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 or write to them. For |
8 | //! this to be safe, these pointers must be *valid* for the given access. Whether a pointer is valid |
9 | //! depends on the operation it is used for (read or write), and the extent of the memory that is |
10 | //! accessed (i.e., how many bytes are read/written) -- it makes no sense to ask "is this pointer |
11 | //! valid"; one has to ask "is this pointer valid for a given access". Most functions use `*mut T` |
12 | //! and `*const T` to access only a single value, in which case the documentation omits the size and |
13 | //! 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::dangling` 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. That is, reference and pointer accesses cannot be |
39 | //! interleaved. |
40 | //! |
41 | //! These axioms, along with careful use of [`offset`] for pointer arithmetic, |
42 | //! are enough to correctly implement many useful things in unsafe code. Stronger guarantees |
43 | //! will be provided eventually, as the [aliasing] rules are being determined. For more |
44 | //! information, see the [book] as well as the section in the reference devoted |
45 | //! to [undefined behavior][ub]. |
46 | //! |
47 | //! We say that a pointer is "dangling" if it is not valid for any non-zero-sized accesses. This |
48 | //! means out-of-bounds pointers, pointers to freed memory, null pointers, and pointers created with |
49 | //! [`NonNull::dangling`] are all dangling. |
50 | //! |
51 | //! ## Alignment |
52 | //! |
53 | //! Valid raw pointers as defined above are not necessarily properly aligned (where |
54 | //! "proper" alignment is defined by the pointee type, i.e., `*const T` must be |
55 | //! aligned to `mem::align_of::<T>()`). However, most functions require their |
56 | //! arguments to be properly aligned, and will explicitly state |
57 | //! this requirement in their documentation. Notable exceptions to this are |
58 | //! [`read_unaligned`] and [`write_unaligned`]. |
59 | //! |
60 | //! When a function requires proper alignment, it does so even if the access |
61 | //! has size 0, i.e., even if memory is not actually touched. Consider using |
62 | //! [`NonNull::dangling`] in such cases. |
63 | //! |
64 | //! ## Allocated object |
65 | //! |
66 | //! For several operations, such as [`offset`] or field projections (`expr.field`), the notion of an |
67 | //! "allocated object" becomes relevant. An allocated object is a contiguous region of memory. |
68 | //! Common examples of allocated objects include stack-allocated variables (each variable is a |
69 | //! separate allocated object), heap allocations (each allocation created by the global allocator is |
70 | //! a separate allocated object), and `static` variables. |
71 | //! |
72 | //! # Strict Provenance |
73 | //! |
74 | //! **The following text is non-normative, insufficiently formal, and is an extremely strict |
75 | //! interpretation of provenance. It's ok if your code doesn't strictly conform to it.** |
76 | //! |
77 | //! [Strict Provenance][] is an experimental set of APIs that help tools that try |
78 | //! to validate the memory-safety of your program's execution. Notably this includes [Miri][] |
79 | //! and [CHERI][], which can detect when you access out of bounds memory or otherwise violate |
80 | //! Rust's memory model. |
81 | //! |
82 | //! Provenance must exist in some form for any programming |
83 | //! language compiled for modern computer architectures, but specifying a model for provenance |
84 | //! in a way that is useful to both compilers and programmers is an ongoing challenge. |
85 | //! The [Strict Provenance][] experiment seeks to explore the question: *what if we just said you |
86 | //! couldn't do all the nasty operations that make provenance so messy?* |
87 | //! |
88 | //! What APIs would have to be removed? What APIs would have to be added? How much would code |
89 | //! have to change, and is it worse or better now? Would any patterns become truly inexpressible? |
90 | //! Could we carve out special exceptions for those patterns? Should we? |
91 | //! |
92 | //! A secondary goal of this project is to see if we can disambiguate the many functions of |
93 | //! pointer<->integer casts enough for the definition of `usize` to be loosened so that it |
94 | //! isn't *pointer*-sized but address-space/offset/allocation-sized (we'll probably continue |
95 | //! to conflate these notions). This would potentially make it possible to more efficiently |
96 | //! target platforms where pointers are larger than offsets, such as CHERI and maybe some |
97 | //! segmented architectures. |
98 | //! |
99 | //! ## Provenance |
100 | //! |
101 | //! **This section is *non-normative* and is part of the [Strict Provenance][] experiment.** |
102 | //! |
103 | //! Pointers are not *simply* an "integer" or "address". For instance, it's uncontroversial |
104 | //! to say that a Use After Free is clearly Undefined Behaviour, even if you "get lucky" |
105 | //! and the freed memory gets reallocated before your read/write (in fact this is the |
106 | //! worst-case scenario, UAFs would be much less concerning if this didn't happen!). |
107 | //! To rationalize this claim, pointers need to somehow be *more* than just their addresses: |
108 | //! they must have provenance. |
109 | //! |
110 | //! When an allocation is created, that allocation has a unique Original Pointer. For alloc |
111 | //! APIs this is literally the pointer the call returns, and for local variables and statics, |
112 | //! this is the name of the variable/static. This is mildly overloading the term "pointer" |
113 | //! for the sake of brevity/exposition. |
114 | //! |
115 | //! The Original Pointer for an allocation is guaranteed to have unique access to the entire |
116 | //! allocation and *only* that allocation. In this sense, an allocation can be thought of |
117 | //! as a "sandbox" that cannot be broken into or out of. *Provenance* is the permission |
118 | //! to access an allocation's sandbox and has both a *spatial* and *temporal* component: |
119 | //! |
120 | //! * Spatial: A range of bytes that the pointer is allowed to access. |
121 | //! * Temporal: The lifetime (of the allocation) that access to these bytes is tied to. |
122 | //! |
123 | //! Spatial provenance makes sure you don't go beyond your sandbox, while temporal provenance |
124 | //! makes sure that you can't "get lucky" after your permission to access some memory |
125 | //! has been revoked (either through deallocations or borrows expiring). |
126 | //! |
127 | //! Provenance is implicitly shared with all pointers transitively derived from |
128 | //! The Original Pointer through operations like [`offset`], borrowing, and pointer casts. |
129 | //! Some operations may *shrink* the derived provenance, limiting how much memory it can |
130 | //! access or how long it's valid for (i.e. borrowing a subfield and subslicing). |
131 | //! |
132 | //! Shrinking provenance cannot be undone: even if you "know" there is a larger allocation, you |
133 | //! can't derive a pointer with a larger provenance. Similarly, you cannot "recombine" |
134 | //! two contiguous provenances back into one (i.e. with a `fn merge(&[T], &[T]) -> &[T]`). |
135 | //! |
136 | //! A reference to a value always has provenance over exactly the memory that field occupies. |
137 | //! A reference to a slice always has provenance over exactly the range that slice describes. |
138 | //! |
139 | //! If an allocation is deallocated, all pointers with provenance to that allocation become |
140 | //! invalidated, and effectively lose their provenance. |
141 | //! |
142 | //! The strict provenance experiment is mostly only interested in exploring stricter *spatial* |
143 | //! provenance. In this sense it can be thought of as a subset of the more ambitious and |
144 | //! formal [Stacked Borrows][] research project, which is what tools like [Miri][] are based on. |
145 | //! In particular, Stacked Borrows is necessary to properly describe what borrows are allowed |
146 | //! to do and when they become invalidated. This necessarily involves much more complex |
147 | //! *temporal* reasoning than simply identifying allocations. Adjusting APIs and code |
148 | //! for the strict provenance experiment will also greatly help Stacked Borrows. |
149 | //! |
150 | //! |
151 | //! ## Pointer Vs Addresses |
152 | //! |
153 | //! **This section is *non-normative* and is part of the [Strict Provenance][] experiment.** |
154 | //! |
155 | //! One of the largest historical issues with trying to define provenance is that programmers |
156 | //! freely convert between pointers and integers. Once you allow for this, it generally becomes |
157 | //! impossible to accurately track and preserve provenance information, and you need to appeal |
158 | //! to very complex and unreliable heuristics. But of course, converting between pointers and |
159 | //! integers is very useful, so what can we do? |
160 | //! |
161 | //! Also did you know WASM is actually a "Harvard Architecture"? As in function pointers are |
162 | //! handled completely differently from data pointers? And we kind of just shipped Rust on WASM |
163 | //! without really addressing the fact that we let you freely convert between function pointers |
164 | //! and data pointers, because it mostly Just Works? Let's just put that on the "pointer casts |
165 | //! are dubious" pile. |
166 | //! |
167 | //! Strict Provenance attempts to square these circles by decoupling Rust's traditional conflation |
168 | //! of pointers and `usize` (and `isize`), and defining a pointer to semantically contain the |
169 | //! following information: |
170 | //! |
171 | //! * The **address-space** it is part of (e.g. "data" vs "code" in WASM). |
172 | //! * The **address** it points to, which can be represented by a `usize`. |
173 | //! * The **provenance** it has, defining the memory it has permission to access. |
174 | //! Provenance can be absent, in which case the pointer does not have permission to access any memory. |
175 | //! |
176 | //! Under Strict Provenance, a usize *cannot* accurately represent a pointer, and converting from |
177 | //! a pointer to a usize is generally an operation which *only* extracts the address. It is |
178 | //! therefore *impossible* to construct a valid pointer from a usize because there is no way |
179 | //! to restore the address-space and provenance. In other words, pointer-integer-pointer |
180 | //! roundtrips are not possible (in the sense that the resulting pointer is not dereferenceable). |
181 | //! |
182 | //! The key insight to making this model *at all* viable is the [`with_addr`][] method: |
183 | //! |
184 | //! ```text |
185 | //! /// Creates a new pointer with the given address. |
186 | //! /// |
187 | //! /// This performs the same operation as an `addr as ptr` cast, but copies |
188 | //! /// the *address-space* and *provenance* of `self` to the new pointer. |
189 | //! /// This allows us to dynamically preserve and propagate this important |
190 | //! /// information in a way that is otherwise impossible with a unary cast. |
191 | //! /// |
192 | //! /// This is equivalent to using `wrapping_offset` to offset `self` to the |
193 | //! /// given address, and therefore has all the same capabilities and restrictions. |
194 | //! pub fn with_addr(self, addr: usize) -> Self; |
195 | //! ``` |
196 | //! |
197 | //! So you're still able to drop down to the address representation and do whatever |
198 | //! clever bit tricks you want *as long as* you're able to keep around a pointer |
199 | //! into the allocation you care about that can "reconstitute" the other parts of the pointer. |
200 | //! Usually this is very easy, because you only are taking a pointer, messing with the address, |
201 | //! and then immediately converting back to a pointer. To make this use case more ergonomic, |
202 | //! we provide the [`map_addr`][] method. |
203 | //! |
204 | //! To help make it clear that code is "following" Strict Provenance semantics, we also provide an |
205 | //! [`addr`][] method which promises that the returned address is not part of a |
206 | //! pointer-usize-pointer roundtrip. In the future we may provide a lint for pointer<->integer |
207 | //! casts to help you audit if your code conforms to strict provenance. |
208 | //! |
209 | //! |
210 | //! ## Using Strict Provenance |
211 | //! |
212 | //! Most code needs no changes to conform to strict provenance, as the only really concerning |
213 | //! operation that *wasn't* obviously already Undefined Behaviour is casts from usize to a |
214 | //! pointer. For code which *does* cast a usize to a pointer, the scope of the change depends |
215 | //! on exactly what you're doing. |
216 | //! |
217 | //! In general you just need to make sure that if you want to convert a usize address to a |
218 | //! pointer and then use that pointer to read/write memory, you need to keep around a pointer |
219 | //! that has sufficient provenance to perform that read/write itself. In this way all of your |
220 | //! casts from an address to a pointer are essentially just applying offsets/indexing. |
221 | //! |
222 | //! This is generally trivial to do for simple cases like tagged pointers *as long as you |
223 | //! represent the tagged pointer as an actual pointer and not a usize*. For instance: |
224 | //! |
225 | //! ``` |
226 | //! #![feature(strict_provenance)] |
227 | //! |
228 | //! unsafe { |
229 | //! // A flag we want to pack into our pointer |
230 | //! static HAS_DATA: usize = 0x1; |
231 | //! static FLAG_MASK: usize = !HAS_DATA; |
232 | //! |
233 | //! // Our value, which must have enough alignment to have spare least-significant-bits. |
234 | //! let my_precious_data: u32 = 17; |
235 | //! assert!(core::mem::align_of::<u32>() > 1); |
236 | //! |
