xref: /llvm-project-15.0.7/lld/ELF/Driver.cpp (revision db49209c)
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(opt::InputArgList &Args) {
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 == "nodelete" ||
354          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(Args);
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, const Symbol *> SymbolNameToSymbol;
654   for (InputFile *File : ObjectFiles)
655     for (Symbol *Sym : File->getSymbols())
656       SymbolNameToSymbol[Sym->getName()] = Sym;
657 
658   auto FindSection = [&](StringRef SymName) -> InputSectionBase * {
659     const Symbol *Sym = SymbolNameToSymbol.lookup(SymName);
660     if (Sym)
661       warnUnorderableSymbol(Sym);
662     else if (Config->WarnSymbolOrdering)
663       warn(MB.getBufferIdentifier() + ": no such symbol: " + SymName);
664 
665     if (const Defined *DR = dyn_cast_or_null<Defined>(Sym))
666       return dyn_cast_or_null<InputSectionBase>(DR->Section);
667     return nullptr;
668   };
669 
670   for (StringRef L : args::getLines(MB)) {
671     SmallVector<StringRef, 3> Fields;
672     L.split(Fields, ' ');
673     uint64_t Count;
674     if (Fields.size() != 3 || !to_integer(Fields[2], Count))
675       fatal(MB.getBufferIdentifier() + ": parse error");
676 
677     if (const InputSectionBase *FromSB = FindSection(Fields[0]))
678       if (const InputSectionBase *ToSB = FindSection(Fields[1]))
679         Config->CallGraphProfile[std::make_pair(FromSB, ToSB)] += Count;
680   }
681 }
682 
683 template <class ELFT> static void readCallGraphsFromObjectFiles() {
684   auto FindSection = [&](const Symbol *Sym) -> const InputSectionBase * {
685     warnUnorderableSymbol(Sym);
686     if (const auto *SymD = dyn_cast<Defined>(Sym))
687       return dyn_cast_or_null<InputSectionBase>(SymD->Section);
688     return nullptr;
689   };
690 
691   for (auto File : ObjectFiles) {
692     auto *Obj = cast<ObjFile<ELFT>>(File);
693     for (const Elf_CGProfile_Impl<ELFT> &CGPE : Obj->CGProfile) {
694       const InputSectionBase *FromSB =
695           FindSection(&Obj->getSymbol(CGPE.cgp_from));
696       const InputSectionBase *ToSB = FindSection(&Obj->getSymbol(CGPE.cgp_to));
697       if (!FromSB || !ToSB)
698         continue;
699       Config->CallGraphProfile[{FromSB, ToSB}] += CGPE.cgp_weight;
700     }
701   }
702 }
703 
704 static bool getCompressDebugSections(opt::InputArgList &Args) {
705   StringRef S = Args.getLastArgValue(OPT_compress_debug_sections, "none");
706   if (S == "none")
707     return false;
708   if (S != "zlib")
709     error("unknown --compress-debug-sections value: " + S);
710   if (!zlib::isAvailable())
711     error("--compress-debug-sections: zlib is not available");
712   return true;
713 }
714 
715 static std::pair<StringRef, StringRef> getOldNewOptions(opt::InputArgList &Args,
716                                                         unsigned Id) {
717   auto *Arg = Args.getLastArg(Id);
718   if (!Arg)
719     return {"", ""};
720 
721   StringRef S = Arg->getValue();
722   std::pair<StringRef, StringRef> Ret = S.split(';');
723   if (Ret.second.empty())
724     error(Arg->getSpelling() + " expects 'old;new' format, but got " + S);
725   return Ret;
726 }
727 
728 // Parse the symbol ordering file and warn for any duplicate entries.
729 static std::vector<StringRef> getSymbolOrderingFile(MemoryBufferRef MB) {
730   SetVector<StringRef> Names;
731   for (StringRef S : args::getLines(MB))
732     if (!Names.insert(S) && Config->WarnSymbolOrdering)
733       warn(MB.getBufferIdentifier() + ": duplicate ordered symbol: " + S);
734 
735   return Names.takeVector();
736 }
737 
738 static void parseClangOption(StringRef Opt, const Twine &Msg) {
739   std::string Err;
740   raw_string_ostream OS(Err);
741 
742   const char *Argv[] = {Config->ProgName.data(), Opt.data()};
743   if (cl::ParseCommandLineOptions(2, Argv, "", &OS))
744     return;
745   OS.flush();
746   error(Msg + ": " + StringRef(Err).trim());
747 }
748 
749 // Initializes Config members by the command line options.
