1 //===- IRSymtab.cpp - implementation of IR symbol tables ------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "llvm/Object/IRSymtab.h"
10 #include "llvm/ADT/ArrayRef.h"
11 #include "llvm/ADT/DenseMap.h"
12 #include "llvm/ADT/SmallPtrSet.h"
13 #include "llvm/ADT/SmallString.h"
14 #include "llvm/ADT/SmallVector.h"
15 #include "llvm/ADT/StringRef.h"
16 #include "llvm/Bitcode/BitcodeReader.h"
17 #include "llvm/Config/llvm-config.h"
18 #include "llvm/IR/Comdat.h"
19 #include "llvm/IR/DataLayout.h"
20 #include "llvm/IR/GlobalAlias.h"
21 #include "llvm/IR/GlobalObject.h"
22 #include "llvm/IR/Mangler.h"
23 #include "llvm/IR/Metadata.h"
24 #include "llvm/IR/Module.h"
25 #include "llvm/MC/StringTableBuilder.h"
26 #include "llvm/Object/ModuleSymbolTable.h"
27 #include "llvm/Object/SymbolicFile.h"
28 #include "llvm/Support/Allocator.h"
29 #include "llvm/Support/Casting.h"
30 #include "llvm/Support/CommandLine.h"
31 #include "llvm/Support/Error.h"
32 #include "llvm/Support/StringSaver.h"
33 #include "llvm/Support/VCSRevision.h"
34 #include "llvm/Support/raw_ostream.h"
35 #include "llvm/TargetParser/Triple.h"
36 #include <cassert>
37 #include <string>
38 #include <utility>
39 #include <vector>
40 
41 using namespace llvm;
42 using namespace irsymtab;
43 
44 static cl::opt<bool> DisableBitcodeVersionUpgrade(
45     "disable-bitcode-version-upgrade", cl::Hidden,
46     cl::desc("Disable automatic bitcode upgrade for version mismatch"));
47 
48 static const char *PreservedSymbols[] = {
49 #define HANDLE_LIBCALL(code, name) name,
50 #include "llvm/IR/RuntimeLibcalls.def"
51 #undef HANDLE_LIBCALL
52     // There are global variables, so put it here instead of in
53     // RuntimeLibcalls.def.
54     // TODO: Are there similar such variables?
55     "__ssp_canary_word",
56     "__stack_chk_guard",
57 };
58 
59 namespace {
60 
getExpectedProducerName()61 const char *getExpectedProducerName() {
62   static char DefaultName[] = LLVM_VERSION_STRING
63 #ifdef LLVM_REVISION
64       " " LLVM_REVISION
65 #endif
66       ;
67   // Allows for testing of the irsymtab writer and upgrade mechanism. This
68   // environment variable should not be set by users.
69   if (char *OverrideName = getenv("LLVM_OVERRIDE_PRODUCER"))
70     return OverrideName;
71   return DefaultName;
72 }
73 
74 const char *kExpectedProducerName = getExpectedProducerName();
75 
76 /// Stores the temporary state that is required to build an IR symbol table.
77 struct Builder {
78   SmallVector<char, 0> &Symtab;
79   StringTableBuilder &StrtabBuilder;
80   StringSaver Saver;
81 
82   // This ctor initializes a StringSaver using the passed in BumpPtrAllocator.
83   // The StringTableBuilder does not create a copy of any strings added to it,
84   // so this provides somewhere to store any strings that we create.
