Summary:
I locked this down a little bit recently, but make
double-extra-super-sure that we aren't sending the user anywhere suspicious or
open-redirecty. This also locks down protocol-relative URIs (//evil.com/path)
although I don't think any browsers do bad stuff with them in this context, and
header injection URIs (although I don't think any of the modern PHP runtimes are
vulnerable).
Test Plan:
- Ran tests.
- Hit redirect page with valid and invalid next URIs; was punted to / for
invalid ones and to the right place for valid ones.
Reviewers: btrahan, jungejason
Reviewed By: btrahan
CC: arice, aran, epriestley, btrahan
Differential Revision: https://secure.phabricator.com/D1369
Summary: If a remote address has too many recent login failures, require they
fill out a captcha before they can attempt to login.
Test Plan: Tried to login a bunch of times, then submitted the CAPTHCA form with
various combinations of valid/invalid passwords and valid/invalid captchas.
Reviewers: btrahan, jungejason
Reviewed By: jungejason
CC: aran, epriestley, jungejason
Maniphest Tasks: T765
Differential Revision: https://secure.phabricator.com/D1379
interfaces
Summary:
- We have a hard-coded minimum length of 3 right now (and 1 in the other
interface), which is sort of silly.
- Provide a more reasonable default, and allow it to be configured.
- We have two password reset interfaces, one of which no longer actually
requires you to verify you own the account. This is more than a bit derp.
- Merge the interfaces into one, using either an email token or the account's
current password to let you change the password.
Test Plan:
- Reset password on an account.
- Changed password on an account.
- Created a new account, logged in, set the password.
- Tried to set a too-short password, got an error.
Reviewers: btrahan, jungejason, nh
Reviewed By: jungejason
CC: aran, jungejason
Maniphest Tasks: T766
Differential Revision: https://secure.phabricator.com/D1374
Summary:
- There are some recent reports of login issues, see T755 and T754. I'm not
really sure what's going on, but this is an attempt at getting some more
information.
- When we login a user by setting 'phusr' and 'phsid', send them to
/login/validate/ to validate that the cookies actually got set.
- Do email password resets in two steps: first, log the user in. Redirect them
through validate, then give them the option to reset their password.
- Don't CSRF logged-out users. It technically sort of works most of the time
right now, but is silly. If we need logged-out CSRF we should generate it in
some more reliable way.
Test Plan:
- Logged in with username/password.
- Logged in with OAuth.
- Logged in with email password reset.
- Sent bad values to /login/validate/, got appropriate errors.
- Reset password.
- Verified next_uri still works.
Reviewers: btrahan, jungejason
Reviewed By: btrahan
CC: aran, btrahan, j3kuntz
Maniphest Tasks: T754, T755
Differential Revision: https://secure.phabricator.com/D1353
Summary: These seem to work relatively reasonably and don't have any known
deal-breaking failures.
Test Plan: shrug~
Reviewers: btrahan, jungejason
Reviewed By: btrahan
CC: aran, btrahan
Differential Revision: https://secure.phabricator.com/D1324
Summary: See comments. A few installs have remarked that their organizations
would prefer buttons labled "Submit" to buttons labeled "Clowncopterize".
Test Plan:
- In "serious" mode, verified Differential and Maniphest have serious strings,
tasks can not be closed out of spite, and reset/welcome emails are extremely
serious.
- In unserious mode, verified Differential and Maniphest have normal strings,
tasks can be closed out of spite, and reset/welcome emails are silly.
- This does not disable the "fax these changes" message in Arcanist (no
reasonable way for it to read the config value) or the rainbow syntax
highlighter (already removable though configuration).
Reviewers: moskov, jungejason, nh, tuomaspelkonen, aran
Reviewed By: moskov
CC: aran, moskov
Differential Revision: 1081
Summary:
This is pretty straightforward, except:
- We need to request read/write access to the address book to get the account
ID (which we MUST have) and real name, email and account name (which we'd like
to have). This is way more access than we should need, but there's apparently no
"get_loggedin_user_basic_information" type of call in the Google API suite (or,
at least, I couldn't find one).
- We can't get the profile picture or profile URI since there's no Plus API
access and Google users don't have meaningful public pages otherwise.
- Google doesn't save the fact that you've authorized the app, so every time
you want to login you need to reaffirm that you want to give us silly amounts of
access. Phabricator sessions are pretty long-duration though so this shouldn't
be a major issue.
Test Plan:
- Registered, logged out, and logged in with Google.
- Registered, logged out, and logged in with Facebook / Github to make sure I
didn't break anything.
- Linked / unlinked Google accounts.
Reviewers: Makinde, jungejason, nh, tuomaspelkonen, aran
Reviewed By: aran
CC: aran, epriestley, Makinde
Differential Revision: 916
Summary:
@tomo ran into an issue where he had some non-SSL-only cookie or whatever, so
"Logout" had no apparent effect. Make sure "Logout" really works by destroying
the session.
