xref: /llvm-project-15.0.7/lld/ELF/Driver.cpp (revision 77a4adc4)
1 //===- Driver.cpp ---------------------------------------------------------===//
2 //
3 //                             The LLVM Linker
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // The driver drives the entire linking process. It is responsible for
11 // parsing command line options and doing whatever it is instructed to do.
12 //
13 // One notable thing in the LLD's driver when compared to other linkers is
14 // that the LLD's driver is agnostic on the host operating system.
15 // Other linkers usually have implicit default values (such as a dynamic
16 // linker path or library paths) for each host OS.
17 //
18 // I don't think implicit default values are useful because they are
19 // usually explicitly specified by the compiler driver. They can even
20 // be harmful when you are doing cross-linking. Therefore, in LLD, we
21 // simply trust the compiler driver to pass all required options and
22 // don't try to make effort on our side.
23 //
24 //===----------------------------------------------------------------------===//
25 
26 #include "Driver.h"
27 #include "Config.h"
28 #include "Filesystem.h"
29 #include "ICF.h"
30 #include "InputFiles.h"
31 #include "InputSection.h"
32 #include "LinkerScript.h"
33 #include "MarkLive.h"
34 #include "OutputSections.h"
35 #include "ScriptParser.h"
36 #include "SymbolTable.h"
37 #include "Symbols.h"
38 #include "SyntheticSections.h"
39 #include "Target.h"
40 #include "Writer.h"
41 #include "lld/Common/Args.h"
42 #include "lld/Common/Driver.h"
43 #include "lld/Common/ErrorHandler.h"
44 #include "lld/Common/Memory.h"
45 #include "lld/Common/Strings.h"
46 #include "lld/Common/TargetOptionsCommandFlags.h"
47 #include "lld/Common/Threads.h"
48 #include "lld/Common/Version.h"
49 #include "llvm/ADT/SetVector.h"
50 #include "llvm/ADT/StringExtras.h"
51 #include "llvm/ADT/StringSwitch.h"
52 #include "llvm/Support/CommandLine.h"
53 #include "llvm/Support/Compression.h"
54 #include "llvm/Support/LEB128.h"
55 #include "llvm/Support/Path.h"
56 #include "llvm/Support/TarWriter.h"
57 #include "llvm/Support/TargetSelect.h"
58 #include "llvm/Support/raw_ostream.h"
59 #include <cstdlib>
60 #include <utility>
61 
62 using namespace llvm;
63 using namespace llvm::ELF;
64 using namespace llvm::object;
65 using namespace llvm::sys;
66 using namespace llvm::support;
67 
68 using namespace lld;
69 using namespace lld::elf;
70 
71 Configuration *elf::Config;
72 LinkerDriver *elf::Driver;
73 
74 static void setConfigs(opt::InputArgList &Args);
75 
76 bool elf::link(ArrayRef<const char *> Args, bool CanExitEarly,
77                raw_ostream &Error) {
78   errorHandler().LogName = args::getFilenameWithoutExe(Args[0]);
79   errorHandler().ErrorLimitExceededMsg =
80       "too many errors emitted, stopping now (use "
81       "-error-limit=0 to see all errors)";
82   errorHandler().ErrorOS = &Error;
83   errorHandler().ExitEarly = CanExitEarly;
84   errorHandler().ColorDiagnostics = Error.has_colors();
85 
86   InputSections.clear();
87   OutputSections.clear();
88   BinaryFiles.clear();
89   BitcodeFiles.clear();
90   ObjectFiles.clear();
91   SharedFiles.clear();
92 
93   Config = make<Configuration>();
94   Driver = make<LinkerDriver>();
95   Script = make<LinkerScript>();
96   Symtab = make<SymbolTable>();
97 
98   Tar = nullptr;
99   memset(&In, 0, sizeof(In));
100 
101   Config->ProgName = Args[0];
102 
103   Driver->main(Args);
104 
105   // Exit immediately if we don't need to return to the caller.
106   // This saves time because the overhead of calling destructors
107   // for all globally-allocated objects is not negligible.
108   if (CanExitEarly)
109     exitLld(errorCount() ? 1 : 0);
110 
111   freeArena();
112   return !errorCount();
113 }
114 
115 // Parses a linker -m option.
116 static std::tuple<ELFKind, uint16_t, uint8_t> parseEmulation(StringRef Emul) {
117   uint8_t OSABI = 0;
118   StringRef S = Emul;
119   if (S.endswith("_fbsd")) {
120     S = S.drop_back(5);
121     OSABI = ELFOSABI_FREEBSD;
122   }
123 
124   std::pair<ELFKind, uint16_t> Ret =
125       StringSwitch<std::pair<ELFKind, uint16_t>>(S)
126           .Cases("aarch64elf", "aarch64linux", "aarch64_elf64_le_vec",
127                  {ELF64LEKind, EM_AARCH64})
128           .Cases("armelf", "armelf_linux_eabi", {ELF32LEKind, EM_ARM})
129           .Case("elf32_x86_64", {ELF32LEKind, EM_X86_64})
130           .Cases("elf32btsmip", "elf32btsmipn32", {ELF32BEKind, EM_MIPS})
131           .Cases("elf32ltsmip", "elf32ltsmipn32", {ELF32LEKind, EM_MIPS})
132           .Case("elf32lriscv", {ELF32LEKind, EM_RISCV})
133           .Case("elf32ppc", {ELF32BEKind, EM_PPC})
134           .Case("elf64btsmip", {ELF64BEKind, EM_MIPS})
135           .Case("elf64ltsmip", {ELF64LEKind, EM_MIPS})
136           .Case("elf64lriscv", {ELF64LEKind, EM_RISCV})
137           .Case("elf64ppc", {ELF64BEKind, EM_PPC64})
138           .Case("elf64lppc", {ELF64LEKind, EM_PPC64})
139           .Cases("elf_amd64", "elf_x86_64", {ELF64LEKind, EM_X86_64})
140           .Case("elf_i386", {ELF32LEKind, EM_386})
141           .Case("elf_iamcu", {ELF32LEKind, EM_IAMCU})
142           .Default({ELFNoneKind, EM_NONE});
143 
144   if (Ret.first == ELFNoneKind)
145     error("unknown emulation: " + Emul);
146   return std::make_tuple(Ret.first, Ret.second, OSABI);
147 }
148 
149 // Returns slices of MB by parsing MB as an archive file.
150 // Each slice consists of a member file in the archive.
151 std::vector<std::pair<MemoryBufferRef, uint64_t>> static getArchiveMembers(
152     MemoryBufferRef MB) {
153   std::unique_ptr<Archive> File =
154       CHECK(Archive::create(MB),
155             MB.getBufferIdentifier() + ": failed to parse archive");
156 
157   std::vector<std::pair<MemoryBufferRef, uint64_t>> V;
158   Error Err = Error::success();
159   bool AddToTar = File->isThin() && Tar;
160   for (const ErrorOr<Archive::Child> &COrErr : File->children(Err)) {
161     Archive::Child C =
162         CHECK(COrErr, MB.getBufferIdentifier() +
163                           ": could not get the child of the archive");
164     MemoryBufferRef MBRef =
165         CHECK(C.getMemoryBufferRef(),
166               MB.getBufferIdentifier() +
167                   ": could not get the buffer for a child of the archive");
168     if (AddToTar)
169       Tar->append(relativeToRoot(check(C.getFullName())), MBRef.getBuffer());
170     V.push_back(std::make_pair(MBRef, C.getChildOffset()));
171   }
172   if (Err)
173     fatal(MB.getBufferIdentifier() + ": Archive::children failed: " +
174           toString(std::move(Err)));
175 
176   // Take ownership of memory buffers created for members of thin archives.
177   for (std::unique_ptr<MemoryBuffer> &MB : File->takeThinBuffers())
178     make<std::unique_ptr<MemoryBuffer>>(std::move(MB));
179 
180   return V;
181 }
182 
183 // Opens a file and create a file object. Path has to be resolved already.
184 void LinkerDriver::addFile(StringRef Path, bool WithLOption) {
185   using namespace sys::fs;
186 
187   Optional<MemoryBufferRef> Buffer = readFile(Path);
188   if (!Buffer.hasValue())
189     return;
190   MemoryBufferRef MBRef = *Buffer;
191 
192   if (Config->FormatBinary) {
193     Files.push_back(make<BinaryFile>(MBRef));
194     return;
195   }
196 
197   switch (identify_magic(MBRef.getBuffer())) {
198   case file_magic::unknown:
199     readLinkerScript(MBRef);
200     return;
201   case file_magic::archive: {
202     // Handle -whole-archive.
203     if (InWholeArchive) {
204       for (const auto &P : getArchiveMembers(MBRef))
205         Files.push_back(createObjectFile(P.first, Path, P.second));
206       return;
207     }
208 
209     std::unique_ptr<Archive> File =
210         CHECK(Archive::create(MBRef), Path + ": failed to parse archive");
211 
212     // If an archive file has no symbol table, it is likely that a user
213     // is attempting LTO and using a default ar command that doesn't
214     // understand the LLVM bitcode file. It is a pretty common error, so
215     // we'll handle it as if it had a symbol table.