237 | //! // Create a tagged pointer |
238 | //! let ptr = &my_precious_data as *const u32; |
239 | //! let tagged = ptr.map_addr(|addr| addr | HAS_DATA); |
240 | //! |
241 | //! // Check the flag: |
242 | //! if tagged.addr() & HAS_DATA != 0 { |
243 | //! // Untag and read the pointer |
244 | //! let data = *tagged.map_addr(|addr| addr & FLAG_MASK); |
245 | //! assert_eq!(data, 17); |
246 | //! } else { |
247 | //! unreachable!() |
248 | //! } |
249 | //! } |
250 | //! ``` |
251 | //! |
252 | //! (Yes, if you've been using AtomicUsize for pointers in concurrent datastructures, you should |
253 | //! be using AtomicPtr instead. If that messes up the way you atomically manipulate pointers, |
254 | //! we would like to know why, and what needs to be done to fix it.) |
255 | //! |
256 | //! Something more complicated and just generally *evil* like an XOR-List requires more significant |
257 | //! changes like allocating all nodes in a pre-allocated Vec or Arena and using a pointer |
258 | //! to the whole allocation to reconstitute the XORed addresses. |
259 | //! |
260 | //! Situations where a valid pointer *must* be created from just an address, such as baremetal code |
261 | //! accessing a memory-mapped interface at a fixed address, are an open question on how to support. |
262 | //! These situations *will* still be allowed, but we might require some kind of "I know what I'm |
263 | //! doing" annotation to explain the situation to the compiler. It's also possible they need no |
264 | //! special attention at all, because they're generally accessing memory outside the scope of |
265 | //! "the abstract machine", or already using "I know what I'm doing" annotations like "volatile". |
266 | //! |
267 | //! Under [Strict Provenance] it is Undefined Behaviour to: |
268 | //! |
269 | //! * Access memory through a pointer that does not have provenance over that memory. |
270 | //! |
271 | //! * [`offset`] a pointer to or from an address it doesn't have provenance over. |
272 | //! This means it's always UB to offset a pointer derived from something deallocated, |
273 | //! even if the offset is 0. Note that a pointer "one past the end" of its provenance |
274 | //! is not actually outside its provenance, it just has 0 bytes it can load/store. |
275 | //! |
276 | //! But it *is* still sound to: |
277 | //! |
278 | //! * Create a pointer without provenance from just an address (see [`ptr::dangling`][]). Such a |
279 | //! pointer cannot be used for memory accesses (except for zero-sized accesses). This can still be |
280 | //! useful for sentinel values like `null` *or* to represent a tagged pointer that will never be |
281 | //! dereferenceable. In general, it is always sound for an integer to pretend to be a pointer "for |
282 | //! fun" as long as you don't use operations on it which require it to be valid (non-zero-sized |
283 | //! offset, read, write, etc). |
284 | //! |
285 | //! * Forge an allocation of size zero at any sufficiently aligned non-null address. |
286 | //! i.e. the usual "ZSTs are fake, do what you want" rules apply *but* this only applies |
287 | //! for actual forgery (integers cast to pointers). If you borrow some struct's field |
288 | //! that *happens* to be zero-sized, the resulting pointer will have provenance tied to |
289 | //! that allocation and it will still get invalidated if the allocation gets deallocated. |
290 | //! In the future we may introduce an API to make such a forged allocation explicit. |
291 | //! |
292 | //! * [`wrapping_offset`][] a pointer outside its provenance. This includes pointers |
293 | //! which have "no" provenance. Unfortunately there may be practical limits on this for a |
294 | //! particular platform, and it's an open question as to how to specify this (if at all). |
295 | //! Notably, [CHERI][] relies on a compression scheme that can't handle a |
296 | //! pointer getting offset "too far" out of bounds. If this happens, the address |
297 | //! returned by `addr` will be the value you expect, but the provenance will get invalidated |
298 | //! and using it to read/write will fault. The details of this are architecture-specific |
299 | //! and based on alignment, but the buffer on either side of the pointer's range is pretty |
300 | //! generous (think kilobytes, not bytes). |
301 | //! |
302 | //! * Compare arbitrary pointers by address. Addresses *are* just integers and so there is |
303 | //! always a coherent answer, even if the pointers are dangling or from different |
304 | //! address-spaces/provenances. Of course, comparing addresses from different address-spaces |
305 | //! is generally going to be *meaningless*, but so is comparing Kilograms to Meters, and Rust |
306 | //! doesn't prevent that either. Similarly, if you get "lucky" and notice that a pointer |
307 | //! one-past-the-end is the "same" address as the start of an unrelated allocation, anything |
308 | //! you do with that fact is *probably* going to be gibberish. The scope of that gibberish |
309 | //! is kept under control by the fact that the two pointers *still* aren't allowed to access |
310 | //! the other's allocation (bytes), because they still have different provenance. |
311 | //! |
312 | //! * Perform pointer tagging tricks. This falls out of [`wrapping_offset`] but is worth |
313 | //! mentioning in more detail because of the limitations of [CHERI][]. Low-bit tagging |
314 | //! is very robust, and often doesn't even go out of bounds because types ensure |
315 | //! size >= align (and over-aligning actually gives CHERI more flexibility). Anything |
316 | //! more complex than this rapidly enters "extremely platform-specific" territory as |
317 | //! certain things may or may not be allowed based on specific supported operations. |
318 | //! For instance, ARM explicitly supports high-bit tagging, and so CHERI on ARM inherits |
319 | //! that and should support it. |
320 | //! |
321 | //! ## Exposed Provenance |
322 | //! |
323 | //! **This section is *non-normative* and is an extension to the [Strict Provenance] experiment.** |
324 | //! |
325 | //! As discussed above, pointer-usize-pointer roundtrips are not possible under [Strict Provenance]. |
326 | //! This is by design: the goal of Strict Provenance is to provide a clear specification that we are |
327 | //! confident can be formalized unambiguously and can be subject to precise formal reasoning. |
328 | //! |
329 | //! However, there exist situations where pointer-usize-pointer roundtrips cannot be avoided, or |
330 | //! where avoiding them would require major refactoring. Legacy platform APIs also regularly assume |
331 | //! that `usize` can capture all the information that makes up a pointer. The goal of Strict |
332 | //! Provenance is not to rule out such code; the goal is to put all the *other* pointer-manipulating |
333 | //! code onto a more solid foundation. Strict Provenance is about improving the situation where |
334 | //! possible (all the code that can be written with Strict Provenance) without making things worse |
335 | //! for situations where Strict Provenance is insufficient. |
336 | //! |
337 | //! For these situations, there is a highly experimental extension to Strict Provenance called |
338 | //! *Exposed Provenance*. This extension permits pointer-usize-pointer roundtrips. However, its |
339 | //! semantics are on much less solid footing than Strict Provenance, and at this point it is not yet |
340 | //! clear where a satisfying unambiguous semantics can be defined for Exposed Provenance. |
341 | //! Furthermore, Exposed Provenance will not work (well) with tools like [Miri] and [CHERI]. |
342 | //! |
343 | //! Exposed Provenance is provided by the [`expose_provenance`] and [`with_exposed_provenance`] methods, |
344 | //! which are meant to replace `as` casts between pointers and integers. [`expose_provenance`] is a lot like |
345 | //! [`addr`], but additionally adds the provenance of the pointer to a global list of 'exposed' |
346 | //! provenances. (This list is purely conceptual, it exists for the purpose of specifying Rust but |
347 | //! is not materialized in actual executions, except in tools like [Miri].) [`with_exposed_provenance`] |
348 | //! can be used to construct a pointer with one of these previously 'exposed' provenances. |
349 | //! [`with_exposed_provenance`] takes only `addr: usize` as arguments, so unlike in [`with_addr`] there is |
350 | //! no indication of what the correct provenance for the returned pointer is -- and that is exactly |
351 | //! what makes pointer-usize-pointer roundtrips so tricky to rigorously specify! There is no |
352 | //! algorithm that decides which provenance will be used. You can think of this as "guessing" the |
353 | //! right provenance, and the guess will be "maximally in your favor", in the sense that if there is |
354 | //! any way to avoid undefined behavior, then that is the guess that will be taken. However, if |
355 | //! there is *no* previously 'exposed' provenance that justifies the way the returned pointer will |
356 | //! be used, the program has undefined behavior. |
357 | //! |
358 | //! Using [`expose_provenance`] or [`with_exposed_provenance`] (or the `as` casts) means that code is |
359 | //! *not* following Strict Provenance rules. The goal of the Strict Provenance experiment is to |
360 | //! determine how far one can get in Rust without the use of [`expose_provenance`] and |
361 | //! [`with_exposed_provenance`], and to encourage code to be written with Strict Provenance APIs only. |
362 | //! Maximizing the amount of such code is a major win for avoiding specification complexity and to |
363 | //! facilitate adoption of tools like [CHERI] and [Miri] that can be a big help in increasing the |
364 | //! confidence in (unsafe) Rust code. |
365 | //! |
366 | //! [aliasing]: ../../nomicon/aliasing.html |
367 | //! [book]: ../../book/ch19-01-unsafe-rust.html#dereferencing-a-raw-pointer |
368 | //! [ub]: ../../reference/behavior-considered-undefined.html |
369 | //! [zst]: ../../nomicon/exotic-sizes.html#zero-sized-types-zsts |
370 | //! [atomic operations]: crate::sync::atomic |
371 | //! [`offset`]: pointer::offset |
372 | //! [`wrapping_offset`]: pointer::wrapping_offset |
373 | //! [`with_addr`]: pointer::with_addr |
374 | //! [`map_addr`]: pointer::map_addr |
375 | //! [`addr`]: pointer::addr |
376 | //! [`ptr::dangling`]: core::ptr::dangling |
377 | //! [`expose_provenance`]: pointer::expose_provenance |
378 | //! [`with_exposed_provenance`]: with_exposed_provenance |
379 | //! [Miri]: https://github.com/rust-lang/miri |
380 | //! [CHERI]: https://www.cl.cam.ac.uk/research/security/ctsrd/cheri/ |
381 | //! [Strict Provenance]: https://github.com/rust-lang/rust/issues/95228 |
382 | //! [Stacked Borrows]: https://plv.mpi-sws.org/rustbelt/stacked-borrows/ |
383 | |
384 | #![stable (feature = "rust1" , since = "1.0.0" )] |
385 | // There are many unsafe functions taking pointers that don't dereference them. |
386 | #![allow (clippy::not_unsafe_ptr_arg_deref)] |
387 | |
388 | use crate::cmp::Ordering; |
389 | use crate::fmt; |
390 | use crate::hash; |
391 | use crate::intrinsics; |
392 | use crate::marker::FnPtr; |
393 | use crate::ub_checks; |
394 | |
395 | use crate::mem::{self, align_of, size_of, MaybeUninit}; |
396 | |
397 | mod alignment; |
398 | #[unstable (feature = "ptr_alignment_type" , issue = "102070" )] |
399 | pub use alignment::Alignment; |
400 | |
401 | #[stable (feature = "rust1" , since = "1.0.0" )] |
402 | #[doc (inline)] |
403 | pub use crate::intrinsics::copy_nonoverlapping; |
404 | |
405 | #[stable (feature = "rust1" , since = "1.0.0" )] |
406 | #[doc (inline)] |
407 | pub use crate::intrinsics::copy; |
408 | |
409 | #[stable (feature = "rust1" , since = "1.0.0" )] |
410 | #[doc (inline)] |
411 | pub use crate::intrinsics::write_bytes; |
412 | |
413 | mod metadata; |
414 | #[unstable (feature = "ptr_metadata" , issue = "81513" )] |
415 | pub use metadata::{from_raw_parts, from_raw_parts_mut, metadata, DynMetadata, Pointee, Thin}; |
416 | |
417 | mod non_null; |
418 | #[stable (feature = "nonnull" , since = "1.25.0" )] |
419 | pub use non_null::NonNull; |
420 | |
421 | mod unique; |
422 | #[unstable (feature = "ptr_internals" , issue = "none" )] |
423 | pub use unique::Unique; |
424 | |
425 | mod const_ptr; |
426 | mod mut_ptr; |
427 | |
428 | /// Executes the destructor (if any) of the pointed-to value. |
429 | /// |
430 | /// This is semantically equivalent to calling [`ptr::read`] and discarding |
431 | /// the result, but has the following advantages: |
432 | /// |
433 | /// * It is *required* to use `drop_in_place` to drop unsized types like |
434 | /// trait objects, because they can't be read out onto the stack and |
435 | /// dropped normally. |
436 | /// |
437 | /// * It is friendlier to the optimizer to do this over [`ptr::read`] when |
438 | /// dropping manually allocated memory (e.g., in the implementations of |
439 | /// `Box`/`Rc`/`Vec`), as the compiler doesn't need to prove that it's |
440 | /// sound to elide the copy. |
441 | /// |
442 | /// * It can be used to drop [pinned] data when `T` is not `repr(packed)` |
443 | /// (pinned data must not be moved before it is dropped). |
444 | /// |
445 | /// Unaligned values cannot be dropped in place, they must be copied to an aligned |
446 | /// location first using [`ptr::read_unaligned`]. For packed structs, this move is |
447 | /// done automatically by the compiler. This means the fields of packed structs |
448 | /// are not dropped in-place. |
449 | /// |
450 | /// [`ptr::read`]: self::read |
451 | /// [`ptr::read_unaligned`]: self::read_unaligned |
452 | /// [pinned]: crate::pin |
453 | /// |
454 | /// # Safety |
455 | /// |
456 | /// Behavior is undefined if any of the following conditions are violated: |
457 | /// |
458 | /// * `to_drop` must be [valid] for both reads and writes. |