Builder__anon960edc490111::Builder85   Builder(SmallVector<char, 0> &Symtab, StringTableBuilder &StrtabBuilder,
86           BumpPtrAllocator &Alloc)
87       : Symtab(Symtab), StrtabBuilder(StrtabBuilder), Saver(Alloc) {}
88 
89   DenseMap<const Comdat *, int> ComdatMap;
90   Mangler Mang;
91   Triple TT;
92 
93   std::vector<storage::Comdat> Comdats;
94   std::vector<storage::Module> Mods;
95   std::vector<storage::Symbol> Syms;
96   std::vector<storage::Uncommon> Uncommons;
97 
98   std::string COFFLinkerOpts;
99   raw_string_ostream COFFLinkerOptsOS{COFFLinkerOpts};
100 
101   std::vector<storage::Str> DependentLibraries;
102 
setStr__anon960edc490111::Builder103   void setStr(storage::Str &S, StringRef Value) {
104     S.Offset = StrtabBuilder.add(Value);
105     S.Size = Value.size();
106   }
107 
108   template <typename T>
writeRange__anon960edc490111::Builder109   void writeRange(storage::Range<T> &R, const std::vector<T> &Objs) {
110     R.Offset = Symtab.size();
111     R.Size = Objs.size();
112     Symtab.insert(Symtab.end(), reinterpret_cast<const char *>(Objs.data()),
113                   reinterpret_cast<const char *>(Objs.data() + Objs.size()));
114   }
115 
116   Expected<int> getComdatIndex(const Comdat *C, const Module *M);
117 
118   Error addModule(Module *M);
119   Error addSymbol(const ModuleSymbolTable &Msymtab,
120                   const SmallPtrSet<GlobalValue *, 4> &Used,
121                   ModuleSymbolTable::Symbol Sym);
122 
123   Error build(ArrayRef<Module *> Mods);
124 };
125 
addModule(Module * M)126 Error Builder::addModule(Module *M) {
127   if (M->getDataLayoutStr().empty())
128     return make_error<StringError>("input module has no datalayout",
129                                    inconvertibleErrorCode());
130 
131   // Symbols in the llvm.used list will get the FB_Used bit and will not be
132   // internalized. We do this for llvm.compiler.used as well:
133   //
134   // IR symbol table tracks module-level asm symbol references but not inline
135   // asm. A symbol only referenced by inline asm is not in the IR symbol table,
136   // so we may not know that the definition (in another translation unit) is
137   // referenced. That definition may have __attribute__((used)) (which lowers to
138   // llvm.compiler.used on ELF targets) to communicate to the compiler that it
139   // may be used by inline asm. The usage is perfectly fine, so we treat
140   // llvm.compiler.used conservatively as llvm.used to work around our own
141   // limitation.
142   SmallVector<GlobalValue *, 4> UsedV;
143   collectUsedGlobalVariables(*M, UsedV, /*CompilerUsed=*/false);
144   collectUsedGlobalVariables(*M, UsedV, /*CompilerUsed=*/true);
145   SmallPtrSet<GlobalValue *, 4> Used(UsedV.begin(), UsedV.end());
146 
147   ModuleSymbolTable Msymtab;
148   Msymtab.addModule(M);
149 
150   storage::Module Mod;
151   Mod.Begin = Syms.size();
152   Mod.End = Syms.size() + Msymtab.symbols().size();
153   Mod.UncBegin = Uncommons.size();
154   Mods.push_back(Mod);
155 
156   if (TT.isOSBinFormatCOFF()) {
157     if (auto E = M->materializeMetadata())
158       return E;
159     if (NamedMDNode *LinkerOptions =
160             M->getNamedMetadata("llvm.linker.options")) {
161       for (MDNode *MDOptions : LinkerOptions->operands())
162         for (const MDOperand &MDOption : cast<MDNode>(MDOptions)->operands())
163           COFFLinkerOptsOS << " " << cast<MDString>(MDOption)->getString();
164     }
165   }
166 
167   if (TT.isOSBinFormatELF()) {
168     if (auto E = M->materializeMetadata())
169       return E;
170     if (NamedMDNode *N = M->getNamedMetadata("llvm.dependent-libraries")) {
171       for (MDNode *MDOptions : N->operands()) {
172         const auto OperandStr =
173             cast<MDString>(cast<MDNode>(MDOptions)->getOperand(0))->getString();
174         storage::Str Specifier;
175         setStr(Specifier, OperandStr);
176         DependentLibraries.emplace_back(Specifier);
177       }
178     }
179   }
180 
181   for (ModuleSymbolTable::Symbol Msym : Msymtab.symbols())
182     if (Error Err = addSymbol(Msymtab, Used, Msym))
183       return Err;
184 
185   return Error::success();
186 }
187 
getComdatIndex(const Comdat * C,const Module * M)188 Expected<int> Builder::getComdatIndex(const Comdat *C, const Module *M) {
189   auto P = ComdatMap.insert(std::make_pair(C, Comdats.size()));
190   if (P.second) {
191     std::string Name;
192     if (TT.isOSBinFormatCOFF()) {
193       const GlobalValue *GV = M->getNamedValue(C->getName());
194       if (!GV)
195         return make_error<StringError>("Could not find leader",
196                                        inconvertibleErrorCode());
197       // Internal leaders do not affect symbol resolution, therefore they do not
198       // appear in the symbol table.
