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
