reynir @reynir@bsd.network
Joined Oct 2020
I got a mail from @\nowsecur0 . com saying Secur0 can do free security audit of open source software. It seems Secur0 is a legit thing (hard to tell these days), but the domain nowsecur0 . com was registered about a month ago and Secur0 seems to use secur0 . com - so likely a scam. Please beware!
I received the email to my email address I use in my public GitHub profile.
reynir
boosted
Thanks to @NGIZero @nlnet, we have funding work on μTCP, our TCP/IP stack for #MirageOS / #mkernel #OCaml #unikernels https://nlnet.nl/project/microTCP/ .
The longer story is this project I first heard about in 2012 at ICFP in Copenhagen where @peter_sewell_ gave an impressive talk "Tales from the Jungle". I was at that moment doing my PhD on software verification (Coq, separation logic) and it was so nice to see real world stories of verification.
For several reasons (MirageOS melting pot, OCaml-TLS with David Kaloper), I went to Cambridge and worked with Peter on that formal model, read our JFP journal paper http://www.cl.cam.ac.uk/~pes20/Netsem/paper3.pdf .
Then I left academia, and worked in my spare time (finding projects, going through team processes, ...) on the OCaml implementation very slowly. Motivated by @dinosaure and @reynir we deployed it in 2024, and fixed issues. Now we have several months of funded work to push it over the finish line.
Big thanks to everyone involved so we can push this forward. 🎉💕
Been looking into how cloud-init uses ISO-9660 (the CD ROM filesystem). Among other files they put a user-data file. This could be a cloud-config yaml file. But it could also be a MIME multipart file containing more files. This strikes me as super weird - you already have a read-only filesystem so why use anothee half-assed filesystem?
But then, cloud-init supports multiple so-called data sources of which ISO-9660 is one of them. In that context it makes a bit more sense.
reynir
boosted
@reynir and the other question is do OS stacks like TCP sessions where port is 0? if not, why not?
So my gut feeling is OS's void these ports for outgoing connections, but I can't explain why they would.
Ports below 1024 are privileged and operating systems will only allow privileged users to listen on those ports. But will operating systems avoid these ports for source port in outgoing connections?
My router's DHCP server sends a DHCPOFFER with a siaddr of 0.0.0.0 and server identifier the actual IP (dhcp option 54).
If my client then replies with a DHCPREQUEST with siaddr 0.0.0.0 and server identifier 0.0.0.0 then it's ignored.
However, if my client replies with siaddr 192.168.0.1 and server identifier 192.168.0.1 it's accepted. Also if siaddr is 0.0.0.0.
So my client should send the correct server identifier, but I am now confused what siaddr even means.
reynir
boosted
A year ago, I started implementing a ntp server in #OCaml. I had finished the big part: the client side, which combines multiple sources. Recently, I got back to this project and implemented the server side, which turned out not to be that difficult at all. Today, the #unikernel is available on osau.re:123, and my ntpd selects it as a synchronization source! The project is available here: https://github.com/robur-coop/chaos. It’s still experimental, so let’s see how it holds up over the long term.
reynir
boosted
I’ve been working for several days now on an old #unikernel that I started back in 2022. Four years on, it incorporates our latest work on mkernel, mnet, #OCaml 5 and vif, which has allowed me to redeploy it on my server. https://paste.osau.re/ is now available again! It’s a good example of how to build a unikernel today. All the code is available here for those who are curious: https://github.com/dinosaure/pasteur. The deployment user experience is also much better than it was with Albatross and aussi!
reynir
boosted
I've finally finished my implementation of SSH for unikernels in #OCaml. For this example, I've used @reynir project to implement a #unikernel that offers a chat service (like IRC) via SSH. So now we have an engine for providing a TUI within a unikernel that doesn’t even recognise the concept of a console (thanks to lwd)!
reynir
boosted
I just made a new release of Solo5 (https://github.com/solo5/solo5) and it improves our #unikernel (specially the networking) for Linux (we can transfer 3 times more data per second)! It paves the good way for next things.
reynir
boosted
I intended to write this for a while: Security issues to look out for in OCaml (based on security issues I discovered or helped fix in the past): https://discuss.ocaml.org/t/security-issues-to-look-out-for-in-ocaml/18158
OCaml prevents a certain category of security issues, but that shouldn't make us over-confident.
I tried to generalize a bit:
* security properties don't compose: module A and B can be secure on their own, but you still need to test/fuzz their interaction
* Look at security bugs fixed in other ecosystems. Chances are the OCaml implementation fell into the same trap that other implementations did.
They responded. It was indeed an LLM. They had given it instructions to only request for more relevant info and to sign off with the above "Issue Agent Bot" because they use an access token to their personal account for the LLM.
No reply on whether the students should ignore the go-ahead or not...
Someone in qemu-project is triaging issues and signing them off with "Issue Agent Bot on behalf of the user". Now I don't know if I can trust what they say. They are saying "go ahead and work on this" to my feature request to some students, and I feel bad for them if some bot just told them to waste their time :/
reynir
boosted
I'm really pleased to announce the next version of Miou (0.6.0) here: https://discuss.ocaml.org/t/ann-miou-a-simple-scheduler-for-ocaml-5/12963/18. Step by step, we build reusable pieces of #OCaml libraries and it's nice to see results all together. Thanks for all contributors on these various projects.
reynir
boosted
The OCaml Users Survey 2026 is now open! 📊
Share your experience with OCaml - from tools you use to new questions on AI/LLM tooling and debugging. Takes 10-15 minutes, open until May 25th.
https://forms.gle/gt5nikqUmoQWeYYQ9
https://discuss.ocaml.org/t/ann-ocaml-users-survey-2026/18026
I used a dedicated email address for registering my PEN. I can say that the email attracts a decent amount of SPAM.
reynir
boosted
I am pleased to announce the launch of our mailing list ptt@mailingl.st. You can view the announcement here: https://discuss.ocaml.org/t/ann-ptt-a-mailing-list-system-as-unikernel/17987. It brings to fruition all the work we have done on #unikernels and email, all in #OCaml!
The OCaml users survey 2026 opened up for feedback on the questions. A few of the questions are on AI usage. I suggested they add an option to one of the questions so they not only have an option for "no plans" on using AI but also an option for people who are morally opposed to using AI. I hope they add that option, and I look forward to seeing the results.
Joined Oct 2020