199       if (GV->hasLocalLinkage()) {
200         P.first->second = -1;
201         return -1;
202       }
203       llvm::raw_string_ostream OS(Name);
204       Mang.getNameWithPrefix(OS, GV, false);
205     } else {
206       Name = std::string(C->getName());
207     }
208 
209     storage::Comdat Comdat;
210     setStr(Comdat.Name, Saver.save(Name));
211     Comdat.SelectionKind = C->getSelectionKind();
212     Comdats.push_back(Comdat);
213   }
214 
215   return P.first->second;
216 }
217 
buildPreservedSymbolsSet()218 static DenseSet<StringRef> buildPreservedSymbolsSet() {
219   return DenseSet<StringRef>(std::begin(PreservedSymbols),
220                              std::end(PreservedSymbols));
221 }
222 
addSymbol(const ModuleSymbolTable & Msymtab,const SmallPtrSet<GlobalValue *,4> & Used,ModuleSymbolTable::Symbol Msym)223 Error Builder::addSymbol(const ModuleSymbolTable &Msymtab,
224                          const SmallPtrSet<GlobalValue *, 4> &Used,
225                          ModuleSymbolTable::Symbol Msym) {
226   Syms.emplace_back();
227   storage::Symbol &Sym = Syms.back();
228   Sym = {};
229 
230   storage::Uncommon *Unc = nullptr;
231   auto Uncommon = [&]() -> storage::Uncommon & {
232     if (Unc)
233       return *Unc;
234     Sym.Flags |= 1 << storage::Symbol::FB_has_uncommon;
235     Uncommons.emplace_back();
236     Unc = &Uncommons.back();
237     *Unc = {};
238     setStr(Unc->COFFWeakExternFallbackName, "");
239     setStr(Unc->SectionName, "");
240     return *Unc;
241   };
242 
243   SmallString<64> Name;
244   {
245     raw_svector_ostream OS(Name);
246     Msymtab.printSymbolName(OS, Msym);
247   }
248   setStr(Sym.Name, Saver.save(Name.str()));
249 
250   auto Flags = Msymtab.getSymbolFlags(Msym);
251   if (Flags & object::BasicSymbolRef::SF_Undefined)
252     Sym.Flags |= 1 << storage::Symbol::FB_undefined;
253   if (Flags & object::BasicSymbolRef::SF_Weak)
254     Sym.Flags |= 1 << storage::Symbol::FB_weak;
255   if (Flags & object::BasicSymbolRef::SF_Common)
256     Sym.Flags |= 1 << storage::Symbol::FB_common;
257   if (Flags & object::BasicSymbolRef::SF_Indirect)
258     Sym.Flags |= 1 << storage::Symbol::FB_indirect;
259   if (Flags & object::BasicSymbolRef::SF_Global)
260     Sym.Flags |= 1 << storage::Symbol::FB_global;
261   if (Flags & object::BasicSymbolRef::SF_FormatSpecific)
262     Sym.Flags |= 1 << storage::Symbol::FB_format_specific;
263   if (Flags & object::BasicSymbolRef::SF_Executable)
264     Sym.Flags |= 1 << storage::Symbol::FB_executable;
265 
266   Sym.ComdatIndex = -1;
267   auto *GV = dyn_cast_if_present<GlobalValue *>(Msym);
268   if (!GV) {
269     // Undefined module asm symbols act as GC roots and are implicitly used.
