1 //===- OutputSections.cpp -------------------------------------------------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "OutputSections.h"
12 #include "LinkerScript.h"
15 #include "SymbolTable.h"
16 #include "SyntheticSections.h"
19 #include "llvm/Support/Compression.h"
20 #include "llvm/Support/Dwarf.h"
21 #include "llvm/Support/MD5.h"
22 #include "llvm/Support/MathExtras.h"
23 #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;
32 using namespace lld::elf;
35 OutputSection *Out::Opd;
37 PhdrEntry *Out::TlsPhdr;
38 OutputSection *Out::DebugInfo;
39 OutputSection *Out::ElfHeader;
40 OutputSection *Out::ProgramHeaders;
41 OutputSection *Out::PreinitArray;
42 OutputSection *Out::InitArray;
43 OutputSection *Out::FiniArray;
45 uint32_t OutputSection::getPhdrFlags() const {
47 if (Flags & SHF_WRITE)
49 if (Flags & SHF_EXECINSTR)
55 void OutputSection::writeHeaderTo(typename ELFT::Shdr *Shdr) {
56 Shdr->sh_entsize = Entsize;
57 Shdr->sh_addralign = Alignment;
59 Shdr->sh_offset = Offset;
60 Shdr->sh_flags = Flags;
65 Shdr->sh_name = ShName;
68 OutputSection::OutputSection(StringRef Name, uint32_t Type, uint64_t Flags)
69 : SectionBase(Output, Name, Flags, /*Entsize*/ 0, /*Alignment*/ 1, Type,
72 SectionIndex(INT_MAX) {}
74 static bool compareByFilePosition(InputSection *A, InputSection *B) {
75 // Synthetic doesn't have link order dependecy, stable_sort will keep it last
76 if (A->kind() == InputSectionBase::Synthetic ||
77 B->kind() == InputSectionBase::Synthetic)
79 auto *LA = cast<InputSection>(A->getLinkOrderDep());
80 auto *LB = cast<InputSection>(B->getLinkOrderDep());
81 OutputSection *AOut = LA->OutSec;
82 OutputSection *BOut = LB->OutSec;
84 return AOut->SectionIndex < BOut->SectionIndex;
85 return LA->OutSecOff < LB->OutSecOff;
88 // Compress section contents if this section contains debug info.
89 template <class ELFT> void OutputSection::maybeCompress() {
90 typedef typename ELFT::Chdr Elf_Chdr;
92 // Compress only DWARF debug sections.
93 if (!Config->CompressDebugSections || (Flags & SHF_ALLOC) ||
94 !Name.startswith(".debug_"))
97 // Create a section header.
98 ZDebugHeader.resize(sizeof(Elf_Chdr));
99 auto *Hdr = reinterpret_cast<Elf_Chdr *>(ZDebugHeader.data());
100 Hdr->ch_type = ELFCOMPRESS_ZLIB;
102 Hdr->ch_addralign = Alignment;
104 // Write section contents to a temporary buffer and compress it.
105 std::vector<uint8_t> Buf(Size);
106 writeTo<ELFT>(Buf.data());
107 if (Error E = zlib::compress(toStringRef(Buf), CompressedData))
108 fatal("compress failed: " + llvm::toString(std::move(E)));
110 // Update section headers.
111 Size = sizeof(Elf_Chdr) + CompressedData.size();
112 Flags |= SHF_COMPRESSED;
115 template <class ELFT> void OutputSection::finalize() {
116 if ((this->Flags & SHF_LINK_ORDER) && !this->Sections.empty()) {
117 std::sort(Sections.begin(), Sections.end(), compareByFilePosition);
120 // We must preserve the link order dependency of sections with the
121 // SHF_LINK_ORDER flag. The dependency is indicated by the sh_link field. We
122 // need to translate the InputSection sh_link to the OutputSection sh_link,
123 // all InputSections in the OutputSection have the same dependency.
124 if (auto *D = this->Sections.front()->getLinkOrderDep())
125 this->Link = D->OutSec->SectionIndex;
128 uint32_t Type = this->Type;
129 if (!Config->CopyRelocs || (Type != SHT_RELA && Type != SHT_REL))
132 InputSection *First = Sections[0];
133 if (isa<SyntheticSection>(First))
136 this->Link = InX::SymTab->OutSec->SectionIndex;
137 // sh_info for SHT_REL[A] sections should contain the section header index of
138 // the section to which the relocation applies.
