Loading...
--- dyld/dyld-1340/common/MachOAnalyzer.cpp
+++ dyld/dyld-955/common/MachOAnalyzer.cpp
@@ -21,46 +21,34 @@
* @APPLE_LICENSE_HEADER_END@
*/
-#include <TargetConditionals.h>
-#include "Defines.h"
-#if !TARGET_OS_EXCLAVEKIT
- #include <sys/types.h>
- #include <sys/fcntl.h>
- #include <sys/mman.h>
- #include <mach/mach.h>
- #include <unistd.h>
-#endif
-
-#if SUPPORT_CLASSIC_RELOCS
- #include <mach-o/reloc.h>
- #include <mach-o/x86_64/reloc.h>
-#endif
-
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include <sys/fcntl.h>
+#include <sys/mman.h>
+#include <mach/mach.h>
#include <assert.h>
#include <limits.h>
#include <stdlib.h>
#include <string.h>
+#include <unistd.h>
+#include <mach-o/reloc.h>
+#include <mach-o/x86_64/reloc.h>
#include <mach-o/nlist.h>
-#include <ptrauth.h>
+#include <TargetConditionals.h>
#include "MachOAnalyzer.h"
#include "CodeSigningTypes.h"
#include "Array.h"
-#include "Header.h"
-#include "Universal.h"
// FIXME: We should get this from cctools
#define DYLD_CACHE_ADJ_V2_FORMAT 0x7F
-using mach_o::Header;
-using mach_o::Platform;
-using mach_o::GradedArchitectures;
-using mach_o::Universal;
-
namespace dyld3 {
+
bool MachOAnalyzer::isValidMainExecutable(Diagnostics& diag, const char* path, uint64_t sliceLength,
- const GradedArchitectures& archs, Platform platform) const
+ const GradedArchs& archs, Platform platform) const
{
if ( !this->validMachOForArchAndPlatform(diag, (size_t)sliceLength, path, archs, platform, true) )
return false;
@@ -76,30 +64,28 @@
return true;
}
-#if !TARGET_OS_EXCLAVEKIT
bool MachOAnalyzer::loadFromBuffer(Diagnostics& diag, const closure::FileSystem& fileSystem,
- const char* path, const GradedArchitectures& archs, Platform platform,
+ const char* path, const GradedArchs& archs, Platform platform,
closure::LoadedFileInfo& info)
{
// if fat, remap just slice needed
- bool fatButMissingSlice = false;
- std::span<const uint8_t> content = {(const uint8_t*)info.fileContent, (size_t)info.fileContentLen};
- if ( const Universal* uni = Universal::isUniversal(content) ) {
- Universal::Slice slice;
- if ( uni->bestSlice(archs, info.isOSBinary, slice) ) {
- uint64_t sliceOffset = slice.buffer.data() - content.data();
- // unmap anything before slice
- fileSystem.unloadPartialFile(info, sliceOffset, slice.buffer.size());
- // Update the info to keep track of the new slice offset.
- info.sliceOffset = sliceOffset;
- info.sliceLen = slice.buffer.size();
- }
- else {
- fatButMissingSlice = true;
- }
- }
-
- if ( fatButMissingSlice ) {
+ bool fatButMissingSlice;
+ const FatFile* fh = (FatFile*)info.fileContent;
+ uint64_t sliceOffset = info.sliceOffset;
+ uint64_t sliceLen = info.sliceLen;
+ if ( fh->isFatFileWithSlice(diag, info.fileContentLen, archs, info.isOSBinary, sliceOffset, sliceLen, fatButMissingSlice) ) {
+ // unmap anything before slice
+ fileSystem.unloadPartialFile(info, sliceOffset, sliceLen);
+ // Update the info to keep track of the new slice offset.
+ info.sliceOffset = sliceOffset;
+ info.sliceLen = sliceLen;
+ }
+ else if ( diag.hasError() ) {
+ // We must have generated an error in the fat file parsing so use that error
+ fileSystem.unloadFile(info);
+ return false;
+ }
+ else if ( fatButMissingSlice ) {
diag.error("missing compatible arch in %s", path);
fileSystem.unloadFile(info);
return false;
@@ -114,8 +100,7 @@
}
// if has zero-fill expansion, re-map
- if ( !mh->isPreload() )
- mh = mh->remapIfZeroFill(diag, fileSystem, info);
+ mh = mh->remapIfZeroFill(diag, fileSystem, info);
// on error, remove mappings and return nullptr
if ( diag.hasError() ) {
@@ -124,8 +109,7 @@
}
// now that LINKEDIT is at expected offset, finish validation
- if ( !mh->isPreload() )
- mh->validLinkedit(diag, path);
+ mh->validLinkedit(diag, path);
// on error, remove mappings and return nullptr
if ( diag.hasError() ) {
@@ -138,7 +122,7 @@
closure::LoadedFileInfo MachOAnalyzer::load(Diagnostics& diag, const closure::FileSystem& fileSystem,
- const char* path, const GradedArchitectures& archs, Platform platform, char realerPath[PATH_MAX])
+ const char* path, const GradedArchs& archs, Platform platform, char realerPath[MAXPATHLEN])
{
// FIXME: This should probably be an assert, but if we happen to have a diagnostic here then something is wrong
// above us and we should quickly return instead of doing unnecessary work.
@@ -146,14 +130,12 @@
return closure::LoadedFileInfo();
closure::LoadedFileInfo info;
- void (^fileErrorLog)(const char *format, ...) __printflike(1, 2)
- = ^(const char *format, ...) __printflike(1, 2) {
- va_list list;
- va_start(list, format);
- diag.error(format, va_list_wrap(list));
- va_end(list);
- };
- if ( !fileSystem.loadFile(path, info, realerPath, fileErrorLog) ) {
+ if (!fileSystem.loadFile(path, info, realerPath, ^(const char *format, ...) {
+ va_list list;
+ va_start(list, format);
+ diag.error(format, list);
+ va_end(list);
+ })) {
return closure::LoadedFileInfo();
}
@@ -168,7 +150,6 @@
return info;
}
-
// for use with already mmap()ed file
bool MachOAnalyzer::isOSBinary(int fd, uint64_t sliceOffset, uint64_t sliceSize) const
{
@@ -178,7 +159,7 @@
uint32_t sigOffset;
uint32_t sigSize;
- if ( !((const Header*)this)->hasCodeSignature(sigOffset, sigSize) )
+ if ( !this->hasCodeSignature(sigOffset, sigSize) )
return false;
// register code signature
@@ -217,7 +198,6 @@
return result;
}
-#endif // !TARGET_OS_EXCLAVEKIT
@@ -230,7 +210,15 @@
}
#endif
-bool MachOAnalyzer::validMachOForArchAndPlatform(Diagnostics& diag, size_t sliceLength, const char* path, const GradedArchitectures& archs, Platform reqPlatform, bool isOSBinary, bool internalInstall) const
+uint64_t MachOAnalyzer::mappedSize() const
+{
+ uint64_t vmSpace;
+ bool hasZeroFill;
+ analyzeSegmentsLayout(vmSpace, hasZeroFill);
+ return vmSpace;
+}
+
+bool MachOAnalyzer::validMachOForArchAndPlatform(Diagnostics& diag, size_t sliceLength, const char* path, const GradedArchs& archs, Platform reqPlatform, bool isOSBinary) const
{
// must start with mach-o magic value
if ( (this->magic != MH_MAGIC) && (this->magic != MH_MAGIC_64) ) {
@@ -238,7 +226,7 @@
return false;
}
- if ( !archs.isCompatible(((const Header*)this)->arch(), isOSBinary) ) {
+ if ( !archs.grade(this->cputype, this->cpusubtype, isOSBinary) ) {
diag.error("could not use '%s' because it is not a compatible arch", path);
return false;
}
@@ -250,11 +238,10 @@
case MH_BUNDLE:
case MH_DYLINKER:
break;
-#if BUILDING_DYLDINFO || BUILDING_APP_CACHE_UTIL
+#if BUILDING_DYLDINFO || BUILDING_APP_CACHE_UTIL || BUILDING_RUN_STATIC
// Allow offline tools to analyze binaries dyld doesn't load
case MH_KEXT_BUNDLE:
case MH_FILESET:
- case MH_PRELOAD:
break;
#endif
default:
@@ -280,23 +267,44 @@
#if BUILDING_DYLDINFO || BUILDING_APP_CACHE_UTIL
if ( isFileSet() ) {
// A statically linked kernel collection should contain a 0 platform
- mach_o::PlatformAndVersions pvs = ((const Header*)this)->platformAndVersions();
- if ( !pvs.platform.empty() ) {
+ __block bool foundPlatform = false;
+ __block bool foundBadPlatform = false;
+ forEachSupportedPlatform(^(Platform platform, uint32_t minOS, uint32_t sdk) {
+ foundPlatform = true;
+ if ( platform != Platform::unknown ) {
+ foundBadPlatform = true;
+ }
+ });
+ if (!foundPlatform) {
+ diag.error("could not use '%s' because we expected it to have a platform", path);
+ return false;
+ }
+ if (foundBadPlatform) {
diag.error("could not use '%s' because is has the wrong platform", path);
return false;
}
- } else if ( reqPlatform.empty() ) {
- // This is handled elsewhere in the kernel collection builder, where we have access
- // to the kernel binary and can infer its platform
+ } else if ( reqPlatform == Platform::unknown ) {
+ // Unfortunately the static kernel has a platform, but kext's don't, so we can't
+ // verify the platform of the kernel.
+ if ( !isStaticExecutable() ) {
+ __block bool foundPlatform = false;
+ forEachSupportedPlatform(^(Platform platform, uint32_t minOS, uint32_t sdk) {
+ foundPlatform = true;
+ });
+ if (foundPlatform) {
+ diag.error("could not use '%s' because we expected it to have no platform", path);
+ return false;
+ }
+ }
} else
#endif
- if ( !((const Header*)this)->loadableIntoProcess(reqPlatform, path, internalInstall) ) {
- diag.error("could not use '%s' because it was not built for platform %s", path, reqPlatform.name().c_str());
+ if ( !this->loadableIntoProcess(reqPlatform, path) ) {
+ diag.error("could not use '%s' because it was not built for platform %s", path, MachOFile::platformName(reqPlatform));
return false;
}
// validate dylib loads
- if ( !validEmbeddedPaths(diag, reqPlatform, path, internalInstall) )
+ if ( !validEmbeddedPaths(diag, reqPlatform, path) )
return false;
// validate segments
@@ -363,7 +371,7 @@
Diagnostics walkDiag;
forEachLoadCommand(walkDiag, ^(const load_command* cmd, bool& stop) {});
if ( walkDiag.hasError() ) {
-#if BUILDING_CACHE_BUILDER || BUILDING_UNIT_TESTS || BUILDING_CACHE_BUILDER_UNIT_TESTS
+#if BUILDING_CACHE_BUILDER
diag.error("in '%s' %s", path, walkDiag.errorMessage().c_str());
#else
diag.error("in '%s' %s", path, walkDiag.errorMessage());
@@ -373,13 +381,13 @@
// check load commands fit in TEXT segment
__block bool foundTEXT = false;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stop) {
- if ( info.segmentName == "__TEXT" ) {
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ if ( strcmp(info.segName, "__TEXT") == 0 ) {
foundTEXT = true;
if ( this->sizeofcmds + machHeaderSize() > info.fileSize ) {
diag.error("in '%s' load commands exceed length of __TEXT segment", path);
}
- if ( (info.fileOffset != 0) && !this->isPreload() ) {
+ if ( info.fileOffset != 0 ) {
diag.error("in '%s' __TEXT segment not start of mach-o", path);
}
stop = true;
@@ -393,7 +401,6 @@
return true;
}
-#if !TARGET_OS_EXCLAVEKIT
const MachOAnalyzer* MachOAnalyzer::remapIfZeroFill(Diagnostics& diag, const closure::FileSystem& fileSystem, closure::LoadedFileInfo& info) const
{
uint64_t vmSpaceRequired;
@@ -411,17 +418,17 @@
#if BUILDING_APP_CACHE_UTIL
// The auxKC is mapped with __DATA first, so we need to get either the __DATA or __TEXT depending on what is earliest
__block uint64_t baseAddress = ~0ULL;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& segmentInfo, bool& stop) {
- baseAddress = std::min(baseAddress, segmentInfo.vmaddr);
+ forEachSegment(^(const SegmentInfo& segmentInfo, bool& stop) {
+ baseAddress = std::min(baseAddress, segmentInfo.vmAddr);
});
- uint64_t textSegVMAddr = ((const Header*)this)->preferredLoadAddress();
+ uint64_t textSegVMAddr = preferredLoadAddress();
#else
- uint64_t baseAddress = ((const Header*)this)->preferredLoadAddress();
+ uint64_t baseAddress = preferredLoadAddress();
#endif
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& segmentInfo, bool& stop) {
- if ( (segmentInfo.fileSize != 0) && (segmentInfo.vmsize != 0) ) {
- kern_return_t r = vm_copy(mach_task_self(), (vm_address_t)((long)info.fileContent+segmentInfo.fileOffset), (vm_size_t)segmentInfo.fileSize, (vm_address_t)(newMappedAddr+segmentInfo.vmaddr-baseAddress));
+ forEachSegment(^(const SegmentInfo& segmentInfo, bool& stop) {
+ if ( (segmentInfo.fileSize != 0) && (segmentInfo.vmSize != 0) ) {
+ kern_return_t r = vm_copy(mach_task_self(), (vm_address_t)((long)info.fileContent+segmentInfo.fileOffset), (vm_size_t)segmentInfo.fileSize, (vm_address_t)(newMappedAddr+segmentInfo.vmAddr-baseAddress));
if ( r != KERN_SUCCESS ) {
diag.error("vm_copy() failure");
stop = true;
@@ -440,12 +447,12 @@
info.unload = [](const closure::LoadedFileInfo& info) {
// Unloading binaries where __DATA is first requires working out the real range of the binary
// The fileContent points at the mach_header, not the actaul start of the file content, unfortunately.
- const Header* hdr = (const Header*)info.fileContent;
+ const MachOAnalyzer* ma = (const MachOAnalyzer*)info.fileContent;
__block uint64_t baseAddress = ~0ULL;
- hdr->forEachSegment(^(const Header::SegmentInfo& segInfo, bool& stop) {
- baseAddress = std::min(baseAddress, segInfo.vmaddr);
+ ma->forEachSegment(^(const SegmentInfo& segInfo, bool& stop) {
+ baseAddress = std::min(baseAddress, segInfo.vmAddr);
});
- uint64_t textSegVMAddr = hdr->preferredLoadAddress();
+ uint64_t textSegVMAddr = ma->preferredLoadAddress();
uint64_t basePointerOffset = textSegVMAddr - baseAddress;
uint8_t* bufferStart = (uint8_t*)info.fileContent - basePointerOffset;
@@ -478,9 +485,222 @@
return this;
}
-#endif // !TARGET_OS_EXCLAVEKIT
-
-bool MachOAnalyzer::validEmbeddedPaths(Diagnostics& diag, Platform platform, const char* path, bool internalInstall) const
+
+void MachOAnalyzer::analyzeSegmentsLayout(uint64_t& vmSpace, bool& hasZeroFill) const
+{
+ __block bool writeExpansion = false;
+ __block uint64_t lowestVmAddr = 0xFFFFFFFFFFFFFFFFULL;
+ __block uint64_t highestVmAddr = 0;
+ __block uint64_t sumVmSizes = 0;
+ forEachSegment(^(const SegmentInfo& segmentInfo, bool& stop) {
+ if ( strcmp(segmentInfo.segName, "__PAGEZERO") == 0 )
+ return;
+ if ( segmentInfo.writable() && (segmentInfo.fileSize != segmentInfo.vmSize) )
+ writeExpansion = true; // zerofill at end of __DATA
+ if ( segmentInfo.vmSize == 0 ) {
+ // Always zero fill if we have zero-sized segments
+ writeExpansion = true;
+ }
+ if ( segmentInfo.vmAddr < lowestVmAddr )
+ lowestVmAddr = segmentInfo.vmAddr;
+ if ( segmentInfo.vmAddr+segmentInfo.vmSize > highestVmAddr )
+ highestVmAddr = segmentInfo.vmAddr+segmentInfo.vmSize;
+ sumVmSizes += segmentInfo.vmSize;
+ });
+ uint64_t totalVmSpace = (highestVmAddr - lowestVmAddr);
+ // LINKEDIT vmSize is not required to be a multiple of page size. Round up if that is the case
+ const uint64_t pageSize = uses16KPages() ? 0x4000 : 0x1000;
+ totalVmSpace = (totalVmSpace + (pageSize - 1)) & ~(pageSize - 1);
+ bool hasHole = (totalVmSpace != sumVmSizes); // segments not contiguous
+
+ // The aux KC may have __DATA first, in which case we always want to vm_copy to the right place
+ bool hasOutOfOrderSegments = false;
+#if BUILDING_APP_CACHE_UTIL
+ uint64_t textSegVMAddr = preferredLoadAddress();
+ hasOutOfOrderSegments = textSegVMAddr != lowestVmAddr;
+#endif
+
+ vmSpace = totalVmSpace;
+ hasZeroFill = writeExpansion || hasHole || hasOutOfOrderSegments;
+}
+
+bool MachOAnalyzer::enforceFormat(Malformed kind) const
+{
+ // TODO: Add a mapping from generic releases to platform versions
+#if BUILDING_DYLDINFO || BUILDING_APP_CACHE_UTIL || BUILDING_RUN_STATIC
+ // HACK: If we are the kernel, we have a different format to enforce
+ if ( isFileSet() ) {
+ bool result = false;
+ switch (kind) {
+ case Malformed::linkeditOrder:
+ case Malformed::linkeditAlignment:
+ case Malformed::dyldInfoAndlocalRelocs:
+ result = true;
+ break;
+ case Malformed::segmentOrder:
+ // The aux KC has __DATA first
+ result = false;
+ break;
+ case Malformed::linkeditPermissions:
+ case Malformed::executableData:
+ case Malformed::writableData:
+ case Malformed::codeSigAlignment:
+ case Malformed::sectionsAddrRangeWithinSegment:
+ case Malformed::loaderPathsAreReal:
+ case Malformed::mainExecInDyldCache:
+ result = true;
+ break;
+ case Malformed::noLinkedDylibs:
+ case Malformed::textPermissions:
+ // The kernel has its own __TEXT_EXEC for executable memory
+ result = false;
+ break;
+ case Malformed::noUUID:
+ result = true;
+ break;
+ }
+ return result;
+ }
+
+ if ( isStaticExecutable() ) {
+ bool result = false;
+ switch (kind) {
+ case Malformed::linkeditOrder:
+ case Malformed::linkeditAlignment:
+ case Malformed::dyldInfoAndlocalRelocs:
+ result = true;
+ break;
+ case Malformed::segmentOrder:
+ result = false;
+ break;
+ case Malformed::linkeditPermissions:
+ case Malformed::executableData:
+ case Malformed::codeSigAlignment:
+ case Malformed::textPermissions:
+ case Malformed::sectionsAddrRangeWithinSegment:
+ case Malformed::loaderPathsAreReal:
+ case Malformed::mainExecInDyldCache:
+ result = true;
+ break;
+ case Malformed::noLinkedDylibs:
+ case Malformed::writableData:
+ case Malformed::noUUID:
+ // The kernel has __DATA_CONST marked as r/o
+ result = false;
+ break;
+ }
+ return result;
+ }
+
+#endif
+
+ __block bool result = false;
+ forEachSupportedPlatform(^(Platform platform, uint32_t minOS, uint32_t sdk) {
+ switch (platform) {
+ case Platform::macOS:
+ switch (kind) {
+ case Malformed::linkeditOrder:
+ case Malformed::linkeditAlignment:
+ case Malformed::dyldInfoAndlocalRelocs:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000A0E00) // macOS 10.14
+ result = true;
+ break;
+ case Malformed::segmentOrder:
+ case Malformed::linkeditPermissions:
+ case Malformed::textPermissions:
+ case Malformed::executableData:
+ case Malformed::writableData:
+ case Malformed::codeSigAlignment:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000A0F00) // macOS 10.15
+ result = true;
+ break;
+ case Malformed::sectionsAddrRangeWithinSegment:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000A1000) // macOS 10.16
+ result = true;
+ break;
+ case Malformed::noLinkedDylibs:
+ case Malformed::loaderPathsAreReal:
+ case Malformed::mainExecInDyldCache:
+ case Malformed::noUUID:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000C0000) // macOS 12.0
+ result = true;
+ break;
+ }
+ break;
+ case Platform::iOS:
+ case Platform::tvOS:
+ case Platform::iOSMac:
+ switch (kind) {
+ case Malformed::linkeditOrder:
+ case Malformed::dyldInfoAndlocalRelocs:
+ case Malformed::textPermissions:
+ case Malformed::executableData:
+ case Malformed::writableData:
+ result = true;
+ break;
+ case Malformed::linkeditAlignment:
+ case Malformed::segmentOrder:
+ case Malformed::linkeditPermissions:
+ case Malformed::codeSigAlignment:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000D0000) // iOS 13
+ result = true;
+ break;
+ case Malformed::sectionsAddrRangeWithinSegment:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000E0000) // iOS 14
+ result = true;
+ break;
+ case Malformed::noLinkedDylibs:
+ case Malformed::loaderPathsAreReal:
+ case Malformed::mainExecInDyldCache:
+ case Malformed::noUUID:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x000F0000) // iOS 15
+ result = true;
+ }
+ break;
+ case Platform::watchOS:
+ switch (kind) {
+ case Malformed::linkeditOrder:
+ case Malformed::dyldInfoAndlocalRelocs:
+ case Malformed::textPermissions:
+ case Malformed::executableData:
+ case Malformed::writableData:
+ result = true;
+ break;
+ case Malformed::linkeditAlignment:
+ case Malformed::segmentOrder:
+ case Malformed::linkeditPermissions:
+ case Malformed::codeSigAlignment:
+ case Malformed::sectionsAddrRangeWithinSegment:
+ case Malformed::noLinkedDylibs:
+ case Malformed::loaderPathsAreReal:
+ case Malformed::mainExecInDyldCache:
+ case Malformed::noUUID:
+ // enforce these checks on new binaries only
+ if (sdk >= 0x00080000) // watchOS 8
+ result = true;
+ break;
+ }
+ break;
+ case Platform::driverKit:
+ result = true;
+ break;
+ default:
+ result = true;
+ break;
+ }
+ });
+ // if binary is so old, there is no platform info, don't enforce malformed errors
+ return result;
+}
+
+bool MachOAnalyzer::validEmbeddedPaths(Diagnostics& diag, Platform platform, const char* path) const
{
__block int index = 1;
__block bool allGood = true;
@@ -583,23 +803,10 @@
// all new binaries must link with something else
if ( (dependentsCount == 0) && enforceFormat(Malformed::noLinkedDylibs) ) {
- const Header *hdr = (const Header*)this;
- const char* libSystemDir = hdr->builtForPlatform(Platform::driverKit, true) ? "/System/DriverKit/usr/lib/system/" : "/usr/lib/system/";
+ const char* libSystemDir = this->builtForPlatform(Platform::driverKit, true) ? "/System/DriverKit/usr/lib/system/" : "/usr/lib/system/";
// except for dylibs in libSystem.dylib which are ok to link with nothing (they are on bottom)
bool isNotLibSystem = (installName == nullptr) || (strncmp(installName, libSystemDir, strlen(libSystemDir)) != 0);
- if ( internalInstall && ( hdr->builtForPlatform(Platform::macOS_exclaveKit, true)
- || hdr->builtForPlatform(Platform::iOS_exclaveKit, true)
- || hdr->builtForPlatform(Platform::tvOS_exclaveKit, true)
- || hdr->builtForPlatform(Platform::watchOS_exclaveKit, true)
- || hdr->builtForPlatform(Platform::visionOS_exclaveKit, true)) ) {
- // The path of ExclaveKit libSystem libraries starts with /System/ExclaveKit
- const size_t prefixLength = 18;
- isNotLibSystem = true;
- if ( installName != nullptr && strlen(installName) > prefixLength )
- if ( strncmp(installName + prefixLength, "/usr/lib/system/", 16) == 0 )
- isNotLibSystem = false;
- }
if ( this->isDyldManaged() && isNotLibSystem ) {
diag.error("in '%s' missing LC_LOAD_DYLIB (must link with at least libSystem.dylib)", path);
return false;
@@ -609,71 +816,318 @@
return true;
}
+bool MachOAnalyzer::validSegments(Diagnostics& diag, const char* path, size_t fileLen) const
+{
+ // check segment load command size
+ __block bool badSegmentLoadCommand = false;
+ forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ if ( cmd->cmd == LC_SEGMENT_64 ) {
+ const segment_command_64* seg = (segment_command_64*)cmd;
+ int32_t sectionsSpace = cmd->cmdsize - sizeof(segment_command_64);
+ if ( sectionsSpace < 0 ) {
+ diag.error("in '%s' load command size too small for LC_SEGMENT_64", path);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( (sectionsSpace % sizeof(section_64)) != 0 ) {
+ diag.error("in '%s' segment load command size 0x%X will not fit whole number of sections", path, cmd->cmdsize);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( sectionsSpace != (seg->nsects * sizeof(section_64)) ) {
+ diag.error("in '%s' load command size 0x%X does not match nsects %d", path, cmd->cmdsize, seg->nsects);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( greaterThanAddOrOverflow(seg->fileoff, seg->filesize, fileLen) ) {
+ diag.error("in '%s' segment load command content extends beyond end of file", path);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( (seg->filesize > seg->vmsize) && ((seg->vmsize != 0) || ((seg->flags & SG_NORELOC) == 0)) ) {
+ // <rdar://problem/19986776> dyld should support non-allocatable __LLVM segment
+ diag.error("in '%s' segment '%s' filesize exceeds vmsize", path, seg->segname);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ }
+ else if ( cmd->cmd == LC_SEGMENT ) {
+ const segment_command* seg = (segment_command*)cmd;
+ int32_t sectionsSpace = cmd->cmdsize - sizeof(segment_command);
+ if ( sectionsSpace < 0 ) {
+ diag.error("in '%s' load command size too small for LC_SEGMENT", path);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( (sectionsSpace % sizeof(section)) != 0 ) {
+ diag.error("in '%s' segment load command size 0x%X will not fit whole number of sections", path, cmd->cmdsize);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( sectionsSpace != (seg->nsects * sizeof(section)) ) {
+ diag.error("in '%s' load command size 0x%X does not match nsects %d", path, cmd->cmdsize, seg->nsects);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ else if ( (seg->filesize > seg->vmsize) && ((seg->vmsize != 0) || ((seg->flags & SG_NORELOC) == 0)) ) {
+ // <rdar://problem/19986776> dyld should support non-allocatable __LLVM segment
+ diag.error("in '%s' segment '%s' filesize exceeds vmsize", path, seg->segname);
+ badSegmentLoadCommand = true;
+ stop = true;
+ }
+ }
+ });
+ if ( badSegmentLoadCommand )
+ return false;
+
+ // check mapping permissions of segments
+ __block bool badPermissions = false;
+ __block bool badSize = false;
+ __block bool hasTEXT = false;
+ __block bool hasLINKEDIT = false;
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ if ( strcmp(info.segName, "__TEXT") == 0 ) {
+ if ( (info.protections != (VM_PROT_READ|VM_PROT_EXECUTE)) && enforceFormat(Malformed::textPermissions) ) {
+ diag.error("in '%s' __TEXT segment permissions is not 'r-x'", path);
+ badPermissions = true;
+ stop = true;
+ }
+ hasTEXT = true;
+ }
+ else if ( strcmp(info.segName, "__LINKEDIT") == 0 ) {
+ if ( (info.protections != VM_PROT_READ) && enforceFormat(Malformed::linkeditPermissions) ) {
+ diag.error("in '%s' __LINKEDIT segment permissions is not 'r--'", path);
+ badPermissions = true;
+ stop = true;
+ }
+ hasLINKEDIT = true;
+ }
+ else if ( (info.protections & 0xFFFFFFF8) != 0 ) {
+ diag.error("in '%s' %s segment permissions has invalid bits set", path, info.segName);
+ badPermissions = true;
+ stop = true;
+ }
+ if ( greaterThanAddOrOverflow(info.fileOffset, info.fileSize, fileLen) ) {
+ diag.error("in '%s' %s segment content extends beyond end of file", path, info.segName);
+ badSize = true;
+ stop = true;
+ }
+ if ( is64() ) {
+ if ( info.vmAddr+info.vmSize < info.vmAddr ) {
+ diag.error("in '%s' %s segment vm range wraps", path, info.segName);
+ badSize = true;
+ stop = true;
+ }
+ }
+ else {
+ if ( (uint32_t)(info.vmAddr+info.vmSize) < (uint32_t)(info.vmAddr) ) {
+ diag.error("in '%s' %s segment vm range wraps", path, info.segName);
+ badSize = true;
+ stop = true;
+ }
+ }
+ });
+ if ( badPermissions || badSize )
+ return false;
+ if ( !hasTEXT ) {
+ diag.error("in '%s' missing __TEXT segment", path);
+ return false;
+ }
+ if ( !hasLINKEDIT ) {
+ diag.error("in '%s' missing __LINKEDIT segment", path);
+ return false;
+ }
+
+ // check for overlapping segments
+ __block bool badSegments = false;
+ forEachSegment(^(const SegmentInfo& info1, bool& stop1) {
+ uint64_t seg1vmEnd = info1.vmAddr + info1.vmSize;
+ uint64_t seg1FileEnd = info1.fileOffset + info1.fileSize;
+ forEachSegment(^(const SegmentInfo& info2, bool& stop2) {
+ if ( info1.segIndex == info2.segIndex )
+ return;
+ uint64_t seg2vmEnd = info2.vmAddr + info2.vmSize;
+ uint64_t seg2FileEnd = info2.fileOffset + info2.fileSize;
+ if ( ((info2.vmAddr <= info1.vmAddr) && (seg2vmEnd > info1.vmAddr) && (seg1vmEnd > info1.vmAddr )) || ((info2.vmAddr >= info1.vmAddr ) && (info2.vmAddr < seg1vmEnd) && (seg2vmEnd > info2.vmAddr)) ) {
+ diag.error("in '%s' segment %s vm range overlaps segment %s", path, info1.segName, info2.segName);
+ badSegments = true;
+ stop1 = true;
+ stop2 = true;
+ }
+ if ( ((info2.fileOffset <= info1.fileOffset) && (seg2FileEnd > info1.fileOffset) && (seg1FileEnd > info1.fileOffset)) || ((info2.fileOffset >= info1.fileOffset) && (info2.fileOffset < seg1FileEnd) && (seg2FileEnd > info2.fileOffset )) ) {
+ if ( !inDyldCache() ) {
+ // HACK: Split shared caches might put the __TEXT in a SubCache, then the __DATA in a later SubCache.
