Safety in Rust
Rust is a systems programming language which attempts to be a “safe, concurrent, and practical language.” Rust uses “Zero-Cost” abstractions to accomplish these aspirations. By catching all memory and concurrency pitfalls early on, in the compilation step by enforcing strong rules on data access and manipulation, as well as fostering good development choices through idiomatic practices, the language re-enforces safety without the cost associated with traditional mechanisms. The root of the memory safety abstractions revolve around a fairly simple set of rules. Rust enforces an ownership system which includes borrowing to allow for guaranteed memory safety, which allows for all compiled programs to be free of data race conditions, as well as memory access errors.
Language Security Features
Immutability by Default
Variables within Rust by default are immutable. By defaulting variables to being immutable, the Rust language forces the developer to really think about if a variable needs to be modified in place.
Within Rust, variable bindings have a concept of ownership over what they are bound to. Logically when an entity is bound to a label, or variable, when the variable goes out of scope, Rust will free the bound resource.
By enforcing this rule very strictly, every time a variable goes out of scope it’s bound resource is deterministically freed. This protection allows us the language to guard against common errors such as double free errors which are common to other languages.
In addition to the deterministic freeing of bound resources, entities bound to a variable must follow the following rules:
- There can be zero or more immutable references
- There can be exactly one reference which is mutable
This means that you can either have as many immutable references to a variable as you want, OR you can have exactly one mutable reference ONLY. These rules enforced within the compiler allow guaranteed data race condition protections.
Data races occur when the following are true:
- Two or more pointers to the same resource
- At least one pointer is written
- There is no synchronization
As demonstrated by the rules, there can only ever be exactly one pointer to a resource at any time which is mutable. This check is performed at compile time. Since you cannot have an immutable pointer and a mutable pointer to the same variable binding at the same time there is no chance of a data race.
Given that there are rules about only having one mutable pointer to a variable binding at a time, rust employs a concept of borrowing. Effectively, when you want to have a block or function manipulate an existing variable binding rust allows the developer the ability to lend the variable binding to the new block or function through the use of references. Example seen below, pulled from the rust book:
The above fails because we have borrowed the reference from x, and given the bound value to y. Based on the rules we can only have one mutable reference so the compiler will fail referencing the fact that we are trying to use x, even though the reference is borrowed to y. A more legitimate example would be the below:
Of course the basic ownership rules still apply, (zero or more immutable, only one mutable reference). Borrowing tells the compiler to not deterministically deallocate the resource when it goes out of scope, as it is just borrowed from the calling scope, and will be deallocated later on.
Borrowing prevents a number of common issues, such as iterator invalidation and use after free errors. Iterator invalidation happens when you mutate an array or vector while you are iterating over said array or vector. After free errors happen when you try to access a variable whose binding had been freed. Since borrowing doesn’t free the variable binding, this is not possible in rust.
As mentioned above, in Rust variable bindings all have a [lifetime][lifetime]. Usually this lifetime is the length of the scope where the variable was defined. But what happens when you need to use variables from different scopes in the same function call that returns something? Which scope lifetime do you need to use?
In Rust you can define explicit lifetimes for function arguments, and struct parameters, which allow the developer to explicitly tell the compiler that this variable binding exists for a given lifetime. Example below:
In the above example, what if the result of
add still references
b? at our
println! we have the potential for a use after free, or dangling pointer
To allot for these situations Rust employs explicit lifetime mappings for function inputs and outputs, so Rust can make sure that a lifetime is appropriate in complicated situations. Take the below correction:
With the above lifetime annotations the compiler can now tell that the second argument’s scope is no factor to the lifetime of the response, which should be the same as the lifetime of the first parameter. By explicitly stating how long the lifetime of the parameters and results should be, the compiler can enforce the scoping rules appropriately.
List of Rust Greatness:
- Prevents Iterator Invalidation with Borrowing
- Prevents Use After Free, dangling pointers with Lifetimes and Borrowing
- Prevents All Data Races with Ownership rules
- Prevents Double Free with Scope Deallocation
- Prevents overwriting variables with immutability by default
- Prevents most Memory Leaks by freeing bound resources after scope