]> git.proxmox.com Git - rustc.git/blob - library/core/src/ptr/mod.rs
New upstream version 1.64.0+dfsg1
[rustc.git] / library / core / src / ptr / mod.rs
1 //! Manually manage memory through raw pointers.
2 //!
3 //! *[See also the pointer primitive types](pointer).*
4 //!
5 //! # Safety
6 //!
7 //! Many functions in this module take raw pointers as arguments and read from
8 //! or write to them. For this to be safe, these pointers must be *valid*.
9 //! Whether a pointer is valid depends on the operation it is used for
10 //! (read or write), and the extent of the memory that is accessed (i.e.,
11 //! how many bytes are read/written). Most functions use `*mut T` and `*const T`
12 //! to access only a single value, in which case the documentation omits the size
13 //! and implicitly assumes it to be `size_of::<T>()` bytes.
14 //!
15 //! The precise rules for validity are not determined yet. The guarantees that are
16 //! provided at this point are very minimal:
17 //!
18 //! * A [null] pointer is *never* valid, not even for accesses of [size zero][zst].
19 //! * For a pointer to be valid, it is necessary, but not always sufficient, that the pointer
20 //! be *dereferenceable*: the memory range of the given size starting at the pointer must all be
21 //! within the bounds of a single allocated object. Note that in Rust,
22 //! every (stack-allocated) variable is considered a separate allocated object.
23 //! * Even for operations of [size zero][zst], the pointer must not be pointing to deallocated
24 //! memory, i.e., deallocation makes pointers invalid even for zero-sized operations. However,
25 //! casting any non-zero integer *literal* to a pointer is valid for zero-sized accesses, even if
26 //! some memory happens to exist at that address and gets deallocated. This corresponds to writing
27 //! your own allocator: allocating zero-sized objects is not very hard. The canonical way to
28 //! obtain a pointer that is valid for zero-sized accesses is [`NonNull::dangling`].
29 //FIXME: mention `ptr::invalid` above, once it is stable.
30 //! * All accesses performed by functions in this module are *non-atomic* in the sense
31 //! of [atomic operations] used to synchronize between threads. This means it is
32 //! undefined behavior to perform two concurrent accesses to the same location from different
33 //! threads unless both accesses only read from memory. Notice that this explicitly
34 //! includes [`read_volatile`] and [`write_volatile`]: Volatile accesses cannot
35 //! be used for inter-thread synchronization.
36 //! * The result of casting a reference to a pointer is valid for as long as the
37 //! underlying object is live and no reference (just raw pointers) is used to
38 //! access the same memory.
39 //!
40 //! These axioms, along with careful use of [`offset`] for pointer arithmetic,
41 //! are enough to correctly implement many useful things in unsafe code. Stronger guarantees
42 //! will be provided eventually, as the [aliasing] rules are being determined. For more
43 //! information, see the [book] as well as the section in the reference devoted
44 //! to [undefined behavior][ub].
45 //!
46 //! ## Alignment
47 //!
48 //! Valid raw pointers as defined above are not necessarily properly aligned (where
49 //! "proper" alignment is defined by the pointee type, i.e., `*const T` must be
50 //! aligned to `mem::align_of::<T>()`). However, most functions require their
51 //! arguments to be properly aligned, and will explicitly state
52 //! this requirement in their documentation. Notable exceptions to this are
53 //! [`read_unaligned`] and [`write_unaligned`].
54 //!
55 //! When a function requires proper alignment, it does so even if the access
56 //! has size 0, i.e., even if memory is not actually touched. Consider using
57 //! [`NonNull::dangling`] in such cases.
58 //!
59 //! ## Allocated object
60 //!
61 //! For several operations, such as [`offset`] or field projections (`expr.field`), the notion of an
62 //! "allocated object" becomes relevant. An allocated object is a contiguous region of memory.
63 //! Common examples of allocated objects include stack-allocated variables (each variable is a
64 //! separate allocated object), heap allocations (each allocation created by the global allocator is
65 //! a separate allocated object), and `static` variables.
66 //!
67 //!
68 //! # Strict Provenance
69 //!
70 //! **The following text is non-normative, insufficiently formal, and is an extremely strict
71 //! interpretation of provenance. It's ok if your code doesn't strictly conform to it.**
72 //!
73 //! [Strict Provenance][] is an experimental set of APIs that help tools that try
74 //! to validate the memory-safety of your program's execution. Notably this includes [Miri][]
75 //! and [CHERI][], which can detect when you access out of bounds memory or otherwise violate
76 //! Rust's memory model.
77 //!
78 //! Provenance must exist in some form for any programming
79 //! language compiled for modern computer architectures, but specifying a model for provenance
80 //! in a way that is useful to both compilers and programmers is an ongoing challenge.
81 //! The [Strict Provenance][] experiment seeks to explore the question: *what if we just said you
82 //! couldn't do all the nasty operations that make provenance so messy?*
83 //!
84 //! What APIs would have to be removed? What APIs would have to be added? How much would code
85 //! have to change, and is it worse or better now? Would any patterns become truly inexpressible?
86 //! Could we carve out special exceptions for those patterns? Should we?
87 //!
88 //! A secondary goal of this project is to see if we can disambiguate the many functions of
89 //! pointer<->integer casts enough for the definition of `usize` to be loosened so that it
90 //! isn't *pointer*-sized but address-space/offset/allocation-sized (we'll probably continue
91 //! to conflate these notions). This would potentially make it possible to more efficiently
92 //! target platforms where pointers are larger than offsets, such as CHERI and maybe some
93 //! segmented architecures.
94 //!
95 //! ## Provenance
96 //!
97 //! **This section is *non-normative* and is part of the [Strict Provenance][] experiment.**
98 //!
99 //! Pointers are not *simply* an "integer" or "address". For instance, it's uncontroversial
100 //! to say that a Use After Free is clearly Undefined Behaviour, even if you "get lucky"
101 //! and the freed memory gets reallocated before your read/write (in fact this is the
102 //! worst-case scenario, UAFs would be much less concerning if this didn't happen!).
103 //! To rationalize this claim, pointers need to somehow be *more* than just their addresses:
104 //! they must have provenance.
105 //!
106 //! When an allocation is created, that allocation has a unique Original Pointer. For alloc
107 //! APIs this is literally the pointer the call returns, and for local variables and statics,
108 //! this is the name of the variable/static. This is mildly overloading the term "pointer"
109 //! for the sake of brevity/exposition.
110 //!
111 //! The Original Pointer for an allocation is guaranteed to have unique access to the entire
112 //! allocation and *only* that allocation. In this sense, an allocation can be thought of
113 //! as a "sandbox" that cannot be broken into or out of. *Provenance* is the permission
114 //! to access an allocation's sandbox and has both a *spatial* and *temporal* component:
115 //!
116 //! * Spatial: A range of bytes that the pointer is allowed to access.
117 //! * Temporal: The lifetime (of the allocation) that access to these bytes is tied to.
118 //!
119 //! Spatial provenance makes sure you don't go beyond your sandbox, while temporal provenance
120 //! makes sure that you can't "get lucky" after your permission to access some memory
121 //! has been revoked (either through deallocations or borrows expiring).
122 //!
123 //! Provenance is implicitly shared with all pointers transitively derived from
124 //! The Original Pointer through operations like [`offset`], borrowing, and pointer casts.
125 //! Some operations may *shrink* the derived provenance, limiting how much memory it can
126 //! access or how long it's valid for (i.e. borrowing a subfield and subslicing).
127 //!
128 //! Shrinking provenance cannot be undone: even if you "know" there is a larger allocation, you
129 //! can't derive a pointer with a larger provenance. Similarly, you cannot "recombine"
130 //! two contiguous provenances back into one (i.e. with a `fn merge(&[T], &[T]) -> &[T]`).
131 //!
132 //! A reference to a value always has provenance over exactly the memory that field occupies.
133 //! A reference to a slice always has provenance over exactly the range that slice describes.
134 //!
135 //! If an allocation is deallocated, all pointers with provenance to that allocation become
136 //! invalidated, and effectively lose their provenance.
137 //!
138 //! The strict provenance experiment is mostly only interested in exploring stricter *spatial*
139 //! provenance. In this sense it can be thought of as a subset of the more ambitious and
140 //! formal [Stacked Borrows][] research project, which is what tools like [Miri][] are based on.
141 //! In particular, Stacked Borrows is necessary to properly describe what borrows are allowed
142 //! to do and when they become invalidated. This necessarily involves much more complex
143 //! *temporal* reasoning than simply identifying allocations. Adjusting APIs and code
144 //! for the strict provenance experiment will also greatly help Stacked Borrows.
145 //!
146 //!
147 //! ## Pointer Vs Addresses
148 //!
149 //! **This section is *non-normative* and is part of the [Strict Provenance][] experiment.**
150 //!
151 //! One of the largest historical issues with trying to define provenance is that programmers
152 //! freely convert between pointers and integers. Once you allow for this, it generally becomes
153 //! impossible to accurately track and preserve provenance information, and you need to appeal
154 //! to very complex and unreliable heuristics. But of course, converting between pointers and
155 //! integers is very useful, so what can we do?
156 //!
157 //! Also did you know WASM is actually a "Harvard Architecture"? As in function pointers are
158 //! handled completely differently from data pointers? And we kind of just shipped Rust on WASM
159 //! without really addressing the fact that we let you freely convert between function pointers
160 //! and data pointers, because it mostly Just Works? Let's just put that on the "pointer casts
161 //! are dubious" pile.
162 //!
163 //! Strict Provenance attempts to square these circles by decoupling Rust's traditional conflation
164 //! of pointers and `usize` (and `isize`), and defining a pointer to semantically contain the
165 //! following information:
166 //!
167 //! * The **address-space** it is part of (e.g. "data" vs "code" in WASM).
168 //! * The **address** it points to, which can be represented by a `usize`.
169 //! * The **provenance** it has, defining the memory it has permission to access.
170 //!
171 //! Under Strict Provenance, a usize *cannot* accurately represent a pointer, and converting from
172 //! a pointer to a usize is generally an operation which *only* extracts the address. It is
173 //! therefore *impossible* to construct a valid pointer from a usize because there is no way
174 //! to restore the address-space and provenance. In other words, pointer-integer-pointer
175 //! roundtrips are not possible (in the sense that the resulting pointer is not dereferencable).
176 //!
177 //! The key insight to making this model *at all* viable is the [`with_addr`][] method:
178 //!
179 //! ```text
180 //! /// Creates a new pointer with the given address.
181 //! ///
182 //! /// This performs the same operation as an `addr as ptr` cast, but copies
183 //! /// the *address-space* and *provenance* of `self` to the new pointer.
184 //! /// This allows us to dynamically preserve and propagate this important
185 //! /// information in a way that is otherwise impossible with a unary cast.
186 //! ///
187 //! /// This is equivalent to using `wrapping_offset` to offset `self` to the
188 //! /// given address, and therefore has all the same capabilities and restrictions.
189 //! pub fn with_addr(self, addr: usize) -> Self;
190 //! ```
191 //!
192 //! So you're still able to drop down to the address representation and do whatever
193 //! clever bit tricks you want *as long as* you're able to keep around a pointer
194 //! into the allocation you care about that can "reconstitute" the other parts of the pointer.
195 //! Usually this is very easy, because you only are taking a pointer, messing with the address,
196 //! and then immediately converting back to a pointer. To make this use case more ergonomic,
197 //! we provide the [`map_addr`][] method.
198 //!
199 //! To help make it clear that code is "following" Strict Provenance semantics, we also provide an
200 //! [`addr`][] method which promises that the returned address is not part of a
201 //! pointer-usize-pointer roundtrip. In the future we may provide a lint for pointer<->integer
202 //! casts to help you audit if your code conforms to strict provenance.
203 //!
204 //!
205 //! ## Using Strict Provenance
206 //!
207 //! Most code needs no changes to conform to strict provenance, as the only really concerning
208 //! operation that *wasn't* obviously already Undefined Behaviour is casts from usize to a
209 //! pointer. For code which *does* cast a usize to a pointer, the scope of the change depends
210 //! on exactly what you're doing.
