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/SmallPtrSet.h"
12#include "llvm/ADT/SmallString.h"
13#include "llvm/ADT/SmallVector.h"
14#include "llvm/ADT/StringRef.h"
15#include "llvm/Bitcode/BitcodeReader.h"
16#include "llvm/Config/llvm-config.h"
17#include "llvm/IR/Comdat.h"
18#include "llvm/IR/DataLayout.h"
19#include "llvm/IR/GlobalAlias.h"
20#include "llvm/IR/GlobalObject.h"
21#include "llvm/IR/Mangler.h"
22#include "llvm/IR/Metadata.h"
23#include "llvm/IR/Module.h"
24#include "llvm/MC/StringTableBuilder.h"
25#include "llvm/Object/ModuleSymbolTable.h"
26#include "llvm/Object/SymbolicFile.h"
27#include "llvm/Support/Allocator.h"
28#include "llvm/Support/Casting.h"
29#include "llvm/Support/CommandLine.h"
30#include "llvm/Support/Error.h"
31#include "llvm/Support/StringSaver.h"
32#include "llvm/Support/VCSRevision.h"
33#include "llvm/Support/raw_ostream.h"
34#include "llvm/TargetParser/Triple.h"
35#include <cassert>
36#include <string>
37#include <utility>
38#include <vector>
39
40using namespace llvm;
41using namespace irsymtab;
42
43static cl::opt<bool> DisableBitcodeVersionUpgrade(
44 "disable-bitcode-version-upgrade", cl::Hidden,
45 cl::desc("Disable automatic bitcode upgrade for version mismatch"));
46
47namespace {
48
49const char *getExpectedProducerName() {
50 static char DefaultName[] = LLVM_VERSION_STRING
51#ifdef LLVM_REVISION
52 " " LLVM_REVISION
53#endif
54 ;
55 // Allows for testing of the irsymtab writer and upgrade mechanism. This
56 // environment variable should not be set by users.
57 if (char *OverrideName = getenv(name: "LLVM_OVERRIDE_PRODUCER"))
58 return OverrideName;
59 return DefaultName;
60}
61
62const char *kExpectedProducerName = getExpectedProducerName();
63
64/// Stores the temporary state that is required to build an IR symbol table.
65struct Builder {
66 SmallVector<char, 0> &Symtab;
67 StringTableBuilder &StrtabBuilder;
68 StringSaver Saver;
69
70 // This ctor initializes a StringSaver using the passed in BumpPtrAllocator.
71 // The StringTableBuilder does not create a copy of any strings added to it,
72 // so this provides somewhere to store any strings that we create.
73 Builder(SmallVector<char, 0> &Symtab, StringTableBuilder &StrtabBuilder,
74 BumpPtrAllocator &Alloc, const Triple &TT)
75 : Symtab(Symtab), StrtabBuilder(StrtabBuilder), Saver(Alloc), TT(TT) {}
76
77 DenseMap<const Comdat *, int> ComdatMap;
78 Mangler Mang;
79 const Triple &TT;
80
81 std::vector<storage::Comdat> Comdats;
82 std::vector<storage::Module> Mods;
83 std::vector<storage::Symbol> Syms;
84 std::vector<storage::Uncommon> Uncommons;
85
86 std::string COFFLinkerOpts;
87 raw_string_ostream COFFLinkerOptsOS{COFFLinkerOpts};
88
89 std::vector<storage::Str> DependentLibraries;
90
91 void setStr(storage::Str &S, StringRef Value) {
92 S.Offset = StrtabBuilder.add(S: Value);
93 S.Size = Value.size();
94 }
95
96 template <typename T>
97 void writeRange(storage::Range<T> &R, const std::vector<T> &Objs) {
98 R.Offset = Symtab.size();
99 R.Size = Objs.size();
100 Symtab.insert(I: Symtab.end(), From: reinterpret_cast<const char *>(Objs.data()),
101 To: reinterpret_cast<const char *>(Objs.data() + Objs.size()));
102 }
103
104 Expected<int> getComdatIndex(const Comdat *C, const Module *M);
