Summary: Ref T6947.
Test Plan: toggled setting in application settings and changes stuck. set policy to admin user a only and could not add a provider as a admin user b.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, epriestley
Maniphest Tasks: T6947
Differential Revision: https://secure.phabricator.com/D11356
Summary:
Ref T6870. Since it does not make sense to redirect the user to the login form after they log in, we try not to set the login form as the `next` cookie.
However, the current check is hard-coded to `/auth/start/`, and the form can also be served at `/login/`. This has no real effect on normal users, but did make debugging T6870 confusing.
Instead of using a hard-coded path check, test if the controller was delegated to. If it was, store the URI. If it's handling the request without delegation, don't.
Test Plan:
- Visited login form at `/login/` and `/auth/start/`, saw it not set a next URI.
- Visited login form at `/settings/` (while logged out), saw it set a next URI.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley, lpriestley
Maniphest Tasks: T6870
Differential Revision: https://secure.phabricator.com/D11292
Summary: So meta it hurts. Fixes T887.
Test Plan: created a second instance of phabricator locally. made an account on oauth server phabricator. set up my normal dev phabricator to use this new oauth phabricator. noted the form worked. created an account via the oauth method and it worked.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, epriestley
Maniphest Tasks: T887
Differential Revision: https://secure.phabricator.com/D11287
Summary: Ref T6822.
Test Plan: Visual inspection. These methods are only called from within `PhabricatorController` subclasses.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin, epriestley
Maniphest Tasks: T6822
Differential Revision: https://secure.phabricator.com/D11241
Summary:
Ref T4712. Thus far, it seems that most "non-standard" things can be done pretty easily in the controller. Aside from deploying, this diff had to fix a few bugs / missing implementations of stuff.
(Notably, PhabricatorAuthProviderConfig, HeraldRule, PhabricatorSlowvotePoll, and AlmanacNetwork needed to implement PhabricatorApplicationTransactionInterface, PhabricatorAuthAuthProviderPHIDType had to be added, and a rendering bug in transactions of type PhabricatorOAuth2AuthProvider had to be fixed.)
Test Plan: Almanac - looked at binding, device, network, and service view controllers and verified timeline displayed properly. Herald - looked at a rule and verified timeline. Slowvote - looked at a vote and verified timeline. Auth - looked at an auth provider (Facebook) and verified proper display of transactions within timeline.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, epriestley
Maniphest Tasks: T4712
Differential Revision: https://secure.phabricator.com/D10921
Summary:
Ref T6240. Some discussion in that task. In instance/cluster environments, daemons need to make Conduit calls that bypass policy checks.
We can't just let anyone add SSH keys with this capability to the web directly, because then an adminstrator could just add a key they own and start signing requests with it, bypassing policy checks.
Add a `bin/almanac trust-key --id <x>` workflow for trusting keys. Only trusted keys can sign requests.
Test Plan:
- Generated a user key.
- Generated a device key.
- Trusted a device key.
- Untrusted a device key.
- Hit the various errors on trust/untrust.
- Tried to edit a trusted key.
{F236010}
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T6240
Differential Revision: https://secure.phabricator.com/D10878
Summary:
Ref T5833. I want to add SSH keys to Almanac devices, but the edit workflows for them are currently bound tightly to users.
Instead, decouple key management from users and the settings panel.
Test Plan:
- Uploaded, generated, edited and deleted SSH keys.
- Hit missing name, missing key, bad key format, duplicate key errors.
- Edited/generated/deleted/etc keys for a bot user as an administrator.
- Got HiSec'd on everything.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5833
Differential Revision: https://secure.phabricator.com/D10824
Summary: Missed this in previous pass. Send these as links in HTML emails.
Test Plan: Register a new user that nees approval.
Reviewers: btrahan, epriestley
Reviewed By: epriestley
Subscribers: Korvin, epriestley
Differential Revision: https://secure.phabricator.com/D10815
Summary:
Fixes T5900. We have some very old code here which does not let you update your password if the `account.editable` flag is set.
This was approximately introduced in D890, and I think it was mostly copy/pasted at that point. I'm not sure this ever really made sense. The option is not documented as affecting this, for example. In the modern environment of auth providers, it definitely does not make sense.
Instead, always allow users to change passwords if the install has a password provider configured.
Test Plan:
- Set `account.editable` to false.
