Contents

• Overview

  • Example library sandboxing

• Core API

  • Creating (and destroying) sandboxes

  • Calling sandboxed library functions

  • Exposing functions to sandboxed code

  • Tainted values

    • Unwrapping tainted values

    • Operating on tainted values

  • Application-sandbox shared memory

  • Standard library

• References

• Indices and tables

Overview

This is a short tutorial on the RLBox API. If you are looking for a reference of all APIs, see [Doxygen].

RLBox is a toolkit for sandboxing third-party libraries. The toolkit consists of (1) a Wasm-based sandbox and (2) an API for retrofitting existing application code to interface with a sandboxed library. The Wasm-based sandbox is documented in its corresponding repository. This documentation focuses on the API and the interface you will use when sandboxing code, independent of the underlying sandboxing mechanism.

Why do we need a sandboxing API? Sandboxing libraries without the RLBox API is both tedious and error-prone. This is especially the case when retrofitting an existing codebase like Firefox where libraries are trusted and thus the application-library boundary is blurry. To sandbox a library – and thus to move to a world where the library is no longer trusted – we need to modify this application-library boundary. For example, we need to add security checks in Firefox to ensure that any value from the sandboxed library is properly validated before it is used. Otherwise, the library (when compromised) may be able to abuse Firefox code to hijack its control flow (see [RLBoxPaper] for details). The RLBox API is explicitly designed to make retrofitting of existing application code simpler and less error-prone.

Sandboxing architecture overview As shown in Fig. 1, RLBox ensures that a sandboxed library is memory isolated from the rest of the application – the library cannot directly access memory outside its designated region – and that all boundary crossings are explicit. This ensures that the library cannot, for example, corrupt Firefox’s address space. It also ensures that Firefox cannot inadvertently expose sensitive data to the library (e.g., pointers that would leak its ASLR).

Fig. 1 Sandboxed libraries are isolated from the application and all communication between the sandboxed library and application code is mediated. This ensures that the application code is robust and does not use untrusted, tainted values without checking them.

Memory isolation is enforced by the underlying sandboxing mechanism (from the start, when you create the sandbox with create_sandbox()). Explicit boundary crossings are enforced by RLBox (either at compile- or and run-time). For example, with RLBox you can’t call library functions directly; instead, you must use the invoke_sandbox_function() method. Similarly, the library cannot call arbitrary Firefox functions; instead, it can only call functions that you expose with the register_callback() method. (To simplify the sandboxing task, though, RLBox does expose a standard library as described in Standard library.)

When calling a library function, RLBox copies simple values into the sandbox memory before calling the function. For larger data types, such as structs and arrays, you can’t simply pass a pointer to the object. This would leak ASLR and, more importantly, would not work: sandboxed code cannot access application memory. So, you must explicitly allocate memory in the sandbox via malloc_in_sandbox() and copy application data to this region of memory (e.g., via strncpy).

RLBox similarly copies simple return values and callback arguments. Larger data structures, however, must (again) be passed by sandbox-reference, i.e., via a reference/pointer to sandbox memory.

To ensure that application code doesn’t use values that originate in the sandbox – and may thus be under the control of an attacker – unsafely, RLBox considers all such values as untrusted and taints them. Tainted values are essentially opaque values (though RLBox does provide some basic operators on tainted values). To use a tainted value, you must unwrap it by copying the value into application memory – and thus out of the reach of the attacker – and verifying it. Indeed, RLBox forces application code to perform the copy and verification in sync using verifiction functions.

Example library sandboxing

To get a feel for what it’s like to use RLBox, we’re going to sandbox a tiny library mylib that has four functions:

// mylib.c:

void hello() {
   printf("Hello world from mylib\n");
}

unsigned add(unsigned a, unsigned b) {
   return a + b;
}

void echo(const char* str) {
   printf("> mylib: %s\n", str);
}

void call_cb(void (*cb) (const char* str)) {
   cb("hi again!");
}

This is not the most interesting library, security-wise, but it is complicated enough to demonstrate various RLBox features.

To get started, in our main application file let’s first import the RLBox library:

// main.cpp:

#define RLBOX_SINGLE_THREADED_INVOCATIONS
#define RLBOX_USE_STATIC_CALLS() rlbox_noop_sandbox_lookup_symbol

#include <stdio.h>
#include "mylib.h"
#include "rlbox.hpp"
#include "rlbox_noop_sandbox.hpp"

using namespace rlbox;
...

