Native File System

Draft Community Group Report,

This version:
https://wicg.github.io/native-file-system/
Issue Tracking:
GitHub
Inline In Spec
Editor:
(Google)

Abstract

This document defines a web platform API that lets websites gain write access to the native file system. It builds on File API, but adds lots of new functionality on top.

Status of this document

This specification was published by the Web Platform Incubator Community Group. It is not a W3C Standard nor is it on the W3C Standards Track. Please note that under the W3C Community Contributor License Agreement (CLA) there is a limited opt-out and other conditions apply. Learn more about W3C Community and Business Groups.

1. Introduction

This section is non-normative.

TODO

This provides similar functionality as earlier drafts of the File API: Directories and System as well as the File and Directory Entries API, but with a more modern API.

2. Files and Directories

2.1. Concepts

An entry is either a file entry or a directory entry.

Each entry has an associated name.

A file entry additionally consists of binary data and a modification timestamp.

A directory entry additionally consists of a set of children, which are themselves entries. Each member is either a file or a directory.

An entry entry should be contained in the children of at most one directory entry, and that directory entry is also known as entry’s parent. An entry's parent is null if no such directory entry exists.

Note: Two different entries can represent the same file or directory on disk, in which case it is possible for both entries to have a different parent, or for one entry to have a parent while the other entry does not have a parent. Typically an entry does not have a parent if it was returned by chooseFileSystemEntries() or getSystemDirectory(), and an entry will have a parent in all other cases.

Entries can (but don’t have to) be backed by files on the systems native file system, so it is possible for the binary data, modification timestamp, and children of entries to be modified by applications outside of this specification. Exactly how external changes are reflected in the data structures defined by this specification, as well as how changes made to the data structures defined here are reflected externally is left up to individual user-agent implementations.

An entry a is the same as an entry b if a is equal to b, or if a and b are backed by the same file or directory on the native file system.

TODO: Explain better how entries map to files on disk (multiple entries can map to the same file or directory on disk but an entry doesn’t have to map to any file on disk).

To resolve an entry child relative to a directory entry root, run the following steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. If child is the same as root, resolve result with an empty list, and abort.

    2. Let childPromises be « ».

    3. For each entry of root’s entry's children:

      1. Let p be the result of resolving child relative to entry.

      2. Append p to childPromises.

      3. Upon fulfillment of p with value path:

        1. If path is not null:

          1. Prepend entry’s name to path.

          2. Resolve result with path.

    4. Wait for all childPromises, with the following success steps:

      1. If result hasn’t been resolved yet, resolve result with null.

  3. Return result.

2.2. Permissions

Figure out some way to integrate this better with the permissions API [permissions], for example by making the entry be a part of the PermissionDescriptor.

An entry entry has associated query permission steps, which take a FileSystemHandlePermissionDescriptor and return a PermissionState. Unless specified otherwise, these steps are:
  1. Let parent be entry’s parent.

  2. Assert: parent is not null.

    Note: § 3.1 Native File System Permissions overrides these steps for entries returned by chooseFileSystemEntries(). Additionally § 4.1 Special File System Concepts overrides these steps for entries returned by getSystemDirectory(). All other entries always have a parent.

  3. Return the result of running parent’s query permission steps.

Note: These steps return the current state synchronously, however these steps are only invoked from "in parallel" sections of algorithms, so this doesn’t have to be implemented in a synchronous manner.

An entry entry also has associated request permission steps, which take a FileSystemHandlePermissionDescriptor. Unless specified otherwise, these steps are:
  1. Let parent be entry’s parent.

  2. If parent is not null, run parent’s request permission steps.

Note: These steps do not return anything. However as a result of executing the request permission steps for an entry the permission state for that entry (and other entries) can be updated, and the new state can be queried by executing the query permission steps.

2.3. The FileSystemHandle interface

dictionary FileSystemHandlePermissionDescriptor {
  boolean writable = false;
};

[Exposed=(Window,Worker), SecureContext, Serializable]
interface FileSystemHandle {
  readonly attribute boolean isFile;
  readonly attribute boolean isDirectory;
  readonly attribute USVString name;

  Promise<boolean> isSameEntry(FileSystemHandle other);

  Promise<PermissionState> queryPermission(optional FileSystemHandlePermissionDescriptor descriptor = {});
  Promise<PermissionState> requestPermission(optional FileSystemHandlePermissionDescriptor descriptor = {});
};

A FileSystemHandle object represents an entry. Each FileSystemHandle object is associated with an entry (an entry). Multiple separate objects implementing the FileSystemHandle interface can all be associated with the same entry simultaneously.

FileSystemHandle objects are serializable objects.

In the Origin Trial as available in Chrome 78, these objects are not yet serializable. In Chrome 82 they are.

Their serialization steps, given value, serialized and forStorage are:

  1. Set serialized.[[Origin]] to value’s relevant settings object's origin.

  2. Set serialized.[[Entry]] to value’s entry.

Their deserialization steps, given serialized and value are:
  1. If serialized.[[Origin]] is not same origin with value’s relevant settings object's origin, then throw a DataCloneError.

  2. Set value’s entry to serialized.[[Entry]]

handle . isFile

Returns true if handle is a FileSystemFileHandle.

handle . isDirectory

Returns true if handle is a FileSystemDirectoryHandle.

handle . name

Returns the name of the entry represented by handle.

The isFile attribute must return true if the associated entry is a file entry, and false otherwise.

The isDirectory attribute must return true if the associated entry is a directory entry, and false otherwise.

The name attribute must return the name of the associated entry.

2.3.1. The isSameEntry() method

same = await handle1 . isSameEntry( handle2 )

Returns true if handle1 and handle2 represent the same file or directory.

