1//===- InputFiles.h ---------------------------------------------*- C++ -*-===//
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#ifndef LLD_COFF_INPUT_FILES_H
10#define LLD_COFF_INPUT_FILES_H
11
12#include "Config.h"
13#include "lld/Common/LLVM.h"
14#include "llvm/ADT/ArrayRef.h"
15#include "llvm/ADT/DenseMap.h"
16#include "llvm/ADT/DenseSet.h"
17#include "llvm/ADT/StringSet.h"
18#include "llvm/BinaryFormat/Magic.h"
19#include "llvm/Object/Archive.h"
20#include "llvm/Object/COFF.h"
21#include "llvm/Support/StringSaver.h"
22#include <memory>
23#include <set>
24#include <vector>
25
26namespace llvm {
27struct DILineInfo;
28namespace pdb {
29class DbiModuleDescriptorBuilder;
30class NativeSession;
31}
32namespace lto {
33class InputFile;
34}
35}
36
37namespace lld {
38class DWARFCache;
39
40namespace coff {
41class COFFLinkerContext;
42
43const COFFSyncStream &operator<<(const COFFSyncStream &, const InputFile *);
44
45std::vector<MemoryBufferRef> getArchiveMembers(COFFLinkerContext &,
46 llvm::object::Archive *file);
47
48using llvm::COFF::IMAGE_FILE_MACHINE_UNKNOWN;
49using llvm::COFF::MachineTypes;
50using llvm::object::Archive;
51using llvm::object::COFFObjectFile;
52using llvm::object::COFFSymbolRef;
53using llvm::object::coff_import_header;
54using llvm::object::coff_section;
55
56class Chunk;
57class Defined;
58class DefinedImportData;
59class DefinedImportThunk;
60class DefinedRegular;
61class ImportThunkChunk;
62class ImportThunkChunkARM64EC;
63class SectionChunk;
64class Symbol;
65class SymbolTable;
66class Undefined;
67class TpiSource;
68
69// The root class of input files.
70class InputFile {
71public:
72 enum Kind {
73 ArchiveKind,
74 ObjectKind,
75 PDBKind,
76 ImportKind,
77 BitcodeKind,
78 DLLKind
79 };
80 Kind kind() const { return fileKind; }
81 virtual ~InputFile() {}
82
83 // Returns the filename.
84 StringRef getName() const { return mb.getBufferIdentifier(); }
85
86 // Reads a file (the constructor doesn't do that).
87 virtual void parse() = 0;
88
89 // Returns the CPU type this file was compiled to.
90 virtual MachineTypes getMachineType() const {
91 return IMAGE_FILE_MACHINE_UNKNOWN;
92 }
93
94 MemoryBufferRef mb;
95
96 // An archive file name if this file is created from an archive.
97 StringRef parentName;
98
99 // Returns .drectve section contents if exist.
100 StringRef getDirectives() { return directives; }
101
102 SymbolTable &symtab;
103
104protected:
105 InputFile(SymbolTable &s, Kind k, MemoryBufferRef m, bool lazy = false)
106 : mb(m), symtab(s), fileKind(k), lazy(lazy) {}
107
108 StringRef directives;
109
110private:
111 const Kind fileKind;
112
113public:
114 // True if this is a lazy ObjFile or BitcodeFile.
115 bool lazy = false;
116};
117
118// .lib or .a file.
119class ArchiveFile : public InputFile {
120public:
121 explicit ArchiveFile(COFFLinkerContext &ctx, MemoryBufferRef m);
122 static bool classof(const InputFile *f) { return f->kind() == ArchiveKind; }
123 void parse() override;
124
125 // Enqueues an archive member load for the given symbol. If we've already
126 // enqueued a load for the same archive member, this function does nothing,
127 // which ensures that we don't load the same member more than once.
128 void addMember(const Archive::Symbol &sym);
129
130private:
131 std::unique_ptr<Archive> file;
132 llvm::DenseSet<uint64_t> seen;
133};
134
135// .obj or .o file. This may be a member of an archive file.
