[LLD] [COFF] Implement a GNU/ELF like -wrap option

Add a simple forwarding option in the MinGW frontend, and implement
the private -wrap option in the COFF linker.

The feature in lld-link isn't gated by the -lldmingw option, but
the option is left as a private, undocumented option primarily
used by the MinGW driver.

The implementation is significantly based on the support for --wrap
in the ELF linker, but many small nuance details are different
between the ELF and COFF linkers, ending up with more than a few
implementation differences.

This fixes https://bugs.llvm.org/show_bug.cgi?id=47384.

Differential Revision: https://reviews.llvm.org/D89004
This commit is contained in:
Martin Storsjö 2020-10-06 13:54:49 +03:00
parent 9803cf57d6
commit a012c704b5
18 changed files with 444 additions and 3 deletions

View File

@ -2008,6 +2008,12 @@ void LinkerDriver::link(ArrayRef<const char *> argsArr) {
while (run());
}
// Create wrapped symbols for -wrap option.
std::vector<WrappedSymbol> wrapped = addWrappedSymbols(args);
// Load more object files that might be needed for wrapped symbols.
if (!wrapped.empty())
while (run());
if (config->autoImport) {
// MinGW specific.
// Load any further object files that might be needed for doing automatic
@ -2051,6 +2057,10 @@ void LinkerDriver::link(ArrayRef<const char *> argsArr) {
// references to the symbols we use from them.
run();
// Apply symbol renames for -wrap.
if (!wrapped.empty())
wrapSymbols(wrapped);
// Resolve remaining undefined symbols and warn about imported locals.
symtab->resolveRemainingUndefines();
if (errorCount())

View File

@ -148,6 +148,8 @@ public:
ArrayRef<SectionChunk *> getGuardLJmpChunks() { return guardLJmpChunks; }
ArrayRef<Symbol *> getSymbols() { return symbols; }
MutableArrayRef<Symbol *> getMutableSymbols() { return symbols; }
ArrayRef<uint8_t> getDebugSection(StringRef secName);
// Returns a Symbol object for the symbolIndex'th symbol in the

View File

@ -139,6 +139,11 @@ void BitcodeCompiler::add(BitcodeFile &f) {
r.VisibleToRegularObj = sym->isUsedInRegularObj;
if (r.Prevailing)
undefine(sym);
// We tell LTO to not apply interprocedural optimization for wrapped
// (with -wrap) symbols because otherwise LTO would inline them while
// their values are still not final.
r.LinkerRedefined = !sym->canInline;
}
checkError(ltoObj->add(std::move(f.obj), resols));
}

