]> git.saurik.com Git - apple/ld64.git/blob - src/ld/parsers/macho_relocatable_file.cpp
projects / apple / ld64.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ld64-264.3.101.tar.gz
[apple/ld64.git] / src / ld / parsers / macho_relocatable_file.cpp
1 /* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-
2 *
3 * Copyright (c) 2009-2010 Apple Inc. All rights reserved.
4 *
5 * @APPLE_LICENSE_HEADER_START@
6 *
7 * This file contains Original Code and/or Modifications of Original Code
8 * as defined in and that are subject to the Apple Public Source License
9 * Version 2.0 (the 'License'). You may not use this file except in
10 * compliance with the License. Please obtain a copy of the License at
11 * http://www.opensource.apple.com/apsl/ and read it before using this
12 * file.
13 *
14 * The Original Code and all software distributed under the License are
15 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
16 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
17 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
19 * Please see the License for the specific language governing rights and
20 * limitations under the License.
21 *
22 * @APPLE_LICENSE_HEADER_END@
23 */
24
25
26 #include <stdint.h>
27 #include <stdlib.h>
28 #include <math.h>
29 #include <unistd.h>
30 #include <fcntl.h>
31 #include <sys/param.h>
32 #include <sys/stat.h>
33 #include <sys/mman.h>
34
35 #include "MachOFileAbstraction.hpp"
36
37 #include "libunwind/DwarfInstructions.hpp"
38 #include "libunwind/AddressSpace.hpp"
39 #include "libunwind/Registers.hpp"
40
41 #include <vector>
42 #include <set>
43 #include <map>
44 #include <algorithm>
45 #include <type_traits>
46
47 #include "dwarf2.h"
48 #include "debugline.h"
49
50 #include "Architectures.hpp"
51 #include "Bitcode.hpp"
52 #include "ld.hpp"
53 #include "macho_relocatable_file.h"
54
55
56
57 extern void throwf(const char* format, ...) __attribute__ ((noreturn,format(printf, 1, 2)));
58 extern void warning(const char* format, ...) __attribute__((format(printf, 1, 2)));
59
60 namespace mach_o {
61 namespace relocatable {
62
63
64 // forward reference
65 template <typename A> class Parser;
66 template <typename A> class Atom;
67 template <typename A> class Section;
68 template <typename A> class CFISection;
69 template <typename A> class CUSection;
70
71 template <typename A>
72 class File : public ld::relocatable::File
73 {
74 public:
75 File(const char* p, time_t mTime, const uint8_t* content, ld::File::Ordinal ord) :
76 ld::relocatable::File(p,mTime,ord), _fileContent(content),
77 _sectionsArray(NULL), _atomsArray(NULL),
78 _sectionsArrayCount(0), _atomsArrayCount(0), _aliasAtomsArrayCount(0),
79 _debugInfoKind(ld::relocatable::File::kDebugInfoNone),
80 _dwarfTranslationUnitPath(NULL),
81 _dwarfDebugInfoSect(NULL), _dwarfDebugAbbrevSect(NULL),
82 _dwarfDebugLineSect(NULL), _dwarfDebugStringSect(NULL),
83 _objConstraint(ld::File::objcConstraintNone),
84 _swiftVersion(0),
85 _cpuSubType(0),
86 _minOSVersion(0),
87 _platform(0),
88 _canScatterAtoms(false),
89 _srcKind(kSourceUnknown) {}
90 virtual ~File();
91
92 // overrides of ld::File
93 virtual bool forEachAtom(ld::File::AtomHandler&) const;
94 virtual bool justInTimeforEachAtom(const char* name, ld::File::AtomHandler&) const
95 { return false; }
96 virtual uint32_t minOSVersion() const { return _minOSVersion; }
97 virtual uint32_t platformLoadCommand() const { return _platform; }
98
99 // overrides of ld::relocatable::File
100 virtual ObjcConstraint objCConstraint() const { return _objConstraint; }
101 virtual uint32_t cpuSubType() const { return _cpuSubType; }
102 virtual DebugInfoKind debugInfo() const { return _debugInfoKind; }
103 virtual const std::vector<ld::relocatable::File::Stab>* stabs() const { return &_stabs; }
104 virtual bool canScatterAtoms() const { return _canScatterAtoms; }
105 virtual const char* translationUnitSource() const;
106 virtual LinkerOptionsList* linkerOptions() const { return &_linkerOptions; }
107 virtual uint8_t swiftVersion() const { return _swiftVersion; }
108 virtual ld::Bitcode* getBitcode() const { return _bitcode.get(); }
109 virtual SourceKind sourceKind() const { return _srcKind; }
110
111 const uint8_t* fileContent() { return _fileContent; }
112 private:
113 friend class Atom<A>;
114 friend class Section<A>;
115 friend class Parser<A>;
116 friend class CFISection<A>::OAS;
117
118 typedef typename A::P P;
119
120 const uint8_t* _fileContent;
121 Section<A>** _sectionsArray;
122 uint8_t* _atomsArray;
123 uint8_t* _aliasAtomsArray;
124 uint32_t _sectionsArrayCount;
125 uint32_t _atomsArrayCount;
126 uint32_t _aliasAtomsArrayCount;
127 std::vector<ld::Fixup> _fixups;
128 std::vector<ld::Atom::UnwindInfo> _unwindInfos;
129 std::vector<ld::Atom::LineInfo> _lineInfos;
130 std::vector<ld::relocatable::File::Stab>_stabs;
131 ld::relocatable::File::DebugInfoKind _debugInfoKind;
132 const char* _dwarfTranslationUnitPath;
133 const macho_section<P>* _dwarfDebugInfoSect;
134 const macho_section<P>* _dwarfDebugAbbrevSect;
135 const macho_section<P>* _dwarfDebugLineSect;
136 const macho_section<P>* _dwarfDebugStringSect;
137 ld::File::ObjcConstraint _objConstraint;
138 uint8_t _swiftVersion;
139 uint32_t _cpuSubType;
140 uint32_t _minOSVersion;
141 uint32_t _platform;
142 bool _canScatterAtoms;
143 std::vector<std::vector<const char*> > _linkerOptions;
144 std::unique_ptr<ld::Bitcode> _bitcode;
145 SourceKind _srcKind;
146 };
147
148
149 template <typename A>
150 class Section : public ld::Section
151 {
152 public:
153 typedef typename A::P::uint_t pint_t;
154 typedef typename A::P P;
155 typedef typename A::P::E E;
156
157 virtual ~Section() { }
158 class File<A>& file() const { return _file; }
159 const macho_section<P>* machoSection() const { return _machOSection; }
160 uint32_t sectionNum(class Parser<A>&) const;
161 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr);
162 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeUnclassified; }
163 virtual bool dontDeadStrip() { return (this->_machOSection->flags() & S_ATTR_NO_DEAD_STRIP); }
164 virtual bool dontDeadStripIfReferencesLive() { return ( (this->_machOSection != NULL) && (this->_machOSection->flags() & S_ATTR_LIVE_SUPPORT) ); }
165 virtual Atom<A>* findAtomByAddress(pint_t addr) { return this->findContentAtomByAddress(addr, this->_beginAtoms, this->_endAtoms); }
166 virtual bool addFollowOnFixups() const { return ! _file.canScatterAtoms(); }
167 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
168 struct Parser<A>::LabelAndCFIBreakIterator& it,
169 const struct Parser<A>::CFI_CU_InfoArrays&) = 0;
170 virtual uint32_t computeAtomCount(class Parser<A>& parser,
171 struct Parser<A>::LabelAndCFIBreakIterator& it,
172 const struct Parser<A>::CFI_CU_InfoArrays&) = 0;
173 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
174 virtual bool addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>*);
175 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const { return 0; }
176 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
177 const ld::IndirectBindingTable& ind) const { return false; }
178 virtual bool ignoreLabel(const char* label) const { return false; }
179 static const char* makeSectionName(const macho_section<typename A::P>* s);
180
181 protected:
182 Section(File<A>& f, const macho_section<typename A::P>* s)
183 : ld::Section(makeSegmentName(s), makeSectionName(s), sectionType(s)),
184 _file(f), _machOSection(s), _beginAtoms(NULL), _endAtoms(NULL), _hasAliases(false) { }
185 Section(File<A>& f, const char* segName, const char* sectName, ld::Section::Type t, bool hidden=false)
186 : ld::Section(segName, sectName, t, hidden), _file(f), _machOSection(NULL),
187 _beginAtoms(NULL), _endAtoms(NULL), _hasAliases(false) { }
188
189
190 Atom<A>* findContentAtomByAddress(pint_t addr, class Atom<A>* start, class Atom<A>* end);
191 uint32_t x86_64PcRelOffset(uint8_t r_type);
192 void addLOH(class Parser<A>& parser, int kind, int count, const uint64_t addrs[]);
193 static const char* makeSegmentName(const macho_section<typename A::P>* s);
194 static bool readable(const macho_section<typename A::P>* s);
195 static bool writable(const macho_section<typename A::P>* s);
196 static bool exectuable(const macho_section<typename A::P>* s);
197 static ld::Section::Type sectionType(const macho_section<typename A::P>* s);
198
199 File<A>& _file;
200 const macho_section<P>* _machOSection;
201 class Atom<A>* _beginAtoms;
202 class Atom<A>* _endAtoms;
203 bool _hasAliases;
204 std::set<const class Atom<A>*> _altEntries;
205 };
206
207
208 template <typename A>
209 class CFISection : public Section<A>
210 {
211 public:
212 CFISection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
213 : Section<A>(f, s) { }
214 uint32_t cfiCount(Parser<A>& parser);
215
216 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeCFI; }
217 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
218 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
219 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
220 virtual bool addFollowOnFixups() const { return false; }
221
222
223 ///
224 /// ObjectFileAddressSpace is used as a template parameter to UnwindCursor for parsing
225 /// dwarf CFI information in an object file.
226 ///
227 class OAS
228 {
229 public:
230 typedef typename A::P::uint_t pint_t;
231 typedef typename A::P P;
232 typedef typename A::P::E E;
233 typedef typename A::P::uint_t sint_t;
234
235 OAS(CFISection<A>& ehFrameSection, const uint8_t* ehFrameBuffer) :
236 _ehFrameSection(ehFrameSection),
237 _ehFrameContent(ehFrameBuffer),
238 _ehFrameStartAddr(ehFrameSection.machoSection()->addr()),
239 _ehFrameEndAddr(ehFrameSection.machoSection()->addr()+ehFrameSection.machoSection()->size()) {}
240
241 uint8_t get8(pint_t addr) { return *((uint8_t*)mappedAddress(addr)); }
242 uint16_t get16(pint_t addr) { return E::get16(*((uint16_t*)mappedAddress(addr))); }
243 uint32_t get32(pint_t addr) { return E::get32(*((uint32_t*)mappedAddress(addr))); }
244 uint64_t get64(pint_t addr) { return E::get64(*((uint64_t*)mappedAddress(addr))); }
245 pint_t getP(pint_t addr) { return P::getP(*((pint_t*)mappedAddress(addr))); }
246 uint64_t getULEB128(pint_t& addr, pint_t end);
247 int64_t getSLEB128(pint_t& addr, pint_t end);
248 pint_t getEncodedP(pint_t& addr, pint_t end, uint8_t encoding);
249 private:
250 const void* mappedAddress(pint_t addr);
251
252 CFISection<A>& _ehFrameSection;
253 const uint8_t* _ehFrameContent;
254 pint_t _ehFrameStartAddr;
255 pint_t _ehFrameEndAddr;
256 };
257
258
259 typedef typename A::P::uint_t pint_t;
260 typedef libunwind::CFI_Atom_Info<OAS> CFI_Atom_Info;
261
262 void cfiParse(class Parser<A>& parser, uint8_t* buffer, CFI_Atom_Info cfiArray[], uint32_t& cfiCount, const pint_t cuStarts[], uint32_t cuCount);
263 bool needsRelocating();
264
265 static bool bigEndian();
266 private:
267 void addCiePersonalityFixups(class Parser<A>& parser, const CFI_Atom_Info* cieInfo);
268 static void warnFunc(void* ref, uint64_t funcAddr, const char* msg);
269 };
270
271
272 template <typename A>
273 class CUSection : public Section<A>
274 {
275 public:
276 CUSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
277 : Section<A>(f, s) { }
278
279 typedef typename A::P::uint_t pint_t;
280 typedef typename A::P P;
281 typedef typename A::P::E E;
282
283 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&) { return 0; }
284 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&) { return 0; }
285 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
286 virtual bool addFollowOnFixups() const { return false; }
287
288 struct Info {
289 pint_t functionStartAddress;
290 uint32_t functionSymbolIndex;
291 uint32_t rangeLength;
292 uint32_t compactUnwindInfo;
293 const char* personality;
294 pint_t lsdaAddress;
295 Atom<A>* function;
296 Atom<A>* lsda;
297 };
298
299 uint32_t count();
300 void parse(class Parser<A>& parser, uint32_t cnt, Info array[]);
301 static bool encodingMeansUseDwarf(compact_unwind_encoding_t enc);
302
303
304 private:
305
306 const char* personalityName(class Parser<A>& parser, const macho_relocation_info<P>* reloc);
307
308 static int infoSorter(const void* l, const void* r);
309
310 };
311
312
313 template <typename A>
314 class TentativeDefinitionSection : public Section<A>
315 {
316 public:
317 TentativeDefinitionSection(Parser<A>& parser, File<A>& f)
318 : Section<A>(f, "__DATA", "__comm/tent", ld::Section::typeTentativeDefs) {}
319
320 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeZeroFill; }
321 virtual bool addFollowOnFixups() const { return false; }
322 virtual Atom<A>* findAtomByAddress(typename A::P::uint_t addr) { throw "TentativeDefinitionSection::findAtomByAddress() should never be called"; }
323 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it,
324 const struct Parser<A>::CFI_CU_InfoArrays&);
325 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
326 struct Parser<A>::LabelAndCFIBreakIterator& it,
327 const struct Parser<A>::CFI_CU_InfoArrays&);
328 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&) {}
329 private:
330 typedef typename A::P::uint_t pint_t;
331 typedef typename A::P P;
332 };
333
334
335 template <typename A>
336 class AbsoluteSymbolSection : public Section<A>
337 {
338 public:
339 AbsoluteSymbolSection(Parser<A>& parser, File<A>& f)
340 : Section<A>(f, "__DATA", "__abs", ld::Section::typeAbsoluteSymbols, true) {}
341
342 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeUnclassified; }
343 virtual bool dontDeadStrip() { return false; }
344 virtual ld::Atom::Alignment alignmentForAddress(typename A::P::uint_t addr) { return ld::Atom::Alignment(0); }
345 virtual bool addFollowOnFixups() const { return false; }
346 virtual Atom<A>* findAtomByAddress(typename A::P::uint_t addr) { throw "AbsoluteSymbolSection::findAtomByAddress() should never be called"; }
347 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it,
348 const struct Parser<A>::CFI_CU_InfoArrays&);
349 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
350 struct Parser<A>::LabelAndCFIBreakIterator& it,
351 const struct Parser<A>::CFI_CU_InfoArrays&);
352 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&) {}
353 virtual Atom<A>* findAbsAtomForValue(typename A::P::uint_t);
354
355 private:
356 typedef typename A::P::uint_t pint_t;
357 typedef typename A::P P;
358 };
359
360
361 template <typename A>
362 class SymboledSection : public Section<A>
363 {
364 public:
365 SymboledSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s);
366 virtual ld::Atom::ContentType contentType() { return _type; }
367 virtual bool dontDeadStrip();
368 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it,
369 const struct Parser<A>::CFI_CU_InfoArrays&);
370 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer,
371 struct Parser<A>::LabelAndCFIBreakIterator& it,
372 const struct Parser<A>::CFI_CU_InfoArrays&);
373 protected:
374 typedef typename A::P::uint_t pint_t;
375 typedef typename A::P P;
376
377 ld::Atom::ContentType _type;
378 };
379
380
381 template <typename A>
382 class TLVDefsSection : public SymboledSection<A>
383 {
384 public:
385 TLVDefsSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s) :
386 SymboledSection<A>(parser, f, s) { }
387
388 private:
389
390 };
391
392
393 template <typename A>
394 class ImplicitSizeSection : public Section<A>
395 {
396 public:
397 ImplicitSizeSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
398 : Section<A>(f, s) { }
399 virtual uint32_t computeAtomCount(class Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
400 virtual uint32_t appendAtoms(class Parser<A>& parser, uint8_t* buffer, struct Parser<A>::LabelAndCFIBreakIterator& it, const struct Parser<A>::CFI_CU_InfoArrays&);
401 protected:
402 typedef typename A::P::uint_t pint_t;
403 typedef typename A::P P;
404
405 virtual bool addFollowOnFixups() const { return false; }
406 virtual const char* unlabeledAtomName(Parser<A>& parser, pint_t addr) = 0;
407 virtual ld::Atom::SymbolTableInclusion symbolTableInclusion();
408 virtual pint_t elementSizeAtAddress(pint_t addr) = 0;
409 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& parser, pint_t addr) { return ld::Atom::scopeLinkageUnit; }
410 virtual bool useElementAt(Parser<A>& parser,
411 struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr) = 0;
412 virtual ld::Atom::Definition definition() { return ld::Atom::definitionRegular; }
413 virtual ld::Atom::Combine combine(Parser<A>& parser, pint_t addr) = 0;
414 virtual bool ignoreLabel(const char* label) const { return (label[0] == 'L'); }
415 };
416
417
418 template <typename A>
419 class FixedSizeSection : public ImplicitSizeSection<A>
420 {
421 public:
422 FixedSizeSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
423 : ImplicitSizeSection<A>(parser, f, s) { }
424 protected:
425 typedef typename A::P::uint_t pint_t;
426 typedef typename A::P P;
427 typedef typename A::P::E E;
428
429 virtual bool useElementAt(Parser<A>& parser,
430 struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr)
431 { return true; }
432 };
433
434
435 template <typename A>
436 class Literal4Section : public FixedSizeSection<A>
437 {
438 public:
439 Literal4Section(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
440 : FixedSizeSection<A>(parser, f, s) {}
441 protected:
442 typedef typename A::P::uint_t pint_t;
443 typedef typename A::P P;
444
445 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(2); }
446 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "4-byte-literal"; }
447 virtual pint_t elementSizeAtAddress(pint_t addr) { return 4; }
448 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
449 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
450 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
451 const ld::IndirectBindingTable& ind) const;
452 virtual bool ignoreLabel(const char* label) const;
453 };
454
455 template <typename A>
456 class Literal8Section : public FixedSizeSection<A>
457 {
458 public:
459 Literal8Section(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
460 : FixedSizeSection<A>(parser, f, s) {}
461 protected:
462 typedef typename A::P::uint_t pint_t;
463 typedef typename A::P P;
464
465 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(3); }
466 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "8-byte-literal"; }
467 virtual pint_t elementSizeAtAddress(pint_t addr) { return 8; }
468 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
469 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
470 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
471 const ld::IndirectBindingTable& ind) const;
472 virtual bool ignoreLabel(const char* label) const;
473 };
474
475 template <typename A>
476 class Literal16Section : public FixedSizeSection<A>
477 {
478 public:
479 Literal16Section(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
480 : FixedSizeSection<A>(parser, f, s) {}
481 protected:
482 typedef typename A::P::uint_t pint_t;
483 typedef typename A::P P;
484
485 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(4); }
486 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "16-byte-literal"; }
487 virtual pint_t elementSizeAtAddress(pint_t addr) { return 16; }
488 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
489 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
490 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
491 const ld::IndirectBindingTable& ind) const;
492 virtual bool ignoreLabel(const char* label) const;
493 };
494
495
496 template <typename A>
497 class NonLazyPointerSection : public FixedSizeSection<A>
498 {
499 public:
500 NonLazyPointerSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
501 : FixedSizeSection<A>(parser, f, s) {}
502 protected:
503 typedef typename A::P::uint_t pint_t;
504 typedef typename A::P P;
505
506 virtual void makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&);
507 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeNonLazyPointer; }
508 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
509 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "non_lazy_ptr"; }
510 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
511 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& parser, pint_t addr);
512 virtual ld::Atom::Combine combine(Parser<A>&, pint_t);
513 virtual bool ignoreLabel(const char* label) const { return true; }
514 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
515 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
516 const ld::IndirectBindingTable& ind) const;
517
518 private:
519 static const char* targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind);
520 static ld::Fixup::Kind fixupKind();
521 };
522
523 template <typename A>
524 class TLVPointerSection : public FixedSizeSection<A>
525 {
526 public:
527 TLVPointerSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
528 : FixedSizeSection<A>(parser, f, s) {}
529 protected:
530 typedef typename A::P::uint_t pint_t;
531 typedef typename A::P P;
532
533 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeTLVPointer; }
534 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
535 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "tlv_lazy_ptr"; }
536 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
537 virtual ld::Atom::Combine combine(Parser<A>&, pint_t);
538 virtual bool ignoreLabel(const char* label) const { return true; }
539 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
540 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
541 const ld::IndirectBindingTable& ind) const;
542
543 private:
544 static const char* targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind, bool* isStatic);
545 };
546
547
548 template <typename A>
549 class CFStringSection : public FixedSizeSection<A>
550 {
551 public:
552 CFStringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
553 : FixedSizeSection<A>(parser, f, s) {}
554 protected:
555 typedef typename A::P::uint_t pint_t;
556
557 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
558 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "CFString"; }
559 virtual pint_t elementSizeAtAddress(pint_t addr) { return 4*sizeof(pint_t); }
560 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndReferences; }
561 virtual bool ignoreLabel(const char* label) const { return true; }
562 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
563 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
564 const ld::IndirectBindingTable& ind) const;
565 private:
566 enum ContentType { contentUTF8, contentUTF16, contentUnknown };
567 static const uint8_t* targetContent(const class Atom<A>* atom, const ld::IndirectBindingTable& ind,
568 ContentType* ct, unsigned int* count);
569 };
570
571
572 template <typename A>
573 class ObjC1ClassSection : public FixedSizeSection<A>
574 {
575 public:
576 ObjC1ClassSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
577 : FixedSizeSection<A>(parser, f, s) {}
578 protected:
579 typedef typename A::P::uint_t pint_t;
580 typedef typename A::P P;
581 typedef typename A::P::E E;
582
583 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& , pint_t ) { return ld::Atom::scopeGlobal; }
584 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(2); }
585 virtual const char* unlabeledAtomName(Parser<A>&, pint_t);
586 virtual ld::Atom::SymbolTableInclusion symbolTableInclusion() { return ld::Atom::symbolTableIn; }
587 virtual pint_t elementSizeAtAddress(pint_t addr);
588 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineNever; }
589 virtual bool ignoreLabel(const char* label) const { return true; }
590 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
591 { return 0; }
592 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
593 const ld::IndirectBindingTable& ind) const { return false; }
594 virtual bool addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>*);
595 };
596
597
598 template <typename A>
599 class ObjC2ClassRefsSection : public FixedSizeSection<A>
600 {
601 public:
602 ObjC2ClassRefsSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
603 : FixedSizeSection<A>(parser, f, s) {}
604 protected:
605 typedef typename A::P::uint_t pint_t;
606
607 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
608 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "objc-class-ref"; }
609 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
610 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndReferences; }
611 virtual bool ignoreLabel(const char* label) const { return true; }
612 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
613 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
614 const ld::IndirectBindingTable& ind) const;
615 private:
616 const char* targetClassName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
617 };
618
619
620 template <typename A>
621 class ObjC2CategoryListSection : public FixedSizeSection<A>
622 {
623 public:
624 ObjC2CategoryListSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
625 : FixedSizeSection<A>(parser, f, s) {}
626 protected:
627 typedef typename A::P::uint_t pint_t;
628
629 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
630 virtual ld::Atom::Scope scopeAtAddress(Parser<A>& parser, pint_t addr) { return ld::Atom::scopeTranslationUnit; }
631 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "objc-cat-list"; }
632 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
633 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineNever; }
634 virtual bool ignoreLabel(const char* label) const { return true; }
635 private:
636 const char* targetClassName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
637 };
638
639
640 template <typename A>
641 class PointerToCStringSection : public FixedSizeSection<A>
642 {
643 public:
644 PointerToCStringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
645 : FixedSizeSection<A>(parser, f, s) {}
646 protected:
647 typedef typename A::P::uint_t pint_t;
648
649 virtual ld::Atom::Alignment alignmentForAddress(pint_t addr) { return ld::Atom::Alignment(log2(sizeof(pint_t))); }
650 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "pointer-to-literal-cstring"; }
651 virtual pint_t elementSizeAtAddress(pint_t addr) { return sizeof(pint_t); }
652 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndReferences; }
653 virtual bool ignoreLabel(const char* label) const { return true; }
654 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
655 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
656 const ld::IndirectBindingTable& ind) const;
657 virtual const char* targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
658 };
659
660
661 template <typename A>
662 class Objc1ClassReferences : public PointerToCStringSection<A>
663 {
664 public:
665 Objc1ClassReferences(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
666 : PointerToCStringSection<A>(parser, f, s) {}
667
668 typedef typename A::P::uint_t pint_t;
669 typedef typename A::P P;
670
671 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "pointer-to-literal-objc-class-name"; }
672 virtual bool addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>*);
673 virtual const char* targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
674 };
675
676
677 template <typename A>
678 class CStringSection : public ImplicitSizeSection<A>
679 {
680 public:
681 CStringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
682 : ImplicitSizeSection<A>(parser, f, s) {}
683 protected:
684 typedef typename A::P::uint_t pint_t;
685 typedef typename A::P P;
686
687 virtual ld::Atom::ContentType contentType() { return ld::Atom::typeCString; }
688 virtual Atom<A>* findAtomByAddress(pint_t addr);
689 virtual const char* unlabeledAtomName(Parser<A>&, pint_t) { return "cstring"; }
690 virtual pint_t elementSizeAtAddress(pint_t addr);
691 virtual bool ignoreLabel(const char* label) const;
692 virtual bool useElementAt(Parser<A>& parser,
693 struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr);
694 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
695 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
696 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
697 const ld::IndirectBindingTable& ind) const;
698
699 };
700
701
702 template <typename A>
703 class UTF16StringSection : public SymboledSection<A>
704 {
705 public:
706 UTF16StringSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
707 : SymboledSection<A>(parser, f, s) {}
708 protected:
709 typedef typename A::P::uint_t pint_t;
710 typedef typename A::P P;
711
712 virtual ld::Atom::Combine combine(Parser<A>&, pint_t) { return ld::Atom::combineByNameAndContent; }
713 virtual unsigned long contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const;
714 virtual bool canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
715 const ld::IndirectBindingTable& ind) const;
716 };
717
718
719 //
720 // Atoms in mach-o files
721 //
722 template <typename A>
723 class Atom : public ld::Atom
724 {
725 public:
726 // overrides of ld::Atom
727 virtual const ld::File* file() const;
728 virtual const char* translationUnitSource() const
729 { return sect().file().translationUnitSource(); }
730 virtual const char* name() const { return _name; }
731 virtual uint64_t size() const { return _size; }
732 virtual uint64_t objectAddress() const { return _objAddress; }
733 virtual void copyRawContent(uint8_t buffer[]) const;
734 virtual const uint8_t* rawContentPointer() const { return contentPointer(); }
735 virtual unsigned long contentHash(const ld::IndirectBindingTable& ind) const
736 { if ( _hash == 0 ) _hash = sect().contentHash(this, ind); return _hash; }
737 virtual bool canCoalesceWith(const ld::Atom& rhs, const ld::IndirectBindingTable& ind) const
738 { return sect().canCoalesceWith(this, rhs, ind); }
739 virtual ld::Fixup::iterator fixupsBegin() const { return &machofile()._fixups[_fixupsStartIndex]; }
740 virtual ld::Fixup::iterator fixupsEnd() const { return &machofile()._fixups[_fixupsStartIndex+_fixupsCount]; }
741 virtual ld::Atom::UnwindInfo::iterator beginUnwind() const { return &machofile()._unwindInfos[_unwindInfoStartIndex]; }
742 virtual ld::Atom::UnwindInfo::iterator endUnwind() const { return &machofile()._unwindInfos[_unwindInfoStartIndex+_unwindInfoCount]; }
743 virtual ld::Atom::LineInfo::iterator beginLineInfo() const{ return &machofile()._lineInfos[_lineInfoStartIndex]; }
744 virtual ld::Atom::LineInfo::iterator endLineInfo() const { return &machofile()._lineInfos[_lineInfoStartIndex+_lineInfoCount]; }
745 virtual void setFile(const ld::File* f);
746
747 private:
748
749 enum { kFixupStartIndexBits = 32,
750 kLineInfoStartIndexBits = 32,
751 kUnwindInfoStartIndexBits = 24,
752 kFixupCountBits = 24,
753 kLineInfoCountBits = 12,
754 kUnwindInfoCountBits = 4
755 }; // must sum to 128
756
757 public:
758 // methods for all atoms from mach-o object file
759 Section<A>& sect() const { return (Section<A>&)section(); }
760 File<A>& machofile() const { return ((Section<A>*)(this->_section))->file(); }
761 void setFixupsRange(uint32_t s, uint32_t c);
762 void setUnwindInfoRange(uint32_t s, uint32_t c);
763 void extendUnwindInfoRange();
764 void setLineInfoRange(uint32_t s, uint32_t c);
765 bool roomForMoreLineInfoCount() { return (_lineInfoCount < ((1<<kLineInfoCountBits)-1)); }
766 void incrementLineInfoCount() { assert(roomForMoreLineInfoCount()); ++_lineInfoCount; }
767 void incrementFixupCount() { if (_fixupsCount == ((1 << kFixupCountBits)-1))
768 throwf("too may fixups in %s", name()); ++_fixupsCount; }
769 const uint8_t* contentPointer() const;
770 uint32_t fixupCount() const { return _fixupsCount; }
771 void verifyAlignment(const macho_section<typename A::P>&) const;
772
773 typedef typename A::P P;
774 typedef typename A::P::E E;
775 typedef typename A::P::uint_t pint_t;
776 // constuct via all attributes
777 Atom(Section<A>& sct, const char* nm, pint_t addr, uint64_t sz,
778 ld::Atom::Definition d, ld::Atom::Combine c, ld::Atom::Scope s,
779 ld::Atom::ContentType ct, ld::Atom::SymbolTableInclusion i,
780 bool dds, bool thumb, bool al, ld::Atom::Alignment a)
781 : ld::Atom((ld::Section&)sct, d, c, s, ct, i, dds, thumb, al, a),
782 _size(sz), _objAddress(addr), _name(nm), _hash(0),
783 _fixupsStartIndex(0), _lineInfoStartIndex(0),
784 _unwindInfoStartIndex(0), _fixupsCount(0),
785 _lineInfoCount(0), _unwindInfoCount(0) { }
786 // construct via symbol table entry
787 Atom(Section<A>& sct, Parser<A>& parser, const macho_nlist<P>& sym,
788 uint64_t sz, bool alias=false)
789 : ld::Atom((ld::Section&)sct, parser.definitionFromSymbol(sym),
790 parser.combineFromSymbol(sym), parser.scopeFromSymbol(sym),
791 parser.resolverFromSymbol(sym) ? ld::Atom::typeResolver : sct.contentType(),
792 parser.inclusionFromSymbol(sym),
793 (parser.dontDeadStripFromSymbol(sym) && !sct.dontDeadStripIfReferencesLive()) || sct.dontDeadStrip(),
794 parser.isThumbFromSymbol(sym), alias,
795 sct.alignmentForAddress(sym.n_value())),
796 _size(sz), _objAddress(sym.n_value()),
797 _name(parser.nameFromSymbol(sym)), _hash(0),
798 _fixupsStartIndex(0), _lineInfoStartIndex(0),
799 _unwindInfoStartIndex(0), _fixupsCount(0),
800 _lineInfoCount(0), _unwindInfoCount(0) {
801 // <rdar://problem/6783167> support auto-hidden weak symbols
802 if ( _scope == ld::Atom::scopeGlobal &&
803 (sym.n_desc() & (N_WEAK_DEF|N_WEAK_REF)) == (N_WEAK_DEF|N_WEAK_REF) )
804 this->setAutoHide();
805 this->verifyAlignment(*sct.machoSection());
806 if ( sct.dontDeadStripIfReferencesLive() )
807 this->setDontDeadStripIfReferencesLive();
808 }
809
810 private:
811 friend class Parser<A>;
812 friend class Section<A>;
813 friend class CStringSection<A>;
814 friend class AbsoluteSymbolSection<A>;
815
816 pint_t _size;
817 pint_t _objAddress;
818 const char* _name;
819 mutable unsigned long _hash;
820
821 uint64_t _fixupsStartIndex : kFixupStartIndexBits,
822 _lineInfoStartIndex : kLineInfoStartIndexBits,
823 _unwindInfoStartIndex : kUnwindInfoStartIndexBits,
824 _fixupsCount : kFixupCountBits,
825 _lineInfoCount : kLineInfoCountBits,
826 _unwindInfoCount : kUnwindInfoCountBits;
827
828 static std::map<const ld::Atom*, const ld::File*> _s_fileOverride;
829 };
830
831 template <typename A>
832 std::map<const ld::Atom*, const ld::File*> Atom<A>::_s_fileOverride;
833
834 template <typename A>
835 void Atom<A>::setFile(const ld::File* f) {
836 _s_fileOverride[this] = f;
837 }
838
839 template <typename A>
840 const ld::File* Atom<A>::file() const
841 {
842 std::map<const ld::Atom*, const ld::File*>::iterator pos = _s_fileOverride.find(this);
843 if ( pos != _s_fileOverride.end() )
844 return pos->second;
845
846 return &sect().file();
847 }
848
849 template <typename A>
850 void Atom<A>::setFixupsRange(uint32_t startIndex, uint32_t count)
851 {
852 if ( count >= (1 << kFixupCountBits) )
853 throwf("too many fixups in function %s", this->name());
854 if ( startIndex >= (1 << kFixupStartIndexBits) )
855 throwf("too many fixups in file");
856 assert(((startIndex+count) <= sect().file()._fixups.size()) && "fixup index out of range");
857 _fixupsStartIndex = startIndex;
858 _fixupsCount = count;
859 }
860
861 template <typename A>
862 void Atom<A>::setUnwindInfoRange(uint32_t startIndex, uint32_t count)
863 {
864 if ( count >= (1 << kUnwindInfoCountBits) )
865 throwf("too many compact unwind infos in function %s", this->name());
866 if ( startIndex >= (1 << kUnwindInfoStartIndexBits) )
867 throwf("too many compact unwind infos (%d) in file", startIndex);
868 assert((startIndex+count) <= sect().file()._unwindInfos.size() && "unwindinfo index out of range");
869 _unwindInfoStartIndex = startIndex;
870 _unwindInfoCount = count;
871 }
872
873 template <typename A>
874 void Atom<A>::extendUnwindInfoRange()
875 {
876 if ( _unwindInfoCount+1 >= (1 << kUnwindInfoCountBits) )
877 throwf("too many compact unwind infos in function %s", this->name());
878 _unwindInfoCount += 1;
879 }
880
881 template <typename A>
882 void Atom<A>::setLineInfoRange(uint32_t startIndex, uint32_t count)
883 {
884 assert((count < (1 << kLineInfoCountBits)) && "too many line infos");
885 assert((startIndex+count) < sect().file()._lineInfos.size() && "line info index out of range");
886 _lineInfoStartIndex = startIndex;
887 _lineInfoCount = count;
888 }
889
890 template <typename A>
891 const uint8_t* Atom<A>::contentPointer() const
892 {
893 const macho_section<P>* sct = this->sect().machoSection();
894 if ( this->_objAddress > sct->addr() + sct->size() )
895 throwf("malformed .o file, symbol has address 0x%0llX which is outside range of its section", (uint64_t)this->_objAddress);
896 uint32_t fileOffset = sct->offset() - sct->addr() + this->_objAddress;
897 return this->sect().file().fileContent()+fileOffset;
898 }
899
900
901 template <typename A>
902 void Atom<A>::copyRawContent(uint8_t buffer[]) const
903 {
904 // copy base bytes
905 if ( this->contentType() == ld::Atom::typeZeroFill ) {
906 bzero(buffer, _size);
907 }
908 else if ( _size != 0 ) {
909 memcpy(buffer, this->contentPointer(), _size);
910 }
911 }
912
913 template <>
914 void Atom<arm>::verifyAlignment(const macho_section<P>&) const
915 {
916 if ( (this->section().type() == ld::Section::typeCode) && ! isThumb() ) {
917 if ( ((_objAddress % 4) != 0) || (this->alignment().powerOf2 < 2) )
918 warning("ARM function not 4-byte aligned: %s from %s", this->name(), this->file()->path());
919 }
920 }
921
922 #if SUPPORT_ARCH_arm64
923 template <>
924 void Atom<arm64>::verifyAlignment(const macho_section<P>& sect) const
925 {
926 if ( (this->section().type() == ld::Section::typeCode) && (sect.size() != 0) ) {
927 if ( ((_objAddress % 4) != 0) || (this->alignment().powerOf2 < 2) )
928 warning("arm64 function not 4-byte aligned: %s from %s", this->name(), this->file()->path());
929 }
930 }
931 #endif
932
933 template <typename A>
934 void Atom<A>::verifyAlignment(const macho_section<P>&) const
935 {
936 }
937
938
939 class AliasAtom : public ld::Atom
940 {
941 public:
942 AliasAtom(const char* name, bool hidden, const ld::File* file, const char* aliasOfName) :
943 ld::Atom(_s_section, ld::Atom::definitionRegular, ld::Atom::combineNever,
944 (hidden ? ld::Atom::scopeLinkageUnit : ld::Atom::scopeGlobal),
945 ld::Atom::typeUnclassified, ld::Atom::symbolTableIn,
946 false, false, true, 0),
947 _file(file),
948 _name(name),
949 _fixup(0, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, ld::Fixup::bindingByNameUnbound, aliasOfName) { }
950
951 virtual const ld::File* file() const { return _file; }
952 virtual const char* translationUnitSource() const
953 { return NULL; }
954 virtual const char* name() const { return _name; }
955 virtual uint64_t size() const { return 0; }
956 virtual uint64_t objectAddress() const { return 0; }
957 virtual void copyRawContent(uint8_t buffer[]) const { }
958 virtual ld::Fixup::iterator fixupsBegin() const { return &((ld::Fixup*)&_fixup)[0]; }
959 virtual ld::Fixup::iterator fixupsEnd() const { return &((ld::Fixup*)&_fixup)[1]; }
960
961 private:
962 static ld::Section _s_section;
963
964 const ld::File* _file;
965 const char* _name;
966 ld::Fixup _fixup;
967 };
968
969 ld::Section AliasAtom::_s_section("__LD", "__aliases", ld::Section::typeTempAlias, true);
970
971
972 template <typename A>
973 class Parser
974 {
975 public:
976 static bool validFile(const uint8_t* fileContent, bool subtypeMustMatch=false,
977 cpu_subtype_t subtype=0);
978 static const char* fileKind(const uint8_t* fileContent);
979 static Options::Platform findPlatform(const macho_header<typename A::P>* header);
980 static bool hasObjC2Categories(const uint8_t* fileContent);
981 static bool hasObjC1Categories(const uint8_t* fileContent);
982 static bool getNonLocalSymbols(const uint8_t* fileContnet, std::vector<const char*> &syms);
983 static ld::relocatable::File* parse(const uint8_t* fileContent, uint64_t fileLength,
984 const char* path, time_t modTime, ld::File::Ordinal ordinal,
985 const ParserOptions& opts) {
986 Parser p(fileContent, fileLength, path, modTime,
987 ordinal, opts.warnUnwindConversionProblems,
988 opts.keepDwarfUnwind, opts.forceDwarfConversion,
989 opts.neverConvertDwarf, opts.verboseOptimizationHints,
990 opts.ignoreMismatchPlatform);
991 return p.parse(opts);
992 }
993
994 typedef typename A::P P;
995 typedef typename A::P::E E;
996 typedef typename A::P::uint_t pint_t;
997
998 struct SourceLocation {
999 SourceLocation() {}
1000 SourceLocation(Atom<A>* a, uint32_t o) : atom(a), offsetInAtom(o) {}
1001 Atom<A>* atom;
1002 uint32_t offsetInAtom;
1003 };
1004
1005 struct TargetDesc {
1006 Atom<A>* atom;
1007 const char* name; // only used if targetAtom is NULL
1008 int64_t addend;
1009 bool weakImport; // only used if targetAtom is NULL
1010 };
1011
1012 struct FixupInAtom {
1013 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, Atom<A>* target) :
1014 fixup(src.offsetInAtom, c, k, target), atom(src.atom) { src.atom->incrementFixupCount(); }
1015
1016 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, Atom<A>* target) :
1017 fixup(src.offsetInAtom, c, k, b, target), atom(src.atom) { src.atom->incrementFixupCount(); }
1018
1019 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, bool wi, const char* name) :
1020 fixup(src.offsetInAtom, c, k, wi, name), atom(src.atom) { src.atom->incrementFixupCount(); }
1021
1022 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, const char* name) :
1023 fixup(src.offsetInAtom, c, k, b, name), atom(src.atom) { src.atom->incrementFixupCount(); }
1024
1025 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, uint64_t addend) :
1026 fixup(src.offsetInAtom, c, k, addend), atom(src.atom) { src.atom->incrementFixupCount(); }
1027
1028 FixupInAtom(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k) :
1029 fixup(src.offsetInAtom, c, k, (uint64_t)0), atom(src.atom) { src.atom->incrementFixupCount(); }
1030
1031 ld::Fixup fixup;
1032 Atom<A>* atom;
1033 };
1034
1035 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, Atom<A>* target) {
1036 _allFixups.push_back(FixupInAtom(src, c, k, target));
1037 }
1038
1039 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, Atom<A>* target) {
1040 _allFixups.push_back(FixupInAtom(src, c, k, b, target));
1041 }
1042
1043 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, bool wi, const char* name) {
1044 _allFixups.push_back(FixupInAtom(src, c, k, wi, name));
1045 }
1046
1047 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, ld::Fixup::TargetBinding b, const char* name) {
1048 _allFixups.push_back(FixupInAtom(src, c, k, b, name));
1049 }
1050
1051 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k, uint64_t addend) {
1052 _allFixups.push_back(FixupInAtom(src, c, k, addend));
1053 }
1054
1055 void addFixup(const SourceLocation& src, ld::Fixup::Cluster c, ld::Fixup::Kind k) {
1056 _allFixups.push_back(FixupInAtom(src, c, k));
1057 }
1058
1059 const char* path() { return _path; }
1060 uint32_t symbolCount() { return _symbolCount; }
1061 uint32_t indirectSymbol(uint32_t indirectIndex);
1062 const macho_nlist<P>& symbolFromIndex(uint32_t index);
1063 const char* nameFromSymbol(const macho_nlist<P>& sym);
1064 ld::Atom::Scope scopeFromSymbol(const macho_nlist<P>& sym);
1065 static ld::Atom::Definition definitionFromSymbol(const macho_nlist<P>& sym);
1066 static ld::Atom::Combine combineFromSymbol(const macho_nlist<P>& sym);
1067 ld::Atom::SymbolTableInclusion inclusionFromSymbol(const macho_nlist<P>& sym);
1068 static bool dontDeadStripFromSymbol(const macho_nlist<P>& sym);
1069 static bool isThumbFromSymbol(const macho_nlist<P>& sym);
1070 static bool weakImportFromSymbol(const macho_nlist<P>& sym);
1071 static bool resolverFromSymbol(const macho_nlist<P>& sym);
1072 static bool altEntryFromSymbol(const macho_nlist<P>& sym);
1073 uint32_t symbolIndexFromIndirectSectionAddress(pint_t,const macho_section<P>*);
1074 const macho_section<P>* firstMachOSection() { return _sectionsStart; }
1075 const macho_section<P>* machOSectionFromSectionIndex(uint32_t index);
