src/librustc_middle/mir/interpret/allocation.rs

   1 //! The virtual memory representation of the MIR interpreter.
   2
   3 use std::borrow::Cow;
   4 use std::convert::TryFrom;
   5 use std::iter;
   6 use std::ops::{Deref, DerefMut, Range};
   7
   8 use rustc_ast::ast::Mutability;
   9 use rustc_data_structures::sorted_map::SortedMap;
  10 use rustc_target::abi::{Align, HasDataLayout, Size};
  11
  12 use super::{
  13     read_target_uint, write_target_uint, AllocId, InterpResult, Pointer, Scalar, ScalarMaybeUndef,
  14 };
  15
  16 // NOTE: When adding new fields, make sure to adjust the `Snapshot` impl in
  17 // `src/librustc_mir/interpret/snapshot.rs`.
  18 #[derive(Clone, Debug, Eq, PartialEq, PartialOrd, Ord, Hash, RustcEncodable, RustcDecodable)]
  19 #[derive(HashStable)]
  20 pub struct Allocation<Tag = (), Extra = ()> {
  21     /// The actual bytes of the allocation.
  22     /// Note that the bytes of a pointer represent the offset of the pointer.
  23     bytes: Vec<u8>,
  24     /// Maps from byte addresses to extra data for each pointer.
  25     /// Only the first byte of a pointer is inserted into the map; i.e.,
  26     /// every entry in this map applies to `pointer_size` consecutive bytes starting
  27     /// at the given offset.
  28     relocations: Relocations<Tag>,
  29     /// Denotes which part of this allocation is initialized.
  30     undef_mask: UndefMask,
  31     /// The size of the allocation. Currently, must always equal `bytes.len()`.
  32     pub size: Size,
  33     /// The alignment of the allocation to detect unaligned reads.
  34     /// (`Align` guarantees that this is a power of two.)
  35     pub align: Align,
  36     /// `true` if the allocation is mutable.
  37     /// Also used by codegen to determine if a static should be put into mutable memory,
  38     /// which happens for `static mut` and `static` with interior mutability.
  39     pub mutability: Mutability,
  40     /// Extra state for the machine.
  41     pub extra: Extra,
  42 }
  43
  44 pub trait AllocationExtra<Tag>: ::std::fmt::Debug + Clone {
  45     // There is no constructor in here because the constructor's type depends
  46     // on `MemoryKind`, and making things sufficiently generic leads to painful
  47     // inference failure.
  48
  49     /// Hook for performing extra checks on a memory read access.
  50     ///
  51     /// Takes read-only access to the allocation so we can keep all the memory read
  52     /// operations take `&self`. Use a `RefCell` in `AllocExtra` if you
  53     /// need to mutate.
  54     #[inline(always)]
  55     fn memory_read(
  56         _alloc: &Allocation<Tag, Self>,
  57         _ptr: Pointer<Tag>,
  58         _size: Size,
  59     ) -> InterpResult<'tcx> {
  60         Ok(())
  61     }
  62
  63     /// Hook for performing extra checks on a memory write access.
  64     #[inline(always)]
  65     fn memory_written(
  66         _alloc: &mut Allocation<Tag, Self>,
  67         _ptr: Pointer<Tag>,
  68         _size: Size,
  69     ) -> InterpResult<'tcx> {
  70         Ok(())
  71     }
  72
  73     /// Hook for performing extra checks on a memory deallocation.
  74     /// `size` will be the size of the allocation.
  75     #[inline(always)]
  76     fn memory_deallocated(
  77         _alloc: &mut Allocation<Tag, Self>,
  78         _ptr: Pointer<Tag>,
  79         _size: Size,
  80     ) -> InterpResult<'tcx> {
  81         Ok(())
  82     }
  83 }
  84
  85 // For `Tag = ()` and no extra state, we have a trivial implementation.
  86 impl AllocationExtra<()> for () {}
  87
  88 // The constructors are all without extra; the extra gets added by a machine hook later.