216     if (!File->isEmpty() && !File->hasSymbolTable()) {
217       for (const auto &P : getArchiveMembers(MBRef))
218         Files.push_back(make<LazyObjFile>(P.first, Path, P.second));
219       return;
220     }
221 
222     // Handle the regular case.
223     Files.push_back(make<ArchiveFile>(std::move(File)));
224     return;
225   }
226   case file_magic::elf_shared_object:
227     if (Config->Relocatable) {
228       error("attempted static link of dynamic object " + Path);
229       return;
230     }
231 
232     // DSOs usually have DT_SONAME tags in their ELF headers, and the
233     // sonames are used to identify DSOs. But if they are missing,
234     // they are identified by filenames. We don't know whether the new
235     // file has a DT_SONAME or not because we haven't parsed it yet.
236     // Here, we set the default soname for the file because we might
237     // need it later.
238     //
239     // If a file was specified by -lfoo, the directory part is not
240     // significant, as a user did not specify it. This behavior is
241     // compatible with GNU.
242     Files.push_back(
243         createSharedFile(MBRef, WithLOption ? path::filename(Path) : Path));
244     return;
245   case file_magic::bitcode:
246   case file_magic::elf_relocatable:
247     if (InLib)
248       Files.push_back(make<LazyObjFile>(MBRef, "", 0));
249     else
250       Files.push_back(createObjectFile(MBRef));
251     break;
252   default:
253     error(Path + ": unknown file type");
254   }
255 }
256 
257 // Add a given library by searching it from input search paths.
258 void LinkerDriver::addLibrary(StringRef Name) {
259   if (Optional<std::string> Path = searchLibrary(Name))
260     addFile(*Path, /*WithLOption=*/true);
261   else
262     error("unable to find library -l" + Name);
263 }
264 
265 // This function is called on startup. We need this for LTO since
266 // LTO calls LLVM functions to compile bitcode files to native code.
267 // Technically this can be delayed until we read bitcode files, but
268 // we don't bother to do lazily because the initialization is fast.
269 static void initLLVM() {
270   InitializeAllTargets();
271   InitializeAllTargetMCs();
272   InitializeAllAsmPrinters();
273   InitializeAllAsmParsers();
274 }
275 
276 // Some command line options or some combinations of them are not allowed.
277 // This function checks for such errors.
278 static void checkOptions() {
279   // The MIPS ABI as of 2016 does not support the GNU-style symbol lookup
280   // table which is a relatively new feature.
281   if (Config->EMachine == EM_MIPS && Config->GnuHash)
282     error("the .gnu.hash section is not compatible with the MIPS target");
283 
284   if (Config->FixCortexA53Errata843419 && Config->EMachine != EM_AARCH64)
285     error("--fix-cortex-a53-843419 is only supported on AArch64 targets");
286 
287   if (Config->TocOptimize && Config->EMachine != EM_PPC64)
288     error("--toc-optimize is only supported on the PowerPC64 target");
289 
290   if (Config->Pie && Config->Shared)
291     error("-shared and -pie may not be used together");
292 
293   if (!Config->Shared && !Config->FilterList.empty())
294     error("-F may not be used without -shared");
295 
296   if (!Config->Shared && !Config->AuxiliaryList.empty())
297     error("-f may not be used without -shared");
298 
299   if (!Config->Relocatable && !Config->DefineCommon)
300     error("-no-define-common not supported in non relocatable output");
301 
302   if (Config->Relocatable) {
303     if (Config->Shared)
304       error("-r and -shared may not be used together");
305     if (Config->GcSections)
306       error("-r and --gc-sections may not be used together");
307     if (Config->GdbIndex)
308       error("-r and --gdb-index may not be used together");
309     if (Config->ICF != ICFLevel::None)
310       error("-r and --icf may not be used together");
311     if (Config->Pie)
312       error("-r and -pie may not be used together");
313   }
314 
315   if (Config->ExecuteOnly) {
316     if (Config->EMachine != EM_AARCH64)
317       error("-execute-only is only supported on AArch64 targets");
318 
319     if (Config->SingleRoRx && !Script->HasSectionsCommand)
320       error("-execute-only and -no-rosegment cannot be used together");
321   }
322 }
323 
324 static const char *getReproduceOption(opt::InputArgList &Args) {
325   if (auto *Arg = Args.getLastArg(OPT_reproduce))
326     return Arg->getValue();
327   return getenv("LLD_REPRODUCE");
328 }
329 
330 static bool hasZOption(opt::InputArgList &Args, StringRef Key) {
331   for (auto *Arg : Args.filtered(OPT_z))
332     if (Key == Arg->getValue())
333       return true;
334   return false;
335 }
336 
337 static bool getZFlag(opt::InputArgList &Args, StringRef K1, StringRef K2,
338                      bool Default) {
339   for (auto *Arg : Args.filtered_reverse(OPT_z)) {
340     if (K1 == Arg->getValue())
341       return true;
342     if (K2 == Arg->getValue())
343       return false;
344   }
345   return Default;
346 }
347 
348 static bool isKnownZFlag(StringRef S) {
349   return S == "combreloc" || S == "copyreloc" || S == "defs" ||
350          S == "execstack" || S == "global" || S == "hazardplt" ||
351          S == "initfirst" || S == "interpose" ||
352          S == "keep-text-section-prefix" || S == "lazy" || S == "muldefs" ||
353          S == "nocombreloc" || S == "nocopyreloc" || S == "nodefaultlib" ||
354          S == "nodelete" || S == "nodlopen" || S == "noexecstack" ||
355          S == "nokeep-text-section-prefix" || S == "norelro" || S == "notext" ||
356          S == "now" || S == "origin" || S == "relro" || S == "retpolineplt" ||
357          S == "rodynamic" || S == "text" || S == "wxneeded" ||
358          S.startswith("max-page-size=") || S.startswith("stack-size=");
359 }
360 
361 // Report an error for an unknown -z option.
362 static void checkZOptions(opt::InputArgList &Args) {
363   for (auto *Arg : Args.filtered(OPT_z))
364     if (!isKnownZFlag(Arg->getValue()))
365       error("unknown -z value: " + StringRef(Arg->getValue()));
366 }
367 
368 void LinkerDriver::main(ArrayRef<const char *> ArgsArr) {
369   ELFOptTable Parser;
370   opt::InputArgList Args = Parser.parse(ArgsArr.slice(1));
371 
372   // Interpret this flag early because error() depends on them.
373   errorHandler().ErrorLimit = args::getInteger(Args, OPT_error_limit, 20);
374 
375   // Handle -help
376   if (Args.hasArg(OPT_help)) {
377     printHelp();
378     return;
379   }
380 
381   // Handle -v or -version.
382   //
383   // A note about "compatible with GNU linkers" message: this is a hack for
384   // scripts generated by GNU Libtool 2.4.6 (released in February 2014 and
385   // still the newest version in March 2017) or earlier to recognize LLD as
386   // a GNU compatible linker. As long as an output for the -v option
387   // contains "GNU" or "with BFD", they recognize us as GNU-compatible.
388   //
389   // This is somewhat ugly hack, but in reality, we had no choice other
390   // than doing this. Considering the very long release cycle of Libtool,
391   // it is not easy to improve it to recognize LLD as a GNU compatible
392   // linker in a timely manner. Even if we can make it, there are still a
393   // lot of "configure" scripts out there that are generated by old version
394   // of Libtool. We cannot convince every software developer to migrate to
395   // the latest version and re-generate scripts. So we have this hack.
396   if (Args.hasArg(OPT_v) || Args.hasArg(OPT_version))
397     message(getLLDVersion() + " (compatible with GNU linkers)");
398 
399   if (const char *Path = getReproduceOption(Args)) {
400     // Note that --reproduce is a debug option so you can ignore it
401     // if you are trying to understand the whole picture of the code.
402     Expected<std::unique_ptr<TarWriter>> ErrOrWriter =
403         TarWriter::create(Path, path::stem(Path));
404     if (ErrOrWriter) {
405       Tar = ErrOrWriter->get();
406       Tar->append("response.txt", createResponseFile(Args));
407       Tar->append("version.txt", getLLDVersion() + "\n");
408       make<std::unique_ptr<TarWriter>>(std::move(*ErrOrWriter));
409     } else {
410       error(Twine("--reproduce: failed to open ") + Path + ": " +
411             toString(ErrOrWriter.takeError()));
412     }
413   }
414 
415   readConfigs(Args);
416   checkZOptions(Args);
417 
418   // The behavior of -v or --version is a bit strange, but this is
419   // needed for compatibility with GNU linkers.
420   if (Args.hasArg(OPT_v) && !Args.hasArg(OPT_INPUT))
421     return;
422   if (Args.hasArg(OPT_version))
423     return;
424 
425   initLLVM();
426   createFiles(Args);
427   if (errorCount())
428     return;
429 
430   inferMachineType();
431   setConfigs(Args);
432   checkOptions();
433   if (errorCount())
434     return;
435 
436   switch (Config->EKind) {
437   case ELF32LEKind:
438     link<ELF32LE>(Args);
439     return;
440   case ELF32BEKind:
441     link<ELF32BE>(Args);
442     return;
443   case ELF64LEKind:
444     link<ELF64LE>(Args);
445     return;
446   case ELF64BEKind:
447     link<ELF64BE>(Args);
448     return;
449   default:
450     llvm_unreachable("unknown Config->EKind");
451   }
452 }
453 
454 static std::string getRpath(opt::InputArgList &Args) {
455   std::vector<StringRef> V = args::getStrings(Args, OPT_rpath);
456   return llvm::join(V.begin(), V.end(), ":");
457 }
458 
459 // Determines what we should do if there are remaining unresolved
460 // symbols after the name resolution.
