Loading...
--- dyld/dyld-132.13/launch-cache/update_dyld_shared_cache.cpp
+++ dyld/dyld-360.22/launch-cache/update_dyld_shared_cache.cpp
@@ -1,6 +1,6 @@
/* -*- mode: C++; c-basic-offset: 4; tab-width: 4 -*-
*
- * Copyright (c) 2006-2009 Apple Inc. All rights reserved.
+ * Copyright (c) 2006-2011 Apple Inc. All rights reserved.
*
* @APPLE_LICENSE_HEADER_START@
*
@@ -31,7 +31,9 @@
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
+#include <math.h>
#include <fcntl.h>
+#include <dlfcn.h>
#include <signal.h>
#include <errno.h>
#include <sys/uio.h>
@@ -44,28 +46,38 @@
#include <servers/bootstrap.h>
#include <mach-o/loader.h>
#include <mach-o/fat.h>
-#include <vproc.h>
-#include <vproc_priv.h>
+#include <CoreFoundation/CoreFoundation.h>
+#include <Security/Security.h>
+#include <Security/SecCodeSigner.h>
+#include <CommonCrypto/CommonDigest.h>
#include "dyld_cache_format.h"
#include <vector>
#include <set>
#include <map>
-#include <ext/hash_map>
+#include <unordered_map>
#include "Architectures.hpp"
#include "MachOLayout.hpp"
#include "MachORebaser.hpp"
#include "MachOBinder.hpp"
#include "CacheFileAbstraction.hpp"
+#include "dyld_cache_config.h"
#define SELOPT_WRITE
-#include <objc/objc-selopt.h>
-
+#include "objc-shared-cache.h"
+
+#define FIRST_DYLIB_TEXT_OFFSET 0x10000
+
+#ifndef LC_FUNCTION_STARTS
+ #define LC_FUNCTION_STARTS 0x26
+#endif
static bool verbose = false;
static bool progress = false;
+static bool iPhoneOS = false;
+static bool rootless = true;
static std::vector<const char*> warnings;
@@ -90,20 +102,40 @@
}
-static uint64_t pageAlign(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
+class CStringHash {
+public:
+ size_t operator()(const char* __s) const {
+ size_t __h = 0;
+ for ( ; *__s; ++__s)
+ __h = 5 * __h + *__s;
+ return __h;
+ };
+};
+class CStringEquals
+{
+public:
+ bool operator()(const char* left, const char* right) const { return (strcmp(left, right) == 0); }
+};
+
+
class ArchGraph
{
public:
+ typedef std::unordered_map<const char*, const char*, CStringHash, CStringEquals> StringToString;
+
static void addArchPair(ArchPair ap);
static void addRoot(const char* vpath, const std::set<ArchPair>& archs);
+ static uint64_t maxCacheSizeForArchPair(ArchPair ap);
static void findSharedDylibs(ArchPair ap);
static ArchGraph* graphForArchPair(ArchPair ap) { return fgPerArchGraph[ap]; }
- static void setFileSystemRoot(const char* root, bool usesOverlay) { fgFileSystemRoot = root; fgUsesOverlay = usesOverlay; }
+ static void setFileSystemRoot(const char* root) { fgFileSystemRoot = root; }
+ static void setFileSystemOverlay(const std::vector<const char*>& overlays);
static const char* archName(ArchPair ap);
ArchPair getArchPair() { return fArchPair; }
std::set<const class MachOLayoutAbstraction*>& getSharedDylibs() { return fSharedDylibs; }
+ StringToString& getDylibAliases() { return fAliasesMap; }
const char* archName() { return archName(fArchPair); }
private:
@@ -118,6 +150,8 @@
const MachOLayoutAbstraction* getLayout() const { return fLayout; }
size_t useCount() const { return fRootsDependentOnThis.size(); }
bool allDependentsFound() const { return !fDependentMissing; }
+ bool dependsOnDylibList() const { return fRootsDependentOnThis.count(const_cast<DependencyNode*>(this)); }
+
private:
ArchGraph* fGraph;
const char* fPath;
@@ -128,10 +162,7 @@
std::set<DependencyNode*> fRootsDependentOnThis;
};
- struct CStringEquals {
- bool operator()(const char* left, const char* right) const { return (strcmp(left, right) == 0); }
- };
- typedef __gnu_cxx::hash_map<const char*, class DependencyNode*, __gnu_cxx::hash<const char*>, CStringEquals> PathToNode;
+ typedef std::unordered_map<const char*, class DependencyNode*, CStringHash, CStringEquals> PathToNode;
ArchGraph(ArchPair ap) : fArchPair(ap) {}
@@ -139,19 +170,21 @@
DependencyNode* getNode(const char* path);
DependencyNode* getNodeForVirtualPath(const char* vpath);
static bool canBeShared(const MachOLayoutAbstraction* layout, ArchPair ap, const std::set<const MachOLayoutAbstraction*>& possibleLibs, std::map<const MachOLayoutAbstraction*, bool>& shareableMap);
+ static bool sharable(const MachOLayoutAbstraction* layout, ArchPair ap, char** msg);
static std::map<ArchPair, ArchGraph*> fgPerArchGraph;
static const char* fgFileSystemRoot;
- static bool fgUsesOverlay;
+ static std::vector<const char*> fgFileSystemOverlays;
ArchPair fArchPair;
std::set<DependencyNode*> fRoots;
PathToNode fNodes;
std::set<const MachOLayoutAbstraction*> fSharedDylibs; // use set to avoid duplicates when installname!=realpath
+ StringToString fAliasesMap;
};
std::map<ArchPair, ArchGraph*> ArchGraph::fgPerArchGraph;
const char* ArchGraph::fgFileSystemRoot = "";
-bool ArchGraph::fgUsesOverlay = false;
+std::vector<const char*> ArchGraph::fgFileSystemOverlays;
void ArchGraph::addArchPair(ArchPair ap)
{
@@ -159,45 +192,56 @@
fgPerArchGraph[ap] = new ArchGraph(ap);
}
+void ArchGraph::setFileSystemOverlay(const std::vector<const char*>& overlays)
+{
+ for (std::vector<const char*>::const_iterator it=overlays.begin(); it != overlays.end(); ++it)
+ fgFileSystemOverlays.push_back(*it);
+}
+
void ArchGraph::addRoot(const char* vpath, const std::set<ArchPair>& onlyArchs)
{
- char completePath[strlen(fgFileSystemRoot)+strlen(vpath)+2];
+ //fprintf(stderr, "addRoot(%s)\n", vpath);
+ char completePath[MAXPATHLEN];
const char* path = NULL;
- if ( strlen(fgFileSystemRoot) == 0 ) {
- path = vpath;
- }
- else {
+ // check -overlay path first
+ for (std::vector<const char*>::const_iterator it=fgFileSystemOverlays.begin(); it != fgFileSystemOverlays.end(); ++it) {
+ strcpy(completePath, *it);
+ strcat(completePath, vpath); // assumes vpath starts with '/'
+ struct stat stat_buf;
+ if ( stat(completePath, &stat_buf) == 0 ) {
+ path = completePath;
+ break;
+ }
+ }
+ // if not found in overlay, check for -root
+ if ( (path == NULL) && (fgFileSystemRoot[0] != '\0') ) {
strcpy(completePath, fgFileSystemRoot);
strcat(completePath, vpath); // assumes vpath starts with '/'
- if ( fgUsesOverlay ) {
- // using -overlay means if /overlay/usr/lib exists use it, otherwise use original path
- struct stat stat_buf;
- if ( stat(completePath, &stat_buf) == 0 )
- path = completePath;
- else
- path = vpath;
- }
- else {
- // using -root means alway redirect /usr/lib to /rootpath/usr/lib
+ struct stat stat_buf;
+ if ( stat(completePath, &stat_buf) == 0 )
path = completePath;
- }
- }
+ }
+ if ( path == NULL )
+ path = vpath;
+
try {
+ //fprintf(stderr, " UniversalMachOLayout::find(%s)\n", path);
const UniversalMachOLayout& uni = UniversalMachOLayout::find(path, &onlyArchs);
for(std::set<ArchPair>::iterator ait = onlyArchs.begin(); ait != onlyArchs.end(); ++ait) {
try {
const MachOLayoutAbstraction* layout = uni.getSlice(*ait);
- fgPerArchGraph[*ait]->addRoot(path, layout);
+ if ( layout != NULL )
+ fgPerArchGraph[*ait]->addRoot(path, layout);
}
catch (const char* msg) {
if ( verbose )
- fprintf(stderr, "update_dyld_shared_cache: warning for %s can't use root %s: %s\n", fgPerArchGraph[*ait]->archName(), path, msg);
+ fprintf(stderr, "update_dyld_shared_cache: warning for %s can't use root '%s': %s\n", fgPerArchGraph[*ait]->archName(), path, msg);
}
}
}
catch (const char* msg) {
- fprintf(stderr, "update_dyld_shared_cache: warning can't use root %s: %s\n", path, msg);
+ fprintf(stderr, "update_dyld_shared_cache: warning can't use root '%s': %s\n", path, msg);
}
}
@@ -221,30 +265,57 @@
// a virtual path does not have the fgFileSystemRoot prefix
ArchGraph::DependencyNode* ArchGraph::getNodeForVirtualPath(const char* vpath)
{
- if ( fgFileSystemRoot == NULL ) {
- return this->getNode(vpath);
- }
- else {
- char completePath[strlen(fgFileSystemRoot)+strlen(vpath)+2];
+ //fprintf(stderr, "getNodeForVirtualPath(%s)\n", vpath);
+ char completePath[MAXPATHLEN];
+ for (std::vector<const char*>::const_iterator it=fgFileSystemOverlays.begin(); it != fgFileSystemOverlays.end(); ++it) {
+ const char* overlayPath = *it;
+ // using -overlay means if /overlay/path/dylib exists use it, otherwise use /path/dylib
+ strcpy(completePath, overlayPath);
+ strcat(completePath, vpath); // assumes vpath starts with '/'
+ struct stat stat_buf;
+ if ( stat(completePath, &stat_buf) == 0 ) {
+ return this->getNode(completePath);
+ }
+ // <rdar://problem/9279770> support when install name is a symlink
+ const char* pathToSymlink = vpath;
+ if ( fgFileSystemRoot[0] != '\0' ) {
+ strcpy(completePath, fgFileSystemRoot);
+ strcat(completePath, vpath);
+ pathToSymlink = completePath;
+ }
+ if ( (lstat(pathToSymlink, &stat_buf) == 0) && S_ISLNK(stat_buf.st_mode) ) {
+ // requested path did not exist in /overlay, but leaf of path is a symlink in /
+ char pathInSymLink[MAXPATHLEN];
+ size_t res = readlink(pathToSymlink, pathInSymLink, sizeof(pathInSymLink));
+ if ( res != -1 ) {
+ pathInSymLink[res] = '\0';
+ if ( pathInSymLink[0] != '/' ) {
+ char symFullPath[MAXPATHLEN];
+ strcpy(symFullPath, vpath);
+ char* lastSlash = strrchr(symFullPath, '/');
+ if ( lastSlash != NULL ) {
+ strcpy(lastSlash+1, pathInSymLink);
+ // (re)try looking for what symlink points to, but in /overlay
+ return this->getNodeForVirtualPath(symFullPath);
+ }
+ }
+ }
+ }
+ }
+
+ if ( fgFileSystemRoot[0] != '\0' ) {
+ // using -root means always use /rootpath/usr/lib
strcpy(completePath, fgFileSystemRoot);
strcat(completePath, vpath); // assumes vpath starts with '/'
- if ( fgUsesOverlay ) {
- // using -overlay means if /overlay/usr/lib exists use it, otherwise use original path
- struct stat stat_buf;
- if ( stat(completePath, &stat_buf) == 0 )
- return this->getNode(completePath);
- else
- return this->getNode(vpath);
- }
- else {
- // using -root means always use /rootpath/usr/lib
- return this->getNode(completePath);
- }
- }
+ return this->getNode(completePath);
+ }
+ // not found in -overlay or -root not used
+ return this->getNode(vpath);
}
ArchGraph::DependencyNode* ArchGraph::getNode(const char* path)
{
+ //fprintf(stderr, "getNode(%s)\n", path);
// look up supplied path to see if node already exists
PathToNode::iterator pos = fNodes.find(path);
if ( pos != fNodes.end() )
@@ -257,8 +328,20 @@
// look up real path to see if node already exists
pos = fNodes.find(realPath);
- if ( pos != fNodes.end() )
+ if ( pos != fNodes.end() ) {
+ // update fAliasesMap with symlinks found
+ const char* aliasPath = path;
+ if ( (fgFileSystemRoot != NULL) && (strncmp(path, fgFileSystemRoot, strlen(fgFileSystemRoot)) == 0) ) {
+ aliasPath = &path[strlen(fgFileSystemRoot)];
+ }
+ if ( fAliasesMap.find(aliasPath) == fAliasesMap.end() ) {
+ if ( strcmp(aliasPath, pos->second->getLayout()->getID().name) != 0 ) {
+ fAliasesMap[strdup(aliasPath)] = pos->second->getLayout()->getID().name;
+ //fprintf(stderr, "getNode() %s: added alias %s -> %s\n", archName(fArchPair), aliasPath, fAliasesMap[aliasPath]);
+ }
+ }
return pos->second;
+ }
// still does not exist, so create a new node
const UniversalMachOLayout& uni = UniversalMachOLayout::find(realPath);
@@ -270,8 +353,50 @@
fNodes[node->getPath()] = node;
// if install name is not real path, add install name to node map
if ( (node->getLayout()->getFileType() == MH_DYLIB) && (strcmp(realPath, node->getLayout()->getID().name) != 0) ) {
- //fprintf(stderr, "adding node alias 0x%08X %s for %s\n", fArch, node->getLayout()->getID().name, realPath);
- fNodes[node->getLayout()->getID().name] = node;
+ //fprintf(stderr, "adding %s node alias %s for %s\n", archName(fArchPair), node->getLayout()->getID().name, realPath);
+ pos = fNodes.find(node->getLayout()->getID().name);
+ if ( pos != fNodes.end() ) {
+ // get uuids of two dylibs to see if this is accidental copy of a dylib or two differnent dylibs with same -install_name
+ uuid_t uuid1;
+ uuid_t uuid2;
+ node->getLayout()->uuid(uuid1);
+ pos->second->getLayout()->uuid(uuid2);
+ if ( memcmp(&uuid1, &uuid2, 16) == 0 ) {
+ // <rdar://problem/8305479> warn if two dylib in cache have same install_name
+ char* msg;
+ asprintf(&msg, "update_dyld_shared_cache: warning, found two copies of the same dylib with same install path: %s\n\t%s\n\t%s\n",
+ node->getLayout()->getID().name, pos->second->getPath(), node->getPath());
+ fprintf(stderr, "%s", msg);
+ warnings.push_back(msg);
+ }
+ else {
+ // <rdar://problem/12763450> update_dyld_shared_cache should fail if two images have same install name
+ fprintf(stderr, "update_dyld_shared_cache: found two different dylibs with same install path: %s\n\t%s\n\t%s\n",
+ node->getLayout()->getID().name, pos->second->getPath(), node->getPath());
+ exit(1);
+ }
+ }
+ else
+ fNodes[node->getLayout()->getID().name] = node;
+ // update fAliasesMap with symlinks found
+ const char* aliasPath = realPath;
+ if ( (fgFileSystemRoot != NULL) && (fgFileSystemRoot[0] != '\0') && (strncmp(realPath, fgFileSystemRoot, strlen(fgFileSystemRoot)) == 0) ) {
+ aliasPath = &realPath[strlen(fgFileSystemRoot)];
+ }
+ // <rdar://problem/11192810> Too many aliases in -overlay mode
+ for (std::vector<const char*>::const_iterator it=fgFileSystemOverlays.begin(); it != fgFileSystemOverlays.end(); ++it) {
+ const char* overlayPath = *it;
+ if ( strncmp(realPath, overlayPath, strlen(overlayPath)) == 0 ) {
+ aliasPath = &realPath[strlen(overlayPath)];
+ break;
+ }
+ }
+ if ( fAliasesMap.find(aliasPath) == fAliasesMap.end() ) {
+ if ( strcmp(aliasPath, node->getLayout()->getID().name) != 0 ) {
+ fAliasesMap[strdup(aliasPath)] = node->getLayout()->getID().name;
+ //fprintf(stderr, "getNode() %s: added alias %s -> %s\n", archName(fArchPair), aliasPath, fAliasesMap[aliasPath]);
+ }
+ }
}
return node;
}
@@ -292,7 +417,13 @@
// expand @executable_path path prefix
const char* executablePath = mainExecutableLayout->getFilePath();
char newPath[strlen(executablePath) + strlen(dependentPath)+2];
- strcpy(newPath, executablePath);
+ if ( (fgFileSystemRoot != NULL) && (strncmp(executablePath, fgFileSystemRoot, strlen(fgFileSystemRoot)) == 0) ) {
+ // executablePath already has rootPath prefix, need to remove that to get to base virtual path
+ strcpy(newPath, &executablePath[strlen(fgFileSystemRoot)]);
+ }
+ else {
+ strcpy(newPath, executablePath);
+ }
char* addPoint = strrchr(newPath,'/');
if ( addPoint != NULL )
strcpy(&addPoint[1], &dependentPath[17]);
@@ -303,7 +434,13 @@
else if ( strncmp(dependentPath, "@loader_path/", 13) == 0 ) {
// expand @loader_path path prefix
char newPath[strlen(fPath) + strlen(dependentPath)+2];
- strcpy(newPath, fPath);
+ if ( (fgFileSystemRoot != NULL) && (strncmp(fPath, fgFileSystemRoot, strlen(fgFileSystemRoot)) == 0) ) {
+ // fPath already has rootPath prefix, need to remove that to get to base virtual path
+ strcpy(newPath, &fPath[strlen(fgFileSystemRoot)]);
+ }
+ else {
+ strcpy(newPath, fPath);
+ }
char* addPoint = strrchr(newPath,'/');
if ( addPoint != NULL )
strcpy(&addPoint[1], &dependentPath[13]);
@@ -314,10 +451,18 @@
else if ( strncmp(dependentPath, "@rpath/", 7) == 0 ) {
throw "@rpath not supported in dyld shared cache";
}
- fDependsOn.insert(fGraph->getNodeForVirtualPath(dependentPath));
+ // <rdar://problem/9161945> silently ignore dependents from main executables that can't be in shared cache
+ bool addDependent = true;
+ if ( fLayout->getFileType() == MH_EXECUTE ) {
+ if ( (strncmp(dependentPath, "/usr/lib/", 9) != 0) && (strncmp(dependentPath, "/System/Library/", 16) != 0) ) {
+ addDependent = false;
+ }
+ }
+ if ( addDependent )
+ fDependsOn.insert(fGraph->getNodeForVirtualPath(dependentPath));
}
catch (const char* msg) {
- if ( it->weakImport && ! fLayout->hasSplitSegInfo() ) {
+ if ( it->weakImport || ! fLayout->hasSplitSegInfo() ) {
// ok to ignore missing weak imported dylibs from things that are
// not going to be in the dyld shared cache
}
@@ -345,46 +490,117 @@
}
+
ArchGraph::DependencyNode::DependencyNode(ArchGraph* graph, const char* path, const MachOLayoutAbstraction* layout)
: fGraph(graph), fPath(strdup(path)), fLayout(layout), fDependenciesLoaded(false), fDependentMissing(false)
{
//fprintf(stderr, "new DependencyNode(0x%08X, %s)\n", graph->fArch, path);
}
+uint64_t ArchGraph::maxCacheSizeForArchPair(ArchPair ap) {
+ switch ( ap.arch ) {
+ case CPU_TYPE_I386:
+ return 0x20000000;
+ case CPU_TYPE_X86_64:
+ return 0x40000000;
+ case CPU_TYPE_ARM:
+ return ARM_SHARED_REGION_SIZE;
+ case CPU_TYPE_ARM64:
+ return ARM64_SHARED_REGION_SIZE;
+ default: return UINT64_MAX;
+ }
+}
+
void ArchGraph::findSharedDylibs(ArchPair ap)
{
const PathToNode& nodes = fgPerArchGraph[ap]->fNodes;
std::set<const MachOLayoutAbstraction*> possibleLibs;
- //fprintf(stderr, "shared for arch 0x%08X\n", arch);
+ std::map<const MachOLayoutAbstraction*, const DependencyNode *> layoutToNode;
+ //fprintf(stderr, "shared for arch %s\n", archName(ap));
for(PathToNode::const_iterator it = nodes.begin(); it != nodes.end(); ++it) {
DependencyNode* node = it->second;
// <rdar://problem/6127437> put all dylibs in shared cache - not just ones used by more than one app
if ( node->allDependentsFound() /*&& (node->useCount() > 1)*/ ) {
const MachOLayoutAbstraction* layout = node->getLayout();
- if ( layout->hasSplitSegInfo() && layout->isRootOwned() && layout->inSharableLocation() )
- possibleLibs.insert(layout);
- //fprintf(stderr, "\t%s\n", it->first);
+ if ( layout->isDylib() ) {
+ char* msg;
+ if ( sharable(layout, ap, &msg) ) {
+ possibleLibs.insert(layout);
+ layoutToNode[layout] = node;
+ }
+ else {
+ if ( !iPhoneOS && (layout->getID().name[0] == '@') ) {
+ // <rdar://problem/7770139> update_dyld_shared_cache should suppress warnings for embedded frameworks
+ }
+ else {
+ warnings.push_back(msg);
+ fprintf(stderr, "update_dyld_shared_cache: for arch %s, %s\n", archName(ap), msg);
+ }
+ }
+ }
}
}
// prune so that all shareable libs depend only on other shareable libs
std::set<const MachOLayoutAbstraction*>& sharedLibs = fgPerArchGraph[ap]->fSharedDylibs;
std::map<const MachOLayoutAbstraction*,bool> shareableMap;
+ uint64_t totalLibSize = 0;
for (std::set<const MachOLayoutAbstraction*>::iterator lit = possibleLibs.begin(); lit != possibleLibs.end(); ++lit) {
- if ( canBeShared(*lit, ap, possibleLibs, shareableMap) )
+ if ( canBeShared(*lit, ap, possibleLibs, shareableMap) ) {
+ totalLibSize += (*lit)->getVMSize();
sharedLibs.insert(*lit);
- }
+ }
+ }
+
+#if 0 // disable auto-eviction because it happens before linkedit optimization which means it is overly conservative.
