1 //===- LinkerScript.cpp ---------------------------------------------------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains the parser/evaluator of the linker script.
12 //===----------------------------------------------------------------------===//
14 #include "LinkerScript.h"
16 #include "InputSection.h"
18 #include "OutputSections.h"
20 #include "SymbolTable.h"
22 #include "SyntheticSections.h"
26 #include "llvm/ADT/STLExtras.h"
27 #include "llvm/ADT/StringRef.h"
28 #include "llvm/BinaryFormat/ELF.h"
29 #include "llvm/Support/Casting.h"
30 #include "llvm/Support/Compression.h"
31 #include "llvm/Support/Endian.h"
32 #include "llvm/Support/ErrorHandling.h"
33 #include "llvm/Support/FileSystem.h"
34 #include "llvm/Support/Path.h"
45 using namespace llvm::ELF;
46 using namespace llvm::object;
47 using namespace llvm::support::endian;
49 using namespace lld::elf;
51 LinkerScript *elf::Script;
53 uint64_t ExprValue::getValue() const {
55 if (OutputSection *OS = Sec->getOutputSection())
56 return alignTo(Sec->getOffset(Val) + OS->Addr, Alignment);
57 error(Loc + ": unable to evaluate expression: input section " + Sec->Name +
58 " has no output section assigned");
60 return alignTo(Val, Alignment);
63 uint64_t ExprValue::getSecAddr() const {
65 return Sec->getOffset(0) + Sec->getOutputSection()->Addr;
69 template <class ELFT> static SymbolBody *addRegular(SymbolAssignment *Cmd) {
71 uint8_t Visibility = Cmd->Hidden ? STV_HIDDEN : STV_DEFAULT;
72 std::tie(Sym, std::ignore) = Symtab<ELFT>::X->insert(
73 Cmd->Name, /*Type*/ 0, Visibility, /*CanOmitFromDynSym*/ false,
75 Sym->Binding = STB_GLOBAL;
76 ExprValue Value = Cmd->Expression();
77 SectionBase *Sec = Value.isAbsolute() ? nullptr : Value.Sec;
79 // We want to set symbol values early if we can. This allows us to use symbols
80 // as variables in linker scripts. Doing so allows us to write expressions
81 // like this: `alignment = 16; . = ALIGN(., alignment)`
82 uint64_t SymValue = Value.isAbsolute() ? Value.getValue() : 0;
83 replaceBody<DefinedRegular>(Sym, Cmd->Name, /*IsLocal=*/false, Visibility,
84 STT_NOTYPE, SymValue, 0, Sec, nullptr);
88 OutputSectionCommand *
89 LinkerScript::createOutputSectionCommand(StringRef Name, StringRef Location) {
90 OutputSectionCommand *&CmdRef = NameToOutputSectionCommand[Name];
91 OutputSectionCommand *Cmd;
92 if (CmdRef && CmdRef->Location.empty()) {
93 // There was a forward reference.
96 Cmd = make<OutputSectionCommand>(Name);
100 Cmd->Location = Location;
104 OutputSectionCommand *
105 LinkerScript::getOrCreateOutputSectionCommand(StringRef Name) {
106 OutputSectionCommand *&CmdRef = NameToOutputSectionCommand[Name];
108 CmdRef = make<OutputSectionCommand>(Name);
112 void LinkerScript::setDot(Expr E, const Twine &Loc, bool InSec) {
113 uint64_t Val = E().getValue();
114 if (Val < Dot && InSec)
115 error(Loc + ": unable to move location counter backward for: " +
116 CurAddressState->OutSec->Name);
118 // Update to location counter means update to section size.
120 CurAddressState->OutSec->Size = Dot - CurAddressState->OutSec->Addr;
123 // Sets value of a symbol. Two kinds of symbols are processed: synthetic
124 // symbols, whose value is an offset from beginning of section and regular
125 // symbols whose value is absolute.
126 void LinkerScript::assignSymbol(SymbolAssignment *Cmd, bool InSec) {
127 if (Cmd->Name == ".") {
128 setDot(Cmd->Expression, Cmd->Location, InSec);
135 auto *Sym = cast<DefinedRegular>(Cmd->Sym);
136 ExprValue V = Cmd->Expression();
137 if (V.isAbsolute()) {
138 Sym->Value = V.getValue();
140 Sym->Section = V.Sec;
141 Sym->Value = alignTo(V.Val, V.Alignment);
145 static SymbolBody *findSymbol(StringRef S) {
146 switch (Config->EKind) {
148 return Symtab<ELF32LE>::X->find(S);
150 return Symtab<ELF32BE>::X->find(S);
152 return Symtab<ELF64LE>::X->find(S);
154 return Symtab<ELF64BE>::X->find(S);
156 llvm_unreachable("unknown Config->EKind");
160 static SymbolBody *addRegularSymbol(SymbolAssignment *Cmd) {
161 switch (Config->EKind) {
163 return addRegular<ELF32LE>(Cmd);
165 return addRegular<ELF32BE>(Cmd);
167 return addRegular<ELF64LE>(Cmd);
169 return addRegular<ELF64BE>(Cmd);
171 llvm_unreachable("unknown Config->EKind");
175 void LinkerScript::addSymbol(SymbolAssignment *Cmd) {
176 if (Cmd->Name == ".")
179 // If a symbol was in PROVIDE(), we need to define it only when
180 // it is a referenced undefined symbol.
181 SymbolBody *B = findSymbol(Cmd->Name);
182 if (Cmd->Provide && (!B || B->isDefined()))
185 Cmd->Sym = addRegularSymbol(Cmd);
188 bool SymbolAssignment::classof(const BaseCommand *C) {
189 return C->Kind == AssignmentKind;
192 bool OutputSectionCommand::classof(const BaseCommand *C) {
193 return C->Kind == OutputSectionKind;
196 // Fill [Buf, Buf + Size) with Filler.
