zeroize/lib.rs
1#![no_std]
2#![cfg_attr(docsrs, feature(doc_auto_cfg))]
3#![doc(
4 html_logo_url = "https://raw.githubusercontent.com/RustCrypto/media/6ee8e381/logo.svg",
5 html_favicon_url = "https://raw.githubusercontent.com/RustCrypto/media/6ee8e381/logo.svg"
6)]
7#![warn(missing_docs, rust_2018_idioms, unused_qualifications)]
8
9//! Securely zero memory with a simple trait ([`Zeroize`]) built on stable Rust
10//! primitives which guarantee the operation will not be "optimized away".
11//!
12//! ## About
13//!
14//! [Zeroing memory securely is hard] - compilers optimize for performance, and
15//! in doing so they love to "optimize away" unnecessary zeroing calls. There are
16//! many documented "tricks" to attempt to avoid these optimizations and ensure
17//! that a zeroing routine is performed reliably.
18//!
19//! This crate isn't about tricks: it uses [`core::ptr::write_volatile`]
20//! and [`core::sync::atomic`] memory fences to provide easy-to-use, portable
21//! zeroing behavior which works on all of Rust's core number types and slices
22//! thereof, implemented in pure Rust with no usage of FFI or assembly.
23//!
24//! - No insecure fallbacks!
25//! - No dependencies!
26//! - No FFI or inline assembly! **WASM friendly** (and tested)!
27//! - `#![no_std]` i.e. **embedded-friendly**!
28//! - No functionality besides securely zeroing memory!
29//! - (Optional) Custom derive support for zeroing complex structures
30//!
31//! ## Minimum Supported Rust Version
32//!
33//! Requires Rust **1.72** or newer.
34//!
35//! In the future, we reserve the right to change MSRV (i.e. MSRV is out-of-scope
36//! for this crate's SemVer guarantees), however when we do it will be accompanied
37//! by a minor version bump.
38//!
39//! ## Usage
40//!
41//! ```
42//! use zeroize::Zeroize;
43//!
44//! // Protip: don't embed secrets in your source code.
45//! // This is just an example.
46//! let mut secret = b"Air shield password: 1,2,3,4,5".to_vec();
47//! // [ ... ] open the air shield here
48//!
49//! // Now that we're done using the secret, zero it out.
50//! secret.zeroize();
51//! ```
52//!
53//! The [`Zeroize`] trait is impl'd on all of Rust's core scalar types including
54//! integers, floats, `bool`, and `char`.
55//!
56//! Additionally, it's implemented on slices and `IterMut`s of the above types.
57//!
58//! When the `alloc` feature is enabled (which it is by default), it's also
59//! impl'd for `Vec<T>` for the above types as well as `String`, where it provides
60//! [`Vec::clear`] / [`String::clear`]-like behavior (truncating to zero-length)
61//! but ensures the backing memory is securely zeroed with some caveats.
62//!
63//! With the `std` feature enabled (which it is **not** by default), [`Zeroize`]
64//! is also implemented for [`CString`]. After calling `zeroize()` on a `CString`,
65//! its internal buffer will contain exactly one nul byte. The backing
66//! memory is zeroed by converting it to a `Vec<u8>` and back into a `CString`.
67//! (NOTE: see "Stack/Heap Zeroing Notes" for important `Vec`/`String`/`CString` details)
68//!
69//! [`CString`]: https://doc.rust-lang.org/std/ffi/struct.CString.html
70//!
71//! The [`DefaultIsZeroes`] marker trait can be impl'd on types which also
72//! impl [`Default`], which implements [`Zeroize`] by overwriting a value with
73//! the default value.
74//!
75//! ## Custom Derive Support
76//!
77//! This crate has custom derive support for the `Zeroize` trait,
78//! gated under the `zeroize` crate's `zeroize_derive` Cargo feature,
79//! which automatically calls `zeroize()` on all members of a struct
80//! or tuple struct.
81//!
82//! Attributes supported for `Zeroize`:
83//!
84//! On the item level:
85//! - `#[zeroize(drop)]`: *deprecated* use `ZeroizeOnDrop` instead
86//! - `#[zeroize(bound = "T: MyTrait")]`: this replaces any trait bounds
87//! inferred by zeroize
88//!
