Remix.run Logo
josephg 4 hours ago

> i dont know what the solution here is other than stop using npm

Personally I think we need to start adding capability based systems into our programming languages. Random code shouldn't have "ambient authority" to just do anything on my computer with the same privileges as me. Like, if a function has this signature:

    function add(a: int, b: int) -> int
Then it should only be able to read its input, and return any integer it wants. But it shouldn't get ambient authority to access anything else on my computer. No network access. No filesystem. Nothing.

Philosophically, I kind of think of it like function arguments and globals. If I call a function foo(someobj), then function foo is explicitly given access to someobj. And it also has access to any globals in my program. But we generally consider globals to be smelly. Passing data explicitly is better.

But the whole filesystem is essentially available as a global that any function, anywhere, can access. With full user permissions. I say no. I want languages where the filesystem itself (or a subset of it) can be passed as an argument. And if a function doesn't get passed a filesystem, it can't access a filesystem. If a function isn't passed a network socket, it can't just create one out of nothing.

I don't think it would be that onerous. The main function would get passed "the whole operating system" in a sense - like the filesystem and so on. And then it can pass files and sockets and whatnot to functions that need access to that stuff.

If we build something like that, we should be able to build something like npm but where you don't need to trust the developers of 3rd party software so much. The current system of trusting everyone with everything is insane.

ratmice 4 hours ago | parent | next [-]

I couldn't agree with you more, the thing is our underlying security models are protecting systems from their users, but do nothing for protecting user data from the programs they run. Capability based security model will fix that.

Gigachad 3 hours ago | parent [-]

Only on desktop. Mobile has this sorted. Programs have access to their own files unrestricted, and then can access the shared file space only through the users specifically selecting them.

josephg 2 hours ago | parent [-]

I think there's 2 kinds of systems we're talking about here:

1. Capabilities given to a program by the user. Eg, "This program wants to access your contacts. Allow / deny". But everything within a program might still have undifferentiated access. This requires support from the operating system to restrict what a program can do. This exists today in iOS and Android.

2. Capabilities within a program. So, if I call a function in a 3rd party library with the signature add(int, int), it can't access the filesystem or open network connections or access any data thats not in its argument list. Enforcing this would require support from the programming language, not the operating system. I don't know of any programming languages today which do this. C and Rust both fail here, as any function in the program can access the memory space of the entire program and make arbitrary syscalls.

Application level permissions are a good start. But we need the second kind of fine-grained capabilities to protect us from malicious packages in npm, pip and cargo.

ratmice 38 minutes ago | parent [-]

I would also say there is a 3rd class, which are distributed capabilities.

When you look at a mobile program such as the GadgetBridge which is synchronizing data between a mobile device and a watch, and number of permissions it requires like contacts, bluetooth pairing, notifications, yadda yadda the list goes on.

Systems like E-Lang wouldn't bundle all these up into a single application. Your watch would have some capabilities, and those would interact directly with capabilities on the phone. I feel like if you want to look at our current popular mobile OS's as capability systems the capabilities are pretty coarse grained.

One thing I would add about compilers, npm, pip, cargo. Is that compilers are transformational programs, they really only need read and write access to a finite set of input, and output. In that sense, even capabilities are overkill because honestly they only need the bare minimum of IO, a batch processing system could do better than our mainstream OS security model.

irishcoffee 4 hours ago | parent | prev [-]

> No network access. No filesystem. Nothing.

Ironically, any c++ app I've written on windows does exactly this. "Are you sure you want to allow this program to access networking?" At least the first time I run it.

I also rarely write/run code for windows.

christophilus 2 hours ago | parent [-]

Yeah, but if that app was built using a malicious dependency that only relied on the same permissions the app already uses, you’d just click “Yes” and move on and be pwned.

irishcoffee an hour ago | parent [-]

Oh, I don't npm.

If I can't yum (et.al.) install it I absolutely review the past major point releases for an hour and do my research on the library.