1 //===- llvm-link.cpp - Low-level LLVM linker ------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This utility may be invoked in the following manner:
10 //  llvm-link a.bc b.bc c.bc -o x.bc
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/ADT/STLExtras.h"
15 #include "llvm/BinaryFormat/Magic.h"
16 #include "llvm/Bitcode/BitcodeReader.h"
17 #include "llvm/Bitcode/BitcodeWriter.h"
18 #include "llvm/IR/AutoUpgrade.h"
19 #include "llvm/IR/DiagnosticInfo.h"
20 #include "llvm/IR/DiagnosticPrinter.h"
21 #include "llvm/IR/LLVMContext.h"
22 #include "llvm/IR/Module.h"
23 #include "llvm/IR/ModuleSummaryIndex.h"
24 #include "llvm/IR/Verifier.h"
25 #include "llvm/IRReader/IRReader.h"
26 #include "llvm/Linker/Linker.h"
27 #include "llvm/Object/Archive.h"
28 #include "llvm/Support/CommandLine.h"
29 #include "llvm/Support/FileSystem.h"
30 #include "llvm/Support/InitLLVM.h"
31 #include "llvm/Support/Path.h"
32 #include "llvm/Support/SourceMgr.h"
33 #include "llvm/Support/SystemUtils.h"
34 #include "llvm/Support/ToolOutputFile.h"
35 #include "llvm/Support/WithColor.h"
36 #include "llvm/Transforms/IPO/FunctionImport.h"
37 #include "llvm/Transforms/IPO/Internalize.h"
38 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
39 
40 #include <memory>
41 #include <utility>
42 using namespace llvm;
43 
44 static cl::list<std::string>
45 InputFilenames(cl::Positional, cl::OneOrMore,
46                cl::desc("<input bitcode files>"));
47 
48 static cl::list<std::string> OverridingInputs(
49     "override", cl::ZeroOrMore, cl::value_desc("filename"),
50     cl::desc(
51         "input bitcode file which can override previously defined symbol(s)"));
52 
53 // Option to simulate function importing for testing. This enables using
54 // llvm-link to simulate ThinLTO backend processes.
55 static cl::list<std::string> Imports(
56     "import", cl::ZeroOrMore, cl::value_desc("function:filename"),
57     cl::desc("Pair of function name and filename, where function should be "
58              "imported from bitcode in filename"));
59 
60 // Option to support testing of function importing. The module summary
61 // must be specified in the case were we request imports via the -import
62 // option, as well as when compiling any module with functions that may be
63 // exported (imported by a different llvm-link -import invocation), to ensure
64 // consistent promotion and renaming of locals.
65 static cl::opt<std::string>
66     SummaryIndex("summary-index", cl::desc("Module summary index filename"),
67                  cl::init(""), cl::value_desc("filename"));
68 
69 static cl::opt<std::string>
70 OutputFilename("o", cl::desc("Override output filename"), cl::init("-"),
71                cl::value_desc("filename"));
72 
73 static cl::opt<bool>
74 Internalize("internalize", cl::desc("Internalize linked symbols"));
75 
76 static cl::opt<bool>
77     DisableDITypeMap("disable-debug-info-type-map",
78                      cl::desc("Don't use a uniquing type map for debug info"));
79 
80 static cl::opt<bool>
81 OnlyNeeded("only-needed", cl::desc("Link only needed symbols"));
82 
83 static cl::opt<bool>
84 Force("f", cl::desc("Enable binary output on terminals"));
85 
86 static cl::opt<bool>
87     DisableLazyLoad("disable-lazy-loading",
88                     cl::desc("Disable lazy module loading"));
89 
90 static cl::opt<bool>
91     OutputAssembly("S", cl::desc("Write output as LLVM assembly"), cl::Hidden);
92 
93 static cl::opt<bool>
94 Verbose("v", cl::desc("Print information about actions taken"));
95 
96 static cl::opt<bool>
97 DumpAsm("d", cl::desc("Print assembly as linked"), cl::Hidden);
98 
99 static cl::opt<bool>
100 SuppressWarnings("suppress-warnings", cl::desc("Suppress all linking warnings"),
101                  cl::init(false));
102 
103 static cl::opt<bool> PreserveBitcodeUseListOrder(
104     "preserve-bc-uselistorder",
105     cl::desc("Preserve use-list order when writing LLVM bitcode."),
106     cl::init(true), cl::Hidden);
107 
108 static cl::opt<bool> PreserveAssemblyUseListOrder(
109     "preserve-ll-uselistorder",
110     cl::desc("Preserve use-list order when writing LLVM assembly."),
111     cl::init(false), cl::Hidden);
112 
113 static ExitOnError ExitOnErr;
114 
115 // Read the specified bitcode file in and return it. This routine searches the
116 // link path for the specified file to try to find it...
