[llvm-objcopy] Add --skip-symbol and --skip-symbols options (#80873)

Add --skip-symbol and --skip-symbols options that allow to skip symbols
when executing other options that can change the symbol's name, binding
or visibility, similar to an existing option --keep-symbol that keeps a
symbol from being removed by other options.

llvm/docs/CommandGuide/llvm-objcopy.rst

@@ -464,6 +464,19 @@ them.
  Read a list of symbols from <filename> and change their visibility to the
  specified value. Visibility values: default, internal, hidden, protected.
 
+.. option:: --skip-symbol <symbol>
+
+ Do not change the parameters of symbol ``<symbol>`` when executing other
+ options that can change the symbol's name, binding or visibility.
+
+.. option:: --skip-symbols <filename>
+
+ Do not change the parameters of symbols named in the file ``<filename>`` when
+ executing other options that can change the symbol's name, binding or
+ visibility. In the file, each line represents a single symbol, with leading
+ and trailing whitespace ignored, as is anything following a '#'.
+ Can be specified multiple times to read names from multiple files.
+
 .. option:: --split-dwo <dwo-file>
 
  Equivalent to running :program:`llvm-objcopy` with :option:`--extract-dwo` and

llvm/docs/ReleaseNotes.rst

@@ -175,6 +175,10 @@ Changes to the LLVM tools
   ``--set-symbols-visibility`` options for ELF input to change the
   visibility of symbols.
 
+* llvm-objcopy now supports ``--skip-symbol`` and ``--skip-symbols`` options
+  for ELF input to skip the specified symbols when executing other options
+  that can change a symbol's name, binding or visibility.
+
 Changes to LLDB
 ---------------------------------
 

llvm/include/llvm/ObjCopy/CommonConfig.h

@@ -233,6 +233,7 @@ struct CommonConfig {
   NameMatcher UnneededSymbolsToRemove;
   NameMatcher SymbolsToWeaken;
   NameMatcher SymbolsToKeepGlobal;
+  NameMatcher SymbolsToSkip;
 
   // Map options
   StringMap<SectionRename> SectionsToRename;

llvm/lib/ObjCopy/ConfigManager.cpp

@@ -15,7 +15,7 @@ namespace objcopy {
 
 Expected<const COFFConfig &> ConfigManager::getCOFFConfig() const {
   if (!Common.SplitDWO.empty() || !Common.SymbolsPrefix.empty() ||
-      !Common.SymbolsPrefixRemove.empty() ||
+      !Common.SymbolsPrefixRemove.empty() || !Common.SymbolsToSkip.empty() ||
       !Common.AllocSectionsPrefix.empty() || !Common.KeepSection.empty() ||
       !Common.SymbolsToGlobalize.empty() || !Common.SymbolsToKeep.empty() ||
       !Common.SymbolsToLocalize.empty() || !Common.SymbolsToWeaken.empty() ||
@@ -34,7 +34,7 @@ Expected<const COFFConfig &> ConfigManager::getCOFFConfig() const {
 
 Expected<const MachOConfig &> ConfigManager::getMachOConfig() const {
   if (!Common.SplitDWO.empty() || !Common.SymbolsPrefix.empty() ||
-      !Common.SymbolsPrefixRemove.empty() ||
+      !Common.SymbolsPrefixRemove.empty() || !Common.SymbolsToSkip.empty() ||
       !Common.AllocSectionsPrefix.empty() || !Common.KeepSection.empty() ||
       !Common.SymbolsToGlobalize.empty() || !Common.SymbolsToKeep.empty() ||
       !Common.SymbolsToLocalize.empty() ||
@@ -56,7 +56,7 @@ Expected<const MachOConfig &> ConfigManager::getMachOConfig() const {
 Expected<const WasmConfig &> ConfigManager::getWasmConfig() const {
   if (!Common.AddGnuDebugLink.empty() || Common.ExtractPartition ||
       !Common.SplitDWO.empty() || !Common.SymbolsPrefix.empty() ||
-      !Common.SymbolsPrefixRemove.empty() ||
+      !Common.SymbolsPrefixRemove.empty() || !Common.SymbolsToSkip.empty() ||
       !Common.AllocSectionsPrefix.empty() ||
       Common.DiscardMode != DiscardType::None || !Common.SymbolsToAdd.empty() ||
       !Common.SymbolsToGlobalize.empty() || !Common.SymbolsToLocalize.empty() ||
@@ -77,7 +77,7 @@ Expected<const WasmConfig &> ConfigManager::getWasmConfig() const {
 Expected<const XCOFFConfig &> ConfigManager::getXCOFFConfig() const {
   if (!Common.AddGnuDebugLink.empty() || Common.ExtractPartition ||
       !Common.SplitDWO.empty() || !Common.SymbolsPrefix.empty() ||
-      !Common.SymbolsPrefixRemove.empty() ||
+      !Common.SymbolsPrefixRemove.empty() || !Common.SymbolsToSkip.empty() ||
       !Common.AllocSectionsPrefix.empty() ||
       Common.DiscardMode != DiscardType::None || !Common.AddSection.empty() ||
       !Common.DumpSection.empty() || !Common.SymbolsToAdd.empty() ||

