Auto merge of #3319 - bjorn3:some_more_shims, r=RalfJung

Directly implement native exception raise methods in miri

This implements the `_Unwind_RaiseException` function used on pretty much every unix system for starting unwinding. This allows removing the miri special case from libpanic_unwind for unix.

Windows still needs `miri_start_unwind` as SEH unwinding isn't supported by miri. Unlike DWARF unwinding, SEH preserves all stack frames until right after the do_catch function has executed. Because of this panic_unwind stack allocates the exception object. Miri can't currently model unwinding without destroying stack frames and as such will report a use-after-free of the exception object.

src/tools/miri/ci/ci.sh

@@ -145,7 +145,7 @@ case $HOST_TARGET in
     TEST_TARGET=s390x-unknown-linux-gnu run_tests # big-endian architecture of choice
     # Partially supported targets (tier 2)
     BASIC="empty_main integer vec string btreemap hello hashmap heap_alloc align" # ensures we have the basics: stdout/stderr, system allocator, randomness (for HashMap initialization)
-    UNIX="panic/panic concurrency/simple atomic libc-mem libc-misc libc-random env num_cpus" # the things that are very similar across all Unixes, and hence easily supported there
+    UNIX="panic/panic panic/unwind concurrency/simple atomic libc-mem libc-misc libc-random env num_cpus" # the things that are very similar across all Unixes, and hence easily supported there
     TEST_TARGET=x86_64-unknown-freebsd run_tests_minimal $BASIC $UNIX threadname libc-time fs
     TEST_TARGET=i686-unknown-freebsd   run_tests_minimal $BASIC $UNIX threadname libc-time fs
     TEST_TARGET=x86_64-unknown-illumos run_tests_minimal $BASIC $UNIX pthread-sync

src/tools/miri/src/intrinsics/atomic.rs

@@ -116,7 +116,7 @@ fn fence_ord(ord: &str) -> AtomicFenceOrd {
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }
 

src/tools/miri/src/intrinsics/mod.rs

@@ -26,7 +26,7 @@ fn call_intrinsic(
         args: &[OpTy<'tcx, Provenance>],
         dest: &MPlaceTy<'tcx, Provenance>,
         ret: Option<mir::BasicBlock>,
-        _unwind: mir::UnwindAction,
+        unwind: mir::UnwindAction,
     ) -> InterpResult<'tcx, Option<ty::Instance<'tcx>>> {
         let this = self.eval_context_mut();
 
@@ -62,11 +62,16 @@ fn call_intrinsic(
                     args: instance.args,
                 }))
             }
-            EmulateItemResult::NeedsJumping => {
+            EmulateItemResult::NeedsReturn => {
                 trace!("{:?}", this.dump_place(&dest.clone().into()));
                 this.return_to_block(ret)?;
                 Ok(None)
             }
+            EmulateItemResult::NeedsUnwind => {
+                // Jump to the unwind block to begin unwinding.
+                this.unwind_to_block(unwind)?;
+                Ok(None)
+            }
             EmulateItemResult::AlreadyJumped => Ok(None),
         }
     }
@@ -441,6 +446,6 @@ fn emulate_intrinsic_by_name(
             _ => return Ok(EmulateItemResult::NotSupported),
         }
 
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/intrinsics/simd.rs

@@ -746,7 +746,7 @@ enum Op {
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 
     fn fminmax_op(

src/tools/miri/src/shims/alloc.rs

@@ -87,7 +87,7 @@ fn emulate_allocator(
             }
             AllocatorKind::Default => {
                 default(this)?;
-                Ok(EmulateItemResult::NeedsJumping)
+                Ok(EmulateItemResult::NeedsReturn)
             }
         }
     }

