Kerla: Monolithic kernel in Rust, aiming for Linux ABI compatibility
github.com
github.com
I know this kind of stuff spark the ``it's meaningless to rewrite everything (especially Linux) in Rust'' debate. I agree 100% that rewriting Linux in Rust (or your favorite language) is just a waste of time and such a project won't live long.
That said, what I'd say here is that it's fun. Really fun. Implementing ABI compatibility requires you to understand under the hood of the kernel. You can learn how printf(3) works. You can learn what happens before main().
Especially, in Rust, it's inspiring to explore ways to write a kernel safely and expressively. One of my favorite part is the implementation of read system call [1].
Lastly, You can try this OS on SSH. Hope you enjoy :)
ssh root@kerla-demo.seiya.me
Considering that this is exactly how Linux was born (just a hobby project for fun), I wouldn't assume so fast that it's useless. Moreover, you don't need to justify yourself if you just want to have fun !
I agree they shouldn't have to justify themselves, but it is handy to preempt that.
In fact in this instance I think it's a little disingenuous to quote Linux and say it could happen again. The industry is totally different now. There's much more competition than there was when Linux was released and that competition is much more mature too. Plus the bar for a production-quality kernel is a lot higher than it was when Linux was released.
Seems apropos to me, given the fact that a Linux ABI compatible hobby project is under discussion - and that everyone here is familiar with the famous Usenet announcement.
> The industry is totally different now. There's much more competition than there was when...
I wonder how you define "competition"? Because there were way more operating systems in use then, and the industry was far more fractured. Fractured in a way that was meaningful - not like today where you can spin up a VM and be productive in short order, thanks to the significant lack of distinguishing difference. That is the really interesting thing about these hobby projects - they introduce possibilities that are either completely ignored by organizations suffering from inertia constraints, or can't be mimicked because they're diametrically opposed to present designs.
There is way more competition now:
+ Linux (countless distributions)
+ FreeBSD
+ OpenBSD
+ NetBSD
+ DragonflyBSD
+ HardenedBSD
+ Darwin
+ Minix 3 (which wasn't free when Linux was released)
+ Illumos
+ OpenIndiana
+ Nexenta OS
+ SmartOS
+ ...and many others based off OpenSolaris / Illumos
Don't conflate standardisation of the industry with a lack of options. More options do exist today and are in use (eg some games consoles run FreeBSD, Netflix uses BSD, Nexenta is used in some enterprise storage solutions, Darwin may not be used in any free capacity but macOS is clearly used heavily by HN readers, and so on and so forth).
Moreover, I've used FreeBSD, Solaris, OpenSolaris, Nexenta and OpenBSD on production systems over the last 10 years (and the list gets more esoteric if we look past 10 years). So just because you might exist in a Linux-only ecosystem it doesn't mean that's the case for the entire IT industry.
> So just because you might exist in a Linux-only...
Obnoxiously presumptuous.
lol it's not nonsense. Every item I've listed offers something unique.
> Just look at the last 5 items... seriously - all these Illumos derivatives are distinct operating systems in competition with one another?
Actually the differences in the Illumos derivatives are quite fundamental:
- Nexentra is aimed at being an enterprise storage solution with Debian user land. That's massively unlike Illumos
- Smart OS is designed to be a smarter virtualisation and containerisation host OS. It even bundles Linux's KVM virtualization. It's not intended to be run as a desktop platform
- OpenIndiana and Illumos are probably the most similar, however OpenIndiana aims to be more true to OpenSolaris while Illumos aims to be more of a hybrid upstream. So while they're both multi-paradigm (like how Debian can be a desktop or server OS) there are some major differences in their commit tree.
Honest question: how close are you to the Illumos projects? Or are you just an outsider looking in and making assumptions about equivalence based on your experience with Linux? I ask because Illumos forks are much more akin to BSD forks than they are Linux distros.
> Be serous, that list only further proves my point about the total lack of distinguishing difference in the present offerings - relative to the pre-linux environment.
Even if you want to pare down the list to upstreams, you still have half a dozen BSDs (I really hope you at least have enough experience with BSD to realise these aren't just respins like Linux distros), Illumos, Minix, and Darwin. Verses your point in the 90s which was basically non-free BSD, non-free SysV, non-free QNX, non-free Minix etc....you still have those options now PLUS the ones I've listed.
This is the problem with your argument. You're assuming the old choices have gone away, which they haven't...well, apart from SCO and Minix is now free. And in addition we have dozens of interesting new platforms, some of which I've exampled, too.
There's no way on Earth we have less choice now than in the 90s. We might have the industry largely standardising on a subset but that's an entirely different argument and it's only representative of the lowest common denominator doing common problems. However if you look slightly outside of the status quo and you'd see there is a lot of variety still happening in the industry. I know this first hand too -- as I've said in my previous post, I've worked in plenty of places that weren't just Linux shops :)
You then go on to characterize components grafted in from another OS as "unique". That is a laughably low bar, making every potential permutation of existing operating system components distinct offerings in competition for mindshare.
> Honest question...
Honest answer: you are running code I've written. I've been using ZFS in production for as long as one possibly can. My home file server has been SmartOS for many years, before that it was FreeBSD.
> I really hope you at least have enough experience with BSD to realise...
