X-Git-Url: https://git.saurik.com/apple/ld64.git/blobdiff_plain/4be885f63e03d78a3780e8041f107657f85eb5cb..bee7e226299dacc6d80d9cf6f1585c3d5d0645e0:/src/ld/ld.cpp diff --git a/src/ld/ld.cpp b/src/ld/ld.cpp index 6d93c0b..a8d2276 100644 --- a/src/ld/ld.cpp +++ b/src/ld/ld.cpp @@ -1,5 +1,6 @@ /* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-* - * Copyright (c) 2005-2007 Apple Inc. All rights reserved. + * + * Copyright (c) 2005-2011 Apple Inc. All rights reserved. * * @APPLE_LICENSE_HEADER_START@ * @@ -23,7 +24,7 @@ // start temp HACK for cross builds extern "C" double log2 ( double ); -#define __MATH__ +//#define __MATH__ // end temp HACK for cross builds @@ -36,11 +37,15 @@ extern "C" double log2 ( double ); #include #include #include +#include #include #include #include #include #include +#include +#include +#include #include #include @@ -49,4046 +54,1341 @@ extern "C" double log2 ( double ); #include #include #include -#include -#include -#include +#include +#include -#include "configure.h" #include "Options.h" -#include "ObjectFile.h" - -#include "MachOReaderRelocatable.hpp" -#include "ArchiveReader.hpp" -#include "MachOReaderDylib.hpp" -#include "MachOWriterExecutable.hpp" - - -#if LTO_SUPPORT -#include "LTOReader.hpp" -#endif - -#include "OpaqueSection.hpp" - - -class CStringComparor -{ -public: - bool operator()(const char* left, const char* right) const { return (strcmp(left, right) < 0); } +#include "MachOFileAbstraction.hpp" +#include "Architectures.hpp" +#include "ld.hpp" + +#include "InputFiles.h" +#include "Resolver.h" +#include "OutputFile.h" +#include "Snapshot.h" + +#include "passes/stubs/make_stubs.h" +#include "passes/dtrace_dof.h" +#include "passes/got.h" +#include "passes/tlvp.h" +#include "passes/huge.h" +#include "passes/compact_unwind.h" +#include "passes/order.h" +#include "passes/branch_island.h" +#include "passes/branch_shim.h" +#include "passes/objc.h" +#include "passes/dylibs.h" +#include "passes/bitcode_bundle.h" +#include "passes/code_dedup.h" + +#include "parsers/archive_file.h" +#include "parsers/macho_relocatable_file.h" +#include "parsers/macho_dylib_file.h" +#include "parsers/lto_file.h" +#include "parsers/opaque_section_file.h" + + +struct PerformanceStatistics { + uint64_t startTool; + uint64_t startInputFileProcessing; + uint64_t startResolver; + uint64_t startDylibs; + uint64_t startPasses; + uint64_t startOutput; + uint64_t startDone; + vm_statistics_data_t vmStart; + vm_statistics_data_t vmEnd; }; -class CStringEquals -{ -public: - bool operator()(const char* left, const char* right) const { return (strcmp(left, right) == 0); } -}; -class Section : public ObjectFile::Section +class InternalState : public ld::Internal { public: - static Section* find(const char* sectionName, const char* segmentName, bool zeroFill, bool untrustedZeroFill, bool createIfNeeded=true); - static void assignIndexes(); - const char* getName() { return fSectionName; } + InternalState(const Options& opts) : _options(opts), _atomsOrderedInSections(false) { } + virtual ld::Internal::FinalSection* addAtom(const ld::Atom& atom); + virtual ld::Internal::FinalSection* getFinalSection(const ld::Section&); + ld::Internal::FinalSection* getFinalSection(const char* seg, const char* sect, ld::Section::Type type); + + uint64_t assignFileOffsets(); + void setSectionSizesAndAlignments(); + void sortSections(); + void markAtomsOrdered() { _atomsOrderedInSections = true; } + bool hasReferenceToWeakExternal(const ld::Atom& atom); + + virtual ~InternalState() {} private: - Section(const char* sectionName, const char* segmentName, bool zeroFill, bool untrustedZeroFill); + bool inMoveRWChain(const ld::Atom& atom, const char* filePath, const char*& dstSeg, bool& wildCardMatch); + bool inMoveROChain(const ld::Atom& atom, const char* filePath, const char*& dstSeg, bool& wildCardMatch); - struct Sorter { - static int segmentOrdinal(const char* segName); - bool operator()(Section* left, Section* right); + class FinalSection : public ld::Internal::FinalSection + { + public: + FinalSection(const ld::Section& sect, uint32_t sectionsSeen, const Options&); + static int sectionComparer(const void* l, const void* r); + static const ld::Section& outputSection(const ld::Section& sect, bool mergeZeroFill); + static const ld::Section& objectOutputSection(const ld::Section& sect, const Options&); + private: + friend class InternalState; + static uint32_t sectionOrder(const ld::Section& sect, uint32_t sectionsSeen, const Options& options); + static uint32_t segmentOrder(const ld::Section& sect, const Options& options); + uint32_t _segmentOrder; + uint32_t _sectionOrder; + + static std::vector _s_segmentsSeen; + static ld::Section _s_DATA_data; + static ld::Section _s_DATA_const; + static ld::Section _s_TEXT_text; + static ld::Section _s_TEXT_const; + static ld::Section _s_DATA_nl_symbol_ptr; + static ld::Section _s_DATA_common; + static ld::Section _s_DATA_zerofill; + static ld::Section _s_DATA_DIRTY_data; + static ld::Section _s_DATA_CONST_const; + }; + + bool hasZeroForFileOffset(const ld::Section* sect); + uint64_t pageAlign(uint64_t addr); + uint64_t pageAlign(uint64_t addr, uint64_t pageSize); + + struct SectionHash { + size_t operator()(const ld::Section*) const; + }; + struct SectionEquals { + bool operator()(const ld::Section* left, const ld::Section* right) const; }; + typedef std::unordered_map SectionInToOut; + - typedef __gnu_cxx::hash_map, CStringEquals> NameToOrdinal; - typedef __gnu_cxx::hash_map, CStringEquals> NameToSection; - //typedef std::map NameToSection; + SectionInToOut _sectionInToFinalMap; + const Options& _options; + bool _atomsOrderedInSections; + std::unordered_map _pendingSegMove; +}; - const char* fSectionName; - const char* fSegmentName; - bool fZeroFill; - bool fUntrustedZeroFill; +ld::Section InternalState::FinalSection::_s_DATA_data( "__DATA", "__data", ld::Section::typeUnclassified); +ld::Section InternalState::FinalSection::_s_DATA_const("__DATA", "__const", ld::Section::typeUnclassified); +ld::Section InternalState::FinalSection::_s_TEXT_text( "__TEXT", "__text", ld::Section::typeCode); +ld::Section InternalState::FinalSection::_s_TEXT_const("__TEXT", "__const", ld::Section::typeUnclassified); +ld::Section InternalState::FinalSection::_s_DATA_nl_symbol_ptr("__DATA", "__nl_symbol_ptr", ld::Section::typeNonLazyPointer); +ld::Section InternalState::FinalSection::_s_DATA_common("__DATA", "__common", ld::Section::typeZeroFill); +ld::Section InternalState::FinalSection::_s_DATA_zerofill("__DATA", "__zerofill", ld::Section::typeZeroFill); +ld::Section InternalState::FinalSection::_s_DATA_DIRTY_data( "__DATA_DIRTY", "__data", ld::Section::typeUnclassified); +ld::Section InternalState::FinalSection::_s_DATA_CONST_const( "__DATA_CONST", "__const", ld::Section::typeUnclassified); - static NameToSection fgMapping; - static std::vector fgSections; - static NameToOrdinal fgSegmentDiscoverOrder; -}; +std::vector InternalState::FinalSection::_s_segmentsSeen; -Section::NameToSection Section::fgMapping; -std::vector Section::fgSections; -Section::NameToOrdinal Section::fgSegmentDiscoverOrder; -Section::Section(const char* sectionName, const char* segmentName, bool zeroFill, bool untrustedZeroFill) - : fSectionName(sectionName), fSegmentName(segmentName), fZeroFill(zeroFill), fUntrustedZeroFill(untrustedZeroFill) +size_t InternalState::SectionHash::operator()(const ld::Section* sect) const { - this->fIndex = fgSections.size() + 20; // room for 20 standard sections - // special placement of some sections - if ( strcmp(segmentName, "__TEXT") == 0 ) { - // sort unwind info to end of segment - if ( strcmp(sectionName, "__eh_frame") == 0 ) - this->fIndex = INT_MAX; - else if ( strcmp(sectionName, "__unwind_info") == 0 ) - this->fIndex = INT_MAX-1; - else if ( strcmp(sectionName, "__gcc_except_tab") == 0 ) - this->fIndex = INT_MAX-2; - } - else if ( strcmp(segmentName, "__DATA") == 0 ) { - // sort sections dyld will touch to start of segment - if ( strcmp(sectionName, "__dyld") == 0 ) - this->fIndex = 1; - else if ( strcmp(sectionName, "__program_vars") == 0 ) - this->fIndex = 1; - else if ( strcmp(sectionName, "__mod_init_func") == 0 ) - this->fIndex = 2; - else if ( strcmp(sectionName, "__nl_symbol_ptr") == 0 ) - this->fIndex = 3; - else if ( strcmp(sectionName, "__la_symbol_ptr") == 0 ) - this->fIndex = 4; - else if ( strcmp(sectionName, "__const") == 0 ) - this->fIndex = 5; - else if ( strcmp(sectionName, "__cfstring") == 0 ) - this->fIndex = 6; - else if ( strcmp(sectionName, "__gcc_except_tab") == 0 ) - this->fIndex = 7; - else if ( strcmp(sectionName, "__objc_data") == 0 ) - this->fIndex = 8; - else if ( strcmp(sectionName, "__objc_msgrefs") == 0 ) - this->fIndex = 9; - else if ( strcmp(sectionName, "__objc_protorefs") == 0 ) - this->fIndex = 10; - else if ( strcmp(sectionName, "__objc_selrefs") == 0 ) - this->fIndex = 11; - else if ( strcmp(sectionName, "__objc_classrefs") == 0 ) - this->fIndex = 12; - else if ( strcmp(sectionName, "__objc_superrefs") == 0 ) - this->fIndex = 13; - else if ( strcmp(sectionName, "__objc_const") == 0 ) - this->fIndex = 14; - else if ( strcmp(sectionName, "__objc_classlist") == 0 ) - this->fIndex = 15; - else if ( strcmp(sectionName, "__objc_nlclslist") == 0 ) - this->fIndex = 16; - else if ( strcmp(sectionName, "__objc_catlist") == 0 ) - this->fIndex = 17; - else if ( strcmp(sectionName, "__objc_protolist") == 0 ) - this->fIndex = 18; - else if ( strcmp(sectionName, "__objc_imageinfo") == 0 ) - this->fIndex = 19; - - } - - //fprintf(stderr, "new Section(%s, %s) => %p, %u\n", sectionName, segmentName, this, this->getIndex()); + size_t hash = 0; + ld::CStringHash temp; + hash += temp.operator()(sect->segmentName()); + hash += temp.operator()(sect->sectionName()); + return hash; } -Section* Section::find(const char* sectionName, const char* segmentName, bool zeroFill, bool untrustedZeroFill, bool createIfNeeded) +bool InternalState::SectionEquals::operator()(const ld::Section* left, const ld::Section* right) const { - NameToSection::iterator pos = fgMapping.find(sectionName); - if ( pos != fgMapping.end() ) { - if ( strcmp(pos->second->fSegmentName, segmentName) == 0 ) { - if ( !untrustedZeroFill && pos->second->fUntrustedZeroFill ) { - pos->second->fZeroFill = zeroFill; - pos->second->fUntrustedZeroFill = false; - } - return pos->second; - } - // otherwise same section name is used in different segments, look slow way - for (std::vector::iterator it=fgSections.begin(); it != fgSections.end(); it++) { - if ( (strcmp((*it)->fSectionName, sectionName) == 0) && (strcmp((*it)->fSegmentName, segmentName) == 0) ) - return *it; - } - } + return (*left == *right); +} - if ( !createIfNeeded ) - return NULL; - - // does not exist, so make a new one - Section* sect = new Section(sectionName, segmentName, zeroFill, untrustedZeroFill); - fgMapping[sectionName] = sect; - fgSections.push_back(sect); - if ( (strcmp(sectionName, "__text") == 0) && (strcmp(segmentName, "__TEXT") == 0) ) { - // special case __StaticInit to be right after __text - find("__StaticInit", "__TEXT", false, true); - } +InternalState::FinalSection::FinalSection(const ld::Section& sect, uint32_t sectionsSeen, const Options& opts) + : ld::Internal::FinalSection(sect), + _segmentOrder(segmentOrder(sect, opts)), + _sectionOrder(sectionOrder(sect, sectionsSeen, opts)) +{ + //fprintf(stderr, "FinalSection(%16s, %16s) _segmentOrder=%3d, _sectionOrder=0x%08X\n", + // this->segmentName(), this->sectionName(), _segmentOrder, _sectionOrder); +} - // remember segment discovery order - if ( fgSegmentDiscoverOrder.find(segmentName) == fgSegmentDiscoverOrder.end() ) - fgSegmentDiscoverOrder[segmentName] = fgSegmentDiscoverOrder.size(); +const ld::Section& InternalState::FinalSection::outputSection(const ld::Section& sect, bool mergeZeroFill) +{ + // merge sections in final linked image + switch ( sect.type() ) { + case ld::Section::typeLiteral4: + case ld::Section::typeLiteral8: + case ld::Section::typeLiteral16: + if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) + return _s_TEXT_const; + break; + case ld::Section::typeUnclassified: + if ( strcmp(sect.segmentName(), "__DATA") == 0 ) { + if ( strcmp(sect.sectionName(), "__datacoal_nt") == 0 ) + return _s_DATA_data; + if ( strcmp(sect.sectionName(), "__const_coal") == 0 ) + return _s_DATA_const; + } + else if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) { + if ( strcmp(sect.sectionName(), "__const_coal") == 0 ) + return _s_TEXT_const; + } + else if ( strcmp(sect.segmentName(), "__DATA_DIRTY") == 0 ) { + if ( strcmp(sect.sectionName(), "__datacoal_nt") == 0 ) + return _s_DATA_DIRTY_data; + } + else if ( strcmp(sect.segmentName(), "__DATA_CONST") == 0 ) { + if ( strcmp(sect.sectionName(), "__const_coal") == 0 ) + return _s_DATA_CONST_const; + } + break; + case ld::Section::typeZeroFill: + if ( mergeZeroFill ) + return _s_DATA_zerofill; + break; + case ld::Section::typeCode: + if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) { + if ( strcmp(sect.sectionName(), "__textcoal_nt") == 0 ) + return _s_TEXT_text; + else if ( strcmp(sect.sectionName(), "__StaticInit") == 0 ) + return _s_TEXT_text; + } + break; + case ld::Section::typeNonLazyPointer: + if ( strcmp(sect.segmentName(), "__DATA") == 0 ) { + if ( strcmp(sect.sectionName(), "__nl_symbol_ptr") == 0 ) + return _s_DATA_nl_symbol_ptr; + } + else if ( strcmp(sect.segmentName(), "__IMPORT") == 0 ) { + if ( strcmp(sect.sectionName(), "__pointers") == 0 ) + return _s_DATA_nl_symbol_ptr; + } + break; + case ld::Section::typeTentativeDefs: + if ( (strcmp(sect.segmentName(), "__DATA") == 0) && (strcmp(sect.sectionName(), "__comm/tent") == 0) ) { + if ( mergeZeroFill ) + return _s_DATA_zerofill; + else + return _s_DATA_common; + } + break; + // FIX ME: more + default: + break; + } + return sect; +} +const ld::Section& InternalState::FinalSection::objectOutputSection(const ld::Section& sect, const Options& options) +{ + // in -r mode the only section that ever changes is __tenative -> __common with -d option + if ( (sect.type() == ld::Section::typeTentativeDefs) && options.makeTentativeDefinitionsReal()) + return _s_DATA_common; return sect; } -int Section::Sorter::segmentOrdinal(const char* segName) +uint32_t InternalState::FinalSection::segmentOrder(const ld::Section& sect, const Options& options) { - if ( strcmp(segName, "__HEADER") == 0 ) - return 1; - if ( strcmp(segName, "__PAGEZERO") == 0 ) + if ( options.outputKind() == Options::kPreload ) { + if ( strcmp(sect.segmentName(), "__HEADER") == 0 ) + return 0; + const std::vector& order = options.segmentOrder(); + for (size_t i=0; i != order.size(); ++i) { + if ( strcmp(sect.segmentName(), order[i]) == 0 ) + return i+1; + } + if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) + return order.size()+1; + if ( strcmp(sect.segmentName(), "__DATA") == 0 ) + return order.size()+2; + } + else if ( options.outputKind() == Options::kStaticExecutable ) { + const std::vector& order = options.segmentOrder(); + for (size_t i=0; i != order.size(); ++i) { + if ( strcmp(sect.segmentName(), order[i]) == 0 ) + return i+1; + } + if ( strcmp(sect.segmentName(), "__PAGEZERO") == 0 ) + return 0; + if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) + return order.size()+1; + if ( strcmp(sect.segmentName(), "__DATA") == 0 ) + return order.size()+2; + } + else { + if ( strcmp(sect.segmentName(), "__PAGEZERO") == 0 ) + return 0; + if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) + return 1; + if ( strcmp(sect.segmentName(), "__TEXT_EXEC") == 0 ) + return 2; + // in -r mode, want __DATA last so zerofill sections are at end + if ( strcmp(sect.segmentName(), "__DATA") == 0 ) + return (options.outputKind() == Options::kObjectFile) ? 6 : 3; + if ( strcmp(sect.segmentName(), "__OBJC") == 0 ) + return 4; + if ( strcmp(sect.segmentName(), "__IMPORT") == 0 ) + return 5; + } + // layout non-standard segments in order seen (+100 to shift beyond standard segments) + for (uint32_t i=0; i < _s_segmentsSeen.size(); ++i) { + if ( strcmp(_s_segmentsSeen[i], sect.segmentName()) == 0 ) + return i+100; + } + _s_segmentsSeen.push_back(sect.segmentName()); + return _s_segmentsSeen.size()-1+100; +} + +uint32_t InternalState::FinalSection::sectionOrder(const ld::Section& sect, uint32_t sectionsSeen, const Options& options) +{ + if ( sect.type() == ld::Section::typeFirstSection ) + return 0; + if ( sect.type() == ld::Section::typeMachHeader ) return 1; - if ( strcmp(segName, "__TEXT") == 0 ) - return 2; - if ( strcmp(segName, "__DATA") == 0 ) - return 3; - if ( strcmp(segName, "__OBJC") == 0 ) - return 4; - if ( strcmp(segName, "__OBJC2") == 0 ) - return 5; - if ( strcmp(segName, "__LINKEDIT") == 0 ) - return INT_MAX; // linkedit segment should always sort last - else - return fgSegmentDiscoverOrder[segName]+6; + if ( sect.type() == ld::Section::typeLastSection ) + return INT_MAX; + const std::vector* sectionList = options.sectionOrder(sect.segmentName()); + if ( ((options.outputKind() == Options::kPreload) || (options.outputKind() == Options::kDyld)) && (sectionList != NULL) ) { + uint32_t count = 10; + for (std::vector::const_iterator it=sectionList->begin(); it != sectionList->end(); ++it, ++count) { + if ( strcmp(*it, sect.sectionName()) == 0 ) + return count; + } + } + if ( strcmp(sect.segmentName(), "__TEXT") == 0 ) { + switch ( sect.type() ) { + case ld::Section::typeCode: + // make __text always be first "code" section + if ( strcmp(sect.sectionName(), "__text") == 0 ) + return 10; + else + return 11; + case ld::Section::typeNonStdCString: + if ( (strcmp(sect.sectionName(), "__oslogstring") == 0) && options.makeEncryptable() ) + return INT_MAX-1; + else + return sectionsSeen+20; + case ld::Section::typeStub: + return 12; + case ld::Section::typeStubHelper: + return 13; + case ld::Section::typeLSDA: + return INT_MAX-4; + case ld::Section::typeUnwindInfo: + return INT_MAX-3; + case ld::Section::typeCFI: + return INT_MAX-2; + case ld::Section::typeStubClose: + return INT_MAX; + default: + return sectionsSeen+20; + } + } + else if ( strncmp(sect.segmentName(), "__DATA", 6) == 0 ) { + switch ( sect.type() ) { + case ld::Section::typeLazyPointerClose: + return 8; + case ld::Section::typeDyldInfo: + return 9; + case ld::Section::typeNonLazyPointer: + return 10; + case ld::Section::typeLazyPointer: + return 11; + case ld::Section::typeInitializerPointers: + return 12; + case ld::Section::typeTerminatorPointers: + return 13; + case ld::Section::typeTLVInitialValues: + return INT_MAX-259; // need TLV zero-fill to follow TLV init values + case ld::Section::typeTLVZeroFill: + return INT_MAX-258; + case ld::Section::typeZeroFill: + // make sure __huge is always last zerofill section + if ( strcmp(sect.sectionName(), "__huge") == 0 ) + return INT_MAX-1; + else + return INT_MAX-256+sectionsSeen; // zero fill need to be last and in "seen" order + default: + // __DATA,__const section should be near __mod_init_func not __data + if ( strcmp(sect.sectionName(), "__const") == 0 ) + return 14; + // Linker should put __cfstring near __const + if ( strcmp(sect.sectionName(), "__cfstring") == 0 ) + return 15; + // Reorder sections to reduce page faults in object files + else if ( strcmp(sect.sectionName(), "__objc_classlist") == 0 ) + return 20; + else if ( strcmp(sect.sectionName(), "__objc_nlclslist") == 0 ) + return 21; + else if ( strcmp(sect.sectionName(), "__objc_catlist") == 0 ) + return 22; + else if ( strcmp(sect.sectionName(), "__objc_nlcatlist") == 0 ) + return 23; + else if ( strcmp(sect.sectionName(), "__objc_protolist") == 0 ) + return 24; + else if ( strcmp(sect.sectionName(), "__objc_imageinfo") == 0 ) + return 25; + else if ( strcmp(sect.sectionName(), "__objc_const") == 0 ) + return 26; + else if ( strcmp(sect.sectionName(), "__objc_selrefs") == 0 ) + return 27; + else if ( strcmp(sect.sectionName(), "__objc_msgrefs") == 0 ) + return 28; + else if ( strcmp(sect.sectionName(), "__objc_protorefs") == 0 ) + return 29; + else if ( strcmp(sect.sectionName(), "__objc_classrefs") == 0 ) + return 30; + else if ( strcmp(sect.sectionName(), "__objc_superrefs") == 0 ) + return 31; + else if ( strcmp(sect.sectionName(), "__objc_ivar") == 0 ) + return 32; + else if ( strcmp(sect.sectionName(), "__objc_data") == 0 ) + return 33; + else + return sectionsSeen+40; + } + } + // make sure zerofill in any other section is at end of segment + if ( sect.type() == ld::Section::typeZeroFill ) + return INT_MAX-256+sectionsSeen; + return sectionsSeen+20; +} + +#ifndef NDEBUG +static void validateFixups(const ld::Atom& atom) +{ + //fprintf(stderr, "validateFixups %s\n", atom.name()); + bool lastWasClusterEnd = true; + ld::Fixup::Cluster lastClusterSize = ld::Fixup::k1of1; + uint32_t curClusterOffsetInAtom = 0; + for (ld::Fixup::iterator fit=atom.fixupsBegin(); fit != atom.fixupsEnd(); ++fit) { + //fprintf(stderr, " fixup offset=%d, cluster=%d\n", fit->offsetInAtom, fit->clusterSize); + assert((fit->offsetInAtom <= atom.size()) || (fit->offsetInAtom == 0)); + if ( fit->firstInCluster() ) { + assert(lastWasClusterEnd); + curClusterOffsetInAtom = fit->offsetInAtom; + lastWasClusterEnd = (fit->clusterSize == ld::Fixup::k1of1); + } + else { + assert(!lastWasClusterEnd); + assert(fit->offsetInAtom == curClusterOffsetInAtom); + switch ((ld::Fixup::Cluster)fit->clusterSize) { + case ld::Fixup::k1of1: + case ld::Fixup::k1of2: + case ld::Fixup::k1of3: + case ld::Fixup::k1of4: + case ld::Fixup::k1of5: + lastWasClusterEnd = false; + break; + case ld::Fixup::k2of2: + assert(lastClusterSize = ld::Fixup::k1of2); + lastWasClusterEnd = true; + break; + case ld::Fixup::k2of3: + assert(lastClusterSize = ld::Fixup::k1of3); + lastWasClusterEnd = false; + break; + case ld::Fixup::k2of4: + assert(lastClusterSize = ld::Fixup::k1of4); + lastWasClusterEnd = false; + break; + case ld::Fixup::k2of5: + assert(lastClusterSize = ld::Fixup::k1of5); + lastWasClusterEnd = false; + break; + case ld::Fixup::k3of3: + assert(lastClusterSize = ld::Fixup::k2of3); + lastWasClusterEnd = true; + break; + case ld::Fixup::k3of4: + assert(lastClusterSize = ld::Fixup::k2of4); + lastWasClusterEnd = false; + break; + case ld::Fixup::k3of5: + assert(lastClusterSize = ld::Fixup::k2of5); + lastWasClusterEnd = false; + break; + case ld::Fixup::k4of4: + assert(lastClusterSize = ld::Fixup::k3of4); + lastWasClusterEnd = true; + break; + case ld::Fixup::k4of5: + assert(lastClusterSize = ld::Fixup::k3of5); + lastWasClusterEnd = false; + break; + case ld::Fixup::k5of5: + assert(lastClusterSize = ld::Fixup::k4of5); + lastWasClusterEnd = true; + break; + } + } + lastClusterSize = fit->clusterSize; + if ( fit->binding == ld::Fixup::bindingDirectlyBound ) { + assert(fit->u.target != NULL); + } + } + switch (lastClusterSize) { + case ld::Fixup::k1of1: + case ld::Fixup::k2of2: + case ld::Fixup::k3of3: + case ld::Fixup::k4of4: + case ld::Fixup::k5of5: + break; + default: + assert(0 && "last fixup was not end of cluster"); + break; + } } +#endif +bool InternalState::hasReferenceToWeakExternal(const ld::Atom& atom) +{ + // if __DATA,__const atom has pointer to weak external symbol, don't move to __DATA_CONST + const ld::Atom* target = NULL; + for (ld::Fixup::iterator fit=atom.fixupsBegin(); fit != atom.fixupsEnd(); ++fit) { + if ( fit->firstInCluster() ) { + target = NULL; + } + switch ( fit->binding ) { + case ld::Fixup::bindingNone: + case ld::Fixup::bindingByNameUnbound: + break; + case ld::Fixup::bindingByContentBound: + case ld::Fixup::bindingDirectlyBound: + target = fit->u.target; + break; + case ld::Fixup::bindingsIndirectlyBound: + target = indirectBindingTable[fit->u.bindingIndex]; + break; + } + if ( (target != NULL) && (target->definition() == ld::Atom::definitionRegular) + && (target->combine() == ld::Atom::combineByName) && (target->scope() == ld::Atom::scopeGlobal) ) { + return true; + } + } + return false; +} -bool Section::Sorter::operator()(Section* left, Section* right) +bool InternalState::inMoveRWChain(const ld::Atom& atom, const char* filePath, const char*& dstSeg, bool& wildCardMatch) { - // Segment is primary sort key - int leftSegOrdinal = segmentOrdinal(left->fSegmentName); - int rightSegOrdinal = segmentOrdinal(right->fSegmentName); - if ( leftSegOrdinal < rightSegOrdinal ) - return true; - if ( leftSegOrdinal > rightSegOrdinal ) + if ( !