1 | //===- InputFiles.cpp -----------------------------------------------------===// |
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 "InputFiles.h" |
10 | #include "Config.h" |
11 | #include "DWARF.h" |
12 | #include "Driver.h" |
13 | #include "InputSection.h" |
14 | #include "LinkerScript.h" |
15 | #include "SymbolTable.h" |
16 | #include "Symbols.h" |
17 | #include "SyntheticSections.h" |
18 | #include "Target.h" |
19 | #include "lld/Common/CommonLinkerContext.h" |
20 | #include "lld/Common/DWARF.h" |
21 | #include "llvm/ADT/CachedHashString.h" |
22 | #include "llvm/ADT/STLExtras.h" |
23 | #include "llvm/LTO/LTO.h" |
24 | #include "llvm/Object/IRObjectFile.h" |
25 | #include "llvm/Support/ARMAttributeParser.h" |
26 | #include "llvm/Support/ARMBuildAttributes.h" |
27 | #include "llvm/Support/Endian.h" |
28 | #include "llvm/Support/FileSystem.h" |
29 | #include "llvm/Support/Path.h" |
30 | #include "llvm/Support/RISCVAttributeParser.h" |
31 | #include "llvm/Support/TarWriter.h" |
32 | #include "llvm/Support/TimeProfiler.h" |
33 | #include "llvm/Support/raw_ostream.h" |
34 | #include <optional> |
35 | |
36 | using namespace llvm; |
37 | using namespace llvm::ELF; |
38 | using namespace llvm::object; |
39 | using namespace llvm::sys; |
40 | using namespace llvm::sys::fs; |
41 | using namespace llvm::support::endian; |
42 | using namespace lld; |
43 | using namespace lld::elf; |
44 | |
45 | // This function is explicitly instantiated in ARM.cpp, don't do it here to |
46 | // avoid warnings with MSVC. |
47 | extern template void ObjFile<ELF32LE>::importCmseSymbols(); |
48 | extern template void ObjFile<ELF32BE>::importCmseSymbols(); |
49 | extern template void ObjFile<ELF64LE>::importCmseSymbols(); |
50 | extern template void ObjFile<ELF64BE>::importCmseSymbols(); |
51 | |
52 | bool InputFile::isInGroup; |
53 | uint32_t InputFile::nextGroupId; |
54 | |
55 | std::unique_ptr<TarWriter> elf::tar; |
56 | |
57 | // Returns "<internal>", "foo.a(bar.o)" or "baz.o". |
58 | std::string lld::toString(const InputFile *f) { |
59 | static std::mutex mu; |
60 | if (!f) |
61 | return "<internal>" ; |
62 | |
63 | { |
64 | std::lock_guard<std::mutex> lock(mu); |
65 | if (f->toStringCache.empty()) { |
66 | if (f->archiveName.empty()) |
67 | f->toStringCache = f->getName(); |
68 | else |
69 | (f->archiveName + "(" + f->getName() + ")" ).toVector(Out&: f->toStringCache); |
70 | } |
71 | } |
72 | return std::string(f->toStringCache); |
73 | } |
74 | |
75 | static ELFKind getELFKind(MemoryBufferRef mb, StringRef archiveName) { |
76 | unsigned char size; |
77 | unsigned char endian; |
78 | std::tie(args&: size, args&: endian) = getElfArchType(Object: mb.getBuffer()); |
79 | |
80 | auto report = [&](StringRef msg) { |
81 | StringRef filename = mb.getBufferIdentifier(); |
82 | if (archiveName.empty()) |
83 | fatal(msg: filename + ": " + msg); |
84 | else |
85 | fatal(msg: archiveName + "(" + filename + "): " + msg); |
86 | }; |
87 | |
88 | if (!mb.getBuffer().starts_with(Prefix: ElfMagic)) |
89 | report("not an ELF file" ); |
90 | if (endian != ELFDATA2LSB && endian != ELFDATA2MSB) |
91 | report("corrupted ELF file: invalid data encoding" ); |
92 | if (size != ELFCLASS32 && size != ELFCLASS64) |
93 | report("corrupted ELF file: invalid file class" ); |
94 | |
95 | size_t bufSize = mb.getBuffer().size(); |
96 | if ((size == ELFCLASS32 && bufSize < sizeof(Elf32_Ehdr)) || |
97 | (size == ELFCLASS64 && bufSize < sizeof(Elf64_Ehdr))) |
98 | report("corrupted ELF file: file is too short" ); |
99 | |
100 | if (size == ELFCLASS32) |
101 | return (endian == ELFDATA2LSB) ? ELF32LEKind : ELF32BEKind; |
102 | return (endian == ELFDATA2LSB) ? ELF64LEKind : ELF64BEKind; |
103 | } |
104 | |
105 | // For ARM only, to set the EF_ARM_ABI_FLOAT_SOFT or EF_ARM_ABI_FLOAT_HARD |
106 | // flag in the ELF Header we need to look at Tag_ABI_VFP_args to find out how |
107 | // the input objects have been compiled. |
108 | static void updateARMVFPArgs(const ARMAttributeParser &attributes, |
109 | const InputFile *f) { |
110 | std::optional<unsigned> attr = |
111 | attributes.getAttributeValue(tag: ARMBuildAttrs::ABI_VFP_args); |
112 | if (!attr) |
113 | // If an ABI tag isn't present then it is implicitly given the value of 0 |
114 | // which maps to ARMBuildAttrs::BaseAAPCS. However many assembler files, |
115 | // including some in glibc that don't use FP args (and should have value 3) |
116 | // don't have the attribute so we do not consider an implicit value of 0 |
117 | // as a clash. |
118 | return; |
119 | |
120 | unsigned vfpArgs = *attr; |
121 | ARMVFPArgKind arg; |
122 | switch (vfpArgs) { |
123 | case ARMBuildAttrs::BaseAAPCS: |
124 | arg = ARMVFPArgKind::Base; |
125 | break; |
126 | case ARMBuildAttrs::HardFPAAPCS: |
127 | arg = ARMVFPArgKind::VFP; |
128 | break; |
129 | case ARMBuildAttrs::ToolChainFPPCS: |
130 | // Tool chain specific convention that conforms to neither AAPCS variant. |
131 | arg = ARMVFPArgKind::ToolChain; |
132 | break; |
133 | case ARMBuildAttrs::CompatibleFPAAPCS: |
134 | // Object compatible with all conventions. |
135 | return; |
136 | default: |
137 | error(msg: toString(f) + ": unknown Tag_ABI_VFP_args value: " + Twine(vfpArgs)); |
138 | return; |
139 | } |
140 | // Follow ld.bfd and error if there is a mix of calling conventions. |
141 | if (config->armVFPArgs != arg && config->armVFPArgs != ARMVFPArgKind::Default) |
142 | error(msg: toString(f) + ": incompatible Tag_ABI_VFP_args" ); |
143 | else |
144 | config->armVFPArgs = arg; |
145 | } |
146 | |
147 | // The ARM support in lld makes some use of instructions that are not available |
148 | // on all ARM architectures. Namely: |
149 | // - Use of BLX instruction for interworking between ARM and Thumb state. |
150 | // - Use of the extended Thumb branch encoding in relocation. |
151 | // - Use of the MOVT/MOVW instructions in Thumb Thunks. |
152 | // The ARM Attributes section contains information about the architecture chosen |
153 | // at compile time. We follow the convention that if at least one input object |
154 | // is compiled with an architecture that supports these features then lld is |
155 | // permitted to use them. |
156 | static void updateSupportedARMFeatures(const ARMAttributeParser &attributes) { |
157 | std::optional<unsigned> attr = |
158 | attributes.getAttributeValue(tag: ARMBuildAttrs::CPU_arch); |
159 | if (!attr) |
160 | return; |
161 | auto arch = *attr; |
162 | switch (arch) { |
163 | case ARMBuildAttrs::Pre_v4: |
164 | case ARMBuildAttrs::v4: |
165 | case ARMBuildAttrs::v4T: |
166 | // Architectures prior to v5 do not support BLX instruction |
167 | break; |
168 | case ARMBuildAttrs::v5T: |
169 | case ARMBuildAttrs::v5TE: |
170 | case ARMBuildAttrs::v5TEJ: |
171 | case ARMBuildAttrs::v6: |
172 | case ARMBuildAttrs::v6KZ: |
173 | case ARMBuildAttrs::v6K: |
174 | config->armHasBlx = true; |
175 | // Architectures used in pre-Cortex processors do not support |
176 | // The J1 = 1 J2 = 1 Thumb branch range extension, with the exception |
177 | // of Architecture v6T2 (arm1156t2-s and arm1156t2f-s) that do. |
178 | break; |
179 | default: |
180 | // All other Architectures have BLX and extended branch encoding |
181 | config->armHasBlx = true; |
182 | config->armJ1J2BranchEncoding = true; |
183 | if (arch != ARMBuildAttrs::v6_M && arch != ARMBuildAttrs::v6S_M) |
184 | // All Architectures used in Cortex processors with the exception |
185 | // of v6-M and v6S-M have the MOVT and MOVW instructions. |
186 | config->armHasMovtMovw = true; |
187 | break; |
188 | } |
189 | |
190 | // Only ARMv8-M or later architectures have CMSE support. |
191 | std::optional<unsigned> profile = |
192 | attributes.getAttributeValue(tag: ARMBuildAttrs::CPU_arch_profile); |
193 | if (!profile) |
194 | return; |
195 | if (arch >= ARMBuildAttrs::CPUArch::v8_M_Base && |
196 | profile == ARMBuildAttrs::MicroControllerProfile) |
197 | config->armCMSESupport = true; |
198 | |
199 | // The thumb PLT entries require Thumb2 which can be used on multiple archs. |
200 | // For now, let's limit it to ones where ARM isn't available and we know have |
201 | // Thumb2. |
202 | std::optional<unsigned> armISA = |
203 | attributes.getAttributeValue(tag: ARMBuildAttrs::ARM_ISA_use); |
204 | std::optional<unsigned> thumb = |
205 | attributes.getAttributeValue(tag: ARMBuildAttrs::THUMB_ISA_use); |
206 | config->armHasArmISA |= armISA && *armISA >= ARMBuildAttrs::Allowed; |
207 | config->armHasThumb2ISA |= thumb && *thumb >= ARMBuildAttrs::AllowThumb32; |
208 | } |
209 | |
210 | InputFile::InputFile(Kind k, MemoryBufferRef m) |
211 | : mb(m), groupId(nextGroupId), fileKind(k) { |
212 | // All files within the same --{start,end}-group get the same group ID. |
213 | // Otherwise, a new file will get a new group ID. |
214 | if (!isInGroup) |
215 | ++nextGroupId; |
216 | } |
217 | |
218 | std::optional<MemoryBufferRef> elf::readFile(StringRef path) { |
219 | llvm::TimeTraceScope timeScope("Load input files" , path); |
220 | |
221 | // The --chroot option changes our virtual root directory. |
222 | // This is useful when you are dealing with files created by --reproduce. |
223 | if (!config->chroot.empty() && path.starts_with(Prefix: "/" )) |
224 | path = saver().save(S: config->chroot + path); |
225 | |
226 | bool remapped = false; |
227 | auto it = config->remapInputs.find(Val: path); |
228 | if (it != config->remapInputs.end()) { |
229 | path = it->second; |
230 | remapped = true; |
231 | } else { |
232 | for (const auto &[pat, toFile] : config->remapInputsWildcards) { |
233 | if (pat.match(S: path)) { |
234 | path = toFile; |
235 | remapped = true; |
236 | break; |
237 | } |
238 | } |
239 | } |
240 | if (remapped) { |
241 | // Use /dev/null to indicate an input file that should be ignored. Change |
242 | // the path to NUL on Windows. |
243 | #ifdef _WIN32 |
244 | if (path == "/dev/null" ) |
245 | path = "NUL" ; |
246 | #endif |
247 | } |
248 | |
249 | log(msg: path); |
250 | config->dependencyFiles.insert(X: llvm::CachedHashString(path)); |
251 | |
252 | auto mbOrErr = MemoryBuffer::getFile(Filename: path, /*IsText=*/false, |
253 | /*RequiresNullTerminator=*/false); |
254 | if (auto ec = mbOrErr.getError()) { |
255 | error(msg: "cannot open " + path + ": " + ec.message()); |
256 | return std::nullopt; |