I originally kept the sessions around to be able to debug session stuff, but we
have a fairly good session log now and no reprorted session bugs except for all
the cookie stuff. It's also slightly more secure to actually destroy sessions,
since it means "logout" breaks any cookies that attackers somehow stole (e.g.,
by reading your requests off a public wifi network).
Test Plan: Commented out the cookie clear and logged out. I was logged out and
given a useful error message about clearing my cookies.
Reviewers: jungejason, nh, tuomaspelkonen, aran
Reviewed By: aran
CC: tomo, aran, epriestley
Differential Revision: 911
Summary: Open AphrontWriteGuard for user login.
Test Plan: verified that the user can log in.
Reviewers: epriestley
Reviewed By: epriestley
CC: aran, epriestley
Differential Revision: 840
Summary:
Provide a catchall mechanism to find unprotected writes.
- Depends on D758.
- Similar to WriteOnHTTPGet stuff from Facebook's stack.
- Since we have a small number of storage mechanisms and highly structured
read/write pathways, we can explicitly answer the question "is this page
performing a write?".
- Never allow writes without CSRF checks.
- This will probably break some things. That's fine: they're CSRF
vulnerabilities or weird edge cases that we can fix. But don't push to Facebook
for a few days unless you're prepared to deal with this.
- **>>> MEGADERP: All Conduit write APIs are currently vulnerable to CSRF!
<<<**
Test Plan:
- Ran some scripts that perform writes (scripts/search indexers), no issues.
- Performed normal CSRF submits.
- Added writes to an un-CSRF'd page, got an exception.
- Executed conduit methods.
- Did login/logout (this works because the logged-out user validates the
logged-out csrf "token").
- Did OAuth login.
- Did OAuth registration.
Reviewers: pedram, andrewjcg, erling, jungejason, tuomaspelkonen, aran,
codeblock
Commenters: pedram
CC: aran, epriestley, pedram
Differential Revision: 777
Summary:
We currently cycle CSRF tokens every hour and check for the last two valid ones.
This means that a form could go stale in as little as an hour, and is certainly
stale after two.
When a stale form is submitted, you basically get a terrible heisen-state where
some of your data might persist if you're lucky but more likely it all just
vanishes. The .js file below outlines some more details.
This is a pretty terrible UX and we don't need to be as conservative about CSRF
validation as we're being. Remedy this problem by:
- Accepting the last 6 CSRF tokens instead of the last 1 (i.e., pages are
valid for at least 6 hours, and for as long as 7).
- Using JS to refresh the CSRF token every 55 minutes (i.e., pages connected
to the internet are valid indefinitely).
- Showing the user an explicit message about what went wrong when CSRF
validation fails so the experience is less bewildering.
They should now only be able to submit with a bad CSRF token if:
- They load a page, disconnect from the internet for 7 hours, reconnect, and
submit the form within 55 minutes; or
- They are actually the victim of a CSRF attack.
We could eventually fix the first one by tracking reconnects, which might be
"free" once the notification server gets built. It will probably never be an
issue in practice.
Test Plan:
- Reduced CSRF cycle frequency to 2 seconds, submitted a form after 15
seconds, got the CSRF exception.
- Reduced csrf-refresh cycle frequency to 3 seconds, submitted a form after 15
seconds, got a clean form post.
- Added debugging code the the csrf refresh to make sure it was doing sensible
things (pulling different tokens, finding all the inputs).
Reviewed By: aran
Reviewers: tuomaspelkonen, jungejason, aran
CC: aran, epriestley
Differential Revision: 660
Summary:
- have files be uploaded by drag+drop instead of browse.
- Files are named by their uploaded filename, the user isn't given a chance to enter a file name. Is this bad?
- Store author PHID now with files
- Allow an ?author=<username> to limit the /files/ list by author.
- If one file is uploaded, the user is taken to its info page.
- If several are uploaded, they are taken to a list of their files.
Test Plan:
- Quickly tested everything and it still worked, I'd recommend some people try this out before it gets committed though. It's a rather huge revision.
Reviewers:
epriestley, Ttech
CC:
Differential Revision: 612
Summary:
Send the user where they were intending to go after github and localized logins.
Before, because Github didn't send oauthState, we would force / upon them.
Test Plan:
Tried all three methods of login successfully.
Reviewers:
epriestley
CC:
Differential Revision: 602
Summary:
- When an administrator creates a user, provide an option to send a welcome
email. Right now this workflow kind of dead-ends.
- Prevent administrators from changing the "System Agent" flag. If they can
change it, they can grab another user's certificate and then act as them. This
is a vaguely weaker security policy than is exhibited elsewhere in the
application. Instead, make user accounts immutably normal users or system agents
at creation time.
- Prevent administrators from changing email addresses after account creation.
