]> git.saurik.com Git - apple/dyld.git/blobdiff - src/dyld_process_info.cpp
dyld-635.2.tar.gz
[apple/dyld.git] / src / dyld_process_info.cpp
index 849f17c1b923876d1d938d75580a0d3a3054f01c..e373fa9a32fbabd7933d5703a114cbb92c8a12de 100644 (file)
 #include "dyld_images.h"
 #include "dyld_priv.h"
 
+#include "Tracing.h"
+
+// this was in dyld_priv.h but it is no longer exported
+extern "C" {
+    const struct dyld_all_image_infos* _dyld_get_all_image_infos();
+}
+
+RemoteBuffer::RemoteBuffer() : _localAddress(0), _size(0) {}
+bool RemoteBuffer::map(task_t task, mach_vm_address_t remote_address, bool shared) {
+    vm_prot_t cur_protection = VM_PROT_NONE;
+    vm_prot_t max_protection = VM_PROT_NONE;
+    if (_size == 0) {
+        _kr = KERN_NO_SPACE;
+        return false;
+    }
+    _localAddress = 0;
+    _kr = mach_vm_remap(mach_task_self(),
+                        &_localAddress,
+                        _size,
+                        0,  // mask
+                        VM_FLAGS_ANYWHERE | VM_FLAGS_RETURN_DATA_ADDR | (shared ? 0 : VM_FLAGS_RESILIENT_CODESIGN),
+                        task,
+                        remote_address,
+                        !shared,
+                        &cur_protection,
+                        &max_protection,
+                        VM_INHERIT_NONE);
+    dyld3::kdebug_trace_dyld_marker(DBG_DYLD_DEBUGGING_VM_REMAP, _localAddress, (uint64_t)_size, _kr, remote_address);
+    if (shared && (cur_protection != (VM_PROT_READ|VM_PROT_WRITE))) {
+        if (_kr == KERN_SUCCESS && _localAddress != 0) {
+            _kr = vm_deallocate(mach_task_self(), _localAddress, _size);
+            dyld3::kdebug_trace_dyld_marker(DBG_DYLD_DEBUGGING_VM_UNMAP, _localAddress, (uint64_t)_size, _kr, 0);
+        }
+        _localAddress = 0;
+        _kr = KERN_PROTECTION_FAILURE;
+    }
+    return (_kr == KERN_SUCCESS);
+}
+
+RemoteBuffer::RemoteBuffer(task_t task, mach_vm_address_t remote_address, size_t remote_size, bool shared, bool allow_truncation)
+        : _localAddress(0), _size(remote_size), _kr(KERN_SUCCESS) {
+    // Try the initial map
+    if (map(task, remote_address, shared)) return;
+    // It failed, try to calculate the largest size that can fit in the same page as the remote_address
+    uint64_t newSize = PAGE_SIZE - remote_address%PAGE_SIZE;;
+    // If truncation is allowed and the newSize is different than the original size try that
+    if (!allow_truncation && newSize != _size) return;
+    _size = newSize;
+    if (map(task, remote_address, shared)) return;
+    // That did not work, null out the buffer
+    _size = 0;
+    _localAddress = 0;
+}
+RemoteBuffer::~RemoteBuffer() {
+    if (_localAddress) {
+        _kr = vm_deallocate(mach_task_self(), _localAddress, _size);
+        dyld3::kdebug_trace_dyld_marker(DBG_DYLD_DEBUGGING_VM_UNMAP, _localAddress, (uint64_t)_size, _kr, 0);
+    }
+}
+void *RemoteBuffer::getLocalAddress() { return (void *)_localAddress; }
+size_t RemoteBuffer::getSize() { return _size; }
+kern_return_t RemoteBuffer::getKernelReturn() { return _kr; }
+
+void withRemoteBuffer(task_t task, mach_vm_address_t remote_address, size_t remote_size, bool shared, bool allow_truncation, kern_return_t *kr, void (^block)(void *buffer, size_t size)) {
+    kern_return_t krSink = KERN_SUCCESS;
+    if (kr == nullptr) {
+        kr = &krSink;
+    }
+    RemoteBuffer buffer(task, remote_address, remote_size, shared, allow_truncation);
+    *kr = buffer.getKernelReturn();
+    if (*kr == KERN_SUCCESS) {
+        block(buffer.getLocalAddress(), buffer.getSize());
+    }
+}
+
 
 //
 // Opaque object returned by _dyld_process_info_create()
 //
 
+struct __attribute__((visibility("hidden"))) dyld_process_info_deleter { // deleter
+    //    dyld_process_info_deleter() {};
+    //    dyld_process_info_deleter(const dyld_process_info_deleter&) { }
+    //    dyld_process_info_deleter(dyld_process_info_deleter&) {}
+    //    dyld_process_info_deleter(dyld_process_info_deleter&&) {}
+    void operator()(dyld_process_info_base* p) const {
+        if (p) {
+            free(p);
+        }
+    };
+};
+
+static dyld_process_info_deleter deleter;
+typedef std::unique_ptr<dyld_process_info_base, dyld_process_info_deleter> dyld_process_info_ptr;
+
 struct __attribute__((visibility("hidden"))) dyld_process_info_base {
-    static dyld_process_info_base* make(task_t task, const dyld_all_image_infos_64& allImageInfo, const dyld_image_info_64 imageArray[], kern_return_t* kr);
-    static dyld_process_info_base* makeSuspended(task_t task, kern_return_t* kr);
+    template<typename T1, typename T2>
+    static dyld_process_info_ptr make(task_t task, const T1& allImageInfo, uint64_t timestamp, kern_return_t* kr);
+    template<typename T>
+    static dyld_process_info_ptr makeSuspended(task_t task, const T& allImageInfo, kern_return_t* kr);
 
