https://github.com/rust-lang/rust/commit/263edd43c5255084292329423c61a9d69715ebfa.patch https://github.com/rust-lang/rust/issues/102155 Issue seen on native builds on riscv64 across multiple Linux Distributions. An alternative workaround appears to be building stage 1 with debug enabled. From 27412d1e3e128349bc515c16ce882860e20f037d Mon Sep 17 00:00:00 2001 From: 5225225 <5225225@mailbox.org> Date: Thu, 14 Jul 2022 22:42:47 +0100 Subject: [PATCH] Use constant eval to do strict validity checks --- Cargo.lock | 1 + .../src/intrinsics/mod.rs | 15 +---- compiler/rustc_codegen_ssa/Cargo.toml | 1 + compiler/rustc_codegen_ssa/src/mir/block.rs | 9 ++- .../src/const_eval/machine.rs | 2 +- .../src/interpret/intrinsics.rs | 56 ++++++++-------- compiler/rustc_const_eval/src/lib.rs | 6 ++ .../src/might_permit_raw_init.rs | 40 +++++++++++ compiler/rustc_middle/src/query/mod.rs | 8 +++ compiler/rustc_middle/src/ty/query.rs | 1 + compiler/rustc_query_impl/src/keys.rs | 12 +++- compiler/rustc_target/src/abi/mod.rs | 38 +++++------ .../intrinsics/panic-uninitialized-zeroed.rs | 66 ++++++++++++------- 13 files changed, 161 insertions(+), 94 deletions(-) create mode 100644 compiler/rustc_const_eval/src/might_permit_raw_init.rs diff --git a/Cargo.lock b/Cargo.lock index 147d47044078a..dd6f0345affd0 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -3664,6 +3664,7 @@ dependencies = [ "rustc_arena", "rustc_ast", "rustc_attr", + "rustc_const_eval", "rustc_data_structures", "rustc_errors", "rustc_fs_util", diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs index eafae1cdc8af0..4b2207f375879 100644 --- a/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs +++ b/compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs @@ -58,7 +58,6 @@ pub(crate) use llvm::codegen_llvm_intrinsic_call; use rustc_middle::ty::print::with_no_trimmed_paths; use rustc_middle::ty::subst::SubstsRef; use rustc_span::symbol::{kw, sym, Symbol}; -use rustc_target::abi::InitKind; use crate::prelude::*; use cranelift_codegen::ir::AtomicRmwOp; @@ -672,12 +671,7 @@ fn codegen_regular_intrinsic_call<'tcx>( return; } - if intrinsic == sym::assert_zero_valid - && !layout.might_permit_raw_init( - fx, - InitKind::Zero, - fx.tcx.sess.opts.unstable_opts.strict_init_checks) { - + if intrinsic == sym::assert_zero_valid && !fx.tcx.permits_zero_init(layout) { with_no_trimmed_paths!({ crate::base::codegen_panic( fx, @@ -688,12 +682,7 @@ fn codegen_regular_intrinsic_call<'tcx>( return; } - if intrinsic == sym::assert_uninit_valid - && !layout.might_permit_raw_init( - fx, - InitKind::Uninit, - fx.tcx.sess.opts.unstable_opts.strict_init_checks) { - + if intrinsic == sym::assert_uninit_valid && !fx.tcx.permits_uninit_init(layout) { with_no_trimmed_paths!({ crate::base::codegen_panic( fx, diff --git a/compiler/rustc_codegen_ssa/Cargo.toml b/compiler/rustc_codegen_ssa/Cargo.toml index faabea92f5a6c..81c8b9ceb136e 100644 --- a/compiler/rustc_codegen_ssa/Cargo.toml +++ b/compiler/rustc_codegen_ssa/Cargo.toml @@ -40,6 +40,7 @@ rustc_metadata = { path = "../rustc_metadata" } rustc_query_system = { path = "../rustc_query_system" } rustc_target = { path = "../rustc_target" } rustc_session = { path = "../rustc_session" } +rustc_const_eval = { path = "../rustc_const_eval" } [dependencies.object] version = "0.29.0" diff --git a/compiler/rustc_codegen_ssa/src/mir/block.rs b/compiler/rustc_codegen_ssa/src/mir/block.rs index 745da821c9d76..773c55cf551d5 100644 --- a/compiler/rustc_codegen_ssa/src/mir/block.rs +++ b/compiler/rustc_codegen_ssa/src/mir/block.rs @@ -22,7 +22,7 @@ use rustc_span::source_map::Span; use rustc_span::{sym, Symbol}; use rustc_symbol_mangling::typeid_for_fnabi; use rustc_target::abi::call::{ArgAbi, FnAbi, PassMode}; -use rustc_target::abi::{self, HasDataLayout, InitKind, WrappingRange}; +use rustc_target::abi::{self, HasDataLayout, WrappingRange}; use rustc_target::spec::abi::Abi; /// Used by `FunctionCx::codegen_terminator` for emitting common patterns @@ -528,7 +528,6 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { source_info: mir::SourceInfo, target: Option<mir::BasicBlock>, cleanup: Option<mir::BasicBlock>, - strict_validity: bool, ) -> bool { // Emit a panic or a no-op for `assert_*` intrinsics. // These are intrinsics that compile to panics so that we can get a message @@ -547,12 +546,13 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { }); if let Some(intrinsic) = panic_intrinsic { use AssertIntrinsic::*; + let ty = instance.unwrap().substs.type_at(0); let layout = bx.layout_of(ty); let do_panic = match intrinsic { Inhabited => layout.abi.is_uninhabited(), - ZeroValid => !layout.might_permit_raw_init(bx, InitKind::Zero, strict_validity), - UninitValid => !layout.might_permit_raw_init(bx, InitKind::Uninit, strict_validity), + ZeroValid => !bx.tcx().permits_zero_init(layout), + UninitValid => !bx.tcx().permits_uninit_init(layout), }; if do_panic { let msg_str = with_no_visible_paths!({ @@ -687,7 +687,6 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> { source_info, target, cleanup, - self.cx.tcx().sess.opts.unstable_opts.strict_init_checks, ) { return; } diff --git a/compiler/rustc_const_eval/src/const_eval/machine.rs b/compiler/rustc_const_eval/src/const_eval/machine.rs index 29ab1d187719c..e00e667fb71e2 100644 --- a/compiler/rustc_const_eval/src/const_eval/machine.rs +++ b/compiler/rustc_const_eval/src/const_eval/machine.rs @@ -104,7 +104,7 @@ pub struct CompileTimeInterpreter<'mir, 'tcx> { } impl<'mir, 'tcx> CompileTimeInterpreter<'mir, 'tcx> { - pub(super) fn new(const_eval_limit: Limit, can_access_statics: bool) -> Self { + pub(crate) fn new(const_eval_limit: Limit, can_access_statics: bool) -> Self { CompileTimeInterpreter { steps_remaining: const_eval_limit.0, stack: Vec::new(), diff --git a/compiler/rustc_const_eval/src/interpret/intrinsics.rs b/compiler/rustc_const_eval/src/interpret/intrinsics.rs index e2a8a9891f72f..7827fb8395b7f 100644 --- a/compiler/rustc_const_eval/src/interpret/intrinsics.rs +++ b/compiler/rustc_const_eval/src/interpret/intrinsics.rs @@ -15,7 +15,7 @@ use rustc_middle::ty::layout::LayoutOf as _; use rustc_middle::ty::subst::SubstsRef; use rustc_middle::ty::{Ty, TyCtxt}; use rustc_span::symbol::{sym, Symbol}; -use rustc_target::abi::{Abi, Align, InitKind, Primitive, Size}; +use rustc_target::abi::{Abi, Align, Primitive, Size}; use super::{ util::ensure_monomorphic_enough, CheckInAllocMsg, ImmTy, InterpCx, Machine, OpTy, PlaceTy, @@ -413,35 +413,33 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> { ), )?; } - if intrinsic_name == sym::assert_zero_valid - && !layout.might_permit_raw_init( - self, - InitKind::Zero, - self.tcx.sess.opts.unstable_opts.strict_init_checks, - ) - { - M::abort( - self, - format!