270     if (Flags & object::BasicSymbolRef::SF_Undefined)
271       Sym.Flags |= 1 << storage::Symbol::FB_used;
272     setStr(Sym.IRName, "");
273     return Error::success();
274   }
275 
276   setStr(Sym.IRName, GV->getName());
277 
278   static const DenseSet<StringRef> PreservedSymbolsSet =
279       buildPreservedSymbolsSet();
280   bool IsPreservedSymbol = PreservedSymbolsSet.contains(GV->getName());
281 
282   if (Used.count(GV) || IsPreservedSymbol)
283     Sym.Flags |= 1 << storage::Symbol::FB_used;
284   if (GV->isThreadLocal())
285     Sym.Flags |= 1 << storage::Symbol::FB_tls;
286   if (GV->hasGlobalUnnamedAddr())
287     Sym.Flags |= 1 << storage::Symbol::FB_unnamed_addr;
288   if (GV->canBeOmittedFromSymbolTable())
289     Sym.Flags |= 1 << storage::Symbol::FB_may_omit;
290   Sym.Flags |= unsigned(GV->getVisibility()) << storage::Symbol::FB_visibility;
291 
292   if (Flags & object::BasicSymbolRef::SF_Common) {
293     auto *GVar = dyn_cast<GlobalVariable>(GV);
294     if (!GVar)
295       return make_error<StringError>("Only variables can have common linkage!",
296                                      inconvertibleErrorCode());
297     Uncommon().CommonSize =
298         GV->getParent()->getDataLayout().getTypeAllocSize(GV->getValueType());
299     Uncommon().CommonAlign = GVar->getAlign() ? GVar->getAlign()->value() : 0;
300   }
301 
302   const GlobalObject *GO = GV->getAliaseeObject();
303   if (!GO) {
304     if (isa<GlobalIFunc>(GV))
305       GO = cast<GlobalIFunc>(GV)->getResolverFunction();
306     if (!GO)
307       return make_error<StringError>("Unable to determine comdat of alias!",
308                                      inconvertibleErrorCode());
309   }
310   if (const Comdat *C = GO->getComdat()) {
311     Expected<int> ComdatIndexOrErr = getComdatIndex(C, GV->getParent());
312     if (!ComdatIndexOrErr)
313       return ComdatIndexOrErr.takeError();
314     Sym.ComdatIndex = *ComdatIndexOrErr;
315   }
316 
317   if (TT.isOSBinFormatCOFF()) {
318     emitLinkerFlagsForGlobalCOFF(COFFLinkerOptsOS, GV, TT, Mang);
319 
320     if ((Flags & object::BasicSymbolRef::SF_Weak) &&
321         (Flags & object::BasicSymbolRef::SF_Indirect)) {
322       auto *Fallback = dyn_cast<GlobalValue>(
323           cast<GlobalAlias>(GV)->getAliasee()->stripPointerCasts());
324       if (!Fallback)
325         return make_error<StringError>("Invalid weak external",
326                                        inconvertibleErrorCode());
327       std::string FallbackName;
328       raw_string_ostream OS(FallbackName);
329       Msymtab.printSymbolName(OS, Fallback);
330       OS.flush();
331       setStr(Uncommon().COFFWeakExternFallbackName, Saver.save(FallbackName));
332     }
333   }
334 
335   if (!GO->getSection().empty())
336     setStr(Uncommon().SectionName, Saver.save(GO->getSection()));
337 
338   return Error::success();
339 }
340 
build(ArrayRef<Module * > IRMods)341 Error Builder::build(ArrayRef<Module *> IRMods) {
342   storage::Header Hdr;
343 
344   assert(!IRMods.empty());
345   Hdr.Version = storage::Header::kCurrentVersion;
346   setStr(Hdr.Producer, kExpectedProducerName);
347   setStr(Hdr.TargetTriple, IRMods[0]->getTargetTriple());
348   setStr(Hdr.SourceFileName, IRMods[0]->getSourceFileName());
349   TT = Triple(IRMods[0]->getTargetTriple());
350 
351   for (auto *M : IRMods)
352     if (Error Err = addModule(M))
353       return Err;
354 
355   COFFLinkerOptsOS.flush();
356   setStr(Hdr.COFFLinkerOpts, Saver.save(COFFLinkerOpts));
357 
358   // We are about to fill in the header's range fields, so reserve space for it
359   // and copy it in afterwards.
