[tialaramex]

Much worse than that, even memory-safe languages like (safe) Rust, and the inevitable suggestion of AUTOSAR and so on aren't the answer. To properly answer your demand for a "practical way to ensure user input does not cause mischief" you want a drastically less capable language which cannot even in principle express the programs that should not exist, that's exactly what WUFFS is for.

https://github.com/google/wuffs

This sort of bug can't happen in WUFFS because you can't express the idea "corrupt the heap memory" even if you desperately wanted to. The tell-tale sign of such languages is that they are not general purpose languages, because those are able to express a wide variety of stupid things you don't want to do.

[dataflow]

Could you expand a bit on what you mean regarding memory-safe Rust not being the answer to "cannot corrupt the heap memory"?

[tialaramex]

Sure, for example a Rust program is allowed to open, read and write files. On Linux one of the files it's allowed to open /proc/self/mem points directly into its own heap, another is a list of what's in that heap and where - it can use these to scrawl on the heap and cause havoc.

That's the price of being a General Purpose programming language. We don't know if you might want to scribble on your own heap, or delete all the files labelled "Important, DO NOT DELETE" or mail a copy of the password database to a throwaway account, and so you can do all those things. Those Linux files pointing into process address space aren't a mistake, I wrote code that needs them (and then I ported it to safe Rust months ago) but with great power comes great box office potential or something like that.

Now you might say, "I'm sure I won't get something so obvious wrong", but the trouble is that's what the people who wrote this GitHub code apparently thought too. Hence I say we should use specialised languages with a deliberately narrower scope where this category of mistake is impossible.

WUFFS as it stands would be pretty exhausting to write a Markdown parser in because WUFFS doesn't believe in strings, at all. But it's already a better fit for this problem than C++ because the worst case scenario can't happen.

[dataflow]

Oh interesting, thanks a lot for elaborating. Question: why would a memory-safe Rust program open a sensitive file like /proc/self/mem? Wouldn't that require corrupting memory somehow, which presumably the Rust compiler would be proving impossible? Or are you suggesting the file name might come from user input, and the user input would somehow find an exploit to redirect it to that file? Would that be possible/likely in any way for something like parsing?

[nextaccountic]

That's because opening and writing to arbitrary files is marked a safe operation in Rust, and sometimes Rust programs open files whose filename were supplied by untrusted, potentially malicious input. And as said in the other comment, this can lead to UB: https://github.com/rust-lang/rust/issues/32670 (it's Linux here that is at fault, really, for offering a hugely unsafe API through the filesystem), so, we could imagine a paranoid (but technically correct) version of Rust where the APIs for opening and/or writing to a file were marked as unsafe.

But, since the operations in actual Rust are marked as safe, the compiler doesn't provide any checks here: we can cause UB in code without any unsafe { }. Moreover, checking if the path starts with /proc isn't enough to make the UB go away: procfs can be mounted on any dir, there can be bind mounts further obscuring the file resolution, etc.

This means that if you really care about memory safety (and correctness in general), the precise way you setup your environment is also critical, down to the minimum details. It's like your Dockerfile had a metaphorical unsafe { } block around it: in a system that doesn't mount /proc you just closed a whole host of bugs, and a buggy system that mounts procfs in other dirs may cause arbitrary havok. (note that mounting procfs is a privileged operation)

There are low level languages that, unlike Rust, completely prevents memory safety errors, like ATS. In ATS you can deal with pointers and pointer arithmetic (like in C or Rust) but to follow a pointer you need to provide a mathematical proof that they are valid. This is enough if we consider the program in isolation, but programs are never run in isolation. A proper mathematical proof of memory safety needs to consider ALL software running in the system, globally: then everything is mathematically verified, and the build step can just reject an unsound system setup.

That way we could theoretically be more precise about our memory safety guarantees: opening and writing to a file is safe, but only if procfs isn't mounted. If procfs is mounted anywhere, then this may go wrong: we need to prove we aren't doing something bad. This means that in a system where sysadmins can just log in and mount random stuff, writing to files must be unsafe!

