Address sanitizer won’t panic/crash your program on all memory safety violations. Attackers know how to achieve remote code execution in processes running Asan. Asan’s docs specifically call out that you should not use it in prod. In other words, Asan is not memory safe. It’s just a bug finding tool.
Fil-C will panic your program, or give some kind of memory safe outcome (that is of no use to the attacker) in all of the cases that attackers use to achieve remote code execution. In other words, Fil-C is memory safe.
The fact that Fil-C achieves memory safety using runtime checks doesn’t make it any less memory safe. Even rust uses runtime checks (most importantly for array bounds). And, type systems that try to prove safety statically often amount to forcing the programmer to write the checks themselves.
> Because thread 2 can observe a mismatch between a pointer and its capability, an attacker controlled index into P2 from thread 2 can access memory of an object other than the one to which P2 points.
Under Fil-C’s memory safety rules, „the object at which P points” is determined entirely by the capability and nothing else.
You got the capability for P1? You can access P1. That’s all there is to it. And the stores and loads of the capability itself never tear. They are atomic and monotonic (LLVM’s way of saying they follow something like the JMM).
This isn’t a violation of memory safety as most folks working in this space understand it. Memory safety is about preventing the weird execution that happens when an attacker can access all memory, not just the memory they happen to get a capability to.
> He claims Java has the same problem. It does not.
It does: in Java, what object you can access is entirely determined by what objects you got to load from memory, just like in Fil-C.
You’re trying to define „object” in terms of the untrusted intval, which for Fil-C’s execution model is just a glorified index.
Just because the nature of the guarantees doesn’t match your specific expectations does not mean that those guarantees are flawed. All type systems allow incorrect programs to do wrong things. Memory safety isn’t about 100% correctness - it’s about bounding the fallout of incorrect execution to a bounded set of memory.
> That's correct but irrelevant: thread 2 has P2 and it's paired with the wrong capability. Kaboom.
Yes, kaboom. The kaboom you get is a safety panic because a nonadversarial program would have had in bounds pointers and the tear that arises from the race causes an OOB pointer that panics on access. No memory safe language prevents adversarial programs from doing bad things (that’s what sandboxes are for, as TFA elucidates).
But that doesn’t matter. What matters is that someone attacking Fil-C cannot use a UAF or OOBA to access all memory. They can only use it to access whatever objects they happen to have visibility into based on local variables and whatever can be transitively loaded from them by the code being attacked.
That’s memory safety.
> He doesn't acknowledge corner cases like the one I've described.
You know about this case because it’s clearly documented in the Fil-C documentation. You’re just disagreeing with the notion that the pointer’s intval is untrusted and irrelevant to the threat model.
You don't always get a panic. An attacker who can get a program to access an offset he controls relative to P2 can access P1 if P2 is torn such that it's still coupled, at the moment of adversarial access, with P1's capability. That's dangerous if a program has made a control decision based on the pointer bits being P2. IOW, an attacker controlled offset can transform P2 back into P1 and access memory using P1's capability even if program control flow has proceeded as though only P2 were accessible at the moment of adversarial access.
That can definitely enable a "weird execution" in the sense that it can let an attacker make the program follow an execution path that a plain reading of the source code suggests it can't.
Is it a corner case that'll seldom come up in practice? No. Is it a weakening of memory safety relative to what the JVM and Rust provide? Yes.
You are trying to define the problem away with sleigh-of-hand about the pointer "really" being its capability while ignoring that programs make decisions based on pointer identity independent of capability -- because they're C programs and can't even observe these capabilities. The JVM doesn't have this problem, because in the JVM, the pointer is the capability.
It's exactly this refusal to acknowledge limitations that spooks me about your whole system.
> An attacker who can get a program to access an offset he controls relative to P2 can access P1 if P2 is torn such that it's still coupled, at the moment of adversarial access, with P1's capability
Only if the program was written in a way that allowed for legitimate access to P1. You’re articulating this as if P1 was out of thin air; it’s not. It’s the capability you loaded because the program was written in a way that let you have access to it. Like if you wrote a Java program in a way where a shared field F sometimes pointed to object P1. Of course that means loaders of F get to access P1.