257 | } |
258 | |
259 | MemoryBufferRef mbref = (*mbOrErr)->getMemBufferRef(); |
260 | ctx.memoryBuffers.push_back(Elt: std::move(*mbOrErr)); // take MB ownership |
261 | |
262 | if (tar) |
263 | tar->append(Path: relativeToRoot(path), Data: mbref.getBuffer()); |
264 | return mbref; |
265 | } |
266 | |
267 | // All input object files must be for the same architecture |
268 | // (e.g. it does not make sense to link x86 object files with |
269 | // MIPS object files.) This function checks for that error. |
270 | static bool isCompatible(InputFile *file) { |
271 | if (!file->isElf() && !isa<BitcodeFile>(Val: file)) |
272 | return true; |
273 | |
274 | if (file->ekind == config->ekind && file->emachine == config->emachine) { |
275 | if (config->emachine != EM_MIPS) |
276 | return true; |
277 | if (isMipsN32Abi(f: file) == config->mipsN32Abi) |
278 | return true; |
279 | } |
280 | |
281 | StringRef target = |
282 | !config->bfdname.empty() ? config->bfdname : config->emulation; |
283 | if (!target.empty()) { |
284 | error(msg: toString(f: file) + " is incompatible with " + target); |
285 | return false; |
286 | } |
287 | |
288 | InputFile *existing = nullptr; |
289 | if (!ctx.objectFiles.empty()) |
290 | existing = ctx.objectFiles[0]; |
291 | else if (!ctx.sharedFiles.empty()) |
292 | existing = ctx.sharedFiles[0]; |
293 | else if (!ctx.bitcodeFiles.empty()) |
294 | existing = ctx.bitcodeFiles[0]; |
295 | std::string with; |
296 | if (existing) |
297 | with = " with " + toString(f: existing); |
298 | error(msg: toString(f: file) + " is incompatible" + with); |
299 | return false; |
300 | } |
301 | |
302 | template <class ELFT> static void doParseFile(InputFile *file) { |
303 | if (!isCompatible(file)) |
304 | return; |
305 | |
306 | // Lazy object file |
307 | if (file->lazy) { |
308 | if (auto *f = dyn_cast<BitcodeFile>(Val: file)) { |
309 | ctx.lazyBitcodeFiles.push_back(Elt: f); |
310 | f->parseLazy(); |
311 | } else { |
312 | cast<ObjFile<ELFT>>(file)->parseLazy(); |
313 | } |
314 | return; |
315 | } |
316 | |
317 | if (config->trace) |
318 | message(msg: toString(f: file)); |
319 | |
320 | if (file->kind() == InputFile::ObjKind) { |
321 | ctx.objectFiles.push_back(Elt: cast<ELFFileBase>(Val: file)); |
322 | cast<ObjFile<ELFT>>(file)->parse(); |
323 | } else if (auto *f = dyn_cast<SharedFile>(Val: file)) { |
324 | f->parse<ELFT>(); |
325 | } else if (auto *f = dyn_cast<BitcodeFile>(Val: file)) { |
326 | ctx.bitcodeFiles.push_back(Elt: f); |
327 | f->parse(); |
328 | } else { |
329 | ctx.binaryFiles.push_back(Elt: cast<BinaryFile>(Val: file)); |
330 | cast<BinaryFile>(Val: file)->parse(); |
331 | } |
332 | } |
333 | |
334 | // Add symbols in File to the symbol table. |
335 | void elf::parseFile(InputFile *file) { invokeELFT(doParseFile, file); } |
336 | |
337 | // This function is explicitly instantiated in ARM.cpp. Mark it extern here, |
338 | // to avoid warnings when building with MSVC. |
339 | extern template void ObjFile<ELF32LE>::importCmseSymbols(); |
340 | extern template void ObjFile<ELF32BE>::importCmseSymbols(); |
341 | extern template void ObjFile<ELF64LE>::importCmseSymbols(); |
342 | extern template void ObjFile<ELF64BE>::importCmseSymbols(); |
343 | |
344 | template <class ELFT> |
345 | static void doParseFiles(const std::vector<InputFile *> &files, |
346 | InputFile *armCmseImpLib) { |
347 | // Add all files to the symbol table. This will add almost all symbols that we |
348 | // need to the symbol table. This process might add files to the link due to |
349 | // addDependentLibrary. |
350 | for (size_t i = 0; i < files.size(); ++i) { |
351 | llvm::TimeTraceScope timeScope("Parse input files" , files[i]->getName()); |
352 | doParseFile<ELFT>(files[i]); |
353 | } |
354 | if (armCmseImpLib) |
355 | cast<ObjFile<ELFT>>(*armCmseImpLib).importCmseSymbols(); |
356 | } |
357 | |
358 | void elf::parseFiles(const std::vector<InputFile *> &files, |
359 | InputFile *armCmseImpLib) { |
360 | llvm::TimeTraceScope timeScope("Parse input files" ); |
361 | invokeELFT(doParseFiles, files, armCmseImpLib); |
362 | } |
363 | |
364 | // Concatenates arguments to construct a string representing an error location. |
365 | static std::string createFileLineMsg(StringRef path, unsigned line) { |
366 | std::string filename = std::string(path::filename(path)); |
367 | std::string lineno = ":" + std::to_string(val: line); |
368 | if (filename == path) |
369 | return filename + lineno; |
370 | return filename + lineno + " (" + path.str() + lineno + ")" ; |
371 | } |
372 | |
373 | template <class ELFT> |
374 | static std::string getSrcMsgAux(ObjFile<ELFT> &file, const Symbol &sym, |
375 | const InputSectionBase &sec, uint64_t offset) { |
376 | // In DWARF, functions and variables are stored to different places. |
377 | // First, look up a function for a given offset. |
378 | if (std::optional<DILineInfo> info = file.getDILineInfo(&sec, offset)) |
379 | return createFileLineMsg(path: info->FileName, line: info->Line); |
380 | |
381 | // If it failed, look up again as a variable. |
382 | if (std::optional<std::pair<std::string, unsigned>> fileLine = |
383 | file.getVariableLoc(sym.getName())) |
384 | return createFileLineMsg(path: fileLine->first, line: fileLine->second); |
385 | |
386 | // File.sourceFile contains STT_FILE symbol, and that is a last resort. |
387 | return std::string(file.sourceFile); |
388 | } |
389 | |
390 | std::string InputFile::getSrcMsg(const Symbol &sym, const InputSectionBase &sec, |
391 | uint64_t offset) { |
392 | if (kind() != ObjKind) |
393 | return "" ; |
394 | switch (ekind) { |
395 | default: |
396 | llvm_unreachable("Invalid kind" ); |
397 | case ELF32LEKind: |
398 | return getSrcMsgAux(file&: cast<ObjFile<ELF32LE>>(Val&: *this), sym, sec, offset); |
399 | case ELF32BEKind: |
400 | return getSrcMsgAux(file&: cast<ObjFile<ELF32BE>>(Val&: *this), sym, sec, offset); |
401 | case ELF64LEKind: |
402 | return getSrcMsgAux(file&: cast<ObjFile<ELF64LE>>(Val&: *this), sym, sec, offset); |
403 | case ELF64BEKind: |
404 | return getSrcMsgAux(file&: cast<ObjFile<ELF64BE>>(Val&: *this), sym, sec, offset); |
405 | } |
406 | } |
407 | |
408 | StringRef InputFile::getNameForScript() const { |
409 | if (archiveName.empty()) |
410 | return getName(); |
411 | |
412 | if (nameForScriptCache.empty()) |
413 | nameForScriptCache = (archiveName + Twine(':') + getName()).str(); |
414 | |
415 | return nameForScriptCache; |
416 | } |
417 | |
418 | // An ELF object file may contain a `.deplibs` section. If it exists, the |
419 | // section contains a list of library specifiers such as `m` for libm. This |
420 | // function resolves a given name by finding the first matching library checking |
421 | // the various ways that a library can be specified to LLD. This ELF extension |
422 | // is a form of autolinking and is called `dependent libraries`. It is currently |
423 | // unique to LLVM and lld. |
424 | static void addDependentLibrary(StringRef specifier, const InputFile *f) { |
425 | if (!config->dependentLibraries) |
426 | return; |
427 | if (std::optional<std::string> s = searchLibraryBaseName(path: specifier)) |
428 | ctx.driver.addFile(path: saver().save(S: *s), /*withLOption=*/true); |
429 | else if (std::optional<std::string> s = findFromSearchPaths(path: specifier)) |
430 | ctx.driver.addFile(path: saver().save(S: *s), /*withLOption=*/true); |
431 | else if (fs::exists(Path: specifier)) |
432 | ctx.driver.addFile(path: specifier, /*withLOption=*/false); |
433 | else |
434 | error(msg: toString(f) + |
435 | ": unable to find library from dependent library specifier: " + |
436 | specifier); |
437 | } |
438 | |
439 | // Record the membership of a section group so that in the garbage collection |
440 | // pass, section group members are kept or discarded as a unit. |
441 | template <class ELFT> |
442 | static void handleSectionGroup(ArrayRef<InputSectionBase *> sections, |
443 | ArrayRef<typename ELFT::Word> entries) { |
444 | bool hasAlloc = false; |
445 | for (uint32_t index : entries.slice(1)) { |
446 | if (index >= sections.size()) |
447 | return; |
448 | if (InputSectionBase *s = sections[index]) |
449 | if (s != &InputSection::discarded && s->flags & SHF_ALLOC) |
450 | hasAlloc = true; |
451 | } |
452 | |
453 | // If any member has the SHF_ALLOC flag, the whole group is subject to garbage |
454 | // collection. See the comment in markLive(). This rule retains .debug_types |
455 | // and .rela.debug_types. |
456 | if (!hasAlloc) |
457 | return; |
458 | |
459 | // Connect the members in a circular doubly-linked list via |
460 | // nextInSectionGroup. |
461 | InputSectionBase *head; |
462 | InputSectionBase *prev = nullptr; |
463 | for (uint32_t index : entries.slice(1)) { |
464 | InputSectionBase *s = sections[index]; |
465 | if (!s || s == &InputSection::discarded) |
466 | continue; |
467 | if (prev) |
468 | prev->nextInSectionGroup = s; |
469 | else |
470 | head = s; |
471 | prev = s; |
472 | } |
473 | if (prev) |
474 | prev->nextInSectionGroup = head; |
475 | } |
476 | |
477 | template <class ELFT> DWARFCache *ObjFile<ELFT>::getDwarf() { |
478 | llvm::call_once(initDwarf, [this]() { |
479 | dwarf = std::make_unique<DWARFCache>(std::make_unique<DWARFContext>( |
480 | std::make_unique<LLDDwarfObj<ELFT>>(this), "" , |
481 | [&](Error err) { warn(getName() + ": " + toString(E: std::move(err))); }, |
482 | [&](Error warning) { |
483 | warn(getName() + ": " + toString(E: std::move(warning))); |
484 | })); |
485 | }); |
486 | |
487 | return dwarf.get(); |
488 | } |
489 | |
490 | // Returns the pair of file name and line number describing location of data |
491 | // object (variable, array, etc) definition. |
492 | template <class ELFT> |
493 | std::optional<std::pair<std::string, unsigned>> |
494 | ObjFile<ELFT>::getVariableLoc(StringRef name) { |
495 | return getDwarf()->getVariableLoc(name); |
496 | } |
497 | |
498 | // Returns source line information for a given offset |
499 | // using DWARF debug info. |
500 | template <class ELFT> |
501 | std::optional<DILineInfo> |
502 | ObjFile<ELFT>::getDILineInfo(const InputSectionBase *s, uint64_t offset) { |
503 | // Detect SectionIndex for specified section. |
504 | uint64_t sectionIndex = object::SectionedAddress::UndefSection; |
505 | ArrayRef<InputSectionBase *> sections = s->file->getSections(); |