Of course that's not very practical. It would be cumbersome to prove you're not doing /proc shenanigans every time you messed with files. And arguably, any program that open arbitrary filenames that came from untrusted input is buggy anyway. You should always do filename validations, specially to confine some input to some directory (when applicable), avoiding paths with ../ that escape it, for example. And, any setup that mounts procfs outside of /proc is irreparably broken. We don't have a tool to automatically check for such issues, but if those two things are followed, we won't have UB here.

How to do better than that? We need better system-level APIs, in which operations that are "obviously" safe can really be 100% memory safe all the time.

[dataflow]

I get all this from a mathematical standpoint, but from a programming standpoint is this really that hard to solve in something like Rust? Like for this class of problems, can't you just (say) route every I/O syscall (like open()/openat()/etc.) through a custom layer that properly validates the path? Like making sure none of the directories cross into procfs or all that? At that point you just need to prove the correctness of the path validation (which should be pretty doable by hand) and let the Rust compiler take care of proving memory safety in the program itself... right?

[nextaccountic]

Rust is too much of a low level language for that. In Rust, operations typically map straightforwardly to system operations, and the layer on top is pretty minimal.

There's some exceptions though. Rust puts a mutex for accessing stdio (to prevent interleaved, broken output when calling println!() from many threads). Rust also has a mutex for accessing the environment [0]

But in this case... see, this /proc thing is pretty niche. With some bizarre combinations of brokenness (either on your application, or on the system setup, or both) it can indeed lead to UB, which can be a serious security bugs, remote code execution even, but it's very, very rare that some program needs to care about this in practice. Rust is a practical language and I think in this case the line it drew was quite sensible.

Sadly, this means that Rust safety guarantees aren't absolute, but Rust doesn't even have yet a precise mathematical definition of UB anyways, so we can't even in principle start formalizing this enough for this to matter.

Rust is also in a though spot here because as I said, there is hardly a foolproof way to check at runtime whether you're accessing procfs: checking for /proc in the path is just an heuristic that doesn't actually close the UB loophole. You would need to inspect all mounts to check for procfs and bind mounts (and also check for symlinks, hardlinks, etc),

Maybe another route is to open the file as normal, but then do a stat and check the device and inode: if the device is a procfs device, and the inode is a bad one, return an error (if you want to open it anyway, you need to use an unsafe API). Or, if we can't check because some system setup shenanigans, default to returning an error. This could be an useful crate for a sufficiently paranoid application, but might not make the cut for the Rust stdlib, even though it ostensibly closes a safety loophole. (or it just might; maybe this should be proposed)

[0] Unlike the stdio one, the environment mutex is actually critical for safety in Rust programs. But you can break this safety by calling C code that reads the environment in a non-threadsafe manner without passing through the Rust mutex. So, accessing the environment from many threads can easily lead to UB, even though the operation is marked as safe in Rust. This can still be sound from Rust's pov because calling C APIs is unsafe, so you "just" need to guarantee that all C code isn't accessing the env behind your back. Except that there's some bad APIs like getaddrinfo that may implicitly access the environment, and tons of libraries call that, so in practice many C libraries can't be given a safe Rust interface. See https://doc.rust-lang.org/std/env/fn.set_var.html and https://internals.rust-lang.org/t/synchronized-ffi-access-to... and https://github.com/rust-lang/rust/issues/27970

Note that for many well behaved programs, environment variables are read only at program startup (before creating any threads), saved to a config struct, and this struct is passed around as needed. This usage can be safe even without the mutex. So an alternative design would be to prevent calling some APIs once you have created threads. To do that you maybe could add some way of tracking at type level whether the program is single threaded or multi-threaded (perhaps with session types, or the typestate pattern: basically a type-level state machine, where spawning a thread makes you go to the multi-threaded state if you're not already there). Also, in the single-threaded state, Arc could be automatically converted to Rc as well, Mutex converted to RefCell, etc. It would be interesting to see a language designed around this.