use rustc_hir::def::DefKind; use rustc_hir::LangItem; use rustc_middle::mir; use rustc_middle::mir::interpret::PointerArithmetic; use rustc_middle::ty::layout::{FnAbiOf, TyAndLayout}; use rustc_middle::ty::{self, TyCtxt}; use rustc_span::Span; use std::borrow::Borrow; use std::hash::Hash; use std::ops::ControlFlow; use rustc_data_structures::fx::FxIndexMap; use rustc_data_structures::fx::IndexEntry; use std::fmt; use rustc_ast::Mutability; use rustc_hir::def_id::DefId; use rustc_middle::mir::AssertMessage; use rustc_span::symbol::{sym, Symbol}; use rustc_target::abi::{Align, Size}; use rustc_target::spec::abi::Abi as CallAbi; use crate::errors::{LongRunning, LongRunningWarn}; use crate::fluent_generated as fluent; use crate::interpret::{ self, compile_time_machine, AllocId, ConstAllocation, FnArg, FnVal, Frame, ImmTy, InterpCx, InterpResult, OpTy, PlaceTy, Pointer, Scalar, }; use super::error::*; /// When hitting this many interpreted terminators we emit a deny by default lint /// that notfies the user that their constant takes a long time to evaluate. If that's /// what they intended, they can just allow the lint. const LINT_TERMINATOR_LIMIT: usize = 2_000_000; /// The limit used by `-Z tiny-const-eval-limit`. This smaller limit is useful for internal /// tests not needing to run 30s or more to show some behaviour. const TINY_LINT_TERMINATOR_LIMIT: usize = 20; /// After this many interpreted terminators, we start emitting progress indicators at every /// power of two of interpreted terminators. const PROGRESS_INDICATOR_START: usize = 4_000_000; /// Extra machine state for CTFE, and the Machine instance pub struct CompileTimeInterpreter<'mir, 'tcx> { /// The number of terminators that have been evaluated. /// /// This is used to produce lints informing the user that the compiler is not stuck. /// Set to `usize::MAX` to never report anything. pub(super) num_evaluated_steps: usize, /// The virtual call stack. pub(super) stack: Vec>, /// We need to make sure consts never point to anything mutable, even recursively. That is /// relied on for pattern matching on consts with references. /// To achieve this, two pieces have to work together: /// * Interning makes everything outside of statics immutable. /// * Pointers to allocations inside of statics can never leak outside, to a non-static global. /// This boolean here controls the second part. pub(super) can_access_statics: CanAccessStatics, /// Whether to check alignment during evaluation. pub(super) check_alignment: CheckAlignment, } #[derive(Copy, Clone)] pub enum CheckAlignment { /// Ignore all alignment requirements. /// This is mainly used in interning. No, /// Hard error when dereferencing a misaligned pointer. Error, } #[derive(Copy, Clone, PartialEq)] pub(crate) enum CanAccessStatics { No, Yes, } impl From for CanAccessStatics { fn from(value: bool) -> Self { if value { Self::Yes } else { Self::No } } } impl<'mir, 'tcx> CompileTimeInterpreter<'mir, 'tcx> { pub(crate) fn new( can_access_statics: CanAccessStatics, check_alignment: CheckAlignment, ) -> Self { CompileTimeInterpreter { num_evaluated_steps: 0, stack: Vec::new(), can_access_statics, check_alignment, } } } impl interpret::AllocMap for FxIndexMap { #[inline(always)] fn contains_key(&mut self, k: &Q) -> bool where K: Borrow, { FxIndexMap::contains_key(self, k) } #[inline(always)] fn insert(&mut self, k: K, v: V) -> Option { FxIndexMap::insert(self, k, v) } #[inline(always)] fn remove(&mut self, k: &Q) -> Option where K: Borrow, { FxIndexMap::remove(self, k) } #[inline(always)] fn filter_map_collect(&self, mut f: impl FnMut(&K, &V) -> Option) -> Vec { self.iter().filter_map(move |(k, v)| f(k, &*v)).collect() } #[inline(always)] fn get_or(&self, k: K, vacant: impl FnOnce() -> Result) -> Result<&V, E> { match self.get(&k) { Some(v) => Ok(v), None => { vacant()?; bug!