  89 impl<Tag> Allocation<Tag> {
  90     /// Creates a read-only allocation initialized by the given bytes
  91     pub fn from_bytes<'a>(slice: impl Into<Cow<'a, [u8]>>, align: Align) -> Self {
  92         let bytes = slice.into().into_owned();
  93         let size = Size::from_bytes(bytes.len());
  94         Self {
  95             bytes,
  96             relocations: Relocations::new(),
  97             undef_mask: UndefMask::new(size, true),
  98             size,
  99             align,
 100             mutability: Mutability::Not,
 101             extra: (),
 102         }
 103     }
 104
 105     pub fn from_byte_aligned_bytes<'a>(slice: impl Into<Cow<'a, [u8]>>) -> Self {
 106         Allocation::from_bytes(slice, Align::from_bytes(1).unwrap())
 107     }
 108
 109     pub fn undef(size: Size, align: Align) -> Self {
 110         Allocation {
 111             bytes: vec![0; size.bytes_usize()],
 112             relocations: Relocations::new(),
 113             undef_mask: UndefMask::new(size, false),
 114             size,
 115             align,
 116             mutability: Mutability::Mut,
 117             extra: (),
 118         }
 119     }
 120 }
 121
 122 impl Allocation<(), ()> {
 123     /// Add Tag and Extra fields
 124     pub fn with_tags_and_extra<T, E>(
 125         self,
 126         mut tagger: impl FnMut(AllocId) -> T,
 127         extra: E,
 128     ) -> Allocation<T, E> {
 129         Allocation {
 130             bytes: self.bytes,
 131             size: self.size,
 132             relocations: Relocations::from_presorted(
 133                 self.relocations
 134                     .iter()
 135                     // The allocations in the relocations (pointers stored *inside* this allocation)
 136                     // all get the base pointer tag.
 137                     .map(|&(offset, ((), alloc))| {
 138                         let tag = tagger(alloc);
 139                         (offset, (tag, alloc))
 140                     })
 141                     .collect(),
 142             ),
 143             undef_mask: self.undef_mask,
 144             align: self.align,
 145             mutability: self.mutability,
 146             extra,
 147         }
 148     }
 149 }
 150
 151 /// Raw accessors. Provide access to otherwise private bytes.
 152 impl<Tag, Extra> Allocation<Tag, Extra> {
 153     pub fn len(&self) -> usize {
 154         self.size.bytes_usize()
 155     }
 156
 157     /// Looks at a slice which may describe undefined bytes or describe a relocation. This differs
 158     /// from `get_bytes_with_undef_and_ptr` in that it does no relocation checks (even on the
 159     /// edges) at all. It further ignores `AllocationExtra` callbacks.
 160     /// This must not be used for reads affecting the interpreter execution.
 161     pub fn inspect_with_undef_and_ptr_outside_interpreter(&self, range: Range<usize>) -> &[u8] {
 162         &self.bytes[range]
 163     }
 164
 165     /// Returns the undef mask.
 166     pub fn undef_mask(&self) -> &UndefMask {
 167         &self.undef_mask
 168     }
 169
 170     /// Returns the relocation list.
 171     pub fn relocations(&self) -> &Relocations<Tag> {
 172         &self.relocations
 173     }
 174 }
 175
 176 impl<'tcx> rustc_serialize::UseSpecializedDecodable for &'tcx Allocation {}
 177
 178 /// Byte accessors.
 179 impl<'tcx, Tag: Copy, Extra: AllocationExtra<Tag>> Allocation<Tag, Extra> {
 180     /// Just a small local helper function to avoid a bit of code repetition.
 181     /// Returns the range of this allocation that was meant.
 182     #[inline]
 183     fn check_bounds(&self, offset: Size, size: Size) -> Range<usize> {
 184         let end = offset + size; // This does overflow checking.
 185         let end = usize::try_from(end.bytes()).expect("access too big for this host architecture");
 186         assert!(
 187             end <= self.len(),
 188             "Out-of-bounds access at offset {}, size {} in allocation of size {}",
 189             offset.bytes(),
 190             size.bytes(),
 191             self.len()
 192         );
 193         offset.bytes_usize()..end
 194     }
 195
 196     /// The last argument controls whether we error out when there are undefined
 197     /// or pointer bytes. You should never call this, call `get_bytes` or
 198     /// `get_bytes_with_undef_and_ptr` instead,
 199     ///
 200     /// This function also guarantees that the resulting pointer will remain stable
 201     /// even when new allocations are pushed to the `HashMap`. `copy_repeatedly` relies
 202     /// on that.
 203     ///
 204     /// It is the caller's responsibility to check bounds and alignment beforehand.
 205     fn get_bytes_internal(
 206         &self,
 207         cx: &impl HasDataLayout,
 208         ptr: Pointer<Tag>,
 209         size: Size,
 210         check_defined_and_ptr: bool,
 211     ) -> InterpResult<'tcx, &[u8]> {
 212         let range = self.check_bounds(ptr.offset, size);
 213
 214         if check_defined_and_ptr {
 215             self.check_defined(ptr, size)?;
 216             self.check_relocations(cx, ptr, size)?;
 217         } else {
 218             // We still don't want relocations on the *edges*.
