1 //===- OutputSections.cpp -------------------------------------------------===//
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
7 //===----------------------------------------------------------------------===//
9 #include "OutputSections.h"
11 #include "LinkerScript.h"
12 #include "SymbolTable.h"
13 #include "SyntheticSections.h"
15 #include "lld/Common/Memory.h"
16 #include "lld/Common/Strings.h"
17 #include "lld/Common/Threads.h"
18 #include "llvm/BinaryFormat/Dwarf.h"
19 #include "llvm/Support/Compression.h"
20 #include "llvm/Support/MD5.h"
21 #include "llvm/Support/MathExtras.h"
22 #include "llvm/Support/SHA1.h"
26 using namespace llvm::dwarf;
27 using namespace llvm::object;
28 using namespace llvm::support::endian;
29 using namespace llvm::ELF;
33 uint8_t *Out::bufferStart;
35 PhdrEntry *Out::tlsPhdr;
36 OutputSection *Out::elfHeader;
37 OutputSection *Out::programHeaders;
38 OutputSection *Out::preinitArray;
39 OutputSection *Out::initArray;
40 OutputSection *Out::finiArray;
42 std::vector<OutputSection *> outputSections;
44 uint32_t OutputSection::getPhdrFlags() const {
46 if (config->emachine != EM_ARM || !(flags & SHF_ARM_PURECODE))
48 if (flags & SHF_WRITE)
50 if (flags & SHF_EXECINSTR)
56 void OutputSection::writeHeaderTo(typename ELFT::Shdr *shdr) {
57 shdr->sh_entsize = entsize;
58 shdr->sh_addralign = alignment;
60 shdr->sh_offset = offset;
61 shdr->sh_flags = flags;
66 shdr->sh_name = shName;
69 OutputSection::OutputSection(StringRef name, uint32_t type, uint64_t flags)
70 : BaseCommand(OutputSectionKind),
71 SectionBase(Output, name, flags, /*Entsize*/ 0, /*Alignment*/ 1, type,
72 /*Info*/ 0, /*Link*/ 0) {}
74 // We allow sections of types listed below to merged into a
75 // single progbits section. This is typically done by linker
76 // scripts. Merging nobits and progbits will force disk space
77 // to be allocated for nobits sections. Other ones don't require
78 // any special treatment on top of progbits, so there doesn't
79 // seem to be a harm in merging them.
80 static bool canMergeToProgbits(unsigned type) {
81 return type == SHT_NOBITS || type == SHT_PROGBITS || type == SHT_INIT_ARRAY ||
82 type == SHT_PREINIT_ARRAY || type == SHT_FINI_ARRAY ||
86 // Record that isec will be placed in the OutputSection. isec does not become
87 // permanent until finalizeInputSections() is called. The function should not be
88 // used after finalizeInputSections() is called. If you need to add an
89 // InputSection post finalizeInputSections(), then you must do the following:
91 // 1. Find or create an InputSectionDescription to hold InputSection.
92 // 2. Add the InputSection to the InputSectionDescription::sections.
93 // 3. Call commitSection(isec).
94 void OutputSection::recordSection(InputSectionBase *isec) {
95 partition = isec->partition;
97 if (sectionCommands.empty() ||
98 !isa<InputSectionDescription>(sectionCommands.back()))
99 sectionCommands.push_back(make<InputSectionDescription>(""));
100 auto *isd = cast<InputSectionDescription>(sectionCommands.back());
101 isd->sectionBases.push_back(isec);
104 // Update fields (type, flags, alignment, etc) according to the InputSection
105 // isec. Also check whether the InputSection flags and type are consistent with
106 // other InputSections.
107 void OutputSection::commitSection(InputSection *isec) {
108 if (!hasInputSections) {
109 // If IS is the first section to be added to this section,
110 // initialize type, entsize and flags from isec.
111 hasInputSections = true;
113 entsize = isec->entsize;
116 // Otherwise, check if new type or flags are compatible with existing ones.
117 unsigned mask = SHF_TLS | SHF_LINK_ORDER;
118 if ((flags & mask) != (isec->flags & mask))
119 error("incompatible section flags for " + name + "\n>>> " + toString(isec) +
120 ": 0x" + utohexstr(isec->flags) + "\n>>> output section " + name +
121 ": 0x" + utohexstr(flags));
123 if (type != isec->type) {
124 if (!canMergeToProgbits(type) || !canMergeToProgbits(isec->type))
125 error("section type mismatch for " + isec->name + "\n>>> " +
126 toString(isec) + ": " +
127 getELFSectionTypeName(config->emachine, isec->type) +
128 "\n>>> output section " + name + ": " +
129 getELFSectionTypeName(config->emachine, type));
138 config->emachine == EM_ARM ? (uint64_t)SHF_ARM_PURECODE : 0;
139 uint64_t orMask = ~andMask;
140 uint64_t andFlags = (flags & isec->flags) & andMask;
141 uint64_t orFlags = (flags | isec->flags) & orMask;
142 flags = andFlags | orFlags;
144 flags &= ~(uint64_t)SHF_ALLOC;
146 alignment = std::max(alignment, isec->alignment);
148 // If this section contains a table of fixed-size entries, sh_entsize
149 // holds the element size. If it contains elements of different size we
150 // set sh_entsize to 0.
