virtual_buffer/
lib.rs

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
//! This crate provides a cross-platform API for dealing with buffers backed by raw virtual memory.
//!
//! Apart from providing protection and isolation between processes, paging, and memory mapped
//! hardware, virtual memory serves to solve another critical issue: the issue of the virtual
//! buffer. It allows us to [reserve] a range of memory only in the process's virtual address
//! space, without actually [committing] any of the memory. This can be used to create a buffer
//! that's infinitely growable and shrinkable *in-place*, without wasting any physical memory, nor
//! even overcommitting any memory. It can also be used to create sparse data structures that don't
//! overcommit memory.
//!
//! The property of growing in-place is very valuable when reallocation is impossible, for example
//! because the data structure needs to be concurrent or otherwise pinned. It may also be of use
//! for single-threaded use cases if reallocation is too expensive (say, tens to hundreds of MB).
//! However, it's probably easier to use something like [`Vec::with_capacity`] in that case.
//!
//! See also [the `vec` module] for an implementation of a concurrent vector.
//!
//! # Reserving
//!
//! Reserving memory involves allocating a range of virtual address space, such that other
//! allocations within the same process can't reserve any of the same virtual address space for
//! anything else. Memory that has been reserved has zero memory cost, however, it can't be
//! accessed. In order to access any of the pages, you will have to commit them first.
//!
//! # Committing
//!
//! A range of reserved memory can be committed to make it accessible. Memory that has been freshly
//! committed doesn't use up any physical memory. It merely counts towards overcommitment, which
//! may increase the likelihood of being OOM-killed, and may take up space for page tables and may
//! use some space in the page file. A committed page is only ever backed by a physical page after
//! being written to for the first time (being "faulted"), or when it was [prefaulted].
//!
//! Committed memory can be committed again without issue, so there is no need to keep track of
//! which pages have been committed in order to safely commit some of them.
//!
//! # Decommitting
//!
//! A range of committed memory can be decommitted, making it inaccessible again, and releasing any
//! physical memory that may have been used for them back to the operating system. Decommitted
//! memory is still reserved.
//!
//! Reserved but uncommitted memory can be decommitted without issue, so there is no need to keep
//! track of which pages have been committed in order to safely decommit some of them.
//!
//! # Unreserving
//!
//! Memory that is unreserved is available for new allocations to reserve again.
//!
//! Committed memory can be unreserved without needing to be decommitted first. However, it's not
//! possible to unreserve a range of reserved memory, only the entire allocation.
//!
//! # Prefaulting
//!
//! By default, each committed page is only ever backed by physical memory after it was first
//! written to. Since this happens for every page, and can be slightly costly due to the overhead
//! of a context switch, operating systems provide a way to *prefault* multiple pages at once.
//!
//! # Pages
//!
//! A page refers to the granularity at which the processor's Memory Management Unit operates and
//! varies between processor architectures. As such, virtual memory operations can only affect
//! ranges that are aligned to the *page size*.
//!
//! # Cargo features
//!
//! | Feature | Description                                       |
//! |---------|---------------------------------------------------|
//! | std     | Enables the use of `std::error` and `std::borrow` |
//!
//! [reserve]: self#reserving
//! [committing]: self#committing
//! [`Vec::with_capacity`]: https://doc.rust-lang.org/std/vec/struct.Vec.html#method.with_capacity
//! [the `vec` module]: self::vec
//! [prefaulted]: self#prefaulting

#![allow(
    unused_unsafe,
    clippy::doc_markdown,
    clippy::inline_always,
    clippy::unused_self
)]
#![forbid(unsafe_op_in_unsafe_fn)]
#![cfg_attr(not(feature = "std"), no_std)]

#[cfg(not(any(unix, windows)))]
compile_error!("unsupported platform");

#[cfg(unix)]
use self::unix as sys;
#[cfg(windows)]
use self::windows as sys;
use core::{fmt, mem};

pub mod vec;

