Merge tag 'trace-v7.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-tracegrafted

Pull tracing fixes from Steven Rostedt:

 - Don't destroy user event fields when removal fails

   User event fields are destroyed before the event is removed from
   visibility. But that can fail leaving the still visible event with no
   fields. Move the destroying of the fields to after the event is
   successfully removed from visibility.

 - Initialize function graph state is fork before calling
   copy_exec_state()

   For non-CLONE_VM forks, copy_exec_state() allocates a new
   task_exec_state. If that allocation fails, ftrace_graph_exit_task()
   will free the tasks ret_stack pointer. Since that pointer is still
   using the parent's ret_stack, it mistakenly frees the parent's
   pointer too.

   Call ftrace_graph_init() on the task first which will NULL out the
   new tasks's ret_stack and if the copy fails, it will not free
   anything.

 - Remove FGRAPH_MAX_INDEX

   The macro FGRAPH_MAX_INDEX was added but never used. Remove it.

 - Save ent_size in function graph printing of nested functions

   The function graph tracer needs to look at the next event to see if
   the next event is the return of the current function entry. If it is,
   it prints a single line:

	ktime_get();

   Otherwise it prints it like a nested function:

	tick_nohz_irq_exit() {
	    ktime_get();
	    kcpustat_irq_exit();
	}

   In order to look at the next event, it must save the current event so
   that it has the information to print from it. It saves the event in
   the iterator descriptor called "ent". What it doesn't save is the
   ent_size of the event which is now used to know if the function graph
   arguments are to be printed. The peek doesn't save the size so the
   size used happens to be that of the size of the last event that was
   seen.

   Save the entry event size in the iterator descriptor so that the
   correct size is used.

 - Fix several errors with freeing data in the histogram code

   The histogram code had a lot of leaked or or incorrect accounting
   when failures happen. Correct them.

 - Fix histogram regression of .percent and .graph modifiers

   Up until 6.3 histogram values could have "percent" or "graph"
   modifiers that changed how they were printed. But a change that added
   restricting histograms values from being strings, stack traces and
   other modifiers inadvertently prevented them from using the percent
   and graph modifiers, which were legal use cases for values.

   Put back the percent and graph modifiers.

 - Fix various typos in the comments

 - Set the trace_clock before initializing a histogram with clock
   argument

   The histogram API allows the user to specific which trace clock to
   use via a "clock=" string. The histogram is set up first before the
   clock is checked. If the passed in clock is not valid, it exits
   without fully fixing up the histogram leaving it on the list and a
   use-after-free can trigger.

   Update the clock argument first and if it fails then exit gracefully
   before the histogram trigger is placed on any lists.

 - Restore :mod: trailer after parsing in ftrace_set_clr_event

   The function ftrace_set_clr_event() modifies the parse string and
   needs to put it back to what was passed in. It searches for ":mod:"
   via a strsep() but fails to put back the first ':' in the string.

   Add back the ':' in the passed in string.

 - Take trace_array reference when opening a tracer options file

   The options files are dynamically created and some tracers add their
   own options. When a tracer adds their own list of options, the
   trace_array holding them has an array to hold the list of options for
   each tracer. This array increases in size via a krealloc(), and the
   new entry gets a newly allocated array to hold the options of the new
   tracer being added.

   The element in each entry of the tracer's option array holds a
   pointer back to the trace_array, a pointer to the tracer it is
   associated to, a pointer to the flags of the option.

   The issue is that these arrays are freed when the trace_array is
   freed when its instance it represents is removed from the instances
   directory. There's a race that an open of one of these options files
   can happen when the instance is being removed.

   Add a new helper function to be called by the open function of the
   options file to iterate all existing trace_arrays under a lock and
   find the one that has the given option element in one of it's tracer
   arrays. If found, then update the associated trace_array's reference
   counter to keep it from being freed. If not found, have the open call
   return -ENODEV.

 - Disable interrupts when acquiring the lock in rb_wake_up_waiters()

   The function rb_wake_up_waiters() assumes it will be called in
   interrupt context and does not disable irqs when taking
   cpu_buffer->reader_lock, which can be called in hard interrupt
   context. The issue is in PREEMPT_RT, this function is called in
   thread context leaving this lock open to a deadlock.

   Take the lock with interrupts disabled.

 - Use rcu_assign_pointer() for tmp_ops filter hash

   The tmp_ops used in update_ftrace_direct_mod() assigns its
   filter_hash field directly, but that field is annotated as __rcu and
   sparse complains. Assign it with rcu_assign_pointer()

 - Fix use-after-free in enable_trigger_private_data_free()

   The trace_event_call is accessed through the event_trigger_data's
   trace_event_file pointer to put the trace_event_call on freeing. The
   issue is that the trace_event_file data may have been freed already
   causing a use-after-free. Add a field to the event_trigger_data that
   points directly to the trace_event_call so that it can decrement its
   reference directly without needing to go through the
   trace_event_file.

 - Fix accounting of buffer data remote headers

   trace_buffer_desc_size() and trace_remote_alloc_buffer() undercount
   the number of pages is needed for the asked for size as it doesn't
   take into account the meta data on each page. Add a helper function
   to do the calculation properly and use that in these functions.

 - Catch nr_page_va overflow in ring_buffer_desc sizing

   The number of pages per remote ring buffer is capped by
   ring_buffer_desc::nr_page_va (32 bits). A buffer_size large enough to
   overflow that field would silently allocate a descriptor smaller than
   what was asked for.

 - Do not resize the subbuf order if any per_cpu buffer is disabled

   The mmapping of ring buffers disables resizing the subbuffers, but it
   is done per-cpu whereas the subbuf size change is done for all the
   per_cpu buffers under the buffer->mutex. It could change the size of
   some while the mapping is happening on others. Have the resize of the
   subbuf order check all the per_cpu buffers under the lock to see if
   any of them is disabled before starting and causing an inconsistency
   between buffers that are being mapped.

* tag 'trace-v7.3-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace: (25 commits)
  ring-buffer: Check resize_disabled before publishing the new subbuf order
  tracing/remotes: Catch nr_page_va overflow in ring_buffer_desc sizing
  tracing/remotes: Account for ring buffer page header in size calculation
  tracing: Don't dereference trace_event_file in deferred trigger free
  ftrace: Use rcu_assign_pointer() for tmp_ops filter hash
  ring-buffer: Acquire the lock with irqsave in rb_wake_up_waiters()
  tracing: Take trace_array reference when opening a tracer options file
  tracing: Fix ring_buffer_read_page_size() kernel-doc
  tracing: Restore :mod: trailer after parsing in ftrace_set_clr_event()
  tracing: Fix memory corruption from a "STACKTRACE" histogram key
  tracing: Fix memory corruption from the histogram stacktrace modifier
  tracing: Undo the registration when enabling the histogram trigger fails
  tracing: Take the reference before publishing the named histogram trigger
  tracing: Set the trace clock before registering the histogram trigger
  tracing: Fix typo "preceeded" in comment
  tracing: Fix typo "availabe" in comment
  tracing: Let histogram values keep the percent and graph modifiers
  tracing: Keep the entry count when the histogram stats allocation fails
  tracing: Free histogram the field rejected for a bad modifier
  tracing: Free histogram the var ref when its initialization fails
  ...

