| 1 | //===- SyntheticSections.h -------------------------------------*- C++ -*-===// |
| 2 | // |
| 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 |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #ifndef LLD_MACHO_SYNTHETIC_SECTIONS_H |
| 10 | #define LLD_MACHO_SYNTHETIC_SECTIONS_H |
| 11 | |
| 12 | #include "Config.h" |
| 13 | #include "ExportTrie.h" |
| 14 | #include "InputSection.h" |
| 15 | #include "OutputSection.h" |
| 16 | #include "OutputSegment.h" |
| 17 | #include "Target.h" |
| 18 | #include "Writer.h" |
| 19 | |
| 20 | #include "llvm/ADT/DenseMap.h" |
| 21 | #include "llvm/ADT/MapVector.h" |
| 22 | #include "llvm/ADT/SetVector.h" |
| 23 | #include "llvm/BinaryFormat/MachO.h" |
| 24 | #include "llvm/Support/MathExtras.h" |
| 25 | #include "llvm/Support/raw_ostream.h" |
| 26 | |
| 27 | namespace llvm { |
| 28 | class DWARFUnit; |
| 29 | } // namespace llvm |
| 30 | |
| 31 | namespace lld::macho { |
| 32 | |
| 33 | class Defined; |
| 34 | class DylibSymbol; |
| 35 | class LoadCommand; |
| 36 | class ObjFile; |
| 37 | class UnwindInfoSection; |
| 38 | |
| 39 | class SyntheticSection : public OutputSection { |
| 40 | public: |
| 41 | SyntheticSection(const char *segname, const char *name); |
| 42 | virtual ~SyntheticSection() = default; |
| 43 | |
| 44 | static bool classof(const OutputSection *sec) { |
| 45 | return sec->kind() == SyntheticKind; |
| 46 | } |
| 47 | |
| 48 | StringRef segname; |
| 49 | // This fake InputSection makes it easier for us to write code that applies |
| 50 | // generically to both user inputs and synthetics. |
| 51 | InputSection *isec; |
| 52 | }; |
| 53 | |
| 54 | // All sections in __LINKEDIT should inherit from this. |
| 55 | class LinkEditSection : public SyntheticSection { |
| 56 | public: |
| 57 | LinkEditSection(const char *segname, const char *name) |
| 58 | : SyntheticSection(segname, name) { |
| 59 | align = target->wordSize; |
| 60 | } |
| 61 | |
| 62 | // Implementations of this method can assume that the regular (non-__LINKEDIT) |
| 63 | // sections already have their addresses assigned. |
| 64 | virtual void finalizeContents() {} |
| 65 | |
| 66 | // Sections in __LINKEDIT are special: their offsets are recorded in the |
| 67 | // load commands like LC_DYLD_INFO_ONLY and LC_SYMTAB, instead of in section |
| 68 | // headers. |
| 69 | bool isHidden() const final { return true; } |
| 70 | |
| 71 | virtual uint64_t getRawSize() const = 0; |
| 72 | |
| 73 | // codesign (or more specifically libstuff) checks that each section in |
| 74 | // __LINKEDIT ends where the next one starts -- no gaps are permitted. We |
| 75 | // therefore align every section's start and end points to WordSize. |
| 76 | // |
| 77 | // NOTE: This assumes that the extra bytes required for alignment can be |
| 78 | // zero-valued bytes. |
| 79 | uint64_t getSize() const final { return llvm::alignTo(Value: getRawSize(), Align: align); } |
| 80 | }; |
| 81 | |
| 82 | // The header of the Mach-O file, which must have a file offset of zero. |
| 83 | class final : public SyntheticSection { |
| 84 | public: |
| 85 | (); |
| 86 | bool () const override { return true; } |
| 87 | uint64_t () const override; |
| 88 | void (uint8_t *buf) const override; |
| 89 | |
| 90 | void addLoadCommand(LoadCommand *); |
| 91 | |
| 92 | protected: |
| 93 | std::vector<LoadCommand *> loadCommands; |
| 94 | uint32_t = 0; |
| 95 | }; |
| 96 | |
| 97 | // A hidden section that exists solely for the purpose of creating the |
| 98 | // __PAGEZERO segment, which is used to catch null pointer dereferences. |
| 99 | class PageZeroSection final : public SyntheticSection { |
| 100 | public: |
| 101 | PageZeroSection(); |
| 102 | bool isHidden() const override { return true; } |
| 103 | bool isNeeded() const override { return target->pageZeroSize != 0; } |
| 104 | uint64_t getSize() const override { return target->pageZeroSize; } |
| 105 | uint64_t getFileSize() const override { return 0; } |
| 106 | void writeTo(uint8_t *buf) const override {} |
| 107 | }; |
| 108 | |
| 109 | // This is the base class for the GOT and TLVPointer sections, which are nearly |
| 110 | // functionally identical -- they will both be populated by dyld with addresses |
| 111 | // to non-lazily-loaded dylib symbols. The main difference is that the |
| 112 | // TLVPointerSection stores references to thread-local variables. |
| 113 | class NonLazyPointerSectionBase : public SyntheticSection { |
| 114 | public: |
| 115 | NonLazyPointerSectionBase(const char *segname, const char *name); |
| 116 | const llvm::SetVector<const Symbol *> &getEntries() const { return entries; } |
| 117 | bool isNeeded() const override { return !entries.empty(); } |
| 118 | uint64_t getSize() const override { |
| 119 | return entries.size() * target->wordSize; |
| 120 | } |
| 121 | void writeTo(uint8_t *buf) const override; |
| 122 | void addEntry(Symbol *sym); |
| 123 | uint64_t getVA(uint32_t gotIndex) const { |
