]> git.saurik.com Git - apple/dyld.git/blobdiff - dyld3/shared-cache/SharedCacheBuilder.cpp
dyld-750.5.tar.gz
[apple/dyld.git] / dyld3 / shared-cache / SharedCacheBuilder.cpp
diff --git a/dyld3/shared-cache/SharedCacheBuilder.cpp b/dyld3/shared-cache/SharedCacheBuilder.cpp
new file mode 100644 (file)
index 0000000..b1523d1
--- /dev/null
@@ -0,0 +1,3535 @@
+/* -*- mode: C++; c-basic-offset: 4; indent-tabs-mode: nil -*-
+ *
+ * Copyright (c) 2014 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 <unistd.h>
+#include <dirent.h>
+#include <sys/errno.h>
+#include <sys/fcntl.h>
+#include <sys/param.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <mach/mach.h>
+#include <mach/mach_vm.h>
+#include <mach/mach_time.h>
+#include <mach/shared_region.h>
+#include <apfs/apfs_fsctl.h>
+
+#include <CommonCrypto/CommonHMAC.h>
+#include <CommonCrypto/CommonDigest.h>
+#include <CommonCrypto/CommonDigestSPI.h>
+
+#include "ClosureBuilder.h"
+#include "Closure.h"
+#include "ClosureFileSystemNull.h"
+#include "CodeSigningTypes.h"
+#include "MachOFileAbstraction.hpp"
+#include "SharedCacheBuilder.h"
+
+#include "FileUtils.h"
+#include "StringUtils.h"
+#include "Trie.hpp"
+
+#if __has_include("dyld_cache_config.h")
+    #include "dyld_cache_config.h"
+#else
+    #define ARM_SHARED_REGION_START      0x1A000000ULL
+    #define ARM_SHARED_REGION_SIZE       0x26000000ULL
+    #define ARM64_SHARED_REGION_START   0x180000000ULL
+    #define ARM64_SHARED_REGION_SIZE     0x100000000ULL
+#endif
+
+#ifndef ARM64_32_SHARED_REGION_START
+    #define ARM64_32_SHARED_REGION_START 0x1A000000ULL
+    #define ARM64_32_SHARED_REGION_SIZE  0x26000000ULL
+#endif
+
+#if ARM_SHARED_REGION_SIZE > 0x26000000ULL
+  #define  ARMV7K_CHAIN_BITS    0xC0000000
+  #define  ARMV7K_MAX           0x0
+#else
+  #define  ARMV7K_CHAIN_BITS    0xE0000000
+  #define  ARMV7K_MAX           0x20000000
+#endif
+
+const SharedCacheBuilder::ArchLayout SharedCacheBuilder::_s_archLayout[] = {
+    { 0x7FFF20000000ULL,            0xEFE00000ULL,              0x0,         0x40000000, 0x00FFFF0000000000, "x86_64",   CS_PAGE_SIZE_4K,  12, 2, true,  true,  true  },
+    { 0x7FFF20000000ULL,            0xEFE00000ULL,              0x0,         0x40000000, 0x00FFFF0000000000, "x86_64h",  CS_PAGE_SIZE_4K,  12, 2, true,  true,  true  },
+    { SHARED_REGION_BASE_I386,      SHARED_REGION_SIZE_I386,    0x0,         0x00200000,                0x0, "i386",     CS_PAGE_SIZE_4K,  12, 0, false, false, true  },
+    { ARM64_SHARED_REGION_START,    ARM64_SHARED_REGION_SIZE,   0x0,         0x02000000, 0x00FFFF0000000000, "arm64",    CS_PAGE_SIZE_4K,  14, 2, false, true,  false },
+#if SUPPORT_ARCH_arm64e
+    { ARM64_SHARED_REGION_START,    ARM64_SHARED_REGION_SIZE,   0x0,         0x02000000, 0x00FFFF0000000000, "arm64e",   CS_PAGE_SIZE_16K, 14, 2, false, true,  false },
+#endif
+#if SUPPORT_ARCH_arm64_32
+    { ARM64_32_SHARED_REGION_START, ARM64_32_SHARED_REGION_SIZE,0x0,         0x02000000,         0xC0000000, "arm64_32", CS_PAGE_SIZE_16K, 14, 6, false, false, true  },
+#endif
+    { ARM_SHARED_REGION_START,      ARM_SHARED_REGION_SIZE,     0x0,         0x02000000,         0xE0000000, "armv7s",   CS_PAGE_SIZE_4K,  14, 4, false, false, true  },
+    { ARM_SHARED_REGION_START,      ARM_SHARED_REGION_SIZE,     ARMV7K_MAX,  0x00400000,  ARMV7K_CHAIN_BITS, "armv7k",   CS_PAGE_SIZE_4K,  14, 4, false, false, true  },
+    { 0x40000000,                   0x40000000,                 0x0,         0x02000000,                0x0, "sim-x86",  CS_PAGE_SIZE_4K,  14, 0, false, false, true  }
+};
+
+
+// These are dylibs that may be interposed, so stubs calling into them should never be bypassed
+const char* const SharedCacheBuilder::_s_neverStubEliminateDylibs[] = {
+    "/usr/lib/system/libdispatch.dylib",
+    nullptr
+};
+
+// These are functions that are interposed by Instruments.app or ASan
+const char* const SharedCacheBuilder::_s_neverStubEliminateSymbols[] = {
+    "___bzero",
+    "___cxa_atexit",
+    "___cxa_throw",
+    "__longjmp",
+    "__objc_autoreleasePoolPop",
+    "_accept",
+    "_access",
+    "_asctime",
+    "_asctime_r",
+    "_asprintf",
+    "_atoi",
+    "_atol",
+    "_atoll",
+    "_calloc",
+    "_chmod",
+    "_chown",
+    "_close",
+    "_confstr",
+    "_ctime",
+    "_ctime_r",
+    "_dispatch_after",
+    "_dispatch_after_f",
+    "_dispatch_async",
+    "_dispatch_async_f",
+    "_dispatch_barrier_async_f",
+    "_dispatch_group_async",
+    "_dispatch_group_async_f",
+    "_dispatch_source_set_cancel_handler",
+    "_dispatch_source_set_event_handler",
+    "_dispatch_sync_f",
+    "_dlclose",
+    "_dlopen",
+    "_dup",
+    "_dup2",
+    "_endgrent",
+    "_endpwent",
+    "_ether_aton",
+    "_ether_hostton",
+    "_ether_line",
+    "_ether_ntoa",
+    "_ether_ntohost",
+    "_fchmod",
+    "_fchown",
+    "_fclose",
+    "_fdopen",
+    "_fflush",
+    "_fopen",
+    "_fork",
+    "_fprintf",
+    "_free",
+    "_freopen",
+    "_frexp",
+    "_frexpf",
+    "_frexpl",
+    "_fscanf",
+    "_fstat",
+    "_fstatfs",
+    "_fstatfs64",
+    "_fsync",
+    "_ftime",
+    "_getaddrinfo",
+    "_getattrlist",
+    "_getcwd",
+    "_getgrent",
+    "_getgrgid",
+    "_getgrgid_r",
+    "_getgrnam",
+    "_getgrnam_r",
+    "_getgroups",
+    "_gethostbyaddr",
+    "_gethostbyname",
+    "_gethostbyname2",
+    "_gethostent",
+    "_getifaddrs",
+    "_getitimer",
+    "_getnameinfo",
+    "_getpass",
+    "_getpeername",
+    "_getpwent",
+    "_getpwnam",
+    "_getpwnam_r",
+    "_getpwuid",
+    "_getpwuid_r",
+    "_getsockname",
+    "_getsockopt",
+    "_gmtime",
+    "_gmtime_r",
+    "_if_indextoname",
+    "_if_nametoindex",
+    "_index",
+    "_inet_aton",
+    "_inet_ntop",
+    "_inet_pton",
+    "_initgroups",
+    "_ioctl",
+    "_lchown",
+    "_lgamma",
+    "_lgammaf",
+    "_lgammal",
+    "_link",
+    "_listxattr",
+    "_localtime",
+    "_localtime_r",
+    "_longjmp",
+    "_lseek",
+    "_lstat",
+    "_malloc",
+    "_malloc_create_zone",
+    "_malloc_default_purgeable_zone",
+    "_malloc_default_zone",
+    "_malloc_good_size",
+    "_malloc_make_nonpurgeable",
+    "_malloc_make_purgeable",
+    "_malloc_set_zone_name",
+    "_mbsnrtowcs",
+    "_mbsrtowcs",
+    "_mbstowcs",
+    "_memchr",
+    "_memcmp",
+    "_memcpy",
+    "_memmove",
+    "_memset",
+    "_mktime",
+    "_mlock",
+    "_mlockall",
+    "_modf",
+    "_modff",
+    "_modfl",
+    "_munlock",
+    "_munlockall",
+    "_objc_autoreleasePoolPop",
+    "_objc_setProperty",
+    "_objc_setProperty_atomic",
+    "_objc_setProperty_atomic_copy",
+    "_objc_setProperty_nonatomic",
+    "_objc_setProperty_nonatomic_copy",
+    "_objc_storeStrong",
+    "_open",
+    "_opendir",
+    "_poll",
+    "_posix_memalign",
+    "_pread",
+    "_printf",
+    "_pthread_attr_getdetachstate",
+    "_pthread_attr_getguardsize",
+    "_pthread_attr_getinheritsched",
+    "_pthread_attr_getschedparam",
+    "_pthread_attr_getschedpolicy",
+    "_pthread_attr_getscope",
+    "_pthread_attr_getstack",
+    "_pthread_attr_getstacksize",
+    "_pthread_condattr_getpshared",
+    "_pthread_create",
+    "_pthread_getschedparam",
+    "_pthread_join",
+    "_pthread_mutex_lock",
+    "_pthread_mutex_unlock",
+    "_pthread_mutexattr_getprioceiling",
+    "_pthread_mutexattr_getprotocol",
+    "_pthread_mutexattr_getpshared",
+    "_pthread_mutexattr_gettype",
+    "_pthread_rwlockattr_getpshared",
+    "_pwrite",
+    "_rand_r",
+    "_read",
+    "_readdir",
+    "_readdir_r",
+    "_readv",
+    "_readv$UNIX2003",
+    "_realloc",
+    "_realpath",
+    "_recv",
+    "_recvfrom",
+    "_recvmsg",
+    "_remquo",
+    "_remquof",
+    "_remquol",
+    "_scanf",
+    "_send",
+    "_sendmsg",
+    "_sendto",
+    "_setattrlist",
+    "_setgrent",
+    "_setitimer",
+    "_setlocale",
+    "_setpwent",
+    "_shm_open",
+    "_shm_unlink",
+    "_sigaction",
+    "_sigemptyset",
+    "_sigfillset",
+    "_siglongjmp",
+    "_signal",
+    "_sigpending",
+    "_sigprocmask",
+    "_sigwait",
+    "_snprintf",
+    "_sprintf",
+    "_sscanf",
+    "_stat",
+    "_statfs",
+    "_statfs64",
+    "_strcasecmp",
+    "_strcat",
+    "_strchr",
+    "_strcmp",
+    "_strcpy",
+    "_strdup",
+    "_strerror",
+    "_strerror_r",
+    "_strlen",
+    "_strncasecmp",
+    "_strncat",
+    "_strncmp",
+    "_strncpy",
+    "_strptime",
+    "_strtoimax",
+    "_strtol",
+    "_strtoll",
+    "_strtoumax",
+    "_tempnam",
+    "_time",
+    "_times",
+    "_tmpnam",
+    "_tsearch",
+    "_unlink",
+    "_valloc",
+    "_vasprintf",
+    "_vfprintf",
+    "_vfscanf",
+    "_vprintf",
+    "_vscanf",
+    "_vsnprintf",
+    "_vsprintf",
+    "_vsscanf",
+    "_wait",
+    "_wait$UNIX2003",
+    "_wait3",
+    "_wait4",
+    "_waitid",
+    "_waitid$UNIX2003",
+    "_waitpid",
+    "_waitpid$UNIX2003",
+    "_wcslen",
+    "_wcsnrtombs",
+    "_wcsrtombs",
+    "_wcstombs",
+    "_wordexp",
+    "_write",
+    "_writev",
+    "_writev$UNIX2003",
+    // <rdar://problem/22050956> always use stubs for C++ symbols that can be overridden
+    "__ZdaPv",
+    "__ZdlPv",
+    "__Znam",
+    "__Znwm",
+
+    nullptr
+};
+
+
+inline uint32_t absolutetime_to_milliseconds(uint64_t abstime)
+{
+    return (uint32_t)(abstime/1000/1000);
+}
+
+// Handles building a list of input files to the SharedCacheBuilder itself.
+class CacheInputBuilder {
+public:
+    CacheInputBuilder(const dyld3::closure::FileSystem& fileSystem,
+                      const dyld3::GradedArchs& archs, dyld3::Platform reqPlatform)
+    : fileSystem(fileSystem), reqArchs(archs), reqPlatform(reqPlatform) { }
+
+    // Loads and maps any MachOs in the given list of files.
+    void loadMachOs(std::vector<CacheBuilder::InputFile>& inputFiles,
+                    std::vector<CacheBuilder::LoadedMachO>& dylibsToCache,
+                    std::vector<CacheBuilder::LoadedMachO>& otherDylibs,
+                    std::vector<CacheBuilder::LoadedMachO>& executables,
+                    std::vector<CacheBuilder::LoadedMachO>& couldNotLoadFiles) {
+
+        std::map<std::string, uint64_t> dylibInstallNameMap;
+        for (CacheBuilder::InputFile& inputFile : inputFiles) {
+            char realerPath[MAXPATHLEN];
+            dyld3::closure::LoadedFileInfo loadedFileInfo = dyld3::MachOAnalyzer::load(inputFile.diag, fileSystem, inputFile.path, reqArchs, reqPlatform, realerPath);
+            const dyld3::MachOAnalyzer* ma = (const dyld3::MachOAnalyzer*)loadedFileInfo.fileContent;
+            if (ma == nullptr) {
+                couldNotLoadFiles.emplace_back((CacheBuilder::LoadedMachO){ DyldSharedCache::MappedMachO(), loadedFileInfo, &inputFile });
+                continue;
+            }
+
+            DyldSharedCache::MappedMachO mappedFile(inputFile.path, ma, loadedFileInfo.sliceLen, false, false,
+                                                    loadedFileInfo.sliceOffset, loadedFileInfo.mtime, loadedFileInfo.inode);
+
+            // The file can be loaded with the given slice, but we may still want to exlude it from the cache.
+            if (ma->isDylib()) {
+                std::string installName = ma->installName();
+
+                // Let the platform exclude the file before we do anything else.
+                if (platformExcludesInstallName(installName)) {
+                    inputFile.diag.verbose("Platform excluded file\n");
+                    fileSystem.unloadFile(loadedFileInfo);
+                    continue;
+                }
+
+                if (!ma->canBePlacedInDyldCache(inputFile.path, ^(const char* msg) {
+                    inputFile.diag.warning("Dylib located at '%s' cannot be placed in cache because: %s", inputFile.path, msg);
+                })) {
+                    // TODO: Add exclusion lists here?
+                    // Probably not as we already applied the dylib exclusion list.
+                    if (!ma->canHavePrecomputedDlopenClosure(inputFile.path, ^(const char* msg) {
+                        inputFile.diag.verbose("Dylib located at '%s' cannot prebuild dlopen closure in cache because: %s", inputFile.path, msg);
+                    }) ) {
+                        fileSystem.unloadFile(loadedFileInfo);
+                        continue;
+                    }
+                    otherDylibs.emplace_back((CacheBuilder::LoadedMachO){ mappedFile, loadedFileInfo, &inputFile });
+                    continue;
+                }
+
+                // Otherwise see if we have another file with this install name
+                auto iteratorAndInserted = dylibInstallNameMap.insert(std::make_pair(installName, dylibsToCache.size()));
+                if (iteratorAndInserted.second) {
+                    // We inserted the dylib so we haven't seen another with this name.
+                    if (installName[0] != '@' && installName != inputFile.path) {
+                        inputFile.diag.warning("Dylib located at '%s' has installname '%s'", inputFile.path, installName.c_str());
+                    }
+
+                    dylibsToCache.emplace_back((CacheBuilder::LoadedMachO){ mappedFile, loadedFileInfo, &inputFile });
+                } else {
+                    // We didn't insert this one so we've seen it before.
+                    CacheBuilder::LoadedMachO& previousLoadedMachO = dylibsToCache[iteratorAndInserted.first->second];
+                    inputFile.diag.warning("Multiple dylibs claim installname '%s' ('%s' and '%s')", installName.c_str(), inputFile.path, previousLoadedMachO.mappedFile.runtimePath.c_str());
+
+                    // This is the "Good" one, overwrite
+                    if (inputFile.path == installName) {
+                        // Unload the old one
+                        fileSystem.unloadFile(previousLoadedMachO.loadedFileInfo);
+
+                        // And replace with this one.
+                        previousLoadedMachO.mappedFile = mappedFile;
+                        previousLoadedMachO.loadedFileInfo = loadedFileInfo;
+                    }
+                }
+            } else if (ma->isBundle()) {
+                // TODO: Add exclusion lists here?
+                if (!ma->canHavePrecomputedDlopenClosure(inputFile.path, ^(const char* msg) {
+                    inputFile.diag.verbose("Dylib located at '%s' cannot prebuild dlopen closure in cache because: %s", inputFile.path, msg);
+                }) ) {
+                    fileSystem.unloadFile(loadedFileInfo);
+                    continue;
+                }
+                otherDylibs.emplace_back((CacheBuilder::LoadedMachO){ mappedFile, loadedFileInfo, &inputFile });
+            } else if (ma->isDynamicExecutable()) {
+                if (platformExcludesExecutablePath_macOS(inputFile.path)) {
+                    inputFile.diag.verbose("Platform excluded file\n");
+                    fileSystem.unloadFile(loadedFileInfo);
+                    continue;
+                }
+                executables.emplace_back((CacheBuilder::LoadedMachO){ mappedFile, loadedFileInfo, &inputFile });
+            } else {
+                inputFile.diag.verbose("Unsupported mach file type\n");
+                fileSystem.unloadFile(loadedFileInfo);
+            }
+        }
+    }
+
+private:
+
+
+
+    static bool platformExcludesInstallName_macOS(const std::string& installName) {
+        return false;
+    }
+
+    static bool platformExcludesInstallName_iOS(const std::string& installName) {
+        if ( installName == "/System/Library/Caches/com.apple.xpc/sdk.dylib" )
+            return true;
+        if ( installName == "/System/Library/Caches/com.apple.xpcd/xpcd_cache.dylib" )
+            return true;
+        return false;
+    }
+
+    static bool platformExcludesInstallName_tvOS(const std::string& installName) {
+        return platformExcludesInstallName_iOS(installName);
+    }
+
+    static bool platformExcludesInstallName_watchOS(const std::string& installName) {
+        return platformExcludesInstallName_iOS(installName);
+    }
+
+    static bool platformExcludesInstallName_bridgeOS(const std::string& installName) {
+        return platformExcludesInstallName_iOS(installName);
+    }
+
+    // Returns true if the current platform requires that this install name be excluded from the shared cache
+    // Note that this overrides any exclusion from anywhere else.
+    bool platformExcludesInstallName(const std::string& installName) {
+        switch (reqPlatform) {
+            case dyld3::Platform::unknown:
+                return false;
+            case dyld3::Platform::macOS:
+                return platformExcludesInstallName_macOS(installName);
+            case dyld3::Platform::iOS:
+                return platformExcludesInstallName_iOS(installName);
+            case dyld3::Platform::tvOS:
+                return platformExcludesInstallName_tvOS(installName);
+            case dyld3::Platform::watchOS:
+                return platformExcludesInstallName_watchOS(installName);
+            case dyld3::Platform::bridgeOS:
+                return platformExcludesInstallName_bridgeOS(installName);
+            case dyld3::Platform::iOSMac:
+                return false;
+            case dyld3::Platform::iOS_simulator:
+                return false;
+            case dyld3::Platform::tvOS_simulator:
+                return false;
+            case dyld3::Platform::watchOS_simulator:
+                return false;
+            case dyld3::Platform::driverKit:
+                return false;
+        }
+    }
+
+
+
+
+    static bool platformExcludesExecutablePath_macOS(const std::string& path) {
+        return false;
+    }
+
+    static bool platformExcludesExecutablePath_iOS(const std::string& path) {
+        //HACK exclude all launchd and installd variants until we can do something about xpcd_cache.dylib and friends
+        if (path == "/sbin/launchd"
+            || path == "/usr/local/sbin/launchd.debug"
+            || path == "/usr/local/sbin/launchd.development"
+            || path == "/usr/libexec/installd") {
+            return true;
+        }
+        return false;
+    }
+
+    static bool platformExcludesExecutablePath_tvOS(const std::string& path) {
+        return platformExcludesExecutablePath_iOS(path);
+    }
+
+    static bool platformExcludesExecutablePath_watchOS(const std::string& path) {
+        return platformExcludesExecutablePath_iOS(path);
+    }
+
+    static bool platformExcludesExecutablePath_bridgeOS(const std::string& path) {
+        return platformExcludesExecutablePath_iOS(path);
+    }
+
+    // Returns true if the current platform requires that this path be excluded from the shared cache
+    // Note that this overrides any exclusion from anywhere else.