105
106 Error addModule(Module *M);
107 Error addSymbol(const ModuleSymbolTable &Msymtab,
108 const SmallPtrSet<GlobalValue *, 4> &Used,
109 ModuleSymbolTable::Symbol Sym);
110
111 Error build(ArrayRef<Module *> Mods);
112};
113
114Error Builder::addModule(Module *M) {
115 if (M->getDataLayoutStr().empty())
116 return make_error<StringError>(Args: "input module has no datalayout",
117 Args: inconvertibleErrorCode());
118
119 // Symbols in the llvm.used list will get the FB_Used bit and will not be
120 // internalized. We do this for llvm.compiler.used as well:
121 //
122 // IR symbol table tracks module-level asm symbol references but not inline
123 // asm. A symbol only referenced by inline asm is not in the IR symbol table,
124 // so we may not know that the definition (in another translation unit) is
125 // referenced. That definition may have __attribute__((used)) (which lowers to
126 // llvm.compiler.used on ELF targets) to communicate to the compiler that it
127 // may be used by inline asm. The usage is perfectly fine, so we treat
128 // llvm.compiler.used conservatively as llvm.used to work around our own
129 // limitation.
130 SmallVector<GlobalValue *, 4> UsedV;
131 collectUsedGlobalVariables(M: *M, Vec&: UsedV, /*CompilerUsed=*/false);
132 collectUsedGlobalVariables(M: *M, Vec&: UsedV, /*CompilerUsed=*/true);
133 SmallPtrSet<GlobalValue *, 4> Used(llvm::from_range, UsedV);
134
135 ModuleSymbolTable Msymtab;
136 Msymtab.addModule(M);
137
138 storage::Module Mod;
139 Mod.Begin = Syms.size();
140 Mod.End = Syms.size() + Msymtab.symbols().size();
141 Mod.UncBegin = Uncommons.size();
142 Mods.push_back(x: Mod);
143
144 if (TT.isOSBinFormatCOFF()) {
145 if (auto E = M->materializeMetadata())
146 return E;
147 if (NamedMDNode *LinkerOptions =
148 M->getNamedMetadata(Name: "llvm.linker.options")) {
149 for (MDNode *MDOptions : LinkerOptions->operands())
150 for (const MDOperand &MDOption : cast<MDNode>(Val: MDOptions)->operands())
151 COFFLinkerOptsOS << " " << cast<MDString>(Val: MDOption)->getString();
152 }
153 }
154
155 if (TT.isOSBinFormatELF()) {
156 if (auto E = M->materializeMetadata())
157 return E;
158 if (NamedMDNode *N = M->getNamedMetadata(Name: "llvm.dependent-libraries")) {
159 for (MDNode *MDOptions : N->operands()) {
160 const auto OperandStr =
161 cast<MDString>(Val: cast<MDNode>(Val: MDOptions)->getOperand(I: 0))->getString();
162 storage::Str Specifier;
163 setStr(S&: Specifier, Value: OperandStr);
164 DependentLibraries.emplace_back(args&: Specifier);
165 }
166 }
167 }
168
169 for (ModuleSymbolTable::Symbol Msym : Msymtab.symbols())
170 if (Error Err = addSymbol(Msymtab, Used, Sym: Msym))
171 return Err;
172
173 return Error::success();
174}
175
176Expected<int> Builder::getComdatIndex(const Comdat *C, const Module *M) {
177 auto P = ComdatMap.insert(KV: std::make_pair(x&: C, y: Comdats.size()));
178 if (P.second) {
179 std::string Name;
180 if (TT.isOSBinFormatCOFF()) {
181 const GlobalValue *GV = M->getNamedValue(Name: C->getName());
182 if (!GV)
183 return make_error<StringError>(Args: "Could not find leader",
184 Args: inconvertibleErrorCode());
185 // Internal leaders do not affect symbol resolution, therefore they do not
186 // appear in the symbol table.