_options.hasDataSymbolMoves() ) return false; - // zerofill section sort to the end - if ( !left->fZeroFill && right->fZeroFill ) + auto pos = _pendingSegMove.find(&atom); + if ( pos != _pendingSegMove.end() ) { + dstSeg = pos->second; return true; - if ( left->fZeroFill && !right->fZeroFill ) - return false; - - // section discovery order is last sort key - return left->fIndex < right->fIndex; -} + } -void Section::assignIndexes() -{ - //printf("unsorted sections:\n"); - //for (std::vector::iterator it=fgSections.begin(); it != fgSections.end(); it++) { - // printf("section: name=%s, segment: name=%s, discovery order=%d\n", (*it)->fSectionName, (*it)->fSegmentName, (*it)->fIndex); - //} + bool result = false; + if ( _options.moveRwSymbol(atom.name(), filePath, dstSeg, wildCardMatch) ) + result = true; - // sort it - std::sort(fgSections.begin(), fgSections.end(), Section::Sorter()); + for (ld::Fixup::iterator fit=atom.fixupsBegin(); fit != atom.fixupsEnd(); ++fit) { + if ( fit->kind == ld::Fixup::kindNoneFollowOn ) { + if ( fit->binding == ld::Fixup::bindingDirectlyBound ) { + if ( inMoveRWChain(*(fit->u.target), filePath, dstSeg, wildCardMatch) ) + result = true; + } + } + } - // assign correct section ordering to each Section object - unsigned int newOrder = 1; - for (std::vector::iterator it=fgSections.begin(); it != fgSections.end(); it++) - (*it)->fIndex = newOrder++; + if ( result ) { + for (ld::Fixup::iterator fit=atom.fixupsBegin(); fit != atom.fixupsEnd(); ++fit) { + if ( fit->kind == ld::Fixup::kindNoneFollowOn ) { + if ( fit->binding == ld::Fixup::bindingDirectlyBound ) { + _pendingSegMove[fit->u.target] = dstSeg; + } + } + } + } - //printf("sorted sections:\n"); - //for (std::vector::iterator it=fgSections.begin(); it != fgSections.end(); it++) { - // printf("section: index=%d, obj=%p, name=%s\n", (*it)->fIndex, (*it), (*it)->fSectionName); - //} + return result; } -class Linker : public ObjectFile::Reader::DylibHander { -public: - Linker(int argc, const char* argv[]); - - const char* getArchPrefix(); - const char* architectureName(); - bool showArchitectureInErrors(); - bool isInferredArchitecture(); - void createReaders(); - void createWriter(); - void addInputFile(ObjectFile::Reader* reader, const Options::FileInfo& ); - void setOutputFile(ExecutableFile::Writer* writer); - void link(); - void optimize(); - - // implemenation from ObjectFile::Reader::DylibHander - virtual ObjectFile::Reader* findDylib(const char* installPath, const char* fromPath); - -private: - struct WhyLiveBackChain - { - WhyLiveBackChain* previous; - ObjectFile::Atom* referer; - }; - - ObjectFile::Reader* createReader(const Options::FileInfo&); - void addAtom(ObjectFile::Atom& atom); - void addAtoms(std::vector& atoms); - void buildAtomList(); - void adjustScope(); - void processDylibs(); - void markDead(ObjectFile::Atom* atom); - void updateConstraints(ObjectFile::Reader* reader); - void loadAndResolve(); - void processDTrace(); - void checkObjC(); - void loadUndefines(); - void checkUndefines(); - void resolveReferences(); - void deadStripResolve(); - void addLiveRoot(const char* name); - void moveToFrontOfSection(ObjectFile::Atom* atom); - ObjectFile::Atom* findAtom(const Options::OrderedSymbol& pair); - void logArchive(ObjectFile::Reader* reader); - void sortSections(); - void sortAtoms(); - void tweakLayout(); - void writeDotOutput(); - static bool minimizeStab(ObjectFile::Reader::Stab& stab); - static const char* truncateStabString(const char* str); - void collectDebugInfo(); - void writeOutput(); - ObjectFile::Atom* entryPoint(bool orInit); - ObjectFile::Atom* dyldClassicHelper(); - ObjectFile::Atom* dyldCompressedHelper(); - ObjectFile::Atom* dyldLazyLibraryHelper(); - const char* assureFullPath(const char* path); - void markLive(ObjectFile::Atom& atom, Linker::WhyLiveBackChain* previous); - void collectStabs(ObjectFile::Reader* reader, std::map& atomOrdinals); - void synthesizeDebugNotes(std::vector& allAtomsByReader); - void printStatistics(); - void printTime(const char* msg, uint64_t partTime, uint64_t totalTime); - char* commatize(uint64_t in, char* out); - void getVMInfo(vm_statistics_data_t& info); - cpu_type_t inferArchitecture(); - void checkDylibClientRestrictions(ObjectFile::Reader* reader); - void logDylib(ObjectFile::Reader* reader, bool indirect); - - void resolve(ObjectFile::Reference* reference); - void resolveFrom(ObjectFile::Reference* reference); - std::vector* addJustInTimeAtoms(const char* name, bool searchDylibs, bool searchArchives, bool okToMakeProxy); - void addJustInTimeAtomsAndMarkLive(const char* name); - - ObjectFile::Reader* addDylib(ObjectFile::Reader* reader, const Options::FileInfo& info, uint64_t mappedLen); - ObjectFile::Reader* addObject(ObjectFile::Reader* reader, const Options::FileInfo& info, uint64_t mappedLen); - ObjectFile::Reader* addArchive(ObjectFile::Reader* reader, const Options::FileInfo& info, uint64_t mappedLen); - - void logTraceInfo(const char* format, ...); - - - class SymbolTable - { - public: - typedef __gnu_cxx::hash_map, CStringEquals> Mapper; - SymbolTable(Linker&); - void require(const char* name); - bool add(ObjectFile::Atom& atom); - ObjectFile::Atom* find(const char* name); - unsigned int getRequireCount() { return fRequireCount; } - void getUndefinesNames(std::vector& undefines); - void getTentativesNames(std::vector& tents); - bool hasExternalTentativeDefinitions() { return fHasExternalTentativeDefinitions; } - bool hasExternalWeakDefinitions() { return fHasExternalWeakDefinitions; } - void setHasExternalWeakDefinitions(bool value) { fHasExternalWeakDefinitions = value; } - Mapper::iterator begin() { return fTable.begin(); } - Mapper::iterator end() { return fTable.end(); } +bool InternalState::inMoveROChain(const ld::Atom& atom, const char* filePath, const char*& dstSeg, bool& wildCardMatch) +{ + if ( !_options.hasCodeSymbolMoves() ) + return false; - private: - Linker& fOwner; - Mapper fTable; - unsigned int fRequireCount; - bool fHasExternalTentativeDefinitions; - bool fHasExternalWeakDefinitions; - }; + auto pos = _pendingSegMove.find(&atom); + if ( pos != _pendingSegMove.end() ) { + dstSeg = pos->second; + return true; + } - class AtomSorter - { - public: - AtomSorter(std::map* map, std::set& inits, - std::set& terms) : - fOverriddenOrdinalMap(map), fInitializerSet(inits), fTerminatorSet(terms) {} - bool operator()(const ObjectFile::Atom* left, const ObjectFile::Atom* right); - private: - std::map* fOverriddenOrdinalMap; - std::set& fInitializerSet; - std::set& fTerminatorSet; - }; + bool result = false; + if ( _options.moveRoSymbol(atom.name(), filePath, dstSeg, wildCardMatch) ) + result = true; - typedef std::map SectionOrder; + for (ld::Fixup::iterator fit=atom.fixupsBegin(); fit != atom.fixupsEnd(); ++fit) { + if ( fit->kind == ld::Fixup::kindNoneFollowOn ) { + if ( fit->binding == ld::Fixup::bindingDirectlyBound ) { + if ( inMoveROChain(*(fit->u.target), filePath, dstSeg, wildCardMatch) ) + result = true; + } + } + } - struct DTraceProbeInfo { - DTraceProbeInfo(const ObjectFile::Atom* a, uint32_t o, const char* n) : atom(a), offset(o), probeName(n) {} - const ObjectFile::Atom* atom; - uint32_t offset; - const char* probeName; - }; - typedef __gnu_cxx::hash_map, __gnu_cxx::hash, CStringEquals> ProviderToProbes; - typedef __gnu_cxx::hash_set, CStringEquals> CStringSet; - typedef __gnu_cxx::hash_map, CStringEquals> InstallNameToReader; + if ( result ) { + for (ld::Fixup::iterator fit=atom.fixupsBegin(); fit != atom.fixupsEnd(); ++fit) { + if ( fit->kind == ld::Fixup::kindNoneFollowOn ) { + if ( fit->binding == ld::Fixup::bindingDirectlyBound ) { + _pendingSegMove[fit->u.target] = dstSeg; + } + } + } + } - struct IndirectLibrary { - const char* path; - uint64_t fileLen; - ObjectFile::Reader* reader; - std::set parents; - ObjectFile::Reader* reExportedViaDirectLibrary; - }; + return result; +} - ObjectFile::Reader* findDirectLibraryWhichReExports(struct IndirectLibrary& indirectLib); - Options fOptions; - SymbolTable fGlobalSymbolTable; - uint32_t fNextInputOrdinal; - std::vector fInputFiles; - ExecutableFile::Writer* fOutputFile; - InstallNameToReader fDylibMap; - std::map fDylibOptionsMap; - std::set fDylibsProcessed; - ObjectFile::Reader* fBundleLoaderReader; - std::vector fReadersThatHaveSuppliedAtoms; - std::vector fAllAtoms; - std::set fArchiveReaders; - std::set fArchiveReadersLogged; - std::set fDeadAtoms; - std::set fLiveAtoms; - std::set fLiveRootAtoms; - std::set fInitializerAtoms; - std::set fTerminatorAtoms; - std::set fRegularDefAtomsThatOverrideADylibsWeakDef; - std::vector fStabs; - std::vector fAtomsWithUnresolvedReferences; - std::set fAtomsOverriddenByLateLoads; - bool fInitialLoadsDone; - bool fCreateUUID; - bool fCanScatter; - SectionOrder fSectionOrder; - cpu_type_t fArchitecture; - const char* fArchitectureName; - bool fArchitectureInferred; - bool fDirectLibrariesComplete; - bool fBiggerThanTwoGigOutput; - uint64_t fOutputFileSize; - uint64_t fTotalZeroFillSize; - uint64_t fTotalSize; - uint64_t fStartTime; - uint64_t fStartCreateReadersTime; - uint64_t fStartCreateWriterTime; - uint64_t fStartBuildAtomsTime; - uint64_t fStartLoadAndResolveTime; - uint64_t fStartSortTime; - uint64_t fStartDebugTime; - uint64_t fStartWriteTime; - uint64_t fEndTime; - uint64_t fTotalObjectSize; - uint64_t fTotalArchiveSize; - uint32_t fTotalObjectLoaded; - uint32_t fTotalArchivesLoaded; - uint32_t fTotalDylibsLoaded; - vm_statistics_data_t fStartVMInfo; - ObjectFile::Reader::ObjcConstraint fCurrentObjCConstraint; - ObjectFile::Reader::CpuConstraint fCurrentCpuConstraint; - bool fObjcReplacmentClasses; - bool fAllDirectDylibsLoaded; -}; -Linker::Linker(int argc, const char* argv[]) - : fOptions(argc, argv), fGlobalSymbolTable(*this), fNextInputOrdinal(1), fOutputFile(NULL), fBundleLoaderReader(NULL), - fInitialLoadsDone(false), fCreateUUID(fOptions.outputKind() != Options::kObjectFile), fCanScatter(true), - fArchitecture(0), fArchitectureInferred(false), fDirectLibrariesComplete(false), fBiggerThanTwoGigOutput(false), - fOutputFileSize(0), fTotalZeroFillSize(0), fTotalSize(0), fTotalObjectSize(0), - fTotalArchiveSize(0), fTotalObjectLoaded(0), fTotalArchivesLoaded(0), fTotalDylibsLoaded(0), - fCurrentObjCConstraint(ObjectFile::Reader::kObjcNone), fCurrentCpuConstraint(ObjectFile::Reader::kCpuAny), - fObjcReplacmentClasses(false), fAllDirectDylibsLoaded(false) +ld::Internal::FinalSection* InternalState::addAtom(const ld::Atom& atom) { - fStartTime = mach_absolute_time(); - if ( fOptions.printStatistics() ) - getVMInfo(fStartVMInfo); + //fprintf(stderr, "addAtom: %s\n", atom.name()); + ld::Internal::FinalSection* fs = NULL; + const char* curSectName = atom.section().sectionName(); + const char* curSegName = atom.section().segmentName(); + ld::Section::Type sectType = atom.section().type(); + const ld::File* f = atom.file(); + const char* path = (f != NULL) ? f->path() : NULL; + if ( atom.section().type() == ld::Section::typeTentativeDefs ) { + // tentative defintions don't have a real section name yet + sectType = ld::Section::typeZeroFill; + if ( _options.mergeZeroFill() ) + curSectName = FinalSection::_s_DATA_zerofill.sectionName(); + else + curSectName = FinalSection::_s_DATA_common.sectionName(); + } + // Support for -move_to_r._segment + if ( atom.symbolTableInclusion() == ld::Atom::symbolTableIn ) { + const char* dstSeg; + bool wildCardMatch; + if ( inMoveRWChain(atom, path, dstSeg, wildCardMatch) ) { + if ( (sectType != ld::Section::typeZeroFill) + && (sectType != ld::Section::typeUnclassified) + && (sectType != ld::Section::typeTentativeDefs) + && (sectType != ld::Section::typeDyldInfo) ) { + if ( !wildCardMatch ) + warning("cannot move symbol '%s' from file %s to segment '%s' because symbol is not data (is %d)", atom.name(), path, dstSeg, sectType); + } + else { + curSegName = dstSeg; + if ( _options.traceSymbolLayout() ) + printf("symbol '%s', -move_to_rw_segment mapped it to %s/%s\n", atom.name(), curSegName, curSectName); + fs = this->getFinalSection(curSegName, curSectName, sectType); + } + } + if ( (fs == NULL) && inMoveROChain(atom, path, dstSeg, wildCardMatch) ) { + if ( (sectType != ld::Section::typeCode) + && (sectType != ld::Section::typeUnclassified) ) { + if ( !wildCardMatch ) + warning("cannot move symbol '%s' from file %s to segment '%s' because symbol is not code (is %d)", atom.name(), path, dstSeg, sectType); + } + else { + curSegName = dstSeg; + if ( _options.traceSymbolLayout() ) + printf("symbol '%s', -move_to_ro_segment mapped it to %s/%s\n", atom.name(), curSegName, curSectName); + fs = this->getFinalSection(curSegName, curSectName, ld::Section::typeCode); + } + } + } + // support for -rename_section and -rename_segment + for (const Options::SectionRename& rename : _options.sectionRenames()) { + if ( (strcmp(curSectName, rename.fromSection) == 0) && (strcmp(curSegName, rename.fromSegment) == 0) ) { + if ( _options.useDataConstSegment() && (strcmp(curSectName, "__const") == 0) && (strcmp(curSegName, "__DATA") == 0) && hasReferenceToWeakExternal(atom) ) { + // if __DATA,__const atom has pointer to weak external symbol, don't move to __DATA_CONST + curSectName = "__const_weak"; + fs = this->getFinalSection(curSegName, curSectName, sectType); + if ( _options.traceSymbolLayout() ) + printf("symbol '%s', contains pointers to weak symbols, so mapped it to __DATA/__const_weak\n", atom.name()); + } + else if ( _options.useDataConstSegment() && (sectType == ld::Section::typeNonLazyPointer) && hasReferenceToWeakExternal(atom) ) { + // if __DATA,__nl_symbol_ptr atom has pointer to weak external symbol, don't move to __DATA_CONST + curSectName = "__got_weak"; + fs = this->getFinalSection("__DATA", curSectName, sectType); + if ( _options.traceSymbolLayout() ) + printf("symbol '%s', contains pointers to weak symbols, so mapped it to __DATA/__got_weak\n", atom.name()); + } + else { + curSegName = rename.toSegment; + curSectName = rename.toSection; + fs = this->getFinalSection(rename.toSegment, rename.toSection, sectType); + if ( _options.traceSymbolLayout() ) + printf("symbol '%s', -rename_section mapped it to %s/%s\n", atom.name(), fs->segmentName(), fs->sectionName()); + } + } + } + for (const Options::SegmentRename& rename : _options.segmentRenames()) { + if ( strcmp(curSegName, rename.fromSegment) == 0 ) { + if ( _options.traceSymbolLayout() ) + printf("symbol '%s', -rename_segment mapped it to %s/%s\n", atom.name(), rename.toSegment, curSectName); + fs = this->getFinalSection(rename.toSegment, curSectName, sectType); + } + } - fArchitecture = fOptions.architecture(); - if ( fArchitecture == 0 ) { - // -arch not specified, scan .o files to figure out what it should be - fArchitecture = inferArchitecture(); - fArchitectureInferred = true; + // if no override, use default location + if ( fs == NULL ) { + fs = this->getFinalSection(atom.section()); + if ( _options.traceSymbolLayout() && (atom.symbolTableInclusion() == ld::Atom::symbolTableIn) ) + printf("symbol '%s', use default mapping to %s/%s\n", atom.name(), fs->segmentName(), fs->sectionName()); } - switch (fArchitecture) { - case CPU_TYPE_POWERPC: - fArchitectureName = "ppc"; - break; - case CPU_TYPE_POWERPC64: - fArchitectureName = "ppc64"; - break; - case CPU_TYPE_I386: - fArchitectureName = "i386"; - break; - case CPU_TYPE_X86_64: - fArchitectureName = "x86_64"; - break; - case CPU_TYPE_ARM: - fArchitectureName = "arm"; - break; - default: - fArchitectureName = "unknown architecture"; - break; + + //fprintf(stderr, "InternalState::doAtom(%p), name=%s, sect=%s, finalseg=%s\n", &atom, atom.name(), atom.section().sectionName(), fs->segmentName()); +#ifndef NDEBUG + validateFixups(atom); +#endif + if ( _atomsOrderedInSections ) { + // make sure this atom is placed before any trailing section$end$ atom + if ( (fs->atoms.size() > 1) && (fs->atoms.back()->contentType() == ld::Atom::typeSectionEnd) ) { + // last atom in section$end$ atom, insert before it + const ld::Atom* endAtom = fs->atoms.back(); + fs->atoms.pop_back(); + fs->atoms.push_back(&atom); + fs->atoms.push_back(endAtom); + } + else { + // not end atom, just append new atom + fs->atoms.push_back(&atom); + } } + else { + // normal case + fs->atoms.push_back(&atom); + } + this->atomToSection[&atom] = fs; + return fs; } -const char* Linker::architectureName() -{ - return fArchitectureName; -} -bool Linker::showArchitectureInErrors() -{ - return fOptions.printArchPrefix(); -} -bool Linker::isInferredArchitecture() -{ - return fArchitectureInferred; +ld::Internal::FinalSection* InternalState::getFinalSection(const char* seg, const char* sect, ld::Section::Type type) +{ + for (std::vector::iterator it=sections.begin(); it != sections.end(); ++it) { + if ( (strcmp((*it)->segmentName(),seg) == 0) && (strcmp((*it)->sectionName(),sect) == 0) ) + return *it; + } + return this->getFinalSection(*new ld::Section(seg, sect, type, false)); } -cpu_type_t Linker::inferArchitecture() -{ - // scan all input files, looking for a thin .o file. - // the first one found is presumably the architecture to link - uint8_t buffer[sizeof(mach_header_64)]; - std::vector& files = fOptions.getInputFiles(); - for (std::vector::iterator it = files.begin(); it != files.end(); ++it) { - int fd = ::open(it->path, O_RDONLY, 0); - if ( fd != -1 ) { - ssize_t amount = read(fd, buffer, sizeof(buffer)); - ::close(fd); - if ( amount >= (ssize_t)sizeof(buffer) ) { - if ( mach_o::relocatable::Reader::validFile(buffer) ) { - //warning("-arch not used, infering -arch ppc based on %s", it->path); - return CPU_TYPE_POWERPC; - } - else if ( mach_o::relocatable::Reader::validFile(buffer) ) { - //warning("-arch not used, infering -arch ppc64 based on %s", it->path); - return CPU_TYPE_POWERPC64; - } - else if ( mach_o::relocatable::Reader::validFile(buffer) ) { - //warning("-arch not used, infering -arch i386 based on %s", it->path); - return CPU_TYPE_I386; - } - else if ( mach_o::relocatable::Reader::validFile(buffer) ) { - //warning("-arch not used, infering -arch x86_64 based on %s", it->path); - return CPU_TYPE_X86_64; +ld::Internal::FinalSection* InternalState::getFinalSection(const ld::Section& inputSection) +{ + const ld::Section* baseForFinalSection = &inputSection; + + // see if input section already has a FinalSection + SectionInToOut::iterator pos = _sectionInToFinalMap.find(&inputSection); + if ( pos != _sectionInToFinalMap.end() ) { + return pos->second; + } + + // otherwise, create a new final section + switch ( _options.outputKind() ) { + case Options::kStaticExecutable: + case Options::kDynamicExecutable: + case Options::kDynamicLibrary: + case Options::kDynamicBundle: + case Options::kDyld: + case Options::kKextBundle: + case Options::kPreload: + { + // coalesce some sections + const ld::Section& outSect = FinalSection::outputSection(inputSection, _options.mergeZeroFill()); + pos = _sectionInToFinalMap.find(&outSect); + if ( pos != _sectionInToFinalMap.end() ) { + _sectionInToFinalMap[&inputSection] = pos->second; + //fprintf(stderr, "_sectionInToFinalMap[%p] = %p\n", &inputSection, pos->second); + return pos->second; } - else if ( mach_o::relocatable::Reader::validFile(buffer) ) { - //warning("-arch not used, infering -arch arm based on %s", it->path); - return CPU_TYPE_ARM; + else if ( outSect != inputSection ) { + // new output section created, but not in map + baseForFinalSection = &outSect; } } - } + break; + case Options::kObjectFile: + baseForFinalSection = &FinalSection::objectOutputSection(inputSection, _options); + pos = _sectionInToFinalMap.find(baseForFinalSection); + if ( pos != _sectionInToFinalMap.end() ) { + _sectionInToFinalMap[&inputSection] = pos->second; + //fprintf(stderr, "_sectionInToFinalMap[%p] = %p\n", &inputSection, pos->second); + return pos->second; + } + break; } - // no thin .o files found, so default to same architecture this was built as - warning("-arch not specified"); -#if __ppc__ - return CPU_TYPE_POWERPC; -#elif __i386__ - return CPU_TYPE_I386; -#elif __ppc64__ - return CPU_TYPE_POWERPC64; -#elif __x86_64__ - return CPU_TYPE_X86_64; -#elif __arm__ - return CPU_TYPE_ARM; -#else - #error unknown default architecture -#endif + InternalState::FinalSection* result = new InternalState::FinalSection(*baseForFinalSection, + _sectionInToFinalMap.size(), _options); + _sectionInToFinalMap[baseForFinalSection] = result; + //fprintf(stderr, "_sectionInToFinalMap[%p(%s)] = %p\n", baseForFinalSection, baseForFinalSection->sectionName(), result); + sections.push_back(result); + return result; } -void Linker::addInputFile(ObjectFile::Reader* reader, const Options::FileInfo& info) +int InternalState::FinalSection::sectionComparer(const void* l, const void* r) { - fInputFiles.push_back(reader); - fDylibOptionsMap[reader] = info.options; + const FinalSection* left = *(FinalSection**)l; + const FinalSection* right = *(FinalSection**)r; + if ( left->_segmentOrder != right->_segmentOrder ) + return (left->_segmentOrder - right->_segmentOrder); + return (left->_sectionOrder - right->_sectionOrder); } -void Linker::setOutputFile(ExecutableFile::Writer* writer) +void InternalState::sortSections() { - fOutputFile = writer; + //fprintf(stderr, "UNSORTED final sections:\n"); + //for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + // fprintf(stderr, "final section %p %s/%s\n", (*it), (*it)->segmentName(), (*it)->sectionName()); + //} + qsort(§ions[0], sections.size(), sizeof(FinalSection*), &InternalState::FinalSection::sectionComparer); + //fprintf(stderr, "SORTED final sections:\n"); + //for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + // fprintf(stderr, "final section %p %s/%s\n", (*it), (*it)->segmentName(), (*it)->sectionName()); + //} + assert((sections[0]->type() == ld::Section::typeMachHeader) + || ((sections[0]->type() == ld::Section::typeFirstSection) && (sections[1]->type() == ld::Section::typeMachHeader)) + || ((sections[0]->type() == ld::Section::typePageZero) && (sections[1]->type() == ld::Section::typeMachHeader)) + || ((sections[0]->type() == ld::Section::typePageZero) && (sections[1]->type() == ld::Section::typeFirstSection) && (sections[2]->type() == ld::Section::typeMachHeader)) ); + } -class InSet -{ -public: - InSet(std::set& deadAtoms) : fDeadAtoms(deadAtoms) {} - - bool operator()(ObjectFile::Atom*& atom) const { - return ( fDeadAtoms.count(atom) != 0 ); - } - -private: - std::set& fDeadAtoms; -}; -void Linker::loadAndResolve() +bool InternalState::hasZeroForFileOffset(const ld::Section* sect) { - fStartLoadAndResolveTime = mach_absolute_time(); - if ( fOptions.deadStrip() == Options::kDeadStripOff ) { - // without dead-code-stripping: - // find atoms to resolve all undefines - this->loadUndefines(); - // verify nothing is missing - this->checkUndefines(); - // once all undefines fulfill, then bind all references - this->resolveReferences(); - // remove atoms weak atoms that have been overridden - fAllAtoms.erase(std::remove_if(fAllAtoms.begin(), fAllAtoms.end(), InSet(fDeadAtoms)), fAllAtoms.end()); - } - else { - // with dead code stripping: - // start binding references from roots, - this->deadStripResolve(); - // verify nothing is missing - this->checkUndefines(); + switch ( sect->type() ) { + case ld::Section::typeZeroFill: + case ld::Section::typeTLVZeroFill: + return _options.optimizeZeroFill(); + case ld::Section::typePageZero: + case ld::Section::typeStack: + case ld::Section::typeTentativeDefs: + return true; + default: + break; } + return false; } -void Linker::optimize() +uint64_t InternalState::pageAlign(uint64_t addr) { - // give each reader a chance to do any optimizations - bool didSomething = false; - std::vector newAtoms; - std::vector additionalUndefines; - std::vector newlyDeadAtoms; - for (std::vector::iterator it=fInputFiles.begin(); it != fInputFiles.end(); it++) { - didSomething |= (*it)->optimize(fAllAtoms, newAtoms, additionalUndefines, fDeadAtoms, newlyDeadAtoms, fNextInputOrdinal, - fOutputFile, entryPoint(true), fOptions.llvmOptions(), - fOptions.allGlobalsAreDeadStripRoots(), (int)fOptions.outputKind(), fOptions.verbose(), - fOptions.saveTempFiles(), fOptions.getOutputFilePath(), fOptions.positionIndependentExecutable(), - fOptions.allowTextRelocs()); - } - - // only do next steps if some optimization was actually done - if ( didSomething ) { - // add all newly created atoms to fAllAtoms and update symbol table - this->addAtoms(newAtoms); - - // add dead atoms to dead list and remove from fAllAtoms - for(std::vector::iterator itr = newlyDeadAtoms.begin(); itr != newlyDeadAtoms.end(); ++itr) - markDead(*itr); - fAllAtoms.erase(std::remove_if(fAllAtoms.begin(), fAllAtoms.end(), InSet(fDeadAtoms)), fAllAtoms.end()); + const uint64_t alignment = _options.segmentAlignment(); + return ((addr+alignment-1) & (-alignment)); +} - // Make sure all atoms have a section. Atoms that were not originally in a mach-o file could - // not have their section set until now. - for(std::vector::iterator itr = fAllAtoms.begin(); itr != fAllAtoms.end(); ++itr) { - ObjectFile::Atom *atom = *itr; - if ( atom->getSection() == NULL ) - atom->setSection(Section::find(atom->getSectionName(), atom->getSegment().getName(), atom->isZeroFill(), true)); - } +uint64_t InternalState::pageAlign(uint64_t addr, uint64_t pageSize) +{ + return ((addr+pageSize-1) & (-pageSize)); +} - // resolve new undefines - for(std::vector::iterator riter = additionalUndefines.begin(); riter != additionalUndefines.end(); ++riter) { - const char *targetName = *riter; - //fprintf(stderr, "LTO additional undefine: %s\n", targetName); - ObjectFile::Atom* target = fGlobalSymbolTable.find(targetName); - if ( target == NULL) { - // mark that this symbol is needed - fGlobalSymbolTable.require(targetName); - // try to find it in some library - this->addJustInTimeAtoms(targetName, true, true, true); +void InternalState::setSectionSizesAndAlignments() +{ + for (std::vector::iterator sit = sections.begin(); sit != sections.end(); ++sit) { + ld::Internal::FinalSection* sect = *sit; + if ( sect->type() == ld::Section::typeAbsoluteSymbols ) { + // absolute symbols need their finalAddress() to their value + for (std::vector::iterator ait = sect->atoms.begin(); ait != sect->atoms.end(); ++ait) { + const ld::Atom* atom = *ait; + (const_cast(atom))->setSectionOffset(atom->objectAddress()); } } - - if ( fOptions.deadStrip() != Options::kDeadStripOff ) { - // LTO may optimize away some atoms, so dead stripping must be redone - fLiveAtoms.clear(); - this->deadStripResolve(); - } else { - // LTO may require new library symbols to be loaded, so redo - this->checkUndefines(); - this->resolveReferences(); - } - } -} - - -void Linker::adjustScope() -{ - // if -exported_symbols_list is used, demoted to hidden, symbols that are not in it - if ( fOptions.hasExportRestrictList() ) { - // The use of an -export file means the previous computation of fHasExternalWeakDefinitions could change - fGlobalSymbolTable.setHasExternalWeakDefinitions(false); - for(std::vector::iterator itr = fAllAtoms.begin(); itr != fAllAtoms.end(); ++itr) { - ObjectFile::Atom *atom = *itr; - ObjectFile::Atom::Scope scope = atom->getScope(); - const char* name = atom->getName(); - if ( name != NULL ) { - if ( scope == ObjectFile::Atom::scopeGlobal ) { - // check for globals that are downgraded to hidden - if ( !fOptions.shouldExport(name) ) { - atom->setScope(ObjectFile::Atom::scopeLinkageUnit); - //fprintf(stderr, "demote %s to hidden\n", name); - } - else if ( atom->getDefinitionKind() == ObjectFile::Atom::kWeakDefinition ) { - // we do have an exported weak symbol, turn WEAK_DEFINES back on - fGlobalSymbolTable.setHasExternalWeakDefinitions(true); + uint16_t maxAlignment = 0; + uint64_t offset = 0; + for (std::vector::iterator ait = sect->atoms.begin(); ait != sect->atoms.end(); ++ait) { + const ld::Atom* atom = *ait; + bool pagePerAtom = false; + uint32_t atomAlignmentPowerOf2 = atom->alignment().powerOf2; + uint32_t atomModulus = atom->alignment().modulus; + if ( _options.pageAlignDataAtoms() && ( strncmp(atom->section().segmentName(), "__DATA", 6) == 0) ) { + // most objc sections cannot be padded + bool contiguousObjCSection = ( strncmp(atom->section().sectionName(), "__objc_", 7) == 0 ); + if ( strcmp(atom->section().sectionName(), "__objc_const") == 0 ) + contiguousObjCSection = false; + if ( strcmp(atom->section().sectionName(), "__objc_data") == 0 ) + contiguousObjCSection = false; + switch ( atom->section().type() ) { + case ld::Section::typeUnclassified: + case ld::Section::typeTentativeDefs: + case ld::Section::typeZeroFill: + if ( contiguousObjCSection ) + break; + pagePerAtom = true; + if ( atomAlignmentPowerOf2 < 12 ) { + atomAlignmentPowerOf2 = 12; + atomModulus = 0; + } + break; + default: + break; } } - else if ( scope == ObjectFile::Atom::scopeLinkageUnit ) { - // check for hiddens that were requested to be exported - if ( fOptions.hasExportMaskList() && fOptions.shouldExport(name) ) { - warning("cannot export hidden symbol %s from %s", name, atom->getFile()->getPath()); + if ( atomAlignmentPowerOf2 > maxAlignment ) + maxAlignment = atomAlignmentPowerOf2; + // calculate section offset for this atom + uint64_t alignment = 1 << atomAlignmentPowerOf2; + uint64_t currentModulus = (offset % alignment); + uint64_t requiredModulus = atomModulus; + if ( currentModulus != requiredModulus ) { + if ( requiredModulus > currentModulus ) + offset += requiredModulus-currentModulus; + else + offset += requiredModulus+alignment-currentModulus; + } + // LINKEDIT atoms are laid out later + if ( sect->type() != ld::Section::typeLinkEdit ) { + (const_cast(atom))->setSectionOffset(offset); + offset += atom->size(); + if ( pagePerAtom ) { + offset = (offset + 4095) & (-4096); // round up to end of page } } + if ( (atom->scope() == ld::Atom::scopeGlobal) + && (atom->definition() == ld::Atom::definitionRegular) + && (atom->combine() == ld::Atom::combineByName) + && ((atom->symbolTableInclusion() == ld::Atom::symbolTableIn) + || (atom->symbolTableInclusion() == ld::Atom::symbolTableInAndNeverStrip)) ) { + this->hasWeakExternalSymbols = true; + if ( _options.warnWeakExports() ) + warning("weak external symbol: %s", atom->name()); + } } + sect->size = offset; + // section alignment is that of a contained atom with the greatest alignment + sect->alignment = maxAlignment; + // unless -sectalign command line option overrides + if ( _options.hasCustomSectionAlignment(sect->segmentName(), sect->sectionName()) ) + sect->alignment = _options.customSectionAlignment(sect->segmentName(), sect->sectionName()); + // each atom in __eh_frame has zero alignment to assure they pack together, + // but compilers usually make the CFIs pointer sized, so we want whole section + // to start on pointer sized boundary. + if ( sect->type() == ld::Section::typeCFI ) + sect->alignment = 3; + if ( sect->type() == ld::Section::typeTLVDefs ) + this->hasThreadLocalVariableDefinitions = true; } } - - // linking is done, so demote hidden symbols to static - if ( (fOptions.outputKind() == Options::kObjectFile) && fOptions.keepPrivateExterns() ) { - // ld -r -keep_private_externs does not move hidden symbols to static - } - else { - for(std::vector::iterator itr = fAllAtoms.begin(); itr != fAllAtoms.end(); ++itr) { - ObjectFile::Atom *atom = *itr; - // hidden common symbols cannot be demoted to static - if ( (atom->getScope() == ObjectFile::Atom::scopeLinkageUnit) && (atom->getDefinitionKind() != ObjectFile::Atom::kTentativeDefinition) ) { - atom->setScope(ObjectFile::Atom::scopeTranslationUnit); - //fprintf(stderr, "demote %s to static\n", atom->getDisplayName()); - } - } - } -} - -void Linker::link() -{ - this->buildAtomList(); - this->loadAndResolve(); - this->optimize(); - this->adjustScope(); - this->checkObjC(); - this->processDTrace(); - this->tweakLayout(); - this->sortSections(); - this->sortAtoms(); - this->writeDotOutput(); - this->collectDebugInfo(); - this->writeOutput(); - this->printStatistics(); - - if ( fOptions.pauseAtEnd() ) - sleep(10); -} - -void Linker::printTime(const char* msg, uint64_t partTime, uint64_t totalTime) -{ - static uint64_t sUnitsPerSecond = 0; - if ( sUnitsPerSecond == 0 ) { - struct mach_timebase_info timeBaseInfo; - if ( mach_timebase_info(&timeBaseInfo) == KERN_SUCCESS ) { - sUnitsPerSecond = 1000000000ULL * timeBaseInfo.denom / timeBaseInfo.numer; - //fprintf(stderr, "sUnitsPerSecond=%llu\n", sUnitsPerSecond); - } - } - if ( partTime < sUnitsPerSecond ) { - uint32_t milliSecondsTimeTen = (partTime*10000)/sUnitsPerSecond; - uint32_t milliSeconds = milliSecondsTimeTen/10; - uint32_t percentTimesTen = (partTime*1000)/totalTime; - uint32_t percent = percentTimesTen/10; - fprintf(stderr, "%s: %u.%u milliseconds (%u.%u%%)\n", msg, milliSeconds, milliSecondsTimeTen-milliSeconds*10, percent, percentTimesTen-percent*10); - } - else { - uint32_t secondsTimeTen = (partTime*10)/sUnitsPerSecond; - uint32_t seconds = secondsTimeTen/10; - uint32_t percentTimesTen = (partTime*1000)/totalTime; - uint32_t percent = percentTimesTen/10; - fprintf(stderr, "%s: %u.%u seconds (%u.%u%%)\n", msg, seconds, secondsTimeTen-seconds*10, percent, percentTimesTen-percent*10); - } -} - -char* Linker::commatize(uint64_t in, char* out) -{ - char* result = out; - char rawNum[30]; - sprintf(rawNum, "%llu", in); - const int rawNumLen = strlen(rawNum); - for(int i=0; i < rawNumLen-1; ++i) { - *out++ = rawNum[i]; - if ( ((rawNumLen-i) % 3) == 1 ) - *out++ = ','; - } - *out++ = rawNum[rawNumLen-1]; - *out = '\0'; - return result; -} - -void Linker::getVMInfo(vm_statistics_data_t& info) -{ - mach_msg_type_number_t count = sizeof(vm_statistics_data_t) / sizeof(natural_t); - kern_return_t error = host_statistics(mach_host_self(), HOST_VM_INFO, - (host_info_t)&info, &count); - if (error != KERN_SUCCESS) { - bzero(&info, sizeof(vm_statistics_data_t)); - } -} - -void Linker::printStatistics() -{ - fEndTime = mach_absolute_time(); - if ( fOptions.printStatistics() ) { - vm_statistics_data_t endVMInfo; - getVMInfo(endVMInfo); - - uint64_t totalTime = fEndTime - fStartTime; - printTime("ld total time", totalTime, totalTime); - printTime(" option parsing time", fStartCreateReadersTime - fStartTime, totalTime); - printTime(" object file processing",fStartCreateWriterTime - fStartCreateReadersTime, totalTime); - printTime(" output file setup", fStartBuildAtomsTime - fStartCreateWriterTime, totalTime); - printTime(" build atom list", fStartLoadAndResolveTime - fStartBuildAtomsTime, totalTime); - printTime(" resolve references", fStartSortTime - fStartLoadAndResolveTime, totalTime); - printTime(" sort output", fStartDebugTime - fStartSortTime, totalTime); - printTime(" process debug info", fStartWriteTime - fStartDebugTime, totalTime); - printTime(" write output", fEndTime - fStartWriteTime, totalTime); - fprintf(stderr, "pageins=%u, pageouts=%u, faults=%u\n", endVMInfo.pageins-fStartVMInfo.pageins, - endVMInfo.pageouts-fStartVMInfo.pageouts, endVMInfo.faults-fStartVMInfo.faults); - char temp[40]; - fprintf(stderr, "processed %3u object files, totaling %15s bytes\n", fTotalObjectLoaded, commatize(fTotalObjectSize, temp)); - fprintf(stderr, "processed %3u archive files, totaling %15s bytes\n", fTotalArchivesLoaded, commatize(fTotalArchiveSize, temp)); - fprintf(stderr, "processed %3u dylib files\n", fTotalDylibsLoaded); - fprintf(stderr, "wrote output file totaling %15s bytes\n", commatize(fOutputFileSize, temp)); - } -} - -inline void Linker::addAtom(ObjectFile::Atom& atom) -{ - // add to list of all atoms - fAllAtoms.push_back(&atom); - - if ( fOptions.deadStrip() == Options::kDeadStripOff ) { - // not dead-stripping code, so add atom's references's names to symbol table as to-be-resolved-later - std::vector& references = atom.getReferences(); - for (std::vector::iterator it=references.begin(); it != references.end(); it++) { - ObjectFile::Reference* reference = *it; - if ( reference->getTargetBinding() == ObjectFile::Reference::kUnboundByName ) - fGlobalSymbolTable.require(reference->getTargetName()); - if ( reference->getFromTargetBinding() == ObjectFile::Reference::kUnboundByName ) - fGlobalSymbolTable.require(reference->getFromTargetName()); - } - // update total size info (except for __ZEROPAGE atom) - if ( atom.getSegment().isContentReadable() ) { - fTotalSize += atom.getSize(); - if ( atom.isZeroFill() ) - fTotalZeroFillSize += atom.getSize(); - } - } - else { - if ( atom.dontDeadStrip() ) - fLiveRootAtoms.insert(&atom); - } - - // if in global namespace, add atom itself to symbol table - ObjectFile::Atom::Scope scope = atom.getScope(); - const char* name = atom.getName(); - if ( (scope != ObjectFile::Atom::scopeTranslationUnit) && (name != NULL) ) { - // add to symbol table - fGlobalSymbolTable.add(atom); - } - - // record section orders so output file can have same order - if (atom.getSectionName()) { - bool untrusted = false; - switch ( atom.getContentType() ) { - case ObjectFile::Atom::kSectionStart: - case ObjectFile::Atom::kSectionEnd: - untrusted = true; - default: - break; - } - atom.setSection(Section::find(atom.getSectionName(), atom.getSegment().getName(), atom.isZeroFill(), untrusted)); - } -} - - -void Linker::markDead(ObjectFile::Atom* atom) -{ - //fprintf(stderr, "markDead(%p) %s from %s\n", atom, atom->getDisplayName(), atom->getFile()->getPath()); - fDeadAtoms.insert(atom); - - // -dead_strip inhibits weak coalescing in no_dead_strip section - if ( fLiveRootAtoms.count(atom) != 0 ) { - fLiveRootAtoms.erase(atom); - } - - // - // The kGroupSubordinate reference kind is used to model group comdat. - // The "signature" atom in the group has a kGroupSubordinate reference to - // all other members of the group. So, if the signature atom is - // coalesced away, all other atoms in the group should also be removed. - // - std::vector& references = atom->getReferences(); - for (std::vector::iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Reference* ref = *rit; - if ( ref->getKind() == 2 /*kGroupSubordinate*/ ) { // FIX FIX - ObjectFile::Atom* targetAtom = &(ref->getTarget()); - //fprintf(stderr, " markDead(%p) subordinate %s\n", targetAtom, targetAtom->getDisplayName()); - if ( targetAtom == NULL ) { - warning("%s has a group reference to %s but is not bound", atom->getDisplayName(), ref->getTargetName()); - } - else { - if ( targetAtom->getScope() != ObjectFile::Atom::scopeTranslationUnit ) { - // ok for .eh symbols to be not static in -r mode - if ( (fOptions.outputKind() != Options::kObjectFile) || (strcmp(targetAtom->getSectionName(), "__eh_frame") != 0) ) - warning("%s is in a comdat group but its scope is not static", targetAtom->getDisplayName()); - } - this->markDead(targetAtom); - } - } - } -} - -void Linker::updateConstraints(ObjectFile::Reader* reader) -{ - // check objc objects were compiled compatibly - ObjectFile::Reader::ObjcConstraint objcAddition = reader->getObjCConstraint(); - if ( reader->getInstallPath() == NULL ) { - // adding a .o file - switch ( objcAddition ) { - case ObjectFile::Reader::kObjcNone: - break; - case ObjectFile::Reader::kObjcRetainRelease: - if ( fCurrentObjCConstraint == ObjectFile::Reader::kObjcGC ) - throwf("%s built with incompatible Garbage Collection settings to link with previous .o files", reader->getPath()); - fCurrentObjCConstraint = ObjectFile::Reader::kObjcRetainRelease; - break; - case ObjectFile::Reader::kObjcRetainReleaseOrGC: - if ( fCurrentObjCConstraint == ObjectFile::Reader::kObjcNone ) - fCurrentObjCConstraint = ObjectFile::Reader::kObjcRetainReleaseOrGC; - break; - case ObjectFile::Reader::kObjcGC: - if ( fCurrentObjCConstraint == ObjectFile::Reader::kObjcRetainRelease ) - throwf("%s built with incompatible Garbage Collection settings to link with previous .o files", reader->getPath()); - fCurrentObjCConstraint = ObjectFile::Reader::kObjcGC; - break; - } - } - if ( reader->objcReplacementClasses() ) - fObjcReplacmentClasses = true; - - // check cpu sub-types for stricter sub-type - fCurrentCpuConstraint = (ObjectFile::Reader::CpuConstraint)reader->updateCpuConstraint(fCurrentCpuConstraint); -} - -inline void Linker::addAtoms(std::vector& atoms) -{ - bool scanAll = fOptions.readerOptions().fFullyLoadArchives || fOptions.readerOptions().fLoadAllObjcObjectsFromArchives; - bool first = true; - for (std::vector::iterator it=atoms.begin(); it != atoms.end(); it++) { - // usually we only need to get the first atom's reader, but - // with -all_load all atoms from all .o files come come back together - // so we need to scan all atoms - if ( first || scanAll ) { - // update fReadersThatHaveSuppliedAtoms - ObjectFile::Reader* reader = (*it)->getFile(); - if ( std::find(fReadersThatHaveSuppliedAtoms.begin(), fReadersThatHaveSuppliedAtoms.end(), reader) - == fReadersThatHaveSuppliedAtoms.end() ) { - fReadersThatHaveSuppliedAtoms.push_back(reader); - updateConstraints(reader); - } - } - this->addAtom(**it); - first = false; - } -} - -void Linker::logArchive(ObjectFile::Reader* reader) -{ - if ( (fArchiveReaders.count(reader) != 0) && (fArchiveReadersLogged.count(reader) == 0) ) { - fArchiveReadersLogged.insert(reader); - const char* fullPath = reader->getPath(); - char realName[MAXPATHLEN]; - if ( realpath(fullPath, realName) != NULL ) - fullPath = realName; - logTraceInfo("[Logging for XBS] Used static archive: %s\n", fullPath); - } -} - - -void Linker::buildAtomList() -{ - fStartBuildAtomsTime = mach_absolute_time(); - // add initial undefines from -u option - std::vector& initialUndefines = fOptions.initialUndefines(); - for (std::vector::iterator it=initialUndefines.begin(); it != initialUndefines.end(); it++) { - fGlobalSymbolTable.require(*it); - } - - // writer can contribute atoms - this->addAtoms(fOutputFile->getAtoms()); - - // each reader contributes atoms - for (std::vector::iterator it=fInputFiles.begin(); it != fInputFiles.end(); it++) { - ObjectFile::Reader* reader = *it; - std::vector& atoms = reader->getAtoms(); - this->addAtoms(atoms); - if ( fOptions.readerOptions().fTraceArchives && (atoms.size() != 0) ) - logArchive(reader); - } - - // extra command line section always at end - std::vector& extraSections = fOptions.extraSections(); - for( std::vector::iterator it=extraSections.begin(); it != extraSections.end(); ++it) { - this->addAtoms((new opaque_section::Reader(it->segmentName, it->sectionName, it->path, it->data, it->dataLen, fNextInputOrdinal))->getAtoms()); - fNextInputOrdinal += it->dataLen; - } - - // done with all .o files on command line - // everything loaded from now on is a just-in-time atom - fInitialLoadsDone = true; -} - -static const char* pathLeafName(const char* path) -{ - const char* shortPath = strrchr(path, '/'); - if ( shortPath == NULL ) - return path; - else - return &shortPath[1]; -} - - -void Linker::loadUndefines() -{ - // keep looping until no more undefines were added in last loop - unsigned int undefineCount = 0xFFFFFFFF; - while ( undefineCount != fGlobalSymbolTable.getRequireCount() ) { - undefineCount = fGlobalSymbolTable.getRequireCount(); - std::vector undefineNames; - fGlobalSymbolTable.getUndefinesNames(undefineNames); - for(std::vector::iterator it = undefineNames.begin(); it != undefineNames.end(); ++it) { - // load for previous undefine may also have loaded this undefine, so check again - if ( fGlobalSymbolTable.find(*it) == NULL ) { - std::vector* atoms = this->addJustInTimeAtoms(*it, true, true, true); - if ( atoms != NULL ) - delete atoms; - } - } - // need to search archives for overrides of common symbols - if ( fGlobalSymbolTable.hasExternalTentativeDefinitions() ) { - bool searchDylibs = (fOptions.commonsMode() == Options::kCommonsOverriddenByDylibs); - std::vector tentativeDefinitionNames; - fGlobalSymbolTable.getTentativesNames(tentativeDefinitionNames); - for(std::vector::iterator it = tentativeDefinitionNames.begin(); it != tentativeDefinitionNames.end(); ++it) { - // load for previous tentative may also have overridden this tentative, so check again - ObjectFile::Atom* tent = fGlobalSymbolTable.find(*it); - if ( (tent != NULL) && (tent->getDefinitionKind() == ObjectFile::Atom::kTentativeDefinition) ) { - std::vector* atoms = this->addJustInTimeAtoms(*it, searchDylibs, true, false); - if ( atoms != NULL ) - delete atoms; - } - } - } - } -} - -// temp hack for rdar://problem/4718189 map ObjC class names to new runtime names -class ExportedObjcClass -{ -public: - ExportedObjcClass(Options& opt) : fOptions(opt) {} - - bool operator()(const char* name) const { - if ( fOptions.shouldExport(name) ) { - if ( strncmp(name, ".objc_class_name_", 17) == 0 ) - return true; - if ( strncmp(name, "_OBJC_CLASS_$_", 14) == 0 ) - return true; - if ( strncmp(name, "_OBJC_METACLASS_$_", 18) == 0 ) - return true; - } - //fprintf(stderr, "%s is not exported\n", name); - return false; - } -private: - Options& fOptions; -}; - - -void Linker::checkUndefines() -{ - // error out on any remaining undefines - bool doPrint = true; - bool doError = true; - switch ( fOptions.undefinedTreatment() ) { - case Options::kUndefinedError: - break; - case Options::kUndefinedDynamicLookup: - doError = false; - break; - case Options::kUndefinedWarning: - doError = false; - break; - case Options::kUndefinedSuppress: - doError = false; - doPrint = false; - break; - } - std::vector unresolvableUndefines; - fGlobalSymbolTable.getUndefinesNames(unresolvableUndefines); - - // temp hack for rdar://problem/4718189 map ObjC class names to new runtime names - // ignore unresolved references to Objc class names that are listed in -exported_symbols_list - if ( fOptions.hasExportRestrictList() ) - unresolvableUndefines.erase(std::remove_if(unresolvableUndefines.begin(), unresolvableUndefines.end(), ExportedObjcClass(fOptions)), unresolvableUndefines.end()); - - const int unresolvableCount = unresolvableUndefines.size(); - int unresolvableExportsCount = 0; - if ( unresolvableCount != 0 ) { - if ( doPrint ) { - if ( fOptions.printArchPrefix() ) - fprintf(stderr, "Undefined symbols for architecture %s:\n", fArchitectureName); - else - fprintf(stderr, "Undefined symbols:\n"); - for (int i=0; i < unresolvableCount; ++i) { - const char* name = unresolvableUndefines[i]; - fprintf(stderr, " \"%s\", referenced from:\n", name); - // scan all atoms for references - bool foundAtomReference = false; - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* atom = *it; - std::vector& references = atom->getReferences(); - for (std::vector::iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Reference* reference = *rit; - if ( reference->getTargetBinding() == ObjectFile::Reference::kUnboundByName ) { - if ( strcmp(reference->getTargetName(), name) == 0 ) { - fprintf(stderr, " %s in %s\n", atom->getDisplayName(), pathLeafName(atom->getFile()->getPath())); - foundAtomReference = true; - } - } - if ( reference->getFromTargetBinding() == ObjectFile::Reference::kUnboundByName ) { - if ( strcmp(reference->getFromTargetName(), name) == 0 ) { - fprintf(stderr, " %s in %s\n", atom->getDisplayName(), pathLeafName(atom->getFile()->getPath())); - foundAtomReference = true; - } - } - } - } - // scan command line options - if ( !foundAtomReference && fOptions.hasExportRestrictList() && fOptions.shouldExport(name) ) { - fprintf(stderr, " -exported_symbols_list command line option\n"); - ++unresolvableExportsCount; - } - } - } - if ( doError ) - throw "symbol(s) not found"; - } - - // for each tentative definition in symbol table look for dylib that exports same symbol name - if ( fGlobalSymbolTable.hasExternalTentativeDefinitions() ) { - for (SymbolTable::Mapper::iterator it=fGlobalSymbolTable.begin(); it != fGlobalSymbolTable.end(); ++it) { - ObjectFile::Atom* atom = it->second; - if ( (atom != NULL) && (atom->getDefinitionKind()==ObjectFile::Atom::kTentativeDefinition) - && (atom->getScope() == ObjectFile::Atom::scopeGlobal) ) { - // look for dylibs that export same name as used by global tentative definition - addJustInTimeAtoms(atom->getName(), true, false, false); - } - } - } - - - // record any overrides of weak symbols any linked dylib - for (SymbolTable::Mapper::iterator it=fGlobalSymbolTable.begin(); it != fGlobalSymbolTable.end(); ++it) { - ObjectFile::Atom* atom = it->second; - if ( (atom != NULL) && (atom->getDefinitionKind()==ObjectFile::Atom::kRegularDefinition) - && (atom->getScope() == ObjectFile::Atom::scopeGlobal) ) { - const char* name = atom->getName(); - //fprintf(stderr, "looking for dylibs with a weak %s\n", name); - // look for dylibs with weak exports of the same name - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - ObjectFile::Reader* reader = it->second; - if ( reader->hasWeakExternals() ) { - std::vector* dylibAtoms = reader->getJustInTimeAtomsFor(name); - if ( dylibAtoms != NULL ) { - //fprintf(stderr, "addJustInTimeAtoms(%s) => found in file %s\n", name, reader->getPath() ); - // if this is a weak definition in a dylib - if ( (dylibAtoms->at(0)->getDefinitionKind() == ObjectFile::Atom::kExternalWeakDefinition) ) { - fRegularDefAtomsThatOverrideADylibsWeakDef.insert(atom); - } - } - } - } - } - } - -} - - - -std::vector* Linker::addJustInTimeAtoms(const