WIP PROOF-OF-CONCEPT: handle all the fallout in the libs

Still working on this, but it seems to largely be a lot of `as usize` -> `.addr()`

Author: Gankra

library/alloc/src/lib.rs

@@ -158,6 +158,7 @@
 #![feature(rustc_allow_const_fn_unstable)]
 #![feature(rustc_attrs)]
 #![feature(staged_api)]
+#![feature(strict_provenance)]
 #![cfg_attr(test, feature(test))]
 #![feature(unboxed_closures)]
 #![feature(unsized_fn_params)]

library/alloc/src/rc.rs

@@ -2115,13 +2115,12 @@ impl<T> Weak<T> {
     #[rustc_const_unstable(feature = "const_weak_new", issue = "95091", reason = "recently added")]
     #[must_use]
     pub const fn new() -> Weak<T> {
-        Weak { ptr: unsafe { NonNull::new_unchecked(usize::MAX as *mut RcBox<T>) } }
+        Weak { ptr: unsafe { NonNull::new_unchecked(ptr::invalid_mut::<RcBox<T>>(usize::MAX)) } }
     }
 }
 
 pub(crate) fn is_dangling<T: ?Sized>(ptr: *mut T) -> bool {
-    let address = ptr as *mut () as usize;
-    address == usize::MAX
+    (ptr as *mut ()).addr() == usize::MAX
 }
 
 /// Helper type to allow accessing the reference counts without

library/alloc/src/slice.rs

@@ -1044,7 +1044,7 @@ where
     impl<T> Drop for MergeHole<T> {
         fn drop(&mut self) {
             // `T` is not a zero-sized type, so it's okay to divide by its size.
-            let len = (self.end as usize - self.start as usize) / mem::size_of::<T>();
+            let len = (self.end.addr() - self.start.addr()) / mem::size_of::<T>();
             unsafe {
                 ptr::copy_nonoverlapping(self.start, self.dest, len);
             }

library/alloc/src/sync.rs

@@ -1745,7 +1745,7 @@ impl<T> Weak<T> {
     #[rustc_const_unstable(feature = "const_weak_new", issue = "95091", reason = "recently added")]
     #[must_use]
     pub const fn new() -> Weak<T> {
-        Weak { ptr: unsafe { NonNull::new_unchecked(usize::MAX as *mut ArcInner<T>) } }
+        Weak { ptr: unsafe { NonNull::new_unchecked(ptr::invalid_mut::<ArcInner<T>>(usize::MAX)) } }
     }
 }
 

library/alloc/src/vec/into_iter.rs

@@ -154,7 +154,7 @@ impl<T, A: Allocator> Iterator for IntoIter<T, A> {
     #[inline]
     fn size_hint(&self) -> (usize, Option<usize>) {
         let exact = if mem::size_of::<T>() == 0 {
-            (self.end as usize).wrapping_sub(self.ptr as usize)
+            self.end.addr().wrapping_sub(self.ptr.addr())
         } else {
             unsafe { self.end.offset_from(self.ptr) as usize }
         };

library/core/src/alloc/layout.rs

@@ -194,7 +194,7 @@ impl Layout {
     #[inline]
     pub const fn dangling(&self) -> NonNull<u8> {
         // SAFETY: align is guaranteed to be non-zero
-        unsafe { NonNull::new_unchecked(self.align() as *mut u8) }
+        unsafe { NonNull::new_unchecked(crate::ptr::invalid_mut::<u8>(self.align())) }
     }
 