506 | for (uint64_t curIndex = 0; curIndex < sections.size(); ++curIndex) { |
507 | if (s == sections[curIndex]) { |
508 | sectionIndex = curIndex; |
509 | break; |
510 | } |
511 | } |
512 | |
513 | return getDwarf()->getDILineInfo(offset, sectionIndex); |
514 | } |
515 | |
516 | ELFFileBase::ELFFileBase(Kind k, ELFKind ekind, MemoryBufferRef mb) |
517 | : InputFile(k, mb) { |
518 | this->ekind = ekind; |
519 | } |
520 | |
521 | template <typename Elf_Shdr> |
522 | static const Elf_Shdr *findSection(ArrayRef<Elf_Shdr> sections, uint32_t type) { |
523 | for (const Elf_Shdr &sec : sections) |
524 | if (sec.sh_type == type) |
525 | return &sec; |
526 | return nullptr; |
527 | } |
528 | |
529 | void ELFFileBase::init() { |
530 | switch (ekind) { |
531 | case ELF32LEKind: |
532 | init<ELF32LE>(k: fileKind); |
533 | break; |
534 | case ELF32BEKind: |
535 | init<ELF32BE>(k: fileKind); |
536 | break; |
537 | case ELF64LEKind: |
538 | init<ELF64LE>(k: fileKind); |
539 | break; |
540 | case ELF64BEKind: |
541 | init<ELF64BE>(k: fileKind); |
542 | break; |
543 | default: |
544 | llvm_unreachable("getELFKind" ); |
545 | } |
546 | } |
547 | |
548 | template <class ELFT> void ELFFileBase::init(InputFile::Kind k) { |
549 | using Elf_Shdr = typename ELFT::Shdr; |
550 | using Elf_Sym = typename ELFT::Sym; |
551 | |
552 | // Initialize trivial attributes. |
553 | const ELFFile<ELFT> &obj = getObj<ELFT>(); |
554 | emachine = obj.getHeader().e_machine; |
555 | osabi = obj.getHeader().e_ident[llvm::ELF::EI_OSABI]; |
556 | abiVersion = obj.getHeader().e_ident[llvm::ELF::EI_ABIVERSION]; |
557 | |
558 | ArrayRef<Elf_Shdr> sections = CHECK(obj.sections(), this); |
559 | elfShdrs = sections.data(); |
560 | numELFShdrs = sections.size(); |
561 | |
562 | // Find a symbol table. |
563 | const Elf_Shdr *symtabSec = |
564 | findSection(sections, k == SharedKind ? SHT_DYNSYM : SHT_SYMTAB); |
565 | |
566 | if (!symtabSec) |
567 | return; |
568 | |
569 | // Initialize members corresponding to a symbol table. |
570 | firstGlobal = symtabSec->sh_info; |
571 | |
572 | ArrayRef<Elf_Sym> eSyms = CHECK(obj.symbols(symtabSec), this); |
573 | if (firstGlobal == 0 || firstGlobal > eSyms.size()) |
574 | fatal(msg: toString(f: this) + ": invalid sh_info in symbol table" ); |
575 | |
576 | elfSyms = reinterpret_cast<const void *>(eSyms.data()); |
577 | numELFSyms = uint32_t(eSyms.size()); |
578 | stringTable = CHECK(obj.getStringTableForSymtab(*symtabSec, sections), this); |
579 | } |
580 | |
581 | template <class ELFT> |
582 | uint32_t ObjFile<ELFT>::getSectionIndex(const Elf_Sym &sym) const { |
583 | return CHECK( |
584 | this->getObj().getSectionIndex(sym, getELFSyms<ELFT>(), shndxTable), |
585 | this); |
586 | } |
587 | |
588 | template <class ELFT> void ObjFile<ELFT>::parse(bool ignoreComdats) { |
589 | object::ELFFile<ELFT> obj = this->getObj(); |
590 | // Read a section table. justSymbols is usually false. |
591 | if (this->justSymbols) { |
592 | initializeJustSymbols(); |
593 | initializeSymbols(obj); |
594 | return; |
595 | } |
596 | |
597 | // Handle dependent libraries and selection of section groups as these are not |
598 | // done in parallel. |
599 | ArrayRef<Elf_Shdr> objSections = getELFShdrs<ELFT>(); |
600 | StringRef shstrtab = CHECK(obj.getSectionStringTable(objSections), this); |
601 | uint64_t size = objSections.size(); |
602 | sections.resize(size); |
603 | for (size_t i = 0; i != size; ++i) { |
604 | const Elf_Shdr &sec = objSections[i]; |
605 | if (sec.sh_type == SHT_LLVM_DEPENDENT_LIBRARIES && !config->relocatable) { |
606 | StringRef name = check(obj.getSectionName(sec, shstrtab)); |
607 | ArrayRef<char> data = CHECK( |
608 | this->getObj().template getSectionContentsAsArray<char>(sec), this); |
609 | if (!data.empty() && data.back() != '\0') { |
610 | error( |
611 | toString(this) + |
612 | ": corrupted dependent libraries section (unterminated string): " + |
613 | name); |
614 | } else { |
615 | for (const char *d = data.begin(), *e = data.end(); d < e;) { |
616 | StringRef s(d); |
617 | addDependentLibrary(s, this); |
618 | d += s.size() + 1; |
619 | } |
620 | } |
621 | this->sections[i] = &InputSection::discarded; |
622 | continue; |
623 | } |
624 | |
625 | if (sec.sh_type == SHT_ARM_ATTRIBUTES && config->emachine == EM_ARM) { |
626 | ARMAttributeParser attributes; |
627 | ArrayRef<uint8_t> contents = |
628 | check(this->getObj().getSectionContents(sec)); |
629 | StringRef name = check(obj.getSectionName(sec, shstrtab)); |
630 | this->sections[i] = &InputSection::discarded; |
631 | if (Error e = attributes.parse(section: contents, endian: ekind == ELF32LEKind |
632 | ? llvm::endianness::little |
633 | : llvm::endianness::big)) { |
634 | InputSection isec(*this, sec, name); |
635 | warn(msg: toString(&isec) + ": " + llvm::toString(E: std::move(e))); |
636 | } else { |
637 | updateSupportedARMFeatures(attributes); |
638 | updateARMVFPArgs(attributes, this); |
639 | |
640 | // FIXME: Retain the first attribute section we see. The eglibc ARM |
641 | // dynamic loaders require the presence of an attribute section for |
642 | // dlopen to work. In a full implementation we would merge all attribute |
643 | // sections. |
644 | if (in.attributes == nullptr) { |
645 | in.attributes = std::make_unique<InputSection>(*this, sec, name); |
646 | this->sections[i] = in.attributes.get(); |
647 | } |
648 | } |
649 | } |
650 | |
651 | // Producing a static binary with MTE globals is not currently supported, |
652 | // remove all SHT_AARCH64_MEMTAG_GLOBALS_STATIC sections as they're unused |
653 | // medatada, and we don't want them to end up in the output file for static |
654 | // executables. |
655 | if (sec.sh_type == SHT_AARCH64_MEMTAG_GLOBALS_STATIC && |
656 | !canHaveMemtagGlobals()) { |
657 | this->sections[i] = &InputSection::discarded; |
658 | continue; |
659 | } |
660 | |
661 | if (sec.sh_type != SHT_GROUP) |
662 | continue; |
663 | StringRef signature = getShtGroupSignature(sections: objSections, sec); |
664 | ArrayRef<Elf_Word> entries = |
665 | CHECK(obj.template getSectionContentsAsArray<Elf_Word>(sec), this); |
666 | if (entries.empty()) |
667 | fatal(toString(this) + ": empty SHT_GROUP" ); |
668 | |
669 | Elf_Word flag = entries[0]; |
670 | if (flag && flag != GRP_COMDAT) |
671 | fatal(toString(this) + ": unsupported SHT_GROUP format" ); |
672 | |
673 | bool keepGroup = |
674 | (flag & GRP_COMDAT) == 0 || ignoreComdats || |
675 | symtab.comdatGroups.try_emplace(CachedHashStringRef(signature), this) |
676 | .second; |
677 | if (keepGroup) { |
678 | if (!config->resolveGroups) |
679 | this->sections[i] = createInputSection( |
680 | idx: i, sec, name: check(obj.getSectionName(sec, shstrtab))); |
681 | continue; |
682 | } |
683 | |
684 | // Otherwise, discard group members. |
685 | for (uint32_t secIndex : entries.slice(1)) { |
686 | if (secIndex >= size) |
687 | fatal(toString(this) + |
688 | ": invalid section index in group: " + Twine(secIndex)); |
689 | this->sections[secIndex] = &InputSection::discarded; |
690 | } |
691 | } |
692 | |
693 | // Read a symbol table. |
694 | initializeSymbols(obj); |
695 | } |
696 | |
697 | // Sections with SHT_GROUP and comdat bits define comdat section groups. |
698 | // They are identified and deduplicated by group name. This function |
699 | // returns a group name. |
700 | template <class ELFT> |
701 | StringRef ObjFile<ELFT>::getShtGroupSignature(ArrayRef<Elf_Shdr> sections, |
702 | const Elf_Shdr &sec) { |
703 | typename ELFT::SymRange symbols = this->getELFSyms<ELFT>(); |
704 | if (sec.sh_info >= symbols.size()) |
705 | fatal(toString(this) + ": invalid symbol index" ); |
706 | const typename ELFT::Sym &sym = symbols[sec.sh_info]; |
707 | return CHECK(sym.getName(this->stringTable), this); |
708 | } |
709 | |
710 | template <class ELFT> |
711 | bool ObjFile<ELFT>::shouldMerge(const Elf_Shdr &sec, StringRef name) { |
712 | // On a regular link we don't merge sections if -O0 (default is -O1). This |
713 | // sometimes makes the linker significantly faster, although the output will |
714 | // be bigger. |
715 | // |
716 | // Doing the same for -r would create a problem as it would combine sections |
717 | // with different sh_entsize. One option would be to just copy every SHF_MERGE |
718 | // section as is to the output. While this would produce a valid ELF file with |
719 | // usable SHF_MERGE sections, tools like (llvm-)?dwarfdump get confused when |
720 | // they see two .debug_str. We could have separate logic for combining |
721 | // SHF_MERGE sections based both on their name and sh_entsize, but that seems |
722 | // to be more trouble than it is worth. Instead, we just use the regular (-O1) |
723 | // logic for -r. |
724 | if (config->optimize == 0 && !config->relocatable) |
725 | return false; |
726 | |
727 | // A mergeable section with size 0 is useless because they don't have |
728 | // any data to merge. A mergeable string section with size 0 can be |
729 | // argued as invalid because it doesn't end with a null character. |
730 | // We'll avoid a mess by handling them as if they were non-mergeable. |
731 | if (sec.sh_size == 0) |
732 | return false; |
733 | |
734 | // Check for sh_entsize. The ELF spec is not clear about the zero |
735 | // sh_entsize. It says that "the member [sh_entsize] contains 0 if |
736 | // the section does not hold a table of fixed-size entries". We know |
737 | // that Rust 1.13 produces a string mergeable section with a zero |
738 | // sh_entsize. Here we just accept it rather than being picky about it. |
739 | uint64_t entSize = sec.sh_entsize; |
740 | if (entSize == 0) |
741 | return false; |
742 | if (sec.sh_size % entSize) |
743 | fatal(toString(this) + ":(" + name + "): SHF_MERGE section size (" + |
744 | Twine(sec.sh_size) + ") must be a multiple of sh_entsize (" + |
745 | Twine(entSize) + ")" ); |
746 | |
747 | if (sec.sh_flags & SHF_WRITE) |
748 | fatal(toString(this) + ":(" + name + |
749 | "): writable SHF_MERGE section is not supported" ); |
750 | |
751 | return true; |
752 | } |
753 | |
754 | // This is for --just-symbols. |
755 | // |
756 | // --just-symbols is a very minor feature that allows you to link your |
757 | // output against other existing program, so that if you load both your |