("The CTFE machine shouldn't ever need to extend the alloc_map when reading") } } } #[inline(always)] fn get_mut_or(&mut self, k: K, vacant: impl FnOnce() -> Result) -> Result<&mut V, E> { match self.entry(k) { IndexEntry::Occupied(e) => Ok(e.into_mut()), IndexEntry::Vacant(e) => { let v = vacant()?; Ok(e.insert(v)) } } } } pub(crate) type CompileTimeEvalContext<'mir, 'tcx> = InterpCx<'mir, 'tcx, CompileTimeInterpreter<'mir, 'tcx>>; #[derive(Debug, PartialEq, Eq, Copy, Clone)] pub enum MemoryKind { Heap, } impl fmt::Display for MemoryKind { fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { match self { MemoryKind::Heap => write!(f, "heap allocation"), } } } impl interpret::MayLeak for MemoryKind { #[inline(always)] fn may_leak(self) -> bool { match self { MemoryKind::Heap => false, } } } impl interpret::MayLeak for ! { #[inline(always)] fn may_leak(self) -> bool { // `self` is uninhabited self } } impl<'mir, 'tcx: 'mir> CompileTimeEvalContext<'mir, 'tcx> { fn location_triple_for_span(&self, span: Span) -> (Symbol, u32, u32) { let topmost = span.ctxt().outer_expn().expansion_cause().unwrap_or(span); let caller = self.tcx.sess.source_map().lookup_char_pos(topmost.lo()); use rustc_session::{config::RemapPathScopeComponents, RemapFileNameExt}; ( Symbol::intern( &caller .file .name .for_scope(&self.tcx.sess, RemapPathScopeComponents::DIAGNOSTICS) .to_string_lossy(), ), u32::try_from(caller.line).unwrap(), u32::try_from(caller.col_display).unwrap().checked_add(1).unwrap(), ) } /// "Intercept" a function call, because we have something special to do for it. /// All `#[rustc_do_not_const_check]` functions should be hooked here. /// If this returns `Some` function, which may be `instance` or a different function with /// compatible arguments, then evaluation should continue with that function. /// If this returns `None`, the function call has been handled and the function has returned. fn hook_special_const_fn( &mut self, instance: ty::Instance<'tcx>, args: &[FnArg<'tcx>], dest: &PlaceTy<'tcx>, ret: Option, ) -> InterpResult<'tcx, Option>> { let def_id = instance.def_id(); if self.tcx.has_attr(def_id, sym::rustc_const_panic_str) || Some(def_id) == self.tcx.lang_items().begin_panic_fn() { let args = self.copy_fn_args(args)?; // &str or &&str assert!(args.len() == 1); let mut msg_place = self.deref_pointer(&args[0])?; while msg_place.layout.ty.is_ref() { msg_place = self.deref_pointer(&msg_place)?; } let msg = Symbol::intern(self.read_str(&msg_place)?); let span = self.find_closest_untracked_caller_location(); let (file, line, col) = self.location_triple_for_span(span); return Err(ConstEvalErrKind::Panic { msg, file, line, col }.into()); } else if Some(def_id) == self.tcx.lang_items().panic_fmt() { // For panic_fmt, call const_panic_fmt instead. let const_def_id = self.tcx.require_lang_item(LangItem::ConstPanicFmt, None); let new_instance = ty::Instance::resolve( *self.tcx, ty::ParamEnv::reveal_all(), const_def_id, instance.args, ) .unwrap() .unwrap(); return Ok(Some(new_instance)); } else if Some(def_id) == self.tcx.lang_items().align_offset_fn() { let args = self.copy_fn_args(args)?; // For align_offset, we replace the function call if the pointer has no address. match self.align_offset(instance, &args, dest, ret)? { ControlFlow::Continue(()) => return Ok(Some(instance)), ControlFlow::Break(()) => return Ok(None), } } Ok(Some(instance)) } /// `align_offset(ptr, target_align)` needs special handling in const eval, because the pointer /// may not have an address. /// /// If `ptr` does have a known address, then we return `Continue(())` and the function call should /// proceed as normal. /// /// If `ptr` doesn't have an address, but its underlying allocation's alignment is at most /// `target_align`, then we call the function again with an dummy address relative to the /// allocation. /// /// If `ptr` doesn't have an address and `target_align` is stricter than the underlying /// allocation's alignment, then we return `usize::MAX` immediately. fn align_offset( &mut self, instance: ty::Instance<'tcx>, args: &[OpTy<'tcx>], dest: &PlaceTy<'tcx>, ret: Option, ) -> InterpResult<'tcx, ControlFlow<()>> { assert_eq!(args.len(), 2); let ptr = self.read_pointer(&args[0])?; let target_align = self.read_scalar(&args[1])?.to_target_usize(self)?; if !target_align.is_power_of_two() { throw_ub_custom!( fluent::const_eval_align_offset_invalid_align, target_align = target_align, ); } match self.ptr_try_get_alloc_id(ptr) { Ok((alloc_id, offset, _extra)) => { let (_size, alloc_align, _kind) = self.get_alloc_info(alloc_id); if target_align <= alloc_align.bytes() { // Extract the address relative to the allocation base that is definitely // sufficiently aligned and call `align_offset` again. let addr = ImmTy::from_uint(offset.bytes(), args[0].layout).into(); let align = ImmTy::from_uint(target_align, args[1].layout).into(); let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty())?; // We replace the entire function call with a "tail call". // Note that this happens before the frame of the original function // is pushed on the stack. self.eval_fn_call( FnVal::Instance(instance), (CallAbi::Rust, fn_abi), &[FnArg::Copy(addr), FnArg::Copy(align)], /* with_caller_location = */ false, dest, ret, mir::UnwindAction::Unreachable, )?; Ok(ControlFlow::Break(())) } else { // Not alignable in const, return `usize::MAX`. let usize_max = Scalar::from_target_usize(self.target_usize_max(), self); self.write_scalar(usize_max, dest)?; self.return_to_block(ret)?; Ok(ControlFlow::Break(())) } } Err(_addr) => { // The pointer has an address, continue with function call. Ok(ControlFlow::Continue(())) } } } /// See documentation on the `ptr_guaranteed_cmp` intrinsic. fn guaranteed_cmp(&mut self, a: Scalar, b: Scalar) -> InterpResult<'tcx, u8> { Ok(match (a, b) { // Comparisons between integers are always known. (Scalar::Int { .. }, Scalar::Int { .. }) => { if a == b { 1 } else { 0 } } // Comparisons of abstract pointers with null pointers are known if the pointer // is in bounds, because if they are in bounds, the pointer can't be null. // Inequality with integers other than null can never be known for sure. (Scalar::Int(int), ptr @ Scalar::Ptr(..)) | (ptr @ Scalar::Ptr(..), Scalar::Int(int)) if int.is_null() && !self.scalar_may_be_null(ptr)? => { 0 } // Equality with integers can never be known for sure. (Scalar::Int { .. }, Scalar::Ptr(..)) | (Scalar::Ptr(..), Scalar::Int { .. }) => 2, // FIXME: return a `1` for when both sides are the same pointer, *except* that // some things (like functions and vtables) do not have stable addresses // so we need to be careful around them (see e.g. #73722). // FIXME: return `0` for at least some comparisons where we can reliably // determine the result of runtime inequality tests at compile-time. // Examples include comparison of addresses in different static items. (Scalar::Ptr(..), Scalar::Ptr(..)) => 2, }) } } impl<'mir, 'tcx> interpret::Machine<'mir, 'tcx> for CompileTimeInterpreter<'mir, 'tcx> { compile_time_machine!(<'mir, 'tcx>); type MemoryKind = MemoryKind; const PANIC_ON_ALLOC_FAIL: bool = false; // will be raised as a proper error #[inline(always)] fn enforce_alignment(ecx: &InterpCx<'mir, 'tcx, Self>) -> bool { matches!