+
+ // Check to see if the unoptimized cache size is too large, if so trim out some libraries
+ uint64_t maxCacheSize = maxCacheSizeForArchPair(ap);
+ if (totalLibSize > maxCacheSize) {
+ fprintf(stderr, "update_dyld_shared_cache: unoptimized %s shared cache overflow, total VM space: %lldMB (max=%lldMB)\n", archName(ap), totalLibSize/(1024*1024), maxCacheSize/(1024*1024));
+ std::vector<const MachOLayoutAbstraction*> removableLibs;
+
+ for (const MachOLayoutAbstraction* layout : sharedLibs) {
+ // Every library uses itself, and every MH_DYLIB has an extra useCount, so we know useCount of 2 implies nothing else in the shared cache uses it
+ if (layoutToNode[layout]->useCount() == 2) {
+ if ( layoutToNode[layout]->dependsOnDylibList() ) {
+ removableLibs.push_back(layout);
+ //fprintf(stderr, " possible to evict: %s\n", layout->getID().name);
+ }
+ }
+ }
+
+ std::sort(removableLibs.begin(), removableLibs.end(), [](const MachOLayoutAbstraction* a, const MachOLayoutAbstraction* b){
+ return a->getVMSize() < b->getVMSize();
+ });
+
+ while ( (totalLibSize > maxCacheSize) && !removableLibs.empty() ) {
+ const MachOLayoutAbstraction* largestRemovableLib = removableLibs.back();
+ removableLibs.pop_back();
+ if ( largestRemovableLib->getVMSize() > 1024*1024 )
+ fprintf(stderr, "update_dyld_shared_cache: evicting % 3lldMB leaf dylib %s\n", largestRemovableLib->getVMSize()/(1024*1024), largestRemovableLib->getID().name);
+ else
+ fprintf(stderr, "update_dyld_shared_cache: evicting % 3lldKB leaf dylib %s\n", largestRemovableLib->getVMSize()/1024, largestRemovableLib->getID().name);
+ sharedLibs.erase(largestRemovableLib);
+ totalLibSize -= largestRemovableLib->getVMSize();
+ }
+ fprintf(stderr, "update_dyld_shared_cache: unoptimized %s shared cache reduced to total VM space: %lldMB\n", archName(ap), totalLibSize/1024/1024);
+ }
+#endif
}
const char* ArchGraph::archName(ArchPair ap)
{
switch ( ap.arch ) {
- case CPU_TYPE_POWERPC:
- return "ppc";
case CPU_TYPE_I386:
return "i386";
case CPU_TYPE_X86_64:
- return "x86_64";
+ switch ( ap.subtype ) {
+ case CPU_SUBTYPE_X86_64_H:
+ return "x86_64h";
+ default:
+ return "x86_64";
+ }
case CPU_TYPE_ARM:
switch ( ap.subtype ) {
case CPU_SUBTYPE_ARM_V4T:
@@ -397,12 +613,40 @@
return "arm-xscale";
case CPU_SUBTYPE_ARM_V7:
return "armv7";
+ case CPU_SUBTYPE_ARM_V7F:
+ return "armv7f";
+ case CPU_SUBTYPE_ARM_V7K:
+ return "armv7k";
+ case CPU_SUBTYPE_ARM_V7S:
+ return "armv7s";
default:
return "arm";
}
+ case CPU_TYPE_ARM64:
+ return "arm64";
default:
return "unknown";
}
+}
+
+bool ArchGraph::sharable(const MachOLayoutAbstraction* layout, ArchPair ap, char** msg)
+{
+ int trustErr = layout->notTrusted();
+ if ( ! layout->isTwoLevelNamespace() )
+ asprintf(msg, "can't put %s in shared cache because it was built -flat_namespace", layout->getID().name);
+ else if ( ! layout->inSharableLocation() )
+ asprintf(msg, "can't put %s in shared cache because its -install_name is not in /usr/lib or /System/Library", layout->getID().name);
+ else if ( ! layout->hasSplitSegInfo() )
+ asprintf(msg, "can't put %s in shared cache because it was not built for %s or later", layout->getID().name, (iPhoneOS ? "iPhoneOS 3.1" : "MacOSX 10.5"));
+ else if ( rootless == true && trustErr != 0 )
+ asprintf(msg, "can't put %s in shared cache because it is not trusted: %s", layout->getFilePath(), strerror(trustErr));
+ else if ( layout->hasDynamicLookupLinkage() )
+ asprintf(msg, "can't put %s in shared cache because it was built with '-undefined dynamic_lookup'", layout->getID().name);
+ else if ( layout->hasMainExecutableLookupLinkage() )
+ asprintf(msg, "can't put %s in shared cache because it was built with '-bundle_loader'", layout->getID().name);
+ else
+ return true;
+ return false;
}
bool ArchGraph::canBeShared(const MachOLayoutAbstraction* layout, ArchPair ap, const std::set<const MachOLayoutAbstraction*>& possibleLibs, std::map<const MachOLayoutAbstraction*, bool>& shareableMap)
@@ -416,14 +660,8 @@
if ( possibleLibs.count(layout) == 0 ) {
shareableMap[layout] = false;
char* msg;
- if ( ! layout->hasSplitSegInfo() )
- asprintf(&msg, "can't put %s in shared cache because it was not built for 10.5 or later", layout->getID().name);
- else if ( ! layout->isRootOwned() )
- asprintf(&msg, "can't put %s in shared cache because it is not owned by root", layout->getID().name);
- else if ( ! layout->inSharableLocation() )
- asprintf(&msg, "can't put %s in shared cache because it is not in /usr/lib or /System/Library", layout->getID().name);
- else
- asprintf(&msg, "can't put %s in shared cache", layout->getID().name);
+ if ( sharable(layout, ap, &msg) )
+ asprintf(&msg, "can't put %s in shared cache, unknown reason", layout->getID().name);
warnings.push_back(msg);
if ( verbose )
fprintf(stderr, "update_dyld_shared_cache: for arch %s, %s\n", archName(ap), msg);
@@ -442,6 +680,9 @@
if ( nodes.find(realPath) != nodes.end() )
continue;
}
+ // handle weak imported dylibs not found
+ if ( dit->weakImport )
+ continue;
shareableMap[layout] = false;
char* msg;
asprintf(&msg, "can't put %s in shared cache because it depends on %s which can't be found", layout->getID().name, dit->name);
@@ -466,48 +707,144 @@
}
+
+class StringPool
+{
+public:
+ StringPool();
+ const char* getBuffer();
+ uint32_t size();
+ uint32_t add(const char* str);
+ uint32_t addUnique(const char* str);
+ const char* stringAtIndex(uint32_t) const;
+
+private:
+ typedef std::unordered_map<const char*, uint32_t, CStringHash, CStringEquals> StringToOffset;
+
+ char* fBuffer;
+ uint32_t fBufferAllocated;
+ uint32_t fBufferUsed;
+ StringToOffset fUniqueStrings;
+};
+
+
+StringPool::StringPool()
+ : fBufferUsed(0), fBufferAllocated(128*1024*1024)
+{
+ fBuffer = (char*)malloc(fBufferAllocated);
+}
+
+uint32_t StringPool::add(const char* str)
+{
+ uint32_t len = strlen(str);
+ if ( (fBufferUsed + len + 1) > fBufferAllocated ) {
+ // grow buffer
+ throw "string buffer exhausted";
+ }
+ strcpy(&fBuffer[fBufferUsed], str);
+ uint32_t result = fBufferUsed;
+ fUniqueStrings[&fBuffer[fBufferUsed]] = result;
+ fBufferUsed += len+1;
+ return result;
+}
+
+uint32_t StringPool::addUnique(const char* str)
+{
+ StringToOffset::iterator pos = fUniqueStrings.find(str);
+ if ( pos != fUniqueStrings.end() )
+ return pos->second;
+ else {
+ //fprintf(stderr, "StringPool::addUnique() new string: %s\n", str);
+ return this->add(str);
+ }
+}
+
+uint32_t StringPool::size()
+{
+ return fBufferUsed;
+}
+
+const char* StringPool::getBuffer()
+{
+ return fBuffer;
+}
+
+const char* StringPool::stringAtIndex(uint32_t index) const
+{
+ return &fBuffer[index];
+}
+
+
+
+struct LocalSymbolInfo
+{
+ uint32_t dylibOffset;
+ uint32_t nlistStartIndex;
+ uint32_t nlistCount;
+};
+
+
template <typename A>
class SharedCache
{
public:
- SharedCache(ArchGraph* graph, const char* rootPath, bool alphaSort, bool verify, bool optimize, bool overlay, uint64_t dyldBaseAddress);
- bool update(bool usesOverlay, bool force, bool optimize, bool deleteExistingFirst, int archIndex,
- int archCount, bool keepSignatures);
+ SharedCache(ArchGraph* graph, const char* rootPath, const std::vector<const char*>& overlayPaths, const char* cacheDir, bool explicitCacheDir,
+ bool alphaSort, bool verify, bool optimize, uint64_t dyldBaseAddress);
+ bool update(bool force, bool optimize, bool deleteExistingFirst, int archIndex,
+ int archCount, bool keepSignatures, bool dontMapLocalSymbols);
+ void writeCacheFile(const char *cacheFilePath, uint8_t *cacheFileBuffer, uint32_t cacheFileSize, bool deleteOldCache);
static const char* cacheFileSuffix(bool optimized, const char* archName);
- uint64_t mappedCacheAddressForAddress(uint64_t addr);
+ // vm address = address AS WRITTEN into the cache
+ // mapped address = address AS MAPPED into the update process only
+ // file offset = offset relative to start of cache file
+ void * mappedAddressForVMAddress(uint64_t vmaddr);
+ uint64_t VMAddressForMappedAddress(const void *mapaddr);
+ uint64_t cacheFileOffsetForVMAddress(uint64_t addr) const;
+ uint64_t VMAddressForCacheFileOffset(uint64_t addr) const;
+
+ static const char* archName();
private:
typedef typename A::P P;
typedef typename A::P::E E;
typedef typename A::P::uint_t pint_t;
- bool notUpToDate(const char* path);
- bool notUpToDate(const void* cache);
- uint8_t* optimizeLINKEDIT(bool keepSignatures);
- void optimizeObjC();
+ bool notUpToDate(const char* path, unsigned int aliasCount);
+ bool notUpToDate(const void* cache, unsigned int aliasCount);
+ uint8_t* optimizeLINKEDIT(bool keepSignatures, bool dontMapLocalSymbols);
+ void optimizeObjC(std::vector<void*>& pointersInData);
static void getSharedCacheBasAddresses(cpu_type_t arch, uint64_t* baseReadOnly, uint64_t* baseWritable);
static cpu_type_t arch();
- static const char* archName();
- static uint64_t sharedRegionReadOnlyStartAddress();
- static uint64_t sharedRegionWritableStartAddress();
- static uint64_t sharedRegionReadOnlySize();
- static uint64_t sharedRegionWritableSize();
+ static uint64_t sharedRegionStartAddress();
+ static uint64_t sharedRegionSize();
+ static uint64_t sharedRegionStartWritableAddress(uint64_t);
+ static uint64_t sharedRegionStartReadOnlyAddress(uint64_t, uint64_t);
static uint64_t getWritableSegmentNewAddress(uint64_t proposedNewAddress, uint64_t originalAddress, uint64_t executableSlide);
-
-
- void assignNewBaseAddresses();
- uint64_t cacheFileOffsetForAddress(uint64_t addr);
+ static bool addCacheSlideInfo();
+ static uint64_t pathHash(const char*);
+
+ static uint64_t pageAlign(uint64_t addr);
+ static uint64_t regionAlign(uint64_t addr);
+ static uint64_t pageAlign4KB(uint64_t addr);
+ void assignNewBaseAddresses(bool verify);
struct LayoutInfo {
const MachOLayoutAbstraction* layout;
+ std::vector<const char*> aliases;
dyld_cache_image_info info;
};
struct ByNameSorter {
bool operator()(const LayoutInfo& left, const LayoutInfo& right)
{ return (strcmp(left.layout->getID().name, right.layout->getID().name) < 0); }
+ };
+
+ struct ByAddressSorter {
+ bool operator()(const LayoutInfo& left, const LayoutInfo& right) {
+ return (left.layout->getSegments()[0].newAddress() < right.layout->getSegments()[0].newAddress());
+ }
};
struct ByCStringSectionSizeSorter {
@@ -520,8 +857,8 @@
// one image has no segments
return segs_l.size() > segs_r.size();
}
- const macho_header<P> *mh_l = (const macho_header<P>*)segs_l[0].mappedAddress();
- const macho_header<P> *mh_r = (const macho_header<P>*)segs_r[0].mappedAddress();
+ const macho_header<P> *mh_l = (macho_header<P>*)segs_l[0].mappedAddress();
+ const macho_header<P> *mh_r = (macho_header<P>*)segs_r[0].mappedAddress();
const macho_section<P> *cstring_l = mh_l->getSection("__TEXT", "__cstring");
const macho_section<P> *cstring_r = mh_r->getSection("__TEXT", "__cstring");
if (!cstring_l || !cstring_r) {
@@ -542,14 +879,19 @@
std::map<const MachOLayoutAbstraction*, uint32_t>& fMap;
};
-
+
ArchGraph* fArchGraph;
const bool fVerify;
bool fExistingIsNotUpToDate;
+ bool fCacheFileInFinalLocation;
const char* fCacheFilePath;
uint8_t* fExistingCacheForVerification;
std::vector<LayoutInfo> fDylibs;
+ std::vector<LayoutInfo> fDylibAliases;
std::vector<shared_file_mapping_np> fMappings;
+ std::vector<macho_nlist<P> > fUnmappedLocalSymbols;
+ StringPool fUnmappedLocalsStringPool;
+ std::vector<LocalSymbolInfo> fLocalSymbolInfos;
uint32_t fHeaderSize;
uint8_t* fInMemoryCache;
uint64_t fDyldBaseAddress;
@@ -561,7 +903,19 @@
uint32_t fOffsetOfLazyBindInfoInCombinedLinkedit;
uint32_t fOffsetOfExportInfoInCombinedLinkedit;
uint32_t fOffsetOfOldSymbolTableInfoInCombinedLinkedit;
+ uint32_t fSizeOfOldSymbolTableInfoInCombinedLinkedit;
+ uint32_t fOffsetOfOldExternalRelocationsInCombinedLinkedit;
+ uint32_t fSizeOfOldExternalRelocationsInCombinedLinkedit;
+ uint32_t fOffsetOfOldIndirectSymbolsInCombinedLinkedit;
+ uint32_t fSizeOfOldIndirectSymbolsInCombinedLinkedit;
+ uint32_t fOffsetOfOldStringPoolInCombinedLinkedit;
+ uint32_t fSizeOfOldStringPoolInCombinedLinkedit;
+ uint32_t fOffsetOfFunctionStartsInCombinedLinkedit;
+ uint32_t fSizeOfFunctionStartsInCombinedLinkedit;
+ uint32_t fOffsetOfDataInCodeInCombinedLinkedit;
+ uint32_t fSizeOfDataInCodeInCombinedLinkedit;
uint32_t fLinkEditsTotalOptimizedSize;
+ uint32_t fUnmappedLocalSymbolsSize;
};
@@ -572,35 +926,49 @@
typedef typename A::P P;
typedef typename A::P::uint_t pint_t;
- SharedCache<A>* const fCache;
- const macho_section<P>* const fSection;
- pint_t * const fBase;
- uint64_t const fCount;
+ SharedCache<A>* const fCache;
+ const macho_section<P>* const fSection;
+ pint_t * const fBase;
+ pint_t fCount;
public:
PointerSection(SharedCache<A>* cache, const macho_header<P>* header,
const char *segname, const char *sectname)
: fCache(cache)
, fSection(header->getSection(segname, sectname))
- , fBase(fSection ? (pint_t *)cache->mappedCacheAddressForAddress(fSection->addr()) : 0)
+ , fBase(fSection ? (pint_t *)cache->mappedAddressForVMAddress(fSection->addr()) : 0)
, fCount(fSection ? fSection->size() / sizeof(pint_t) : 0)
{
}
- uint64_t count() const { return fCount; }
-
- uint64_t getUnmapped(uint64_t index) const {
+ pint_t count() const { return fCount; }
+
+ pint_t getVMAddress(pint_t index) const {
if (index >= fCount) throwf("index out of range");
return P::getP(fBase[index]);
}
- T get(uint64_t index) const {
- return (T)fCache->mappedCacheAddressForAddress(getUnmapped(index));
+ T get(pint_t index) const {
+ return (T)fCache->mappedAddressForVMAddress(getVMAddress(index));
}
- void set(uint64_t index, uint64_t value) {
+ void setVMAddress(pint_t index, pint_t value) {
if (index >= fCount) throwf("index out of range");
P::setP(fBase[index], value);
+ }
+
+ void removeNulls() {
+ pint_t shift = 0;
+ for (pint_t i = 0; i < fCount; i++) {
+ pint_t value = fBase[i];
+ if (value) {
+ fBase[i-shift] = value;
+ } else {
+ shift++;
+ }
+ }
+ fCount -= shift;
+ const_cast<macho_section<P>*>(fSection)->set_size(fCount * sizeof(pint_t));
}
};
@@ -620,7 +988,7 @@
const char *segname, const char *sectname)
: fCache(cache)
, fSection(header->getSection(segname, sectname))
- , fBase(fSection ? (T *)cache->mappedCacheAddressForAddress(fSection->addr()) : 0)
+ , fBase(fSection ? (T *)cache->mappedAddressForVMAddress(fSection->addr()) : 0)
, fCount(fSection ? fSection->size() / sizeof(T) : 0)
{
}
@@ -640,79 +1008,153 @@
-template <> cpu_type_t SharedCache<ppc>::arch() { return CPU_TYPE_POWERPC; }
template <> cpu_type_t SharedCache<x86>::arch() { return CPU_TYPE_I386; }
template <> cpu_type_t SharedCache<x86_64>::arch() { return CPU_TYPE_X86_64; }
template <> cpu_type_t SharedCache<arm>::arch() { return CPU_TYPE_ARM; }
-
-template <> uint64_t SharedCache<ppc>::sharedRegionReadOnlyStartAddress() { return 0x90000000; }
-template <> uint64_t SharedCache<x86>::sharedRegionReadOnlyStartAddress() { return 0x90000000; }
-template <> uint64_t SharedCache<x86_64>::sharedRegionReadOnlyStartAddress() { return 0x7FFF80000000LL; }
-template <> uint64_t SharedCache<arm>::sharedRegionReadOnlyStartAddress() { return 0x30000000; }
-
-template <> uint64_t SharedCache<ppc>::sharedRegionWritableStartAddress() { return 0xA0000000; }
-template <> uint64_t SharedCache<x86>::sharedRegionWritableStartAddress() { return 0xA0000000; }
-template <> uint64_t SharedCache<x86_64>::sharedRegionWritableStartAddress() { return 0x7FFF70000000LL; }
-template <> uint64_t SharedCache<arm>::sharedRegionWritableStartAddress() { return 0x38000000; }
-
-template <> uint64_t SharedCache<ppc>::sharedRegionReadOnlySize() { return 0x10000000; }
-template <> uint64_t SharedCache<x86>::sharedRegionReadOnlySize() { return 0x10000000; }
-template <> uint64_t SharedCache<x86_64>::sharedRegionReadOnlySize() { return 0x7FE00000; }
-template <> uint64_t SharedCache<arm>::sharedRegionReadOnlySize() { return 0x08000000; }
-
-template <> uint64_t SharedCache<ppc>::sharedRegionWritableSize() { return 0x10000000; }
-template <> uint64_t SharedCache<x86>::sharedRegionWritableSize() { return 0x10000000; }
-template <> uint64_t SharedCache<x86_64>::sharedRegionWritableSize() { return 0x20000000; }
-template <> uint64_t SharedCache<arm>::sharedRegionWritableSize() { return 0x08000000; }
-
-
-template <> const char* SharedCache<ppc>::archName() { return "ppc"; }
+template <> cpu_type_t SharedCache<arm64>::arch() { return CPU_TYPE_ARM64; }
+
+template <> uint64_t SharedCache<x86>::sharedRegionStartAddress() { return 0x90000000; }
+template <> uint64_t SharedCache<x86_64>::sharedRegionStartAddress() { return 0x7FFF80000000LL; }
+template <> uint64_t SharedCache<arm>::sharedRegionStartAddress() { return ARM_SHARED_REGION_START; }
+template <> uint64_t SharedCache<arm64>::sharedRegionStartAddress() { return ARM64_SHARED_REGION_START; }
+
+template <> uint64_t SharedCache<x86>::sharedRegionSize() { return 0x20000000; }
+template <> uint64_t SharedCache<x86_64>::sharedRegionSize() { return 0x40000000; }
+template <> uint64_t SharedCache<arm>::sharedRegionSize() { return ARM_SHARED_REGION_SIZE; }
+template <> uint64_t SharedCache<arm64>::sharedRegionSize() { return ARM64_SHARED_REGION_SIZE; }
+
+template <> uint64_t SharedCache<x86>::sharedRegionStartWritableAddress(uint64_t exEnd) { return exEnd + 0x04000000; }
+template <> uint64_t SharedCache<x86_64>::sharedRegionStartWritableAddress(uint64_t exEnd) { return 0x7FFF70000000LL; }
+template <> uint64_t SharedCache<arm>::sharedRegionStartWritableAddress(uint64_t exEnd) { return (exEnd + 16383) & (-16384); }
+template <> uint64_t SharedCache<arm64>::sharedRegionStartWritableAddress(uint64_t exEnd) { return exEnd; }
+
+template <> uint64_t SharedCache<x86>::sharedRegionStartReadOnlyAddress(uint64_t wrEnd, uint64_t exEnd) { return wrEnd + 0x04000000; }
+template <> uint64_t SharedCache<x86_64>::sharedRegionStartReadOnlyAddress(uint64_t wrEnd, uint64_t exEnd){ return exEnd; }
+template <> uint64_t SharedCache<arm>::sharedRegionStartReadOnlyAddress(uint64_t wrEnd, uint64_t exEnd) { return (wrEnd + 16383) & (-16384); }
+template <> uint64_t SharedCache<arm64>::sharedRegionStartReadOnlyAddress(uint64_t wrEnd, uint64_t exEnd) { return (wrEnd + 16383) & (-16384); }
+
template <> const char* SharedCache<x86>::archName() { return "i386"; }
template <> const char* SharedCache<x86_64>::archName() { return "x86_64"; }
template <> const char* SharedCache<arm>::archName() { return "arm"; }
-
-template <> const char* SharedCache<ppc>::cacheFileSuffix(bool optimized, const char*) { return optimized ? "ppc" : "rosetta"; }
+template <> const char* SharedCache<arm64>::archName() { return "arm64"; }
+
template <> const char* SharedCache<x86>::cacheFileSuffix(bool, const char* archName) { return archName; }
template <> const char* SharedCache<x86_64>::cacheFileSuffix(bool, const char* archName){ return archName; }
template <> const char* SharedCache<arm>::cacheFileSuffix(bool, const char* archName) { return archName; }
+template <> const char* SharedCache<arm64>::cacheFileSuffix(bool, const char* archName) { return archName; }
+
+template <> uint64_t SharedCache<x86>::pageAlign(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
+template <> uint64_t SharedCache<x86_64>::pageAlign(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
+template <> uint64_t SharedCache<arm>::pageAlign(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
+template <> uint64_t SharedCache<arm64>::pageAlign(uint64_t addr) { return ( (addr + 16383) & (-16384) ); }
+
+template <> uint64_t SharedCache<x86>::regionAlign(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
+template <> uint64_t SharedCache<x86_64>::regionAlign(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
+template <> uint64_t SharedCache<arm>::regionAlign(uint64_t addr) { return ( (addr + 16383) & (-16384) ); }
+template <> uint64_t SharedCache<arm64>::regionAlign(uint64_t addr) { return ( (addr + 16383) & (-16384) ); }
+
+
+template <typename A>
+uint64_t SharedCache<A>::pageAlign4KB(uint64_t addr) { return ( (addr + 4095) & (-4096) ); }
template <typename A>
-SharedCache<A>::SharedCache(ArchGraph* graph, const char* rootPath, bool alphaSort, bool verify, bool optimize, bool overlay, uint64_t dyldBaseAddress)
- : fArchGraph(graph), fVerify(verify), fExistingIsNotUpToDate(true), fCacheFilePath(NULL),
- fExistingCacheForVerification(NULL), fDyldBaseAddress(dyldBaseAddress)
+SharedCache<A>::SharedCache(ArchGraph* graph, const char* rootPath, const std::vector<const char*>& overlayPaths, const char* cacheDir, bool explicitCacheDir, bool alphaSort, bool verify, bool optimize, uint64_t dyldBaseAddress)
+ : fArchGraph(graph), fVerify(verify), fExistingIsNotUpToDate(true),
+ fCacheFileInFinalLocation(rootPath[0] == '\0'), fCacheFilePath(NULL),
+ fExistingCacheForVerification(NULL), fDyldBaseAddress(dyldBaseAddress),
+ fOffsetOfBindInfoInCombinedLinkedit(0), fOffsetOfWeakBindInfoInCombinedLinkedit(0),
+ fOffsetOfLazyBindInfoInCombinedLinkedit(0), fOffsetOfExportInfoInCombinedLinkedit(0),
+ fOffsetOfOldSymbolTableInfoInCombinedLinkedit(0), fSizeOfOldSymbolTableInfoInCombinedLinkedit(0),
+ fOffsetOfOldExternalRelocationsInCombinedLinkedit(0), fSizeOfOldExternalRelocationsInCombinedLinkedit(0),
+ fOffsetOfOldIndirectSymbolsInCombinedLinkedit(0), fSizeOfOldIndirectSymbolsInCombinedLinkedit(0),
+ fOffsetOfOldStringPoolInCombinedLinkedit(0), fSizeOfOldStringPoolInCombinedLinkedit(0),
+ fOffsetOfFunctionStartsInCombinedLinkedit(0), fSizeOfFunctionStartsInCombinedLinkedit(0),
+ fOffsetOfDataInCodeInCombinedLinkedit(0), fSizeOfDataInCodeInCombinedLinkedit(0),
+ fUnmappedLocalSymbolsSize(0)
{
if ( fArchGraph->getArchPair().arch != arch() )
throwf("SharedCache object is wrong architecture: 0x%08X vs 0x%08X", fArchGraph->getArchPair().arch, arch());
// build vector of all shared dylibs
+ unsigned int aliasCount = 0;
std::set<const MachOLayoutAbstraction*>& dylibs = fArchGraph->getSharedDylibs();
+ ArchGraph::StringToString& aliases = fArchGraph->getDylibAliases();
for(std::set<const MachOLayoutAbstraction*>::iterator it = dylibs.begin(); it != dylibs.end(); ++it) {
const MachOLayoutAbstraction* lib = *it;
LayoutInfo temp;
temp.layout = lib;
temp.info.address = 0;
+ temp.info.inode = lib->getInode();
temp.info.modTime = lib->getLastModTime();
- temp.info.inode = lib->getInode();
- temp.info.pathFileOffset = lib->getNameFileOffset();
+ if ( iPhoneOS ) {
+ temp.info.inode = pathHash(lib->getID().name);
+ temp.info.modTime = 0;
+ }
+ temp.info.pathFileOffset = lib->getNameFileOffset(); // for now this is the offset within the dylib
+ for(ArchGraph::StringToString::iterator ait = aliases.begin(); ait != aliases.end(); ++ait) {
+ if ( strcmp(ait->second, lib->getID().name) == 0 ) {
+ temp.aliases.push_back(ait->first);
+ ++aliasCount;
+ }
+ }
fDylibs.push_back(temp);
}
- // examine the existing shared cache file
- char cachePath[1024];
- strcpy(cachePath, rootPath);
- strcat(cachePath, DYLD_SHARED_CACHE_DIR);
- strcat(cachePath, DYLD_SHARED_CACHE_BASE_NAME);
- strcat(cachePath, cacheFileSuffix(optimize, fArchGraph->archName()));
- fCacheFilePath = strdup(cachePath);
- const char* pathToExistingCacheFile = fCacheFilePath;
- char cachePathNonOverlay[1024];
- if ( overlay ) {
- strcpy(cachePathNonOverlay, DYLD_SHARED_CACHE_DIR);
- strcat(cachePathNonOverlay, DYLD_SHARED_CACHE_BASE_NAME);
- strcat(cachePathNonOverlay, cacheFileSuffix(optimize, fArchGraph->archName()));
- pathToExistingCacheFile = cachePathNonOverlay;
- }
- fExistingIsNotUpToDate = this->notUpToDate(pathToExistingCacheFile);
+ // create path to cache file
+ char cachePathCanonical[MAXPATHLEN];
+ strcpy(cachePathCanonical, cacheDir);
+ if ( cachePathCanonical[strlen(cachePathCanonical)-1] != '/' )
+ strcat(cachePathCanonical, "/");
+ strcat(cachePathCanonical, DYLD_SHARED_CACHE_BASE_NAME);
+ strcat(cachePathCanonical, cacheFileSuffix(optimize, fArchGraph->archName()));
+ char cachePath[MAXPATHLEN];
+ if ( explicitCacheDir ) {
+ fCacheFilePath = strdup(cachePathCanonical);
+ }
+ else if ( overlayPaths.size() == 1 ) {
+ // if no -cache_dir and exactly on -overlay, write cache file into that overlay dir
+ strcpy(cachePath, overlayPaths[0]);
+ strcat(cachePath, "/");
+ strcat(cachePath, cachePathCanonical);
+ fCacheFilePath = strdup(cachePath);
+ }
+ else if ( rootPath[0] != '\0' ) {
+ strcpy(cachePath, rootPath);
+ strcat(cachePath, "/");
+ strcat(cachePath, cachePathCanonical);
+ fCacheFilePath = strdup(cachePath);
+ }
+ else {
+ fCacheFilePath = strdup(cachePathCanonical);
+ }
+
+ // If the path we are writing to is trusted then our sources need to be trusted
+ // <rdar://problem/21166835> Can't update the update_dyld_shared_cache on a non-boot volume
+ rootless = rootless_check_trusted(fCacheFilePath);
+
+ if ( overlayPaths.size() == 1 ) {
+ // in overlay mode if there already is a cache file in the overlay,
+ // check if it is up to date.
