.--~~~~~~~~~~~~~------.
/--===============------\
| |```````````````| |
| | | |
| | >_< | |
| | | |
| |_______________| |
| ::::|
'======================='
//-"-"-"-"-"-"-"-"-"-"-\\
//_"_"_"_"_"_"_"_"_"_"_"_\\
[-------------------------]
\_________________________/
Secure Shell Developer Guide
Secure Shell is a Chrome extension that combines hterm with a NaCl build of OpenSSH to provide a PuTTY-like app for Chrome users.
See /HACK.md for general information about working with the source control setup.
The Secure Shell app depends on some library code from libapps/libdot/ and the hterm terminal emulator from in libapps/hterm/. To build these external dependencies, run...
nassh$ ./bin/mkdeps
This will create the nassh/js/nassh_deps.concat.js file containing all of the necessary libdot and hterm source. If you find yourself changing a lot of libdot or hterm code and testing those changes in Secure Shell you can run this script with the “--forever” (aka -f) option. When run in this manner it will automatically re-create nassh_deps.concat.js file whenever one of the source files is changed.
Secure Shell depends on a NaCl (Native Client) plugin to function. This plugin is a port of OpenSSH. You'll have to find or create a version of this plugin, and copy it into libapps/nassh/plugin/.
Your options are (pick one):
# In the ssh_client/ directory. $ cp -a output/hterm/plugin/ ../nassh/
# In the nassh/ directory. $ wget https://commondatastorage.googleapis.com/chromeos-localmirror/secureshell/releases/0.27.tar.xz $ tar -xvf 0.27.tar.xz plugin/
plugin/ directory from the latest version of Secure Shell. If you have Secure Shell installed, the plugin can be found in your profile directory, under Default/Extensions/pnhechapfaindjhompbnflcldabbghjo/<version>/plugin/.Loading directly from the checked out nassh directory is the normal way of testing. It will use the dev extension id to avoid conflicts with the stable extension id, although it will still conflict if you have the dev version installed from the Chrome Web Store (CWS).
You will need to manually select the variant you want to work on. The extension is defined by the manifest_ext.json file while the app is defined by the manifest_v1.5.json file. Simply symlink it to manifest.json:
nassh$ ln -s manifest_ext.json manifest.json
For details on the different manifests and modes, see the next section (and the FAQ). You probably want to start with the extension version if you're not going to be hacking on Chrome OS features.
The extension id is controlled by the key field in the manifest.json. See the manifest key docs for more details.
Now that your checkout is ready, you can load it into Chrome.
chrome://extensions page.nassh/ directory.If you‘re not running on Chrome OS device, and loading the app, you might see warnings right away about certain permissions (see the whitelisted sections below). You can ignore those. It’s unfortunate they show up with the same level/color as legitmate errors.
* 'file_system_provider_capabilities' is only allowed for extensions and packaged apps, but this is a legacy packaged app. * 'terminalPrivate' is not allowed for specified platform. * 'fileSystemProvider' is not allowed for specified platform.
There are three manifest files in here currently. The manifest_v1.5.json is used to build the Secure Shell App and has been what we've used for the longest time, but is largely for Chrome OS only now. The manifest_ext.json is used to build the Secure Shell Extension which works on all platforms (but lacks any Chrome OS specific features).
The “v1.5” and “v2” app formats should not be confused with the “v1” and “v2” manifest formats. Secure Shell uses the legacy/deprecated “v1.5” app style to launch itself rather than the “v2” style. It means that, in many ways, the “v1.5” app behaves more like an extension (e.g. it shares cookies with your main browser instance and can run inside a tab) rather than an app (e.g. it doesn't get access to many newer chrome.app.* APIs).
See the FAQ for more details on the differences between the extension & app.
If you‘re updating these files, you’ll sometimes also need to update the manifest for [crosh] which lives in the Chromium tree.