+    bool platformExcludesExecutablePath(const std::string& path) {
+        switch (reqPlatform) {
+            case dyld3::Platform::unknown:
+                return false;
+            case dyld3::Platform::macOS:
+                return platformExcludesExecutablePath_macOS(path);
+            case dyld3::Platform::iOS:
+                return platformExcludesExecutablePath_iOS(path);
+            case dyld3::Platform::tvOS:
+                return platformExcludesExecutablePath_tvOS(path);
+            case dyld3::Platform::watchOS:
+                return platformExcludesExecutablePath_watchOS(path);
+            case dyld3::Platform::bridgeOS:
+                return platformExcludesExecutablePath_bridgeOS(path);
+            case dyld3::Platform::iOSMac:
+                return false;
+            case dyld3::Platform::iOS_simulator:
+                return false;
+            case dyld3::Platform::tvOS_simulator:
+                return false;
+            case dyld3::Platform::watchOS_simulator:
+                return false;
+            case dyld3::Platform::driverKit:
+                return false;
+        }
+    }
+
+    const dyld3::closure::FileSystem&                   fileSystem;
+    const dyld3::GradedArchs&                           reqArchs;
+    dyld3::Platform                                     reqPlatform;
+};
+
+SharedCacheBuilder::SharedCacheBuilder(const DyldSharedCache::CreateOptions& options,
+                                       const dyld3::closure::FileSystem& fileSystem)
+    : CacheBuilder(options, fileSystem) {
+
+    std::string targetArch = options.archs->name();
+    if ( options.forSimulator && (options.archs == &dyld3::GradedArchs::i386) )
+        targetArch = "sim-x86";
+
+    for (const ArchLayout& layout : _s_archLayout) {
+        if ( layout.archName == targetArch ) {
+            _archLayout = &layout;
+            _is64 = _archLayout->is64;
+            break;
+        }
+    }
+
+    if (!_archLayout) {
+        _diagnostics.error("Tool was built without support for: '%s'", targetArch.c_str());
+    }
+}
+
+static void verifySelfContained(std::vector<CacheBuilder::LoadedMachO>& dylibsToCache,
+                                std::vector<CacheBuilder::LoadedMachO>& otherDylibs,
+                                std::vector<CacheBuilder::LoadedMachO>& couldNotLoadFiles)
+{
+    // build map of dylibs
+    __block std::map<std::string, const CacheBuilder::LoadedMachO*> knownDylibs;
+    __block std::map<std::string, const CacheBuilder::LoadedMachO*> allDylibs;
+    for (const CacheBuilder::LoadedMachO& dylib : dylibsToCache) {
+        knownDylibs.insert({ dylib.mappedFile.runtimePath, &dylib });
+        allDylibs.insert({ dylib.mappedFile.runtimePath, &dylib });
+        if (const char* installName = dylib.mappedFile.mh->installName()) {
+            knownDylibs.insert({ installName, &dylib });
+            allDylibs.insert({ installName, &dylib });
+        }
+    }
+
+    for (const CacheBuilder::LoadedMachO& dylib : otherDylibs) {
+        allDylibs.insert({ dylib.mappedFile.runtimePath, &dylib });
+        if (const char* installName = dylib.mappedFile.mh->installName())
+            allDylibs.insert({ installName, &dylib });
+    }
+
+    for (const CacheBuilder::LoadedMachO& dylib : couldNotLoadFiles) {
+        allDylibs.insert({ dylib.inputFile->path, &dylib });
+    }
+
+    // check all dependencies to assure every dylib in cache only depends on other dylibs in cache
+    __block std::map<std::string, std::set<std::string>> badDylibs;
+    __block bool doAgain = true;
+    while ( doAgain ) {
+        doAgain = false;
+        // scan dylib list making sure all dependents are in dylib list
+        for (const CacheBuilder::LoadedMachO& dylib : dylibsToCache) {
+            if ( badDylibs.count(dylib.mappedFile.runtimePath) != 0 )
+                continue;
+            dylib.mappedFile.mh->forEachDependentDylib(^(const char* loadPath, bool isWeak, bool isReExport, bool isUpward, uint32_t compatVersion, uint32_t curVersion, bool& stop) {
+                if (isWeak)
+                    return;
+                if ( knownDylibs.count(loadPath) == 0 ) {
+                    badDylibs[dylib.mappedFile.runtimePath].insert(std::string("Could not find dependency '") + loadPath + "'");
+                    knownDylibs.erase(dylib.mappedFile.runtimePath);
+                    knownDylibs.erase(dylib.mappedFile.mh->installName());
+                    doAgain = true;
+                }
+            });
+        }
+    }
+
+    // Now walk the dylibs which depend on missing dylibs and see if any of them are required binaries.
+    for (auto badDylibsIterator : badDylibs) {
+        const std::string& dylibRuntimePath = badDylibsIterator.first;
+        auto requiredDylibIterator = allDylibs.find(dylibRuntimePath);
+        if (requiredDylibIterator == allDylibs.end())
+            continue;
+        if (!requiredDylibIterator->second->inputFile->mustBeIncluded())
+            continue;
+        // This dylib is required so mark all dependencies as requried too
+        __block std::vector<const CacheBuilder::LoadedMachO*> worklist;
+        worklist.push_back(requiredDylibIterator->second);
+        while (!worklist.empty()) {
+            const CacheBuilder::LoadedMachO* dylib = worklist.back();
+            worklist.pop_back();
+            if (!dylib->mappedFile.mh)
+                continue;
+            dylib->mappedFile.mh->forEachDependentDylib(^(const char* loadPath, bool isWeak, bool isReExport, bool isUpward, uint32_t compatVersion, uint32_t curVersion, bool& stop) {
+                if (isWeak)
+                    return;
+                auto dylibIterator = allDylibs.find(loadPath);
+                if (dylibIterator != allDylibs.end()) {
+                    if (dylibIterator->second->inputFile->state == CacheBuilder::InputFile::Unset) {
+                        dylibIterator->second->inputFile->state = CacheBuilder::InputFile::MustBeIncludedForDependent;
+                        worklist.push_back(dylibIterator->second);
+                    }
+                }
+            });
+        }
+    }
+
+    // FIXME: Make this an option we can pass in
+    const bool evictLeafDylibs = true;
+    if (evictLeafDylibs) {
+        doAgain = true;
+        while ( doAgain ) {
+            doAgain = false;
+
+            // build count of how many references there are to each dylib
+            __block std::set<std::string> referencedDylibs;
+            for (const CacheBuilder::LoadedMachO& dylib : dylibsToCache) {
+                if ( badDylibs.count(dylib.mappedFile.runtimePath) != 0 )
+                    continue;
+                dylib.mappedFile.mh->forEachDependentDylib(^(const char* loadPath, bool isWeak, bool isReExport, bool isUpward, uint32_t compatVersion, uint32_t curVersion, bool &stop) {
+                    referencedDylibs.insert(loadPath);
+                });
+            }
+
+            // find all dylibs not referenced
+            for (const CacheBuilder::LoadedMachO& dylib : dylibsToCache) {
+                if ( badDylibs.count(dylib.mappedFile.runtimePath) != 0 )
+                    continue;
+                const char* installName = dylib.mappedFile.mh->installName();
+                if ( (referencedDylibs.count(installName) == 0) && (dylib.inputFile->state == CacheBuilder::InputFile::MustBeExcludedIfUnused) ) {
+                    badDylibs[dylib.mappedFile.runtimePath].insert(std::string("It has been explicitly excluded as it is unused"));
+                    doAgain = true;
+                }
+            }
+        }
+    }
+
+    // Move bad dylibs from dylibs to cache to other dylibs.
+    for (const CacheBuilder::LoadedMachO& dylib : dylibsToCache) {
+        auto i = badDylibs.find(dylib.mappedFile.runtimePath);
+        if ( i != badDylibs.end()) {
+            otherDylibs.push_back(dylib);
+            for (const std::string& reason : i->second )
+                otherDylibs.back().inputFile->diag.warning("Dylib located at '%s' not placed in shared cache because: %s", dylib.mappedFile.runtimePath.c_str(), reason.c_str());
+        }
+    }
+
+    const auto& badDylibsLambdaRef = badDylibs;
+    dylibsToCache.erase(std::remove_if(dylibsToCache.begin(), dylibsToCache.end(), [&](const CacheBuilder::LoadedMachO& dylib) {
+        if (badDylibsLambdaRef.find(dylib.mappedFile.runtimePath) != badDylibsLambdaRef.end())
+            return true;
+        return false;
+    }), dylibsToCache.end());
+}
+
+// This is the new build API which takes the raw files (which could be FAT) and tries to build a cache from them.
+// We should remove the other build() method, or make it private so that this can wrap it.
+void SharedCacheBuilder::build(std::vector<CacheBuilder::InputFile>& inputFiles,
+                               std::vector<DyldSharedCache::FileAlias>& aliases) {
+    // First filter down to files which are actually MachO's
+    CacheInputBuilder cacheInputBuilder(_fileSystem, *_options.archs, _options.platform);
+
+    std::vector<LoadedMachO> dylibsToCache;
+    std::vector<LoadedMachO> otherDylibs;
+    std::vector<LoadedMachO> executables;
+    std::vector<LoadedMachO> couldNotLoadFiles;
+    cacheInputBuilder.loadMachOs(inputFiles, dylibsToCache, otherDylibs, executables, couldNotLoadFiles);
+
+    verifySelfContained(dylibsToCache, otherDylibs, couldNotLoadFiles);
+
+    // Check for required binaries before we try to build the cache
+    if (!_diagnostics.hasError()) {
+        // If we succeeded in building, then now see if there was a missing required file, and if so why its missing.
+        std::string errorString;
+        for (const LoadedMachO& dylib : otherDylibs) {
+            if (dylib.inputFile->mustBeIncluded()) {
+                // An error loading a required file must be propagated up to the top level diagnostic handler.
+                bool gotWarning = false;
+                for (const std::string& warning : dylib.inputFile->diag.warnings()) {
+                    gotWarning = true;
+                    std::string message = warning;
+                    if (message.back() == '\n')
+                        message.pop_back();
+                    if (!errorString.empty())
+                        errorString += "ERROR: ";
+                    errorString += "Required binary was not included in the shared cache '" + std::string(dylib.inputFile->path) + "' because: " + message + "\n";
+                }
+                if (!gotWarning) {
+                    if (!errorString.empty())
+                        errorString += "ERROR: ";
+                    errorString += "Required binary was not included in the shared cache '" + std::string(dylib.inputFile->path) + "' because: 'unknown error.  Please report to dyld'\n";
+                }
+            }
+        }
+        for (const LoadedMachO& dylib : couldNotLoadFiles) {
+            if (dylib.inputFile->mustBeIncluded()) {
+                if (dylib.inputFile->diag.hasError()) {
+                    if (!errorString.empty())
+                        errorString += "ERROR: ";
+                    errorString += "Required binary was not included in the shared cache '" + std::string(dylib.inputFile->path) + "' because: " + dylib.inputFile->diag.errorMessage() + "\n";
+                } else {
+                    if (!errorString.empty())
+                        errorString += "ERROR: ";
+                    errorString += "Required binary was not included in the shared cache '" + std::string(dylib.inputFile->path) + "' because: 'unknown error.  Please report to dyld'\n";
+
+                }
+            }
+        }
+        if (!errorString.empty()) {
+            _diagnostics.error("%s", errorString.c_str());
+        }
+    }
+
+    if (!_diagnostics.hasError())
+        build(dylibsToCache, otherDylibs, executables, aliases);
+
+    if (!_diagnostics.hasError()) {
+        // If we succeeded in building, then now see if there was a missing required file, and if so why its missing.
+        std::string errorString;
+        for (CacheBuilder::InputFile& inputFile : inputFiles) {
+            if (inputFile.mustBeIncluded() && inputFile.diag.hasError()) {
+                // An error loading a required file must be propagated up to the top level diagnostic handler.
+                std::string message = inputFile.diag.errorMessage();
+                if (message.back() == '\n')
+                    message.pop_back();
+                errorString += "Required binary was not included in the shared cache '" + std::string(inputFile.path) + "' because: " + message + "\n";
+            }
+        }
+        if (!errorString.empty()) {
+            _diagnostics.error("%s", errorString.c_str());
+        }
+    }
+
+    // Add all the warnings from the input files to the top level warnings on the main diagnostics object.
+    for (CacheBuilder::InputFile& inputFile : inputFiles) {
+        for (const std::string& warning : inputFile.diag.warnings())
+            _diagnostics.warning("%s", warning.c_str());
+    }
+
+    // Clean up the loaded files
+    for (LoadedMachO& loadedMachO : dylibsToCache)
+        _fileSystem.unloadFile(loadedMachO.loadedFileInfo);
+    for (LoadedMachO& loadedMachO : otherDylibs)
+        _fileSystem.unloadFile(loadedMachO.loadedFileInfo);
+    for (LoadedMachO& loadedMachO : executables)
+        _fileSystem.unloadFile(loadedMachO.loadedFileInfo);
+}
+
+void SharedCacheBuilder::build(const std::vector<DyldSharedCache::MappedMachO>& dylibs,
+                               const std::vector<DyldSharedCache::MappedMachO>& otherOsDylibsInput,
+                               const std::vector<DyldSharedCache::MappedMachO>& osExecutables,
+                               std::vector<DyldSharedCache::FileAlias>& aliases) {
+
+    std::vector<LoadedMachO> dylibsToCache;
+    std::vector<LoadedMachO> otherDylibs;
+    std::vector<LoadedMachO> executables;
+
+    for (const DyldSharedCache::MappedMachO& mappedMachO : dylibs) {
+        dyld3::closure::LoadedFileInfo loadedFileInfo;
+        loadedFileInfo.fileContent      = mappedMachO.mh;
+        loadedFileInfo.fileContentLen   = mappedMachO.length;
+        loadedFileInfo.sliceOffset      = mappedMachO.sliceFileOffset;
+        loadedFileInfo.sliceLen         = mappedMachO.length;
+        loadedFileInfo.inode            = mappedMachO.inode;
+        loadedFileInfo.mtime            = mappedMachO.modTime;
+        loadedFileInfo.path             = mappedMachO.runtimePath.c_str();
+        dylibsToCache.emplace_back((LoadedMachO){ mappedMachO, loadedFileInfo, nullptr });
+    }
+
+    for (const DyldSharedCache::MappedMachO& mappedMachO : otherOsDylibsInput) {
+        dyld3::closure::LoadedFileInfo loadedFileInfo;
+        loadedFileInfo.fileContent      = mappedMachO.mh;
+        loadedFileInfo.fileContentLen   = mappedMachO.length;
+        loadedFileInfo.sliceOffset      = mappedMachO.sliceFileOffset;
+        loadedFileInfo.sliceLen         = mappedMachO.length;
+        loadedFileInfo.inode            = mappedMachO.inode;
+        loadedFileInfo.mtime            = mappedMachO.modTime;
+        loadedFileInfo.path             = mappedMachO.runtimePath.c_str();
+        otherDylibs.emplace_back((LoadedMachO){ mappedMachO, loadedFileInfo, nullptr });
+    }
+
+    for (const DyldSharedCache::MappedMachO& mappedMachO : osExecutables) {
+        dyld3::closure::LoadedFileInfo loadedFileInfo;
+        loadedFileInfo.fileContent      = mappedMachO.mh;
+        loadedFileInfo.fileContentLen   = mappedMachO.length;
+        loadedFileInfo.sliceOffset      = mappedMachO.sliceFileOffset;
+        loadedFileInfo.sliceLen         = mappedMachO.length;
+        loadedFileInfo.inode            = mappedMachO.inode;
+        loadedFileInfo.mtime            = mappedMachO.modTime;
+        loadedFileInfo.path             = mappedMachO.runtimePath.c_str();
+        executables.emplace_back((LoadedMachO){ mappedMachO, loadedFileInfo, nullptr });
+    }
+
+    build(dylibsToCache, otherDylibs, executables, aliases);
+}
+
+void SharedCacheBuilder::build(const std::vector<LoadedMachO>& dylibs,
+                               const std::vector<LoadedMachO>& otherOsDylibsInput,
+                               const std::vector<LoadedMachO>& osExecutables,
+                               std::vector<DyldSharedCache::FileAlias>& aliases)
+{
+    // <rdar://problem/21317611> error out instead of crash if cache has no dylibs
+    // FIXME: plist should specify required vs optional dylibs
+    if ( dylibs.size() < 30 ) {
+        _diagnostics.error("missing required minimum set of dylibs");
+        return;
+    }
+    uint64_t t1 = mach_absolute_time();
+
+    // make copy of dylib list and sort
+    makeSortedDylibs(dylibs, _options.dylibOrdering);
+
+    // allocate space used by largest possible cache plus room for LINKEDITS before optimization
+    _allocatedBufferSize = _archLayout->sharedMemorySize * 1.50;
+    if ( vm_allocate(mach_task_self(), &_fullAllocatedBuffer, _allocatedBufferSize, VM_FLAGS_ANYWHERE) != 0 ) {
+        _diagnostics.error("could not allocate buffer");
+        return;
+    }
+
+    // assign addresses for each segment of each dylib in new cache
+    parseCoalescableSegments();
+    processSelectorStrings(osExecutables);
+    assignSegmentAddresses();
+    std::vector<const LoadedMachO*> overflowDylibs;
+    while ( cacheOverflowAmount() != 0 ) {
+        if ( !_options.evictLeafDylibsOnOverflow ) {
+            _diagnostics.error("cache overflow by %lluMB", cacheOverflowAmount() / 1024 / 1024);
+            return;
+        }
+        size_t evictionCount = evictLeafDylibs(cacheOverflowAmount(), overflowDylibs);
+        // re-layout cache
+        for (DylibInfo& dylib : _sortedDylibs)
+            dylib.cacheLocation.clear();
+        _coalescedText.clear();
+        parseCoalescableSegments();
+        processSelectorStrings(osExecutables);
+        assignSegmentAddresses();
+
+        _diagnostics.verbose("cache overflow, evicted %lu leaf dylibs\n", evictionCount);
+    }
+    markPaddingInaccessible();
+
+     // copy all segments into cache
+    uint64_t t2 = mach_absolute_time();
+    writeCacheHeader();
+    copyRawSegments();
+
+    // rebase all dylibs for new location in cache
+    uint64_t t3 = mach_absolute_time();
+    _aslrTracker.setDataRegion(_readWriteRegion.buffer, _readWriteRegion.sizeInUse);
+    if ( !_options.cacheSupportsASLR )
+        _aslrTracker.disable();
+    adjustAllImagesForNewSegmentLocations();
+    if ( _diagnostics.hasError() )
+        return;
+
+    // build ImageArray for dyld3, which has side effect of binding all cached dylibs
+    uint64_t t4 = mach_absolute_time();
+    buildImageArray(aliases);
+    if ( _diagnostics.hasError() )
+        return;
+
+    // optimize ObjC
+    uint64_t t5 = mach_absolute_time();
+    DyldSharedCache* dyldCache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    if ( _options.optimizeObjC )
+        optimizeObjC();
+    if ( _diagnostics.hasError() )
+        return;
+
+
+    // optimize away stubs
+    uint64_t t6 = mach_absolute_time();
+    if ( _options.optimizeStubs )
+         optimizeAwayStubs();
+
+
+    // FIPS seal corecrypto, This must be done after stub elimination (so that __TEXT,__text is not changed after sealing)
+    fipsSign();
+
+    // merge and compact LINKEDIT segments
+    uint64_t t7 = mach_absolute_time();
+    optimizeLinkedit();
+
+    // copy ImageArray to end of read-only region
+    addImageArray();
+    if ( _diagnostics.hasError() )
+        return;
+    uint64_t t8 = mach_absolute_time();
+
+    // don't add dyld3 closures to simulator cache
+    if ( !dyld3::MachOFile::isSimulatorPlatform(_options.platform) ) {
+        // compute and add dlopen closures for all other dylibs
+        addOtherImageArray(otherOsDylibsInput, overflowDylibs);
+        if ( _diagnostics.hasError() )
+            return;
+
+        // compute and add launch closures to end of read-only region
+        addClosures(osExecutables);
+        if ( _diagnostics.hasError() )
+            return;
+    }
+
+    // update final readOnly region size
+    dyld_cache_mapping_info* mappings = (dyld_cache_mapping_info*)(_readExecuteRegion.buffer + dyldCache->header.mappingOffset);
+    mappings[2].size = _readOnlyRegion.sizeInUse;
+    if ( _options.excludeLocalSymbols )
+        dyldCache->header.localSymbolsOffset = _readOnlyRegion.cacheFileOffset + _readOnlyRegion.sizeInUse;
+
+    // record max slide now that final size is established
+    if ( _archLayout->sharedRegionsAreDiscontiguous ) {
+        // special case x86_64 which has three non-contiguous chunks each in their own 1GB regions
+        uint64_t maxSlide0 = 0x60000000 - _readExecuteRegion.sizeInUse; // TEXT region has 1.5GB region
+        uint64_t maxSlide1 = 0x40000000 - _readWriteRegion.sizeInUse;
+        uint64_t maxSlide2 = 0x3FE00000 - _readOnlyRegion.sizeInUse;
+        dyldCache->header.maxSlide = std::min(std::min(maxSlide0, maxSlide1), maxSlide2);
+    }
+    else {
+        // <rdar://problem/49852839> branch predictor on arm64 currently only looks at low 32-bits, so don't slide cache more than 2GB
+        if ( (_archLayout->sharedMemorySize == 0x100000000) && (_readExecuteRegion.sizeInUse < 0x80000000) )
+            dyldCache->header.maxSlide = 0x80000000 - _readExecuteRegion.sizeInUse;
+        else
+            dyldCache->header.maxSlide = (_archLayout->sharedMemoryStart + _archLayout->sharedMemorySize) - (_readOnlyRegion.unslidLoadAddress + _readOnlyRegion.sizeInUse);
+    }
+
+    // mark if any input dylibs were built with chained fixups
+    dyldCache->header.builtFromChainedFixups = _someDylibsUsedChainedFixups;
+
+    uint64_t t9 = mach_absolute_time();
+
+    // fill in slide info at start of region[2]
+    // do this last because it modifies pointers in DATA segments
+    if ( _options.cacheSupportsASLR ) {
+#if SUPPORT_ARCH_arm64e
+        if ( strcmp(_archLayout->archName, "arm64e") == 0 )
+            writeSlideInfoV3(_aslrTracker.bitmap(), _aslrTracker.dataPageCount());
+        else
+#endif
+        if ( _archLayout->is64 )
+            writeSlideInfoV2<Pointer64<LittleEndian>>(_aslrTracker.bitmap(), _aslrTracker.dataPageCount());
+#if SUPPORT_ARCH_arm64_32 || SUPPORT_ARCH_armv7k
+        else if ( _archLayout->pointerDeltaMask == 0xC0000000 )
+            writeSlideInfoV4<Pointer32<LittleEndian>>(_aslrTracker.bitmap(), _aslrTracker.dataPageCount());
+#endif
+        else
+            writeSlideInfoV2<Pointer32<LittleEndian>>(_aslrTracker.bitmap(), _aslrTracker.dataPageCount());
+    }
+
+    uint64_t t10 = mach_absolute_time();
+
+    // last sanity check on size
+    if ( cacheOverflowAmount() != 0 ) {
+        _diagnostics.error("cache overflow after optimizations 0x%llX -> 0x%llX", _readExecuteRegion.unslidLoadAddress, _readOnlyRegion.unslidLoadAddress + _readOnlyRegion.sizeInUse);
+        return;
+    }
+
+    // codesignature is part of file, but is not mapped
+    codeSign();
+    if ( _diagnostics.hasError() )
+        return;
+
+    uint64_t t11 = mach_absolute_time();
+
+    if ( _options.verbose ) {
+        fprintf(stderr, "time to layout cache: %ums\n", absolutetime_to_milliseconds(t2-t1));
+        fprintf(stderr, "time to copy cached dylibs into buffer: %ums\n", absolutetime_to_milliseconds(t3-t2));
+        fprintf(stderr, "time to adjust segments for new split locations: %ums\n", absolutetime_to_milliseconds(t4-t3));
+        fprintf(stderr, "time to bind all images: %ums\n", absolutetime_to_milliseconds(t5-t4));
+        fprintf(stderr, "time to optimize Objective-C: %ums\n", absolutetime_to_milliseconds(t6-t5));
+        fprintf(stderr, "time to do stub elimination: %ums\n", absolutetime_to_milliseconds(t7-t6));
+        fprintf(stderr, "time to optimize LINKEDITs: %ums\n", absolutetime_to_milliseconds(t8-t7));
+        fprintf(stderr, "time to build %lu closures: %ums\n", osExecutables.size(), absolutetime_to_milliseconds(t9-t8));
+        fprintf(stderr, "time to compute slide info: %ums\n", absolutetime_to_milliseconds(t10-t9));
+        fprintf(stderr, "time to compute UUID and codesign cache file: %ums\n", absolutetime_to_milliseconds(t11-t10));
+    }
+
+    return;
+}
+
+const std::set<std::string> SharedCacheBuilder::warnings()
+{
+    return _diagnostics.warnings();
+}
+
+const std::set<const dyld3::MachOAnalyzer*> SharedCacheBuilder::evictions()
+{
+    return _evictions;
+}
+
+void SharedCacheBuilder::deleteBuffer()
+{
+    // Cache buffer
+    vm_deallocate(mach_task_self(), _fullAllocatedBuffer, _archLayout->sharedMemorySize);
+    _fullAllocatedBuffer = 0;
+    _allocatedBufferSize = 0;
+    // Local symbols buffer
+    if ( _localSymbolsRegion.bufferSize != 0 ) {
+        vm_deallocate(mach_task_self(), (vm_address_t)_localSymbolsRegion.buffer, _localSymbolsRegion.bufferSize);
+        _localSymbolsRegion.buffer = 0;
+        _localSymbolsRegion.bufferSize = 0;
+    }
+    // Code singatures
+    vm_deallocate(mach_task_self(), (vm_address_t)_codeSignatureRegion.buffer, _codeSignatureRegion.bufferSize);
+    _codeSignatureRegion.buffer = 0;
+    _codeSignatureRegion.bufferSize = 0;
+}
+
+
+void SharedCacheBuilder::makeSortedDylibs(const std::vector<LoadedMachO>& dylibs, const std::unordered_map<std::string, unsigned> sortOrder)
+{
+    for (const LoadedMachO& dylib : dylibs) {
+        _sortedDylibs.push_back({ &dylib, dylib.mappedFile.runtimePath, {} });
+    }
+
+    std::sort(_sortedDylibs.begin(), _sortedDylibs.end(), [&](const DylibInfo& a, const DylibInfo& b) {
+        const auto& orderA = sortOrder.find(a.input->mappedFile.runtimePath);
+        const auto& orderB = sortOrder.find(b.input->mappedFile.runtimePath);
+        bool foundA = (orderA != sortOrder.end());
+        bool foundB = (orderB != sortOrder.end());
+
+        // Order all __DATA_DIRTY segments specified in the order file first, in
+        // the order specified in the file, followed by any other __DATA_DIRTY
+        // segments in lexicographic order.