758 | // program and the other program into memory, your output can refer the |
759 | // other program's symbols. |
760 | // |
761 | // When the option is given, we link "just symbols". The section table is |
762 | // initialized with null pointers. |
763 | template <class ELFT> void ObjFile<ELFT>::initializeJustSymbols() { |
764 | sections.resize(numELFShdrs); |
765 | } |
766 | |
767 | static bool isKnownSpecificSectionType(uint32_t t, uint32_t flags) { |
768 | if (SHT_LOUSER <= t && t <= SHT_HIUSER && !(flags & SHF_ALLOC)) |
769 | return true; |
770 | if (SHT_LOOS <= t && t <= SHT_HIOS && !(flags & SHF_OS_NONCONFORMING)) |
771 | return true; |
772 | // Allow all processor-specific types. This is different from GNU ld. |
773 | return SHT_LOPROC <= t && t <= SHT_HIPROC; |
774 | } |
775 | |
776 | template <class ELFT> |
777 | void ObjFile<ELFT>::initializeSections(bool ignoreComdats, |
778 | const llvm::object::ELFFile<ELFT> &obj) { |
779 | ArrayRef<Elf_Shdr> objSections = getELFShdrs<ELFT>(); |
780 | StringRef shstrtab = CHECK(obj.getSectionStringTable(objSections), this); |
781 | uint64_t size = objSections.size(); |
782 | SmallVector<ArrayRef<Elf_Word>, 0> selectedGroups; |
783 | for (size_t i = 0; i != size; ++i) { |
784 | if (this->sections[i] == &InputSection::discarded) |
785 | continue; |
786 | const Elf_Shdr &sec = objSections[i]; |
787 | const uint32_t type = sec.sh_type; |
788 | |
789 | // SHF_EXCLUDE'ed sections are discarded by the linker. However, |
790 | // if -r is given, we'll let the final link discard such sections. |
791 | // This is compatible with GNU. |
792 | if ((sec.sh_flags & SHF_EXCLUDE) && !config->relocatable) { |
793 | if (type == SHT_LLVM_CALL_GRAPH_PROFILE) |
794 | cgProfileSectionIndex = i; |
795 | if (type == SHT_LLVM_ADDRSIG) { |
796 | // We ignore the address-significance table if we know that the object |
797 | // file was created by objcopy or ld -r. This is because these tools |
798 | // will reorder the symbols in the symbol table, invalidating the data |
799 | // in the address-significance table, which refers to symbols by index. |
800 | if (sec.sh_link != 0) |
801 | this->addrsigSec = &sec; |
802 | else if (config->icf == ICFLevel::Safe) |
803 | warn(toString(this) + |
804 | ": --icf=safe conservatively ignores " |
805 | "SHT_LLVM_ADDRSIG [index " + |
806 | Twine(i) + |
807 | "] with sh_link=0 " |
808 | "(likely created using objcopy or ld -r)" ); |
809 | } |
810 | this->sections[i] = &InputSection::discarded; |
811 | continue; |
812 | } |
813 | |
814 | switch (type) { |
815 | case SHT_GROUP: { |
816 | if (!config->relocatable) |
817 | sections[i] = &InputSection::discarded; |
818 | StringRef signature = |
819 | cantFail(this->getELFSyms<ELFT>()[sec.sh_info].getName(stringTable)); |
820 | ArrayRef<Elf_Word> entries = |
821 | cantFail(obj.template getSectionContentsAsArray<Elf_Word>(sec)); |
822 | if ((entries[0] & GRP_COMDAT) == 0 || ignoreComdats || |
823 | symtab.comdatGroups.find(Val: CachedHashStringRef(signature))->second == |
824 | this) |
825 | selectedGroups.push_back(entries); |
826 | break; |
827 | } |
828 | case SHT_SYMTAB_SHNDX: |
829 | shndxTable = CHECK(obj.getSHNDXTable(sec, objSections), this); |
830 | break; |
831 | case SHT_SYMTAB: |
832 | case SHT_STRTAB: |
833 | case SHT_REL: |
834 | case SHT_RELA: |
835 | case SHT_CREL: |
836 | case SHT_NULL: |
837 | break; |
838 | case SHT_PROGBITS: |
839 | case SHT_NOTE: |
840 | case SHT_NOBITS: |
841 | case SHT_INIT_ARRAY: |
842 | case SHT_FINI_ARRAY: |
843 | case SHT_PREINIT_ARRAY: |
844 | this->sections[i] = |
845 | createInputSection(idx: i, sec, name: check(obj.getSectionName(sec, shstrtab))); |
846 | break; |
847 | case SHT_LLVM_LTO: |
848 | // Discard .llvm.lto in a relocatable link that does not use the bitcode. |
849 | // The concatenated output does not properly reflect the linking |
850 | // semantics. In addition, since we do not use the bitcode wrapper format, |
851 | // the concatenated raw bitcode would be invalid. |
852 | if (config->relocatable && !config->fatLTOObjects) { |
853 | sections[i] = &InputSection::discarded; |
854 | break; |
855 | } |
856 | [[fallthrough]]; |
857 | default: |
858 | this->sections[i] = |
859 | createInputSection(idx: i, sec, name: check(obj.getSectionName(sec, shstrtab))); |
860 | if (type == SHT_LLVM_SYMPART) |
861 | ctx.hasSympart.store(i: true, m: std::memory_order_relaxed); |
862 | else if (config->rejectMismatch && |
863 | !isKnownSpecificSectionType(type, sec.sh_flags)) |
864 | errorOrWarn(toString(this->sections[i]) + ": unknown section type 0x" + |
865 | Twine::utohexstr(Val: type)); |
866 | break; |
867 | } |
868 | } |
869 | |
870 | // We have a second loop. It is used to: |
871 | // 1) handle SHF_LINK_ORDER sections. |
872 | // 2) create relocation sections. In some cases the section header index of a |
873 | // relocation section may be smaller than that of the relocated section. In |
874 | // such cases, the relocation section would attempt to reference a target |
875 | // section that has not yet been created. For simplicity, delay creation of |
876 | // relocation sections until now. |
877 | for (size_t i = 0; i != size; ++i) { |
878 | if (this->sections[i] == &InputSection::discarded) |
879 | continue; |
880 | const Elf_Shdr &sec = objSections[i]; |
881 | |
882 | if (isStaticRelSecType(sec.sh_type)) { |
883 | // Find a relocation target section and associate this section with that. |
884 | // Target may have been discarded if it is in a different section group |
885 | // and the group is discarded, even though it's a violation of the spec. |
886 | // We handle that situation gracefully by discarding dangling relocation |
887 | // sections. |
888 | const uint32_t info = sec.sh_info; |
889 | InputSectionBase *s = getRelocTarget(idx: i, info); |
890 | if (!s) |
891 | continue; |
892 | |
893 | // ELF spec allows mergeable sections with relocations, but they are rare, |
894 | // and it is in practice hard to merge such sections by contents, because |
895 | // applying relocations at end of linking changes section contents. So, we |
896 | // simply handle such sections as non-mergeable ones. Degrading like this |
897 | // is acceptable because section merging is optional. |
898 | if (auto *ms = dyn_cast<MergeInputSection>(Val: s)) { |
899 | s = makeThreadLocal<InputSection>( |
900 | args&: ms->file, args&: ms->flags, args&: ms->type, args&: ms->addralign, |
901 | args: ms->contentMaybeDecompress(), args&: ms->name); |
902 | sections[info] = s; |
903 | } |
904 | |
905 | if (s->relSecIdx != 0) |
906 | error( |
907 | msg: toString(s) + |
908 | ": multiple relocation sections to one section are not supported" ); |
909 | s->relSecIdx = i; |
910 | |
911 | // Relocation sections are usually removed from the output, so return |
912 | // `nullptr` for the normal case. However, if -r or --emit-relocs is |
913 | // specified, we need to copy them to the output. (Some post link analysis |
914 | // tools specify --emit-relocs to obtain the information.) |
915 | if (config->copyRelocs) { |
916 | auto *isec = makeThreadLocal<InputSection>( |
917 | *this, sec, check(obj.getSectionName(sec, shstrtab))); |
918 | // If the relocated section is discarded (due to /DISCARD/ or |
919 | // --gc-sections), the relocation section should be discarded as well. |
920 | s->dependentSections.push_back(NewVal: isec); |
921 | sections[i] = isec; |
922 | } |
923 | continue; |
924 | } |
925 | |
926 | // A SHF_LINK_ORDER section with sh_link=0 is handled as if it did not have |
927 | // the flag. |
928 | if (!sec.sh_link || !(sec.sh_flags & SHF_LINK_ORDER)) |
929 | continue; |
930 | |
931 | InputSectionBase *linkSec = nullptr; |
932 | if (sec.sh_link < size) |
933 | linkSec = this->sections[sec.sh_link]; |
934 | if (!linkSec) |
935 | fatal(toString(this) + ": invalid sh_link index: " + Twine(sec.sh_link)); |
936 | |
937 | // A SHF_LINK_ORDER section is discarded if its linked-to section is |
938 | // discarded. |
939 | InputSection *isec = cast<InputSection>(this->sections[i]); |
940 | linkSec->dependentSections.push_back(NewVal: isec); |
941 | if (!isa<InputSection>(Val: linkSec)) |
942 | error(msg: "a section " + isec->name + |
943 | " with SHF_LINK_ORDER should not refer a non-regular section: " + |
944 | toString(linkSec)); |
945 | } |
946 | |
947 | for (ArrayRef<Elf_Word> entries : selectedGroups) |
948 | handleSectionGroup<ELFT>(this->sections, entries); |
949 | } |
950 | |
951 | // Read the following info from the .note.gnu.property section and write it to |
952 | // the corresponding fields in `ObjFile`: |
953 | // - Feature flags (32 bits) representing x86 or AArch64 features for |
954 | // hardware-assisted call flow control; |
955 | // - AArch64 PAuth ABI core info (16 bytes). |
956 | template <class ELFT> |
957 | void readGnuProperty(const InputSection &sec, ObjFile<ELFT> &f) { |
958 | using Elf_Nhdr = typename ELFT::Nhdr; |
959 | using Elf_Note = typename ELFT::Note; |
960 | |
961 | ArrayRef<uint8_t> data = sec.content(); |
962 | auto reportFatal = [&](const uint8_t *place, const Twine &msg) { |
963 | fatal(msg: toString(f: sec.file) + ":(" + sec.name + "+0x" + |
964 | Twine::utohexstr(Val: place - sec.content().data()) + "): " + msg); |
965 | }; |
966 | while (!data.empty()) { |
967 | // Read one NOTE record. |
968 | auto *nhdr = reinterpret_cast<const Elf_Nhdr *>(data.data()); |
969 | if (data.size() < sizeof(Elf_Nhdr) || |
970 | data.size() < nhdr->getSize(sec.addralign)) |
971 | reportFatal(data.data(), "data is too short" ); |
972 | |
973 | Elf_Note note(*nhdr); |
974 | if (nhdr->n_type != NT_GNU_PROPERTY_TYPE_0 || note.getName() != "GNU" ) { |
975 | data = data.slice(nhdr->getSize(sec.addralign)); |
976 | continue; |
977 | } |
978 | |
979 | uint32_t featureAndType = config->emachine == EM_AARCH64 |
980 | ? GNU_PROPERTY_AARCH64_FEATURE_1_AND |
981 | : GNU_PROPERTY_X86_FEATURE_1_AND; |
982 | |
983 | // Read a body of a NOTE record, which consists of type-length-value fields. |
984 | ArrayRef<uint8_t> desc = note.getDesc(sec.addralign); |
985 | while (!desc.empty()) { |
986 | const uint8_t *place = desc.data(); |
987 | if (desc.size() < 8) |
988 | reportFatal(place, "program property is too short" ); |
989 | uint32_t type = read32<ELFT::Endianness>(desc.data()); |
990 | uint32_t size = read32<ELFT::Endianness>(desc.data() + 4); |
