Auto merge of #95604 - nbdd0121:used2, r=petrochenkov

Generate synthetic object file to ensure all exported and used symbols participate in the linking

Fix #50007 and #47384

This is the synthetic object file approach that I described in #95363 (comment), allowing all exported and used symbols to be linked while still allowing them to be GCed.

Related #93791, #95363

r? `@petrochenkov`
cc `@carbotaniuman`

Author: bors

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -16,7 +16,7 @@ use rustc_errors::{FatalError, Handler};
 use rustc_hir::def_id::LOCAL_CRATE;
 use rustc_middle::bug;
 use rustc_middle::dep_graph::WorkProduct;
-use rustc_middle::middle::exported_symbols::SymbolExportLevel;
+use rustc_middle::middle::exported_symbols::{SymbolExportInfo, SymbolExportLevel};
 use rustc_session::cgu_reuse_tracker::CguReuse;
 use rustc_session::config::{self, CrateType, Lto};
 use tracing::{debug, info};
@@ -55,8 +55,8 @@ fn prepare_lto(
         Lto::No => panic!("didn't request LTO but we're doing LTO"),
     };
 
-    let symbol_filter = &|&(ref name, level): &(String, SymbolExportLevel)| {
-        if level.is_below_threshold(export_threshold) {
+    let symbol_filter = &|&(ref name, info): &(String, SymbolExportInfo)| {
+        if info.level.is_below_threshold(export_threshold) || info.used {
             Some(CString::new(name.as_str()).unwrap())
         } else {
             None

compiler/rustc_codegen_ssa/src/back/link.rs

@@ -7,6 +7,7 @@ use rustc_errors::{ErrorGuaranteed, Handler};
 use rustc_fs_util::fix_windows_verbatim_for_gcc;
 use rustc_hir::def_id::CrateNum;
 use rustc_middle::middle::dependency_format::Linkage;
+use rustc_middle::middle::exported_symbols::SymbolExportKind;
 use rustc_session::config::{self, CFGuard, CrateType, DebugInfo, LdImpl, Strip};
 use rustc_session::config::{OutputFilenames, OutputType, PrintRequest, SplitDwarfKind};
 use rustc_session::cstore::DllImport;
@@ -1655,6 +1656,73 @@ fn add_post_link_args(cmd: &mut dyn Linker, sess: &Session, flavor: LinkerFlavor
     }
 }
 
+/// Add a synthetic object file that contains reference to all symbols that we want to expose to
+/// the linker.
+///
+/// Background: we implement rlibs as static library (archives). Linkers treat archives
+/// differently from object files: all object files participate in linking, while archives will
+/// only participate in linking if they can satisfy at least one undefined reference (version
+/// scripts doesn't count). This causes `#[no_mangle]` or `#[used]` items to be ignored by the
+/// linker, and since they never participate in the linking, using `KEEP` in the linker scripts
+/// can't keep them either. This causes #47384.
+///
+/// To keep them around, we could use `--whole-archive` and equivalents to force rlib to
+/// participate in linking like object files, but this proves to be expensive (#93791). Therefore
+/// we instead just introduce an undefined reference to them. This could be done by `-u` command
+/// line option to the linker or `EXTERN(...)` in linker scripts, however they does not only
+/// introduce an undefined reference, but also make them the GC roots, preventing `--gc-sections`
+/// from removing them, and this is especially problematic for embedded programming where every
+/// byte counts.
+///
+/// This method creates a synthetic object file, which contains undefined references to all symbols
+/// that are necessary for the linking. They are only present in symbol table but not actually
+/// used in any sections, so the linker will therefore pick relevant rlibs for linking, but
+/// unused `#[no_mangle]` or `#[used]` can still be discard by GC sections.
+fn add_linked_symbol_object(
+    cmd: &mut dyn Linker,
+    sess: &Session,
+    tmpdir: &Path,
+    symbols: &[(String, SymbolExportKind)],
+) {
+    if symbols.is_empty() {
+        return;
+    }
+
+    let Some(mut file) = super::metadata::create_object_file(sess) else {
+        return;
+    };
+
+    // NOTE(nbdd0121): MSVC will hang if the input object file contains no sections,
+    // so add an empty section.
+    if file.format() == object::BinaryFormat::Coff {
+        file.add_section(Vec::new(), ".text".into(), object::SectionKind::Text);
+    }
+
+    for (sym, kind) in symbols.iter() {
+        file.add_symbol(object::write::Symbol {
+            name: sym.clone().into(),
+            value: 0,
+            size: 0,
+            kind: match kind {
+                SymbolExportKind::Text => object::SymbolKind::Text,
+                SymbolExportKind::Data => object::SymbolKind::Data,
+                SymbolExportKind::Tls => object::SymbolKind::Tls,
+            },
+            scope: object::SymbolScope::Unknown,
+            weak: false,
+            section: object::write::SymbolSection::Undefined,
+            flags: object::SymbolFlags::None,
+        });
+    }
+
+    let path = tmpdir.join("symbols.o");
+    let result = std::fs::write(&path, file.write().unwrap());
+    if let Err(e) = result {
+        sess.fatal(&format!("failed to write {}: {}", path.display(), e));
+    }
+    cmd.add_object(&path);
+}
+
 /// Add object files containing code from the current crate.
 fn add_local_crate_regular_objects(cmd: &mut dyn Linker, codegen_results: &CodegenResults) {
     for obj in codegen_results.modules.iter().filter_map(|m| m.object.as_ref()) {
@@ -1795,6 +1863,13 @@ fn linker_with_args<'a, B: ArchiveBuilder<'a>>(
     // Pre-link CRT objects.
     add_pre_link_objects(cmd, sess, link_output_kind, crt_objects_fallback);
 