In our main function, let’s now create a new sandbox (for this example we’re going to use the NULL sandbox) and call the hello function:

...
int main(int argc, char const *argv[]) {

   // Create a new sandbox
   rlbox::rlbox_sandbox<rlbox_noop_sandbox> sandbox;
   sandbox.create_sandbox();

   // call the library hello function
   sandbox.invoke_sandbox_function(hello);
...

Note that we do not call hello() directly. Instead, we use the invoke_sandbox_function() method. We can similarly call the add function:

...
   // call the add function and check the result:
   auto ok = sandbox.invoke_sandbox_function(add, 3, 4).copy_and_verify([](unsigned ret){
         printf("Adding... 3+4 = %d\n", ret);
         return ret == 7;
   });
   printf("OK? = %d\n", ok);
...

This invocation is a bit more interesting. First, we call add with arguments. Second, RLBox ensures that the unsigned return value that add returns is tainted and thus cannot be used without verification. Here, we call the copy_and_verify function which copies the value into application memory and runs our verifier function:

[](unsigned ret){
      printf("Adding... 3+4 = %d\n", ret);
      return ret == 7;
}

This lambda simply prints the tainted value and returns true if it is 7. A compromised library could return any value and if we use this value to, say, index an array this could potentially introduce an out-of-bounds memory access.

Let’s now call the echo function which takes a slightly more interesting argument: a string. Here, we can’t simply pass a string literal as an argument: the sandbox cannot access application memory where this would be allocated. Instead, we must allocate a buffer in sandbox memory and copy the string we want to pass to echo into this region:

...
   const char* helloStr = "hi hi!";
   size_t helloSize = strlen(helloStr) + 1;
   // allocate memory in the sandbox:
   auto taintedStr = sandbox.malloc_in_sandbox<char>(helloSize);
   // copy helloStr into the sandbox:
   std::strncpy(taintedStr.unverified_safe_pointer_because(helloSize, "writing to region"), helloStr, helloSize);
...

Note that taintedStr is actually a tainted string: it lives in the sandbox memory and could be written to by the (compromised) library code concurrently. As such, it’s unsafe for us to use this pointer without verification. Above, we use the “unverified_safe_pointer_because” verifier which basically removes the taint without any verification. This is safe because we copy the helloStr to sandbox memory: at worst, the sandboxed library can overwrite the memory region pointed to by taintedStr and crash when it tries to print it.

Now, we can just call the function and free the allocated string:

...
   sandbox.invoke_sandbox_function(echo, taintedStr);
   sandbox.free_in_sandbox(taintedStr);
...

Finally, let’s call the call_cb function. To do this, let’s first define a callback for the function to call. We define this function above the main function:

...
void hello_cb(rlbox_sandbox<rlbox_noop_sandbox>& _,
            tainted<const char*, rlbox_noop_sandbox> str) {
   auto checked_string =
      str.copy_and_verify_string([](std::unique_ptr<char[]> val) {
         return std::strlen(val.get()) < 1024 ? std::move(val) : nullptr;
      });
   printf("hello_cb: %s\n", checked_string.get());
}
...

This callback is called with a string. We thus call the string verification function with a simple verifier:

...
   [](std::unique_ptr<char[]> val) {
        return std::strlen(val.get()) < 1024 ? std::move(val) : nullptr;
    }
...

This verifier moves the string if it’s length is less than 1KB and otherwise returns the nullptr. In the callback we simply print this (potentially null) string.

Let’s now continue in main, register the callback – otherwise RLBox will disallow the library-application call – and pass the callback to the call_cb function:

...
   // register callback and call it
   auto cb = sandbox.register_callback(hello_cb);
   sandbox.invoke_sandbox_function(call_cb, cb);
...

Finally, let’s destroy the sandbox and exit:

...
   // destroy sandbox
   sandbox.destroy_sandbox();

   return 0;
}

Core API

In this section we describe a large part of the RLBox API you are likely to encounter when porting libraries. The API has some more advanced features and types that are necessary but not as commonly used (see [Doxygen]). In most cases the RLBox type system will give you an informative error if and how to use these features.