| 124 | return addr + gotIndex * target->wordSize; |
| 125 | } |
| 126 | |
| 127 | private: |
| 128 | llvm::SetVector<const Symbol *> entries; |
| 129 | }; |
| 130 | |
| 131 | class GotSection final : public NonLazyPointerSectionBase { |
| 132 | public: |
| 133 | GotSection(); |
| 134 | }; |
| 135 | |
| 136 | class TlvPointerSection final : public NonLazyPointerSectionBase { |
| 137 | public: |
| 138 | TlvPointerSection(); |
| 139 | }; |
| 140 | |
| 141 | struct Location { |
| 142 | const InputSection *isec; |
| 143 | uint64_t offset; |
| 144 | |
| 145 | Location(const InputSection *isec, uint64_t offset) |
| 146 | : isec(isec), offset(offset) {} |
| 147 | uint64_t getVA() const { return isec->getVA(off: offset); } |
| 148 | }; |
| 149 | |
| 150 | // Stores rebase opcodes, which tell dyld where absolute addresses have been |
| 151 | // encoded in the binary. If the binary is not loaded at its preferred address, |
| 152 | // dyld has to rebase these addresses by adding an offset to them. |
| 153 | class RebaseSection final : public LinkEditSection { |
| 154 | public: |
| 155 | RebaseSection(); |
| 156 | void finalizeContents() override; |
| 157 | uint64_t getRawSize() const override { return contents.size(); } |
| 158 | bool isNeeded() const override { return !locations.empty(); } |
| 159 | void writeTo(uint8_t *buf) const override; |
| 160 | |
| 161 | void addEntry(const InputSection *isec, uint64_t offset) { |
| 162 | if (config->isPic) |
| 163 | locations.emplace_back(args&: isec, args&: offset); |
| 164 | } |
| 165 | |
| 166 | private: |
| 167 | std::vector<Location> locations; |
| 168 | SmallVector<char, 128> contents; |
| 169 | }; |
| 170 | |
| 171 | struct BindingEntry { |
| 172 | int64_t addend; |
| 173 | Location target; |
| 174 | BindingEntry(int64_t addend, Location target) |
| 175 | : addend(addend), target(target) {} |
| 176 | }; |
| 177 | |
| 178 | template <class Sym> |
| 179 | using BindingsMap = llvm::DenseMap<Sym, std::vector<BindingEntry>>; |
| 180 | |
| 181 | // Stores bind opcodes for telling dyld which symbols to load non-lazily. |
| 182 | class BindingSection final : public LinkEditSection { |
| 183 | public: |
| 184 | BindingSection(); |
| 185 | void finalizeContents() override; |
| 186 | uint64_t getRawSize() const override { return contents.size(); } |
| 187 | bool isNeeded() const override { return !bindingsMap.empty(); } |
| 188 | void writeTo(uint8_t *buf) const override; |
| 189 | |
| 190 | void addEntry(const Symbol *dysym, const InputSection *isec, uint64_t offset, |
| 191 | int64_t addend = 0) { |
| 192 | bindingsMap[dysym].emplace_back(args&: addend, args: Location(isec, offset)); |
| 193 | } |
| 194 | |
| 195 | private: |
| 196 | BindingsMap<const Symbol *> bindingsMap; |
| 197 | SmallVector<char, 128> contents; |
| 198 | }; |
| 199 | |
| 200 | // Stores bind opcodes for telling dyld which weak symbols need coalescing. |
| 201 | // There are two types of entries in this section: |
| 202 | // |
| 203 | // 1) Non-weak definitions: This is a symbol definition that weak symbols in |
| 204 | // other dylibs should coalesce to. |
| 205 | // |
| 206 | // 2) Weak bindings: These tell dyld that a given symbol reference should |
| 207 | // coalesce to a non-weak definition if one is found. Note that unlike the |
| 208 | // entries in the BindingSection, the bindings here only refer to these |
| 209 | // symbols by name, but do not specify which dylib to load them from. |
| 210 | class WeakBindingSection final : public LinkEditSection { |
| 211 | public: |
| 212 | WeakBindingSection(); |
| 213 | void finalizeContents() override; |
| 214 | uint64_t getRawSize() const override { return contents.size(); } |
| 215 | bool isNeeded() const override { |
| 216 | return !bindingsMap.empty() || !definitions.empty(); |
| 217 | } |
| 218 | |
| 219 | void writeTo(uint8_t *buf) const override; |
| 220 | |
| 221 | void addEntry(const Symbol *symbol, const InputSection *isec, uint64_t offset, |
| 222 | int64_t addend = 0) { |
| 223 | bindingsMap[symbol].emplace_back(args&: addend, args: Location(isec, offset)); |
| 224 | } |
| 225 | |
| 226 | bool hasEntry() const { return !bindingsMap.empty(); } |
| 227 | |
| 228 | void addNonWeakDefinition(const Defined *defined) { |
| 229 | definitions.emplace_back(args&: defined); |
| 230 | } |
| 231 | |
| 232 | bool hasNonWeakDefinition() const { return !definitions.empty(); } |
| 233 | |
| 234 | private: |
| 235 | BindingsMap<const Symbol *> bindingsMap; |
| 236 | std::vector<const Defined *> definitions; |
| 237 | SmallVector<char, 128> contents; |
| 238 | }; |
| 239 | |
| 240 | // The following sections implement lazy symbol binding -- very similar to the |
| 241 | // PLT mechanism in ELF. |
| 242 | // |
| 243 | // ELF's .plt section is broken up into two sections in Mach-O: StubsSection |
| 244 | // and StubHelperSection. Calls to functions in dylibs will end up calling into |