I'm keenly aware of their differences and commonality, having tracked the introduction of a terminal mode bug all they way back to a time when "Melbourne getty" was a thing.
> There's no way on Earth we have less choice now than in the 90s.
Well, I suppose that if you are including abandonware and dead product offerings in the list of competing operating systems...
> We might have the industry largely standardising on...
It seems like you are trying really hard to avoid using the more appropriate word for what has happened: "consolidating".
That's literally no different to how Unix evolved. You talk about diversity in the 90s when most of the platforms were being trolled by SCO for containing the same code base. Don't you see the hypocrisy of your argument there?
> Honest answer: you are running code I've written. I've been using ZFS in production for as long as one possibly can. My home file server has been SmartOS for many years, before that it was FreeBSD.
Then it seems very strange that you don't acknowledge the differences in current platforms while proclaiming that Unix was more diversified in the 90s (if it was that clear cut SCO wouldn't have been trolling everyone).
> Well, I suppose that if you are including abandonware and dead product offerings in the list of competing operating systems...
You're overstating things once again. :)
> It seems like you are trying really hard to avoid using the more appropriate word for what has happened: "consolidating".
No. Consolidating means the removal of options. Those options still exist they're just not as commonly used. Thus term I used of "standardisation" is more apt.
Look, I do understand the rose tinted glasses you're wearing. There are aspects of 90s era systems administration and development that I miss too. But I still feel you're way off the mark with your opinions here. In some places you are exaggerating a nuanced point to such an extent that as much as I'd love to cheer on for the "good old days of computing", your comments simply don't represent my experiences then nor now.
Uh, that is exactly my point. You know that you were just arguing that they were unique operating systems, right? You don't see how the comparison totally undermines that position?
> ...while proclaiming that Unix was more diversified in the 90s...
I'll never understand why anyone bothers fabricating strawmen in thread based mediums - I never made that claim, and that is plainly obvious to anybody who has the ability to scroll up. I have a hard time believing that you could, in good faith, read "there were way more operating systems in use then" and honestly think "oh, he is obviously only talking about Unix!"
> No. Consolidating means the removal of options.
Yes... you remember what the option was selecting from? I'll save you the trouble of manipulating your scroll wheel: "The industry is totally different now. There's much more competition... the bar for a production-quality kernel is a lot higher..."
You still wanna keep saying that stuff from before, about options?
I never said they were unique operating systems. I said we have more choice now than we ever with regards to Unix-like systems and I said they were distinct platforms, code bases, even upstreams. All of which is true. I have no interest in entering a philosophical question of what changes constitute a "unique operating system" however we can at least have a technical discussion about choice and build of the Unix-ecosystem.
> > ...while proclaiming that Unix was more diversified in the 90s...
> I'll never understand why anyone bothers fabricating strawmen in thread based mediums - I never made that claim, and that is plainly obvious to anybody who has the ability to scroll up.
All these comments you make of strawman arguments and now I can see that the original reason for our disagreement was that you replied arguing a different point to me from the outset!
I was only ever talking about Unix-like platforms (given we're talking about Linux ABI) and you misunderstood that post to think we were talking about operating systems in the broader sense. In fact the language I used should have made the scope explicit so I'm giving you the benefit of the doubt here that you weren't intentionally just pulling an asshole move of subtly changing the scope to win a different argument (which is the very definition of a straw man -- since we're already arguing about who's trying to pull a straw man)
> I have a hard time believing that you could, in good faith, read "there were way more operating systems in use then" and honestly think "oh, he is obviously only talking about Unix!"
I was though. You were the one who replied to my point where I was only talking about Unix so good faith would dictate you continue with the same scope and context as the conversation started out with. So yes, I did assume you were still only talking about Unix. Because that's what the conversation was about prior to you joining it.
-----
If we are to discuss your point then I agree there are fewer operating systems in general. And I agree that is a great loss. I'm also happy to talk at lengths about that too..... but I don't really see what relevance that has within a discussion about whether yet another POSIX kernel will "become the next Linux" because even if Kerla was to "become the next Linux" it still wouldn't satisfy your argument about diversity. So why make it in the first place?
Like I said before, I'm using a lot of good faith here assuming your misunderstanding was a genuine one and that you weren't intentionally trying to derail the conversation.
Indeed, there's a word for it which I find particularly lovely, autotelic.
I think it's the opposite. When Linux was released, basically every major tech company had their own variant of Unix. Now almost everyone has migrated to Linux or Windows.
Competing with modern Linux by creating a drop-in replacement for Linux would be a pretty difficult task given that people have been optimizing Linux for decades now, but on the other hand I think there are ways to improve dramatically on the traditional POSIX-style API that Linux mostly adheres to. For a random example, why can't a given process have more than one "current working directory" at a time? That seems like an arbitrary limitation imposed by an implementation detail in early Unix system, and it causes problems for modularity. There are many other little details like that. I think if Linux is replaced by something eventually, it'll most likely be because the new thing has a cleaner, more powerful, or more generally usable API.
(Kerla is apparently not trying to invent a new API; I'm just saying that's the direction I would recommend to anyone project with a goal of seriously competing with Linux.)
