Loading...
--- dyld/dyld-239.3/launch-cache/dsc_extractor.cpp
+++ dyld/dyld-360.21/launch-cache/dsc_extractor.cpp
@@ -45,8 +45,10 @@
#include "dsc_iterator.h"
#include "dsc_extractor.h"
+#include "MachOTrie.hpp"
#include <vector>
+#include <set>
#include <map>
#include <unordered_map>
#include <algorithm>
@@ -76,9 +78,36 @@
};
typedef std::unordered_map<const char*, std::vector<seg_info>, CStringHash, CStringEquals> NameToSegments;
+// Filter to find individual symbol re-exports in trie
+class NotReExportSymbol {
+public:
+ NotReExportSymbol(const std::set<int> &rd) :_reexportDeps(rd) {}
+ bool operator()(const mach_o::trie::Entry &entry) const {
+ bool result = isSymbolReExport(entry);
+ if (result) {
+ // <rdar://problem/17671438> Xcode 6 leaks in dyld_shared_cache_extract_dylibs
+ ::free((void*)entry.name);
+ const_cast<mach_o::trie::Entry*>(&entry)->name = NULL;
+ }
+ return result;
+ }
+private:
+ bool isSymbolReExport(const mach_o::trie::Entry &entry) const {
+ if ( (entry.flags & EXPORT_SYMBOL_FLAGS_KIND_MASK) != EXPORT_SYMBOL_FLAGS_KIND_REGULAR )
+ return true;
+ if ( (entry.flags & EXPORT_SYMBOL_FLAGS_REEXPORT) == 0 )
+ return true;
+ // If the symbol comes from a dylib that is re-exported, this is not an individual symbol re-export
+ if ( _reexportDeps.count(entry.other) != 0 )
+ return true;
+ return false;
+ }
+ const std::set<int> &_reexportDeps;
+};
+
template <typename A>
-int optimize_linkedit(macho_header<typename A::P>* mh, uint32_t textOffsetInCache, const void* mapped_cache, uint64_t* newSize)
+int optimize_linkedit(macho_header<typename A::P>* mh, uint64_t textOffsetInCache, const void* mapped_cache, uint64_t* newSize)
{
typedef typename A::P P;
typedef typename A::P::E E;
@@ -97,8 +126,14 @@
macho_dysymtab_command<P>* dynamicSymTab = NULL;
macho_linkedit_data_command<P>* functionStarts = NULL;
macho_linkedit_data_command<P>* dataInCode = NULL;
+ uint32_t exportsTrieOffset = 0;
+ uint32_t exportsTrieSize = 0;
+ std::set<int> reexportDeps;
+ int depIndex = 0;
for (uint32_t i = 0; i < cmd_count; ++i) {
- if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
+ switch ( cmd->cmd() ) {
+ case macho_segment_command<P>::CMD:
+ {
// update segment/section file offsets
macho_segment_command<P>* segCmd = (macho_segment_command<P>*)cmd;
segCmd->set_fileoff(cumulativeFileSize);
@@ -106,16 +141,20 @@
macho_section<P>* const sectionsEnd = §ionsStart[segCmd->nsects()];
for(macho_section<P>* sect = sectionsStart; sect < sectionsEnd; ++sect) {
if ( sect->offset() != 0 )
- sect->set_offset(cumulativeFileSize+sect->addr()-segCmd->vmaddr());
+ sect->set_offset((uint32_t)(cumulativeFileSize+sect->addr()-segCmd->vmaddr()));
}
if ( strcmp(segCmd->segname(), "__LINKEDIT") == 0 ) {
linkEditSegCmd = segCmd;
}
cumulativeFileSize += segCmd->filesize();
- }
- else if ( cmd->cmd() == LC_DYLD_INFO_ONLY ) {
+ }
+ break;
+ case LC_DYLD_INFO_ONLY:
+ {
// zero out all dyld info