| 245 | // StubsSection, which contains indirect jumps to addresses stored in the |
| 246 | // LazyPointerSection (the counterpart to ELF's .plt.got). |
| 247 | // |
| 248 | // We will first describe how non-weak symbols are handled. |
| 249 | // |
| 250 | // At program start, the LazyPointerSection contains addresses that point into |
| 251 | // one of the entry points in the middle of the StubHelperSection. The code in |
| 252 | // StubHelperSection will push on the stack an offset into the |
| 253 | // LazyBindingSection. The push is followed by a jump to the beginning of the |
| 254 | // StubHelperSection (similar to PLT0), which then calls into dyld_stub_binder. |
| 255 | // dyld_stub_binder is a non-lazily-bound symbol, so this call looks it up in |
| 256 | // the GOT. |
| 257 | // |
| 258 | // The stub binder will look up the bind opcodes in the LazyBindingSection at |
| 259 | // the given offset. The bind opcodes will tell the binder to update the |
| 260 | // address in the LazyPointerSection to point to the symbol, so that subsequent |
| 261 | // calls don't have to redo the symbol resolution. The binder will then jump to |
| 262 | // the resolved symbol. |
| 263 | // |
| 264 | // With weak symbols, the situation is slightly different. Since there is no |
| 265 | // "weak lazy" lookup, function calls to weak symbols are always non-lazily |
| 266 | // bound. We emit both regular non-lazy bindings as well as weak bindings, in |
| 267 | // order that the weak bindings may overwrite the non-lazy bindings if an |
| 268 | // appropriate symbol is found at runtime. However, the bound addresses will |
| 269 | // still be written (non-lazily) into the LazyPointerSection. |
| 270 | // |
| 271 | // Symbols are always bound eagerly when chained fixups are used. In that case, |
| 272 | // StubsSection contains indirect jumps to addresses stored in the GotSection. |
| 273 | // The GOT directly contains the fixup entries, which will be replaced by the |
| 274 | // address of the target symbols on load. LazyPointerSection and |
| 275 | // StubHelperSection are not used. |
| 276 | |
| 277 | class StubsSection final : public SyntheticSection { |
| 278 | public: |
| 279 | StubsSection(); |
| 280 | uint64_t getSize() const override; |
| 281 | bool isNeeded() const override { return !entries.empty(); } |
| 282 | void finalize() override; |
| 283 | void writeTo(uint8_t *buf) const override; |
| 284 | const llvm::SetVector<Symbol *> &getEntries() const { return entries; } |
| 285 | // Creates a stub for the symbol and the corresponding entry in the |
| 286 | // LazyPointerSection. |
| 287 | void addEntry(Symbol *); |
| 288 | uint64_t getVA(uint32_t stubsIndex) const { |
| 289 | assert(isFinal || target->usesThunks()); |
| 290 | // ConcatOutputSection::finalize() can seek the address of a |
| 291 | // stub before its address is assigned. Before __stubs is |
| 292 | // finalized, return a contrived out-of-range address. |
| 293 | return isFinal ? addr + stubsIndex * target->stubSize |
| 294 | : TargetInfo::outOfRangeVA; |
| 295 | } |
| 296 | |
| 297 | bool isFinal = false; // is address assigned? |
| 298 | |
| 299 | private: |
| 300 | llvm::SetVector<Symbol *> entries; |
| 301 | }; |
| 302 | |
| 303 | class StubHelperSection final : public SyntheticSection { |
| 304 | public: |
| 305 | StubHelperSection(); |
| 306 | uint64_t getSize() const override; |
| 307 | bool isNeeded() const override; |
| 308 | void writeTo(uint8_t *buf) const override; |
| 309 | |
| 310 | void setUp(); |
| 311 | |
| 312 | DylibSymbol *stubBinder = nullptr; |
| 313 | Defined *dyldPrivate = nullptr; |
| 314 | }; |
| 315 | |
| 316 | class ObjCSelRefsHelper { |
| 317 | public: |
| 318 | static void initialize(); |
| 319 | static void cleanup(); |
| 320 | |
| 321 | static ConcatInputSection *getSelRef(StringRef methname); |
| 322 | static ConcatInputSection *makeSelRef(StringRef methname); |
| 323 | |
| 324 | private: |
| 325 | static llvm::DenseMap<llvm::CachedHashStringRef, ConcatInputSection *> |
| 326 | methnameToSelref; |
| 327 | }; |
| 328 | |
| 329 | // Objective-C stubs are hoisted objc_msgSend calls per selector called in the |
| 330 | // program. Apple Clang produces undefined symbols to each stub, such as |
| 331 | // '_objc_msgSend$foo', which are then synthesized by the linker. The stubs |
| 332 | // load the particular selector 'foo' from __objc_selrefs, setting it to the |
| 333 | // first argument of the objc_msgSend call, and then jumps to objc_msgSend. The |
| 334 | // actual stub contents are mirrored from ld64. |
| 335 | class ObjCStubsSection final : public SyntheticSection { |
| 336 | public: |
| 337 | ObjCStubsSection(); |
| 338 | void addEntry(Symbol *sym); |
| 339 | uint64_t getSize() const override; |
| 340 | bool isNeeded() const override { return !symbols.empty(); } |
| 341 | void finalize() override { isec->isFinal = true; } |