+ struct stat stat_buf;
+ if ( stat(fCacheFilePath, &stat_buf) == 0 ) {
+ fExistingIsNotUpToDate = this->notUpToDate(fCacheFilePath, aliasCount);
+ }
+ else if ( rootPath[0] != '\0' ) {
+ // using -root and -overlay, but no cache file in overlay, check one in -root
+ char cachePathRoot[MAXPATHLEN];
+ strcpy(cachePathRoot, rootPath);
+ strcat(cachePathRoot, "/");
+ strcat(cachePathRoot, cachePathCanonical);
+ fExistingIsNotUpToDate = this->notUpToDate(cachePathRoot, aliasCount);
+ }
+ else {
+ // uisng -overlay, but no cache file in overlay, check one in boot volume
+ fExistingIsNotUpToDate = this->notUpToDate(cachePathCanonical, aliasCount);
+ }
+ }
+ else {
+ fExistingIsNotUpToDate = this->notUpToDate(fCacheFilePath, aliasCount);
+ }
// sort shared dylibs
if ( verify ) {
@@ -730,33 +1172,56 @@
}
// assign segments in each dylib a new address
- this->assignNewBaseAddresses();
+ this->assignNewBaseAddresses(verify);
+
+ // calculate where string pool offset will start
+ // calculate cache file header size
+ fHeaderSize = sizeof(dyld_cache_header)
+ + fMappings.size()*sizeof(shared_file_mapping_np)
+ + (fDylibs.size()+aliasCount)*sizeof(dyld_cache_image_info);
+ const uint64_t baseHeaderSize = fHeaderSize;
+ //fprintf(stderr, "aliasCount=%d, fHeaderSize=0x%08X\n", aliasCount, fHeaderSize);
+ // build list of aliases and compute where each ones path string will go
+ for(typename std::vector<struct LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
+ for(std::vector<const char*>::const_iterator ait = it->aliases.begin(); ait != it->aliases.end(); ++ait) {
+ LayoutInfo temp = *it;
+ // alias looks just like real dylib, but has a different name string
+ const char* aliasPath = *ait;
+ temp.aliases.clear();
+ temp.aliases.push_back(aliasPath);
+ temp.info.pathFileOffset = fHeaderSize;
+ if ( iPhoneOS ) {
+ temp.info.inode = pathHash(aliasPath);
+ temp.info.modTime = 0;
+ }
+ fDylibAliases.push_back(temp);
+ fHeaderSize += strlen(aliasPath)+1;
+ }
+ }
+ std::sort(fDylibAliases.begin(), fDylibAliases.end(), ByNameSorter());
+ //fprintf(stderr, "fHeaderSize=0x%08X, fDylibAliases.size()=%lu\n", fHeaderSize, fDylibAliases.size());
+ fHeaderSize = pageAlign(fHeaderSize);
// check that cache we are about to create for verification purposes has same layout as existing cache
if ( verify ) {
// if no existing cache, say so
if ( fExistingCacheForVerification == NULL ) {
throwf("update_dyld_shared_cache[%u] for arch=%s, could not verify because cache file does not exist in /var/db/dyld/\n",
- getpid(), archName());
+ getpid(), fArchGraph->archName());
}
const dyldCacheHeader<E>* header = (dyldCacheHeader<E>*)fExistingCacheForVerification;
const dyldCacheImageInfo<E>* cacheEntry = (dyldCacheImageInfo<E>*)(fExistingCacheForVerification + header->imagesOffset());
for(typename std::vector<LayoutInfo>::iterator it = fDylibs.begin(); it != fDylibs.end(); ++it, ++cacheEntry) {
if ( cacheEntry->address() != it->layout->getSegments()[0].newAddress() ) {
throwf("update_dyld_shared_cache[%u] warning: for arch=%s, could not verify cache because start address of %s is 0x%llX in cache, but should be 0x%llX\n",
- getpid(), archName(), it->layout->getID().name, cacheEntry->address(), it->layout->getSegments()[0].newAddress());
- }
- }
- }
-
- // calculate cache file header size
- fHeaderSize = pageAlign(sizeof(dyld_cache_header)
- + fMappings.size()*sizeof(shared_file_mapping_np)
- + fDylibs.size()*sizeof(dyld_cache_image_info) );
- //+ fDependencyPool.size()*sizeof(uint16_t));
-
- if ( fHeaderSize > 0x3000 )
- throwf("header size miscalculation 0x%08X", fHeaderSize);
+ getpid(), fArchGraph->archName(), it->layout->getID().name, cacheEntry->address(), it->layout->getSegments()[0].newAddress());
+ }
+ }
+ }
+
+
+ if ( fHeaderSize > FIRST_DYLIB_TEXT_OFFSET )
+ throwf("header size overflow: allowed=0x%08X, base=0x%08llX, aliases=0x%08llX", FIRST_DYLIB_TEXT_OFFSET, baseHeaderSize, fHeaderSize-baseHeaderSize);
}
@@ -766,106 +1231,159 @@
return proposedNewAddress;
}
-template <>
-uint64_t SharedCache<ppc>::getWritableSegmentNewAddress(uint64_t proposedNewAddress, uint64_t originalAddress, uint64_t executableSlide)
-{
- // for ppc writable segments can only move in increments of 64K (so only hi16 instruction needs to be modified)
- return (((executableSlide & 0x000000000000F000ULL) - ((proposedNewAddress - originalAddress) & 0x000000000000F000ULL)) & 0x000000000000F000ULL) + proposedNewAddress;
-}
-
-
template <typename A>
-void SharedCache<A>::assignNewBaseAddresses()
-{
- // first layout TEXT and DATA for split-seg (or can be split-seg) dylibs
- uint64_t currentExecuteAddress = sharedRegionReadOnlyStartAddress() + 0x3000;
- uint64_t currentWritableAddress = sharedRegionWritableStartAddress();
+uint64_t SharedCache<A>::pathHash(const char* path)
+{
+ uint64_t sum = 0;
+ for (const char* s=path; *s != '\0'; ++s)
+ sum += sum*4 + *s;
+ return sum;
+}
+
+
+template <typename A>
+void SharedCache<A>::assignNewBaseAddresses(bool verify)
+{
+ // first layout TEXT for dylibs
+ const uint64_t startExecuteAddress = sharedRegionStartAddress();
+ uint64_t currentExecuteAddress = startExecuteAddress + FIRST_DYLIB_TEXT_OFFSET;
for(typename std::vector<LayoutInfo>::iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
std::vector<MachOLayoutAbstraction::Segment>& segs = ((MachOLayoutAbstraction*)(it->layout))->getSegments();
- MachOLayoutAbstraction::Segment* executableSegment = NULL;
for (int i=0; i < segs.size(); ++i) {
MachOLayoutAbstraction::Segment& seg = segs[i];
seg.reset();
+ if ( seg.executable() && !seg.writable() ) {
+ // <rdar://problem/15947734> Some dylib require extra alignment
+ currentExecuteAddress = (currentExecuteAddress + seg.alignment() - 1) & (-seg.alignment());
+ // __TEXT segment
+ if ( it->info.address == 0 )
+ it->info.address = currentExecuteAddress;
+ seg.setNewAddress(currentExecuteAddress);
+ currentExecuteAddress += pageAlign(seg.size());
+ }
+ }
+ }
+ // align __TEXT region
+ currentExecuteAddress = regionAlign(currentExecuteAddress);
+
+#define DENSE_PACK 0
+ // layout __DATA* segments
+ std::vector<MachOLayoutAbstraction::Segment*> dataSegs;
+ std::vector<MachOLayoutAbstraction::Segment*> dataConstSegs;
+ std::vector<MachOLayoutAbstraction::Segment*> dataDirtySegs;
+ const uint64_t startWritableAddress = sharedRegionStartWritableAddress(currentExecuteAddress);
+ uint64_t currentWritableAddress = startWritableAddress;
+ for (const LayoutInfo& info : fDylibs ) {
+ for (MachOLayoutAbstraction::Segment& seg : ((MachOLayoutAbstraction*)(info.layout))->getSegments()) {
if ( seg.writable() ) {
- if ( seg.executable() && it->layout->hasSplitSegInfo() ) {
- // skip __IMPORT segments in this pass
- }
- else {
- // __DATA segment
- // for ppc, writable segments have to move in 64K increments
- if ( it->layout->hasSplitSegInfo() ) {
- if ( executableSegment == NULL )
- throwf("first segment in dylib is not executable for %s", it->layout->getID().name);
- seg.setNewAddress(getWritableSegmentNewAddress(currentWritableAddress, seg.address(), executableSegment->newAddress() - executableSegment->address()));
- }
- else
- seg.setNewAddress(currentWritableAddress);
- currentWritableAddress = pageAlign(seg.newAddress() + seg.size());
- }
- }
- else {
- if ( seg.executable() ) {
- // __TEXT segment
- if ( it->info.address == 0 )
- it->info.address = currentExecuteAddress;
- executableSegment = &seg;
- seg.setNewAddress(currentExecuteAddress);
- currentExecuteAddress += pageAlign(seg.size());
- }
- else {
- // skip read-only segments in this pass
- }
- }
- }
- }
-
- // append all read-only (but not LINKEDIT) segments at end of all TEXT segments
- // append all IMPORT segments at end of all DATA segments rounded to next 2MB
- uint64_t currentReadOnlyAddress = currentExecuteAddress;
- uint64_t startWritableExecutableAddress = (currentWritableAddress + 0x200000 - 1) & (-0x200000);
- uint64_t currentWritableExecutableAddress = startWritableExecutableAddress;
+ if ( seg.executable() )
+ throw "found writable and executable segment";
+ seg.reset();
+ if ( strcmp(seg.name(), "__DATA_CONST") == 0 )
+ dataConstSegs.push_back(&seg);
+ else if ( strcmp(seg.name(), "__DATA_DIRTY") == 0 )
+ dataDirtySegs.push_back(&seg);
+ else
+ dataSegs.push_back(&seg);
+ }
+ }
+ }
+ // coalesce all __DATA_CONST segments
+ for (MachOLayoutAbstraction::Segment* seg : dataConstSegs) {
+ #if DENSE_PACK
+ // start segment at needed alignment
+ currentWritableAddress = (currentWritableAddress + seg->sectionsAlignment() - 1) & (-seg->sectionsAlignment());
+ seg->setNewAddress(currentWritableAddress);
+ // pack together
+ uint64_t justSectionsSize = seg->sectionsSize();
+ currentWritableAddress = seg->newAddress() + justSectionsSize;
+ seg->setSize(justSectionsSize);
+ if ( seg->fileSize() > justSectionsSize )
+ seg->setFileSize(justSectionsSize);
+ #else
+ seg->setNewAddress(currentWritableAddress);
+ // pack to 4KB pages
+ currentWritableAddress = pageAlign4KB(seg->newAddress() + seg->size());
+ #endif
+ }
+ #if DENSE_PACK
+ currentWritableAddress = pageAlign4KB(currentWritableAddress);
+ #endif
+ // coalesce all __DATA segments
+ for (MachOLayoutAbstraction::Segment* seg : dataSegs) {
+ #if DENSE_PACK
+ // start segment at needed alignment
+ currentWritableAddress = (currentWritableAddress + seg->sectionsAlignment() - 1) & (-seg->sectionsAlignment());
+ seg->setNewAddress(currentWritableAddress);
+ // pack together
+ uint64_t justSectionsSize = seg->sectionsSize();
+ currentWritableAddress = seg->newAddress() + justSectionsSize;
+ seg->setSize(justSectionsSize);
+ if ( seg->fileSize() > justSectionsSize )
+ seg->setFileSize(justSectionsSize);
+ #else
+ seg->setNewAddress(currentWritableAddress);
+ // pack to 4KB pages
+ currentWritableAddress = pageAlign4KB(seg->newAddress() + seg->size());
+ #endif
+ }
+ #if DENSE_PACK
+ currentWritableAddress = pageAlign4KB(currentWritableAddress);
+ #endif
+ // coalesce all __DATA_DIRTY segments
+ for (MachOLayoutAbstraction::Segment* seg : dataDirtySegs) {
+ // start segment at needed alignment
+ currentWritableAddress = (currentWritableAddress + seg->sectionsAlignment() - 1) & (-seg->sectionsAlignment());
+ seg->setNewAddress(currentWritableAddress);
+ // pack together
+ uint64_t justSectionsSize = seg->sectionsSize();
+ currentWritableAddress = seg->newAddress() + justSectionsSize;
+ seg->setSize(justSectionsSize);
+ if ( seg->fileSize() > justSectionsSize )
+ seg->setFileSize(justSectionsSize);
+ }
+ // align __DATA region
+ currentWritableAddress = regionAlign(currentWritableAddress);
+
+ // layout all read-only (but not LINKEDIT) segments
+ const uint64_t startReadOnlyAddress = sharedRegionStartReadOnlyAddress(currentWritableAddress, currentExecuteAddress);
+ uint64_t currentReadOnlyAddress = startReadOnlyAddress;
for(typename std::vector<LayoutInfo>::iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
std::vector<MachOLayoutAbstraction::Segment>& segs = ((MachOLayoutAbstraction*)(it->layout))->getSegments();
for(int i=0; i < segs.size(); ++i) {
MachOLayoutAbstraction::Segment& seg = segs[i];
- if ( !seg.writable() && !seg.executable() && (strcmp(seg.name(), "__LINKEDIT") != 0) ) {
- // allocate non-executable,read-only segments from end of read only shared region
+ if ( seg.readable() && !seg.writable() && !seg.executable() && (strcmp(seg.name(), "__LINKEDIT") != 0) ) {
+ // __UNICODE segment
seg.setNewAddress(currentReadOnlyAddress);
currentReadOnlyAddress += pageAlign(seg.size());
}
- else if ( seg.writable() && seg.executable() && it->layout->hasSplitSegInfo() ) {
- // allocate IMPORT segments to end of writable shared region
- seg.setNewAddress(currentWritableExecutableAddress);
- currentWritableExecutableAddress += pageAlign(seg.size());
- }
}
}
- // append all LINKEDIT segments at end of all read-only segments
+ // layout all LINKEDIT segments at end of all read-only segments
+ currentReadOnlyAddress = regionAlign(currentReadOnlyAddress); // <rdar://problem/16491435>
fLinkEditsStartAddress = currentReadOnlyAddress;
fFirstLinkEditSegment = NULL;
for(typename std::vector<LayoutInfo>::iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
std::vector<MachOLayoutAbstraction::Segment>& segs = ((MachOLayoutAbstraction*)(it->layout))->getSegments();
for(int i=0; i < segs.size(); ++i) {
MachOLayoutAbstraction::Segment& seg = segs[i];
- if ( !seg.writable() && !seg.executable() && (strcmp(seg.name(), "__LINKEDIT") == 0) ) {
+ if ( seg.readable() && !seg.writable() && !seg.executable() && (strcmp(seg.name(), "__LINKEDIT") == 0) ) {
if ( fFirstLinkEditSegment == NULL )
fFirstLinkEditSegment = &seg;
- // allocate non-executable,read-only segments from end of read only shared region
seg.setNewAddress(currentReadOnlyAddress);
currentReadOnlyAddress += pageAlign(seg.size());
}
}
}
- fLinkEditsTotalUnoptimizedSize = (currentReadOnlyAddress - fLinkEditsStartAddress + 4095) & (-4096);
-
+ fLinkEditsTotalUnoptimizedSize = pageAlign(currentReadOnlyAddress - fLinkEditsStartAddress);
// populate large mappings
uint64_t cacheFileOffset = 0;
- if ( currentExecuteAddress > sharedRegionReadOnlyStartAddress() + 0x3000 ) {
+ if ( currentExecuteAddress > startExecuteAddress ) {
shared_file_mapping_np executeMapping;
- executeMapping.sfm_address = sharedRegionReadOnlyStartAddress();
- executeMapping.sfm_size = currentExecuteAddress - sharedRegionReadOnlyStartAddress();
+ executeMapping.sfm_address = startExecuteAddress;
+ executeMapping.sfm_size = currentExecuteAddress - startExecuteAddress;
executeMapping.sfm_file_offset = cacheFileOffset;
executeMapping.sfm_max_prot = VM_PROT_READ | VM_PROT_EXECUTE;
executeMapping.sfm_init_prot = VM_PROT_READ | VM_PROT_EXECUTE;
@@ -873,29 +1391,18 @@
cacheFileOffset += executeMapping.sfm_size;
shared_file_mapping_np writableMapping;
- writableMapping.sfm_address = sharedRegionWritableStartAddress();
- writableMapping.sfm_size = currentWritableAddress - sharedRegionWritableStartAddress();
+ writableMapping.sfm_address = startWritableAddress;
+ writableMapping.sfm_size = currentWritableAddress - startWritableAddress;
writableMapping.sfm_file_offset = cacheFileOffset;
writableMapping.sfm_max_prot = VM_PROT_READ | VM_PROT_WRITE;
writableMapping.sfm_init_prot = VM_PROT_READ | VM_PROT_WRITE;
fMappings.push_back(writableMapping);
cacheFileOffset += writableMapping.sfm_size;
-
- if ( currentWritableExecutableAddress > startWritableExecutableAddress ) {
- shared_file_mapping_np writableExecutableMapping;
- writableExecutableMapping.sfm_address = startWritableExecutableAddress;
- writableExecutableMapping.sfm_size = currentWritableExecutableAddress - startWritableExecutableAddress;
- writableExecutableMapping.sfm_file_offset= cacheFileOffset;
- writableExecutableMapping.sfm_max_prot = VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
- writableExecutableMapping.sfm_init_prot = VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
- fMappings.push_back(writableExecutableMapping);
- cacheFileOffset += writableExecutableMapping.sfm_size;
- }
-
+
// make read-only (contains LINKEDIT segments) last, so it can be cut back when optimized
shared_file_mapping_np readOnlyMapping;
- readOnlyMapping.sfm_address = currentExecuteAddress;
- readOnlyMapping.sfm_size = currentReadOnlyAddress - currentExecuteAddress;
+ readOnlyMapping.sfm_address = startReadOnlyAddress;
+ readOnlyMapping.sfm_size = currentReadOnlyAddress - startReadOnlyAddress;
readOnlyMapping.sfm_file_offset = cacheFileOffset;
readOnlyMapping.sfm_max_prot = VM_PROT_READ;
readOnlyMapping.sfm_init_prot = VM_PROT_READ;
@@ -905,8 +1412,8 @@
else {
// empty cache
shared_file_mapping_np cacheHeaderMapping;
- cacheHeaderMapping.sfm_address = sharedRegionWritableStartAddress();
- cacheHeaderMapping.sfm_size = 0x3000;
+ cacheHeaderMapping.sfm_address = startExecuteAddress;
+ cacheHeaderMapping.sfm_size = FIRST_DYLIB_TEXT_OFFSET;
cacheHeaderMapping.sfm_file_offset = cacheFileOffset;
cacheHeaderMapping.sfm_max_prot = VM_PROT_READ;
cacheHeaderMapping.sfm_init_prot = VM_PROT_READ;
@@ -917,26 +1424,43 @@
template <typename A>
-uint64_t SharedCache<A>::cacheFileOffsetForAddress(uint64_t addr)
-{
- for(std::vector<shared_file_mapping_np>::iterator it = fMappings.begin(); it != fMappings.end(); ++it) {
- if ( (it->sfm_address <= addr) && (addr < it->sfm_address+it->sfm_size) )
- return it->sfm_file_offset + addr - it->sfm_address;
- }
- throwf("address 0x%0llX is not in cache", addr);
-}
-
+uint64_t SharedCache<A>::cacheFileOffsetForVMAddress(uint64_t vmaddr) const
+{
+ for(std::vector<shared_file_mapping_np>::const_iterator it = fMappings.begin(); it != fMappings.end(); ++it) {
+ if ( (it->sfm_address <= vmaddr) && (vmaddr < it->sfm_address+it->sfm_size) )
+ return it->sfm_file_offset + vmaddr - it->sfm_address;
+ }
+ throwf("address 0x%0llX is not in cache", vmaddr);
+}
template <typename A>
-uint64_t SharedCache<A>::mappedCacheAddressForAddress(uint64_t addr)
-{
- if (!addr) return 0;
- else return (uint64_t)(fInMemoryCache + cacheFileOffsetForAddress(addr));
-}
-
+uint64_t SharedCache<A>::VMAddressForCacheFileOffset(uint64_t offset) const
+{
+ for(std::vector<shared_file_mapping_np>::const_iterator it = fMappings.begin(); it != fMappings.end(); ++it) {
+ if ( (it->sfm_file_offset <= offset) && (offset < it->sfm_file_offset+it->sfm_size) )
+ return it->sfm_address + offset - it->sfm_file_offset;
+ }
+ throwf("offset 0x%0llX is not in cache", offset);
+}
template <typename A>
-bool SharedCache<A>::notUpToDate(const void* cache)
+void *SharedCache<A>::mappedAddressForVMAddress(uint64_t vmaddr)
+{
+ if (!vmaddr) return NULL;
+ else return fInMemoryCache + cacheFileOffsetForVMAddress(vmaddr);
+}
+
+template <typename A>
+uint64_t SharedCache<A>::VMAddressForMappedAddress(const void *mapaddr)
+{
+ if (!mapaddr) return 0;
+ uint64_t offset = (uint8_t *)mapaddr - (uint8_t *)fInMemoryCache;
+ return VMAddressForCacheFileOffset(offset);
+}
+
+
+template <typename A>
+bool SharedCache<A>::notUpToDate(const void* cache, unsigned int aliasCount)
{
dyldCacheHeader<E>* header = (dyldCacheHeader<E>*)cache;
// not valid if header signature is wrong
@@ -946,25 +1470,29 @@
strcpy(&temp[15-strlen(archPairName)], archPairName);
if ( strcmp(header->magic(), temp) != 0 ) {
if ( fVerify ) {
- fprintf(stderr, "update_dyld_shared_cache[%u] cannot verify %s because current cache file has invalid header\n", getpid(), archName());