197 // This is used for linker script "=fillexp" command.
198 static void fill(uint8_t *Buf, size_t Size, uint32_t Filler) {
200 for (; I + 4 < Size; I += 4)
201 memcpy(Buf + I, &Filler, 4);
202 memcpy(Buf + I, &Filler, Size - I);
205 bool InputSectionDescription::classof(const BaseCommand *C) {
206 return C->Kind == InputSectionKind;
209 bool AssertCommand::classof(const BaseCommand *C) {
210 return C->Kind == AssertKind;
213 bool BytesDataCommand::classof(const BaseCommand *C) {
214 return C->Kind == BytesDataKind;
217 static StringRef basename(InputSectionBase *S) {
219 return sys::path::filename(S->File->getName());
223 bool LinkerScript::shouldKeep(InputSectionBase *S) {
224 for (InputSectionDescription *ID : Opt.KeptSections)
225 if (ID->FilePat.match(basename(S)))
226 for (SectionPattern &P : ID->SectionPatterns)
227 if (P.SectionPat.match(S->Name))
232 // If an input string is in the form of "foo.N" where N is a number,
233 // return N. Otherwise, returns 65536, which is one greater than the
235 static int getPriority(StringRef S) {
236 size_t Pos = S.rfind('.');
237 if (Pos == StringRef::npos)
240 if (!to_integer(S.substr(Pos + 1), V, 10))
245 // A helper function for the SORT() command.
246 static std::function<bool(InputSectionBase *, InputSectionBase *)>
247 getComparator(SortSectionPolicy K) {
249 case SortSectionPolicy::Alignment:
250 return [](InputSectionBase *A, InputSectionBase *B) {
251 // ">" is not a mistake. Sections with larger alignments are placed
252 // before sections with smaller alignments in order to reduce the
253 // amount of padding necessary. This is compatible with GNU.
254 return A->Alignment > B->Alignment;
256 case SortSectionPolicy::Name:
257 return [](InputSectionBase *A, InputSectionBase *B) {
258 return A->Name < B->Name;
260 case SortSectionPolicy::Priority:
261 return [](InputSectionBase *A, InputSectionBase *B) {
262 return getPriority(A->Name) < getPriority(B->Name);
265 llvm_unreachable("unknown sort policy");
269 // A helper function for the SORT() command.
270 static bool matchConstraints(ArrayRef<InputSectionBase *> Sections,
271 ConstraintKind Kind) {
272 if (Kind == ConstraintKind::NoConstraint)
275 bool IsRW = llvm::any_of(Sections, [](InputSectionBase *Sec) {
276 return static_cast<InputSectionBase *>(Sec)->Flags & SHF_WRITE;
279 return (IsRW && Kind == ConstraintKind::ReadWrite) ||
280 (!IsRW && Kind == ConstraintKind::ReadOnly);
283 static void sortSections(InputSection **Begin, InputSection **End,
284 SortSectionPolicy K) {
285 if (K != SortSectionPolicy::Default && K != SortSectionPolicy::None)
286 std::stable_sort(Begin, End, getComparator(K));
289 // Compute and remember which sections the InputSectionDescription matches.
290 std::vector<InputSection *>
291 LinkerScript::computeInputSections(const InputSectionDescription *Cmd) {
292 std::vector<InputSection *> Ret;
294 // Collects all sections that satisfy constraints of Cmd.
295 for (const SectionPattern &Pat : Cmd->SectionPatterns) {
296 size_t SizeBefore = Ret.size();
298 for (InputSectionBase *Sec : InputSections) {
303 reportDiscarded(Sec);
307 // For -emit-relocs we have to ignore entries like
308 // .rela.dyn : { *(.rela.data) }
309 // which are common because they are in the default bfd script.
310 if (Sec->Type == SHT_REL || Sec->Type == SHT_RELA)
313 StringRef Filename = basename(Sec);
314 if (!Cmd->FilePat.match(Filename) ||
315 Pat.ExcludedFilePat.match(Filename) ||
316 !Pat.SectionPat.match(Sec->Name))
319 Ret.push_back(cast<InputSection>(Sec));
320 Sec->Assigned = true;
323 // Sort sections as instructed by SORT-family commands and --sort-section
324 // option. Because SORT-family commands can be nested at most two depth
325 // (e.g. SORT_BY_NAME(SORT_BY_ALIGNMENT(.text.*))) and because the command
326 // line option is respected even if a SORT command is given, the exact
327 // behavior we have here is a bit complicated. Here are the rules.
329 // 1. If two SORT commands are given, --sort-section is ignored.
330 // 2. If one SORT command is given, and if it is not SORT_NONE,
331 // --sort-section is handled as an inner SORT command.
332 // 3. If one SORT command is given, and if it is SORT_NONE, don't sort.
333 // 4. If no SORT command is given, sort according to --sort-section.
