OpenBSD: Removing syscall(2) from libc and kernel
marc.info
marc.info
https://github.com/openbsd/src/commit/312e26c80be876012ae979...
The ports tree is being cleansed of syscall(2) usage, until they're all gone.
msyscall, pinsyscall, recent mandatory IBT/BTI, xonly. OpenBSD is making some waves, but people aren't really seeing them yet.
I wonder if there could be some way to sign a dynamic library to allow it to create direct system calls and then pass that as a kernel command line argument at boot?
It is not the only operating system in the "unstable kernel interface" group though. Linux is actually the only one with a stable system call interface.
I've written somewhat at length about this:
It's wrong to say that Linux stands alone in having a stable syscall interface. FreeBSD [0] and NetBSD [1] both retain syscall compatibility for old binaries (the former apparently with some exceptions permitted); DragonFly BSD also appears to keep old syscalls in place. In fact, I only know of Windows, OpenBSD, and presumably macOS as mainstream desktop OSes without mostly-stable syscalls.
[0] https://wiki.freebsd.org/AddingSyscalls#Backward_compatibily
[1] https://www.netbsd.org/docs/internals/en/chap-processes.html...
https://forums.freebsd.org/threads/what-is-meant-by-abi.8622...
> I keep reading that the ABI of FreeBSD is constantly changing.
> Someone's system has random crashes and someone else says it's that pesky ABI changing again.
It seems FreeBSD has a reputation of changing binary interfaces. Even when they clarified that they meant the kernel-userspace interface, it was pointed out they're only stable for current releases. There seems to be some emulation but I can't figure out how hard of a guarantee that is.
https://docs.freebsd.org/en/books/handbook/kernelconfig/#ker...
> If the kernel version differs from the one that the system utilities have been built with
> for example, a kernel built from -CURRENT sources is installed on a -RELEASE system
> many system status commands like ps(1) and vmstat(8) will not work.
> To fix this, recompile and install a world built with the same version of the source tree as the kernel.
> It is never a good idea to use a different version of the kernel than the rest of the operating system.
How could the ABI be stable if you have to recompile user space when you update the kernel? Linux is said to be able to run binaries from the 90s.
https://old.reddit.com/r/BSD/comments/8vysxg/a_question_abou...
> FreeBSD breaks syscall ABI compatibility with some regularity.
> Like Windows and unlike Linux, the compatibility layer is considered to be libc, rather than the syscall interface.
I appreciate the feedback. My intention was to maximize the use of space with big fonts even on the desktop but there's no point if the resulting text is unreadable.
Also, FYI, OpenBSD is never going to stop being written in C, and is never going to introduce a language like Rust into the kernel [1], so there's little point in wishing for this.
If you wish to rid yourself of legacy of C (and therefore that of Unix), then OpenBSD, which is Unix (or derived from it) and will always be written in C, is not a good operating system for you.
Edit: Changed to be less rude; it wasn't my intention to be rude.
So in short, it's a pragmatic thing more than anything else.
(Aside: that thread is a little bit outdated by the way, as I do believe there's a reasonably complete Rust coreutils implementation now. Not that I think Rust is a good fit for OpenBSD though – I wish people would stop conflating "safety" with "Rust".)
> Aside: that thread is a little bit outdated by the way, ...
Yeah, that thread was more to show that they really dislike the idea of replacing C with memory safe language, and instead say that people who can't program safely in C shouldn't be programming.
bytevolcano:
> I've always subscribed to the idea that too much safety results in too may idiots, and the same is true for all these "safe" programming languages. "Oh I don't have to write any form of bounds-checking, because the language will do it for me."
Nick Holland:
> Idiots who shouldn't be coding, coding. "safe" languages being trusted to be safe when in the hands of idiots. Like you said.