136class ObjFile : public InputFile {
137public:
138 static ObjFile *create(COFFLinkerContext &ctx, MemoryBufferRef mb,
139 bool lazy = false);
140 explicit ObjFile(SymbolTable &symtab, COFFObjectFile *coffObj, bool lazy);
141
142 static bool classof(const InputFile *f) { return f->kind() == ObjectKind; }
143 void parse() override;
144 void parseLazy();
145 MachineTypes getMachineType() const override;
146 ArrayRef<Chunk *> getChunks() { return chunks; }
147 ArrayRef<SectionChunk *> getDebugChunks() { return debugChunks; }
148 ArrayRef<SectionChunk *> getSXDataChunks() { return sxDataChunks; }
149 ArrayRef<SectionChunk *> getGuardFidChunks() { return guardFidChunks; }
150 ArrayRef<SectionChunk *> getGuardIATChunks() { return guardIATChunks; }
151 ArrayRef<SectionChunk *> getGuardLJmpChunks() { return guardLJmpChunks; }
152 ArrayRef<SectionChunk *> getGuardEHContChunks() { return guardEHContChunks; }
153 ArrayRef<Symbol *> getSymbols() { return symbols; }
154
155 MutableArrayRef<Symbol *> getMutableSymbols() { return symbols; }
156
157 ArrayRef<uint8_t> getDebugSection(StringRef secName);
158
159 // Returns a Symbol object for the symbolIndex'th symbol in the
160 // underlying object file.
161 Symbol *getSymbol(uint32_t symbolIndex) {
162 return symbols[symbolIndex];
163 }
164
165 // Returns the underlying COFF file.
166 COFFObjectFile *getCOFFObj() { return coffObj.get(); }
167
168 // Add a symbol for a range extension thunk. Return the new symbol table
169 // index. This index can be used to modify a relocation.
170 uint32_t addRangeThunkSymbol(Symbol *thunk) {
171 symbols.push_back(x: thunk);
172 return symbols.size() - 1;
173 }
174
175 void includeResourceChunks();
176
177 bool isResourceObjFile() const { return !resourceChunks.empty(); }
178
179 // Flags in the absolute @feat.00 symbol if it is present. These usually
180 // indicate if an object was compiled with certain security features enabled
181 // like stack guard, safeseh, /guard:cf, or other things.
182 uint32_t feat00Flags = 0;
183
184 // True if this object file is compatible with SEH. COFF-specific and
185 // x86-only. COFF spec 5.10.1. The .sxdata section.
186 bool hasSafeSEH() { return feat00Flags & 0x1; }
187
188 // True if this file was compiled with /guard:cf.
189 bool hasGuardCF() { return feat00Flags & 0x800; }
190
191 // True if this file was compiled with /guard:ehcont.
192 bool hasGuardEHCont() { return feat00Flags & 0x4000; }
193
194 // Pointer to the PDB module descriptor builder. Various debug info records
195 // will reference object files by "module index", which is here. Things like
196 // source files and section contributions are also recorded here. Will be null
197 // if we are not producing a PDB.
198 llvm::pdb::DbiModuleDescriptorBuilder *moduleDBI = nullptr;
199
200 const coff_section *addrsigSec = nullptr;
201
202 const coff_section *callgraphSec = nullptr;
203
204 // When using Microsoft precompiled headers, this is the PCH's key.
205 // The same key is used by both the precompiled object, and objects using the
206 // precompiled object. Any difference indicates out-of-date objects.
207 std::optional<uint32_t> pchSignature;
208
209 // Whether this file was compiled with /hotpatch.
210 bool hotPatchable = false;
211
212 // Whether the object was already merged into the final PDB.
213 bool mergedIntoPDB = false;
214
215 // If the OBJ has a .debug$T stream, this tells how it will be handled.
216 TpiSource *debugTypesObj = nullptr;
217
218 // The .debug$P or .debug$T section data if present. Empty otherwise.
