Auto merge of #142681 - 1c3t3a:sanitize-off-on, r=<try>

Remove the `#[no_sanitize]` attribute in favor of `#[sanitize(xyz = "on|off")]`

This came up during the sanitizer stabilization (#123617). Instead of a `#[no_sanitize(xyz)]` attribute, we would like to have a `#[sanitize(xyz = "on|off")]` attribute, which is more powerful and allows to be extended in the future (instead
of just focusing on turning sanitizers off). The implementation is done according to what was [discussed on Zulip](https://rust-lang.zulipchat.com/#narrow/channel/343119-project-exploit-mitigations/topic/Stabilize.20the.20.60no_sanitize.60.20attribute/with/495377292)).

The new attribute also works on modules, traits and impl items and thus enables usage as the following:
```rust
#[sanitize(address = "off")]
mod foo {
    fn unsanitized(..) {}

    #[sanitize(address = "on")]
    fn sanitized(..) {}
}

trait MyTrait {
  #[sanitize(address = "off")]
  fn unsanitized_default(..) {}
}

#[sanitize(thread = "off")]
impl MyTrait for () {
    ...
}
```

r? `@rcvalle`

Author: rust-bors[bot]

compiler/rustc_codegen_ssa/messages.ftl

@@ -165,8 +165,8 @@ codegen_ssa_invalid_monomorphization_unsupported_symbol = invalid monomorphizati
 
 codegen_ssa_invalid_monomorphization_unsupported_symbol_of_size = invalid monomorphization of `{$name}` intrinsic: unsupported {$symbol} from `{$in_ty}` with element `{$in_elem}` of size `{$size}` to `{$ret_ty}`
 
-codegen_ssa_invalid_no_sanitize = invalid argument for `no_sanitize`
-    .note = expected one of: `address`, `cfi`, `hwaddress`, `kcfi`, `memory`, `memtag`, `shadow-call-stack`, or `thread`
+codegen_ssa_invalid_sanitize = invalid argument for `sanitize`
+    .note = expected one of: `address`, `kernel_address`, `cfi`, `hwaddress`, `kcfi`, `memory`, `memtag`, `shadow_call_stack`, or `thread`
 
 codegen_ssa_invalid_windows_subsystem = invalid windows subsystem `{$subsystem}`, only `windows` and `console` are allowed
 

compiler/rustc_codegen_ssa/src/codegen_attrs.rs

@@ -9,7 +9,7 @@ use rustc_attr_data_structures::{
 use rustc_hir::def::DefKind;
 use rustc_hir::def_id::{DefId, LOCAL_CRATE, LocalDefId};
 use rustc_hir::weak_lang_items::WEAK_LANG_ITEMS;
-use rustc_hir::{self as hir, LangItem, lang_items};
+use rustc_hir::{self as hir, Attribute, LangItem, lang_items};
 use rustc_middle::middle::codegen_fn_attrs::{
     CodegenFnAttrFlags, CodegenFnAttrs, PatchableFunctionEntry,
 };
@@ -86,7 +86,7 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: LocalDefId) -> CodegenFnAttrs {
     let rust_target_features = tcx.rust_target_features(LOCAL_CRATE);
 
     let mut link_ordinal_span = None;
-    let mut no_sanitize_span = None;
+    let mut sanitize_span = None;
 