+ fprintf(stderr, "update_dyld_shared_cache[%u] cannot verify %s because current cache file has invalid header\n", getpid(), archPairName);
return false;
}
else {
- fprintf(stderr, "update_dyld_shared_cache[%u] current cache file has invalid header\n", getpid());
+ fprintf(stderr, "update_dyld_shared_cache[%u] updating cache because current cache file has invalid header\n", getpid());
return true;
}
}
// not valid if count of images does not match current images needed
- if ( header->imagesCount() != fDylibs.size() ) {
+ if ( header->imagesCount() != (fDylibs.size()+aliasCount) ) {
if ( fVerify ) {
- fprintf(stderr, "update_dyld_shared_cache[%u] cannot verify %s because current cache file contains a different set of dylibs\n", getpid(), archName());
+ fprintf(stderr, "update_dyld_shared_cache[%u] cannot verify %s because current cache file contains a different set of dylibs\n", getpid(), archPairName);
return false;
}
else {
- fprintf(stderr, "update_dyld_shared_cache[%u] current cache file is invalid because it contains a different set of dylibs\n", getpid());
+ fprintf(stderr, "update_dyld_shared_cache[%u] updating %s cache because current cache file contains a different set of dylibs\n", getpid(), archPairName);
return true;
}
}
+ // get end of TEXT region
+ const dyldCacheFileMapping<E>* textMapping = (dyldCacheFileMapping<E>*)((uint8_t*)cache+sizeof(dyldCacheHeader<E>));
+ const uint32_t textSize = textMapping->size();
+
// verify every dylib in constructed graph is in existing cache with same inode and modTime
std::map<const MachOLayoutAbstraction*, uint32_t> sortingMap;
const dyldCacheImageInfo<E>* imagesStart = (dyldCacheImageInfo<E>*)((uint8_t*)cache + header->imagesOffset());
@@ -974,18 +1502,26 @@
//fprintf(stderr, "inode=0x%llX, mTime=0x%llX, path=%s\n", it->info.inode, it->info.modTime, it->layout->getID().name);
for(const dyldCacheImageInfo<E>* cacheEntry = imagesStart; cacheEntry < imagesEnd; ++cacheEntry) {
if ( fVerify ) {
+ if ( cacheEntry->pathFileOffset() > textSize ) {
+ throwf("update_dyld_shared_cache[%u]: for arch=%s, image entries corrupt, bad path offset in %s\n",
+ getpid(), archPairName, it->layout->getID().name);
+ }
// in -verify mode, just match by path and warn if file looks different
if ( strcmp((char*)cache+cacheEntry->pathFileOffset(), it->layout->getID().name) == 0 ) {
found = true;
sortingMap[it->layout] = cacheEntry-imagesStart;
if ( (cacheEntry->inode() != it->info.inode) || (cacheEntry->modTime() != it->info.modTime) ) {
fprintf(stderr, "update_dyld_shared_cache[%u] warning: for arch=%s, %s has changed since cache was built\n",
- getpid(), archName(), it->layout->getID().name);
+ getpid(), archPairName, it->layout->getID().name);
}
break;
}
}
else {
+ if ( cacheEntry->pathFileOffset() > textSize ) {
+ // cache corrupt, needs to be regenerated
+ return true;
+ }
// in normal update mode, everything has to match for cache to be up-to-date
if ( (cacheEntry->inode() == it->info.inode)
&& (cacheEntry->modTime() == it->info.modTime)
@@ -997,10 +1533,10 @@
}
if ( !found ) {
if ( fVerify ) {
- throwf("update_dyld_shared_cache[%u] can't verify %s cache because %s is not in existing cache\n", getpid(), archName(), it->layout->getID().name);
+ throwf("update_dyld_shared_cache[%u] can't verify %s cache because %s is not in existing cache\n", getpid(), archPairName, it->layout->getID().name);
}
else {
- fprintf(stderr, "update_dyld_shared_cache[%u] current %s cache file invalid because %s has changed\n", getpid(), archName(), it->layout->getID().name);
+ fprintf(stderr, "update_dyld_shared_cache[%u] updating %s cache because dylib at %s has changed\n", getpid(), archPairName, it->layout->getID().name);
return true;
}
}
@@ -1024,7 +1560,7 @@
template <typename A>
-bool SharedCache<A>::notUpToDate(const char* path)
+bool SharedCache<A>::notUpToDate(const char* path, unsigned int aliasCount)
{
// mmap existing cache file
int fd = ::open(path, O_RDONLY);
@@ -1032,114 +1568,56 @@
return true;
struct stat stat_buf;
::fstat(fd, &stat_buf);
- uint8_t* mappingAddr = (uint8_t*)mmap(NULL, stat_buf.st_size, PROT_READ , MAP_FILE | MAP_PRIVATE, fd, 0);
+ uint32_t cacheFileSize = stat_buf.st_size;
+ uint32_t cacheAllocatedSize = pageAlign(cacheFileSize);
+ uint8_t* mappingAddr = NULL;
+ if ( vm_allocate(mach_task_self(), (vm_address_t*)(&mappingAddr), cacheAllocatedSize, VM_FLAGS_ANYWHERE) != KERN_SUCCESS )
+ throwf("can't vm_allocate cache of size %u", cacheFileSize);
+ // <rdar://problem/8960832> update_dyld_shared_cache -verify finds differences
+ (void)fcntl(fd, F_NOCACHE, 1);
+ ssize_t readResult = pread(fd, mappingAddr, cacheFileSize, 0);
+ if ( readResult != cacheFileSize )
+ throwf("can't read all of existing cache file (%lu of %u): %s", readResult, cacheFileSize, path);
::close(fd);
- if ( mappingAddr == (uint8_t*)(-1) )
- return true;
// validate it
- bool result = this->notUpToDate(mappingAddr);
+ bool result = this->notUpToDate(mappingAddr, aliasCount);
if ( fVerify ) {
// don't unmap yet, leave so it can be verified later
fExistingCacheForVerification = mappingAddr;
}
else {
// unmap
- ::munmap(mappingAddr, stat_buf.st_size);
+ vm_deallocate(mach_task_self(), (vm_address_t)mappingAddr, cacheAllocatedSize);
if ( verbose && !result )
fprintf(stderr, "update_dyld_shared_cache: %s is up-to-date\n", path);
}
return result;
}
-class CStringEquals
-{
-public:
- bool operator()(const char* left, const char* right) const { return (strcmp(left, right) == 0); }
-};
-
-class StringPool
-{
-public:
- StringPool();
- const char* getBuffer();
- uint32_t size();
- uint32_t add(const char* str);
- uint32_t addUnique(const char* str);
- const char* stringAtIndex(uint32_t) const;
-private:
- typedef __gnu_cxx::hash_map<const char*, uint32_t, __gnu_cxx::hash<const char*>, CStringEquals> StringToOffset;
-
- char* fBuffer;
- uint32_t fBufferAllocated;
- uint32_t fBufferUsed;
- StringToOffset fUniqueStrings;
-};
-
-
-StringPool::StringPool()
- : fBufferUsed(0), fBufferAllocated(32*1024*1024)
-{
- fBuffer = (char*)malloc(fBufferAllocated);
-}
-
-uint32_t StringPool::add(const char* str)
-{
- uint32_t len = strlen(str);
- if ( (fBufferUsed + len + 1) > fBufferAllocated ) {
- // grow buffer
- throw "string buffer exhausted";
- }
- strcpy(&fBuffer[fBufferUsed], str);
- uint32_t result = fBufferUsed;
- fUniqueStrings[&fBuffer[fBufferUsed]] = result;
- fBufferUsed += len+1;
- return result;
-}
-
-uint32_t StringPool::addUnique(const char* str)
-{
- StringToOffset::iterator pos = fUniqueStrings.find(str);
- if ( pos != fUniqueStrings.end() )
- return pos->second;
- else {
- //fprintf(stderr, "StringPool::addUnique() new string: %s\n", str);
- return this->add(str);
- }
-}
-
-uint32_t StringPool::size()
-{
- return fBufferUsed;
-}
-
-const char* StringPool::getBuffer()
-{
- return fBuffer;
-}
-
-const char* StringPool::stringAtIndex(uint32_t index) const
-{
- return &fBuffer[index];
-}
template <typename A>
class LinkEditOptimizer
{
public:
- LinkEditOptimizer(const MachOLayoutAbstraction&, uint8_t*, StringPool&);
+ LinkEditOptimizer(const MachOLayoutAbstraction&, const SharedCache<A>&, uint8_t*, StringPool&);
virtual ~LinkEditOptimizer() {}
void copyBindInfo(uint32_t&);
void copyWeakBindInfo(uint32_t&);
void copyLazyBindInfo(uint32_t&);
void copyExportInfo(uint32_t&);
- void copyLocalSymbols(uint32_t symbolTableOffset, uint32_t&);
+ void copyLocalSymbols(uint32_t symbolTableOffset, uint32_t&, bool dontMapLocalSymbols,
+ uint8_t* cacheStart, StringPool& unmappedLocalsStringPool,
+ std::vector<macho_nlist<typename A::P> >& unmappedSymbols,
+ std::vector<LocalSymbolInfo>& info);
void copyExportedSymbols(uint32_t symbolTableOffset, uint32_t&);
void copyImportedSymbols(uint32_t symbolTableOffset, uint32_t&);
void copyExternalRelocations(uint32_t& offset);
void copyIndirectSymbolTable(uint32_t& offset);
+ void copyFunctionStarts(uint32_t& offset);
+ void copyDataInCode(uint32_t& offset);
void updateLoadCommands(uint64_t newVMAddress, uint64_t size, uint32_t stringPoolOffset,
uint32_t linkEditsFileOffset, bool keepSignatures);
@@ -1151,12 +1629,15 @@
private:
+ const SharedCache<A>& fSharedCache;
const macho_header<P>* fHeader;
uint8_t* fNewLinkEditStart;
uint8_t* fLinkEditBase;
const MachOLayoutAbstraction& fLayout;
macho_dyld_info_command<P>* fDyldInfo;
macho_dysymtab_command<P>* fDynamicSymbolTable;
+ macho_linkedit_data_command<P>* fFunctionStarts;
+ macho_linkedit_data_command<P>* fDataInCode;
macho_symtab_command<P>* fSymbolTableLoadCommand;
const macho_nlist<P>* fSymbolTable;
const char* fStrings;
@@ -1179,14 +1660,19 @@
uint32_t fImportedSymbolsCountInNewLinkEdit;
uint32_t fExternalRelocationsOffsetIntoNewLinkEdit;
uint32_t fIndirectSymbolTableOffsetInfoNewLinkEdit;
+ uint32_t fFunctionStartsOffsetInNewLinkEdit;
+ uint32_t fDataInCodeOffsetInNewLinkEdit;
+ uint32_t fUnmappedLocalSymbolsStartIndexInNewLinkEdit;
+ uint32_t fUnmappedLocalSymbolsCountInNewLinkEdit;
};
template <typename A>
-LinkEditOptimizer<A>::LinkEditOptimizer(const MachOLayoutAbstraction& layout, uint8_t* newLinkEdit, StringPool& stringPool)
- : fLayout(layout), fLinkEditBase(NULL), fNewLinkEditStart(newLinkEdit), fDyldInfo(NULL),
- fDynamicSymbolTable(NULL), fSymbolTableLoadCommand(NULL), fSymbolTable(NULL), fStrings(NULL), fNewStringPool(stringPool),
+LinkEditOptimizer<A>::LinkEditOptimizer(const MachOLayoutAbstraction& layout, const SharedCache<A>& sharedCache, uint8_t* newLinkEdit, StringPool& stringPool)
+ : fSharedCache(sharedCache), fLayout(layout), fLinkEditBase(NULL), fNewLinkEditStart(newLinkEdit), fDyldInfo(NULL),
+ fDynamicSymbolTable(NULL), fFunctionStarts(NULL), fDataInCode(NULL),
+ fSymbolTableLoadCommand(NULL), fSymbolTable(NULL), fStrings(NULL), fNewStringPool(stringPool),
fBindInfoOffsetIntoNewLinkEdit(0), fBindInfoSizeInNewLinkEdit(0),
fWeakBindInfoOffsetIntoNewLinkEdit(0), fWeakBindInfoSizeInNewLinkEdit(0),
fLazyBindInfoOffsetIntoNewLinkEdit(0), fLazyBindInfoSizeInNewLinkEdit(0),
@@ -1195,7 +1681,9 @@
fLocalSymbolsStartIndexInNewLinkEdit(0), fLocalSymbolsCountInNewLinkEdit(0),
fExportedSymbolsStartIndexInNewLinkEdit(0), fExportedSymbolsCountInNewLinkEdit(0),
fImportSymbolsStartIndexInNewLinkEdit(0), fImportedSymbolsCountInNewLinkEdit(0),
- fExternalRelocationsOffsetIntoNewLinkEdit(0), fIndirectSymbolTableOffsetInfoNewLinkEdit(0)
+ fExternalRelocationsOffsetIntoNewLinkEdit(0), fIndirectSymbolTableOffsetInfoNewLinkEdit(0),
+ fFunctionStartsOffsetInNewLinkEdit(0), fDataInCodeOffsetInNewLinkEdit(0),
+ fUnmappedLocalSymbolsStartIndexInNewLinkEdit(0), fUnmappedLocalSymbolsCountInNewLinkEdit(0)
{
fHeader = (const macho_header<P>*)fLayout.getSegments()[0].mappedAddress();
@@ -1228,6 +1716,11 @@
case LC_DYLD_INFO_ONLY:
fDyldInfo = (macho_dyld_info_command<P>*)cmd;
break;
+ case LC_FUNCTION_STARTS:
+ fFunctionStarts = (macho_linkedit_data_command<P>*)cmd;
+ case LC_DATA_IN_CODE:
+ fDataInCode = (macho_linkedit_data_command<P>*)cmd;
+ break;
}
cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
}
@@ -1290,35 +1783,55 @@
template <typename A>
void LinkEditOptimizer<A>::copyExportInfo(uint32_t& offset)
{
- if ( (fDyldInfo != NULL) && (fDyldInfo->export_off() != 0) ) {
+ if ( (fDyldInfo != NULL) && (fLayout.getDyldInfoExports() != NULL) ) {
fExportInfoOffsetIntoNewLinkEdit = offset;
fExportInfoSizeInNewLinkEdit = fDyldInfo->export_size();
- // warning, export_off is only 32-bits so if the trie grows it must be allocated with 32-bits of fLinkeditBase
- memcpy(fNewLinkEditStart+offset, fLinkEditBase+(int32_t)fDyldInfo->export_off(), fDyldInfo->export_size());
+ memcpy(fNewLinkEditStart+offset, fLayout.getDyldInfoExports(), fDyldInfo->export_size());
offset += fDyldInfo->export_size();
}
}
-
template <typename A>
-void LinkEditOptimizer<A>::copyLocalSymbols(uint32_t symbolTableOffset, uint32_t& symbolIndex)
+void LinkEditOptimizer<A>::copyLocalSymbols(uint32_t symbolTableOffset, uint32_t& symbolIndex, bool dontMapLocalSymbols, uint8_t* cacheStart,
+ StringPool& unmappedLocalsStringPool, std::vector<macho_nlist<P> >& unmappedSymbols,
+ std::vector<LocalSymbolInfo>& dylibInfos)
{
fLocalSymbolsStartIndexInNewLinkEdit = symbolIndex;
+ LocalSymbolInfo localInfo;
+ localInfo.dylibOffset = ((uint8_t*)fHeader) - cacheStart;
+ localInfo.nlistStartIndex = unmappedSymbols.size();
+ localInfo.nlistCount = 0;
fSymbolTableStartOffsetInNewLinkEdit = symbolTableOffset + symbolIndex*sizeof(macho_nlist<P>);
macho_nlist<P>* const newSymbolTableStart = (macho_nlist<P>*)(fNewLinkEditStart+symbolTableOffset);
const macho_nlist<P>* const firstLocal = &fSymbolTable[fDynamicSymbolTable->ilocalsym()];
const macho_nlist<P>* const lastLocal = &fSymbolTable[fDynamicSymbolTable->ilocalsym()+fDynamicSymbolTable->nlocalsym()];
uint32_t oldIndex = fDynamicSymbolTable->ilocalsym();
for (const macho_nlist<P>* entry = firstLocal; entry < lastLocal; ++entry, ++oldIndex) {
- if ( (entry->n_type() & N_TYPE) == N_SECT ) {
+ // <rdar://problem/12237639> don't copy stab symbols
+ if ( (entry->n_sect() != NO_SECT) && ((entry->n_type() & N_STAB) == 0) ) {
+ const char* name = &fStrings[entry->n_strx()];
macho_nlist<P>* newSymbolEntry = &newSymbolTableStart[symbolIndex];
*newSymbolEntry = *entry;
- newSymbolEntry->set_n_strx(fNewStringPool.add(&fStrings[entry->n_strx()]));
- ++symbolIndex;
+ if ( dontMapLocalSymbols ) {
+ // if local in __text, add <redacted> symbol name to shared cache so backtraces don't have bogus names
+ if ( entry->n_sect() == 1 ) {
+ newSymbolEntry->set_n_strx(fNewStringPool.addUnique("<redacted>"));
+ ++symbolIndex;
+ }
+ // copy local symbol to unmmapped locals area
+ unmappedSymbols.push_back(*entry);
+ unmappedSymbols.back().set_n_strx(unmappedLocalsStringPool.addUnique(name));
+ }
+ else {
+ newSymbolEntry->set_n_strx(fNewStringPool.addUnique(name));
+ ++symbolIndex;
+ }
}
}
fLocalSymbolsCountInNewLinkEdit = symbolIndex - fLocalSymbolsStartIndexInNewLinkEdit;
+ localInfo.nlistCount = unmappedSymbols.size() - localInfo.nlistStartIndex;
+ dylibInfos.push_back(localInfo);
//fprintf(stderr, "%u locals starting at %u for %s\n", fLocalSymbolsCountInNewLinkEdit, fLocalSymbolsStartIndexInNewLinkEdit, fLayout.getFilePath());
}
@@ -1336,7 +1849,7 @@
&& (strncmp(&fStrings[entry->n_strx()], "$ld$", 4) != 0) ) {
macho_nlist<P>* newSymbolEntry = &newSymbolTableStart[symbolIndex];
*newSymbolEntry = *entry;
- newSymbolEntry->set_n_strx(fNewStringPool.add(&fStrings[entry->n_strx()]));
+ newSymbolEntry->set_n_strx(fNewStringPool.addUnique(&fStrings[entry->n_strx()]));
fOldToNewSymbolIndexes[oldIndex] = symbolIndex-fLocalSymbolsStartIndexInNewLinkEdit;
++symbolIndex;
}
@@ -1394,6 +1907,27 @@
offset += sizeof(macho_relocation_info<P>);
}
}
+
+template <typename A>
+void LinkEditOptimizer<A>::copyFunctionStarts(uint32_t& offset)
+{
+ if ( fFunctionStarts != NULL ) {
+ fFunctionStartsOffsetInNewLinkEdit = offset;
+ memcpy(&fNewLinkEditStart[offset], &fLinkEditBase[fFunctionStarts->dataoff()], fFunctionStarts->datasize());
+ offset += fFunctionStarts->datasize();
+ }
+}
+
+template <typename A>
+void LinkEditOptimizer<A>::copyDataInCode(uint32_t& offset)
+{
+ if ( fDataInCode != NULL ) {
+ fDataInCodeOffsetInNewLinkEdit = offset;
+ memcpy(&fNewLinkEditStart[offset], &fLinkEditBase[fDataInCode->dataoff()], fDataInCode->datasize());
+ offset += fDataInCode->datasize();
+ }
+}
+
template <typename A>
void LinkEditOptimizer<A>::copyIndirectSymbolTable(uint32_t& offset)
@@ -1415,7 +1949,7 @@
}
template <typename A>
-void LinkEditOptimizer<A>::updateLoadCommands(uint64_t newVMAddress, uint64_t size, uint32_t stringPoolOffset,
+void LinkEditOptimizer<A>::updateLoadCommands(uint64_t newVMAddress, uint64_t leSize, uint32_t stringPoolOffset,
uint32_t linkEditsFileOffset, bool keepSignatures)
{
// set LINKEDIT segment commmand to new merged LINKEDIT
@@ -1427,9 +1961,40 @@
macho_segment_command<P>* seg = (macho_segment_command<P>*)cmd;
if ( strcmp(seg->segname(), "__LINKEDIT") == 0 ) {
seg->set_vmaddr(newVMAddress);
- seg->set_vmsize(size);
- seg->set_filesize(size);
+ seg->set_vmsize(leSize);
+ seg->set_filesize(leSize);
seg->set_fileoff(linkEditsFileOffset);
+ }
+ else {
+ pint_t oldFileOff = seg->fileoff();
+ // don't alter __TEXT until <rdar://problem/7022345> is fixed
+ if ( strcmp(seg->segname(), "__TEXT") != 0 ) {
+ // update all other segments fileoff to be offset from start of cache file
+ seg->set_fileoff(fSharedCache.cacheFileOffsetForVMAddress(seg->vmaddr()));
+ }
+ pint_t fileOffsetDelta = seg->fileoff() - oldFileOff;
+ const MachOLayoutAbstraction::Segment* layoutSeg = fLayout.getSegment(seg->segname());
+ if ( layoutSeg != NULL ) {
+ //if ( seg->filesize() != layoutSeg->fileSize() ) {
+ // fprintf(stderr, "LC filesize=0x%08llX, trimmed seg file size=0x%08llX, seg=%s, path=%s\n",
+ // seg->filesize(), layoutSeg->fileSize(), seg->segname(), fLayout.getFilePath());
+ //}
+ //if ( seg->vmsize() != layoutSeg->size() ) {
+ // fprintf(stderr, "LC vmsize=0x%08llX, trimmed seg size=0x%08llX, seg=%s, path=%s\n",
+ // seg->vmsize(), layoutSeg->size(), seg->segname(), fLayout.getFilePath());
+ //}
+ seg->set_vmsize(layoutSeg->size());
+ seg->set_filesize(layoutSeg->fileSize());
+ }
+ // update all sections in this segment
+ macho_section<P>* const sectionsStart = (macho_section<P>*)((char*)seg + sizeof(macho_segment_command<P>));
+ macho_section<P>* const sectionsEnd = §ionsStart[seg->nsects()];
+ for(macho_section<P>* sect = sectionsStart; sect < sectionsEnd; ++sect) {
+ if ( sect->offset() != 0 )
+ sect->set_offset(sect->offset()+fileOffsetDelta);
+ //if ( (sect->flags() & SECTION_TYPE) == S_MOD_INIT_FUNC_POINTERS )
+ // fprintf(stderr, "found initializer(s) in %s\n", fLayout.getFilePath());
+ }
}
}
cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
@@ -1475,6 +2040,15 @@
fDynamicSymbolTable->set_locreloff(0);
fDynamicSymbolTable->set_nlocrel(0);
+ // update function starts
+ if ( fFunctionStarts != NULL ) {
+ fFunctionStarts->set_dataoff(linkEditsFileOffset+fFunctionStartsOffsetInNewLinkEdit);
+ }
+ // update data-in-code info
+ if ( fDataInCode != NULL ) {
+ fDataInCode->set_dataoff(linkEditsFileOffset+fDataInCodeOffsetInNewLinkEdit);
+ }
+
// now remove load commands no longer needed
const macho_load_command<P>* srcCmd = cmds;
macho_load_command<P>* dstCmd = (macho_load_command<P>*)cmds;