> That can definitely enable a "weird execution"
Accessing a non-free object pointed by a pointer you loaded from the heap is not weird.
I get the feeling that you’re not following me on what „weird execution” is. It’s when the attacker can use a bug in one part of the software to control the entire program’s behavior. Your example ain’t that.
> Is it a corner case that'll seldom come up in practice? No. Is it a weakening of memory safety relative to what the JVM and Rust provide? Yes.
I don’t care about whether it’s a corner case.
My point is that there’s no capability model violation and no weird execution in your example.
It’s exactly like what the JVM provides if you think of the intval as just a field selector.
I’m not claiming it’s like what rust provides. Rust has stricter rules that are enforced less strictly (you can and do use the unsafe escape hatch in rust code to an extent that has no equal in Fil-C).
I think his argument is that you can have code this:
user = s->user;
if(user == bob)
user->acls[s->idx]->has_all_privileges = true;
And this happens:
1. s->user is initialized to alice
2. Thread 1 sets s->idx to ((alice - bob) / sizeof(...)) and s->user to Bob, but only the intval portion is executed and the capability still points to Alice
3. Thread 2 executes the if, which succeeds, and then gives all privileges to Alice unexpectedly since the bob intval plus the idx points to Alice, while the capability is still for Alice
It does seem a real issue although perhaps not very likely to be present and exploitable.
Seems perhaps fixable by making pointer equality require that capabilities are also equal.
1. I’m not claiming that Fil-C fixes all security bugs. I’m only claiming that it’s memory safe and I am defining what that means with high precision. As with all definitions of memory safety, it doesn’t catch all things that all people consider to be bad.
2. Your program would crash with a safety panic in the absence of a race. Security bugs are when the program runs fine normally, but is exploitable under adversarial use. Your program crashes normally, and is exploitable under adversarial use.
So not only is it not likely to be present or exploitable, but if you wrote that code then you’d be crashing in Fil-C in whatever tests you ran at your desk or whenever a normal user tried to use your code.
But perhaps point 1 is still the most important: of course you can write code with security bugs in Fil-C, Rust, or Java. Memory safety is just about making a local bug not result in control of arbitrary memory in the whole program. Fil-C achieves that key property here, hence its memory safe.
In my understanding the program can work correctly in normal use.
It is buggy because it fails to check that s->idx is in bounds, but that isn't problem if non-adversarial use of s->idx is in bounds (for example, if the program is a server with an accompanying client and s->idx is always in bounds when coming from the unmodified client).
It is also potentially buggy because it doesn't use atomic pointers despite comcurrent use, but I think non-atomic pointers work reliably on most compiler/arch combinations, so this is commonplace in C code.
A somewhat related issue if that since Fil-C capabilities currently are only at the object level, such an out-of-bounds access can access other parts of the object (e.g. an out-of-bounds access in an array contained in an array element can overwrite other either of the outer array)
It is true though that this doesn't give arbitrary access to any memory, just to the whole object referred to by any capability write that the read may map to, with pointer value checks being unrelated to the accessed object.
If you set the index to `((alice - bob) / sizeof(...))` then that will fail under Fil-C’s rules (unless you get lucky with the torn capability and the capability refers to Alice).
Exactly. I agree that this specific problem is hard to exploit.
> Seems perhaps fixable by making pointer equality require that capabilities are also equal
You'd need 128-bit atomics or something. You'd ruin performance. I think Fil-C is actually making the right engineering tradeoff here.
My point is that the way Pizlo communicates about this issue and others makes me disinclined to trust his system.
- His incorrect claims about the JVM worry me.
- His schtick about how Fil-C is safer than Rust because the latter has the "unsafe" keyword and the former does not is more definitional shenanigans. Both Fil-C and Rust have unsafe code: it's just that in the Fil-C case, only Pizlo gets to write unsafe code and he calls it a runtime.
What other caveats are hiding behind Pizlo's broadly confident but narrowly true assertions?