macho_dyld_info_command<P>* dyldInfo = (macho_dyld_info_command<P>*)cmd;
+ exportsTrieOffset = dyldInfo->export_off();
+ exportsTrieSize = dyldInfo->export_size();
dyldInfo->set_rebase_off(0);
dyldInfo->set_rebase_size(0);
dyldInfo->set_bind_off(0);
@@ -126,18 +165,29 @@
dyldInfo->set_lazy_bind_size(0);
dyldInfo->set_export_off(0);
dyldInfo->set_export_size(0);
- }
- else if ( cmd->cmd() == LC_SYMTAB ) {
+ }
+ break;
+ case LC_SYMTAB:
symtab = (macho_symtab_command<P>*)cmd;
- }
- else if ( cmd->cmd() == LC_DYSYMTAB ) {
+ break;
+ case LC_DYSYMTAB:
dynamicSymTab = (macho_dysymtab_command<P>*)cmd;
- }
- else if ( cmd->cmd() == LC_FUNCTION_STARTS ) {
+ break;
+ case LC_FUNCTION_STARTS:
functionStarts = (macho_linkedit_data_command<P>*)cmd;
- }
- else if ( cmd->cmd() == LC_DATA_IN_CODE ) {
+ break;
+ case LC_DATA_IN_CODE:
dataInCode = (macho_linkedit_data_command<P>*)cmd;
+ break;
+ case LC_LOAD_DYLIB:
+ case LC_LOAD_WEAK_DYLIB:
+ case LC_REEXPORT_DYLIB:
+ case LC_LOAD_UPWARD_DYLIB:
+ ++depIndex;
+ if ( cmd->cmd() == LC_REEXPORT_DYLIB ) {
+ reexportDeps.insert(depIndex);
+ }
+ break;
}
cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
}
@@ -156,26 +206,35 @@
return -1;
}
- const uint32_t newFunctionStartsOffset = linkEditSegCmd->fileoff();
+ const uint64_t newFunctionStartsOffset = linkEditSegCmd->fileoff();
uint32_t functionStartsSize = 0;
if ( functionStarts != NULL ) {
// copy function starts from original cache file to new mapped dylib file
functionStartsSize = functionStarts->datasize();
memcpy((char*)mh + newFunctionStartsOffset, (char*)mapped_cache + functionStarts->dataoff(), functionStartsSize);
}
- const uint32_t newDataInCodeOffset = (newFunctionStartsOffset + functionStartsSize + sizeof(pint_t) - 1) & (-sizeof(pint_t)); // pointer align
+ const uint64_t newDataInCodeOffset = (newFunctionStartsOffset + functionStartsSize + sizeof(pint_t) - 1) & (-sizeof(pint_t)); // pointer align
uint32_t dataInCodeSize = 0;
if ( dataInCode != NULL ) {
// copy data-in-code info from original cache file to new mapped dylib file
dataInCodeSize = dataInCode->datasize();
memcpy((char*)mh + newDataInCodeOffset, (char*)mapped_cache + dataInCode->dataoff(), dataInCodeSize);
}
-
+
+ std::vector<mach_o::trie::Entry> exports;
+ if ( exportsTrieSize != 0 ) {
+ const uint8_t* exportsStart = ((uint8_t*)mapped_cache) + exportsTrieOffset;
+ const uint8_t* exportsEnd = &exportsStart[exportsTrieSize];
+ mach_o::trie::parseTrie(exportsStart, exportsEnd, exports);
+ exports.erase(std::remove_if(exports.begin(), exports.end(), NotReExportSymbol(reexportDeps)), exports.end());
+ }
+
// look for local symbol info in unmapped part of shared cache
dyldCacheHeader<E>* header = (dyldCacheHeader<E>*)mapped_cache;
macho_nlist<P>* localNlists = NULL;
uint32_t localNlistCount = 0;
const char* localStrings = NULL;
+ const char* localStringsEnd = NULL;
if ( header->mappingOffset() > offsetof(dyld_cache_header,localSymbolsSize) ) {