187 if (GV->hasLocalLinkage()) {
188 P.first->second = -1;
189 return -1;
190 }
191 llvm::raw_string_ostream OS(Name);
192 Mang.getNameWithPrefix(OS, GV, CannotUsePrivateLabel: false);
193 } else {
194 Name = std::string(C->getName());
195 }
196
197 storage::Comdat Comdat;
198 setStr(S&: Comdat.Name, Value: Saver.save(S: Name));
199 Comdat.SelectionKind = C->getSelectionKind();
200 Comdats.push_back(x: Comdat);
201 }
202
203 return P.first->second;
204}
205
206Error Builder::addSymbol(const ModuleSymbolTable &Msymtab,
207 const SmallPtrSet<GlobalValue *, 4> &Used,
208 ModuleSymbolTable::Symbol Msym) {
209 Syms.emplace_back();
210 storage::Symbol &Sym = Syms.back();
211 Sym = {};
212
213 storage::Uncommon *Unc = nullptr;
214 auto Uncommon = [&]() -> storage::Uncommon & {
215 if (Unc)
216 return *Unc;
217 Sym.Flags |= 1 << storage::Symbol::FB_has_uncommon;
218 Uncommons.emplace_back();
219 Unc = &Uncommons.back();
220 *Unc = {};
221 setStr(S&: Unc->COFFWeakExternFallbackName, Value: "");
222 setStr(S&: Unc->SectionName, Value: "");
223 return *Unc;
224 };
225
226 SmallString<64> Name;
227 {
228 raw_svector_ostream OS(Name);
229 Msymtab.printSymbolName(OS, S: Msym);
230 }
231 setStr(S&: Sym.Name, Value: Saver.save(S: Name.str()));
232
233 auto Flags = Msymtab.getSymbolFlags(S: Msym);
234 if (Flags & object::BasicSymbolRef::SF_Undefined)
235 Sym.Flags |= 1 << storage::Symbol::FB_undefined;
236 if (Flags & object::BasicSymbolRef::SF_Weak)
237 Sym.Flags |= 1 << storage::Symbol::FB_weak;
238 if (Flags & object::BasicSymbolRef::SF_Common)
239 Sym.Flags |= 1 << storage::Symbol::FB_common;
240 if (Flags & object::BasicSymbolRef::SF_Indirect)
241 Sym.Flags |= 1 << storage::Symbol::FB_indirect;
242 if (Flags & object::BasicSymbolRef::SF_Global)
243 Sym.Flags |= 1 << storage::Symbol::FB_global;
244 if (Flags & object::BasicSymbolRef::SF_FormatSpecific)
245 Sym.Flags |= 1 << storage::Symbol::FB_format_specific;
246 if (Flags & object::BasicSymbolRef::SF_Executable)
247 Sym.Flags |= 1 << storage::Symbol::FB_executable;
248
249 Sym.ComdatIndex = -1;
250 auto *GV = dyn_cast_if_present<GlobalValue *>(Val&: Msym);
251 if (!GV) {
252 // Undefined module asm symbols act as GC roots and are implicitly used.