@@ -1483,6 +2057,8 @@
uint32_t cmdSize = srcCmd->cmdsize();
switch ( srcCmd->cmd() ) {
case LC_SEGMENT_SPLIT_INFO:
+ case LC_DYLIB_CODE_SIGN_DRS:
+ case LC_RPATH:
// don't copy
break;
case LC_CODE_SIGNATURE:
@@ -1516,7 +2092,7 @@
template <typename A>
-uint8_t* SharedCache<A>::optimizeLINKEDIT(bool keepSignatures)
+uint8_t* SharedCache<A>::optimizeLINKEDIT(bool keepSignatures, bool dontMapLocalSymbols)
{
// allocate space for optimized LINKEDIT area
uint8_t* newLinkEdit = new uint8_t[fLinkEditsTotalUnoptimizedSize];
@@ -1528,7 +2104,7 @@
// create optimizer object for each LINKEDIT segment
std::vector<LinkEditOptimizer<A>*> optimizers;
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
- optimizers.push_back(new LinkEditOptimizer<A>(*it->layout, newLinkEdit, stringPool));
+ optimizers.push_back(new LinkEditOptimizer<A>(*it->layout, *this, newLinkEdit, stringPool));
}
// rebase info is not copied because images in shared cache are never rebased
@@ -1562,38 +2138,60 @@
fOffsetOfOldSymbolTableInfoInCombinedLinkedit = offset;
uint32_t symbolTableOffset = offset;
uint32_t symbolTableIndex = 0;
+ if ( dontMapLocalSymbols )
+ fUnmappedLocalSymbols.reserve(16384);
for(typename std::vector<LinkEditOptimizer<A>*>::iterator it = optimizers.begin(); it != optimizers.end(); ++it) {
- (*it)->copyLocalSymbols(symbolTableOffset, symbolTableIndex);
+ (*it)->copyLocalSymbols(symbolTableOffset, symbolTableIndex, dontMapLocalSymbols, fInMemoryCache,
+ fUnmappedLocalsStringPool, fUnmappedLocalSymbols, fLocalSymbolInfos);
(*it)->copyExportedSymbols(symbolTableOffset, symbolTableIndex);
(*it)->copyImportedSymbols(symbolTableOffset, symbolTableIndex);
}
+ fSizeOfOldSymbolTableInfoInCombinedLinkedit = symbolTableIndex * sizeof(macho_nlist<typename A::P>);
+ offset = symbolTableOffset + fSizeOfOldSymbolTableInfoInCombinedLinkedit & (-8);
// copy external relocations, 8-byte aligned after end of symbol table
- uint32_t externalRelocsOffset = symbolTableOffset + (symbolTableIndex * sizeof(macho_nlist<typename A::P>) + 7) & (-8);
- //uint32_t externalRelocsStartOffset = externalRelocsOffset;
+ fOffsetOfOldExternalRelocationsInCombinedLinkedit = offset;
for(typename std::vector<LinkEditOptimizer<A>*>::iterator it = optimizers.begin(); it != optimizers.end(); ++it) {
- (*it)->copyExternalRelocations(externalRelocsOffset);
- }
-
+ (*it)->copyExternalRelocations(offset);
+ }
+ fSizeOfOldExternalRelocationsInCombinedLinkedit = offset - fOffsetOfOldExternalRelocationsInCombinedLinkedit;
+
+ // copy function starts
+ fOffsetOfFunctionStartsInCombinedLinkedit = offset;
+ for(typename std::vector<LinkEditOptimizer<A>*>::iterator it = optimizers.begin(); it != optimizers.end(); ++it) {
+ (*it)->copyFunctionStarts(offset);
+ }
+ fSizeOfFunctionStartsInCombinedLinkedit = offset - fOffsetOfFunctionStartsInCombinedLinkedit;
+
+ // copy data-in-code info
+ fOffsetOfDataInCodeInCombinedLinkedit = offset;
+ for(typename std::vector<LinkEditOptimizer<A>*>::iterator it = optimizers.begin(); it != optimizers.end(); ++it) {
+ (*it)->copyDataInCode(offset);
+ }
+ fSizeOfDataInCodeInCombinedLinkedit = offset - fOffsetOfDataInCodeInCombinedLinkedit;
+
// copy indirect symbol tables
- uint32_t indirectSymbolTableOffset = externalRelocsOffset;
+ fOffsetOfOldIndirectSymbolsInCombinedLinkedit = offset;
for(typename std::vector<LinkEditOptimizer<A>*>::iterator it = optimizers.begin(); it != optimizers.end(); ++it) {
- (*it)->copyIndirectSymbolTable(indirectSymbolTableOffset);
- }
-
+ (*it)->copyIndirectSymbolTable(offset);
+ }
+ fSizeOfOldIndirectSymbolsInCombinedLinkedit = offset - fOffsetOfOldIndirectSymbolsInCombinedLinkedit;
+
// copy string pool
- uint32_t stringPoolOffset = indirectSymbolTableOffset;
- memcpy(&newLinkEdit[stringPoolOffset], stringPool.getBuffer(), stringPool.size());
-
- // find new size
- fLinkEditsTotalOptimizedSize = (stringPoolOffset + stringPool.size() + 4095) & (-4096);
-
- // choose new linkedit file offset
- uint32_t linkEditsFileOffset = fLinkEditsStartAddress - sharedRegionReadOnlyStartAddress();
-
+ fOffsetOfOldStringPoolInCombinedLinkedit = offset;
+ memcpy(&newLinkEdit[offset], stringPool.getBuffer(), stringPool.size());
+ fSizeOfOldStringPoolInCombinedLinkedit = stringPool.size();
+
+ // total new size round up to page size
+ fLinkEditsTotalOptimizedSize = pageAlign(fOffsetOfOldStringPoolInCombinedLinkedit + fSizeOfOldStringPoolInCombinedLinkedit);
+
+ // choose new linkedit file offset
+ uint32_t linkEditsFileOffset = cacheFileOffsetForVMAddress(fLinkEditsStartAddress);
+// uint32_t linkEditsFileOffset = fLinkEditsStartAddress - sharedRegionStartAddress();
+
// update load commands so that all dylibs shared different areas of the same LINKEDIT segment
for(typename std::vector<LinkEditOptimizer<A>*>::iterator it = optimizers.begin(); it != optimizers.end(); ++it) {
- (*it)->updateLoadCommands(fLinkEditsStartAddress, fLinkEditsTotalUnoptimizedSize, stringPoolOffset, linkEditsFileOffset, keepSignatures);
+ (*it)->updateLoadCommands(fLinkEditsStartAddress, fLinkEditsTotalOptimizedSize, fOffsetOfOldStringPoolInCombinedLinkedit, linkEditsFileOffset, keepSignatures);
}
//fprintf(stderr, "fLinkEditsTotalUnoptimizedSize=%llu, fLinkEditsTotalOptimizedSize=%u\n", fLinkEditsTotalUnoptimizedSize, fLinkEditsTotalOptimizedSize);
@@ -1615,19 +2213,18 @@
seg.setFileOffset(linkEditsFileOffset);
}
}
- }
-
+ }
+
// return new end of cache
- return (uint8_t*)fFirstLinkEditSegment->mappedAddress() + fLinkEditsTotalOptimizedSize;
-}
-
+ return (uint8_t*)fFirstLinkEditSegment->mappedAddress() + regionAlign(fLinkEditsTotalOptimizedSize);
+}
template <typename A>
class ObjCSelectorUniquer
{
private:
- objc_selopt::string_map fSelectorStrings;
+ objc_opt::string_map fSelectorStrings;
SharedCache<A> *fCache;
size_t fCount;
@@ -1643,95 +2240,488 @@
{
fCount++;
const char *s = (const char *)
- fCache->mappedCacheAddressForAddress(oldValue);
- objc_selopt::string_map::iterator element =
- fSelectorStrings.insert(objc_selopt::string_map::value_type(s, oldValue)).first;
+ fCache->mappedAddressForVMAddress(oldValue);
+ objc_opt::string_map::iterator element =
+ fSelectorStrings.insert(objc_opt::string_map::value_type(s, oldValue)).first;
return (typename A::P::uint_t)element->second;
}
- objc_selopt::string_map& strings() {
+ objc_opt::string_map& strings() {
return fSelectorStrings;
}
size_t count() const { return fCount; }
};
-template <>
-void SharedCache<arm>::optimizeObjC()
-{
- // objc optimizations on arm not yet supported
-}
template <typename A>
-void SharedCache<A>::optimizeObjC()
-{
+class ClassListBuilder
+{
+private:
+ typedef typename A::P P;
+
+ objc_opt::string_map fClassNames;
+ objc_opt::class_map fClasses;
+ size_t fCount;
+ HeaderInfoOptimizer<A>& fHinfos;
+
+public:
+
+ ClassListBuilder(HeaderInfoOptimizer<A>& hinfos)
+ : fClassNames()
+ , fClasses()
+ , fCount(0)
+ , fHinfos(hinfos)
+ { }
+
+ void visitClass(SharedCache<A>* cache,
+ const macho_header<P>* header,
+ objc_class_t<A>* cls)
+ {
+ if (cls->isMetaClass(cache)) return;
+
+ const char *name = cls->getName(cache);
+ uint64_t name_vmaddr = cache->VMAddressForMappedAddress(name);
+ uint64_t cls_vmaddr = cache->VMAddressForMappedAddress(cls);
+ uint64_t hinfo_vmaddr = cache->VMAddressForMappedAddress(fHinfos.hinfoForHeader(cache, header));
+ fClassNames.insert(objc_opt::string_map::value_type(name, name_vmaddr));
+ fClasses.insert(objc_opt::class_map::value_type(name, std::pair<uint64_t, uint64_t>(cls_vmaddr, hinfo_vmaddr)));
+ fCount++;
+ }
+
+ objc_opt::string_map& classNames() {
+ return fClassNames;
+ }
+
+ objc_opt::class_map& classes() {
+ return fClasses;
+ }
+
+ size_t count() const { return fCount; }
+};
+
+
+template <typename A>
+class ProtocolOptimizer
+{
+private:
+ typedef typename A::P P;
+ typedef typename A::P::uint_t pint_t;
+
+ objc_opt::string_map fProtocolNames;
+ objc_opt::protocol_map fProtocols;
+ size_t fProtocolCount;
+ size_t fProtocolReferenceCount;
+
+ friend class ProtocolReferenceWalker<A, ProtocolOptimizer<A>>;
+ pint_t visitProtocolReference(SharedCache<A>* cache, pint_t oldValue)
+ {
+ objc_protocol_t<A>* proto = (objc_protocol_t<A>*)
+ cache->mappedAddressForVMAddress(oldValue);
+ pint_t newValue = fProtocols[proto->getName(cache)];
+ if (oldValue != newValue) fProtocolReferenceCount++;
+ return newValue;
+ }
+
+public:
+
+ ProtocolOptimizer()
+ : fProtocolNames()
+ , fProtocols()
+ , fProtocolCount(0)
+ , fProtocolReferenceCount(0)
+ { }
+
+ void addProtocols(SharedCache<A>* cache,
+ const macho_header<P>* header)
+ {
+ PointerSection<A, objc_protocol_t<A> *>
+ protocols(cache, header, "__DATA", "__objc_protolist");
+
+ for (pint_t i = 0; i < protocols.count(); i++) {
+ objc_protocol_t<A> *proto = protocols.get(i);
+
+ const char *name = proto->getName(cache);
+ if (fProtocolNames.count(name) == 0) {
+ // Need a Swift demangler API in OS before we can handle this
+ if (0 == strncmp(name, "_TtP", 4)) {
+ throw "objc protocol has Swift name";
+ }
+ if (proto->getSize() > sizeof(objc_protocol_t<A>)) {
+ throw "objc protocol is too big";
+ }
+
+ uint64_t name_vmaddr = cache->VMAddressForMappedAddress(name);
+ uint64_t proto_vmaddr = cache->VMAddressForMappedAddress(proto);
+ fProtocolNames.insert(objc_opt::string_map::value_type(name, name_vmaddr));
+ fProtocols.insert(objc_opt::protocol_map::value_type(name, proto_vmaddr));
+ fProtocolCount++;
+ }
+ }
+ }
+
+ const char *writeProtocols(SharedCache<A>* cache,
+ uint8_t *& dest, size_t& remaining,
+ std::vector<void*>& pointersInData,
+ pint_t protocolClassVMAddr)
+ {
+ if (fProtocolCount == 0) return NULL;
+
+ if (protocolClassVMAddr == 0) {
+ return "libobjc's Protocol class symbol not found (metadata not optimized)";
+ }
+
+ size_t required = fProtocolCount * sizeof(objc_protocol_t<A>);
+ if (remaining < required) {
+ return "libobjc's read-write section is too small (metadata not optimized)";
+ }
+
+ for (objc_opt::protocol_map::iterator iter = fProtocols.begin();
+ iter != fProtocols.end();
+ ++iter)
+ {
+ objc_protocol_t<A>* oldProto = (objc_protocol_t<A>*)
+ cache->mappedAddressForVMAddress(iter->second);
+
+ // Create a new protocol object.
+ objc_protocol_t<A>* proto = (objc_protocol_t<A>*)dest;
+ dest += sizeof(*proto);
+ remaining -= sizeof(*proto);
+
+ // Initialize it.
+ uint32_t oldSize = oldProto->getSize();
+ memcpy(proto, oldProto, oldSize);
+ if (!proto->getIsaVMAddr()) {
+ proto->setIsaVMAddr(protocolClassVMAddr);
+ }
+ if (oldSize < sizeof(*proto)) {
+ // Protocol object is old. Populate new fields.
+ proto->setSize(sizeof(objc_protocol_t<A>));
+ // missing extendedMethodTypes is already nil
+ }
+ // Some protocol objects are big enough to have the
+ // demangledName field but don't initialize it.
+ if (! proto->getDemangledName(cache)) {
+ proto->setDemangledName(cache, proto->getName(cache));
+ }
+ proto->setFixedUp();
+
+ // Redirect the protocol table at our new object.
+ iter->second = cache->VMAddressForMappedAddress(proto);
+
+ // Add new rebase entries.
+ proto->addPointers(pointersInData);
+ }
+
+ return NULL;
+ }
+
+ void updateReferences(SharedCache<A>* cache, const macho_header<P>* header)
+ {
+ ProtocolReferenceWalker<A, ProtocolOptimizer<A>> refs(*this);
+ refs.walk(cache, header);
+ }
+
+ objc_opt::string_map& protocolNames() {
+ return fProtocolNames;
+ }
+
+ objc_opt::protocol_map& protocols() {
+ return fProtocols;
+ }
+
+ size_t protocolCount() const { return fProtocolCount; }
+ size_t protocolReferenceCount() const { return fProtocolReferenceCount; }
+};
+
+
+static int percent(size_t num, size_t denom) {
+ if (denom) return (int)(num / (double)denom * 100);
+ else return 100;
+}
+
+template <typename A>
+void SharedCache<A>::optimizeObjC(std::vector<void*>& pointersInData)
+{
+ const char *err;
+
if ( verbose ) {
- fprintf(stderr, "update_dyld_shared_cache: for %s, uniquing objc selectors\n", archName());
+ fprintf(stderr, "update_dyld_shared_cache: for %s, optimizing objc metadata\n", archName());
}
- // Find libobjc's __TEXT,__objc_selopt section
- const macho_section<P> *seloptSection = NULL;
+ size_t headerSize = P::round_up(sizeof(objc_opt::objc_opt_t));
+ if (headerSize != sizeof(objc_opt::objc_opt_t)) {
+ warn(archName(), "libobjc's optimization structure size is wrong (metadata not optimized)");
+ }
+
+ // Find libobjc's empty sections to fill in.
+ // Find libobjc's list of pointers for us to use.
+ const macho_section<P> *optROSection = NULL;
+ const macho_section<P> *optRWSection = NULL;
+ const macho_section<P> *optPointerListSection = NULL;
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
- if (0 == strstr(it->layout->getFilePath(), "libobjc")) continue;
+ if ( strstr(it->layout->getFilePath(), "/libobjc.") != NULL ) {
+ const macho_header<P>* mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ optROSection = mh->getSection("__TEXT", "__objc_opt_ro");
+ optRWSection = mh->getSection("__DATA", "__objc_opt_rw");
+ optPointerListSection = mh->getSection("__DATA", "__objc_opt_ptrs");
+ break;
+ }
+ }
+
+ if ( optROSection == NULL ) {
+ warn(archName(), "libobjc's read-only section missing (metadata not optimized)");
+ return;
+ }
+
+ if ( optRWSection == NULL ) {
+ warn(archName(), "libobjc's read/write section missing (metadata not optimized)");
+ return;
+ }
+
+ if ( optPointerListSection == NULL ) {
+ warn(archName(), "libobjc's pointer list section missing (metadata not optimized)");
+ return;
+ }
+
+ uint8_t* optROData = (uint8_t*)mappedAddressForVMAddress(optROSection->addr());
+ size_t optRORemaining = optROSection->size();
+
+ uint8_t* optRWData = (uint8_t*)mappedAddressForVMAddress(optRWSection->addr());
+ size_t optRWRemaining = optRWSection->size();
+
+ if (optRORemaining < headerSize) {
+ warn(archName(), "libobjc's read-only section is too small (metadata not optimized)");
+ return;
+ }
+ objc_opt::objc_opt_t* optROHeader = (objc_opt::objc_opt_t *)optROData;
+ optROData += headerSize;
+ optRORemaining -= headerSize;
+
+ if (E::get32(optROHeader->version) != objc_opt::VERSION) {
+ warn(archName(), "libobjc's read-only section version is unrecognized (metadata not optimized)");
+ return;
+ }
+
+ if (optPointerListSection->size() < sizeof(objc_opt::objc_opt_pointerlist_tt<pint_t>)) {
+ warn(archName(), "libobjc's pointer list section is too small (metadata not optimized)");
+ return;
+ }
+ const objc_opt::objc_opt_pointerlist_tt<pint_t> *optPointerList = (const objc_opt::objc_opt_pointerlist_tt<pint_t> *)mappedAddressForVMAddress(optPointerListSection->addr());
+
+ // Write nothing to optROHeader until everything else is written.
+ // If something fails below, libobjc will not use the section.
+
+ // Find objc-containing dylibs
+ std::vector<LayoutInfo> objcDylibs;
+ for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
+ macho_header<P> *mh = (macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ if (mh->getSection("__DATA", "__objc_imageinfo") || mh->getSegment("__OBJC")) {
+ objcDylibs.push_back(*it);
+ }
+ }
+
+ // Build image list
+
+ // This is SAFE: the binaries themselves are unmodified.
+
+ std::vector<LayoutInfo> addressSortedDylibs = objcDylibs;
+ std::sort(addressSortedDylibs.begin(), addressSortedDylibs.end(), ByAddressSorter());
+
+ uint64_t hinfoVMAddr = optRWSection->addr() + optRWSection->size() - optRWRemaining;
+ HeaderInfoOptimizer<A> hinfoOptimizer;
+ err = hinfoOptimizer.init(objcDylibs.size(), optRWData, optRWRemaining);
+ if (err) {
+ warn(archName(), err);
+ return;
+ }
+ for(typename std::vector<LayoutInfo>::const_iterator it = addressSortedDylibs.begin(); it != addressSortedDylibs.end(); ++it) {
const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
- if ((seloptSection = mh->getSection("__TEXT", "__objc_selopt"))) break;
- }
-
- if (!seloptSection) {
- warn(archName(), "couldn't find libobjc's unique selector section (selectors not optimized)");
- return;
+ hinfoOptimizer.update(this, mh, pointersInData);
}
- objc_selopt::objc_selopt_t *seloptData = (objc_selopt::objc_selopt_t *)
- mappedCacheAddressForAddress(seloptSection->addr());
- if (seloptSection->size() < sizeof(seloptData->version)) {
- warn(archName(), "libobjc's unique selector section is too small (selectors not optimized)");
- return;
- }
-
- if (E::get32(seloptData->version) != objc_selopt::VERSION) {
- warn(archName(), "libobjc's unique selector section version is unrecognized (selectors not optimized)");
- return;
- }
-
// Update selector references and build selector list
- ObjCSelectorUniquer<A> uniq(this);
+
+ // This is SAFE: if we run out of room for the selector table,
+ // the modified binaries are still usable.
// Heuristic: choose selectors from libraries with more cstring data first.
// This tries to localize selector cstring memory.
- std::vector<LayoutInfo> sortedDylibs = fDylibs;
- std::sort(sortedDylibs.begin(), sortedDylibs.end(), ByCStringSectionSizeSorter());
-
- for(typename std::vector<LayoutInfo>::const_iterator it = sortedDylibs.begin(); it != sortedDylibs.end(); ++it) {
+ ObjCSelectorUniquer<A> uniq(this);
+ std::vector<LayoutInfo> sizeSortedDylibs = objcDylibs;
+ std::sort(sizeSortedDylibs.begin(), sizeSortedDylibs.end(), ByCStringSectionSizeSorter());
+
+ SelectorOptimizer<A, ObjCSelectorUniquer<A> > selOptimizer(uniq);
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
LegacySelectorUpdater<A, ObjCSelectorUniquer<A> >::update(this, mh, uniq);
- SelectorUpdater<A, ObjCSelectorUniquer<A> >::update(this, mh, uniq);
- }
-
- if ( verbose ) {
- fprintf(stderr, "update_dyld_shared_cache: for %s, found %zu unique objc selectors\n", archName(), uniq.strings().size());
- }
-
- // Write selector hash table to libobjc's __TEXT,__objc_selopt section
- size_t bytesUsed;
- const char *err =
- objc_selopt::write_selopt(seloptData, seloptSection->addr(),
- seloptSection->size(), uniq.strings(),
- E::little_endian, &bytesUsed);
+ selOptimizer.optimize(this, mh);
+ }
+
+ uint64_t seloptVMAddr = optROSection->addr() + optROSection->size() - optRORemaining;
+ objc_opt::objc_selopt_t *selopt = new(optROData) objc_opt::objc_selopt_t;
+ err = selopt->write(seloptVMAddr, optRORemaining, uniq.strings());
if (err) {
warn(archName(), err);
return;
}
+ optROData += selopt->size();
+ optRORemaining -= selopt->size();
+ selopt->byteswap(E::little_endian), selopt = NULL;
+
+
+ // Build class table.