459 | /// |
460 | /// * `to_drop` must be properly aligned, even if `T` has size 0. |
461 | /// |
462 | /// * `to_drop` must be nonnull, even if `T` has size 0. |
463 | /// |
464 | /// * The value `to_drop` points to must be valid for dropping, which may mean |
465 | /// it must uphold additional invariants. These invariants depend on the type |
466 | /// of the value being dropped. For instance, when dropping a Box, the box's |
467 | /// pointer to the heap must be valid. |
468 | /// |
469 | /// * While `drop_in_place` is executing, the only way to access parts of |
470 | /// `to_drop` is through the `&mut self` references supplied to the |
471 | /// `Drop::drop` methods that `drop_in_place` invokes. |
472 | /// |
473 | /// Additionally, if `T` is not [`Copy`], using the pointed-to value after |
474 | /// calling `drop_in_place` can cause undefined behavior. Note that `*to_drop = |
475 | /// foo` counts as a use because it will cause the value to be dropped |
476 | /// again. [`write()`] can be used to overwrite data without causing it to be |
477 | /// dropped. |
478 | /// |
479 | /// [valid]: self#safety |
480 | /// |
481 | /// # Examples |
482 | /// |
483 | /// Manually remove the last item from a vector: |
484 | /// |
485 | /// ``` |
486 | /// use std::ptr; |
487 | /// use std::rc::Rc; |
488 | /// |
489 | /// let last = Rc::new(1); |
490 | /// let weak = Rc::downgrade(&last); |
491 | /// |
492 | /// let mut v = vec![Rc::new(0), last]; |
493 | /// |
494 | /// unsafe { |
495 | /// // Get a raw pointer to the last element in `v`. |
496 | /// let ptr = &mut v[1] as *mut _; |
497 | /// // Shorten `v` to prevent the last item from being dropped. We do that first, |
498 | /// // to prevent issues if the `drop_in_place` below panics. |
499 | /// v.set_len(1); |
500 | /// // Without a call `drop_in_place`, the last item would never be dropped, |
501 | /// // and the memory it manages would be leaked. |
502 | /// ptr::drop_in_place(ptr); |
503 | /// } |
504 | /// |
505 | /// assert_eq!(v, &[0.into()]); |
506 | /// |
507 | /// // Ensure that the last item was dropped. |
508 | /// assert!(weak.upgrade().is_none()); |
509 | /// ``` |
510 | #[stable (feature = "drop_in_place" , since = "1.8.0" )] |
511 | #[lang = "drop_in_place" ] |
512 | #[allow (unconditional_recursion)] |
513 | #[rustc_diagnostic_item = "ptr_drop_in_place" ] |
514 | pub unsafe fn drop_in_place<T: ?Sized>(to_drop: *mut T) { |
515 | // Code here does not matter - this is replaced by the |
516 | // real drop glue by the compiler. |
517 | |
518 | // SAFETY: see comment above |
519 | unsafe { drop_in_place(to_drop) } |
520 | } |
521 | |
522 | /// Creates a null raw pointer. |
523 | /// |
524 | /// This function is equivalent to zero-initializing the pointer: |
525 | /// `MaybeUninit::<*const T>::zeroed().assume_init()`. |
526 | /// The resulting pointer has the address 0. |
527 | /// |
528 | /// # Examples |
529 | /// |
530 | /// ``` |
531 | /// use std::ptr; |
532 | /// |
533 | /// let p: *const i32 = ptr::null(); |
534 | /// assert!(p.is_null()); |
535 | /// assert_eq!(p as usize, 0); // this pointer has the address 0 |
536 | /// ``` |
537 | #[inline (always)] |
538 | #[must_use ] |
539 | #[stable (feature = "rust1" , since = "1.0.0" )] |
540 | #[rustc_promotable ] |
541 | #[rustc_const_stable (feature = "const_ptr_null" , since = "1.24.0" )] |
542 | #[rustc_allow_const_fn_unstable (ptr_metadata)] |
543 | #[rustc_diagnostic_item = "ptr_null" ] |
544 | pub const fn null<T: ?Sized + Thin>() -> *const T { |
545 | from_raw_parts(data_pointer:without_provenance(0), ()) |
546 | } |
547 | |
548 | /// Creates a null mutable raw pointer. |
549 | /// |
550 | /// This function is equivalent to zero-initializing the pointer: |
551 | /// `MaybeUninit::<*mut T>::zeroed().assume_init()`. |
552 | /// The resulting pointer has the address 0. |
553 | /// |
554 | /// # Examples |
555 | /// |
556 | /// ``` |
557 | /// use std::ptr; |
558 | /// |
559 | /// let p: *mut i32 = ptr::null_mut(); |
560 | /// assert!(p.is_null()); |
561 | /// assert_eq!(p as usize, 0); // this pointer has the address 0 |
562 | /// ``` |
563 | #[inline (always)] |
564 | #[must_use ] |
565 | #[stable (feature = "rust1" , since = "1.0.0" )] |
566 | #[rustc_promotable ] |
567 | #[rustc_const_stable (feature = "const_ptr_null" , since = "1.24.0" )] |
568 | #[rustc_allow_const_fn_unstable (ptr_metadata)] |
569 | #[rustc_diagnostic_item = "ptr_null_mut" ] |
570 | pub const fn null_mut<T: ?Sized + Thin>() -> *mut T { |
571 | from_raw_parts_mut(data_pointer:without_provenance_mut(0), ()) |
572 | } |
573 | |
574 | /// Creates a pointer with the given address and no provenance. |
575 | /// |
576 | /// This is equivalent to `ptr::null().with_addr(addr)`. |
577 | /// |
578 | /// Without provenance, this pointer is not associated with any actual allocation. Such a |
579 | /// no-provenance pointer may be used for zero-sized memory accesses (if suitably aligned), but |
580 | /// non-zero-sized memory accesses with a no-provenance pointer are UB. No-provenance pointers are |
581 | /// little more than a usize address in disguise. |
582 | /// |
583 | /// This is different from `addr as *const T`, which creates a pointer that picks up a previously |
584 | /// exposed provenance. See [`with_exposed_provenance`] for more details on that operation. |
585 | /// |
586 | /// This API and its claimed semantics are part of the Strict Provenance experiment, |
587 | /// see the [module documentation][crate::ptr] for details. |
588 | #[inline (always)] |
589 | #[must_use ] |
590 | #[rustc_const_stable (feature = "stable_things_using_strict_provenance" , since = "1.61.0" )] |
591 | #[unstable (feature = "strict_provenance" , issue = "95228" )] |
592 | pub const fn without_provenance<T>(addr: usize) -> *const T { |
593 | // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic. |
594 | // We use transmute rather than a cast so tools like Miri can tell that this |
595 | // is *not* the same as with_exposed_provenance. |
596 | // SAFETY: every valid integer is also a valid pointer (as long as you don't dereference that |
597 | // pointer). |
598 | unsafe { mem::transmute(src:addr) } |
599 | } |
600 | |
601 | /// Creates a new pointer that is dangling, but well-aligned. |
602 | /// |
603 | /// This is useful for initializing types which lazily allocate, like |
604 | /// `Vec::new` does. |
605 | /// |
606 | /// Note that the pointer value may potentially represent a valid pointer to |
607 | /// a `T`, which means this must not be used as a "not yet initialized" |
608 | /// sentinel value. Types that lazily allocate must track initialization by |
609 | /// some other means. |
610 | #[inline (always)] |
611 | #[must_use ] |
612 | #[rustc_const_stable (feature = "stable_things_using_strict_provenance" , since = "1.61.0" )] |
613 | #[unstable (feature = "strict_provenance" , issue = "95228" )] |
614 | pub const fn dangling<T>() -> *const T { |
615 | without_provenance(addr:mem::align_of::<T>()) |
616 | } |
617 | |
618 | /// Creates a pointer with the given address and no provenance. |
619 | /// |
620 | /// This is equivalent to `ptr::null_mut().with_addr(addr)`. |
621 | /// |
622 | /// Without provenance, this pointer is not associated with any actual allocation. Such a |
623 | /// no-provenance pointer may be used for zero-sized memory accesses (if suitably aligned), but |
624 | /// non-zero-sized memory accesses with a no-provenance pointer are UB. No-provenance pointers are |
625 | /// little more than a usize address in disguise. |
626 | /// |
627 | /// This is different from `addr as *mut T`, which creates a pointer that picks up a previously |
628 | /// exposed provenance. See [`with_exposed_provenance_mut`] for more details on that operation. |
629 | /// |
630 | /// This API and its claimed semantics are part of the Strict Provenance experiment, |
631 | /// see the [module documentation][crate::ptr] for details. |
632 | #[inline (always)] |
633 | #[must_use ] |
634 | #[rustc_const_stable (feature = "stable_things_using_strict_provenance" , since = "1.61.0" )] |
635 | #[unstable (feature = "strict_provenance" , issue = "95228" )] |
636 | pub const fn without_provenance_mut<T>(addr: usize) -> *mut T { |
637 | // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic. |
638 | // We use transmute rather than a cast so tools like Miri can tell that this |
639 | // is *not* the same as with_exposed_provenance. |
640 | // SAFETY: every valid integer is also a valid pointer (as long as you don't dereference that |
641 | // pointer). |
642 | unsafe { mem::transmute(src:addr) } |
643 | } |
644 | |
645 | /// Creates a new pointer that is dangling, but well-aligned. |
646 | /// |
647 | /// This is useful for initializing types which lazily allocate, like |
648 | /// `Vec::new` does. |
649 | /// |
650 | /// Note that the pointer value may potentially represent a valid pointer to |
651 | /// a `T`, which means this must not be used as a "not yet initialized" |
652 | /// sentinel value. Types that lazily allocate must track initialization by |
653 | /// some other means. |
654 | #[inline (always)] |
655 | #[must_use ] |
656 | #[rustc_const_stable (feature = "stable_things_using_strict_provenance" , since = "1.61.0" )] |
657 | #[unstable (feature = "strict_provenance" , issue = "95228" )] |
658 | pub const fn dangling_mut<T>() -> *mut T { |
659 | without_provenance_mut(addr:mem::align_of::<T>()) |
660 | } |
661 | |
662 | /// Convert an address back to a pointer, picking up a previously 'exposed' provenance. |
663 | /// |
664 | /// This is a more rigorously specified alternative to `addr as *const T`. The provenance of the |
665 | /// returned pointer is that of *any* pointer that was previously exposed by passing it to |
666 | /// [`expose_provenance`][pointer::expose_provenance], or a `ptr as usize` cast. In addition, memory which is |
667 | /// outside the control of the Rust abstract machine (MMIO registers, for example) is always |
668 | /// considered to be exposed, so long as this memory is disjoint from memory that will be used by |
669 | /// the abstract machine such as the stack, heap, and statics. |
670 | /// |
671 | /// If there is no 'exposed' provenance that justifies the way this pointer will be used, |
672 | /// the program has undefined behavior. In particular, the aliasing rules still apply: pointers |
673 | /// and references that have been invalidated due to aliasing accesses cannot be used any more, |
674 | /// even if they have been exposed! |
675 | /// |
676 | /// Note that there is no algorithm that decides which provenance will be used. You can think of this |
677 | /// as "guessing" the right provenance, and the guess will be "maximally in your favor", in the sense |
678 | /// that if there is any way to avoid undefined behavior (while upholding all aliasing requirements), |
679 | /// then that is the guess that will be taken. |
680 | /// |
681 | /// On platforms with multiple address spaces, it is your responsibility to ensure that the |
682 | /// address makes sense in the address space that this pointer will be used with. |
683 | /// |
684 | /// Using this function means that code is *not* following [Strict |
685 | /// Provenance][self#strict-provenance] rules. "Guessing" a |
686 | /// suitable provenance complicates specification and reasoning and may not be supported by |
687 | /// tools that help you to stay conformant with the Rust memory model, so it is recommended to |
688 | /// use [`with_addr`][pointer::with_addr] wherever possible. |
689 | /// |
690 | /// On most platforms this will produce a value with the same bytes as the address. Platforms |
691 | /// which need to store additional information in a pointer may not support this operation, |
692 | /// since it is generally not possible to actually *compute* which provenance the returned |
693 | /// pointer has to pick up. |
694 | /// |
695 | /// It is unclear whether this function can be given a satisfying unambiguous specification. This |
696 | /// API and its claimed semantics are part of [Exposed Provenance][self#exposed-provenance]. |
697 | #[must_use ] |
698 | #[inline (always)] |
699 | #[unstable (feature = "exposed_provenance" , issue = "95228" )] |
700 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
701 | #[allow (fuzzy_provenance_casts)] // this *is* the explicit provenance API one should use instead |
702 | pub fn with_exposed_provenance<T>(addr: usize) -> *const T |
703 | where |
704 | T: Sized, |
705 | { |
706 | // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic. |
707 | addr as *const T |
708 | } |
709 | |
710 | /// Convert an address back to a mutable pointer, picking up a previously 'exposed' provenance. |
711 | /// |
712 | /// This is a more rigorously specified alternative to `addr as *mut T`. The provenance of the |
713 | /// returned pointer is that of *any* pointer that was previously passed to |
714 | /// [`expose_provenance`][pointer::expose_provenance] or a `ptr as usize` cast. If there is no previously |
715 | /// 'exposed' provenance that justifies the way this pointer will be used, the program has undefined |
716 | /// behavior. Note that there is no algorithm that decides which provenance will be used. You can |
717 | /// think of this as "guessing" the right provenance, and the guess will be "maximally in your |
718 | /// favor", in the sense that if there is any way to avoid undefined behavior, then that is the |
719 | /// guess that will be taken. |
720 | /// |
721 | /// On platforms with multiple address spaces, it is your responsibility to ensure that the |
722 | /// address makes sense in the address space that this pointer will be used with. |
723 | /// |
724 | /// Using this function means that code is *not* following [Strict |
725 | /// Provenance][self#strict-provenance] rules. "Guessing" a |