This method is first available in Chrome 82.

The isSameEntry(other) method, when invoked, must run these steps:
  1. Let realm be this's relevant Realm.

  2. Let p be a new promise in realm.

  3. Run the following steps in parallel:

    1. If this's entry is the same as other’s entry, resolve p with true.

    2. Else resolve p with false.

  4. Return p.

2.3.2. The queryPermission() method

the currently described API here assumes a model where it is not possible to have a write-only handle. I.e. it is not possible to have or request write access without also having read access. There definitely are use cases for write-only handles (e.g. directory downloads), so we might have to reconsider this. <https://github.com/wicg/native-file-system/issues/119>

status = await handle . queryPermission({ writable = false })
status = await handle . queryPermission()

Queries the current state of the read permission of this handle. If this returns "prompt" the website will have to call requestPermission() before any operations on the handle can be done. If this returns "denied" any operations will reject.

Usually handles returned by chooseFileSystemEntries will initially return "granted" for their read permission state, however other than through the user revoking permission, a handle retrieved from IndexedDB is also likely to return "prompt".

status = await handle . queryPermission({ writable = true })

Queries the current state of the write permission of this handle. If this returns "prompt", attempting to modify the file or directory this handle represents will require user activation and will result in a confirmation prompt being shown to the user. However if the state of the read permission of this handle is also "prompt" the website will need to call requestPermission(). There is no automatic prompting for read access when attempting to read from a file or directory.

The queryPermission(descriptor) method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Resolve result with the result of running this’s entry's query permission steps given descriptor.

  3. Return result.

2.3.3. The requestPermission() method

status = await handle . requestPermission({ writable = false })
status = await handle . requestPermission()

If the state of the read permission of this handle is anything other than "prompt", this will return that state directly. If it is "prompt" however, user activation is needed and this will show a confirmation prompt to the user. The new read permission state is then returned, depending on the user’s response to the prompt.

status = await handle . requestPermission({ writable = true })

If the state of the write permission of this handle is anything other than "prompt", this will return that state directly. If the status of the read permission of this handle is "denied" this will return that.

Otherwise the state of the write permission is "prompt" and this will show a confirmation prompt to the user. The new write permission state is then returned, depending on what the user selected.

The requestPermission(descriptor) method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Run this’s entry's request permission steps given descriptor. If that throws an exception, reject result with that exception and abort.

    2. Resolve result with the result of running this’s entry's query permission steps given descriptor.

  3. Return result.

2.4. The FileSystemFileHandle interface

dictionary FileSystemCreateWritableOptions {
  boolean keepExistingData = false;
};

[Exposed=(Window,Worker), SecureContext, Serializable]
interface FileSystemFileHandle : FileSystemHandle {
  Promise<File> getFile();
  Promise<FileSystemWritableFileStream> createWritable(optional FileSystemCreateWritableOptions options = {});
};

A FileSystemFileHandle's associated entry must be a file entry.

FileSystemFileHandle objects are serializable objects. Their serialization steps and deserialization steps are the same as those for FileSystemHandle.

In the Origin Trial as available in Chrome 78, these objects are not yet serializable. In Chrome 82 they are.

2.4.1. The getFile() method

file = await fileHandle . getFile()

Returns a File representing the state on disk of the entry represented by handle. If the file on disk changes or is removed after this method is called, the returned File object will likely be no longer readable.

The getFile() method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Let permissionStatus be the result of running this’s entry's query permission steps given «[ "writable" → false ]».

    2. If permissionStatus is not "granted", reject result with a NotAllowedError and abort.

    3. TODO

  3. Return result.

2.4.2. The createWritable() method

In the Origin Trial as available in Chrome 82, createWritable replaces the createWriter method.

stream = await fileHandle . createWritable()
stream = await fileHandle . createWritable({ keepExistingData: true/false })

Returns a FileSystemWritableFileStream that can be used to write to the file. Any changes made through stream won’t be reflected in the file represented by fileHandle until the stream has been closed. User agents try to ensure that no partial writes happen, i.e. the file represented by fileHandle will either contains its old contents or it will contain whatever data was written through stream up until the stream has been closed.

This is typically implemented by writing data to a temporary file, and only replacing the file represented by fileHandle with the temporary file when the writable filestream is closed.

If keepExistingData is false or not specified, the temporary file starts out empty, otherwise the existing file is first copied to this temporary file.

There has been some discussion around and desire for a "inPlace" mode for createWritable (where changes will be written to the actual underlying file as they are written to the writer, for example to support in-place modification of large files or things like databases). This is not currently implemented in Chrome. Implementing this is currently blocked on figuring out how to combine the desire to run malware checks with the desire to let websites make fast in-place modifications to existing large files. <https://github.com/wicg/native-file-system/issues/67>

The createWritable(options) method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Let entry be this’s entry.

    2. Run entry’s request permission steps given «[ "writable" → true ]». If that throws an exception, reject result with that exception and abort.