211 //!
212 //! In general you just need to make sure that if you want to convert a usize address to a
213 //! pointer and then use that pointer to read/write memory, you need to keep around a pointer
214 //! that has sufficient provenance to perform that read/write itself. In this way all of your
215 //! casts from an address to a pointer are essentially just applying offsets/indexing.
216 //!
217 //! This is generally trivial to do for simple cases like tagged pointers *as long as you
218 //! represent the tagged pointer as an actual pointer and not a usize*. For instance:
219 //!
220 //! ```
221 //! #![feature(strict_provenance)]
222 //!
223 //! unsafe {
224 //! // A flag we want to pack into our pointer
225 //! static HAS_DATA: usize = 0x1;
226 //! static FLAG_MASK: usize = !HAS_DATA;
227 //!
228 //! // Our value, which must have enough alignment to have spare least-significant-bits.
229 //! let my_precious_data: u32 = 17;
230 //! assert!(core::mem::align_of::<u32>() > 1);
231 //!
232 //! // Create a tagged pointer
233 //! let ptr = &my_precious_data as *const u32;
234 //! let tagged = ptr.map_addr(|addr| addr | HAS_DATA);
235 //!
236 //! // Check the flag:
237 //! if tagged.addr() & HAS_DATA != 0 {
238 //! // Untag and read the pointer
239 //! let data = *tagged.map_addr(|addr| addr & FLAG_MASK);
240 //! assert_eq!(data, 17);
241 //! } else {
242 //! unreachable!()
243 //! }
244 //! }
245 //! ```
246 //!
247 //! (Yes, if you've been using AtomicUsize for pointers in concurrent datastructures, you should
248 //! be using AtomicPtr instead. If that messes up the way you atomically manipulate pointers,
249 //! we would like to know why, and what needs to be done to fix it.)
250 //!
251 //! Something more complicated and just generally *evil* like an XOR-List requires more significant
252 //! changes like allocating all nodes in a pre-allocated Vec or Arena and using a pointer
253 //! to the whole allocation to reconstitute the XORed addresses.
254 //!
255 //! Situations where a valid pointer *must* be created from just an address, such as baremetal code
256 //! accessing a memory-mapped interface at a fixed address, are an open question on how to support.
257 //! These situations *will* still be allowed, but we might require some kind of "I know what I'm
258 //! doing" annotation to explain the situation to the compiler. It's also possible they need no
259 //! special attention at all, because they're generally accessing memory outside the scope of
260 //! "the abstract machine", or already using "I know what I'm doing" annotations like "volatile".
261 //!
262 //! Under [Strict Provenance] it is Undefined Behaviour to:
263 //!
264 //! * Access memory through a pointer that does not have provenance over that memory.
265 //!
266 //! * [`offset`] a pointer to or from an address it doesn't have provenance over.
267 //! This means it's always UB to offset a pointer derived from something deallocated,
268 //! even if the offset is 0. Note that a pointer "one past the end" of its provenance
269 //! is not actually outside its provenance, it just has 0 bytes it can load/store.
270 //!
271 //! But it *is* still sound to:
272 //!
273 //! * Create an invalid pointer from just an address (see [`ptr::invalid`][]). This can
274 //! be used for sentinel values like `null` *or* to represent a tagged pointer that will
275 //! never be dereferencable. In general, it is always sound for an integer to pretend
276 //! to be a pointer "for fun" as long as you don't use operations on it which require
277 //! it to be valid (offset, read, write, etc).
278 //!
279 //! * Forge an allocation of size zero at any sufficiently aligned non-null address.
280 //! i.e. the usual "ZSTs are fake, do what you want" rules apply *but* this only applies
281 //! for actual forgery (integers cast to pointers). If you borrow some struct's field
282 //! that *happens* to be zero-sized, the resulting pointer will have provenance tied to
283 //! that allocation and it will still get invalidated if the allocation gets deallocated.
284 //! In the future we may introduce an API to make such a forged allocation explicit.
285 //!
286 //! * [`wrapping_offset`][] a pointer outside its provenance. This includes invalid pointers
287 //! which have "no" provenance. Unfortunately there may be practical limits on this for a
288 //! particular platform, and it's an open question as to how to specify this (if at all).
289 //! Notably, [CHERI][] relies on a compression scheme that can't handle a
290 //! pointer getting offset "too far" out of bounds. If this happens, the address
291 //! returned by `addr` will be the value you expect, but the provenance will get invalidated
292 //! and using it to read/write will fault. The details of this are architecture-specific
293 //! and based on alignment, but the buffer on either side of the pointer's range is pretty
294 //! generous (think kilobytes, not bytes).
295 //!
296 //! * Compare arbitrary pointers by address. Addresses *are* just integers and so there is
297 //! always a coherent answer, even if the pointers are invalid or from different
298 //! address-spaces/provenances. Of course, comparing addresses from different address-spaces
299 //! is generally going to be *meaningless*, but so is comparing Kilograms to Meters, and Rust
300 //! doesn't prevent that either. Similarly, if you get "lucky" and notice that a pointer
301 //! one-past-the-end is the "same" address as the start of an unrelated allocation, anything
302 //! you do with that fact is *probably* going to be gibberish. The scope of that gibberish
303 //! is kept under control by the fact that the two pointers *still* aren't allowed to access
304 //! the other's allocation (bytes), because they still have different provenance.
305 //!
306 //! * Perform pointer tagging tricks. This falls out of [`wrapping_offset`] but is worth
307 //! mentioning in more detail because of the limitations of [CHERI][]. Low-bit tagging
308 //! is very robust, and often doesn't even go out of bounds because types ensure
309 //! size >= align (and over-aligning actually gives CHERI more flexibility). Anything
310 //! more complex than this rapidly enters "extremely platform-specific" territory as
311 //! certain things may or may not be allowed based on specific supported operations.
312 //! For instance, ARM explicitly supports high-bit tagging, and so CHERI on ARM inherits
313 //! that and should support it.
314 //!
315 //! ## Pointer-usize-pointer roundtrips and 'exposed' provenance
316 //!
317 //! **This section is *non-normative* and is part of the [Strict Provenance] experiment.**
318 //!
319 //! As discussed above, pointer-usize-pointer roundtrips are not possible under [Strict Provenance].
320 //! However, there exists legacy Rust code that is full of such roundtrips, and legacy platform APIs
321 //! regularly assume that `usize` can capture all the information that makes up a pointer. There
322 //! also might be code that cannot be ported to Strict Provenance (which is something we would [like
323 //! to hear about][Strict Provenance]).
324 //!
325 //! For situations like this, there is a fallback plan, a way to 'opt out' of Strict Provenance.
326 //! However, note that this makes your code a lot harder to specify, and the code will not work
327 //! (well) with tools like [Miri] and [CHERI].
328 //!
329 //! This fallback plan is provided by the [`expose_addr`] and [`from_exposed_addr`] methods (which
330 //! are equivalent to `as` casts between pointers and integers). [`expose_addr`] is a lot like
331 //! [`addr`], but additionally adds the provenance of the pointer to a global list of 'exposed'
332 //! provenances. (This list is purely conceptual, it exists for the purpose of specifying Rust but
333 //! is not materialized in actual executions, except in tools like [Miri].) [`from_exposed_addr`]
334 //! can be used to construct a pointer with one of these previously 'exposed' provenances.
335 //! [`from_exposed_addr`] takes only `addr: usize` as arguments, so unlike in [`with_addr`] there is
336 //! no indication of what the correct provenance for the returned pointer is -- and that is exactly
337 //! what makes pointer-usize-pointer roundtrips so tricky to rigorously specify! There is no
338 //! algorithm that decides which provenance will be used. You can think of this as "guessing" the
339 //! right provenance, and the guess will be "maximally in your favor", in the sense that if there is
340 //! any way to avoid undefined behavior, then that is the guess that will be taken. However, if
341 //! there is *no* previously 'exposed' provenance that justifies the way the returned pointer will
342 //! be used, the program has undefined behavior.
343 //!
344 //! Using [`expose_addr`] or [`from_exposed_addr`] (or the equivalent `as` casts) means that code is
345 //! *not* following Strict Provenance rules. The goal of the Strict Provenance experiment is to
346 //! determine whether it is possible to use Rust without [`expose_addr`] and [`from_exposed_addr`].
347 //! If this is successful, it would be a major win for avoiding specification complexity and to
348 //! facilitate adoption of tools like [CHERI] and [Miri] that can be a big help in increasing the
349 //! confidence in (unsafe) Rust code.
350 //!
351 //! [aliasing]: ../../nomicon/aliasing.html
352 //! [book]: ../../book/ch19-01-unsafe-rust.html#dereferencing-a-raw-pointer
353 //! [ub]: ../../reference/behavior-considered-undefined.html
354 //! [zst]: ../../nomicon/exotic-sizes.html#zero-sized-types-zsts
355 //! [atomic operations]: crate::sync::atomic
356 //! [`offset`]: pointer::offset
357 //! [`wrapping_offset`]: pointer::wrapping_offset
358 //! [`with_addr`]: pointer::with_addr
359 //! [`map_addr`]: pointer::map_addr
360 //! [`addr`]: pointer::addr
361 //! [`ptr::invalid`]: core::ptr::invalid
362 //! [`expose_addr`]: pointer::expose_addr
363 //! [`from_exposed_addr`]: from_exposed_addr
364 //! [Miri]: https://github.com/rust-lang/miri
365 //! [CHERI]: https://www.cl.cam.ac.uk/research/security/ctsrd/cheri/
366 //! [Strict Provenance]: https://github.com/rust-lang/rust/issues/95228
367 //! [Stacked Borrows]: https://plv.mpi-sws.org/rustbelt/stacked-borrows/
368
369 #![stable(feature = "rust1", since = "1.0.0")]
370
371 use crate::cmp::Ordering;
372 use crate::fmt;
373 use crate::hash;
374 use crate::intrinsics::{
375 self, assert_unsafe_precondition, is_aligned_and_not_null, is_nonoverlapping,
376 };
377
378 use crate::mem::{self, MaybeUninit};
379
380 #[stable(feature = "rust1", since = "1.0.0")]
381 #[doc(inline)]
382 pub use crate::intrinsics::copy_nonoverlapping;
383
384 #[stable(feature = "rust1", since = "1.0.0")]
385 #[doc(inline)]
386 pub use crate::intrinsics::copy;
387
388 #[stable(feature = "rust1", since = "1.0.0")]
389 #[doc(inline)]
390 pub use crate::intrinsics::write_bytes;
391
392 mod metadata;
393 pub(crate) use metadata::PtrRepr;
394 #[unstable(feature = "ptr_metadata", issue = "81513")]
395 pub use metadata::{from_raw_parts, from_raw_parts_mut, metadata, DynMetadata, Pointee, Thin};
396
397 mod non_null;
398 #[stable(feature = "nonnull", since = "1.25.0")]
399 pub use non_null::NonNull;
400
401 mod unique;
402 #[unstable(feature = "ptr_internals", issue = "none")]
403 pub use unique::Unique;
404
405 mod const_ptr;
406 mod mut_ptr;
407
408 /// Executes the destructor (if any) of the pointed-to value.
409 ///
410 /// This is semantically equivalent to calling [`ptr::read`] and discarding
411 /// the result, but has the following advantages:
412 ///
413 /// * It is *required* to use `drop_in_place` to drop unsized types like
414 /// trait objects, because they can't be read out onto the stack and
415 /// dropped normally.
416 ///
417 /// * It is friendlier to the optimizer to do this over [`ptr::read`] when
418 /// dropping manually allocated memory (e.g., in the implementations of
419 /// `Box`/`Rc`/`Vec`), as the compiler doesn't need to prove that it's
420 /// sound to elide the copy.
421 ///
422 /// * It can be used to drop [pinned] data when `T` is not `repr(packed)`
423 /// (pinned data must not be moved before it is dropped).