/// An allocation backed by raw virtual memory, giving you the power to directly manipulate the
/// pages within it.
///
/// See also [the crate-level documentation] for more information about virtual memory.
///
/// [the crate-level documentation]: self
pub struct Allocation {
    inner: sys::Allocation,
}

impl Allocation {
    /// Allocates a new region in the process's virtual address space.
    ///
    /// `size` is the size to [reserve] in bytes. This number can be excessively huge, as none of
    /// the memory is [committed] until you call [`commit`]. The memory is [unreserved] when the
    /// `Allocation` is dropped.
    ///
    /// # Errors
    ///
    /// Returns an error if the operating system returns an error.
    ///
    /// # Panics
    ///
    /// - Panics if `size` is not aligned to the [page size].
    /// - Panics if `size` is zero.
    ///
    /// [reserve]: self#reserving
    /// [page size]: self#pages
    /// [committed]: self#committing
    /// [unreserved]: self#unreserving
    /// [`commit`]: Self::commit
    pub fn new(size: usize) -> Result<Self> {
        assert!(is_aligned(size, page_size()));
        assert_ne!(size, 0);

        let inner = sys::Allocation::new(size)?;

        Ok(Allocation { inner })
    }

    /// Creates a dangling `Allocation`, that is, an allocation with a dangling pointer and zero
    /// size.
    ///
    /// This is useful as a placeholder value to defer allocation until later or if no allocation
    /// is needed.
    ///
    /// `alignment` is the alignment of the allocation's pointer, and must be a power of two.
    ///
    /// # Panics
    ///
    /// Panics if `alignment` is not a power of two.
    #[inline]
    #[must_use]
    pub const fn dangling(alignment: usize) -> Allocation {
        let inner = sys::Allocation::dangling(alignment);

        Allocation { inner }
    }

    /// Returns the pointer to the beginning of the allocation.
    ///
    /// The returned pointer is always valid, including [dangling allocations], for reads and
    /// writes of [`size()`] bytes in the sense that it can never lead to undefined behavior.
    /// However, doing a read or write access to [pages] that have not been [committed] will result
    /// in the process receiving SIGSEGV / STATUS_ACCESS_VIOLATION.
    ///
    /// The pointer must not be accessed after `self` has been dropped.
    ///
    /// [dangling allocations]: Self::dangling
    /// [pages]: self#pages
    /// [committed]: self#committing
    /// [`size()`]: Self::size
    #[inline(always)]
    #[must_use]
    pub const fn ptr(&self) -> *mut u8 {
        self.inner.ptr().cast()
    }

    /// Returns the size that was used to [allocate] `self` aligned up to the [page size].
    ///
    /// [allocate]: Self::new
    /// [page size]: self#pages
    #[inline(always)]
    #[must_use]
    pub const fn size(&self) -> usize {
        self.inner.size()
    }

    /// [Commits] the given region of memory.
    ///
    /// # Errors
    ///
    /// Returns an error if the operating system returns an error.
    ///
    /// # Panics
    ///
    /// - Panics if the allocation is [dangling].
    /// - Panics if `ptr` and `size` denote a region that is out of bounds of the allocation.
    /// - Panics if `ptr` and/or `size` is not aligned to the [page size].
    /// - Panics if `size` is zero.
    ///
    /// [Commits]: self#committing
    /// [dangling]: Self::dangling
    /// [page size]: self#pages
    pub fn commit(&self, ptr: *mut u8, size: usize) -> Result<()> {
        self.check_range(ptr, size);

        // SAFETY: Enforced by the `check_range` call above.
        unsafe { self.inner.commit(ptr.cast(), size) }
    }

    /// [Decommits] the given region of memory.
    ///
    /// # Errors
    ///
    /// Returns an error if the operating system returns an error.
    ///
    /// # Panics
    ///
    /// - Panics if the allocation is [dangling].
    /// - Panics if `ptr` and `size` denote a region that is out of bounds of the allocation.
    /// - Panics if `ptr` and/or `size` is not aligned to the [page size].
    /// - Panics if `size` is zero.
    ///
    /// [Decommits]: self#decommitting
    /// [dangling]: Self::dangling
    /// [page size]: self#pages
    pub fn decommit(&self, ptr: *mut u8, size: usize) -> Result<()> {
        self.check_range(ptr, size);