89//! On the field level:
90//! - `#[zeroize(skip)]`: skips this field or variant when calling `zeroize()`
91//!
92//! Attributes supported for `ZeroizeOnDrop`:
93//!
94//! On the field level:
95//! - `#[zeroize(skip)]`: skips this field or variant when calling `zeroize()`
96//!
97//! Example which derives `Drop`:
98//!
99//! ```
100//! # #[cfg(feature = "zeroize_derive")]
101//! # {
102//! use zeroize::{Zeroize, ZeroizeOnDrop};
103//!
104//! // This struct will be zeroized on drop
105//! #[derive(Zeroize, ZeroizeOnDrop)]
106//! struct MyStruct([u8; 32]);
107//! # }
108//! ```
109//!
110//! Example which does not derive `Drop` (useful for e.g. `Copy` types)
111//!
112//! ```
113//! #[cfg(feature = "zeroize_derive")]
114//! # {
115//! use zeroize::Zeroize;
116//!
117//! // This struct will *NOT* be zeroized on drop
118//! #[derive(Copy, Clone, Zeroize)]
119//! struct MyStruct([u8; 32]);
120//! # }
121//! ```
122//!
123//! Example which only derives `Drop`:
124//!
125//! ```
126//! # #[cfg(feature = "zeroize_derive")]
127//! # {
128//! use zeroize::ZeroizeOnDrop;
129//!
130//! // This struct will be zeroized on drop
131//! #[derive(ZeroizeOnDrop)]
132//! struct MyStruct([u8; 32]);
133//! # }
134//! ```
135//!
136//! ## `Zeroizing<Z>`: wrapper for zeroizing arbitrary values on drop
137//!
138//! `Zeroizing<Z: Zeroize>` is a generic wrapper type that impls `Deref`
139//! and `DerefMut`, allowing access to an inner value of type `Z`, and also
140//! impls a `Drop` handler which calls `zeroize()` on its contents:
141//!
142//! ```
143//! use zeroize::Zeroizing;
144//!
145//! fn use_secret() {
146//! let mut secret = Zeroizing::new([0u8; 5]);
147//!
148//! // Set the air shield password
149//! // Protip (again): don't embed secrets in your source code.
150//! secret.copy_from_slice(&[1, 2, 3, 4, 5]);
151//! assert_eq!(secret.as_ref(), &[1, 2, 3, 4, 5]);
152//!
153//! // The contents of `secret` will be automatically zeroized on drop
154//! }
155//!
156//! # use_secret()
157//! ```
158//!
159//! ## What guarantees does this crate provide?
160//!
161//! This crate guarantees the following:
162//!
163//! 1. The zeroing operation can't be "optimized away" by the compiler.
164//! 2. All subsequent reads to memory will see "zeroized" values.
165//!
166//! LLVM's volatile semantics ensure #1 is true.
167//!
168//! Additionally, thanks to work by the [Unsafe Code Guidelines Working Group],
169//! we can now fairly confidently say #2 is true as well. Previously there were
170//! worries that the approach used by this crate (mixing volatile and
171//! non-volatile accesses) was undefined behavior due to language contained
172//! in the documentation for `write_volatile`, however after some discussion
173//! [these remarks have been removed] and the specific usage pattern in this
174//! crate is considered to be well-defined.
175//!
176//! Additionally this crate leverages [`core::sync::atomic::compiler_fence`]
177//! with the strictest ordering
178//! ([`Ordering::SeqCst`]) as a
179//! precaution to help ensure reads are not reordered before memory has been
180//! zeroed.
181//!
182//! All of that said, there is still potential for microarchitectural attacks
183//! (ala Spectre/Meltdown) to leak "zeroized" secrets through covert channels.
184//! This crate makes no guarantees that zeroized values cannot be leaked
185//! through such channels, as they represent flaws in the underlying hardware.
186//!
187//! ## Stack/Heap Zeroing Notes
188//!
189//! This crate can be used to zero values from either the stack or the heap.
190//!
191//! However, be aware several operations in Rust can unintentionally leave
192//! copies of data in memory. This includes but is not limited to:
193//!