    3. Let permissionStatus be the result of running entry’s query permission steps given «[ "writable" → true ]».

    4. If permissionStatus is not "granted", reject result with a NotAllowedError and abort.

    5. Let stream be the result of creating a new FileSystemWritableFileStream for entry.

    6. If options.keepExistingData is true:

      1. Set stream.[[buffer]] to a copy of entry’s binary data.

    7. Resolve result with stream.

  3. Return result.

2.5. The FileSystemDirectoryHandle interface

dictionary FileSystemGetFileOptions {
  boolean create = false;
};

dictionary FileSystemGetDirectoryOptions {
  boolean create = false;
};

dictionary FileSystemRemoveOptions {
  boolean recursive = false;
};

[Exposed=(Window,Worker), SecureContext, Serializable]
interface FileSystemDirectoryHandle : FileSystemHandle {
  Promise<FileSystemFileHandle> getFile(USVString name, optional FileSystemGetFileOptions options = {});
  Promise<FileSystemDirectoryHandle> getDirectory(USVString name, optional FileSystemGetDirectoryOptions options = {});

  // This really returns an async iterable, but that is not yet expressable in WebIDL.
  object getEntries();

  Promise<void> removeEntry(USVString name, optional FileSystemRemoveOptions options = {});

  Promise<sequence<USVString>?> resolve(FileSystemHandle possibleDescendant);
};

A FileSystemDirectoryHandle's associated entry must be a directory entry.

FileSystemDirectoryHandle objects are serializable objects. Their serialization steps and deserialization steps are the same as those for FileSystemHandle.

In the Origin Trial as available in Chrome 78, these objects are not yet serializable. In Chrome 82 they are.

Should we have separate getFile and getDirectory methods, or just a single getChild/getEntry method?

Having getFile methods in both FileSystemDirectoryHandle and FileSystemFileHandle, but with very different behavior might be confusing? Perhaps rename at least one of them (but see also previous issue). <https://github.com/wicg/native-file-system/issues/98>

Should getEntries be its own method, or should FileSystemDirectoryHandle just be an async iterable itself? <https://github.com/wicg/native-file-system/issues/47>

2.5.1. The getFile() method

fileHandle = await directoryHandle . getFile(name)
fileHandle = await directoryHandle . getFile(name, { create: false })

Returns a handle for a file named name in the directory represented by directoryHandle. If no such file exists, this rejects.

fileHandle = await directoryHandle . getFile(name, { create: true })

Returns a handle for a file named name in the directory represented by directoryHandle. If no such file exists, this creates a new file. If no file with named name can be created this rejects. Creation can fail because there already is a directory with the same name, because the name uses characters that aren’t supported in file names on the underlying file system, or because the user agent for security reasons decided not to allow creation of the file.

This operation requires write permission, even if the file being returned already exists. If this handle doesn’t already have write permission, this could result in a prompt being shown to the user. To get an existing file without needing write permission, call this method with { create: false }.

The getFile(name, options) method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Let entry be this’s entry.

    2. If options.create is true:

      1. Run entry’s request permission steps given «[ "writable" → true ]». If that throws an exception, reject result with that exception and abort.

      2. Let permissionStatus be the result of running entry’s query permission steps given «[ "writable" → true ]».

    3. Otherwise:

      1. Let permissionStatus be the result of running entry’s query permission steps given «[ "writable" → false ]».

    4. If permissionStatus is not "granted", reject result with a NotAllowedError and abort.

    5. TODO

  3. Return result.

2.5.2. The getDirectory() method

subdirHandle = await directoryHandle . getDirectory(name)
subdirHandle = await directoryHandle . getDirectory(name, { create: false })

Returns a handle for a directory named name in the directory represented by directoryHandle. If no such directory exists, this rejects.

subdirHandle = await directoryHandle . getDirectory(name, { create: true })

Returns a handle for a directory named name in the directory represented by directoryHandle. If no such directory exists, this creates a new directory. If creating the directory failed, this rejects. Creation can fail because there already is a file with the same name, or because the name uses characters that aren’t supported in file names on the underlying file system.

This operation requires write permission, even if the directory being returned already exists. If this handle doesn’t already have write permission, this could result in a prompt being shown to the user. To get an existing directory without needing write permission, call this method with { create: false }.

The getDirectory(name, options) method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Let entry be this’s entry.

    2. If options.create is true:

      1. Run entry’s request permission steps given «[ "writable" → true ]». If that throws an exception, reject result with that exception and abort.

      2. Let permissionStatus be the result of running entry’s query permission steps given «[ "writable" → true ]».

    3. Else:

      1. Let permissionStatus be the result of running entry’s query permission steps given «[ "writable" → false ]».

    4. If permissionStatus is not "granted", reject result with a NotAllowedError and abort.

    5. TODO

  3. Return result.

2.5.3. The getEntries() method

for await (const handle of directoryHandle . getEntries()) {}

Iterates over all entries whose parent is the entry represented by directoryHandle.

The getEntries() method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Let permissionStatus be the result of running this’s entry's query permission steps given «[ "writable" → false ]».

    2. If permissionStatus is not "granted", reject result with a NotAllowedError and abort.

    3. TODO

  3. Return result.

2.5.4. The removeEntry() method

await directoryHandle . removeEntry(name)
await directoryHandle . removeEntry(name, { recursive: false })

If the directory represented by directoryHandle contains a file named name, or an empty directory named name, this will attempt to delete that file or directory.

Attempting to delete a file or directory that does not exist is considered success, while attempting to delete a non-empty directory will result in a promise rejection.

await directoryHandle . removeEntry(name, { recursive: true })

Removes the entry named name in the directory represented by directoryHandle. If that entry is a directory, its contents will also be deleted recursively. recursively.

Attempting to delete a file or directory that does not exist is considered success.

The removeEntry(name, options) method, when invoked, must run these steps:
  1. Let result be a new promise.

  2. Run the following steps in parallel:

    1. Let entry be this’s entry.

    2. Run entry’s request permission steps given «[ "writable" → true ]». If that throws an exception, reject result with that exception.

    3. Let permissionStatus be the result of running entry’s query permission steps given «[ "writable" → true ]».

    4. If permissionStatus is not "granted", reject result with a NotAllowedError and abort.

    5. TODO

  3. Return result.

2.5.5. The resolve() method

path = await directory . resolve( child )

If child is equal to directory, path will be an empty array.