991 | desc = desc.slice(N: 8); |
992 | if (desc.size() < size) |
993 | reportFatal(place, "program property is too short" ); |
994 | |
995 | if (type == featureAndType) { |
996 | // We found a FEATURE_1_AND field. There may be more than one of these |
997 | // in a .note.gnu.property section, for a relocatable object we |
998 | // accumulate the bits set. |
999 | if (size < 4) |
1000 | reportFatal(place, "FEATURE_1_AND entry is too short" ); |
1001 | f.andFeatures |= read32<ELFT::Endianness>(desc.data()); |
1002 | } else if (config->emachine == EM_AARCH64 && |
1003 | type == GNU_PROPERTY_AARCH64_FEATURE_PAUTH) { |
1004 | if (!f.aarch64PauthAbiCoreInfo.empty()) { |
1005 | reportFatal(data.data(), |
1006 | "multiple GNU_PROPERTY_AARCH64_FEATURE_PAUTH entries are " |
1007 | "not supported" ); |
1008 | } else if (size != 16) { |
1009 | reportFatal(data.data(), "GNU_PROPERTY_AARCH64_FEATURE_PAUTH entry " |
1010 | "is invalid: expected 16 bytes, but got " + |
1011 | Twine(size)); |
1012 | } |
1013 | f.aarch64PauthAbiCoreInfo = desc; |
1014 | } |
1015 | |
1016 | // Padding is present in the note descriptor, if necessary. |
1017 | desc = desc.slice(alignTo<(ELFT::Is64Bits ? 8 : 4)>(size)); |
1018 | } |
1019 | |
1020 | // Go to next NOTE record to look for more FEATURE_1_AND descriptions. |
1021 | data = data.slice(nhdr->getSize(sec.addralign)); |
1022 | } |
1023 | } |
1024 | |
1025 | template <class ELFT> |
1026 | InputSectionBase *ObjFile<ELFT>::getRelocTarget(uint32_t idx, uint32_t info) { |
1027 | if (info < this->sections.size()) { |
1028 | InputSectionBase *target = this->sections[info]; |
1029 | |
1030 | // Strictly speaking, a relocation section must be included in the |
1031 | // group of the section it relocates. However, LLVM 3.3 and earlier |
1032 | // would fail to do so, so we gracefully handle that case. |
1033 | if (target == &InputSection::discarded) |
1034 | return nullptr; |
1035 | |
1036 | if (target != nullptr) |
1037 | return target; |
1038 | } |
1039 | |
1040 | error(toString(this) + Twine(": relocation section (index " ) + Twine(idx) + |
1041 | ") has invalid sh_info (" + Twine(info) + ")" ); |
1042 | return nullptr; |
1043 | } |
1044 | |
1045 | // The function may be called concurrently for different input files. For |
1046 | // allocation, prefer makeThreadLocal which does not require holding a lock. |
1047 | template <class ELFT> |
1048 | InputSectionBase *ObjFile<ELFT>::createInputSection(uint32_t idx, |
1049 | const Elf_Shdr &sec, |
1050 | StringRef name) { |
1051 | if (name.starts_with(Prefix: ".n" )) { |
1052 | // The GNU linker uses .note.GNU-stack section as a marker indicating |
1053 | // that the code in the object file does not expect that the stack is |
1054 | // executable (in terms of NX bit). If all input files have the marker, |
1055 | // the GNU linker adds a PT_GNU_STACK segment to tells the loader to |
1056 | // make the stack non-executable. Most object files have this section as |
1057 | // of 2017. |
1058 | // |
1059 | // But making the stack non-executable is a norm today for security |
1060 | // reasons. Failure to do so may result in a serious security issue. |
1061 | // Therefore, we make LLD always add PT_GNU_STACK unless it is |
1062 | // explicitly told to do otherwise (by -z execstack). Because the stack |
1063 | // executable-ness is controlled solely by command line options, |
1064 | // .note.GNU-stack sections are simply ignored. |
1065 | if (name == ".note.GNU-stack" ) |
1066 | return &InputSection::discarded; |
1067 | |
1068 | // Object files that use processor features such as Intel Control-Flow |
1069 | // Enforcement (CET) or AArch64 Branch Target Identification BTI, use a |
1070 | // .note.gnu.property section containing a bitfield of feature bits like the |
1071 | // GNU_PROPERTY_X86_FEATURE_1_IBT flag. Read a bitmap containing the flag. |
1072 | // |
1073 | // Since we merge bitmaps from multiple object files to create a new |
1074 | // .note.gnu.property containing a single AND'ed bitmap, we discard an input |
1075 | // file's .note.gnu.property section. |
1076 | if (name == ".note.gnu.property" ) { |
1077 | readGnuProperty<ELFT>(InputSection(*this, sec, name), *this); |
1078 | return &InputSection::discarded; |
1079 | } |
1080 | |
1081 | // Split stacks is a feature to support a discontiguous stack, |
1082 | // commonly used in the programming language Go. For the details, |
1083 | // see https://gcc.gnu.org/wiki/SplitStacks. An object file compiled |
1084 | // for split stack will include a .note.GNU-split-stack section. |
1085 | if (name == ".note.GNU-split-stack" ) { |
1086 | if (config->relocatable) { |
1087 | error( |
1088 | msg: "cannot mix split-stack and non-split-stack in a relocatable link" ); |
1089 | return &InputSection::discarded; |
1090 | } |
1091 | this->splitStack = true; |
1092 | return &InputSection::discarded; |
1093 | } |
1094 | |
1095 | // An object file compiled for split stack, but where some of the |
1096 | // functions were compiled with the no_split_stack_attribute will |
1097 | // include a .note.GNU-no-split-stack section. |
1098 | if (name == ".note.GNU-no-split-stack" ) { |
1099 | this->someNoSplitStack = true; |
1100 | return &InputSection::discarded; |
1101 | } |
1102 | |
1103 | // Strip existing .note.gnu.build-id sections so that the output won't have |
1104 | // more than one build-id. This is not usually a problem because input |
1105 | // object files normally don't have .build-id sections, but you can create |
1106 | // such files by "ld.{bfd,gold,lld} -r --build-id", and we want to guard |
1107 | // against it. |
1108 | if (name == ".note.gnu.build-id" ) |
1109 | return &InputSection::discarded; |
1110 | } |
1111 | |
1112 | // The linker merges EH (exception handling) frames and creates a |
1113 | // .eh_frame_hdr section for runtime. So we handle them with a special |
1114 | // class. For relocatable outputs, they are just passed through. |
1115 | if (name == ".eh_frame" && !config->relocatable) |
1116 | return makeThreadLocal<EhInputSection>(*this, sec, name); |
1117 | |
1118 | if ((sec.sh_flags & SHF_MERGE) && shouldMerge(sec, name)) |
1119 | return makeThreadLocal<MergeInputSection>(*this, sec, name); |
1120 | return makeThreadLocal<InputSection>(*this, sec, name); |
1121 | } |
1122 | |
1123 | // Initialize symbols. symbols is a parallel array to the corresponding ELF |
1124 | // symbol table. |
1125 | template <class ELFT> |
1126 | void ObjFile<ELFT>::initializeSymbols(const object::ELFFile<ELFT> &obj) { |
1127 | ArrayRef<Elf_Sym> eSyms = this->getELFSyms<ELFT>(); |
1128 | if (numSymbols == 0) { |
1129 | numSymbols = eSyms.size(); |
1130 | symbols = std::make_unique<Symbol *[]>(numSymbols); |
1131 | } |
1132 | |
1133 | // Some entries have been filled by LazyObjFile. |
1134 | for (size_t i = firstGlobal, end = eSyms.size(); i != end; ++i) |
1135 | if (!symbols[i]) |
1136 | symbols[i] = symtab.insert(CHECK(eSyms[i].getName(stringTable), this)); |
1137 | |
1138 | // Perform symbol resolution on non-local symbols. |
1139 | SmallVector<unsigned, 32> undefineds; |
1140 | for (size_t i = firstGlobal, end = eSyms.size(); i != end; ++i) { |
1141 | const Elf_Sym &eSym = eSyms[i]; |
1142 | uint32_t secIdx = eSym.st_shndx; |
1143 | if (secIdx == SHN_UNDEF) { |
1144 | undefineds.push_back(Elt: i); |
1145 | continue; |
1146 | } |
1147 | |
1148 | uint8_t binding = eSym.getBinding(); |
1149 | uint8_t stOther = eSym.st_other; |
1150 | uint8_t type = eSym.getType(); |
1151 | uint64_t value = eSym.st_value; |
1152 | uint64_t size = eSym.st_size; |
1153 | |
1154 | Symbol *sym = symbols[i]; |
1155 | sym->isUsedInRegularObj = true; |
1156 | if (LLVM_UNLIKELY(eSym.st_shndx == SHN_COMMON)) { |
1157 | if (value == 0 || value >= UINT32_MAX) |
1158 | fatal(toString(this) + ": common symbol '" + sym->getName() + |
1159 | "' has invalid alignment: " + Twine(value)); |
1160 | hasCommonSyms = true; |
1161 | sym->resolve( |
1162 | other: CommonSymbol{this, StringRef(), binding, stOther, type, value, size}); |
1163 | continue; |
1164 | } |
1165 | |
1166 | // Handle global defined symbols. Defined::section will be set in postParse. |
1167 | sym->resolve(other: Defined{this, StringRef(), binding, stOther, type, value, size, |
1168 | nullptr}); |
1169 | } |
1170 | |
1171 | // Undefined symbols (excluding those defined relative to non-prevailing |
1172 | // sections) can trigger recursive extract. Process defined symbols first so |
1173 | // that the relative order between a defined symbol and an undefined symbol |
1174 | // does not change the symbol resolution behavior. In addition, a set of |
1175 | // interconnected symbols will all be resolved to the same file, instead of |
1176 | // being resolved to different files. |
1177 | for (unsigned i : undefineds) { |
1178 | const Elf_Sym &eSym = eSyms[i]; |
1179 | Symbol *sym = symbols[i]; |
1180 | sym->resolve(other: Undefined{this, StringRef(), eSym.getBinding(), eSym.st_other, |
1181 | eSym.getType()}); |
1182 | sym->isUsedInRegularObj = true; |
1183 | sym->referenced = true; |
1184 | } |
1185 | } |
1186 | |
1187 | template <class ELFT> |
1188 | void ObjFile<ELFT>::initSectionsAndLocalSyms(bool ignoreComdats) { |
1189 | if (!justSymbols) |
1190 | initializeSections(ignoreComdats, obj: getObj()); |
1191 | |
1192 | if (!firstGlobal) |
1193 | return; |
1194 | SymbolUnion *locals = makeThreadLocalN<SymbolUnion>(firstGlobal); |
1195 | memset(locals, 0, sizeof(SymbolUnion) * firstGlobal); |
1196 | |
1197 | ArrayRef<Elf_Sym> eSyms = this->getELFSyms<ELFT>(); |
1198 | for (size_t i = 0, end = firstGlobal; i != end; ++i) { |
1199 | const Elf_Sym &eSym = eSyms[i]; |
1200 | uint32_t secIdx = eSym.st_shndx; |
1201 | if (LLVM_UNLIKELY(secIdx == SHN_XINDEX)) |
1202 | secIdx = check(getExtendedSymbolTableIndex<ELFT>(eSym, i, shndxTable)); |
1203 | else if (secIdx >= SHN_LORESERVE) |
1204 | secIdx = 0; |
1205 | if (LLVM_UNLIKELY(secIdx >= sections.size())) |
1206 | fatal(toString(this) + ": invalid section index: " + Twine(secIdx)); |
1207 | if (LLVM_UNLIKELY(eSym.getBinding() != STB_LOCAL)) |
1208 | error(toString(this) + ": non-local symbol (" + Twine(i) + |
1209 | ") found at index < .symtab's sh_info (" + Twine(end) + ")" ); |
1210 | |
1211 | InputSectionBase *sec = sections[secIdx]; |