194//! - Moves and [`Copy`]
195//! - Heap reallocation when using [`Vec`] and [`String`]
196//! - Borrowers of a reference making copies of the data
197//!
198//! [`Pin`][`core::pin::Pin`] can be leveraged in conjunction with this crate
199//! to ensure data kept on the stack isn't moved.
200//!
201//! The `Zeroize` impls for `Vec`, `String` and `CString` zeroize the entire
202//! capacity of their backing buffer, but cannot guarantee copies of the data
203//! were not previously made by buffer reallocation. It's therefore important
204//! when attempting to zeroize such buffers to initialize them to the correct
205//! capacity, and take care to prevent subsequent reallocation.
206//!
207//! The `secrecy` crate provides higher-level abstractions for eliminating
208//! usage patterns which can cause reallocations:
209//!
210//! <https://crates.io/crates/secrecy>
211//!
212//! ## What about: clearing registers, mlock, mprotect, etc?
213//!
214//! This crate is focused on providing simple, unobtrusive support for reliably
215//! zeroing memory using the best approach possible on stable Rust.
216//!
217//! Clearing registers is a difficult problem that can't easily be solved by
218//! something like a crate, and requires either inline ASM or rustc support.
219//! See <https://github.com/rust-lang/rust/issues/17046> for background on
220//! this particular problem.
221//!
222//! Other memory protection mechanisms are interesting and useful, but often
223//! overkill (e.g. defending against RAM scraping or attackers with swap access).
224//! In as much as there may be merit to these approaches, there are also many
225//! other crates that already implement more sophisticated memory protections.
226//! Such protections are explicitly out-of-scope for this crate.
227//!
228//! Zeroing memory is [good cryptographic hygiene] and this crate seeks to promote
229//! it in the most unobtrusive manner possible. This includes omitting complex
230//! `unsafe` memory protection systems and just trying to make the best memory
231//! zeroing crate available.
232//!
233//! [Zeroing memory securely is hard]: http://www.daemonology.net/blog/2014-09-04-how-to-zero-a-buffer.html
234//! [Unsafe Code Guidelines Working Group]: https://github.com/rust-lang/unsafe-code-guidelines
235//! [these remarks have been removed]: https://github.com/rust-lang/rust/pull/60972
236//! [good cryptographic hygiene]: https://github.com/veorq/cryptocoding#clean-memory-of-secret-data
237//! [`Ordering::SeqCst`]: core::sync::atomic::Ordering::SeqCst
238
239#[cfg(feature = "alloc")]
240extern crate alloc;
241
242#[cfg(feature = "std")]
243extern crate std;
244
245#[cfg(feature = "zeroize_derive")]
246pub use zeroize_derive::{Zeroize, ZeroizeOnDrop};
247
248#[cfg(target_arch = "aarch64")]
249mod aarch64;
250#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
251mod x86;
252
253use core::{
254 marker::{PhantomData, PhantomPinned},
255 mem::{self, MaybeUninit},
256 num::{
257 self, NonZeroI128, NonZeroI16, NonZeroI32, NonZeroI64, NonZeroI8, NonZeroIsize,
258 NonZeroU128, NonZeroU16, NonZeroU32, NonZeroU64, NonZeroU8, NonZeroUsize,
259 },
260 ops, ptr,
261 slice::IterMut,
262 sync::atomic,
263};
264
265#[cfg(feature = "alloc")]
266use alloc::{boxed::Box, string::String, vec::Vec};
267
268#[cfg(feature = "std")]
269use std::ffi::CString;
270
271/// Trait for securely erasing values from memory.
272pub trait Zeroize {
273 /// Zero out this object from memory using Rust intrinsics which ensure the
274 /// zeroization operation is not "optimized away" by the compiler.
275 fn zeroize(&mut self);
276}
277
278/// Marker trait signifying that this type will [`Zeroize::zeroize`] itself on [`Drop`].
279pub trait ZeroizeOnDrop {}
280
281/// Marker trait for types whose [`Default`] is the desired zeroization result
282pub trait DefaultIsZeroes: Copy + Default + Sized {}
283
284/// Fallible trait for representing cases where zeroization may or may not be
285/// possible.