1076 uint32_t machOSectionCount() { return _machOSectionsCount; }
1077 uint32_t undefinedStartIndex() { return _undefinedStartIndex; }
1078 uint32_t undefinedEndIndex() { return _undefinedEndIndex; }
1079 void addFixup(FixupInAtom f) { _allFixups.push_back(f); }
1080 Section<A>* sectionForNum(unsigned int sectNum);
1081 Section<A>* sectionForAddress(pint_t addr);
1082 Atom<A>* findAtomByAddress(pint_t addr);
1083 Atom<A>* findAtomByAddressOrNullIfStub(pint_t addr);
1084 Atom<A>* findAtomByAddressOrLocalTargetOfStub(pint_t addr, uint32_t* offsetInAtom);
1085 Atom<A>* findAtomByName(const char* name); // slow!
1086 void findTargetFromAddress(pint_t addr, TargetDesc& target);
1087 void findTargetFromAddress(pint_t baseAddr, pint_t addr, TargetDesc& target);
1088 void findTargetFromAddressAndSectionNum(pint_t addr, unsigned int sectNum,
1089 TargetDesc& target);
1090 uint32_t tentativeDefinitionCount() { return _tentativeDefinitionCount; }
1091 uint32_t absoluteSymbolCount() { return _absoluteSymbolCount; }
1092
1093 uint32_t fileLength() const { return _fileLength; }
1094 bool hasStubsSection() { return (_stubsSectionNum != 0); }
1095 unsigned int stubsSectionNum() { return _stubsSectionNum; }
1096 void addDtraceExtraInfos(const SourceLocation& src, const char* provider);
1097 const char* scanSymbolTableForAddress(uint64_t addr);
1098 bool warnUnwindConversionProblems() { return _warnUnwindConversionProblems; }
1099 bool hasDataInCodeLabels() { return _hasDataInCodeLabels; }
1100 bool keepDwarfUnwind() { return _keepDwarfUnwind; }
1101 bool forceDwarfConversion() { return _forceDwarfConversion; }
1102 bool verboseOptimizationHints() { return _verboseOptimizationHints; }
1103 bool neverConvertDwarf() { return _neverConvertDwarf; }
1104 bool armUsesZeroCostExceptions() { return _armUsesZeroCostExceptions; }
1105 uint8_t maxDefaultCommonAlignment() { return _maxDefaultCommonAlignment; }
1106
1107
1108 macho_data_in_code_entry<P>* dataInCodeStart() { return _dataInCodeStart; }
1109 macho_data_in_code_entry<P>* dataInCodeEnd() { return _dataInCodeEnd; }
1110 const uint8_t* optimizationHintsStart() { return _lohStart; }
1111 const uint8_t* optimizationHintsEnd() { return _lohEnd; }
1112 bool hasOptimizationHints() { return _lohStart != _lohEnd; }
1113
1114
1115 void addFixups(const SourceLocation& src, ld::Fixup::Kind kind, const TargetDesc& target);
1116 void addFixups(const SourceLocation& src, ld::Fixup::Kind kind, const TargetDesc& target, const TargetDesc& picBase);
1117
1118
1119
1120 struct LabelAndCFIBreakIterator {
1121 typedef typename CFISection<A>::CFI_Atom_Info CFI_Atom_Info;
1122 LabelAndCFIBreakIterator(const uint32_t* ssa, uint32_t ssc, const pint_t* cfisa,
1123 uint32_t cfisc, bool ols)
1124 : sortedSymbolIndexes(ssa), sortedSymbolCount(ssc), cfiStartsArray(cfisa),
1125 cfiStartsCount(cfisc), fileHasOverlappingSymbols(ols),
1126 newSection(false), cfiIndex(0), symIndex(0) {}
1127 bool next(Parser<A>& parser, const Section<A>& sect, uint32_t sectNum, pint_t startAddr, pint_t endAddr,
1128 pint_t* addr, pint_t* size, const macho_nlist<P>** sym);
1129 pint_t peek(Parser<A>& parser, pint_t startAddr, pint_t endAddr);
1130 void beginSection() { newSection = true; symIndex = 0; }
1131
1132 const uint32_t* const sortedSymbolIndexes;
1133 const uint32_t sortedSymbolCount;
1134 const pint_t* cfiStartsArray;
1135 const uint32_t cfiStartsCount;
1136 const bool fileHasOverlappingSymbols;
1137 bool newSection;
1138 uint32_t cfiIndex;
1139 uint32_t symIndex;
1140 };
1141
1142 struct CFI_CU_InfoArrays {
1143 typedef typename CFISection<A>::CFI_Atom_Info CFI_Atom_Info;
1144 typedef typename CUSection<A>::Info CU_Info;
1145 CFI_CU_InfoArrays(const CFI_Atom_Info* cfiAr, uint32_t cfiC, CU_Info* cuAr, uint32_t cuC)
1146 : cfiArray(cfiAr), cuArray(cuAr), cfiCount(cfiC), cuCount(cuC) {}
1147 const CFI_Atom_Info* const cfiArray;
1148 CU_Info* const cuArray;
1149 const uint32_t cfiCount;
1150 const uint32_t cuCount;
1151 };
1152
1153
1154
1155 private:
1156 friend class Section<A>;
1157
1158 enum SectionType { sectionTypeIgnore, sectionTypeLiteral4, sectionTypeLiteral8, sectionTypeLiteral16,
1159 sectionTypeNonLazy, sectionTypeCFI, sectionTypeCString, sectionTypeCStringPointer,
1160 sectionTypeUTF16Strings, sectionTypeCFString, sectionTypeObjC2ClassRefs, typeObjC2CategoryList,
1161 sectionTypeObjC1Classes, sectionTypeSymboled, sectionTypeObjC1ClassRefs,
1162 sectionTypeTentativeDefinitions, sectionTypeAbsoluteSymbols, sectionTypeTLVDefs,
1163 sectionTypeCompactUnwind, sectionTypeTLVPointers};
1164
1165 template <typename P>
1166 struct MachOSectionAndSectionClass
1167 {
1168 const macho_section<P>* sect;
1169 SectionType type;
1170
1171 static int sorter(const void* l, const void* r) {
1172 const MachOSectionAndSectionClass<P>* left = (MachOSectionAndSectionClass<P>*)l;
1173 const MachOSectionAndSectionClass<P>* right = (MachOSectionAndSectionClass<P>*)r;
1174 int64_t diff = left->sect->addr() - right->sect->addr();
1175 if ( diff == 0 )
1176 return 0;
1177 if ( diff < 0 )
1178 return -1;
1179 else
1180 return 1;
1181 }
1182 };
1183
1184 struct ParserAndSectionsArray { Parser* parser; const uint32_t* sortedSectionsArray; };
1185
1186
1187 Parser(const uint8_t* fileContent, uint64_t fileLength,
1188 const char* path, time_t modTime, ld::File::Ordinal ordinal,
1189 bool warnUnwindConversionProblems, bool keepDwarfUnwind,
1190 bool forceDwarfConversion, bool neverConvertDwarf,
1191 bool verboseOptimizationHints, bool ignoreMismatchPlatform);
1192 ld::relocatable::File* parse(const ParserOptions& opts);
1193 static uint8_t loadCommandSizeMask();
1194 bool parseLoadCommands(Options::Platform platform, uint32_t minOSVersion, bool simulator, bool ignoreMismatchPlatform);
1195 void makeSections();
1196 void prescanSymbolTable();
1197 void makeSortedSymbolsArray(uint32_t symArray[], const uint32_t sectionArray[]);
1198 void makeSortedSectionsArray(uint32_t array[]);
1199 static int pointerSorter(const void* l, const void* r);
1200 static int symbolIndexSorter(void* extra, const void* l, const void* r);
1201 static int sectionIndexSorter(void* extra, const void* l, const void* r);
1202
1203 void parseDebugInfo();
1204 void parseStabs();
1205 void appendAliasAtoms(uint8_t* atomBuffer);
1206 static bool isConstFunStabs(const char *stabStr);
1207 bool read_comp_unit(const char ** name, const char ** comp_dir,
1208 uint64_t *stmt_list);
1209 pint_t realAddr(pint_t addr);
1210 const char* getDwarfString(uint64_t form, const uint8_t*& p);
1211 uint64_t getDwarfOffset(uint64_t form, const uint8_t*& di, bool dwarf64);
1212 bool skip_form(const uint8_t ** offset, const uint8_t * end,
1213 uint64_t form, uint8_t addr_size, bool dwarf64);
1214
1215
1216 // filled in by constructor
1217 const uint8_t* _fileContent;
1218 uint32_t _fileLength;
1219 const char* _path;
1220 time_t _modTime;
1221 ld::File::Ordinal _ordinal;
1222
1223 // filled in by parseLoadCommands()
1224 File<A>* _file;
1225 const macho_nlist<P>* _symbols;
1226 uint32_t _symbolCount;
1227 uint32_t _indirectSymbolCount;
1228 const char* _strings;
1229 uint32_t _stringsSize;
1230 const uint32_t* _indirectTable;
1231 uint32_t _indirectTableCount;
1232 uint32_t _undefinedStartIndex;
1233 uint32_t _undefinedEndIndex;
1234 const macho_section<P>* _sectionsStart;
1235 uint32_t _machOSectionsCount;
1236 bool _hasUUID;
1237 macho_data_in_code_entry<P>* _dataInCodeStart;
1238 macho_data_in_code_entry<P>* _dataInCodeEnd;
1239 const uint8_t* _lohStart;
1240 const uint8_t* _lohEnd;
1241
1242 // filled in by parse()
1243 CFISection<A>* _EHFrameSection;
1244 CUSection<A>* _compactUnwindSection;
1245 AbsoluteSymbolSection<A>* _absoluteSection;
1246 uint32_t _tentativeDefinitionCount;
1247 uint32_t _absoluteSymbolCount;
1248 uint32_t _symbolsInSections;
1249 bool _hasLongBranchStubs;
1250 bool _AppleObjc; // FSF has objc that uses different data layout
1251 bool _overlappingSymbols;
1252 bool _warnUnwindConversionProblems;
1253 bool _hasDataInCodeLabels;
1254 bool _keepDwarfUnwind;
1255 bool _forceDwarfConversion;
1256 bool _neverConvertDwarf;
1257 bool _verboseOptimizationHints;
1258 bool _armUsesZeroCostExceptions;
1259 bool _ignoreMismatchPlatform;
1260 bool _treateBitcodeAsData;
1261 bool _usingBitcode;
1262 uint8_t _maxDefaultCommonAlignment;
1263 unsigned int _stubsSectionNum;
1264 const macho_section<P>* _stubsMachOSection;
1265 std::vector<const char*> _dtraceProviderInfo;
1266 std::vector<FixupInAtom> _allFixups;
1267 };
1268
1269
1270
1271 template <typename A>
1272 Parser<A>::Parser(const uint8_t* fileContent, uint64_t fileLength, const char* path, time_t modTime,
1273 ld::File::Ordinal ordinal, bool convertDUI, bool keepDwarfUnwind, bool forceDwarfConversion,
1274 bool neverConvertDwarf, bool verboseOptimizationHints, bool ignoreMismatchPlatform)
1275 : _fileContent(fileContent), _fileLength(fileLength), _path(path), _modTime(modTime),
1276 _ordinal(ordinal), _file(NULL),
1277 _symbols(NULL), _symbolCount(0), _indirectSymbolCount(0), _strings(NULL), _stringsSize(0),
1278 _indirectTable(NULL), _indirectTableCount(0),
1279 _undefinedStartIndex(0), _undefinedEndIndex(0),
1280 _sectionsStart(NULL), _machOSectionsCount(0), _hasUUID(false),
1281 _dataInCodeStart(NULL), _dataInCodeEnd(NULL),
1282 _lohStart(NULL), _lohEnd(NULL),
1283 _EHFrameSection(NULL), _compactUnwindSection(NULL), _absoluteSection(NULL),
1284 _tentativeDefinitionCount(0), _absoluteSymbolCount(0),
1285 _symbolsInSections(0), _hasLongBranchStubs(false), _AppleObjc(false),
1286 _overlappingSymbols(false), _warnUnwindConversionProblems(convertDUI), _hasDataInCodeLabels(false),
1287 _keepDwarfUnwind(keepDwarfUnwind), _forceDwarfConversion(forceDwarfConversion),
1288 _neverConvertDwarf(neverConvertDwarf),
1289 _verboseOptimizationHints(verboseOptimizationHints),
1290 _ignoreMismatchPlatform(ignoreMismatchPlatform),
1291 _stubsSectionNum(0), _stubsMachOSection(NULL)
1292 {
1293 }
1294
1295
1296 template <>
1297 bool Parser<x86>::validFile(const uint8_t* fileContent, bool, cpu_subtype_t)
1298 {
1299 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1300 if ( header->magic() != MH_MAGIC )
1301 return false;
1302 if ( header->cputype() != CPU_TYPE_I386 )
1303 return false;
1304 if ( header->filetype() != MH_OBJECT )
1305 return false;
1306 return true;
1307 }
1308
1309 template <>
1310 bool Parser<x86_64>::validFile(const uint8_t* fileContent, bool, cpu_subtype_t)
1311 {
1312 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1313 if ( header->magic() != MH_MAGIC_64 )
1314 return false;
1315 if ( header->cputype() != CPU_TYPE_X86_64 )
1316 return false;
1317 if ( header->filetype() != MH_OBJECT )
1318 return false;
1319 return true;
1320 }
1321
1322 template <>
1323 bool Parser<arm>::validFile(const uint8_t* fileContent, bool subtypeMustMatch, cpu_subtype_t subtype)
1324 {
1325 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1326 if ( header->magic() != MH_MAGIC )
1327 return false;
1328 if ( header->cputype() != CPU_TYPE_ARM )
1329 return false;
1330 if ( header->filetype() != MH_OBJECT )
1331 return false;
1332 if ( subtypeMustMatch ) {
1333 if ( (cpu_subtype_t)header->cpusubtype() == subtype )
1334 return true;
1335 // hack until libcc_kext.a is made fat
1336 if ( header->cpusubtype() == CPU_SUBTYPE_ARM_ALL )
1337 return true;
1338 return false;
1339 }
1340 return true;
1341 }
1342
1343
1344 template <>
1345 bool Parser<arm64>::validFile(const uint8_t* fileContent, bool subtypeMustMatch, cpu_subtype_t subtype)
1346 {
1347 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1348 if ( header->magic() != MH_MAGIC_64 )
1349 return false;
1350 if ( header->cputype() != CPU_TYPE_ARM64 )
1351 return false;
1352 if ( header->filetype() != MH_OBJECT )
1353 return false;
1354 return true;
1355 }
1356
1357
1358 template <>
1359 const char* Parser<x86>::fileKind(const uint8_t* fileContent)
1360 {
1361 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1362 if ( header->magic() != MH_MAGIC )
1363 return NULL;
1364 if ( header->cputype() != CPU_TYPE_I386 )
1365 return NULL;
1366 return "i386";
1367 }
1368
1369 template <>
1370 const char* Parser<x86_64>::fileKind(const uint8_t* fileContent)
1371 {
1372 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1373 if ( header->magic() != MH_MAGIC_64 )
1374 return NULL;
1375 if ( header->cputype() != CPU_TYPE_X86_64 )
1376 return NULL;
1377 return "x86_64";
1378 }
1379
1380 template <>
1381 const char* Parser<arm>::fileKind(const uint8_t* fileContent)
1382 {
1383 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1384 if ( header->magic() != MH_MAGIC )
1385 return NULL;
1386 if ( header->cputype() != CPU_TYPE_ARM )
1387 return NULL;
1388 for (const ArchInfo* t=archInfoArray; t->archName != NULL; ++t) {
1389 if ( (t->cpuType == CPU_TYPE_ARM) && ((cpu_subtype_t)header->cpusubtype() == t->cpuSubType) ) {
1390 return t->archName;
1391 }
1392 }
1393 return "arm???";
1394 }
1395
1396 #if SUPPORT_ARCH_arm64
1397 template <>
1398 const char* Parser<arm64>::fileKind(const uint8_t* fileContent)
1399 {
1400 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1401 if ( header->magic() != MH_MAGIC_64 )
1402 return NULL;
1403 if ( header->cputype() != CPU_TYPE_ARM64 )
1404 return NULL;
1405 return "arm64";
1406 }
1407 #endif
1408
1409 template <typename A>
1410 bool Parser<A>::hasObjC2Categories(const uint8_t* fileContent)
1411 {
1412 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1413 const uint32_t cmd_count = header->ncmds();
1414 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1415 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1416 const macho_load_command<P>* cmd = cmds;
1417 for (uint32_t i = 0; i < cmd_count; ++i) {
1418 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
1419 const macho_segment_command<P>* segment = (macho_segment_command<P>*)cmd;
1420 const macho_section<P>* sectionsStart = (macho_section<P>*)((char*)segment + sizeof(macho_segment_command<P>));
1421 for (uint32_t si=0; si < segment->nsects(); ++si) {
1422 const macho_section<P>* sect = &sectionsStart[si];
1423 if ( (sect->size() > 0)
1424 && (strcmp(sect->sectname(), "__objc_catlist") == 0)
1425 && (strcmp(sect->segname(), "__DATA") == 0) ) {
1426 return true;
1427 }
1428 }
1429 }
1430 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
1431 if ( cmd > cmdsEnd )
1432 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
1433 }
1434 return false;
1435 }
1436
1437
1438 template <typename A>
1439 bool Parser<A>::hasObjC1Categories(const uint8_t* fileContent)
1440 {
1441 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1442 const uint32_t cmd_count = header->ncmds();
1443 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1444 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1445 const macho_load_command<P>* cmd = cmds;
1446 for (uint32_t i = 0; i < cmd_count; ++i) {
1447 if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
1448 const macho_segment_command<P>* segment = (macho_segment_command<P>*)cmd;
1449 const macho_section<P>* sectionsStart = (macho_section<P>*)((char*)segment + sizeof(macho_segment_command<P>));
1450 for (uint32_t si=0; si < segment->nsects(); ++si) {
1451 const macho_section<P>* sect = &sectionsStart[si];
1452 if ( (sect->size() > 0)
1453 && (strcmp(sect->sectname(), "__category") == 0)
1454 && (strcmp(sect->segname(), "__OBJC") == 0) ) {
1455 return true;
1456 }
1457 }
1458 }
1459 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
1460 if ( cmd > cmdsEnd )
1461 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
1462 }
1463 return false;
1464 }
1465
1466
1467 template <typename A>
1468 bool Parser<A>::getNonLocalSymbols(const uint8_t* fileContent, std::vector<const char*> &syms)
1469 {
1470 const macho_header<P>* header = (const macho_header<P>*)fileContent;
1471 const uint32_t cmd_count = header->ncmds();
1472 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1473 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1474 const macho_load_command<P>* cmd = cmds;
1475 for (uint32_t i = 0; i < cmd_count; ++i) {
1476 if ( cmd->cmd() == LC_SYMTAB ) {
1477 const macho_symtab_command<P>* symtab = (macho_symtab_command<P>*)cmd;
1478 uint32_t symbolCount = symtab->nsyms();
1479 const macho_nlist<P>* symbols = (const macho_nlist<P>*)(fileContent + symtab->symoff());
1480 const char* strings = (char*)fileContent + symtab->stroff();
1481 for (uint32_t j = 0; j < symbolCount; ++j) {
1482 // ignore stabs and count only ext symbols
1483 if ( (symbols[j].n_type() & N_STAB) == 0 &&
1484 (symbols[j].n_type() & N_EXT) != 0 ) {
1485 const char* symName = &strings[symbols[j].n_strx()];
1486 syms.push_back(symName);
1487 }
1488 }
1489 return true;
1490 }
1491 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
1492 if ( cmd > cmdsEnd )
1493 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
1494 }
1495 return false;
1496 }
1497
1498
1499 template <typename A>
1500 int Parser<A>::pointerSorter(const void* l, const void* r)
1501 {
1502 // sort references by address
1503 const pint_t* left = (pint_t*)l;
1504 const pint_t* right = (pint_t*)r;
1505 return (*left - *right);
1506 }
1507
1508 template <typename A>
1509 typename A::P::uint_t Parser<A>::LabelAndCFIBreakIterator::peek(Parser<A>& parser, pint_t startAddr, pint_t endAddr)
1510 {
1511 pint_t symbolAddr;
1512 if ( symIndex < sortedSymbolCount )
1513 symbolAddr = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]).n_value();
1514 else
1515 symbolAddr = endAddr;
1516 pint_t cfiAddr;
1517 if ( cfiIndex < cfiStartsCount )
1518 cfiAddr = cfiStartsArray[cfiIndex];
1519 else
1520 cfiAddr = endAddr;
1521 if ( (cfiAddr < symbolAddr) && (cfiAddr >= startAddr) ) {
1522 if ( cfiAddr < endAddr )
1523 return cfiAddr;
1524 else
1525 return endAddr;
1526 }
1527 else {
1528 if ( symbolAddr < endAddr )
1529 return symbolAddr;
1530 else
1531 return endAddr;
1532 }
1533 }
1534
1535 //
1536 // Parses up a section into chunks based on labels and CFI information.
1537 // Each call returns the next chunk address and size, and (if the break
1538 // was becuase of a label, the symbol). Returns false when no more chunks.
1539 //
1540 template <typename A>
1541 bool Parser<A>::LabelAndCFIBreakIterator::next(Parser<A>& parser, const Section<A>& sect, uint32_t sectNum, pint_t startAddr, pint_t endAddr,
1542 pint_t* addr, pint_t* size, const macho_nlist<P>** symbol)
1543 {
1544 // may not be a label on start of section, but need atom demarcation there
1545 if ( newSection ) {
1546 newSection = false;
1547 // advance symIndex until we get to the first label at or past the start of this section
1548 while ( symIndex < sortedSymbolCount ) {
1549 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1550 if ( ! sect.ignoreLabel(parser.nameFromSymbol(sym)) ) {
1551 pint_t nextSymbolAddr = sym.n_value();
1552 //fprintf(stderr, "sectNum=%d, nextSymbolAddr=0x%08llX, name=%s\n", sectNum, (uint64_t)nextSymbolAddr, parser.nameFromSymbol(sym));
1553 if ( (nextSymbolAddr > startAddr) || ((nextSymbolAddr == startAddr) && (sym.n_sect() == sectNum)) )
1554 break;
1555 }
1556 ++symIndex;
1557 }
1558 if ( symIndex < sortedSymbolCount ) {
1559 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1560 pint_t nextSymbolAddr = sym.n_value();
1561 // if next symbol found is not in this section
1562 if ( sym.n_sect() != sectNum ) {
1563 // check for CFI break instead of symbol break
1564 if ( cfiIndex < cfiStartsCount ) {
1565 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1566 if ( nextCfiAddr < endAddr ) {
1567 // use cfi
1568 ++cfiIndex;
1569 *addr = nextCfiAddr;
1570 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1571 *symbol = NULL;
1572 return true;
1573 }
1574 }
1575 *addr = startAddr;
1576 *size = endAddr - startAddr;
1577 *symbol = NULL;
1578 if ( startAddr == endAddr )
1579 return false; // zero size section
1580 else
1581 return true; // whole section is one atom with no label
1582 }
1583 // if also CFI break here, eat it
1584 if ( cfiIndex < cfiStartsCount ) {
1585 if ( cfiStartsArray[cfiIndex] == nextSymbolAddr )
1586 ++cfiIndex;
1587 }
1588 if ( nextSymbolAddr == startAddr ) {
1589 // label at start of section, return it as chunk
1590 ++symIndex;
1591 *addr = startAddr;
1592 *size = peek(parser, startAddr, endAddr) - startAddr;
1593 *symbol = &sym;
1594 return true;
1595 }
1596 // return chunk before first symbol
1597 *addr = startAddr;
1598 *size = nextSymbolAddr - startAddr;
1599 *symbol = NULL;
1600 return true;
1601 }
1602 // no symbols in section, check CFI
1603 if ( cfiIndex < cfiStartsCount ) {
1604 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1605 if ( nextCfiAddr < endAddr ) {
1606 // use cfi
1607 ++cfiIndex;
1608 *addr = nextCfiAddr;
1609 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1610 *symbol = NULL;
1611 return true;
1612 }
1613 }
1614 // no cfi, so whole section is one chunk
1615 *addr = startAddr;
1616 *size = endAddr - startAddr;
1617 *symbol = NULL;
1618 if ( startAddr == endAddr )
1619 return false; // zero size section
1620 else
1621 return true; // whole section is one atom with no label
1622 }
1623
1624 while ( (symIndex < sortedSymbolCount) && (cfiIndex < cfiStartsCount) ) {
1625 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1626 pint_t nextSymbolAddr = sym.n_value();
1627 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1628 if ( nextSymbolAddr < nextCfiAddr ) {
1629 if ( nextSymbolAddr >= endAddr )
1630 return false;
1631 ++symIndex;
1632 if ( nextSymbolAddr < startAddr )
1633 continue;
1634 *addr = nextSymbolAddr;
1635 *size = peek(parser, startAddr, endAddr) - nextSymbolAddr;
1636 *symbol = &sym;
1637 return true;
1638 }
1639 else if ( nextCfiAddr < nextSymbolAddr ) {
1640 if ( nextCfiAddr >= endAddr )
1641 return false;
1642 ++cfiIndex;
1643 if ( nextCfiAddr < startAddr )
1644 continue;
1645 *addr = nextCfiAddr;
1646 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1647 *symbol = NULL;
1648 return true;
1649 }
1650 else {
1651 if ( nextCfiAddr >= endAddr )
1652 return false;
1653 ++symIndex;
1654 ++cfiIndex;
1655 if ( nextCfiAddr < startAddr )
1656 continue;
1657 *addr = nextCfiAddr;
1658 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1659 *symbol = &sym;
1660 return true;
1661 }
1662 }
1663 while ( symIndex < sortedSymbolCount ) {
1664 const macho_nlist<P>& sym = parser.symbolFromIndex(sortedSymbolIndexes[symIndex]);
1665 pint_t nextSymbolAddr = sym.n_value();
1666 // if next symbol found is not in this section, then done with iteration
1667 if ( sym.n_sect() != sectNum )
1668 return false;
1669 ++symIndex;
1670 if ( nextSymbolAddr < startAddr )
1671 continue;
1672 *addr = nextSymbolAddr;
1673 *size = peek(parser, startAddr, endAddr) - nextSymbolAddr;
1674 *symbol = &sym;
1675 return true;
1676 }
1677 while ( cfiIndex < cfiStartsCount ) {
1678 pint_t nextCfiAddr = cfiStartsArray[cfiIndex];
1679 if ( nextCfiAddr >= endAddr )
1680 return false;
1681 ++cfiIndex;
1682 if ( nextCfiAddr < startAddr )
1683 continue;
1684 *addr = nextCfiAddr;
1685 *size = peek(parser, startAddr, endAddr) - nextCfiAddr;
1686 *symbol = NULL;
1687 return true;
1688 }
1689 return false;
1690 }
1691
1692 template <>
1693 typename arm::P::uint_t Parser<arm>::realAddr(typename arm::P::uint_t addr)
1694 {
1695 return addr & (-2);
1696 }
1697
1698 template <typename A>
1699 typename A::P::uint_t Parser<A>::realAddr(typename A::P::uint_t addr)
1700 {
1701 return addr;
1702 }
1703
1704 #define STACK_ALLOC_IF_SMALL(_type, _name, _actual_count, _maxCount) \
1705 _type* _name = NULL; \
1706 uint32_t _name##_count = 1; \
1707 if ( _actual_count > _maxCount ) \
1708 _name = (_type*)malloc(sizeof(_type) * _actual_count); \
1709 else \
1710 _name##_count = _actual_count; \
1711 _type _name##_buffer[_name##_count]; \
1712 if ( _name == NULL ) \
1713 _name = _name##_buffer;
1714
1715
1716 template <typename A>
1717 ld::relocatable::File* Parser<A>::parse(const ParserOptions& opts)
1718 {
1719 // create file object
1720 _file = new File<A>(_path, _modTime, _fileContent, _ordinal);
1721
1722 // set sourceKind
1723 _file->_srcKind = opts.srcKind;
1724 // set treatBitcodeAsData
1725 _treateBitcodeAsData = opts.treateBitcodeAsData;
1726 _usingBitcode = opts.usingBitcode;
1727
1728 // respond to -t option
1729 if ( opts.logAllFiles )
1730 printf("%s\n", _path);
1731
1732 _armUsesZeroCostExceptions = opts.armUsesZeroCostExceptions;
1733 _maxDefaultCommonAlignment = opts.maxDefaultCommonAlignment;
1734
1735 // parse start of mach-o file
1736 if ( ! parseLoadCommands(opts.platform, opts.minOSVersion, opts.simulator, opts.ignoreMismatchPlatform) )
1737 return _file;
1738
1739 // make array of
1740 uint32_t sortedSectionIndexes[_machOSectionsCount];
1741 this->makeSortedSectionsArray(sortedSectionIndexes);
1742
1743 // make symbol table sorted by address
1744 this->prescanSymbolTable();
1745 uint32_t sortedSymbolIndexes[_symbolsInSections];
1746 this->makeSortedSymbolsArray(sortedSymbolIndexes, sortedSectionIndexes);
1747
1748 // allocate Section<A> object for each mach-o section
1749 makeSections();
1750
1751 // if it exists, do special early parsing of __compact_unwind section
1752 uint32_t countOfCUs = 0;
1753 if ( _compactUnwindSection != NULL )
1754 countOfCUs = _compactUnwindSection->count();
1755 // stack allocate (if not too large) cuInfoBuffer
1756 STACK_ALLOC_IF_SMALL(typename CUSection<A>::Info, cuInfoArray, countOfCUs, 1024);
1757 if ( countOfCUs != 0 )
1758 _compactUnwindSection->parse(*this, countOfCUs, cuInfoArray);
1759
1760 // create lists of address that already have compact unwind and thus don't need the dwarf parsed
1761 unsigned cuLsdaCount = 0;
1762 pint_t cuStarts[countOfCUs];
1763 for (uint32_t i=0; i < countOfCUs; ++i) {
1764 if ( CUSection<A>::encodingMeansUseDwarf(cuInfoArray[i].compactUnwindInfo) )
1765 cuStarts[i] = -1;
1766 else
1767 cuStarts[i] = cuInfoArray[i].functionStartAddress;
1768 if ( cuInfoArray[i].lsdaAddress != 0 )
1769 ++cuLsdaCount;
1770 }
1771
1772
1773 // if it exists, do special early parsing of __eh_frame section
1774 // stack allocate (if not too large) array of CFI_Atom_Info
1775 uint32_t countOfCFIs = 0;
1776 if ( _EHFrameSection != NULL )
1777 countOfCFIs = _EHFrameSection->cfiCount(*this);
1778 STACK_ALLOC_IF_SMALL(typename CFISection<A>::CFI_Atom_Info, cfiArray, countOfCFIs, 1024);
1779
1780 // stack allocate (if not too large) a copy of __eh_frame to apply relocations to
1781 uint32_t sectSize = 4;
1782 if ( (countOfCFIs != 0) && _EHFrameSection->needsRelocating() )
1783 sectSize = _EHFrameSection->machoSection()->size()+4;
1784 STACK_ALLOC_IF_SMALL(uint8_t, ehBuffer, sectSize, 50*1024);
1785 uint32_t cfiStartsCount = 0;
1786 if ( countOfCFIs != 0 ) {
1787 _EHFrameSection->cfiParse(*this, ehBuffer, cfiArray, countOfCFIs, cuStarts, countOfCUs);
1788 // count functions and lsdas
1789 for(uint32_t i=0; i < countOfCFIs; ++i) {
1790 if ( cfiArray[i].isCIE )
1791 continue;
1792 //fprintf(stderr, "cfiArray[i].func = 0x%08llX, cfiArray[i].lsda = 0x%08llX, encoding=0x%08X\n",
1793 // (uint64_t)cfiArray[i].u.fdeInfo.function.targetAddress,
1794 // (uint64_t)cfiArray[i].u.fdeInfo.lsda.targetAddress,
1795 // cfiArray[i].u.fdeInfo.compactUnwindInfo);
1796 if ( cfiArray[i].u.fdeInfo.function.targetAddress != CFI_INVALID_ADDRESS )
1797 ++cfiStartsCount;
1798 if ( cfiArray[i].u.fdeInfo.lsda.targetAddress != CFI_INVALID_ADDRESS )
1799 ++cfiStartsCount;
1800 }
1801 }
1802 CFI_CU_InfoArrays cfis(cfiArray, countOfCFIs, cuInfoArray, countOfCUs);
1803
1804 // create sorted array of function starts and lsda starts
1805 pint_t cfiStartsArray[cfiStartsCount+cuLsdaCount];
1806 uint32_t countOfFDEs = 0;
1807 uint32_t cfiStartsArrayCount = 0;
1808 if ( countOfCFIs != 0 ) {
1809 for(uint32_t i=0; i < countOfCFIs; ++i) {
1810 if ( cfiArray[i].isCIE )
1811 continue;
1812 if ( cfiArray[i].u.fdeInfo.function.targetAddress != CFI_INVALID_ADDRESS )
1813 cfiStartsArray[cfiStartsArrayCount++] = realAddr(cfiArray[i].u.fdeInfo.function.targetAddress);
1814 if ( cfiArray[i].u.fdeInfo.lsda.targetAddress != CFI_INVALID_ADDRESS )
1815 cfiStartsArray[cfiStartsArrayCount++] = cfiArray[i].u.fdeInfo.lsda.targetAddress;
1816 ++countOfFDEs;
1817 }
1818 }
1819 if ( cuLsdaCount != 0 ) {
1820 // merge in an lsda info from compact unwind
1821 for (uint32_t i=0; i < countOfCUs; ++i) {
1822 if ( cuInfoArray[i].lsdaAddress == 0 )
1823 continue;
1824 // append to cfiStartsArray if not already in that list
1825 bool found = false;
1826 for(uint32_t j=0; j < cfiStartsArrayCount; ++j) {
1827 if ( cfiStartsArray[j] == cuInfoArray[i].lsdaAddress )
1828 found = true;
1829 }
1830 if ( ! found ) {
1831 cfiStartsArray[cfiStartsArrayCount++] = cuInfoArray[i].lsdaAddress;
1832 }
1833 }
1834 }
1835 if ( cfiStartsArrayCount != 0 ) {
1836 ::qsort(cfiStartsArray, cfiStartsArrayCount, sizeof(pint_t), pointerSorter);
1837 #ifndef NDEBUG
1838 // scan for FDEs claming the same function
1839 for(uint32_t i=1; i < cfiStartsArrayCount; ++i) {
1840 assert( cfiStartsArray[i] != cfiStartsArray[i-1] );
1841 }
1842 #endif
1843 }
1844
1845 Section<A>** sections = _file->_sectionsArray;
1846 uint32_t sectionsCount = _file->_sectionsArrayCount;
1847
1848 // figure out how many atoms will be allocated and allocate
1849 LabelAndCFIBreakIterator breakIterator(sortedSymbolIndexes, _symbolsInSections, cfiStartsArray,
1850 cfiStartsArrayCount, _overlappingSymbols);
1851 uint32_t computedAtomCount = 0;
1852 for (uint32_t i=0; i < sectionsCount; ++i ) {
1853 breakIterator.beginSection();
1854 uint32_t count = sections[i]->computeAtomCount(*this, breakIterator, cfis);
1855 //const macho_section<P>* sect = sections[i]->machoSection();
1856 //fprintf(stderr, "computed count=%u for section %s size=%llu\n", count, sect->sectname(), (sect != NULL) ? sect->size() : 0);
1857 computedAtomCount += count;
1858 }
1859 //fprintf(stderr, "allocating %d atoms * sizeof(Atom<A>)=%ld, sizeof(ld::Atom)=%ld\n", computedAtomCount, sizeof(Atom<A>), sizeof(ld::Atom));
1860 _file->_atomsArray = new uint8_t[computedAtomCount*sizeof(Atom<A>)];
1861 _file->_atomsArrayCount = 0;
1862
1863 // have each section append atoms to _atomsArray
1864 LabelAndCFIBreakIterator breakIterator2(sortedSymbolIndexes, _symbolsInSections, cfiStartsArray,
1865 cfiStartsArrayCount, _overlappingSymbols);
1866 for (uint32_t i=0; i < sectionsCount; ++i ) {
1867 uint8_t* atoms = _file->_atomsArray + _file->_atomsArrayCount*sizeof(Atom<A>);
1868 breakIterator2.beginSection();
1869 uint32_t count = sections[i]->appendAtoms(*this, atoms, breakIterator2, cfis);
1870 //fprintf(stderr, "append count=%u for section %s/%s\n", count, sections[i]->machoSection()->segname(), sections[i]->machoSection()->sectname());
1871 _file->_atomsArrayCount += count;
1872 }
1873 assert( _file->_atomsArrayCount == computedAtomCount && "more atoms allocated than expected");
1874
1875
1876 // have each section add all fix-ups for its atoms
1877 _allFixups.reserve(computedAtomCount*5);
1878 for (uint32_t i=0; i < sectionsCount; ++i )
1879 sections[i]->makeFixups(*this, cfis);
1880
1881 // assign fixups start offset for each atom
1882 uint8_t* p = _file->_atomsArray;
1883 uint32_t fixupOffset = 0;
1884 for(int i=_file->_atomsArrayCount; i > 0; --i) {
1885 Atom<A>* atom = (Atom<A>*)p;
1886 atom->_fixupsStartIndex = fixupOffset;
1887 fixupOffset += atom->_fixupsCount;
1888 atom->_fixupsCount = 0;
1889 p += sizeof(Atom<A>);
1890 }
1891 assert(fixupOffset == _allFixups.size());
1892 _file->_fixups.resize(fixupOffset);
1893
1894 // copy each fixup for each atom
1895 for(typename std::vector<FixupInAtom>::iterator it=_allFixups.begin(); it != _allFixups.end(); ++it) {
1896 uint32_t slot = it->atom->_fixupsStartIndex + it->atom->_fixupsCount;
1897 _file->_fixups[slot] = it->fixup;
1898 it->atom->_fixupsCount++;
1899 }
1900
1901 // done with temp vector
1902 _allFixups.clear();
1903
1904 // add unwind info
1905 _file->_unwindInfos.reserve(countOfFDEs+countOfCUs);
1906 for(uint32_t i=0; i < countOfCFIs; ++i) {
1907 if ( cfiArray[i].isCIE )
1908 continue;
1909 if ( cfiArray[i].u.fdeInfo.function.targetAddress != CFI_INVALID_ADDRESS ) {
1910 ld::Atom::UnwindInfo info;
1911 info.startOffset = 0;
1912 info.unwindInfo = cfiArray[i].u.fdeInfo.compactUnwindInfo;
1913 _file->_unwindInfos.push_back(info);
1914 Atom<A>* func = findAtomByAddress(cfiArray[i].u.fdeInfo.function.targetAddress);
1915 func->setUnwindInfoRange(_file->_unwindInfos.size()-1, 1);
1916 //fprintf(stderr, "cu from dwarf =0x%08X, atom=%s\n", info.unwindInfo, func->name());
1917 }
1918 }
1919 // apply compact infos in __LD,__compact_unwind section to each function
1920 // if function also has dwarf unwind, CU will override it
1921 Atom<A>* lastFunc = NULL;
1922 uint32_t lastEnd = 0;
1923 for(uint32_t i=0; i < countOfCUs; ++i) {
1924 typename CUSection<A>::Info* info = &cuInfoArray[i];
1925 assert(info->function != NULL);
1926 ld::Atom::UnwindInfo ui;
1927 ui.startOffset = info->functionStartAddress - info->function->objectAddress();
1928 ui.unwindInfo = info->compactUnwindInfo;
1929 _file->_unwindInfos.push_back(ui);
1930 // don't override with converted cu with "use dwarf" cu, if forcing dwarf conversion
1931 if ( !_forceDwarfConversion || !CUSection<A>::encodingMeansUseDwarf(info->compactUnwindInfo) ) {
1932 //fprintf(stderr, "cu=0x%08X, atom=%s\n", ui.unwindInfo, info->function->name());
1933 // if previous is for same function, extend range
1934 if ( info->function == lastFunc ) {
1935 if ( lastEnd != ui.startOffset ) {
1936 if ( lastEnd < ui.startOffset )
1937 warning("__LD,__compact_unwind entries for %s have a gap at offset 0x%0X", info->function->name(), lastEnd);
1938 else
1939 warning("__LD,__compact_unwind entries for %s overlap at offset 0x%0X", info->function->name(), lastEnd);
1940 }
1941 lastFunc->extendUnwindInfoRange();
1942 }
1943 else
1944 info->function->setUnwindInfoRange(_file->_unwindInfos.size()-1, 1);
1945 lastFunc = info->function;
1946 lastEnd = ui.startOffset + info->rangeLength;
1947 }
1948 }
1949
1950 // process indirect symbols which become AliasAtoms
1951 _file->_aliasAtomsArray = NULL;
1952 _file->_aliasAtomsArrayCount = 0;
1953 if ( _indirectSymbolCount != 0 ) {
1954 _file->_aliasAtomsArrayCount = _indirectSymbolCount;
1955 _file->_aliasAtomsArray = new uint8_t[_file->_aliasAtomsArrayCount*sizeof(AliasAtom)];
1956 this->appendAliasAtoms(_file->_aliasAtomsArray);
1957 }
1958
1959
1960 // parse dwarf debug info to get line info
1961 this->parseDebugInfo();
1962
1963 return _file;
1964 }
1965
1966 static void versionToString(uint32_t value, char buffer[32])
1967 {
1968 if ( value & 0xFF )
1969 sprintf(buffer, "%d.%d.%d", value >> 16, (value >> 8) & 0xFF, value & 0xFF);
1970 else
1971 sprintf(buffer, "%d.%d", value >> 16, (value >> 8) & 0xFF);
1972 }
1973
1974 template <> uint8_t Parser<x86>::loadCommandSizeMask() { return 0x03; }
1975 template <> uint8_t Parser<x86_64>::loadCommandSizeMask() { return 0x07; }
1976 template <> uint8_t Parser<arm>::loadCommandSizeMask() { return 0x03; }
1977 template <> uint8_t Parser<arm64>::loadCommandSizeMask() { return 0x07; }
1978
1979 template <typename A>
1980 bool Parser<A>::parseLoadCommands(Options::Platform platform, uint32_t linkMinOSVersion, bool simulator, bool ignoreMismatchPlatform)
1981 {
1982 const macho_header<P>* header = (const macho_header<P>*)_fileContent;
1983
1984 // set File attributes
1985 _file->_canScatterAtoms = (header->flags() & MH_SUBSECTIONS_VIA_SYMBOLS);
1986 _file->_cpuSubType = header->cpusubtype();
1987
1988 const macho_segment_command<P>* segment = NULL;
1989 const uint8_t* const endOfFile = _fileContent + _fileLength;
1990 const uint32_t cmd_count = header->ncmds();
1991 // <rdar://problem/5394172> an empty .o file with zero load commands will crash linker
1992 if ( cmd_count == 0 )
1993 return false;
1994 Options::Platform lcPlatform = Options::kPlatformUnknown;
1995 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
1996 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
1997 const macho_load_command<P>* cmd = cmds;
1998 for (uint32_t i = 0; i < cmd_count; ++i) {
1999 uint32_t size = cmd->cmdsize();
2000 if ( (size & this->loadCommandSizeMask()) != 0 )
2001 throwf("load command #%d has a unaligned size", i);
2002 const uint8_t* endOfCmd = ((uint8_t*)cmd)+cmd->cmdsize();
2003 if ( endOfCmd > (uint8_t*)cmdsEnd )
2004 throwf("load command #%d extends beyond the end of the load commands", i);
2005 if ( endOfCmd > endOfFile )
2006 throwf("load command #%d extends beyond the end of the file", i);
2007 switch (cmd->cmd()) {
2008 case LC_SYMTAB:
2009 {
2010 const macho_symtab_command<P>* symtab = (macho_symtab_command<P>*)cmd;
2011 _symbolCount = symtab->nsyms();
2012 _symbols = (const macho_nlist<P>*)(_fileContent + symtab->symoff());
2013 _strings = (char*)_fileContent + symtab->stroff();
2014 _stringsSize = symtab->strsize();
2015 if ( (symtab->symoff() + _symbolCount*sizeof(macho_nlist<P>)) > _fileLength )
2016 throw "mach-o symbol table extends beyond end of file";
2017 if ( (_strings + _stringsSize) > (char*)endOfFile )
2018 throw "mach-o string pool extends beyond end of file";
2019 if ( _indirectTable == NULL ) {
2020 if ( _undefinedEndIndex == 0 ) {
2021 _undefinedStartIndex = 0;
2022 _undefinedEndIndex = symtab->nsyms();
2023 }
2024 }
2025 }
2026 break;
2027 case LC_DYSYMTAB:
2028 {
2029 const macho_dysymtab_command<P>* dsymtab = (macho_dysymtab_command<P>*)cmd;
2030 _indirectTable = (uint32_t*)(_fileContent + dsymtab->indirectsymoff());
2031 _indirectTableCount = dsymtab->nindirectsyms();
2032 if ( &_indirectTable[_indirectTableCount] > (uint32_t*)endOfFile )
2033 throw "indirect symbol table extends beyond end of file";
2034 _undefinedStartIndex = dsymtab->iundefsym();
2035 _undefinedEndIndex = _undefinedStartIndex + dsymtab->nundefsym();
2036 }
2037 break;
2038 case LC_UUID:
2039 _hasUUID = true;
2040 break;
2041 case LC_DATA_IN_CODE:
2042 {
2043 const macho_linkedit_data_command<P>* dc = (macho_linkedit_data_command<P>*)cmd;
2044 _dataInCodeStart = (macho_data_in_code_entry<P>*)(_fileContent + dc->dataoff());
2045 _dataInCodeEnd = (macho_data_in_code_entry<P>*)(_fileContent + dc->dataoff() + dc->datasize());
2046 if ( _dataInCodeEnd > (macho_data_in_code_entry<P>*)endOfFile )
2047 throw "LC_DATA_IN_CODE table extends beyond end of file";
2048 }
2049 break;
2050 case LC_LINKER_OPTION:
2051 {
2052 const macho_linker_option_command<P>* loc = (macho_linker_option_command<P>*)cmd;
2053 const char* buffer = loc->buffer();
2054 _file->_linkerOptions.resize(_file->_linkerOptions.size() + 1);
2055 std::vector<const char*>& vec = _file->_linkerOptions.back();
2056 for (uint32_t j=0; j < loc->count(); ++j) {
2057 vec.push_back(buffer);
2058 buffer += strlen(buffer) + 1;
2059 }
2060 if ( buffer > ((char*)cmd + loc->cmdsize()) )
2061 throw "malformed LC_LINKER_OPTION";
2062 }
2063 break;
2064 case LC_LINKER_OPTIMIZATION_HINTS:
2065 {
2066 const macho_linkedit_data_command<P>* loh = (macho_linkedit_data_command<P>*)cmd;
2067 _lohStart = _fileContent + loh->dataoff();
2068 _lohEnd = _fileContent + loh->dataoff() + loh->datasize();
2069 if ( _lohEnd > endOfFile )
2070 throw "LC_LINKER_OPTIMIZATION_HINTS table extends beyond end of file";
2071 }
2072 break;
2073 case LC_VERSION_MIN_MACOSX:
2074 case LC_VERSION_MIN_IPHONEOS:
2075 case LC_VERSION_MIN_WATCHOS:
2076 #if SUPPORT_APPLE_TV
2077 case LC_VERSION_MIN_TVOS:
2078 #endif
2079 if ( ignoreMismatchPlatform )
2080 break;
2081 _file->_platform = cmd->cmd();
2082 lcPlatform = Options::platformForLoadCommand(cmd->cmd());
2083 _file->_minOSVersion = ((macho_version_min_command<P>*)cmd)->version();
2084 break;
2085 case macho_segment_command<P>::CMD:
2086 if ( segment != NULL )
2087 throw "more than one LC_SEGMENT found in object file";
2088 segment = (macho_segment_command<P>*)cmd;
2089 break;
2090 default:
2091 // ignore unknown load commands
2092 break;
2093 }
2094 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
2095 if ( cmd > cmdsEnd )
2096 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
2097 }
2098 // arm/arm64 objects are default to ios platform if not set.