I really want to like Fil-C. It's good technology and something like it can really improve the baseline level of information security in society. But Pizlo is either going to have to learn to be less grandiose and knock it off with the word games. If he doesn't, he'll be remembered not as the guy who finally fixed C security but merely as an inspiration for the guy who does.
All I’m really hearing is that this guy rubs you the wrong way, so you’re not going to give him the benefit of the doubt that you’d give to others.
I mean, maybe you’re right that his personality will turn everyone off and none of this stuff will ever make it upstream. But that kind of seems like a problem you’re actively trying to create via your discourse.
> Only if the program was written in a way that allowed for legitimate access to P1. You’re articulating this as if P1 was out of thin air; it’s not.
My program:
if (p == P2) return p[attacker_controlled_index];
If the return statement can access P1, disjoint from P2, that's a weird execution for any useful definition of "weird". You can't just define the problem away.
Your central claim is that you can take any old C program, compile it with Fil-C, and get a memory-safe C program. Turns out you get memory safety only if you write that C program with Fil-C's memory model and its limits in mind. If someone's going to do that, why not write instead with Rust's memory model in mind and not pay a 4x performance penalty?
> that's a weird execution for any useful definition of "weird".
Weird execution is a term of art in the security biz. This is not that.
Weird execution happens when the attacker can control all of memory, not just objects the victim program rightly loaded from the heap.
> Your central claim is that you can take any old C program, compile it with Fil-C, and get a memory-safe C program.
Yes. Your program is memory safe. You get to access P1 if p pointed at P1.
You don’t get to define what memory safety means in Fil-C. I have defined it here: https://fil-c.org/gimso
Not every memory safe language defines it the same way. Python and JavaScript have a weaker definition since they both have powerful reflection including eval and similar superpowers. Rust has a weaker definition if you consider that you can use `unsafe`. Go has a weaker definition if you consider that tearing in Go leads to actual weird execution (attacker gets to pop the entire Go type system). Java’s definition is most similar to Fil-C’s, but even there you could argue both ways (Java has more unsafe code in its implementation while Fil-C doesn’t have the strict aliasing of Java’s type system).
You can always argue that someone else’s language isn’t memory safe if you allow yourself to define memory safety in a different way. That’s not a super useful line of argumentation, though it is amusing and fun
Sorry to intrude on the discussion, but I have a hard time grasping how to produce the behavior mentioned by quotemstr. From what I understand the following program would do it:
int arr1[] = {1, 2, 3, 4, 5};
int arr2[] = {10, 20, 30, 40, 50};
int *p1 = &arr1[1];
int *p2 = &arr2[2];
int *p = choose_between(p1,p2);
//then sometime later, a function gets passed p
// and this snippet runs
if (p == p2) {
//p gets torn by another thread
return p; // this allows an illegal index/pointer combo, possibly returning p1[1]
}
Is this program demonstrating the issue? Does this execute under Fil-C's rules without a memory fault? If not, could you provide some pseudocode that causes the described behavior?
Fil-C lets programs access objects through the wrong pointer under data race. All over the Internet, you've responded to the tearing critique (and I'm not the only one making it) by alternatively 1) asserting that racing code will panic safely on tear, which is factually incorrect, and 2) asserting that a program can access memory only through its loaded capabilities, which is factually correct but a non sequitur for the subject at hand.
You're shredding your credibility for nothing. You can instead just acknowledge Fil-C provides memory safety only for code correctly synchronized under the C memory model. That's still plenty useful and nobody will think less of you for it. They'll think more, honestly.
> asserting that racing code will panic safely on tear, which is factually incorrect
Try it. That’s what happens.
> through its loaded capabilities, which is factually correct but a non sequitur for the subject at hand.
It’s literally the safety property that Fil-C guarantees.
Safety properties provided by languages aren’t about preventing every bad thing that users can imagine. Just because the language does something different than what you expect - even if it allows you to write a program with a security bug - doesn’t mean that the language in question isn’t memory safe.
> You're shredding your credibility for nothing. You can instead just acknowledge Fil-C provides memory safety only for code correctly synchronized under the C memory model.