+    add_linked_symbol_object(
+        cmd,
+        sess,
+        tmpdir,
+        &codegen_results.crate_info.linked_symbols[&crate_type],
+    );
+
     // Sanitizer libraries.
     add_sanitizer_libraries(sess, crate_type, cmd);
 

compiler/rustc_codegen_ssa/src/back/linker.rs

@@ -12,6 +12,7 @@ use std::{env, mem, str};
 
 use rustc_hir::def_id::{CrateNum, LOCAL_CRATE};
 use rustc_middle::middle::dependency_format::Linkage;
+use rustc_middle::middle::exported_symbols::{ExportedSymbol, SymbolExportInfo, SymbolExportKind};
 use rustc_middle::ty::TyCtxt;
 use rustc_serialize::{json, Encoder};
 use rustc_session::config::{self, CrateType, DebugInfo, LinkerPluginLto, Lto, OptLevel, Strip};
@@ -1518,6 +1519,29 @@ impl<'a> L4Bender<'a> {
     }
 }
 
+fn for_each_exported_symbols_include_dep<'tcx>(
+    tcx: TyCtxt<'tcx>,
+    crate_type: CrateType,
+    mut callback: impl FnMut(ExportedSymbol<'tcx>, SymbolExportInfo, CrateNum),
+) {
+    for &(symbol, info) in tcx.exported_symbols(LOCAL_CRATE).iter() {
+        callback(symbol, info, LOCAL_CRATE);
+    }
+
+    let formats = tcx.dependency_formats(());
+    let deps = formats.iter().find_map(|(t, list)| (*t == crate_type).then_some(list)).unwrap();
+
+    for (index, dep_format) in deps.iter().enumerate() {
+        let cnum = CrateNum::new(index + 1);
+        // For each dependency that we are linking to statically ...
+        if *dep_format == Linkage::Static {
+            for &(symbol, info) in tcx.exported_symbols(cnum).iter() {
+                callback(symbol, info, cnum);
+            }
+        }
+    }
+}
+
 pub(crate) fn exported_symbols(tcx: TyCtxt<'_>, crate_type: CrateType) -> Vec<String> {
     if let Some(ref exports) = tcx.sess.target.override_export_symbols {
         return exports.iter().map(ToString::to_string).collect();
@@ -1526,34 +1550,38 @@ pub(crate) fn exported_symbols(tcx: TyCtxt<'_>, crate_type: CrateType) -> Vec<St
     let mut symbols = Vec::new();
 
     let export_threshold = symbol_export::crates_export_threshold(&[crate_type]);
-    for &(symbol, level) in tcx.exported_symbols(LOCAL_CRATE).iter() {
-        if level.is_below_threshold(export_threshold) {
-            symbols.push(symbol_export::symbol_name_for_instance_in_crate(
-                tcx,
-                symbol,
-                LOCAL_CRATE,
-            ));
+    for_each_exported_symbols_include_dep(tcx, crate_type, |symbol, info, cnum| {
+        if info.level.is_below_threshold(export_threshold) {
+            symbols.push(symbol_export::symbol_name_for_instance_in_crate(tcx, symbol, cnum));
         }
-    }
+    });
 
-    let formats = tcx.dependency_formats(());
-    let deps = formats.iter().find_map(|(t, list)| (*t == crate_type).then_some(list)).unwrap();
-
-    for (index, dep_format) in deps.iter().enumerate() {
-        let cnum = CrateNum::new(index + 1);
-        // For each dependency that we are linking to statically ...
-        if *dep_format == Linkage::Static {
-            // ... we add its symbol list to our export list.
-            for &(symbol, level) in tcx.exported_symbols(cnum).iter() {
-                if !level.is_below_threshold(export_threshold) {
-                    continue;
-                }
+    symbols
+}
 
-                symbols.push(symbol_export::symbol_name_for_instance_in_crate(tcx, symbol, cnum));
-            }
+pub(crate) fn linked_symbols(
+    tcx: TyCtxt<'_>,
+    crate_type: CrateType,
+) -> Vec<(String, SymbolExportKind)> {
+    match crate_type {
+        CrateType::Executable | CrateType::Cdylib => (),
+        CrateType::Staticlib | CrateType::ProcMacro | CrateType::Rlib | CrateType::Dylib => {
+            return Vec::new();
         }
     }
 
+    let mut symbols = Vec::new();
+
+    let export_threshold = symbol_export::crates_export_threshold(&[crate_type]);
+    for_each_exported_symbols_include_dep(tcx, crate_type, |symbol, info, cnum| {
+        if info.level.is_below_threshold(export_threshold) || info.used {
+            symbols.push((
+                symbol_export::symbol_name_for_instance_in_crate(tcx, symbol, cnum),
+                info.kind,
+            ));
+        }
+    });
+
     symbols
 }
 

compiler/rustc_codegen_ssa/src/back/metadata.rs

@@ -94,7 +94,7 @@ fn search_for_metadata<'a>(
         .map_err(|e| format!("failed to read {} section in '{}': {}", section, path.display(), e))
 }
 
-fn create_object_file(sess: &Session) -> Option<write::Object<'static>> {
+pub(crate) fn create_object_file(sess: &Session) -> Option<write::Object<'static>> {
     let endianness = match sess.target.options.endian {
         Endian::Little => Endianness::Little,
         Endian::Big => Endianness::Big,