286///
287/// This is primarily useful for scenarios like reference counted data, where
288/// zeroization is only possible when the last reference is dropped.
289pub trait TryZeroize {
290 /// Try to zero out this object from memory using Rust intrinsics which
291 /// ensure the zeroization operation is not "optimized away" by the
292 /// compiler.
293 #[must_use]
294 fn try_zeroize(&mut self) -> bool;
295}
296
297impl<Z> Zeroize for Z
298where
299 Z: DefaultIsZeroes,
300{
301 fn zeroize(&mut self) {
302 volatile_write(self, Z::default());
303 atomic_fence();
304 }
305}
306
307macro_rules! impl_zeroize_with_default {
308 ($($type:ty),+) => {
309 $(impl DefaultIsZeroes for $type {})+
310 };
311}
312
313#[rustfmt::skip]
314impl_zeroize_with_default! {
315 PhantomPinned, (), bool, char,
316 f32, f64,
317 i8, i16, i32, i64, i128, isize,
318 u8, u16, u32, u64, u128, usize
319}
320
321/// `PhantomPinned` is zero sized so provide a ZeroizeOnDrop implementation.
322impl ZeroizeOnDrop for PhantomPinned {}
323
324/// `()` is zero sized so provide a ZeroizeOnDrop implementation.
325impl ZeroizeOnDrop for () {}
326
327macro_rules! impl_zeroize_for_non_zero {
328 ($($type:ty),+) => {
329 $(
330 impl Zeroize for $type {
331 fn zeroize(&mut self) {
332 const ONE: $type = match <$type>::new(1) {
333 Some(one) => one,
334 None => unreachable!(),
335 };
336 volatile_write(self, ONE);
337 atomic_fence();
338 }
339 }
340 )+
341 };
342}
343
344impl_zeroize_for_non_zero!(
345 NonZeroI8,
346 NonZeroI16,
347 NonZeroI32,
348 NonZeroI64,
349 NonZeroI128,
350 NonZeroIsize,
351 NonZeroU8,
352 NonZeroU16,
353 NonZeroU32,
354 NonZeroU64,
355 NonZeroU128,
356 NonZeroUsize
357);
358
359impl<Z> Zeroize for num::Wrapping<Z>
360where
361 Z: Zeroize,
362{
363 fn zeroize(&mut self) {
364 self.0.zeroize();
365 }
366}
367
368/// Impl [`Zeroize`] on arrays of types that impl [`Zeroize`].
369impl<Z, const N: usize> Zeroize for [Z; N]
370where
371 Z: Zeroize,
372{
373 fn zeroize(&mut self) {
374 self.iter_mut().zeroize();
375 }
376}
377
378/// Impl [`ZeroizeOnDrop`] on arrays of types that impl [`ZeroizeOnDrop`].
379impl<Z, const N: usize> ZeroizeOnDrop for [Z; N] where Z: ZeroizeOnDrop {}
380
381impl<Z> Zeroize for IterMut<'_, Z>
382where
383 Z: Zeroize,
384{
385 fn zeroize(&mut self) {
386 for elem in self {
387 elem.zeroize();
388 }
389 }
390}
391
392impl<Z> Zeroize for Option<Z>
393where
394 Z: Zeroize,
395{
396 fn zeroize(&mut self) {
397 if let Some(value) = self {
398 value.zeroize();
399
400 // Ensures self is None and that the value was dropped. Without the take, the drop
401 // of the (zeroized) value isn't called, which might lead to a leak or other
402 // unexpected behavior. For example, if this were Option<Vec<T>>, the above call to
403 // zeroize would not free the allocated memory, but the the `take` call will.
404 self.take();
405 }
406
407 // Ensure that if the `Option` were previously `Some` but a value was copied/moved out
408 // that the remaining space in the `Option` is zeroized.
409 //
410 // Safety:
411 //
412 // The memory pointed to by `self` is valid for `mem::size_of::<Self>()` bytes.
413 // It is also properly aligned, because `u8` has an alignment of `1`.
414 unsafe {
415 volatile_set((self as *mut Self).cast::<u8>(), 0, mem::size_of::<Self>());
416 }
417
418 // Ensures self is overwritten with the `None` bit pattern. volatile_write can't be
419 // used because Option<Z> is not copy.