Creating (and destroying) sandboxes

RLBox encapsulates sandboxes with rlbox_sandbox class. For now, RLBox supports two sandboxes: a Wasm-based sandboxed and the null sandbox. The null sandbox doesn’t actually enforce any isolation, but is very useful for migrating an existing codebase to use the RLBox API. In fact, in most cases you want to port the existing code to use RLBox when interfacing with a particular library and only then switch over to the Wasm-based sandbox.

template<typename T_Sbx>
class rlbox_sandbox : protected T_Sbx

Encapsulation for sandboxes.

tparam T_Sbx

Type of sandbox. For the null sandbox this is rlbox_noop_sandbox

class rlbox_noop_sandbox

Class that implements the null sandbox. This sandbox doesn’t actually provide any isolation and only serves as a stepping stone towards migrating an application to use the RLBox API.

template<typename ...T_Args>
inline auto rlbox::rlbox_sandbox::create_sandbox(T_Args... args)

Create a new sandbox.

Template Parameters

T_args – Arguments passed to the underlying sandbox implementation. For the null sandbox, no arguments are necessary.

Creating sandboxes is mostly straightforward. For the null sandbox, however, you need to add a #define at the top of your entry file, before you include the RLBox headers:

#define RLBOX_USE_STATIC_CALLS() rlbox_noop_sandbox_lookup_symbol
...
rlbox::rlbox_sandbox<rlbox_noop_sandbox> sandbox;
sandbox.create_sandbox();

inline auto rlbox::rlbox_sandbox::destroy_sandbox()

Destroy sandbox and reclaim any memory.

It’s important to destroy a sandbox after you are done with it. This ensures that the memory footprint of sandboxing remains low. Once you destroy a sandbox though, it is an error to use the sandbox object.

Calling sandboxed library functions

RLBox disallows code from calling sandboxed library functions directly. Instead, application code must use the invoke_sandbox_function() method.

invoke_sandbox_function(func_name, ...)

Call sandbox function.

Parameters

• func_name – The sandboxed library function to call.

• ... – Arguments to function should be simple or tainted values.

Returns

Tainted value or void.

Though this function is defined via macros, RLBox uses some template and macro magic to make this look like a sandbox method. So, in general, you can call sandboxed library functions as:

// call foo(4)
auto result = sandbox.invoke_sandbox_function(foo, 4);

Exposing functions to sandboxed code

Application code can expose callback functions to sandbox via register_callback(). These functions can be called by the sandboxed code until they are unregistered.

The type signatures of register_callback() function is a bit daunting. In short, the function takes a callback function and returns a function pointer that can be passed to the sandbox (e.g., via invoke_sandbox_function()).

A callback function is a function that has a special type:

• The first argument of the function must be a reference a sandbox object.

• The remaining arguments must be tainted.

• The return value must be tainted or void. This ensures that the application cannot accidentally leak data to the sandbox.

Forcing arguments to be tainted forces the application to handled values coming from the sandbox with care. Dually, the return type ensures that the application cannot accidentally leak data to the sandbox.

template<typename T_Ret, typename ...T_Args>
inline void rlbox::rlbox_sandbox::unregister_callback(void *key)

Unregister a callback function and disallow the sandbox from calling this function henceforth.

Tainted values

Values that originate in the sandbox are tainted. We use a special tainted type tainted to encapsulate such values and prevent the application from using tainted values unsafely.

template<typename T, typename T_Sbx>
class tainted

RLBox has several kinds of tainted values, beyond tainted. Thse, however, are slightly less pervasive in the surface API.

template<typename T, typename T_Sbx>
class tainted_volatile : public rlbox::tainted_base_impl<tainted_volatile, T, T_Sbx>

Tainted volatile values are like tainted values but still point to sandbox memory. Dereferencing a tainted pointer produces a tainted_volatile.

class tainted_boolean_hint

Tainted boolean value that serves as a “hint” and not a definite answer. Comparisons with a tainted_volatile return such hints. They are not tainted<bool> values because a compromised sandbox can modify tainted_volatile data at any time.

Unwrapping tainted values

To use tainted values, the application can copy the value to application memory, verify the value, and unwrap it. RLBox provides several functions to do this.

template<typename T_Func>
inline auto rlbox::tainted_base_impl::copy_and_verify(T_Func verifier) const

Copy tainted value from sandbox and verify it.

Parameters

verifer – Function used to verify the copied value.

Template Parameters

T_Func – the type of the verifier.

Returns

Whatever the verifier function returns.