139 InputSectionBase *S = First->getRelocatedSection();
140 this->Info = S->OutSec->SectionIndex;
143 static uint64_t updateOffset(uint64_t Off, InputSection *S) {
144 Off = alignTo(Off, S->Alignment);
146 return Off + S->getSize();
149 void OutputSection::addSection(InputSection *S) {
151 Sections.push_back(S);
153 this->updateAlignment(S->Alignment);
155 // The actual offsets will be computed by assignAddresses. For now, use
156 // crude approximation so that it is at least easy for other code to know the
157 // section order. It is also used to calculate the output section size early
158 // for compressed debug sections.
159 this->Size = updateOffset(Size, S);
161 // If this section contains a table of fixed-size entries, sh_entsize
162 // holds the element size. Consequently, if this contains two or more
163 // input sections, all of them must have the same sh_entsize. However,
164 // you can put different types of input sections into one output
165 // sectin by using linker scripts. I don't know what to do here.
166 // Probably we sholuld handle that as an error. But for now we just
167 // pick the largest sh_entsize.
168 this->Entsize = std::max(this->Entsize, S->Entsize);
171 // This function is called after we sort input sections
172 // and scan relocations to setup sections' offsets.
173 void OutputSection::assignOffsets() {
175 for (InputSection *S : Sections)
176 Off = updateOffset(Off, S);
180 void OutputSection::sort(std::function<int(InputSectionBase *S)> Order) {
181 typedef std::pair<unsigned, InputSection *> Pair;
182 auto Comp = [](const Pair &A, const Pair &B) { return A.first < B.first; };
185 for (InputSection *S : Sections)
186 V.push_back({Order(S), S});
187 std::stable_sort(V.begin(), V.end(), Comp);
190 Sections.push_back(P.second);
193 // Sorts input sections by section name suffixes, so that .foo.N comes
194 // before .foo.M if N < M. Used to sort .{init,fini}_array.N sections.
195 // We want to keep the original order if the priorities are the same
196 // because the compiler keeps the original initialization order in a
197 // translation unit and we need to respect that.
198 // For more detail, read the section of the GCC's manual about init_priority.
199 void OutputSection::sortInitFini() {
200 // Sort sections by priority.
201 sort([](InputSectionBase *S) { return getPriority(S->Name); });
204 // Returns true if S matches /Filename.?\.o$/.
205 static bool isCrtBeginEnd(StringRef S, StringRef Filename) {
206 if (!S.endswith(".o"))
209 if (S.endswith(Filename))
211 return !S.empty() && S.drop_back().endswith(Filename);
214 static bool isCrtbegin(StringRef S) { return isCrtBeginEnd(S, "crtbegin"); }
215 static bool isCrtend(StringRef S) { return isCrtBeginEnd(S, "crtend"); }
217 // .ctors and .dtors are sorted by this priority from highest to lowest.
219 // 1. The section was contained in crtbegin (crtbegin contains
220 // some sentinel value in its .ctors and .dtors so that the runtime
221 // can find the beginning of the sections.)
223 // 2. The section has an optional priority value in the form of ".ctors.N"
224 // or ".dtors.N" where N is a number. Unlike .{init,fini}_array,
225 // they are compared as string rather than number.
227 // 3. The section is just ".ctors" or ".dtors".
229 // 4. The section was contained in crtend, which contains an end marker.
231 // In an ideal world, we don't need this function because .init_array and
232 // .ctors are duplicate features (and .init_array is newer.) However, there
233 // are too many real-world use cases of .ctors, so we had no choice to
234 // support that with this rather ad-hoc semantics.
235 static bool compCtors(const InputSection *A, const InputSection *B) {
236 bool BeginA = isCrtbegin(A->File->getName());
237 bool BeginB = isCrtbegin(B->File->getName());
238 if (BeginA != BeginB)
240 bool EndA = isCrtend(A->File->getName());
241 bool EndB = isCrtend(B->File->getName());
244 StringRef X = A->Name;
245 StringRef Y = B->Name;
246 assert(X.startswith(".ctors") || X.startswith(".dtors"));
247 assert(Y.startswith(".ctors") || Y.startswith(".dtors"));
250 if (X.empty() && Y.empty())
255 // Sorts input sections by the special rules for .ctors and .dtors.
256 // Unfortunately, the rules are different from the one for .{init,fini}_array.
257 // Read the comment above.
258 void OutputSection::sortCtorsDtors() {
259 std::stable_sort(Sections.begin(), Sections.end(), compCtors);
262 // Fill [Buf, Buf + Size) with Filler.
263 // This is used for linker script "=fillexp" command.