Of course that's hard to pull off in practice. Linux might have classes of errors that wouldn't exist if it were written in Rust, but even if using Rust eliminates three fourths of the code defects, the end result could be less secure if Linux gets ten or a hundred times as much scrutiny from people actively looking through the code for bugs to fix.
There wasn't many UNIXes that targeted x86 aside from 386BSD. And even those that were available were often expensive. Even Minix wasn't free at that time. Which is exactly the reason Linus created his hobby OS.
Now we have a dozen different flavours of BSD, countless Linux distributions and several OpenSolaris spin offs too. Plus many other underdogs used in speciality domains. And that's before even looking at the commercial offerings like QNX, Solaris and macOS.
Just because there are a couple of industry heavyweights that take up the majority of common use cases, don't be fooled into thinking there is a lack of choice nor even that the industry is scared to use anything outside of Linux and Windows.
> Competing with modern Linux by creating a drop-in replacement for Linux would be a pretty difficult task given that people have been optimizing Linux for decades now, but on the other hand I think there are ways to improve dramatically on the traditional POSIX-style API that Linux mostly adheres to. For a random example, why can't a given process have more than one "current working directory" at a time? That seems like an arbitrary limitation imposed by an implementation detail in early Unix system, and it causes problems for modularity. There are many other little details like that. I think if Linux is replaced by something eventually, it'll most likely be because the new thing has a cleaner, more powerful, or more generally usable API.
That's not the goal of this nor any other project that aims for ABI compatibility with Linux and there are plenty of research kernels out there if that's your thing.
> (Kerla is apparently not trying to invent a new API; I'm just saying that's the direction I would recommend to anyone project with a goal of seriously competing with Linux.)
The only way to seriously compete with Linux would be to get large commercial backing. And even then, good luck. Linux won not because it is the best but instead because it's "good enough". Same is true with Windows. Any engineer with aspirations of creating a "better kernel" or "better OS" needs to remember that.
[0]: https://gitlab.com/0xTJ/mosys/-/blob/master/src/vfs.c#L469
[1]: https://gitlab.com/0xTJ/mosys/-/blob/master/src/vfs.c#L907
If only you could write an entirely new (meant to be better, for some cases at least) kernel compatible with Linux hardware drivers - this would be not waste at all but in fact fantastic.
An idea in my mind is to use Kerla for virtualized environments like KVM where only limited device drivers are needed (virtio for example).
IMHO getting a minimal set that works is good enough to get people on board. This is still nontrivial. But for example, getting FUSE working would be useful. Even if the driver itself was not ABI compatible, it would bring functionality and someone might then aim for ABI compatibility afterward which would open even more doors.
Especially since, as I understand it, even Linux isn't compatible with the Linux driver API across versions; they can and will change internals at will and just update in-tree drivers to match. Hence some of the difficulty doing things like getting a newer kernel on assorted embedded devices (ex. 99% of phones) because you have to port the vendor's drivers to the new version and both sides changed stuff.
What do you think about other Rust OS projects like Redox?
Aside from Redox, I'd mention Tock [2], an embedded operating system. It introduces a novel components isolation using Rust's power. I believe this field is where Rust shines and am looking forward to seeing it in production.
[1]: https://www.redox-os.org/screens/ [2]: https://www.tockos.org
https://github.com/betrusted-io/xous-core
As a embedded service processor OS for a big server rack, Oxide Computer is working on 'HubrisOS'. They seem to have not released it yet, but that will be open sourced.
https://github.com/oxidecomputer
Those are two efforts where I know real resources are going into.
This is a great talk about what they do and why: https://www.youtube.com/watch?v=vvZA9n3e5pc
Does that mean the product will not have an FPGA? I kind of liked that idea of updating the hardware.
P.S: Twitter spaces about the history of computing are really fun. Love to hear more about all the dead computer companies you researched. There is not enough content about computer history out there.
1. ASICs are out of the question for us for a bunch of economic reasons
2. FPGAs by and large do not have a good security story
3. The ones that do (or rather, the one that does) has an entirely proprietary ecosystem and a terrible toolchain -- and seems to rely on security-through-obscurity
4. Once you are away from a softcore, the instruction set is frankly less material than the SoC -- and the RISC-V SoC space is still immature in lots of little ways relative to the Cortex space
5. Frankly, there's a lot to like about ST: good parts, good docs, good eval boards, pretty good availability (!), minimum of proprietary garbage.
We tried really hard (and indeed, I think my colleagues would say that I probably tried a little too hard) to get FPGAs to be viable, and then to get FPGAs + hardcores to be viable, and then multicore hardcores to be viable (one for the SP, one for the RoT). Ultimately, all of these paths proved to be not yet ready. And while we're big believers in FPGAs and RISC-V, we're even bigger believers in our need to ship a product! ;)
On Hubris and Humility: when "write your own OS" isn't the worst idea
Hubris is a small open-source operating system for deeply-embedded computer systems, such as our server's replacement for the Baseboard Management Controller. Because our BMC replacement uses a lower-complexity microcontroller with region-based memory protection instead of virtual memory, our options were limited. We were unable to find an off-the-shelf option that met our requirements around safety, security, and correctness, so we wrote one.