        // SAFETY: Enforced by the `check_range` call above.
        unsafe { self.inner.decommit(ptr.cast(), size) }
    }

    /// [Prefaults] the given region of memory.
    ///
    /// # Errors
    ///
    /// Returns an error if the operating system returns an error.
    ///
    /// # Panics
    ///
    /// - Panics if the allocation is [dangling].
    /// - Panics if `ptr` and `size` denote a region that is out of bounds of the allocation.
    /// - Panics if `ptr` and/or `size` is not aligned to the [page size].
    /// - Panics if `size` is zero.
    ///
    /// [Prefaults]: self#prefaulting
    /// [dangling]: Self::dangling
    /// [page size]: self#pages
    pub fn prefault(&self, ptr: *mut u8, size: usize) -> Result<()> {
        self.check_range(ptr, size);

        // SAFETY: Enforced by the `check_range` call above.
        unsafe { self.inner.prefault(ptr.cast(), size) }
    }

    #[inline(never)]
    fn check_range(&self, ptr: *mut u8, size: usize) {
        assert_ne!(self.size(), 0, "the allocation is dangling");
        assert_ne!(size, 0);

        let allocated_range = addr(self.ptr())..addr(self.ptr()) + self.size();
        let requested_range = addr(ptr)..addr(ptr).checked_add(size).unwrap();
        assert!(allocated_range.start <= requested_range.start);
        assert!(requested_range.end <= allocated_range.end);

        let page_size = page_size();
        assert!(is_aligned(addr(ptr), page_size));
        assert!(is_aligned(size, page_size));
    }
}

impl fmt::Debug for Allocation {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        fmt::Debug::fmt(&self.inner, f)
    }
}

/// Returns the [page size] of the system.
///
/// The value is cached globally and very fast to retrieve.
///
/// [page size]: self#pages
#[inline(always)]
#[must_use]
pub fn page_size() -> usize {
    sys::page_size()
}

/// Returns the smallest value greater or equal to `val` that is a multiple of `alignment`. Returns
/// zero on overflow.
///
/// You may use this together with [`page_size`] to align your regions for committing/decommitting.
///
/// `alignment` must be a power of two (which implies that it must be non-zero).
#[inline(always)]
#[must_use]
pub const fn align_up(val: usize, alignment: usize) -> usize {
    debug_assert!(alignment.is_power_of_two());

    val.wrapping_add(alignment - 1) & !(alignment - 1)
}

/// Returns the largest value smaller or equal to `val` that is a multiple of `alignment`.
///
/// You may use this together with [`page_size`] to align your regions for committing/decommitting.
///
/// `alignment` must be a power of two (which implies that it must be non-zero).
#[inline(always)]
#[must_use]
pub const fn align_down(val: usize, alignment: usize) -> usize {
    debug_assert!(alignment.is_power_of_two());

    val & !(alignment - 1)
}

fn is_aligned(val: usize, alignment: usize) -> bool {
    debug_assert!(alignment.is_power_of_two());

    val & (alignment - 1) == 0
}

/// The type returned by the various [`Allocation`] methods.
pub type Result<T, E = Error> = ::core::result::Result<T, E>;

/// Represents an OS error that can be returned by the by the various [`Allocation`] methods.
#[derive(Debug)]
pub struct Error {
    code: i32,
}

impl Error {
    /// Returns the OS error that this error represents.
    #[inline]
    #[must_use]
    pub fn as_raw_os_error(&self) -> i32 {
        self.code
    }
}

impl fmt::Display for Error {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        sys::format_error(self.code, f)
    }
}

#[cfg(feature = "std")]
impl std::error::Error for Error {}

#[cfg(unix)]
mod unix {
    #![allow(non_camel_case_types)]

    use super::{without_provenance_mut, Result};
    use core::{
        ffi::{c_char, c_int, c_void, CStr},
        fmt,
        ptr::{self, NonNull},
        str,
        sync::atomic::{AtomicUsize, Ordering},
    };

