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"
29 class SyntheticSection;
30 class MergeSyntheticSection;
31 template <class ELFT> class ObjFile;
34 // This is the base class of all sections that lld handles. Some are sections in
35 // input files, some are sections in the produced output file and some exist
36 // just as a convenience for implementing special ways of combining some
40 enum Kind { Regular, EHFrame, Merge, Synthetic, Output };
42 Kind kind() const { return (Kind)SectionKind; }
46 // This pointer points to the "real" instance of this instance.
47 // Usually Repl == this. However, if ICF merges two sections,
48 // Repl pointer of one section points to another section. So,
49 // if you need to get a pointer to this instance, do not use
50 // this but instead this->Repl.
53 unsigned SectionKind : 3;
55 // The next two bit fields are only used by InputSectionBase, but we
56 // put them here so the struct packs better.
58 // The garbage collector sets sections' Live bits.
59 // If GC is disabled, all sections are considered live by default.
64 // Set for sections that should not be folded by ICF.
65 unsigned KeepUnique : 1;
67 // These corresponds to the fields in Elf_Shdr.
75 OutputSection *getOutputSection();
76 const OutputSection *getOutputSection() const {
77 return const_cast<SectionBase *>(this)->getOutputSection();
80 // Translate an offset in the input section to an offset in the output
82 uint64_t getOffset(uint64_t Offset) const;
84 uint64_t getVA(uint64_t Offset = 0) const;
87 SectionBase(Kind SectionKind, StringRef Name, uint64_t Flags,
88 uint64_t Entsize, uint64_t Alignment, uint32_t Type,
89 uint32_t Info, uint32_t Link)
90 : Name(Name), Repl(this), SectionKind(SectionKind), Live(false),
91 Bss(false), KeepUnique(false), Alignment(Alignment), Flags(Flags),
92 Entsize(Entsize), Type(Type), Link(Link), Info(Info) {}
95 // This corresponds to a section of an input file.
96 class InputSectionBase : public SectionBase {
99 InputSectionBase(ObjFile<ELFT> &File, const typename ELFT::Shdr &Header,
100 StringRef Name, Kind SectionKind);
102 InputSectionBase(InputFile *File, uint64_t Flags, uint32_t Type,
103 uint64_t Entsize, uint32_t Link, uint32_t Info,
104 uint32_t Alignment, ArrayRef<uint8_t> Data, StringRef Name,
107 static bool classof(const SectionBase *S) { return S->kind() != Output; }
109 // The file which contains this section. Its dynamic type is always
110 // ObjFile<ELFT>, but in order to avoid ELFT, we use InputFile as
114 template <class ELFT> ObjFile<ELFT> *getFile() const {
115 return cast_or_null<ObjFile<ELFT>>(File);
118 ArrayRef<uint8_t> Data;
119 uint64_t getOffsetInFile() const;
121 // True if this section has already been placed to a linker script
122 // output section. This is needed because, in a linker script, you
123 // can refer to the same section more than once. For example, in
124 // the following linker script,
126 // .foo : { *(.text) }
127 // .bar : { *(.text) }
129 // .foo takes all .text sections, and .bar becomes empty. To achieve
130 // this, we need to memorize whether a section has been placed or
131 // not for each input section.
132 bool Assigned = false;
134 // Input sections are part of an output section. Special sections
135 // like .eh_frame and merge sections are first combined into a
136 // synthetic section that is then added to an output section. In all
137 // cases this points one level up.
138 SectionBase *Parent = nullptr;
140 // Relocations that refer to this section.
141 const void *FirstRelocation = nullptr;
142 unsigned NumRelocations : 31;
143 unsigned AreRelocsRela : 1;
145 template <class ELFT> ArrayRef<typename ELFT::Rel> rels() const {
146 assert(!AreRelocsRela);
147 return llvm::makeArrayRef(
148 static_cast<const typename ELFT::Rel *>(FirstRelocation),
152 template <class ELFT> ArrayRef<typename ELFT::Rela> relas() const {
153 assert(AreRelocsRela);
154 return llvm::makeArrayRef(
155 static_cast<const typename ELFT::Rela *>(FirstRelocation),
159 // InputSections that are dependent on us (reverse dependency for GC)
160 llvm::TinyPtrVector<InputSection *> DependentSections;
162 // Returns the size of this section (even if this is a common or BSS.)