     /// Creates a layout describing the record that can hold a value

library/core/src/fmt/mod.rs

@@ -352,7 +352,11 @@ impl<'a> ArgumentV1<'a> {
     }
 
     fn as_usize(&self) -> Option<usize> {
-        if self.formatter as usize == USIZE_MARKER as usize {
+        // We are type punning a bit here: USIZE_MARKER only takes an &usize but
+        // formatter takes an &Opaque. Rust understandably doesn't think we should compare
+        // the function pointers if they don't have the same signature, so we cast to
+        // pointers to convince it that we know what we're doing.
+        if self.formatter as *mut u8 == USIZE_MARKER as *mut u8 {
             // SAFETY: The `formatter` field is only set to USIZE_MARKER if
             // the value is a usize, so this is safe
             Some(unsafe { *(self.value as *const _ as *const usize) })
@@ -2246,7 +2250,7 @@ impl<T: ?Sized> Pointer for *const T {
             }
             f.flags |= 1 << (FlagV1::Alternate as u32);
 
-            let ret = LowerHex::fmt(&(ptr as usize), f);
+            let ret = LowerHex::fmt(&(ptr.addr()), f);
 
             f.width = old_width;
             f.flags = old_flags;

library/core/src/hash/mod.rs

@@ -793,7 +793,7 @@ mod impls {
         #[inline]
         fn hash<H: Hasher>(&self, state: &mut H) {
             let (address, metadata) = self.to_raw_parts();
-            state.write_usize(address as usize);
+            state.write_usize(address.addr());
             metadata.hash(state);
         }
     }
@@ -803,7 +803,7 @@ mod impls {
         #[inline]
         fn hash<H: Hasher>(&self, state: &mut H) {
             let (address, metadata) = self.to_raw_parts();
-            state.write_usize(address as usize);
+            state.write_usize(address.addr());
             metadata.hash(state);
         }
     }

library/core/src/intrinsics.rs

@@ -987,8 +987,8 @@ extern "rust-intrinsic" {
     ///     std::mem::transmute::<&i32, usize>(ptr)
     /// };
     ///
-    /// // Use an `as` cast instead
-    /// let ptr_num_cast = ptr as *const i32 as usize;
+    /// // Use `.addr()` instead
+    /// let ptr_num_cast = (ptr as *const i32).addr();
     /// ```
     ///
     /// Turning a `*mut T` into an `&mut T`:
@@ -1972,15 +1972,15 @@ extern "rust-intrinsic" {
 /// Checks whether `ptr` is properly aligned with respect to
 /// `align_of::<T>()`.
 pub(crate) fn is_aligned_and_not_null<T>(ptr: *const T) -> bool {
-    !ptr.is_null() && ptr as usize % mem::align_of::<T>() == 0
+    !ptr.is_null() && ptr.addr() % mem::align_of::<T>() == 0
 }
 
 /// Checks whether the regions of memory starting at `src` and `dst` of size
 /// `count * size_of::<T>()` do *not* overlap.
 #[cfg(debug_assertions)]
 pub(crate) fn is_nonoverlapping<T>(src: *const T, dst: *const T, count: usize) -> bool {
-    let src_usize = src as usize;
-    let dst_usize = dst as usize;
+    let src_usize = src.addr();
+    let dst_usize = dst.addr();
     let size = mem::size_of::<T>().checked_mul(count).unwrap();
     let diff = if src_usize > dst_usize { src_usize - dst_usize } else { dst_usize - src_usize };
     // If the absolute distance between the ptrs is at least as big as the size of the buffer,

library/core/src/ptr/non_null.rs

@@ -90,7 +90,7 @@ impl<T: Sized> NonNull<T> {
         // to a *mut T. Therefore, `ptr` is not null and the conditions for
         // calling new_unchecked() are respected.
         unsafe {
-            let ptr = mem::align_of::<T>() as *mut T;
+            let ptr = crate::ptr::invalid_mut::<T>(mem::align_of::<T>());
             NonNull::new_unchecked(ptr)
         }
     }
@@ -469,7 +469,7 @@ impl<T> NonNull<[T]> {
     /// use std::ptr::NonNull;
     ///
     /// let slice: NonNull<[i8]> = NonNull::slice_from_raw_parts(NonNull::dangling(), 3);
-    /// assert_eq!(slice.as_non_null_ptr(), NonNull::new(1 as *mut i8).unwrap());
+    /// assert_eq!(slice.as_non_null_ptr(), NonNull::<i8>::dangling());
     /// ```
     #[inline]
     #[must_use]
@@ -489,7 +489,7 @@ impl<T> NonNull<[T]> {
     /// use std::ptr::NonNull;
     ///
     /// let slice: NonNull<[i8]> = NonNull::slice_from_raw_parts(NonNull::dangling(), 3);
-    /// assert_eq!(slice.as_mut_ptr(), 1 as *mut i8);
+    /// assert_eq!(slice.as_mut_ptr(), NonNull::<i8>::dangling());
     /// ```
     #[inline]
     #[must_use]