2099 // rdar://problem/21746314
2100 if (lcPlatform == Options::kPlatformUnknown &&
2101 (std::is_same<A, arm>::value || std::is_same<A, arm64>::value))
2102 lcPlatform = Options::kPlatformiOS;
2103
2104 // Check platform cross-linking.
2105 if ( !ignoreMismatchPlatform ) {
2106 if ( lcPlatform != platform ) {
2107 switch (platform) {
2108 case Options::kPlatformOSX:
2109 case Options::kPlatformiOS:
2110 if ( lcPlatform == Options::kPlatformUnknown )
2111 break;
2112 // fall through if the Platform is not Unknown
2113 case Options::kPlatformWatchOS:
2114 // Error when using bitcocde, warning otherwise.
2115 if (_usingBitcode)
2116 throwf("building for %s%s, but linking in object file built for %s,",
2117 Options::platformName(platform), (simulator ? " simulator" : ""),
2118 Options::platformName(lcPlatform));
2119 else
2120 warning("URGENT: building for %s%s, but linking in object file (%s) built for %s. "
2121 "Note: This will be an error in the future.",
2122 Options::platformName(platform), (simulator ? " simulator" : ""), path(),
2123 Options::platformName(lcPlatform));
2124 break;
2125 #if SUPPORT_APPLE_TV
2126 case Options::kPlatform_tvOS:
2127 // Error when using bitcocde, warning otherwise.
2128 if (_usingBitcode)
2129 throwf("building for %s%s, but linking in object file built for %s,",
2130 Options::platformName(platform), (simulator ? " simulator" : ""),
2131 Options::platformName(lcPlatform));
2132 else
2133 warning("URGENT: building for %s%s, but linking in object file (%s) built for %s. "
2134 "Note: This will be an error in the future.",
2135 Options::platformName(platform), (simulator ? " simulator" : ""), path(),
2136 Options::platformName(lcPlatform));
2137 break;
2138 #endif
2139 case Options::kPlatformUnknown:
2140 // skip if the target platform is unknown
2141 break;
2142 }
2143 }
2144 if ( linkMinOSVersion && (_file->_minOSVersion > linkMinOSVersion) ) {
2145 char t1[32];
2146 char t2[32];
2147 versionToString(_file->_minOSVersion, t1);
2148 versionToString(linkMinOSVersion, t2);
2149 warning("object file (%s) was built for newer %s version (%s) than being linked (%s)",
2150 _path, Options::platformName(lcPlatform), t1, t2);
2151 }
2152 }
2153
2154
2155 // record range of sections
2156 if ( segment == NULL )
2157 throw "missing LC_SEGMENT";
2158 _sectionsStart = (macho_section<P>*)((char*)segment + sizeof(macho_segment_command<P>));
2159 _machOSectionsCount = segment->nsects();
2160 if ( (sizeof(macho_segment_command<P>) + _machOSectionsCount * sizeof(macho_section<P>)) > segment->cmdsize() )
2161 throw "too many sections for size of LC_SEGMENT command";
2162 return true;
2163 }
2164
2165 template <typename A>
2166 Options::Platform Parser<A>::findPlatform(const macho_header<P>* header)
2167 {
2168 const uint32_t cmd_count = header->ncmds();
2169 if ( cmd_count == 0 )
2170 return Options::kPlatformUnknown;
2171 const macho_load_command<P>* const cmds = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>));
2172 const macho_load_command<P>* const cmdsEnd = (macho_load_command<P>*)((char*)header + sizeof(macho_header<P>) + header->sizeofcmds());
2173 const macho_load_command<P>* cmd = cmds;
2174 for (uint32_t i = 0; i < cmd_count; ++i) {
2175 uint32_t size = cmd->cmdsize();
2176 if ( (size & loadCommandSizeMask()) != 0 )
2177 throwf("load command #%d has a unaligned size", i);
2178 const uint8_t* endOfCmd = ((uint8_t*)cmd)+cmd->cmdsize();
2179 if ( endOfCmd > (uint8_t*)cmdsEnd )
2180 throwf("load command #%d extends beyond the end of the load commands", i);
2181 switch (cmd->cmd()) {
2182 case LC_VERSION_MIN_MACOSX:
2183 return Options::kPlatformOSX;
2184 case LC_VERSION_MIN_IPHONEOS:
2185 return Options::kPlatformiOS;
2186 }
2187 cmd = (const macho_load_command<P>*)(((char*)cmd)+cmd->cmdsize());
2188 if ( cmd > cmdsEnd )
2189 throwf("malformed mach-o file, load command #%d is outside size of load commands", i);
2190 }
2191 return Options::kPlatformUnknown;
2192 }
2193
2194
2195 template <typename A>
2196 void Parser<A>::prescanSymbolTable()
2197 {
2198 _tentativeDefinitionCount = 0;
2199 _absoluteSymbolCount = 0;
2200 _symbolsInSections = 0;
2201 _hasDataInCodeLabels = false;
2202 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2203 const macho_nlist<P>& sym = symbolFromIndex(i);
2204 // ignore stabs
2205 if ( (sym.n_type() & N_STAB) != 0 )
2206 continue;
2207
2208 // look at undefines
2209 const char* symbolName = this->nameFromSymbol(sym);
2210 if ( (sym.n_type() & N_TYPE) == N_UNDF ) {
2211 if ( sym.n_value() != 0 ) {
2212 // count tentative definitions
2213 ++_tentativeDefinitionCount;
2214 }
2215 else if ( strncmp(symbolName, "___dtrace_", 10) == 0 ) {
2216 // any undefined starting with __dtrace_*$ that is not ___dtrace_probe$* or ___dtrace_isenabled$*
2217 // is extra provider info
2218 if ( (strncmp(&symbolName[10], "probe$", 6) != 0) && (strncmp(&symbolName[10], "isenabled$", 10) != 0) ) {
2219 _dtraceProviderInfo.push_back(symbolName);
2220 }
2221 }
2222 continue;
2223 }
2224 else if ( ((sym.n_type() & N_TYPE) == N_INDR) && ((sym.n_type() & N_EXT) != 0) ) {
2225 _indirectSymbolCount++;
2226 continue;
2227 }
2228
2229 // count absolute symbols
2230 if ( (sym.n_type() & N_TYPE) == N_ABS ) {
2231 const char* absName = this->nameFromSymbol(sym);
2232 // ignore .objc_class_name_* symbols
2233 if ( strncmp(absName, ".objc_class_name_", 17) == 0 ) {
2234 _AppleObjc = true;
2235 continue;
2236 }
2237 // ignore .objc_class_name_* symbols
2238 if ( strncmp(absName, ".objc_category_name_", 20) == 0 )
2239 continue;
2240 // ignore empty *.eh symbols
2241 if ( strcmp(&absName[strlen(absName)-3], ".eh") == 0 )
2242 continue;
2243 ++_absoluteSymbolCount;
2244 }
2245
2246 // only look at definitions
2247 if ( (sym.n_type() & N_TYPE) != N_SECT )
2248 continue;
2249
2250 // 'L' labels do not denote atom breaks
2251 if ( symbolName[0] == 'L' ) {
2252 // <rdar://problem/9218847> Formalize data in code with L$start$ labels
2253 if ( strncmp(symbolName, "L$start$", 8) == 0 )
2254 _hasDataInCodeLabels = true;
2255 continue;
2256 }
2257 // how many def syms in each section
2258 if ( sym.n_sect() > _machOSectionsCount )
2259 throw "bad n_sect in symbol table";
2260
2261 _symbolsInSections++;
2262 }
2263 }
2264
2265 template <typename A>
2266 void Parser<A>::appendAliasAtoms(uint8_t* p)
2267 {
2268 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2269 const macho_nlist<P>& sym = symbolFromIndex(i);
2270 // ignore stabs
2271 if ( (sym.n_type() & N_STAB) != 0 )
2272 continue;
2273
2274 // only look at N_INDR symbols
2275 if ( (sym.n_type() & N_TYPE) != N_INDR )
2276 continue;
2277
2278 // skip non-external aliases
2279 if ( (sym.n_type() & N_EXT) == 0 )
2280 continue;
2281
2282 const char* symbolName = this->nameFromSymbol(sym);
2283 const char* aliasOfName = &_strings[sym.n_value()];
2284 bool isHiddenVisibility = (sym.n_type() & N_PEXT);
2285 AliasAtom* allocatedSpace = (AliasAtom*)p;
2286 new (allocatedSpace) AliasAtom(symbolName, isHiddenVisibility, _file, aliasOfName);
2287 p += sizeof(AliasAtom);
2288 }
2289 }
2290
2291
2292
2293 template <typename A>
2294 int Parser<A>::sectionIndexSorter(void* extra, const void* l, const void* r)
2295 {
2296 Parser<A>* parser = (Parser<A>*)extra;
2297 const uint32_t* left = (uint32_t*)l;
2298 const uint32_t* right = (uint32_t*)r;
2299 const macho_section<P>* leftSect = parser->machOSectionFromSectionIndex(*left);
2300 const macho_section<P>* rightSect = parser->machOSectionFromSectionIndex(*right);
2301
2302 // can't just return difference because 64-bit diff does not fit in 32-bit return type
2303 int64_t result = leftSect->addr() - rightSect->addr();
2304 if ( result == 0 ) {
2305 // two sections with same start address
2306 // one with zero size goes first
2307 bool leftEmpty = ( leftSect->size() == 0 );
2308 bool rightEmpty = ( rightSect->size() == 0 );
2309 if ( leftEmpty != rightEmpty ) {
2310 return ( rightEmpty ? 1 : -1 );
2311 }
2312 if ( !leftEmpty && !rightEmpty )
2313 throwf("overlapping sections");
2314 // both empty, so chose file order
2315 return ( rightSect - leftSect );
2316 }
2317 else if ( result < 0 )
2318 return -1;
2319 else
2320 return 1;
2321 }
2322
2323 template <typename A>
2324 void Parser<A>::makeSortedSectionsArray(uint32_t array[])
2325 {
2326 const bool log = false;
2327
2328 if ( log ) {
2329 fprintf(stderr, "unsorted sections:\n");
2330 for(unsigned int i=0; i < _machOSectionsCount; ++i )
2331 fprintf(stderr, "0x%08llX %s %s\n", _sectionsStart[i].addr(), _sectionsStart[i].segname(), _sectionsStart[i].sectname());
2332 }
2333
2334 // sort by symbol table address
2335 for (uint32_t i=0; i < _machOSectionsCount; ++i)
2336 array[i] = i;
2337 ::qsort_r(array, _machOSectionsCount, sizeof(uint32_t), this, &sectionIndexSorter);
2338
2339 if ( log ) {
2340 fprintf(stderr, "sorted sections:\n");
2341 for(unsigned int i=0; i < _machOSectionsCount; ++i )
2342 fprintf(stderr, "0x%08llX %s %s\n", _sectionsStart[array[i]].addr(), _sectionsStart[array[i]].segname(), _sectionsStart[array[i]].sectname());
2343 }
2344 }
2345
2346
2347
2348 template <typename A>
2349 int Parser<A>::symbolIndexSorter(void* extra, const void* l, const void* r)
2350 {
2351 ParserAndSectionsArray* extraInfo = (ParserAndSectionsArray*)extra;
2352 Parser<A>* parser = extraInfo->parser;
2353 const uint32_t* sortedSectionsArray = extraInfo->sortedSectionsArray;
2354 const uint32_t* left = (uint32_t*)l;
2355 const uint32_t* right = (uint32_t*)r;
2356 const macho_nlist<P>& leftSym = parser->symbolFromIndex(*left);
2357 const macho_nlist<P>& rightSym = parser->symbolFromIndex(*right);
2358 // can't just return difference because 64-bit diff does not fit in 32-bit return type
2359 int64_t result = leftSym.n_value() - rightSym.n_value();
2360 if ( result == 0 ) {
2361 // two symbols with same address
2362 // if in different sections, sort earlier section first
2363 if ( leftSym.n_sect() != rightSym.n_sect() ) {
2364 for (uint32_t i=0; i < parser->machOSectionCount(); ++i) {
2365 if ( sortedSectionsArray[i]+1 == leftSym.n_sect() )
2366 return -1;
2367 if ( sortedSectionsArray[i]+1 == rightSym.n_sect() )
2368 return 1;
2369 }
2370 }
2371 // two symbols in same section, means one is an alias
2372 // if one is ltmp*, make it an alias (sort first)
2373 const char* leftName = parser->nameFromSymbol(leftSym);
2374 const char* rightName = parser->nameFromSymbol(rightSym);
2375 bool leftIsTmp = strncmp(leftName, "ltmp", 4);
2376 bool rightIsTmp = strncmp(rightName, "ltmp", 4);
2377 if ( leftIsTmp != rightIsTmp ) {
2378 return (rightIsTmp ? -1 : 1);
2379 }
2380
2381 // if only one is global, make the other an alias (sort first)
2382 if ( (leftSym.n_type() & N_EXT) != (rightSym.n_type() & N_EXT) ) {
2383 if ( (rightSym.n_type() & N_EXT) != 0 )
2384 return -1;
2385 else
2386 return 1;
2387 }
2388 // if both are global, sort alphabetically. earlier one will be the alias
2389 return ( strcmp(rightName, leftName) );
2390 }
2391 else if ( result < 0 )
2392 return -1;
2393 else
2394 return 1;
2395 }
2396
2397
2398 template <typename A>
2399 void Parser<A>::makeSortedSymbolsArray(uint32_t array[], const uint32_t sectionArray[])
2400 {
2401 const bool log = false;
2402
2403 uint32_t* p = array;
2404 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2405 const macho_nlist<P>& sym = symbolFromIndex(i);
2406 // ignore stabs
2407 if ( (sym.n_type() & N_STAB) != 0 )
2408 continue;
2409
2410 // only look at definitions
2411 if ( (sym.n_type() & N_TYPE) != N_SECT )
2412 continue;
2413
2414 // 'L' labels do not denote atom breaks
2415 const char* symbolName = this->nameFromSymbol(sym);
2416 if ( symbolName[0] == 'L' )
2417 continue;
2418
2419 // how many def syms in each section
2420 if ( sym.n_sect() > _machOSectionsCount )
2421 throw "bad n_sect in symbol table";
2422
2423 // append to array
2424 *p++ = i;
2425 }
2426 assert(p == &array[_symbolsInSections] && "second pass over symbol table yield a different number of symbols");
2427
2428 // sort by symbol table address
2429 ParserAndSectionsArray extra = { this, sectionArray };
2430 ::qsort_r(array, _symbolsInSections, sizeof(uint32_t), &extra, &symbolIndexSorter);
2431
2432
2433 // look for two symbols at same address
2434 _overlappingSymbols = false;
2435 for (unsigned int i=1; i < _symbolsInSections; ++i) {
2436 if ( symbolFromIndex(array[i-1]).n_value() == symbolFromIndex(array[i]).n_value() ) {
2437 //fprintf(stderr, "overlapping symbols at 0x%08llX\n", symbolFromIndex(array[i-1]).n_value());
2438 _overlappingSymbols = true;
2439 break;
2440 }
2441 }
2442
2443 if ( log ) {
2444 fprintf(stderr, "sorted symbols:\n");
2445 for(unsigned int i=0; i < _symbolsInSections; ++i )
2446 fprintf(stderr, "0x%09llX symIndex=%d sectNum=%2d, %s\n", symbolFromIndex(array[i]).n_value(), array[i], symbolFromIndex(array[i]).n_sect(), nameFromSymbol(symbolFromIndex(array[i])) );
2447 }
2448 }
2449
2450 template <typename A>
2451 void Parser<A>::makeSections()
2452 {
2453 // classify each section by type
2454 // compute how many Section objects will be needed and total size for all
2455 unsigned int totalSectionsSize = 0;
2456 uint8_t machOSectsStorage[sizeof(MachOSectionAndSectionClass<P>)*(_machOSectionsCount+2)]; // also room for tentative-defs and absolute symbols
2457 // allocate raw storage for all section objects on stack
2458 MachOSectionAndSectionClass<P>* machOSects = (MachOSectionAndSectionClass<P>*)machOSectsStorage;
2459 unsigned int count = 0;
2460 // local variable for bitcode parsing
2461 const macho_section<P>* bitcodeSect = NULL;
2462 const macho_section<P>* cmdlineSect = NULL;
2463 const macho_section<P>* swiftCmdlineSect = NULL;
2464 const macho_section<P>* bundleSect = NULL;
2465 bool bitcodeAsm = false;
2466
2467 for (uint32_t i=0; i < _machOSectionsCount; ++i) {
2468 const macho_section<P>* sect = &_sectionsStart[i];
2469 if ( (sect->offset() + sect->size() > _fileLength) && ((sect->flags() & SECTION_TYPE) != S_ZEROFILL) )
2470 throwf("section %s/%s extends beyond end of file,", sect->segname(), sect->sectname());
2471
2472 if ( (sect->flags() & S_ATTR_DEBUG) != 0 ) {
2473 if ( strcmp(sect->segname(), "__DWARF") == 0 ) {
2474 // note that .o file has dwarf
2475 _file->_debugInfoKind = ld::relocatable::File::kDebugInfoDwarf;
2476 // save off iteresting dwarf sections
2477 if ( strcmp(sect->sectname(), "__debug_info") == 0 )
2478 _file->_dwarfDebugInfoSect = sect;
2479 else if ( strcmp(sect->sectname(), "__debug_abbrev") == 0 )
2480 _file->_dwarfDebugAbbrevSect = sect;
2481 else if ( strcmp(sect->sectname(), "__debug_line") == 0 )
2482 _file->_dwarfDebugLineSect = sect;
2483 else if ( strcmp(sect->sectname(), "__debug_str") == 0 )
2484 _file->_dwarfDebugStringSect = sect;
2485 // linker does not propagate dwarf sections to output file
2486 continue;
2487 }
2488 else if ( strcmp(sect->segname(), "__LD") == 0 ) {
2489 if ( strncmp(sect->sectname(), "__compact_unwind", 16) == 0 ) {
2490 machOSects[count].sect = sect;
2491 totalSectionsSize += sizeof(CUSection<A>);
2492 machOSects[count++].type = sectionTypeCompactUnwind;
2493 continue;
2494 }
2495 }
2496 }
2497 if ( strcmp(sect->segname(), "__LLVM") == 0 ) {
2498 // Process bitcode segement
2499 if ( strncmp(sect->sectname(), "__bitcode", 9) == 0 ) {
2500 bitcodeSect = sect;
2501 } else if ( strncmp(sect->sectname(), "__cmdline", 9) == 0 ) {
2502 cmdlineSect = sect;
2503 } else if ( strncmp(sect->sectname(), "__swift_cmdline", 15) == 0 ) {
2504 swiftCmdlineSect = sect;
2505 } else if ( strncmp(sect->sectname(), "__bundle", 8) == 0 ) {
2506 bundleSect = sect;
2507 } else if ( strncmp(sect->sectname(), "__asm", 5) == 0 ) {
2508 bitcodeAsm = true;
2509 }
2510 // If treat the bitcode as data, continue to parse as a normal section.
2511 if ( !_treateBitcodeAsData )
2512 continue;
2513 }
2514 // ignore empty __OBJC sections
2515 if ( (sect->size() == 0) && (strcmp(sect->segname(), "__OBJC") == 0) )
2516 continue;
2517 // objc image info section is really attributes and not content
2518 if ( ((strcmp(sect->sectname(), "__image_info") == 0) && (strcmp(sect->segname(), "__OBJC") == 0))
2519 || ((strncmp(sect->sectname(), "__objc_imageinfo", 16) == 0) && (strcmp(sect->segname(), "__DATA") == 0)) ) {
2520 // struct objc_image_info {
2521 // uint32_t version; // initially 0
2522 // uint32_t flags;
2523 // };
2524 // #define OBJC_IMAGE_SUPPORTS_GC 2
2525 // #define OBJC_IMAGE_GC_ONLY 4
2526 // #define OBJC_IMAGE_IS_SIMULATED 32
2527 //
2528 const uint32_t* contents = (uint32_t*)(_file->fileContent()+sect->offset());
2529 if ( (sect->size() >= 8) && (contents[0] == 0) ) {
2530 uint32_t flags = E::get32(contents[1]);
2531 if ( (flags & 4) == 4 )
2532 _file->_objConstraint = ld::File::objcConstraintGC;
2533 else if ( (flags & 2) == 2 )
2534 _file->_objConstraint = ld::File::objcConstraintRetainReleaseOrGC;
2535 else if ( (flags & 32) == 32 )
2536 _file->_objConstraint = ld::File::objcConstraintRetainReleaseForSimulator;
2537 else
2538 _file->_objConstraint = ld::File::objcConstraintRetainRelease;
2539 _file->_swiftVersion = ((flags >> 8) & 0xFF);
2540 if ( sect->size() > 8 ) {
2541 warning("section %s/%s has unexpectedly large size %llu in %s",
2542 sect->segname(), Section<A>::makeSectionName(sect), sect->size(), _file->path());
2543 }
2544 }
2545 else {
2546 warning("can't parse %s/%s section in %s", sect->segname(), Section<A>::makeSectionName(sect), _file->path());
2547 }
2548 continue;
2549 }
2550 machOSects[count].sect = sect;
2551 switch ( sect->flags() & SECTION_TYPE ) {
2552 case S_SYMBOL_STUBS:
2553 if ( _stubsSectionNum == 0 ) {
2554 _stubsSectionNum = i+1;
2555 _stubsMachOSection = sect;
2556 }
2557 else
2558 assert(1 && "multiple S_SYMBOL_STUBS sections");
2559 case S_LAZY_SYMBOL_POINTERS:
2560 break;
2561 case S_4BYTE_LITERALS:
2562 totalSectionsSize += sizeof(Literal4Section<A>);
2563 machOSects[count++].type = sectionTypeLiteral4;
2564 break;
2565 case S_8BYTE_LITERALS:
2566 totalSectionsSize += sizeof(Literal8Section<A>);
2567 machOSects[count++].type = sectionTypeLiteral8;
2568 break;
2569 case S_16BYTE_LITERALS:
2570 totalSectionsSize += sizeof(Literal16Section<A>);
2571 machOSects[count++].type = sectionTypeLiteral16;
2572 break;
2573 case S_NON_LAZY_SYMBOL_POINTERS:
2574 totalSectionsSize += sizeof(NonLazyPointerSection<A>);
2575 machOSects[count++].type = sectionTypeNonLazy;
2576 break;
2577 case S_THREAD_LOCAL_VARIABLE_POINTERS:
2578 totalSectionsSize += sizeof(TLVPointerSection<A>);
2579 machOSects[count++].type = sectionTypeTLVPointers;
2580 break;
2581 case S_LITERAL_POINTERS:
2582 if ( (strcmp(sect->segname(), "__OBJC") == 0) && (strcmp(sect->sectname(), "__cls_refs") == 0) ) {
2583 totalSectionsSize += sizeof(Objc1ClassReferences<A>);
2584 machOSects[count++].type = sectionTypeObjC1ClassRefs;
2585 }
2586 else {
2587 totalSectionsSize += sizeof(PointerToCStringSection<A>);
2588 machOSects[count++].type = sectionTypeCStringPointer;
2589 }
2590 break;
2591 case S_CSTRING_LITERALS:
2592 totalSectionsSize += sizeof(CStringSection<A>);
2593 machOSects[count++].type = sectionTypeCString;
2594 break;
2595 case S_MOD_INIT_FUNC_POINTERS:
2596 case S_MOD_TERM_FUNC_POINTERS:
2597 case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
2598 case S_INTERPOSING:
2599 case S_ZEROFILL:
2600 case S_REGULAR:
2601 case S_COALESCED:
2602 case S_THREAD_LOCAL_REGULAR:
2603 case S_THREAD_LOCAL_ZEROFILL:
2604 if ( (strcmp(sect->segname(), "__TEXT") == 0) && (strcmp(sect->sectname(), "__eh_frame") == 0) ) {
2605 totalSectionsSize += sizeof(CFISection<A>);
2606 machOSects[count++].type = sectionTypeCFI;
2607 }
2608 else if ( (strcmp(sect->segname(), "__DATA") == 0) && (strcmp(sect->sectname(), "__cfstring") == 0) ) {
2609 totalSectionsSize += sizeof(CFStringSection<A>);
2610 machOSects[count++].type = sectionTypeCFString;
2611 }
2612 else if ( (strcmp(sect->segname(), "__TEXT") == 0) && (strcmp(sect->sectname(), "__ustring") == 0) ) {
2613 totalSectionsSize += sizeof(UTF16StringSection<A>);
2614 machOSects[count++].type = sectionTypeUTF16Strings;
2615 }
2616 else if ( (strcmp(sect->segname(), "__DATA") == 0) && (strncmp(sect->sectname(), "__objc_classrefs", 16) == 0) ) {
2617 totalSectionsSize += sizeof(ObjC2ClassRefsSection<A>);
2618 machOSects[count++].type = sectionTypeObjC2ClassRefs;
2619 }
2620 else if ( (strcmp(sect->segname(), "__DATA") == 0) && (strcmp(sect->sectname(), "__objc_catlist") == 0) ) {
2621 totalSectionsSize += sizeof(ObjC2CategoryListSection<A>);
2622 machOSects[count++].type = typeObjC2CategoryList;
2623 }
2624 else if ( _AppleObjc && (strcmp(sect->segname(), "__OBJC") == 0) && (strcmp(sect->sectname(), "__class") == 0) ) {
2625 totalSectionsSize += sizeof(ObjC1ClassSection<A>);
2626 machOSects[count++].type = sectionTypeObjC1Classes;
2627 }
2628 else {
2629 totalSectionsSize += sizeof(SymboledSection<A>);
2630 machOSects[count++].type = sectionTypeSymboled;
2631 }
2632 break;
2633 case S_THREAD_LOCAL_VARIABLES:
2634 totalSectionsSize += sizeof(TLVDefsSection<A>);
2635 machOSects[count++].type = sectionTypeTLVDefs;
2636 break;
2637 default:
2638 throwf("unknown section type %d", sect->flags() & SECTION_TYPE);
2639 }
2640 }
2641
2642 // Create bitcode
2643 if ( bitcodeSect != NULL ) {
2644 if ( cmdlineSect != NULL )
2645 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::ClangBitcode(&_fileContent[bitcodeSect->offset()], bitcodeSect->size(),
2646 &_fileContent[cmdlineSect->offset()], cmdlineSect->size()));
2647 else if ( swiftCmdlineSect != NULL )
2648 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::SwiftBitcode(&_fileContent[bitcodeSect->offset()], bitcodeSect->size(),
2649 &_fileContent[swiftCmdlineSect->offset()], swiftCmdlineSect->size()));
2650 else
2651 throwf("Object file with bitcode missing cmdline options: %s", _file->path());
2652 }
2653 else if ( bundleSect != NULL )
2654 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::BundleBitcode(&_fileContent[bundleSect->offset()], bundleSect->size()));
2655 else if ( bitcodeAsm )
2656 _file->_bitcode = std::unique_ptr<ld::Bitcode>(new ld::AsmBitcode(_fileContent, _fileLength));
2657
2658 // sort by address (mach-o object files don't aways have sections sorted)
2659 ::qsort(machOSects, count, sizeof(MachOSectionAndSectionClass<P>), MachOSectionAndSectionClass<P>::sorter);
2660
2661 // we will synthesize a dummy Section<A> object for tentative definitions
2662 if ( _tentativeDefinitionCount > 0 ) {
2663 totalSectionsSize += sizeof(TentativeDefinitionSection<A>);
2664 machOSects[count++].type = sectionTypeTentativeDefinitions;
2665 }
2666
2667 // we will synthesize a dummy Section<A> object for Absolute symbols
2668 if ( _absoluteSymbolCount > 0 ) {
2669 totalSectionsSize += sizeof(AbsoluteSymbolSection<A>);
2670 machOSects[count++].type = sectionTypeAbsoluteSymbols;
2671 }
2672
2673 // allocate one block for all Section objects as well as pointers to each
2674 uint8_t* space = new uint8_t[totalSectionsSize+count*sizeof(Section<A>*)];
2675 _file->_sectionsArray = (Section<A>**)space;
2676 _file->_sectionsArrayCount = count;
2677 Section<A>** objects = _file->_sectionsArray;
2678 space += count*sizeof(Section<A>*);
2679 for (uint32_t i=0; i < count; ++i) {
2680 switch ( machOSects[i].type ) {
2681 case sectionTypeIgnore:
2682 break;
2683 case sectionTypeLiteral4:
2684 *objects++ = new (space) Literal4Section<A>(*this, *_file, machOSects[i].sect);
2685 space += sizeof(Literal4Section<A>);
2686 break;
2687 case sectionTypeLiteral8:
2688 *objects++ = new (space) Literal8Section<A>(*this, *_file, machOSects[i].sect);
2689 space += sizeof(Literal8Section<A>);
2690 break;
2691 case sectionTypeLiteral16:
2692 *objects++ = new (space) Literal16Section<A>(*this, *_file, machOSects[i].sect);
2693 space += sizeof(Literal16Section<A>);
2694 break;
2695 case sectionTypeNonLazy:
2696 *objects++ = new (space) NonLazyPointerSection<A>(*this, *_file, machOSects[i].sect);
2697 space += sizeof(NonLazyPointerSection<A>);
2698 break;
2699 case sectionTypeTLVPointers:
2700 *objects++ = new (space) TLVPointerSection<A>(*this, *_file, machOSects[i].sect);
2701 space += sizeof(TLVPointerSection<A>);
2702 break;
2703 case sectionTypeCFI:
2704 _EHFrameSection = new (space) CFISection<A>(*this, *_file, machOSects[i].sect);
2705 *objects++ = _EHFrameSection;
2706 space += sizeof(CFISection<A>);
2707 break;
2708 case sectionTypeCString:
2709 *objects++ = new (space) CStringSection<A>(*this, *_file, machOSects[i].sect);
2710 space += sizeof(CStringSection<A>);
2711 break;
2712 case sectionTypeCStringPointer:
2713 *objects++ = new (space) PointerToCStringSection<A>(*this, *_file, machOSects[i].sect);
2714 space += sizeof(PointerToCStringSection<A>);
2715 break;
2716 case sectionTypeObjC1ClassRefs:
2717 *objects++ = new (space) Objc1ClassReferences<A>(*this, *_file, machOSects[i].sect);
2718 space += sizeof(Objc1ClassReferences<A>);
2719 break;
2720 case sectionTypeUTF16Strings:
2721 *objects++ = new (space) UTF16StringSection<A>(*this, *_file, machOSects[i].sect);
2722 space += sizeof(UTF16StringSection<A>);
2723 break;
2724 case sectionTypeCFString:
2725 *objects++ = new (space) CFStringSection<A>(*this, *_file, machOSects[i].sect);
2726 space += sizeof(CFStringSection<A>);
2727 break;
2728 case sectionTypeObjC2ClassRefs:
2729 *objects++ = new (space) ObjC2ClassRefsSection<A>(*this, *_file, machOSects[i].sect);
2730 space += sizeof(ObjC2ClassRefsSection<A>);
2731 break;
2732 case typeObjC2CategoryList:
2733 *objects++ = new (space) ObjC2CategoryListSection<A>(*this, *_file, machOSects[i].sect);
2734 space += sizeof(ObjC2CategoryListSection<A>);
2735 break;
2736 case sectionTypeObjC1Classes:
2737 *objects++ = new (space) ObjC1ClassSection<A>(*this, *_file, machOSects[i].sect);
2738 space += sizeof(ObjC1ClassSection<A>);
2739 break;
2740 case sectionTypeSymboled:
2741 *objects++ = new (space) SymboledSection<A>(*this, *_file, machOSects[i].sect);
2742 space += sizeof(SymboledSection<A>);
2743 break;
2744 case sectionTypeTLVDefs:
2745 *objects++ = new (space) TLVDefsSection<A>(*this, *_file, machOSects[i].sect);
2746 space += sizeof(TLVDefsSection<A>);
2747 break;
2748 case sectionTypeCompactUnwind:
2749 _compactUnwindSection = new (space) CUSection<A>(*this, *_file, machOSects[i].sect);
2750 *objects++ = _compactUnwindSection;
2751 space += sizeof(CUSection<A>);
2752 break;
2753 case sectionTypeTentativeDefinitions:
2754 *objects++ = new (space) TentativeDefinitionSection<A>(*this, *_file);
2755 space += sizeof(TentativeDefinitionSection<A>);
2756 break;
2757 case sectionTypeAbsoluteSymbols:
2758 _absoluteSection = new (space) AbsoluteSymbolSection<A>(*this, *_file);
2759 *objects++ = _absoluteSection;
2760 space += sizeof(AbsoluteSymbolSection<A>);
2761 break;
2762 default:
2763 throw "internal error uknown SectionType";
2764 }
2765 }
2766 }
2767
2768
2769 template <typename A>
2770 Section<A>* Parser<A>::sectionForAddress(typename A::P::uint_t addr)
2771 {
2772 for (uint32_t i=0; i < _file->_sectionsArrayCount; ++i ) {
2773 const macho_section<typename A::P>* sect = _file->_sectionsArray[i]->machoSection();
2774 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
2775 if ( sect != NULL ) {
2776 if ( (sect->addr() <= addr) && (addr < (sect->addr()+sect->size())) ) {
2777 return _file->_sectionsArray[i];
2778 }
2779 }
2780 }
2781 // not strictly in any section
2782 // may be in a zero length section
2783 for (uint32_t i=0; i < _file->_sectionsArrayCount; ++i ) {
2784 const macho_section<typename A::P>* sect = _file->_sectionsArray[i]->machoSection();
2785 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
2786 if ( sect != NULL ) {
2787 if ( (sect->addr() == addr) && (sect->size() == 0) ) {
2788 return _file->_sectionsArray[i];
2789 }
2790 }
2791 }
2792
2793 throwf("sectionForAddress(0x%llX) address not in any section", (uint64_t)addr);
2794 }
2795
2796 template <typename A>
2797 Section<A>* Parser<A>::sectionForNum(unsigned int num)
2798 {
2799 for (uint32_t i=0; i < _file->_sectionsArrayCount; ++i ) {
2800 const macho_section<typename A::P>* sect = _file->_sectionsArray[i]->machoSection();
2801 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
2802 if ( sect != NULL ) {
2803 if ( num == (unsigned int)((sect - _sectionsStart)+1) )
2804 return _file->_sectionsArray[i];
2805 }
2806 }
2807 throwf("sectionForNum(%u) section number not for any section", num);
2808 }
2809
2810 template <typename A>
2811 Atom<A>* Parser<A>::findAtomByAddress(pint_t addr)
2812 {
2813 Section<A>* section = this->sectionForAddress(addr);
2814 return section->findAtomByAddress(addr);
2815 }
2816
2817 template <typename A>
2818 Atom<A>* Parser<A>::findAtomByAddressOrNullIfStub(pint_t addr)
2819 {
2820 if ( hasStubsSection() && (_stubsMachOSection->addr() <= addr) && (addr < (_stubsMachOSection->addr()+_stubsMachOSection->size())) )
2821 return NULL;
2822 return findAtomByAddress(addr);
2823 }
2824
2825 template <typename A>
2826 Atom<A>* Parser<A>::findAtomByAddressOrLocalTargetOfStub(pint_t addr, uint32_t* offsetInAtom)
2827 {
2828 if ( hasStubsSection() && (_stubsMachOSection->addr() <= addr) && (addr < (_stubsMachOSection->addr()+_stubsMachOSection->size())) ) {
2829 // target is a stub, remove indirection
2830 uint32_t symbolIndex = this->symbolIndexFromIndirectSectionAddress(addr, _stubsMachOSection);
2831 assert(symbolIndex != INDIRECT_SYMBOL_LOCAL);
2832 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
2833 // can't be to external weak symbol
2834 assert( (this->combineFromSymbol(sym) != ld::Atom::combineByName) || (this->scopeFromSymbol(sym) != ld::Atom::scopeGlobal) );
2835 *offsetInAtom = 0;
2836 return this->findAtomByName(this->nameFromSymbol(sym));
2837 }
2838 Atom<A>* target = this->findAtomByAddress(addr);
2839 *offsetInAtom = addr - target->_objAddress;
2840 return target;
2841 }
2842
2843 template <typename A>
2844 Atom<A>* Parser<A>::findAtomByName(const char* name)
2845 {
2846 uint8_t* p = _file->_atomsArray;
2847 for(int i=_file->_atomsArrayCount; i > 0; --i) {
2848 Atom<A>* atom = (Atom<A>*)p;
2849 if ( strcmp(name, atom->name()) == 0 )
2850 return atom;
2851 p += sizeof(Atom<A>);
2852 }
2853 return NULL;
2854 }
2855
2856 template <typename A>
2857 void Parser<A>::findTargetFromAddress(pint_t addr, TargetDesc& target)
2858 {
2859 if ( hasStubsSection() && (_stubsMachOSection->addr() <= addr) && (addr < (_stubsMachOSection->addr()+_stubsMachOSection->size())) ) {
2860 // target is a stub, remove indirection
2861 uint32_t symbolIndex = this->symbolIndexFromIndirectSectionAddress(addr, _stubsMachOSection);
2862 assert(symbolIndex != INDIRECT_SYMBOL_LOCAL);
2863 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
2864 target.atom = NULL;
2865 target.name = this->nameFromSymbol(sym);
2866 target.weakImport = this->weakImportFromSymbol(sym);
2867 target.addend = 0;
2868 return;
2869 }
2870 Section<A>* section = this->sectionForAddress(addr);
2871 target.atom = section->findAtomByAddress(addr);
2872 target.addend = addr - target.atom->_objAddress;
2873 target.weakImport = false;
2874 target.name = NULL;
2875 }
2876
2877 template <typename A>
2878 void Parser<A>::findTargetFromAddress(pint_t baseAddr, pint_t addr, TargetDesc& target)
2879 {
2880 findTargetFromAddress(baseAddr, target);
2881 target.addend = addr - target.atom->_objAddress;
2882 }
2883
2884 template <typename A>
2885 void Parser<A>::findTargetFromAddressAndSectionNum(pint_t addr, unsigned int sectNum, TargetDesc& target)
2886 {
2887 if ( sectNum == R_ABS ) {
2888 // target is absolute symbol that corresponds to addr
2889 if ( _absoluteSection != NULL ) {
2890 target.atom = _absoluteSection->findAbsAtomForValue(addr);
2891 if ( target.atom != NULL ) {
2892 target.name = NULL;
2893 target.weakImport = false;
2894 target.addend = 0;
2895 return;
2896 }
2897 }
2898 throwf("R_ABS reloc but no absolute symbol at target address");
2899 }
2900
2901 if ( hasStubsSection() && (stubsSectionNum() == sectNum) ) {
2902 // target is a stub, remove indirection
2903 uint32_t symbolIndex = this->symbolIndexFromIndirectSectionAddress(addr, _stubsMachOSection);
2904 assert(symbolIndex != INDIRECT_SYMBOL_LOCAL);
2905 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
2906 // use direct reference when stub is to a static function
2907 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (this->nameFromSymbol(sym)[0] == 'L')) ) {
2908 this->findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
2909 }
2910 else {
2911 target.atom = NULL;
2912 target.name = this->nameFromSymbol(sym);
2913 target.weakImport = this->weakImportFromSymbol(sym);
2914 target.addend = 0;
2915 }
2916 return;
2917 }
2918 Section<A>* section = this->sectionForNum(sectNum);
2919 target.atom = section->findAtomByAddress(addr);
2920 if ( target.atom == NULL ) {
2921 typedef typename A::P::sint_t sint_t;
2922 sint_t a = (sint_t)addr;
2923 sint_t sectStart = (sint_t)(section->machoSection()->addr());
2924 sint_t sectEnd = sectStart + section->machoSection()->size();
2925 if ( a < sectStart ) {
2926 // target address is before start of section, so must be negative addend
2927 target.atom = section->findAtomByAddress(sectStart);
2928 target.addend = a - sectStart;
2929 target.weakImport = false;
2930 target.name = NULL;
2931 return;
2932 }
2933 else if ( a >= sectEnd ) {
2934 target.atom = section->findAtomByAddress(sectEnd-1);
2935 target.addend = a - sectEnd;
2936 target.weakImport = false;
2937 target.name = NULL;
2938 return;
2939 }
2940 }
2941 assert(target.atom != NULL);
2942 target.addend = addr - target.atom->_objAddress;
2943 target.weakImport = false;
2944 target.name = NULL;
2945 }
2946
2947 template <typename A>
2948 void Parser<A>::addDtraceExtraInfos(const SourceLocation& src, const char* providerName)
2949 {
2950 // for every ___dtrace_stability$* and ___dtrace_typedefs$* undefine with
2951 // a matching provider name, add a by-name kDtraceTypeReference at probe site
2952 const char* dollar = strchr(providerName, '$');
2953 if ( dollar != NULL ) {
2954 int providerNameLen = dollar-providerName+1;
2955 for ( std::vector<const char*>::iterator it = _dtraceProviderInfo.begin(); it != _dtraceProviderInfo.end(); ++it) {