Using the dev extension id is necessary in order to access some APIs that are whitelisted only for Secure Shell. If you don't need these features, you can get by with using a different id (and delete the settings from the manifest_v1.5.json for the app to avoid warnings at runtime). These settings are already removed from the manifest_ext.json for the extension.
terminalPrivate). (1) (2) (3)chrome.crashReportPrivate APIs. This allows users to opt-in to providing crash-reports. Not required to be able to use Secure Shell of course.chrome.metricsPrivate APIs. This allows users to opt-in to metrics/UMA collection about basic features. Not required to be able to use Secure Shell of course.fileSystemProvider and file_system_provider_capabilities). (1) To double check what & where things are whitelisted, search the Chromium code base for our extension ids:
The hashes are the SHA1's of the (lower case) extension id.
$ ext_id_hash() { printf "$1" | tr '[:upper:]' '[:lower:]' | sha1sum | tr '[:lower:]' '[:upper:]'; } $ ext_id_hash pnhechapfaindjhompbnflcldabbghjo 0EA6B717932AD64C469C1CCB6911457733295907
There are a few extensions we talk to for various services at runtime which need to whitelist our extension ids (for source verification). If you don't need any of these services, then you can ignore it.
If you try to load an unpacked extension using the stable extension id, you might run into problems if your administrator installs it via enterprise policy. If you see the error below, you won't be able to bypass it. Just use the dev extension id instead.
Secure Shell (extension ID "pnhechapfaindjhompbnflcldabbghjo") is blocked by the administrator.
While most of the UI code for [crosh] lives here (e.g. HTML/CSS/JS), the backend code and manifest lives in Chrome.
chrome.terminalPrivate JavaScript API.chrome.terminalPrivate JavaScript API.api/terminal/ code.See the libapps hacking document for details.
Keep in mind that the NaCl [ssh_client] code does not live here.
The vast majority of the code here lives under js/.
*.concat.js: Compiled JS output of other projects we use.nassh object setup and glue code to Chrome runtime.nassh.App code.nassh.CommandInstance launching code.nassh.relay.Corp code for proxying connections via [Corp Relay] protocol.nassh.Stream streams.*_tests.js: Each module has a corresponding set of unittests. The filename follows the convention of adding _tests. e.g. nassh_tests.js contains the tests for nassh.js.There are a few specialized modules that are not relevant to the core Secure Shell logic.
When the extension is launched (e.g. a new connection is opened), the background page is automatically created. This is used to monitor global state like extension updates and coordinate SFTP mounts. The logic lives in nassh_background.js and takes care of creating a new instance of nassh.App which it saves in the background page‘s app variable. If you aren’t looking at the SFTP logic, you can probably ignore this code.
When the extension is run, a new nassh.html window is shown. If no connection info is provided via the URL, then an iframe is created to show nassh_connect_dialog.html. Here the user manages their saved list of connections and picks the one they want to connect to. This logic is in nassh_connect_dialog.js. Once the user makes a selection (either connecting or mounting), a message is sent to nassh_command_instance.js. There the connection dialog is closed, the NaCl plugin is loaded, and the streams are connected to hterm.
Since the ssh program uses positive exit statuses, we tend to use -1 for internal exit states in the JS code. It doesn't matter too much as the exit values are purely for showing the user.
Here is the API that the JS code uses to communicate with the NaCl [ssh_client] module.
The nassh.CommandInstance.prototype.sendToPlugin_ function in nassh_command_instance.js is used to package up and make all the calls. Helper functions are also provided in that file to avoid a JS API to callers.
At the lowest level, we pass a dictionary to the plugin. It has two fields, both of which must be specified (even if arguments is just []).
name: The function we want to call (as a string).arguments: An array of arguments to the function.The name field can be any one of:
| Function name | Description | Arguments |
|---|---|---|
startSession | Start a new ssh connection! | (object session) |
onOpenFile | Open a new file. | (int fd, bool success, bool is_atty) |
onOpenSocket | Open a new socket. | (int fd, bool success, bool is_atty) |
onRead | Send new data to the plugin. | (int fd, ArrayBuffer data) |
onWriteAcknowledge | Tell plugin we've read data. | (int fd, number count) |
onClose | Close an existing fd. | (int fd) |
onReadReady | Notify plugin data is available. | (int fd, bool result) |
onResize | Notify terminal size changes. | (int width, int height) |
onExitAcknowledge | Used to quit the plugin. | () |
The session object currently has these members:
username: Username for accessing the remote system.host: Hostname for accessing the remote system.port: Port number for accessing the remote system.terminalWidth: Initial width of the terminal window.terminalHeight: Initial height of the terminal window.useJsSocket: Whether to use JS for network traffic.environment: A key/value object of environment variables.arguments: Extra command line options for ssh.writeWindow: Size of the write window.authAgentAppID: Extension id to use as the ssh-agent.subsystem: Which subsystem to launch.The onWriteAcknowledge count field tracks the total byte count sent for the connection, not the count from the most recent write request. It supports up to Number.MAX_SAFE_INTEGER bytes.