424 ///
425 /// Unaligned values cannot be dropped in place, they must be copied to an aligned
426 /// location first using [`ptr::read_unaligned`]. For packed structs, this move is
427 /// done automatically by the compiler. This means the fields of packed structs
428 /// are not dropped in-place.
429 ///
430 /// [`ptr::read`]: self::read
431 /// [`ptr::read_unaligned`]: self::read_unaligned
432 /// [pinned]: crate::pin
433 ///
434 /// # Safety
435 ///
436 /// Behavior is undefined if any of the following conditions are violated:
437 ///
438 /// * `to_drop` must be [valid] for both reads and writes.
439 ///
440 /// * `to_drop` must be properly aligned.
441 ///
442 /// * The value `to_drop` points to must be valid for dropping, which may mean it must uphold
443 /// additional invariants - this is type-dependent.
444 ///
445 /// Additionally, if `T` is not [`Copy`], using the pointed-to value after
446 /// calling `drop_in_place` can cause undefined behavior. Note that `*to_drop =
447 /// foo` counts as a use because it will cause the value to be dropped
448 /// again. [`write()`] can be used to overwrite data without causing it to be
449 /// dropped.
450 ///
451 /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned.
452 ///
453 /// [valid]: self#safety
454 ///
455 /// # Examples
456 ///
457 /// Manually remove the last item from a vector:
458 ///
459 /// ```
460 /// use std::ptr;
461 /// use std::rc::Rc;
462 ///
463 /// let last = Rc::new(1);
464 /// let weak = Rc::downgrade(&last);
465 ///
466 /// let mut v = vec![Rc::new(0), last];
467 ///
468 /// unsafe {
469 /// // Get a raw pointer to the last element in `v`.
470 /// let ptr = &mut v[1] as *mut _;
471 /// // Shorten `v` to prevent the last item from being dropped. We do that first,
472 /// // to prevent issues if the `drop_in_place` below panics.
473 /// v.set_len(1);
474 /// // Without a call `drop_in_place`, the last item would never be dropped,
475 /// // and the memory it manages would be leaked.
476 /// ptr::drop_in_place(ptr);
477 /// }
478 ///
479 /// assert_eq!(v, &[0.into()]);
480 ///
481 /// // Ensure that the last item was dropped.
482 /// assert!(weak.upgrade().is_none());
483 /// ```
484 #[stable(feature = "drop_in_place", since = "1.8.0")]
485 #[lang = "drop_in_place"]
486 #[allow(unconditional_recursion)]
487 pub unsafe fn drop_in_place<T: ?Sized>(to_drop: *mut T) {
488 // Code here does not matter - this is replaced by the
489 // real drop glue by the compiler.
490
491 // SAFETY: see comment above
492 unsafe { drop_in_place(to_drop) }
493 }
494
495 /// Creates a null raw pointer.
496 ///
497 /// # Examples
498 ///
499 /// ```
500 /// use std::ptr;
501 ///
502 /// let p: *const i32 = ptr::null();
503 /// assert!(p.is_null());
504 /// ```
505 #[inline(always)]
506 #[must_use]
507 #[stable(feature = "rust1", since = "1.0.0")]
508 #[rustc_promotable]
509 #[rustc_const_stable(feature = "const_ptr_null", since = "1.24.0")]
510 #[rustc_allow_const_fn_unstable(ptr_metadata)]
511 #[rustc_diagnostic_item = "ptr_null"]
512 pub const fn null<T: ?Sized + Thin>() -> *const T {
513 from_raw_parts(invalid(0), ())
514 }
515
516 /// Creates an invalid pointer with the given address.
517 ///
518 /// This is different from `addr as *const T`, which creates a pointer that picks up a previously
519 /// exposed provenance. See [`from_exposed_addr`] for more details on that operation.
520 ///
521 /// The module's top-level documentation discusses the precise meaning of an "invalid"
522 /// pointer but essentially this expresses that the pointer is not associated
523 /// with any actual allocation and is little more than a usize address in disguise.
524 ///
525 /// This pointer will have no provenance associated with it and is therefore
526 /// UB to read/write/offset. This mostly exists to facilitate things
527 /// like `ptr::null` and `NonNull::dangling` which make invalid pointers.
528 ///
529 /// (Standard "Zero-Sized-Types get to cheat and lie" caveats apply, although it
530 /// may be desirable to give them their own API just to make that 100% clear.)
531 ///
532 /// This API and its claimed semantics are part of the Strict Provenance experiment,
533 /// see the [module documentation][crate::ptr] for details.
534 #[inline(always)]
535 #[must_use]
536 #[rustc_const_stable(feature = "stable_things_using_strict_provenance", since = "1.61.0")]
537 #[unstable(feature = "strict_provenance", issue = "95228")]
538 pub const fn invalid<T>(addr: usize) -> *const T {
539 // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic.
540 // We use transmute rather than a cast so tools like Miri can tell that this
541 // is *not* the same as from_exposed_addr.
542 // SAFETY: every valid integer is also a valid pointer (as long as you don't dereference that
543 // pointer).
544 unsafe { mem::transmute(addr) }
545 }
546
547 /// Creates an invalid mutable pointer with the given address.
548 ///
549 /// This is different from `addr as *mut T`, which creates a pointer that picks up a previously
550 /// exposed provenance. See [`from_exposed_addr_mut`] for more details on that operation.
551 ///
552 /// The module's top-level documentation discusses the precise meaning of an "invalid"
553 /// pointer but essentially this expresses that the pointer is not associated
554 /// with any actual allocation and is little more than a usize address in disguise.
555 ///
556 /// This pointer will have no provenance associated with it and is therefore
557 /// UB to read/write/offset. This mostly exists to facilitate things
558 /// like `ptr::null` and `NonNull::dangling` which make invalid pointers.
559 ///
560 /// (Standard "Zero-Sized-Types get to cheat and lie" caveats apply, although it
561 /// may be desirable to give them their own API just to make that 100% clear.)
562 ///
563 /// This API and its claimed semantics are part of the Strict Provenance experiment,
564 /// see the [module documentation][crate::ptr] for details.
565 #[inline(always)]
566 #[must_use]
567 #[rustc_const_stable(feature = "stable_things_using_strict_provenance", since = "1.61.0")]
568 #[unstable(feature = "strict_provenance", issue = "95228")]
569 pub const fn invalid_mut<T>(addr: usize) -> *mut T {
570 // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic.
571 // We use transmute rather than a cast so tools like Miri can tell that this
572 // is *not* the same as from_exposed_addr.
573 // SAFETY: every valid integer is also a valid pointer (as long as you don't dereference that
574 // pointer).
575 unsafe { mem::transmute(addr) }
576 }
577
578 /// Convert an address back to a pointer, picking up a previously 'exposed' provenance.
579 ///
580 /// This is equivalent to `addr as *const T`. The provenance of the returned pointer is that of *any*
581 /// pointer that was previously passed to [`expose_addr`][pointer::expose_addr] or a `ptr as usize`
582 /// cast. If there is no previously 'exposed' provenance that justifies the way this pointer will be
583 /// used, the program has undefined behavior. Note that there is no algorithm that decides which
584 /// provenance will be used. You can think of this as "guessing" the right provenance, and the guess
585 /// will be "maximally in your favor", in the sense that if there is any way to avoid undefined
586 /// behavior, then that is the guess that will be taken.
587 ///
588 /// On platforms with multiple address spaces, it is your responsibility to ensure that the
589 /// address makes sense in the address space that this pointer will be used with.
590 ///
591 /// Using this method means that code is *not* following strict provenance rules. "Guessing" a
592 /// suitable provenance complicates specification and reasoning and may not be supported by
593 /// tools that help you to stay conformant with the Rust memory model, so it is recommended to
594 /// use [`with_addr`][pointer::with_addr] wherever possible.
595 ///
596 /// On most platforms this will produce a value with the same bytes as the address. Platforms
597 /// which need to store additional information in a pointer may not support this operation,
598 /// since it is generally not possible to actually *compute* which provenance the returned
599 /// pointer has to pick up.
600 ///
601 /// This API and its claimed semantics are part of the Strict Provenance experiment, see the
602 /// [module documentation][crate::ptr] for details.
603 #[must_use]
604 #[inline]
605 #[unstable(feature = "strict_provenance", issue = "95228")]
606 pub fn from_exposed_addr<T>(addr: usize) -> *const T
607 where
608 T: Sized,
609 {
610 // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic.
611 addr as *const T
612 }
613
614 /// Convert an address back to a mutable pointer, picking up a previously 'exposed' provenance.
615 ///
616 /// This is equivalent to `addr as *mut T`. The provenance of the returned pointer is that of *any*
617 /// pointer that was previously passed to [`expose_addr`][pointer::expose_addr] or a `ptr as usize`
618 /// cast. If there is no previously 'exposed' provenance that justifies the way this pointer will be
619 /// used, the program has undefined behavior. Note that there is no algorithm that decides which
620 /// provenance will be used. You can think of this as "guessing" the right provenance, and the guess
621 /// will be "maximally in your favor", in the sense that if there is any way to avoid undefined
622 /// behavior, then that is the guess that will be taken.
623 ///
624 /// On platforms with multiple address spaces, it is your responsibility to ensure that the
625 /// address makes sense in the address space that this pointer will be used with.
626 ///
627 /// Using this method means that code is *not* following strict provenance rules. "Guessing" a
628 /// suitable provenance complicates specification and reasoning and may not be supported by
629 /// tools that help you to stay conformant with the Rust memory model, so it is recommended to
630 /// use [`with_addr`][pointer::with_addr] wherever possible.
631 ///
632 /// On most platforms this will produce a value with the same bytes as the address. Platforms
633 /// which need to store additional information in a pointer may not support this operation,
634 /// since it is generally not possible to actually *compute* which provenance the returned
635 /// pointer has to pick up.
636 ///
637 /// This API and its claimed semantics are part of the Strict Provenance experiment, see the
638 /// [module documentation][crate::ptr] for details.
639 #[must_use]
640 #[inline]
641 #[unstable(feature = "strict_provenance", issue = "95228")]
642 pub fn from_exposed_addr_mut<T>(addr: usize) -> *mut T
643 where
644 T: Sized,
645 {
646 // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic.
647 addr as *mut T
648 }
649
650 /// Creates a null mutable raw pointer.
651 ///
652 /// # Examples
653 ///
654 /// ```
655 /// use std::ptr;
656 ///
657 /// let p: *mut i32 = ptr::null_mut();
658 /// assert!(p.is_null());
659 /// ```
660 #[inline(always)]
661 #[must_use]
662 #[stable(feature = "rust1", since = "1.0.0")]
663 #[rustc_promotable]
664 #[rustc_const_stable(feature = "const_ptr_null", since = "1.24.0")]
665 #[rustc_allow_const_fn_unstable(ptr_metadata)]
666 #[rustc_diagnostic_item = "ptr_null_mut"]
667 pub const fn null_mut<T: ?Sized + Thin>() -> *mut T {
668 from_raw_parts_mut(invalid_mut(0), ())
669 }
670
671 /// Forms a raw slice from a pointer and a length.
672 ///
673 /// The `len` argument is the number of **elements**, not the number of bytes.
674 ///
675 /// This function is safe, but actually using the return value is unsafe.
676 /// See the documentation of [`slice::from_raw_parts`] for slice safety requirements.
677 ///
678 /// [`slice::from_raw_parts`]: crate::slice::from_raw_parts
679 ///
680 /// # Examples
681 ///
682 /// ```rust
683 /// use std::ptr;
684 ///
685 /// // create a slice pointer when starting out with a pointer to the first element
686 /// let x = [5, 6, 7];
687 /// let raw_pointer = x.as_ptr();
688 /// let slice = ptr::slice_from_raw_parts(raw_pointer, 3);
689 /// assert_eq!(unsafe { &*slice }[2], 7);
690 /// ```
691 #[inline]
692 #[stable(feature = "slice_from_raw_parts", since = "1.42.0")]
693 #[rustc_const_stable(feature = "const_slice_from_raw_parts", since = "1.64.0")]
694 #[rustc_allow_const_fn_unstable(ptr_metadata)]
695 pub const fn slice_from_raw_parts<T>(data: *const T, len: usize) -> *const [T] {
696 from_raw_parts(data.cast(), len)
697 }
698
699 /// Performs the same functionality as [`slice_from_raw_parts`], except that a
700 /// raw mutable slice is returned, as opposed to a raw immutable slice.