 219             self.check_relocation_edges(cx, ptr, size)?;
 220         }
 221
 222         AllocationExtra::memory_read(self, ptr, size)?;
 223
 224         Ok(&self.bytes[range])
 225     }
 226
 227     /// Checks that these bytes are initialized and not pointer bytes, and then return them
 228     /// as a slice.
 229     ///
 230     /// It is the caller's responsibility to check bounds and alignment beforehand.
 231     /// Most likely, you want to use the `PlaceTy` and `OperandTy`-based methods
 232     /// on `InterpCx` instead.
 233     #[inline]
 234     pub fn get_bytes(
 235         &self,
 236         cx: &impl HasDataLayout,
 237         ptr: Pointer<Tag>,
 238         size: Size,
 239     ) -> InterpResult<'tcx, &[u8]> {
 240         self.get_bytes_internal(cx, ptr, size, true)
 241     }
 242
 243     /// It is the caller's responsibility to handle undefined and pointer bytes.
 244     /// However, this still checks that there are no relocations on the *edges*.
 245     ///
 246     /// It is the caller's responsibility to check bounds and alignment beforehand.
 247     #[inline]
 248     pub fn get_bytes_with_undef_and_ptr(
 249         &self,
 250         cx: &impl HasDataLayout,
 251         ptr: Pointer<Tag>,
 252         size: Size,
 253     ) -> InterpResult<'tcx, &[u8]> {
 254         self.get_bytes_internal(cx, ptr, size, false)
 255     }
 256
 257     /// Just calling this already marks everything as defined and removes relocations,
 258     /// so be sure to actually put data there!
 259     ///
 260     /// It is the caller's responsibility to check bounds and alignment beforehand.
 261     /// Most likely, you want to use the `PlaceTy` and `OperandTy`-based methods
 262     /// on `InterpCx` instead.
 263     pub fn get_bytes_mut(
 264         &mut self,
 265         cx: &impl HasDataLayout,
 266         ptr: Pointer<Tag>,
 267         size: Size,
 268     ) -> InterpResult<'tcx, &mut [u8]> {
 269         let range = self.check_bounds(ptr.offset, size);
 270
 271         self.mark_definedness(ptr, size, true);
 272         self.clear_relocations(cx, ptr, size)?;
 273
 274         AllocationExtra::memory_written(self, ptr, size)?;
 275
 276         Ok(&mut self.bytes[range])
 277     }
 278 }
 279
 280 /// Reading and writing.
 281 impl<'tcx, Tag: Copy, Extra: AllocationExtra<Tag>> Allocation<Tag, Extra> {
 282     /// Reads bytes until a `0` is encountered. Will error if the end of the allocation is reached
 283     /// before a `0` is found.
 284     ///
 285     /// Most likely, you want to call `Memory::read_c_str` instead of this method.
 286     pub fn read_c_str(
 287         &self,
 288         cx: &impl HasDataLayout,
 289         ptr: Pointer<Tag>,
 290     ) -> InterpResult<'tcx, &[u8]> {
 291         let offset = ptr.offset.bytes_usize();
 292         Ok(match self.bytes[offset..].iter().position(|&c| c == 0) {
 293             Some(size) => {
 294                 let size_with_null = Size::from_bytes(size) + Size::from_bytes(1);
 295                 // Go through `get_bytes` for checks and AllocationExtra hooks.
 296                 // We read the null, so we include it in the request, but we want it removed
 297                 // from the result, so we do subslicing.
 298                 &self.get_bytes(cx, ptr, size_with_null)?[..size]
 299             }
 300             // This includes the case where `offset` is out-of-bounds to begin with.
 301             None => throw_ub!(UnterminatedCString(ptr.erase_tag())),
 302         })
 303     }
 304
 305     /// Validates that `ptr.offset` and `ptr.offset + size` do not point to the middle of a
 306     /// relocation. If `allow_ptr_and_undef` is `false`, also enforces that the memory in the
 307     /// given range contains neither relocations nor undef bytes.
 308     pub fn check_bytes(
 309         &self,
 310         cx: &impl HasDataLayout,
 311         ptr: Pointer<Tag>,
 312         size: Size,
 313         allow_ptr_and_undef: bool,
 314     ) -> InterpResult<'tcx> {
 315         // Check bounds and relocations on the edges.
 316         self.get_bytes_with_undef_and_ptr(cx, ptr, size)?;
 317         // Check undef and ptr.
 318         if !allow_ptr_and_undef {
 319             self.check_defined(ptr, size)?;
 320             self.check_relocations(cx, ptr, size)?;
 321         }
 322         Ok(())
 323     }
 324
 325     /// Writes `src` to the memory starting at `ptr.offset`.
 326     ///
 327     /// It is the caller's responsibility to check bounds and alignment beforehand.