Fil-C provides memory safety even for incorrectly synchronized code. That safety guarantee is easy to understand and easy to verify: you only get to access the memory of the capability you actually loaded. You’re trying to evade this definition by getting hung up on what the pointer’s intval was, and your PoC uses a pointer comparison to illustrate that. You’re right that the intval is untrusted under Fil-C rules.
I’m not going to downplay the guarantees of my technology just to appease you. Whether or not you find me credible is less important to me than being honest about what Fil-C guarantees.
> If you set the index to `((alice - bob) / sizeof(...))` then that will fail under Fil-C’s rules (unless you get lucky with the torn capability and the capability refers to Alice).
In the comment above, you write, referring to a fault on access through a torn capability
> Try it. That’s what happens.
Your position would be clearer if you could resolve this contradiction. Yes or no: does an access through a pointer with an arbitrary offset under a data race that results in that pointer's capability tearing always fault?
> You’re right that the intval is untrusted under Fil-C rules.
Can Fil-C compile C?
You can't argue, simultaneously,
1) it's the capability, not your "intval", that is the real pointer with respect to execution flow and simultaneously, and
2) that Fil-C compiles normal C in which the "intval" has semantic meaning.
Your argument is that Fil-C is correct with respect to capabilities even if pointers are transiently incorrect under data races. The trouble is that Fil-C programs can't observe these capabilities and can observe pointers, and so make control flow decisions based on these transient incorrect (you call them "untrusted") inputs.
Can you show an actual minimal C program which has this problem? I’m trying to follow along here, but it’s very hard for me to understand the exact scenario you’re talking about.
> Rust has a weaker definition if you consider that you can use `unsafe`
I don't see it. Rust makes the same guarantees regardless of the unsafe keyword. The difference is only that with the unsafe keyword you the programmer are responsible for upholding those guarantees whereas the compiler can check safe Rust.
But the definition is what we're talking about, not whether you make mistakes. Of course it's important that safe Rust is checked by the compiler, but that's crucially not part of how safety is defined.
I would guess that somebody more on the pulse of C's safety efforts could tell you whether they have a definition of memory safety for C or whether they're comfortable with an existing definition from somebody else.
I'm curious what you make of quotemastr's point about a race causing a mismatch between the pointer's capability and its index. First off, in your estimation can this realistically be exploited to wreak havoc on extant C programs compiled using Fil-C? Second, is such a mismatch able to happen in safe Rust? Third, is such a mismatch able to happen in unsafe Rust?
Edit: clarification to narrow the question even further
"Wreak havoc" is a very vague claim. Instinctively the tearing feels like something very difficult to usefully exploit, but, we know historically that the only people who can reliably tell you whether it was difficult are the attackers actually trying to do it. Don't believe the defenders.
AIUI this capability versus value distinction is a Fil-C thing. So, that's not a thing in Rust at all. In Safe Rust the pointer types, which is what we care about here, aren't very interesting because safe Rust can't dereference them, safe Rust is fine with you making a pointer from the word "LAUGHING" (not a pointer to the string, just the literal bytes in ASCII, but treated as a pointer) or from just some random bytes you found in a data file, because it's not allowed to dereference them so, cool, whatever, no harm no foul.
In unsafe Rust we're allowed to dereference valid pointers, but it's our job to ensure we obey that rule about validity, it being our job to obey rules is what "unsafe" means. So, that silly "LAUGHING" pointer isn't valid, it's just pointer-shaped toxic material. Even if, by coincidence, a pointer you have happened to have the same address as that pointer, in both C and Rust it's not OK to just go around dereferencing invalid pointers, they are not offsets into an imaginary huge array of all memory even though some C programmers act like they are.
Ignoring the Fil-C specific capabilities, in Rust the tearing issue is a matter of synchronization, which is something Rust cares about as part of delivering "fearless concurrency". Rust's marker traits Send and Sync are good place to start learning about that. Yes, we could unsafely implement these marker traits in unsafe Rust when we shouldn't, and thus enable what I imagine you'd call havoc.
So, mostly the problem is that your question is (unintentionally) too vague to answer well but I hope I was at least somewhat helpful.