750 void LinkerDriver::readConfigs(opt::InputArgList &Args) {
751   errorHandler().Verbose = Args.hasArg(OPT_verbose);
752   errorHandler().FatalWarnings =
753       Args.hasFlag(OPT_fatal_warnings, OPT_no_fatal_warnings, false);
754   ThreadsEnabled = Args.hasFlag(OPT_threads, OPT_no_threads, true);
755 
756   Config->AllowMultipleDefinition =
757       Args.hasFlag(OPT_allow_multiple_definition,
758                    OPT_no_allow_multiple_definition, false) ||
759       hasZOption(Args, "muldefs");
760   Config->AuxiliaryList = args::getStrings(Args, OPT_auxiliary);
761   Config->Bsymbolic = Args.hasArg(OPT_Bsymbolic);
762   Config->BsymbolicFunctions = Args.hasArg(OPT_Bsymbolic_functions);
763   Config->CheckSections =
764       Args.hasFlag(OPT_check_sections, OPT_no_check_sections, true);
765   Config->Chroot = Args.getLastArgValue(OPT_chroot);
766   Config->CompressDebugSections = getCompressDebugSections(Args);
767   Config->Cref = Args.hasFlag(OPT_cref, OPT_no_cref, false);
768   Config->DefineCommon = Args.hasFlag(OPT_define_common, OPT_no_define_common,
769                                       !Args.hasArg(OPT_relocatable));
770   Config->Demangle = Args.hasFlag(OPT_demangle, OPT_no_demangle, true);
771   Config->DisableVerify = Args.hasArg(OPT_disable_verify);
772   Config->Discard = getDiscard(Args);
773   Config->DwoDir = Args.getLastArgValue(OPT_plugin_opt_dwo_dir_eq);
774   Config->DynamicLinker = getDynamicLinker(Args);
775   Config->EhFrameHdr =
776       Args.hasFlag(OPT_eh_frame_hdr, OPT_no_eh_frame_hdr, false);
777   Config->EmitRelocs = Args.hasArg(OPT_emit_relocs);
778   Config->EnableNewDtags =
779       Args.hasFlag(OPT_enable_new_dtags, OPT_disable_new_dtags, true);
780   Config->Entry = Args.getLastArgValue(OPT_entry);
781   Config->ExecuteOnly =
782       Args.hasFlag(OPT_execute_only, OPT_no_execute_only, false);
783   Config->ExportDynamic =
784       Args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false);
785   Config->FilterList = args::getStrings(Args, OPT_filter);
786   Config->Fini = Args.getLastArgValue(OPT_fini, "_fini");
787   Config->FixCortexA53Errata843419 = Args.hasArg(OPT_fix_cortex_a53_843419);
788   Config->GcSections = Args.hasFlag(OPT_gc_sections, OPT_no_gc_sections, false);
789   Config->GnuUnique = Args.hasFlag(OPT_gnu_unique, OPT_no_gnu_unique, true);
790   Config->GdbIndex = Args.hasFlag(OPT_gdb_index, OPT_no_gdb_index, false);
791   Config->ICF = getICF(Args);
792   Config->IgnoreDataAddressEquality =
793       Args.hasArg(OPT_ignore_data_address_equality);
794   Config->IgnoreFunctionAddressEquality =
795       Args.hasArg(OPT_ignore_function_address_equality);
796   Config->Init = Args.getLastArgValue(OPT_init, "_init");
797   Config->LTOAAPipeline = Args.getLastArgValue(OPT_lto_aa_pipeline);
798   Config->LTODebugPassManager = Args.hasArg(OPT_lto_debug_pass_manager);
799   Config->LTONewPassManager = Args.hasArg(OPT_lto_new_pass_manager);
800   Config->LTONewPmPasses = Args.getLastArgValue(OPT_lto_newpm_passes);
801   Config->LTOO = args::getInteger(Args, OPT_lto_O, 2);
802   Config->LTOObjPath = Args.getLastArgValue(OPT_plugin_opt_obj_path_eq);
803   Config->LTOPartitions = args::getInteger(Args, OPT_lto_partitions, 1);
804   Config->LTOSampleProfile = Args.getLastArgValue(OPT_lto_sample_profile);
805   Config->MapFile = Args.getLastArgValue(OPT_Map);
806   Config->MipsGotSize = args::getInteger(Args, OPT_mips_got_size, 0xfff0);
807   Config->MergeArmExidx =
808       Args.hasFlag(OPT_merge_exidx_entries, OPT_no_merge_exidx_entries, true);
809   Config->NoinhibitExec = Args.hasArg(OPT_noinhibit_exec);
810   Config->Nostdlib = Args.hasArg(OPT_nostdlib);
811   Config->OFormatBinary = isOutputFormatBinary(Args);
812   Config->Omagic = Args.hasFlag(OPT_omagic, OPT_no_omagic, false);
813   Config->OptRemarksFilename = Args.getLastArgValue(OPT_opt_remarks_filename);
814   Config->OptRemarksWithHotness = Args.hasArg(OPT_opt_remarks_with_hotness);
815   Config->Optimize = args::getInteger(Args, OPT_O, 1);
816   Config->OrphanHandling = getOrphanHandling(Args);
817   Config->OutputFile = Args.getLastArgValue(OPT_o);
818   Config->Pie = Args.hasFlag(OPT_pie, OPT_no_pie, false);
819   Config->PrintIcfSections =
820       Args.hasFlag(OPT_print_icf_sections, OPT_no_print_icf_sections, false);
821   Config->PrintGcSections =
822       Args.hasFlag(OPT_print_gc_sections, OPT_no_print_gc_sections, false);
823   Config->Rpath = getRpath(Args);
824   Config->Relocatable = Args.hasArg(OPT_relocatable);
825   Config->SaveTemps = Args.hasArg(OPT_save_temps);
826   Config->SearchPaths = args::getStrings(Args, OPT_library_path);
827   Config->SectionStartMap = getSectionStartMap(Args);
828   Config->Shared = Args.hasArg(OPT_shared);
829   Config->SingleRoRx = Args.hasArg(OPT_no_rosegment);
830   Config->SoName = Args.getLastArgValue(OPT_soname);
831   Config->SortSection = getSortSection(Args);
832   Config->Strip = getStrip(Args);
833   Config->Sysroot = Args.getLastArgValue(OPT_sysroot);
834   Config->Target1Rel = Args.hasFlag(OPT_target1_rel, OPT_target1_abs, false);
835   Config->Target2 = getTarget2(Args);
836   Config->ThinLTOCacheDir = Args.getLastArgValue(OPT_thinlto_cache_dir);
837   Config->ThinLTOCachePolicy = CHECK(
838       parseCachePruningPolicy(Args.getLastArgValue(OPT_thinlto_cache_policy)),