(ecx.machine.check_alignment, CheckAlignment::Error) } #[inline(always)] fn enforce_validity(ecx: &InterpCx<'mir, 'tcx, Self>, layout: TyAndLayout<'tcx>) -> bool { ecx.tcx.sess.opts.unstable_opts.extra_const_ub_checks || layout.abi.is_uninhabited() } fn load_mir( ecx: &InterpCx<'mir, 'tcx, Self>, instance: ty::InstanceDef<'tcx>, ) -> InterpResult<'tcx, &'tcx mir::Body<'tcx>> { match instance { ty::InstanceDef::Item(def) => { if ecx.tcx.is_ctfe_mir_available(def) { Ok(ecx.tcx.mir_for_ctfe(def)) } else if ecx.tcx.def_kind(def) == DefKind::AssocConst { let guar = ecx.tcx.sess.delay_span_bug( rustc_span::DUMMY_SP, "This is likely a const item that is missing from its impl", ); throw_inval!(AlreadyReported(guar.into())); } else { // `find_mir_or_eval_fn` checks that this is a const fn before even calling us, // so this should be unreachable. let path = ecx.tcx.def_path_str(def); bug!("trying to call extern function `{path}` at compile-time"); } } _ => Ok(ecx.tcx.instance_mir(instance)), } } fn find_mir_or_eval_fn( ecx: &mut InterpCx<'mir, 'tcx, Self>, orig_instance: ty::Instance<'tcx>, _abi: CallAbi, args: &[FnArg<'tcx>], dest: &PlaceTy<'tcx>, ret: Option, _unwind: mir::UnwindAction, // unwinding is not supported in consts ) -> InterpResult<'tcx, Option<(&'mir mir::Body<'tcx>, ty::Instance<'tcx>)>> { debug!("find_mir_or_eval_fn: {:?}", orig_instance); // Replace some functions. let Some(instance) = ecx.hook_special_const_fn(orig_instance, args, dest, ret)? else { // Call has already been handled. return Ok(None); }; // Only check non-glue functions if let ty::InstanceDef::Item(def) = instance.def { // Execution might have wandered off into other crates, so we cannot do a stability- // sensitive check here. But we can at least rule out functions that are not const at // all. That said, we have to allow calling functions inside a trait marked with // #[const_trait]. These *are* const-checked! // FIXME: why does `is_const_fn_raw` not classify them as const? if (!ecx.tcx.is_const_fn_raw(def) && !ecx.tcx.is_const_default_method(def)) || ecx.tcx.has_attr(def, sym::rustc_do_not_const_check) { // We certainly do *not* want to actually call the fn // though, so be sure we return here. throw_unsup_format!("calling non-const function `{}`", instance) } } // This is a const fn. Call it. // In case of replacement, we return the *original* instance to make backtraces work out // (and we hope this does not confuse the FnAbi checks too much). Ok(Some((ecx.load_mir(instance.def, None)?, orig_instance))) } fn panic_nounwind(ecx: &mut InterpCx<'mir, 'tcx, Self>, msg: &str) -> InterpResult<'tcx> { let msg = Symbol::intern(msg); let span = ecx.find_closest_untracked_caller_location(); let (file, line, col) = ecx.location_triple_for_span(span); Err(ConstEvalErrKind::Panic { msg, file, line, col }.into()) } fn call_intrinsic( ecx: &mut InterpCx<'mir, 'tcx, Self>, instance: ty::Instance<'tcx>, args: &[OpTy<'tcx>], dest: &PlaceTy<'tcx, Self::Provenance>, target: Option, _unwind: mir::UnwindAction, ) -> InterpResult<'tcx> { // Shared intrinsics. if ecx.emulate_intrinsic(instance, args, dest, target)? { return Ok(()); } let intrinsic_name = ecx.tcx.item_name(instance.def_id()); // CTFE-specific intrinsics. let Some(ret) = target else { throw_unsup_format!("intrinsic `{intrinsic_name}` is not supported at compile-time"); }; match intrinsic_name { sym::ptr_guaranteed_cmp => { let a = ecx.read_scalar(&args[0])?; let b = ecx.read_scalar(&args[1])?; let cmp = ecx.guaranteed_cmp(a, b)?; ecx.write_scalar(Scalar::from_u8(cmp), dest)?; } sym::const_allocate => { let size = ecx.read_scalar(&args[0])?.to_target_usize(ecx)?; let align = ecx.read_scalar(&args[1])?.to_target_usize(ecx)?; let align = match Align::from_bytes(align) { Ok(a) => a, Err(err) => throw_ub_custom!