Hubris provides preemptive multitasking, memory isolation between separately-compiled components, the ability to isolate crashing drivers and restart them without affecting the rest of the system, and flexible inter-component messaging that eliminates the need for most syscalls -- in about 2000 lines of Rust. The Hubris debugger, Humility, allows us to walk up to a running system and inspect the interaction of all tasks, or capture a dump for offline debugging.
However, Hubris may be more interesting for what it doesn't have. There are no operations for creating or destroying tasks at runtime, no dynamic resource allocation, no driver code running in privileged mode, and no C code in the system. This removes, by construction, a lot of the attack surface normally present in similar systems.
This talk will provide an overview of Hubris's design, the structure of a Hubris application, and some highlights of things we learned along the way.
I have to say, I feel a bit “dirty” carrying so much unused and legacy code around with Linux, so I like people trying to reinvent the wheel just for the pleasure of a fresh start. For the aesthetics. Even if it’s merely a fantasy and not replacing anything soon, realistically. They are also keeping OS development accessible to new generations of geeks. The unfriendliness of C, the gigantic codebase and seemingly distinct culture make the Linux kernel quite off putting, filtering possible engagement by unfortunate parameters IMO. Novel OS development in Rust takes away at least some of those barriers and some of the gained knowledge may be applicable with the Linux kernel later.
For example, I've written a Mandelbrot visualizer so many times I've lost count. Not because the world needs another poorly written or optimized rainbow-ladybug-simulator, but because it serves as a slightly-non-trivial hello-world. For example, it's the first end-to-end thing I made in Common Lisp. https://git.sr.ht/~amtunlimited/mandelbrot-plot
I just rewrite them all the time as means to get a feeling about programming languages.
I dumped a couple of other stuff on GitHub so that HR people are happy to get a link that they never read anyway.
Then I get back to Java and .NET at the office. :)
https://www.cs.princeton.edu/~appel/modern/
It covers most areas quite alright even if a bit aged.
IMO, writing and running tests in the kernel space is pretty easy thanks to Rust’s flexible testing feature [1].
> You can learn how printf(3) works. You can learn what happens before main().
Yes! It's such a wonderful experience. This is exactly what I most enjoy doing, just seeing how things work, maybe making my own version. I hope you have lots of fun.
You're reimplementing Linux's kernel-userspace binary interface, right? The system call interface is stable and language agnostic, it's really nice. Some pointers for anyone who'd like to know more:
https://man7.org/linux/man-pages/man2/syscalls.2.html
https://man7.org/linux/man-pages/man2/syscall.2.html
https://github.com/torvalds/linux/blob/master/Documentation/...
* Some people bake bread even though there's a good bakery nearby.
* Some people grow gardens even though there's a farmers' market down the street.
* Some people restore old cars even though there's a good restoration professional in town.
* Some people make k8s clusters on rpi even though they could rent that for cheap.
You're making an OS that's linux compatible even though there's already linux. And that is awesome!
The permissive license would also be interesting in some applications.
"I completely agree that rewriting existing large, feature-rich, and robust operating system kernels is not a good idea. However, a question came to my mind: what are the pros and cons of writing a practical UNIX-like OS kernel in Rust from scratch? How does it look like? This is why I started this project."[1]
I would love to see a solid implementation of a real micro kernel based OS based on a message passing core in Rust to become a viable alternative to the now 50 year old Unix architecture. This would give us a possibility of some real progress on the security and process scheduling front, both of which are severely lacking in the Unix model.
https://en.wikipedia.org/wiki/QNX
https://en.wikipedia.org/wiki/Fuchsia_(operating_system)
https://lobste.rs/s/p8bizb/getting_know_fuchsia_google_s_ope...
In fact, micro kernels have a lot of issues and limit the possibilities of what an OS can do! One of the most important issues is they prevents resource sharing in a rich and efficient way between the components of the kernel. They make building rich relationships between kernel data structures and kernel subsystems very messy, complicated, impracticable and very inefficient.
Micro kernels are delusional and people should stop implicitly assuming their superiority.
But this does argue against message passing, since such systems preferentially try to use messages over sharing whenever it's possible.
Rust's memory model largely makes those isolated address spaces unnecessary, and sending a message at runtime basically amounts to copying a pointer. So, with these performance issues resolved, I think it makes sense to look at microkernels again. The reasons why Linus was opposed to writing Linux as a microkernel in the 90's don't necessarily still apply now in 2021.
There may some things the Linux kernel does that can't be efficiently expressed using a message-passing style, but I don't know if that's actually true or not.
I think you're confusing several concepts. Rust's "memory model" is the C++11 memory model. Maybe you meant the borrow checker and the distinction between shared and exclusive references? Yes, those make it much harder to accidentally screw up other parts of your address space, but it's downright trivial to do it on purpose – you just need an unsafe block and std::ptr::write.
There's nothing about Rust that could replace isolated address spaces. Rust is typically compiled to native machine instructions, there's no sandbox or virtual machine that will protect different modules from each other.
What about that WiFi driver with 10,000s of lines of code? Are you absolutely sure there isn't some code in it that accidentally or purposefully leaks the rest of your kernel data structures to the outside world? With drivers in their own isolated address space, you wouldn't have to be.
Can't find it right now, but I recall reading they implemented a network driver with minimal overhead compared to a traditional OS. While they used message passing, a lot of overhead could be eliminated due to the assumptions of safe code.