839       "--thinlto-cache-policy: invalid cache policy");
840   Config->ThinLTOEmitImportsFiles =
841       Args.hasArg(OPT_plugin_opt_thinlto_emit_imports_files);
842   Config->ThinLTOIndexOnly = Args.hasArg(OPT_plugin_opt_thinlto_index_only) ||
843                              Args.hasArg(OPT_plugin_opt_thinlto_index_only_eq);
844   Config->ThinLTOIndexOnlyArg =
845       Args.getLastArgValue(OPT_plugin_opt_thinlto_index_only_eq);
846   Config->ThinLTOJobs = args::getInteger(Args, OPT_thinlto_jobs, -1u);
847   Config->ThinLTOObjectSuffixReplace =
848       getOldNewOptions(Args, OPT_plugin_opt_thinlto_object_suffix_replace_eq);
849   Config->ThinLTOPrefixReplace =
850       getOldNewOptions(Args, OPT_plugin_opt_thinlto_prefix_replace_eq);
851   Config->Trace = Args.hasArg(OPT_trace);
852   Config->Undefined = args::getStrings(Args, OPT_undefined);
853   Config->UndefinedVersion =
854       Args.hasFlag(OPT_undefined_version, OPT_no_undefined_version, true);
855   Config->UseAndroidRelrTags = Args.hasFlag(
856       OPT_use_android_relr_tags, OPT_no_use_android_relr_tags, false);
857   Config->UnresolvedSymbols = getUnresolvedSymbolPolicy(Args);
858   Config->WarnBackrefs =
859       Args.hasFlag(OPT_warn_backrefs, OPT_no_warn_backrefs, false);
860   Config->WarnCommon = Args.hasFlag(OPT_warn_common, OPT_no_warn_common, false);
861   Config->WarnIfuncTextrel =
862       Args.hasFlag(OPT_warn_ifunc_textrel, OPT_no_warn_ifunc_textrel, false);
863   Config->WarnSymbolOrdering =
864       Args.hasFlag(OPT_warn_symbol_ordering, OPT_no_warn_symbol_ordering, true);
865   Config->ZCombreloc = getZFlag(Args, "combreloc", "nocombreloc", true);
866   Config->ZCopyreloc = getZFlag(Args, "copyreloc", "nocopyreloc", true);
867   Config->ZExecstack = getZFlag(Args, "execstack", "noexecstack", false);
868   Config->ZGlobal = hasZOption(Args, "global");
869   Config->ZHazardplt = hasZOption(Args, "hazardplt");
870   Config->ZInitfirst = hasZOption(Args, "initfirst");
871   Config->ZInterpose = hasZOption(Args, "interpose");
872   Config->ZKeepTextSectionPrefix = getZFlag(
873       Args, "keep-text-section-prefix", "nokeep-text-section-prefix", false);
874   Config->ZNodelete = hasZOption(Args, "nodelete");
875   Config->ZNodlopen = hasZOption(Args, "nodlopen");
876   Config->ZNow = getZFlag(Args, "now", "lazy", false);
877   Config->ZOrigin = hasZOption(Args, "origin");
878   Config->ZRelro = getZFlag(Args, "relro", "norelro", true);
879   Config->ZRetpolineplt = hasZOption(Args, "retpolineplt");
880   Config->ZRodynamic = hasZOption(Args, "rodynamic");
881   Config->ZStackSize = args::getZOptionValue(Args, OPT_z, "stack-size", 0);
882   Config->ZText = getZFlag(Args, "text", "notext", true);
883   Config->ZWxneeded = hasZOption(Args, "wxneeded");
884 
885   // Parse LTO options.
886   if (auto *Arg = Args.getLastArg(OPT_plugin_opt_mcpu_eq))
887     parseClangOption(Saver.save("-mcpu=" + StringRef(Arg->getValue())),
888                      Arg->getSpelling());
889 
890   for (auto *Arg : Args.filtered(OPT_plugin_opt))
891     parseClangOption(Arg->getValue(), Arg->getSpelling());
892 
893   // Parse -mllvm options.
894   for (auto *Arg : Args.filtered(OPT_mllvm))
895     parseClangOption(Arg->getValue(), Arg->getSpelling());
896 
897   if (Config->LTOO > 3)
898     error("invalid optimization level for LTO: " + Twine(Config->LTOO));
899   if (Config->LTOPartitions == 0)
900     error("--lto-partitions: number of threads must be > 0");
901   if (Config->ThinLTOJobs == 0)
902     error("--thinlto-jobs: number of threads must be > 0");
903 
904   // Parse ELF{32,64}{LE,BE} and CPU type.
905   if (auto *Arg = Args.getLastArg(OPT_m)) {
906     StringRef S = Arg->getValue();
907     std::tie(Config->EKind, Config->EMachine, Config->OSABI) =
908         parseEmulation(S);
909     Config->MipsN32Abi = (S == "elf32btsmipn32" || S == "elf32ltsmipn32");
910     Config->Emulation = S;
911   }
912 
913   // Parse -hash-style={sysv,gnu,both}.
914   if (auto *Arg = Args.getLastArg(OPT_hash_style)) {
915     StringRef S = Arg->getValue();
916     if (S == "sysv")
917       Config->SysvHash = true;
918     else if (S == "gnu")
919       Config->GnuHash = true;
920     else if (S == "both")
921       Config->SysvHash = Config->GnuHash = true;
922     else
923       error("unknown -hash-style: " + S);
924   }
925 
926   if (Args.hasArg(OPT_print_map))
927     Config->MapFile = "-";
928 
929   // --omagic is an option to create old-fashioned executables in which
930   // .text segments are writable. Today, the option is still in use to
931   // create special-purpose programs such as boot loaders. It doesn't
932   // make sense to create PT_GNU_RELRO for such executables.
933   if (Config->Omagic)
934     Config->ZRelro = false;
935 
936   std::tie(Config->BuildId, Config->BuildIdVector) = getBuildId(Args);
937 
938   std::tie(Config->AndroidPackDynRelocs, Config->RelrPackDynRelocs) =
939       getPackDynRelocs(Args);
940 
941   if (auto *Arg = Args.getLastArg(OPT_symbol_ordering_file))
942     if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
943       Config->SymbolOrderingFile = getSymbolOrderingFile(*Buffer);
944 
945   // If --retain-symbol-file is used, we'll keep only the symbols listed in
946   // the file and discard all others.
