Loading...
libdyld_introspection/ProcessAtlas.cpp /dev/null dyld-941.5
--- /dev/null
+++ dyld/dyld-941.5/libdyld_introspection/ProcessAtlas.cpp
@@ -0,0 +1,1128 @@
+/* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-
+ *
+ * Copyright (c) 2016 Apple Inc. All rights reserved.
+ *
+ * @APPLE_LICENSE_HEADER_START@
+ *
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this
+ * file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
+ *
+ * @APPLE_LICENSE_HEADER_END@
+ */
+
+#include <TargetConditionals.h>
+
+#include <Block.h>
+#include <ctype.h>
+#include <dlfcn.h>
+#include <fcntl.h>
+#include <dirent.h>
+#include <unistd.h>
+#include <libproc.h>
+
+#include <sys/attr.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <sys/fsgetpath.h>
+
+#include <mach/mach_vm.h>
+#include <mach-o/dyld_priv.h> // FIXME: We can remove this once we fully integrate into dyld4
+#include "dyld_cache_format.h"
+//FIXME: We should remove this header
+#include "dyld_process_info_internal.h" // For dyld_all_image_infos_{32,64}
+
+#include "Defines.h"
+#include "MachOFile.h"
+#include "ProcessAtlas.h"
+
+#include "ProcessAtlas.h"
+
+using namespace dyld4;
+// TODO: forEach shared cache needs to filter out subcaches and skip them
+
+namespace {
+static Allocator* libDylAllocator() {
+    static Allocator* allocator = nullptr;
+    if (!allocator) {
+        allocator = Allocator::bootstrap();
+    }
+    return allocator;
+}
+
+static const size_t kCachePeekSize = 0x4000;
+
+static const dyld_cache_header* cacheFilePeek(int fd, uint8_t* firstPage) {
+    // sanity check header
+    if ( pread(fd, firstPage, kCachePeekSize, 0) != kCachePeekSize ) {
+        return nullptr;
+    }
+    const dyld_cache_header* cache = (dyld_cache_header*)firstPage;
+    if ( strncmp(cache->magic, "dyld_v1", strlen("dyld_v1")) != 0 ) {
+        return nullptr;
+    }
+    return cache;
+}
+
+static void getCacheInfo(const dyld_cache_header *cache, uint64_t &headerSize, bool& splitCache) {
+    // If we have sub caches, then the cache header itself tells us how much space we need to cover all caches
+    if ( cache->mappingOffset >= __offsetof(dyld_cache_header, subCacheArrayCount) ) {
+        // New style cache
+        headerSize = cache->subCacheArrayOffset + (sizeof(dyld_subcache_entry)*cache->subCacheArrayCount);
+        splitCache = true;
+    } else {
+        // Old style cache
+        headerSize = cache->imagesOffsetOld + (sizeof(dyld_cache_image_info)*cache->imagesCountOld);
+        splitCache = false;
+    }
+}
+}
+
+namespace dyld4 {
+namespace Atlas {
+
+#pragma mark -
+#pragma mark Mappers
+
+
+static
+void printMapping(dyld_cache_mapping_and_slide_info* mapping, uint8_t index, uint64_t slide) {
+#if 0
+    const char* mappingName = "*unknown*";
+    if ( mapping->maxProt & VM_PROT_EXECUTE ) {
+        mappingName = "__TEXT";
+    } else if ( mapping->maxProt & VM_PROT_WRITE ) {
+        if ( mapping->flags & DYLD_CACHE_MAPPING_AUTH_DATA ) {
+            if ( mapping->flags & DYLD_CACHE_MAPPING_DIRTY_DATA )
+                mappingName = "__AUTH_DIRTY";
+            else if ( mapping->flags & DYLD_CACHE_MAPPING_CONST_DATA )
+                mappingName = "__AUTH_CONST";
+            else
+                mappingName = "__AUTH";
+        } else {
+            if ( mapping->flags & DYLD_CACHE_MAPPING_DIRTY_DATA )
+                mappingName = "__DATA_DIRTY";
+            else if ( mapping->flags & DYLD_CACHE_MAPPING_CONST_DATA )
+                mappingName = "__DATA_CONST";
+            else
+                mappingName = "__DATA";
+        }
+    }
+    else if ( mapping->maxProt & VM_PROT_READ ) {
+        mappingName = "__LINKEDIT";
+    }
+
+    fprintf(stderr, "%16s %4lluMB,  file offset: #%u/0x%08llX -> 0x%08llX,  address: 0x%08llX -> 0x%08llX\n",
+            mappingName, mapping->size / (1024*1024), index, mapping->fileOffset,
+            mapping->fileOffset + mapping->size, mapping->address + slide,
+            mapping->address + mapping->size + slide);
+#endif
+}
+
+SharedPtr<Mapper> Mapper::mapperForSharedCache(const char* cachePath, const DRL::UUID& uuid, const void* baseAddress) {
+    bool        useLocalCache   = false;
+    size_t      length          = 0;
+    uint64_t    slide     = 0;
+    int fd = open(cachePath, O_RDONLY);
+    if ( fd == -1 ) {
+        return nullptr;
+    }
+    const void* localBaseAddress = _dyld_get_shared_cache_range(&length);
+    if (localBaseAddress) {
+        auto localCacheHeader = ((dyld_cache_header*)localBaseAddress);
+        auto localUUID = DRL::UUID(&localCacheHeader->uuid[0]);
+        if (localUUID == uuid) {
+            useLocalCache = true;
+        }
+    }
+    uint8_t firstPage[kCachePeekSize];
+    const dyld_cache_header* onDiskCacheHeader = cacheFilePeek(fd, &firstPage[0]);
+    if (!onDiskCacheHeader) {
+        close(fd);
+        return nullptr;
+    }
+    if (baseAddress) {
+        slide = (uint64_t)baseAddress-(uint64_t)onDiskCacheHeader->sharedRegionStart;
+    }
+    uint64_t headerSize = 0;
+    bool splitCache = false;
+    getCacheInfo(onDiskCacheHeader, headerSize, splitCache);
+    if (splitCache && (onDiskCacheHeader->imagesCount == 0)) {
+        //This is a subcache, bail
+        close(fd);
+        return nullptr;
+    }
+    void* mapping = mmap(nullptr, (size_t)headerSize, PROT_READ, MAP_FILE | MAP_PRIVATE, fd, 0);