src/tools/miri/src/shims/foreign_items.rs

@@ -82,11 +82,15 @@ fn emulate_foreign_item(
         }
 
         // The rest either implements the logic, or falls back to `lookup_exported_symbol`.
-        match this.emulate_foreign_item_inner(link_name, abi, args, dest, unwind)? {
-            EmulateItemResult::NeedsJumping => {
+        match this.emulate_foreign_item_inner(link_name, abi, args, dest)? {
+            EmulateItemResult::NeedsReturn => {
                 trace!("{:?}", this.dump_place(&dest.clone().into()));
                 this.return_to_block(ret)?;
             }
+            EmulateItemResult::NeedsUnwind => {
+                // Jump to the unwind block to begin unwinding.
+                this.unwind_to_block(unwind)?;
+            }
             EmulateItemResult::AlreadyJumped => (),
             EmulateItemResult::NotSupported => {
                 if let Some(body) = this.lookup_exported_symbol(link_name)? {
@@ -206,7 +210,6 @@ fn emulate_foreign_item_inner(
         abi: Abi,
         args: &[OpTy<'tcx, Provenance>],
         dest: &MPlaceTy<'tcx, Provenance>,
-        unwind: mir::UnwindAction,
     ) -> InterpResult<'tcx, EmulateItemResult> {
         let this = self.eval_context_mut();
 
@@ -218,7 +221,7 @@ fn emulate_foreign_item_inner(
             // by the specified `.so` file; we should continue and check if it corresponds to
             // a provided shim.
             if this.call_native_fn(link_name, dest, args)? {
-                return Ok(EmulateItemResult::NeedsJumping);
+                return Ok(EmulateItemResult::NeedsReturn);
             }
         }
 
@@ -263,9 +266,9 @@ fn emulate_foreign_item_inner(
         match link_name.as_str() {
             // Miri-specific extern functions
             "miri_start_unwind" => {
-                // `check_shim` happens inside `handle_miri_start_unwind`.
-                this.handle_miri_start_unwind(abi, link_name, args, unwind)?;
-                return Ok(EmulateItemResult::AlreadyJumped);
+                let [payload] = this.check_shim(abi, Abi::Rust, link_name, args)?;
+                this.handle_miri_start_unwind(payload)?;
+                return Ok(EmulateItemResult::NeedsUnwind);
             }
             "miri_run_provenance_gc" => {
                 let [] = this.check_shim(abi, Abi::Rust, link_name, args)?;
@@ -480,7 +483,7 @@ fn emulate_foreign_item_inner(
                     "__rust_alloc" => return this.emulate_allocator(default),
                     "miri_alloc" => {
                         default(this)?;
-                        return Ok(EmulateItemResult::NeedsJumping);
+                        return Ok(EmulateItemResult::NeedsReturn);
                     }
                     _ => unreachable!(),
                 }
@@ -540,7 +543,7 @@ fn emulate_foreign_item_inner(
                     }
                     "miri_dealloc" => {
                         default(this)?;
-                        return Ok(EmulateItemResult::NeedsJumping);
+                        return Ok(EmulateItemResult::NeedsReturn);
                     }
                     _ => unreachable!(),
                 }
@@ -961,6 +964,6 @@ fn emulate_foreign_item_inner(
         };
         // We only fall through to here if we did *not* hit the `_` arm above,
         // i.e., if we actually emulated the function with one of the shims.
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/mod.rs

@@ -22,8 +22,10 @@
 /// What needs to be done after emulating an item (a shim or an intrinsic) is done.
 pub enum EmulateItemResult {
     /// The caller is expected to jump to the return block.
-    NeedsJumping,
-    /// Jumping has already been taken care of.
+    NeedsReturn,
+    /// The caller is expected to jump to the unwind block.
+    NeedsUnwind,
+    /// Jumping to the next block has already been taken care of.
     AlreadyJumped,
     /// The item is not supported.
     NotSupported,

src/tools/miri/src/shims/panic.rs

@@ -13,7 +13,6 @@
 
 use rustc_ast::Mutability;
 use rustc_middle::{mir, ty};
-use rustc_span::Symbol;
 use rustc_target::spec::abi::Abi;
 use rustc_target::spec::PanicStrategy;
 
@@ -46,25 +45,15 @@ impl<'mir, 'tcx: 'mir> EvalContextExt<'mir, 'tcx> for crate::MiriInterpCx<'mir,
 pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
     /// Handles the special `miri_start_unwind` intrinsic, which is called
     /// by libpanic_unwind to delegate the actual unwinding process to Miri.
-    fn handle_miri_start_unwind(
-        &mut self,
-        abi: Abi,
-        link_name: Symbol,
-        args: &[OpTy<'tcx, Provenance>],
-        unwind: mir::UnwindAction,
-    ) -> InterpResult<'tcx> {
+    fn handle_miri_start_unwind(&mut self, payload: &OpTy<'tcx, Provenance>) -> InterpResult<'tcx> {
         let this = self.eval_context_mut();
 
         trace!("miri_start_unwind: {:?}", this.frame().instance);
 