 328     /// Most likely, you want to call `Memory::write_bytes` instead of this method.
 329     pub fn write_bytes(
 330         &mut self,
 331         cx: &impl HasDataLayout,
 332         ptr: Pointer<Tag>,
 333         src: impl IntoIterator<Item = u8>,
 334     ) -> InterpResult<'tcx> {
 335         let mut src = src.into_iter();
 336         let (lower, upper) = src.size_hint();
 337         let len = upper.expect("can only write bounded iterators");
 338         assert_eq!(lower, len, "can only write iterators with a precise length");
 339         let bytes = self.get_bytes_mut(cx, ptr, Size::from_bytes(len))?;
 340         // `zip` would stop when the first iterator ends; we want to definitely
 341         // cover all of `bytes`.
 342         for dest in bytes {
 343             *dest = src.next().expect("iterator was shorter than it said it would be");
 344         }
 345         src.next().expect_none("iterator was longer than it said it would be");
 346         Ok(())
 347     }
 348
 349     /// Reads a *non-ZST* scalar.
 350     ///
 351     /// ZSTs can't be read for two reasons:
 352     /// * byte-order cannot work with zero-element buffers;
 353     /// * in order to obtain a `Pointer`, we need to check for ZSTness anyway due to integer
 354     ///   pointers being valid for ZSTs.
 355     ///
 356     /// It is the caller's responsibility to check bounds and alignment beforehand.
 357     /// Most likely, you want to call `InterpCx::read_scalar` instead of this method.
 358     pub fn read_scalar(
 359         &self,
 360         cx: &impl HasDataLayout,
 361         ptr: Pointer<Tag>,
 362         size: Size,
 363     ) -> InterpResult<'tcx, ScalarMaybeUndef<Tag>> {
 364         // `get_bytes_unchecked` tests relocation edges.
 365         let bytes = self.get_bytes_with_undef_and_ptr(cx, ptr, size)?;
 366         // Undef check happens *after* we established that the alignment is correct.
 367         // We must not return `Ok()` for unaligned pointers!
 368         if self.is_defined(ptr, size).is_err() {
 369             // This inflates undefined bytes to the entire scalar, even if only a few
 370             // bytes are undefined.
 371             return Ok(ScalarMaybeUndef::Undef);
 372         }
 373         // Now we do the actual reading.
 374         let bits = read_target_uint(cx.data_layout().endian, bytes).unwrap();
 375         // See if we got a pointer.
 376         if size != cx.data_layout().pointer_size {
 377             // *Now*, we better make sure that the inside is free of relocations too.
 378             self.check_relocations(cx, ptr, size)?;
 379         } else {
 380             if let Some(&(tag, alloc_id)) = self.relocations.get(&ptr.offset) {
 381                 let ptr = Pointer::new_with_tag(alloc_id, Size::from_bytes(bits), tag);
 382                 return Ok(ScalarMaybeUndef::Scalar(ptr.into()));
 383             }
 384         }
 385         // We don't. Just return the bits.
 386         Ok(ScalarMaybeUndef::Scalar(Scalar::from_uint(bits, size)))
 387     }
 388
 389     /// Reads a pointer-sized scalar.
 390     ///
 391     /// It is the caller's responsibility to check bounds and alignment beforehand.
 392     /// Most likely, you want to call `InterpCx::read_scalar` instead of this method.
 393     pub fn read_ptr_sized(
 394         &self,
 395         cx: &impl HasDataLayout,
 396         ptr: Pointer<Tag>,
 397     ) -> InterpResult<'tcx, ScalarMaybeUndef<Tag>> {
 398         self.read_scalar(cx, ptr, cx.data_layout().pointer_size)
 399     }
 400
 401     /// Writes a *non-ZST* scalar.
 402     ///
 403     /// ZSTs can't be read for two reasons:
 404     /// * byte-order cannot work with zero-element buffers;
 405     /// * in order to obtain a `Pointer`, we need to check for ZSTness anyway due to integer
 406     ///   pointers being valid for ZSTs.
 407     ///
 408     /// It is the caller's responsibility to check bounds and alignment beforehand.
 409     /// Most likely, you want to call `InterpCx::write_scalar` instead of this method.