334 InputSection **Begin = Ret.data() + SizeBefore;
335 InputSection **End = Ret.data() + Ret.size();
336 if (Pat.SortOuter != SortSectionPolicy::None) {
337 if (Pat.SortInner == SortSectionPolicy::Default)
338 sortSections(Begin, End, Config->SortSection);
340 sortSections(Begin, End, Pat.SortInner);
341 sortSections(Begin, End, Pat.SortOuter);
347 void LinkerScript::discard(ArrayRef<InputSectionBase *> V) {
348 for (InputSectionBase *S : V) {
350 if (S == InX::ShStrTab || S == InX::Dynamic || S == InX::DynSymTab ||
352 error("discarding " + S->Name + " section is not allowed");
353 discard(S->DependentSections);
357 std::vector<InputSectionBase *>
358 LinkerScript::createInputSectionList(OutputSectionCommand &OutCmd) {
359 std::vector<InputSectionBase *> Ret;
361 for (BaseCommand *Base : OutCmd.Commands) {
362 auto *Cmd = dyn_cast<InputSectionDescription>(Base);
366 Cmd->Sections = computeInputSections(Cmd);
367 Ret.insert(Ret.end(), Cmd->Sections.begin(), Cmd->Sections.end());
373 void LinkerScript::processCommands(OutputSectionFactory &Factory) {
374 // A symbol can be assigned before any section is mentioned in the linker
375 // script. In an DSO, the symbol values are addresses, so the only important
376 // section values are:
379 // * Any value meaning a regular section.
380 // To handle that, create a dummy aether section that fills the void before
381 // the linker scripts switches to another section. It has an index of one
382 // which will map to whatever the first actual section is.
383 Aether = make<OutputSection>("", 0, SHF_ALLOC);
384 Aether->SectionIndex = 1;
385 auto State = make_unique<AddressState>(Opt);
386 // CurAddressState captures the local AddressState and makes it accessible
387 // deliberately. This is needed as there are some cases where we cannot just
388 // thread the current state through to a lambda function created by the
390 CurAddressState = State.get();
391 CurAddressState->OutSec = Aether;
394 for (size_t I = 0; I < Opt.Commands.size(); ++I) {
395 // Handle symbol assignments outside of any output section.
396 if (auto *Cmd = dyn_cast<SymbolAssignment>(Opt.Commands[I])) {
401 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I])) {
402 std::vector<InputSectionBase *> V = createInputSectionList(*Cmd);
404 // The output section name `/DISCARD/' is special.
405 // Any input section assigned to it is discarded.
406 if (Cmd->Name == "/DISCARD/") {
411 // This is for ONLY_IF_RO and ONLY_IF_RW. An output section directive
412 // ".foo : ONLY_IF_R[OW] { ... }" is handled only if all member input
413 // sections satisfy a given constraint. If not, a directive is handled
414 // as if it wasn't present from the beginning.
416 // Because we'll iterate over Commands many more times, the easiest
417 // way to "make it as if it wasn't present" is to just remove it.
418 if (!matchConstraints(V, Cmd->Constraint)) {
419 for (InputSectionBase *S : V)
421 Opt.Commands.erase(Opt.Commands.begin() + I);
426 // A directive may contain symbol definitions like this:
427 // ".foo : { ...; bar = .; }". Handle them.
428 for (BaseCommand *Base : Cmd->Commands)
429 if (auto *OutCmd = dyn_cast<SymbolAssignment>(Base))
432 // Handle subalign (e.g. ".foo : SUBALIGN(32) { ... }"). If subalign
433 // is given, input sections are aligned to that value, whether the
434 // given value is larger or smaller than the original section alignment.
435 if (Cmd->SubalignExpr) {
436 uint32_t Subalign = Cmd->SubalignExpr().getValue();
437 for (InputSectionBase *S : V)
438 S->Alignment = Subalign;
441 // Add input sections to an output section.
442 for (InputSectionBase *S : V)
443 Factory.addInputSec(S, Cmd->Name, Cmd->Sec);
444 if (OutputSection *Sec = Cmd->Sec) {
445 assert(Sec->SectionIndex == INT_MAX);
446 Sec->SectionIndex = I;
448 Sec->Type = SHT_NOBITS;
449 SecToCommand[Sec] = Cmd;
453 CurAddressState = nullptr;
456 void LinkerScript::fabricateDefaultCommands() {
457 std::vector<BaseCommand *> Commands;
459 // Define start address
460 uint64_t StartAddr = -1;
462 // The Sections with -T<section> have been sorted in order of ascending
463 // address. We must lower StartAddr if the lowest -T<section address> as
464 // calls to setDot() must be monotonically increasing.
465 for (auto &KV : Config->SectionStartMap)
466 StartAddr = std::min(StartAddr, KV.second);
468 Commands.push_back(make<SymbolAssignment>(
471 return std::min(StartAddr, Config->ImageBase + elf::getHeaderSize());
475 // For each OutputSection that needs a VA fabricate an OutputSectionCommand
476 // with an InputSectionDescription describing the InputSections
477 for (OutputSection *Sec : OutputSections) {
478 auto *OSCmd = createOutputSectionCommand(Sec->Name, "<internal>");
480 SecToCommand[Sec] = OSCmd;
482 Commands.push_back(OSCmd);
483 if (Sec->Sections.size()) {
484 auto *ISD = make<InputSectionDescription>("");
485 OSCmd->Commands.push_back(ISD);
486 for (InputSection *ISec : Sec->Sections) {
487 ISD->Sections.push_back(ISec);
488 ISec->Assigned = true;
492 // SECTIONS commands run before other non SECTIONS commands
493 Commands.insert(Commands.end(), Opt.Commands.begin(), Opt.Commands.end());
494 Opt.Commands = std::move(Commands);
497 // Add sections that didn't match any sections command.