What I mean is, what’s to stop us saying that C upholds all the same guarantees that Rust does and that it’s the programmer that’s responsible for upholding them (just as the programmer is responsible in the case of Rust code marked ‘unsafe’)? This seems like a semantic game to avoid acknowledging that unsafe Rust comes with some of (though not all) of the same risks as C code.
In short, the definitions are not important. What matters are the risks that you do or don’t run. And if your Rust code contains unsafe blocks, you are running risks that you wouldn’t be if you used Fil-C, which has no such escape hatch. (Of course this goes both ways – your Fil-C code is more likely to fail, safely, with a runtime error due to a mistake that Rust would have caught at compile time.)
Real world C software does not read like software written by people who are in fact upholding those guarantees you say C could equally have. It reads as though they think such a guarantee is a joke or an irrelevance. It's not rare for me to run into people who think C's pointers are just indexing into a massive array of all RAM (or its equivalent on today's systems with virtual addressing), that's not just not in the same ballpark as a safe C program, that's playing a different sport on another continent.
You seem to be suggesting that a language being safe or unsafe is a social contract rather than a technical property of the language.
>And do you say that C offers these guarantees ?
No, that would be silly, and it's an illustration of why it is silly to say that a language guarantees X if it is the programmer who must check that X holds. If we go down that route (which, to repeat, would be silly), then we can make C safe without any technical changes just by adding some language to the standard saying that C programmers are obliged to ensure that their code maintains a certain list of invariants. When you say that "Rust makes the same guarantees regardless of the unsafe keyword", it seems to me that you are doing something equally pointless.
> You seem to be suggesting that a language being safe or unsafe is a social contract rather than a technical property of the language.
Quite some way up this thread pizlonator insists that each programming language defines memory safety differently, quantifying some as "weaker" or "stronger" and giving the example that Rust has the `unsafe` keyword and so that's weaker than Fil-C.
That's what we were discussing when you jumped in with your C hypothetical.
You apparently instead believe in a single universal "safety" and every language is either absolutely safe or unsafe according to foldr for whatever that's worth - but that's not what we were talking about.
> ... then we can make C safe without any technical changes just by adding some language to the standard saying that C programmers are obliged to ensure that their code maintains a certain list of invariants.
In Rust you can use #![forbid(unsafe_code)]
to totally forbid unsafe code in your codebase. Rust also checks for memory safety at compile time, these are strong guarantees that ensure that if the code compiles it is memory safe.
I'm aware of that, but I'm responding to the original claim that "Rust makes the same guarantees regardless of the unsafe keyword" (see https://news.ycombinator.com/item?id=46262774)
Ah. I agree with you. When unsafe is used the borrow checker cannot check for memory safety, the programmer has to provide the guarantees by making sure their code does not violate memory safety, similar to programming in C.
But unsafe Rust is still far better than C because the unsafe keyword is visible and one can grep it and audit the unsafe parts. Idiomatic Rust also requires that the programmer provides comments as to why that part is unsafe.
I think making things more explicit with "unsafe" is an advantage of Rust, but I think "far better" is a bit of an exaggeration. In C you need to audit pointer arithmetic, malloc/free, casts and unons. If you limit pointer arithmetic to a few safe accessor functions and have a documented lifetime rules, this is also relatively simple to do (more difficult than "grep" but not much). Vice versa, if you use a lot of "unsafe" in Rust or in complicated ways, it can also easily become possible to guarantee safety. In contrast to what people seem to believe, the bug does not need to be inside in unsafe block (a logic error outside can cause the UB inside unsafe or a violation of some of Rust's invariants inside unsafe can allow UB outside of unsafe) and can result even from the interaction of unsafe blocks.
The practical memory safety we see in Rust is much more the result of trying hard to avoid memory safety issues and requiring comments for unsafe blocks is part of this culture.
You may define "memory safety" as you like. I will define "trustworthy system" as one in which the author acknowledges and owns limitations instead of iteratively refining private definitions until the limitations disappear. You can define a mathematical notation in which 2+3=9, but I'm under no obligation to accept it, and I'll take the attempt into consideration when evaluating the credibility of proofs in this strange notation.