- Used a password reset link.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5900
Differential Revision: https://secure.phabricator.com/D10331
Summary: Ref T5861. Adds an option to opt out of all notification email. We'll still send you password resets, email verifications, etc.
Test Plan:
{F189484}
- Added unit tests.
- With preference set to different things, tried to send myself mail. Mail respected preferences.
- Sent password reset email, which got through the preference.
Reviewers: btrahan, chad
Reviewed By: chad
Subscribers: rush898, epriestley
Maniphest Tasks: T5861
Differential Revision: https://secure.phabricator.com/D10237
Summary:
Via HackerOne. If a user adds an email address and typos it, entering `alinculne@gmailo.com`, and it happens to be a valid address which an evil user controls, the evil user can request a password reset and compromise the account.
This strains the imagination, but we can implement a better behavior cheaply.
- If an account has any verified addresses, only send to verified addresses.
- If an account has no verified addresses (e.g., is a new account), send to any address.
We've also received several reports about reset links not being destroyed as aggressively as researchers expect. While there's no specific scenario where this does any harm, revoke all outstanding reset tokens when a reset link is used to improve the signal/noise ratio of the reporting channel.
Test Plan:
- Tried to send a reset link to an unverified address on an account with a verified address (got new error).
- Tried to send a reset link to a verified adddress on an account with a verified address (got email).
- Tried to send a reset link to an invalid address (got old error).
- Tried to send a reset link to an unverified address on an account with only unverified addresses -- a new user (got email).
- Requested several reset links, used one, verified all the others were revoked.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D10206
Summary:
Fixes T5509. Currently, existing sessions live on even if you change your password.
Over the course of the program, we've recieved a lot of HackerOne reports that sessions do not terminate when users change their passwords. I hold that this isn't a security vulnerability: users can explicitly manage sessions, and this is more general and more powerful than tying session termination to password resets. In particular, many installs do not use a password provider at all (and no researcher has reported this in a general, application-aware way that discusses multiple authentication providers).
That said, dealing with these false positives is vaguely time consuming, and the "expected" behavior isn't bad for users, so just align behavior with researcher expectations: when passwords are changed, providers are removed, or multi-factor authentication is added to an account, terminate all other active login sessions.
Test Plan:
- Using two browsers, established multiple login sessions.
- In one browser, changed account password. Saw session terminate and logout in the second browser.
- In one browser, removed an authentication provider. Saw session terminate and logout in the second browser.
- In one browser, added MFA. Saw session terminate and logout in the second browser.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5509
Differential Revision: https://secure.phabricator.com/D10135
Summary:
Fixes T5506. Depends on D10133. When users remove an email address or change their primary email address, invalidate any outstanding password reset links.
This is a very small security risk, but the current behavior is somewhat surprising, and an attacker could sit on a reset link for up to 24 hours and then use it to re-compromise an account.
Test Plan:
- Changed primary address and removed addreses.
- Verified these actions invalidated outstanding one-time login temporary tokens.
- Tried to use revoked reset links.
- Revoked normally from new UI panel.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5506
Differential Revision: https://secure.phabricator.com/D10134
Summary:
Ref T5506. This makes it easier to understand and manage temporary tokens.
Eventually this could be more user-friendly, since it's relatively difficult to understand what this screen means. My short-term goal is just to make the next change easier to implement and test.
The next diff will close a small security weakness: if you change your email address, password reset links which were sent to the old address are still valid. Although an attacker would need substantial access to exploit this (essentially, it would just make it easier for them to re-compromise an already compromised account), it's a bit surprising. In the next diff, email address changes will invalidate outstanding password reset links.
Test Plan:
- Viewed outstanding tokens.
- Added tokens to the list by making "Forgot your password?" requests.
- Revoked tokens individually.
- Revoked all tokens.
- Tried to use a revoked token.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5506
Differential Revision: https://secure.phabricator.com/D10133
Summary: Ref T4896. Instead of using custom stuff, use standard stuff.
Test Plan: Viewed a bunch of feed stories and published some over the Asana bridge.
Reviewers: joshuaspence, btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4896
Differential Revision: https://secure.phabricator.com/D10114
Summary:
Fixes T3732. Ref T1205. Ref T3116.
External accounts (like emails used as identities, Facebook accounts, LDAP accounts, etc.) are stored in "ExternalAccount" objects.