( fluent::const_eval_invalid_align_details, name = "const_allocate", err_kind = err.diag_ident(), align = err.align() ), }; let ptr = ecx.allocate_ptr( Size::from_bytes(size as u64), align, interpret::MemoryKind::Machine(MemoryKind::Heap), )?; ecx.write_pointer(ptr, dest)?; } sym::const_deallocate => { let ptr = ecx.read_pointer(&args[0])?; let size = ecx.read_scalar(&args[1])?.to_target_usize(ecx)?; let align = ecx.read_scalar(&args[2])?.to_target_usize(ecx)?; let size = Size::from_bytes(size); let align = match Align::from_bytes(align) { Ok(a) => a, Err(err) => throw_ub_custom!( fluent::const_eval_invalid_align_details, name = "const_deallocate", err_kind = err.diag_ident(), align = err.align() ), }; // If an allocation is created in an another const, // we don't deallocate it. let (alloc_id, _, _) = ecx.ptr_get_alloc_id(ptr)?; let is_allocated_in_another_const = matches!( ecx.tcx.try_get_global_alloc(alloc_id), Some(interpret::GlobalAlloc::Memory(_)) ); if !is_allocated_in_another_const { ecx.deallocate_ptr( ptr, Some((size, align)), interpret::MemoryKind::Machine(MemoryKind::Heap), )?; } } _ => { throw_unsup_format!( "intrinsic `{intrinsic_name}` is not supported at compile-time" ); } } ecx.go_to_block(ret); Ok(()) } fn assert_panic( ecx: &mut InterpCx<'mir, 'tcx, Self>, msg: &AssertMessage<'tcx>, _unwind: mir::UnwindAction, ) -> InterpResult<'tcx> { use rustc_middle::mir::AssertKind::*; // Convert `AssertKind` to `AssertKind`. let eval_to_int = |op| ecx.read_immediate(&ecx.eval_operand(op, None)?).map(|x| x.to_const_int()); let err = match msg { BoundsCheck { len, index } => { let len = eval_to_int(len)?; let index = eval_to_int(index)?; BoundsCheck { len, index } } Overflow(op, l, r) => Overflow(*op, eval_to_int(l)?, eval_to_int(r)?), OverflowNeg(op) => OverflowNeg(eval_to_int(op)?), DivisionByZero(op) => DivisionByZero(eval_to_int(op)?), RemainderByZero(op) => RemainderByZero(eval_to_int(op)?), ResumedAfterReturn(coroutine_kind) => ResumedAfterReturn(*coroutine_kind), ResumedAfterPanic(coroutine_kind) => ResumedAfterPanic(*coroutine_kind), MisalignedPointerDereference { ref required, ref found } => { MisalignedPointerDereference { required: eval_to_int(required)?, found: eval_to_int(found)?, } } }; Err(ConstEvalErrKind::AssertFailure(err).into()) } fn binary_ptr_op( _ecx: &InterpCx<'mir, 'tcx, Self>, _bin_op: mir::BinOp, _left: &ImmTy<'tcx>, _right: &ImmTy<'tcx>, ) -> InterpResult<'tcx, (ImmTy<'tcx>, bool)> { throw_unsup_format!("pointer arithmetic or comparison is not supported at compile-time"); } fn increment_const_eval_counter(ecx: &mut InterpCx<'mir, 'tcx, Self>) -> InterpResult<'tcx> { // The step limit has already been hit in a previous call to `increment_const_eval_counter`. if let Some(new_steps) = ecx.machine.num_evaluated_steps.checked_add(1) { let (limit, start) = if ecx.tcx.sess.opts.unstable_opts.tiny_const_eval_limit { (TINY_LINT_TERMINATOR_LIMIT, TINY_LINT_TERMINATOR_LIMIT) } else { (LINT_TERMINATOR_LIMIT, PROGRESS_INDICATOR_START) }; ecx.machine.num_evaluated_steps = new_steps; // By default, we have a *deny* lint kicking in after some time // to ensure `loop {}` doesn't just go forever. // In case that lint got reduced, in particular for `--cap-lint` situations, we also // have a hard warning shown every now and then for really long executions. if new_steps == limit { // By default, we stop after a million steps, but the user can disable this lint // to be