 410     pub fn write_scalar(
 411         &mut self,
 412         cx: &impl HasDataLayout,
 413         ptr: Pointer<Tag>,
 414         val: ScalarMaybeUndef<Tag>,
 415         type_size: Size,
 416     ) -> InterpResult<'tcx> {
 417         let val = match val {
 418             ScalarMaybeUndef::Scalar(scalar) => scalar,
 419             ScalarMaybeUndef::Undef => {
 420                 self.mark_definedness(ptr, type_size, false);
 421                 return Ok(());
 422             }
 423         };
 424
 425         let bytes = match val.to_bits_or_ptr(type_size, cx) {
 426             Err(val) => u128::from(val.offset.bytes()),
 427             Ok(data) => data,
 428         };
 429
 430         let endian = cx.data_layout().endian;
 431         let dst = self.get_bytes_mut(cx, ptr, type_size)?;
 432         write_target_uint(endian, dst, bytes).unwrap();
 433
 434         // See if we have to also write a relocation.
 435         if let Scalar::Ptr(val) = val {
 436             self.relocations.insert(ptr.offset, (val.tag, val.alloc_id));
 437         }
 438
 439         Ok(())
 440     }
 441
 442     /// Writes a pointer-sized scalar.
 443     ///
 444     /// It is the caller's responsibility to check bounds and alignment beforehand.
 445     /// Most likely, you want to call `InterpCx::write_scalar` instead of this method.
 446     pub fn write_ptr_sized(
 447         &mut self,
 448         cx: &impl HasDataLayout,
 449         ptr: Pointer<Tag>,
 450         val: ScalarMaybeUndef<Tag>,
 451     ) -> InterpResult<'tcx> {
 452         let ptr_size = cx.data_layout().pointer_size;
 453         self.write_scalar(cx, ptr, val, ptr_size)
 454     }
 455 }
 456
 457 /// Relocations.
 458 impl<'tcx, Tag: Copy, Extra> Allocation<Tag, Extra> {
 459     /// Returns all relocations overlapping with the given pointer-offset pair.
 460     pub fn get_relocations(
 461         &self,
 462         cx: &impl HasDataLayout,
 463         ptr: Pointer<Tag>,
 464         size: Size,
 465     ) -> &[(Size, (Tag, AllocId))] {
 466         // We have to go back `pointer_size - 1` bytes, as that one would still overlap with
 467         // the beginning of this range.
 468         let start = ptr.offset.bytes().saturating_sub(cx.data_layout().pointer_size.bytes() - 1);
 469         let end = ptr.offset + size; // This does overflow checking.
 470         self.relocations.range(Size::from_bytes(start)..end)
 471     }
 472
 473     /// Checks that there are no relocations overlapping with the given range.
 474     #[inline(always)]
 475     fn check_relocations(
 476         &self,
 477         cx: &impl HasDataLayout,
 478         ptr: Pointer<Tag>,
 479         size: Size,
 480     ) -> InterpResult<'tcx> {
 481         if self.get_relocations(cx, ptr, size).is_empty() {
 482             Ok(())
 483         } else {
 484             throw_unsup!(ReadPointerAsBytes)
 485         }
 486     }
 487
 488     /// Removes all relocations inside the given range.
 489     /// If there are relocations overlapping with the edges, they
 490     /// are removed as well *and* the bytes they cover are marked as
 491     /// uninitialized. This is a somewhat odd "spooky action at a distance",
 492     /// but it allows strictly more code to run than if we would just error
 493     /// immediately in that case.
 494     fn clear_relocations(
 495         &mut self,
 496         cx: &impl HasDataLayout,
 497         ptr: Pointer<Tag>,
 498         size: Size,
 499     ) -> InterpResult<'tcx> {
 500         // Find the start and end of the given range and its outermost relocations.
 501         let (first, last) = {
 502             // Find all relocations overlapping the given range.
 503             let relocations = self.get_relocations(cx, ptr, size);
 504             if relocations.is_empty() {
 505                 return Ok(());
 506             }
 507
 508             (
 509                 relocations.first().unwrap().0,
 510                 relocations.last().unwrap().0 + cx.data_layout().pointer_size,
 511             )
 512         };
 513         let start = ptr.offset;
 514         let end = start + size; // `Size` addition
 515
 516         // Mark parts of the outermost relocations as undefined if they partially fall outside the
 517         // given range.
 518         if first < start {
 519             self.undef_mask.set_range(first, start, false);
 520         }
 521         if last > end {
 522             self.undef_mask.set_range(end, last, false);
 523         }
 524
 525         // Forget all the relocations.
 526         self.relocations.remove_range(first..last);
 527
 528         Ok(())
 529     }
 530
 531     /// Errors if there are relocations overlapping with the edges of the
 532     /// given memory range.
 533     #[inline]
 534     fn check_relocation_edges(
 535         &self,
 536         cx: &impl HasDataLayout,
 537         ptr: Pointer<Tag>,
 538         size: Size,
 539     ) -> InterpResult<'tcx> {
 540         self.check_relocations(cx, ptr, Size::ZERO)?;
 541         self.check_relocations(cx, ptr.offset(size, cx)?, Size::ZERO)?;
 542         Ok(())
 543     }
 544 }
 545
 546 /// Undefined bytes.