View File

@ -7,9 +7,14 @@
//===----------------------------------------------------------------------===//
#include "MinGW.h"
#include "Driver.h"
#include "InputFiles.h"
#include "SymbolTable.h"
#include "lld/Common/ErrorHandler.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/Object/COFF.h"
#include "llvm/Support/Parallel.h"
#include "llvm/Support/Path.h"
#include "llvm/Support/raw_ostream.h"
@ -173,3 +178,73 @@ void lld::coff::writeDefFile(StringRef name) {
os << "\n";
}
}
static StringRef mangle(Twine sym) {
assert(config->machine != IMAGE_FILE_MACHINE_UNKNOWN);
if (config->machine == I386)
return saver.save("_" + sym);
return saver.save(sym);
}
// Handles -wrap option.
//
// This function instantiates wrapper symbols. At this point, they seem
// like they are not being used at all, so we explicitly set some flags so
// that LTO won't eliminate them.
std::vector<WrappedSymbol>
lld::coff::addWrappedSymbols(opt::InputArgList &args) {
std::vector<WrappedSymbol> v;
DenseSet<StringRef> seen;
for (auto *arg : args.filtered(OPT_wrap)) {
StringRef name = arg->getValue();
if (!seen.insert(name).second)
continue;
Symbol *sym = symtab->findUnderscore(name);
if (!sym)
continue;
Symbol *real = symtab->addUndefined(mangle("__real_" + name));
Symbol *wrap = symtab->addUndefined(mangle("__wrap_" + name));
v.push_back({sym, real, wrap});
// These symbols may seem undefined initially, but don't bail out
// at symtab->reportUnresolvable() due to them, but let wrapSymbols
// below sort things out before checking finally with
// symtab->resolveRemainingUndefines().
sym->deferUndefined = true;
real->deferUndefined = true;
// We want to tell LTO not to inline symbols to be overwritten
// because LTO doesn't know the final symbol contents after renaming.
real->canInline = false;
sym->canInline = false;
// Tell LTO not to eliminate these symbols.
sym->isUsedInRegularObj = true;
if (!isa<Undefined>(wrap))
wrap->isUsedInRegularObj = true;
}
return v;
}
// Do renaming for -wrap by updating pointers to symbols.
//
// When this function is executed, only InputFiles and symbol table
// contain pointers to symbol objects. We visit them to replace pointers,
// so that wrapped symbols are swapped as instructed by the command line.
void lld::coff::wrapSymbols(ArrayRef<WrappedSymbol> wrapped) {
DenseMap<Symbol *, Symbol *> map;
for (const WrappedSymbol &w : wrapped) {
map[w.sym] = w.wrap;
map[w.real] = w.sym;
}
// Update pointers in input files.
parallelForEach(ObjFile::instances, [&](ObjFile *file) {
MutableArrayRef<Symbol *> syms = file->getMutableSymbols();
for (size_t i = 0, e = syms.size(); i != e; ++i)
if (Symbol *s = map.lookup(syms[i]))
syms[i] = s;
});
}

View File

@ -12,7 +12,10 @@
#include "Config.h"
#include "Symbols.h"
#include "lld/Common/LLVM.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/StringSet.h"
#include "llvm/Option/ArgList.h"
#include <vector>
namespace lld {
namespace coff {
@ -36,6 +39,24 @@ public:
void writeDefFile(StringRef name);
// The -wrap option is a feature to rename symbols so that you can write
// wrappers for existing functions. If you pass `-wrap:foo`, all
// occurrences of symbol `foo` are resolved to `__wrap_foo` (so, you are
// expected to write `__wrap_foo` function as a wrapper). The original
// symbol becomes accessible as `__real_foo`, so you can call that from your
// wrapper.
//
// This data structure is instantiated for each -wrap option.
struct WrappedSymbol {
Symbol *sym;
Symbol *real;
Symbol *wrap;
};
std::vector<WrappedSymbol> addWrappedSymbols(llvm::opt::InputArgList &args);
void wrapSymbols(ArrayRef<WrappedSymbol> wrapped);
} // namespace coff
} // namespace lld

View File

@ -252,6 +252,7 @@ def print_symbol_order: P<
"print-symbol-order",
"Print a symbol order specified by /call-graph-ordering-file and "
"/call-graph-profile-sort into the specified file">;
def wrap : P_priv<"wrap">;
// Flags for debugging
def lldmap : F<"lldmap">;

View File

@ -390,7 +390,7 @@ void SymbolTable::reportUnresolvable() {
for (auto &i : symMap) {
Symbol *sym = i.second;
auto *undef = dyn_cast<Undefined>(sym);
if (!undef)
if (!undef || sym->deferUndefined)
continue;
if (undef->getWeakAlias())
continue;
@ -482,6 +482,7 @@ std::pair<Symbol *, bool> SymbolTable::insert(StringRef name) {
sym = reinterpret_cast<Symbol *>(make<SymbolUnion>());
sym->isUsedInRegularObj = false;
sym->pendingArchiveLoad = false;
sym->canInline = true;
inserted = true;
}
return {sym, inserted};