461 static UnresolvedPolicy getUnresolvedSymbolPolicy(opt::InputArgList &Args) {
462   UnresolvedPolicy ErrorOrWarn = Args.hasFlag(OPT_error_unresolved_symbols,
463                                               OPT_warn_unresolved_symbols, true)
464                                      ? UnresolvedPolicy::ReportError
465                                      : UnresolvedPolicy::Warn;
466 
467   // Process the last of -unresolved-symbols, -no-undefined or -z defs.
468   for (auto *Arg : llvm::reverse(Args)) {
469     switch (Arg->getOption().getID()) {
470     case OPT_unresolved_symbols: {
471       StringRef S = Arg->getValue();
472       if (S == "ignore-all" || S == "ignore-in-object-files")
473         return UnresolvedPolicy::Ignore;
474       if (S == "ignore-in-shared-libs" || S == "report-all")
475         return ErrorOrWarn;
476       error("unknown --unresolved-symbols value: " + S);
477       continue;
478     }
479     case OPT_no_undefined:
480       return ErrorOrWarn;
481     case OPT_z:
482       if (StringRef(Arg->getValue()) == "defs")
483         return ErrorOrWarn;
484       continue;
485     }
486   }
487 
488   // -shared implies -unresolved-symbols=ignore-all because missing
489   // symbols are likely to be resolved at runtime using other DSOs.
490   if (Config->Shared)
491     return UnresolvedPolicy::Ignore;
492   return ErrorOrWarn;
493 }
494 
495 static Target2Policy getTarget2(opt::InputArgList &Args) {
496   StringRef S = Args.getLastArgValue(OPT_target2, "got-rel");
497   if (S == "rel")
498     return Target2Policy::Rel;
499   if (S == "abs")
500     return Target2Policy::Abs;
501   if (S == "got-rel")
502     return Target2Policy::GotRel;
503   error("unknown --target2 option: " + S);
504   return Target2Policy::GotRel;
505 }
506 
507 static bool isOutputFormatBinary(opt::InputArgList &Args) {
508   StringRef S = Args.getLastArgValue(OPT_oformat, "elf");
509   if (S == "binary")
510     return true;
511   if (!S.startswith("elf"))
512     error("unknown --oformat value: " + S);
513   return false;
514 }
515 
516 static DiscardPolicy getDiscard(opt::InputArgList &Args) {
517   if (Args.hasArg(OPT_relocatable))
518     return DiscardPolicy::None;
519 
520   auto *Arg =
521       Args.getLastArg(OPT_discard_all, OPT_discard_locals, OPT_discard_none);
522   if (!Arg)
523     return DiscardPolicy::Default;
524   if (Arg->getOption().getID() == OPT_discard_all)
525     return DiscardPolicy::All;
526   if (Arg->getOption().getID() == OPT_discard_locals)
527     return DiscardPolicy::Locals;
528   return DiscardPolicy::None;
529 }
530 
531 static StringRef getDynamicLinker(opt::InputArgList &Args) {
532   auto *Arg = Args.getLastArg(OPT_dynamic_linker, OPT_no_dynamic_linker);
533   if (!Arg || Arg->getOption().getID() == OPT_no_dynamic_linker)
534     return "";
535   return Arg->getValue();
536 }
537 
538 static ICFLevel getICF(opt::InputArgList &Args) {
539   auto *Arg = Args.getLastArg(OPT_icf_none, OPT_icf_safe, OPT_icf_all);
540   if (!Arg || Arg->getOption().getID() == OPT_icf_none)
541     return ICFLevel::None;
542   if (Arg->getOption().getID() == OPT_icf_safe)
543     return ICFLevel::Safe;
544   return ICFLevel::All;
545 }
546 
547 static StripPolicy getStrip(opt::InputArgList &Args) {
548   if (Args.hasArg(OPT_relocatable))
549     return StripPolicy::None;
550 
551   auto *Arg = Args.getLastArg(OPT_strip_all, OPT_strip_debug);
552   if (!Arg)
553     return StripPolicy::None;
554   if (Arg->getOption().getID() == OPT_strip_all)
555     return StripPolicy::All;
556   return StripPolicy::Debug;
557 }
558 
559 static uint64_t parseSectionAddress(StringRef S, const opt::Arg &Arg) {
560   uint64_t VA = 0;
561   if (S.startswith("0x"))
562     S = S.drop_front(2);
563   if (!to_integer(S, VA, 16))
564     error("invalid argument: " + toString(Arg));
565   return VA;
566 }
567 
568 static StringMap<uint64_t> getSectionStartMap(opt::InputArgList &Args) {
569   StringMap<uint64_t> Ret;
570   for (auto *Arg : Args.filtered(OPT_section_start)) {
571     StringRef Name;
572     StringRef Addr;
573     std::tie(Name, Addr) = StringRef(Arg->getValue()).split('=');
574     Ret[Name] = parseSectionAddress(Addr, *Arg);
575   }
576 
577   if (auto *Arg = Args.getLastArg(OPT_Ttext))
578     Ret[".text"] = parseSectionAddress(Arg->getValue(), *Arg);
579   if (auto *Arg = Args.getLastArg(OPT_Tdata))
580     Ret[".data"] = parseSectionAddress(Arg->getValue(), *Arg);
581   if (auto *Arg = Args.getLastArg(OPT_Tbss))
582     Ret[".bss"] = parseSectionAddress(Arg->getValue(), *Arg);
583   return Ret;
584 }
585 
586 static SortSectionPolicy getSortSection(opt::InputArgList &Args) {
587   StringRef S = Args.getLastArgValue(OPT_sort_section);
588   if (S == "alignment")
589     return SortSectionPolicy::Alignment;
590   if (S == "name")
591     return SortSectionPolicy::Name;
592   if (!S.empty())
593     error("unknown --sort-section rule: " + S);
594   return SortSectionPolicy::Default;
595 }
596 
597 static OrphanHandlingPolicy getOrphanHandling(opt::InputArgList &Args) {
598   StringRef S = Args.getLastArgValue(OPT_orphan_handling, "place");
599   if (S == "warn")
600     return OrphanHandlingPolicy::Warn;
601   if (S == "error")
602     return OrphanHandlingPolicy::Error;
603   if (S != "place")
604     error("unknown --orphan-handling mode: " + S);
605   return OrphanHandlingPolicy::Place;
606 }
607 
608 // Parse --build-id or --build-id=<style>. We handle "tree" as a
609 // synonym for "sha1" because all our hash functions including
610 // -build-id=sha1 are actually tree hashes for performance reasons.
611 static std::pair<BuildIdKind, std::vector<uint8_t>>
612 getBuildId(opt::InputArgList &Args) {
613   auto *Arg = Args.getLastArg(OPT_build_id, OPT_build_id_eq);
614   if (!Arg)
615     return {BuildIdKind::None, {}};
616 
617   if (Arg->getOption().getID() == OPT_build_id)
618     return {BuildIdKind::Fast, {}};
619 
620   StringRef S = Arg->getValue();
621   if (S == "fast")
622     return {BuildIdKind::Fast, {}};
623   if (S == "md5")
624     return {BuildIdKind::Md5, {}};
625   if (S == "sha1" || S == "tree")
626     return {BuildIdKind::Sha1, {}};
627   if (S == "uuid")
628     return {BuildIdKind::Uuid, {}};
629   if (S.startswith("0x"))
630     return {BuildIdKind::Hexstring, parseHex(S.substr(2))};
631 
632   if (S != "none")
633     error("unknown --build-id style: " + S);
634   return {BuildIdKind::None, {}};
635 }
636 
637 static std::pair<bool, bool> getPackDynRelocs(opt::InputArgList &Args) {
638   StringRef S = Args.getLastArgValue(OPT_pack_dyn_relocs, "none");
639   if (S == "android")
640     return {true, false};
641   if (S == "relr")
642     return {false, true};
643   if (S == "android+relr")
644     return {true, true};
645 
646   if (S != "none")
647     error("unknown -pack-dyn-relocs format: " + S);
648   return {false, false};
649 }
650 
651 static void readCallGraph(MemoryBufferRef MB) {
652   // Build a map from symbol name to section
653   DenseMap<StringRef, Symbol *> Map;
654   for (InputFile *File : ObjectFiles)
655     for (Symbol *Sym : File->getSymbols())
656       Map[Sym->getName()] = Sym;
657 
658   auto FindSection = [&](StringRef Name) -> InputSectionBase * {
659     Symbol *Sym = Map.lookup(Name);
660     if (!Sym) {
661       if (Config->WarnSymbolOrdering)
662         warn(MB.getBufferIdentifier() + ": no such symbol: " + Name);
663       return nullptr;
664     }
665     maybeWarnUnorderableSymbol(Sym);
666 
667     if (Defined *DR = dyn_cast_or_null<Defined>(Sym))
668       return dyn_cast_or_null<InputSectionBase>(DR->Section);
669     return nullptr;
670   };
671 
672   for (StringRef Line : args::getLines(MB)) {
673     SmallVector<StringRef, 3> Fields;
674     Line.split(Fields, ' ');
675     uint64_t Count;
676 
677     if (Fields.size() != 3 || !to_integer(Fields[2], Count)) {
678       error(MB.getBufferIdentifier() + ": parse error");
679       return;
680     }
681 
682     if (InputSectionBase *From = FindSection(Fields[0]))
683       if (InputSectionBase *To = FindSection(Fields[1]))
684         Config->CallGraphProfile[std::make_pair(From, To)] += Count;
685   }
686 }
687 
688 template <class ELFT> static void readCallGraphsFromObjectFiles() {
689   for (auto File : ObjectFiles) {
690     auto *Obj = cast<ObjFile<ELFT>>(File);
691 
692     for (const Elf_CGProfile_Impl<ELFT> &CGPE : Obj->CGProfile) {
693       auto *FromSym = dyn_cast<Defined>(&Obj->getSymbol(CGPE.cgp_from));
694       auto *ToSym = dyn_cast<Defined>(&Obj->getSymbol(CGPE.cgp_to));
695       if (!FromSym || !ToSym)
696         continue;
697 
698       auto *From = dyn_cast_or_null<InputSectionBase>(FromSym->Section);
699       auto *To = dyn_cast_or_null<InputSectionBase>(ToSym->Section);
700       if (From && To)
701         Config->CallGraphProfile[{From, To}] += CGPE.cgp_weight;
702     }
703   }
704 }
705 
706 static bool getCompressDebugSections(opt::InputArgList &Args) {
707   StringRef S = Args.getLastArgValue(OPT_compress_debug_sections, "none");
708   if (S == "none")
709     return false;
710   if (S != "zlib")
711     error("unknown --compress-debug-sections value: " + S);
712   if (!zlib::isAvailable())
713     error("--compress-debug-sections: zlib is not available");
714   return true;
715 }
716 
717 static std::pair<StringRef, StringRef> getOldNewOptions(opt::InputArgList &Args,
718                                                         unsigned Id) {
719   auto *Arg = Args.getLastArg(Id);
720   if (!Arg)
721     return {"", ""};
722 
723   StringRef S = Arg->getValue();
724   std::pair<StringRef, StringRef> Ret = S.split(';');
725   if (Ret.second.empty())
726     error(Arg->getSpelling() + " expects 'old;new' format, but got " + S);
727   return Ret;
728 }
729 
730 // Parse the symbol ordering file and warn for any duplicate entries.