+    if (mapping == MAP_FAILED) {
+        close(fd);
+        return nullptr;
+    }
+    auto onDiskHeaderBytes = (uint8_t*)mapping;
+    auto onDiskCacheMappings = (dyld_cache_mapping_and_slide_info*)&onDiskHeaderBytes[onDiskCacheHeader->mappingWithSlideOffset];
+    Vector<Mapper::Mapping> mappings(libDylAllocator());
+    for (auto i = 0; i < onDiskCacheHeader->mappingWithSlideCount; ++i) {
+        if (useLocalCache && ((onDiskCacheMappings[i].maxProt & VM_PROT_WRITE) != VM_PROT_WRITE)) {
+            // This region is immutable, use in memory version
+            printMapping(&onDiskCacheMappings[i], 255, slide);
+            mappings.emplace_back((Mapper::Mapping){
+                .address    = onDiskCacheMappings[i].address + slide,
+                .size       = onDiskCacheMappings[i].size,
+                .offset     = onDiskCacheMappings[i].fileOffset,
+                .fd         = -1
+            });
+        } else {
+            printMapping(&onDiskCacheMappings[i], 0, slide);
+            mappings.emplace_back((Mapper::Mapping){
+                .address    = onDiskCacheMappings[i].address + slide,
+                .size       = onDiskCacheMappings[i].size,
+                .offset     = onDiskCacheMappings[i].fileOffset,
+                .fd         = fd
+            });
+        }
+    }
+    if (splitCache) {
+        auto subCaches = (dyld_subcache_entry*)&onDiskHeaderBytes[onDiskCacheHeader->subCacheArrayOffset];
+        for (auto i = 0; i < onDiskCacheHeader->subCacheArrayCount; ++i) {
+            char subCachePath[PATH_MAX];
+            snprintf(&subCachePath[0], PATH_MAX, "%s.%u", cachePath, i+1);
+            fd = open(subCachePath, O_RDONLY);
+            if ( fd == -1 ) {
+                break;
+            }
+            // TODO: We should check we have enough space, but for now just allocate a page
+            uint8_t firstSubPage[kCachePeekSize];
+            const dyld_cache_header* subCache = cacheFilePeek(fd, &firstSubPage[0]);
+            if (!subCache) {
+                close(fd);
+                continue;
+            }
+            auto subCacheheaderBytes = (uint8_t*)subCache;
+            auto subCacheMappings = (dyld_cache_mapping_and_slide_info*)&subCacheheaderBytes[subCache->mappingWithSlideOffset];
+
+            auto onDiskSubcacheUUID = DRL::UUID(subCache->uuid);
+            auto subcacheUUID = DRL::UUID(subCaches[i].uuid);
+            if (subcacheUUID != onDiskSubcacheUUID) {
+                //TODO: Replace this with a set
+                Vector<int> fds(libDylAllocator());
+                for (auto& deadMapping : mappings) {
+                    if (deadMapping.fd == -1) { continue; }
+                    if (std::find(fds.begin(), fds.end(), deadMapping.fd) == fds.end()) {
+                        fds.push_back(deadMapping.fd);
+                    }
+                }
+                for (auto& deadFd : fds) {
+                    close(deadFd);
+                }
+                return nullptr;
+            }
+
+            for (auto j = 0; j < subCache->mappingWithSlideCount; ++j) {
+                if (useLocalCache && ((onDiskCacheMappings[j].maxProt & VM_PROT_WRITE) != VM_PROT_WRITE)) {
+                    // This region is immutable, use in memory version
+                    printMapping(&subCacheMappings[j], 255, slide);;
+                    mappings.emplace_back((Mapper::Mapping){
+                        .address    = subCacheMappings[j].address + slide,
+                        .size       = subCacheMappings[j].size,
+                        .offset     = subCacheMappings[j].fileOffset,
+                        .fd         = -1
+                    });
+                } else {
+                    printMapping(&subCacheMappings[j], j+1, slide);;
+                    mappings.emplace_back((Mapper::Mapping){
+                        .address    = subCacheMappings[j].address + slide,
+                        .size       = subCacheMappings[j].size,
+                        .offset     = subCacheMappings[j].fileOffset,
+                        .fd         = fd
+                    });
+                }
+            }
+        }
+    }
+    munmap(mapping,(size_t)headerSize);
+    return SharedPtr<Mapper>(new (libDylAllocator()) Mapper(mappings));
+}
+
+
+std::pair<SharedPtr<Mapper>,uint64_t> Mapper::mapperForSharedCacheLocals(const char* filePath) {
+    struct stat statbuf;
+    if ( ::stat(filePath, &statbuf) != 0 ) {
+        return { SharedPtr<Mapper>(), 0};
+    }
+
+    int fd = open(filePath, O_RDONLY);
+    if ( fd == -1 ) {
+        return { SharedPtr<Mapper>(), 0};
+    }
+
+    // sanity check header
+    uint8_t firstPage[kCachePeekSize];
+    const dyld_cache_header* cache = cacheFilePeek(fd, &firstPage[0]);
+    if (!cache) {
+        close(fd);
+        return { SharedPtr<Mapper>(), 0};
+    }
+    uint64_t baseAddress = 0;
+
+    // We want the cache header, which is at the start of the, and the locals, which are at the end.
+    // Just map the whole file as a single range, as we need file offsets in the mappings anyway
+    // With split caches, this is more reasonable as the locals are in their own file, so we want more or
+    // less the whole file anyway, and there's no wasted space for __TEXT, __DATA, etc.