+ // The file offsets are in to each SubCache file, which means that they might overlap
+ // For now we have no choice but to disable this error
+ diag.error("in '%s' segment %s file content overlaps segment %s", path, info1.segName, info2.segName);
+ badSegments = true;
+ stop1 = true;
+ stop2 = true;
+ }
+ }
+ if ( (info1.segIndex < info2.segIndex) && !stop1 ) {
+ if ( (info1.vmAddr > info2.vmAddr) || ((info1.fileOffset > info2.fileOffset ) && (info1.fileOffset != 0) && (info2.fileOffset != 0)) ){
+ if ( !inDyldCache() && enforceFormat(Malformed::segmentOrder) && !isStaticExecutable() ) {
+ //<rdar://80084852> whitelist go libraries __DWARF segments
+ if ( (strcmp(info1.segName, "__DWARF") != 0 && strcmp(info2.segName, "__DWARF") != 0) ) {
+ // dyld cache __DATA_* segments are moved around
+ // The static kernel also has segments with vmAddr's before __TEXT
+ diag.error("in '%s' segment load commands out of order with respect to layout for %s and %s", path, info1.segName, info2.segName);
+ badSegments = true;
+ stop1 = true;
+ stop2 = true;
+ }
+ }
+ }
+ }
+ });
+ });
+ if ( badSegments )
+ return false;
+
+ // check sections are within segment
+ __block bool badSections = false;
+ forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ if ( cmd->cmd == LC_SEGMENT_64 ) {
+ const segment_command_64* seg = (segment_command_64*)cmd;
+ const section_64* const sectionsStart = (section_64*)((char*)seg + sizeof(struct segment_command_64));
+ const section_64* const sectionsEnd = §ionsStart[seg->nsects];
+ for (const section_64* sect=sectionsStart; (sect < sectionsEnd); ++sect) {
+ if ( (int64_t)(sect->size) < 0 ) {
+ diag.error("in '%s' section '%s' size too large 0x%llX", path, sect->sectname, sect->size);
+ badSections = true;
+ }
+ else if ( sect->addr < seg->vmaddr ) {
+ diag.error("in '%s' section '%s' start address 0x%llX is before containing segment's address 0x%0llX", path, sect->sectname, sect->addr, seg->vmaddr);
+ badSections = true;
+ }
+ else if ( sect->addr+sect->size > seg->vmaddr+seg->vmsize ) {
+ bool ignoreError = !enforceFormat(Malformed::sectionsAddrRangeWithinSegment);
+#if BUILDING_APP_CACHE_UTIL
+ if ( (seg->vmsize == 0) && !strcmp(seg->segname, "__CTF") )
+ ignoreError = true;
+#endif
+ if ( !ignoreError ) {
+ diag.error("in '%s' section '%s' end address 0x%llX is beyond containing segment's end address 0x%0llX", path, sect->sectname, sect->addr+sect->size, seg->vmaddr+seg->vmsize);
+ badSections = true;
+ }
+ }
+ }
+ }
+ else if ( cmd->cmd == LC_SEGMENT ) {
+ const segment_command* seg = (segment_command*)cmd;
+ const section* const sectionsStart = (section*)((char*)seg + sizeof(struct segment_command));
+ const section* const sectionsEnd = §ionsStart[seg->nsects];
+ for (const section* sect=sectionsStart; !stop && (sect < sectionsEnd); ++sect) {
+ if ( (int64_t)(sect->size) < 0 ) {
+ diag.error("in '%s' section %s size too large 0x%X", path, sect->sectname, sect->size);
+ badSections = true;
+ }
+ else if ( sect->addr < seg->vmaddr ) {
+ diag.error("in '%s' section %s start address 0x%X is before containing segment's address 0x%0X", path, sect->sectname, sect->addr, seg->vmaddr);
+ badSections = true;
+ }
+ else if ( sect->addr+sect->size > seg->vmaddr+seg->vmsize ) {
+ diag.error("in '%s' section %s end address 0x%X is beyond containing segment's end address 0x%0X", path, sect->sectname, sect->addr+sect->size, seg->vmaddr+seg->vmsize);
+ badSections = true;
+ }
+ }
+ }
+ });
+
+ return !badSections;
+}
+
bool MachOAnalyzer::validMain(Diagnostics& diag, const char* path) const
{
+ // FIXME: Also check that dylibs with inDyldCache() set are actaully in the cache
if ( this->inDyldCache() && MachOAnalyzer::enforceFormat(Malformed::mainExecInDyldCache) ) {
diag.error("MH_EXECUTE is in dyld shared cache");
return false;
}
+ const char* executableTextSegmentName = "__TEXT";
+#if BUILDING_APP_CACHE_UTIL
+ // The kernel has __start in __TEXT_EXEC, or for x86_64 it's __HIB
+ if ( isStaticExecutable() ) {
+ if ( isArch("x86_64") || isArch("x86_64h") )
+ executableTextSegmentName = "__HIB";
+ else
+ executableTextSegmentName = "__TEXT_EXEC";
+ }
+#endif
+
+ __block uint64_t textSegStartAddr = 0;
+ __block uint64_t textSegStartSize = 0;
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ if ( strcmp(info.segName, executableTextSegmentName) == 0 ) {
+ textSegStartAddr = info.vmAddr;
+ textSegStartSize = info.vmSize;
+ stop = true;
+ }
+ });
__block int mainCount = 0;
__block int threadCount = 0;
forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ entry_point_command* mainCmd;
+ uint64_t startAddress;
switch (cmd->cmd) {
- case LC_MAIN: {
+ case LC_MAIN:
++mainCount;
- entry_point_command* mainCmd = (entry_point_command*)cmd;
- uint64_t startAddress = ((const Header*)this)->preferredLoadAddress() + mainCmd->entryoff;
-
- __block bool foundSegment = false;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stopSegment) {
- // Skip segments which don't contain this address
- if ( (startAddress < info.vmaddr) || (startAddress >= info.vmaddr+info.vmsize) )
- return;
- foundSegment = true;
- if ( !info.executable() )
- diag.error("LC_MAIN points to non-executable segment");
- stopSegment = true;
- });
- if (!foundSegment)
- diag.error("LC_MAIN entryoff is out of range");
-
- stop = true;
- break;
- }
- case LC_UNIXTHREAD: {
+ mainCmd = (entry_point_command*)cmd;
+ if ( mainCmd->entryoff >= textSegStartSize ) {
+ startAddress = preferredLoadAddress() + mainCmd->entryoff;
+ __block bool foundSegment = false;
+ forEachSegment(^(const SegmentInfo& info, bool& stopSegment) {
+ // Skip segments which don't contain this address
+ if ( (startAddress < info.vmAddr) || (startAddress >= info.vmAddr+info.vmSize) )
+ return;
+ foundSegment = true;
+ if ( (info.protections & VM_PROT_EXECUTE) == 0 )
+ diag.error("LC_MAIN points to non-executable segment");
+ stopSegment = true;
+ });
+ if (!foundSegment)
+ diag.error("LC_MAIN entryoff is out of range");
+ stop = true;
+ }
+ break;
+ case LC_UNIXTHREAD:
++threadCount;
- uint64_t startAddress = entryAddrFromThreadCmd((thread_command*)cmd);
+ startAddress = entryAddrFromThreadCmd((thread_command*)cmd);
if ( startAddress == 0 ) {
diag.error("LC_UNIXTHREAD not valid for arch %s", archName());
stop = true;
- } else {
+ }
+#if BUILDING_DYLDINFO
+ else if ( isStaticExecutable() ) {
__block bool foundSegment = false;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stopSegment) {
+ forEachSegment(^(const SegmentInfo& info, bool& stopSegment) {
// Skip segments which don't contain this address
- if ( (startAddress < info.vmaddr) || (startAddress >= info.vmaddr+info.vmsize) )
+ if ( (startAddress < info.vmAddr) || (startAddress >= info.vmAddr+info.vmSize) )
return;
foundSegment = true;
- if ( !info.executable() ) {
- // Suppress this error for the x86_64 kernel
- if ( !this->isStaticExecutable() )
- diag.error("LC_UNIXTHREAD points to non-executable segment");
- }
+ if ( (info.protections & VM_PROT_EXECUTE) == 0 )
+ diag.error("LC_UNIXTHREAD points to non-executable segment");
stopSegment = true;
});
if (!foundSegment)
diag.error("LC_UNIXTHREAD entry is out of range");
stop = true;
}
- break;
- }
+#endif
+ else if ( (startAddress < textSegStartAddr) || (startAddress >= textSegStartAddr+textSegStartSize) ) {
+ diag.error("LC_UNIXTHREAD entry not in %s segment", executableTextSegmentName);
+ stop = true;
+ }
+ break;
}
});
if ( diag.hasError() )
return false;
- if ( ((const Header*)this)->builtForPlatform(Platform::driverKit) ) {
+ if ( this->builtForPlatform(Platform::driverKit) ) {
if ( mainCount + threadCount == 0 )
return true;
diag.error("LC_MAIN not allowed for driverkit");
@@ -682,6 +1136,7 @@
if ( mainCount+threadCount == 1 )
return true;
+
if ( mainCount + threadCount == 0 )
diag.error("missing LC_MAIN or LC_UNIXTHREAD");
else
@@ -722,16 +1177,174 @@
bool MachOAnalyzer::validLinkeditLayout(Diagnostics& diag, const char* path) const
{
- __block bool result = false;
- this->withVMLayout(diag, ^(const mach_o::Layout &layout) {
- result = layout.isValidLinkeditLayout(diag, path);
- });
- return result;
-}
-
-
-
-bool MachOAnalyzer::invalidRebaseState(Diagnostics& diag, const char* opcodeName, const char* path, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ LinkEditInfo leInfo;
+ getLinkEditPointers(diag, leInfo);
+ if ( diag.hasError() )
+ return false;
+ const uint32_t ptrSize = pointerSize();
+
+ // build vector of all blobs in LINKEDIT
+ LinkEditContentChunk blobs[32];
+ LinkEditContentChunk* bp = blobs;
+ if ( leInfo.dyldInfo != nullptr ) {
+ if ( leInfo.dyldInfo->rebase_size != 0 )
+ *bp++ = {"rebase opcodes", ptrSize, leInfo.dyldInfo->rebase_off, leInfo.dyldInfo->rebase_size};
+ if ( leInfo.dyldInfo->bind_size != 0 )
+ *bp++ = {"bind opcodes", ptrSize, leInfo.dyldInfo->bind_off, leInfo.dyldInfo->bind_size};
+ if ( leInfo.dyldInfo->weak_bind_size != 0 )
+ *bp++ = {"weak bind opcodes", ptrSize, leInfo.dyldInfo->weak_bind_off, leInfo.dyldInfo->weak_bind_size};
+ if ( leInfo.dyldInfo->lazy_bind_size != 0 )
+ *bp++ = {"lazy bind opcodes", ptrSize, leInfo.dyldInfo->lazy_bind_off, leInfo.dyldInfo->lazy_bind_size};
+ if ( leInfo.dyldInfo->export_size!= 0 )
+ *bp++ = {"exports trie", ptrSize, leInfo.dyldInfo->export_off, leInfo.dyldInfo->export_size};
+ }
+ if ( leInfo.exportsTrie != nullptr ) {
+ if ( leInfo.exportsTrie->datasize != 0 )
+ *bp++ = {"exports trie", ptrSize, leInfo.exportsTrie->dataoff, leInfo.exportsTrie->datasize};
+ }
+ if ( leInfo.chainedFixups != nullptr ) {
+ if ( leInfo.chainedFixups->datasize != 0 )
+ *bp++ = {"chained fixups", ptrSize, leInfo.chainedFixups->dataoff, leInfo.chainedFixups->datasize};
+ }
+
+ if ( leInfo.dynSymTab != nullptr ) {
+ if ( leInfo.dynSymTab->nlocrel != 0 )
+ *bp++ = {"local relocations", ptrSize, leInfo.dynSymTab->locreloff, static_cast<uint32_t>(leInfo.dynSymTab->nlocrel*sizeof(relocation_info))};
+ if ( leInfo.dynSymTab->nextrel != 0 )
+ *bp++ = {"external relocations", ptrSize, leInfo.dynSymTab->extreloff, static_cast<uint32_t>(leInfo.dynSymTab->nextrel*sizeof(relocation_info))};
+ if ( leInfo.dynSymTab->nindirectsyms != 0 )
+ *bp++ = {"indirect symbol table", 4, leInfo.dynSymTab->indirectsymoff, leInfo.dynSymTab->nindirectsyms*4};
+ }
+ if ( leInfo.splitSegInfo != nullptr ) {
+ if ( leInfo.splitSegInfo->datasize != 0 )
+ *bp++ = {"shared cache info", ptrSize, leInfo.splitSegInfo->dataoff, leInfo.splitSegInfo->datasize};
+ }
+ if ( leInfo.functionStarts != nullptr ) {
+ if ( leInfo.functionStarts->datasize != 0 )
+ *bp++ = {"function starts", ptrSize, leInfo.functionStarts->dataoff, leInfo.functionStarts->datasize};
+ }
+ if ( leInfo.dataInCode != nullptr ) {
+ if ( leInfo.dataInCode->datasize != 0 )
+ *bp++ = {"data in code", ptrSize, leInfo.dataInCode->dataoff, leInfo.dataInCode->datasize};
+ }
+ if ( leInfo.symTab != nullptr ) {
+ if ( leInfo.symTab->nsyms != 0 )
+ *bp++ = {"symbol table", ptrSize, leInfo.symTab->symoff, static_cast<uint32_t>(leInfo.symTab->nsyms*(ptrSize == 8 ? sizeof(nlist_64) : sizeof(struct nlist)))};
+ if ( leInfo.symTab->strsize != 0 )
+ *bp++ = {"symbol table strings", 1, leInfo.symTab->stroff, leInfo.symTab->strsize};
+ }
+ if ( leInfo.codeSig != nullptr ) {
+ if ( leInfo.codeSig->datasize != 0 )
+ *bp++ = {"code signature", ptrSize, leInfo.codeSig->dataoff, leInfo.codeSig->datasize};
+ }
+
+ // check for bad combinations
+ if ( (leInfo.dyldInfo != nullptr) && (leInfo.dyldInfo->cmd == LC_DYLD_INFO_ONLY) && (leInfo.dynSymTab != nullptr) ) {
+ if ( (leInfo.dynSymTab->nlocrel != 0) && enforceFormat(Malformed::dyldInfoAndlocalRelocs) ) {
+ diag.error("in '%s' malformed mach-o contains LC_DYLD_INFO_ONLY and local relocations", path);
+ return false;
+ }
+ if ( leInfo.dynSymTab->nextrel != 0 ) {
+ diag.error("in '%s' malformed mach-o contains LC_DYLD_INFO_ONLY and external relocations", path);
+ return false;
+ }
+ }
+
+ bool checkMissingDyldInfo = true;
+#if BUILDING_DYLDINFO || BUILDING_APP_CACHE_UTIL
+ checkMissingDyldInfo = !isFileSet() && !isStaticExecutable() && !isKextBundle();
+#endif
+ if ( (leInfo.dyldInfo == nullptr) && (leInfo.dynSymTab == nullptr) && checkMissingDyldInfo ) {
+ diag.error("in '%s' malformed mach-o misssing LC_DYLD_INFO and LC_DYSYMTAB", path);
+ return false;
+ }
+
+ // FIXME: Remove this hack
+#if BUILDING_APP_CACHE_UTIL
+ if ( isFileSet() )
+ return true;
+#endif
+
+ const unsigned long blobCount = bp - blobs;
+ if ( blobCount == 0 ) {
+ diag.error("in '%s' malformed mach-o missing LINKEDIT", path);
+ return false;
+ }
+
+ uint32_t linkeditFileEnd = leInfo.layout.linkeditFileOffset + leInfo.layout.linkeditFileSize;
+
+
+ // sort blobs by file-offset and error on overlaps
+ LinkEditContentChunk::sort(blobs, blobCount);
+ uint32_t prevEnd = leInfo.layout.linkeditFileOffset;
+ const char* prevName = "start of LINKEDIT";
+ for (unsigned long i=0; i < blobCount; ++i) {
+ const LinkEditContentChunk& blob = blobs[i];
+ if ( blob.fileOffsetStart < prevEnd ) {
+ diag.error("in '%s' LINKEDIT overlap of %s and %s", path, prevName, blob.name);
+ return false;
+ }
+ if (greaterThanAddOrOverflow(blob.fileOffsetStart, blob.size, linkeditFileEnd)) {
+ diag.error("in '%s' LINKEDIT content '%s' extends beyond end of segment", path, blob.name);
+ return false;
+ }
+ if ( (blob.fileOffsetStart & (blob.alignment-1)) != 0 ) {
+ // <rdar://problem/51115705> relax code sig alignment for pre iOS13
+ Malformed kind = (strcmp(blob.name, "code signature") == 0) ? Malformed::codeSigAlignment : Malformed::linkeditAlignment;
+ if ( enforceFormat(kind) )
+ diag.error("in '%s' mis-aligned LINKEDIT content '%s'", path, blob.name);
+ }
+ prevEnd = blob.fileOffsetStart + blob.size;
+ prevName = blob.name;
+ }
+
+ // Check for invalid symbol table sizes
+ if ( leInfo.symTab != nullptr ) {
+ if ( leInfo.symTab->nsyms > 0x10000000 ) {
+ diag.error("in '%s' malformed mach-o image: symbol table too large", path);
+ return false;
+ }
+ if ( leInfo.dynSymTab != nullptr ) {
+ // validate indirect symbol table
+ if ( leInfo.dynSymTab->nindirectsyms != 0 ) {
+ if ( leInfo.dynSymTab->nindirectsyms > 0x10000000 ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table too large", path);
+ return false;
+ }
+ }
+ if ( (leInfo.dynSymTab->nlocalsym > leInfo.symTab->nsyms) || (leInfo.dynSymTab->ilocalsym > leInfo.symTab->nsyms) ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table local symbol count exceeds total symbols", path);
+ return false;
+ }
+ if ( leInfo.dynSymTab->ilocalsym + leInfo.dynSymTab->nlocalsym < leInfo.dynSymTab->ilocalsym ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table local symbol count wraps", path);
+ return false;
+ }
+ if ( (leInfo.dynSymTab->nextdefsym > leInfo.symTab->nsyms) || (leInfo.dynSymTab->iextdefsym > leInfo.symTab->nsyms) ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table extern symbol count exceeds total symbols", path);
+ return false;
+ }
+ if ( leInfo.dynSymTab->iextdefsym + leInfo.dynSymTab->nextdefsym < leInfo.dynSymTab->iextdefsym ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table extern symbol count wraps", path);
+ return false;
+ }
+ if ( (leInfo.dynSymTab->nundefsym > leInfo.symTab->nsyms) || (leInfo.dynSymTab->iundefsym > leInfo.symTab->nsyms) ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table undefined symbol count exceeds total symbols", path);
+ return false;
+ }
+ if ( leInfo.dynSymTab->iundefsym + leInfo.dynSymTab->nundefsym < leInfo.dynSymTab->iundefsym ) {
+ diag.error("in '%s' malformed mach-o image: indirect symbol table undefined symbol count wraps", path);
+ return false;
+ }
+ }
+ }
+
+ return true;
+}
+
+
+
+bool MachOAnalyzer::invalidRebaseState(Diagnostics& diag, const char* opcodeName, const char* path, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset, Rebase kind) const
{
if ( !segIndexSet ) {
@@ -742,8 +1355,8 @@
diag.error("in '%s' %s segment index %d too large", path, opcodeName, segmentIndex);
return true;
}
- if ( segmentOffset > (segments[segmentIndex].vmsize - ptrSize) ) {
- diag.error("in '%s' %s current segment offset 0x%08llX beyond segment size (0x%08llX)", path, opcodeName, segmentOffset, segments[segmentIndex].vmsize);
+ if ( segmentOffset > (segments[segmentIndex].vmSize-ptrSize) ) {
+ diag.error("in '%s' %s current segment offset 0x%08llX beyond segment size (0x%08llX)", path, opcodeName, segmentOffset, segments[segmentIndex].vmSize);
return true;
}
switch ( kind ) {
@@ -760,6 +1373,10 @@
break;
case Rebase::textAbsolute32:
case Rebase::textPCrel32:
+ if ( !segments[segmentIndex].textRelocs ) {
+ diag.error("in '%s' %s text rebase is in segment that does not support text relocations", path, opcodeName);
+ return true;
+ }
if ( segments[segmentIndex].writable() ) {
diag.error("in '%s' %s text rebase is in writable segment", path, opcodeName);
return true;
@@ -777,17 +1394,17 @@
}
-void MachOAnalyzer::getAllSegmentsInfos(Diagnostics& diag, Header::SegmentInfo segments[]) const
-{
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stop) {
- segments[info.segmentIndex] = info;
+void MachOAnalyzer::getAllSegmentsInfos(Diagnostics& diag, SegmentInfo segments[]) const
+{
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ segments[info.segIndex] = info;
});
}
bool MachOAnalyzer::validRebaseInfo(Diagnostics& diag, const char* path) const
{
- forEachRebase(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ forEachRebase(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset, Rebase kind, bool& stop) {
if ( invalidRebaseState(diag, opcodeName, path, leInfo, segments, segIndexSet, ptrSize, segmentIndex, segmentOffset, kind) )
stop = true;
@@ -800,20 +1417,20 @@
{
__block bool startVmAddrSet = false;
__block uint64_t startVmAddr = 0;
- forEachRebase(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ forEachRebase(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset, Rebase kind, bool& stop) {
if ( kind != Rebase::textAbsolute32 )
return;
if ( !startVmAddrSet ) {
for (int i=0; i <= segmentIndex; ++i) {
- if ( segments[i].segmentName == "__TEXT" ) {
- startVmAddr = segments[i].vmaddr;
+ if ( strcmp(segments[i].segName, "__TEXT") == 0 ) {
+ startVmAddr = segments[i].vmAddr;
startVmAddrSet = true;
break;
}
}
}
- uint64_t rebaseVmAddr = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t rebaseVmAddr = segments[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = rebaseVmAddr - startVmAddr;
handler(runtimeOffset, stop);
});
@@ -827,17 +1444,17 @@
__block uint64_t lpEndVmAddr = 0;
__block uint64_t shVmAddr = 0;
__block uint64_t shEndVmAddr = 0;
- forEachSection(^(const Header::SectionInfo& info, bool &stop) {
- if ( (info.flags & SECTION_TYPE) == S_LAZY_SYMBOL_POINTERS ) {
- lpVmAddr = info.address;
- lpEndVmAddr = info.address + info.size;
- }
- else if ( (info.flags & S_ATTR_PURE_INSTRUCTIONS) && (info.sectionName == "__stub_helper") ) {
- shVmAddr = info.address;
- shEndVmAddr = info.address + info.size;
+ forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo& info, bool malformedSectionRange, bool &stop) {
+ if ( (info.sectFlags & SECTION_TYPE) == S_LAZY_SYMBOL_POINTERS ) {
+ lpVmAddr = info.sectAddr;
+ lpEndVmAddr = info.sectAddr + info.sectSize;
+ }
+ else if ( (info.sectFlags & S_ATTR_PURE_INSTRUCTIONS) && (strcmp(info.sectName, "__stub_helper") == 0) ) {
+ shVmAddr = info.sectAddr;
+ shEndVmAddr = info.sectAddr + info.sectSize;
}
});
- forEachRebase(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ forEachRebase(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset, Rebase kind, bool& stop) {
switch ( kind ) {
case Rebase::unknown:
@@ -852,14 +1469,14 @@
}
if ( !startVmAddrSet ) {
for (int i=0; i < segmentIndex; ++i) {
- if ( segments[i].segmentName == "__TEXT" ) {
- startVmAddr = segments[i].vmaddr;
+ if ( strcmp(segments[i].segName, "__TEXT") == 0 ) {
+ startVmAddr = segments[i].vmAddr;
startVmAddrSet = true;
break;
}
}
}
- uint64_t rebaseVmAddr = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t rebaseVmAddr = segments[segmentIndex].vmAddr + segmentOffset;