| 342 | void writeTo(uint8_t *buf) const override; |
| 343 | void setUp(); |
| 344 | |
| 345 | static constexpr llvm::StringLiteral symbolPrefix = "_objc_msgSend$" ; |
| 346 | static bool isObjCStubSymbol(Symbol *sym); |
| 347 | static StringRef getMethname(Symbol *sym); |
| 348 | |
| 349 | private: |
| 350 | std::vector<Defined *> symbols; |
| 351 | Symbol *objcMsgSend = nullptr; |
| 352 | }; |
| 353 | |
| 354 | // Note that this section may also be targeted by non-lazy bindings. In |
| 355 | // particular, this happens when branch relocations target weak symbols. |
| 356 | class LazyPointerSection final : public SyntheticSection { |
| 357 | public: |
| 358 | LazyPointerSection(); |
| 359 | uint64_t getSize() const override; |
| 360 | bool isNeeded() const override; |
| 361 | void writeTo(uint8_t *buf) const override; |
| 362 | uint64_t getVA(uint32_t index) const { |
| 363 | return addr + (index << target->p2WordSize); |
| 364 | } |
| 365 | }; |
| 366 | |
| 367 | class LazyBindingSection final : public LinkEditSection { |
| 368 | public: |
| 369 | LazyBindingSection(); |
| 370 | void finalizeContents() override; |
| 371 | uint64_t getRawSize() const override { return contents.size(); } |
| 372 | bool isNeeded() const override { return !entries.empty(); } |
| 373 | void writeTo(uint8_t *buf) const override; |
| 374 | // Note that every entry here will by referenced by a corresponding entry in |
| 375 | // the StubHelperSection. |
| 376 | void addEntry(Symbol *dysym); |
| 377 | const llvm::SetVector<Symbol *> &getEntries() const { return entries; } |
| 378 | |
| 379 | private: |
| 380 | uint32_t encode(const Symbol &); |
| 381 | |
| 382 | llvm::SetVector<Symbol *> entries; |
| 383 | SmallVector<char, 128> contents; |
| 384 | llvm::raw_svector_ostream os{contents}; |
| 385 | }; |
| 386 | |
| 387 | // Stores a trie that describes the set of exported symbols. |
| 388 | class ExportSection final : public LinkEditSection { |
| 389 | public: |
| 390 | ExportSection(); |
| 391 | void finalizeContents() override; |
| 392 | uint64_t getRawSize() const override { return size; } |
| 393 | bool isNeeded() const override { return size; } |
| 394 | void writeTo(uint8_t *buf) const override; |
| 395 | |
| 396 | bool hasWeakSymbol = false; |
| 397 | |
| 398 | private: |
| 399 | TrieBuilder trieBuilder; |
| 400 | size_t size = 0; |
| 401 | }; |
| 402 | |
| 403 | // Stores 'data in code' entries that describe the locations of data regions |
| 404 | // inside code sections. This is used by llvm-objdump to distinguish jump tables |
| 405 | // and stop them from being disassembled as instructions. |
| 406 | class DataInCodeSection final : public LinkEditSection { |
| 407 | public: |
| 408 | DataInCodeSection(); |
| 409 | void finalizeContents() override; |
| 410 | uint64_t getRawSize() const override { |
| 411 | return sizeof(llvm::MachO::data_in_code_entry) * entries.size(); |
| 412 | } |
| 413 | void writeTo(uint8_t *buf) const override; |
| 414 | |
| 415 | private: |
| 416 | std::vector<llvm::MachO::data_in_code_entry> entries; |
| 417 | }; |
| 418 | |
| 419 | // Stores ULEB128 delta encoded addresses of functions. |
| 420 | class FunctionStartsSection final : public LinkEditSection { |
| 421 | public: |
| 422 | FunctionStartsSection(); |
| 423 | void finalizeContents() override; |
| 424 | uint64_t getRawSize() const override { return contents.size(); } |
| 425 | void writeTo(uint8_t *buf) const override; |
| 426 | |
| 427 | private: |
| 428 | SmallVector<char, 128> contents; |
| 429 | }; |
| 430 | |
| 431 | // Stores the strings referenced by the symbol table. |
| 432 | class StringTableSection final : public LinkEditSection { |
| 433 | public: |
| 434 | StringTableSection(); |
| 435 | // Returns the start offset of the added string. |
| 436 | uint32_t addString(StringRef); |
| 437 | uint64_t getRawSize() const override { return size; } |
| 438 | void writeTo(uint8_t *buf) const override; |
| 439 | |
| 440 | static constexpr size_t emptyStringIndex = 1; |
| 441 | |
| 442 | private: |
| 443 | // ld64 emits string tables which start with a space and a zero byte. We |
| 444 | // match its behavior here since some tools depend on it. |
| 445 | // Consequently, the empty string will be at index 1, not zero. |
| 446 | std::vector<StringRef> strings{" " }; |
| 447 | llvm::DenseMap<llvm::CachedHashStringRef, uint32_t> stringMap; |
| 448 | size_t size = 2; |
| 449 | }; |
| 450 | |
| 451 | struct SymtabEntry { |
| 452 | Symbol *sym; |
| 453 | size_t strx; |
| 454 | }; |
| 455 | |
| 456 | struct StabsEntry { |
| 457 | uint8_t type = 0; |
| 458 | uint32_t strx = StringTableSection::emptyStringIndex; |
| 459 | uint8_t sect = 0; |
| 460 | uint16_t desc = 0; |
| 461 | uint64_t value = 0; |
| 462 | |
| 463 | StabsEntry() = default; |
| 464 | explicit StabsEntry(uint8_t type) : type(type) {} |
| 465 | }; |
| 466 | |
| 467 | // Symbols of the same type must be laid out contiguously: we choose to emit |