420 //
421 // Safety:
422 //
423 // self is safe to replace with `None`, which the take() call above should have
424 // already done semantically. Any value which needed to be dropped will have been
425 // done so by take().
426 unsafe { ptr::write_volatile(self, None) }
427
428 atomic_fence();
429 }
430}
431
432impl<Z> ZeroizeOnDrop for Option<Z> where Z: ZeroizeOnDrop {}
433
434/// Impl [`Zeroize`] on [`MaybeUninit`] types.
435///
436/// This fills the memory with zeroes.
437/// Note that this ignore invariants that `Z` might have, because
438/// [`MaybeUninit`] removes all invariants.
439impl<Z> Zeroize for MaybeUninit<Z> {
440 fn zeroize(&mut self) {
441 // Safety:
442 // `MaybeUninit` is valid for any byte pattern, including zeros.
443 unsafe { ptr::write_volatile(self, MaybeUninit::zeroed()) }
444 atomic_fence();
445 }
446}
447
448/// Impl [`Zeroize`] on slices of [`MaybeUninit`] types.
449///
450/// This impl can eventually be optimized using an memset intrinsic,
451/// such as [`core::intrinsics::volatile_set_memory`].
452///
453/// This fills the slice with zeroes.
454///
455/// Note that this ignore invariants that `Z` might have, because
456/// [`MaybeUninit`] removes all invariants.
457impl<Z> Zeroize for [MaybeUninit<Z>] {
458 fn zeroize(&mut self) {
459 let ptr = self.as_mut_ptr().cast::<MaybeUninit<u8>>();
460 let size = self.len().checked_mul(mem::size_of::<Z>()).unwrap();
461 assert!(size <= isize::MAX as usize);
462
463 // Safety:
464 //
465 // This is safe, because every valid pointer is well aligned for u8
466 // and it is backed by a single allocated object for at least `self.len() * size_pf::<Z>()` bytes.
467 // and 0 is a valid value for `MaybeUninit<Z>`
468 // The memory of the slice should not wrap around the address space.
469 unsafe { volatile_set(ptr, MaybeUninit::zeroed(), size) }
470 atomic_fence();
471 }
472}
473
474/// Impl [`Zeroize`] on slices of types that can be zeroized with [`Default`].
475///
476/// This impl can eventually be optimized using an memset intrinsic,
477/// such as [`core::intrinsics::volatile_set_memory`]. For that reason the
478/// blanket impl on slices is bounded by [`DefaultIsZeroes`].
479///
480/// To zeroize a mut slice of `Z: Zeroize` which does not impl
481/// [`DefaultIsZeroes`], call `iter_mut().zeroize()`.
482impl<Z> Zeroize for [Z]
483where
484 Z: DefaultIsZeroes,
485{
486 fn zeroize(&mut self) {
487 assert!(self.len() <= isize::MAX as usize);
488
489 // Safety:
490 //
491 // This is safe, because the slice is well aligned and is backed by a single allocated
492 // object for at least `self.len()` elements of type `Z`.
493 // `self.len()` is also not larger than an `isize`, because of the assertion above.
494 // The memory of the slice should not wrap around the address space.
495 unsafe { volatile_set(self.as_mut_ptr(), Z::default(), self.len()) };
496 atomic_fence();
497 }
498}
499
500impl Zeroize for str {
501 fn zeroize(&mut self) {
502 // Safety:
503 // A zeroized byte slice is a valid UTF-8 string.
504 unsafe { self.as_bytes_mut().zeroize() }
505 }
506}
507
508/// [`PhantomData`] is always zero sized so provide a [`Zeroize`] implementation.
509impl<Z> Zeroize for PhantomData<Z> {
510 fn zeroize(&mut self) {}
511}
512
513/// [`PhantomData` is always zero sized so provide a ZeroizeOnDrop implementation.