117 //
118 static std::unique_ptr<Module> loadFile(const char *argv0,
119                                         std::unique_ptr<MemoryBuffer> Buffer,
120                                         LLVMContext &Context,
121                                         bool MaterializeMetadata = true) {
122   SMDiagnostic Err;
123   if (Verbose)
124     errs() << "Loading '" << Buffer->getBufferIdentifier() << "'\n";
125   std::unique_ptr<Module> Result;
126   if (DisableLazyLoad)
127     Result = parseIR(*Buffer, Err, Context);
128   else
129     Result =
130         getLazyIRModule(std::move(Buffer), Err, Context, !MaterializeMetadata);
131 
132   if (!Result) {
133     Err.print(argv0, errs());
134     return nullptr;
135   }
136 
137   if (MaterializeMetadata) {
138     ExitOnErr(Result->materializeMetadata());
139     UpgradeDebugInfo(*Result);
140   }
141 
142   return Result;
143 }
144 
145 static std::unique_ptr<Module> loadArFile(const char *Argv0,
146                                           std::unique_ptr<MemoryBuffer> Buffer,
147                                           LLVMContext &Context) {
148   std::unique_ptr<Module> Result(new Module("ArchiveModule", Context));
149   StringRef ArchiveName = Buffer->getBufferIdentifier();
150   if (Verbose)
151     errs() << "Reading library archive file '" << ArchiveName
152            << "' to memory\n";
153   Error Err = Error::success();
154   object::Archive Archive(*Buffer, Err);
155   ExitOnErr(std::move(Err));
156   Linker L(*Result);
157   for (const object::Archive::Child &C : Archive.children(Err)) {
158     Expected<StringRef> Ename = C.getName();
159     if (Error E = Ename.takeError()) {
160       errs() << Argv0 << ": ";
161       WithColor::error()
162           << " failed to read name of archive member"
163           << ArchiveName << "'\n";
164       return nullptr;
165     }
166     std::string ChildName = Ename.get().str();
167     if (Verbose)
168       errs() << "Parsing member '" << ChildName
169              << "' of archive library to module.\n";
170     SMDiagnostic ParseErr;
171     Expected<MemoryBufferRef> MemBuf = C.getMemoryBufferRef();
172     if (Error E = MemBuf.takeError()) {
173       errs() << Argv0 << ": ";
174       WithColor::error() << " loading memory for member '" << ChildName
175                          << "' of archive library failed'" << ArchiveName
176                          << "'\n";
177       return nullptr;
178     };
179 
180     if (!isBitcode(reinterpret_cast<const unsigned char *>
181                    (MemBuf.get().getBufferStart()),
182                    reinterpret_cast<const unsigned char *>
183                    (MemBuf.get().getBufferEnd()))) {
184       errs() << Argv0 << ": ";
185       WithColor::error() << "  member of archive is not a bitcode file: '"
186                          << ChildName << "'\n";
187       return nullptr;
188     }
189 
190     std::unique_ptr<Module> M;
191     if (DisableLazyLoad)
192       M = parseIR(MemBuf.get(), ParseErr, Context);
193     else
194       M = getLazyIRModule(MemoryBuffer::getMemBuffer(MemBuf.get(), false),
195                           ParseErr, Context);
196 
197     if (!M.get()) {
198       errs() << Argv0 << ": ";
199       WithColor::error() << " parsing member '" << ChildName
200                          << "' of archive library failed'" << ArchiveName
201                          << "'\n";
202       return nullptr;
203     }
204     if (Verbose)
205       errs() << "Linking member '" << ChildName << "' of archive library.\n";
206     if (L.linkInModule(std::move(M)))
207       return nullptr;
208   } // end for each child
209   ExitOnErr(std::move(Err));
210   return Result;
211 }
212 
213 namespace {
214 
215 /// Helper to load on demand a Module from file and cache it for subsequent
216 /// queries during function importing.
217 class ModuleLazyLoaderCache {
218   /// Cache of lazily loaded module for import.
219   StringMap<std::unique_ptr<Module>> ModuleMap;
220 
221   /// Retrieve a Module from the cache or lazily load it on demand.
222   std::function<std::unique_ptr<Module>(const char *argv0,
223                                         const std::string &FileName)>
224       createLazyModule;
225 
226 public:
227   /// Create the loader, Module will be initialized in \p Context.
228   ModuleLazyLoaderCache(std::function<std::unique_ptr<Module>(
229                             const char *argv0, const std::string &FileName)>
230                             createLazyModule)
231       : createLazyModule(std::move(createLazyModule)) {}
232 
233   /// Retrieve a Module from the cache or lazily load it on demand.