| 468 | // all local symbols first, then external symbols, and finally undefined |
| 469 | // symbols. For each symbol type, the LC_DYSYMTAB load command will record the |
| 470 | // range (start index and total number) of those symbols in the symbol table. |
| 471 | class SymtabSection : public LinkEditSection { |
| 472 | public: |
| 473 | void finalizeContents() override; |
| 474 | uint32_t getNumSymbols() const; |
| 475 | uint32_t getNumLocalSymbols() const { |
| 476 | return stabs.size() + localSymbols.size(); |
| 477 | } |
| 478 | uint32_t getNumExternalSymbols() const { return externalSymbols.size(); } |
| 479 | uint32_t getNumUndefinedSymbols() const { return undefinedSymbols.size(); } |
| 480 | |
| 481 | private: |
| 482 | void emitBeginSourceStab(StringRef); |
| 483 | void emitEndSourceStab(); |
| 484 | void emitObjectFileStab(ObjFile *); |
| 485 | void emitEndFunStab(Defined *); |
| 486 | Defined *getFuncBodySym(Defined *); |
| 487 | void emitStabs(); |
| 488 | |
| 489 | protected: |
| 490 | SymtabSection(StringTableSection &); |
| 491 | |
| 492 | StringTableSection &stringTableSection; |
| 493 | // STABS symbols are always local symbols, but we represent them with special |
| 494 | // entries because they may use fields like n_sect and n_desc differently. |
| 495 | std::vector<StabsEntry> stabs; |
| 496 | std::vector<SymtabEntry> localSymbols; |
| 497 | std::vector<SymtabEntry> externalSymbols; |
| 498 | std::vector<SymtabEntry> undefinedSymbols; |
| 499 | }; |
| 500 | |
| 501 | template <class LP> SymtabSection *makeSymtabSection(StringTableSection &); |
| 502 | |
| 503 | // The indirect symbol table is a list of 32-bit integers that serve as indices |
| 504 | // into the (actual) symbol table. The indirect symbol table is a |
| 505 | // concatenation of several sub-arrays of indices, each sub-array belonging to |
| 506 | // a separate section. The starting offset of each sub-array is stored in the |
| 507 | // reserved1 header field of the respective section. |
| 508 | // |
| 509 | // These sub-arrays provide symbol information for sections that store |
| 510 | // contiguous sequences of symbol references. These references can be pointers |
| 511 | // (e.g. those in the GOT and TLVP sections) or assembly sequences (e.g. |
| 512 | // function stubs). |
| 513 | class IndirectSymtabSection final : public LinkEditSection { |
| 514 | public: |
| 515 | IndirectSymtabSection(); |
| 516 | void finalizeContents() override; |
| 517 | uint32_t getNumSymbols() const; |
| 518 | uint64_t getRawSize() const override { |
| 519 | return getNumSymbols() * sizeof(uint32_t); |
| 520 | } |
| 521 | bool isNeeded() const override; |
| 522 | void writeTo(uint8_t *buf) const override; |
| 523 | }; |
| 524 | |
| 525 | // The code signature comes at the very end of the linked output file. |
| 526 | class CodeSignatureSection final : public LinkEditSection { |
| 527 | public: |
| 528 | // NOTE: These values are duplicated in llvm-objcopy's MachO/Object.h file |
| 529 | // and any changes here, should be repeated there. |
| 530 | static constexpr uint8_t blockSizeShift = 12; |
| 531 | static constexpr size_t blockSize = (1 << blockSizeShift); // 4 KiB |
| 532 | static constexpr size_t hashSize = 256 / 8; |
| 533 | static constexpr size_t = llvm::alignTo<8>( |
| 534 | Value: sizeof(llvm::MachO::CS_SuperBlob) + sizeof(llvm::MachO::CS_BlobIndex)); |
| 535 | static constexpr uint32_t = |
| 536 | blobHeadersSize + sizeof(llvm::MachO::CS_CodeDirectory); |
| 537 | |
| 538 | uint32_t fileNamePad = 0; |
| 539 | uint32_t = 0; |
| 540 | StringRef fileName; |
| 541 | |
| 542 | CodeSignatureSection(); |
| 543 | uint64_t getRawSize() const override; |
| 544 | bool isNeeded() const override { return true; } |
| 545 | void writeTo(uint8_t *buf) const override; |
| 546 | uint32_t getBlockCount() const; |
| 547 | void writeHashes(uint8_t *buf) const; |
| 548 | }; |
| 549 | |
| 550 | class CStringSection : public SyntheticSection { |
| 551 | public: |
| 552 | CStringSection(const char *name); |
| 553 | void addInput(CStringInputSection *); |
| 554 | uint64_t getSize() const override { return size; } |
| 555 | virtual void finalizeContents(); |
| 556 | bool isNeeded() const override { return !inputs.empty(); } |
| 557 | void writeTo(uint8_t *buf) const override; |
| 558 | |
| 559 | std::vector<CStringInputSection *> inputs; |
| 560 | |
| 561 | private: |
| 562 | uint64_t size; |
| 563 | }; |
| 564 | |
| 565 | class DeduplicatedCStringSection final : public CStringSection { |
| 566 | public: |
| 567 | DeduplicatedCStringSection(const char *name) : CStringSection(name){}; |
| 568 | uint64_t getSize() const override { return size; } |
| 569 | void finalizeContents() override; |
| 570 | void writeTo(uint8_t *buf) const override; |
| 571 | uint64_t getStringOffset(StringRef str) const; |
| 572 | |
| 573 | private: |