726 | /// suitable provenance complicates specification and reasoning and may not be supported by |
727 | /// tools that help you to stay conformant with the Rust memory model, so it is recommended to |
728 | /// use [`with_addr`][pointer::with_addr] wherever possible. |
729 | /// |
730 | /// On most platforms this will produce a value with the same bytes as the address. Platforms |
731 | /// which need to store additional information in a pointer may not support this operation, |
732 | /// since it is generally not possible to actually *compute* which provenance the returned |
733 | /// pointer has to pick up. |
734 | /// |
735 | /// It is unclear whether this function can be given a satisfying unambiguous specification. This |
736 | /// API and its claimed semantics are part of [Exposed Provenance][self#exposed-provenance]. |
737 | #[must_use ] |
738 | #[inline (always)] |
739 | #[unstable (feature = "exposed_provenance" , issue = "95228" )] |
740 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
741 | #[allow (fuzzy_provenance_casts)] // this *is* the explicit provenance API one should use instead |
742 | pub fn with_exposed_provenance_mut<T>(addr: usize) -> *mut T |
743 | where |
744 | T: Sized, |
745 | { |
746 | // FIXME(strict_provenance_magic): I am magic and should be a compiler intrinsic. |
747 | addr as *mut T |
748 | } |
749 | |
750 | /// Convert a reference to a raw pointer. |
751 | /// |
752 | /// This is equivalent to `r as *const T`, but is a bit safer since it will never silently change |
753 | /// type or mutability, in particular if the code is refactored. |
754 | #[inline (always)] |
755 | #[must_use ] |
756 | #[stable (feature = "ptr_from_ref" , since = "1.76.0" )] |
757 | #[rustc_const_stable (feature = "ptr_from_ref" , since = "1.76.0" )] |
758 | #[rustc_never_returns_null_ptr ] |
759 | #[rustc_diagnostic_item = "ptr_from_ref" ] |
760 | pub const fn from_ref<T: ?Sized>(r: &T) -> *const T { |
761 | r |
762 | } |
763 | |
764 | /// Convert a mutable reference to a raw pointer. |
765 | /// |
766 | /// This is equivalent to `r as *mut T`, but is a bit safer since it will never silently change |
767 | /// type or mutability, in particular if the code is refactored. |
768 | #[inline (always)] |
769 | #[must_use ] |
770 | #[stable (feature = "ptr_from_ref" , since = "1.76.0" )] |
771 | #[rustc_const_stable (feature = "ptr_from_ref" , since = "1.76.0" )] |
772 | #[rustc_allow_const_fn_unstable (const_mut_refs)] |
773 | #[rustc_never_returns_null_ptr ] |
774 | pub const fn from_mut<T: ?Sized>(r: &mut T) -> *mut T { |
775 | r |
776 | } |
777 | |
778 | /// Forms a raw slice from a pointer and a length. |
779 | /// |
780 | /// The `len` argument is the number of **elements**, not the number of bytes. |
781 | /// |
782 | /// This function is safe, but actually using the return value is unsafe. |
783 | /// See the documentation of [`slice::from_raw_parts`] for slice safety requirements. |
784 | /// |
785 | /// [`slice::from_raw_parts`]: crate::slice::from_raw_parts |
786 | /// |
787 | /// # Examples |
788 | /// |
789 | /// ```rust |
790 | /// use std::ptr; |
791 | /// |
792 | /// // create a slice pointer when starting out with a pointer to the first element |
793 | /// let x = [5, 6, 7]; |
794 | /// let raw_pointer = x.as_ptr(); |
795 | /// let slice = ptr::slice_from_raw_parts(raw_pointer, 3); |
796 | /// assert_eq!(unsafe { &*slice }[2], 7); |
797 | /// ``` |
798 | /// |
799 | /// You must ensure that the pointer is valid and not null before dereferencing |
800 | /// the raw slice. A slice reference must never have a null pointer, even if it's empty. |
801 | /// |
802 | /// ```rust,should_panic |
803 | /// use std::ptr; |
804 | /// let danger: *const [u8] = ptr::slice_from_raw_parts(ptr::null(), 0); |
805 | /// unsafe { |
806 | /// danger.as_ref().expect("references must not be null" ); |
807 | /// } |
808 | /// ``` |
809 | #[inline ] |
810 | #[stable (feature = "slice_from_raw_parts" , since = "1.42.0" )] |
811 | #[rustc_const_stable (feature = "const_slice_from_raw_parts" , since = "1.64.0" )] |
812 | #[rustc_allow_const_fn_unstable (ptr_metadata)] |
813 | #[rustc_diagnostic_item = "ptr_slice_from_raw_parts" ] |
814 | pub const fn slice_from_raw_parts<T>(data: *const T, len: usize) -> *const [T] { |
815 | from_raw_parts(data_pointer:data.cast(), metadata:len) |
816 | } |
817 | |
818 | /// Forms a raw mutable slice from a pointer and a length. |
819 | /// |
820 | /// The `len` argument is the number of **elements**, not the number of bytes. |
821 | /// |
822 | /// Performs the same functionality as [`slice_from_raw_parts`], except that a |
823 | /// raw mutable slice is returned, as opposed to a raw immutable slice. |
824 | /// |
825 | /// This function is safe, but actually using the return value is unsafe. |
826 | /// See the documentation of [`slice::from_raw_parts_mut`] for slice safety requirements. |
827 | /// |
828 | /// [`slice::from_raw_parts_mut`]: crate::slice::from_raw_parts_mut |
829 | /// |
830 | /// # Examples |
831 | /// |
832 | /// ```rust |
833 | /// use std::ptr; |
834 | /// |
835 | /// let x = &mut [5, 6, 7]; |
836 | /// let raw_pointer = x.as_mut_ptr(); |
837 | /// let slice = ptr::slice_from_raw_parts_mut(raw_pointer, 3); |
838 | /// |
839 | /// unsafe { |
840 | /// (*slice)[2] = 99; // assign a value at an index in the slice |
841 | /// }; |
842 | /// |
843 | /// assert_eq!(unsafe { &*slice }[2], 99); |
844 | /// ``` |
845 | /// |
846 | /// You must ensure that the pointer is valid and not null before dereferencing |
847 | /// the raw slice. A slice reference must never have a null pointer, even if it's empty. |
848 | /// |
849 | /// ```rust,should_panic |
850 | /// use std::ptr; |
851 | /// let danger: *mut [u8] = ptr::slice_from_raw_parts_mut(ptr::null_mut(), 0); |
852 | /// unsafe { |
853 | /// danger.as_mut().expect("references must not be null" ); |
854 | /// } |
855 | /// ``` |
856 | #[inline ] |
857 | #[stable (feature = "slice_from_raw_parts" , since = "1.42.0" )] |
858 | #[rustc_const_unstable (feature = "const_slice_from_raw_parts_mut" , issue = "67456" )] |
859 | #[rustc_diagnostic_item = "ptr_slice_from_raw_parts_mut" ] |
860 | pub const fn slice_from_raw_parts_mut<T>(data: *mut T, len: usize) -> *mut [T] { |
861 | from_raw_parts_mut(data_pointer:data.cast(), metadata:len) |
862 | } |
863 | |
864 | /// Swaps the values at two mutable locations of the same type, without |
865 | /// deinitializing either. |
866 | /// |
867 | /// But for the following exceptions, this function is semantically |
868 | /// equivalent to [`mem::swap`]: |
869 | /// |
870 | /// * It operates on raw pointers instead of references. When references are |
871 | /// available, [`mem::swap`] should be preferred. |
872 | /// |
873 | /// * The two pointed-to values may overlap. If the values do overlap, then the |
874 | /// overlapping region of memory from `x` will be used. This is demonstrated |
875 | /// in the second example below. |
876 | /// |
877 | /// * The operation is "untyped" in the sense that data may be uninitialized or otherwise violate |
878 | /// the requirements of `T`. The initialization state is preserved exactly. |
879 | /// |
880 | /// # Safety |
881 | /// |
882 | /// Behavior is undefined if any of the following conditions are violated: |
883 | /// |
884 | /// * Both `x` and `y` must be [valid] for both reads and writes. They must remain valid even when the |
885 | /// other pointer is written. (This means if the memory ranges overlap, the two pointers must not |
886 | /// be subject to aliasing restrictions relative to each other.) |
887 | /// |
888 | /// * Both `x` and `y` must be properly aligned. |
889 | /// |
890 | /// Note that even if `T` has size `0`, the pointers must be non-null and properly aligned. |
891 | /// |
892 | /// [valid]: self#safety |
893 | /// |
894 | /// # Examples |
895 | /// |
896 | /// Swapping two non-overlapping regions: |
897 | /// |
898 | /// ``` |
899 | /// use std::ptr; |
900 | /// |
901 | /// let mut array = [0, 1, 2, 3]; |
902 | /// |
903 | /// let (x, y) = array.split_at_mut(2); |
904 | /// let x = x.as_mut_ptr().cast::<[u32; 2]>(); // this is `array[0..2]` |
905 | /// let y = y.as_mut_ptr().cast::<[u32; 2]>(); // this is `array[2..4]` |
906 | /// |
907 | /// unsafe { |
908 | /// ptr::swap(x, y); |
909 | /// assert_eq!([2, 3, 0, 1], array); |
910 | /// } |
911 | /// ``` |
912 | /// |
913 | /// Swapping two overlapping regions: |
914 | /// |
915 | /// ``` |
916 | /// use std::ptr; |
917 | /// |
918 | /// let mut array: [i32; 4] = [0, 1, 2, 3]; |
919 | /// |
920 | /// let array_ptr: *mut i32 = array.as_mut_ptr(); |
921 | /// |
922 | /// let x = array_ptr as *mut [i32; 3]; // this is `array[0..3]` |
923 | /// let y = unsafe { array_ptr.add(1) } as *mut [i32; 3]; // this is `array[1..4]` |
924 | /// |
925 | /// unsafe { |
926 | /// ptr::swap(x, y); |
927 | /// // The indices `1..3` of the slice overlap between `x` and `y`. |
928 | /// // Reasonable results would be for to them be `[2, 3]`, so that indices `0..3` are |
929 | /// // `[1, 2, 3]` (matching `y` before the `swap`); or for them to be `[0, 1]` |
930 | /// // so that indices `1..4` are `[0, 1, 2]` (matching `x` before the `swap`). |
931 | /// // This implementation is defined to make the latter choice. |
932 | /// assert_eq!([1, 0, 1, 2], array); |
933 | /// } |
934 | /// ``` |
935 | #[inline ] |
936 | #[stable (feature = "rust1" , since = "1.0.0" )] |
937 | #[rustc_const_unstable (feature = "const_swap" , issue = "83163" )] |
938 | #[rustc_diagnostic_item = "ptr_swap" ] |
939 | pub const unsafe fn swap<T>(x: *mut T, y: *mut T) { |
940 | // Give ourselves some scratch space to work with. |
941 | // We do not have to worry about drops: `MaybeUninit` does nothing when dropped. |
942 | let mut tmp: MaybeUninit = MaybeUninit::<T>::uninit(); |
943 | |
944 | // Perform the swap |
945 | // SAFETY: the caller must guarantee that `x` and `y` are |
946 | // valid for writes and properly aligned. `tmp` cannot be |
947 | // overlapping either `x` or `y` because `tmp` was just allocated |
948 | // on the stack as a separate allocated object. |
949 | unsafe { |
950 | copy_nonoverlapping(src:x, dst:tmp.as_mut_ptr(), count:1); |
951 | copy(src:y, dst:x, count:1); // `x` and `y` may overlap |
952 | copy_nonoverlapping(src:tmp.as_ptr(), dst:y, count:1); |
953 | } |
954 | } |
955 | |
956 | /// Swaps `count * size_of::<T>()` bytes between the two regions of memory |
957 | /// beginning at `x` and `y`. The two regions must *not* overlap. |
958 | /// |
959 | /// The operation is "untyped" in the sense that data may be uninitialized or otherwise violate the |
960 | /// requirements of `T`. The initialization state is preserved exactly. |
961 | /// |
962 | /// # Safety |
963 | /// |
964 | /// Behavior is undefined if any of the following conditions are violated: |
965 | /// |
966 | /// * Both `x` and `y` must be [valid] for both reads and writes of `count * |
967 | /// size_of::<T>()` bytes. |
968 | /// |
969 | /// * Both `x` and `y` must be properly aligned. |
970 | /// |
971 | /// * The region of memory beginning at `x` with a size of `count * |
972 | /// size_of::<T>()` bytes must *not* overlap with the region of memory |
973 | /// beginning at `y` with the same size. |
974 | /// |
975 | /// Note that even if the effectively copied size (`count * size_of::<T>()`) is `0`, |
976 | /// the pointers must be non-null and properly aligned. |
977 | /// |
978 | /// [valid]: self#safety |
979 | /// |
980 | /// # Examples |
981 | /// |
982 | /// Basic usage: |
983 | /// |
984 | /// ``` |
985 | /// use std::ptr; |
986 | /// |
987 | /// let mut x = [1, 2, 3, 4]; |
988 | /// let mut y = [7, 8, 9]; |
989 | /// |
990 | /// unsafe { |
991 | /// ptr::swap_nonoverlapping(x.as_mut_ptr(), y.as_mut_ptr(), 2); |
992 | /// } |
993 | /// |
994 | /// assert_eq!(x, [7, 8, 3, 4]); |
995 | /// assert_eq!(y, [1, 2, 9]); |
996 | /// ``` |
997 | #[inline ] |
998 | #[stable (feature = "swap_nonoverlapping" , since = "1.27.0" )] |
999 | #[rustc_const_unstable (feature = "const_swap" , issue = "83163" )] |
1000 | #[rustc_diagnostic_item = "ptr_swap_nonoverlapping" ] |
1001 | pub const unsafe fn swap_nonoverlapping<T>(x: *mut T, y: *mut T, count: usize) { |
1002 | #[allow (unused)] |
1003 | macro_rules! attempt_swap_as_chunks { |
1004 | ($ChunkTy:ty) => { |
1005 | if mem::align_of::<T>() >= mem::align_of::<$ChunkTy>() |
1006 | && mem::size_of::<T>() % mem::size_of::<$ChunkTy>() == 0 |
1007 | { |
1008 | let x: *mut $ChunkTy = x.cast(); |
1009 | let y: *mut $ChunkTy = y.cast(); |
1010 | let count = count * (mem::size_of::<T>() / mem::size_of::<$ChunkTy>()); |
1011 | // SAFETY: these are the same bytes that the caller promised were |
1012 | // ok, just typed as `MaybeUninit<ChunkTy>`s instead of as `T`s. |
1013 | // The `if` condition above ensures that we're not violating |
1014 | // alignment requirements, and that the division is exact so |
1015 | // that we don't lose any bytes off the end. |
1016 | return unsafe { swap_nonoverlapping_simple_untyped(x, y, count) }; |
1017 | } |
1018 | }; |
1019 | } |
1020 | |
1021 | ub_checks::assert_unsafe_precondition!( |
1022 | check_language_ub, |
1023 | "ptr::swap_nonoverlapping requires that both pointer arguments are aligned and non-null \ |
1024 | and the specified memory ranges do not overlap" , |
1025 | ( |
1026 | x: *mut () = x as *mut (), |
1027 | y: *mut () = y as *mut (), |
1028 | size: usize = size_of::<T>(), |
1029 | align: usize = align_of::<T>(), |
1030 | count: usize = count, |
1031 | ) => |
1032 | ub_checks::is_aligned_and_not_null(x, align) |
1033 | && ub_checks::is_aligned_and_not_null(y, align) |
1034 | && ub_checks::is_nonoverlapping(x, y, size, count) |
1035 | ); |
1036 | |
1037 | // Split up the slice into small power-of-two-sized chunks that LLVM is able |
1038 | // to vectorize (unless it's a special type with more-than-pointer alignment, |