498 void LinkerScript::addOrphanSections(OutputSectionFactory &Factory) {
499 unsigned NumCommands = Opt.Commands.size();
500 for (InputSectionBase *S : InputSections) {
501 if (!S->Live || S->Parent)
503 StringRef Name = getOutputSectionName(S->Name);
504 auto End = Opt.Commands.begin() + NumCommands;
505 auto I = std::find_if(Opt.Commands.begin(), End, [&](BaseCommand *Base) {
506 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
507 return Cmd->Name == Name;
510 OutputSectionCommand *Cmd;
512 Factory.addInputSec(S, Name);
513 OutputSection *Sec = S->getOutputSection();
514 assert(Sec->SectionIndex == INT_MAX);
515 OutputSectionCommand *&CmdRef = SecToCommand[Sec];
517 CmdRef = createOutputSectionCommand(Sec->Name, "<internal>");
519 Opt.Commands.push_back(CmdRef);
523 Cmd = cast<OutputSectionCommand>(*I);
524 Factory.addInputSec(S, Name, Cmd->Sec);
525 if (OutputSection *Sec = Cmd->Sec) {
526 SecToCommand[Sec] = Cmd;
527 unsigned Index = std::distance(Opt.Commands.begin(), I);
528 assert(Sec->SectionIndex == INT_MAX || Sec->SectionIndex == Index);
529 Sec->SectionIndex = Index;
532 auto *ISD = make<InputSectionDescription>("");
533 ISD->Sections.push_back(cast<InputSection>(S));
534 Cmd->Commands.push_back(ISD);
538 uint64_t LinkerScript::advance(uint64_t Size, unsigned Align) {
539 bool IsTbss = (CurAddressState->OutSec->Flags & SHF_TLS) &&
540 CurAddressState->OutSec->Type == SHT_NOBITS;
541 uint64_t Start = IsTbss ? Dot + CurAddressState->ThreadBssOffset : Dot;
542 Start = alignTo(Start, Align);
543 uint64_t End = Start + Size;
546 CurAddressState->ThreadBssOffset = End - Dot;
552 void LinkerScript::output(InputSection *S) {
553 uint64_t Pos = advance(S->getSize(), S->Alignment);
554 S->OutSecOff = Pos - S->getSize() - CurAddressState->OutSec->Addr;
556 // Update output section size after adding each section. This is so that
557 // SIZEOF works correctly in the case below:
558 // .foo { *(.aaa) a = SIZEOF(.foo); *(.bbb) }
559 CurAddressState->OutSec->Size = Pos - CurAddressState->OutSec->Addr;
561 // If there is a memory region associated with this input section, then
562 // place the section in that region and update the region index.
563 if (CurAddressState->MemRegion) {
564 uint64_t &CurOffset =
565 CurAddressState->MemRegionOffset[CurAddressState->MemRegion];
566 CurOffset += CurAddressState->OutSec->Size;
567 uint64_t CurSize = CurOffset - CurAddressState->MemRegion->Origin;
568 if (CurSize > CurAddressState->MemRegion->Length) {
569 uint64_t OverflowAmt = CurSize - CurAddressState->MemRegion->Length;
570 error("section '" + CurAddressState->OutSec->Name +
571 "' will not fit in region '" + CurAddressState->MemRegion->Name +
572 "': overflowed by " + Twine(OverflowAmt) + " bytes");
577 void LinkerScript::switchTo(OutputSection *Sec) {
578 if (CurAddressState->OutSec == Sec)
581 CurAddressState->OutSec = Sec;
582 CurAddressState->OutSec->Addr =
583 advance(0, CurAddressState->OutSec->Alignment);
585 // If neither AT nor AT> is specified for an allocatable section, the linker
586 // will set the LMA such that the difference between VMA and LMA for the
587 // section is the same as the preceding output section in the same region
588 // https://sourceware.org/binutils/docs-2.20/ld/Output-Section-LMA.html
589 if (CurAddressState->LMAOffset)
590 CurAddressState->OutSec->LMAOffset = CurAddressState->LMAOffset();
593 void LinkerScript::process(BaseCommand &Base) {
594 // This handles the assignments to symbol or to the dot.
595 if (auto *Cmd = dyn_cast<SymbolAssignment>(&Base)) {
596 assignSymbol(Cmd, true);
600 // Handle BYTE(), SHORT(), LONG(), or QUAD().
601 if (auto *Cmd = dyn_cast<BytesDataCommand>(&Base)) {
602 Cmd->Offset = Dot - CurAddressState->OutSec->Addr;
604 CurAddressState->OutSec->Size = Dot - CurAddressState->OutSec->Addr;
609 if (auto *Cmd = dyn_cast<AssertCommand>(&Base)) {
614 // Handle a single input section description command.
615 // It calculates and assigns the offsets for each section and also
616 // updates the output section size.
617 auto &Cmd = cast<InputSectionDescription>(Base);
618 for (InputSection *Sec : Cmd.Sections) {
619 // We tentatively added all synthetic sections at the beginning and removed
620 // empty ones afterwards (because there is no way to know whether they were
621 // going be empty or not other than actually running linker scripts.)
622 // We need to ignore remains of empty sections.
623 if (auto *S = dyn_cast<SyntheticSection>(Sec))
629 assert(CurAddressState->OutSec == Sec->getParent());
634 // This function searches for a memory region to place the given output
635 // section in. If found, a pointer to the appropriate memory region is
636 // returned. Otherwise, a nullptr is returned.
637 MemoryRegion *LinkerScript::findMemoryRegion(OutputSectionCommand *Cmd) {
638 // If a memory region name was specified in the output section command,
639 // then try to find that region first.
640 if (!Cmd->MemoryRegionName.empty()) {
641 auto It = Opt.MemoryRegions.find(Cmd->MemoryRegionName);
642 if (It != Opt.MemoryRegions.end())
644 error("memory region '" + Cmd->MemoryRegionName + "' not declared");
648 // If at least one memory region is defined, all sections must
649 // belong to some memory region. Otherwise, we don't need to do
650 // anything for memory regions.
651 if (Opt.MemoryRegions.empty())
654 OutputSection *Sec = Cmd->Sec;
655 // See if a region can be found by matching section flags.
656 for (auto &Pair : Opt.MemoryRegions) {
657 MemoryRegion &M = Pair.second;
658 if ((M.Flags & Sec->Flags) && (M.NegFlags & Sec->Flags) == 0)
662 // Otherwise, no suitable region was found.