For a given tainted type, the verifier should have the following signature:

Tainted type kind	Example type	Example verifier
Simple type	int	T_Ret(*)(int)
Pointer to simple type	int*	T_Ret(*)(unique_ptr<int>)
Pointer to class type	Foo*	T_Ret(*)(unique_ptr<Foo>)
Pointer to array	int[4]	T_Ret(*)(std::array<int, 4>)
Class type	Foo	T_Ret(*)(tainted<Foo>)

In general, the return type of the verifier T_Ret is not constrained and can be anything the caller chooses.

template<typename T_Func>
inline auto rlbox::tainted_base_impl::copy_and_verify_range(T_Func verifier, std::size_t count) const

Copy a range of tainted values from sandbox and verify them.

Parameters

• verifer – Function used to verify the copied value.

• count – Number of elements to copy.

Template Parameters

T_Func – the type of the verifier. If the tainted type is int* then T_Func = T_Ret(*)(unique_ptr<int[]>).

Returns

Whatever the verifier function returns.

template<typename T_Func>
inline auto rlbox::tainted_base_impl::copy_and_verify_string(T_Func verifier) const

Copy a tainted string from sandbox and verify it.

Parameters

verifer – Function used to verify the copied value.

Template Parameters

T_Func – the type of the verifier T_Ret(*)(unique_ptr<char[]>)

Returns

Whatever the verifier function returns.

template<typename T_Func>
inline auto rlbox::tainted_base_impl::copy_and_verify_address(T_Func verifier)

Copy a tainted pointer from sandbox and verify the address.

This function is useful if you need to verify physical bits representing the address of a pointer. Other APIs such as copy_and_verify performs a deep copy and changes the address bits.

Parameters

verifier – Function used to verify the copied value.

Template Parameters

T_Func – the type of the verifier T_Ret(*)(uintptr_t)

Returns

Whatever the verifier function returns.

In some cases it’s useful to unwrap tainted values without verification. Sometimes this is safe to do and RLBox provides a method for doing so called unverified_safe_because

Since pointers are special (sandbox code may modify the data the pointer points to), we have a similar function for pointers called unverified_safe_pointer_because. This API requires specifying the number of elements being pointed to for safety.

We however provide additional functions that are especially useful during migration:

template<template<typename, typename> typename T_Wrap, typename T, typename T_Sbx>
class rlbox::tainted_base_impl

Public Functions

inline auto UNSAFE_unverified()

Unwrap a tainted value without verification. This is an unsafe operation and should be used with care.

inline auto UNSAFE_sandboxed(rlbox_sandbox<T_Sbx> &sandbox)

Like UNSAFE_unverified, but get the underlying sandbox representation.

For the Wasm-based sandbox, this function additionally validates the unwrapped value against the machine model of the sandbox (LP32).

Parameters

sandbox – Reference to sandbox.

These functions are also available for callback

Danger

Unchecked unwrapped tainted values can be abused by a compromised or malicious library to potentially compromise the application.

Operating on tainted values

Unwrapping tainted values requires care – getting a verifier wrong could lead to a security vulnerability. It’s also not cheap: we need to copy data to the application memory to ensure that the sandboxed code cannot modify the data we’re tyring to verify. Lucikly, it’s not always necessary to copy and verify: sometimes we can compute on tainted values directly. To this end, RLBox defines different kinds of operators on tainted values, which produce tainted values. This allows you to perform some computations on tainted values, pass the values back into the sandbox, and only later unwrap a tainted value when you need to. operators like + and - on tainted values.

Class of operator	Supported operators
Arithmetic operators	=, +, -, *, /, %, ++, --
Relational operators	==, !=, <, <=, >, >=
Logical operators	!, && (limited), || (limited)
Bitwise operators	~, &, |, ^, <<, >>
Compound operators	+=, -=, *=, /=, %=, &=, |=, ^=, <<=, >>=
Pointer operators	[], *, &, ->

When applying a binary operator like << to a tainted value and an untainted values the result is always tainted.

RLBox also defines several comparison operators on tainted values that sometime unwrap the result:

• Operators ==, != on tainted pointers is allowed if the rhs is nullptr_t and return unwrapped bool.

• Operator ! on tainted pointers retruns an unwrapped bool.

• Operators ==, !=, ! on non-pointer tainted values return a tainted<bool>

• Operators ==, !=, ! on tainted_volatile values returns a tainted_boolean_hint

• Operators && and || on booleans are only permitted when arguments are variables (not expressions). This is because C++ does not permit safe overloading of && and || operations with expression arguments as this affects the short circuiting behaviour of these operations.