-    uint32_t&                   retainCount() const { return _retainCount; }
+    std::atomic<uint32_t>&      retainCount() const { return _retainCount; }
     dyld_process_cache_info*    cacheInfo() const { return (dyld_process_cache_info*)(((char*)this) + _cacheInfoOffset); }
     dyld_process_state_info*    stateInfo() const { return (dyld_process_state_info*)(((char*)this) + _stateInfoOffset); }
     void                        forEachImage(void (^callback)(uint64_t machHeaderAddress, const uuid_t uuid, const char* path)) const;
     void                        forEachSegment(uint64_t machHeaderAddress, void (^callback)(uint64_t segmentAddress, uint64_t segmentSize, const char* segmentName)) const;
 
+    void retain()
+    {
+        _retainCount++;
+    }
+
+    void release()
+    {
+        uint32_t newCount = --_retainCount;
+
+        if ( newCount == 0 ) {
+            free(this);
+        }
+    }
+
 private:
     struct ImageInfo {
         uuid_t                  uuid;
@@ -73,19 +179,21 @@ private:
 
     static bool                 inCache(uint64_t addr) { return (addr > SHARED_REGION_BASE) && (addr < SHARED_REGION_BASE+SHARED_REGION_SIZE); }
     kern_return_t               addImage(task_t task, bool sameCacheAsThisProcess, uint64_t imageAddress, uint64_t imagePath, const char* imagePathLocal);
+
     kern_return_t               addDyldImage(task_t task, uint64_t dyldAddress, uint64_t dyldPathAddress, const char* localPath);
 
     bool                        invalid() { return ((char*)_stringRevBumpPtr < (char*)_curSegment); }
     const char*                 copyPath(task_t task, uint64_t pathAddr, kern_return_t* kr);
-    const char*                 addString(const char*);
+    const char*                 addString(const char*, size_t);
     const char*                 copySegmentName(const char*);
 
     void                        addInfoFromLoadCommands(const mach_header* mh, uint64_t addressInTask, size_t size);
+    kern_return_t               addInfoFromRemoteLoadCommands(task_t task, uint64_t remoteMH);
 
     void                        inspectLocalImageLoadCommands(uint64_t imageAddress, void* func);
     kern_return_t               inspectRemoteImageLoadCommands(task_t task, uint64_t imageAddress, void* func);
 
-    mutable uint32_t            _retainCount;
+    mutable std::atomic<uint32_t>            _retainCount;
     const uint32_t              _cacheInfoOffset;
     const uint32_t              _stateInfoOffset;
     const uint32_t              _imageInfosOffset;
@@ -118,131 +226,216 @@ dyld_process_info_base::dyld_process_info_base(unsigned imageCount, size_t total
 {
 }
 
-
-dyld_process_info_base* dyld_process_info_base::make(task_t task, const dyld_all_image_infos_64& allImageInfo, const dyld_image_info_64 imageArray[], kern_return_t* kr)
+template<typename T1, typename T2>
+dyld_process_info_ptr dyld_process_info_base::make(task_t task, const T1& allImageInfo, uint64_t timestamp, kern_return_t* kr)
 {
-    // figure out how many path strings will need to be copied and their size
-    const dyld_all_image_infos* myInfo = _dyld_get_all_image_infos();
-    bool sameCacheAsThisProcess = ((memcmp(myInfo->sharedCacheUUID, allImageInfo.sharedCacheUUID, 16) == 0) && (myInfo->sharedCacheSlide == allImageInfo.sharedCacheSlide));
-    unsigned countOfPathsNeedingCopying = 0;
-    if ( sameCacheAsThisProcess ) {
-        for (int i=0; i < allImageInfo.infoArrayCount; ++i) {
-            if ( !inCache(imageArray[i].imageFilePath) )
-                ++countOfPathsNeedingCopying;
+    __block dyld_process_info_ptr result = nullptr;
+
+    // bail out of dyld is too old
+    if ( allImageInfo.version < 15 ) {
+        *kr = KERN_FAILURE;
+        return nullptr;
+    }
+
+    // Check if the process is suspended
+    if (allImageInfo.infoArrayChangeTimestamp == 0) {
+        result = dyld_process_info_base::makeSuspended<T1>(task, allImageInfo, kr);
+        // If we have a result return it, otherwise rescan
+        if (result) {
+            // If it returned the process is suspended and there is nothing more to do
+            return std::move(result);
+        } else {
+            // Check to see if the process change timestamp is greater than 0, if not then sleep to let the process
+            // finish initializing
+            if (allImageInfo.infoArrayChangeTimestamp == 0) {
+                usleep(1000 * 50); // 50ms
+            }
         }
     }
-    else {
-        countOfPathsNeedingCopying = allImageInfo.infoArrayCount+1;
-    }
-    unsigned imageCountWithDyld = allImageInfo.infoArrayCount+1;
 
-    // allocate result object
-    size_t allocationSize =   sizeof(dyld_process_info_base)
-                            + sizeof(dyld_process_cache_info)
-                            + sizeof(dyld_process_state_info)
-                            + sizeof(ImageInfo)*(imageCountWithDyld)
-                            + sizeof(SegmentInfo)*imageCountWithDyld*5
-                            + countOfPathsNeedingCopying*PATH_MAX;
-    void* storage = malloc(allocationSize);
-    dyld_process_info_base* obj =  new (storage) dyld_process_info_base(imageCountWithDyld, allocationSize); // placement new()
+    // Test to see if there are no changes and we can exit early
+    if (timestamp != 0 && timestamp == allImageInfo.infoArrayChangeTimestamp) {
+        *kr = KERN_SUCCESS;
+        return nullptr;
+    }
+
+    for(uint32_t i = 0; i < 10; ++i) {
+        uint64_t currentTimestamp = allImageInfo.infoArrayChangeTimestamp;
+        mach_vm_address_t infoArray = allImageInfo.infoArray;
+        if (currentTimestamp == 0) continue;
+        if (infoArray == 0) {
+            // Check if the task is suspended mid dylib load and exit early
+            mach_task_basic_info ti;
+            mach_msg_type_number_t count = MACH_TASK_BASIC_INFO_COUNT;
+            if ((*kr = task_info(task, MACH_TASK_BASIC_INFO, (task_info_t)&ti, &count))) {
+                continue;
+            }
 