able to run until the heat death of the universe or power loss, whichever // comes first. let hir_id = ecx.best_lint_scope(); let is_error = ecx .tcx .lint_level_at_node( rustc_session::lint::builtin::LONG_RUNNING_CONST_EVAL, hir_id, ) .0 .is_error(); let span = ecx.cur_span(); ecx.tcx.emit_spanned_lint( rustc_session::lint::builtin::LONG_RUNNING_CONST_EVAL, hir_id, span, LongRunning { item_span: ecx.tcx.span }, ); // If this was a hard error, don't bother continuing evaluation. if is_error { let guard = ecx .tcx .sess .delay_span_bug(span, "The deny lint should have already errored"); throw_inval!(AlreadyReported(guard.into())); } } else if new_steps > start && new_steps.is_power_of_two() { // Only report after a certain number of terminators have been evaluated and the // current number of evaluated terminators is a power of 2. The latter gives us a cheap // way to implement exponential backoff. let span = ecx.cur_span(); ecx.tcx.sess.emit_warning(LongRunningWarn { span, item_span: ecx.tcx.span }); } } Ok(()) } #[inline(always)] fn expose_ptr( _ecx: &mut InterpCx<'mir, 'tcx, Self>, _ptr: Pointer, ) -> InterpResult<'tcx> { // This is only reachable with -Zunleash-the-miri-inside-of-you. throw_unsup_format!("exposing pointers is not possible at compile-time") } #[inline(always)] fn init_frame_extra( ecx: &mut InterpCx<'mir, 'tcx, Self>, frame: Frame<'mir, 'tcx>, ) -> InterpResult<'tcx, Frame<'mir, 'tcx>> { // Enforce stack size limit. Add 1 because this is run before the new frame is pushed. if !ecx.recursion_limit.value_within_limit(ecx.stack().len() + 1) { throw_exhaust!(StackFrameLimitReached) } else { Ok(frame) } } #[inline(always)] fn stack<'a>( ecx: &'a InterpCx<'mir, 'tcx, Self>, ) -> &'a [Frame<'mir, 'tcx, Self::Provenance, Self::FrameExtra>] { &ecx.machine.stack } #[inline(always)] fn stack_mut<'a>( ecx: &'a mut InterpCx<'mir, 'tcx, Self>, ) -> &'a mut Vec> { &mut ecx.machine.stack } fn before_access_global( _tcx: TyCtxt<'tcx>, machine: &Self, alloc_id: AllocId, alloc: ConstAllocation<'tcx>, static_def_id: Option, is_write: bool, ) -> InterpResult<'tcx> { let alloc = alloc.inner(); if is_write { // Write access. These are never allowed, but we give a targeted error message. match alloc.mutability { Mutability::Not => Err(err_ub!(WriteToReadOnly(alloc_id)).into()), Mutability::Mut => Err(ConstEvalErrKind::ModifiedGlobal.into()), } } else { // Read access. These are usually allowed, with some exceptions. if machine.can_access_statics == CanAccessStatics::Yes { // Machine configuration allows us read from anything (e.g., `static` initializer). Ok(()) } else if static_def_id.is_some() { // Machine configuration does not allow us to read statics // (e.g., `const` initializer). // See const_eval::machine::MemoryExtra::can_access_statics for why // this check is so important: if we could read statics, we could read pointers // to mutable allocations *inside* statics. These allocations are not themselves // statics, so pointers to them can get around the check in `validity.rs`. Err(ConstEvalErrKind::ConstAccessesStatic.into()) } else { // Immutable global, this read is fine. // But make sure we never accept a read from something mutable, that would be // unsound. The reason is that as the content of this allocation may be different // now and at run-time, so if we permit reading now we might return the wrong value. assert_eq!(alloc.mutability, Mutability::Not); Ok(()) } } } } // Please do not add any code below the above `Machine` trait impl. I (oli-obk) plan more cleanups // so we can end up having a file with just that impl, but for now, let's keep the impl discoverable // at the bottom of this file.