    #[derive(Debug)]
    pub struct Allocation {
        ptr: NonNull<c_void>,
        size: usize,
    }

    // SAFETY: It is safe to send `Allocation::ptr` to another thread because the user would have
    // to use unsafe code themself by dereferencing it.
    unsafe impl Send for Allocation {}

    // SAFETY: It is safe to share `Allocation::ptr` between threads because the user would have to
    // use unsafe code themself by dereferencing it.
    unsafe impl Sync for Allocation {}

    impl Allocation {
        pub fn new(size: usize) -> Result<Self> {
            // Miri doesn't support protections other than read/write.
            #[cfg(not(miri))]
            let prot = libc::PROT_NONE;
            #[cfg(miri)]
            let prot = libc::PROT_READ | libc::PROT_WRITE;

            let flags = libc::MAP_PRIVATE | libc::MAP_ANONYMOUS;

            // SAFETY: Enforced by the fact that we are passing in a null pointer as the address,
            // so that no existing mappings can be affected in any way.
            let ptr = unsafe { libc::mmap(ptr::null_mut(), size, prot, flags, -1, 0) };

            result(ptr != libc::MAP_FAILED)?;

            Ok(Allocation {
                ptr: NonNull::new(ptr).unwrap(),
                size,
            })
        }

        #[inline]
        pub const fn dangling(alignment: usize) -> Self {
            assert!(alignment.is_power_of_two());

            Allocation {
                // SAFETY: We checked that `alignment` is a power of two, which means it must be
                // non-zero.
                ptr: unsafe { NonNull::new_unchecked(without_provenance_mut(alignment).cast()) },
                size: 0,
            }
        }

        #[inline(always)]
        pub const fn ptr(&self) -> *mut c_void {
            self.ptr.as_ptr()
        }

        #[inline(always)]
        pub const fn size(&self) -> usize {
            self.size
        }

        #[cfg(not(miri))]
        pub unsafe fn commit(&self, ptr: *mut c_void, size: usize) -> Result<()> {
            // SAFETY: The caller must guarantee that `ptr` and `size` are in bounds of the
            // allocation such that no other allocations can be affected and that `ptr` is aligned
            // to the page size. As for this allocation, the only way to access it is by unsafely
            // dererencing its pointer, where the user has the responsibility to make sure that
            // that is valid.
            result(unsafe { libc::mprotect(ptr, size, libc::PROT_READ | libc::PROT_WRITE) } == 0)
        }

        #[cfg(miri)]
        pub unsafe fn commit(&self, _ptr: *mut c_void, _size: usize) -> Result<()> {
            // Committing memory has no effect on the operational semantics, so there's nothing for
            // Miri to test anyway except hitting a segmentation fault which is perfectly defined
            // behavior.
            Ok(())
        }

        #[cfg(not(miri))]
        pub unsafe fn decommit(&self, ptr: *mut c_void, size: usize) -> Result<()> {
            // God forbid this be one syscall :ferrisPensive:

            // SAFETY: The caller must guarantee that `ptr` and `size` are in bounds of the
            // allocation such that no other allocations can be affected and that `ptr` is aligned
            // to the page size. As for this allocation, the only way to access it is by unsafely
            // dererencing its pointer, where the user has the responsibility to make sure that
            // that is valid.
            result(unsafe { libc::madvise(ptr, size, libc::MADV_DONTNEED) } == 0)?;

            // SAFETY: Same as the previous.
            result(unsafe { libc::mprotect(ptr, size, libc::PROT_NONE) } == 0)?;

            Ok(())
        }

        #[cfg(miri)]
        pub unsafe fn decommit(&self, _ptr: *mut c_void, _size: usize) -> Result<()> {
            // Decommitting memory has no effect on the operational semantics, so there's nothing
            // for Miri to test anyway except hitting a segmentation fault which is perfectly
            // defined behavior.
            Ok(())
        }