731 static std::vector<StringRef> getSymbolOrderingFile(MemoryBufferRef MB) {
732   SetVector<StringRef> Names;
733   for (StringRef S : args::getLines(MB))
734     if (!Names.insert(S) && Config->WarnSymbolOrdering)
735       warn(MB.getBufferIdentifier() + ": duplicate ordered symbol: " + S);
736 
737   return Names.takeVector();
738 }
739 
740 static void parseClangOption(StringRef Opt, const Twine &Msg) {
741   std::string Err;
742   raw_string_ostream OS(Err);
743 
744   const char *Argv[] = {Config->ProgName.data(), Opt.data()};
745   if (cl::ParseCommandLineOptions(2, Argv, "", &OS))
746     return;
747   OS.flush();
748   error(Msg + ": " + StringRef(Err).trim());
749 }
750 
751 // Initializes Config members by the command line options.
752 void LinkerDriver::readConfigs(opt::InputArgList &Args) {
753   errorHandler().Verbose = Args.hasArg(OPT_verbose);
754   errorHandler().FatalWarnings =
755       Args.hasFlag(OPT_fatal_warnings, OPT_no_fatal_warnings, false);
756   ThreadsEnabled = Args.hasFlag(OPT_threads, OPT_no_threads, true);
757 
758   Config->AllowMultipleDefinition =
759       Args.hasFlag(OPT_allow_multiple_definition,
760                    OPT_no_allow_multiple_definition, false) ||
761       hasZOption(Args, "muldefs");
762   Config->AuxiliaryList = args::getStrings(Args, OPT_auxiliary);
763   Config->Bsymbolic = Args.hasArg(OPT_Bsymbolic);
764   Config->BsymbolicFunctions = Args.hasArg(OPT_Bsymbolic_functions);
765   Config->CheckSections =
766       Args.hasFlag(OPT_check_sections, OPT_no_check_sections, true);
767   Config->Chroot = Args.getLastArgValue(OPT_chroot);
768   Config->CompressDebugSections = getCompressDebugSections(Args);
769   Config->Cref = Args.hasFlag(OPT_cref, OPT_no_cref, false);
770   Config->DefineCommon = Args.hasFlag(OPT_define_common, OPT_no_define_common,
771                                       !Args.hasArg(OPT_relocatable));
772   Config->Demangle = Args.hasFlag(OPT_demangle, OPT_no_demangle, true);
773   Config->DisableVerify = Args.hasArg(OPT_disable_verify);
774   Config->Discard = getDiscard(Args);
775   Config->DwoDir = Args.getLastArgValue(OPT_plugin_opt_dwo_dir_eq);
776   Config->DynamicLinker = getDynamicLinker(Args);
777   Config->EhFrameHdr =
778       Args.hasFlag(OPT_eh_frame_hdr, OPT_no_eh_frame_hdr, false);
779   Config->EmitRelocs = Args.hasArg(OPT_emit_relocs);
780   Config->CallGraphProfileSort = Args.hasFlag(
781       OPT_call_graph_profile_sort, OPT_no_call_graph_profile_sort, true);
782   Config->EnableNewDtags =
783       Args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true);
784   Config->Entry = Args.getLastArgValue(OPT_entry);
785   Config->ExecuteOnly =
786       Args.hasFlag(OPT_execute_only, OPT_no_execute_only, false);
787   Config->ExportDynamic =
788       Args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false);
789   Config->FilterList = args::getStrings(Args, OPT_filter);
790   Config->Fini = Args.getLastArgValue(OPT_fini, "_fini");
791   Config->FixCortexA53Errata843419 = Args.hasArg(OPT_fix_cortex_a53_843419);
792   Config->GcSections = Args.hasFlag(OPT_gc_sections, OPT_no_gc_sections, false);
793   Config->GnuUnique = Args.hasFlag(OPT_gnu_unique, OPT_no_gnu_unique, true);
794   Config->GdbIndex = Args.hasFlag(OPT_gdb_index, OPT_no_gdb_index, false);
795   Config->ICF = getICF(Args);
796   Config->IgnoreDataAddressEquality =
797       Args.hasArg(OPT_ignore_data_address_equality);
798   Config->IgnoreFunctionAddressEquality =
799       Args.hasArg(OPT_ignore_function_address_equality);
800   Config->Init = Args.getLastArgValue(OPT_init, "_init");
801   Config->LTOAAPipeline = Args.getLastArgValue(OPT_lto_aa_pipeline);
802   Config->LTODebugPassManager = Args.hasArg(OPT_lto_debug_pass_manager);
803   Config->LTONewPassManager = Args.hasArg(OPT_lto_new_pass_manager);
804   Config->LTONewPmPasses = Args.getLastArgValue(OPT_lto_newpm_passes);
805   Config->LTOO = args::getInteger(Args, OPT_lto_O, 2);
806   Config->LTOObjPath = Args.getLastArgValue(OPT_plugin_opt_obj_path_eq);
807   Config->LTOPartitions = args::getInteger(Args, OPT_lto_partitions, 1);
808   Config->LTOSampleProfile = Args.getLastArgValue(OPT_lto_sample_profile);
809   Config->MapFile = Args.getLastArgValue(OPT_Map);
810   Config->MipsGotSize = args::getInteger(Args, OPT_mips_got_size, 0xfff0);
811   Config->MergeArmExidx =
812       Args.hasFlag(OPT_merge_exidx_entries, OPT_no_merge_exidx_entries, true);
813   Config->NoinhibitExec = Args.hasArg(OPT_noinhibit_exec);
814   Config->Nostdlib = Args.hasArg(OPT_nostdlib);
815   Config->OFormatBinary = isOutputFormatBinary(Args);
816   Config->Omagic = Args.hasFlag(OPT_omagic, OPT_no_omagic, false);
817   Config->OptRemarksFilename = Args.getLastArgValue(OPT_opt_remarks_filename);
818   Config->OptRemarksWithHotness = Args.hasArg(OPT_opt_remarks_with_hotness);
819   Config->Optimize = args::getInteger(Args, OPT_O, 1);
820   Config->OrphanHandling = getOrphanHandling(Args);
821   Config->OutputFile = Args.getLastArgValue(OPT_o);
822   Config->Pie = Args.hasFlag(OPT_pie, OPT_no_pie, false);
823   Config->PrintIcfSections =
824       Args.hasFlag(OPT_print_icf_sections, OPT_no_print_icf_sections, false);
825   Config->PrintGcSections =
826       Args.hasFlag(OPT_print_gc_sections, OPT_no_print_gc_sections, false);
827   Config->Rpath = getRpath(Args);
828   Config->Relocatable = Args.hasArg(OPT_relocatable);
829   Config->SaveTemps = Args.hasArg(OPT_save_temps);
830   Config->SearchPaths = args::getStrings(Args, OPT_library_path);
831   Config->SectionStartMap = getSectionStartMap(Args);
832   Config->Shared = Args.hasArg(OPT_shared);
833   Config->SingleRoRx = Args.hasArg(OPT_no_rosegment);
834   Config->SoName = Args.getLastArgValue(OPT_soname);
835   Config->SortSection = getSortSection(Args);
836   Config->SplitStackAdjustSize = args::getInteger(Args, OPT_split_stack_adjust_size, 16384);
837   Config->Strip = getStrip(Args);
838   Config->Sysroot = Args.getLastArgValue(OPT_sysroot);
839   Config->Target1Rel = Args.hasFlag(OPT_target1_rel, OPT_target1_abs, false);
840   Config->Target2 = getTarget2(Args);
841   Config->ThinLTOCacheDir = Args.getLastArgValue(OPT_thinlto_cache_dir);
842   Config->ThinLTOCachePolicy = CHECK(
843       parseCachePruningPolicy(Args.getLastArgValue(OPT_thinlto_cache_policy)),
844       "--thinlto-cache-policy: invalid cache policy");
845   Config->ThinLTOEmitImportsFiles =
846       Args.hasArg(OPT_plugin_opt_thinlto_emit_imports_files);
847   Config->ThinLTOIndexOnly = Args.hasArg(OPT_plugin_opt_thinlto_index_only) ||
848                              Args.hasArg(OPT_plugin_opt_thinlto_index_only_eq);
849   Config->ThinLTOIndexOnlyArg =
850       Args.getLastArgValue(OPT_plugin_opt_thinlto_index_only_eq);
851   Config->ThinLTOJobs = args::getInteger(Args, OPT_thinlto_jobs, -1u);
852   Config->ThinLTOObjectSuffixReplace =