151 if (entsize != isec->entsize)
155 // This function scans over the InputSectionBase list sectionBases to create
156 // InputSectionDescription::sections.
158 // It removes MergeInputSections from the input section array and adds
159 // new synthetic sections at the location of the first input section
160 // that it replaces. It then finalizes each synthetic section in order
161 // to compute an output offset for each piece of each input section.
162 void OutputSection::finalizeInputSections() {
163 std::vector<MergeSyntheticSection *> mergeSections;
164 for (BaseCommand *base : sectionCommands) {
165 auto *cmd = dyn_cast<InputSectionDescription>(base);
168 cmd->sections.reserve(cmd->sectionBases.size());
169 for (InputSectionBase *s : cmd->sectionBases) {
170 MergeInputSection *ms = dyn_cast<MergeInputSection>(s);
172 cmd->sections.push_back(cast<InputSection>(s));
176 // We do not want to handle sections that are not alive, so just remove
177 // them instead of trying to merge.
181 auto i = llvm::find_if(mergeSections, [=](MergeSyntheticSection *sec) {
182 // While we could create a single synthetic section for two different
183 // values of Entsize, it is better to take Entsize into consideration.
185 // With a single synthetic section no two pieces with different Entsize
186 // could be equal, so we may as well have two sections.
188 // Using Entsize in here also allows us to propagate it to the synthetic
191 // SHF_STRINGS section with different alignments should not be merged.
192 return sec->flags == ms->flags && sec->entsize == ms->entsize &&
193 (sec->alignment == ms->alignment || !(sec->flags & SHF_STRINGS));
195 if (i == mergeSections.end()) {
196 MergeSyntheticSection *syn =
197 createMergeSynthetic(name, ms->type, ms->flags, ms->alignment);
198 mergeSections.push_back(syn);
199 i = std::prev(mergeSections.end());
200 syn->entsize = ms->entsize;
201 cmd->sections.push_back(syn);
203 (*i)->addSection(ms);
206 // sectionBases should not be used from this point onwards. Clear it to
208 cmd->sectionBases.clear();
210 // Some input sections may be removed from the list after ICF.
211 for (InputSection *s : cmd->sections)
214 for (auto *ms : mergeSections)
215 ms->finalizeContents();
218 static void sortByOrder(MutableArrayRef<InputSection *> in,
219 llvm::function_ref<int(InputSectionBase *s)> order) {
220 std::vector<std::pair<int, InputSection *>> v;
221 for (InputSection *s : in)
222 v.push_back({order(s), s});
223 llvm::stable_sort(v, less_first());
225 for (size_t i = 0; i < v.size(); ++i)
229 uint64_t getHeaderSize() {
230 if (config->oFormatBinary)
232 return Out::elfHeader->size + Out::programHeaders->size;
235 bool OutputSection::classof(const BaseCommand *c) {
236 return c->kind == OutputSectionKind;
239 void OutputSection::sort(llvm::function_ref<int(InputSectionBase *s)> order) {
241 for (BaseCommand *b : sectionCommands)
242 if (auto *isd = dyn_cast<InputSectionDescription>(b))
243 sortByOrder(isd->sections, order);
246 // Fill [Buf, Buf + Size) with Filler.
247 // This is used for linker script "=fillexp" command.
248 static void fill(uint8_t *buf, size_t size,
249 const std::array<uint8_t, 4> &filler) {
251 for (; i + 4 < size; i += 4)
252 memcpy(buf + i, filler.data(), 4);
253 memcpy(buf + i, filler.data(), size - i);
256 // Compress section contents if this section contains debug info.
257 template <class ELFT> void OutputSection::maybeCompress() {
258 using Elf_Chdr = typename ELFT::Chdr;
260 // Compress only DWARF debug sections.
261 if (!config->compressDebugSections || (flags & SHF_ALLOC) ||
262 !name.startswith(".debug_"))
265 // Create a section header.