library/core/src/ptr/unique.rs

@@ -73,7 +73,7 @@ impl<T: Sized> Unique<T> {
     pub const fn dangling() -> Self {
         // SAFETY: mem::align_of() returns a valid, non-null pointer. The
         // conditions to call new_unchecked() are thus respected.
-        unsafe { Unique::new_unchecked(mem::align_of::<T>() as *mut T) }
+        unsafe { Unique::new_unchecked(crate::ptr::invalid_mut::<T>(mem::align_of::<T>())) }
     }
 }
 

library/core/src/slice/ascii.rs

@@ -293,17 +293,17 @@ fn is_ascii(s: &[u8]) -> bool {
     // Paranoia check about alignment, since we're about to do a bunch of
     // unaligned loads. In practice this should be impossible barring a bug in
     // `align_offset` though.
-    debug_assert_eq!((word_ptr as usize) % mem::align_of::<usize>(), 0);
+    debug_assert_eq!((word_ptr.addr()) % mem::align_of::<usize>(), 0);
 
     // Read subsequent words until the last aligned word, excluding the last
     // aligned word by itself to be done in tail check later, to ensure that
     // tail is always one `usize` at most to extra branch `byte_pos == len`.
     while byte_pos < len - USIZE_SIZE {
         debug_assert!(
             // Sanity check that the read is in bounds
-            (word_ptr as usize + USIZE_SIZE) <= (start.wrapping_add(len) as usize) &&
+            (word_ptr.addr() + USIZE_SIZE) <= (start.wrapping_add(len).addr()) &&
             // And that our assumptions about `byte_pos` hold.
-            (word_ptr as usize) - (start as usize) == byte_pos
+            (word_ptr.addr()) - (start.addr()) == byte_pos
         );
 
         // SAFETY: We know `word_ptr` is properly aligned (because of

library/core/src/slice/iter/macros.rs

@@ -20,13 +20,13 @@ macro_rules! len {
         if size == 0 {
             // This _cannot_ use `unchecked_sub` because we depend on wrapping
             // to represent the length of long ZST slice iterators.
-            ($self.end as usize).wrapping_sub(start.as_ptr() as usize)
+            ($self.end.addr()).wrapping_sub(start.as_ptr().addr())
         } else {
             // We know that `start <= end`, so can do better than `offset_from`,
             // which needs to deal in signed.  By setting appropriate flags here
             // we can tell LLVM this, which helps it remove bounds checks.
             // SAFETY: By the type invariant, `start <= end`
-            let diff = unsafe { unchecked_sub($self.end as usize, start.as_ptr() as usize) };
+            let diff = unsafe { unchecked_sub($self.end.addr(), start.as_ptr().addr()) };
             // By also telling LLVM that the pointers are apart by an exact
             // multiple of the type size, it can optimize `len() == 0` down to
             // `start == end` instead of `(end - start) < size`.

