fixes potential XSS issue with TOTP exception lists (which need escaping, see also #1185).
refactors Template escaping to use TemplateEscape enum (rather than the previous boolean).
thanks @TrapoSAMA
Someone wishing to remain anonymous emailed @cboltz a number of security issues, summarised below.
1. fix sql injection bug in getPageBrowser - search keys bypass the field filter, due to 2a6f27c3 / #1127
2. when performing a password change, do not start the session before the new password has been validated for compexity etc - as things stand someone with a recovery code gets logged in even if their password gets
rejected. Also invalid recovery code on failure.
3. broadcast-message.php was using smtp_mail() with incorrect args, and passing the SMTP password in as the mail body. - probably related to commit 7f1d78e and 42a664e9c2
4. a regular (non-admin) user could submit an arbitrary username in the POST data and create/edit a TOTP exception for another user or entire domain
PR includes refactoring of the password change code (into PasswordChangeAttempt) to make it easier to test/verify.
Wenhao Wu emailed with the following security report, this affects master and postfixadmin-4.0.
High — mailbox takeover of mail flow for a 2FA-enrolled account by a password holder. CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N ≈ 8.1 (PR:L models possession of the first factor; impact is limited to the attacker's own mailbox account). CWE-308 (Use of Single-Factor Authentication with Two Factors), secondary CWE-287. Not related to the fixed TOTP-exception issue #1014.
....
With `$CONF['totp']='YES'` and a mailbox that has TOTP enrolled, logging in with **only the mailbox password** creates an MFA-incomplete session (`username` set, `roles=[]`, `mfa_complete=false`) that is parked at `users/login-mfa.php`. That half-session cannot open `users/main.php` (it has no `user` role), but `public/edit.php` never checks MFA completion:
- `public/edit.php:30` — `authentication_get_username()` (comment "enforce login") treats any session with a username as logged in; it never calls `authentication_mfa_incomplete()`.
- `public/edit.php:57-63` — for a non-admin session with a handler that has `user_hardcoded_field`, `require_role('admin')` is skipped entirely.
....
Wenhao Wu, Southeast University
thanks @d3do-23 (@cboltz and TrapoSAMA for reviewing etc)
- require a completed user or admin role before opening TOTP management
- remove TOTP through the existing password-verifying change path
- keep configured post-change hooks active when a secret is removed
- add regression coverage proving an invalid password cannot remove the stored secret
This is the first focused change for #1128.
thanks @TrapoSAMA
Require a validated canonical site_url for recovery links, keep responses uniform, and persist a five-minute per-account request limit.
The token update is atomic across supported database engines and performs the same password hashing work for unknown accounts.
Refs #1128
thanks @TrapoSAMA
* Fix privilege check ordering in TotpPf::addException() (#1012)
Global admins were incorrectly downgraded to regular admins because
authentication_has_role('admin') was checked before 'global-admin',
and global admins have both roles. Reorder to check global-admin first.
* Add missing isset() check on $_GET['abort'] in users/login-mfa.php (#1013)
Matches the admin version at public/login-mfa.php which already has
the isset() check.
possible fix for https://github.com/postfixadmin/postfixadmin/issues/967
CSRF Token changes : rather than have one PFA_Token value in $_SESSION, store a list of them, where each value has a time stamp it was issued on. The aim is to allow the user to have multiple browser tabs open at once - and not have the last tab that loaded trample on others. Tokens expire after 1 hour.
Adds a Smarty helper function `{CSRF_Token}` which will output `<input type=hidden name=CSRF_Token value=asdf123>` or `{CSRF_Token type=url}` which outputs a literal string like 'asdf123' suitable for placing in a URL.
Adds a PHP exception handler in public/common.php, which could be used to display a more friendly error message/response (but at the moment is fairly minimal).
Removes most uses of `die()` within the codebase, and instead moves to throw an exception