1039 | // because we don't want to pessimize things like slices of SIMD vectors.) |
1040 | if mem::align_of::<T>() <= mem::size_of::<usize>() |
1041 | && (!mem::size_of::<T>().is_power_of_two() |
1042 | || mem::size_of::<T>() > mem::size_of::<usize>() * 2) |
1043 | { |
1044 | attempt_swap_as_chunks!(usize); |
1045 | attempt_swap_as_chunks!(u8); |
1046 | } |
1047 | |
1048 | // SAFETY: Same preconditions as this function |
1049 | unsafe { swap_nonoverlapping_simple_untyped(x, y, count) } |
1050 | } |
1051 | |
1052 | /// Same behaviour and safety conditions as [`swap_nonoverlapping`] |
1053 | /// |
1054 | /// LLVM can vectorize this (at least it can for the power-of-two-sized types |
1055 | /// `swap_nonoverlapping` tries to use) so no need to manually SIMD it. |
1056 | #[inline ] |
1057 | #[rustc_const_unstable (feature = "const_swap" , issue = "83163" )] |
1058 | const unsafe fn swap_nonoverlapping_simple_untyped<T>(x: *mut T, y: *mut T, count: usize) { |
1059 | let x = x.cast::<MaybeUninit<T>>(); |
1060 | let y = y.cast::<MaybeUninit<T>>(); |
1061 | let mut i = 0; |
1062 | while i < count { |
1063 | // SAFETY: By precondition, `i` is in-bounds because it's below `n` |
1064 | let x = unsafe { x.add(i) }; |
1065 | // SAFETY: By precondition, `i` is in-bounds because it's below `n` |
1066 | // and it's distinct from `x` since the ranges are non-overlapping |
1067 | let y = unsafe { y.add(i) }; |
1068 | |
1069 | // If we end up here, it's because we're using a simple type -- like |
1070 | // a small power-of-two-sized thing -- or a special type with particularly |
1071 | // large alignment, particularly SIMD types. |
1072 | // Thus we're fine just reading-and-writing it, as either it's small |
1073 | // and that works well anyway or it's special and the type's author |
1074 | // presumably wanted things to be done in the larger chunk. |
1075 | |
1076 | // SAFETY: we're only ever given pointers that are valid to read/write, |
1077 | // including being aligned, and nothing here panics so it's drop-safe. |
1078 | unsafe { |
1079 | let a: MaybeUninit<T> = read(x); |
1080 | let b: MaybeUninit<T> = read(y); |
1081 | write(x, b); |
1082 | write(y, a); |
1083 | } |
1084 | |
1085 | i += 1; |
1086 | } |
1087 | } |
1088 | |
1089 | /// Moves `src` into the pointed `dst`, returning the previous `dst` value. |
1090 | /// |
1091 | /// Neither value is dropped. |
1092 | /// |
1093 | /// This function is semantically equivalent to [`mem::replace`] except that it |
1094 | /// operates on raw pointers instead of references. When references are |
1095 | /// available, [`mem::replace`] should be preferred. |
1096 | /// |
1097 | /// # Safety |
1098 | /// |
1099 | /// Behavior is undefined if any of the following conditions are violated: |
1100 | /// |
1101 | /// * `dst` must be [valid] for both reads and writes. |
1102 | /// |
1103 | /// * `dst` must be properly aligned. |
1104 | /// |
1105 | /// * `dst` must point to a properly initialized value of type `T`. |
1106 | /// |
1107 | /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned. |
1108 | /// |
1109 | /// [valid]: self#safety |
1110 | /// |
1111 | /// # Examples |
1112 | /// |
1113 | /// ``` |
1114 | /// use std::ptr; |
1115 | /// |
1116 | /// let mut rust = vec!['b' , 'u' , 's' , 't' ]; |
1117 | /// |
1118 | /// // `mem::replace` would have the same effect without requiring the unsafe |
1119 | /// // block. |
1120 | /// let b = unsafe { |
1121 | /// ptr::replace(&mut rust[0], 'r' ) |
1122 | /// }; |
1123 | /// |
1124 | /// assert_eq!(b, 'b' ); |
1125 | /// assert_eq!(rust, &['r' , 'u' , 's' , 't' ]); |
1126 | /// ``` |
1127 | #[inline ] |
1128 | #[stable (feature = "rust1" , since = "1.0.0" )] |
1129 | #[rustc_const_unstable (feature = "const_replace" , issue = "83164" )] |
1130 | #[rustc_diagnostic_item = "ptr_replace" ] |
1131 | pub const unsafe fn replace<T>(dst: *mut T, src: T) -> T { |
1132 | // SAFETY: the caller must guarantee that `dst` is valid to be |
1133 | // cast to a mutable reference (valid for writes, aligned, initialized), |
1134 | // and cannot overlap `src` since `dst` must point to a distinct |
1135 | // allocated object. |
1136 | unsafe { |
1137 | ub_checks::assert_unsafe_precondition!( |
1138 | check_language_ub, |
1139 | "ptr::replace requires that the pointer argument is aligned and non-null" , |
1140 | ( |
1141 | addr: *const () = dst as *const (), |
1142 | align: usize = align_of::<T>(), |
1143 | ) => ub_checks::is_aligned_and_not_null(addr, align) |
1144 | ); |
1145 | mem::replace(&mut *dst, src) |
1146 | } |
1147 | } |
1148 | |
1149 | /// Reads the value from `src` without moving it. This leaves the |
1150 | /// memory in `src` unchanged. |
1151 | /// |
1152 | /// # Safety |
1153 | /// |
1154 | /// Behavior is undefined if any of the following conditions are violated: |
1155 | /// |
1156 | /// * `src` must be [valid] for reads. |
1157 | /// |
1158 | /// * `src` must be properly aligned. Use [`read_unaligned`] if this is not the |
1159 | /// case. |
1160 | /// |
1161 | /// * `src` must point to a properly initialized value of type `T`. |
1162 | /// |
1163 | /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned. |
1164 | /// |
1165 | /// # Examples |
1166 | /// |
1167 | /// Basic usage: |
1168 | /// |
1169 | /// ``` |
1170 | /// let x = 12; |
1171 | /// let y = &x as *const i32; |
1172 | /// |
1173 | /// unsafe { |
1174 | /// assert_eq!(std::ptr::read(y), 12); |
1175 | /// } |
1176 | /// ``` |
1177 | /// |
1178 | /// Manually implement [`mem::swap`]: |
1179 | /// |
1180 | /// ``` |
1181 | /// use std::ptr; |
1182 | /// |
1183 | /// fn swap<T>(a: &mut T, b: &mut T) { |
1184 | /// unsafe { |
1185 | /// // Create a bitwise copy of the value at `a` in `tmp`. |
1186 | /// let tmp = ptr::read(a); |
1187 | /// |
1188 | /// // Exiting at this point (either by explicitly returning or by |
1189 | /// // calling a function which panics) would cause the value in `tmp` to |
1190 | /// // be dropped while the same value is still referenced by `a`. This |
1191 | /// // could trigger undefined behavior if `T` is not `Copy`. |
1192 | /// |
1193 | /// // Create a bitwise copy of the value at `b` in `a`. |
1194 | /// // This is safe because mutable references cannot alias. |
1195 | /// ptr::copy_nonoverlapping(b, a, 1); |
1196 | /// |
1197 | /// // As above, exiting here could trigger undefined behavior because |
1198 | /// // the same value is referenced by `a` and `b`. |
1199 | /// |
1200 | /// // Move `tmp` into `b`. |
1201 | /// ptr::write(b, tmp); |
1202 | /// |
1203 | /// // `tmp` has been moved (`write` takes ownership of its second argument), |
1204 | /// // so nothing is dropped implicitly here. |
1205 | /// } |
1206 | /// } |
1207 | /// |
1208 | /// let mut foo = "foo" .to_owned(); |
1209 | /// let mut bar = "bar" .to_owned(); |
1210 | /// |
1211 | /// swap(&mut foo, &mut bar); |
1212 | /// |
1213 | /// assert_eq!(foo, "bar" ); |
1214 | /// assert_eq!(bar, "foo" ); |
1215 | /// ``` |
1216 | /// |
1217 | /// ## Ownership of the Returned Value |
1218 | /// |
1219 | /// `read` creates a bitwise copy of `T`, regardless of whether `T` is [`Copy`]. |
1220 | /// If `T` is not [`Copy`], using both the returned value and the value at |
1221 | /// `*src` can violate memory safety. Note that assigning to `*src` counts as a |
1222 | /// use because it will attempt to drop the value at `*src`. |
1223 | /// |
1224 | /// [`write()`] can be used to overwrite data without causing it to be dropped. |
1225 | /// |
1226 | /// ``` |
1227 | /// use std::ptr; |
1228 | /// |
1229 | /// let mut s = String::from("foo" ); |
1230 | /// unsafe { |
1231 | /// // `s2` now points to the same underlying memory as `s`. |
1232 | /// let mut s2: String = ptr::read(&s); |
1233 | /// |
1234 | /// assert_eq!(s2, "foo" ); |
1235 | /// |
1236 | /// // Assigning to `s2` causes its original value to be dropped. Beyond |
1237 | /// // this point, `s` must no longer be used, as the underlying memory has |
1238 | /// // been freed. |
1239 | /// s2 = String::default(); |
1240 | /// assert_eq!(s2, "" ); |
1241 | /// |
1242 | /// // Assigning to `s` would cause the old value to be dropped again, |
1243 | /// // resulting in undefined behavior. |
1244 | /// // s = String::from("bar"); // ERROR |
1245 | /// |
1246 | /// // `ptr::write` can be used to overwrite a value without dropping it. |
1247 | /// ptr::write(&mut s, String::from("bar" )); |
1248 | /// } |
1249 | /// |
1250 | /// assert_eq!(s, "bar" ); |
1251 | /// ``` |
1252 | /// |
1253 | /// [valid]: self#safety |
1254 | #[inline ] |
1255 | #[stable (feature = "rust1" , since = "1.0.0" )] |
1256 | #[rustc_const_stable (feature = "const_ptr_read" , since = "1.71.0" )] |
1257 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
1258 | #[rustc_diagnostic_item = "ptr_read" ] |
1259 | pub const unsafe fn read<T>(src: *const T) -> T { |
1260 | // It would be semantically correct to implement this via `copy_nonoverlapping` |
1261 | // and `MaybeUninit`, as was done before PR #109035. Calling `assume_init` |
1262 | // provides enough information to know that this is a typed operation. |
1263 | |
1264 | // However, as of March 2023 the compiler was not capable of taking advantage |
1265 | // of that information. Thus the implementation here switched to an intrinsic, |
1266 | // which lowers to `_0 = *src` in MIR, to address a few issues: |
1267 | // |
1268 | // - Using `MaybeUninit::assume_init` after a `copy_nonoverlapping` was not |
1269 | // turning the untyped copy into a typed load. As such, the generated |
1270 | // `load` in LLVM didn't get various metadata, such as `!range` (#73258), |
1271 | // `!nonnull`, and `!noundef`, resulting in poorer optimization. |
1272 | // - Going through the extra local resulted in multiple extra copies, even |
1273 | // in optimized MIR. (Ignoring StorageLive/Dead, the intrinsic is one |
1274 | // MIR statement, while the previous implementation was eight.) LLVM |
1275 | // could sometimes optimize them away, but because `read` is at the core |
1276 | // of so many things, not having them in the first place improves what we |
1277 | // hand off to the backend. For example, `mem::replace::<Big>` previously |
1278 | // emitted 4 `alloca` and 6 `memcpy`s, but is now 1 `alloc` and 3 `memcpy`s. |
1279 | // - In general, this approach keeps us from getting any more bugs (like |
1280 | // #106369) that boil down to "`read(p)` is worse than `*p`", as this |
1281 | // makes them look identical to the backend (or other MIR consumers). |
1282 | // |
1283 | // Future enhancements to MIR optimizations might well allow this to return |
1284 | // to the previous implementation, rather than using an intrinsic. |
1285 | |
1286 | // SAFETY: the caller must guarantee that `src` is valid for reads. |
1287 | unsafe { |
1288 | #[cfg (debug_assertions)] // Too expensive to always enable (for now?) |
1289 | ub_checks::assert_unsafe_precondition!( |
1290 | check_language_ub, |
1291 | "ptr::read requires that the pointer argument is aligned and non-null" , |
1292 | ( |
1293 | addr: *const () = src as *const (), |
1294 | align: usize = align_of::<T>(), |
1295 | ) => ub_checks::is_aligned_and_not_null(addr, align) |
1296 | ); |
1297 | crate::intrinsics::read_via_copy(src) |
1298 | } |
1299 | } |
1300 | |
1301 | /// Reads the value from `src` without moving it. This leaves the |
1302 | /// memory in `src` unchanged. |
1303 | /// |
1304 | /// Unlike [`read`], `read_unaligned` works with unaligned pointers. |
1305 | /// |
1306 | /// # Safety |
1307 | /// |
1308 | /// Behavior is undefined if any of the following conditions are violated: |
1309 | /// |
1310 | /// * `src` must be [valid] for reads. |
1311 | /// |
1312 | /// * `src` must point to a properly initialized value of type `T`. |
1313 | /// |
1314 | /// Like [`read`], `read_unaligned` creates a bitwise copy of `T`, regardless of |
1315 | /// whether `T` is [`Copy`]. If `T` is not [`Copy`], using both the returned |
1316 | /// value and the value at `*src` can [violate memory safety][read-ownership]. |
1317 | /// |
1318 | /// Note that even if `T` has size `0`, the pointer must be non-null. |
1319 | /// |
1320 | /// [read-ownership]: read#ownership-of-the-returned-value |
1321 | /// [valid]: self#safety |
1322 | /// |
1323 | /// ## On `packed` structs |
1324 | /// |
1325 | /// Attempting to create a raw pointer to an `unaligned` struct field with |
1326 | /// an expression such as `&packed.unaligned as *const FieldType` creates an |
1327 | /// intermediate unaligned reference before converting that to a raw pointer. |
1328 | /// That this reference is temporary and immediately cast is inconsequential |
1329 | /// as the compiler always expects references to be properly aligned. |
1330 | /// As a result, using `&packed.unaligned as *const FieldType` causes immediate |
1331 | /// *undefined behavior* in your program. |
1332 | /// |
1333 | /// Instead you must use the [`ptr::addr_of!`](addr_of) macro to |
1334 | /// create the pointer. You may use that returned pointer together with this |
1335 | /// function. |
1336 | /// |
1337 | /// An example of what not to do and how this relates to `read_unaligned` is: |
1338 | /// |
1339 | /// ``` |
1340 | /// #[repr(packed, C)] |
1341 | /// struct Packed { |
1342 | /// _padding: u8, |
1343 | /// unaligned: u32, |
1344 | /// } |
1345 | /// |
1346 | /// let packed = Packed { |
1347 | /// _padding: 0x00, |
1348 | /// unaligned: 0x01020304, |
1349 | /// }; |
1350 | /// |
1351 | /// // Take the address of a 32-bit integer which is not aligned. |
1352 | /// // In contrast to `&packed.unaligned as *const _`, this has no undefined behavior. |
1353 | /// let unaligned = std::ptr::addr_of!(packed.unaligned); |
1354 | /// |
1355 | /// let v = unsafe { std::ptr::read_unaligned(unaligned) }; |
1356 | /// assert_eq!(v, 0x01020304); |
1357 | /// ``` |
1358 | /// |