Application-sandbox shared memory

Since sandboxed code cannot access application memory, to share objects across the boundary you need to explicitly allocate memory that both the application and sandbox can access. To this end, malloc_in_sandbox() allocates memory within the sandbox region and returns a tainted pointer that can be used by both the application and sandbox (e.g., by passing the pointer as an argument to a function).

template<typename T>
inline tainted<T*, T_Sbx> rlbox::rlbox_sandbox::malloc_in_sandbox()

Allocate a new pointer that is accessible to both the application and sandbox. The pointer is allocated in sandbox memory.

Template Parameters

T – The type of the pointer you want to create. If T=int, this would return a pointer to an int.

Returns

tainted<T*, T_Sbx> Tainted pointer accessible to the application and sandbox.

template<typename T>
inline tainted<T*, T_Sbx> rlbox::rlbox_sandbox::malloc_in_sandbox(uint32_t count)

Allocate an array that is accessible to both the application and sandbox. The pointer is allocated in sandbox memory.

Template Parameters

T – The type of the array elements you want to create. If T=int, this would return a pointer to an array of ints.

Parameters

count – The number of array elements to allocate.

Returns

tainted<T*, T_Sbx> Tainted pointer accessible to the application and sandbox.

To distinguish between different pointer types, RLBox also provides some helper functions:

inline bool rlbox::rlbox_sandbox::is_pointer_in_app_memory(const void *p)

Check if the pointer points to application memory. For the null-sandbox, this always returns true.

inline bool rlbox::rlbox_sandbox::is_pointer_in_sandbox_memory(const void *p)

Check if the pointer points to this sandbox’s memory. For the null-sandbox, this always returns true.

static inline bool rlbox::rlbox_sandbox::is_in_same_sandbox(const void *p1, const void *p2)

Check if two pointers are in the same sandbox. For the null-sandbox, this always returns true.

Standard library

RLBox provides several helper functions to application for handling sandboxed memory regions and values.

template<typename T_Sbx, typename T_Rhs, typename T_Val, typename T_Num, template<typename, typename> typename T_Wrap>
inline T_Wrap<T_Rhs*, T_Sbx> rlbox::memset(rlbox_sandbox<T_Sbx> &sandbox, T_Wrap<T_Rhs*, T_Sbx> ptr, T_Val value, T_Num num)

Fill sandbox memory with a constant byte.

template<typename T_Sbx, typename T_Rhs, typename T_Lhs, typename T_Num, template<typename, typename> typename T_Wrap>
inline T_Wrap<T_Rhs*, T_Sbx> rlbox::memcpy(rlbox_sandbox<T_Sbx> &sandbox, T_Wrap<T_Rhs*, T_Sbx> dest, T_Lhs src, T_Num num)

Copy to sandbox memory area. Note that memcpy is meant to be called on byte arrays does not adjust data according to ABI differences. If the programmer does accidentally call memcpy on buffers that needs ABI adjustment, this may cause compatibility issues, but will not cause a security issue as the destination is always a tainted or tainted_volatile pointer.

template<typename T_Lhs, typename T_Rhs, typename T_Sbx, template<typename, typename> typename T_Wrap>
inline tainted<T_Lhs, T_Sbx> rlbox::sandbox_reinterpret_cast(const T_Wrap<T_Rhs, T_Sbx> &rhs) noexcept

The equivalent of a reinterpret_cast but operates on sandboxed values.

template<typename T_Lhs, typename T_Rhs, typename T_Sbx, template<typename, typename> typename T_Wrap>
inline tainted<T_Lhs, T_Sbx> rlbox::sandbox_const_cast(const T_Wrap<T_Rhs, T_Sbx> &rhs) noexcept

The equivalent of a const_cast but operates on sandboxed values.

template<typename T_Lhs, typename T_Rhs, typename T_Sbx, template<typename, typename> typename T_Wrap>
inline tainted<T_Lhs, T_Sbx> rlbox::sandbox_static_cast(const T_Wrap<T_Rhs, T_Sbx> &rhs) noexcept

The equivalent of a static_cast but operates on sandboxed values.

References

RLBoxPaper

*Retrofitting Fine Grain Isolation in the Firefox Renderer*. S. Narayan, C. Disselkoen, T. Garfinkel, S. Lerner, H. Shacham, D. Stefan.

Doxygen(1,2)

RLBox Doxygen Documentation

Indices and tables

• Index

• Module Index

• Search Page