+        if ( foundA && foundB )
+            return orderA->second < orderB->second;
+        else if ( foundA )
+            return true;
+        else if ( foundB )
+             return false;
+
+        // Sort mac before iOSMac
+        bool isIOSMacA = strncmp(a.input->mappedFile.runtimePath.c_str(), "/System/iOSSupport/", 19) == 0;
+        bool isIOSMacB = strncmp(b.input->mappedFile.runtimePath.c_str(), "/System/iOSSupport/", 19) == 0;
+        if (isIOSMacA != isIOSMacB)
+            return !isIOSMacA;
+
+        // Finally sort by path
+        return a.input->mappedFile.runtimePath < b.input->mappedFile.runtimePath;
+    });
+}
+
+struct DylibAndSize
+{
+    const CacheBuilder::LoadedMachO*    input;
+    const char*                         installName;
+    uint64_t                            size;
+};
+
+uint64_t SharedCacheBuilder::cacheOverflowAmount()
+{
+    if ( _archLayout->sharedRegionsAreDiscontiguous ) {
+        // for macOS x86_64 cache, need to check each region for overflow
+        if ( _readExecuteRegion.sizeInUse > 0x60000000 )
+            return (_readExecuteRegion.sizeInUse - 0x60000000);
+
+        if ( _readWriteRegion.sizeInUse > 0x40000000 )
+            return (_readWriteRegion.sizeInUse - 0x40000000);
+
+        if ( _readOnlyRegion.sizeInUse > 0x3FE00000 )
+            return (_readOnlyRegion.sizeInUse - 0x3FE00000);
+    }
+    else if ( _archLayout->textAndDataMaxSize != 0 ) {
+        // for armv7k, limit is 512MB of TEX+DATA
+        uint64_t totalTextAndData = _readWriteRegion.unslidLoadAddress + _readWriteRegion.sizeInUse - _readExecuteRegion.unslidLoadAddress;
+        if ( totalTextAndData < _archLayout->textAndDataMaxSize )
+            return 0;
+        else
+            return totalTextAndData - _archLayout->textAndDataMaxSize;
+    }
+    else {
+        bool alreadyOptimized = (_readOnlyRegion.sizeInUse != _readOnlyRegion.bufferSize);
+        uint64_t vmSize = _readOnlyRegion.unslidLoadAddress - _readExecuteRegion.unslidLoadAddress;
+        if ( alreadyOptimized )
+            vmSize += _readOnlyRegion.sizeInUse;
+        else if ( _options.excludeLocalSymbols )
+            vmSize += (_readOnlyRegion.sizeInUse * 37/100); // assume locals removal and LINKEDIT optimzation reduces LINKEDITs %37 of original size
+        else
+            vmSize += (_readOnlyRegion.sizeInUse * 80/100); // assume LINKEDIT optimzation reduces LINKEDITs to %80 of original size
+        if ( vmSize > _archLayout->sharedMemorySize )
+            return vmSize - _archLayout->sharedMemorySize;
+    }
+    // fits in shared region
+    return 0;
+}
+
+size_t SharedCacheBuilder::evictLeafDylibs(uint64_t reductionTarget, std::vector<const LoadedMachO*>& overflowDylibs)
+{
+    // build a reverse map of all dylib dependencies
+    __block std::map<std::string, std::set<std::string>> references;
+    std::map<std::string, std::set<std::string>>* referencesPtr = &references;
+    for (const DylibInfo& dylib : _sortedDylibs) {
+        // Esnure we have an entry (even if it is empty)
+        if (references.count(dylib.input->mappedFile.mh->installName()) == 0) {
+            references[dylib.input->mappedFile.mh->installName()] = std::set<std::string>();
+        };
+        dylib.input->mappedFile.mh->forEachDependentDylib(^(const char* loadPath, bool isWeak, bool isReExport, bool isUpward, uint32_t compatVersion, uint32_t curVersion, bool &stop) {
+            references[loadPath].insert(dylib.input->mappedFile.mh->installName());
+        });
+    }
+
+    // Find the sizes of all the dylibs
+    std::vector<DylibAndSize> dylibsToSort;
+    std::vector<DylibAndSize> sortedDylibs;
+    for (const DylibInfo& dylib : _sortedDylibs) {
+        const char* installName = dylib.input->mappedFile.mh->installName();
+        __block uint64_t segsSize = 0;
+        dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& info, bool& stop) {
+            if ( strcmp(info.segName, "__LINKEDIT") != 0 )
+                segsSize += info.vmSize;
+        });
+        dylibsToSort.push_back({ dylib.input, installName, segsSize });
+    }
+
+    // Build an ordered list of what to remove. At each step we do following
+    // 1) Find all dylibs that nothing else depends on
+    // 2a) If any of those dylibs are not in the order select the largest one of them
+    // 2b) If all the leaf dylibs are in the order file select the last dylib that appears last in the order file
+    // 3) Remove all entries to the removed file from the reverse dependency map
+    // 4) Go back to one and repeat until there are no more evictable dylibs
+    // This results in us always choosing the locally optimal selection, and then taking into account how that impacts
+    // the dependency graph for subsequent selections
+
+    bool candidateFound = true;
+    while (candidateFound) {
+        candidateFound = false;
+        DylibAndSize candidate;
+        uint64_t candidateOrder = 0;
+        for(const auto& dylib : dylibsToSort) {
+            const auto &i = referencesPtr->find(dylib.installName);
+            assert(i != referencesPtr->end());
+            if (!i->second.empty()) {
+                continue;
+            }
+            const auto& j = _options.dylibOrdering.find(dylib.input->mappedFile.runtimePath);
+            uint64_t order = 0;
+            if (j != _options.dylibOrdering.end()) {
+                order = j->second;
+            } else {
+                // Not in the order file, set order sot it goes to the front of the list
+                order = UINT64_MAX;
+            }
+            if (order > candidateOrder ||
+                (order == UINT64_MAX && candidate.size < dylib.size)) {
+                    // The new file is either a lower priority in the order file
+                    // or the same priority as the candidate but larger
+                    candidate = dylib;
+                    candidateOrder = order;
+                    candidateFound = true;
+            }
+        }
+        if (candidateFound) {
+            sortedDylibs.push_back(candidate);
+            referencesPtr->erase(candidate.installName);
+            for (auto& dependent : references) {
+                (void)dependent.second.erase(candidate.installName);
+            }
+            auto j = std::find_if(dylibsToSort.begin(), dylibsToSort.end(), [&candidate](const DylibAndSize& dylib) {
+                return (strcmp(candidate.installName, dylib.installName) == 0);
+            });
+            if (j != dylibsToSort.end()) {
+                dylibsToSort.erase(j);
+            }
+        }
+    }
+
+     // build set of dylibs that if removed will allow cache to build
+    for (DylibAndSize& dylib : sortedDylibs) {
+        if ( _options.verbose )
+            _diagnostics.warning("to prevent cache overflow, not caching %s", dylib.installName);
+        _evictions.insert(dylib.input->mappedFile.mh);
+        // Track the evicted dylibs so we can try build "other" dlopen closures for them.
+        overflowDylibs.push_back(dylib.input);
+        if ( dylib.size > reductionTarget )
+            break;
+        reductionTarget -= dylib.size;
+    }
+
+    // prune _sortedDylibs
+    _sortedDylibs.erase(std::remove_if(_sortedDylibs.begin(), _sortedDylibs.end(), [&](const DylibInfo& dylib) {
+        return (_evictions.count(dylib.input->mappedFile.mh) != 0);
+    }),_sortedDylibs.end());
+
+    return _evictions.size();
+}
+
+
+void SharedCacheBuilder::writeCacheHeader()
+{
+    // "dyld_v1" + spaces + archName(), with enough spaces to pad to 15 bytes
+    std::string magic = "dyld_v1";
+    magic.append(15 - magic.length() - strlen(_options.archs->name()), ' ');
+    magic.append(_options.archs->name());
+    assert(magic.length() == 15);
+
+    // fill in header
+    dyld_cache_header* dyldCacheHeader = (dyld_cache_header*)_readExecuteRegion.buffer;
+    memcpy(dyldCacheHeader->magic, magic.c_str(), 16);
+    dyldCacheHeader->mappingOffset        = sizeof(dyld_cache_header);
+    dyldCacheHeader->mappingCount         = 3;
+    dyldCacheHeader->imagesOffset         = (uint32_t)(dyldCacheHeader->mappingOffset + 3*sizeof(dyld_cache_mapping_info));
+    dyldCacheHeader->imagesCount          = (uint32_t)_sortedDylibs.size() + _aliasCount;
+    dyldCacheHeader->dyldBaseAddress      = 0;
+    dyldCacheHeader->codeSignatureOffset  = 0;
+    dyldCacheHeader->codeSignatureSize    = 0;
+    dyldCacheHeader->slideInfoOffset      = _slideInfoFileOffset;
+    dyldCacheHeader->slideInfoSize        = _slideInfoBufferSizeAllocated;
+    dyldCacheHeader->localSymbolsOffset   = 0;
+    dyldCacheHeader->localSymbolsSize     = 0;
+    dyldCacheHeader->cacheType            = _options.optimizeStubs ? kDyldSharedCacheTypeProduction : kDyldSharedCacheTypeDevelopment;
+    dyldCacheHeader->accelerateInfoAddr   = 0;
+    dyldCacheHeader->accelerateInfoSize   = 0;
+    bzero(dyldCacheHeader->uuid, 16);// overwritten later by recomputeCacheUUID()
+    dyldCacheHeader->branchPoolsOffset    = 0;
+    dyldCacheHeader->branchPoolsCount     = 0;
+    dyldCacheHeader->imagesTextOffset     = dyldCacheHeader->imagesOffset + sizeof(dyld_cache_image_info)*dyldCacheHeader->imagesCount;
+    dyldCacheHeader->imagesTextCount      = _sortedDylibs.size();
+    dyldCacheHeader->patchInfoAddr        = 0;
+    dyldCacheHeader->patchInfoSize        = 0;
+    dyldCacheHeader->otherImageGroupAddrUnused  = 0;
+    dyldCacheHeader->otherImageGroupSizeUnused  = 0;
+    dyldCacheHeader->progClosuresAddr     = 0;
+    dyldCacheHeader->progClosuresSize     = 0;
+    dyldCacheHeader->progClosuresTrieAddr = 0;
+    dyldCacheHeader->progClosuresTrieSize = 0;
+    dyldCacheHeader->platform             = (uint8_t)_options.platform;
+    dyldCacheHeader->formatVersion        = dyld3::closure::kFormatVersion;
+    dyldCacheHeader->dylibsExpectedOnDisk = !_options.dylibsRemovedDuringMastering;
+    dyldCacheHeader->simulator            = _options.forSimulator;
+    dyldCacheHeader->locallyBuiltCache    = _options.isLocallyBuiltCache;
+    dyldCacheHeader->builtFromChainedFixups= false;
+    dyldCacheHeader->formatVersion        = dyld3::closure::kFormatVersion;
+    dyldCacheHeader->sharedRegionStart    = _archLayout->sharedMemoryStart;
+    dyldCacheHeader->sharedRegionSize     = _archLayout->sharedMemorySize;
+
+   // fill in mappings
+    dyld_cache_mapping_info* mappings = (dyld_cache_mapping_info*)(_readExecuteRegion.buffer + dyldCacheHeader->mappingOffset);
+    mappings[0].address    = _readExecuteRegion.unslidLoadAddress;
+    mappings[0].fileOffset = 0;
+    mappings[0].size       = _readExecuteRegion.sizeInUse;
+    mappings[0].maxProt    = VM_PROT_READ | VM_PROT_EXECUTE;
+    mappings[0].initProt   = VM_PROT_READ | VM_PROT_EXECUTE;
+    mappings[1].address    = _readWriteRegion.unslidLoadAddress;
+    mappings[1].fileOffset = _readExecuteRegion.sizeInUse;
+    mappings[1].size       = _readWriteRegion.sizeInUse;
+    mappings[1].maxProt    = VM_PROT_READ | VM_PROT_WRITE;
+    mappings[1].initProt   = VM_PROT_READ | VM_PROT_WRITE;
+    mappings[2].address    = _readOnlyRegion.unslidLoadAddress;
+    mappings[2].fileOffset = _readExecuteRegion.sizeInUse + _readWriteRegion.sizeInUse;
+    mappings[2].size       = _readOnlyRegion.sizeInUse;
+    mappings[2].maxProt    = VM_PROT_READ;
+    mappings[2].initProt   = VM_PROT_READ;
+
+    // fill in image table
+    dyld_cache_image_info* images = (dyld_cache_image_info*)(_readExecuteRegion.buffer + dyldCacheHeader->imagesOffset);
+    for (const DylibInfo& dylib : _sortedDylibs) {
+        const char* installName = dylib.input->mappedFile.mh->installName();
+        images->address = dylib.cacheLocation[0].dstCacheUnslidAddress;
+        if ( _options.dylibsRemovedDuringMastering ) {
+            images->modTime = 0;
+            images->inode   = pathHash(installName);
+        }
+        else {
+            images->modTime = dylib.input->mappedFile.modTime;
+            images->inode   = dylib.input->mappedFile.inode;
+        }
+        uint32_t installNameOffsetInTEXT =  (uint32_t)(installName - (char*)dylib.input->mappedFile.mh);
+        images->pathFileOffset = (uint32_t)dylib.cacheLocation[0].dstCacheFileOffset + installNameOffsetInTEXT;
+        ++images;
+    }
+    // append aliases image records and strings
+/*
+    for (auto &dylib : _dylibs) {
+        if (!dylib->installNameAliases.empty()) {
+            for (const std::string& alias : dylib->installNameAliases) {
+                images->set_address(_segmentMap[dylib][0].address);
+                if (_manifest.platform() == "osx") {
+                    images->modTime = dylib->lastModTime;
+                    images->inode = dylib->inode;
+                }
+                else {
+                    images->modTime = 0;
+                    images->inode = pathHash(alias.c_str());
+                }
+                images->pathFileOffset = offset;
+                //fprintf(stderr, "adding alias %s for %s\n", alias.c_str(), dylib->installName.c_str());
+                ::strcpy((char*)&_buffer[offset], alias.c_str());
+                offset += alias.size() + 1;
+                ++images;
+            }
+        }
+    }
+*/
+    // calculate start of text image array and trailing string pool
+    dyld_cache_image_text_info* textImages = (dyld_cache_image_text_info*)(_readExecuteRegion.buffer + dyldCacheHeader->imagesTextOffset);
+    uint32_t stringOffset = (uint32_t)(dyldCacheHeader->imagesTextOffset + sizeof(dyld_cache_image_text_info) * _sortedDylibs.size());
+
+    // write text image array and image names pool at same time
+    for (const DylibInfo& dylib : _sortedDylibs) {
+        dylib.input->mappedFile.mh->getUuid(textImages->uuid);
+        textImages->loadAddress     = dylib.cacheLocation[0].dstCacheUnslidAddress;
+        textImages->textSegmentSize = (uint32_t)dylib.cacheLocation[0].dstCacheSegmentSize;
+        textImages->pathOffset      = stringOffset;
+        const char* installName = dylib.input->mappedFile.mh->installName();
+        ::strcpy((char*)_readExecuteRegion.buffer + stringOffset, installName);
+        stringOffset += (uint32_t)strlen(installName)+1;
+        ++textImages;
+    }
+
+    // make sure header did not overflow into first mapped image
+    const dyld_cache_image_info* firstImage = (dyld_cache_image_info*)(_readExecuteRegion.buffer + dyldCacheHeader->imagesOffset);
+    assert(stringOffset <= (firstImage->address - mappings[0].address));
+}
+
+void SharedCacheBuilder::processSelectorStrings(const std::vector<LoadedMachO>& executables) {
+    const bool log = false;
+
+    _selectorStringsFromExecutables = 0;
+
+    // Don't do this optimisation on watchOS where the shared cache is too small
+    if (_options.platform == dyld3::Platform::watchOS)
+        return;
+
+    // Get the method name coalesced section as that is where we need to put these strings
+    CacheBuilder::CacheCoalescedText::StringSection& cacheStringSection = _coalescedText.getSectionData("__objc_methname");
+    for (const LoadedMachO& executable : executables) {
+        const dyld3::MachOAnalyzer* ma = (const dyld3::MachOAnalyzer*)executable.loadedFileInfo.fileContent;
+
+        uint64_t sizeBeforeProcessing = cacheStringSection.bufferSize;
+
+        ma->forEachObjCMethodName(^(const char *methodName) {
+            std::string_view str = methodName;
+            auto itAndInserted = cacheStringSection.stringsToOffsets.insert({ str, cacheStringSection.bufferSize });
+            if (itAndInserted.second) {
+                // If we inserted the string then we need to include it in the total
+                cacheStringSection.bufferSize += str.size() + 1;
+                // if (log) printf("Selector: %s -> %s\n", ma->installName(), methodName);
+                ++_selectorStringsFromExecutables;
+            }
+        });
+
+        uint64_t sizeAfterProcessing = cacheStringSection.bufferSize;
+        if ( log && (sizeBeforeProcessing != sizeAfterProcessing) ) {
+            printf("Pulled in % 6lld bytes of selectors from %s\n",
+                   sizeAfterProcessing - sizeBeforeProcessing, executable.loadedFileInfo.path);
+        }
+    }
+
+    _diagnostics.verbose("Pulled in %lld selector strings from executables\n", _selectorStringsFromExecutables);
+}
+
+void SharedCacheBuilder::parseCoalescableSegments() {
+    const bool log = false;
+
+    for (DylibInfo& dylib : _sortedDylibs)
+        _coalescedText.parseCoalescableText(dylib.input->mappedFile.mh, dylib.textCoalescer);
+
+    if (log) {
+        for (const char* section : CacheCoalescedText::SupportedSections) {
+            CacheCoalescedText::StringSection& sectionData = _coalescedText.getSectionData(section);
+            printf("Coalesced %s from % 10lld -> % 10d, saving % 10lld bytes\n", section,
+                   sectionData.bufferSize + sectionData.savedSpace, sectionData.bufferSize, sectionData.savedSpace);
+        }
+    }
+}
+
+void SharedCacheBuilder::assignSegmentAddresses()
+{
+    // calculate size of header info and where first dylib's mach_header should start
+    size_t startOffset = sizeof(dyld_cache_header) + 3*sizeof(dyld_cache_mapping_info);
+    startOffset += sizeof(dyld_cache_image_info) * _sortedDylibs.size();
+    startOffset += sizeof(dyld_cache_image_text_info) * _sortedDylibs.size();
+    for (const DylibInfo& dylib : _sortedDylibs) {
+        startOffset += (strlen(dylib.input->mappedFile.mh->installName()) + 1);
+    }
+    //fprintf(stderr, "%s total header size = 0x%08lX\n", _options.archName.c_str(), startOffset);
+    startOffset = align(startOffset, 12);
+
+    // HACK!: Rebase v4 assumes that values below 0x8000 are not pointers (encoding as offsets from the cache header).
+    // If using a minimal cache, we need to pad out the cache header to make sure a pointer doesn't fall within that range
+#if SUPPORT_ARCH_arm64_32 || SUPPORT_ARCH_armv7k
+    if ( _options.cacheSupportsASLR && !_archLayout->is64 ) {
+        if ( _archLayout->pointerDeltaMask == 0xC0000000 )
+            startOffset = std::max(startOffset, (size_t)0x8000);
+    }
+#endif
+
+    // assign TEXT segment addresses
+    _readExecuteRegion.buffer               = (uint8_t*)_fullAllocatedBuffer;
+    _readExecuteRegion.bufferSize           = 0;
+    _readExecuteRegion.sizeInUse            = 0;
+    _readExecuteRegion.unslidLoadAddress    = _archLayout->sharedMemoryStart;
+    _readExecuteRegion.cacheFileOffset      = 0;
+    __block uint64_t addr = _readExecuteRegion.unslidLoadAddress + startOffset; // header
+    for (DylibInfo& dylib : _sortedDylibs) {
+        __block uint64_t textSegVmAddr = 0;
+        dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+            if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                textSegVmAddr = segInfo.vmAddr;
+            if ( segInfo.protections != (VM_PROT_READ | VM_PROT_EXECUTE) )
+                return;
+            // We may have coalesced the sections at the end of this segment.  In that case, shrink the segment to remove them.