 547 impl<'tcx, Tag: Copy, Extra> Allocation<Tag, Extra> {
 548     /// Checks whether the given range  is entirely defined.
 549     ///
 550     /// Returns `Ok(())` if it's defined. Otherwise returns the index of the byte
 551     /// at which the first undefined access begins.
 552     fn is_defined(&self, ptr: Pointer<Tag>, size: Size) -> Result<(), Size> {
 553         self.undef_mask.is_range_defined(ptr.offset, ptr.offset + size) // `Size` addition
 554     }
 555
 556     /// Checks that a range of bytes is defined. If not, returns the `ReadUndefBytes`
 557     /// error which will report the first byte which is undefined.
 558     fn check_defined(&self, ptr: Pointer<Tag>, size: Size) -> InterpResult<'tcx> {
 559         self.is_defined(ptr, size)
 560             .or_else(|idx| throw_ub!(InvalidUndefBytes(Some(Pointer::new(ptr.alloc_id, idx)))))
 561     }
 562
 563     pub fn mark_definedness(&mut self, ptr: Pointer<Tag>, size: Size, new_state: bool) {
 564         if size.bytes() == 0 {
 565             return;
 566         }
 567         self.undef_mask.set_range(ptr.offset, ptr.offset + size, new_state);
 568     }
 569 }
 570
 571 /// Run-length encoding of the undef mask.
 572 /// Used to copy parts of a mask multiple times to another allocation.
 573 pub struct AllocationDefinedness {
 574     /// The definedness of the first range.
 575     initial: bool,
 576     /// The lengths of ranges that are run-length encoded.
 577     /// The definedness of the ranges alternate starting with `initial`.
 578     ranges: smallvec::SmallVec<[u64; 1]>,
 579 }
 580
 581 impl AllocationDefinedness {
 582     pub fn all_bytes_undef(&self) -> bool {
 583         // The `ranges` are run-length encoded and of alternating definedness.
 584         // So if `ranges.len() > 1` then the second block is a range of defined.
 585         !self.initial && self.ranges.len() == 1
 586     }
 587 }
 588
 589 /// Transferring the definedness mask to other allocations.
 590 impl<Tag, Extra> Allocation<Tag, Extra> {
 591     /// Creates a run-length encoding of the undef mask.
 592     pub fn compress_undef_range(&self, src: Pointer<Tag>, size: Size) -> AllocationDefinedness {
 593         // Since we are copying `size` bytes from `src` to `dest + i * size` (`for i in 0..repeat`),
 594         // a naive undef mask copying algorithm would repeatedly have to read the undef mask from
 595         // the source and write it to the destination. Even if we optimized the memory accesses,
 596         // we'd be doing all of this `repeat` times.
 597         // Therefore we precompute a compressed version of the undef mask of the source value and
 598         // then write it back `repeat` times without computing any more information from the source.
 599
 600         // A precomputed cache for ranges of defined/undefined bits
 601         // 0000010010001110 will become
 602         // `[5, 1, 2, 1, 3, 3, 1]`,
 603         // where each element toggles the state.
 604
 605         let mut ranges = smallvec::SmallVec::<[u64; 1]>::new();
 606         let initial = self.undef_mask.get(src.offset);
 607         let mut cur_len = 1;
 608         let mut cur = initial;
 609
 610         for i in 1..size.bytes() {
 611             // FIXME: optimize to bitshift the current undef block's bits and read the top bit.
 612             if self.undef_mask.get(src.offset + Size::from_bytes(i)) == cur {
 613                 cur_len += 1;
 614             } else {
 615                 ranges.push(cur_len);
 616                 cur_len = 1;
 617                 cur = !cur;
 618             }
 619         }
 620
 621         ranges.push(cur_len);
 622
 623         AllocationDefinedness { ranges, initial }
 624     }
 625
 626     /// Applies multiple instances of the run-length encoding to the undef mask.
 627     pub fn mark_compressed_undef_range(
 628         &mut self,
 629         defined: &AllocationDefinedness,
 630         dest: Pointer<Tag>,
 631         size: Size,
 632         repeat: u64,
 633     ) {
 634         // An optimization where we can just overwrite an entire range of definedness bits if
 635         // they are going to be uniformly `1` or `0`.