llvm/lib/ObjCopy/ELF/ELFObjcopy.cpp

@@ -291,6 +291,9 @@ static Error updateAndRemoveSymbols(const CommonConfig &Config,
     return Error::success();
 
   Obj.SymbolTable->updateSymbols([&](Symbol &Sym) {
+    if (Config.SymbolsToSkip.matches(Sym.Name))
+      return;
+
     // Common and undefined symbols don't make sense as local symbols, and can
     // even cause crashes if we localize those, so skip them.
     if (!Sym.isCommon() && Sym.getShndx() != SHN_UNDEF &&

llvm/test/tools/llvm-objcopy/ELF/skip-symbol.test

@@ -0,0 +1,100 @@
+## This test checks the functionality of options --skip-symbol and --skip-symbols.
+# RUN: yaml2obj %s -o %t.o
+# RUN: echo 'foo[2-3]' > %t.skip.regex
+
+## Check --skip-symbol functionality when changing symbol bindings.
+# RUN: llvm-objcopy %t.o %t2.o --localize-hidden --skip-symbol=foo3
+# RUN: llvm-readelf -s %t2.o | FileCheck %s --check-prefix=LH-SYM
+# LH-SYM-DAG: LOCAL  HIDDEN      1 foo1
+# LH-SYM-DAG: LOCAL  HIDDEN      1 foo2
+# LH-SYM-DAG: GLOBAL HIDDEN      1 foo3
+# LH-SYM-DAG: LOCAL  HIDDEN      1 foo4
+# LH-SYM-DAG: LOCAL  HIDDEN      1 foo5
+
+## Check --skip-symbols functionality when changing symbol bindings.
+# RUN: llvm-objcopy %t.o %t1.o --localize-hidden --skip-symbols=%t.skip.regex --regex
+# RUN: llvm-readelf -s %t1.o | FileCheck %s --check-prefix=LH-SYMS
+# LH-SYMS-DAG: LOCAL  HIDDEN      1 foo1
+# LH-SYMS-DAG: GLOBAL HIDDEN      1 foo2
+# LH-SYMS-DAG: GLOBAL HIDDEN      1 foo3
+# LH-SYMS-DAG: LOCAL  HIDDEN      1 foo4
+# LH-SYMS-DAG: LOCAL  HIDDEN      1 foo5
+
+## Check --skip-symbol functionality when changing symbol names.
+# RUN: echo -e "foo1 bar1\nfoo2 bar2" > %t.renames.list
+# RUN: llvm-objcopy %t.o %t4.o --redefine-syms=%t.renames.list \
+# RUN:   --skip-symbol='fo*' --wildcard
+# RUN: llvm-readelf -s %t4.o | FileCheck %s --check-prefix=RS-SYM
+# RS-SYM-DAG: foo1
+# RS-SYM-DAG: foo2
+# RS-SYM-DAG: foo3
+# RS-SYM-DAG: foo4
+# RS-SYM-DAG: foo5
+
+## Check --skip-symbols functionality when changing symbol names.
+# RUN: llvm-objcopy %t.o %t3.o --redefine-syms=%t.renames.list \
+# RUN:   --skip-symbols=%t.skip.regex --regex
+# RUN: llvm-readelf -s %t3.o | FileCheck %s --check-prefix=RS-SYMS
+# RS-SYMS-DAG: bar1
+# RS-SYMS-DAG: foo2
+# RS-SYMS-DAG: foo3
+# RS-SYMS-DAG: foo4
+# RS-SYMS-DAG: foo5
+
+## Check the functionality when using skip options multiple times.
+# RUN: echo "foo3" > %t.symbol0.list
+# RUN: echo "foo4" > %t.symbol1.list
+# RUN: llvm-objcopy %t.o %t5.o --set-symbol-visibility='foo*'=internal --wildcard \
+# RUN:   --skip-symbol=foo1 --skip-symbol=foo2 \
+# RUN:   --skip-symbols=%t.symbol0.list --skip-symbols=%t.symbol1.list
+# RUN: llvm-readelf -s %t5.o | FileCheck %s --check-prefix=BOTH
+# BOTH-DAG: GLOBAL HIDDEN      1 foo1
+# BOTH-DAG: GLOBAL HIDDEN      1 foo2
+# BOTH-DAG: GLOBAL HIDDEN      1 foo3
+# BOTH-DAG: GLOBAL HIDDEN      1 foo4
+## Only foo5 is not skipped.
+# BOTH-DAG: GLOBAL INTERNAL    1 foo5
+
+## Check that using an invalid symbol name regex generates an error.
+# RUN: echo '*.' > %t.symbols.regex
+# RUN: not llvm-objcopy %t.o --skip-symbols=%t.symbols.regex --regex 2>&1 | \
+# RUN:   FileCheck %s --check-prefix=SYMBOL
+# RUN: not llvm-objcopy %t.o --skip-symbol='*.' --regex 2>&1 | \
+# RUN:   FileCheck %s --check-prefix=SYMBOL
+# SYMBOL: error: cannot compile regular expression '*.': repetition-operator operand invalid
+
+## Check passing an invalid filename generates an error.
+# RUN: not llvm-objcopy %t.o --skip-symbols=no_file 2>&1 | \
+# RUN:   FileCheck %s --check-prefix=FILE -DMSG=%errc_ENOENT
+# FILE: error: 'no_file': [[MSG]]
+
+!ELF
+FileHeader:
+  Class:   ELFCLASS64
+  Data:    ELFDATA2LSB
+  Type:    ET_REL
+  Machine: EM_X86_64
+Sections:
+  - Name: .text
+    Type: SHT_PROGBITS
+Symbols:
+  - Name:    foo1
+    Section: .text
+    Binding: STB_GLOBAL
+    Other:   [ STV_HIDDEN ]
+  - Name:    foo2
+    Section: .text
+    Binding: STB_GLOBAL
+    Other:   [ STV_HIDDEN ]
+  - Name:    foo3
+    Section: .text
+    Binding: STB_GLOBAL
+    Other:   [ STV_HIDDEN ]
+  - Name:    foo4
+    Section: .text
+    Binding: STB_GLOBAL
+    Other:   [ STV_HIDDEN ]
+  - Name:    foo5
+    Section: .text
+    Binding: STB_GLOBAL
+    Other:   [ STV_HIDDEN ]