[1]: https://en.wikipedia.org/wiki/Singularity_%28operating_syste...
The concept is interesting (if maybe untested) it allows to have explicit opt-in sharing while blocking accidental sharing
Rust's compile time checks cannot replace the MMU. There were experimental SAS (single address space) operating systems which didn't depend on the MMU for protection, but they used "managed" languages, i.e., all code was executed in a sandboxed JIT.
I'm also imagining a usage model where all the various components of the microkernel come from the same place and have passed various minimum standards like "doesn't have any unsafe blocks". If some program really needs an unsafe block to do something, then that goes into a (hopefully small) library of heavily-scrutinized functions that do dangerous things but expose safe interfaces.
If we compare with the Linux kernel: in Linux, any part of the kernel can clobber memory in any other part; there's no protection at all. Users generally accept that because the Linux kernel is a pretty well-run project, but still bugs slip in from time to time. If instead you imagine a system where the compiler simply doesn't allow one part of the kernel to clobber memory that belongs to some other part, that would be a significant improvement over what Linux offers.
Anyways, security mechanisms don't necessarily have to be enforced at run-time if they can be enforced at compile-time. That assumes you compile the code yourself with a trusted compiler, or you get the binaries from a trusted source. (Though unfortunately Spectre has become a big problem for compiler-enforced data isolation within one big shared address space; I'm not sure whether compilers these days can do code generation in such a way that it doesn't suffer from sidechannel attacks, or if for the foreseeable future it's just something that software developers have to be wary of.)
In theory you might eventually be able to run device drivers from random people or run arbitrary applications in the same address space as the kernel, but that would require a lot of faith in the compiler to reject anything that could be a security hole. It's an interesting model, though; a pure unikernel approach everything shares the same page tables and there's no need to swap them in or out when context switches happen. And invoking a system call can be as cheap as executing an ordinary function. It might even be inlined by the compiler.
Device drivers make up the largest part of the Linux kernel, and they would by necessity contain lots of unsafe blocks – the compiler can't reason about the interaction with some device. This is typically different from device to device, and there's no way this could be separated into a small library or module.
Look, I agree that a kernel written in Rust could be much safer than one written in C, and even those instances of 'unsafe' blocks would be easier to audit than C's "everything is unsafe". But it still couldn't replace all the advantages you get from micro-kernels with separate address spaces.
For example, it's impossible in Rust to safely unload a dynamically loaded library. The reason for this is, that such a library could contain static data, like string literals, which are of type &'static str. Passing a 'static reference from the library to the program doesn't require 'unsafe', but will leave a dangling reference when you unload the library. Of course, the unloading operation is itself 'unsafe', but that doesn't help you prevent, locate, or track the transfer of a 'static reference.
Another advantage is that there is no such thing as a kernel crash, worst case you get a device driver that needs restarting. The whole loadable module model is broken by design and replicating it in Rust isn't going to solve that particular problem (and it isn't one that Rust is trying to solve).
There have been several other "real micro kernel based OS based on a message passing core", and they too, by almost any metric, have been a failure.
Process scheduling in Linux in 2021 is far, far ahead of anything in any other OS, ukernel or MP-based. The idea that there hasn't been progress in this area is really completely absurd. In fact, the general idea that Linux follows "the now 50 year old Unix architecture" is also pushing the boundaries quite a lot.
BeOS wasn't half bad. The failure was probably mostly commercial. It is true that they moved their networking stack into the kernel, but it's not entirely clear to me that they had to do that, it was just the most expedient way to get acceptable performance and stability given the (programmer-time) resourcing constraints of the project.
Fair point. You could probably say this of other attempted microkernels too, to be even fairer. But that doesn't really change the fundamental point that just cooking up "a better OS design" doesn't lead to it being successful. There's a lot more in play that "mere quality".
you did say "by almost any metric".
But you're right, QnX is the one (probably the only?) micro-kernel that has seen widespread adoption. Almost all of it has been in the context of embedded systems, which doesn't invalidate the substantial success, but it does leave QnX as the exception that proves the rule. It also leaves QnX as yet another example of the general failure of microkernels as general purpose computing platforms. QnX design is excellent for the contexts where it is deployed, but there's a reason you don't run it on PCs, tablets, data crunchers, smartphones and many other sorts of computing contexts.
I used it daily on my PC for years and it was hands down the best environment I've used with distance (mostly compared to IRIX, Windows and Linux). Super responsive, never locks up, no weird delays it just works, 24x7, year after year. The whole throughput argument never worked anyway, that was just Linus talking about something that he didn't have direct experience with, it's latency that is far more important than raw throughput in interactive computing because interactive computing is a real time task.
I've taken an intro to OS class and am currently going through Linux From Scratch [1], which is interesting and is teaching me a lot, but it's more about how to setup a Linux distro using existing packages and not really about reading/writing the code involved.
Any recommendations?
For more general information, there are also a handful of articles and/or blog posts that I've come across when I was getting started that gave a very good introduction to many of the low-level components of how computers work and what the OS needs to do to run them. One of my favorites was Many But Finite [1]. Check out a bunch of his posts about the memory map, the computer boot process [2], cpu rings, etc... Not exhaustive by any means, but good, well-presented info. Many other such sources exist; maybe others will share their favorites.