> The more I see of "safe" languages, the more I love assembly. Most people who call themselves programmers...shouldn't.
bytevolcano again (in the context of memory safe languages):
> A good programmer won't even need these languages in the first place. Case in point, the entire OpenBSD dev team. :)
They'd probably be fine if Go dropped support for OpenBSD altogether; they don't seem to have a high opinion of managed languages or their users.
In the end it's just random people posting stuff; you see that on HN as well, but also "it should literally be forbidden to use non-safe languages" on the other extreme end. bytevolcano seems to have just a few posts to the mailing list. Nick Holland is more involved (nick@), but mostly in documentation and that kind of stuff. There are no actual code commits beyond "add fstab example" and that kind of thing.
I happen to know tedu@ (actual OpenBSD developer) does a lot of Go stuff. I don't want to put words in his mouth, but I'd be surprised if he would come close to agreeing with that kind of stuff.
Unfortunately OpenBSD has always attracted its share of "ima very smart for using openbsd unlike poor dumb linux penguins!11" type of people.
In general I don't think OpenBSD people in general would be against safer languages. I mean, they did a lot of work to make a "safer libc" and it includes Perl in base.
I had assumed you needed permission or else you'd be blocked for posting on a mailing thread, or something like that, so I perceived everyone there except the asker of the question as an OpenBSD developers.
I'm sorry; I didn't realize I was spreading borderline libel.
> They … dislike modern programming languages.
You based this statement on the words of random commenters. And yet OpenBSD maintains packages for these modern programming languages, and OpenBSD developers spend their free time actively sending patches and bug reports to these language projects to keep them running on OpenBSD. Why not look up the opinions of those people before writing sentences like that?
It's interesting, though, that at least initially, the only examples of memory-safe languages were ones like Java or Python, which are both very high-level and come with a lot of runtime tradeoffs that would practically make them unsuitable to tools like ls or cat that need to be run quickly and a lot. (Also interesting: this never seems to be given as a reason not to use memory-safe languages.) Later on Rust and Ada are discussed, but it feels much more cursory, except in the final comment.
The big reason to use C seems to be that it provides a signal that the developer is competent - you need to be at least good enough to learn C in order to contribute to OpenBSD. And more specifically, if you would rather write code in a non-C, memory-safe language, you are probably a bad developer who should not contribute to OpenBSD. To me, this intuitively feels like quite a weak argument (are there so many people clamouring to contribute to OpenBSD that they need this kind of gatekeeping mechanism?) but it's pretty much the main one that gets repeated. The more convincing argument at the end is that, practically speaking, most of the benefits of memory-safe, low-level languages like Rust or Ada can be provided by static C linters/checkers, and so they don't add much value overall. This seems very much the opposite of what some other companies are saying, e.g. Google and Microsoft, which is that even with static analysis of C/C++, Rust is still dramatically reducing the number of memory safety issues they're running into.
I find it interesting that they don't bring up what seems to be the more obvious reasons not to bring in memory-safe languages, which are mainly the classic "rewrites are hard, here be dragons" arguments - the existing tools have been hardened through years of use, and rewriting them is more likely to add new bugs than remove old ones; new languages/tools means new knowledge which means having people on hand familiar with the new languages to understand what's going on; reimplantation tends to be slow initially, which eats resources and prevents ongoing maintenance and development; etc.
My main takeaways from that discussion are (a) I really hope I never have to work with the OpenBSD maintainers, and (b) their main arguments seem more about maintaining the exclusivity of their club than about the technical details of any transition.
It's pretty much on-brand for them. The OpenBSD project started when the founder was kicked off of the NetBSD core team, allegedly for being rude and abrasive to users. At one point, both the FreeBSD[0] and NetBSD[1] mailing lists blocked traffic from this individual because they claimed he threatened to aggressively spam their mailing lists.
0 - https://mail-archive.freebsd.org/cgi/getmsg.cgi?fetch=56044+... 1 - http://mail-index.netbsd.org/current-users/1996/10/20/0004.h...