663 if (Sec->Flags & SHF_ALLOC)
664 error("no memory region specified for section '" + Sec->Name + "'");
668 // This function assigns offsets to input sections and an output section
669 // for a single sections command (e.g. ".text { *(.text); }").
670 void LinkerScript::assignOffsets(OutputSectionCommand *Cmd) {
671 OutputSection *Sec = Cmd->Sec;
675 if (!(Sec->Flags & SHF_ALLOC))
677 else if (Cmd->AddrExpr)
678 setDot(Cmd->AddrExpr, Cmd->Location, false);
682 CurAddressState->LMAOffset = [=] { return Cmd->LMAExpr().getValue() - D; };
685 CurAddressState->MemRegion = Cmd->MemRegion;
686 if (CurAddressState->MemRegion)
687 Dot = CurAddressState->MemRegionOffset[CurAddressState->MemRegion];
690 // We do not support custom layout for compressed debug sectons.
691 // At this point we already know their size and have compressed content.
692 if (CurAddressState->OutSec->Flags & SHF_COMPRESSED)
695 for (BaseCommand *C : Cmd->Commands)
699 void LinkerScript::removeEmptyCommands() {
700 // It is common practice to use very generic linker scripts. So for any
701 // given run some of the output sections in the script will be empty.
702 // We could create corresponding empty output sections, but that would
703 // clutter the output.
704 // We instead remove trivially empty sections. The bfd linker seems even
705 // more aggressive at removing them.
706 auto Pos = std::remove_if(
707 Opt.Commands.begin(), Opt.Commands.end(), [&](BaseCommand *Base) {
708 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base))
709 return Cmd->Sec == nullptr;
712 Opt.Commands.erase(Pos, Opt.Commands.end());
715 static bool isAllSectionDescription(const OutputSectionCommand &Cmd) {
716 for (BaseCommand *Base : Cmd.Commands)
717 if (!isa<InputSectionDescription>(*Base))
722 void LinkerScript::adjustSectionsBeforeSorting() {
723 // If the output section contains only symbol assignments, create a
724 // corresponding output section. The bfd linker seems to only create them if
725 // '.' is assigned to, but creating these section should not have any bad
726 // consequeces and gives us a section to put the symbol in.
727 uint64_t Flags = SHF_ALLOC;
729 for (int I = 0, E = Opt.Commands.size(); I != E; ++I) {
730 auto *Cmd = dyn_cast<OutputSectionCommand>(Opt.Commands[I]);
733 if (OutputSection *Sec = Cmd->Sec) {
738 if (isAllSectionDescription(*Cmd))
741 auto *OutSec = make<OutputSection>(Cmd->Name, SHT_PROGBITS, Flags);
742 OutSec->SectionIndex = I;
744 SecToCommand[OutSec] = Cmd;
748 void LinkerScript::adjustSectionsAfterSorting() {
749 // Try and find an appropriate memory region to assign offsets in.
750 for (BaseCommand *Base : Opt.Commands) {
751 if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base)) {
752 Cmd->MemRegion = findMemoryRegion(Cmd);
753 // Handle align (e.g. ".foo : ALIGN(16) { ... }").
755 Cmd->Sec->updateAlignment(Cmd->AlignExpr().getValue());
759 // If output section command doesn't specify any segments,
760 // and we haven't previously assigned any section to segment,
761 // then we simply assign section to the very first load segment.
762 // Below is an example of such linker script:
763 // PHDRS { seg PT_LOAD; }
764 // SECTIONS { .aaa : { *(.aaa) } }
765 std::vector<StringRef> DefPhdrs;
767 std::find_if(Opt.PhdrsCommands.begin(), Opt.PhdrsCommands.end(),
768 [](const PhdrsCommand &Cmd) { return Cmd.Type == PT_LOAD; });
769 if (FirstPtLoad != Opt.PhdrsCommands.end())
770 DefPhdrs.push_back(FirstPtLoad->Name);
772 // Walk the commands and propagate the program headers to commands that don't
773 // explicitly specify them.
774 for (BaseCommand *Base : Opt.Commands) {
775 auto *Cmd = dyn_cast<OutputSectionCommand>(Base);
779 if (Cmd->Phdrs.empty()) {
780 OutputSection *Sec = Cmd->Sec;
781 // To match the bfd linker script behaviour, only propagate program
782 // headers to sections that are allocated.
783 if (Sec && (Sec->Flags & SHF_ALLOC))
784 Cmd->Phdrs = DefPhdrs;
786 DefPhdrs = Cmd->Phdrs;
790 removeEmptyCommands();
793 void LinkerScript::processNonSectionCommands() {
794 for (BaseCommand *Base : Opt.Commands) {
795 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base))
796 assignSymbol(Cmd, false);
797 else if (auto *Cmd = dyn_cast<AssertCommand>(Base))
802 void LinkerScript::allocateHeaders(std::vector<PhdrEntry> &Phdrs) {
803 uint64_t Min = std::numeric_limits<uint64_t>::max();
804 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
805 OutputSection *Sec = Cmd->Sec;
806 if (Sec->Flags & SHF_ALLOC)
807 Min = std::min<uint64_t>(Min, Sec->Addr);
810 auto FirstPTLoad = llvm::find_if(
811 Phdrs, [](const PhdrEntry &E) { return E.p_type == PT_LOAD; });
812 if (FirstPTLoad == Phdrs.end())
815 uint64_t HeaderSize = getHeaderSize();
816 if (HeaderSize <= Min || Script->hasPhdrsCommands()) {
817 Min = alignDown(Min - HeaderSize, Config->MaxPageSize);
818 Out::ElfHeader->Addr = Min;
819 Out::ProgramHeaders->Addr = Min + Out::ElfHeader->Size;
823 assert(FirstPTLoad->First == Out::ElfHeader);
824 OutputSection *ActualFirst = nullptr;
825 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
826 OutputSection *Sec = Cmd->Sec;
827 if (Sec->FirstInPtLoad == Out::ElfHeader) {
833 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
834 OutputSection *Sec = Cmd->Sec;
835 if (Sec->FirstInPtLoad == Out::ElfHeader)
836 Sec->FirstInPtLoad = ActualFirst;
838 FirstPTLoad->First = ActualFirst;
840 Phdrs.erase(FirstPTLoad);
843 auto PhdrI = llvm::find_if(
844 Phdrs, [](const PhdrEntry &E) { return E.p_type == PT_PHDR; });
845 if (PhdrI != Phdrs.end())
849 LinkerScript::AddressState::AddressState(const ScriptConfiguration &Opt) {
850 for (auto &MRI : Opt.MemoryRegions) {
851 const MemoryRegion *MR = &MRI.second;
852 MemRegionOffset[MR] = MR->Origin;
856 void LinkerScript::assignAddresses() {
857 // Assign addresses as instructed by linker script SECTIONS sub-commands.