library/core/src/slice/sort.rs

@@ -269,7 +269,7 @@ where
     // Returns the number of elements between pointers `l` (inclusive) and `r` (exclusive).
     fn width<T>(l: *mut T, r: *mut T) -> usize {
         assert!(mem::size_of::<T>() > 0);
-        (r as usize - l as usize) / mem::size_of::<T>()
+        (r.addr() - l.addr()) / mem::size_of::<T>()
     }
 
     loop {

library/panic_unwind/src/dwarf/eh.rs

@@ -10,6 +10,16 @@
 
 #![allow(non_upper_case_globals)]
 #![allow(unused)]
+// FIXME(strict_provenance_magic): the unwinder has special permissions and semantics.
+//
+// This is at worst an Interesting Case Study that is worth doing a deep dive on.
+//
+// This implementation pretty aggressively puns *offsets* and actual addresses. It can
+// probably be made "strict" but requires a lot of very careful thinking about the
+// precise semantics of this format (I think it's basically shoe-horning a bunch of
+// things into a common path for simplicity, when in reality there are many impossible
+// combinations).
+#![cfg_attr(not(bootstrap), allow(fuzzy_provenance_casts))]
 
 use crate::dwarf::DwarfReader;
 use core::mem;

library/panic_unwind/src/seh.rs

@@ -45,6 +45,12 @@
 //! [llvm]: https://llvm.org/docs/ExceptionHandling.html#background-on-windows-exceptions
 
 #![allow(nonstandard_style)]
+// FIXME(strict_provenance_magic): the unwinder has special permissions and semantics.
+//
+// This is at worst an Interesting Case Study that is worth doing a deep dive on.
+//
+// I haven't looked closely at this implementation yet.
+#![cfg_attr(not(bootstrap), allow(fuzzy_provenance_casts))]
 
 use alloc::boxed::Box;
 use core::any::Any;

library/std/src/backtrace.rs

@@ -293,7 +293,7 @@ impl Backtrace {
         if !Backtrace::enabled() {
             return Backtrace { inner: Inner::Disabled };
         }
-        Backtrace::create(Backtrace::capture as usize)
+        Backtrace::create((Backtrace::capture as *mut ()).addr())
     }
 
     /// Forcibly captures a full backtrace, regardless of environment variable
@@ -308,7 +308,7 @@ impl Backtrace {
     /// parts of code.
     #[inline(never)] // want to make sure there's a frame here to remove
     pub fn force_capture() -> Backtrace {
-        Backtrace::create(Backtrace::force_capture as usize)
+        Backtrace::create((Backtrace::force_capture as *mut ()).addr())
     }
 
     /// Forcibly captures a disabled backtrace, regardless of environment
@@ -330,7 +330,7 @@ impl Backtrace {
                     frame: RawFrame::Actual(frame.clone()),
                     symbols: Vec::new(),
                 });
-                if frame.symbol_address() as usize == ip && actual_start.is_none() {
+                if frame.symbol_address().addr() == ip && actual_start.is_none() {
                     actual_start = Some(frames.len());
                 }
                 true
@@ -493,7 +493,7 @@ impl RawFrame {
         match self {
             RawFrame::Actual(frame) => frame.ip(),
             #[cfg(test)]
-            RawFrame::Fake => 1 as *mut c_void,
+            RawFrame::Fake => ptr::invalid_mut(1),
         }
     }
 }

library/std/src/io/error/repr_bitpacked.rs

@@ -105,7 +105,7 @@
 use super::{Custom, ErrorData, ErrorKind, SimpleMessage};
 use alloc::boxed::Box;
 use core::mem::{align_of, size_of};
-use core::ptr::NonNull;
+use core::ptr::{self, NonNull};
 