        #[cfg(not(miri))]
        pub unsafe fn prefault(&self, ptr: *mut c_void, size: usize) -> Result<()> {
            // SAFETY: The caller must guarantee that `ptr` and `size` are in bounds of the
            // allocation such that no other allocations can be affected and that `ptr` is aligned
            // to the page size. This call is otherwise purely an optimization hint and can't
            // change program behavior.
            result(unsafe { libc::madvise(ptr, size, libc::MADV_WILLNEED) } == 0)
        }

        #[cfg(miri)]
        pub unsafe fn prefault(&self, _ptr: *mut c_void, _size: usize) -> Result<()> {
            Ok(())
        }
    }

    impl Drop for Allocation {
        fn drop(&mut self) {
            if self.size != 0 {
                // SAFETY: It is the responsibility of the user who is unsafely derefercing the
                // allocation's pointer to ensure that those accesses don't happen after the
                // allocation has been dropped. We know the pointer and its size is valid because
                // we allocated it.
                unsafe { libc::munmap(self.ptr(), self.size) };
            }
        }
    }

    #[inline(always)]
    pub fn page_size() -> usize {
        static PAGE_SIZE: AtomicUsize = AtomicUsize::new(0);

        #[cold]
        fn page_size_slow() -> usize {
            let page_size = usize::try_from(unsafe { libc::sysconf(libc::_SC_PAGE_SIZE) }).unwrap();
            PAGE_SIZE.store(page_size, Ordering::Relaxed);

            page_size
        }

        let cached = PAGE_SIZE.load(Ordering::Relaxed);

        if cached != 0 {
            cached
        } else {
            page_size_slow()
        }
    }

    fn result(condition: bool) -> Result<()> {
        if condition {
            Ok(())
        } else {
            Err(super::Error { code: errno() })
        }
    }

    #[cfg(not(target_os = "vxworks"))]
    fn errno() -> i32 {
        unsafe { *errno_location() as i32 }
    }

    #[cfg(target_os = "vxworks")]
    fn errno() -> i32 {
        unsafe { libc::errnoGet() as i32 }
    }

    pub fn format_error(errnum: i32, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        let mut buf = [0 as c_char; 128];

        let res = unsafe {
            libc::strerror_r(errnum as c_int, buf.as_mut_ptr(), buf.len() as libc::size_t)
        };

        assert!(res >= 0, "strerror_r failure");

        let buf = unsafe { CStr::from_ptr(buf.as_ptr()) }.to_bytes();

        let s = str::from_utf8(buf).unwrap_or_else(|err| {
            // SAFETY: The `from_utf8` call above checked that `err.valid_up_to()` bytes are valid.
            unsafe { str::from_utf8_unchecked(&buf[..err.valid_up_to()]) }
        });

        f.write_str(s)
    }

    extern "C" {
        #[cfg(not(target_os = "vxworks"))]
        #[cfg_attr(
            any(
                target_os = "linux",
                target_os = "emscripten",
                target_os = "fuchsia",
                target_os = "l4re",
                target_os = "hurd",
                target_os = "dragonfly"
            ),
            link_name = "__errno_location"
        )]
        #[cfg_attr(
            any(
                target_os = "netbsd",
                target_os = "openbsd",
                target_os = "android",
                target_os = "redox",
                target_env = "newlib"
            ),
            link_name = "__errno"
        )]
        #[cfg_attr(
            any(target_os = "solaris", target_os = "illumos"),
            link_name = "___errno"
        )]
        #[cfg_attr(target_os = "nto", link_name = "__get_errno_ptr")]
        #[cfg_attr(
            any(target_os = "freebsd", target_vendor = "apple"),
            link_name = "__error"
        )]
        #[cfg_attr(target_os = "haiku", link_name = "_errnop")]
        #[cfg_attr(target_os = "aix", link_name = "_Errno")]
        fn errno_location() -> *mut c_int;
    }
}

#[cfg(windows)]
mod windows {
    #![allow(non_camel_case_types, non_snake_case)]

    use super::{without_provenance_mut, Result};
    use core::{
        ffi::c_void,
        fmt, mem,
        ptr::{self, NonNull},
        str,
        sync::atomic::{AtomicUsize, Ordering},
    };