947   if (auto *Arg = Args.getLastArg(OPT_retain_symbols_file)) {
948     Config->DefaultSymbolVersion = VER_NDX_LOCAL;
949     if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
950       for (StringRef S : args::getLines(*Buffer))
951         Config->VersionScriptGlobals.push_back(
952             {S, /*IsExternCpp*/ false, /*HasWildcard*/ false});
953   }
954 
955   bool HasExportDynamic =
956       Args.hasFlag(OPT_export_dynamic, OPT_no_export_dynamic, false);
957 
958   // Parses -dynamic-list and -export-dynamic-symbol. They make some
959   // symbols private. Note that -export-dynamic takes precedence over them
960   // as it says all symbols should be exported.
961   if (!HasExportDynamic) {
962     for (auto *Arg : Args.filtered(OPT_dynamic_list))
963       if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
964         readDynamicList(*Buffer);
965 
966     for (auto *Arg : Args.filtered(OPT_export_dynamic_symbol))
967       Config->DynamicList.push_back(
968           {Arg->getValue(), /*IsExternCpp*/ false, /*HasWildcard*/ false});
969   }
970 
971   // If --export-dynamic-symbol=foo is given and symbol foo is defined in
972   // an object file in an archive file, that object file should be pulled
973   // out and linked. (It doesn't have to behave like that from technical
974   // point of view, but this is needed for compatibility with GNU.)
975   for (auto *Arg : Args.filtered(OPT_export_dynamic_symbol))
976     Config->Undefined.push_back(Arg->getValue());
977 
978   for (auto *Arg : Args.filtered(OPT_version_script))
979     if (Optional<std::string> Path = searchScript(Arg->getValue())) {
980       if (Optional<MemoryBufferRef> Buffer = readFile(*Path))
981         readVersionScript(*Buffer);
982     } else {
983       error(Twine("cannot find version script ") + Arg->getValue());
984     }
985 }
986 
987 // Some Config members do not directly correspond to any particular
988 // command line options, but computed based on other Config values.
989 // This function initialize such members. See Config.h for the details
990 // of these values.
991 static void setConfigs(opt::InputArgList &Args) {
992   ELFKind K = Config->EKind;
993   uint16_t M = Config->EMachine;
994 
995   Config->CopyRelocs = (Config->Relocatable || Config->EmitRelocs);
996   Config->Is64 = (K == ELF64LEKind || K == ELF64BEKind);
997   Config->IsLE = (K == ELF32LEKind || K == ELF64LEKind);
998   Config->Endianness = Config->IsLE ? endianness::little : endianness::big;
999   Config->IsMips64EL = (K == ELF64LEKind && M == EM_MIPS);
1000   Config->Pic = Config->Pie || Config->Shared;
1001   Config->Wordsize = Config->Is64 ? 8 : 4;
1002 
1003   // ELF defines two different ways to store relocation addends as shown below:
1004   //
1005   //  Rel:  Addends are stored to the location where relocations are applied.
1006   //  Rela: Addends are stored as part of relocation entry.
1007   //
1008   // In other words, Rela makes it easy to read addends at the price of extra
1009   // 4 or 8 byte for each relocation entry. We don't know why ELF defined two
1010   // different mechanisms in the first place, but this is how the spec is
1011   // defined.
1012   //
1013   // You cannot choose which one, Rel or Rela, you want to use. Instead each
1014   // ABI defines which one you need to use. The following expression expresses
1015   // that.
1016   Config->IsRela = M == EM_AARCH64 || M == EM_AMDGPU || M == EM_HEXAGON ||
1017                    M == EM_PPC || M == EM_PPC64 || M == EM_RISCV ||
1018                    M == EM_X86_64;
1019 
1020   // If the output uses REL relocations we must store the dynamic relocation
1021   // addends to the output sections. We also store addends for RELA relocations
1022   // if --apply-dynamic-relocs is used.
1023   // We default to not writing the addends when using RELA relocations since
1024   // any standard conforming tool can find it in r_addend.
1025   Config->WriteAddends = Args.hasFlag(OPT_apply_dynamic_relocs,
1026                                       OPT_no_apply_dynamic_relocs, false) ||
1027                          !Config->IsRela;
1028 
1029   Config->TocOptimize =
1030       Args.hasFlag(OPT_toc_optimize, OPT_no_toc_optimize, M == EM_PPC64);
1031 }
1032 
1033 // Returns a value of "-format" option.
1034 static bool isFormatBinary(StringRef S) {
1035   if (S == "binary")
1036     return true;
1037   if (S == "elf" || S == "default")
1038     return false;
1039   error("unknown -format value: " + S +
1040         " (supported formats: elf, default, binary)");
1041   return false;
1042 }
1043 
1044 void LinkerDriver::createFiles(opt::InputArgList &Args) {
1045   // For --{push,pop}-state.
1046   std::vector<std::tuple<bool, bool, bool>> Stack;
1047 
1048   // Iterate over argv to process input files and positional arguments.