701 ///
702 /// See the documentation of [`slice_from_raw_parts`] for more details.
703 ///
704 /// This function is safe, but actually using the return value is unsafe.
705 /// See the documentation of [`slice::from_raw_parts_mut`] for slice safety requirements.
706 ///
707 /// [`slice::from_raw_parts_mut`]: crate::slice::from_raw_parts_mut
708 ///
709 /// # Examples
710 ///
711 /// ```rust
712 /// use std::ptr;
713 ///
714 /// let x = &mut [5, 6, 7];
715 /// let raw_pointer = x.as_mut_ptr();
716 /// let slice = ptr::slice_from_raw_parts_mut(raw_pointer, 3);
717 ///
718 /// unsafe {
719 /// (*slice)[2] = 99; // assign a value at an index in the slice
720 /// };
721 ///
722 /// assert_eq!(unsafe { &*slice }[2], 99);
723 /// ```
724 #[inline]
725 #[stable(feature = "slice_from_raw_parts", since = "1.42.0")]
726 #[rustc_const_unstable(feature = "const_slice_from_raw_parts_mut", issue = "67456")]
727 pub const fn slice_from_raw_parts_mut<T>(data: *mut T, len: usize) -> *mut [T] {
728 from_raw_parts_mut(data.cast(), len)
729 }
730
731 /// Swaps the values at two mutable locations of the same type, without
732 /// deinitializing either.
733 ///
734 /// But for the following exceptions, this function is semantically
735 /// equivalent to [`mem::swap`]:
736 ///
737 /// * It operates on raw pointers instead of references. When references are
738 /// available, [`mem::swap`] should be preferred.
739 ///
740 /// * The two pointed-to values may overlap. If the values do overlap, then the
741 /// overlapping region of memory from `x` will be used. This is demonstrated
742 /// in the second example below.
743 ///
744 /// * The operation is "untyped" in the sense that data may be uninitialized or otherwise violate
745 /// the requirements of `T`. The initialization state is preserved exactly.
746 ///
747 /// # Safety
748 ///
749 /// Behavior is undefined if any of the following conditions are violated:
750 ///
751 /// * Both `x` and `y` must be [valid] for both reads and writes.
752 ///
753 /// * Both `x` and `y` must be properly aligned.
754 ///
755 /// Note that even if `T` has size `0`, the pointers must be non-null and properly aligned.
756 ///
757 /// [valid]: self#safety
758 ///
759 /// # Examples
760 ///
761 /// Swapping two non-overlapping regions:
762 ///
763 /// ```
764 /// use std::ptr;
765 ///
766 /// let mut array = [0, 1, 2, 3];
767 ///
768 /// let (x, y) = array.split_at_mut(2);
769 /// let x = x.as_mut_ptr().cast::<[u32; 2]>(); // this is `array[0..2]`
770 /// let y = y.as_mut_ptr().cast::<[u32; 2]>(); // this is `array[2..4]`
771 ///
772 /// unsafe {
773 /// ptr::swap(x, y);
774 /// assert_eq!([2, 3, 0, 1], array);
775 /// }
776 /// ```
777 ///
778 /// Swapping two overlapping regions:
779 ///
780 /// ```
781 /// use std::ptr;
782 ///
783 /// let mut array: [i32; 4] = [0, 1, 2, 3];
784 ///
785 /// let array_ptr: *mut i32 = array.as_mut_ptr();
786 ///
787 /// let x = array_ptr as *mut [i32; 3]; // this is `array[0..3]`
788 /// let y = unsafe { array_ptr.add(1) } as *mut [i32; 3]; // this is `array[1..4]`
789 ///
790 /// unsafe {
791 /// ptr::swap(x, y);
792 /// // The indices `1..3` of the slice overlap between `x` and `y`.
793 /// // Reasonable results would be for to them be `[2, 3]`, so that indices `0..3` are
794 /// // `[1, 2, 3]` (matching `y` before the `swap`); or for them to be `[0, 1]`
795 /// // so that indices `1..4` are `[0, 1, 2]` (matching `x` before the `swap`).
796 /// // This implementation is defined to make the latter choice.
797 /// assert_eq!([1, 0, 1, 2], array);
798 /// }
799 /// ```
800 #[inline]
801 #[stable(feature = "rust1", since = "1.0.0")]
802 #[rustc_const_unstable(feature = "const_swap", issue = "83163")]
803 pub const unsafe fn swap<T>(x: *mut T, y: *mut T) {
804 // Give ourselves some scratch space to work with.
805 // We do not have to worry about drops: `MaybeUninit` does nothing when dropped.
806 let mut tmp = MaybeUninit::<T>::uninit();
807
808 // Perform the swap
809 // SAFETY: the caller must guarantee that `x` and `y` are
810 // valid for writes and properly aligned. `tmp` cannot be
811 // overlapping either `x` or `y` because `tmp` was just allocated
812 // on the stack as a separate allocated object.
813 unsafe {
814 copy_nonoverlapping(x, tmp.as_mut_ptr(), 1);
815 copy(y, x, 1); // `x` and `y` may overlap
816 copy_nonoverlapping(tmp.as_ptr(), y, 1);
817 }
818 }
819
820 /// Swaps `count * size_of::<T>()` bytes between the two regions of memory
821 /// beginning at `x` and `y`. The two regions must *not* overlap.
822 ///
823 /// The operation is "untyped" in the sense that data may be uninitialized or otherwise violate the
824 /// requirements of `T`. The initialization state is preserved exactly.
825 ///
826 /// # Safety
827 ///
828 /// Behavior is undefined if any of the following conditions are violated:
829 ///
830 /// * Both `x` and `y` must be [valid] for both reads and writes of `count *
831 /// size_of::<T>()` bytes.
832 ///
833 /// * Both `x` and `y` must be properly aligned.
834 ///
835 /// * The region of memory beginning at `x` with a size of `count *
836 /// size_of::<T>()` bytes must *not* overlap with the region of memory
837 /// beginning at `y` with the same size.
838 ///
839 /// Note that even if the effectively copied size (`count * size_of::<T>()`) is `0`,
840 /// the pointers must be non-null and properly aligned.
841 ///
842 /// [valid]: self#safety
843 ///
844 /// # Examples
845 ///
846 /// Basic usage:
847 ///
848 /// ```
849 /// use std::ptr;
850 ///
851 /// let mut x = [1, 2, 3, 4];
852 /// let mut y = [7, 8, 9];
853 ///
854 /// unsafe {
855 /// ptr::swap_nonoverlapping(x.as_mut_ptr(), y.as_mut_ptr(), 2);
856 /// }
857 ///
858 /// assert_eq!(x, [7, 8, 3, 4]);
859 /// assert_eq!(y, [1, 2, 9]);
860 /// ```
861 #[inline]
862 #[stable(feature = "swap_nonoverlapping", since = "1.27.0")]
863 #[rustc_const_unstable(feature = "const_swap", issue = "83163")]
864 pub const unsafe fn swap_nonoverlapping<T>(x: *mut T, y: *mut T, count: usize) {
865 #[allow(unused)]
866 macro_rules! attempt_swap_as_chunks {
867 ($ChunkTy:ty) => {
868 if mem::align_of::<T>() >= mem::align_of::<$ChunkTy>()
869 && mem::size_of::<T>() % mem::size_of::<$ChunkTy>() == 0
870 {
871 let x: *mut $ChunkTy = x.cast();
872 let y: *mut $ChunkTy = y.cast();
873 let count = count * (mem::size_of::<T>() / mem::size_of::<$ChunkTy>());
874 // SAFETY: these are the same bytes that the caller promised were
875 // ok, just typed as `MaybeUninit<ChunkTy>`s instead of as `T`s.
876 // The `if` condition above ensures that we're not violating
877 // alignment requirements, and that the division is exact so
878 // that we don't lose any bytes off the end.
879 return unsafe { swap_nonoverlapping_simple_untyped(x, y, count) };
880 }
881 };
882 }
883
884 // SAFETY: the caller must guarantee that `x` and `y` are
885 // valid for writes and properly aligned.
886 unsafe {
887 assert_unsafe_precondition!(
888 is_aligned_and_not_null(x)
889 && is_aligned_and_not_null(y)
890 && is_nonoverlapping(x, y, count)
891 );
892 }
893
894 // NOTE(scottmcm) Miri is disabled here as reading in smaller units is a
895 // pessimization for it. Also, if the type contains any unaligned pointers,
896 // copying those over multiple reads is difficult to support.
897 #[cfg(not(miri))]
898 {
899 // Split up the slice into small power-of-two-sized chunks that LLVM is able
900 // to vectorize (unless it's a special type with more-than-pointer alignment,
901 // because we don't want to pessimize things like slices of SIMD vectors.)
902 if mem::align_of::<T>() <= mem::size_of::<usize>()
903 && (!mem::size_of::<T>().is_power_of_two()
904 || mem::size_of::<T>() > mem::size_of::<usize>() * 2)
905 {
906 attempt_swap_as_chunks!(usize);
907 attempt_swap_as_chunks!(u8);
908 }
909 }
910
911 // SAFETY: Same preconditions as this function
912 unsafe { swap_nonoverlapping_simple_untyped(x, y, count) }
913 }
914
915 /// Same behaviour and safety conditions as [`swap_nonoverlapping`]
916 ///
917 /// LLVM can vectorize this (at least it can for the power-of-two-sized types
918 /// `swap_nonoverlapping` tries to use) so no need to manually SIMD it.
919 #[inline]
920 #[rustc_const_unstable(feature = "const_swap", issue = "83163")]
921 const unsafe fn swap_nonoverlapping_simple_untyped<T>(x: *mut T, y: *mut T, count: usize) {
922 let x = x.cast::<MaybeUninit<T>>();
923 let y = y.cast::<MaybeUninit<T>>();
924 let mut i = 0;
925 while i < count {
926 // SAFETY: By precondition, `i` is in-bounds because it's below `n`
927 let x = unsafe { &mut *x.add(i) };
928 // SAFETY: By precondition, `i` is in-bounds because it's below `n`
929 // and it's distinct from `x` since the ranges are non-overlapping
930 let y = unsafe { &mut *y.add(i) };
931 mem::swap_simple::<MaybeUninit<T>>(x, y);
932
933 i += 1;
934 }
935 }
936
937 /// Moves `src` into the pointed `dst`, returning the previous `dst` value.
938 ///
939 /// Neither value is dropped.
940 ///
941 /// This function is semantically equivalent to [`mem::replace`] except that it
942 /// operates on raw pointers instead of references. When references are
943 /// available, [`mem::replace`] should be preferred.
944 ///
945 /// # Safety
946 ///
947 /// Behavior is undefined if any of the following conditions are violated:
948 ///
949 /// * `dst` must be [valid] for both reads and writes.
950 ///
951 /// * `dst` must be properly aligned.
952 ///
953 /// * `dst` must point to a properly initialized value of type `T`.
954 ///
955 /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned.
956 ///
957 /// [valid]: self#safety
958 ///
959 /// # Examples
960 ///
961 /// ```
962 /// use std::ptr;
963 ///
964 /// let mut rust = vec!['b', 'u', 's', 't'];
965 ///
966 /// // `mem::replace` would have the same effect without requiring the unsafe
967 /// // block.
968 /// let b = unsafe {
969 /// ptr::replace(&mut rust[0], 'r')
970 /// };
971 ///
972 /// assert_eq!(b, 'b');
973 /// assert_eq!(rust, &['r', 'u', 's', 't']);
974 /// ```
975 #[inline]
976 #[stable(feature = "rust1", since = "1.0.0")]
977 #[rustc_const_unstable(feature = "const_replace", issue = "83164")]
978 pub const unsafe fn replace<T>(dst: *mut T, mut src: T) -> T {
979 // SAFETY: the caller must guarantee that `dst` is valid to be
980 // cast to a mutable reference (valid for writes, aligned, initialized),
981 // and cannot overlap `src` since `dst` must point to a distinct
982 // allocated object.