859 auto State = make_unique<AddressState>(Opt);
860 // CurAddressState captures the local AddressState and makes it accessible
861 // deliberately. This is needed as there are some cases where we cannot just
862 // thread the current state through to a lambda function created by the
864 CurAddressState = State.get();
865 ErrorOnMissingSection = true;
868 for (BaseCommand *Base : Opt.Commands) {
869 if (auto *Cmd = dyn_cast<SymbolAssignment>(Base)) {
870 assignSymbol(Cmd, false);
874 if (auto *Cmd = dyn_cast<AssertCommand>(Base)) {
879 auto *Cmd = cast<OutputSectionCommand>(Base);
882 CurAddressState = nullptr;
885 // Creates program headers as instructed by PHDRS linker script command.
886 std::vector<PhdrEntry> LinkerScript::createPhdrs() {
887 std::vector<PhdrEntry> Ret;
889 // Process PHDRS and FILEHDR keywords because they are not
890 // real output sections and cannot be added in the following loop.
891 for (const PhdrsCommand &Cmd : Opt.PhdrsCommands) {
892 Ret.emplace_back(Cmd.Type, Cmd.Flags == UINT_MAX ? PF_R : Cmd.Flags);
893 PhdrEntry &Phdr = Ret.back();
896 Phdr.add(Out::ElfHeader);
898 Phdr.add(Out::ProgramHeaders);
901 Phdr.p_paddr = Cmd.LMAExpr().getValue();
906 // Add output sections to program headers.
907 for (OutputSectionCommand *Cmd : OutputSectionCommands) {
908 // Assign headers specified by linker script
909 for (size_t Id : getPhdrIndices(Cmd)) {
910 OutputSection *Sec = Cmd->Sec;
912 if (Opt.PhdrsCommands[Id].Flags == UINT_MAX)
913 Ret[Id].p_flags |= Sec->getPhdrFlags();
919 bool LinkerScript::ignoreInterpSection() {
920 // Ignore .interp section in case we have PHDRS specification
921 // and PT_INTERP isn't listed.
922 if (Opt.PhdrsCommands.empty())
924 for (PhdrsCommand &Cmd : Opt.PhdrsCommands)
925 if (Cmd.Type == PT_INTERP)
930 OutputSectionCommand *LinkerScript::getCmd(OutputSection *Sec) const {
931 auto I = SecToCommand.find(Sec);
932 if (I == SecToCommand.end())
937 void OutputSectionCommand::sort(std::function<int(InputSectionBase *S)> Order) {
938 typedef std::pair<unsigned, InputSection *> Pair;
939 auto Comp = [](const Pair &A, const Pair &B) { return A.first < B.first; };
942 assert(Commands.size() == 1);
943 auto *ISD = cast<InputSectionDescription>(Commands[0]);
944 for (InputSection *S : ISD->Sections)
945 V.push_back({Order(S), S});
946 std::stable_sort(V.begin(), V.end(), Comp);
947 ISD->Sections.clear();
949 ISD->Sections.push_back(P.second);
952 // Returns true if S matches /Filename.?\.o$/.
953 static bool isCrtBeginEnd(StringRef S, StringRef Filename) {
954 if (!S.endswith(".o"))
957 if (S.endswith(Filename))
959 return !S.empty() && S.drop_back().endswith(Filename);
962 static bool isCrtbegin(StringRef S) { return isCrtBeginEnd(S, "crtbegin"); }
963 static bool isCrtend(StringRef S) { return isCrtBeginEnd(S, "crtend"); }
965 // .ctors and .dtors are sorted by this priority from highest to lowest.
967 // 1. The section was contained in crtbegin (crtbegin contains
968 // some sentinel value in its .ctors and .dtors so that the runtime
969 // can find the beginning of the sections.)
971 // 2. The section has an optional priority value in the form of ".ctors.N"
972 // or ".dtors.N" where N is a number. Unlike .{init,fini}_array,
973 // they are compared as string rather than number.
975 // 3. The section is just ".ctors" or ".dtors".
977 // 4. The section was contained in crtend, which contains an end marker.
979 // In an ideal world, we don't need this function because .init_array and
980 // .ctors are duplicate features (and .init_array is newer.) However, there
981 // are too many real-world use cases of .ctors, so we had no choice to
982 // support that with this rather ad-hoc semantics.