Here is the API that the NaCl [ssh_client] code uses to communicate with the JS layers.
At the lowest level, we pass a dictionary to the JS code. It has two fields, both of which must be specified (even if arguments is just []).
name: The function we want to call (as a string).arguments: An array of arguments to the function.The name field can be any one of:
| Function name | Description | Arguments |
|---|---|---|
openFile | Plugin wants to open a file. | (int fd, str path, int mode) |
openSocket | Plugin wants to open a socket. | (int fd, str host, int port) |
read | Plugin wants to read data. | (int fd, int count) |
write | Plugin wants to write data. | (int fd, ArrayBuffer data) |
close | Plugin wants to close an fd. | (int fd) |
exit | The plugin is exiting. | (int code) |
printLog | Send a string to console.log. | (str str) |
On Chrome OS, it is possible to mount remote paths via SFTP and the Files app. We currently support [version 3][SFTPv3] of the protocol. We don't support newer standards because the most popular implementation is [OpenSSH]'s which only supports SFTPv3, and for the majority of Secure Shell users, they only interact with [OpenSSH].
We could support newer versions, but they wouldn‘t be well tested, and not a lot of people would even use it, and it’s not like we'd see any performance improvements (as our operations tend to be basic open/read/write/close).
We support a few optional extensions to improve behavior or performance.
The [SFTPv3] protocol says the argument order should be linkpath then the targetpath, but OpenSSH made a mistake and reversed the order. They can't change the order without breaking existing clients or servers, so they document it and leave it as-is.
We follow the [OpenSSH SFTP Protocol] here, as do many other clients.
[SFTPv6] noted this desync between implementations and the specification and replaced the SSH_FXP_SYMLINK packet type with a new SSH_FXP_LINK.
The [OpenSSH SFTP Protocol] defines a fsync@openssh.com extension which takes an open handle and instructs the remote server to use the fsync() syscall.
We use this extension when the user has requested syncing after writes.
The [OpenSSH SFTP Protocol] defines a hardlink@openssh.com extension with the same API as SSH_FXP_SYMLINK.
We use this extension when the user has requested hardlinks explicitly.
The [OpenSSH SFTP Protocol] defines a posix-rename@openssh.com extension with the same API as SSH_FXP_RENAME, but uses the simple rename(2) semantics on the server. Otherwise, the default rename operation can be a bit buggy/racy.
We use this extension when available, or fallback to SSH_FXP_RENAME if not.
The [OpenSSH SFTP Protocol] defines a statvfs@openssh.com extension which returns the struct statvfs data from statvfs() syscall.
We use this extension when the user has requested filesystem statistics.
The copy-data extension is great for speeding up remote copies as it avoids having to download data from one file and uploading to a different one.
Here's a random list of documents which would be useful to people.
[Corp Relay: relay-protocol.md#corp-relay [crosh]: chromeos-crosh.md [gnubbyd]: https://chrome.google.com/webstore/detail/beknehfpfkghjoafdifaflglpjkojoco [NaCl]: https://developer.chrome.com/native-client [Native Sockets]: https://crbug.com/909927 [OpenSSH]: https://www.openssh.com/ [OpenSSH SFTP Protocol]: https://github.com/openssh/openssh-portable/blob/master/PROTOCOL [SFTPv3]: https://tools.ietf.org/html/draft-ietf-secsh-filexfer-02 [SFTPv6]: https://tools.ietf.org/html/draft-ietf-secsh-filexfer-13 [ssh_client]: ../../ssh_client/ [wash]: ../../wash/