Forcing usage of libc is actually an effort towards type safety. `syscall` just takes a syscall number and an unspecified amount of arguments of unconstrained type. It's the opposite of being type safe.
Forcing libc has absolutely nothing to do with type safety. How do you think they invoke syscalls in libc? There's no more type checking of syscalls there than there is in Go's standard library. From a user's point of view it's identical, except you don't have to worry about endless glibc version errors with Go.
The libc can enforce some minimal sanity checks on the syscall arguments. For example the number of arguments and some early validation to avoid entering the kernel if failure can be foreseen. But the main utility is to make it more difficult to use Return Oriented Programming to build exploits.
Something they wouldn't have to do if they'd heeded the warnings they got since they first started going raw syscalls on non linux systems.
But as usual, go is uniquely american, only doing the right thing after it has tried everything else.
No, you could not.
On pretty much every system but linux the libc (or equivalent) is the officially supported API to the kernel. Bypassing it is not supported and thus definitionally can not be sane, that's like declaring that doing the sane thing is avoiding hardcoding against the front door and entering your home through the roof instead. The front door is what you're supposed to use, do hardcode against it. Same with libc.
And Go has been going "ouch" then putting its fingers straight back in the socket from the start, as it keeps trying to work around libcs on all the platform where it is not supported.
Edit: NT with its split OS API dll and separate libc is its own can of worms. You can have multiple instances of libc in the same address space, which among other things implies that you cannot safely malloc() in one module and free() in another, and don't even think about passing FILE* around.
That might be your new point after having yeeted the goalposts out, but it's definitely not your original statement.
> and appeals to tradition don't make it less insane.
This is not an appeal to tradition, it's how things work. If the authors and maintainers of the system tell you something is the supported API, wilfully bypassing that supported API is not "the sane thing", especially when you complain that your bypass blows your face off down the line, then you're an asshole too.
That doesn't seem like anything common to an American way of doing things. What an odd statement.
Perhaps this is why most innovations are American. We don't automatically fall for the bully pulpit of the gnostic class.
> Something they wouldn't have to do
Yea, but they'll get it done anyways, and the language will continue to be excellent. I'm sure Google can absorb the engineering challenge without subtracting anything from us or other languages.
Which is a reasonable thing, given that the libc interface is defined by a widely used IEEE standard while the kernel interface is not.
And that's just the platforms which technically enforce it. Linux is essentially the only platform which actually supports raw syscalls, in the sense that it's considered a normal system API.
That sounds like an additional knob to tweak. The OpenBSD project is famously opposed to such knobs.
It can dynamically link the system32 DLLs in Windows and probably could OpenBSD's crt, but dynamic linking in general doesn't work. `dlopen` is no-op, and this has made Cosmo graphics extremely difficult so far.
As for why the userspace system service library is part of libc instead of being a separate libsyscall or libkernel, that probably is due to Unix being a C operating system--written at a time when most operating systems also came with their own system language. It's definitely not the case for all OS's that the C runtime library is the same as the libsyscall/libkernel--most notably, on Windows, the former is MSVCRT*.dll and the latter is kernel32.dll (or ntdll.dll if you're looking very specifically at syscalls themselves, but ntdll.dll is largely an unstable interface).
Thanks!
A big motivation for it was security posture; it means that Microsoft can now ship security updates to UCRT that everyone can rely on rather than a ton of extra surface area through various multiple versions of runtime libraries.
It is different, uniquely so, on linux: on most unices the kernel and libc are developed as two sides of an entire system, both being updated in lockstep when the system is updated. As such there is no real concern about keeping the syscalls stable, if you need to change it on the kernel side you update it to match on the libc side and you're done, everybody is supposed to use the libc.
Not so on linux, the kernel and the libc (most commonly glibc) are developed by entirely different groups which don't necessarily like or communicate with one another. As a result, on linux syscalls are a stable API, and direct syscalls are an officially supported method of interaction. In fact its sometimes necessary as the libc might decide not to expose a syscall.