If child is a direct child of directory, path will be an array containing child’s name.

If child is a descendant of directory, path will be an array containing the names of all the intermediate directories and child’s name as last element.

Otherwise (directory and child are not related), path will be null.

This method is first available in Chrome 82.

// Assume we at some point got a valid directory handle.
const dir_ref = current_project_dir;
if (!dir_ref) return;

// Now get a file reference by showing a file picker:
const file_ref = await self.chooseFileSystemEntries({type: 'openFile'});
if (!file_ref) {
    // User cancelled, or otherwise failed to open a file.
    return;
}

// Check if file_ref exists inside dir_ref:
const relative_path = await dir_ref.resolve(file_ref);
if (relative_path === null) {
    // Not inside dir_ref
} else {
    // relative_path is an array of names, giving the relative path
    // from dir_ref to the file that is represented by file_ref:
    assert relative_path.pop() == file_ref.name;

    let entry = dir_ref;
    for (const name of relative_path) {
        entry = await entry.getDirectory(name);
    }
    entry = await entry.getFile(file_ref.name);

    // Now |entry| will represent the same file on disk as |file_ref|.
    assert await entry.isSameEntry(file_ref) == true;
}
The resolve(possibleDescendant) method, when invoked, must return the result of resolving possibleDescendant’s entry relative to this's entry.

2.6. The FileSystemWritableFileStream interface

enum WriteCommandType {
  "write",
  "seek",
  "truncate",
};

dictionary WriteParams {
  required WriteCommandType type;
  unsigned long long? size;
  unsigned long long? position;
  (BufferSource or Blob or USVString)? data;
};

typedef (BufferSource or Blob or USVString or WriteParams) FileSystemWriteChunkType;

[Exposed=(Window,Worker), SecureContext]
interface FileSystemWritableFileStream : WritableStream {
  Promise<void> write(FileSystemWriteChunkType data);
  Promise<void> seek(unsigned long long position);
  Promise<void> truncate(unsigned long long size);
};

A FileSystemWritableFileStream has an associated [[file]] (a file entry).

A FileSystemWritableFileStream has an associated [[buffer]] (a byte sequence). It is initially empty.

Note: This buffer can get arbitrarily large, so it is expected that implementations will not keep this in memory, but instead use a temporary file for this. All access to [[buffer]] is done in promise returning methods and algorithms, so even though operations on it seem sync, implementations can implement them async.

A FileSystemWritableFileStream has an associated [[seekOffset]] (a number). It is initially 0.

A FileSystemWritableFileStream object is a WritableStream object with additional convenience methods, which operates on a single file on disk.

Upon creation, an underlying sink will have been created and the stream will be usable. All operations executed on the stream are queuable and producers will be able to respond to backpressure.

The underlying sink’s write method, and therefore WritableStreamDefaultWriter’s write() method, will accept byte-like data or WriteParams as input.

The FileSystemWritableFileStream has a file position cursor initialized at byte offset 0 from the top of the file. When using write() or by using WritableStream capabilities through the WritableStreamDefaultWriter’s write() method, this position will be advanced based on the number of bytes written through the stream object.

Similarly, when piping a ReadableStream into a FileSystemWritableFileStream object, this position is updated with the number of bytes that passed through the stream.

getWriter() returns an instance of WritableStreamDefaultWriter.

To create a new FileSystemWritableFileStream given a file entry file, perform the following steps:
  1. Let stream be a new FileSystemWritableFileStream.

  2. Perform InitializeWritableStream(stream)

  3. Set stream.[[file]] to file.

  4. Let controller be a new WritableStreamDefaultController.

  5. Let startAlgorithm be an algorithm that returns undefined.

  6. Let writeAlgorithm be an algorithm which takes a chunk argument and returns the result of running the write a chunk algorithm with stream and chunk.

  7. Let closeAlgorithm be the following steps:

    1. Let closeResult be a new promise.

    2. Run the following steps in parallel:

      1. Let permissionStatus be the result of running stream.[[file]]'s query permission steps given «[ "writable" → true ]».

      2. If permissionStatus is not "granted", reject closeResult with a NotAllowedError and abort.

      3. TODO: optional UA defined security checks

      4. Set stream.[[file]]'s binary data to stream.[[buffer]]. If that throws an exception, reject closeResult with that exception and abort.

        Note: It is expected that this atomically updates the contents of the file on disk being written to.

      5. Resolve closeResult with undefined.

    3. Return closeResult.

  8. Let abortAlgorithm be an algorithm that returns a promise resolved with undefined.

  9. Let highWaterMark be 1.

  10. Let sizeAlgorithm be an algorithm that returns 1.

  11. Perform SetUpWritableStreamDefaultController(stream, controller, startAlgorithm, writeAlgorithm, closeAlgorithm, abortAlgorithm, highWaterMark, sizeAlgorithm).

  12. Return stream.

The write a chunk algorithm, given a FileSystemWritableFileStream stream and chunk, runs these steps:
  1. Let input be the result of converting chunk to a FileSystemWriteChunkType. If this throws an exception, then return a promise rejected with that exception.