234   Module &operator()(const char *argv0, const std::string &FileName);
235 
236   std::unique_ptr<Module> takeModule(const std::string &FileName) {
237     auto I = ModuleMap.find(FileName);
238     assert(I != ModuleMap.end());
239     std::unique_ptr<Module> Ret = std::move(I->second);
240     ModuleMap.erase(I);
241     return Ret;
242   }
243 };
244 
245 // Get a Module for \p FileName from the cache, or load it lazily.
246 Module &ModuleLazyLoaderCache::operator()(const char *argv0,
247                                           const std::string &Identifier) {
248   auto &Module = ModuleMap[Identifier];
249   if (!Module)
250     Module = createLazyModule(argv0, Identifier);
251   return *Module;
252 }
253 } // anonymous namespace
254 
255 namespace {
256 struct LLVMLinkDiagnosticHandler : public DiagnosticHandler {
257   bool handleDiagnostics(const DiagnosticInfo &DI) override {
258     unsigned Severity = DI.getSeverity();
259     switch (Severity) {
260     case DS_Error:
261       WithColor::error();
262       break;
263     case DS_Warning:
264       if (SuppressWarnings)
265         return true;
266       WithColor::warning();
267       break;
268     case DS_Remark:
269     case DS_Note:
270       llvm_unreachable("Only expecting warnings and errors");
271     }
272 
273     DiagnosticPrinterRawOStream DP(errs());
274     DI.print(DP);
275     errs() << '\n';
276     return true;
277   }
278 };
279 }
280 
281 /// Import any functions requested via the -import option.
282 static bool importFunctions(const char *argv0, Module &DestModule) {
283   if (SummaryIndex.empty())
284     return true;
285   std::unique_ptr<ModuleSummaryIndex> Index =
286       ExitOnErr(llvm::getModuleSummaryIndexForFile(SummaryIndex));
287 
288   // Map of Module -> List of globals to import from the Module
289   FunctionImporter::ImportMapTy ImportList;
290 
291   auto ModuleLoader = [&DestModule](const char *argv0,
292                                     const std::string &Identifier) {
293     std::unique_ptr<MemoryBuffer> Buffer =
294         ExitOnErr(errorOrToExpected(MemoryBuffer::getFileOrSTDIN(Identifier)));
295     return loadFile(argv0, std::move(Buffer), DestModule.getContext(), false);
296   };
297 
298   ModuleLazyLoaderCache ModuleLoaderCache(ModuleLoader);
299   for (const auto &Import : Imports) {
300     // Identify the requested function and its bitcode source file.
301     size_t Idx = Import.find(':');
302     if (Idx == std::string::npos) {
303       errs() << "Import parameter bad format: " << Import << "\n";
304       return false;
305     }
306     std::string FunctionName = Import.substr(0, Idx);
307     std::string FileName = Import.substr(Idx + 1, std::string::npos);
308 
309     // Load the specified source module.
310     auto &SrcModule = ModuleLoaderCache(argv0, FileName);
311 
312     if (verifyModule(SrcModule, &errs())) {
313       errs() << argv0 << ": " << FileName;
314       WithColor::error() << "input module is broken!\n";
315       return false;
316     }
317 
318     Function *F = SrcModule.getFunction(FunctionName);
319     if (!F) {
320       errs() << "Ignoring import request for non-existent function "
321              << FunctionName << " from " << FileName << "\n";
322       continue;
323     }
324     // We cannot import weak_any functions without possibly affecting the
325     // order they are seen and selected by the linker, changing program
326     // semantics.
327     if (F->hasWeakAnyLinkage()) {
328       errs() << "Ignoring import request for weak-any function " << FunctionName
329              << " from " << FileName << "\n";
330       continue;
331     }
332 
333     if (Verbose)
334       errs() << "Importing " << FunctionName << " from " << FileName << "\n";
335 
336     auto &Entry = ImportList[FileName];
337     Entry.insert(F->getGUID());
338   }
339   auto CachedModuleLoader = [&](StringRef Identifier) {
340     return ModuleLoaderCache.takeModule(std::string(Identifier));
341   };
342   FunctionImporter Importer(*Index, CachedModuleLoader,
343                             /*ClearDSOLocalOnDeclarations=*/false);
344   ExitOnErr(Importer.importFunctions(DestModule, ImportList));
345 
346   return true;
347 }
348 
349 static bool linkFiles(const char *argv0, LLVMContext &Context, Linker &L,
350                       const cl::list<std::string> &Files,
351                       unsigned Flags) {
352   // Filter out flags that don't apply to the first file we load.
353   unsigned ApplicableFlags = Flags & Linker::Flags::OverrideFromSrc;
354   // Similar to some flags, internalization doesn't apply to the first file.