266 zDebugHeader.resize(sizeof(Elf_Chdr));
267 auto *hdr = reinterpret_cast<Elf_Chdr *>(zDebugHeader.data());
268 hdr->ch_type = ELFCOMPRESS_ZLIB;
270 hdr->ch_addralign = alignment;
272 // Write section contents to a temporary buffer and compress it.
273 std::vector<uint8_t> buf(size);
274 writeTo<ELFT>(buf.data());
275 // We chose 1 as the default compression level because it is the fastest. If
276 // -O2 is given, we use level 6 to compress debug info more by ~15%. We found
277 // that level 7 to 9 doesn't make much difference (~1% more compression) while
278 // they take significant amount of time (~2x), so level 6 seems enough.
279 if (Error e = zlib::compress(toStringRef(buf), compressedData,
280 config->optimize >= 2 ? 6 : 1))
281 fatal("compress failed: " + llvm::toString(std::move(e)));
283 // Update section headers.
284 size = sizeof(Elf_Chdr) + compressedData.size();
285 flags |= SHF_COMPRESSED;
288 static void writeInt(uint8_t *buf, uint64_t data, uint64_t size) {
298 llvm_unreachable("unsupported Size argument");
301 template <class ELFT> void OutputSection::writeTo(uint8_t *buf) {
302 if (type == SHT_NOBITS)
305 // If -compress-debug-section is specified and if this is a debug section,
306 // we've already compressed section contents. If that's the case,
307 // just write it down.
308 if (!compressedData.empty()) {
309 memcpy(buf, zDebugHeader.data(), zDebugHeader.size());
310 memcpy(buf + zDebugHeader.size(), compressedData.data(),
311 compressedData.size());
315 // Write leading padding.
316 std::vector<InputSection *> sections = getInputSections(this);
317 std::array<uint8_t, 4> filler = getFiller();
318 bool nonZeroFiller = read32(filler.data()) != 0;
320 fill(buf, sections.empty() ? size : sections[0]->outSecOff, filler);
322 parallelForEachN(0, sections.size(), [&](size_t i) {
323 InputSection *isec = sections[i];
324 isec->writeTo<ELFT>(buf);
326 // Fill gaps between sections.
328 uint8_t *start = buf + isec->outSecOff + isec->getSize();
330 if (i + 1 == sections.size())
333 end = buf + sections[i + 1]->outSecOff;
334 fill(start, end - start, filler);
338 // Linker scripts may have BYTE()-family commands with which you
339 // can write arbitrary bytes to the output. Process them if any.
340 for (BaseCommand *base : sectionCommands)
341 if (auto *data = dyn_cast<ByteCommand>(base))
342 writeInt(buf + data->offset, data->expression().getValue(), data->size);
345 static void finalizeShtGroup(OutputSection *os,
346 InputSection *section) {
347 assert(config->relocatable);
349 // sh_link field for SHT_GROUP sections should contain the section index of
351 os->link = in.symTab->getParent()->sectionIndex;
353 // sh_info then contain index of an entry in symbol table section which
354 // provides signature of the section group.
355 ArrayRef<Symbol *> symbols = section->file->getSymbols();
356 os->info = in.symTab->getSymbolIndex(symbols[section->info]);
359 void OutputSection::finalize() {
360 std::vector<InputSection *> v = getInputSections(this);
361 InputSection *first = v.empty() ? nullptr : v[0];
363 if (flags & SHF_LINK_ORDER) {
364 // We must preserve the link order dependency of sections with the
365 // SHF_LINK_ORDER flag. The dependency is indicated by the sh_link field. We
366 // need to translate the InputSection sh_link to the OutputSection sh_link,
367 // all InputSections in the OutputSection have the same dependency.
368 if (auto *ex = dyn_cast<ARMExidxSyntheticSection>(first))
369 link = ex->getLinkOrderDep()->getParent()->sectionIndex;
370 else if (auto *d = first->getLinkOrderDep())
371 link = d->getParent()->sectionIndex;
374 if (type == SHT_GROUP) {
375 finalizeShtGroup(this, first);
379 if (!config->copyRelocs || (type != SHT_RELA && type != SHT_REL))
382 if (isa<SyntheticSection>(first))
385 link = in.symTab->getParent()->sectionIndex;
386 // sh_info for SHT_REL[A] sections should contain the section header index of
387 // the section to which the relocation applies.
388 InputSectionBase *s = first->getRelocatedSection();
389 info = s->getOutputSection()->sectionIndex;
390 flags |= SHF_INFO_LINK;
393 // Returns true if S is in one of the many forms the compiler driver may pass
396 // Gcc uses any of crtbegin[<empty>|S|T].o.
397 // Clang uses Gcc's plus clang_rt.crtbegin[<empty>|S|T][-<arch>|<empty>].o.