2956 const char* typeDollar = strchr(*it, '$');
2957 if ( typeDollar != NULL ) {
2958 if ( strncmp(typeDollar+1, providerName, providerNameLen) == 0 ) {
2959 addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindDtraceExtra,false, *it);
2960 }
2961 }
2962 }
2963 }
2964 }
2965
2966 template <typename A>
2967 const char* Parser<A>::scanSymbolTableForAddress(uint64_t addr)
2968 {
2969 uint64_t closestSymAddr = 0;
2970 const char* closestSymName = NULL;
2971 for (uint32_t i=0; i < this->_symbolCount; ++i) {
2972 const macho_nlist<P>& sym = symbolFromIndex(i);
2973 // ignore stabs
2974 if ( (sym.n_type() & N_STAB) != 0 )
2975 continue;
2976
2977 // only look at definitions
2978 if ( (sym.n_type() & N_TYPE) != N_SECT )
2979 continue;
2980
2981 // return with exact match
2982 if ( sym.n_value() == addr ) {
2983 const char* name = nameFromSymbol(sym);
2984 if ( strncmp(name, "ltmp", 4) != 0 )
2985 return name;
2986 // treat 'ltmp*' labels as close match
2987 closestSymAddr = sym.n_value();
2988 closestSymName = name;
2989 }
2990
2991 // record closest seen so far
2992 if ( (sym.n_value() < addr) && ((sym.n_value() > closestSymAddr) || (closestSymName == NULL)) )
2993 closestSymName = nameFromSymbol(sym);
2994 }
2995
2996 return (closestSymName != NULL) ? closestSymName : "unknown";
2997 }
2998
2999
3000 template <typename A>
3001 void Parser<A>::addFixups(const SourceLocation& src, ld::Fixup::Kind setKind, const TargetDesc& target)
3002 {
3003 // some fixup pairs can be combined
3004 ld::Fixup::Cluster cl = ld::Fixup::k1of3;
3005 ld::Fixup::Kind firstKind = ld::Fixup::kindSetTargetAddress;
3006 bool combined = false;
3007 if ( target.addend == 0 ) {
3008 cl = ld::Fixup::k1of1;
3009 combined = true;
3010 switch ( setKind ) {
3011 case ld::Fixup::kindStoreLittleEndian32:
3012 firstKind = ld::Fixup::kindStoreTargetAddressLittleEndian32;
3013 break;
3014 case ld::Fixup::kindStoreLittleEndian64:
3015 firstKind = ld::Fixup::kindStoreTargetAddressLittleEndian64;
3016 break;
3017 case ld::Fixup::kindStoreBigEndian32:
3018 firstKind = ld::Fixup::kindStoreTargetAddressBigEndian32;
3019 break;
3020 case ld::Fixup::kindStoreBigEndian64:
3021 firstKind = ld::Fixup::kindStoreTargetAddressBigEndian64;
3022 break;
3023 case ld::Fixup::kindStoreX86BranchPCRel32:
3024 firstKind = ld::Fixup::kindStoreTargetAddressX86BranchPCRel32;
3025 break;
3026 case ld::Fixup::kindStoreX86PCRel32:
3027 firstKind = ld::Fixup::kindStoreTargetAddressX86PCRel32;
3028 break;
3029 case ld::Fixup::kindStoreX86PCRel32GOTLoad:
3030 firstKind = ld::Fixup::kindStoreTargetAddressX86PCRel32GOTLoad;
3031 break;
3032 case ld::Fixup::kindStoreX86PCRel32TLVLoad:
3033 firstKind = ld::Fixup::kindStoreTargetAddressX86PCRel32TLVLoad;
3034 break;
3035 case ld::Fixup::kindStoreX86Abs32TLVLoad:
3036 firstKind = ld::Fixup::kindStoreTargetAddressX86Abs32TLVLoad;
3037 break;
3038 case ld::Fixup::kindStoreARMBranch24:
3039 firstKind = ld::Fixup::kindStoreTargetAddressARMBranch24;
3040 break;
3041 case ld::Fixup::kindStoreThumbBranch22:
3042 firstKind = ld::Fixup::kindStoreTargetAddressThumbBranch22;
3043 break;
3044 #if SUPPORT_ARCH_arm64
3045 case ld::Fixup::kindStoreARM64Branch26:
3046 firstKind = ld::Fixup::kindStoreTargetAddressARM64Branch26;
3047 break;
3048 case ld::Fixup::kindStoreARM64Page21:
3049 firstKind = ld::Fixup::kindStoreTargetAddressARM64Page21;
3050 break;
3051 case ld::Fixup::kindStoreARM64PageOff12:
3052 firstKind = ld::Fixup::kindStoreTargetAddressARM64PageOff12;
3053 break;
3054 case ld::Fixup::kindStoreARM64GOTLoadPage21:
3055 firstKind = ld::Fixup::kindStoreTargetAddressARM64GOTLoadPage21;
3056 break;
3057 case ld::Fixup::kindStoreARM64GOTLoadPageOff12:
3058 firstKind = ld::Fixup::kindStoreTargetAddressARM64GOTLoadPageOff12;
3059 break;
3060 case ld::Fixup::kindStoreARM64TLVPLoadPage21:
3061 firstKind = ld::Fixup::kindStoreTargetAddressARM64TLVPLoadPage21;
3062 break;
3063 case ld::Fixup::kindStoreARM64TLVPLoadPageOff12:
3064 firstKind = ld::Fixup::kindStoreTargetAddressARM64TLVPLoadPageOff12;
3065 break;
3066 #endif
3067 default:
3068 combined = false;
3069 cl = ld::Fixup::k1of2;
3070 break;
3071 }
3072 }
3073
3074 if ( target.atom != NULL ) {
3075 if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
3076 addFixup(src, cl, firstKind, target.atom);
3077 }
3078 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
3079 addFixup(src, cl, firstKind, ld::Fixup::bindingByContentBound, target.atom);
3080 }
3081 else if ( (src.atom->section().type() == ld::Section::typeCFString) && (src.offsetInAtom != 0) ) {
3082 // backing string in CFStrings should always be direct
3083 addFixup(src, cl, firstKind, target.atom);
3084 }
3085 else if ( (src.atom == target.atom) && (target.atom->combine() == ld::Atom::combineByName) ) {
3086 // reference to self should always be direct
3087 addFixup(src, cl, firstKind, target.atom);
3088 }
3089 else {
3090 // change direct fixup to by-name fixup
3091 addFixup(src, cl, firstKind, false, target.atom->name());
3092 }
3093 }
3094 else {
3095 addFixup(src, cl, firstKind, target.weakImport, target.name);
3096 }
3097 if ( target.addend == 0 ) {
3098 if ( ! combined )
3099 addFixup(src, ld::Fixup::k2of2, setKind);
3100 }
3101 else {
3102 addFixup(src, ld::Fixup::k2of3, ld::Fixup::kindAddAddend, target.addend);
3103 addFixup(src, ld::Fixup::k3of3, setKind);
3104 }
3105 }
3106
3107 template <typename A>
3108 void Parser<A>::addFixups(const SourceLocation& src, ld::Fixup::Kind kind, const TargetDesc& target, const TargetDesc& picBase)
3109 {
3110 ld::Fixup::Cluster cl = (target.addend == 0) ? ld::Fixup::k1of4 : ld::Fixup::k1of5;
3111 if ( target.atom != NULL ) {
3112 if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
3113 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, target.atom);
3114 }
3115 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
3116 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, target.atom);
3117 }
3118 else {
3119 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, false, target.atom->name());
3120 }
3121 }
3122 else {
3123 addFixup(src, cl, ld::Fixup::kindSetTargetAddress, target.weakImport, target.name);
3124 }
3125 if ( target.addend == 0 ) {
3126 assert(picBase.atom != NULL);
3127 addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, picBase.atom);
3128 addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, picBase.addend);
3129 addFixup(src, ld::Fixup::k4of4, kind);
3130 }
3131 else {
3132 addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, target.addend);
3133 addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, picBase.atom);
3134 addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, picBase.addend);
3135 addFixup(src, ld::Fixup::k5of5, kind);
3136 }
3137 }
3138
3139
3140
3141 template <typename A>
3142 uint32_t TentativeDefinitionSection<A>::computeAtomCount(class Parser<A>& parser,
3143 struct Parser<A>::LabelAndCFIBreakIterator& it,
3144 const struct Parser<A>::CFI_CU_InfoArrays&)
3145 {
3146 return parser.tentativeDefinitionCount();
3147 }
3148
3149 template <typename A>
3150 uint32_t TentativeDefinitionSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
3151 struct Parser<A>::LabelAndCFIBreakIterator& it,
3152 const struct Parser<A>::CFI_CU_InfoArrays&)
3153 {
3154 this->_beginAtoms = (Atom<A>*)p;
3155 uint32_t count = 0;
3156 for (uint32_t i=parser.undefinedStartIndex(); i < parser.undefinedEndIndex(); ++i) {
3157 const macho_nlist<P>& sym = parser.symbolFromIndex(i);
3158 if ( ((sym.n_type() & N_TYPE) == N_UNDF) && (sym.n_value() != 0) ) {
3159 uint64_t size = sym.n_value();
3160 uint8_t alignP2 = GET_COMM_ALIGN(sym.n_desc());
3161 if ( alignP2 == 0 ) {
3162 // common symbols align to their size
3163 // that is, a 4-byte common aligns to 4-bytes
3164 // if this size is not a power of two,
3165 // then round up to the next power of two
3166 alignP2 = 63 - (uint8_t)__builtin_clzll(size);
3167 if ( size != (1ULL << alignP2) )
3168 ++alignP2;
3169 }
3170 // limit alignment of extremely large commons to 2^15 bytes (8-page)
3171 if ( alignP2 > parser.maxDefaultCommonAlignment() )
3172 alignP2 = parser.maxDefaultCommonAlignment();
3173 Atom<A>* allocatedSpace = (Atom<A>*)p;
3174 new (allocatedSpace) Atom<A>(*this, parser.nameFromSymbol(sym), (pint_t)ULLONG_MAX, size,
3175 ld::Atom::definitionTentative, ld::Atom::combineByName,
3176 parser.scopeFromSymbol(sym), ld::Atom::typeZeroFill, ld::Atom::symbolTableIn,
3177 parser.dontDeadStripFromSymbol(sym), false, false, ld::Atom::Alignment(alignP2) );
3178 p += sizeof(Atom<A>);
3179 ++count;
3180 }
3181 }
3182 this->_endAtoms = (Atom<A>*)p;
3183 return count;
3184 }
3185
3186
3187 template <typename A>
3188 uint32_t AbsoluteSymbolSection<A>::computeAtomCount(class Parser<A>& parser,
3189 struct Parser<A>::LabelAndCFIBreakIterator& it,
3190 const struct Parser<A>::CFI_CU_InfoArrays&)
3191 {
3192 return parser.absoluteSymbolCount();
3193 }
3194
3195 template <typename A>
3196 uint32_t AbsoluteSymbolSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
3197 struct Parser<A>::LabelAndCFIBreakIterator& it,
3198 const struct Parser<A>::CFI_CU_InfoArrays&)
3199 {
3200 this->_beginAtoms = (Atom<A>*)p;
3201 uint32_t count = 0;
3202 for (uint32_t i=0; i < parser.symbolCount(); ++i) {
3203 const macho_nlist<P>& sym = parser.symbolFromIndex(i);
3204 if ( (sym.n_type() & N_TYPE) != N_ABS )
3205 continue;
3206 const char* absName = parser.nameFromSymbol(sym);
3207 // ignore .objc_class_name_* symbols
3208 if ( strncmp(absName, ".objc_class_name_", 17) == 0 )
3209 continue;
3210 // ignore .objc_class_name_* symbols
3211 if ( strncmp(absName, ".objc_category_name_", 20) == 0 )
3212 continue;
3213 // ignore empty *.eh symbols
3214 if ( strcmp(&absName[strlen(absName)-3], ".eh") == 0 )
3215 continue;
3216
3217 Atom<A>* allocatedSpace = (Atom<A>*)p;
3218 new (allocatedSpace) Atom<A>(*this, parser, sym, 0);
3219 p += sizeof(Atom<A>);
3220 ++count;
3221 }
3222 this->_endAtoms = (Atom<A>*)p;
3223 return count;
3224 }
3225
3226 template <typename A>
3227 Atom<A>* AbsoluteSymbolSection<A>::findAbsAtomForValue(typename A::P::uint_t value)
3228 {
3229 Atom<A>* end = this->_endAtoms;
3230 for(Atom<A>* p = this->_beginAtoms; p < end; ++p) {
3231 if ( p->_objAddress == value )
3232 return p;
3233 }
3234 return NULL;
3235 }
3236
3237
3238 template <typename A>
3239 uint32_t Parser<A>::indirectSymbol(uint32_t indirectIndex)
3240 {
3241 if ( indirectIndex >= _indirectTableCount )
3242 throw "indirect symbol index out of range";
3243 return E::get32(_indirectTable[indirectIndex]);
3244 }
3245
3246 template <typename A>
3247 const macho_nlist<typename A::P>& Parser<A>::symbolFromIndex(uint32_t index)
3248 {
3249 if ( index > _symbolCount )
3250 throw "symbol index out of range";
3251 return _symbols[index];
3252 }
3253
3254 template <typename A>
3255 const macho_section<typename A::P>* Parser<A>::machOSectionFromSectionIndex(uint32_t index)
3256 {
3257 if ( index >= _machOSectionsCount )
3258 throw "section index out of range";
3259 return &_sectionsStart[index];
3260 }
3261
3262 template <typename A>
3263 uint32_t Parser<A>::symbolIndexFromIndirectSectionAddress(pint_t addr, const macho_section<P>* sect)
3264 {
3265 uint32_t elementSize = 0;
3266 switch ( sect->flags() & SECTION_TYPE ) {
3267 case S_SYMBOL_STUBS:
3268 elementSize = sect->reserved2();
3269 break;
3270 case S_LAZY_SYMBOL_POINTERS:
3271 case S_NON_LAZY_SYMBOL_POINTERS:
3272 elementSize = sizeof(pint_t);
3273 break;
3274 default:
3275 throw "section does not use inirect symbol table";
3276 }
3277 uint32_t indexInSection = (addr - sect->addr()) / elementSize;
3278 uint32_t indexIntoIndirectTable = sect->reserved1() + indexInSection;
3279 return this->indirectSymbol(indexIntoIndirectTable);
3280 }
3281
3282
3283
3284 template <typename A>
3285 const char* Parser<A>::nameFromSymbol(const macho_nlist<P>& sym)
3286 {
3287 return &_strings[sym.n_strx()];
3288 }
3289
3290 template <typename A>
3291 ld::Atom::Scope Parser<A>::scopeFromSymbol(const macho_nlist<P>& sym)
3292 {
3293 if ( (sym.n_type() & N_EXT) == 0 )
3294 return ld::Atom::scopeTranslationUnit;
3295 else if ( (sym.n_type() & N_PEXT) != 0 )
3296 return ld::Atom::scopeLinkageUnit;
3297 else if ( this->nameFromSymbol(sym)[0] == 'l' ) // since all 'l' symbols will be remove, don't make them global
3298 return ld::Atom::scopeLinkageUnit;
3299 else
3300 return ld::Atom::scopeGlobal;
3301 }
3302
3303 template <typename A>
3304 ld::Atom::Definition Parser<A>::definitionFromSymbol(const macho_nlist<P>& sym)
3305 {
3306 switch ( sym.n_type() & N_TYPE ) {
3307 case N_ABS:
3308 return ld::Atom::definitionAbsolute;
3309 case N_SECT:
3310 return ld::Atom::definitionRegular;
3311 case N_UNDF:
3312 if ( sym.n_value() != 0 )
3313 return ld::Atom::definitionTentative;
3314 }
3315 throw "definitionFromSymbol() bad symbol";
3316 }
3317
3318 template <typename A>
3319 ld::Atom::Combine Parser<A>::combineFromSymbol(const macho_nlist<P>& sym)
3320 {
3321 if ( sym.n_desc() & N_WEAK_DEF )
3322 return ld::Atom::combineByName;
3323 else
3324 return ld::Atom::combineNever;
3325 }
3326
3327
3328 template <typename A>
3329 ld::Atom::SymbolTableInclusion Parser<A>::inclusionFromSymbol(const macho_nlist<P>& sym)
3330 {
3331 const char* symbolName = nameFromSymbol(sym);
3332 // labels beginning with 'l' (lowercase ell) are automatically removed in final linked images <rdar://problem/4571042>
3333 // labels beginning with 'L' should have been stripped by the assembler, so are stripped now
3334 if ( sym.n_desc() & REFERENCED_DYNAMICALLY )
3335 return ld::Atom::symbolTableInAndNeverStrip;
3336 else if ( symbolName[0] == 'l' )
3337 return ld::Atom::symbolTableNotInFinalLinkedImages;
3338 else if ( symbolName[0] == 'L' )
3339 return ld::Atom::symbolTableNotIn;
3340 else
3341 return ld::Atom::symbolTableIn;
3342 }
3343
3344 template <typename A>
3345 bool Parser<A>::dontDeadStripFromSymbol(const macho_nlist<P>& sym)
3346 {
3347 return ( (sym.n_desc() & (N_NO_DEAD_STRIP|REFERENCED_DYNAMICALLY)) != 0 );
3348 }
3349
3350 template <typename A>
3351 bool Parser<A>::isThumbFromSymbol(const macho_nlist<P>& sym)
3352 {
3353 return ( sym.n_desc() & N_ARM_THUMB_DEF );
3354 }
3355
3356 template <typename A>
3357 bool Parser<A>::weakImportFromSymbol(const macho_nlist<P>& sym)
3358 {
3359 return ( ((sym.n_type() & N_TYPE) == N_UNDF) && ((sym.n_desc() & N_WEAK_REF) != 0) );
3360 }
3361
3362 template <typename A>
3363 bool Parser<A>::resolverFromSymbol(const macho_nlist<P>& sym)
3364 {
3365 return ( sym.n_desc() & N_SYMBOL_RESOLVER );
3366 }
3367
3368 template <typename A>
3369 bool Parser<A>::altEntryFromSymbol(const macho_nlist<P>& sym)
3370 {
3371 return ( sym.n_desc() & N_ALT_ENTRY );
3372 }
3373
3374
3375 /* Skip over a LEB128 value (signed or unsigned). */
3376 static void
3377 skip_leb128 (const uint8_t ** offset, const uint8_t * end)
3378 {
3379 while (*offset != end && **offset >= 0x80)
3380 (*offset)++;
3381 if (*offset != end)
3382 (*offset)++;
3383 }
3384
3385 /* Read a ULEB128 into a 64-bit word. Return (uint64_t)-1 on overflow
3386 or error. On overflow, skip past the rest of the uleb128. */
3387 static uint64_t
3388 read_uleb128 (const uint8_t ** offset, const uint8_t * end)
3389 {
3390 uint64_t result = 0;
3391 int bit = 0;
3392
3393 do {
3394 uint64_t b;
3395
3396 if (*offset == end)
3397 return (uint64_t) -1;
3398
3399 b = **offset & 0x7f;
3400
3401 if (bit >= 64 || b << bit >> bit != b)
3402 result = (uint64_t) -1;
3403 else
3404 result |= b << bit, bit += 7;
3405 } while (*(*offset)++ >= 0x80);
3406 return result;
3407 }
3408
3409
3410 /* Skip over a DWARF attribute of form FORM. */
3411 template <typename A>
3412 bool Parser<A>::skip_form(const uint8_t ** offset, const uint8_t * end, uint64_t form,
3413 uint8_t addr_size, bool dwarf64)
3414 {
3415 int64_t sz=0;
3416
3417 switch (form)
3418 {
3419 case DW_FORM_addr:
3420 sz = addr_size;
3421 break;
3422
3423 case DW_FORM_block2:
3424 if (end - *offset < 2)
3425 return false;
3426 sz = 2 + A::P::E::get16(*(uint16_t*)offset);
3427 break;
3428
3429 case DW_FORM_block4:
3430 if (end - *offset < 4)
3431 return false;
3432 sz = 2 + A::P::E::get32(*(uint32_t*)offset);
3433 break;
3434
3435 case DW_FORM_data2:
3436 case DW_FORM_ref2:
3437 sz = 2;
3438 break;
3439
3440 case DW_FORM_data4:
3441 case DW_FORM_ref4:
3442 sz = 4;
3443 break;
3444
3445 case DW_FORM_data8:
3446 case DW_FORM_ref8:
3447 sz = 8;
3448 break;
3449
3450 case DW_FORM_string:
3451 while (*offset != end && **offset)
3452 ++*offset;
3453 case DW_FORM_data1:
3454 case DW_FORM_flag:
3455 case DW_FORM_ref1:
3456 sz = 1;
3457 break;
3458
3459 case DW_FORM_block:
3460 sz = read_uleb128 (offset, end);
3461 break;
3462
3463 case DW_FORM_block1:
3464 if (*offset == end)
3465 return false;
3466 sz = 1 + **offset;
3467 break;
3468
3469 case DW_FORM_sdata:
3470 case DW_FORM_udata:
3471 case DW_FORM_ref_udata:
3472 skip_leb128 (offset, end);
3473 return true;
3474
3475 case DW_FORM_strp:
3476 case DW_FORM_ref_addr:
3477 sz = 4;
3478 break;
3479
3480 case DW_FORM_sec_offset:
3481 sz = sizeof(typename A::P::uint_t);
3482 break;
3483
3484 case DW_FORM_exprloc:
3485 sz = read_uleb128 (offset, end);
3486 break;
3487
3488 case DW_FORM_flag_present:
3489 sz = 0;
3490 break;
3491
3492 case DW_FORM_ref_sig8:
3493 sz = 8;
3494 break;
3495
3496 default:
3497 return false;
3498 }
3499 if (end - *offset < sz)
3500 return false;
3501 *offset += sz;
3502 return true;
3503 }
3504
3505
3506 template <typename A>
3507 const char* Parser<A>::getDwarfString(uint64_t form, const uint8_t*& di)
3508 {
3509 uint32_t offset;
3510 const char* dwarfStrings;
3511 const char* result = NULL;
3512 switch (form) {
3513 case DW_FORM_string:
3514 result = (const char*)di;
3515 di += strlen(result) + 1;
3516 break;
3517 case DW_FORM_strp:
3518 offset = E::get32(*((uint32_t*)di));
3519 dwarfStrings = (char*)_file->fileContent() + _file->_dwarfDebugStringSect->offset();
3520 if ( offset < _file->_dwarfDebugStringSect->size() )
3521 result = &dwarfStrings[offset];
3522 else
3523 warning("dwarf DW_FORM_strp (offset=0x%08X) is too big in %s", offset, this->_path);
3524 di += 4;
3525 break;
3526 default:
3527 warning("unknown dwarf string encoding (form=%lld) in %s", form, this->_path);
3528 break;
3529 }
3530 return result;
3531 }
3532
3533 template <typename A>
3534 uint64_t Parser<A>::getDwarfOffset(uint64_t form, const uint8_t*& di, bool dwarf64)
3535 {
3536 if ( form == DW_FORM_sec_offset )
3537 form = (dwarf64 ? DW_FORM_data8 : DW_FORM_data4);
3538 uint64_t result = -1;
3539 switch (form) {
3540 case DW_FORM_data4:
3541 result = A::P::E::get32(*(uint32_t*)di);
3542 di += 4;
3543 break;
3544 case DW_FORM_data8:
3545 result = A::P::E::get64(*(uint64_t*)di);
3546 di += 8;
3547 break;
3548 default:
3549 warning("unknown dwarf DW_FORM_ for DW_AT_stmt_list in %s", this->_path);
3550 }
3551 return result;
3552 }
3553
3554
3555 template <typename A>
3556 struct AtomAndLineInfo {
3557 Atom<A>* atom;
3558 ld::Atom::LineInfo info;
3559 };
3560
3561
3562 // <rdar://problem/5591394> Add support to ld64 for N_FUN stabs when used for symbolic constants
3563 // Returns whether a stabStr belonging to an N_FUN stab represents a
3564 // symbolic constant rather than a function
3565 template <typename A>
3566 bool Parser<A>::isConstFunStabs(const char *stabStr)
3567 {
3568 const char* colon;
3569 // N_FUN can be used for both constants and for functions. In case it's a constant,
3570 // the format of the stabs string is "symname:c=<value>;"
3571 // ':' cannot appear in the symbol name, except if it's an Objective-C method
3572 // (in which case the symbol name starts with + or -, and then it's definitely
3573 // not a constant)
3574 return (stabStr != NULL) && (stabStr[0] != '+') && (stabStr[0] != '-')
3575 && ((colon = strchr(stabStr, ':')) != NULL)
3576 && (colon[1] == 'c') && (colon[2] == '=');
3577 }
3578
3579
3580 template <typename A>
3581 void Parser<A>::parseDebugInfo()
3582 {
3583 // check for dwarf __debug_info section
3584 if ( _file->_dwarfDebugInfoSect == NULL ) {
3585 // if no DWARF debug info, look for stabs
3586 this->parseStabs();
3587 return;
3588 }
3589 if ( _file->_dwarfDebugInfoSect->size() == 0 )
3590 return;
3591
3592 uint64_t stmtList;
3593 const char* tuDir;
3594 const char* tuName;
3595 if ( !read_comp_unit(&tuName, &tuDir, &stmtList) ) {
3596 // if can't parse dwarf, warn and give up
3597 _file->_dwarfTranslationUnitPath = NULL;
3598 warning("can't parse dwarf compilation unit info in %s", _path);
3599 _file->_debugInfoKind = ld::relocatable::File::kDebugInfoNone;
3600 return;
3601 }
3602 if ( (tuName != NULL) && (tuName[0] == '/') ) {
3603 _file->_dwarfTranslationUnitPath = tuName;
3604 }
3605 else if ( (tuDir != NULL) && (tuName != NULL) ) {
3606 asprintf((char**)&(_file->_dwarfTranslationUnitPath), "%s/%s", tuDir, tuName);
3607 }
3608 else if ( tuDir == NULL ) {
3609 _file->_dwarfTranslationUnitPath = tuName;
3610 }
3611 else {
3612 _file->_dwarfTranslationUnitPath = NULL;
3613 }
3614
3615 // add line number info to atoms from dwarf
3616 std::vector<AtomAndLineInfo<A> > entries;
3617 entries.reserve(64);
3618 if ( _file->_debugInfoKind == ld::relocatable::File::kDebugInfoDwarf ) {
3619 // file with just data will have no __debug_line info
3620 if ( (_file->_dwarfDebugLineSect != NULL) && (_file->_dwarfDebugLineSect->size() != 0) ) {
3621 // validate stmt_list
3622 if ( (stmtList != (uint64_t)-1) && (stmtList < _file->_dwarfDebugLineSect->size()) ) {
3623 const uint8_t* debug_line = (uint8_t*)_file->fileContent() + _file->_dwarfDebugLineSect->offset();
3624 struct line_reader_data* lines = line_open(&debug_line[stmtList],
3625 _file->_dwarfDebugLineSect->size() - stmtList, E::little_endian);
3626 struct line_info result;
3627 Atom<A>* curAtom = NULL;
3628 uint32_t curAtomOffset = 0;
3629 uint32_t curAtomAddress = 0;
3630 uint32_t curAtomSize = 0;
3631 std::map<uint32_t,const char*> dwarfIndexToFile;
3632 if ( lines != NULL ) {
3633 while ( line_next(lines, &result, line_stop_pc) ) {
3634 //fprintf(stderr, "curAtom=%p, result.pc=0x%llX, result.line=%llu, result.end_of_sequence=%d,"
3635 // " curAtomAddress=0x%X, curAtomSize=0x%X\n",
3636 // curAtom, result.pc, result.line, result.end_of_sequence, curAtomAddress, curAtomSize);
3637 // work around weird debug line table compiler generates if no functions in __text section
3638 if ( (curAtom == NULL) && (result.pc == 0) && result.end_of_sequence && (result.file == 1))
3639 continue;
3640 // for performance, see if in next pc is in current atom
3641 if ( (curAtom != NULL) && (curAtomAddress <= result.pc) && (result.pc < (curAtomAddress+curAtomSize)) ) {
3642 curAtomOffset = result.pc - curAtomAddress;
3643 }
3644 // or pc at end of current atom
3645 else if ( result.end_of_sequence && (curAtom != NULL) && (result.pc == (curAtomAddress+curAtomSize)) ) {
3646 curAtomOffset = result.pc - curAtomAddress;
3647 }
3648 // or only one function that is a one line function
3649 else if ( result.end_of_sequence && (curAtom == NULL) && (this->findAtomByAddress(0) != NULL) && (result.pc == this->findAtomByAddress(0)->size()) ) {
3650 curAtom = this->findAtomByAddress(0);
3651 curAtomOffset = result.pc - curAtom->objectAddress();
3652 curAtomAddress = curAtom->objectAddress();
3653 curAtomSize = curAtom->size();
3654 }
3655 else {
3656 // do slow look up of atom by address
3657 try {
3658 curAtom = this->findAtomByAddress(result.pc);
3659 }
3660 catch (...) {
3661 // in case of bug in debug info, don't abort link, just limp on
3662 curAtom = NULL;
3663 }
3664 if ( curAtom == NULL )
3665 break; // file has line info but no functions
3666 if ( result.end_of_sequence && (curAtomAddress+curAtomSize < result.pc) ) {
3667 // a one line function can be returned by line_next() as one entry with pc at end of blob
3668 // look for alt atom starting at end of previous atom
3669 uint32_t previousEnd = curAtomAddress+curAtomSize;
3670 Atom<A>* alt = this->findAtomByAddressOrNullIfStub(previousEnd);
3671 if ( alt == NULL )
3672 continue; // ignore spurious debug info for stubs
3673 if ( result.pc <= alt->objectAddress() + alt->size() ) {
3674 curAtom = alt;
3675 curAtomOffset = result.pc - alt->objectAddress();
3676 curAtomAddress = alt->objectAddress();
3677 curAtomSize = alt->size();
3678 }
3679 else {
3680 curAtomOffset = result.pc - curAtom->objectAddress();
3681 curAtomAddress = curAtom->objectAddress();
3682 curAtomSize = curAtom->size();
3683 }
3684 }
3685 else {
3686 curAtomOffset = result.pc - curAtom->objectAddress();
3687 curAtomAddress = curAtom->objectAddress();
3688 curAtomSize = curAtom->size();
3689 }
3690 }
3691 const char* filename;
3692 std::map<uint32_t,const char*>::iterator pos = dwarfIndexToFile.find(result.file);
3693 if ( pos == dwarfIndexToFile.end() ) {
3694 filename = line_file(lines, result.file);
3695 dwarfIndexToFile[result.file] = filename;
3696 }
3697 else {
3698 filename = pos->second;
3699 }
3700 // only record for ~8000 line info records per function
3701 if ( curAtom->roomForMoreLineInfoCount() ) {
3702 AtomAndLineInfo<A> entry;
3703 entry.atom = curAtom;
3704 entry.info.atomOffset = curAtomOffset;
3705 entry.info.fileName = filename;
3706 entry.info.lineNumber = result.line;
3707 //fprintf(stderr, "addr=0x%08llX, line=%lld, file=%s, atom=%s, atom.size=0x%X, end=%d\n",
3708 // result.pc, result.line, filename, curAtom->name(), curAtomSize, result.end_of_sequence);
3709 entries.push_back(entry);
3710 curAtom->incrementLineInfoCount();
3711 }
3712 if ( result.end_of_sequence ) {
3713 curAtom = NULL;
3714 }
3715 }
3716 line_free(lines);
3717 }
3718 }
3719 }
3720 }
3721
3722 // assign line info start offset for each atom
3723 uint8_t* p = _file->_atomsArray;
3724 uint32_t liOffset = 0;
3725 for(int i=_file->_atomsArrayCount; i > 0; --i) {
3726 Atom<A>* atom = (Atom<A>*)p;
3727 atom->_lineInfoStartIndex = liOffset;
3728 liOffset += atom->_lineInfoCount;
3729 atom->_lineInfoCount = 0;
3730 p += sizeof(Atom<A>);
3731 }
3732 assert(liOffset == entries.size());
3733 _file->_lineInfos.resize(liOffset);
3734
3735 // copy each line info for each atom
3736 for (typename std::vector<AtomAndLineInfo<A> >::iterator it = entries.begin(); it != entries.end(); ++it) {
3737 uint32_t slot = it->atom->_lineInfoStartIndex + it->atom->_lineInfoCount;
3738 _file->_lineInfos[slot] = it->info;
3739 it->atom->_lineInfoCount++;
3740 }
3741
3742 // done with temp vector
3743 entries.clear();
3744 }
3745
3746 template <typename A>
3747 void Parser<A>::parseStabs()
3748 {
3749 // scan symbol table for stabs entries
3750 Atom<A>* currentAtom = NULL;
3751 pint_t currentAtomAddress = 0;
3752 enum { start, inBeginEnd, inFun } state = start;
3753 for (uint32_t symbolIndex = 0; symbolIndex < _symbolCount; ++symbolIndex ) {
3754 const macho_nlist<P>& sym = this->symbolFromIndex(symbolIndex);
3755 bool useStab = true;
3756 uint8_t type = sym.n_type();
3757 const char* symString = (sym.n_strx() != 0) ? this->nameFromSymbol(sym) : NULL;
3758 if ( (type & N_STAB) != 0 ) {
3759 _file->_debugInfoKind = (_hasUUID ? ld::relocatable::File::kDebugInfoStabsUUID : ld::relocatable::File::kDebugInfoStabs);
3760 ld::relocatable::File::Stab stab;
3761 stab.atom = NULL;
3762 stab.type = type;
3763 stab.other = sym.n_sect();
3764 stab.desc = sym.n_desc();
3765 stab.value = sym.n_value();
3766 stab.string = NULL;
3767 switch (state) {
3768 case start:
3769 switch (type) {
3770 case N_BNSYM:
3771 // beginning of function block
3772 state = inBeginEnd;
3773 // fall into case to lookup atom by addresss
3774 case N_LCSYM:
3775 case N_STSYM:
3776 currentAtomAddress = sym.n_value();
3777 currentAtom = this->findAtomByAddress(currentAtomAddress);
3778 if ( currentAtom != NULL ) {
3779 stab.atom = currentAtom;
3780 stab.string = symString;
3781 }
3782 else {
3783 fprintf(stderr, "can't find atom for stabs BNSYM at %08llX in %s",
3784 (uint64_t)sym.n_value(), _path);
3785 }
3786 break;
3787 case N_SO:
3788 case N_OSO:
3789 case N_OPT:
3790 case N_LSYM:
3791 case N_RSYM:
3792 case N_PSYM:
3793 case N_AST:
3794 // not associated with an atom, just copy
3795 stab.string = symString;
3796 break;
3797 case N_GSYM:
3798 {
3799 // n_value field is NOT atom address ;-(
3800 // need to find atom by name match
3801 const char* colon = strchr(symString, ':');
3802 if ( colon != NULL ) {
3803 // build underscore leading name
3804 int nameLen = colon - symString;
3805 char symName[nameLen+2];
3806 strlcpy(&symName[1], symString, nameLen+1);
3807 symName[0] = '_';
3808 symName[nameLen+1] = '\0';
3809 currentAtom = this->findAtomByName(symName);
3810 if ( currentAtom != NULL ) {
3811 stab.atom = currentAtom;
3812 stab.string = symString;
3813 }
3814 }
3815 else {
3816 // might be a debug-note without trailing :G()
3817 currentAtom = this->findAtomByName(symString);
3818 if ( currentAtom != NULL ) {
3819 stab.atom = currentAtom;
3820 stab.string = symString;
3821 }
3822 }
3823 if ( stab.atom == NULL ) {
3824 // ld_classic added bogus GSYM stabs for old style dtrace probes
3825 if ( (strncmp(symString, "__dtrace_probe$", 15) != 0) )
3826 warning("can't find atom for N_GSYM stabs %s in %s", symString, _path);
3827 useStab = false;
3828 }
3829 break;
3830 }
3831 case N_FUN:
3832 if ( isConstFunStabs(symString) ) {
3833 // constant not associated with a function
3834 stab.string = symString;
3835 }
3836 else {
3837 // old style stabs without BNSYM
3838 state = inFun;
3839 currentAtomAddress = sym.n_value();
3840 currentAtom = this->findAtomByAddress(currentAtomAddress);
3841 if ( currentAtom != NULL ) {
3842 stab.atom = currentAtom;
3843 stab.string = symString;
3844 }
3845 else {
3846 warning("can't find atom for stabs FUN at %08llX in %s",
3847 (uint64_t)currentAtomAddress, _path);
3848 }
3849 }
3850 break;
3851 case N_SOL:
3852 case N_SLINE:
3853 stab.string = symString;
3854 // old stabs
3855 break;
3856 case N_BINCL:
3857 case N_EINCL:
3858 case N_EXCL:
3859 stab.string = symString;
3860 // -gfull built .o file
3861 break;
3862 default:
3863 warning("unknown stabs type 0x%X in %s", type, _path);
3864 }
3865 break;
3866 case inBeginEnd:
3867 stab.atom = currentAtom;
3868 switch (type) {
3869 case N_ENSYM:
3870 state = start;
3871 currentAtom = NULL;
3872 break;
3873 case N_LCSYM:
3874 case N_STSYM:
3875 {
3876 Atom<A>* nestedAtom = this->findAtomByAddress(sym.n_value());
3877 if ( nestedAtom != NULL ) {
3878 stab.atom = nestedAtom;
3879 stab.string = symString;
3880 }
3881 else {
3882 warning("can't find atom for stabs 0x%X at %08llX in %s",
3883 type, (uint64_t)sym.n_value(), _path);
3884 }
3885 break;
3886 }
3887 case N_LBRAC:
3888 case N_RBRAC:
3889 case N_SLINE:
3890 // adjust value to be offset in atom
3891 stab.value -= currentAtomAddress;
3892 default:
3893 stab.string = symString;
3894 break;
3895 }
3896 break;
3897 case inFun:
3898 switch (type) {
3899 case N_FUN:
3900 if ( isConstFunStabs(symString) ) {
3901 stab.atom = currentAtom;
3902 stab.string = symString;
3903 }
3904 else {
3905 if ( sym.n_sect() != 0 ) {
3906 // found another start stab, must be really old stabs...
3907 currentAtomAddress = sym.n_value();
3908 currentAtom = this->findAtomByAddress(currentAtomAddress);
3909 if ( currentAtom != NULL ) {
3910 stab.atom = currentAtom;
3911 stab.string = symString;
3912 }
3913 else {
3914 warning("can't find atom for stabs FUN at %08llX in %s",
3915 (uint64_t)currentAtomAddress, _path);
3916 }
3917 }
3918 else {
3919 // found ending stab, switch back to start state
3920 stab.string = symString;
3921 stab.atom = currentAtom;
3922 state = start;
3923 currentAtom = NULL;
3924 }
3925 }
3926 break;
3927 case N_LBRAC:
3928 case N_RBRAC:
3929 case N_SLINE:
3930 // adjust value to be offset in atom
3931 stab.value -= currentAtomAddress;
3932 stab.atom = currentAtom;
3933 break;
3934 case N_SO:
3935 stab.string = symString;
3936 state = start;
3937 break;
3938 default:
3939 stab.atom = currentAtom;
3940 stab.string = symString;
3941 break;
3942 }
3943 break;
3944 }
3945 // add to list of stabs for this .o file
3946 if ( useStab )
3947 _file->_stabs.push_back(stab);
3948 }
3949 }
3950 }
3951
3952
3953
3954 // Look at the compilation unit DIE and determine
3955 // its NAME, compilation directory (in COMP_DIR) and its
3956 // line number information offset (in STMT_LIST). NAME and COMP_DIR
3957 // may be NULL (especially COMP_DIR) if they are not in the .o file;
3958 // STMT_LIST will be (uint64_t) -1.
3959 //
3960 // At present this assumes that there's only one compilation unit DIE.
3961 //
3962 template <typename A>
3963 bool Parser<A>::read_comp_unit(const char ** name, const char ** comp_dir,
3964 uint64_t *stmt_list)
3965 {
3966 const uint8_t * debug_info;
3967 const uint8_t * debug_abbrev;
3968 const uint8_t * di;
3969 const uint8_t * next_cu;
3970 const uint8_t * da;
3971 const uint8_t * end;
3972 const uint8_t * enda;
3973 uint64_t sz;
3974 uint16_t vers;
3975 uint64_t abbrev_base;
3976 uint64_t abbrev;
3977 uint8_t address_size;
3978 bool dwarf64;
3979
3980 *name = NULL;
3981 *comp_dir = NULL;
3982 *stmt_list = (uint64_t) -1;
3983
3984 if ( (_file->_dwarfDebugInfoSect == NULL) || (_file->_dwarfDebugAbbrevSect == NULL) )
3985 return false;
3986
3987 if (_file->_dwarfDebugInfoSect->size() < 12)
3988 /* Too small to be a real debug_info section. */
3989 return false;
3990
3991 debug_info = (uint8_t*)_file->fileContent() + _file->_dwarfDebugInfoSect->offset();
3992 debug_abbrev = (uint8_t*)_file->fileContent() + _file->_dwarfDebugAbbrevSect->offset();
3993 next_cu = debug_info;
3994
3995 while ((uint64_t)(next_cu - debug_info) < _file->_dwarfDebugInfoSect->size()) {
3996 di = next_cu;
3997 sz = A::P::E::get32(*(uint32_t*)di);
3998 di += 4;
3999 dwarf64 = sz == 0xffffffff;
4000 if (dwarf64)
4001 sz = A::P::E::get64(*(uint64_t*)di), di += 8;
4002 else if (sz > 0xffffff00)
4003 /* Unknown dwarf format. */
4004 return false;
4005
4006 /* Verify claimed size. */
4007 if (sz + (di - debug_info) > _file->_dwarfDebugInfoSect->size() || sz <= (dwarf64 ? 23 : 11))
4008 return false;
4009
4010 next_cu = di + sz;
4011
4012 vers = A::P::E::get16(*(uint16_t*)di);
4013 if (vers < 2 || vers > 4)
4014 /* DWARF version wrong for this code.