-    // fill in base info
-    dyld_process_cache_info* cacheInfo = obj->cacheInfo();
-    memcpy(cacheInfo->cacheUUID, allImageInfo.sharedCacheUUID, 16);
-    cacheInfo->cacheBaseAddress    = allImageInfo.sharedCacheBaseAddress;
-    cacheInfo->privateCache        = allImageInfo.processDetachedFromSharedRegion;
-    // if no cache is used, allImageInfo has all zeros for cache UUID
-    cacheInfo->noCache = true;
-    for (int i=0; i < 16; ++i) {
-        if ( cacheInfo->cacheUUID[i] != 0 ) {
-            cacheInfo->noCache = false;
-        }
-    }
+            // The task is suspended, exit
+            if (ti.suspend_count != 0) {
+                // Not exactly correct, but conveys that operation may succeed in the future
+                *kr = KERN_RESOURCE_SHORTAGE;
+                return  nullptr;
+            }
+            continue;
+        };
+
+        // For the moment we are going to truncate any image list longer than 8192 because some programs do
+        // terrible things that corrupt their own image lists and we need to stop clients from crashing
+        // reading them. We can try to do something more advanced in the future. rdar://27446361
+        uint32_t imageCount = allImageInfo.infoArrayCount;
+        imageCount = MIN(imageCount, 8192);
+        size_t imageArraySize = imageCount * sizeof(T2);
+
+        withRemoteBuffer(task, infoArray, imageArraySize, false, false, kr, ^(void *buffer, size_t size) {
+            // figure out how many path strings will need to be copied and their size
+            T2* imageArray = (T2 *)buffer;
+            const dyld_all_image_infos* myInfo = _dyld_get_all_image_infos();
+            bool sameCacheAsThisProcess = !allImageInfo.processDetachedFromSharedRegion
+                && !myInfo->processDetachedFromSharedRegion
+                && ((memcmp(myInfo->sharedCacheUUID, &allImageInfo.sharedCacheUUID[0], 16) == 0)
+                && (myInfo->sharedCacheSlide == allImageInfo.sharedCacheSlide));
+            unsigned countOfPathsNeedingCopying = 0;
+            if ( sameCacheAsThisProcess ) {
+                for (uint32_t i=0; i < imageCount; ++i) {
+                    if ( !inCache(imageArray[i].imageFilePath) )
+                        ++countOfPathsNeedingCopying;
+                }
+            }
+            else {
+                countOfPathsNeedingCopying = imageCount+1;
+            }
+            unsigned imageCountWithDyld = imageCount+1;
+
+            // allocate result object
+            size_t allocationSize = sizeof(dyld_process_info_base)
+                                        + sizeof(dyld_process_cache_info)
+                                        + sizeof(dyld_process_state_info)
+                                        + sizeof(ImageInfo)*(imageCountWithDyld)
+                                        + sizeof(SegmentInfo)*imageCountWithDyld*5
+                                        + countOfPathsNeedingCopying*PATH_MAX;
+            void* storage = malloc(allocationSize);
+            auto info = dyld_process_info_ptr(new (storage) dyld_process_info_base(imageCountWithDyld, allocationSize), deleter);
+            //info =  new (storage) dyld_process_info_base(imageCountWithDyld, allocationSize); // placement new()
+
+            // fill in base info
+            dyld_process_cache_info* cacheInfo = info->cacheInfo();
+            memcpy(cacheInfo->cacheUUID, &allImageInfo.sharedCacheUUID[0], 16);
+            cacheInfo->cacheBaseAddress    = allImageInfo.sharedCacheBaseAddress;
+            cacheInfo->privateCache        = allImageInfo.processDetachedFromSharedRegion;
+            // if no cache is used, allImageInfo has all zeros for cache UUID
+            cacheInfo->noCache = true;
+            for (int i=0; i < 16; ++i) {
+                if ( cacheInfo->cacheUUID[i] != 0 ) {
+                    cacheInfo->noCache = false;
+                }
+            }
 
-    dyld_process_state_info* stateInfo = obj->stateInfo();
-    stateInfo->timestamp           = allImageInfo.infoArrayChangeTimestamp;
-    stateInfo->imageCount          = imageCountWithDyld;
-    stateInfo->initialImageCount   = (uint32_t)(allImageInfo.initialImageCount+1);
-    if ( allImageInfo.infoArray != 0 )
-        stateInfo->dyldState = dyld_process_state_dyld_initialized;
-    if ( allImageInfo.libSystemInitialized != 0 ) {
-        stateInfo->dyldState = dyld_process_state_libSystem_initialized;
-        if ( allImageInfo.initialImageCount != allImageInfo.infoArrayCount )
-            stateInfo->dyldState = dyld_process_state_program_running;
-    }
-    if ( allImageInfo.errorMessage != 0 )
-        stateInfo->dyldState = allImageInfo.terminationFlags ? dyld_process_state_terminated_before_inits : dyld_process_state_dyld_terminated;
+            dyld_process_state_info* stateInfo = info->stateInfo();
+            stateInfo->timestamp           = currentTimestamp;
+            stateInfo->imageCount          = imageCountWithDyld;
+            stateInfo->initialImageCount   = (uint32_t)(allImageInfo.initialImageCount+1);
+            stateInfo->dyldState = dyld_process_state_dyld_initialized;
 