+//    fprintf(stderr, "Mapping\n");
+//    fprintf(stderr, "fd\tAddress\tFile Offset\tSize\n");
+//    fprintf(stderr, "%u\t0x%llx\t0x%x\t%llu\n", fd, baseAddress, 0, (uint64_t)statbuf.st_size);
+    Vector<Mapper::Mapping> mappings(libDylAllocator());
+    mappings.emplace_back((Mapper::Mapping){
+        .address = baseAddress,
+        .size = (uint64_t)statbuf.st_size,
+        .offset = 0,
+        .fd = fd
+    });
+    return  {SharedPtr<Mapper>(new (libDylAllocator()) Mapper(mappings)), baseAddress};
+}
+
+Mapper::Mapper() : _mappings({{ .address = 0, .size = std::numeric_limits<uint64_t>::max(), .offset = 0, .fd = -1 }}, libDylAllocator()), _flatMapping(nullptr) {}
+Mapper::Mapper(const Vector<Mapping>& M) : _mappings(M), _flatMapping(nullptr) {}
+
+Mapper::~Mapper() {
+    assert(_flatMapping == nullptr);
+    //TODO: Replace this with a set
+    Vector<int> fds(libDylAllocator());
+    for (auto& mapping : _mappings) {
+        if (mapping.fd == -1) { continue; }
+        if (std::find(fds.begin(), fds.end(), mapping.fd) == fds.end()) {
+            fds.push_back(mapping.fd);
+        }
+    }
+    for (auto& fd : fds) {
+        close(fd);
+    }
+}
+
+std::pair<void*,bool> Mapper::map(const void* addr, uint64_t size) const {
+    if (_flatMapping) {
+        uint64_t offset = (uint64_t)addr-(uint64_t)baseAddress();
+        return {(void*)((uintptr_t)_flatMapping+offset),false};
+    }
+    for (const auto& mapping : _mappings) {
+        if (((uint64_t)addr >= mapping.address) && ((uint64_t)addr < (mapping.address + mapping.size))) {
+            if (mapping.fd == -1) {
+                return {(void*)((uint64_t)addr+mapping.offset), false};
+            }
+            assert(((uint64_t)addr + size) <= mapping.address + mapping.size);
+            off_t offset = (off_t)addr - mapping.address + mapping.offset;
+            // Handle unaligned mmap
+            void* newMapping = nullptr;
+            size_t extraBytes = 0;
+            off_t roundedOffset = offset & (-1*PAGE_SIZE);
+            extraBytes = (size_t)offset - (size_t)roundedOffset;
+            newMapping = mmap(nullptr, (size_t)size+extraBytes, PROT_READ, MAP_FILE | MAP_PRIVATE, mapping.fd, roundedOffset);
+            if (newMapping == (void*)-1) {
+                printf("mmap failed: %s (%d)\n", strerror(errno), errno);
+                return {(void*)1, false};
+            }
+            return {(void*)((uintptr_t)newMapping+extraBytes),true};
+        }
+    }
+    return {(void*)-1, false};
+}
+
+void Mapper::unmap(const void* addr, uint64_t size) const {
+    void* roundedAddr = (void*)((intptr_t)addr & (-1*PAGE_SIZE));
+    size_t extraBytes = (uintptr_t)addr - (uintptr_t)roundedAddr;
+    munmap(roundedAddr, (size_t)size+extraBytes);
+}
+
+const void* Mapper::baseAddress() const {
+    return (const void*)_mappings[0].address;
+}
+
+const uint64_t Mapper::size() const {
+    return (_mappings.back().address - _mappings[0].address) + _mappings.back().size;
+}
+
+bool Mapper::pin() {
+    assert(_flatMapping == nullptr);
+    //TODO: Move onto dyld allocators once we merge the large allocations support
+    if (vm_allocate(mach_task_self(), (vm_address_t*)&_flatMapping, (vm_size_t)size(), VM_FLAGS_ANYWHERE) != KERN_SUCCESS) {
+        return false;
+    }
+    for (const auto& mapping : _mappings) {
+        uint64_t destAddr = (mapping.address - _mappings[0].address) + (uint64_t)_flatMapping;
+        if (mapping.fd == -1) {
+            if (vm_copy(mach_task_self(), (vm_address_t)mapping.address, (vm_size_t)mapping.size, (vm_address_t)destAddr) != KERN_SUCCESS) {
+                unpin();
+                return false;
+            }
+        } else {
+            if (mmap((void*)destAddr, (vm_size_t)mapping.size, PROT_READ, MAP_FILE | MAP_PRIVATE | MAP_FIXED, mapping.fd, mapping.offset) == MAP_FAILED) {
+                unpin();
+                return false;
+            }
+        }
+    }
+    return true;
+}
+void Mapper::unpin() {
+    assert(_flatMapping != nullptr);
+    vm_deallocate(mach_task_self(), (vm_address_t)_flatMapping, (vm_size_t)size());
+    _flatMapping = nullptr;
+}
+
+
+#pragma mark -
+#pragma mark Image
+
+Image::Image(SharedPtr<Mapper>& M, void* A, uint64_t S, const SharedCache* SC) : _slide(S), _address(A), _mapper(M), _sharedCache(SC) {}
+
+const MachOLoaded* Image::ml() {
+    void* slidML = (void*)((uintptr_t)_address+_slide);
+    if (!_ml) {
+        // Note, using 4k here as we might be an arm64e process inspecting an x86_64 image, which uses 4k pages
+        _ml = _mapper->map<MachOLoaded>(slidML, 4096);
+        size_t size = _ml->sizeofcmds;
+        if ( _ml->magic == MH_MAGIC_64 ) {
+            size += sizeof(mach_header_64);
+        } else {
+            size += sizeof(mach_header);
+        }
+        if (size > 4096) {
+            _ml = _mapper->map<MachOLoaded>(slidML, size);
+        }
+    }
+    // This is a bit of a mess. With compact info this will be unified, but for now we use a lot of hacky abstactions here to deal with
+    // in process / vs out of process / vs shared cache.
+    return &*_ml;
+}
+
+//const MachOLoaded*              ml();
+//DRL::UUID                       _uuid;
+//Mapper::Pointer<MachOLoaded>    _ml;
+//const uint64_t                  _slide              = 0;
+//const void*                     _address            = nullptr;
+//SharedPtr<Mapper>               _mapper;
+//const SharedCache*              _sharedCache        = nullptr;
+//const char*                     _installname        = nullptr;
+//const char*                     _filename           = nullptr;
+//bool                            _uuidLoaded         = false;
+//bool                            _installnameLoaded  = false;
+//bool                            _filenameLoaded     = false;
+
+const DRL::UUID& Image::uuid() {
+    if (!_uuidLoaded) {
+        uuid_t fileUUID;
+        if (ml()->getUuid(fileUUID)) {
+            _uuid = DRL::UUID(fileUUID);
+        }
+        _uuidLoaded = true;
+    }
+    return _uuid;
+}
+
+const char* Image::installname() {
+    if (!_installnameLoaded) {
+        _installname = ml()->installName();
+        _installnameLoaded = true;
+    }
+    return _installname;
+}
+const char* Image::filename() {
+    if (!_filenameLoaded) {
+        //TODO: The filename can be derived via the fsid objects in all image info
+        _filenameLoaded = true;
+    }
+    return _filename;
+}
+
+const SharedCache* Image::sharedCache() const {
+    return _sharedCache;
+}
+
+uint64_t Image::sharedCacheVMOffset() const {
+    return (uint64_t)_address - sharedCache()->baseAddress();
+}
+
+uint32_t Image::pointerSize() {
+    return ml()->pointerSize();
+}
+
+
+bool Image::forEachSegment(void (^block)(const char* segmentName, uint64_t vmAddr, uint64_t vmSize, int perm)) {
+    ml()->forEachSegment(^(const MachOLoaded::SegmentInfo &info, bool &stop) {
+        block(info.segName, info.vmAddr + _slide, info.vmSize, info.protections);
+    });
+    return true;
+}
+
+bool Image::forEachSection(void (^block)(const char* segmentName, const char* sectionName, uint64_t vmAddr, uint64_t vmSize)) {
+    ml()->forEachSection(^(const MachOLoaded::SectionInfo &info, bool malformedSectionRange, bool &stop) {
+        block(info.segInfo.segName, info.sectName, info.sectAddr + _slide, info.sectSize);
+    });
+    return true;
+}
+
+bool Image::contentForSegment(const char* segmentName, void (^contentReader)(const void* content, uint64_t vmAddr, uint64_t vmSize)) {
+    __block bool result = false;
+    ml()->forEachSegment(^(const MachOLoaded::SegmentInfo &info, bool &stop) {
+        if (strcmp(segmentName, info.segName) != 0) { return; }
+        if (info.vmSize) {
+            auto content = _mapper->map<uint8_t>((void*)(info.vmAddr+_slide), info.vmSize);
+            contentReader((void*)&*content, info.vmAddr + _slide, info.vmSize);
+        } else {
+            contentReader(nullptr, info.vmAddr + _slide, 0);
+        }
+        result = true;
+        stop = true;
+    });
+    return result;
+}
+
+bool Image::contentForSection(const char* segmentName, const char* sectionName,
+                              void (^contentReader)(const void* content, uint64_t vmAddr, uint64_t vmSize)) {
+    __block bool result = false;
+    ml()->forEachSection(^(const MachOLoaded::SectionInfo &info, bool malformedRange, bool &stop) {
+        if (strcmp(segmentName, info.segInfo.segName) != 0) { return; }
+        if (strcmp(sectionName, info.sectName) != 0) { return; }
+        if (info.sectSize) {
+            auto content = _mapper->map<uint8_t>((void*)(info.sectAddr+_slide), info.sectSize);
+            contentReader((void*)&*content, info.sectAddr + _slide, info.sectSize);
+        } else {
+            contentReader(nullptr, info.sectAddr + _slide, 0);
+        }
+        result = true;
+        stop = true;
+    });
+    return result;
+}
+
+#pragma mark -
+#pragma mark Shared Cache Locals
+
+SharedCacheLocals::SharedCacheLocals(SharedPtr<Mapper>& M, bool use64BitDylibOffsets)
+    : _mapper(M), _use64BitDylibOffsets(use64BitDylibOffsets) {
+    auto header = _mapper->map<dyld_cache_header>((void*)0, sizeof(dyld_cache_header));
+
+    // Map in the whole locals buffer.