163 size_t getSize() const;
165 InputSection *getLinkOrderDep() const;
167 // Get the function symbol that encloses this offset from within the
169 template <class ELFT>
170 Defined *getEnclosingFunction(uint64_t Offset);
172 // Compilers emit zlib-compressed debug sections if the -gz option
173 // is given. This function checks if this section is compressed, and
174 // if so, decompress in memory.
175 void maybeDecompress();
177 // Returns a source location string. Used to construct an error message.
178 template <class ELFT> std::string getLocation(uint64_t Offset);
179 std::string getSrcMsg(const Symbol &Sym, uint64_t Offset);
180 std::string getObjMsg(uint64_t Offset);
182 // Each section knows how to relocate itself. These functions apply
183 // relocations, assuming that Buf points to this section's copy in
184 // the mmap'ed output buffer.
185 template <class ELFT> void relocate(uint8_t *Buf, uint8_t *BufEnd);
186 void relocateAlloc(uint8_t *Buf, uint8_t *BufEnd);
188 // The native ELF reloc data type is not very convenient to handle.
189 // So we convert ELF reloc records to our own records in Relocations.cpp.
190 // This vector contains such "cooked" relocations.
191 std::vector<Relocation> Relocations;
193 // A function compiled with -fsplit-stack calling a function
194 // compiled without -fsplit-stack needs its prologue adjusted. Find
195 // such functions and adjust their prologues. This is very similar
196 // to relocation. See https://gcc.gnu.org/wiki/SplitStacks for more
198 template <typename ELFT>
199 void adjustSplitStackFunctionPrologues(uint8_t *Buf, uint8_t *End);
202 template <typename T> llvm::ArrayRef<T> getDataAs() const {
203 size_t S = Data.size();
204 assert(S % sizeof(T) == 0);
205 return llvm::makeArrayRef<T>((const T *)Data.data(), S / sizeof(T));
209 // A pointer that owns decompressed data if a section is compressed by zlib.
210 // Since the feature is not used often, this is usually a nullptr.
211 std::unique_ptr<char[]> DecompressBuf;
214 // SectionPiece represents a piece of splittable section contents.
215 // We allocate a lot of these and binary search on them. This means that they
216 // have to be as compact as possible, which is why we don't store the size (can
217 // be found by looking at the next one).
218 struct SectionPiece {
219 SectionPiece(size_t Off, uint32_t Hash, bool Live)
220 : InputOff(Off), Hash(Hash), OutputOff(0),
221 Live(Live || !Config->GcSections) {}
225 int64_t OutputOff : 63;
229 static_assert(sizeof(SectionPiece) == 16, "SectionPiece is too big");
231 // This corresponds to a SHF_MERGE section of an input file.
232 class MergeInputSection : public InputSectionBase {
234 template <class ELFT>
235 MergeInputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
237 MergeInputSection(uint64_t Flags, uint32_t Type, uint64_t Entsize,
238 ArrayRef<uint8_t> Data, StringRef Name);
240 static bool classof(const SectionBase *S) { return S->kind() == Merge; }
241 void splitIntoPieces();
243 // Translate an offset in the input section to an offset in the parent
244 // MergeSyntheticSection.
245 uint64_t getParentOffset(uint64_t Offset) const;
247 // Splittable sections are handled as a sequence of data
248 // rather than a single large blob of data.
249 std::vector<SectionPiece> Pieces;
250 llvm::DenseMap<uint32_t, uint32_t> OffsetMap;
252 // Returns I'th piece's data. This function is very hot when
253 // string merging is enabled, so we want to inline.