+
+ // This is SAFE: the binaries themselves are unmodified.
+
+ ClassListBuilder<A> classes(hinfoOptimizer);
+ ClassWalker< A, ClassListBuilder<A> > classWalker(classes);
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ classWalker.walk(this, mh);
+ }
+
+ uint64_t clsoptVMAddr = optROSection->addr() + optROSection->size() - optRORemaining;
+ objc_opt::objc_clsopt_t *clsopt = new(optROData) objc_opt::objc_clsopt_t;
+ err = clsopt->write(clsoptVMAddr, optRORemaining,
+ classes.classNames(), classes.classes(), verbose);
+ if (err) {
+ warn(archName(), err);
+ return;
+ }
+ optROData += clsopt->size();
+ optRORemaining -= clsopt->size();
+ size_t duplicateCount = clsopt->duplicateCount();
+ clsopt->byteswap(E::little_endian), clsopt = NULL;
+
+
+ // Sort method lists.
+
+ // This is SAFE: modified binaries are still usable as unsorted lists.
+ // This must be done AFTER uniquing selectors.
+
+ MethodListSorter<A> methodSorter;
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ macho_header<P> *mh = (macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ methodSorter.optimize(this, mh);
+ }
+
+
+ // Unique protocols and build protocol table.
+
+ // This is SAFE: no protocol references are updated yet
+ // This must be done AFTER updating method lists.
+
+ ProtocolOptimizer<A> protocolOptimizer;
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ protocolOptimizer.addProtocols(this, mh);
+ }
+
+ pint_t protocolClassVMAddr = P::getP(optPointerList->protocolClass);
+ err = protocolOptimizer.writeProtocols(this, optRWData, optRWRemaining,
+ pointersInData, protocolClassVMAddr);
+ if (err) {
+ warn(archName(), err);
+ return;
+ }
+
+ uint64_t protocoloptVMAddr = optROSection->addr() + optROSection->size() - optRORemaining;
+ objc_opt::objc_protocolopt_t *protocolopt = new(optROData) objc_opt::objc_protocolopt_t;
+ err = protocolopt->write(protocoloptVMAddr, optRORemaining,
+ protocolOptimizer.protocolNames(),
+ protocolOptimizer.protocols(), verbose);
+ if (err) {
+ warn(archName(), err);
+ return;
+ }
+ optROData += protocolopt->size();
+ optRORemaining -= protocolopt->size();
+ protocolopt->byteswap(E::little_endian), protocolopt = NULL;
+
+
+ // Redirect protocol references to the uniqued protocols.
+
+ // This is SAFE: the new protocol objects are still usable as-is.
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ protocolOptimizer.updateReferences(this, mh);
+ }
+
+
+ // Repair ivar offsets.
+
+ // This is SAFE: the runtime always validates ivar offsets at runtime.
+
+ IvarOffsetOptimizer<A> ivarOffsetOptimizer;
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ ivarOffsetOptimizer.findGCClasses(this, mh);
+ }
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ ivarOffsetOptimizer.optimize(this, mh);
+ }
+
+
+ // Success. Mark dylibs as optimized.
+ for(typename std::vector<LayoutInfo>::const_iterator it = sizeSortedDylibs.begin(); it != sizeSortedDylibs.end(); ++it) {
+ const macho_header<P> *mh = (const macho_header<P>*)(*it->layout).getSegments()[0].mappedAddress();
+ const macho_section<P> *imageInfoSection;
+ imageInfoSection = mh->getSection("__DATA", "__objc_imageinfo");
+ if (!imageInfoSection) {
+ imageInfoSection = mh->getSection("__OBJC", "__image_info");
+ }
+ if (imageInfoSection) {
+ objc_image_info<A> *info = (objc_image_info<A> *)
+ mappedAddressForVMAddress(imageInfoSection->addr());
+ info->setOptimizedByDyld();
+ }
+ }
+
+
+ // Success. Update RO header last.
+ E::set32(optROHeader->selopt_offset, seloptVMAddr - optROSection->addr());
+ E::set32(optROHeader->clsopt_offset, clsoptVMAddr - optROSection->addr());
+ E::set32(optROHeader->protocolopt_offset, protocoloptVMAddr - optROSection->addr());
+ E::set32(optROHeader->headeropt_offset, hinfoVMAddr - optROSection->addr());
if ( verbose ) {
+ size_t roSize = optROSection->size() - optRORemaining;
+ size_t rwSize = optRWSection->size() - optRWRemaining;
fprintf(stderr, "update_dyld_shared_cache: for %s, %zu/%llu bytes "
- "(%d%%) used in libobjc unique selector section\n",
- archName(), bytesUsed, seloptSection->size(),
- (int)(bytesUsed / (double)seloptSection->size() * 100));
+ "(%d%%) used in libobjc read-only optimization section\n",
+ archName(), roSize, optROSection->size(),
+ percent(roSize, optROSection->size()));
+ fprintf(stderr, "update_dyld_shared_cache: for %s, %zu/%llu bytes "
+ "(%d%%) used in libobjc read/write optimization section\n",
+ archName(), rwSize, optRWSection->size(),
+ percent(rwSize, optRWSection->size()));
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "uniqued %zu selectors\n",
+ archName(), uniq.strings().size());
fprintf(stderr, "update_dyld_shared_cache: for %s, "
"updated %zu selector references\n",
archName(), uniq.count());
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "uniqued %zu protocols\n",
+ archName(), protocolOptimizer.protocolCount());
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "updated %zu protocol references\n",
+ archName(), protocolOptimizer.protocolReferenceCount());
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "updated %zu ivar offsets\n",
+ archName(), ivarOffsetOptimizer.optimized());
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "sorted %zu method lists\n",
+ archName(), methodSorter.optimized());
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "recorded %zu classes (%zu duplicates)\n",
+ archName(), classes.classNames().size(), duplicateCount);
+ fprintf(stderr, "update_dyld_shared_cache: for %s, "
+ "wrote objc metadata optimization version %d\n",
+ archName(), objc_opt::VERSION);
}
return;
@@ -1751,11 +2741,183 @@
}
+// <rdar://problem/10730767> update_dyld_shared_cache should use sync_volume_np() instead of sync()
+static void sync_volume(const char* volumePath)
+{
+#if __MAC_OS_X_VERSION_MIN_REQUIRED >= 1080
+ int error = sync_volume_np(volumePath, SYNC_VOLUME_FULLSYNC|SYNC_VOLUME_FULLSYNC);
+#else
+ int full_sync = 3; // SYNC_VOLUME_FULLSYNC | SYNC_VOLUME_FULLSYNC
+ int error = 0;
+ if ( fsctl(volumePath, 0x80004101 /*FSCTL_SYNC_VOLUME*/, &full_sync, 0) == -1)
+ error = errno;
+#endif
+ if ( error )
+ ::sync();
+}
+
+
+// <rdar://problem/12552226> update shared cache should sign the shared cache
+static bool adhoc_codesign_share_cache(const char* path)
+{
+ CFURLRef target = ::CFURLCreateFromFileSystemRepresentation(NULL, (const UInt8 *)path, strlen(path), FALSE);
+ if ( target == NULL )
+ return false;
+
+ SecStaticCodeRef code;
+ OSStatus status = ::SecStaticCodeCreateWithPath(target, kSecCSDefaultFlags, &code);
+ CFRelease(target);
+ if ( status ) {
+ ::fprintf(stderr, "codesign: failed to create url to signed object\n");
+ return false;
+ }
+
+ const void * keys[1] = { (void *)kSecCodeSignerIdentity } ;
+ const void * values[1] = { (void *)kCFNull };
+ CFDictionaryRef params = ::CFDictionaryCreate(NULL, keys, values, 1, &kCFTypeDictionaryKeyCallBacks, &kCFTypeDictionaryValueCallBacks);
+ if ( params == NULL ) {
+ CFRelease(code);
+ return false;
+ }
+
+ SecCodeSignerRef signer;
+ status = ::SecCodeSignerCreate(params, kSecCSDefaultFlags, &signer);
+ CFRelease(params);
+ if ( status ) {
+ CFRelease(code);
+ ::fprintf(stderr, "codesign: failed to create signer object\n");
+ return false;
+ }
+
+ status = ::SecCodeSignerAddSignatureWithErrors(signer, code, kSecCSDefaultFlags, NULL);
+ CFRelease(code);
+ CFRelease(signer);
+ if ( status ) {
+ ::fprintf(stderr, "codesign: failed to sign object: %s\n", path);
+ return false;
+ }
+
+ if ( verbose )
+ ::fprintf(stderr, "codesigning complete of %s\n", path);
+
+ return true;
+}
+
template <typename A>
-bool SharedCache<A>::update(bool usesOverlay, bool force, bool optimize, bool deleteExistingFirst, int archIndex,
- int archCount, bool keepSignatures)
+void SharedCache<A>::writeCacheFile(const char *cacheFilePath, uint8_t *cacheFileBuffer, uint32_t cacheFileSize, bool deleteOldCache) {
+ char tempCachePath[strlen(cacheFilePath)+16];
+ sprintf(tempCachePath, "%s.tmp%u", cacheFilePath, getpid());
+
+ try {
+ // install signal handlers to delete temp file if program is killed
+ sCleanupFile = tempCachePath;
+ ::signal(SIGINT, cleanup);
+ ::signal(SIGBUS, cleanup);
+ ::signal(SIGSEGV, cleanup);
+
+ // compute UUID of whole cache
+ uint8_t digest[16];
+ CC_MD5(cacheFileBuffer, cacheFileSize, digest);
+ // <rdar://problem/6723729> uuids should conform to RFC 4122 UUID version 4 & UUID version 5 formats
+ digest[6] = ( digest[6] & 0x0F ) | ( 3 << 4 );
+ digest[8] = ( digest[8] & 0x3F ) | 0x80;
+ ((dyldCacheHeader<E>*)cacheFileBuffer)->set_uuid(digest);
+
+ // create var/db/dyld dirs if needed
+ char dyldDirs[1024];
+ strcpy(dyldDirs, cacheFilePath);
+ char* lastSlash = strrchr(dyldDirs, '/');
+ if ( lastSlash != NULL )
+ lastSlash[1] = '\0';
+ struct stat stat_buf;
+ if ( stat(dyldDirs, &stat_buf) != 0 ) {
+ const char* afterSlash = &dyldDirs[1];
+ char* slash;
+ while ( (slash = strchr(afterSlash, '/')) != NULL ) {
+ *slash = '\0';
+ ::mkdir(dyldDirs, S_IRWXU | S_IRGRP|S_IXGRP | S_IROTH|S_IXOTH);
+ *slash = '/';
+ afterSlash = slash+1;
+ }
+ }
+
+ // create temp file for cache
+ int fd = ::open(tempCachePath, O_CREAT | O_RDWR | O_TRUNC, 0644);
+ if ( fd == -1 )
+ throwf("can't create temp file %s, errno=%d", tempCachePath, errno);
+
+ // try to allocate whole cache file contiguously
+ fstore_t fcntlSpec = { F_ALLOCATECONTIG|F_ALLOCATEALL, F_PEOFPOSMODE, 0, cacheFileSize, 0 };
+ ::fcntl(fd, F_PREALLOCATE, &fcntlSpec);
+
+ // write out cache file
+ if ( verbose )
+ fprintf(stderr, "update_dyld_shared_cache: writing cache to disk: %s\n", tempCachePath);
+ if ( ::pwrite(fd, cacheFileBuffer, cacheFileSize, 0) != cacheFileSize )
+ throwf("write() failure creating cache file, errno=%d", errno);
+
+ // flush to disk and close
+ int result = ::fcntl(fd, F_FULLFSYNC, NULL);
+ if ( result == -1 )
+ fprintf(stderr, "update_dyld_shared_cache: warning, fcntl(F_FULLFSYNC) failed with errno=%d for %s\n", errno, tempCachePath);
+ result = ::close(fd);
+ if ( result != 0 )
+ fprintf(stderr, "update_dyld_shared_cache: warning, close() failed with errno=%d for %s\n", errno, tempCachePath);
+
+ if ( !iPhoneOS )
+ adhoc_codesign_share_cache(tempCachePath);
+
+ if ( deleteOldCache ) {
+ const char* pathLastSlash = strrchr(cacheFilePath, '/');
+ if ( pathLastSlash != NULL ) {
+ result = ::unlink(cacheFilePath);
+ if ( result != 0 ) {
+ if ( errno != ENOENT )
+ fprintf(stderr, "update_dyld_shared_cache: warning, unable to remove existing cache %s because errno=%d\n", cacheFilePath, errno);
+ }
+ }
+ }
+
+ // move new cache file to correct location for use after reboot
+ if ( verbose )
+ fprintf(stderr, "update_dyld_shared_cache: atomically moving cache file into place: %s\n", cacheFilePath);
+ result = ::rename(tempCachePath, cacheFilePath);
+ if ( result != 0 )
+ throwf("can't swap newly create dyld shared cache file: rename(%s,%s) returned errno=%d", tempCachePath, cacheFilePath, errno);
+
+ // flush everything to disk to assure rename() gets recorded
+ sync_volume(cacheFilePath);
+
+ // restore default signal handlers
+ ::signal(SIGINT, SIG_DFL);
+ ::signal(SIGBUS, SIG_DFL);
+ ::signal(SIGSEGV, SIG_DFL);
+ }
+ catch (...){
+ // remove temp cache file
+ ::unlink(tempCachePath);
+ throw;
+ }
+}
+
+
+template <> bool SharedCache<x86_64>::addCacheSlideInfo(){ return true; }
+template <> bool SharedCache<arm>::addCacheSlideInfo() { return true; }
+template <> bool SharedCache<x86>::addCacheSlideInfo() { return false; }
+template <> bool SharedCache<arm64>::addCacheSlideInfo() { return true; }
+
+
+template <typename A>
+bool SharedCache<A>::update(bool force, bool optimize, bool deleteExistingFirst, int archIndex,
+ int archCount, bool keepSignatures, bool dontMapLocalSymbols)
{
bool didUpdate = false;
+ bool canEmitDevelopmentCache = true;
+ char devCacheFilePath[strlen(fCacheFilePath)+strlen(".development")];
+ char fileListFilePath[strlen(fCacheFilePath)+strlen(".list")];
+ sprintf(devCacheFilePath, "%s.development", fCacheFilePath);
+ sprintf(fileListFilePath, "%s.list", fCacheFilePath);
+ std::vector<const char *> paths;
// already up to date?
if ( force || fExistingIsNotUpToDate ) {
@@ -1771,14 +2933,12 @@
::unlink(fCacheFilePath);
uint8_t* inMemoryCache = NULL;
uint32_t allocatedCacheSize = 0;
- char tempCachePath[strlen(fCacheFilePath)+16];
- sprintf(tempCachePath, "%s.tmp%u", fCacheFilePath, getpid());
try {
// allocate a memory block to hold cache
uint32_t cacheFileSize = 0;
for(std::vector<shared_file_mapping_np>::iterator it = fMappings.begin(); it != fMappings.end(); ++it) {
uint32_t end = it->sfm_file_offset + it->sfm_size;
- if ( end > cacheFileSize )
+ if ( end > cacheFileSize )
cacheFileSize = end;
}
if ( vm_allocate(mach_task_self(), (vm_address_t*)(&inMemoryCache), cacheFileSize, VM_FLAGS_ANYWHERE) != KERN_SUCCESS )
@@ -1797,10 +2957,14 @@
header->set_mappingOffset(sizeof(dyldCacheHeader<E>));
header->set_mappingCount(fMappings.size());
header->set_imagesOffset(header->mappingOffset() + fMappings.size()*sizeof(dyldCacheFileMapping<E>));
- header->set_imagesCount(fDylibs.size());
+ header->set_imagesCount(fDylibs.size()+fDylibAliases.size());
header->set_dyldBaseAddress(fDyldBaseAddress);
- //header->set_dependenciesOffset(sizeof(dyldCacheHeader<E>) + fMappings.size()*sizeof(dyldCacheFileMapping<E>) + fDylibs.size()*sizeof(dyldCacheImageInfo<E>));
- //header->set_dependenciesCount(fDependencyPool.size());
+ header->set_codeSignatureOffset(cacheFileSize);
+ header->set_codeSignatureSize(0);
+ header->set_slideInfoOffset(0);
+ header->set_slideInfoSize(0);
+ header->set_localSymbolsOffset(0);
+ header->set_localSymbolsSize(0);
// fill in mappings
dyldCacheFileMapping<E>* mapping = (dyldCacheFileMapping<E>*)&inMemoryCache[sizeof(dyldCacheHeader<E>)];
@@ -1822,8 +2986,18 @@
image->set_address(it->info.address);
image->set_modTime(it->info.modTime);
image->set_inode(it->info.inode);
- image->set_pathFileOffset(cacheFileOffsetForAddress(it->info.address+it->info.pathFileOffset));
- //image->set_dependenciesStartOffset(it->info.dependenciesStartOffset);
+ image->set_pathFileOffset(cacheFileOffsetForVMAddress(it->info.address+it->info.pathFileOffset));
+ ++image;
+ }
+
+ // add aliases to end of image table
+ for(typename std::vector<LayoutInfo>::iterator it = fDylibAliases.begin(); it != fDylibAliases.end(); ++it) {
+ image->set_address(it->info.address);
+ image->set_modTime(it->info.modTime);
+ image->set_inode(it->info.inode);
+ image->set_pathFileOffset(it->info.pathFileOffset);
+ strcpy((char*)inMemoryCache+it->info.pathFileOffset, it->aliases[0]);
+ //fprintf(stderr, "adding alias to offset 0x%08X %s\n", it->info.pathFileOffset, it->aliases[0]);
++image;
}
@@ -1831,32 +3005,38 @@
const int dylibCount = fDylibs.size();
int dylibIndex = 0;
int progressIndex = 0;
+ bool foundLibSystem = false;
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it, ++dylibIndex) {
const char* path = it->layout->getFilePath();
int src = ::open(path, O_RDONLY, 0);
if ( src == -1 )
- throwf("can't open file %s, errnor=%d", it->layout->getID().name, errno);
+ throwf("can't open file %s, errno=%d", it->layout->getID().name, errno);
// mark source as "don't cache"
(void)fcntl(src, F_NOCACHE, 1);
// verify file has not changed since dependency analysis
struct stat stat_buf;
if ( fstat(src, &stat_buf) == -1)
throwf("can't stat open file %s, errno=%d", path, errno);
- if ( (it->layout->getInode() != stat_buf.st_ino) || (it->layout->getLastModTime() != stat_buf.st_mtime) )
- throwf("file modified during cache creation: %s", path);
-
+ if ( (it->layout->getInode() != stat_buf.st_ino) )
+ throwf("file inode changed from %llu to %llu during cache creation: %s", it->layout->getInode(), stat_buf.st_ino, path);
+ else if ( it->layout->getLastModTime() != stat_buf.st_mtime )
+ throwf("file mtime changed from 0x%lX to 0x%lX during cache creation: %s", it->layout->getLastModTime(), stat_buf.st_mtime, path);
+ if ( strcmp(it->layout->getID().name, "/usr/lib/libSystem.B.dylib") == 0 )
+ foundLibSystem = true;
if ( verbose )
- fprintf(stderr, "update_dyld_shared_cache: copying %s to cache\n", it->layout->getID().name);
+ fprintf(stderr, "update_dyld_shared_cache: copying %s to cache\n", it->layout->getFilePath());
try {
const std::vector<MachOLayoutAbstraction::Segment>& segs = it->layout->getSegments();
for (int i=0; i < segs.size(); ++i) {
const MachOLayoutAbstraction::Segment& seg = segs[i];
- if ( verbose )
- fprintf(stderr, "\t\tsegment %s, size=0x%0llX, cache address=0x%0llX\n", seg.name(), seg.fileSize(), seg.newAddress());
+ if ( verbose ) {
+ fprintf(stderr, "\t\tsegment %s, size=0x%0llX, cache address=0x%0llX, buffer address=%p\n",
+ seg.name(), seg.size(), seg.newAddress(), &inMemoryCache[cacheFileOffsetForVMAddress(seg.newAddress())]);
+ }
if ( seg.size() > 0 ) {
const uint64_t segmentSrcStartOffset = it->layout->getOffsetInUniversalFile()+seg.fileOffset();
const uint64_t segmentSize = seg.fileSize();
- const uint64_t segmentDstStartOffset = cacheFileOffsetForAddress(seg.newAddress());
+ const uint64_t segmentDstStartOffset = cacheFileOffsetForVMAddress(seg.newAddress());
ssize_t readResult = ::pread(src, &inMemoryCache[segmentDstStartOffset], segmentSize, segmentSrcStartOffset);
if ( readResult != segmentSize ) {
if ( readResult == -1 )
@@ -1872,6 +3052,7 @@
throwf("%s while copying %s to shared cache", msg, it->layout->getID().name);
}
::close(src);
+ paths.push_back(it->layout->getID().name);
if ( progress ) {
// assuming read takes 40% of time
int nextProgressIndex = archIndex*100+(40*dylibIndex)/dylibCount;
@@ -1880,43 +3061,38 @@
progressIndex = nextProgressIndex;
}
}
-
+ if ( !foundLibSystem )
+ throw "cache would be missing required dylib /usr/lib/libSystem.B.dylib";
+
// set mapped address for each segment
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
std::vector<MachOLayoutAbstraction::Segment>& segs = ((MachOLayoutAbstraction*)(it->layout))->getSegments();
for (int i=0; i < segs.size(); ++i) {
MachOLayoutAbstraction::Segment& seg = segs[i];
if ( seg.size() > 0 )
- seg.setMappedAddress(inMemoryCache + cacheFileOffsetForAddress(seg.newAddress()));
+ seg.setMappedAddress(inMemoryCache + cacheFileOffsetForVMAddress(seg.newAddress()));
//fprintf(stderr, "%s at %p to %p for %s\n", seg.name(), seg.mappedAddress(), (char*)seg.mappedAddress()+ seg.size(), it->layout->getID().name);
}
}
-
+
+ // also construct list of all pointers in cache to other things in cache
+ std::vector<void*> pointersInData;
+ pointersInData.reserve(1024);
+
// rebase each dylib in shared cache
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
try {
Rebaser<A> r(*it->layout);
- r.rebase();
+ if (!r.rebase(pointersInData)) {
+ canEmitDevelopmentCache = false;
+ fprintf(stderr, "update_dyld_shared_cache: Omitting development cache for %s, cannot rebase dylib into place for %s\n", archName(), it->layout->getID().name);
+ }
//if ( verbose )
// fprintf(stderr, "update_dyld_shared_cache: for %s, rebasing dylib into cache for %s\n", archName(), it->layout->getID().name);
}
catch (const char* msg) {
throwf("%s in %s", msg, it->layout->getID().name);