char* name, bool searchDylibs, bool searchArchives, bool okToMakeProxy) -{ - //fprintf(stderr, "addJustInTimeAtoms(%s, searchDylibs=%d, searchArchives=%d)\n", name, searchDylibs, searchArchives ); - // when creating final linked image, writer gets first chance - if ( fOptions.outputKind() != Options::kObjectFile ) { - std::vector* atoms = fOutputFile->getJustInTimeAtomsFor(name); - if ( atoms != NULL ) { - this->addAtoms(*atoms); - //fprintf(stderr, "addJustInTimeAtoms(%s) => found in file %s\n", name, fOutputFile->getPath() ); - return atoms; // found a definition, no need to search anymore - } - } - - // give readers a chance - for (std::vector::iterator it=fInputFiles.begin(); it != fInputFiles.end(); it++) { - ObjectFile::Reader* reader = *it; - if ( reader != NULL ) { - // if this reader is a static archive that has the symbol we need, pull in all atoms in that module - // if this reader is a dylib that exports the symbol we need, have it synthesize an atom for us. - //fprintf(stderr, "addJustInTimeAtoms(%s), looking in reader %s\n", name, reader->getPath() ); - bool isDylibReader = (reader->getInstallPath() != NULL); - if ( isDylibReader ? searchDylibs : searchArchives ) { - std::vector* atoms = reader->getJustInTimeAtomsFor(name); - if ( atoms != NULL ) { - this->addAtoms(*atoms); - //fprintf(stderr, "addJustInTimeAtoms(%s) => found in file %s\n", name, reader->getPath() ); - if ( !isDylibReader && fOptions.readerOptions().fTraceArchives ) { - logArchive(reader); - } - // if this is a weak definition in a dylib - if ( isDylibReader && (atoms->size() == 1) && (atoms->at(0)->getDefinitionKind() == ObjectFile::Atom::kExternalWeakDefinition) ) { - // keep looking for a non-weak definition - } - else { - // found a definition, no need to search anymore - return atoms; - } - } - } - } - } - - // for two level namesapce, give all implicitly link dylibs a chance - if ( fOptions.nameSpace() == Options::kTwoLevelNameSpace ) { - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - if ( it->second->implicitlyLinked() ) { - //fprintf(stderr, "addJustInTimeAtoms(%s), looking in implicitly linked %s\n", name, it->second->getPath() ); - std::vector* atoms = it->second->getJustInTimeAtomsFor(name); - if ( atoms != NULL ) { - this->addAtoms(*atoms); - //fprintf(stderr, "addJustInTimeAtoms(%s) => found in file %s\n", name, reader->getPath() ); - // if this is a weak definition in a dylib - if ( (atoms->size() == 1) && (atoms->at(0)->getDefinitionKind() == ObjectFile::Atom::kExternalWeakDefinition) ) { - // keep looking for a non-weak definition - } - else { - // found a definition, no need to search anymore - return atoms; - } - } - } - } - } - - // for flat namespace, give indirect dylibs - if ( fOptions.nameSpace() != Options::kTwoLevelNameSpace ) { - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - if ( ! it->second->explicitlyLinked() ) { - std::vector* atoms = it->second->getJustInTimeAtomsFor(name); - if ( atoms != NULL ) { - this->addAtoms(*atoms); - //fprintf(stderr, "addJustInTimeAtoms(%s) => found in file %s\n", name, reader->getPath() ); - return atoms; // found a definition, no need to search anymore - } - } - } - } - - // writer creates a proxy in two cases: - // 1) ld -r is being used to create a .o file - // 2) -undefined dynamic_lookup is being used - // 3) -U _foo is being used - // 4) x86_64 kext bundle is being created - if ( (fOptions.outputKind() == Options::kObjectFile) - || ((fOptions.undefinedTreatment() != Options::kUndefinedError) && okToMakeProxy) - || (fOptions.someAllowedUndefines() && okToMakeProxy) - || (fOptions.outputKind() == Options::kKextBundle) ) { - ObjectFile::Atom* atom = fOutputFile->getUndefinedProxyAtom(name); - if ( atom != NULL ) { - this->addAtom(*atom); - return NULL; - } - } - //fprintf(stderr, "addJustInTimeAtoms(%s) => not found\n", name); - return NULL; -} - -void Linker::resolve(ObjectFile::Reference* reference) -{ - // look in global symbol table - const char* targetName = reference->getTargetName(); - ObjectFile::Atom* target = fGlobalSymbolTable.find(targetName); - if ( target == NULL ) { - throwf("unexpected undefined symbol: %s", targetName); - } - reference->setTarget(*target, reference->getTargetOffset()); -} - -void Linker::resolveFrom(ObjectFile::Reference* reference) -{ - // handle references that have two (from and to) targets - const char* fromTargetName = reference->getFromTargetName(); - ObjectFile::Atom* fromTarget = fGlobalSymbolTable.find(fromTargetName); - if ( fromTarget == NULL ) { - throwf("unexpected undefined symbol: %s", fromTargetName); - } - reference->setFromTarget(*fromTarget); -} - - -void Linker::resolveReferences() -{ - // note: the atom list may grow during this loop as libraries supply needed atoms - for (unsigned int j=0; j < fAllAtoms.size(); ++j) { - ObjectFile::Atom* atom = fAllAtoms[j]; - std::vector& references = atom->getReferences(); - for (std::vector::iterator it=references.begin(); it != references.end(); it++) { - ObjectFile::Reference* reference = *it; - if ( reference->getTargetBinding() == ObjectFile::Reference::kUnboundByName ) - this->resolve(reference); - if ( reference->getFromTargetBinding() == ObjectFile::Reference::kUnboundByName ) - this->resolveFrom(reference); - } - } -} - - -// used to remove stabs associated with atoms that won't be in output file -class NotInSet -{ -public: - NotInSet(std::set& theSet) : fSet(theSet) {} - - bool operator()(const ObjectFile::Reader::Stab& stab) const { - if ( stab.atom == NULL ) - return false; // leave stabs that are not associated with any atome - else - return ( fSet.count(stab.atom) == 0 ); - } - -private: - std::set& fSet; -}; - - -class NotLive -{ -public: - NotLive(std::set& set) : fLiveAtoms(set) {} - - bool operator()(ObjectFile::Atom*& atom) const { - //if ( fLiveAtoms.count(atom) == 0 ) - // fprintf(stderr, "dead strip %s\n", atom->getDisplayName()); - return ( fLiveAtoms.count(atom) == 0 ); - } -private: - std::set& fLiveAtoms; -}; - - -void Linker::addJustInTimeAtomsAndMarkLive(const char* name) -{ - //fprintf(stderr, "addJustInTimeAtomsAndMarkLive(%s)\n", name); - std::vector* atoms = this->addJustInTimeAtoms(name, true, true, true); - if ( atoms != NULL ) { - if ( fOptions.allGlobalsAreDeadStripRoots() ) { - for (std::vector::iterator it=atoms->begin(); it != atoms->end(); it++) { - ObjectFile::Atom* atom = *it; - if ( atom->getScope() == ObjectFile::Atom::scopeGlobal ) { - WhyLiveBackChain rootChain; - rootChain.previous = NULL; - rootChain.referer = atom; - this->markLive(*atom, &rootChain); - } - } - } - delete atoms; - } -} - -void Linker::markLive(ObjectFile::Atom& atom, struct Linker::WhyLiveBackChain* previous) -{ - //fprintf(stderr, "markLive(%p)\n", &atom); - if ( fLiveAtoms.count(&atom) == 0 ) { - // if -why_live cares about this symbol, then dump chain - if ( (previous->referer != NULL) && fOptions.printWhyLive(previous->referer->getDisplayName()) ) { - int depth = 0; - for(WhyLiveBackChain* p = previous; p != NULL; p = p->previous, ++depth) { - for(int i=depth; i > 0; --i) - fprintf(stderr, " "); - fprintf(stderr, "%p %s from %s\n", p->referer, p->referer->getDisplayName(), p->referer->getFile()->getPath()); - } - } - // set up next chain - WhyLiveBackChain thisChain; - thisChain.previous = previous; - // this atom is live - fLiveAtoms.insert(&atom); - // update total size info (except for __ZEROPAGE atom) - if ( atom.getSegment().isContentReadable() ) { - fTotalSize += atom.getSize(); - if ( atom.isZeroFill() ) - fTotalZeroFillSize += atom.getSize(); - } - // and all atoms it references - std::vector& references = atom.getReferences(); - for (std::vector::iterator it=references.begin(); it != references.end(); it++) { - ObjectFile::Reference* reference = *it; - if ( reference->getTargetBinding() == ObjectFile::Reference::kUnboundByName ) { - // look in global symbol table - const char* targetName = reference->getTargetName(); - ObjectFile::Atom* target = fGlobalSymbolTable.find(targetName); - if ( (target == NULL) || (target->getDefinitionKind() == ObjectFile::Atom::kTentativeDefinition) ) { - // load archives or dylibs - this->addJustInTimeAtomsAndMarkLive(targetName); - } - // look again - target = fGlobalSymbolTable.find(targetName); - if ( target != NULL ) { - reference->setTarget(*target, reference->getTargetOffset()); - } - else { - // mark as undefined, for later error processing - fAtomsWithUnresolvedReferences.push_back(&atom); - fGlobalSymbolTable.require(targetName); - } - } - switch ( reference->getTargetBinding() ) { - case ObjectFile::Reference::kBoundDirectly: - case ObjectFile::Reference::kBoundByName: - thisChain.referer = &reference->getTarget(); - markLive(reference->getTarget(), &thisChain); - break; - case ObjectFile::Reference::kDontBind: - case ObjectFile::Reference::kUnboundByName: - // do nothing - break; - } - // do the same as above, for "from target" - if ( reference->getFromTargetBinding() == ObjectFile::Reference::kUnboundByName ) { - // look in global symbol table - const char* targetName = reference->getFromTargetName(); - ObjectFile::Atom* target = fGlobalSymbolTable.find(targetName); - if ( (target == NULL) || (target->getDefinitionKind() == ObjectFile::Atom::kTentativeDefinition) ) { - // load archives or dylibs - this->addJustInTimeAtomsAndMarkLive(targetName); - } - // look again - target = fGlobalSymbolTable.find(targetName); - if ( target != NULL ) { - reference->setFromTarget(*target); - } - else { - // mark as undefined, for later error processing - fGlobalSymbolTable.require(targetName); - } - } - switch ( reference->getFromTargetBinding() ) { - case ObjectFile::Reference::kBoundDirectly: - case ObjectFile::Reference::kBoundByName: - thisChain.referer = &reference->getFromTarget(); - markLive(reference->getFromTarget(), &thisChain); - break; - case ObjectFile::Reference::kUnboundByName: - case ObjectFile::Reference::kDontBind: - // do nothing - break; - } - } - } -} - - -void Linker::addLiveRoot(const char* name) -{ - ObjectFile::Atom* target = fGlobalSymbolTable.find(name); - if ( target == NULL ) { - this->addJustInTimeAtomsAndMarkLive(name); - target = fGlobalSymbolTable.find(name); - } - if ( target != NULL ) - fLiveRootAtoms.insert(target); -} - -void Linker::moveToFrontOfSection(ObjectFile::Atom* atom) -{ - // check if already moved to front - if ( fInitializerAtoms.find(atom) == fInitializerAtoms.end() ) { - // don't re-order initializers from .o files without MH_SUBSECTIONS_VIA_SYMBOLS - // since that could make all atoms in the file look like initializers - if ( atom->getFile()->canScatterAtoms() ) { - //fprintf(stdout, "marking as initializer: %s\n", atom->getDisplayName()); - fInitializerAtoms.insert(atom); - // mark all functions that this function references - std::vector& references = atom->getReferences(); - for (std::vector::const_iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Atom* childAtom = &((*rit)->getTarget()); - if ( childAtom != NULL ) { - if ( (*rit)->isBranch() ) { - this->moveToFrontOfSection(childAtom); - } - else if ( (childAtom->getName() != NULL) && (strncmp(childAtom->getName(), "___tcf_", 7) == 0) ) { - //fprintf(stdout, "marking as terminator: %s\n", childAtom->getDisplayName()); - fTerminatorAtoms.insert(childAtom); - } - } - } - } - } -} - -void Linker::deadStripResolve() -{ - // add main() to live roots - ObjectFile::Atom* entryPoint = this->entryPoint(false); - if ( entryPoint != NULL ) - fLiveRootAtoms.insert(entryPoint); - - // add dyld_stub_binding_helper/dyld_stub_binder to live roots - ObjectFile::Atom* dyldHelper = this->dyldClassicHelper(); - if ( dyldHelper != NULL ) - fLiveRootAtoms.insert(dyldHelper); - dyldHelper = this->dyldCompressedHelper(); - if ( dyldHelper != NULL ) - fLiveRootAtoms.insert(dyldHelper); - - // if using lazy dylib loading, add dyld_lazy_dylib_stub_binding_helper() to live roots - if ( fOptions.usingLazyDylibLinking() ) { - ObjectFile::Atom* dyldLazyDylibHelper = this->dyldLazyLibraryHelper(); - if ( dyldLazyDylibHelper != NULL ) - fLiveRootAtoms.insert(dyldLazyDylibHelper); - } - - // add -exported_symbols_list, -init, and -u entries to live roots - std::vector& initialUndefines = fOptions.initialUndefines(); - for (std::vector::iterator it=initialUndefines.begin(); it != initialUndefines.end(); it++) - addLiveRoot(*it); - - // if -exported_symbols_list that has wildcards, we need to find all matches and make them the roots - // - if ( fOptions.hasWildCardExportRestrictList() ) { - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* atom = *it; - if ( (atom->getScope() == ObjectFile::Atom::scopeGlobal) - && (fDeadAtoms.count(atom) == 0) - && fOptions.shouldExport(atom->getName()) ) - fLiveRootAtoms.insert(atom); - } - } - - // in some cases, every global scope atom in initial .o files is a root - if ( fOptions.allGlobalsAreDeadStripRoots() ) { - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* atom = *it; - if ( (atom->getScope() == ObjectFile::Atom::scopeGlobal) && (fDeadAtoms.count(atom) == 0) ) - fLiveRootAtoms.insert(atom); - } - } - - // mark all roots as live, and all atoms they reference - for (std::set::iterator it=fLiveRootAtoms.begin(); it != fLiveRootAtoms.end(); it++) { - WhyLiveBackChain rootChain; - rootChain.previous = NULL; - rootChain.referer = *it; - markLive(**it, &rootChain); - } - - // it is possible that there are unresolved references that can be resolved now - // this can happen if the first reference to a common symbol in an archive. - // common symbols are not in the archive TOC, but the .o could have been pulled in later. - // ld64 while linking cc1 [ when dead_strip is ON] - for (std::vector::iterator it=fAtomsWithUnresolvedReferences.begin(); it != fAtomsWithUnresolvedReferences.end(); it++) { - std::vector& references = (*it)->getReferences(); - for (std::vector::iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Reference* reference = *rit; - if ( reference->getTargetBinding() == ObjectFile::Reference::kUnboundByName ) { - ObjectFile::Atom* target = fGlobalSymbolTable.find(reference->getTargetName()); - if ( target != NULL ) { - reference->setTarget(*target, reference->getTargetOffset()); - fLiveAtoms.insert(target); - // by just adding this atom to fLiveAtoms set, we are assuming it has no - // references, which is true for commons. - if ( target->getDefinitionKind() != ObjectFile::Atom::kTentativeDefinition ) - warning("internal error %s is not a tentative definition", target->getDisplayName()); - } - } - if ( reference->getFromTargetBinding() == ObjectFile::Reference::kUnboundByName ) { - ObjectFile::Atom* target = fGlobalSymbolTable.find(reference->getFromTargetName()); - if ( target != NULL ) { - reference->setFromTarget(*target); - fLiveAtoms.insert(target); - // by just adding this atom to fLiveAtoms set, we are assuming it has no - // references, which is true for commons. - if ( target->getDefinitionKind() != ObjectFile::Atom::kTentativeDefinition ) - warning("internal error %s is not a tentative definition", target->getDisplayName()); - } - } - } - } - - // It is possible that some weak symbols were overridden by lazily load objects from archives - // and we have some atoms that still refer to the overridden ones. - // In that case we need to go back and rebind - if ( fAtomsOverriddenByLateLoads.size() > 0 ) { - for (std::set::iterator it=fLiveAtoms.begin(); it != fLiveAtoms.end(); ++it) { - ObjectFile::Atom* atom = *it; - std::vector& references = atom->getReferences(); - for (std::vector::iterator rit=references.begin(); rit != references.end(); ++rit) { - ObjectFile::Reference* reference = *rit; - ObjectFile::Atom* toTarget = &reference->getTarget(); - if ( fAtomsOverriddenByLateLoads.count(toTarget) ) { - //fprintf(stderr, "change reference in %p from %p to %p\n", atom, toTarget, fGlobalSymbolTable.find(toTarget->getName())); - reference->setTarget(*fGlobalSymbolTable.find(toTarget->getName()), reference->getTargetOffset()); - } - ObjectFile::Atom* fromTarget = &reference->getFromTarget(); - if ( (fromTarget != NULL) && fAtomsOverriddenByLateLoads.count(fromTarget) ) { - //fprintf(stderr, "change from reference in %p from %p to %p\n", atom, fromTarget, fGlobalSymbolTable.find(fromTarget->getName())); - reference->setTarget(*fGlobalSymbolTable.find(fromTarget->getName()), reference->getFromTargetOffset()); - } - } - } - - // make sure overriders are live if the atom they overrid was live - for (std::set::iterator it=fAtomsOverriddenByLateLoads.begin(); it != fAtomsOverriddenByLateLoads.end(); ++it) { - ObjectFile::Atom* overriderAtom = *it; - if ( fLiveAtoms.count(overriderAtom) ) { - WhyLiveBackChain rootChain; - rootChain.previous = NULL; - rootChain.referer = *it; - markLive(*fGlobalSymbolTable.find(overriderAtom->getName()), &rootChain); - } - } - - // remove overridden atoms from fLiveAtoms - fAllAtoms.erase(std::remove_if(fAllAtoms.begin(), fAllAtoms.end(), InSet(fAtomsOverriddenByLateLoads)), fAllAtoms.end()); - fAtomsOverriddenByLateLoads.clear(); - // remove dead atoms from fLiveAtoms - fAllAtoms.erase(std::remove_if(fAllAtoms.begin(), fAllAtoms.end(), InSet(fDeadAtoms)), fAllAtoms.end()); - } - - // now remove all non-live atoms from fAllAtoms - fAllAtoms.erase(std::remove_if(fAllAtoms.begin(), fAllAtoms.end(), NotLive(fLiveAtoms)), fAllAtoms.end()); -} - -void Linker::checkObjC() -{ - // check dylibs - switch ( fCurrentObjCConstraint ) { - case ObjectFile::Reader::kObjcNone: - // can link against any dylib - break; - case ObjectFile::Reader::kObjcRetainRelease: - // cannot link against GC-only dylibs - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - if ( it->second->explicitlyLinked() ) { - if ( it->second->getObjCConstraint() == ObjectFile::Reader::kObjcGC ) - throwf("this linkage unit uses Retain/Release. It cannot link against the GC-only dylib: %s", it->second->getPath()); - } - } - break; - case ObjectFile::Reader::kObjcRetainReleaseOrGC: - // can link against GC or RR dylibs - break; - case ObjectFile::Reader::kObjcGC: - // cannot link against RR-only dylibs - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - if ( it->second->explicitlyLinked() ) { - if ( it->second->getObjCConstraint() == ObjectFile::Reader::kObjcRetainRelease ) - throwf("this linkage unit requires GC. It cannot link against Retain/Release dylib: %s", it->second->getPath()); - } - } - break; - } - - // synthesize __OBJC __image_info atom if needed - if ( fCurrentObjCConstraint != ObjectFile::Reader::kObjcNone ) { - this->addAtom(fOutputFile->makeObjcInfoAtom(fCurrentObjCConstraint, fObjcReplacmentClasses)); - } -} - - -static uint8_t pcRelKind(cpu_type_t arch) -{ - switch ( arch ) { - case CPU_TYPE_POWERPC: - return ppc::kPointerDiff32; - case CPU_TYPE_POWERPC64: - return ppc64::kPointerDiff32; - case CPU_TYPE_I386: - return x86::kPointerDiff; - case CPU_TYPE_X86_64: - return x86_64::kPointerDiff32; - case CPU_TYPE_ARM: - return arm::kPointerDiff; - } - throw "uknown architecture"; -} - -typedef uint8_t* (*oldcreatedof_func_t) (const char*, cpu_type_t, unsigned int, const char*[], const char*[], uint64_t offsetsInDOF[], size_t* size); -typedef uint8_t* (*createdof_func_t)(cpu_type_t, unsigned int, const char*[], unsigned int, const char*[], const char*[], uint64_t offsetsInDOF[], size_t* size); - - -void Linker::processDTrace() -{ - // only make __dof section in final linked images - if ( fOptions.outputKind() == Options::kObjectFile ) - return; - - // scan all atoms looking for dtrace probes - std::vector probeSites; - std::vector isEnabledSites; - std::map atomToDtraceTypes; - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); ++it) { - ObjectFile::Atom* atom = *it; - std::vector& references = atom->getReferences(); - for (std::vector::iterator rit=references.begin(); rit != references.end(); ++rit) { - ObjectFile::Reference* ref = *rit; - if ( ref->getTargetBinding() == ObjectFile::Reference::kDontBind ) { - const char* probeName = ref->getTargetName(); - if ( probeName != NULL ) { - uint32_t offsetInAtom = ref->getFixUpOffset(); - if ( strncmp(probeName, "___dtrace_probe$", 16) == 0 ) - probeSites.push_back(DTraceProbeInfo(atom, offsetInAtom, probeName)); - else if ( strncmp(probeName, "___dtrace_isenabled$", 20) == 0 ) - isEnabledSites.push_back(DTraceProbeInfo(atom, offsetInAtom, probeName)); - else if ( strncmp(probeName, "___dtrace_", 10) == 0 ) - atomToDtraceTypes[atom].insert(probeName); - } - } - } - } - - // if no probes, we're done - if ( (probeSites.size() == 0) && (isEnabledSites.size() == 0) ) - return; - - // partition probes by provider name - // The symbol names looks like: - // "___dtrace_probe$" provider-name "$" probe-name [ "$"... ] - // "___dtrace_isenabled$" provider-name "$" probe-name [ "$"... ] - ProviderToProbes providerToProbes; - std::vector emptyList; - for(std::vector::iterator it = probeSites.begin(); it != probeSites.end(); ++it) { - // ignore probes in functions that were coalesed away rdar://problem/5628149 - if ( fDeadAtoms.count((ObjectFile::Atom*)(it->atom)) == 0 ) { - const char* providerStart = &it->probeName[16]; - const char* providerEnd = strchr(providerStart, '$'); - if ( providerEnd != NULL ) { - char providerName[providerEnd-providerStart+1]; - strlcpy(providerName, providerStart, providerEnd-providerStart+1); - ProviderToProbes::iterator pos = providerToProbes.find(providerName); - if ( pos == providerToProbes.end() ) { - const char* dup = strdup(providerName); - providerToProbes[dup] = emptyList; - } - providerToProbes[providerName].push_back(*it); - } - } - } - for(std::vector::iterator it = isEnabledSites.begin(); it != isEnabledSites.end(); ++it) { - // ignore probes in functions that were coalesed away rdar://problem/5628149 - if ( fDeadAtoms.count((ObjectFile::Atom*)(it->atom)) == 0 ) { - const char* providerStart = &it->probeName[20]; - const char* providerEnd = strchr(providerStart, '$'); - if ( providerEnd != NULL ) { - char providerName[providerEnd-providerStart+1]; - strlcpy(providerName, providerStart, providerEnd-providerStart+1); - ProviderToProbes::iterator pos = providerToProbes.find(providerName); - if ( pos == providerToProbes.end() ) { - const char* dup = strdup(providerName); - providerToProbes[dup] = emptyList; - } - providerToProbes[providerName].push_back(*it); - } - } - } - - // create a DOF section for each provider - int dofIndex=1; - CStringSet sectionNamesUsed; - for(ProviderToProbes::iterator pit = providerToProbes.begin(); pit != providerToProbes.end(); ++pit, ++dofIndex) { - const char* providerName = pit->first; - const std::vector& probes = pit->second; - - // open library and find dtrace_create_dof() - void* handle = dlopen("/usr/lib/libdtrace.dylib", RTLD_LAZY); - if ( handle == NULL ) - throwf("couldn't dlopen() /usr/lib/libdtrace.dylib: %s", dlerror()); - createdof_func_t pCreateDOF = (createdof_func_t)dlsym(handle, "dtrace_ld_create_dof"); - if ( pCreateDOF == NULL ) - throwf("couldn't find \"dtrace_ld_create_dof\" in /usr/lib/libdtrace.dylib: %s", dlerror()); - // build list of typedefs/stability infos for this provider - CStringSet types; - for(std::vector::const_iterator it = probes.begin(); it != probes.end(); ++it) { - std::map::iterator pos = atomToDtraceTypes.find(it->atom); - if ( pos != atomToDtraceTypes.end() ) { - for(CStringSet::iterator sit = pos->second.begin(); sit != pos->second.end(); ++sit) { - const char* providerStart = strchr(*sit, '$')+1; - const char* providerEnd = strchr(providerStart, '$'); - if ( providerEnd != NULL ) { - char aProviderName[providerEnd-providerStart+1]; - strlcpy(aProviderName, providerStart, providerEnd-providerStart+1); - if ( strcmp(aProviderName, providerName) == 0 ) - types.insert(*sit); - } - } - } - } - int typeCount = types.size(); - const char* typeNames[typeCount]; - //fprintf(stderr, "types for %s:\n", providerName); - uint32_t index = 0; - for(CStringSet::iterator it = types.begin(); it != types.end(); ++it) { - typeNames[index] = *it; - //fprintf(stderr, "\t%s\n", *it); - ++index; - } - - // build list of probe/isenabled sites - const uint32_t probeCount = probes.size(); - const char* probeNames[probeCount]; - const char* funtionNames[probeCount]; - uint64_t offsetsInDOF[probeCount]; - index = 0; - for(std::vector::const_iterator it = probes.begin(); it != probes.end(); ++it) { - probeNames[index] = it->probeName; - funtionNames[index] = it->atom->getName(); - offsetsInDOF[index] = 0; - ++index; - } - //fprintf(stderr, "calling libtrace to create DOF\n"); - //for(uint32_t i=0; i < probeCount; ++i) - // fprintf(stderr, " [%u]\t %s\t%s\n", i, probeNames[i], funtionNames[i]); - // call dtrace library to create DOF section - size_t dofSectionSize; - uint8_t* p = (*pCreateDOF)(fArchitecture, typeCount, typeNames, probeCount, probeNames, funtionNames, offsetsInDOF, &dofSectionSize); - if ( p != NULL ) { - char sectionName[18]; - strcpy(sectionName, "__dof_"); - strlcpy(§ionName[6], providerName, 10); - // create unique section name so each DOF is in its own section - if ( sectionNamesUsed.count(sectionName) != 0 ) { - sectionName[15] = '0'; - sectionName[16] = '\0'; - while ( sectionNamesUsed.count(sectionName) != 0 ) - ++sectionName[15]; - } - sectionNamesUsed.insert(sectionName); - char symbolName[strlen(providerName)+64]; - sprintf(symbolName, "__dtrace_dof_for_provider_%s", providerName); - opaque_section::Reader* reader = new opaque_section::Reader("__TEXT", sectionName, - "dtrace", p, dofSectionSize, fNextInputOrdinal, symbolName); - fNextInputOrdinal += dofSectionSize; - // add references - for (uint32_t i=0; i < probeCount; ++i) { - uint64_t offset = offsetsInDOF[i]; - //fprintf(stderr, "%s offset[%d]=0x%08llX\n", providerName, i, offset); - if ( offset > dofSectionSize ) - throwf("offsetsInDOF[i]=%0llX > dofSectionSize=%0lX\n", i, offset, dofSectionSize); - reader->addSectionReference(pcRelKind(fArchitecture), offset, probes[i].atom, probes[i].offset, reader->getAtoms()[0], 0); - } - this->addAtoms(reader->getAtoms()); - } - else { - throw "error creating dtrace DOF section"; - } - } -} - - -static bool matchesObjectFile(ObjectFile::Atom* atom, const char* objectFileLeafName) -{ - if ( objectFileLeafName == NULL ) - return true; - const char* atomFullPath = atom->getFile()->getPath(); - const char* lastSlash = strrchr(atomFullPath, '/'); - if ( lastSlash != NULL ) { - if ( strcmp(&lastSlash[1], objectFileLeafName) == 0 ) - return true; - } - else { - if ( strcmp(atomFullPath, objectFileLeafName) == 0 ) - return true; - } - return false; -} - - -static bool usesAnonymousNamespace(const char* symbol) -{ - return ( (strncmp(symbol, "__Z", 3) == 0) && (strstr(symbol, "_GLOBAL__N_") != NULL) ); -} - - -// -// convert: -// __ZN20_GLOBAL__N__Z5main2v3barEv => _ZN-3barEv -// __ZN37_GLOBAL__N_main.cxx_00000000_493A01A33barEv => _ZN-3barEv -// -static void canonicalizeAnonymousName(const char* inSymbol, char outSymbol[]) -{ - const char* globPtr = strstr(inSymbol, "_GLOBAL__N_"); - while ( isdigit(*(--globPtr)) ) - ; // loop - char* endptr; - unsigned long length = strtoul(globPtr+1, &endptr, 10); - const char* globEndPtr = endptr + length; - int startLen = globPtr-inSymbol+1; - memcpy(outSymbol, inSymbol, startLen); - outSymbol[startLen] = '-'; - strcpy(&outSymbol[startLen+1], globEndPtr); -} - - -ObjectFile::Atom* Linker::findAtom(const Options::OrderedSymbol& orderedSymbol) -{ - ObjectFile::Atom* atom = fGlobalSymbolTable.find(orderedSymbol.symbolName); - if ( atom != NULL ) { - if ( matchesObjectFile(atom, orderedSymbol.objectFileName) ) - return atom; - } - else { - // slow case. The requested symbol is not in symbol table, so might be static function - static SymbolTable::Mapper hashTableOfTranslationUnitScopedSymbols; - static SymbolTable::Mapper hashTableOfSymbolsWithAnonymousNamespace; - static bool built = false; - // build a hash_map the first time - if ( !built ) { - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - atom = *it; - const char* name = atom->getName(); - if ( name != NULL) { - if ( usesAnonymousNamespace(name) ) { - // symbol that uses anonymous namespace - char canonicalName[strlen(name)+2]; - canonicalizeAnonymousName(name, canonicalName); - const char* hashName = strdup(canonicalName); - SymbolTable::Mapper::iterator pos = hashTableOfSymbolsWithAnonymousNamespace.find(hashName); - if ( pos == hashTableOfSymbolsWithAnonymousNamespace.end() ) - hashTableOfSymbolsWithAnonymousNamespace[hashName] = atom; - else - hashTableOfSymbolsWithAnonymousNamespace[hashName] = NULL; // collision, denote with NULL - } - else if ( atom->getScope() == ObjectFile::Atom::scopeTranslationUnit ) { - // static function or data - SymbolTable::Mapper::iterator pos = hashTableOfTranslationUnitScopedSymbols.find(name); - if ( pos == hashTableOfTranslationUnitScopedSymbols.end() ) - hashTableOfTranslationUnitScopedSymbols[name] = atom; - else - hashTableOfTranslationUnitScopedSymbols[name] = NULL; // collision, denote with NULL - } - } - } - //fprintf(stderr, "built hash table of %lu static functions\n", hashTableOfTranslationUnitScopedSymbols.size()); - built = true; - } - - // look for name in hashTableOfTranslationUnitScopedSymbols - SymbolTable::Mapper::iterator pos = hashTableOfTranslationUnitScopedSymbols.find(orderedSymbol.symbolName); - if ( pos != hashTableOfTranslationUnitScopedSymbols.end() ) { - if ( (pos->second != NULL) && matchesObjectFile(pos->second, orderedSymbol.objectFileName) ) { - //fprintf(stderr, "found %s in hash table\n", orderedSymbol.symbolName); - return pos->second; - } - if ( pos->second == NULL ) - // name is in hash table, but atom is NULL, so that means there are duplicates, so we use super slow way - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - atom = *it; - if ( atom->getScope() == ObjectFile::Atom::scopeTranslationUnit ) { - const char* name = atom->getName(); - if ( (name != NULL) && (strcmp(name, orderedSymbol.symbolName) == 0) ) { - if ( matchesObjectFile(atom, orderedSymbol.objectFileName) ) { - if ( fOptions.printOrderFileStatistics() ) - warning("%s specified in order_file but it exists in multiple .o files. " - "Prefix symbol with .o filename in order_file to disambiguate", orderedSymbol.symbolName); - return atom; - } - } - } - } - } - - // look for name in hashTableOfSymbolsWithAnonymousNamespace - if ( usesAnonymousNamespace(orderedSymbol.symbolName) ) { - // symbol that uses anonymous namespace - char canonicalName[strlen(orderedSymbol.symbolName)+2]; - canonicalizeAnonymousName(orderedSymbol.symbolName, canonicalName); - SymbolTable::Mapper::iterator pos = hashTableOfSymbolsWithAnonymousNamespace.find(canonicalName); - if ( pos != hashTableOfSymbolsWithAnonymousNamespace.end() ) { - if ( (pos->second != NULL) && matchesObjectFile(pos->second, orderedSymbol.objectFileName) ) { - //fprintf(stderr, "found %s in anonymous namespace hash table\n", canonicalName); - return pos->second; - } - if ( pos->second == NULL ) - // name is in hash table, but atom is NULL, so that means there are duplicates, so we use super slow way - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - atom = *it; - const char* name = atom->getName(); - if ( (name != NULL) && usesAnonymousNamespace(name) ) { - char canonicalAtomName[strlen(name)+2]; - canonicalizeAnonymousName(name, canonicalAtomName); - if ( strcmp(canonicalAtomName, canonicalName) == 0 ) { - if ( matchesObjectFile(atom, orderedSymbol.objectFileName) ) { - if ( fOptions.printOrderFileStatistics() ) - warning("%s specified in order_file but it exists in multiple .o files. " - "Prefix symbol with .o filename in order_file to disambiguate", orderedSymbol.symbolName); - return atom; - } - } - } - } - } - } - } - return NULL; -} - - -void Linker::sortSections() -{ - Section::assignIndexes(); -} - - -// -// Linker::sortAtoms() -// -// The purpose of this method is to take the graph of all Atoms and produce an ordered -// sequence of atoms. The constraints are that: 1) all Atoms of the same Segment must -// be contiguous, 2) all Atoms of the same Section must be contigous, 3) Atoms specified -// in an order_file are seqenced as in the order_file and before Atoms not specified, -// 4) Atoms in the same section from the same .o file should be contiguous and sequenced -// in the same order they were in the .o file, 5) Atoms in the same Section but which came -// from different .o files should be sequenced in the same order that the .o files -// were passed to the linker (i.e. command line order). -// -// The way this is implemented is that the linker passes a "base ordinal" to each Reader -// as it is constructed. The reader should construct it Atoms so that calling getOrdinal() -// on its atoms returns a contiguous range of values starting at the base ordinal. Then -// sorting is just sorting by section, then by ordinal. -// -// If an order_file is specified, it gets more complicated. First, an override-ordinal map -// is created. It causes the sort routine to ignore the value returned by getOrdinal() and -// use the override value instead. Next some Atoms must be layed out consecutively -// (e.g. hand written assembly that does not end with return, but rather falls into -// the next label). This is modeled in Readers via a "kFollowOn" reference. The use of -// kFollowOn refernces produces "clusters" of atoms that must stay together. -// If an order_file tries to move one atom, it may need to move a whole cluster. The -// algorithm to do this models clusters using two maps. The "starts" maps maps any -// atom in a cluster to the first Atom in the cluster. The "nexts" maps an Atom in a -// cluster to the next Atom in the cluster. With this in place, while processing an -// order_file, if any entry is in a cluster (in "starts" map), then the entire cluster is -// given ordinal overrides. -// -void Linker::sortAtoms() -{ - fStartSortTime = mach_absolute_time(); - // if -order_file is used, build map of atom ordinal overrides - std::map* ordinalOverrideMap = NULL; - std::map theOrdinalOverrideMap; - const bool log = false; - if ( fOptions.orderedSymbols().size() != 0 ) { - // first make a pass to find all follow-on references and build start/next maps - // which are a way to represent clusters of atoms that must layout together - std::map followOnStarts; - std::map followOnNexts; - for (std::vector::iterator ait=fAllAtoms.begin(); ait != fAllAtoms.end(); ait++) { - ObjectFile::Atom* atom = *ait; - std::vector& references = atom->getReferences(); - for (std::vector::iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Reference* ref = *rit; - if ( ref->getKind() == 1 ) { // FIX FIX - ObjectFile::Atom* targetAtom = &ref->getTarget(); - if ( log ) fprintf(stderr, "ref %s -> %s", atom->getDisplayName(), targetAtom->getDisplayName()); - std::map::iterator startFrom = followOnStarts.find(atom); - std::map::iterator startTo = followOnStarts.find(targetAtom); - if ( (startFrom == followOnStarts.end()) && (startTo == followOnStarts.end()) ) { - // this is first time we've seen either atom, make simple cluster of the two - if ( log ) fprintf(stderr, " new cluster\n"); - followOnStarts[atom] = atom; - followOnStarts[targetAtom] = atom; - followOnNexts[atom] = targetAtom; - followOnNexts[targetAtom] = NULL; - } - else if ( (startFrom != followOnStarts.end()) && (startTo == followOnStarts.end()) && (followOnNexts[atom] == NULL) ) { - // atom is at end of an existing cluster, so append target to end of cluster - if ( log ) fprintf(stderr, " end of cluster starting with %s\n", followOnStarts[atom]->getDisplayName()); - followOnNexts[atom] = targetAtom; - followOnNexts[targetAtom] = NULL; - followOnStarts[targetAtom] = followOnStarts[atom]; - } - else { - // gerneral case of inserting into an existing cluster - if ( followOnNexts[atom] != NULL ) { - // an atom with two follow-ons is illegal - warning("can't order %s because both %s and %s must follow it", - atom->getDisplayName(), targetAtom->getDisplayName(), followOnNexts[atom]->getDisplayName()); - } - else { - // there already exists an atom that says target must be its follow-on - const ObjectFile::Atom* originalStart = startTo->second; - const ObjectFile::Atom* originalPrevious = originalStart; - while ( followOnNexts[originalPrevious] != targetAtom ) - originalPrevious = followOnNexts[originalPrevious]; - bool otherIsAlias = (originalPrevious->getSize() == 0); - bool thisIsAlias = (atom->getSize() == 0); - if ( !otherIsAlias && !thisIsAlias ) { - warning("can't order %s because both %s and %s must preceed it", - targetAtom->getDisplayName(), originalPrevious->getDisplayName(), atom->getDisplayName()); - } - else if ( otherIsAlias ) { - if ( originalPrevious == originalStart ) { - // other is alias at start of cluster, make this the new start of cluster - if ( log ) fprintf(stderr, " becomes new start of cluster previous starting with %s\n", originalStart->getDisplayName()); - followOnNexts[atom] = originalPrevious; - for(const ObjectFile::Atom* nextAtom = atom; nextAtom != NULL; nextAtom = followOnNexts[nextAtom]) - followOnStarts[nextAtom] = atom; - } - else { - // other is alias in middle of cluster, insert new atom before it - if ( log ) fprintf(stderr, " insert into cluster starting with %s before alias %s\n", originalStart->getDisplayName(), originalPrevious->getDisplayName()); - followOnStarts[atom] = originalStart; - followOnNexts[atom] = originalPrevious; - for(const ObjectFile::Atom* a = originalStart; a != NULL; a = followOnNexts[a]) { - if ( followOnNexts[a] == originalPrevious ) { - followOnNexts[a] = atom; - break; - } - } - } - } - else { - // this is alias, so it can go inbetween originalPrevious and targetAtom - if ( log ) fprintf(stderr, " insert into cluster starting with %s after %s\n", originalStart->getDisplayName(), originalPrevious->getDisplayName()); - followOnStarts[atom] = originalStart; - followOnNexts[atom] = followOnNexts[originalPrevious]; - followOnNexts[originalPrevious] = atom; - } - } - } - } - } - } - - if ( log ) { - for(std::map::iterator it = followOnStarts.begin(); it != followOnStarts.end(); ++it) - fprintf(stderr, "start %s -> %s\n", it->first->getDisplayName(), it->second->getDisplayName()); - - for(std::map::iterator it = followOnNexts.begin(); it != followOnNexts.end(); ++it) - fprintf(stderr, "next %s -> %s\n", it->first->getDisplayName(), (it->second != NULL) ? it->second->getDisplayName() : "null"); - } - - // with the start/next maps of follow-on atoms we can process the order file and produce override ordinals - ordinalOverrideMap = &theOrdinalOverrideMap; - uint32_t index = 0; - uint32_t matchCount = 0; - std::vector& orderedSymbols = fOptions.orderedSymbols(); - for(std::vector::iterator it = orderedSymbols.begin(); it != orderedSymbols.end(); ++it) { - ObjectFile::Atom* atom = this->findAtom(*it); - if ( atom != NULL ) { - std::map::iterator start = followOnStarts.find(atom); - if ( start != followOnStarts.end() ) { - // this symbol for the order file corresponds to an atom that is in a cluster that must lay out together - for(const ObjectFile::Atom* nextAtom = start->second; nextAtom != NULL; nextAtom = followOnNexts[nextAtom]) { - std::map::iterator pos = theOrdinalOverrideMap.find(nextAtom); - if ( pos == theOrdinalOverrideMap.end() ) { - theOrdinalOverrideMap[nextAtom] = index++; - if (log ) fprintf(stderr, "override ordinal %u assigned to %s in cluster from %s\n", index, nextAtom->getDisplayName(), nextAtom->getFile()->getPath()); - } - else { - if (log ) fprintf(stderr, "could not order %s as %u because it was already laid out earlier by %s as %u\n", - atom->getDisplayName(), index, followOnStarts[atom]->getDisplayName(), theOrdinalOverrideMap[atom] ); - } - } - } - else { - theOrdinalOverrideMap[atom] = index; - if (log ) fprintf(stderr, "override ordinal %u assigned to %s from %s\n", index, atom->getDisplayName(), atom->getFile()->getPath()); - } - } - else { - ++matchCount; - //fprintf(stderr, "can't find match for order_file entry %s/%s\n", it->objectFileName, it->symbolName); - } - ++index; - } - if ( fOptions.printOrderFileStatistics() && (fOptions.orderedSymbols().size() != matchCount) ) { - warning("only %u out of %lu order_file symbols were applicable", matchCount, fOptions.orderedSymbols().size() ); - } - } - - // sort atoms - std::sort(fAllAtoms.begin(), fAllAtoms.end(), Linker::AtomSorter(ordinalOverrideMap, fInitializerAtoms, fTerminatorAtoms)); - - //fprintf(stderr, "Sorted atoms:\n"); - //for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - // fprintf(stderr, "\t%p, %u %s\t%s\n", (*it)->getSection(), (*it)->getSection()->getIndex(), (*it)->getDisplayName(), (*it)->getFile()->getPath()); - //} -} - - -// make sure given addresses are within reach of branches, etc -void Linker::tweakLayout() -{ - // > 2GB images need their large zero fill atoms sorted to the end to keep access with +/- 2GB - if ( fTotalSize > 0x7F000000 ) { - fBiggerThanTwoGigOutput = true; - - if ( (fTotalSize-fTotalZeroFillSize) > 0x7F000000 ) - throwf("total output size exceeds 2GB (%lldMB)", (fTotalSize-fTotalZeroFillSize)/(1024*1024)); - - // move very large (>1MB) zero fill atoms to a new section at very end of __DATA segment - Section* hugeZeroFills = Section::find("__huge", "__DATA", true, true); - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* atom = *it; - if ( atom->isZeroFill() && (atom->getSize() > 1024*1024) && (strcmp(atom->getSegment().getName(), "__DATA") == 0) ) - atom->setSection(hugeZeroFills); - } - } - - // move all initializers to start of __text section - if ( fOptions.readerOptions().fAutoOrderInitializers ) { - // move -init function to front of __text - if ( fOptions.initFunctionName() != NULL ) { - ObjectFile::Atom* initAtom = fGlobalSymbolTable.find(fOptions.initFunctionName()); - if ( initAtom == NULL ) - throwf("could not find -init function: \"%s\"", fOptions.initFunctionName()); - moveToFrontOfSection(initAtom); - } - - // move all functions pointed to by __mod_init_func section to front of __text - Section* initSection = Section::find("__mod_init_func", "__DATA", false, true, false); - if ( initSection != NULL ) { - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); ++it) { - if ( (*it)->getSection() == initSection ) { - std::vector& references = (*it)->getReferences(); - if ( references.size() == 1 ) - moveToFrontOfSection(&(references[0]->getTarget())); - } - } - } - } - - // move atoms with relocations to start of __DATA,__data section - // linker should order __DATA segment to reduce dyld dirtied pages - if ( fOptions.orderData() ) { - bool slideable = false; - switch ( fOptions.outputKind() ) { - case Options::kDynamicExecutable: - case Options::kStaticExecutable: - case Options::kDyld: - case Options::kPreload: - case Options::kObjectFile: - case Options::kKextBundle: - slideable = false; - break; - case Options::kDynamicLibrary: - case Options::kDynamicBundle: - slideable = true; - break; - } - const bool hasPreferredLoadAddress = (fOptions.baseAddress() != 0); - Section* dataSection = Section::find("__data", "__DATA", false, true, false); - if ( dataSection != NULL ) { - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); ++it) { - ObjectFile::Atom* dataAtom = *it; - if ( dataAtom->getSection() == dataSection ) { - std::vector& references = dataAtom->getReferences(); - if ( references.size() > 0 ) { - if ( slideable && !hasPreferredLoadAddress ) { - // in a slidable image dyld will need to rebase and bind so any references will need runtime fixups - // if image has preferred base address, assume it will load there and not rebase - moveToFrontOfSection(dataAtom); - } - else { - // in a non-slideable image, dyld will only do binding, so only references to - // symbols in another dylib will need runtime fixups - //fprintf(stderr, "reference from atom %s\n", dataAtom->getDisplayName()); - for (std::vector::iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Reference* reference = *rit; - //fprintf(stderr, "\t%d %s\n", reference->getTarget().getDefinitionKind(), reference->getTarget().getDisplayName()); - if ( (reference->getTarget().getDefinitionKind() == ObjectFile::Atom::kExternalDefinition) - || (reference->getTarget().getDefinitionKind() == ObjectFile::Atom::kExternalWeakDefinition) ) { - moveToFrontOfSection(dataAtom); - break; - } - } - } - } - } - } - } - } - -} - - -void Linker::writeDotOutput() -{ - const char* dotOutFilePath = fOptions.dotOutputFile(); - if ( dotOutFilePath != NULL ) { - FILE* out = fopen(dotOutFilePath, "w"); - if ( out != NULL ) { - // print header - fprintf(out, "digraph dg\n{\n"); - fprintf(out, "\tconcentrate = true;\n"); - fprintf(out, "\trankdir = LR;\n"); - - // print each atom as a node - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* atom = *it; - if ( atom->getFile() != fOutputFile ) { - const char* name = atom->getDisplayName(); - if ( (atom->getDefinitionKind() == ObjectFile::Atom::kExternalDefinition) - || (atom->getDefinitionKind() == ObjectFile::Atom::kExternalWeakDefinition) ) { - fprintf(out, "\taddr%p [ shape = plaintext, label = \"%s\" ];\n", atom, name); - } - else if ( strcmp(atom->getSectionName(), "__cstring") == 0 ) { - char cstring[atom->getSize()+2]; - atom->copyRawContent((uint8_t*)cstring); - fprintf(out, "\taddr%p [ label = \"string: '", atom); - for (const char* s=cstring; *s != '\0'; ++s) { - if ( *s == '\n' ) - fprintf(out, "\\\\n"); - else - fputc(*s, out); - } - fprintf(out, "'\" ];\n"); - } - else { - fprintf(out, "\taddr%p [ label = \"%s\" ];\n", atom, name); - } - } - } - fprintf(out, "\n"); - - // print each reference as an edge - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* fromAtom = *it; - if ( fromAtom->getFile() != fOutputFile ) { - std::vector& references = fromAtom->getReferences(); - std::set seenTargets; - for (std::vector::iterator rit=references.begin(); rit != references.end(); rit++) { - ObjectFile::Reference* reference = *rit; - ObjectFile::Atom* toAtom = &(reference->getTarget()); - if ( seenTargets.count(toAtom) == 0 ) { - seenTargets.insert(toAtom); - fprintf(out, "\taddr%p -> addr%p;\n", fromAtom, toAtom); - } - } - } - } - fprintf(out, "\n"); - - // push all imports to bottom of graph - fprintf(out, "{ rank = same; "); - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - ObjectFile::Atom* atom = *it; - if ( atom->getFile() != fOutputFile ) - if ( (atom->getDefinitionKind() == ObjectFile::Atom::kExternalDefinition) - || (atom->getDefinitionKind() == ObjectFile::Atom::kExternalWeakDefinition) ) { - fprintf(out, "addr%p; ", atom); - } - } - fprintf(out, "};\n "); - - // print footer - fprintf(out, "}\n"); - fclose(out); - } - else { - warning("could not write dot output file: %s", dotOutFilePath); - } - } -} - -ObjectFile::Atom* Linker::entryPoint(bool orInit) -{ - // if main executable, find entry point atom - ObjectFile::Atom* entryPoint = NULL; - switch ( fOptions.outputKind() ) { - case Options::kDynamicExecutable: - case Options::kStaticExecutable: - case Options::kDyld: - case Options::kPreload: - entryPoint = fGlobalSymbolTable.find(fOptions.entryName()); - if ( entryPoint == NULL ) { - throwf("could not find entry point \"%s\" (perhaps missing crt1.o)", fOptions.entryName()); - } - break; - case Options::kDynamicLibrary: - if ( orInit && (fOptions.initFunctionName() != NULL) ) { - entryPoint = fGlobalSymbolTable.find(fOptions.initFunctionName()); - if ( entryPoint == NULL ) { - throwf("could not find -init function: \"%s\"", fOptions.initFunctionName()); - } - } - break; - case Options::kObjectFile: - case Options::kDynamicBundle: - case Options::kKextBundle: - entryPoint = NULL; - break; - } - return entryPoint; -} - -ObjectFile::Atom* Linker::dyldClassicHelper() -{ - if ( fOptions.makeClassicDyldInfo() ) - return fGlobalSymbolTable.find("dyld_stub_binding_helper"); - else - return NULL; -} - -ObjectFile::Atom* Linker::dyldCompressedHelper() -{ - if ( fOptions.makeCompressedDyldInfo() ) { - // dyld_stub_binder is in libSystem.B.dylib - ObjectFile::Atom* atom = fGlobalSymbolTable.find("dyld_stub_binder"); - if ( atom == NULL ) { - this->addJustInTimeAtoms("dyld_stub_binder", true, false, true); - } - atom = fGlobalSymbolTable.find("dyld_stub_binder"); - return atom; - } - else - return NULL; -} - -ObjectFile::Atom* Linker::dyldLazyLibraryHelper() -{ - return fGlobalSymbolTable.find("dyld_lazy_dylib_stub_binding_helper"); -} - -const char* Linker::assureFullPath(const char* path) -{ - if ( path[0] == '/' ) - return path; - char cwdbuff[MAXPATHLEN]; - if ( getcwd(cwdbuff, MAXPATHLEN) != NULL ) { - char* result; - asprintf(&result, "%s/%s", cwdbuff, path); - if ( result != NULL ) - return result; - } - return path; -} - - -// -// The stab strings are of the form: -// ':' -// but the contain a colon. -// For C++ may contain a double colon (e.g. std::string:f(0,1) ) -// For Objective-C