983 static bool compCtors(const InputSection *A, const InputSection *B) {
984 bool BeginA = isCrtbegin(A->File->getName());
985 bool BeginB = isCrtbegin(B->File->getName());
986 if (BeginA != BeginB)
988 bool EndA = isCrtend(A->File->getName());
989 bool EndB = isCrtend(B->File->getName());
992 StringRef X = A->Name;
993 StringRef Y = B->Name;
994 assert(X.startswith(".ctors") || X.startswith(".dtors"));
995 assert(Y.startswith(".ctors") || Y.startswith(".dtors"));
998 if (X.empty() && Y.empty())
1003 // Sorts input sections by the special rules for .ctors and .dtors.
1004 // Unfortunately, the rules are different from the one for .{init,fini}_array.
1005 // Read the comment above.
1006 void OutputSectionCommand::sortCtorsDtors() {
1007 assert(Commands.size() == 1);
1008 auto *ISD = cast<InputSectionDescription>(Commands[0]);
1009 std::stable_sort(ISD->Sections.begin(), ISD->Sections.end(), compCtors);
1012 // Sorts input sections by section name suffixes, so that .foo.N comes
1013 // before .foo.M if N < M. Used to sort .{init,fini}_array.N sections.
1014 // We want to keep the original order if the priorities are the same
1015 // because the compiler keeps the original initialization order in a
1016 // translation unit and we need to respect that.
1017 // For more detail, read the section of the GCC's manual about init_priority.
1018 void OutputSectionCommand::sortInitFini() {
1019 // Sort sections by priority.
1020 sort([](InputSectionBase *S) { return getPriority(S->Name); });
1023 uint32_t OutputSectionCommand::getFiller() {
1026 if (Sec->Flags & SHF_EXECINSTR)
1027 return Target->TrapInstr;
1031 static void writeInt(uint8_t *Buf, uint64_t Data, uint64_t Size) {
1035 write16(Buf, Data, Config->Endianness);
1037 write32(Buf, Data, Config->Endianness);
1039 write64(Buf, Data, Config->Endianness);
1041 llvm_unreachable("unsupported Size argument");
1044 static bool compareByFilePosition(InputSection *A, InputSection *B) {
1045 // Synthetic doesn't have link order dependecy, stable_sort will keep it last
1046 if (A->kind() == InputSectionBase::Synthetic ||
1047 B->kind() == InputSectionBase::Synthetic)
1049 InputSection *LA = A->getLinkOrderDep();
1050 InputSection *LB = B->getLinkOrderDep();
1051 OutputSection *AOut = LA->getParent();
1052 OutputSection *BOut = LB->getParent();
1054 return AOut->SectionIndex < BOut->SectionIndex;
1055 return LA->OutSecOff < LB->OutSecOff;
1058 template <class ELFT>
1059 static void finalizeShtGroup(OutputSection *OS,
1060 ArrayRef<InputSection *> Sections) {
1061 assert(Config->Relocatable && Sections.size() == 1);
1063 // sh_link field for SHT_GROUP sections should contain the section index of
1064 // the symbol table.
1065 OS->Link = InX::SymTab->getParent()->SectionIndex;
1067 // sh_info then contain index of an entry in symbol table section which
1068 // provides signature of the section group.
1069 elf::ObjectFile<ELFT> *Obj = Sections[0]->getFile<ELFT>();
1070 ArrayRef<SymbolBody *> Symbols = Obj->getSymbols();
1071 OS->Info = InX::SymTab->getSymbolIndex(Symbols[Sections[0]->Info - 1]);
1074 template <class ELFT> void OutputSectionCommand::finalize() {
1075 // Link order may be distributed across several InputSectionDescriptions
1076 // but sort must consider them all at once.
1077 std::vector<InputSection **> ScriptSections;
1078 std::vector<InputSection *> Sections;
1079 for (BaseCommand *Base : Commands)
1080 if (auto *ISD = dyn_cast<InputSectionDescription>(Base))
1081 for (InputSection *&IS : ISD->Sections) {
1082 ScriptSections.push_back(&IS);
1083 Sections.push_back(IS);
1086 if ((Sec->Flags & SHF_LINK_ORDER)) {
1087 std::stable_sort(Sections.begin(), Sections.end(), compareByFilePosition);
1088 for (int I = 0, N = Sections.size(); I < N; ++I)
1089 *ScriptSections[I] = Sections[I];
1091 // We must preserve the link order dependency of sections with the
1092 // SHF_LINK_ORDER flag. The dependency is indicated by the sh_link field. We
1093 // need to translate the InputSection sh_link to the OutputSection sh_link,
1094 // all InputSections in the OutputSection have the same dependency.
1095 if (auto *D = Sections.front()->getLinkOrderDep())
1096 Sec->Link = D->getParent()->SectionIndex;
1099 uint32_t Type = Sec->Type;
1100 if (Type == SHT_GROUP) {
1101 finalizeShtGroup<ELFT>(Sec, Sections);
1105 if (!Config->CopyRelocs || (Type != SHT_RELA && Type != SHT_REL))
1108 InputSection *First = Sections[0];
1109 if (isa<SyntheticSection>(First))
1112 Sec->Link = InX::SymTab->getParent()->SectionIndex;
1113 // sh_info for SHT_REL[A] sections should contain the section header index of
1114 // the section to which the relocation applies.
1115 InputSectionBase *S = First->getRelocatedSection();
1116 Sec->Info = S->getOutputSection()->SectionIndex;
1117 Sec->Flags |= SHF_INFO_LINK;
1120 // Compress section contents if this section contains debug info.
1121 template <class ELFT> void OutputSectionCommand::maybeCompress() {
1122 typedef typename ELFT::Chdr Elf_Chdr;
1124 // Compress only DWARF debug sections.
1125 if (!Config->CompressDebugSections || (Sec->Flags & SHF_ALLOC) ||
1126 !Name.startswith(".debug_"))
1129 // Create a section header.