dyldCacheLocalSymbolsInfo<E>* localInfo = (dyldCacheLocalSymbolsInfo<E>*)(((uint8_t*)mapped_cache) + header->localSymbolsOffset());
dyldCacheLocalSymbolEntry<E>* entries = (dyldCacheLocalSymbolEntry<E>*)(((uint8_t*)mapped_cache) + header->localSymbolsOffset() + localInfo->entriesOffset());
@@ -187,11 +246,11 @@
localNlistCount = entries[i].nlistCount();
localNlists = &allLocalNlists[localNlistStart];
localStrings = ((char*)localInfo) + localInfo->stringsOffset();
+ localStringsEnd = &localStrings[localInfo->stringsSize()];
break;
}
}
}
-
// compute number of symbols in new symbol table
const macho_nlist<P>* const mergedSymTabStart = (macho_nlist<P>*)(((uint8_t*)mapped_cache) + symtab->symoff());
const macho_nlist<P>* const mergedSymTabend = &mergedSymTabStart[symtab->nsyms()];
@@ -206,14 +265,18 @@
}
}
+ // add room for N_INDR symbols for re-exported symbols
+ newSymCount += exports.size();
+
// copy symbol entries and strings from original cache file to new mapped dylib file
- const uint32_t newSymTabOffset = (newDataInCodeOffset + dataInCodeSize + sizeof(pint_t) - 1) & (-sizeof(pint_t)); // pointer align
- const uint32_t newIndSymTabOffset = newSymTabOffset + newSymCount*sizeof(macho_nlist<P>);
- const uint32_t newStringPoolOffset = newIndSymTabOffset + dynamicSymTab->nindirectsyms()*sizeof(uint32_t);
+ const uint64_t newSymTabOffset = (newDataInCodeOffset + dataInCodeSize + sizeof(pint_t) - 1) & (-sizeof(pint_t)); // pointer align
+ const uint64_t newIndSymTabOffset = newSymTabOffset + newSymCount*sizeof(macho_nlist<P>);
+ const uint64_t newStringPoolOffset = newIndSymTabOffset + dynamicSymTab->nindirectsyms()*sizeof(uint32_t);
macho_nlist<P>* const newSymTabStart = (macho_nlist<P>*)(((uint8_t*)mh) + newSymTabOffset);
char* const newStringPoolStart = (char*)mh + newStringPoolOffset;
const uint32_t* mergedIndSymTab = (uint32_t*)((char*)mapped_cache + dynamicSymTab->indirectsymoff());
const char* mergedStringPoolStart = (char*)mapped_cache + symtab->stroff();
+ const char* mergedStringPoolEnd = &mergedStringPoolStart[symtab->strsize()];
macho_nlist<P>* t = newSymTabStart;
int poolOffset = 0;
uint32_t symbolsCopied = 0;
@@ -224,8 +287,28 @@
continue;
*t = *s;
t->set_n_strx(poolOffset);
- strcpy(&newStringPoolStart[poolOffset], &mergedStringPoolStart[s->n_strx()]);
- poolOffset += (strlen(&newStringPoolStart[poolOffset]) + 1);
+ const char* symName = &mergedStringPoolStart[s->n_strx()];
+ if ( symName > mergedStringPoolEnd )
+ symName = "<corrupt symbol name>";
+ strcpy(&newStringPoolStart[poolOffset], symName);
+ poolOffset += (strlen(symName) + 1);
+ ++t;
+ ++symbolsCopied;
+ }
+ // <rdar://problem/16529213> recreate N_INDR symbols in extracted dylibs for debugger
+ for (std::vector<mach_o::trie::Entry>::iterator it = exports.begin(); it != exports.end(); ++it) {
+ strcpy(&newStringPoolStart[poolOffset], it->name);
+ t->set_n_strx(poolOffset);
+ poolOffset += (strlen(it->name) + 1);
+ t->set_n_type(N_INDR | N_EXT);