Currently, we have a very restrictive `CAN_VIEW` policy for ExternalAccounts, to add an extra layer of protection to make sure users can't use them in unintended ways. For example, it would be bad if a user could link their Phabricator account to a Facebook account without proper authentication. All of the controllers which do sensitive things have checks anyway, but a restrictive CAN_VIEW provided an extra layer of protection. Se T3116 for some discussion.
However, this means that when grey/external users take actions (via email, or via applications like Legalpad) other users can't load the account handles and can't see anything about the actor (they just see "Restricted External Account" or similar).
Balancing these concerns is mostly about not making a huge mess while doing it. This seems like a reasonable approach:
- Add `CAN_EDIT` on these objects.
- Make that very restricted, but open up `CAN_VIEW`.
- Require `CAN_EDIT` any time we're going to do something authentication/identity related.
This is slightly easier to get wrong (forget CAN_EDIT) than other approaches, but pretty simple, and we always have extra checks in place anyway -- this is just a safety net.
I'm not quite sure how we should identify external accounts, so for now we're just rendering "Email User" or similar -- clearly not a bug, but not identifying. We can figure out what to render in the long term elsewhere.
Test Plan:
- Viewed external accounts.
- Linked an external account.
- Refreshed an external account.
- Edited profile picture.
- Viewed sessions panel.
- Published a bunch of stuff to Asana/JIRA.
- Legalpad signature page now shows external accounts.
{F171595}
Reviewers: chad, btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T3732, T1205, T3116
Differential Revision: https://secure.phabricator.com/D9767
Summary: Ran `arc lint --apply-patches --everything` over rP, mainly to change double quotes to single quotes where appropriate. These changes also validate that the `ArcanistXHPASTLinter::LINT_DOUBLE_QUOTE` rule is working as expected.
Test Plan: Eyeballed it.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin, hach-que
Differential Revision: https://secure.phabricator.com/D9431
Summary: Ref T5089. Adds a `security.require-multi-factor-auth` which forces all users to enroll in MFA before they can use their accounts.
Test Plan:
Config:
{F159750}
Roadblock:
{F159748}
After configuration:
{F159749}
- Required MFA, got roadblocked, added MFA, got unblocked.
- Removed MFA, got blocked again.
- Used `bin/auth strip` to strip MFA, got blocked.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T5089
Differential Revision: https://secure.phabricator.com/D9285
Summary: Both email verify and welcome links now verify email, centralize them and record them in the user activity log.
Test Plan:
- Followed a "verify email" link and got verified.
- Followed a "welcome" (verifying) link.
- Followed a "reset" (non-verifying) link.
- Looked in the activity log for the verifications.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D9284
Summary:
Ref T4398. This code hadn't been touched in a while and had a few crufty bits.
**One Time Resets**: Currently, password reset (and similar links) are valid for about 48 hours, but we always use one token to generate them (it's bound to the account). This isn't horrible, but it could be better, and it produces a lot of false positives on HackerOne.
Instead, use TemporaryTokens to make each link one-time only and good for no more than 24 hours.
**Coupling of Email Verification and One-Time Login**: Currently, one-time login links ("password reset links") are tightly bound to an email address, and using a link verifies that email address.
This is convenient for "Welcome" emails, so the user doesn't need to go through two rounds of checking email in order to login, then very their email, then actually get access to Phabricator.
However, for other types of these links (like those generated by `bin/auth recover`) there's no need to do any email verification.
Instead, make the email verification part optional, and use it on welcome links but not other types of links.
**Message Customization**: These links can come out of several workflows: welcome, password reset, username change, or `bin/auth recover`. Add a hint to the URI so the text on the page can be customized a bit to help users through the workflow.
**Reset Emails Going to Main Account Email**: Previously, we would send password reset email to the user's primary account email. However, since we verify email coming from reset links this isn't correct and could allow a user to verify an email without actually controlling it.
Since the user needs a real account in the first place this does not seem useful on its own, but might be a component in some other attack. The user might also no longer have access to their primary account, in which case this wouldn't be wrong, but would not be very useful.
Mitigate this in two ways:
- First, send to the actual email address the user entered, not the primary account email address.
- Second, don't let these links verify emails: they're just login links. This primarily makes it more difficult for an attacker to add someone else's email to their account, send them a reset link, get them to login and implicitly verify the email by not reading very carefully, and then figure out something interesting to do (there's currently no followup attack here, but allowing this does seem undesirable).
