1 //===- InputSection.h -------------------------------------------*- C++ -*-===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #ifndef LLD_ELF_INPUT_SECTION_H
11 #define LLD_ELF_INPUT_SECTION_H
14 #include "Relocations.h"
16 #include "lld/Common/LLVM.h"
17 #include "llvm/ADT/CachedHashString.h"
18 #include "llvm/ADT/DenseSet.h"
19 #include "llvm/ADT/TinyPtrVector.h"
20 #include "llvm/Object/ELF.h"
21 #include "llvm/Support/Threading.h"
31 class SyntheticSection;
32 class MergeSyntheticSection;
33 template <class ELFT> class ObjFile;
36 // This is the base class of all sections that lld handles. Some are sections in
37 // input files, some are sections in the produced output file and some exist
38 // just as a convenience for implementing special ways of combining some
42 enum Kind { Regular, EHFrame, Merge, Synthetic, Output };
44 Kind kind() const { return (Kind)SectionKind; }
48 // This pointer points to the "real" instance of this instance.
49 // Usually Repl == this. However, if ICF merges two sections,
50 // Repl pointer of one section points to another section. So,
51 // if you need to get a pointer to this instance, do not use
52 // this but instead this->Repl.
55 unsigned SectionKind : 3;
57 // The next two bit fields are only used by InputSectionBase, but we
58 // put them here so the struct packs better.
60 // The garbage collector sets sections' Live bits.
61 // If GC is disabled, all sections are considered live by default.
66 // These corresponds to the fields in Elf_Shdr.
74 OutputSection *getOutputSection();
75 const OutputSection *getOutputSection() const {
76 return const_cast<SectionBase *>(this)->getOutputSection();
79 // Translate an offset in the input section to an offset in the output
81 uint64_t getOffset(uint64_t Offset) const;
84 SectionBase(Kind SectionKind, StringRef Name, uint64_t Flags,
85 uint64_t Entsize, uint64_t Alignment, uint32_t Type,
86 uint32_t Info, uint32_t Link)
87 : Name(Name), Repl(this), SectionKind(SectionKind), Live(false),
88 Bss(false), Alignment(Alignment), Flags(Flags), Entsize(Entsize),
89 Type(Type), Link(Link), Info(Info) {}
92 // This corresponds to a section of an input file.
93 class InputSectionBase : public SectionBase {
96 InputSectionBase(ObjFile<ELFT> &File, const typename ELFT::Shdr &Header,
97 StringRef Name, Kind SectionKind);
99 InputSectionBase(InputFile *File, uint64_t Flags, uint32_t Type,
100 uint64_t Entsize, uint32_t Link, uint32_t Info,
101 uint32_t Alignment, ArrayRef<uint8_t> Data, StringRef Name,
104 static bool classof(const SectionBase *S) { return S->kind() != Output; }
106 // The file which contains this section. It's dynamic type is always
107 // ObjFile<ELFT>, but in order to avoid ELFT, we use InputFile as
111 template <class ELFT> ObjFile<ELFT> *getFile() const {
112 return cast_or_null<ObjFile<ELFT>>(File);
115 ArrayRef<uint8_t> Data;
116 uint64_t getOffsetInFile() const;
118 // True if this section has already been placed to a linker script
119 // output section. This is needed because, in a linker script, you
120 // can refer to the same section more than once. For example, in
121 // the following linker script,
123 // .foo : { *(.text) }
124 // .bar : { *(.text) }
126 // .foo takes all .text sections, and .bar becomes empty. To achieve
127 // this, we need to memorize whether a section has been placed or
128 // not for each input section.
129 bool Assigned = false;
131 // Input sections are part of an output section. Special sections
132 // like .eh_frame and merge sections are first combined into a
133 // synthetic section that is then added to an output section. In all
134 // cases this points one level up.
135 SectionBase *Parent = nullptr;
137 // Relocations that refer to this section.