    #[derive(Debug)]
    pub struct Allocation {
        ptr: NonNull<c_void>,
        size: usize,
    }

    // SAFETY: It is safe to send `Allocation::ptr` to another thread because the user would have
    // to use unsafe code themself by dereferencing it.
    unsafe impl Send for Allocation {}

    // SAFETY: It is safe to share `Allocation::ptr` between threads because the user would have to
    // use unsafe code themself by dereferencing it.
    unsafe impl Sync for Allocation {}

    impl Allocation {
        pub fn new(size: usize) -> Result<Self> {
            // Miri doesn't support protections other than read/write.
            #[cfg(not(miri))]
            let protect = PAGE_NOACCESS;
            #[cfg(miri)]
            let protect = PAGE_READWRITE;

            // SAFETY: Enforced by the fact that we are passing in a null pointer as the address,
            // so that no existing mappings can be affected in any way.
            let ptr = unsafe { VirtualAlloc(ptr::null_mut(), size, MEM_RESERVE, protect) };

            result(!ptr.is_null())?;

            Ok(Allocation {
                ptr: NonNull::new(ptr).unwrap(),
                size,
            })
        }

        #[inline]
        pub const fn dangling(alignment: usize) -> Self {
            assert!(alignment.is_power_of_two());

            Allocation {
                // SAFETY: We checked that `alignment` is a power of two, which means it must be
                // non-zero.
                ptr: unsafe { NonNull::new_unchecked(without_provenance_mut(alignment).cast()) },
                size: 0,
            }
        }

        #[inline(always)]
        pub const fn ptr(&self) -> *mut c_void {
            self.ptr.as_ptr()
        }

        #[inline(always)]
        pub const fn size(&self) -> usize {
            self.size
        }

        #[cfg(not(miri))]
        pub unsafe fn commit(&self, ptr: *mut c_void, size: usize) -> Result<()> {
            // SAFETY: The caller must guarantee that `ptr` and `size` are in bounds of the
            // allocation such that no other allocations can be affected. As for this allocation,
            // the only way to access it is by unsafely dererencing its pointer, where the user
            // has the responsibility to make sure that that is valid.
            result(!unsafe { VirtualAlloc(ptr, size, MEM_COMMIT, PAGE_READWRITE) }.is_null())
        }

        #[cfg(miri)]
        pub unsafe fn commit(&self, _ptr: *mut c_void, _size: usize) -> Result<()> {
            // Committing memory has no effect on the operational semantics, so there's nothing for
            // Miri to test anyway except hitting a segmentation fault which is perfectly defined
            // behavior.
            Ok(())
        }

        #[cfg(not(miri))]
        pub unsafe fn decommit(&self, ptr: *mut c_void, size: usize) -> Result<()> {
            // SAFETY: The caller must guarantee that `ptr` and `size` are in bounds of the
            // allocation such that no other allocations can be affected. As for this allocation,
            // the only way to access it is by unsafely dererencing its pointer, where the user
            // has the responsibility to make sure that that is valid.
            result(unsafe { VirtualFree(ptr, size, MEM_DECOMMIT) } != 0)
        }

        #[cfg(miri)]
        pub unsafe fn decommit(&self, _ptr: *mut c_void, _size: usize) -> Result<()> {
            // Decommitting memory has no effect on the operational semantics, so there's nothing
            // for Miri to test anyway except hitting a segmentation fault which is perfectly
            // defined behavior.
            Ok(())
        }

        #[cfg(all(not(miri), not(target_vendor = "win7")))]
        pub unsafe fn prefault(&self, ptr: *mut c_void, size: usize) -> Result<()> {
            let entry = WIN32_MEMORY_RANGE_ENTRY {
                VirtualAddress: ptr,
                NumberOfBytes: size,
            };

            // SAFETY: The caller must guarantee that `ptr` and `size` are in bounds of the
            // allocation such that no other allocations can be affected. We are targetting our own
            // process, the pointer points to a valid and initialized memory location above, and
            // the size of 1 is correct as there is only one entry. This call is otherwise purely
            // an optimization hint and can't change program behavior.
            result(unsafe { PrefetchVirtualMemory(GetCurrentProcess(), 1, &entry, 0) } != 0)
        }

