Auto merge of #2516 - RalfJung:read-pointer-as-bytes, r=RalfJung

Adjust for supporting more implicit ptr-to-int transmutation

This is the Miri side of rust-lang/rust#101101.
Fixes #2456.

Author: bors

rust-version

@@ -1 +1 @@
-94b2b15e63c5d2b2a6a0910e3dae554ce9415bf9
+0631ea5d73f4a3199c776687b12c20c50a91f0d2

src/diagnostics.rs

@@ -224,13 +224,14 @@ pub fn report_error<'tcx, 'mir>(
             let helps = match e.kind() {
                 Unsupported(
                     UnsupportedOpInfo::ThreadLocalStatic(_) |
-                    UnsupportedOpInfo::ReadExternStatic(_)
+                    UnsupportedOpInfo::ReadExternStatic(_) |
+                    UnsupportedOpInfo::PartialPointerOverwrite(_) | // we make memory uninit instead
+                    UnsupportedOpInfo::ReadPointerAsBytes
                 ) =>
                     panic!("Error should never be raised by Miri: {kind:?}", kind = e.kind()),
                 Unsupported(
                     UnsupportedOpInfo::Unsupported(_) |
-                    UnsupportedOpInfo::PartialPointerOverwrite(_) |
-                    UnsupportedOpInfo::ReadPointerAsBytes
+                    UnsupportedOpInfo::PartialPointerCopy(_)
                 ) =>
                     vec![(None, format!("this is likely not a bug in the program; it indicates that the program performed an operation that the interpreter does not support"))],
                 UndefinedBehavior(UndefinedBehaviorInfo::AlignmentCheckFailed { .. })

src/helpers.rs

@@ -757,7 +757,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         }
 
         // Step 2: get the bytes.
-        this.read_bytes_ptr(ptr, len)
+        this.read_bytes_ptr_strip_provenance(ptr, len)
     }
 
     fn read_wide_str(&self, mut ptr: Pointer<Option<Provenance>>) -> InterpResult<'tcx, Vec<u16>> {

src/machine.rs

@@ -205,6 +205,21 @@ impl interpret::Provenance for Provenance {
             Provenance::Wildcard => None,
         }
     }
+
+    fn join(left: Option<Self>, right: Option<Self>) -> Option<Self> {
+        match (left, right) {
+            // If both are the *same* concrete tag, that is the result.
+            (
+                Some(Provenance::Concrete { alloc_id: left_alloc, sb: left_sb }),
+                Some(Provenance::Concrete { alloc_id: right_alloc, sb: right_sb }),
+            ) if left_alloc == right_alloc && left_sb == right_sb => left,
+            // If one side is a wildcard, the best possible outcome is that it is equal to the other
+            // one, and we use that.
+            (Some(Provenance::Wildcard), o) | (o, Some(Provenance::Wildcard)) => o,
+            // Otherwise, fall back to `None`.
+            _ => None,
+        }
+    }
 }
 
 impl fmt::Debug for ProvenanceExtra {

src/shims/foreign_items.rs

@@ -560,8 +560,8 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
                 let n = Size::from_bytes(this.read_scalar(n)?.to_machine_usize(this)?);
 
                 let result = {
-                    let left_bytes = this.read_bytes_ptr(left, n)?;
-                    let right_bytes = this.read_bytes_ptr(right, n)?;
+                    let left_bytes = this.read_bytes_ptr_strip_provenance(left, n)?;
+                    let right_bytes = this.read_bytes_ptr_strip_provenance(right, n)?;
 
                     use std::cmp::Ordering::*;
                     match left_bytes.cmp(right_bytes) {
@@ -583,7 +583,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
                 let val = val as u8;
 
                 if let Some(idx) = this
-                    .read_bytes_ptr(ptr, Size::from_bytes(num))?
+                    .read_bytes_ptr_strip_provenance(ptr, Size::from_bytes(num))?
                     .iter()
                     .rev()
                     .position(|&c| c == val)
@@ -606,7 +606,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
                 let val = val as u8;
 