-    // fill in info for dyld
-    if ( allImageInfo.dyldPath != 0 ) {
-        if ( kern_return_t r = obj->addDyldImage(task, allImageInfo.dyldImageLoadAddress, allImageInfo.dyldPath, NULL) ) {
-            if ( kr != NULL )
-                *kr = r;
-            goto fail;
-        }
-    }
-    
-    // fill in info for each image
-    for (uint32_t i=0; i < allImageInfo.infoArrayCount; ++i) {
-        if ( kern_return_t r = obj->addImage(task, sameCacheAsThisProcess, imageArray[i].imageLoadAddress, imageArray[i].imageFilePath, NULL) ) {
-           if ( kr != NULL )
-                *kr = r;
-            goto fail;
-        }
-    }
+            if ( allImageInfo.libSystemInitialized != 0 ) {
+                stateInfo->dyldState = dyld_process_state_libSystem_initialized;
+                if ( allImageInfo.initialImageCount != imageCount ) {
+                    stateInfo->dyldState = dyld_process_state_program_running;
+                }
+            }
+            if ( allImageInfo.errorMessage != 0 ) {
+                stateInfo->dyldState = allImageInfo.terminationFlags ? dyld_process_state_terminated_before_inits : dyld_process_state_dyld_terminated;
+            }
+            // fill in info for dyld
+            if ( allImageInfo.dyldPath != 0 ) {
+                if ((*kr = info->addDyldImage(task, allImageInfo.dyldImageLoadAddress, allImageInfo.dyldPath, NULL))) {
+                    result = nullptr;
+                    return;
+                }
+            }
+            // fill in info for each image
+            for (uint32_t i=0; i < imageCount; ++i) {
+                if ((*kr = info->addImage(task, sameCacheAsThisProcess, imageArray[i].imageLoadAddress, imageArray[i].imageFilePath, NULL))) {
+                    result = nullptr;
+                    return;
+                }
+            }
+            // sanity check internal data did not overflow
+            if ( info->invalid() ) {
+                *kr = KERN_FAILURE;
+                result = nullptr;
+                return;
+            }
 
-    // sanity check internal data did not overflow
-    if ( obj->invalid() )
-        goto fail;
+            result = std::move(info);
+        });
 
-    return obj;
+        if (result) break;
+    }
 
-fail:
-    free(obj);
-    return NULL;
+    return std::move(result);
 }
 
-dyld_process_info_base* dyld_process_info_base::makeSuspended(task_t task, kern_return_t* kr)
+template<typename T>
+dyld_process_info_ptr dyld_process_info_base::makeSuspended(task_t task, const T& allImageInfo, kern_return_t* kr)
 {
     pid_t   pid;
-    kern_return_t result = pid_for_task(task, &pid);
-    if ( result != KERN_SUCCESS ) {
-        if ( kr != NULL )
-            *kr = result;
+    if ((*kr = pid_for_task(task, &pid))) {
         return NULL;
     }
 
-    unsigned imageCount = 0;  // main executable and dyld
-    uint64_t            mainExecutableAddress = 0;
-    uint64_t            dyldAddress = 0;
+    mach_task_basic_info ti;
+    mach_msg_type_number_t count = MACH_TASK_BASIC_INFO_COUNT;
+    if ((*kr = task_info(task, MACH_TASK_BASIC_INFO, (task_info_t)&ti, &count))) {
+        return  nullptr;
+    }
+
+    // The task is not suspended, exit
+    if (ti.suspend_count == 0) {
+        return  nullptr;
+    }
+
+    __block unsigned    imageCount = 0;  // main executable and dyld
+    __block uint64_t    mainExecutableAddress = 0;
+    __block uint64_t    dyldAddress = 0;
     char                dyldPathBuffer[PATH_MAX+1];
     char                mainExecutablePathBuffer[PATH_MAX+1];
+    __block char *      dyldPath = &dyldPathBuffer[0];
+    __block char *      mainExecutablePath = &mainExecutablePathBuffer[0];
     mach_vm_size_t      size;
     for (mach_vm_address_t address = 0; ; address += size) {
         vm_region_basic_info_data_64_t  info;
         mach_port_t                     objectName;
         unsigned int                    infoCount = VM_REGION_BASIC_INFO_COUNT_64;
-        result = mach_vm_region(task, &address, &size, VM_REGION_BASIC_INFO,
-                         (vm_region_info_t)&info, &infoCount, &objectName);
-        if ( result != KERN_SUCCESS )
+        if (kern_return_t r = mach_vm_region(task, &address, &size, VM_REGION_BASIC_INFO,
+                                             (vm_region_info_t)&info, &infoCount, &objectName)) {
             break;
-        if ( info.protection == (VM_PROT_READ|VM_PROT_EXECUTE) ) {
-            if ( mainExecutableAddress == 0 ) {
-                mainExecutableAddress = address;
-                int len = proc_regionfilename(pid, mainExecutableAddress, mainExecutablePathBuffer, PATH_MAX);
-                if ( len != 0 )
-                    mainExecutablePathBuffer[len] = '\0';
-                ++imageCount;
-            }
-            else if ( dyldAddress == 0 ) {
-                dyldAddress = address;
-                int len = proc_regionfilename(pid, dyldAddress, dyldPathBuffer, PATH_MAX);
-                if ( len != 0 )
-                    dyldPathBuffer[len] = '\0';
-                ++imageCount;
-            }
-            //fprintf(stderr, "vm region: addr=0x%llX, size=0x%llX, prot=0x%X\n", (uint64_t)address, (uint64_t)size, info.protection);
         }
+        if ( info.protection != (VM_PROT_READ|VM_PROT_EXECUTE) )
+            continue;
+            // read start of vm region to verify it is a mach header
+            withRemoteObject(task, address, false, NULL, ^(mach_header_64 mhBuffer){
+                if ( (mhBuffer.magic != MH_MAGIC) && (mhBuffer.magic != MH_MAGIC_64) )
+                    return;
+                // now know the region is the start of a mach-o file
+                if ( mhBuffer.filetype == MH_EXECUTE ) {
+                    mainExecutableAddress = address;
+                    int len = proc_regionfilename(pid, mainExecutableAddress, mainExecutablePath, PATH_MAX);
+                    if ( len != 0 ) {
+                        mainExecutablePath[len] = '\0';
+                    }
+                    ++imageCount;
+                }
+                else if ( mhBuffer.filetype == MH_DYLINKER ) {
+                    dyldAddress = address;
+                    int len = proc_regionfilename(pid, dyldAddress, dyldPath, PATH_MAX);
+                    if ( len != 0 ) {
+                        dyldPath[len] = '\0';
+                    }
+                    ++imageCount;
+                }
+            });
+            //fprintf(stderr, "vm region: addr=0x%llX, size=0x%llX, prot=0x%X\n", (uint64_t)address, (uint64_t)size, info.protection);
     }
     //fprintf(stderr, "dyld: addr=0x%llX, path=%s\n", dyldAddress, dyldPathBuffer);
     //fprintf(stderr, "app:  addr=0x%llX, path=%s\n", mainExecutableAddress, mainExecutablePathBuffer);
@@ -255,8 +448,7 @@ dyld_process_info_base* dyld_process_info_base::makeSuspended(task_t task, kern_
                             + sizeof(SegmentInfo)*imageCount*5
                             + imageCount*PATH_MAX;
     void* storage = malloc(allocationSize);
-    dyld_process_info_base* obj =  new (storage) dyld_process_info_base(imageCount, allocationSize); // placement new()
-
+    auto obj = dyld_process_info_ptr(new (storage) dyld_process_info_base(imageCount, allocationSize), deleter);
     // fill in base info
     dyld_process_cache_info* cacheInfo = obj->cacheInfo();
     bzero(cacheInfo->cacheUUID, 16);
@@ -272,108 +464,77 @@ dyld_process_info_base* dyld_process_info_base::makeSuspended(task_t task, kern_
 