4015 Chances are we could continue anyway, but we don't know for sure. */
4016 return false;
4017 di += 2;
4018
4019 /* Find the debug_abbrev section. */
4020 abbrev_base = dwarf64 ? A::P::E::get64(*(uint64_t*)di) : A::P::E::get32(*(uint32_t*)di);
4021 di += dwarf64 ? 8 : 4;
4022
4023 if (abbrev_base > _file->_dwarfDebugAbbrevSect->size())
4024 return false;
4025 da = debug_abbrev + abbrev_base;
4026 enda = debug_abbrev + _file->_dwarfDebugAbbrevSect->size();
4027
4028 address_size = *di++;
4029
4030 /* Find the abbrev number we're looking for. */
4031 end = di + sz;
4032 abbrev = read_uleb128 (&di, end);
4033 if (abbrev == (uint64_t) -1)
4034 return false;
4035
4036 /* Skip through the debug_abbrev section looking for that abbrev. */
4037 for (;;)
4038 {
4039 uint64_t this_abbrev = read_uleb128 (&da, enda);
4040 uint64_t attr;
4041
4042 if (this_abbrev == abbrev)
4043 /* This is almost always taken. */
4044 break;
4045 skip_leb128 (&da, enda); /* Skip the tag. */
4046 if (da == enda)
4047 return false;
4048 da++; /* Skip the DW_CHILDREN_* value. */
4049
4050 do {
4051 attr = read_uleb128 (&da, enda);
4052 skip_leb128 (&da, enda);
4053 } while (attr != 0 && attr != (uint64_t) -1);
4054 if (attr != 0)
4055 return false;
4056 }
4057
4058 /* Check that the abbrev is one for a DW_TAG_compile_unit. */
4059 if (read_uleb128 (&da, enda) != DW_TAG_compile_unit)
4060 return false;
4061 if (da == enda)
4062 return false;
4063 da++; /* Skip the DW_CHILDREN_* value. */
4064
4065 /* Now, go through the DIE looking for DW_AT_name,
4066 DW_AT_comp_dir, and DW_AT_stmt_list. */
4067 bool skip_to_next_cu = false;
4068 while (!skip_to_next_cu) {
4069
4070 uint64_t attr = read_uleb128 (&da, enda);
4071 uint64_t form = read_uleb128 (&da, enda);
4072
4073 if (attr == (uint64_t) -1)
4074 return false;
4075 else if (attr == 0)
4076 return true;
4077 if (form == DW_FORM_indirect)
4078 form = read_uleb128 (&di, end);
4079
4080 switch (attr) {
4081 case DW_AT_name:
4082 *name = getDwarfString(form, di);
4083 /* Swift object files may contain two CUs: One
4084 describes the Swift code, one is created by the
4085 clang importer. Skip over the CU created by the
4086 clang importer as it may be empty. */
4087 if (std::string(*name) == "<swift-imported-modules>")
4088 skip_to_next_cu = true;
4089 break;
4090 case DW_AT_comp_dir:
4091 *comp_dir = getDwarfString(form, di);
4092 break;
4093 case DW_AT_stmt_list:
4094 *stmt_list = getDwarfOffset(form, di, dwarf64);
4095 break;
4096 default:
4097 if (! skip_form (&di, end, form, address_size, dwarf64))
4098 return false;
4099 }
4100 }
4101 }
4102 return false;
4103 }
4104
4105
4106
4107 template <typename A>
4108 File<A>::~File()
4109 {
4110 free(_sectionsArray);
4111 free(_atomsArray);
4112 }
4113
4114 template <typename A>
4115 const char* File<A>::translationUnitSource() const
4116 {
4117 return _dwarfTranslationUnitPath;
4118 }
4119
4120 template <typename A>
4121 bool File<A>::forEachAtom(ld::File::AtomHandler& handler) const
4122 {
4123 handler.doFile(*this);
4124 uint8_t* p = _atomsArray;
4125 for(int i=_atomsArrayCount; i > 0; --i) {
4126 handler.doAtom(*((Atom<A>*)p));
4127 p += sizeof(Atom<A>);
4128 }
4129 p = _aliasAtomsArray;
4130 for(int i=_aliasAtomsArrayCount; i > 0; --i) {
4131 handler.doAtom(*((AliasAtom*)p));
4132 p += sizeof(AliasAtom);
4133 }
4134
4135 return (_atomsArrayCount != 0) || (_aliasAtomsArrayCount != 0);
4136 }
4137
4138 template <typename A>
4139 const char* Section<A>::makeSegmentName(const macho_section<typename A::P>* sect)
4140 {
4141 // mach-o section record only has room for 16-byte seg/sect names
4142 // so a 16-byte name has no trailing zero
4143 const char* name = sect->segname();
4144 if ( strlen(name) < 16 )
4145 return name;
4146 char* tmp = new char[17];
4147 strlcpy(tmp, name, 17);
4148 return tmp;
4149 }
4150
4151 template <typename A>
4152 const char* Section<A>::makeSectionName(const macho_section<typename A::P>* sect)
4153 {
4154 const char* name = sect->sectname();
4155 if ( strlen(name) < 16 )
4156 return name;
4157
4158 // special case common long section names so we don't have to malloc
4159 if ( strncmp(sect->sectname(), "__objc_classrefs", 16) == 0 )
4160 return "__objc_classrefs";
4161 if ( strncmp(sect->sectname(), "__objc_classlist", 16) == 0 )
4162 return "__objc_classlist";
4163 if ( strncmp(sect->sectname(), "__objc_nlclslist", 16) == 0 )
4164 return "__objc_nlclslist";
4165 if ( strncmp(sect->sectname(), "__objc_nlcatlist", 16) == 0 )
4166 return "__objc_nlcatlist";
4167 if ( strncmp(sect->sectname(), "__objc_protolist", 16) == 0 )
4168 return "__objc_protolist";
4169 if ( strncmp(sect->sectname(), "__objc_protorefs", 16) == 0 )
4170 return "__objc_protorefs";
4171 if ( strncmp(sect->sectname(), "__objc_superrefs", 16) == 0 )
4172 return "__objc_superrefs";
4173 if ( strncmp(sect->sectname(), "__objc_imageinfo", 16) == 0 )
4174 return "__objc_imageinfo";
4175 if ( strncmp(sect->sectname(), "__objc_stringobj", 16) == 0 )
4176 return "__objc_stringobj";
4177 if ( strncmp(sect->sectname(), "__gcc_except_tab", 16) == 0 )
4178 return "__gcc_except_tab";
4179
4180 char* tmp = new char[17];
4181 strlcpy(tmp, name, 17);
4182 return tmp;
4183 }
4184
4185 template <typename A>
4186 bool Section<A>::readable(const macho_section<typename A::P>* sect)
4187 {
4188 return true;
4189 }
4190
4191 template <typename A>
4192 bool Section<A>::writable(const macho_section<typename A::P>* sect)
4193 {
4194 // mach-o .o files do not contain segment permissions
4195 // we just know TEXT is special
4196 return ( strcmp(sect->segname(), "__TEXT") != 0 );
4197 }
4198
4199 template <typename A>
4200 bool Section<A>::exectuable(const macho_section<typename A::P>* sect)
4201 {
4202 // mach-o .o files do not contain segment permissions
4203 // we just know TEXT is special
4204 return ( strcmp(sect->segname(), "__TEXT") == 0 );
4205 }
4206
4207
4208 template <typename A>
4209 ld::Section::Type Section<A>::sectionType(const macho_section<typename A::P>* sect)
4210 {
4211 switch ( sect->flags() & SECTION_TYPE ) {
4212 case S_ZEROFILL:
4213 return ld::Section::typeZeroFill;
4214 case S_CSTRING_LITERALS:
4215 if ( (strcmp(sect->sectname(), "__cstring") == 0) && (strcmp(sect->segname(), "__TEXT") == 0) )
4216 return ld::Section::typeCString;
4217 else
4218 return ld::Section::typeNonStdCString;
4219 case S_4BYTE_LITERALS:
4220 return ld::Section::typeLiteral4;
4221 case S_8BYTE_LITERALS:
4222 return ld::Section::typeLiteral8;
4223 case S_LITERAL_POINTERS:
4224 return ld::Section::typeCStringPointer;
4225 case S_NON_LAZY_SYMBOL_POINTERS:
4226 return ld::Section::typeNonLazyPointer;
4227 case S_LAZY_SYMBOL_POINTERS:
4228 return ld::Section::typeLazyPointer;
4229 case S_SYMBOL_STUBS:
4230 return ld::Section::typeStub;
4231 case S_MOD_INIT_FUNC_POINTERS:
4232 return ld::Section::typeInitializerPointers;
4233 case S_MOD_TERM_FUNC_POINTERS:
4234 return ld::Section::typeTerminatorPointers;
4235 case S_INTERPOSING:
4236 return ld::Section::typeUnclassified;
4237 case S_16BYTE_LITERALS:
4238 return ld::Section::typeLiteral16;
4239 case S_REGULAR:
4240 case S_COALESCED:
4241 if ( sect->flags() & S_ATTR_PURE_INSTRUCTIONS ) {
4242 return ld::Section::typeCode;
4243 }
4244 else if ( strcmp(sect->segname(), "__TEXT") == 0 ) {
4245 if ( strcmp(sect->sectname(), "__eh_frame") == 0 )
4246 return ld::Section::typeCFI;
4247 else if ( strcmp(sect->sectname(), "__ustring") == 0 )
4248 return ld::Section::typeUTF16Strings;
4249 else if ( strcmp(sect->sectname(), "__textcoal_nt") == 0 )
4250 return ld::Section::typeCode;
4251 else if ( strcmp(sect->sectname(), "__StaticInit") == 0 )
4252 return ld::Section::typeCode;
4253 else if ( strcmp(sect->sectname(), "__constructor") == 0 )
4254 return ld::Section::typeInitializerPointers;
4255 }
4256 else if ( strcmp(sect->segname(), "__DATA") == 0 ) {
4257 if ( strcmp(sect->sectname(), "__cfstring") == 0 )
4258 return ld::Section::typeCFString;
4259 else if ( strcmp(sect->sectname(), "__dyld") == 0 )
4260 return ld::Section::typeDyldInfo;
4261 else if ( strcmp(sect->sectname(), "__program_vars") == 0 )
4262 return ld::Section::typeDyldInfo;
4263 else if ( strncmp(sect->sectname(), "__objc_classrefs", 16) == 0 )
4264 return ld::Section::typeObjCClassRefs;
4265 else if ( strcmp(sect->sectname(), "__objc_catlist") == 0 )
4266 return ld::Section::typeObjC2CategoryList;
4267 }
4268 else if ( strcmp(sect->segname(), "__OBJC") == 0 ) {
4269 if ( strcmp(sect->sectname(), "__class") == 0 )
4270 return ld::Section::typeObjC1Classes;
4271 }
4272 break;
4273 case S_THREAD_LOCAL_REGULAR:
4274 return ld::Section::typeTLVInitialValues;
4275 case S_THREAD_LOCAL_ZEROFILL:
4276 return ld::Section::typeTLVZeroFill;
4277 case S_THREAD_LOCAL_VARIABLES:
4278 return ld::Section::typeTLVDefs;
4279 case S_THREAD_LOCAL_VARIABLE_POINTERS:
4280 return ld::Section::typeTLVPointers;
4281 case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
4282 return ld::Section::typeTLVInitializerPointers;
4283 }
4284 return ld::Section::typeUnclassified;
4285 }
4286
4287
4288 template <typename A>
4289 Atom<A>* Section<A>::findContentAtomByAddress(pint_t addr, class Atom<A>* start, class Atom<A>* end)
4290 {
4291 // do a binary search of atom array
4292 uint32_t atomCount = end - start;
4293 Atom<A>* base = start;
4294 for (uint32_t n = atomCount; n > 0; n /= 2) {
4295 Atom<A>* pivot = &base[n/2];
4296 pint_t atomStartAddr = pivot->_objAddress;
4297 pint_t atomEndAddr = atomStartAddr + pivot->_size;
4298 if ( atomStartAddr <= addr ) {
4299 // address in normal atom
4300 if (addr < atomEndAddr)
4301 return pivot;
4302 // address in "end" label (but not in alias)
4303 if ( (pivot->_size == 0) && (addr == atomEndAddr) && !pivot->isAlias() )
4304 return pivot;
4305 }
4306 if ( addr >= atomEndAddr ) {
4307 // key > pivot
4308 // move base to atom after pivot
4309 base = &pivot[1];
4310 --n;
4311 }
4312 else {
4313 // key < pivot
4314 // keep same base
4315 }
4316 }
4317 return NULL;
4318 }
4319
4320 template <typename A>
4321 ld::Atom::Alignment Section<A>::alignmentForAddress(pint_t addr)
4322 {
4323 const uint32_t sectionAlignment = this->_machOSection->align();
4324 uint32_t modulus = (addr % (1 << sectionAlignment));
4325 if ( modulus > 0xFFFF )
4326 warning("alignment for symbol at address 0x%08llX in %s exceeds 2^16", (uint64_t)addr, this->file().path());
4327 return ld::Atom::Alignment(sectionAlignment, modulus);
4328 }
4329
4330 template <typename A>
4331 uint32_t Section<A>::sectionNum(class Parser<A>& parser) const
4332 {
4333 if ( _machOSection == NULL )
4334 return 0;
4335 else
4336 return 1 + (this->_machOSection - parser.firstMachOSection());
4337 }
4338
4339 // arm does not have zero cost exceptions
4340 template <>
4341 uint32_t CFISection<arm>::cfiCount(Parser<arm>& parser)
4342 {
4343 if ( parser.armUsesZeroCostExceptions() ) {
4344 // create ObjectAddressSpace object for use by libunwind
4345 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4346 return libunwind::CFI_Parser<OAS>::getCFICount(oas,
4347 this->_machOSection->addr(), this->_machOSection->size());
4348 }
4349 return 0;
4350 }
4351
4352 template <typename A>
4353 uint32_t CFISection<A>::cfiCount(Parser<A>& parser)
4354 {
4355 // create ObjectAddressSpace object for use by libunwind
4356 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4357 return libunwind::CFI_Parser<OAS>::getCFICount(oas,
4358 this->_machOSection->addr(), this->_machOSection->size());
4359 }
4360
4361 template <typename A>
4362 void CFISection<A>::warnFunc(void* ref, uint64_t funcAddr, const char* msg)
4363 {
4364 Parser<A>* parser = (Parser<A>*)ref;
4365 if ( ! parser->warnUnwindConversionProblems() )
4366 return;
4367 if ( funcAddr != CFI_INVALID_ADDRESS ) {
4368 // atoms are not constructed yet, so scan symbol table for labels
4369 const char* name = parser->scanSymbolTableForAddress(funcAddr);
4370 warning("could not create compact unwind for %s: %s", name, msg);
4371 }
4372 else {
4373 warning("could not create compact unwind: %s", msg);
4374 }
4375 }
4376
4377 template <>
4378 bool CFISection<x86_64>::needsRelocating()
4379 {
4380 return true;
4381 }
4382
4383 template <>
4384 bool CFISection<arm64>::needsRelocating()
4385 {
4386 return true;
4387 }
4388
4389 template <typename A>
4390 bool CFISection<A>::needsRelocating()
4391 {
4392 return false;
4393 }
4394
4395 template <>
4396 void CFISection<x86_64>::cfiParse(class Parser<x86_64>& parser, uint8_t* buffer,
4397 libunwind::CFI_Atom_Info<CFISection<x86_64>::OAS>::CFI_Atom_Info cfiArray[],
4398 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4399 {
4400 const uint32_t sectionSize = this->_machOSection->size();
4401 // copy __eh_frame data to buffer
4402 memcpy(buffer, file().fileContent() + this->_machOSection->offset(), sectionSize);
4403
4404 // and apply relocations
4405 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(file().fileContent() + this->_machOSection->reloff());
4406 const macho_relocation_info<P>* relocsEnd = &relocs[this->_machOSection->nreloc()];
4407 for (const macho_relocation_info<P>* reloc = relocs; reloc < relocsEnd; ++reloc) {
4408 uint64_t value = 0;
4409 switch ( reloc->r_type() ) {
4410 case X86_64_RELOC_SUBTRACTOR:
4411 value = 0 - parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4412 ++reloc;
4413 if ( reloc->r_extern() )
4414 value += parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4415 break;
4416 case X86_64_RELOC_UNSIGNED:
4417 value = parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4418 break;
4419 case X86_64_RELOC_GOT:
4420 // this is used for the reference to the personality function in CIEs
4421 // store the symbol number of the personality function for later use as a Fixup
4422 value = reloc->r_symbolnum();
4423 break;
4424 default:
4425 fprintf(stderr, "CFISection::cfiParse() unexpected relocation type at r_address=0x%08X\n", reloc->r_address());
4426 break;
4427 }
4428 if ( reloc->r_address() > sectionSize )
4429 throwf("malformed __eh_frame relocation, offset (0x%08X) is beyond end of section,", reloc->r_address());
4430 uint64_t* p64;
4431 uint32_t* p32;
4432 switch ( reloc->r_length() ) {
4433 case 3:
4434 p64 = (uint64_t*)&buffer[reloc->r_address()];
4435 E::set64(*p64, value + E::get64(*p64));
4436 break;
4437 case 2:
4438 p32 = (uint32_t*)&buffer[reloc->r_address()];
4439 E::set32(*p32, value + E::get32(*p32));
4440 break;
4441 default:
4442 fprintf(stderr, "CFISection::cfiParse() unexpected relocation size at r_address=0x%08X\n", reloc->r_address());
4443 break;
4444 }
4445 }
4446
4447 // create ObjectAddressSpace object for use by libunwind
4448 OAS oas(*this, buffer);
4449
4450 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4451 const char* msg;
4452 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_x86_64>::parseCFIs(
4453 oas, this->_machOSection->addr(), this->_machOSection->size(),
4454 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4455 cfiArray, count, (void*)&parser, warnFunc);
4456 if ( msg != NULL )
4457 throwf("malformed __eh_frame section: %s", msg);
4458 }
4459
4460 template <>
4461 void CFISection<x86>::cfiParse(class Parser<x86>& parser, uint8_t* buffer,
4462 libunwind::CFI_Atom_Info<CFISection<x86>::OAS>::CFI_Atom_Info cfiArray[],
4463 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4464 {
4465 // create ObjectAddressSpace object for use by libunwind
4466 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4467
4468 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4469 const char* msg;
4470 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_x86>::parseCFIs(
4471 oas, this->_machOSection->addr(), this->_machOSection->size(),
4472 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4473 cfiArray, count, (void*)&parser, warnFunc);
4474 if ( msg != NULL )
4475 throwf("malformed __eh_frame section: %s", msg);
4476 }
4477
4478
4479
4480
4481 template <>
4482 void CFISection<arm>::cfiParse(class Parser<arm>& parser, uint8_t* buffer,
4483 libunwind::CFI_Atom_Info<CFISection<arm>::OAS>::CFI_Atom_Info cfiArray[],
4484 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4485 {
4486 if ( !parser.armUsesZeroCostExceptions() ) {
4487 // most arm do not use zero cost exceptions
4488 assert(count == 0);
4489 return;
4490 }
4491 // create ObjectAddressSpace object for use by libunwind
4492 OAS oas(*this, (uint8_t*)this->file().fileContent()+this->_machOSection->offset());
4493
4494 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4495 const char* msg;
4496 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_arm>::parseCFIs(
4497 oas, this->_machOSection->addr(), this->_machOSection->size(),
4498 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4499 cfiArray, count, (void*)&parser, warnFunc);
4500 if ( msg != NULL )
4501 throwf("malformed __eh_frame section: %s", msg);
4502 }
4503
4504
4505
4506
4507 template <>
4508 void CFISection<arm64>::cfiParse(class Parser<arm64>& parser, uint8_t* buffer,
4509 libunwind::CFI_Atom_Info<CFISection<arm64>::OAS>::CFI_Atom_Info cfiArray[],
4510 uint32_t& count, const pint_t cuStarts[], uint32_t cuCount)
4511 {
4512 // copy __eh_frame data to buffer
4513 const uint32_t sectionSize = this->_machOSection->size();
4514 memcpy(buffer, file().fileContent() + this->_machOSection->offset(), sectionSize);
4515
4516 // and apply relocations
4517 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(file().fileContent() + this->_machOSection->reloff());
4518 const macho_relocation_info<P>* relocsEnd = &relocs[this->_machOSection->nreloc()];
4519 for (const macho_relocation_info<P>* reloc = relocs; reloc < relocsEnd; ++reloc) {
4520 uint64_t* p64 = (uint64_t*)&buffer[reloc->r_address()];
4521 uint32_t* p32 = (uint32_t*)&buffer[reloc->r_address()];
4522 uint32_t addend32 = E::get32(*p32);
4523 uint64_t addend64 = E::get64(*p64);
4524 uint64_t value = 0;
4525 switch ( reloc->r_type() ) {
4526 case ARM64_RELOC_SUBTRACTOR:
4527 value = 0 - parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4528 ++reloc;
4529 if ( reloc->r_extern() )
4530 value += parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4531 break;
4532 case ARM64_RELOC_UNSIGNED:
4533 value = parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
4534 break;
4535 case ARM64_RELOC_POINTER_TO_GOT:
4536 // this is used for the reference to the personality function in CIEs
4537 // store the symbol number of the personality function for later use as a Fixup
4538 value = reloc->r_symbolnum();
4539 addend32 = 0;
4540 addend64 = 0;
4541 break;
4542 default:
4543 fprintf(stderr, "CFISection::cfiParse() unexpected relocation type at r_address=0x%08X\n", reloc->r_address());
4544 break;
4545 }
4546 if ( reloc->r_address() > sectionSize )
4547 throwf("malformed __eh_frame relocation, offset (0x%08X) is beyond end of section,", reloc->r_address());
4548 switch ( reloc->r_length() ) {
4549 case 3:
4550 E::set64(*p64, value + addend64);
4551 break;
4552 case 2:
4553 E::set32(*p32, value + addend32);
4554 break;
4555 default:
4556 fprintf(stderr, "CFISection::cfiParse() unexpected relocation size at r_address=0x%08X\n", reloc->r_address());
4557 break;
4558 }
4559 }
4560
4561
4562 // create ObjectAddressSpace object for use by libunwind
4563 OAS oas(*this, buffer);
4564
4565 // use libuwind to parse __eh_frame data into array of CFI_Atom_Info
4566 const char* msg;
4567 msg = libunwind::DwarfInstructions<OAS, libunwind::Registers_arm64>::parseCFIs(
4568 oas, this->_machOSection->addr(), this->_machOSection->size(),
4569 cuStarts, cuCount, parser.keepDwarfUnwind(), parser.forceDwarfConversion(), parser.neverConvertDwarf(),
4570 cfiArray, count, (void*)&parser, warnFunc);
4571 if ( msg != NULL )
4572 throwf("malformed __eh_frame section: %s", msg);
4573 }
4574
4575
4576 template <typename A>
4577 uint32_t CFISection<A>::computeAtomCount(class Parser<A>& parser,
4578 struct Parser<A>::LabelAndCFIBreakIterator& it,
4579 const struct Parser<A>::CFI_CU_InfoArrays& cfis)
4580 {
4581 return cfis.cfiCount;
4582 }
4583
4584
4585
4586 template <typename A>
4587 uint32_t CFISection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
4588 struct Parser<A>::LabelAndCFIBreakIterator& it,
4589 const struct Parser<A>::CFI_CU_InfoArrays& cfis)
4590 {
4591 this->_beginAtoms = (Atom<A>*)p;
4592 // walk CFI_Atom_Info array and create atom for each entry
4593 const CFI_Atom_Info* start = &cfis.cfiArray[0];
4594 const CFI_Atom_Info* end = &cfis.cfiArray[cfis.cfiCount];
4595 for(const CFI_Atom_Info* a=start; a < end; ++a) {
4596 Atom<A>* space = (Atom<A>*)p;
4597 new (space) Atom<A>(*this, (a->isCIE ? "CIE" : "FDE"), a->address, a->size,
4598 ld::Atom::definitionRegular, ld::Atom::combineNever, ld::Atom::scopeTranslationUnit,
4599 ld::Atom::typeCFI, ld::Atom::symbolTableNotInFinalLinkedImages,
4600 false, false, false, ld::Atom::Alignment(0));
4601 p += sizeof(Atom<A>);
4602 }
4603 this->_endAtoms = (Atom<A>*)p;
4604 return cfis.cfiCount;
4605 }
4606
4607
4608 template <> bool CFISection<x86_64>::bigEndian() { return false; }
4609 template <> bool CFISection<x86>::bigEndian() { return false; }
4610 template <> bool CFISection<arm>::bigEndian() { return false; }
4611 template <> bool CFISection<arm64>::bigEndian() { return false; }
4612
4613
4614 template <>
4615 void CFISection<x86_64>::addCiePersonalityFixups(class Parser<x86_64>& parser, const CFI_Atom_Info* cieInfo)
4616 {
4617 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4618 if ( personalityEncoding == 0x9B ) {
4619 // compiler always produces X86_64_RELOC_GOT with addend of 4 to personality function
4620 // CFISection<x86_64>::cfiParse() set targetAddress to be symbolIndex + 4 + addressInCIE
4621 uint32_t symbolIndex = cieInfo->u.cieInfo.personality.targetAddress - 4
4622 - cieInfo->address - cieInfo->u.cieInfo.personality.offsetInCFI;
4623 const macho_nlist<P>& sym = parser.symbolFromIndex(symbolIndex);
4624 const char* personalityName = parser.nameFromSymbol(sym);
4625
4626 Atom<x86_64>* cieAtom = this->findAtomByAddress(cieInfo->address);
4627 Parser<x86_64>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4628 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, false, personalityName);
4629 parser.addFixup(src, ld::Fixup::k2of3, ld::Fixup::kindAddAddend, 4);
4630 parser.addFixup(src, ld::Fixup::k3of3, ld::Fixup::kindStoreX86PCRel32GOT);
4631 }
4632 else if ( personalityEncoding != 0 ) {
4633 throwf("unsupported address encoding (%02X) of personality function in CIE",
4634 personalityEncoding);
4635 }
4636 }
4637
4638 template <>
4639 void CFISection<x86>::addCiePersonalityFixups(class Parser<x86>& parser, const CFI_Atom_Info* cieInfo)
4640 {
4641 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4642 if ( (personalityEncoding == 0x9B) || (personalityEncoding == 0x90) ) {
4643 uint32_t offsetInCFI = cieInfo->u.cieInfo.personality.offsetInCFI;
4644 uint32_t nlpAddr = cieInfo->u.cieInfo.personality.targetAddress;
4645 Atom<x86>* cieAtom = this->findAtomByAddress(cieInfo->address);
4646 Atom<x86>* nlpAtom = parser.findAtomByAddress(nlpAddr);
4647 assert(nlpAtom->contentType() == ld::Atom::typeNonLazyPointer);
4648 Parser<x86>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4649
4650 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, nlpAtom);
4651 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, cieAtom);
4652 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, offsetInCFI);
4653 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
4654 }
4655 else if ( personalityEncoding != 0 ) {
4656 throwf("unsupported address encoding (%02X) of personality function in CIE", personalityEncoding);
4657 }
4658 }
4659
4660 #if SUPPORT_ARCH_arm64
4661 template <>
4662 void CFISection<arm64>::addCiePersonalityFixups(class Parser<arm64>& parser, const CFI_Atom_Info* cieInfo)
4663 {
4664 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4665 if ( personalityEncoding == 0x9B ) {
4666 // compiler always produces ARM64_RELOC_GOT r_pcrel=1 to personality function
4667 // CFISection<arm64>::cfiParse() set targetAddress to be symbolIndex + addressInCIE
4668 uint32_t symbolIndex = cieInfo->u.cieInfo.personality.targetAddress
4669 - cieInfo->address - cieInfo->u.cieInfo.personality.offsetInCFI;
4670 const macho_nlist<P>& sym = parser.symbolFromIndex(symbolIndex);
4671 const char* personalityName = parser.nameFromSymbol(sym);
4672
4673 Atom<arm64>* cieAtom = this->findAtomByAddress(cieInfo->address);
4674 Parser<arm64>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4675 parser.addFixup(src, ld::Fixup::k1of2, ld::Fixup::kindSetTargetAddress, false, personalityName);
4676 parser.addFixup(src, ld::Fixup::k2of2, ld::Fixup::kindStoreARM64PCRelToGOT);
4677 }
4678 else if ( personalityEncoding != 0 ) {
4679 throwf("unsupported address encoding (%02X) of personality function in CIE",
4680 personalityEncoding);
4681 }
4682 }
4683 #endif
4684
4685 template <>
4686 void CFISection<arm>::addCiePersonalityFixups(class Parser<arm>& parser, const CFI_Atom_Info* cieInfo)
4687 {
4688 uint8_t personalityEncoding = cieInfo->u.cieInfo.personality.encodingOfTargetAddress;
4689 if ( (personalityEncoding == 0x9B) || (personalityEncoding == 0x90) ) {
4690 uint32_t offsetInCFI = cieInfo->u.cieInfo.personality.offsetInCFI;
4691 uint32_t nlpAddr = cieInfo->u.cieInfo.personality.targetAddress;
4692 Atom<arm>* cieAtom = this->findAtomByAddress(cieInfo->address);
4693 Atom<arm>* nlpAtom = parser.findAtomByAddress(nlpAddr);
4694 assert(nlpAtom->contentType() == ld::Atom::typeNonLazyPointer);
4695 Parser<arm>::SourceLocation src(cieAtom, cieInfo->u.cieInfo.personality.offsetInCFI);
4696
4697 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, nlpAtom);
4698 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, cieAtom);
4699 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, offsetInCFI);
4700 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
4701 }
4702 else if ( personalityEncoding != 0 ) {
4703 throwf("unsupported address encoding (%02X) of personality function in CIE", personalityEncoding);
4704 }
4705 }
4706
4707
4708
4709 template <typename A>
4710 void CFISection<A>::addCiePersonalityFixups(class Parser<A>& parser, const CFI_Atom_Info* cieInfo)
4711 {
4712 assert(0 && "addCiePersonalityFixups() not implemented for arch");
4713 }
4714
4715 template <typename A>
4716 void CFISection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays& cfis)
4717 {
4718 ld::Fixup::Kind store32 = bigEndian() ? ld::Fixup::kindStoreBigEndian32 : ld::Fixup::kindStoreLittleEndian32;
4719 ld::Fixup::Kind store64 = bigEndian() ? ld::Fixup::kindStoreBigEndian64 : ld::Fixup::kindStoreLittleEndian64;
4720
4721 // add all references for FDEs, including implicit group references
4722 const CFI_Atom_Info* end = &cfis.cfiArray[cfis.cfiCount];
4723 for(const CFI_Atom_Info* p = &cfis.cfiArray[0]; p < end; ++p) {
4724 if ( p->isCIE ) {
4725 // add reference to personality function if used
4726 if ( p->u.cieInfo.personality.targetAddress != CFI_INVALID_ADDRESS ) {
4727 this->addCiePersonalityFixups(parser, p);
4728 }
4729 }
4730 else {
4731 // find FDE Atom
4732 Atom<A>* fdeAtom = this->findAtomByAddress(p->address);
4733 // find function Atom
4734 Atom<A>* functionAtom = parser.findAtomByAddress(p->u.fdeInfo.function.targetAddress);
4735 // find CIE Atom
4736 Atom<A>* cieAtom = this->findAtomByAddress(p->u.fdeInfo.cie.targetAddress);
4737 // find LSDA Atom
4738 Atom<A>* lsdaAtom = NULL;
4739 if ( p->u.fdeInfo.lsda.targetAddress != CFI_INVALID_ADDRESS ) {
4740 lsdaAtom = parser.findAtomByAddress(p->u.fdeInfo.lsda.targetAddress);
4741 }
4742 // add reference from FDE to CIE (always 32-bit pc-rel)
4743 typename Parser<A>::SourceLocation fdeToCieSrc(fdeAtom, p->u.fdeInfo.cie.offsetInCFI);
4744 parser.addFixup(fdeToCieSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, fdeAtom);
4745 parser.addFixup(fdeToCieSrc, ld::Fixup::k2of4, ld::Fixup::kindAddAddend, p->u.fdeInfo.cie.offsetInCFI);
4746 parser.addFixup(fdeToCieSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, cieAtom);
4747 parser.addFixup(fdeToCieSrc, ld::Fixup::k4of4, store32, cieAtom);
4748
4749 // add reference from FDE to function
4750 typename Parser<A>::SourceLocation fdeToFuncSrc(fdeAtom, p->u.fdeInfo.function.offsetInCFI);
4751 switch (p->u.fdeInfo.function.encodingOfTargetAddress) {
4752 case DW_EH_PE_pcrel|DW_EH_PE_ptr:
4753 if ( sizeof(typename A::P::uint_t) == 8 ) {
4754 parser.addFixup(fdeToFuncSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, functionAtom);
4755 parser.addFixup(fdeToFuncSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4756 parser.addFixup(fdeToFuncSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.function.offsetInCFI);
4757 parser.addFixup(fdeToFuncSrc, ld::Fixup::k4of4, store64);
4758 break;
4759 }
4760 // else fall into 32-bit case
4761 case DW_EH_PE_pcrel|DW_EH_PE_sdata4:
4762 parser.addFixup(fdeToFuncSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, functionAtom);
4763 parser.addFixup(fdeToFuncSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4764 parser.addFixup(fdeToFuncSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.function.offsetInCFI);
4765 parser.addFixup(fdeToFuncSrc, ld::Fixup::k4of4, store32);
4766 break;
4767 default:
4768 throw "unsupported encoding in FDE of pointer to function";
4769 }
4770
4771 // add reference from FDE to LSDA
4772 typename Parser<A>::SourceLocation fdeToLsdaSrc(fdeAtom, p->u.fdeInfo.lsda.offsetInCFI);
4773 if ( lsdaAtom != NULL ) {
4774 switch (p->u.fdeInfo.lsda.encodingOfTargetAddress) {
4775 case DW_EH_PE_pcrel|DW_EH_PE_ptr:
4776 if ( sizeof(typename A::P::uint_t) == 8 ) {
4777 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, lsdaAtom);
4778 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4779 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.lsda.offsetInCFI);
4780 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k4of4, store64);
4781 break;
4782 }
4783 // else fall into 32-bit case
4784 case DW_EH_PE_pcrel|DW_EH_PE_sdata4:
4785 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, lsdaAtom);
4786 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k2of4, ld::Fixup::kindSubtractTargetAddress, fdeAtom);
4787 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k3of4, ld::Fixup::kindSubtractAddend, p->u.fdeInfo.lsda.offsetInCFI);
4788 parser.addFixup(fdeToLsdaSrc, ld::Fixup::k4of4, store32);
4789 break;
4790 default:
4791 throw "unsupported encoding in FDE of pointer to LSDA";
4792 }
4793 }
4794
4795 // FDE is in group lead by function atom
4796 typename Parser<A>::SourceLocation fdeSrc(functionAtom,0);
4797 parser.addFixup(fdeSrc, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateFDE, fdeAtom);
4798
4799 // LSDA is in group lead by function atom
4800 if ( lsdaAtom != NULL ) {
4801 parser.addFixup(fdeSrc, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, lsdaAtom);
4802 }
4803 }
4804 }
4805 }
4806
4807
4808
4809
4810 template <typename A>
4811 const void* CFISection<A>::OAS::mappedAddress(pint_t addr)
4812 {
4813 if ( (_ehFrameStartAddr <= addr) && (addr < _ehFrameEndAddr) )
4814 return &_ehFrameContent[addr-_ehFrameStartAddr];
4815 else {
4816 // requested bytes are not in __eh_frame section
4817 // this can occur when examining the instruction bytes in the __text
4818 File<A>& file = _ehFrameSection.file();
4819 for (uint32_t i=0; i < file._sectionsArrayCount; ++i ) {
4820 const macho_section<typename A::P>* sect = file._sectionsArray[i]->machoSection();
4821 // TentativeDefinitionSection and AbsoluteSymbolSection have no mach-o section
4822 if ( sect != NULL ) {
4823 if ( (sect->addr() <= addr) && (addr < (sect->addr()+sect->size())) ) {
4824 return file.fileContent() + sect->offset() + addr - sect->addr();
4825 }
4826 }
4827 }
4828 throwf("__eh_frame parsing problem. Can't find target of reference to address 0x%08llX", (uint64_t)addr);
4829 }
4830 }
4831
4832
4833 template <typename A>
4834 uint64_t CFISection<A>::OAS::getULEB128(pint_t& logicalAddr, pint_t end)
4835 {
4836 uintptr_t size = (end - logicalAddr);
4837 libunwind::LocalAddressSpace::pint_t laddr = (libunwind::LocalAddressSpace::pint_t)mappedAddress(logicalAddr);
4838 libunwind::LocalAddressSpace::pint_t sladdr = laddr;
4839 uint64_t result = libunwind::LocalAddressSpace::getULEB128(laddr, laddr+size);
4840 logicalAddr += (laddr-sladdr);
4841 return result;
4842 }
4843
4844 template <typename A>
4845 int64_t CFISection<A>::OAS::getSLEB128(pint_t& logicalAddr, pint_t end)
4846 {
4847 uintptr_t size = (end - logicalAddr);
4848 libunwind::LocalAddressSpace::pint_t laddr = (libunwind::LocalAddressSpace::pint_t)mappedAddress(logicalAddr);
4849 libunwind::LocalAddressSpace::pint_t sladdr = laddr;
4850 int64_t result = libunwind::LocalAddressSpace::getSLEB128(laddr, laddr+size);
4851 logicalAddr += (laddr-sladdr);
4852 return result;
4853 }
4854
4855 template <typename A>
4856 typename A::P::uint_t CFISection<A>::OAS::getEncodedP(pint_t& addr, pint_t end, uint8_t encoding)
4857 {
4858 pint_t startAddr = addr;
4859 pint_t p = addr;
4860 pint_t result;
4861
4862 // first get value
4863 switch (encoding & 0x0F) {
4864 case DW_EH_PE_ptr:
4865 result = getP(addr);
4866 p += sizeof(pint_t);
4867 addr = (pint_t)p;
4868 break;
4869 case DW_EH_PE_uleb128:
4870 result = getULEB128(addr, end);
4871 break;
4872 case DW_EH_PE_udata2:
4873 result = get16(addr);
4874 p += 2;
4875 addr = (pint_t)p;
4876 break;
4877 case DW_EH_PE_udata4:
4878 result = get32(addr);
4879 p += 4;
4880 addr = (pint_t)p;
4881 break;
4882 case DW_EH_PE_udata8:
4883 result = get64(addr);
4884 p += 8;
4885 addr = (pint_t)p;
4886 break;
4887 case DW_EH_PE_sleb128:
4888 result = getSLEB128(addr, end);
4889 break;
4890 case DW_EH_PE_sdata2:
4891 result = (int16_t)get16(addr);
4892 p += 2;
4893 addr = (pint_t)p;
4894 break;
4895 case DW_EH_PE_sdata4:
4896 result = (int32_t)get32(addr);
4897 p += 4;
4898 addr = (pint_t)p;
4899 break;
4900 case DW_EH_PE_sdata8:
4901 result = get64(addr);
4902 p += 8;
4903 addr = (pint_t)p;
4904 break;
4905 default:
4906 throwf("ObjectFileAddressSpace<A>::getEncodedP() encoding 0x%08X not supported", encoding);
4907 }
4908
4909 // then add relative offset
4910 switch ( encoding & 0x70 ) {
4911 case DW_EH_PE_absptr:
4912 // do nothing
4913 break;
4914 case DW_EH_PE_pcrel:
4915 result += startAddr;
4916 break;
4917 case DW_EH_PE_textrel:
4918 throw "DW_EH_PE_textrel pointer encoding not supported";
4919 break;
4920 case DW_EH_PE_datarel:
4921 throw "DW_EH_PE_datarel pointer encoding not supported";
4922 break;
4923 case DW_EH_PE_funcrel:
4924 throw "DW_EH_PE_funcrel pointer encoding not supported";
4925 break;
4926 case DW_EH_PE_aligned:
4927 throw "DW_EH_PE_aligned pointer encoding not supported";
4928 break;
4929 default:
4930 throwf("ObjectFileAddressSpace<A>::getEncodedP() encoding 0x%08X not supported", encoding);
4931 break;
4932 }
4933
4934 // Note: DW_EH_PE_indirect is only used in CIEs to refernce the personality pointer
4935 // When parsing .o files that pointer contains zero, so we don't to return that.
4936 // Instead we skip the dereference and return the address of the pointer.
