File I/O

Deprecation of the technologies described here has been announced for platforms other than ChromeOS. Please visit our migration guide for details.


This section describes how to use the FileIO API to read and write files using a local secure data store.

You might use the File IO API with the URL Loading APIs to create an overall data download and caching solution for your NaCl applications. For example:

  1. Use the File IO APIs to check the local disk to see if a file exists that your program needs.
  2. If the file exists locally, load it into memory using the File IO API. If the file doesn’t exist locally, use the URL Loading API to retrieve the file from the server.
  3. Use the File IO API to write the file to disk.
  4. Load the file into memory using the File IO API when needed by your application.

The example discussed in this section is included in the SDK in the directory examples/api/file_io.

Reference information

For reference information related to FileIO, see the following documentation:

  • file_io.h - API to create a FileIO object
  • file_ref.h - API to create a file reference or “weak pointer” to a file in a file system
  • file_system.h - API to create a file system associated with a file

Local file I/O

Chrome provides an obfuscated, restricted area on disk to which a web app can safely read and write files. The Pepper FileIO, FileRef, and FileSystem APIs (collectively called the File IO APIs) allow you to access this sandboxed local disk so you can read and write files and manage caching yourself. The data is persistent between launches of Chrome, and is not removed unless your application deletes it or the user manually deletes it. There is no limit to the amount of local data you can use, other than the actual space available on the local drive.

Enabling local file I/O

The easiest way to enable the writing of persistent local data is to include the unlimitedStorage permission in your Chrome Web Store manifest file. With this permission you can use the Pepper FileIO API without the need to request disk space at run time. When the user installs the app Chrome displays a message announcing that the app writes to the local disk.

If you do not use the unlimitedStorage permission you must include JavaScript code that calls the HTML5 Quota Management API to explicitly request local disk space before using the FileIO API. In this case Chrome will prompt the user to accept a requestQuota call every time one is made.

Testing local file I/O

You should be aware that using the unlimitedStorage manifest permission constrains the way you can test your app. Three of the four techniques described in Running Native Client Applications read the Chrome Web Store manifest file and enable the unlimitedStorage permission when it appears, but the first technique (local server) does not. If you want to test the file IO portion of your app with a simple local server, you need to include JavaScript code that calls the HTML5 Quota Management API. When you deliver your application you can replace this code with the unlimitedStorage manifest permission.

The file_io example

The Native Client SDK includes an example, file_io, that demonstrates how to read and write a local disk file. Since you will probably run the example from a local server without a Chrome Web Store manifest file, the example’s index file uses JavaScript to perform the Quota Management setup as described above. The example has these primary files:

  • index.html - The HTML code that launches the Native Client module and displays the user interface.
  • example.js - JavaScript code that requests quota (as described above). It also listens for user interaction with the user interface, and forwards the requests to the Native Client module.
  • - The code that sets up and provides an entry point to the Native Client module.

The remainder of this section covers the code in the file for reading and writing files.

File I/O overview

Like many Pepper APIs, the File IO API includes a set of methods that execute asynchronously and that invoke callback functions in your Native Client module. Unlike most other examples, the file_io example also demonstrates how to make Pepper calls synchronously on a worker thread.

It is illegal to make blocking calls to Pepper on the module’s main thread. This restriction is lifted when running on a worker thread—this is called “calling Pepper off the main thread”. This often simplifies the logic of your code; multiple asynchronous Pepper functions can be called from one function on your worker thread, so you can use the stack and standard control flow structures normally.

The high-level flow for the file_io example is described below. Note that methods in the namespace pp are part of the Pepper C++ API.

Creating and writing a file

Following are the high-level steps involved in creating and writing to a file:

  1. pp::FileIO::Open is called with the PP_FILEOPEN_FLAG_CREATE flag to create a file. Because the callback function is pp::BlockUntilComplete, this thread is blocked until Open succeeds or fails.
  2. pp::FileIO::Write is called to write the contents. Again, the thread is blocked until the call to Write completes. If there is more data to write, Write is called again.
  3. When there is no more data to write, call pp::FileIO::Flush.

Opening and reading a file

Following are the high-level steps involved in opening and reading a file:

  1. pp::FileIO::Open is called to open the file. Because the callback function is pp::BlockUntilComplete, this thread is blocked until Open succeeds or fails.
  2. pp::FileIO::Query is called to query information about the file, such as its file size. The thread is blocked until Query completes.
  3. pp::FileIO::Read is called to read the contents. The thread is blocked until Read completes. If there is more data to read, Read is called again.

Deleting a file

Deleting a file is straightforward: call pp::FileRef::Delete. The thread is blocked until Delete completes.

Making a directory

Making a directory is also straightforward: call pp::File::MakeDirectory. The thread is blocked until MakeDirectory completes.

