211 lines
5.0 KiB
Rust
211 lines
5.0 KiB
Rust
//! zeroize integration tests.
|
|
|
|
use std::{
|
|
marker::{PhantomData, PhantomPinned},
|
|
mem::{size_of, MaybeUninit},
|
|
num::*,
|
|
};
|
|
use zeroize::*;
|
|
|
|
#[cfg(feature = "std")]
|
|
use std::ffi::CString;
|
|
|
|
#[derive(Clone, Debug, PartialEq)]
|
|
struct ZeroizedOnDrop(u64);
|
|
|
|
impl Drop for ZeroizedOnDrop {
|
|
fn drop(&mut self) {
|
|
self.0.zeroize();
|
|
}
|
|
}
|
|
|
|
#[test]
|
|
fn non_zero() {
|
|
macro_rules! non_zero_test {
|
|
($($type:ty),+) => {
|
|
$(let mut value = <$type>::new(42).unwrap();
|
|
value.zeroize();
|
|
assert_eq!(value.get(), 1);)+
|
|
};
|
|
}
|
|
|
|
non_zero_test!(
|
|
NonZeroI8,
|
|
NonZeroI16,
|
|
NonZeroI32,
|
|
NonZeroI64,
|
|
NonZeroI128,
|
|
NonZeroIsize,
|
|
NonZeroU8,
|
|
NonZeroU16,
|
|
NonZeroU32,
|
|
NonZeroU64,
|
|
NonZeroU128,
|
|
NonZeroUsize
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn zeroize_byte_arrays() {
|
|
let mut arr = [42u8; 137];
|
|
arr.zeroize();
|
|
assert_eq!(arr.as_ref(), [0u8; 137].as_ref());
|
|
}
|
|
|
|
#[test]
|
|
fn zeroize_on_drop_byte_arrays() {
|
|
let mut arr = [ZeroizedOnDrop(42); 1];
|
|
unsafe { core::ptr::drop_in_place(&mut arr) };
|
|
assert_eq!(arr.as_ref(), [ZeroizedOnDrop(0); 1].as_ref());
|
|
}
|
|
|
|
#[test]
|
|
fn zeroize_maybeuninit_byte_arrays() {
|
|
let mut arr = [MaybeUninit::new(42u64); 64];
|
|
arr.zeroize();
|
|
let arr_init: [u64; 64] = unsafe { core::mem::transmute(arr) };
|
|
assert_eq!(arr_init, [0u64; 64]);
|
|
}
|
|
|
|
#[test]
|
|
fn zeroize_check_zerosize_types() {
|
|
// Since we assume these types have zero size, we test this holds for
|
|
// the current version of Rust.
|
|
assert_eq!(size_of::<()>(), 0);
|
|
assert_eq!(size_of::<PhantomPinned>(), 0);
|
|
assert_eq!(size_of::<PhantomData<usize>>(), 0);
|
|
}
|
|
|
|
#[test]
|
|
fn zeroize_check_tuple() {
|
|
let mut tup1 = (42u8,);
|
|
tup1.zeroize();
|
|
assert_eq!(tup1, (0u8,));
|
|
|
|
let mut tup2 = (42u8, 42u8);
|
|
tup2.zeroize();
|
|
assert_eq!(tup2, (0u8, 0u8));
|
|
}
|
|
|
|
#[test]
|
|
fn zeroize_on_drop_check_tuple() {
|
|
let mut tup1 = (ZeroizedOnDrop(42),);
|
|
unsafe { core::ptr::drop_in_place(&mut tup1) };
|
|
assert_eq!(tup1, (ZeroizedOnDrop(0),));
|
|
|
|
let mut tup2 = (ZeroizedOnDrop(42), ZeroizedOnDrop(42));
|
|
unsafe { core::ptr::drop_in_place(&mut tup2) };
|
|
assert_eq!(tup2, (ZeroizedOnDrop(0), ZeroizedOnDrop(0)));
|
|
}
|
|
|
|
#[cfg(feature = "alloc")]
|
|
#[test]
|
|
fn zeroize_vec() {
|
|
let mut vec = vec![42; 3];
|
|
vec.zeroize();
|
|
assert!(vec.is_empty());
|
|
}
|
|
|
|
#[cfg(feature = "alloc")]
|
|
#[test]
|
|
fn zeroize_vec_entire_capacity() {
|
|
#[derive(Clone)]