+    // TODO: Once we have the symbols in their own file, simplify this to just map the whole file
+    // and not do the header and locals separately
+    _locals = _mapper->map<uint8_t>((void*)header->localSymbolsOffset, header->localSymbolsSize);
+}
+
+const dyld_cache_local_symbols_info* SharedCacheLocals::localInfo() const {
+    return (const dyld_cache_local_symbols_info*)(&*_locals);
+}
+
+bool SharedCacheLocals::use64BitDylibOffsets() const {
+    return _use64BitDylibOffsets;
+}
+
+#pragma mark -
+#pragma mark Shared Cache
+
+SharedCache::SharedCache(SharedPtr<Mapper>& M, const char* FP, bool P) :
+    _files(libDylAllocator()), _private(P), _images(libDylAllocator()), _mapper(M)
+{
+    assert(_mapper);
+    auto baseAddress = _mapper->baseAddress();
+    _files.emplace_back(libDylAllocator()->strdup(FP));
+    _header = _mapper->map<dyld_cache_header>(baseAddress, PAGE_SIZE);
+    uint64_t headerSize = 0;
+    bool splitCache = false;
+    getCacheInfo(&*_header, headerSize, splitCache);
+    if (headerSize > PAGE_SIZE) {
+        _header = _mapper->map<dyld_cache_header>(baseAddress, headerSize);
+    }
+    _uuid = DRL::UUID(&_header->uuid[0]);
+    _slide = (uint64_t)baseAddress -  _header->sharedRegionStart;
+    auto headerBytes = (uint8_t*)&*_header;
+    auto mappings = (dyld_cache_mapping_and_slide_info*)&headerBytes[_header->mappingWithSlideOffset];
+    uint64_t endAddress = 0;
+    for (auto i = 0; i < _header->mappingWithSlideCount; ++i) {
+        if (endAddress < mappings[i].address + mappings[i].size) {
+            endAddress = mappings[i].address + mappings[i].size;
+        }
+    }
+    auto images = (dyld_cache_image_info*)&headerBytes[_header->imagesOffsetOld];
+    uint32_t imagesCount = _header->imagesCountOld;
+    if ( _header->mappingOffset >= __offsetof(dyld_cache_header, imagesCount) ) {
+        images = (dyld_cache_image_info*)&headerBytes[_header->imagesOffset];
+        imagesCount = _header->imagesCount;
+    }
+    for (auto i = 0; i < imagesCount; ++i) {
+        _images.emplace_back(libDylAllocator()->makeUnique<Image>(_mapper, (void*)images[i].address, _slide, this));
+    }
+
+    if (splitCache) {
+        char cachePath[PATH_MAX];
+        auto subCaches = (dyld_subcache_entry*)&headerBytes[_header->subCacheArrayOffset];
+        for (auto i = 0; i < _header->subCacheArrayCount; ++i) {
+            auto subCacheHeader = _mapper->map<dyld_cache_header>((void*)(subCaches[i].cacheVMOffset + (uint64_t)baseAddress), PAGE_SIZE);
+            uint64_t subCacheHeaderSize = 0;
+            bool splitCacheUnused;
+            getCacheInfo(&*subCacheHeader, subCacheHeaderSize, splitCacheUnused);
+            if (subCacheHeaderSize > PAGE_SIZE) {
+                subCacheHeader = _mapper->map<dyld_cache_header>((void*)(subCaches[i].cacheVMOffset + (uint64_t)baseAddress), subCacheHeaderSize);
+            }
+            auto subCacheHeaderBytes = (uint8_t*)&*subCacheHeader;
+            auto subCacheMappings = (dyld_cache_mapping_and_slide_info*)&subCacheHeaderBytes[subCacheHeader->mappingWithSlideOffset];
+            for (auto j = 0; j < subCacheHeader->mappingWithSlideCount; ++j) {
+                if (endAddress < subCacheMappings[j].address + subCacheMappings[j].size) {
+                    endAddress = subCacheMappings[j].address + subCacheMappings[j].size;
+                }
+            }
+            snprintf(&cachePath[0], PATH_MAX, "%s.%u", &*_files[0], i+1);
+            _files.emplace_back(libDylAllocator()->strdup(cachePath));
+        }
+        if ( (_header->mappingOffset >= __offsetof(dyld_cache_header, symbolFileUUID)) && !uuid_is_null(_header->symbolFileUUID) ) {
+            strlcpy(&cachePath[0], &*_files[0], PATH_MAX);
+            // On new caches, the locals come from a new subCache file
+            if (strstr(cachePath, ".development") != nullptr) {
+                cachePath[strlen(cachePath)-(strlen(".development"))] = 0;
+            }
+            strlcat(cachePath, ".symbols", PATH_MAX);
+            _files.emplace_back(libDylAllocator()->strdup(cachePath));
+        }
+    }
+    _size = endAddress - _header->sharedRegionStart;
+}
+
+UniquePtr<SharedCache> SharedCache::createForTask(task_read_t task, kern_return_t *kr) {
+    kern_return_t krSink = KERN_SUCCESS;
+    if (kr == nullptr) {
+        kr = &krSink;
+    }
+    mach_msg_type_number_t count = TASK_DYLD_INFO_COUNT;
+    task_dyld_info_data_t task_dyld_info;
+    if ( task_info(task, TASK_DYLD_INFO, (task_info_t)&task_dyld_info, &count) != KERN_SUCCESS) { return nullptr; }
+    //The kernel will return MACH_VM_MIN_ADDRESS for an executable that has not had dyld loaded
+    if (task_dyld_info.all_image_info_addr == MACH_VM_MIN_ADDRESS) { return nullptr; }
+    uint8_t remoteBuffer[16*1024];
+    mach_vm_size_t readSize = 0;
+    // Using mach_vm_read_overwrite because this is part of dyld. If the file is removed or the codesignature is invalid
+    // then the system is broken beyond recovery anyway
+    *kr = mach_vm_read_overwrite(task, task_dyld_info.all_image_info_addr, task_dyld_info.all_image_info_size,
+                                 (mach_vm_address_t)&remoteBuffer[0], &readSize);
+    if (*kr != KERN_SUCCESS) {
+        return nullptr;
+    }
+    DRL::UUID uuid;
+    uint64_t baseAddress                    = 0;
+    uint64_t size                           = 0;
+    uint64_t FSID                           = 0;
+    uint64_t FSObjID                        = 0;
+    bool processDetachedFromSharedRegion    = false;
+    if (task_dyld_info.all_image_info_format == TASK_DYLD_ALL_IMAGE_INFO_32 ) {