1359 | /// Accessing unaligned fields directly with e.g. `packed.unaligned` is safe however. |
1360 | /// |
1361 | /// # Examples |
1362 | /// |
1363 | /// Read a usize value from a byte buffer: |
1364 | /// |
1365 | /// ``` |
1366 | /// use std::mem; |
1367 | /// |
1368 | /// fn read_usize(x: &[u8]) -> usize { |
1369 | /// assert!(x.len() >= mem::size_of::<usize>()); |
1370 | /// |
1371 | /// let ptr = x.as_ptr() as *const usize; |
1372 | /// |
1373 | /// unsafe { ptr.read_unaligned() } |
1374 | /// } |
1375 | /// ``` |
1376 | #[inline ] |
1377 | #[stable (feature = "ptr_unaligned" , since = "1.17.0" )] |
1378 | #[rustc_const_stable (feature = "const_ptr_read" , since = "1.71.0" )] |
1379 | #[rustc_allow_const_fn_unstable ( |
1380 | const_mut_refs, |
1381 | const_maybe_uninit_as_mut_ptr, |
1382 | const_intrinsic_copy |
1383 | )] |
1384 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
1385 | #[rustc_diagnostic_item = "ptr_read_unaligned" ] |
1386 | pub const unsafe fn read_unaligned<T>(src: *const T) -> T { |
1387 | let mut tmp: MaybeUninit = MaybeUninit::<T>::uninit(); |
1388 | // SAFETY: the caller must guarantee that `src` is valid for reads. |
1389 | // `src` cannot overlap `tmp` because `tmp` was just allocated on |
1390 | // the stack as a separate allocated object. |
1391 | // |
1392 | // Also, since we just wrote a valid value into `tmp`, it is guaranteed |
1393 | // to be properly initialized. |
1394 | unsafe { |
1395 | copy_nonoverlapping(src as *const u8, dst:tmp.as_mut_ptr() as *mut u8, count:mem::size_of::<T>()); |
1396 | tmp.assume_init() |
1397 | } |
1398 | } |
1399 | |
1400 | /// Overwrites a memory location with the given value without reading or |
1401 | /// dropping the old value. |
1402 | /// |
1403 | /// `write` does not drop the contents of `dst`. This is safe, but it could leak |
1404 | /// allocations or resources, so care should be taken not to overwrite an object |
1405 | /// that should be dropped. |
1406 | /// |
1407 | /// Additionally, it does not drop `src`. Semantically, `src` is moved into the |
1408 | /// location pointed to by `dst`. |
1409 | /// |
1410 | /// This is appropriate for initializing uninitialized memory, or overwriting |
1411 | /// memory that has previously been [`read`] from. |
1412 | /// |
1413 | /// # Safety |
1414 | /// |
1415 | /// Behavior is undefined if any of the following conditions are violated: |
1416 | /// |
1417 | /// * `dst` must be [valid] for writes. |
1418 | /// |
1419 | /// * `dst` must be properly aligned. Use [`write_unaligned`] if this is not the |
1420 | /// case. |
1421 | /// |
1422 | /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned. |
1423 | /// |
1424 | /// [valid]: self#safety |
1425 | /// |
1426 | /// # Examples |
1427 | /// |
1428 | /// Basic usage: |
1429 | /// |
1430 | /// ``` |
1431 | /// let mut x = 0; |
1432 | /// let y = &mut x as *mut i32; |
1433 | /// let z = 12; |
1434 | /// |
1435 | /// unsafe { |
1436 | /// std::ptr::write(y, z); |
1437 | /// assert_eq!(std::ptr::read(y), 12); |
1438 | /// } |
1439 | /// ``` |
1440 | /// |
1441 | /// Manually implement [`mem::swap`]: |
1442 | /// |
1443 | /// ``` |
1444 | /// use std::ptr; |
1445 | /// |
1446 | /// fn swap<T>(a: &mut T, b: &mut T) { |
1447 | /// unsafe { |
1448 | /// // Create a bitwise copy of the value at `a` in `tmp`. |
1449 | /// let tmp = ptr::read(a); |
1450 | /// |
1451 | /// // Exiting at this point (either by explicitly returning or by |
1452 | /// // calling a function which panics) would cause the value in `tmp` to |
1453 | /// // be dropped while the same value is still referenced by `a`. This |
1454 | /// // could trigger undefined behavior if `T` is not `Copy`. |
1455 | /// |
1456 | /// // Create a bitwise copy of the value at `b` in `a`. |
1457 | /// // This is safe because mutable references cannot alias. |
1458 | /// ptr::copy_nonoverlapping(b, a, 1); |
1459 | /// |
1460 | /// // As above, exiting here could trigger undefined behavior because |
1461 | /// // the same value is referenced by `a` and `b`. |
1462 | /// |
1463 | /// // Move `tmp` into `b`. |
1464 | /// ptr::write(b, tmp); |
1465 | /// |
1466 | /// // `tmp` has been moved (`write` takes ownership of its second argument), |
1467 | /// // so nothing is dropped implicitly here. |
1468 | /// } |
1469 | /// } |
1470 | /// |
1471 | /// let mut foo = "foo" .to_owned(); |
1472 | /// let mut bar = "bar" .to_owned(); |
1473 | /// |
1474 | /// swap(&mut foo, &mut bar); |
1475 | /// |
1476 | /// assert_eq!(foo, "bar" ); |
1477 | /// assert_eq!(bar, "foo" ); |
1478 | /// ``` |
1479 | #[inline ] |
1480 | #[stable (feature = "rust1" , since = "1.0.0" )] |
1481 | #[rustc_const_unstable (feature = "const_ptr_write" , issue = "86302" )] |
1482 | #[rustc_diagnostic_item = "ptr_write" ] |
1483 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
1484 | pub const unsafe fn write<T>(dst: *mut T, src: T) { |
1485 | // Semantically, it would be fine for this to be implemented as a |
1486 | // `copy_nonoverlapping` and appropriate drop suppression of `src`. |
1487 | |
1488 | // However, implementing via that currently produces more MIR than is ideal. |
1489 | // Using an intrinsic keeps it down to just the simple `*dst = move src` in |
1490 | // MIR (11 statements shorter, at the time of writing), and also allows |
1491 | // `src` to stay an SSA value in codegen_ssa, rather than a memory one. |
1492 | |
1493 | // SAFETY: the caller must guarantee that `dst` is valid for writes. |
1494 | // `dst` cannot overlap `src` because the caller has mutable access |
1495 | // to `dst` while `src` is owned by this function. |
1496 | unsafe { |
1497 | #[cfg (debug_assertions)] // Too expensive to always enable (for now?) |
1498 | ub_checks::assert_unsafe_precondition!( |
1499 | check_language_ub, |
1500 | "ptr::write requires that the pointer argument is aligned and non-null" , |
1501 | ( |
1502 | addr: *mut () = dst as *mut (), |
1503 | align: usize = align_of::<T>(), |
1504 | ) => ub_checks::is_aligned_and_not_null(addr, align) |
1505 | ); |
1506 | intrinsics::write_via_move(dst, src) |
1507 | } |
1508 | } |
1509 | |
1510 | /// Overwrites a memory location with the given value without reading or |
1511 | /// dropping the old value. |
1512 | /// |
1513 | /// Unlike [`write()`], the pointer may be unaligned. |
1514 | /// |
1515 | /// `write_unaligned` does not drop the contents of `dst`. This is safe, but it |
1516 | /// could leak allocations or resources, so care should be taken not to overwrite |
1517 | /// an object that should be dropped. |
1518 | /// |
1519 | /// Additionally, it does not drop `src`. Semantically, `src` is moved into the |
1520 | /// location pointed to by `dst`. |
1521 | /// |
1522 | /// This is appropriate for initializing uninitialized memory, or overwriting |
1523 | /// memory that has previously been read with [`read_unaligned`]. |
1524 | /// |
1525 | /// # Safety |
1526 | /// |
1527 | /// Behavior is undefined if any of the following conditions are violated: |
1528 | /// |
1529 | /// * `dst` must be [valid] for writes. |
1530 | /// |
1531 | /// Note that even if `T` has size `0`, the pointer must be non-null. |
1532 | /// |
1533 | /// [valid]: self#safety |
1534 | /// |
1535 | /// ## On `packed` structs |
1536 | /// |
1537 | /// Attempting to create a raw pointer to an `unaligned` struct field with |
1538 | /// an expression such as `&packed.unaligned as *const FieldType` creates an |
1539 | /// intermediate unaligned reference before converting that to a raw pointer. |
1540 | /// That this reference is temporary and immediately cast is inconsequential |
1541 | /// as the compiler always expects references to be properly aligned. |
1542 | /// As a result, using `&packed.unaligned as *const FieldType` causes immediate |
1543 | /// *undefined behavior* in your program. |
1544 | /// |
1545 | /// Instead you must use the [`ptr::addr_of_mut!`](addr_of_mut) |
1546 | /// macro to create the pointer. You may use that returned pointer together with |
1547 | /// this function. |
1548 | /// |
1549 | /// An example of how to do it and how this relates to `write_unaligned` is: |
1550 | /// |
1551 | /// ``` |
1552 | /// #[repr(packed, C)] |
1553 | /// struct Packed { |
1554 | /// _padding: u8, |
1555 | /// unaligned: u32, |
1556 | /// } |
1557 | /// |
1558 | /// let mut packed: Packed = unsafe { std::mem::zeroed() }; |
1559 | /// |
1560 | /// // Take the address of a 32-bit integer which is not aligned. |
1561 | /// // In contrast to `&packed.unaligned as *mut _`, this has no undefined behavior. |
1562 | /// let unaligned = std::ptr::addr_of_mut!(packed.unaligned); |
1563 | /// |
1564 | /// unsafe { std::ptr::write_unaligned(unaligned, 42) }; |
1565 | /// |
1566 | /// assert_eq!({packed.unaligned}, 42); // `{...}` forces copying the field instead of creating a reference. |
1567 | /// ``` |
1568 | /// |
1569 | /// Accessing unaligned fields directly with e.g. `packed.unaligned` is safe however |
1570 | /// (as can be seen in the `assert_eq!` above). |
1571 | /// |
1572 | /// # Examples |
1573 | /// |
1574 | /// Write a usize value to a byte buffer: |
1575 | /// |
1576 | /// ``` |
1577 | /// use std::mem; |
1578 | /// |
1579 | /// fn write_usize(x: &mut [u8], val: usize) { |
1580 | /// assert!(x.len() >= mem::size_of::<usize>()); |
1581 | /// |
1582 | /// let ptr = x.as_mut_ptr() as *mut usize; |
1583 | /// |
1584 | /// unsafe { ptr.write_unaligned(val) } |
1585 | /// } |
1586 | /// ``` |
1587 | #[inline ] |
1588 | #[stable (feature = "ptr_unaligned" , since = "1.17.0" )] |
1589 | #[rustc_const_unstable (feature = "const_ptr_write" , issue = "86302" )] |
1590 | #[rustc_diagnostic_item = "ptr_write_unaligned" ] |
1591 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
1592 | pub const unsafe fn write_unaligned<T>(dst: *mut T, src: T) { |
1593 | // SAFETY: the caller must guarantee that `dst` is valid for writes. |
1594 | // `dst` cannot overlap `src` because the caller has mutable access |
1595 | // to `dst` while `src` is owned by this function. |
1596 | unsafe { |
1597 | copy_nonoverlapping(src:addr_of!(src) as *const u8, dst as *mut u8, count:mem::size_of::<T>()); |
1598 | // We are calling the intrinsic directly to avoid function calls in the generated code. |
1599 | intrinsics::forget(src); |
1600 | } |
1601 | } |
1602 | |
1603 | /// Performs a volatile read of the value from `src` without moving it. This |
1604 | /// leaves the memory in `src` unchanged. |
1605 | /// |
1606 | /// Volatile operations are intended to act on I/O memory, and are guaranteed |
1607 | /// to not be elided or reordered by the compiler across other volatile |
1608 | /// operations. |
1609 | /// |
1610 | /// # Notes |
1611 | /// |
1612 | /// Rust does not currently have a rigorously and formally defined memory model, |
1613 | /// so the precise semantics of what "volatile" means here is subject to change |
1614 | /// over time. That being said, the semantics will almost always end up pretty |
1615 | /// similar to [C11's definition of volatile][c11]. |
1616 | /// |
1617 | /// The compiler shouldn't change the relative order or number of volatile |
1618 | /// memory operations. However, volatile memory operations on zero-sized types |
1619 | /// (e.g., if a zero-sized type is passed to `read_volatile`) are noops |
1620 | /// and may be ignored. |
1621 | /// |
1622 | /// [c11]: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf |
1623 | /// |
1624 | /// # Safety |
1625 | /// |
1626 | /// Behavior is undefined if any of the following conditions are violated: |
1627 | /// |
1628 | /// * `src` must be [valid] for reads. |
1629 | /// |
1630 | /// * `src` must be properly aligned. |
1631 | /// |
1632 | /// * `src` must point to a properly initialized value of type `T`. |
1633 | /// |
1634 | /// Like [`read`], `read_volatile` creates a bitwise copy of `T`, regardless of |
1635 | /// whether `T` is [`Copy`]. If `T` is not [`Copy`], using both the returned |
1636 | /// value and the value at `*src` can [violate memory safety][read-ownership]. |
1637 | /// However, storing non-[`Copy`] types in volatile memory is almost certainly |
1638 | /// incorrect. |
1639 | /// |
1640 | /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned. |
1641 | /// |
1642 | /// [valid]: self#safety |
1643 | /// [read-ownership]: read#ownership-of-the-returned-value |
1644 | /// |
1645 | /// Just like in C, whether an operation is volatile has no bearing whatsoever |
1646 | /// on questions involving concurrent access from multiple threads. Volatile |
1647 | /// accesses behave exactly like non-atomic accesses in that regard. In particular, |
1648 | /// a race between a `read_volatile` and any write operation to the same location |
1649 | /// is undefined behavior. |
1650 | /// |
1651 | /// # Examples |
1652 | /// |
1653 | /// Basic usage: |
1654 | /// |
1655 | /// ``` |
1656 | /// let x = 12; |
1657 | /// let y = &x as *const i32; |
1658 | /// |
1659 | /// unsafe { |
1660 | /// assert_eq!(std::ptr::read_volatile(y), 12); |
1661 | /// } |
1662 | /// ``` |
1663 | #[inline ] |
1664 | #[stable (feature = "volatile" , since = "1.9.0" )] |
1665 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
1666 | #[rustc_diagnostic_item = "ptr_read_volatile" ] |
1667 | pub unsafe fn read_volatile<T>(src: *const T) -> T { |
1668 | // SAFETY: the caller must uphold the safety contract for `volatile_load`. |
1669 | unsafe { |
1670 | ub_checks::assert_unsafe_precondition!( |
1671 | check_language_ub, |
1672 | "ptr::read_volatile requires that the pointer argument is aligned and non-null" , |
1673 | ( |
1674 | addr: *const () = src as *const (), |
1675 | align: usize = align_of::<T>(), |
1676 | ) => ub_checks::is_aligned_and_not_null(addr, align) |
1677 | ); |
1678 | intrinsics::volatile_load(src) |
1679 | } |
1680 | } |
1681 | |
1682 | /// Performs a volatile write of a memory location with the given value without |
1683 | /// reading or dropping the old value. |