1 files changed, 1034 insertions, 0 deletions

diff --git a/rust/zerocopy/src/wrappers.rs b/rust/zerocopy/src/wrappers.rs
new file mode 100644
index 000000000..1a8cf2b41
--- /dev/null
+++ b/rust/zerocopy/src/wrappers.rs
@@ -0,0 +1,1034 @@
+// SPDX-License-Identifier: (BSD-2-Clause OR Apache-2.0) OR MIT
+//
+// Copyright 2023 The Fuchsia Authors
+//
+// Licensed under a BSD-style license <LICENSE-BSD>, Apache License, Version 2.0
+// <LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0>, or the MIT
+// license <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your option.
+// This file may not be copied, modified, or distributed except according to
+// those terms.
+
+use core::{fmt, hash::Hash};
+
+use super::*;
+use crate::pointer::{invariant::Valid, SizeEq, TransmuteFrom};
+
+/// A type with no alignment requirement.
+///
+/// An `Unalign` wraps a `T`, removing any alignment requirement. `Unalign<T>`
+/// has the same size and bit validity as `T`, but not necessarily the same
+/// alignment [or ABI]. This is useful if a type with an alignment requirement
+/// needs to be read from a chunk of memory which provides no alignment
+/// guarantees.
+///
+/// Since `Unalign` has no alignment requirement, the inner `T` may not be
+/// properly aligned in memory. There are five ways to access the inner `T`:
+/// - by value, using [`get`] or [`into_inner`]
+/// - by reference inside of a callback, using [`update`]
+/// - fallibly by reference, using [`try_deref`] or [`try_deref_mut`]; these can
+///   fail if the `Unalign` does not satisfy `T`'s alignment requirement at
+///   runtime
+/// - unsafely by reference, using [`deref_unchecked`] or
+///   [`deref_mut_unchecked`]; it is the caller's responsibility to ensure that
+///   the `Unalign` satisfies `T`'s alignment requirement
+/// - (where `T: Unaligned`) infallibly by reference, using [`Deref::deref`] or
+///   [`DerefMut::deref_mut`]
+///
+/// [or ABI]: https://github.com/google/zerocopy/issues/164
+/// [`get`]: Unalign::get
+/// [`into_inner`]: Unalign::into_inner
+/// [`update`]: Unalign::update
+/// [`try_deref`]: Unalign::try_deref
+/// [`try_deref_mut`]: Unalign::try_deref_mut
+/// [`deref_unchecked`]: Unalign::deref_unchecked
+/// [`deref_mut_unchecked`]: Unalign::deref_mut_unchecked
+///
+/// # Example
+///
+/// In this example, we need `EthernetFrame` to have no alignment requirement -
+/// and thus implement [`Unaligned`]. `EtherType` is `#[repr(u16)]` and so
+/// cannot implement `Unaligned`. We use `Unalign` to relax `EtherType`'s
+/// alignment requirement so that `EthernetFrame` has no alignment requirement
+/// and can implement `Unaligned`.
+///
+/// ```rust
+/// use zerocopy::*;
+/// # use zerocopy_derive::*;
+/// # #[derive(FromBytes, KnownLayout, Immutable, Unaligned)] #[repr(C)] struct Mac([u8; 6]);
+///
+/// # #[derive(PartialEq, Copy, Clone, Debug)]
+/// #[derive(TryFromBytes, KnownLayout, Immutable)]
+/// #[repr(u16)]
+/// enum EtherType {
+///     Ipv4 = 0x0800u16.to_be(),
+///     Arp = 0x0806u16.to_be(),
+///     Ipv6 = 0x86DDu16.to_be(),
+///     # /*
+///     ...
+///     # */
+/// }
+///
+/// #[derive(TryFromBytes, KnownLayout, Immutable, Unaligned)]
+/// #[repr(C)]
+/// struct EthernetFrame {
+///     src: Mac,
+///     dst: Mac,
+///     ethertype: Unalign<EtherType>,
+///     payload: [u8],
+/// }
+///
+/// let bytes = &[
+///     # 0, 1, 2, 3, 4, 5,
+///     # 6, 7, 8, 9, 10, 11,
+///     # /*
+///     ...
+///     # */
+///     0x86, 0xDD,            // EtherType
+///     0xDE, 0xAD, 0xBE, 0xEF // Payload
+/// ][..];
+///
+/// // PANICS: Guaranteed not to panic because `bytes` is of the right
+/// // length, has the right contents, and `EthernetFrame` has no
+/// // alignment requirement.
+/// let packet = EthernetFrame::try_ref_from_bytes(&bytes).unwrap();
+///
+/// assert_eq!(packet.ethertype.get(), EtherType::Ipv6);
+/// assert_eq!(packet.payload, [0xDE, 0xAD, 0xBE, 0xEF]);
+/// ```
+///
+/// # Safety
+///
+/// `Unalign<T>` is guaranteed to have the same size and bit validity as `T`,
+/// and to have [`UnsafeCell`]s covering the same byte ranges as `T`.
+/// `Unalign<T>` is guaranteed to have alignment 1.
+// NOTE: This type is sound to use with types that need to be dropped. The
+// reason is that the compiler-generated drop code automatically moves all
+// values to aligned memory slots before dropping them in-place. This is not
+// well-documented, but it's hinted at in places like [1] and [2]. However, this
+// also means that `T` must be `Sized`; unless something changes, we can never
+// support unsized `T`. [3]
+//
+// [1] https://github.com/rust-lang/rust/issues/54148#issuecomment-420529646
+// [2] https://github.com/google/zerocopy/pull/126#discussion_r1018512323
+// [3] https://github.com/google/zerocopy/issues/209
+#[allow(missing_debug_implementations)]
+#[derive(Default, Copy)]
+#[cfg_attr(any(feature = "derive", test), derive(Immutable, FromBytes, IntoBytes, Unaligned))]
+#[repr(C, packed)]
+pub struct Unalign<T>(T);
+
+// We do not use `derive(KnownLayout)` on `Unalign`, because the derive is not
+// smart enough to realize that `Unalign<T>` is always sized and thus emits a
+// `KnownLayout` impl bounded on `T: KnownLayout.` This is overly restrictive.
+impl_known_layout!(T => Unalign<T>);
+
+// FIXME(https://github.com/rust-lang/rust-clippy/issues/16087): Move these
+// attributes below the comment once this Clippy bug is fixed.
+#[cfg_attr(
+    all(__ZEROCOPY_INTERNAL_USE_ONLY_NIGHTLY_FEATURES_IN_TESTS, any(feature = "derive", test)),
+    expect(unused_unsafe)
+)]
+#[cfg_attr(
+    all(
+        not(__ZEROCOPY_INTERNAL_USE_ONLY_NIGHTLY_FEATURES_IN_TESTS),
+        any(feature = "derive", test)
+    ),
+    allow(unused_unsafe)
+)]
+// SAFETY:
+// - `Unalign<T>` promises to have alignment 1, and so we don't require that `T:
+//   Unaligned`.