View File

@ -103,8 +103,8 @@ protected:
explicit Symbol(Kind k, StringRef n = "")
: symbolKind(k), isExternal(true), isCOMDAT(false),
writtenToSymtab(false), pendingArchiveLoad(false), isGCRoot(false),
isRuntimePseudoReloc(false), nameSize(n.size()),
nameData(n.empty() ? nullptr : n.data()) {}
isRuntimePseudoReloc(false), deferUndefined(false), canInline(true),
nameSize(n.size()), nameData(n.empty() ? nullptr : n.data()) {}
const unsigned symbolKind : 8;
unsigned isExternal : 1;
@ -130,6 +130,16 @@ public:
unsigned isRuntimePseudoReloc : 1;
// True if we want to allow this symbol to be undefined in the early
// undefined check pass in SymbolTable::reportUnresolvable(), as it
// might be fixed up later.
unsigned deferUndefined : 1;
// False if LTO shouldn't inline whatever this symbol points to. If a symbol
// is overwritten after LTO, LTO shouldn't inline the symbol because it
// doesn't know the final contents of the symbol.
uint8_t canInline : 1;
protected:
// Symbol name length. Assume symbol lengths fit in a 32-bit integer.
uint32_t nameSize;
@ -468,7 +478,9 @@ void replaceSymbol(Symbol *s, ArgT &&... arg) {
"SymbolUnion not aligned enough");
assert(static_cast<Symbol *>(static_cast<T *>(nullptr)) == nullptr &&
"Not a Symbol");
bool canInline = s->canInline;
new (s) T(std::forward<ArgT>(arg)...);
s->canInline = canInline;
}
} // namespace coff

View File

@ -377,6 +377,8 @@ bool mingw::link(ArrayRef<const char *> argsArr, bool canExitEarly,
add("-includeoptional:" + StringRef(a->getValue()));
for (auto *a : args.filtered(OPT_delayload))
add("-delayload:" + StringRef(a->getValue()));
for (auto *a : args.filtered(OPT_wrap))
add("-wrap:" + StringRef(a->getValue()));
std::vector<StringRef> searchPaths;
for (auto *a : args.filtered(OPT_L)) {

View File

@ -91,6 +91,8 @@ defm whole_archive: B<"whole-archive",
def v: Flag<["-"], "v">, HelpText<"Display the version number">;
def verbose: F<"verbose">, HelpText<"Verbose mode">;
def version: F<"version">, HelpText<"Display the version number and exit">;
defm wrap: Eq<"wrap", "Use wrapper functions for symbol">,
MetaVarName<"<symbol>">;
// LLD specific options
def _HASH_HASH_HASH : Flag<["-"], "###">,

49
lld/test/COFF/wrap-i386.s Normal file
View File

@ -0,0 +1,49 @@
// REQUIRES: x86
// RUN: split-file %s %t.dir
// RUN: llvm-mc -filetype=obj -triple=i686-win32-gnu %t.dir/main.s -o %t.main.obj
// RUN: llvm-mc -filetype=obj -triple=i686-win32-gnu %t.dir/other.s -o %t.other.obj
// RUN: lld-link -out:%t.exe %t.main.obj %t.other.obj -entry:entry -subsystem:console -debug:symtab -safeseh:no -wrap:foo -wrap:nosuchsym
// RUN: llvm-objdump -d --print-imm-hex %t.exe | FileCheck %s
// CHECK: <_entry>:
// CHECK-NEXT: movl $0x11010, %edx
// CHECK-NEXT: movl $0x11010, %edx
// CHECK-NEXT: movl $0x11000, %edx
// RUN: llvm-readobj --symbols %t.exe > %t.dump
// RUN: FileCheck --check-prefix=SYM1 %s < %t.dump
// RUN: FileCheck --check-prefix=SYM2 %s < %t.dump
// RUN: FileCheck --check-prefix=SYM3 %s < %t.dump
// _foo = 0xffc11000 = 4290842624
// ___wrap_foo = ffc11010 = 4290842640
// SYM1: Name: _foo
// SYM1-NEXT: Value: 4290842624
// SYM1-NEXT: Section: IMAGE_SYM_ABSOLUTE
// SYM1-NEXT: BaseType: Null
// SYM1-NEXT: ComplexType: Null
// SYM1-NEXT: StorageClass: External
// SYM2: Name: ___wrap_foo
// SYM2-NEXT: Value: 4290842640
// SYM2-NEXT: Section: IMAGE_SYM_ABSOLUTE
// SYM2-NEXT: BaseType: Null
// SYM2-NEXT: ComplexType: Null
// SYM2-NEXT: StorageClass: External
// SYM3-NOT: Name: ___real_foo
#--- main.s
.global _entry
_entry:
movl $_foo, %edx
movl $___wrap_foo, %edx
movl $___real_foo, %edx
#--- other.s
.global _foo
.global ___wrap_foo
.global ___real_foo
_foo = 0x11000
___wrap_foo = 0x11010
___real_foo = 0x11020