399 static bool isCrtbegin(StringRef s) {
400 static std::regex re(R"((clang_rt\.)?crtbegin[ST]?(-.*)?\.o)");
401 s = sys::path::filename(s);
402 return std::regex_match(s.begin(), s.end(), re);
405 static bool isCrtend(StringRef s) {
406 static std::regex re(R"((clang_rt\.)?crtend[ST]?(-.*)?\.o)");
407 s = sys::path::filename(s);
408 return std::regex_match(s.begin(), s.end(), re);
411 // .ctors and .dtors are sorted by this priority from highest to lowest.
413 // 1. The section was contained in crtbegin (crtbegin contains
414 // some sentinel value in its .ctors and .dtors so that the runtime
415 // can find the beginning of the sections.)
417 // 2. The section has an optional priority value in the form of ".ctors.N"
418 // or ".dtors.N" where N is a number. Unlike .{init,fini}_array,
419 // they are compared as string rather than number.
421 // 3. The section is just ".ctors" or ".dtors".
423 // 4. The section was contained in crtend, which contains an end marker.
425 // In an ideal world, we don't need this function because .init_array and
426 // .ctors are duplicate features (and .init_array is newer.) However, there
427 // are too many real-world use cases of .ctors, so we had no choice to
428 // support that with this rather ad-hoc semantics.
429 static bool compCtors(const InputSection *a, const InputSection *b) {
430 bool beginA = isCrtbegin(a->file->getName());
431 bool beginB = isCrtbegin(b->file->getName());
432 if (beginA != beginB)
434 bool endA = isCrtend(a->file->getName());
435 bool endB = isCrtend(b->file->getName());
438 StringRef x = a->name;
439 StringRef y = b->name;
440 assert(x.startswith(".ctors") || x.startswith(".dtors"));
441 assert(y.startswith(".ctors") || y.startswith(".dtors"));
447 // Sorts input sections by the special rules for .ctors and .dtors.
448 // Unfortunately, the rules are different from the one for .{init,fini}_array.
449 // Read the comment above.
450 void OutputSection::sortCtorsDtors() {
451 assert(sectionCommands.size() == 1);
452 auto *isd = cast<InputSectionDescription>(sectionCommands[0]);
453 llvm::stable_sort(isd->sections, compCtors);
456 // If an input string is in the form of "foo.N" where N is a number,
457 // return N. Otherwise, returns 65536, which is one greater than the
459 int getPriority(StringRef s) {
460 size_t pos = s.rfind('.');
461 if (pos == StringRef::npos)
464 if (!to_integer(s.substr(pos + 1), v, 10))
469 std::vector<InputSection *> getInputSections(OutputSection *os) {
470 std::vector<InputSection *> ret;
471 for (BaseCommand *base : os->sectionCommands)
472 if (auto *isd = dyn_cast<InputSectionDescription>(base))
473 ret.insert(ret.end(), isd->sections.begin(), isd->sections.end());
477 // Sorts input sections by section name suffixes, so that .foo.N comes
478 // before .foo.M if N < M. Used to sort .{init,fini}_array.N sections.
479 // We want to keep the original order if the priorities are the same
480 // because the compiler keeps the original initialization order in a
481 // translation unit and we need to respect that.
482 // For more detail, read the section of the GCC's manual about init_priority.
483 void OutputSection::sortInitFini() {
484 // Sort sections by priority.
485 sort([](InputSectionBase *s) { return getPriority(s->name); });
488 std::array<uint8_t, 4> OutputSection::getFiller() {
491 if (flags & SHF_EXECINSTR)
492 return target->trapInstr;
496 template void OutputSection::writeHeaderTo<ELF32LE>(ELF32LE::Shdr *Shdr);
497 template void OutputSection::writeHeaderTo<ELF32BE>(ELF32BE::Shdr *Shdr);
498 template void OutputSection::writeHeaderTo<ELF64LE>(ELF64LE::Shdr *Shdr);
499 template void OutputSection::writeHeaderTo<ELF64BE>(ELF64BE::Shdr *Shdr);
501 template void OutputSection::writeTo<ELF32LE>(uint8_t *Buf);
502 template void OutputSection::writeTo<ELF32BE>(uint8_t *Buf);
503 template void OutputSection::writeTo<ELF64LE>(uint8_t *Buf);
504 template void OutputSection::writeTo<ELF64BE>(uint8_t *Buf);
506 template void OutputSection::maybeCompress<ELF32LE>();
507 template void OutputSection::maybeCompress<ELF32BE>();
508 template void OutputSection::maybeCompress<ELF64LE>();
509 template void OutputSection::maybeCompress<ELF64BE>();