+// - `Unalign<T>` has the same bit validity as `T`, and so it is `FromZeros`,
+//   `FromBytes`, or `IntoBytes` exactly when `T` is as well.
+// - `Immutable`: `Unalign<T>` has the same fields as `T`, so it permits
+//   interior mutation exactly when `T` does.
+// - `TryFromBytes`: `Unalign<T>` has the same the same bit validity as `T`, so
+//   `T::is_bit_valid` is a sound implementation of `is_bit_valid`.
+//
+#[allow(clippy::multiple_unsafe_ops_per_block)]
+const _: () = unsafe {
+    impl_or_verify!(T => Unaligned for Unalign<T>);
+    impl_or_verify!(T: Immutable => Immutable for Unalign<T>);
+    impl_or_verify!(
+        T: TryFromBytes => TryFromBytes for Unalign<T>;
+        |c| T::is_bit_valid(c.transmute::<_, _, BecauseImmutable>())
+    );
+    impl_or_verify!(T: FromZeros => FromZeros for Unalign<T>);
+    impl_or_verify!(T: FromBytes => FromBytes for Unalign<T>);
+    impl_or_verify!(T: IntoBytes => IntoBytes for Unalign<T>);
+};
+
+// Note that `Unalign: Clone` only if `T: Copy`. Since the inner `T` may not be
+// aligned, there's no way to safely call `T::clone`, and so a `T: Clone` bound
+// is not sufficient to implement `Clone` for `Unalign`.
+impl<T: Copy> Clone for Unalign<T> {
+    #[inline(always)]
+    fn clone(&self) -> Unalign<T> {
+        *self
+    }
+}
+
+impl<T> Unalign<T> {
+    /// Constructs a new `Unalign`.
+    #[inline(always)]
+    pub const fn new(val: T) -> Unalign<T> {
+        Unalign(val)
+    }
+
+    /// Consumes `self`, returning the inner `T`.
+    #[inline(always)]
+    pub const fn into_inner(self) -> T {
+        // SAFETY: Since `Unalign` is `#[repr(C, packed)]`, it has the same size
+        // and bit validity as `T`.
+        //
+        // We do this instead of just destructuring in order to prevent
+        // `Unalign`'s `Drop::drop` from being run, since dropping is not
+        // supported in `const fn`s.
+        //
+        // FIXME(https://github.com/rust-lang/rust/issues/73255): Destructure
+        // instead of using unsafe.
+        unsafe { crate::util::transmute_unchecked(self) }
+    }
+
+    /// Attempts to return a reference to the wrapped `T`, failing if `self` is
+    /// not properly aligned.
+    ///
+    /// If `self` does not satisfy `align_of::<T>()`, then `try_deref` returns
+    /// `Err`.
+    ///
+    /// If `T: Unaligned`, then `Unalign<T>` implements [`Deref`], and callers
+    /// may prefer [`Deref::deref`], which is infallible.
+    #[inline(always)]
+    pub fn try_deref(&self) -> Result<&T, AlignmentError<&Self, T>> {
+        let inner = Ptr::from_ref(self).transmute();
+        match inner.try_into_aligned() {
+            Ok(aligned) => Ok(aligned.as_ref()),
+            Err(err) => Err(err.map_src(
+                #[inline(always)]
+                |src| src.into_unalign().as_ref(),
+            )),
+        }
+    }
+
+    /// Attempts to return a mutable reference to the wrapped `T`, failing if
+    /// `self` is not properly aligned.
+    ///
+    /// If `self` does not satisfy `align_of::<T>()`, then `try_deref` returns
+    /// `Err`.
+    ///
+    /// If `T: Unaligned`, then `Unalign<T>` implements [`DerefMut`], and
+    /// callers may prefer [`DerefMut::deref_mut`], which is infallible.
+    #[inline(always)]
+    pub fn try_deref_mut(&mut self) -> Result<&mut T, AlignmentError<&mut Self, T>> {
+        let inner = Ptr::from_mut(self).transmute::<_, _, (_, (_, _))>();
+        match inner.try_into_aligned() {
+            Ok(aligned) => Ok(aligned.as_mut()),
+            Err(err) => Err(err.map_src(|src| src.into_unalign().as_mut())),
+        }
+    }
+
+    /// Returns a reference to the wrapped `T` without checking alignment.
+    ///
+    /// If `T: Unaligned`, then `Unalign<T>` implements[ `Deref`], and callers
+    /// may prefer [`Deref::deref`], which is safe.
+    ///
+    /// # Safety
+    ///
+    /// The caller must guarantee that `self` satisfies `align_of::<T>()`.
+    #[inline(always)]
+    pub const unsafe fn deref_unchecked(&self) -> &T {
+        // SAFETY: `Unalign<T>` is `repr(transparent)`, so there is a valid `T`
+        // at the same memory location as `self`. It has no alignment guarantee,
+        // but the caller has promised that `self` is properly aligned, so we
+        // know that it is sound to create a reference to `T` at this memory
+        // location.
+        //
+        // We use `mem::transmute` instead of `&*self.get_ptr()` because
+        // dereferencing pointers is not stable in `const` on our current MSRV
+        // (1.56 as of this writing).
+        unsafe { mem::transmute(self) }
+    }
+
+    /// Returns a mutable reference to the wrapped `T` without checking
+    /// alignment.
+    ///
+    /// If `T: Unaligned`, then `Unalign<T>` implements[ `DerefMut`], and
+    /// callers may prefer [`DerefMut::deref_mut`], which is safe.
+    ///
+    /// # Safety
+    ///
+    /// The caller must guarantee that `self` satisfies `align_of::<T>()`.
+    #[inline(always)]
+    pub unsafe fn deref_mut_unchecked(&mut self) -> &mut T {
+        // SAFETY: `self.get_mut_ptr()` returns a raw pointer to a valid `T` at
+        // the same memory location as `self`. It has no alignment guarantee,
+        // but the caller has promised that `self` is properly aligned, so we
+        // know that the pointer itself is aligned, and thus that it is sound to
+        // create a reference to a `T` at this memory location.
+        unsafe { &mut *self.get_mut_ptr() }
+    }
+
+    /// Gets an unaligned raw pointer to the inner `T`.
+    ///
+    /// # Safety
+    ///
+    /// The returned raw pointer is not necessarily aligned to
+    /// `align_of::<T>()`. Most functions which operate on raw pointers require
+    /// those pointers to be aligned, so calling those functions with the result
+    /// of `get_ptr` will result in undefined behavior if alignment is not