983 unsafe {
984 assert_unsafe_precondition!(is_aligned_and_not_null(dst));
985 mem::swap(&mut *dst, &mut src); // cannot overlap
986 }
987 src
988 }
989
990 /// Reads the value from `src` without moving it. This leaves the
991 /// memory in `src` unchanged.
992 ///
993 /// # Safety
994 ///
995 /// Behavior is undefined if any of the following conditions are violated:
996 ///
997 /// * `src` must be [valid] for reads.
998 ///
999 /// * `src` must be properly aligned. Use [`read_unaligned`] if this is not the
1000 /// case.
1001 ///
1002 /// * `src` must point to a properly initialized value of type `T`.
1003 ///
1004 /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned.
1005 ///
1006 /// # Examples
1007 ///
1008 /// Basic usage:
1009 ///
1010 /// ```
1011 /// let x = 12;
1012 /// let y = &x as *const i32;
1013 ///
1014 /// unsafe {
1015 /// assert_eq!(std::ptr::read(y), 12);
1016 /// }
1017 /// ```
1018 ///
1019 /// Manually implement [`mem::swap`]:
1020 ///
1021 /// ```
1022 /// use std::ptr;
1023 ///
1024 /// fn swap<T>(a: &mut T, b: &mut T) {
1025 /// unsafe {
1026 /// // Create a bitwise copy of the value at `a` in `tmp`.
1027 /// let tmp = ptr::read(a);
1028 ///
1029 /// // Exiting at this point (either by explicitly returning or by
1030 /// // calling a function which panics) would cause the value in `tmp` to
1031 /// // be dropped while the same value is still referenced by `a`. This
1032 /// // could trigger undefined behavior if `T` is not `Copy`.
1033 ///
1034 /// // Create a bitwise copy of the value at `b` in `a`.
1035 /// // This is safe because mutable references cannot alias.
1036 /// ptr::copy_nonoverlapping(b, a, 1);
1037 ///
1038 /// // As above, exiting here could trigger undefined behavior because
1039 /// // the same value is referenced by `a` and `b`.
1040 ///
1041 /// // Move `tmp` into `b`.
1042 /// ptr::write(b, tmp);
1043 ///
1044 /// // `tmp` has been moved (`write` takes ownership of its second argument),
1045 /// // so nothing is dropped implicitly here.
1046 /// }
1047 /// }
1048 ///
1049 /// let mut foo = "foo".to_owned();
1050 /// let mut bar = "bar".to_owned();
1051 ///
1052 /// swap(&mut foo, &mut bar);
1053 ///
1054 /// assert_eq!(foo, "bar");
1055 /// assert_eq!(bar, "foo");
1056 /// ```
1057 ///
1058 /// ## Ownership of the Returned Value
1059 ///
1060 /// `read` creates a bitwise copy of `T`, regardless of whether `T` is [`Copy`].
1061 /// If `T` is not [`Copy`], using both the returned value and the value at
1062 /// `*src` can violate memory safety. Note that assigning to `*src` counts as a
1063 /// use because it will attempt to drop the value at `*src`.
1064 ///
1065 /// [`write()`] can be used to overwrite data without causing it to be dropped.
1066 ///
1067 /// ```
1068 /// use std::ptr;
1069 ///
1070 /// let mut s = String::from("foo");
1071 /// unsafe {
1072 /// // `s2` now points to the same underlying memory as `s`.
1073 /// let mut s2: String = ptr::read(&s);
1074 ///
1075 /// assert_eq!(s2, "foo");
1076 ///
1077 /// // Assigning to `s2` causes its original value to be dropped. Beyond
1078 /// // this point, `s` must no longer be used, as the underlying memory has
1079 /// // been freed.
1080 /// s2 = String::default();
1081 /// assert_eq!(s2, "");
1082 ///
1083 /// // Assigning to `s` would cause the old value to be dropped again,
1084 /// // resulting in undefined behavior.
1085 /// // s = String::from("bar"); // ERROR
1086 ///
1087 /// // `ptr::write` can be used to overwrite a value without dropping it.
1088 /// ptr::write(&mut s, String::from("bar"));
1089 /// }
1090 ///
1091 /// assert_eq!(s, "bar");
1092 /// ```
1093 ///
1094 /// [valid]: self#safety
1095 #[inline]
1096 #[stable(feature = "rust1", since = "1.0.0")]
1097 #[rustc_const_unstable(feature = "const_ptr_read", issue = "80377")]
1098 #[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
1099 pub const unsafe fn read<T>(src: *const T) -> T {
1100 // We are calling the intrinsics directly to avoid function calls in the generated code
1101 // as `intrinsics::copy_nonoverlapping` is a wrapper function.
1102 extern "rust-intrinsic" {
1103 #[rustc_const_stable(feature = "const_intrinsic_copy", since = "1.63.0")]
1104 fn copy_nonoverlapping<T>(src: *const T, dst: *mut T, count: usize);
1105 }
1106
1107 let mut tmp = MaybeUninit::<T>::uninit();
1108 // SAFETY: the caller must guarantee that `src` is valid for reads.
1109 // `src` cannot overlap `tmp` because `tmp` was just allocated on
1110 // the stack as a separate allocated object.
1111 //
1112 // Also, since we just wrote a valid value into `tmp`, it is guaranteed
1113 // to be properly initialized.
1114 unsafe {
1115 copy_nonoverlapping(src, tmp.as_mut_ptr(), 1);
1116 tmp.assume_init()
1117 }
1118 }
1119
1120 /// Reads the value from `src` without moving it. This leaves the
1121 /// memory in `src` unchanged.
1122 ///
1123 /// Unlike [`read`], `read_unaligned` works with unaligned pointers.
1124 ///
1125 /// # Safety
1126 ///
1127 /// Behavior is undefined if any of the following conditions are violated:
1128 ///
1129 /// * `src` must be [valid] for reads.
1130 ///
1131 /// * `src` must point to a properly initialized value of type `T`.
1132 ///
1133 /// Like [`read`], `read_unaligned` creates a bitwise copy of `T`, regardless of
1134 /// whether `T` is [`Copy`]. If `T` is not [`Copy`], using both the returned
1135 /// value and the value at `*src` can [violate memory safety][read-ownership].
1136 ///
1137 /// Note that even if `T` has size `0`, the pointer must be non-null.
1138 ///
1139 /// [read-ownership]: read#ownership-of-the-returned-value
1140 /// [valid]: self#safety
1141 ///
1142 /// ## On `packed` structs
1143 ///
1144 /// Attempting to create a raw pointer to an `unaligned` struct field with
1145 /// an expression such as `&packed.unaligned as *const FieldType` creates an
1146 /// intermediate unaligned reference before converting that to a raw pointer.
1147 /// That this reference is temporary and immediately cast is inconsequential
1148 /// as the compiler always expects references to be properly aligned.
1149 /// As a result, using `&packed.unaligned as *const FieldType` causes immediate
1150 /// *undefined behavior* in your program.
1151 ///
1152 /// Instead you must use the [`ptr::addr_of!`](addr_of) macro to
1153 /// create the pointer. You may use that returned pointer together with this
1154 /// function.
1155 ///
1156 /// An example of what not to do and how this relates to `read_unaligned` is:
1157 ///
1158 /// ```
1159 /// #[repr(packed, C)]
1160 /// struct Packed {
1161 /// _padding: u8,
1162 /// unaligned: u32,
1163 /// }
1164 ///
1165 /// let packed = Packed {
1166 /// _padding: 0x00,
1167 /// unaligned: 0x01020304,
1168 /// };
1169 ///
1170 /// // Take the address of a 32-bit integer which is not aligned.
1171 /// // In contrast to `&packed.unaligned as *const _`, this has no undefined behavior.
1172 /// let unaligned = std::ptr::addr_of!(packed.unaligned);
1173 ///
1174 /// let v = unsafe { std::ptr::read_unaligned(unaligned) };
1175 /// assert_eq!(v, 0x01020304);
1176 /// ```
1177 ///
1178 /// Accessing unaligned fields directly with e.g. `packed.unaligned` is safe however.
1179 ///
1180 /// # Examples
1181 ///
1182 /// Read a usize value from a byte buffer:
1183 ///
1184 /// ```
1185 /// use std::mem;
1186 ///
1187 /// fn read_usize(x: &[u8]) -> usize {
1188 /// assert!(x.len() >= mem::size_of::<usize>());
1189 ///
1190 /// let ptr = x.as_ptr() as *const usize;
1191 ///
1192 /// unsafe { ptr.read_unaligned() }
1193 /// }
1194 /// ```
1195 #[inline]
1196 #[stable(feature = "ptr_unaligned", since = "1.17.0")]
1197 #[rustc_const_unstable(feature = "const_ptr_read", issue = "80377")]
1198 #[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
1199 pub const unsafe fn read_unaligned<T>(src: *const T) -> T {
1200 let mut tmp = MaybeUninit::<T>::uninit();
1201 // SAFETY: the caller must guarantee that `src` is valid for reads.
1202 // `src` cannot overlap `tmp` because `tmp` was just allocated on
1203 // the stack as a separate allocated object.
1204 //
1205 // Also, since we just wrote a valid value into `tmp`, it is guaranteed
1206 // to be properly initialized.
1207 unsafe {
1208 copy_nonoverlapping(src as *const u8, tmp.as_mut_ptr() as *mut u8, mem::size_of::<T>());
1209 tmp.assume_init()
1210 }
1211 }
1212
1213 /// Overwrites a memory location with the given value without reading or
1214 /// dropping the old value.
1215 ///
1216 /// `write` does not drop the contents of `dst`. This is safe, but it could leak
1217 /// allocations or resources, so care should be taken not to overwrite an object
1218 /// that should be dropped.
1219 ///
1220 /// Additionally, it does not drop `src`. Semantically, `src` is moved into the
1221 /// location pointed to by `dst`.
1222 ///
1223 /// This is appropriate for initializing uninitialized memory, or overwriting
1224 /// memory that has previously been [`read`] from.
1225 ///
1226 /// # Safety
1227 ///
1228 /// Behavior is undefined if any of the following conditions are violated:
1229 ///
1230 /// * `dst` must be [valid] for writes.
1231 ///
1232 /// * `dst` must be properly aligned. Use [`write_unaligned`] if this is not the
1233 /// case.
1234 ///
1235 /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned.
1236 ///
1237 /// [valid]: self#safety
1238 ///
1239 /// # Examples
1240 ///
1241 /// Basic usage:
1242 ///
1243 /// ```
1244 /// let mut x = 0;
1245 /// let y = &mut x as *mut i32;
1246 /// let z = 12;
1247 ///
1248 /// unsafe {
1249 /// std::ptr::write(y, z);
1250 /// assert_eq!(std::ptr::read(y), 12);
1251 /// }
1252 /// ```
1253 ///
1254 /// Manually implement [`mem::swap`]:
1255 ///
1256 /// ```
1257 /// use std::ptr;
1258 ///
1259 /// fn swap<T>(a: &mut T, b: &mut T) {
1260 /// unsafe {
1261 /// // Create a bitwise copy of the value at `a` in `tmp`.
1262 /// let tmp = ptr::read(a);
1263 ///
1264 /// // Exiting at this point (either by explicitly returning or by
1265 /// // calling a function which panics) would cause the value in `tmp` to
1266 /// // be dropped while the same value is still referenced by `a`. This
1267 /// // could trigger undefined behavior if `T` is not `Copy`.
1268 ///
1269 /// // Create a bitwise copy of the value at `b` in `a`.
1270 /// // This is safe because mutable references cannot alias.