Do FreeBSD [0] and NetBSD [1] not fall under "most unices"? They similarly retain backward compatibility in their syscall interface, so that old binaries with old libcs can run on newer kernels. Linux does not stand alone here.
[0] https://wiki.freebsd.org/AddingSyscalls#Backward_compatibily
[1] https://www.netbsd.org/docs/internals/en/chap-processes.html...
I guess people don't do it because the difference is minimal and they use a lot of other features from the C runtime too.
EDIT: Ops. Not on BSD! The entire thread is about BSD and here I am mindlessly talking about Linux.
[0] I seem to remember that this changed in a recent Windows version, but I couldn't immediately find a source.
[1] msyscall(2) or, if you don't have an OpenBSD system at hand, https://man.openbsd.org/msyscall
Turns out, having a dedicated syscall instruction and trap pathway is just better design. Unlike a regular exception, this is a deliberate change of control to the kernel, so you can enforce a much stronger ABI requirement. In particular, you can define it to use a standard function call ABI with respect to preserved and non-preserved registers making it literally look like a standard function call.
For similar reasons, having a dedicated hardware pathway like on x86-64 is also just better design. System calls are a synchronous, voluntary transfer of control that is expected to return in contrast to (1) interrupts which are a asynchronous involuntary transfer of control and (2) instruction stream exceptions which are a synchronous involuntary transfer of control with no guarantee of return. This fundamental distinction can be leveraged for more efficient and simpler implementations.
https://devblogs.microsoft.com/oldnewthing/20041215-00/?p=37...
IIRC, PDP11 and VAX used traps as the way to call the supervisor, and it was pretty cheap.
But interesting nevertheless.
The list would be editable, but there would be an operation which seals it from further editing.
With this feature, we could load a new plugin into a C program with dlopen, and it would not be able to make syscalls of its own, other than through the existing C library, no matter how prim and proper its call sequences look like.
Possibly, ranges could be used instead: syscalls can originate from several registered address ranges, and that's it. Programs that create dynamic code on the fly, which can make system calls, could allocate trampolines in a registered area, for that purpose.
This work is removing a very indirect morph of syscall where the arguments/sysnum are in a struct in memory, making it harder for exploits to invoke weird versions of syscalls on their own terms.
Under the hood, it's all assembly language instructions invoking an operating system written in C. Operating systems care very much about types, but can't offer type safety. Programming languages can.
This is awesome.
Also, it’s rarely used in my experience. I wonder what kind of code would even notice?
> I recognize we can never completely remove all mechanisms they use. However, I hope I am forcing attack coders into using increasingly more complicated methods
It's honestly so ridiculous an OS that claims to have security a a focus refuses to add even some sort of basic MAC/RBAC implementation. Even both OSX and Windows have had something for ages now.
OpenBSD is like a very hardened safe, made of steel and huge bolts and locks. Very polished, very smooth and hard surface.
MAC/RBAC is like having security officers, interviews, checking of IDs, filling in forms and getting an OK from ones boss before performing work someplace in the building and so on.
Both these things can be good. But OpenBSD was always about making a small system as hardened as possible. Evidently, they aren't completely done yet with making the core as hardened as possible.
Windows has great architecture, but lacks instead severerly this hardness which OpenBSD possesses. What use is MAC/RBAC if someone can gain kernel access with a 0 day exploit?
Nah they won't. The devs have an irrational resistance to the very idea.
I disagree with your analogies. OpenBSD has a focus on auditing to remove all bugs, which is great, but they provide very little to help prevent what can be done if a bug is exploited, and they've certainly had no shortage of serious bugs.
> What use is MAC/RBAC if someone can gain kernel access with a 0 day exploit?
Kernel exploits are pretty rare. Most exploits are in userland.
Their safe is very hard, but once you are in, you are in. And I think I agree with your assessment, they aren’t likely to start creating MAC/RBAC solutions.