+            __block size_t sizeOfSections = 0;
+            __block bool foundCoalescedSection = false;
+            dylib.input->mappedFile.mh->forEachSection(^(const dyld3::MachOAnalyzer::SectionInfo &sectInfo, bool malformedSectionRange, bool &stopSection) {
+                if (strcmp(sectInfo.segInfo.segName, segInfo.segName) != 0)
+                    return;
+                if ( dylib.textCoalescer.sectionWasCoalesced(sectInfo.sectName)) {
+                    foundCoalescedSection = true;
+                } else {
+                    sizeOfSections = sectInfo.sectAddr + sectInfo.sectSize - segInfo.vmAddr;
+                }
+            });
+            if (!foundCoalescedSection)
+                sizeOfSections = segInfo.sizeOfSections;
+
+            // Keep __TEXT segments 4K or more aligned
+            addr = align(addr, std::max((int)segInfo.p2align, (int)12));
+            uint64_t offsetInRegion = addr - _readExecuteRegion.unslidLoadAddress;
+            SegmentMappingInfo loc;
+            loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+            loc.segName                = segInfo.segName;
+            loc.dstSegment             = _readExecuteRegion.buffer + offsetInRegion;
+            loc.dstCacheUnslidAddress  = addr;
+            loc.dstCacheFileOffset     = (uint32_t)offsetInRegion;
+            loc.dstCacheSegmentSize    = (uint32_t)align(sizeOfSections, 12);
+            loc.dstCacheFileSize       = (uint32_t)align(sizeOfSections, 12);
+            loc.copySegmentSize        = (uint32_t)sizeOfSections;
+            loc.srcSegmentIndex        = segInfo.segIndex;
+            dylib.cacheLocation.push_back(loc);
+            addr += loc.dstCacheSegmentSize;
+        });
+    }
+    // move read-only segments to end of TEXT
+    if ( _archLayout->textAndDataMaxSize != 0 ) {
+       for (DylibInfo& dylib : _sortedDylibs) {
+            __block uint64_t textSegVmAddr = 0;
+            dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+                if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                    textSegVmAddr = segInfo.vmAddr;
+                if ( segInfo.protections != VM_PROT_READ )
+                    return;
+                if ( strcmp(segInfo.segName, "__LINKEDIT") == 0 )
+                    return;
+
+                // Keep segments segments 4K or more aligned
+                addr = align(addr, std::max((int)segInfo.p2align, (int)12));
+                uint64_t offsetInRegion = addr - _readExecuteRegion.unslidLoadAddress;
+                SegmentMappingInfo loc;
+                loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+                loc.segName                = segInfo.segName;
+                loc.dstSegment             = _readExecuteRegion.buffer + offsetInRegion;
+                loc.dstCacheUnslidAddress  = addr;
+                loc.dstCacheFileOffset     = (uint32_t)(_readExecuteRegion.cacheFileOffset + offsetInRegion);
+                loc.dstCacheSegmentSize    = (uint32_t)align(segInfo.sizeOfSections, 12);
+                loc.dstCacheFileSize       = (uint32_t)segInfo.sizeOfSections;
+                loc.copySegmentSize        = (uint32_t)segInfo.sizeOfSections;
+                loc.srcSegmentIndex        = segInfo.segIndex;
+                dylib.cacheLocation.push_back(loc);
+                addr += loc.dstCacheSegmentSize;
+            });
+        }
+    }
+
+    // reserve space for objc optimization tables and deduped strings
+    uint64_t objcReadOnlyBufferVMAddr = addr;
+    _objcReadOnlyBuffer = _readExecuteRegion.buffer + (addr - _readExecuteRegion.unslidLoadAddress);
+
+    // First the strings as we'll fill in the objc tables later in the optimizer
+    for (const char* section: CacheCoalescedText::SupportedSections) {
+        CacheCoalescedText::StringSection& cacheStringSection = _coalescedText.getSectionData(section);
+        cacheStringSection.bufferAddr = _readExecuteRegion.buffer + (addr - _readExecuteRegion.unslidLoadAddress);
+        cacheStringSection.bufferVMAddr = addr;
+        addr += cacheStringSection.bufferSize;
+    }
+
+    addr = align(addr, 14);
+    _objcReadOnlyBufferSizeUsed = addr - objcReadOnlyBufferVMAddr;
+
+    uint32_t totalSelectorRefCount = (uint32_t)_selectorStringsFromExecutables;
+    uint32_t totalClassDefCount    = 0;
+    uint32_t totalProtocolDefCount = 0;
+    for (DylibInfo& dylib : _sortedDylibs) {
+        dyld3::MachOAnalyzer::ObjCInfo info = dylib.input->mappedFile.mh->getObjCInfo();
+        totalSelectorRefCount   += info.selRefCount;
+        totalClassDefCount      += info.classDefCount;
+        totalProtocolDefCount   += info.protocolDefCount;
+    }
+
+    // now that shared cache coalesces all selector strings, use that better count
+    uint32_t coalescedSelectorCount = (uint32_t)_coalescedText.objcMethNames.stringsToOffsets.size();
+    if ( coalescedSelectorCount > totalSelectorRefCount )
+        totalSelectorRefCount = coalescedSelectorCount;
+    addr += align(computeReadOnlyObjC(totalSelectorRefCount, totalClassDefCount, totalProtocolDefCount), 14);
+    _objcReadOnlyBufferSizeAllocated = addr - objcReadOnlyBufferVMAddr;
+
+
+    // align TEXT region end
+    uint64_t endTextAddress = align(addr, _archLayout->sharedRegionAlignP2);
+    _readExecuteRegion.bufferSize = endTextAddress - _readExecuteRegion.unslidLoadAddress;
+    _readExecuteRegion.sizeInUse  = _readExecuteRegion.bufferSize;
+
+
+    // assign __DATA* addresses
+    if ( _archLayout->sharedRegionsAreDiscontiguous )
+        addr = _archLayout->sharedMemoryStart + 0x60000000;
+    else
+        addr = align((addr + _archLayout->sharedRegionPadding), _archLayout->sharedRegionAlignP2);
+    _readWriteRegion.buffer               = (uint8_t*)_fullAllocatedBuffer + addr - _archLayout->sharedMemoryStart;
+    _readWriteRegion.bufferSize           = 0;
+    _readWriteRegion.sizeInUse            = 0;
+    _readWriteRegion.unslidLoadAddress    = addr;
+    _readWriteRegion.cacheFileOffset      = _readExecuteRegion.sizeInUse;
+
+    // layout all __DATA_CONST segments
+    __block int dataConstSegmentCount = 0;
+    for (DylibInfo& dylib : _sortedDylibs) {
+        __block uint64_t textSegVmAddr = 0;
+       dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+           if ( _options.platform == dyld3::Platform::watchOS_simulator)
+               return;
+            if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                textSegVmAddr = segInfo.vmAddr;
+            if ( segInfo.protections != (VM_PROT_READ | VM_PROT_WRITE) )
+                return;
+            if ( strcmp(segInfo.segName, "__DATA_CONST") != 0 )
+                return;
+            ++dataConstSegmentCount;
+            // Pack __DATA_CONST segments
+            addr = align(addr, segInfo.p2align);
+            size_t copySize = std::min((size_t)segInfo.fileSize, (size_t)segInfo.sizeOfSections);
+            uint64_t offsetInRegion = addr - _readWriteRegion.unslidLoadAddress;
+            SegmentMappingInfo loc;
+            loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+            loc.segName                = segInfo.segName;
+            loc.dstSegment             = _readWriteRegion.buffer + offsetInRegion;
+            loc.dstCacheUnslidAddress  = addr;
+            loc.dstCacheFileOffset     = (uint32_t)(_readWriteRegion.cacheFileOffset + offsetInRegion);
+            loc.dstCacheSegmentSize    = (uint32_t)segInfo.sizeOfSections;
+            loc.dstCacheFileSize       = (uint32_t)copySize;
+            loc.copySegmentSize        = (uint32_t)copySize;
+            loc.srcSegmentIndex        = segInfo.segIndex;
+            dylib.cacheLocation.push_back(loc);
+            addr += loc.dstCacheSegmentSize;
+        });
+    }
+
+    // layout all __DATA segments (and other r/w non-dirty, non-const) segments
+    for (DylibInfo& dylib : _sortedDylibs) {
+        __block uint64_t textSegVmAddr = 0;
+       dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+            if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                textSegVmAddr = segInfo.vmAddr;
+            if ( segInfo.protections != (VM_PROT_READ | VM_PROT_WRITE) )
+                return;
+            if ( _options.platform != dyld3::Platform::watchOS_simulator) {
+                if ( strcmp(segInfo.segName, "__DATA_CONST") == 0 )
+                    return;
+                if ( strcmp(segInfo.segName, "__DATA_DIRTY") == 0 )
+                    return;
+            }
+            bool forcePageAlignedData = false;
+            if (_options.platform == dyld3::Platform::macOS) {
+                forcePageAlignedData = dylib.input->mappedFile.mh->hasUnalignedPointerFixups();
+                //if ( forcePageAlignedData )
+                //    warning("unaligned pointer in %s\n", dylib.input->mappedFile.runtimePath.c_str());
+            }
+            if ( (dataConstSegmentCount > 10) && !forcePageAlignedData ) {
+                // Pack __DATA segments only if we also have __DATA_CONST segments
+                addr = align(addr, segInfo.p2align);
+            }
+            else {
+                // Keep __DATA segments 4K or more aligned
+                addr = align(addr, std::max((int)segInfo.p2align, (int)12));
+            }
+            size_t copySize = std::min((size_t)segInfo.fileSize, (size_t)segInfo.sizeOfSections);
+            uint64_t offsetInRegion = addr - _readWriteRegion.unslidLoadAddress;
+            SegmentMappingInfo loc;
+            loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+            loc.segName                = segInfo.segName;
+            loc.dstSegment             = _readWriteRegion.buffer + offsetInRegion;
+            loc.dstCacheUnslidAddress  = addr;
+            loc.dstCacheFileOffset     = (uint32_t)(_readWriteRegion.cacheFileOffset + offsetInRegion);
+            loc.dstCacheSegmentSize    = (uint32_t)segInfo.sizeOfSections;
+            loc.dstCacheFileSize       = (uint32_t)copySize;
+            loc.copySegmentSize        = (uint32_t)copySize;
+            loc.srcSegmentIndex        = segInfo.segIndex;
+            dylib.cacheLocation.push_back(loc);
+            addr += loc.dstCacheSegmentSize;
+        });
+    }
+
+    // layout all __DATA_DIRTY segments, sorted (FIXME)
+    const size_t dylibCount = _sortedDylibs.size();
+    uint32_t dirtyDataSortIndexes[dylibCount];
+    for (size_t i=0; i < dylibCount; ++i)
+        dirtyDataSortIndexes[i] = (uint32_t)i;
+    std::sort(&dirtyDataSortIndexes[0], &dirtyDataSortIndexes[dylibCount], [&](const uint32_t& a, const uint32_t& b) {
+        const auto& orderA = _options.dirtyDataSegmentOrdering.find(_sortedDylibs[a].input->mappedFile.runtimePath);
+        const auto& orderB = _options.dirtyDataSegmentOrdering.find(_sortedDylibs[b].input->mappedFile.runtimePath);
+        bool foundA = (orderA != _options.dirtyDataSegmentOrdering.end());
+        bool foundB = (orderB != _options.dirtyDataSegmentOrdering.end());
+
+        // Order all __DATA_DIRTY segments specified in the order file first, in the order specified in the file,
+        // followed by any other __DATA_DIRTY segments in lexicographic order.
+        if ( foundA && foundB )
+            return orderA->second < orderB->second;
+        else if ( foundA )
+            return true;
+        else if ( foundB )
+             return false;
+        else
+             return _sortedDylibs[a].input->mappedFile.runtimePath < _sortedDylibs[b].input->mappedFile.runtimePath;
+    });
+    addr = align(addr, 12);
+    for (size_t i=0; i < dylibCount; ++i) {
+        DylibInfo& dylib  = _sortedDylibs[dirtyDataSortIndexes[i]];
+        __block uint64_t textSegVmAddr = 0;
+        dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+            if ( _options.platform == dyld3::Platform::watchOS_simulator)
+                return;
+            if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                textSegVmAddr = segInfo.vmAddr;
+            if ( segInfo.protections != (VM_PROT_READ | VM_PROT_WRITE) )
+                return;
+            if ( strcmp(segInfo.segName, "__DATA_DIRTY") != 0 )
+                return;
+            // Pack __DATA_DIRTY segments
+            addr = align(addr, segInfo.p2align);
+            size_t copySize = std::min((size_t)segInfo.fileSize, (size_t)segInfo.sizeOfSections);
+            uint64_t offsetInRegion = addr - _readWriteRegion.unslidLoadAddress;
+            SegmentMappingInfo loc;
+            loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+            loc.segName                = segInfo.segName;
+            loc.dstSegment             = _readWriteRegion.buffer + offsetInRegion;
+            loc.dstCacheUnslidAddress  = addr;
+            loc.dstCacheFileOffset     = (uint32_t)(_readWriteRegion.cacheFileOffset + offsetInRegion);
+            loc.dstCacheSegmentSize    = (uint32_t)segInfo.sizeOfSections;
+            loc.dstCacheFileSize       = (uint32_t)copySize;
+            loc.copySegmentSize        = (uint32_t)copySize;
+            loc.srcSegmentIndex        = segInfo.segIndex;
+            dylib.cacheLocation.push_back(loc);
+            addr += loc.dstCacheSegmentSize;
+        });
+    }
+
+    // reserve space for objc r/w optimization tables
+    _objcReadWriteBufferSizeAllocated = align(computeReadWriteObjC((uint32_t)_sortedDylibs.size(), totalProtocolDefCount), 14);
+    addr = align(addr, 4); // objc r/w section contains pointer and must be at least pointer align
+    _objcReadWriteBuffer = _readWriteRegion.buffer + (addr - _readWriteRegion.unslidLoadAddress);
+    addr += _objcReadWriteBufferSizeAllocated;
+
+    // align DATA region end
+    uint64_t endDataAddress = align(addr, _archLayout->sharedRegionAlignP2);
+    _readWriteRegion.bufferSize   = endDataAddress - _readWriteRegion.unslidLoadAddress;
+    _readWriteRegion.sizeInUse    = _readWriteRegion.bufferSize;
+
+    // start read-only region
+    if ( _archLayout->sharedRegionsAreDiscontiguous )
+        addr = _archLayout->sharedMemoryStart + 0xA0000000;
+    else
+        addr = align((addr + _archLayout->sharedRegionPadding), _archLayout->sharedRegionAlignP2);
+    _readOnlyRegion.buffer               = (uint8_t*)_fullAllocatedBuffer + addr - _archLayout->sharedMemoryStart;
+    _readOnlyRegion.bufferSize           = 0;
+    _readOnlyRegion.sizeInUse            = 0;
+    _readOnlyRegion.unslidLoadAddress    = addr;
+    _readOnlyRegion.cacheFileOffset      = _readWriteRegion.cacheFileOffset + _readWriteRegion.sizeInUse;
+
+    // reserve space for kernel ASLR slide info at start of r/o region
+    if ( _options.cacheSupportsASLR ) {
+        size_t slideInfoSize = sizeof(dyld_cache_slide_info);
+        slideInfoSize = std::max(slideInfoSize, sizeof(dyld_cache_slide_info2));
+        slideInfoSize = std::max(slideInfoSize, sizeof(dyld_cache_slide_info3));
+        slideInfoSize = std::max(slideInfoSize, sizeof(dyld_cache_slide_info4));
+        _slideInfoBufferSizeAllocated = align(slideInfoSize + (_readWriteRegion.sizeInUse/4096) * _archLayout->slideInfoBytesPerPage + 0x4000, _archLayout->sharedRegionAlignP2);
+        _slideInfoFileOffset = _readOnlyRegion.cacheFileOffset;
+        addr += _slideInfoBufferSizeAllocated;
+    }
+
+    // layout all read-only (but not LINKEDIT) segments
+    if ( _archLayout->textAndDataMaxSize == 0 ) {
+        for (DylibInfo& dylib : _sortedDylibs) {
+            __block uint64_t textSegVmAddr = 0;
+            dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+                if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                    textSegVmAddr = segInfo.vmAddr;
+                if ( segInfo.protections != VM_PROT_READ )
+                    return;
+                if ( strcmp(segInfo.segName, "__LINKEDIT") == 0 )
+                    return;
+
+                // Keep segments segments 4K or more aligned
+                addr = align(addr, std::max((int)segInfo.p2align, (int)12));
+                uint64_t offsetInRegion = addr - _readOnlyRegion.unslidLoadAddress;
+                SegmentMappingInfo loc;
+                loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+                loc.segName                = segInfo.segName;
+                loc.dstSegment             = _readOnlyRegion.buffer + offsetInRegion;
+                loc.dstCacheUnslidAddress  = addr;
+                loc.dstCacheFileOffset     = (uint32_t)(_readOnlyRegion.cacheFileOffset + offsetInRegion);
+                loc.dstCacheSegmentSize    = (uint32_t)align(segInfo.sizeOfSections, 12);
+                loc.dstCacheFileSize       = (uint32_t)segInfo.sizeOfSections;
+                loc.copySegmentSize        = (uint32_t)segInfo.sizeOfSections;
+                loc.srcSegmentIndex        = segInfo.segIndex;
+                dylib.cacheLocation.push_back(loc);
+                addr += loc.dstCacheSegmentSize;
+            });
+        }
+    }
+    // layout all LINKEDIT segments (after other read-only segments), aligned to 16KB
+    addr = align(addr, 14);
+    _nonLinkEditReadOnlySize =  addr - _readOnlyRegion.unslidLoadAddress;
+    for (DylibInfo& dylib : _sortedDylibs) {
+        __block uint64_t textSegVmAddr = 0;
+        dylib.input->mappedFile.mh->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& segInfo, bool& stop) {
+            if ( strcmp(segInfo.segName, "__TEXT") == 0 )
+                textSegVmAddr = segInfo.vmAddr;
+            if ( segInfo.protections != VM_PROT_READ )
+                return;
+            if ( strcmp(segInfo.segName, "__LINKEDIT") != 0 )
+                return;
+            // Keep segments segments 4K or more aligned
+            addr = align(addr, std::max((int)segInfo.p2align, (int)12));
+            size_t copySize = std::min((size_t)segInfo.fileSize, (size_t)segInfo.sizeOfSections);
+            uint64_t offsetInRegion = addr - _readOnlyRegion.unslidLoadAddress;
+            SegmentMappingInfo loc;
+            loc.srcSegment             = (uint8_t*)dylib.input->mappedFile.mh + segInfo.vmAddr - textSegVmAddr;
+            loc.segName                = segInfo.segName;
+            loc.dstSegment             = _readOnlyRegion.buffer + offsetInRegion;
+            loc.dstCacheUnslidAddress  = addr;
+            loc.dstCacheFileOffset     = (uint32_t)(_readOnlyRegion.cacheFileOffset + offsetInRegion);
+            loc.dstCacheSegmentSize    = (uint32_t)align(segInfo.sizeOfSections, 12);
+            loc.dstCacheFileSize       = (uint32_t)copySize;
+            loc.copySegmentSize        = (uint32_t)copySize;
+            loc.srcSegmentIndex        = segInfo.segIndex;
+            dylib.cacheLocation.push_back(loc);
+            addr += loc.dstCacheSegmentSize;
+        });
+    }
+
+    // align r/o region end
+    uint64_t endReadOnlyAddress = align(addr, _archLayout->sharedRegionAlignP2);
+    _readOnlyRegion.bufferSize  = endReadOnlyAddress - _readOnlyRegion.unslidLoadAddress;
+    _readOnlyRegion.sizeInUse   = _readOnlyRegion.bufferSize;
+
+    //fprintf(stderr, "RX region=%p -> %p, logical addr=0x%llX\n", _readExecuteRegion.buffer, _readExecuteRegion.buffer+_readExecuteRegion.bufferSize, _readExecuteRegion.unslidLoadAddress);
+    //fprintf(stderr, "RW region=%p -> %p, logical addr=0x%llX\n", _readWriteRegion.buffer,   _readWriteRegion.buffer+_readWriteRegion.bufferSize, _readWriteRegion.unslidLoadAddress);
+    //fprintf(stderr, "RO region=%p -> %p, logical addr=0x%llX\n", _readOnlyRegion.buffer,    _readOnlyRegion.buffer+_readOnlyRegion.bufferSize, _readOnlyRegion.unslidLoadAddress);
+
+    // sort SegmentMappingInfo for each image to be in the same order as original segments
+    for (DylibInfo& dylib : _sortedDylibs) {
+        std::sort(dylib.cacheLocation.begin(), dylib.cacheLocation.end(), [&](const SegmentMappingInfo& a, const SegmentMappingInfo& b) {
+            return a.srcSegmentIndex < b.srcSegmentIndex;
+        });
+    }
+}
+
+static dyld_cache_patchable_location makePatchLocation(size_t cacheOff, uint64_t ad) {
+    int64_t signedAddend = (int64_t)ad;
+    assert(((signedAddend << 52) >> 52) == signedAddend);
+    dyld_cache_patchable_location patch;
+    patch.cacheOffset           = cacheOff;
+    patch.addend                = ad;
+    patch.authenticated         = 0;
+    patch.usesAddressDiversity  = 0;
+    patch.key                   = 0;
+    patch.discriminator         = 0;
+    return patch;
+}
+
+static dyld_cache_patchable_location makePatchLocation(size_t cacheOff, uint64_t ad,
+                                                       dyld3::MachOLoaded::ChainedFixupPointerOnDisk authInfo) {
+    int64_t signedAddend = (int64_t)ad;
+    assert(((signedAddend << 52) >> 52) == signedAddend);
+    dyld_cache_patchable_location patch;
+    patch.cacheOffset           = cacheOff;
+    patch.addend                = ad;
+    patch.authenticated         = authInfo.arm64e.authBind.auth;
+    patch.usesAddressDiversity  = authInfo.arm64e.authBind.addrDiv;
+    patch.key                   = authInfo.arm64e.authBind.key;
+    patch.discriminator         = authInfo.arm64e.authBind.diversity;
+    return patch;
+}
+
+void SharedCacheBuilder::buildImageArray(std::vector<DyldSharedCache::FileAlias>& aliases)
+{
+    typedef dyld3::closure::ClosureBuilder::CachedDylibInfo         CachedDylibInfo;
+
+    // convert STL data structures to simple arrays to passe to makeDyldCacheImageArray()
+    __block std::vector<CachedDylibInfo> dylibInfos;
+    __block std::unordered_map<dyld3::closure::ImageNum, const dyld3::MachOLoaded*> imageNumToML;
+    DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    cache->forEachImage(^(const mach_header* mh, const char* installName) {
+        uint64_t mtime;
+        uint64_t inode;
+        cache->getIndexedImageEntry((uint32_t)dylibInfos.size(), mtime, inode);
+        CachedDylibInfo entry;
+        entry.fileInfo.fileContent  = mh;
+        entry.fileInfo.path         = installName;
+        entry.fileInfo.sliceOffset  = 0;
+        entry.fileInfo.inode        = inode;
+        entry.fileInfo.mtime        = mtime;
+        dylibInfos.push_back(entry);
+        imageNumToML[(dyld3::closure::ImageNum)(dylibInfos.size())] = (dyld3::MachOLoaded*)mh;
+    });
+
+    // Convert symlinks from STL to simple char pointers.