853       getOldNewOptions(Args, OPT_plugin_opt_thinlto_object_suffix_replace_eq);
854   Config->ThinLTOPrefixReplace =
855       getOldNewOptions(Args, OPT_plugin_opt_thinlto_prefix_replace_eq);
856   Config->Trace = Args.hasArg(OPT_trace);
857   Config->Undefined = args::getStrings(Args, OPT_undefined);
858   Config->UndefinedVersion =
859       Args.hasFlag(OPT_undefined_version, OPT_no_undefined_version, true);
860   Config->UseAndroidRelrTags = Args.hasFlag(
861       OPT_use_android_relr_tags, OPT_no_use_android_relr_tags, false);
862   Config->UnresolvedSymbols = getUnresolvedSymbolPolicy(Args);
863   Config->WarnBackrefs =
864       Args.hasFlag(OPT_warn_backrefs, OPT_no_warn_backrefs, false);
865   Config->WarnCommon = Args.hasFlag(OPT_warn_common, OPT_no_warn_common, false);
866   Config->WarnIfuncTextrel =
867       Args.hasFlag(OPT_warn_ifunc_textrel, OPT_no_warn_ifunc_textrel, false);
868   Config->WarnSymbolOrdering =
869       Args.hasFlag(OPT_warn_symbol_ordering, OPT_no_warn_symbol_ordering, true);
870   Config->ZCombreloc = getZFlag(Args, "combreloc", "nocombreloc", true);
871   Config->ZCopyreloc = getZFlag(Args, "copyreloc", "nocopyreloc", true);
872   Config->ZExecstack = getZFlag(Args, "execstack", "noexecstack", false);
873   Config->ZGlobal = hasZOption(Args, "global");
874   Config->ZHazardplt = hasZOption(Args, "hazardplt");
875   Config->ZInitfirst = hasZOption(Args, "initfirst");
876   Config->ZInterpose = hasZOption(Args, "interpose");
877   Config->ZKeepTextSectionPrefix = getZFlag(
878       Args, "keep-text-section-prefix", "nokeep-text-section-prefix", false);
879   Config->ZNodefaultlib = hasZOption(Args, "nodefaultlib");
880   Config->ZNodelete = hasZOption(Args, "nodelete");
881   Config->ZNodlopen = hasZOption(Args, "nodlopen");
882   Config->ZNow = getZFlag(Args, "now", "lazy", false);
883   Config->ZOrigin = hasZOption(Args, "origin");
884   Config->ZRelro = getZFlag(Args, "relro", "norelro", true);
885   Config->ZRetpolineplt = hasZOption(Args, "retpolineplt");
886   Config->ZRodynamic = hasZOption(Args, "rodynamic");
887   Config->ZStackSize = args::getZOptionValue(Args, OPT_z, "stack-size", 0);
888   Config->ZText = getZFlag(Args, "text", "notext", true);
889   Config->ZWxneeded = hasZOption(Args, "wxneeded");
890 
891   // Parse LTO options.
892   if (auto *Arg = Args.getLastArg(OPT_plugin_opt_mcpu_eq))
893     parseClangOption(Saver.save("-mcpu=" + StringRef(Arg->getValue())),
894                      Arg->getSpelling());
895 
896   for (auto *Arg : Args.filtered(OPT_plugin_opt))
897     parseClangOption(Arg->getValue(), Arg->getSpelling());
898 
899   // Parse -mllvm options.
900   for (auto *Arg : Args.filtered(OPT_mllvm))
901     parseClangOption(Arg->getValue(), Arg->getSpelling());
902 
903   if (Config->LTOO > 3)
904     error("invalid optimization level for LTO: " + Twine(Config->LTOO));
905   if (Config->LTOPartitions == 0)
906     error("--lto-partitions: number of threads must be > 0");
907   if (Config->ThinLTOJobs == 0)
908     error("--thinlto-jobs: number of threads must be > 0");
909 
910   if (Config->SplitStackAdjustSize < 0)
911     error("--split-stack-adjust-size: size must be >= 0");
912 
913   // Parse ELF{32,64}{LE,BE} and CPU type.
914   if (auto *Arg = Args.getLastArg(OPT_m)) {
915     StringRef S = Arg->getValue();
916     std::tie(Config->EKind, Config->EMachine, Config->OSABI) =
917         parseEmulation(S);
918     Config->MipsN32Abi = (S == "elf32btsmipn32" || S == "elf32ltsmipn32");
919     Config->Emulation = S;
920   }
921 
922   // Parse -hash-style={sysv,gnu,both}.
923   if (auto *Arg = Args.getLastArg(OPT_hash_style)) {
924     StringRef S = Arg->getValue();
925     if (S == "sysv")
926       Config->SysvHash = true;
927     else if (S == "gnu")
928       Config->GnuHash = true;
929     else if (S == "both")
930       Config->SysvHash = Config->GnuHash = true;
931     else
932       error("unknown -hash-style: " + S);
933   }
934 
935   if (Args.hasArg(OPT_print_map))
936     Config->MapFile = "-";
937 
938   // --omagic is an option to create old-fashioned executables in which
939   // .text segments are writable. Today, the option is still in use to
940   // create special-purpose programs such as boot loaders. It doesn't
941   // make sense to create PT_GNU_RELRO for such executables.
942   if (Config->Omagic)
943     Config->ZRelro = false;
944 
945   std::tie(Config->BuildId, Config->BuildIdVector) = getBuildId(Args);
946 
947   std::tie(Config->AndroidPackDynRelocs, Config->RelrPackDynRelocs) =
948       getPackDynRelocs(Args);
949 
950   if (auto *Arg = Args.getLastArg(OPT_symbol_ordering_file))
951     if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
952       Config->SymbolOrderingFile = getSymbolOrderingFile(*Buffer);
953 
954   // If --retain-symbol-file is used, we'll keep only the symbols listed in
955   // the file and discard all others.
956   if (auto *Arg = Args.getLastArg(OPT_retain_symbols_file)) {
957     Config->DefaultSymbolVersion = VER_NDX_LOCAL;
958     if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
959       for (StringRef S : args::getLines(*Buffer))
960         Config->VersionScriptGlobals.push_back(
961             {S, /*IsExternCpp*/ false, /*HasWildcard*/ false});
962   }
963 
964   bool HasExportDynamic =
965       Args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false);
966 
967   // Parses -dynamic-list and -export-dynamic-symbol. They make some
968   // symbols private. Note that -export-dynamic takes precedence over them
969   // as it says all symbols should be exported.
970   if (!HasExportDynamic) {
971     for (auto *Arg : Args.filtered(OPT_dynamic_list))
972       if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
973         readDynamicList(*Buffer);
974 
975     for (auto *Arg : Args.filtered(OPT_export_dynamic_symbol))
976       Config->DynamicList.push_back(
977           {Arg->getValue(), /*IsExternCpp*/ false, /*HasWildcard*/ false});
978   }
979 
980   // If --export-dynamic-symbol=foo is given and symbol foo is defined in
981   // an object file in an archive file, that object file should be pulled
982   // out and linked. (It doesn't have to behave like that from technical
983   // point of view, but this is needed for compatibility with GNU.)
984   for (auto *Arg : Args.filtered(OPT_export_dynamic_symbol))
985     Config->Undefined.push_back(Arg->getValue());
986 
987   for (auto *Arg : Args.filtered(OPT_version_script))
988     if (Optional<std::string> Path = searchScript(Arg->getValue())) {
989       if (Optional<MemoryBufferRef> Buffer = readFile(*Path))
990         readVersionScript(*Buffer);
991     } else {
992       error(Twine("cannot find version script ") + Arg->getValue());
993     }
994 }
995 
996 // Some Config members do not directly correspond to any particular
997 // command line options, but computed based on other Config values.