                 let idx = this
-                    .read_bytes_ptr(ptr, Size::from_bytes(num))?
+                    .read_bytes_ptr_strip_provenance(ptr, Size::from_bytes(num))?
                     .iter()
                     .position(|&c| c == val);
                 if let Some(idx) = idx {

src/shims/unix/fs.rs

@@ -761,7 +761,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
         let communicate = this.machine.communicate();
 
         if let Some(file_descriptor) = this.machine.file_handler.handles.get(&fd) {
-            let bytes = this.read_bytes_ptr(buf, Size::from_bytes(count))?;
+            let bytes = this.read_bytes_ptr_strip_provenance(buf, Size::from_bytes(count))?;
             let result =
                 file_descriptor.write(communicate, bytes)?.map(|c| i64::try_from(c).unwrap());
             this.try_unwrap_io_result(result)

src/shims/windows/dlsym.rs

@@ -78,7 +78,8 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriEvalContextExt<'mir, 'tcx
                     // stdout/stderr
                     use std::io::{self, Write};
 
-                    let buf_cont = this.read_bytes_ptr(buf, Size::from_bytes(u64::from(n)))?;
+                    let buf_cont =
+                        this.read_bytes_ptr_strip_provenance(buf, Size::from_bytes(u64::from(n)))?;
                     let res = if this.machine.mute_stdout_stderr {
                         Ok(buf_cont.len())
                     } else if handle == -11 {

tests/fail/copy_half_a_pointer.rs

@@ -0,0 +1,20 @@
+#![allow(dead_code)]
+
+// We use packed structs to get around alignment restrictions
+#[repr(packed)]
+struct Data {
+    pad: u8,
+    ptr: &'static i32,
+}
+
+static G: i32 = 0;
+
+fn main() {
+    let mut d = Data { pad: 0, ptr: &G };
+
+    // Get a pointer to the beginning of the Data struct (one u8 byte, then the pointer bytes).
+    let d_alias = &mut d as *mut _ as *mut *const u8;
+    unsafe {
+        let _x = d_alias.read_unaligned(); //~ERROR: unable to copy parts of a pointer
+    }
+}

tests/fail/copy_half_a_pointer.stderr

@@ -0,0 +1,14 @@
+error: unsupported operation: unable to copy parts of a pointer from memory at ALLOC+0x8
+  --> $DIR/copy_half_a_pointer.rs:LL:CC
+   |
+LL |         let _x = d_alias.read_unaligned();
+   |                  ^^^^^^^^^^^^^^^^^^^^^^^^ unable to copy parts of a pointer from memory at ALLOC+0x8
+   |
+   = help: this is likely not a bug in the program; it indicates that the program performed an operation that the interpreter does not support
+   = note: backtrace:
+   = note: inside `main` at $DIR/copy_half_a_pointer.rs:LL:CC
+
+note: some details are omitted, run with `MIRIFLAGS=-Zmiri-backtrace=full` for a verbose backtrace
+
+error: aborting due to previous error
+

tests/fail/intrinsics/raw_eq_on_ptr.rs

@@ -6,6 +6,5 @@ extern "rust-intrinsic" {
 
 fn main() {
     let x = &0;
-    // FIXME: the error message is not great (should be UB rather than 'unsupported')
-    unsafe { raw_eq(&x, &x) }; //~ERROR: unsupported operation
+    unsafe { raw_eq(&x, &x) }; //~ERROR: `raw_eq` on bytes with provenance
 }

tests/fail/intrinsics/raw_eq_on_ptr.stderr

@@ -1,10 +1,11 @@
-error: unsupported operation: unable to turn pointer into raw bytes
+error: Undefined Behavior: `raw_eq` on bytes with provenance
   --> $DIR/raw_eq_on_ptr.rs:LL:CC
    |
 LL |     unsafe { raw_eq(&x, &x) };
-   |              ^^^^^^^^^^^^^^ unable to turn pointer into raw bytes
+   |              ^^^^^^^^^^^^^^ `raw_eq` on bytes with provenance
    |
-   = help: this is likely not a bug in the program; it indicates that the program performed an operation that the interpreter does not support
+   = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior
+   = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information
    = note: backtrace:
    = note: inside `main` at $DIR/raw_eq_on_ptr.rs:LL:CC
 