4937 // if ( encoding & DW_EH_PE_indirect )
4938 // result = getP(result);
4939
4940 return result;
4941 }
4942
4943 template <>
4944 const char* CUSection<x86_64>::personalityName(class Parser<x86_64>& parser, const macho_relocation_info<x86_64::P>* reloc)
4945 {
4946 if ( reloc->r_extern() ) {
4947 assert((reloc->r_type() == X86_64_RELOC_UNSIGNED) && "wrong reloc type on personality column in __compact_unwind section");
4948 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
4949 return parser.nameFromSymbol(sym);
4950 }
4951 else {
4952 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
4953 pint_t personalityAddr = *content;
4954 assert((parser.sectionForAddress(personalityAddr)->type() == ld::Section::typeCode) && "personality column in __compact_unwind section is not pointer to function");
4955 // atoms may not be constructed yet, so scan symbol table for labels
4956 const char* name = parser.scanSymbolTableForAddress(personalityAddr);
4957 return name;
4958 }
4959 }
4960
4961 template <>
4962 const char* CUSection<x86>::personalityName(class Parser<x86>& parser, const macho_relocation_info<x86::P>* reloc)
4963 {
4964 if ( reloc->r_extern() ) {
4965 assert((reloc->r_type() == GENERIC_RELOC_VANILLA) && "wrong reloc type on personality column in __compact_unwind section");
4966 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
4967 return parser.nameFromSymbol(sym);
4968 }
4969 else {
4970 // support __LD, __compact_unwind personality entries which are pointer to personality non-lazy pointer
4971 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
4972 pint_t nlPointerAddr = *content;
4973 Section<x86>* nlSection = parser.sectionForAddress(nlPointerAddr);
4974 if ( nlSection->type() == ld::Section::typeCode ) {
4975 // personality function is defined in this .o file, so this is a direct reference to it
4976 // atoms may not be constructed yet, so scan symbol table for labels
4977 const char* name = parser.scanSymbolTableForAddress(nlPointerAddr);
4978 return name;
4979 }
4980 else {
4981 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(nlPointerAddr, nlSection->machoSection());
4982 const macho_nlist<P>& nlSymbol = parser.symbolFromIndex(symIndex);
4983 return parser.nameFromSymbol(nlSymbol);
4984 }
4985 }
4986 }
4987
4988 #if SUPPORT_ARCH_arm64
4989 template <>
4990 const char* CUSection<arm64>::personalityName(class Parser<arm64>& parser, const macho_relocation_info<arm64::P>* reloc)
4991 {
4992 if ( reloc->r_extern() ) {
4993 assert((reloc->r_type() == ARM64_RELOC_UNSIGNED) && "wrong reloc type on personality column in __compact_unwind section");
4994 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
4995 return parser.nameFromSymbol(sym);
4996 }
4997 else {
4998 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
4999 pint_t personalityAddr = *content;
5000 Section<arm64>* personalitySection = parser.sectionForAddress(personalityAddr);
5001 (void)personalitySection;
5002 assert((personalitySection->type() == ld::Section::typeCode) && "personality column in __compact_unwind section is not pointer to function");
5003 // atoms may not be constructed yet, so scan symbol table for labels
5004 const char* name = parser.scanSymbolTableForAddress(personalityAddr);
5005 return name;
5006 }
5007 }
5008 #endif
5009
5010 #if SUPPORT_ARCH_arm_any
5011 template <>
5012 const char* CUSection<arm>::personalityName(class Parser<arm>& parser, const macho_relocation_info<arm::P>* reloc)
5013 {
5014 if ( reloc->r_extern() ) {
5015 assert((reloc->r_type() == ARM_RELOC_VANILLA) && "wrong reloc type on personality column in __compact_unwind section");
5016 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
5017 return parser.nameFromSymbol(sym);
5018 }
5019 else {
5020 // support __LD, __compact_unwind personality entries which are pointer to personality non-lazy pointer
5021 const pint_t* content = (pint_t*)(this->file().fileContent() + this->_machOSection->offset() + reloc->r_address());
5022 pint_t nlPointerAddr = *content;
5023 Section<arm>* nlSection = parser.sectionForAddress(nlPointerAddr);
5024 if ( nlSection->type() == ld::Section::typeCode ) {
5025 // personality function is defined in this .o file, so this is a direct reference to it
5026 // atoms may not be constructed yet, so scan symbol table for labels
5027 const char* name = parser.scanSymbolTableForAddress(nlPointerAddr);
5028 return name;
5029 }
5030 else {
5031 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(nlPointerAddr, nlSection->machoSection());
5032 const macho_nlist<P>& nlSymbol = parser.symbolFromIndex(symIndex);
5033 return parser.nameFromSymbol(nlSymbol);
5034 }
5035 }
5036 }
5037 #endif
5038
5039
5040 template <typename A>
5041 const char* CUSection<A>::personalityName(class Parser<A>& parser, const macho_relocation_info<P>* reloc)
5042 {
5043 return NULL;
5044 }
5045
5046 template <>
5047 bool CUSection<x86>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5048 {
5049 return ((enc & UNWIND_X86_MODE_MASK) == UNWIND_X86_MODE_DWARF);
5050 }
5051
5052 template <>
5053 bool CUSection<x86_64>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5054 {
5055 return ((enc & UNWIND_X86_64_MODE_MASK) == UNWIND_X86_64_MODE_DWARF);
5056 }
5057
5058 #if SUPPORT_ARCH_arm_any
5059 template <>
5060 bool CUSection<arm>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5061 {
5062 return ((enc & UNWIND_ARM_MODE_MASK) == UNWIND_ARM_MODE_DWARF);
5063 }
5064 #endif
5065
5066 #if SUPPORT_ARCH_arm64
5067 template <>
5068 bool CUSection<arm64>::encodingMeansUseDwarf(compact_unwind_encoding_t enc)
5069 {
5070 return ((enc & UNWIND_ARM64_MODE_MASK) == UNWIND_ARM64_MODE_DWARF);
5071 }
5072 #endif
5073
5074 template <typename A>
5075 int CUSection<A>::infoSorter(const void* l, const void* r)
5076 {
5077 // sort references by symbol index, then address
5078 const Info* left = (Info*)l;
5079 const Info* right = (Info*)r;
5080 if ( left->functionSymbolIndex == right->functionSymbolIndex )
5081 return (left->functionStartAddress - right->functionStartAddress);
5082 else
5083 return (left->functionSymbolIndex - right->functionSymbolIndex);
5084 }
5085
5086 template <typename A>
5087 void CUSection<A>::parse(class Parser<A>& parser, uint32_t cnt, Info array[])
5088 {
5089 // walk section content and copy to Info array
5090 const macho_compact_unwind_entry<P>* const entries = (macho_compact_unwind_entry<P>*)(this->file().fileContent() + this->_machOSection->offset());
5091 for (uint32_t i=0; i < cnt; ++i) {
5092 Info* info = &array[i];
5093 const macho_compact_unwind_entry<P>* entry = &entries[i];
5094 info->functionStartAddress = entry->codeStart();
5095 info->functionSymbolIndex = 0xFFFFFFFF;
5096 info->rangeLength = entry->codeLen();
5097 info->compactUnwindInfo = entry->compactUnwindInfo();
5098 info->personality = NULL;
5099 info->lsdaAddress = entry->lsda();
5100 info->function = NULL;
5101 info->lsda = NULL;
5102 if ( (info->compactUnwindInfo & UNWIND_PERSONALITY_MASK) != 0 )
5103 warning("no bits should be set in UNWIND_PERSONALITY_MASK of compact unwind encoding in __LD,__compact_unwind section");
5104 if ( info->lsdaAddress != 0 ) {
5105 info->compactUnwindInfo |= UNWIND_HAS_LSDA;
5106 }
5107 }
5108
5109 // scan relocs, extern relocs are needed for personality references (possibly for function/lsda refs??)
5110 const uint32_t sectionSize = this->_machOSection->size();
5111 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(this->file().fileContent() + this->_machOSection->reloff());
5112 const macho_relocation_info<P>* relocsEnd = &relocs[this->_machOSection->nreloc()];
5113 for (const macho_relocation_info<P>* reloc = relocs; reloc < relocsEnd; ++reloc) {
5114 if ( reloc->r_address() & R_SCATTERED )
5115 continue;
5116 if ( reloc->r_address() > sectionSize )
5117 throwf("malformed __compact_unwind relocation, offset (0x%08X) is beyond end of section,", reloc->r_address());
5118 if ( reloc->r_extern() ) {
5119 // only expect external relocs on some colummns
5120 if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::personalityFieldOffset() ) {
5121 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5122 array[entryIndex].personality = this->personalityName(parser, reloc);
5123 }
5124 else if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::lsdaFieldOffset() ) {
5125 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5126 const macho_nlist<P>& lsdaSym = parser.symbolFromIndex(reloc->r_symbolnum());
5127 if ( (lsdaSym.n_type() & N_TYPE) == N_SECT )
5128 array[entryIndex].lsdaAddress = lsdaSym.n_value();
5129 else
5130 warning("unexpected extern relocation to lsda in __compact_unwind section");
5131 }
5132 else if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::codeStartFieldOffset() ) {
5133 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5134 array[entryIndex].functionSymbolIndex = reloc->r_symbolnum();
5135 array[entryIndex].functionStartAddress += parser.symbolFromIndex(reloc->r_symbolnum()).n_value();
5136 }
5137 else {
5138 warning("unexpected extern relocation in __compact_unwind section");
5139 }
5140 }
5141 else {
5142 if ( (reloc->r_address() % sizeof(macho_compact_unwind_entry<P>)) == macho_compact_unwind_entry<P>::personalityFieldOffset() ) {
5143 uint32_t entryIndex = reloc->r_address() / sizeof(macho_compact_unwind_entry<P>);
5144 array[entryIndex].personality = this->personalityName(parser, reloc);
5145 }
5146 }
5147 }
5148
5149 // sort array by function start address so unwind infos will be contiguous for a given function
5150 ::qsort(array, cnt, sizeof(Info), infoSorter);
5151 }
5152
5153 template <typename A>
5154 uint32_t CUSection<A>::count()
5155 {
5156 const macho_section<P>* machoSect = this->machoSection();
5157 if ( (machoSect->size() % sizeof(macho_compact_unwind_entry<P>)) != 0 )
5158 throw "malformed __LD,__compact_unwind section, bad length";
5159
5160 return machoSect->size() / sizeof(macho_compact_unwind_entry<P>);
5161 }
5162
5163 template <typename A>
5164 void CUSection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays& cus)
5165 {
5166 Info* const arrayStart = cus.cuArray;
5167 Info* const arrayEnd = &cus.cuArray[cus.cuCount];
5168 for (Info* info=arrayStart; info < arrayEnd; ++info) {
5169 // find function atom from address
5170 info->function = parser.findAtomByAddress(info->functionStartAddress);
5171 // find lsda atom from address
5172 if ( info->lsdaAddress != 0 ) {
5173 info->lsda = parser.findAtomByAddress(info->lsdaAddress);
5174 // add lsda subordinate
5175 typename Parser<A>::SourceLocation src(info->function, info->functionStartAddress - info->function->objectAddress());
5176 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, info->lsda);
5177 }
5178 if ( info->personality != NULL ) {
5179 // add personality subordinate
5180 typename Parser<A>::SourceLocation src(info->function, info->functionStartAddress - info->function->objectAddress());
5181 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinatePersonality, false, info->personality);
5182 }
5183 }
5184
5185 }
5186
5187 template <typename A>
5188 SymboledSection<A>::SymboledSection(Parser<A>& parser, File<A>& f, const macho_section<typename A::P>* s)
5189 : Section<A>(f, s), _type(ld::Atom::typeUnclassified)
5190 {
5191 switch ( s->flags() & SECTION_TYPE ) {
5192 case S_ZEROFILL:
5193 _type = ld::Atom::typeZeroFill;
5194 break;
5195 case S_MOD_INIT_FUNC_POINTERS:
5196 _type = ld::Atom::typeInitializerPointers;
5197 break;
5198 case S_MOD_TERM_FUNC_POINTERS:
5199 _type = ld::Atom::typeTerminatorPointers;
5200 break;
5201 case S_THREAD_LOCAL_VARIABLES:
5202 _type = ld::Atom::typeTLV;
5203 break;
5204 case S_THREAD_LOCAL_ZEROFILL:
5205 _type = ld::Atom::typeTLVZeroFill;
5206 break;
5207 case S_THREAD_LOCAL_REGULAR:
5208 _type = ld::Atom::typeTLVInitialValue;
5209 break;
5210 case S_THREAD_LOCAL_INIT_FUNCTION_POINTERS:
5211 _type = ld::Atom::typeTLVInitializerPointers;
5212 break;
5213 case S_REGULAR:
5214 if ( strncmp(s->sectname(), "__gcc_except_tab", 16) == 0 )
5215 _type = ld::Atom::typeLSDA;
5216 else if ( this->type() == ld::Section::typeInitializerPointers )
5217 _type = ld::Atom::typeInitializerPointers;
5218 break;
5219 }
5220 }
5221
5222
5223 template <typename A>
5224 bool SymboledSection<A>::dontDeadStrip()
5225 {
5226 switch ( _type ) {
5227 case ld::Atom::typeInitializerPointers:
5228 case ld::Atom::typeTerminatorPointers:
5229 return true;
5230 default:
5231 // model an object file without MH_SUBSECTIONS_VIA_SYMBOLS as one in which nothing can be dead stripped
5232 if ( ! this->_file.canScatterAtoms() )
5233 return true;
5234 // call inherited
5235 return Section<A>::dontDeadStrip();
5236 }
5237 return false;
5238 }
5239
5240
5241 template <typename A>
5242 uint32_t SymboledSection<A>::computeAtomCount(class Parser<A>& parser,
5243 struct Parser<A>::LabelAndCFIBreakIterator& it,
5244 const struct Parser<A>::CFI_CU_InfoArrays&)
5245 {
5246 const pint_t startAddr = this->_machOSection->addr();
5247 const pint_t endAddr = startAddr + this->_machOSection->size();
5248 const uint32_t sectNum = this->sectionNum(parser);
5249
5250 uint32_t count = 0;
5251 pint_t addr;
5252 pint_t size;
5253 const macho_nlist<P>* sym;
5254 while ( it.next(parser, *this, sectNum, startAddr, endAddr, &addr, &size, &sym) ) {
5255 ++count;
5256 }
5257 //fprintf(stderr, "computeAtomCount(%s,%s) => %d\n", this->segmentName(), this->sectionName(), count);
5258 return count;
5259 }
5260
5261 template <typename A>
5262 uint32_t SymboledSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
5263 struct Parser<A>::LabelAndCFIBreakIterator& it,
5264 const struct Parser<A>::CFI_CU_InfoArrays&)
5265 {
5266 this->_beginAtoms = (Atom<A>*)p;
5267
5268 //fprintf(stderr, "SymboledSection::appendAtoms() in section %s\n", this->_machOSection->sectname());
5269 const pint_t startAddr = this->_machOSection->addr();
5270 const pint_t endAddr = startAddr + this->_machOSection->size();
5271 const uint32_t sectNum = this->sectionNum(parser);
5272
5273 uint32_t count = 0;
5274 pint_t addr;
5275 pint_t size;
5276 const macho_nlist<P>* label;
5277 while ( it.next(parser, *this, sectNum, startAddr, endAddr, &addr, &size, &label) ) {
5278 Atom<A>* allocatedSpace = (Atom<A>*)p;
5279 // is break because of label or CFI?
5280 if ( label != NULL ) {
5281 // The size is computed based on the address of the next label (or the end of the section for the last label)
5282 // If there are two labels at the same address, we want them one to be an alias of the other.
5283 // If the label is at the end of a section, it is has zero size, but is not an alias
5284 const bool isAlias = ( (size == 0) && (addr < endAddr) );
5285 new (allocatedSpace) Atom<A>(*this, parser, *label, size, isAlias);
5286 if ( isAlias )
5287 this->_hasAliases = true;
5288 if ( parser.altEntryFromSymbol(*label) )
5289 this->_altEntries.insert(allocatedSpace);
5290 }
5291 else {
5292 ld::Atom::SymbolTableInclusion inclusion = ld::Atom::symbolTableNotIn;
5293 ld::Atom::ContentType ctype = this->contentType();
5294 if ( ctype == ld::Atom::typeLSDA )
5295 inclusion = ld::Atom::symbolTableInWithRandomAutoStripLabel;
5296 new (allocatedSpace) Atom<A>(*this, "anon", addr, size, ld::Atom::definitionRegular, ld::Atom::combineNever,
5297 ld::Atom::scopeTranslationUnit, ctype, inclusion,
5298 this->dontDeadStrip(), false, false, this->alignmentForAddress(addr));
5299 }
5300 p += sizeof(Atom<A>);
5301 ++count;
5302 }
5303
5304 this->_endAtoms = (Atom<A>*)p;
5305 return count;
5306 }
5307
5308
5309 template <>
5310 ld::Atom::SymbolTableInclusion ImplicitSizeSection<arm64>::symbolTableInclusion()
5311 {
5312 return ld::Atom::symbolTableInWithRandomAutoStripLabel;
5313 }
5314
5315 template <typename A>
5316 ld::Atom::SymbolTableInclusion ImplicitSizeSection<A>::symbolTableInclusion()
5317 {
5318 return ld::Atom::symbolTableNotIn;
5319 }
5320
5321
5322 template <typename A>
5323 uint32_t ImplicitSizeSection<A>::computeAtomCount(class Parser<A>& parser,
5324 struct Parser<A>::LabelAndCFIBreakIterator& it,
5325 const struct Parser<A>::CFI_CU_InfoArrays&)
5326 {
5327 uint32_t count = 0;
5328 const macho_section<P>* sect = this->machoSection();
5329 const pint_t startAddr = sect->addr();
5330 const pint_t endAddr = startAddr + sect->size();
5331 for (pint_t addr = startAddr; addr < endAddr; addr += elementSizeAtAddress(addr) ) {
5332 if ( useElementAt(parser, it, addr) )
5333 ++count;
5334 }
5335 if ( it.fileHasOverlappingSymbols && (sect->size() != 0) && (this->combine(parser, startAddr) == ld::Atom::combineByNameAndContent) ) {
5336 // if there are multiple labels in this section for the same address, then clone them into multi atoms
5337 pint_t prevSymbolAddr = (pint_t)(-1);
5338 uint8_t prevSymbolSectNum = 0;
5339 bool prevIgnore = false;
5340 for(uint32_t i=0; i < it.sortedSymbolCount; ++i) {
5341 const macho_nlist<P>& sym = parser.symbolFromIndex(it.sortedSymbolIndexes[i]);
5342 const pint_t symbolAddr = sym.n_value();
5343 const uint8_t symbolSectNum = sym.n_sect();
5344 const bool ignore = this->ignoreLabel(parser.nameFromSymbol(sym));
5345 if ( !ignore && !prevIgnore && (symbolAddr == prevSymbolAddr) && (prevSymbolSectNum == symbolSectNum) && (symbolSectNum == this->sectionNum(parser)) ) {
5346 ++count;
5347 }
5348 prevSymbolAddr = symbolAddr;
5349 prevSymbolSectNum = symbolSectNum;
5350 prevIgnore = ignore;
5351 }
5352 }
5353 return count;
5354 }
5355
5356 template <typename A>
5357 uint32_t ImplicitSizeSection<A>::appendAtoms(class Parser<A>& parser, uint8_t* p,
5358 struct Parser<A>::LabelAndCFIBreakIterator& it,
5359 const struct Parser<A>::CFI_CU_InfoArrays&)
5360 {
5361 this->_beginAtoms = (Atom<A>*)p;
5362
5363 const macho_section<P>* sect = this->machoSection();
5364 const pint_t startAddr = sect->addr();
5365 const pint_t endAddr = startAddr + sect->size();
5366 const uint32_t sectNum = this->sectionNum(parser);
5367 //fprintf(stderr, "ImplicitSizeSection::appendAtoms() in section %s\n", sect->sectname());
5368 uint32_t count = 0;
5369 pint_t foundAddr;
5370 pint_t size;
5371 const macho_nlist<P>* foundLabel;
5372 Atom<A>* allocatedSpace;
5373 while ( it.next(parser, *this, sectNum, startAddr, endAddr, &foundAddr, &size, &foundLabel) ) {
5374 if ( foundLabel != NULL ) {
5375 bool skip = false;
5376 pint_t labeledAtomSize = this->elementSizeAtAddress(foundAddr);
5377 allocatedSpace = (Atom<A>*)p;
5378 if ( this->ignoreLabel(parser.nameFromSymbol(*foundLabel)) ) {
5379 if ( size == 0 ) {
5380 // <rdar://problem/10018737>
5381 // a size of zero means there is another label at same location
5382 // and we are supposed to ignore this label
5383 skip = true;
5384 }
5385 else {
5386 //fprintf(stderr, " 0x%08llX make annon, size=%lld\n", (uint64_t)foundAddr, (uint64_t)size);
5387 new (allocatedSpace) Atom<A>(*this, this->unlabeledAtomName(parser, foundAddr), foundAddr,
5388 this->elementSizeAtAddress(foundAddr), this->definition(),
5389 this->combine(parser, foundAddr), this->scopeAtAddress(parser, foundAddr),
5390 this->contentType(), this->symbolTableInclusion(),
5391 this->dontDeadStrip(), false, false, this->alignmentForAddress(foundAddr));
5392 }
5393 }
5394 else {
5395 // make named atom for label
5396 //fprintf(stderr, " 0x%08llX make labeled\n", (uint64_t)foundAddr);
5397 new (allocatedSpace) Atom<A>(*this, parser, *foundLabel, labeledAtomSize);
5398 }
5399 if ( !skip ) {
5400 ++count;
5401 p += sizeof(Atom<A>);
5402 foundAddr += labeledAtomSize;
5403 size -= labeledAtomSize;
5404 }
5405 }
5406 // some number of anonymous atoms
5407 for (pint_t addr = foundAddr; addr < (foundAddr+size); addr += elementSizeAtAddress(addr) ) {
5408 // make anon atoms for area before label
5409 if ( this->useElementAt(parser, it, addr) ) {
5410 //fprintf(stderr, " 0x%08llX make annon, size=%lld\n", (uint64_t)addr, (uint64_t)elementSizeAtAddress(addr));
5411 allocatedSpace = (Atom<A>*)p;
5412 new (allocatedSpace) Atom<A>(*this, this->unlabeledAtomName(parser, addr), addr, this->elementSizeAtAddress(addr),
5413 this->definition(), this->combine(parser, addr), this->scopeAtAddress(parser, addr),
5414 this->contentType(), this->symbolTableInclusion(),
5415 this->dontDeadStrip(), false, false, this->alignmentForAddress(addr));
5416 ++count;
5417 p += sizeof(Atom<A>);
5418 }
5419 }
5420 }
5421
5422 this->_endAtoms = (Atom<A>*)p;
5423
5424 return count;
5425 }
5426
5427 template <typename A>
5428 bool Literal4Section<A>::ignoreLabel(const char* label) const
5429 {
5430 return (label[0] == 'L') || (label[0] == 'l');
5431 }
5432
5433 template <typename A>
5434 unsigned long Literal4Section<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5435 {
5436 const uint32_t* literalContent = (uint32_t*)atom->contentPointer();
5437 return *literalContent;
5438 }
5439
5440 template <typename A>
5441 bool Literal4Section<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5442 const ld::IndirectBindingTable& ind) const
5443 {
5444 assert(this->type() == rhs.section().type());
5445 const uint32_t* literalContent = (uint32_t*)atom->contentPointer();
5446
5447 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5448 assert(rhsAtom != NULL);
5449 if ( rhsAtom != NULL ) {
5450 const uint32_t* rhsLiteralContent = (uint32_t*)rhsAtom->contentPointer();
5451 return (*literalContent == *rhsLiteralContent);
5452 }
5453 return false;
5454 }
5455
5456
5457 template <typename A>
5458 bool Literal8Section<A>::ignoreLabel(const char* label) const
5459 {
5460 return (label[0] == 'L') || (label[0] == 'l');
5461 }
5462
5463 template <typename A>
5464 unsigned long Literal8Section<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5465 {
5466 #if __LP64__
5467 const uint64_t* literalContent = (uint64_t*)atom->contentPointer();
5468 return *literalContent;
5469 #else
5470 unsigned long hash = 5381;
5471 const uint8_t* byteContent = atom->contentPointer();
5472 for (int i=0; i < 8; ++i) {
5473 hash = hash * 33 + byteContent[i];
5474 }
5475 return hash;
5476 #endif
5477 }
5478
5479 template <typename A>
5480 bool Literal8Section<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5481 const ld::IndirectBindingTable& ind) const
5482 {
5483 if ( rhs.section().type() != ld::Section::typeLiteral8 )
5484 return false;
5485 assert(this->type() == rhs.section().type());
5486 const uint64_t* literalContent = (uint64_t*)atom->contentPointer();
5487
5488 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5489 assert(rhsAtom != NULL);
5490 if ( rhsAtom != NULL ) {
5491 const uint64_t* rhsLiteralContent = (uint64_t*)rhsAtom->contentPointer();
5492 return (*literalContent == *rhsLiteralContent);
5493 }
5494 return false;
5495 }
5496
5497 template <typename A>
5498 bool Literal16Section<A>::ignoreLabel(const char* label) const
5499 {
5500 return (label[0] == 'L') || (label[0] == 'l');
5501 }
5502
5503 template <typename A>
5504 unsigned long Literal16Section<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5505 {
5506 unsigned long hash = 5381;
5507 const uint8_t* byteContent = atom->contentPointer();
5508 for (int i=0; i < 16; ++i) {
5509 hash = hash * 33 + byteContent[i];
5510 }
5511 return hash;
5512 }
5513
5514 template <typename A>
5515 bool Literal16Section<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5516 const ld::IndirectBindingTable& ind) const
5517 {
5518 if ( rhs.section().type() != ld::Section::typeLiteral16 )
5519 return false;
5520 assert(this->type() == rhs.section().type());
5521 const uint64_t* literalContent = (uint64_t*)atom->contentPointer();
5522
5523 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5524 assert(rhsAtom != NULL);
5525 if ( rhsAtom != NULL ) {
5526 const uint64_t* rhsLiteralContent = (uint64_t*)rhsAtom->contentPointer();
5527 return ((literalContent[0] == rhsLiteralContent[0]) && (literalContent[1] == rhsLiteralContent[1]));
5528 }
5529 return false;
5530 }
5531
5532
5533
5534 template <typename A>
5535 typename A::P::uint_t CStringSection<A>::elementSizeAtAddress(pint_t addr)
5536 {
5537 const macho_section<P>* sect = this->machoSection();
5538 const char* stringContent = (char*)(this->file().fileContent() + sect->offset() + addr - sect->addr());
5539 return strlen(stringContent) + 1;
5540 }
5541
5542 template <typename A>
5543 bool CStringSection<A>::useElementAt(Parser<A>& parser, struct Parser<A>::LabelAndCFIBreakIterator& it, pint_t addr)
5544 {
5545 return true;
5546 }
5547
5548 template <typename A>
5549 bool CStringSection<A>::ignoreLabel(const char* label) const
5550 {
5551 return (label[0] == 'L') || (label[0] == 'l');
5552 }
5553
5554
5555 template <typename A>
5556 Atom<A>* CStringSection<A>::findAtomByAddress(pint_t addr)
5557 {
5558 Atom<A>* result = this->findContentAtomByAddress(addr, this->_beginAtoms, this->_endAtoms);
5559 return result;
5560 }
5561
5562 template <typename A>
5563 unsigned long CStringSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5564 {
5565 unsigned long hash = 5381;
5566 const char* stringContent = (char*)atom->contentPointer();
5567 for (const char* s = stringContent; *s != '\0'; ++s) {
5568 hash = hash * 33 + *s;
5569 }
5570 return hash;
5571 }
5572
5573
5574 template <typename A>
5575 bool CStringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5576 const ld::IndirectBindingTable& ind) const
5577 {
5578 if ( rhs.section().type() != ld::Section::typeCString )
5579 return false;
5580 assert(this->type() == rhs.section().type());
5581 assert(strcmp(this->sectionName(), rhs.section().sectionName())== 0);
5582 assert(strcmp(this->segmentName(), rhs.section().segmentName())== 0);
5583 const char* stringContent = (char*)atom->contentPointer();
5584
5585 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5586 assert(rhsAtom != NULL);
5587 if ( rhsAtom != NULL ) {
5588 if ( atom->_size != rhsAtom->_size )
5589 return false;
5590 const char* rhsStringContent = (char*)rhsAtom->contentPointer();
5591 return (strcmp(stringContent, rhsStringContent) == 0);
5592 }
5593 return false;
5594 }
5595
5596
5597 template <>
5598 ld::Fixup::Kind NonLazyPointerSection<x86>::fixupKind()
5599 {
5600 return ld::Fixup::kindStoreLittleEndian32;
5601 }
5602
5603 template <>
5604 ld::Fixup::Kind NonLazyPointerSection<arm>::fixupKind()
5605 {
5606 return ld::Fixup::kindStoreLittleEndian32;
5607 }
5608
5609 template <>
5610 ld::Fixup::Kind NonLazyPointerSection<arm64>::fixupKind()
5611 {
5612 return ld::Fixup::kindStoreLittleEndian64;
5613 }
5614
5615
5616 template <>
5617 void NonLazyPointerSection<x86_64>::makeFixups(class Parser<x86_64>& parser, const struct Parser<x86_64>::CFI_CU_InfoArrays&)
5618 {
5619 assert(0 && "x86_64 should not have non-lazy-pointer sections in .o files");
5620 }
5621
5622 template <typename A>
5623 void NonLazyPointerSection<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&)
5624 {
5625 // add references for each NLP atom based on indirect symbol table
5626 const macho_section<P>* sect = this->machoSection();
5627 const pint_t endAddr = sect->addr() + sect->size();
5628 for( pint_t addr = sect->addr(); addr < endAddr; addr += sizeof(pint_t)) {
5629 typename Parser<A>::SourceLocation src;
5630 typename Parser<A>::TargetDesc target;
5631 src.atom = this->findAtomByAddress(addr);
5632 src.offsetInAtom = 0;
5633 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5634 target.atom = NULL;
5635 target.name = NULL;
5636 target.weakImport = false;
5637 target.addend = 0;
5638 if ( symIndex == INDIRECT_SYMBOL_LOCAL ) {
5639 // use direct reference for local symbols
5640 const pint_t* nlpContent = (pint_t*)(this->file().fileContent() + sect->offset() + addr - sect->addr());
5641 pint_t targetAddr = P::getP(*nlpContent);
5642 target.atom = parser.findAtomByAddress(targetAddr);
5643 target.weakImport = false;
5644 target.addend = (targetAddr - target.atom->objectAddress());
5645 // <rdar://problem/8385011> if pointer to thumb function, mask of thumb bit (not an addend of +1)
5646 if ( target.atom->isThumb() )
5647 target.addend &= (-2);
5648 assert(src.atom->combine() == ld::Atom::combineNever);
5649 }
5650 else {
5651 const macho_nlist<P>& sym = parser.symbolFromIndex(symIndex);
5652 // use direct reference for local symbols
5653 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) ) {
5654 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
5655 assert(src.atom->combine() == ld::Atom::combineNever);
5656 }
5657 else {
5658 target.name = parser.nameFromSymbol(sym);
5659 target.weakImport = parser.weakImportFromSymbol(sym);
5660 assert(src.atom->combine() == ld::Atom::combineByNameAndReferences);
5661 }
5662 }
5663 parser.addFixups(src, this->fixupKind(), target);
5664 }
5665 }
5666
5667 template <typename A>
5668 ld::Atom::Combine NonLazyPointerSection<A>::combine(Parser<A>& parser, pint_t addr)
5669 {
5670 const macho_section<P>* sect = this->machoSection();
5671 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5672 if ( symIndex == INDIRECT_SYMBOL_LOCAL)
5673 return ld::Atom::combineNever;
5674
5675 // don't coalesce non-lazy-pointers to local symbols
5676 const macho_nlist<P>& sym = parser.symbolFromIndex(symIndex);
5677 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) )
5678 return ld::Atom::combineNever;
5679
5680 return ld::Atom::combineByNameAndReferences;
5681 }
5682
5683 template <typename A>
5684 const char* NonLazyPointerSection<A>::targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind)
5685 {
5686 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5687 assert(atom->fixupCount() == 1);
5688 ld::Fixup::iterator fit = atom->fixupsBegin();
5689 const char* name = NULL;
5690 switch ( fit->binding ) {
5691 case ld::Fixup::bindingByNameUnbound:
5692 name = fit->u.name;
5693 break;
5694 case ld::Fixup::bindingByContentBound:
5695 name = fit->u.target->name();
5696 break;
5697 case ld::Fixup::bindingsIndirectlyBound:
5698 name = ind.indirectName(fit->u.bindingIndex);
5699 break;
5700 default:
5701 assert(0);
5702 }
5703 assert(name != NULL);
5704 return name;
5705 }
5706
5707 template <typename A>
5708 unsigned long NonLazyPointerSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5709 {
5710 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5711 unsigned long hash = 9508;
5712 for (const char* s = this->targetName(atom, ind); *s != '\0'; ++s) {
5713 hash = hash * 33 + *s;
5714 }
5715 return hash;
5716 }
5717
5718 template <typename A>
5719 bool NonLazyPointerSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5720 const ld::IndirectBindingTable& indirectBindingTable) const
5721 {
5722 if ( rhs.section().type() != ld::Section::typeNonLazyPointer )
5723 return false;
5724 assert(this->type() == rhs.section().type());
5725 // there can be many non-lazy pointer in different section names
5726 // we only want to coalesce in same section name
5727 if ( *this != rhs.section() )
5728 return false;
5729 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5730 assert(rhsAtom != NULL);
5731 const char* thisName = this->targetName(atom, indirectBindingTable);
5732 const char* rhsName = this->targetName(rhsAtom, indirectBindingTable);
5733 return (strcmp(thisName, rhsName) == 0);
5734 }
5735
5736 template <typename A>
5737 ld::Atom::Scope NonLazyPointerSection<A>::scopeAtAddress(Parser<A>& parser, pint_t addr)
5738 {
5739 const macho_section<P>* sect = this->machoSection();
5740 uint32_t symIndex = parser.symbolIndexFromIndirectSectionAddress(addr, sect);
5741 if ( symIndex == INDIRECT_SYMBOL_LOCAL)
5742 return ld::Atom::scopeTranslationUnit;
5743 else
5744 return ld::Atom::scopeLinkageUnit;
5745 }
5746
5747
5748
5749 template <typename A>
5750 ld::Atom::Combine TLVPointerSection<A>::combine(Parser<A>& parser, pint_t addr)
5751 {
5752 return ld::Atom::combineByNameAndReferences;
5753 }
5754
5755
5756 template <typename A>
5757 const char* TLVPointerSection<A>::targetName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind, bool* isStatic)
5758 {
5759 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5760 assert(atom->fixupCount() == 1);
5761 *isStatic = false;
5762 ld::Fixup::iterator fit = atom->fixupsBegin();
5763 const char* name = NULL;
5764 switch ( fit->binding ) {
5765 case ld::Fixup::bindingByNameUnbound:
5766 name = fit->u.name;
5767 break;
5768 case ld::Fixup::bindingByContentBound:
5769 name = fit->u.target->name();
5770 break;
5771 case ld::Fixup::bindingsIndirectlyBound:
5772 name = ind.indirectName(fit->u.bindingIndex);
5773 break;
5774 case ld::Fixup::bindingDirectlyBound:
5775 name = fit->u.target->name();
5776 *isStatic = (fit->u.target->scope() == ld::Atom::scopeTranslationUnit);
5777 break;
5778 default:
5779 assert(0);
5780 }
5781 assert(name != NULL);
5782 return name;
5783 }
5784
5785 template <typename A>
5786 unsigned long TLVPointerSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5787 {
5788 assert(atom->combine() == ld::Atom::combineByNameAndReferences);
5789 unsigned long hash = 9508;
5790 bool isStatic;
5791 for (const char* s = this->targetName(atom, ind, &isStatic); *s != '\0'; ++s) {
5792 hash = hash * 33 + *s;
5793 }
5794 return hash;
5795 }
5796
5797 template <typename A>
5798 bool TLVPointerSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5799 const ld::IndirectBindingTable& indirectBindingTable) const
5800 {
5801 if ( rhs.section().type() != ld::Section::typeTLVPointers )
5802 return false;
5803 assert(this->type() == rhs.section().type());
5804 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5805 assert(rhsAtom != NULL);
5806 bool thisIsStatic;
5807 bool rhsIsStatic;
5808 const char* thisName = this->targetName(atom, indirectBindingTable, &thisIsStatic);
5809 const char* rhsName = this->targetName(rhsAtom, indirectBindingTable, &rhsIsStatic);
5810 return !thisIsStatic && !rhsIsStatic && (strcmp(thisName, rhsName) == 0);
5811 }
5812
5813
5814 template <typename A>
5815 const uint8_t* CFStringSection<A>::targetContent(const class Atom<A>* atom, const ld::IndirectBindingTable& ind,
5816 ContentType* ct, unsigned int* count)
5817 {
5818 *ct = contentUnknown;
5819 for (ld::Fixup::iterator fit=atom->fixupsBegin(), end=atom->fixupsEnd(); fit != end; ++fit) {
5820 const ld::Atom* targetAtom = NULL;
5821 switch ( fit->binding ) {
5822 case ld::Fixup::bindingByNameUnbound:
5823 // ignore reference to ___CFConstantStringClassReference
5824 // we are just looking for reference to backing string data
5825 assert(fit->offsetInAtom == 0);
5826 assert(strcmp(fit->u.name, "___CFConstantStringClassReference") == 0);
5827 break;
5828 case ld::Fixup::bindingDirectlyBound:
5829 case ld::Fixup::bindingByContentBound:
5830 targetAtom = fit->u.target;
5831 break;
5832 case ld::Fixup::bindingsIndirectlyBound:
5833 targetAtom = ind.indirectAtom(fit->u.bindingIndex);
5834 break;
5835 default:
5836 assert(0 && "bad binding type");
5837 }
5838 assert(targetAtom != NULL);
5839 const Atom<A>* target = dynamic_cast<const Atom<A>*>(targetAtom);
5840 if ( targetAtom->section().type() == ld::Section::typeCString ) {
5841 *ct = contentUTF8;
5842 *count = targetAtom->size();
5843 }
5844 else if ( targetAtom->section().type() == ld::Section::typeUTF16Strings ) {
5845 *ct = contentUTF16;
5846 *count = (targetAtom->size()+1)/2; // round up incase of buggy compiler that has only one trailing zero byte
5847 }
5848 else {
5849 *ct = contentUnknown;
5850 *count = 0;
5851 return NULL;
5852 }
5853 return target->contentPointer();
5854 }
5855 assert(0);
5856 return NULL;
5857 }
5858
5859 template <typename A>
5860 unsigned long CFStringSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5861 {
5862 // base hash of CFString on hash of cstring it wraps
5863 ContentType cType;
5864 unsigned long hash;
5865 unsigned int charCount;
5866 const uint8_t* content = this->targetContent(atom, ind, &cType, &charCount);
5867 switch ( cType ) {
5868 case contentUTF8:
5869 hash = 9408;
5870 for (const char* s = (char*)content; *s != '\0'; ++s) {
5871 hash = hash * 33 + *s;
5872 }
5873 return hash;
5874 case contentUTF16:
5875 hash = 407955;
5876 --charCount; // don't add last 0x0000 to hash because some buggy compilers only have trailing single byte
5877 for (const uint16_t* s = (uint16_t*)content; charCount > 0; ++s, --charCount) {
5878 hash = hash * 1025 + *s;
5879 }
5880 return hash;
5881 case contentUnknown:
5882 // <rdar://problem/14134211> For malformed CFStrings, hash to address of atom so they have unique hashes
5883 return ULONG_MAX - (unsigned long)(atom);
5884 }
5885 return 0;
5886 }
5887
5888
5889 template <typename A>
5890 bool CFStringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
5891 const ld::IndirectBindingTable& indirectBindingTable) const
5892 {
5893 if ( atom == &rhs )
5894 return true;
5895 if ( rhs.section().type() != ld::Section::typeCFString)
5896 return false;
5897 assert(this->type() == rhs.section().type());
5898 assert(strcmp(this->sectionName(), "__cfstring") == 0);
5899
5900 ContentType thisType;
5901 unsigned int charCount;
5902 const uint8_t* cstringContent = this->targetContent(atom, indirectBindingTable, &thisType, &charCount);
5903 ContentType rhsType;
5904 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
5905 assert(rhsAtom != NULL);
5906 unsigned int rhsCharCount;
5907 const uint8_t* rhsStringContent = this->targetContent(rhsAtom, indirectBindingTable, &rhsType, &rhsCharCount);
5908
5909 if ( thisType != rhsType )
5910 return false;
5911
5912 if ( thisType == contentUnknown )
5913 return false;
5914
5915 if ( rhsType == contentUnknown )
5916 return false;
5917
5918 // no need to compare content of pointers are already the same
5919 if ( cstringContent == rhsStringContent )
5920 return true;
5921
5922 // no need to compare content if size is different
5923 if ( charCount != rhsCharCount )
5924 return false;
5925
5926 switch ( thisType ) {
5927 case contentUTF8:
5928 return (strcmp((char*)cstringContent, (char*)rhsStringContent) == 0);
5929 case contentUTF16:
5930 {
5931 const uint16_t* cstringContent16 = (uint16_t*)cstringContent;
5932 const uint16_t* rhsStringContent16 = (uint16_t*)rhsStringContent;
5933 for (unsigned int i = 0; i < charCount; ++i) {
5934 if ( cstringContent16[i] != rhsStringContent16[i] )
5935 return false;
5936 }
5937 return true;
5938 }
5939 case contentUnknown:
5940 return false;
5941 }
5942 return false;
5943 }
5944
5945
5946 template <typename A>
5947 typename A::P::uint_t ObjC1ClassSection<A>::elementSizeAtAddress(pint_t addr)
5948 {
5949 // nominal size for each class is 48 bytes, but sometimes the compiler
5950 // over aligns and there is padding after class data
5951 const macho_section<P>* sct = this->machoSection();
5952 uint32_t align = 1 << sct->align();
5953 uint32_t size = ((12 * sizeof(pint_t)) + align-1) & (-align);
5954 return size;
5955 }
5956
5957 template <typename A>
5958 const char* ObjC1ClassSection<A>::unlabeledAtomName(Parser<A>& parser, pint_t addr)
5959 {
5960 // 8-bytes into class object is pointer to class name
5961 const macho_section<P>* sct = this->machoSection();
5962 uint32_t classObjcFileOffset = sct->offset() - sct->addr() + addr;
5963 const uint8_t* mappedFileContent = this->file().fileContent();
5964 pint_t nameAddr = P::getP(*((pint_t*)(mappedFileContent+classObjcFileOffset+2*sizeof(pint_t))));
5965
5966 // find section containing string address to get string bytes