 // The 2 least-significant bits are used as tag.
 const TAG_MASK: usize = 0b11;
@@ -126,7 +126,7 @@ impl Repr {
         let p = Box::into_raw(b).cast::<u8>();
         // Should only be possible if an allocator handed out a pointer with
         // wrong alignment.
-        debug_assert_eq!((p as usize & TAG_MASK), 0);
+        debug_assert_eq!((p.addr() & TAG_MASK), 0);
         // Note: We know `TAG_CUSTOM <= size_of::<Custom>()` (static_assert at
         // end of file), and both the start and end of the expression must be
         // valid without address space wraparound due to `Box`'s semantics.
@@ -156,7 +156,7 @@ impl Repr {
     pub(super) fn new_os(code: i32) -> Self {
         let utagged = ((code as usize) << 32) | TAG_OS;
         // Safety: `TAG_OS` is not zero, so the result of the `|` is not 0.
-        let res = Self(unsafe { NonNull::new_unchecked(utagged as *mut ()) });
+        let res = Self(unsafe { NonNull::new_unchecked(ptr::invalid_mut(utagged)) });
         // quickly smoke-check we encoded the right thing (This generally will
         // only run in libstd's tests, unless the user uses -Zbuild-std)
         debug_assert!(
@@ -170,7 +170,7 @@ impl Repr {
     pub(super) fn new_simple(kind: ErrorKind) -> Self {
         let utagged = ((kind as usize) << 32) | TAG_SIMPLE;
         // Safety: `TAG_SIMPLE` is not zero, so the result of the `|` is not 0.
-        let res = Self(unsafe { NonNull::new_unchecked(utagged as *mut ()) });
+        let res = Self(unsafe { NonNull::new_unchecked(ptr::invalid_mut(utagged)) });
         // quickly smoke-check we encoded the right thing (This generally will
         // only run in libstd's tests, unless the user uses -Zbuild-std)
         debug_assert!(
@@ -228,7 +228,7 @@ unsafe fn decode_repr<C, F>(ptr: NonNull<()>, make_custom: F) -> ErrorData<C>
 where
     F: FnOnce(*mut Custom) -> C,
 {
-    let bits = ptr.as_ptr() as usize;
+    let bits = ptr.as_ptr().addr();
     match bits & TAG_MASK {
         TAG_OS => {
             let code = ((bits as i64) >> 32) as i32;

library/std/src/lib.rs

@@ -275,6 +275,7 @@
 #![feature(extend_one)]
 #![feature(float_minimum_maximum)]
 #![feature(format_args_nl)]
+#![feature(strict_provenance)]
 #![feature(get_mut_unchecked)]
 #![feature(hashmap_internals)]
 #![feature(int_error_internals)]

library/std/src/os/windows/io/handle.rs

@@ -9,6 +9,7 @@ use crate::fs;
 use crate::io;
 use crate::marker::PhantomData;
 use crate::mem::forget;
+use crate::ptr;
 use crate::sys::c;
 use crate::sys::cvt;
 use crate::sys_common::{AsInner, FromInner, IntoInner};
@@ -174,7 +175,7 @@ impl OwnedHandle {
             return unsafe { Ok(Self::from_raw_handle(handle)) };
         }
 
-        let mut ret = 0 as c::HANDLE;
+        let mut ret = ptr::null_mut();
         cvt(unsafe {
             let cur_proc = c::GetCurrentProcess();
             c::DuplicateHandle(

library/std/src/os/windows/io/socket.rs

@@ -129,6 +129,8 @@ impl OwnedSocket {
         }
     }
 
+    // FIXME(strict_provenance_magic): we defined RawSocket to be a u64 ;-;
+    #[cfg_attr(not(bootstrap), allow(fuzzy_provenance_casts))]
     #[cfg(not(target_vendor = "uwp"))]
     pub(crate) fn set_no_inherit(&self) -> io::Result<()> {
         cvt(unsafe {

library/std/src/path.rs

@@ -1449,8 +1449,8 @@ impl PathBuf {
         };
 
         // truncate until right after the file stem
-        let end_file_stem = file_stem[file_stem.len()..].as_ptr() as usize;
-        let start = os_str_as_u8_slice(&self.inner).as_ptr() as usize;
+        let end_file_stem = file_stem[file_stem.len()..].as_ptr().addr();
+        let start = os_str_as_u8_slice(&self.inner).as_ptr().addr();
         let v = self.as_mut_vec();
         v.truncate(end_file_stem.wrapping_sub(start));