        #[cfg(any(miri, target_vendor = "win7"))]
        pub unsafe fn prefault(&self, _ptr: *mut c_void, _size: usize) -> Result<()> {
            Ok(())
        }
    }

    impl Drop for Allocation {
        fn drop(&mut self) {
            if self.size != 0 {
                // SAFETY: It is the responsibility of the user who is unsafely derefercing the
                // allocation's pointer to ensure that those accesses don't happen after the
                // allocation has been dropped. We know that the pointer is valid because we
                // allocated it, and we are passing in 0 as the size as required for `MEM_RELEASE`.
                unsafe { VirtualFree(self.ptr(), 0, MEM_RELEASE) };
            }
        }
    }

    #[inline(always)]
    pub fn page_size() -> usize {
        static PAGE_SIZE: AtomicUsize = AtomicUsize::new(0);

        #[cold]
        fn page_size_slow() -> usize {
            // SAFETY: `SYSTEM_INFO` is composed only of primitive types.
            let mut system_info = unsafe { mem::zeroed() };
            // SAFETY: The pointer points to a valid memory location above.
            unsafe { GetSystemInfo(&mut system_info) };
            let page_size = usize::try_from(system_info.dwPageSize).unwrap();
            PAGE_SIZE.store(page_size, Ordering::Relaxed);

            page_size
        }

        let cached = PAGE_SIZE.load(Ordering::Relaxed);

        if cached != 0 {
            cached
        } else {
            page_size_slow()
        }
    }

    fn result(condition: bool) -> Result<()> {
        if condition {
            Ok(())
        } else {
            Err(super::Error { code: errno() })
        }
    }

    fn errno() -> i32 {
        unsafe { GetLastError() as i32 }
    }

    pub fn format_error(mut errnum: i32, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        let mut buf = [0u16; 2048];
        let mut module = ptr::null_mut();
        let mut flags = 0;

        // NTSTATUS errors may be encoded as HRESULT, which may returned from
        // GetLastError. For more information about Windows error codes, see
        // `[MS-ERREF]`: https://docs.microsoft.com/en-us/openspecs/windows_protocols/ms-erref/0642cb2f-2075-4469-918c-4441e69c548a
        if (errnum & FACILITY_NT_BIT as i32) != 0 {
            // format according to https://support.microsoft.com/en-us/help/259693
            const NTDLL_DLL: &[u16] = &[
                'N' as _, 'T' as _, 'D' as _, 'L' as _, 'L' as _, '.' as _, 'D' as _, 'L' as _,
                'L' as _, 0,
            ];

            module = unsafe { GetModuleHandleW(NTDLL_DLL.as_ptr()) };

            if !module.is_null() {
                errnum ^= FACILITY_NT_BIT as i32;
                flags = FORMAT_MESSAGE_FROM_HMODULE;
            }
        }

        let res = unsafe {
            FormatMessageW(
                flags | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS,
                module,
                errnum as u32,
                0,
                buf.as_mut_ptr(),
                buf.len() as u32,
                ptr::null(),
            ) as usize
        };

        if res == 0 {
            // Sometimes FormatMessageW can fail e.g., system doesn't like 0 as langId,
            let fm_err = errno();
            return write!(
                f,
                "OS Error {errnum} (FormatMessageW() returned error {fm_err})",
            );
        }

        let mut output_len = 0;
        let mut output = [0u8; 2048];

        for c in char::decode_utf16(buf[..res].iter().copied()) {
            let Ok(c) = c else {
                return write!(
                    f,
                    "OS Error {errnum} (FormatMessageW() returned invalid UTF-16)",
                );
            };

            let len = c.len_utf8();

            if len > output.len() - output_len {
                break;
            }

            c.encode_utf8(&mut output[output_len..]);
            output_len += len;
        }