+    /// guaranteed using some out-of-band mechanism. In general, the only
+    /// functions which are safe to call with this pointer are those which are
+    /// explicitly documented as being sound to use with an unaligned pointer,
+    /// such as [`read_unaligned`].
+    ///
+    /// Even if the caller is permitted to mutate `self` (e.g. they have
+    /// ownership or a mutable borrow), it is not guaranteed to be sound to
+    /// write through the returned pointer. If writing is required, prefer
+    /// [`get_mut_ptr`] instead.
+    ///
+    /// [`read_unaligned`]: core::ptr::read_unaligned
+    /// [`get_mut_ptr`]: Unalign::get_mut_ptr
+    #[inline(always)]
+    pub const fn get_ptr(&self) -> *const T {
+        ptr::addr_of!(self.0)
+    }
+
+    /// Gets an unaligned mutable raw pointer to the inner `T`.
+    ///
+    /// # Safety
+    ///
+    /// The returned raw pointer is not necessarily aligned to
+    /// `align_of::<T>()`. Most functions which operate on raw pointers require
+    /// those pointers to be aligned, so calling those functions with the result
+    /// of `get_ptr` will result in undefined behavior if alignment is not
+    /// guaranteed using some out-of-band mechanism. In general, the only
+    /// functions which are safe to call with this pointer are those which are
+    /// explicitly documented as being sound to use with an unaligned pointer,
+    /// such as [`read_unaligned`].
+    ///
+    /// [`read_unaligned`]: core::ptr::read_unaligned
+    // FIXME(https://github.com/rust-lang/rust/issues/57349): Make this `const`.
+    #[inline(always)]
+    pub fn get_mut_ptr(&mut self) -> *mut T {
+        ptr::addr_of_mut!(self.0)
+    }
+
+    /// Sets the inner `T`, dropping the previous value.
+    // FIXME(https://github.com/rust-lang/rust/issues/57349): Make this `const`.
+    #[inline(always)]
+    pub fn set(&mut self, t: T) {
+        *self = Unalign::new(t);
+    }
+
+    /// Updates the inner `T` by calling a function on it.
+    ///
+    /// If [`T: Unaligned`], then `Unalign<T>` implements [`DerefMut`], and that
+    /// impl should be preferred over this method when performing updates, as it
+    /// will usually be faster and more ergonomic.
+    ///
+    /// For large types, this method may be expensive, as it requires copying
+    /// `2 * size_of::<T>()` bytes. \[1\]
+    ///
+    /// \[1\] Since the inner `T` may not be aligned, it would not be sound to
+    /// invoke `f` on it directly. Instead, `update` moves it into a
+    /// properly-aligned location in the local stack frame, calls `f` on it, and
+    /// then moves it back to its original location in `self`.
+    ///
+    /// [`T: Unaligned`]: Unaligned
+    #[inline]
+    pub fn update<O, F: FnOnce(&mut T) -> O>(&mut self, f: F) -> O {
+        if mem::align_of::<T>() == 1 {
+            // While we advise callers to use `DerefMut` when `T: Unaligned`,
+            // not all callers will be able to guarantee `T: Unaligned` in all
+            // cases. In particular, callers who are themselves providing an API
+            // which is generic over `T` may sometimes be called by *their*
+            // callers with `T` such that `align_of::<T>() == 1`, but cannot
+            // guarantee this in the general case. Thus, this optimization may
+            // sometimes be helpful.
+
+            // SAFETY: Since `T`'s alignment is 1, `self` satisfies its
+            // alignment by definition.
+            let t = unsafe { self.deref_mut_unchecked() };
+            return f(t);
+        }
+
+        // On drop, this moves `copy` out of itself and uses `ptr::write` to
+        // overwrite `slf`.
+        struct WriteBackOnDrop<T> {
+            copy: ManuallyDrop<T>,
+            slf: *mut Unalign<T>,
+        }
+
+        impl<T> Drop for WriteBackOnDrop<T> {
+            fn drop(&mut self) {
+                // SAFETY: We never use `copy` again as required by
+                // `ManuallyDrop::take`.
+                let copy = unsafe { ManuallyDrop::take(&mut self.copy) };
+                // SAFETY: `slf` is the raw pointer value of `self`. We know it
+                // is valid for writes and properly aligned because `self` is a
+                // mutable reference, which guarantees both of these properties.
+                unsafe { ptr::write(self.slf, Unalign::new(copy)) };
+            }
+        }
+
+        // SAFETY: We know that `self` is valid for reads, properly aligned, and
+        // points to an initialized `Unalign<T>` because it is a mutable
+        // reference, which guarantees all of these properties.
+        //
+        // Since `T: !Copy`, it would be unsound in the general case to allow
+        // both the original `Unalign<T>` and the copy to be used by safe code.
+        // We guarantee that the copy is used to overwrite the original in the
+        // `Drop::drop` impl of `WriteBackOnDrop`. So long as this `drop` is
+        // called before any other safe code executes, soundness is upheld.
+        // While this method can terminate in two ways (by returning normally or
+        // by unwinding due to a panic in `f`), in both cases, `write_back` is
+        // dropped - and its `drop` called - before any other safe code can
+        // execute.
+        let copy = unsafe { ptr::read(self) }.into_inner();
+        let mut write_back = WriteBackOnDrop { copy: ManuallyDrop::new(copy), slf: self };
+
+        let ret = f(&mut write_back.copy);
+
+        drop(write_back);
+        ret
+    }
+}
+
+impl<T: Copy> Unalign<T> {
+    /// Gets a copy of the inner `T`.
+    // FIXME(https://github.com/rust-lang/rust/issues/57349): Make this `const`.
+    #[inline(always)]
+    pub fn get(&self) -> T {
+        let Unalign(val) = *self;
+        val
+    }
+}
+
+impl<T: Unaligned> Deref for Unalign<T> {