1049   for (auto *Arg : Args) {
1050     switch (Arg->getOption().getUnaliasedOption().getID()) {
1051     case OPT_library:
1052       addLibrary(Arg->getValue());
1053       break;
1054     case OPT_INPUT:
1055       addFile(Arg->getValue(), /*WithLOption=*/false);
1056       break;
1057     case OPT_defsym: {
1058       StringRef From;
1059       StringRef To;
1060       std::tie(From, To) = StringRef(Arg->getValue()).split('=');
1061       if (From.empty() || To.empty())
1062         error("-defsym: syntax error: " + StringRef(Arg->getValue()));
1063       else
1064         readDefsym(From, MemoryBufferRef(To, "-defsym"));
1065       break;
1066     }
1067     case OPT_script:
1068       if (Optional<std::string> Path = searchScript(Arg->getValue())) {
1069         if (Optional<MemoryBufferRef> MB = readFile(*Path))
1070           readLinkerScript(*MB);
1071         break;
1072       }
1073       error(Twine("cannot find linker script ") + Arg->getValue());
1074       break;
1075     case OPT_as_needed:
1076       Config->AsNeeded = true;
1077       break;
1078     case OPT_format:
1079       Config->FormatBinary = isFormatBinary(Arg->getValue());
1080       break;
1081     case OPT_no_as_needed:
1082       Config->AsNeeded = false;
1083       break;
1084     case OPT_Bstatic:
1085       Config->Static = true;
1086       break;
1087     case OPT_Bdynamic:
1088       Config->Static = false;
1089       break;
1090     case OPT_whole_archive:
1091       InWholeArchive = true;
1092       break;
1093     case OPT_no_whole_archive:
1094       InWholeArchive = false;
1095       break;
1096     case OPT_just_symbols:
1097       if (Optional<MemoryBufferRef> MB = readFile(Arg->getValue())) {
1098         Files.push_back(createObjectFile(*MB));
1099         Files.back()->JustSymbols = true;
1100       }
1101       break;
1102     case OPT_start_group:
1103       if (InputFile::IsInGroup)
1104         error("nested --start-group");
1105       InputFile::IsInGroup = true;
1106       break;
1107     case OPT_end_group:
1108       if (!InputFile::IsInGroup)
1109         error("stray --end-group");
1110       InputFile::IsInGroup = false;
1111       ++InputFile::NextGroupId;
1112       break;
1113     case OPT_start_lib:
1114       if (InLib)
1115         error("nested --start-lib");
1116       if (InputFile::IsInGroup)
1117         error("may not nest --start-lib in --start-group");
1118       InLib = true;
1119       InputFile::IsInGroup = true;
1120       break;
1121     case OPT_end_lib:
1122       if (!InLib)
1123         error("stray --end-lib");
1124       InLib = false;
1125       InputFile::IsInGroup = false;
1126       ++InputFile::NextGroupId;
1127       break;
1128     case OPT_push_state:
1129       Stack.emplace_back(Config->AsNeeded, Config->Static, InWholeArchive);
1130       break;
1131     case OPT_pop_state:
1132       if (Stack.empty()) {
1133         error("unbalanced --push-state/--pop-state");
1134         break;
1135       }
1136       std::tie(Config->AsNeeded, Config->Static, InWholeArchive) = Stack.back();
1137       Stack.pop_back();
1138       break;
1139     }
1140   }
1141 
1142   if (Files.empty() && errorCount() == 0)
1143     error("no input files");
1144 }
1145 
1146 // If -m <machine_type> was not given, infer it from object files.
1147 void LinkerDriver::inferMachineType() {
1148   if (Config->EKind != ELFNoneKind)
1149     return;
1150 
1151   for (InputFile *F : Files) {
1152     if (F->EKind == ELFNoneKind)
1153       continue;
1154     Config->EKind = F->EKind;
1155     Config->EMachine = F->EMachine;
1156     Config->OSABI = F->OSABI;
1157     Config->MipsN32Abi = Config->EMachine == EM_MIPS && isMipsN32Abi(F);
1158     return;
1159   }
1160   error("target emulation unknown: -m or at least one .o file required");
1161 }
1162 
1163 // Parse -z max-page-size=<value>. The default value is defined by
1164 // each target.
1165 static uint64_t getMaxPageSize(opt::InputArgList &Args) {
1166   uint64_t Val = args::getZOptionValue(Args, OPT_z, "max-page-size",
1167                                        Target->DefaultMaxPageSize);
1168   if (!isPowerOf2_64(Val))
1169     error("max-page-size: value isn't a power of 2");
1170   return Val;
1171 }
1172 
1173 // Parses -image-base option.
1174 static Optional<uint64_t> getImageBase(opt::InputArgList &Args) {
1175   // Because we are using "Config->MaxPageSize" here, this function has to be
1176   // called after the variable is initialized.
1177   auto *Arg = Args.getLastArg(OPT_image_base);
1178   if (!Arg)
1179     return None;
1180 
1181   StringRef S = Arg->getValue();
1182   uint64_t V;
1183   if (!to_integer(S, V)) {
1184     error("-image-base: number expected, but got " + S);
1185     return 0;
1186   }
1187   if ((V % Config->MaxPageSize) != 0)
1188     warn("-image-base: address isn't multiple of page size: " + S);
1189   return V;
1190 }
1191 
1192 // Parses `--exclude-libs=lib,lib,...`.
1193 // The library names may be delimited by commas or colons.
1194 static DenseSet<StringRef> getExcludeLibs(opt::InputArgList &Args) {
1195   DenseSet<StringRef> Ret;
1196   for (auto *Arg : Args.filtered(OPT_exclude_libs)) {
1197     StringRef S = Arg->getValue();
1198     for (;;) {
1199       size_t Pos = S.find_first_of(",:");
1200       if (Pos == StringRef::npos)
1201         break;
1202       Ret.insert(S.substr(0, Pos));
1203       S = S.substr(Pos + 1);
1204     }
1205     Ret.insert(S);
1206   }
1207   return Ret;
1208 }
1209 
1210 // Handles the -exclude-libs option. If a static library file is specified
1211 // by the -exclude-libs option, all public symbols from the archive become
1212 // private unless otherwise specified by version scripts or something.
1213 // A special library name "ALL" means all archive files.
1214 //
1215 // This is not a popular option, but some programs such as bionic libc use it.
1216 template <class ELFT>
1217 static void excludeLibs(opt::InputArgList &Args) {
1218   DenseSet<StringRef> Libs = getExcludeLibs(Args);
1219   bool All = Libs.count("ALL");
1220 
1221   auto Visit = [&](InputFile *File) {
1222     if (!File->ArchiveName.empty())
1223       if (All || Libs.count(path::filename(File->ArchiveName)))
1224         for (Symbol *Sym : File->getSymbols())
1225           if (!Sym->isLocal() && Sym->File == File)
1226             Sym->VersionId = VER_NDX_LOCAL;
1227   };
1228 
1229   for (InputFile *File : ObjectFiles)
1230     Visit(File);
1231 
1232   for (BitcodeFile *File : BitcodeFiles)
1233     Visit(File);
1234 }
1235 
1236 // Force Sym to be entered in the output. Used for -u or equivalent.