name may contain a colon instead square bracket (e.g. [Foo doit:]:f(0,1) ) -// -const char* Linker::truncateStabString(const char* str) -{ - enum { start, inObjc } state = start; - for (const char* s = str; *s != 0; ++s) { - char c = *s; - switch (state) { - case start: - if ( c == '[' ) { - state = inObjc; - } - else { - if ( c == ':' ) { - if ( s[1] == ':' ) { - ++s; - } - else { - // found colon - // Duplicate strndup behavior here. - int trunStrLen = s-str+2; - char* temp = new char[trunStrLen+1]; - memcpy(temp, str, trunStrLen); - temp[trunStrLen] = '\0'; - return temp; - } - } - } - break; - case inObjc: - if ( c == ']' ) { - state = start; - } - break; - } - } - // malformed - return str; -} - - -bool Linker::minimizeStab(ObjectFile::Reader::Stab& stab) -{ - switch(stab.type){ - case N_GSYM: - case N_STSYM: - case N_LCSYM: - case N_FUN: - // these all need truncated strings - stab.string = truncateStabString(stab.string); - return true; - case N_SO: - case N_OSO: - case N_OPT: - case N_SOL: - // these are included in the minimal stabs, but they keep their full string - return true; - default: - return false; - } -} - - -struct HeaderRange { - std::vector::iterator begin; - std::vector::iterator end; - int parentRangeIndex; - uint32_t sum; - bool sumPrecomputed; - bool useEXCL; - bool cannotEXCL; // because of SLINE, etc stabs -}; - - -typedef __gnu_cxx::hash_map, __gnu_cxx::hash, CStringEquals> PathToSums; - -// hash table that maps header path to a vector of known checksums for that path -static PathToSums sKnownBINCLs; - - -void Linker::collectStabs(ObjectFile::Reader* reader, std::map& atomOrdinals) -{ - const bool log = false; - bool minimal = ( fOptions.readerOptions().fDebugInfoStripping == ObjectFile::ReaderOptions::kDebugInfoMinimal ); - std::vector* readerStabs = reader->getStabs(); - if ( readerStabs == NULL ) - return; - - if ( log ) fprintf(stderr, "processesing %lu stabs for %s\n", readerStabs->size(), reader->getPath()); - std::vector ranges; - int curRangeIndex = -1; - int count = 0; - ObjectFile::Atom* atomWithLowestOrdinal = NULL; - ObjectFile::Atom* atomWithHighestOrdinal = NULL; - uint32_t highestOrdinal = 0; - uint32_t lowestOrdinal = UINT_MAX; - std::vector > soRanges; - // 1) find all (possibly nested) BINCL/EINCL ranges and their checksums - // 2) find all SO/SO ranges and the first/last atom own by a FUN stab therein - for(std::vector::iterator it=readerStabs->begin(); it != readerStabs->end(); ++it) { - ++count; - switch ( it->type ) { - case N_BINCL: - { - HeaderRange range; - range.begin = it; - range.end = readerStabs->end(); - range.parentRangeIndex = curRangeIndex; - range.sum = it->value; - range.sumPrecomputed = (range.sum != 0); - range.useEXCL = false; - range.cannotEXCL = false; - curRangeIndex = ranges.size(); - if ( log ) fprintf(stderr, "[%d]BINCL %s\n", curRangeIndex, it->string); - ranges.push_back(range); - } - break; - case N_EINCL: - if ( curRangeIndex == -1 ) { - warning("EINCL missing BINCL in %s", reader->getPath()); - } - else { - ranges[curRangeIndex].end = it+1; - if ( log ) fprintf(stderr, "[%d->%d]EINCL %s\n", curRangeIndex, ranges[curRangeIndex].parentRangeIndex, it->string); - curRangeIndex = ranges[curRangeIndex].parentRangeIndex; - } - break; - case N_FUN: - { - std::map::iterator pos = atomOrdinals.find(it->atom); - if ( pos != atomOrdinals.end() ) { - uint32_t ordinal = pos->second; - if ( ordinal > highestOrdinal ) { - highestOrdinal = ordinal; - atomWithHighestOrdinal = it->atom; - } - if ( ordinal < lowestOrdinal ) { - lowestOrdinal = ordinal; - atomWithLowestOrdinal = it->atom; - } - } - } - // fall through - case N_BNSYM: - case N_ENSYM: - case N_LBRAC: - case N_RBRAC: - case N_SLINE: - case N_STSYM: - case N_LCSYM: - if ( curRangeIndex != -1 ) { - ranges[curRangeIndex].cannotEXCL = true; - if ( fOptions.warnStabs() ) - warning("cannot do BINCL/EINCL optimzation because of stabs kinds in %s for %s\n", ranges[curRangeIndex].begin->string, reader->getPath()); - } - break; - case N_SO: - if ( (it->string != NULL) && (strlen(it->string) > 0) ) { - // start SO, reset hi/low FUN tracking - atomWithLowestOrdinal = NULL; - atomWithHighestOrdinal = NULL; - highestOrdinal = 0; - lowestOrdinal = UINT_MAX; - } - else { - // end SO, record hi/low atoms for this SO range - soRanges.push_back(std::make_pair(atomWithLowestOrdinal, atomWithHighestOrdinal)); - } - // fall through - default: - if ( curRangeIndex != -1 ) { - if ( ! ranges[curRangeIndex].sumPrecomputed ) { - uint32_t sum = 0; - const char* s = it->string; - char c; - while ( (c = *s++) != 0 ) { - sum += c; - // don't checkusm first number (file index) after open paren in string - if ( c == '(' ) { - while(isdigit(*s)) - ++s; - } - } - ranges[curRangeIndex].sum += sum; - } - } - - } - } - if ( log ) fprintf(stderr, "processesed %d stabs for %s\n", count, reader->getPath()); - if ( curRangeIndex != -1 ) - warning("BINCL (%s) missing EINCL in %s", ranges[curRangeIndex].begin->string, reader->getPath()); - - // if no BINCLs - if ( ranges.size() == 0 ) { - unsigned int soIndex = 0; - for(std::vector::iterator it=readerStabs->begin(); it != readerStabs->end(); ++it) { - // copy minimal or all stabs - ObjectFile::Reader::Stab stab = *it; - if ( !minimal || minimizeStab(stab) ) { - if ( stab.type == N_SO ) { - if ( soIndex < soRanges.size() ) { - if ( (stab.string != NULL) && (strlen(stab.string) > 0) ) { - // starting SO is associated with first atom - stab.atom = soRanges[soIndex].first; - } - else { - // ending SO is associated with last atom - stab.atom = soRanges[soIndex].second; - ++soIndex; - } - } - } - fStabs.push_back(stab); - } - } - return; - } - - //fprintf(stderr, "BINCL/EINCL info for %s\n", reader->getPath()); - //for(std::vector::iterator it=ranges.begin(); it != ranges.end(); ++it) { - // fprintf(stderr, "%08X %s\n", it->sum, it->begin->string); - //} - - // see if any of these BINCL/EINCL ranges have already been seen and therefore can be replaced with EXCL - for(std::vector::iterator it=ranges.begin(); it != ranges.end(); ++it) { - if ( ! it->cannotEXCL ) { - const char* header = it->begin->string; - uint32_t sum = it->sum; - PathToSums::iterator pos = sKnownBINCLs.find(header); - if ( pos != sKnownBINCLs.end() ) { - std::vector& sums = pos->second; - for(std::vector::iterator sit=sums.begin(); sit != sums.end(); ++sit) { - if (*sit == sum) { - //fprintf(stderr, "use EXCL for %s in %s\n", header, reader->getPath()); - it->useEXCL = true; - break; - } - } - if ( ! it->useEXCL ) { - // have seen this path, but not this checksum - //fprintf(stderr, "registering another checksum %08X for %s\n", sum, header); - sums.push_back(sum); - } - } - else { - // have not seen this path, so add to known BINCLs - std::vector empty; - sKnownBINCLs[header] = empty; - sKnownBINCLs[header].push_back(sum); - //fprintf(stderr, "registering checksum %08X for %s\n", sum, header); - } - } - } - - // add a new set of stabs with BINCL/EINCL runs that have been seen before, replaced with EXCLs - curRangeIndex = -1; - const int maxRangeIndex = ranges.size(); - int soIndex = 0; - for(std::vector::iterator it=readerStabs->begin(); it != readerStabs->end(); ++it) { - switch ( it->type ) { - case N_BINCL: - for(int i=curRangeIndex+1; i < maxRangeIndex; ++i) { - if ( ranges[i].begin == it ) { - curRangeIndex = i; - HeaderRange& range = ranges[curRangeIndex]; - ObjectFile::Reader::Stab stab = *it; - stab.value = range.sum; // BINCL and EXCL have n_value set to checksum - if ( range.useEXCL ) - stab.type = N_EXCL; // transform BINCL into EXCL - if ( !minimal ) - fStabs.push_back(stab); - break; - } - } - break; - case N_EINCL: - if ( curRangeIndex != -1 ) { - if ( !ranges[curRangeIndex].useEXCL && !minimal ) - fStabs.push_back(*it); - curRangeIndex = ranges[curRangeIndex].parentRangeIndex; - } - break; - default: - if ( (curRangeIndex == -1) || !ranges[curRangeIndex].useEXCL ) { - ObjectFile::Reader::Stab stab = *it; - if ( !minimal || minimizeStab(stab) ) { - if ( stab.type == N_SO ) { - if ( (stab.string != NULL) && (strlen(stab.string) > 0) ) { - // starting SO is associated with first atom - stab.atom = soRanges[soIndex].first; - } - else { - // ending SO is associated with last atom - stab.atom = soRanges[soIndex].second; - ++soIndex; - } - } - fStabs.push_back(stab); - } - } - } - } - -} - - -// used to prune out atoms that don't need debug notes generated -class NoDebugNoteAtom -{ -public: - NoDebugNoteAtom(const std::map& readersWithDwarfOrdinals) - : fReadersWithDwarfOrdinals(readersWithDwarfOrdinals) {} - - bool operator()(const ObjectFile::Atom* atom) const { - if ( atom->getSymbolTableInclusion() == ObjectFile::Atom::kSymbolTableNotIn ) - return true; - if ( atom->getName() == NULL ) - return true; - if ( fReadersWithDwarfOrdinals.find(atom->getFile()) == fReadersWithDwarfOrdinals.end() ) - return true; - return false; - } - -private: - const std::map& fReadersWithDwarfOrdinals; -}; - -// used to sort atoms with debug notes -class ReadersWithDwarfSorter -{ -public: - ReadersWithDwarfSorter(const std::map& readersWithDwarfOrdinals, - const std::map& atomOrdinals) - : fReadersWithDwarfOrdinals(readersWithDwarfOrdinals), fAtomOrdinals(atomOrdinals) {} - - bool operator()(const ObjectFile::Atom* left, const ObjectFile::Atom* right) const - { - // first sort by reader - unsigned int leftReaderIndex = fReadersWithDwarfOrdinals.find(left->getFile())->second; - unsigned int rightReaderIndex = fReadersWithDwarfOrdinals.find(right->getFile())->second; - if ( leftReaderIndex != rightReaderIndex ) - return (leftReaderIndex < rightReaderIndex); - - // then sort by atom ordinal - unsigned int leftAtomIndex = fAtomOrdinals.find(left)->second; - unsigned int rightAtomIndex = fAtomOrdinals.find(right)->second; - return leftAtomIndex < rightAtomIndex; - } - -private: - const std::map& fReadersWithDwarfOrdinals; - const std::map& fAtomOrdinals; -}; - - - - - -void Linker::synthesizeDebugNotes(std::vector& allAtomsByReader) -{ - // synthesize "debug notes" and add them to master stabs vector - const char* dirPath = NULL; - const char* filename = NULL; - bool wroteStartSO = false; - bool useZeroOSOModTime = (getenv("RC_RELEASE") != NULL); - __gnu_cxx::hash_set, CStringEquals> seenFiles; - for (std::vector::iterator it=allAtomsByReader.begin(); it != allAtomsByReader.end(); it++) { - ObjectFile::Atom* atom = *it; - const char* newDirPath; - const char* newFilename; - //fprintf(stderr, "debug note for %s\n", atom->getDisplayName()); - if ( atom->getTranslationUnitSource(&newDirPath, &newFilename) ) { - // need SO's whenever the translation unit source file changes - if ( newFilename != filename ) { - // gdb like directory SO's to end in '/', but dwarf DW_AT_comp_dir usually does not have trailing '/' - if ( (newDirPath != NULL) && (strlen(newDirPath) > 1 ) && (newDirPath[strlen(newDirPath)-1] != '/') ) - asprintf((char**)&newDirPath, "%s/", newDirPath); - if ( filename != NULL ) { - // translation unit change, emit ending SO - ObjectFile::Reader::Stab endFileStab; - endFileStab.atom = NULL; - endFileStab.type = N_SO; - endFileStab.other = 1; - endFileStab.desc = 0; - endFileStab.value = 0; - endFileStab.string = ""; - fStabs.push_back(endFileStab); - } - // new translation unit, emit start SO's - ObjectFile::Reader::Stab dirPathStab; - dirPathStab.atom = NULL; - dirPathStab.type = N_SO; - dirPathStab.other = 0; - dirPathStab.desc = 0; - dirPathStab.value = 0; - dirPathStab.string = newDirPath; - fStabs.push_back(dirPathStab); - ObjectFile::Reader::Stab fileStab; - fileStab.atom = NULL; - fileStab.type = N_SO; - fileStab.other = 0; - fileStab.desc = 0; - fileStab.value = 0; - fileStab.string = newFilename; - fStabs.push_back(fileStab); - // Synthesize OSO for start of file - ObjectFile::Reader::Stab objStab; - objStab.atom = NULL; - objStab.type = N_OSO; - // linker should put cpusubtype in n_sect field of nlist entry for N_OSO debug note entries - objStab.other = atom->getFile()->updateCpuConstraint(0); - objStab.desc = 1; - objStab.value = useZeroOSOModTime ? 0 : atom->getFile()->getModificationTime(); - objStab.string = assureFullPath(atom->getFile()->getPath()); - fStabs.push_back(objStab); - wroteStartSO = true; - // add the source file path to seenFiles so it does not show up in SOLs - seenFiles.insert(newFilename); - } - filename = newFilename; - dirPath = newDirPath; - if ( atom->getSegment().isContentExecutable() && (strncmp(atom->getSectionName(), "__text", 6) == 0) ) { - // Synthesize BNSYM and start FUN stabs - ObjectFile::Reader::Stab beginSym; - beginSym.atom = atom; - beginSym.type = N_BNSYM; - beginSym.other = 1; - beginSym.desc = 0; - beginSym.value = 0; - beginSym.string = ""; - fStabs.push_back(beginSym); - ObjectFile::Reader::Stab startFun; - startFun.atom = atom; - startFun.type = N_FUN; - startFun.other = 1; - startFun.desc = 0; - startFun.value = 0; - startFun.string = atom->getName(); - fStabs.push_back(startFun); - // Synthesize any SOL stabs needed - std::vector* lineInfo = atom->getLineInfo(); - if ( lineInfo != NULL ) { - const char* curFile = NULL; - for (std::vector::iterator it = lineInfo->begin(); it != lineInfo->end(); ++it) { - if ( it->fileName != curFile ) { - if ( seenFiles.count(it->fileName) == 0 ) { - seenFiles.insert(it->fileName); - ObjectFile::Reader::Stab sol; - sol.atom = 0; - sol.type = N_SOL; - sol.other = 0; - sol.desc = 0; - sol.value = 0; - sol.string = it->fileName; - fStabs.push_back(sol); - } - curFile = it->fileName; - } - } - } - // Synthesize end FUN and ENSYM stabs - ObjectFile::Reader::Stab endFun; - endFun.atom = atom; - endFun.type = N_FUN; - endFun.other = 0; - endFun.desc = 0; - endFun.value = 0; - endFun.string = ""; - fStabs.push_back(endFun); - ObjectFile::Reader::Stab endSym; - endSym.atom = atom; - endSym.type = N_ENSYM; - endSym.other = 1; - endSym.desc = 0; - endSym.value = 0; - endSym.string = ""; - fStabs.push_back(endSym); - } - else if ( atom->getSymbolTableInclusion() == ObjectFile::Atom::kSymbolTableNotIn ) { - // no stabs for atoms that would not be in the symbol table - } - else if ( atom->getSymbolTableInclusion() == ObjectFile::Atom::kSymbolTableInAsAbsolute ) { - // no stabs for absolute symbols - } - else if ( (strcmp(atom->getSectionName(), "__eh_frame") == 0) ) { - // no stabs for .eh atoms - } - else if ( (strncmp(atom->getName(), "__dtrace_probe$", 15) == 0) ) { - // no stabs for old style dtrace probes - } - else { - ObjectFile::Reader::Stab globalsStab; - const char* name = atom->getName(); - if ( atom->getScope() == ObjectFile::Atom::scopeTranslationUnit ) { - // Synthesize STSYM stab for statics - globalsStab.atom = atom; - globalsStab.type = N_STSYM; - globalsStab.other = 1; - globalsStab.desc = 0; - globalsStab.value = 0; - globalsStab.string = name; - fStabs.push_back(globalsStab); - } - else { - // Synthesize GSYM stab for other globals - globalsStab.atom = atom; - globalsStab.type = N_GSYM; - globalsStab.other = 1; - globalsStab.desc = 0; - globalsStab.value = 0; - globalsStab.string = name; - fStabs.push_back(globalsStab); - } - } - } - } - - if ( wroteStartSO ) { - // emit ending SO - ObjectFile::Reader::Stab endFileStab; - endFileStab.atom = NULL; - endFileStab.type = N_SO; - endFileStab.other = 1; - endFileStab.desc = 0; - endFileStab.value = 0; - endFileStab.string = ""; - fStabs.push_back(endFileStab); - } -} - - - - -void Linker::collectDebugInfo() -{ - std::map atomOrdinals; - fStartDebugTime = mach_absolute_time(); - if ( fOptions.readerOptions().fDebugInfoStripping != ObjectFile::ReaderOptions::kDebugInfoNone ) { - - // determine mixture of stabs and dwarf - bool someStabs = false; - bool someDwarf = false; - for (std::vector::iterator it=fReadersThatHaveSuppliedAtoms.begin(); - it != fReadersThatHaveSuppliedAtoms.end(); - it++) { - ObjectFile::Reader* reader = *it; - if ( reader != NULL ) { - switch ( reader->getDebugInfoKind() ) { - case ObjectFile::Reader::kDebugInfoNone: - break; - case ObjectFile::Reader::kDebugInfoStabs: - someStabs = true; - break; - case ObjectFile::Reader::kDebugInfoDwarf: - someDwarf = true; - fCreateUUID = true; - break; - case ObjectFile::Reader::kDebugInfoStabsUUID: - someStabs = true; - fCreateUUID = true; - break; - default: - throw "Unhandled type of debug information"; - } - } - } - - if ( someDwarf || someStabs ) { - // try to minimize re-allocations - fStabs.reserve(1024); - - // make mapping from atoms to ordinal - uint32_t ordinal = 1; - for (std::vector::iterator it=fAllAtoms.begin(); it != fAllAtoms.end(); it++) { - atomOrdinals[*it] = ordinal++; - } - } - - // process all dwarf .o files as a batch - if ( someDwarf ) { - // make mapping from readers with dwarf to ordinal - std::map readersWithDwarfOrdinals; - uint32_t readerOrdinal = 1; - for (std::vector::iterator it=fReadersThatHaveSuppliedAtoms.begin(); - it != fReadersThatHaveSuppliedAtoms.end(); - it++) { - ObjectFile::Reader* reader = *it; - if ( (reader != NULL) && (reader->getDebugInfoKind() == ObjectFile::Reader::kDebugInfoDwarf) ) { - readersWithDwarfOrdinals[reader] = readerOrdinal++; - } - } - - // make a vector of atoms - std::vector allAtomsByReader(fAllAtoms.begin(), fAllAtoms.end()); - // remove those not from a reader that has dwarf - allAtomsByReader.erase(std::remove_if(allAtomsByReader.begin(), allAtomsByReader.end(), - NoDebugNoteAtom(readersWithDwarfOrdinals)), allAtomsByReader.end()); - // sort by reader then atom ordinal - std::sort(allAtomsByReader.begin(), allAtomsByReader.end(), ReadersWithDwarfSorter(readersWithDwarfOrdinals, atomOrdinals)); - // add debug notes for each atom - this->synthesizeDebugNotes(allAtomsByReader); - } - - // process all stabs .o files one by one - if ( someStabs ) { - // get stabs from each reader, in command line order - for (std::vector::iterator it=fReadersThatHaveSuppliedAtoms.begin(); - it != fReadersThatHaveSuppliedAtoms.end(); - it++) { - ObjectFile::Reader* reader = *it; - if ( reader != NULL ) { - switch ( reader->getDebugInfoKind() ) { - case ObjectFile::Reader::kDebugInfoDwarf: - case ObjectFile::Reader::kDebugInfoNone: - // do nothing - break; - case ObjectFile::Reader::kDebugInfoStabs: - case ObjectFile::Reader::kDebugInfoStabsUUID: - collectStabs(reader, atomOrdinals); - break; - default: - throw "Unhandled type of debug information"; - } - } - } - // remove stabs associated with atoms that won't be in output - std::set allAtomsSet; - allAtomsSet.insert(fAllAtoms.begin(), fAllAtoms.end()); - fStabs.erase(std::remove_if(fStabs.begin(), fStabs.end(), NotInSet(allAtomsSet)), fStabs.end()); - } - } -} - -void Linker::writeOutput() -{ - if ( fOptions.forceCpuSubtypeAll() ) - fCurrentCpuConstraint = ObjectFile::Reader::kCpuAny; - - fStartWriteTime = mach_absolute_time(); - // tell writer about each segment's atoms - fOutputFileSize = fOutputFile->write(fAllAtoms, fStabs, this->entryPoint(true), - this->dyldClassicHelper(),this->dyldCompressedHelper(), this->dyldLazyLibraryHelper(), - fCreateUUID, fCanScatter, - fCurrentCpuConstraint, fBiggerThanTwoGigOutput, - fRegularDefAtomsThatOverrideADylibsWeakDef, - fGlobalSymbolTable.hasExternalWeakDefinitions()); -} - -ObjectFile::Reader* Linker::createReader(const Options::FileInfo& info) -{ - // map in whole file - uint64_t len = info.fileLen; - int fd = ::open(info.path, O_RDONLY, 0); - if ( fd == -1 ) - throwf("can't open file, errno=%d", errno); - if ( info.fileLen < 20 ) - throw "file too small"; - - uint8_t* p = (uint8_t*)::mmap(NULL, info.fileLen, PROT_READ, MAP_FILE | MAP_PRIVATE, fd, 0); - if ( p == (uint8_t*)(-1) ) - throwf("can't map file, errno=%d", errno); - - // if fat file, skip to architecture we want - // Note: fat header is always big-endian - const fat_header* fh = (fat_header*)p; - if ( fh->magic == OSSwapBigToHostInt32(FAT_MAGIC) ) { - const struct fat_arch* archs = (struct fat_arch*)(p + sizeof(struct fat_header)); - uint32_t sliceToUse; - bool sliceFound = false; - if ( fOptions.preferSubArchitecture() ) { - // first try to find a slice that match cpu-type and cpu-sub-type - for (uint32_t i=0; i < OSSwapBigToHostInt32(fh->nfat_arch); ++i) { - if ( (OSSwapBigToHostInt32(archs[i].cputype) == (uint32_t)fArchitecture) - && (OSSwapBigToHostInt32(archs[i].cpusubtype) == (uint32_t)fOptions.subArchitecture()) ) { - sliceToUse = i; - sliceFound = true; - break; - } - } - } - if ( !sliceFound ) { - // look for any slice that matches just cpu-type - for (uint32_t i=0; i < OSSwapBigToHostInt32(fh->nfat_arch); ++i) { - if ( OSSwapBigToHostInt32(archs[i].cputype) == (uint32_t)fArchitecture ) { - sliceToUse = i; - sliceFound = true; - break; - } - } - } - if ( sliceFound ) { - uint32_t fileOffset = OSSwapBigToHostInt32(archs[sliceToUse].offset); - len = OSSwapBigToHostInt32(archs[sliceToUse].size); - // if requested architecture is page aligned within fat file, then remap just that portion of file - if ( (fileOffset & 0x00000FFF) == 0 ) { - // unmap whole file - munmap((caddr_t)p, info.fileLen); - // re-map just part we need - p = (uint8_t*)::mmap(NULL, len, PROT_READ, MAP_FILE | MAP_PRIVATE, fd, fileOffset); - if ( p == (uint8_t*)(-1) ) - throwf("can't re-map file, errno=%d", errno); - } - else { - p = &p[fileOffset]; - } - } - } - ::close(fd); - - bool objSubtypeMustMatch = (fOptions.preferSubArchitecture() && !fOptions.allowSubArchitectureMismatches()); - switch (fArchitecture) { - case CPU_TYPE_POWERPC: - if ( mach_o::relocatable::Reader::validFile(p) ) - return this->addObject(new mach_o::relocatable::Reader::Reader(p, info.path, info.modTime, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( mach_o::dylib::Reader::validFile(p, info.options.fBundleLoader) ) - return this->addDylib(new mach_o::dylib::Reader::Reader(p, len, info.path, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( archive::Reader::validFile(p, len) ) - return this->addArchive(new archive::Reader::Reader(p, len, info.path, info.modTime, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - break; - case CPU_TYPE_POWERPC64: - if ( mach_o::relocatable::Reader::validFile(p) ) - return this->addObject(new mach_o::relocatable::Reader::Reader(p, info.path, info.modTime, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( mach_o::dylib::Reader::validFile(p, info.options.fBundleLoader) ) - return this->addDylib(new mach_o::dylib::Reader::Reader(p, len, info.path, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( archive::Reader::validFile(p, len) ) - return this->addArchive(new archive::Reader::Reader(p, len, info.path, info.modTime, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - break; - case CPU_TYPE_I386: - if ( mach_o::relocatable::Reader::validFile(p) ) - return this->addObject(new mach_o::relocatable::Reader::Reader(p, info.path, info.modTime, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( mach_o::dylib::Reader::validFile(p, info.options.fBundleLoader) ) - return this->addDylib(new mach_o::dylib::Reader::Reader(p, len, info.path, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( archive::Reader::validFile(p, len) ) - return this->addArchive(new archive::Reader::Reader(p, len, info.path, info.modTime, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - break; - case CPU_TYPE_X86_64: - if ( mach_o::relocatable::Reader::validFile(p) ) - return this->addObject(new mach_o::relocatable::Reader::Reader(p, info.path, info.modTime, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( mach_o::dylib::Reader::validFile(p, info.options.fBundleLoader) ) - return this->addDylib(new mach_o::dylib::Reader::Reader(p, len, info.path, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( archive::Reader::validFile(p, len) ) - return this->addArchive(new archive::Reader::Reader(p, len, info.path, info.modTime, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - case CPU_TYPE_ARM: - if ( mach_o::relocatable::Reader::validFile(p, objSubtypeMustMatch, fOptions.subArchitecture()) ) - return this->addObject(new mach_o::relocatable::Reader::Reader(p, info.path, info.modTime, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( mach_o::dylib::Reader::validFile(p, info.options.fBundleLoader) ) - return this->addDylib(new mach_o::dylib::Reader::Reader(p, len, info.path, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - else if ( archive::Reader::validFile(p, len) ) - return this->addArchive(new archive::Reader::Reader(p, len, info.path, info.modTime, info.options, fOptions.readerOptions(), fNextInputOrdinal), info, len); - break; - } - -#if