View File

@ -0,0 +1,36 @@
// REQUIRES: x86
// Check that wrapping works when the wrapped symbol is imported from a
// different DLL.
// RUN: split-file %s %t.dir
// RUN: llc %t.dir/main.ll -o %t.main.obj --filetype=obj
// RUN: llvm-as %t.dir/main.ll -o %t.main.bc
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %t.dir/lib.s -o %t.lib.obj
// RUN: lld-link -dll -out:%t.lib.dll %t.lib.obj -noentry -export:func -implib:%t.lib.lib
// RUN: lld-link -out:%t.exe %t.main.obj %t.lib.lib -entry:entry -subsystem:console -wrap:func
// RUN: lld-link -out:%t.exe %t.main.bc %t.lib.lib -entry:entry -subsystem:console -wrap:func
#--- main.ll
target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-w64-windows-gnu"
declare void @func()
define void @entry() {
call void @func()
ret void
}
declare void @__real_func()
define void @__wrap_func() {
call void @__real_func()
ret void
}
#--- lib.s
.global func
func:
ret

View File

@ -0,0 +1,36 @@
; REQUIRES: x86
; LTO
; RUN: llvm-as %s -o %t.obj
; RUN: lld-link -out:%t.exe %t.obj -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: cat %t.exe.resolution.txt | FileCheck -check-prefix=RESOLS %s
; ThinLTO
; RUN: opt -module-summary %s -o %t.obj
; RUN: lld-link -out:%t.exe %t.obj -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: cat %t.exe.resolution.txt | FileCheck -check-prefix=RESOLS %s
; Make sure that the 'r' (linker redefined) bit is set for bar and __real_bar
; in the resolutions file. The calls to bar and __real_bar will be routed to
; __wrap_bar and bar, respectively. So they cannot be inlined.
; RESOLS: ,bar,pxr{{$}}
; RESOLS: ,__real_bar,xr{{$}}
; RESOLS: ,__wrap_bar,px{{$}}
target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-w64-windows-gnu"
define void @bar() {
ret void
}
define void @entry() {
call void @bar()
ret void
}
declare void @__real_bar()
define void @__wrap_bar() {
call void @__real_bar()
ret void
}