+    type Target = T;
+
+    #[inline(always)]
+    fn deref(&self) -> &T {
+        Ptr::from_ref(self).transmute().bikeshed_recall_aligned().as_ref()
+    }
+}
+
+impl<T: Unaligned> DerefMut for Unalign<T> {
+    #[inline(always)]
+    fn deref_mut(&mut self) -> &mut T {
+        Ptr::from_mut(self).transmute::<_, _, (_, (_, _))>().bikeshed_recall_aligned().as_mut()
+    }
+}
+
+impl<T: Unaligned + PartialOrd> PartialOrd<Unalign<T>> for Unalign<T> {
+    #[inline(always)]
+    fn partial_cmp(&self, other: &Unalign<T>) -> Option<Ordering> {
+        PartialOrd::partial_cmp(self.deref(), other.deref())
+    }
+}
+
+impl<T: Unaligned + Ord> Ord for Unalign<T> {
+    #[inline(always)]
+    fn cmp(&self, other: &Unalign<T>) -> Ordering {
+        Ord::cmp(self.deref(), other.deref())
+    }
+}
+
+impl<T: Unaligned + PartialEq> PartialEq<Unalign<T>> for Unalign<T> {
+    #[inline(always)]
+    fn eq(&self, other: &Unalign<T>) -> bool {
+        PartialEq::eq(self.deref(), other.deref())
+    }
+}
+
+impl<T: Unaligned + Eq> Eq for Unalign<T> {}
+
+impl<T: Unaligned + Hash> Hash for Unalign<T> {
+    #[inline(always)]
+    fn hash<H>(&self, state: &mut H)
+    where
+        H: Hasher,
+    {
+        self.deref().hash(state);
+    }
+}
+
+impl<T: Unaligned + Debug> Debug for Unalign<T> {
+    #[inline(always)]
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        Debug::fmt(self.deref(), f)
+    }
+}
+
+impl<T: Unaligned + Display> Display for Unalign<T> {
+    #[inline(always)]
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        Display::fmt(self.deref(), f)
+    }
+}
+
+/// A wrapper type to construct uninitialized instances of `T`.
+///
+/// `MaybeUninit` is identical to the [standard library
+/// `MaybeUninit`][core-maybe-uninit] type except that it supports unsized
+/// types.
+///
+/// # Layout
+///
+/// The same layout guarantees and caveats apply to `MaybeUninit<T>` as apply to
+/// the [standard library `MaybeUninit`][core-maybe-uninit] with one exception:
+/// for `T: !Sized`, there is no single value for `T`'s size. Instead, for such
+/// types, the following are guaranteed:
+/// - Every [valid size][valid-size] for `T` is a valid size for
+///   `MaybeUninit<T>` and vice versa
+/// - Given `t: *const T` and `m: *const MaybeUninit<T>` with identical fat
+///   pointer metadata, `t` and `m` address the same number of bytes (and
+///   likewise for `*mut`)
+///
+/// [core-maybe-uninit]: core::mem::MaybeUninit
+/// [valid-size]: crate::KnownLayout#what-is-a-valid-size
+#[repr(transparent)]
+#[doc(hidden)]
+pub struct MaybeUninit<T: ?Sized + KnownLayout>(
+    // SAFETY: `MaybeUninit<T>` has the same size as `T`, because (by invariant
+    // on `T::MaybeUninit`) `T::MaybeUninit` has `T::LAYOUT` identical to `T`,
+    // and because (invariant on `T::LAYOUT`) we can trust that `LAYOUT`
+    // accurately reflects the layout of `T`. By invariant on `T::MaybeUninit`,
+    // it admits uninitialized bytes in all positions. Because `MaybeUninit` is
+    // marked `repr(transparent)`, these properties additionally hold true for
+    // `Self`.
+    T::MaybeUninit,
+);
+
+#[doc(hidden)]
+impl<T: ?Sized + KnownLayout> MaybeUninit<T> {
+    /// Constructs a `MaybeUninit<T>` initialized with the given value.
+    #[inline(always)]
+    pub fn new(val: T) -> Self
+    where
+        T: Sized,
+        Self: Sized,
+    {
+        // SAFETY: It is valid to transmute `val` to `MaybeUninit<T>` because it
+        // is both valid to transmute `val` to `T::MaybeUninit`, and it is valid
+        // to transmute from `T::MaybeUninit` to `MaybeUninit<T>`.
+        //
+        // First, it is valid to transmute `val` to `T::MaybeUninit` because, by
+        // invariant on `T::MaybeUninit`:
+        // - For `T: Sized`, `T` and `T::MaybeUninit` have the same size.
+        // - All byte sequences of the correct size are valid values of
+        //   `T::MaybeUninit`.
+        //
+        // Second, it is additionally valid to transmute from `T::MaybeUninit`
+        // to `MaybeUninit<T>`, because `MaybeUninit<T>` is a
+        // `repr(transparent)` wrapper around `T::MaybeUninit`.
+        //
+        // These two transmutes are collapsed into one so we don't need to add a
+        // `T::MaybeUninit: Sized` bound to this function's `where` clause.
+        unsafe { crate::util::transmute_unchecked(val) }
+    }
+
+    /// Constructs an uninitialized `MaybeUninit<T>`.
+    #[must_use]
+    #[inline(always)]
+    pub fn uninit() -> Self
+    where
+        T: Sized,
+        Self: Sized,
+    {
+        let uninit = CoreMaybeUninit::<T>::uninit();
+        // SAFETY: It is valid to transmute from `CoreMaybeUninit<T>` to
+        // `MaybeUninit<T>` since they both admit uninitialized bytes in all
+        // positions, and they have the same size (i.e., that of `T`).
+        //
+        // `MaybeUninit<T>` has the same size as `T`, because (by invariant on
+        // `T::MaybeUninit`) `T::MaybeUninit` has `T::LAYOUT` identical to `T`,
+        // and because (invariant on `T::LAYOUT`) we can trust that `LAYOUT`
+        // accurately reflects the layout of `T`.
+        //
+        // `CoreMaybeUninit<T>` has the same size as `T` [1] and admits
+        // uninitialized bytes in all positions.
+        //
+        // [1] Per https://doc.rust-lang.org/1.81.0/std/mem/union.MaybeUninit.html#layout-1:
+        //
+        //   `MaybeUninit<T>` is guaranteed to have the same size, alignment,
+        //   and ABI as `T`
+        unsafe { crate::util::transmute_unchecked(uninit) }
+    }
+
+    /// Creates a `Box<MaybeUninit<T>>`.
+    ///
+    /// This function is useful for allocating large, uninit values on the heap
+    /// without ever creating a temporary instance of `Self` on the stack.
+    ///
+    /// # Errors
+    ///
