317.html

2025-09-22 07:12:49 UTC

You obviously can't use LLMs to review code written by LLMs. So you still need people who know how to read code. How do you get those people without having people write code?

2025-09-11 19:53:22 UTC

Spending time with my in their 70s parents and they're very successfully managing not to be racist or transphobic so I still don't buy the "They're a product of their age" argument

2025-09-06 21:54:04 UTC

A fun thing you can do right now to test whether your system will stop booting next week because of a secure boot certificate expiring!
1) Does your system currently have secure boot enabled? If not, go to step 5
2) Download Fedora 42
3) Does it refuse to boot with a secure boot violation? If not, go to step 5
4) This is interesting and unexpected! Please let me know, you will literally be the first
5) Nothing is going to happen to you next week

2025-09-04 17:14:36 UTC

Only one more week to take advantage of https://nondeterministic.computer/@mjg59/115048295075274818 (so far nobody has, so I assume at this point everyone is on board with the idea that nothing whatsoever is going to happen as a result of non-existent secure boot certificate expiry on 2025-09-11)

2025-08-18 06:14:26 UTC

People who are still inclined to believe that Linux systems will stop booting next month because of secure boot rollover! Send me evidence that you have donated to a charity and, if Linux stops booting on any system after 2025-09-11 because of some sort of certificate rollover bullshit, I will (your choice) either match that donation or pay you back your donation (you will need to deal with the tax consequences), up to a total of $50,000.

2025-08-15 17:21:20 UTC

2012: "Secure boot is a plot by Microsoft to kill Linux"
2025: EA's insistence on invasive anti-cheat results in a bunch of Windows users managing to get their secure boot configuration into a state where their GPUs no longer work and there's no recovery path: https://www.reddit.com/r/Battlefield/comments/1miaynl/secure_boot_megathread_guide_community_support/
Microsoft would have to be *very* bad at this for a plot to have backfired this badly

2025-08-11 05:00:03 UTC

I suddenly need to care about Amiga DMA cycle timings and this isn't an improvement in my quality of life

2025-07-27 22:20:29 UTC

Need my GB driver's license details For Reasons, can't find the card, remember it's saved in my Hertz profile, go there, page shows me most the the number ***d out, check developer tools, API is returning the full number and just obfuscating it client side, hurrah

2025-07-23 08:42:23 UTC

Anyone sharing anything about phones always listening to your conversations and showing you adverts about that:
(a) Do you realise how incredibly easy it is to monitor all the traffic coming from your phone?
(b) Do you understand how giant a scandal it would be if that were demonstrated?
(c) Do you know how many people there are who would love to be at the confluence of (a) and (b) and yet have not been because it's not a thing that's happening?

2025-07-18 23:08:13 UTC

Today's non-sensical TPM-related bullshit: someone telling me that TPMs can prevent you from clearing firmware variables

2025-07-07 06:57:37 UTC

"Huh why is logging into the server in the room next door so slow" because my eeros rebooted overnight and my laptop roamed onto the other SSID that's VPNed back to the UK and all my packets have been travelling 10,000 miles to go a few metres

2025-06-25 22:28:25 UTC

"Why bother writing free software if nobody's going to use it" why bother producing art or documenting the life cycle of an extinct insect or sharing anything that you find interesting or fun - if people find something I've done useful then wonderful, but even if they don't I want to be able to share it with them

2025-06-25 19:42:30 UTC

Writing my first actual Rust project tell my family I loved them

2025-06-20 07:30:00 UTC

My first paid software development job was on accessibility software. I remember visiting a user to help set up our code, and the sheer joy he had at being able to communicate more quickly than he had been able to for years.
Seeing the effort put into improving modern Linux accessibility is heartwarming. There's been almost 20 years of almost nobody caring. It's important. It's worthwhile.
Say thank you to the people doing that work. Stop amplifying the people saying that work isn't happening.

2025-06-12 11:08:52 UTC

In the odd limbo space of having cleared US immigration on a different continent, so while I'm not strictly back in the US yet I'd rate the probability of arrest as low

2025-06-09 21:03:28 UTC

- reply

no meaningful secure messenger asserts that messages cannot be shared by the recipient, and any developer of such an app that made that claim would be an indication that app shouldn't be trusted

2025-06-09 20:51:05 UTC

- reply

are you arguing that it should be impossible to take screenshots of messages sent to you? Because that's a really bad line of argument to start down

2025-06-09 20:39:59 UTC

- reply

a client running on an end user's system is, as it should be, under that end user's control. If you send someone messages then they are going to be able to choose how they're retained. The alternative implies computers that are not under the owner's control.

2025-06-09 11:44:31 UTC

- reply

https://signal.org/blog/private-contact-discovery/ is the writeup of it, I don't know if there's a further spec other than the implementation

2025-06-09 11:40:50 UTC

- reply

the enclave attests itself to the client, the server admin can't fake it or monitor the communication

2025-06-09 10:24:20 UTC

- reply

Which Signal logs?