253 if (Flags & object::BasicSymbolRef::SF_Undefined)
254 Sym.Flags |= 1 << storage::Symbol::FB_used;
255 setStr(S&: Sym.IRName, Value: "");
256 return Error::success();
257 }
258
259 StringRef GVName = GV->getName();
260 setStr(S&: Sym.IRName, Value: GVName);
261
262 if (Used.count(Ptr: GV))
263 Sym.Flags |= 1 << storage::Symbol::FB_used;
264 if (GV->isThreadLocal())
265 Sym.Flags |= 1 << storage::Symbol::FB_tls;
266 if (GV->hasGlobalUnnamedAddr())
267 Sym.Flags |= 1 << storage::Symbol::FB_unnamed_addr;
268 if (GV->canBeOmittedFromSymbolTable())
269 Sym.Flags |= 1 << storage::Symbol::FB_may_omit;
270 Sym.Flags |= unsigned(GV->getVisibility()) << storage::Symbol::FB_visibility;
271
272 if (Flags & object::BasicSymbolRef::SF_Common) {
273 auto *GVar = dyn_cast<GlobalVariable>(Val: GV);
274 if (!GVar)
275 return make_error<StringError>(Args: "Only variables can have common linkage!",
276 Args: inconvertibleErrorCode());
277 Uncommon().CommonSize = GVar->getGlobalSize(DL: GV->getDataLayout());
278 Uncommon().CommonAlign = GVar->getAlign() ? GVar->getAlign()->value() : 0;
279 }
280
281 const GlobalObject *GO = GV->getAliaseeObject();
282 if (!GO) {
283 if (isa<GlobalIFunc>(Val: GV))
284 GO = cast<GlobalIFunc>(Val: GV)->getResolverFunction();
285 if (!GO)
286 return make_error<StringError>(Args: "Unable to determine comdat of alias!",
287 Args: inconvertibleErrorCode());
288 }
289 if (const Comdat *C = GO->getComdat()) {
290 Expected<int> ComdatIndexOrErr = getComdatIndex(C, M: GV->getParent());
291 if (!ComdatIndexOrErr)
292 return ComdatIndexOrErr.takeError();
293 Sym.ComdatIndex = *ComdatIndexOrErr;
294 }
295
296 if (TT.isOSBinFormatCOFF()) {
297 emitLinkerFlagsForGlobalCOFF(OS&: COFFLinkerOptsOS, GV, TT, Mangler&: Mang);
298
299 if ((Flags & object::BasicSymbolRef::SF_Weak) &&
300 (Flags & object::BasicSymbolRef::SF_Indirect)) {
301 auto *Fallback = dyn_cast<GlobalValue>(
302 Val: cast<GlobalAlias>(Val: GV)->getAliasee()->stripPointerCasts());
303 if (!Fallback)
304 return make_error<StringError>(Args: "Invalid weak external",
305 Args: inconvertibleErrorCode());
306 std::string FallbackName;
307 raw_string_ostream OS(FallbackName);
308 Msymtab.printSymbolName(OS, S: Fallback);
309 setStr(S&: Uncommon().COFFWeakExternFallbackName, Value: Saver.save(S: FallbackName));
310 }
311 }
312
313 if (!GO->getSection().empty())
314 setStr(S&: Uncommon().SectionName, Value: Saver.save(S: GO->getSection()));
315
316 return Error::success();
317}
318
319Error Builder::build(ArrayRef<Module *> IRMods) {
320 storage::Header Hdr;
321
322 assert(!IRMods.empty());
323 Hdr.Version = storage::Header::kCurrentVersion;
324 setStr(S&: Hdr.Producer, Value: kExpectedProducerName);
325 setStr(S&: Hdr.TargetTriple, Value: IRMods[0]->getTargetTriple().str());
326 setStr(S&: Hdr.SourceFileName, Value: IRMods[0]->getSourceFileName());
327
328 for (auto *M : IRMods)
329 if (Error Err = addModule(M))
330 return Err;
331
332 setStr(S&: Hdr.COFFLinkerOpts, Value: Saver.save(S: COFFLinkerOpts));
333
334 // We are about to fill in the header's range fields, so reserve space for it
335 // and copy it in afterwards.