1684 | /// |
1685 | /// Volatile operations are intended to act on I/O memory, and are guaranteed |
1686 | /// to not be elided or reordered by the compiler across other volatile |
1687 | /// operations. |
1688 | /// |
1689 | /// `write_volatile` does not drop the contents of `dst`. This is safe, but it |
1690 | /// could leak allocations or resources, so care should be taken not to overwrite |
1691 | /// an object that should be dropped. |
1692 | /// |
1693 | /// Additionally, it does not drop `src`. Semantically, `src` is moved into the |
1694 | /// location pointed to by `dst`. |
1695 | /// |
1696 | /// # Notes |
1697 | /// |
1698 | /// Rust does not currently have a rigorously and formally defined memory model, |
1699 | /// so the precise semantics of what "volatile" means here is subject to change |
1700 | /// over time. That being said, the semantics will almost always end up pretty |
1701 | /// similar to [C11's definition of volatile][c11]. |
1702 | /// |
1703 | /// The compiler shouldn't change the relative order or number of volatile |
1704 | /// memory operations. However, volatile memory operations on zero-sized types |
1705 | /// (e.g., if a zero-sized type is passed to `write_volatile`) are noops |
1706 | /// and may be ignored. |
1707 | /// |
1708 | /// [c11]: http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf |
1709 | /// |
1710 | /// # Safety |
1711 | /// |
1712 | /// Behavior is undefined if any of the following conditions are violated: |
1713 | /// |
1714 | /// * `dst` must be [valid] for writes. |
1715 | /// |
1716 | /// * `dst` must be properly aligned. |
1717 | /// |
1718 | /// Note that even if `T` has size `0`, the pointer must be non-null and properly aligned. |
1719 | /// |
1720 | /// [valid]: self#safety |
1721 | /// |
1722 | /// Just like in C, whether an operation is volatile has no bearing whatsoever |
1723 | /// on questions involving concurrent access from multiple threads. Volatile |
1724 | /// accesses behave exactly like non-atomic accesses in that regard. In particular, |
1725 | /// a race between a `write_volatile` and any other operation (reading or writing) |
1726 | /// on the same location is undefined behavior. |
1727 | /// |
1728 | /// # Examples |
1729 | /// |
1730 | /// Basic usage: |
1731 | /// |
1732 | /// ``` |
1733 | /// let mut x = 0; |
1734 | /// let y = &mut x as *mut i32; |
1735 | /// let z = 12; |
1736 | /// |
1737 | /// unsafe { |
1738 | /// std::ptr::write_volatile(y, z); |
1739 | /// assert_eq!(std::ptr::read_volatile(y), 12); |
1740 | /// } |
1741 | /// ``` |
1742 | #[inline ] |
1743 | #[stable (feature = "volatile" , since = "1.9.0" )] |
1744 | #[rustc_diagnostic_item = "ptr_write_volatile" ] |
1745 | #[cfg_attr (miri, track_caller)] // even without panics, this helps for Miri backtraces |
1746 | pub unsafe fn write_volatile<T>(dst: *mut T, src: T) { |
1747 | // SAFETY: the caller must uphold the safety contract for `volatile_store`. |
1748 | unsafe { |
1749 | ub_checks::assert_unsafe_precondition!( |
1750 | check_language_ub, |
1751 | "ptr::write_volatile requires that the pointer argument is aligned and non-null" , |
1752 | ( |
1753 | addr: *mut () = dst as *mut (), |
1754 | align: usize = align_of::<T>(), |
1755 | ) => ub_checks::is_aligned_and_not_null(addr, align) |
1756 | ); |
1757 | intrinsics::volatile_store(dst, val:src); |
1758 | } |
1759 | } |
1760 | |
1761 | /// Align pointer `p`. |
1762 | /// |
1763 | /// Calculate offset (in terms of elements of `size_of::<T>()` stride) that has to be applied |
1764 | /// to pointer `p` so that pointer `p` would get aligned to `a`. |
1765 | /// |
1766 | /// # Safety |
1767 | /// `a` must be a power of two. |
1768 | /// |
1769 | /// # Notes |
1770 | /// This implementation has been carefully tailored to not panic. It is UB for this to panic. |
1771 | /// The only real change that can be made here is change of `INV_TABLE_MOD_16` and associated |
1772 | /// constants. |
1773 | /// |
1774 | /// If we ever decide to make it possible to call the intrinsic with `a` that is not a |
1775 | /// power-of-two, it will probably be more prudent to just change to a naive implementation rather |
1776 | /// than trying to adapt this to accommodate that change. |
1777 | /// |
1778 | /// Any questions go to @nagisa. |
1779 | #[lang = "align_offset" ] |
1780 | pub(crate) const unsafe fn align_offset<T: Sized>(p: *const T, a: usize) -> usize { |
1781 | // FIXME(#75598): Direct use of these intrinsics improves codegen significantly at opt-level <= |
1782 | // 1, where the method versions of these operations are not inlined. |
1783 | use intrinsics::{ |
1784 | assume, cttz_nonzero, exact_div, mul_with_overflow, unchecked_rem, unchecked_sub, |
1785 | wrapping_add, wrapping_mul, wrapping_sub, |
1786 | }; |
1787 | #[cfg (bootstrap)] |
1788 | const unsafe fn unchecked_shl(value: usize, shift: usize) -> usize { |
1789 | value << shift |
1790 | } |
1791 | #[cfg (bootstrap)] |
1792 | const unsafe fn unchecked_shr(value: usize, shift: usize) -> usize { |
1793 | value >> shift |
1794 | } |
1795 | #[cfg (not(bootstrap))] |
1796 | use intrinsics::{unchecked_shl, unchecked_shr}; |
1797 | |
1798 | /// Calculate multiplicative modular inverse of `x` modulo `m`. |
1799 | /// |
1800 | /// This implementation is tailored for `align_offset` and has following preconditions: |
1801 | /// |
1802 | /// * `m` is a power-of-two; |
1803 | /// * `x < m`; (if `x ≥ m`, pass in `x % m` instead) |
1804 | /// |
1805 | /// Implementation of this function shall not panic. Ever. |
1806 | #[inline ] |
1807 | const unsafe fn mod_inv(x: usize, m: usize) -> usize { |
1808 | /// Multiplicative modular inverse table modulo 2⁴ = 16. |
1809 | /// |
1810 | /// Note, that this table does not contain values where inverse does not exist (i.e., for |
1811 | /// `0⁻¹ mod 16`, `2⁻¹ mod 16`, etc.) |
1812 | const INV_TABLE_MOD_16: [u8; 8] = [1, 11, 13, 7, 9, 3, 5, 15]; |
1813 | /// Modulo for which the `INV_TABLE_MOD_16` is intended. |
1814 | const INV_TABLE_MOD: usize = 16; |
1815 | |
1816 | // SAFETY: `m` is required to be a power-of-two, hence non-zero. |
1817 | let m_minus_one = unsafe { unchecked_sub(m, 1) }; |
1818 | let mut inverse = INV_TABLE_MOD_16[(x & (INV_TABLE_MOD - 1)) >> 1] as usize; |
1819 | let mut mod_gate = INV_TABLE_MOD; |
1820 | // We iterate "up" using the following formula: |
1821 | // |
1822 | // $$ xy ≡ 1 (mod 2ⁿ) → xy (2 - xy) ≡ 1 (mod 2²ⁿ) $$ |
1823 | // |
1824 | // This application needs to be applied at least until `2²ⁿ ≥ m`, at which point we can |
1825 | // finally reduce the computation to our desired `m` by taking `inverse mod m`. |
1826 | // |
1827 | // This computation is `O(log log m)`, which is to say, that on 64-bit machines this loop |
1828 | // will always finish in at most 4 iterations. |
1829 | loop { |
1830 | // y = y * (2 - xy) mod n |
1831 | // |
1832 | // Note, that we use wrapping operations here intentionally – the original formula |
1833 | // uses e.g., subtraction `mod n`. It is entirely fine to do them `mod |
1834 | // usize::MAX` instead, because we take the result `mod n` at the end |
1835 | // anyway. |
1836 | if mod_gate >= m { |
1837 | break; |
1838 | } |
1839 | inverse = wrapping_mul(inverse, wrapping_sub(2usize, wrapping_mul(x, inverse))); |
1840 | let (new_gate, overflow) = mul_with_overflow(mod_gate, mod_gate); |
1841 | if overflow { |
1842 | break; |
1843 | } |
1844 | mod_gate = new_gate; |
1845 | } |
1846 | inverse & m_minus_one |
1847 | } |
1848 | |
1849 | let stride = mem::size_of::<T>(); |
1850 | |
1851 | // SAFETY: This is just an inlined `p.addr()` (which is not |
1852 | // a `const fn` so we cannot call it). |
1853 | // During const eval, we hook this function to ensure that the pointer never |
1854 | // has provenance, making this sound. |
1855 | let addr: usize = unsafe { mem::transmute(p) }; |
1856 | |
1857 | // SAFETY: `a` is a power-of-two, therefore non-zero. |
1858 | let a_minus_one = unsafe { unchecked_sub(a, 1) }; |
1859 | |
1860 | if stride == 0 { |
1861 | // SPECIAL_CASE: handle 0-sized types. No matter how many times we step, the address will |
1862 | // stay the same, so no offset will be able to align the pointer unless it is already |
1863 | // aligned. This branch _will_ be optimized out as `stride` is known at compile-time. |
1864 | let p_mod_a = addr & a_minus_one; |
1865 | return if p_mod_a == 0 { 0 } else { usize::MAX }; |
1866 | } |
1867 | |
1868 | // SAFETY: `stride == 0` case has been handled by the special case above. |
1869 | let a_mod_stride = unsafe { unchecked_rem(a, stride) }; |
1870 | if a_mod_stride == 0 { |
1871 | // SPECIAL_CASE: In cases where the `a` is divisible by `stride`, byte offset to align a |
1872 | // pointer can be computed more simply through `-p (mod a)`. In the off-chance the byte |
1873 | // offset is not a multiple of `stride`, the input pointer was misaligned and no pointer |
1874 | // offset will be able to produce a `p` aligned to the specified `a`. |
1875 | // |
1876 | // The naive `-p (mod a)` equation inhibits LLVM's ability to select instructions |
1877 | // like `lea`. We compute `(round_up_to_next_alignment(p, a) - p)` instead. This |
1878 | // redistributes operations around the load-bearing, but pessimizing `and` instruction |
1879 | // sufficiently for LLVM to be able to utilize the various optimizations it knows about. |
1880 | // |
1881 | // LLVM handles the branch here particularly nicely. If this branch needs to be evaluated |
1882 | // at runtime, it will produce a mask `if addr_mod_stride == 0 { 0 } else { usize::MAX }` |
1883 | // in a branch-free way and then bitwise-OR it with whatever result the `-p mod a` |
1884 | // computation produces. |
1885 | |
1886 | let aligned_address = wrapping_add(addr, a_minus_one) & wrapping_sub(0, a); |
1887 | let byte_offset = wrapping_sub(aligned_address, addr); |
1888 | // FIXME: Remove the assume after <https://github.com/llvm/llvm-project/issues/62502> |
1889 | // SAFETY: Masking by `-a` can only affect the low bits, and thus cannot have reduced |
1890 | // the value by more than `a-1`, so even though the intermediate values might have |
1891 | // wrapped, the byte_offset is always in `[0, a)`. |
1892 | unsafe { assume(byte_offset < a) }; |
1893 | |
1894 | // SAFETY: `stride == 0` case has been handled by the special case above. |
1895 | let addr_mod_stride = unsafe { unchecked_rem(addr, stride) }; |
1896 | |
1897 | return if addr_mod_stride == 0 { |
1898 | // SAFETY: `stride` is non-zero. This is guaranteed to divide exactly as well, because |
1899 | // addr has been verified to be aligned to the original type’s alignment requirements. |
1900 | unsafe { exact_div(byte_offset, stride) } |
1901 | } else { |
1902 | usize::MAX |
1903 | }; |
1904 | } |
1905 | |
1906 | // GENERAL_CASE: From here on we’re handling the very general case where `addr` may be |
1907 | // misaligned, there isn’t an obvious relationship between `stride` and `a` that we can take an |
1908 | // advantage of, etc. This case produces machine code that isn’t particularly high quality, |
1909 | // compared to the special cases above. The code produced here is still within the realm of |
1910 | // miracles, given the situations this case has to deal with. |
1911 | |
1912 | // SAFETY: a is power-of-two hence non-zero. stride == 0 case is handled above. |
1913 | // FIXME(const-hack) replace with min |
1914 | let gcdpow = unsafe { |
1915 | let x = cttz_nonzero(stride); |
1916 | let y = cttz_nonzero(a); |
1917 | if x < y { x } else { y } |
1918 | }; |
1919 | // SAFETY: gcdpow has an upper-bound that’s at most the number of bits in a usize. |
1920 | let gcd = unsafe { unchecked_shl(1usize, gcdpow) }; |
1921 | // SAFETY: gcd is always greater or equal to 1. |
1922 | if addr & unsafe { unchecked_sub(gcd, 1) } == 0 { |
1923 | // This branch solves for the following linear congruence equation: |
1924 | // |
1925 | // ` p + so = 0 mod a ` |
1926 | // |
1927 | // `p` here is the pointer value, `s` - stride of `T`, `o` offset in `T`s, and `a` - the |
1928 | // requested alignment. |
1929 | // |
1930 | // With `g = gcd(a, s)`, and the above condition asserting that `p` is also divisible by |
1931 | // `g`, we can denote `a' = a/g`, `s' = s/g`, `p' = p/g`, then this becomes equivalent to: |
1932 | // |
1933 | // ` p' + s'o = 0 mod a' ` |
1934 | // ` o = (a' - (p' mod a')) * (s'^-1 mod a') ` |
1935 | // |
1936 | // The first term is "the relative alignment of `p` to `a`" (divided by the `g`), the |
1937 | // second term is "how does incrementing `p` by `s` bytes change the relative alignment of |
1938 | // `p`" (again divided by `g`). Division by `g` is necessary to make the inverse well |
1939 | // formed if `a` and `s` are not co-prime. |
1940 | // |
1941 | // Furthermore, the result produced by this solution is not "minimal", so it is necessary |
1942 | // to take the result `o mod lcm(s, a)`. This `lcm(s, a)` is the same as `a'`. |
1943 | |
1944 | // SAFETY: `gcdpow` has an upper-bound not greater than the number of trailing 0-bits in |
1945 | // `a`. |
1946 | let a2 = unsafe { unchecked_shr(a, gcdpow) }; |
1947 | // SAFETY: `a2` is non-zero. Shifting `a` by `gcdpow` cannot shift out any of the set bits |
1948 | // in `a` (of which it has exactly one). |
1949 | let a2minus1 = unsafe { unchecked_sub(a2, 1) }; |
1950 | // SAFETY: `gcdpow` has an upper-bound not greater than the number of trailing 0-bits in |
1951 | // `a`. |
1952 | let s2 = unsafe { unchecked_shr(stride & a_minus_one, gcdpow) }; |
1953 | // SAFETY: `gcdpow` has an upper-bound not greater than the number of trailing 0-bits in |
1954 | // `a`. Furthermore, the subtraction cannot overflow, because `a2 = a >> gcdpow` will |
1955 | // always be strictly greater than `(p % a) >> gcdpow`. |
1956 | let minusp2 = unsafe { unchecked_sub(a2, unchecked_shr(addr & a_minus_one, gcdpow)) }; |