138 const void *FirstRelocation = nullptr;
139 unsigned NumRelocations : 31;
140 unsigned AreRelocsRela : 1;
142 template <class ELFT> ArrayRef<typename ELFT::Rel> rels() const {
143 assert(!AreRelocsRela);
144 return llvm::makeArrayRef(
145 static_cast<const typename ELFT::Rel *>(FirstRelocation),
149 template <class ELFT> ArrayRef<typename ELFT::Rela> relas() const {
150 assert(AreRelocsRela);
151 return llvm::makeArrayRef(
152 static_cast<const typename ELFT::Rela *>(FirstRelocation),
156 // InputSections that are dependent on us (reverse dependency for GC)
157 llvm::TinyPtrVector<InputSection *> DependentSections;
159 // Returns the size of this section (even if this is a common or BSS.)
160 size_t getSize() const;
162 InputSection *getLinkOrderDep() const;
164 // Compilers emit zlib-compressed debug sections if the -gz option
165 // is given. This function checks if this section is compressed, and
166 // if so, decompress in memory.
167 void maybeUncompress();
169 // Returns a source location string. Used to construct an error message.
170 template <class ELFT> std::string getLocation(uint64_t Offset);
171 std::string getSrcMsg(const Symbol &Sym, uint64_t Offset);
172 std::string getObjMsg(uint64_t Offset);
174 // Each section knows how to relocate itself. These functions apply
175 // relocations, assuming that Buf points to this section's copy in
176 // the mmap'ed output buffer.
177 template <class ELFT> void relocate(uint8_t *Buf, uint8_t *BufEnd);
178 void relocateAlloc(uint8_t *Buf, uint8_t *BufEnd);
180 // The native ELF reloc data type is not very convenient to handle.
181 // So we convert ELF reloc records to our own records in Relocations.cpp.
182 // This vector contains such "cooked" relocations.
183 std::vector<Relocation> Relocations;
185 template <typename T> llvm::ArrayRef<T> getDataAs() const {
186 size_t S = Data.size();
187 assert(S % sizeof(T) == 0);
188 return llvm::makeArrayRef<T>((const T *)Data.data(), S / sizeof(T));
192 // A pointer that owns uncompressed data if a section is compressed by zlib.
193 // Since the feature is not used often, this is usually a nullptr.
194 std::unique_ptr<char[]> UncompressBuf;
197 // SectionPiece represents a piece of splittable section contents.
198 // We allocate a lot of these and binary search on them. This means that they
199 // have to be as compact as possible, which is why we don't store the size (can
200 // be found by looking at the next one).
201 struct SectionPiece {
202 SectionPiece(size_t Off, uint32_t Hash, bool Live)
203 : InputOff(Off), Hash(Hash), OutputOff(-1),
204 Live(Live || !Config->GcSections) {}
208 int64_t OutputOff : 63;
212 static_assert(sizeof(SectionPiece) == 16, "SectionPiece is too big");
214 // This corresponds to a SHF_MERGE section of an input file.
215 class MergeInputSection : public InputSectionBase {
217 template <class ELFT>
218 MergeInputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
220 MergeInputSection(uint64_t Flags, uint32_t Type, uint64_t Entsize,
221 ArrayRef<uint8_t> Data, StringRef Name);
223 static bool classof(const SectionBase *S) { return S->kind() == Merge; }
224 void splitIntoPieces();
226 // Mark the piece at a given offset live. Used by GC.
227 void markLiveAt(uint64_t Offset) {
228 if (this->Flags & llvm::ELF::SHF_ALLOC)
229 LiveOffsets.insert(Offset);
232 // Translate an offset in the input section to an offset
233 // in the output section.
234 uint64_t getOffset(uint64_t Offset) const;
236 // Splittable sections are handled as a sequence of data
237 // rather than a single large blob of data.
238 std::vector<SectionPiece> Pieces;
240 // Returns I'th piece's data. This function is very hot when
241 // string merging is enabled, so we want to inline.