Listing the contents of a directory

Following are the high-level steps involved in listing a directory:

  1. pp::FileRef::ReadDirectoryEntries is called, and given a directory entry to list. A callback is given as well; many of the other functions use pp::BlockUntilComplete, but ReadDirectoryEntries returns results in its callback, so it must be specified.
  2. When the call to ReadDirectoryEntries completes, it calls ListCallback which packages up the results into a string message, and sends it to JavaScript.

file_io deep dive

The file_io example displays a user interface with a couple of fields and several buttons. Following is a screenshot of the file_io example:


Each radio button is a file operation you can perform, with some reasonable default values for filenames. Try typing a message in the large input box and clicking Save, then switching to the Load File operation, and clicking Load.

Let’s take a look at what is going on under the hood.

Opening a file system and preparing for file I/O

pp::Instance::Init is called when an instance of a module is created. In this example, Init starts a new thread (via the pp::SimpleThread class), and tells it to open the filesystem:

virtual bool Init(uint32_t /*argc*/,
                  const char * /*argn*/ [],
                  const char * /*argv*/ []) {
  // Open the file system on the file_thread_. Since this is the first
  // operation we perform there, and because we do everything on the
  // file_thread_ synchronously, this ensures that the FileSystem is open
  // before any FileIO operations execute.
  return true;

When the file thread starts running, it will call OpenFileSystem. This calls pp::FileSystem::Open and blocks the file thread until the function returns.

void OpenFileSystem(int32_t /*result*/) {
  int32_t rv = file_system_.Open(1024 * 1024, pp::BlockUntilComplete());
  if (rv == PP_OK) {
    file_system_ready_ = true;
    // Notify the user interface that we're ready
  } else {
    ShowErrorMessage("Failed to open file system", rv);

Handling messages from JavaScript

When you click the Save button, JavaScript posts a message to the NaCl module with the file operation to perform sent as a string (See Messaging System for more details on message passing). The string is parsed by HandleMessage, and new work is added to the file thread:

virtual void HandleMessage(const pp::Var& var_message) {
  if (!var_message.is_string())

  // Parse message into: instruction file_name_length file_name [file_text]
  std::string message = var_message.AsString();
  std::string instruction;
  std::string file_name;
  std::stringstream reader(message);
  int file_name_length;

  reader >> instruction >> file_name_length;
  reader.ignore(1);  // Eat the delimiter[0], file_name_length);


  // Dispatch the instruction
  if (instruction == kLoadPrefix) {
        callback_factory_.NewCallback(&FileIoInstance::Load, file_name));
  } else if (instruction == kSavePrefix) {

Saving a file

FileIoInstance::Save is called when the Save button is pressed. First, it checks to see that the FileSystem has been successfully opened:

if (!file_system_ready_) {
  ShowErrorMessage("File system is not open", PP_ERROR_FAILED);

It then creates a pp::FileRef resource with the name of the file. A FileRef resource is a weak reference to a file in the FileSystem; that is, a file can still be deleted even if there are outstanding FileRef resources.

pp::FileRef ref(file_system_, file_name.c_str());

Next, a pp::FileIO resource is created and opened. The call to pp::FileIO::Open passes PP_FILEOPEFLAG_WRITE to open the file for writing, PP_FILEOPENFLAG_CREATE to create a new file if it doesn’t already exist and PP_FILEOPENFLAG_TRUNCATE to clear the file of any previous content:

pp::FileIO file(this);

int32_t open_result =
if (open_result != PP_OK) {
  ShowErrorMessage("File open for write failed", open_result);

Now that the file is opened, it is written to in chunks. In an asynchronous model, this would require writing a separate function, storing the current state on the free store and a chain of callbacks. Because this function is called off the main thread, pp::FileIO::Write can be called synchronously and a conventional do/while loop can be used:

int64_t offset = 0;
int32_t bytes_written = 0;
do {
  bytes_written = file.Write(offset,
                    + offset,
  if (bytes_written > 0) {
    offset += bytes_written;
  } else {
    ShowErrorMessage("File write failed", bytes_written);
} while (bytes_written < static_cast<int64_t>(file_contents.length()));

Finally, the file is flushed to push all changes to disk:

int32_t flush_result = file.Flush(pp::BlockUntilComplete());
if (flush_result != PP_OK) {
  ShowErrorMessage("File fail to flush", flush_result);

Loading a file

FileIoInstance::Load is called when the Load button is pressed. Like the Save function, Load first checks to see if the FileSystem has been successfully opened, and creates a new FileRef:

if (!file_system_ready_) {
  ShowErrorMessage("File system is not open", PP_ERROR_FAILED);
pp::FileRef ref(file_system_, file_name.c_str());