-        // Get the raw pointer stored in arg[0] (the panic payload).
-        let [payload] = this.check_shim(abi, Abi::Rust, link_name, args)?;
         let payload = this.read_scalar(payload)?;
         let thread = this.active_thread_mut();
         thread.panic_payloads.push(payload);
 
-        // Jump to the unwind block to begin unwinding.
-        this.unwind_to_block(unwind)?;
         Ok(())
     }
 

src/tools/miri/src/shims/unix/android/foreign_items.rs

@@ -27,6 +27,6 @@ fn emulate_foreign_item_inner(
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/unix/foreign_items.rs

@@ -639,6 +639,31 @@ fn emulate_foreign_item_inner(
                 this.gen_random(ptr, len)?;
                 this.write_scalar(Scalar::from_target_usize(len, this), dest)?;
             }
+            "_Unwind_RaiseException" => {
+                // This is not formally part of POSIX, but it is very wide-spread on POSIX systems.
+                // It was originally specified as part of the Itanium C++ ABI:
+                // https://itanium-cxx-abi.github.io/cxx-abi/abi-eh.html#base-throw.
+                // On Linux it is
+                // documented as part of the LSB:
+                // https://refspecs.linuxfoundation.org/LSB_5.0.0/LSB-Core-generic/LSB-Core-generic/baselib--unwind-raiseexception.html
+                // Basically every other UNIX uses the exact same api though. Arm also references
+                // back to the Itanium C++ ABI for the definition of `_Unwind_RaiseException` for
+                // arm64:
+                // https://github.com/ARM-software/abi-aa/blob/main/cppabi64/cppabi64.rst#toc-entry-35
+                // For arm32 they did something custom, but similar enough that the same
+                // `_Unwind_RaiseException` impl in miri should work:
+                // https://github.com/ARM-software/abi-aa/blob/main/ehabi32/ehabi32.rst
+                if !matches!(&*this.tcx.sess.target.os, "linux" | "freebsd" | "illumos" | "solaris" | "android" | "macos") {
+                    throw_unsup_format!(
+                        "`_Unwind_RaiseException` is not supported on {}",
+                        this.tcx.sess.target.os
+                    );
+                }
+                // This function looks and behaves excatly like miri_start_unwind.
+                let [payload] = this.check_shim(abi, Abi::C { unwind: true }, link_name, args)?;
+                this.handle_miri_start_unwind(payload)?;
+                return Ok(EmulateItemResult::NeedsUnwind);
+            }
 
             // Incomplete shims that we "stub out" just to get pre-main initialization code to work.
             // These shims are enabled only when the caller is in the standard library.
@@ -760,6 +785,6 @@ fn emulate_foreign_item_inner(
             }
         };
 
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/unix/freebsd/foreign_items.rs

@@ -86,6 +86,6 @@ fn emulate_foreign_item_inner(
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/unix/linux/foreign_items.rs

@@ -203,6 +203,6 @@ fn emulate_foreign_item_inner(
             _ => return Ok(EmulateItemResult::NotSupported),
         };
 
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/unix/macos/foreign_items.rs

@@ -177,6 +177,6 @@ fn emulate_foreign_item_inner(
             _ => return Ok(EmulateItemResult::NotSupported),
         };
 
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/unix/solarish/foreign_items.rs

@@ -45,6 +45,6 @@ fn emulate_foreign_item_inner(
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/wasi/foreign_items.rs

@@ -35,6 +35,6 @@ fn emulate_foreign_item_inner(
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/windows/foreign_items.rs

@@ -762,6 +762,6 @@ fn emulate_foreign_item_inner(
             _ => return Ok(EmulateItemResult::NotSupported),
         }
 
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/aesni.rs

@@ -127,7 +127,7 @@ fn emulate_x86_aesni_intrinsic(
             // with an external crate.
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }
 

src/tools/miri/src/shims/x86/avx.rs

@@ -344,6 +344,6 @@ fn emulate_x86_avx_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/avx2.rs

@@ -440,6 +440,6 @@ fn emulate_x86_avx2_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/mod.rs

@@ -144,7 +144,7 @@ fn emulate_x86_intrinsic(
 
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }
 

src/tools/miri/src/shims/x86/sse.rs

@@ -212,6 +212,6 @@ fn emulate_x86_sse_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/sse2.rs