 636         if defined.ranges.len() <= 1 {
 637             self.undef_mask.set_range_inbounds(
 638                 dest.offset,
 639                 dest.offset + size * repeat, // `Size` operations
 640                 defined.initial,
 641             );
 642             return;
 643         }
 644
 645         for mut j in 0..repeat {
 646             j *= size.bytes();
 647             j += dest.offset.bytes();
 648             let mut cur = defined.initial;
 649             for range in &defined.ranges {
 650                 let old_j = j;
 651                 j += range;
 652                 self.undef_mask.set_range_inbounds(
 653                     Size::from_bytes(old_j),
 654                     Size::from_bytes(j),
 655                     cur,
 656                 );
 657                 cur = !cur;
 658             }
 659         }
 660     }
 661 }
 662
 663 /// Relocations.
 664 #[derive(Clone, PartialEq, Eq, PartialOrd, Ord, Hash, Debug, RustcEncodable, RustcDecodable)]
 665 pub struct Relocations<Tag = (), Id = AllocId>(SortedMap<Size, (Tag, Id)>);
 666
 667 impl<Tag, Id> Relocations<Tag, Id> {
 668     pub fn new() -> Self {
 669         Relocations(SortedMap::new())
 670     }
 671
 672     // The caller must guarantee that the given relocations are already sorted
 673     // by address and contain no duplicates.
 674     pub fn from_presorted(r: Vec<(Size, (Tag, Id))>) -> Self {
 675         Relocations(SortedMap::from_presorted_elements(r))
 676     }
 677 }
 678
 679 impl<Tag> Deref for Relocations<Tag> {
 680     type Target = SortedMap<Size, (Tag, AllocId)>;
 681
 682     fn deref(&self) -> &Self::Target {
 683         &self.0
 684     }
 685 }
 686
 687 impl<Tag> DerefMut for Relocations<Tag> {
 688     fn deref_mut(&mut self) -> &mut Self::Target {
 689         &mut self.0
 690     }
 691 }
 692
 693 /// A partial, owned list of relocations to transfer into another allocation.
 694 pub struct AllocationRelocations<Tag> {
 695     relative_relocations: Vec<(Size, (Tag, AllocId))>,
 696 }
 697
 698 impl<Tag: Copy, Extra> Allocation<Tag, Extra> {
 699     pub fn prepare_relocation_copy(
 700         &self,
 701         cx: &impl HasDataLayout,
 702         src: Pointer<Tag>,
 703         size: Size,
 704         dest: Pointer<Tag>,
 705         length: u64,
 706     ) -> AllocationRelocations<Tag> {
 707         let relocations = self.get_relocations(cx, src, size);
 708         if relocations.is_empty() {
 709             return AllocationRelocations { relative_relocations: Vec::new() };
 710         }
 711
 712         let mut new_relocations = Vec::with_capacity(relocations.len() * (length as usize));
 713
 714         for i in 0..length {
 715             new_relocations.extend(relocations.iter().map(|&(offset, reloc)| {
 716                 // compute offset for current repetition
 717                 let dest_offset = dest.offset + size * i; // `Size` operations
 718                 (
 719                     // shift offsets from source allocation to destination allocation
 720                     (offset + dest_offset) - src.offset, // `Size` operations
 721                     reloc,
 722                 )
 723             }));
 724         }
 725
 726         AllocationRelocations { relative_relocations: new_relocations }
 727     }
 728
 729     /// Applies a relocation copy.
 730     /// The affected range, as defined in the parameters to `prepare_relocation_copy` is expected
 731     /// to be clear of relocations.
 732     pub fn mark_relocation_range(&mut self, relocations: AllocationRelocations<Tag>) {
 733         self.relocations.insert_presorted(relocations.relative_relocations);
 734     }
 735 }
 736
 737 ////////////////////////////////////////////////////////////////////////////////
 738 // Undefined byte tracking
 739 ////////////////////////////////////////////////////////////////////////////////
 740
 741 type Block = u64;
 742
 743 /// A bitmask where each bit refers to the byte with the same index. If the bit is `true`, the byte
 744 /// is defined. If it is `false` the byte is undefined.
 745 #[derive(Clone, Debug, Eq, PartialEq, PartialOrd, Ord, Hash, RustcEncodable, RustcDecodable)]
 746 #[derive(HashStable)]
 747 pub struct UndefMask {
 748     blocks: Vec<Block>,
 749     len: Size,
 750 }
 751
 752 impl UndefMask {
 753     pub const BLOCK_SIZE: u64 = 64;
 754
 755     pub fn new(size: Size, state: bool) -> Self {
 756         let mut m = UndefMask { blocks: vec![], len: Size::ZERO };
 757         m.grow(size, state);
 758         m
 759     }
 760
 761     /// Checks whether the range `start..end` (end-exclusive) is entirely defined.