2025-06-09 10:23:27 UTC

- reply

It's storing the phone number/XMPP ID, and it claims that the app will periodically re-query, so if you run the server you can just log the lookup

2025-06-07 12:16:34 UTC

- reply

Signal's social graphing is done in secure enclaves, that information isn't extractable

2025-06-07 11:57:58 UTC

- reply

yes, you can trade off functionality to gain privacy on a fairly arbitrary scale, but teaching people to make informed decisions about where they should be on that scale is hard and you still end up with metadata transfer between servers with very different security properties. Is this a good choice for some people? Yes! Are most people going to end up making well informed choices? Not even slightly

2025-06-07 10:54:20 UTC

- reply

you're trading off centralised server infrastructure for contact discovery that's less privacy preserving (quicksy has a list of every Jabber ID and phone number combo!), which may be worth it for some use cases but certainly isn't for all!

2025-06-06 11:29:34 UTC

- reply

what does the server being in the US have to do with anything?

2025-06-05 11:12:20 UTC

Twitter's new encrypted DM system stores your private key material on Twitter-owned services, protected with nothing more than a 4-digit PIN. If hostile, or if legally compelled to, Twitter could easily decrypt all your messages. It's also MITMable and doesn't secure metadata. Use Signal.
https://mjg59.dreamwidth.org/71646.html

2025-05-24 06:40:21 UTC

- reply

Yeah great ok the manufacturer also has no ability to patch my fridge but how does that help me this differentiation between software in ROM and software in flash is absolute bullshit, it's all software and it should all be free and modifiable

2025-05-24 06:38:46 UTC

RMS printer moment, but it's about my fridge failing to run the ice maker fill tube heater for long enough so it freezes so the ice maker gets no water and makes no ice, but also the software controlling this is in ROM so the FSF says it's fine even though there's no way I can fix it even though I am entirely capable of reverse engineering and patching it otherwise

2025-05-19 07:47:48 UTC

- reply

sadly my girlfriend is from outside Gothenburg so I think I'm actually contractually forbidden

2025-05-19 07:34:52 UTC

Ah good flight to the UK booked for next month

2025-04-25 06:15:37 UTC

I'm kind of surprised that there's still no hypercall for mark_kernel_ro() that marks the kernel code read-only in the hypervisor page table such that even if someone finds a gadget to make it writable again it won't work in a VM situation - are there hypervisors that automatically detect this case and do it implicitly?

2025-04-17 08:52:25 UTC

- reply

This is hilariously Northern Ireland (both the initial fuckup, but also the somewhat confused response of everyone else, including the journalist)

2025-04-05 07:03:26 UTC

Extremely niche content, but: back in the early 2000s the AmigaOne was released using a northbridge produced by a company called MAI Logic. It apparently had awful cache coherence bugs when using DMA, but the details are lost in a sea of ludicrous infighting between different Amiga splinter groups. Does anyone actually know the technical details of what was broken with it?

2025-03-21 16:10:30 UTC

- reply

If the membership chose a candidate that the board refuses to appoint, the membership deserves to know that.

2025-03-21 16:07:14 UTC

- reply

There is a well defined process for how elections are run (https://opensource.org/about/board-of-directors/elections). This year, several candidates were removed from the election results for supposedly refusing to use proprietary software to sign the board agreement. The requirement to do so is not stated anywhere in the election process. The bylaws allow the board to ignore the will of the electorate, but removing candidates who didn't violate any election rules is just fucking ridiculous.

2025-03-21 16:00:25 UTC

Like many nonprofits, membership of the @npub1djq…zj5f board of directors is technically down to the existing board choosing to appoint new directors. OSI "membership" (in the sense that donors can become "members" of the organisation) isn't a thing that's defined by the by-laws, it's a process that the directors have chosen to adopt. As such, the elections that members participate in are just guidance for the board's decisions, rather than anything they're obliged to respect.

2025-03-19 00:01:10 UTC

When Twitter launched encrypted DMs they were bad. They haven't improved. The person behind them is now a senior member of DOGE and getting appointed to the board of a government-backed mortgage giant: https://mjg59.dreamwidth.org/71188.html

2025-03-18 03:19:08 UTC

- reply

Nobody is going to try to make money on a proprietary fork of an MIT Coreutils. Nobody is hiding their trade secrets there. This isn't the 80s.
What is a bigger issue is the more symbolic nature of things. People had the opportunity to pick a copyleft licence and chose not to. We can view this as an attack on copyleft (albeit one that's likely symbolic at best), or we can accept that the copyleft community has been doing a poor job winning the hearts and minds of new generations of developers

2025-03-18 03:14:06 UTC

- reply

What Coreutils-adjacent GPL enforcement there has been centred around Busybox, a GPLed implementation of many POSIX and Unixish tools, commonly used in embedded devices. Busybox-related enforcement has been an effective tool in obtaining compliance, to the extent that it's been reimplemented under a permissive licence with the explicit goal of reducing enforcement risk. Coreutils has simply never been subject to enforcement in the same way, so there's no significant impact.

2025-03-18 03:05:47 UTC

I'm a huge proponent of copyleft licensing, I'm in favour of using the GPL as a tool to ensure users have the ability to modify the software on their devices, and I'm just having trouble getting too worked up about the Rust reimplementation of Coreutils being MIT. Philosophically? Yeah, it sucks. Practical outcomes? Almost certainly none. The GPL violators aren't going to change coreutils implementation to avoid being sued, the FSF wasn't going to do that anyway

2025-03-08 18:46:45 UTC

The discovery that lifepo4 power banks are now at the point where they can switch from AC fast enough that they're usable as UPSes is about to change my life in a short term expensive but long term cheap way