998 // This function initialize such members. See Config.h for the details
999 // of these values.
1000 static void setConfigs(opt::InputArgList &Args) {
1001   ELFKind K = Config->EKind;
1002   uint16_t M = Config->EMachine;
1003 
1004   Config->CopyRelocs = (Config->Relocatable || Config->EmitRelocs);
1005   Config->Is64 = (K == ELF64LEKind || K == ELF64BEKind);
1006   Config->IsLE = (K == ELF32LEKind || K == ELF64LEKind);
1007   Config->Endianness = Config->IsLE ? endianness::little : endianness::big;
1008   Config->IsMips64EL = (K == ELF64LEKind && M == EM_MIPS);
1009   Config->Pic = Config->Pie || Config->Shared;
1010   Config->Wordsize = Config->Is64 ? 8 : 4;
1011 
1012   // ELF defines two different ways to store relocation addends as shown below:
1013   //
1014   //  Rel:  Addends are stored to the location where relocations are applied.
1015   //  Rela: Addends are stored as part of relocation entry.
1016   //
1017   // In other words, Rela makes it easy to read addends at the price of extra
1018   // 4 or 8 byte for each relocation entry. We don't know why ELF defined two
1019   // different mechanisms in the first place, but this is how the spec is
1020   // defined.
1021   //
1022   // You cannot choose which one, Rel or Rela, you want to use. Instead each
1023   // ABI defines which one you need to use. The following expression expresses
1024   // that.
1025   Config->IsRela = M == EM_AARCH64 || M == EM_AMDGPU || M == EM_HEXAGON ||
1026                    M == EM_PPC || M == EM_PPC64 || M == EM_RISCV ||
1027                    M == EM_X86_64;
1028 
1029   // If the output uses REL relocations we must store the dynamic relocation
1030   // addends to the output sections. We also store addends for RELA relocations
1031   // if --apply-dynamic-relocs is used.
1032   // We default to not writing the addends when using RELA relocations since
1033   // any standard conforming tool can find it in r_addend.
1034   Config->WriteAddends = Args.hasFlag(OPT_apply_dynamic_relocs,
1035                                       OPT_no_apply_dynamic_relocs, false) ||
1036                          !Config->IsRela;
1037 
1038   Config->TocOptimize =
1039       Args.hasFlag(OPT_toc_optimize, OPT_no_toc_optimize, M == EM_PPC64);
1040 }
1041 
1042 // Returns a value of "-format" option.
1043 static bool isFormatBinary(StringRef S) {
1044   if (S == "binary")
1045     return true;
1046   if (S == "elf" || S == "default")
1047     return false;
1048   error("unknown -format value: " + S +
1049         " (supported formats: elf, default, binary)");
1050   return false;
1051 }
1052 
1053 void LinkerDriver::createFiles(opt::InputArgList &Args) {
1054   // For --{push,pop}-state.
1055   std::vector<std::tuple<bool, bool, bool>> Stack;
1056 
1057   // Iterate over argv to process input files and positional arguments.
1058   for (auto *Arg : Args) {
1059     switch (Arg->getOption().getUnaliasedOption().getID()) {
1060     case OPT_library:
1061       addLibrary(Arg->getValue());
1062       break;
1063     case OPT_INPUT:
1064       addFile(Arg->getValue(), /*WithLOption=*/false);
1065       break;
1066     case OPT_defsym: {
1067       StringRef From;
1068       StringRef To;
1069       std::tie(From, To) = StringRef(Arg->getValue()).split('=');
1070       if (From.empty() || To.empty())
1071         error("-defsym: syntax error: " + StringRef(Arg->getValue()));
1072       else
1073         readDefsym(From, MemoryBufferRef(To, "-defsym"));
1074       break;
1075     }
1076     case OPT_script:
1077       if (Optional<std::string> Path = searchScript(Arg->getValue())) {
1078         if (Optional<MemoryBufferRef> MB = readFile(*Path))
1079           readLinkerScript(*MB);
1080         break;
1081       }
1082       error(Twine("cannot find linker script ") + Arg->getValue());
1083       break;
1084     case OPT_as_needed:
1085       Config->AsNeeded = true;
1086       break;
1087     case OPT_format:
1088       Config->FormatBinary = isFormatBinary(Arg->getValue());
1089       break;
1090     case OPT_no_as_needed:
1091       Config->AsNeeded = false;
1092       break;
1093     case OPT_Bstatic:
1094       Config->Static = true;
1095       break;
1096     case OPT_Bdynamic:
1097       Config->Static = false;
1098       break;
1099     case OPT_whole_archive:
1100       InWholeArchive = true;
1101       break;
1102     case OPT_no_whole_archive:
1103       InWholeArchive = false;
1104       break;
1105     case OPT_just_symbols:
1106       if (Optional<MemoryBufferRef> MB = readFile(Arg->getValue())) {
1107         Files.push_back(createObjectFile(*MB));
1108         Files.back()->JustSymbols = true;
1109       }
1110       break;
1111     case OPT_start_group:
1112       if (InputFile::IsInGroup)
1113         error("nested --start-group");
1114       InputFile::IsInGroup = true;
1115       break;
1116     case OPT_end_group:
1117       if (!InputFile::IsInGroup)
1118         error("stray --end-group");
1119       InputFile::IsInGroup = false;
1120       ++InputFile::NextGroupId;
1121       break;
1122     case OPT_start_lib:
1123       if (InLib)
1124         error("nested --start-lib");
1125       if (InputFile::IsInGroup)
1126         error("may not nest --start-lib in --start-group");
1127       InLib = true;
1128       InputFile::IsInGroup = true;
1129       break;
1130     case OPT_end_lib:
1131       if (!InLib)
1132         error("stray --end-lib");
1133       InLib = false;
1134       InputFile::IsInGroup = false;
1135       ++InputFile::NextGroupId;
1136       break;
1137     case OPT_push_state:
1138       Stack.emplace_back(Config->AsNeeded, Config->Static, InWholeArchive);
1139       break;
1140     case OPT_pop_state:
1141       if (Stack.empty()) {
1142         error("unbalanced --push-state/--pop-state");
1143         break;
1144       }
1145       std::tie(Config->AsNeeded, Config->Static, InWholeArchive) = Stack.back();
1146       Stack.pop_back();
1147       break;
1148     }
1149   }
1150 
1151   if (Files.empty() && errorCount() == 0)
1152     error("no input files");
1153 }
1154 
1155 // If -m <machine_type> was not given, infer it from object files.
1156 void LinkerDriver::inferMachineType() {
1157   if (Config->EKind != ELFNoneKind)
1158     return;
1159 
1160   for (InputFile *F : Files) {
1161     if (F->EKind == ELFNoneKind)
1162       continue;
1163     Config->EKind = F->EKind;
1164     Config->EMachine = F->EMachine;
1165     Config->OSABI = F->OSABI;
1166     Config->MipsN32Abi = Config->EMachine == EM_MIPS && isMipsN32Abi(F);
1167     return;
1168   }
1169   error("target emulation unknown: -m or at least one .o file required");
1170 }
1171 
1172 // Parse -z max-page-size=<value>. The default value is defined by
1173 // each target.
1174 static uint64_t getMaxPageSize(opt::InputArgList &Args) {
1175   uint64_t Val = args::getZOptionValue(Args, OPT_z, "max-page-size",
1176                                        Target->DefaultMaxPageSize);
1177   if (!isPowerOf2_64(Val))
1178     error("max-page-size: value isn't a power of 2");
1179   return Val;
1180 }
1181 
1182 // Parses -image-base option.
1183 static Optional<uint64_t> getImageBase(opt::InputArgList &Args) {
1184   // Because we are using "Config->MaxPageSize" here, this function has to be
1185   // called after the variable is initialized.
1186   auto *Arg = Args.getLastArg(OPT_image_base);
1187   if (!Arg)
1188     return None;
1189 
1190   StringRef S = Arg->getValue();
1191   uint64_t V;
1192   if (!to_integer(S, V)) {
1193     error("-image-base: number expected, but got " + S);
1194     return 0;
1195   }
1196   if ((V % Config->MaxPageSize) != 0)
1197     warn("-image-base: address isn't multiple of page size: " + S);
1198   return V;
1199 }
1200 
1201 // Parses `--exclude-libs=lib,lib,...`.
1202 // The library names may be delimited by commas or colons.
1203 static DenseSet<StringRef> getExcludeLibs(opt::InputArgList &Args) {
1204   DenseSet<StringRef> Ret;
1205   for (auto *Arg : Args.filtered(OPT_exclude_libs)) {
1206     StringRef S = Arg->getValue();
1207     for (;;) {
1208       size_t Pos = S.find_first_of(",:");
1209       if (Pos == StringRef::npos)
1210         break;
1211       Ret.insert(S.substr(0, Pos));
1212       S = S.substr(Pos + 1);
1213     }
1214     Ret.insert(S);
1215   }
1216   return Ret;
1217 }
1218 
1219 // Handles the -exclude-libs option. If a static library file is specified
1220 // by the -exclude-libs option, all public symbols from the archive become
1221 // private unless otherwise specified by version scripts or something.
1222 // A special library name "ALL" means all archive files.