+    /// Returns an error on allocation failure. Allocation failure is guaranteed
+    /// never to cause a panic or an abort.
+    #[cfg(feature = "alloc")]
+    #[inline]
+    pub fn new_boxed_uninit(meta: T::PointerMetadata) -> Result<Box<Self>, AllocError> {
+        // SAFETY: `alloc::alloc::alloc_zeroed` is a valid argument of
+        // `new_box`. The referent of the pointer returned by `alloc` (and,
+        // consequently, the `Box` derived from it) is a valid instance of
+        // `Self`, because `Self` is `MaybeUninit` and thus admits arbitrary
+        // (un)initialized bytes.
+        unsafe { crate::util::new_box(meta, alloc::alloc::alloc) }
+    }
+
+    /// Extracts the value from the `MaybeUninit<T>` container.
+    ///
+    /// # Safety
+    ///
+    /// The caller must ensure that `self` is in an bit-valid state. Depending
+    /// on subsequent use, it may also need to be in a library-valid state.
+    #[inline(always)]
+    pub unsafe fn assume_init(self) -> T
+    where
+        T: Sized,
+        Self: Sized,
+    {
+        // SAFETY: The caller guarantees that `self` is in an bit-valid state.
+        unsafe { crate::util::transmute_unchecked(self) }
+    }
+}
+
+impl<T: ?Sized + KnownLayout> fmt::Debug for MaybeUninit<T> {
+    #[inline]
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.pad(core::any::type_name::<Self>())
+    }
+}
+
+#[allow(unreachable_pub)] // False positive on MSRV
+#[doc(hidden)]
+pub use read_only_def::*;
+mod read_only_def {
+    /// A read-only wrapper.
+    ///
+    /// A `ReadOnly<T>` disables any interior mutability in `T`, ensuring that
+    /// a `&ReadOnly<T>` is genuinely read-only. Thus, `ReadOnly<T>` is
+    /// [`Immutable`] regardless of whether `T` is.
+    ///
+    /// Note that `&mut ReadOnly<T>` still permits mutation – the read-only
+    /// property only applies to shared references.
+    ///
+    /// [`Immutable`]: crate::Immutable
+    #[repr(transparent)]
+    pub struct ReadOnly<T: ?Sized> {
+        // INVARIANT: `inner` is never mutated through a `&ReadOnly<T>`
+        // reference.
+        inner: T,
+    }
+
+    impl<T> ReadOnly<T> {
+        /// Creates a new `ReadOnly`.
+        #[must_use]
+        #[inline(always)]
+        pub const fn new(t: T) -> ReadOnly<T> {
+            ReadOnly { inner: t }
+        }
+
+        /// Returns the inner value.
+        #[must_use]
+        #[inline(always)]
+        pub fn into_inner(r: ReadOnly<T>) -> T {
+            r.inner
+        }
+    }
+
+    impl<T: ?Sized> ReadOnly<T> {
+        #[inline(always)]
+        pub(crate) fn as_mut(r: &mut ReadOnly<T>) -> &mut T {
+            // SAFETY: `r: &mut ReadOnly`, so this doesn't violate the invariant
+            // that `inner` is never mutated through a `&ReadOnly<T>` reference.
+            &mut r.inner
+        }
+
+        /// # Safety
+        ///
+        /// The caller promises not to mutate the referent (i.e., via interior
+        /// mutation).
+        pub(crate) const unsafe fn as_ref_unchecked(r: &ReadOnly<T>) -> &T {
+            // SAFETY: The caller promises not to mutate the referent.
+            &r.inner
+        }
+    }
+}
+
+// SAFETY: `ReadOnly<T>` is a `#[repr(transparent)` wrapper around `T`.
+const _: () = unsafe {
+    unsafe_impl_known_layout!(T: ?Sized + KnownLayout => #[repr(T)] ReadOnly<T>);
+};
+
+#[allow(clippy::multiple_unsafe_ops_per_block)]
+// SAFETY:
+// - `ReadOnly<T>` has the same alignment as `T`, and so it is `Unaligned`
+//   exactly when `T` is as well.
+// - `ReadOnly<T>` has the same bit validity as `T`, and so this `is_bit_valid`
+//   implementation is correct, and thus the `TryFromBytes` impl is sound.
+// - `ReadOnly<T>` has the same bit validity as `T`, and so it is `FromZeros`,
+//   `FromBytes`, and `IntoBytes` exactly when `T` is as well.
+const _: () = unsafe {
+    unsafe_impl!(T: ?Sized + Unaligned => Unaligned for ReadOnly<T>);
+    unsafe_impl!(
+        T: ?Sized + TryFromBytes => TryFromBytes for ReadOnly<T>;
+        |c| T::is_bit_valid(c.cast::<_, <ReadOnly<T> as SizeEq<ReadOnly<ReadOnly<T>>>>::CastFrom, _>())
+    );
+    unsafe_impl!(T: ?Sized + FromZeros => FromZeros for ReadOnly<T>);
+    unsafe_impl!(T: ?Sized + FromBytes => FromBytes for ReadOnly<T>);
+    unsafe_impl!(T: ?Sized + IntoBytes => IntoBytes for ReadOnly<T>);
+};
+
+// SAFETY: By invariant, `inner` is never mutated through a `&ReadOnly<T>`
+// reference.
+const _: () = unsafe {
+    unsafe_impl!(T: ?Sized => Immutable for ReadOnly<T>);
+};
+
+const _: () = {
+    use crate::pointer::cast::CastExact;
+
+    // SAFETY: `ReadOnly<T>` has the same layout as `T`.
+    define_cast!(unsafe { pub CastFromReadOnly<T: ?Sized> = ReadOnly<T> => T});
+    // SAFETY: `ReadOnly<T>` has the same layout as `T`.
+    unsafe impl<T: ?Sized> CastExact<ReadOnly<T>, T> for CastFromReadOnly {}
+    // SAFETY: `ReadOnly<T>` has the same layout as `T`.
+    define_cast!(unsafe { pub CastToReadOnly<T: ?Sized> = T => ReadOnly<T>});
+    // SAFETY: `ReadOnly<T>` has the same layout as `T`.
+    unsafe impl<T: ?Sized> CastExact<T, ReadOnly<T>> for CastToReadOnly {}
+
+    impl<T: ?Sized> SizeEq<ReadOnly<T>> for T {
+        type CastFrom = CastFromReadOnly;
+    }
+
+    impl<T: ?Sized> SizeEq<T> for ReadOnly<T> {
+        type CastFrom = CastToReadOnly;
+    }
+};
+
+// SAFETY: `ReadOnly<T>` is a `#[repr(transparent)]` wrapper around `T`, and so
+// it has the same bit validity as `T`.
+unsafe impl<T: ?Sized> TransmuteFrom<T, Valid, Valid> for ReadOnly<T> {}
+
+// SAFETY: `ReadOnly<T>` is a `#[repr(transparent)]` wrapper around `T`, and so
+// it has the same bit validity as `T`.
+unsafe impl<T: ?Sized> TransmuteFrom<ReadOnly<T>, Valid, Valid> for T {}
+
+impl<'a, T: ?Sized + Immutable> From<&'a T> for &'a ReadOnly<T> {
+    #[inline(always)]
+    fn from(t: &'a T) -> &'a ReadOnly<T> {