Nobody is trying to hide the existence of "eval" or "unsafe". You're making a categorical claim of safety that's true only under a tendentious reading of common English words. Users reading your claims will come away with a mistaken faith in your system's guarantees.
> I will define "trustworthy system" as one in which the author acknowledges and owns limitations instead of iteratively refining private definitions until the limitations disappear.
You know about this limitation that you keep going on about because it’s extremely well documented on fil-c.org
[Woman walking on beach at sunset, holding hands with husband]
Voiceover: "Miracurol cures cancer."
[Couple now laughing over dinner with friends]
"Ask your doctor if Miracurol is right for you."
[Same footage continues, voice accelerates]
"In clinical trials, five mice with lymphoma received Miracurol. All five were cured. One exploded. Not tested in humans. Side effects include headache, itchiness, impotence, explosion, and death. Miracurol's cancer-free guarantee applies only to cancers covered under Miracurol's definition of cancer, available at miracurol.org. Manufacturer not responsible for outcomes following improper use. Consult your doctor."
[Couple walking golden retriever, sun flare]
Voiceover: "Miracurol. Because you deserve to live cancer-free."
Patient: "I exploded."
Miracurol: "That's extremely well documented on miracurol.org."
> I will define "trustworthy system" as one in which the author acknowledges and owns limitations
You can't then go on to complain that the author does document the limitations but considers the overall system good. Fil-C, by the definition you just espoused, is a "trustworthy system".
It’s really sad to see your posts on this thread. Fil-C is an incredible achievement and absolutely full of interesting technical details to dig into. I’m not a mod, but as a reader of the site and someone who takes a curious interest in the progress of Fil-C, can you please stop attacking its creator like this. It’s tedious, needlessly rude, and lessens the opportunity for the rest of us to actually learn something from an expert.
Facts are facts and exist independent of who discovers them. If you'd like to learn, the last thing you want to do is stop people poking at contradictions and pressure-testing claims. If Fil-C is really the "incredible achievement" you say it is, it can withstand scrutiny.
I'm not an expert here but I have to say this feels like a very weak objection.
p points to P1. One thread reads through p. Another thread races with that and mutates p to point to P2. The result is the first thread reads from either P1 or P2 (but no other object).
This seems totally fine and expected to me? If there's a data race on a pointer, you might read one or the other values, but not garbage and not out of bounds. I mean, if it could guarantee a panic that's nice, but that's a bonus, not required for safety.
Say your C program has sensitive information in module A and a memory safety bug in module B. Running that program in wasm won’t prevent the attacker from using the bug in B to get read/write access to the data in A.
In practice what the attacker will really do is use the memory safety bug to achieve weird execution: even without control over the program counter, the fact that a memory safety bug inside the wasm memory gives read write access to all of that memory means the attacker can make the program do whatever they want, subject to the wasm sandbox limits (ie whatever the host allows the wasm guest to do).
Basically wasm amounts to a lightweight and portable replacement for running native code in a sufficiently sandboxed process
Your general point stands - wasm's original goal was mainly sandboxing - but
1. Wasm does provide some amount of memory safety even to compiled C code. For example, the call stack is entirely protected. Also, indirect calls are type-checked, etc.
2. Wasm can provide memory safety if you compile to WasmGC. But, you can't really compile C to that, of course...
Correct me if I'm wrong, but with LLVM on Wasm, I think casting a function pointer to the wrong type will result in you calling some totally unrelated function of the correct type? That sounds like the opposite of safety to me.
I agree about the call stack, and don't know about GC.
That is incorrect about function pointers: The VM does check that you are calling the right function type, and it will trap if the type does not match.
Here it is in the spec:
> The call_indirect instruction calls a function indirectly through an operand indexing into a table that is denoted by a table index and must have type funcref. Since it may contain functions of heterogeneous type, the callee is dynamically checked against the function type indexed by the instruction’s second immediate, and the call is aborted with a trap if it does not match.
(Other sandboxing approaches, including related ones like asm.js, do other things, some closer to what you mentioned. But wasm has strict checking here.)
It’s true that a full blown VM is an excellent sandbox.