        // SAFETY: The `encode_utf8` calls above were used to encode valid UTF-8.
        let s = unsafe { str::from_utf8_unchecked(&output[..output_len]) };

        f.write_str(s)
    }

    windows_targets::link!("kernel32.dll" "system" fn GetSystemInfo(
        lpSystemInfo: *mut SYSTEM_INFO,
    ));

    windows_targets::link!("kernel32.dll" "system" fn VirtualAlloc(
        lpAddress: *mut c_void,
        dwSize: usize,
        flAllocationType: u32,
        flProtect: u32,
    ) -> *mut c_void);

    windows_targets::link!("kernel32.dll" "system" fn VirtualFree(
        lpAddress: *mut c_void,
        dwSize: usize,
        dwFreeType: u32,
    ) -> i32);

    #[cfg(not(target_vendor = "win7"))]
    windows_targets::link!("kernel32.dll" "system" fn GetCurrentProcess() -> HANDLE);

    #[cfg(not(target_vendor = "win7"))]
    windows_targets::link!("kernel32.dll" "system" fn PrefetchVirtualMemory(
        hProcess: HANDLE,
        NumberOfEntries: usize,
        VirtualAddresses: *const WIN32_MEMORY_RANGE_ENTRY,
        Flags: u32,
    ) -> i32);

    windows_targets::link!("kernel32.dll" "system" fn GetLastError() -> u32);

    windows_targets::link!("kernel32.dll" "system" fn FormatMessageW(
        dwFlags: u32,
        lpSource: *const c_void,
        dwMessageId: u32,
        dwLanguageId: u32,
        lpBuffer: *mut u16,
        nSize: u32,
        arguments: *const *const i8,
    ) -> u32);

    windows_targets::link!("kernel32.dll" "system" fn GetModuleHandleW(
        lpModuleName: *const u16,
    ) -> HMODULE);

    #[repr(C)]
    struct SYSTEM_INFO {
        wProcessorArchitecture: u16,
        wReserved: u16,
        dwPageSize: u32,
        lpMinimumApplicationAddress: *mut c_void,
        lpMaximumApplicationAddress: *mut c_void,
        dwActiveProcessorMask: usize,
        dwNumberOfProcessors: u32,
        dwProcessorType: u32,
        dwAllocationGranularity: u32,
        wProcessorLevel: u16,
        wProcessorRevision: u16,
    }

    const MEM_COMMIT: u32 = 1 << 12;
    const MEM_RESERVE: u32 = 1 << 13;
    const MEM_DECOMMIT: u32 = 1 << 14;
    const MEM_RELEASE: u32 = 1 << 15;

    const PAGE_NOACCESS: u32 = 1 << 0;
    const PAGE_READWRITE: u32 = 1 << 2;

    #[cfg(not(target_vendor = "win7"))]
    type HANDLE = isize;

    #[cfg(not(target_vendor = "win7"))]
    #[repr(C)]
    struct WIN32_MEMORY_RANGE_ENTRY {
        VirtualAddress: *mut c_void,
        NumberOfBytes: usize,
    }

    const FACILITY_NT_BIT: u32 = 1 << 28;

    const FORMAT_MESSAGE_FROM_HMODULE: u32 = 1 << 11;
    const FORMAT_MESSAGE_FROM_SYSTEM: u32 = 1 << 12;
    const FORMAT_MESSAGE_IGNORE_INSERTS: u32 = 1 << 9;

    type HMODULE = *mut c_void;
}

// TODO: Replace this with `<*const u8>::addr` once it's stable.
#[allow(clippy::transmutes_expressible_as_ptr_casts)]
fn addr(ptr: *const u8) -> usize {
    // SAFETY: `*const u8` and `usize` have the same layout.
    unsafe { mem::transmute::<*const u8, usize>(ptr) }
}

// TODO: Replace this with `ptr::without_provenance_mut` once it's stable.
#[allow(clippy::useless_transmute)]
const fn without_provenance_mut(addr: usize) -> *mut u8 {
    // SAFETY: `usize` and `*mut u8` have the same layout.
    unsafe { mem::transmute::<usize, *mut u8>(addr) }
}