 762     ///
 763     /// Returns `Ok(())` if it's defined. Otherwise returns the index of the byte
 764     /// at which the first undefined access begins.
 765     #[inline]
 766     pub fn is_range_defined(&self, start: Size, end: Size) -> Result<(), Size> {
 767         if end > self.len {
 768             return Err(self.len);
 769         }
 770
 771         // FIXME(oli-obk): optimize this for allocations larger than a block.
 772         let idx = (start.bytes()..end.bytes()).map(Size::from_bytes).find(|&i| !self.get(i));
 773
 774         match idx {
 775             Some(idx) => Err(idx),
 776             None => Ok(()),
 777         }
 778     }
 779
 780     pub fn set_range(&mut self, start: Size, end: Size, new_state: bool) {
 781         let len = self.len;
 782         if end > len {
 783             self.grow(end - len, new_state);
 784         }
 785         self.set_range_inbounds(start, end, new_state);
 786     }
 787
 788     pub fn set_range_inbounds(&mut self, start: Size, end: Size, new_state: bool) {
 789         let (blocka, bita) = bit_index(start);
 790         let (blockb, bitb) = bit_index(end);
 791         if blocka == blockb {
 792             // First set all bits except the first `bita`,
 793             // then unset the last `64 - bitb` bits.
 794             let range = if bitb == 0 {
 795                 u64::MAX << bita
 796             } else {
 797                 (u64::MAX << bita) & (u64::MAX >> (64 - bitb))
 798             };
 799             if new_state {
 800                 self.blocks[blocka] |= range;
 801             } else {
 802                 self.blocks[blocka] &= !range;
 803             }
 804             return;
 805         }
 806         // across block boundaries
 807         if new_state {
 808             // Set `bita..64` to `1`.
 809             self.blocks[blocka] |= u64::MAX << bita;
 810             // Set `0..bitb` to `1`.
 811             if bitb != 0 {
 812                 self.blocks[blockb] |= u64::MAX >> (64 - bitb);
 813             }
 814             // Fill in all the other blocks (much faster than one bit at a time).
 815             for block in (blocka + 1)..blockb {
 816                 self.blocks[block] = u64::MAX;
 817             }
 818         } else {
 819             // Set `bita..64` to `0`.
 820             self.blocks[blocka] &= !(u64::MAX << bita);
 821             // Set `0..bitb` to `0`.
 822             if bitb != 0 {
 823                 self.blocks[blockb] &= !(u64::MAX >> (64 - bitb));
 824             }
 825             // Fill in all the other blocks (much faster than one bit at a time).
 826             for block in (blocka + 1)..blockb {
 827                 self.blocks[block] = 0;
 828             }
 829         }
 830     }
 831
 832     #[inline]
 833     pub fn get(&self, i: Size) -> bool {
 834         let (block, bit) = bit_index(i);
 835         (self.blocks[block] & (1 << bit)) != 0
 836     }
 837
 838     #[inline]
 839     pub fn set(&mut self, i: Size, new_state: bool) {
 840         let (block, bit) = bit_index(i);
 841         self.set_bit(block, bit, new_state);
 842     }
 843
 844     #[inline]
 845     fn set_bit(&mut self, block: usize, bit: usize, new_state: bool) {
 846         if new_state {
 847             self.blocks[block] |= 1 << bit;
 848         } else {
 849             self.blocks[block] &= !(1 << bit);
 850         }
 851     }
 852
 853     pub fn grow(&mut self, amount: Size, new_state: bool) {
 854         if amount.bytes() == 0 {
 855             return;
 856         }
 857         let unused_trailing_bits =
 858             u64::try_from(self.blocks.len()).unwrap() * Self::BLOCK_SIZE - self.len.bytes();
 859         if amount.bytes() > unused_trailing_bits {
 860             let additional_blocks = amount.bytes() / Self::BLOCK_SIZE + 1;
 861             self.blocks.extend(
 862                 // FIXME(oli-obk): optimize this by repeating `new_state as Block`.
 863                 iter::repeat(0).take(usize::try_from(additional_blocks).unwrap()),
 864             );
 865         }
 866         let start = self.len;
 867         self.len += amount;
 868         self.set_range_inbounds(start, start + amount, new_state); // `Size` operation
 869     }
 870 }
 871
 872 #[inline]
 873 fn bit_index(bits: Size) -> (usize, usize) {
 874     let bits = bits.bytes();
 875     let a = bits / UndefMask::BLOCK_SIZE;
 876     let b = bits % UndefMask::BLOCK_SIZE;
 877     (usize::try_from(a).unwrap(), usize::try_from(b).unwrap())
 878 }