  2. Let p be a new promise.

  3. Run the following steps in parallel:

    1. Let permissionStatus be the result of running stream.[[file]]'s query permission steps given «[ "writable" → true ]».

    2. If permissionStatus is not "granted", reject p with a NotAllowedError and abort.

    3. Let command be input.type if input is a WriteParams, and "write" otherwise.

    4. If command is "write":

      1. Let data be input.data if input is a WriteParams, and input otherwise.

      2. If data is undefined, reject p with a TypeError and abort.

      3. Let writePosition be stream.[[seekOffset]].

      4. If input is a WriteParams and input.position is not undefined, set writePosition to input.position.

      5. Let oldSize be stream.[[buffer]]'s length.

      6. If writePosition is larger than oldSize, reject p with a InvalidStateError and abort.

        Not clear if this should reject, and if it does, is this really the right error type? Chrome’s implementation is actually inconsistent about this rejecting or not.

      7. If data is a BufferSource, let dataBytes be a copy of data.

      8. Else if data is a Blob:

        1. Let dataBytes be the result of performing the read operation on data. If this throws an exception, reject p with that exception and abort.

      9. Else:

        1. Assert: data is a USVString.

        2. Let dataBytes be the result of UTF-8 encoding data.

      10. Let head be a byte sequence containing the first writePosition bytes of stream.[[buffer]].

      11. Let tail be an empty byte sequence.

      12. If writePosition + data.length is smaller than oldSize:

        1. Let tail be a byte sequence containing the last oldSize - (writePosition + data.length) bytes of stream.[[buffer]].

      13. Set stream.[[buffer]] to the concatenation of head, data and tail.

      14. Set stream.[[seekOffset]] to writePosition + data.length.

      15. Resolve p.

    5. Else if command is "seek":

      1. If chunk.position is undefined, reject p with a TypeError and abort.

      2. Set stream.[[seekOffset]] to chunk.position.

      3. Resolve p.

    6. Else if command is "truncate":

      1. If chunk.size is undefined, reject p with a TypeError and abort.

      2. Let newSize be chunk.size.

      3. Let oldSize be stream.[[buffer]]'s length.

      4. If newSize is larger than oldSize:

        1. Set stream.[[buffer]] to a byte sequence formed by concating stream.[[buffer]] with a byte sequence containing newSize-oldSize 0x00 bytes.

      5. Else if newSize is smaller than oldSize:

        1. Set stream.[[buffer]] to a byte sequence containing the first newSize bytes in stream.[[buffer]].

      6. If stream.[[seekOffset]] is bigger than newSize, set stream.[[seekOffset]] to newSize.

      7. Resolve p.

  4. Return p.

2.6.1. The write() method

await stream . write(data)
await stream . write({ type: "write", data: data })

Writes the content of data into the file associated with stream at the current file cursor offset.

No changes are written to the actual file on disk until the stream has been closed. Changes are typically written to a temporary file instead.

await stream . write({ type: "write", position: position, data: data })

Writes the content of data into the file associated with stream at position bytes from the top of the file. Also updates the current file cursor offset to the end of the written data.

No changes are written to the actual file on disk until the stream has been closed. Changes are typically written to a temporary file instead.

await stream . write({ type: "seek", position: position })

Updates the current file cursor offset the position bytes from the top of the file.

await stream . write({ type: "truncate", size: size })

Resizes the file associated with stream to be size bytes long. If size is larger than the current file size this pads the file with null bytes, otherwise it truncates the file.

The file cursor is updated when truncate is called. If the offset is smaller than offset, it remains unchanged. If the offset is larger than size, the offset is set to size to ensure that subsequent writes do not error.

No changes are written to the actual file until on disk until the stream has been closed. Changes are typically written to a temporary file instead.

The write(data) method, when invoked, must run these steps:
  1. Let writer be the result of calling AcquireWritableStreamDefaultWriter(this).

  2. Let result be WritableStreamDefaultWriterWrite(writer, data).

  3. Perform WritableStreamDefaultWriterRelease(writer).

  4. Return result.

2.6.2. The seek() method

await stream . seek(position)

Updates the current file cursor offset the position bytes from the top of the file.

The seek(position) method, when invoked, must run these steps:
  1. Let writer be the result of calling AcquireWritableStreamDefaultWriter(this).

  2. Let result be WritableStreamDefaultWriterWrite(writer, {type: "seek", position: position, size: undefined, data: undefined}).

  3. Perform WritableStreamDefaultWriterRelease(writer).

  4. Return result.

2.6.3. The truncate() method

await stream . truncate(size)

Resizes the file associated with stream to be size bytes long. If size is larger than the current file size this pads the file with null bytes, otherwise it truncates the file.

The file cursor is updated when truncate is called. If the offset is smaller than offset, it remains unchanged. If the offset is larger than size, the offset is set to size to ensure that subsequent writes do not error.

No changes are written to the actual file until on disk until the stream has been closed. Changes are typically written to a temporary file instead.

The truncate(size) method, when invoked, must run these steps:
  1. Let writer be the result of calling AcquireWritableStreamDefaultWriter(this).

  2. Let result be WritableStreamDefaultWriterWrite(writer, {type: "truncate", size: size, position: undefined, data: undefined}).

  3. Perform WritableStreamDefaultWriterRelease(writer).

  4. Return result.

3. Accessing Native File System

3.1. Native File System Permissions

The native file system query permission steps for an entry entry given a FileSystemHandlePermissionDescriptor descriptor are:
  1. If descriptor.writable is true:

    1. Let readState be the result of running the query permission steps for entry given «[ "writable" → false ]».

    2. If readState is not "granted", return readState.

  2. Let settings be the current settings object.

  3. If there was a previous invocation of this algorithm for the same entry, descriptor and settings, returning previousResult, and the UA has not received new information about the user’s intent since that invocation, return previousResult.

  4. Return whichever of the following options most accurately reflects the user’s intent for the calling algorithm:

    succeed without prompting the user

    "granted"

    show the user a prompt to decide whether to succeed

    "prompt"

    fail without prompting the user

    "denied"

The native file system request permission steps for an entry entry given a FileSystemHandlePermissionDescriptor descriptor are:
  1. Let status be the result of running the query permission steps for entry given descriptor.