bool isLazyPointerRebase = false;
if ( (rebaseVmAddr >= lpVmAddr) && (rebaseVmAddr < lpEndVmAddr) ) {
// rebase is in lazy pointer section
@@ -915,7 +1532,7 @@
void MachOAnalyzer::forEachRebase(Diagnostics& diag,
- void (^handler)(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ void (^handler)(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset,
Rebase kind, bool& stop)) const
{
@@ -924,7 +1541,7 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
@@ -1030,7 +1647,6 @@
return;
}
-#if SUPPORT_CLASSIC_RELOCS
if ( leInfo.dynSymTab != nullptr ) {
// old binary, walk relocations
const uint64_t relocsStartAddress = localRelocBaseAddress(segmentsInfo, leInfo.layout.linkeditSegIndex);
@@ -1043,7 +1659,7 @@
for (const relocation_info* reloc=relocsStart; (reloc < relocsEnd) && !stop; ++reloc) {
if ( reloc->r_length != relocSize ) {
bool shouldEmitError = true;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
if ( usesClassicRelocationsInKernelCollection() && (reloc->r_length == 2) && (relocSize == 3) )
shouldEmitError = false;
#endif
@@ -1065,7 +1681,7 @@
uint32_t segIndex = 0;
uint64_t segOffset = 0;
uint64_t addr = 0;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
// xnu for x86_64 has __HIB mapped before __DATA, so offsets appear to be
// negative
if ( isStaticExecutable() || isFileSet() ) {
@@ -1107,54 +1723,49 @@
}
});
}
-#endif // SUPPORT_CLASSIC_RELOCS
-}
-
-bool MachOAnalyzer::segIndexAndOffsetForAddress(uint64_t addr, const Header::SegmentInfo segmentsInfos[], uint32_t segCount, uint32_t& segIndex, uint64_t& segOffset) const
+}
+
+bool MachOAnalyzer::segIndexAndOffsetForAddress(uint64_t addr, const SegmentInfo segmentsInfos[], uint32_t segCount, uint32_t& segIndex, uint64_t& segOffset) const
{
for (uint32_t i=0; i < segCount; ++i) {
- if ( (segmentsInfos[i].vmaddr <= addr) && (addr < segmentsInfos[i].vmaddr+segmentsInfos[i].vmsize) ) {
+ if ( (segmentsInfos[i].vmAddr <= addr) && (addr < segmentsInfos[i].vmAddr+segmentsInfos[i].vmSize) ) {
segIndex = i;
- segOffset = addr - segmentsInfos[i].vmaddr;
+ segOffset = addr - segmentsInfos[i].vmAddr;
return true;
}
}
return false;
}
-uint64_t MachOAnalyzer::localRelocBaseAddress(const Header::SegmentInfo segmentsInfos[], uint32_t segCount) const
+uint64_t MachOAnalyzer::localRelocBaseAddress(const SegmentInfo segmentsInfos[], uint32_t segCount) const
{
if ( isArch("x86_64") || isArch("x86_64h") ) {
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
if ( isKextBundle() ) {
// for kext bundles the reloc base address starts at __TEXT segment
- return segmentsInfos[0].vmaddr;
+ return segmentsInfos[0].vmAddr;
}
#endif
// for all other kinds, the x86_64 reloc base address starts at first writable segment (usually __DATA)
for (uint32_t i=0; i < segCount; ++i) {
if ( segmentsInfos[i].writable() )
- return segmentsInfos[i].vmaddr;
- }
- }
- // reloc base address is start of TEXT segment
- if ( this->isMainExecutable() && (segmentsInfos[0].initProt == 0) )
- return segmentsInfos[1].vmaddr;
- else
- return segmentsInfos[0].vmaddr;
-}
-
-uint64_t MachOAnalyzer::externalRelocBaseAddress(const Header::SegmentInfo segmentsInfos[], uint32_t segCount) const
+ return segmentsInfos[i].vmAddr;
+ }
+ }
+ return segmentsInfos[0].vmAddr;
+}
+
+uint64_t MachOAnalyzer::externalRelocBaseAddress(const SegmentInfo segmentsInfos[], uint32_t segCount) const
{
// Dyld caches are too large for a raw r_address, so everything is an offset from the base address
if ( inDyldCache() ) {
- return ((const Header*)this)->preferredLoadAddress();
- }
-
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+ return preferredLoadAddress();
+ }
+
+#if BUILDING_APP_CACHE_UTIL
if ( isKextBundle() ) {
// for kext bundles the reloc base address starts at __TEXT segment
- return ((const Header*)this)->preferredLoadAddress();
+ return preferredLoadAddress();
}
#endif
@@ -1162,7 +1773,7 @@
// for x86_64 reloc base address starts at first writable segment (usually __DATA)
for (uint32_t i=0; i < segCount; ++i) {
if ( segmentsInfos[i].writable() )
- return segmentsInfos[i].vmaddr;
+ return segmentsInfos[i].vmAddr;
}
}
// For everyone else we start at 0
@@ -1198,9 +1809,9 @@
if ( (indirectSymbolTableCount == 0) && isKextBundle() )
return;
- forEachSection(^(const Header::SectionInfo& sectInfo, bool& sectionStop) {
- uint8_t sectionType = (sectInfo.flags & SECTION_TYPE);
- bool selfModifyingStub = (sectionType == S_SYMBOL_STUBS) && (sectInfo.flags & S_ATTR_SELF_MODIFYING_CODE) && (sectInfo.reserved2 == 5) && (this->cputype == CPU_TYPE_I386);
+ forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo& sectInfo, bool malformedSectionRange, bool& sectionStop) {
+ uint8_t sectionType = (sectInfo.sectFlags & SECTION_TYPE);
+ bool selfModifyingStub = (sectionType == S_SYMBOL_STUBS) && (sectInfo.sectFlags & S_ATTR_SELF_MODIFYING_CODE) && (sectInfo.reserved2 == 5) && (this->cputype == CPU_TYPE_I386);
if ( (sectionType != S_LAZY_SYMBOL_POINTERS) && (sectionType != S_NON_LAZY_SYMBOL_POINTERS) && !selfModifyingStub )
return;
if ( (flags & S_ATTR_SELF_MODIFYING_CODE) && !selfModifyingStub ) {
@@ -1209,9 +1820,9 @@
return;
}
uint32_t elementSize = selfModifyingStub ? sectInfo.reserved2 : ptrSize;
- uint32_t elementCount = (uint32_t)(sectInfo.size/elementSize);
+ uint32_t elementCount = (uint32_t)(sectInfo.sectSize/elementSize);
if ( greaterThanAddOrOverflow(sectInfo.reserved1, elementCount, indirectSymbolTableCount) ) {
- diag.error("section %.*s overflows indirect symbol table", (int)sectInfo.sectionName.size(), sectInfo.sectionName.data());
+ diag.error("section %s overflows indirect symbol table", sectInfo.sectName);
sectionStop = true;
return;
}
@@ -1221,7 +1832,7 @@
if ( symNum == INDIRECT_SYMBOL_ABS )
continue;
if ( symNum == INDIRECT_SYMBOL_LOCAL ) {
- handler(sectInfo.address+i*elementSize, false, 0, "", false, false, false, stop);
+ handler(sectInfo.sectAddr+i*elementSize, false, 0, "", false, false, false, stop);
continue;
}
if ( symNum > symCount ) {
@@ -1250,7 +1861,7 @@
// Note we only want to change the value in memory once, before rebases are applied. We don't want to accidentally
// change it again later.
if ( supportPrivateExternsWorkaround ) {
- uintptr_t* ptr = (uintptr_t*)((uint8_t*)(sectInfo.address+i*elementSize) + this->getSlide());
+ uintptr_t* ptr = (uintptr_t*)((uint8_t*)(sectInfo.sectAddr+i*elementSize) + this->getSlide());
uint64_t n_value = is64Bit ? symbols64[symNum].n_value : symbols32[symNum].n_value;
*ptr = (uintptr_t)n_value;
}
@@ -1260,7 +1871,7 @@
// Handle defined weak def symbols which need to get a special ordinal
if ( ((n_type & N_TYPE) == N_SECT) && ((n_type & N_EXT) != 0) && ((n_desc & N_WEAK_DEF) != 0) )
libOrdinal = BIND_SPECIAL_DYLIB_WEAK_LOOKUP;
- handler(sectInfo.address+i*elementSize, true, libOrdinal, symbolName, weakImport, lazy, selfModifyingStub, stop);
+ handler(sectInfo.sectAddr+i*elementSize, true, libOrdinal, symbolName, weakImport, lazy, selfModifyingStub, stop);
}
sectionStop = stop;
});
@@ -1290,7 +1901,7 @@
bool MachOAnalyzer::validBindInfo(Diagnostics& diag, const char* path) const
{
- forEachBind(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ forEachBind(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset,
uint8_t type, const char* symbolName, bool weakImport, bool lazyBind, uint64_t addend, bool& stop) {
@@ -1303,7 +1914,7 @@
return diag.noError();
}
-bool MachOAnalyzer::invalidBindState(Diagnostics& diag, const char* opcodeName, const char* path, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+bool MachOAnalyzer::invalidBindState(Diagnostics& diag, const char* opcodeName, const char* path, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal, uint32_t ptrSize,
uint8_t segmentIndex, uint64_t segmentOffset, uint8_t type, const char* symbolName) const
{
@@ -1315,8 +1926,8 @@
diag.error("in '%s' %s segment index %d too large", path, opcodeName, segmentIndex);
return true;
}
- if ( segmentOffset > (segments[segmentIndex].vmsize - ptrSize) ) {
- diag.error("in '%s' %s current segment offset 0x%08llX beyond segment size (0x%08llX)", path, opcodeName, segmentOffset, segments[segmentIndex].vmsize);
+ if ( segmentOffset > (segments[segmentIndex].vmSize-ptrSize) ) {
+ diag.error("in '%s' %s current segment offset 0x%08llX beyond segment size (0x%08llX)", path, opcodeName, segmentOffset, segments[segmentIndex].vmSize);
return true;
}
if ( symbolName == NULL ) {
@@ -1350,11 +1961,11 @@
case BIND_TYPE_TEXT_PCREL32: {
// Text relocations are permitted in x86_64 kexts
bool forceAllowTextRelocs = false;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
if ( isKextBundle() && (isArch("x86_64") || isArch("x86_64h")) )
forceAllowTextRelocs = true;
#endif
- if ( !forceAllowTextRelocs ) {
+ if ( !forceAllowTextRelocs && !segments[segmentIndex].textRelocs ) {
diag.error("in '%s' %s text bind is in segment that does not support text relocations", path, opcodeName);
return true;
}
@@ -1381,20 +1992,20 @@
{
__block bool startVmAddrSet = false;
__block uint64_t startVmAddr = 0;
- forEachBind(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ forEachBind(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset,
uint8_t type, const char* symbolName, bool weakImport, bool lazyBind, uint64_t addend, bool& stop) {
if ( !startVmAddrSet ) {
for (int i=0; i <= segmentIndex; ++i) {
- if ( segments[i].segmentName == "__TEXT" ) {
- startVmAddr = segments[i].vmaddr;
+ if ( strcmp(segments[i].segName, "__TEXT") == 0 ) {
+ startVmAddr = segments[i].vmAddr;
startVmAddrSet = true;
break;
}
}
}
- uint64_t bindVmOffset = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t bindVmOffset = segments[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = bindVmOffset - startVmAddr;
handler(runtimeOffset, libOrdinal, type, symbolName, weakImport, lazyBind, addend, stop);
}, ^(const char* symbolName) {
@@ -1413,7 +2024,7 @@
}
void MachOAnalyzer::forEachBind(Diagnostics& diag,
- void (^handler)(const char* opcodeName, const LinkEditInfo& leInfo, const Header::SegmentInfo segments[],
+ void (^handler)(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset, uint8_t type,
const char* symbolName, bool weakImport, bool lazyBind, uint64_t addend, bool& stop),
@@ -1427,7 +2038,7 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
@@ -1699,7 +2310,6 @@
else if ( leInfo.chainedFixups != nullptr ) {
// binary uses chained fixups, so do nothing
}
-#if SUPPORT_CLASSIC_RELOCS
else if ( leInfo.dynSymTab != nullptr ) {
// old binary, process external relocations
const uint64_t relocsStartAddress = externalRelocBaseAddress(segmentsInfo, leInfo.layout.linkeditSegIndex);
@@ -1715,7 +2325,7 @@
uint32_t poolSize = leInfo.symTab->strsize;
for (const relocation_info* reloc=relocsStart; (reloc < relocsEnd) && !stop; ++reloc) {
bool isBranch = false;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
if ( isKextBundle() ) {
// kext's may have other kinds of relocations, eg, branch relocs. Skip them
if ( isArch("x86_64") || isArch("x86_64h") ) {
@@ -1764,7 +2374,7 @@
else {
const char* symbolName = stringPool + strOffset;
bool weakImport = (n_desc & N_WEAK_REF);
- const uint8_t* content = (uint8_t*)this + segmentsInfo[segIndex].vmaddr - leInfo.layout.textUnslidVMAddr + segOffset;
+ const uint8_t* content = (uint8_t*)this + segmentsInfo[segIndex].vmAddr - leInfo.layout.textUnslidVMAddr + segOffset;
uint64_t addend = (reloc->r_length == 3) ? *((uint64_t*)content) : *((uint32_t*)content);
// Handle defined weak def symbols which need to get a special ordinal
if ( ((n_type & N_TYPE) == N_SECT) && ((n_type & N_EXT) != 0) && ((n_desc & N_WEAK_DEF) != 0) )
@@ -1798,7 +2408,7 @@
}
});
}
-#endif // SUPPORT_CLASSIC_RELOCS
+
}
bool MachOAnalyzer::validChainedFixupsInfo(Diagnostics& diag, const char* path) const
@@ -1808,7 +2418,7 @@
if ( diag.hasError() )
return false;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return false;
@@ -1868,8 +2478,8 @@
uint32_t numExtraSegments = (leInfo.layout.lastSegIndex + 1) - startsInfo->seg_count;
for (unsigned i = 0; i != numExtraSegments; ++i) {
// Check each extra segment before linkedit
- const Header::SegmentInfo& segInfo = segmentsInfo[leInfo.layout.linkeditSegIndex - (i + 1)];
- if ( segInfo.vmsize == 0 )
+ const SegmentInfo& segInfo = segmentsInfo[leInfo.layout.linkeditSegIndex - (i + 1)];
+ if ( segInfo.vmSize == 0 )
++numNoRelocSegments;
}
@@ -1878,7 +2488,7 @@
return false;
}
}
- const uint64_t baseAddress = ((const Header*)this)->preferredLoadAddress();
+ const uint64_t baseAddress = preferredLoadAddress();
uint32_t maxValidPointerSeen = 0;
uint16_t pointer_format_for_all = 0;
bool pointer_format_found = false;
@@ -1899,7 +2509,7 @@
diag.error("chained fixups, page_size not 4KB or 16KB in segment #%d", i);
return false;
}
- if ( segInfo->pointer_format > 13 ) {
+ if ( segInfo->pointer_format > 12 ) {
diag.error("chained fixups, unknown pointer_format in segment #%d", i);
return false;
}
@@ -1911,7 +2521,7 @@
diag.error("chained fixups, pointer_format not same for all segments %d and %d", segInfo->pointer_format, pointer_format_for_all);
return false;
}
- if ( segInfo->segment_offset != (segmentsInfo[i].vmaddr - baseAddress) ) {
+ if ( segInfo->segment_offset != (segmentsInfo[i].vmAddr - baseAddress) ) {
diag.error("chained fixups, segment_offset does not match vmaddr from LC_SEGMENT in segment #%d", i);
return false;
}
@@ -1930,7 +2540,7 @@
diag.error("chained fixups, page_start array overflows size");
return false;
}
- uint32_t maxOverflowIndex = (uint32_t)(segInfo->size - offsetof(dyld_chained_starts_in_segment, page_start[0]))/sizeof(uint16_t);
+ uint32_t maxOverflowIndex = (uint32_t)(segInfo->size - offsetof(dyld_chained_starts_in_segment, page_start[segInfo->page_count]))/sizeof(uint16_t);
for (int pageIndex=0; pageIndex < segInfo->page_count; ++pageIndex) {
uint16_t offsetInPage = segInfo->page_start[pageIndex];
if ( offsetInPage == DYLD_CHAINED_PTR_START_NONE )
@@ -1948,8 +2558,7 @@
uint16_t lastOffsetInPage = 0;
do {
if ( overflowIndex > maxOverflowIndex ) {
- diag.error("chain overflow index out of range %d (max=%d) in segment %.*s", overflowIndex, maxOverflowIndex,
- (int)segmentName(i).size(), segmentName(i).data());
+ diag.error("chain overflow index out of range %d (max=%d) in segment %s", overflowIndex, maxOverflowIndex, segmentName(i));
return false;
}
offsetInPage = (segInfo->page_start[overflowIndex] & ~DYLD_CHAINED_PTR_START_LAST);
@@ -1994,7 +2603,7 @@
// validate max_valid_pointer is larger than last segment
if ( (maxValidPointerSeen != 0) && !inDyldCache() ) {
- uint64_t lastSegmentLastVMAddr = segmentsInfo[leInfo.layout.linkeditSegIndex-1].vmaddr + segmentsInfo[leInfo.layout.linkeditSegIndex-1].vmsize;
+ uint64_t lastSegmentLastVMAddr = segmentsInfo[leInfo.layout.linkeditSegIndex-1].vmAddr + segmentsInfo[leInfo.layout.linkeditSegIndex-1].vmSize;
if ( maxValidPointerSeen < lastSegmentLastVMAddr ) {
diag.error("chained fixups, max_valid_pointer too small for image");
return false;
@@ -2012,7 +2621,7 @@
^(uint32_t totalTargets, bool& stop) {
maxTargetCount = totalTargets;
},
- ^(const LinkEditInfo& leInfo, const Header::SegmentInfo segments[], bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal, uint8_t type, const char* symbolName, uint64_t addend, bool weakImport, bool& stop) {
+ ^(const LinkEditInfo& leInfo, const SegmentInfo segments[], bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal, uint8_t type, const char* symbolName, uint64_t addend, bool weakImport, bool& stop) {
if ( symbolName == NULL ) {
diag.error("in '%s' missing BIND_OPCODE_SET_SYMBOL_TRAILING_FLAGS_IMM", path);
}
@@ -2035,15 +2644,15 @@
if ( diag.hasError() )
stop = true;
},
- ^(const LinkEditInfo& leInfo, const Header::SegmentInfo segments[], uint8_t segmentIndex, bool segIndexSet, uint64_t segmentOffset, uint16_t format, bool& stop) {
+ ^(const LinkEditInfo& leInfo, const SegmentInfo segments[], uint8_t segmentIndex, bool segIndexSet, uint64_t segmentOffset, uint16_t format, bool& stop) {
if ( !segIndexSet ) {
diag.error("in '%s' missing BIND_OPCODE_SET_SEGMENT_AND_OFFSET_ULEB", path);
}
else if ( segmentIndex >= leInfo.layout.linkeditSegIndex ) {
diag.error("in '%s' segment index %d too large", path, segmentIndex);
}
- else if ( segmentOffset > (segments[segmentIndex].vmsize-8) ) {
- diag.error("in '%s' current segment offset 0x%08llX beyond segment size (0x%08llX)", path, segmentOffset, segments[segmentIndex].vmsize);
+ else if ( segmentOffset > (segments[segmentIndex].vmSize-8) ) {
+ diag.error("in '%s' current segment offset 0x%08llX beyond segment size (0x%08llX)", path, segmentOffset, segments[segmentIndex].vmSize);
}
else if ( !segments[segmentIndex].writable() ) {
diag.error("in '%s' pointer bind is in non-writable segment", path);
@@ -2062,8 +2671,8 @@
void MachOAnalyzer::parseOrgArm64eChainedFixups(Diagnostics& diag, void (^targetCount)(uint32_t totalTargets, bool& stop),
- void (^addTarget)(const LinkEditInfo& leInfo, const Header::SegmentInfo segments[], bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal, uint8_t type, const char* symbolName, uint64_t addend, bool weakImport, bool& stop),
- void (^addChainStart)(const LinkEditInfo& leInfo, const Header::SegmentInfo segments[], uint8_t segmentIndex, bool segIndexSet, uint64_t segmentOffset, uint16_t format, bool& stop)) const
+ void (^addTarget)(const LinkEditInfo& leInfo, const SegmentInfo segments[], bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal, uint8_t type, const char* symbolName, uint64_t addend, bool weakImport, bool& stop),
+ void (^addChainStart)(const LinkEditInfo& leInfo, const SegmentInfo segments[], uint8_t segmentIndex, bool segIndexSet, uint64_t segmentOffset, uint16_t format, bool& stop)) const
{
bool stop = false;
@@ -2072,7 +2681,7 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
@@ -2178,26 +2787,174 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
+ bool stop = false;
if ( leInfo.dyldInfo != nullptr ) {
- parseOrgArm64eChainedFixups(diag, nullptr, ^(const LinkEditInfo& leInfo2, const Header::SegmentInfo segments[], bool libraryOrdinalSet, uint32_t dylibCount,
+ parseOrgArm64eChainedFixups(diag, nullptr, ^(const LinkEditInfo& leInfo2, const SegmentInfo segments[], bool libraryOrdinalSet, uint32_t dylibCount,
int libOrdinal, uint8_t type, const char* symbolName, uint64_t fixAddend, bool weakImport, bool& stopChain) {
callback(libOrdinal, symbolName, fixAddend, weakImport, stopChain);
}, nullptr);
}
else if ( leInfo.chainedFixups != nullptr ) {
const dyld_chained_fixups_header* header = (dyld_chained_fixups_header*)getLinkEditContent(leInfo.layout, leInfo.chainedFixups->dataoff);
- MachOFile::forEachChainedFixupTarget(diag, header, leInfo.chainedFixups, callback);
- }
-}
-
+ if ( (header->imports_offset > leInfo.chainedFixups->datasize) || (header->symbols_offset > leInfo.chainedFixups->datasize) ) {
+ diag.error("malformed import table");
+ return;
+ }
+ const dyld_chained_import* imports;
+ const dyld_chained_import_addend* importsA32;
+ const dyld_chained_import_addend64* importsA64;
+ const char* symbolsPool = (char*)header + header->symbols_offset;
+ uint32_t maxSymbolOffset = leInfo.chainedFixups->datasize - header->symbols_offset;
+ int libOrdinal;
+ switch (header->imports_format) {
+ case DYLD_CHAINED_IMPORT:
+ imports = (dyld_chained_import*)((uint8_t*)header + header->imports_offset);
+ for (uint32_t i=0; i < header->imports_count && !stop; ++i) {
+ const char* symbolName = &symbolsPool[imports[i].name_offset];
+ if ( imports[i].name_offset > maxSymbolOffset ) {
+ diag.error("malformed import table, string overflow");
+ return;
+ }
+ uint8_t libVal = imports[i].lib_ordinal;
+ if ( libVal > 0xF0 )
+ libOrdinal = (int8_t)libVal;
+ else
+ libOrdinal = libVal;
+ callback(libOrdinal, symbolName, 0, imports[i].weak_import, stop);
+ }
+ break;
+ case DYLD_CHAINED_IMPORT_ADDEND:
+ importsA32 = (dyld_chained_import_addend*)((uint8_t*)header + header->imports_offset);
+ for (uint32_t i=0; i < header->imports_count && !stop; ++i) {
+ const char* symbolName = &symbolsPool[importsA32[i].name_offset];
+ if ( importsA32[i].name_offset > maxSymbolOffset ) {
+ diag.error("malformed import table, string overflow");
+ return;
+ }
+ uint8_t libVal = importsA32[i].lib_ordinal;
+ if ( libVal > 0xF0 )
+ libOrdinal = (int8_t)libVal;
+ else
+ libOrdinal = libVal;
+ callback(libOrdinal, symbolName, importsA32[i].addend, importsA32[i].weak_import, stop);
+ }
+ break;
+ case DYLD_CHAINED_IMPORT_ADDEND64:
+ importsA64 = (dyld_chained_import_addend64*)((uint8_t*)header + header->imports_offset);
+ for (uint32_t i=0; i < header->imports_count && !stop; ++i) {
+ const char* symbolName = &symbolsPool[importsA64[i].name_offset];
+ if ( importsA64[i].name_offset > maxSymbolOffset ) {
+ diag.error("malformed import table, string overflow");
+ return;
+ }
+ uint16_t libVal = importsA64[i].lib_ordinal;
+ if ( libVal > 0xFFF0 )
+ libOrdinal = (int16_t)libVal;
+ else
+ libOrdinal = libVal;
+ callback(libOrdinal, symbolName, importsA64[i].addend, importsA64[i].weak_import, stop);
+ }
+ break;
+ default:
+ diag.error("unknown imports format");
+ return;
+ }
+ }
+}
+
+uint32_t MachOAnalyzer::segmentCount() const
+{
+ __block uint32_t count = 0;
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ ++count;
+ });
+ return count;
+}
+
+bool MachOAnalyzer::hasCodeSignature(uint32_t& fileOffset, uint32_t& size) const
+{
+ fileOffset = 0;
+ size = 0;
+
+ Diagnostics diag;
+ forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ if ( cmd->cmd == LC_CODE_SIGNATURE ) {
+ const linkedit_data_command* sigCmd = (linkedit_data_command*)cmd;
+ fileOffset = sigCmd->dataoff;
+ size = sigCmd->datasize;
+ stop = true;
+ }
+ });
+ diag.assertNoError(); // any malformations in the file should have been caught by earlier validate() call
+
+ // early exist if no LC_CODE_SIGNATURE
+ if ( fileOffset == 0 )
+ return false;
+
+ // <rdar://problem/13622786> ignore code signatures in macOS binaries built with pre-10.9 tools
+ if ( (this->cputype == CPU_TYPE_X86_64) || (this->cputype == CPU_TYPE_I386) ) {
+ __block bool foundPlatform = false;
+ __block bool badSignature = false;
+ forEachSupportedPlatform(^(Platform platform, uint32_t minOS, uint32_t sdk) {
+ foundPlatform = true;
+ if ( (platform == Platform::macOS) && (sdk < 0x000A0900) )
+ badSignature = true;
+ });
+ return foundPlatform && !badSignature;
+ }
+
+ return true;
+}
+
+bool MachOAnalyzer::hasProgramVars(uint32_t& progVarsOffset, bool& crtRunsInitializers, DyldLookFunc*& dyldLookupFuncAddr) const
+{
+ crtRunsInitializers = false;
+ if ( this->filetype != MH_EXECUTE )
+ return false;
+
+ // macOS 10.8+ program uses LC_MAIN and ProgramVars are in libdyld.dylib
+ // macOS 10.6 -> 10.7 ProgramVars are in __program_vars section in main executable
+ // macOS 10.5 ProgramVars are in __dyld section in main executable and 7 pointers in size
+ // macOS 10.4 and earlier ProgramVars need to be looked up by name in nlist of main executable
+
+ uint64_t offset;
+ bool usesCRT;
+ if ( getEntry(offset, usesCRT) && usesCRT ) {
+ // is pre-10.8 program
+ uint64_t sectionSize;
+ if ( const void* progVarsSection = findSectionContent("__DATA", "__program_vars", sectionSize) ) {
+ // macOS 10.6 or 10.7 binary
+ progVarsOffset = (uint32_t)((uint8_t*)progVarsSection - (uint8_t*)this);
+ return true;
+ }
+ else if ( const void* dyldSection = findSectionContent("__DATA", "__dyld", sectionSize) ) {
+#if SUPPPORT_PRE_LC_MAIN
+ if ( sectionSize >= 2*pointerSize() ) {
+ dyldLookupFuncAddr = (DyldLookFunc*)((uint8_t*)dyldSection + pointerSize());
+ }
+#endif
+ if ( sectionSize >= 7*pointerSize() ) {
+ // macOS 10.5 binary
+ progVarsOffset = (uint32_t)((uint8_t*)dyldSection - (uint8_t*)this) + 2*pointerSize();
+ return true;
+ }
+ else {
+ // macOS 10.4 binary
+ crtRunsInitializers = true;
+ }
+ }
+ return false;
+ }
+ return false;
+}
// Convert from a (possibly) live pointer to a vmAddr
-uint64_t MachOAnalyzer::VMAddrConverter::convertToVMAddr(uint64_t value, const Array<uint64_t>& bindTargets) const {
+uint64_t MachOAnalyzer::VMAddrConverter::convertToVMAddr(uint64_t value) const {
if ( contentRebased ) {
if ( value == 0 )
return 0;
@@ -2221,14 +2978,6 @@
if ( chainedValue->isRebase(chainedPointerFormat, preferredLoadAddress, targetRuntimeOffset) ) {
value = preferredLoadAddress + targetRuntimeOffset;
}
-
-#if !BUILDING_DYLD
- // Patchable objc classes use binds to self. Support them in offline tools
- uint32_t bindOrdinal = 0;
- int64_t addend = 0;
- if ( !bindTargets.empty() && chainedValue->isBind(chainedPointerFormat, bindOrdinal, addend))
- value = bindTargets[bindOrdinal] + addend;
-#endif
return value;
}
@@ -2238,10 +2987,6 @@
switch ( sharedCacheChainedPointerFormat ) {
case VMAddrConverter::SharedCacheFormat::none:
assert(false);
- case VMAddrConverter::SharedCacheFormat::v1: {
- // Nothing to do here. We don't have chained fixup bits to remove, or a value_add to apply
- break;
- }
case VMAddrConverter::SharedCacheFormat::v2_x86_64_tbi: {
const uint64_t deltaMask = 0x00FFFF0000000000;
const uint64_t valueMask = ~deltaMask;
@@ -2272,40 +3017,84 @@
}
break;
}
- case VMAddrConverter::SharedCacheFormat::v5: {
- // Just use the chained pointer format for arm64e
- if ( value == 0 )
- return 0;
- auto* chainedValue = (MachOAnalyzer::ChainedFixupPointerOnDisk*)&value;
- uint64_t targetRuntimeOffset;
- if ( chainedValue->isRebase(DYLD_CHAINED_PTR_ARM64E_SHARED_CACHE, preferredLoadAddress,
- targetRuntimeOffset) ) {
- value = preferredLoadAddress + targetRuntimeOffset;
- }
- break;
- }
}
return value;
}
#endif
return value;
-}
-
-uint64_t MachOAnalyzer::VMAddrConverter::convertToVMAddr(uint64_t v) const
-{
- return this->convertToVMAddr(v, {});
}
MachOAnalyzer::VMAddrConverter MachOAnalyzer::makeVMAddrConverter(bool contentRebased) const {
MachOAnalyzer::VMAddrConverter vmAddrConverter;
- vmAddrConverter.preferredLoadAddress = ((const Header*)this)->preferredLoadAddress();
+ vmAddrConverter.preferredLoadAddress = preferredLoadAddress();
vmAddrConverter.slide = getSlide();
vmAddrConverter.chainedPointerFormat = hasChainedFixups() ? chainedPointerFormat() : 0;
vmAddrConverter.contentRebased = contentRebased;
return vmAddrConverter;
}
+bool MachOAnalyzer::hasInitializer(Diagnostics& diag) const
+{
+ __block bool result = false;
+
+ // if dylib linked with -init linker option, that initializer is first
+ forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ if ( (cmd->cmd == LC_ROUTINES) || (cmd->cmd == LC_ROUTINES_64) ) {
+ result = true;
+ stop = true;
+ }
+ });
+
+ if ( result )
+ return true;
+
+ // next any function pointers in mod-init section
+ forEachInitializerPointerSection(diag, ^(uint32_t sectionOffset, uint32_t sectionSize, const uint8_t* content, bool& stop) {
+ result = true;
+ stop = true;
+ });
+
+ if ( result )
+ return true;
+
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( (info.sectFlags & SECTION_TYPE) != S_INIT_FUNC_OFFSETS )
+ return;
+ result = true;
+ stop = true;
+ });
+
+ return result;
+}
+
+void MachOAnalyzer::forEachInitializerPointerSection(Diagnostics& diag, void (^callback)(uint32_t sectionOffset, uint32_t sectionSize, const uint8_t* content, bool& stop)) const
+{
+ const unsigned ptrSize = pointerSize();
+ const uint64_t baseAddress = preferredLoadAddress();
+ const uint64_t slide = (uint64_t)this - baseAddress;
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& sectStop) {
+ if ( (info.sectFlags & SECTION_TYPE) == S_MOD_INIT_FUNC_POINTERS ) {
+ if ( (info.sectSize % ptrSize) != 0 ) {
+ diag.error("initializer section %s/%s has bad size", info.segInfo.segName, info.sectName);
+ sectStop = true;
+ return;
+ }
+ if ( malformedSectionRange ) {
+ diag.error("initializer section %s/%s extends beyond its segment", info.segInfo.segName, info.sectName);
+ sectStop = true;
+ return;
+ }
+ const uint8_t* content = (uint8_t*)(info.sectAddr + slide);
+ if ( ((long)content % ptrSize) != 0 ) {
+ diag.error("initializer section %s/%s is not pointer aligned", info.segInfo.segName, info.sectName);
+ sectStop = true;
+ return;
+ }
+ callback((uint32_t)(info.sectAddr - baseAddress), (uint32_t)info.sectSize, content, sectStop);
+ }
+ });
+}
struct VIS_HIDDEN SegmentRanges
{
@@ -2333,9 +3122,9 @@
void MachOAnalyzer::forEachInitializer(Diagnostics& diag, const VMAddrConverter& vmAddrConverter, void (^callback)(uint32_t offset), const void* dyldCache) const
{
__block SegmentRanges executableSegments;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stop) {
- if ( (info.initProt & VM_PROT_EXECUTE) != 0 ) {
- executableSegments.segments.push_back({ info.vmaddr, info.vmaddr + info.vmsize, (uint32_t)info.fileSize });
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ if ( (info.protections & VM_PROT_EXECUTE) != 0 ) {
+ executableSegments.segments.push_back({ info.vmAddr, info.vmAddr + info.vmSize, (uint32_t)info.fileSize });
}
});
@@ -2344,7 +3133,7 @@
return;
}
- uint64_t loadAddress = ((const Header*)this)->preferredLoadAddress();
+ uint64_t loadAddress = preferredLoadAddress();
intptr_t slide = getSlide();
// if dylib linked with -init linker option, that initializer is first
@@ -2369,8 +3158,7 @@
// next any function pointers in mod-init section
const unsigned ptrSize = pointerSize();
- forEachInitializerPointerSection(diag, ^(uint32_t sectionOffset, uint32_t sectionSize, bool& stop) {
- const uint8_t* content = (uint8_t*)this + sectionOffset;
+ forEachInitializerPointerSection(diag, ^(uint32_t sectionOffset, uint32_t sectionSize, const uint8_t* content, bool& stop) {
if ( ptrSize == 8 ) {
const uint64_t* initsStart = (uint64_t*)content;
const uint64_t* initsEnd = (uint64_t*)((uint8_t*)content + sectionSize);
@@ -2399,33 +3187,32 @@
}
});
- forEachSection(^(const Header::SectionInfo& info, bool& stop) {
- if ( (info.flags & SECTION_TYPE) != S_INIT_FUNC_OFFSETS )
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( (info.sectFlags & SECTION_TYPE) != S_INIT_FUNC_OFFSETS )
return;
- const uint8_t* content = (uint8_t*)(info.address + slide);
- if ( info.segInitProt & VM_PROT_WRITE ) {
- diag.error("initializer offsets section %.*s/%.*s must be in read-only segment",
- (int)info.segmentName.size(), info.segmentName.data(),
- (int)info.sectionName.size(), info.sectionName.data());
+ const uint8_t* content = (uint8_t*)(info.sectAddr + slide);
+ if ( info.segInfo.writable() ) {
+ diag.error("initializer offsets section %s/%s must be in read-only segment", info.segInfo.segName, info.sectName);
stop = true;
return;
}
- if ( (info.size % 4) != 0 ) {
- diag.error("initializer offsets section %.*s/%.*s has bad size",
- (int)info.segmentName.size(), info.segmentName.data(),
- (int)info.sectionName.size(), info.sectionName.data());
+ if ( (info.sectSize % 4) != 0 ) {
+ diag.error("initializer offsets section %s/%s has bad size", info.segInfo.segName, info.sectName);
stop = true;
return;
}
- if ( (info.address % 4) != 0 ) {
- diag.error("initializer offsets section %.*s/%.*s is not 4-byte aligned",
- (int)info.segmentName.size(), info.segmentName.data(),
- (int)info.sectionName.size(), info.sectionName.data());
+ if ( malformedSectionRange ) {
+ diag.error("initializer offsets section %s/%s extends beyond the end of the segment", info.segInfo.segName, info.sectName);
stop = true;
return;
}
+ if ( (info.sectAddr % 4) != 0 ) {
+ diag.error("initializer offsets section %s/%s is not 4-byte aligned", info.segInfo.segName, info.sectName);
+ stop = true;
+ return;
+ }
const uint32_t* initsStart = (uint32_t*)content;
- const uint32_t* initsEnd = (uint32_t*)((uint8_t*)content + info.size);
+ const uint32_t* initsEnd = (uint32_t*)((uint8_t*)content + info.sectSize);
for (const uint32_t* p=initsStart; p < initsEnd; ++p) {
uint32_t anInitOffset = *p;
if ( !executableSegments.contains(loadAddress + anInitOffset) ) {
@@ -2450,9 +3237,9 @@
void MachOAnalyzer::forEachTerminator(Diagnostics& diag, const VMAddrConverter& vmAddrConverter, void (^callback)(uint32_t offset)) const
{
__block SegmentRanges executableSegments;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stop) {
- if ( (info.initProt & VM_PROT_EXECUTE) != 0 ) {
- executableSegments.segments.push_back({ info.vmaddr, info.vmaddr + info.vmsize, (uint32_t)info.fileSize });
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ if ( (info.protections & VM_PROT_EXECUTE) != 0 ) {
+ executableSegments.segments.push_back({ info.vmAddr, info.vmAddr + info.vmSize, (uint32_t)info.fileSize });
}
});
@@ -2461,32 +3248,33 @@
return;
}
- uint64_t loadAddress = ((const Header*)this)->preferredLoadAddress();
+ uint64_t loadAddress = preferredLoadAddress();
intptr_t slide = getSlide();
// next any function pointers in mod-term section
const unsigned ptrSize = pointerSize();
- forEachSection(^(const Header::SectionInfo& info, bool& stop) {
- if ( (info.flags & SECTION_TYPE) == S_MOD_TERM_FUNC_POINTERS ) {
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( (info.sectFlags & SECTION_TYPE) == S_MOD_TERM_FUNC_POINTERS ) {
const uint8_t* content;
- content = (uint8_t*)(info.address + slide);
- if ( (info.size % ptrSize) != 0 ) {
- diag.error("terminator section %.*s/%.*s has bad size",
- (int)info.segmentName.size(), info.segmentName.data(),
- (int)info.sectionName.size(), info.sectionName.data());
+ content = (uint8_t*)(info.sectAddr + slide);
+ if ( (info.sectSize % ptrSize) != 0 ) {
+ diag.error("terminator section %s/%s has bad size", info.segInfo.segName, info.sectName);
stop = true;
return;
}
- if ( ((long)content % ptrSize) != 0 ) {
- diag.error("terminator section %.*s/%.*s is not pointer aligned",
- (int)info.segmentName.size(), info.segmentName.data(),
- (int)info.sectionName.size(), info.sectionName.data());
+ if ( malformedSectionRange ) {
+ diag.error("terminator section %s/%s extends beyond its segment", info.segInfo.segName, info.sectName);
stop = true;
return;
}
+ if ( ((long)content % ptrSize) != 0 ) {
+ diag.error("terminator section %s/%s is not pointer aligned", info.segInfo.segName, info.sectName);
+ stop = true;
+ return;
+ }
if ( ptrSize == 8 ) {
const uint64_t* initsStart = (uint64_t*)content;
- const uint64_t* initsEnd = (uint64_t*)((uint8_t*)content + info.size);
+ const uint64_t* initsEnd = (uint64_t*)((uint8_t*)content + info.sectSize);
for (const uint64_t* p=initsStart; p < initsEnd; ++p) {
uint64_t rawContent = *p;
#if __has_feature(ptrauth_calls)
@@ -2503,7 +3291,7 @@
}
else {
const uint32_t* initsStart = (uint32_t*)content;
- const uint32_t* initsEnd = (uint32_t*)((uint8_t*)content + info.size);
+ const uint32_t* initsEnd = (uint32_t*)((uint8_t*)content + info.sectSize);
for (const uint32_t* p=initsStart; p < initsEnd; ++p) {
uint32_t anInit = (uint32_t)vmAddrConverter.convertToVMAddr(*p);
if ( !executableSegments.contains(anInit) ) {
@@ -2518,57 +3306,114 @@
});
}
-bool MachOAnalyzer::hasSwiftOrObjC(bool* hasSwift) const
+
+
+void MachOAnalyzer::forEachRPath(void (^callback)(const char* rPath, bool& stop)) const
+{
+ Diagnostics diag;
+ forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ if ( cmd->cmd == LC_RPATH ) {
+ const char* rpath = (char*)cmd + ((struct rpath_command*)cmd)->path.offset;
+ callback(rpath, stop);
+ }
+ });
+ diag.assertNoError(); // any malformations in the file should have been caught by earlier validate() call
+}
+
+
+bool MachOAnalyzer::hasObjC() const
+{
+ __block bool result = false;
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( (strcmp(info.sectName, "__objc_imageinfo") == 0) && (strncmp(info.segInfo.segName, "__DATA", 6) == 0) ) {
+ result = true;
+ stop = true;
+ }
+ if ( (this->cputype == CPU_TYPE_I386) && (strcmp(info.sectName, "__image_info") == 0) && (strcmp(info.segInfo.segName, "__OBJC") == 0) ) {
+ result = true;
+ stop = true;
+ }
+ });
+ return result;
+}
+
+bool MachOAnalyzer::usesObjCGarbageCollection() const
+{
+ __block bool result = false;
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( (strcmp(info.sectName, "__objc_imageinfo") == 0) && (strncmp(info.segInfo.segName, "__DATA", 6) == 0) ) {
+ const uint64_t slide = (uint64_t)this - preferredLoadAddress();
+ const uint32_t* flags = (uint32_t*)(info.sectAddr + slide);
+ if ( flags[1] & 4 )
+ result = true;
+ stop = true;
+ }
+ });
+ return result;
+}
+
+
+bool MachOAnalyzer::hasPlusLoadMethod(Diagnostics& diag) const
+{
+ __block bool result = false;
+ if ( (this->cputype == CPU_TYPE_I386) && this->builtForPlatform(Platform::macOS) ) {
+ // old objc runtime has no special section for +load methods, scan for string
+ int64_t slide = getSlide();
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( ( (info.sectFlags & SECTION_TYPE) == S_CSTRING_LITERALS ) ) {
+ if ( malformedSectionRange ) {
+ diag.error("cstring section %s/%s extends beyond the end of the segment", info.segInfo.segName, info.sectName);
+ stop = true;
+ return;
+ }
+ const uint8_t* content = (uint8_t*)(info.sectAddr + slide);
+ const char* s = (char*)content;
+ const char* end = s + info.sectSize;
+ while ( s < end ) {
+ if ( strcmp(s, "load") == 0 ) {
+ result = true;
+ stop = true;
+ return;
+ }
+ while (*s != '\0' )
+ ++s;
+ ++s;
+ }
+ }
+ });
+ }
+ else {
+ // in new objc runtime compiler puts classes/categories with +load method in specical section
+ forEachSection(^(const SectionInfo& info, bool malformedSectionRange, bool& stop) {
+ if ( strncmp(info.segInfo.segName, "__DATA", 6) != 0 )
+ return;
+ if ( (strcmp(info.sectName, "__objc_nlclslist") == 0) || (strcmp(info.sectName, "__objc_nlcatlist") == 0)) {
+ result = true;
+ stop = true;
+ }
+ });
+ }
+ return result;
+}
+
+bool MachOAnalyzer::isSwiftLibrary() const
{
struct objc_image_info {
int32_t version;
uint32_t flags;
};
-
- if ( hasSwift != nullptr )
- *hasSwift = false;
-
- uintptr_t slide = getSlide();
+
+ int64_t slide = getSlide();
__block bool result = false;
- forEachSection(^(const Header::SectionInfo& sectInfo, bool& stop) {
- if ( (sectInfo.sectionName == "__objc_imageinfo") && sectInfo.segmentName.starts_with("__DATA") ) {
- if ( hasSwift != nullptr ) {
- objc_image_info* info = (objc_image_info*)((uintptr_t)sectInfo.address + slide);
- uint32_t swiftVersion = ((info->flags >> 8) & 0xFF);
- if ( swiftVersion )
- *hasSwift = true;
- }
-
- result = true;
- stop = true;
- }
- if ( (this->cputype == CPU_TYPE_I386) && (sectInfo.sectionName == "__image_info") && (sectInfo.sectionName == "__OBJC") ) {
- result = true;
- stop = true;
- }
- });
- return result;
-}
-
-bool MachOAnalyzer::hasSwift() const {
- bool hasSwift = false;
- this->hasSwiftOrObjC(&hasSwift);
-
- return hasSwift;
-}
-
-bool MachOAnalyzer::usesObjCGarbageCollection() const
-{
- __block bool result = false;
- forEachSection(^(const Header::SectionInfo& info, bool& stop) {
- if ( (info.sectionName == "__objc_imageinfo") && info.segmentName.starts_with("__DATA") ) {
- const uint64_t slide = (uint64_t)this - ((const Header*)this)->preferredLoadAddress();
- const uint32_t* flags = (uint32_t*)(info.address + slide);
- if ( flags[1] & 4 )
+ forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ if ( (strncmp(sectInfo.sectName, "__objc_imageinfo", 16) == 0) && (strncmp(sectInfo.segInfo.segName, "__DATA", 6) == 0) ) {
+ objc_image_info* info = (objc_image_info*)((uint8_t*)sectInfo.sectAddr + slide);
+ uint32_t swiftVersion = ((info->flags >> 8) & 0xFF);
+ if ( swiftVersion )
result = true;
stop = true;
}
- });
+ });
return result;
}
@@ -2649,11 +3494,11 @@
{
__block uint64_t textVmAddr = 0;
__block uint64_t result = 0;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& info, bool& stop) {
- if ( info.segmentName == "__TEXT" )
- textVmAddr = info.vmaddr;
- if ( info.segmentIndex == targetSegIndex ) {
- result = (info.vmaddr - textVmAddr) + targetSegOffset;
+ forEachSegment(^(const SegmentInfo& info, bool& stop) {
+ if ( strcmp(info.segName, "__TEXT") == 0 )
+ textVmAddr = info.vmAddr;