1212 | uint8_t type = eSym.getType(); |
1213 | if (type == STT_FILE) |
1214 | sourceFile = CHECK(eSym.getName(stringTable), this); |
1215 | if (LLVM_UNLIKELY(stringTable.size() <= eSym.st_name)) |
1216 | fatal(toString(this) + ": invalid symbol name offset" ); |
1217 | StringRef name(stringTable.data() + eSym.st_name); |
1218 | |
1219 | symbols[i] = reinterpret_cast<Symbol *>(locals + i); |
1220 | if (eSym.st_shndx == SHN_UNDEF || sec == &InputSection::discarded) |
1221 | new (symbols[i]) Undefined(this, name, STB_LOCAL, eSym.st_other, type, |
1222 | /*discardedSecIdx=*/secIdx); |
1223 | else |
1224 | new (symbols[i]) Defined(this, name, STB_LOCAL, eSym.st_other, type, |
1225 | eSym.st_value, eSym.st_size, sec); |
1226 | symbols[i]->partition = 1; |
1227 | symbols[i]->isUsedInRegularObj = true; |
1228 | } |
1229 | } |
1230 | |
1231 | // Called after all ObjFile::parse is called for all ObjFiles. This checks |
1232 | // duplicate symbols and may do symbol property merge in the future. |
1233 | template <class ELFT> void ObjFile<ELFT>::postParse() { |
1234 | static std::mutex mu; |
1235 | ArrayRef<Elf_Sym> eSyms = this->getELFSyms<ELFT>(); |
1236 | for (size_t i = firstGlobal, end = eSyms.size(); i != end; ++i) { |
1237 | const Elf_Sym &eSym = eSyms[i]; |
1238 | Symbol &sym = *symbols[i]; |
1239 | uint32_t secIdx = eSym.st_shndx; |
1240 | uint8_t binding = eSym.getBinding(); |
1241 | if (LLVM_UNLIKELY(binding != STB_GLOBAL && binding != STB_WEAK && |
1242 | binding != STB_GNU_UNIQUE)) |
1243 | errorOrWarn(toString(this) + ": symbol (" + Twine(i) + |
1244 | ") has invalid binding: " + Twine((int)binding)); |
1245 | |
1246 | // st_value of STT_TLS represents the assigned offset, not the actual |
1247 | // address which is used by STT_FUNC and STT_OBJECT. STT_TLS symbols can |
1248 | // only be referenced by special TLS relocations. It is usually an error if |
1249 | // a STT_TLS symbol is replaced by a non-STT_TLS symbol, vice versa. |
1250 | if (LLVM_UNLIKELY(sym.isTls()) && eSym.getType() != STT_TLS && |
1251 | eSym.getType() != STT_NOTYPE) |
1252 | errorOrWarn("TLS attribute mismatch: " + toString(sym) + "\n>>> in " + |
1253 | toString(f: sym.file) + "\n>>> in " + toString(this)); |
1254 | |
1255 | // Handle non-COMMON defined symbol below. !sym.file allows a symbol |
1256 | // assignment to redefine a symbol without an error. |
1257 | if (!sym.file || !sym.isDefined() || secIdx == SHN_UNDEF || |
1258 | secIdx == SHN_COMMON) |
1259 | continue; |
1260 | |
1261 | if (LLVM_UNLIKELY(secIdx == SHN_XINDEX)) |
1262 | secIdx = check(getExtendedSymbolTableIndex<ELFT>(eSym, i, shndxTable)); |
1263 | else if (secIdx >= SHN_LORESERVE) |
1264 | secIdx = 0; |
1265 | if (LLVM_UNLIKELY(secIdx >= sections.size())) |
1266 | fatal(toString(this) + ": invalid section index: " + Twine(secIdx)); |
1267 | InputSectionBase *sec = sections[secIdx]; |
1268 | if (sec == &InputSection::discarded) { |
1269 | if (sym.traced) { |
1270 | printTraceSymbol(sym: Undefined{this, sym.getName(), sym.binding, |
1271 | sym.stOther, sym.type, secIdx}, |
1272 | name: sym.getName()); |
1273 | } |
1274 | if (sym.file == this) { |
1275 | std::lock_guard<std::mutex> lock(mu); |
1276 | ctx.nonPrevailingSyms.emplace_back(Args: &sym, Args&: secIdx); |
1277 | } |
1278 | continue; |
1279 | } |
1280 | |
1281 | if (sym.file == this) { |
1282 | cast<Defined>(Val&: sym).section = sec; |
1283 | continue; |
1284 | } |
1285 | |
1286 | if (sym.binding == STB_WEAK || binding == STB_WEAK) |
1287 | continue; |
1288 | std::lock_guard<std::mutex> lock(mu); |
1289 | ctx.duplicates.push_back(Elt: {&sym, this, sec, eSym.st_value}); |
1290 | } |
1291 | } |
1292 | |
1293 | // The handling of tentative definitions (COMMON symbols) in archives is murky. |
1294 | // A tentative definition will be promoted to a global definition if there are |
1295 | // no non-tentative definitions to dominate it. When we hold a tentative |
1296 | // definition to a symbol and are inspecting archive members for inclusion |
1297 | // there are 2 ways we can proceed: |
1298 | // |
1299 | // 1) Consider the tentative definition a 'real' definition (ie promotion from |
1300 | // tentative to real definition has already happened) and not inspect |
1301 | // archive members for Global/Weak definitions to replace the tentative |
1302 | // definition. An archive member would only be included if it satisfies some |
1303 | // other undefined symbol. This is the behavior Gold uses. |
1304 | // |
1305 | // 2) Consider the tentative definition as still undefined (ie the promotion to |
1306 | // a real definition happens only after all symbol resolution is done). |
1307 | // The linker searches archive members for STB_GLOBAL definitions to |
1308 | // replace the tentative definition with. This is the behavior used by |
1309 | // GNU ld. |
1310 | // |
1311 | // The second behavior is inherited from SysVR4, which based it on the FORTRAN |
1312 | // COMMON BLOCK model. This behavior is needed for proper initialization in old |
1313 | // (pre F90) FORTRAN code that is packaged into an archive. |
1314 | // |
1315 | // The following functions search archive members for definitions to replace |
1316 | // tentative definitions (implementing behavior 2). |
1317 | static bool isBitcodeNonCommonDef(MemoryBufferRef mb, StringRef symName, |
1318 | StringRef archiveName) { |
1319 | IRSymtabFile symtabFile = check(e: readIRSymtab(MBRef: mb)); |
1320 | for (const irsymtab::Reader::SymbolRef &sym : |
1321 | symtabFile.TheReader.symbols()) { |
1322 | if (sym.isGlobal() && sym.getName() == symName) |
1323 | return !sym.isUndefined() && !sym.isWeak() && !sym.isCommon(); |
1324 | } |
1325 | return false; |
1326 | } |
1327 | |
1328 | template <class ELFT> |
1329 | static bool isNonCommonDef(ELFKind ekind, MemoryBufferRef mb, StringRef symName, |
1330 | StringRef archiveName) { |
1331 | ObjFile<ELFT> *obj = make<ObjFile<ELFT>>(ekind, mb, archiveName); |
1332 | obj->init(); |
1333 | StringRef stringtable = obj->getStringTable(); |
1334 | |
1335 | for (auto sym : obj->template getGlobalELFSyms<ELFT>()) { |
1336 | Expected<StringRef> name = sym.getName(stringtable); |
1337 | if (name && name.get() == symName) |
1338 | return sym.isDefined() && sym.getBinding() == STB_GLOBAL && |
1339 | !sym.isCommon(); |
1340 | } |
1341 | return false; |
1342 | } |
1343 | |
1344 | static bool isNonCommonDef(MemoryBufferRef mb, StringRef symName, |
1345 | StringRef archiveName) { |
1346 | switch (getELFKind(mb, archiveName)) { |
1347 | case ELF32LEKind: |
1348 | return isNonCommonDef<ELF32LE>(ekind: ELF32LEKind, mb, symName, archiveName); |
1349 | case ELF32BEKind: |
1350 | return isNonCommonDef<ELF32BE>(ekind: ELF32BEKind, mb, symName, archiveName); |
1351 | case ELF64LEKind: |
1352 | return isNonCommonDef<ELF64LE>(ekind: ELF64LEKind, mb, symName, archiveName); |
1353 | case ELF64BEKind: |
1354 | return isNonCommonDef<ELF64BE>(ekind: ELF64BEKind, mb, symName, archiveName); |
1355 | default: |
1356 | llvm_unreachable("getELFKind" ); |
1357 | } |
1358 | } |
1359 | |
1360 | unsigned SharedFile::vernauxNum; |
1361 | |
1362 | SharedFile::SharedFile(MemoryBufferRef m, StringRef defaultSoName) |
1363 | : ELFFileBase(SharedKind, getELFKind(mb: m, archiveName: "" ), m), soName(defaultSoName), |
1364 | isNeeded(!config->asNeeded) {} |
1365 | |
1366 | // Parse the version definitions in the object file if present, and return a |
1367 | // vector whose nth element contains a pointer to the Elf_Verdef for version |
1368 | // identifier n. Version identifiers that are not definitions map to nullptr. |
1369 | template <typename ELFT> |
1370 | static SmallVector<const void *, 0> |
1371 | parseVerdefs(const uint8_t *base, const typename ELFT::Shdr *sec) { |
1372 | if (!sec) |
1373 | return {}; |
1374 | |
1375 | // Build the Verdefs array by following the chain of Elf_Verdef objects |
1376 | // from the start of the .gnu.version_d section. |
1377 | SmallVector<const void *, 0> verdefs; |
1378 | const uint8_t *verdef = base + sec->sh_offset; |
1379 | for (unsigned i = 0, e = sec->sh_info; i != e; ++i) { |
1380 | auto *curVerdef = reinterpret_cast<const typename ELFT::Verdef *>(verdef); |
1381 | verdef += curVerdef->vd_next; |
1382 | unsigned verdefIndex = curVerdef->vd_ndx; |
1383 | if (verdefIndex >= verdefs.size()) |
1384 | verdefs.resize(N: verdefIndex + 1); |
1385 | verdefs[verdefIndex] = curVerdef; |
1386 | } |
1387 | return verdefs; |
1388 | } |
1389 | |
1390 | // Parse SHT_GNU_verneed to properly set the name of a versioned undefined |
1391 | // symbol. We detect fatal issues which would cause vulnerabilities, but do not |
1392 | // implement sophisticated error checking like in llvm-readobj because the value |
1393 | // of such diagnostics is low. |
1394 | template <typename ELFT> |
1395 | std::vector<uint32_t> SharedFile::parseVerneed(const ELFFile<ELFT> &obj, |
1396 | const typename ELFT::Shdr *sec) { |
1397 | if (!sec) |
1398 | return {}; |
1399 | std::vector<uint32_t> verneeds; |
1400 | ArrayRef<uint8_t> data = CHECK(obj.getSectionContents(*sec), this); |
1401 | const uint8_t *verneedBuf = data.begin(); |
1402 | for (unsigned i = 0; i != sec->sh_info; ++i) { |
1403 | if (verneedBuf + sizeof(typename ELFT::Verneed) > data.end()) |
1404 | fatal(msg: toString(f: this) + " has an invalid Verneed" ); |
1405 | auto *vn = reinterpret_cast<const typename ELFT::Verneed *>(verneedBuf); |
1406 | const uint8_t *vernauxBuf = verneedBuf + vn->vn_aux; |
1407 | for (unsigned j = 0; j != vn->vn_cnt; ++j) { |
1408 | if (vernauxBuf + sizeof(typename ELFT::Vernaux) > data.end()) |
1409 | fatal(msg: toString(f: this) + " has an invalid Vernaux" ); |
1410 | auto *aux = reinterpret_cast<const typename ELFT::Vernaux *>(vernauxBuf); |
1411 | if (aux->vna_name >= this->stringTable.size()) |
1412 | fatal(msg: toString(f: this) + " has a Vernaux with an invalid vna_name" ); |
1413 | uint16_t version = aux->vna_other & VERSYM_VERSION; |
1414 | if (version >= verneeds.size()) |
1415 | verneeds.resize(new_size: version + 1); |
1416 | verneeds[version] = aux->vna_name; |
1417 | vernauxBuf += aux->vna_next; |
1418 | } |
1419 | verneedBuf += vn->vn_next; |
1420 | } |
1421 | return verneeds; |
1422 | } |
1423 | |
1424 | // We do not usually care about alignments of data in shared object |