254 LLVM_ATTRIBUTE_ALWAYS_INLINE
255 llvm::CachedHashStringRef getData(size_t I) const {
256 size_t Begin = Pieces[I].InputOff;
258 (Pieces.size() - 1 == I) ? Data.size() : Pieces[I + 1].InputOff;
259 return {toStringRef(Data.slice(Begin, End - Begin)), Pieces[I].Hash};
262 // Returns the SectionPiece at a given input section offset.
263 SectionPiece *getSectionPiece(uint64_t Offset);
264 const SectionPiece *getSectionPiece(uint64_t Offset) const {
265 return const_cast<MergeInputSection *>(this)->getSectionPiece(Offset);
268 SyntheticSection *getParent() const;
271 void splitStrings(ArrayRef<uint8_t> A, size_t Size);
272 void splitNonStrings(ArrayRef<uint8_t> A, size_t Size);
275 struct EhSectionPiece {
276 EhSectionPiece(size_t Off, InputSectionBase *Sec, uint32_t Size,
277 unsigned FirstRelocation)
278 : InputOff(Off), Sec(Sec), Size(Size), FirstRelocation(FirstRelocation) {}
280 ArrayRef<uint8_t> data() { return {Sec->Data.data() + this->InputOff, Size}; }
283 ssize_t OutputOff = -1;
284 InputSectionBase *Sec;
286 unsigned FirstRelocation;
289 // This corresponds to a .eh_frame section of an input file.
290 class EhInputSection : public InputSectionBase {
292 template <class ELFT>
293 EhInputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
295 static bool classof(const SectionBase *S) { return S->kind() == EHFrame; }
296 template <class ELFT> void split();
297 template <class ELFT, class RelTy> void split(ArrayRef<RelTy> Rels);
299 // Splittable sections are handled as a sequence of data
300 // rather than a single large blob of data.
301 std::vector<EhSectionPiece> Pieces;
303 SyntheticSection *getParent() const;
306 // This is a section that is added directly to an output section
307 // instead of needing special combination via a synthetic section. This
308 // includes all input sections with the exceptions of SHF_MERGE and
309 // .eh_frame. It also includes the synthetic sections themselves.
310 class InputSection : public InputSectionBase {
312 InputSection(InputFile *F, uint64_t Flags, uint32_t Type, uint32_t Alignment,
313 ArrayRef<uint8_t> Data, StringRef Name, Kind K = Regular);
314 template <class ELFT>
315 InputSection(ObjFile<ELFT> &F, const typename ELFT::Shdr &Header,
318 // Write this section to a mmap'ed file, assuming Buf is pointing to
319 // beginning of the output section.
320 template <class ELFT> void writeTo(uint8_t *Buf);
322 uint64_t getOffset(uint64_t Offset) const { return OutSecOff + Offset; }
324 OutputSection *getParent() const;
326 // This variable has two usages. Initially, it represents an index in the
327 // OutputSection's InputSection list, and is used when ordering SHF_LINK_ORDER
328 // sections. After assignAddresses is called, it represents the offset from
329 // the beginning of the output section this section was assigned to.
330 uint64_t OutSecOff = 0;
332 static bool classof(const SectionBase *S);
334 InputSectionBase *getRelocatedSection() const;
336 template <class ELFT, class RelTy>
337 void relocateNonAlloc(uint8_t *Buf, llvm::ArrayRef<RelTy> Rels);
340 uint32_t Class[2] = {0, 0};
342 // Called by ICF to merge two input sections.
343 void replace(InputSection *Other);
345 static InputSection Discarded;
348 template <class ELFT, class RelTy>
349 void copyRelocations(uint8_t *Buf, llvm::ArrayRef<RelTy> Rels);
351 template <class ELFT> void copyShtGroup(uint8_t *Buf);
354 // The list of all input sections.
355 extern std::vector<InputSectionBase *> InputSections;
358 std::string toString(const elf::InputSectionBase *);