+ if ( info.segIndex == targetSegIndex ) {
+ result = (info.vmAddr - textVmAddr) + targetSegOffset;
}
});
return result;
@@ -2662,17 +3507,48 @@
bool MachOAnalyzer::hasLazyPointers(uint32_t& runtimeOffset, uint32_t& size) const
{
size = 0;
- forEachSection(^(const Header::SectionInfo& info, bool &stop) {
- if ( (info.flags & SECTION_TYPE) == S_LAZY_SYMBOL_POINTERS ) {
- runtimeOffset = (uint32_t)(info.address - ((const Header*)this)->preferredLoadAddress());
- size = (uint32_t)info.size;
+ forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo& info, bool malformedSectionRange, bool &stop) {
+ if ( (info.sectFlags & SECTION_TYPE) == S_LAZY_SYMBOL_POINTERS ) {
+ runtimeOffset = (uint32_t)(info.sectAddr - preferredLoadAddress());
+ size = (uint32_t)info.sectSize;
stop = true;
}
});
return (size != 0);
}
-#if !TARGET_OS_EXCLAVEKIT
+
+bool MachOAnalyzer::getEntry(uint64_t& offset, bool& usesCRT) const
+{
+ Diagnostics diag;
+ offset = 0;
+ forEachLoadCommand(diag, ^(const load_command* cmd, bool& stop) {
+ if ( cmd->cmd == LC_MAIN ) {
+ entry_point_command* mainCmd = (entry_point_command*)cmd;
+ usesCRT = false;
+ offset = mainCmd->entryoff;
+ stop = true;
+ }
+ else if ( cmd->cmd == LC_UNIXTHREAD ) {
+ stop = true;
+ usesCRT = true;
+ uint64_t startAddress = entryAddrFromThreadCmd((thread_command*)cmd);
+ offset = startAddress - preferredLoadAddress();
+ }
+ });
+ return (offset != 0);
+}
+
+
+void MachOAnalyzer::forEachDOFSection(Diagnostics& diag, void (^callback)(uint32_t offset)) const
+{
+ forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo& info, bool malformedSectionRange, bool &stop) {
+ if ( ( (info.sectFlags & SECTION_TYPE) == S_DTRACE_DOF ) && !malformedSectionRange ) {
+ callback((uint32_t)(info.sectAddr - info.segInfo.vmAddr));
+ }
+ });
+}
+
void MachOAnalyzer::forEachCDHash(void (^handler)(const uint8_t cdHash[20])) const
{
Diagnostics diag;
@@ -2705,7 +3581,60 @@
return requiresLV;
}
-#endif // !TARGET_OS_EXCLAVEKIT
+
+bool MachOAnalyzer::canHavePrecomputedDlopenClosure(const char* path, void (^failureReason)(const char*)) const
+{
+ if (!MachOFile::canHavePrecomputedDlopenClosure(path, failureReason))
+ return false;
+
+ // prebuilt closures use the cdhash of the dylib to verify that the dylib is still the same
+ // at runtime as when the shared cache processed it. We must have a code signature to record this information
+ uint32_t codeSigFileOffset;
+ uint32_t codeSigSize;
+ if ( !hasCodeSignature(codeSigFileOffset, codeSigSize) ) {
+ failureReason("no code signature");
+ return false;
+ }
+
+ __block bool retval = true;
+
+ // images that use dynamic_lookup, bundle_loader, or have weak-defs cannot have dlopen closure pre-computed
+ Diagnostics diag;
+ auto checkBind = ^(int libOrdinal, bool& stop) {
+ switch (libOrdinal) {
+ case BIND_SPECIAL_DYLIB_WEAK_LOOKUP:
+ failureReason("has weak externals");
+ retval = false;
+ stop = true;
+ break;
+ case BIND_SPECIAL_DYLIB_FLAT_LOOKUP:
+ failureReason("has dynamic_lookup binds");
+ retval = false;
+ stop = true;
+ break;
+ case BIND_SPECIAL_DYLIB_MAIN_EXECUTABLE:
+ failureReason("has reference to main executable (bundle loader)");
+ retval = false;
+ stop = true;
+ break;
+ }
+ };
+
+ if (hasChainedFixups()) {
+ forEachChainedFixupTarget(diag, ^(int libOrdinal, const char *symbolName, uint64_t addend, bool weakImport, bool &stop) {
+ checkBind(libOrdinal, stop);
+ });
+ } else {
+ forEachBind(diag, ^(uint64_t runtimeOffset, int libOrdinal, const char* symbolName, bool weakImport, bool lazyBind, uint64_t addend, bool& stop) {
+ checkBind(libOrdinal, stop);
+ },
+ ^(const char* symbolName) {
+ });
+ }
+
+ return retval;
+}
+
bool MachOAnalyzer::hasUnalignedPointerFixups() const
{
@@ -2828,16 +3757,10 @@
bool MachOAnalyzer::neverUnload() const
{
- bool hasSwift = false;
- if ( this->hasSwiftOrObjC(&hasSwift) ) {
- // Policy: images with ObjC or Swift are never unloaded
- // except MH_BUNDLE *without* Swift can be unloaded
- if ( hasSwift || !this->isBundle() ) {
- return true;
- }
- }
-
- if ( (this->flags & MH_HAS_TLV_DESCRIPTORS) ) {
+ if ( this->isDylib() && this->hasObjC() ) {
+ return true;
+ }
+ else if ( this->hasThreadLocalVariables() ) {
return true;
}
else {
@@ -2853,6 +3776,127 @@
return false;
}
+bool MachOAnalyzer::canBePlacedInDyldCache(const char* path, void (^failureReason)(const char*)) const
+{
+ if (!MachOFile::canBePlacedInDyldCache(path, failureReason))
+ return false;
+
+ // arm64e requires split seg v2 as the split seg code can't handle chained fixups for split seg v1
+ if ( isArch("arm64e") ) {
+ uint32_t splitSegSize = 0;
+ const uint8_t* infoStart = (const uint8_t*)getSplitSeg(splitSegSize);
+ if ( *infoStart != DYLD_CACHE_ADJ_V2_FORMAT ) {
+ failureReason("chained fixups requires split seg v2");
+ return false;
+ }
+ }
+
+ // <rdar://problem/57769033> dyld_cache_patchable_location only supports addend in range 0..31
+ const bool is64bit = is64();
+ __block Diagnostics diag;
+ __block bool addendTooLarge = false;
+ if ( this->hasChainedFixups() ) {
+ // with chained fixups, addends can be in the import table or embedded in a bind pointer
+ forEachChainedFixupTarget(diag, ^(int libOrdinal, const char* symbolName, uint64_t addend, bool weakImport, bool& stop) {
+ if ( is64bit )
+ addend &= 0x00FFFFFFFFFFFFFF; // ignore TBI
+ if ( addend > 31 ) {
+ addendTooLarge = true;
+ stop = true;
+ }
+ });
+ // check each pointer for embedded addend
+ withChainStarts(diag, 0, ^(const dyld_chained_starts_in_image* starts) {
+ forEachFixupInAllChains(diag, starts, false, ^(ChainedFixupPointerOnDisk* fixupLoc, const dyld_chained_starts_in_segment* segInfo, bool& stop) {
+ switch (segInfo->pointer_format) {
+ case DYLD_CHAINED_PTR_ARM64E:
+ case DYLD_CHAINED_PTR_ARM64E_USERLAND:
+ case DYLD_CHAINED_PTR_ARM64E_USERLAND24:
+ if ( fixupLoc->arm64e.bind.bind && !fixupLoc->arm64e.authBind.auth ) {
+ if ( fixupLoc->arm64e.bind.addend > 31 ) {
+ addendTooLarge = true;
+ stop = true;
+ }
+ }
+ break;
+ case DYLD_CHAINED_PTR_64:
+ case DYLD_CHAINED_PTR_64_OFFSET:
+ if ( fixupLoc->generic64.rebase.bind ) {
+ if ( fixupLoc->generic64.bind.addend > 31 ) {
+ addendTooLarge = true;
+ stop = true;
+ }
+ }
+ break;
+ case DYLD_CHAINED_PTR_32:
+ if ( fixupLoc->generic32.bind.bind ) {
+ if ( fixupLoc->generic32.bind.addend > 31 ) {
+ addendTooLarge = true;
+ stop = true;
+ }
+ }
+ break;
+ }
+ });
+ });
+ }
+ else {
+ // scan bind opcodes for large addend
+ forEachBind(diag, ^(const char* opcodeName, const LinkEditInfo& leInfo, const SegmentInfo* segments, bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
+ uint32_t ptrSize, uint8_t segmentIndex, uint64_t segmentOffset, uint8_t type, const char* symbolName, bool weakImport, bool lazyBind, uint64_t addend, bool& stop) {
+ if ( is64bit )
+ addend &= 0x00FFFFFFFFFFFFFF; // ignore TBI
+ if ( addend > 31 ) {
+ addendTooLarge = true;
+ stop = true;
+ }
+ },
+ ^(const char* symbolName) {
+ });
+ }
+ if ( addendTooLarge ) {
+ failureReason("bind addend too large");
+ return false;
+ }
+
+ // evict swift dylibs with split seg v1 info
+ if ( this->isSwiftLibrary() && this->isSplitSegV1() )
+ return false;
+
+ if ( hasChainedFixups() ) {
+ // Chained fixups assumes split seg v2. This is true for now as chained fixups is arm64e only
+ return true; // this->isSplitSegV2();
+ }
+
+ if ( !(isArch("x86_64") || isArch("x86_64h")) )
+ return true;
+
+ __block bool rebasesOk = true;
+ uint64_t startVMAddr = preferredLoadAddress();
+ uint64_t endVMAddr = startVMAddr + mappedSize();
+ forEachRebase(diag, false, ^(uint64_t runtimeOffset, bool &stop) {
+ // We allow TBI for x86_64 dylibs, but then require that the remainder of the offset
+ // is a 32-bit offset from the mach-header.
+ uint64_t value = *(uint64_t*)((uint8_t*)this + runtimeOffset);
+ value &= 0x00FFFFFFFFFFFFFFULL;
+ if ( (value < startVMAddr) || (value >= endVMAddr) ) {
+ failureReason("rebase value out of range of dylib");
+ rebasesOk = false;
+ stop = true;
+ return;
+ }
+
+ // Also error if the rebase location is anything other than 4/8 byte aligned
+ if ( (runtimeOffset & 0x3) != 0 ) {
+ failureReason("rebase value is not 4-byte aligned");
+ rebasesOk = false;
+ stop = true;
+ return;
+ }
+ });
+ return rebasesOk;
+}
+
#if BUILDING_APP_CACHE_UTIL
bool MachOAnalyzer::canBePlacedInKernelCollection(const char* path, void (^failureReason)(const char*)) const
{
@@ -2883,17 +3927,17 @@
__block bool foundHIB = false;
__block uint64_t hibernateVMAddr = 0;
__block uint64_t hibernateVMSize = 0;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo &segmentInfo, bool &stop) {
- if ( segmentInfo.segmentName == "__TEXT" ) {
+ forEachSegment(^(const SegmentInfo &segmentInfo, bool &stop) {
+ if ( strcmp(segmentInfo.segName, "__TEXT") == 0 ) {
foundText = true;
}
- if ( segmentInfo.segmentName == "__TEXT_EXEC" ) {
+ if ( strcmp(segmentInfo.segName, "__TEXT_EXEC") == 0 ) {
foundTextExec = true;
}
- if ( segmentInfo.segmentName == "__HIB" ) {
+ if ( strcmp(segmentInfo.segName, "__HIB") == 0 ) {
foundHIB = true;
- hibernateVMAddr = segmentInfo.vmaddr;
- hibernateVMSize = segmentInfo.vmsize;
+ hibernateVMAddr = segmentInfo.vmAddr;
+ hibernateVMSize = segmentInfo.vmSize;
}
});
if (!foundText) {
@@ -2911,7 +3955,7 @@
// The hibernate segment should be mapped before the base address
if ( foundHIB ) {
- uint64_t baseAddress = ((const Header*)this)->preferredLoadAddress();
+ uint64_t baseAddress = preferredLoadAddress();
if ( greaterThanAddOrOverflow(hibernateVMAddr, hibernateVMSize, baseAddress) ) {
failureReason("__HIB segment should be mapped before base address");
return false;
@@ -2920,34 +3964,9 @@
}
// Don't allow kext's to have load addresses
- if ( isKextBundle() && (((const Header*)this)->preferredLoadAddress() != 0) ) {
+ if ( isKextBundle() && (preferredLoadAddress() != 0) ) {
failureReason("Has load address");
return false;
- }
-
- // All kexts with an executable must have a kmod_info
- if ( isKextBundle() ) {
- __block bool found = false;
- __block Diagnostics diag;
-
- // Check for a global first
- FoundSymbol foundInfo;
- found = findExportedSymbol(diag, "_kmod_info", true, foundInfo, nullptr);
- if ( !found ) {
- // And fall back to a local if we need to
- forEachLocalSymbol(diag, ^(const char* aSymbolName, uint64_t n_value, uint8_t n_type,
- uint8_t n_sect, uint16_t n_desc, bool& stop) {
- if ( strcmp(aSymbolName, "_kmod_info") == 0 ) {
- found = true;
- stop = true;
- }
- });
- }
-
- if ( !found ) {
- failureReason("kexts must have a _kmod_info symbol");
- return false;
- }
}
if (hasChainedFixups()) {
@@ -3033,10 +4052,10 @@
// range of the number of bits we have
if ( isStaticExecutable() ) {
__block uint64_t baseAddress = ~0ULL;
- ((const Header*)this)->forEachSegment(^(const Header::SegmentInfo& sinfo, bool& segStop) {
- baseAddress = std::min(baseAddress, sinfo.vmaddr);
+ forEachSegment(^(const SegmentInfo& sinfo, bool& segStop) {
+ baseAddress = std::min(baseAddress, sinfo.vmAddr);
});
- uint64_t textSegVMAddr = ((const Header*)this)->preferredLoadAddress();
+ uint64_t textSegVMAddr = preferredLoadAddress();
runtimeOffset = (textSegVMAddr + runtimeOffset) - baseAddress;
}
#endif
@@ -3086,6 +4105,15 @@
#endif
+bool MachOAnalyzer::usesClassicRelocationsInKernelCollection() const {
+ // The xnu x86_64 static executable needs to do the i386->x86_64 transition
+ // so will be emitted with classic relocations
+ if ( isArch("x86_64") || isArch("x86_64h") ) {
+ return isStaticExecutable() || isFileSet();
+ }
+ return false;
+}
+
uint64_t MachOAnalyzer::chainStartsOffset() const
{
const dyld_chained_fixups_header* header = chainedFixupsHeader();
@@ -3106,12 +4134,27 @@
return (dyld_chained_fixups_header*)getLinkEditContent(leInfo.layout, leInfo.chainedFixups->dataoff);
}
+uint16_t MachOAnalyzer::chainedPointerFormat(const dyld_chained_fixups_header* header)
+{
+ const dyld_chained_starts_in_image* startsInfo = (dyld_chained_starts_in_image*)((uint8_t*)header + header->starts_offset);
+ for (uint32_t i=0; i < startsInfo->seg_count; ++i) {
+ uint32_t segInfoOffset = startsInfo->seg_info_offset[i];
+ // 0 offset means this segment has no fixups
+ if ( segInfoOffset == 0 )
+ continue;
+ const dyld_chained_starts_in_segment* segInfo = (dyld_chained_starts_in_segment*)((uint8_t*)startsInfo + segInfoOffset);
+ if ( segInfo->page_count != 0 )
+ return segInfo->pointer_format;
+ }
+ return 0; // no chains (perhaps no __DATA segment)
+}
+
uint16_t MachOAnalyzer::chainedPointerFormat() const
{
const dyld_chained_fixups_header* header = chainedFixupsHeader();
if ( header != nullptr ) {
// get pointer format from chain info struct in LINKEDIT
- return MachOFile::chainedPointerFormat(header);
+ return chainedPointerFormat(header);
}
assert(this->cputype == CPU_TYPE_ARM64 && (this->maskedCpuSubtype() == CPU_SUBTYPE_ARM64E) && "chainedPointerFormat() called on non-chained binary");
return DYLD_CHAINED_PTR_ARM64E;
@@ -3147,8 +4190,8 @@
// don't want this code in non-arm64e dyld because it causes a stack protector which dereferences a GOT pointer before GOT is set up
else if ( (leInfo.dyldInfo != nullptr) && (this->cputype == CPU_TYPE_ARM64) && (this->maskedCpuSubtype() == CPU_SUBTYPE_ARM64E) ) {
// old arm64e binary, create a dyld_chained_starts_in_image for caller
- uint64_t baseAddress = ((const Header*)this)->preferredLoadAddress();
- uint64_t imagePageCount = this->mappedSize()/0x1000;
+ uint64_t baseAddress = preferredLoadAddress();
+ uint64_t imagePageCount = this->mappedSize()/0x4000;
size_t bufferSize = leInfo.dyldInfo->bind_size + (size_t)imagePageCount*sizeof(uint16_t) + 512;
BLOCK_ACCCESSIBLE_ARRAY(uint8_t, buffer, bufferSize);
uint8_t* bufferEnd = &buffer[bufferSize];
@@ -3158,7 +4201,7 @@
header->seg_info_offset[i] = 0;
__block uint8_t curSegIndex = 0;
__block dyld_chained_starts_in_segment* curSeg = (dyld_chained_starts_in_segment*)(&(header->seg_info_offset[header->seg_count]));
- parseOrgArm64eChainedFixups(diag, nullptr, nullptr, ^(const LinkEditInfo& leInfo2, const Header::SegmentInfo segments[], uint8_t segmentIndex,
+ parseOrgArm64eChainedFixups(diag, nullptr, nullptr, ^(const LinkEditInfo& leInfo2, const SegmentInfo segments[], uint8_t segmentIndex,
bool segIndexSet, uint64_t segmentOffset, uint16_t format, bool& stop) {
uint32_t pageIndex = (uint32_t)(segmentOffset/0x1000);
if ( segmentIndex != curSegIndex ) {
@@ -3181,7 +4224,7 @@
curSeg->size = (uint32_t)((uint8_t*)(&curSeg->page_start[pageIndex]) - (uint8_t*)curSeg);
curSeg->page_size = 0x1000; // old arm64e encoding used 4KB pages
curSeg->pointer_format = DYLD_CHAINED_PTR_ARM64E;
- curSeg->segment_offset = segments[segmentIndex].vmaddr - baseAddress;
+ curSeg->segment_offset = segments[segmentIndex].vmAddr - baseAddress;
curSeg->max_valid_pointer = 0;
curSeg->page_count = pageIndex+1;
assert((uint8_t*)(&curSeg->page_start[pageIndex]) < bufferEnd);
@@ -3243,64 +4286,6 @@
return false;
}
-bool MachOAnalyzer::hasRebaseRuns(const void** runs, size_t* runsSize) const
-{
- if ( !this->isPreload() )
- return false;
-
- uint64_t sectionSize;
- if (const void* sect = this->findSectionContent("__TEXT", "__rebase_info", sectionSize) ) {
- *runs = sect;
- *runsSize = (size_t)sectionSize;
- return true;
- }
- return false;
-}
-
-struct RebaseRuns
-{
- uint32_t startAddress;
- uint8_t runs[]; // value of even indexes is how many pointers in a row are rebases, value of odd indexes times 4 is memory to skip over
- // two zero values in a row signals the end of the run
-};
-
-void MachOAnalyzer::forEachRebaseRunAddress(const void* runs, size_t runsSize, void (^handler)(uint32_t address)) const
-{
- const RebaseRuns* rr = (RebaseRuns*)runs;
- const RebaseRuns* end = (RebaseRuns*)((uint8_t*)runs + runsSize);
- while ( rr < end ) {
- uint32_t address = rr->startAddress;
- int index = 0;
- bool done = false;
- while ( !done ) {
- uint8_t count = rr->runs[index];
- if ( count == 0 ) {
- // two 0x00 in a row mean the run is complete
- if ( rr->runs[index+1] == 0 ) {
- ++index;
- done = true;
- }
- }
- else {
- if ( index & 1 ) {
- // odd runs index => how much to jump forward
- address += ((count-1) * 4);
- }
- else {
- // even runs index => how many pointers in a row that need rebasing
- for (int i=0; i < count; ++i) {
- handler(address);
- address += 4;
- }
- }
- }
- ++index;
- }
- // 4-byte align for next run
- index = (index+3) & (-4);
- rr = (RebaseRuns*)(&rr->runs[index]);
- }
-}
MachOAnalyzer::ObjCInfo MachOAnalyzer::getObjCInfo() const
{
@@ -3310,22 +4295,22 @@
result.protocolDefCount = 0;
const uint32_t ptrSize = pointerSize();
- forEachSection(^(const Header::SectionInfo& sectInfo, bool& stop) {
- if ( sectInfo.segmentName.starts_with("__DATA") ) {
- if ( sectInfo.sectionName == "__objc_selrefs" )
- result.selRefCount += (sectInfo.size/ptrSize);
- else if ( sectInfo.sectionName == "__objc_classlist" )
- result.classDefCount += (sectInfo.size/ptrSize);
- else if ( sectInfo.sectionName == "__objc_protolist" )
- result.protocolDefCount += (sectInfo.size/ptrSize);
- }
- else if ( (this->cputype == CPU_TYPE_I386) && (sectInfo.sectionName == "__OBJC") ) {
- if ( sectInfo.sectionName == "__message_refs" )
- result.selRefCount += (sectInfo.size/4);
- else if ( sectInfo.sectionName == "__class" )
- result.classDefCount += (sectInfo.size/48);
- else if ( sectInfo.sectionName == "__protocol" )
- result.protocolDefCount += (sectInfo.size/20);
+ forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ if ( strncmp(sectInfo.segInfo.segName, "__DATA", 6) == 0 ) {
+ if ( strcmp(sectInfo.sectName, "__objc_selrefs") == 0 )
+ result.selRefCount += (sectInfo.sectSize/ptrSize);
+ else if ( strcmp(sectInfo.sectName, "__objc_classlist") == 0 )
+ result.classDefCount += (sectInfo.sectSize/ptrSize);
+ else if ( strcmp(sectInfo.sectName, "__objc_protolist") == 0 )
+ result.protocolDefCount += (sectInfo.sectSize/ptrSize);
+ }
+ else if ( (this->cputype == CPU_TYPE_I386) && (strcmp(sectInfo.segInfo.segName, "__OBJC") == 0) ) {
+ if ( strcmp(sectInfo.sectName, "__message_refs") == 0 )
+ result.selRefCount += (sectInfo.sectSize/4);
+ else if ( strcmp(sectInfo.sectName, "__class") == 0 )
+ result.classDefCount += (sectInfo.sectSize/48);
+ else if ( strcmp(sectInfo.sectName, "__protocol") == 0 )
+ result.protocolDefCount += (sectInfo.sectSize/20);
}
});
@@ -3406,32 +4391,32 @@
uint32_t fairplayTextOffsetStart;
uint32_t fairplayTextOffsetEnd;
uint32_t fairplaySize;
- if ( ((const Header*)this)->isFairPlayEncrypted(fairplayTextOffsetStart, fairplaySize) ) {
+ if ( isFairPlayEncrypted(fairplayTextOffsetStart, fairplaySize) ) {
fairplayTextOffsetEnd = fairplayTextOffsetStart + fairplaySize;
} else {
fairplayTextOffsetEnd = 0;
}
result = PrintableStringResult::UnknownSection;
- forEachSection(^(const Header::SegmentInfo& segInfo, const Header::SectionInfo §Info, bool &stop) {
- if ( stringVMAddr < sectInfo.address ) {
+ forEachSection(^(const MachOAnalyzer::SectionInfo §Info, bool malformedSectionRange, bool &stop) {
+ if ( stringVMAddr < sectInfo.sectAddr ) {
return;
}
- if ( stringVMAddr >= ( sectInfo.address + sectInfo.size) ) {
+ if ( stringVMAddr >= ( sectInfo.sectAddr + sectInfo.sectSize) ) {
return;
}
// We can't scan this section if its protected
- if ( segInfo.isProtected() ) {
+ if ( sectInfo.segInfo.isProtected ) {
result = PrintableStringResult::ProtectedSection;
stop = true;
return;
}
// We can't scan this section if it overlaps with the fairplay range
- if ( fairplayTextOffsetEnd < sectInfo.fileOffset ) {
+ if ( fairplayTextOffsetEnd < sectInfo.sectFileOffset ) {
// Fairplay range ends before section
- } else if ( fairplayTextOffsetStart > (sectInfo.fileOffset + sectInfo.size) ) {
+ } else if ( fairplayTextOffsetStart > (sectInfo.sectFileOffset + sectInfo.sectSize) ) {
// Fairplay range starts after section
} else {
// Must overlap
@@ -3447,7 +4432,7 @@
#if BUILDING_SHARED_CACHE_UTIL || BUILDING_DYLDINFO
// The shared cache coalesces strings in to their own section.