     // fill in info for dyld
     if ( dyldAddress != 0 ) {
-        if ( kern_return_t r = obj->addDyldImage(task, dyldAddress, 0, dyldPathBuffer) ) {
-            if ( kr != NULL )
-                *kr = r;
-            free(obj);
-            return NULL;
+        if ((*kr = obj->addDyldImage(task, dyldAddress, 0, dyldPath))) {
+            return nullptr;
         }
     }
 
     // fill in info for each image
     if ( mainExecutableAddress != 0 ) {
-        if ( kern_return_t r = obj->addImage(task, false, mainExecutableAddress, 0, mainExecutablePathBuffer) ) {
-            if ( kr != NULL )
-                *kr = r;
-            free(obj);
-            return NULL;
+        if ((*kr = obj->addImage(task, false, mainExecutableAddress, 0, mainExecutablePath))) {
+            return nullptr;
         }
     }
 
+    if (allImageInfo.infoArrayChangeTimestamp != 0) {
+        return  nullptr;
+    }
+
+    count = MACH_TASK_BASIC_INFO_COUNT;
+    if ((*kr = task_info(task, MACH_TASK_BASIC_INFO, (task_info_t)&ti, &count))) {
+        return  nullptr;
+    }
+
+    // The task is not suspended, exit
+    if (ti.suspend_count == 0) {
+        return  nullptr;
+    }
+
     return obj;
 }
 
 
 
-const char* dyld_process_info_base::addString(const char* str)
+const char* dyld_process_info_base::addString(const char* str, size_t maxlen)
 {
-    size_t len = strlen(str) + 1;
+    size_t len = strnlen(str, maxlen) + 1;
     _stringRevBumpPtr -= len;
-    strcpy(_stringRevBumpPtr, str);
+    strlcpy(_stringRevBumpPtr, str, len);
     return _stringRevBumpPtr;
 }
 
 const char* dyld_process_info_base::copyPath(task_t task, uint64_t stringAddressInTask, kern_return_t* kr)
 {
-    char temp[PATH_MAX+8]; // +8 is to allow '\0' at temp[PATH_MAX]
-    mach_vm_size_t readSize = PATH_MAX;
-    if ( ((stringAddressInTask & 0xFFF) + PATH_MAX) < 4096 ) {
-        // string fits within page, only one vm_read needed
-        if ( kern_return_t r = mach_vm_read_overwrite(task, stringAddressInTask, PATH_MAX, (vm_address_t)&temp, &readSize) ) {
-            if ( kr != NULL )
-                *kr = r;
-            return NULL;
-        }
-    }
-    else {
-        // string may cross page boundary, split into two reads
-        size_t firstLen = 4096 - (stringAddressInTask & 0xFFF);
-        readSize = firstLen;
-        if ( kern_return_t r = mach_vm_read_overwrite(task, stringAddressInTask, firstLen, (vm_address_t)&temp, &readSize) ) {
-            if ( kr != NULL )
-                *kr = r;
-            return NULL;
-        }
-        temp[firstLen] = '\0';
-        if ( strlen(temp) >= firstLen ) {
-            readSize = PATH_MAX-firstLen;
-            if ( kern_return_t r = mach_vm_read_overwrite(task, stringAddressInTask+firstLen, PATH_MAX-firstLen, (vm_address_t)&temp+firstLen, &readSize) ) {
-                if ( kr != NULL )
-                    *kr = r;
-                return NULL;
-                temp[PATH_MAX] = '\0'; // truncate any string that is too long
-            }
-        }
-    }
-    if ( kr != NULL )
-        *kr = KERN_SUCCESS;
-    return addString(temp);
+    __block const char* retval = NULL;
+    withRemoteBuffer(task, stringAddressInTask, PATH_MAX, false, true, kr, ^(void *buffer, size_t size) {
+        retval = addString(static_cast<const char *>(buffer), size);
+    });
+    return retval;
 }
 