     for attr in attrs.iter() {
         // In some cases, attribute are only valid on functions, but it's the `check_attr`
@@ -254,39 +254,7 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: LocalDefId) -> CodegenFnAttrs {
                     }
                 }
             }
-            sym::no_sanitize => {
-                no_sanitize_span = Some(attr.span());
-                if let Some(list) = attr.meta_item_list() {
-                    for item in list.iter() {
-                        match item.name() {
-                            Some(sym::address) => {
-                                codegen_fn_attrs.no_sanitize |=
-                                    SanitizerSet::ADDRESS | SanitizerSet::KERNELADDRESS
-                            }
-                            Some(sym::cfi) => codegen_fn_attrs.no_sanitize |= SanitizerSet::CFI,
-                            Some(sym::kcfi) => codegen_fn_attrs.no_sanitize |= SanitizerSet::KCFI,
-                            Some(sym::memory) => {
-                                codegen_fn_attrs.no_sanitize |= SanitizerSet::MEMORY
-                            }
-                            Some(sym::memtag) => {
-                                codegen_fn_attrs.no_sanitize |= SanitizerSet::MEMTAG
-                            }
-                            Some(sym::shadow_call_stack) => {
-                                codegen_fn_attrs.no_sanitize |= SanitizerSet::SHADOWCALLSTACK
-                            }
-                            Some(sym::thread) => {
-                                codegen_fn_attrs.no_sanitize |= SanitizerSet::THREAD
-                            }
-                            Some(sym::hwaddress) => {
-                                codegen_fn_attrs.no_sanitize |= SanitizerSet::HWADDRESS
-                            }
-                            _ => {
-                                tcx.dcx().emit_err(errors::InvalidNoSanitize { span: item.span() });
-                            }
-                        }
-                    }
-                }
-            }
+            sym::sanitize => sanitize_span = Some(attr.span()),
             sym::instruction_set => {
                 codegen_fn_attrs.instruction_set =
                     attr.meta_item_list().and_then(|l| match &l[..] {
@@ -387,6 +355,8 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: LocalDefId) -> CodegenFnAttrs {
     codegen_fn_attrs.alignment =
         Ord::max(codegen_fn_attrs.alignment, tcx.sess.opts.unstable_opts.min_function_alignment);
 
+    // Compute the disabled sanitizers.
+    codegen_fn_attrs.no_sanitize |= tcx.disabled_sanitizers_for(did);
     // On trait methods, inherit the `#[align]` of the trait's method prototype.
     codegen_fn_attrs.alignment = Ord::max(codegen_fn_attrs.alignment, tcx.inherited_align(did));
 
@@ -453,11 +423,11 @@ fn codegen_fn_attrs(tcx: TyCtxt<'_>, did: LocalDefId) -> CodegenFnAttrs {
 
     if !codegen_fn_attrs.no_sanitize.is_empty()
         && codegen_fn_attrs.inline.always()
-        && let (Some(no_sanitize_span), Some(inline_span)) = (no_sanitize_span, inline_span)
+        && let (Some(sanitize_span), Some(inline_span)) = (sanitize_span, inline_span)
     {
         let hir_id = tcx.local_def_id_to_hir_id(did);
-        tcx.node_span_lint(lint::builtin::INLINE_NO_SANITIZE, hir_id, no_sanitize_span, |lint| {
-            lint.primary_message("`no_sanitize` will have no effect after inlining");
+        tcx.node_span_lint(lint::builtin::INLINE_NO_SANITIZE, hir_id, sanitize_span, |lint| {
+            lint.primary_message("setting `sanitize` off will have no effect after inlining");
             lint.span_note(inline_span, "inlining requested here");
         })
     }
@@ -553,6 +523,87 @@ fn opt_trait_item(tcx: TyCtxt<'_>, def_id: DefId) -> Option<DefId> {
     }
 }
 