llvm/tools/llvm-objcopy/ObjcopyOptions.cpp

@@ -978,6 +978,15 @@ objcopy::parseObjcopyOptions(ArrayRef<const char *> RawArgsArr,
             addSymbolsFromFile(Config.SymbolsToKeep, DC.Alloc, Arg->getValue(),
                                SymbolMatchStyle, ErrorCallback))
       return std::move(E);
+  for (auto *Arg : InputArgs.filtered(OBJCOPY_skip_symbol))
+    if (Error E = Config.SymbolsToSkip.addMatcher(NameOrPattern::create(
+            Arg->getValue(), SymbolMatchStyle, ErrorCallback)))
+      return std::move(E);
+  for (auto *Arg : InputArgs.filtered(OBJCOPY_skip_symbols))
+    if (Error E =
+            addSymbolsFromFile(Config.SymbolsToSkip, DC.Alloc, Arg->getValue(),
+                               SymbolMatchStyle, ErrorCallback))
+      return std::move(E);
   for (auto *Arg : InputArgs.filtered(OBJCOPY_add_symbol)) {
     Expected<NewSymbolInfo> SymInfo = parseNewSymbolInfo(Arg->getValue());
     if (!SymInfo)

llvm/tools/llvm-objcopy/ObjcopyOpts.td

@@ -206,6 +206,20 @@ defm keep_symbols
          "be repeated to read symbols from many files">,
       MetaVarName<"filename">;
 
+defm skip_symbol : Eq<"skip-symbol", "Do not change parameters of symbol <symbol> "
+                        "when executing other options that can change the symbol's "
+                        "name, binding or visibility">,
+                     MetaVarName<"symbol">;
+
+defm skip_symbols
+    : Eq<"skip-symbols",
+         "Read a list of symbols from <filename> and run as if "
+         "--skip-symbol=<symbol> is set for each one. <filename> "
+         "contains one symbol per line and may contain comments beginning with "
+         "'#'. Leading and trailing whitespace is stripped from each line. May "
+         "be repeated to read symbols from many files">,
+      MetaVarName<"filename">;
+
 defm dump_section
     : Eq<"dump-section",
          "Dump contents of section named <section> into file <file>">,