( - "aborted execution: attempted to zero-initialize type `{}`, which is invalid", - ty - ), - )?; + + if intrinsic_name == sym::assert_zero_valid { + let should_panic = !self.tcx.permits_zero_init(layout); + + if should_panic { + M::abort( + self, + format!( + "aborted execution: attempted to zero-initialize type `{}`, which is invalid", + ty + ), + )?; + } } - if intrinsic_name == sym::assert_uninit_valid - && !layout.might_permit_raw_init( - self, - InitKind::Uninit, - self.tcx.sess.opts.unstable_opts.strict_init_checks, - ) - { - M::abort( - self, - format!( - "aborted execution: attempted to leave type `{}` uninitialized, which is invalid", - ty - ), - )?; + + if intrinsic_name == sym::assert_uninit_valid { + let should_panic = !self.tcx.permits_uninit_init(layout); + + if should_panic { + M::abort( + self, + format!( + "aborted execution: attempted to leave type `{}` uninitialized, which is invalid", + ty + ), + )?; + } } } sym::simd_insert => { diff --git a/compiler/rustc_const_eval/src/lib.rs b/compiler/rustc_const_eval/src/lib.rs index d65d4f7eb720e..72ac6af685dc4 100644 --- a/compiler/rustc_const_eval/src/lib.rs +++ b/compiler/rustc_const_eval/src/lib.rs @@ -33,11 +33,13 @@ extern crate rustc_middle; pub mod const_eval; mod errors; pub mod interpret; +mod might_permit_raw_init; pub mod transform; pub mod util; use rustc_middle::ty; use rustc_middle::ty::query::Providers; +use rustc_target::abi::InitKind; pub fn provide(providers: &mut Providers) { const_eval::provide(providers); @@ -59,4 +61,8 @@ pub fn provide(providers: &mut Providers) { let (param_env, value) = param_env_and_value.into_parts(); const_eval::deref_mir_constant(tcx, param_env, value) }; + providers.permits_uninit_init = + |tcx, ty| might_permit_raw_init::might_permit_raw_init(tcx, ty, InitKind::Uninit); + providers.permits_zero_init = + |tcx, ty| might_permit_raw_init::might_permit_raw_init(tcx, ty, InitKind::Zero); } diff --git a/compiler/rustc_const_eval/src/might_permit_raw_init.rs b/compiler/rustc_const_eval/src/might_permit_raw_init.rs new file mode 100644 index 0000000000000..f971c2238c7bb --- /dev/null +++ b/compiler/rustc_const_eval/src/might_permit_raw_init.rs @@ -0,0 +1,40 @@ +use crate::const_eval::CompileTimeInterpreter; +use crate::interpret::{InterpCx, MemoryKind, OpTy}; +use rustc_middle::ty::layout::LayoutCx; +use rustc_middle::ty::{layout::TyAndLayout, ParamEnv, TyCtxt}; +use rustc_session::Limit; +use rustc_target::abi::InitKind; + +pub fn might_permit_raw_init<'tcx>( + tcx: TyCtxt<'tcx>, + ty: TyAndLayout<'tcx>, + kind: InitKind, +) -> bool { + let strict = tcx.sess.opts.unstable_opts.strict_init_checks; + + if strict { + let machine = CompileTimeInterpreter::new(Limit::new(0), false); + + let mut cx = InterpCx::new(tcx, rustc_span::DUMMY_SP, ParamEnv::reveal_all(), machine); + + let allocated = cx + .allocate(ty, MemoryKind::Machine(crate::const_eval::MemoryKind::Heap)) + .expect("OOM: failed to allocate for uninit check"); + + if kind == InitKind::Zero { + cx.write_bytes_ptr( + allocated.ptr, + std::iter::repeat(0_u8).take(ty.layout.size().bytes_usize()), + ) + .expect("failed to write bytes for zero valid check"); + } + + let