514impl<Z> ZeroizeOnDrop for PhantomData<Z> {}
515
516macro_rules! impl_zeroize_tuple {
517 ( $( $type_name:ident ),+ ) => {
518 impl<$($type_name: Zeroize),+> Zeroize for ($($type_name,)+) {
519 fn zeroize(&mut self) {
520 #[allow(non_snake_case)]
521 let ($($type_name,)+) = self;
522 $($type_name.zeroize());+
523 }
524 }
525
526 impl<$($type_name: ZeroizeOnDrop),+> ZeroizeOnDrop for ($($type_name,)+) { }
527 }
528}
529
530// Generic implementations for tuples up to 10 parameters.
531impl_zeroize_tuple!(A);
532impl_zeroize_tuple!(A, B);
533impl_zeroize_tuple!(A, B, C);
534impl_zeroize_tuple!(A, B, C, D);
535impl_zeroize_tuple!(A, B, C, D, E);
536impl_zeroize_tuple!(A, B, C, D, E, F);
537impl_zeroize_tuple!(A, B, C, D, E, F, G);
538impl_zeroize_tuple!(A, B, C, D, E, F, G, H);
539impl_zeroize_tuple!(A, B, C, D, E, F, G, H, I);
540impl_zeroize_tuple!(A, B, C, D, E, F, G, H, I, J);
541
542#[cfg(feature = "alloc")]
543impl<Z> Zeroize for Vec<Z>
544where
545 Z: Zeroize,
546{
547 /// "Best effort" zeroization for `Vec`.
548 ///
549 /// Ensures the entire capacity of the `Vec` is zeroed. Cannot ensure that
550 /// previous reallocations did not leave values on the heap.
551 fn zeroize(&mut self) {
552 // Zeroize all the initialized elements.
553 self.iter_mut().zeroize();
554
555 // Set the Vec's length to 0 and drop all the elements.
556 self.clear();
557
558 // Zero the full capacity of `Vec`.
559 self.spare_capacity_mut().zeroize();
560 }
561}
562
563#[cfg(feature = "alloc")]
564impl<Z> ZeroizeOnDrop for Vec<Z> where Z: ZeroizeOnDrop {}
565
566#[cfg(feature = "alloc")]
567impl<Z> Zeroize for Box<[Z]>
568where
569 Z: Zeroize,
570{
571 /// Unlike `Vec`, `Box<[Z]>` cannot reallocate, so we can be sure that we are not leaving
572 /// values on the heap.
573 fn zeroize(&mut self) {
574 self.iter_mut().zeroize();
575 }
576}
577
578#[cfg(feature = "alloc")]
579impl<Z> ZeroizeOnDrop for Box<[Z]> where Z: ZeroizeOnDrop {}
580
581#[cfg(feature = "alloc")]
582impl Zeroize for Box<str> {
583 fn zeroize(&mut self) {
584 self.as_mut().zeroize();
585 }
586}
587
588#[cfg(feature = "alloc")]
589impl Zeroize for String {
590 fn zeroize(&mut self) {
591 unsafe { self.as_mut_vec() }.zeroize();
592 }
593}
594
595#[cfg(feature = "std")]
596impl Zeroize for CString {
597 fn zeroize(&mut self) {
598 // mem::take uses replace internally to swap the pointer
599 // Unfortunately this results in an allocation for a Box::new(&[0]) as CString must
600 // contain a trailing zero byte
601 let this = mem::take(self);
602
603 // - CString::into_bytes_with_nul calls ::into_vec which takes ownership of the heap pointer
604 // as a Vec<u8>
605 // - Calling .zeroize() on the resulting vector clears out the bytes
606 // From: https://github.com/RustCrypto/utils/pull/759#issuecomment-1087976570
607 let mut buf = this.into_bytes_with_nul();
608 buf.zeroize();
609
610 // expect() should never fail, because zeroize() truncates the Vec
611 let zeroed = CString::new(buf).expect("buf not truncated");
612
613 // Replace self by the zeroed CString to maintain the original ptr of the buffer
614 let _ = mem::replace(self, zeroed);
615 }
616}
617
618/// `Zeroizing` is a a wrapper for any `Z: Zeroize` type which implements a
619/// `Drop` handler which zeroizes dropped values.