1237 template <class ELFT> static void handleUndefined(StringRef Name) {
1238   Symbol *Sym = Symtab->find(Name);
1239   if (!Sym)
1240     return;
1241 
1242   // Since symbol S may not be used inside the program, LTO may
1243   // eliminate it. Mark the symbol as "used" to prevent it.
1244   Sym->IsUsedInRegularObj = true;
1245 
1246   if (Sym->isLazy())
1247     Symtab->fetchLazy<ELFT>(Sym);
1248 }
1249 
1250 template <class ELFT> static void handleLibcall(StringRef Name) {
1251   Symbol *Sym = Symtab->find(Name);
1252   if (!Sym || !Sym->isLazy())
1253     return;
1254 
1255   MemoryBufferRef MB;
1256   if (auto *LO = dyn_cast<LazyObject>(Sym))
1257     MB = LO->File->MB;
1258   else
1259     MB = cast<LazyArchive>(Sym)->getMemberBuffer();
1260 
1261   if (isBitcode(MB))
1262     Symtab->fetchLazy<ELFT>(Sym);
1263 }
1264 
1265 // If all references to a DSO happen to be weak, the DSO is not added
1266 // to DT_NEEDED. If that happens, we need to eliminate shared symbols
1267 // created from the DSO. Otherwise, they become dangling references
1268 // that point to a non-existent DSO.
1269 template <class ELFT> static void demoteSharedSymbols() {
1270   for (Symbol *Sym : Symtab->getSymbols()) {
1271     if (auto *S = dyn_cast<SharedSymbol>(Sym)) {
1272       if (!S->getFile<ELFT>().IsNeeded) {
1273         bool Used = S->Used;
1274         replaceSymbol<Undefined>(S, nullptr, S->getName(), STB_WEAK, S->StOther,
1275                                  S->Type);
1276         S->Used = Used;
1277       }
1278     }
1279   }
1280 }
1281 
1282 // The section referred to by S is considered address-significant. Set the
1283 // KeepUnique flag on the section if appropriate.
1284 static void markAddrsig(Symbol *S) {
1285   if (auto *D = dyn_cast_or_null<Defined>(S))
1286     if (D->Section)
1287       // We don't need to keep text sections unique under --icf=all even if they
1288       // are address-significant.
1289       if (Config->ICF == ICFLevel::Safe || !(D->Section->Flags & SHF_EXECINSTR))
1290         D->Section->KeepUnique = true;
1291 }
1292 
1293 // Record sections that define symbols mentioned in --keep-unique <symbol>
1294 // and symbols referred to by address-significance tables. These sections are
1295 // ineligible for ICF.
1296 template <class ELFT>
1297 static void findKeepUniqueSections(opt::InputArgList &Args) {
1298   for (auto *Arg : Args.filtered(OPT_keep_unique)) {
1299     StringRef Name = Arg->getValue();
1300     auto *D = dyn_cast_or_null<Defined>(Symtab->find(Name));
1301     if (!D || !D->Section) {
1302       warn("could not find symbol " + Name + " to keep unique");
1303       continue;
1304     }
1305     D->Section->KeepUnique = true;
1306   }
1307 
1308   // --icf=all --ignore-data-address-equality means that we can ignore
1309   // the dynsym and address-significance tables entirely.
1310   if (Config->ICF == ICFLevel::All && Config->IgnoreDataAddressEquality)
1311     return;
1312 
1313   // Symbols in the dynsym could be address-significant in other executables
1314   // or DSOs, so we conservatively mark them as address-significant.
1315   for (Symbol *S : Symtab->getSymbols())
1316     if (S->includeInDynsym())
1317       markAddrsig(S);
1318 
1319   // Visit the address-significance table in each object file and mark each
1320   // referenced symbol as address-significant.
1321   for (InputFile *F : ObjectFiles) {
1322     auto *Obj = cast<ObjFile<ELFT>>(F);
1323     ArrayRef<Symbol *> Syms = Obj->getSymbols();
1324     if (Obj->AddrsigSec) {
1325       ArrayRef<uint8_t> Contents =
1326           check(Obj->getObj().getSectionContents(Obj->AddrsigSec));
1327       const uint8_t *Cur = Contents.begin();
1328       while (Cur != Contents.end()) {
1329         unsigned Size;
1330         const char *Err;
1331         uint64_t SymIndex = decodeULEB128(Cur, &Size, Contents.end(), &Err);
1332         if (Err)
1333           fatal(toString(F) + ": could not decode addrsig section: " + Err);
1334         markAddrsig(Syms[SymIndex]);
1335         Cur += Size;
1336       }
1337     } else {
1338       // If an object file does not have an address-significance table,
1339       // conservatively mark all of its symbols as address-significant.
1340       for (Symbol *S : Syms)
1341         markAddrsig(S);
1342     }
1343   }
1344 }
1345 
1346 template <class ELFT> static Symbol *addUndefined(StringRef Name) {
1347   return Symtab->addUndefined<ELFT>(Name, STB_GLOBAL, STV_DEFAULT, 0, false,
1348                                     nullptr);
1349 }
1350 
1351 // The --wrap option is a feature to rename symbols so that you can write
1352 // wrappers for existing functions. If you pass `-wrap=foo`, all
1353 // occurrences of symbol `foo` are resolved to `wrap_foo` (so, you are
1354 // expected to write `wrap_foo` function as a wrapper). The original
1355 // symbol becomes accessible as `real_foo`, so you can call that from your
1356 // wrapper.
1357 //
1358 // This data structure is instantiated for each -wrap option.
1359 struct WrappedSymbol {
1360   Symbol *Sym;
1361   Symbol *Real;
1362   Symbol *Wrap;
1363 };
1364 
1365 // Handles -wrap option.
1366 //
1367 // This function instantiates wrapper symbols. At this point, they seem
1368 // like they are not being used at all, so we explicitly set some flags so
1369 // that LTO won't eliminate them.