1271 /// ptr::copy_nonoverlapping(b, a, 1);
1272 ///
1273 /// // As above, exiting here could trigger undefined behavior because
1274 /// // the same value is referenced by `a` and `b`.
1275 ///
1276 /// // Move `tmp` into `b`.
1277 /// ptr::write(b, tmp);
1278 ///
1279 /// // `tmp` has been moved (`write` takes ownership of its second argument),
1280 /// // so nothing is dropped implicitly here.
1281 /// }
1282 /// }
1283 ///
1284 /// let mut foo = "foo".to_owned();
1285 /// let mut bar = "bar".to_owned();
1286 ///
1287 /// swap(&mut foo, &mut bar);
1288 ///
1289 /// assert_eq!(foo, "bar");
1290 /// assert_eq!(bar, "foo");
1291 /// ```
1292 #[inline]
1293 #[stable(feature = "rust1", since = "1.0.0")]
1294 #[rustc_const_unstable(feature = "const_ptr_write", issue = "86302")]
1295 #[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
1296 pub const unsafe fn write<T>(dst: *mut T, src: T) {
1297 // We are calling the intrinsics directly to avoid function calls in the generated code
1298 // as `intrinsics::copy_nonoverlapping` is a wrapper function.
1299 extern "rust-intrinsic" {
1300 #[rustc_const_stable(feature = "const_intrinsic_copy", since = "1.63.0")]
1301 fn copy_nonoverlapping<T>(src: *const T, dst: *mut T, count: usize);
1302 }
1303
1304 // SAFETY: the caller must guarantee that `dst` is valid for writes.
1305 // `dst` cannot overlap `src` because the caller has mutable access
1306 // to `dst` while `src` is owned by this function.
1307 unsafe {
1308 copy_nonoverlapping(&src as *const T, dst, 1);
1309 intrinsics::forget(src);
1310 }
1311 }
1312
1313 /// Overwrites a memory location with the given value without reading or
1314 /// dropping the old value.
1315 ///
1316 /// Unlike [`write()`], the pointer may be unaligned.
1317 ///
1318 /// `write_unaligned` does not drop the contents of `dst`. This is safe, but it
1319 /// could leak allocations or resources, so care should be taken not to overwrite
1320 /// an object that should be dropped.
1321 ///
1322 /// Additionally, it does not drop `src`. Semantically, `src` is moved into the
1323 /// location pointed to by `dst`.
1324 ///
1325 /// This is appropriate for initializing uninitialized memory, or overwriting
1326 /// memory that has previously been read with [`read_unaligned`].
1327 ///
1328 /// # Safety
1329 ///
1330 /// Behavior is undefined if any of the following conditions are violated:
1331 ///
1332 /// * `dst` must be [valid] for writes.
1333 ///
1334 /// Note that even if `T` has size `0`, the pointer must be non-null.
1335 ///
1336 /// [valid]: self#safety
1337 ///
1338 /// ## On `packed` structs
1339 ///
1340 /// Attempting to create a raw pointer to an `unaligned` struct field with
1341 /// an expression such as `&packed.unaligned as *const FieldType` creates an
1342 /// intermediate unaligned reference before converting that to a raw pointer.
1343 /// That this reference is temporary and immediately cast is inconsequential
1344 /// as the compiler always expects references to be properly aligned.
1345 /// As a result, using `&packed.unaligned as *const FieldType` causes immediate
1346 /// *undefined behavior* in your program.
1347 ///
1348 /// Instead you must use the [`ptr::addr_of_mut!`](addr_of_mut)
1349 /// macro to create the pointer. You may use that returned pointer together with
1350 /// this function.
1351 ///
1352 /// An example of how to do it and how this relates to `write_unaligned` is:
1353 ///
1354 /// ```
1355 /// #[repr(packed, C)]
1356 /// struct Packed {
1357 /// _padding: u8,
1358 /// unaligned: u32,
1359 /// }
1360 ///
1361 /// let mut packed: Packed = unsafe { std::mem::zeroed() };
1362 ///
1363 /// // Take the address of a 32-bit integer which is not aligned.
1364 /// // In contrast to `&packed.unaligned as *mut _`, this has no undefined behavior.
1365 /// let unaligned = std::ptr::addr_of_mut!(packed.unaligned);
1366 ///
1367 /// unsafe { std::ptr::write_unaligned(unaligned, 42) };
1368 ///
1369 /// assert_eq!({packed.unaligned}, 42); // `{...}` forces copying the field instead of creating a reference.
1370 /// ```
1371 ///
1372 /// Accessing unaligned fields directly with e.g. `packed.unaligned` is safe however
1373 /// (as can be seen in the `assert_eq!` above).
1374 ///
1375 /// # Examples
1376 ///
1377 /// Write a usize value to a byte buffer:
1378 ///
1379 /// ```
1380 /// use std::mem;
1381 ///
1382 /// fn write_usize(x: &mut [u8], val: usize) {
1383 /// assert!(x.len() >= mem::size_of::<usize>());
1384 ///
1385 /// let ptr = x.as_mut_ptr() as *mut usize;
1386 ///
1387 /// unsafe { ptr.write_unaligned(val) }
1388 /// }
1389 /// ```
1390 #[inline]
1391 #[stable(feature = "ptr_unaligned", since = "1.17.0")]
1392 #[rustc_const_unstable(feature = "const_ptr_write", issue = "86302")]
1393 #[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
1394 pub const unsafe fn write_unaligned<T>(dst: *mut T, src: T) {
1395 // SAFETY: the caller must guarantee that `dst` is valid for writes.
1396 // `dst` cannot overlap `src` because the caller has mutable access
1397 // to `dst` while `src` is owned by this function.
1398 unsafe {
1399 copy_nonoverlapping(&src as *const T as *const u8, dst as *mut u8, mem::size_of::<T>());
1400 // We are calling the intrinsic directly to avoid function calls in the generated code.
1401 intrinsics::forget(src);
1402 }
1403 }
1404
1405 /// Performs a volatile read of the value from `src` without moving it. This
1406 /// leaves the memory in `src` unchanged.
1407 ///
1408 /// Volatile operations are intended to act on I/O memory, and are guaranteed
1409 /// to not be elided or reordered by the compiler across other volatile
1410 /// operations.
1411 ///
1412 /// # Notes
1413 ///
1414 /// Rust does not currently have a rigorously and formally defined memory model,
1415 /// so the precise semantics of what "volatile" means here is subject to change
1416 /// over time. That being said, the semantics will almost always end up pretty
1417 /// similar to [C11's definition of volatile][c11].
1418 ///
1419 /// The compiler shouldn't change the relative order or number of volatile
1420 /// memory operations. However, volatile memory operations on zero-sized types
1421 /// (e.g., if a zero-sized type is passed to `read_volatile`) are noops
1422 /// and may be ignored.
1423 ///
1424 /// [c11]: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf
1425 ///
1426 /// # Safety
1427 ///
1428 /// Behavior is undefined if any of the following conditions are violated:
1429 ///
1430 /// * `src` must be [valid] for reads.
1431 ///
1432 /// * `src` must be properly aligned.
1433 ///
1434 /// * `src` must point to a properly initialized value of type `T`.
1435 ///
1436 /// Like [`read`], `read_volatile` creates a bitwise copy of `T`, regardless of
1437 /// whether `T` is [`Copy`]. If `T` is not [`Copy`], using both the returned
1438 /// value and the value at `*src` can [violate memory safety][read-ownership].
1439 /// However, storing non-[`Copy`] types in volatile memory is almost certainly
1440 /// incorrect.
1441 ///
1442 /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned.
1443 ///
1444 /// [valid]: self#safety
1445 /// [read-ownership]: read#ownership-of-the-returned-value
1446 ///
1447 /// Just like in C, whether an operation is volatile has no bearing whatsoever
1448 /// on questions involving concurrent access from multiple threads. Volatile
1449 /// accesses behave exactly like non-atomic accesses in that regard. In particular,
1450 /// a race between a `read_volatile` and any write operation to the same location
1451 /// is undefined behavior.
1452 ///
1453 /// # Examples
1454 ///
1455 /// Basic usage:
1456 ///
1457 /// ```
1458 /// let x = 12;
1459 /// let y = &x as *const i32;
1460 ///
1461 /// unsafe {
1462 /// assert_eq!(std::ptr::read_volatile(y), 12);
1463 /// }
1464 /// ```
1465 #[inline]
1466 #[stable(feature = "volatile", since = "1.9.0")]
1467 #[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
1468 pub unsafe fn read_volatile<T>(src: *const T) -> T {
1469 // SAFETY: the caller must uphold the safety contract for `volatile_load`.
1470 unsafe {
1471 assert_unsafe_precondition!(is_aligned_and_not_null(src));
1472 intrinsics::volatile_load(src)
1473 }
1474 }
1475
1476 /// Performs a volatile write of a memory location with the given value without
1477 /// reading or dropping the old value.
1478 ///
1479 /// Volatile operations are intended to act on I/O memory, and are guaranteed
1480 /// to not be elided or reordered by the compiler across other volatile
1481 /// operations.
1482 ///
1483 /// `write_volatile` does not drop the contents of `dst`. This is safe, but it
1484 /// could leak allocations or resources, so care should be taken not to overwrite
1485 /// an object that should be dropped.
1486 ///
1487 /// Additionally, it does not drop `src`. Semantically, `src` is moved into the
1488 /// location pointed to by `dst`.
1489 ///
1490 /// # Notes
1491 ///
1492 /// Rust does not currently have a rigorously and formally defined memory model,
1493 /// so the precise semantics of what "volatile" means here is subject to change
1494 /// over time. That being said, the semantics will almost always end up pretty
1495 /// similar to [C11's definition of volatile][c11].
1496 ///
1497 /// The compiler shouldn't change the relative order or number of volatile
1498 /// memory operations. However, volatile memory operations on zero-sized types
1499 /// (e.g., if a zero-sized type is passed to `write_volatile`) are noops
1500 /// and may be ignored.
1501 ///
1502 /// [c11]: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf
1503 ///
1504 /// # Safety
1505 ///
1506 /// Behavior is undefined if any of the following conditions are violated:
1507 ///
1508 /// * `dst` must be [valid] for writes.
1509 ///
1510 /// * `dst` must be properly aligned.
1511 ///
1512 /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned.
1513 ///
1514 /// [valid]: self#safety
1515 ///
1516 /// Just like in C, whether an operation is volatile has no bearing whatsoever
1517 /// on questions involving concurrent access from multiple threads. Volatile
1518 /// accesses behave exactly like non-atomic accesses in that regard. In particular,
1519 /// a race between a `write_volatile` and any other operation (reading or writing)
1520 /// on the same location is undefined behavior.
1521 ///
1522 /// # Examples
1523 ///
1524 /// Basic usage:
1525 ///
1526 /// ```
1527 /// let mut x = 0;
1528 /// let y = &mut x as *mut i32;
1529 /// let z = 12;
1530 ///
1531 /// unsafe {
1532 /// std::ptr::write_volatile(y, z);
1533 /// assert_eq!(std::ptr::read_volatile(y), 12);
1534 /// }
1535 /// ```
1536 #[inline]
1537 #[stable(feature = "volatile", since = "1.9.0")]
1538 #[cfg_attr(miri, track_caller)] // even without panics, this helps for Miri backtraces
1539 pub unsafe fn write_volatile<T>(dst: *mut T, src: T) {
1540 // SAFETY: the caller must uphold the safety contract for `volatile_store`.
1541 unsafe {
1542 assert_unsafe_precondition!(is_aligned_and_not_null(dst));
1543 intrinsics::volatile_store(dst, src);
1544 }
1545 }
1546
1547 /// Align pointer `p`.
1548 ///
1549 /// Calculate offset (in terms of elements of `stride` stride) that has to be applied
1550 /// to pointer `p` so that pointer `p` would get aligned to `a`.
1551 ///
1552 /// Note: This implementation has been carefully tailored to not panic. It is UB for this to panic.
1553 /// The only real change that can be made here is change of `INV_TABLE_MOD_16` and associated
1554 /// constants.