1223 //
1224 // This is not a popular option, but some programs such as bionic libc use it.
1225 template <class ELFT>
1226 static void excludeLibs(opt::InputArgList &Args) {
1227   DenseSet<StringRef> Libs = getExcludeLibs(Args);
1228   bool All = Libs.count("ALL");
1229 
1230   auto Visit = [&](InputFile *File) {
1231     if (!File->ArchiveName.empty())
1232       if (All || Libs.count(path::filename(File->ArchiveName)))
1233         for (Symbol *Sym : File->getSymbols())
1234           if (!Sym->isLocal() && Sym->File == File)
1235             Sym->VersionId = VER_NDX_LOCAL;
1236   };
1237 
1238   for (InputFile *File : ObjectFiles)
1239     Visit(File);
1240 
1241   for (BitcodeFile *File : BitcodeFiles)
1242     Visit(File);
1243 }
1244 
1245 // Force Sym to be entered in the output. Used for -u or equivalent.
1246 template <class ELFT> static void handleUndefined(StringRef Name) {
1247   Symbol *Sym = Symtab->find(Name);
1248   if (!Sym)
1249     return;
1250 
1251   // Since symbol S may not be used inside the program, LTO may
1252   // eliminate it. Mark the symbol as "used" to prevent it.
1253   Sym->IsUsedInRegularObj = true;
1254 
1255   if (Sym->isLazy())
1256     Symtab->fetchLazy<ELFT>(Sym);
1257 }
1258 
1259 template <class ELFT> static void handleLibcall(StringRef Name) {
1260   Symbol *Sym = Symtab->find(Name);
1261   if (!Sym || !Sym->isLazy())
1262     return;
1263 
1264   MemoryBufferRef MB;
1265   if (auto *LO = dyn_cast<LazyObject>(Sym))
1266     MB = LO->File->MB;
1267   else
1268     MB = cast<LazyArchive>(Sym)->getMemberBuffer();
1269 
1270   if (isBitcode(MB))
1271     Symtab->fetchLazy<ELFT>(Sym);
1272 }
1273 
1274 // If all references to a DSO happen to be weak, the DSO is not added
1275 // to DT_NEEDED. If that happens, we need to eliminate shared symbols
1276 // created from the DSO. Otherwise, they become dangling references
1277 // that point to a non-existent DSO.
1278 template <class ELFT> static void demoteSharedSymbols() {
1279   for (Symbol *Sym : Symtab->getSymbols()) {
1280     if (auto *S = dyn_cast<SharedSymbol>(Sym)) {
1281       if (!S->getFile<ELFT>().IsNeeded) {
1282         bool Used = S->Used;
1283         replaceSymbol<Undefined>(S, nullptr, S->getName(), STB_WEAK, S->StOther,
1284                                  S->Type);
1285         S->Used = Used;
1286       }
1287     }
1288   }
1289 }
1290 
1291 // The section referred to by S is considered address-significant. Set the
1292 // KeepUnique flag on the section if appropriate.
1293 static void markAddrsig(Symbol *S) {
1294   if (auto *D = dyn_cast_or_null<Defined>(S))
1295     if (D->Section)
1296       // We don't need to keep text sections unique under --icf=all even if they
1297       // are address-significant.
1298       if (Config->ICF == ICFLevel::Safe || !(D->Section->Flags & SHF_EXECINSTR))
1299         D->Section->KeepUnique = true;
1300 }
1301 
1302 // Record sections that define symbols mentioned in --keep-unique <symbol>
1303 // and symbols referred to by address-significance tables. These sections are
1304 // ineligible for ICF.
1305 template <class ELFT>
1306 static void findKeepUniqueSections(opt::InputArgList &Args) {
1307   for (auto *Arg : Args.filtered(OPT_keep_unique)) {
1308     StringRef Name = Arg->getValue();
1309     auto *D = dyn_cast_or_null<Defined>(Symtab->find(Name));
1310     if (!D || !D->Section) {
1311       warn("could not find symbol " + Name + " to keep unique");
1312       continue;
1313     }
1314     D->Section->KeepUnique = true;
1315   }
1316 
1317   // --icf=all --ignore-data-address-equality means that we can ignore
1318   // the dynsym and address-significance tables entirely.
1319   if (Config->ICF == ICFLevel::All && Config->IgnoreDataAddressEquality)
1320     return;
1321 
1322   // Symbols in the dynsym could be address-significant in other executables
1323   // or DSOs, so we conservatively mark them as address-significant.
1324   for (Symbol *S : Symtab->getSymbols())
1325     if (S->includeInDynsym())
1326       markAddrsig(S);
1327 
1328   // Visit the address-significance table in each object file and mark each
1329   // referenced symbol as address-significant.
1330   for (InputFile *F : ObjectFiles) {
1331     auto *Obj = cast<ObjFile<ELFT>>(F);
1332     ArrayRef<Symbol *> Syms = Obj->getSymbols();
1333     if (Obj->AddrsigSec) {
1334       ArrayRef<uint8_t> Contents =
1335           check(Obj->getObj().getSectionContents(Obj->AddrsigSec));
1336       const uint8_t *Cur = Contents.begin();
1337       while (Cur != Contents.end()) {
1338         unsigned Size;
1339         const char *Err;
1340         uint64_t SymIndex = decodeULEB128(Cur, &Size, Contents.end(), &Err);
1341         if (Err)
1342           fatal(toString(F) + ": could not decode addrsig section: " + Err);
1343         markAddrsig(Syms[SymIndex]);
1344         Cur += Size;
1345       }
1346     } else {
1347       // If an object file does not have an address-significance table,
1348       // conservatively mark all of its symbols as address-significant.
1349       for (Symbol *S : Syms)
1350         markAddrsig(S);
1351     }
1352   }
1353 }
1354 
1355 template <class ELFT> static Symbol *addUndefined(StringRef Name) {
1356   return Symtab->addUndefined<ELFT>(Name, STB_GLOBAL, STV_DEFAULT, 0, false,
1357                                     nullptr);
1358 }
1359 
1360 // The --wrap option is a feature to rename symbols so that you can write
1361 // wrappers for existing functions. If you pass `-wrap=foo`, all
1362 // occurrences of symbol `foo` are resolved to `wrap_foo` (so, you are
1363 // expected to write `wrap_foo` function as a wrapper). The original
1364 // symbol becomes accessible as `real_foo`, so you can call that from your
1365 // wrapper.
1366 //
1367 // This data structure is instantiated for each -wrap option.
1368 struct WrappedSymbol {
1369   Symbol *Sym;
1370   Symbol *Real;
1371   Symbol *Wrap;
1372 };
1373 
1374 // Handles -wrap option.
1375 //
1376 // This function instantiates wrapper symbols. At this point, they seem
1377 // like they are not being used at all, so we explicitly set some flags so
1378 // that LTO won't eliminate them.
1379 template <class ELFT>
1380 static std::vector<WrappedSymbol> addWrappedSymbols(opt::InputArgList &Args) {
1381   std::vector<WrappedSymbol> V;
1382   DenseSet<StringRef> Seen;
1383 
1384   for (auto *Arg : Args.filtered(OPT_wrap)) {
1385     StringRef Name = Arg->getValue();
1386     if (!Seen.insert(Name).second)
1387       continue;
1388 
1389     Symbol *Sym = Symtab->find(Name);
1390     if (!Sym)
1391       continue;
1392 
1393     Symbol *Real = addUndefined<ELFT>(Saver.save("__real_" + Name));
1394     Symbol *Wrap = addUndefined<ELFT>(Saver.save("__wrap_" + Name));
1395     V.push_back({Sym, Real, Wrap});
1396 
1397     // We want to tell LTO not to inline symbols to be overwritten
1398     // because LTO doesn't know the final symbol contents after renaming.
1399     Real->CanInline = false;
1400     Sym->CanInline = false;
1401 
1402     // Tell LTO not to eliminate these symbols.
1403     Sym->IsUsedInRegularObj = true;
1404     Wrap->IsUsedInRegularObj = true;
1405   }
1406   return V;
1407 }
1408 
1409 // Do renaming for -wrap by updating pointers to symbols.
1410 //
1411 // When this function is executed, only InputFiles and symbol table
1412 // contain pointers to symbol objects. We visit them to replace pointers,
1413 // so that wrapped symbols are swapped as instructed by the command line.
1414 template <class ELFT> static void wrapSymbols(ArrayRef<WrappedSymbol> Wrapped) {
1415   DenseMap<Symbol *, Symbol *> Map;
1416   for (const WrappedSymbol &W : Wrapped) {
1417     Map[W.Sym] = W.Wrap;
1418     Map[W.Real] = W.Sym;
1419   }
1420 
1421   // Update pointers in input files.
1422   parallelForEach(ObjectFiles, [&](InputFile *File) {
1423     std::vector<Symbol *> &Syms = File->getMutableSymbols();
1424     for (size_t I = 0, E = Syms.size(); I != E; ++I)
1425       if (Symbol *S = Map.lookup(Syms[I]))
1426         Syms[I] = S;
1427   });
1428 
1429   // Update pointers in the symbol table.