|
|
struct PanicOnNonZeroDrop(u64);
|
|
|
|
impl Zeroize for PanicOnNonZeroDrop {
|
|
fn zeroize(&mut self) {
|
|
self.0 = 0;
|
|
}
|
|
}
|
|
|
|
impl Drop for PanicOnNonZeroDrop {
|
|
fn drop(&mut self) {
|
|
if self.0 != 0 {
|
|
panic!("dropped non-zeroized data");
|
|
}
|
|
}
|
|
}
|
|
|
|
// Ensure that the entire capacity of the vec is zeroized and that no unitinialized data
|
|
// is ever interpreted as initialized
|
|
let mut vec = vec![PanicOnNonZeroDrop(42); 2];
|
|
|
|
unsafe {
|
|
vec.set_len(1);
|
|
}
|
|
|
|
vec.zeroize();
|
|
|
|
unsafe {
|
|
vec.set_len(2);
|
|
}
|
|
|
|
drop(vec);
|
|
}
|
|
|
|
#[cfg(feature = "alloc")]
|
|
#[test]
|
|
fn zeroize_string() {
|
|
let mut string = String::from("Hello, world!");
|
|
string.zeroize();
|
|
assert!(string.is_empty());
|
|
}
|
|
|
|
#[cfg(feature = "alloc")]
|
|
#[test]
|
|
fn zeroize_string_entire_capacity() {
|
|
let mut string = String::from("Hello, world!");
|
|
string.truncate(5);
|
|
|
|
string.zeroize();
|
|
|
|
// convert the string to a vec to easily access the unused capacity
|
|
let mut as_vec = string.into_bytes();
|
|
unsafe { as_vec.set_len(as_vec.capacity()) };
|
|
|
|
assert!(as_vec.iter().all(|byte| *byte == 0));
|
|
}
|
|
|
|
// TODO(tarcieri): debug flaky test (with potential UB?) See: RustCrypto/utils#774
|
|
#[cfg(feature = "std")]
|
|
#[ignore]
|
|
#[test]
|
|
fn zeroize_c_string() {
|
|
let mut cstring = CString::new("Hello, world!").expect("CString::new failed");
|
|
let orig_len = cstring.as_bytes().len();
|
|
let orig_ptr = cstring.as_bytes().as_ptr();
|
|
cstring.zeroize();
|
|
// This doesn't quite test that the original memory has been cleared, but only that
|
|
// cstring now owns an empty buffer
|
|
assert!(cstring.as_bytes().is_empty());
|
|
for i in 0..orig_len {
|
|
unsafe {
|
|
// Using a simple deref, only one iteration of the loop is performed
|
|
// presumably because after zeroize, the internal buffer has a length of one/
|
|
// `read_volatile` seems to "fix" this
|
|
// Note that this is very likely UB
|
|
assert_eq!(orig_ptr.add(i).read_volatile(), 0);
|
|
}
|
|
}
|
|
}
|
|
|
|
#[cfg(feature = "alloc")]
|
|
#[test]
|
|
fn zeroize_box() {
|
|
let mut boxed_arr = Box::new([42u8; 3]);
|
|
boxed_arr.zeroize();
|
|
assert_eq!(boxed_arr.as_ref(), &[0u8; 3]);
|
|
}
|
|
|
|
#[cfg(feature = "alloc")]
|
|
#[test]
|
|
fn asref() {
|
|
let mut buffer: Zeroizing<Vec<u8>> = Default::default();
|
|
let _asmut: &mut [u8] = buffer.as_mut();
|
|
let _asref: &[u8] = buffer.as_ref();
|
|
|
|
let mut buffer: Zeroizing<Box<[u8]>> = Default::default();
|
|
let _asmut: &mut [u8] = buffer.as_mut();
|
|
let _asref: &[u8] = buffer.as_ref();
|
|
}
|