Same deal as conduit certs. The 'bin/accountadmin' script can still do this if a
user has a real problem.
- Prevent administrators from resetting passwords. There's no need for this
anymore with welcome emails plus email login and it raises the same issues.
Test Plan:
- Created a new account, selected "send welcome email", got a welcome email,
logged in with the link inside it.
- Created a new system agent.
- Reset an account's password.
Reviewed By: aran
Reviewers: tuomaspelkonen, jungejason, aran
CC: anjali, aran, epriestley
Differential Revision: 379
Summary:
Allows you to link to comments with "D123#3" or "T123#3", then adds a pile of JS
to try to make it not terrible. :/
The thing I'm trying to avoid here is when someone says "look at this!
http://blog.com/#comment-239291" and you click and your browser jumps somewhere
random and you have no idea which comment they meant. Since I really hate this,
I've tried to avoid it by making sure the comment is always highlighted.
Test Plan:
Put T1#1 and D1#1 in remarkup and verified they linked properly.
Clicked anchors on individual comments.
Faked all comments hidden in Differential and verified they expanded on anchor
or anchor change.
Reviewed By: aran
Reviewers: aran, tomo, mroch, jungejason, tuomaspelkonen
CC: aran, epriestley
Differential Revision: 383
Summary:
Despite the form's claims that you can login with username or email, it actually
accepted only username.
Test Plan:
Logged in using my email.
Reviewed By: jungejason
Reviewers: tuomaspelkonen, jungejason, aran
CC: jr, anjali, aran, jungejason
Differential Revision: 354
Summary: This isn't complete, but I figured I'd ship it for review while it's still smallish.
Provide an activity log for high-level system actions (logins, admin actions). This basically allows two things to happen:
- The log itself is useful if there are shenanigans.
- Password login can check it and start CAPTCHA'ing users after a few failed attempts.
I'm going to change how the admin stuff works a little bit too, since right now you can make someone an agent, grab their certificate, revert them back to a normal user, and then act on their behalf over Conduit. This is a little silly, I'm going to move "agent" to the create workflow instead. I'll also add a confirm/email step to the administrative password reset flow.
Test Plan: Took various administrative and non-administrative actions, they appeared in the logs. Filtered the logs in a bunch of different ways.
Reviewers: jungejason, tuomaspelkonen, aran
CC:
Differential Revision: 302
Summary:
Provide an "isAdmin" flag for users, to designate administrative users.
Restore the account editing interface and allow it to set role flags and reset
passwords.
Provide an "isDisabled" flag for users and shut down all system access for them.
Test Plan:
Created "admin" and "disabled" users. Did administrative things with the admin
user. Tried to do stuff with the disabled user and was rebuffed. Tried to access
administrative interfaces with a normal non-admin user and was denied.
Reviewed By: aran
Reviewers: tuomaspelkonen, jungejason, aran
CC: ccheever, aran
Differential Revision: 278
Summary:
Github allows you to have an account without a real name. The OAuth controller
actually handles this fine, mostly, except that it calls a bogus method. Also
there is some null vs empty string confusion.
Test Plan:
Deleted my name on Github and then registered for an account on Phabricator.
Reviewed By: tuomaspelkonen
Reviewers: jungejason, tuomaspelkonen, aran
CC: anjali, aran, tuomaspelkonen
Differential Revision: 247
Summary:
I pretty shortsightedly made sending a side effect of save() in the case that a
server is configured for immediate sending. Move this out, make it explicit, and
get rid of all the tangles surrounding it.
The web tool now ignores the server setting and only repsects the checkbox,
which makes far more sense.
Test Plan:
Sent mails from Maniphest, Differential, and the web console. Also ran all the
unit tests. Verified headers from Maniphest.
Reviewed By: rm
Reviewers: aran, rm
CC: tuomaspelkonen, rm, jungejason, aran
Differential Revision: 200
Summary:
There's an OAuth diagnostics page at /oauth/facebook/diagnose/, which
shows some diagnostic information. Currently, it attempts to establish an
application token session and shows the token if it is successful. An attacker
could use this to do vaguely nefarious things (retreive application statistics,
I think?).
This interface was originally admin-only but then I threw out the very silly
admin mode patch I had at the time and we currently have no admin mode, and
thus this interface is public. This token isn't useful in diagnosis anyway,
so don't reveal it.
Test Plan:
Visited oauth diagnostics page, no token revealed
Reviewed By: tuomaspelkonen
Reviewers: tuomaspelkonen, jungejason
CC: tuomaspelkonen
Differential Revision: 136
Summary:
When a user clicks a link like /T32 and has to login, redirect them
to the resource once they've authenticated if possible. OAuth has a param
specifically for this, called 'state', so use it if possible. Facebook
supports it but Github does not.
Test Plan:
logged in with facebook after viewing /D20
Reviewed By: aran
Reviewers: aran
CC: aran, epriestley
Differential Revision: 61