}
- }
-
- // merge/optimize all LINKEDIT segments
- if ( optimize ) {
- //fprintf(stderr, "update_dyld_shared_cache: original cache file size %uMB\n", cacheFileSize/(1024*1024));
- cacheFileSize = (this->optimizeLINKEDIT(keepSignatures) - inMemoryCache);
- //fprintf(stderr, "update_dyld_shared_cache: optimized cache file size %uMB\n", cacheFileSize/(1024*1024));
- // update header to reduce mapping size
- dyldCacheHeader<E>* cacheHeader = (dyldCacheHeader<E>*)inMemoryCache;
- dyldCacheFileMapping<E>* mappings = (dyldCacheFileMapping<E>*)&inMemoryCache[sizeof(dyldCacheHeader<E>)];
- dyldCacheFileMapping<E>* lastMapping = &mappings[cacheHeader->mappingCount()-1];
- lastMapping->set_size(cacheFileSize-lastMapping->file_offset());
- // update fMappings so .map file will print correctly
- fMappings.back().sfm_size = cacheFileSize-fMappings.back().sfm_file_offset;
}
if ( verbose )
@@ -1926,13 +3102,12 @@
std::vector<Binder<A>*> binders;
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
//fprintf(stderr, "binding %s\n", it->layout->getID().name);
- Binder<A>* binder = new Binder<A>(*it->layout, fDyldBaseAddress);
+ Binder<A>* binder = new Binder<A>(*it->layout);
binders.push_back(binder);
// only add dylibs to map
if ( it->layout->getID().name != NULL )
map[it->layout->getID().name] = binder;
}
-
// tell each Binder about the others
for(typename std::vector<Binder<A>*>::iterator it = binders.begin(); it != binders.end(); ++it) {
(*it)->setDependentBinders(map);
@@ -1942,27 +3117,254 @@
if ( verbose )
fprintf(stderr, "update_dyld_shared_cache: for %s, updating binding information in cache for %s\n", archName(), (*it)->getDylibID());
try {
- (*it)->bind();
+ (*it)->bind(pointersInData);
}
catch (const char* msg) {
throwf("%s in %s", msg, (*it)->getDylibID());
}
}
+
+ for(typename std::vector<LayoutInfo>::iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
+ const macho_header<P>* fHeader = (const macho_header<P>*)it->layout->getSegments()[0].mappedAddress();
+ const macho_load_command<P>* const cmds = (macho_load_command<P>*)((uint8_t*)fHeader + sizeof(macho_header<P>));
+ const uint32_t cmd_count = fHeader->ncmds();
+ const macho_load_command<P>* cmd = cmds;
+ macho_dyld_info_command<P>* fDyldInfo;
+ uint64_t originalLinkEditVMAddr = 0;
+ for (uint32_t i = 0; i < cmd_count; ++i) {
+ if ( cmd->cmd() == macho_segment_command<P>::CMD ) {
+ macho_segment_command<P>* seg = (macho_segment_command<P>*)cmd;
+ if ( strcmp(seg->segname(), "__LINKEDIT") != 0 ) {
+ pint_t oldFileOff = seg->fileoff();
+ originalLinkEditVMAddr += seg->vmsize();
+ // don't alter __TEXT until <rdar://problem/7022345> is fixed
+ if ( strcmp(seg->segname(), "__TEXT") != 0 ) {
+ // update all other segments fileoff to be offset from start of cache file
+ seg->set_fileoff(cacheFileOffsetForVMAddress(seg->vmaddr()));
+ }
+ pint_t fileOffsetDelta = seg->fileoff() - oldFileOff;
+ const MachOLayoutAbstraction::Segment* layoutSeg = it->layout->getSegment(seg->segname());
+ if ( layoutSeg != NULL ) {
+ seg->set_vmsize(layoutSeg->size());
+ seg->set_filesize(layoutSeg->fileSize());
+ }
+ // update all sections in this segment
+ macho_section<P>* const sectionsStart = (macho_section<P>*)((char*)seg + sizeof(macho_segment_command<P>));
+ macho_section<P>* const sectionsEnd = §ionsStart[seg->nsects()];
+ for(macho_section<P>* sect = sectionsStart; sect < sectionsEnd; ++sect) {
+ if ( sect->offset() != 0 )
+ sect->set_offset(sect->offset()+fileOffsetDelta);
+ }
+ }
+ } else if (cmd->cmd() == LC_DYLD_INFO || cmd->cmd() == LC_DYLD_INFO_ONLY) {
+ fDyldInfo = (macho_dyld_info_command<P>*)cmd;
+ }
+ cmd = (const macho_load_command<P>*)(((uint8_t*)cmd)+cmd->cmdsize());
+ }
+ }
+
+ // optimize binding
+ for(typename std::vector<Binder<A>*>::iterator it = binders.begin(); it != binders.end(); ++it) {
+ try {
+ (*it)->optimize();
+ }
+ catch (const char* msg) {
+ throwf("%s in %s", msg, (*it)->getDylibID());
+ }
+ }
+
// delete binders
for(typename std::vector<Binder<A>*>::iterator it = binders.begin(); it != binders.end(); ++it) {
delete *it;
}
+ // merge/optimize all LINKEDIT segments
+ if ( optimize ) {
+ if ( verbose )
+ fprintf(stderr, "update_dyld_shared_cache: original cache file size %uMB\n", cacheFileSize/(1024*1024));
+ cacheFileSize = (this->optimizeLINKEDIT(keepSignatures, dontMapLocalSymbols) - inMemoryCache);
+ if ( verbose )
+ fprintf(stderr, "update_dyld_shared_cache: optimized cache file size %uMB\n", cacheFileSize/(1024*1024));
+ // update header to reduce mapping size
+ dyldCacheHeader<E>* cacheHeader = (dyldCacheHeader<E>*)inMemoryCache;
+ dyldCacheFileMapping<E>* mappings = (dyldCacheFileMapping<E>*)&inMemoryCache[sizeof(dyldCacheHeader<E>)];
+ dyldCacheFileMapping<E>* lastMapping = &mappings[cacheHeader->mappingCount()-1];
+ lastMapping->set_size(cacheFileSize-lastMapping->file_offset());
+ // update fMappings so .map file will print correctly
+ fMappings.back().sfm_size = cacheFileSize-fMappings.back().sfm_file_offset;
+ // update header
+ //fprintf(stderr, "update_dyld_shared_cache: changing end of cache address from 0x%08llX to 0x%08llX\n",
+ // header->codeSignatureOffset(), fMappings.back().sfm_address + fMappings.back().sfm_size);
+ header->set_codeSignatureOffset(fMappings.back().sfm_file_offset + fMappings.back().sfm_size);
+ }
+
+ // dump dev cache with optimized linkedit, but not ObjC optimizations
+ if (iPhoneOS && canEmitDevelopmentCache) {
+ int fileListFD = ::open(fileListFilePath, O_WRONLY | O_CREAT | O_TRUNC, 0644);
+ if ( fileListFD != -1 ) {
+ for (const char* path : paths) {
+ write(fileListFD, path, strlen(path)+1);
+ write(fileListFD, "\n", 1);
+ }
+ close(fileListFD);
+ }
+
+ ((dyldCacheHeader<E>*)inMemoryCache)->set_cacheType(1);
+ writeCacheFile(devCacheFilePath, inMemoryCache, cacheFileSize, fCacheFileInFinalLocation);
+ }
+
// unique objc selectors and update other objc metadata
- if (optimize) {
- optimizeObjC();
- }
- if ( progress ) {
- // assuming objc optimizations takes 15% of time
- fprintf(stdout, "%3u/%u\n", (archIndex+1)*55, archCount*100);
- }
-
+ if ( optimize ) {
+ optimizeObjC(pointersInData);
+ if ( progress ) {
+ // assuming objc optimizations takes 15% of time
+ fprintf(stdout, "%3u/%u\n", (archIndex+1)*55, archCount*100);
+ }
+ }
+
+ if ( addCacheSlideInfo() ) {
+ // build bitmap of which pointers need sliding
+ uint8_t* const dataStart = &inMemoryCache[fMappings[1].sfm_file_offset]; // R/W mapping is always second
+ uint8_t* const dataEnd = &inMemoryCache[fMappings[1].sfm_file_offset+fMappings[1].sfm_size];
+ const int bitmapSize = (dataEnd - dataStart)/(4*8);
+ uint8_t* bitmap = (uint8_t*)calloc(bitmapSize, 1);
+ void* lastPointer = inMemoryCache;
+ for(std::vector<void*>::iterator pit=pointersInData.begin(); pit != pointersInData.end(); ++pit) {
+ if ( *pit != lastPointer ) {
+ void* p = *pit;
+ if ( (p < dataStart) || ( p > dataEnd) )
+ throwf("DATA pointer for sliding, out of range 0x%08lX\n", (long)((uint8_t*)p-inMemoryCache));
+ long offset = (long)((uint8_t*)p - dataStart);
+ if ( (offset % 4) != 0 )
+ throwf("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);
+ lastPointer = p;
+ }
+ }
+
+ // allocate worst case size block of all slide info
+ const int entry_size = 4096/(8*4); // 8 bits per byte, possible pointer every 4 bytes.
+ const int toc_count = bitmapSize/entry_size;
+ int slideInfoSize = sizeof(dyldCacheSlideInfo<E>) + 2*toc_count + entry_size*(toc_count+1);
+ dyldCacheSlideInfo<E>* slideInfo = (dyldCacheSlideInfo<E>*)calloc(slideInfoSize, 1);
+ slideInfo->set_version(1);
+ slideInfo->set_toc_offset(sizeof(dyldCacheSlideInfo<E>));
+ slideInfo->set_toc_count(toc_count);
+ slideInfo->set_entries_offset((slideInfo->toc_offset()+2*toc_count+127)&(-128));
+ slideInfo->set_entries_count(0);
+ slideInfo->set_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 ) {
+ //fprintf(stderr, "toc[%d] optimized to %d\n", i, j);
+ 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->set_entries_count(entry_count);
+
+ int slideInfoPageSize = regionAlign(slideInfo->entries_offset() + entry_count*entry_size);
+ cacheFileSize += slideInfoPageSize;
+
+ // update mappings to increase RO size
+ dyldCacheHeader<E>* cacheHeader = (dyldCacheHeader<E>*)inMemoryCache;
+ dyldCacheFileMapping<E>* mappings = (dyldCacheFileMapping<E>*)&inMemoryCache[sizeof(dyldCacheHeader<E>)];
+ dyldCacheFileMapping<E>* lastMapping = &mappings[cacheHeader->mappingCount()-1];
+ lastMapping->set_size(lastMapping->size()+slideInfoPageSize);
+
+ // update header to show location of slidePointers
+ cacheHeader->set_slideInfoOffset(cacheHeader->codeSignatureOffset());
+ cacheHeader->set_slideInfoSize(slideInfoPageSize);
+ cacheHeader->set_codeSignatureOffset(cacheHeader->codeSignatureOffset()+slideInfoPageSize);
+
+ // update fMappings so .map file will print correctly
+ fMappings.back().sfm_size = cacheFileSize-fMappings.back().sfm_file_offset;
+
+ // copy compressed into into buffer
+ memcpy(&inMemoryCache[cacheHeader->slideInfoOffset()], slideInfo, slideInfoPageSize);
+ }
+
+ // append local symbol info in an unmapped region
+ if ( dontMapLocalSymbols ) {
+ uint32_t spaceAtEnd = allocatedCacheSize - cacheFileSize;
+ uint32_t localSymbolsOffset = pageAlign(cacheFileSize);
+ dyldCacheLocalSymbolsInfo<E>* infoHeader = (dyldCacheLocalSymbolsInfo<E>*)(&inMemoryCache[localSymbolsOffset]);
+ const uint32_t entriesOffset = sizeof(dyldCacheLocalSymbolsInfo<E>);
+ const uint32_t entriesCount = fLocalSymbolInfos.size();
+ const uint32_t nlistOffset = entriesOffset + entriesCount * sizeof(dyldCacheLocalSymbolEntry<E>);
+ const uint32_t nlistCount = fUnmappedLocalSymbols.size();
+ const uint32_t stringsOffset = nlistOffset + nlistCount * sizeof(macho_nlist<P>);
+ const uint32_t stringsSize = fUnmappedLocalsStringPool.size();
+ if ( stringsOffset+stringsSize > spaceAtEnd )
+ throwf("update_dyld_shared_cache[%u] for arch=%s, out of space for local symbols. Have 0x%X, Need 0x%X\n",
+ getpid(), fArchGraph->archName(), spaceAtEnd, stringsOffset+stringsSize);
+ // fill in local symbols info
+ infoHeader->set_nlistOffset(nlistOffset);
+ infoHeader->set_nlistCount(nlistCount);
+ infoHeader->set_stringsOffset(stringsOffset);
+ infoHeader->set_stringsSize(stringsSize);
+ infoHeader->set_entriesOffset(entriesOffset);
+ infoHeader->set_entriesCount(entriesCount);
+ // copy info for each dylib
+ dyldCacheLocalSymbolEntry<E>* entries = (dyldCacheLocalSymbolEntry<E>*)(&inMemoryCache[localSymbolsOffset+entriesOffset]);
+ for (int i=0; i < entriesCount; ++i) {
+ entries[i].set_dylibOffset(fLocalSymbolInfos[i].dylibOffset);
+ entries[i].set_nlistStartIndex(fLocalSymbolInfos[i].nlistStartIndex);
+ entries[i].set_nlistCount(fLocalSymbolInfos[i].nlistCount);
+ }
+ // copy nlists
+ memcpy(&inMemoryCache[localSymbolsOffset+nlistOffset], &fUnmappedLocalSymbols[0], nlistCount*sizeof(macho_nlist<P>));
+ // copy string pool
+ memcpy(&inMemoryCache[localSymbolsOffset+stringsOffset], fUnmappedLocalsStringPool.getBuffer(), stringsSize);
+
+ // update state
+ fUnmappedLocalSymbolsSize = pageAlign(stringsOffset + stringsSize);
+ cacheFileSize = regionAlign(localSymbolsOffset + fUnmappedLocalSymbolsSize);
+
+ // update header to show location of slidePointers
+ dyldCacheHeader<E>* cacheHeader = (dyldCacheHeader<E>*)inMemoryCache;
+ cacheHeader->set_localSymbolsOffset(localSymbolsOffset);
+ cacheHeader->set_localSymbolsSize(stringsOffset+stringsSize);
+ cacheHeader->set_codeSignatureOffset(cacheFileSize);
+ }
+
+ // make sure after all optimizations, that whole cache file fits into shared region address range
+ {
+ dyldCacheHeader<E>* cacheHeader = (dyldCacheHeader<E>*)inMemoryCache;
+ dyldCacheFileMapping<E>* mappings = (dyldCacheFileMapping<E>*)&inMemoryCache[cacheHeader->mappingOffset()];
+ // <rdar://problem/16128830> incorporate code signature size into overflow check
+ uint32_t estCodeSigSize = regionAlign(cacheFileSize/200); // guess 0.5% for code signature
+ for (int i=0; i < cacheHeader->mappingCount(); ++i) {
+ uint64_t endAddr = mappings[i].address() + mappings[i].size() + estCodeSigSize;
+ if ( endAddr > (sharedRegionStartAddress() + sharedRegionSize()) ) {
+ throwf("update_dyld_shared_cache[%u] for arch=%s, shared cache will not fit in shared regions address space. Overflow amount: %lluKB\n",
+ getpid(), fArchGraph->archName(), (endAddr-(sharedRegionStartAddress() + sharedRegionSize()))/1024);
+ }
+ }
+ }
+
if ( fVerify ) {
+ // if no existing cache, say so
+ if ( fExistingCacheForVerification == NULL ) {
+ throwf("update_dyld_shared_cache[%u] for arch=%s, could not verify because cache file does not exist in /var/db/dyld/\n",
+ getpid(), archName());
+ }
// new cache is built, compare header entries
const dyldCacheHeader<E>* newHeader = (dyldCacheHeader<E>*)inMemoryCache;
const dyldCacheHeader<E>* oldHeader = (dyldCacheHeader<E>*)fExistingCacheForVerification;
@@ -2019,71 +3421,9 @@
}
}
else {
- // install signal handlers to delete temp file if program is killed
- sCleanupFile = tempCachePath;
- ::signal(SIGINT, cleanup);
- ::signal(SIGBUS, cleanup);
- ::signal(SIGSEGV, cleanup);
-
- // create var/db/dyld dirs if needed
- char dyldDirs[1024];
- strcpy(dyldDirs, fCacheFilePath);
- char* lastSlash = strrchr(dyldDirs, '/');
- if ( lastSlash != NULL )
- lastSlash[1] = '\0';
- struct stat stat_buf;
- if ( stat(dyldDirs, &stat_buf) != 0 ) {
- const char* afterSlash = &dyldDirs[1];
- char* slash;
- while ( (slash = strchr(afterSlash, '/')) != NULL ) {
- *slash = '\0';
- ::mkdir(dyldDirs, S_IRWXU | S_IRGRP|S_IXGRP | S_IROTH|S_IXOTH);
- *slash = '/';
- afterSlash = slash+1;
- }
- }
-
- // create temp file for cache
- int fd = ::open(tempCachePath, O_CREAT | O_RDWR | O_TRUNC, 0644);
- if ( fd == -1 )
- throwf("can't create temp file %s, errnor=%d", tempCachePath, errno);
-
- // try to allocate whole cache file contiguously
- fstore_t fcntlSpec = { F_ALLOCATECONTIG|F_ALLOCATEALL, F_PEOFPOSMODE, 0, cacheFileSize, 0 };
- ::fcntl(fd, F_PREALLOCATE, &fcntlSpec);
-
- // write out cache file
- if ( verbose )
- fprintf(stderr, "update_dyld_shared_cache: writing cache to disk\n");
- if ( ::pwrite(fd, inMemoryCache, cacheFileSize, 0) != cacheFileSize )
- throwf("write() failure creating cache file, errno=%d", errno);
- if ( progress ) {
- // assuming write takes 35% of time
- fprintf(stdout, "%3u/%u\n", (archIndex+1)*90, archCount*100);
- }
-
- // flush to disk and close
- int result = ::fcntl(fd, F_FULLFSYNC, NULL);
- if ( result == -1 )
- fprintf(stderr, "update_dyld_shared_cache: warning, fcntl(F_FULLFSYNC) failed with errno=%d for %s\n", errno, tempCachePath);
- result = ::close(fd);
- if ( result != 0 )
- fprintf(stderr, "update_dyld_shared_cache: warning, close() failed with errno=%d for %s\n", errno, tempCachePath);
-
- // atomically swap in new cache file, do this after F_FULLFSYNC
- result = ::rename(tempCachePath, fCacheFilePath);
- if ( result != 0 )
- throwf("can't swap newly create dyld shared cache file: rename(%s,%s) returned errno=%d", tempCachePath, fCacheFilePath, errno);
-
- // flush everything to disk to assure rename() gets recorded
- ::sync();
+ ((dyldCacheHeader<E>*)inMemoryCache)->set_cacheType(0);
+ writeCacheFile(fCacheFilePath, inMemoryCache, cacheFileSize, fCacheFileInFinalLocation);
didUpdate = true;
-
- // restore default signal handlers
- ::signal(SIGINT, SIG_DFL);
- ::signal(SIGBUS, SIG_DFL);
- ::signal(SIGSEGV, SIG_DFL);
-
// generate human readable "map" file that shows the layout of the cache file
if ( verbose )
fprintf(stderr, "update_dyld_shared_cache: writing .map file to disk\n");
@@ -2093,7 +3433,7 @@
sprintf(tempMapFilePath, "%s.map%u", fCacheFilePath, getpid());
FILE* fmap = ::fopen(tempMapFilePath, "w");
if ( fmap == NULL ) {
- fprintf(stderr, "can't create map file %s, errnor=%d", tempCachePath, errno);
+ fprintf(stderr, "can't create map file %s, errno=%d", tempMapFilePath, errno);
}
else {
for(std::vector<shared_file_mapping_np>::iterator it = fMappings.begin(); it != fMappings.end(); ++it) {
@@ -2105,36 +3445,80 @@
else if ( it->sfm_init_prot == (VM_PROT_EXECUTE|VM_PROT_WRITE|VM_PROT_READ) )
prot = "WX";
if ( it->sfm_size > 1024*1024 )
- fprintf(fmap, "mapping %s %4lluMB 0x%0llX -> 0x%0llX\n", prot, it->sfm_size/(1024*1024),
+ fprintf(fmap, "mapping %s %4lluMB 0x%0llX -> 0x%0llX\n", prot, it->sfm_size/(1024*1024),
it->sfm_address, it->sfm_address+it->sfm_size);
else
- fprintf(fmap, "mapping %s %4lluKB 0x%0llX -> 0x%0llX\n", prot, it->sfm_size/1024,
+ fprintf(fmap, "mapping %s %4lluKB 0x%0llX -> 0x%0llX\n", prot, it->sfm_size/1024,
it->sfm_address, it->sfm_address+it->sfm_size);
}
- fprintf(fmap, "linkedit %4uKB 0x%0llX -> 0x%0llX weak binding info\n",
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX weak binding info\n",
(fOffsetOfExportInfoInCombinedLinkedit-fOffsetOfWeakBindInfoInCombinedLinkedit)/1024,
fLinkEditsStartAddress+fOffsetOfWeakBindInfoInCombinedLinkedit,
fLinkEditsStartAddress+fOffsetOfExportInfoInCombinedLinkedit);
- fprintf(fmap, "linkedit %4uKB 0x%0llX -> 0x%0llX export info\n",
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX export info\n",
(fOffsetOfBindInfoInCombinedLinkedit-fOffsetOfExportInfoInCombinedLinkedit)/1024,
fLinkEditsStartAddress+fOffsetOfExportInfoInCombinedLinkedit,
fLinkEditsStartAddress+fOffsetOfBindInfoInCombinedLinkedit);
- fprintf(fmap, "linkedit %4uKB 0x%0llX -> 0x%0llX binding info\n",
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX binding info\n",
(fOffsetOfLazyBindInfoInCombinedLinkedit-fOffsetOfBindInfoInCombinedLinkedit)/1024,
fLinkEditsStartAddress+fOffsetOfBindInfoInCombinedLinkedit,
fLinkEditsStartAddress+fOffsetOfLazyBindInfoInCombinedLinkedit);
- fprintf(fmap, "linkedit %4uKB 0x%0llX -> 0x%0llX lazy binding info\n",
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX lazy binding info\n",
(fOffsetOfOldSymbolTableInfoInCombinedLinkedit-fOffsetOfLazyBindInfoInCombinedLinkedit)/1024,
fLinkEditsStartAddress+fOffsetOfLazyBindInfoInCombinedLinkedit,
fLinkEditsStartAddress+fOffsetOfOldSymbolTableInfoInCombinedLinkedit);
- fprintf(fmap, "linkedit %4uMB 0x%0llX -> 0x%0llX non-dyld symbol table info\n",
- (fLinkEditsTotalOptimizedSize-fOffsetOfOldSymbolTableInfoInCombinedLinkedit)/(1024*1024),
+ fprintf(fmap, " linkedit %4uMB 0x%0llX -> 0x%0llX non-dyld symbol table size\n",
+ (fSizeOfOldSymbolTableInfoInCombinedLinkedit)/(1024*1024),
fLinkEditsStartAddress+fOffsetOfOldSymbolTableInfoInCombinedLinkedit,
- fLinkEditsStartAddress+fLinkEditsTotalOptimizedSize);
+ fLinkEditsStartAddress+fOffsetOfOldSymbolTableInfoInCombinedLinkedit+fSizeOfOldSymbolTableInfoInCombinedLinkedit);
+ if ( fSizeOfFunctionStartsInCombinedLinkedit != 0 )
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX non-dyld functions starts size\n",
+ fSizeOfFunctionStartsInCombinedLinkedit/1024,
+ fLinkEditsStartAddress+fOffsetOfFunctionStartsInCombinedLinkedit,
+ fLinkEditsStartAddress+fOffsetOfFunctionStartsInCombinedLinkedit+fSizeOfFunctionStartsInCombinedLinkedit);
+ if ( fSizeOfDataInCodeInCombinedLinkedit != 0 )
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX non-dyld data-in-code info size\n",