264 static void fill(uint8_t *Buf, size_t Size, uint32_t Filler) {
266 for (; I + 4 < Size; I += 4)
267 memcpy(Buf + I, &Filler, 4);
268 memcpy(Buf + I, &Filler, Size - I);
271 uint32_t OutputSection::getFiller() {
272 // Determine what to fill gaps between InputSections with, as specified by the
273 // linker script. If nothing is specified and this is an executable section,
274 // fall back to trap instructions to prevent bad diassembly and detect invalid
276 if (Optional<uint32_t> Filler = Script->getFiller(this))
278 if (Flags & SHF_EXECINSTR)
279 return Target->TrapInstr;
283 template <class ELFT> void OutputSection::writeTo(uint8_t *Buf) {
286 // We may have already rendered compressed content when using
287 // -compress-debug-sections option. Write it together with header.
288 if (!CompressedData.empty()) {
289 memcpy(Buf, ZDebugHeader.data(), ZDebugHeader.size());
290 memcpy(Buf + ZDebugHeader.size(), CompressedData.data(),
291 CompressedData.size());
295 // Write leading padding.
296 uint32_t Filler = getFiller();
298 fill(Buf, Sections.empty() ? Size : Sections[0]->OutSecOff, Filler);
300 parallelForEachN(0, Sections.size(), [=](size_t I) {
301 InputSection *Sec = Sections[I];
302 Sec->writeTo<ELFT>(Buf);
304 // Fill gaps between sections.
306 uint8_t *Start = Buf + Sec->OutSecOff + Sec->getSize();
308 if (I + 1 == Sections.size())
311 End = Buf + Sections[I + 1]->OutSecOff;
312 fill(Start, End - Start, Filler);
316 // Linker scripts may have BYTE()-family commands with which you
317 // can write arbitrary bytes to the output. Process them if any.
318 Script->writeDataBytes(this, Buf);
321 static uint64_t getOutFlags(InputSectionBase *S) {
322 return S->Flags & ~SHF_GROUP & ~SHF_COMPRESSED;
325 static SectionKey createKey(InputSectionBase *C, StringRef OutsecName) {
326 // The ELF spec just says
327 // ----------------------------------------------------------------
328 // In the first phase, input sections that match in name, type and
329 // attribute flags should be concatenated into single sections.
330 // ----------------------------------------------------------------
332 // However, it is clear that at least some flags have to be ignored for
333 // section merging. At the very least SHF_GROUP and SHF_COMPRESSED have to be
334 // ignored. We should not have two output .text sections just because one was
335 // in a group and another was not for example.
337 // It also seems that that wording was a late addition and didn't get the
338 // necessary scrutiny.
340 // Merging sections with different flags is expected by some users. One
341 // reason is that if one file has
343 // int *const bar __attribute__((section(".foo"))) = (int *)0;
345 // gcc with -fPIC will produce a read only .foo section. But if another
349 // int *const bar __attribute__((section(".foo"))) = (int *)&zed;
351 // gcc with -fPIC will produce a read write section.
353 // Last but not least, when using linker script the merge rules are forced by
354 // the script. Unfortunately, linker scripts are name based. This means that
355 // expressions like *(.foo*) can refer to multiple input sections with
356 // different flags. We cannot put them in different output sections or we
357 // would produce wrong results for
359 // start = .; *(.foo.*) end = .; *(.bar)
361 // and a mapping of .foo1 and .bar1 to one section and .foo2 and .bar2 to
362 // another. The problem is that there is no way to layout those output
363 // sections such that the .foo sections are the only thing between the start
366 // Given the above issues, we instead merge sections by name and error on
367 // incompatible types and flags.
369 uint32_t Alignment = 0;
371 if (Config->Relocatable && (C->Flags & SHF_MERGE)) {
372 Alignment = std::max<uint64_t>(C->Alignment, C->Entsize);
373 Flags = C->Flags & (SHF_MERGE | SHF_STRINGS);
376 return SectionKey{OutsecName, Flags, Alignment};
379 OutputSectionFactory::OutputSectionFactory(
380 std::vector<OutputSection *> &OutputSections)
381 : OutputSections(OutputSections) {}
383 static uint64_t getIncompatibleFlags(uint64_t Flags) {
384 return Flags & (SHF_ALLOC | SHF_TLS);
387 // We allow sections of types listed below to merged into a
388 // single progbits section. This is typically done by linker
389 // scripts. Merging nobits and progbits will force disk space
390 // to be allocated for nobits sections. Other ones don't require
391 // any special treatment on top of progbits, so there doesn't
392 // seem to be a harm in merging them.