View File

@ -0,0 +1,84 @@
; REQUIRES: x86
; RUN: split-file %s %t.dir
;; LTO
; RUN: llvm-as %t.dir/main.ll -o %t.main.bc
; RUN: llvm-as %t.dir/wrap.ll -o %t.wrap.bc
; RUN: llvm-as %t.dir/other.ll -o %t.other.bc
; RUN: rm -f %t.bc.lib
; RUN: llvm-ar rcs %t.bc.lib %t.wrap.bc %t.other.bc
;; ThinLTO
; RUN: opt -module-summary %t.dir/main.ll -o %t.main.thin
; RUN: opt -module-summary %t.dir/wrap.ll -o %t.wrap.thin
; RUN: opt -module-summary %t.dir/other.ll -o %t.other.thin
; RUN: rm -f %t.thin.lib
; RUN: llvm-ar rcs %t.thin.lib %t.wrap.thin %t.other.thin
;; Object
; RUN: llc %t.dir/main.ll -o %t.main.obj --filetype=obj
; RUN: llc %t.dir/wrap.ll -o %t.wrap.obj --filetype=obj
; RUN: llc %t.dir/other.ll -o %t.other.obj --filetype=obj
; RUN: rm -f %t.obj.lib
; RUN: llvm-ar rcs %t.obj.lib %t.wrap.obj %t.other.obj
;; This test verifies that -wrap works correctly for inter-module references to
;; the wrapped symbol, when LTO or ThinLTO is involved. It checks for various
;; combinations of bitcode and regular objects.
;; LTO + LTO
; RUN: lld-link -out:%t.bc-bc.exe %t.main.bc -libpath:%T %t.bc.lib -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: llvm-objdump -d %t.bc-bc.exe | FileCheck %s --check-prefixes=CHECK,JMP
;; LTO + Object
; RUN: lld-link -out:%t.bc-obj.exe %t.main.bc -libpath:%T %t.obj.lib -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: llvm-objdump -d %t.bc-obj.exe | FileCheck %s --check-prefixes=CHECK,JMP
;; Object + LTO
; RUN: lld-link -out:%t.obj-bc.exe %t.main.obj -libpath:%T %t.bc.lib -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: llvm-objdump -d %t.obj-bc.exe | FileCheck %s --check-prefixes=CHECK,CALL
;; ThinLTO + ThinLTO
; RUN: lld-link -out:%t.thin-thin.exe %t.main.thin -libpath:%T %t.thin.lib -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: llvm-objdump -d %t.thin-thin.exe | FileCheck %s --check-prefixes=CHECK,JMP
;; ThinLTO + Object
; RUN: lld-link -out:%t.thin-obj.exe %t.main.thin -libpath:%T %t.obj.lib -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: llvm-objdump -d %t.thin-obj.exe | FileCheck %s --check-prefixes=CHECK,JMP
;; Object + ThinLTO
; RUN: lld-link -out:%t.obj-thin.exe %t.main.obj -libpath:%T %t.thin.lib -entry:entry -subsystem:console -wrap:bar -debug:symtab -lldsavetemps
; RUN: llvm-objdump -d %t.obj-thin.exe | FileCheck %s --check-prefixes=CHECK,CALL
;; Make sure that calls in entry() are not eliminated and that bar is
;; routed to __wrap_bar.
; CHECK: <entry>:
; CHECK: {{jmp|callq}}{{.*}}<__wrap_bar>
;--- main.ll
target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-w64-windows-gnu"
declare void @bar()
define void @entry() {
call void @bar()
ret void
}
;--- wrap.ll
target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-w64-windows-gnu"
declare void @other()
define void @__wrap_bar() {
call void @other()
ret void
}
;--- other.ll
target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-w64-windows-gnu"
define void @other() {
ret void
}

View File

@ -0,0 +1,21 @@
// REQUIRES: x86
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %s -o %t.obj
// RUN: not lld-link -lldmingw -out:%t.exe %t.obj -entry:entry -subsystem:console -wrap:foo 2>&1 | FileCheck %s
// Check that we error out properly with an undefined symbol, if
// __real_foo is referenced and missing, even if the -lldmingw flag is set
// (which otherwise tolerates certain cases of references to missing
// sections, to tolerate certain GCC pecularities).
// CHECK: error: undefined symbol: foo
.global entry
entry:
call foo
ret
.global __wrap_foo
__wrap_foo:
call __real_foo
ret