And if you'd just like to get a more gentle introduction into actual OS code than the monster that is Linux, maybe try looking at some of the BSD's. NetBSD [3] and OpenBSD [4] are remarkably easy to navigate and follow from a code flow perspective, and if you just want to see how things work in a real OS, it's both a good reference and a good place to start.
[0] https://wiki.osdev.org/Main_Page
[1] https://manybutfinite.com/archives
Also, you could try reading the Plan 9 source code.
(I would say the code is what you're after now, but in case you are interested in more of the theory of why it's designed that way, you can check out the research paper here: https://dedis.cs.yale.edu/2010/det/)
ssh root@kerla-demo.seiya.meAlso, as owner of the instances are you able to see what commands people are running? First thing I did was to type hello, and it said hello world to me :)
The mechanism is pretty simple: a Node.js server (running on genuine Linux) listens on tcp:22. Once you connect, it boots a dedicated Firecracker microVM instance and forwards network packets between your SSH client and VM.
Regarding the command history, others (including I) can't see what you type. If you could, it must be a vulnerability.
I think a killer feature would be to emulate the Linux kernel device driver model such that existing C driver modules could be used with Kerla. Further, borrow (no pun intended!) the Linux device driver Rust wrappers from the Rust for Linux kernel project to then enable writing Linux device drivers in Rust.
The question is whether emulating the Linux device driver model (by which I kind of mean mimicking the set of core kernel services needed to support drivers - kernel threads, sync primitives, allocators etc) ends up being a massive task akin to writing the core kernel itself.
I think there might be reasonable trade offs that make the problem more tractable.
Having the ability to have a significantly simpler kernel with Rust’s memory and thread safety guarantees plus the ability to use even a small set of Linux’s massive set of drivers would be very compelling.
While I completely agree that NetBSD's Rump kernel would be a great way to go (I've played with it in the past and quite like it!) I do wish there was a similar initiative bound to the Linux kernel.
The closest I've seen is Octavian Purdila's Linux-Kernel-Library project: https://lkml.org/lkml/2015/11/3/706
Not sure how far they've come though. Perhaps Kerla could be a prime mover for them ?
Heck, it'd probably be easier to emulate the Windows driver architecture - at least those have proper ABI compat.
My take away was that while 'new' driver classes/subsystems do involve more churn in the core kernel support subsystems, there is a clear trend towards convergence in terms of significant change.
If the change truly was significant and continuous then that would impede productisation with the Rust kernel. I'm not saying that the A{B,P}Is are in any way stable but I personally don't think that keeping up will be an intractable problem.
In the spirit of Kerla's authors prime motive - which appears to be hacking for knowledge - this might be something to try, perhaps with a driver class that is known to be more stable.
And better drivers.
The kernel has (deliberately) never had a simple or stable driver ABI, so I think this is potentially a huge task. And then you've also left a C-based attack surface in place.
I'm not sure what the minimal set is if you decide to only support 2021-era hardware. PCIe+SATA+framebuffer+USBhost+USBHID?
I think using the Rust-for-linux wrappers to implement the drivers in Rust would eventually help.
Also, even if initially Linux C drivers were transplanted into Kerla that could be a net benefit if the tradeoff of wide device driver availability vs increase in the trusted compute base was acceptable.
I personally think that tradeoff would be acceptable - if the transplanting ability was tractable. It's a hard one to absolutely agree on, admittedly.
Towards the minimal set of hardware, fair point! I think that one should be able to get quite far with judicious selection of device driver classes and leave the truly in-flux classes alone ?
:)
The only technical problem I can think of is that Rust may not be available for all CPU architectures Linux supports, but this is just speculation on my part having done no research on the matter.
The limitation of that approach is that all internal interfaces (which includes most data structures) have to be C-compatible. You can get much of the safety benefits of Rust, but lose a lot of Rust's ergonomics.
Just like Android Linux compiles just fine with clang for the last five years or so, it now makes use of Rust.
https://source.android.com/setup/build/rust/building-rust-mo...
If upstream ever cares to support clang or Rust, that is another matter.
See: https://lwn.net/Articles/862018/
The issue of Rust's LLVM based compiler toolchain not targeting all CPU archs is intended to be solved by the gcc-rs project.
See: https://lwn.net/Articles/871283/
I think the approach the Rust for Linux project is taking is wise: Not about outright re-writes but more about focusing on those subsystems where Rust's intrinsic safety and security properties helps the most.
Your ordinary userspace apps use std (the standard library) which relies on the alloc crate, and that provides heap allocation which indeed panics if the allocation fails. Because it correctly guesses that your "clever" strategy to handle allocation failure actually isn't and will just triple fault anyway so it should cut to the chase.
The kernel obviously doesn't have std, and Rust for Linux implements its own alloc crate.
Linus' requirement that you can fail memory allocation just means all the calls in alloc that can actually allocate memory now return Result to indicate whether the allocation was successful. This isn't how you'd do things in userspace, but, this isn't userspace so fine.
Naive me - not a kernel dev at all - would argue that returning Result<Memory, AllocationError> is always better, even for userspace because it would allow me to additionally log something or gracefully deal with this.