ot: OpTy<'_, _> = allocated.into(); + + // Assume that if it failed, it's a validation failure. + cx.validate_operand(&ot).is_ok() + } else { + let layout_cx = LayoutCx { tcx, param_env: ParamEnv::reveal_all() }; + ty.might_permit_raw_init(&layout_cx, kind) + } +} diff --git a/compiler/rustc_middle/src/query/mod.rs b/compiler/rustc_middle/src/query/mod.rs index bdae7e5fcd6b1..0581ef41f66c2 100644 --- a/compiler/rustc_middle/src/query/mod.rs +++ b/compiler/rustc_middle/src/query/mod.rs @@ -2053,4 +2053,12 @@ rustc_queries! { desc { |tcx| "looking up generator diagnostic data of `{}`", tcx.def_path_str(key) } separate_provide_extern } + + query permits_uninit_init(key: TyAndLayout<'tcx>) -> bool { + desc { "checking to see if {:?} permits being left uninit", key.ty } + } + + query permits_zero_init(key: TyAndLayout<'tcx>) -> bool { + desc { "checking to see if {:?} permits being left zeroed", key.ty } + } } diff --git a/compiler/rustc_middle/src/ty/query.rs b/compiler/rustc_middle/src/ty/query.rs index 3d662ed5de4ba..2452bcf6a61b8 100644 --- a/compiler/rustc_middle/src/ty/query.rs +++ b/compiler/rustc_middle/src/ty/query.rs @@ -28,6 +28,7 @@ use crate::traits::query::{ use crate::traits::specialization_graph; use crate::traits::{self, ImplSource}; use crate::ty::fast_reject::SimplifiedType; +use crate::ty::layout::TyAndLayout; use crate::ty::subst::{GenericArg, SubstsRef}; use crate::ty::util::AlwaysRequiresDrop; use crate::ty::GeneratorDiagnosticData; diff --git a/compiler/rustc_query_impl/src/keys.rs b/compiler/rustc_query_impl/src/keys.rs index 6fbafeb1d32b3..5477431431374 100644 --- a/compiler/rustc_query_impl/src/keys.rs +++ b/compiler/rustc_query_impl/src/keys.rs @@ -6,7 +6,7 @@ use rustc_middle::mir; use rustc_middle::traits; use rustc_middle::ty::fast_reject::SimplifiedType; use rustc_middle::ty::subst::{GenericArg, SubstsRef}; -use rustc_middle::ty::{self, Ty, TyCtxt}; +use rustc_middle::ty::{self, layout::TyAndLayout, Ty, TyCtxt}; use rustc_span::symbol::{Ident, Symbol}; use rustc_span::{Span, DUMMY_SP}; @@ -385,6 +385,16 @@ impl<'tcx> Key for Ty<'tcx> { } } +impl<'tcx> Key for TyAndLayout<'tcx> { + #[inline(always)] + fn query_crate_is_local(&self) -> bool { + true + } + fn default_span(&self, _: TyCtxt<'_>) -> Span { + DUMMY_SP + } +} + impl<'tcx> Key for (Ty<'tcx>, Ty<'tcx>) { #[inline(always)] fn query_crate_is_local(&self) -> bool { diff --git a/compiler/rustc_target/src/abi/mod.rs b/compiler/rustc_target/src/abi/mod.rs index d1eafd6ac5fb8..6f4d073d70486 100644 --- a/compiler/rustc_target/src/abi/mod.rs +++ b/compiler/rustc_target/src/abi/mod.rs @@ -1372,7 +1372,7 @@ pub struct PointeeInfo { /// Used in `might_permit_raw_init` to indicate the kind of initialisation /// that is checked to be valid -#[derive(Copy, Clone, Debug)] +#[derive(Copy, Clone, Debug, PartialEq, Eq)] pub enum InitKind { Zero, Uninit, @@ -1487,14 +1487,18 @@ impl<'a, Ty> TyAndLayout<'a, Ty> { /// /// `init_kind` indicates if the memory is zero-initialized or left uninitialized. /// - /// `strict` is an opt-in debugging flag added in #97323 that enables more checks. + /// This code is intentionally conservative, and will not detect + /// * zero init of an enum whose 0 variant does not allow zero initialization + /// * making uninitialized types who have a