+        const dyld_all_image_infos_32* info = (const dyld_all_image_infos_32*)&remoteBuffer[0];
+        baseAddress                     = info->sharedCacheBaseAddress;
+        processDetachedFromSharedRegion = info->processDetachedFromSharedRegion;
+        FSID                            = info->cacheFSID;
+        FSObjID                         = info->cacheFSObjID;
+    } else {
+        const dyld_all_image_infos_64* info = (const dyld_all_image_infos_64*)&remoteBuffer[0];
+        baseAddress                     = info->sharedCacheBaseAddress;
+        processDetachedFromSharedRegion = info->processDetachedFromSharedRegion;
+        FSID                            = info->cacheFSID;
+        FSObjID                         = info->cacheFSObjID;
+    }
+    *kr = mach_vm_read_overwrite(task, baseAddress, 16*1024, (mach_vm_address_t)&remoteBuffer, &readSize);
+    if (*kr != KERN_SUCCESS) {
+        return nullptr;
+    }
+    auto header = ((dyld_cache_header*)&remoteBuffer[0]);
+    uuid = DRL::UUID(&header->uuid[0]);
+    for(auto i = 0; i < header->mappingCount; ++i) {
+        auto mapping = (dyld_cache_mapping_info*)&remoteBuffer[header->mappingOffset+(i*sizeof(dyld_cache_mapping_info))];
+        uint64_t regionEndSize = mapping->address + mapping->size - header->sharedRegionStart;
+        if (size < regionEndSize) {
+            size = regionEndSize;
+        }
+    }
+    char cachePath[PATH_MAX];
+
+    if (FSID && FSObjID) {
+        // Some older dyld_sims do not set processDetachedFromSharedRegion, so check for the presence of path info and use it if present
+        if (fsgetpath(cachePath, PATH_MAX, (fsid_t*)&FSID, FSObjID) == -1) {
+            *kr = KERN_FAILURE;
+            return nullptr;
+        }
+    } else {
+        if (header->platform == PLATFORM_DRIVERKIT) {
+            strlcpy(cachePath, DRIVERKIT_DYLD_SHARED_CACHE_DIR, PATH_MAX);
+        } else if constexpr(TARGET_OS_IPHONE) {
+            strlcpy(cachePath, IPHONE_DYLD_SHARED_CACHE_DIR, PATH_MAX);
+        } else {
+            strlcpy(cachePath, MACOSX_MRM_DYLD_SHARED_CACHE_DIR, PATH_MAX);
+        }
+
+        if ( strcmp(header->magic, "dyld_v1  x86_64") == 0 ) {
+            strlcat(cachePath, "dyld_shared_cache_x86_64", PATH_MAX);
+        } else if ( strcmp(header->magic, "dyld_v1 x86_64h") == 0 ) {
+            strlcat(cachePath, "dyld_shared_cache_x86_64h", PATH_MAX);
+        } else if ( strcmp(header->magic, "dyld_v1  arm64e") == 0 ) {
+            strlcat(cachePath, "dyld_shared_cache_arm64e", PATH_MAX);
+        } else if ( strcmp(header->magic, "dyld_v1   arm64") == 0 ) {
+            strlcat(cachePath, "dyld_shared_cache_arm64", PATH_MAX);
+        } else if ( strcmp(header->magic, "dyld_v1  armv7k") == 0 ) {
+            strlcat(cachePath, "dyld_shared_cache_armv7k", PATH_MAX);
+        } else if ( strcmp(header->magic, "dyld_v1arm64_32") == 0 ) {
+            strlcat(cachePath, "dyld_shared_cache_arm64_32", PATH_MAX);
+        }
+        if constexpr(TARGET_OS_IPHONE) {
+            if (header->cacheType == 0 && header->platform != PLATFORM_DRIVERKIT) {
+                strlcat(cachePath, ".development", PATH_MAX);
+            }
+        }
+    }
+
+    // Use placement new since operator new is not available
+    // TODO: We open the files to make the mapper and then again to find the paths, this can be made more efficient
+    auto mapper = Mapper::mapperForSharedCache(cachePath, uuid, (void*)baseAddress);
+    char pathBuffer[PATH_MAX];
+    if (mapper && realpath(cachePath, &pathBuffer[0])) {
+        return libDylAllocator()->makeUnique<SharedCache>(mapper, pathBuffer, processDetachedFromSharedRegion);
+    }
+    *kr = KERN_FAILURE;
+    return nullptr;
+}
+
+static bool isSubCachePath(const char* path)
+{
+    size_t pathLen = strnlen(path, PATH_MAX);
+    return ((pathLen > 1) && (path[pathLen-2] == '.') && isdigit(path[pathLen-1]));
+}
+
+void SharedCache::forEachInstalledCacheWithSystemPath(const char* systemPath, void (^block)(SharedCache* cache)) {
+    // TODO: We can make this more resilient by encoding all the paths in a special section /usr/lib/dyld, and then parsing them out
+
+    static const char* cacheDirPaths[] = {
+#if TARGET_OS_IPHONE
+        IPHONE_DYLD_SHARED_CACHE_DIR,
+#else
+        MACOSX_MRM_DYLD_SHARED_CACHE_DIR,
+#endif
+        DRIVERKIT_DYLD_SHARED_CACHE_DIR
+    };
+    for ( int i = 0; i < sizeof(cacheDirPaths)/sizeof(char*); i++ ) {
+        char systemCacheDirPath[PATH_MAX];
+        strlcpy(systemCacheDirPath, systemPath, PATH_MAX);
+        strlcat(systemCacheDirPath, cacheDirPaths[i], PATH_MAX);
+        DIR* dirp = ::opendir(systemCacheDirPath);
+        if ( dirp != NULL) {
+            dirent entry;
+            dirent* entp = NULL;
+            char cachePath[PATH_MAX];
+            while ( ::readdir_r(dirp, &entry, &entp) == 0 ) {
+                if ( entp == NULL )
+                    break;
+                if ( entp->d_type != DT_REG )
+                    continue;
+                if ( strlcpy(cachePath, systemCacheDirPath, PATH_MAX) >= PATH_MAX )
+                    continue;
+                if ( strlcat(cachePath, entp->d_name, PATH_MAX) >= PATH_MAX )
+                    continue;
+                if ( isSubCachePath(cachePath) )
+                    continue;
+                // FIXME: The memory managemnt here is awful, fix with allocators
+                auto cache = Atlas::SharedCache::createForFilePath(cachePath);
+                if (cache) {
+                    block(cache.get());
+                }
+            }
+            closedir(dirp);