+/// For an attr that has the `sanitize` attribute, read the list of
+/// disabled sanitizers.
+fn parse_sanitize_attr(tcx: TyCtxt<'_>, attr: &Attribute) -> SanitizerSet {
+    let mut result = SanitizerSet::empty();
+    if let Some(list) = attr.meta_item_list() {
+        for item in list.iter() {
+            let MetaItemInner::MetaItem(set) = item else {
+                tcx.dcx().emit_err(errors::InvalidSanitize { span: attr.span() });
+                break;
+            };
+            let segments = set.path.segments.iter().map(|x| x.ident.name).collect::<Vec<_>>();
+            match segments.as_slice() {
+                // Similar to clang, sanitize(address = ..) and
+                // sanitize(kernel_address = ..) control both ASan and KASan
+                // Source: https://reviews.llvm.org/D44981.
+                [sym::address] | [sym::kernel_address] if set.value_str() == Some(sym::off) => {
+                    result |= SanitizerSet::ADDRESS | SanitizerSet::KERNELADDRESS
+                }
+                [sym::address] | [sym::kernel_address] if set.value_str() == Some(sym::on) => {
+                    result &= !SanitizerSet::ADDRESS;
+                    result &= !SanitizerSet::KERNELADDRESS;
+                }
+                [sym::cfi] if set.value_str() == Some(sym::off) => result |= SanitizerSet::CFI,
+                [sym::cfi] if set.value_str() == Some(sym::on) => result &= !SanitizerSet::CFI,
+                [sym::kcfi] if set.value_str() == Some(sym::off) => result |= SanitizerSet::KCFI,
+                [sym::kcfi] if set.value_str() == Some(sym::on) => result &= !SanitizerSet::KCFI,
+                [sym::memory] if set.value_str() == Some(sym::off) => {
+                    result |= SanitizerSet::MEMORY
+                }
+                [sym::memory] if set.value_str() == Some(sym::on) => {
+                    result &= !SanitizerSet::MEMORY
+                }
+                [sym::memtag] if set.value_str() == Some(sym::off) => {
+                    result |= SanitizerSet::MEMTAG
+                }
+                [sym::memtag] if set.value_str() == Some(sym::on) => {
+                    result &= !SanitizerSet::MEMTAG
+                }
+                [sym::shadow_call_stack] if set.value_str() == Some(sym::off) => {
+                    result |= SanitizerSet::SHADOWCALLSTACK
+                }
+                [sym::shadow_call_stack] if set.value_str() == Some(sym::on) => {
+                    result &= !SanitizerSet::SHADOWCALLSTACK
+                }
+                [sym::thread] if set.value_str() == Some(sym::off) => {
+                    result |= SanitizerSet::THREAD
+                }
+                [sym::thread] if set.value_str() == Some(sym::on) => {
+                    result &= !SanitizerSet::THREAD
+                }
+                [sym::hwaddress] if set.value_str() == Some(sym::off) => {
+                    result |= SanitizerSet::HWADDRESS
+                }
+                [sym::hwaddress] if set.value_str() == Some(sym::on) => {
+                    result &= !SanitizerSet::HWADDRESS
+                }
+                _ => {
+                    tcx.dcx().emit_err(errors::InvalidSanitize { span: attr.span() });
+                }
+            }
+        }
+    }
+    result
+}
+
+fn disabled_sanitizers_for(tcx: TyCtxt<'_>, did: LocalDefId) -> SanitizerSet {
+    // Check for a sanitize annotation directly on this def.
+    if let Some(attr) = tcx.get_attr(did, sym::sanitize) {
+        return parse_sanitize_attr(tcx, attr);
+    }
+
+    // Otherwise backtrack.
+    match tcx.opt_local_parent(did) {
+        // Check the parent (recursively).
+        Some(parent) => tcx.disabled_sanitizers_for(parent),
+        // We reached the crate root without seeing an attribute, so
+        // there is no sanitizers to exclude.
+        None => SanitizerSet::empty(),
+    }
+}
+
 /// Checks if the provided DefId is a method in a trait impl for a trait which has track_caller
 /// applied to the method prototype.
 fn should_inherit_track_caller(tcx: TyCtxt<'_>, def_id: DefId) -> bool {
@@ -692,6 +743,11 @@ fn autodiff_attrs(tcx: TyCtxt<'_>, id: DefId) -> Option<AutoDiffAttrs> {
 }
 
 pub(crate) fn provide(providers: &mut Providers) {
-    *providers =
-        Providers { codegen_fn_attrs, should_inherit_track_caller, inherited_align, ..*providers };
+    *providers = Providers {
+        codegen_fn_attrs,
+        should_inherit_track_caller,
+        inherited_align,
+        disabled_sanitizers_for,
+        ..*providers
+    };
 }