full valid range (ints, floats, raw pointers) + /// * Any form of invalid value being made inside an array (unless the value is uninhabited) /// - /// This is conservative: in doubt, it will answer `true`. + /// A strict form of these checks that uses const evaluation exists in + /// `rustc_const_eval::might_permit_raw_init`, and a tracking issue for making these checks + /// stricter is <https://github.com/rust-lang/rust/issues/66151>. /// - /// FIXME: Once we removed all the conservatism, we could alternatively - /// create an all-0/all-undef constant and run the const value validator to see if - /// this is a valid value for the given type. - pub fn might_permit_raw_init<C>(self, cx: &C, init_kind: InitKind, strict: bool) -> bool + /// FIXME: Once all the conservatism is removed from here, and the checks are ran by default, + /// we can use the const evaluation checks always instead. + pub fn might_permit_raw_init<C>(self, cx: &C, init_kind: InitKind) -> bool where Self: Copy, Ty: TyAbiInterface<'a, C>, @@ -1507,13 +1511,8 @@ impl<'a, Ty> TyAndLayout<'a, Ty> { s.valid_range(cx).contains(0) } InitKind::Uninit => { - if strict { - // The type must be allowed to be uninit (which means "is a union"). - s.is_uninit_valid() - } else { - // The range must include all values. - s.is_always_valid(cx) - } + // The range must include all values. + s.is_always_valid(cx) } } }; @@ -1534,19 +1533,12 @@ impl<'a, Ty> TyAndLayout<'a, Ty> { // If we have not found an error yet, we need to recursively descend into fields. match &self.fields { FieldsShape::Primitive | FieldsShape::Union { .. } => {} - FieldsShape::Array { count, .. } => { + FieldsShape::Array { .. } => { // FIXME(#66151): For now, we are conservative and do not check arrays by default. - if strict - && *count > 0 - && !self.field(cx, 0).might_permit_raw_init(cx, init_kind, strict) - { - // Found non empty array with a type that is unhappy about this kind of initialization - return false; - } } FieldsShape::Arbitrary { offsets, .. } => { for idx in 0..offsets.len() { - if !self.field(cx, idx).might_permit_raw_init(cx, init_kind, strict) { + if !self.field(cx, idx).might_permit_raw_init(cx, init_kind) { // We found a field that is unhappy with this kind of initialization. return false; } diff --git a/src/test/ui/intrinsics/panic-uninitialized-zeroed.rs b/src/test/ui/intrinsics/panic-uninitialized-zeroed.rs index 3ffd35ecdb8da..255151a96032c 100644 --- a/src/test/ui/intrinsics/panic-uninitialized-zeroed.rs +++ b/src/test/ui/intrinsics/panic-uninitialized-zeroed.rs @@ -57,6 +57,13 @@ enum LR_NonZero { struct ZeroSized; +#[allow(dead_code)] +#[repr(i32)] +enum ZeroIsValid { + Zero(u8) = 0, + One(NonNull<()>) = 1, +} + fn test_panic_msg<T>(op: impl (FnOnce() -> T) + panic::UnwindSafe, msg: &str) { let err = panic::catch_unwind(op).err(); assert_eq!