| 574 | llvm::DenseMap<llvm::CachedHashStringRef, uint64_t> stringOffsetMap; |
| 575 | size_t size = 0; |
| 576 | }; |
| 577 | |
| 578 | /* |
| 579 | * This section contains deduplicated literal values. The 16-byte values are |
| 580 | * laid out first, followed by the 8- and then the 4-byte ones. |
| 581 | */ |
| 582 | class WordLiteralSection final : public SyntheticSection { |
| 583 | public: |
| 584 | using UInt128 = std::pair<uint64_t, uint64_t>; |
| 585 | // I don't think the standard guarantees the size of a pair, so let's make |
| 586 | // sure it's exact -- that way we can construct it via `mmap`. |
| 587 | static_assert(sizeof(UInt128) == 16); |
| 588 | |
| 589 | WordLiteralSection(); |
| 590 | void addInput(WordLiteralInputSection *); |
| 591 | void finalizeContents(); |
| 592 | void writeTo(uint8_t *buf) const override; |
| 593 | |
| 594 | uint64_t getSize() const override { |
| 595 | return literal16Map.size() * 16 + literal8Map.size() * 8 + |
| 596 | literal4Map.size() * 4; |
| 597 | } |
| 598 | |
| 599 | bool isNeeded() const override { |
| 600 | return !literal16Map.empty() || !literal4Map.empty() || |
| 601 | !literal8Map.empty(); |
| 602 | } |
| 603 | |
| 604 | uint64_t getLiteral16Offset(uintptr_t buf) const { |
| 605 | return literal16Map.at(Val: *reinterpret_cast<const UInt128 *>(buf)) * 16; |
| 606 | } |
| 607 | |
| 608 | uint64_t getLiteral8Offset(uintptr_t buf) const { |
| 609 | return literal16Map.size() * 16 + |
| 610 | literal8Map.at(Val: *reinterpret_cast<const uint64_t *>(buf)) * 8; |
| 611 | } |
| 612 | |
| 613 | uint64_t getLiteral4Offset(uintptr_t buf) const { |
| 614 | return literal16Map.size() * 16 + literal8Map.size() * 8 + |
| 615 | literal4Map.at(Val: *reinterpret_cast<const uint32_t *>(buf)) * 4; |
| 616 | } |
| 617 | |
| 618 | private: |
| 619 | std::vector<WordLiteralInputSection *> inputs; |
| 620 | |
| 621 | // Literal values can be any bit pattern. |
| 622 | llvm::DenseMap<UInt128, uint64_t> literal16Map; |
| 623 | llvm::DenseMap<uint64_t, uint64_t> literal8Map; |
| 624 | llvm::DenseMap<uint32_t, uint64_t> literal4Map; |
| 625 | }; |
| 626 | |
| 627 | class ObjCImageInfoSection final : public SyntheticSection { |
| 628 | public: |
| 629 | ObjCImageInfoSection(); |
| 630 | bool isNeeded() const override { return !files.empty(); } |
| 631 | uint64_t getSize() const override { return 8; } |
| 632 | void addFile(const InputFile *file) { |
| 633 | assert(!file->objCImageInfo.empty()); |
| 634 | files.push_back(x: file); |
| 635 | } |
| 636 | void finalizeContents(); |
| 637 | void writeTo(uint8_t *buf) const override; |
| 638 | |
| 639 | private: |
| 640 | struct ImageInfo { |
| 641 | uint8_t swiftVersion = 0; |
| 642 | bool hasCategoryClassProperties = false; |
| 643 | } info; |
| 644 | static ImageInfo parseImageInfo(const InputFile *); |
| 645 | std::vector<const InputFile *> files; // files with image info |
| 646 | }; |
| 647 | |
| 648 | // This section stores 32-bit __TEXT segment offsets of initializer functions. |
| 649 | // |
| 650 | // The compiler stores pointers to initializers in __mod_init_func. These need |
| 651 | // to be fixed up at load time, which takes time and dirties memory. By |
| 652 | // synthesizing InitOffsetsSection from them, this data can live in the |
| 653 | // read-only __TEXT segment instead. This section is used by default when |
| 654 | // chained fixups are enabled. |
| 655 | // |
| 656 | // There is no similar counterpart to __mod_term_func, as that section is |
| 657 | // deprecated, and static destructors are instead handled by registering them |
| 658 | // via __cxa_atexit from an autogenerated initializer function (see D121736). |
| 659 | class InitOffsetsSection final : public SyntheticSection { |
| 660 | public: |
| 661 | InitOffsetsSection(); |
| 662 | bool isNeeded() const override { return !sections.empty(); } |
| 663 | uint64_t getSize() const override; |
| 664 | void writeTo(uint8_t *buf) const override; |
| 665 | void setUp(); |
| 666 | |
| 667 | void addInput(ConcatInputSection *isec) { sections.push_back(x: isec); } |
| 668 | const std::vector<ConcatInputSection *> &inputs() const { return sections; } |
| 669 | |
| 670 | private: |
| 671 | std::vector<ConcatInputSection *> sections; |
| 672 | }; |
| 673 | |
| 674 | // This SyntheticSection is for the __objc_methlist section, which contains |
| 675 | // relative method lists if the -objc_relative_method_lists option is enabled. |
| 676 | class ObjCMethListSection final : public SyntheticSection { |
| 677 | public: |
| 678 | ObjCMethListSection(); |
| 679 | |
| 680 | static bool isMethodList(const ConcatInputSection *isec); |
| 681 | void addInput(ConcatInputSection *isec) { inputs.push_back(x: isec); } |
| 682 | std::vector<ConcatInputSection *> getInputs() { return inputs; } |
| 683 | |
| 684 | void setUp(); |
| 685 | void finalize() override; |
| 686 | bool isNeeded() const override { return !inputs.empty(); } |