1370 template <class ELFT>
1371 static std::vector<WrappedSymbol> addWrappedSymbols(opt::InputArgList &Args) {
1372   std::vector<WrappedSymbol> V;
1373   DenseSet<StringRef> Seen;
1374 
1375   for (auto *Arg : Args.filtered(OPT_wrap)) {
1376     StringRef Name = Arg->getValue();
1377     if (!Seen.insert(Name).second)
1378       continue;
1379 
1380     Symbol *Sym = Symtab->find(Name);
1381     if (!Sym)
1382       continue;
1383 
1384     Symbol *Real = addUndefined<ELFT>(Saver.save("__real_" + Name));
1385     Symbol *Wrap = addUndefined<ELFT>(Saver.save("__wrap_" + Name));
1386     V.push_back({Sym, Real, Wrap});
1387 
1388     // We want to tell LTO not to inline symbols to be overwritten
1389     // because LTO doesn't know the final symbol contents after renaming.
1390     Real->CanInline = false;
1391     Sym->CanInline = false;
1392 
1393     // Tell LTO not to eliminate these symbols.
1394     Sym->IsUsedInRegularObj = true;
1395     Wrap->IsUsedInRegularObj = true;
1396   }
1397   return V;
1398 }
1399 
1400 // Do renaming for -wrap by updating pointers to symbols.
1401 //
1402 // When this function is executed, only InputFiles and symbol table
1403 // contain pointers to symbol objects. We visit them to replace pointers,
1404 // so that wrapped symbols are swapped as instructed by the command line.
1405 template <class ELFT> static void wrapSymbols(ArrayRef<WrappedSymbol> Wrapped) {
1406   DenseMap<Symbol *, Symbol *> Map;
1407   for (const WrappedSymbol &W : Wrapped) {
1408     Map[W.Sym] = W.Wrap;
1409     Map[W.Real] = W.Sym;
1410   }
1411 
1412   // Update pointers in input files.
1413   parallelForEach(ObjectFiles, [&](InputFile *File) {
1414     std::vector<Symbol *> &Syms = File->getMutableSymbols();
1415     for (size_t I = 0, E = Syms.size(); I != E; ++I)
1416       if (Symbol *S = Map.lookup(Syms[I]))
1417         Syms[I] = S;
1418   });
1419 
1420   // Update pointers in the symbol table.
1421   for (const WrappedSymbol &W : Wrapped)
1422     Symtab->wrap(W.Sym, W.Real, W.Wrap);
1423 }
1424 
1425 static const char *LibcallRoutineNames[] = {
1426 #define HANDLE_LIBCALL(code, name) name,
1427 #include "llvm/IR/RuntimeLibcalls.def"
1428 #undef HANDLE_LIBCALL
1429 };
1430 
1431 // Do actual linking. Note that when this function is called,
1432 // all linker scripts have already been parsed.
1433 template <class ELFT> void LinkerDriver::link(opt::InputArgList &Args) {
1434   Target = getTarget();
1435   InX<ELFT>::VerSym = nullptr;
1436   InX<ELFT>::VerNeed = nullptr;
1437 
1438   Config->MaxPageSize = getMaxPageSize(Args);
1439   Config->ImageBase = getImageBase(Args);
1440 
1441   // If a -hash-style option was not given, set to a default value,
1442   // which varies depending on the target.
1443   if (!Args.hasArg(OPT_hash_style)) {
1444     if (Config->EMachine == EM_MIPS)
1445       Config->SysvHash = true;
1446     else
1447       Config->SysvHash = Config->GnuHash = true;
1448   }
1449 
1450   // Default output filename is "a.out" by the Unix tradition.
1451   if (Config->OutputFile.empty())
1452     Config->OutputFile = "a.out";
1453 
1454   // Fail early if the output file or map file is not writable. If a user has a
1455   // long link, e.g. due to a large LTO link, they do not wish to run it and
1456   // find that it failed because there was a mistake in their command-line.
1457   if (auto E = tryCreateFile(Config->OutputFile))
1458     error("cannot open output file " + Config->OutputFile + ": " + E.message());
1459   if (auto E = tryCreateFile(Config->MapFile))
1460     error("cannot open map file " + Config->MapFile + ": " + E.message());
1461   if (errorCount())
1462     return;
1463 
1464   // Use default entry point name if no name was given via the command
1465   // line nor linker scripts. For some reason, MIPS entry point name is
1466   // different from others.
1467   Config->WarnMissingEntry =
1468       (!Config->Entry.empty() || (!Config->Shared && !Config->Relocatable));
1469   if (Config->Entry.empty() && !Config->Relocatable)
1470     Config->Entry = (Config->EMachine == EM_MIPS) ? "__start" : "_start";
1471 
1472   // Handle --trace-symbol.
1473   for (auto *Arg : Args.filtered(OPT_trace_symbol))
1474     Symtab->trace(Arg->getValue());
1475 
1476   // Add all files to the symbol table. This will add almost all
1477   // symbols that we need to the symbol table.
1478   for (InputFile *F : Files)
1479     Symtab->addFile<ELFT>(F);
1480 
1481   // Now that we have every file, we can decide if we will need a
1482   // dynamic symbol table.
1483   // We need one if we were asked to export dynamic symbols or if we are
1484   // producing a shared library.
1485   // We also need one if any shared libraries are used and for pie executables
1486   // (probably because the dynamic linker needs it).
1487   Config->HasDynSymTab =
1488       !SharedFiles.empty() || Config->Pic || Config->ExportDynamic;
1489 
1490   // Some symbols (such as __ehdr_start) are defined lazily only when there
1491   // are undefined symbols for them, so we add these to trigger that logic.
1492   for (StringRef Name : Script->ReferencedSymbols)
1493     addUndefined<ELFT>(Name);
1494 
1495   // Handle the `--undefined <sym>` options.
1496   for (StringRef S : Config->Undefined)
1497     handleUndefined<ELFT>(S);
1498 
1499   // If an entry symbol is in a static archive, pull out that file now.