**Password Reset Without Old Password**: After a user logs in via email, we send them to the password settings panel (if passwords are enabled) with a code that lets them set a new password without knowing the old one.
Previously, this code was static and based on the email address. Instead, issue a one-time code.
**Jump Into Hisec**: Normally, when a user who has multi-factor auth on their account logs in, we prompt them for factors but don't put them in high security. You usually don't want to go do high-security stuff immediately after login, and it would be confusing and annoying if normal logins gave you a "YOU ARE IN HIGH SECURITY" alert bubble.
However, if we're taking you to the password reset screen, we //do// want to put the user in high security, since that screen requires high security. If we don't do this, the user gets two factor prompts in a row.
To accomplish this, we set a cookie when we know we're sending the user into a high security workflow. This cookie makes login finalization upgrade all the way from "partial" to "high security", instead of stopping halfway at "normal". This is safe because the user has just passed a factor check; the only reason we don't normally do this is to reduce annoyance.
**Some UI Cleanup**: Some of this was using really old UI. Modernize it a bit.
Test Plan:
- **One Time Resets**
- Used a reset link.
- Tried to reuse a reset link, got denied.
- Verified each link is different.
- **Coupling of Email Verification and One-Time Login**
- Verified that `bin/auth`, password reset, and username change links do not have an email verifying URI component.
- Tried to tack one on, got denied.
- Used the welcome email link to login + verify.
- Tried to mutate the URI to not verify, or verify something else: got denied.
- **Message Customization**
- Viewed messages on the different workflows. They seemed OK.
- **Reset Emails Going to Main Account Email**
- Sent password reset email to non-primary email.
- Received email at specified address.
- Verified it does not verify the address.
- **Password Reset Without Old Password**
- Reset password without knowledge of old one after email reset.
- Tried to do that without a key, got denied.
- Tried to reuse a key, got denied.
- **Jump Into Hisec**
- Logged in with MFA user, got factor'd, jumped directly into hisec.
- Logged in with non-MFA user, no factors, normal password reset.
- **Some UI Cleanup**
- Viewed new UI.
- **Misc**
- Created accounts, logged in with welcome link, got verified.
- Changed a username, used link to log back in.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4398
Differential Revision: https://secure.phabricator.com/D9252
Summary: This is useful when you're trying to onboard an entire office and you end up using the Google OAuth anyway.
Test Plan: tested locally. Maybe I should write some tests?
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Differential Revision: https://secure.phabricator.com/D9150
Summary: Did a more exhaustive grep on setIcon and found 99.9% of the icons.
Test Plan: I verified icon names on UIExamples, but unable to test some of the more complex flows visually. Mostly a read and replace.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: epriestley, Korvin, hach-que
Differential Revision: https://secure.phabricator.com/D9088
Summary: The removes the sprite sheet 'icons' and replaces it with FontAwesome fonts.
Test Plan:
- Grep for SPRITE_ICONS and replace
- Grep for sprite-icons and replace
- Grep for PhabricatorActionList and choose all new icons
- Grep for Crumbs and fix icons
- Test/Replace PHUIList Icon support
- Test/Replace ObjectList Icon support (foot, epoch, etc)
- Browse as many pages as I could get to
- Remove sprite-icons and move remarkup to own sheet
- Review this diff in Differential
Reviewers: btrahan, epriestley
Reviewed By: epriestley
Subscribers: epriestley, Korvin, hach-que
Differential Revision: https://secure.phabricator.com/D9052
Summary: Fixes T4728, first pass, Make real name optional on user accounts
Test Plan: Default real name config should be false (not required). Create new user, real name should not be required. Toggle config, real name should be required. Users with no real name should be always listed by their usernames.
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: epriestley, Korvin
Maniphest Tasks: T4728
Differential Revision: https://secure.phabricator.com/D9027
Summary:
Ref T4398. This prompts users for multi-factor auth on login.
Roughly, this introduces the idea of "partial" sessions, which we haven't finished constructing yet. In practice, this means the session has made it through primary auth but not through multi-factor auth. Add a workflow for bringing a partial session up to a full one.
Test Plan:
- Used Conduit.
- Logged in as multi-factor user.
- Logged in as no-factor user.
- Tried to do non-login-things with a partial session.