  2. If status is not "prompt" return.

  3. If the current global object is not a Window, return.

    Should this throw instead of returning silently?

  4. If the current global object does not have transient activation, throw a NotAllowedError.

  5. Let environment be the current settings object.

  6. If environment’s origin is not same origin with environment’s top-level origin, throw a NotAllowedError.

  7. Display a prompt to the user requesting access as described by descriptor to entry.

  8. Wait for the user to accept or reject the prompt. The user’s interaction may provide new information about the user’s intent for this descriptor and entry. Additionally if descriptor.writable is true, the user’s interaction may also provide new information about the user’s intent for the same entry with «[ "writable" → false ]».

Note: This is intentionally vague about the details of the permission UI and how the UA infers user intent. UAs should be able to explore a variety of UI approaches within this framework.

3.2. The chooseFileSystemEntries() method

enum ChooseFileSystemEntriesType { "open-file", "save-file", "open-directory" };

dictionary ChooseFileSystemEntriesOptionsAccepts {
  USVString description;
  sequence<USVString> mimeTypes;
  sequence<USVString> extensions;
};

dictionary ChooseFileSystemEntriesOptions {
    ChooseFileSystemEntriesType type = "open-file";
    boolean multiple = false;
    sequence<ChooseFileSystemEntriesOptionsAccepts> accepts;
    boolean excludeAcceptAllOption = false;
};

[SecureContext]
partial interface Window {
    Promise<(FileSystemHandle or sequence<FileSystemHandle>)>
        chooseFileSystemEntries(optional ChooseFileSystemEntriesOptions options = {});
};
result = await window . chooseFileSystemEntries(options)

Shows a file picker dialog to the user and returns handles for the selected files or directories.

The options argument sets options that influence the behavior of the shown file picker.

options.type specifies the type of the entry the website wants the user to pick. When set to "open-file" (the default), the user can select only existing files. When set to "save-file" the dialog will additionally let the user select files that don’t yet exist, and if the user selects a file that does exist already, its contents will be cleared before the handle is returned to the website. Finally when set to "open-directory", the dialog will let the user select directories instead of files.

If options.multiple is false (or absent) the user can only select a single file, and the result will be a single FileSystemHandle. If on the other hand options.multiple is true, the dialog can let the user select more than one file, and result will be an array of FileSystemHandle instances (even if the user did select a single file, if multiple is true this will be returned as a single-element array).

Finally options.accepts and options.excludeAcceptAllOption specify the types of files the dialog will let the user select. Each entry in options.accepts describes a single type of file, consisting of a description, zero or more mimeTypes and zero or more extensions. Options with no valid mimeTypes and no extensions are invalid and are ignored. If no description is provided one will be generated.

If options.excludeAcceptAllOption is true, or if no valid entries exist in options.accepts, an option matching all files will be included in the file types the dialog lets the user select.

The chooseFileSystemEntries(options) method, when invoked, must run these steps:
  1. Let environment be the current settings object.

  2. If environment’s origin is an opaque origin, return a promise rejected with a SecurityError.

  3. Let browsing context be environment’s responsible browsing context.

  4. Let top-level context be browsing context’s top-level browsing context.

  5. If environment’s origin is not same origin with browsing context’s top-level browsing context's active document's origin, return a promise rejected with a SecurityError.

    There must be a better way to express this "no third-party iframes" constraint.

  6. TODO (Use the native file system query permission steps and native file system request permission steps in the returned entries).

4. Accessing special file systems

4.1. Special File System Concepts

A bucket contains a sandboxed file system root, a directory entry.

The sandboxed file system root for one bucket must be completely separate from that of any other bucket.

Note: While user agents will typically implement this by persisting the contents of this sandboxed file system root to disk, it is not intended that the contents are easily user accessible. Similarly there is no expectation that files or directories with names matching the names of children of the sandboxed file system root exist.

Note: In Chrome this sandboxed file system root refers to the same storage as the temporary file system as used to be defined in File API: Directories and System.

The sandboxed file system root's query permission steps are the following:
  1. Return "granted".

4.2. The getSystemDirectory() method

enum SystemDirectoryType {
  "sandbox"
};

dictionary GetSystemDirectoryOptions {
  required SystemDirectoryType type;
};

[SecureContext]
partial interface FileSystemDirectoryHandle {
  static Promise<FileSystemDirectoryHandle> getSystemDirectory(GetSystemDirectoryOptions options);
};
directoryHandle = FileSystemDirectoryHandle . getSystemDirectory({ type: "sandbox" })

Returns the sandboxed file system.

getSystemDirectory might not be the best name. Also perhaps should be on Window rather than on FileSystemDirectoryHandle. <https://github.com/wicg/native-file-system/issues/27>

The getSystemDirectory(options) method, when invoked, must run these steps:
  1. Let environment be the current settings object.

  2. If environment’s origin is an opaque origin, return a promise rejected with a SecurityError.

  3. Assert: options.type is "sandbox".

  4. Let storage be environment’s origin's site storage unit.

  5. Let bucket be storage’s bucket.

  6. Return a promise resolved with a new FileSystemDirectoryHandle, whose associated entry is bucket’s sandboxed file system root.

5. Privacy Considerations

This section is non-normative.

This API does not give websites any more read access to data than the existing <input type=file> and <input type=file webkitdirectory> APIs already do. Furthermore similarly to those APIs, all access to files and directories is explicitly gated behind a file or directory picker.