The usual situation is like what chrome or OpenSSH want:
- They want to be able to do dangerous things by design. Chrome wants to save downloads. Chrome wants to call rendering APIs. OpenSSH wants to pop a shell.
- They want to deal with untrusted inputs. Chrome downloads things off the internet and parses them. OpenSSH has a protocol that it parses.
So you want to split your process into two with privilege separation:
- one process has zero privileges and does the parsing of untrusted inputs.
- another process has high privilege but never deals with untrusted inputs.
And then the two processes have some carefully engineered IPC protocol for talking to one another.
Could you run the deprivileged process in a VM for maximum security? Yeah that’s one way to do it. But it’s cleaner to run it as a normal process, ask the OS to sandbox it (deprivilege it), and then have a local domain socket (or whatever) that the two processes can use to communicate.
If you used a VM for deprivileging then:
- There’d be more overhead. Chrome wants to do this per origin per tab. OpenSSH wants to do it per connection. Maybe a VM is too much
- You could put the whole browser into the VM but then you’d still need something outside it for saving files. And probably for talking to the GPU. You could run OpenSSH in the VM but then that defeats the purpose (you want to use it to pop a shell after all).
- You can use vsocks and other things to communicate between host and guest but it’s much more gross than the options available when using traditional process sandboxing
Does it even work with openssh example? Pwning the parser progress will let attacker spoof arbitrary communication, which in case of SSH lets them execute arbitrary commands. Or is there a smart way to work around that?
> "If I don't make big claims, no one will notice."
I am making big claims because there are big claims to be made.
> he statements about the actual security benefits should be independently verified -this hasn't happened yet
I don't know what this means. Folks other than me have independently verified my claims, just not exhaustively. No memory safe language runtime has been exhaustively verified, save maybe Spark. So you're either saying something that isn't true at all, or that could be said for any memory safe language runtime.
To clarify the position, my concern isn't that the project is bad - it's that security engineering is a two-front war. You have to add new protections (memory safety) without breaking existing contracts (like ld.so behavior).
When a project makes 'big claims' about safety, less technical users might interpret that as 'production ready'. My caution is caused by the fact that modifying the runtime is high-risk territory where regressions can introduce vulns that are distinct from the memory safety issues you are solving.
The goal is to prevent the regression in the first place. I'm looking forward to seeing how the verification matures and rooting for it.
> without breaking existing contracts (like ld.so behavior)
If you think that Fil-C regresses ld.so then get specific. Otherwise what you’re doing is spreading fear, uncertainty, and doubt for no good reason.
Fil-C has always honored the setuid behavior provided by ld.so. There was a bug - since fixed - that the Fil-C runtime called getenv instead of secure_getenv.
> When a project makes 'big claims' about safety, less technical users might interpret that as 'production ready'.
Fil-C is production ready and already has production users.
Running ffmpeg compiled for wasm and watching as most codec selections lead to runtime crashes due to invalid memory accesses is fun. But, yeah, it’s runtime safety, so going to wasm as a middle step doesn’t do much.
> Running ffmpeg compiled for wasm and watching as most codec selections lead to runtime crashes due to invalid memory accesses is fun.
For all you know that’s a bug in the wasm port of the codec.
> it’s runtime safety
So is Fil-C
The problem with wasm is that an OOBA in one C allocation in the wasm guest can still give the attacker the power to clobber any memory in the guest. All that’s protected is the host. That’s enough to achieve weird execution.
Hence why I say that wasm is a sandbox. It’s not memory safety.
Finally reality is catching up with the WASM sales pitch against other bytecode formats introduced since 1958, regarding security and how great it is over anything else.
Warm was great because it was lightweight and easy to target from any language and create any custom interaction API with the host. That's becoming less true as they bolt on features no one needed (GC) and popularize standardized interfaces that contain the kitchen sink (WASI) but these things can still be treated as optional so it can still be used for much more flexible use cases than java or .net
There’s no guarantee the toolchains will support WASM “preview” forever and make the bloat optional, and even if they do you could still end up in an ecosystem where it would be unviable. At some point you’re probably better off just compiling to RISCV and using an emulator library instead.