620#[derive(Debug, Default, Eq, PartialEq)]
621pub struct Zeroizing<Z: Zeroize>(Z);
622
623impl<Z> Zeroizing<Z>
624where
625 Z: Zeroize,
626{
627 /// Move value inside a `Zeroizing` wrapper which ensures it will be
628 /// zeroized when it's dropped.
629 #[inline(always)]
630 pub fn new(value: Z) -> Self {
631 Self(value)
632 }
633}
634
635impl<Z: Zeroize + Clone> Clone for Zeroizing<Z> {
636 #[inline(always)]
637 fn clone(&self) -> Self {
638 Self(self.0.clone())
639 }
640
641 #[inline(always)]
642 fn clone_from(&mut self, source: &Self) {
643 self.0.zeroize();
644 self.0.clone_from(&source.0);
645 }
646}
647
648impl<Z> From<Z> for Zeroizing<Z>
649where
650 Z: Zeroize,
651{
652 #[inline(always)]
653 fn from(value: Z) -> Zeroizing<Z> {
654 Zeroizing(value)
655 }
656}
657
658impl<Z> ops::Deref for Zeroizing<Z>
659where
660 Z: Zeroize,
661{
662 type Target = Z;
663
664 #[inline(always)]
665 fn deref(&self) -> &Z {
666 &self.0
667 }
668}
669
670impl<Z> ops::DerefMut for Zeroizing<Z>
671where
672 Z: Zeroize,
673{
674 #[inline(always)]
675 fn deref_mut(&mut self) -> &mut Z {
676 &mut self.0
677 }
678}
679
680impl<T, Z> AsRef<T> for Zeroizing<Z>
681where
682 T: ?Sized,
683 Z: AsRef<T> + Zeroize,
684{
685 #[inline(always)]
686 fn as_ref(&self) -> &T {
687 self.0.as_ref()
688 }
689}
690
691impl<T, Z> AsMut<T> for Zeroizing<Z>
692where
693 T: ?Sized,
694 Z: AsMut<T> + Zeroize,
695{
696 #[inline(always)]
697 fn as_mut(&mut self) -> &mut T {
698 self.0.as_mut()
699 }
700}
701
702impl<Z> Zeroize for Zeroizing<Z>
703where
704 Z: Zeroize,
705{
706 fn zeroize(&mut self) {
707 self.0.zeroize();
708 }
709}
710
711impl<Z> ZeroizeOnDrop for Zeroizing<Z> where Z: Zeroize {}
712
713impl<Z> Drop for Zeroizing<Z>
714where
715 Z: Zeroize,
716{
717 fn drop(&mut self) {
718 self.0.zeroize()
719 }
720}
721
722#[cfg(feature = "serde")]
723impl<Z> serde::Serialize for Zeroizing<Z>
724where
725 Z: Zeroize + serde::Serialize,
726{
727 #[inline(always)]
728 fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
729 where
730 S: serde::Serializer,
731 {
732 self.0.serialize(serializer)
733 }
734}
735
736#[cfg(feature = "serde")]
737impl<'de, Z> serde::Deserialize<'de> for Zeroizing<Z>
738where
739 Z: Zeroize + serde::Deserialize<'de>,
740{
741 #[inline(always)]
742 fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
743 where
744 D: serde::Deserializer<'de>,
745 {
746 Ok(Self(Z::deserialize(deserializer)?))
747 }
748}
749
750/// Use fences to prevent accesses from being reordered before this
751/// point, which should hopefully help ensure that all accessors
752/// see zeroes after this point.
753#[inline(always)]
754fn atomic_fence() {
755 atomic::compiler_fence(atomic::Ordering::SeqCst);
756}
757
758/// Perform a volatile write to the destination
759#[inline(always)]
760fn volatile_write<T: Copy + Sized>(dst: &mut T, src: T) {
761 unsafe { ptr::write_volatile(dst, src) }
762}
763
764/// Perform a volatile `memset` operation which fills a slice with a value
765///
766/// Safety:
767/// The memory pointed to by `dst` must be a single allocated object that is valid for `count`
768/// contiguous elements of `T`.
769/// `count` must not be larger than an `isize`.
770/// `dst` being offset by `mem::size_of::<T> * count` bytes must not wrap around the address space.
771/// Also `dst` must be properly aligned.