+        let ro = Ptr::from_ref(t).transmute::<_, _, (_, _)>();
+        // SAFETY: `ReadOnly<T>` has the same alignment as `T`, and
+        // `Ptr::from_ref` produces an aligned `Ptr`.
+        let ro = unsafe { ro.assume_alignment() };
+        ro.as_ref()
+    }
+}
+
+impl<T: ?Sized + Immutable> Deref for ReadOnly<T> {
+    type Target = T;
+
+    #[inline(always)]
+    fn deref(&self) -> &Self::Target {
+        // SAFETY: By `T: Immutable`, `&T` doesn't permit interior mutation.
+        unsafe { ReadOnly::as_ref_unchecked(self) }
+    }
+}
+
+impl<T: ?Sized + Immutable> DerefMut for ReadOnly<T> {
+    #[inline(always)]
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        ReadOnly::as_mut(self)
+    }
+}
+
+impl<T: ?Sized + Immutable + Debug> Debug for ReadOnly<T> {
+    #[inline(always)]
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        self.deref().fmt(f)
+    }
+}
+
+// SAFETY: See safety comment on `ProjectToTag`.
+unsafe impl<T: HasTag + ?Sized> HasTag for ReadOnly<T> {
+    #[allow(clippy::missing_inline_in_public_items)]
+    fn only_derive_is_allowed_to_implement_this_trait()
+    where
+        Self: Sized,
+    {
+    }
+
+    type Tag = T::Tag;
+
+    // SAFETY: `<T as SizeEq<ReadOnly<T>>>::CastFrom` is a no-op projection that
+    // produces a pointer with the same referent. By invariant, for any `Ptr<'_,
+    // T, I>` it is sound to use `T::ProjectToTag` to project to a `Ptr<'_,
+    // T::Tag, I>`. Since `ReadOnly<T>` has the same layout and validity as `T`,
+    // the same is true of projecting from a `Ptr<'_, ReadOnly<T>, I>`.
+    type ProjectToTag = crate::pointer::cast::TransitiveProject<
+        T,
+        <T as SizeEq<ReadOnly<T>>>::CastFrom,
+        T::ProjectToTag,
+    >;
+}
+
+// SAFETY: `ReadOnly<T>` is a `#[repr(transparent)]` wrapper around `T`, and so
+// has the same fields at the same offsets. Thus, it satisfies the safety
+// invariants of `HasField<Field, VARIANT_ID, FIELD_ID>` for field `f` exactly
+// when `T` does, as guaranteed by the `T: HasField` bound:
+// - If `VARIANT_ID` is `STRUCT_VARIANT_ID` or `UNION_VARIANT_ID`, then `T` has
+//   the layout of a struct or union type. Since `ReadOnly<T>` is a transparent
+//   wrapper around `T`, it does too. Otherwise, if `VARIANT_ID` is an enum
+//   variant index, then `T` has the layout of an enum type, and `ReadOnly<T>`
+//   does too.
+// - By `T: HasField<_, _, FIELD_ID>`:
+//   - `T` has a field `f` with name `n` such that
+//     `FIELD_ID = zerocopy::ident_id!(n)` or at index `i` such that
+//     `FIELD_ID = zerocopy::ident_id!(i)`.
+//   - `Field` has the same visibility as `f`.
+//   - `T::Type` has the same type as `f`. Thus, `ReadOnly<T::Type>` has the
+//     same type as `f`, wrapped in `ReadOnly`.
+//
+// `project` satisfies its post-condition – namely, that the returned pointer
+// refers to a non-strict subset of the bytes of `slf`'s referent, and has the
+// same provenance as `slf` – because all intermediate operations satisfy those
+// same conditions.
+unsafe impl<T, Field, const VARIANT_ID: i128, const FIELD_ID: i128>
+    HasField<Field, VARIANT_ID, FIELD_ID> for ReadOnly<T>
+where
+    T: HasField<Field, VARIANT_ID, FIELD_ID> + ?Sized,
+{
+    #[allow(clippy::missing_inline_in_public_items)]
+    fn only_derive_is_allowed_to_implement_this_trait()
+    where
+        Self: Sized,
+    {
+    }
+
+    type Type = ReadOnly<T::Type>;
+
+    #[inline(always)]
+    fn project(slf: PtrInner<'_, Self>) -> *mut ReadOnly<T::Type> {
+        slf.project::<_, <T as SizeEq<ReadOnly<T>>>::CastFrom>()
+            .project::<_, crate::pointer::cast::Projection<Field, VARIANT_ID, FIELD_ID>>()
+            .project::<_, <ReadOnly<T::Type> as SizeEq<T::Type>>::CastFrom>()
+            .as_non_null()
+            .as_ptr()
+    }
+}
+
+// SAFETY: `ReadOnly<T>` is a `#[repr(transparent)]` wrapper around `T`, and so
+// has the same fields at the same offsets. `is_projectable` simply delegates to
+// `T::is_projectable`, which is sound because a `Ptr<'_, ReadOnly<T>, I>` will
+// be projectable exactly when a `Ptr<'_, T, I>` referent is.
+unsafe impl<T, Field, I, const VARIANT_ID: i128, const FIELD_ID: i128>
+    ProjectField<Field, I, VARIANT_ID, FIELD_ID> for ReadOnly<T>
+where
+    T: ProjectField<Field, I, VARIANT_ID, FIELD_ID> + ?Sized,
+    I: invariant::Invariants,
+{
+    #[allow(clippy::missing_inline_in_public_items)]
+    fn only_derive_is_allowed_to_implement_this_trait()
+    where
+        Self: Sized,
+    {
+    }
+
+    type Invariants = T::Invariants;
+
+    type Error = T::Error;
+
+    #[inline(always)]
+    fn is_projectable<'a>(ptr: Ptr<'a, Self::Tag, I>) -> Result<(), Self::Error> {
+        T::is_projectable(ptr)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use core::panic::AssertUnwindSafe;
+
+    use super::*;
+    use crate::util::testutil::*;
+
+    #[test]
+    fn test_unalign() {
+        // Test methods that don't depend on alignment.
+        let mut u = Unalign::new(AU64(123));
+        assert_eq!(u.get(), AU64(123));
+        assert_eq!(u.into_inner(), AU64(123));
+        assert_eq!(u.get_ptr(), <*const _>::cast::<AU64>(&u));
+        assert_eq!(u.get_mut_ptr(), <*mut _>::cast::<AU64>(&mut u));
+        u.set(AU64(321));
+        assert_eq!(u.get(), AU64(321));
+
+        // Test methods that depend on alignment (when alignment is satisfied).
+        let mut u: Align<_, AU64> = Align::new(Unalign::new(AU64(123)));
+        assert_eq!(u.t.try_deref().unwrap(), &AU64(123));
+        assert_eq!(u.t.try_deref_mut().unwrap(), &mut AU64(123));
+        // SAFETY: The `Align<_, AU64>` guarantees proper alignment.
+        assert_eq!(unsafe { u.t.deref_unchecked() }, &AU64(123));
+        // SAFETY: The `Align<_, AU64>` guarantees proper alignment.
+        assert_eq!(unsafe { u.t.deref_mut_unchecked() }, &mut AU64(123));