1430   for (const WrappedSymbol &W : Wrapped)
1431     Symtab->wrap(W.Sym, W.Real, W.Wrap);
1432 }
1433 
1434 static const char *LibcallRoutineNames[] = {
1435 #define HANDLE_LIBCALL(code, name) name,
1436 #include "llvm/IR/RuntimeLibcalls.def"
1437 #undef HANDLE_LIBCALL
1438 };
1439 
1440 // Do actual linking. Note that when this function is called,
1441 // all linker scripts have already been parsed.
1442 template <class ELFT> void LinkerDriver::link(opt::InputArgList &Args) {
1443   Target = getTarget();
1444   InX<ELFT>::VerSym = nullptr;
1445   InX<ELFT>::VerNeed = nullptr;
1446 
1447   Config->MaxPageSize = getMaxPageSize(Args);
1448   Config->ImageBase = getImageBase(Args);
1449 
1450   // If a -hash-style option was not given, set to a default value,
1451   // which varies depending on the target.
1452   if (!Args.hasArg(OPT_hash_style)) {
1453     if (Config->EMachine == EM_MIPS)
1454       Config->SysvHash = true;
1455     else
1456       Config->SysvHash = Config->GnuHash = true;
1457   }
1458 
1459   // Default output filename is "a.out" by the Unix tradition.
1460   if (Config->OutputFile.empty())
1461     Config->OutputFile = "a.out";
1462 
1463   // Fail early if the output file or map file is not writable. If a user has a
1464   // long link, e.g. due to a large LTO link, they do not wish to run it and
1465   // find that it failed because there was a mistake in their command-line.
1466   if (auto E = tryCreateFile(Config->OutputFile))
1467     error("cannot open output file " + Config->OutputFile + ": " + E.message());
1468   if (auto E = tryCreateFile(Config->MapFile))
1469     error("cannot open map file " + Config->MapFile + ": " + E.message());
1470   if (errorCount())
1471     return;
1472 
1473   // Use default entry point name if no name was given via the command
1474   // line nor linker scripts. For some reason, MIPS entry point name is
1475   // different from others.
1476   Config->WarnMissingEntry =
1477       (!Config->Entry.empty() || (!Config->Shared && !Config->Relocatable));
1478   if (Config->Entry.empty() && !Config->Relocatable)
1479     Config->Entry = (Config->EMachine == EM_MIPS) ? "__start" : "_start";
1480 
1481   // Handle --trace-symbol.
1482   for (auto *Arg : Args.filtered(OPT_trace_symbol))
1483     Symtab->trace(Arg->getValue());
1484 
1485   // Add all files to the symbol table. This will add almost all
1486   // symbols that we need to the symbol table.
1487   for (InputFile *F : Files)
1488     Symtab->addFile<ELFT>(F);
1489 
1490   // Now that we have every file, we can decide if we will need a
1491   // dynamic symbol table.
1492   // We need one if we were asked to export dynamic symbols or if we are
1493   // producing a shared library.
1494   // We also need one if any shared libraries are used and for pie executables
1495   // (probably because the dynamic linker needs it).
1496   Config->HasDynSymTab =
1497       !SharedFiles.empty() || Config->Pic || Config->ExportDynamic;
1498 
1499   // Some symbols (such as __ehdr_start) are defined lazily only when there
1500   // are undefined symbols for them, so we add these to trigger that logic.
1501   for (StringRef Name : Script->ReferencedSymbols)
1502     addUndefined<ELFT>(Name);
1503 
1504   // Handle the `--undefined <sym>` options.
1505   for (StringRef S : Config->Undefined)
1506     handleUndefined<ELFT>(S);
1507 
1508   // If an entry symbol is in a static archive, pull out that file now.
1509   handleUndefined<ELFT>(Config->Entry);
1510 
1511   // If any of our inputs are bitcode files, the LTO code generator may create
1512   // references to certain library functions that might not be explicit in the
1513   // bitcode file's symbol table. If any of those library functions are defined
1514   // in a bitcode file in an archive member, we need to arrange to use LTO to
1515   // compile those archive members by adding them to the link beforehand.
1516   //
1517   // However, adding all libcall symbols to the link can have undesired
1518   // consequences. For example, the libgcc implementation of
1519   // __sync_val_compare_and_swap_8 on 32-bit ARM pulls in an .init_array entry
1520   // that aborts the program if the Linux kernel does not support 64-bit
1521   // atomics, which would prevent the program from running even if it does not
1522   // use 64-bit atomics.
1523   //
1524   // Therefore, we only add libcall symbols to the link before LTO if we have
1525   // to, i.e. if the symbol's definition is in bitcode. Any other required
1526   // libcall symbols will be added to the link after LTO when we add the LTO
1527   // object file to the link.
1528   if (!BitcodeFiles.empty())
1529     for (const char *S : LibcallRoutineNames)
1530       handleLibcall<ELFT>(S);
1531 
1532   // Return if there were name resolution errors.
1533   if (errorCount())
1534     return;
1535 
1536   // Now when we read all script files, we want to finalize order of linker
1537   // script commands, which can be not yet final because of INSERT commands.
1538   Script->processInsertCommands();
1539 
1540   // We want to declare linker script's symbols early,
1541   // so that we can version them.
1542   // They also might be exported if referenced by DSOs.
1543   Script->declareSymbols();
1544 
1545   // Handle the -exclude-libs option.
1546   if (Args.hasArg(OPT_exclude_libs))
1547     excludeLibs<ELFT>(Args);
1548 
1549   // Create ElfHeader early. We need a dummy section in
1550   // addReservedSymbols to mark the created symbols as not absolute.
1551   Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC);
1552   Out::ElfHeader->Size = sizeof(typename ELFT::Ehdr);
1553 
1554   // We need to create some reserved symbols such as _end. Create them.
1555   if (!Config->Relocatable)
1556     addReservedSymbols();
1557 
1558   // Apply version scripts.
1559   //
1560   // For a relocatable output, version scripts don't make sense, and
1561   // parsing a symbol version string (e.g. dropping "@ver1" from a symbol
1562   // name "foo@ver1") rather do harm, so we don't call this if -r is given.
1563   if (!Config->Relocatable)
1564     Symtab->scanVersionScript();
1565 
1566   // Create wrapped symbols for -wrap option.
1567   std::vector<WrappedSymbol> Wrapped = addWrappedSymbols<ELFT>(Args);
1568 
1569   // Do link-time optimization if given files are LLVM bitcode files.
1570   // This compiles bitcode files into real object files.
1571   //
1572   // With this the symbol table should be complete. After this, no new names
1573   // except a few linker-synthesized ones will be added to the symbol table.
1574   Symtab->addCombinedLTOObject<ELFT>();
1575   if (errorCount())
1576     return;
1577 
1578   // If -thinlto-index-only is given, we should create only "index
1579   // files" and not object files. Index file creation is already done
1580   // in addCombinedLTOObject, so we are done if that's the case.
1581   if (Config->ThinLTOIndexOnly)
1582     return;
1583 
1584   // Apply symbol renames for -wrap.
1585   if (!Wrapped.empty())
1586     wrapSymbols<ELFT>(Wrapped);
1587 
1588   // Now that we have a complete list of input files.
1589   // Beyond this point, no new files are added.
1590   // Aggregate all input sections into one place.
1591   for (InputFile *F : ObjectFiles)
1592     for (InputSectionBase *S : F->getSections())
1593       if (S && S != &InputSection::Discarded)
1594         InputSections.push_back(S);
1595   for (BinaryFile *F : BinaryFiles)
1596     for (InputSectionBase *S : F->getSections())
1597       InputSections.push_back(cast<InputSection>(S));
1598 
1599   // We do not want to emit debug sections if --strip-all
1600   // or -strip-debug are given.
1601   if (Config->Strip != StripPolicy::None)
1602     llvm::erase_if(InputSections, [](InputSectionBase *S) {
1603       return S->Name.startswith(".debug") || S->Name.startswith(".zdebug");
1604     });
1605 
1606   Config->EFlags = Target->calcEFlags();
1607 
1608   if (Config->EMachine == EM_ARM) {
1609     // FIXME: These warnings can be removed when lld only uses these features
1610     // when the input objects have been compiled with an architecture that
1611     // supports them.
1612     if (Config->ARMHasBlx == false)
1613       warn("lld uses blx instruction, no object with architecture supporting "
1614            "feature detected");
1615   }
1616 
1617   // This adds a .comment section containing a version string. We have to add it
1618   // before mergeSections because the .comment section is a mergeable section.
1619   if (!Config->Relocatable)
1620     InputSections.push_back(createCommentSection());
1621 
1622   // Do size optimizations: garbage collection, merging of SHF_MERGE sections
1623   // and identical code folding.
1624   splitSections<ELFT>();
1625   markLive<ELFT>();
1626   demoteSharedSymbols<ELFT>();
1627   mergeSections();
1628   if (Config->ICF != ICFLevel::None) {
1629     findKeepUniqueSections<ELFT>(Args);
1630     doIcf<ELFT>();
1631   }
1632 
1633   // Read the callgraph now that we know what was gced or icfed
1634   if (Config->CallGraphProfileSort) {
1635     if (auto *Arg = Args.getLastArg(OPT_call_graph_ordering_file))
1636       if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
1637         readCallGraph(*Buffer);
1638     readCallGraphsFromObjectFiles<ELFT>();
1639   }
1640 
1641   // Write the result to the file.
1642   writeResult<ELFT>();
1643 }
1644