+        }
+    }
+}
+
+UniquePtr<SharedCache> SharedCache::createForFilePath(const char* filePath) {
+    auto uuid = DRL::UUID();
+    auto fileMapper = Mapper::mapperForSharedCache(filePath, uuid, 0);
+    if (!fileMapper) { return nullptr; }
+    // Use placement new since operator new is not available
+    char pathBuffer[PATH_MAX];
+    if (realpath(filePath, &pathBuffer[0])) {
+        return libDylAllocator()->makeUnique<SharedCache>(fileMapper, pathBuffer, true);
+    }
+    return nullptr;
+}
+
+
+const DRL::UUID& SharedCache::uuid() const {
+    return _uuid;
+}
+
+uint64_t SharedCache::baseAddress() const {
+    return (uint64_t)_mapper->baseAddress();
+}
+
+uint64_t SharedCache::size() const {
+    return _size;
+}
+
+void SharedCache::forEachFilePath(void (^block)(const char* file_path)) const {
+    for (auto& file : _files) {
+        block(&*file);
+    }
+}
+
+bool SharedCache::isPrivateMapped() const {
+    return _private;
+}
+
+void SharedCache::forEachImage(void (^block)(Image* image)) {
+    for(auto& image : _images) {
+        block(&*image);
+    }
+}
+
+// Maps the local symbols for this shared cache.
+// Locals are in an unmapped part of the file, so we have to map then in separately
+UniquePtr<SharedCacheLocals> SharedCache::localSymbols() const {
+    // The locals might be in their own locals file, or in the main cache file.
+    // Where it is depends on the cache header
+    char localSymbolsCachePath[PATH_MAX];
+    strlcpy(&localSymbolsCachePath[0], &*_files[0], PATH_MAX);
+    bool useSymbolsFile = (_header->mappingOffset >= __offsetof(dyld_cache_header, symbolFileUUID));
+    if ( useSymbolsFile ) {
+        if ( uuid_is_null(_header->symbolFileUUID) )
+            return nullptr;
+
+        // On new caches, the locals come from a new subCache file
+        if (strstr(localSymbolsCachePath, ".development") != nullptr) {
+            localSymbolsCachePath[strlen(localSymbolsCachePath)-(strlen(".development"))] = 0;
+        }
+        strlcat(localSymbolsCachePath, ".symbols", PATH_MAX);
+    } else {
+        if ( (_header->localSymbolsSize == 0) || (_header->localSymbolsOffset == 0) )
+            return nullptr;
+    }
+
+    auto [fileMapper, baseAddress] = Mapper::mapperForSharedCacheLocals(localSymbolsCachePath);
+    if (!fileMapper) { return nullptr; }
+    // Use placement new since operator new is not available
+    return libDylAllocator()->makeUnique<SharedCacheLocals>(fileMapper, useSymbolsFile);
+}
+
+bool SharedCache::pin() {
+    return _mapper->pin();
+}
+
+void SharedCache::unpin() {
+    return _mapper->unpin();
+}
+
+#ifdef TARGET_OS_OSX
+bool SharedCache::mapSubCacheAndInvokeBlock(const dyld_cache_header* subCacheHeader,
+                                            void (^block)(const void* cacheBuffer, size_t size)) {
+    auto subCacheHeaderBytes = (uint8_t*)subCacheHeader;
+    uint64_t fileSize = 0;
+    for(auto i = 0; i < subCacheHeader->mappingCount; ++i) {
+        auto mapping = (dyld_cache_mapping_info*)&subCacheHeaderBytes[subCacheHeader->mappingOffset+(i*sizeof(dyld_cache_mapping_info))];
+        uint64_t regionEndSize = mapping->fileOffset + mapping->size;
+        if (fileSize < regionEndSize) {
+            fileSize = regionEndSize;
+        }
+    }
+    vm_address_t mappedSubCache = 0;
+    if (vm_allocate(mach_task_self(), &mappedSubCache, (size_t)fileSize, VM_FLAGS_ANYWHERE) != KERN_SUCCESS) {
+        return false;
+    }
+    for(auto i = 0; i < _header->mappingCount; ++i) {
+        //    _slide = _baseAddress -  _header->sharedRegionStart;
+        auto mapping = (dyld_cache_mapping_info*)&subCacheHeaderBytes[subCacheHeader->mappingOffset+(i*sizeof(dyld_cache_mapping_info))];
+        auto mappingBytes = _mapper->map<uint8_t>((void*)(mapping->address - _slide), mapping->size);
+        vm_copy(mach_task_self(), (vm_address_t)&*mappingBytes, (vm_size_t)mapping->size, (vm_address_t)(mappedSubCache+mapping->fileOffset));
+    }
+    block((void*)mappedSubCache, (size_t)fileSize);
+    assert(vm_deallocate(mach_task_self(), (vm_address_t)mappedSubCache, (vm_size_t)fileSize) == KERN_SUCCESS);
+    return true;
+}
+
+bool SharedCache::forEachSubcache4Rosetta(void (^block)(const void* cacheBuffer, size_t size)) {
+    if (strcmp(_header->magic, "dyld_v1  x86_64") != 0) {
+        return false;
+    }
+    uint64_t headerSize;
+    bool splitCache = false;
+    getCacheInfo(&*_header, headerSize, splitCache);
+    mapSubCacheAndInvokeBlock(&*_header, block);
+    auto headerBytes = (uint8_t*)&*_header;
+    if (splitCache) {
+        auto subCaches = (dyld_subcache_entry*)&headerBytes[_header->subCacheArrayOffset];
+        for (auto i = 0; i < _header->subCacheArrayCount; ++i) {
+            auto subCacheHeader = _mapper->map<dyld_cache_header>((void*)(baseAddress() + subCaches[i].cacheVMOffset), PAGE_SIZE);
+            uint64_t subCacheHeaderSize = subCacheHeader->mappingOffset+subCacheHeader->mappingCount*sizeof(dyld_cache_mapping_info);
+            getCacheInfo(&*_header, headerSize, splitCache);
+            if (subCacheHeaderSize > PAGE_SIZE) {
+                subCacheHeader = _mapper->map<dyld_cache_header>((void*)(baseAddress() + subCaches[i].cacheVMOffset), subCacheHeaderSize);
+            }
+//            printf("Subcache Offset: %lx\n", (uintptr_t)&headerBytes[subCaches[i].cacheVMOffset]);
+//            printf("subCacheHeader: %lx\n", (uintptr_t)&*subCacheHeader);
+//            printf("Subcache magic: %s\n", subCacheHeader->magic);
+            mapSubCacheAndInvokeBlock(&*subCacheHeader, block);
+        }
+    }