- Reviewed account activity logs.
{F149295}
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4398
Differential Revision: https://secure.phabricator.com/D8922
Summary:
Ref T4398. This adds a settings panel for account activity so users can review activity on their own account. Some goals are:
- Make it easier for us to develop and support auth and credential information, see T4398. This is the primary driver.
- Make it easier for users to understand and review auth and credential information (see T4842 for an example -- this isn't there yet, but builds toward it).
- Improve user confidence in security by making logging more apparent and accessible.
Minor corresponding changes:
- Entering and exiting hisec mode is now logged.
- This, sessions, and OAuth authorizations have moved to a new "Sessions and Logs" area, since "Authentication" was getting huge.
Test Plan:
- Viewed new panel.
- Viewed old UI.
- Entered/exited hisec and got prompted.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4398
Differential Revision: https://secure.phabricator.com/D8871
Summary:
Ref T4398. This is roughly a "sudo" mode, like GitHub has for accessing SSH keys, or Facebook has for managing credit cards. GitHub actually calls theirs "sudo" mode, but I think that's too technical for big parts of our audience. I've gone with "high security mode".
This doesn't actually get exposed in the UI yet (and we don't have any meaningful auth factors to prompt the user for) but the workflow works overall. I'll go through it in a comment, since I need to arrange some screenshots.
Test Plan: See guided walkthrough.
Reviewers: chad, btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T4398
Differential Revision: https://secure.phabricator.com/D8851
Summary:
- Dialog pages currently have no titles or crumbs, and look shoddy. Add titles and crumbs.
- Dialog titles aren't always great for crumbs, add an optional "short title" for crumbs.
- `AphrontDialogResponse` is pure boilerplate. Allow controllers to just return a `DialogView` instead and get the same effect.
- Building dialogs requires a bit of boilerplate, and we generally construct them with no explicit `"action"`, which has some issues with T4593. Provide a convenience method to set the viewer and get a reasonable, explict submit URI.
Test Plan:
- Viewed dialog on its own.
- Viewed dialog as a dialog.
{F132353}
Reviewers: btrahan, chad
Reviewed By: chad
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D8577
Summary:
This is partly a good feature, and partly should reduce false positives on HackerOne reporting things vaguely related to this.
Allow a user to terminate login sessions from the settings panel.
Test Plan:
- Terminated a session.
- Terminated all sessions.
- Tried to terminate all sessions again.
- Logged in with two browsers, terminated the other browser's session, reloaded, got kicked out.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Differential Revision: https://secure.phabricator.com/D8556
Summary:
Fixes T3471. Specific issues:
- Add the ability to set a temporary cookie (expires when the browser closes).
- We overwrote 'phcid' on every page load. This creates some issues with browser extensions. Instead, only write it if isn't set. To counterbalance this, make it temporary.
- Make the 'next_uri' cookie temporary.
- Make the 'phreg' cookie temporary.
- Fix an issue where deleted cookies would persist after 302 (?) in some cases (this is/was 100% for me locally).
Test Plan:
- Closed my browser, reopned it, verified temporary cookies were gone.
- Logged in, authed, linked, logged out.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: epriestley
Maniphest Tasks: T3471
Differential Revision: https://secure.phabricator.com/D8537
Summary:
See <https://github.com/facebook/phabricator/issues/541>.
- If a provider returns the email `""` or `"0"`, we currently don't let the user edit it and thus don't let them register.
- If a provider returns an invalid email like `"!!!"` (permitted by GitHub, e.g.), we show them a nonsense error message.
Instead:
- Pretend we didn't get an address if we get an invalid address.
- Test the address strictly against `null`.
Test Plan: Registered on Phabricator with my GitHub email set to `""` (empty string) and `"!!!"` (bang bang bang).
Reviewers: btrahan, chad
Reviewed By: chad
Subscribers: aran, epriestley
Differential Revision: https://secure.phabricator.com/D8528
Summary:
Ref T4593. There are a variety of clever attacks against OAuth which involve changing the redirect URI to some other URI on the same domain which exhibits unexpected behavior in response to an OAuth request. The best approach to dealing with this is for providers to lock to a specific path and refuse to redirect elsewhere, but not all providers do this.
We haven't had any specific issues related to this, but the anchor issue in T4593 was only a step away.