+        *u.t.try_deref_mut().unwrap() = AU64(321);
+        assert_eq!(u.t.get(), AU64(321));
+
+        // Test methods that depend on alignment (when alignment is not
+        // satisfied).
+        let mut u: ForceUnalign<_, AU64> = ForceUnalign::new(Unalign::new(AU64(123)));
+        assert!(matches!(u.t.try_deref(), Err(AlignmentError { .. })));
+        assert!(matches!(u.t.try_deref_mut(), Err(AlignmentError { .. })));
+
+        // Test methods that depend on `T: Unaligned`.
+        let mut u = Unalign::new(123u8);
+        assert_eq!(u.try_deref(), Ok(&123));
+        assert_eq!(u.try_deref_mut(), Ok(&mut 123));
+        assert_eq!(u.deref(), &123);
+        assert_eq!(u.deref_mut(), &mut 123);
+        *u = 21;
+        assert_eq!(u.get(), 21);
+
+        // Test that some `Unalign` functions and methods are `const`.
+        const _UNALIGN: Unalign<u64> = Unalign::new(0);
+        const _UNALIGN_PTR: *const u64 = _UNALIGN.get_ptr();
+        const _U64: u64 = _UNALIGN.into_inner();
+        // Make sure all code is considered "used".
+        //
+        // FIXME(https://github.com/rust-lang/rust/issues/104084): Remove this
+        // attribute.
+        #[allow(dead_code)]
+        const _: () = {
+            let x: Align<_, AU64> = Align::new(Unalign::new(AU64(123)));
+            // Make sure that `deref_unchecked` is `const`.
+            //
+            // SAFETY: The `Align<_, AU64>` guarantees proper alignment.
+            let au64 = unsafe { x.t.deref_unchecked() };
+            match au64 {
+                AU64(123) => {}
+                _ => const_unreachable!(),
+            }
+        };
+    }
+
+    #[test]
+    fn test_unalign_update() {
+        let mut u = Unalign::new(AU64(123));
+        u.update(|a| a.0 += 1);
+        assert_eq!(u.get(), AU64(124));
+
+        // Test that, even if the callback panics, the original is still
+        // correctly overwritten. Use a `Box` so that Miri is more likely to
+        // catch any unsoundness (which would likely result in two `Box`es for
+        // the same heap object, which is the sort of thing that Miri would
+        // probably catch).
+        let mut u = Unalign::new(Box::new(AU64(123)));
+        let res = std::panic::catch_unwind(AssertUnwindSafe(|| {
+            u.update(|a| {
+                a.0 += 1;
+                panic!();
+            })
+        }));
+        assert!(res.is_err());
+        assert_eq!(u.into_inner(), Box::new(AU64(124)));
+
+        // Test the align_of::<T>() == 1 optimization.
+        let mut u = Unalign::new([0u8, 1]);
+        u.update(|a| a[0] += 1);
+        assert_eq!(u.get(), [1u8, 1]);
+    }
+
+    #[test]
+    fn test_unalign_copy_clone() {
+        // Test that `Copy` and `Clone` do not cause soundness issues. This test
+        // is mainly meant to exercise UB that would be caught by Miri.
+
+        // `u.t` is definitely not validly-aligned for `AU64`'s alignment of 8.
+        let u = ForceUnalign::<_, AU64>::new(Unalign::new(AU64(123)));
+        #[allow(clippy::clone_on_copy)]
+        let v = u.t.clone();
+        let w = u.t;
+        assert_eq!(u.t.get(), v.get());
+        assert_eq!(u.t.get(), w.get());
+        assert_eq!(v.get(), w.get());
+    }
+
+    #[test]
+    fn test_unalign_trait_impls() {
+        let zero = Unalign::new(0u8);
+        let one = Unalign::new(1u8);
+
+        assert!(zero < one);
+        assert_eq!(PartialOrd::partial_cmp(&zero, &one), Some(Ordering::Less));
+        assert_eq!(Ord::cmp(&zero, &one), Ordering::Less);
+
+        assert_ne!(zero, one);
+        assert_eq!(zero, zero);
+        assert!(!PartialEq::eq(&zero, &one));
+        assert!(PartialEq::eq(&zero, &zero));
+
+        fn hash<T: Hash>(t: &T) -> u64 {
+            let mut h = std::collections::hash_map::DefaultHasher::new();
+            t.hash(&mut h);
+            h.finish()
+        }
+
+        assert_eq!(hash(&zero), hash(&0u8));
+        assert_eq!(hash(&one), hash(&1u8));
+
+        assert_eq!(format!("{:?}", zero), format!("{:?}", 0u8));
+        assert_eq!(format!("{:?}", one), format!("{:?}", 1u8));
+        assert_eq!(format!("{}", zero), format!("{}", 0u8));
+        assert_eq!(format!("{}", one), format!("{}", 1u8));
+    }
+
+    #[test]
+    #[allow(clippy::as_conversions)]
+    fn test_maybe_uninit() {
+        // int
+        {
+            let input = 42;
+            let uninit = MaybeUninit::new(input);
+            // SAFETY: `uninit` is in an initialized state
+            let output = unsafe { uninit.assume_init() };
+            assert_eq!(input, output);
+        }
+
+        // thin ref
+        {
+            let input = 42;
+            let uninit = MaybeUninit::new(&input);
+            // SAFETY: `uninit` is in an initialized state
+            let output = unsafe { uninit.assume_init() };
+            assert_eq!(&input as *const _, output as *const _);
+            assert_eq!(input, *output);
+        }
+
+        // wide ref
+        {
+            let input = [1, 2, 3, 4];
+            let uninit = MaybeUninit::new(&input[..]);
+            // SAFETY: `uninit` is in an initialized state
+            let output = unsafe { uninit.assume_init() };
+            assert_eq!(&input[..] as *const _, output as *const _);
+            assert_eq!(input, *output);
+        }
+    }
+    #[test]
+    fn test_maybe_uninit_uninit() {
+        let _uninit = MaybeUninit::<u8>::uninit();
+        // Cannot check value, but can check it compiles and runs
+    }
+
+    #[test]
+    #[cfg(feature = "alloc")]
+    fn test_maybe_uninit_new_boxed_uninit() {
+        let _boxed = MaybeUninit::<u8>::new_boxed_uninit(()).unwrap();
+    }
+
+    #[test]
+    fn test_maybe_uninit_debug() {
+        let uninit = MaybeUninit::<u8>::uninit();
+        assert!(format!("{:?}", uninit).contains("MaybeUninit"));
+    }
+}