@@ -388,6 +388,6 @@ fn emulate_x86_sse2_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/sse3.rs

@@ -51,6 +51,6 @@ fn emulate_x86_sse3_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/sse41.rs

@@ -176,6 +176,6 @@ fn emulate_x86_sse41_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/src/shims/x86/ssse3.rs

@@ -137,6 +137,6 @@ fn emulate_x86_ssse3_intrinsic(
             }
             _ => return Ok(EmulateItemResult::NotSupported),
         }
-        Ok(EmulateItemResult::NeedsJumping)
+        Ok(EmulateItemResult::NeedsReturn)
     }
 }

src/tools/miri/tests/pass/panic/unwind_dwarf.rs

@@ -0,0 +1,96 @@
+//@ignore-target-windows: Windows uses a different unwinding mechanism
+#![feature(core_intrinsics, panic_unwind, rustc_attrs)]
+#![allow(internal_features)]
+
+//! Unwinding using `_Unwind_RaiseException`
+
+extern crate unwind as uw;
+
+use std::any::Any;
+use std::ptr;
+
+#[repr(C)]
+struct Exception {
+    _uwe: uw::_Unwind_Exception,
+    cause: Box<dyn Any + Send>,
+}
+
+pub fn panic(data: Box<dyn Any + Send>) -> u32 {
+    extern "C" fn exception_cleanup(
+        _unwind_code: uw::_Unwind_Reason_Code,
+        _exception: *mut uw::_Unwind_Exception,
+    ) {
+        std::process::abort();
+    }
+
+    let exception = Box::new(Exception {
+        _uwe: uw::_Unwind_Exception {
+            exception_class: miri_exception_class(),
+            exception_cleanup: Some(exception_cleanup),
+            private: [core::ptr::null(); uw::unwinder_private_data_size],
+        },
+        cause: data,
+    });
+    let exception_param = Box::into_raw(exception) as *mut uw::_Unwind_Exception;
+    return unsafe { uw::_Unwind_RaiseException(exception_param) as u32 };
+}
+
+pub unsafe fn rust_panic_cleanup(ptr: *mut u8) -> Box<dyn Any + Send> {
+    let exception = ptr as *mut uw::_Unwind_Exception;
+    if (*exception).exception_class != miri_exception_class() {
+        std::process::abort();
+    }
+
+    let exception = exception.cast::<Exception>();
+
+    let exception = Box::from_raw(exception as *mut Exception);
+    exception.cause
+}
+
+fn miri_exception_class() -> uw::_Unwind_Exception_Class {
+    // M O Z \0  M I R I -- vendor, language
+    // (Miri's own exception class is just used for testing)
+    0x4d4f5a_00_4d495249
+}
+
+pub fn catch_unwind<R, F: FnOnce() -> R>(f: F) -> Result<R, Box<dyn Any + Send>> {
+    struct Data<F, R> {
+        f: Option<F>,
+        r: Option<R>,
+        p: Option<Box<dyn Any + Send>>,
+    }
+
+    let mut data = Data { f: Some(f), r: None, p: None };
+
+    let data_ptr = ptr::addr_of_mut!(data) as *mut u8;
+    unsafe {
+        return if std::intrinsics::catch_unwind(do_call::<F, R>, data_ptr, do_catch::<F, R>) == 0 {
+            Ok(data.r.take().unwrap())
+        } else {
+            Err(data.p.take().unwrap())
+        };
+    }
+
+    fn do_call<F: FnOnce() -> R, R>(data: *mut u8) {
+        unsafe {
+            let data = &mut *data.cast::<Data<F, R>>();
+            let f = data.f.take().unwrap();
+            data.r = Some(f());
+        }
+    }
+
+    #[rustc_nounwind]
+    fn do_catch<F: FnOnce() -> R, R>(data: *mut u8, payload: *mut u8) {
+        unsafe {
+            let obj = rust_panic_cleanup(payload);
+            (*data.cast::<Data<F, R>>()).p = Some(obj);
+        }
+    }
+}
+
+fn main() {
+    assert_eq!(
+        catch_unwind(|| panic(Box::new(42))).unwrap_err().downcast::<i32>().unwrap(),
+        Box::new(42)
+    );
+}