Security/Features/PasswordManagerImprovements

Please use "Edit with form" above to edit this page.

Status

Security Improvements to Password Manager
Stage	Draft
Status	`
Release target	`
Health	OK
Status note	`

Team

Product manager	`
Directly Responsible Individual	`
Lead engineer	`
Security lead	Tanvi Vyas
Privacy lead	`
Localization lead	`
Accessibility lead	`
QA lead	`
UX lead	`
Product marketing lead	`
Operations lead	`
Additional members	`

Open issues/risks

`

Stage 1: Definition

1. Feature overview

Hardening the Password Manager.

2. Users & use cases

• See https://www.usenix.org/conference/usenixsecurity14/technical-sessions/presentation/silver
• Include improvements to the Password Manager to prevent attacks such as "Lupin" in the paper "Automated Password Extraction Attack on Modern Password Managers". When a user is on an insecure network and makes any http request, a mitm could add login iframes to the http request. The mitm can then add javascript to the responses from the login pages that send the attacker the autocompleted passwords from password manager.

3. Dependencies

`

4. Requirements

• Preventing active network attacks:
  1. Do not autofill the username/password stored in password manager in certain situations (listed below). Provide alternative UX - UX help needed
     • For http sites (IE 11 has this security feature)
     • https sites that have mixed active content
     • https sites that require a cert override (chrome does this)
     • in iframed sites (where the parent and frame are not same orign, or always?) (safari does this for non-same origin, chrome does this for all frames). (Open Issue - what about third party widgets that allow users to login and post comments.)
     • Invisible form fields (visibility and opacity, although this isn't going to prevent clickjacking attacks to autofill the passwords)
  2. Highlight Cleartext Passwords - Warn users about entering their passwords on insecure pages - Bug 759860. UX help needed for this.
  3. Secure Filling - Passwords that are saved by the password manager should not be available to javascript on the page. The actual password values should only be sent on submit. This protects the password from attacks via xss, 3rd party javascript, etc.

• Preventing local attacks:
  • Explore solutions for encrypting the passwords stored locally in the password manager (for example, make use of keychain or encryption mechanisms that come with the OS).

• Duplicate Passwords - Protecting users from password reuse attacks
  • If an identical password is stored for both any http site and any https site (not necessarily same domain), and it is not used on the http site for X months, expire the http version. (This helps in cases where the hostname has changed, e.g. from mail.google.com to www.google.com.) If the http site and the https site are same domain, we can expire the http password in <X months (this can help in cases where the website has upgraded to https, but the user's http password manager entry still exists and is open to attack).

• HSTS: If a site is HSTS, then there is no reason to have http data for that site. Hence, if a site is marked HSTS, and the user has any data (cookies, passwords, etc) that are not https-only/secure, immediately mark that data as https-only. (This helps if a site uses STS, but the user's privacy settings cause the password storage to outlive the STS storage.)

---

Further ideas/thoughts that are not fully investigated.

• Assume we have implemented secure filling (javascript on the page can't read the password). If the user prompts the password manager to fill in a password on an HTTP page and the form action has changed since the password was stored, don't send the password (might be tricky to implement). Perhaps with an UI for user override. (Secure autofilling causes issues with ajax logins, a technique used mostly in china per https://www.usenix.org/conference/usenixsecurity14/technical-sessions/presentation/silver)

• When the password field name is different from the name when the password was saved, don't allow secure filling & submit. Moreover, don't allow javascript to dynamically change the name of the form?? See section 5.2, Preventing self exfiltration attacks.

• Question - If we don't allow javascript access to all password fields, could javascript on the page still grab key strokes?

• Working on the Credential Management Specification can help us in the long run with password detection and secure filling.

• Only autofill forms after user interaction with the login page, not necessarily the form - https://code.google.com/p/chromium/issues/detail?id=257156. We need more details here.

• Consider not autofilling passwords in following cases.
  1. If there are multiple login forms on a page??
  2. If the form action has changed since the password was stored. (To do this, we need to include the form action in the key, along with the url. We'd need to check the action both on load and on submit. What if it's a javascript action? More investigation needed here.)

• More on HSTS - When clearing history, retain the STS bit if any other data associated with the site is retained. For example, in the common case of clearing all history other than passwords, retain the STS bits for sites that have passwords stored. (Potential privacy issue)

• Prefer secure origins - If a password is stored in an http version of a site, see if the https version exists. If it does, prompt the user to redirect to the https version of the site and store their password there instead. (Issue here is that we don't always know if changing the url to https will work, or if a site is set up to have a different domain or path for their secure version)

• Secure Filling 2.0
  • Do not give javascript access to any password fields (regardless of whether the password manager saves the password) - the actual password the user enters is only used on submit. The problem with this is with registration pages that use javascript to test the password strength. Bug 653132

Non-goals

`

Stage 2: Design

5. Functional specification

`

6. User experience design

• What should the UX be when we do not autofill?
  1. Today the experience is that the user first has to type the username, and then the password will be filled in. [NB: it fails utterly on "password-only" forms (where the username is known to the site already, such as by a cookie) because there's no username field interaction to trigger the fill-in]
  2. We could have an fill-and-submit option, so that there is still only one click for the user. Instead of clicking "submit" they would click "fill and submit". This way, we've added security without decreasing usability.
  3. To prevent clickjacking attacks, the UI for this could live somewhere in chrome; the user could interact with the some password manager directly instead of the webpage.

• What should the UX be when we alert users of duplicate passwords?

• Can we somehow expose the about:config option signon.autofill to users to prevent attacks on saved passwords in HTTPS pages (ex: XSS bugs on the victim site). Alternatively, if we come up with a fill-and-submit that is more usable than autofill, we can deploy it on both HTTP and HTTPS pages.

Stage 3: Planning

7. Implementation plan

`

8. Reviews

Security review

`

Privacy review

`

Localization review

`

Accessibility

`

Quality Assurance review

`

Operations review

`

Stage 4: Development

9. Implementation

`

Stage 5: Release

10. Landing criteria

`

Feature details

Priority	Unprioritized
Rank	999
Theme / Goal	`
Roadmap	Security
Secondary roadmap	`
Feature list	`
Project	`
Engineering team	`

Team status notes

	status	notes
Products	`	`
Engineering	`	`
Security	`	`
Privacy	`	`
Localization	`	`
Accessibility	`	`
Quality assurance	`	`
User experience	`	`
Product marketing	`	`
Operations	`	`