View File

@ -0,0 +1,29 @@
// REQUIRES: x86
// RUN: split-file %s %t.dir
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %t.dir/main.s -o %t.main.obj
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %t.dir/wrap.s -o %t.wrap.obj
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %t.dir/other.s -o %t.other.obj
// RUN: rm -f %t.lib
// RUN: llvm-ar rcs %t.lib %t.wrap.obj %t.other.obj
// RUN: lld-link -out:%t.exe %t.main.obj -libpath:%T %t.lib -entry:entry -subsystem:console -wrap:foo
// Note: No real definition of foo exists here, but that works fine as long
// as there's no actual references to __real_foo.
#--- main.s
.global entry
entry:
call foo
ret
#--- wrap.s
.global __wrap_foo
__wrap_foo:
call other_func
ret
#--- other.s
.global other_func
other_func:
ret

51
lld/test/COFF/wrap.s Normal file
View File

@ -0,0 +1,51 @@
// REQUIRES: x86
// RUN: split-file %s %t.dir
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %t.dir/main.s -o %t.main.obj
// RUN: llvm-mc -filetype=obj -triple=x86_64-win32-gnu %t.dir/other.s -o %t.other.obj
// RUN: lld-link -out:%t.exe %t.main.obj %t.other.obj -entry:entry -subsystem:console -debug:symtab -wrap:foo -wrap:nosuchsym
// RUN: llvm-objdump -d --print-imm-hex %t.exe | FileCheck %s
// RUN: lld-link -out:%t.exe %t.main.obj %t.other.obj -entry:entry -subsystem:console -debug:symtab -wrap:foo -wrap:foo -wrap:nosuchsym
// RUN: llvm-objdump -d --print-imm-hex %t.exe | FileCheck %s
// CHECK: <entry>:
// CHECK-NEXT: movl $0x11010, %edx
// CHECK-NEXT: movl $0x11010, %edx
// CHECK-NEXT: movl $0x11000, %edx
// RUN: llvm-readobj --symbols %t.exe > %t.dump
// RUN: FileCheck --check-prefix=SYM1 %s < %t.dump
// RUN: FileCheck --check-prefix=SYM2 %s < %t.dump
// RUN: FileCheck --check-prefix=SYM3 %s < %t.dump
// foo = 0xC0011000 = 3221295104
// __wrap_foo = 0xC0011010 = 3221295120
// SYM1: Name: foo
// SYM1-NEXT: Value: 3221295104
// SYM1-NEXT: Section: IMAGE_SYM_ABSOLUTE
// SYM1-NEXT: BaseType: Null
// SYM1-NEXT: ComplexType: Null
// SYM1-NEXT: StorageClass: External
// SYM2: Name: __wrap_foo
// SYM2-NEXT: Value: 3221295120
// SYM2-NEXT: Section: IMAGE_SYM_ABSOLUTE
// SYM2-NEXT: BaseType: Null
// SYM2-NEXT: ComplexType: Null
// SYM2-NEXT: StorageClass: External
// SYM3-NOT: Name: __real_foo
#--- main.s
.global entry
entry:
movl $foo, %edx
movl $__wrap_foo, %edx
movl $__real_foo, %edx
#--- other.s
.global foo
.global __wrap_foo
.global __real_foo
foo = 0x11000
__wrap_foo = 0x11010
__real_foo = 0x11020

View File

@ -281,3 +281,7 @@ ALLOW_MULTIPLE_DEFINITION: -force:multiple
RUN: ld.lld -### -m i386pep foo.o --allow-multiple-definition --no-allow-multiple-definition | FileCheck -check-prefix NO_ALLOW_MULTIPLE_DEFINITION %s
RUN: ld.lld -### -m i386pep foo.o -allow-multiple-definition -no-allow-multiple-definition | FileCheck -check-prefix NO_ALLOW_MULTIPLE_DEFINITION %s
NO_ALLOW_MULTIPLE_DEFINITION-NOT: -force:multiple
RUN: ld.lld -### -m i386pep foo.o -wrap foo1 --wrap foo2 | FileCheck -check-prefix WRAP %s
RUN: ld.lld -### -m i386pep foo.o -wrap=foo1 --wrap=foo2 | FileCheck -check-prefix WRAP %s
WRAP: -wrap:foo1 -wrap:foo2