Next, Load creates and opens a new FileIO resource, passing PP_FILEOPENFLAG_READ to open the file for reading. The result is compared to PP_ERROR_FILENOTFOUND to give a better error message when the file doesn’t exist:

int32_t open_result =
    file.Open(ref, PP_FILEOPENFLAG_READ, pp::BlockUntilComplete());
if (open_result == PP_ERROR_FILENOTFOUND) {
  ShowErrorMessage("File not found", open_result);
} else if (open_result != PP_OK) {
  ShowErrorMessage("File open for read failed", open_result);

Then Load calls pp::FileIO::Query to get metadata about the file, such as its size. This is used to allocate a std::vector buffer that holds the data from the file in memory:

int32_t query_result = file.Query(&info, pp::BlockUntilComplete());
if (query_result != PP_OK) {
  ShowErrorMessage("File query failed", query_result);


std::vector<char> data(info.size);

Similar to Save, a conventional while loop is used to read the file into the newly allocated buffer:

int64_t offset = 0;
int32_t bytes_read = 0;
int32_t bytes_to_read = info.size;
while (bytes_to_read > 0) {
  bytes_read = file.Read(offset,
                         data.size() - offset,
  if (bytes_read > 0) {
    offset += bytes_read;
    bytes_to_read -= bytes_read;
  } else if (bytes_read < 0) {
    // If bytes_read < PP_OK then it indicates the error code.
    ShowErrorMessage("File read failed", bytes_read);

Finally, the contents of the file are sent back to JavaScript, to be displayed on the page. This example uses “DISP|” as a prefix command for display information:

std::string string_data(data.begin(), data.end());
PostMessage("DISP|" + string_data);
ShowStatusMessage("Load success");

Deleting a file

FileIoInstance::Delete is called when the Delete button is pressed. First, it checks whether the FileSystem has been opened, and creates a new FileRef:

if (!file_system_ready_) {
  ShowErrorMessage("File system is not open", PP_ERROR_FAILED);
pp::FileRef ref(file_system_, file_name.c_str());

Unlike Save and Load, Delete is called on the FileRef resource, not a FileIO resource. Note that the result is checked for PP_ERROR_FILENOTFOUND to give a better error message when trying to delete a non-existent file:

int32_t result = ref.Delete(pp::BlockUntilComplete());
if (result == PP_ERROR_FILENOTFOUND) {
  ShowStatusMessage("File/Directory not found");
} else if (result != PP_OK) {
  ShowErrorMessage("Deletion failed", result);

Listing files in a directory

FileIoInstance::List is called when the List Directory button is pressed. Like all other operations, it checks whether the FileSystem has been opened and creates a new FileRef:

if (!file_system_ready_) {
  ShowErrorMessage("File system is not open", PP_ERROR_FAILED);

pp::FileRef ref(file_system_, dir_name.c_str());

Unlike the other operations, it does not make a blocking call to pp::FileRef::ReadDirectoryEntries. Since ReadDirectoryEntries returns the resulting directory entries in its callback, a new callback object is created pointing to FileIoInstance::ListCallback.

The pp::CompletionCallbackFactory template class is used to instantiate a new callback. Notice that the FileRef resource is passed as a parameter; this will add a reference count to the callback object, to keep the FileRef resource from being destroyed when the function finishes.

// Pass ref along to keep it alive.
    &FileIoInstance::ListCallback, ref));

FileIoInstance::ListCallback then gets the results passed as a std::vector of pp::DirectoryEntry objects, and sends them to JavaScript:

void ListCallback(int32_t result,
                  const std::vector<pp::DirectoryEntry>& entries,
                  pp::FileRef /*unused_ref*/) {
  if (result != PP_OK) {
    ShowErrorMessage("List failed", result);

  std::stringstream ss;
  ss << "LIST";
  for (size_t i = 0; i < entries.size(); ++i) {
    pp::Var name = entries[i].file_ref().GetName();
    if (name.is_string()) {
      ss << "|" << name.AsString();
  ShowStatusMessage("List success");

Making a new directory

FileIoInstance::MakeDir is called when the Make Directory button is pressed. Like all other operations, it checks whether the FileSystem has been opened and creates a new FileRef:

if (!file_system_ready_) {
  ShowErrorMessage("File system is not open", PP_ERROR_FAILED);
pp::FileRef ref(file_system_, dir_name.c_str());

Then the pp::FileRef::MakeDirectory function is called.

int32_t result = ref.MakeDirectory(
    PP_MAKEDIRECTORYFLAG_NONE, pp::BlockUntilComplete());
if (result != PP_OK) {
  ShowErrorMessage("Make directory failed", result);
ShowStatusMessage("Make directory success");
We serve cookies on this site to analyze traffic, remember your preferences, and optimize your experience.