1500   handleUndefined<ELFT>(Config->Entry);
1501 
1502   // If any of our inputs are bitcode files, the LTO code generator may create
1503   // references to certain library functions that might not be explicit in the
1504   // bitcode file's symbol table. If any of those library functions are defined
1505   // in a bitcode file in an archive member, we need to arrange to use LTO to
1506   // compile those archive members by adding them to the link beforehand.
1507   //
1508   // However, adding all libcall symbols to the link can have undesired
1509   // consequences. For example, the libgcc implementation of
1510   // __sync_val_compare_and_swap_8 on 32-bit ARM pulls in an .init_array entry
1511   // that aborts the program if the Linux kernel does not support 64-bit
1512   // atomics, which would prevent the program from running even if it does not
1513   // use 64-bit atomics.
1514   //
1515   // Therefore, we only add libcall symbols to the link before LTO if we have
1516   // to, i.e. if the symbol's definition is in bitcode. Any other required
1517   // libcall symbols will be added to the link after LTO when we add the LTO
1518   // object file to the link.
1519   if (!BitcodeFiles.empty())
1520     for (const char *S : LibcallRoutineNames)
1521       handleLibcall<ELFT>(S);
1522 
1523   // Return if there were name resolution errors.
1524   if (errorCount())
1525     return;
1526 
1527   // Now when we read all script files, we want to finalize order of linker
1528   // script commands, which can be not yet final because of INSERT commands.
1529   Script->processInsertCommands();
1530 
1531   // We want to declare linker script's symbols early,
1532   // so that we can version them.
1533   // They also might be exported if referenced by DSOs.
1534   Script->declareSymbols();
1535 
1536   // Handle the -exclude-libs option.
1537   if (Args.hasArg(OPT_exclude_libs))
1538     excludeLibs<ELFT>(Args);
1539 
1540   // Create ElfHeader early. We need a dummy section in
1541   // addReservedSymbols to mark the created symbols as not absolute.
1542   Out::ElfHeader = make<OutputSection>("", 0, SHF_ALLOC);
1543   Out::ElfHeader->Size = sizeof(typename ELFT::Ehdr);
1544 
1545   // We need to create some reserved symbols such as _end. Create them.
1546   if (!Config->Relocatable)
1547     addReservedSymbols();
1548 
1549   // Apply version scripts.
1550   //
1551   // For a relocatable output, version scripts don't make sense, and
1552   // parsing a symbol version string (e.g. dropping "@ver1" from a symbol
1553   // name "foo@ver1") rather do harm, so we don't call this if -r is given.
1554   if (!Config->Relocatable)
1555     Symtab->scanVersionScript();
1556 
1557   // Create wrapped symbols for -wrap option.
1558   std::vector<WrappedSymbol> Wrapped = addWrappedSymbols<ELFT>(Args);
1559 
1560   // Do link-time optimization if given files are LLVM bitcode files.
1561   // This compiles bitcode files into real object files.
1562   //
1563   // With this the symbol table should be complete. After this, no new names
1564   // except a few linker-synthesized ones will be added to the symbol table.
1565   Symtab->addCombinedLTOObject<ELFT>();
1566   if (errorCount())
1567     return;
1568 
1569   // If -thinlto-index-only is given, we should create only "index
1570   // files" and not object files. Index file creation is already done
1571   // in addCombinedLTOObject, so we are done if that's the case.
1572   if (Config->ThinLTOIndexOnly)
1573     return;
1574 
1575   // Apply symbol renames for -wrap.
1576   if (!Wrapped.empty())
1577     wrapSymbols<ELFT>(Wrapped);
1578 
1579   // Now that we have a complete list of input files.
1580   // Beyond this point, no new files are added.
1581   // Aggregate all input sections into one place.
1582   for (InputFile *F : ObjectFiles)
1583     for (InputSectionBase *S : F->getSections())
1584       if (S && S != &InputSection::Discarded)
1585         InputSections.push_back(S);
1586   for (BinaryFile *F : BinaryFiles)
1587     for (InputSectionBase *S : F->getSections())
1588       InputSections.push_back(cast<InputSection>(S));
1589 
1590   // We do not want to emit debug sections if --strip-all
1591   // or -strip-debug are given.
1592   if (Config->Strip != StripPolicy::None)
1593     llvm::erase_if(InputSections, [](InputSectionBase *S) {
1594       return S->Name.startswith(".debug") || S->Name.startswith(".zdebug");
1595     });
1596 
1597   Config->EFlags = Target->calcEFlags();
1598 
1599   if (Config->EMachine == EM_ARM) {
1600     // FIXME: These warnings can be removed when lld only uses these features
1601     // when the input objects have been compiled with an architecture that
1602     // supports them.
1603     if (Config->ARMHasBlx == false)
1604       warn("lld uses blx instruction, no object with architecture supporting "
1605            "feature detected.");
1606   }
1607 
1608   // This adds a .comment section containing a version string. We have to add it
1609   // before mergeSections because the .comment section is a mergeable section.
1610   if (!Config->Relocatable)
1611     InputSections.push_back(createCommentSection());
1612 
1613   // Do size optimizations: garbage collection, merging of SHF_MERGE sections
1614   // and identical code folding.
1615   splitSections<ELFT>();
1616   markLive<ELFT>();
1617   demoteSharedSymbols<ELFT>();
1618   mergeSections();
1619   if (Config->ICF != ICFLevel::None) {
1620     findKeepUniqueSections<ELFT>(Args);
1621     doIcf<ELFT>();
1622   }
1623 
1624   // Read the callgraph now that we know what was gced or icfed
1625   if (auto *Arg = Args.getLastArg(OPT_call_graph_ordering_file))
1626     if (Optional<MemoryBufferRef> Buffer = readFile(Arg->getValue()))
1627       readCallGraph(*Buffer);
1628   readCallGraphsFromObjectFiles<ELFT>();
1629 
1630   // Write the result to the file.
1631   writeResult<ELFT>();
1632 }
1633