| 687 | uint64_t getSize() const override { return sectionSize; } |
| 688 | void writeTo(uint8_t *bufStart) const override; |
| 689 | |
| 690 | private: |
| 691 | void (const uint8_t *buf, uint32_t &structSizeAndFlags, |
| 692 | uint32_t &structCount) const; |
| 693 | void (uint8_t *buf, uint32_t structSizeAndFlags, |
| 694 | uint32_t structCount) const; |
| 695 | uint32_t computeRelativeMethodListSize(uint32_t absoluteMethodListSize) const; |
| 696 | void writeRelativeOffsetForIsec(const ConcatInputSection *isec, uint8_t *buf, |
| 697 | uint32_t &inSecOff, uint32_t &outSecOff, |
| 698 | bool useSelRef) const; |
| 699 | uint32_t writeRelativeMethodList(const ConcatInputSection *isec, |
| 700 | uint8_t *buf) const; |
| 701 | |
| 702 | static constexpr uint32_t = |
| 703 | /*structSizeAndFlags*/ sizeof(uint32_t) + |
| 704 | /*structCount*/ sizeof(uint32_t); |
| 705 | // Relative method lists are supported only for 3-pointer method lists |
| 706 | static constexpr uint32_t pointersPerStruct = 3; |
| 707 | // The runtime identifies relative method lists via this magic value |
| 708 | static constexpr uint32_t = 0x80000000; |
| 709 | // In the method list header, the first 2 bytes are the size of struct |
| 710 | static constexpr uint32_t structSizeMask = 0x0000FFFF; |
| 711 | // In the method list header, the last 2 bytes are the flags for the struct |
| 712 | static constexpr uint32_t structFlagsMask = 0xFFFF0000; |
| 713 | // Relative method lists have 4 byte alignment as all data in the InputSection |
| 714 | // is 4 byte |
| 715 | static constexpr uint32_t relativeOffsetSize = sizeof(uint32_t); |
| 716 | |
| 717 | // The output size of the __objc_methlist section, computed during finalize() |
| 718 | uint32_t sectionSize = 0; |
| 719 | std::vector<ConcatInputSection *> inputs; |
| 720 | }; |
| 721 | |
| 722 | // Chained fixups are a replacement for classic dyld opcodes. In this format, |
| 723 | // most of the metadata necessary for binding symbols and rebasing addresses is |
| 724 | // stored directly in the memory location that will have the fixup applied. |
| 725 | // |
| 726 | // The fixups form singly linked lists; each one covering a single page in |
| 727 | // memory. The __LINKEDIT,__chainfixups section stores the page offset of the |
| 728 | // first fixup of each page; the rest can be found by walking the chain using |
| 729 | // the offset that is embedded in each entry. |
| 730 | // |
| 731 | // This setup allows pages to be relocated lazily at page-in time and without |
| 732 | // being dirtied. The kernel can discard and load them again as needed. This |
| 733 | // technique, called page-in linking, was introduced in macOS 13. |
| 734 | // |
| 735 | // The benefits of this format are: |
| 736 | // - smaller __LINKEDIT segment, as most of the fixup information is stored in |
| 737 | // the data segment |
| 738 | // - faster startup, since not all relocations need to be done upfront |
| 739 | // - slightly lower memory usage, as fewer pages are dirtied |
| 740 | // |
| 741 | // Userspace x86_64 and arm64 binaries have two types of fixup entries: |
| 742 | // - Rebase entries contain an absolute address, to which the object's load |
| 743 | // address will be added to get the final value. This is used for loading |
| 744 | // the address of a symbol defined in the same binary. |
| 745 | // - Binding entries are mostly used for symbols imported from other dylibs, |
| 746 | // but for weakly bound and interposable symbols as well. They are looked up |
| 747 | // by a (symbol name, library) pair stored in __chainfixups. This import |
| 748 | // entry also encodes whether the import is weak (i.e. if the symbol is |
| 749 | // missing, it should be set to null instead of producing a load error). |
| 750 | // The fixup encodes an ordinal associated with the import, and an optional |
| 751 | // addend. |
| 752 | // |
| 753 | // The entries are tightly packed 64-bit bitfields. One of the bits specifies |
| 754 | // which kind of fixup to interpret them as. |
| 755 | // |
| 756 | // LLD generates the fixup data in 5 stages: |
| 757 | // 1. While scanning relocations, we make a note of each location that needs |
| 758 | // a fixup by calling addRebase() or addBinding(). During this, we assign |
| 759 | // a unique ordinal for each (symbol name, library, addend) import tuple. |
| 760 | // 2. After addresses have been assigned to all sections, and thus the memory |
| 761 | // layout of the linked image is final; finalizeContents() is called. Here, |
| 762 | // the page offsets of the chain start entries are calculated. |
| 763 | // 3. ChainedFixupsSection::writeTo() writes the page start offsets and the |
| 764 | // imports table to the output file. |
| 765 | // 4. Each section's fixup entries are encoded and written to disk in |
| 766 | // ConcatInputSection::writeTo(), but without writing the offsets that form |
| 767 | // the chain. |
