By default, Google’s externally accessible account service handles authentication for Google accounts. When an unauthenticated user visits a Google page that requires authentication, the Google login form requests the user’s email address and password. After the user submits their email and password, Google authentication verifies that the credentials entered are correct. If the credentials are correct, Google authentication sets the user’s login cookies.
Some enterprises use a more sophisticated model in which a third-party identity provider (IdP) handles authentication. Google authentication supports this model via the industry-standard Security Assertion Markup Language (SAML) protocol. An administrator can configure a domain to use SAML authentication.
ChromeOS needs to identify the user’s password entered during login to:
- Encrypt the user’s data stored on the disk drive.
- Protect the lock screen.
- Enable offline login when there is no network accessibility.
When using SAML, the password is not entered on a ChromeOS system dialog directly, but inside a webview hosted by the identity provider. While ChromeOS has access to the HTML, there is no simple, canonical way to get the password as it is unclear which form fields contain the data.
There are two ways to get a user’s password when using SAML: the Credentials Passing API and password scraping.
A SAML identity provider may use password scraping when they don’t support the Credentials Passing API.
In this method:
- The authentication screen injects a content script into the webview that hosts the login process.
- The content script identifies HTML input fields of type password and copies their contents into an array. The array is updated whenever the contents of a password field changes.
- Scraped passwords are sent to a background page that accumulates them. This way, the password can be captured even if the login flow spans multiple redirects to different HTML pages.
At the end of the login flow, the array of scraped passwords is retrieved from the background page. Three cases are possible: no password was scraped, exactly one password was scraped, or more than one password was scraped.
The content script fails to locate the password in the HTML pages served by the identity provider. The identity provider may not use traditional passwords.
In this scenario, ChromeOS will prompt the user to pick a manual password for the device. If the password does not exist (e.g. authentication by smart cards, NFC, biometry), the ChromeOS authentication process may proceed without the password.
The content script identifies exactly one password. Most likely, this is the user’s password used for authentication.
In this scenario, we most likely scraped the user’s password correctly. ChromeOS will use the scraped password as the user’s password to continue the authentication process.
The content script identifies multiple passwords. This may happen under circumstances such as an identity provider requiring a user to enter a permanent password and a one-time password into the login form.
In this scenario, we probably scraped the user’s actual password and some additional password fields that are not of interest to ChromeOS. To determine which one is the correct password, ChromeOS will prompt the user to enter the password once more into an additional password prompt.
If the password entered matches one of the scraped passwords, the user’s actual password has been identified and the authentication process will continue. If there is no match, the user will be prompted to enter their password again. After two mismatches, login fails with an error message.
For Enterprise enrollment, the enrolling user’s email address is needed to associate the device with the correct domain. The email is sent from the Device Management (DM) Server to Chrome in the username field of the PolicyData message during device policy fetch. There is no need to determine the user’s password.