242 LLVM_ATTRIBUTE_ALWAYS_INLINE
243 llvm::CachedHashStringRef getData(size_t I) const {
244 size_t Begin = Pieces[I].InputOff;
246 (Pieces.size() - 1 == I) ? Data.size() : Pieces[I + 1].InputOff;
247 return {toStringRef(Data.slice(Begin, End - Begin)), Pieces[I].Hash};
250 // Returns the SectionPiece at a given input section offset.
251 SectionPiece *getSectionPiece(uint64_t Offset);
252 const SectionPiece *getSectionPiece(uint64_t Offset) const;
254 SyntheticSection *getParent() const;
257 void splitStrings(ArrayRef<uint8_t> A, size_t Size);
258 void splitNonStrings(ArrayRef<uint8_t> A, size_t Size);
260 mutable llvm::DenseMap<uint32_t, uint32_t> OffsetMap;
261 mutable llvm::once_flag InitOffsetMap;
263 llvm::DenseSet<uint64_t> LiveOffsets;
266 struct EhSectionPiece {
267 EhSectionPiece(size_t Off, InputSectionBase *Sec, uint32_t Size,
268 unsigned FirstRelocation)
269 : InputOff(Off), Sec(Sec), Size(Size), FirstRelocation(FirstRelocation) {}
271 ArrayRef<uint8_t> data() { return {Sec->Data.data() + this->InputOff, Size}; }
274 ssize_t OutputOff = -1;
275 InputSectionBase *Sec;
277 unsigned FirstRelocation;
280 // This corresponds to a .eh_frame section of an input file.
281 class EhInputSection : public InputSectionBase {
283 template <class ELFT>
284 EhInputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
286 static bool classof(const SectionBase *S) { return S->kind() == EHFrame; }
287 template <class ELFT> void split();
288 template <class ELFT, class RelTy> void split(ArrayRef<RelTy> Rels);
290 // Splittable sections are handled as a sequence of data
291 // rather than a single large blob of data.
292 std::vector<EhSectionPiece> Pieces;
294 SyntheticSection *getParent() const;
297 // This is a section that is added directly to an output section
298 // instead of needing special combination via a synthetic section. This
299 // includes all input sections with the exceptions of SHF_MERGE and
300 // .eh_frame. It also includes the synthetic sections themselves.
301 class InputSection : public InputSectionBase {
303 InputSection(InputFile *F, uint64_t Flags, uint32_t Type, uint32_t Alignment,
304 ArrayRef<uint8_t> Data, StringRef Name, Kind K = Regular);
305 template <class ELFT>
306 InputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
309 // Write this section to a mmap'ed file, assuming Buf is pointing to
310 // beginning of the output section.
311 template <class ELFT> void writeTo(uint8_t *Buf);
313 OutputSection *getParent() const;
315 // This variable has two usages. Initially, it represents an index in the
316 // OutputSection's InputSection list, and is used when ordering SHF_LINK_ORDER
317 // sections. After assignAddresses is called, it represents the offset from
318 // the beginning of the output section this section was assigned to.
319 uint64_t OutSecOff = 0;
321 static bool classof(const SectionBase *S);
323 InputSectionBase *getRelocatedSection();
325 template <class ELFT, class RelTy>
326 void relocateNonAlloc(uint8_t *Buf, llvm::ArrayRef<RelTy> Rels);
329 uint32_t Class[2] = {0, 0};
331 // Called by ICF to merge two input sections.
332 void replace(InputSection *Other);
334 static InputSection Discarded;
337 template <class ELFT, class RelTy>
338 void copyRelocations(uint8_t *Buf, llvm::ArrayRef<RelTy> Rels);
340 template <class ELFT> void copyShtGroup(uint8_t *Buf);
343 // The list of all input sections.
344 extern std::vector<InputSectionBase *> InputSections;
346 // Builds section order for handling --symbol-ordering-file.
347 llvm::DenseMap<SectionBase *, int> buildSectionOrder();
351 std::string toString(const elf::InputSectionBase *);