+    std::vector<dyld3::closure::ClosureBuilder::CachedDylibAlias> dylibAliases;
+    dylibAliases.reserve(aliases.size());
+    for (const auto& alias : aliases)
+        dylibAliases.push_back({ alias.realPath.c_str(), alias.aliasPath.c_str() });
+
+    dyld3::closure::ClosureBuilder::CacheDylibsBindingHandlers handlers;
+
+    handlers.chainedBind = ^(dyld3::closure::ImageNum, const dyld3::MachOLoaded* imageLoadAddress,
+                             const  dyld_chained_starts_in_image* starts,
+                             const dyld3::Array<dyld3::closure::Image::ResolvedSymbolTarget>& targets,
+                             const dyld3::Array<dyld3::closure::ClosureBuilder::ResolvedTargetInfo>& targetInfos) {
+        _someDylibsUsedChainedFixups = true;
+        imageLoadAddress->forEachFixupInAllChains(_diagnostics, starts, true, ^(dyld3::MachOLoaded::ChainedFixupPointerOnDisk* fixupLoc, const dyld_chained_starts_in_segment* segInfo, bool& stop) {
+            uint64_t                                                  offsetInCache;
+            dyld3::closure::Image::ResolvedSymbolTarget               target;
+            const dyld3::closure::ClosureBuilder::ResolvedTargetInfo* targetInfo;
+            dyld3::MachOLoaded::ChainedFixupPointerOnDisk             orgValue;
+            uint64_t                                                  targetVMAddr;
+            uint64_t                                                  addend;
+            bool                                                      pointerIntoCache = true;
+            switch (segInfo->pointer_format) {
+                case DYLD_CHAINED_PTR_ARM64E:
+                    orgValue = *fixupLoc;
+                    if ( fixupLoc->arm64e.bind.bind ) {
+                        target     = targets[fixupLoc->arm64e.bind.ordinal];
+                        targetInfo = &targetInfos[fixupLoc->arm64e.bind.ordinal];
+                        switch ( target.sharedCache.kind ) {
+                            case dyld3::closure::Image::ResolvedSymbolTarget::kindSharedCache:
+                                offsetInCache = target.sharedCache.offset - targetInfo->addend;
+                                _dylibToItsExports[targetInfo->foundInDylib].insert(offsetInCache);
+                                _exportsToName[offsetInCache] = targetInfo->foundSymbolName;
+                                addend = targetInfo->addend;
+                                {
+                                    uint16_t top16 = addend >> 48;
+                                    uint8_t  top8  = addend >> 56;
+                                    // if top byte is non-zero and this is not a negative addend, pull off high8
+                                    if ( (top16 != 0xFFFF) && (top8 != 0) ) {
+                                        _aslrTracker.setHigh8(fixupLoc, top8);
+                                        addend &= 0x00FFFFFFFFFFFFFFULL;
+                                    }
+                                }
+                                targetVMAddr = _archLayout->sharedMemoryStart + target.sharedCache.offset;
+                                if ( fixupLoc->arm64e.authBind.auth ) {
+                                    // store auth data in side table
+                                    _aslrTracker.setAuthData(fixupLoc, fixupLoc->arm64e.authBind.diversity, fixupLoc->arm64e.authBind.addrDiv, fixupLoc->arm64e.authBind.key);
+                                    _exportsToUses[offsetInCache].push_back(makePatchLocation((uint8_t*)fixupLoc - _readExecuteRegion.buffer, addend, *fixupLoc));
+                                }
+                                else {
+                                    _exportsToUses[offsetInCache].push_back(makePatchLocation((uint8_t*)fixupLoc - _readExecuteRegion.buffer, addend));
+                                    // plain binds can have addend in chain
+                                    targetVMAddr += fixupLoc->arm64e.bind.addend;
+                                }
+                                // change location from a chain ptr into a raw pointer to the target vmaddr
+                                fixupLoc->raw64 = targetVMAddr;
+                                _aslrTracker.add(fixupLoc);
+                                break;
+                            case dyld3::closure::Image::ResolvedSymbolTarget::kindAbsolute:
+                                fixupLoc->raw64 = target.absolute.value;
+                                pointerIntoCache = false;
+                                // don't record absolute targets for ASLR
+                                _aslrTracker.remove(fixupLoc);
+                                if ( (targetInfo->libOrdinal > 0) && (targetInfo->libOrdinal <= (int)(imageLoadAddress->dependentDylibCount())) ) {
+                                    _missingWeakImports[fixupLoc] = imageLoadAddress->dependentDylibLoadPath(targetInfo->libOrdinal - 1);
+                                }
+                               break;
+                            default:
+                                assert(0 && "unsupported ResolvedSymbolTarget kind in dyld cache");
+                        }
+                    }
+                    else {
+                        // convert rebase chain entry to raw pointer to target vmaddr
+                        if ( fixupLoc->arm64e.rebase.auth ) {
+                            // store auth data in side table
+                            _aslrTracker.setAuthData(fixupLoc, fixupLoc->arm64e.authRebase.diversity, fixupLoc->arm64e.authRebase.addrDiv, fixupLoc->arm64e.authRebase.key);
+                            targetVMAddr = fixupLoc->arm64e.authRebase.target;
+                            fixupLoc->raw64 = targetVMAddr;
+                        }
+                        else {
+                            targetVMAddr = fixupLoc->arm64e.rebase.target;
+                            if ( targetVMAddr == CacheBuilder::kRebaseTargetInSideTableArm64e ) {
+                                // target was stored in side table
+                                _aslrTracker.hasRebaseTarget64(fixupLoc, &targetVMAddr);
+                            }
+                            // store high8 in side table
+                            if ( fixupLoc->arm64e.rebase.high8 )
+                                _aslrTracker.setHigh8(fixupLoc, fixupLoc->arm64e.rebase.high8);
+                            fixupLoc->raw64 = targetVMAddr;
+                        }
+                    }
+                    if ( pointerIntoCache ) {
+                        assert(fixupLoc->raw64 > _readExecuteRegion.unslidLoadAddress);
+                        assert(fixupLoc->raw64 < _readOnlyRegion.unslidLoadAddress+_readOnlyRegion.sizeInUse);
+                    }
+                    break;
+                case DYLD_CHAINED_PTR_64:
+                    if ( fixupLoc->generic64.bind.bind ) {
+                        target     = targets[fixupLoc->generic64.bind.ordinal];
+                        targetInfo = &targetInfos[fixupLoc->generic64.bind.ordinal];
+                        switch ( target.sharedCache.kind ) {
+                            case dyld3::closure::Image::ResolvedSymbolTarget::kindSharedCache:
+                                offsetInCache = target.sharedCache.offset - targetInfo->addend;
+                                _dylibToItsExports[targetInfo->foundInDylib].insert(offsetInCache);
+                                _exportsToName[offsetInCache] = targetInfo->foundSymbolName;
+                                addend = targetInfo->addend + fixupLoc->generic64.bind.addend;
+                                {
+                                    uint16_t top16 = addend >> 48;
+                                    uint8_t  top8  = addend >> 56;
+                                    // if top byte is non-zero and this is not a negative addend, pull off high8
+                                    if ( (top16 != 0xFFFF) && (top8 != 0) ) {
+                                        _aslrTracker.setHigh8(fixupLoc, top8);
+                                        addend &= 0x00FFFFFFFFFFFFFFULL;
+                                    }
+                                }
+                                // turn this bind into a flat vmaddr
+                                fixupLoc->raw64 = _archLayout->sharedMemoryStart + target.sharedCache.offset + fixupLoc->generic64.bind.addend;
+                                _exportsToUses[offsetInCache].push_back(makePatchLocation((uint8_t*)fixupLoc - _readExecuteRegion.buffer, addend));
+                                _aslrTracker.add(fixupLoc);
+                                break;
+                            case dyld3::closure::Image::ResolvedSymbolTarget::kindAbsolute:
+                                fixupLoc->raw64 = target.absolute.value;
+                                pointerIntoCache = false;
+                                // don't record absolute targets for ASLR
+                                _aslrTracker.remove(fixupLoc);
+                                if ( (targetInfo->libOrdinal > 0) && (targetInfo->libOrdinal <= (int)(imageLoadAddress->dependentDylibCount())) ) {
+                                    _missingWeakImports[fixupLoc] = imageLoadAddress->dependentDylibLoadPath(targetInfo->libOrdinal - 1);
+                                }
+                                break;
+                            default:
+                                assert(0 && "unsupported ResolvedSymbolTarget kind in dyld cache");
+                        }
+                    }
+                    else {
+                        // convert rebase chain entry to raw pointer to target vmaddr
+                        targetVMAddr = fixupLoc->generic64.rebase.target;
+                        if ( targetVMAddr == CacheBuilder::kRebaseTargetInSideTableArm64 ) {
+                            // target was stored in side table
+                            _aslrTracker.hasRebaseTarget64(fixupLoc, &targetVMAddr);
+                        }
+                        // store high8 in side table
+                        if ( fixupLoc->generic64.rebase.high8 )
+                            _aslrTracker.setHigh8(fixupLoc, fixupLoc->generic64.rebase.high8);
+                        fixupLoc->raw64 = targetVMAddr;
+                    }
+                    if ( pointerIntoCache ) {
+                        assert(fixupLoc->raw64 > _readExecuteRegion.unslidLoadAddress);
+                        assert(fixupLoc->raw64 < _readOnlyRegion.unslidLoadAddress+_readOnlyRegion.sizeInUse);
+                    }
+                    break;
+                case DYLD_CHAINED_PTR_32:
+                    if ( fixupLoc->generic32.bind.bind ) {
+                        // turn this bind into a flat vmaddr pointer
+                        target     = targets[fixupLoc->generic32.bind.ordinal];
+                        targetInfo = &targetInfos[fixupLoc->generic32.bind.ordinal];
+                        switch ( target.sharedCache.kind ) {
+                            case dyld3::closure::Image::ResolvedSymbolTarget::kindSharedCache:
+                                offsetInCache = target.sharedCache.offset - targetInfo->addend;
+                                _dylibToItsExports[targetInfo->foundInDylib].insert(offsetInCache);
+                                _exportsToName[offsetInCache] = targetInfo->foundSymbolName;
+                                fixupLoc->raw32 = (uint32_t)(_archLayout->sharedMemoryStart + target.sharedCache.offset + fixupLoc->generic32.bind.addend);
+                                _exportsToUses[offsetInCache].push_back(makePatchLocation((uint8_t*)fixupLoc - _readExecuteRegion.buffer, targetInfo->addend+fixupLoc->generic32.bind.addend));
+                                _aslrTracker.add(fixupLoc);
+                                break;
+                            case dyld3::closure::Image::ResolvedSymbolTarget::kindAbsolute:
+                                fixupLoc->raw32 = (uint32_t)target.absolute.value;
+                                pointerIntoCache = false;
+                                // don't record absolute targets for ASLR
+                                _aslrTracker.remove(fixupLoc);
+                                if ( (targetInfo->libOrdinal > 0) && (targetInfo->libOrdinal <= (int)(imageLoadAddress->dependentDylibCount())) ) {
+                                    _missingWeakImports[fixupLoc] = imageLoadAddress->dependentDylibLoadPath(targetInfo->libOrdinal - 1);
+                                }
+                                break;
+                            default:
+                                assert(0 && "unsupported ResolvedSymbolTarget kind in dyld cache");
+                        }
+                    }
+                    else if ( fixupLoc->generic32.rebase.target == CacheBuilder::kRebaseTargetInSideTableGeneric32 ) {
+                        uint32_t targetVmAddr;
+                        if ( _aslrTracker.hasRebaseTarget32(fixupLoc, &targetVmAddr) )
+                            fixupLoc->raw32 = targetVmAddr;
+                        else
+                            assert(0 && "missing target for rebase");
+                    }
+                    else if ( fixupLoc->generic32.rebase.target > segInfo->max_valid_pointer ) {
+                        __block const char* badImagePath = nullptr;
+                        cache->forEachImage(^(const mach_header* mh, const char* installName) {
+                            if ( mh == imageLoadAddress )
+                                badImagePath = installName;
+                        });
+                        _diagnostics.error("unexpected non-pointer in chain for image at %s", badImagePath);
+                        stop = true;
+                        pointerIntoCache = false;
+                    }
+                    if ( pointerIntoCache ) {
+                        assert(fixupLoc->raw32 > _readExecuteRegion.unslidLoadAddress);
+                        assert(fixupLoc->raw32 < _readOnlyRegion.unslidLoadAddress+_readOnlyRegion.sizeInUse);
+                    }
+                    break;
+                case DYLD_CHAINED_PTR_64_OFFSET:
+                case DYLD_CHAINED_PTR_ARM64E_OFFSET:
+                    assert(0 && "unsupported chained bind type");
+                    break;
+                default:
+                    assert(0 && "unsupported chained bind type");
+            }
+
+        });
+    };
+
+    handlers.rebase = ^(dyld3::closure::ImageNum imageNum, const dyld3::MachOLoaded* imageToFix, uint32_t runtimeOffset) {
+        // record location in aslr tracker so kernel can slide this on page-in
+        uint8_t* fixupLoc = (uint8_t*)imageToFix+runtimeOffset;
+        _aslrTracker.add(fixupLoc);
+    };
+
+    handlers.bind = ^(dyld3::closure::ImageNum imageNum, const dyld3::MachOLoaded* mh,
+                      uint32_t runtimeOffset, dyld3::closure::Image::ResolvedSymbolTarget target,
+                      const dyld3::closure::ClosureBuilder::ResolvedTargetInfo& targetInfo) {
+        uint8_t* fixupLoc = (uint8_t*)mh+runtimeOffset;
+
+        // binder is called a second time for weak_bind info, which we ignore when building cache
+        const bool weakDefUseAlreadySet = targetInfo.weakBindCoalese && _aslrTracker.has(fixupLoc);
+
+        // do actual bind that sets pointer in image to unslid target address
+        uint64_t offsetInCache;
+        switch ( target.sharedCache.kind ) {
+            case dyld3::closure::Image::ResolvedSymbolTarget::kindSharedCache:
+                offsetInCache = target.sharedCache.offset - targetInfo.addend;
+                _dylibToItsExports[targetInfo.foundInDylib].insert(offsetInCache);
+                if (targetInfo.isWeakDef)
+                    _dylibWeakExports.insert({ targetInfo.foundInDylib, offsetInCache });
+                _exportsToUses[offsetInCache].push_back(makePatchLocation(fixupLoc - _readExecuteRegion.buffer, targetInfo.addend));
+                _exportsToName[offsetInCache] = targetInfo.foundSymbolName;
+                if ( !weakDefUseAlreadySet ) {
+                    if ( _archLayout->is64 )
+                        *((uint64_t*)fixupLoc) = _archLayout->sharedMemoryStart + target.sharedCache.offset;
+                    else
+                        *((uint32_t*)fixupLoc) = (uint32_t)(_archLayout->sharedMemoryStart + target.sharedCache.offset);
+                    // record location in aslr tracker so kernel can slide this on page-in
+                    _aslrTracker.add(fixupLoc);
+                }
+                break;
+            case dyld3::closure::Image::ResolvedSymbolTarget::kindAbsolute:
+                 if ( _archLayout->is64 )
+                    *((uint64_t*)fixupLoc) = target.absolute.value;
+                else
+                    *((uint32_t*)fixupLoc) = (uint32_t)(target.absolute.value);
+                // don't record absolute targets for ASLR
+                // HACK: Split seg may have added a target.  Remove it
+                _aslrTracker.remove(fixupLoc);
+                if ( (targetInfo.libOrdinal > 0) && (targetInfo.libOrdinal <= (int)(mh->dependentDylibCount())) ) {
+                    _missingWeakImports[fixupLoc] = mh->dependentDylibLoadPath(targetInfo.libOrdinal - 1);
+                }
+                break;
+            default:
+                assert(0 && "unsupported ResolvedSymbolTarget kind in dyld cache");
+        }
+    };
+
+    // build ImageArray for all dylibs in dyld cache
+    dyld3::closure::PathOverrides pathOverrides;
+    dyld3::closure::ClosureBuilder cb(dyld3::closure::kFirstDyldCacheImageNum, _fileSystem, cache, false, *_options.archs, pathOverrides,
+                                      dyld3::closure::ClosureBuilder::AtPath::none, false, nullptr, _options.platform, &handlers);
+    dyld3::Array<CachedDylibInfo> dylibs(&dylibInfos[0], dylibInfos.size(), dylibInfos.size());
+    const dyld3::Array<dyld3::closure::ClosureBuilder::CachedDylibAlias> aliasesArray(dylibAliases.data(), dylibAliases.size(), dylibAliases.size());
+    _imageArray = cb.makeDyldCacheImageArray(_options.optimizeStubs, dylibs, aliasesArray);
+    if ( cb.diagnostics().hasError() ) {
+        _diagnostics.error("%s", cb.diagnostics().errorMessage().c_str());
+        return;
+    }
+
+}
+
+static bool operator==(const dyld_cache_patchable_location& a, const dyld_cache_patchable_location& b) {
+    return a.cacheOffset == b.cacheOffset;
+}
+
+void SharedCacheBuilder::addImageArray()
+{
+    // build trie of dylib paths
+    __block std::vector<DylibIndexTrie::Entry> dylibEntrys;
+    _imageArray->forEachImage(^(const dyld3::closure::Image* image, bool& stop) {
+        dylibEntrys.push_back(DylibIndexTrie::Entry(image->path(), DylibIndex(image->imageNum()-1)));
+        image->forEachAlias(^(const char *aliasPath, bool &innerStop) {
+            dylibEntrys.push_back(DylibIndexTrie::Entry(aliasPath, DylibIndex(image->imageNum()-1)));
+        });
+    });
+    DylibIndexTrie dylibsTrie(dylibEntrys);
+    std::vector<uint8_t> trieBytes;
+    dylibsTrie.emit(trieBytes);
+    while ( (trieBytes.size() % 4) != 0 )
+        trieBytes.push_back(0);
+
+    // build set of functions to never stub-eliminate because tools may need to override them
+    std::unordered_set<std::string> alwaysGeneratePatch;
+    for (const char* const* p=_s_neverStubEliminateSymbols; *p != nullptr; ++p)
+        alwaysGeneratePatch.insert(*p);
+
+    // Add the patches for the image array.
+    __block uint64_t numPatchImages             = _imageArray->size();
+    __block uint64_t numPatchExports            = 0;
+    __block uint64_t numPatchLocations          = 0;
+    __block uint64_t numPatchExportNameBytes    = 0;
+
+    auto needsPatch = [&](bool dylibNeedsPatching, const dyld3::MachOLoaded* mh,
+                          CacheOffset offset) -> bool {
+        if (dylibNeedsPatching)
+            return true;
+        if (_dylibWeakExports.find({ mh, offset }) != _dylibWeakExports.end())
+            return true;
+        const std::string& exportName = _exportsToName[offset];
+        return alwaysGeneratePatch.find(exportName) != alwaysGeneratePatch.end();
+    };
+
+    std::set<std::string> alwaysPatchDylibs;
+    for (const char* const* d= _s_neverStubEliminateDylibs; *d != nullptr; ++d)
+        alwaysPatchDylibs.insert(*d);
+
+    // First calculate how much space we need
+    const DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    cache->forEachImage(^(const mach_header* mh, const char* installName) {
+        const dyld3::MachOLoaded* ml = (const dyld3::MachOLoaded*)mh;
+        const std::set<CacheOffset>& dylibExports = _dylibToItsExports[ml];
+
+        // On a customer cache, only store patch locations for interposable dylibs and weak binding
+        bool dylibNeedsPatching = !_options.optimizeStubs || alwaysPatchDylibs.count(installName);
+
+        uint64_t numDylibExports = 0;
+        for (CacheOffset exportCacheOffset : dylibExports) {
+            if (!needsPatch(dylibNeedsPatching, ml, exportCacheOffset))
+                continue;
+            std::vector<dyld_cache_patchable_location>& uses = _exportsToUses[exportCacheOffset];
+            uses.erase(std::unique(uses.begin(), uses.end()), uses.end());
+            numPatchLocations += uses.size();
+
+            std::string exportName = _exportsToName[exportCacheOffset];
+            numPatchExportNameBytes += exportName.size() + 1;
+        }
+        numPatchExports += numDylibExports;
+    });
+
+    // Now reserve the space
+    __block std::vector<dyld_cache_image_patches>       patchImages;
+    __block std::vector<dyld_cache_patchable_export>    patchExports;
+    __block std::vector<dyld_cache_patchable_location>  patchLocations;
+    __block std::vector<char>                           patchExportNames;
+
+    patchImages.reserve(numPatchImages);
+    patchExports.reserve(numPatchExports);
+    patchLocations.reserve(numPatchLocations);
+    patchExportNames.reserve(numPatchExportNameBytes);
+
+    // And now fill it with the patch data
+    cache->forEachImage(^(const mach_header* mh, const char* installName) {
+        const dyld3::MachOLoaded* ml = (const dyld3::MachOLoaded*)mh;
+        const std::set<CacheOffset>& dylibExports = _dylibToItsExports[ml];
+
+        // On a customer cache, only store patch locations for interposable dylibs and weak binding
+        bool dylibNeedsPatching = !_options.optimizeStubs || alwaysPatchDylibs.count(installName);
+
+        // Add the patch image which points in to the exports
+        dyld_cache_image_patches patchImage;
+        patchImage.patchExportsStartIndex   = (uint32_t)patchExports.size();
+        patchImage.patchExportsCount        = 0;
+
+        // Then add each export which points to a list of locations and a name
+        for (CacheOffset exportCacheOffset : dylibExports) {
+            if (!needsPatch(dylibNeedsPatching, ml, exportCacheOffset))
+                continue;
+            ++patchImage.patchExportsCount;
+            std::vector<dyld_cache_patchable_location>& uses = _exportsToUses[exportCacheOffset];
+
+            dyld_cache_patchable_export cacheExport;
+            cacheExport.cacheOffsetOfImpl           = (uint32_t)exportCacheOffset;
+            cacheExport.patchLocationsStartIndex    = (uint32_t)patchLocations.size();
+            cacheExport.patchLocationsCount         = (uint32_t)uses.size();
+            cacheExport.exportNameOffset            = (uint32_t)patchExportNames.size();
+            patchExports.push_back(cacheExport);
+
+            // Now add the list of locations.