LTO_SUPPORT - if ( lto::Reader::validFile(p, len, fArchitecture) ) { - return this->addObject(new lto::Reader(p, len, info.path, info.modTime, fOptions.readerOptions(), fArchitecture), info, len); - } - else if ( !lto::Reader::loaded() && (p[0] == 'B') && (p[1] == 'C') ) { - throw "could not process object file. Looks like an llvm bitcode object file, but libLTO.dylib could not be loaded"; - } -#endif - // error handling - if ( ((fat_header*)p)->magic == OSSwapBigToHostInt32(FAT_MAGIC) ) { - throwf("missing required architecture %s in file", fArchitectureName); - } - else { - throw "file is not of required architecture"; - } -} - -void Linker::logDylib(ObjectFile::Reader* reader, bool indirect) -{ - if ( fOptions.readerOptions().fTraceDylibs ) { - const char* fullPath = reader->getPath(); - char realName[MAXPATHLEN]; - if ( realpath(fullPath, realName) != NULL ) - fullPath = realName; - if ( indirect ) - logTraceInfo("[Logging for XBS] Used indirect dynamic library: %s\n", fullPath); - else - logTraceInfo("[Logging for XBS] Used dynamic library: %s\n", fullPath); - } -} - - - -ObjectFile::Reader* Linker::findDylib(const char* installPath, const char* fromPath) -{ - //fprintf(stderr, "findDylib(%s, %s)\n", installPath, fromPath); - InstallNameToReader::iterator pos = fDylibMap.find(installPath); - if ( pos != fDylibMap.end() ) { - return pos->second; - } - else { - // allow -dylib_path option to override indirect library to use - for (std::vector::iterator dit = fOptions.dylibOverrides().begin(); dit != fOptions.dylibOverrides().end(); ++dit) { - if ( strcmp(dit->installName,installPath) == 0 ) {\ - try { - Options::FileInfo info = fOptions.findFile(dit->useInstead); - ObjectFile::Reader* reader = this->createReader(info); - fDylibMap[strdup(installPath)] = reader; - this->logDylib(reader, true); - return reader; - } - catch (const char* msg) { - warning("ignoring -dylib_file option, %s", msg); - } - } - } - char newPath[MAXPATHLEN]; - // handle @loader_path - if ( strncmp(installPath, "@loader_path/", 13) == 0 ) { - strcpy(newPath, fromPath); - char* addPoint = strrchr(newPath,'/'); - if ( addPoint != NULL ) - strcpy(&addPoint[1], &installPath[13]); - else - strcpy(newPath, &installPath[13]); - installPath = newPath; - } - // note: @executable_path case is handled inside findFileUsingPaths() - // search for dylib using -F and -L paths - Options::FileInfo info = fOptions.findFileUsingPaths(installPath); - try { - ObjectFile::Reader* reader = this->createReader(info); - fDylibMap[strdup(installPath)] = reader; - this->logDylib(reader, true); - return reader; - } - catch (const char* msg) { - throwf("in %s, %s", info.path, msg); - } - } -} - - -void Linker::processDylibs() -{ - fAllDirectDylibsLoaded = true; - - // mark all dylibs initially specified as required and check if they can be used - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - it->second->setExplicitlyLinked(); - this->checkDylibClientRestrictions(it->second); - } - - // keep processing dylibs until no more dylibs are added - unsigned long lastMapSize = 0; - while ( lastMapSize != fDylibMap.size() ) { - lastMapSize = fDylibMap.size(); - // can't iterator fDylibMap while modifying it, so use temp buffer - std::vector currentUnprocessedReaders; - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - if ( fDylibsProcessed.count(it->second) == 0 ) - currentUnprocessedReaders.push_back(it->second); - } - for (std::vector::iterator it=currentUnprocessedReaders.begin(); it != currentUnprocessedReaders.end(); it++) { - fDylibsProcessed.insert(*it); - (*it)->processIndirectLibraries(this); - } - } - - // go back over original dylibs and mark sub frameworks as re-exported - if ( fOptions.outputKind() == Options::kDynamicLibrary ) { - const char* myLeaf = strrchr(fOptions.installPath(), '/'); - if ( myLeaf != NULL ) { - for (std::vector::iterator it=fInputFiles.begin(); it != fInputFiles.end(); it++) { - ObjectFile::Reader* reader = *it; - const char* childParent = reader->parentUmbrella(); - if ( childParent != NULL ) { - if ( strcmp(childParent, &myLeaf[1]) == 0 ) { - // set re-export bit of info - std::map::iterator pos = fDylibOptionsMap.find(reader); - if ( pos != fDylibOptionsMap.end() ) { - pos->second.fReExport = true; - } - } - } - } + + // All __thread_data and __thread_bss sections must have same alignment + uint8_t maxThreadAlign = 0; + for (ld::Internal::FinalSection* sect : sections) { + if ( (sect->type() == ld::Section::typeTLVInitialValues) || (sect->type() == ld::Section::typeTLVZeroFill) ) { + if ( sect->alignment > maxThreadAlign ) + maxThreadAlign = sect->alignment; } } - -} - - - -void Linker::createReaders() -{ - fStartCreateReadersTime = mach_absolute_time(); - std::vector& files = fOptions.getInputFiles(); - const int count = files.size(); - if ( count == 0 ) - throw "no object files specified"; - // add all direct object, archives, and dylibs - for (int i=0; i < count; ++i) { - Options::FileInfo& entry = files[i]; - // ignore /usr/lib/dyld on command line in crt.o build - if ( strcmp(entry.path, "/usr/lib/dyld") != 0 ) { - try { - this->addInputFile(this->createReader(entry), entry); - } - catch (const char* msg) { - if ( (strstr(msg, "architecture") != NULL) && !fOptions.errorOnOtherArchFiles() ) { - if ( fOptions.ignoreOtherArchInputFiles() ) { - // ignore, because this is about an architecture not in use - } - else { - warning("in %s, %s", entry.path, msg); - } - } - else { - throwf("in %s, %s", entry.path, msg); - } - } + for (ld::Internal::FinalSection* sect : sections) { + if ( (sect->type() == ld::Section::typeTLVInitialValues) || (sect->type() == ld::Section::typeTLVZeroFill) ) { + sect->alignment = maxThreadAlign; } } - this->processDylibs(); -} - - - -ObjectFile::Reader* Linker::addArchive(ObjectFile::Reader* reader, const Options::FileInfo& info, uint64_t mappedLen) -{ - fNextInputOrdinal += mappedLen; - // remember which readers are archives because they are logged differently - fArchiveReaders.insert(reader); - - // update stats - fTotalArchiveSize += mappedLen; - ++fTotalArchivesLoaded; - return reader; } -ObjectFile::Reader* Linker::addObject(ObjectFile::Reader* reader, const Options::FileInfo& info, uint64_t mappedLen) +uint64_t InternalState::assignFileOffsets() { - fNextInputOrdinal += mappedLen; - // any .o files that don't have MH_SUBSECTIONS_VIA_SYMBOLS, that means a generated .o file can't - if ( (fOptions.outputKind() == Options::kObjectFile) && !reader->canScatterAtoms() ) - fCanScatter = false; - - // update stats - fTotalObjectSize += mappedLen; - ++fTotalObjectLoaded; - return reader; -} + const bool log = false; + const bool hiddenSectionsOccupyAddressSpace = ((_options.outputKind() != Options::kObjectFile) + && (_options.outputKind() != Options::kPreload)); + const bool segmentsArePageAligned = (_options.outputKind() != Options::kObjectFile); - -void Linker::checkDylibClientRestrictions(ObjectFile::Reader* reader) -{ - // Check for any restrictions on who can link with this dylib - const char* readerParentName = reader->parentUmbrella() ; - std::vector* clients = reader->getAllowableClients(); - if ( (readerParentName != NULL) || (clients != NULL) ) { - // only dylibs that are in an umbrella or have a client list need verification - const char* installName = fOptions.installPath(); - const char* installNameLastSlash = strrchr(installName, '/'); - bool isParent = false; - bool isSibling = false; - bool isAllowableClient = false; - // There are three cases: - if ( (readerParentName != NULL) && (installNameLastSlash != NULL) ) { - // case 1) The dylib has a parent umbrella, and we are creating the parent umbrella - isParent = ( strcmp(&installNameLastSlash[1], readerParentName) == 0 ); - - // hack to support umbrella variants that encode the variant name in the install name - // e.g. CoreServices_profile - if ( !isParent ) { - const char* underscore = strchr(&installNameLastSlash[1], '_'); - if ( underscore != NULL ) { - isParent = ( strncmp(&installNameLastSlash[1], readerParentName, underscore-installNameLastSlash-1) == 0 ); - } + uint64_t address = 0; + const char* lastSegName = ""; + uint64_t floatingAddressStart = _options.baseAddress(); + bool haveFixedSegments = false; + + // mark all sections as not having an address yet + for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + sect->alignmentPaddingBytes = 0; + sect->address = ULLONG_MAX; + } + + // first pass, assign addresses to sections in segments with fixed start addresses + if ( log ) fprintf(stderr, "Fixed address segments:\n"); + for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + if ( ! _options.hasCustomSegmentAddress(sect->segmentName()) ) + continue; + haveFixedSegments = true; + if ( segmentsArePageAligned ) { + if ( strcmp(lastSegName, sect->segmentName()) != 0 ) { + address = _options.customSegmentAddress(sect->segmentName()); + lastSegName = sect->segmentName(); + } + } + // adjust section address based on alignment + uint64_t unalignedAddress = address; + uint64_t alignment = (1 << sect->alignment); + address = ( (unalignedAddress+alignment-1) & (-alignment) ); + + // update section info + sect->address = address; + sect->alignmentPaddingBytes = (address - unalignedAddress); + + // sanity check size + if ( ((address + sect->size) > _options.maxAddress()) && (_options.outputKind() != Options::kObjectFile) + && (_options.outputKind() != Options::kStaticExecutable) ) + throwf("section %s (address=0x%08llX, size=%llu) would make the output executable exceed available address range", + sect->sectionName(), address, sect->size); + + if ( log ) fprintf(stderr, " address=0x%08llX, hidden=%d, alignment=%02d, section=%s,%s\n", + sect->address, sect->isSectionHidden(), sect->alignment, sect->segmentName(), sect->sectionName()); + // update running totals + if ( !sect->isSectionHidden() || hiddenSectionsOccupyAddressSpace ) + address += sect->size; + + // if TEXT segment address is fixed, then flow other segments after it + if ( strcmp(sect->segmentName(), "__TEXT") == 0 ) { + floatingAddressStart = address; + } + } + + // second pass, assign section addresses to sections in segments that are ordered after a segment with a fixed address + if ( haveFixedSegments && !_options.segmentOrder().empty() ) { + if ( log ) fprintf(stderr, "After Fixed address segments:\n"); + lastSegName = ""; + ld::Internal::FinalSection* lastSect = NULL; + for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + if ( (sect->address == ULLONG_MAX) && _options.segmentOrderAfterFixedAddressSegment(sect->segmentName()) ) { + address = lastSect->address + lastSect->size; + if ( (strcmp(lastSegName, sect->segmentName()) != 0) && segmentsArePageAligned ) { + // round up size of last segment + address = pageAlign(address, _options.segPageSize(lastSegName)); + } + // adjust section address based on alignment + uint64_t unalignedAddress = address; + uint64_t alignment = (1 << sect->alignment); + address = ( (unalignedAddress+alignment-1) & (-alignment) ); + sect->alignmentPaddingBytes = (address - unalignedAddress); + sect->address = address; + if ( log ) fprintf(stderr, " address=0x%08llX, hidden=%d, alignment=%02d, section=%s,%s\n", + sect->address, sect->isSectionHidden(), sect->alignment, sect->segmentName(), sect->sectionName()); + // update running totals + if ( !sect->isSectionHidden() || hiddenSectionsOccupyAddressSpace ) + address += sect->size; + } + lastSegName = sect->segmentName(); + lastSect = sect; + } + } + + // last pass, assign addresses to remaining sections + address = floatingAddressStart; + lastSegName = ""; + ld::Internal::FinalSection* overlappingFixedSection = NULL; + ld::Internal::FinalSection* overlappingFlowSection = NULL; + ld::Internal::FinalSection* prevSect = NULL; + if ( log ) fprintf(stderr, "Regular layout segments:\n"); + for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + if ( sect->address != ULLONG_MAX ) + continue; + if ( (_options.outputKind() == Options::kPreload) && (sect->type() == ld::Section::typeMachHeader) ) { + sect->alignmentPaddingBytes = 0; + continue; + } + if ( segmentsArePageAligned ) { + if ( strcmp(lastSegName, sect->segmentName()) != 0 ) { + // round up size of last segment if needed + if ( *lastSegName != '\0' ) { + address = pageAlign(address, _options.segPageSize(lastSegName)); + } + // set segment address based on end of last segment + address = pageAlign(address); + lastSegName = sect->segmentName(); } - - // case 2) The dylib has a parent umbrella, and we are creating a sibling with the same parent - isSibling = ( (fOptions.umbrellaName() != NULL) && (strcmp(fOptions.umbrellaName(), readerParentName) == 0) ); } - - if ( !isParent && !isSibling && (clients != NULL) ) { - // case 3) the dylib has a list of allowable clients, and we are creating one of them - const char* clientName = fOptions.clientName(); - int clientNameLen = 0; - if ( clientName != NULL ) { - // use client name as specified on command line - clientNameLen = strlen(clientName); + + // adjust section address based on alignment + uint64_t unalignedAddress = address; + uint64_t alignment = (1 << sect->alignment); + address = ( (unalignedAddress+alignment-1) & (-alignment) ); + + // update section info + sect->address = address; + sect->alignmentPaddingBytes = (address - unalignedAddress); + + // if first section is more aligned than segment, move segment start up to match + if ( (prevSect != NULL) && (prevSect->type() == ld::Section::typeFirstSection) && (strcmp(prevSect->segmentName(), sect->segmentName()) == 0) ) { + assert(prevSect->size == 0); + if ( prevSect->address != sect->address ) { + prevSect->alignmentPaddingBytes += (sect->address - prevSect->address); + prevSect->address = sect->address; + } + } + + // sanity check size + if ( ((address + sect->size) > _options.maxAddress()) && (_options.outputKind() != Options::kObjectFile) + && (_options.outputKind() != Options::kStaticExecutable) ) + throwf("section %s (address=0x%08llX, size=%llu) would make the output executable exceed available address range", + sect->sectionName(), address, sect->size); + + // sanity check it does not overlap a fixed address segment + for (std::vector::iterator sit = sections.begin(); sit != sections.end(); ++sit) { + ld::Internal::FinalSection* otherSect = *sit; + if ( ! _options.hasCustomSegmentAddress(otherSect->segmentName()) ) + continue; + if ( otherSect->size == 0 ) + continue; + if ( sect->size == 0 ) + continue; + if ( sect->address > otherSect->address ) { + if ( (otherSect->address+otherSect->size) > sect->address ) { + overlappingFixedSection = otherSect; + overlappingFlowSection = sect; + } } else { - // infer client name from output path (e.g. xxx/libfoo_variant.A.dylib --> foo, Bar.framework/Bar_variant --> Bar) - clientName = installName; - clientNameLen = strlen(clientName); - // starts after last slash - if ( installNameLastSlash != NULL ) - clientName = &installNameLastSlash[1]; - if ( strncmp(clientName, "lib", 3) == 0 ) - clientName = &clientName[3]; - // up to first dot - const char* firstDot = strchr(clientName, '.'); - if ( firstDot != NULL ) - clientNameLen = firstDot - clientName; - // up to first underscore - const char* firstUnderscore = strchr(clientName, '_'); - if ( (firstUnderscore != NULL) && ((firstUnderscore - clientName) < clientNameLen) ) - clientNameLen = firstUnderscore - clientName; - } - - // Use clientName to check if this dylib is able to link against the allowable clients. - for (std::vector::iterator it = clients->begin(); it != clients->end(); it++) { - if ( strncmp(*it, clientName, clientNameLen) == 0 ) - isAllowableClient = true; + if ( (sect->address+sect->size) > otherSect->address ) { + overlappingFixedSection = otherSect; + overlappingFlowSection = sect; + } } } + + if ( log ) fprintf(stderr, " address=0x%08llX, size=0x%08llX, hidden=%d, alignment=%02d, padBytes=%d, section=%s,%s\n", + sect->address, sect->size, sect->isSectionHidden(), sect->alignment, sect->alignmentPaddingBytes, + sect->segmentName(), sect->sectionName()); + // update running totals + if ( !sect->isSectionHidden() || hiddenSectionsOccupyAddressSpace ) + address += sect->size; + prevSect = sect; + } + if ( overlappingFixedSection != NULL ) { + fprintf(stderr, "Section layout:\n"); + for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + //if ( sect->isSectionHidden() ) + // continue; + fprintf(stderr, " address:0x%08llX, alignment:2^%d, size:0x%08llX, padBytes:%d, section:%s/%s\n", + sect->address, sect->alignment, sect->size, sect->alignmentPaddingBytes, + sect->segmentName(), sect->sectionName()); - if ( !isParent && !isSibling && !isAllowableClient ) { - if ( readerParentName != NULL ) { - throwf("cannot link directly with %s. Link against the umbrella framework '%s.framework' instead.", - reader->getPath(), readerParentName); - } - else { - throwf("cannot link directly with %s", reader->getPath()); - } } + throwf("Section (%s/%s) overlaps fixed address section (%s/%s)", + overlappingFlowSection->segmentName(), overlappingFlowSection->sectionName(), + overlappingFixedSection->segmentName(), overlappingFixedSection->sectionName()); } -} - - -ObjectFile::Reader* Linker::addDylib(ObjectFile::Reader* reader, const Options::FileInfo& info, uint64_t mappedLen) -{ - switch ( fOptions.outputKind() ) { - case Options::kDynamicExecutable: - case Options::kDynamicLibrary: - case Options::kDynamicBundle: - break; - case Options::kStaticExecutable: - case Options::kDyld: - case Options::kPreload: - case Options::kObjectFile: - case Options::kKextBundle: - warning("unexpected dylib (%s) on link line", reader->getPath()); - break; - } - - fNextInputOrdinal += mappedLen; - if ( (reader->getInstallPath() == NULL) && !info.options.fBundleLoader ) { - // this is a "blank" stub - // silently ignore it - return reader; - } - // add to map of loaded dylibs - const char* installPath = reader->getInstallPath(); - if ( installPath != NULL ) { - InstallNameToReader::iterator pos = fDylibMap.find(installPath); - if ( pos == fDylibMap.end() ) { - fDylibMap[strdup(installPath)] = reader; + + + // third pass, assign section file offsets + uint64_t fileOffset = 0; + lastSegName = ""; + if ( log ) fprintf(stderr, "All segments with file offsets:\n"); + for (std::vector::iterator it = sections.begin(); it != sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + if ( hasZeroForFileOffset(sect) ) { + // fileoff of zerofill sections is moot, but historically it is set to zero + sect->fileOffset = 0; + + // align file offset with address layout + fileOffset += sect->alignmentPaddingBytes; } else { - InstallNameToReader::iterator pos2 = fDylibMap.find(reader->getPath()); - if ( pos2 == fDylibMap.end() ) - fDylibMap[strdup(reader->getPath())] = reader; - else - warning("duplicate dylib %s", reader->getPath()); - } - } - else if ( info.options.fBundleLoader ) - fBundleLoaderReader = reader; - - // log direct readers - if ( !fAllDirectDylibsLoaded ) - this->logDylib(reader, false); - - // update stats - ++fTotalDylibsLoaded; - - return reader; -} - - -void Linker::logTraceInfo (const char* format, ...) -{ - static int trace_file = -1; - char trace_buffer[MAXPATHLEN * 2]; - char *buffer_ptr; - int length; - ssize_t amount_written; - const char *trace_file_path = fOptions.readerOptions().fTraceOutputFile; + // page align file offset at start of each segment + if ( segmentsArePageAligned && (*lastSegName != '\0') && (strcmp(lastSegName, sect->segmentName()) != 0) ) { + fileOffset = pageAlign(fileOffset, _options.segPageSize(lastSegName)); + } + lastSegName = sect->segmentName(); - if(trace_file == -1) { - if(trace_file_path != NULL) { - trace_file = open(trace_file_path, O_WRONLY | O_APPEND | O_CREAT, 0666); - if(trace_file == -1) - throwf("Could not open or create trace file: %s", trace_file_path); - } - else { - trace_file = fileno(stderr); + // align file offset with address layout + fileOffset += sect->alignmentPaddingBytes; + + // update section info + sect->fileOffset = fileOffset; + + // update running total + fileOffset += sect->size; } + + if ( log ) fprintf(stderr, " fileoffset=0x%08llX, address=0x%08llX, hidden=%d, size=%lld, alignment=%02d, section=%s,%s\n", + sect->fileOffset, sect->address, sect->isSectionHidden(), sect->size, sect->alignment, + sect->segmentName(), sect->sectionName()); } - va_list ap; - va_start(ap, format); - length = vsnprintf(trace_buffer, sizeof(trace_buffer), format, ap); - va_end(ap); - buffer_ptr = trace_buffer; - - while(length > 0) { - amount_written = write(trace_file, buffer_ptr, length); - if(amount_written == -1) - /* Failure to write shouldn't fail the build. */ - return; - buffer_ptr += amount_written; - length -= amount_written; - } -} - - - -void Linker::createWriter() -{ - fStartCreateWriterTime = mach_absolute_time(); - - // make a vector out of all required dylibs in fDylibMap - std::vector dynamicLibraries; - // need to preserve command line order - for (std::vector::iterator it=fInputFiles.begin(); it != fInputFiles.end(); it++) { - ObjectFile::Reader* reader = *it; - for (InstallNameToReader::iterator mit=fDylibMap.begin(); mit != fDylibMap.end(); mit++) { - if ( reader == mit->second ) { - ExecutableFile::DyLibUsed dylibInfo; - dylibInfo.reader = reader; - dylibInfo.options = fDylibOptionsMap[reader]; - dynamicLibraries.push_back(dylibInfo); - break; - } - } - } - // then add any other dylibs - for (InstallNameToReader::iterator it=fDylibMap.begin(); it != fDylibMap.end(); it++) { - if ( it->second->implicitlyLinked() ) { - // if not already in dynamicLibraries - bool alreadyInDynamicLibraries = false; - for (std::vector::iterator dit=dynamicLibraries.begin(); dit != dynamicLibraries.end(); dit++) { - if ( dit->reader == it->second ) { - alreadyInDynamicLibraries = true; - break; - } - } - if ( ! alreadyInDynamicLibraries ) { - ExecutableFile::DyLibUsed dylibInfo; - dylibInfo.reader = it->second; - std::map::iterator pos = fDylibOptionsMap.find(it->second); - if ( pos != fDylibOptionsMap.end() ) { - dylibInfo.options = pos->second; - } - else { - dylibInfo.options.fWeakImport = false; // FIX ME - dylibInfo.options.fReExport = false; - dylibInfo.options.fBundleLoader = false; - } - dynamicLibraries.push_back(dylibInfo); +#if 0 + // for encrypted iPhoneOS apps + if ( _options.makeEncryptable() ) { + // remember end of __TEXT for later use by load command + for (std::vector::iterator it = state.sections.begin(); it != state.sections.end(); ++it) { + ld::Internal::FinalSection* sect = *it; + if ( strcmp(sect->segmentName(), "__TEXT") == 0 ) { + _encryptedTEXTendOffset = pageAlign(sect->fileOffset + sect->size); } } } - if ( fBundleLoaderReader != NULL ) { - ExecutableFile::DyLibUsed dylibInfo; - dylibInfo.reader = fBundleLoaderReader; - dylibInfo.options.fWeakImport = false; - dylibInfo.options.fReExport = false; - dylibInfo.options.fBundleLoader = true; - dynamicLibraries.push_back(dylibInfo); - } - - const char* path = fOptions.getOutputFilePath(); - switch ( fArchitecture ) { - case CPU_TYPE_POWERPC: - this->setOutputFile(new mach_o::executable::Writer(path, fOptions, dynamicLibraries)); - break; - case CPU_TYPE_POWERPC64: - this->setOutputFile(new mach_o::executable::Writer(path, fOptions, dynamicLibraries)); - break; - case CPU_TYPE_I386: - this->setOutputFile(new mach_o::executable::Writer(path, fOptions, dynamicLibraries)); - break; - case CPU_TYPE_X86_64: - this->setOutputFile(new mach_o::executable::Writer(path, fOptions, dynamicLibraries)); - break; - case CPU_TYPE_ARM: - this->setOutputFile(new mach_o::executable::Writer(path, fOptions, dynamicLibraries)); - break; - default: - throw "unknown architecture"; - } -} - +#endif -Linker::SymbolTable::SymbolTable(Linker& owner) - : fOwner(owner), fRequireCount(0), fHasExternalTentativeDefinitions(false), fHasExternalWeakDefinitions(false) -{ + // return total file size + return fileOffset; } -void Linker::SymbolTable::require(const char* name) +static char* commatize(uint64_t in, char* out) { - //fprintf(stderr, "require(%s)\n", name); - Mapper::iterator pos = fTable.find(name); - if ( pos == fTable.end() ) { - fTable[name] = NULL; - ++fRequireCount; + char* result = out; + char rawNum[30]; + sprintf(rawNum, "%llu", in); + const int rawNumLen = strlen(rawNum); + for(int i=0; i < rawNumLen-1; ++i) { + *out++ = rawNum[i]; + if ( ((rawNumLen-i) % 3) == 1 ) + *out++ = ','; } + *out++ = rawNum[rawNumLen-1]; + *out = '\0'; + return