tests/fail/reading_half_a_pointer.rs

@@ -24,6 +24,7 @@ fn main() {
     // starts 1 byte to the right, so using it would actually be wrong!
     let d_alias = &mut w.data as *mut _ as *mut *const u8;
     unsafe {
-        let _x = *d_alias; //~ ERROR: unable to turn pointer into raw bytes
+        let x = *d_alias;
+        let _val = *x; //~ERROR: is a dangling pointer (it has no provenance)
     }
 }

tests/fail/reading_half_a_pointer.stderr

@@ -1,10 +1,11 @@
-error: unsupported operation: unable to turn pointer into raw bytes
+error: Undefined Behavior: dereferencing pointer failed: $HEX[noalloc] is a dangling pointer (it has no provenance)
   --> $DIR/reading_half_a_pointer.rs:LL:CC
    |
-LL |         let _x = *d_alias;
-   |                  ^^^^^^^^ unable to turn pointer into raw bytes
+LL |         let _val = *x;
+   |                    ^^ dereferencing pointer failed: $HEX[noalloc] is a dangling pointer (it has no provenance)
    |
-   = help: this is likely not a bug in the program; it indicates that the program performed an operation that the interpreter does not support
+   = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior
+   = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information
    = note: backtrace:
    = note: inside `main` at $DIR/reading_half_a_pointer.rs:LL:CC
 

tests/fail/transmute_fat1.rs

@@ -0,0 +1,52 @@
+#![feature(strict_provenance)]
+use std::{mem, ptr};
+
+fn t1() {
+    // If we are careful, we can exploit data layout...
+    // This is a tricky case since we are transmuting a ScalarPair type to a non-ScalarPair type.
+    let raw = unsafe { mem::transmute::<&[u8], [*const u8; 2]>(&[42]) };
+    let ptr: *const u8 = unsafe { mem::transmute_copy(&raw) };
+    assert_eq!(unsafe { *ptr }, 42);
+}
+
+#[cfg(target_pointer_width = "64")]
+const PTR_SIZE: usize = 8;
+#[cfg(target_pointer_width = "32")]
+const PTR_SIZE: usize = 4;
+
+fn t2() {
+    let bad = unsafe { mem::transmute::<&[u8], [u8; 2 * PTR_SIZE]>(&[1u8]) };
+    let _val = bad[0] + bad[bad.len() - 1];
+}
+
+fn ptr_integer_array() {
+    let r = &mut 42;
+    let _i: [usize; 1] = unsafe { mem::transmute(r) };
+
+    let _x: [u8; PTR_SIZE] = unsafe { mem::transmute(&0) };
+}
+
+fn ptr_in_two_halves() {
+    unsafe {
+        let ptr = &0 as *const i32;
+        let arr = [ptr; 2];
+        // We want to do a scalar read of a pointer at offset PTR_SIZE/2 into this array. But we
+        // cannot use a packed struct or `read_unaligned`, as those use the memcpy code path in
+        // Miri. So instead we shift the entire array by a bit and then the actual read we want to
+        // do is perfectly aligned.
+        let mut target_arr = [ptr::null::<i32>(); 3];
+        let target = target_arr.as_mut_ptr().cast::<u8>();
+        target.add(PTR_SIZE / 2).cast::<[*const i32; 2]>().write_unaligned(arr);
+        // Now target_arr[1] is a mix of the two `ptr` we had stored in `arr`.
+        let strange_ptr = target_arr[1];
+        // Check that the provenance works out.
+        assert_eq!(*strange_ptr.with_addr(ptr.addr()), 0);
+    }
+}
+
+fn main() {
+    t1();
+    t2();
+    ptr_integer_array();
+    ptr_in_two_halves();
+}