+    return true;
+}
+#endif
+
+#if BUILDING_LIBDYLD_INTROSPECTION || BUILDING_LIBDYLD || BUILDING_UNIT_TESTS
+#pragma mark -
+#pragma mark Process
+
+Process::Process(task_read_t task, kern_return_t *kr) : _task(task),
+                                                        _queue(dispatch_queue_create("com.apple.dyld.introspection", NULL)),
+                                                        _registeredNotifiers(libDylAllocator())  {}
+
+Process::~Process() {
+    dispatch_async_and_wait(_queue, ^{
+        if (_state == Connected) {
+            teardownNotifications();
+        }
+    });
+    dispatch_release(_queue);
+}
+
+UniquePtr<Process> Process::createForCurrentTask() {
+    //FIXME: We should special case this when we do full process info
+    return createForTask(mach_task_self(), nullptr);
+}
+
+UniquePtr<Process> Process::createForTask(task_read_t task, kern_return_t *kr) {
+    return libDylAllocator()->makeUnique<Process>(task, kr);
+}
+
+void Process::setupNotifications(kern_return_t *kr) {
+    assert(kr != NULL);
+    assert(_state == Disconnected);
+    // Allocate a port to listen on in this monitoring task
+    mach_port_options_t options = { .flags = MPO_IMPORTANCE_RECEIVER | MPO_CONTEXT_AS_GUARD | MPO_STRICT, .mpl = { MACH_PORT_QLIMIT_DEFAULT }};
+    *kr = mach_port_construct(mach_task_self(), &options, (mach_port_context_t)this, &_port);
+    if (*kr != KERN_SUCCESS) {
+        return;
+    }
+    // Setup notifications in case the send goes away
+    mach_port_t previous = MACH_PORT_NULL;
+    *kr = mach_port_request_notification(mach_task_self(), _port, MACH_NOTIFY_NO_SENDERS, 1, _port, MACH_MSG_TYPE_MAKE_SEND_ONCE, &previous);
+    if ((*kr != KERN_SUCCESS) || previous != MACH_PORT_NULL) {
+        (void)mach_port_destruct(mach_task_self(), _port, 0, (mach_port_context_t)this);
+        return;
+    }
+    if constexpr(TARGET_OS_SIMULATOR) {
+        static dispatch_once_t onceToken;
+        static kern_return_t (*tdpinr)(task_t, mach_port_t) = nullptr;
+        dispatch_once(&onceToken, ^{
+            tdpinr = (kern_return_t (*)(task_t, mach_port_t))dlsym(RTLD_DEFAULT, "task_dyld_process_info_notify_register");
+        });
+        if (tdpinr) {
+            *kr = tdpinr(_task, _port);
+        } else {
+            // We can fail silently here. It is a new SPI no one is using, and the new simulators will only supported on macOS's new enough
+            // to have the task_dyld_process_info_notify_register(). The only reason not to abort() is internal developers who might accidentally
+            // hit this as we transition.
+            (void)mach_port_destruct(mach_task_self(), _port, 0, (mach_port_context_t)this);
+            return;
+        }
+    } else {
+        *kr = task_dyld_process_info_notify_register(_task, _port);
+    }
+    if (*kr != KERN_SUCCESS) {
+        (void)mach_port_destruct(mach_task_self(), _port, 0, (mach_port_context_t)this);
+        return;
+    }
+    _machSource = dispatch_source_create(DISPATCH_SOURCE_TYPE_MACH_RECV, _port, 0, _queue);
+    if (_machSource == nullptr) {
+        (void)mach_port_destruct(mach_task_self(), _port, 0, (mach_port_context_t)this);
+        return;
+    }
+    dispatch_source_set_event_handler(_machSource, ^{ handleNotifications(); });
+    // Copy these into locals so the block captures them as const instead of implicitly referring to the members via this
+    task_read_t blockTask = _task;
+    mach_port_t blockPort = _port;
+    dispatch_source_t blockSource = _machSource;
+    dispatch_source_set_cancel_handler(_machSource, ^{
+        if constexpr(TARGET_OS_SIMULATOR) {
+            static dispatch_once_t onceToken;
+            static kern_return_t (*tdpind)(task_t, mach_port_t) = nullptr;
+            dispatch_once(&onceToken, ^{
+                tdpind = (kern_return_t (*)(task_t, mach_port_t))dlsym(RTLD_DEFAULT, "task_dyld_process_info_notify_deregister");
+            });
+            if (tdpind) {
+                (void)tdpind(blockTask, blockPort);
+            }
+        } else {
+            (void)task_dyld_process_info_notify_deregister(blockTask, blockPort);
+        }
+        (void)mach_port_destruct(mach_task_self(), blockPort, 0, (mach_port_context_t)this);
+        dispatch_release(blockSource);
+    });
+    dispatch_activate(_machSource);
+    _state = Connected;
+}
+
+void Process::teardownNotifications() {
+    assert(_state == Connected);
+    if (_machSource) {
+        dispatch_source_cancel(_machSource);
+        _port = 0;
+        _machSource = NULL;
+        _state = Disconnected;
+        for (auto& notiferRecord : _registeredNotifiers) {
+            if (notiferRecord.notifierID != 0) {
+                assert(notiferRecord.queue != NULL);
+                assert(notiferRecord.block != NULL);
+                dispatch_release(notiferRecord.queue);
+                Block_release(notiferRecord.block);
+                // Leaving a tombstone
+                notiferRecord = (ProcessNotifierRecord){NULL, NULL, 0};
+            }
+        }
+    }
+}
+
+void Process::handleNotifications() {
+    if (_state != Connected) { return; }
+    // This event handler block has an implicit reference to "this"
+    // if incrementing the count goes to one, that means the object may have already been destroyed
+    uint8_t messageBuffer[DYLD_PROCESS_INFO_NOTIFY_MAX_BUFFER_SIZE] = {};
+    mach_msg_header_t* h = (mach_msg_header_t*)messageBuffer;
+
+    kern_return_t r = mach_msg(h, MACH_RCV_MSG | MACH_RCV_VOUCHER| MACH_RCV_TRAILER_ELEMENTS(MACH_RCV_TRAILER_AUDIT) | MACH_RCV_TRAILER_TYPE(MACH_MSG_TRAILER_FORMAT_0), 0, sizeof(messageBuffer)-sizeof(mach_msg_audit_trailer_t), _port, MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