Even if I don't want to deal with it, I could just `.unwrap()` or `.expect('my error message')` it.
Note: I am not trying to be snarky here, I genuinely don't know and would like to.
If answering this is too complex, maybe you can point me in the right direction so I can ask the right questions to find answers myself? Thanks in any case!
> I could just .unwrap() or .expect('my error message') it.
Panicking can allocate. Allocating can fail. Failing can panic. Panicking can allocate. Allocating can fail. You can bite yourself in the ass like a real Ourobouros.
IMO, a prerequisite to using fallible allocation APIs should be attempting to write your own allocator, handling the weird and wacky problem of initialising a data structure (for the heap) in such a way that if it fails, it fails without allocating but leaves some hint as to what went wrong.
So for `Rust for Linux` they also need to implement a `Result-like` type that is stack only based to solve this issue, right?
If so, cool, thanks, you just made my day by tickling my learning nerves! :)
It actually has to do with `panic!(...)`. When you use `unwrap()`/`expect("...")`, you use the panic macro under the hood; parts of the panicking infrastructure use a boxed trait object which could contain a static string or formatted String or anything else really. The box can allocate if it is not a ZST. I believe the alloc crate's default handler tries to avoid this kind of thing, so that it can't fail to allocate AGAIN in the failure-handling routine. It will likely do a better job than you could.
This is a live issue at the moment, so to go into any more detail I'd have to read a bunch of recent Rust issues/PRs.
They can kill entire misbehaving processes. What are you going to do in your little program, clear a cache whose objects are sprinkled evenly across 150 different pages? You would need more control than you get from blindly using malloc/free/rust_alloc globally. Something like memcached would be able to use these APIs, because it uses its own allocator, and knows enough about its layout to predictably free entire pages at once.
That isn't very accurate. In Rust when programming in no_std, you can (must?) define your own panic handler:
https://doc.rust-lang.org/nomicon/panic-handler.html
Which you would define in the kernel. While I'm not going to speculate on exactly what the implementation would look like, you definitely do not need to "crash" the computer. I haven't done any kernel programming, but I'm guessing the kernel could do some things at that point with shared memory space that is already allocated to deal with this situation and try to recover in some way.
Edit: for example, I just found this in the kerla project: https://github.com/nuta/kerla/blob/88fd40823852a63bd639e602b...
That halts now, but it probably doesn't need to, or could do it conditionally based on the contents of PanicInfo.
> panic handler [...] Which you would define in the kernel. While I'm not going to speculate on exactly what the implementation would look like, you definitely do not need to "crash" the computer.
The panic handler loses so much of the context that crashing the computer is the only thing you can practically achieve. You can't retry an operation generically from with a panic handler, it doesn't know anything about the operation you were attempting. The OOM handler gets a Layout struct only. You could try unwinding or something, but within a syscall handler, I don't see how anything good can come from that. Unwinding in the kernel is simply a terrible idea. What else are you going to do?
Now there is a lot that the allocator could do. If you wanted something to be retriable, it could be interesting if the thing that failed was an async task. If so, that panic info could carry enough information to say, the failure was an OOM, here’s the task that failed, and it is marked as retriable. Yes, this would require a store of tasks somewhere in the kernel. Then based on it being an OOM, see if any memory can be reallocated before retrying, or wait to retry until it is.
This is where theoretically a new async based Rust kernel, especially a micro-kernel, could be interesting. Is stack unwinding in the kernel a bad idea? Maybe. Can it be done in Linux? Maybe not, maybe it’s too much work to track all this information, but I disagree with the conviction with which you right it off.
The Error type currently isn’t in core, but for other reasons, that just got resolved: https://twitter.com/bascule/status/1452029363197784071?s=21
If you don't have any memory your allocations are all failing. When you assemble the log message, the allocation needed to do that fails. Bang, double fault.
Now, often people don't really mean they want allocations to be able to fail generally, they're just thinking about that code they wrote that reads an entire file into RAM. If it was a 100GB file that would be a bad idea. But the best answer is: Guard the allocation you're actually worried about, don't ladle this into the fast path everybody has to deal with on every allocation.
It never occurred to me (being in non-embedded land) that returning an enum as the error or a &'static str instead of a heap structure like String, could also fail.
Seeing that Result isn't part of core, but of std, this makes sense.
Just to tickle my nerve though: theoretically speaking, with your example, it would work, right?
I couldn't allocate 100GB (because OOM or not even enough RAM to begin with) but it could be that the system can allocate the needed memory for error message just fine.
Very interesting.
Add to that that several operating systems will lie about whether you're out of memory, so the 'error' or failure will often not be on the Result() value but come in a SIGKILL instead, it's just adding complexity.
People that are actually worried about it and no how to deal with it, will be coding with a different style and can use the alloc library where/when they need to. (at least when it gets stabilized in Rust)
I've never checked core before, so I did when checking up for this discussion.
I somehow missed Result. Silly me didn't search on that page, but ofc I found it on std
https://doc.rust-lang.org/std/result/index.html
Also thanks for clarifying that values of Result can be stack-only!
Like sit down, figure out all the things you'll want to do on an allocation failure, and once you have determined that you slice a little chunk of memory when you start your app (and maybe _that_ fails and you can't do anything). and when you hit a failure you do your think, then tear stuff down.