360   Symtab.resize(sizeof(storage::Header));
361   writeRange(Hdr.Modules, Mods);
362   writeRange(Hdr.Comdats, Comdats);
363   writeRange(Hdr.Symbols, Syms);
364   writeRange(Hdr.Uncommons, Uncommons);
365   writeRange(Hdr.DependentLibraries, DependentLibraries);
366   *reinterpret_cast<storage::Header *>(Symtab.data()) = Hdr;
367   return Error::success();
368 }
369 
370 } // end anonymous namespace
371 
build(ArrayRef<Module * > Mods,SmallVector<char,0> & Symtab,StringTableBuilder & StrtabBuilder,BumpPtrAllocator & Alloc)372 Error irsymtab::build(ArrayRef<Module *> Mods, SmallVector<char, 0> &Symtab,
373                       StringTableBuilder &StrtabBuilder,
374                       BumpPtrAllocator &Alloc) {
375   return Builder(Symtab, StrtabBuilder, Alloc).build(Mods);
376 }
377 
378 // Upgrade a vector of bitcode modules created by an old version of LLVM by
379 // creating an irsymtab for them in the current format.
upgrade(ArrayRef<BitcodeModule> BMs)380 static Expected<FileContents> upgrade(ArrayRef<BitcodeModule> BMs) {
381   FileContents FC;
382 
383   LLVMContext Ctx;
384   std::vector<Module *> Mods;
385   std::vector<std::unique_ptr<Module>> OwnedMods;
386   for (auto BM : BMs) {
387     Expected<std::unique_ptr<Module>> MOrErr =
388         BM.getLazyModule(Ctx, /*ShouldLazyLoadMetadata*/ true,
389                          /*IsImporting*/ false);
390     if (!MOrErr)
391       return MOrErr.takeError();
392 
393     Mods.push_back(MOrErr->get());
394     OwnedMods.push_back(std::move(*MOrErr));
395   }
396 
397   StringTableBuilder StrtabBuilder(StringTableBuilder::RAW);
398   BumpPtrAllocator Alloc;
399   if (Error E = build(Mods, FC.Symtab, StrtabBuilder, Alloc))
400     return std::move(E);
401 
402   StrtabBuilder.finalizeInOrder();
403   FC.Strtab.resize(StrtabBuilder.getSize());
404   StrtabBuilder.write((uint8_t *)FC.Strtab.data());
405 
406   FC.TheReader = {{FC.Symtab.data(), FC.Symtab.size()},
407                   {FC.Strtab.data(), FC.Strtab.size()}};
408   return std::move(FC);
409 }
410 
readBitcode(const BitcodeFileContents & BFC)411 Expected<FileContents> irsymtab::readBitcode(const BitcodeFileContents &BFC) {
412   if (BFC.Mods.empty())
413     return make_error<StringError>("Bitcode file does not contain any modules",
414                                    inconvertibleErrorCode());
415 
416   if (!DisableBitcodeVersionUpgrade) {
417     if (BFC.StrtabForSymtab.empty() ||
418         BFC.Symtab.size() < sizeof(storage::Header))
419       return upgrade(BFC.Mods);
420 
421     // We cannot use the regular reader to read the version and producer,
422     // because it will expect the header to be in the current format. The only
423     // thing we can rely on is that the version and producer will be present as
424     // the first struct elements.
425     auto *Hdr = reinterpret_cast<const storage::Header *>(BFC.Symtab.data());
426     unsigned Version = Hdr->Version;
427     StringRef Producer = Hdr->Producer.get(BFC.StrtabForSymtab);
428     if (Version != storage::Header::kCurrentVersion ||
429         Producer != kExpectedProducerName)
430       return upgrade(BFC.Mods);
431   }
432 
433   FileContents FC;
434   FC.TheReader = {{BFC.Symtab.data(), BFC.Symtab.size()},
435                   {BFC.StrtabForSymtab.data(), BFC.StrtabForSymtab.size()}};
436 
437   // Finally, make sure that the number of modules in the symbol table matches
438   // the number of modules in the bitcode file. If they differ, it may mean that
439   // the bitcode file was created by binary concatenation, so we need to create
440   // a new symbol table from scratch.
441   if (FC.TheReader.getNumModules() != BFC.Mods.size())
442     return upgrade(std::move(BFC.Mods));
443 
444   return std::move(FC);
445 }
446