| 768 | // 5. Finally, each page's (which might correspond to multiple sections) |
| 769 | // fixups are linked together in Writer::buildFixupChains(). |
| 770 | class ChainedFixupsSection final : public LinkEditSection { |
| 771 | public: |
| 772 | ChainedFixupsSection(); |
| 773 | void finalizeContents() override; |
| 774 | uint64_t getRawSize() const override { return size; } |
| 775 | bool isNeeded() const override; |
| 776 | void writeTo(uint8_t *buf) const override; |
| 777 | |
| 778 | void addRebase(const InputSection *isec, uint64_t offset) { |
| 779 | locations.emplace_back(args&: isec, args&: offset); |
| 780 | } |
| 781 | void addBinding(const Symbol *dysym, const InputSection *isec, |
| 782 | uint64_t offset, int64_t addend = 0); |
| 783 | |
| 784 | void setHasNonWeakDefinition() { hasNonWeakDef = true; } |
| 785 | |
| 786 | // Returns an (ordinal, inline addend) tuple used by dyld_chained_ptr_64_bind. |
| 787 | std::pair<uint32_t, uint8_t> getBinding(const Symbol *sym, |
| 788 | int64_t addend) const; |
| 789 | |
| 790 | const std::vector<Location> &getLocations() const { return locations; } |
| 791 | |
| 792 | bool hasWeakBinding() const { return hasWeakBind; } |
| 793 | bool hasNonWeakDefinition() const { return hasNonWeakDef; } |
| 794 | |
| 795 | private: |
| 796 | // Location::offset initially stores the offset within an InputSection, but |
| 797 | // contains output segment offsets after finalizeContents(). |
| 798 | std::vector<Location> locations; |
| 799 | // (target symbol, addend) => import ordinal |
| 800 | llvm::MapVector<std::pair<const Symbol *, int64_t>, uint32_t> bindings; |
| 801 | |
| 802 | struct SegmentInfo { |
| 803 | SegmentInfo(const OutputSegment *oseg) : oseg(oseg) {} |
| 804 | |
| 805 | const OutputSegment *oseg; |
| 806 | // (page index, fixup starts offset) |
| 807 | llvm::SmallVector<std::pair<uint16_t, uint16_t>> pageStarts; |
| 808 | |
| 809 | size_t getSize() const; |
| 810 | size_t writeTo(uint8_t *buf) const; |
| 811 | }; |
| 812 | llvm::SmallVector<SegmentInfo, 4> fixupSegments; |
| 813 | |
| 814 | size_t symtabSize = 0; |
| 815 | size_t size = 0; |
| 816 | |
| 817 | bool needsAddend = false; |
| 818 | bool needsLargeAddend = false; |
| 819 | bool hasWeakBind = false; |
| 820 | bool hasNonWeakDef = false; |
| 821 | llvm::MachO::ChainedImportFormat importFormat; |
| 822 | }; |
| 823 | |
| 824 | void writeChainedRebase(uint8_t *buf, uint64_t targetVA); |
| 825 | void writeChainedFixup(uint8_t *buf, const Symbol *sym, int64_t addend); |
| 826 | |
| 827 | struct InStruct { |
| 828 | const uint8_t *bufferStart = nullptr; |
| 829 | MachHeaderSection * = nullptr; |
| 830 | /// The list of cstring sections. Note that this includes \p cStringSection |
| 831 | /// and \p objcMethnameSection already. |
| 832 | llvm::SmallVector<CStringSection *> cStringSections; |
| 833 | CStringSection *cStringSection = nullptr; |
| 834 | DeduplicatedCStringSection *objcMethnameSection = nullptr; |
| 835 | WordLiteralSection *wordLiteralSection = nullptr; |
| 836 | RebaseSection *rebase = nullptr; |
| 837 | BindingSection *binding = nullptr; |
| 838 | WeakBindingSection *weakBinding = nullptr; |
| 839 | LazyBindingSection *lazyBinding = nullptr; |
| 840 | ExportSection *exports = nullptr; |
| 841 | GotSection *got = nullptr; |
| 842 | TlvPointerSection *tlvPointers = nullptr; |
| 843 | LazyPointerSection *lazyPointers = nullptr; |
| 844 | StubsSection *stubs = nullptr; |
| 845 | StubHelperSection *stubHelper = nullptr; |
| 846 | ObjCStubsSection *objcStubs = nullptr; |
| 847 | UnwindInfoSection *unwindInfo = nullptr; |
| 848 | ObjCImageInfoSection *objCImageInfo = nullptr; |
| 849 | ConcatInputSection *imageLoaderCache = nullptr; |
| 850 | InitOffsetsSection *initOffsets = nullptr; |
| 851 | ObjCMethListSection *objcMethList = nullptr; |
| 852 | ChainedFixupsSection *chainedFixups = nullptr; |
| 853 | |
| 854 | CStringSection *getOrCreateCStringSection(StringRef name, |
| 855 | bool forceDedupStrings = false) { |
| 856 | auto [it, didEmplace] = |
| 857 | cStringSectionMap.try_emplace(Key: name, Args: cStringSections.size()); |
| 858 | if (!didEmplace) |
| 859 | return cStringSections[it->getValue()]; |
| 860 | |
| 861 | std::string &nameData = *make<std::string>(args&: name); |
| 862 | CStringSection *sec; |
| 863 | if (config->dedupStrings || forceDedupStrings) |
| 864 | sec = make<DeduplicatedCStringSection>(args: nameData.c_str()); |
| 865 | else |
| 866 | sec = make<CStringSection>(args: nameData.c_str()); |
| 867 | cStringSections.push_back(Elt: sec); |
| 868 | return sec; |
| 869 | } |
| 870 | |
| 871 | private: |
| 872 | llvm::StringMap<unsigned> cStringSectionMap; |
| 873 | }; |
| 874 | |
| 875 | extern InStruct in; |
| 876 | extern std::vector<SyntheticSection *> syntheticSections; |
| 877 | |
| 878 | void createSyntheticSymbols(); |
| 879 | |
| 880 | } // namespace lld::macho |
| 881 | |
| 882 | #endif |
| 883 | |