5967 const macho_section<P>* const sections = parser.firstMachOSection();
5968 const uint32_t sectionCount = parser.machOSectionCount();
5969 for (uint32_t i=0; i < sectionCount; ++i) {
5970 const macho_section<P>* aSect = &sections[i];
5971 if ( (aSect->addr() <= nameAddr) && (nameAddr < (aSect->addr()+aSect->size())) ) {
5972 assert((aSect->flags() & SECTION_TYPE) == S_CSTRING_LITERALS);
5973 uint32_t nameFileOffset = aSect->offset() - aSect->addr() + nameAddr;
5974 const char* name = (char*)mappedFileContent + nameFileOffset;
5975 // spin through symbol table to find absolute symbol corresponding to this class
5976 for (uint32_t s=0; s < parser.symbolCount(); ++s) {
5977 const macho_nlist<P>& sym = parser.symbolFromIndex(s);
5978 if ( (sym.n_type() & N_TYPE) != N_ABS )
5979 continue;
5980 const char* absName = parser.nameFromSymbol(sym);
5981 if ( strncmp(absName, ".objc_class_name_", 17) == 0 ) {
5982 if ( strcmp(&absName[17], name) == 0 )
5983 return absName;
5984 }
5985 }
5986 assert(0 && "obj class name not found in symbol table");
5987 }
5988 }
5989 assert(0 && "obj class name not found");
5990 return "unknown objc class";
5991 }
5992
5993
5994 template <typename A>
5995 const char* ObjC2ClassRefsSection<A>::targetClassName(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
5996 {
5997 assert(atom->fixupCount() == 1);
5998 ld::Fixup::iterator fit = atom->fixupsBegin();
5999 const char* className = NULL;
6000 switch ( fit->binding ) {
6001 case ld::Fixup::bindingByNameUnbound:
6002 className = fit->u.name;
6003 break;
6004 case ld::Fixup::bindingDirectlyBound:
6005 case ld::Fixup::bindingByContentBound:
6006 className = fit->u.target->name();
6007 break;
6008 case ld::Fixup::bindingsIndirectlyBound:
6009 className = ind.indirectName(fit->u.bindingIndex);
6010 break;
6011 default:
6012 assert(0 && "unsupported binding in objc2 class ref section");
6013 }
6014 assert(className != NULL);
6015 return className;
6016 }
6017
6018
6019 template <typename A>
6020 unsigned long ObjC2ClassRefsSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6021 {
6022 unsigned long hash = 978;
6023 for (const char* s = targetClassName(atom, ind); *s != '\0'; ++s) {
6024 hash = hash * 33 + *s;
6025 }
6026 return hash;
6027 }
6028
6029 template <typename A>
6030 bool ObjC2ClassRefsSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
6031 const ld::IndirectBindingTable& indirectBindingTable) const
6032 {
6033 assert(this->type() == rhs.section().type());
6034 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
6035 assert(rhsAtom != NULL);
6036 const char* thisClassName = targetClassName(atom, indirectBindingTable);
6037 const char* rhsClassName = targetClassName(rhsAtom, indirectBindingTable);
6038 return (strcmp(thisClassName, rhsClassName) == 0);
6039 }
6040
6041
6042 template <typename A>
6043 const char* Objc1ClassReferences<A>::targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6044 {
6045 assert(atom->fixupCount() == 2);
6046 ld::Fixup::iterator fit = atom->fixupsBegin();
6047 if ( fit->kind == ld::Fixup::kindSetTargetAddress )
6048 ++fit;
6049 const ld::Atom* targetAtom = NULL;
6050 switch ( fit->binding ) {
6051 case ld::Fixup::bindingByContentBound:
6052 targetAtom = fit->u.target;
6053 break;
6054 case ld::Fixup::bindingsIndirectlyBound:
6055 targetAtom = ind.indirectAtom(fit->u.bindingIndex);
6056 if ( targetAtom == NULL ) {
6057 fprintf(stderr, "missing target named %s\n", ind.indirectName(fit->u.bindingIndex));
6058 }
6059 break;
6060 default:
6061 assert(0);
6062 }
6063 assert(targetAtom != NULL);
6064 const Atom<A>* target = dynamic_cast<const Atom<A>*>(targetAtom);
6065 assert(target != NULL);
6066 return (char*)target->contentPointer();
6067 }
6068
6069
6070 template <typename A>
6071 const char* PointerToCStringSection<A>::targetCString(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6072 {
6073 assert(atom->fixupCount() == 1);
6074 ld::Fixup::iterator fit = atom->fixupsBegin();
6075 const ld::Atom* targetAtom = NULL;
6076 switch ( fit->binding ) {
6077 case ld::Fixup::bindingByContentBound:
6078 targetAtom = fit->u.target;
6079 break;
6080 case ld::Fixup::bindingsIndirectlyBound:
6081 targetAtom = ind.indirectAtom(fit->u.bindingIndex);
6082 break;
6083 case ld::Fixup::bindingDirectlyBound:
6084 targetAtom = fit->u.target;
6085 break;
6086 default:
6087 assert(0 && "unsupported reference to selector");
6088 }
6089 assert(targetAtom != NULL);
6090 const Atom<A>* target = dynamic_cast<const Atom<A>*>(targetAtom);
6091 assert(target != NULL);
6092 assert(target->contentType() == ld::Atom::typeCString);
6093 return (char*)target->contentPointer();
6094 }
6095
6096 template <typename A>
6097 unsigned long PointerToCStringSection<A>::contentHash(const class Atom<A>* atom,
6098 const ld::IndirectBindingTable& indirectBindingTable) const
6099 {
6100 // make hash from section name and target cstring name
6101 unsigned long hash = 123;
6102 for (const char* s = this->sectionName(); *s != '\0'; ++s) {
6103 hash = hash * 33 + *s;
6104 }
6105 for (const char* s = this->targetCString(atom, indirectBindingTable); *s != '\0'; ++s) {
6106 hash = hash * 33 + *s;
6107 }
6108 return hash;
6109 }
6110
6111 template <typename A>
6112 bool PointerToCStringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
6113 const ld::IndirectBindingTable& indirectBindingTable) const
6114 {
6115 assert(this->type() == rhs.section().type());
6116 // there can be pointers-to-cstrings in different section names
6117 // we only want to coalesce in same section name
6118 if ( *this != rhs.section() )
6119 return false;
6120
6121 // get string content for this
6122 const char* cstringContent = this->targetCString(atom, indirectBindingTable);
6123 const Atom<A>* rhsAtom = dynamic_cast<const Atom<A>*>(&rhs);
6124 assert(rhsAtom != NULL);
6125 const char* rhsCstringContent = this->targetCString(rhsAtom, indirectBindingTable);
6126
6127 assert(cstringContent != NULL);
6128 assert(rhsCstringContent != NULL);
6129 return (strcmp(cstringContent, rhsCstringContent) == 0);
6130 }
6131
6132
6133
6134 template <typename A>
6135 unsigned long UTF16StringSection<A>::contentHash(const class Atom<A>* atom, const ld::IndirectBindingTable& ind) const
6136 {
6137 unsigned long hash = 5381;
6138 const uint16_t* stringContent = (uint16_t*)atom->contentPointer();
6139 // some buggy compilers end utf16 data with single byte, so don't use last word in hash computation
6140 unsigned int count = (atom->size()/2) - 1;
6141 for (const uint16_t* s = stringContent; count > 0; ++s, --count) {
6142 hash = hash * 33 + *s;
6143 }
6144 return hash;
6145 }
6146
6147 template <typename A>
6148 bool UTF16StringSection<A>::canCoalesceWith(const class Atom<A>* atom, const ld::Atom& rhs,
6149 const ld::IndirectBindingTable& ind) const
6150 {
6151 if ( rhs.section().type() != ld::Section::typeUTF16Strings )
6152 return false;
6153 assert(0);
6154 return false;
6155 }
6156
6157
6158
6159
6160
6161
6162
6163 template <>
6164 uint32_t Section<x86_64>::x86_64PcRelOffset(uint8_t r_type)
6165 {
6166 switch ( r_type ) {
6167 case X86_64_RELOC_SIGNED:
6168 return 4;
6169 case X86_64_RELOC_SIGNED_1:
6170 return 5;
6171 case X86_64_RELOC_SIGNED_2:
6172 return 6;
6173 case X86_64_RELOC_SIGNED_4:
6174 return 8;
6175 }
6176 return 0;
6177 }
6178
6179
6180 template <>
6181 bool Section<x86_64>::addRelocFixup(class Parser<x86_64>& parser, const macho_relocation_info<P>* reloc)
6182 {
6183 const macho_section<P>* sect = this->machoSection();
6184 uint64_t srcAddr = sect->addr() + reloc->r_address();
6185 Parser<x86_64>::SourceLocation src;
6186 Parser<x86_64>::TargetDesc target;
6187 Parser<x86_64>::TargetDesc toTarget;
6188 src.atom = this->findAtomByAddress(srcAddr);
6189 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6190 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
6191 uint64_t contentValue = 0;
6192 const macho_relocation_info<x86_64::P>* nextReloc = &reloc[1];
6193 bool result = false;
6194 bool useDirectBinding;
6195 switch ( reloc->r_length() ) {
6196 case 0:
6197 contentValue = *fixUpPtr;
6198 break;
6199 case 1:
6200 contentValue = (int64_t)(int16_t)E::get16(*((uint16_t*)fixUpPtr));
6201 break;
6202 case 2:
6203 contentValue = (int64_t)(int32_t)E::get32(*((uint32_t*)fixUpPtr));
6204 break;
6205 case 3:
6206 contentValue = E::get64(*((uint64_t*)fixUpPtr));
6207 break;
6208 }
6209 target.atom = NULL;
6210 target.name = NULL;
6211 target.weakImport = false;
6212 target.addend = 0;
6213 if ( reloc->r_extern() ) {
6214 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
6215 // use direct reference for local symbols
6216 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(sym)[0] == 'L')) ) {
6217 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
6218 target.addend += contentValue;
6219 }
6220 else {
6221 target.name = parser.nameFromSymbol(sym);
6222 target.weakImport = parser.weakImportFromSymbol(sym);
6223 target.addend = contentValue;
6224 }
6225 // cfstrings should always use direct reference to backing store
6226 if ( (this->type() == ld::Section::typeCFString) && (src.offsetInAtom != 0) ) {
6227 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
6228 target.addend = contentValue;
6229 }
6230 }
6231 else {
6232 if ( reloc->r_pcrel() )
6233 contentValue += srcAddr + x86_64PcRelOffset(reloc->r_type());
6234 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
6235 }
6236 switch ( reloc->r_type() ) {
6237 case X86_64_RELOC_UNSIGNED:
6238 if ( reloc->r_pcrel() )
6239 throw "pcrel and X86_64_RELOC_UNSIGNED not supported";
6240 switch ( reloc->r_length() ) {
6241 case 0:
6242 case 1:
6243 throw "length < 2 and X86_64_RELOC_UNSIGNED not supported";
6244 case 2:
6245 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6246 break;
6247 case 3:
6248 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian64, target);
6249 break;
6250 }
6251 break;
6252 case X86_64_RELOC_SIGNED:
6253 case X86_64_RELOC_SIGNED_1:
6254 case X86_64_RELOC_SIGNED_2:
6255 case X86_64_RELOC_SIGNED_4:
6256 if ( ! reloc->r_pcrel() )
6257 throw "not pcrel and X86_64_RELOC_SIGNED* not supported";
6258 if ( reloc->r_length() != 2 )
6259 throw "length != 2 and X86_64_RELOC_SIGNED* not supported";
6260 switch ( reloc->r_type() ) {
6261 case X86_64_RELOC_SIGNED:
6262 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32, target);
6263 break;
6264 case X86_64_RELOC_SIGNED_1:
6265 if ( reloc->r_extern() )
6266 target.addend += 1;
6267 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32_1, target);
6268 break;
6269 case X86_64_RELOC_SIGNED_2:
6270 if ( reloc->r_extern() )
6271 target.addend += 2;
6272 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32_2, target);
6273 break;
6274 case X86_64_RELOC_SIGNED_4:
6275 if ( reloc->r_extern() )
6276 target.addend += 4;
6277 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32_4, target);
6278 break;
6279 }
6280 break;
6281 case X86_64_RELOC_BRANCH:
6282 if ( ! reloc->r_pcrel() )
6283 throw "not pcrel and X86_64_RELOC_BRANCH not supported";
6284 switch ( reloc->r_length() ) {
6285 case 2:
6286 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
6287 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceCallSiteNop, false, target.name);
6288 parser.addDtraceExtraInfos(src, &target.name[16]);
6289 }
6290 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
6291 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceIsEnableSiteClear, false, target.name);
6292 parser.addDtraceExtraInfos(src, &target.name[20]);
6293 }
6294 else {
6295 parser.addFixups(src, ld::Fixup::kindStoreX86BranchPCRel32, target);
6296 }
6297 break;
6298 case 0:
6299 parser.addFixups(src, ld::Fixup::kindStoreX86BranchPCRel8, target);
6300 break;
6301 default:
6302 throwf("length=%d and X86_64_RELOC_BRANCH not supported", reloc->r_length());
6303 }
6304 break;
6305 case X86_64_RELOC_GOT:
6306 if ( ! reloc->r_extern() )
6307 throw "not extern and X86_64_RELOC_GOT not supported";
6308 if ( ! reloc->r_pcrel() )
6309 throw "not pcrel and X86_64_RELOC_GOT not supported";
6310 if ( reloc->r_length() != 2 )
6311 throw "length != 2 and X86_64_RELOC_GOT not supported";
6312 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32GOT, target);
6313 break;
6314 case X86_64_RELOC_GOT_LOAD:
6315 if ( ! reloc->r_extern() )
6316 throw "not extern and X86_64_RELOC_GOT_LOAD not supported";
6317 if ( ! reloc->r_pcrel() )
6318 throw "not pcrel and X86_64_RELOC_GOT_LOAD not supported";
6319 if ( reloc->r_length() != 2 )
6320 throw "length != 2 and X86_64_RELOC_GOT_LOAD not supported";
6321 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32GOTLoad, target);
6322 break;
6323 case X86_64_RELOC_SUBTRACTOR:
6324 if ( reloc->r_pcrel() )
6325 throw "X86_64_RELOC_SUBTRACTOR cannot be pc-relative";
6326 if ( reloc->r_length() < 2 )
6327 throw "X86_64_RELOC_SUBTRACTOR must have r_length of 2 or 3";
6328 if ( !reloc->r_extern() )
6329 throw "X86_64_RELOC_SUBTRACTOR must have r_extern=1";
6330 if ( nextReloc->r_type() != X86_64_RELOC_UNSIGNED )
6331 throw "X86_64_RELOC_SUBTRACTOR must be followed by X86_64_RELOC_UNSIGNED";
6332 result = true;
6333 if ( nextReloc->r_pcrel() )
6334 throw "X86_64_RELOC_UNSIGNED following a X86_64_RELOC_SUBTRACTOR cannot be pc-relative";
6335 if ( nextReloc->r_length() != reloc->r_length() )
6336 throw "X86_64_RELOC_UNSIGNED following a X86_64_RELOC_SUBTRACTOR must have same r_length";
6337 if ( nextReloc->r_extern() ) {
6338 const macho_nlist<P>& sym = parser.symbolFromIndex(nextReloc->r_symbolnum());
6339 // use direct reference for local symbols
6340 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(sym)[0] == 'L')) ) {
6341 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), toTarget);
6342 toTarget.addend = contentValue;
6343 useDirectBinding = true;
6344 }
6345 else {
6346 toTarget.name = parser.nameFromSymbol(sym);
6347 toTarget.weakImport = parser.weakImportFromSymbol(sym);
6348 toTarget.addend = contentValue;
6349 useDirectBinding = false;
6350 }
6351 }
6352 else {
6353 parser.findTargetFromAddressAndSectionNum(contentValue, nextReloc->r_symbolnum(), toTarget);
6354 useDirectBinding = (toTarget.atom->scope() == ld::Atom::scopeTranslationUnit) || ((toTarget.atom->combine() == ld::Atom::combineByNameAndContent) || (toTarget.atom->combine() == ld::Atom::combineByNameAndReferences));
6355 }
6356 if ( useDirectBinding ) {
6357 if ( (toTarget.atom->combine() == ld::Atom::combineByNameAndContent) || (toTarget.atom->combine() == ld::Atom::combineByNameAndReferences) )
6358 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, toTarget.atom);
6359 else
6360 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.atom);
6361 }
6362 else
6363 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.weakImport, toTarget.name);
6364 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindAddAddend, toTarget.addend);
6365 if ( target.atom == NULL )
6366 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, false, target.name);
6367 else
6368 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, target.atom);
6369 if ( reloc->r_length() == 2 )
6370 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
6371 else
6372 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian64);
6373 break;
6374 case X86_64_RELOC_TLV:
6375 if ( ! reloc->r_extern() )
6376 throw "not extern and X86_64_RELOC_TLV not supported";
6377 if ( ! reloc->r_pcrel() )
6378 throw "not pcrel and X86_64_RELOC_TLV not supported";
6379 if ( reloc->r_length() != 2 )
6380 throw "length != 2 and X86_64_RELOC_TLV not supported";
6381 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32TLVLoad, target);
6382 break;
6383 default:
6384 throwf("unknown relocation type %d", reloc->r_type());
6385 }
6386 return result;
6387 }
6388
6389
6390
6391 template <>
6392 bool Section<x86>::addRelocFixup(class Parser<x86>& parser, const macho_relocation_info<P>* reloc)
6393 {
6394 const macho_section<P>* sect = this->machoSection();
6395 uint32_t srcAddr;
6396 const uint8_t* fixUpPtr;
6397 uint32_t contentValue = 0;
6398 ld::Fixup::Kind kind = ld::Fixup::kindNone;
6399 Parser<x86>::SourceLocation src;
6400 Parser<x86>::TargetDesc target;
6401
6402 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
6403 srcAddr = sect->addr() + reloc->r_address();
6404 src.atom = this->findAtomByAddress(srcAddr);
6405 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6406 fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
6407 switch ( reloc->r_type() ) {
6408 case GENERIC_RELOC_VANILLA:
6409 switch ( reloc->r_length() ) {
6410 case 0:
6411 contentValue = (int32_t)(int8_t)*fixUpPtr;
6412 if ( reloc->r_pcrel() ) {
6413 kind = ld::Fixup::kindStoreX86BranchPCRel8;
6414 contentValue += srcAddr + sizeof(uint8_t);
6415 }
6416 else
6417 throw "r_length=0 and r_pcrel=0 not supported";
6418 break;
6419 case 1:
6420 contentValue = (int32_t)(int16_t)E::get16(*((uint16_t*)fixUpPtr));
6421 if ( reloc->r_pcrel() ) {
6422 kind = ld::Fixup::kindStoreX86PCRel16;
6423 contentValue += srcAddr + sizeof(uint16_t);
6424 }
6425 else
6426 kind = ld::Fixup::kindStoreLittleEndian16;
6427 break;
6428 case 2:
6429 contentValue = E::get32(*((uint32_t*)fixUpPtr));
6430 if ( reloc->r_pcrel() ) {
6431 kind = ld::Fixup::kindStoreX86BranchPCRel32;
6432 contentValue += srcAddr + sizeof(uint32_t);
6433 }
6434 else
6435 kind = ld::Fixup::kindStoreLittleEndian32;
6436 break;
6437 case 3:
6438 throw "r_length=3 not supported";
6439 }
6440 if ( reloc->r_extern() ) {
6441 target.atom = NULL;
6442 const macho_nlist<P>& targetSymbol = parser.symbolFromIndex(reloc->r_symbolnum());
6443 target.name = parser.nameFromSymbol(targetSymbol);
6444 target.weakImport = parser.weakImportFromSymbol(targetSymbol);
6445 target.addend = (int32_t)contentValue;
6446 }
6447 else {
6448 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
6449 }
6450 if ( (kind == ld::Fixup::kindStoreX86BranchPCRel32) && (target.name != NULL) ) {
6451 if ( strncmp(target.name, "___dtrace_probe$", 16) == 0 ) {
6452 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceCallSiteNop, false, target.name);
6453 parser.addDtraceExtraInfos(src, &target.name[16]);
6454 return false;
6455 }
6456 else if ( strncmp(target.name, "___dtrace_isenabled$", 20) == 0 ) {
6457 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreX86DtraceIsEnableSiteClear, false, target.name);
6458 parser.addDtraceExtraInfos(src, &target.name[20]);
6459 return false;
6460 }
6461 }
6462 parser.addFixups(src, kind, target);
6463 return false;
6464 break;
6465 case GENERIC_RLEOC_TLV:
6466 {
6467 if ( !reloc->r_extern() )
6468 throw "r_extern=0 and r_type=GENERIC_RLEOC_TLV not supported";
6469 if ( reloc->r_length() != 2 )
6470 throw "r_length!=2 and r_type=GENERIC_RLEOC_TLV not supported";
6471 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
6472 // use direct reference for local symbols
6473 if ( ((sym.n_type() & N_TYPE) == N_SECT) && ((sym.n_type() & N_EXT) == 0) ) {
6474 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
6475 }
6476 else {
6477 target.atom = NULL;
6478 target.name = parser.nameFromSymbol(sym);
6479 target.weakImport = parser.weakImportFromSymbol(sym);
6480 }
6481 target.addend = (int64_t)(int32_t)E::get32(*((uint32_t*)fixUpPtr));
6482 if ( reloc->r_pcrel() ) {
6483 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32TLVLoad, target);
6484 }
6485 else {
6486 parser.addFixups(src, ld::Fixup::kindStoreX86Abs32TLVLoad, target);
6487 }
6488 return false;
6489 }
6490 break;
6491 default:
6492 throwf("unsupported i386 relocation type (%d)", reloc->r_type());
6493 }
6494 }
6495 else {
6496 // scattered relocation
6497 const macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
6498 srcAddr = sect->addr() + sreloc->r_address();
6499 src.atom = this->findAtomByAddress(srcAddr);
6500 assert(src.atom != NULL);
6501 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6502 fixUpPtr = file().fileContent() + sect->offset() + sreloc->r_address();
6503 uint32_t relocValue = sreloc->r_value();
6504 bool result = false;
6505 // file format allows pair to be scattered or not
6506 const macho_scattered_relocation_info<P>* nextSReloc = &sreloc[1];
6507 const macho_relocation_info<P>* nextReloc = &reloc[1];
6508 bool nextRelocIsPair = false;
6509 uint32_t nextRelocAddress = 0;
6510 uint32_t nextRelocValue = 0;
6511 if ( (nextReloc->r_address() & R_SCATTERED) == 0 ) {
6512 if ( nextReloc->r_type() == GENERIC_RELOC_PAIR ) {
6513 nextRelocIsPair = true;
6514 nextRelocAddress = nextReloc->r_address();
6515 result = true; // iterator should skip next reloc, since we've consumed it here
6516 }
6517 }
6518 else {
6519 if ( nextSReloc->r_type() == GENERIC_RELOC_PAIR ) {
6520 nextRelocIsPair = true;
6521 nextRelocAddress = nextSReloc->r_address();
6522 nextRelocValue = nextSReloc->r_value();
6523 }
6524 }
6525 switch (sreloc->r_type()) {
6526 case GENERIC_RELOC_VANILLA:
6527 // with a scattered relocation we get both the target (sreloc->r_value()) and the target+offset (*fixUpPtr)
6528 target.atom = parser.findAtomByAddress(relocValue);
6529 if ( sreloc->r_pcrel() ) {
6530 switch ( sreloc->r_length() ) {
6531 case 0:
6532 contentValue = srcAddr + 1 + *fixUpPtr;
6533 target.addend = (int32_t)contentValue - (int32_t)relocValue;
6534 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel8, target);
6535 break;
6536 case 1:
6537 contentValue = srcAddr + 2 + LittleEndian::get16(*((uint16_t*)fixUpPtr));
6538 target.addend = (int32_t)contentValue - (int32_t)relocValue;
6539 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel16, target);
6540 break;
6541 case 2:
6542 contentValue = srcAddr + 4 + LittleEndian::get32(*((uint32_t*)fixUpPtr));
6543 target.addend = (int32_t)contentValue - (int32_t)relocValue;
6544 parser.addFixups(src, ld::Fixup::kindStoreX86PCRel32, target);
6545 break;
6546 case 3:
6547 throw "unsupported r_length=3 for scattered pc-rel vanilla reloc";
6548 break;
6549 }
6550 }
6551 else {
6552 if ( sreloc->r_length() != 2 )
6553 throwf("unsupported r_length=%d for scattered vanilla reloc", sreloc->r_length());
6554 contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
6555 target.addend = (int32_t)contentValue - (int32_t)(target.atom->objectAddress());
6556 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6557 }
6558 break;
6559 case GENERIC_RELOC_SECTDIFF:
6560 case GENERIC_RELOC_LOCAL_SECTDIFF:
6561 {
6562 if ( !nextRelocIsPair )
6563 throw "GENERIC_RELOC_SECTDIFF missing following pair";
6564 switch ( sreloc->r_length() ) {
6565 case 0:
6566 case 3:
6567 throw "bad length for GENERIC_RELOC_SECTDIFF";
6568 case 1:
6569 contentValue = (int32_t)(int16_t)LittleEndian::get16(*((uint16_t*)fixUpPtr));
6570 kind = ld::Fixup::kindStoreLittleEndian16;
6571 break;
6572 case 2:
6573 contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
6574 kind = ld::Fixup::kindStoreLittleEndian32;
6575 break;
6576 }
6577 Atom<x86>* fromAtom = parser.findAtomByAddress(nextRelocValue);
6578 uint32_t offsetInFrom = nextRelocValue - fromAtom->_objAddress;
6579 parser.findTargetFromAddress(sreloc->r_value(), target);
6580 // check for addend encoded in the section content
6581 int64_t addend = (int32_t)contentValue - (int32_t)(sreloc->r_value() - nextRelocValue);
6582 if ( addend < 0 ) {
6583 // switch binding base on coalescing
6584 if ( target.atom == NULL ) {
6585 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.name);
6586 }
6587 else if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
6588 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, target.atom);
6589 }
6590 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
6591 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, target.atom);
6592 }
6593 else {
6594 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.atom->name());
6595 }
6596 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, target.addend);
6597 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
6598 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom-addend);
6599 parser.addFixup(src, ld::Fixup::k5of5, kind);
6600 }
6601 else {
6602 // switch binding base on coalescing
6603 if ( target.atom == NULL ) {
6604 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.name);
6605 }
6606 else if ( target.atom->scope() == ld::Atom::scopeTranslationUnit ) {
6607 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, target.atom);
6608 }
6609 else if ( (target.atom->combine() == ld::Atom::combineByNameAndContent) || (target.atom->combine() == ld::Atom::combineByNameAndReferences) ) {
6610 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, target.atom);
6611 }
6612 else {
6613 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, target.atom->name());
6614 }
6615 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, target.addend+addend);
6616 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
6617 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom);
6618 parser.addFixup(src, ld::Fixup::k5of5, kind);
6619 }
6620 }
6621 break;
6622 }
6623 return result;
6624 }
6625 }
6626
6627
6628
6629
6630
6631 #if SUPPORT_ARCH_arm_any
6632 template <>
6633 bool Section<arm>::addRelocFixup(class Parser<arm>& parser, const macho_relocation_info<P>* reloc)
6634 {
6635 const macho_section<P>* sect = this->machoSection();
6636 bool result = false;
6637 uint32_t srcAddr;
6638 uint32_t dstAddr;
6639 uint32_t* fixUpPtr;
6640 int32_t displacement = 0;
6641 uint32_t instruction = 0;
6642 pint_t contentValue = 0;
6643 Parser<arm>::SourceLocation src;
6644 Parser<arm>::TargetDesc target;
6645 const macho_relocation_info<P>* nextReloc;
6646
6647 if ( (reloc->r_address() & R_SCATTERED) == 0 ) {
6648 bool externSymbolIsThumbDef = false;
6649 srcAddr = sect->addr() + reloc->r_address();
6650 src.atom = this->findAtomByAddress(srcAddr);
6651 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6652 fixUpPtr = (uint32_t*)(file().fileContent() + sect->offset() + reloc->r_address());
6653 if ( reloc->r_type() != ARM_RELOC_PAIR )
6654 instruction = LittleEndian::get32(*fixUpPtr);
6655 if ( reloc->r_extern() ) {
6656 const macho_nlist<P>& targetSymbol = parser.symbolFromIndex(reloc->r_symbolnum());
6657 // use direct reference for local symbols
6658 if ( ((targetSymbol.n_type() & N_TYPE) == N_SECT) && (((targetSymbol.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(targetSymbol)[0] == 'L')) ) {
6659 parser.findTargetFromAddressAndSectionNum(targetSymbol.n_value(), targetSymbol.n_sect(), target);
6660 }
6661 else {
6662 target.atom = NULL;
6663 target.name = parser.nameFromSymbol(targetSymbol);
6664 target.weakImport = parser.weakImportFromSymbol(targetSymbol);
6665 if ( ((targetSymbol.n_type() & N_TYPE) == N_SECT) && (targetSymbol.n_desc() & N_ARM_THUMB_DEF) )
6666 externSymbolIsThumbDef = true;
6667 }
6668 }
6669 switch ( reloc->r_type() ) {
6670 case ARM_RELOC_BR24:
6671 // Sign-extend displacement
6672 displacement = (instruction & 0x00FFFFFF) << 2;
6673 if ( (displacement & 0x02000000) != 0 )
6674 displacement |= 0xFC000000;
6675 // The pc added will be +8 from the pc
6676 displacement += 8;
6677 // If this is BLX add H << 1
6678 if ((instruction & 0xFE000000) == 0xFA000000)
6679 displacement += ((instruction & 0x01000000) >> 23);
6680 if ( reloc->r_extern() ) {
6681 dstAddr = srcAddr + displacement;
6682 // <rdar://problem/16652542> support large .o files
6683 if ( srcAddr > 0x2000000 ) {
6684 dstAddr -= ((srcAddr + 0x1FFFFFF) & 0xFC000000);
6685 }
6686 target.addend = dstAddr;
6687 if ( externSymbolIsThumbDef )
6688 target.addend &= -2; // remove thumb bit
6689 }
6690 else {
6691 dstAddr = srcAddr + displacement;
6692 parser.findTargetFromAddressAndSectionNum(dstAddr, reloc->r_symbolnum(), target);
6693 }
6694 // special case "calls" for dtrace
6695 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
6696 parser.addFixup(src, ld::Fixup::k1of1,
6697 ld::Fixup::kindStoreARMDtraceCallSiteNop, false, target.name);
6698 parser.addDtraceExtraInfos(src, &target.name[16]);
6699 }
6700 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
6701 parser.addFixup(src, ld::Fixup::k1of1,
6702 ld::Fixup::kindStoreARMDtraceIsEnableSiteClear, false, target.name);
6703 parser.addDtraceExtraInfos(src, &target.name[20]);
6704 }
6705 else {
6706 parser.addFixups(src, ld::Fixup::kindStoreARMBranch24, target);
6707 }
6708 break;
6709 case ARM_THUMB_RELOC_BR22:
6710 // thumb2 added two more bits to displacement, complicating the displacement decoding
6711 {
6712 uint32_t s = (instruction >> 10) & 0x1;
6713 uint32_t j1 = (instruction >> 29) & 0x1;
6714 uint32_t j2 = (instruction >> 27) & 0x1;
6715 uint32_t imm10 = instruction & 0x3FF;
6716 uint32_t imm11 = (instruction >> 16) & 0x7FF;
6717 uint32_t i1 = (j1 == s);
6718 uint32_t i2 = (j2 == s);
6719 uint32_t dis = (s << 24) | (i1 << 23) | (i2 << 22) | (imm10 << 12) | (imm11 << 1);
6720 int32_t sdis = dis;
6721 if ( s )
6722 sdis |= 0xFE000000;
6723 displacement = sdis;
6724 }
6725 // The pc added will be +4 from the pc
6726 displacement += 4;
6727 // If the instruction was blx, force the low 2 bits to be clear
6728 dstAddr = srcAddr + displacement;
6729 if ((instruction & 0xD0000000) == 0xC0000000)
6730 dstAddr &= 0xFFFFFFFC;
6731
6732 if ( reloc->r_extern() ) {
6733 // <rdar://problem/16652542> support large .o files
6734 if ( srcAddr > 0x1000000 ) {
6735 dstAddr -= ((srcAddr + 0xFFFFFF) & 0xFE000000);
6736 }
6737 target.addend = (int64_t)(int32_t)dstAddr;
6738 }
6739 else {
6740 parser.findTargetFromAddressAndSectionNum(dstAddr, reloc->r_symbolnum(), target);
6741 }
6742 // special case "calls" for dtrace
6743 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
6744 parser.addFixup(src, ld::Fixup::k1of1,
6745 ld::Fixup::kindStoreThumbDtraceCallSiteNop, false, target.name);
6746 parser.addDtraceExtraInfos(src, &target.name[16]);
6747 }
6748 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
6749 parser.addFixup(src, ld::Fixup::k1of1,
6750 ld::Fixup::kindStoreThumbDtraceIsEnableSiteClear, false, target.name);
6751 parser.addDtraceExtraInfos(src, &target.name[20]);
6752 }
6753 else {
6754 parser.addFixups(src, ld::Fixup::kindStoreThumbBranch22, target);
6755 }
6756 break;
6757 case ARM_RELOC_VANILLA:
6758 if ( reloc->r_length() != 2 )
6759 throw "bad length for ARM_RELOC_VANILLA";
6760 contentValue = LittleEndian::get32(*fixUpPtr);
6761 if ( reloc->r_extern() ) {
6762 target.addend = (int32_t)contentValue;
6763 if ( externSymbolIsThumbDef )
6764 target.addend &= -2; // remove thumb bit
6765 }
6766 else {
6767 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
6768 // possible non-extern relocation turned into by-name ref because target is a weak-def
6769 if ( target.atom != NULL ) {
6770 if ( target.atom->isThumb() )
6771 target.addend &= -2; // remove thumb bit
6772 // if reference to LSDA, add group subordinate fixup
6773 if ( target.atom->contentType() == ld::Atom::typeLSDA ) {
6774 Parser<arm>::SourceLocation src2;
6775 src2.atom = src.atom;
6776 src2.offsetInAtom = 0;
6777 parser.addFixup(src2, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, target.atom);
6778 }
6779 }
6780 }
6781 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6782 break;
6783 case ARM_THUMB_32BIT_BRANCH:
6784 // silently ignore old unnecessary reloc
6785 break;
6786 case ARM_RELOC_HALF:
6787 nextReloc = &reloc[1];
6788 if ( nextReloc->r_type() == ARM_RELOC_PAIR ) {
6789 uint32_t instruction16;
6790 uint32_t other16 = (nextReloc->r_address() & 0xFFFF);
6791 bool isThumb;
6792 if ( reloc->r_length() & 2 ) {
6793 isThumb = true;
6794 uint32_t i = ((instruction & 0x00000400) >> 10);
6795 uint32_t imm4 = (instruction & 0x0000000F);
6796 uint32_t imm3 = ((instruction & 0x70000000) >> 28);
6797 uint32_t imm8 = ((instruction & 0x00FF0000) >> 16);
6798 instruction16 = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
6799 }
6800 else {
6801 isThumb = false;
6802 uint32_t imm4 = ((instruction & 0x000F0000) >> 16);
6803 uint32_t imm12 = (instruction & 0x00000FFF);
6804 instruction16 = (imm4 << 12) | imm12;
6805 }
6806 if ( reloc->r_length() & 1 ) {
6807 // high 16
6808 dstAddr = ((instruction16 << 16) | other16);
6809 if ( reloc->r_extern() ) {
6810 target.addend = dstAddr;
6811 if ( externSymbolIsThumbDef )
6812 target.addend &= -2; // remove thumb bit
6813 }
6814 else {
6815 parser.findTargetFromAddress(dstAddr, target);
6816 if ( target.atom->isThumb() )
6817 target.addend &= (-2); // remove thumb bit
6818 }
6819 parser.addFixups(src, (isThumb ? ld::Fixup::kindStoreThumbHigh16 : ld::Fixup::kindStoreARMHigh16), target);
6820 }
6821 else {
6822 // low 16
6823 dstAddr = (other16 << 16) | instruction16;
6824 if ( reloc->r_extern() ) {
6825 target.addend = dstAddr;
6826 if ( externSymbolIsThumbDef )
6827 target.addend &= -2; // remove thumb bit
6828 }
6829 else {
6830 parser.findTargetFromAddress(dstAddr, target);
6831 if ( target.atom->isThumb() )
6832 target.addend &= (-2); // remove thumb bit
6833 }
6834 parser.addFixups(src, (isThumb ? ld::Fixup::kindStoreThumbLow16 : ld::Fixup::kindStoreARMLow16), target);
6835 }
6836 result = true;
6837 }
6838 else
6839 throw "for ARM_RELOC_HALF, next reloc is not ARM_RELOC_PAIR";
6840 break;
6841 default:
6842 throwf("unknown relocation type %d", reloc->r_type());
6843 break;
6844 }
6845 }
6846 else {
6847 const macho_scattered_relocation_info<P>* sreloc = (macho_scattered_relocation_info<P>*)reloc;
6848 // file format allows pair to be scattered or not
6849 const macho_scattered_relocation_info<P>* nextSReloc = &sreloc[1];
6850 nextReloc = &reloc[1];
6851 srcAddr = sect->addr() + sreloc->r_address();
6852 dstAddr = sreloc->r_value();
6853 fixUpPtr = (uint32_t*)(file().fileContent() + sect->offset() + sreloc->r_address());
6854 instruction = LittleEndian::get32(*fixUpPtr);
6855 src.atom = this->findAtomByAddress(srcAddr);
6856 src.offsetInAtom = srcAddr - src.atom->_objAddress;
6857 bool nextRelocIsPair = false;
6858 uint32_t nextRelocAddress = 0;
6859 uint32_t nextRelocValue = 0;
6860 if ( (nextReloc->r_address() & R_SCATTERED) == 0 ) {
6861 if ( nextReloc->r_type() == ARM_RELOC_PAIR ) {
6862 nextRelocIsPair = true;
6863 nextRelocAddress = nextReloc->r_address();
6864 result = true;
6865 }
6866 }
6867 else {
6868 if ( nextSReloc->r_type() == ARM_RELOC_PAIR ) {
6869 nextRelocIsPair = true;
6870 nextRelocAddress = nextSReloc->r_address();
6871 nextRelocValue = nextSReloc->r_value();
6872 result = true;
6873 }
6874 }
6875 switch ( sreloc->r_type() ) {
6876 case ARM_RELOC_VANILLA:
6877 // with a scattered relocation we get both the target (sreloc->r_value()) and the target+offset (*fixUpPtr)
6878 if ( sreloc->r_length() != 2 )
6879 throw "bad length for ARM_RELOC_VANILLA";
6880 target.atom = parser.findAtomByAddress(sreloc->r_value());
6881 if ( target.atom == NULL )
6882 throwf("bad r_value (0x%08X) for ARM_RELOC_VANILLA\n", sreloc->r_value());
6883 contentValue = LittleEndian::get32(*fixUpPtr);
6884 target.addend = contentValue - target.atom->_objAddress;
6885 if ( target.atom->isThumb() )
6886 target.addend &= -2; // remove thumb bit
6887 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
6888 break;
6889 case ARM_RELOC_BR24:
6890 // Sign-extend displacement
6891 displacement = (instruction & 0x00FFFFFF) << 2;
6892 if ( (displacement & 0x02000000) != 0 )
6893 displacement |= 0xFC000000;
6894 // The pc added will be +8 from the pc
6895 displacement += 8;
6896 // If this is BLX add H << 1
6897 if ((instruction & 0xFE000000) == 0xFA000000)
6898 displacement += ((instruction & 0x01000000) >> 23);
6899 target.atom = parser.findAtomByAddress(sreloc->r_value());
6900 target.addend = (int64_t)(srcAddr + displacement) - (int64_t)(target.atom->_objAddress);
6901 parser.addFixups(src, ld::Fixup::kindStoreARMBranch24, target);
6902 break;
6903 case ARM_THUMB_RELOC_BR22:
6904 // thumb2 added two more bits to displacement, complicating the displacement decoding
6905 {
6906 uint32_t s = (instruction >> 10) & 0x1;
6907 uint32_t j1 = (instruction >> 29) & 0x1;
6908 uint32_t j2 = (instruction >> 27) & 0x1;
6909 uint32_t imm10 = instruction & 0x3FF;
6910 uint32_t imm11 = (instruction >> 16) & 0x7FF;
6911 uint32_t i1 = (j1 == s);
6912 uint32_t i2 = (j2 == s);
6913 uint32_t dis = (s << 24) | (i1 << 23) | (i2 << 22) | (imm10 << 12) | (imm11 << 1);
6914 int32_t sdis = dis;
6915 if ( s )
6916 sdis |= 0xFE000000;
6917 displacement = sdis;
6918 }
6919 // The pc added will be +4 from the pc
6920 displacement += 4;
6921 dstAddr = srcAddr+displacement;
6922 // If the instruction was blx, force the low 2 bits to be clear
6923 if ((instruction & 0xF8000000) == 0xE8000000)
6924 dstAddr &= 0xFFFFFFFC;
6925 target.atom = parser.findAtomByAddress(sreloc->r_value());
6926 target.addend = dstAddr - target.atom->_objAddress;
6927 parser.addFixups(src, ld::Fixup::kindStoreThumbBranch22, target);
6928 break;
6929 case ARM_RELOC_SECTDIFF:
6930 case ARM_RELOC_LOCAL_SECTDIFF:
6931 {
6932 if ( ! nextRelocIsPair )
6933 throw "ARM_RELOC_SECTDIFF missing following pair";
6934 if ( sreloc->r_length() != 2 )
6935 throw "bad length for ARM_RELOC_SECTDIFF";
6936 contentValue = LittleEndian::get32(*fixUpPtr);
6937 Atom<arm>* fromAtom = parser.findAtomByAddress(nextRelocValue);
6938 uint32_t offsetInFrom = nextRelocValue - fromAtom->_objAddress;
6939 uint32_t offsetInTarget;
6940 Atom<arm>* targetAtom = parser.findAtomByAddressOrLocalTargetOfStub(sreloc->r_value(), &offsetInTarget);
6941 // check for addend encoded in the section content
6942 int64_t addend = (int32_t)contentValue - (int32_t)(sreloc->r_value() - nextRelocValue);
6943 if ( targetAtom->isThumb() )
6944 addend &= -2; // remove thumb bit
6945 // if reference to LSDA, add group subordinate fixup
6946 if ( targetAtom->contentType() == ld::Atom::typeLSDA ) {
6947 Parser<arm>::SourceLocation src2;
6948 src2.atom = src.atom;
6949 src2.offsetInAtom = 0;
6950 parser.addFixup(src2, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinateLSDA, targetAtom);
6951 }
6952 if ( addend < 0 ) {
6953 // switch binding base on coalescing
6954 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
6955 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, targetAtom);
6956 }
6957 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
6958 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
6959 }
6960 else {
6961 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
6962 }
6963 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, offsetInTarget);
6964 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
6965 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom-addend);
6966 parser.addFixup(src, ld::Fixup::k5of5, ld::Fixup::kindStoreLittleEndian32);
6967 }
6968 else {
6969 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
6970 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, targetAtom);
6971 }
6972 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
6973 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
6974 }
6975 else {
6976 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
6977 }
6978 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, (uint32_t)(offsetInTarget+addend));
6979 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
6980 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom);
6981 parser.addFixup(src, ld::Fixup::k5of5, ld::Fixup::kindStoreLittleEndian32);
6982 }