+            patchLocations.insert(patchLocations.end(), uses.begin(), uses.end());
+
+            // And add the export name
+            const std::string& exportName = _exportsToName[exportCacheOffset];
+            patchExportNames.insert(patchExportNames.end(), &exportName[0], &exportName[0] + exportName.size() + 1);
+        }
+        patchImages.push_back(patchImage);
+    });
+
+    while ( (patchExportNames.size() % 4) != 0 )
+        patchExportNames.push_back('\0');
+
+    uint64_t patchInfoSize = sizeof(dyld_cache_patch_info);
+    patchInfoSize += sizeof(dyld_cache_image_patches) * patchImages.size();
+    patchInfoSize += sizeof(dyld_cache_patchable_export) * patchExports.size();
+    patchInfoSize += sizeof(dyld_cache_patchable_location) * patchLocations.size();
+    patchInfoSize += patchExportNames.size();
+
+    // check for fit
+    uint64_t imageArraySize = _imageArray->size();
+    size_t freeSpace = _readOnlyRegion.bufferSize - _readOnlyRegion.sizeInUse;
+    if ( (imageArraySize+trieBytes.size()+patchInfoSize) > freeSpace ) {
+        _diagnostics.error("cache buffer too small to hold ImageArray and Trie (buffer size=%lldMB, imageArray size=%lldMB, trie size=%luKB, patch size=%lluKB, free space=%ldMB)",
+                            _allocatedBufferSize/1024/1024, imageArraySize/1024/1024, trieBytes.size()/1024, patchInfoSize/1024, freeSpace/1024/1024);
+        return;
+    }
+
+    // copy into cache and update header
+    DyldSharedCache* dyldCache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    dyldCache->header.dylibsImageArrayAddr = _readOnlyRegion.unslidLoadAddress + _readOnlyRegion.sizeInUse;
+    dyldCache->header.dylibsImageArraySize = imageArraySize;
+    dyldCache->header.dylibsTrieAddr       = dyldCache->header.dylibsImageArrayAddr + imageArraySize;
+    dyldCache->header.dylibsTrieSize       = trieBytes.size();
+    ::memcpy(_readOnlyRegion.buffer + _readOnlyRegion.sizeInUse, _imageArray, imageArraySize);
+    ::memcpy(_readOnlyRegion.buffer + _readOnlyRegion.sizeInUse + imageArraySize, &trieBytes[0], trieBytes.size());
+
+    // Also write out the patch info
+    dyldCache->header.patchInfoAddr = dyldCache->header.dylibsTrieAddr + dyldCache->header.dylibsTrieSize;
+    dyldCache->header.patchInfoSize = patchInfoSize;
+    dyld_cache_patch_info patchInfo;
+    patchInfo.patchTableArrayAddr = dyldCache->header.patchInfoAddr + sizeof(dyld_cache_patch_info);
+    patchInfo.patchTableArrayCount = patchImages.size();
+    patchInfo.patchExportArrayAddr = patchInfo.patchTableArrayAddr + (patchInfo.patchTableArrayCount * sizeof(dyld_cache_image_patches));
+    patchInfo.patchExportArrayCount = patchExports.size();
+    patchInfo.patchLocationArrayAddr = patchInfo.patchExportArrayAddr + (patchInfo.patchExportArrayCount * sizeof(dyld_cache_patchable_export));
+    patchInfo.patchLocationArrayCount = patchLocations.size();
+    patchInfo.patchExportNamesAddr = patchInfo.patchLocationArrayAddr + (patchInfo.patchLocationArrayCount * sizeof(dyld_cache_patchable_location));
+    patchInfo.patchExportNamesSize = patchExportNames.size();
+    ::memcpy(_readOnlyRegion.buffer + dyldCache->header.patchInfoAddr - _readOnlyRegion.unslidLoadAddress,
+             &patchInfo, sizeof(dyld_cache_patch_info));
+    ::memcpy(_readOnlyRegion.buffer + patchInfo.patchTableArrayAddr - _readOnlyRegion.unslidLoadAddress,
+             &patchImages[0], sizeof(patchImages[0]) * patchImages.size());
+    ::memcpy(_readOnlyRegion.buffer + patchInfo.patchExportArrayAddr - _readOnlyRegion.unslidLoadAddress,
+             &patchExports[0], sizeof(patchExports[0]) * patchExports.size());
+    ::memcpy(_readOnlyRegion.buffer + patchInfo.patchLocationArrayAddr - _readOnlyRegion.unslidLoadAddress,
+             &patchLocations[0], sizeof(patchLocations[0]) * patchLocations.size());
+    ::memcpy(_readOnlyRegion.buffer + patchInfo.patchExportNamesAddr - _readOnlyRegion.unslidLoadAddress,
+             &patchExportNames[0], patchExportNames.size());
+
+    _readOnlyRegion.sizeInUse += align(imageArraySize+trieBytes.size()+patchInfoSize,14);
+
+    // Free the underlying image array buffer
+    _imageArray->deallocate();
+}
+
+void SharedCacheBuilder::addOtherImageArray(const std::vector<LoadedMachO>& otherDylibsAndBundles, std::vector<const LoadedMachO*>& overflowDylibs)
+{
+    DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    dyld3::closure::PathOverrides pathOverrides;
+    dyld3::closure::FileSystemNull nullFileSystem;
+    dyld3::closure::ClosureBuilder cb(dyld3::closure::kFirstOtherOSImageNum, nullFileSystem, cache, false, *_options.archs, pathOverrides,
+                                      dyld3::closure::ClosureBuilder::AtPath::none, false, nullptr, _options.platform);
+
+    // make ImageArray for other dylibs and bundles
+    STACK_ALLOC_ARRAY(dyld3::closure::LoadedFileInfo, others, otherDylibsAndBundles.size() + overflowDylibs.size());
+    for (const LoadedMachO& other : otherDylibsAndBundles) {
+        if ( !contains(other.loadedFileInfo.path, "staged_system_apps/") )
+            others.push_back(other.loadedFileInfo);
+    }
+
+    for (const LoadedMachO* dylib : overflowDylibs) {
+        if (dylib->mappedFile.mh->canHavePrecomputedDlopenClosure(dylib->mappedFile.runtimePath.c_str(), ^(const char*) {}) )
+            others.push_back(dylib->loadedFileInfo);
+    }
+
+    // Sort the others array by name so that it is deterministic
+    std::sort(others.begin(), others.end(),
+              [](const dyld3::closure::LoadedFileInfo& a, const dyld3::closure::LoadedFileInfo& b) {
+                  // Sort mac before iOSMac
+                  bool isIOSMacA = strncmp(a.path, "/System/iOSSupport/", 19) == 0;
+                  bool isIOSMacB = strncmp(b.path, "/System/iOSSupport/", 19) == 0;
+                  if (isIOSMacA != isIOSMacB)
+                      return !isIOSMacA;
+                  return strcmp(a.path, b.path) < 0;
+              });
+
+    const dyld3::closure::ImageArray* otherImageArray = cb.makeOtherDylibsImageArray(others, (uint32_t)_sortedDylibs.size());
+
+    // build trie of paths
+    __block std::vector<DylibIndexTrie::Entry> otherEntrys;
+    otherImageArray->forEachImage(^(const dyld3::closure::Image* image, bool& stop) {
+        if ( !image->isInvalid() )
+            otherEntrys.push_back(DylibIndexTrie::Entry(image->path(), DylibIndex(image->imageNum())));
+    });
+    DylibIndexTrie dylibsTrie(otherEntrys);
+    std::vector<uint8_t> trieBytes;
+    dylibsTrie.emit(trieBytes);
+    while ( (trieBytes.size() % 4) != 0 )
+        trieBytes.push_back(0);
+
+    // check for fit
+    uint64_t imageArraySize = otherImageArray->size();
+    size_t freeSpace = _readOnlyRegion.bufferSize - _readOnlyRegion.sizeInUse;
+    if ( imageArraySize+trieBytes.size() > freeSpace ) {
+        _diagnostics.error("cache buffer too small to hold ImageArray and Trie (buffer size=%lldMB, imageArray size=%lldMB, trie size=%luKB, free space=%ldMB)",
+                           _allocatedBufferSize/1024/1024, imageArraySize/1024/1024, trieBytes.size()/1024, freeSpace/1024/1024);
+        return;
+    }
+
+    // copy into cache and update header
+    DyldSharedCache* dyldCache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    dyldCache->header.otherImageArrayAddr = _readOnlyRegion.unslidLoadAddress + _readOnlyRegion.sizeInUse;
+    dyldCache->header.otherImageArraySize = imageArraySize;
+    dyldCache->header.otherTrieAddr       = dyldCache->header.otherImageArrayAddr + imageArraySize;
+    dyldCache->header.otherTrieSize       = trieBytes.size();
+    ::memcpy(_readOnlyRegion.buffer + _readOnlyRegion.sizeInUse, otherImageArray, imageArraySize);
+    ::memcpy(_readOnlyRegion.buffer + _readOnlyRegion.sizeInUse + imageArraySize, &trieBytes[0], trieBytes.size());
+    _readOnlyRegion.sizeInUse += align(imageArraySize+trieBytes.size(),14);
+
+    // Free the underlying buffer
+    otherImageArray->deallocate();
+}
+
+
+void SharedCacheBuilder::addClosures(const std::vector<LoadedMachO>& osExecutables)
+{
+    const DyldSharedCache* dyldCache = (DyldSharedCache*)_readExecuteRegion.buffer;
+
+    __block std::vector<Diagnostics> osExecutablesDiags;
+    __block std::vector<const dyld3::closure::LaunchClosure*> osExecutablesClosures;
+    osExecutablesDiags.resize(osExecutables.size());
+    osExecutablesClosures.resize(osExecutables.size());
+
+    dispatch_apply(osExecutables.size(), DISPATCH_APPLY_AUTO, ^(size_t index) {
+        const LoadedMachO& loadedMachO = osExecutables[index];
+        // don't pre-build closures for staged apps into dyld cache, since they won't run from that location
+        if ( startsWith(loadedMachO.mappedFile.runtimePath, "/private/var/staged_system_apps/") ) {
+            return;
+        }
+        dyld3::closure::PathOverrides pathOverrides;
+        dyld3::closure::ClosureBuilder builder(dyld3::closure::kFirstLaunchClosureImageNum, _fileSystem, dyldCache, false, *_options.archs, pathOverrides,
+                                               dyld3::closure::ClosureBuilder::AtPath::all, false, nullptr, _options.platform, nullptr);
+        bool issetuid = false;
+        if ( this->_options.platform == dyld3::Platform::macOS || dyld3::MachOFile::isSimulatorPlatform(this->_options.platform) )
+            _fileSystem.fileExists(loadedMachO.loadedFileInfo.path, nullptr, nullptr, &issetuid, nullptr);
+        const dyld3::closure::LaunchClosure* mainClosure = builder.makeLaunchClosure(loadedMachO.loadedFileInfo, issetuid);
+        if ( builder.diagnostics().hasError() ) {
+           osExecutablesDiags[index].error("%s", builder.diagnostics().errorMessage().c_str());
+        }
+        else {
+            assert(mainClosure != nullptr);
+            osExecutablesClosures[index] = mainClosure;
+        }
+    });
+
+    std::map<std::string, const dyld3::closure::LaunchClosure*> closures;
+    for (uint64_t i = 0, e = osExecutables.size(); i != e; ++i) {
+        const LoadedMachO& loadedMachO = osExecutables[i];
+        const Diagnostics& diag = osExecutablesDiags[i];
+        if (diag.hasError()) {
+            if ( _options.verbose ) {
+                _diagnostics.warning("building closure for '%s': %s", loadedMachO.mappedFile.runtimePath.c_str(), diag.errorMessage().c_str());
+                for (const std::string& warn : diag.warnings() )
+                    _diagnostics.warning("%s", warn.c_str());
+            }
+            if ( loadedMachO.inputFile && (loadedMachO.inputFile->mustBeIncluded()) ) {
+                loadedMachO.inputFile->diag.error("%s", diag.errorMessage().c_str());
+            }
+        } else {
+            // Note, a closure could be null here if it has a path we skip.
+            if (osExecutablesClosures[i] != nullptr)
+                closures[loadedMachO.mappedFile.runtimePath] = osExecutablesClosures[i];
+        }
+    }
+
+    osExecutablesDiags.clear();
+    osExecutablesClosures.clear();
+
+    // preflight space needed
+    size_t closuresSpace = 0;
+    for (const auto& entry : closures) {
+        closuresSpace += entry.second->size();
+    }
+    size_t freeSpace = _readOnlyRegion.bufferSize - _readOnlyRegion.sizeInUse;
+    if ( closuresSpace > freeSpace ) {
+        _diagnostics.error("cache buffer too small to hold all closures (buffer size=%lldMB, closures size=%ldMB, free space=%ldMB)",
+                            _allocatedBufferSize/1024/1024, closuresSpace/1024/1024, freeSpace/1024/1024);
+        return;
+    }
+    DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    cache->header.progClosuresAddr = _readOnlyRegion.unslidLoadAddress + _readOnlyRegion.sizeInUse;
+    uint8_t* closuresBase = _readOnlyRegion.buffer + _readOnlyRegion.sizeInUse;
+    std::vector<DylibIndexTrie::Entry> closureEntrys;
+    uint32_t currentClosureOffset = 0;
+    for (const auto& entry : closures) {
+        const dyld3::closure::LaunchClosure* closure = entry.second;
+        closureEntrys.push_back(DylibIndexTrie::Entry(entry.first, DylibIndex(currentClosureOffset)));
+        size_t size = closure->size();
+        assert((size % 4) == 0);
+        memcpy(closuresBase+currentClosureOffset, closure, size);
+        currentClosureOffset += size;
+        freeSpace -= size;
+        closure->deallocate();
+    }
+    cache->header.progClosuresSize = currentClosureOffset;
+    _readOnlyRegion.sizeInUse += currentClosureOffset;
+    freeSpace = _readOnlyRegion.bufferSize - _readOnlyRegion.sizeInUse;
+    // build trie of indexes into closures list
+    DylibIndexTrie closureTrie(closureEntrys);
+    std::vector<uint8_t> trieBytes;
+    closureTrie.emit(trieBytes);
+    while ( (trieBytes.size() % 8) != 0 )
+        trieBytes.push_back(0);
+    if ( trieBytes.size() > freeSpace ) {
+        _diagnostics.error("cache buffer too small to hold all closures trie (buffer size=%lldMB, trie size=%ldMB, free space=%ldMB)",
+                            _allocatedBufferSize/1024/1024, trieBytes.size()/1024/1024, freeSpace/1024/1024);
+        return;
+    }
+    memcpy(_readOnlyRegion.buffer + _readOnlyRegion.sizeInUse, &trieBytes[0], trieBytes.size());
+    cache->header.progClosuresTrieAddr = _readOnlyRegion.unslidLoadAddress + _readOnlyRegion.sizeInUse;
+    cache->header.progClosuresTrieSize = trieBytes.size();
+    _readOnlyRegion.sizeInUse += trieBytes.size();
+    _readOnlyRegion.sizeInUse = align(_readOnlyRegion.sizeInUse, 14);
+}
+
+
+bool SharedCacheBuilder::writeCache(void (^cacheSizeCallback)(uint64_t size), bool (^copyCallback)(const uint8_t* src, uint64_t size, uint64_t dstOffset))
+{
+    const dyld_cache_header*       cacheHeader = (dyld_cache_header*)_readExecuteRegion.buffer;
+    const dyld_cache_mapping_info* mappings = (dyld_cache_mapping_info*)(_readExecuteRegion.buffer + cacheHeader->mappingOffset);
+    assert(_readExecuteRegion.sizeInUse       == mappings[0].size);
+    assert(_readWriteRegion.sizeInUse         == mappings[1].size);
+    assert(_readOnlyRegion.sizeInUse          == mappings[2].size);
+    assert(_readExecuteRegion.cacheFileOffset == mappings[0].fileOffset);
+    assert(_readWriteRegion.cacheFileOffset   == mappings[1].fileOffset);
+    assert(_readOnlyRegion.cacheFileOffset    == mappings[2].fileOffset);
+    assert(_codeSignatureRegion.sizeInUse     == cacheHeader->codeSignatureSize);
+    assert(cacheHeader->codeSignatureOffset   == mappings[2].fileOffset+_readOnlyRegion.sizeInUse+_localSymbolsRegion.sizeInUse);
+    cacheSizeCallback(_readExecuteRegion.sizeInUse+_readWriteRegion.sizeInUse+_readOnlyRegion.sizeInUse+_localSymbolsRegion.sizeInUse+_codeSignatureRegion.sizeInUse);
+    bool fullyWritten = copyCallback(_readExecuteRegion.buffer, _readExecuteRegion.sizeInUse, mappings[0].fileOffset);
+    fullyWritten &= copyCallback(_readWriteRegion.buffer, _readWriteRegion.sizeInUse, mappings[1].fileOffset);
+    fullyWritten &= copyCallback(_readOnlyRegion.buffer, _readOnlyRegion.sizeInUse, mappings[2].fileOffset);
+    if ( _localSymbolsRegion.sizeInUse != 0 ) {
+        assert(cacheHeader->localSymbolsOffset == mappings[2].fileOffset+_readOnlyRegion.sizeInUse);
+        fullyWritten &= copyCallback(_localSymbolsRegion.buffer, _localSymbolsRegion.sizeInUse, cacheHeader->localSymbolsOffset);
+    }
+    fullyWritten &= copyCallback(_codeSignatureRegion.buffer, _codeSignatureRegion.sizeInUse, cacheHeader->codeSignatureOffset);
+    return fullyWritten;
+}
+
+
+void SharedCacheBuilder::writeFile(const std::string& path)
+{
+    std::string pathTemplate = path + "-XXXXXX";
+    size_t templateLen = strlen(pathTemplate.c_str())+2;
+    BLOCK_ACCCESSIBLE_ARRAY(char, pathTemplateSpace, templateLen);
+    strlcpy(pathTemplateSpace, pathTemplate.c_str(), templateLen);
+    int fd = mkstemp(pathTemplateSpace);
+    if ( fd != -1 ) {
+        auto cacheSizeCallback = ^(uint64_t size) {
+            // if making macOS dyld cache for current OS into standard location
+            if ( (_options.platform == dyld3::Platform::macOS) && startsWith(path, MACOSX_DYLD_SHARED_CACHE_DIR) ) {
+                // <rdar://48687550> pin cache file to SSD on fusion drives
+                apfs_data_pin_location_t where = APFS_PIN_DATA_TO_MAIN;
+                ::fsctl(pathTemplateSpace, APFSIOC_PIN_DATA, &where, 0);
+            }
+            // set final cache file size (may help defragment file)
+            ::ftruncate(fd, size);
+        };
+        auto copyCallback = ^(const uint8_t* src, uint64_t size, uint64_t dstOffset) {
+            uint64_t writtenSize = pwrite(fd, src, size, dstOffset);
+            return writtenSize == size;
+        };
+        // <rdar://problem/55370916> TOCTOU: verify path is still a realpath (not changed)
+        char tempPath[MAXPATHLEN];
+        if ( ::fcntl(fd, F_GETPATH, tempPath) == 0 ) {
+            size_t tempPathLen = strlen(tempPath);
+            if ( tempPathLen > 7 )
+                tempPath[tempPathLen-7] = '\0'; // remove trailing -xxxxxx
+            if ( path != tempPath ) {
+                _diagnostics.error("output file path changed from: '%s' to: '%s'", path.c_str(), tempPath);
+                ::close(fd);
+                return;
+            }
+        }
+        else {
+            _diagnostics.error("unable to fcntl(fd, F_GETPATH) on output file");
+            ::close(fd);
+            return;
+        }
+        bool fullyWritten = writeCache(cacheSizeCallback, copyCallback);
+        if ( fullyWritten ) {
+            ::fchmod(fd, S_IRUSR|S_IRGRP|S_IROTH); // mkstemp() makes file "rw-------", switch it to "r--r--r--"
+            // <rdar://problem/55370916> TOCTOU: verify path is still a realpath (not changed)
+            char resolvedPath[PATH_MAX];
+            ::realpath(path.c_str(), resolvedPath);
+            // Note: if the target cache file does not already exist, realpath() will return NULL, but still fill in the path buffer
+            if ( path != resolvedPath ) {
+                _diagnostics.error("output file path changed from: '%s' to: '%s'", path.c_str(), resolvedPath);
+                return;
+            }
+            if ( ::rename(pathTemplateSpace, path.c_str()) == 0) {
+                ::close(fd);
+                return; // success
+            }
+        }
+        else {
+            _diagnostics.error("could not write file %s", pathTemplateSpace);
+        }
+        ::close(fd);
+        ::unlink(pathTemplateSpace);
+    }
+    else {
+        _diagnostics.error("could not open file %s", pathTemplateSpace);
+    }
+}
+
+void SharedCacheBuilder::writeBuffer(uint8_t*& buffer, uint64_t& bufferSize) {
+    auto cacheSizeCallback = ^(uint64_t size) {
+        buffer = (uint8_t*)malloc(size);
+        bufferSize = size;
+    };
+    auto copyCallback = ^(const uint8_t* src, uint64_t size, uint64_t dstOffset) {
+        memcpy(buffer + dstOffset, src, size);
+        return true;
+    };
+    bool fullyWritten = writeCache(cacheSizeCallback, copyCallback);
+    assert(fullyWritten);
+}
+
+void SharedCacheBuilder::writeMapFile(const std::string& path)
+{
+    const DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    std::string mapContent = cache->mapFile();
+    safeSave(mapContent.c_str(), mapContent.size(), path);
+}
+
+std::string SharedCacheBuilder::getMapFileBuffer(const std::string& cacheDisposition) const
+{
+    const DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    return cache->generateJSONMap(cacheDisposition.c_str());
+}
+
+void SharedCacheBuilder::markPaddingInaccessible()
+{
+    // region between RX and RW
+    uint8_t* startPad1 = _readExecuteRegion.buffer+_readExecuteRegion.sizeInUse;
+    uint8_t* endPad1   = _readWriteRegion.buffer;
+    ::vm_protect(mach_task_self(), (vm_address_t)startPad1, endPad1-startPad1, false, 0);
+
+    // region between RW and RO
+    uint8_t* startPad2 = _readWriteRegion.buffer+_readWriteRegion.sizeInUse;
+    uint8_t* endPad2   = _readOnlyRegion.buffer;
+    ::vm_protect(mach_task_self(), (vm_address_t)startPad2, endPad2-startPad2, false, 0);
+}
+
+
+void SharedCacheBuilder::forEachCacheDylib(void (^callback)(const std::string& path)) {
+    for (const DylibInfo& dylibInfo : _sortedDylibs)
+        callback(dylibInfo.runtimePath);
+}
+
+
+uint64_t SharedCacheBuilder::pathHash(const char* path)
+{
+    uint64_t sum = 0;
+    for (const char* s=path; *s != '\0'; ++s)
+        sum += sum*4 + *s;
+    return sum;
+}
+
+
+void SharedCacheBuilder::findDylibAndSegment(const void* contentPtr, std::string& foundDylibName, std::string& foundSegName)
+{
+    foundDylibName = "???";
+    foundSegName   = "???";
+    uint64_t unslidVmAddr = ((uint8_t*)contentPtr - _readExecuteRegion.buffer) + _readExecuteRegion.unslidLoadAddress;
+    const DyldSharedCache* cache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    cache->forEachImage(^(const mach_header* mh, const char* installName) {
+        ((dyld3::MachOLoaded*)mh)->forEachSegment(^(const dyld3::MachOFile::SegmentInfo& info, bool &stop) {
+            if ( (unslidVmAddr >= info.vmAddr) && (unslidVmAddr < (info.vmAddr+info.vmSize)) ) {
+                foundDylibName = installName;
+                foundSegName   = info.segName;
+                stop           = true;
+            }
+        });
+    });
+}
+
+
+void SharedCacheBuilder::fipsSign()
+{
+    // find libcorecrypto.dylib in cache being built
+    DyldSharedCache* dyldCache = (DyldSharedCache*)_readExecuteRegion.buffer;
+    __block const dyld3::MachOLoaded* ml = nullptr;
+    dyldCache->forEachImage(^(const mach_header* mh, const char* installName) {
+        if ( strcmp(installName, "/usr/lib/system/libcorecrypto.dylib") == 0 )
+            ml = (dyld3::MachOLoaded*)mh;
+    });
+    if ( ml == nullptr ) {
+        _diagnostics.warning("Could not find libcorecrypto.dylib, skipping FIPS sealing");
+        return;
+    }
+
+    // find location in libcorecrypto.dylib to store hash of __text section
+    uint64_t hashStoreSize;
+    const void* hashStoreLocation = ml->findSectionContent("__TEXT", "__fips_hmacs", hashStoreSize);
+    if ( hashStoreLocation == nullptr ) {
+        _diagnostics.warning("Could not find __TEXT/__fips_hmacs section in libcorecrypto.dylib, skipping FIPS sealing");
+        return;
+    }
+    if ( hashStoreSize != 32 ) {
+        _diagnostics.warning("__TEXT/__fips_hmacs section in libcorecrypto.dylib is not 32 bytes in size, skipping FIPS sealing");
+        return;
+    }
+
+    // compute hmac hash of __text section
+    uint64_t textSize;
+    const void* textLocation = ml->findSectionContent("__TEXT", "__text", textSize);
+    if ( textLocation == nullptr ) {
+        _diagnostics.warning("Could not find __TEXT/__text section in libcorecrypto.dylib, skipping FIPS sealing");
+        return;
+    }
+    unsigned char hmac_key = 0;
+    CCHmac(kCCHmacAlgSHA256, &hmac_key, 1, textLocation, textSize, (void*)hashStoreLocation); // store hash directly into hashStoreLocation
+}
+
+void SharedCacheBuilder::codeSign()
+{
+    uint8_t  dscHashType;
+    uint8_t  dscHashSize;
+    uint32_t dscDigestFormat;
+    bool agile = false;
+
+    // select which codesigning hash
+    switch (_options.codeSigningDigestMode) {
+        case DyldSharedCache::Agile:
+            agile = true;
+            // Fall through to SHA1, because the main code directory remains SHA1 for compatibility.