// Assume its a valid pointer
- if (result == PrintableStringResult::UnknownSection && this->inDyldCache()) {
+ if (result == PrintableStringResult::UnknownSection) {
result = PrintableStringResult::CanPrint;
return (const char*)(stringVMAddr + getSlide());
}
@@ -3459,58 +4444,22 @@
}
void MachOAnalyzer::forEachObjCClass(uint64_t classListRuntimeOffset, uint64_t classListCount,
- const VMAddrConverter& vmAddrConverter,
- ClassCallback& callback) const {
-#if !BUILDING_DYLD
- // ObjC patching needs the bind targets for interposable references to the classes
- // build targets table
- STACK_ALLOC_OVERFLOW_SAFE_ARRAY(uint64_t, bindTargets, 32);
- if ( this->hasChainedFixups() ) {
- intptr_t slide = this->getSlide();
- __block Diagnostics diag;
- this->forEachBindTarget(diag, false, ^(const BindTargetInfo& info, bool& stop) {
- if ( diag.hasError() ) {
- stop = true;
- return;
- }
-
- if ( info.libOrdinal == BIND_SPECIAL_DYLIB_SELF ) {
- void* result = nullptr;
- bool resultPointsToInstructions = false;
- if ( this->hasExportedSymbol(info.symbolName, nullptr, &result, &resultPointsToInstructions) ) {
- uint64_t resultVMAddr = (uint64_t)result - (uint64_t)slide;
- bindTargets.push_back(resultVMAddr);
- } else {
- bindTargets.push_back(0);
- }
- } else {
- bindTargets.push_back(0);
- }
- }, ^(const MachOAnalyzer::BindTargetInfo& info, bool& stop) {
- });
- }
-#else
- // dyld always analyzes objc after fixups, so we don't need the bind targets
- Array<uint64_t> bindTargets;
-#endif
-
+ const VMAddrConverter& vmAddrConverter, ClassCallback& callback) const {
const uint64_t ptrSize = pointerSize();
const uint8_t* arrayBase = (uint8_t*)this + classListRuntimeOffset;
if ( ptrSize == 8 ) {
typedef uint64_t PtrTy;
__block bool stop = false;
for ( uint64_t i = 0; i != classListCount; ++i ) {
- uint64_t classVMAddr = vmAddrConverter.convertToVMAddr(*(PtrTy*)(arrayBase + (i * sizeof(PtrTy))), bindTargets);
- parseObjCClass(vmAddrConverter, classVMAddr, bindTargets,
- ^(uint64_t classSuperclassVMAddr, uint64_t classDataVMAddr, const ObjCClassInfo& objcClass) {
+ uint64_t classVMAddr = vmAddrConverter.convertToVMAddr(*(PtrTy*)(arrayBase + (i * sizeof(PtrTy))));
+ parseObjCClass(vmAddrConverter, classVMAddr, ^(uint64_t classSuperclassVMAddr, uint64_t classDataVMAddr, const ObjCClassInfo& objcClass) {
callback(classVMAddr, classSuperclassVMAddr, classDataVMAddr, objcClass, false, stop);
if ( stop )
return;
// Then parse and call for the metaclass
uint64_t isaVMAddr = objcClass.isaVMAddr;
- parseObjCClass(vmAddrConverter, isaVMAddr, bindTargets,
- ^(uint64_t metaclassSuperclassVMAddr, uint64_t metaclassDataVMAddr, const ObjCClassInfo& objcMetaclass) {
+ parseObjCClass(vmAddrConverter, isaVMAddr, ^(uint64_t metaclassSuperclassVMAddr, uint64_t metaclassDataVMAddr, const ObjCClassInfo& objcMetaclass) {
callback(isaVMAddr, metaclassSuperclassVMAddr, metaclassDataVMAddr, objcMetaclass, true, stop);
});
});
@@ -3521,17 +4470,15 @@
typedef uint32_t PtrTy;
__block bool stop = false;
for ( uint64_t i = 0; i != classListCount; ++i ) {
- uint64_t classVMAddr = vmAddrConverter.convertToVMAddr(*(PtrTy*)(arrayBase + (i * sizeof(PtrTy))), bindTargets);
- parseObjCClass(vmAddrConverter, classVMAddr, bindTargets,
- ^(uint64_t classSuperclassVMAddr, uint64_t classDataVMAddr, const ObjCClassInfo& objcClass) {
+ uint64_t classVMAddr = vmAddrConverter.convertToVMAddr(*(PtrTy*)(arrayBase + (i * sizeof(PtrTy))));
+ parseObjCClass(vmAddrConverter, classVMAddr, ^(uint64_t classSuperclassVMAddr, uint64_t classDataVMAddr, const ObjCClassInfo& objcClass) {
callback(classVMAddr, classSuperclassVMAddr, classDataVMAddr, objcClass, false, stop);
if ( stop )
return;
// Then parse and call for the metaclass
uint64_t isaVMAddr = objcClass.isaVMAddr;
- parseObjCClass(vmAddrConverter, isaVMAddr, bindTargets,
- ^(uint64_t metaclassSuperclassVMAddr, uint64_t metaclassDataVMAddr, const ObjCClassInfo& objcMetaclass) {
+ parseObjCClass(vmAddrConverter, isaVMAddr, ^(uint64_t metaclassSuperclassVMAddr, uint64_t metaclassDataVMAddr, const ObjCClassInfo& objcMetaclass) {
callback(isaVMAddr, metaclassSuperclassVMAddr, metaclassDataVMAddr, objcMetaclass, true, stop);
});
});
@@ -3545,7 +4492,7 @@
ClassCallback& callback) const {
uint64_t classListRuntimeOffset;
uint64_t classListSize;
- bool foundSection = ((const Header*)this)->findObjCDataSection("__objc_classlist", classListRuntimeOffset, classListSize);
+ bool foundSection = findObjCDataSection("__objc_classlist", classListRuntimeOffset, classListSize);
if ( !foundSection )
return;
@@ -3559,7 +4506,7 @@
}
void MachOAnalyzer::parseObjCClass(const VMAddrConverter& vmAddrConverter,
- uint64_t classVMAddr, const Array<uint64_t>& bindTargets,
+ uint64_t classVMAddr,
void (^handler)(uint64_t classSuperclassVMAddr,
uint64_t classDataVMAddr,
const ObjCClassInfo& objcClass)) const {
@@ -3591,7 +4538,7 @@
// First call the handler on the class
const objc_class_t* classPtr = (const objc_class_t*)(classVMAddr + slide);
const swift_class_metadata_t* swiftClassPtr = (const swift_class_metadata_t*)classPtr;
- objcClass.isaVMAddr = vmAddrConverter.convertToVMAddr(classPtr->isaVMAddr, bindTargets);
+ objcClass.isaVMAddr = vmAddrConverter.convertToVMAddr(classPtr->isaVMAddr);
objcClass.superclassVMAddr = vmAddrConverter.convertToVMAddr(classPtr->superclassVMAddr);
objcClass.methodCacheVMAddr = classPtr->methodCacheProperties == 0 ? 0 : vmAddrConverter.convertToVMAddr(classPtr->methodCacheProperties);
objcClass.dataVMAddr = vmAddrConverter.convertToVMAddr(classPtr->dataVMAddrAndFastFlags) & FAST_DATA_MASK;
@@ -3621,7 +4568,7 @@
// First call the handler on the class
const objc_class_t* classPtr = (const objc_class_t*)(classVMAddr + slide);
const swift_class_metadata_t* swiftClassPtr = (const swift_class_metadata_t*)classPtr;
- objcClass.isaVMAddr = vmAddrConverter.convertToVMAddr(classPtr->isaVMAddr, bindTargets);
+ objcClass.isaVMAddr = vmAddrConverter.convertToVMAddr(classPtr->isaVMAddr);
objcClass.superclassVMAddr = vmAddrConverter.convertToVMAddr(classPtr->superclassVMAddr);
objcClass.methodCacheVMAddr = classPtr->methodCacheProperties == 0 ? 0 : vmAddrConverter.convertToVMAddr(classPtr->methodCacheProperties);
objcClass.dataVMAddr = vmAddrConverter.convertToVMAddr(classPtr->dataVMAddrAndFastFlags) & FAST_DATA_MASK;
@@ -3633,48 +4580,6 @@
}
handler(classSuperclassVMAddr, classDataVMAddr, objcClass);
-}
-
-bool MachOAnalyzer::isSwiftClass(const void* classLocation) const
-{
- const uint64_t ptrSize = pointerSize();
- if ( ptrSize == 8 ) {
- struct objc_class_t {
- uint64_t isaVMAddr;
- uint64_t superclassVMAddr;
- uint64_t methodCacheBuckets;
- uint64_t methodCacheProperties;
- uint64_t dataVMAddrAndFastFlags;
- };
- enum : uint64_t {
- FAST_DATA_MASK = 0x00007ffffffffff8ULL
- };
-
- const objc_class_t* classPtr = (const objc_class_t*)classLocation;
- bool isSwiftLegacy = classPtr->dataVMAddrAndFastFlags & ObjCClassInfo::FAST_IS_SWIFT_LEGACY;
- bool isSwiftStable = classPtr->dataVMAddrAndFastFlags & ObjCClassInfo::FAST_IS_SWIFT_STABLE;
-
- // The Swift class flags are only present if the class is swift
- return (isSwiftLegacy || isSwiftStable);
- } else {
- struct objc_class_t {
- uint32_t isaVMAddr;
- uint32_t superclassVMAddr;
- uint32_t methodCacheBuckets;
- uint32_t methodCacheProperties;
- uint32_t dataVMAddrAndFastFlags;
- };
- enum : uint32_t {
- FAST_DATA_MASK = 0xfffffffcUL
- };
-
- const objc_class_t* classPtr = (const objc_class_t*)classLocation;
- bool isSwiftLegacy = classPtr->dataVMAddrAndFastFlags & ObjCClassInfo::FAST_IS_SWIFT_LEGACY;
- bool isSwiftStable = classPtr->dataVMAddrAndFastFlags & ObjCClassInfo::FAST_IS_SWIFT_STABLE;
-
- // The Swift class flags are only present if the class is swift
- return (isSwiftLegacy || isSwiftStable);
- }
}
void MachOAnalyzer::forEachObjCCategory(uint64_t categoryListRuntimeOffset, uint64_t categoryListCount,
@@ -3740,7 +4645,7 @@
CategoryCallback& callback) const {
uint64_t categoryListRuntimeOffset;
uint64_t categoryListSize;
- bool foundSection = ((const Header*)this)->findObjCDataSection("__objc_catlist", categoryListRuntimeOffset, categoryListSize);
+ bool foundSection = findObjCDataSection("__objc_catlist", categoryListRuntimeOffset, categoryListSize);
if ( !foundSection )
return;
@@ -3835,7 +4740,7 @@
ProtocolCallback& callback) const {
uint64_t protocolListRuntimeOffset;
uint64_t protocolListSize;
- bool foundSection = ((const Header*)this)->findObjCDataSection("__objc_protolist", protocolListRuntimeOffset, protocolListSize);
+ bool foundSection = findObjCDataSection("__objc_protolist", protocolListRuntimeOffset, protocolListSize);
if ( !foundSection )
return;
@@ -3848,38 +4753,6 @@
forEachObjCProtocol(protocolListRuntimeOffset, protocolListSize / ptrSize, vmAddrConverter, callback);
}
-static void ignorePreoptimizedListsOfLists(uint64_t& listVMAddr, intptr_t slide)
-{
- // If this is a list of lists, then we likely just want the class list. So go to the end which is where
- // we emitted it
- if ( listVMAddr & 1 ) {
- struct ListOfListsEntry {
- union {
- struct {
- uint64_t imageIndex: 16;
- int64_t offset: 48;
- };
- struct {
- uint32_t entsize;
- uint32_t count;
- };
- };
- };
-
- listVMAddr = listVMAddr & ~1;
- const ListOfListsEntry* listHeader = (const ListOfListsEntry*)(listVMAddr + slide);
- if ( listHeader->count != 0 ) {
- const ListOfListsEntry& listEntry = (listHeader + 1)[listHeader->count - 1];
-
- // The list entry is a relative offset to the target
- // Work out the VMAddress of that target
- uint64_t listEntryVMOffset{(uint64_t)&listEntry - (uint64_t)listHeader};
- uint64_t listEntryVMAddr = listVMAddr + listEntryVMOffset;
- listVMAddr = listEntryVMAddr + (uint64_t)listEntry.offset;
- }
- }
-}
-
void MachOAnalyzer::forEachObjCMethod(uint64_t methodListVMAddr, const VMAddrConverter& vmAddrConverter,
uint64_t sharedCacheRelativeSelectorBaseVMAddress,
void (^handler)(uint64_t methodVMAddr, const ObjCMethod& method, bool& stop)) const
@@ -3889,8 +4762,6 @@
const uint64_t ptrSize = pointerSize();
intptr_t slide = getSlide();
-
- ignorePreoptimizedListsOfLists(methodListVMAddr, slide);
if ( ptrSize == 8 ) {
typedef uint64_t PtrTy;
@@ -4059,8 +4930,6 @@
const uint64_t ptrSize = pointerSize();
intptr_t slide = getSlide();
- ignorePreoptimizedListsOfLists(propertyListVMAddr, slide);
-
if ( ptrSize == 8 ) {
typedef uint64_t PtrTy;
struct property_list_t {
@@ -4128,8 +4997,6 @@
auto ptrSize = pointerSize();
intptr_t slide = getSlide();
-
- ignorePreoptimizedListsOfLists(protocolListVMAddr, slide);
if ( ptrSize == 8 ) {
typedef uint64_t PtrTy;
@@ -4222,7 +5089,7 @@
void MachOAnalyzer::forEachObjCSelectorReference(uint64_t selRefsRuntimeOffset, uint64_t selRefsCount, const VMAddrConverter& vmAddrConverter,
void (^handler)(uint64_t selRefVMAddr, uint64_t selRefTargetVMAddr, bool& stop)) const
{
- uint64_t baseAddress = ((const Header*)this)->preferredLoadAddress();
+ uint64_t baseAddress = preferredLoadAddress();
const uint64_t ptrSize = pointerSize();
const uint8_t* selRefs = (uint8_t*)this + selRefsRuntimeOffset;
if ( ptrSize == 8 ) {
@@ -4248,13 +5115,33 @@
}
}
+bool MachOAnalyzer::findObjCDataSection(const char *sectionName, uint64_t& sectionRuntimeOffset, uint64_t& sectionSize) const
+{
+ uint64_t baseAddress = preferredLoadAddress();
+
+ __block bool foundSection = false;
+ forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ if ( (strcmp(sectInfo.segInfo.segName, "__DATA") != 0) &&
+ (strcmp(sectInfo.segInfo.segName, "__DATA_CONST") != 0) &&
+ (strcmp(sectInfo.segInfo.segName, "__DATA_DIRTY") != 0) )
+ return;
+ if ( strcmp(sectInfo.sectName, sectionName) != 0 )
+ return;
+ foundSection = true;
+ sectionRuntimeOffset = sectInfo.sectAddr - baseAddress;
+ sectionSize = sectInfo.sectSize;
+ stop = true;
+ });
+ return foundSection;
+}
+
void MachOAnalyzer::forEachObjCSelectorReference(Diagnostics& diag, const VMAddrConverter& vmAddrConverter,
void (^handler)(uint64_t selRefVMAddr, uint64_t selRefTargetVMAddr, bool& stop)) const
{
uint64_t selRefsRuntimeOffset;
uint64_t selRefsSize;
- bool foundSection = ((const Header*)this)->findObjCDataSection("__objc_selrefs", selRefsRuntimeOffset, selRefsSize);
+ bool foundSection = findObjCDataSection("__objc_selrefs", selRefsRuntimeOffset, selRefsSize);
if ( !foundSection )
return;
@@ -4269,18 +5156,22 @@
void MachOAnalyzer::forEachObjCMethodName(void (^handler)(const char* methodName)) const {
intptr_t slide = getSlide();
- forEachSection(^(const Header::SegmentInfo& segInfo, const Header::SectionInfo& sectInfo, bool& stop) {
- if ( sectInfo.sectionName != "__TEXT" )
+ forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ if ( strcmp(sectInfo.segInfo.segName, "__TEXT") != 0 )
return;
- if ( sectInfo.sectionName != "__objc_methname" )
+ if ( strcmp(sectInfo.sectName, "__objc_methname") != 0 )
return;
- if ( segInfo.isProtected() || ( (sectInfo.flags & SECTION_TYPE) != S_CSTRING_LITERALS ) ) {
+ if ( sectInfo.segInfo.isProtected || ( (sectInfo.sectFlags & SECTION_TYPE) != S_CSTRING_LITERALS ) ) {
stop = true;
return;
}
-
- const char* content = (const char*)(sectInfo.address + slide);
- uint64_t sectionSize = sectInfo.size;
+ if ( malformedSectionRange ) {
+ stop = true;
+ return;
+ }
+
+ const char* content = (const char*)(sectInfo.sectAddr + slide);
+ uint64_t sectionSize = sectInfo.sectSize;
const char* s = (const char*)content;
const char* end = s + sectionSize;
@@ -4320,21 +5211,40 @@
}
#endif
+
+bool MachOAnalyzer::hasObjCMessageReferences() const {
+
+ __block bool foundSection = false;
+ forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ if ( strncmp(sectInfo.segInfo.segName, "__DATA", 6) != 0 )
+ return;
+ if ( strcmp(sectInfo.sectName, "__objc_msgrefs") != 0 )
+ return;
+ foundSection = true;
+ stop = true;
+ });
+ return foundSection;
+}
+
const MachOAnalyzer::ObjCImageInfo* MachOAnalyzer::objcImageInfo() const {
- uintptr_t slide = getSlide();
+ int64_t slide = getSlide();
__block bool foundInvalidObjCImageInfo = false;
__block const ObjCImageInfo* imageInfo = nullptr;
- forEachSection(^(const Header::SectionInfo& sectionInfo, bool& stop) {
- if ( !sectionInfo.segmentName.starts_with("__DATA") )
+ forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo& sectionInfo, bool malformedSectionRange, bool& stop) {
+ if ( strncmp(sectionInfo.segInfo.segName, "__DATA", 6) != 0 )
return;
- if ( sectionInfo.sectionName != "__objc_imageinfo" )
+ if (strcmp(sectionInfo.sectName, "__objc_imageinfo") != 0)
return;
- if ( sectionInfo.size != 8 ) {
+ if ( malformedSectionRange ) {
stop = true;
return;
}
- imageInfo = (const ObjCImageInfo*)((uintptr_t)sectionInfo.address + slide);
+ if ( sectionInfo.sectSize != 8 ) {
+ stop = true;
+ return;
+ }
+ imageInfo = (const ObjCImageInfo*)(sectionInfo.sectAddr + slide);
if ( (imageInfo->flags & ObjCImageInfo::dyldPreoptimized) != 0 ) {
foundInvalidObjCImageInfo = true;
stop = true;
@@ -4347,9 +5257,301 @@
return imageInfo;
}
+// A 32-bit relative pointer to a value. The offset value is either:
+// - a direct 32-bit offset to the value, if the low bit is 0, or
+// - an offset to a pointer sized slot, if the low bit is 1
+class SwiftRelativePointer
+{
+public:
+ uint64_t getTargetRuntimeOffset(const MachOAnalyzer* ma,
+ const MachOAnalyzer::VMAddrConverter& vmAddrConverter,
+ uint32_t ptrSize) const {
+ if ( (relativeOffset & 0x1) == 0 ) {
+ // Relative offset directly to the target value
+ uint64_t relativePointerRuntimeOffset = (uint64_t)this - (uint64_t)ma;
+ return relativePointerRuntimeOffset + (int64_t)relativeOffset;
+ } else {
+ // Relative offset to a pointer. The pointer contains the target value
+ int32_t offset = relativeOffset & ~0x1ULL;
+ uint64_t relativePointerRuntimeOffset = (uint64_t)this - (uint64_t)ma;
+ relativePointerRuntimeOffset += (int64_t)offset;
+ uint64_t pointerValue = 0;
+ if ( ptrSize == 8 ) {
+ pointerValue = *(uint64_t*)((uint8_t*)ma + relativePointerRuntimeOffset);
+ } else {
+ pointerValue = *(uint32_t*)((uint8_t*)ma + relativePointerRuntimeOffset);
+ }
+ // Convert from the encoded pointer to a runtime offset
+ return vmAddrConverter.convertToVMAddr(pointerValue) - vmAddrConverter.preferredLoadAddress;
+ }
+ }
+
+private:
+ int32_t relativeOffset = 0;
+};
+
+class SwiftProtocolConformanceFlags
+{
+public:
+ // Taken from MetadataValues.h
+ enum class TypeMetadataKind : uint32_t {
+ mask = 0x7 << 3, // 8 type reference kinds
+ shift = 3,
+ };
+
+ // Taken from MetadataValues.h
+ enum class TypeReferenceKind : uint32_t {
+ // The conformance is for a nominal type referenced directly;
+ // getTypeDescriptor() points to the type context descriptor.