compiler/rustc_codegen_ssa/src/errors.rs

@@ -1101,9 +1101,9 @@ impl IntoDiagArg for ExpectedPointerMutability {
 }
 
 #[derive(Diagnostic)]
-#[diag(codegen_ssa_invalid_no_sanitize)]
+#[diag(codegen_ssa_invalid_sanitize)]
 #[note]
-pub(crate) struct InvalidNoSanitize {
+pub(crate) struct InvalidSanitize {
     #[primary_span]
     pub span: Span,
 }

compiler/rustc_feature/src/builtin_attrs.rs

@@ -540,9 +540,8 @@ pub static BUILTIN_ATTRIBUTES: &[BuiltinAttribute] = &[
     ungated!(track_caller, Normal, template!(Word), WarnFollowing, EncodeCrossCrate::Yes),
     ungated!(instruction_set, Normal, template!(List: "set"), ErrorPreceding, EncodeCrossCrate::No),
     gated!(
-        no_sanitize, Normal,
-        template!(List: "address, kcfi, memory, thread"), DuplicatesOk,
-        EncodeCrossCrate::No, experimental!(no_sanitize)
+        sanitize, Normal, template!(List: r#"address = "on|off", cfi = "on|off""#), ErrorPreceding,
+        EncodeCrossCrate::No, sanitize, experimental!(sanitize),
     ),
     gated!(
         coverage, Normal, template!(OneOf: &[sym::off, sym::on]),

compiler/rustc_feature/src/removed.rs

@@ -190,6 +190,9 @@ declare_features! (
     (removed, no_coverage, "1.74.0", Some(84605), Some("renamed to `coverage_attribute`"), 114656),
     /// Allows `#[no_debug]`.
     (removed, no_debug, "1.43.0", Some(29721), Some("removed due to lack of demand"), 69667),
+    // Allows the use of `no_sanitize` attribute.
+    /// The feature was renamed to `sanitize` and the attribute to `#[sanitize(xyz = "on|off")]`
+    (removed, no_sanitize, "CURRENT_RUSTC_VERSION", Some(39699), Some(r#"renamed to sanitize(xyz = "on|off")"#), 1234),
     /// Note: this feature was previously recorded in a separate
     /// `STABLE_REMOVED` list because it, uniquely, was once stable but was
     /// then removed. But there was no utility storing it separately, so now

compiler/rustc_feature/src/unstable.rs

@@ -591,8 +591,6 @@ declare_features! (
     (unstable, new_range, "1.86.0", Some(123741)),
     /// Allows `#![no_core]`.
     (unstable, no_core, "1.3.0", Some(29639)),
-    /// Allows the use of `no_sanitize` attribute.
-    (unstable, no_sanitize, "1.42.0", Some(39699)),
     /// Allows using the `non_exhaustive_omitted_patterns` lint.
     (unstable, non_exhaustive_omitted_patterns_lint, "1.57.0", Some(89554)),
     /// Allows `for<T>` binders in where-clauses
@@ -625,6 +623,8 @@ declare_features! (
     (unstable, return_type_notation, "1.70.0", Some(109417)),
     /// Allows `extern "rust-cold"`.
     (unstable, rust_cold_cc, "1.63.0", Some(97544)),
+    /// Allows the use of the `sanitize` attribute.
+    (unstable, sanitize, "CURRENT_RUSTC_VERSION", Some(39699)),
     /// Allows the use of SIMD types in functions declared in `extern` blocks.
     (unstable, simd_ffi, "1.0.0", Some(27731)),
     /// Allows specialization of implementations (RFC 1210).