There are however several major privacy risks with this new API:

5.1. Users giving access to more, or more sensitive files than they intended.

This isn’t a new risk with this API, but user agents should try to make sure that users are aware of what exactly they’re giving websites access to. This is particularly important when giving access to a directory, where it might not be immediately clear to a user just how many files actually exist in that directory.

A related risk is having a user give access to particularly sensitive data. This could include some of a user agent’s configuration data, network cache or cookie store, or operating system configuration data such as password files. To protect against this, user agents are encouraged to restrict which directories a user is allowed to select in a directory picker, and potentially even restrict which files the user is allowed to select. This will make it much harder to accidentally give access to a directory that contains particularly sensitive data. Care must be taken to strike the right balance between restricting what the API can access while still having the API be useful. After all, this API intentionally lets the user use websites to interact with some of their most private personal data.

5.2. Websites trying to use this API for tracking.

This API could be used by websites to track the user across clearing browsing data. This is because, in contrast with existing file access APIs, user agents are able to grant persistent access to files or directories and can re-prompt. In combination with the ability to write to files, websites will be able to persist an identifier on the users' disk. Clearing browsing data will not affect those files in any way, making these identifiers persist through those actions.

This risk is somewhat mitigated by the fact that clearing browsing data will also clear IndexedDB, so websites won’t have any handles to re-prompt for permission after browsing data was cleared. Furthermore user agents are encouraged to make it clear what files and directories a website has access to, and to automatically expire permission grants except for particularly well trusted origins (for example persistent permissions could be limited to "installed" web applications).

User agents also are encouraged to provide a way for users to revoke permissions granted. Clearing browsing data is expected to revoke all permissions as well.

5.3. First-party vs third-party contexts.

In third-party contexts (e.g. an iframe whose origin does not match that of the top-level frame) websites can’t gain access to data they don’t already have access to. This includes both getting access to new files or directories via the chooseFileSystemEntries API, as well as requesting more permissions to existing handles via the requestPermission API.

Handles can also only be post-messaged to same-origin destinations. Attempts to send a handle to a cross-origin destination will result in a messageerror event.

6. Security Considerations

This section is non-normative.

This API gives websites the ability to modify existing files on disk, as well as write to new files. This has a couple of important security considerations:

6.1. Malware

This API could be used by websites to try to store and/or execute malware on the users system. To mitigate this risk, this API does not provide any way to mark files as executable (on the other hand files that are already executable likely remain that way, even after the files are modified through this API). Furthermore user agents are encouraged to apply things like Mark-of-the-Web to files created or modified by this API.

Finally, user agents are encouraged to verify the contents of files modified by this API via malware scans and safe browsing checks, unless some kind of external strong trust relation already exists. This of course has effects on the performance characteristics of this API.

"Atomic writes" attempts to make it explicit what this API can and can’t do, and how performance can be effected by safe browsing checks. <https://github.com/wicg/native-file-system/issues/51>

6.2. Ransomware attacks

Another risk factor is that of ransomware attacks. The limitations described above regarding blocking access to certain sensitive directories helps limit the damage such an attack can do. Additionally user agents can grant write access to files at whatever granularity they deem appropriate.

Conformance

Conformance requirements are expressed with a combination of descriptive assertions and RFC 2119 terminology. The key words “MUST”, “MUST NOT”, “REQUIRED”, “SHALL”, “SHALL NOT”, “SHOULD”, “SHOULD NOT”, “RECOMMENDED”, “MAY”, and “OPTIONAL” in the normative parts of this document are to be interpreted as described in RFC 2119. However, for readability, these words do not appear in all uppercase letters in this specification.

All of the text of this specification is normative except sections explicitly marked as non-normative, examples, and notes. [RFC2119]

Examples in this specification are introduced with the words “for example” or are set apart from the normative text with class="example", like this:

This is an example of an informative example.

Informative notes begin with the word “Note” and are set apart from the normative text with class="note", like this:

Note, this is an informative note.

Index

Terms defined by this specification

Terms defined by reference

References

Normative References

[ENCODING]
Anne van Kesteren. Encoding Standard. Living Standard. URL: https://encoding.spec.whatwg.org/
[FILE-API]
Marijn Kruisselbrink; Arun Ranganathan. File API. 11 September 2019. WD. URL: https://www.w3.org/TR/FileAPI/
[HTML]
Anne van Kesteren; et al. HTML Standard. Living Standard. URL: https://html.spec.whatwg.org/multipage/
[INFRA]
Anne van Kesteren; Domenic Denicola. Infra Standard. Living Standard. URL: https://infra.spec.whatwg.org/
[PERMISSIONS]
Mounir Lamouri; Marcos Caceres; Jeffrey Yasskin. Permissions. 25 September 2017. WD. URL: https://www.w3.org/TR/permissions/
[RFC2119]
S. Bradner. Key words for use in RFCs to Indicate Requirement Levels. March 1997. Best Current Practice. URL: https://tools.ietf.org/html/rfc2119
[STORAGE]
Anne van Kesteren. Storage Standard. Living Standard. URL: https://storage.spec.whatwg.org/
[STREAMS]
Adam Rice; Domenic Denicola; 吉野剛史 (Takeshi Yoshino). Streams Standard. Living Standard. URL: https://streams.spec.whatwg.org/
[WebIDL]
Boris Zbarsky. Web IDL. 15 December 2016. ED. URL: https://heycam.github.io/webidl/

Informative References

[ENTRIES-API]
File and Directory Entries API. cg-draft. URL: https://wicg.github.io/entries-api/
[FETCH]
Anne van Kesteren. Fetch Standard. Living Standard. URL: https://fetch.spec.whatwg.org/
[FILE-SYSTEM-API]
Eric Uhrhane. File API: Directories and System. 24 April 2014. NOTE. URL: https://www.w3.org/TR/file-system-api/