+            [[clang::fallthrough]];
+        case DyldSharedCache::SHA1only:
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wdeprecated-declarations"
+            dscHashType     = CS_HASHTYPE_SHA1;
+            dscHashSize     = CS_HASH_SIZE_SHA1;
+            dscDigestFormat = kCCDigestSHA1;
+#pragma clang diagnostic pop
+            break;
+        case DyldSharedCache::SHA256only:
+            dscHashType     = CS_HASHTYPE_SHA256;
+            dscHashSize     = CS_HASH_SIZE_SHA256;
+            dscDigestFormat = kCCDigestSHA256;
+            break;
+        default:
+            _diagnostics.error("codeSigningDigestMode has unknown, unexpected value %d, bailing out.",
+                               _options.codeSigningDigestMode);
+            return;
+    }
+
+    std::string cacheIdentifier = "com.apple.dyld.cache.";
+    cacheIdentifier +=  _options.archs->name();
+    if ( _options.dylibsRemovedDuringMastering ) {
+        if ( _options.optimizeStubs  )
+            cacheIdentifier +=  ".release";
+        else
+            cacheIdentifier += ".development";
+    }
+    // get pointers into shared cache buffer
+    size_t          inBbufferSize = _readExecuteRegion.sizeInUse+_readWriteRegion.sizeInUse+_readOnlyRegion.sizeInUse+_localSymbolsRegion.sizeInUse;
+
+    const uint16_t pageSize = _archLayout->csPageSize;
+
+    // layout code signature contents
+    uint32_t blobCount     = agile ? 4 : 3;
+    size_t   idSize        = cacheIdentifier.size()+1; // +1 for terminating 0
+    uint32_t slotCount     = (uint32_t)((inBbufferSize + pageSize - 1) / pageSize);
+    uint32_t xSlotCount    = CSSLOT_REQUIREMENTS;
+    size_t   idOffset      = offsetof(CS_CodeDirectory, end_withExecSeg);
+    size_t   hashOffset    = idOffset+idSize + dscHashSize*xSlotCount;
+    size_t   hash256Offset = idOffset+idSize + CS_HASH_SIZE_SHA256*xSlotCount;
+    size_t   cdSize        = hashOffset + (slotCount * dscHashSize);
+    size_t   cd256Size     = agile ? hash256Offset + (slotCount * CS_HASH_SIZE_SHA256) : 0;
+    size_t   reqsSize      = 12;
+    size_t   cmsSize       = sizeof(CS_Blob);
+    size_t   cdOffset      = sizeof(CS_SuperBlob) + blobCount*sizeof(CS_BlobIndex);
+    size_t   cd256Offset   = cdOffset + cdSize;
+    size_t   reqsOffset    = cd256Offset + cd256Size; // equals cdOffset + cdSize if not agile
+    size_t   cmsOffset     = reqsOffset + reqsSize;
+    size_t   sbSize        = cmsOffset + cmsSize;
+    size_t   sigSize       = align(sbSize, 14);       // keep whole cache 16KB aligned
+
+    // allocate space for blob
+    vm_address_t codeSigAlloc;
+    if ( vm_allocate(mach_task_self(), &codeSigAlloc, sigSize, VM_FLAGS_ANYWHERE) != 0 ) {
+        _diagnostics.error("could not allocate code signature buffer");
+        return;
+    }
+    _codeSignatureRegion.buffer     = (uint8_t*)codeSigAlloc;
+    _codeSignatureRegion.bufferSize = sigSize;
+    _codeSignatureRegion.sizeInUse  = sigSize;
+
+    // create overall code signature which is a superblob
+    CS_SuperBlob* sb = reinterpret_cast<CS_SuperBlob*>(_codeSignatureRegion.buffer);
+    sb->magic           = htonl(CSMAGIC_EMBEDDED_SIGNATURE);
+    sb->length          = htonl(sbSize);
+    sb->count           = htonl(blobCount);
+    sb->index[0].type   = htonl(CSSLOT_CODEDIRECTORY);
+    sb->index[0].offset = htonl(cdOffset);
+    sb->index[1].type   = htonl(CSSLOT_REQUIREMENTS);
+    sb->index[1].offset = htonl(reqsOffset);
+    sb->index[2].type   = htonl(CSSLOT_CMS_SIGNATURE);
+    sb->index[2].offset = htonl(cmsOffset);
+    if ( agile ) {
+        sb->index[3].type = htonl(CSSLOT_ALTERNATE_CODEDIRECTORIES + 0);
+        sb->index[3].offset = htonl(cd256Offset);
+    }
+
+    // fill in empty requirements
+    CS_RequirementsBlob* reqs = (CS_RequirementsBlob*)(((char*)sb)+reqsOffset);
+    reqs->magic  = htonl(CSMAGIC_REQUIREMENTS);
+    reqs->length = htonl(sizeof(CS_RequirementsBlob));
+    reqs->data   = 0;
+
+    // initialize fixed fields of Code Directory
+    CS_CodeDirectory* cd = (CS_CodeDirectory*)(((char*)sb)+cdOffset);
+    cd->magic           = htonl(CSMAGIC_CODEDIRECTORY);
+    cd->length          = htonl(cdSize);
+    cd->version         = htonl(0x20400);               // supports exec segment
+    cd->flags           = htonl(kSecCodeSignatureAdhoc);
+    cd->hashOffset      = htonl(hashOffset);
+    cd->identOffset     = htonl(idOffset);
+    cd->nSpecialSlots   = htonl(xSlotCount);
+    cd->nCodeSlots      = htonl(slotCount);
+    cd->codeLimit       = htonl(inBbufferSize);
+    cd->hashSize        = dscHashSize;
+    cd->hashType        = dscHashType;
+    cd->platform        = 0;                            // not platform binary
+    cd->pageSize        = __builtin_ctz(pageSize);      // log2(CS_PAGE_SIZE);
+    cd->spare2          = 0;                            // unused (must be zero)
+    cd->scatterOffset   = 0;                            // not supported anymore
+    cd->teamOffset      = 0;                            // no team ID
+    cd->spare3          = 0;                            // unused (must be zero)
+    cd->codeLimit64     = 0;                            // falls back to codeLimit
+
+    // executable segment info
+    cd->execSegBase     = htonll(_readExecuteRegion.cacheFileOffset); // base of TEXT segment
+    cd->execSegLimit    = htonll(_readExecuteRegion.sizeInUse);       // size of TEXT segment
+    cd->execSegFlags    = 0;                                          // not a main binary
+
+    // initialize dynamic fields of Code Directory
+    strcpy((char*)cd + idOffset, cacheIdentifier.c_str());
+
+    // add special slot hashes
+    uint8_t* hashSlot = (uint8_t*)cd + hashOffset;
+    uint8_t* reqsHashSlot = &hashSlot[-CSSLOT_REQUIREMENTS*dscHashSize];
+    CCDigest(dscDigestFormat, (uint8_t*)reqs, sizeof(CS_RequirementsBlob), reqsHashSlot);
+
+    CS_CodeDirectory* cd256;
+    uint8_t* hash256Slot;
+    uint8_t* reqsHash256Slot;
+    if ( agile ) {
+        // Note that the assumption here is that the size up to the hashes is the same as for
+        // sha1 code directory, and that they come last, after everything else.
+
+        cd256 = (CS_CodeDirectory*)(((char*)sb)+cd256Offset);
+        cd256->magic           = htonl(CSMAGIC_CODEDIRECTORY);
+        cd256->length          = htonl(cd256Size);
+        cd256->version         = htonl(0x20400);               // supports exec segment
+        cd256->flags           = htonl(kSecCodeSignatureAdhoc);
+        cd256->hashOffset      = htonl(hash256Offset);
+        cd256->identOffset     = htonl(idOffset);
+        cd256->nSpecialSlots   = htonl(xSlotCount);
+        cd256->nCodeSlots      = htonl(slotCount);
+        cd256->codeLimit       = htonl(inBbufferSize);
+        cd256->hashSize        = CS_HASH_SIZE_SHA256;
+        cd256->hashType        = CS_HASHTYPE_SHA256;
+        cd256->platform        = 0;                            // not platform binary
+        cd256->pageSize        = __builtin_ctz(pageSize);      // log2(CS_PAGE_SIZE);
+        cd256->spare2          = 0;                            // unused (must be zero)
+        cd256->scatterOffset   = 0;                            // not supported anymore
+        cd256->teamOffset      = 0;                            // no team ID
+        cd256->spare3          = 0;                            // unused (must be zero)
+        cd256->codeLimit64     = 0;                            // falls back to codeLimit
+
+        // executable segment info
+        cd256->execSegBase     = cd->execSegBase;
+        cd256->execSegLimit    = cd->execSegLimit;
+        cd256->execSegFlags    = cd->execSegFlags;
+
+        // initialize dynamic fields of Code Directory
+        strcpy((char*)cd256 + idOffset, cacheIdentifier.c_str());
+
+        // add special slot hashes
+        hash256Slot = (uint8_t*)cd256 + hash256Offset;
+        reqsHash256Slot = &hash256Slot[-CSSLOT_REQUIREMENTS*CS_HASH_SIZE_SHA256];
+        CCDigest(kCCDigestSHA256, (uint8_t*)reqs, sizeof(CS_RequirementsBlob), reqsHash256Slot);
+    }
+    else {
+        cd256 = NULL;
+        hash256Slot = NULL;
+        reqsHash256Slot = NULL;
+    }
+
+    // fill in empty CMS blob for ad-hoc signing
+    CS_Blob* cms = (CS_Blob*)(((char*)sb)+cmsOffset);
+    cms->magic  = htonl(CSMAGIC_BLOBWRAPPER);
+    cms->length = htonl(sizeof(CS_Blob));
+
+
+    // alter header of cache to record size and location of code signature
+    // do this *before* hashing each page
+    dyld_cache_header* cache = (dyld_cache_header*)_readExecuteRegion.buffer;
+    cache->codeSignatureOffset  = inBbufferSize;
+    cache->codeSignatureSize    = sigSize;
+
+    const uint32_t rwSlotStart = (uint32_t)(_readExecuteRegion.sizeInUse / pageSize);
+    const uint32_t roSlotStart = (uint32_t)(rwSlotStart + _readWriteRegion.sizeInUse / pageSize);
+    const uint32_t localsSlotStart = (uint32_t)(roSlotStart + _readOnlyRegion.sizeInUse / pageSize);
+    auto codeSignPage = ^(size_t i) {
+        const uint8_t* code = nullptr;
+        // move to correct region
+        if ( i < rwSlotStart )
+            code = _readExecuteRegion.buffer + (i * pageSize);
+        else if ( i >= rwSlotStart && i < roSlotStart )
+            code = _readWriteRegion.buffer + ((i - rwSlotStart) * pageSize);
+        else if ( i >= roSlotStart && i < localsSlotStart )
+            code = _readOnlyRegion.buffer + ((i - roSlotStart) * pageSize);
+        else
+            code = _localSymbolsRegion.buffer + ((i - localsSlotStart) * pageSize);
+
+        CCDigest(dscDigestFormat, code, pageSize, hashSlot + (i * dscHashSize));
+
+        if ( agile ) {
+            CCDigest(kCCDigestSHA256, code, pageSize, hash256Slot + (i * CS_HASH_SIZE_SHA256));
+        }
+    };
+
+    // compute hashes
+    dispatch_apply(slotCount, DISPATCH_APPLY_AUTO, ^(size_t i) {
+        codeSignPage(i);
+    });
+
+    // Now that we have a code signature, compute a cache UUID by hashing the code signature blob
+    {
+        uint8_t* uuidLoc = cache->uuid;
+        assert(uuid_is_null(uuidLoc));
+        static_assert(offsetof(dyld_cache_header, uuid) / CS_PAGE_SIZE_4K == 0, "uuid is expected in the first page of the cache");
+        uint8_t fullDigest[CC_SHA256_DIGEST_LENGTH];
+        CC_SHA256((const void*)cd, (unsigned)cdSize, fullDigest);
+        memcpy(uuidLoc, fullDigest, 16);
+        // <rdar://problem/6723729> uuids should conform to RFC 4122 UUID version 4 & UUID version 5 formats
+        uuidLoc[6] = ( uuidLoc[6] & 0x0F ) | ( 3 << 4 );
+        uuidLoc[8] = ( uuidLoc[8] & 0x3F ) | 0x80;
+
+        // Now codesign page 0 again, because we modified it by setting uuid in header
+        codeSignPage(0);
+    }
+
+    // hash of entire code directory (cdHash) uses same hash as each page
+    uint8_t fullCdHash[dscHashSize];
+    CCDigest(dscDigestFormat, (const uint8_t*)cd, cdSize, fullCdHash);
+    // Note: cdHash is defined as first 20 bytes of hash
+    memcpy(_cdHashFirst, fullCdHash, 20);
+    if ( agile ) {
+        uint8_t fullCdHash256[CS_HASH_SIZE_SHA256];
+        CCDigest(kCCDigestSHA256, (const uint8_t*)cd256, cd256Size, fullCdHash256);
+        // Note: cdHash is defined as first 20 bytes of hash, even for sha256
+        memcpy(_cdHashSecond, fullCdHash256, 20);
+    }
+    else {
+        memset(_cdHashSecond, 0, 20);
+    }
+}
+
+const bool SharedCacheBuilder::agileSignature()
+{
+    return _options.codeSigningDigestMode == DyldSharedCache::Agile;
+}
+
+static const std::string cdHash(uint8_t hash[20])
+{
+    char buff[48];
+    for (int i = 0; i < 20; ++i)
+        sprintf(&buff[2*i], "%2.2x", hash[i]);
+    return buff;
+}
+
+const std::string SharedCacheBuilder::cdHashFirst()
+{
+    return cdHash(_cdHashFirst);
+}
+
+const std::string SharedCacheBuilder::cdHashSecond()
+{
+    return cdHash(_cdHashSecond);
+}
+
+const std::string SharedCacheBuilder::uuid() const
+{
+    dyld_cache_header* cache = (dyld_cache_header*)_readExecuteRegion.buffer;
+    uuid_string_t uuidStr;
+    uuid_unparse(cache->uuid, uuidStr);
+    return uuidStr;
+}
+
+
+
+template <typename P>
+bool SharedCacheBuilder::makeRebaseChainV2(uint8_t* pageContent, uint16_t lastLocationOffset, uint16_t offset, const dyld_cache_slide_info2* info)
+{
+    typedef typename P::uint_t     pint_t;
+
+    const pint_t   deltaMask    = (pint_t)(info->delta_mask);
+    const pint_t   valueMask    = ~deltaMask;
+    const pint_t   valueAdd     = (pint_t)(info->value_add);
+    const unsigned deltaShift   = __builtin_ctzll(deltaMask) - 2;
+    const uint32_t maxDelta     = (uint32_t)(deltaMask >> deltaShift);
+
+    pint_t* lastLoc = (pint_t*)&pageContent[lastLocationOffset+0];
+    pint_t lastValue = (pint_t)P::getP(*lastLoc);
+    if ( (lastValue - valueAdd) & deltaMask ) {
+        std::string dylibName;
+        std::string segName;
+        findDylibAndSegment((void*)pageContent, dylibName, segName);
+        _diagnostics.error("rebase pointer (0x%0lX) does not point within cache. lastOffset=0x%04X, seg=%s, dylib=%s\n",
+                            (long)lastValue, lastLocationOffset, segName.c_str(), dylibName.c_str());
+        return false;
+    }
+    if ( offset <= (lastLocationOffset+maxDelta) ) {
+        // previous location in range, make link from it
+        // encode this location into last value
+        pint_t delta = offset - lastLocationOffset;
+        pint_t newLastValue = ((lastValue - valueAdd) & valueMask) | (delta << deltaShift);
+        //warning("  add chain: delta = %d, lastOffset=0x%03X, offset=0x%03X, org value=0x%08lX, new value=0x%08lX",
+        //                    offset - lastLocationOffset, lastLocationOffset, offset, (long)lastValue, (long)newLastValue);
+        uint8_t highByte;
+        if ( _aslrTracker.hasHigh8(lastLoc, &highByte) ) {
+            uint64_t tbi = (uint64_t)highByte << 56;
+            newLastValue |= tbi;
+        }
+        P::setP(*lastLoc, newLastValue);
+        return true;
+    }
+    //fprintf(stderr, "  too big delta = %d, lastOffset=0x%03X, offset=0x%03X\n", offset - lastLocationOffset, lastLocationOffset, offset);
+
+    // distance between rebase locations is too far
+    // see if we can make a chain from non-rebase locations
+    uint16_t nonRebaseLocationOffsets[1024];
+    unsigned nrIndex = 0;
+    for (uint16_t i = lastLocationOffset; i < offset-maxDelta; ) {
+        nonRebaseLocationOffsets[nrIndex] = 0;
+        for (int j=maxDelta; j > 0; j -= 4) {
+            pint_t value = (pint_t)P::getP(*(pint_t*)&pageContent[i+j]);
+            if ( value == 0 ) {
+                // Steal values of 0 to be used in the rebase chain
+                nonRebaseLocationOffsets[nrIndex] = i+j;
+                break;
+            }
+        }
+        if ( nonRebaseLocationOffsets[nrIndex] == 0 ) {
+            lastValue = (pint_t)P::getP(*lastLoc);
+            pint_t newValue = ((lastValue - valueAdd) & valueMask);
+            //warning("   no way to make non-rebase delta chain, terminate off=0x%03X, old value=0x%08lX, new value=0x%08lX", lastLocationOffset, (long)value, (long)newValue);
+            P::setP(*lastLoc, newValue);
+            return false;
+        }
+        i = nonRebaseLocationOffsets[nrIndex];
+        ++nrIndex;
+    }
+
+    // we can make chain. go back and add each non-rebase location to chain
+    uint16_t prevOffset = lastLocationOffset;
+    pint_t* prevLoc = (pint_t*)&pageContent[prevOffset];
+    for (unsigned n=0; n < nrIndex; ++n) {
+        uint16_t nOffset = nonRebaseLocationOffsets[n];
+        assert(nOffset != 0);
+        pint_t* nLoc = (pint_t*)&pageContent[nOffset];
+        uint32_t delta2 = nOffset - prevOffset;
+        pint_t value = (pint_t)P::getP(*prevLoc);
+        pint_t newValue;
+        if ( value == 0 )
+            newValue = (delta2 << deltaShift);
+        else
+            newValue = ((value - valueAdd) & valueMask) | (delta2 << deltaShift);
+        //warning("    non-rebase delta = %d, to off=0x%03X, old value=0x%08lX, new value=0x%08lX", delta2, nOffset, (long)value, (long)newValue);
+        P::setP(*prevLoc, newValue);
+        prevOffset = nOffset;
+        prevLoc = nLoc;
+    }
+    uint32_t delta3 = offset - prevOffset;
+    pint_t value = (pint_t)P::getP(*prevLoc);
+    pint_t newValue;
+    if ( value == 0 )
+        newValue = (delta3 << deltaShift);
+    else
+        newValue = ((value - valueAdd) & valueMask) | (delta3 << deltaShift);
+    //warning("    non-rebase delta = %d, to off=0x%03X, old value=0x%08lX, new value=0x%08lX", delta3, offset, (long)value, (long)newValue);
+    P::setP(*prevLoc, newValue);
+
+    return true;
+}
+
+
+template <typename P>
+void SharedCacheBuilder::addPageStartsV2(uint8_t* pageContent, const bool bitmap[], const dyld_cache_slide_info2* info,
+                                         std::vector<uint16_t>& pageStarts, std::vector<uint16_t>& pageExtras)
+{
+    typedef typename P::uint_t     pint_t;
+
+    const pint_t   deltaMask    = (pint_t)(info->delta_mask);
+    const pint_t   valueMask    = ~deltaMask;
+    const uint32_t pageSize     = info->page_size;
+    const pint_t   valueAdd     = (pint_t)(info->value_add);
+
+    uint16_t startValue = DYLD_CACHE_SLIDE_PAGE_ATTR_NO_REBASE;
+    uint16_t lastLocationOffset = 0xFFFF;
+    for(uint32_t i=0; i < pageSize/4; ++i) {
+        unsigned offset = i*4;
+        if ( bitmap[i] ) {
+            if ( startValue == DYLD_CACHE_SLIDE_PAGE_ATTR_NO_REBASE ) {
+                // found first rebase location in page
+                startValue = i;
+            }
+            else if ( !makeRebaseChainV2<P>(pageContent, lastLocationOffset, offset, info) ) {
+                // can't record all rebasings in one chain
+                if ( (startValue & DYLD_CACHE_SLIDE_PAGE_ATTR_EXTRA) == 0 ) {
+                    // switch page_start to "extras" which is a list of chain starts
+                    unsigned indexInExtras = (unsigned)pageExtras.size();
+                    if ( indexInExtras > 0x3FFF ) {
+                        _diagnostics.error("rebase overflow in v2 page extras");
+                        return;
+                    }
+                    pageExtras.push_back(startValue);
+                    startValue = indexInExtras | DYLD_CACHE_SLIDE_PAGE_ATTR_EXTRA;
+                }
+                pageExtras.push_back(i);
+            }
+            lastLocationOffset = offset;
+        }
+    }
+    if ( lastLocationOffset != 0xFFFF ) {
+        // mark end of chain
+        pint_t* lastLoc = (pint_t*)&pageContent[lastLocationOffset];
+        pint_t lastValue = (pint_t)P::getP(*lastLoc);
+        pint_t newValue = ((lastValue - valueAdd) & valueMask);
+        P::setP(*lastLoc, newValue);
+    }
+    if ( startValue & DYLD_CACHE_SLIDE_PAGE_ATTR_EXTRA ) {
+        // add end bit to extras
+        pageExtras.back() |= DYLD_CACHE_SLIDE_PAGE_ATTR_END;
+    }
+    pageStarts.push_back(startValue);
+}
+
+template <typename P>