+ t->set_n_sect(0);
+ t->set_n_desc(0);
+ const char* importName = it->importName;
+ if ( *importName == '\0' )
+ importName = it->name;
+ strcpy(&newStringPoolStart[poolOffset], importName);
+ t->set_n_value(poolOffset);
+ poolOffset += (strlen(importName) + 1);
++t;
++symbolsCopied;
}
@@ -236,6 +319,8 @@
// copy local symbols
for (uint32_t i=0; i < localNlistCount; ++i) {
const char* localName = &localStrings[localNlists[i].n_strx()];
+ if ( localName > localStringsEnd )
+ localName = "<corrupt local symbol name>";
*t = localNlists[i];
t->set_n_strx(poolOffset);
strcpy(&newStringPoolStart[poolOffset], localName);
@@ -259,27 +344,33 @@
// update load commands
if ( functionStarts != NULL ) {
- functionStarts->set_dataoff(newFunctionStartsOffset);
+ functionStarts->set_dataoff((uint32_t)newFunctionStartsOffset);
functionStarts->set_datasize(functionStartsSize);
}
if ( dataInCode != NULL ) {
- dataInCode->set_dataoff(newDataInCodeOffset);
+ dataInCode->set_dataoff((uint32_t)newDataInCodeOffset);
dataInCode->set_datasize(dataInCodeSize);
}
symtab->set_nsyms(symbolsCopied);
- symtab->set_symoff(newSymTabOffset);
- symtab->set_stroff(newStringPoolOffset);
+ symtab->set_symoff((uint32_t)newSymTabOffset);
+ symtab->set_stroff((uint32_t)newStringPoolOffset);
symtab->set_strsize(poolOffset);
dynamicSymTab->set_extreloff(0);
dynamicSymTab->set_nextrel(0);
dynamicSymTab->set_locreloff(0);
dynamicSymTab->set_nlocrel(0);
- dynamicSymTab->set_indirectsymoff(newIndSymTabOffset);
+ dynamicSymTab->set_indirectsymoff((uint32_t)newIndSymTabOffset);
linkEditSegCmd->set_filesize(symtab->stroff()+symtab->strsize() - linkEditSegCmd->fileoff());
linkEditSegCmd->set_vmsize( (linkEditSegCmd->filesize()+4095) & (-4096) );
// return new size
*newSize = (symtab->stroff()+symtab->strsize()+4095) & (-4096);
+
+ // <rdar://problem/17671438> Xcode 6 leaks in dyld_shared_cache_extract_dylibs
+ for (std::vector<mach_o::trie::Entry>::iterator it = exports.begin(); it != exports.end(); ++it) {
+ ::free((void*)(it->name));
+ }
+
return 0;
}
@@ -348,8 +439,8 @@
FA->align = OSSwapHostToBigInt32(12);
// Write regular segments into the buffer
- uint32_t totalSize = 0;
- uint32_t textOffsetInCache = 0;
+ uint64_t totalSize = 0;
+ uint64_t textOffsetInCache = 0;
for( std::vector<seg_info>::const_iterator it=segments.begin(); it != segments.end(); ++it) {
if(strcmp(it->segName, "__TEXT") == 0 ) {
@@ -364,7 +455,7 @@
if( afa->cputype == FA->cputype
&& afa->cpusubtype == FA->cpusubtype) {
- fprintf(stderr, "arch already exists in fat dylib\n");
+ //fprintf(stderr, "arch already exists in fat dylib\n");
dylib_data.resize(offsetInFatFile);
return offsetInFatFile;
}
@@ -422,6 +513,8 @@
dylib_create_func = dylib_maker<x86>;
else if ( strcmp((char*)mapped_cache, "dyld_v1 x86_64") == 0 )
dylib_create_func = dylib_maker<x86_64>;
+ else if ( strcmp((char*)mapped_cache, "dyld_v1 x86_64h") == 0 )
+ dylib_create_func = dylib_maker<x86_64>;