1555 ///
1556 /// If we ever decide to make it possible to call the intrinsic with `a` that is not a
1557 /// power-of-two, it will probably be more prudent to just change to a naive implementation rather
1558 /// than trying to adapt this to accommodate that change.
1559 ///
1560 /// Any questions go to @nagisa.
1561 #[lang = "align_offset"]
1562 pub(crate) unsafe fn align_offset<T: Sized>(p: *const T, a: usize) -> usize {
1563 // FIXME(#75598): Direct use of these intrinsics improves codegen significantly at opt-level <=
1564 // 1, where the method versions of these operations are not inlined.
1565 use intrinsics::{
1566 cttz_nonzero, exact_div, unchecked_rem, unchecked_shl, unchecked_shr, unchecked_sub,
1567 wrapping_add, wrapping_mul, wrapping_sub,
1568 };
1569
1570 /// Calculate multiplicative modular inverse of `x` modulo `m`.
1571 ///
1572 /// This implementation is tailored for `align_offset` and has following preconditions:
1573 ///
1574 /// * `m` is a power-of-two;
1575 /// * `x < m`; (if `x ≥ m`, pass in `x % m` instead)
1576 ///
1577 /// Implementation of this function shall not panic. Ever.
1578 #[inline]
1579 unsafe fn mod_inv(x: usize, m: usize) -> usize {
1580 /// Multiplicative modular inverse table modulo 2⁴ = 16.
1581 ///
1582 /// Note, that this table does not contain values where inverse does not exist (i.e., for
1583 /// `0⁻¹ mod 16`, `2⁻¹ mod 16`, etc.)
1584 const INV_TABLE_MOD_16: [u8; 8] = [1, 11, 13, 7, 9, 3, 5, 15];
1585 /// Modulo for which the `INV_TABLE_MOD_16` is intended.
1586 const INV_TABLE_MOD: usize = 16;
1587 /// INV_TABLE_MOD²
1588 const INV_TABLE_MOD_SQUARED: usize = INV_TABLE_MOD * INV_TABLE_MOD;
1589
1590 let table_inverse = INV_TABLE_MOD_16[(x & (INV_TABLE_MOD - 1)) >> 1] as usize;
1591 // SAFETY: `m` is required to be a power-of-two, hence non-zero.
1592 let m_minus_one = unsafe { unchecked_sub(m, 1) };
1593 if m <= INV_TABLE_MOD {
1594 table_inverse & m_minus_one
1595 } else {
1596 // We iterate "up" using the following formula:
1597 //
1598 // $$ xy ≡ 1 (mod 2ⁿ) → xy (2 - xy) ≡ 1 (mod 2²ⁿ) $$
1599 //
1600 // until 2²ⁿ ≥ m. Then we can reduce to our desired `m` by taking the result `mod m`.
1601 let mut inverse = table_inverse;
1602 let mut going_mod = INV_TABLE_MOD_SQUARED;
1603 loop {
1604 // y = y * (2 - xy) mod n
1605 //
1606 // Note, that we use wrapping operations here intentionally – the original formula
1607 // uses e.g., subtraction `mod n`. It is entirely fine to do them `mod
1608 // usize::MAX` instead, because we take the result `mod n` at the end
1609 // anyway.
1610 inverse = wrapping_mul(inverse, wrapping_sub(2usize, wrapping_mul(x, inverse)));
1611 if going_mod >= m {
1612 return inverse & m_minus_one;
1613 }
1614 going_mod = wrapping_mul(going_mod, going_mod);
1615 }
1616 }
1617 }
1618
1619 let addr = p.addr();
1620 let stride = mem::size_of::<T>();
1621 // SAFETY: `a` is a power-of-two, therefore non-zero.
1622 let a_minus_one = unsafe { unchecked_sub(a, 1) };
1623
1624 if stride == 0 {
1625 // SPECIAL_CASE: handle 0-sized types. No matter how many times we step, the address will
1626 // stay the same, so no offset will be able to align the pointer unless it is already
1627 // aligned. This branch _will_ be optimized out as `stride` is known at compile-time.
1628 let p_mod_a = addr & a_minus_one;
1629 return if p_mod_a == 0 { 0 } else { usize::MAX };
1630 }
1631
1632 // SAFETY: `stride == 0` case has been handled by the special case above.
1633 let a_mod_stride = unsafe { unchecked_rem(a, stride) };
1634 if a_mod_stride == 0 {
1635 // SPECIAL_CASE: In cases where the `a` is divisible by `stride`, byte offset to align a
1636 // pointer can be computed more simply through `-p (mod a)`. In the off-chance the byte
1637 // offset is not a multiple of `stride`, the input pointer was misaligned and no pointer
1638 // offset will be able to produce a `p` aligned to the specified `a`.
1639 //
1640 // The naive `-p (mod a)` equation inhibits LLVM's ability to select instructions
1641 // like `lea`. We compute `(round_up_to_next_alignment(p, a) - p)` instead. This
1642 // redistributes operations around the load-bearing, but pessimizing `and` instruction
1643 // sufficiently for LLVM to be able to utilize the various optimizations it knows about.
1644 //
1645 // LLVM handles the branch here particularly nicely. If this branch needs to be evaluated
1646 // at runtime, it will produce a mask `if addr_mod_stride == 0 { 0 } else { usize::MAX }`
1647 // in a branch-free way and then bitwise-OR it with whatever result the `-p mod a`
1648 // computation produces.
1649
1650 // SAFETY: `stride == 0` case has been handled by the special case above.
1651 let addr_mod_stride = unsafe { unchecked_rem(addr, stride) };
1652
1653 return if addr_mod_stride == 0 {
1654 let aligned_address = wrapping_add(addr, a_minus_one) & wrapping_sub(0, a);
1655 let byte_offset = wrapping_sub(aligned_address, addr);
1656 // SAFETY: `stride` is non-zero. This is guaranteed to divide exactly as well, because
1657 // addr has been verified to be aligned to the original type’s alignment requirements.
1658 unsafe { exact_div(byte_offset, stride) }
1659 } else {
1660 usize::MAX
1661 };
1662 }
1663
1664 // GENERAL_CASE: From here on we’re handling the very general case where `addr` may be
1665 // misaligned, there isn’t an obvious relationship between `stride` and `a` that we can take an
1666 // advantage of, etc. This case produces machine code that isn’t particularly high quality,
1667 // compared to the special cases above. The code produced here is still within the realm of
1668 // miracles, given the situations this case has to deal with.
1669
1670 // SAFETY: a is power-of-two hence non-zero. stride == 0 case is handled above.
1671 let gcdpow = unsafe { cttz_nonzero(stride).min(cttz_nonzero(a)) };
1672 // SAFETY: gcdpow has an upper-bound that’s at most the number of bits in a usize.
1673 let gcd = unsafe { unchecked_shl(1usize, gcdpow) };
1674 // SAFETY: gcd is always greater or equal to 1.
1675 if addr & unsafe { unchecked_sub(gcd, 1) } == 0 {
1676 // This branch solves for the following linear congruence equation:
1677 //
1678 // ` p + so = 0 mod a `
1679 //
1680 // `p` here is the pointer value, `s` - stride of `T`, `o` offset in `T`s, and `a` - the
1681 // requested alignment.
1682 //
1683 // With `g = gcd(a, s)`, and the above condition asserting that `p` is also divisible by
1684 // `g`, we can denote `a' = a/g`, `s' = s/g`, `p' = p/g`, then this becomes equivalent to:
1685 //
1686 // ` p' + s'o = 0 mod a' `
1687 // ` o = (a' - (p' mod a')) * (s'^-1 mod a') `
1688 //
1689 // The first term is "the relative alignment of `p` to `a`" (divided by the `g`), the
1690 // second term is "how does incrementing `p` by `s` bytes change the relative alignment of
1691 // `p`" (again divided by `g`). Division by `g` is necessary to make the inverse well
1692 // formed if `a` and `s` are not co-prime.
1693 //
1694 // Furthermore, the result produced by this solution is not "minimal", so it is necessary
1695 // to take the result `o mod lcm(s, a)`. This `lcm(s, a)` is the same as `a'`.
1696
1697 // SAFETY: `gcdpow` has an upper-bound not greater than the number of trailing 0-bits in
1698 // `a`.
1699 let a2 = unsafe { unchecked_shr(a, gcdpow) };
1700 // SAFETY: `a2` is non-zero. Shifting `a` by `gcdpow` cannot shift out any of the set bits
1701 // in `a` (of which it has exactly one).
1702 let a2minus1 = unsafe { unchecked_sub(a2, 1) };
1703 // SAFETY: `gcdpow` has an upper-bound not greater than the number of trailing 0-bits in
1704 // `a`.
1705 let s2 = unsafe { unchecked_shr(stride & a_minus_one, gcdpow) };
1706 // SAFETY: `gcdpow` has an upper-bound not greater than the number of trailing 0-bits in
1707 // `a`. Furthermore, the subtraction cannot overflow, because `a2 = a >> gcdpow` will
1708 // always be strictly greater than `(p % a) >> gcdpow`.
1709 let minusp2 = unsafe { unchecked_sub(a2, unchecked_shr(addr & a_minus_one, gcdpow)) };
1710 // SAFETY: `a2` is a power-of-two, as proven above. `s2` is strictly less than `a2`
1711 // because `(s % a) >> gcdpow` is strictly less than `a >> gcdpow`.
1712 return wrapping_mul(minusp2, unsafe { mod_inv(s2, a2) }) & a2minus1;
1713 }
1714
1715 // Cannot be aligned at all.
1716 usize::MAX
1717 }
1718
1719 /// Compares raw pointers for equality.
1720 ///
1721 /// This is the same as using the `==` operator, but less generic:
1722 /// the arguments have to be `*const T` raw pointers,
1723 /// not anything that implements `PartialEq`.
1724 ///
1725 /// This can be used to compare `&T` references (which coerce to `*const T` implicitly)
1726 /// by their address rather than comparing the values they point to
1727 /// (which is what the `PartialEq for &T` implementation does).
1728 ///
1729 /// # Examples
1730 ///
1731 /// ```
1732 /// use std::ptr;
1733 ///
1734 /// let five = 5;
1735 /// let other_five = 5;
1736 /// let five_ref = &five;
1737 /// let same_five_ref = &five;
1738 /// let other_five_ref = &other_five;
1739 ///
1740 /// assert!(five_ref == same_five_ref);
1741 /// assert!(ptr::eq(five_ref, same_five_ref));
1742 ///
1743 /// assert!(five_ref == other_five_ref);
1744 /// assert!(!ptr::eq(five_ref, other_five_ref));
1745 /// ```
1746 ///
1747 /// Slices are also compared by their length (fat pointers):
1748 ///
1749 /// ```
1750 /// let a = [1, 2, 3];
1751 /// assert!(std::ptr::eq(&a[..3], &a[..3]));
1752 /// assert!(!std::ptr::eq(&a[..2], &a[..3]));
1753 /// assert!(!std::ptr::eq(&a[0..2], &a[1..3]));
1754 /// ```
1755 ///
1756 /// Traits are also compared by their implementation:
1757 ///
1758 /// ```
1759 /// #[repr(transparent)]
1760 /// struct Wrapper { member: i32 }
1761 ///
1762 /// trait Trait {}
1763 /// impl Trait for Wrapper {}
1764 /// impl Trait for i32 {}
1765 ///
1766 /// let wrapper = Wrapper { member: 10 };
1767 ///
1768 /// // Pointers have equal addresses.
1769 /// assert!(std::ptr::eq(
1770 /// &wrapper as *const Wrapper as *const u8,
1771 /// &wrapper.member as *const i32 as *const u8
1772 /// ));
1773 ///
1774 /// // Objects have equal addresses, but `Trait` has different implementations.
1775 /// assert!(!std::ptr::eq(
1776 /// &wrapper as &dyn Trait,
1777 /// &wrapper.member as &dyn Trait,
1778 /// ));
1779 /// assert!(!std::ptr::eq(
1780 /// &wrapper as &dyn Trait as *const dyn Trait,
1781 /// &wrapper.member as &dyn Trait as *const dyn Trait,
1782 /// ));
1783 ///
1784 /// // Converting the reference to a `*const u8` compares by address.