219 ArrayRef<uint8_t> debugTypes;
220
221 std::optional<std::pair<StringRef, uint32_t>>
222 getVariableLocation(StringRef var);
223
224 std::optional<llvm::DILineInfo> getDILineInfo(uint32_t offset,
225 uint32_t sectionIndex);
226
227private:
228 const coff_section* getSection(uint32_t i);
229 const coff_section *getSection(COFFSymbolRef sym) {
230 return getSection(i: sym.getSectionNumber());
231 }
232
233 void enqueuePdbFile(StringRef path, ObjFile *fromFile);
234
235 void initializeChunks();
236 void initializeSymbols();
237 void initializeFlags();
238 void initializeDependencies();
239 void initializeECThunks();
240
241 SectionChunk *
242 readSection(uint32_t sectionNumber,
243 const llvm::object::coff_aux_section_definition *def,
244 StringRef leaderName);
245
246 void readAssociativeDefinition(
247 COFFSymbolRef coffSym,
248 const llvm::object::coff_aux_section_definition *def);
249
250 void readAssociativeDefinition(
251 COFFSymbolRef coffSym,
252 const llvm::object::coff_aux_section_definition *def,
253 uint32_t parentSection);
254
255 void recordPrevailingSymbolForMingw(
256 COFFSymbolRef coffSym,
257 llvm::DenseMap<StringRef, uint32_t> &prevailingSectionMap);
258
259 void maybeAssociateSEHForMingw(
260 COFFSymbolRef sym, const llvm::object::coff_aux_section_definition *def,
261 const llvm::DenseMap<StringRef, uint32_t> &prevailingSectionMap);
262
263 // Given a new symbol Sym with comdat selection Selection, if the new
264 // symbol is not (yet) Prevailing and the existing comdat leader set to
265 // Leader, emits a diagnostic if the new symbol and its selection doesn't
266 // match the existing symbol and its selection. If either old or new
267 // symbol have selection IMAGE_COMDAT_SELECT_LARGEST, Sym might replace
268 // the existing leader. In that case, Prevailing is set to true.
269 void
270 handleComdatSelection(COFFSymbolRef sym, llvm::COFF::COMDATType &selection,
271 bool &prevailing, DefinedRegular *leader,
272 const llvm::object::coff_aux_section_definition *def);
273
274 std::optional<Symbol *>
275 createDefined(COFFSymbolRef sym,
276 std::vector<const llvm::object::coff_aux_section_definition *>
277 &comdatDefs,
278 bool &prevailingComdat);
279 Symbol *createRegular(COFFSymbolRef sym);
280 Symbol *createUndefined(COFFSymbolRef sym, bool overrideLazy);
281
282 std::unique_ptr<COFFObjectFile> coffObj;
283
284 // List of all chunks defined by this file. This includes both section
285 // chunks and non-section chunks for common symbols.
286 std::vector<Chunk *> chunks;
287
288 std::vector<SectionChunk *> resourceChunks;
289
290 // CodeView debug info sections.
291 std::vector<SectionChunk *> debugChunks;
292
293 // Chunks containing symbol table indices of exception handlers. Only used for
294 // 32-bit x86.
295 std::vector<SectionChunk *> sxDataChunks;
296
297 // Chunks containing symbol table indices of address taken symbols, address
298 // taken IAT entries, longjmp and ehcont targets. These are not linked into
299 // the final binary when /guard:cf is set.
300 std::vector<SectionChunk *> guardFidChunks;
301 std::vector<SectionChunk *> guardIATChunks;
302 std::vector<SectionChunk *> guardLJmpChunks;
303 std::vector<SectionChunk *> guardEHContChunks;
304
305 std::vector<SectionChunk *> hybmpChunks;
306
307 // This vector contains a list of all symbols defined or referenced by this
308 // file. They are indexed such that you can get a Symbol by symbol
309 // index. Nonexistent indices (which are occupied by auxiliary
310 // symbols in the real symbol table) are filled with null pointers.
311 std::vector<Symbol *> symbols;
312
313 // This vector contains the same chunks as Chunks, but they are
314 // indexed such that you can get a SectionChunk by section index.
315 // Nonexistent section indices are filled with null pointers.
316 // (Because section number is 1-based, the first slot is always a
317 // null pointer.) This vector is only valid during initialization.
318 std::vector<SectionChunk *> sparseChunks;
319
320 DWARFCache *dwarf = nullptr;
321};
322
323// This is a PDB type server dependency, that is not a input file per se, but
324// needs to be treated like one. Such files are discovered from the debug type
325// stream.
326class PDBInputFile : public InputFile {
327public:
328 explicit PDBInputFile(COFFLinkerContext &ctx, MemoryBufferRef m);
329 ~PDBInputFile();
330 static bool classof(const InputFile *f) { return f->kind() == PDBKind; }
331 void parse() override;
332
333 static PDBInputFile *findFromRecordPath(const COFFLinkerContext &ctx,
334 StringRef path, ObjFile *fromFile);
335
336 // Record possible errors while opening the PDB file
337 std::optional<std::string> loadErrorStr;
338
339 // This is the actual interface to the PDB (if it was opened successfully)
340 std::unique_ptr<llvm::pdb::NativeSession> session;
341
342 // If the PDB has a .debug$T stream, this tells how it will be handled.