+void SharedCacheBuilder::writeSlideInfoV2(const bool bitmap[], unsigned dataPageCount)
+{
+    typedef typename P::uint_t    pint_t;
+    typedef typename P::E         E;
+    const uint32_t pageSize = 4096;
+
+    // fill in fixed info
+    assert(_slideInfoFileOffset != 0);
+    dyld_cache_slide_info2* info = (dyld_cache_slide_info2*)_readOnlyRegion.buffer;
+    info->version    = 2;
+    info->page_size  = pageSize;
+    info->delta_mask = _archLayout->pointerDeltaMask;
+    info->value_add  = _archLayout->useValueAdd ? _archLayout->sharedMemoryStart : 0;
+
+    // set page starts and extras for each page
+    std::vector<uint16_t> pageStarts;
+    std::vector<uint16_t> pageExtras;
+    pageStarts.reserve(dataPageCount);
+    uint8_t* pageContent = _readWriteRegion.buffer;
+    const bool* bitmapForPage = bitmap;
+    for (unsigned i=0; i < dataPageCount; ++i) {
+        //warning("page[%d]", i);
+        addPageStartsV2<P>(pageContent, bitmapForPage, info, pageStarts, pageExtras);
+        if ( _diagnostics.hasError() ) {
+            return;
+        }
+        pageContent += pageSize;
+        bitmapForPage += (sizeof(bool)*(pageSize/4));
+    }
+
+    // fill in computed info
+    info->page_starts_offset = sizeof(dyld_cache_slide_info2);
+    info->page_starts_count  = (unsigned)pageStarts.size();
+    info->page_extras_offset = (unsigned)(sizeof(dyld_cache_slide_info2)+pageStarts.size()*sizeof(uint16_t));
+    info->page_extras_count  = (unsigned)pageExtras.size();
+    uint16_t* pageStartsBuffer = (uint16_t*)((char*)info + info->page_starts_offset);
+    uint16_t* pageExtrasBuffer = (uint16_t*)((char*)info + info->page_extras_offset);
+    for (unsigned i=0; i < pageStarts.size(); ++i)
+        pageStartsBuffer[i] = pageStarts[i];
+    for (unsigned i=0; i < pageExtras.size(); ++i)
+        pageExtrasBuffer[i] = pageExtras[i];
+    // update header with final size
+    uint64_t slideInfoSize = align(info->page_extras_offset + pageExtras.size()*sizeof(uint16_t), _archLayout->sharedRegionAlignP2);
+    if ( slideInfoSize > _slideInfoBufferSizeAllocated ) {
+        _diagnostics.error("kernel slide info overflow buffer");
+    }
+    ((dyld_cache_header*)_readExecuteRegion.buffer)->slideInfoSize = slideInfoSize;
+    //fprintf(stderr, "pageCount=%u, page_starts_count=%lu, page_extras_count=%lu\n", dataPageCount, pageStarts.size(), pageExtras.size());
+}
+
+#if SUPPORT_ARCH_arm64_32 || SUPPORT_ARCH_armv7k
+// fits in to int16_t
+static bool smallValue(uint64_t value)
+{
+    uint32_t high = (value & 0xFFFF8000);
+    return (high == 0) || (high == 0xFFFF8000);
+}
+
+template <typename P>
+bool SharedCacheBuilder::makeRebaseChainV4(uint8_t* pageContent, uint16_t lastLocationOffset, uint16_t offset, const dyld_cache_slide_info4* info)
+{
+    typedef typename P::uint_t     pint_t;
+
+    const pint_t   deltaMask    = (pint_t)(info->delta_mask);
+    const pint_t   valueMask    = ~deltaMask;
+    const pint_t   valueAdd     = (pint_t)(info->value_add);
+    const unsigned deltaShift   = __builtin_ctzll(deltaMask) - 2;
+    const uint32_t maxDelta     = (uint32_t)(deltaMask >> deltaShift);
+
+    pint_t* lastLoc = (pint_t*)&pageContent[lastLocationOffset+0];
+    pint_t lastValue = (pint_t)P::getP(*lastLoc);
+    if ( (lastValue - valueAdd) & deltaMask ) {
+        std::string dylibName;
+        std::string segName;
+        findDylibAndSegment((void*)pageContent, dylibName, segName);
+        _diagnostics.error("rebase pointer does not point within cache. lastOffset=0x%04X, seg=%s, dylib=%s\n",
+                            lastLocationOffset, segName.c_str(), dylibName.c_str());
+        return false;
+    }
+    if ( offset <= (lastLocationOffset+maxDelta) ) {
+        // previous location in range, make link from it
+        // encode this location into last value
+        pint_t delta = offset - lastLocationOffset;
+        pint_t newLastValue = ((lastValue - valueAdd) & valueMask) | (delta << deltaShift);
+        //warning("  add chain: delta = %d, lastOffset=0x%03X, offset=0x%03X, org value=0x%08lX, new value=0x%08lX",
+        //                    offset - lastLocationOffset, lastLocationOffset, offset, (long)lastValue, (long)newLastValue);
+        P::setP(*lastLoc, newLastValue);
+        return true;
+    }
+    //fprintf(stderr, "  too big delta = %d, lastOffset=0x%03X, offset=0x%03X\n", offset - lastLocationOffset, lastLocationOffset, offset);
+
+    // distance between rebase locations is too far
+    // see if we can make a chain from non-rebase locations
+    uint16_t nonRebaseLocationOffsets[1024];
+    unsigned nrIndex = 0;
+    for (uint16_t i = lastLocationOffset; i < offset-maxDelta; ) {
+        nonRebaseLocationOffsets[nrIndex] = 0;
+        for (int j=maxDelta; j > 0; j -= 4) {
+            pint_t value = (pint_t)P::getP(*(pint_t*)&pageContent[i+j]);
+            if ( smallValue(value) ) {
+                // Steal values of 0 to be used in the rebase chain
+                nonRebaseLocationOffsets[nrIndex] = i+j;
+                break;
+            }
+        }
+        if ( nonRebaseLocationOffsets[nrIndex] == 0 ) {
+            lastValue = (pint_t)P::getP(*lastLoc);
+            pint_t newValue = ((lastValue - valueAdd) & valueMask);
+            //fprintf(stderr, "   no way to make non-rebase delta chain, terminate off=0x%03X, old value=0x%08lX, new value=0x%08lX\n",
+            //                lastLocationOffset, (long)lastValue, (long)newValue);
+            P::setP(*lastLoc, newValue);
+            return false;
+        }
+        i = nonRebaseLocationOffsets[nrIndex];
+        ++nrIndex;
+    }
+
+    // we can make chain. go back and add each non-rebase location to chain
+    uint16_t prevOffset = lastLocationOffset;
+    pint_t* prevLoc = (pint_t*)&pageContent[prevOffset];
+    for (unsigned n=0; n < nrIndex; ++n) {
+        uint16_t nOffset = nonRebaseLocationOffsets[n];
+        assert(nOffset != 0);
+        pint_t* nLoc = (pint_t*)&pageContent[nOffset];
+        uint32_t delta2 = nOffset - prevOffset;
+        pint_t value = (pint_t)P::getP(*prevLoc);
+        pint_t newValue;
+        if ( smallValue(value) )
+            newValue = (value & valueMask) | (delta2 << deltaShift);
+        else
+            newValue = ((value - valueAdd) & valueMask) | (delta2 << deltaShift);
+        //warning("    non-rebase delta = %d, to off=0x%03X, old value=0x%08lX, new value=0x%08lX", delta2, nOffset, (long)value, (long)newValue);
+        P::setP(*prevLoc, newValue);
+        prevOffset = nOffset;
+        prevLoc = nLoc;
+    }
+    uint32_t delta3 = offset - prevOffset;
+    pint_t value = (pint_t)P::getP(*prevLoc);
+    pint_t newValue;
+    if ( smallValue(value) )
+        newValue = (value & valueMask) | (delta3 << deltaShift);
+    else
+        newValue = ((value - valueAdd) & valueMask) | (delta3 << deltaShift);
+    //warning("    non-rebase delta = %d, to off=0x%03X, old value=0x%08lX, new value=0x%08lX", delta3, offset, (long)value, (long)newValue);
+    P::setP(*prevLoc, newValue);
+
+    return true;
+}
+
+
+template <typename P>
+void SharedCacheBuilder::addPageStartsV4(uint8_t* pageContent, const bool bitmap[], const dyld_cache_slide_info4* info,
+                                         std::vector<uint16_t>& pageStarts, std::vector<uint16_t>& pageExtras)
+{
+    typedef typename P::uint_t     pint_t;
+
+    const pint_t   deltaMask    = (pint_t)(info->delta_mask);
+    const pint_t   valueMask    = ~deltaMask;
+    const uint32_t pageSize     = info->page_size;
+    const pint_t   valueAdd     = (pint_t)(info->value_add);
+
+    uint16_t startValue = DYLD_CACHE_SLIDE4_PAGE_NO_REBASE;
+    uint16_t lastLocationOffset = 0xFFFF;
+    for(uint32_t i=0; i < pageSize/4; ++i) {
+        unsigned offset = i*4;
+        if ( bitmap[i] ) {
+            if ( startValue == DYLD_CACHE_SLIDE4_PAGE_NO_REBASE ) {
+                // found first rebase location in page
+                startValue = i;
+            }
+            else if ( !makeRebaseChainV4<P>(pageContent, lastLocationOffset, offset, info) ) {
+                // can't record all rebasings in one chain
+                if ( (startValue & DYLD_CACHE_SLIDE4_PAGE_USE_EXTRA) == 0 ) {
+                    // switch page_start to "extras" which is a list of chain starts
+                    unsigned indexInExtras = (unsigned)pageExtras.size();
+                    if ( indexInExtras >= DYLD_CACHE_SLIDE4_PAGE_INDEX ) {
+                        _diagnostics.error("rebase overflow in v4 page extras");
+                        return;
+                    }
+                    pageExtras.push_back(startValue);
+                    startValue = indexInExtras | DYLD_CACHE_SLIDE4_PAGE_USE_EXTRA;
+                }
+                pageExtras.push_back(i);
+            }
+            lastLocationOffset = offset;
+        }
+    }
+    if ( lastLocationOffset != 0xFFFF ) {
+        // mark end of chain
+        pint_t* lastLoc = (pint_t*)&pageContent[lastLocationOffset];
+        pint_t lastValue = (pint_t)P::getP(*lastLoc);
+        pint_t newValue = ((lastValue - valueAdd) & valueMask);
+        P::setP(*lastLoc, newValue);
+    }
+    if ( startValue & DYLD_CACHE_SLIDE4_PAGE_USE_EXTRA ) {
+        // add end bit to extras
+        pageExtras.back() |= DYLD_CACHE_SLIDE4_PAGE_EXTRA_END;
+    }
+    pageStarts.push_back(startValue);
+}
+
+
+
+template <typename P>
+void SharedCacheBuilder::writeSlideInfoV4(const bool bitmap[], unsigned dataPageCount)
+{
+    typedef typename P::uint_t    pint_t;
+    typedef typename P::E         E;
+    const uint32_t pageSize = 4096;
+
+    // fill in fixed info
+    assert(_slideInfoFileOffset != 0);
+    dyld_cache_slide_info4* info = (dyld_cache_slide_info4*)_readOnlyRegion.buffer;
+    info->version    = 4;
+    info->page_size  = pageSize;
+    info->delta_mask = _archLayout->pointerDeltaMask;
+    info->value_add  = info->value_add  = _archLayout->useValueAdd ? _archLayout->sharedMemoryStart : 0;
+
+    // set page starts and extras for each page
+    std::vector<uint16_t> pageStarts;
+    std::vector<uint16_t> pageExtras;
+    pageStarts.reserve(dataPageCount);
+    uint8_t* pageContent = _readWriteRegion.buffer;
+    const bool* bitmapForPage = bitmap;
+    for (unsigned i=0; i < dataPageCount; ++i) {
+        addPageStartsV4<P>(pageContent, bitmapForPage, info, pageStarts, pageExtras);
+        if ( _diagnostics.hasError() ) {
+            return;
+        }
+        pageContent += pageSize;
+        bitmapForPage += (sizeof(bool)*(pageSize/4));
+    }
+    // fill in computed info
+    info->page_starts_offset = sizeof(dyld_cache_slide_info4);
+    info->page_starts_count  = (unsigned)pageStarts.size();
+    info->page_extras_offset = (unsigned)(sizeof(dyld_cache_slide_info4)+pageStarts.size()*sizeof(uint16_t));
+    info->page_extras_count  = (unsigned)pageExtras.size();
+    uint16_t* pageStartsBuffer = (uint16_t*)((char*)info + info->page_starts_offset);
+    uint16_t* pageExtrasBuffer = (uint16_t*)((char*)info + info->page_extras_offset);
+    for (unsigned i=0; i < pageStarts.size(); ++i)
+        pageStartsBuffer[i] = pageStarts[i];
+    for (unsigned i=0; i < pageExtras.size(); ++i)
+        pageExtrasBuffer[i] = pageExtras[i];
+    // update header with final size
+    uint64_t slideInfoSize = align(info->page_extras_offset + pageExtras.size()*sizeof(uint16_t), _archLayout->sharedRegionAlignP2);
+    if ( slideInfoSize > _slideInfoBufferSizeAllocated ) {
+        _diagnostics.error("kernel slide info v4 overflow buffer, need %lldKB, have room for %lldKB", slideInfoSize, _slideInfoBufferSizeAllocated);
+    }
+    ((dyld_cache_header*)_readExecuteRegion.buffer)->slideInfoSize = slideInfoSize;
+    //fprintf(stderr, "pageCount=%u, page_starts_count=%lu, page_extras_count=%lu\n", dataPageCount, pageStarts.size(), pageExtras.size());
+}
+#endif
+
+/*
+void CacheBuilder::writeSlideInfoV1()
+{
+    // build one 128-byte bitmap per page (4096) of DATA
+    uint8_t* const dataStart = (uint8_t*)_buffer.get() + regions[1].fileOffset;
+    uint8_t* const dataEnd   = dataStart + regions[1].size;
+    const long bitmapSize = (dataEnd - dataStart)/(4*8);
+    uint8_t* bitmap = (uint8_t*)calloc(bitmapSize, 1);
+    for (void* p : _pointersForASLR) {
+        if ( (p < dataStart) || ( p > dataEnd) )
+            terminate("DATA pointer for sliding, out of range\n");
+        long offset = (long)((uint8_t*)p - dataStart);
+        if ( (offset % 4) != 0 )
+            terminate("pointer not 4-byte aligned in DATA offset 0x%08lX\n", offset);
+        long byteIndex = offset / (4*8);
+        long bitInByte =  (offset % 32) >> 2;
+        bitmap[byteIndex] |= (1 << bitInByte);
+    }
+
+    // allocate worst case size block of all slide info
+    const unsigned entry_size = 4096/(8*4); // 8 bits per byte, possible pointer every 4 bytes.
+    const unsigned toc_count = (unsigned)bitmapSize/entry_size;
+    dyld_cache_slide_info* slideInfo = (dyld_cache_slide_info*)((uint8_t*)_buffer + _slideInfoFileOffset);
+    slideInfo->version          = 1;
+    slideInfo->toc_offset       = sizeof(dyld_cache_slide_info);
+    slideInfo->toc_count        = toc_count;
+    slideInfo->entries_offset   = (slideInfo->toc_offset+2*toc_count+127)&(-128);
+    slideInfo->entries_count    = 0;
+    slideInfo->entries_size     = entry_size;
+    // append each unique entry
+    const dyldCacheSlideInfoEntry* bitmapAsEntries = (dyldCacheSlideInfoEntry*)bitmap;
+    dyldCacheSlideInfoEntry* const entriesInSlidInfo = (dyldCacheSlideInfoEntry*)((char*)slideInfo+slideInfo->entries_offset());
+    int entry_count = 0;
+    for (int i=0; i < toc_count; ++i) {
+        const dyldCacheSlideInfoEntry* thisEntry = &bitmapAsEntries[i];
+        // see if it is same as one already added
+        bool found = false;
+        for (int j=0; j < entry_count; ++j) {
+            if ( memcmp(thisEntry, &entriesInSlidInfo[j], entry_size) == 0 ) {
+                slideInfo->set_toc(i, j);
+                found = true;
+                break;
+            }
+        }
+        if ( !found ) {
+            // append to end
+            memcpy(&entriesInSlidInfo[entry_count], thisEntry, entry_size);
+            slideInfo->set_toc(i, entry_count++);
+        }
+    }
+    slideInfo->entries_count  = entry_count;
+    ::free((void*)bitmap);
+
+    _buffer.header->slideInfoSize = align(slideInfo->entries_offset + entry_count*entry_size, _archLayout->sharedRegionAlignP2);
+}
+
+*/
+
+
+void SharedCacheBuilder::setPointerContentV3(dyld3::MachOLoaded::ChainedFixupPointerOnDisk* loc, uint64_t targetVMAddr, size_t next)
+{
+    assert(targetVMAddr > _readExecuteRegion.unslidLoadAddress);
+    assert(targetVMAddr < _readOnlyRegion.unslidLoadAddress+_readOnlyRegion.sizeInUse);
+    dyld3::MachOLoaded::ChainedFixupPointerOnDisk tmp;
+    uint16_t diversity;
+    bool     hasAddrDiv;
+    uint8_t  key;
+    if ( _aslrTracker.hasAuthData(loc, &diversity, &hasAddrDiv, &key) ) {
+        // if base cache address cannot fit into target, then use offset
+        tmp.arm64e.authRebase.target = _readExecuteRegion.unslidLoadAddress;
+        if (  tmp.arm64e.authRebase.target != _readExecuteRegion.unslidLoadAddress )
+            targetVMAddr -= _readExecuteRegion.unslidLoadAddress;
+        loc->arm64e.authRebase.target    = targetVMAddr;
+        loc->arm64e.authRebase.diversity = diversity;
+        loc->arm64e.authRebase.addrDiv   = hasAddrDiv;
+        loc->arm64e.authRebase.key       = key;
+        loc->arm64e.authRebase.next      = next;
+        loc->arm64e.authRebase.bind      = 0;
+        loc->arm64e.authRebase.auth      = 1;
+        assert(loc->arm64e.authRebase.target == targetVMAddr && "target truncated");
+        assert(loc->arm64e.authRebase.next == next && "next location truncated");
+    }
+    else {
+        uint8_t highByte = 0;
+        _aslrTracker.hasHigh8(loc, &highByte);
+        // if base cache address cannot fit into target, then use offset
+        tmp.arm64e.rebase.target = _readExecuteRegion.unslidLoadAddress;
+        if ( tmp.arm64e.rebase.target != _readExecuteRegion.unslidLoadAddress )
+            targetVMAddr -= _readExecuteRegion.unslidLoadAddress;
+        loc->arm64e.rebase.target   = targetVMAddr;
+        loc->arm64e.rebase.high8    = highByte;
+        loc->arm64e.rebase.next     = next;
+        loc->arm64e.rebase.bind     = 0;
+        loc->arm64e.rebase.auth     = 0;
+        assert(loc->arm64e.rebase.target == targetVMAddr && "target truncated");
+        assert(loc->arm64e.rebase.next == next && "next location truncated");
+    }
+}
+
+uint16_t SharedCacheBuilder::pageStartV3(uint8_t* pageContent, uint32_t pageSize, const bool bitmap[])
+{
+    const int maxPerPage = pageSize / 4;
+    uint16_t result = DYLD_CACHE_SLIDE_V3_PAGE_ATTR_NO_REBASE;
+    dyld3::MachOLoaded::ChainedFixupPointerOnDisk* lastLoc = nullptr;
+    for (int i=0; i < maxPerPage; ++i) {
+        if ( bitmap[i] ) {
+            if ( result == DYLD_CACHE_SLIDE_V3_PAGE_ATTR_NO_REBASE ) {
+                // found first rebase location in page
+                result = i * 4;
+            }
+            dyld3::MachOLoaded::ChainedFixupPointerOnDisk* loc = (dyld3::MachOLoaded::ChainedFixupPointerOnDisk*)(pageContent + i*4);;
+            if ( lastLoc != nullptr ) {
+                // convert vmaddr based pointers to arm64e dyld cache chains
+                setPointerContentV3(lastLoc, lastLoc->raw64, loc - lastLoc);
+            }
+            lastLoc = loc;
+        }
+    }
+    if ( lastLoc != nullptr ) {
+        // convert vmaddr based pointers to arm64e dyld cache chain, and mark end of chain
+        setPointerContentV3(lastLoc, lastLoc->raw64, 0);
+    }
+    return result;
+}
+
+
+void SharedCacheBuilder::writeSlideInfoV3(const bool bitmap[], unsigned dataPageCount)
+{
+    const uint32_t pageSize = 4096;
+
+    // fill in fixed info
+    assert(_slideInfoFileOffset != 0);
+    dyld_cache_slide_info3* info = (dyld_cache_slide_info3*)_readOnlyRegion.buffer;
+    info->version           = 3;
+    info->page_size         = pageSize;
+    info->page_starts_count = dataPageCount;
+    info->auth_value_add    = _archLayout->sharedMemoryStart;
+
+    // fill in per-page starts
+    uint8_t* pageContent = _readWriteRegion.buffer;
+    const bool* bitmapForPage = bitmap;
+    for (unsigned i=0; i < dataPageCount; ++i) {
+        info->page_starts[i] = pageStartV3(pageContent, pageSize, bitmapForPage);
+        pageContent += pageSize;
+        bitmapForPage += (sizeof(bool)*(pageSize/4));
+    }
+
+    // update header with final size
+    dyld_cache_header* dyldCacheHeader = (dyld_cache_header*)_readExecuteRegion.buffer;
+    dyldCacheHeader->slideInfoSize = align(__offsetof(dyld_cache_slide_info3, page_starts[dataPageCount]), _archLayout->sharedRegionAlignP2);
+    if ( dyldCacheHeader->slideInfoSize > _slideInfoBufferSizeAllocated ) {
+        _diagnostics.error("kernel slide info overflow buffer");
+    }
+}