772#[inline(always)]
773unsafe fn volatile_set<T: Copy + Sized>(dst: *mut T, src: T, count: usize) {
774 // TODO(tarcieri): use `volatile_set_memory` when stabilized
775 for i in 0..count {
776 // Safety:
777 //
778 // This is safe because there is room for at least `count` objects of type `T` in the
779 // allocation pointed to by `dst`, because `count <= isize::MAX` and because
780 // `dst.add(count)` must not wrap around the address space.
781 let ptr = dst.add(i);
782
783 // Safety:
784 //
785 // This is safe, because the pointer is valid and because `dst` is well aligned for `T` and
786 // `ptr` is an offset of `dst` by a multiple of `mem::size_of::<T>()` bytes.
787 ptr::write_volatile(ptr, src);
788 }
789}
790
791/// Zeroizes a flat type/struct. Only zeroizes the values that it owns, and it does not work on
792/// dynamically sized values or trait objects. It would be inefficient to use this function on a
793/// type that already implements `ZeroizeOnDrop`.
794///
795/// # Safety
796/// - The type must not contain references to outside data or dynamically sized data, such as
797/// `Vec<T>` or `String`.
798/// - Values stored in the type must not have `Drop` impls.
799/// - This function can invalidate the type if it is used after this function is called on it.
800/// It is advisable to call this function only in `impl Drop`.
801/// - The bit pattern of all zeroes must be valid for the data being zeroized. This may not be
802/// true for enums and pointers.
803///
804/// # Incompatible data types
805/// Some data types that cannot be safely zeroized using `zeroize_flat_type` include,
806/// but are not limited to:
807/// - References: `&T` and `&mut T`
808/// - Non-nullable types: `NonNull<T>`, `NonZeroU32`, etc.
809/// - Enums with explicit non-zero tags.
810/// - Smart pointers and collections: `Arc<T>`, `Box<T>`, `Vec<T>`, `HashMap<K, V>`, `String`, etc.
811///
812/// # Examples
813/// Safe usage for a struct containing strictly flat data:
814/// ```
815/// use zeroize::{ZeroizeOnDrop, zeroize_flat_type};
816///
817/// struct DataToZeroize {
818/// flat_data_1: [u8; 32],
819/// flat_data_2: SomeMoreFlatData,
820/// }
821///
822/// struct SomeMoreFlatData(u64);
823///
824/// impl Drop for DataToZeroize {
825/// fn drop(&mut self) {
826/// unsafe { zeroize_flat_type(self as *mut Self) }
827/// }
828/// }
829/// impl ZeroizeOnDrop for DataToZeroize {}
830///
831/// let mut data = DataToZeroize {
832/// flat_data_1: [3u8; 32],
833/// flat_data_2: SomeMoreFlatData(123u64)
834/// };
835///
836/// // data gets zeroized when dropped
837/// ```
838#[inline(always)]
839pub unsafe fn zeroize_flat_type<F: Sized>(data: *mut F) {
840 let size = mem::size_of::<F>();
841 // Safety:
842 //
843 // This is safe because `mem::size_of<T>()` returns the exact size of the object in memory, and
844 // `data_ptr` points directly to the first byte of the data.
845 volatile_set(data as *mut u8, 0, size);
846 atomic_fence()
847}
848
849/// Internal module used as support for `AssertZeroizeOnDrop`.
850#[doc(hidden)]
851pub mod __internal {
852 use super::*;
853
854 /// Auto-deref workaround for deriving `ZeroizeOnDrop`.
855 pub trait AssertZeroizeOnDrop {
856 fn zeroize_or_on_drop(self);
857 }
858
859 impl<T: ZeroizeOnDrop + ?Sized> AssertZeroizeOnDrop for &&mut T {
860 fn zeroize_or_on_drop(self) {}
861 }
862
863 /// Auto-deref workaround for deriving `ZeroizeOnDrop`.
864 pub trait AssertZeroize {
865 fn zeroize_or_on_drop(&mut self);
866 }
867
868 impl<T: Zeroize + ?Sized> AssertZeroize for T {
869 fn zeroize_or_on_drop(&mut self) {
870 self.zeroize()
871 }
872 }
873}