336 Symtab.resize(N: sizeof(storage::Header));
337 writeRange(R&: Hdr.Modules, Objs: Mods);
338 writeRange(R&: Hdr.Comdats, Objs: Comdats);
339 writeRange(R&: Hdr.Symbols, Objs: Syms);
340 writeRange(R&: Hdr.Uncommons, Objs: Uncommons);
341 writeRange(R&: Hdr.DependentLibraries, Objs: DependentLibraries);
342 *reinterpret_cast<storage::Header *>(Symtab.data()) = Hdr;
343 return Error::success();
344}
345
346} // end anonymous namespace
347
348Error irsymtab::build(ArrayRef<Module *> Mods, SmallVector<char, 0> &Symtab,
349 StringTableBuilder &StrtabBuilder,
350 BumpPtrAllocator &Alloc) {
351 const Triple &TT = Mods[0]->getTargetTriple();
352 return Builder(Symtab, StrtabBuilder, Alloc, TT).build(IRMods: Mods);
353}
354
355// Upgrade a vector of bitcode modules created by an old version of LLVM by
356// creating an irsymtab for them in the current format.
357static Expected<FileContents> upgrade(ArrayRef<BitcodeModule> BMs) {
358 FileContents FC;
359
360 LLVMContext Ctx;
361 std::vector<Module *> Mods;
362 std::vector<std::unique_ptr<Module>> OwnedMods;
363 for (auto BM : BMs) {
364 Expected<std::unique_ptr<Module>> MOrErr =
365 BM.getLazyModule(Context&: Ctx, /*ShouldLazyLoadMetadata*/ true,
366 /*IsImporting*/ false);
367 if (!MOrErr)
368 return MOrErr.takeError();
369
370 Mods.push_back(x: MOrErr->get());
371 OwnedMods.push_back(x: std::move(*MOrErr));
372 }
373
374 StringTableBuilder StrtabBuilder(StringTableBuilder::RAW);
375 BumpPtrAllocator Alloc;
376 if (Error E = build(Mods, Symtab&: FC.Symtab, StrtabBuilder, Alloc))
377 return std::move(E);
378
379 StrtabBuilder.finalizeInOrder();
380 FC.Strtab.resize(N: StrtabBuilder.getSize());
381 StrtabBuilder.write(Buf: (uint8_t *)FC.Strtab.data());
382
383 FC.TheReader = {{FC.Symtab.data(), FC.Symtab.size()},
384 {FC.Strtab.data(), FC.Strtab.size()}};
385 return std::move(FC);
386}
387
388Expected<FileContents> irsymtab::readBitcode(const BitcodeFileContents &BFC) {
389 if (BFC.Mods.empty())
390 return make_error<StringError>(Args: "Bitcode file does not contain any modules",
391 Args: inconvertibleErrorCode());
392
393 if (!DisableBitcodeVersionUpgrade) {
394 if (BFC.StrtabForSymtab.empty() ||
395 BFC.Symtab.size() < sizeof(storage::Header))
396 return upgrade(BMs: BFC.Mods);
397
398 // We cannot use the regular reader to read the version and producer,
399 // because it will expect the header to be in the current format. The only
400 // thing we can rely on is that the version and producer will be present as
401 // the first struct elements.
402 auto *Hdr = reinterpret_cast<const storage::Header *>(BFC.Symtab.data());
403 unsigned Version = Hdr->Version;
404 StringRef Producer = Hdr->Producer.get(Strtab: BFC.StrtabForSymtab);
405 if (Version != storage::Header::kCurrentVersion ||
406 Producer != kExpectedProducerName)
407 return upgrade(BMs: BFC.Mods);
408 }
409
410 FileContents FC;
411 FC.TheReader = {{BFC.Symtab.data(), BFC.Symtab.size()},
412 {BFC.StrtabForSymtab.data(), BFC.StrtabForSymtab.size()}};
413
414 // Finally, make sure that the number of modules in the symbol table matches
415 // the number of modules in the bitcode file. If they differ, it may mean that
416 // the bitcode file was created by binary concatenation, so we need to create
417 // a new symbol table from scratch.
418 if (FC.TheReader.getNumModules() != BFC.Mods.size())
419 return upgrade(BMs: std::move(BFC.Mods));
420
421 return std::move(FC);
422}
423