1957 | // SAFETY: `a2` is a power-of-two, as proven above. `s2` is strictly less than `a2` |
1958 | // because `(s % a) >> gcdpow` is strictly less than `a >> gcdpow`. |
1959 | return wrapping_mul(minusp2, unsafe { mod_inv(s2, a2) }) & a2minus1; |
1960 | } |
1961 | |
1962 | // Cannot be aligned at all. |
1963 | usize::MAX |
1964 | } |
1965 | |
1966 | /// Compares raw pointers for equality. |
1967 | /// |
1968 | /// This is the same as using the `==` operator, but less generic: |
1969 | /// the arguments have to be `*const T` raw pointers, |
1970 | /// not anything that implements `PartialEq`. |
1971 | /// |
1972 | /// This can be used to compare `&T` references (which coerce to `*const T` implicitly) |
1973 | /// by their address rather than comparing the values they point to |
1974 | /// (which is what the `PartialEq for &T` implementation does). |
1975 | /// |
1976 | /// When comparing wide pointers, both the address and the metadata are tested for equality. |
1977 | /// However, note that comparing trait object pointers (`*const dyn Trait`) is unreliable: pointers |
1978 | /// to values of the same underlying type can compare inequal (because vtables are duplicated in |
1979 | /// multiple codegen units), and pointers to values of *different* underlying type can compare equal |
1980 | /// (since identical vtables can be deduplicated within a codegen unit). |
1981 | /// |
1982 | /// # Examples |
1983 | /// |
1984 | /// ``` |
1985 | /// use std::ptr; |
1986 | /// |
1987 | /// let five = 5; |
1988 | /// let other_five = 5; |
1989 | /// let five_ref = &five; |
1990 | /// let same_five_ref = &five; |
1991 | /// let other_five_ref = &other_five; |
1992 | /// |
1993 | /// assert!(five_ref == same_five_ref); |
1994 | /// assert!(ptr::eq(five_ref, same_five_ref)); |
1995 | /// |
1996 | /// assert!(five_ref == other_five_ref); |
1997 | /// assert!(!ptr::eq(five_ref, other_five_ref)); |
1998 | /// ``` |
1999 | /// |
2000 | /// Slices are also compared by their length (fat pointers): |
2001 | /// |
2002 | /// ``` |
2003 | /// let a = [1, 2, 3]; |
2004 | /// assert!(std::ptr::eq(&a[..3], &a[..3])); |
2005 | /// assert!(!std::ptr::eq(&a[..2], &a[..3])); |
2006 | /// assert!(!std::ptr::eq(&a[0..2], &a[1..3])); |
2007 | /// ``` |
2008 | #[stable (feature = "ptr_eq" , since = "1.17.0" )] |
2009 | #[inline (always)] |
2010 | #[must_use = "pointer comparison produces a value" ] |
2011 | #[rustc_diagnostic_item = "ptr_eq" ] |
2012 | #[allow (ambiguous_wide_pointer_comparisons)] // it's actually clear here |
2013 | pub fn eq<T: ?Sized>(a: *const T, b: *const T) -> bool { |
2014 | a == b |
2015 | } |
2016 | |
2017 | /// Compares the *addresses* of the two pointers for equality, |
2018 | /// ignoring any metadata in fat pointers. |
2019 | /// |
2020 | /// If the arguments are thin pointers of the same type, |
2021 | /// then this is the same as [`eq`]. |
2022 | /// |
2023 | /// # Examples |
2024 | /// |
2025 | /// ``` |
2026 | /// use std::ptr; |
2027 | /// |
2028 | /// let whole: &[i32; 3] = &[1, 2, 3]; |
2029 | /// let first: &i32 = &whole[0]; |
2030 | /// |
2031 | /// assert!(ptr::addr_eq(whole, first)); |
2032 | /// assert!(!ptr::eq::<dyn std::fmt::Debug>(whole, first)); |
2033 | /// ``` |
2034 | #[stable (feature = "ptr_addr_eq" , since = "1.76.0" )] |
2035 | #[inline (always)] |
2036 | #[must_use = "pointer comparison produces a value" ] |
2037 | pub fn addr_eq<T: ?Sized, U: ?Sized>(p: *const T, q: *const U) -> bool { |
2038 | (p as *const ()) == (q as *const ()) |
2039 | } |
2040 | |
2041 | /// Hash a raw pointer. |
2042 | /// |
2043 | /// This can be used to hash a `&T` reference (which coerces to `*const T` implicitly) |
2044 | /// by its address rather than the value it points to |
2045 | /// (which is what the `Hash for &T` implementation does). |
2046 | /// |
2047 | /// # Examples |
2048 | /// |
2049 | /// ``` |
2050 | /// use std::hash::{DefaultHasher, Hash, Hasher}; |
2051 | /// use std::ptr; |
2052 | /// |
2053 | /// let five = 5; |
2054 | /// let five_ref = &five; |
2055 | /// |
2056 | /// let mut hasher = DefaultHasher::new(); |
2057 | /// ptr::hash(five_ref, &mut hasher); |
2058 | /// let actual = hasher.finish(); |
2059 | /// |
2060 | /// let mut hasher = DefaultHasher::new(); |
2061 | /// (five_ref as *const i32).hash(&mut hasher); |
2062 | /// let expected = hasher.finish(); |
2063 | /// |
2064 | /// assert_eq!(actual, expected); |
2065 | /// ``` |
2066 | #[stable (feature = "ptr_hash" , since = "1.35.0" )] |
2067 | pub fn hash<T: ?Sized, S: hash::Hasher>(hashee: *const T, into: &mut S) { |
2068 | use crate::hash::Hash; |
2069 | hashee.hash(state:into); |
2070 | } |
2071 | |
2072 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2073 | impl<F: FnPtr> PartialEq for F { |
2074 | #[inline ] |
2075 | fn eq(&self, other: &Self) -> bool { |
2076 | self.addr() == other.addr() |
2077 | } |
2078 | } |
2079 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2080 | impl<F: FnPtr> Eq for F {} |
2081 | |
2082 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2083 | impl<F: FnPtr> PartialOrd for F { |
2084 | #[inline ] |
2085 | fn partial_cmp(&self, other: &Self) -> Option<Ordering> { |
2086 | self.addr().partial_cmp(&other.addr()) |
2087 | } |
2088 | } |
2089 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2090 | impl<F: FnPtr> Ord for F { |
2091 | #[inline ] |
2092 | fn cmp(&self, other: &Self) -> Ordering { |
2093 | self.addr().cmp(&other.addr()) |
2094 | } |
2095 | } |
2096 | |
2097 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2098 | impl<F: FnPtr> hash::Hash for F { |
2099 | fn hash<HH: hash::Hasher>(&self, state: &mut HH) { |
2100 | state.write_usize(self.addr() as _) |
2101 | } |
2102 | } |
2103 | |
2104 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2105 | impl<F: FnPtr> fmt::Pointer for F { |
2106 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
2107 | fmt::pointer_fmt_inner(self.addr() as _, f) |
2108 | } |
2109 | } |
2110 | |
2111 | #[stable (feature = "fnptr_impls" , since = "1.4.0" )] |
2112 | impl<F: FnPtr> fmt::Debug for F { |
2113 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
2114 | fmt::pointer_fmt_inner(self.addr() as _, f) |
2115 | } |
2116 | } |
2117 | |
2118 | /// Create a `const` raw pointer to a place, without creating an intermediate reference. |
2119 | /// |
2120 | /// Creating a reference with `&`/`&mut` is only allowed if the pointer is properly aligned |
2121 | /// and points to initialized data. For cases where those requirements do not hold, |
2122 | /// raw pointers should be used instead. However, `&expr as *const _` creates a reference |
2123 | /// before casting it to a raw pointer, and that reference is subject to the same rules |
2124 | /// as all other references. This macro can create a raw pointer *without* creating |
2125 | /// a reference first. |
2126 | /// |
2127 | /// See [`addr_of_mut`] for how to create a pointer to uninitialized data. |
2128 | /// Doing that with `addr_of` would not make much sense since one could only |
2129 | /// read the data, and that would be Undefined Behavior. |
2130 | /// |
2131 | /// # Safety |
2132 | /// |
2133 | /// The `expr` in `addr_of!(expr)` is evaluated as a place expression, but never loads from the |
2134 | /// place or requires the place to be dereferenceable. This means that `addr_of!((*ptr).field)` |
2135 | /// still requires the projection to `field` to be in-bounds, using the same rules as [`offset`]. |
2136 | /// However, `addr_of!(*ptr)` is defined behavior even if `ptr` is null, dangling, or misaligned. |
2137 | /// |
2138 | /// Note that `Deref`/`Index` coercions (and their mutable counterparts) are applied inside |
2139 | /// `addr_of!` like everywhere else, in which case a reference is created to call `Deref::deref` or |
2140 | /// `Index::index`, respectively. The statements above only apply when no such coercions are |
2141 | /// applied. |
2142 | /// |
2143 | /// [`offset`]: pointer::offset |
2144 | /// |
2145 | /// # Example |
2146 | /// |
2147 | /// **Correct usage: Creating a pointer to unaligned data** |
2148 | /// |
2149 | /// ``` |
2150 | /// use std::ptr; |
2151 | /// |
2152 | /// #[repr(packed)] |
2153 | /// struct Packed { |
2154 | /// f1: u8, |
2155 | /// f2: u16, |
2156 | /// } |
2157 | /// |
2158 | /// let packed = Packed { f1: 1, f2: 2 }; |
2159 | /// // `&packed.f2` would create an unaligned reference, and thus be Undefined Behavior! |
2160 | /// let raw_f2 = ptr::addr_of!(packed.f2); |
2161 | /// assert_eq!(unsafe { raw_f2.read_unaligned() }, 2); |
2162 | /// ``` |
2163 | /// |
2164 | /// **Incorrect usage: Out-of-bounds fields projection** |
2165 | /// |
2166 | /// ```rust,no_run |
2167 | /// use std::ptr; |
2168 | /// |
2169 | /// #[repr(C)] |
2170 | /// struct MyStruct { |
2171 | /// field1: i32, |
2172 | /// field2: i32, |
2173 | /// } |
2174 | /// |
2175 | /// let ptr: *const MyStruct = ptr::null(); |
2176 | /// let fieldptr = unsafe { ptr::addr_of!((*ptr).field2) }; // Undefined Behavior ⚠️ |
2177 | /// ``` |
2178 | /// |
2179 | /// The field projection `.field2` would offset the pointer by 4 bytes, |
2180 | /// but the pointer is not in-bounds of an allocation for 4 bytes, |
2181 | /// so this offset is Undefined Behavior. |
2182 | /// See the [`offset`] docs for a full list of requirements for inbounds pointer arithmetic; the |
2183 | /// same requirements apply to field projections, even inside `addr_of!`. (In particular, it makes |
2184 | /// no difference whether the pointer is null or dangling.) |
2185 | #[stable (feature = "raw_ref_macros" , since = "1.51.0" )] |
2186 | #[rustc_macro_transparency = "semitransparent" ] |
2187 | #[allow_internal_unstable (raw_ref_op)] |
2188 | pub macro addr_of($place:expr) { |
2189 | &raw const $place |
2190 | } |
2191 | |
2192 | /// Create a `mut` raw pointer to a place, without creating an intermediate reference. |
2193 | /// |
2194 | /// Creating a reference with `&`/`&mut` is only allowed if the pointer is properly aligned |
2195 | /// and points to initialized data. For cases where those requirements do not hold, |
2196 | /// raw pointers should be used instead. However, `&mut expr as *mut _` creates a reference |
2197 | /// before casting it to a raw pointer, and that reference is subject to the same rules |
2198 | /// as all other references. This macro can create a raw pointer *without* creating |
2199 | /// a reference first. |
2200 | /// |
2201 | /// # Safety |
2202 | /// |
2203 | /// The `expr` in `addr_of_mut!(expr)` is evaluated as a place expression, but never loads from the |
2204 | /// place or requires the place to be dereferenceable. This means that `addr_of_mut!((*ptr).field)` |
2205 | /// still requires the projection to `field` to be in-bounds, using the same rules as [`offset`]. |
2206 | /// However, `addr_of_mut!(*ptr)` is defined behavior even if `ptr` is null, dangling, or misaligned. |
2207 | /// |
2208 | /// Note that `Deref`/`Index` coercions (and their mutable counterparts) are applied inside |
2209 | /// `addr_of_mut!` like everywhere else, in which case a reference is created to call `Deref::deref` |
2210 | /// or `Index::index`, respectively. The statements above only apply when no such coercions are |
2211 | /// applied. |
2212 | /// |
2213 | /// [`offset`]: pointer::offset |
2214 | /// |
2215 | /// # Examples |
2216 | /// |
2217 | /// **Correct usage: Creating a pointer to unaligned data** |
2218 | /// |
2219 | /// ``` |
2220 | /// use std::ptr; |
2221 | /// |
2222 | /// #[repr(packed)] |
2223 | /// struct Packed { |
2224 | /// f1: u8, |
2225 | /// f2: u16, |
2226 | /// } |
2227 | /// |
2228 | /// let mut packed = Packed { f1: 1, f2: 2 }; |
2229 | /// // `&mut packed.f2` would create an unaligned reference, and thus be Undefined Behavior! |
2230 | /// let raw_f2 = ptr::addr_of_mut!(packed.f2); |
2231 | /// unsafe { raw_f2.write_unaligned(42); } |
2232 | /// assert_eq!({packed.f2}, 42); // `{...}` forces copying the field instead of creating a reference. |
2233 | /// ``` |
2234 | /// |
2235 | /// **Correct usage: Creating a pointer to uninitialized data** |
2236 | /// |
2237 | /// ```rust |
2238 | /// use std::{ptr, mem::MaybeUninit}; |
2239 | /// |
2240 | /// struct Demo { |
2241 | /// field: bool, |
2242 | /// } |
2243 | /// |
2244 | /// let mut uninit = MaybeUninit::<Demo>::uninit(); |
2245 | /// // `&uninit.as_mut().field` would create a reference to an uninitialized `bool`, |
2246 | /// // and thus be Undefined Behavior! |
2247 | /// let f1_ptr = unsafe { ptr::addr_of_mut!((*uninit.as_mut_ptr()).field) }; |
2248 | /// unsafe { f1_ptr.write(true); } |
2249 | /// let init = unsafe { uninit.assume_init() }; |
2250 | /// ``` |
2251 | /// |
2252 | /// **Incorrect usage: Out-of-bounds fields projection** |
2253 | /// |
2254 | /// ```rust,no_run |
2255 | /// use std::ptr; |
2256 | /// |
2257 | /// #[repr(C)] |
2258 | /// struct MyStruct { |
2259 | /// field1: i32, |
2260 | /// field2: i32, |
2261 | /// } |
2262 | /// |
2263 | /// let ptr: *mut MyStruct = ptr::null_mut(); |
2264 | /// let fieldptr = unsafe { ptr::addr_of_mut!((*ptr).field2) }; // Undefined Behavior ⚠️ |
2265 | /// ``` |
2266 | /// |
2267 | /// The field projection `.field2` would offset the pointer by 4 bytes, |
2268 | /// but the pointer is not in-bounds of an allocation for 4 bytes, |
2269 | /// so this offset is Undefined Behavior. |
2270 | /// See the [`offset`] docs for a full list of requirements for inbounds pointer arithmetic; the |
2271 | /// same requirements apply to field projections, even inside `addr_of_mut!`. (In particular, it |
2272 | /// makes no difference whether the pointer is null or dangling.) |
2273 | #[stable (feature = "raw_ref_macros" , since = "1.51.0" )] |
2274 | #[rustc_macro_transparency = "semitransparent" ] |
2275 | #[allow_internal_unstable (raw_ref_op)] |
2276 | pub macro addr_of_mut($place:expr) { |
2277 | &raw mut $place |
2278 | } |
2279 | |