result; } -// convenience labels for 2-dimensional switch statement -enum AllDefinitionCombinations { - kRegAndReg = (ObjectFile::Atom::kRegularDefinition << 3) | ObjectFile::Atom::kRegularDefinition, - kRegAndWeak = (ObjectFile::Atom::kRegularDefinition << 3) | ObjectFile::Atom::kWeakDefinition, - kRegAndTent = (ObjectFile::Atom::kRegularDefinition << 3) | ObjectFile::Atom::kTentativeDefinition, - kRegAndExtern = (ObjectFile::Atom::kRegularDefinition << 3) | ObjectFile::Atom::kExternalDefinition, - kRegAndExternWeak = (ObjectFile::Atom::kRegularDefinition << 3) | ObjectFile::Atom::kExternalWeakDefinition, - kRegAndAbsolute = (ObjectFile::Atom::kRegularDefinition << 3) | ObjectFile::Atom::kAbsoluteSymbol, - kWeakAndReg = (ObjectFile::Atom::kWeakDefinition << 3) | ObjectFile::Atom::kRegularDefinition, - kWeakAndWeak = (ObjectFile::Atom::kWeakDefinition << 3) | ObjectFile::Atom::kWeakDefinition, - kWeakAndTent = (ObjectFile::Atom::kWeakDefinition << 3) | ObjectFile::Atom::kTentativeDefinition, - kWeakAndExtern = (ObjectFile::Atom::kWeakDefinition << 3) | ObjectFile::Atom::kExternalDefinition, - kWeakAndExternWeak = (ObjectFile::Atom::kWeakDefinition << 3) | ObjectFile::Atom::kExternalWeakDefinition, - kWeakAndAbsolute = (ObjectFile::Atom::kWeakDefinition << 3) | ObjectFile::Atom::kAbsoluteSymbol, - kTentAndReg = (ObjectFile::Atom::kTentativeDefinition << 3) | ObjectFile::Atom::kRegularDefinition, - kTentAndWeak = (ObjectFile::Atom::kTentativeDefinition << 3) | ObjectFile::Atom::kWeakDefinition, - kTentAndTent = (ObjectFile::Atom::kTentativeDefinition << 3) | ObjectFile::Atom::kTentativeDefinition, - kTentAndExtern = (ObjectFile::Atom::kTentativeDefinition << 3) | ObjectFile::Atom::kExternalDefinition, - kTentAndExternWeak = (ObjectFile::Atom::kTentativeDefinition << 3) | ObjectFile::Atom::kExternalWeakDefinition, - kTentAndAbsolute = (ObjectFile::Atom::kTentativeDefinition << 3) | ObjectFile::Atom::kAbsoluteSymbol, - kExternAndReg = (ObjectFile::Atom::kExternalDefinition << 3) | ObjectFile::Atom::kRegularDefinition, - kExternAndWeak = (ObjectFile::Atom::kExternalDefinition << 3) | ObjectFile::Atom::kWeakDefinition, - kExternAndTent = (ObjectFile::Atom::kExternalDefinition << 3) | ObjectFile::Atom::kTentativeDefinition, - kExternAndExtern = (ObjectFile::Atom::kExternalDefinition << 3) | ObjectFile::Atom::kExternalDefinition, - kExternAndExternWeak = (ObjectFile::Atom::kExternalDefinition << 3) | ObjectFile::Atom::kExternalWeakDefinition, - kExternAndAbsolute = (ObjectFile::Atom::kExternalDefinition << 3) | ObjectFile::Atom::kAbsoluteSymbol, - kExternWeakAndReg = (ObjectFile::Atom::kExternalWeakDefinition << 3) | ObjectFile::Atom::kRegularDefinition, - kExternWeakAndWeak = (ObjectFile::Atom::kExternalWeakDefinition << 3) | ObjectFile::Atom::kWeakDefinition, - kExternWeakAndTent = (ObjectFile::Atom::kExternalWeakDefinition << 3) | ObjectFile::Atom::kTentativeDefinition, - kExternWeakAndExtern = (ObjectFile::Atom::kExternalWeakDefinition << 3) | ObjectFile::Atom::kExternalDefinition, - kExternWeakAndExternWeak= (ObjectFile::Atom::kExternalWeakDefinition << 3) | ObjectFile::Atom::kExternalWeakDefinition, - kExternWeakAndAbsolute = (ObjectFile::Atom::kExternalWeakDefinition << 3) | ObjectFile::Atom::kAbsoluteSymbol, - kAbsoluteAndReg = (ObjectFile::Atom::kAbsoluteSymbol << 3) | ObjectFile::Atom::kRegularDefinition, - kAbsoluteAndWeak = (ObjectFile::Atom::kAbsoluteSymbol << 3) | ObjectFile::Atom::kWeakDefinition, - kAbsoluteAndTent = (ObjectFile::Atom::kAbsoluteSymbol << 3) | ObjectFile::Atom::kTentativeDefinition, - kAbsoluteAndExtern = (ObjectFile::Atom::kAbsoluteSymbol << 3) | ObjectFile::Atom::kExternalDefinition, - kAbsoluteAndExternWeak = (ObjectFile::Atom::kAbsoluteSymbol << 3) | ObjectFile::Atom::kExternalWeakDefinition, - kAbsoluteAndAbsolute = (ObjectFile::Atom::kAbsoluteSymbol << 3) | ObjectFile::Atom::kAbsoluteSymbol -}; - -bool Linker::SymbolTable::add(ObjectFile::Atom& newAtom) +static void printTime(const char* msg, uint64_t partTime, uint64_t totalTime) { - bool useNew = true; - bool checkVisibilityMismatch = false; - const char* name = newAtom.getName(); - //fprintf(stderr, "map.add(%s => %p from %s)\n", name, &newAtom, newAtom.getFile()->getPath()); - Mapper::iterator pos = fTable.find(name); - ObjectFile::Atom* existingAtom = NULL; - if ( pos != fTable.end() ) - existingAtom = pos->second; - if ( existingAtom != NULL ) { - // already have atom with same name in symbol table - switch ( (AllDefinitionCombinations)((existingAtom->getDefinitionKind() << 3) | newAtom.getDefinitionKind()) ) { - case kRegAndReg: - throwf("duplicate symbol %s in %s and %s", name, newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - case kRegAndWeak: - // ignore new weak atom, because we already have a non-weak one - useNew = false; - break; - case kRegAndTent: - // ignore new tentative atom, because we already have a regular one - useNew = false; - checkVisibilityMismatch = true; - if ( newAtom.getSize() > existingAtom->getSize() ) { - warning("for symbol %s tentative definition of size %llu from %s is " - "is smaller than the real definition of size %llu from %s", - newAtom.getDisplayName(), newAtom.getSize(), newAtom.getFile()->getPath(), - existingAtom->getSize(), existingAtom->getFile()->getPath()); - } - break; - case kRegAndExtern: - // ignore external atom, because we already have a one - useNew = false; - break; - case kRegAndExternWeak: - // ignore external atom, because we already have a one - useNew = false; - break; - case kRegAndAbsolute: - throwf("duplicate symbol %s in %s and %s", name, newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - break; - case kWeakAndReg: - // replace existing weak atom with regular one - break; - case kWeakAndWeak: - // have another weak atom, use whichever has largest alignment requirement - // because codegen of some client may require alignment - useNew = ( newAtom.getAlignment().trailingZeros() > existingAtom->getAlignment().trailingZeros() ); - checkVisibilityMismatch = true; - break; - case kWeakAndTent: - // replace existing weak atom with tentative one ??? - break; - case kWeakAndExtern: - // keep weak atom, at runtime external one may override - useNew = false; - break; - case kWeakAndExternWeak: - // keep weak atom, at runtime external one may override - useNew = false; - break; - case kWeakAndAbsolute: - // replace existing weak atom with absolute one - break; - case kTentAndReg: - // replace existing tentative atom with regular one - checkVisibilityMismatch = true; - if ( newAtom.getSize() < existingAtom->getSize() ) { - warning("for symbol %s tentative definition of size %llu from %s is " - "being replaced by a real definition of size %llu from %s", - newAtom.getDisplayName(), existingAtom->getSize(), existingAtom->getFile()->getPath(), - newAtom.getSize(), newAtom.getFile()->getPath()); - } - break; - case kTentAndWeak: - // replace existing tentative atom with weak one ??? - break; - case kTentAndTent: - // use largest - checkVisibilityMismatch = true; - if ( newAtom.getSize() < existingAtom->getSize() ) { - useNew = false; - } - else { - if ( newAtom.getAlignment().trailingZeros() < existingAtom->getAlignment().trailingZeros() ) - warning("alignment lost in merging tentative definition %s", newAtom.getDisplayName()); - } - break; - case kTentAndExtern: - case kTentAndExternWeak: - // a tentative definition and a dylib definition, so commons-mode decides how to handle - switch ( fOwner.fOptions.commonsMode() ) { - case Options::kCommonsIgnoreDylibs: - if ( fOwner.fOptions.warnCommons() ) - warning("using common symbol %s from %s and ignoring defintion from dylib %s", - existingAtom->getName(), existingAtom->getFile()->getPath(), newAtom.getFile()->getPath()); - useNew = false; - break; - case Options::kCommonsOverriddenByDylibs: - if ( fOwner.fOptions.warnCommons() ) - warning("replacing common symbol %s from %s with true definition from dylib %s", - existingAtom->getName(), existingAtom->getFile()->getPath(), newAtom.getFile()->getPath()); - break; - case Options::kCommonsConflictsDylibsError: - throwf("common symbol %s from %s conflicts with defintion from dylib %s", - existingAtom->getName(), existingAtom->getFile()->getPath(), newAtom.getFile()->getPath()); - } - break; - case kTentAndAbsolute: - // replace tentative with absolute (can't size check because absolutes have no size) - break; - case kExternAndReg: - // replace external atom with regular one - break; - case kExternAndWeak: - // replace external atom with weak one - break; - case kExternAndTent: - // a tentative definition and a dylib definition, so commons-mode decides how to handle - switch ( fOwner.fOptions.commonsMode() ) { - case Options::kCommonsIgnoreDylibs: - if ( fOwner.fOptions.warnCommons() ) - warning("using common symbol %s from %s and ignoring defintion from dylib %s", - newAtom.getName(), newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - break; - case Options::kCommonsOverriddenByDylibs: - if ( fOwner.fOptions.warnCommons() ) - warning("replacing defintion of %s from dylib %s with common symbol from %s", - newAtom.getName(), existingAtom->getFile()->getPath(), newAtom.getFile()->getPath()); - useNew = false; - break; - case Options::kCommonsConflictsDylibsError: - throwf("common symbol %s from %s conflicts with defintion from dylib %s", - newAtom.getName(), newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - } - break; - case kExternAndExtern: - throwf("duplicate symbol %s in %s and %s\n", name, newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - case kExternAndExternWeak: - // keep strong dylib atom, ignore weak one - useNew = false; - break; - case kExternAndAbsolute: - // replace external atom with absolute one - break; - case kExternWeakAndReg: - // replace existing weak external with regular - break; - case kExternWeakAndWeak: - // replace existing weak external with weak (let dyld decide at runtime which to use) - break; - case kExternWeakAndTent: - // a tentative definition and a dylib definition, so commons-mode decides how to handle - switch ( fOwner.fOptions.commonsMode() ) { - case Options::kCommonsIgnoreDylibs: - if ( fOwner.fOptions.warnCommons() ) - warning("using common symbol %s from %s and ignoring defintion from dylib %s", - newAtom.getName(), newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - break; - case Options::kCommonsOverriddenByDylibs: - if ( fOwner.fOptions.warnCommons() ) - warning("replacing defintion of %s from dylib %s with common symbol from %s", - newAtom.getName(), existingAtom->getFile()->getPath(), newAtom.getFile()->getPath()); - useNew = false; - break; - case Options::kCommonsConflictsDylibsError: - throwf("common symbol %s from %s conflicts with defintion from dylib %s", - newAtom.getName(), newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - } - break; - case kExternWeakAndExtern: - // replace existing weak external with external - break; - case kExternWeakAndExternWeak: - // keep existing external weak - useNew = false; - break; - case kExternWeakAndAbsolute: - // replace existing weak external with absolute - break; - case kAbsoluteAndReg: - throwf("duplicate symbol %s in %s and %s", name, newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - case kAbsoluteAndWeak: - // ignore new weak atom, because we already have a non-weak one - useNew = false; - break; - case kAbsoluteAndTent: - // ignore new tentative atom, because we already have a regular one - useNew = false; - break; - case kAbsoluteAndExtern: - // ignore external atom, because we already have a one - useNew = false; - break; - case kAbsoluteAndExternWeak: - // ignore external atom, because we already have a one - useNew = false; - break; - case kAbsoluteAndAbsolute: - throwf("duplicate symbol %s in %s and %s", name, newAtom.getFile()->getPath(), existingAtom->getFile()->getPath()); - break; - } - } - if ( (existingAtom != NULL) && checkVisibilityMismatch && (newAtom.getScope() != existingAtom->getScope()) ) { - warning("%s has different visibility (%s) in %s and (%s) in %s", - newAtom.getDisplayName(), (newAtom.getScope() == 1 ? "hidden" : "default"), newAtom.getFile()->getPath(), (existingAtom->getScope() == 1 ? "hidden" : "default"), existingAtom->getFile()->getPath()); + static uint64_t sUnitsPerSecond = 0; + if ( sUnitsPerSecond == 0 ) { + struct mach_timebase_info timeBaseInfo; + if ( mach_timebase_info(&timeBaseInfo) != KERN_SUCCESS ) + return; + sUnitsPerSecond = 1000000000ULL * timeBaseInfo.denom / timeBaseInfo.numer; } - if ( useNew ) { - fTable[name] = &newAtom; - if ( existingAtom != NULL ) { - fOwner.markDead(existingAtom); - if ( fOwner.fInitialLoadsDone ) { - //fprintf(stderr, "existing %p %s overridden by %p\n", existingAtom, existingAtom->getName(), &newAtom); - fOwner.fAtomsOverriddenByLateLoads.insert(existingAtom); - } - } - if ( newAtom.getScope() == ObjectFile::Atom::scopeGlobal ) { - switch ( newAtom.getDefinitionKind() ) { - case ObjectFile::Atom::kTentativeDefinition: - fHasExternalTentativeDefinitions = true; - ++fRequireCount; // added a tentative definition means loadUndefines() needs to continue - break; - case ObjectFile::Atom::kWeakDefinition: - fHasExternalWeakDefinitions = true; - break; - default: - break; - } - } + if ( partTime < sUnitsPerSecond ) { + uint32_t milliSecondsTimeTen = (partTime*10000)/sUnitsPerSecond; + uint32_t milliSeconds = milliSecondsTimeTen/10; + uint32_t percentTimesTen = (partTime*1000)/totalTime; + uint32_t percent = percentTimesTen/10; + fprintf(stderr, "%24s: % 4d.%d milliseconds (% 4d.%d%%)\n", msg, milliSeconds, milliSecondsTimeTen-milliSeconds*10, percent, percentTimesTen-percent*10); } else { - fOwner.markDead(&newAtom); - } - return useNew; -} - - - -ObjectFile::Atom* Linker::SymbolTable::find(const char* name) -{ - Mapper::iterator pos = fTable.find(name); - if ( pos != fTable.end() ) { - return pos->second; + uint32_t secondsTimeTen = (partTime*10)/sUnitsPerSecond; + uint32_t seconds = secondsTimeTen/10; + uint32_t percentTimesTen = (partTime*1000)/totalTime; + uint32_t percent = percentTimesTen/10; + fprintf(stderr, "%24s: % 4d.%d seconds (% 4d.%d%%)\n", msg, seconds, secondsTimeTen-seconds*10, percent, percentTimesTen-percent*10); } - return NULL; } -void Linker::SymbolTable::getUndefinesNames(std::vector& undefines) +static void getVMInfo(vm_statistics_data_t& info) { - for (Mapper::iterator it=fTable.begin(); it != fTable.end(); it++) { - if ( it->second == NULL ) { - undefines.push_back(it->first); - } + mach_msg_type_number_t count = sizeof(vm_statistics_data_t) / sizeof(natural_t); + kern_return_t error = host_statistics(mach_host_self(), HOST_VM_INFO, + (host_info_t)&info, &count); + if (error != KERN_SUCCESS) { + bzero(&info, sizeof(vm_statistics_data_t)); } } -void Linker::SymbolTable::getTentativesNames(std::vector& tents) -{ - for (Mapper::iterator it=fTable.begin(); it != fTable.end(); it++) { - if ( it->second != NULL ) { - if ( (it->second->getDefinitionKind() == ObjectFile::Atom::kTentativeDefinition) - && (it->second->getScope() == ObjectFile::Atom::scopeGlobal) ) { - tents.push_back(it->first); - } - } - } -} +static const char* sOverridePathlibLTO = NULL; -bool Linker::AtomSorter::operator()(const ObjectFile::Atom* left, const ObjectFile::Atom* right) +// +// This is magic glue that overrides the default behaviour +// of lazydylib1.o which is used to lazily load libLTO.dylib. +// +extern "C" const char* dyld_lazy_dylib_path_fix(const char* path); +const char* dyld_lazy_dylib_path_fix(const char* path) { - if ( left == right ) - return false; - - // first sort by section order (which is already sorted by segment) - unsigned int leftSectionIndex = left->getSection()->getIndex(); - unsigned int rightSectionIndex = right->getSection()->getIndex(); - if ( leftSectionIndex != rightSectionIndex) - return (leftSectionIndex < rightSectionIndex); - - // magic section$start symbol always sorts to the start of its section - if ( left->getContentType() == ObjectFile::Atom::kSectionStart ) - return true; - if ( right->getContentType() == ObjectFile::Atom::kSectionStart ) - return false; - - // if a -order_file is specified, then sorting is altered to sort those symbols first - if ( fOverriddenOrdinalMap != NULL ) { - std::map::iterator leftPos = fOverriddenOrdinalMap->find(left); - std::map::iterator rightPos = fOverriddenOrdinalMap->find(right); - std::map::iterator end = fOverriddenOrdinalMap->end(); - if ( leftPos != end ) { - if ( rightPos != end ) { - // both left and right are overridden, so compare overridden ordinals - return leftPos->second < rightPos->second; - } - else { - // left is overridden and right is not, so left < right - return true; - } - } - else { - if ( rightPos != end ) { - // right is overridden and left is not, so right < left - return false; - } - else { - // neither are overridden, do default sort - // fall into default sorting below - } - } - } - - // the __common section can have real or tentative definitions - // we want the real ones to sort before tentative ones - bool leftIsTent = (left->getDefinitionKind() == ObjectFile::Atom::kTentativeDefinition); - bool rightIsTent = (right->getDefinitionKind() == ObjectFile::Atom::kTentativeDefinition); - if ( leftIsTent != rightIsTent ) - return rightIsTent; - - // magic section$end symbol always sorts to the end of its section - if ( left->getContentType() == ObjectFile::Atom::kSectionEnd ) - return false; - if ( right->getContentType() == ObjectFile::Atom::kSectionEnd ) - return true; - - // initializers are auto sorted to start of section - if ( !fInitializerSet.empty() ) { - bool leftFirst = (fInitializerSet.count(left) != 0); - bool rightFirst = (fInitializerSet.count(right) != 0); - if ( leftFirst != rightFirst ) - return leftFirst; - } - - // terminators are auto sorted to end of section - if ( !fTerminatorSet.empty() ) { - bool leftLast = (fTerminatorSet.count(left) != 0); - bool rightLast = (fTerminatorSet.count(right) != 0); - if ( leftLast != rightLast ) - return rightLast; - } - - // lastly sort by atom ordinal. this is already sorted by .o order - return left->getOrdinal() < right->getOrdinal(); + if ( sOverridePathlibLTO != NULL ) + return sOverridePathlibLTO; + else + return path; } + int main(int argc, const char* argv[]) { const char* archName = NULL; bool showArch = false; bool archInferred = false; try { - // create linker object given command line arguments - Linker ld(argc, argv); - - // save error message prefix - archName = ld.architectureName(); - archInferred = ld.isInferredArchitecture(); - showArch = ld.showArchitectureInErrors(); - - // open all input files - ld.createReaders(); - - // open output file - ld.createWriter(); - - // do linking - ld.link(); + PerformanceStatistics statistics; + statistics.startTool = mach_absolute_time(); + + // create object to track command line arguments + Options options(argc, argv); + InternalState state(options); + + // allow libLTO to be overridden by command line -lto_library + sOverridePathlibLTO = options.overridePathlibLTO(); + + // gather vm stats + if ( options.printStatistics() ) + getVMInfo(statistics.vmStart); + + // update strings for error messages + showArch = options.printArchPrefix(); + archName = options.architectureName(); + archInferred = (options.architecture() == 0); + + // open and parse input files + statistics.startInputFileProcessing = mach_absolute_time(); + ld::tool::InputFiles inputFiles(options, &archName); + + // load and resolve all references + statistics.startResolver = mach_absolute_time(); + ld::tool::Resolver resolver(options, inputFiles, state); + resolver.resolve(); + + // add dylibs used + statistics.startDylibs = mach_absolute_time(); + inputFiles.dylibs(state); + + // do initial section sorting so passes have rough idea of the layout + state.sortSections(); + + // run passes + statistics.startPasses = mach_absolute_time(); + ld::passes::objc::doPass(options, state); + ld::passes::stubs::doPass(options, state); + ld::passes::huge::doPass(options, state); + ld::passes::got::doPass(options, state); + ld::passes::tlvp::doPass(options, state); + ld::passes::dylibs::doPass(options, state); // must be after stubs and GOT passes + ld::passes::order::doPass(options, state); + state.markAtomsOrdered(); + ld::passes::dedup::doPass(options, state); + ld::passes::branch_shim::doPass(options, state); // must be after stubs + ld::passes::branch_island::doPass(options, state); // must be after stubs and order pass + ld::passes::dtrace::doPass(options, state); + ld::passes::compact_unwind::doPass(options, state); // must be after order pass + ld::passes::bitcode_bundle::doPass(options, state); // must be after dylib + + // sort final sections + state.sortSections(); + + // write output file + statistics.startOutput = mach_absolute_time(); + ld::tool::OutputFile out(options); + out.write(state); + statistics.startDone = mach_absolute_time(); + + // print statistics + //mach_o::relocatable::printCounts(); + if ( options.printStatistics() ) { + getVMInfo(statistics.vmEnd); + uint64_t totalTime = statistics.startDone - statistics.startTool; + printTime("ld total time", totalTime, totalTime); + printTime(" option parsing time", statistics.startInputFileProcessing - statistics.startTool, totalTime); + printTime(" object file processing", statistics.startResolver - statistics.startInputFileProcessing,totalTime); + printTime(" resolve symbols", statistics.startDylibs - statistics.startResolver, totalTime); + printTime(" build atom list", statistics.startPasses - statistics.startDylibs, totalTime); + printTime(" passess", statistics.startOutput - statistics.startPasses, totalTime); + printTime(" write output", statistics.startDone - statistics.startOutput, totalTime); + fprintf(stderr, "pageins=%u, pageouts=%u, faults=%u\n", + statistics.vmEnd.pageins-statistics.vmStart.pageins, + statistics.vmEnd.pageouts-statistics.vmStart.pageouts, + statistics.vmEnd.faults-statistics.vmStart.faults); + char temp[40]; + fprintf(stderr, "processed %3u object files, totaling %15s bytes\n", inputFiles._totalObjectLoaded, commatize(inputFiles._totalObjectSize, temp)); + fprintf(stderr, "processed %3u archive files, totaling %15s bytes\n", inputFiles._totalArchivesLoaded, commatize(inputFiles._totalArchiveSize, temp)); + fprintf(stderr, "processed %3u dylib files\n", inputFiles._totalDylibsLoaded); + fprintf(stderr, "wrote output file totaling %15s bytes\n", commatize(out.fileSize(), temp)); + } + // Would like linker warning to be build error. + if ( options.errorBecauseOfWarnings() ) { + fprintf(stderr, "ld: fatal warning(s) induced error (-fatal_warnings)\n"); + return 1; + } } catch (const char* msg) { - if ( archInferred ) + if ( strstr(msg, "malformed") != NULL ) + fprintf(stderr, "ld: %s\n", msg); + else if ( archInferred ) fprintf(stderr, "ld: %s for inferred architecture %s\n", msg, archName); else if ( showArch ) fprintf(stderr, "ld: %s for architecture %s\n", msg, archName); @@ -4099,3 +1399,40 @@ int main(int argc, const char* argv[]) return 0; } + + +#ifndef NDEBUG +// implement assert() function to print out a backtrace before aborting +void __assert_rtn(const char* func, const char* file, int line, const char* failedexpr) +{ + Snapshot *snapshot = Snapshot::globalSnapshot; + + snapshot->setSnapshotMode(Snapshot::SNAPSHOT_DEBUG); + snapshot->createSnapshot(); + snapshot->recordAssertionMessage("Assertion failed: (%s), function %s, file %s, line %d.\n", failedexpr, func, file, line); + + void* callStack[128]; + int depth = ::backtrace(callStack, 128); + char* buffer = (char*)malloc(1024); + for(int i=0; i < depth-1; ++i) { + Dl_info info; + dladdr(callStack[i], &info); + const char* symboName = info.dli_sname; + if ( (symboName != NULL) && (strncmp(symboName, "_Z", 2) == 0) ) { + size_t bufLen = 1024; + int result; + char* unmangled = abi::__cxa_demangle(symboName, buffer, &bufLen, &result); + if ( unmangled != NULL ) + symboName = unmangled; + } + long offset = (uintptr_t)callStack[i] - (uintptr_t)info.dli_saddr; + fprintf(stderr, "%d %p %s + %ld\n", i, callStack[i], symboName, offset); + snapshot->recordAssertionMessage("%d %p %s + %ld\n", i, callStack[i], symboName, offset); + } + fprintf(stderr, "A linker snapshot was created at:\n\t%s\n", snapshot->rootDir()); + fprintf(stderr, "ld: Assertion failed: (%s), function %s, file %s, line %d.\n", failedexpr, func, file, line); + exit(1); +} +#endif + +