1785 /// assert!(std::ptr::eq(
1786 /// &wrapper as &dyn Trait as *const dyn Trait as *const u8,
1787 /// &wrapper.member as &dyn Trait as *const dyn Trait as *const u8,
1788 /// ));
1789 /// ```
1790 #[stable(feature = "ptr_eq", since = "1.17.0")]
1791 #[inline]
1792 pub fn eq<T: ?Sized>(a: *const T, b: *const T) -> bool {
1793 a == b
1794 }
1795
1796 /// Hash a raw pointer.
1797 ///
1798 /// This can be used to hash a `&T` reference (which coerces to `*const T` implicitly)
1799 /// by its address rather than the value it points to
1800 /// (which is what the `Hash for &T` implementation does).
1801 ///
1802 /// # Examples
1803 ///
1804 /// ```
1805 /// use std::collections::hash_map::DefaultHasher;
1806 /// use std::hash::{Hash, Hasher};
1807 /// use std::ptr;
1808 ///
1809 /// let five = 5;
1810 /// let five_ref = &five;
1811 ///
1812 /// let mut hasher = DefaultHasher::new();
1813 /// ptr::hash(five_ref, &mut hasher);
1814 /// let actual = hasher.finish();
1815 ///
1816 /// let mut hasher = DefaultHasher::new();
1817 /// (five_ref as *const i32).hash(&mut hasher);
1818 /// let expected = hasher.finish();
1819 ///
1820 /// assert_eq!(actual, expected);
1821 /// ```
1822 #[stable(feature = "ptr_hash", since = "1.35.0")]
1823 pub fn hash<T: ?Sized, S: hash::Hasher>(hashee: *const T, into: &mut S) {
1824 use crate::hash::Hash;
1825 hashee.hash(into);
1826 }
1827
1828 // If this is a unary fn pointer, it adds a doc comment.
1829 // Otherwise, it hides the docs entirely.
1830 macro_rules! maybe_fnptr_doc {
1831 (@ #[$meta:meta] $item:item) => {
1832 #[doc(hidden)]
1833 #[$meta]
1834 $item
1835 };
1836 ($a:ident @ #[$meta:meta] $item:item) => {
1837 #[cfg_attr(not(bootstrap), doc(fake_variadic))]
1838 #[doc = "This trait is implemented for function pointers with up to twelve arguments."]
1839 #[$meta]
1840 $item
1841 };
1842 ($a:ident $($rest_a:ident)+ @ #[$meta:meta] $item:item) => {
1843 #[doc(hidden)]
1844 #[$meta]
1845 $item
1846 };
1847 }
1848
1849 // FIXME(strict_provenance_magic): function pointers have buggy codegen that
1850 // necessitates casting to a usize to get the backend to do the right thing.
1851 // for now I will break AVR to silence *a billion* lints. We should probably
1852 // have a proper "opaque function pointer type" to handle this kind of thing.
1853
1854 // Impls for function pointers
1855 macro_rules! fnptr_impls_safety_abi {
1856 ($FnTy: ty, $($Arg: ident),*) => {
1857 maybe_fnptr_doc! {
1858 $($Arg)* @
1859 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1860 impl<Ret, $($Arg),*> PartialEq for $FnTy {
1861 #[inline]
1862 fn eq(&self, other: &Self) -> bool {
1863 *self as usize == *other as usize
1864 }
1865 }
1866 }
1867
1868 maybe_fnptr_doc! {
1869 $($Arg)* @
1870 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1871 impl<Ret, $($Arg),*> Eq for $FnTy {}
1872 }
1873
1874 maybe_fnptr_doc! {
1875 $($Arg)* @
1876 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1877 impl<Ret, $($Arg),*> PartialOrd for $FnTy {
1878 #[inline]
1879 fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
1880 (*self as usize).partial_cmp(&(*other as usize))
1881 }
1882 }
1883 }
1884
1885 maybe_fnptr_doc! {
1886 $($Arg)* @
1887 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1888 impl<Ret, $($Arg),*> Ord for $FnTy {
1889 #[inline]
1890 fn cmp(&self, other: &Self) -> Ordering {
1891 (*self as usize).cmp(&(*other as usize))
1892 }
1893 }
1894 }
1895
1896 maybe_fnptr_doc! {
1897 $($Arg)* @
1898 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1899 impl<Ret, $($Arg),*> hash::Hash for $FnTy {
1900 fn hash<HH: hash::Hasher>(&self, state: &mut HH) {
1901 state.write_usize(*self as usize)
1902 }
1903 }
1904 }
1905
1906 maybe_fnptr_doc! {
1907 $($Arg)* @
1908 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1909 impl<Ret, $($Arg),*> fmt::Pointer for $FnTy {
1910 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1911 fmt::pointer_fmt_inner(*self as usize, f)
1912 }
1913 }
1914 }
1915
1916 maybe_fnptr_doc! {
1917 $($Arg)* @
1918 #[stable(feature = "fnptr_impls", since = "1.4.0")]
1919 impl<Ret, $($Arg),*> fmt::Debug for $FnTy {
1920 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1921 fmt::pointer_fmt_inner(*self as usize, f)
1922 }
1923 }
1924 }
1925 }
1926 }
1927
1928 macro_rules! fnptr_impls_args {
1929 ($($Arg: ident),+) => {
1930 fnptr_impls_safety_abi! { extern "Rust" fn($($Arg),+) -> Ret, $($Arg),+ }
1931 fnptr_impls_safety_abi! { extern "C" fn($($Arg),+) -> Ret, $($Arg),+ }
1932 fnptr_impls_safety_abi! { extern "C" fn($($Arg),+ , ...) -> Ret, $($Arg),+ }
1933 fnptr_impls_safety_abi! { unsafe extern "Rust" fn($($Arg),+) -> Ret, $($Arg),+ }
1934 fnptr_impls_safety_abi! { unsafe extern "C" fn($($Arg),+) -> Ret, $($Arg),+ }
1935 fnptr_impls_safety_abi! { unsafe extern "C" fn($($Arg),+ , ...) -> Ret, $($Arg),+ }
1936 };
1937 () => {
1938 // No variadic functions with 0 parameters
1939 fnptr_impls_safety_abi! { extern "Rust" fn() -> Ret, }
1940 fnptr_impls_safety_abi! { extern "C" fn() -> Ret, }
1941 fnptr_impls_safety_abi! { unsafe extern "Rust" fn() -> Ret, }
1942 fnptr_impls_safety_abi! { unsafe extern "C" fn() -> Ret, }
1943 };
1944 }
1945
1946 fnptr_impls_args! {}
1947 fnptr_impls_args! { T }
1948 fnptr_impls_args! { A, B }
1949 fnptr_impls_args! { A, B, C }
1950 fnptr_impls_args! { A, B, C, D }
1951 fnptr_impls_args! { A, B, C, D, E }
1952 fnptr_impls_args! { A, B, C, D, E, F }
1953 fnptr_impls_args! { A, B, C, D, E, F, G }
1954 fnptr_impls_args! { A, B, C, D, E, F, G, H }
1955 fnptr_impls_args! { A, B, C, D, E, F, G, H, I }
1956 fnptr_impls_args! { A, B, C, D, E, F, G, H, I, J }
1957 fnptr_impls_args! { A, B, C, D, E, F, G, H, I, J, K }
1958 fnptr_impls_args! { A, B, C, D, E, F, G, H, I, J, K, L }
1959
1960 /// Create a `const` raw pointer to a place, without creating an intermediate reference.
1961 ///
1962 /// Creating a reference with `&`/`&mut` is only allowed if the pointer is properly aligned
1963 /// and points to initialized data. For cases where those requirements do not hold,
1964 /// raw pointers should be used instead. However, `&expr as *const _` creates a reference
1965 /// before casting it to a raw pointer, and that reference is subject to the same rules
1966 /// as all other references. This macro can create a raw pointer *without* creating
1967 /// a reference first.
1968 ///
1969 /// Note, however, that the `expr` in `addr_of!(expr)` is still subject to all
1970 /// the usual rules. In particular, `addr_of!(*ptr::null())` is Undefined
1971 /// Behavior because it dereferences a null pointer.
1972 ///
1973 /// # Example
1974 ///
1975 /// ```
1976 /// use std::ptr;
1977 ///
1978 /// #[repr(packed)]
1979 /// struct Packed {
1980 /// f1: u8,
1981 /// f2: u16,
1982 /// }
1983 ///
1984 /// let packed = Packed { f1: 1, f2: 2 };
1985 /// // `&packed.f2` would create an unaligned reference, and thus be Undefined Behavior!
1986 /// let raw_f2 = ptr::addr_of!(packed.f2);
1987 /// assert_eq!(unsafe { raw_f2.read_unaligned() }, 2);
1988 /// ```
1989 ///
1990 /// See [`addr_of_mut`] for how to create a pointer to unininitialized data.
1991 /// Doing that with `addr_of` would not make much sense since one could only
1992 /// read the data, and that would be Undefined Behavior.
1993 #[stable(feature = "raw_ref_macros", since = "1.51.0")]
1994 #[rustc_macro_transparency = "semitransparent"]
1995 #[allow_internal_unstable(raw_ref_op)]
1996 pub macro addr_of($place:expr) {
1997 &raw const $place
1998 }
1999
2000 /// Create a `mut` raw pointer to a place, without creating an intermediate reference.
2001 ///
2002 /// Creating a reference with `&`/`&mut` is only allowed if the pointer is properly aligned
2003 /// and points to initialized data. For cases where those requirements do not hold,
2004 /// raw pointers should be used instead. However, `&mut expr as *mut _` creates a reference
2005 /// before casting it to a raw pointer, and that reference is subject to the same rules
2006 /// as all other references. This macro can create a raw pointer *without* creating
2007 /// a reference first.
2008 ///
2009 /// Note, however, that the `expr` in `addr_of_mut!(expr)` is still subject to all
2010 /// the usual rules. In particular, `addr_of_mut!(*ptr::null_mut())` is Undefined
2011 /// Behavior because it dereferences a null pointer.
2012 ///
2013 /// # Examples
2014 ///
2015 /// **Creating a pointer to unaligned data:**
2016 ///
2017 /// ```
2018 /// use std::ptr;
2019 ///
2020 /// #[repr(packed)]
2021 /// struct Packed {
2022 /// f1: u8,
2023 /// f2: u16,
2024 /// }
2025 ///
2026 /// let mut packed = Packed { f1: 1, f2: 2 };
2027 /// // `&mut packed.f2` would create an unaligned reference, and thus be Undefined Behavior!
2028 /// let raw_f2 = ptr::addr_of_mut!(packed.f2);
2029 /// unsafe { raw_f2.write_unaligned(42); }
2030 /// assert_eq!({packed.f2}, 42); // `{...}` forces copying the field instead of creating a reference.
2031 /// ```
2032 ///
2033 /// **Creating a pointer to uninitialized data:**
2034 ///
2035 /// ```rust
2036 /// use std::{ptr, mem::MaybeUninit};
2037 ///
2038 /// struct Demo {
2039 /// field: bool,
2040 /// }
2041 ///
2042 /// let mut uninit = MaybeUninit::<Demo>::uninit();
2043 /// // `&uninit.as_mut().field` would create a reference to an uninitialized `bool`,
2044 /// // and thus be Undefined Behavior!
2045 /// let f1_ptr = unsafe { ptr::addr_of_mut!((*uninit.as_mut_ptr()).field) };
2046 /// unsafe { f1_ptr.write(true); }
2047 /// let init = unsafe { uninit.assume_init() };
2048 /// ```
2049 #[stable(feature = "raw_ref_macros", since = "1.51.0")]
2050 #[rustc_macro_transparency = "semitransparent"]
2051 #[allow_internal_unstable(raw_ref_op)]
2052 pub macro addr_of_mut($place:expr) {
2053 &raw mut $place
2054 }