( @@ -152,33 +159,12 @@ fn main() { "attempted to zero-initialize type `*const dyn core::marker::Send`, which is invalid" ); - /* FIXME(#66151) we conservatively do not error here yet. - test_panic_msg( - || mem::uninitialized::<LR_NonZero>(), - "attempted to leave type `LR_NonZero` uninitialized, which is invalid" - ); - test_panic_msg( - || mem::zeroed::<LR_NonZero>(), - "attempted to zero-initialize type `LR_NonZero`, which is invalid" - ); - - test_panic_msg( - || mem::uninitialized::<ManuallyDrop<LR_NonZero>>(), - "attempted to leave type `std::mem::ManuallyDrop<LR_NonZero>` uninitialized, \ - which is invalid" - ); - test_panic_msg( - || mem::zeroed::<ManuallyDrop<LR_NonZero>>(), - "attempted to zero-initialize type `std::mem::ManuallyDrop<LR_NonZero>`, \ - which is invalid" - ); - */ - test_panic_msg( || mem::uninitialized::<(NonNull<u32>, u32, u32)>(), "attempted to leave type `(core::ptr::non_null::NonNull<u32>, u32, u32)` uninitialized, \ which is invalid" ); + test_panic_msg( || mem::zeroed::<(NonNull<u32>, u32, u32)>(), "attempted to zero-initialize type `(core::ptr::non_null::NonNull<u32>, u32, u32)`, \ @@ -196,11 +182,23 @@ fn main() { which is invalid" ); + test_panic_msg( + || mem::uninitialized::<LR_NonZero>(), + "attempted to leave type `LR_NonZero` uninitialized, which is invalid" + ); + + test_panic_msg( + || mem::uninitialized::<ManuallyDrop<LR_NonZero>>(), + "attempted to leave type `core::mem::manually_drop::ManuallyDrop<LR_NonZero>` uninitialized, \ + which is invalid" + ); + test_panic_msg( || mem::uninitialized::<NoNullVariant>(), "attempted to leave type `NoNullVariant` uninitialized, \ which is invalid" ); + test_panic_msg( || mem::zeroed::<NoNullVariant>(), "attempted to zero-initialize type `NoNullVariant`, \ @@ -212,10 +210,12 @@ fn main() { || mem::uninitialized::<bool>(), "attempted to leave type `bool` uninitialized, which is invalid" ); + test_panic_msg( || mem::uninitialized::<LR>(), "attempted to leave type `LR` uninitialized, which is invalid" ); + test_panic_msg( || mem::uninitialized::<ManuallyDrop<LR>>(), "attempted to leave type `core::mem::manually_drop::ManuallyDrop<LR>` uninitialized, which is invalid" @@ -229,6 +229,7 @@ fn main() { let _val = mem::zeroed::<Option<&'static i32>>(); let _val = mem::zeroed::<MaybeUninit<NonNull<u32>>>(); let _val = mem::zeroed::<[!; 0]>(); + let _val = mem::zeroed::<ZeroIsValid>(); let _val = mem::uninitialized::<MaybeUninit<bool>>(); let _val = mem::uninitialized::<[!; 0]>(); let _val = mem::uninitialized::<()>(); @@ -259,12 +260,33 @@ fn main() { || mem::zeroed::<[NonNull<()>; 1]>(), "attempted to zero-initialize type `[core::ptr::non_null::NonNull<()>; 1]`, which is invalid" ); + + // FIXME(#66151) we conservatively do not error here yet (by default). + test_panic_msg( + || mem::zeroed::<LR_NonZero>(), + "attempted to zero-initialize type `LR_NonZero`, which is invalid" + ); + + test_panic_msg( + || mem::zeroed::<ManuallyDrop<LR_NonZero>>(), + "attempted to zero-initialize type `core::mem::manually_drop::ManuallyDrop<LR_NonZero>`, \ + which is invalid" + ); } else { // These are UB because they have not been officially blessed, but we await the resolution // of <https://github.com/rust-lang/unsafe-code-guidelines/issues/71> before doing // anything about that. let _val = mem::uninitialized::<i32>(); let _val = mem::uninitialized::<*const ()>(); + + // These are UB, but best to test them to ensure we don't become unintentionally + // stricter. + + // It's currently unchecked to create invalid enums and values inside arrays. + let _val = mem::zeroed::<LR_NonZero>(); + let _val = mem::zeroed::<[LR_NonZero; 1]>(); + let _val = mem::zeroed::<[NonNull<()>; 1]>(); + let _val = mem::uninitialized::<[NonNull<()>; 1]>(); } } }