355   bool InternalizeLinkedSymbols = false;
356   for (const auto &File : Files) {
357     std::unique_ptr<MemoryBuffer> Buffer =
358         ExitOnErr(errorOrToExpected(MemoryBuffer::getFileOrSTDIN(File)));
359 
360     std::unique_ptr<Module> M =
361         identify_magic(Buffer->getBuffer()) == file_magic::archive
362             ? loadArFile(argv0, std::move(Buffer), Context)
363             : loadFile(argv0, std::move(Buffer), Context);
364     if (!M.get()) {
365       errs() << argv0 << ": ";
366       WithColor::error() << " loading file '" << File << "'\n";
367       return false;
368     }
369 
370     // Note that when ODR merging types cannot verify input files in here When
371     // doing that debug metadata in the src module might already be pointing to
372     // the destination.
373     if (DisableDITypeMap && verifyModule(*M, &errs())) {
374       errs() << argv0 << ": " << File << ": ";
375       WithColor::error() << "input module is broken!\n";
376       return false;
377     }
378 
379     // If a module summary index is supplied, load it so linkInModule can treat
380     // local functions/variables as exported and promote if necessary.
381     if (!SummaryIndex.empty()) {
382       std::unique_ptr<ModuleSummaryIndex> Index =
383           ExitOnErr(llvm::getModuleSummaryIndexForFile(SummaryIndex));
384 
385       // Conservatively mark all internal values as promoted, since this tool
386       // does not do the ThinLink that would normally determine what values to
387       // promote.
388       for (auto &I : *Index) {
389         for (auto &S : I.second.SummaryList) {
390           if (GlobalValue::isLocalLinkage(S->linkage()))
391             S->setLinkage(GlobalValue::ExternalLinkage);
392         }
393       }
394 
395       // Promotion
396       if (renameModuleForThinLTO(*M, *Index,
397                                  /*ClearDSOLocalOnDeclarations=*/false))
398         return true;
399     }
400 
401     if (Verbose)
402       errs() << "Linking in '" << File << "'\n";
403 
404     bool Err = false;
405     if (InternalizeLinkedSymbols) {
406       Err = L.linkInModule(
407           std::move(M), ApplicableFlags, [](Module &M, const StringSet<> &GVS) {
408             internalizeModule(M, [&GVS](const GlobalValue &GV) {
409               return !GV.hasName() || (GVS.count(GV.getName()) == 0);
410             });
411           });
412     } else {
413       Err = L.linkInModule(std::move(M), ApplicableFlags);
414     }
415 
416     if (Err)
417       return false;
418 
419     // Internalization applies to linking of subsequent files.
420     InternalizeLinkedSymbols = Internalize;
421 
422     // All linker flags apply to linking of subsequent files.
423     ApplicableFlags = Flags;
424   }
425 
426   return true;
427 }
428 
429 int main(int argc, char **argv) {
430   InitLLVM X(argc, argv);
431   ExitOnErr.setBanner(std::string(argv[0]) + ": ");
432 
433   LLVMContext Context;
434   Context.setDiagnosticHandler(
435     std::make_unique<LLVMLinkDiagnosticHandler>(), true);
436   cl::ParseCommandLineOptions(argc, argv, "llvm linker\n");
437 
438   if (!DisableDITypeMap)
439     Context.enableDebugTypeODRUniquing();
440 
441   auto Composite = std::make_unique<Module>("llvm-link", Context);
442   Linker L(*Composite);
443 
444   unsigned Flags = Linker::Flags::None;
445   if (OnlyNeeded)
446     Flags |= Linker::Flags::LinkOnlyNeeded;
447 
448   // First add all the regular input files
449   if (!linkFiles(argv[0], Context, L, InputFilenames, Flags))
450     return 1;
451 
452   // Next the -override ones.
453   if (!linkFiles(argv[0], Context, L, OverridingInputs,
454                  Flags | Linker::Flags::OverrideFromSrc))
455     return 1;
456 
457   // Import any functions requested via -import
458   if (!importFunctions(argv[0], *Composite))
459     return 1;
460 
461   if (DumpAsm)
462     errs() << "Here's the assembly:\n" << *Composite;
463 
464   std::error_code EC;
465   ToolOutputFile Out(OutputFilename, EC,
466                      OutputAssembly ? sys::fs::OF_Text : sys::fs::OF_None);
467   if (EC) {
468     WithColor::error() << EC.message() << '\n';
469     return 1;
470   }
471 
472   if (verifyModule(*Composite, &errs())) {
473     errs() << argv[0] << ": ";
474     WithColor::error() << "linked module is broken!\n";
475     return 1;
476   }
477 
478   if (Verbose)
479     errs() << "Writing bitcode...\n";
480   if (OutputAssembly) {
481     Composite->print(Out.os(), nullptr, PreserveAssemblyUseListOrder);
482   } else if (Force || !CheckBitcodeOutputToConsole(Out.os()))
483     WriteBitcodeToFile(*Composite, Out.os(), PreserveBitcodeUseListOrder);
484 
485   // Declare success.
486   Out.keep();
487 
488   return 0;
489 }
490