+ fSizeOfDataInCodeInCombinedLinkedit/1024,
+ fLinkEditsStartAddress+fOffsetOfDataInCodeInCombinedLinkedit,
+ fLinkEditsStartAddress+fOffsetOfDataInCodeInCombinedLinkedit+fSizeOfDataInCodeInCombinedLinkedit);
+ if ( fSizeOfOldExternalRelocationsInCombinedLinkedit != 0 )
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX non-dyld external relocs size\n",
+ fSizeOfOldExternalRelocationsInCombinedLinkedit/1024,
+ fLinkEditsStartAddress+fOffsetOfOldExternalRelocationsInCombinedLinkedit,
+ fLinkEditsStartAddress+fOffsetOfOldExternalRelocationsInCombinedLinkedit+fSizeOfOldExternalRelocationsInCombinedLinkedit);
+ fprintf(fmap, " linkedit %4uKB 0x%0llX -> 0x%0llX non-dyld indirect symbol table size\n",
+ fSizeOfOldIndirectSymbolsInCombinedLinkedit/1024,
+ fLinkEditsStartAddress+fOffsetOfOldIndirectSymbolsInCombinedLinkedit,
+ fLinkEditsStartAddress+fOffsetOfOldIndirectSymbolsInCombinedLinkedit+fSizeOfOldIndirectSymbolsInCombinedLinkedit);
+ fprintf(fmap, " linkedit %4uMB 0x%0llX -> 0x%0llX non-dyld string pool\n",
+ (fSizeOfOldStringPoolInCombinedLinkedit)/(1024*1024),
+ fLinkEditsStartAddress+fOffsetOfOldStringPoolInCombinedLinkedit,
+ fLinkEditsStartAddress+fOffsetOfOldStringPoolInCombinedLinkedit+fSizeOfOldStringPoolInCombinedLinkedit);
+
+ dyldCacheHeader<E>* cacheHeader = (dyldCacheHeader<E>*)inMemoryCache;
+ if ( cacheHeader->slideInfoSize() != 0 ) {
+ fprintf(fmap, " linkedit %4lluKB kernel slide info\n", (cacheHeader->slideInfoSize())/1024);
+ }
+
+ fprintf(fmap, "unmapped -- %4uMB local symbol info\n", fUnmappedLocalSymbolsSize/(1024*1024));
+
+ uint64_t endMappingAddr = fMappings[2].sfm_address + fMappings[2].sfm_size;
+ fprintf(fmap, "total map %4lluMB\n", (endMappingAddr - sharedRegionStartAddress())/(1024*1024));
+ if ( sharedRegionStartWritableAddress(0) == 0x7FFF70000000LL ) {
+ // x86_64 has different slide constraints
+ uint64_t freeSpace = 256*1024*1024 - fMappings[1].sfm_size;
+ fprintf(fmap, "r/w space %4lluMB -> %d bits of entropy for ASLR\n\n", freeSpace/(1024*1024), (int)log2(freeSpace/4096));
+ }
+ else {
+ uint64_t freeSpace = sharedRegionStartAddress() + sharedRegionSize() - endMappingAddr;
+ fprintf(fmap, "free space %4lluMB -> %d bits of entropy for ASLR\n\n", freeSpace/(1024*1024), (int)log2(freeSpace/4096));
+ }
for(typename std::vector<LayoutInfo>::const_iterator it = fDylibs.begin(); it != fDylibs.end(); ++it) {
fprintf(fmap, "%s\n", it->layout->getID().name);
+ for (std::vector<const char*>::const_iterator ait = it->aliases.begin(); ait != it->aliases.end(); ++ait)
+ fprintf(fmap, "%s\n", *ait);
const std::vector<MachOLayoutAbstraction::Segment>& segs = it->layout->getSegments();
for (int i=0; i < segs.size(); ++i) {
const MachOLayoutAbstraction::Segment& seg = segs[i];
@@ -2148,7 +3532,7 @@
}
}
fclose(fmap);
- result = ::rename(tempMapFilePath, mapFilePath);
+ ::rename(tempMapFilePath, mapFilePath);
}
}
@@ -2161,8 +3545,6 @@
}
}
catch (...){
- // remove temp cache file
- ::unlink(tempCachePath);
// remove in memory cache
if ( inMemoryCache != NULL )
vm_deallocate(mach_task_self(), (vm_address_t)inMemoryCache, allocatedCacheSize);
@@ -2243,10 +3625,11 @@
-static void setSharedDylibs(const char* rootPath, bool usesOverlay, const std::set<ArchPair>& onlyArchs, std::vector<const char*> rootsPaths)
+static void setSharedDylibs(const char* rootPath, const std::vector<const char*>& overlayPaths, const std::set<ArchPair>& onlyArchs, std::vector<const char*> rootsPaths)
{
// set file system root
- ArchGraph::setFileSystemRoot(rootPath, usesOverlay);
+ ArchGraph::setFileSystemRoot(rootPath);
+ ArchGraph::setFileSystemOverlay(overlayPaths);
// initialize all architectures requested
for(std::set<ArchPair>::iterator a = onlyArchs.begin(); a != onlyArchs.end(); ++a)
@@ -2262,12 +3645,11 @@
}
-static void scanForSharedDylibs(const char* rootPath, bool usesOverlay, const char* dirOfPathFiles, const std::set<ArchPair>& onlyArchs)
+static void scanForSharedDylibs(const char* rootPath, const std::vector<const char*>& overlayPaths, const char* dirOfPathFiles, const std::set<ArchPair>& onlyArchs)
{
char rootDirOfPathFiles[strlen(rootPath)+strlen(dirOfPathFiles)+2];
- // in -overlay mode, still look for roots in /var/db/dyld
// in -root mode, look for roots in /rootpath/var/db/dyld
- if ( !usesOverlay && (strlen(rootPath) != 0) ) {
+ if ( rootPath[0] != '\0' ) {
strcpy(rootDirOfPathFiles, rootPath);
strcat(rootDirOfPathFiles, dirOfPathFiles);
dirOfPathFiles = rootDirOfPathFiles;
@@ -2281,14 +3663,23 @@
if ( dir == NULL )
throwf("%s does not exist, errno=%d\n", dirOfPathFiles, errno);
for (dirent* entry = ::readdir(dir); entry != NULL; entry = ::readdir(dir)) {
- if ( entry->d_type == DT_REG ) {
- // only look at files ending in .paths
+ if ( entry->d_type == DT_REG || entry->d_type == DT_UNKNOWN ) {
+ // only look at regular files ending in .paths
if ( strcmp(&entry->d_name[entry->d_namlen-6], ".paths") == 0 ) {
+ struct stat tmpStatPathsFile;
char fullPath[strlen(dirOfPathFiles)+entry->d_namlen+2];
strcpy(fullPath, dirOfPathFiles);
strcat(fullPath, "/");
strcat(fullPath, entry->d_name);
- parsePathsFile(fullPath, rootsPaths);
+ if ( lstat(fullPath, &tmpStatPathsFile) == -1 ) {
+ fprintf(stderr, "update_dyld_shared_cache: can't access %s\n", fullPath);
+ }
+ else if ( S_ISREG(tmpStatPathsFile.st_mode) ) {
+ parsePathsFile(fullPath, rootsPaths);
+ }
+ else {
+ fprintf(stderr, "update_dyld_shared_cache: wrong file type for %s\n", fullPath);
+ }
}
else {
fprintf(stderr, "update_dyld_shared_cache: warning, ignore file with wrong extension: %s\n", entry->d_name);
@@ -2299,14 +3690,14 @@
if ( rootsPaths.size() == 0 )
fprintf(stderr, "update_dyld_shared_cache: warning, no entries found in shared_region_roots\n");
- setSharedDylibs(rootPath, usesOverlay, onlyArchs, rootsPaths);
-}
-
-static void setSharedDylibs(const char* rootPath, bool usesOverlay, const char* pathsFile, const std::set<ArchPair>& onlyArchs)
+ setSharedDylibs(rootPath, overlayPaths, onlyArchs, rootsPaths);
+}
+
+static void setSharedDylibs(const char* rootPath, const std::vector<const char*>& overlayPaths, const char* pathsFile, const std::set<ArchPair>& onlyArchs)
{
std::vector<const char*> rootsPaths;
parsePathsFile(pathsFile, rootsPaths);
- setSharedDylibs(rootPath, usesOverlay, onlyArchs, rootsPaths);
+ setSharedDylibs(rootPath, overlayPaths, onlyArchs, rootsPaths);
}
@@ -2317,15 +3708,15 @@
// creation time was before the last restart of this machine.
static void deleteOrphanTempCacheFiles()
{
- DIR* dir = ::opendir(DYLD_SHARED_CACHE_DIR);
+ DIR* dir = ::opendir(MACOSX_DYLD_SHARED_CACHE_DIR);
if ( dir != NULL ) {
std::vector<const char*> filesToDelete;
for (dirent* entry = ::readdir(dir); entry != NULL; entry = ::readdir(dir)) {
if ( entry->d_type == DT_REG ) {
// only look at files with .tmp in name
if ( strstr(entry->d_name, ".tmp") != NULL ) {
- char fullPath[strlen(DYLD_SHARED_CACHE_DIR)+entry->d_namlen+2];
- strcpy(fullPath, DYLD_SHARED_CACHE_DIR);
+ char fullPath[strlen(MACOSX_DYLD_SHARED_CACHE_DIR)+entry->d_namlen+2];
+ strcpy(fullPath, MACOSX_DYLD_SHARED_CACHE_DIR);
strcat(fullPath, "/");
strcat(fullPath, entry->d_name);
struct stat tmpFileStatInfo;
@@ -2353,8 +3744,8 @@
-static bool updateSharedeCacheFile(const char* rootPath, bool usesOverlay, const char* cacheDir, const std::set<ArchPair>& onlyArchs,
- bool force, bool alphaSort, bool optimize, bool deleteExistingFirst, bool verify, bool keepSignatures)
+static bool updateSharedeCacheFile(const char* rootPath, const std::vector<const char*>& overlayPaths, const char* cacheDir, bool explicitCacheDir, const std::set<ArchPair>& onlyArchs,
+ bool force, bool alphaSort, bool optimize, bool deleteExistingFirst, bool verify, bool keepSignatures, bool dontMapLocalSymbols)
{
bool didUpdate = false;
// get dyld load address info
@@ -2379,40 +3770,35 @@
else
fprintf(stderr, "update_dyld_shared_cache: warning, dyld not available for specified architectures\n");
switch ( a->arch ) {
- case CPU_TYPE_POWERPC:
- {
- #if __i386__ || __x86_64__
- // <rdar://problem/5217377> Rosetta does not work with optimized dyld shared cache
- SharedCache<ppc> cache(ArchGraph::graphForArchPair(*a), rootPath, alphaSort, verify, false, usesOverlay, dyldBaseAddress);
- didUpdate |= cache.update(usesOverlay, force, false, deleteExistingFirst, index, archCount, keepSignatures);
- #else
- SharedCache<ppc> cache(ArchGraph::graphForArchPair(*a), rootPath, alphaSort, verify, optimize, usesOverlay, dyldBaseAddress);
- didUpdate |= cache.update(usesOverlay, force, optimize, deleteExistingFirst, index, archCount, keepSignatures);
- #endif
- }
- break;
case CPU_TYPE_I386:
{
- SharedCache<x86> cache(ArchGraph::graphForArchPair(*a), rootPath, alphaSort, verify, optimize, usesOverlay, dyldBaseAddress);
- didUpdate |= cache.update(usesOverlay, force, optimize, deleteExistingFirst, index, archCount, keepSignatures);
+ SharedCache<x86> cache(ArchGraph::graphForArchPair(*a), rootPath, overlayPaths, cacheDir, explicitCacheDir, alphaSort, verify, optimize, dyldBaseAddress);
+ didUpdate |= cache.update(force, optimize, deleteExistingFirst, index, archCount, keepSignatures, dontMapLocalSymbols);
}
break;
case CPU_TYPE_X86_64:
{
- SharedCache<x86_64> cache(ArchGraph::graphForArchPair(*a), rootPath, alphaSort, verify, optimize, usesOverlay, dyldBaseAddress);
- didUpdate |= cache.update(usesOverlay, force, optimize, deleteExistingFirst, index, archCount, keepSignatures);
+ SharedCache<x86_64> cache(ArchGraph::graphForArchPair(*a), rootPath, overlayPaths, cacheDir, explicitCacheDir, alphaSort, verify, optimize, dyldBaseAddress);
+ didUpdate |= cache.update(force, optimize, deleteExistingFirst, index, archCount, keepSignatures, dontMapLocalSymbols);
}
break;
case CPU_TYPE_ARM:
{
- SharedCache<arm> cache(ArchGraph::graphForArchPair(*a), rootPath, alphaSort, verify, optimize, usesOverlay, dyldBaseAddress);
- didUpdate |= cache.update(usesOverlay, force, optimize, deleteExistingFirst, index, archCount, keepSignatures);
+ SharedCache<arm> cache(ArchGraph::graphForArchPair(*a), rootPath, overlayPaths, cacheDir, explicitCacheDir, alphaSort, verify, optimize, dyldBaseAddress);
+ didUpdate |= cache.update(force, optimize, deleteExistingFirst, index, archCount, keepSignatures, dontMapLocalSymbols);
}
break;
- }
- }
-
- deleteOrphanTempCacheFiles();
+ case CPU_TYPE_ARM64:
+ {
+ SharedCache<arm64> cache(ArchGraph::graphForArchPair(*a), rootPath, overlayPaths, cacheDir, explicitCacheDir, alphaSort, verify, optimize, dyldBaseAddress);
+ didUpdate |= cache.update(force, optimize, deleteExistingFirst, index, archCount, keepSignatures, dontMapLocalSymbols);
+ }
+ break;
+ }
+ }
+
+ if ( !iPhoneOS )
+ deleteOrphanTempCacheFiles();
return didUpdate;
}
@@ -2428,14 +3814,17 @@
{
std::set<ArchPair> onlyArchs;
const char* rootPath = "";
+ std::vector<const char*> overlayPaths;
const char* dylibListFile = NULL;
bool force = false;
bool alphaSort = false;
bool optimize = true;
- bool hasRoot = false;
- bool hasOverlay = false;
bool verify = false;
bool keepSignatures = false;
+ bool explicitCacheDir = false;
+ bool dontMapLocalSymbols = false;
+ bool relaunchForHaswell = false;
+ const char* cacheDir = NULL;
try {
// parse command line options
@@ -2463,36 +3852,43 @@
else if ( strcmp(arg, "-no_opt") == 0 ) {
optimize = false;
}
+ else if ( strcmp(arg, "-dont_map_local_symbols") == 0 ) {
+ dontMapLocalSymbols = true;
+ }
+ else if ( strcmp(arg, "-iPhone") == 0 ) {
+ iPhoneOS = true;
+ alphaSort = true;
+ }
else if ( strcmp(arg, "-dylib_list") == 0 ) {
dylibListFile = argv[++i];
if ( dylibListFile == NULL )
throw "-dylib_list missing path argument";
- keepSignatures = true;
}
else if ( (strcmp(arg, "-root") == 0) || (strcmp(arg, "--root") == 0) ) {
- if ( hasOverlay )
- throw "cannot use both -root and -overlay";
rootPath = argv[++i];
if ( rootPath == NULL )
throw "-root missing path argument";
- hasRoot = true;
}
else if ( strcmp(arg, "-overlay") == 0 ) {
- if ( hasRoot )
- throw "cannot use both -root and -overlay";
- rootPath = argv[++i];
- if ( rootPath == NULL )
- throw "-root missing path argument";
- hasOverlay = true;
+ const char* path = argv[++i];
+ if ( path == NULL )
+ throw "-overlay missing path argument";
+ overlayPaths.push_back(path);
+ }
+ else if ( strcmp(arg, "-cache_dir") == 0 ) {
+ cacheDir = argv[++i];
+ if ( cacheDir == NULL )
+ throw "-cache_dir missing path argument";
+ explicitCacheDir = true;
}
else if ( strcmp(arg, "-arch") == 0 ) {
const char* arch = argv[++i];
- if ( strcmp(arch, "ppc") == 0 )
- onlyArchs.insert(ArchPair(CPU_TYPE_POWERPC, CPU_SUBTYPE_POWERPC_ALL));
- else if ( strcmp(arch, "i386") == 0 )
+ if ( strcmp(arch, "i386") == 0 )
onlyArchs.insert(ArchPair(CPU_TYPE_I386, CPU_SUBTYPE_I386_ALL));
else if ( strcmp(arch, "x86_64") == 0 )
onlyArchs.insert(ArchPair(CPU_TYPE_X86_64, CPU_SUBTYPE_X86_64_ALL));
+ else if ( strcmp(arch, "x86_64h") == 0 )
+ onlyArchs.insert(ArchPair(CPU_TYPE_X86_64, CPU_SUBTYPE_X86_64_H));
else if ( strcmp(arch, "armv4t") == 0 )
onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V4T));
else if ( strcmp(arch, "armv5") == 0 )
@@ -2501,15 +3897,21 @@
onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V6));
else if ( strcmp(arch, "armv7") == 0 )
onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V7));
+ else if ( strcmp(arch, "armv7f") == 0 )
+ onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V7F));
+ else if ( strcmp(arch, "armv7k") == 0 )
+ onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V7K));
+ else if ( strcmp(arch, "armv7s") == 0 )
+ onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V7S));
+ else if ( strcmp(arch, "arm64") == 0 )
+ onlyArchs.insert(ArchPair(CPU_TYPE_ARM64, CPU_SUBTYPE_ARM64_ALL));
else
throwf("unknown architecture %s", arch);
}
else if ( strcmp(arg, "-universal_boot") == 0 ) {
- #if __ppc__
- throwf("universal_boot option can only be used on Intel machines");
- #endif
onlyArchs.insert(ArchPair(CPU_TYPE_X86_64, CPU_SUBTYPE_X86_64_ALL));
onlyArchs.insert(ArchPair(CPU_TYPE_I386, CPU_SUBTYPE_I386_ALL));
+ relaunchForHaswell = true;
}
else {
usage();
@@ -2522,42 +3924,57 @@
}
}
- // strip tailing slashes on -root or -overlay
+ // strip tailing slashes on -root
+ // make it a real path so as to not make all dylibs look like symlink aliases
if ( rootPath[0] != '\0' ) {
- int len = strlen(rootPath)-1;
- if ( rootPath[len] == '/' ) {
- char* newRootPath = strdup(rootPath);
- while ( newRootPath[len] == '/' )
- newRootPath[len--] = '\0';
- rootPath = newRootPath;
- }
+ char realRootPath[MAXPATHLEN];
+ if ( realpath(rootPath, realRootPath) == NULL )
+ throwf("realpath() failed on %s\n", rootPath);
+ rootPath = strdup(realRootPath);
}
+ // strip tailing slashes on -overlay
+ for (std::vector<const char*>::iterator it=overlayPaths.begin(); it != overlayPaths.end(); ++it) {
+ char realOverlayPath[MAXPATHLEN];
+ if ( realpath(*it, realOverlayPath) == NULL )
+ throwf("realpath() failed on %s\n", *it);
+ *it = strdup(realOverlayPath);
+ }
+
+ // set default location to write cache dir
+ if ( cacheDir == NULL )
+ cacheDir = (iPhoneOS ? IPHONE_DYLD_SHARED_CACHE_DIR : MACOSX_DYLD_SHARED_CACHE_DIR);
+
// if no restrictions specified, use architectures that work on this machine
if ( onlyArchs.size() == 0 ) {
- int available;
- size_t len = sizeof(int);
- #if __i386__ || __x86_64__
- onlyArchs.insert(ArchPair(CPU_TYPE_I386, CPU_SUBTYPE_I386_ALL));
- // check rosetta is installed
- char rosettaPath[1024];
- strlcpy(rosettaPath, rootPath, 1024);
- strlcat(rosettaPath, "/usr/libexec/oah/translate", 1024);
- struct stat stat_buf;
- if ( stat(rosettaPath, &stat_buf) == 0 ) {
- onlyArchs.insert(ArchPair(CPU_TYPE_POWERPC, CPU_SUBTYPE_POWERPC_ALL));
- }
- else if ( hasOverlay ) {
- // in overlay mode, rosetta may be installed on base system, but is not in update root
- if ( stat("/usr/libexec/oah/translate", &stat_buf) == 0 )
- onlyArchs.insert(ArchPair(CPU_TYPE_POWERPC, CPU_SUBTYPE_POWERPC_ALL));
- }
- // check system is capable of running 64-bit programs
- if ( (sysctlbyname("hw.optional.x86_64", &available, &len, NULL, 0) == 0) && available )
- onlyArchs.insert(ArchPair(CPU_TYPE_X86_64, CPU_SUBTYPE_X86_64_ALL));
- #else
- #error unknown architecture
- #endif
+ if ( iPhoneOS ) {
+ onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V6));
+ onlyArchs.insert(ArchPair(CPU_TYPE_ARM, CPU_SUBTYPE_ARM_V7));
+ }
+ else {
+ int available;
+ size_t len = sizeof(int);
+ #if __i386__ || __x86_64__
+ onlyArchs.insert(ArchPair(CPU_TYPE_I386, CPU_SUBTYPE_I386_ALL));
+ // check system is capable of running 64-bit programs
+ if ( (sysctlbyname("hw.optional.x86_64", &available, &len, NULL, 0) == 0) && available ) {
+ // check system is capable of running x86_64h code
+ struct host_basic_info info;
+ mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT;
+ mach_port_t hostPort = mach_host_self();
+ kern_return_t result = host_info(hostPort, HOST_BASIC_INFO, (host_info_t)&info, &count);
+ mach_port_deallocate(mach_task_self(), hostPort);
+ if ( result != KERN_SUCCESS )
+ throw "host_info() failed";
+ if ( info.cpu_subtype == CPU_SUBTYPE_X86_64_H )
+ onlyArchs.insert(ArchPair(CPU_TYPE_X86_64, CPU_SUBTYPE_X86_64_H));
+ else
+ onlyArchs.insert(ArchPair(CPU_TYPE_X86_64, CPU_SUBTYPE_X86_ALL));
+ }
+ #else
+ #error unsupported architecture
+ #endif
+ }
}
if ( !verify && (geteuid() != 0) )
@@ -2565,11 +3982,44 @@
// build list of shared dylibs
if ( dylibListFile != NULL )
- setSharedDylibs(rootPath, hasOverlay, dylibListFile, onlyArchs);
+ setSharedDylibs(rootPath, overlayPaths, dylibListFile, onlyArchs);
else
- scanForSharedDylibs(rootPath, hasOverlay, "/var/db/dyld/shared_region_roots/", onlyArchs);
- updateSharedeCacheFile(rootPath, hasOverlay, DYLD_SHARED_CACHE_DIR, onlyArchs, force, alphaSort, optimize,
- false, verify, keepSignatures);
+ scanForSharedDylibs(rootPath, overlayPaths, "/var/db/dyld/shared_region_roots/", onlyArchs);
+ bool didUpdate = updateSharedeCacheFile(rootPath, overlayPaths, cacheDir, explicitCacheDir, onlyArchs, force, alphaSort, optimize,
+ false, verify, keepSignatures, dontMapLocalSymbols);
+
+ if ( didUpdate && !iPhoneOS ) {
+ void* handle = dlopen("/usr/lib/libspindump.dylib", RTLD_LAZY);
+ if ( handle != NULL ) {
+ typedef bool (*dscsym_proc_t)(const char *root);
+ dscsym_proc_t proc = (dscsym_proc_t)dlsym(handle, "dscsym_save_nuggets_for_current_caches");
+ const char* nuggetRootPath = "/";
+ if ( !overlayPaths.empty() )
+ nuggetRootPath = overlayPaths[0];
+ else if ( rootPath[0] != '\0' )
+ nuggetRootPath = rootPath;
+ (*proc)(nuggetRootPath);
+ }
+ dlclose(handle);
+ }
+
+ if ( relaunchForHaswell ) {
+ char cmd[2048];
+ strlcpy(cmd, argv[0], 2048);
+ strlcat(cmd, " -arch x86_64h", 2048);
+ if ( force )
+ strlcat(cmd, " -force", 2048);
+ if ( verify )
+ strlcat(cmd, " -verify", 2048);
+ if ( alphaSort )
+ strlcat(cmd, " -sort_by_name", 2048);
+ if ( (rootPath != NULL) && (rootPath[0] != '\0') ) {
+ strlcat(cmd, " -root ", 2048);
+ strlcat(cmd, rootPath, 2048);
+ }
+ return system(cmd);
+ }
+
}
catch (const char* msg) {
fprintf(stderr, "update_dyld_shared_cache failed: %s\n", msg);