compiler/rustc_lint_defs/src/builtin.rs

@@ -2298,18 +2298,18 @@ declare_lint! {
 
 declare_lint! {
     /// The `inline_no_sanitize` lint detects incompatible use of
-    /// [`#[inline(always)]`][inline] and [`#[no_sanitize(...)]`][no_sanitize].
+    /// [`#[inline(always)]`][inline] and [`#[sanitize(xyz = "off")]`][sanitize].
     ///
     /// [inline]: https://doc.rust-lang.org/reference/attributes/codegen.html#the-inline-attribute
-    /// [no_sanitize]: https://doc.rust-lang.org/nightly/unstable-book/language-features/no-sanitize.html
+    /// [sanitize]: https://doc.rust-lang.org/nightly/unstable-book/language-features/no-sanitize.html
     ///
     /// ### Example
     ///
     /// ```rust
-    /// #![feature(no_sanitize)]
+    /// #![feature(sanitize)]
     ///
     /// #[inline(always)]
-    /// #[no_sanitize(address)]
+    /// #[sanitize(address = "off")]
     /// fn x() {}
     ///
     /// fn main() {
@@ -2322,11 +2322,11 @@ declare_lint! {
     /// ### Explanation
     ///
     /// The use of the [`#[inline(always)]`][inline] attribute prevents the
-    /// the [`#[no_sanitize(...)]`][no_sanitize] attribute from working.
+    /// the [`#[sanitize(xyz = "off")]`][sanitize] attribute from working.
     /// Consider temporarily removing `inline` attribute.
     pub INLINE_NO_SANITIZE,
     Warn,
-    "detects incompatible use of `#[inline(always)]` and `#[no_sanitize(...)]`",
+    r#"detects incompatible use of `#[inline(always)]` and `#[sanitize(... = "off")]`"#,
 }
 
 declare_lint! {

compiler/rustc_middle/src/middle/codegen_fn_attrs.rs

@@ -40,8 +40,8 @@ pub struct CodegenFnAttrs {
     /// The `#[link_section = "..."]` attribute, or what executable section this
     /// should be placed in.
     pub link_section: Option<Symbol>,
-    /// The `#[no_sanitize(...)]` attribute. Indicates sanitizers for which
-    /// instrumentation should be disabled inside the annotated function.
+    /// The `#[sanitize(xyz = "off")]` attribute. Indicates sanitizers for which
+    /// instrumentation should be disabled inside the function.
     pub no_sanitize: SanitizerSet,
     /// The `#[instruction_set(set)]` attribute. Indicates if the generated code should
     /// be generated against a specific instruction set. Only usable on architectures which allow

compiler/rustc_middle/src/query/erase.rs

@@ -342,6 +342,7 @@ trivial! {
     rustc_span::Symbol,
     rustc_span::Ident,
     rustc_target::spec::PanicStrategy,
+    rustc_target::spec::SanitizerSet,
     rustc_type_ir::Variance,
     u32,
     usize,

compiler/rustc_middle/src/query/mod.rs

@@ -100,7 +100,7 @@ use rustc_session::lint::LintExpectationId;
 use rustc_span::def_id::LOCAL_CRATE;
 use rustc_span::source_map::Spanned;
 use rustc_span::{DUMMY_SP, Span, Symbol};
-use rustc_target::spec::PanicStrategy;
+use rustc_target::spec::{PanicStrategy, SanitizerSet};
 use {rustc_abi as abi, rustc_ast as ast, rustc_attr_data_structures as attr, rustc_hir as hir};
 
 use crate::infer::canonical::{self, Canonical};
@@ -2669,6 +2669,16 @@ rustc_queries! {
         desc { |tcx| "looking up anon const kind of `{}`", tcx.def_path_str(def_id) }
         separate_provide_extern
     }
+
+    /// Checks for the nearest `#[sanitize(xyz = "off")]` or
+    /// `#[sanitize(xyz = "on")]` on this def and any enclosing defs, up to the
+    /// crate root.
+    ///
+    /// Returns the set of sanitizers that is explicitly disabled for this def.
+    query disabled_sanitizers_for(key: LocalDefId) -> SanitizerSet {
+        desc { |tcx| "checking what set of sanitizers are enabled on `{}`", tcx.def_path_str(key) }
+        feedable
+    }
 }
 
 rustc_with_all_queries! { define_callbacks! }