+    if ( r == KERN_SUCCESS && !(h->msgh_bits & MACH_MSGH_BITS_COMPLEX)) {
+        //fprintf(stderr, "received message id=0x%X, size=%d\n", h->msgh_id, h->msgh_size);
+        if ( (h->msgh_id & 0xFFFFF000) == DYLD_PROCESS_EVENT_ID_BASE ) {
+            if (h->msgh_size != sizeof(mach_msg_header_t)) {
+                teardownNotifications();
+            } else {
+                for (auto& notifier : _registeredNotifiers) {
+                    if ((h->msgh_id & ~0xFFFFF000) == notifier.notifierID) {
+                        dispatch_async_and_wait(notifier.queue, notifier.block);
+                    }
+                }
+            }
+            mach_msg_header_t replyHeader;
+            replyHeader.msgh_bits        = MACH_MSGH_BITS_SET(MACH_MSGH_BITS_REMOTE(h->msgh_bits), 0, 0, 0);
+            replyHeader.msgh_id          = 0;
+            replyHeader.msgh_local_port  = MACH_PORT_NULL;
+            replyHeader.msgh_remote_port  = h->msgh_remote_port;
+            replyHeader.msgh_reserved    = 0;
+            replyHeader.msgh_size        = sizeof(replyHeader);
+            r = mach_msg(&replyHeader, MACH_SEND_MSG, replyHeader.msgh_size, 0, MACH_PORT_NULL, 0, MACH_PORT_NULL);
+            if (r == KERN_SUCCESS) {
+                h->msgh_remote_port = MACH_PORT_NULL;
+            } else {
+                teardownNotifications();
+            }
+        } else if ( h->msgh_id == MACH_NOTIFY_NO_SENDERS ) {
+            // Validate this notification came from the kernel
+            const mach_msg_audit_trailer_t *audit_tlr = (mach_msg_audit_trailer_t *)((uint8_t *)h + round_msg(h->msgh_size));
+            if (audit_tlr->msgh_trailer_type == MACH_MSG_TRAILER_FORMAT_0
+                && audit_tlr->msgh_trailer_size >= sizeof(mach_msg_audit_trailer_t)
+                // We cannot link to libbsm, so we are hardcoding the audit token offset (5)
+                // And the value the represents the kernel (0)
+                && audit_tlr->msgh_audit.val[5] == 0) {
+                teardownNotifications();
+            }
+        } else if ( h->msgh_id != DYLD_PROCESS_INFO_NOTIFY_LOAD_ID
+                   && h->msgh_id != DYLD_PROCESS_INFO_NOTIFY_UNLOAD_ID
+                   && h->msgh_id != DYLD_PROCESS_INFO_NOTIFY_MAIN_ID) {
+            fprintf(stderr, "dyld: received unknown message id=0x%X, size=%d\n", h->msgh_id, h->msgh_size);
+        }
+    } else {
+        fprintf(stderr, "dyld: received unknown message id=0x%X, size=%d\n", h->msgh_id, h->msgh_size);
+    }
+    mach_msg_destroy(h);
+}
+
+
+uint32_t Process::registerEventHandler(kern_return_t *kr, uint32_t event, dispatch_queue_t queue, void (^block)()) {
+    __block uint32_t result = 0;
+    dispatch_async_and_wait(_queue, ^{
+        if (_state == Disconnected) {
+            setupNotifications(kr);
+            if (*kr != KERN_SUCCESS) {
+                return;
+            }
+        }
+        assert(_state == Connected);
+        dispatch_retain(queue);
+        _registeredNotifiers.emplace_back((ProcessNotifierRecord){queue, Block_copy(block), event});
+//        fprintf(stderr, "Entered:\n");
+//        for (auto i = 0 ; i < _registeredNotifiers.size(); ++i) {
+//            fprintf(stderr, "%u: event %u\n", i, _registeredNotifiers[i].notifierID);
+//        }
+        result = (uint32_t)_registeredNotifiers.size();
+//        fprintf(stderr, "result: %u\n", result);
+
+    });
+//    fprintf(stderr, "result2: %u\n", result);
+    return result;
+}
+
+void Process::unregisterEventHandler(uint32_t handle) {
+    dispatch_async_and_wait(_queue, ^{
+//        fprintf(stderr, "Trying to remove handle (%u) from %lu\n", handle, _registeredNotifiers.size());
+        assert(_registeredNotifiers.size() >= handle);
+        auto& notiferRecord = _registeredNotifiers[handle-1];
+        if (notiferRecord.notifierID == 0) {
+            // Already torndown
+            return;
+        }
+        assert(notiferRecord.queue != NULL);
+        assert(notiferRecord.block != NULL);
+        dispatch_release(notiferRecord.queue);
+        Block_release(notiferRecord.block);
+        // Leave a tombstone
+        //FIXME: Will not be necssary if we move to Map
+        notiferRecord = (ProcessNotifierRecord){NULL, NULL, 0};
+
+        bool liveNotifiers = false;
+        for (auto& notifier : _registeredNotifiers) {
+            if (notifier.notifierID != 0) {
+                liveNotifiers = true;
+                break;
+            }
+        }
+        if (!liveNotifiers) {
+            teardownNotifications();
+        }
+    });
+}
+
+#pragma mark -
+#pragma mark Process Snapshot
+
+UniquePtr<ProcessSnapshot> ProcessSnapshot::createForTask(task_read_t task, kern_return_t *kr) {
+    return libDylAllocator()->makeUnique<ProcessSnapshot>(task, kr);
+}
+
+ProcessSnapshot::ProcessSnapshot(task_read_t task, kern_return_t *kr) : _task(task), _sharedCache(SharedCache::createForTask(_task, nullptr)),
+                                                                        _images(libDylAllocator()) {
+}
+
+void ProcessSnapshot::forEachImage(void (^block)(Image* image)) {
+//    for(auto& image : _images) {
+//        block(&*image);
+//    }
+}
+
+UniquePtr<ProcessSnapshot> Process::createSnapshot(kern_return_t *kr) {
+    return ProcessSnapshot::createForTask(_task, kr);
+}
+
+UniquePtr<SharedCache>& ProcessSnapshot::sharedCache() {
+    return _sharedCache;
+}
+#endif /* BUILDING_LIBDYLD_INTROSPECTION || BUILDING_LIBDYLD */
+
+};
+};