-
 kern_return_t dyld_process_info_base::addImage(task_t task, bool sameCacheAsThisProcess, uint64_t imageAddress, uint64_t imagePath, const char* imagePathLocal)
 {
+    kern_return_t kr = KERN_SUCCESS;
     _curImage->loadAddress = imageAddress;
     _curImage->segmentStartIndex = _curSegmentIndex;
     if ( imagePathLocal != NULL ) {
-        _curImage->path = addString(imagePathLocal);
+        _curImage->path = addString(imagePathLocal, PATH_MAX);
     }
     else if ( sameCacheAsThisProcess && inCache(imagePath) ) {
         _curImage->path = (const char*)imagePath;
     }
     else {
-        kern_return_t kr;
         _curImage->path = copyPath(task, imagePath, &kr);
-        if ( kr )
+        if ( kr != KERN_SUCCESS)
             return kr;
     }
     if ( sameCacheAsThisProcess && inCache(imageAddress) ) {
         addInfoFromLoadCommands((mach_header*)imageAddress, imageAddress, 32*1024);
     }
     else {
-        mach_vm_size_t readSize = sizeof(mach_header_64);
-        mach_header_64 mhBuffer;
-        if ( kern_return_t r = mach_vm_read_overwrite(task, imageAddress, sizeof(mach_header_64), (vm_address_t)&mhBuffer, &readSize) ) {
-            return r;
-        }
-        size_t          headerPagesSize = (sizeof(mach_header_64) + mhBuffer.sizeofcmds + 4095) & (-4096);
-        vm_address_t    localCopyBuffer;
-        unsigned int    localCopyBufferSize;
-        if ( kern_return_t r = mach_vm_read(task, imageAddress, headerPagesSize, &localCopyBuffer, &localCopyBufferSize) ) {
-            return r;
-        }
-        addInfoFromLoadCommands((mach_header*)localCopyBuffer, imageAddress, localCopyBufferSize);
-        vm_deallocate(mach_task_self(), localCopyBuffer, localCopyBufferSize);
+        kr = addInfoFromRemoteLoadCommands(task, imageAddress);
+        if ( kr != KERN_SUCCESS)
+            return kr;
     }
     _curImage->segmentsCount = _curSegmentIndex - _curImage->segmentStartIndex;
     _curImage++;
@@ -381,33 +542,52 @@ kern_return_t dyld_process_info_base::addImage(task_t task, bool sameCacheAsThis
 }
 
 
+kern_return_t dyld_process_info_base::addInfoFromRemoteLoadCommands(task_t task, uint64_t remoteMH) {
+    __block kern_return_t kr = KERN_SUCCESS;
+    __block size_t headerPagesSize = 0;
+    __block bool done = false;
+
+    //Since the minimum we can reasonably map is a page, map that.
+    withRemoteBuffer(task, remoteMH, PAGE_SIZE, false, false, &kr, ^(void * buffer, size_t size) {
+        const mach_header* mh = (const mach_header*)buffer;
+        headerPagesSize = sizeof(mach_header) + mh->sizeofcmds;
+        if (headerPagesSize <= PAGE_SIZE) {
+            addInfoFromLoadCommands(mh, remoteMH, size);
+            done = true;
+        }
+    });
+
+    //The load commands did not fit in the first page, but now we know the size, so remap and try again
+    if (!done) {
+        if (kr != KERN_SUCCESS) {
+            return kr;
+        }
+        withRemoteBuffer(task, remoteMH, headerPagesSize, false, false, &kr, ^(void * buffer, size_t size) {
+            addInfoFromLoadCommands((mach_header*)buffer, remoteMH, size);
+        });
+    }
+
+    return kr;
+}
+
 kern_return_t dyld_process_info_base::addDyldImage(task_t task, uint64_t dyldAddress, uint64_t dyldPathAddress, const char* localPath)
 {
-    kern_return_t kr;
+    __block kern_return_t kr = KERN_SUCCESS;
     _curImage->loadAddress = dyldAddress;
     _curImage->segmentStartIndex = _curSegmentIndex;
     if ( localPath != NULL ) {
-        _curImage->path = addString(localPath);
+        _curImage->path = addString(localPath, PATH_MAX);
     }
     else {
         _curImage->path = copyPath(task, dyldPathAddress, &kr);
-        if ( kr )
+        if ( kr != KERN_SUCCESS)
             return kr;
     }
 