Now that I see interactive programs taking a gigabyte and the system being ok, I guess we're in a different regime.
In general Don't assume anything about your global process state just because one allocator fails.
We have a light / temporary fork of the Rust stdlib allocator with fallible allocation support: https://github.com/Rust-for-Linux/linux/tree/rust/rust/alloc
See e.g. https://github.com/Rust-for-Linux/linux/commit/487d7578bd036...
If the kernel is running out of memory, IMHO that's a bug. The kernel is ultimately responsible for memory management right? It needs to prioritize itself over everything else or the system is in trouble.
I still think the project is really cool and worthwhile.
And then, in the longer term, someone ought to write a book about operating system implementation in Rust that ignores Linux interoperability and focuses on readability, maintainability by showing use of Rust's idiomatic style.
Using an incremental approach at least gives us some benefits in the near future.
I'm not even saying a fresh start would be a bad idea. But incrementally replacing parts of Linux seems like a more promising approach, IMHO.
C rewritten in Rust is still very C-like, and requires refactorings that you can't do until it's all Rust.
BTW: There are two GCC-based Rust implementations in the works, so compatibility with exotic platforms is going to be solved.
It’s just not an interesting or unique perspective, and pretty silly to suggest seriously.
Thanks so much for sharing!
However, we have seen *many* memory-safety crashes in the drivers to talk to network interfaces. Those really are full of security vulnerabilities. The most prominent recent one was broadcom's wifi driver having heap buffer overflows allowing remote code execution by anyone within wifi range in 2019[0].
[0]: https://www.bleepingcomputer.com/news/security/broadcom-wifi...
Is it possible to write C code that has the safety properties of Rust code which provides the same functionality? Absolutely, in theory. But will you be able to accomplish this and then guarantee this safety and have the confidence that future code changes won't break those guarantees? That becomes exceedingly more difficult to do, nevermind the extra time and effort it would take. In most cases this would be prohibitively expensive, and you'd just get on with life hoping for the best and mitigating what you can.
In my view, one of the killer features of Rust is its ability to be easily ABI-compatible with C as needed, which I see as a pragmatic feature of Rust that acknowledges the messy software reality of our world and the fact that C (or any other language) isn't just going to be "replaced" by rewriting stuff everywhere, neither anytime soon nor probably ever in totality. This lets us more readily combine the great capabilities of Rust with the enormous volumes of useful code that already exist today without resorting to dogmatic approaches of always thinking we need to rewrite the latter. The "rewrite everything in Rust" idea is fun in an academic kind of way, but in the real world economics will prevent that from happening everywhere, in a similar way that economics usually prevents you from validating and providing certain safety guarantees about your C code.
Mozilla decided to create Rust after years of using C/C++. Why?
Because they spent a fortune fixing the same problems over and over again, and they realized that no matter how clever their developers were, those problems were always going to arise as long as they kept using C/C++.
Not because C/C++ is inherently bad, but because is more flexible than strictly needed in some ways.
Rust was designed to prevent many classes of problems.
1. the toolchain is better. (cargo ala)
2. maintenance of the software is better. (static linked binaries, easier deploys)
3. memory safety.
4. resource consumption (cpu/ram)
when I don't need 4 I use golang (most cases) because of 1, 2, and 3.
and I choose tools (all other things being equal) written in rust for the same reasons.
Must be an awesome language to work on.
If you want Linux to survive, it needs to be re-written from scratch, optimised in Rust and compatible with the existing Linux binaries or drivers.
This project might have a chance.
From: [0]
> TL;DR: I'm writing a Linux clone in Rust just for fun. It does NOT aim to replace the Linux kernel.
A certain someone thought that their hobby OS Minix clone won't be 'big or professional like GNU'.
It's time for 'change'.
Linux is going to survive. And no it does not need to be re-written from scratch in Rust. Someone lied to you at the church of rust.
My assumption is that all titles with Rust get blind upvotes, sure a new kernel is cool but would be more cool if is more then someone toy project, this should probably be posted and upvoted in Rust forums so it gets support from the fans and hopefully the Rust guys will decide what kernel should they focus on and stop promoting every student weekend project here.
How on earth have you come to that one?
Why do you think the Linux developers are bolting on Rust in the Linux kernel themselves? [0]
It has already been admitted.
[0] https://lore.kernel.org/lkml/20210704202756.29107-1-ojeda@ke...
A big problem in the Rust ecosystem is the lack of continued work and support on libraries (understandably, since many are new and aren't commercially supported).
For example, pyroute2 is still better for Linux admin that the equivalent Rust crates. Whilst the latter often fully support async, none of them cover all of the features of pyroute2 nor the easy-to-use API.
Hopefully things will improve over time, it's always a shame to see great projects be abandoned.
This is Hacker News, not Enterprise IT News.
Even Rocket https://github.com/SergioBenitez/Rocket seems to have greatly slowed development unfortunately.
https://github.com/seanmonstar/warp
https://github.com/tokio-rs/axum
re headcrab, https://github.com/headcrab-rs/headcrab/issues/132
Warp and axum are certainly not the only choices, and saying they are the best choices is a matter of opinion. Actix-web, rocket, tide, warp, axum are all great web frameworks.