else if ( strcmp((char*)mapped_cache, "dyld_v1 armv5") == 0 )
dylib_create_func = dylib_maker<arm>;
else if ( strcmp((char*)mapped_cache, "dyld_v1 armv6") == 0 )
@@ -430,6 +523,8 @@
dylib_create_func = dylib_maker<arm>;
else if ( strncmp((char*)mapped_cache, "dyld_v1 armv7", 14) == 0 )
dylib_create_func = dylib_maker<arm>;
+ else if ( strcmp((char*)mapped_cache, "dyld_v1 arm64") == 0 )
+ dylib_create_func = dylib_maker<arm64>;
else {
fprintf(stderr, "Error: unrecognized dyld shared cache magic.\n");
munmap(mapped_cache, statbuf.st_size);
@@ -438,7 +533,7 @@
// iterate through all images in cache and build map of dylibs and segments
__block NameToSegments map;
- __block int result = dyld_shared_cache_iterate(mapped_cache, statbuf.st_size, ^(const dyld_shared_cache_dylib_info* dylibInfo, const dyld_shared_cache_segment_info* segInfo) {
+ __block int result = dyld_shared_cache_iterate(mapped_cache, (uint32_t)statbuf.st_size, ^(const dyld_shared_cache_dylib_info* dylibInfo, const dyld_shared_cache_segment_info* segInfo) {
map[dylibInfo->path].push_back(seg_info(segInfo->name, segInfo->fileOffset, segInfo->fileSize));
});
@@ -454,11 +549,10 @@
dispatch_queue_t process_queue = dispatch_get_global_queue(DISPATCH_QUEUE_PRIORITY_LOW, 0);
dispatch_queue_t writer_queue = dispatch_queue_create("dyld writer queue", 0);
- __block int cumulativeResult = 0;
__block unsigned count = 0;
for ( NameToSegments::iterator it = map.begin(); it != map.end(); ++it) {
- dispatch_semaphore_wait(sema, DISPATCH_TIME_FOREVER);
+ dispatch_semaphore_wait(sema, DISPATCH_TIME_FOREVER);
dispatch_group_async(group, process_queue, ^{
char dylib_path[PATH_MAX];
@@ -474,7 +568,7 @@
int fd = ::open(dylib_path, O_CREAT | O_EXLOCK | O_RDWR, 0644);
if ( fd == -1 ) {
fprintf(stderr, "can't open or create dylib file %s, errnor=%d\n", dylib_path, errno);
- cumulativeResult = -1;
+ result = -1;
return;
}
@@ -482,7 +576,7 @@
if (fstat(fd, &statbuf)) {
fprintf(stderr, "Error: stat failed for dyld file %s, errnor=%d\n", dylib_path, errno);
close(fd);
- cumulativeResult = -1;
+ result = -1;
return;
}
@@ -490,21 +584,21 @@
if(pread(fd, &vec->front(), vec->size(), 0) != (long)vec->size()) {
fprintf(stderr, "can't read dylib file %s, errnor=%d\n", dylib_path, errno);
close(fd);
- cumulativeResult = -1;
+ result = -1;
return;
}
const size_t offset = dylib_create_func(mapped_cache, *vec, it->second);
dispatch_group_async(group, writer_queue, ^{
- progress(count++, map.size());
+ progress(count++, (unsigned)map.size());
if(offset != vec->size()) {
//Write out the first page, and everything after offset
if( pwrite(fd, &vec->front(), 4096, 0) == -1
|| pwrite(fd, &vec->front() + offset, vec->size() - offset, offset) == -1) {
fprintf(stderr, "error writing, errnor=%d\n", errno);
- cumulativeResult = -1;
+ result = -1;
}
}
@@ -520,7 +614,7 @@
dispatch_release(writer_queue);
munmap(mapped_cache, statbuf.st_size);
- return cumulativeResult;
+ return result;
}