-    mach_vm_size_t readSize = sizeof(mach_header_64);
-    mach_header_64 mhBuffer;
-    if ( kern_return_t r = mach_vm_read_overwrite(task, dyldAddress, sizeof(mach_header_64), (vm_address_t)&mhBuffer, &readSize) ) {
-        return r;
-    }
-    size_t          headerPagesSize = (sizeof(mach_header_64) + mhBuffer.sizeofcmds + 4095) & (-4096);
-    vm_address_t    localCopyBuffer;
-    unsigned int    localCopyBufferSize;
-    if ( kern_return_t r = mach_vm_read(task, dyldAddress, headerPagesSize, &localCopyBuffer, &localCopyBufferSize) ) {
-        return r;
-    }
-    addInfoFromLoadCommands((mach_header*)localCopyBuffer, dyldAddress, localCopyBufferSize);
-    vm_deallocate(mach_task_self(), localCopyBuffer, localCopyBufferSize);
+    kr = addInfoFromRemoteLoadCommands(task, dyldAddress);
+    if ( kr != KERN_SUCCESS)
+        return kr;
+
     _curImage->segmentsCount = _curSegmentIndex - _curImage->segmentStartIndex;
     _curImage++;
     return KERN_SUCCESS;
@@ -423,7 +603,7 @@ void dyld_process_info_base::addInfoFromLoadCommands(const mach_header* mh, uint
         startCmds = (load_command*)((char *)mh + sizeof(mach_header));
     else
         return;  // not a mach-o file, or wrong endianness
-        
+
     const load_command* const cmdsEnd = (load_command*)((char*)startCmds + mh->sizeofcmds);
     const load_command* cmd = startCmds;
     for(uint32_t i = 0; i < mh->ncmds; ++i) {
@@ -464,7 +644,7 @@ const char* dyld_process_info_base::copySegmentName(const char* name)
         return *s;
     }
     // copy custom segment names into string pool
-    return addString(name);
+    return addString(name, 16);
 }
 
 void dyld_process_info_base::forEachImage(void (^callback)(uint64_t machHeaderAddress, const uuid_t uuid, const char* path)) const
@@ -479,14 +659,14 @@ void dyld_process_info_base::forEachSegment(uint64_t machHeaderAddress, void (^c
     for (const ImageInfo* p = _firstImage; p < _curImage; ++p) {
         if ( p->loadAddress == machHeaderAddress ) {
             uint64_t slide = 0;
-            for (int i=0; i < p->segmentsCount; ++i) {
+            for (uint32_t i=0; i < p->segmentsCount; ++i) {
                 const SegmentInfo* seg = &_firstSegment[p->segmentStartIndex+i];
                 if ( strcmp(seg->name, "__TEXT") == 0 ) {
                     slide = machHeaderAddress - seg->addr;
                     break;
                 }
             }
-            for (int i=0; i < p->segmentsCount; ++i) {
+            for (uint32_t i=0; i < p->segmentsCount; ++i) {
                 const SegmentInfo* seg = &_firstSegment[p->segmentStartIndex+i];
                 callback(seg->addr + slide, seg->size, seg->name);
             }
@@ -495,122 +675,49 @@ void dyld_process_info_base::forEachSegment(uint64_t machHeaderAddress, void (^c
     }
 }
 
-
-
-
-
-// Implementation that works with existing dyld data structures
 dyld_process_info _dyld_process_info_create(task_t task, uint64_t timestamp, kern_return_t* kr)
 {
-    if ( kr != NULL )
-        *kr = KERN_SUCCESS;
+    __block dyld_process_info result = nullptr;
+    kern_return_t krSink = KERN_SUCCESS;
+    if (kr == nullptr) {
+        kr = &krSink;
+    }
+    *kr = KERN_SUCCESS;
 
     task_dyld_info_data_t task_dyld_info;
     mach_msg_type_number_t count = TASK_DYLD_INFO_COUNT;
     if ( kern_return_t r = task_info(task, TASK_DYLD_INFO, (task_info_t)&task_dyld_info, &count) ) {
-        if ( kr != NULL )
-            *kr = r;
-        return  NULL;
+        *kr = r;
+        return  nullptr;
     }
 
     //The kernel will return MACH_VM_MIN_ADDRESS for an executable that has not had dyld loaded
     if (task_dyld_info.all_image_info_addr == MACH_VM_MIN_ADDRESS)
-        return NULL;
+        return nullptr;
 