+ directTypeDescriptor = 0x00,
+
+ // The conformance is for a nominal type referenced indirectly;
+ // getTypeDescriptor() points to the type context descriptor.
+ indirectTypeDescriptor = 0x01,
+
+ // The conformance is for an Objective-C class that should be looked up
+ // by class name.
+ directObjCClassName = 0x02,
+
+ // The conformance is for an Objective-C class that has no nominal type
+ // descriptor.
+ // getIndirectObjCClass() points to a variable that contains the pointer to
+ // the class object, which then requires a runtime call to get metadata.
+ //
+ // On platforms without Objective-C interoperability, this case is
+ // unused.
+ indirectObjCClass = 0x03,
+ };
+
+ TypeReferenceKind typeReferenceKind() const {
+ return (TypeReferenceKind)((flags & (uint32_t)TypeMetadataKind::mask) >> (uint32_t)TypeMetadataKind::shift);
+ }
+private:
+ uint32_t flags = 0;
+};
+
+// This represents the type descriptors pointed to by protocol conformances
+class TypeContextDescriptor
+{
+public:
+
+ const SwiftRelativePointer& name() const {
+ return _name;
+ }
+
+ bool isForeignMetadata() const {
+ // The botton 2 bits have the flags
+ return (getKindSpecificFlags() & 0x3) == ForeignMetadataInitialization;
+ }
+
+ bool hasImportInfo() const {
+ // Bit 2 tells us if we have import info, ie, a name containing NULLs
+ return (getKindSpecificFlags() & (1 << 2)) != 0;
+ }
+
+private:
+
+ enum TypeContextDescriptorFlags : uint16_t {
+ ForeignMetadataInitialization = 0x2
+ };
+
+ // The most significant two bytes of the flags word, which can have
+ // kind-specific meaning.
+ uint16_t getKindSpecificFlags() const {
+ return (_flags >> 16u) & 0xFFFFu;
+ }
+
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wunused-private-field"
+ uint32_t _flags;
+ int32_t _parent;
+ SwiftRelativePointer _name;
+ int32_t _accessFunction;
+ int32_t _fields;
+#pragma clang diagnostic pop
+};
+
+// A 32-bit relative pointer to a value
+// The type the pointer depends on the SwiftProtocolConformanceFlags::TypeReferenceKind
+class SwiftTypeRefPointer
+{
+public:
+ // A type descriptor is one of the types described by TypeReferenceKind.
+ // This works out which of the give types we are, and populates the given out parameters to match
+ void getTypeDescriptor(SwiftProtocolConformanceFlags::TypeReferenceKind refKind,
+ const MachOAnalyzer* ma,
+ const MachOAnalyzer::VMAddrConverter& vmAddrConverter,
+ uint32_t ptrSize,
+ bool canAnalyzeIndirectPointers,
+ uint64_t& descriptorRuntimeOffset,
+ uint64_t& objcClassNameRuntimeOffset,
+ uint64_t& objcClassRuntimeOffset,
+ uint64_t& foreignMetadataNameRuntimeOffset,
+ bool& foreignMetadataNameHasImportInfo) const {
+ descriptorRuntimeOffset = 0;
+ objcClassNameRuntimeOffset = 0;
+ objcClassRuntimeOffset = 0;
+ switch ( refKind ) {
+ case SwiftProtocolConformanceFlags::TypeReferenceKind::directTypeDescriptor: {
+ // Relative offset directly to the target value
+ uint64_t relativePointerRuntimeOffset = (uint64_t)this - (uint64_t)ma;
+ descriptorRuntimeOffset = relativePointerRuntimeOffset + (int64_t)relativeOffset;
+ break;
+ }
+ case SwiftProtocolConformanceFlags::TypeReferenceKind::indirectTypeDescriptor: {
+ // Relative offset to a pointer. The pointer contains the target value
+ uint64_t relativePointerRuntimeOffset = (uint64_t)this - (uint64_t)ma;
+ relativePointerRuntimeOffset += (int64_t)relativeOffset;
+ uint64_t pointerValue = 0;
+ if ( ptrSize == 8 ) {
+ pointerValue = *(uint64_t*)((uint8_t*)ma + relativePointerRuntimeOffset);
+ } else {
+ pointerValue = *(uint32_t*)((uint8_t*)ma + relativePointerRuntimeOffset);
+ }
+ // This might be a pointer to a missing weak import. If that is the case, just skip it
+ if ( pointerValue == 0 )
+ break;
+ // Convert from the encoded pointer to a runtime offset
+ descriptorRuntimeOffset = vmAddrConverter.convertToVMAddr(pointerValue) - vmAddrConverter.preferredLoadAddress;
+ break;
+ }
+ case SwiftProtocolConformanceFlags::TypeReferenceKind::directObjCClassName: {
+ // Relative offset directly to the class name string
+ uint64_t relativePointerRuntimeOffset = (uint64_t)this - (uint64_t)ma;
+ objcClassNameRuntimeOffset = relativePointerRuntimeOffset + (int64_t)relativeOffset;
+ break;
+ }
+ case SwiftProtocolConformanceFlags::TypeReferenceKind::indirectObjCClass: {
+ // Relative offset to a pointer. The pointer contains the target class value
+ uint64_t relativePointerRuntimeOffset = (uint64_t)this - (uint64_t)ma;
+ relativePointerRuntimeOffset += (int64_t)relativeOffset;
+ uint64_t pointerValue = 0;
+ if ( ptrSize == 8 ) {
+ pointerValue = *(uint64_t*)((uint8_t*)ma + relativePointerRuntimeOffset);
+ } else {
+ pointerValue = *(uint32_t*)((uint8_t*)ma + relativePointerRuntimeOffset);
+ }
+ // This might be a pointer to a missing weak import. If that is the case, just skip it
+ if ( pointerValue == 0 )
+ break;
+ // Convert from the encoded pointer to a runtime offset
+ objcClassRuntimeOffset = vmAddrConverter.convertToVMAddr(pointerValue) - vmAddrConverter.preferredLoadAddress;
+ break;
+ }
+ }
+
+ // Type descriptors might be foreign. This means that the runtime needs to use their name to identify them
+ if ( descriptorRuntimeOffset != 0 ) {
+ // Once we've bound all dylibs in the shared cache, we can visit indirect type descriptors, which are binds.
+ // Prior to that, just add a placeholder value for foriegn metadata to suggest that it might be set. We don't know
+ // for sure, but need to make pace for it
+ bool canAnalyze = canAnalyzeIndirectPointers;
+ if ( refKind == SwiftProtocolConformanceFlags::TypeReferenceKind::directTypeDescriptor )
+ canAnalyze = true;
+
+ if ( canAnalyze ) {
+ const TypeContextDescriptor* typeDesc = (const TypeContextDescriptor*)((uint8_t*)ma + descriptorRuntimeOffset);
+ if ( typeDesc->isForeignMetadata() ) {
+ foreignMetadataNameRuntimeOffset = typeDesc->name().getTargetRuntimeOffset(ma, vmAddrConverter, ptrSize);
+ foreignMetadataNameHasImportInfo = typeDesc->hasImportInfo();
+ // printf("Name: %s\n", (const char*)ma + foreignMetadatNameRuntimeOffset);
+ }
+ } else {
+ // HACK: Add a placeholder
+ foreignMetadataNameRuntimeOffset = ~0ULL;
+ }
+ }
+ }
+
+private:
+ int32_t relativeOffset = 0;
+};
+
+void MachOAnalyzer::forEachSwiftProtocolConformance(uint64_t protocolConformanceListRuntimeOffset,
+ uint64_t numProtocolConformances,
+ const VMAddrConverter& vmAddrConverter,
+ bool canAnalyzeIndirectPointers,
+ void (^handler)(uint64_t protocolConformanceRuntimeOffset,
+ const SwiftProtocolConformance& protocolConformance,
+ bool& stopProtocolConformance)) const
+{
+ // Protocol conformances are a 32-bit offset from the list entry
+ const int32_t* protocolConformanceList = (const int32_t*)((uint8_t*)this + protocolConformanceListRuntimeOffset);
+
+ struct Conformance
+ {
+ const SwiftRelativePointer protocolRelativePointer;
+ const SwiftTypeRefPointer typeRef;
+ int32_t witnessTable;
+ const SwiftProtocolConformanceFlags flags;
+ };
+
+ const uint32_t ptrSize = MachOFile::pointerSize();
+ for (uint64_t i = 0; i != numProtocolConformances; ++i) {
+ int32_t relativeOffset = protocolConformanceList[i];
+ uint64_t listEntryRuntimeOffset = protocolConformanceListRuntimeOffset + (i * 4);
+ uint64_t protocolConformanceRuntimeOffset = listEntryRuntimeOffset + (int64_t)relativeOffset;
+
+ const Conformance* conformance = (const Conformance*)((uint8_t*)this + protocolConformanceRuntimeOffset);
+
+ SwiftProtocolConformance swiftProtocolConformance;
+ swiftProtocolConformance.protocolRuntimeOffset = conformance->protocolRelativePointer.getTargetRuntimeOffset(this, vmAddrConverter, ptrSize);
+ conformance->typeRef.getTypeDescriptor(conformance->flags.typeReferenceKind(), this, vmAddrConverter, ptrSize,
+ canAnalyzeIndirectPointers,
+ swiftProtocolConformance.typeConformanceRuntimeOffset,
+ swiftProtocolConformance.typeConformanceObjCClassNameRuntimeOffset,
+ swiftProtocolConformance.typeConformanceObjCClassRuntimeOffset,
+ swiftProtocolConformance.foreignMetadataNameRuntimeOffset,
+ swiftProtocolConformance.foreignMetadataNameHasImportInfo);
+
+ bool stopProtocolConformance = false;
+ handler(protocolConformanceRuntimeOffset, swiftProtocolConformance, stopProtocolConformance);
+ if ( stopProtocolConformance )
+ break;
+ }
+}
+
+void MachOAnalyzer::forEachSwiftProtocolConformance(Diagnostics& diag, const VMAddrConverter& vmAddrConverter,
+ bool canAnalyzeIndirectPointers,
+ void (^handler)(uint64_t protocolConformanceRuntimeOffset,
+ const SwiftProtocolConformance& protocolConformance,
+ bool& stopProtocolConformance)) const
+{
+ uint64_t protocolListRuntimeOffset;
+ uint64_t protocolListSize;
+ const void* sectionContent = findSectionContent("__TEXT", "__swift5_proto", protocolListSize);
+ if ( sectionContent == nullptr )
+ return;
+ protocolListRuntimeOffset = (uint64_t)sectionContent - (uint64_t)this;
+
+ // Each entry is a 32-bit offset to a protocol conformance
+ const uint32_t entrySize = 4;
+ if ( (protocolListSize % entrySize) != 0 ) {
+ diag.error("Invalid Swift protocol conformance section size");
+ return;
+ }
+
+ forEachSwiftProtocolConformance(protocolListRuntimeOffset, protocolListSize / entrySize, vmAddrConverter, canAnalyzeIndirectPointers, handler);
+}
+
+uint32_t MachOAnalyzer::loadCommandsFreeSpace() const
+{
+ __block uint32_t firstSectionFileOffset = 0;
+ __block uint32_t firstSegmentFileOffset = 0;
+ forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ firstSectionFileOffset = sectInfo.sectFileOffset;
+ firstSegmentFileOffset = (uint32_t)sectInfo.segInfo.fileOffset;
+ stop = true;
+ });
+
+ uint32_t headerSize = (this->magic == MH_MAGIC_64) ? sizeof(mach_header_64) : sizeof(mach_header);
+ uint32_t existSpaceUsed = this->sizeofcmds + headerSize;
+ return firstSectionFileOffset - firstSegmentFileOffset - existSpaceUsed;
+}
+
void MachOAnalyzer::forEachWeakDef(Diagnostics& diag,
void (^handler)(const char* symbolName, uint64_t imageOffset, bool isFromExportTrie)) const {
- uint64_t baseAddress = ((const Header*)this)->preferredLoadAddress();
+ uint64_t baseAddress = preferredLoadAddress();
forEachGlobalSymbol(diag, ^(const char *symbolName, uint64_t n_value, uint8_t n_type, uint8_t n_sect, uint16_t n_desc, bool &stop) {
if ( (n_desc & N_WEAK_DEF) != 0 ) {
handler(symbolName, n_value - baseAddress, false);
@@ -4367,21 +5569,111 @@
});
}
+uint32_t MachOAnalyzer::getFixupsLoadCommandFileOffset() const
+{
+ Diagnostics diag;
+ LinkEditInfo linkeditInfo;
+ getLinkEditLoadCommands(diag, linkeditInfo);
+ if ( diag.hasError() )
+ return 0;
+ if ( linkeditInfo.dyldInfo )
+ return (uint32_t)( (uint8_t*)linkeditInfo.dyldInfo - (uint8_t*)this );
+ else if ( linkeditInfo.chainedFixups )
+ return (uint32_t)( (uint8_t*)linkeditInfo.chainedFixups - (uint8_t*)this );
+ return 0;
+}
+
+
+bool MachOAnalyzer::inCodeSection(uint32_t runtimeOffset) const
+{
+ // only needed for arm64e code to know to sign pointers
+ if ( (this->cputype != CPU_TYPE_ARM64) || (this->maskedCpuSubtype() != CPU_SUBTYPE_ARM64E) )
+ return false;
+
+ __block bool result = false;
+ uint64_t baseAddress = this->preferredLoadAddress();
+ this->forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ if ( ((sectInfo.sectAddr-baseAddress) <= runtimeOffset) && (runtimeOffset < (sectInfo.sectAddr+sectInfo.sectSize-baseAddress)) ) {
+ result = ( (sectInfo.sectFlags & S_ATTR_PURE_INSTRUCTIONS) || (sectInfo.sectFlags & S_ATTR_SOME_INSTRUCTIONS) );
+ stop = true;
+ }
+ });
+ return result;
+}
+
+// These are mangled symbols for all the variants of operator new and delete
+// which a main executable can define (non-weak) and override the
+// weak-def implementation in the OS.