393 static bool canMergeToProgbits(unsigned Type) {
394 return Type == SHT_NOBITS || Type == SHT_PROGBITS || Type == SHT_INIT_ARRAY ||
395 Type == SHT_PREINIT_ARRAY || Type == SHT_FINI_ARRAY ||
399 static void reportDiscarded(InputSectionBase *IS) {
400 if (!Config->PrintGcSections)
402 message("removing unused section from '" + IS->Name + "' in file '" +
403 IS->File->getName());
406 void OutputSectionFactory::addInputSec(InputSectionBase *IS,
407 StringRef OutsecName) {
408 SectionKey Key = createKey(IS, OutsecName);
409 OutputSection *&Sec = Map[Key];
410 return addInputSec(IS, OutsecName, Sec);
413 void OutputSectionFactory::addInputSec(InputSectionBase *IS,
414 StringRef OutsecName,
415 OutputSection *&Sec) {
421 uint64_t Flags = getOutFlags(IS);
423 if (getIncompatibleFlags(Sec->Flags) != getIncompatibleFlags(IS->Flags))
424 error("incompatible section flags for " + Sec->Name +
425 "\n>>> " + toString(IS) + ": 0x" + utohexstr(IS->Flags) +
426 "\n>>> output section " + Sec->Name + ": 0x" +
427 utohexstr(Sec->Flags));
428 if (Sec->Type != IS->Type) {
429 if (canMergeToProgbits(Sec->Type) && canMergeToProgbits(IS->Type))
430 Sec->Type = SHT_PROGBITS;
432 error("section type mismatch for " + IS->Name +
433 "\n>>> " + toString(IS) + ": " +
434 getELFSectionTypeName(Config->EMachine, IS->Type) +
435 "\n>>> output section " + Sec->Name + ": " +
436 getELFSectionTypeName(Config->EMachine, Sec->Type));
440 Sec = make<OutputSection>(OutsecName, IS->Type, Flags);
441 OutputSections.push_back(Sec);
444 Sec->addSection(cast<InputSection>(IS));
447 OutputSectionFactory::~OutputSectionFactory() {}
449 SectionKey DenseMapInfo<SectionKey>::getEmptyKey() {
450 return SectionKey{DenseMapInfo<StringRef>::getEmptyKey(), 0, 0};
453 SectionKey DenseMapInfo<SectionKey>::getTombstoneKey() {
454 return SectionKey{DenseMapInfo<StringRef>::getTombstoneKey(), 0, 0};
457 unsigned DenseMapInfo<SectionKey>::getHashValue(const SectionKey &Val) {
458 return hash_combine(Val.Name, Val.Flags, Val.Alignment);
461 bool DenseMapInfo<SectionKey>::isEqual(const SectionKey &LHS,
462 const SectionKey &RHS) {
463 return DenseMapInfo<StringRef>::isEqual(LHS.Name, RHS.Name) &&
464 LHS.Flags == RHS.Flags && LHS.Alignment == RHS.Alignment;
467 uint64_t elf::getHeaderSize() {
468 if (Config->OFormatBinary)
470 return Out::ElfHeader->Size + Out::ProgramHeaders->Size;
473 template void OutputSection::writeHeaderTo<ELF32LE>(ELF32LE::Shdr *Shdr);
474 template void OutputSection::writeHeaderTo<ELF32BE>(ELF32BE::Shdr *Shdr);
475 template void OutputSection::writeHeaderTo<ELF64LE>(ELF64LE::Shdr *Shdr);
476 template void OutputSection::writeHeaderTo<ELF64BE>(ELF64BE::Shdr *Shdr);
478 template void OutputSection::finalize<ELF32LE>();
479 template void OutputSection::finalize<ELF32BE>();
480 template void OutputSection::finalize<ELF64LE>();
481 template void OutputSection::finalize<ELF64BE>();
483 template void OutputSection::maybeCompress<ELF32LE>();
484 template void OutputSection::maybeCompress<ELF32BE>();
485 template void OutputSection::maybeCompress<ELF64LE>();
486 template void OutputSection::maybeCompress<ELF64BE>();
488 template void OutputSection::writeTo<ELF32LE>(uint8_t *Buf);
489 template void OutputSection::writeTo<ELF32BE>(uint8_t *Buf);
490 template void OutputSection::writeTo<ELF64LE>(uint8_t *Buf);
491 template void OutputSection::writeTo<ELF64BE>(uint8_t *Buf);