Wasm now supports multiple modules and multiple linear memories per module, so it ought to be quite possible to compile C to Wasm in a way that enforces C's object access rules, much like CHERI if perhaps not Fil-C itself.
You wouldn't be able to get quite as fine-grained. One memory per object is probably horrifically slow. And I don't know about Fil-C, but CHERI at least allows capabilities (pointers with bounds) to overlap and subset each other. I.e. you could allocate an arena and get a capability for that, and then allocate an object inside that arena and get a smaller capability for that, and then get a pointer to a field in that object and get capability just for that field.
Fil-C has like one "linear memory" per object and each capability gives read/write access to the whole object.
But Fil-C has its compiler which does analysis passes for eliding bounds-checks where they are not needed, and I think it could theoretically do a better job at that than a WASM compiler with multi-memories, because C source code could contain more information.
Unlike WASM, but like CHERI, every pointer in memory is also tagged, and would lose its pointer status if overwritten by an integer, so it is still more memory-safe in that way.
One would probably just need to define WASM extensions that allow for such subsetting. Performance will probably be competitive with software implementations of CHERI (perhaps with varying levels of hardware acceleration down the road) which isn't that bad.
Posts like the one I made about how to do sandboxing are specifically to make the runtime transparent to folks so that meaningful auditing can happen.
> For example, Filip mentioned that some setuid programs can be compiled with it, but it also makes changes to ld.so. I pointed this out to the author on Twitter, as it could be problematic.
The changes to ld.so are tiny and don’t affect anything interesting to setuid. Basically it’s just one change: teaching the ld.so that the layout of libc is different.
More than a month ago, I fixed a setuid bug where the Fil-C runtime was calling getenv rather than secure_getenv. Now I’m just using secure_getenv.
> In other words, these are still teething problems with Fil-C, which will be reviewed and fixed over time. I just want to point out that using it for real-world "infrastructures" might be somewhat risky at this point. We need unix nerds to experiment with.
There’s some truth to what you’re saying and there’s also some FUD to what you’re saying. Like a perfectly ambiguous mix of truth and FUD. Good job I guess?
Is it FUD? Approximately speaking, all software has bugs. Being an early adopter for security critical things is bound to carry significant risk. It seems like a relevant topic to bring up in this sort of venue for a project of this sort.
It's true. I used to promote high-assurance kernels. They had low odds of coding errors but the specs could be wrong. Many problems Linux et al. solved are essentially spec-level. So, we just apply all of that to the secure designs, right?
Well, those spec issues are usually not documented or new engineers won't know where to find a full list. That means the architecturally-insecure OS's might be more secure in specific areas due to all the investment put into them over time. So, recommending the "higher-security design" might actually lower security.
For techniques like Fil-C, the issues include abstraction gap attacks and implementation problems. For the former, the model of Fil-C might mismatch the legacy code in some ways. (Ex: Ada/C FFI with trampolines.) Also, the interactions between legacy and Fil-C might introduce new bugs because integrations are essentially a new program. This problem did occur in practice in a few, research works.
I haven't reviewed Fil-C. I've forgotten too much C and the author was really clever. It might be hard to prove the absence of bugs in it. However, it might still be very helpful in securing C programs.
I think Rust is great for sandboxing because of how Rust has basically no runtime. This is one of the nice things about rust!
Go has the same problems I’m describing in my post. Maybe those folks haven’t done the work to make the Go runtime safe for sandboxing, like what I did for Fil-C.
Sure, but even just setuiding to a restrictive uid or chrooting would go a long way, even in a managed runtime language where syscall restrictions are more challenging.
It would require a bit of porting (since Fil-C currently assumes you have all of the Linux syscalls). But you could probably even lift some of the microVM’s functionality into Fil-C’s memory safe userland.
Fil-C will panic your program, or give some kind of memory safe outcome (that is of no use to the attacker) in all of the cases that attackers use to achieve remote code execution. In other words, Fil-C is memory safe.
The fact that Fil-C achieves memory safety using runtime checks doesn’t make it any less memory safe. Even rust uses runtime checks (most importantly for array bounds). And, type systems that try to prove safety statically often amount to forcing the programmer to write the checks themselves.
reply