6983 }
6984 break;
6985 case ARM_RELOC_HALF_SECTDIFF:
6986 if ( nextRelocIsPair ) {
6987 instruction = LittleEndian::get32(*fixUpPtr);
6988 Atom<arm>* fromAtom = parser.findAtomByAddress(nextRelocValue);
6989 uint32_t offsetInFrom = nextRelocValue - fromAtom->_objAddress;
6990 Atom<arm>* targetAtom = parser.findAtomByAddress(sreloc->r_value());
6991 uint32_t offsetInTarget = sreloc->r_value() - targetAtom->_objAddress;
6992 uint32_t instruction16;
6993 uint32_t other16 = (nextRelocAddress & 0xFFFF);
6994 bool isThumb;
6995 if ( sreloc->r_length() & 2 ) {
6996 isThumb = true;
6997 uint32_t i = ((instruction & 0x00000400) >> 10);
6998 uint32_t imm4 = (instruction & 0x0000000F);
6999 uint32_t imm3 = ((instruction & 0x70000000) >> 28);
7000 uint32_t imm8 = ((instruction & 0x00FF0000) >> 16);
7001 instruction16 = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
7002 }
7003 else {
7004 isThumb = false;
7005 uint32_t imm4 = ((instruction & 0x000F0000) >> 16);
7006 uint32_t imm12 = (instruction & 0x00000FFF);
7007 instruction16 = (imm4 << 12) | imm12;
7008 }
7009 if ( sreloc->r_length() & 1 )
7010 dstAddr = ((instruction16 << 16) | other16);
7011 else
7012 dstAddr = (other16 << 16) | instruction16;
7013 if ( targetAtom->isThumb() )
7014 dstAddr &= (-2); // remove thumb bit
7015 int32_t addend = dstAddr - (sreloc->r_value() - nextRelocValue);
7016 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
7017 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, targetAtom);
7018 }
7019 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
7020 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
7021 }
7022 else {
7023 parser.addFixup(src, ld::Fixup::k1of5, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
7024 }
7025 parser.addFixup(src, ld::Fixup::k2of5, ld::Fixup::kindAddAddend, (uint32_t)offsetInTarget+addend);
7026 parser.addFixup(src, ld::Fixup::k3of5, ld::Fixup::kindSubtractTargetAddress, fromAtom);
7027 parser.addFixup(src, ld::Fixup::k4of5, ld::Fixup::kindSubtractAddend, offsetInFrom);
7028 if ( sreloc->r_length() & 1 ) {
7029 // high 16
7030 parser.addFixup(src, ld::Fixup::k5of5, (isThumb ? ld::Fixup::kindStoreThumbHigh16 : ld::Fixup::kindStoreARMHigh16));
7031 }
7032 else {
7033 // low 16
7034 parser.addFixup(src, ld::Fixup::k5of5, (isThumb ? ld::Fixup::kindStoreThumbLow16 : ld::Fixup::kindStoreARMLow16));
7035 }
7036 result = true;
7037 }
7038 else
7039 throw "ARM_RELOC_HALF_SECTDIFF reloc missing following pair";
7040 break;
7041 case ARM_RELOC_HALF:
7042 if ( nextRelocIsPair ) {
7043 instruction = LittleEndian::get32(*fixUpPtr);
7044 Atom<arm>* targetAtom = parser.findAtomByAddress(sreloc->r_value());
7045 uint32_t instruction16;
7046 uint32_t other16 = (nextRelocAddress & 0xFFFF);
7047 bool isThumb;
7048 if ( sreloc->r_length() & 2 ) {
7049 isThumb = true;
7050 uint32_t i = ((instruction & 0x00000400) >> 10);
7051 uint32_t imm4 = (instruction & 0x0000000F);
7052 uint32_t imm3 = ((instruction & 0x70000000) >> 28);
7053 uint32_t imm8 = ((instruction & 0x00FF0000) >> 16);
7054 instruction16 = (imm4 << 12) | (i << 11) | (imm3 << 8) | imm8;
7055 }
7056 else {
7057 isThumb = false;
7058 uint32_t imm4 = ((instruction & 0x000F0000) >> 16);
7059 uint32_t imm12 = (instruction & 0x00000FFF);
7060 instruction16 = (imm4 << 12) | imm12;
7061 }
7062 if ( sreloc->r_length() & 1 )
7063 dstAddr = ((instruction16 << 16) | other16);
7064 else
7065 dstAddr = (other16 << 16) | instruction16;
7066 if ( targetAtom->scope() == ld::Atom::scopeTranslationUnit ) {
7067 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, targetAtom);
7068 }
7069 else if ( (targetAtom->combine() == ld::Atom::combineByNameAndContent) || (targetAtom->combine() == ld::Atom::combineByNameAndReferences) ) {
7070 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, ld::Fixup::bindingByContentBound, targetAtom);
7071 }
7072 else {
7073 parser.addFixup(src, ld::Fixup::k1of3, ld::Fixup::kindSetTargetAddress, false, targetAtom->name());
7074 }
7075 parser.addFixup(src, ld::Fixup::k2of3, ld::Fixup::kindAddAddend, dstAddr - targetAtom->_objAddress);
7076 if ( sreloc->r_length() & 1 ) {
7077 // high 16
7078 parser.addFixup(src, ld::Fixup::k3of3, (isThumb ? ld::Fixup::kindStoreThumbHigh16 : ld::Fixup::kindStoreARMHigh16));
7079 }
7080 else {
7081 // low 16
7082 parser.addFixup(src, ld::Fixup::k3of3, (isThumb ? ld::Fixup::kindStoreThumbLow16 : ld::Fixup::kindStoreARMLow16));
7083 }
7084 result = true;
7085 }
7086 else
7087 throw "scattered ARM_RELOC_HALF reloc missing following pair";
7088 break;
7089 default:
7090 throwf("unknown ARM scattered relocation type %d", sreloc->r_type());
7091 }
7092 }
7093 return result;
7094 }
7095 #endif
7096
7097
7098 #if SUPPORT_ARCH_arm64
7099 template <>
7100 bool Section<arm64>::addRelocFixup(class Parser<arm64>& parser, const macho_relocation_info<P>* reloc)
7101 {
7102 bool result = false;
7103 Parser<arm64>::SourceLocation src;
7104 Parser<arm64>::TargetDesc target = { NULL, NULL, false, 0 };
7105 Parser<arm64>::TargetDesc toTarget;
7106 int32_t prefixRelocAddend = 0;
7107 if ( reloc->r_type() == ARM64_RELOC_ADDEND ) {
7108 uint32_t rawAddend = reloc->r_symbolnum();
7109 prefixRelocAddend = rawAddend;
7110 if ( rawAddend & 0x00800000 )
7111 prefixRelocAddend |= 0xFF000000; // sign extend 24-bit signed int to 32-bits
7112 uint32_t addendAddress = reloc->r_address();
7113 ++reloc; //advance to next reloc record
7114 result = true;
7115 if ( reloc->r_address() != addendAddress )
7116 throw "ARM64_RELOC_ADDEND r_address does not match next reloc's r_address";
7117 }
7118 const macho_section<P>* sect = this->machoSection();
7119 uint64_t srcAddr = sect->addr() + reloc->r_address();
7120 src.atom = this->findAtomByAddress(srcAddr);
7121 src.offsetInAtom = srcAddr - src.atom->_objAddress;
7122 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
7123 uint64_t contentValue = 0;
7124 const macho_relocation_info<arm64::P>* nextReloc = &reloc[1];
7125 bool useDirectBinding;
7126 uint32_t instruction;
7127 uint32_t encodedAddend;
7128 switch ( reloc->r_length() ) {
7129 case 0:
7130 contentValue = *fixUpPtr;
7131 break;
7132 case 1:
7133 contentValue = (int64_t)(int16_t)E::get16(*((uint16_t*)fixUpPtr));
7134 break;
7135 case 2:
7136 contentValue = (int64_t)(int32_t)E::get32(*((uint32_t*)fixUpPtr));
7137 break;
7138 case 3:
7139 contentValue = E::get64(*((uint64_t*)fixUpPtr));
7140 break;
7141 }
7142 if ( reloc->r_extern() ) {
7143 const macho_nlist<P>& sym = parser.symbolFromIndex(reloc->r_symbolnum());
7144 const char* symbolName = parser.nameFromSymbol(sym);
7145 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (symbolName[0] == 'L') || (symbolName[0] == 'l')) ) {
7146 // use direct reference for local symbols
7147 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
7148 //target.addend += contentValue;
7149 }
7150 else if ( ((sym.n_type() & N_TYPE) == N_SECT) && (src.atom->_objAddress <= sym.n_value()) && (sym.n_value() < (src.atom->_objAddress+src.atom->size())) ) {
7151 // <rdar://problem/13700961> spurious warning when weak function has reference to itself
7152 // use direct reference when atom targets itself
7153 target.atom = src.atom;
7154 target.name = NULL;
7155 }
7156 else {
7157 target.name = symbolName;
7158 target.weakImport = parser.weakImportFromSymbol(sym);
7159 //target.addend = contentValue;
7160 }
7161 // cfstrings should always use direct reference to backing store
7162 if ( (this->type() == ld::Section::typeCFString) && (src.offsetInAtom != 0) ) {
7163 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), target);
7164 //target.addend = contentValue;
7165 }
7166 }
7167 else {
7168 if ( reloc->r_pcrel() )
7169 contentValue += srcAddr;
7170 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), target);
7171 }
7172 switch ( reloc->r_type() ) {
7173 case ARM64_RELOC_UNSIGNED:
7174 if ( reloc->r_pcrel() )
7175 throw "pcrel and ARM64_RELOC_UNSIGNED not supported";
7176 if ( reloc->r_extern() )
7177 target.addend = contentValue;
7178 switch ( reloc->r_length() ) {
7179 case 0:
7180 case 1:
7181 throw "length < 2 and ARM64_RELOC_UNSIGNED not supported";
7182 case 2:
7183 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian32, target);
7184 break;
7185 case 3:
7186 parser.addFixups(src, ld::Fixup::kindStoreLittleEndian64, target);
7187 break;
7188 }
7189 break;
7190 case ARM64_RELOC_BRANCH26:
7191 if ( ! reloc->r_pcrel() )
7192 throw "not pcrel and ARM64_RELOC_BRANCH26 not supported";
7193 if ( ! reloc->r_extern() )
7194 throw "r_extern == 0 and ARM64_RELOC_BRANCH26 not supported";
7195 if ( reloc->r_length() != 2 )
7196 throw "r_length != 2 and ARM64_RELOC_BRANCH26 not supported";
7197 if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_probe$", 16) == 0) ) {
7198 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreARM64DtraceCallSiteNop, false, target.name);
7199 parser.addDtraceExtraInfos(src, &target.name[16]);
7200 }
7201 else if ( (target.name != NULL) && (strncmp(target.name, "___dtrace_isenabled$", 20) == 0) ) {
7202 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindStoreARM64DtraceIsEnableSiteClear, false, target.name);
7203 parser.addDtraceExtraInfos(src, &target.name[20]);
7204 }
7205 else {
7206 target.addend = prefixRelocAddend;
7207 instruction = contentValue;
7208 encodedAddend = (instruction & 0x03FFFFFF) << 2;
7209 if ( encodedAddend != 0 ) {
7210 if ( prefixRelocAddend == 0 ) {
7211 warning("branch26 instruction at 0x%08X has embedded addend. ARM64_RELOC_ADDEND should be used instead", reloc->r_address());
7212 target.addend = encodedAddend;
7213 }
7214 else {
7215 throwf("branch26 instruction at 0x%08X has embedded addend and ARM64_RELOC_ADDEND also used", reloc->r_address());
7216 }
7217 }
7218 parser.addFixups(src, ld::Fixup::kindStoreARM64Branch26, target);
7219 }
7220 break;
7221 case ARM64_RELOC_PAGE21:
7222 if ( ! reloc->r_pcrel() )
7223 throw "not pcrel and ARM64_RELOC_PAGE21 not supported";
7224 if ( ! reloc->r_extern() )
7225 throw "r_extern == 0 and ARM64_RELOC_PAGE21 not supported";
7226 if ( reloc->r_length() != 2 )
7227 throw "length != 2 and ARM64_RELOC_PAGE21 not supported";
7228 target.addend = prefixRelocAddend;
7229 instruction = contentValue;
7230 encodedAddend = ((instruction & 0x60000000) >> 29) | ((instruction & 0x01FFFFE0) >> 3);
7231 encodedAddend *= 4096; // internally addend is in bytes, so scale
7232 if ( encodedAddend != 0 ) {
7233 if ( prefixRelocAddend == 0 ) {
7234 warning("adrp instruction at 0x%08X has embedded addend. ARM64_RELOC_ADDEND should be used instead", reloc->r_address());
7235 target.addend = encodedAddend;
7236 }
7237 else {
7238 throwf("adrp instruction at 0x%08X has embedded addend and ARM64_RELOC_ADDEND also used", reloc->r_address());
7239 }
7240 }
7241 parser.addFixups(src, ld::Fixup::kindStoreARM64Page21, target);
7242 break;
7243 case ARM64_RELOC_PAGEOFF12:
7244 if ( reloc->r_pcrel() )
7245 throw "pcrel and ARM64_RELOC_PAGEOFF12 not supported";
7246 if ( ! reloc->r_extern() )
7247 throw "r_extern == 0 and ARM64_RELOC_PAGEOFF12 not supported";
7248 if ( reloc->r_length() != 2 )
7249 throw "length != 2 and ARM64_RELOC_PAGEOFF12 not supported";
7250 target.addend = prefixRelocAddend;
7251 instruction = contentValue;
7252 encodedAddend = ((instruction & 0x003FFC00) >> 10);
7253 // internally addend is in bytes. Some instructions have an implicit scale factor
7254 if ( (instruction & 0x3B000000) == 0x39000000 ) {
7255 switch ( instruction & 0xC0000000 ) {
7256 case 0x00000000:
7257 break;
7258 case 0x40000000:
7259 encodedAddend *= 2;
7260 break;
7261 case 0x80000000:
7262 encodedAddend *= 4;
7263 break;
7264 case 0xC0000000:
7265 encodedAddend *= 8;
7266 break;
7267 }
7268 }
7269 if ( encodedAddend != 0 ) {
7270 if ( prefixRelocAddend == 0 ) {
7271 warning("pageoff12 instruction at 0x%08X has embedded addend. ARM64_RELOC_ADDEND should be used instead", reloc->r_address());
7272 target.addend = encodedAddend;
7273 }
7274 else {
7275 throwf("pageoff12 instruction at 0x%08X has embedded addend and ARM64_RELOC_ADDEND also used", reloc->r_address());
7276 }
7277 }
7278 parser.addFixups(src, ld::Fixup::kindStoreARM64PageOff12, target);
7279 break;
7280 case ARM64_RELOC_GOT_LOAD_PAGE21:
7281 if ( ! reloc->r_pcrel() )
7282 throw "not pcrel and ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7283 if ( ! reloc->r_extern() )
7284 throw "r_extern == 0 and ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7285 if ( reloc->r_length() != 2 )
7286 throw "length != 2 and ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7287 if ( prefixRelocAddend != 0 )
7288 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_GOT_LOAD_PAGE21 not supported";
7289 instruction = contentValue;
7290 target.addend = ((instruction & 0x60000000) >> 29) | ((instruction & 0x01FFFFE0) >> 3);
7291 if ( target.addend != 0 )
7292 throw "non-zero addend with ARM64_RELOC_GOT_LOAD_PAGE21 is not supported";
7293 parser.addFixups(src, ld::Fixup::kindStoreARM64GOTLoadPage21, target);
7294 break;
7295 case ARM64_RELOC_GOT_LOAD_PAGEOFF12:
7296 if ( reloc->r_pcrel() )
7297 throw "pcrel and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7298 if ( ! reloc->r_extern() )
7299 throw "r_extern == 0 and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7300 if ( reloc->r_length() != 2 )
7301 throw "length != 2 and ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7302 if ( prefixRelocAddend != 0 )
7303 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_GOT_LOAD_PAGEOFF12 not supported";
7304 instruction = contentValue;
7305 target.addend = ((instruction & 0x003FFC00) >> 10);
7306 parser.addFixups(src, ld::Fixup::kindStoreARM64GOTLoadPageOff12, target);
7307 break;
7308 case ARM64_RELOC_TLVP_LOAD_PAGE21:
7309 if ( ! reloc->r_pcrel() )
7310 throw "not pcrel and ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7311 if ( ! reloc->r_extern() )
7312 throw "r_extern == 0 and ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7313 if ( reloc->r_length() != 2 )
7314 throw "length != 2 and ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7315 if ( prefixRelocAddend != 0 )
7316 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_TLVP_LOAD_PAGE21 not supported";
7317 instruction = contentValue;
7318 target.addend = ((instruction & 0x60000000) >> 29) | ((instruction & 0x01FFFFE0) >> 3);
7319 if ( target.addend != 0 )
7320 throw "non-zero addend with ARM64_RELOC_GOT_LOAD_PAGE21 is not supported";
7321 parser.addFixups(src, ld::Fixup::kindStoreARM64TLVPLoadPage21, target);
7322 break;
7323 case ARM64_RELOC_TLVP_LOAD_PAGEOFF12:
7324 if ( reloc->r_pcrel() )
7325 throw "pcrel and ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7326 if ( ! reloc->r_extern() )
7327 throw "r_extern == 0 and ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7328 if ( reloc->r_length() != 2 )
7329 throw "length != 2 and ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7330 if ( prefixRelocAddend != 0 )
7331 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_TLVP_LOAD_PAGEOFF12 not supported";
7332 instruction = contentValue;
7333 target.addend = ((instruction & 0x003FFC00) >> 10);
7334 parser.addFixups(src, ld::Fixup::kindStoreARM64TLVPLoadPageOff12, target);
7335 break;
7336 case ARM64_RELOC_SUBTRACTOR:
7337 if ( reloc->r_pcrel() )
7338 throw "ARM64_RELOC_SUBTRACTOR cannot be pc-relative";
7339 if ( reloc->r_length() < 2 )
7340 throw "ARM64_RELOC_SUBTRACTOR must have r_length of 2 or 3";
7341 if ( !reloc->r_extern() )
7342 throw "ARM64_RELOC_SUBTRACTOR must have r_extern=1";
7343 if ( nextReloc->r_type() != ARM64_RELOC_UNSIGNED )
7344 throw "ARM64_RELOC_SUBTRACTOR must be followed by ARM64_RELOC_UNSIGNED";
7345 if ( prefixRelocAddend != 0 )
7346 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_SUBTRACTOR not supported";
7347 result = true;
7348 if ( nextReloc->r_pcrel() )
7349 throw "ARM64_RELOC_UNSIGNED following a ARM64_RELOC_SUBTRACTOR cannot be pc-relative";
7350 if ( nextReloc->r_length() != reloc->r_length() )
7351 throw "ARM64_RELOC_UNSIGNED following a ARM64_RELOC_SUBTRACTOR must have same r_length";
7352 if ( nextReloc->r_extern() ) {
7353 const macho_nlist<P>& sym = parser.symbolFromIndex(nextReloc->r_symbolnum());
7354 // use direct reference for local symbols
7355 if ( ((sym.n_type() & N_TYPE) == N_SECT) && (((sym.n_type() & N_EXT) == 0) || (parser.nameFromSymbol(sym)[0] == 'L')) ) {
7356 parser.findTargetFromAddressAndSectionNum(sym.n_value(), sym.n_sect(), toTarget);
7357 toTarget.addend = contentValue;
7358 useDirectBinding = true;
7359 }
7360 else {
7361 toTarget.name = parser.nameFromSymbol(sym);
7362 toTarget.weakImport = parser.weakImportFromSymbol(sym);
7363 toTarget.addend = contentValue;
7364 useDirectBinding = false;
7365 }
7366 }
7367 else {
7368 parser.findTargetFromAddressAndSectionNum(contentValue, nextReloc->r_symbolnum(), toTarget);
7369 useDirectBinding = (toTarget.atom->scope() == ld::Atom::scopeTranslationUnit);
7370 }
7371 if ( useDirectBinding )
7372 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.atom);
7373 else
7374 parser.addFixup(src, ld::Fixup::k1of4, ld::Fixup::kindSetTargetAddress, toTarget.weakImport, toTarget.name);
7375 parser.addFixup(src, ld::Fixup::k2of4, ld::Fixup::kindAddAddend, toTarget.addend);
7376 if ( target.atom == NULL )
7377 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, false, target.name);
7378 else
7379 parser.addFixup(src, ld::Fixup::k3of4, ld::Fixup::kindSubtractTargetAddress, target.atom);
7380 if ( reloc->r_length() == 2 )
7381 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian32);
7382 else
7383 parser.addFixup(src, ld::Fixup::k4of4, ld::Fixup::kindStoreLittleEndian64);
7384 break;
7385 case ARM64_RELOC_POINTER_TO_GOT:
7386 if ( ! reloc->r_extern() )
7387 throw "r_extern == 0 and ARM64_RELOC_POINTER_TO_GOT not supported";
7388 if ( prefixRelocAddend != 0 )
7389 throw "ARM64_RELOC_ADDEND followed by ARM64_RELOC_POINTER_TO_GOT not supported";
7390 if ( reloc->r_pcrel() ) {
7391 if ( reloc->r_length() != 2 )
7392 throw "r_length != 2 and r_extern = 1 and ARM64_RELOC_POINTER_TO_GOT not supported";
7393 parser.addFixups(src, ld::Fixup::kindStoreARM64PCRelToGOT, target);
7394 }
7395 else {
7396 if ( reloc->r_length() != 3 )
7397 throw "r_length != 3 and r_extern = 0 and ARM64_RELOC_POINTER_TO_GOT not supported";
7398 parser.addFixups(src, ld::Fixup::kindStoreARM64PointerToGOT, target);
7399 }
7400 break;
7401 default:
7402 throwf("unknown relocation type %d", reloc->r_type());
7403 }
7404 return result;
7405 }
7406 #endif
7407
7408 template <typename A>
7409 bool ObjC1ClassSection<A>::addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>* reloc)
7410 {
7411 // inherited
7412 FixedSizeSection<A>::addRelocFixup(parser, reloc);
7413
7414 assert(0 && "needs template specialization");
7415 return false;
7416 }
7417
7418 template <>
7419 bool ObjC1ClassSection<x86>::addRelocFixup(class Parser<x86>& parser, const macho_relocation_info<x86::P>* reloc)
7420 {
7421 // if this is the reloc for the super class name string, add implicit reference to super class
7422 if ( ((reloc->r_address() & R_SCATTERED) == 0) && (reloc->r_type() == GENERIC_RELOC_VANILLA) ) {
7423 assert( reloc->r_length() == 2 );
7424 assert( ! reloc->r_pcrel() );
7425
7426 const macho_section<P>* sect = this->machoSection();
7427 Parser<x86>::SourceLocation src;
7428 uint32_t srcAddr = sect->addr() + reloc->r_address();
7429 src.atom = this->findAtomByAddress(srcAddr);
7430 src.offsetInAtom = srcAddr - src.atom->objectAddress();
7431 if ( src.offsetInAtom == 4 ) {
7432 Parser<x86>::TargetDesc stringTarget;
7433 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
7434 uint32_t contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
7435 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), stringTarget);
7436
7437 assert(stringTarget.atom != NULL);
7438 assert(stringTarget.atom->contentType() == ld::Atom::typeCString);
7439 const char* superClassBaseName = (char*)stringTarget.atom->rawContentPointer();
7440 char* superClassName = new char[strlen(superClassBaseName) + 20];
7441 strcpy(superClassName, ".objc_class_name_");
7442 strcat(superClassName, superClassBaseName);
7443
7444 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindSetTargetAddress, false, superClassName);
7445 }
7446 }
7447 // inherited
7448 return FixedSizeSection<x86>::addRelocFixup(parser, reloc);
7449 }
7450
7451
7452
7453 template <typename A>
7454 bool Objc1ClassReferences<A>::addRelocFixup(class Parser<A>& parser, const macho_relocation_info<P>* reloc)
7455 {
7456 // inherited
7457 PointerToCStringSection<A>::addRelocFixup(parser, reloc);
7458
7459 assert(0 && "needs template specialization");
7460 return false;
7461 }
7462
7463
7464
7465 template <>
7466 bool Objc1ClassReferences<x86>::addRelocFixup(class Parser<x86>& parser, const macho_relocation_info<x86::P>* reloc)
7467 {
7468 // add implict class refs, fixups not usable yet, so look at relocations
7469 assert( (reloc->r_address() & R_SCATTERED) == 0 );
7470 assert( reloc->r_type() == GENERIC_RELOC_VANILLA );
7471 assert( reloc->r_length() == 2 );
7472 assert( ! reloc->r_pcrel() );
7473
7474 const macho_section<P>* sect = this->machoSection();
7475 Parser<x86>::SourceLocation src;
7476 uint32_t srcAddr = sect->addr() + reloc->r_address();
7477 src.atom = this->findAtomByAddress(srcAddr);
7478 src.offsetInAtom = srcAddr - src.atom->objectAddress();
7479 Parser<x86>::TargetDesc stringTarget;
7480 const uint8_t* fixUpPtr = file().fileContent() + sect->offset() + reloc->r_address();
7481 uint32_t contentValue = LittleEndian::get32(*((uint32_t*)fixUpPtr));
7482 parser.findTargetFromAddressAndSectionNum(contentValue, reloc->r_symbolnum(), stringTarget);
7483
7484 assert(stringTarget.atom != NULL);
7485 assert(stringTarget.atom->contentType() == ld::Atom::typeCString);
7486 const char* baseClassName = (char*)stringTarget.atom->rawContentPointer();
7487 char* objcClassName = new char[strlen(baseClassName) + 20];
7488 strcpy(objcClassName, ".objc_class_name_");
7489 strcat(objcClassName, baseClassName);
7490
7491 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindSetTargetAddress, false, objcClassName);
7492
7493 // inherited
7494 return PointerToCStringSection<x86>::addRelocFixup(parser, reloc);
7495 }
7496
7497 #if SUPPORT_ARCH_arm64
7498 template <>
7499 void Section<arm64>::addLOH(class Parser<arm64>& parser, int kind, int count, const uint64_t addrs[]) {
7500 switch (kind) {
7501 case LOH_ARM64_ADRP_ADRP:
7502 case LOH_ARM64_ADRP_LDR:
7503 case LOH_ARM64_ADRP_ADD:
7504 case LOH_ARM64_ADRP_LDR_GOT:
7505 if ( count != 2 )
7506 warning("arm64 Linker Optimiztion Hint %d has wrong number of arguments", kind);
7507 break;
7508 case LOH_ARM64_ADRP_ADD_LDR:
7509 case LOH_ARM64_ADRP_LDR_GOT_LDR:
7510 case LOH_ARM64_ADRP_ADD_STR:
7511 case LOH_ARM64_ADRP_LDR_GOT_STR:
7512 if ( count != 3 )
7513 warning("arm64 Linker Optimiztion Hint %d has wrong number of arguments", kind);
7514 }
7515
7516 // pick lowest address in tuple for use as offsetInAtom
7517 uint64_t lowestAddress = addrs[0];
7518 for(int i=1; i < count; ++i) {
7519 if ( addrs[i] < lowestAddress )
7520 lowestAddress = addrs[i];
7521 }
7522 // verify all other address are in same atom
7523 Atom<arm64>* inAtom = parser.findAtomByAddress(lowestAddress);
7524 const uint64_t atomStartAddr = inAtom->objectAddress();
7525 const uint64_t atomEndAddr = atomStartAddr + inAtom->size();
7526 for(int i=0; i < count; ++i) {
7527 if ( (addrs[i] < atomStartAddr) || (addrs[i] >= atomEndAddr) ) {
7528 warning("arm64 Linker Optimiztion Hint addresses are not in same atom: 0x%08llX and 0x%08llX",
7529 lowestAddress, addrs[i]);
7530 return; // skip this LOH
7531 }
7532 if ( (addrs[i] & 0x3) != 0 ) {
7533 warning("arm64 Linker Optimiztion Hint address is not 4-byte aligned: 0x%08llX", addrs[i]);
7534 return; // skip this LOH
7535 }
7536 if ( (addrs[i] - lowestAddress) > 0xFFFF ) {
7537 if ( parser.verboseOptimizationHints() ) {
7538 warning("arm64 Linker Optimiztion Hint addresses are too far apart: 0x%08llX and 0x%08llX",
7539 lowestAddress, addrs[i]);
7540 }
7541 return; // skip this LOH
7542 }
7543 }
7544
7545 // encoded kind, count, and address deltas in 64-bit addend
7546 ld::Fixup::LOH_arm64 extra;
7547 extra.addend = 0;
7548 extra.info.kind = kind;
7549 extra.info.count = count-1;
7550 extra.info.delta1 = (addrs[0] - lowestAddress) >> 2;
7551 extra.info.delta2 = (count > 1) ? ((addrs[1] - lowestAddress) >> 2) : 0;
7552 extra.info.delta3 = (count > 2) ? ((addrs[2] - lowestAddress) >> 2) : 0;
7553 extra.info.delta4 = (count > 3) ? ((addrs[3] - lowestAddress) >> 2) : 0;
7554 typename Parser<arm64>::SourceLocation src(inAtom, lowestAddress- inAtom->objectAddress());
7555 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindLinkerOptimizationHint, extra.addend);
7556 }
7557 #endif
7558
7559 template <typename A>
7560 void Section<A>::addLOH(class Parser<A>& parser, int kind, int count, const uint64_t addrs[]) {
7561
7562 }
7563
7564 template <typename A>
7565 void Section<A>::makeFixups(class Parser<A>& parser, const struct Parser<A>::CFI_CU_InfoArrays&)
7566 {
7567 const macho_section<P>* sect = this->machoSection();
7568 if ( sect->reloff() + (sect->nreloc() * sizeof(macho_relocation_info<P>)) > parser.fileLength() )
7569 throwf("relocations for section %s/%s extends beyond end of file,", sect->segname(), Section<A>::makeSectionName(sect) );
7570 const macho_relocation_info<P>* relocs = (macho_relocation_info<P>*)(file().fileContent() + sect->reloff());
7571 const uint32_t relocCount = sect->nreloc();
7572 for (uint32_t r = 0; r < relocCount; ++r) {
7573 try {
7574 if ( this->addRelocFixup(parser, &relocs[r]) )
7575 ++r; // skip next
7576 }
7577 catch (const char* msg) {
7578 throwf("in section %s,%s reloc %u: %s", sect->segname(), Section<A>::makeSectionName(sect), r, msg);
7579 }
7580 }
7581
7582 // add follow-on fixups if .o file is missing .subsections_via_symbols
7583 if ( this->addFollowOnFixups() ) {
7584 Atom<A>* end = &_endAtoms[-1];
7585 for(Atom<A>* p = _beginAtoms; p < end; ++p) {
7586 typename Parser<A>::SourceLocation src(p, 0);
7587 Atom<A>* nextAtom = &p[1];
7588 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, nextAtom);
7589 }
7590 }
7591 else if ( this->type() == ld::Section::typeCode ) {
7592 // if FDE broke text not at a symbol, use followOn to keep code together
7593 Atom<A>* end = &_endAtoms[-1];
7594 for(Atom<A>* p = _beginAtoms; p < end; ++p) {
7595 typename Parser<A>::SourceLocation src(p, 0);
7596 Atom<A>* nextAtom = &p[1];
7597 if ( (p->symbolTableInclusion() == ld::Atom::symbolTableIn) && (nextAtom->symbolTableInclusion() == ld::Atom::symbolTableNotIn) ) {
7598 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, nextAtom);
7599 }
7600 }
7601 }
7602 if ( !this->_altEntries.empty() && !this->addFollowOnFixups() ) {
7603 if ( _altEntries.count(_beginAtoms) != 0 )
7604 warning("N_ALT_ENTRY bit set on first atom in section %s/%s", sect->segname(), Section<A>::makeSectionName(sect));
7605
7606 Atom<A>* end = &_endAtoms[-1];
7607 for(Atom<A>* p = _beginAtoms; p < end; ++p) {
7608 Atom<A>* nextAtom = &p[1];
7609 // <rdar://problem/22960070> support alt_entry aliases (alias process already added followOn, don't repeat)
7610 if ( (_altEntries.count(nextAtom) != 0) && (p->_objAddress != nextAtom->_objAddress) ) {
7611 typename Parser<A>::SourceLocation src(p, 0);
7612 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, nextAtom);
7613 typename Parser<A>::SourceLocation src2(nextAtom, 0);
7614 parser.addFixup(src2, ld::Fixup::k1of1, ld::Fixup::kindNoneGroupSubordinate, p);
7615 }
7616 }
7617 }
7618
7619 // <rdar://problem/9218847> track data-in-code
7620 if ( parser.hasDataInCodeLabels() && (this->type() == ld::Section::typeCode) ) {
7621 for (uint32_t i=0; i < parser.symbolCount(); ++i) {
7622 const macho_nlist<P>& sym = parser.symbolFromIndex(i);
7623 // ignore stabs
7624 if ( (sym.n_type() & N_STAB) != 0 )
7625 continue;
7626 // ignore non-definitions
7627 if ( (sym.n_type() & N_TYPE) != N_SECT )
7628 continue;
7629
7630 // 'L' labels do not denote atom breaks
7631 const char* symbolName = parser.nameFromSymbol(sym);
7632 if ( symbolName[0] == 'L' ) {
7633 if ( strncmp(symbolName, "L$start$", 8) == 0 ) {
7634 ld::Fixup::Kind kind = ld::Fixup::kindNone;
7635 if ( strncmp(&symbolName[8], "data$", 5) == 0 )
7636 kind = ld::Fixup::kindDataInCodeStartData;
7637 else if ( strncmp(&symbolName[8], "code$", 5) == 0 )
7638 kind = ld::Fixup::kindDataInCodeEnd;
7639 else if ( strncmp(&symbolName[8], "jt8$", 4) == 0 )
7640 kind = ld::Fixup::kindDataInCodeStartJT8;
7641 else if ( strncmp(&symbolName[8], "jt16$", 4) == 0 )
7642 kind = ld::Fixup::kindDataInCodeStartJT16;
7643 else if ( strncmp(&symbolName[8], "jt32$", 4) == 0 )
7644 kind = ld::Fixup::kindDataInCodeStartJT32;
7645 else if ( strncmp(&symbolName[8], "jta32$", 4) == 0 )
7646 kind = ld::Fixup::kindDataInCodeStartJTA32;
7647 else
7648 warning("unknown L$start$ label %s in file %s", symbolName, this->file().path());
7649 if ( kind != ld::Fixup::kindNone ) {
7650 Atom<A>* inAtom = parser.findAtomByAddress(sym.n_value());
7651 typename Parser<A>::SourceLocation src(inAtom, sym.n_value() - inAtom->objectAddress());
7652 parser.addFixup(src, ld::Fixup::k1of1, kind);
7653 }
7654 }
7655 }
7656 }
7657 }
7658
7659 // <rdar://problem/11150575> Handle LC_DATA_IN_CODE in object files
7660 if ( this->type() == ld::Section::typeCode ) {
7661 const pint_t startAddr = this->_machOSection->addr();
7662 const pint_t endAddr = startAddr + this->_machOSection->size();
7663 for ( const macho_data_in_code_entry<P>* p = parser.dataInCodeStart(); p != parser.dataInCodeEnd(); ++p ) {
7664 if ( (p->offset() >= startAddr) && (p->offset() < endAddr) ) {
7665 ld::Fixup::Kind kind = ld::Fixup::kindNone;
7666 switch ( p->kind() ) {
7667 case DICE_KIND_DATA:
7668 kind = ld::Fixup::kindDataInCodeStartData;
7669 break;
7670 case DICE_KIND_JUMP_TABLE8:
7671 kind = ld::Fixup::kindDataInCodeStartJT8;
7672 break;
7673 case DICE_KIND_JUMP_TABLE16:
7674 kind = ld::Fixup::kindDataInCodeStartJT16;
7675 break;
7676 case DICE_KIND_JUMP_TABLE32:
7677 kind = ld::Fixup::kindDataInCodeStartJT32;
7678 break;
7679 case DICE_KIND_ABS_JUMP_TABLE32:
7680 kind = ld::Fixup::kindDataInCodeStartJTA32;
7681 break;
7682 default:
7683 kind = ld::Fixup::kindDataInCodeStartData;
7684 warning("uknown LC_DATA_IN_CODE kind (%d) at offset 0x%08X", p->kind(), p->offset());
7685 break;
7686 }
7687 Atom<A>* inAtom = parser.findAtomByAddress(p->offset());
7688 typename Parser<A>::SourceLocation srcStart(inAtom, p->offset() - inAtom->objectAddress());
7689 parser.addFixup(srcStart, ld::Fixup::k1of1, kind);
7690 typename Parser<A>::SourceLocation srcEnd(inAtom, p->offset() + p->length() - inAtom->objectAddress());
7691 parser.addFixup(srcEnd, ld::Fixup::k1of1, ld::Fixup::kindDataInCodeEnd);
7692 }
7693 }
7694 }
7695
7696 // <rdar://problem/11945700> convert linker optimization hints into internal format
7697 if ( this->type() == ld::Section::typeCode && parser.hasOptimizationHints() ) {
7698 const pint_t startAddr = this->_machOSection->addr();
7699 const pint_t endAddr = startAddr + this->_machOSection->size();
7700 for (const uint8_t* p = parser.optimizationHintsStart(); p < parser.optimizationHintsEnd(); ) {
7701 uint64_t addrs[4];
7702 int32_t kind = read_uleb128(&p, parser.optimizationHintsEnd());
7703 if ( kind == 0 ) // padding at end of loh buffer
7704 break;
7705 if ( kind == -1 ) {
7706 warning("malformed uleb128 kind in LC_LINKER_OPTIMIZATION_HINTS");
7707 break;
7708 }
7709 int32_t count = read_uleb128(&p, parser.optimizationHintsEnd());
7710 if ( count == -1 ) {
7711 warning("malformed uleb128 count in LC_LINKER_OPTIMIZATION_HINTS");
7712 break;
7713 }
7714 if ( count > 3 ) {
7715 warning("address count > 3 in LC_LINKER_OPTIMIZATION_HINTS");
7716 break;
7717 }
7718 for (int32_t i=0; i < count; ++i) {
7719 addrs[i] = read_uleb128(&p, parser.optimizationHintsEnd());
7720 }
7721 if ( (startAddr <= addrs[0]) && (addrs[0] < endAddr) ) {
7722 this->addLOH(parser, kind, count, addrs);
7723 //fprintf(stderr, "kind=%d", kind);
7724 //for (int32_t i=0; i < count; ++i) {
7725 // fprintf(stderr, ", addr=0x%08llX", addrs[i]);
7726 //}
7727 //fprintf(stderr, "\n");
7728 }
7729 }
7730 }
7731
7732
7733 // add follow-on fixups for aliases
7734 if ( _hasAliases ) {
7735 for(Atom<A>* p = _beginAtoms; p < _endAtoms; ++p) {
7736 if ( p->isAlias() && ! this->addFollowOnFixups() ) {
7737 Atom<A>* targetOfAlias = &p[1];
7738 assert(p < &_endAtoms[-1]);
7739 assert(p->_objAddress == targetOfAlias->_objAddress);
7740 typename Parser<A>::SourceLocation src(p, 0);
7741 parser.addFixup(src, ld::Fixup::k1of1, ld::Fixup::kindNoneFollowOn, targetOfAlias);
7742 }
7743 }
7744 }
7745 }
7746
7747
7748
7749 //
7750 // main function used by linker to instantiate ld::Files
7751 //
7752 ld::relocatable::File* parse(const uint8_t* fileContent, uint64_t fileLength,
7753 const char* path, time_t modTime, ld::File::Ordinal ordinal, const ParserOptions& opts)
7754 {
7755 switch ( opts.architecture ) {
7756 #if SUPPORT_ARCH_x86_64
7757 case CPU_TYPE_X86_64:
7758 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) )
7759 return mach_o::relocatable::Parser<x86_64>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7760 break;
7761 #endif
7762 #if SUPPORT_ARCH_i386
7763 case CPU_TYPE_I386:
7764 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent) )
7765 return mach_o::relocatable::Parser<x86>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7766 break;
7767 #endif
7768 #if SUPPORT_ARCH_arm_any
7769 case CPU_TYPE_ARM:
7770 if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) )
7771 return mach_o::relocatable::Parser<arm>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7772 break;
7773 #endif
7774 #if SUPPORT_ARCH_arm64
7775 case CPU_TYPE_ARM64:
7776 if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) )
7777 return mach_o::relocatable::Parser<arm64>::parse(fileContent, fileLength, path, modTime, ordinal, opts);
7778 break;
7779 #endif
7780 }
7781 return NULL;
7782 }
7783
7784 //
7785 // used by archive reader to validate member object file
7786 //
7787 bool isObjectFile(const uint8_t* fileContent, uint64_t fileLength, const ParserOptions& opts)
7788 {
7789 switch ( opts.architecture ) {
7790 case CPU_TYPE_X86_64:
7791 return ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) );
7792 case CPU_TYPE_I386:
7793 return ( mach_o::relocatable::Parser<x86>::validFile(fileContent) );
7794 case CPU_TYPE_ARM:
7795 return ( mach_o::relocatable::Parser<arm>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) );
7796 case CPU_TYPE_ARM64:
7797 return ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, opts.objSubtypeMustMatch, opts.subType) );
7798 }
7799 return false;
7800 }
7801
7802 //
7803 // used by linker to infer architecture when no -arch is on command line
7804 //
7805 bool isObjectFile(const uint8_t* fileContent, cpu_type_t* result, cpu_subtype_t* subResult, Options::Platform* platform)
7806 {
7807 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7808 *result = CPU_TYPE_X86_64;
7809 const macho_header<Pointer64<LittleEndian> >* header = (const macho_header<Pointer64<LittleEndian> >*)fileContent;
7810 *subResult = header->cpusubtype();
7811 *platform = Parser<x86_64>::findPlatform(header);
7812 return true;
7813 }
7814 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent) ) {
7815 const macho_header<Pointer32<LittleEndian> >* header = (const macho_header<Pointer32<LittleEndian> >*)fileContent;
7816 *result = CPU_TYPE_I386;
7817 *subResult = CPU_SUBTYPE_X86_ALL;
7818 *platform = Parser<x86>::findPlatform(header);
7819 return true;
7820 }
7821 if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7822 const macho_header<Pointer32<LittleEndian> >* header = (const macho_header<Pointer32<LittleEndian> >*)fileContent;
7823 *result = CPU_TYPE_ARM;
7824 *subResult = header->cpusubtype();
7825 *platform = Parser<arm>::findPlatform(header);
7826 return true;
7827 }
7828 if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, false, 0) ) {
7829 const macho_header<Pointer64<LittleEndian> >* header = (const macho_header<Pointer64<LittleEndian> >*)fileContent;
7830 *result = CPU_TYPE_ARM64;
7831 *subResult = CPU_SUBTYPE_ARM64_ALL;
7832 *platform = Parser<arm64>::findPlatform(header);
7833 return true;
7834 }
7835 return false;
7836 }
7837
7838 //
7839 // used by linker is error messages to describe bad .o file
7840 //
7841 const char* archName(const uint8_t* fileContent)
7842 {
7843 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7844 return mach_o::relocatable::Parser<x86_64>::fileKind(fileContent);
7845 }
7846 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent) ) {
7847 return mach_o::relocatable::Parser<x86>::fileKind(fileContent);
7848 }
7849 if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7850 return mach_o::relocatable::Parser<arm>::fileKind(fileContent);
7851 }
7852 return NULL;
7853 }
7854
7855 //
7856 // Used by archive reader when -ObjC option is specified
7857 //
7858 bool hasObjC2Categories(const uint8_t* fileContent)
7859 {
7860 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7861 return mach_o::relocatable::Parser<x86_64>::hasObjC2Categories(fileContent);
7862 }
7863 else if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7864 return mach_o::relocatable::Parser<arm>::hasObjC2Categories(fileContent);
7865 }
7866 else if ( mach_o::relocatable::Parser<x86>::validFile(fileContent, false, 0) ) {
7867 return mach_o::relocatable::Parser<x86>::hasObjC2Categories(fileContent);
7868 }
7869 #if SUPPORT_ARCH_arm64
7870 else if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, false, 0) ) {
7871 return mach_o::relocatable::Parser<arm64>::hasObjC2Categories(fileContent);
7872 }
7873 #endif
7874 return false;
7875 }
7876
7877 //
7878 // Used by archive reader when -ObjC option is specified
7879 //
7880 bool hasObjC1Categories(const uint8_t* fileContent)
7881 {
7882 if ( mach_o::relocatable::Parser<x86>::validFile(fileContent, false, 0) ) {
7883 return mach_o::relocatable::Parser<x86>::hasObjC1Categories(fileContent);
7884 }
7885 return false;
7886 }
7887
7888 //
7889 // Used by bitcode obfuscator to get a list of non local symbols from object file
7890 //
7891 bool getNonLocalSymbols(const uint8_t* fileContent, std::vector<const char*> &syms)
7892 {
7893 if ( mach_o::relocatable::Parser<x86_64>::validFile(fileContent) ) {
7894 return mach_o::relocatable::Parser<x86_64>::getNonLocalSymbols(fileContent, syms);
7895 }
7896 else if ( mach_o::relocatable::Parser<arm>::validFile(fileContent, false, 0) ) {
7897 return mach_o::relocatable::Parser<arm>::getNonLocalSymbols(fileContent, syms);
7898 }
7899 else if ( mach_o::relocatable::Parser<x86>::validFile(fileContent, false, 0) ) {
7900 return mach_o::relocatable::Parser<x86>::getNonLocalSymbols(fileContent, syms);
7901 }
7902 else if ( mach_o::relocatable::Parser<arm64>::validFile(fileContent, false, 0) ) {
7903 return mach_o::relocatable::Parser<arm64>::getNonLocalSymbols(fileContent, syms);
7904 }
7905 return false;
7906 }
7907
7908
7909
7910 } // namespace relocatable
7911 } // namespace mach_o
7912
7913
mirror of ld64