1425 | // files because the loader takes care of it. However, if we promote a |
1426 | // DSO symbol to point to .bss due to copy relocation, we need to keep |
1427 | // the original alignment requirements. We infer it in this function. |
1428 | template <typename ELFT> |
1429 | static uint64_t getAlignment(ArrayRef<typename ELFT::Shdr> sections, |
1430 | const typename ELFT::Sym &sym) { |
1431 | uint64_t ret = UINT64_MAX; |
1432 | if (sym.st_value) |
1433 | ret = 1ULL << llvm::countr_zero(Val: (uint64_t)sym.st_value); |
1434 | if (0 < sym.st_shndx && sym.st_shndx < sections.size()) |
1435 | ret = std::min<uint64_t>(ret, sections[sym.st_shndx].sh_addralign); |
1436 | return (ret > UINT32_MAX) ? 0 : ret; |
1437 | } |
1438 | |
1439 | // Fully parse the shared object file. |
1440 | // |
1441 | // This function parses symbol versions. If a DSO has version information, |
1442 | // the file has a ".gnu.version_d" section which contains symbol version |
1443 | // definitions. Each symbol is associated to one version through a table in |
1444 | // ".gnu.version" section. That table is a parallel array for the symbol |
1445 | // table, and each table entry contains an index in ".gnu.version_d". |
1446 | // |
1447 | // The special index 0 is reserved for VERF_NDX_LOCAL and 1 is for |
1448 | // VER_NDX_GLOBAL. There's no table entry for these special versions in |
1449 | // ".gnu.version_d". |
1450 | // |
1451 | // The file format for symbol versioning is perhaps a bit more complicated |
1452 | // than necessary, but you can easily understand the code if you wrap your |
1453 | // head around the data structure described above. |
1454 | template <class ELFT> void SharedFile::parse() { |
1455 | using Elf_Dyn = typename ELFT::Dyn; |
1456 | using Elf_Shdr = typename ELFT::Shdr; |
1457 | using Elf_Sym = typename ELFT::Sym; |
1458 | using Elf_Verdef = typename ELFT::Verdef; |
1459 | using Elf_Versym = typename ELFT::Versym; |
1460 | |
1461 | ArrayRef<Elf_Dyn> dynamicTags; |
1462 | const ELFFile<ELFT> obj = this->getObj<ELFT>(); |
1463 | ArrayRef<Elf_Shdr> sections = getELFShdrs<ELFT>(); |
1464 | |
1465 | const Elf_Shdr *versymSec = nullptr; |
1466 | const Elf_Shdr *verdefSec = nullptr; |
1467 | const Elf_Shdr *verneedSec = nullptr; |
1468 | |
1469 | // Search for .dynsym, .dynamic, .symtab, .gnu.version and .gnu.version_d. |
1470 | for (const Elf_Shdr &sec : sections) { |
1471 | switch (sec.sh_type) { |
1472 | default: |
1473 | continue; |
1474 | case SHT_DYNAMIC: |
1475 | dynamicTags = |
1476 | CHECK(obj.template getSectionContentsAsArray<Elf_Dyn>(sec), this); |
1477 | break; |
1478 | case SHT_GNU_versym: |
1479 | versymSec = &sec; |
1480 | break; |
1481 | case SHT_GNU_verdef: |
1482 | verdefSec = &sec; |
1483 | break; |
1484 | case SHT_GNU_verneed: |
1485 | verneedSec = &sec; |
1486 | break; |
1487 | } |
1488 | } |
1489 | |
1490 | if (versymSec && numELFSyms == 0) { |
1491 | error(msg: "SHT_GNU_versym should be associated with symbol table" ); |
1492 | return; |
1493 | } |
1494 | |
1495 | // Search for a DT_SONAME tag to initialize this->soName. |
1496 | for (const Elf_Dyn &dyn : dynamicTags) { |
1497 | if (dyn.d_tag == DT_NEEDED) { |
1498 | uint64_t val = dyn.getVal(); |
1499 | if (val >= this->stringTable.size()) |
1500 | fatal(msg: toString(f: this) + ": invalid DT_NEEDED entry" ); |
1501 | dtNeeded.push_back(Elt: this->stringTable.data() + val); |
1502 | } else if (dyn.d_tag == DT_SONAME) { |
1503 | uint64_t val = dyn.getVal(); |
1504 | if (val >= this->stringTable.size()) |
1505 | fatal(msg: toString(f: this) + ": invalid DT_SONAME entry" ); |
1506 | soName = this->stringTable.data() + val; |
1507 | } |
1508 | } |
1509 | |
1510 | // DSOs are uniquified not by filename but by soname. |
1511 | DenseMap<CachedHashStringRef, SharedFile *>::iterator it; |
1512 | bool wasInserted; |
1513 | std::tie(args&: it, args&: wasInserted) = |
1514 | symtab.soNames.try_emplace(Key: CachedHashStringRef(soName), Args: this); |
1515 | |
1516 | // If a DSO appears more than once on the command line with and without |
1517 | // --as-needed, --no-as-needed takes precedence over --as-needed because a |
1518 | // user can add an extra DSO with --no-as-needed to force it to be added to |
1519 | // the dependency list. |
1520 | it->second->isNeeded |= isNeeded; |
1521 | if (!wasInserted) |
1522 | return; |
1523 | |
1524 | ctx.sharedFiles.push_back(Elt: this); |
1525 | |
1526 | verdefs = parseVerdefs<ELFT>(obj.base(), verdefSec); |
1527 | std::vector<uint32_t> verneeds = parseVerneed<ELFT>(obj, verneedSec); |
1528 | |
1529 | // Parse ".gnu.version" section which is a parallel array for the symbol |
1530 | // table. If a given file doesn't have a ".gnu.version" section, we use |
1531 | // VER_NDX_GLOBAL. |
1532 | size_t size = numELFSyms - firstGlobal; |
1533 | std::vector<uint16_t> versyms(size, VER_NDX_GLOBAL); |
1534 | if (versymSec) { |
1535 | ArrayRef<Elf_Versym> versym = |
1536 | CHECK(obj.template getSectionContentsAsArray<Elf_Versym>(*versymSec), |
1537 | this) |
1538 | .slice(firstGlobal); |
1539 | for (size_t i = 0; i < size; ++i) |
1540 | versyms[i] = versym[i].vs_index; |
1541 | } |
1542 | |
1543 | // System libraries can have a lot of symbols with versions. Using a |
1544 | // fixed buffer for computing the versions name (foo@ver) can save a |
1545 | // lot of allocations. |
1546 | SmallString<0> versionedNameBuffer; |
1547 | |
1548 | // Add symbols to the symbol table. |
1549 | ArrayRef<Elf_Sym> syms = this->getGlobalELFSyms<ELFT>(); |
1550 | for (size_t i = 0, e = syms.size(); i != e; ++i) { |
1551 | const Elf_Sym &sym = syms[i]; |
1552 | |
1553 | // ELF spec requires that all local symbols precede weak or global |
1554 | // symbols in each symbol table, and the index of first non-local symbol |
1555 | // is stored to sh_info. If a local symbol appears after some non-local |
1556 | // symbol, that's a violation of the spec. |
1557 | StringRef name = CHECK(sym.getName(stringTable), this); |
1558 | if (sym.getBinding() == STB_LOCAL) { |
1559 | errorOrWarn(msg: toString(f: this) + ": invalid local symbol '" + name + |
1560 | "' in global part of symbol table" ); |
1561 | continue; |
1562 | } |
1563 | |
1564 | const uint16_t ver = versyms[i], idx = ver & ~VERSYM_HIDDEN; |
1565 | if (sym.isUndefined()) { |
1566 | // For unversioned undefined symbols, VER_NDX_GLOBAL makes more sense but |
1567 | // as of binutils 2.34, GNU ld produces VER_NDX_LOCAL. |
1568 | if (ver != VER_NDX_LOCAL && ver != VER_NDX_GLOBAL) { |
1569 | if (idx >= verneeds.size()) { |
1570 | error(msg: "corrupt input file: version need index " + Twine(idx) + |
1571 | " for symbol " + name + " is out of bounds\n>>> defined in " + |
1572 | toString(f: this)); |
1573 | continue; |
1574 | } |
1575 | StringRef verName = stringTable.data() + verneeds[idx]; |
1576 | versionedNameBuffer.clear(); |
1577 | name = saver().save( |
1578 | S: (name + "@" + verName).toStringRef(Out&: versionedNameBuffer)); |
1579 | } |
1580 | Symbol *s = symtab.addSymbol( |
1581 | newSym: Undefined{this, name, sym.getBinding(), sym.st_other, sym.getType()}); |
1582 | s->exportDynamic = true; |
1583 | if (sym.getBinding() != STB_WEAK && |
1584 | config->unresolvedSymbolsInShlib != UnresolvedPolicy::Ignore) |
1585 | requiredSymbols.push_back(Elt: s); |
1586 | continue; |
1587 | } |
1588 | |
1589 | if (ver == VER_NDX_LOCAL || |
1590 | (ver != VER_NDX_GLOBAL && idx >= verdefs.size())) { |
1591 | // In GNU ld < 2.31 (before 3be08ea4728b56d35e136af4e6fd3086ade17764), the |
1592 | // MIPS port puts _gp_disp symbol into DSO files and incorrectly assigns |
1593 | // VER_NDX_LOCAL. Workaround this bug. |
1594 | if (config->emachine == EM_MIPS && name == "_gp_disp" ) |
1595 | continue; |
1596 | error(msg: "corrupt input file: version definition index " + Twine(idx) + |
1597 | " for symbol " + name + " is out of bounds\n>>> defined in " + |
1598 | toString(f: this)); |
1599 | continue; |
1600 | } |
1601 | |
1602 | uint32_t alignment = getAlignment<ELFT>(sections, sym); |
1603 | if (ver == idx) { |
1604 | auto *s = symtab.addSymbol( |
1605 | newSym: SharedSymbol{*this, name, sym.getBinding(), sym.st_other, |
1606 | sym.getType(), sym.st_value, sym.st_size, alignment}); |
1607 | s->dsoDefined = true; |
1608 | if (s->file == this) |
1609 | s->versionId = ver; |
1610 | } |
1611 | |
1612 | // Also add the symbol with the versioned name to handle undefined symbols |
1613 | // with explicit versions. |
1614 | if (ver == VER_NDX_GLOBAL) |
1615 | continue; |
1616 | |
1617 | StringRef verName = |
1618 | stringTable.data() + |
1619 | reinterpret_cast<const Elf_Verdef *>(verdefs[idx])->getAux()->vda_name; |
1620 | versionedNameBuffer.clear(); |
1621 | name = (name + "@" + verName).toStringRef(Out&: versionedNameBuffer); |
1622 | auto *s = symtab.addSymbol( |
1623 | newSym: SharedSymbol{*this, saver().save(S: name), sym.getBinding(), sym.st_other, |
1624 | sym.getType(), sym.st_value, sym.st_size, alignment}); |
1625 | s->dsoDefined = true; |
1626 | if (s->file == this) |
1627 | s->versionId = idx; |
1628 | } |
1629 | } |
1630 | |
1631 | static ELFKind getBitcodeELFKind(const Triple &t) { |
1632 | if (t.isLittleEndian()) |
1633 | return t.isArch64Bit() ? ELF64LEKind : ELF32LEKind; |
1634 | return t.isArch64Bit() ? ELF64BEKind : ELF32BEKind; |
1635 | } |
1636 | |
1637 | static uint16_t getBitcodeMachineKind(StringRef path, const Triple &t) { |
1638 | switch (t.getArch()) { |
1639 | case Triple::aarch64: |
1640 | case Triple::aarch64_be: |
1641 | return EM_AARCH64; |
1642 | case Triple::amdgcn: |
1643 | case Triple::r600: |
1644 | return EM_AMDGPU; |
1645 | case Triple::arm: |
1646 | case Triple::armeb: |
1647 | case Triple::thumb: |
1648 | case Triple::thumbeb: |
1649 | return EM_ARM; |
1650 | case Triple::avr: |
1651 | return EM_AVR; |
1652 | case Triple::hexagon: |
1653 | return EM_HEXAGON; |
1654 | case Triple::loongarch32: |
1655 | case Triple::loongarch64: |
1656 | return EM_LOONGARCH; |
1657 | case Triple::mips: |
1658 | case Triple::mipsel: |
1659 | case Triple::mips64: |
1660 | case Triple::mips64el: |
1661 | return EM_MIPS; |
1662 | case Triple::msp430: |
1663 | return EM_MSP430; |
1664 | case Triple::ppc: |
1665 | case Triple::ppcle: |
1666 | return EM_PPC; |
1667 | case Triple::ppc64: |
1668 | case Triple::ppc64le: |
1669 | return EM_PPC64; |