To mitigate this in general, we can reject the OAuth2 `'code'` parameter on //every// page by default, and then whitelist it on the tiny number of controllers which should be able to receive it.
This is very coarse, kind of overkill, and has some fallout (we can't use `'code'` as a normal parameter in the application), but I think it's relatively well-contained and seems reasonable. A better approach might be to whitelist parameters on every controller (i.e., have each controller specify the parameters it can receive), but that would be a ton of work and probably cause a lot of false positives for a long time.
Since we don't use `'code'` normally anywhere (as far as I can tell), the coarseness of this approach seems reasonable.
Test Plan:
- Logged in with OAuth.
- Hit any other page with `?code=...` in the URL, got an exception.
- Grepped for `'code'` and `"code"`, and examined each use to see if it was impacted.
Reviewers: btrahan
Reviewed By: btrahan
Subscribers: aran, epriestley
Maniphest Tasks: T4593
Differential Revision: https://secure.phabricator.com/D8499
Summary: Although the defaults don't require a verified email address, it's easy to lock yourself out by accident by configuring `auth.require-email-verification` or `auth.email-domains` before setting up email. Just force-verify the initial/setup account's address.
Test Plan: Went through setup on a fresh install, saw address verify.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D8365
Summary: If an attacker somehow intercepts a verification URL for an email address, they can hypothetically CSRF the account owner into verifying it. What you'd do before (how do you get the link?) and after (why do you care that you tricked them into verifying) performing this attack is unclear, but in theory we should require a CSRF submission here; add one.
Test Plan: {F118691}
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D8351
Summary: If you copy the registration URL, then register, then load the URL again while logged out (i.e., attempt to reuse the registration URL), we try to show you a tailored error message. However, this call is not correct so we show you a not-so tailored exception instead.
Test Plan:
- Get to the registration screen.
- Save URL.
- Complete registration.
- Log out.
- Return to saved URL.
Previously, exception. Now, readable error.
{F117585}
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Differential Revision: https://secure.phabricator.com/D8322
Summary:
OAuth1 doesn't have anything like the `state` parameter, and I overlooked that we need to shove one in there somewhere. Append it to the callback URI. This functions like `state` in OAuth2.
Without this, an attacker can trick a user into logging into Phabricator with an account the attacker controls.
Test Plan:
- Logged in with JIRA.
- Logged in with Twitter.
- Logged in with Facebook (an OAuth2 provider).
- Linked a Twitter account.
- Linked a Facebook account.
- Jiggered codes in URIs and verified that I got the exceptions I expected.
Reviewers: btrahan, arice
Reviewed By: arice
CC: arice, chad, aran
Differential Revision: https://secure.phabricator.com/D8318
Summary:
Via HackerOne. An attacker can bypass `auth.email-domains` by registering with an email like:
aaaaa...aaaaa@evil.com@company.com
We'll validate the full string, then insert it into the database where it will be truncated, removing the `@company.com` part. Then we'll send an email to `@evil.com`.
Instead, reject email addresses which won't fit in the table.
`STRICT_ALL_TABLES` stops this attack, I'm going to add a setup warning encouraging it.
Test Plan:
- Set `auth.email-domains` to `@company.com`.
- Registered with `aaa...aaa@evil.com@company.com`. Previously this worked, now it is rejected.
- Did a valid registration.
- Tried to add `aaa...aaaa@evil.com@company.com` as an email address. Previously this worked, now it is rejected.
- Did a valid email add.
- Added and executed unit tests.
Reviewers: btrahan, arice
Reviewed By: arice
CC: aran, chad
Differential Revision: https://secure.phabricator.com/D8308
Summary:
Ref T4443. Make hashing algorithms pluggable and extensible so we can deal with the attendant complexities more easily.
This moves "Iterated MD5" to a modular implementation, and adds a tiny bit of hack-glue so we don't need to migrate the DB in this patch. I'll migrate in the next patch, then add bcrypt.
Test Plan:
- Verified that the same stuff gets stored in the DB (i.e., no functional changes):
- Logged into an old password account.
- Changed password.
- Registered a new account.
- Changed password.
- Switched back to master.
- Logged in / out, changed password.
- Switched back, logged in.
- Ran unit tests (they aren't super extensive, but cover some of the basics).