1130 Sec->ZDebugHeader.resize(sizeof(Elf_Chdr));
1131 auto *Hdr = reinterpret_cast<Elf_Chdr *>(Sec->ZDebugHeader.data());
1132 Hdr->ch_type = ELFCOMPRESS_ZLIB;
1133 Hdr->ch_size = Sec->Size;
1134 Hdr->ch_addralign = Sec->Alignment;
1136 // Write section contents to a temporary buffer and compress it.
1137 std::vector<uint8_t> Buf(Sec->Size);
1138 writeTo<ELFT>(Buf.data());
1139 if (Error E = zlib::compress(toStringRef(Buf), Sec->CompressedData))
1140 fatal("compress failed: " + llvm::toString(std::move(E)));
1142 // Update section headers.
1143 Sec->Size = sizeof(Elf_Chdr) + Sec->CompressedData.size();
1144 Sec->Flags |= SHF_COMPRESSED;
1147 template <class ELFT> void OutputSectionCommand::writeTo(uint8_t *Buf) {
1148 if (Sec->Type == SHT_NOBITS)
1153 // If -compress-debug-section is specified and if this is a debug seciton,
1154 // we've already compressed section contents. If that's the case,
1155 // just write it down.
1156 if (!Sec->CompressedData.empty()) {
1157 memcpy(Buf, Sec->ZDebugHeader.data(), Sec->ZDebugHeader.size());
1158 memcpy(Buf + Sec->ZDebugHeader.size(), Sec->CompressedData.data(),
1159 Sec->CompressedData.size());
1163 // Write leading padding.
1164 std::vector<InputSection *> Sections;
1165 for (BaseCommand *Cmd : Commands)
1166 if (auto *ISD = dyn_cast<InputSectionDescription>(Cmd))
1167 for (InputSection *IS : ISD->Sections)
1169 Sections.push_back(IS);
1170 uint32_t Filler = getFiller();
1172 fill(Buf, Sections.empty() ? Sec->Size : Sections[0]->OutSecOff, Filler);
1174 parallelForEachN(0, Sections.size(), [=](size_t I) {
1175 InputSection *IS = Sections[I];
1176 IS->writeTo<ELFT>(Buf);
1178 // Fill gaps between sections.
1180 uint8_t *Start = Buf + IS->OutSecOff + IS->getSize();
1182 if (I + 1 == Sections.size())
1183 End = Buf + Sec->Size;
1185 End = Buf + Sections[I + 1]->OutSecOff;
1186 fill(Start, End - Start, Filler);
1190 // Linker scripts may have BYTE()-family commands with which you
1191 // can write arbitrary bytes to the output. Process them if any.
1192 for (BaseCommand *Base : Commands)
1193 if (auto *Data = dyn_cast<BytesDataCommand>(Base))
1194 writeInt(Buf + Data->Offset, Data->Expression().getValue(), Data->Size);
1197 ExprValue LinkerScript::getSymbolValue(const Twine &Loc, StringRef S) {
1199 return {CurAddressState->OutSec, Dot - CurAddressState->OutSec->Addr, Loc};
1200 if (SymbolBody *B = findSymbol(S)) {
1201 if (auto *D = dyn_cast<DefinedRegular>(B))
1202 return {D->Section, D->Value, Loc};
1203 if (auto *C = dyn_cast<DefinedCommon>(B))
1204 return {InX::Common, C->Offset, Loc};
1206 error(Loc + ": symbol not found: " + S);
1210 bool LinkerScript::isDefined(StringRef S) { return findSymbol(S) != nullptr; }
1212 static const size_t NoPhdr = -1;
1214 // Returns indices of ELF headers containing specific section. Each index is a
1215 // zero based number of ELF header listed within PHDRS {} script block.
1216 std::vector<size_t> LinkerScript::getPhdrIndices(OutputSectionCommand *Cmd) {
1217 std::vector<size_t> Ret;
1218 for (StringRef PhdrName : Cmd->Phdrs) {
1219 size_t Index = getPhdrIndex(Cmd->Location, PhdrName);
1220 if (Index != NoPhdr)
1221 Ret.push_back(Index);
1226 // Returns the index of the segment named PhdrName if found otherwise
1227 // NoPhdr. When not found, if PhdrName is not the special case value 'NONE'
1228 // (which can be used to explicitly specify that a section isn't assigned to a
1229 // segment) then error.
1230 size_t LinkerScript::getPhdrIndex(const Twine &Loc, StringRef PhdrName) {
1232 for (PhdrsCommand &Cmd : Opt.PhdrsCommands) {
1233 if (Cmd.Name == PhdrName)
1237 if (PhdrName != "NONE")
1238 error(Loc + ": section header '" + PhdrName + "' is not listed in PHDRS");
1242 template void OutputSectionCommand::writeTo<ELF32LE>(uint8_t *Buf);
1243 template void OutputSectionCommand::writeTo<ELF32BE>(uint8_t *Buf);
1244 template void OutputSectionCommand::writeTo<ELF64LE>(uint8_t *Buf);
1245 template void OutputSectionCommand::writeTo<ELF64BE>(uint8_t *Buf);
1247 template void OutputSectionCommand::maybeCompress<ELF32LE>();
1248 template void OutputSectionCommand::maybeCompress<ELF32BE>();
1249 template void OutputSectionCommand::maybeCompress<ELF64LE>();
1250 template void OutputSectionCommand::maybeCompress<ELF64BE>();
1252 template void OutputSectionCommand::finalize<ELF32LE>();
1253 template void OutputSectionCommand::finalize<ELF32BE>();
1254 template void OutputSectionCommand::finalize<ELF64LE>();
1255 template void OutputSectionCommand::finalize<ELF64BE>();