IDL Index

dictionary FileSystemHandlePermissionDescriptor {
  boolean writable = false;
};

[Exposed=(Window,Worker), SecureContext, Serializable]
interface FileSystemHandle {
  readonly attribute boolean isFile;
  readonly attribute boolean isDirectory;
  readonly attribute USVString name;

  Promise<boolean> isSameEntry(FileSystemHandle other);

  Promise<PermissionState> queryPermission(optional FileSystemHandlePermissionDescriptor descriptor = {});
  Promise<PermissionState> requestPermission(optional FileSystemHandlePermissionDescriptor descriptor = {});
};

dictionary FileSystemCreateWritableOptions {
  boolean keepExistingData = false;
};

[Exposed=(Window,Worker), SecureContext, Serializable]
interface FileSystemFileHandle : FileSystemHandle {
  Promise<File> getFile();
  Promise<FileSystemWritableFileStream> createWritable(optional FileSystemCreateWritableOptions options = {});
};

dictionary FileSystemGetFileOptions {
  boolean create = false;
};

dictionary FileSystemGetDirectoryOptions {
  boolean create = false;
};

dictionary FileSystemRemoveOptions {
  boolean recursive = false;
};

[Exposed=(Window,Worker), SecureContext, Serializable]
interface FileSystemDirectoryHandle : FileSystemHandle {
  Promise<FileSystemFileHandle> getFile(USVString name, optional FileSystemGetFileOptions options = {});
  Promise<FileSystemDirectoryHandle> getDirectory(USVString name, optional FileSystemGetDirectoryOptions options = {});

  // This really returns an async iterable, but that is not yet expressable in WebIDL.
  object getEntries();

  Promise<void> removeEntry(USVString name, optional FileSystemRemoveOptions options = {});

  Promise<sequence<USVString>?> resolve(FileSystemHandle possibleDescendant);
};

enum WriteCommandType {
  "write",
  "seek",
  "truncate",
};

dictionary WriteParams {
  required WriteCommandType type;
  unsigned long long? size;
  unsigned long long? position;
  (BufferSource or Blob or USVString)? data;
};

typedef (BufferSource or Blob or USVString or WriteParams) FileSystemWriteChunkType;

[Exposed=(Window,Worker), SecureContext]
interface FileSystemWritableFileStream : WritableStream {
  Promise<void> write(FileSystemWriteChunkType data);
  Promise<void> seek(unsigned long long position);
  Promise<void> truncate(unsigned long long size);
};

enum ChooseFileSystemEntriesType { "open-file", "save-file", "open-directory" };

dictionary ChooseFileSystemEntriesOptionsAccepts {
  USVString description;
  sequence<USVString> mimeTypes;
  sequence<USVString> extensions;
};

dictionary ChooseFileSystemEntriesOptions {
    ChooseFileSystemEntriesType type = "open-file";
    boolean multiple = false;
    sequence<ChooseFileSystemEntriesOptionsAccepts> accepts;
    boolean excludeAcceptAllOption = false;
};

[SecureContext]
partial interface Window {
    Promise<(FileSystemHandle or sequence<FileSystemHandle>)>
        chooseFileSystemEntries(optional ChooseFileSystemEntriesOptions options = {});
};

enum SystemDirectoryType {
  "sandbox"
};

dictionary GetSystemDirectoryOptions {
  required SystemDirectoryType type;
};

[SecureContext]
partial interface FileSystemDirectoryHandle {
  static Promise<FileSystemDirectoryHandle> getSystemDirectory(GetSystemDirectoryOptions options);
};

Issues Index

TODO: Explain better how entries map to files on disk (multiple entries can map to the same file or directory on disk but an entry doesn’t have to map to any file on disk).
Figure out some way to integrate this better with the permissions API [permissions], for example by making the entry be a part of the PermissionDescriptor.
the currently described API here assumes a model where it is not possible to have a write-only handle. I.e. it is not possible to have or request write access without also having read access. There definitely are use cases for write-only handles (e.g. directory downloads), so we might have to reconsider this. <https://github.com/wicg/native-file-system/issues/119>
There has been some discussion around and desire for a "inPlace" mode for createWritable (where changes will be written to the actual underlying file as they are written to the writer, for example to support in-place modification of large files or things like databases). This is not currently implemented in Chrome. Implementing this is currently blocked on figuring out how to combine the desire to run malware checks with the desire to let websites make fast in-place modifications to existing large files. <https://github.com/wicg/native-file-system/issues/67>
Should we have separate getFile and getDirectory methods, or just a single getChild/getEntry method?
Having getFile methods in both FileSystemDirectoryHandle and FileSystemFileHandle, but with very different behavior might be confusing? Perhaps rename at least one of them (but see also previous issue). <https://github.com/wicg/native-file-system/issues/98>
Should getEntries be its own method, or should FileSystemDirectoryHandle just be an async iterable itself? <https://github.com/wicg/native-file-system/issues/47>
Not clear if this should reject, and if it does, is this really the right error type? Chrome’s implementation is actually inconsistent about this rejecting or not.
Should this throw instead of returning silently?
There must be a better way to express this "no third-party iframes" constraint.
getSystemDirectory might not be the best name. Also perhaps should be on Window rather than on FileSystemDirectoryHandle. <https://github.com/wicg/native-file-system/issues/27>
"Atomic writes" attempts to make it explicit what this API can and can’t do, and how performance can be effected by safe browsing checks. <https://github.com/wicg/native-file-system/issues/51>