     if ( task_dyld_info.all_image_info_size > sizeof(dyld_all_image_infos_64) )
-        return NULL;
-
-    // read all_image_infos struct
-    dyld_all_image_infos_64 allImageInfo64;
-    mach_vm_size_t readSize = task_dyld_info.all_image_info_size;
-    if ( kern_return_t r = mach_vm_read_overwrite(task, task_dyld_info.all_image_info_addr, task_dyld_info.all_image_info_size, (vm_address_t)&allImageInfo64, &readSize) ) {
-        if ( kr != NULL )
-            *kr = r;
-        return  NULL;
-    }
-    if ( allImageInfo64.infoArrayCount == 0 ) {
-        // could be task was launch suspended or still launching, wait a moment to see
-        usleep(1000 * 50); // 50ms
-        if ( kern_return_t r = mach_vm_read_overwrite(task, task_dyld_info.all_image_info_addr, task_dyld_info.all_image_info_size, (vm_address_t)&allImageInfo64, &readSize) ) {
-            if ( kr != NULL )
-                *kr = r;
-            return  NULL;
+        return nullptr;
+
+    // We use a true shared memory buffer here, that way by making sure that libdyld in both processes
+    // reads and writes the the timestamp atomically we can make sure we get a coherent view of the
+    // remote process.
+    // That also means that we *MUST* directly read the memory, which is why we template the make() call
+    withRemoteBuffer(task, task_dyld_info.all_image_info_addr, task_dyld_info.all_image_info_size, true, false, kr, ^(void *buffer, size_t size) {
+        dyld_process_info_ptr base;
+        if (task_dyld_info.all_image_info_format == TASK_DYLD_ALL_IMAGE_INFO_32 ) {
+            const dyld_all_image_infos_32* info = (const dyld_all_image_infos_32*)buffer;
+            base = dyld_process_info_base::make<dyld_all_image_infos_32, dyld_image_info_32>(task, *info, timestamp, kr);
+        } else {
+            const dyld_all_image_infos_64* info = (const dyld_all_image_infos_64*)buffer;
+            base = dyld_process_info_base::make<dyld_all_image_infos_64, dyld_image_info_64>(task, *info, timestamp, kr);
         }
-        // if infoArrayCount is still zero, then target was most likely launched suspended
-        if ( allImageInfo64.infoArrayCount == 0 )
-            return dyld_process_info_base::makeSuspended(task, kr);
-    }
-
-    // bail out of dyld is too old
-    if ( allImageInfo64.version < 15 ) {
-        if ( kr != NULL )
-            *kr = KERN_INVALID_HOST;
-        return NULL;
-    }
-
-    // normalize by expanding 32-bit all_image_infos into 64-bit one
-    uint32_t imageCount = allImageInfo64.infoArrayCount;
-    size_t imageArraySize = imageCount * sizeof(dyld_image_info_64);
-    if ( task_dyld_info.all_image_info_format == TASK_DYLD_ALL_IMAGE_INFO_32 ) {
-        const dyld_all_image_infos_32* allImageInfo32 = (dyld_all_image_infos_32*)&allImageInfo64;
-        dyld_all_image_infos_64 info64;
-        bzero(&info64, sizeof(info64));
-        info64.version                  = allImageInfo32->version;
-        info64.infoArrayCount           = allImageInfo32->infoArrayCount;
-        info64.infoArray                = allImageInfo32->infoArray;
-        info64.processDetachedFromSharedRegion = allImageInfo32->processDetachedFromSharedRegion;
-        info64.libSystemInitialized     = allImageInfo32->libSystemInitialized;
-        info64.dyldImageLoadAddress     = allImageInfo32->dyldImageLoadAddress;
-        info64.initialImageCount        = allImageInfo32->initialImageCount;
-        info64.uuidArrayCount           = allImageInfo32->uuidArrayCount;
-        info64.uuidArray                = allImageInfo32->uuidArray;
-        info64.dyldAllImageInfosAddress = allImageInfo32->dyldAllImageInfosAddress;
-        info64.sharedCacheSlide         = allImageInfo32->sharedCacheSlide;
-        info64.infoArrayChangeTimestamp = allImageInfo32->infoArrayChangeTimestamp;
-        info64.sharedCacheBaseAddress   = allImageInfo32->sharedCacheBaseAddress;
-        info64.dyldPath                 = allImageInfo32->dyldPath;
-        memcpy((void*)(info64.sharedCacheUUID), (void*)(allImageInfo32->sharedCacheUUID), 16);
-        allImageInfo64 = info64;
-        imageCount = allImageInfo64.infoArrayCount;
-        imageArraySize = imageCount * sizeof(dyld_image_info_32);
-    }
-
-    // don't do any (more) work if target process's dyld timestamp has not changed since previous query
-    if ( (timestamp != 0) && (timestamp == allImageInfo64.infoArrayChangeTimestamp) ) {
-        if ( kr != NULL )
-            *kr = KERN_SUCCESS;
-        return NULL;
-    }
-
-    // For the moment we are going to truncate any image list longer than 8192 because some programs do
-    // terrible things that corrupt their own image lists and we need to stop clients from crashing
-    // reading them. We can try to do something more advanced in the future. rdar://27446361
-    imageCount = MIN(imageCount, 8192);
-
-    // read image array
-    dyld_image_info_64 imageArray64[imageCount];
-    if ( kern_return_t r = mach_vm_read_overwrite(task, allImageInfo64.infoArray, imageArraySize, (vm_address_t)&imageArray64, &readSize) ) {
-         if ( kr != NULL )
-            *kr = r;
-        return  NULL;
-    }
-    // normalize by expanding 32-bit image_infos into 64-bit ones
-    if ( task_dyld_info.all_image_info_format == TASK_DYLD_ALL_IMAGE_INFO_32 ) {
-        const dyld_image_info_32* imageArray32 = (dyld_image_info_32*)&imageArray64;
-        dyld_image_info_64 tempArray[imageCount];
-        for (uint32_t i=0; i < imageCount; ++i) {
-            tempArray[i].imageLoadAddress = imageArray32[i].imageLoadAddress;
-            tempArray[i].imageFilePath    = imageArray32[i].imageFilePath;
-            tempArray[i].imageFileModDate = imageArray32[i].imageFileModDate;
+        if (base) {
+            result = base.release();
         }
-        memcpy(imageArray64, tempArray, sizeof(dyld_image_info_64)*imageCount);
-    }
-
-    // create object based on local copy of all image infos and image array
-    return dyld_process_info_base::make(task, allImageInfo64, imageArray64, kr);
+    });
+    return  result;
 }
 
-
 void _dyld_process_info_get_state(dyld_process_info info, dyld_process_state_info* stateInfo)
 {
     *stateInfo = *info->stateInfo();
@@ -621,16 +728,14 @@ void _dyld_process_info_get_cache(dyld_process_info info, dyld_process_cache_inf
     *cacheInfo = *info->cacheInfo();
 }
 
-void _dyld_process_info_retain(dyld_process_info info)
+void _dyld_process_info_retain(dyld_process_info object)
 {
-    info->retainCount() += 1;
+    const_cast<dyld_process_info_base*>(object)->retain();
 }
 
-void _dyld_process_info_release(dyld_process_info info)
+void _dyld_process_info_release(dyld_process_info object)
 {
-    info->retainCount() -= 1;
-    if ( info->retainCount() == 0 )
-        free((void*)info);
+    const_cast<dyld_process_info_base*>(object)->release();
 }
 
 void _dyld_process_info_for_each_image(dyld_process_info info, void (^callback)(uint64_t machHeaderAddress, const uuid_t uuid, const char* path))