343 TpiSource *debugTypesObj = nullptr;
344};
345
346// This type represents import library members that contain DLL names
347// and symbols exported from the DLLs. See Microsoft PE/COFF spec. 7
348// for details about the format.
349class ImportFile : public InputFile {
350public:
351 explicit ImportFile(COFFLinkerContext &ctx, MemoryBufferRef m);
352
353 static bool classof(const InputFile *f) { return f->kind() == ImportKind; }
354 MachineTypes getMachineType() const override { return getMachineType(m: mb); }
355 static MachineTypes getMachineType(MemoryBufferRef m);
356 bool isSameImport(const ImportFile *other) const;
357 bool isEC() const { return impECSym != nullptr; }
358
359 DefinedImportData *impSym = nullptr;
360 Defined *thunkSym = nullptr;
361 ImportThunkChunkARM64EC *impchkThunk = nullptr;
362 ImportFile *hybridFile = nullptr;
363 std::string dllName;
364
365private:
366 void parse() override;
367 ImportThunkChunk *makeImportThunk();
368
369public:
370 StringRef externalName;
371 const coff_import_header *hdr;
372 Chunk *location = nullptr;
373
374 // Auxiliary IAT symbols and chunks on ARM64EC.
375 DefinedImportData *impECSym = nullptr;
376 Chunk *auxLocation = nullptr;
377 Defined *auxThunkSym = nullptr;
378 DefinedImportData *auxImpCopySym = nullptr;
379 Chunk *auxCopyLocation = nullptr;
380
381 // We want to eliminate dllimported symbols if no one actually refers to them.
382 // These "Live" bits are used to keep track of which import library members
383 // are actually in use.
384 //
385 // If the Live bit is turned off by MarkLive, Writer will ignore dllimported
386 // symbols provided by this import library member.
387 bool live;
388};
389
390// Used for LTO.
391class BitcodeFile : public InputFile {
392public:
393 explicit BitcodeFile(SymbolTable &symtab, MemoryBufferRef mb,
394 std::unique_ptr<llvm::lto::InputFile> &obj, bool lazy);
395 ~BitcodeFile();
396
397 static BitcodeFile *create(COFFLinkerContext &ctx, MemoryBufferRef mb,
398 StringRef archiveName, uint64_t offsetInArchive,
399 bool lazy);
400 static bool classof(const InputFile *f) { return f->kind() == BitcodeKind; }
401 ArrayRef<Symbol *> getSymbols() { return symbols; }
402 MachineTypes getMachineType() const override {
403 return getMachineType(obj: obj.get());
404 }
405 static MachineTypes getMachineType(const llvm::lto::InputFile *obj);
406 void parseLazy();
407 std::unique_ptr<llvm::lto::InputFile> obj;
408
409private:
410 void parse() override;
411
412 std::vector<Symbol *> symbols;
413};
414
415// .dll file. MinGW only.
416class DLLFile : public InputFile {
417public:
418 explicit DLLFile(SymbolTable &symtab, MemoryBufferRef m)
419 : InputFile(symtab, DLLKind, m) {}
420 static bool classof(const InputFile *f) { return f->kind() == DLLKind; }
421 void parse() override;
422 MachineTypes getMachineType() const override;
423
424 struct Symbol {
425 StringRef dllName;
426 StringRef symbolName;
427 llvm::COFF::ImportNameType nameType;
428 llvm::COFF::ImportType importType;
429 };
430
431 void makeImport(Symbol *s);
432
433private:
434 std::unique_ptr<COFFObjectFile> coffObj;
435 llvm::StringSet<> seen;
436};
437
438inline bool isBitcode(MemoryBufferRef mb) {
439 return identify_magic(magic: mb.getBuffer()) == llvm::file_magic::bitcode;
440}
441
442std::string replaceThinLTOSuffix(StringRef path, StringRef suffix,
443 StringRef repl);
444} // namespace coff
445
446std::string toString(const coff::InputFile *file);
447} // namespace lld
448
449#endif
450