Reviewers: btrahan
Reviewed By: btrahan
CC: aran, kofalt
Maniphest Tasks: T4443
Differential Revision: https://secure.phabricator.com/D8268
Summary:
Via HackerOne, there are two related low-severity issues with this workflow:
- We don't check if you're already logged in, so an attacker can trick a victim (whether they're logged in or not) into clicking a reset link for an account the attacker controls (maybe via an invisible iframe) and log the user in under a different account.
- We don't check CSRF tokens either, so after fixing the first thing, an attacker can still trick a //logged-out// victim in the same way.
It's not really clear that doing this opens up any significant attacks afterward, but both of these behaviors aren't good.
I'll probably land this for audit in a few hours if @btrahan doesn't have a chance to take a look at it since he's probably on a plane for most of the day, I'm pretty confident it doesn't break anything.
Test Plan:
- As a logged-in user, clicked another user's password reset link and was not logged in.
- As a logged-out user, clicked a password reset link and needed to submit a form to complete the workflow.
Reviewers: btrahan
CC: chad, btrahan, aran
Differential Revision: https://secure.phabricator.com/D8079
Summary:
Fixes T3793. There's a lot of history here, see D4012, T2102. Basically, the problem is that things used to work like this:
- User is logged out and accesses `/xyz/`. After they login, we'd like to send them back to `/xyz/`, so we set a `next_uri` cookie.
- User's browser has a bunch of extensions and now makes a ton of requests for stuff that doesn't exist, like `humans.txt` and `apple-touch-icon.png`. We can't distinguish between these requests and normal requests in a general way, so we write `next_uri` cookies, overwriting the user's intent (`/xyz/`).
To fix this, we made the 404 page not set `next_uri`, in D4012. So if the browser requests `humans.txt`, we 404 with no cookie, and the `/xyz/` cookie is preserved. However, this is bad because an attacker can determine if objects exist and applications are installed, by visiting, e.g., `/T123` and seeing if they get a 404 page (resource really does not exist) or a login page (resource exists). We'd rather not leak this information.
The comment in the body text describes this in more detail.
This diff sort of tries to do the right thing most of the time: we write the cookie only if we haven't written it in the last 2 minutes. Generally, this should mean that the original request to `/xyz/` writes it, all the `humans.txt` requests don't write it, and things work like users expect. This may occasionally do the wrong thing, but it should be very rare, and we stop leaking information about applications and objects.
Test Plan: Logged out, clicked around / logged in, used Charles to verify that cookies were set in the expected way.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T3793
Differential Revision: https://secure.phabricator.com/D8047
Summary:
Ref T4339. Ref T4310. Currently, sessions look like `"afad85d675fda87a4fadd54"`, and are only issued for logged-in users. To support logged-out CSRF and (eventually) external user sessions, I made two small changes:
- First, sessions now have a "kind", which is indicated by a prefix, like `"A/ab987asdcas7dca"`. This mostly allows us to issue session queries more efficiently: we don't have to issue a query at all for anonymous sessions, and can join the correct table for user and external sessions and save a query. Generally, this gives us more debugging information and more opportunity to recover from issues in a user-friendly way, as with the "invalid session" error in this diff.
- Secondly, if you load a page and don't have a session, we give you an anonymous session. This is just a secret with no special significance.
This does not implement CSRF yet, but gives us a client secret we can use to implement it.
Test Plan:
- Logged in.
- Logged out.
- Browsed around.
- Logged in again.
- Went through link/register.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T4310, T4339
Differential Revision: https://secure.phabricator.com/D8043
Summary: Ref T4339. We have more magical cookie names than we should, move them all to a central location.
Test Plan: Registered, logged in, linked account, logged out. See inlines.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran
Maniphest Tasks: T4339
Differential Revision: https://secure.phabricator.com/D8041
Summary:
Fixes T4143. This mitigates the "use a botnet to slowly try to login to every user account using the passwords '1234', 'password', 'asdfasdf', ..." attack, like the one that hit GitHub.
(I also donated some money to Openwall as a thanks for compiling this wordlist.)
Test Plan:
- Tried to register with a weak password; registered with a strong password.
- Tried to set VCS password to a weak password; set VCS password to a strong password.
- Tried to change password to a weak password; changed password to a strong password.
Reviewers: btrahan
Reviewed By: btrahan
CC: aran, chad
Maniphest Tasks: T4143
Differential Revision: https://secure.phabricator.com/D8048