+static const char* const sTreatAsWeak[] = {
+ "__Znwm", "__ZnwmRKSt9nothrow_t",
+ "__Znam", "__ZnamRKSt9nothrow_t",
+ "__ZdlPv", "__ZdlPvRKSt9nothrow_t", "__ZdlPvm",
+ "__ZdaPv", "__ZdaPvRKSt9nothrow_t", "__ZdaPvm",
+ "__ZnwmSt11align_val_t", "__ZnwmSt11align_val_tRKSt9nothrow_t",
+ "__ZnamSt11align_val_t", "__ZnamSt11align_val_tRKSt9nothrow_t",
+ "__ZdlPvSt11align_val_t", "__ZdlPvSt11align_val_tRKSt9nothrow_t", "__ZdlPvmSt11align_val_t",
+ "__ZdaPvSt11align_val_t", "__ZdaPvSt11align_val_tRKSt9nothrow_t", "__ZdaPvmSt11align_val_t"
+};
+
+void MachOAnalyzer::forEachTreatAsWeakDef(void (^handler)(const char* symbolName))
+{
+ for (const char* sym : sTreatAsWeak)
+ handler(sym);
+}
+
+MachOAnalyzer::TLV_InitialContent MachOAnalyzer::forEachThreadLocalVariable(Diagnostics& diag, void (^handler)(TLV_Thunk& slot)) const
+{
+ __block TLV_InitialContent result = { 0, 0 };
+
+ // most images don't use thread locals, so quickly skip those
+ if ( !this->hasThreadLocalVariables() )
+ return result;
+
+ uintptr_t baseAddress = (uintptr_t)this->preferredLoadAddress();
+ intptr_t slide = (uintptr_t)this - baseAddress;
+ this->forEachSection(^(const SectionInfo& sectInfo, bool malformedSectionRange, bool& stop) {
+ switch (sectInfo.sectFlags & SECTION_TYPE) {
+ case S_THREAD_LOCAL_VARIABLES:
+ {
+ unsigned count = (unsigned)(sectInfo.sectSize / sizeof(TLV_Thunk));
+ const uint8_t* content = (uint8_t*)(sectInfo.sectAddr + slide);
+ TLV_Thunk* slotsStart = (TLV_Thunk*)content;
+ TLV_Thunk* slotsEnd = &slotsStart[count];
+ for (TLV_Thunk* p=slotsStart; p < slotsEnd; ++p) {
+ handler(*p);
+ }
+ }
+ break;
+ case S_THREAD_LOCAL_ZEROFILL:
+ case S_THREAD_LOCAL_REGULAR:
+ if ( result.runtimeOffset == 0 ) {
+ // first of N contiguous TLV template sections, record as if this was only section
+ result.runtimeOffset = sectInfo.sectAddr - baseAddress;
+ result.size = sectInfo.sectSize;
+ }
+ else {
+ // non-first of N contiguous TLV template sections, accumlate values
+ result.size = sectInfo.sectAddr + sectInfo.sectSize - baseAddress - result.runtimeOffset;
+ }
+ break;
+ }
+ });
+
+ return result;
+}
+
+
void MachOAnalyzer::forEachBindTarget(Diagnostics& diag, bool allowLazyBinds,
void (^handler)(const BindTargetInfo& info, bool& stop),
void (^overrideHandler)(const BindTargetInfo& info, bool& stop)) const
{
- if ( this->isPreload() )
- return;
if ( this->hasChainedFixups() )
this->forEachBindTarget_ChainedFixups(diag, handler);
else if ( this->hasOpcodeFixups() )
this->forEachBindTarget_Opcodes(diag, allowLazyBinds, handler, overrideHandler);
-#if SUPPORT_CLASSIC_RELOCS
else
this->forEachBindTarget_Relocations(diag, handler);
-#endif
}
struct WeakBindInfo
@@ -4400,7 +5692,7 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
@@ -4408,12 +5700,12 @@
{
__block unsigned targetIndex = 0;
__block BindTargetInfo targetInfo;
- BindDetailedHandler binder = ^(const char* opcodeName, const LinkEditInfo&, const Header::SegmentInfo segments[],
+ BindDetailedHandler binder = ^(const char* opcodeName, const LinkEditInfo&, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t pointerSize, uint8_t segmentIndex, uint64_t segmentOffset,
uint8_t type, const char* symbolName, bool weakImport, bool lazyBind,
uint64_t addend, bool targetOrAddendChanged, bool& stop) {
- uint64_t bindVmOffset = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t bindVmOffset = segments[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = bindVmOffset - leInfo.layout.textUnslidVMAddr;
if ( targetOrAddendChanged ) {
targetInfo.targetIndex = targetIndex++;
@@ -4439,15 +5731,15 @@
{
__block unsigned weakTargetIndex = 0;
__block BindTargetInfo weakTargetInfo;
- BindDetailedHandler weakBinder = ^(const char* opcodeName, const LinkEditInfo&, const Header::SegmentInfo segments[],
+ BindDetailedHandler weakBinder = ^(const char* opcodeName, const LinkEditInfo&, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t pointerSize, uint8_t segmentIndex, uint64_t segmentOffset,
uint8_t type, const char* symbolName, bool weakImport, bool lazyBind,
uint64_t addend, bool targetOrAddendChanged, bool& stop) {
- uint64_t bindVmOffset = segmentsInfo[segmentIndex].vmaddr + segmentOffset;
+ uint64_t bindVmOffset = segmentsInfo[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = bindVmOffset - leInfo.layout.textUnslidVMAddr;
- if ( (weakTargetIndex == 0) || (symbolName != weakTargetInfo.symbolName) || (strcmp(symbolName, weakTargetInfo.symbolName) != 0) || (weakTargetInfo.addend != addend) ) {
+ if ( (symbolName != weakTargetInfo.symbolName) || (strcmp(symbolName, weakTargetInfo.symbolName) != 0) || (weakTargetInfo.addend != addend) ) {
weakTargetInfo.targetIndex = weakTargetIndex++;
weakTargetInfo.libOrdinal = BIND_SPECIAL_DYLIB_WEAK_LOOKUP;
weakTargetInfo.symbolName = symbolName;
@@ -4518,7 +5810,6 @@
}
-#if SUPPORT_CLASSIC_RELOCS
// old binary, walk external relocations and indirect symbol table
void MachOAnalyzer::forEachBindTarget_Relocations(Diagnostics& diag, void (^handler)(const BindTargetInfo& info, bool& stop)) const
{
@@ -4527,14 +5818,14 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
__block unsigned targetIndex = 0;
this->forEachBind_Relocations(diag, leInfo, segmentsInfo, true,
- ^(const char* opcodeName, const LinkEditInfo&, const Header::SegmentInfo segments[],
+ ^(const char* opcodeName, const LinkEditInfo&, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t pointerSize, uint8_t segmentIndex, uint64_t segmentOffset,
uint8_t type, const char* symbolName, bool weakImport, bool lazyBind,
@@ -4552,7 +5843,6 @@
}
});
}
-#endif // SUPPORT_CLASSIC_RELOCS
void MachOAnalyzer::forEachBindLocation_Relocations(Diagnostics& diag, void (^handler)(uint64_t runtimeOffset, unsigned targetIndex, bool& stop)) const
@@ -4562,34 +5852,30 @@
if ( diag.hasError() )
return;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return;
__block int targetIndex = -1;
this->forEachBind_Relocations(diag, leInfo, segmentsInfo, false,
- ^(const char* opcodeName, const LinkEditInfo&, const Header::SegmentInfo segments[],
+ ^(const char* opcodeName, const LinkEditInfo&, const SegmentInfo segments[],
bool segIndexSet, bool libraryOrdinalSet, uint32_t dylibCount, int libOrdinal,
uint32_t pointerSize, uint8_t segmentIndex, uint64_t segmentOffset,
uint8_t type, const char* symbolName, bool weakImport, bool lazyBind,
uint64_t addend, bool targetOrAddendChanged, bool& stop) {
if ( targetOrAddendChanged )
++targetIndex;
- uint64_t bindVmOffset = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t bindVmOffset = segments[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = bindVmOffset - leInfo.layout.textUnslidVMAddr;
handler(runtimeOffset, targetIndex, stop);
});
}
-#if SUPPORT_CLASSIC_RELOCS
-bool MachOAnalyzer::forEachBind_Relocations(Diagnostics& diag, const LinkEditInfo& leInfo, const Header::SegmentInfo segmentsInfo[],
+
+bool MachOAnalyzer::forEachBind_Relocations(Diagnostics& diag, const LinkEditInfo& leInfo, const SegmentInfo segmentsInfo[],
bool supportPrivateExternsWorkaround, BindDetailedHandler handler) const
{
- // Firmare binaries won't have a dynSymTab
- if ( leInfo.dynSymTab == nullptr )
- return false;
-
const uint64_t relocsStartAddress = externalRelocBaseAddress(segmentsInfo, leInfo.layout.linkeditSegIndex);
const relocation_info* const relocsStart = (relocation_info*)getLinkEditContent(leInfo.layout, leInfo.dynSymTab->extreloff);
const relocation_info* const relocsEnd = &relocsStart[leInfo.dynSymTab->nextrel];
@@ -4608,7 +5894,7 @@
bool stop = false;
for (const relocation_info* reloc=relocsStart; (reloc < relocsEnd) && !stop; ++reloc) {
bool isBranch = false;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
if ( isKextBundle() ) {
// kext's may have other kinds of relocations, eg, branch relocs. Skip them
if ( isArch("x86_64") || isArch("x86_64h") ) {
@@ -4656,7 +5942,7 @@
else {
const char* symbolName = stringPool + strOffset;
bool weakImport = (n_desc & N_WEAK_REF);
- const uint8_t* content = (uint8_t*)this + segmentsInfo[segIndex].vmaddr - leInfo.layout.textUnslidVMAddr + segOffset;
+ const uint8_t* content = (uint8_t*)this + segmentsInfo[segIndex].vmAddr - leInfo.layout.textUnslidVMAddr + segOffset;
uint64_t addend = (reloc->r_length == 3) ? *((uint64_t*)content) : *((uint32_t*)content);
// Handle defined weak def symbols which need to get a special ordinal
if ( ((n_type & N_TYPE) == N_SECT) && ((n_type & N_EXT) != 0) && ((n_desc & N_WEAK_DEF) != 0) )
@@ -4695,7 +5981,6 @@
return false;
}
-#endif
void MachOAnalyzer::forEachBindLocation_Opcodes(Diagnostics& diag, void (^handler)(uint64_t runtimeOffset, unsigned targetIndex, bool& stop),
void (^overrideHandler)(uint64_t runtimeOffset, unsigned overrideBindTargetIndex, bool& stop)) const
@@ -4707,7 +5992,7 @@
});
}
-bool MachOAnalyzer::forEachBind_OpcodesLazy(Diagnostics& diag, const LinkEditInfo& leInfo, const Header::SegmentInfo segmentsInfo[], BindDetailedHandler handler) const
+bool MachOAnalyzer::forEachBind_OpcodesLazy(Diagnostics& diag, const LinkEditInfo& leInfo, const SegmentInfo segmentsInfo[], BindDetailedHandler handler) const
{
if ( (leInfo.dyldInfo == nullptr) || (leInfo.dyldInfo->lazy_bind_size == 0) )
return false;
@@ -4794,7 +6079,7 @@
-bool MachOAnalyzer::forEachBind_OpcodesWeak(Diagnostics& diag, const LinkEditInfo& leInfo, const Header::SegmentInfo segmentsInfo[], BindDetailedHandler handler, void (^strongHandler)(const char* symbolName)) const
+bool MachOAnalyzer::forEachBind_OpcodesWeak(Diagnostics& diag, const LinkEditInfo& leInfo, const SegmentInfo segmentsInfo[], BindDetailedHandler handler, void (^strongHandler)(const char* symbolName)) const
{
if ( (leInfo.dyldInfo == nullptr) || (leInfo.dyldInfo->weak_bind_size == 0) )
return false;
@@ -4893,7 +6178,7 @@
return stop;
}
-bool MachOAnalyzer::forEachBind_OpcodesRegular(Diagnostics& diag, const LinkEditInfo& leInfo, const Header::SegmentInfo segmentsInfo[], BindDetailedHandler handler) const
+bool MachOAnalyzer::forEachBind_OpcodesRegular(Diagnostics& diag, const LinkEditInfo& leInfo, const SegmentInfo segmentsInfo[], BindDetailedHandler handler) const
{
if ( (leInfo.dyldInfo == nullptr) || (leInfo.dyldInfo->bind_size == 0) )
return false;
@@ -5009,20 +6294,20 @@
if ( diag.hasError() )
return false;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return false;
- return this->forEachRebase_Opcodes(diag, leInfo, segmentsInfo, ^(const char* opcodeName, const LinkEditInfo& rleInfo, const Header::SegmentInfo segments[],
+ return this->forEachRebase_Opcodes(diag, leInfo, segmentsInfo, ^(const char* opcodeName, const LinkEditInfo& rleInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t pointerSize, uint8_t segmentIndex, uint64_t segmentOffset, Rebase kind, bool& stop) {
- uint64_t rebaseVmOffset = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t rebaseVmOffset = segments[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = rebaseVmOffset - leInfo.layout.textUnslidVMAddr;
handler(runtimeOffset, stop);
});
}
-bool MachOAnalyzer::forEachRebase_Opcodes(Diagnostics& diag, const LinkEditInfo& leInfo, const Header::SegmentInfo segmentsInfo[], RebaseDetailHandler handler) const
+bool MachOAnalyzer::forEachRebase_Opcodes(Diagnostics& diag, const LinkEditInfo& leInfo, const SegmentInfo segmentsInfo[], RebaseDetailHandler handler) const
{
const Rebase pointerRebaseKind = is64() ? Rebase::pointer64 : Rebase::pointer32;
assert(leInfo.dyldInfo != nullptr);
@@ -5087,7 +6372,7 @@
case REBASE_OPCODE_DO_REBASE_ULEB_TIMES:
count = read_uleb128(diag, p, end);
for (uint32_t i=0; i < count; ++i) {
- handler("REBASE_OPCODE_DO_REBASE_ULEB_TIMES", leInfo, segmentsInfo, segIndexSet, ptrSize, segIndex, segOffset, kind, stop);
+ handler("REBASE_OPCODE_DO_REBASE_ADD_ADDR_ULEB", leInfo, segmentsInfo, segIndexSet, ptrSize, segIndex, segOffset, kind, stop);
segOffset += ptrSize;
if ( stop )
break;
@@ -5116,7 +6401,6 @@
return stop;
}
-#if SUPPORT_CLASSIC_RELOCS
bool MachOAnalyzer::forEachRebaseLocation_Relocations(Diagnostics& diag, void (^handler)(uint64_t runtimeOffset, bool& stop)) const
{
LinkEditInfo leInfo;
@@ -5124,14 +6408,14 @@
if ( diag.hasError() )
return false;
- BLOCK_ACCCESSIBLE_ARRAY(Header::SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
+ BLOCK_ACCCESSIBLE_ARRAY(SegmentInfo, segmentsInfo, leInfo.layout.lastSegIndex+1);
getAllSegmentsInfos(diag, segmentsInfo);
if ( diag.hasError() )
return false;
- return this->forEachRebase_Relocations(diag, leInfo, segmentsInfo, ^(const char* opcodeName, const LinkEditInfo& rleInfo, const Header::SegmentInfo segments[],
+ return this->forEachRebase_Relocations(diag, leInfo, segmentsInfo, ^(const char* opcodeName, const LinkEditInfo& rleInfo, const SegmentInfo segments[],
bool segIndexSet, uint32_t pointerSize, uint8_t segmentIndex, uint64_t segmentOffset, Rebase kind, bool& stop) {
- uint64_t rebaseVmOffset = segments[segmentIndex].vmaddr + segmentOffset;
+ uint64_t rebaseVmOffset = segments[segmentIndex].vmAddr + segmentOffset;
uint64_t runtimeOffset = rebaseVmOffset - leInfo.layout.textUnslidVMAddr;
handler(runtimeOffset, stop);
});
@@ -5141,8 +6425,8 @@
void MachOAnalyzer::sortRelocations(Array<relocation_info>& relocs) const
{
// The kernel linker has malloc, and old-style relocations are extremely common. So use qsort
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
- ::qsort(&relocs[0], (size_t)relocs.count(), sizeof(relocation_info),
+#if BUILDING_APP_CACHE_UTIL
+ ::qsort(&relocs[0], relocs.count(), sizeof(relocation_info),
[](const void* l, const void* r) -> int {
if ( ((relocation_info*)l)->r_address < ((relocation_info*)r)->r_address )
return -1;
@@ -5150,10 +6434,10 @@
return 1;
});
#else
- uint64_t count = relocs.count();
- for (uint64_t i=0; i < count-1; ++i) {
+ uintptr_t count = relocs.count();
+ for (uintptr_t i=0; i < count-1; ++i) {
bool done = true;
- for (uint64_t j=0; j < count-i-1; ++j) {
+ for (uintptr_t j=0; j < count-i-1; ++j) {
if ( relocs[j].r_address > relocs[j+1].r_address ) {
relocation_info temp = relocs[j];
relocs[j] = relocs[j+1];
@@ -5168,7 +6452,7 @@
}
-bool MachOAnalyzer::forEachRebase_Relocations(Diagnostics& diag, const LinkEditInfo& leInfo, const Header::SegmentInfo segmentsInfo[], RebaseDetailHandler handler) const
+bool MachOAnalyzer::forEachRebase_Relocations(Diagnostics& diag, const LinkEditInfo& leInfo, const SegmentInfo segmentsInfo[], RebaseDetailHandler handler) const
{
// old binary, walk relocations
const uint64_t relocsStartAddress = localRelocBaseAddress(segmentsInfo, leInfo.layout.linkeditSegIndex);
@@ -5181,7 +6465,7 @@
for (const relocation_info* reloc=relocsStart; (reloc < relocsEnd) && !stop; ++reloc) {
if ( reloc->r_length != relocSize ) {
bool shouldEmitError = true;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
if ( usesClassicRelocationsInKernelCollection() && (reloc->r_length == 2) && (relocSize == 3) )
shouldEmitError = false;
#endif
@@ -5203,7 +6487,7 @@
uint32_t segIndex = 0;
uint64_t segOffset = 0;
uint64_t addr = 0;
-#if BUILDING_APP_CACHE_UTIL || BUILDING_DYLDINFO
+#if BUILDING_APP_CACHE_UTIL
// xnu for x86_64 has __HIB mapped before __DATA, so offsets appear to be
// negative
if ( isStaticExecutable() || isFileSet() ) {
@@ -5248,243 +6532,8 @@
return stop;
}
-#endif // SUPPORT_CLASSIC_RELOCS
-
-bool MachOAnalyzer::getLinkeditLayout(Diagnostics& diag, uint64_t linkeditFileOffset,
- const uint8_t* linkeditStartAddr, mach_o::LinkeditLayout& layout) const
-{
- // Note, in VM layout all linkedit offsets are adjusted from file offsets.
- // It is essential no-one calls this on an object in file layout. It must be in VM layout
-
- auto getLinkEditContent = [&](uint32_t fileOffset)
- {
- uint64_t offsetInLinkedit = fileOffset - linkeditFileOffset;
- return linkeditStartAddr + offsetInLinkedit;
- };
-
- // FIXME: Other load commands
- this->forEachLoadCommand(diag, ^(const load_command *cmd, bool &stop) {
- switch ( cmd->cmd ) {
- case LC_SYMTAB: {
- const symtab_command* symTabCmd = (const symtab_command*)cmd;
-
- // Record that we found a LC_SYMTAB
- layout.hasSymTab = true;
-
- // NList
- uint64_t nlistEntrySize = this->is64() ? sizeof(struct nlist_64) : sizeof(struct nlist);
- layout.symbolTable.fileOffset = symTabCmd->symoff;
- layout.symbolTable.buffer = getLinkEditContent(symTabCmd->symoff);
- layout.symbolTable.bufferSize = (uint32_t)(symTabCmd->nsyms * nlistEntrySize);
- layout.symbolTable.entryCount = symTabCmd->nsyms;
- layout.symbolTable.hasLinkedit = true;
-
- // Symbol strings
- layout.symbolStrings.fileOffset = symTabCmd->stroff;
- layout.symbolStrings.buffer = getLinkEditContent(symTabCmd->stroff);
- layout.symbolStrings.bufferSize = symTabCmd->strsize;
- layout.symbolStrings.hasLinkedit = true;
- break;
- }
- case LC_DYSYMTAB: {
- const dysymtab_command* dynSymTabCmd = (const dysymtab_command*)cmd;
-
- // Record that we found a LC_DYSYMTAB
- layout.hasDynSymTab = true;
-
- // Local relocs
- layout.localRelocs.fileOffset = dynSymTabCmd->locreloff;
- layout.localRelocs.buffer = getLinkEditContent(dynSymTabCmd->locreloff);
- layout.localRelocs.bufferSize = 0; // Use entryCount instead
- layout.localRelocs.entryIndex = 0; // Use buffer instead
- layout.localRelocs.entryCount = dynSymTabCmd->nlocrel;
- layout.localRelocs.hasLinkedit = true;
-
- // Extern relocs
- layout.externRelocs.fileOffset = dynSymTabCmd->extreloff;
- layout.externRelocs.buffer = getLinkEditContent(dynSymTabCmd->extreloff);
- layout.externRelocs.bufferSize = 0; // Use entryCount instead
- layout.externRelocs.entryIndex = 0; // Use buffer instead
- layout.externRelocs.entryCount = dynSymTabCmd->nextrel;
- layout.externRelocs.hasLinkedit = true;
-
- // Indirect symbol table
- layout.indirectSymbolTable.fileOffset = dynSymTabCmd->indirectsymoff;
- layout.indirectSymbolTable.buffer = getLinkEditContent(dynSymTabCmd->indirectsymoff);
- layout.indirectSymbolTable.bufferSize = 0; // Use entryCount instead
- layout.indirectSymbolTable.entryIndex = 0; // Use buffer instead
- layout.indirectSymbolTable.entryCount = dynSymTabCmd->nindirectsyms;
- layout.indirectSymbolTable.hasLinkedit = true;
-
- // Locals
- layout.localSymbolTable.fileOffset = 0; // unused
- layout.localSymbolTable.buffer = nullptr; // Use entryIndex instead
- layout.localSymbolTable.bufferSize = 0; // Use entryCount instead
- layout.localSymbolTable.entryIndex = dynSymTabCmd->ilocalsym;
- layout.localSymbolTable.entryCount = dynSymTabCmd->nlocalsym;
- layout.localSymbolTable.hasLinkedit = true;
-
- // Globals
- layout.globalSymbolTable.fileOffset = 0; // unused
- layout.globalSymbolTable.buffer = nullptr; // Use entryIndex instead
- layout.globalSymbolTable.bufferSize = 0; // Use entryCount instead
- layout.globalSymbolTable.entryIndex = dynSymTabCmd->iextdefsym;
- layout.globalSymbolTable.entryCount = dynSymTabCmd->nextdefsym;
- layout.globalSymbolTable.hasLinkedit = true;
-
- // Imports
- layout.undefSymbolTable.fileOffset = 0; // unused
- layout.undefSymbolTable.buffer = nullptr; // Use entryIndex instead
- layout.undefSymbolTable.bufferSize = 0; // Use entryCount instead
- layout.undefSymbolTable.entryIndex = dynSymTabCmd->iundefsym;
- layout.undefSymbolTable.entryCount = dynSymTabCmd->nundefsym;
- layout.undefSymbolTable.hasLinkedit = true;
- break;
- }
- case LC_DYLD_INFO:
- case LC_DYLD_INFO_ONLY: {
- const dyld_info_command* linkeditCmd = (const dyld_info_command*)cmd;
-
- // Record what kind of DYLD_INFO we found
- layout.dyldInfoCmd = cmd->cmd;
-
- // Rebase
- layout.rebaseOpcodes.fileOffset = linkeditCmd->rebase_off;
- layout.rebaseOpcodes.buffer = getLinkEditContent(linkeditCmd->rebase_off);
- layout.rebaseOpcodes.bufferSize = linkeditCmd->rebase_size;
- layout.rebaseOpcodes.hasLinkedit = true;
-
- // Bind
- layout.regularBindOpcodes.fileOffset = linkeditCmd->bind_off;
- layout.regularBindOpcodes.buffer = getLinkEditContent(linkeditCmd->bind_off);
- layout.regularBindOpcodes.bufferSize = linkeditCmd->bind_size;
- layout.regularBindOpcodes.hasLinkedit = true;
-
- // Lazy bind
- layout.lazyBindOpcodes.fileOffset = linkeditCmd->lazy_bind_off;
- layout.lazyBindOpcodes.buffer = getLinkEditContent(linkeditCmd->lazy_bind_off);
- layout.lazyBindOpcodes.bufferSize = linkeditCmd->lazy_bind_size;
- layout.lazyBindOpcodes.hasLinkedit = true;
-
- // Weak bind
- layout.weakBindOpcodes.fileOffset = linkeditCmd->weak_bind_off;
- layout.weakBindOpcodes.buffer = getLinkEditContent(linkeditCmd->weak_bind_off);
- layout.weakBindOpcodes.bufferSize = linkeditCmd->weak_bind_size;
- layout.weakBindOpcodes.hasLinkedit = true;
-
- // Export trie
- layout.exportsTrie.fileOffset = linkeditCmd->export_off;
- layout.exportsTrie.buffer = getLinkEditContent(linkeditCmd->export_off);
- layout.exportsTrie.bufferSize = linkeditCmd->export_size;
- layout.exportsTrie.hasLinkedit = true;
- break;
- }
- case LC_DYLD_CHAINED_FIXUPS: {
- const linkedit_data_command* linkeditCmd = (const linkedit_data_command*)cmd;
-
- layout.chainedFixups.fileOffset = linkeditCmd->dataoff;
- layout.chainedFixups.buffer = getLinkEditContent(linkeditCmd->dataoff);
- layout.chainedFixups.bufferSize = linkeditCmd->datasize;
- layout.chainedFixups.entryCount = 0; // Not needed here
- layout.chainedFixups.hasLinkedit = true;
- layout.chainedFixups.cmd = linkeditCmd;
- break;
- }
- case LC_DYLD_EXPORTS_TRIE: {
- const linkedit_data_command* linkeditCmd = (const linkedit_data_command*)cmd;
-
- layout.exportsTrie.fileOffset = linkeditCmd->dataoff;
- layout.exportsTrie.buffer = getLinkEditContent(linkeditCmd->dataoff);
- layout.exportsTrie.bufferSize = linkeditCmd->datasize;
- layout.exportsTrie.entryCount = 0; // Not needed here
- layout.exportsTrie.hasLinkedit = true;
- break;
- }
- case LC_SEGMENT_SPLIT_INFO: {
- const linkedit_data_command* linkeditCmd = (const linkedit_data_command*)cmd;
-
- layout.splitSegInfo.fileOffset = linkeditCmd->dataoff;
- layout.splitSegInfo.buffer = getLinkEditContent(linkeditCmd->dataoff);
- layout.splitSegInfo.bufferSize = linkeditCmd->datasize;
- layout.splitSegInfo.entryCount = 0; // Not needed here
- layout.splitSegInfo.hasLinkedit = true;
- break;
- }
- case LC_FUNCTION_STARTS: {
- const linkedit_data_command* linkeditCmd = (const linkedit_data_command*)cmd;
-
- layout.functionStarts.fileOffset = linkeditCmd->dataoff;
- layout.functionStarts.buffer = getLinkEditContent(linkeditCmd->dataoff);
- layout.functionStarts.bufferSize = linkeditCmd->datasize;
- layout.functionStarts.entryCount = 0; // Not needed here
- layout.functionStarts.hasLinkedit = true;
- break;
- }
- case LC_DATA_IN_CODE: {
- const linkedit_data_command* linkeditCmd = (const linkedit_data_command*)cmd;
-
- layout.dataInCode.fileOffset = linkeditCmd->dataoff;
- layout.dataInCode.buffer = getLinkEditContent(linkeditCmd->dataoff);
- layout.dataInCode.bufferSize = linkeditCmd->datasize;
- layout.dataInCode.entryCount = 0; // Not needed here
- layout.dataInCode.hasLinkedit = true;
- break;
- }
- case LC_CODE_SIGNATURE: {
- const linkedit_data_command* linkeditCmd = (const linkedit_data_command*)cmd;
-
- layout.codeSignature.fileOffset = linkeditCmd->dataoff;
- layout.codeSignature.buffer = getLinkEditContent(linkeditCmd->dataoff);
- layout.codeSignature.bufferSize = linkeditCmd->datasize;
- layout.codeSignature.entryCount = 0; // Not needed here
- layout.codeSignature.hasLinkedit = true;
- break;
- }
- }
- });
-
- return true;
-}
-
-void MachOAnalyzer::withVMLayout(Diagnostics &diag, void (^callback)(const mach_o::Layout &layout)) const
-{
- intptr_t slide = this->getSlide();
- __block uint64_t linkeditFileOffset = 0;
- __block const uint8_t* linkeditStartAddr = nullptr;
-
- const Header* hdr = ((const Header*)this);
- uint32_t numSegments = hdr->segmentCount();
- BLOCK_ACCCESSIBLE_ARRAY(mach_o::SegmentLayout, segmentLayout, numSegments);
- hdr->forEachSegment(^(const Header::SegmentInfo &info, bool &stop) {
- mach_o::SegmentLayout segment;
- segment.vmAddr = info.vmaddr;
- segment.vmSize = info.vmsize;
- segment.fileOffset = info.fileOffset;
- segment.fileSize = info.fileSize;
- segment.buffer = (uint8_t*)(info.vmaddr + slide);
- segment.protections = info.initProt;
-
- segment.kind = mach_o::SegmentLayout::Kind::unknown;
- if ( info.segmentName == "__TEXT" ) {
- segment.kind = mach_o::SegmentLayout::Kind::text;
- } else if ( info.segmentName == "__LINKEDIT" ) {
- segment.kind = mach_o::SegmentLayout::Kind::linkedit;
- linkeditFileOffset = info.fileOffset;
- linkeditStartAddr = segment.buffer;
- }
-
- segmentLayout[info.segmentIndex] = segment;
- });
-
- mach_o::LinkeditLayout linkedit;
- if ( !this->getLinkeditLayout(diag, linkeditFileOffset, linkeditStartAddr, linkedit) ) {
- diag.error("Couldn't get dylib layout");
- return;
- }
-
- mach_o::Layout layout(this, { &segmentLayout[0], &segmentLayout[numSegments] }, linkedit);
- callback(layout);
-}
+
+
} // dyld3