1670 | case Triple::riscv32: |
1671 | case Triple::riscv64: |
1672 | return EM_RISCV; |
1673 | case Triple::sparcv9: |
1674 | return EM_SPARCV9; |
1675 | case Triple::systemz: |
1676 | return EM_S390; |
1677 | case Triple::x86: |
1678 | return t.isOSIAMCU() ? EM_IAMCU : EM_386; |
1679 | case Triple::x86_64: |
1680 | return EM_X86_64; |
1681 | default: |
1682 | error(msg: path + ": could not infer e_machine from bitcode target triple " + |
1683 | t.str()); |
1684 | return EM_NONE; |
1685 | } |
1686 | } |
1687 | |
1688 | static uint8_t getOsAbi(const Triple &t) { |
1689 | switch (t.getOS()) { |
1690 | case Triple::AMDHSA: |
1691 | return ELF::ELFOSABI_AMDGPU_HSA; |
1692 | case Triple::AMDPAL: |
1693 | return ELF::ELFOSABI_AMDGPU_PAL; |
1694 | case Triple::Mesa3D: |
1695 | return ELF::ELFOSABI_AMDGPU_MESA3D; |
1696 | default: |
1697 | return ELF::ELFOSABI_NONE; |
1698 | } |
1699 | } |
1700 | |
1701 | BitcodeFile::BitcodeFile(MemoryBufferRef mb, StringRef archiveName, |
1702 | uint64_t offsetInArchive, bool lazy) |
1703 | : InputFile(BitcodeKind, mb) { |
1704 | this->archiveName = archiveName; |
1705 | this->lazy = lazy; |
1706 | |
1707 | std::string path = mb.getBufferIdentifier().str(); |
1708 | if (config->thinLTOIndexOnly) |
1709 | path = replaceThinLTOSuffix(path: mb.getBufferIdentifier()); |
1710 | |
1711 | // ThinLTO assumes that all MemoryBufferRefs given to it have a unique |
1712 | // name. If two archives define two members with the same name, this |
1713 | // causes a collision which result in only one of the objects being taken |
1714 | // into consideration at LTO time (which very likely causes undefined |
1715 | // symbols later in the link stage). So we append file offset to make |
1716 | // filename unique. |
1717 | StringRef name = archiveName.empty() |
1718 | ? saver().save(S: path) |
1719 | : saver().save(S: archiveName + "(" + path::filename(path) + |
1720 | " at " + utostr(X: offsetInArchive) + ")" ); |
1721 | MemoryBufferRef mbref(mb.getBuffer(), name); |
1722 | |
1723 | obj = CHECK(lto::InputFile::create(mbref), this); |
1724 | |
1725 | Triple t(obj->getTargetTriple()); |
1726 | ekind = getBitcodeELFKind(t); |
1727 | emachine = getBitcodeMachineKind(path: mb.getBufferIdentifier(), t); |
1728 | osabi = getOsAbi(t); |
1729 | } |
1730 | |
1731 | static uint8_t mapVisibility(GlobalValue::VisibilityTypes gvVisibility) { |
1732 | switch (gvVisibility) { |
1733 | case GlobalValue::DefaultVisibility: |
1734 | return STV_DEFAULT; |
1735 | case GlobalValue::HiddenVisibility: |
1736 | return STV_HIDDEN; |
1737 | case GlobalValue::ProtectedVisibility: |
1738 | return STV_PROTECTED; |
1739 | } |
1740 | llvm_unreachable("unknown visibility" ); |
1741 | } |
1742 | |
1743 | static void |
1744 | createBitcodeSymbol(Symbol *&sym, const std::vector<bool> &keptComdats, |
1745 | const lto::InputFile::Symbol &objSym, BitcodeFile &f) { |
1746 | uint8_t binding = objSym.isWeak() ? STB_WEAK : STB_GLOBAL; |
1747 | uint8_t type = objSym.isTLS() ? STT_TLS : STT_NOTYPE; |
1748 | uint8_t visibility = mapVisibility(gvVisibility: objSym.getVisibility()); |
1749 | |
1750 | if (!sym) |
1751 | sym = symtab.insert(name: saver().save(S: objSym.getName())); |
1752 | |
1753 | int c = objSym.getComdatIndex(); |
1754 | if (objSym.isUndefined() || (c != -1 && !keptComdats[c])) { |
1755 | Undefined newSym(&f, StringRef(), binding, visibility, type); |
1756 | sym->resolve(other: newSym); |
1757 | sym->referenced = true; |
1758 | return; |
1759 | } |
1760 | |
1761 | if (objSym.isCommon()) { |
1762 | sym->resolve(other: CommonSymbol{&f, StringRef(), binding, visibility, STT_OBJECT, |
1763 | objSym.getCommonAlignment(), |
1764 | objSym.getCommonSize()}); |
1765 | } else { |
1766 | Defined newSym(&f, StringRef(), binding, visibility, type, 0, 0, nullptr); |
1767 | if (objSym.canBeOmittedFromSymbolTable()) |
1768 | newSym.exportDynamic = false; |
1769 | sym->resolve(other: newSym); |
1770 | } |
1771 | } |
1772 | |
1773 | void BitcodeFile::parse() { |
1774 | for (std::pair<StringRef, Comdat::SelectionKind> s : obj->getComdatTable()) { |
1775 | keptComdats.push_back( |
1776 | x: s.second == Comdat::NoDeduplicate || |
1777 | symtab.comdatGroups.try_emplace(Key: CachedHashStringRef(s.first), Args: this) |
1778 | .second); |
1779 | } |
1780 | |
1781 | if (numSymbols == 0) { |
1782 | numSymbols = obj->symbols().size(); |
1783 | symbols = std::make_unique<Symbol *[]>(num: numSymbols); |
1784 | } |
1785 | // Process defined symbols first. See the comment in |
1786 | // ObjFile<ELFT>::initializeSymbols. |
1787 | for (auto [i, irSym] : llvm::enumerate(First: obj->symbols())) |
1788 | if (!irSym.isUndefined()) |
1789 | createBitcodeSymbol(sym&: symbols[i], keptComdats, objSym: irSym, f&: *this); |
1790 | for (auto [i, irSym] : llvm::enumerate(First: obj->symbols())) |
1791 | if (irSym.isUndefined()) |
1792 | createBitcodeSymbol(sym&: symbols[i], keptComdats, objSym: irSym, f&: *this); |
1793 | |
1794 | for (auto l : obj->getDependentLibraries()) |
1795 | addDependentLibrary(specifier: l, f: this); |
1796 | } |
1797 | |
1798 | void BitcodeFile::parseLazy() { |
1799 | numSymbols = obj->symbols().size(); |
1800 | symbols = std::make_unique<Symbol *[]>(num: numSymbols); |
1801 | for (auto [i, irSym] : llvm::enumerate(First: obj->symbols())) |
1802 | if (!irSym.isUndefined()) { |
1803 | auto *sym = symtab.insert(name: saver().save(S: irSym.getName())); |
1804 | sym->resolve(other: LazySymbol{*this}); |
1805 | symbols[i] = sym; |
1806 | } |
1807 | } |
1808 | |
1809 | void BitcodeFile::postParse() { |
1810 | for (auto [i, irSym] : llvm::enumerate(First: obj->symbols())) { |
1811 | const Symbol &sym = *symbols[i]; |
1812 | if (sym.file == this || !sym.isDefined() || irSym.isUndefined() || |
1813 | irSym.isCommon() || irSym.isWeak()) |
1814 | continue; |
1815 | int c = irSym.getComdatIndex(); |
1816 | if (c != -1 && !keptComdats[c]) |
1817 | continue; |
1818 | reportDuplicate(sym, newFile: this, errSec: nullptr, errOffset: 0); |
1819 | } |
1820 | } |
1821 | |
1822 | void BinaryFile::parse() { |
1823 | ArrayRef<uint8_t> data = arrayRefFromStringRef(Input: mb.getBuffer()); |
1824 | auto *section = make<InputSection>(args: this, args: SHF_ALLOC | SHF_WRITE, args: SHT_PROGBITS, |
1825 | args: 8, args&: data, args: ".data" ); |
1826 | sections.push_back(Elt: section); |
1827 | |
1828 | // For each input file foo that is embedded to a result as a binary |
1829 | // blob, we define _binary_foo_{start,end,size} symbols, so that |
1830 | // user programs can access blobs by name. Non-alphanumeric |
1831 | // characters in a filename are replaced with underscore. |
1832 | std::string s = "_binary_" + mb.getBufferIdentifier().str(); |
1833 | for (char &c : s) |
1834 | if (!isAlnum(C: c)) |
1835 | c = '_'; |
1836 | |
1837 | llvm::StringSaver &saver = lld::saver(); |
1838 | |
1839 | symtab.addAndCheckDuplicate(newSym: Defined{this, saver.save(S: s + "_start" ), |
1840 | STB_GLOBAL, STV_DEFAULT, STT_OBJECT, 0, 0, |
1841 | section}); |
1842 | symtab.addAndCheckDuplicate(newSym: Defined{this, saver.save(S: s + "_end" ), STB_GLOBAL, |
1843 | STV_DEFAULT, STT_OBJECT, data.size(), 0, |
1844 | section}); |
1845 | symtab.addAndCheckDuplicate(newSym: Defined{this, saver.save(S: s + "_size" ), STB_GLOBAL, |
1846 | STV_DEFAULT, STT_OBJECT, data.size(), 0, |
1847 | nullptr}); |
1848 | } |
1849 | |
1850 | InputFile *elf::createInternalFile(StringRef name) { |
1851 | auto *file = |
1852 | make<InputFile>(args: InputFile::InternalKind, args: MemoryBufferRef("" , name)); |
1853 | // References from an internal file do not lead to --warn-backrefs |
1854 | // diagnostics. |
1855 | file->groupId = 0; |
1856 | return file; |
1857 | } |
1858 | |
1859 | ELFFileBase *elf::createObjFile(MemoryBufferRef mb, StringRef archiveName, |
1860 | bool lazy) { |
1861 | ELFFileBase *f; |
1862 | switch (getELFKind(mb, archiveName)) { |
1863 | case ELF32LEKind: |
1864 | f = make<ObjFile<ELF32LE>>(args: ELF32LEKind, args&: mb, args&: archiveName); |
1865 | break; |
1866 | case ELF32BEKind: |
1867 | f = make<ObjFile<ELF32BE>>(args: ELF32BEKind, args&: mb, args&: archiveName); |
1868 | break; |
1869 | case ELF64LEKind: |
1870 | f = make<ObjFile<ELF64LE>>(args: ELF64LEKind, args&: mb, args&: archiveName); |
1871 | break; |
1872 | case ELF64BEKind: |
1873 | f = make<ObjFile<ELF64BE>>(args: ELF64BEKind, args&: mb, args&: archiveName); |
1874 | break; |
1875 | default: |
1876 | llvm_unreachable("getELFKind" ); |
1877 | } |
1878 | f->init(); |
1879 | f->lazy = lazy; |
1880 | return f; |
1881 | } |
1882 | |
1883 | template <class ELFT> void ObjFile<ELFT>::parseLazy() { |
1884 | const ArrayRef<typename ELFT::Sym> eSyms = this->getELFSyms<ELFT>(); |
1885 | numSymbols = eSyms.size(); |
1886 | symbols = std::make_unique<Symbol *[]>(numSymbols); |
1887 | |
1888 | // resolve() may trigger this->extract() if an existing symbol is an undefined |
1889 | // symbol. If that happens, this function has served its purpose, and we can |
1890 | // exit from the loop early. |
1891 | for (size_t i = firstGlobal, end = eSyms.size(); i != end; ++i) { |
1892 | if (eSyms[i].st_shndx == SHN_UNDEF) |
1893 | continue; |
1894 | symbols[i] = symtab.insert(CHECK(eSyms[i].getName(stringTable), this)); |
1895 | symbols[i]->resolve(LazySymbol{*this}); |
1896 | if (!lazy) |
1897 | break; |
1898 | } |
1899 | } |
1900 | |
1901 | bool InputFile::(StringRef name) const { |
1902 | if (isa<BitcodeFile>(Val: this)) |
1903 | return isBitcodeNonCommonDef(mb, symName: name, archiveName); |
1904 | |
1905 | return isNonCommonDef(mb, symName: name, archiveName); |
1906 | } |
1907 | |
1908 | std::string elf::replaceThinLTOSuffix(StringRef path) { |
1909 | auto [suffix, repl] = config->thinLTOObjectSuffixReplace; |
1910 | if (path.consume_back(Suffix: suffix)) |
1911 | return (path + repl).str(); |
1912 | return std::string(path); |
1913 | } |
1914 | |
1915 | template class elf::ObjFile<ELF32LE>; |
1916 | template class elf::ObjFile<ELF32BE>; |
1917 | template class elf::ObjFile<ELF64LE>; |
1918 | template class elf::ObjFile<ELF64BE>; |
1919 | |
1920 | template void SharedFile::parse<ELF32LE>(); |
1921 | template void SharedFile::parse<ELF32BE>(); |
1922 | template void SharedFile::parse<ELF64LE>(); |
1923 | template void SharedFile::parse<ELF64BE>(); |
1924 | |